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  *
  31  * lustre/llite/llite_lib.c
  32  *
  33  * Lustre Light Super operations
  34  */
  35
  36 #define DEBUG_SUBSYSTEM S_LLITE
  37
  38 #include <linux/cpu.h>
  39 #include <linux/module.h>
  40 #include <linux/random.h>
  41 #include <linux/statfs.h>
  42 #include <linux/time.h>
  43 #include <linux/file.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/fs_struct.h>
  52
  53 #ifndef HAVE_CPUS_READ_LOCK
  54 #include <libcfs/linux/linux-cpu.h>
  55 #endif
  56 #include <libcfs/linux/linux-misc.h>
  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
  97         ENTRY;
  98
  99         OBD_ALLOC_PTR(sbi);
 100         if (sbi == NULL)
 101                 RETURN(ERR_PTR(-ENOMEM));
 102
 103         rc = pcc_super_init(&sbi->ll_pcc_super);
 104         if (rc < 0)
 105                 GOTO(out_sbi, rc);
 106
 107         spin_lock_init(&sbi->ll_lock);
 108         mutex_init(&sbi->ll_lco.lco_lock);
 109         spin_lock_init(&sbi->ll_pp_extent_lock);
 110         spin_lock_init(&sbi->ll_process_lock);
 111         sbi->ll_rw_stats_on = 0;
 112         sbi->ll_statfs_max_age = OBD_STATFS_CACHE_SECONDS;
 113
 114         si_meminfo(&si);
 115         pages = si.totalram - si.totalhigh;
 116         lru_page_max = pages / 2;
 117
 118         sbi->ll_ra_info.ra_async_max_active = ll_get_ra_async_max_active();
 119         sbi->ll_ra_info.ll_readahead_wq =
 120                 cfs_cpt_bind_workqueue("ll-readahead-wq", cfs_cpt_tab,
 121                                        0, CFS_CPT_ANY,
 122                                        sbi->ll_ra_info.ra_async_max_active);
 123         if (IS_ERR(sbi->ll_ra_info.ll_readahead_wq))
 124                 GOTO(out_pcc, rc = PTR_ERR(sbi->ll_ra_info.ll_readahead_wq));
 125
 126         /* initialize ll_cache data */
 127         sbi->ll_cache = cl_cache_init(lru_page_max);
 128         if (sbi->ll_cache == NULL)
 129                 GOTO(out_destroy_ra, rc = -ENOMEM);
 130
 131         /* initialize foreign symlink prefix path */
 132         OBD_ALLOC(sbi->ll_foreign_symlink_prefix, sizeof("/mnt/"));
 133         if (sbi->ll_foreign_symlink_prefix == NULL)
 134                 GOTO(out_destroy_ra, rc = -ENOMEM);
 135         memcpy(sbi->ll_foreign_symlink_prefix, "/mnt/", sizeof("/mnt/"));
 136         sbi->ll_foreign_symlink_prefix_size = sizeof("/mnt/");
 137
 138         /* initialize foreign symlink upcall path, none by default */
 139         OBD_ALLOC(sbi->ll_foreign_symlink_upcall, sizeof("none"));
 140         if (sbi->ll_foreign_symlink_upcall == NULL)
 141                 GOTO(out_destroy_ra, rc = -ENOMEM);
 142         memcpy(sbi->ll_foreign_symlink_upcall, "none", sizeof("none"));
 143         sbi->ll_foreign_symlink_upcall_items = NULL;
 144         sbi->ll_foreign_symlink_upcall_nb_items = 0;
 145         init_rwsem(&sbi->ll_foreign_symlink_sem);
 146         /* foreign symlink support (LL_SBI_FOREIGN_SYMLINK in ll_flags)
 147          * not enabled by default
 148          */
 149
 150         sbi->ll_ra_info.ra_max_pages =
 151                 min(pages / 32, SBI_DEFAULT_READ_AHEAD_MAX);
 152         sbi->ll_ra_info.ra_max_pages_per_file =
 153                 min(sbi->ll_ra_info.ra_max_pages / 4,
 154                     SBI_DEFAULT_READ_AHEAD_PER_FILE_MAX);
 155         sbi->ll_ra_info.ra_async_pages_per_file_threshold =
 156                                 sbi->ll_ra_info.ra_max_pages_per_file;
 157         sbi->ll_ra_info.ra_range_pages = SBI_DEFAULT_RA_RANGE_PAGES;
 158         sbi->ll_ra_info.ra_max_read_ahead_whole_pages = -1;
 159         atomic_set(&sbi->ll_ra_info.ra_async_inflight, 0);
 160
 161         set_bit(LL_SBI_VERBOSE, sbi->ll_flags);
 162 #ifdef ENABLE_CHECKSUM
 163         set_bit(LL_SBI_CHECKSUM, sbi->ll_flags);
 164 #endif
 165 #ifdef ENABLE_FLOCK
 166         set_bit(LL_SBI_FLOCK, sbi->ll_flags);
 167 #endif
 168
 169 #ifdef HAVE_LRU_RESIZE_SUPPORT
 170         set_bit(LL_SBI_LRU_RESIZE, sbi->ll_flags);
 171 #endif
 172         set_bit(LL_SBI_LAZYSTATFS, sbi->ll_flags);
 173
 174         /* metadata statahead is enabled by default */
 175         sbi->ll_sa_running_max = LL_SA_RUNNING_DEF;
 176         sbi->ll_sa_max = LL_SA_RPC_DEF;
 177         atomic_set(&sbi->ll_sa_total, 0);
 178         atomic_set(&sbi->ll_sa_wrong, 0);
 179         atomic_set(&sbi->ll_sa_running, 0);
 180         atomic_set(&sbi->ll_agl_total, 0);
 181         set_bit(LL_SBI_AGL_ENABLED, sbi->ll_flags);
 182         set_bit(LL_SBI_FAST_READ, sbi->ll_flags);
 183         set_bit(LL_SBI_TINY_WRITE, sbi->ll_flags);
 184         set_bit(LL_SBI_PARALLEL_DIO, sbi->ll_flags);
 185         ll_sbi_set_encrypt(sbi, true);
 186
 187         /* root squash */
 188         sbi->ll_squash.rsi_uid = 0;
 189         sbi->ll_squash.rsi_gid = 0;
 190         INIT_LIST_HEAD(&sbi->ll_squash.rsi_nosquash_nids);
 191         spin_lock_init(&sbi->ll_squash.rsi_lock);
 192
 193         /* Per-filesystem file heat */
 194         sbi->ll_heat_decay_weight = SBI_DEFAULT_HEAT_DECAY_WEIGHT;
 195         sbi->ll_heat_period_second = SBI_DEFAULT_HEAT_PERIOD_SECOND;
 196
 197         /* Per-fs open heat level before requesting open lock */
 198         sbi->ll_oc_thrsh_count = SBI_DEFAULT_OPENCACHE_THRESHOLD_COUNT;
 199         sbi->ll_oc_max_ms = SBI_DEFAULT_OPENCACHE_THRESHOLD_MAX_MS;
 200         sbi->ll_oc_thrsh_ms = SBI_DEFAULT_OPENCACHE_THRESHOLD_MS;
 201         RETURN(sbi);
 202 out_destroy_ra:
 203         if (sbi->ll_foreign_symlink_prefix)
 204                 OBD_FREE(sbi->ll_foreign_symlink_prefix, sizeof("/mnt/"));
 205         if (sbi->ll_cache) {
 206                 cl_cache_decref(sbi->ll_cache);
 207                 sbi->ll_cache = NULL;
 208         }
 209         destroy_workqueue(sbi->ll_ra_info.ll_readahead_wq);
 210 out_pcc:
 211         pcc_super_fini(&sbi->ll_pcc_super);
 212 out_sbi:
 213         OBD_FREE_PTR(sbi);
 214         RETURN(ERR_PTR(rc));
 215 }
 216
 217 static void ll_free_sbi(struct super_block *sb)
 218 {
 219         struct ll_sb_info *sbi = ll_s2sbi(sb);
 220         ENTRY;
 221
 222         if (sbi != NULL) {
 223                 if (!list_empty(&sbi->ll_squash.rsi_nosquash_nids))
 224                         cfs_free_nidlist(&sbi->ll_squash.rsi_nosquash_nids);
 225                 if (sbi->ll_ra_info.ll_readahead_wq)
 226                         destroy_workqueue(sbi->ll_ra_info.ll_readahead_wq);
 227                 if (sbi->ll_cache != NULL) {
 228                         cl_cache_decref(sbi->ll_cache);
 229                         sbi->ll_cache = NULL;
 230                 }
 231                 if (sbi->ll_foreign_symlink_prefix) {
 232                         OBD_FREE(sbi->ll_foreign_symlink_prefix,
 233                                  sbi->ll_foreign_symlink_prefix_size);
 234                         sbi->ll_foreign_symlink_prefix = NULL;
 235                 }
 236                 if (sbi->ll_foreign_symlink_upcall) {
 237                         OBD_FREE(sbi->ll_foreign_symlink_upcall,
 238                                  strlen(sbi->ll_foreign_symlink_upcall) +
 239                                        1);
 240                         sbi->ll_foreign_symlink_upcall = NULL;
 241                 }
 242                 if (sbi->ll_foreign_symlink_upcall_items) {
 243                         int i;
 244                         int nb_items = sbi->ll_foreign_symlink_upcall_nb_items;
 245                         struct ll_foreign_symlink_upcall_item *items =
 246                                 sbi->ll_foreign_symlink_upcall_items;
 247
 248                         for (i = 0 ; i < nb_items; i++)
 249                                 if (items[i].type == STRING_TYPE)
 250                                         OBD_FREE(items[i].string,
 251                                                        items[i].size);
 252
 253                         OBD_FREE_LARGE(items, nb_items *
 254                                 sizeof(struct ll_foreign_symlink_upcall_item));
 255                         sbi->ll_foreign_symlink_upcall_items = NULL;
 256                 }
 257                 ll_free_rw_stats_info(sbi);
 258                 pcc_super_fini(&sbi->ll_pcc_super);
 259                 OBD_FREE(sbi, sizeof(*sbi));
 260         }
 261         EXIT;
 262 }
 263
 264 static int client_common_fill_super(struct super_block *sb, char *md, char *dt)
 265 {
 266         struct inode *root = NULL;
 267         struct ll_sb_info *sbi = ll_s2sbi(sb);
 268         struct obd_statfs *osfs = NULL;
 269         struct ptlrpc_request *request = NULL;
 270         struct obd_connect_data *data = NULL;
 271         struct obd_uuid *uuid;
 272         struct md_op_data *op_data;
 273         struct lustre_md lmd;
 274         u64 valid;
 275         int size, err, checksum;
 276         bool api32;
 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  |
 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_GRANT_SHRINK |
 322                                   OBD_CONNECT_SHORTIO | OBD_CONNECT_FLAGS2;
 323
 324         data->ocd_connect_flags2 = OBD_CONNECT2_DIR_MIGRATE |
 325                                    OBD_CONNECT2_SUM_STATFS |
 326                                    OBD_CONNECT2_OVERSTRIPING |
 327                                    OBD_CONNECT2_FLR |
 328                                    OBD_CONNECT2_LOCK_CONVERT |
 329                                    OBD_CONNECT2_ARCHIVE_ID_ARRAY |
 330                                    OBD_CONNECT2_INC_XID |
 331                                    OBD_CONNECT2_LSOM |
 332                                    OBD_CONNECT2_ASYNC_DISCARD |
 333                                    OBD_CONNECT2_PCC |
 334                                    OBD_CONNECT2_CRUSH | OBD_CONNECT2_LSEEK |
 335                                    OBD_CONNECT2_GETATTR_PFID |
 336                                    OBD_CONNECT2_DOM_LVB |
 337                                    OBD_CONNECT2_REP_MBITS |
 338                                    OBD_CONNECT2_ATOMIC_OPEN_LOCK;
 339
 340 #ifdef HAVE_LRU_RESIZE_SUPPORT
 341         if (test_bit(LL_SBI_LRU_RESIZE, sbi->ll_flags))
 342                 data->ocd_connect_flags |= OBD_CONNECT_LRU_RESIZE;
 343 #endif
 344         data->ocd_connect_flags |= OBD_CONNECT_ACL_FLAGS;
 345
 346         data->ocd_cksum_types = obd_cksum_types_supported_client();
 347
 348         if (OBD_FAIL_CHECK(OBD_FAIL_MDC_LIGHTWEIGHT))
 349                 /* flag mdc connection as lightweight, only used for test
 350                  * purpose, use with care */
 351                 data->ocd_connect_flags |= OBD_CONNECT_LIGHTWEIGHT;
 352
 353         data->ocd_ibits_known = MDS_INODELOCK_FULL;
 354         data->ocd_version = LUSTRE_VERSION_CODE;
 355
 356         if (sb->s_flags & SB_RDONLY)
 357                 data->ocd_connect_flags |= OBD_CONNECT_RDONLY;
 358         if (test_bit(LL_SBI_USER_XATTR, sbi->ll_flags))
 359                 data->ocd_connect_flags |= OBD_CONNECT_XATTR;
 360
 361 #ifdef SB_NOSEC
 362         /* Setting this indicates we correctly support S_NOSEC (See kernel
 363          * commit 9e1f1de02c2275d7172e18dc4e7c2065777611bf)
 364          */
 365         sb->s_flags |= SB_NOSEC;
 366 #endif
 367         sbi->ll_fop = ll_select_file_operations(sbi);
 368
 369         /* always ping even if server suppress_pings */
 370         if (test_bit(LL_SBI_ALWAYS_PING, sbi->ll_flags))
 371                 data->ocd_connect_flags &= ~OBD_CONNECT_PINGLESS;
 372
 373         obd_connect_set_secctx(data);
 374         if (ll_sbi_has_encrypt(sbi))
 375                 obd_connect_set_enc(data);
 376
 377 #if defined(CONFIG_SECURITY)
 378         data->ocd_connect_flags2 |= OBD_CONNECT2_SELINUX_POLICY;
 379 #endif
 380
 381         data->ocd_brw_size = MD_MAX_BRW_SIZE;
 382
 383         err = obd_connect(NULL, &sbi->ll_md_exp, sbi->ll_md_obd,
 384                           &sbi->ll_sb_uuid, data, sbi->ll_cache);
 385         if (err == -EBUSY) {
 386                 LCONSOLE_ERROR_MSG(0x14f, "An MDT (md %s) is performing "
 387                                    "recovery, of which this client is not a "
 388                                    "part. Please wait for recovery to complete,"
 389                                    " abort, or time out.\n", md);
 390                 GOTO(out, err);
 391         } else if (err) {
 392                 CERROR("cannot connect to %s: rc = %d\n", md, err);
 393                 GOTO(out, err);
 394         }
 395
 396         sbi->ll_md_exp->exp_connect_data = *data;
 397
 398         err = obd_fid_init(sbi->ll_md_exp->exp_obd, sbi->ll_md_exp,
 399                            LUSTRE_SEQ_METADATA);
 400         if (err) {
 401                 CERROR("%s: Can't init metadata layer FID infrastructure, "
 402                        "rc = %d\n", sbi->ll_md_exp->exp_obd->obd_name, err);
 403                 GOTO(out_md, err);
 404         }
 405
 406         /* For mount, we only need fs info from MDT0, and also in DNE, it
 407          * can make sure the client can be mounted as long as MDT0 is
 408          * avaible */
 409         err = obd_statfs(NULL, sbi->ll_md_exp, osfs,
 410                         ktime_get_seconds() - sbi->ll_statfs_max_age,
 411                         OBD_STATFS_FOR_MDT0);
 412         if (err)
 413                 GOTO(out_md_fid, err);
 414
 415         /* This needs to be after statfs to ensure connect has finished.
 416          * Note that "data" does NOT contain the valid connect reply.
 417          * If connecting to a 1.8 server there will be no LMV device, so
 418          * we can access the MDC export directly and exp_connect_flags will
 419          * be non-zero, but if accessing an upgraded 2.1 server it will
 420          * have the correct flags filled in.
 421          * XXX: fill in the LMV exp_connect_flags from MDC(s). */
 422         valid = exp_connect_flags(sbi->ll_md_exp) & CLIENT_CONNECT_MDT_REQD;
 423         if (exp_connect_flags(sbi->ll_md_exp) != 0 &&
 424             valid != CLIENT_CONNECT_MDT_REQD) {
 425                 char *buf;
 426
 427                 OBD_ALLOC_WAIT(buf, PAGE_SIZE);
 428                 obd_connect_flags2str(buf, PAGE_SIZE,
 429                                       valid ^ CLIENT_CONNECT_MDT_REQD, 0, ",");
 430                 LCONSOLE_ERROR_MSG(0x170, "Server %s does not support "
 431                                    "feature(s) needed for correct operation "
 432                                    "of this client (%s). Please upgrade "
 433                                    "server or downgrade client.\n",
 434                                    sbi->ll_md_exp->exp_obd->obd_name, buf);
 435                 OBD_FREE(buf, PAGE_SIZE);
 436                 GOTO(out_md_fid, err = -EPROTO);
 437         }
 438
 439         size = sizeof(*data);
 440         err = obd_get_info(NULL, sbi->ll_md_exp, sizeof(KEY_CONN_DATA),
 441                            KEY_CONN_DATA,  &size, data);
 442         if (err) {
 443                 CERROR("%s: Get connect data failed: rc = %d\n",
 444                        sbi->ll_md_exp->exp_obd->obd_name, err);
 445                 GOTO(out_md_fid, err);
 446         }
 447
 448         LASSERT(osfs->os_bsize);
 449         sb->s_blocksize = osfs->os_bsize;
 450         sb->s_blocksize_bits = log2(osfs->os_bsize);
 451         sb->s_magic = LL_SUPER_MAGIC;
 452         sb->s_maxbytes = MAX_LFS_FILESIZE;
 453         sbi->ll_namelen = osfs->os_namelen;
 454         sbi->ll_mnt.mnt = current->fs->root.mnt;
 455
 456         if (test_bit(LL_SBI_USER_XATTR, sbi->ll_flags) &&
 457             !(data->ocd_connect_flags & OBD_CONNECT_XATTR)) {
 458                 LCONSOLE_INFO("Disabling user_xattr feature because "
 459                               "it is not supported on the server\n");
 460                 clear_bit(LL_SBI_USER_XATTR, sbi->ll_flags);
 461         }
 462
 463         if (data->ocd_connect_flags & OBD_CONNECT_ACL) {
 464 #ifdef SB_POSIXACL
 465                 sb->s_flags |= SB_POSIXACL;
 466 #endif
 467                 set_bit(LL_SBI_ACL, sbi->ll_flags);
 468         } else {
 469                 LCONSOLE_INFO("client wants to enable acl, but mdt not!\n");
 470 #ifdef SB_POSIXACL
 471                 sb->s_flags &= ~SB_POSIXACL;
 472 #endif
 473                 clear_bit(LL_SBI_ACL, sbi->ll_flags);
 474         }
 475
 476         if (data->ocd_connect_flags & OBD_CONNECT_64BITHASH)
 477                 set_bit(LL_SBI_64BIT_HASH, sbi->ll_flags);
 478
 479         if (data->ocd_connect_flags & OBD_CONNECT_LAYOUTLOCK)
 480                 set_bit(LL_SBI_LAYOUT_LOCK, sbi->ll_flags);
 481
 482         if (obd_connect_has_secctx(data))
 483                 set_bit(LL_SBI_FILE_SECCTX, sbi->ll_flags);
 484
 485         if (ll_sbi_has_encrypt(sbi) && !obd_connect_has_enc(data)) {
 486                 if (ll_sbi_has_test_dummy_encryption(sbi))
 487                         LCONSOLE_WARN("%s: server %s does not support encryption feature, encryption deactivated.\n",
 488                                       sbi->ll_fsname,
 489                                       sbi->ll_md_exp->exp_obd->obd_name);
 490                 ll_sbi_set_encrypt(sbi, false);
 491         }
 492
 493         if (data->ocd_ibits_known & MDS_INODELOCK_XATTR) {
 494                 if (!(data->ocd_connect_flags & OBD_CONNECT_MAX_EASIZE)) {
 495                         LCONSOLE_INFO("%s: disabling xattr cache due to "
 496                                       "unknown maximum xattr size.\n", dt);
 497                 } else if (!sbi->ll_xattr_cache_set) {
 498                         /* If xattr_cache is already set (no matter 0 or 1)
 499                          * during processing llog, it won't be enabled here. */
 500                         set_bit(LL_SBI_XATTR_CACHE, sbi->ll_flags);
 501                         sbi->ll_xattr_cache_enabled = 1;
 502                 }
 503         }
 504
 505         sbi->ll_dt_obd = class_name2obd(dt);
 506         if (!sbi->ll_dt_obd) {
 507                 CERROR("DT %s: not setup or attached\n", dt);
 508                 GOTO(out_md_fid, err = -ENODEV);
 509         }
 510
 511         /* pass client page size via ocd_grant_blkbits, the server should report
 512          * back its backend blocksize for grant calculation purpose */
 513         data->ocd_grant_blkbits = PAGE_SHIFT;
 514
 515         /* indicate OST features supported by this client */
 516         data->ocd_connect_flags = OBD_CONNECT_GRANT | OBD_CONNECT_VERSION |
 517                                   OBD_CONNECT_REQPORTAL | OBD_CONNECT_BRW_SIZE |
 518                                   OBD_CONNECT_CANCELSET | OBD_CONNECT_FID |
 519                                   OBD_CONNECT_SRVLOCK |
 520                                   OBD_CONNECT_AT | OBD_CONNECT_OSS_CAPA |
 521                                   OBD_CONNECT_VBR | OBD_CONNECT_FULL20 |
 522                                   OBD_CONNECT_64BITHASH | OBD_CONNECT_MAXBYTES |
 523                                   OBD_CONNECT_EINPROGRESS |
 524                                   OBD_CONNECT_JOBSTATS | OBD_CONNECT_LVB_TYPE |
 525                                   OBD_CONNECT_LAYOUTLOCK |
 526                                   OBD_CONNECT_PINGLESS | OBD_CONNECT_LFSCK |
 527                                   OBD_CONNECT_BULK_MBITS | OBD_CONNECT_SHORTIO |
 528                                   OBD_CONNECT_FLAGS2 | OBD_CONNECT_GRANT_SHRINK;
 529         data->ocd_connect_flags2 = OBD_CONNECT2_LOCKAHEAD |
 530                                    OBD_CONNECT2_INC_XID | OBD_CONNECT2_LSEEK |
 531                                    OBD_CONNECT2_REP_MBITS;
 532
 533         if (!OBD_FAIL_CHECK(OBD_FAIL_OSC_CONNECT_GRANT_PARAM))
 534                 data->ocd_connect_flags |= OBD_CONNECT_GRANT_PARAM;
 535
 536         /* OBD_CONNECT_CKSUM should always be set, even if checksums are
 537          * disabled by default, because it can still be enabled on the
 538          * fly via /sys. As a consequence, we still need to come to an
 539          * agreement on the supported algorithms at connect time
 540          */
 541         data->ocd_connect_flags |= OBD_CONNECT_CKSUM;
 542
 543         if (OBD_FAIL_CHECK(OBD_FAIL_OSC_CKSUM_ADLER_ONLY))
 544                 data->ocd_cksum_types = OBD_CKSUM_ADLER;
 545         else
 546                 data->ocd_cksum_types = obd_cksum_types_supported_client();
 547
 548 #ifdef HAVE_LRU_RESIZE_SUPPORT
 549         data->ocd_connect_flags |= OBD_CONNECT_LRU_RESIZE;
 550 #endif
 551         /* always ping even if server suppress_pings */
 552         if (test_bit(LL_SBI_ALWAYS_PING, sbi->ll_flags))
 553                 data->ocd_connect_flags &= ~OBD_CONNECT_PINGLESS;
 554
 555         if (ll_sbi_has_encrypt(sbi))
 556                 obd_connect_set_enc(data);
 557
 558         CDEBUG(D_RPCTRACE, "ocd_connect_flags: %#llx ocd_version: %d "
 559                "ocd_grant: %d\n", data->ocd_connect_flags,
 560                data->ocd_version, data->ocd_grant);
 561
 562         sbi->ll_dt_obd->obd_upcall.onu_owner = &sbi->ll_lco;
 563         sbi->ll_dt_obd->obd_upcall.onu_upcall = cl_ocd_update;
 564
 565         data->ocd_brw_size = DT_MAX_BRW_SIZE;
 566
 567         err = obd_connect(NULL, &sbi->ll_dt_exp, sbi->ll_dt_obd,
 568                           &sbi->ll_sb_uuid, data, sbi->ll_cache);
 569         if (err == -EBUSY) {
 570                 LCONSOLE_ERROR_MSG(0x150, "An OST (dt %s) is performing "
 571                                    "recovery, of which this client is not a "
 572                                    "part.  Please wait for recovery to "
 573                                    "complete, abort, or time out.\n", dt);
 574                 GOTO(out_md, err);
 575         } else if (err) {
 576                 CERROR("%s: Cannot connect to %s: rc = %d\n",
 577                        sbi->ll_dt_exp->exp_obd->obd_name, dt, err);
 578                 GOTO(out_md, err);
 579         }
 580
 581         if (ll_sbi_has_encrypt(sbi) &&
 582             !obd_connect_has_enc(&sbi->ll_dt_obd->u.lov.lov_ocd)) {
 583                 if (ll_sbi_has_test_dummy_encryption(sbi))
 584                         LCONSOLE_WARN("%s: server %s does not support encryption feature, encryption deactivated.\n",
 585                                       sbi->ll_fsname, dt);
 586                 ll_sbi_set_encrypt(sbi, false);
 587         } else if (ll_sbi_has_test_dummy_encryption(sbi)) {
 588                 LCONSOLE_WARN("Test dummy encryption mode enabled\n");
 589         }
 590
 591         sbi->ll_dt_exp->exp_connect_data = *data;
 592
 593         /* Don't change value if it was specified in the config log */
 594         if (sbi->ll_ra_info.ra_max_read_ahead_whole_pages == -1) {
 595                 sbi->ll_ra_info.ra_max_read_ahead_whole_pages =
 596                         max_t(unsigned long, SBI_DEFAULT_READ_AHEAD_WHOLE_MAX,
 597                               (data->ocd_brw_size >> PAGE_SHIFT));
 598                 if (sbi->ll_ra_info.ra_max_read_ahead_whole_pages >
 599                     sbi->ll_ra_info.ra_max_pages_per_file)
 600                         sbi->ll_ra_info.ra_max_read_ahead_whole_pages =
 601                                 sbi->ll_ra_info.ra_max_pages_per_file;
 602         }
 603
 604         err = obd_fid_init(sbi->ll_dt_exp->exp_obd, sbi->ll_dt_exp,
 605                            LUSTRE_SEQ_METADATA);
 606         if (err) {
 607                 CERROR("%s: Can't init data layer FID infrastructure, "
 608                        "rc = %d\n", sbi->ll_dt_exp->exp_obd->obd_name, err);
 609                 GOTO(out_dt, err);
 610         }
 611
 612         mutex_lock(&sbi->ll_lco.lco_lock);
 613         sbi->ll_lco.lco_flags = data->ocd_connect_flags;
 614         sbi->ll_lco.lco_md_exp = sbi->ll_md_exp;
 615         sbi->ll_lco.lco_dt_exp = sbi->ll_dt_exp;
 616         mutex_unlock(&sbi->ll_lco.lco_lock);
 617
 618         fid_zero(&sbi->ll_root_fid);
 619         err = md_get_root(sbi->ll_md_exp, get_mount_fileset(sb),
 620                            &sbi->ll_root_fid);
 621         if (err) {
 622                 CERROR("cannot mds_connect: rc = %d\n", err);
 623                 GOTO(out_lock_cn_cb, err);
 624         }
 625         if (!fid_is_sane(&sbi->ll_root_fid)) {
 626                 CERROR("%s: Invalid root fid "DFID" during mount\n",
 627                        sbi->ll_md_exp->exp_obd->obd_name,
 628                        PFID(&sbi->ll_root_fid));
 629                 GOTO(out_lock_cn_cb, err = -EINVAL);
 630         }
 631         CDEBUG(D_SUPER, "rootfid "DFID"\n", PFID(&sbi->ll_root_fid));
 632
 633         sb->s_op = &lustre_super_operations;
 634         sb->s_xattr = ll_xattr_handlers;
 635 #if THREAD_SIZE >= 8192 /*b=17630*/
 636         sb->s_export_op = &lustre_export_operations;
 637 #endif
 638 #ifdef HAVE_LUSTRE_CRYPTO
 639         llcrypt_set_ops(sb, &lustre_cryptops);
 640 #endif
 641
 642         /* make root inode
 643          * XXX: move this to after cbd setup? */
 644         valid = OBD_MD_FLGETATTR | OBD_MD_FLBLOCKS | OBD_MD_FLMODEASIZE;
 645         if (test_bit(LL_SBI_ACL, sbi->ll_flags))
 646                 valid |= OBD_MD_FLACL;
 647
 648         OBD_ALLOC_PTR(op_data);
 649         if (op_data == NULL)
 650                 GOTO(out_lock_cn_cb, err = -ENOMEM);
 651
 652         op_data->op_fid1 = sbi->ll_root_fid;
 653         op_data->op_mode = 0;
 654         op_data->op_valid = valid;
 655
 656         err = md_getattr(sbi->ll_md_exp, op_data, &request);
 657
 658         OBD_FREE_PTR(op_data);
 659         if (err) {
 660                 CERROR("%s: md_getattr failed for root: rc = %d\n",
 661                        sbi->ll_md_exp->exp_obd->obd_name, err);
 662                 GOTO(out_lock_cn_cb, err);
 663         }
 664
 665         err = md_get_lustre_md(sbi->ll_md_exp, &request->rq_pill,
 666                                sbi->ll_dt_exp, sbi->ll_md_exp, &lmd);
 667         if (err) {
 668                 CERROR("failed to understand root inode md: rc = %d\n", err);
 669                 ptlrpc_req_finished(request);
 670                 GOTO(out_lock_cn_cb, err);
 671         }
 672
 673         LASSERT(fid_is_sane(&sbi->ll_root_fid));
 674         api32 = test_bit(LL_SBI_32BIT_API, sbi->ll_flags);
 675         root = ll_iget(sb, cl_fid_build_ino(&sbi->ll_root_fid, api32), &lmd);
 676         md_free_lustre_md(sbi->ll_md_exp, &lmd);
 677         ptlrpc_req_finished(request);
 678
 679         if (IS_ERR(root)) {
 680                 lmd_clear_acl(&lmd);
 681                 err = IS_ERR(root) ? PTR_ERR(root) : -EBADF;
 682                 root = NULL;
 683                 CERROR("%s: bad ll_iget() for root: rc = %d\n",
 684                        sbi->ll_fsname, err);
 685                 GOTO(out_root, err);
 686         }
 687
 688         checksum = test_bit(LL_SBI_CHECKSUM, sbi->ll_flags);
 689         if (sbi->ll_checksum_set) {
 690                 err = obd_set_info_async(NULL, sbi->ll_dt_exp,
 691                                          sizeof(KEY_CHECKSUM), KEY_CHECKSUM,
 692                                          sizeof(checksum), &checksum, NULL);
 693                 if (err) {
 694                         CERROR("%s: Set checksum failed: rc = %d\n",
 695                                sbi->ll_dt_exp->exp_obd->obd_name, err);
 696                         GOTO(out_root, err);
 697                 }
 698         }
 699         cl_sb_init(sb);
 700
 701         sb->s_root = d_make_root(root);
 702         if (sb->s_root == NULL) {
 703                 err = -ENOMEM;
 704                 CERROR("%s: can't make root dentry: rc = %d\n",
 705                        sbi->ll_fsname, err);
 706                 GOTO(out_root, err);
 707         }
 708
 709         sbi->ll_sdev_orig = sb->s_dev;
 710
 711         /* We set sb->s_dev equal on all lustre clients in order to support
 712          * NFS export clustering.  NFSD requires that the FSID be the same
 713          * on all clients. */
 714         /* s_dev is also used in lt_compare() to compare two fs, but that is
 715          * only a node-local comparison. */
 716         uuid = obd_get_uuid(sbi->ll_md_exp);
 717         if (uuid != NULL)
 718                 sb->s_dev = get_uuid2int(uuid->uuid, strlen(uuid->uuid));
 719
 720         if (data != NULL)
 721                 OBD_FREE_PTR(data);
 722         if (osfs != NULL)
 723                 OBD_FREE_PTR(osfs);
 724
 725         if (sbi->ll_dt_obd) {
 726                 err = sysfs_create_link(&sbi->ll_kset.kobj,
 727                                         &sbi->ll_dt_obd->obd_kset.kobj,
 728                                         sbi->ll_dt_obd->obd_type->typ_name);
 729                 if (err < 0) {
 730                         CERROR("%s: could not register %s in llite: rc = %d\n",
 731                                dt, sbi->ll_fsname, err);
 732                         err = 0;
 733                 }
 734         }
 735
 736         if (sbi->ll_md_obd) {
 737                 err = sysfs_create_link(&sbi->ll_kset.kobj,
 738                                         &sbi->ll_md_obd->obd_kset.kobj,
 739                                         sbi->ll_md_obd->obd_type->typ_name);
 740                 if (err < 0) {
 741                         CERROR("%s: could not register %s in llite: rc = %d\n",
 742                                md, sbi->ll_fsname, err);
 743                         err = 0;
 744                 }
 745         }
 746
 747         RETURN(err);
 748 out_root:
 749         iput(root);
 750 out_lock_cn_cb:
 751         obd_fid_fini(sbi->ll_dt_exp->exp_obd);
 752 out_dt:
 753         obd_disconnect(sbi->ll_dt_exp);
 754         sbi->ll_dt_exp = NULL;
 755         sbi->ll_dt_obd = NULL;
 756 out_md_fid:
 757         obd_fid_fini(sbi->ll_md_exp->exp_obd);
 758 out_md:
 759         obd_disconnect(sbi->ll_md_exp);
 760         sbi->ll_md_exp = NULL;
 761         sbi->ll_md_obd = NULL;
 762 out:
 763         if (data != NULL)
 764                 OBD_FREE_PTR(data);
 765         if (osfs != NULL)
 766                 OBD_FREE_PTR(osfs);
 767         return err;
 768 }
 769
 770 int ll_get_max_mdsize(struct ll_sb_info *sbi, int *lmmsize)
 771 {
 772         int size, rc;
 773
 774         size = sizeof(*lmmsize);
 775         rc = obd_get_info(NULL, sbi->ll_dt_exp, sizeof(KEY_MAX_EASIZE),
 776                           KEY_MAX_EASIZE, &size, lmmsize);
 777         if (rc != 0) {
 778                 CERROR("%s: cannot get max LOV EA size: rc = %d\n",
 779                        sbi->ll_dt_exp->exp_obd->obd_name, rc);
 780                 RETURN(rc);
 781         }
 782
 783         CDEBUG(D_INFO, "max LOV ea size: %d\n", *lmmsize);
 784
 785         size = sizeof(int);
 786         rc = obd_get_info(NULL, sbi->ll_md_exp, sizeof(KEY_MAX_EASIZE),
 787                           KEY_MAX_EASIZE, &size, lmmsize);
 788         if (rc)
 789                 CERROR("Get max mdsize error rc %d\n", rc);
 790
 791         CDEBUG(D_INFO, "max LMV ea size: %d\n", *lmmsize);
 792
 793         RETURN(rc);
 794 }
 795
 796 /**
 797  * Get the value of the default_easize parameter.
 798  *
 799  * \see client_obd::cl_default_mds_easize
 800  *
 801  * \param[in] sbi       superblock info for this filesystem
 802  * \param[out] lmmsize  pointer to storage location for value
 803  *
 804  * \retval 0            on success
 805  * \retval negative     negated errno on failure
 806  */
 807 int ll_get_default_mdsize(struct ll_sb_info *sbi, int *lmmsize)
 808 {
 809         int size, rc;
 810
 811         size = sizeof(int);
 812         rc = obd_get_info(NULL, sbi->ll_md_exp, sizeof(KEY_DEFAULT_EASIZE),
 813                          KEY_DEFAULT_EASIZE, &size, lmmsize);
 814         if (rc)
 815                 CERROR("Get default mdsize error rc %d\n", rc);
 816
 817         RETURN(rc);
 818 }
 819
 820 /**
 821  * Set the default_easize parameter to the given value.
 822  *
 823  * \see client_obd::cl_default_mds_easize
 824  *
 825  * \param[in] sbi       superblock info for this filesystem
 826  * \param[in] lmmsize   the size to set
 827  *
 828  * \retval 0            on success
 829  * \retval negative     negated errno on failure
 830  */
 831 int ll_set_default_mdsize(struct ll_sb_info *sbi, int lmmsize)
 832 {
 833         int rc;
 834
 835         if (lmmsize < sizeof(struct lov_mds_md) ||
 836             lmmsize > OBD_MAX_DEFAULT_EA_SIZE)
 837                 return -EINVAL;
 838
 839         rc = obd_set_info_async(NULL, sbi->ll_md_exp,
 840                                 sizeof(KEY_DEFAULT_EASIZE), KEY_DEFAULT_EASIZE,
 841                                 sizeof(int), &lmmsize, NULL);
 842
 843         RETURN(rc);
 844 }
 845
 846 static void client_common_put_super(struct super_block *sb)
 847 {
 848         struct ll_sb_info *sbi = ll_s2sbi(sb);
 849         ENTRY;
 850
 851         cl_sb_fini(sb);
 852
 853         obd_fid_fini(sbi->ll_dt_exp->exp_obd);
 854         obd_disconnect(sbi->ll_dt_exp);
 855         sbi->ll_dt_exp = NULL;
 856
 857         ll_debugfs_unregister_super(sb);
 858
 859         obd_fid_fini(sbi->ll_md_exp->exp_obd);
 860         obd_disconnect(sbi->ll_md_exp);
 861         sbi->ll_md_exp = NULL;
 862
 863         EXIT;
 864 }
 865
 866 void ll_kill_super(struct super_block *sb)
 867 {
 868         struct ll_sb_info *sbi;
 869         ENTRY;
 870
 871         /* not init sb ?*/
 872         if (!(sb->s_flags & SB_ACTIVE))
 873                 return;
 874
 875         sbi = ll_s2sbi(sb);
 876         /* we need restore s_dev from changed for clustred NFS before put_super
 877          * because new kernels have cached s_dev and change sb->s_dev in
 878          * put_super not affected real removing devices */
 879         if (sbi) {
 880                 sb->s_dev = sbi->ll_sdev_orig;
 881
 882                 /* wait running statahead threads to quit */
 883                 while (atomic_read(&sbi->ll_sa_running) > 0)
 884                         schedule_timeout_uninterruptible(
 885                                 cfs_time_seconds(1) >> 3);
 886         }
 887
 888         EXIT;
 889 }
 890
 891 /* Since we use this table for ll_sbi_flags_seq_show make
 892  * sure what you want displayed for a specific token that
 893  * is listed more than once below be listed first. For
 894  * example we want "checksum" displayed, not "nochecksum"
 895  * for the sbi_flags.
 896  */
 897 static const match_table_t ll_sbi_flags_name = {
 898         {LL_SBI_NOLCK,                  "nolock"},
 899         {LL_SBI_CHECKSUM,               "checksum"},
 900         {LL_SBI_CHECKSUM,               "nochecksum"},
 901         {LL_SBI_LOCALFLOCK,             "localflock"},
 902         {LL_SBI_FLOCK,                  "flock"},
 903         {LL_SBI_FLOCK,                  "noflock"},
 904         {LL_SBI_USER_XATTR,             "user_xattr"},
 905         {LL_SBI_USER_XATTR,             "nouser_xattr"},
 906         {LL_SBI_LRU_RESIZE,             "lruresize"},
 907         {LL_SBI_LRU_RESIZE,             "nolruresize"},
 908         {LL_SBI_LAZYSTATFS,             "lazystatfs"},
 909         {LL_SBI_LAZYSTATFS,             "nolazystatfs"},
 910         {LL_SBI_32BIT_API,              "32bitapi"},
 911         {LL_SBI_USER_FID2PATH,          "user_fid2path"},
 912         {LL_SBI_USER_FID2PATH,          "nouser_fid2path"},
 913         {LL_SBI_VERBOSE,                "verbose"},
 914         {LL_SBI_VERBOSE,                "noverbose"},
 915         {LL_SBI_ALWAYS_PING,            "always_ping"},
 916         {LL_SBI_TEST_DUMMY_ENCRYPTION,  "test_dummy_encryption"},
 917         {LL_SBI_ENCRYPT,                "encrypt"},
 918         {LL_SBI_ENCRYPT,                "noencrypt"},
 919         {LL_SBI_FOREIGN_SYMLINK,        "foreign_symlink=%s"},
 920         {LL_SBI_NUM_MOUNT_OPT,          NULL},
 921
 922         {LL_SBI_ACL,                    "acl"},
 923         {LL_SBI_AGL_ENABLED,            "agl"},
 924         {LL_SBI_64BIT_HASH,             "64bit_hash"},
 925         {LL_SBI_LAYOUT_LOCK,            "layout"},
 926         {LL_SBI_XATTR_CACHE,            "xattr_cache"},
 927         {LL_SBI_NOROOTSQUASH,           "norootsquash"},
 928         {LL_SBI_FAST_READ,              "fast_read"},
 929         {LL_SBI_FILE_SECCTX,            "file_secctx"},
 930         {LL_SBI_TINY_WRITE,             "tiny_write"},
 931         {LL_SBI_FILE_HEAT,              "file_heat"},
 932         {LL_SBI_PARALLEL_DIO,           "parallel_dio"},
 933 };
 934
 935 int ll_sbi_flags_seq_show(struct seq_file *m, void *v)
 936 {
 937         struct super_block *sb = m->private;
 938         int i;
 939
 940         for (i = 0; i < LL_SBI_NUM_FLAGS; i++) {
 941                 int j;
 942
 943                 if (!test_bit(i, ll_s2sbi(sb)->ll_flags))
 944                         continue;
 945
 946                 for (j = 0; j < ARRAY_SIZE(ll_sbi_flags_name); j++) {
 947                         if (ll_sbi_flags_name[j].token == i &&
 948                             ll_sbi_flags_name[j].pattern) {
 949                                 seq_printf(m, "%s ",
 950                                            ll_sbi_flags_name[j].pattern);
 951                                 break;
 952                         }
 953                 }
 954         }
 955         seq_puts(m, "\b\n");
 956         return 0;
 957 }
 958
 959 /* non-client-specific mount options are parsed in lmd_parse */
 960 static int ll_options(char *options, struct super_block *sb)
 961 {
 962         struct ll_sb_info *sbi = ll_s2sbi(sb);
 963         char *s2, *s1, *opts;
 964
 965         ENTRY;
 966         if (!options)
 967                 RETURN(0);
 968
 969         /* Don't stomp on lmd_opts */
 970         opts = kstrdup(options, GFP_KERNEL);
 971         if (!opts)
 972                 RETURN(-ENOMEM);
 973         s1 = opts;
 974         s2 = opts;
 975
 976         CDEBUG(D_CONFIG, "Parsing opts %s\n", options);
 977
 978         while ((s1 = strsep(&opts, ",")) != NULL) {
 979                 substring_t args[MAX_OPT_ARGS];
 980                 bool turn_off = false;
 981                 int token;
 982
 983                 if (!*s1)
 984                         continue;
 985
 986                 CDEBUG(D_SUPER, "next opt=%s\n", s1);
 987
 988                 if (strncmp(s1, "no", 2) == 0)
 989                         turn_off = true;
 990
 991                 /*
 992                  * Initialize args struct so we know whether arg was
 993                  * found; some options take optional arguments.
 994                  */
 995                 args[0].to = NULL;
 996                 args[0].from = NULL;
 997                 token = match_token(s1, ll_sbi_flags_name, args);
 998                 if (token == LL_SBI_NUM_MOUNT_OPT) {
 999                         if (match_wildcard("context", s1) ||
1000                             match_wildcard("fscontext", s1) ||
1001                             match_wildcard("defcontext", s1) ||
1002                             match_wildcard("rootcontext",s1))
1003                                 continue;
1004
1005                         LCONSOLE_ERROR_MSG(0x152,
1006                                            "Unknown option '%s', won't mount.\n",
1007                                            s1);
1008                         RETURN(-EINVAL);
1009                 }
1010
1011                 switch (token) {
1012                 case LL_SBI_NOLCK:
1013                 case LL_SBI_32BIT_API:
1014                 case LL_SBI_64BIT_HASH:
1015                 case LL_SBI_ALWAYS_PING:
1016                         set_bit(token, sbi->ll_flags);
1017                         break;
1018
1019                 case LL_SBI_FLOCK:
1020                         clear_bit(LL_SBI_LOCALFLOCK, sbi->ll_flags);
1021                         if (turn_off)
1022                                 clear_bit(LL_SBI_FLOCK, sbi->ll_flags);
1023                         else
1024                                 set_bit(token, sbi->ll_flags);
1025                         break;
1026
1027                 case LL_SBI_LOCALFLOCK:
1028                         clear_bit(LL_SBI_FLOCK, sbi->ll_flags);
1029                         set_bit(token, sbi->ll_flags);
1030                         break;
1031
1032                 case LL_SBI_CHECKSUM:
1033                         sbi->ll_checksum_set = 1;
1034                         /* fall through */
1035                 case LL_SBI_USER_XATTR:
1036                 case LL_SBI_USER_FID2PATH:
1037                 case LL_SBI_LRU_RESIZE:
1038                 case LL_SBI_LAZYSTATFS:
1039                 case LL_SBI_VERBOSE:
1040                         if (turn_off)
1041                                 clear_bit(token, sbi->ll_flags);
1042                         else
1043                                 set_bit(token, sbi->ll_flags);
1044                         break;
1045                 case LL_SBI_TEST_DUMMY_ENCRYPTION: {
1046 #ifdef HAVE_LUSTRE_CRYPTO
1047                         set_bit(token, sbi->ll_flags);
1048 #else
1049                         LCONSOLE_WARN("Test dummy encryption mount option ignored: encryption not supported\n");
1050 #endif
1051                         break;
1052                 }
1053                 case LL_SBI_ENCRYPT:
1054 #ifdef HAVE_LUSTRE_CRYPTO
1055                         if (turn_off)
1056                                 clear_bit(token, sbi->ll_flags);
1057                         else
1058                                 set_bit(token, sbi->ll_flags);
1059 #else
1060                         LCONSOLE_WARN("noencrypt or encrypt mount option ignored: encryption not supported\n");
1061 #endif
1062                         break;
1063                 case LL_SBI_FOREIGN_SYMLINK:
1064                         /* non-default prefix provided ? */
1065                         if (args->from) {
1066                                 size_t old_len;
1067                                 char *old;
1068
1069                                 /* path must be absolute */
1070                                 if (args->from[0] != '/') {
1071                                         LCONSOLE_ERROR_MSG(0x152,
1072                                                            "foreign prefix '%s' must be an absolute path\n",
1073                                                            args->from);
1074                                         RETURN(-EINVAL);
1075                                 }
1076
1077                                 old_len = sbi->ll_foreign_symlink_prefix_size;
1078                                 old = sbi->ll_foreign_symlink_prefix;
1079                                 /* alloc for path length and '\0' */
1080                                 sbi->ll_foreign_symlink_prefix = match_strdup(args);
1081                                 if (!sbi->ll_foreign_symlink_prefix) {
1082                                         /* restore previous */
1083                                         sbi->ll_foreign_symlink_prefix = old;
1084                                         sbi->ll_foreign_symlink_prefix_size =
1085                                                 old_len;
1086                                         RETURN(-ENOMEM);
1087                                 }
1088                                 sbi->ll_foreign_symlink_prefix_size =
1089                                         args->to - args->from + 1;
1090                                 OBD_ALLOC_POST(sbi->ll_foreign_symlink_prefix,
1091                                                sbi->ll_foreign_symlink_prefix_size,
1092                                                "kmalloced");
1093                                 if (old)
1094                                         OBD_FREE(old, old_len);
1095
1096                                 /* enable foreign symlink support */
1097                                 set_bit(token, sbi->ll_flags);
1098                         } else {
1099                                 LCONSOLE_ERROR_MSG(0x152,
1100                                                    "invalid %s option\n", s1);
1101                         }
1102                 /* fall through */
1103                 default:
1104                         break;
1105                 }
1106         }
1107         kfree(opts);
1108         RETURN(0);
1109 }
1110
1111 void ll_lli_init(struct ll_inode_info *lli)
1112 {
1113         lli->lli_inode_magic = LLI_INODE_MAGIC;
1114         lli->lli_flags = 0;
1115         rwlock_init(&lli->lli_lock);
1116         lli->lli_posix_acl = NULL;
1117         /* Do not set lli_fid, it has been initialized already. */
1118         fid_zero(&lli->lli_pfid);
1119         lli->lli_mds_read_och = NULL;
1120         lli->lli_mds_write_och = NULL;
1121         lli->lli_mds_exec_och = NULL;
1122         lli->lli_open_fd_read_count = 0;
1123         lli->lli_open_fd_write_count = 0;
1124         lli->lli_open_fd_exec_count = 0;
1125         mutex_init(&lli->lli_och_mutex);
1126         spin_lock_init(&lli->lli_agl_lock);
1127         spin_lock_init(&lli->lli_layout_lock);
1128         ll_layout_version_set(lli, CL_LAYOUT_GEN_NONE);
1129         lli->lli_clob = NULL;
1130
1131         init_rwsem(&lli->lli_xattrs_list_rwsem);
1132         mutex_init(&lli->lli_xattrs_enq_lock);
1133
1134         LASSERT(lli->lli_vfs_inode.i_mode != 0);
1135         if (S_ISDIR(lli->lli_vfs_inode.i_mode)) {
1136                 lli->lli_opendir_key = NULL;
1137                 lli->lli_sai = NULL;
1138                 spin_lock_init(&lli->lli_sa_lock);
1139                 lli->lli_opendir_pid = 0;
1140                 lli->lli_sa_enabled = 0;
1141                 init_rwsem(&lli->lli_lsm_sem);
1142         } else {
1143                 mutex_init(&lli->lli_size_mutex);
1144                 mutex_init(&lli->lli_setattr_mutex);
1145                 lli->lli_symlink_name = NULL;
1146                 ll_trunc_sem_init(&lli->lli_trunc_sem);
1147                 range_lock_tree_init(&lli->lli_write_tree);
1148                 init_rwsem(&lli->lli_glimpse_sem);
1149                 lli->lli_glimpse_time = ktime_set(0, 0);
1150                 INIT_LIST_HEAD(&lli->lli_agl_list);
1151                 lli->lli_agl_index = 0;
1152                 lli->lli_async_rc = 0;
1153                 spin_lock_init(&lli->lli_heat_lock);
1154                 obd_heat_clear(lli->lli_heat_instances, OBD_HEAT_COUNT);
1155                 lli->lli_heat_flags = 0;
1156                 mutex_init(&lli->lli_pcc_lock);
1157                 lli->lli_pcc_state = PCC_STATE_FL_NONE;
1158                 lli->lli_pcc_inode = NULL;
1159                 lli->lli_pcc_dsflags = PCC_DATASET_INVALID;
1160                 lli->lli_pcc_generation = 0;
1161                 mutex_init(&lli->lli_group_mutex);
1162                 lli->lli_group_users = 0;
1163                 lli->lli_group_gid = 0;
1164         }
1165         mutex_init(&lli->lli_layout_mutex);
1166         memset(lli->lli_jobid, 0, sizeof(lli->lli_jobid));
1167         /* ll_cl_context initialize */
1168         INIT_LIST_HEAD(&lli->lli_lccs);
1169 }
1170
1171 #define MAX_STRING_SIZE 128
1172
1173 #ifndef HAVE_SUPER_SETUP_BDI_NAME
1174
1175 #define LSI_BDI_INITIALIZED     0x00400000
1176
1177 #ifndef HAVE_BDI_CAP_MAP_COPY
1178 # define BDI_CAP_MAP_COPY       0
1179 #endif
1180
1181 static int super_setup_bdi_name(struct super_block *sb, char *fmt, ...)
1182 {
1183         struct  lustre_sb_info *lsi = s2lsi(sb);
1184         char buf[MAX_STRING_SIZE];
1185         va_list args;
1186         int err;
1187
1188         err = bdi_init(&lsi->lsi_bdi);
1189         if (err)
1190                 return err;
1191
1192         lsi->lsi_flags |= LSI_BDI_INITIALIZED;
1193         lsi->lsi_bdi.capabilities = BDI_CAP_MAP_COPY;
1194         lsi->lsi_bdi.name = "lustre";
1195         va_start(args, fmt);
1196         vsnprintf(buf, MAX_STRING_SIZE, fmt, args);
1197         va_end(args);
1198         err = bdi_register(&lsi->lsi_bdi, NULL, "%s", buf);
1199         va_end(args);
1200         if (!err)
1201                 sb->s_bdi = &lsi->lsi_bdi;
1202
1203         return err;
1204 }
1205 #endif /* !HAVE_SUPER_SETUP_BDI_NAME */
1206
1207 int ll_fill_super(struct super_block *sb)
1208 {
1209         struct  lustre_profile *lprof = NULL;
1210         struct  lustre_sb_info *lsi = s2lsi(sb);
1211         struct  ll_sb_info *sbi = NULL;
1212         char    *dt = NULL, *md = NULL;
1213         char    *profilenm = get_profile_name(sb);
1214         struct config_llog_instance *cfg;
1215         /* %p for void* in printf needs 16+2 characters: 0xffffffffffffffff */
1216         const int instlen = LUSTRE_MAXINSTANCE + 2;
1217         unsigned long cfg_instance = ll_get_cfg_instance(sb);
1218         char name[MAX_STRING_SIZE];
1219         int md_len = 0;
1220         int dt_len = 0;
1221         uuid_t uuid;
1222         char *ptr;
1223         int len;
1224         int err;
1225
1226         ENTRY;
1227         /* for ASLR, to map between cfg_instance and hashed ptr */
1228         CDEBUG(D_VFSTRACE, "VFS Op: cfg_instance %s-%016lx (sb %p)\n",
1229                profilenm, cfg_instance, sb);
1230
1231         OBD_RACE(OBD_FAIL_LLITE_RACE_MOUNT);
1232
1233         OBD_ALLOC_PTR(cfg);
1234         if (cfg == NULL)
1235                 GOTO(out_free_cfg, err = -ENOMEM);
1236
1237         /* client additional sb info */
1238         lsi->lsi_llsbi = sbi = ll_init_sbi();
1239         if (IS_ERR(sbi))
1240                 GOTO(out_free_cfg, err = PTR_ERR(sbi));
1241
1242         err = ll_options(lsi->lsi_lmd->lmd_opts, sb);
1243         if (err)
1244                 GOTO(out_free_cfg, err);
1245
1246         /* kernel >= 2.6.38 store dentry operations in sb->s_d_op. */
1247         sb->s_d_op = &ll_d_ops;
1248
1249         /* UUID handling */
1250         generate_random_uuid(uuid.b);
1251         snprintf(sbi->ll_sb_uuid.uuid, sizeof(sbi->ll_sb_uuid), "%pU", uuid.b);
1252
1253         CDEBUG(D_CONFIG, "llite sb uuid: %s\n", sbi->ll_sb_uuid.uuid);
1254
1255         /* Get fsname */
1256         len = strlen(profilenm);
1257         ptr = strrchr(profilenm, '-');
1258         if (ptr && (strcmp(ptr, "-client") == 0))
1259                 len -= 7;
1260
1261         if (len > LUSTRE_MAXFSNAME) {
1262                 if (unlikely(len >= MAX_STRING_SIZE))
1263                         len = MAX_STRING_SIZE - 1;
1264                 strncpy(name, profilenm, len);
1265                 name[len] = '\0';
1266                 err = -ENAMETOOLONG;
1267                 CERROR("%s: fsname longer than %u characters: rc = %d\n",
1268                        name, LUSTRE_MAXFSNAME, err);
1269                 GOTO(out_free_cfg, err);
1270         }
1271         strncpy(sbi->ll_fsname, profilenm, len);
1272         sbi->ll_fsname[len] = '\0';
1273
1274         /* Mount info */
1275         snprintf(name, sizeof(name), "%.*s-%016lx", len,
1276                  profilenm, cfg_instance);
1277
1278         err = super_setup_bdi_name(sb, "%s", name);
1279         if (err)
1280                 GOTO(out_free_cfg, err);
1281
1282         /* disable kernel readahead */
1283         sb->s_bdi->ra_pages = 0;
1284
1285         /* Call ll_debugfs_register_super() before lustre_process_log()
1286          * so that "llite.*.*" params can be processed correctly.
1287          */
1288         err = ll_debugfs_register_super(sb, name);
1289         if (err < 0) {
1290                 CERROR("%s: could not register mountpoint in llite: rc = %d\n",
1291                        sbi->ll_fsname, err);
1292                 err = 0;
1293         }
1294
1295         /* The cfg_instance is a value unique to this super, in case some
1296          * joker tries to mount the same fs at two mount points.
1297          */
1298         cfg->cfg_instance = cfg_instance;
1299         cfg->cfg_uuid = lsi->lsi_llsbi->ll_sb_uuid;
1300         cfg->cfg_callback = class_config_llog_handler;
1301         cfg->cfg_sub_clds = CONFIG_SUB_CLIENT;
1302         /* set up client obds */
1303         err = lustre_process_log(sb, profilenm, cfg);
1304         if (err < 0)
1305                 GOTO(out_debugfs, err);
1306
1307         /* Profile set with LCFG_MOUNTOPT so we can find our mdc and osc obds */
1308         lprof = class_get_profile(profilenm);
1309         if (lprof == NULL) {
1310                 LCONSOLE_ERROR_MSG(0x156, "The client profile '%s' could not be"
1311                                    " read from the MGS.  Does that filesystem "
1312                                    "exist?\n", profilenm);
1313                 GOTO(out_debugfs, err = -EINVAL);
1314         }
1315         CDEBUG(D_CONFIG, "Found profile %s: mdc=%s osc=%s\n", profilenm,
1316                lprof->lp_md, lprof->lp_dt);
1317
1318         dt_len = strlen(lprof->lp_dt) + instlen + 2;
1319         OBD_ALLOC(dt, dt_len);
1320         if (!dt)
1321                 GOTO(out_profile, err = -ENOMEM);
1322         snprintf(dt, dt_len - 1, "%s-%016lx", lprof->lp_dt, cfg_instance);
1323
1324         md_len = strlen(lprof->lp_md) + instlen + 2;
1325         OBD_ALLOC(md, md_len);
1326         if (!md)
1327                 GOTO(out_free_dt, err = -ENOMEM);
1328         snprintf(md, md_len - 1, "%s-%016lx", lprof->lp_md, cfg_instance);
1329
1330         /* connections, registrations, sb setup */
1331         err = client_common_fill_super(sb, md, dt);
1332         if (err < 0)
1333                 GOTO(out_free_md, err);
1334
1335         sbi->ll_client_common_fill_super_succeeded = 1;
1336
1337 out_free_md:
1338         if (md)
1339                 OBD_FREE(md, md_len);
1340 out_free_dt:
1341         if (dt)
1342                 OBD_FREE(dt, dt_len);
1343 out_profile:
1344         if (lprof)
1345                 class_put_profile(lprof);
1346 out_debugfs:
1347         if (err < 0)
1348                 ll_debugfs_unregister_super(sb);
1349 out_free_cfg:
1350         if (cfg)
1351                 OBD_FREE_PTR(cfg);
1352
1353         if (err)
1354                 ll_put_super(sb);
1355         else if (test_bit(LL_SBI_VERBOSE, sbi->ll_flags))
1356                 LCONSOLE_WARN("Mounted %s\n", profilenm);
1357         RETURN(err);
1358 } /* ll_fill_super */
1359
1360 void ll_put_super(struct super_block *sb)
1361 {
1362         struct config_llog_instance cfg, params_cfg;
1363         struct obd_device *obd;
1364         struct lustre_sb_info *lsi = s2lsi(sb);
1365         struct ll_sb_info *sbi = ll_s2sbi(sb);
1366         char *profilenm = get_profile_name(sb);
1367         unsigned long cfg_instance = ll_get_cfg_instance(sb);
1368         long ccc_count;
1369         int next, force = 1, rc = 0;
1370         ENTRY;
1371
1372         if (IS_ERR(sbi))
1373                 GOTO(out_no_sbi, 0);
1374
1375         /* Should replace instance_id with something better for ASLR */
1376         CDEBUG(D_VFSTRACE, "VFS Op: cfg_instance %s-%016lx (sb %p)\n",
1377                profilenm, cfg_instance, sb);
1378
1379         cfg.cfg_instance = cfg_instance;
1380         lustre_end_log(sb, profilenm, &cfg);
1381
1382         params_cfg.cfg_instance = cfg_instance;
1383         lustre_end_log(sb, PARAMS_FILENAME, &params_cfg);
1384
1385         if (sbi->ll_md_exp) {
1386                 obd = class_exp2obd(sbi->ll_md_exp);
1387                 if (obd)
1388                         force = obd->obd_force;
1389         }
1390
1391         /* Wait for unstable pages to be committed to stable storage */
1392         if (force == 0) {
1393                 rc = l_wait_event_abortable(
1394                         sbi->ll_cache->ccc_unstable_waitq,
1395                         atomic_long_read(&sbi->ll_cache->ccc_unstable_nr) == 0);
1396         }
1397
1398         ccc_count = atomic_long_read(&sbi->ll_cache->ccc_unstable_nr);
1399         if (force == 0 && rc != -ERESTARTSYS)
1400                 LASSERTF(ccc_count == 0, "count: %li\n", ccc_count);
1401
1402         /* We need to set force before the lov_disconnect in
1403          * lustre_common_put_super, since l_d cleans up osc's as well.
1404          */
1405         if (force) {
1406                 next = 0;
1407                 while ((obd = class_devices_in_group(&sbi->ll_sb_uuid,
1408                                                      &next)) != NULL) {
1409                         obd->obd_force = force;
1410                 }
1411         }
1412
1413         if (sbi->ll_client_common_fill_super_succeeded) {
1414                 /* Only if client_common_fill_super succeeded */
1415                 client_common_put_super(sb);
1416         }
1417
1418         next = 0;
1419         while ((obd = class_devices_in_group(&sbi->ll_sb_uuid, &next)))
1420                 class_manual_cleanup(obd);
1421
1422         if (test_bit(LL_SBI_VERBOSE, sbi->ll_flags))
1423                 LCONSOLE_WARN("Unmounted %s\n", profilenm ? profilenm : "");
1424
1425         if (profilenm)
1426                 class_del_profile(profilenm);
1427
1428 #ifndef HAVE_SUPER_SETUP_BDI_NAME
1429         if (lsi->lsi_flags & LSI_BDI_INITIALIZED) {
1430                 bdi_destroy(&lsi->lsi_bdi);
1431                 lsi->lsi_flags &= ~LSI_BDI_INITIALIZED;
1432         }
1433 #endif
1434
1435         ll_free_sbi(sb);
1436         lsi->lsi_llsbi = NULL;
1437 out_no_sbi:
1438         lustre_common_put_super(sb);
1439
1440         cl_env_cache_purge(~0);
1441
1442         EXIT;
1443 } /* client_put_super */
1444
1445 struct inode *ll_inode_from_resource_lock(struct ldlm_lock *lock)
1446 {
1447         struct inode *inode = NULL;
1448
1449         /* NOTE: we depend on atomic igrab() -bzzz */
1450         lock_res_and_lock(lock);
1451         if (lock->l_resource->lr_lvb_inode) {
1452                 struct ll_inode_info * lli;
1453                 lli = ll_i2info(lock->l_resource->lr_lvb_inode);
1454                 if (lli->lli_inode_magic == LLI_INODE_MAGIC) {
1455                         inode = igrab(lock->l_resource->lr_lvb_inode);
1456                 } else {
1457                         inode = lock->l_resource->lr_lvb_inode;
1458                         LDLM_DEBUG_LIMIT(inode->i_state & I_FREEING ?  D_INFO :
1459                                          D_WARNING, lock, "lr_lvb_inode %p is "
1460                                          "bogus: magic %08x",
1461                                          lock->l_resource->lr_lvb_inode,
1462                                          lli->lli_inode_magic);
1463                         inode = NULL;
1464                 }
1465         }
1466         unlock_res_and_lock(lock);
1467         return inode;
1468 }
1469
1470 void ll_dir_clear_lsm_md(struct inode *inode)
1471 {
1472         struct ll_inode_info *lli = ll_i2info(inode);
1473
1474         LASSERT(S_ISDIR(inode->i_mode));
1475
1476         if (lli->lli_lsm_md) {
1477                 lmv_free_memmd(lli->lli_lsm_md);
1478                 lli->lli_lsm_md = NULL;
1479         }
1480
1481         if (lli->lli_default_lsm_md) {
1482                 lmv_free_memmd(lli->lli_default_lsm_md);
1483                 lli->lli_default_lsm_md = NULL;
1484         }
1485 }
1486
1487 static struct inode *ll_iget_anon_dir(struct super_block *sb,
1488                                       const struct lu_fid *fid,
1489                                       struct lustre_md *md)
1490 {
1491         struct ll_sb_info *sbi = ll_s2sbi(sb);
1492         struct ll_inode_info *lli;
1493         struct mdt_body *body = md->body;
1494         struct inode *inode;
1495         ino_t ino;
1496
1497         ENTRY;
1498
1499         LASSERT(md->lmv);
1500         ino = cl_fid_build_ino(fid, test_bit(LL_SBI_32BIT_API, sbi->ll_flags));
1501         inode = iget_locked(sb, ino);
1502         if (inode == NULL) {
1503                 CERROR("%s: failed get simple inode "DFID": rc = -ENOENT\n",
1504                        sbi->ll_fsname, PFID(fid));
1505                 RETURN(ERR_PTR(-ENOENT));
1506         }
1507
1508         lli = ll_i2info(inode);
1509         if (inode->i_state & I_NEW) {
1510                 inode->i_mode = (inode->i_mode & ~S_IFMT) |
1511                                 (body->mbo_mode & S_IFMT);
1512                 LASSERTF(S_ISDIR(inode->i_mode), "Not slave inode "DFID"\n",
1513                          PFID(fid));
1514
1515                 inode->i_mtime.tv_sec = 0;
1516                 inode->i_atime.tv_sec = 0;
1517                 inode->i_ctime.tv_sec = 0;
1518                 inode->i_rdev = 0;
1519
1520 #ifdef HAVE_BACKING_DEV_INFO
1521                 /* initializing backing dev info. */
1522                 inode->i_mapping->backing_dev_info =
1523                                                 &s2lsi(inode->i_sb)->lsi_bdi;
1524 #endif
1525                 inode->i_op = &ll_dir_inode_operations;
1526                 inode->i_fop = &ll_dir_operations;
1527                 lli->lli_fid = *fid;
1528                 ll_lli_init(lli);
1529
1530                 /* master object FID */
1531                 lli->lli_pfid = body->mbo_fid1;
1532                 CDEBUG(D_INODE, "lli %p slave "DFID" master "DFID"\n",
1533                        lli, PFID(fid), PFID(&lli->lli_pfid));
1534                 unlock_new_inode(inode);
1535         } else {
1536                 /* in directory restripe/auto-split, a directory will be
1537                  * transformed to a stripe if it's plain, set its pfid here,
1538                  * otherwise ll_lock_cancel_bits() can't find the master inode.
1539                  */
1540                 lli->lli_pfid = body->mbo_fid1;
1541         }
1542
1543         RETURN(inode);
1544 }
1545
1546 static int ll_init_lsm_md(struct inode *inode, struct lustre_md *md)
1547 {
1548         struct lu_fid *fid;
1549         struct lmv_stripe_md *lsm = md->lmv;
1550         struct ll_inode_info *lli = ll_i2info(inode);
1551         int i;
1552
1553         LASSERT(lsm != NULL);
1554
1555         CDEBUG(D_INODE, "%s: "DFID" set dir layout:\n",
1556                ll_i2sbi(inode)->ll_fsname, PFID(&lli->lli_fid));
1557         lsm_md_dump(D_INODE, lsm);
1558
1559         if (!lmv_dir_striped(lsm))
1560                 goto out;
1561
1562         /* XXX sigh, this lsm_root initialization should be in
1563          * LMV layer, but it needs ll_iget right now, so we
1564          * put this here right now. */
1565         for (i = 0; i < lsm->lsm_md_stripe_count; i++) {
1566                 fid = &lsm->lsm_md_oinfo[i].lmo_fid;
1567                 LASSERT(lsm->lsm_md_oinfo[i].lmo_root == NULL);
1568
1569                 if (!fid_is_sane(fid))
1570                         continue;
1571
1572                 /* Unfortunately ll_iget will call ll_update_inode,
1573                  * where the initialization of slave inode is slightly
1574                  * different, so it reset lsm_md to NULL to avoid
1575                  * initializing lsm for slave inode. */
1576                 lsm->lsm_md_oinfo[i].lmo_root =
1577                                 ll_iget_anon_dir(inode->i_sb, fid, md);
1578                 if (IS_ERR(lsm->lsm_md_oinfo[i].lmo_root)) {
1579                         int rc = PTR_ERR(lsm->lsm_md_oinfo[i].lmo_root);
1580
1581                         lsm->lsm_md_oinfo[i].lmo_root = NULL;
1582                         while (i-- > 0) {
1583                                 iput(lsm->lsm_md_oinfo[i].lmo_root);
1584                                 lsm->lsm_md_oinfo[i].lmo_root = NULL;
1585                         }
1586                         return rc;
1587                 }
1588         }
1589 out:
1590         lli->lli_lsm_md = lsm;
1591
1592         return 0;
1593 }
1594
1595 static void ll_update_default_lsm_md(struct inode *inode, struct lustre_md *md)
1596 {
1597         struct ll_inode_info *lli = ll_i2info(inode);
1598
1599         ENTRY;
1600
1601         if (!md->default_lmv) {
1602                 /* clear default lsm */
1603                 if (lli->lli_default_lsm_md) {
1604                         down_write(&lli->lli_lsm_sem);
1605                         if (lli->lli_default_lsm_md) {
1606                                 lmv_free_memmd(lli->lli_default_lsm_md);
1607                                 lli->lli_default_lsm_md = NULL;
1608                         }
1609                         up_write(&lli->lli_lsm_sem);
1610                 }
1611                 RETURN_EXIT;
1612         }
1613
1614         if (lli->lli_default_lsm_md) {
1615                 /* do nonthing if default lsm isn't changed */
1616                 down_read(&lli->lli_lsm_sem);
1617                 if (lli->lli_default_lsm_md &&
1618                     lsm_md_eq(lli->lli_default_lsm_md, md->default_lmv)) {
1619                         up_read(&lli->lli_lsm_sem);
1620                         RETURN_EXIT;
1621                 }
1622                 up_read(&lli->lli_lsm_sem);
1623         }
1624
1625         down_write(&lli->lli_lsm_sem);
1626         if (lli->lli_default_lsm_md)
1627                 lmv_free_memmd(lli->lli_default_lsm_md);
1628         lli->lli_default_lsm_md = md->default_lmv;
1629         lsm_md_dump(D_INODE, md->default_lmv);
1630         md->default_lmv = NULL;
1631         up_write(&lli->lli_lsm_sem);
1632         RETURN_EXIT;
1633 }
1634
1635 static int ll_update_lsm_md(struct inode *inode, struct lustre_md *md)
1636 {
1637         struct ll_inode_info *lli = ll_i2info(inode);
1638         struct lmv_stripe_md *lsm = md->lmv;
1639         struct cl_attr  *attr;
1640         int rc = 0;
1641
1642         ENTRY;
1643
1644         LASSERT(S_ISDIR(inode->i_mode));
1645         CDEBUG(D_INODE, "update lsm %p of "DFID"\n", lli->lli_lsm_md,
1646                PFID(ll_inode2fid(inode)));
1647
1648         /* update default LMV */
1649         if (md->default_lmv)
1650                 ll_update_default_lsm_md(inode, md);
1651
1652         /* after dir migration/restripe, a stripe may be turned into a
1653          * directory, in this case, zero out its lli_pfid.
1654          */
1655         if (unlikely(fid_is_norm(&lli->lli_pfid)))
1656                 fid_zero(&lli->lli_pfid);
1657
1658         /*
1659          * no striped information from request, lustre_md from req does not
1660          * include stripeEA, see ll_md_setattr()
1661          */
1662         if (!lsm)
1663                 RETURN(0);
1664
1665         /*
1666          * normally dir layout doesn't change, only take read lock to check
1667          * that to avoid blocking other MD operations.
1668          */
1669         down_read(&lli->lli_lsm_sem);
1670
1671         /* some current lookup initialized lsm, and unchanged */
1672         if (lli->lli_lsm_md && lsm_md_eq(lli->lli_lsm_md, lsm))
1673                 GOTO(unlock, rc = 0);
1674
1675         /* if dir layout doesn't match, check whether version is increased,
1676          * which means layout is changed, this happens in dir split/merge and
1677          * lfsck.
1678          *
1679          * foreign LMV should not change.
1680          */
1681         if (lli->lli_lsm_md && lmv_dir_striped(lli->lli_lsm_md) &&
1682             lsm->lsm_md_layout_version <=
1683             lli->lli_lsm_md->lsm_md_layout_version) {
1684                 CERROR("%s: "DFID" dir layout mismatch:\n",
1685                        ll_i2sbi(inode)->ll_fsname, PFID(&lli->lli_fid));
1686                 lsm_md_dump(D_ERROR, lli->lli_lsm_md);
1687                 lsm_md_dump(D_ERROR, lsm);
1688                 GOTO(unlock, rc = -EINVAL);
1689         }
1690
1691         up_read(&lli->lli_lsm_sem);
1692         down_write(&lli->lli_lsm_sem);
1693         /* clear existing lsm */
1694         if (lli->lli_lsm_md) {
1695                 lmv_free_memmd(lli->lli_lsm_md);
1696                 lli->lli_lsm_md = NULL;
1697         }
1698
1699         rc = ll_init_lsm_md(inode, md);
1700         up_write(&lli->lli_lsm_sem);
1701
1702         if (rc)
1703                 RETURN(rc);
1704
1705         /* set md->lmv to NULL, so the following free lustre_md will not free
1706          * this lsm.
1707          */
1708         md->lmv = NULL;
1709
1710         /* md_merge_attr() may take long, since lsm is already set, switch to
1711          * read lock.
1712          */
1713         down_read(&lli->lli_lsm_sem);
1714
1715         if (!lmv_dir_striped(lli->lli_lsm_md))
1716                 GOTO(unlock, rc = 0);
1717
1718         OBD_ALLOC_PTR(attr);
1719         if (!attr)
1720                 GOTO(unlock, rc = -ENOMEM);
1721
1722         /* validate the lsm */
1723         rc = md_merge_attr(ll_i2mdexp(inode), lli->lli_lsm_md, attr,
1724                            ll_md_blocking_ast);
1725         if (!rc) {
1726                 if (md->body->mbo_valid & OBD_MD_FLNLINK)
1727                         md->body->mbo_nlink = attr->cat_nlink;
1728                 if (md->body->mbo_valid & OBD_MD_FLSIZE)
1729                         md->body->mbo_size = attr->cat_size;
1730                 if (md->body->mbo_valid & OBD_MD_FLATIME)
1731                         md->body->mbo_atime = attr->cat_atime;
1732                 if (md->body->mbo_valid & OBD_MD_FLCTIME)
1733                         md->body->mbo_ctime = attr->cat_ctime;
1734                 if (md->body->mbo_valid & OBD_MD_FLMTIME)
1735                         md->body->mbo_mtime = attr->cat_mtime;
1736         }
1737
1738         OBD_FREE_PTR(attr);
1739         GOTO(unlock, rc);
1740 unlock:
1741         up_read(&lli->lli_lsm_sem);
1742
1743         return rc;
1744 }
1745
1746 void ll_clear_inode(struct inode *inode)
1747 {
1748         struct ll_inode_info *lli = ll_i2info(inode);
1749         struct ll_sb_info *sbi = ll_i2sbi(inode);
1750
1751         ENTRY;
1752
1753         CDEBUG(D_VFSTRACE, "VFS Op:inode="DFID"(%p)\n",
1754                PFID(ll_inode2fid(inode)), inode);
1755
1756         if (S_ISDIR(inode->i_mode)) {
1757                 /* these should have been cleared in ll_file_release */
1758                 LASSERT(lli->lli_opendir_key == NULL);
1759                 LASSERT(lli->lli_sai == NULL);
1760                 LASSERT(lli->lli_opendir_pid == 0);
1761         } else {
1762                 pcc_inode_free(inode);
1763         }
1764
1765         md_null_inode(sbi->ll_md_exp, ll_inode2fid(inode));
1766
1767         LASSERT(!lli->lli_open_fd_write_count);
1768         LASSERT(!lli->lli_open_fd_read_count);
1769         LASSERT(!lli->lli_open_fd_exec_count);
1770
1771         if (lli->lli_mds_write_och)
1772                 ll_md_real_close(inode, FMODE_WRITE);
1773         if (lli->lli_mds_exec_och)
1774                 ll_md_real_close(inode, FMODE_EXEC);
1775         if (lli->lli_mds_read_och)
1776                 ll_md_real_close(inode, FMODE_READ);
1777
1778         if (S_ISLNK(inode->i_mode) && lli->lli_symlink_name) {
1779                 OBD_FREE(lli->lli_symlink_name,
1780                          strlen(lli->lli_symlink_name) + 1);
1781                 lli->lli_symlink_name = NULL;
1782         }
1783
1784         ll_xattr_cache_destroy(inode);
1785
1786         forget_all_cached_acls(inode);
1787         lli_clear_acl(lli);
1788         lli->lli_inode_magic = LLI_INODE_DEAD;
1789
1790         if (S_ISDIR(inode->i_mode))
1791                 ll_dir_clear_lsm_md(inode);
1792         else if (S_ISREG(inode->i_mode) && !is_bad_inode(inode))
1793                 LASSERT(list_empty(&lli->lli_agl_list));
1794
1795         /*
1796          * XXX This has to be done before lsm is freed below, because
1797          * cl_object still uses inode lsm.
1798          */
1799         cl_inode_fini(inode);
1800
1801         llcrypt_put_encryption_info(inode);
1802
1803         EXIT;
1804 }
1805
1806 static int ll_md_setattr(struct dentry *dentry, struct md_op_data *op_data)
1807 {
1808         struct lustre_md md;
1809         struct inode *inode = dentry->d_inode;
1810         struct ll_sb_info *sbi = ll_i2sbi(inode);
1811         struct ptlrpc_request *request = NULL;
1812         int rc, ia_valid;
1813
1814         ENTRY;
1815
1816         op_data = ll_prep_md_op_data(op_data, inode, NULL, NULL, 0, 0,
1817                                      LUSTRE_OPC_ANY, NULL);
1818         if (IS_ERR(op_data))
1819                 RETURN(PTR_ERR(op_data));
1820
1821         /* If this is a chgrp of a regular file, we want to reserve enough
1822          * quota to cover the entire file size.
1823          */
1824         if (S_ISREG(inode->i_mode) && op_data->op_attr.ia_valid & ATTR_GID &&
1825             from_kgid(&init_user_ns, op_data->op_attr.ia_gid) !=
1826             from_kgid(&init_user_ns, inode->i_gid)) {
1827                 op_data->op_xvalid |= OP_XVALID_BLOCKS;
1828                 op_data->op_attr_blocks = inode->i_blocks;
1829         }
1830
1831
1832         rc = md_setattr(sbi->ll_md_exp, op_data, NULL, 0, &request);
1833         if (rc) {
1834                 ptlrpc_req_finished(request);
1835                 if (rc == -ENOENT) {
1836                         clear_nlink(inode);
1837                         /* Unlinked special device node? Or just a race?
1838                          * Pretend we done everything. */
1839                         if (!S_ISREG(inode->i_mode) &&
1840                             !S_ISDIR(inode->i_mode)) {
1841                                 ia_valid = op_data->op_attr.ia_valid;
1842                                 op_data->op_attr.ia_valid &= ~TIMES_SET_FLAGS;
1843                                 rc = simple_setattr(dentry, &op_data->op_attr);
1844                                 op_data->op_attr.ia_valid = ia_valid;
1845                         }
1846                 } else if (rc != -EPERM && rc != -EACCES && rc != -ETXTBSY) {
1847                         CERROR("md_setattr fails: rc = %d\n", rc);
1848                 }
1849                 RETURN(rc);
1850         }
1851
1852         rc = md_get_lustre_md(sbi->ll_md_exp, &request->rq_pill, sbi->ll_dt_exp,
1853                               sbi->ll_md_exp, &md);
1854         if (rc) {
1855                 ptlrpc_req_finished(request);
1856                 RETURN(rc);
1857         }
1858
1859         ia_valid = op_data->op_attr.ia_valid;
1860         /* inode size will be in ll_setattr_ost, can't do it now since dirty
1861          * cache is not cleared yet. */
1862         op_data->op_attr.ia_valid &= ~(TIMES_SET_FLAGS | ATTR_SIZE);
1863         if (S_ISREG(inode->i_mode))
1864                 inode_lock(inode);
1865         rc = simple_setattr(dentry, &op_data->op_attr);
1866         if (S_ISREG(inode->i_mode))
1867                 inode_unlock(inode);
1868         op_data->op_attr.ia_valid = ia_valid;
1869
1870         rc = ll_update_inode(inode, &md);
1871         ptlrpc_req_finished(request);
1872
1873         RETURN(rc);
1874 }
1875
1876 /**
1877  * Zero portion of page that is part of @inode.
1878  * This implies, if necessary:
1879  * - taking cl_lock on range corresponding to concerned page
1880  * - grabbing vm page
1881  * - associating cl_page
1882  * - proceeding to clio read
1883  * - zeroing range in page
1884  * - proceeding to cl_page flush
1885  * - releasing cl_lock
1886  *
1887  * \param[in] inode     inode
1888  * \param[in] index     page index
1889  * \param[in] offset    offset in page to start zero from
1890  * \param[in] len       len to zero
1891  *
1892  * \retval 0            on success
1893  * \retval negative     errno on failure
1894  */
1895 int ll_io_zero_page(struct inode *inode, pgoff_t index, pgoff_t offset,
1896                     unsigned len)
1897 {
1898         struct ll_inode_info *lli = ll_i2info(inode);
1899         struct cl_object *clob = lli->lli_clob;
1900         __u16 refcheck;
1901         struct lu_env *env = NULL;
1902         struct cl_io *io = NULL;
1903         struct cl_page *clpage = NULL;
1904         struct page *vmpage = NULL;
1905         unsigned from = index << PAGE_SHIFT;
1906         struct cl_lock *lock = NULL;
1907         struct cl_lock_descr *descr = NULL;
1908         struct cl_2queue *queue = NULL;
1909         struct cl_sync_io *anchor = NULL;
1910         bool holdinglock = false;
1911         bool lockedbymyself = true;
1912         int rc;
1913
1914         ENTRY;
1915
1916         env = cl_env_get(&refcheck);
1917         if (IS_ERR(env))
1918                 RETURN(PTR_ERR(env));
1919
1920         io = vvp_env_thread_io(env);
1921         io->ci_obj = clob;
1922         rc = cl_io_rw_init(env, io, CIT_WRITE, from, PAGE_SIZE);
1923         if (rc)
1924                 GOTO(putenv, rc);
1925
1926         lock = vvp_env_lock(env);
1927         descr = &lock->cll_descr;
1928         descr->cld_obj   = io->ci_obj;
1929         descr->cld_start = cl_index(io->ci_obj, from);
1930         descr->cld_end   = cl_index(io->ci_obj, from + PAGE_SIZE - 1);
1931         descr->cld_mode  = CLM_WRITE;
1932         descr->cld_enq_flags = CEF_MUST | CEF_NONBLOCK;
1933
1934         /* request lock for page */
1935         rc = cl_lock_request(env, io, lock);
1936         /* -ECANCELED indicates a matching lock with a different extent
1937          * was already present, and -EEXIST indicates a matching lock
1938          * on exactly the same extent was already present.
1939          * In both cases it means we are covered.
1940          */
1941         if (rc == -ECANCELED || rc == -EEXIST)
1942                 rc = 0;
1943         else if (rc < 0)
1944                 GOTO(iofini, rc);
1945         else
1946                 holdinglock = true;
1947
1948         /* grab page */
1949         vmpage = grab_cache_page_nowait(inode->i_mapping, index);
1950         if (vmpage == NULL)
1951                 GOTO(rellock, rc = -EOPNOTSUPP);
1952
1953         if (!PageDirty(vmpage)) {
1954                 /* associate cl_page */
1955                 clpage = cl_page_find(env, clob, vmpage->index,
1956                                       vmpage, CPT_CACHEABLE);
1957                 if (IS_ERR(clpage))
1958                         GOTO(pagefini, rc = PTR_ERR(clpage));
1959
1960                 cl_page_assume(env, io, clpage);
1961         }
1962
1963         if (!PageUptodate(vmpage) && !PageDirty(vmpage) &&
1964             !PageWriteback(vmpage)) {
1965                 /* read page */
1966                 /* set PagePrivate2 to detect special case of empty page
1967                  * in osc_brw_fini_request()
1968                  */
1969                 SetPagePrivate2(vmpage);
1970                 rc = ll_io_read_page(env, io, clpage, NULL);
1971                 if (!PagePrivate2(vmpage))
1972                         /* PagePrivate2 was cleared in osc_brw_fini_request()
1973                          * meaning we read an empty page. In this case, in order
1974                          * to avoid allocating unnecessary block in truncated
1975                          * file, we must not zero and write as below. Subsequent
1976                          * server-side truncate will handle things correctly.
1977                          */
1978                         GOTO(clpfini, rc = 0);
1979                 ClearPagePrivate2(vmpage);
1980                 if (rc)
1981                         GOTO(clpfini, rc);
1982                 lockedbymyself = trylock_page(vmpage);
1983                 cl_page_assume(env, io, clpage);
1984         }
1985
1986         /* zero range in page */
1987         zero_user(vmpage, offset, len);
1988
1989         if (holdinglock && clpage) {
1990                 /* explicitly write newly modified page */
1991                 queue = &io->ci_queue;
1992                 cl_2queue_init(queue);
1993                 anchor = &vvp_env_info(env)->vti_anchor;
1994                 cl_sync_io_init(anchor, 1);
1995                 clpage->cp_sync_io = anchor;
1996                 cl_2queue_add(queue, clpage, true);
1997                 rc = cl_io_submit_rw(env, io, CRT_WRITE, queue);
1998                 if (rc)
1999                         GOTO(queuefini1, rc);
2000                 rc = cl_sync_io_wait(env, anchor, 0);
2001                 if (rc)
2002                         GOTO(queuefini2, rc);
2003                 cl_page_assume(env, io, clpage);
2004
2005 queuefini2:
2006                 cl_2queue_discard(env, io, queue);
2007 queuefini1:
2008                 cl_2queue_disown(env, io, queue);
2009                 cl_2queue_fini(env, queue);
2010         }
2011
2012 clpfini:
2013         if (clpage)
2014                 cl_page_put(env, clpage);
2015 pagefini:
2016         if (lockedbymyself) {
2017                 unlock_page(vmpage);
2018                 put_page(vmpage);
2019         }
2020 rellock:
2021         if (holdinglock)
2022                 cl_lock_release(env, lock);
2023 iofini:
2024         cl_io_fini(env, io);
2025 putenv:
2026         if (env)
2027                 cl_env_put(env, &refcheck);
2028
2029         RETURN(rc);
2030 }
2031
2032 /**
2033  * Get reference file from volatile file name.
2034  * Volatile file name may look like:
2035  * <parent>/LUSTRE_VOLATILE_HDR:<mdt_index>:<random>:fd=<fd>
2036  * where fd is opened descriptor of reference file.
2037  *
2038  * \param[in] volatile_name     volatile file name
2039  * \param[in] volatile_len      volatile file name length
2040  * \param[out] ref_file         pointer to struct file of reference file
2041  *
2042  * \retval 0            on success
2043  * \retval negative     errno on failure
2044  */
2045 int volatile_ref_file(const char *volatile_name, int volatile_len,
2046                       struct file **ref_file)
2047 {
2048         char *p, *q, *fd_str;
2049         int fd, rc;
2050
2051         p = strnstr(volatile_name, ":fd=", volatile_len);
2052         if (!p || strlen(p + 4) == 0)
2053                 return -EINVAL;
2054
2055         q = strchrnul(p + 4, ':');
2056         fd_str = kstrndup(p + 4, q - p - 4, GFP_NOFS);
2057         if (!fd_str)
2058                 return -ENOMEM;
2059         rc = kstrtouint(fd_str, 10, &fd);
2060         kfree(fd_str);
2061         if (rc)
2062                 return -EINVAL;
2063
2064         *ref_file = fget(fd);
2065         if (!(*ref_file))
2066                 return -EINVAL;
2067         return 0;
2068 }
2069
2070 /* If this inode has objects allocated to it (lsm != NULL), then the OST
2071  * object(s) determine the file size and mtime.  Otherwise, the MDS will
2072  * keep these values until such a time that objects are allocated for it.
2073  * We do the MDS operations first, as it is checking permissions for us.
2074  * We don't to the MDS RPC if there is nothing that we want to store there,
2075  * otherwise there is no harm in updating mtime/atime on the MDS if we are
2076  * going to do an RPC anyways.
2077  *
2078  * If we are doing a truncate, we will send the mtime and ctime updates
2079  * to the OST with the punch RPC, otherwise we do an explicit setattr RPC.
2080  * I don't believe it is possible to get e.g. ATTR_MTIME_SET and ATTR_SIZE
2081  * at the same time.
2082  *
2083  * In case of HSMimport, we only set attr on MDS.
2084  */
2085 int ll_setattr_raw(struct dentry *dentry, struct iattr *attr,
2086                    enum op_xvalid xvalid, bool hsm_import)
2087 {
2088         struct inode *inode = dentry->d_inode;
2089         struct ll_inode_info *lli = ll_i2info(inode);
2090         struct md_op_data *op_data = NULL;
2091         ktime_t kstart = ktime_get();
2092         int rc = 0;
2093
2094         ENTRY;
2095
2096         CDEBUG(D_VFSTRACE, "%s: setattr inode "DFID"(%p) from %llu to %llu, "
2097                "valid %x, hsm_import %d\n",
2098                ll_i2sbi(inode)->ll_fsname, PFID(&lli->lli_fid),
2099                inode, i_size_read(inode), attr->ia_size, attr->ia_valid,
2100                hsm_import);
2101
2102         if (attr->ia_valid & ATTR_SIZE) {
2103                 /* Check new size against VFS/VM file size limit and rlimit */
2104                 rc = inode_newsize_ok(inode, attr->ia_size);
2105                 if (rc)
2106                         RETURN(rc);
2107
2108                 /* The maximum Lustre file size is variable, based on the
2109                  * OST maximum object size and number of stripes.  This
2110                  * needs another check in addition to the VFS check above. */
2111                 if (attr->ia_size > ll_file_maxbytes(inode)) {
2112                         CDEBUG(D_INODE,"file "DFID" too large %llu > %llu\n",
2113                                PFID(&lli->lli_fid), attr->ia_size,
2114                                ll_file_maxbytes(inode));
2115                         RETURN(-EFBIG);
2116                 }
2117
2118                 attr->ia_valid |= ATTR_MTIME | ATTR_CTIME;
2119         }
2120
2121         /* POSIX: check before ATTR_*TIME_SET set (from inode_change_ok) */
2122         if (attr->ia_valid & TIMES_SET_FLAGS) {
2123                 if ((!uid_eq(current_fsuid(), inode->i_uid)) &&
2124                     !capable(CAP_FOWNER))
2125                         RETURN(-EPERM);
2126         }
2127
2128         /* We mark all of the fields "set" so MDS/OST does not re-set them */
2129         if (!(xvalid & OP_XVALID_CTIME_SET) &&
2130              (attr->ia_valid & ATTR_CTIME)) {
2131                 attr->ia_ctime = current_time(inode);
2132                 xvalid |= OP_XVALID_CTIME_SET;
2133         }
2134         if (!(attr->ia_valid & ATTR_ATIME_SET) &&
2135             (attr->ia_valid & ATTR_ATIME)) {
2136                 attr->ia_atime = current_time(inode);
2137                 attr->ia_valid |= ATTR_ATIME_SET;
2138         }
2139         if (!(attr->ia_valid & ATTR_MTIME_SET) &&
2140             (attr->ia_valid & ATTR_MTIME)) {
2141                 attr->ia_mtime = current_time(inode);
2142                 attr->ia_valid |= ATTR_MTIME_SET;
2143         }
2144
2145         if (attr->ia_valid & (ATTR_MTIME | ATTR_CTIME))
2146                 CDEBUG(D_INODE, "setting mtime %lld, ctime %lld, now = %lld\n",
2147                        (s64)attr->ia_mtime.tv_sec, (s64)attr->ia_ctime.tv_sec,
2148                        ktime_get_real_seconds());
2149
2150         if (S_ISREG(inode->i_mode))
2151                 inode_unlock(inode);
2152
2153         /* We always do an MDS RPC, even if we're only changing the size;
2154          * only the MDS knows whether truncate() should fail with -ETXTBUSY */
2155
2156         OBD_ALLOC_PTR(op_data);
2157         if (op_data == NULL)
2158                 GOTO(out, rc = -ENOMEM);
2159
2160         if (!hsm_import && attr->ia_valid & ATTR_SIZE) {
2161                 /* If we are changing file size, file content is
2162                  * modified, flag it.
2163                  */
2164                 xvalid |= OP_XVALID_OWNEROVERRIDE;
2165                 op_data->op_bias |= MDS_DATA_MODIFIED;
2166                 clear_bit(LLIF_DATA_MODIFIED, &lli->lli_flags);
2167         }
2168
2169         if (attr->ia_valid & ATTR_FILE) {
2170                 struct ll_file_data *fd = attr->ia_file->private_data;
2171
2172                 if (fd->fd_lease_och)
2173                         op_data->op_bias |= MDS_TRUNC_KEEP_LEASE;
2174         }
2175
2176         op_data->op_attr = *attr;
2177         op_data->op_xvalid = xvalid;
2178
2179         rc = ll_md_setattr(dentry, op_data);
2180         if (rc)
2181                 GOTO(out, rc);
2182
2183         if (!S_ISREG(inode->i_mode) || hsm_import)
2184                 GOTO(out, rc = 0);
2185
2186         if (attr->ia_valid & (ATTR_SIZE | ATTR_ATIME | ATTR_ATIME_SET |
2187                               ATTR_MTIME | ATTR_MTIME_SET | ATTR_CTIME) ||
2188             xvalid & OP_XVALID_CTIME_SET) {
2189                 bool cached = false;
2190
2191                 rc = pcc_inode_setattr(inode, attr, &cached);
2192                 if (cached) {
2193                         if (rc) {
2194                                 CERROR("%s: PCC inode "DFID" setattr failed: "
2195                                        "rc = %d\n",
2196                                        ll_i2sbi(inode)->ll_fsname,
2197                                        PFID(&lli->lli_fid), rc);
2198                                 GOTO(out, rc);
2199                         }
2200                 } else {
2201                         unsigned int flags = 0;
2202
2203                         /* For truncate and utimes sending attributes to OSTs,
2204                          * setting mtime/atime to the past will be performed
2205                          * under PW [0:EOF] extent lock (new_size:EOF for
2206                          * truncate). It may seem excessive to send mtime/atime
2207                          * updates to OSTs when not setting times to past, but
2208                          * it is necessary due to possible time
2209                          * de-synchronization between MDT inode and OST objects
2210                          */
2211                         if (S_ISREG(inode->i_mode) && IS_ENCRYPTED(inode)) {
2212                                 xvalid |= OP_XVALID_FLAGS;
2213                                 flags = LUSTRE_ENCRYPT_FL;
2214                                 /* Call to ll_io_zero_page is not necessary if
2215                                  * truncating on PAGE_SIZE boundary, because
2216                                  * whole pages will be wiped.
2217                                  * In case of Direct IO, all we need is to set
2218                                  * new size.
2219                                  */
2220                                 if (attr->ia_valid & ATTR_SIZE &&
2221                                     attr->ia_size & ~PAGE_MASK &&
2222                                     !(attr->ia_valid & ATTR_FILE &&
2223                                       attr->ia_file->f_flags & O_DIRECT)) {
2224                                         pgoff_t offset =
2225                                                 attr->ia_size & (PAGE_SIZE - 1);
2226
2227                                         rc = ll_io_zero_page(inode,
2228                                                     attr->ia_size >> PAGE_SHIFT,
2229                                                     offset, PAGE_SIZE - offset);
2230                                         if (rc)
2231                                                 GOTO(out, rc);
2232                                 }
2233                                 /* If encrypted volatile file without the key,
2234                                  * we need to fetch size from reference file,
2235                                  * and set it on OST objects. This happens when
2236                                  * migrating or extending an encrypted file
2237                                  * without the key.
2238                                  */
2239                                 if (filename_is_volatile(dentry->d_name.name,
2240                                                          dentry->d_name.len,
2241                                                          NULL) &&
2242                                     llcrypt_require_key(inode) == -ENOKEY) {
2243                                         struct file *ref_file;
2244                                         struct inode *ref_inode;
2245                                         struct ll_inode_info *ref_lli;
2246                                         struct cl_object *ref_obj;
2247                                         struct cl_attr ref_attr = { 0 };
2248                                         struct lu_env *env;
2249                                         __u16 refcheck;
2250
2251                                         rc = volatile_ref_file(
2252                                                 dentry->d_name.name,
2253                                                 dentry->d_name.len,
2254                                                 &ref_file);
2255                                         if (rc)
2256                                                 GOTO(out, rc);
2257
2258                                         ref_inode = file_inode(ref_file);
2259                                         if (!ref_inode) {
2260                                                 fput(ref_file);
2261                                                 GOTO(out, rc = -EINVAL);
2262                                         }
2263
2264                                         env = cl_env_get(&refcheck);
2265                                         if (IS_ERR(env))
2266                                                 GOTO(out, rc = PTR_ERR(env));
2267
2268                                         ref_lli = ll_i2info(ref_inode);
2269                                         ref_obj = ref_lli->lli_clob;
2270                                         cl_object_attr_lock(ref_obj);
2271                                         rc = cl_object_attr_get(env, ref_obj,
2272                                                                 &ref_attr);
2273                                         cl_object_attr_unlock(ref_obj);
2274                                         cl_env_put(env, &refcheck);
2275                                         fput(ref_file);
2276                                         if (rc)
2277                                                 GOTO(out, rc);
2278
2279                                         attr->ia_valid |= ATTR_SIZE;
2280                                         attr->ia_size = ref_attr.cat_size;
2281                                 }
2282                         }
2283                         rc = cl_setattr_ost(lli->lli_clob, attr, xvalid, flags);
2284                 }
2285         }
2286
2287         /* If the file was restored, it needs to set dirty flag.
2288          *
2289          * We've already sent MDS_DATA_MODIFIED flag in
2290          * ll_md_setattr() for truncate. However, the MDT refuses to
2291          * set the HS_DIRTY flag on released files, so we have to set
2292          * it again if the file has been restored. Please check how
2293          * LLIF_DATA_MODIFIED is set in vvp_io_setattr_fini().
2294          *
2295          * Please notice that if the file is not released, the previous
2296          * MDS_DATA_MODIFIED has taken effect and usually
2297          * LLIF_DATA_MODIFIED is not set(see vvp_io_setattr_fini()).
2298          * This way we can save an RPC for common open + trunc
2299          * operation. */
2300         if (test_and_clear_bit(LLIF_DATA_MODIFIED, &lli->lli_flags)) {
2301                 struct hsm_state_set hss = {
2302                         .hss_valid = HSS_SETMASK,
2303                         .hss_setmask = HS_DIRTY,
2304                 };
2305                 int rc2;
2306
2307                 rc2 = ll_hsm_state_set(inode, &hss);
2308                 /* truncate and write can happen at the same time, so that
2309                  * the file can be set modified even though the file is not
2310                  * restored from released state, and ll_hsm_state_set() is
2311                  * not applicable for the file, and rc2 < 0 is normal in this
2312                  * case. */
2313                 if (rc2 < 0)
2314                         CDEBUG(D_INFO, DFID "HSM set dirty failed: rc2 = %d\n",
2315                                PFID(ll_inode2fid(inode)), rc2);
2316         }
2317
2318         EXIT;
2319 out:
2320         if (op_data != NULL)
2321                 ll_finish_md_op_data(op_data);
2322
2323         if (S_ISREG(inode->i_mode)) {
2324                 inode_lock(inode);
2325                 if ((attr->ia_valid & ATTR_SIZE) && !hsm_import)
2326                         inode_dio_wait(inode);
2327                 /* Once we've got the i_mutex, it's safe to set the S_NOSEC
2328                  * flag.  ll_update_inode (called from ll_md_setattr), clears
2329                  * inode flags, so there is a gap where S_NOSEC is not set.
2330                  * This can cause a writer to take the i_mutex unnecessarily,
2331                  * but this is safe to do and should be rare. */
2332                 inode_has_no_xattr(inode);
2333         }
2334
2335         if (!rc)
2336                 ll_stats_ops_tally(ll_i2sbi(inode), attr->ia_valid & ATTR_SIZE ?
2337                                         LPROC_LL_TRUNC : LPROC_LL_SETATTR,
2338                                    ktime_us_delta(ktime_get(), kstart));
2339
2340         RETURN(rc);
2341 }
2342
2343 int ll_setattr(struct dentry *de, struct iattr *attr)
2344 {
2345         int mode = de->d_inode->i_mode;
2346         enum op_xvalid xvalid = 0;
2347         int rc;
2348
2349         rc = llcrypt_prepare_setattr(de, attr);
2350         if (rc)
2351                 return rc;
2352
2353         if ((attr->ia_valid & (ATTR_CTIME|ATTR_SIZE|ATTR_MODE)) ==
2354                               (ATTR_CTIME|ATTR_SIZE|ATTR_MODE))
2355                 xvalid |= OP_XVALID_OWNEROVERRIDE;
2356
2357         if (((attr->ia_valid & (ATTR_MODE|ATTR_FORCE|ATTR_SIZE)) ==
2358                                (ATTR_SIZE|ATTR_MODE)) &&
2359             (((mode & S_ISUID) && !(attr->ia_mode & S_ISUID)) ||
2360              (((mode & (S_ISGID|S_IXGRP)) == (S_ISGID|S_IXGRP)) &&
2361               !(attr->ia_mode & S_ISGID))))
2362                 attr->ia_valid |= ATTR_FORCE;
2363
2364         if ((attr->ia_valid & ATTR_MODE) &&
2365             (mode & S_ISUID) &&
2366             !(attr->ia_mode & S_ISUID) &&
2367             !(attr->ia_valid & ATTR_KILL_SUID))
2368                 attr->ia_valid |= ATTR_KILL_SUID;
2369
2370         if ((attr->ia_valid & ATTR_MODE) &&
2371             ((mode & (S_ISGID|S_IXGRP)) == (S_ISGID|S_IXGRP)) &&
2372             !(attr->ia_mode & S_ISGID) &&
2373             !(attr->ia_valid & ATTR_KILL_SGID))
2374                 attr->ia_valid |= ATTR_KILL_SGID;
2375
2376         return ll_setattr_raw(de, attr, xvalid, false);
2377 }
2378
2379 int ll_statfs_internal(struct ll_sb_info *sbi, struct obd_statfs *osfs,
2380                        u32 flags)
2381 {
2382         struct obd_statfs obd_osfs = { 0 };
2383         time64_t max_age;
2384         int rc;
2385
2386         ENTRY;
2387         max_age = ktime_get_seconds() - sbi->ll_statfs_max_age;
2388
2389         if (test_bit(LL_SBI_LAZYSTATFS, sbi->ll_flags))
2390                 flags |= OBD_STATFS_NODELAY;
2391
2392         rc = obd_statfs(NULL, sbi->ll_md_exp, osfs, max_age, flags);
2393         if (rc)
2394                 RETURN(rc);
2395
2396         osfs->os_type = LL_SUPER_MAGIC;
2397
2398         CDEBUG(D_SUPER, "MDC blocks %llu/%llu objects %llu/%llu\n",
2399               osfs->os_bavail, osfs->os_blocks, osfs->os_ffree, osfs->os_files);
2400
2401         if (osfs->os_state & OS_STATFS_SUM)
2402                 GOTO(out, rc);
2403
2404         rc = obd_statfs(NULL, sbi->ll_dt_exp, &obd_osfs, max_age, flags);
2405         if (rc) /* Possibly a filesystem with no OSTs.  Report MDT totals. */
2406                 GOTO(out, rc = 0);
2407
2408         CDEBUG(D_SUPER, "OSC blocks %llu/%llu objects %llu/%llu\n",
2409                obd_osfs.os_bavail, obd_osfs.os_blocks, obd_osfs.os_ffree,
2410                obd_osfs.os_files);
2411
2412         osfs->os_bsize = obd_osfs.os_bsize;
2413         osfs->os_blocks = obd_osfs.os_blocks;
2414         osfs->os_bfree = obd_osfs.os_bfree;
2415         osfs->os_bavail = obd_osfs.os_bavail;
2416
2417         /* If we have _some_ OSTs, but don't have as many free objects on the
2418          * OSTs as inodes on the MDTs, reduce the reported number of inodes
2419          * to compensate, so that the "inodes in use" number is correct.
2420          * This should be kept in sync with lod_statfs() behaviour.
2421          */
2422         if (obd_osfs.os_files && obd_osfs.os_ffree < osfs->os_ffree) {
2423                 osfs->os_files = (osfs->os_files - osfs->os_ffree) +
2424                                  obd_osfs.os_ffree;
2425                 osfs->os_ffree = obd_osfs.os_ffree;
2426         }
2427
2428 out:
2429         RETURN(rc);
2430 }
2431
2432 static int ll_statfs_project(struct inode *inode, struct kstatfs *sfs)
2433 {
2434         struct if_quotactl qctl = {
2435                 .qc_cmd = LUSTRE_Q_GETQUOTA,
2436                 .qc_type = PRJQUOTA,
2437                 .qc_valid = QC_GENERAL,
2438         };
2439         u64 limit, curblock;
2440         int ret;
2441
2442         qctl.qc_id = ll_i2info(inode)->lli_projid;
2443         ret = quotactl_ioctl(inode->i_sb, &qctl);
2444         if (ret) {
2445                 /* ignore errors if project ID does not have
2446                  * a quota limit or feature unsupported.
2447                  */
2448                 if (ret == -ESRCH || ret == -EOPNOTSUPP)
2449                         ret = 0;
2450                 return ret;
2451         }
2452
2453         limit = ((qctl.qc_dqblk.dqb_bsoftlimit ?
2454                  qctl.qc_dqblk.dqb_bsoftlimit :
2455                  qctl.qc_dqblk.dqb_bhardlimit) * 1024) / sfs->f_bsize;
2456         if (limit && sfs->f_blocks > limit) {
2457                 curblock = (qctl.qc_dqblk.dqb_curspace +
2458                                 sfs->f_bsize - 1) / sfs->f_bsize;
2459                 sfs->f_blocks = limit;
2460                 sfs->f_bfree = sfs->f_bavail =
2461                         (sfs->f_blocks > curblock) ?
2462                         (sfs->f_blocks - curblock) : 0;
2463         }
2464
2465         limit = qctl.qc_dqblk.dqb_isoftlimit ?
2466                 qctl.qc_dqblk.dqb_isoftlimit :
2467                 qctl.qc_dqblk.dqb_ihardlimit;
2468         if (limit && sfs->f_files > limit) {
2469                 sfs->f_files = limit;
2470                 sfs->f_ffree = (sfs->f_files >
2471                         qctl.qc_dqblk.dqb_curinodes) ?
2472                         (sfs->f_files - qctl.qc_dqblk.dqb_curinodes) : 0;
2473         }
2474
2475         return 0;
2476 }
2477
2478 int ll_statfs(struct dentry *de, struct kstatfs *sfs)
2479 {
2480         struct super_block *sb = de->d_sb;
2481         struct obd_statfs osfs;
2482         __u64 fsid = huge_encode_dev(sb->s_dev);
2483         ktime_t kstart = ktime_get();
2484         int rc;
2485
2486         CDEBUG(D_VFSTRACE, "VFS Op:sb=%s (%p)\n", sb->s_id, sb);
2487
2488         /* Some amount of caching on the client is allowed */
2489         rc = ll_statfs_internal(ll_s2sbi(sb), &osfs, OBD_STATFS_SUM);
2490         if (rc)
2491                 return rc;
2492
2493         statfs_unpack(sfs, &osfs);
2494
2495         /* We need to downshift for all 32-bit kernels, because we can't
2496          * tell if the kernel is being called via sys_statfs64() or not.
2497          * Stop before overflowing f_bsize - in which case it is better
2498          * to just risk EOVERFLOW if caller is using old sys_statfs(). */
2499         if (sizeof(long) < 8) {
2500                 while (osfs.os_blocks > ~0UL && sfs->f_bsize < 0x40000000) {
2501                         sfs->f_bsize <<= 1;
2502
2503                         osfs.os_blocks >>= 1;
2504                         osfs.os_bfree >>= 1;
2505                         osfs.os_bavail >>= 1;
2506                 }
2507         }
2508
2509         sfs->f_blocks = osfs.os_blocks;
2510         sfs->f_bfree = osfs.os_bfree;
2511         sfs->f_bavail = osfs.os_bavail;
2512         sfs->f_fsid.val[0] = (__u32)fsid;
2513         sfs->f_fsid.val[1] = (__u32)(fsid >> 32);
2514         if (ll_i2info(de->d_inode)->lli_projid)
2515                 return ll_statfs_project(de->d_inode, sfs);
2516
2517         ll_stats_ops_tally(ll_s2sbi(sb), LPROC_LL_STATFS,
2518                            ktime_us_delta(ktime_get(), kstart));
2519
2520         return 0;
2521 }
2522
2523 void ll_inode_size_lock(struct inode *inode)
2524 {
2525         struct ll_inode_info *lli;
2526
2527         LASSERT(!S_ISDIR(inode->i_mode));
2528
2529         lli = ll_i2info(inode);
2530         mutex_lock(&lli->lli_size_mutex);
2531 }
2532
2533 void ll_inode_size_unlock(struct inode *inode)
2534 {
2535         struct ll_inode_info *lli;
2536
2537         lli = ll_i2info(inode);
2538         mutex_unlock(&lli->lli_size_mutex);
2539 }
2540
2541 void ll_update_inode_flags(struct inode *inode, unsigned int ext_flags)
2542 {
2543         /* do not clear encryption flag */
2544         ext_flags |= ll_inode_to_ext_flags(inode->i_flags) & LUSTRE_ENCRYPT_FL;
2545         inode->i_flags = ll_ext_to_inode_flags(ext_flags);
2546         if (ext_flags & LUSTRE_PROJINHERIT_FL)
2547                 set_bit(LLIF_PROJECT_INHERIT, &ll_i2info(inode)->lli_flags);
2548         else
2549                 clear_bit(LLIF_PROJECT_INHERIT, &ll_i2info(inode)->lli_flags);
2550 }
2551
2552 int ll_update_inode(struct inode *inode, struct lustre_md *md)
2553 {
2554         struct ll_inode_info *lli = ll_i2info(inode);
2555         struct mdt_body *body = md->body;
2556         struct ll_sb_info *sbi = ll_i2sbi(inode);
2557         bool api32;
2558         int rc = 0;
2559
2560         if (body->mbo_valid & OBD_MD_FLEASIZE) {
2561                 rc = cl_file_inode_init(inode, md);
2562                 if (rc)
2563                         return rc;
2564         }
2565
2566         if (S_ISDIR(inode->i_mode)) {
2567                 rc = ll_update_lsm_md(inode, md);
2568                 if (rc != 0)
2569                         return rc;
2570         }
2571
2572         if (body->mbo_valid & OBD_MD_FLACL)
2573                 lli_replace_acl(lli, md);
2574
2575         api32 = test_bit(LL_SBI_32BIT_API, sbi->ll_flags);
2576         inode->i_ino = cl_fid_build_ino(&body->mbo_fid1, api32);
2577         inode->i_generation = cl_fid_build_gen(&body->mbo_fid1);
2578
2579         if (body->mbo_valid & OBD_MD_FLATIME) {
2580                 if (body->mbo_atime > inode->i_atime.tv_sec)
2581                         inode->i_atime.tv_sec = body->mbo_atime;
2582                 lli->lli_atime = body->mbo_atime;
2583         }
2584
2585         if (body->mbo_valid & OBD_MD_FLMTIME) {
2586                 if (body->mbo_mtime > inode->i_mtime.tv_sec) {
2587                         CDEBUG(D_INODE,
2588                                "setting ino %lu mtime from %lld to %llu\n",
2589                                inode->i_ino, (s64)inode->i_mtime.tv_sec,
2590                                body->mbo_mtime);
2591                         inode->i_mtime.tv_sec = body->mbo_mtime;
2592                 }
2593                 lli->lli_mtime = body->mbo_mtime;
2594         }
2595
2596         if (body->mbo_valid & OBD_MD_FLCTIME) {
2597                 if (body->mbo_ctime > inode->i_ctime.tv_sec)
2598                         inode->i_ctime.tv_sec = body->mbo_ctime;
2599                 lli->lli_ctime = body->mbo_ctime;
2600         }
2601
2602         if (body->mbo_valid & OBD_MD_FLBTIME)
2603                 lli->lli_btime = body->mbo_btime;
2604
2605         /* Clear i_flags to remove S_NOSEC before permissions are updated */
2606         if (body->mbo_valid & OBD_MD_FLFLAGS)
2607                 ll_update_inode_flags(inode, body->mbo_flags);
2608         if (body->mbo_valid & OBD_MD_FLMODE)
2609                 inode->i_mode = (inode->i_mode & S_IFMT) |
2610                                 (body->mbo_mode & ~S_IFMT);
2611
2612         if (body->mbo_valid & OBD_MD_FLTYPE)
2613                 inode->i_mode = (inode->i_mode & ~S_IFMT) |
2614                                 (body->mbo_mode & S_IFMT);
2615
2616         LASSERT(inode->i_mode != 0);
2617         if (body->mbo_valid & OBD_MD_FLUID)
2618                 inode->i_uid = make_kuid(&init_user_ns, body->mbo_uid);
2619         if (body->mbo_valid & OBD_MD_FLGID)
2620                 inode->i_gid = make_kgid(&init_user_ns, body->mbo_gid);
2621         if (body->mbo_valid & OBD_MD_FLPROJID)
2622                 lli->lli_projid = body->mbo_projid;
2623         if (body->mbo_valid & OBD_MD_FLNLINK) {
2624                 spin_lock(&inode->i_lock);
2625                 set_nlink(inode, body->mbo_nlink);
2626                 spin_unlock(&inode->i_lock);
2627         }
2628         if (body->mbo_valid & OBD_MD_FLRDEV)
2629                 inode->i_rdev = old_decode_dev(body->mbo_rdev);
2630
2631         if (body->mbo_valid & OBD_MD_FLID) {
2632                 /* FID shouldn't be changed! */
2633                 if (fid_is_sane(&lli->lli_fid)) {
2634                         LASSERTF(lu_fid_eq(&lli->lli_fid, &body->mbo_fid1),
2635                                  "Trying to change FID "DFID
2636                                  " to the "DFID", inode "DFID"(%p)\n",
2637                                  PFID(&lli->lli_fid), PFID(&body->mbo_fid1),
2638                                  PFID(ll_inode2fid(inode)), inode);
2639                 } else {
2640                         lli->lli_fid = body->mbo_fid1;
2641                 }
2642         }
2643
2644         LASSERT(fid_seq(&lli->lli_fid) != 0);
2645
2646         /* In case of encrypted file without the key, please do not lose
2647          * clear text size stored into lli_lazysize in ll_merge_attr(),
2648          * we will need it in ll_prepare_close().
2649          */
2650         if (lli->lli_attr_valid & OBD_MD_FLLAZYSIZE && lli->lli_lazysize &&
2651             llcrypt_require_key(inode) == -ENOKEY)
2652                 lli->lli_attr_valid = body->mbo_valid | OBD_MD_FLLAZYSIZE;
2653         else
2654                 lli->lli_attr_valid = body->mbo_valid;
2655         if (body->mbo_valid & OBD_MD_FLSIZE) {
2656                 i_size_write(inode, body->mbo_size);
2657
2658                 CDEBUG(D_VFSTRACE, "inode="DFID", updating i_size %llu\n",
2659                        PFID(ll_inode2fid(inode)),
2660                        (unsigned long long)body->mbo_size);
2661
2662                 if (body->mbo_valid & OBD_MD_FLBLOCKS)
2663                         inode->i_blocks = body->mbo_blocks;
2664         } else {
2665                 if (body->mbo_valid & OBD_MD_FLLAZYSIZE)
2666                         lli->lli_lazysize = body->mbo_size;
2667                 if (body->mbo_valid & OBD_MD_FLLAZYBLOCKS)
2668                         lli->lli_lazyblocks = body->mbo_blocks;
2669         }
2670
2671         if (body->mbo_valid & OBD_MD_TSTATE) {
2672                 /* Set LLIF_FILE_RESTORING if restore ongoing and
2673                  * clear it when done to ensure to start again
2674                  * glimpsing updated attrs
2675                  */
2676                 if (body->mbo_t_state & MS_RESTORE)
2677                         set_bit(LLIF_FILE_RESTORING, &lli->lli_flags);
2678                 else
2679                         clear_bit(LLIF_FILE_RESTORING, &lli->lli_flags);
2680         }
2681
2682         return 0;
2683 }
2684
2685 /* update directory depth to ROOT, called after LOOKUP lock is fetched. */
2686 void ll_update_dir_depth(struct inode *dir, struct inode *inode)
2687 {
2688         struct ll_inode_info *lli;
2689
2690         if (!S_ISDIR(inode->i_mode))
2691                 return;
2692
2693         if (inode == dir)
2694                 return;
2695
2696         lli = ll_i2info(inode);
2697         lli->lli_depth = ll_i2info(dir)->lli_depth + 1;
2698         CDEBUG(D_INODE, DFID" depth %hu\n", PFID(&lli->lli_fid), lli->lli_depth);
2699 }
2700
2701 void ll_truncate_inode_pages_final(struct inode *inode)
2702 {
2703         struct address_space *mapping = &inode->i_data;
2704         unsigned long nrpages;
2705         unsigned long flags;
2706
2707         truncate_inode_pages_final(mapping);
2708
2709         /* Workaround for LU-118: Note nrpages may not be totally updated when
2710          * truncate_inode_pages() returns, as there can be a page in the process
2711          * of deletion (inside __delete_from_page_cache()) in the specified
2712          * range. Thus mapping->nrpages can be non-zero when this function
2713          * returns even after truncation of the whole mapping.  Only do this if
2714          * npages isn't already zero.
2715          */
2716         nrpages = mapping->nrpages;
2717         if (nrpages) {
2718                 ll_xa_lock_irqsave(&mapping->i_pages, flags);
2719                 nrpages = mapping->nrpages;
2720                 ll_xa_unlock_irqrestore(&mapping->i_pages, flags);
2721         } /* Workaround end */
2722
2723         LASSERTF(nrpages == 0, "%s: inode="DFID"(%p) nrpages=%lu, "
2724                  "see https://jira.whamcloud.com/browse/LU-118\n",
2725                  ll_i2sbi(inode)->ll_fsname,
2726                  PFID(ll_inode2fid(inode)), inode, nrpages);
2727 }
2728
2729 int ll_read_inode2(struct inode *inode, void *opaque)
2730 {
2731         struct lustre_md *md = opaque;
2732         struct ll_inode_info *lli = ll_i2info(inode);
2733         int     rc;
2734         ENTRY;
2735
2736         CDEBUG(D_VFSTRACE, "VFS Op:inode="DFID"(%p)\n",
2737                PFID(&lli->lli_fid), inode);
2738
2739         /* Core attributes from the MDS first.  This is a new inode, and
2740          * the VFS doesn't zero times in the core inode so we have to do
2741          * it ourselves.  They will be overwritten by either MDS or OST
2742          * attributes - we just need to make sure they aren't newer.
2743          */
2744         inode->i_mtime.tv_sec = 0;
2745         inode->i_atime.tv_sec = 0;
2746         inode->i_ctime.tv_sec = 0;
2747         inode->i_rdev = 0;
2748         rc = ll_update_inode(inode, md);
2749         if (rc != 0)
2750                 RETURN(rc);
2751
2752         /* OIDEBUG(inode); */
2753
2754 #ifdef HAVE_BACKING_DEV_INFO
2755         /* initializing backing dev info. */
2756         inode->i_mapping->backing_dev_info = &s2lsi(inode->i_sb)->lsi_bdi;
2757 #endif
2758         if (S_ISREG(inode->i_mode)) {
2759                 struct ll_sb_info *sbi = ll_i2sbi(inode);
2760                 inode->i_op = &ll_file_inode_operations;
2761                 inode->i_fop = sbi->ll_fop;
2762                 inode->i_mapping->a_ops = (struct address_space_operations *)&ll_aops;
2763                 EXIT;
2764         } else if (S_ISDIR(inode->i_mode)) {
2765                 inode->i_op = &ll_dir_inode_operations;
2766                 inode->i_fop = &ll_dir_operations;
2767                 EXIT;
2768         } else if (S_ISLNK(inode->i_mode)) {
2769                 inode->i_op = &ll_fast_symlink_inode_operations;
2770                 EXIT;
2771         } else {
2772                 inode->i_op = &ll_special_inode_operations;
2773
2774                 init_special_inode(inode, inode->i_mode,
2775                                    inode->i_rdev);
2776
2777                 EXIT;
2778         }
2779
2780         return 0;
2781 }
2782
2783 void ll_delete_inode(struct inode *inode)
2784 {
2785         struct ll_inode_info *lli = ll_i2info(inode);
2786         ENTRY;
2787
2788         if (S_ISREG(inode->i_mode) && lli->lli_clob != NULL) {
2789                 /* It is last chance to write out dirty pages,
2790                  * otherwise we may lose data while umount.
2791                  *
2792                  * If i_nlink is 0 then just discard data. This is safe because
2793                  * local inode gets i_nlink 0 from server only for the last
2794                  * unlink, so that file is not opened somewhere else
2795                  */
2796                 cl_sync_file_range(inode, 0, OBD_OBJECT_EOF, inode->i_nlink ?
2797                                    CL_FSYNC_LOCAL : CL_FSYNC_DISCARD, 1);
2798         }
2799
2800         ll_truncate_inode_pages_final(inode);
2801         ll_clear_inode(inode);
2802         clear_inode(inode);
2803
2804         EXIT;
2805 }
2806
2807 int ll_iocontrol(struct inode *inode, struct file *file,
2808                  unsigned int cmd, unsigned long arg)
2809 {
2810         struct ll_sb_info *sbi = ll_i2sbi(inode);
2811         struct ptlrpc_request *req = NULL;
2812         int rc, flags = 0;
2813         ENTRY;
2814
2815         switch (cmd) {
2816         case FS_IOC_GETFLAGS: {
2817                 struct mdt_body *body;
2818                 struct md_op_data *op_data;
2819
2820                 op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL,
2821                                              0, 0, LUSTRE_OPC_ANY,
2822                                              NULL);
2823                 if (IS_ERR(op_data))
2824                         RETURN(PTR_ERR(op_data));
2825
2826                 op_data->op_valid = OBD_MD_FLFLAGS;
2827                 rc = md_getattr(sbi->ll_md_exp, op_data, &req);
2828                 ll_finish_md_op_data(op_data);
2829                 if (rc) {
2830                         CERROR("%s: failure inode "DFID": rc = %d\n",
2831                                sbi->ll_md_exp->exp_obd->obd_name,
2832                                PFID(ll_inode2fid(inode)), rc);
2833                         RETURN(-abs(rc));
2834                 }
2835
2836                 body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
2837
2838                 flags = body->mbo_flags;
2839
2840                 ptlrpc_req_finished(req);
2841
2842                 RETURN(put_user(flags, (int __user *)arg));
2843         }
2844         case FS_IOC_SETFLAGS: {
2845                 struct iattr *attr;
2846                 struct md_op_data *op_data;
2847                 struct cl_object *obj;
2848                 struct fsxattr fa = { 0 };
2849
2850                 if (get_user(flags, (int __user *)arg))
2851                         RETURN(-EFAULT);
2852
2853                 fa.fsx_projid = ll_i2info(inode)->lli_projid;
2854                 if (flags & LUSTRE_PROJINHERIT_FL)
2855                         fa.fsx_xflags = FS_XFLAG_PROJINHERIT;
2856
2857                 rc = ll_ioctl_check_project(inode, fa.fsx_xflags,
2858                                             fa.fsx_projid);
2859                 if (rc)
2860                         RETURN(rc);
2861
2862                 op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL, 0, 0,
2863                                              LUSTRE_OPC_ANY, NULL);
2864                 if (IS_ERR(op_data))
2865                         RETURN(PTR_ERR(op_data));
2866
2867                 op_data->op_attr_flags = flags;
2868                 op_data->op_xvalid |= OP_XVALID_FLAGS;
2869                 rc = md_setattr(sbi->ll_md_exp, op_data, NULL, 0, &req);
2870                 ll_finish_md_op_data(op_data);
2871                 ptlrpc_req_finished(req);
2872                 if (rc)
2873                         RETURN(rc);
2874
2875                 ll_update_inode_flags(inode, flags);
2876
2877                 obj = ll_i2info(inode)->lli_clob;
2878                 if (obj == NULL)
2879                         RETURN(0);
2880
2881                 OBD_ALLOC_PTR(attr);
2882                 if (attr == NULL)
2883                         RETURN(-ENOMEM);
2884
2885                 rc = cl_setattr_ost(obj, attr, OP_XVALID_FLAGS, flags);
2886
2887                 OBD_FREE_PTR(attr);
2888                 RETURN(rc);
2889         }
2890         default:
2891                 RETURN(-ENOSYS);
2892         }
2893
2894         RETURN(0);
2895 }
2896
2897 int ll_flush_ctx(struct inode *inode)
2898 {
2899         struct ll_sb_info  *sbi = ll_i2sbi(inode);
2900
2901         CDEBUG(D_SEC, "flush context for user %d\n",
2902                from_kuid(&init_user_ns, current_uid()));
2903
2904         obd_set_info_async(NULL, sbi->ll_md_exp,
2905                            sizeof(KEY_FLUSH_CTX), KEY_FLUSH_CTX,
2906                            0, NULL, NULL);
2907         obd_set_info_async(NULL, sbi->ll_dt_exp,
2908                            sizeof(KEY_FLUSH_CTX), KEY_FLUSH_CTX,
2909                            0, NULL, NULL);
2910         return 0;
2911 }
2912
2913 /* umount -f client means force down, don't save state */
2914 void ll_umount_begin(struct super_block *sb)
2915 {
2916         struct ll_sb_info *sbi = ll_s2sbi(sb);
2917         struct obd_device *obd;
2918         struct obd_ioctl_data *ioc_data;
2919         int cnt;
2920         ENTRY;
2921
2922         CDEBUG(D_VFSTRACE, "VFS Op: superblock %p count %d active %d\n", sb,
2923                sb->s_count, atomic_read(&sb->s_active));
2924
2925         obd = class_exp2obd(sbi->ll_md_exp);
2926         if (obd == NULL) {
2927                 CERROR("Invalid MDC connection handle %#llx\n",
2928                        sbi->ll_md_exp->exp_handle.h_cookie);
2929                 EXIT;
2930                 return;
2931         }
2932         obd->obd_force = 1;
2933
2934         obd = class_exp2obd(sbi->ll_dt_exp);
2935         if (obd == NULL) {
2936                 CERROR("Invalid LOV connection handle %#llx\n",
2937                        sbi->ll_dt_exp->exp_handle.h_cookie);
2938                 EXIT;
2939                 return;
2940         }
2941         obd->obd_force = 1;
2942
2943         OBD_ALLOC_PTR(ioc_data);
2944         if (ioc_data) {
2945                 obd_iocontrol(IOC_OSC_SET_ACTIVE, sbi->ll_md_exp,
2946                               sizeof *ioc_data, ioc_data, NULL);
2947
2948                 obd_iocontrol(IOC_OSC_SET_ACTIVE, sbi->ll_dt_exp,
2949                               sizeof *ioc_data, ioc_data, NULL);
2950
2951                 OBD_FREE_PTR(ioc_data);
2952         }
2953
2954         /* Really, we'd like to wait until there are no requests outstanding,
2955          * and then continue.  For now, we just periodically checking for vfs
2956          * to decrement mnt_cnt and hope to finish it within 10sec.
2957          */
2958         cnt = 10;
2959         while (cnt > 0 &&
2960                !may_umount(sbi->ll_mnt.mnt)) {
2961                 ssleep(1);
2962                 cnt -= 1;
2963         }
2964
2965         EXIT;
2966 }
2967
2968 int ll_remount_fs(struct super_block *sb, int *flags, char *data)
2969 {
2970         struct ll_sb_info *sbi = ll_s2sbi(sb);
2971         char *profilenm = get_profile_name(sb);
2972         int err;
2973         __u32 read_only;
2974
2975         if ((*flags & MS_RDONLY) != (sb->s_flags & SB_RDONLY)) {
2976                 read_only = *flags & MS_RDONLY;
2977                 err = obd_set_info_async(NULL, sbi->ll_md_exp,
2978                                          sizeof(KEY_READ_ONLY),
2979                                          KEY_READ_ONLY, sizeof(read_only),
2980                                          &read_only, NULL);
2981                 if (err) {
2982                         LCONSOLE_WARN("Failed to remount %s %s (%d)\n",
2983                                       profilenm, read_only ?
2984                                       "read-only" : "read-write", err);
2985                         return err;
2986                 }
2987
2988                 if (read_only)
2989                         sb->s_flags |= SB_RDONLY;
2990                 else
2991                         sb->s_flags &= ~SB_RDONLY;
2992
2993                 if (test_bit(LL_SBI_VERBOSE, sbi->ll_flags))
2994                         LCONSOLE_WARN("Remounted %s %s\n", profilenm,
2995                                       read_only ?  "read-only" : "read-write");
2996         }
2997         return 0;
2998 }
2999
3000 /**
3001  * Cleanup the open handle that is cached on MDT-side.
3002  *
3003  * For open case, the client side open handling thread may hit error
3004  * after the MDT grant the open. Under such case, the client should
3005  * send close RPC to the MDT as cleanup; otherwise, the open handle
3006  * on the MDT will be leaked there until the client umount or evicted.
3007  *
3008  * In further, if someone unlinked the file, because the open handle
3009  * holds the reference on such file/object, then it will block the
3010  * subsequent threads that want to locate such object via FID.
3011  *
3012  * \param[in] sb        super block for this file-system
3013  * \param[in] open_req  pointer to the original open request
3014  */
3015 void ll_open_cleanup(struct super_block *sb, struct req_capsule *pill)
3016 {
3017         struct mdt_body                 *body;
3018         struct md_op_data               *op_data;
3019         struct ptlrpc_request           *close_req = NULL;
3020         struct obd_export               *exp       = ll_s2sbi(sb)->ll_md_exp;
3021         ENTRY;
3022
3023         body = req_capsule_server_get(pill, &RMF_MDT_BODY);
3024         OBD_ALLOC_PTR(op_data);
3025         if (op_data == NULL) {
3026                 CWARN("%s: cannot allocate op_data to release open handle for "
3027                       DFID"\n", ll_s2sbi(sb)->ll_fsname, PFID(&body->mbo_fid1));
3028
3029                 RETURN_EXIT;
3030         }
3031
3032         op_data->op_fid1 = body->mbo_fid1;
3033         op_data->op_open_handle = body->mbo_open_handle;
3034         op_data->op_mod_time = ktime_get_real_seconds();
3035         md_close(exp, op_data, NULL, &close_req);
3036         ptlrpc_req_finished(close_req);
3037         ll_finish_md_op_data(op_data);
3038
3039         EXIT;
3040 }
3041
3042 int ll_prep_inode(struct inode **inode, struct req_capsule *pill,
3043                   struct super_block *sb, struct lookup_intent *it)
3044 {
3045         struct ll_sb_info *sbi = NULL;
3046         struct lustre_md md = { NULL };
3047         bool default_lmv_deleted = false;
3048         int rc;
3049
3050         ENTRY;
3051
3052         LASSERT(*inode || sb);
3053         sbi = sb ? ll_s2sbi(sb) : ll_i2sbi(*inode);
3054         rc = md_get_lustre_md(sbi->ll_md_exp, pill, sbi->ll_dt_exp,
3055                               sbi->ll_md_exp, &md);
3056         if (rc != 0)
3057                 GOTO(out, rc);
3058
3059         /*
3060          * clear default_lmv only if intent_getattr reply doesn't contain it.
3061          * but it needs to be done after iget, check this early because
3062          * ll_update_lsm_md() may change md.
3063          */
3064         if (it && (it->it_op & (IT_LOOKUP | IT_GETATTR)) &&
3065             S_ISDIR(md.body->mbo_mode) && !md.default_lmv)
3066                 default_lmv_deleted = true;
3067
3068         if (*inode) {
3069                 rc = ll_update_inode(*inode, &md);
3070                 if (rc != 0)
3071                         GOTO(out, rc);
3072         } else {
3073                 bool api32 = test_bit(LL_SBI_32BIT_API, sbi->ll_flags);
3074                 struct lu_fid *fid1 = &md.body->mbo_fid1;
3075
3076                 LASSERT(sb != NULL);
3077
3078                 /*
3079                  * At this point server returns to client's same fid as client
3080                  * generated for creating. So using ->fid1 is okay here.
3081                  */
3082                 if (!fid_is_sane(fid1)) {
3083                         CERROR("%s: Fid is insane "DFID"\n",
3084                                 sbi->ll_fsname, PFID(fid1));
3085                         GOTO(out, rc = -EINVAL);
3086                 }
3087
3088                 *inode = ll_iget(sb, cl_fid_build_ino(fid1, api32), &md);
3089                 if (IS_ERR(*inode)) {
3090                         lmd_clear_acl(&md);
3091                         rc = IS_ERR(*inode) ? PTR_ERR(*inode) : -ENOMEM;
3092                         *inode = NULL;
3093                         CERROR("new_inode -fatal: rc %d\n", rc);
3094                         GOTO(out, rc);
3095                 }
3096         }
3097
3098         /* Handling piggyback layout lock.
3099          * Layout lock can be piggybacked by getattr and open request.
3100          * The lsm can be applied to inode only if it comes with a layout lock
3101          * otherwise correct layout may be overwritten, for example:
3102          * 1. proc1: mdt returns a lsm but not granting layout
3103          * 2. layout was changed by another client
3104          * 3. proc2: refresh layout and layout lock granted
3105          * 4. proc1: to apply a stale layout */
3106         if (it != NULL && it->it_lock_mode != 0) {
3107                 struct lustre_handle lockh;
3108                 struct ldlm_lock *lock;
3109
3110                 lockh.cookie = it->it_lock_handle;
3111                 lock = ldlm_handle2lock(&lockh);
3112                 LASSERT(lock != NULL);
3113                 if (ldlm_has_layout(lock)) {
3114                         struct cl_object_conf conf;
3115
3116                         memset(&conf, 0, sizeof(conf));
3117                         conf.coc_opc = OBJECT_CONF_SET;
3118                         conf.coc_inode = *inode;
3119                         conf.coc_lock = lock;
3120                         conf.u.coc_layout = md.layout;
3121                         (void)ll_layout_conf(*inode, &conf);
3122                 }
3123                 LDLM_LOCK_PUT(lock);
3124         }
3125
3126         if (default_lmv_deleted)
3127                 ll_update_default_lsm_md(*inode, &md);
3128
3129         /* we may want to apply some policy for foreign file/dir */
3130         if (ll_sbi_has_foreign_symlink(sbi)) {
3131                 rc = ll_manage_foreign(*inode, &md);
3132                 if (rc < 0)
3133                         GOTO(out, rc);
3134         }
3135
3136         GOTO(out, rc = 0);
3137
3138 out:
3139         /* cleanup will be done if necessary */
3140         md_free_lustre_md(sbi->ll_md_exp, &md);
3141
3142         if (rc != 0 && it != NULL && it->it_op & IT_OPEN) {
3143                 ll_intent_drop_lock(it);
3144                 ll_open_cleanup(sb != NULL ? sb : (*inode)->i_sb, pill);
3145         }
3146
3147         return rc;
3148 }
3149
3150 int ll_obd_statfs(struct inode *inode, void __user *arg)
3151 {
3152         struct ll_sb_info *sbi = NULL;
3153         struct obd_export *exp;
3154         struct obd_ioctl_data *data = NULL;
3155         __u32 type;
3156         int len = 0, rc;
3157
3158         if (inode)
3159                 sbi = ll_i2sbi(inode);
3160         if (!sbi)
3161                 GOTO(out_statfs, rc = -EINVAL);
3162
3163         rc = obd_ioctl_getdata(&data, &len, arg);
3164         if (rc)
3165                 GOTO(out_statfs, rc);
3166
3167         if (!data->ioc_inlbuf1 || !data->ioc_inlbuf2 ||
3168             !data->ioc_pbuf1 || !data->ioc_pbuf2)
3169                 GOTO(out_statfs, rc = -EINVAL);
3170
3171         if (data->ioc_inllen1 != sizeof(__u32) ||
3172             data->ioc_inllen2 != sizeof(__u32) ||
3173             data->ioc_plen1 != sizeof(struct obd_statfs) ||
3174             data->ioc_plen2 != sizeof(struct obd_uuid))
3175                 GOTO(out_statfs, rc = -EINVAL);
3176
3177         memcpy(&type, data->ioc_inlbuf1, sizeof(__u32));
3178         if (type & LL_STATFS_LMV)
3179                 exp = sbi->ll_md_exp;
3180         else if (type & LL_STATFS_LOV)
3181                 exp = sbi->ll_dt_exp;
3182         else
3183                 GOTO(out_statfs, rc = -ENODEV);
3184
3185         rc = obd_iocontrol(IOC_OBD_STATFS, exp, len, data, NULL);
3186         if (rc)
3187                 GOTO(out_statfs, rc);
3188 out_statfs:
3189         OBD_FREE_LARGE(data, len);
3190         return rc;
3191 }
3192
3193 /*
3194  * this is normally called in ll_fini_md_op_data(), but sometimes it needs to
3195  * be called early to avoid deadlock.
3196  */
3197 void ll_unlock_md_op_lsm(struct md_op_data *op_data)
3198 {
3199         if (op_data->op_mea2_sem) {
3200                 up_read_non_owner(op_data->op_mea2_sem);
3201                 op_data->op_mea2_sem = NULL;
3202         }
3203
3204         if (op_data->op_mea1_sem) {
3205                 up_read_non_owner(op_data->op_mea1_sem);
3206                 op_data->op_mea1_sem = NULL;
3207         }
3208 }
3209
3210 /* this function prepares md_op_data hint for passing it down to MD stack. */
3211 struct md_op_data *ll_prep_md_op_data(struct md_op_data *op_data,
3212                                       struct inode *i1, struct inode *i2,
3213                                       const char *name, size_t namelen,
3214                                       __u32 mode, enum md_op_code opc,
3215                                       void *data)
3216 {
3217         struct llcrypt_name fname = { 0 };
3218         int rc;
3219
3220         LASSERT(i1 != NULL);
3221
3222         if (name == NULL) {
3223                 /* Do not reuse namelen for something else. */
3224                 if (namelen != 0)
3225                         return ERR_PTR(-EINVAL);
3226         } else {
3227                 if ((!IS_ENCRYPTED(i1) ||
3228                      (opc != LUSTRE_OPC_LOOKUP && opc != LUSTRE_OPC_CREATE)) &&
3229                     namelen > ll_i2sbi(i1)->ll_namelen)
3230                         return ERR_PTR(-ENAMETOOLONG);
3231
3232                 /* "/" is not valid name, but it's allowed */
3233                 if (!lu_name_is_valid_2(name, namelen) &&
3234                     strncmp("/", name, namelen) != 0)
3235                         return ERR_PTR(-EINVAL);
3236         }
3237
3238         if (op_data == NULL)
3239                 OBD_ALLOC_PTR(op_data);
3240
3241         if (op_data == NULL)
3242                 return ERR_PTR(-ENOMEM);
3243
3244         ll_i2gids(op_data->op_suppgids, i1, i2);
3245         op_data->op_fid1 = *ll_inode2fid(i1);
3246
3247         if (S_ISDIR(i1->i_mode)) {
3248                 down_read_non_owner(&ll_i2info(i1)->lli_lsm_sem);
3249                 op_data->op_mea1_sem = &ll_i2info(i1)->lli_lsm_sem;
3250                 op_data->op_mea1 = ll_i2info(i1)->lli_lsm_md;
3251                 op_data->op_default_mea1 = ll_i2info(i1)->lli_default_lsm_md;
3252         }
3253
3254         if (i2) {
3255                 op_data->op_fid2 = *ll_inode2fid(i2);
3256                 if (S_ISDIR(i2->i_mode)) {
3257                         if (i2 != i1) {
3258                                 /* i2 is typically a child of i1, and MUST be
3259                                  * further from the root to avoid deadlocks.
3260                                  */
3261                                 down_read_non_owner(&ll_i2info(i2)->lli_lsm_sem);
3262                                 op_data->op_mea2_sem =
3263                                                 &ll_i2info(i2)->lli_lsm_sem;
3264                         }
3265                         op_data->op_mea2 = ll_i2info(i2)->lli_lsm_md;
3266                 }
3267         } else {
3268                 fid_zero(&op_data->op_fid2);
3269         }
3270
3271         if (test_bit(LL_SBI_64BIT_HASH, ll_i2sbi(i1)->ll_flags))
3272                 op_data->op_cli_flags |= CLI_HASH64;
3273
3274         if (ll_need_32bit_api(ll_i2sbi(i1)))
3275                 op_data->op_cli_flags |= CLI_API32;
3276
3277         if (opc == LUSTRE_OPC_LOOKUP || opc == LUSTRE_OPC_CREATE) {
3278                 /* In case of lookup, ll_setup_filename() has already been
3279                  * called in ll_lookup_it(), so just take provided name.
3280                  */
3281                 fname.disk_name.name = (unsigned char *)name;
3282                 fname.disk_name.len = namelen;
3283         } else if (name && namelen) {
3284                 struct qstr dname = QSTR_INIT(name, namelen);
3285                 struct inode *dir;
3286                 struct lu_fid *pfid = NULL;
3287                 struct lu_fid fid;
3288                 int lookup;
3289
3290                 if (!S_ISDIR(i1->i_mode) && i2 && S_ISDIR(i2->i_mode)) {
3291                         /* special case when called from ll_link() */
3292                         dir = i2;
3293                         lookup = 0;
3294                 } else {
3295                         dir = i1;
3296                         lookup = (int)(opc == LUSTRE_OPC_ANY);
3297                 }
3298                 if (opc == LUSTRE_OPC_ANY && lookup)
3299                         pfid = &fid;
3300                 rc = ll_setup_filename(dir, &dname, lookup, &fname, pfid);
3301                 if (rc) {
3302                         ll_finish_md_op_data(op_data);
3303                         return ERR_PTR(rc);
3304                 }
3305                 if (pfid && !fid_is_zero(pfid)) {
3306                         if (i2 == NULL)
3307                                 op_data->op_fid2 = fid;
3308                         op_data->op_bias = MDS_FID_OP;
3309                 }
3310                 if (fname.disk_name.name &&
3311                     fname.disk_name.name != (unsigned char *)name)
3312                         /* op_data->op_name must be freed after use */
3313                         op_data->op_flags |= MF_OPNAME_KMALLOCED;
3314         }
3315
3316         /* In fact LUSTRE_OPC_LOOKUP, LUSTRE_OPC_OPEN
3317          * are LUSTRE_OPC_ANY
3318          */
3319         if (opc == LUSTRE_OPC_LOOKUP || opc == LUSTRE_OPC_OPEN)
3320                 op_data->op_code = LUSTRE_OPC_ANY;
3321         else
3322                 op_data->op_code = opc;
3323         op_data->op_name = fname.disk_name.name;
3324         op_data->op_namelen = fname.disk_name.len;
3325         op_data->op_mode = mode;
3326         op_data->op_mod_time = ktime_get_real_seconds();
3327         op_data->op_fsuid = from_kuid(&init_user_ns, current_fsuid());
3328         op_data->op_fsgid = from_kgid(&init_user_ns, current_fsgid());
3329         op_data->op_cap = current_cap();
3330         op_data->op_mds = 0;
3331         if ((opc == LUSTRE_OPC_CREATE) && (name != NULL) &&
3332              filename_is_volatile(name, namelen, &op_data->op_mds)) {
3333                 op_data->op_bias |= MDS_CREATE_VOLATILE;
3334         }
3335         op_data->op_data = data;
3336
3337         return op_data;
3338 }
3339
3340 void ll_finish_md_op_data(struct md_op_data *op_data)
3341 {
3342         ll_unlock_md_op_lsm(op_data);
3343         ll_security_release_secctx(op_data->op_file_secctx,
3344                                    op_data->op_file_secctx_size);
3345         if (op_data->op_flags & MF_OPNAME_KMALLOCED)
3346                 /* allocated via ll_setup_filename called
3347                  * from ll_prep_md_op_data
3348                  */
3349                 kfree(op_data->op_name);
3350         llcrypt_free_ctx(op_data->op_file_encctx, op_data->op_file_encctx_size);
3351         OBD_FREE_PTR(op_data);
3352 }
3353
3354 int ll_show_options(struct seq_file *seq, struct dentry *dentry)
3355 {
3356         struct ll_sb_info *sbi;
3357         int i;
3358
3359         LASSERT(seq && dentry);
3360         sbi = ll_s2sbi(dentry->d_sb);
3361
3362         if (test_bit(LL_SBI_NOLCK, sbi->ll_flags))
3363                 seq_puts(seq, "nolock");
3364
3365         for (i = 1; ll_sbi_flags_name[i].token != LL_SBI_NUM_MOUNT_OPT; i++) {
3366                 /* match_table in some cases has patterns for both enabled and
3367                  * disabled cases. Ignore 'no'xxx versions if bit is set.
3368                  */
3369                 if (test_bit(ll_sbi_flags_name[i].token, sbi->ll_flags) &&
3370                     strncmp(ll_sbi_flags_name[i].pattern, "no", 2)) {
3371                         if (ll_sbi_flags_name[i].token ==
3372                             LL_SBI_FOREIGN_SYMLINK) {
3373                                 seq_show_option(seq, "foreign_symlink",
3374                                                 sbi->ll_foreign_symlink_prefix);
3375                         } else {
3376                                 seq_printf(seq, ",%s",
3377                                            ll_sbi_flags_name[i].pattern);
3378                         }
3379
3380                         /* You can have either localflock or flock but not
3381                          * both. If localflock is set don't print flock or
3382                          * noflock.
3383                          */
3384                         if (ll_sbi_flags_name[i].token == LL_SBI_LOCALFLOCK)
3385                                 i += 2;
3386                 } else if (!test_bit(ll_sbi_flags_name[i].token, sbi->ll_flags) &&
3387                            !strncmp(ll_sbi_flags_name[i].pattern, "no", 2)) {
3388                         seq_printf(seq, ",%s",
3389                                    ll_sbi_flags_name[i].pattern);
3390                 }
3391         }
3392
3393         RETURN(0);
3394 }
3395
3396 /**
3397  * Get obd name by cmd, and copy out to user space
3398  */
3399 int ll_get_obd_name(struct inode *inode, unsigned int cmd, unsigned long arg)
3400 {
3401         struct ll_sb_info *sbi = ll_i2sbi(inode);
3402         struct obd_device *obd;
3403         ENTRY;
3404
3405         if (cmd == OBD_IOC_GETNAME_OLD || cmd == OBD_IOC_GETDTNAME)
3406                 obd = class_exp2obd(sbi->ll_dt_exp);
3407         else if (cmd == OBD_IOC_GETMDNAME)
3408                 obd = class_exp2obd(sbi->ll_md_exp);
3409         else
3410                 RETURN(-EINVAL);
3411
3412         if (!obd)
3413                 RETURN(-ENOENT);
3414
3415         if (copy_to_user((void __user *)arg, obd->obd_name,
3416                          strlen(obd->obd_name) + 1))
3417                 RETURN(-EFAULT);
3418
3419         RETURN(0);
3420 }
3421
3422 static char* ll_d_path(struct dentry *dentry, char *buf, int bufsize)
3423 {
3424         char *path = NULL;
3425
3426         struct path p;
3427
3428         p.dentry = dentry;
3429         p.mnt = current->fs->root.mnt;
3430         path_get(&p);
3431         path = d_path(&p, buf, bufsize);
3432         path_put(&p);
3433         return path;
3434 }
3435
3436 void ll_dirty_page_discard_warn(struct page *page, int ioret)
3437 {
3438         char *buf, *path = NULL;
3439         struct dentry *dentry = NULL;
3440         struct inode *inode = page->mapping->host;
3441
3442         /* this can be called inside spin lock so use GFP_ATOMIC. */
3443         buf = (char *)__get_free_page(GFP_ATOMIC);
3444         if (buf != NULL) {
3445                 dentry = d_find_alias(page->mapping->host);
3446                 if (dentry != NULL)
3447                         path = ll_d_path(dentry, buf, PAGE_SIZE);
3448         }
3449
3450         /* The below message is checked in recovery-small.sh test_24b */
3451         CDEBUG(D_WARNING,
3452                "%s: dirty page discard: %s/fid: "DFID"/%s may get corrupted "
3453                "(rc %d)\n", ll_i2sbi(inode)->ll_fsname,
3454                s2lsi(page->mapping->host->i_sb)->lsi_lmd->lmd_dev,
3455                PFID(ll_inode2fid(inode)),
3456                (path && !IS_ERR(path)) ? path : "", ioret);
3457
3458         if (dentry != NULL)
3459                 dput(dentry);
3460
3461         if (buf != NULL)
3462                 free_page((unsigned long)buf);
3463 }
3464
3465 ssize_t ll_copy_user_md(const struct lov_user_md __user *md,
3466                         struct lov_user_md **kbuf)
3467 {
3468         struct lov_user_md      lum;
3469         ssize_t                 lum_size;
3470         ENTRY;
3471
3472         if (copy_from_user(&lum, md, sizeof(lum)))
3473                 RETURN(-EFAULT);
3474
3475         lum_size = ll_lov_user_md_size(&lum);
3476         if (lum_size < 0)
3477                 RETURN(lum_size);
3478
3479         OBD_ALLOC_LARGE(*kbuf, lum_size);
3480         if (*kbuf == NULL)
3481                 RETURN(-ENOMEM);
3482
3483         if (copy_from_user(*kbuf, md, lum_size) != 0) {
3484                 OBD_FREE_LARGE(*kbuf, lum_size);
3485                 RETURN(-EFAULT);
3486         }
3487
3488         RETURN(lum_size);
3489 }
3490
3491 /*
3492  * Compute llite root squash state after a change of root squash
3493  * configuration setting or add/remove of a lnet nid
3494  */
3495 void ll_compute_rootsquash_state(struct ll_sb_info *sbi)
3496 {
3497         struct root_squash_info *squash = &sbi->ll_squash;
3498         int i;
3499         bool matched;
3500         struct lnet_process_id id;
3501
3502         /* Update norootsquash flag */
3503         spin_lock(&squash->rsi_lock);
3504         if (list_empty(&squash->rsi_nosquash_nids))
3505                 clear_bit(LL_SBI_NOROOTSQUASH, sbi->ll_flags);
3506         else {
3507                 /* Do not apply root squash as soon as one of our NIDs is
3508                  * in the nosquash_nids list */
3509                 matched = false;
3510                 i = 0;
3511                 while (LNetGetId(i++, &id) != -ENOENT) {
3512                         if (id.nid == LNET_NID_LO_0)
3513                                 continue;
3514                         if (cfs_match_nid(id.nid, &squash->rsi_nosquash_nids)) {
3515                                 matched = true;
3516                                 break;
3517                         }
3518                 }
3519                 if (matched)
3520                         set_bit(LL_SBI_NOROOTSQUASH, sbi->ll_flags);
3521                 else
3522                         clear_bit(LL_SBI_NOROOTSQUASH, sbi->ll_flags);
3523         }
3524         spin_unlock(&squash->rsi_lock);
3525 }
3526
3527 /**
3528  * Parse linkea content to extract information about a given hardlink
3529  *
3530  * \param[in]   ldata      - Initialized linkea data
3531  * \param[in]   linkno     - Link identifier
3532  * \param[out]  parent_fid - The entry's parent FID
3533  * \param[out]  ln         - Entry name destination buffer
3534  *
3535  * \retval 0 on success
3536  * \retval Appropriate negative error code on failure
3537  */
3538 static int ll_linkea_decode(struct linkea_data *ldata, unsigned int linkno,
3539                             struct lu_fid *parent_fid, struct lu_name *ln)
3540 {
3541         unsigned int    idx;
3542         int             rc;
3543         ENTRY;
3544
3545         rc = linkea_init_with_rec(ldata);
3546         if (rc < 0)
3547                 RETURN(rc);
3548
3549         if (linkno >= ldata->ld_leh->leh_reccount)
3550                 /* beyond last link */
3551                 RETURN(-ENODATA);
3552
3553         linkea_first_entry(ldata);
3554         for (idx = 0; ldata->ld_lee != NULL; idx++) {
3555                 linkea_entry_unpack(ldata->ld_lee, &ldata->ld_reclen, ln,
3556                                     parent_fid);
3557                 if (idx == linkno)
3558                         break;
3559
3560                 linkea_next_entry(ldata);
3561         }
3562
3563         if (idx < linkno)
3564                 RETURN(-ENODATA);
3565
3566         RETURN(0);
3567 }
3568
3569 /**
3570  * Get parent FID and name of an identified link. Operation is performed for
3571  * a given link number, letting the caller iterate over linkno to list one or
3572  * all links of an entry.
3573  *
3574  * \param[in]     file - File descriptor against which to perform the operation
3575  * \param[in,out] arg  - User-filled structure containing the linkno to operate
3576  *                       on and the available size. It is eventually filled with
3577  *                       the requested information or left untouched on error
3578  *
3579  * \retval - 0 on success
3580  * \retval - Appropriate negative error code on failure
3581  */
3582 int ll_getparent(struct file *file, struct getparent __user *arg)
3583 {
3584         struct inode            *inode = file_inode(file);
3585         struct linkea_data      *ldata;
3586         struct lu_buf            buf = LU_BUF_NULL;
3587         struct lu_name           ln;
3588         struct lu_fid            parent_fid;
3589         __u32                    linkno;
3590         __u32                    name_size;
3591         int                      rc;
3592
3593         ENTRY;
3594
3595         if (!capable(CAP_DAC_READ_SEARCH) &&
3596             !test_bit(LL_SBI_USER_FID2PATH, ll_i2sbi(inode)->ll_flags))
3597                 RETURN(-EPERM);
3598
3599         if (get_user(name_size, &arg->gp_name_size))
3600                 RETURN(-EFAULT);
3601
3602         if (get_user(linkno, &arg->gp_linkno))
3603                 RETURN(-EFAULT);
3604
3605         if (name_size > PATH_MAX)
3606                 RETURN(-EINVAL);
3607
3608         OBD_ALLOC(ldata, sizeof(*ldata));
3609         if (ldata == NULL)
3610                 RETURN(-ENOMEM);
3611
3612         rc = linkea_data_new(ldata, &buf);
3613         if (rc < 0)
3614                 GOTO(ldata_free, rc);
3615
3616         rc = ll_xattr_list(inode, XATTR_NAME_LINK, XATTR_TRUSTED_T, buf.lb_buf,
3617                            buf.lb_len, OBD_MD_FLXATTR);
3618         if (rc < 0)
3619                 GOTO(lb_free, rc);
3620
3621         rc = ll_linkea_decode(ldata, linkno, &parent_fid, &ln);
3622         if (rc < 0)
3623                 GOTO(lb_free, rc);
3624
3625         if (ln.ln_namelen >= name_size)
3626                 GOTO(lb_free, rc = -EOVERFLOW);
3627
3628         if (copy_to_user(&arg->gp_fid, &parent_fid, sizeof(arg->gp_fid)))
3629                 GOTO(lb_free, rc = -EFAULT);
3630
3631         if (copy_to_user(&arg->gp_name, ln.ln_name, ln.ln_namelen))
3632                 GOTO(lb_free, rc = -EFAULT);
3633
3634         if (put_user('\0', arg->gp_name + ln.ln_namelen))
3635                 GOTO(lb_free, rc = -EFAULT);
3636
3637 lb_free:
3638         lu_buf_free(&buf);
3639 ldata_free:
3640         OBD_FREE(ldata, sizeof(*ldata));
3641
3642         RETURN(rc);
3643 }