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/types.h>
  44 #include <libcfs/linux/linux-uuid.h>
  45 #include <linux/version.h>
  46 #include <linux/mm.h>
  47 #include <linux/user_namespace.h>
  48 #include <linux/delay.h>
  49 #include <linux/uidgid.h>
  50 #include <linux/fs_struct.h>
  51
  52 #ifndef HAVE_CPUS_READ_LOCK
  53 #include <libcfs/linux/linux-cpu.h>
  54 #endif
  55 #include <libcfs/linux/linux-misc.h>
  56 #include <uapi/linux/lustre/lustre_ioctl.h>
  57 #ifdef HAVE_UAPI_LINUX_MOUNT_H
  58 #include <uapi/linux/mount.h>
  59 #endif
  60
  61 #include <lustre_ha.h>
  62 #include <lustre_dlm.h>
  63 #include <lprocfs_status.h>
  64 #include <lustre_disk.h>
  65 #include <uapi/linux/lustre/lustre_param.h>
  66 #include <lustre_log.h>
  67 #include <cl_object.h>
  68 #include <obd_cksum.h>
  69 #include "llite_internal.h"
  70
  71 struct kmem_cache *ll_file_data_slab;
  72
  73 #ifndef log2
  74 #define log2(n) ffz(~(n))
  75 #endif
  76
  77 /**
  78  * If there is only one number of core visible to Lustre,
  79  * async readahead will be disabled, to avoid massive over
  80  * subscription, we use 1/2 of active cores as default max
  81  * async readahead requests.
  82  */
  83 static inline unsigned int ll_get_ra_async_max_active(void)
  84 {
  85         return cfs_cpt_weight(cfs_cpt_tab, CFS_CPT_ANY) >> 1;
  86 }
  87
  88 static struct ll_sb_info *ll_init_sbi(void)
  89 {
  90         struct ll_sb_info *sbi = NULL;
  91         unsigned long pages;
  92         unsigned long lru_page_max;
  93         struct sysinfo si;
  94         int rc;
  95         int i;
  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         for (i = 0; i <= LL_PROCESS_HIST_MAX; i++) {
 175                 spin_lock_init(&sbi->ll_rw_extents_info.pp_extents[i].
 176                                pp_r_hist.oh_lock);
 177                 spin_lock_init(&sbi->ll_rw_extents_info.pp_extents[i].
 178                                pp_w_hist.oh_lock);
 179         }
 180
 181         /* metadata statahead is enabled by default */
 182         sbi->ll_sa_running_max = LL_SA_RUNNING_DEF;
 183         sbi->ll_sa_max = LL_SA_RPC_DEF;
 184         atomic_set(&sbi->ll_sa_total, 0);
 185         atomic_set(&sbi->ll_sa_wrong, 0);
 186         atomic_set(&sbi->ll_sa_running, 0);
 187         atomic_set(&sbi->ll_agl_total, 0);
 188         set_bit(LL_SBI_AGL_ENABLED, sbi->ll_flags);
 189         set_bit(LL_SBI_FAST_READ, sbi->ll_flags);
 190         set_bit(LL_SBI_TINY_WRITE, sbi->ll_flags);
 191         set_bit(LL_SBI_PARALLEL_DIO, sbi->ll_flags);
 192         ll_sbi_set_encrypt(sbi, true);
 193
 194         /* root squash */
 195         sbi->ll_squash.rsi_uid = 0;
 196         sbi->ll_squash.rsi_gid = 0;
 197         INIT_LIST_HEAD(&sbi->ll_squash.rsi_nosquash_nids);
 198         spin_lock_init(&sbi->ll_squash.rsi_lock);
 199
 200         /* Per-filesystem file heat */
 201         sbi->ll_heat_decay_weight = SBI_DEFAULT_HEAT_DECAY_WEIGHT;
 202         sbi->ll_heat_period_second = SBI_DEFAULT_HEAT_PERIOD_SECOND;
 203
 204         /* Per-fs open heat level before requesting open lock */
 205         sbi->ll_oc_thrsh_count = SBI_DEFAULT_OPENCACHE_THRESHOLD_COUNT;
 206         sbi->ll_oc_max_ms = SBI_DEFAULT_OPENCACHE_THRESHOLD_MAX_MS;
 207         sbi->ll_oc_thrsh_ms = SBI_DEFAULT_OPENCACHE_THRESHOLD_MS;
 208         RETURN(sbi);
 209 out_destroy_ra:
 210         if (sbi->ll_foreign_symlink_prefix)
 211                 OBD_FREE(sbi->ll_foreign_symlink_prefix, sizeof("/mnt/"));
 212         if (sbi->ll_cache) {
 213                 cl_cache_decref(sbi->ll_cache);
 214                 sbi->ll_cache = NULL;
 215         }
 216         destroy_workqueue(sbi->ll_ra_info.ll_readahead_wq);
 217 out_pcc:
 218         pcc_super_fini(&sbi->ll_pcc_super);
 219 out_sbi:
 220         OBD_FREE_PTR(sbi);
 221         RETURN(ERR_PTR(rc));
 222 }
 223
 224 static void ll_free_sbi(struct super_block *sb)
 225 {
 226         struct ll_sb_info *sbi = ll_s2sbi(sb);
 227         ENTRY;
 228
 229         if (sbi != NULL) {
 230                 if (!list_empty(&sbi->ll_squash.rsi_nosquash_nids))
 231                         cfs_free_nidlist(&sbi->ll_squash.rsi_nosquash_nids);
 232                 if (sbi->ll_ra_info.ll_readahead_wq)
 233                         destroy_workqueue(sbi->ll_ra_info.ll_readahead_wq);
 234                 if (sbi->ll_cache != NULL) {
 235                         cl_cache_decref(sbi->ll_cache);
 236                         sbi->ll_cache = NULL;
 237                 }
 238                 if (sbi->ll_foreign_symlink_prefix) {
 239                         OBD_FREE(sbi->ll_foreign_symlink_prefix,
 240                                  sbi->ll_foreign_symlink_prefix_size);
 241                         sbi->ll_foreign_symlink_prefix = NULL;
 242                 }
 243                 if (sbi->ll_foreign_symlink_upcall) {
 244                         OBD_FREE(sbi->ll_foreign_symlink_upcall,
 245                                  strlen(sbi->ll_foreign_symlink_upcall) +
 246                                        1);
 247                         sbi->ll_foreign_symlink_upcall = NULL;
 248                 }
 249                 if (sbi->ll_foreign_symlink_upcall_items) {
 250                         int i;
 251                         int nb_items = sbi->ll_foreign_symlink_upcall_nb_items;
 252                         struct ll_foreign_symlink_upcall_item *items =
 253                                 sbi->ll_foreign_symlink_upcall_items;
 254
 255                         for (i = 0 ; i < nb_items; i++)
 256                                 if (items[i].type == STRING_TYPE)
 257                                         OBD_FREE(items[i].string,
 258                                                        items[i].size);
 259
 260                         OBD_FREE_LARGE(items, nb_items *
 261                                 sizeof(struct ll_foreign_symlink_upcall_item));
 262                         sbi->ll_foreign_symlink_upcall_items = NULL;
 263                 }
 264                 pcc_super_fini(&sbi->ll_pcc_super);
 265                 OBD_FREE(sbi, sizeof(*sbi));
 266         }
 267         EXIT;
 268 }
 269
 270 static int client_common_fill_super(struct super_block *sb, char *md, char *dt)
 271 {
 272         struct inode *root = NULL;
 273         struct ll_sb_info *sbi = ll_s2sbi(sb);
 274         struct obd_statfs *osfs = NULL;
 275         struct ptlrpc_request *request = NULL;
 276         struct obd_connect_data *data = NULL;
 277         struct obd_uuid *uuid;
 278         struct md_op_data *op_data;
 279         struct lustre_md lmd;
 280         u64 valid;
 281         int size, err, checksum;
 282         bool api32;
 283
 284         ENTRY;
 285         sbi->ll_md_obd = class_name2obd(md);
 286         if (!sbi->ll_md_obd) {
 287                 CERROR("MD %s: not setup or attached\n", md);
 288                 RETURN(-EINVAL);
 289         }
 290
 291         OBD_ALLOC_PTR(data);
 292         if (data == NULL)
 293                 RETURN(-ENOMEM);
 294
 295         OBD_ALLOC_PTR(osfs);
 296         if (osfs == NULL) {
 297                 OBD_FREE_PTR(data);
 298                 RETURN(-ENOMEM);
 299         }
 300
 301         /* pass client page size via ocd_grant_blkbits, the server should report
 302          * back its backend blocksize for grant calculation purpose */
 303         data->ocd_grant_blkbits = PAGE_SHIFT;
 304
 305         /* indicate MDT features supported by this client */
 306         data->ocd_connect_flags = OBD_CONNECT_IBITS    | OBD_CONNECT_NODEVOH  |
 307                                   OBD_CONNECT_ATTRFID  | OBD_CONNECT_GRANT |
 308                                   OBD_CONNECT_VERSION  | OBD_CONNECT_BRW_SIZE |
 309                                   OBD_CONNECT_SRVLOCK  |
 310                                   OBD_CONNECT_MDS_CAPA | OBD_CONNECT_OSS_CAPA |
 311                                   OBD_CONNECT_CANCELSET | OBD_CONNECT_FID     |
 312                                   OBD_CONNECT_AT       | OBD_CONNECT_LOV_V3   |
 313                                   OBD_CONNECT_VBR | OBD_CONNECT_FULL20 |
 314                                   OBD_CONNECT_64BITHASH |
 315                                   OBD_CONNECT_EINPROGRESS |
 316                                   OBD_CONNECT_JOBSTATS | OBD_CONNECT_LVB_TYPE |
 317                                   OBD_CONNECT_LAYOUTLOCK | OBD_CONNECT_PINGLESS|
 318                                   OBD_CONNECT_MAX_EASIZE |
 319                                   OBD_CONNECT_FLOCK_DEAD |
 320                                   OBD_CONNECT_DISP_STRIPE | OBD_CONNECT_LFSCK |
 321                                   OBD_CONNECT_OPEN_BY_FID |
 322                                   OBD_CONNECT_DIR_STRIPE |
 323                                   OBD_CONNECT_BULK_MBITS | OBD_CONNECT_CKSUM |
 324                                   OBD_CONNECT_SUBTREE |
 325                                   OBD_CONNECT_MULTIMODRPCS |
 326                                   OBD_CONNECT_GRANT_PARAM |
 327                                   OBD_CONNECT_GRANT_SHRINK |
 328                                   OBD_CONNECT_SHORTIO | OBD_CONNECT_FLAGS2;
 329
 330         data->ocd_connect_flags2 = OBD_CONNECT2_DIR_MIGRATE |
 331                                    OBD_CONNECT2_SUM_STATFS |
 332                                    OBD_CONNECT2_OVERSTRIPING |
 333                                    OBD_CONNECT2_FLR |
 334                                    OBD_CONNECT2_LOCK_CONVERT |
 335                                    OBD_CONNECT2_ARCHIVE_ID_ARRAY |
 336                                    OBD_CONNECT2_INC_XID |
 337                                    OBD_CONNECT2_LSOM |
 338                                    OBD_CONNECT2_ASYNC_DISCARD |
 339                                    OBD_CONNECT2_PCC |
 340                                    OBD_CONNECT2_CRUSH | OBD_CONNECT2_LSEEK |
 341                                    OBD_CONNECT2_GETATTR_PFID |
 342                                    OBD_CONNECT2_DOM_LVB |
 343                                    OBD_CONNECT2_REP_MBITS |
 344                                    OBD_CONNECT2_ATOMIC_OPEN_LOCK;
 345
 346 #ifdef HAVE_LRU_RESIZE_SUPPORT
 347         if (test_bit(LL_SBI_LRU_RESIZE, sbi->ll_flags))
 348                 data->ocd_connect_flags |= OBD_CONNECT_LRU_RESIZE;
 349 #endif
 350         data->ocd_connect_flags |= OBD_CONNECT_ACL_FLAGS;
 351
 352         data->ocd_cksum_types = obd_cksum_types_supported_client();
 353
 354         if (OBD_FAIL_CHECK(OBD_FAIL_MDC_LIGHTWEIGHT))
 355                 /* flag mdc connection as lightweight, only used for test
 356                  * purpose, use with care */
 357                 data->ocd_connect_flags |= OBD_CONNECT_LIGHTWEIGHT;
 358
 359         data->ocd_ibits_known = MDS_INODELOCK_FULL;
 360         data->ocd_version = LUSTRE_VERSION_CODE;
 361
 362         if (sb->s_flags & SB_RDONLY)
 363                 data->ocd_connect_flags |= OBD_CONNECT_RDONLY;
 364         if (test_bit(LL_SBI_USER_XATTR, sbi->ll_flags))
 365                 data->ocd_connect_flags |= OBD_CONNECT_XATTR;
 366
 367 #ifdef SB_NOSEC
 368         /* Setting this indicates we correctly support S_NOSEC (See kernel
 369          * commit 9e1f1de02c2275d7172e18dc4e7c2065777611bf)
 370          */
 371         sb->s_flags |= SB_NOSEC;
 372 #endif
 373         sbi->ll_fop = ll_select_file_operations(sbi);
 374
 375         /* always ping even if server suppress_pings */
 376         if (test_bit(LL_SBI_ALWAYS_PING, sbi->ll_flags))
 377                 data->ocd_connect_flags &= ~OBD_CONNECT_PINGLESS;
 378
 379         obd_connect_set_secctx(data);
 380         if (ll_sbi_has_encrypt(sbi))
 381                 obd_connect_set_enc(data);
 382
 383 #if defined(CONFIG_SECURITY)
 384         data->ocd_connect_flags2 |= OBD_CONNECT2_SELINUX_POLICY;
 385 #endif
 386
 387         data->ocd_brw_size = MD_MAX_BRW_SIZE;
 388
 389         err = obd_connect(NULL, &sbi->ll_md_exp, sbi->ll_md_obd,
 390                           &sbi->ll_sb_uuid, data, sbi->ll_cache);
 391         if (err == -EBUSY) {
 392                 LCONSOLE_ERROR_MSG(0x14f, "An MDT (md %s) is performing "
 393                                    "recovery, of which this client is not a "
 394                                    "part. Please wait for recovery to complete,"
 395                                    " abort, or time out.\n", md);
 396                 GOTO(out, err);
 397         } else if (err) {
 398                 CERROR("cannot connect to %s: rc = %d\n", md, err);
 399                 GOTO(out, err);
 400         }
 401
 402         sbi->ll_md_exp->exp_connect_data = *data;
 403
 404         err = obd_fid_init(sbi->ll_md_exp->exp_obd, sbi->ll_md_exp,
 405                            LUSTRE_SEQ_METADATA);
 406         if (err) {
 407                 CERROR("%s: Can't init metadata layer FID infrastructure, "
 408                        "rc = %d\n", sbi->ll_md_exp->exp_obd->obd_name, err);
 409                 GOTO(out_md, err);
 410         }
 411
 412         /* For mount, we only need fs info from MDT0, and also in DNE, it
 413          * can make sure the client can be mounted as long as MDT0 is
 414          * avaible */
 415         err = obd_statfs(NULL, sbi->ll_md_exp, osfs,
 416                         ktime_get_seconds() - sbi->ll_statfs_max_age,
 417                         OBD_STATFS_FOR_MDT0);
 418         if (err)
 419                 GOTO(out_md_fid, err);
 420
 421         /* This needs to be after statfs to ensure connect has finished.
 422          * Note that "data" does NOT contain the valid connect reply.
 423          * If connecting to a 1.8 server there will be no LMV device, so
 424          * we can access the MDC export directly and exp_connect_flags will
 425          * be non-zero, but if accessing an upgraded 2.1 server it will
 426          * have the correct flags filled in.
 427          * XXX: fill in the LMV exp_connect_flags from MDC(s). */
 428         valid = exp_connect_flags(sbi->ll_md_exp) & CLIENT_CONNECT_MDT_REQD;
 429         if (exp_connect_flags(sbi->ll_md_exp) != 0 &&
 430             valid != CLIENT_CONNECT_MDT_REQD) {
 431                 char *buf;
 432
 433                 OBD_ALLOC_WAIT(buf, PAGE_SIZE);
 434                 obd_connect_flags2str(buf, PAGE_SIZE,
 435                                       valid ^ CLIENT_CONNECT_MDT_REQD, 0, ",");
 436                 LCONSOLE_ERROR_MSG(0x170, "Server %s does not support "
 437                                    "feature(s) needed for correct operation "
 438                                    "of this client (%s). Please upgrade "
 439                                    "server or downgrade client.\n",
 440                                    sbi->ll_md_exp->exp_obd->obd_name, buf);
 441                 OBD_FREE(buf, PAGE_SIZE);
 442                 GOTO(out_md_fid, err = -EPROTO);
 443         }
 444
 445         size = sizeof(*data);
 446         err = obd_get_info(NULL, sbi->ll_md_exp, sizeof(KEY_CONN_DATA),
 447                            KEY_CONN_DATA,  &size, data);
 448         if (err) {
 449                 CERROR("%s: Get connect data failed: rc = %d\n",
 450                        sbi->ll_md_exp->exp_obd->obd_name, err);
 451                 GOTO(out_md_fid, err);
 452         }
 453
 454         LASSERT(osfs->os_bsize);
 455         sb->s_blocksize = osfs->os_bsize;
 456         sb->s_blocksize_bits = log2(osfs->os_bsize);
 457         sb->s_magic = LL_SUPER_MAGIC;
 458         sb->s_maxbytes = MAX_LFS_FILESIZE;
 459         sbi->ll_namelen = osfs->os_namelen;
 460         sbi->ll_mnt.mnt = current->fs->root.mnt;
 461
 462         if (test_bit(LL_SBI_USER_XATTR, sbi->ll_flags) &&
 463             !(data->ocd_connect_flags & OBD_CONNECT_XATTR)) {
 464                 LCONSOLE_INFO("Disabling user_xattr feature because "
 465                               "it is not supported on the server\n");
 466                 clear_bit(LL_SBI_USER_XATTR, sbi->ll_flags);
 467         }
 468
 469         if (data->ocd_connect_flags & OBD_CONNECT_ACL) {
 470 #ifdef SB_POSIXACL
 471                 sb->s_flags |= SB_POSIXACL;
 472 #endif
 473                 set_bit(LL_SBI_ACL, sbi->ll_flags);
 474         } else {
 475                 LCONSOLE_INFO("client wants to enable acl, but mdt not!\n");
 476 #ifdef SB_POSIXACL
 477                 sb->s_flags &= ~SB_POSIXACL;
 478 #endif
 479                 clear_bit(LL_SBI_ACL, sbi->ll_flags);
 480         }
 481
 482         if (data->ocd_connect_flags & OBD_CONNECT_64BITHASH)
 483                 set_bit(LL_SBI_64BIT_HASH, sbi->ll_flags);
 484
 485         if (data->ocd_connect_flags & OBD_CONNECT_LAYOUTLOCK)
 486                 set_bit(LL_SBI_LAYOUT_LOCK, sbi->ll_flags);
 487
 488         if (obd_connect_has_secctx(data))
 489                 set_bit(LL_SBI_FILE_SECCTX, sbi->ll_flags);
 490
 491         if (ll_sbi_has_encrypt(sbi) && !obd_connect_has_enc(data)) {
 492                 if (ll_sbi_has_test_dummy_encryption(sbi))
 493                         LCONSOLE_WARN("%s: server %s does not support encryption feature, encryption deactivated.\n",
 494                                       sbi->ll_fsname,
 495                                       sbi->ll_md_exp->exp_obd->obd_name);
 496                 ll_sbi_set_encrypt(sbi, false);
 497         }
 498
 499         if (data->ocd_ibits_known & MDS_INODELOCK_XATTR) {
 500                 if (!(data->ocd_connect_flags & OBD_CONNECT_MAX_EASIZE)) {
 501                         LCONSOLE_INFO("%s: disabling xattr cache due to "
 502                                       "unknown maximum xattr size.\n", dt);
 503                 } else if (!sbi->ll_xattr_cache_set) {
 504                         /* If xattr_cache is already set (no matter 0 or 1)
 505                          * during processing llog, it won't be enabled here. */
 506                         set_bit(LL_SBI_XATTR_CACHE, sbi->ll_flags);
 507                         sbi->ll_xattr_cache_enabled = 1;
 508                 }
 509         }
 510
 511         sbi->ll_dt_obd = class_name2obd(dt);
 512         if (!sbi->ll_dt_obd) {
 513                 CERROR("DT %s: not setup or attached\n", dt);
 514                 GOTO(out_md_fid, err = -ENODEV);
 515         }
 516
 517         /* pass client page size via ocd_grant_blkbits, the server should report
 518          * back its backend blocksize for grant calculation purpose */
 519         data->ocd_grant_blkbits = PAGE_SHIFT;
 520
 521         /* indicate OST features supported by this client */
 522         data->ocd_connect_flags = OBD_CONNECT_GRANT | OBD_CONNECT_VERSION |
 523                                   OBD_CONNECT_REQPORTAL | OBD_CONNECT_BRW_SIZE |
 524                                   OBD_CONNECT_CANCELSET | OBD_CONNECT_FID |
 525                                   OBD_CONNECT_SRVLOCK |
 526                                   OBD_CONNECT_AT | OBD_CONNECT_OSS_CAPA |
 527                                   OBD_CONNECT_VBR | OBD_CONNECT_FULL20 |
 528                                   OBD_CONNECT_64BITHASH | OBD_CONNECT_MAXBYTES |
 529                                   OBD_CONNECT_EINPROGRESS |
 530                                   OBD_CONNECT_JOBSTATS | OBD_CONNECT_LVB_TYPE |
 531                                   OBD_CONNECT_LAYOUTLOCK |
 532                                   OBD_CONNECT_PINGLESS | OBD_CONNECT_LFSCK |
 533                                   OBD_CONNECT_BULK_MBITS | OBD_CONNECT_SHORTIO |
 534                                   OBD_CONNECT_FLAGS2 | OBD_CONNECT_GRANT_SHRINK;
 535         data->ocd_connect_flags2 = OBD_CONNECT2_LOCKAHEAD |
 536                                    OBD_CONNECT2_INC_XID | OBD_CONNECT2_LSEEK |
 537                                    OBD_CONNECT2_REP_MBITS;
 538
 539         if (!OBD_FAIL_CHECK(OBD_FAIL_OSC_CONNECT_GRANT_PARAM))
 540                 data->ocd_connect_flags |= OBD_CONNECT_GRANT_PARAM;
 541
 542         /* OBD_CONNECT_CKSUM should always be set, even if checksums are
 543          * disabled by default, because it can still be enabled on the
 544          * fly via /sys. As a consequence, we still need to come to an
 545          * agreement on the supported algorithms at connect time
 546          */
 547         data->ocd_connect_flags |= OBD_CONNECT_CKSUM;
 548
 549         if (OBD_FAIL_CHECK(OBD_FAIL_OSC_CKSUM_ADLER_ONLY))
 550                 data->ocd_cksum_types = OBD_CKSUM_ADLER;
 551         else
 552                 data->ocd_cksum_types = obd_cksum_types_supported_client();
 553
 554 #ifdef HAVE_LRU_RESIZE_SUPPORT
 555         data->ocd_connect_flags |= OBD_CONNECT_LRU_RESIZE;
 556 #endif
 557         /* always ping even if server suppress_pings */
 558         if (test_bit(LL_SBI_ALWAYS_PING, sbi->ll_flags))
 559                 data->ocd_connect_flags &= ~OBD_CONNECT_PINGLESS;
 560
 561         if (ll_sbi_has_encrypt(sbi))
 562                 obd_connect_set_enc(data);
 563
 564         CDEBUG(D_RPCTRACE, "ocd_connect_flags: %#llx ocd_version: %d "
 565                "ocd_grant: %d\n", data->ocd_connect_flags,
 566                data->ocd_version, data->ocd_grant);
 567
 568         sbi->ll_dt_obd->obd_upcall.onu_owner = &sbi->ll_lco;
 569         sbi->ll_dt_obd->obd_upcall.onu_upcall = cl_ocd_update;
 570
 571         data->ocd_brw_size = DT_MAX_BRW_SIZE;
 572
 573         err = obd_connect(NULL, &sbi->ll_dt_exp, sbi->ll_dt_obd,
 574                           &sbi->ll_sb_uuid, data, sbi->ll_cache);
 575         if (err == -EBUSY) {
 576                 LCONSOLE_ERROR_MSG(0x150, "An OST (dt %s) is performing "
 577                                    "recovery, of which this client is not a "
 578                                    "part.  Please wait for recovery to "
 579                                    "complete, abort, or time out.\n", dt);
 580                 GOTO(out_md, err);
 581         } else if (err) {
 582                 CERROR("%s: Cannot connect to %s: rc = %d\n",
 583                        sbi->ll_dt_exp->exp_obd->obd_name, dt, err);
 584                 GOTO(out_md, err);
 585         }
 586
 587         if (ll_sbi_has_encrypt(sbi) &&
 588             !obd_connect_has_enc(&sbi->ll_dt_obd->u.lov.lov_ocd)) {
 589                 if (ll_sbi_has_test_dummy_encryption(sbi))
 590                         LCONSOLE_WARN("%s: server %s does not support encryption feature, encryption deactivated.\n",
 591                                       sbi->ll_fsname, dt);
 592                 ll_sbi_set_encrypt(sbi, false);
 593         } else if (ll_sbi_has_test_dummy_encryption(sbi)) {
 594                 LCONSOLE_WARN("Test dummy encryption mode enabled\n");
 595         }
 596
 597         sbi->ll_dt_exp->exp_connect_data = *data;
 598
 599         /* Don't change value if it was specified in the config log */
 600         if (sbi->ll_ra_info.ra_max_read_ahead_whole_pages == -1) {
 601                 sbi->ll_ra_info.ra_max_read_ahead_whole_pages =
 602                         max_t(unsigned long, SBI_DEFAULT_READ_AHEAD_WHOLE_MAX,
 603                               (data->ocd_brw_size >> PAGE_SHIFT));
 604                 if (sbi->ll_ra_info.ra_max_read_ahead_whole_pages >
 605                     sbi->ll_ra_info.ra_max_pages_per_file)
 606                         sbi->ll_ra_info.ra_max_read_ahead_whole_pages =
 607                                 sbi->ll_ra_info.ra_max_pages_per_file;
 608         }
 609
 610         err = obd_fid_init(sbi->ll_dt_exp->exp_obd, sbi->ll_dt_exp,
 611                            LUSTRE_SEQ_METADATA);
 612         if (err) {
 613                 CERROR("%s: Can't init data layer FID infrastructure, "
 614                        "rc = %d\n", sbi->ll_dt_exp->exp_obd->obd_name, err);
 615                 GOTO(out_dt, err);
 616         }
 617
 618         mutex_lock(&sbi->ll_lco.lco_lock);
 619         sbi->ll_lco.lco_flags = data->ocd_connect_flags;
 620         sbi->ll_lco.lco_md_exp = sbi->ll_md_exp;
 621         sbi->ll_lco.lco_dt_exp = sbi->ll_dt_exp;
 622         mutex_unlock(&sbi->ll_lco.lco_lock);
 623
 624         fid_zero(&sbi->ll_root_fid);
 625         err = md_get_root(sbi->ll_md_exp, get_mount_fileset(sb),
 626                            &sbi->ll_root_fid);
 627         if (err) {
 628                 CERROR("cannot mds_connect: rc = %d\n", err);
 629                 GOTO(out_lock_cn_cb, err);
 630         }
 631         if (!fid_is_sane(&sbi->ll_root_fid)) {
 632                 CERROR("%s: Invalid root fid "DFID" during mount\n",
 633                        sbi->ll_md_exp->exp_obd->obd_name,
 634                        PFID(&sbi->ll_root_fid));
 635                 GOTO(out_lock_cn_cb, err = -EINVAL);
 636         }
 637         CDEBUG(D_SUPER, "rootfid "DFID"\n", PFID(&sbi->ll_root_fid));
 638
 639         sb->s_op = &lustre_super_operations;
 640         sb->s_xattr = ll_xattr_handlers;
 641 #if THREAD_SIZE >= 8192 /*b=17630*/
 642         sb->s_export_op = &lustre_export_operations;
 643 #endif
 644 #ifdef HAVE_LUSTRE_CRYPTO
 645         llcrypt_set_ops(sb, &lustre_cryptops);
 646 #endif
 647
 648         /* make root inode
 649          * XXX: move this to after cbd setup? */
 650         valid = OBD_MD_FLGETATTR | OBD_MD_FLBLOCKS | OBD_MD_FLMODEASIZE;
 651         if (test_bit(LL_SBI_ACL, sbi->ll_flags))
 652                 valid |= OBD_MD_FLACL;
 653
 654         OBD_ALLOC_PTR(op_data);
 655         if (op_data == NULL)
 656                 GOTO(out_lock_cn_cb, err = -ENOMEM);
 657
 658         op_data->op_fid1 = sbi->ll_root_fid;
 659         op_data->op_mode = 0;
 660         op_data->op_valid = valid;
 661
 662         err = md_getattr(sbi->ll_md_exp, op_data, &request);
 663
 664         OBD_FREE_PTR(op_data);
 665         if (err) {
 666                 CERROR("%s: md_getattr failed for root: rc = %d\n",
 667                        sbi->ll_md_exp->exp_obd->obd_name, err);
 668                 GOTO(out_lock_cn_cb, err);
 669         }
 670
 671         err = md_get_lustre_md(sbi->ll_md_exp, &request->rq_pill,
 672                                sbi->ll_dt_exp, sbi->ll_md_exp, &lmd);
 673         if (err) {
 674                 CERROR("failed to understand root inode md: rc = %d\n", err);
 675                 ptlrpc_req_finished(request);
 676                 GOTO(out_lock_cn_cb, err);
 677         }
 678
 679         LASSERT(fid_is_sane(&sbi->ll_root_fid));
 680         api32 = test_bit(LL_SBI_32BIT_API, sbi->ll_flags);
 681         root = ll_iget(sb, cl_fid_build_ino(&sbi->ll_root_fid, api32), &lmd);
 682         md_free_lustre_md(sbi->ll_md_exp, &lmd);
 683         ptlrpc_req_finished(request);
 684
 685         if (IS_ERR(root)) {
 686                 lmd_clear_acl(&lmd);
 687                 err = IS_ERR(root) ? PTR_ERR(root) : -EBADF;
 688                 root = NULL;
 689                 CERROR("%s: bad ll_iget() for root: rc = %d\n",
 690                        sbi->ll_fsname, err);
 691                 GOTO(out_root, err);
 692         }
 693
 694         checksum = test_bit(LL_SBI_CHECKSUM, sbi->ll_flags);
 695         if (sbi->ll_checksum_set) {
 696                 err = obd_set_info_async(NULL, sbi->ll_dt_exp,
 697                                          sizeof(KEY_CHECKSUM), KEY_CHECKSUM,
 698                                          sizeof(checksum), &checksum, NULL);
 699                 if (err) {
 700                         CERROR("%s: Set checksum failed: rc = %d\n",
 701                                sbi->ll_dt_exp->exp_obd->obd_name, err);
 702                         GOTO(out_root, err);
 703                 }
 704         }
 705         cl_sb_init(sb);
 706
 707         sb->s_root = d_make_root(root);
 708         if (sb->s_root == NULL) {
 709                 err = -ENOMEM;
 710                 CERROR("%s: can't make root dentry: rc = %d\n",
 711                        sbi->ll_fsname, err);
 712                 GOTO(out_root, err);
 713         }
 714
 715         sbi->ll_sdev_orig = sb->s_dev;
 716
 717         /* We set sb->s_dev equal on all lustre clients in order to support
 718          * NFS export clustering.  NFSD requires that the FSID be the same
 719          * on all clients. */
 720         /* s_dev is also used in lt_compare() to compare two fs, but that is
 721          * only a node-local comparison. */
 722         uuid = obd_get_uuid(sbi->ll_md_exp);
 723         if (uuid != NULL)
 724                 sb->s_dev = get_uuid2int(uuid->uuid, strlen(uuid->uuid));
 725
 726         if (data != NULL)
 727                 OBD_FREE_PTR(data);
 728         if (osfs != NULL)
 729                 OBD_FREE_PTR(osfs);
 730
 731         if (sbi->ll_dt_obd) {
 732                 err = sysfs_create_link(&sbi->ll_kset.kobj,
 733                                         &sbi->ll_dt_obd->obd_kset.kobj,
 734                                         sbi->ll_dt_obd->obd_type->typ_name);
 735                 if (err < 0) {
 736                         CERROR("%s: could not register %s in llite: rc = %d\n",
 737                                dt, sbi->ll_fsname, err);
 738                         err = 0;
 739                 }
 740         }
 741
 742         if (sbi->ll_md_obd) {
 743                 err = sysfs_create_link(&sbi->ll_kset.kobj,
 744                                         &sbi->ll_md_obd->obd_kset.kobj,
 745                                         sbi->ll_md_obd->obd_type->typ_name);
 746                 if (err < 0) {
 747                         CERROR("%s: could not register %s in llite: rc = %d\n",
 748                                md, sbi->ll_fsname, err);
 749                         err = 0;
 750                 }
 751         }
 752
 753         RETURN(err);
 754 out_root:
 755         iput(root);
 756 out_lock_cn_cb:
 757         obd_fid_fini(sbi->ll_dt_exp->exp_obd);
 758 out_dt:
 759         obd_disconnect(sbi->ll_dt_exp);
 760         sbi->ll_dt_exp = NULL;
 761         sbi->ll_dt_obd = NULL;
 762 out_md_fid:
 763         obd_fid_fini(sbi->ll_md_exp->exp_obd);
 764 out_md:
 765         obd_disconnect(sbi->ll_md_exp);
 766         sbi->ll_md_exp = NULL;
 767         sbi->ll_md_obd = NULL;
 768 out:
 769         if (data != NULL)
 770                 OBD_FREE_PTR(data);
 771         if (osfs != NULL)
 772                 OBD_FREE_PTR(osfs);
 773         return err;
 774 }
 775
 776 int ll_get_max_mdsize(struct ll_sb_info *sbi, int *lmmsize)
 777 {
 778         int size, rc;
 779
 780         size = sizeof(*lmmsize);
 781         rc = obd_get_info(NULL, sbi->ll_dt_exp, sizeof(KEY_MAX_EASIZE),
 782                           KEY_MAX_EASIZE, &size, lmmsize);
 783         if (rc != 0) {
 784                 CERROR("%s: cannot get max LOV EA size: rc = %d\n",
 785                        sbi->ll_dt_exp->exp_obd->obd_name, rc);
 786                 RETURN(rc);
 787         }
 788
 789         CDEBUG(D_INFO, "max LOV ea size: %d\n", *lmmsize);
 790
 791         size = sizeof(int);
 792         rc = obd_get_info(NULL, sbi->ll_md_exp, sizeof(KEY_MAX_EASIZE),
 793                           KEY_MAX_EASIZE, &size, lmmsize);
 794         if (rc)
 795                 CERROR("Get max mdsize error rc %d\n", rc);
 796
 797         CDEBUG(D_INFO, "max LMV ea size: %d\n", *lmmsize);
 798
 799         RETURN(rc);
 800 }
 801
 802 /**
 803  * Get the value of the default_easize parameter.
 804  *
 805  * \see client_obd::cl_default_mds_easize
 806  *
 807  * \param[in] sbi       superblock info for this filesystem
 808  * \param[out] lmmsize  pointer to storage location for value
 809  *
 810  * \retval 0            on success
 811  * \retval negative     negated errno on failure
 812  */
 813 int ll_get_default_mdsize(struct ll_sb_info *sbi, int *lmmsize)
 814 {
 815         int size, rc;
 816
 817         size = sizeof(int);
 818         rc = obd_get_info(NULL, sbi->ll_md_exp, sizeof(KEY_DEFAULT_EASIZE),
 819                          KEY_DEFAULT_EASIZE, &size, lmmsize);
 820         if (rc)
 821                 CERROR("Get default mdsize error rc %d\n", rc);
 822
 823         RETURN(rc);
 824 }
 825
 826 /**
 827  * Set the default_easize parameter to the given value.
 828  *
 829  * \see client_obd::cl_default_mds_easize
 830  *
 831  * \param[in] sbi       superblock info for this filesystem
 832  * \param[in] lmmsize   the size to set
 833  *
 834  * \retval 0            on success
 835  * \retval negative     negated errno on failure
 836  */
 837 int ll_set_default_mdsize(struct ll_sb_info *sbi, int lmmsize)
 838 {
 839         int rc;
 840
 841         if (lmmsize < sizeof(struct lov_mds_md) ||
 842             lmmsize > OBD_MAX_DEFAULT_EA_SIZE)
 843                 return -EINVAL;
 844
 845         rc = obd_set_info_async(NULL, sbi->ll_md_exp,
 846                                 sizeof(KEY_DEFAULT_EASIZE), KEY_DEFAULT_EASIZE,
 847                                 sizeof(int), &lmmsize, NULL);
 848
 849         RETURN(rc);
 850 }
 851
 852 static void client_common_put_super(struct super_block *sb)
 853 {
 854         struct ll_sb_info *sbi = ll_s2sbi(sb);
 855         ENTRY;
 856
 857         cl_sb_fini(sb);
 858
 859         obd_fid_fini(sbi->ll_dt_exp->exp_obd);
 860         obd_disconnect(sbi->ll_dt_exp);
 861         sbi->ll_dt_exp = NULL;
 862
 863         ll_debugfs_unregister_super(sb);
 864
 865         obd_fid_fini(sbi->ll_md_exp->exp_obd);
 866         obd_disconnect(sbi->ll_md_exp);
 867         sbi->ll_md_exp = NULL;
 868
 869         EXIT;
 870 }
 871
 872 void ll_kill_super(struct super_block *sb)
 873 {
 874         struct ll_sb_info *sbi;
 875         ENTRY;
 876
 877         /* not init sb ?*/
 878         if (!(sb->s_flags & SB_ACTIVE))
 879                 return;
 880
 881         sbi = ll_s2sbi(sb);
 882         /* we need restore s_dev from changed for clustred NFS before put_super
 883          * because new kernels have cached s_dev and change sb->s_dev in
 884          * put_super not affected real removing devices */
 885         if (sbi) {
 886                 sb->s_dev = sbi->ll_sdev_orig;
 887
 888                 /* wait running statahead threads to quit */
 889                 while (atomic_read(&sbi->ll_sa_running) > 0)
 890                         schedule_timeout_uninterruptible(
 891                                 cfs_time_seconds(1) >> 3);
 892         }
 893
 894         EXIT;
 895 }
 896
 897 /* Since we use this table for ll_sbi_flags_seq_show make
 898  * sure what you want displayed for a specific token that
 899  * is listed more than once below be listed first. For
 900  * example we want "checksum" displayed, not "nochecksum"
 901  * for the sbi_flags.
 902  */
 903 static const match_table_t ll_sbi_flags_name = {
 904         {LL_SBI_NOLCK,                  "nolock"},
 905         {LL_SBI_CHECKSUM,               "checksum"},
 906         {LL_SBI_CHECKSUM,               "nochecksum"},
 907         {LL_SBI_LOCALFLOCK,             "localflock"},
 908         {LL_SBI_FLOCK,                  "flock"},
 909         {LL_SBI_FLOCK,                  "noflock"},
 910         {LL_SBI_USER_XATTR,             "user_xattr"},
 911         {LL_SBI_USER_XATTR,             "nouser_xattr"},
 912         {LL_SBI_LRU_RESIZE,             "lruresize"},
 913         {LL_SBI_LRU_RESIZE,             "nolruresize"},
 914         {LL_SBI_LAZYSTATFS,             "lazystatfs"},
 915         {LL_SBI_LAZYSTATFS,             "nolazystatfs"},
 916         {LL_SBI_32BIT_API,              "32bitapi"},
 917         {LL_SBI_USER_FID2PATH,          "user_fid2path"},
 918         {LL_SBI_USER_FID2PATH,          "nouser_fid2path"},
 919         {LL_SBI_VERBOSE,                "verbose"},
 920         {LL_SBI_VERBOSE,                "noverbose"},
 921         {LL_SBI_ALWAYS_PING,            "always_ping"},
 922         {LL_SBI_TEST_DUMMY_ENCRYPTION,  "test_dummy_encryption"},
 923         {LL_SBI_ENCRYPT,                "encrypt"},
 924         {LL_SBI_ENCRYPT,                "noencrypt"},
 925         {LL_SBI_FOREIGN_SYMLINK,        "foreign_symlink=%s"},
 926         {LL_SBI_NUM_MOUNT_OPT,          NULL},
 927
 928         {LL_SBI_ACL,                    "acl"},
 929         {LL_SBI_AGL_ENABLED,            "agl"},
 930         {LL_SBI_64BIT_HASH,             "64bit_hash"},
 931         {LL_SBI_LAYOUT_LOCK,            "layout"},
 932         {LL_SBI_XATTR_CACHE,            "xattr_cache"},
 933         {LL_SBI_NOROOTSQUASH,           "norootsquash"},
 934         {LL_SBI_FAST_READ,              "fast_read"},
 935         {LL_SBI_FILE_SECCTX,            "file_secctx"},
 936         {LL_SBI_TINY_WRITE,             "tiny_write"},
 937         {LL_SBI_FILE_HEAT,              "file_heat"},
 938         {LL_SBI_PARALLEL_DIO,           "parallel_dio"},
 939 };
 940
 941 int ll_sbi_flags_seq_show(struct seq_file *m, void *v)
 942 {
 943         struct super_block *sb = m->private;
 944         int i;
 945
 946         for (i = 0; i < LL_SBI_NUM_FLAGS; i++) {
 947                 int j;
 948
 949                 if (!test_bit(i, ll_s2sbi(sb)->ll_flags))
 950                         continue;
 951
 952                 for (j = 0; j < ARRAY_SIZE(ll_sbi_flags_name); j++) {
 953                         if (ll_sbi_flags_name[j].token == i &&
 954                             ll_sbi_flags_name[j].pattern) {
 955                                 seq_printf(m, "%s ",
 956                                            ll_sbi_flags_name[j].pattern);
 957                                 break;
 958                         }
 959                 }
 960         }
 961         seq_puts(m, "\b\n");
 962         return 0;
 963 }
 964
 965 /* non-client-specific mount options are parsed in lmd_parse */
 966 static int ll_options(char *options, struct super_block *sb)
 967 {
 968         struct ll_sb_info *sbi = ll_s2sbi(sb);
 969         char *s2, *s1, *opts;
 970
 971         ENTRY;
 972         if (!options)
 973                 RETURN(0);
 974
 975         /* Don't stomp on lmd_opts */
 976         opts = kstrdup(options, GFP_KERNEL);
 977         if (!opts)
 978                 RETURN(-ENOMEM);
 979         s1 = opts;
 980         s2 = opts;
 981
 982         CDEBUG(D_CONFIG, "Parsing opts %s\n", options);
 983
 984         while ((s1 = strsep(&opts, ",")) != NULL) {
 985                 substring_t args[MAX_OPT_ARGS];
 986                 bool turn_off = false;
 987                 int token;
 988
 989                 if (!*s1)
 990                         continue;
 991
 992                 CDEBUG(D_SUPER, "next opt=%s\n", s1);
 993
 994                 if (strncmp(s1, "no", 2) == 0)
 995                         turn_off = true;
 996
 997                 /*
 998                  * Initialize args struct so we know whether arg was
 999                  * found; some options take optional arguments.
1000                  */
1001                 args[0].to = NULL;
1002                 args[0].from = NULL;
1003                 token = match_token(s1, ll_sbi_flags_name, args);
1004                 if (token == LL_SBI_NUM_MOUNT_OPT) {
1005                         if (match_wildcard("context", s1) ||
1006                             match_wildcard("fscontext", s1) ||
1007                             match_wildcard("defcontext", s1) ||
1008                             match_wildcard("rootcontext",s1))
1009                                 continue;
1010
1011                         LCONSOLE_ERROR_MSG(0x152,
1012                                            "Unknown option '%s', won't mount.\n",
1013                                            s1);
1014                         RETURN(-EINVAL);
1015                 }
1016
1017                 switch (token) {
1018                 case LL_SBI_NOLCK:
1019                 case LL_SBI_32BIT_API:
1020                 case LL_SBI_64BIT_HASH:
1021                 case LL_SBI_ALWAYS_PING:
1022                         set_bit(token, sbi->ll_flags);
1023                         break;
1024
1025                 case LL_SBI_FLOCK:
1026                         clear_bit(LL_SBI_LOCALFLOCK, sbi->ll_flags);
1027                         if (turn_off)
1028                                 clear_bit(LL_SBI_FLOCK, sbi->ll_flags);
1029                         else
1030                                 set_bit(token, sbi->ll_flags);
1031                         break;
1032
1033                 case LL_SBI_LOCALFLOCK:
1034                         clear_bit(LL_SBI_FLOCK, sbi->ll_flags);
1035                         set_bit(token, sbi->ll_flags);
1036                         break;
1037
1038                 case LL_SBI_CHECKSUM:
1039                         sbi->ll_checksum_set = 1;
1040                         /* fall through */
1041                 case LL_SBI_USER_XATTR:
1042                 case LL_SBI_USER_FID2PATH:
1043                 case LL_SBI_LRU_RESIZE:
1044                 case LL_SBI_LAZYSTATFS:
1045                 case LL_SBI_VERBOSE:
1046                         if (turn_off)
1047                                 clear_bit(token, sbi->ll_flags);
1048                         else
1049                                 set_bit(token, sbi->ll_flags);
1050                         break;
1051                 case LL_SBI_TEST_DUMMY_ENCRYPTION: {
1052 #ifdef HAVE_LUSTRE_CRYPTO
1053                         set_bit(token, sbi->ll_flags);
1054 #else
1055                         LCONSOLE_WARN("Test dummy encryption mount option ignored: encryption not supported\n");
1056 #endif
1057                         break;
1058                 }
1059                 case LL_SBI_ENCRYPT:
1060 #ifdef HAVE_LUSTRE_CRYPTO
1061                         if (turn_off)
1062                                 clear_bit(token, sbi->ll_flags);
1063                         else
1064                                 set_bit(token, sbi->ll_flags);
1065 #else
1066                         LCONSOLE_WARN("noencrypt or encrypt mount option ignored: encryption not supported\n");
1067 #endif
1068                         break;
1069                 case LL_SBI_FOREIGN_SYMLINK:
1070                         /* non-default prefix provided ? */
1071                         if (args->from) {
1072                                 size_t old_len;
1073                                 char *old;
1074
1075                                 /* path must be absolute */
1076                                 if (args->from[0] != '/') {
1077                                         LCONSOLE_ERROR_MSG(0x152,
1078                                                            "foreign prefix '%s' must be an absolute path\n",
1079                                                            args->from);
1080                                         RETURN(-EINVAL);
1081                                 }
1082
1083                                 old_len = sbi->ll_foreign_symlink_prefix_size;
1084                                 old = sbi->ll_foreign_symlink_prefix;
1085                                 /* alloc for path length and '\0' */
1086                                 sbi->ll_foreign_symlink_prefix = match_strdup(args);
1087                                 if (!sbi->ll_foreign_symlink_prefix) {
1088                                         /* restore previous */
1089                                         sbi->ll_foreign_symlink_prefix = old;
1090                                         sbi->ll_foreign_symlink_prefix_size =
1091                                                 old_len;
1092                                         RETURN(-ENOMEM);
1093                                 }
1094                                 sbi->ll_foreign_symlink_prefix_size =
1095                                         args->to - args->from + 1;
1096                                 OBD_ALLOC_POST(sbi->ll_foreign_symlink_prefix,
1097                                                sbi->ll_foreign_symlink_prefix_size,
1098                                                "kmalloced");
1099                                 if (old)
1100                                         OBD_FREE(old, old_len);
1101
1102                                 /* enable foreign symlink support */
1103                                 set_bit(token, sbi->ll_flags);
1104                         } else {
1105                                 LCONSOLE_ERROR_MSG(0x152,
1106                                                    "invalid %s option\n", s1);
1107                         }
1108                 /* fall through */
1109                 default:
1110                         break;
1111                 }
1112         }
1113         kfree(opts);
1114         RETURN(0);
1115 }
1116
1117 void ll_lli_init(struct ll_inode_info *lli)
1118 {
1119         lli->lli_inode_magic = LLI_INODE_MAGIC;
1120         lli->lli_flags = 0;
1121         rwlock_init(&lli->lli_lock);
1122         lli->lli_posix_acl = NULL;
1123         /* Do not set lli_fid, it has been initialized already. */
1124         fid_zero(&lli->lli_pfid);
1125         lli->lli_mds_read_och = NULL;
1126         lli->lli_mds_write_och = NULL;
1127         lli->lli_mds_exec_och = NULL;
1128         lli->lli_open_fd_read_count = 0;
1129         lli->lli_open_fd_write_count = 0;
1130         lli->lli_open_fd_exec_count = 0;
1131         mutex_init(&lli->lli_och_mutex);
1132         spin_lock_init(&lli->lli_agl_lock);
1133         spin_lock_init(&lli->lli_layout_lock);
1134         ll_layout_version_set(lli, CL_LAYOUT_GEN_NONE);
1135         lli->lli_clob = NULL;
1136
1137         init_rwsem(&lli->lli_xattrs_list_rwsem);
1138         mutex_init(&lli->lli_xattrs_enq_lock);
1139
1140         LASSERT(lli->lli_vfs_inode.i_mode != 0);
1141         if (S_ISDIR(lli->lli_vfs_inode.i_mode)) {
1142                 lli->lli_opendir_key = NULL;
1143                 lli->lli_sai = NULL;
1144                 spin_lock_init(&lli->lli_sa_lock);
1145                 lli->lli_opendir_pid = 0;
1146                 lli->lli_sa_enabled = 0;
1147                 init_rwsem(&lli->lli_lsm_sem);
1148         } else {
1149                 mutex_init(&lli->lli_size_mutex);
1150                 mutex_init(&lli->lli_setattr_mutex);
1151                 lli->lli_symlink_name = NULL;
1152                 ll_trunc_sem_init(&lli->lli_trunc_sem);
1153                 range_lock_tree_init(&lli->lli_write_tree);
1154                 init_rwsem(&lli->lli_glimpse_sem);
1155                 lli->lli_glimpse_time = ktime_set(0, 0);
1156                 INIT_LIST_HEAD(&lli->lli_agl_list);
1157                 lli->lli_agl_index = 0;
1158                 lli->lli_async_rc = 0;
1159                 spin_lock_init(&lli->lli_heat_lock);
1160                 obd_heat_clear(lli->lli_heat_instances, OBD_HEAT_COUNT);
1161                 lli->lli_heat_flags = 0;
1162                 mutex_init(&lli->lli_pcc_lock);
1163                 lli->lli_pcc_state = PCC_STATE_FL_NONE;
1164                 lli->lli_pcc_inode = NULL;
1165                 lli->lli_pcc_dsflags = PCC_DATASET_INVALID;
1166                 lli->lli_pcc_generation = 0;
1167                 mutex_init(&lli->lli_group_mutex);
1168                 lli->lli_group_users = 0;
1169                 lli->lli_group_gid = 0;
1170         }
1171         mutex_init(&lli->lli_layout_mutex);
1172         memset(lli->lli_jobid, 0, sizeof(lli->lli_jobid));
1173         /* ll_cl_context initialize */
1174         INIT_LIST_HEAD(&lli->lli_lccs);
1175 }
1176
1177 #define MAX_STRING_SIZE 128
1178
1179 #ifndef HAVE_SUPER_SETUP_BDI_NAME
1180
1181 #define LSI_BDI_INITIALIZED     0x00400000
1182
1183 #ifndef HAVE_BDI_CAP_MAP_COPY
1184 # define BDI_CAP_MAP_COPY       0
1185 #endif
1186
1187 static int super_setup_bdi_name(struct super_block *sb, char *fmt, ...)
1188 {
1189         struct  lustre_sb_info *lsi = s2lsi(sb);
1190         char buf[MAX_STRING_SIZE];
1191         va_list args;
1192         int err;
1193
1194         err = bdi_init(&lsi->lsi_bdi);
1195         if (err)
1196                 return err;
1197
1198         lsi->lsi_flags |= LSI_BDI_INITIALIZED;
1199         lsi->lsi_bdi.capabilities = BDI_CAP_MAP_COPY;
1200         lsi->lsi_bdi.name = "lustre";
1201         va_start(args, fmt);
1202         vsnprintf(buf, MAX_STRING_SIZE, fmt, args);
1203         va_end(args);
1204         err = bdi_register(&lsi->lsi_bdi, NULL, "%s", buf);
1205         va_end(args);
1206         if (!err)
1207                 sb->s_bdi = &lsi->lsi_bdi;
1208
1209         return err;
1210 }
1211 #endif /* !HAVE_SUPER_SETUP_BDI_NAME */
1212
1213 int ll_fill_super(struct super_block *sb)
1214 {
1215         struct  lustre_profile *lprof = NULL;
1216         struct  lustre_sb_info *lsi = s2lsi(sb);
1217         struct  ll_sb_info *sbi = NULL;
1218         char    *dt = NULL, *md = NULL;
1219         char    *profilenm = get_profile_name(sb);
1220         struct config_llog_instance *cfg;
1221         /* %p for void* in printf needs 16+2 characters: 0xffffffffffffffff */
1222         const int instlen = LUSTRE_MAXINSTANCE + 2;
1223         unsigned long cfg_instance = ll_get_cfg_instance(sb);
1224         char name[MAX_STRING_SIZE];
1225         int md_len = 0;
1226         int dt_len = 0;
1227         uuid_t uuid;
1228         char *ptr;
1229         int len;
1230         int err;
1231
1232         ENTRY;
1233         /* for ASLR, to map between cfg_instance and hashed ptr */
1234         CDEBUG(D_VFSTRACE, "VFS Op: cfg_instance %s-%016lx (sb %p)\n",
1235                profilenm, cfg_instance, sb);
1236
1237         OBD_RACE(OBD_FAIL_LLITE_RACE_MOUNT);
1238
1239         OBD_ALLOC_PTR(cfg);
1240         if (cfg == NULL)
1241                 GOTO(out_free_cfg, err = -ENOMEM);
1242
1243         /* client additional sb info */
1244         lsi->lsi_llsbi = sbi = ll_init_sbi();
1245         if (IS_ERR(sbi))
1246                 GOTO(out_free_cfg, err = PTR_ERR(sbi));
1247
1248         err = ll_options(lsi->lsi_lmd->lmd_opts, sb);
1249         if (err)
1250                 GOTO(out_free_cfg, err);
1251
1252         /* kernel >= 2.6.38 store dentry operations in sb->s_d_op. */
1253         sb->s_d_op = &ll_d_ops;
1254
1255         /* UUID handling */
1256         generate_random_uuid(uuid.b);
1257         snprintf(sbi->ll_sb_uuid.uuid, sizeof(sbi->ll_sb_uuid), "%pU", uuid.b);
1258
1259         CDEBUG(D_CONFIG, "llite sb uuid: %s\n", sbi->ll_sb_uuid.uuid);
1260
1261         /* Get fsname */
1262         len = strlen(profilenm);
1263         ptr = strrchr(profilenm, '-');
1264         if (ptr && (strcmp(ptr, "-client") == 0))
1265                 len -= 7;
1266
1267         if (len > LUSTRE_MAXFSNAME) {
1268                 if (unlikely(len >= MAX_STRING_SIZE))
1269                         len = MAX_STRING_SIZE - 1;
1270                 strncpy(name, profilenm, len);
1271                 name[len] = '\0';
1272                 err = -ENAMETOOLONG;
1273                 CERROR("%s: fsname longer than %u characters: rc = %d\n",
1274                        name, LUSTRE_MAXFSNAME, err);
1275                 GOTO(out_free_cfg, err);
1276         }
1277         strncpy(sbi->ll_fsname, profilenm, len);
1278         sbi->ll_fsname[len] = '\0';
1279
1280         /* Mount info */
1281         snprintf(name, sizeof(name), "%.*s-%016lx", len,
1282                  profilenm, cfg_instance);
1283
1284         err = super_setup_bdi_name(sb, "%s", name);
1285         if (err)
1286                 GOTO(out_free_cfg, err);
1287
1288         /* Call ll_debugfs_register_super() before lustre_process_log()
1289          * so that "llite.*.*" params can be processed correctly.
1290          */
1291         err = ll_debugfs_register_super(sb, name);
1292         if (err < 0) {
1293                 CERROR("%s: could not register mountpoint in llite: rc = %d\n",
1294                        sbi->ll_fsname, err);
1295                 err = 0;
1296         }
1297
1298         /* The cfg_instance is a value unique to this super, in case some
1299          * joker tries to mount the same fs at two mount points.
1300          */
1301         cfg->cfg_instance = cfg_instance;
1302         cfg->cfg_uuid = lsi->lsi_llsbi->ll_sb_uuid;
1303         cfg->cfg_callback = class_config_llog_handler;
1304         cfg->cfg_sub_clds = CONFIG_SUB_CLIENT;
1305         /* set up client obds */
1306         err = lustre_process_log(sb, profilenm, cfg);
1307         if (err < 0)
1308                 GOTO(out_debugfs, err);
1309
1310         /* Profile set with LCFG_MOUNTOPT so we can find our mdc and osc obds */
1311         lprof = class_get_profile(profilenm);
1312         if (lprof == NULL) {
1313                 LCONSOLE_ERROR_MSG(0x156, "The client profile '%s' could not be"
1314                                    " read from the MGS.  Does that filesystem "
1315                                    "exist?\n", profilenm);
1316                 GOTO(out_debugfs, err = -EINVAL);
1317         }
1318         CDEBUG(D_CONFIG, "Found profile %s: mdc=%s osc=%s\n", profilenm,
1319                lprof->lp_md, lprof->lp_dt);
1320
1321         dt_len = strlen(lprof->lp_dt) + instlen + 2;
1322         OBD_ALLOC(dt, dt_len);
1323         if (!dt)
1324                 GOTO(out_profile, err = -ENOMEM);
1325         snprintf(dt, dt_len - 1, "%s-%016lx", lprof->lp_dt, cfg_instance);
1326
1327         md_len = strlen(lprof->lp_md) + instlen + 2;
1328         OBD_ALLOC(md, md_len);
1329         if (!md)
1330                 GOTO(out_free_dt, err = -ENOMEM);
1331         snprintf(md, md_len - 1, "%s-%016lx", lprof->lp_md, cfg_instance);
1332
1333         /* connections, registrations, sb setup */
1334         err = client_common_fill_super(sb, md, dt);
1335         if (err < 0)
1336                 GOTO(out_free_md, err);
1337
1338         sbi->ll_client_common_fill_super_succeeded = 1;
1339
1340 out_free_md:
1341         if (md)
1342                 OBD_FREE(md, md_len);
1343 out_free_dt:
1344         if (dt)
1345                 OBD_FREE(dt, dt_len);
1346 out_profile:
1347         if (lprof)
1348                 class_put_profile(lprof);
1349 out_debugfs:
1350         if (err < 0)
1351                 ll_debugfs_unregister_super(sb);
1352 out_free_cfg:
1353         if (cfg)
1354                 OBD_FREE_PTR(cfg);
1355
1356         if (err)
1357                 ll_put_super(sb);
1358         else if (test_bit(LL_SBI_VERBOSE, sbi->ll_flags))
1359                 LCONSOLE_WARN("Mounted %s\n", profilenm);
1360         RETURN(err);
1361 } /* ll_fill_super */
1362
1363 void ll_put_super(struct super_block *sb)
1364 {
1365         struct config_llog_instance cfg, params_cfg;
1366         struct obd_device *obd;
1367         struct lustre_sb_info *lsi = s2lsi(sb);
1368         struct ll_sb_info *sbi = ll_s2sbi(sb);
1369         char *profilenm = get_profile_name(sb);
1370         unsigned long cfg_instance = ll_get_cfg_instance(sb);
1371         long ccc_count;
1372         int next, force = 1, rc = 0;
1373         ENTRY;
1374
1375         if (IS_ERR(sbi))
1376                 GOTO(out_no_sbi, 0);
1377
1378         /* Should replace instance_id with something better for ASLR */
1379         CDEBUG(D_VFSTRACE, "VFS Op: cfg_instance %s-%016lx (sb %p)\n",
1380                profilenm, cfg_instance, sb);
1381
1382         cfg.cfg_instance = cfg_instance;
1383         lustre_end_log(sb, profilenm, &cfg);
1384
1385         params_cfg.cfg_instance = cfg_instance;
1386         lustre_end_log(sb, PARAMS_FILENAME, &params_cfg);
1387
1388         if (sbi->ll_md_exp) {
1389                 obd = class_exp2obd(sbi->ll_md_exp);
1390                 if (obd)
1391                         force = obd->obd_force;
1392         }
1393
1394         /* Wait for unstable pages to be committed to stable storage */
1395         if (force == 0) {
1396                 rc = l_wait_event_abortable(
1397                         sbi->ll_cache->ccc_unstable_waitq,
1398                         atomic_long_read(&sbi->ll_cache->ccc_unstable_nr) == 0);
1399         }
1400
1401         ccc_count = atomic_long_read(&sbi->ll_cache->ccc_unstable_nr);
1402         if (force == 0 && rc != -ERESTARTSYS)
1403                 LASSERTF(ccc_count == 0, "count: %li\n", ccc_count);
1404
1405         /* We need to set force before the lov_disconnect in
1406          * lustre_common_put_super, since l_d cleans up osc's as well.
1407          */
1408         if (force) {
1409                 next = 0;
1410                 while ((obd = class_devices_in_group(&sbi->ll_sb_uuid,
1411                                                      &next)) != NULL) {
1412                         obd->obd_force = force;
1413                 }
1414         }
1415
1416         if (sbi->ll_client_common_fill_super_succeeded) {
1417                 /* Only if client_common_fill_super succeeded */
1418                 client_common_put_super(sb);
1419         }
1420
1421         next = 0;
1422         while ((obd = class_devices_in_group(&sbi->ll_sb_uuid, &next)))
1423                 class_manual_cleanup(obd);
1424
1425         if (test_bit(LL_SBI_VERBOSE, sbi->ll_flags))
1426                 LCONSOLE_WARN("Unmounted %s\n", profilenm ? profilenm : "");
1427
1428         if (profilenm)
1429                 class_del_profile(profilenm);
1430
1431 #ifndef HAVE_SUPER_SETUP_BDI_NAME
1432         if (lsi->lsi_flags & LSI_BDI_INITIALIZED) {
1433                 bdi_destroy(&lsi->lsi_bdi);
1434                 lsi->lsi_flags &= ~LSI_BDI_INITIALIZED;
1435         }
1436 #endif
1437
1438         ll_free_sbi(sb);
1439         lsi->lsi_llsbi = NULL;
1440 out_no_sbi:
1441         lustre_common_put_super(sb);
1442
1443         cl_env_cache_purge(~0);
1444
1445         EXIT;
1446 } /* client_put_super */
1447
1448 struct inode *ll_inode_from_resource_lock(struct ldlm_lock *lock)
1449 {
1450         struct inode *inode = NULL;
1451
1452         /* NOTE: we depend on atomic igrab() -bzzz */
1453         lock_res_and_lock(lock);
1454         if (lock->l_resource->lr_lvb_inode) {
1455                 struct ll_inode_info * lli;
1456                 lli = ll_i2info(lock->l_resource->lr_lvb_inode);
1457                 if (lli->lli_inode_magic == LLI_INODE_MAGIC) {
1458                         inode = igrab(lock->l_resource->lr_lvb_inode);
1459                 } else {
1460                         inode = lock->l_resource->lr_lvb_inode;
1461                         LDLM_DEBUG_LIMIT(inode->i_state & I_FREEING ?  D_INFO :
1462                                          D_WARNING, lock, "lr_lvb_inode %p is "
1463                                          "bogus: magic %08x",
1464                                          lock->l_resource->lr_lvb_inode,
1465                                          lli->lli_inode_magic);
1466                         inode = NULL;
1467                 }
1468         }
1469         unlock_res_and_lock(lock);
1470         return inode;
1471 }
1472
1473 void ll_dir_clear_lsm_md(struct inode *inode)
1474 {
1475         struct ll_inode_info *lli = ll_i2info(inode);
1476
1477         LASSERT(S_ISDIR(inode->i_mode));
1478
1479         if (lli->lli_lsm_md) {
1480                 lmv_free_memmd(lli->lli_lsm_md);
1481                 lli->lli_lsm_md = NULL;
1482         }
1483
1484         if (lli->lli_default_lsm_md) {
1485                 lmv_free_memmd(lli->lli_default_lsm_md);
1486                 lli->lli_default_lsm_md = NULL;
1487         }
1488 }
1489
1490 static struct inode *ll_iget_anon_dir(struct super_block *sb,
1491                                       const struct lu_fid *fid,
1492                                       struct lustre_md *md)
1493 {
1494         struct ll_sb_info *sbi = ll_s2sbi(sb);
1495         struct ll_inode_info *lli;
1496         struct mdt_body *body = md->body;
1497         struct inode *inode;
1498         ino_t ino;
1499
1500         ENTRY;
1501
1502         LASSERT(md->lmv);
1503         ino = cl_fid_build_ino(fid, test_bit(LL_SBI_32BIT_API, sbi->ll_flags));
1504         inode = iget_locked(sb, ino);
1505         if (inode == NULL) {
1506                 CERROR("%s: failed get simple inode "DFID": rc = -ENOENT\n",
1507                        sbi->ll_fsname, PFID(fid));
1508                 RETURN(ERR_PTR(-ENOENT));
1509         }
1510
1511         lli = ll_i2info(inode);
1512         if (inode->i_state & I_NEW) {
1513                 inode->i_mode = (inode->i_mode & ~S_IFMT) |
1514                                 (body->mbo_mode & S_IFMT);
1515                 LASSERTF(S_ISDIR(inode->i_mode), "Not slave inode "DFID"\n",
1516                          PFID(fid));
1517
1518                 inode->i_mtime.tv_sec = 0;
1519                 inode->i_atime.tv_sec = 0;
1520                 inode->i_ctime.tv_sec = 0;
1521                 inode->i_rdev = 0;
1522
1523 #ifdef HAVE_BACKING_DEV_INFO
1524                 /* initializing backing dev info. */
1525                 inode->i_mapping->backing_dev_info =
1526                                                 &s2lsi(inode->i_sb)->lsi_bdi;
1527 #endif
1528                 inode->i_op = &ll_dir_inode_operations;
1529                 inode->i_fop = &ll_dir_operations;
1530                 lli->lli_fid = *fid;
1531                 ll_lli_init(lli);
1532
1533                 /* master object FID */
1534                 lli->lli_pfid = body->mbo_fid1;
1535                 CDEBUG(D_INODE, "lli %p slave "DFID" master "DFID"\n",
1536                        lli, PFID(fid), PFID(&lli->lli_pfid));
1537                 unlock_new_inode(inode);
1538         } else {
1539                 /* in directory restripe/auto-split, a directory will be
1540                  * transformed to a stripe if it's plain, set its pfid here,
1541                  * otherwise ll_lock_cancel_bits() can't find the master inode.
1542                  */
1543                 lli->lli_pfid = body->mbo_fid1;
1544         }
1545
1546         RETURN(inode);
1547 }
1548
1549 static int ll_init_lsm_md(struct inode *inode, struct lustre_md *md)
1550 {
1551         struct lu_fid *fid;
1552         struct lmv_stripe_md *lsm = md->lmv;
1553         struct ll_inode_info *lli = ll_i2info(inode);
1554         int i;
1555
1556         LASSERT(lsm != NULL);
1557
1558         CDEBUG(D_INODE, "%s: "DFID" set dir layout:\n",
1559                ll_i2sbi(inode)->ll_fsname, PFID(&lli->lli_fid));
1560         lsm_md_dump(D_INODE, lsm);
1561
1562         if (!lmv_dir_striped(lsm))
1563                 goto out;
1564
1565         /* XXX sigh, this lsm_root initialization should be in
1566          * LMV layer, but it needs ll_iget right now, so we
1567          * put this here right now. */
1568         for (i = 0; i < lsm->lsm_md_stripe_count; i++) {
1569                 fid = &lsm->lsm_md_oinfo[i].lmo_fid;
1570                 LASSERT(lsm->lsm_md_oinfo[i].lmo_root == NULL);
1571
1572                 if (!fid_is_sane(fid))
1573                         continue;
1574
1575                 /* Unfortunately ll_iget will call ll_update_inode,
1576                  * where the initialization of slave inode is slightly
1577                  * different, so it reset lsm_md to NULL to avoid
1578                  * initializing lsm for slave inode. */
1579                 lsm->lsm_md_oinfo[i].lmo_root =
1580                                 ll_iget_anon_dir(inode->i_sb, fid, md);
1581                 if (IS_ERR(lsm->lsm_md_oinfo[i].lmo_root)) {
1582                         int rc = PTR_ERR(lsm->lsm_md_oinfo[i].lmo_root);
1583
1584                         lsm->lsm_md_oinfo[i].lmo_root = NULL;
1585                         while (i-- > 0) {
1586                                 iput(lsm->lsm_md_oinfo[i].lmo_root);
1587                                 lsm->lsm_md_oinfo[i].lmo_root = NULL;
1588                         }
1589                         return rc;
1590                 }
1591         }
1592 out:
1593         lli->lli_lsm_md = lsm;
1594
1595         return 0;
1596 }
1597
1598 static void ll_update_default_lsm_md(struct inode *inode, struct lustre_md *md)
1599 {
1600         struct ll_inode_info *lli = ll_i2info(inode);
1601
1602         if (!md->default_lmv) {
1603                 /* clear default lsm */
1604                 if (lli->lli_default_lsm_md) {
1605                         down_write(&lli->lli_lsm_sem);
1606                         if (lli->lli_default_lsm_md) {
1607                                 lmv_free_memmd(lli->lli_default_lsm_md);
1608                                 lli->lli_default_lsm_md = NULL;
1609                         }
1610                         up_write(&lli->lli_lsm_sem);
1611                 }
1612                 return;
1613         }
1614
1615         if (lli->lli_default_lsm_md) {
1616                 /* do nonthing if default lsm isn't changed */
1617                 down_read(&lli->lli_lsm_sem);
1618                 if (lli->lli_default_lsm_md &&
1619                     lsm_md_eq(lli->lli_default_lsm_md, md->default_lmv)) {
1620                         up_read(&lli->lli_lsm_sem);
1621                         return;
1622                 }
1623                 up_read(&lli->lli_lsm_sem);
1624         }
1625
1626         down_write(&lli->lli_lsm_sem);
1627         if (lli->lli_default_lsm_md)
1628                 lmv_free_memmd(lli->lli_default_lsm_md);
1629         lli->lli_default_lsm_md = md->default_lmv;
1630         lsm_md_dump(D_INODE, md->default_lmv);
1631         md->default_lmv = NULL;
1632         up_write(&lli->lli_lsm_sem);
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 /* If this inode has objects allocated to it (lsm != NULL), then the OST
2033  * object(s) determine the file size and mtime.  Otherwise, the MDS will
2034  * keep these values until such a time that objects are allocated for it.
2035  * We do the MDS operations first, as it is checking permissions for us.
2036  * We don't to the MDS RPC if there is nothing that we want to store there,
2037  * otherwise there is no harm in updating mtime/atime on the MDS if we are
2038  * going to do an RPC anyways.
2039  *
2040  * If we are doing a truncate, we will send the mtime and ctime updates
2041  * to the OST with the punch RPC, otherwise we do an explicit setattr RPC.
2042  * I don't believe it is possible to get e.g. ATTR_MTIME_SET and ATTR_SIZE
2043  * at the same time.
2044  *
2045  * In case of HSMimport, we only set attr on MDS.
2046  */
2047 int ll_setattr_raw(struct dentry *dentry, struct iattr *attr,
2048                    enum op_xvalid xvalid, bool hsm_import)
2049 {
2050         struct inode *inode = dentry->d_inode;
2051         struct ll_inode_info *lli = ll_i2info(inode);
2052         struct md_op_data *op_data = NULL;
2053         ktime_t kstart = ktime_get();
2054         int rc = 0;
2055
2056         ENTRY;
2057
2058         CDEBUG(D_VFSTRACE, "%s: setattr inode "DFID"(%p) from %llu to %llu, "
2059                "valid %x, hsm_import %d\n",
2060                ll_i2sbi(inode)->ll_fsname, PFID(&lli->lli_fid),
2061                inode, i_size_read(inode), attr->ia_size, attr->ia_valid,
2062                hsm_import);
2063
2064         if (attr->ia_valid & ATTR_SIZE) {
2065                 /* Check new size against VFS/VM file size limit and rlimit */
2066                 rc = inode_newsize_ok(inode, attr->ia_size);
2067                 if (rc)
2068                         RETURN(rc);
2069
2070                 /* The maximum Lustre file size is variable, based on the
2071                  * OST maximum object size and number of stripes.  This
2072                  * needs another check in addition to the VFS check above. */
2073                 if (attr->ia_size > ll_file_maxbytes(inode)) {
2074                         CDEBUG(D_INODE,"file "DFID" too large %llu > %llu\n",
2075                                PFID(&lli->lli_fid), attr->ia_size,
2076                                ll_file_maxbytes(inode));
2077                         RETURN(-EFBIG);
2078                 }
2079
2080                 attr->ia_valid |= ATTR_MTIME | ATTR_CTIME;
2081         }
2082
2083         /* POSIX: check before ATTR_*TIME_SET set (from inode_change_ok) */
2084         if (attr->ia_valid & TIMES_SET_FLAGS) {
2085                 if ((!uid_eq(current_fsuid(), inode->i_uid)) &&
2086                     !capable(CAP_FOWNER))
2087                         RETURN(-EPERM);
2088         }
2089
2090         /* We mark all of the fields "set" so MDS/OST does not re-set them */
2091         if (!(xvalid & OP_XVALID_CTIME_SET) &&
2092              (attr->ia_valid & ATTR_CTIME)) {
2093                 attr->ia_ctime = current_time(inode);
2094                 xvalid |= OP_XVALID_CTIME_SET;
2095         }
2096         if (!(attr->ia_valid & ATTR_ATIME_SET) &&
2097             (attr->ia_valid & ATTR_ATIME)) {
2098                 attr->ia_atime = current_time(inode);
2099                 attr->ia_valid |= ATTR_ATIME_SET;
2100         }
2101         if (!(attr->ia_valid & ATTR_MTIME_SET) &&
2102             (attr->ia_valid & ATTR_MTIME)) {
2103                 attr->ia_mtime = current_time(inode);
2104                 attr->ia_valid |= ATTR_MTIME_SET;
2105         }
2106
2107         if (attr->ia_valid & (ATTR_MTIME | ATTR_CTIME))
2108                 CDEBUG(D_INODE, "setting mtime %lld, ctime %lld, now = %lld\n",
2109                        (s64)attr->ia_mtime.tv_sec, (s64)attr->ia_ctime.tv_sec,
2110                        ktime_get_real_seconds());
2111
2112         if (S_ISREG(inode->i_mode))
2113                 inode_unlock(inode);
2114
2115         /* We always do an MDS RPC, even if we're only changing the size;
2116          * only the MDS knows whether truncate() should fail with -ETXTBUSY */
2117
2118         OBD_ALLOC_PTR(op_data);
2119         if (op_data == NULL)
2120                 GOTO(out, rc = -ENOMEM);
2121
2122         if (!hsm_import && attr->ia_valid & ATTR_SIZE) {
2123                 /* If we are changing file size, file content is
2124                  * modified, flag it.
2125                  */
2126                 xvalid |= OP_XVALID_OWNEROVERRIDE;
2127                 op_data->op_bias |= MDS_DATA_MODIFIED;
2128                 clear_bit(LLIF_DATA_MODIFIED, &lli->lli_flags);
2129         }
2130
2131         if (attr->ia_valid & ATTR_FILE) {
2132                 struct ll_file_data *fd = attr->ia_file->private_data;
2133
2134                 if (fd->fd_lease_och)
2135                         op_data->op_bias |= MDS_TRUNC_KEEP_LEASE;
2136         }
2137
2138         op_data->op_attr = *attr;
2139         op_data->op_xvalid = xvalid;
2140
2141         rc = ll_md_setattr(dentry, op_data);
2142         if (rc)
2143                 GOTO(out, rc);
2144
2145         if (!S_ISREG(inode->i_mode) || hsm_import)
2146                 GOTO(out, rc = 0);
2147
2148         if (attr->ia_valid & (ATTR_SIZE | ATTR_ATIME | ATTR_ATIME_SET |
2149                               ATTR_MTIME | ATTR_MTIME_SET | ATTR_CTIME) ||
2150             xvalid & OP_XVALID_CTIME_SET) {
2151                 bool cached = false;
2152
2153                 rc = pcc_inode_setattr(inode, attr, &cached);
2154                 if (cached) {
2155                         if (rc) {
2156                                 CERROR("%s: PCC inode "DFID" setattr failed: "
2157                                        "rc = %d\n",
2158                                        ll_i2sbi(inode)->ll_fsname,
2159                                        PFID(&lli->lli_fid), rc);
2160                                 GOTO(out, rc);
2161                         }
2162                 } else {
2163                         unsigned int flags = 0;
2164
2165                         /* For truncate and utimes sending attributes to OSTs,
2166                          * setting mtime/atime to the past will be performed
2167                          * under PW [0:EOF] extent lock (new_size:EOF for
2168                          * truncate). It may seem excessive to send mtime/atime
2169                          * updates to OSTs when not setting times to past, but
2170                          * it is necessary due to possible time
2171                          * de-synchronization between MDT inode and OST objects
2172                          */
2173                         if (S_ISREG(inode->i_mode) && IS_ENCRYPTED(inode)) {
2174                                 xvalid |= OP_XVALID_FLAGS;
2175                                 flags = LUSTRE_ENCRYPT_FL;
2176                                 /* Call to ll_io_zero_page is not necessary if
2177                                  * truncating on PAGE_SIZE boundary, because
2178                                  * whole pages will be wiped.
2179                                  * In case of Direct IO, all we need is to set
2180                                  * new size.
2181                                  */
2182                                 if (attr->ia_valid & ATTR_SIZE &&
2183                                     attr->ia_size & ~PAGE_MASK &&
2184                                     !(attr->ia_valid & ATTR_FILE &&
2185                                       attr->ia_file->f_flags & O_DIRECT)) {
2186                                         pgoff_t offset =
2187                                                 attr->ia_size & (PAGE_SIZE - 1);
2188
2189                                         rc = ll_io_zero_page(inode,
2190                                                     attr->ia_size >> PAGE_SHIFT,
2191                                                     offset, PAGE_SIZE - offset);
2192                                         if (rc)
2193                                                 GOTO(out, rc);
2194                                 }
2195                         }
2196                         rc = cl_setattr_ost(lli->lli_clob, attr, xvalid, flags);
2197                 }
2198         }
2199
2200         /* If the file was restored, it needs to set dirty flag.
2201          *
2202          * We've already sent MDS_DATA_MODIFIED flag in
2203          * ll_md_setattr() for truncate. However, the MDT refuses to
2204          * set the HS_DIRTY flag on released files, so we have to set
2205          * it again if the file has been restored. Please check how
2206          * LLIF_DATA_MODIFIED is set in vvp_io_setattr_fini().
2207          *
2208          * Please notice that if the file is not released, the previous
2209          * MDS_DATA_MODIFIED has taken effect and usually
2210          * LLIF_DATA_MODIFIED is not set(see vvp_io_setattr_fini()).
2211          * This way we can save an RPC for common open + trunc
2212          * operation. */
2213         if (test_and_clear_bit(LLIF_DATA_MODIFIED, &lli->lli_flags)) {
2214                 struct hsm_state_set hss = {
2215                         .hss_valid = HSS_SETMASK,
2216                         .hss_setmask = HS_DIRTY,
2217                 };
2218                 int rc2;
2219
2220                 rc2 = ll_hsm_state_set(inode, &hss);
2221                 /* truncate and write can happen at the same time, so that
2222                  * the file can be set modified even though the file is not
2223                  * restored from released state, and ll_hsm_state_set() is
2224                  * not applicable for the file, and rc2 < 0 is normal in this
2225                  * case. */
2226                 if (rc2 < 0)
2227                         CDEBUG(D_INFO, DFID "HSM set dirty failed: rc2 = %d\n",
2228                                PFID(ll_inode2fid(inode)), rc2);
2229         }
2230
2231         EXIT;
2232 out:
2233         if (op_data != NULL)
2234                 ll_finish_md_op_data(op_data);
2235
2236         if (S_ISREG(inode->i_mode)) {
2237                 inode_lock(inode);
2238                 if ((attr->ia_valid & ATTR_SIZE) && !hsm_import)
2239                         inode_dio_wait(inode);
2240                 /* Once we've got the i_mutex, it's safe to set the S_NOSEC
2241                  * flag.  ll_update_inode (called from ll_md_setattr), clears
2242                  * inode flags, so there is a gap where S_NOSEC is not set.
2243                  * This can cause a writer to take the i_mutex unnecessarily,
2244                  * but this is safe to do and should be rare. */
2245                 inode_has_no_xattr(inode);
2246         }
2247
2248         if (!rc)
2249                 ll_stats_ops_tally(ll_i2sbi(inode), attr->ia_valid & ATTR_SIZE ?
2250                                         LPROC_LL_TRUNC : LPROC_LL_SETATTR,
2251                                    ktime_us_delta(ktime_get(), kstart));
2252
2253         return rc;
2254 }
2255
2256 int ll_setattr(struct dentry *de, struct iattr *attr)
2257 {
2258         int mode = de->d_inode->i_mode;
2259         enum op_xvalid xvalid = 0;
2260         int rc;
2261
2262         rc = llcrypt_prepare_setattr(de, attr);
2263         if (rc)
2264                 return rc;
2265
2266         if ((attr->ia_valid & (ATTR_CTIME|ATTR_SIZE|ATTR_MODE)) ==
2267                               (ATTR_CTIME|ATTR_SIZE|ATTR_MODE))
2268                 xvalid |= OP_XVALID_OWNEROVERRIDE;
2269
2270         if (((attr->ia_valid & (ATTR_MODE|ATTR_FORCE|ATTR_SIZE)) ==
2271                                (ATTR_SIZE|ATTR_MODE)) &&
2272             (((mode & S_ISUID) && !(attr->ia_mode & S_ISUID)) ||
2273              (((mode & (S_ISGID|S_IXGRP)) == (S_ISGID|S_IXGRP)) &&
2274               !(attr->ia_mode & S_ISGID))))
2275                 attr->ia_valid |= ATTR_FORCE;
2276
2277         if ((attr->ia_valid & ATTR_MODE) &&
2278             (mode & S_ISUID) &&
2279             !(attr->ia_mode & S_ISUID) &&
2280             !(attr->ia_valid & ATTR_KILL_SUID))
2281                 attr->ia_valid |= ATTR_KILL_SUID;
2282
2283         if ((attr->ia_valid & ATTR_MODE) &&
2284             ((mode & (S_ISGID|S_IXGRP)) == (S_ISGID|S_IXGRP)) &&
2285             !(attr->ia_mode & S_ISGID) &&
2286             !(attr->ia_valid & ATTR_KILL_SGID))
2287                 attr->ia_valid |= ATTR_KILL_SGID;
2288
2289         return ll_setattr_raw(de, attr, xvalid, false);
2290 }
2291
2292 int ll_statfs_internal(struct ll_sb_info *sbi, struct obd_statfs *osfs,
2293                        u32 flags)
2294 {
2295         struct obd_statfs obd_osfs = { 0 };
2296         time64_t max_age;
2297         int rc;
2298
2299         ENTRY;
2300         max_age = ktime_get_seconds() - sbi->ll_statfs_max_age;
2301
2302         if (test_bit(LL_SBI_LAZYSTATFS, sbi->ll_flags))
2303                 flags |= OBD_STATFS_NODELAY;
2304
2305         rc = obd_statfs(NULL, sbi->ll_md_exp, osfs, max_age, flags);
2306         if (rc)
2307                 RETURN(rc);
2308
2309         osfs->os_type = LL_SUPER_MAGIC;
2310
2311         CDEBUG(D_SUPER, "MDC blocks %llu/%llu objects %llu/%llu\n",
2312               osfs->os_bavail, osfs->os_blocks, osfs->os_ffree, osfs->os_files);
2313
2314         if (osfs->os_state & OS_STATFS_SUM)
2315                 GOTO(out, rc);
2316
2317         rc = obd_statfs(NULL, sbi->ll_dt_exp, &obd_osfs, max_age, flags);
2318         if (rc) /* Possibly a filesystem with no OSTs.  Report MDT totals. */
2319                 GOTO(out, rc = 0);
2320
2321         CDEBUG(D_SUPER, "OSC blocks %llu/%llu objects %llu/%llu\n",
2322                obd_osfs.os_bavail, obd_osfs.os_blocks, obd_osfs.os_ffree,
2323                obd_osfs.os_files);
2324
2325         osfs->os_bsize = obd_osfs.os_bsize;
2326         osfs->os_blocks = obd_osfs.os_blocks;
2327         osfs->os_bfree = obd_osfs.os_bfree;
2328         osfs->os_bavail = obd_osfs.os_bavail;
2329
2330         /* If we have _some_ OSTs, but don't have as many free objects on the
2331          * OSTs as inodes on the MDTs, reduce the reported number of inodes
2332          * to compensate, so that the "inodes in use" number is correct.
2333          * This should be kept in sync with lod_statfs() behaviour.
2334          */
2335         if (obd_osfs.os_files && obd_osfs.os_ffree < osfs->os_ffree) {
2336                 osfs->os_files = (osfs->os_files - osfs->os_ffree) +
2337                                  obd_osfs.os_ffree;
2338                 osfs->os_ffree = obd_osfs.os_ffree;
2339         }
2340
2341 out:
2342         RETURN(rc);
2343 }
2344
2345 static int ll_statfs_project(struct inode *inode, struct kstatfs *sfs)
2346 {
2347         struct if_quotactl qctl = {
2348                 .qc_cmd = LUSTRE_Q_GETQUOTA,
2349                 .qc_type = PRJQUOTA,
2350                 .qc_valid = QC_GENERAL,
2351         };
2352         u64 limit, curblock;
2353         int ret;
2354
2355         qctl.qc_id = ll_i2info(inode)->lli_projid;
2356         ret = quotactl_ioctl(inode->i_sb, &qctl);
2357         if (ret) {
2358                 /* ignore errors if project ID does not have
2359                  * a quota limit or feature unsupported.
2360                  */
2361                 if (ret == -ESRCH || ret == -EOPNOTSUPP)
2362                         ret = 0;
2363                 return ret;
2364         }
2365
2366         limit = ((qctl.qc_dqblk.dqb_bsoftlimit ?
2367                  qctl.qc_dqblk.dqb_bsoftlimit :
2368                  qctl.qc_dqblk.dqb_bhardlimit) * 1024) / sfs->f_bsize;
2369         if (limit && sfs->f_blocks > limit) {
2370                 curblock = (qctl.qc_dqblk.dqb_curspace +
2371                                 sfs->f_bsize - 1) / sfs->f_bsize;
2372                 sfs->f_blocks = limit;
2373                 sfs->f_bfree = sfs->f_bavail =
2374                         (sfs->f_blocks > curblock) ?
2375                         (sfs->f_blocks - curblock) : 0;
2376         }
2377
2378         limit = qctl.qc_dqblk.dqb_isoftlimit ?
2379                 qctl.qc_dqblk.dqb_isoftlimit :
2380                 qctl.qc_dqblk.dqb_ihardlimit;
2381         if (limit && sfs->f_files > limit) {
2382                 sfs->f_files = limit;
2383                 sfs->f_ffree = (sfs->f_files >
2384                         qctl.qc_dqblk.dqb_curinodes) ?
2385                         (sfs->f_files - qctl.qc_dqblk.dqb_curinodes) : 0;
2386         }
2387
2388         return 0;
2389 }
2390
2391 int ll_statfs(struct dentry *de, struct kstatfs *sfs)
2392 {
2393         struct super_block *sb = de->d_sb;
2394         struct obd_statfs osfs;
2395         __u64 fsid = huge_encode_dev(sb->s_dev);
2396         ktime_t kstart = ktime_get();
2397         int rc;
2398
2399         CDEBUG(D_VFSTRACE, "VFS Op:sb=%s (%p)\n", sb->s_id, sb);
2400
2401         /* Some amount of caching on the client is allowed */
2402         rc = ll_statfs_internal(ll_s2sbi(sb), &osfs, OBD_STATFS_SUM);
2403         if (rc)
2404                 return rc;
2405
2406         statfs_unpack(sfs, &osfs);
2407
2408         /* We need to downshift for all 32-bit kernels, because we can't
2409          * tell if the kernel is being called via sys_statfs64() or not.
2410          * Stop before overflowing f_bsize - in which case it is better
2411          * to just risk EOVERFLOW if caller is using old sys_statfs(). */
2412         if (sizeof(long) < 8) {
2413                 while (osfs.os_blocks > ~0UL && sfs->f_bsize < 0x40000000) {
2414                         sfs->f_bsize <<= 1;
2415
2416                         osfs.os_blocks >>= 1;
2417                         osfs.os_bfree >>= 1;
2418                         osfs.os_bavail >>= 1;
2419                 }
2420         }
2421
2422         sfs->f_blocks = osfs.os_blocks;
2423         sfs->f_bfree = osfs.os_bfree;
2424         sfs->f_bavail = osfs.os_bavail;
2425         sfs->f_fsid.val[0] = (__u32)fsid;
2426         sfs->f_fsid.val[1] = (__u32)(fsid >> 32);
2427         if (ll_i2info(de->d_inode)->lli_projid)
2428                 return ll_statfs_project(de->d_inode, sfs);
2429
2430         ll_stats_ops_tally(ll_s2sbi(sb), LPROC_LL_STATFS,
2431                            ktime_us_delta(ktime_get(), kstart));
2432
2433         return 0;
2434 }
2435
2436 void ll_inode_size_lock(struct inode *inode)
2437 {
2438         struct ll_inode_info *lli;
2439
2440         LASSERT(!S_ISDIR(inode->i_mode));
2441
2442         lli = ll_i2info(inode);
2443         mutex_lock(&lli->lli_size_mutex);
2444 }
2445
2446 void ll_inode_size_unlock(struct inode *inode)
2447 {
2448         struct ll_inode_info *lli;
2449
2450         lli = ll_i2info(inode);
2451         mutex_unlock(&lli->lli_size_mutex);
2452 }
2453
2454 void ll_update_inode_flags(struct inode *inode, unsigned int ext_flags)
2455 {
2456         /* do not clear encryption flag */
2457         ext_flags |= ll_inode_to_ext_flags(inode->i_flags) & LUSTRE_ENCRYPT_FL;
2458         inode->i_flags = ll_ext_to_inode_flags(ext_flags);
2459         if (ext_flags & LUSTRE_PROJINHERIT_FL)
2460                 set_bit(LLIF_PROJECT_INHERIT, &ll_i2info(inode)->lli_flags);
2461         else
2462                 clear_bit(LLIF_PROJECT_INHERIT, &ll_i2info(inode)->lli_flags);
2463 }
2464
2465 int ll_update_inode(struct inode *inode, struct lustre_md *md)
2466 {
2467         struct ll_inode_info *lli = ll_i2info(inode);
2468         struct mdt_body *body = md->body;
2469         struct ll_sb_info *sbi = ll_i2sbi(inode);
2470         bool api32;
2471         int rc = 0;
2472
2473         if (body->mbo_valid & OBD_MD_FLEASIZE) {
2474                 rc = cl_file_inode_init(inode, md);
2475                 if (rc)
2476                         return rc;
2477         }
2478
2479         if (S_ISDIR(inode->i_mode)) {
2480                 rc = ll_update_lsm_md(inode, md);
2481                 if (rc != 0)
2482                         return rc;
2483         }
2484
2485         if (body->mbo_valid & OBD_MD_FLACL)
2486                 lli_replace_acl(lli, md);
2487
2488         api32 = test_bit(LL_SBI_32BIT_API, sbi->ll_flags);
2489         inode->i_ino = cl_fid_build_ino(&body->mbo_fid1, api32);
2490         inode->i_generation = cl_fid_build_gen(&body->mbo_fid1);
2491
2492         if (body->mbo_valid & OBD_MD_FLATIME) {
2493                 if (body->mbo_atime > inode->i_atime.tv_sec)
2494                         inode->i_atime.tv_sec = body->mbo_atime;
2495                 lli->lli_atime = body->mbo_atime;
2496         }
2497
2498         if (body->mbo_valid & OBD_MD_FLMTIME) {
2499                 if (body->mbo_mtime > inode->i_mtime.tv_sec) {
2500                         CDEBUG(D_INODE,
2501                                "setting ino %lu mtime from %lld to %llu\n",
2502                                inode->i_ino, (s64)inode->i_mtime.tv_sec,
2503                                body->mbo_mtime);
2504                         inode->i_mtime.tv_sec = body->mbo_mtime;
2505                 }
2506                 lli->lli_mtime = body->mbo_mtime;
2507         }
2508
2509         if (body->mbo_valid & OBD_MD_FLCTIME) {
2510                 if (body->mbo_ctime > inode->i_ctime.tv_sec)
2511                         inode->i_ctime.tv_sec = body->mbo_ctime;
2512                 lli->lli_ctime = body->mbo_ctime;
2513         }
2514
2515         if (body->mbo_valid & OBD_MD_FLBTIME)
2516                 lli->lli_btime = body->mbo_btime;
2517
2518         /* Clear i_flags to remove S_NOSEC before permissions are updated */
2519         if (body->mbo_valid & OBD_MD_FLFLAGS)
2520                 ll_update_inode_flags(inode, body->mbo_flags);
2521         if (body->mbo_valid & OBD_MD_FLMODE)
2522                 inode->i_mode = (inode->i_mode & S_IFMT) |
2523                                 (body->mbo_mode & ~S_IFMT);
2524
2525         if (body->mbo_valid & OBD_MD_FLTYPE)
2526                 inode->i_mode = (inode->i_mode & ~S_IFMT) |
2527                                 (body->mbo_mode & S_IFMT);
2528
2529         LASSERT(inode->i_mode != 0);
2530         if (body->mbo_valid & OBD_MD_FLUID)
2531                 inode->i_uid = make_kuid(&init_user_ns, body->mbo_uid);
2532         if (body->mbo_valid & OBD_MD_FLGID)
2533                 inode->i_gid = make_kgid(&init_user_ns, body->mbo_gid);
2534         if (body->mbo_valid & OBD_MD_FLPROJID)
2535                 lli->lli_projid = body->mbo_projid;
2536         if (body->mbo_valid & OBD_MD_FLNLINK) {
2537                 spin_lock(&inode->i_lock);
2538                 set_nlink(inode, body->mbo_nlink);
2539                 spin_unlock(&inode->i_lock);
2540         }
2541         if (body->mbo_valid & OBD_MD_FLRDEV)
2542                 inode->i_rdev = old_decode_dev(body->mbo_rdev);
2543
2544         if (body->mbo_valid & OBD_MD_FLID) {
2545                 /* FID shouldn't be changed! */
2546                 if (fid_is_sane(&lli->lli_fid)) {
2547                         LASSERTF(lu_fid_eq(&lli->lli_fid, &body->mbo_fid1),
2548                                  "Trying to change FID "DFID
2549                                  " to the "DFID", inode "DFID"(%p)\n",
2550                                  PFID(&lli->lli_fid), PFID(&body->mbo_fid1),
2551                                  PFID(ll_inode2fid(inode)), inode);
2552                 } else {
2553                         lli->lli_fid = body->mbo_fid1;
2554                 }
2555         }
2556
2557         LASSERT(fid_seq(&lli->lli_fid) != 0);
2558
2559         lli->lli_attr_valid = body->mbo_valid;
2560         if (body->mbo_valid & OBD_MD_FLSIZE) {
2561                 i_size_write(inode, body->mbo_size);
2562
2563                 CDEBUG(D_VFSTRACE, "inode="DFID", updating i_size %llu\n",
2564                        PFID(ll_inode2fid(inode)),
2565                        (unsigned long long)body->mbo_size);
2566
2567                 if (body->mbo_valid & OBD_MD_FLBLOCKS)
2568                         inode->i_blocks = body->mbo_blocks;
2569         } else {
2570                 if (body->mbo_valid & OBD_MD_FLLAZYSIZE)
2571                         lli->lli_lazysize = body->mbo_size;
2572                 if (body->mbo_valid & OBD_MD_FLLAZYBLOCKS)
2573                         lli->lli_lazyblocks = body->mbo_blocks;
2574         }
2575
2576         if (body->mbo_valid & OBD_MD_TSTATE) {
2577                 /* Set LLIF_FILE_RESTORING if restore ongoing and
2578                  * clear it when done to ensure to start again
2579                  * glimpsing updated attrs
2580                  */
2581                 if (body->mbo_t_state & MS_RESTORE)
2582                         set_bit(LLIF_FILE_RESTORING, &lli->lli_flags);
2583                 else
2584                         clear_bit(LLIF_FILE_RESTORING, &lli->lli_flags);
2585         }
2586
2587         return 0;
2588 }
2589
2590 /* update directory depth to ROOT, called after LOOKUP lock is fetched. */
2591 void ll_update_dir_depth(struct inode *dir, struct inode *inode)
2592 {
2593         struct ll_inode_info *lli;
2594
2595         if (!S_ISDIR(inode->i_mode))
2596                 return;
2597
2598         if (inode == dir)
2599                 return;
2600
2601         lli = ll_i2info(inode);
2602         lli->lli_depth = ll_i2info(dir)->lli_depth + 1;
2603         CDEBUG(D_INODE, DFID" depth %hu\n", PFID(&lli->lli_fid), lli->lli_depth);
2604 }
2605
2606 void ll_truncate_inode_pages_final(struct inode *inode)
2607 {
2608         struct address_space *mapping = &inode->i_data;
2609         unsigned long nrpages;
2610         unsigned long flags;
2611
2612         truncate_inode_pages_final(mapping);
2613
2614         /* Workaround for LU-118: Note nrpages may not be totally updated when
2615          * truncate_inode_pages() returns, as there can be a page in the process
2616          * of deletion (inside __delete_from_page_cache()) in the specified
2617          * range. Thus mapping->nrpages can be non-zero when this function
2618          * returns even after truncation of the whole mapping.  Only do this if
2619          * npages isn't already zero.
2620          */
2621         nrpages = mapping->nrpages;
2622         if (nrpages) {
2623                 ll_xa_lock_irqsave(&mapping->i_pages, flags);
2624                 nrpages = mapping->nrpages;
2625                 ll_xa_unlock_irqrestore(&mapping->i_pages, flags);
2626         } /* Workaround end */
2627
2628         LASSERTF(nrpages == 0, "%s: inode="DFID"(%p) nrpages=%lu, "
2629                  "see https://jira.whamcloud.com/browse/LU-118\n",
2630                  ll_i2sbi(inode)->ll_fsname,
2631                  PFID(ll_inode2fid(inode)), inode, nrpages);
2632 }
2633
2634 int ll_read_inode2(struct inode *inode, void *opaque)
2635 {
2636         struct lustre_md *md = opaque;
2637         struct ll_inode_info *lli = ll_i2info(inode);
2638         int     rc;
2639         ENTRY;
2640
2641         CDEBUG(D_VFSTRACE, "VFS Op:inode="DFID"(%p)\n",
2642                PFID(&lli->lli_fid), inode);
2643
2644         /* Core attributes from the MDS first.  This is a new inode, and
2645          * the VFS doesn't zero times in the core inode so we have to do
2646          * it ourselves.  They will be overwritten by either MDS or OST
2647          * attributes - we just need to make sure they aren't newer.
2648          */
2649         inode->i_mtime.tv_sec = 0;
2650         inode->i_atime.tv_sec = 0;
2651         inode->i_ctime.tv_sec = 0;
2652         inode->i_rdev = 0;
2653         rc = ll_update_inode(inode, md);
2654         if (rc != 0)
2655                 RETURN(rc);
2656
2657         /* OIDEBUG(inode); */
2658
2659 #ifdef HAVE_BACKING_DEV_INFO
2660         /* initializing backing dev info. */
2661         inode->i_mapping->backing_dev_info = &s2lsi(inode->i_sb)->lsi_bdi;
2662 #endif
2663         if (S_ISREG(inode->i_mode)) {
2664                 struct ll_sb_info *sbi = ll_i2sbi(inode);
2665                 inode->i_op = &ll_file_inode_operations;
2666                 inode->i_fop = sbi->ll_fop;
2667                 inode->i_mapping->a_ops = (struct address_space_operations *)&ll_aops;
2668                 EXIT;
2669         } else if (S_ISDIR(inode->i_mode)) {
2670                 inode->i_op = &ll_dir_inode_operations;
2671                 inode->i_fop = &ll_dir_operations;
2672                 EXIT;
2673         } else if (S_ISLNK(inode->i_mode)) {
2674                 inode->i_op = &ll_fast_symlink_inode_operations;
2675                 EXIT;
2676         } else {
2677                 inode->i_op = &ll_special_inode_operations;
2678
2679                 init_special_inode(inode, inode->i_mode,
2680                                    inode->i_rdev);
2681
2682                 EXIT;
2683         }
2684
2685         return 0;
2686 }
2687
2688 void ll_delete_inode(struct inode *inode)
2689 {
2690         struct ll_inode_info *lli = ll_i2info(inode);
2691         ENTRY;
2692
2693         if (S_ISREG(inode->i_mode) && lli->lli_clob != NULL) {
2694                 /* It is last chance to write out dirty pages,
2695                  * otherwise we may lose data while umount.
2696                  *
2697                  * If i_nlink is 0 then just discard data. This is safe because
2698                  * local inode gets i_nlink 0 from server only for the last
2699                  * unlink, so that file is not opened somewhere else
2700                  */
2701                 cl_sync_file_range(inode, 0, OBD_OBJECT_EOF, inode->i_nlink ?
2702                                    CL_FSYNC_LOCAL : CL_FSYNC_DISCARD, 1);
2703         }
2704
2705         ll_truncate_inode_pages_final(inode);
2706         ll_clear_inode(inode);
2707         clear_inode(inode);
2708
2709         EXIT;
2710 }
2711
2712 int ll_iocontrol(struct inode *inode, struct file *file,
2713                  unsigned int cmd, unsigned long arg)
2714 {
2715         struct ll_sb_info *sbi = ll_i2sbi(inode);
2716         struct ptlrpc_request *req = NULL;
2717         int rc, flags = 0;
2718         ENTRY;
2719
2720         switch (cmd) {
2721         case FS_IOC_GETFLAGS: {
2722                 struct mdt_body *body;
2723                 struct md_op_data *op_data;
2724
2725                 op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL,
2726                                              0, 0, LUSTRE_OPC_ANY,
2727                                              NULL);
2728                 if (IS_ERR(op_data))
2729                         RETURN(PTR_ERR(op_data));
2730
2731                 op_data->op_valid = OBD_MD_FLFLAGS;
2732                 rc = md_getattr(sbi->ll_md_exp, op_data, &req);
2733                 ll_finish_md_op_data(op_data);
2734                 if (rc) {
2735                         CERROR("%s: failure inode "DFID": rc = %d\n",
2736                                sbi->ll_md_exp->exp_obd->obd_name,
2737                                PFID(ll_inode2fid(inode)), rc);
2738                         RETURN(-abs(rc));
2739                 }
2740
2741                 body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
2742
2743                 flags = body->mbo_flags;
2744
2745                 ptlrpc_req_finished(req);
2746
2747                 RETURN(put_user(flags, (int __user *)arg));
2748         }
2749         case FS_IOC_SETFLAGS: {
2750                 struct iattr *attr;
2751                 struct md_op_data *op_data;
2752                 struct cl_object *obj;
2753                 struct fsxattr fa = { 0 };
2754
2755                 if (get_user(flags, (int __user *)arg))
2756                         RETURN(-EFAULT);
2757
2758                 fa.fsx_projid = ll_i2info(inode)->lli_projid;
2759                 if (flags & LUSTRE_PROJINHERIT_FL)
2760                         fa.fsx_xflags = FS_XFLAG_PROJINHERIT;
2761
2762                 rc = ll_ioctl_check_project(inode, fa.fsx_xflags,
2763                                             fa.fsx_projid);
2764                 if (rc)
2765                         RETURN(rc);
2766
2767                 op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL, 0, 0,
2768                                              LUSTRE_OPC_ANY, NULL);
2769                 if (IS_ERR(op_data))
2770                         RETURN(PTR_ERR(op_data));
2771
2772                 op_data->op_attr_flags = flags;
2773                 op_data->op_xvalid |= OP_XVALID_FLAGS;
2774                 rc = md_setattr(sbi->ll_md_exp, op_data, NULL, 0, &req);
2775                 ll_finish_md_op_data(op_data);
2776                 ptlrpc_req_finished(req);
2777                 if (rc)
2778                         RETURN(rc);
2779
2780                 ll_update_inode_flags(inode, flags);
2781
2782                 obj = ll_i2info(inode)->lli_clob;
2783                 if (obj == NULL)
2784                         RETURN(0);
2785
2786                 OBD_ALLOC_PTR(attr);
2787                 if (attr == NULL)
2788                         RETURN(-ENOMEM);
2789
2790                 rc = cl_setattr_ost(obj, attr, OP_XVALID_FLAGS, flags);
2791
2792                 OBD_FREE_PTR(attr);
2793                 RETURN(rc);
2794         }
2795         default:
2796                 RETURN(-ENOSYS);
2797         }
2798
2799         RETURN(0);
2800 }
2801
2802 int ll_flush_ctx(struct inode *inode)
2803 {
2804         struct ll_sb_info  *sbi = ll_i2sbi(inode);
2805
2806         CDEBUG(D_SEC, "flush context for user %d\n",
2807                from_kuid(&init_user_ns, current_uid()));
2808
2809         obd_set_info_async(NULL, sbi->ll_md_exp,
2810                            sizeof(KEY_FLUSH_CTX), KEY_FLUSH_CTX,
2811                            0, NULL, NULL);
2812         obd_set_info_async(NULL, sbi->ll_dt_exp,
2813                            sizeof(KEY_FLUSH_CTX), KEY_FLUSH_CTX,
2814                            0, NULL, NULL);
2815         return 0;
2816 }
2817
2818 /* umount -f client means force down, don't save state */
2819 void ll_umount_begin(struct super_block *sb)
2820 {
2821         struct ll_sb_info *sbi = ll_s2sbi(sb);
2822         struct obd_device *obd;
2823         struct obd_ioctl_data *ioc_data;
2824         int cnt;
2825         ENTRY;
2826
2827         CDEBUG(D_VFSTRACE, "VFS Op: superblock %p count %d active %d\n", sb,
2828                sb->s_count, atomic_read(&sb->s_active));
2829
2830         obd = class_exp2obd(sbi->ll_md_exp);
2831         if (obd == NULL) {
2832                 CERROR("Invalid MDC connection handle %#llx\n",
2833                        sbi->ll_md_exp->exp_handle.h_cookie);
2834                 EXIT;
2835                 return;
2836         }
2837         obd->obd_force = 1;
2838
2839         obd = class_exp2obd(sbi->ll_dt_exp);
2840         if (obd == NULL) {
2841                 CERROR("Invalid LOV connection handle %#llx\n",
2842                        sbi->ll_dt_exp->exp_handle.h_cookie);
2843                 EXIT;
2844                 return;
2845         }
2846         obd->obd_force = 1;
2847
2848         OBD_ALLOC_PTR(ioc_data);
2849         if (ioc_data) {
2850                 obd_iocontrol(IOC_OSC_SET_ACTIVE, sbi->ll_md_exp,
2851                               sizeof *ioc_data, ioc_data, NULL);
2852
2853                 obd_iocontrol(IOC_OSC_SET_ACTIVE, sbi->ll_dt_exp,
2854                               sizeof *ioc_data, ioc_data, NULL);
2855
2856                 OBD_FREE_PTR(ioc_data);
2857         }
2858
2859         /* Really, we'd like to wait until there are no requests outstanding,
2860          * and then continue.  For now, we just periodically checking for vfs
2861          * to decrement mnt_cnt and hope to finish it within 10sec.
2862          */
2863         cnt = 10;
2864         while (cnt > 0 &&
2865                !may_umount(sbi->ll_mnt.mnt)) {
2866                 ssleep(1);
2867                 cnt -= 1;
2868         }
2869
2870         EXIT;
2871 }
2872
2873 int ll_remount_fs(struct super_block *sb, int *flags, char *data)
2874 {
2875         struct ll_sb_info *sbi = ll_s2sbi(sb);
2876         char *profilenm = get_profile_name(sb);
2877         int err;
2878         __u32 read_only;
2879
2880         if ((*flags & MS_RDONLY) != (sb->s_flags & SB_RDONLY)) {
2881                 read_only = *flags & MS_RDONLY;
2882                 err = obd_set_info_async(NULL, sbi->ll_md_exp,
2883                                          sizeof(KEY_READ_ONLY),
2884                                          KEY_READ_ONLY, sizeof(read_only),
2885                                          &read_only, NULL);
2886                 if (err) {
2887                         LCONSOLE_WARN("Failed to remount %s %s (%d)\n",
2888                                       profilenm, read_only ?
2889                                       "read-only" : "read-write", err);
2890                         return err;
2891                 }
2892
2893                 if (read_only)
2894                         sb->s_flags |= SB_RDONLY;
2895                 else
2896                         sb->s_flags &= ~SB_RDONLY;
2897
2898                 if (test_bit(LL_SBI_VERBOSE, sbi->ll_flags))
2899                         LCONSOLE_WARN("Remounted %s %s\n", profilenm,
2900                                       read_only ?  "read-only" : "read-write");
2901         }
2902         return 0;
2903 }
2904
2905 /**
2906  * Cleanup the open handle that is cached on MDT-side.
2907  *
2908  * For open case, the client side open handling thread may hit error
2909  * after the MDT grant the open. Under such case, the client should
2910  * send close RPC to the MDT as cleanup; otherwise, the open handle
2911  * on the MDT will be leaked there until the client umount or evicted.
2912  *
2913  * In further, if someone unlinked the file, because the open handle
2914  * holds the reference on such file/object, then it will block the
2915  * subsequent threads that want to locate such object via FID.
2916  *
2917  * \param[in] sb        super block for this file-system
2918  * \param[in] open_req  pointer to the original open request
2919  */
2920 void ll_open_cleanup(struct super_block *sb, struct req_capsule *pill)
2921 {
2922         struct mdt_body                 *body;
2923         struct md_op_data               *op_data;
2924         struct ptlrpc_request           *close_req = NULL;
2925         struct obd_export               *exp       = ll_s2sbi(sb)->ll_md_exp;
2926         ENTRY;
2927
2928         body = req_capsule_server_get(pill, &RMF_MDT_BODY);
2929         OBD_ALLOC_PTR(op_data);
2930         if (op_data == NULL) {
2931                 CWARN("%s: cannot allocate op_data to release open handle for "
2932                       DFID"\n", ll_s2sbi(sb)->ll_fsname, PFID(&body->mbo_fid1));
2933
2934                 RETURN_EXIT;
2935         }
2936
2937         op_data->op_fid1 = body->mbo_fid1;
2938         op_data->op_open_handle = body->mbo_open_handle;
2939         op_data->op_mod_time = ktime_get_real_seconds();
2940         md_close(exp, op_data, NULL, &close_req);
2941         ptlrpc_req_finished(close_req);
2942         ll_finish_md_op_data(op_data);
2943
2944         EXIT;
2945 }
2946
2947 int ll_prep_inode(struct inode **inode, struct req_capsule *pill,
2948                   struct super_block *sb, struct lookup_intent *it)
2949 {
2950         struct ll_sb_info *sbi = NULL;
2951         struct lustre_md md = { NULL };
2952         bool default_lmv_deleted = false;
2953         int rc;
2954
2955         ENTRY;
2956
2957         LASSERT(*inode || sb);
2958         sbi = sb ? ll_s2sbi(sb) : ll_i2sbi(*inode);
2959         rc = md_get_lustre_md(sbi->ll_md_exp, pill, sbi->ll_dt_exp,
2960                               sbi->ll_md_exp, &md);
2961         if (rc != 0)
2962                 GOTO(out, rc);
2963
2964         /*
2965          * clear default_lmv only if intent_getattr reply doesn't contain it.
2966          * but it needs to be done after iget, check this early because
2967          * ll_update_lsm_md() may change md.
2968          */
2969         if (it && (it->it_op & (IT_LOOKUP | IT_GETATTR)) &&
2970             S_ISDIR(md.body->mbo_mode) && !md.default_lmv)
2971                 default_lmv_deleted = true;
2972
2973         if (*inode) {
2974                 rc = ll_update_inode(*inode, &md);
2975                 if (rc != 0)
2976                         GOTO(out, rc);
2977         } else {
2978                 bool api32 = test_bit(LL_SBI_32BIT_API, sbi->ll_flags);
2979                 struct lu_fid *fid1 = &md.body->mbo_fid1;
2980
2981                 LASSERT(sb != NULL);
2982
2983                 /*
2984                  * At this point server returns to client's same fid as client
2985                  * generated for creating. So using ->fid1 is okay here.
2986                  */
2987                 if (!fid_is_sane(fid1)) {
2988                         CERROR("%s: Fid is insane "DFID"\n",
2989                                 sbi->ll_fsname, PFID(fid1));
2990                         GOTO(out, rc = -EINVAL);
2991                 }
2992
2993                 *inode = ll_iget(sb, cl_fid_build_ino(fid1, api32), &md);
2994                 if (IS_ERR(*inode)) {
2995                         lmd_clear_acl(&md);
2996                         rc = IS_ERR(*inode) ? PTR_ERR(*inode) : -ENOMEM;
2997                         *inode = NULL;
2998                         CERROR("new_inode -fatal: rc %d\n", rc);
2999                         GOTO(out, rc);
3000                 }
3001         }
3002
3003         /* Handling piggyback layout lock.
3004          * Layout lock can be piggybacked by getattr and open request.
3005          * The lsm can be applied to inode only if it comes with a layout lock
3006          * otherwise correct layout may be overwritten, for example:
3007          * 1. proc1: mdt returns a lsm but not granting layout
3008          * 2. layout was changed by another client
3009          * 3. proc2: refresh layout and layout lock granted
3010          * 4. proc1: to apply a stale layout */
3011         if (it != NULL && it->it_lock_mode != 0) {
3012                 struct lustre_handle lockh;
3013                 struct ldlm_lock *lock;
3014
3015                 lockh.cookie = it->it_lock_handle;
3016                 lock = ldlm_handle2lock(&lockh);
3017                 LASSERT(lock != NULL);
3018                 if (ldlm_has_layout(lock)) {
3019                         struct cl_object_conf conf;
3020
3021                         memset(&conf, 0, sizeof(conf));
3022                         conf.coc_opc = OBJECT_CONF_SET;
3023                         conf.coc_inode = *inode;
3024                         conf.coc_lock = lock;
3025                         conf.u.coc_layout = md.layout;
3026                         (void)ll_layout_conf(*inode, &conf);
3027                 }
3028                 LDLM_LOCK_PUT(lock);
3029         }
3030
3031         if (default_lmv_deleted)
3032                 ll_update_default_lsm_md(*inode, &md);
3033
3034         /* we may want to apply some policy for foreign file/dir */
3035         if (ll_sbi_has_foreign_symlink(sbi)) {
3036                 rc = ll_manage_foreign(*inode, &md);
3037                 if (rc < 0)
3038                         GOTO(out, rc);
3039         }
3040
3041         GOTO(out, rc = 0);
3042
3043 out:
3044         /* cleanup will be done if necessary */
3045         md_free_lustre_md(sbi->ll_md_exp, &md);
3046
3047         if (rc != 0 && it != NULL && it->it_op & IT_OPEN) {
3048                 ll_intent_drop_lock(it);
3049                 ll_open_cleanup(sb != NULL ? sb : (*inode)->i_sb, pill);
3050         }
3051
3052         return rc;
3053 }
3054
3055 int ll_obd_statfs(struct inode *inode, void __user *arg)
3056 {
3057         struct ll_sb_info *sbi = NULL;
3058         struct obd_export *exp;
3059         struct obd_ioctl_data *data = NULL;
3060         __u32 type;
3061         int len = 0, rc;
3062
3063         if (inode)
3064                 sbi = ll_i2sbi(inode);
3065         if (!sbi)
3066                 GOTO(out_statfs, rc = -EINVAL);
3067
3068         rc = obd_ioctl_getdata(&data, &len, arg);
3069         if (rc)
3070                 GOTO(out_statfs, rc);
3071
3072         if (!data->ioc_inlbuf1 || !data->ioc_inlbuf2 ||
3073             !data->ioc_pbuf1 || !data->ioc_pbuf2)
3074                 GOTO(out_statfs, rc = -EINVAL);
3075
3076         if (data->ioc_inllen1 != sizeof(__u32) ||
3077             data->ioc_inllen2 != sizeof(__u32) ||
3078             data->ioc_plen1 != sizeof(struct obd_statfs) ||
3079             data->ioc_plen2 != sizeof(struct obd_uuid))
3080                 GOTO(out_statfs, rc = -EINVAL);
3081
3082         memcpy(&type, data->ioc_inlbuf1, sizeof(__u32));
3083         if (type & LL_STATFS_LMV)
3084                 exp = sbi->ll_md_exp;
3085         else if (type & LL_STATFS_LOV)
3086                 exp = sbi->ll_dt_exp;
3087         else
3088                 GOTO(out_statfs, rc = -ENODEV);
3089
3090         rc = obd_iocontrol(IOC_OBD_STATFS, exp, len, data, NULL);
3091         if (rc)
3092                 GOTO(out_statfs, rc);
3093 out_statfs:
3094         OBD_FREE_LARGE(data, len);
3095         return rc;
3096 }
3097
3098 /*
3099  * this is normally called in ll_fini_md_op_data(), but sometimes it needs to
3100  * be called early to avoid deadlock.
3101  */
3102 void ll_unlock_md_op_lsm(struct md_op_data *op_data)
3103 {
3104         if (op_data->op_mea2_sem) {
3105                 up_read_non_owner(op_data->op_mea2_sem);
3106                 op_data->op_mea2_sem = NULL;
3107         }
3108
3109         if (op_data->op_mea1_sem) {
3110                 up_read_non_owner(op_data->op_mea1_sem);
3111                 op_data->op_mea1_sem = NULL;
3112         }
3113 }
3114
3115 /* this function prepares md_op_data hint for passing it down to MD stack. */
3116 struct md_op_data *ll_prep_md_op_data(struct md_op_data *op_data,
3117                                       struct inode *i1, struct inode *i2,
3118                                       const char *name, size_t namelen,
3119                                       __u32 mode, enum md_op_code opc,
3120                                       void *data)
3121 {
3122         struct llcrypt_name fname = { 0 };
3123         int rc;
3124
3125         LASSERT(i1 != NULL);
3126
3127         if (name == NULL) {
3128                 /* Do not reuse namelen for something else. */
3129                 if (namelen != 0)
3130                         return ERR_PTR(-EINVAL);
3131         } else {
3132                 if (namelen > ll_i2sbi(i1)->ll_namelen)
3133                         return ERR_PTR(-ENAMETOOLONG);
3134
3135                 /* "/" is not valid name, but it's allowed */
3136                 if (!lu_name_is_valid_2(name, namelen) &&
3137                     strncmp("/", name, namelen) != 0)
3138                         return ERR_PTR(-EINVAL);
3139         }
3140
3141         if (op_data == NULL)
3142                 OBD_ALLOC_PTR(op_data);
3143
3144         if (op_data == NULL)
3145                 return ERR_PTR(-ENOMEM);
3146
3147         ll_i2gids(op_data->op_suppgids, i1, i2);
3148         op_data->op_fid1 = *ll_inode2fid(i1);
3149
3150         if (S_ISDIR(i1->i_mode)) {
3151                 down_read_non_owner(&ll_i2info(i1)->lli_lsm_sem);
3152                 op_data->op_mea1_sem = &ll_i2info(i1)->lli_lsm_sem;
3153                 op_data->op_mea1 = ll_i2info(i1)->lli_lsm_md;
3154                 op_data->op_default_mea1 = ll_i2info(i1)->lli_default_lsm_md;
3155         }
3156
3157         if (i2) {
3158                 op_data->op_fid2 = *ll_inode2fid(i2);
3159                 if (S_ISDIR(i2->i_mode)) {
3160                         if (i2 != i1) {
3161                                 /* i2 is typically a child of i1, and MUST be
3162                                  * further from the root to avoid deadlocks.
3163                                  */
3164                                 down_read_non_owner(&ll_i2info(i2)->lli_lsm_sem);
3165                                 op_data->op_mea2_sem =
3166                                                 &ll_i2info(i2)->lli_lsm_sem;
3167                         }
3168                         op_data->op_mea2 = ll_i2info(i2)->lli_lsm_md;
3169                 }
3170         } else {
3171                 fid_zero(&op_data->op_fid2);
3172         }
3173
3174         if (test_bit(LL_SBI_64BIT_HASH, ll_i2sbi(i1)->ll_flags))
3175                 op_data->op_cli_flags |= CLI_HASH64;
3176
3177         if (ll_need_32bit_api(ll_i2sbi(i1)))
3178                 op_data->op_cli_flags |= CLI_API32;
3179
3180         if (opc == LUSTRE_OPC_LOOKUP || opc == LUSTRE_OPC_CREATE) {
3181                 /* In case of lookup, ll_setup_filename() has already been
3182                  * called in ll_lookup_it(), so just take provided name.
3183                  */
3184                 fname.disk_name.name = (unsigned char *)name;
3185                 fname.disk_name.len = namelen;
3186         } else if (name && namelen) {
3187                 struct qstr dname = QSTR_INIT(name, namelen);
3188                 struct inode *dir;
3189                 struct lu_fid *pfid = NULL;
3190                 struct lu_fid fid;
3191                 int lookup;
3192
3193                 if (!S_ISDIR(i1->i_mode) && i2 && S_ISDIR(i2->i_mode)) {
3194                         /* special case when called from ll_link() */
3195                         dir = i2;
3196                         lookup = 0;
3197                 } else {
3198                         dir = i1;
3199                         lookup = (int)(opc == LUSTRE_OPC_ANY);
3200                 }
3201                 if (opc == LUSTRE_OPC_ANY && lookup)
3202                         pfid = &fid;
3203                 rc = ll_setup_filename(dir, &dname, lookup, &fname, pfid);
3204                 if (rc) {
3205                         ll_finish_md_op_data(op_data);
3206                         return ERR_PTR(rc);
3207                 }
3208                 if (pfid && !fid_is_zero(pfid)) {
3209                         if (i2 == NULL)
3210                                 op_data->op_fid2 = fid;
3211                         op_data->op_bias = MDS_FID_OP;
3212                 }
3213                 if (fname.disk_name.name &&
3214                     fname.disk_name.name != (unsigned char *)name)
3215                         /* op_data->op_name must be freed after use */
3216                         op_data->op_flags |= MF_OPNAME_KMALLOCED;
3217         }
3218
3219         /* In fact LUSTRE_OPC_LOOKUP, LUSTRE_OPC_OPEN, LUSTRE_OPC_MIGR
3220          * are LUSTRE_OPC_ANY
3221          */
3222         if (opc == LUSTRE_OPC_LOOKUP || opc == LUSTRE_OPC_OPEN ||
3223             opc == LUSTRE_OPC_MIGR)
3224                 op_data->op_code = LUSTRE_OPC_ANY;
3225         else
3226                 op_data->op_code = opc;
3227         op_data->op_name = fname.disk_name.name;
3228         op_data->op_namelen = fname.disk_name.len;
3229         op_data->op_mode = mode;
3230         op_data->op_mod_time = ktime_get_real_seconds();
3231         op_data->op_fsuid = from_kuid(&init_user_ns, current_fsuid());
3232         op_data->op_fsgid = from_kgid(&init_user_ns, current_fsgid());
3233         op_data->op_cap = current_cap();
3234         op_data->op_mds = 0;
3235         if ((opc == LUSTRE_OPC_CREATE) && (name != NULL) &&
3236              filename_is_volatile(name, namelen, &op_data->op_mds)) {
3237                 op_data->op_bias |= MDS_CREATE_VOLATILE;
3238         }
3239         op_data->op_data = data;
3240
3241         return op_data;
3242 }
3243
3244 void ll_finish_md_op_data(struct md_op_data *op_data)
3245 {
3246         ll_unlock_md_op_lsm(op_data);
3247         ll_security_release_secctx(op_data->op_file_secctx,
3248                                    op_data->op_file_secctx_size);
3249         if (op_data->op_flags & MF_OPNAME_KMALLOCED)
3250                 /* allocated via ll_setup_filename called
3251                  * from ll_prep_md_op_data
3252                  */
3253                 kfree(op_data->op_name);
3254         llcrypt_free_ctx(op_data->op_file_encctx, op_data->op_file_encctx_size);
3255         OBD_FREE_PTR(op_data);
3256 }
3257
3258 int ll_show_options(struct seq_file *seq, struct dentry *dentry)
3259 {
3260         struct ll_sb_info *sbi;
3261         int i;
3262
3263         LASSERT(seq && dentry);
3264         sbi = ll_s2sbi(dentry->d_sb);
3265
3266         if (test_bit(LL_SBI_NOLCK, sbi->ll_flags))
3267                 seq_puts(seq, "nolock");
3268
3269         for (i = 1; ll_sbi_flags_name[i].token != LL_SBI_NUM_MOUNT_OPT; i++) {
3270                 /* match_table in some cases has patterns for both enabled and
3271                  * disabled cases. Ignore 'no'xxx versions if bit is set.
3272                  */
3273                 if (test_bit(ll_sbi_flags_name[i].token, sbi->ll_flags) &&
3274                     strncmp(ll_sbi_flags_name[i].pattern, "no", 2)) {
3275                         if (ll_sbi_flags_name[i].token ==
3276                             LL_SBI_FOREIGN_SYMLINK) {
3277                                 seq_show_option(seq, "foreign_symlink",
3278                                                 sbi->ll_foreign_symlink_prefix);
3279                         } else {
3280                                 seq_printf(seq, ",%s",
3281                                            ll_sbi_flags_name[i].pattern);
3282                         }
3283
3284                         /* You can have either localflock or flock but not
3285                          * both. If localflock is set don't print flock or
3286                          * noflock.
3287                          */
3288                         if (ll_sbi_flags_name[i].token == LL_SBI_LOCALFLOCK)
3289                                 i += 2;
3290                 } else if (!test_bit(ll_sbi_flags_name[i].token, sbi->ll_flags) &&
3291                            !strncmp(ll_sbi_flags_name[i].pattern, "no", 2)) {
3292                         seq_printf(seq, ",%s",
3293                                    ll_sbi_flags_name[i].pattern);
3294                 }
3295         }
3296
3297         RETURN(0);
3298 }
3299
3300 /**
3301  * Get obd name by cmd, and copy out to user space
3302  */
3303 int ll_get_obd_name(struct inode *inode, unsigned int cmd, unsigned long arg)
3304 {
3305         struct ll_sb_info *sbi = ll_i2sbi(inode);
3306         struct obd_device *obd;
3307         ENTRY;
3308
3309         if (cmd == OBD_IOC_GETNAME_OLD || cmd == OBD_IOC_GETDTNAME)
3310                 obd = class_exp2obd(sbi->ll_dt_exp);
3311         else if (cmd == OBD_IOC_GETMDNAME)
3312                 obd = class_exp2obd(sbi->ll_md_exp);
3313         else
3314                 RETURN(-EINVAL);
3315
3316         if (!obd)
3317                 RETURN(-ENOENT);
3318
3319         if (copy_to_user((void __user *)arg, obd->obd_name,
3320                          strlen(obd->obd_name) + 1))
3321                 RETURN(-EFAULT);
3322
3323         RETURN(0);
3324 }
3325
3326 static char* ll_d_path(struct dentry *dentry, char *buf, int bufsize)
3327 {
3328         char *path = NULL;
3329
3330         struct path p;
3331
3332         p.dentry = dentry;
3333         p.mnt = current->fs->root.mnt;
3334         path_get(&p);
3335         path = d_path(&p, buf, bufsize);
3336         path_put(&p);
3337         return path;
3338 }
3339
3340 void ll_dirty_page_discard_warn(struct page *page, int ioret)
3341 {
3342         char *buf, *path = NULL;
3343         struct dentry *dentry = NULL;
3344         struct inode *inode = page->mapping->host;
3345
3346         /* this can be called inside spin lock so use GFP_ATOMIC. */
3347         buf = (char *)__get_free_page(GFP_ATOMIC);
3348         if (buf != NULL) {
3349                 dentry = d_find_alias(page->mapping->host);
3350                 if (dentry != NULL)
3351                         path = ll_d_path(dentry, buf, PAGE_SIZE);
3352         }
3353
3354         /* The below message is checked in recovery-small.sh test_24b */
3355         CDEBUG(D_WARNING,
3356                "%s: dirty page discard: %s/fid: "DFID"/%s may get corrupted "
3357                "(rc %d)\n", ll_i2sbi(inode)->ll_fsname,
3358                s2lsi(page->mapping->host->i_sb)->lsi_lmd->lmd_dev,
3359                PFID(ll_inode2fid(inode)),
3360                (path && !IS_ERR(path)) ? path : "", ioret);
3361
3362         if (dentry != NULL)
3363                 dput(dentry);
3364
3365         if (buf != NULL)
3366                 free_page((unsigned long)buf);
3367 }
3368
3369 ssize_t ll_copy_user_md(const struct lov_user_md __user *md,
3370                         struct lov_user_md **kbuf)
3371 {
3372         struct lov_user_md      lum;
3373         ssize_t                 lum_size;
3374         ENTRY;
3375
3376         if (copy_from_user(&lum, md, sizeof(lum)))
3377                 RETURN(-EFAULT);
3378
3379         lum_size = ll_lov_user_md_size(&lum);
3380         if (lum_size < 0)
3381                 RETURN(lum_size);
3382
3383         OBD_ALLOC_LARGE(*kbuf, lum_size);
3384         if (*kbuf == NULL)
3385                 RETURN(-ENOMEM);
3386
3387         if (copy_from_user(*kbuf, md, lum_size) != 0) {
3388                 OBD_FREE_LARGE(*kbuf, lum_size);
3389                 RETURN(-EFAULT);
3390         }
3391
3392         RETURN(lum_size);
3393 }
3394
3395 /*
3396  * Compute llite root squash state after a change of root squash
3397  * configuration setting or add/remove of a lnet nid
3398  */
3399 void ll_compute_rootsquash_state(struct ll_sb_info *sbi)
3400 {
3401         struct root_squash_info *squash = &sbi->ll_squash;
3402         int i;
3403         bool matched;
3404         struct lnet_process_id id;
3405
3406         /* Update norootsquash flag */
3407         spin_lock(&squash->rsi_lock);
3408         if (list_empty(&squash->rsi_nosquash_nids))
3409                 clear_bit(LL_SBI_NOROOTSQUASH, sbi->ll_flags);
3410         else {
3411                 /* Do not apply root squash as soon as one of our NIDs is
3412                  * in the nosquash_nids list */
3413                 matched = false;
3414                 i = 0;
3415                 while (LNetGetId(i++, &id) != -ENOENT) {
3416                         if (id.nid == LNET_NID_LO_0)
3417                                 continue;
3418                         if (cfs_match_nid(id.nid, &squash->rsi_nosquash_nids)) {
3419                                 matched = true;
3420                                 break;
3421                         }
3422                 }
3423                 if (matched)
3424                         set_bit(LL_SBI_NOROOTSQUASH, sbi->ll_flags);
3425                 else
3426                         clear_bit(LL_SBI_NOROOTSQUASH, sbi->ll_flags);
3427         }
3428         spin_unlock(&squash->rsi_lock);
3429 }
3430
3431 /**
3432  * Parse linkea content to extract information about a given hardlink
3433  *
3434  * \param[in]   ldata      - Initialized linkea data
3435  * \param[in]   linkno     - Link identifier
3436  * \param[out]  parent_fid - The entry's parent FID
3437  * \param[out]  ln         - Entry name destination buffer
3438  *
3439  * \retval 0 on success
3440  * \retval Appropriate negative error code on failure
3441  */
3442 static int ll_linkea_decode(struct linkea_data *ldata, unsigned int linkno,
3443                             struct lu_fid *parent_fid, struct lu_name *ln)
3444 {
3445         unsigned int    idx;
3446         int             rc;
3447         ENTRY;
3448
3449         rc = linkea_init_with_rec(ldata);
3450         if (rc < 0)
3451                 RETURN(rc);
3452
3453         if (linkno >= ldata->ld_leh->leh_reccount)
3454                 /* beyond last link */
3455                 RETURN(-ENODATA);
3456
3457         linkea_first_entry(ldata);
3458         for (idx = 0; ldata->ld_lee != NULL; idx++) {
3459                 linkea_entry_unpack(ldata->ld_lee, &ldata->ld_reclen, ln,
3460                                     parent_fid);
3461                 if (idx == linkno)
3462                         break;
3463
3464                 linkea_next_entry(ldata);
3465         }
3466
3467         if (idx < linkno)
3468                 RETURN(-ENODATA);
3469
3470         RETURN(0);
3471 }
3472
3473 /**
3474  * Get parent FID and name of an identified link. Operation is performed for
3475  * a given link number, letting the caller iterate over linkno to list one or
3476  * all links of an entry.
3477  *
3478  * \param[in]     file - File descriptor against which to perform the operation
3479  * \param[in,out] arg  - User-filled structure containing the linkno to operate
3480  *                       on and the available size. It is eventually filled with
3481  *                       the requested information or left untouched on error
3482  *
3483  * \retval - 0 on success
3484  * \retval - Appropriate negative error code on failure
3485  */
3486 int ll_getparent(struct file *file, struct getparent __user *arg)
3487 {
3488         struct inode            *inode = file_inode(file);
3489         struct linkea_data      *ldata;
3490         struct lu_buf            buf = LU_BUF_NULL;
3491         struct lu_name           ln;
3492         struct lu_fid            parent_fid;
3493         __u32                    linkno;
3494         __u32                    name_size;
3495         int                      rc;
3496
3497         ENTRY;
3498
3499         if (!capable(CAP_DAC_READ_SEARCH) &&
3500             !test_bit(LL_SBI_USER_FID2PATH, ll_i2sbi(inode)->ll_flags))
3501                 RETURN(-EPERM);
3502
3503         if (get_user(name_size, &arg->gp_name_size))
3504                 RETURN(-EFAULT);
3505
3506         if (get_user(linkno, &arg->gp_linkno))
3507                 RETURN(-EFAULT);
3508
3509         if (name_size > PATH_MAX)
3510                 RETURN(-EINVAL);
3511
3512         OBD_ALLOC(ldata, sizeof(*ldata));
3513         if (ldata == NULL)
3514                 RETURN(-ENOMEM);
3515
3516         rc = linkea_data_new(ldata, &buf);
3517         if (rc < 0)
3518                 GOTO(ldata_free, rc);
3519
3520         rc = ll_xattr_list(inode, XATTR_NAME_LINK, XATTR_TRUSTED_T, buf.lb_buf,
3521                            buf.lb_len, OBD_MD_FLXATTR);
3522         if (rc < 0)
3523                 GOTO(lb_free, rc);
3524
3525         rc = ll_linkea_decode(ldata, linkno, &parent_fid, &ln);
3526         if (rc < 0)
3527                 GOTO(lb_free, rc);
3528
3529         if (ln.ln_namelen >= name_size)
3530                 GOTO(lb_free, rc = -EOVERFLOW);
3531
3532         if (copy_to_user(&arg->gp_fid, &parent_fid, sizeof(arg->gp_fid)))
3533                 GOTO(lb_free, rc = -EFAULT);
3534
3535         if (copy_to_user(&arg->gp_name, ln.ln_name, ln.ln_namelen))
3536                 GOTO(lb_free, rc = -EFAULT);
3537
3538         if (put_user('\0', arg->gp_name + ln.ln_namelen))
3539                 GOTO(lb_free, rc = -EFAULT);
3540
3541 lb_free:
3542         lu_buf_free(&buf);
3543 ldata_free:
3544         OBD_FREE(ldata, sizeof(*ldata));
3545
3546         RETURN(rc);
3547 }