lustre/llite/llite_lib.c

   1 /*
   2  * GPL HEADER START
   3  *
   4  * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
   5  *
   6  * This program is free software; you can redistribute it and/or modify
   7  * it under the terms of the GNU General Public License version 2 only,
   8  * as published by the Free Software Foundation.
   9  *
  10  * This program is distributed in the hope that it will be useful, but
  11  * WITHOUT ANY WARRANTY; without even the implied warranty of
  12  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  13  * General Public License version 2 for more details (a copy is included
  14  * in the LICENSE file that accompanied this code).
  15  *
  16  * You should have received a copy of the GNU General Public License
  17  * version 2 along with this program; If not, see
  18  * http://www.gnu.org/licenses/gpl-2.0.html
  19  *
  20  * GPL HEADER END
  21  */
  22 /*
  23  * Copyright (c) 2003, 2010, Oracle and/or its affiliates. All rights reserved.
  24  * Use is subject to license terms.
  25  *
  26  * Copyright (c) 2011, 2017, Intel Corporation.
  27  */
  28 /*
  29  * This file is part of Lustre, http://www.lustre.org/
  30  * Lustre is a trademark of Sun Microsystems, Inc.
  31  *
  32  * lustre/llite/llite_lib.c
  33  *
  34  * Lustre Light Super operations
  35  */
  36
  37 #define DEBUG_SUBSYSTEM S_LLITE
  38
  39 #include <linux/cpu.h>
  40 #include <linux/module.h>
  41 #include <linux/random.h>
  42 #include <linux/statfs.h>
  43 #include <linux/time.h>
  44 #include <linux/types.h>
  45 #include <libcfs/linux/linux-uuid.h>
  46 #include <linux/version.h>
  47 #include <linux/mm.h>
  48 #include <linux/user_namespace.h>
  49 #include <linux/delay.h>
  50 #include <linux/uidgid.h>
  51 #include <linux/security.h>
  52
  53 #ifndef HAVE_CPUS_READ_LOCK
  54 #include <libcfs/linux/linux-cpu.h>
  55 #endif
  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         sbi->ll_ra_info.ra_max_pages_per_file = min(pages / 32,
 132                                                     SBI_DEFAULT_READ_AHEAD_MAX);
 133         sbi->ll_ra_info.ra_async_pages_per_file_threshold =
 134                                 sbi->ll_ra_info.ra_max_pages_per_file;
 135         sbi->ll_ra_info.ra_max_pages = sbi->ll_ra_info.ra_max_pages_per_file;
 136         sbi->ll_ra_info.ra_max_read_ahead_whole_pages = -1;
 137         atomic_set(&sbi->ll_ra_info.ra_async_inflight, 0);
 138
 139         sbi->ll_flags |= LL_SBI_VERBOSE;
 140 #ifdef ENABLE_CHECKSUM
 141         sbi->ll_flags |= LL_SBI_CHECKSUM;
 142 #endif
 143 #ifdef ENABLE_FLOCK
 144         sbi->ll_flags |= LL_SBI_FLOCK;
 145 #endif
 146
 147 #ifdef HAVE_LRU_RESIZE_SUPPORT
 148         sbi->ll_flags |= LL_SBI_LRU_RESIZE;
 149 #endif
 150         sbi->ll_flags |= LL_SBI_LAZYSTATFS;
 151
 152         for (i = 0; i <= LL_PROCESS_HIST_MAX; i++) {
 153                 spin_lock_init(&sbi->ll_rw_extents_info.pp_extents[i].
 154                                pp_r_hist.oh_lock);
 155                 spin_lock_init(&sbi->ll_rw_extents_info.pp_extents[i].
 156                                pp_w_hist.oh_lock);
 157         }
 158
 159         /* metadata statahead is enabled by default */
 160         sbi->ll_sa_running_max = LL_SA_RUNNING_DEF;
 161         sbi->ll_sa_max = LL_SA_RPC_DEF;
 162         atomic_set(&sbi->ll_sa_total, 0);
 163         atomic_set(&sbi->ll_sa_wrong, 0);
 164         atomic_set(&sbi->ll_sa_running, 0);
 165         atomic_set(&sbi->ll_agl_total, 0);
 166         sbi->ll_flags |= LL_SBI_AGL_ENABLED;
 167         sbi->ll_flags |= LL_SBI_FAST_READ;
 168         sbi->ll_flags |= LL_SBI_TINY_WRITE;
 169         ll_sbi_set_encrypt(sbi, true);
 170
 171         /* root squash */
 172         sbi->ll_squash.rsi_uid = 0;
 173         sbi->ll_squash.rsi_gid = 0;
 174         INIT_LIST_HEAD(&sbi->ll_squash.rsi_nosquash_nids);
 175         spin_lock_init(&sbi->ll_squash.rsi_lock);
 176
 177         /* Per-filesystem file heat */
 178         sbi->ll_heat_decay_weight = SBI_DEFAULT_HEAT_DECAY_WEIGHT;
 179         sbi->ll_heat_period_second = SBI_DEFAULT_HEAT_PERIOD_SECOND;
 180         RETURN(sbi);
 181 out_destroy_ra:
 182         destroy_workqueue(sbi->ll_ra_info.ll_readahead_wq);
 183 out_pcc:
 184         pcc_super_fini(&sbi->ll_pcc_super);
 185 out_sbi:
 186         OBD_FREE_PTR(sbi);
 187         RETURN(ERR_PTR(rc));
 188 }
 189
 190 static void ll_free_sbi(struct super_block *sb)
 191 {
 192         struct ll_sb_info *sbi = ll_s2sbi(sb);
 193         ENTRY;
 194
 195         if (sbi != NULL) {
 196                 if (!list_empty(&sbi->ll_squash.rsi_nosquash_nids))
 197                         cfs_free_nidlist(&sbi->ll_squash.rsi_nosquash_nids);
 198                 if (sbi->ll_ra_info.ll_readahead_wq)
 199                         destroy_workqueue(sbi->ll_ra_info.ll_readahead_wq);
 200                 if (sbi->ll_cache != NULL) {
 201                         cl_cache_decref(sbi->ll_cache);
 202                         sbi->ll_cache = NULL;
 203                 }
 204                 pcc_super_fini(&sbi->ll_pcc_super);
 205                 OBD_FREE(sbi, sizeof(*sbi));
 206         }
 207         EXIT;
 208 }
 209
 210 static int client_common_fill_super(struct super_block *sb, char *md, char *dt)
 211 {
 212         struct inode *root = NULL;
 213         struct ll_sb_info *sbi = ll_s2sbi(sb);
 214         struct obd_statfs *osfs = NULL;
 215         struct ptlrpc_request *request = NULL;
 216         struct obd_connect_data *data = NULL;
 217         struct obd_uuid *uuid;
 218         struct md_op_data *op_data;
 219         struct lustre_md lmd;
 220         u64 valid;
 221         int size, err, checksum;
 222
 223         ENTRY;
 224         sbi->ll_md_obd = class_name2obd(md);
 225         if (!sbi->ll_md_obd) {
 226                 CERROR("MD %s: not setup or attached\n", md);
 227                 RETURN(-EINVAL);
 228         }
 229
 230         OBD_ALLOC_PTR(data);
 231         if (data == NULL)
 232                 RETURN(-ENOMEM);
 233
 234         OBD_ALLOC_PTR(osfs);
 235         if (osfs == NULL) {
 236                 OBD_FREE_PTR(data);
 237                 RETURN(-ENOMEM);
 238         }
 239
 240         /* pass client page size via ocd_grant_blkbits, the server should report
 241          * back its backend blocksize for grant calculation purpose */
 242         data->ocd_grant_blkbits = PAGE_SHIFT;
 243
 244         /* indicate MDT features supported by this client */
 245         data->ocd_connect_flags = OBD_CONNECT_IBITS    | OBD_CONNECT_NODEVOH  |
 246                                   OBD_CONNECT_ATTRFID  | OBD_CONNECT_GRANT |
 247                                   OBD_CONNECT_VERSION  | OBD_CONNECT_BRW_SIZE |
 248                                   OBD_CONNECT_SRVLOCK  | OBD_CONNECT_TRUNCLOCK|
 249                                   OBD_CONNECT_MDS_CAPA | OBD_CONNECT_OSS_CAPA |
 250                                   OBD_CONNECT_CANCELSET | OBD_CONNECT_FID     |
 251                                   OBD_CONNECT_AT       | OBD_CONNECT_LOV_V3   |
 252                                   OBD_CONNECT_VBR | OBD_CONNECT_FULL20 |
 253                                   OBD_CONNECT_64BITHASH |
 254                                   OBD_CONNECT_EINPROGRESS |
 255                                   OBD_CONNECT_JOBSTATS | OBD_CONNECT_LVB_TYPE |
 256                                   OBD_CONNECT_LAYOUTLOCK | OBD_CONNECT_PINGLESS|
 257                                   OBD_CONNECT_MAX_EASIZE |
 258                                   OBD_CONNECT_FLOCK_DEAD |
 259                                   OBD_CONNECT_DISP_STRIPE | OBD_CONNECT_LFSCK |
 260                                   OBD_CONNECT_OPEN_BY_FID |
 261                                   OBD_CONNECT_DIR_STRIPE |
 262                                   OBD_CONNECT_BULK_MBITS | OBD_CONNECT_CKSUM |
 263                                   OBD_CONNECT_SUBTREE |
 264                                   OBD_CONNECT_MULTIMODRPCS |
 265                                   OBD_CONNECT_GRANT_PARAM |
 266                                   OBD_CONNECT_SHORTIO | OBD_CONNECT_FLAGS2;
 267
 268         data->ocd_connect_flags2 = OBD_CONNECT2_DIR_MIGRATE |
 269                                    OBD_CONNECT2_SUM_STATFS |
 270                                    OBD_CONNECT2_OVERSTRIPING |
 271                                    OBD_CONNECT2_FLR |
 272                                    OBD_CONNECT2_LOCK_CONVERT |
 273                                    OBD_CONNECT2_ARCHIVE_ID_ARRAY |
 274                                    OBD_CONNECT2_INC_XID |
 275                                    OBD_CONNECT2_LSOM |
 276                                    OBD_CONNECT2_ASYNC_DISCARD |
 277                                    OBD_CONNECT2_PCC |
 278                                    OBD_CONNECT2_CRUSH;
 279
 280 #ifdef HAVE_LRU_RESIZE_SUPPORT
 281         if (sbi->ll_flags & LL_SBI_LRU_RESIZE)
 282                 data->ocd_connect_flags |= OBD_CONNECT_LRU_RESIZE;
 283 #endif
 284 #ifdef CONFIG_LUSTRE_FS_POSIX_ACL
 285         data->ocd_connect_flags |= OBD_CONNECT_ACL | OBD_CONNECT_UMASK |
 286                                    OBD_CONNECT_LARGE_ACL;
 287 #endif
 288
 289         data->ocd_cksum_types = obd_cksum_types_supported_client();
 290
 291         if (OBD_FAIL_CHECK(OBD_FAIL_MDC_LIGHTWEIGHT))
 292                 /* flag mdc connection as lightweight, only used for test
 293                  * purpose, use with care */
 294                 data->ocd_connect_flags |= OBD_CONNECT_LIGHTWEIGHT;
 295
 296         data->ocd_ibits_known = MDS_INODELOCK_FULL;
 297         data->ocd_version = LUSTRE_VERSION_CODE;
 298
 299         if (sb->s_flags & SB_RDONLY)
 300                 data->ocd_connect_flags |= OBD_CONNECT_RDONLY;
 301         if (sbi->ll_flags & LL_SBI_USER_XATTR)
 302                 data->ocd_connect_flags |= OBD_CONNECT_XATTR;
 303
 304 #ifdef SB_NOSEC
 305         /* Setting this indicates we correctly support S_NOSEC (See kernel
 306          * commit 9e1f1de02c2275d7172e18dc4e7c2065777611bf)
 307          */
 308         sb->s_flags |= SB_NOSEC;
 309 #endif
 310
 311         if (sbi->ll_flags & LL_SBI_FLOCK)
 312                 sbi->ll_fop = &ll_file_operations_flock;
 313         else if (sbi->ll_flags & LL_SBI_LOCALFLOCK)
 314                 sbi->ll_fop = &ll_file_operations;
 315         else
 316                 sbi->ll_fop = &ll_file_operations_noflock;
 317
 318         /* always ping even if server suppress_pings */
 319         if (sbi->ll_flags & LL_SBI_ALWAYS_PING)
 320                 data->ocd_connect_flags &= ~OBD_CONNECT_PINGLESS;
 321
 322         obd_connect_set_secctx(data);
 323         if (ll_sbi_has_encrypt(sbi))
 324                 obd_connect_set_enc(data);
 325
 326 #if defined(CONFIG_SECURITY)
 327         data->ocd_connect_flags2 |= OBD_CONNECT2_SELINUX_POLICY;
 328 #endif
 329
 330         data->ocd_brw_size = MD_MAX_BRW_SIZE;
 331
 332         err = obd_connect(NULL, &sbi->ll_md_exp, sbi->ll_md_obd,
 333                           &sbi->ll_sb_uuid, data, sbi->ll_cache);
 334         if (err == -EBUSY) {
 335                 LCONSOLE_ERROR_MSG(0x14f, "An MDT (md %s) is performing "
 336                                    "recovery, of which this client is not a "
 337                                    "part. Please wait for recovery to complete,"
 338                                    " abort, or time out.\n", md);
 339                 GOTO(out, err);
 340         } else if (err) {
 341                 CERROR("cannot connect to %s: rc = %d\n", md, err);
 342                 GOTO(out, err);
 343         }
 344
 345         sbi->ll_md_exp->exp_connect_data = *data;
 346
 347         err = obd_fid_init(sbi->ll_md_exp->exp_obd, sbi->ll_md_exp,
 348                            LUSTRE_SEQ_METADATA);
 349         if (err) {
 350                 CERROR("%s: Can't init metadata layer FID infrastructure, "
 351                        "rc = %d\n", sbi->ll_md_exp->exp_obd->obd_name, err);
 352                 GOTO(out_md, err);
 353         }
 354
 355         /* For mount, we only need fs info from MDT0, and also in DNE, it
 356          * can make sure the client can be mounted as long as MDT0 is
 357          * avaible */
 358         err = obd_statfs(NULL, sbi->ll_md_exp, osfs,
 359                         ktime_get_seconds() - sbi->ll_statfs_max_age,
 360                         OBD_STATFS_FOR_MDT0);
 361         if (err)
 362                 GOTO(out_md_fid, err);
 363
 364         /* This needs to be after statfs to ensure connect has finished.
 365          * Note that "data" does NOT contain the valid connect reply.
 366          * If connecting to a 1.8 server there will be no LMV device, so
 367          * we can access the MDC export directly and exp_connect_flags will
 368          * be non-zero, but if accessing an upgraded 2.1 server it will
 369          * have the correct flags filled in.
 370          * XXX: fill in the LMV exp_connect_flags from MDC(s). */
 371         valid = exp_connect_flags(sbi->ll_md_exp) & CLIENT_CONNECT_MDT_REQD;
 372         if (exp_connect_flags(sbi->ll_md_exp) != 0 &&
 373             valid != CLIENT_CONNECT_MDT_REQD) {
 374                 char *buf;
 375
 376                 OBD_ALLOC_WAIT(buf, PAGE_SIZE);
 377                 obd_connect_flags2str(buf, PAGE_SIZE,
 378                                       valid ^ CLIENT_CONNECT_MDT_REQD, 0, ",");
 379                 LCONSOLE_ERROR_MSG(0x170, "Server %s does not support "
 380                                    "feature(s) needed for correct operation "
 381                                    "of this client (%s). Please upgrade "
 382                                    "server or downgrade client.\n",
 383                                    sbi->ll_md_exp->exp_obd->obd_name, buf);
 384                 OBD_FREE(buf, PAGE_SIZE);
 385                 GOTO(out_md_fid, err = -EPROTO);
 386         }
 387
 388         size = sizeof(*data);
 389         err = obd_get_info(NULL, sbi->ll_md_exp, sizeof(KEY_CONN_DATA),
 390                            KEY_CONN_DATA,  &size, data);
 391         if (err) {
 392                 CERROR("%s: Get connect data failed: rc = %d\n",
 393                        sbi->ll_md_exp->exp_obd->obd_name, err);
 394                 GOTO(out_md_fid, err);
 395         }
 396
 397         LASSERT(osfs->os_bsize);
 398         sb->s_blocksize = osfs->os_bsize;
 399         sb->s_blocksize_bits = log2(osfs->os_bsize);
 400         sb->s_magic = LL_SUPER_MAGIC;
 401         sb->s_maxbytes = MAX_LFS_FILESIZE;
 402         sbi->ll_namelen = osfs->os_namelen;
 403         sbi->ll_mnt.mnt = current->fs->root.mnt;
 404
 405         if ((sbi->ll_flags & LL_SBI_USER_XATTR) &&
 406             !(data->ocd_connect_flags & OBD_CONNECT_XATTR)) {
 407                 LCONSOLE_INFO("Disabling user_xattr feature because "
 408                               "it is not supported on the server\n");
 409                 sbi->ll_flags &= ~LL_SBI_USER_XATTR;
 410         }
 411
 412         if (data->ocd_connect_flags & OBD_CONNECT_ACL) {
 413 #ifdef SB_POSIXACL
 414                 sb->s_flags |= SB_POSIXACL;
 415 #endif
 416                 sbi->ll_flags |= LL_SBI_ACL;
 417         } else {
 418                 LCONSOLE_INFO("client wants to enable acl, but mdt not!\n");
 419 #ifdef SB_POSIXACL
 420                 sb->s_flags &= ~SB_POSIXACL;
 421 #endif
 422                 sbi->ll_flags &= ~LL_SBI_ACL;
 423         }
 424
 425         if (data->ocd_connect_flags & OBD_CONNECT_64BITHASH)
 426                 sbi->ll_flags |= LL_SBI_64BIT_HASH;
 427
 428         if (data->ocd_connect_flags & OBD_CONNECT_LAYOUTLOCK)
 429                 sbi->ll_flags |= LL_SBI_LAYOUT_LOCK;
 430
 431         if (obd_connect_has_secctx(data))
 432                 sbi->ll_flags |= LL_SBI_FILE_SECCTX;
 433
 434         if (ll_sbi_has_encrypt(sbi) && !obd_connect_has_enc(data)) {
 435                 if (ll_sbi_has_test_dummy_encryption(sbi))
 436                         LCONSOLE_WARN("%s: server %s does not support encryption feature, encryption deactivated.\n",
 437                                       sbi->ll_fsname,
 438                                       sbi->ll_md_exp->exp_obd->obd_name);
 439                 ll_sbi_set_encrypt(sbi, false);
 440         }
 441
 442         if (data->ocd_ibits_known & MDS_INODELOCK_XATTR) {
 443                 if (!(data->ocd_connect_flags & OBD_CONNECT_MAX_EASIZE)) {
 444                         LCONSOLE_INFO("%s: disabling xattr cache due to "
 445                                       "unknown maximum xattr size.\n", dt);
 446                 } else if (!sbi->ll_xattr_cache_set) {
 447                         /* If xattr_cache is already set (no matter 0 or 1)
 448                          * during processing llog, it won't be enabled here. */
 449                         sbi->ll_flags |= LL_SBI_XATTR_CACHE;
 450                         sbi->ll_xattr_cache_enabled = 1;
 451                 }
 452         }
 453
 454         sbi->ll_dt_obd = class_name2obd(dt);
 455         if (!sbi->ll_dt_obd) {
 456                 CERROR("DT %s: not setup or attached\n", dt);
 457                 GOTO(out_md_fid, err = -ENODEV);
 458         }
 459
 460         /* pass client page size via ocd_grant_blkbits, the server should report
 461          * back its backend blocksize for grant calculation purpose */
 462         data->ocd_grant_blkbits = PAGE_SHIFT;
 463
 464         /* indicate OST features supported by this client */
 465         data->ocd_connect_flags = OBD_CONNECT_GRANT | OBD_CONNECT_VERSION |
 466                                   OBD_CONNECT_REQPORTAL | OBD_CONNECT_BRW_SIZE |
 467                                   OBD_CONNECT_CANCELSET | OBD_CONNECT_FID |
 468                                   OBD_CONNECT_SRVLOCK | OBD_CONNECT_TRUNCLOCK|
 469                                   OBD_CONNECT_AT | OBD_CONNECT_OSS_CAPA |
 470                                   OBD_CONNECT_VBR | OBD_CONNECT_FULL20 |
 471                                   OBD_CONNECT_64BITHASH | OBD_CONNECT_MAXBYTES |
 472                                   OBD_CONNECT_EINPROGRESS |
 473                                   OBD_CONNECT_JOBSTATS | OBD_CONNECT_LVB_TYPE |
 474                                   OBD_CONNECT_LAYOUTLOCK |
 475                                   OBD_CONNECT_PINGLESS | OBD_CONNECT_LFSCK |
 476                                   OBD_CONNECT_BULK_MBITS | OBD_CONNECT_SHORTIO |
 477                                   OBD_CONNECT_FLAGS2 | OBD_CONNECT_GRANT_SHRINK;
 478         data->ocd_connect_flags2 = OBD_CONNECT2_LOCKAHEAD |
 479                                    OBD_CONNECT2_INC_XID;
 480
 481         if (!OBD_FAIL_CHECK(OBD_FAIL_OSC_CONNECT_GRANT_PARAM))
 482                 data->ocd_connect_flags |= OBD_CONNECT_GRANT_PARAM;
 483
 484         /* OBD_CONNECT_CKSUM should always be set, even if checksums are
 485          * disabled by default, because it can still be enabled on the
 486          * fly via /sys. As a consequence, we still need to come to an
 487          * agreement on the supported algorithms at connect time
 488          */
 489         data->ocd_connect_flags |= OBD_CONNECT_CKSUM;
 490
 491         if (OBD_FAIL_CHECK(OBD_FAIL_OSC_CKSUM_ADLER_ONLY))
 492                 data->ocd_cksum_types = OBD_CKSUM_ADLER;
 493         else
 494                 data->ocd_cksum_types = obd_cksum_types_supported_client();
 495
 496 #ifdef HAVE_LRU_RESIZE_SUPPORT
 497         data->ocd_connect_flags |= OBD_CONNECT_LRU_RESIZE;
 498 #endif
 499         /* always ping even if server suppress_pings */
 500         if (sbi->ll_flags & LL_SBI_ALWAYS_PING)
 501                 data->ocd_connect_flags &= ~OBD_CONNECT_PINGLESS;
 502
 503         if (ll_sbi_has_encrypt(sbi))
 504                 obd_connect_set_enc(data);
 505
 506         CDEBUG(D_RPCTRACE, "ocd_connect_flags: %#llx ocd_version: %d "
 507                "ocd_grant: %d\n", data->ocd_connect_flags,
 508                data->ocd_version, data->ocd_grant);
 509
 510         sbi->ll_dt_obd->obd_upcall.onu_owner = &sbi->ll_lco;
 511         sbi->ll_dt_obd->obd_upcall.onu_upcall = cl_ocd_update;
 512
 513         data->ocd_brw_size = DT_MAX_BRW_SIZE;
 514
 515         err = obd_connect(NULL, &sbi->ll_dt_exp, sbi->ll_dt_obd,
 516                           &sbi->ll_sb_uuid, data, sbi->ll_cache);
 517         if (err == -EBUSY) {
 518                 LCONSOLE_ERROR_MSG(0x150, "An OST (dt %s) is performing "
 519                                    "recovery, of which this client is not a "
 520                                    "part.  Please wait for recovery to "
 521                                    "complete, abort, or time out.\n", dt);
 522                 GOTO(out_md, err);
 523         } else if (err) {
 524                 CERROR("%s: Cannot connect to %s: rc = %d\n",
 525                        sbi->ll_dt_exp->exp_obd->obd_name, dt, err);
 526                 GOTO(out_md, err);
 527         }
 528
 529         if (ll_sbi_has_encrypt(sbi) &&
 530             !obd_connect_has_enc(&sbi->ll_dt_obd->u.lov.lov_ocd)) {
 531                 if (ll_sbi_has_test_dummy_encryption(sbi))
 532                         LCONSOLE_WARN("%s: server %s does not support encryption feature, encryption deactivated.\n",
 533                                       sbi->ll_fsname, dt);
 534                 ll_sbi_set_encrypt(sbi, false);
 535         } else if (ll_sbi_has_test_dummy_encryption(sbi)) {
 536                 LCONSOLE_WARN("Test dummy encryption mode enabled\n");
 537         }
 538
 539         sbi->ll_dt_exp->exp_connect_data = *data;
 540
 541         /* Don't change value if it was specified in the config log */
 542         if (sbi->ll_ra_info.ra_max_read_ahead_whole_pages == -1) {
 543                 sbi->ll_ra_info.ra_max_read_ahead_whole_pages =
 544                         max_t(unsigned long, SBI_DEFAULT_READ_AHEAD_WHOLE_MAX,
 545                               (data->ocd_brw_size >> PAGE_SHIFT));
 546                 if (sbi->ll_ra_info.ra_max_read_ahead_whole_pages >
 547                     sbi->ll_ra_info.ra_max_pages_per_file)
 548                         sbi->ll_ra_info.ra_max_read_ahead_whole_pages =
 549                                 sbi->ll_ra_info.ra_max_pages_per_file;
 550         }
 551
 552         err = obd_fid_init(sbi->ll_dt_exp->exp_obd, sbi->ll_dt_exp,
 553                            LUSTRE_SEQ_METADATA);
 554         if (err) {
 555                 CERROR("%s: Can't init data layer FID infrastructure, "
 556                        "rc = %d\n", sbi->ll_dt_exp->exp_obd->obd_name, err);
 557                 GOTO(out_dt, err);
 558         }
 559
 560         mutex_lock(&sbi->ll_lco.lco_lock);
 561         sbi->ll_lco.lco_flags = data->ocd_connect_flags;
 562         sbi->ll_lco.lco_md_exp = sbi->ll_md_exp;
 563         sbi->ll_lco.lco_dt_exp = sbi->ll_dt_exp;
 564         mutex_unlock(&sbi->ll_lco.lco_lock);
 565
 566         fid_zero(&sbi->ll_root_fid);
 567         err = md_get_root(sbi->ll_md_exp, get_mount_fileset(sb),
 568                            &sbi->ll_root_fid);
 569         if (err) {
 570                 CERROR("cannot mds_connect: rc = %d\n", err);
 571                 GOTO(out_lock_cn_cb, err);
 572         }
 573         if (!fid_is_sane(&sbi->ll_root_fid)) {
 574                 CERROR("%s: Invalid root fid "DFID" during mount\n",
 575                        sbi->ll_md_exp->exp_obd->obd_name,
 576                        PFID(&sbi->ll_root_fid));
 577                 GOTO(out_lock_cn_cb, err = -EINVAL);
 578         }
 579         CDEBUG(D_SUPER, "rootfid "DFID"\n", PFID(&sbi->ll_root_fid));
 580
 581         sb->s_op = &lustre_super_operations;
 582         sb->s_xattr = ll_xattr_handlers;
 583 #if THREAD_SIZE >= 8192 /*b=17630*/
 584         sb->s_export_op = &lustre_export_operations;
 585 #endif
 586 #ifdef HAVE_LUSTRE_CRYPTO
 587         llcrypt_set_ops(sb, &lustre_cryptops);
 588 #endif
 589
 590         /* make root inode
 591          * XXX: move this to after cbd setup? */
 592         valid = OBD_MD_FLGETATTR | OBD_MD_FLBLOCKS | OBD_MD_FLMODEASIZE;
 593         if (sbi->ll_flags & LL_SBI_ACL)
 594                 valid |= OBD_MD_FLACL;
 595
 596         OBD_ALLOC_PTR(op_data);
 597         if (op_data == NULL)
 598                 GOTO(out_lock_cn_cb, err = -ENOMEM);
 599
 600         op_data->op_fid1 = sbi->ll_root_fid;
 601         op_data->op_mode = 0;
 602         op_data->op_valid = valid;
 603
 604         err = md_getattr(sbi->ll_md_exp, op_data, &request);
 605
 606         OBD_FREE_PTR(op_data);
 607         if (err) {
 608                 CERROR("%s: md_getattr failed for root: rc = %d\n",
 609                        sbi->ll_md_exp->exp_obd->obd_name, err);
 610                 GOTO(out_lock_cn_cb, err);
 611         }
 612
 613         err = md_get_lustre_md(sbi->ll_md_exp, request, sbi->ll_dt_exp,
 614                                sbi->ll_md_exp, &lmd);
 615         if (err) {
 616                 CERROR("failed to understand root inode md: rc = %d\n", err);
 617                 ptlrpc_req_finished(request);
 618                 GOTO(out_lock_cn_cb, err);
 619         }
 620
 621         LASSERT(fid_is_sane(&sbi->ll_root_fid));
 622         root = ll_iget(sb, cl_fid_build_ino(&sbi->ll_root_fid,
 623                                             sbi->ll_flags & LL_SBI_32BIT_API),
 624                        &lmd);
 625         md_free_lustre_md(sbi->ll_md_exp, &lmd);
 626         ptlrpc_req_finished(request);
 627
 628         if (IS_ERR(root)) {
 629 #ifdef CONFIG_LUSTRE_FS_POSIX_ACL
 630                 if (lmd.posix_acl) {
 631                         posix_acl_release(lmd.posix_acl);
 632                         lmd.posix_acl = NULL;
 633                 }
 634 #endif
 635                 err = IS_ERR(root) ? PTR_ERR(root) : -EBADF;
 636                 root = NULL;
 637                 CERROR("%s: bad ll_iget() for root: rc = %d\n",
 638                        sbi->ll_fsname, err);
 639                 GOTO(out_root, err);
 640         }
 641
 642         checksum = sbi->ll_flags & LL_SBI_CHECKSUM;
 643         if (sbi->ll_checksum_set) {
 644                 err = obd_set_info_async(NULL, sbi->ll_dt_exp,
 645                                          sizeof(KEY_CHECKSUM), KEY_CHECKSUM,
 646                                          sizeof(checksum), &checksum, NULL);
 647                 if (err) {
 648                         CERROR("%s: Set checksum failed: rc = %d\n",
 649                                sbi->ll_dt_exp->exp_obd->obd_name, err);
 650                         GOTO(out_root, err);
 651                 }
 652         }
 653         cl_sb_init(sb);
 654
 655         sb->s_root = d_make_root(root);
 656         if (sb->s_root == NULL) {
 657                 err = -ENOMEM;
 658                 CERROR("%s: can't make root dentry: rc = %d\n",
 659                        sbi->ll_fsname, err);
 660                 GOTO(out_root, err);
 661         }
 662
 663         sbi->ll_sdev_orig = sb->s_dev;
 664
 665         /* We set sb->s_dev equal on all lustre clients in order to support
 666          * NFS export clustering.  NFSD requires that the FSID be the same
 667          * on all clients. */
 668         /* s_dev is also used in lt_compare() to compare two fs, but that is
 669          * only a node-local comparison. */
 670         uuid = obd_get_uuid(sbi->ll_md_exp);
 671         if (uuid != NULL)
 672                 sb->s_dev = get_uuid2int(uuid->uuid, strlen(uuid->uuid));
 673
 674         if (data != NULL)
 675                 OBD_FREE_PTR(data);
 676         if (osfs != NULL)
 677                 OBD_FREE_PTR(osfs);
 678
 679         if (sbi->ll_dt_obd) {
 680                 err = sysfs_create_link(&sbi->ll_kset.kobj,
 681                                         &sbi->ll_dt_obd->obd_kset.kobj,
 682                                         sbi->ll_dt_obd->obd_type->typ_name);
 683                 if (err < 0) {
 684                         CERROR("%s: could not register %s in llite: rc = %d\n",
 685                                dt, sbi->ll_fsname, err);
 686                         err = 0;
 687                 }
 688         }
 689
 690         if (sbi->ll_md_obd) {
 691                 err = sysfs_create_link(&sbi->ll_kset.kobj,
 692                                         &sbi->ll_md_obd->obd_kset.kobj,
 693                                         sbi->ll_md_obd->obd_type->typ_name);
 694                 if (err < 0) {
 695                         CERROR("%s: could not register %s in llite: rc = %d\n",
 696                                md, sbi->ll_fsname, err);
 697                         err = 0;
 698                 }
 699         }
 700
 701         RETURN(err);
 702 out_root:
 703         if (root)
 704                 iput(root);
 705 out_lock_cn_cb:
 706         obd_fid_fini(sbi->ll_dt_exp->exp_obd);
 707 out_dt:
 708         obd_disconnect(sbi->ll_dt_exp);
 709         sbi->ll_dt_exp = NULL;
 710         sbi->ll_dt_obd = NULL;
 711 out_md_fid:
 712         obd_fid_fini(sbi->ll_md_exp->exp_obd);
 713 out_md:
 714         obd_disconnect(sbi->ll_md_exp);
 715         sbi->ll_md_exp = NULL;
 716         sbi->ll_md_obd = NULL;
 717 out:
 718         if (data != NULL)
 719                 OBD_FREE_PTR(data);
 720         if (osfs != NULL)
 721                 OBD_FREE_PTR(osfs);
 722         return err;
 723 }
 724
 725 int ll_get_max_mdsize(struct ll_sb_info *sbi, int *lmmsize)
 726 {
 727         int size, rc;
 728
 729         size = sizeof(*lmmsize);
 730         rc = obd_get_info(NULL, sbi->ll_dt_exp, sizeof(KEY_MAX_EASIZE),
 731                           KEY_MAX_EASIZE, &size, lmmsize);
 732         if (rc != 0) {
 733                 CERROR("%s: cannot get max LOV EA size: rc = %d\n",
 734                        sbi->ll_dt_exp->exp_obd->obd_name, rc);
 735                 RETURN(rc);
 736         }
 737
 738         CDEBUG(D_INFO, "max LOV ea size: %d\n", *lmmsize);
 739
 740         size = sizeof(int);
 741         rc = obd_get_info(NULL, sbi->ll_md_exp, sizeof(KEY_MAX_EASIZE),
 742                           KEY_MAX_EASIZE, &size, lmmsize);
 743         if (rc)
 744                 CERROR("Get max mdsize error rc %d\n", rc);
 745
 746         CDEBUG(D_INFO, "max LMV ea size: %d\n", *lmmsize);
 747
 748         RETURN(rc);
 749 }
 750
 751 /**
 752  * Get the value of the default_easize parameter.
 753  *
 754  * \see client_obd::cl_default_mds_easize
 755  *
 756  * \param[in] sbi       superblock info for this filesystem
 757  * \param[out] lmmsize  pointer to storage location for value
 758  *
 759  * \retval 0            on success
 760  * \retval negative     negated errno on failure
 761  */
 762 int ll_get_default_mdsize(struct ll_sb_info *sbi, int *lmmsize)
 763 {
 764         int size, rc;
 765
 766         size = sizeof(int);
 767         rc = obd_get_info(NULL, sbi->ll_md_exp, sizeof(KEY_DEFAULT_EASIZE),
 768                          KEY_DEFAULT_EASIZE, &size, lmmsize);
 769         if (rc)
 770                 CERROR("Get default mdsize error rc %d\n", rc);
 771
 772         RETURN(rc);
 773 }
 774
 775 /**
 776  * Set the default_easize parameter to the given value.
 777  *
 778  * \see client_obd::cl_default_mds_easize
 779  *
 780  * \param[in] sbi       superblock info for this filesystem
 781  * \param[in] lmmsize   the size to set
 782  *
 783  * \retval 0            on success
 784  * \retval negative     negated errno on failure
 785  */
 786 int ll_set_default_mdsize(struct ll_sb_info *sbi, int lmmsize)
 787 {
 788         int rc;
 789
 790         if (lmmsize < sizeof(struct lov_mds_md) ||
 791             lmmsize > OBD_MAX_DEFAULT_EA_SIZE)
 792                 return -EINVAL;
 793
 794         rc = obd_set_info_async(NULL, sbi->ll_md_exp,
 795                                 sizeof(KEY_DEFAULT_EASIZE), KEY_DEFAULT_EASIZE,
 796                                 sizeof(int), &lmmsize, NULL);
 797
 798         RETURN(rc);
 799 }
 800
 801 static void client_common_put_super(struct super_block *sb)
 802 {
 803         struct ll_sb_info *sbi = ll_s2sbi(sb);
 804         ENTRY;
 805
 806         cl_sb_fini(sb);
 807
 808         obd_fid_fini(sbi->ll_dt_exp->exp_obd);
 809         obd_disconnect(sbi->ll_dt_exp);
 810         sbi->ll_dt_exp = NULL;
 811
 812         ll_debugfs_unregister_super(sb);
 813
 814         obd_fid_fini(sbi->ll_md_exp->exp_obd);
 815         obd_disconnect(sbi->ll_md_exp);
 816         sbi->ll_md_exp = NULL;
 817
 818         EXIT;
 819 }
 820
 821 void ll_kill_super(struct super_block *sb)
 822 {
 823         struct ll_sb_info *sbi;
 824         ENTRY;
 825
 826         /* not init sb ?*/
 827         if (!(sb->s_flags & SB_ACTIVE))
 828                 return;
 829
 830         sbi = ll_s2sbi(sb);
 831         /* we need restore s_dev from changed for clustred NFS before put_super
 832          * because new kernels have cached s_dev and change sb->s_dev in
 833          * put_super not affected real removing devices */
 834         if (sbi) {
 835                 sb->s_dev = sbi->ll_sdev_orig;
 836
 837                 /* wait running statahead threads to quit */
 838                 while (atomic_read(&sbi->ll_sa_running) > 0)
 839                         schedule_timeout_uninterruptible(
 840                                 cfs_time_seconds(1) >> 3);
 841         }
 842
 843         EXIT;
 844 }
 845
 846 static inline int ll_set_opt(const char *opt, char *data, int fl)
 847 {
 848         if (strncmp(opt, data, strlen(opt)) != 0)
 849                 return 0;
 850         else
 851                 return fl;
 852 }
 853
 854 /* non-client-specific mount options are parsed in lmd_parse */
 855 static int ll_options(char *options, struct ll_sb_info *sbi)
 856 {
 857         int tmp;
 858         char *s1 = options, *s2;
 859         int *flags = &sbi->ll_flags;
 860         ENTRY;
 861
 862         if (!options)
 863                 RETURN(0);
 864
 865         CDEBUG(D_CONFIG, "Parsing opts %s\n", options);
 866
 867         while (*s1) {
 868                 CDEBUG(D_SUPER, "next opt=%s\n", s1);
 869                 tmp = ll_set_opt("nolock", s1, LL_SBI_NOLCK);
 870                 if (tmp) {
 871                         *flags |= tmp;
 872                         goto next;
 873                 }
 874                 tmp = ll_set_opt("flock", s1, LL_SBI_FLOCK);
 875                 if (tmp) {
 876                         *flags = (*flags & ~LL_SBI_LOCALFLOCK) | tmp;
 877                         goto next;
 878                 }
 879                 tmp = ll_set_opt("localflock", s1, LL_SBI_LOCALFLOCK);
 880                 if (tmp) {
 881                         *flags = (*flags & ~LL_SBI_FLOCK) | tmp;
 882                         goto next;
 883                 }
 884                 tmp = ll_set_opt("noflock", s1, LL_SBI_FLOCK|LL_SBI_LOCALFLOCK);
 885                 if (tmp) {
 886                         *flags &= ~tmp;
 887                         goto next;
 888                 }
 889                 tmp = ll_set_opt("user_xattr", s1, LL_SBI_USER_XATTR);
 890                 if (tmp) {
 891                         *flags |= tmp;
 892                         goto next;
 893                 }
 894                 tmp = ll_set_opt("nouser_xattr", s1, LL_SBI_USER_XATTR);
 895                 if (tmp) {
 896                         *flags &= ~tmp;
 897                         goto next;
 898                 }
 899                 tmp = ll_set_opt("context", s1, 1);
 900                 if (tmp)
 901                         goto next;
 902                 tmp = ll_set_opt("fscontext", s1, 1);
 903                 if (tmp)
 904                         goto next;
 905                 tmp = ll_set_opt("defcontext", s1, 1);
 906                 if (tmp)
 907                         goto next;
 908                 tmp = ll_set_opt("rootcontext", s1, 1);
 909                 if (tmp)
 910                         goto next;
 911                 tmp = ll_set_opt("user_fid2path", s1, LL_SBI_USER_FID2PATH);
 912                 if (tmp) {
 913                         *flags |= tmp;
 914                         goto next;
 915                 }
 916                 tmp = ll_set_opt("nouser_fid2path", s1, LL_SBI_USER_FID2PATH);
 917                 if (tmp) {
 918                         *flags &= ~tmp;
 919                         goto next;
 920                 }
 921
 922                 tmp = ll_set_opt("checksum", s1, LL_SBI_CHECKSUM);
 923                 if (tmp) {
 924                         *flags |= tmp;
 925                         sbi->ll_checksum_set = 1;
 926                         goto next;
 927                 }
 928                 tmp = ll_set_opt("nochecksum", s1, LL_SBI_CHECKSUM);
 929                 if (tmp) {
 930                         *flags &= ~tmp;
 931                         sbi->ll_checksum_set = 1;
 932                         goto next;
 933                 }
 934                 tmp = ll_set_opt("lruresize", s1, LL_SBI_LRU_RESIZE);
 935                 if (tmp) {
 936                         *flags |= tmp;
 937                         goto next;
 938                 }
 939                 tmp = ll_set_opt("nolruresize", s1, LL_SBI_LRU_RESIZE);
 940                 if (tmp) {
 941                         *flags &= ~tmp;
 942                         goto next;
 943                 }
 944                 tmp = ll_set_opt("lazystatfs", s1, LL_SBI_LAZYSTATFS);
 945                 if (tmp) {
 946                         *flags |= tmp;
 947                         goto next;
 948                 }
 949                 tmp = ll_set_opt("nolazystatfs", s1, LL_SBI_LAZYSTATFS);
 950                 if (tmp) {
 951                         *flags &= ~tmp;
 952                         goto next;
 953                 }
 954                 tmp = ll_set_opt("32bitapi", s1, LL_SBI_32BIT_API);
 955                 if (tmp) {
 956                         *flags |= tmp;
 957                         goto next;
 958                 }
 959                 tmp = ll_set_opt("verbose", s1, LL_SBI_VERBOSE);
 960                 if (tmp) {
 961                         *flags |= tmp;
 962                         goto next;
 963                 }
 964                 tmp = ll_set_opt("noverbose", s1, LL_SBI_VERBOSE);
 965                 if (tmp) {
 966                         *flags &= ~tmp;
 967                         goto next;
 968                 }
 969                 tmp = ll_set_opt("always_ping", s1, LL_SBI_ALWAYS_PING);
 970                 if (tmp) {
 971                         *flags |= tmp;
 972                         goto next;
 973                 }
 974                 tmp = ll_set_opt("test_dummy_encryption", s1,
 975                                  LL_SBI_TEST_DUMMY_ENCRYPTION);
 976                 if (tmp) {
 977 #ifdef HAVE_LUSTRE_CRYPTO
 978                         *flags |= tmp;
 979 #else
 980                         LCONSOLE_WARN("Test dummy encryption mount option ignored: encryption not supported\n");
 981 #endif
 982                         goto next;
 983                 }
 984                 tmp = ll_set_opt("noencrypt", s1, LL_SBI_ENCRYPT);
 985                 if (tmp) {
 986 #ifdef HAVE_LUSTRE_CRYPTO
 987                         *flags &= ~tmp;
 988 #else
 989                         LCONSOLE_WARN("noencrypt mount option ignored: encryption not supported\n");
 990 #endif
 991                         goto next;
 992                 }
 993                 LCONSOLE_ERROR_MSG(0x152, "Unknown option '%s', won't mount.\n",
 994                                    s1);
 995                 RETURN(-EINVAL);
 996
 997 next:
 998                 /* Find next opt */
 999                 s2 = strchr(s1, ',');
1000                 if (s2 == NULL)
1001                         break;
1002                 s1 = s2 + 1;
1003         }
1004         RETURN(0);
1005 }
1006
1007 void ll_lli_init(struct ll_inode_info *lli)
1008 {
1009         lli->lli_inode_magic = LLI_INODE_MAGIC;
1010         lli->lli_flags = 0;
1011         spin_lock_init(&lli->lli_lock);
1012         lli->lli_posix_acl = NULL;
1013         /* Do not set lli_fid, it has been initialized already. */
1014         fid_zero(&lli->lli_pfid);
1015         lli->lli_mds_read_och = NULL;
1016         lli->lli_mds_write_och = NULL;
1017         lli->lli_mds_exec_och = NULL;
1018         lli->lli_open_fd_read_count = 0;
1019         lli->lli_open_fd_write_count = 0;
1020         lli->lli_open_fd_exec_count = 0;
1021         mutex_init(&lli->lli_och_mutex);
1022         spin_lock_init(&lli->lli_agl_lock);
1023         spin_lock_init(&lli->lli_layout_lock);
1024         ll_layout_version_set(lli, CL_LAYOUT_GEN_NONE);
1025         lli->lli_clob = NULL;
1026
1027         init_rwsem(&lli->lli_xattrs_list_rwsem);
1028         mutex_init(&lli->lli_xattrs_enq_lock);
1029
1030         LASSERT(lli->lli_vfs_inode.i_mode != 0);
1031         if (S_ISDIR(lli->lli_vfs_inode.i_mode)) {
1032                 lli->lli_opendir_key = NULL;
1033                 lli->lli_sai = NULL;
1034                 spin_lock_init(&lli->lli_sa_lock);
1035                 lli->lli_opendir_pid = 0;
1036                 lli->lli_sa_enabled = 0;
1037                 init_rwsem(&lli->lli_lsm_sem);
1038         } else {
1039                 mutex_init(&lli->lli_size_mutex);
1040                 lli->lli_symlink_name = NULL;
1041                 ll_trunc_sem_init(&lli->lli_trunc_sem);
1042                 range_lock_tree_init(&lli->lli_write_tree);
1043                 init_rwsem(&lli->lli_glimpse_sem);
1044                 lli->lli_glimpse_time = ktime_set(0, 0);
1045                 INIT_LIST_HEAD(&lli->lli_agl_list);
1046                 lli->lli_agl_index = 0;
1047                 lli->lli_async_rc = 0;
1048                 spin_lock_init(&lli->lli_heat_lock);
1049                 obd_heat_clear(lli->lli_heat_instances, OBD_HEAT_COUNT);
1050                 lli->lli_heat_flags = 0;
1051                 mutex_init(&lli->lli_pcc_lock);
1052                 lli->lli_pcc_state = PCC_STATE_FL_NONE;
1053                 lli->lli_pcc_inode = NULL;
1054                 lli->lli_pcc_dsflags = PCC_DATASET_INVALID;
1055                 lli->lli_pcc_generation = 0;
1056                 mutex_init(&lli->lli_group_mutex);
1057                 lli->lli_group_users = 0;
1058                 lli->lli_group_gid = 0;
1059         }
1060         mutex_init(&lli->lli_layout_mutex);
1061         memset(lli->lli_jobid, 0, sizeof(lli->lli_jobid));
1062 }
1063
1064 #define MAX_STRING_SIZE 128
1065
1066 #ifndef HAVE_SUPER_SETUP_BDI_NAME
1067
1068 #define LSI_BDI_INITIALIZED     0x00400000
1069
1070 #ifndef HAVE_BDI_CAP_MAP_COPY
1071 # define BDI_CAP_MAP_COPY       0
1072 #endif
1073
1074 static int super_setup_bdi_name(struct super_block *sb, char *fmt, ...)
1075 {
1076         struct  lustre_sb_info *lsi = s2lsi(sb);
1077         char buf[MAX_STRING_SIZE];
1078         va_list args;
1079         int err;
1080
1081         err = bdi_init(&lsi->lsi_bdi);
1082         if (err)
1083                 return err;
1084
1085         lsi->lsi_flags |= LSI_BDI_INITIALIZED;
1086         lsi->lsi_bdi.capabilities = BDI_CAP_MAP_COPY;
1087         lsi->lsi_bdi.name = "lustre";
1088         va_start(args, fmt);
1089         vsnprintf(buf, MAX_STRING_SIZE, fmt, args);
1090         va_end(args);
1091         err = bdi_register(&lsi->lsi_bdi, NULL, "%s", buf);
1092         va_end(args);
1093         if (!err)
1094                 sb->s_bdi = &lsi->lsi_bdi;
1095
1096         return err;
1097 }
1098 #endif /* !HAVE_SUPER_SETUP_BDI_NAME */
1099
1100 int ll_fill_super(struct super_block *sb)
1101 {
1102         struct  lustre_profile *lprof = NULL;
1103         struct  lustre_sb_info *lsi = s2lsi(sb);
1104         struct  ll_sb_info *sbi = NULL;
1105         char    *dt = NULL, *md = NULL;
1106         char    *profilenm = get_profile_name(sb);
1107         struct config_llog_instance *cfg;
1108         /* %p for void* in printf needs 16+2 characters: 0xffffffffffffffff */
1109         const int instlen = LUSTRE_MAXINSTANCE + 2;
1110         unsigned long cfg_instance = ll_get_cfg_instance(sb);
1111         char name[MAX_STRING_SIZE];
1112         int md_len = 0;
1113         int dt_len = 0;
1114         uuid_t uuid;
1115         char *ptr;
1116         int len;
1117         int err;
1118
1119         ENTRY;
1120         /* for ASLR, to map between cfg_instance and hashed ptr */
1121         CDEBUG(D_VFSTRACE, "VFS Op: cfg_instance %s-%016lx (sb %p)\n",
1122                profilenm, cfg_instance, sb);
1123
1124         OBD_ALLOC_PTR(cfg);
1125         if (cfg == NULL)
1126                 GOTO(out_free_cfg, err = -ENOMEM);
1127
1128         /* client additional sb info */
1129         lsi->lsi_llsbi = sbi = ll_init_sbi();
1130         if (IS_ERR(sbi))
1131                 GOTO(out_free_cfg, err = PTR_ERR(sbi));
1132
1133         err = ll_options(lsi->lsi_lmd->lmd_opts, sbi);
1134         if (err)
1135                 GOTO(out_free_cfg, err);
1136
1137         /* kernel >= 2.6.38 store dentry operations in sb->s_d_op. */
1138         sb->s_d_op = &ll_d_ops;
1139
1140         /* UUID handling */
1141         generate_random_uuid(uuid.b);
1142         snprintf(sbi->ll_sb_uuid.uuid, sizeof(sbi->ll_sb_uuid), "%pU", uuid.b);
1143
1144         CDEBUG(D_CONFIG, "llite sb uuid: %s\n", sbi->ll_sb_uuid.uuid);
1145
1146         /* Get fsname */
1147         len = strlen(profilenm);
1148         ptr = strrchr(profilenm, '-');
1149         if (ptr && (strcmp(ptr, "-client") == 0))
1150                 len -= 7;
1151
1152         if (len > LUSTRE_MAXFSNAME) {
1153                 if (unlikely(len >= MAX_STRING_SIZE))
1154                         len = MAX_STRING_SIZE - 1;
1155                 strncpy(name, profilenm, len);
1156                 name[len] = '\0';
1157                 err = -ENAMETOOLONG;
1158                 CERROR("%s: fsname longer than %u characters: rc = %d\n",
1159                        name, LUSTRE_MAXFSNAME, err);
1160                 GOTO(out_free_cfg, err);
1161         }
1162         strncpy(sbi->ll_fsname, profilenm, len);
1163         sbi->ll_fsname[len] = '\0';
1164
1165         /* Mount info */
1166         snprintf(name, sizeof(name), "%.*s-%016lx", len,
1167                  profilenm, cfg_instance);
1168
1169         err = super_setup_bdi_name(sb, "%s", name);
1170         if (err)
1171                 GOTO(out_free_cfg, err);
1172
1173         /* Call ll_debugfs_register_super() before lustre_process_log()
1174          * so that "llite.*.*" params can be processed correctly.
1175          */
1176         err = ll_debugfs_register_super(sb, name);
1177         if (err < 0) {
1178                 CERROR("%s: could not register mountpoint in llite: rc = %d\n",
1179                        sbi->ll_fsname, err);
1180                 err = 0;
1181         }
1182
1183         /* The cfg_instance is a value unique to this super, in case some
1184          * joker tries to mount the same fs at two mount points.
1185          */
1186         cfg->cfg_instance = cfg_instance;
1187         cfg->cfg_uuid = lsi->lsi_llsbi->ll_sb_uuid;
1188         cfg->cfg_callback = class_config_llog_handler;
1189         cfg->cfg_sub_clds = CONFIG_SUB_CLIENT;
1190         /* set up client obds */
1191         err = lustre_process_log(sb, profilenm, cfg);
1192         if (err < 0)
1193                 GOTO(out_debugfs, err);
1194
1195         /* Profile set with LCFG_MOUNTOPT so we can find our mdc and osc obds */
1196         lprof = class_get_profile(profilenm);
1197         if (lprof == NULL) {
1198                 LCONSOLE_ERROR_MSG(0x156, "The client profile '%s' could not be"
1199                                    " read from the MGS.  Does that filesystem "
1200                                    "exist?\n", profilenm);
1201                 GOTO(out_debugfs, err = -EINVAL);
1202         }
1203         CDEBUG(D_CONFIG, "Found profile %s: mdc=%s osc=%s\n", profilenm,
1204                lprof->lp_md, lprof->lp_dt);
1205
1206         dt_len = strlen(lprof->lp_dt) + instlen + 2;
1207         OBD_ALLOC(dt, dt_len);
1208         if (!dt)
1209                 GOTO(out_profile, err = -ENOMEM);
1210         snprintf(dt, dt_len - 1, "%s-%016lx", lprof->lp_dt, cfg_instance);
1211
1212         md_len = strlen(lprof->lp_md) + instlen + 2;
1213         OBD_ALLOC(md, md_len);
1214         if (!md)
1215                 GOTO(out_free_dt, err = -ENOMEM);
1216         snprintf(md, md_len - 1, "%s-%016lx", lprof->lp_md, cfg_instance);
1217
1218         /* connections, registrations, sb setup */
1219         err = client_common_fill_super(sb, md, dt);
1220         if (err < 0)
1221                 GOTO(out_free_md, err);
1222
1223         sbi->ll_client_common_fill_super_succeeded = 1;
1224
1225 out_free_md:
1226         if (md)
1227                 OBD_FREE(md, md_len);
1228 out_free_dt:
1229         if (dt)
1230                 OBD_FREE(dt, dt_len);
1231 out_profile:
1232         if (lprof)
1233                 class_put_profile(lprof);
1234 out_debugfs:
1235         if (err < 0)
1236                 ll_debugfs_unregister_super(sb);
1237 out_free_cfg:
1238         if (cfg)
1239                 OBD_FREE_PTR(cfg);
1240
1241         if (err)
1242                 ll_put_super(sb);
1243         else if (sbi->ll_flags & LL_SBI_VERBOSE)
1244                 LCONSOLE_WARN("Mounted %s\n", profilenm);
1245         RETURN(err);
1246 } /* ll_fill_super */
1247
1248 void ll_put_super(struct super_block *sb)
1249 {
1250         struct config_llog_instance cfg, params_cfg;
1251         struct obd_device *obd;
1252         struct lustre_sb_info *lsi = s2lsi(sb);
1253         struct ll_sb_info *sbi = ll_s2sbi(sb);
1254         char *profilenm = get_profile_name(sb);
1255         unsigned long cfg_instance = ll_get_cfg_instance(sb);
1256         long ccc_count;
1257         int next, force = 1, rc = 0;
1258         ENTRY;
1259
1260         if (IS_ERR(sbi))
1261                 GOTO(out_no_sbi, 0);
1262
1263         /* Should replace instance_id with something better for ASLR */
1264         CDEBUG(D_VFSTRACE, "VFS Op: cfg_instance %s-%016lx (sb %p)\n",
1265                profilenm, cfg_instance, sb);
1266
1267         cfg.cfg_instance = cfg_instance;
1268         lustre_end_log(sb, profilenm, &cfg);
1269
1270         params_cfg.cfg_instance = cfg_instance;
1271         lustre_end_log(sb, PARAMS_FILENAME, &params_cfg);
1272
1273         if (sbi->ll_md_exp) {
1274                 obd = class_exp2obd(sbi->ll_md_exp);
1275                 if (obd)
1276                         force = obd->obd_force;
1277         }
1278
1279         /* Wait for unstable pages to be committed to stable storage */
1280         if (force == 0) {
1281                 rc = l_wait_event_abortable(
1282                         sbi->ll_cache->ccc_unstable_waitq,
1283                         atomic_long_read(&sbi->ll_cache->ccc_unstable_nr) == 0);
1284         }
1285
1286         ccc_count = atomic_long_read(&sbi->ll_cache->ccc_unstable_nr);
1287         if (force == 0 && rc != -ERESTARTSYS)
1288                 LASSERTF(ccc_count == 0, "count: %li\n", ccc_count);
1289
1290         /* We need to set force before the lov_disconnect in
1291          * lustre_common_put_super, since l_d cleans up osc's as well.
1292          */
1293         if (force) {
1294                 next = 0;
1295                 while ((obd = class_devices_in_group(&sbi->ll_sb_uuid,
1296                                                      &next)) != NULL) {
1297                         obd->obd_force = force;
1298                 }
1299         }
1300
1301         if (sbi->ll_client_common_fill_super_succeeded) {
1302                 /* Only if client_common_fill_super succeeded */
1303                 client_common_put_super(sb);
1304         }
1305
1306         next = 0;
1307         while ((obd = class_devices_in_group(&sbi->ll_sb_uuid, &next)))
1308                 class_manual_cleanup(obd);
1309
1310         if (sbi->ll_flags & LL_SBI_VERBOSE)
1311                 LCONSOLE_WARN("Unmounted %s\n", profilenm ? profilenm : "");
1312
1313         if (profilenm)
1314                 class_del_profile(profilenm);
1315
1316 #ifndef HAVE_SUPER_SETUP_BDI_NAME
1317         if (lsi->lsi_flags & LSI_BDI_INITIALIZED) {
1318                 bdi_destroy(&lsi->lsi_bdi);
1319                 lsi->lsi_flags &= ~LSI_BDI_INITIALIZED;
1320         }
1321 #endif
1322
1323         ll_free_sbi(sb);
1324         lsi->lsi_llsbi = NULL;
1325 out_no_sbi:
1326         lustre_common_put_super(sb);
1327
1328         cl_env_cache_purge(~0);
1329
1330         module_put(THIS_MODULE);
1331
1332         EXIT;
1333 } /* client_put_super */
1334
1335 struct inode *ll_inode_from_resource_lock(struct ldlm_lock *lock)
1336 {
1337         struct inode *inode = NULL;
1338
1339         /* NOTE: we depend on atomic igrab() -bzzz */
1340         lock_res_and_lock(lock);
1341         if (lock->l_resource->lr_lvb_inode) {
1342                 struct ll_inode_info * lli;
1343                 lli = ll_i2info(lock->l_resource->lr_lvb_inode);
1344                 if (lli->lli_inode_magic == LLI_INODE_MAGIC) {
1345                         inode = igrab(lock->l_resource->lr_lvb_inode);
1346                 } else {
1347                         inode = lock->l_resource->lr_lvb_inode;
1348                         LDLM_DEBUG_LIMIT(inode->i_state & I_FREEING ?  D_INFO :
1349                                          D_WARNING, lock, "lr_lvb_inode %p is "
1350                                          "bogus: magic %08x",
1351                                          lock->l_resource->lr_lvb_inode,
1352                                          lli->lli_inode_magic);
1353                         inode = NULL;
1354                 }
1355         }
1356         unlock_res_and_lock(lock);
1357         return inode;
1358 }
1359
1360 void ll_dir_clear_lsm_md(struct inode *inode)
1361 {
1362         struct ll_inode_info *lli = ll_i2info(inode);
1363
1364         LASSERT(S_ISDIR(inode->i_mode));
1365
1366         if (lli->lli_lsm_md) {
1367                 lmv_free_memmd(lli->lli_lsm_md);
1368                 lli->lli_lsm_md = NULL;
1369         }
1370
1371         if (lli->lli_default_lsm_md) {
1372                 lmv_free_memmd(lli->lli_default_lsm_md);
1373                 lli->lli_default_lsm_md = NULL;
1374         }
1375 }
1376
1377 static struct inode *ll_iget_anon_dir(struct super_block *sb,
1378                                       const struct lu_fid *fid,
1379                                       struct lustre_md *md)
1380 {
1381         struct ll_sb_info       *sbi = ll_s2sbi(sb);
1382         struct mdt_body         *body = md->body;
1383         struct inode            *inode;
1384         ino_t                   ino;
1385         ENTRY;
1386
1387         ino = cl_fid_build_ino(fid, sbi->ll_flags & LL_SBI_32BIT_API);
1388         inode = iget_locked(sb, ino);
1389         if (inode == NULL) {
1390                 CERROR("%s: failed get simple inode "DFID": rc = -ENOENT\n",
1391                        sbi->ll_fsname, PFID(fid));
1392                 RETURN(ERR_PTR(-ENOENT));
1393         }
1394
1395         if (inode->i_state & I_NEW) {
1396                 struct ll_inode_info *lli = ll_i2info(inode);
1397                 struct lmv_stripe_md *lsm = md->lmv;
1398
1399                 inode->i_mode = (inode->i_mode & ~S_IFMT) |
1400                                 (body->mbo_mode & S_IFMT);
1401                 LASSERTF(S_ISDIR(inode->i_mode), "Not slave inode "DFID"\n",
1402                          PFID(fid));
1403
1404                 inode->i_mtime.tv_sec = 0;
1405                 inode->i_atime.tv_sec = 0;
1406                 inode->i_ctime.tv_sec = 0;
1407                 inode->i_rdev = 0;
1408
1409 #ifdef HAVE_BACKING_DEV_INFO
1410                 /* initializing backing dev info. */
1411                 inode->i_mapping->backing_dev_info =
1412                                                 &s2lsi(inode->i_sb)->lsi_bdi;
1413 #endif
1414                 inode->i_op = &ll_dir_inode_operations;
1415                 inode->i_fop = &ll_dir_operations;
1416                 lli->lli_fid = *fid;
1417                 ll_lli_init(lli);
1418
1419                 LASSERT(lsm != NULL);
1420                 /* master object FID */
1421                 lli->lli_pfid = body->mbo_fid1;
1422                 CDEBUG(D_INODE, "lli %p slave "DFID" master "DFID"\n",
1423                        lli, PFID(fid), PFID(&lli->lli_pfid));
1424                 unlock_new_inode(inode);
1425         }
1426
1427         RETURN(inode);
1428 }
1429
1430 static int ll_init_lsm_md(struct inode *inode, struct lustre_md *md)
1431 {
1432         struct lu_fid *fid;
1433         struct lmv_stripe_md *lsm = md->lmv;
1434         struct ll_inode_info *lli = ll_i2info(inode);
1435         int i;
1436
1437         LASSERT(lsm != NULL);
1438
1439         CDEBUG(D_INODE, "%s: "DFID" set dir layout:\n",
1440                ll_i2sbi(inode)->ll_fsname, PFID(&lli->lli_fid));
1441         lsm_md_dump(D_INODE, lsm);
1442
1443         if (!lmv_dir_striped(lsm))
1444                 goto out;
1445
1446         /* XXX sigh, this lsm_root initialization should be in
1447          * LMV layer, but it needs ll_iget right now, so we
1448          * put this here right now. */
1449         for (i = 0; i < lsm->lsm_md_stripe_count; i++) {
1450                 fid = &lsm->lsm_md_oinfo[i].lmo_fid;
1451                 LASSERT(lsm->lsm_md_oinfo[i].lmo_root == NULL);
1452
1453                 if (!fid_is_sane(fid))
1454                         continue;
1455
1456                 /* Unfortunately ll_iget will call ll_update_inode,
1457                  * where the initialization of slave inode is slightly
1458                  * different, so it reset lsm_md to NULL to avoid
1459                  * initializing lsm for slave inode. */
1460                 lsm->lsm_md_oinfo[i].lmo_root =
1461                                 ll_iget_anon_dir(inode->i_sb, fid, md);
1462                 if (IS_ERR(lsm->lsm_md_oinfo[i].lmo_root)) {
1463                         int rc = PTR_ERR(lsm->lsm_md_oinfo[i].lmo_root);
1464
1465                         lsm->lsm_md_oinfo[i].lmo_root = NULL;
1466                         while (i-- > 0) {
1467                                 iput(lsm->lsm_md_oinfo[i].lmo_root);
1468                                 lsm->lsm_md_oinfo[i].lmo_root = NULL;
1469                         }
1470                         return rc;
1471                 }
1472         }
1473 out:
1474         lli->lli_lsm_md = lsm;
1475
1476         return 0;
1477 }
1478
1479 static void ll_update_default_lsm_md(struct inode *inode, struct lustre_md *md)
1480 {
1481         struct ll_inode_info *lli = ll_i2info(inode);
1482
1483         if (!md->default_lmv) {
1484                 /* clear default lsm */
1485                 if (lli->lli_default_lsm_md) {
1486                         down_write(&lli->lli_lsm_sem);
1487                         if (lli->lli_default_lsm_md) {
1488                                 lmv_free_memmd(lli->lli_default_lsm_md);
1489                                 lli->lli_default_lsm_md = NULL;
1490                         }
1491                         up_write(&lli->lli_lsm_sem);
1492                 }
1493         } else if (lli->lli_default_lsm_md) {
1494                 /* update default lsm if it changes */
1495                 down_read(&lli->lli_lsm_sem);
1496                 if (lli->lli_default_lsm_md &&
1497                     !lsm_md_eq(lli->lli_default_lsm_md, md->default_lmv)) {
1498                         up_read(&lli->lli_lsm_sem);
1499                         down_write(&lli->lli_lsm_sem);
1500                         if (lli->lli_default_lsm_md)
1501                                 lmv_free_memmd(lli->lli_default_lsm_md);
1502                         lli->lli_default_lsm_md = md->default_lmv;
1503                         lsm_md_dump(D_INODE, md->default_lmv);
1504                         md->default_lmv = NULL;
1505                         up_write(&lli->lli_lsm_sem);
1506                 } else {
1507                         up_read(&lli->lli_lsm_sem);
1508                 }
1509         } else {
1510                 /* init default lsm */
1511                 down_write(&lli->lli_lsm_sem);
1512                 lli->lli_default_lsm_md = md->default_lmv;
1513                 lsm_md_dump(D_INODE, md->default_lmv);
1514                 md->default_lmv = NULL;
1515                 up_write(&lli->lli_lsm_sem);
1516         }
1517 }
1518
1519 static int ll_update_lsm_md(struct inode *inode, struct lustre_md *md)
1520 {
1521         struct ll_inode_info *lli = ll_i2info(inode);
1522         struct lmv_stripe_md *lsm = md->lmv;
1523         struct cl_attr  *attr;
1524         int rc = 0;
1525
1526         ENTRY;
1527
1528         LASSERT(S_ISDIR(inode->i_mode));
1529         CDEBUG(D_INODE, "update lsm %p of "DFID"\n", lli->lli_lsm_md,
1530                PFID(ll_inode2fid(inode)));
1531
1532         /* update default LMV */
1533         if (md->default_lmv)
1534                 ll_update_default_lsm_md(inode, md);
1535
1536         /*
1537          * no striped information from request, lustre_md from req does not
1538          * include stripeEA, see ll_md_setattr()
1539          */
1540         if (!lsm)
1541                 RETURN(0);
1542
1543         /*
1544          * normally dir layout doesn't change, only take read lock to check
1545          * that to avoid blocking other MD operations.
1546          */
1547         down_read(&lli->lli_lsm_sem);
1548
1549         /* some current lookup initialized lsm, and unchanged */
1550         if (lli->lli_lsm_md && lsm_md_eq(lli->lli_lsm_md, lsm))
1551                 GOTO(unlock, rc = 0);
1552
1553         /* if dir layout doesn't match, check whether version is increased,
1554          * which means layout is changed, this happens in dir split/merge and
1555          * lfsck.
1556          *
1557          * foreign LMV should not change.
1558          */
1559         if (lli->lli_lsm_md && lmv_dir_striped(lli->lli_lsm_md) &&
1560             lsm->lsm_md_layout_version <=
1561             lli->lli_lsm_md->lsm_md_layout_version) {
1562                 CERROR("%s: "DFID" dir layout mismatch:\n",
1563                        ll_i2sbi(inode)->ll_fsname, PFID(&lli->lli_fid));
1564                 lsm_md_dump(D_ERROR, lli->lli_lsm_md);
1565                 lsm_md_dump(D_ERROR, lsm);
1566                 GOTO(unlock, rc = -EINVAL);
1567         }
1568
1569         up_read(&lli->lli_lsm_sem);
1570         down_write(&lli->lli_lsm_sem);
1571         /* clear existing lsm */
1572         if (lli->lli_lsm_md) {
1573                 lmv_free_memmd(lli->lli_lsm_md);
1574                 lli->lli_lsm_md = NULL;
1575         }
1576
1577         rc = ll_init_lsm_md(inode, md);
1578         up_write(&lli->lli_lsm_sem);
1579
1580         if (rc)
1581                 RETURN(rc);
1582
1583         /* set md->lmv to NULL, so the following free lustre_md will not free
1584          * this lsm.
1585          */
1586         md->lmv = NULL;
1587
1588         /* md_merge_attr() may take long, since lsm is already set, switch to
1589          * read lock.
1590          */
1591         down_read(&lli->lli_lsm_sem);
1592
1593         if (!lmv_dir_striped(lli->lli_lsm_md))
1594                 GOTO(unlock, rc = 0);
1595
1596         OBD_ALLOC_PTR(attr);
1597         if (!attr)
1598                 GOTO(unlock, rc = -ENOMEM);
1599
1600         /* validate the lsm */
1601         rc = md_merge_attr(ll_i2mdexp(inode), lli->lli_lsm_md, attr,
1602                            ll_md_blocking_ast);
1603         if (!rc) {
1604                 if (md->body->mbo_valid & OBD_MD_FLNLINK)
1605                         md->body->mbo_nlink = attr->cat_nlink;
1606                 if (md->body->mbo_valid & OBD_MD_FLSIZE)
1607                         md->body->mbo_size = attr->cat_size;
1608                 if (md->body->mbo_valid & OBD_MD_FLATIME)
1609                         md->body->mbo_atime = attr->cat_atime;
1610                 if (md->body->mbo_valid & OBD_MD_FLCTIME)
1611                         md->body->mbo_ctime = attr->cat_ctime;
1612                 if (md->body->mbo_valid & OBD_MD_FLMTIME)
1613                         md->body->mbo_mtime = attr->cat_mtime;
1614         }
1615
1616         OBD_FREE_PTR(attr);
1617         GOTO(unlock, rc);
1618 unlock:
1619         up_read(&lli->lli_lsm_sem);
1620
1621         return rc;
1622 }
1623
1624 void ll_clear_inode(struct inode *inode)
1625 {
1626         struct ll_inode_info *lli = ll_i2info(inode);
1627         struct ll_sb_info *sbi = ll_i2sbi(inode);
1628
1629         ENTRY;
1630
1631         CDEBUG(D_VFSTRACE, "VFS Op:inode="DFID"(%p)\n",
1632                PFID(ll_inode2fid(inode)), inode);
1633
1634         if (S_ISDIR(inode->i_mode)) {
1635                 /* these should have been cleared in ll_file_release */
1636                 LASSERT(lli->lli_opendir_key == NULL);
1637                 LASSERT(lli->lli_sai == NULL);
1638                 LASSERT(lli->lli_opendir_pid == 0);
1639         } else {
1640                 pcc_inode_free(inode);
1641         }
1642
1643         md_null_inode(sbi->ll_md_exp, ll_inode2fid(inode));
1644
1645         LASSERT(!lli->lli_open_fd_write_count);
1646         LASSERT(!lli->lli_open_fd_read_count);
1647         LASSERT(!lli->lli_open_fd_exec_count);
1648
1649         if (lli->lli_mds_write_och)
1650                 ll_md_real_close(inode, FMODE_WRITE);
1651         if (lli->lli_mds_exec_och)
1652                 ll_md_real_close(inode, FMODE_EXEC);
1653         if (lli->lli_mds_read_och)
1654                 ll_md_real_close(inode, FMODE_READ);
1655
1656         if (S_ISLNK(inode->i_mode) && lli->lli_symlink_name) {
1657                 OBD_FREE(lli->lli_symlink_name,
1658                          strlen(lli->lli_symlink_name) + 1);
1659                 lli->lli_symlink_name = NULL;
1660         }
1661
1662         ll_xattr_cache_destroy(inode);
1663
1664 #ifdef CONFIG_LUSTRE_FS_POSIX_ACL
1665         forget_all_cached_acls(inode);
1666         if (lli->lli_posix_acl) {
1667                 posix_acl_release(lli->lli_posix_acl);
1668                 lli->lli_posix_acl = NULL;
1669         }
1670 #endif
1671         lli->lli_inode_magic = LLI_INODE_DEAD;
1672
1673         if (S_ISDIR(inode->i_mode))
1674                 ll_dir_clear_lsm_md(inode);
1675         else if (S_ISREG(inode->i_mode) && !is_bad_inode(inode))
1676                 LASSERT(list_empty(&lli->lli_agl_list));
1677
1678         /*
1679          * XXX This has to be done before lsm is freed below, because
1680          * cl_object still uses inode lsm.
1681          */
1682         cl_inode_fini(inode);
1683
1684         llcrypt_put_encryption_info(inode);
1685
1686         EXIT;
1687 }
1688
1689 static int ll_md_setattr(struct dentry *dentry, struct md_op_data *op_data)
1690 {
1691         struct lustre_md md;
1692         struct inode *inode = dentry->d_inode;
1693         struct ll_sb_info *sbi = ll_i2sbi(inode);
1694         struct ptlrpc_request *request = NULL;
1695         int rc, ia_valid;
1696         ENTRY;
1697
1698         op_data = ll_prep_md_op_data(op_data, inode, NULL, NULL, 0, 0,
1699                                      LUSTRE_OPC_ANY, NULL);
1700         if (IS_ERR(op_data))
1701                 RETURN(PTR_ERR(op_data));
1702
1703         rc = md_setattr(sbi->ll_md_exp, op_data, NULL, 0, &request);
1704         if (rc) {
1705                 ptlrpc_req_finished(request);
1706                 if (rc == -ENOENT) {
1707                         clear_nlink(inode);
1708                         /* Unlinked special device node? Or just a race?
1709                          * Pretend we done everything. */
1710                         if (!S_ISREG(inode->i_mode) &&
1711                             !S_ISDIR(inode->i_mode)) {
1712                                 ia_valid = op_data->op_attr.ia_valid;
1713                                 op_data->op_attr.ia_valid &= ~TIMES_SET_FLAGS;
1714                                 rc = simple_setattr(dentry, &op_data->op_attr);
1715                                 op_data->op_attr.ia_valid = ia_valid;
1716                         }
1717                 } else if (rc != -EPERM && rc != -EACCES && rc != -ETXTBSY) {
1718                         CERROR("md_setattr fails: rc = %d\n", rc);
1719                 }
1720                 RETURN(rc);
1721         }
1722
1723         rc = md_get_lustre_md(sbi->ll_md_exp, request, sbi->ll_dt_exp,
1724                               sbi->ll_md_exp, &md);
1725         if (rc) {
1726                 ptlrpc_req_finished(request);
1727                 RETURN(rc);
1728         }
1729
1730         ia_valid = op_data->op_attr.ia_valid;
1731         /* inode size will be in ll_setattr_ost, can't do it now since dirty
1732          * cache is not cleared yet. */
1733         op_data->op_attr.ia_valid &= ~(TIMES_SET_FLAGS | ATTR_SIZE);
1734         if (S_ISREG(inode->i_mode))
1735                 inode_lock(inode);
1736         rc = simple_setattr(dentry, &op_data->op_attr);
1737         if (S_ISREG(inode->i_mode))
1738                 inode_unlock(inode);
1739         op_data->op_attr.ia_valid = ia_valid;
1740
1741         rc = ll_update_inode(inode, &md);
1742         ptlrpc_req_finished(request);
1743
1744         RETURN(rc);
1745 }
1746
1747 /**
1748  * Zero portion of page that is part of @inode.
1749  * This implies, if necessary:
1750  * - taking cl_lock on range corresponding to concerned page
1751  * - grabbing vm page
1752  * - associating cl_page
1753  * - proceeding to clio read
1754  * - zeroing range in page
1755  * - proceeding to cl_page flush
1756  * - releasing cl_lock
1757  *
1758  * \param[in] inode     inode
1759  * \param[in] index     page index
1760  * \param[in] offset    offset in page to start zero from
1761  * \param[in] len       len to zero
1762  *
1763  * \retval 0            on success
1764  * \retval negative     errno on failure
1765  */
1766 int ll_io_zero_page(struct inode *inode, pgoff_t index, pgoff_t offset,
1767                     unsigned len)
1768 {
1769         struct ll_inode_info *lli = ll_i2info(inode);
1770         struct cl_object *clob = lli->lli_clob;
1771         __u16 refcheck;
1772         struct lu_env *env = NULL;
1773         struct cl_io *io = NULL;
1774         struct cl_page *clpage = NULL;
1775         struct page *vmpage = NULL;
1776         unsigned from = index << PAGE_SHIFT;
1777         struct cl_lock *lock = NULL;
1778         struct cl_lock_descr *descr = NULL;
1779         struct cl_2queue *queue = NULL;
1780         struct cl_sync_io *anchor = NULL;
1781         bool holdinglock = false;
1782         bool lockedbymyself = true;
1783         int rc;
1784
1785         ENTRY;
1786
1787         env = cl_env_get(&refcheck);
1788         if (IS_ERR(env))
1789                 RETURN(PTR_ERR(env));
1790
1791         io = vvp_env_thread_io(env);
1792         io->ci_obj = clob;
1793         rc = cl_io_rw_init(env, io, CIT_WRITE, from, PAGE_SIZE);
1794         if (rc)
1795                 GOTO(putenv, rc);
1796
1797         lock = vvp_env_lock(env);
1798         descr = &lock->cll_descr;
1799         descr->cld_obj   = io->ci_obj;
1800         descr->cld_start = cl_index(io->ci_obj, from);
1801         descr->cld_end   = cl_index(io->ci_obj, from + PAGE_SIZE - 1);
1802         descr->cld_mode  = CLM_WRITE;
1803         descr->cld_enq_flags = CEF_MUST | CEF_NONBLOCK;
1804
1805         /* request lock for page */
1806         rc = cl_lock_request(env, io, lock);
1807         /* -ECANCELED indicates a matching lock with a different extent
1808          * was already present, and -EEXIST indicates a matching lock
1809          * on exactly the same extent was already present.
1810          * In both cases it means we are covered.
1811          */
1812         if (rc == -ECANCELED || rc == -EEXIST)
1813                 rc = 0;
1814         else if (rc < 0)
1815                 GOTO(iofini, rc);
1816         else
1817                 holdinglock = true;
1818
1819         /* grab page */
1820         vmpage = grab_cache_page_nowait(inode->i_mapping, index);
1821         if (vmpage == NULL)
1822                 GOTO(rellock, rc = -EOPNOTSUPP);
1823
1824         if (!PageDirty(vmpage)) {
1825                 /* associate cl_page */
1826                 clpage = cl_page_find(env, clob, vmpage->index,
1827                                       vmpage, CPT_CACHEABLE);
1828                 if (IS_ERR(clpage))
1829                         GOTO(pagefini, rc = PTR_ERR(clpage));
1830
1831                 cl_page_assume(env, io, clpage);
1832         }
1833
1834         if (!PageUptodate(vmpage) && !PageDirty(vmpage) &&
1835             !PageWriteback(vmpage)) {
1836                 /* read page */
1837                 /* set PagePrivate2 to detect special case of empty page
1838                  * in osc_brw_fini_request()
1839                  */
1840                 SetPagePrivate2(vmpage);
1841                 rc = ll_io_read_page(env, io, clpage, NULL);
1842                 if (!PagePrivate2(vmpage))
1843                         /* PagePrivate2 was cleared in osc_brw_fini_request()
1844                          * meaning we read an empty page. In this case, in order
1845                          * to avoid allocating unnecessary block in truncated
1846                          * file, we must not zero and write as below. Subsequent
1847                          * server-side truncate will handle things correctly.
1848                          */
1849                         GOTO(clpfini, rc = 0);
1850                 ClearPagePrivate2(vmpage);
1851                 if (rc)
1852                         GOTO(clpfini, rc);
1853                 lockedbymyself = trylock_page(vmpage);
1854                 cl_page_assume(env, io, clpage);
1855         }
1856
1857         /* zero range in page */
1858         zero_user(vmpage, offset, len);
1859
1860         if (holdinglock && clpage) {
1861                 /* explicitly write newly modified page */
1862                 queue = &io->ci_queue;
1863                 cl_2queue_init(queue);
1864                 anchor = &vvp_env_info(env)->vti_anchor;
1865                 cl_sync_io_init(anchor, 1);
1866                 clpage->cp_sync_io = anchor;
1867                 cl_2queue_add(queue, clpage);
1868                 rc = cl_io_submit_rw(env, io, CRT_WRITE, queue);
1869                 if (rc)
1870                         GOTO(queuefini1, rc);
1871                 rc = cl_sync_io_wait(env, anchor, 0);
1872                 if (rc)
1873                         GOTO(queuefini2, rc);
1874                 cl_page_assume(env, io, clpage);
1875
1876 queuefini2:
1877                 cl_2queue_discard(env, io, queue);
1878 queuefini1:
1879                 cl_2queue_disown(env, io, queue);
1880                 cl_2queue_fini(env, queue);
1881         }
1882
1883 clpfini:
1884         if (clpage)
1885                 cl_page_put(env, clpage);
1886 pagefini:
1887         if (lockedbymyself) {
1888                 unlock_page(vmpage);
1889                 put_page(vmpage);
1890         }
1891 rellock:
1892         if (holdinglock)
1893                 cl_lock_release(env, lock);
1894 iofini:
1895         cl_io_fini(env, io);
1896 putenv:
1897         if (env)
1898                 cl_env_put(env, &refcheck);
1899
1900         RETURN(rc);
1901 }
1902
1903 /* If this inode has objects allocated to it (lsm != NULL), then the OST
1904  * object(s) determine the file size and mtime.  Otherwise, the MDS will
1905  * keep these values until such a time that objects are allocated for it.
1906  * We do the MDS operations first, as it is checking permissions for us.
1907  * We don't to the MDS RPC if there is nothing that we want to store there,
1908  * otherwise there is no harm in updating mtime/atime on the MDS if we are
1909  * going to do an RPC anyways.
1910  *
1911  * If we are doing a truncate, we will send the mtime and ctime updates
1912  * to the OST with the punch RPC, otherwise we do an explicit setattr RPC.
1913  * I don't believe it is possible to get e.g. ATTR_MTIME_SET and ATTR_SIZE
1914  * at the same time.
1915  *
1916  * In case of HSMimport, we only set attr on MDS.
1917  */
1918 int ll_setattr_raw(struct dentry *dentry, struct iattr *attr,
1919                    enum op_xvalid xvalid, bool hsm_import)
1920 {
1921         struct inode *inode = dentry->d_inode;
1922         struct ll_inode_info *lli = ll_i2info(inode);
1923         struct md_op_data *op_data = NULL;
1924         ktime_t kstart = ktime_get();
1925         int rc = 0;
1926
1927         ENTRY;
1928
1929         CDEBUG(D_VFSTRACE, "%s: setattr inode "DFID"(%p) from %llu to %llu, "
1930                "valid %x, hsm_import %d\n",
1931                ll_i2sbi(inode)->ll_fsname, PFID(&lli->lli_fid),
1932                inode, i_size_read(inode), attr->ia_size, attr->ia_valid,
1933                hsm_import);
1934
1935         if (attr->ia_valid & ATTR_SIZE) {
1936                 /* Check new size against VFS/VM file size limit and rlimit */
1937                 rc = inode_newsize_ok(inode, attr->ia_size);
1938                 if (rc)
1939                         RETURN(rc);
1940
1941                 /* The maximum Lustre file size is variable, based on the
1942                  * OST maximum object size and number of stripes.  This
1943                  * needs another check in addition to the VFS check above. */
1944                 if (attr->ia_size > ll_file_maxbytes(inode)) {
1945                         CDEBUG(D_INODE,"file "DFID" too large %llu > %llu\n",
1946                                PFID(&lli->lli_fid), attr->ia_size,
1947                                ll_file_maxbytes(inode));
1948                         RETURN(-EFBIG);
1949                 }
1950
1951                 attr->ia_valid |= ATTR_MTIME | ATTR_CTIME;
1952         }
1953
1954         /* POSIX: check before ATTR_*TIME_SET set (from inode_change_ok) */
1955         if (attr->ia_valid & TIMES_SET_FLAGS) {
1956                 if ((!uid_eq(current_fsuid(), inode->i_uid)) &&
1957                     !cfs_capable(CFS_CAP_FOWNER))
1958                         RETURN(-EPERM);
1959         }
1960
1961         /* We mark all of the fields "set" so MDS/OST does not re-set them */
1962         if (!(xvalid & OP_XVALID_CTIME_SET) &&
1963              (attr->ia_valid & ATTR_CTIME)) {
1964                 attr->ia_ctime = current_time(inode);
1965                 xvalid |= OP_XVALID_CTIME_SET;
1966         }
1967         if (!(attr->ia_valid & ATTR_ATIME_SET) &&
1968             (attr->ia_valid & ATTR_ATIME)) {
1969                 attr->ia_atime = current_time(inode);
1970                 attr->ia_valid |= ATTR_ATIME_SET;
1971         }
1972         if (!(attr->ia_valid & ATTR_MTIME_SET) &&
1973             (attr->ia_valid & ATTR_MTIME)) {
1974                 attr->ia_mtime = current_time(inode);
1975                 attr->ia_valid |= ATTR_MTIME_SET;
1976         }
1977
1978         if (attr->ia_valid & (ATTR_MTIME | ATTR_CTIME))
1979                 CDEBUG(D_INODE, "setting mtime %lld, ctime %lld, now = %lld\n",
1980                        (s64)attr->ia_mtime.tv_sec, (s64)attr->ia_ctime.tv_sec,
1981                        ktime_get_real_seconds());
1982
1983         if (S_ISREG(inode->i_mode))
1984                 inode_unlock(inode);
1985
1986         /* We always do an MDS RPC, even if we're only changing the size;
1987          * only the MDS knows whether truncate() should fail with -ETXTBUSY */
1988
1989         OBD_ALLOC_PTR(op_data);
1990         if (op_data == NULL)
1991                 GOTO(out, rc = -ENOMEM);
1992
1993         if (!hsm_import && attr->ia_valid & ATTR_SIZE) {
1994                 /* If we are changing file size, file content is
1995                  * modified, flag it.
1996                  */
1997                 xvalid |= OP_XVALID_OWNEROVERRIDE;
1998                 op_data->op_bias |= MDS_DATA_MODIFIED;
1999                 ll_file_clear_flag(lli, LLIF_DATA_MODIFIED);
2000         }
2001
2002         if (attr->ia_valid & ATTR_FILE) {
2003                 struct ll_file_data *fd = attr->ia_file->private_data;
2004
2005                 if (fd->fd_lease_och)
2006                         op_data->op_bias |= MDS_TRUNC_KEEP_LEASE;
2007         }
2008
2009         op_data->op_attr = *attr;
2010         op_data->op_xvalid = xvalid;
2011
2012         rc = ll_md_setattr(dentry, op_data);
2013         if (rc)
2014                 GOTO(out, rc);
2015
2016         if (!S_ISREG(inode->i_mode) || hsm_import)
2017                 GOTO(out, rc = 0);
2018
2019         if (attr->ia_valid & (ATTR_SIZE | ATTR_ATIME | ATTR_ATIME_SET |
2020                               ATTR_MTIME | ATTR_MTIME_SET | ATTR_CTIME) ||
2021             xvalid & OP_XVALID_CTIME_SET) {
2022                 bool cached = false;
2023
2024                 rc = pcc_inode_setattr(inode, attr, &cached);
2025                 if (cached) {
2026                         if (rc) {
2027                                 CERROR("%s: PCC inode "DFID" setattr failed: "
2028                                        "rc = %d\n",
2029                                        ll_i2sbi(inode)->ll_fsname,
2030                                        PFID(&lli->lli_fid), rc);
2031                                 GOTO(out, rc);
2032                         }
2033                 } else {
2034                         unsigned int flags = 0;
2035
2036                         /* For truncate and utimes sending attributes to OSTs,
2037                          * setting mtime/atime to the past will be performed
2038                          * under PW [0:EOF] extent lock (new_size:EOF for
2039                          * truncate). It may seem excessive to send mtime/atime
2040                          * updates to OSTs when not setting times to past, but
2041                          * it is necessary due to possible time
2042                          * de-synchronization between MDT inode and OST objects
2043                          */
2044                         if (S_ISREG(inode->i_mode) && IS_ENCRYPTED(inode) &&
2045                             attr->ia_valid & ATTR_SIZE) {
2046                                 xvalid |= OP_XVALID_FLAGS;
2047                                 flags = LUSTRE_ENCRYPT_FL;
2048                                 if (attr->ia_size & ~PAGE_MASK) {
2049                                         pgoff_t offset =
2050                                                 attr->ia_size & (PAGE_SIZE - 1);
2051
2052                                         rc = ll_io_zero_page(inode,
2053                                                     attr->ia_size >> PAGE_SHIFT,
2054                                                     offset, PAGE_SIZE - offset);
2055                                         if (rc)
2056                                                 GOTO(out, rc);
2057                                 }
2058                         }
2059                         rc = cl_setattr_ost(lli->lli_clob, attr, xvalid, flags);
2060                 }
2061         }
2062
2063         /* If the file was restored, it needs to set dirty flag.
2064          *
2065          * We've already sent MDS_DATA_MODIFIED flag in
2066          * ll_md_setattr() for truncate. However, the MDT refuses to
2067          * set the HS_DIRTY flag on released files, so we have to set
2068          * it again if the file has been restored. Please check how
2069          * LLIF_DATA_MODIFIED is set in vvp_io_setattr_fini().
2070          *
2071          * Please notice that if the file is not released, the previous
2072          * MDS_DATA_MODIFIED has taken effect and usually
2073          * LLIF_DATA_MODIFIED is not set(see vvp_io_setattr_fini()).
2074          * This way we can save an RPC for common open + trunc
2075          * operation. */
2076         if (ll_file_test_and_clear_flag(lli, LLIF_DATA_MODIFIED)) {
2077                 struct hsm_state_set hss = {
2078                         .hss_valid = HSS_SETMASK,
2079                         .hss_setmask = HS_DIRTY,
2080                 };
2081                 int rc2;
2082
2083                 rc2 = ll_hsm_state_set(inode, &hss);
2084                 /* truncate and write can happen at the same time, so that
2085                  * the file can be set modified even though the file is not
2086                  * restored from released state, and ll_hsm_state_set() is
2087                  * not applicable for the file, and rc2 < 0 is normal in this
2088                  * case. */
2089                 if (rc2 < 0)
2090                         CDEBUG(D_INFO, DFID "HSM set dirty failed: rc2 = %d\n",
2091                                PFID(ll_inode2fid(inode)), rc2);
2092         }
2093
2094         EXIT;
2095 out:
2096         if (op_data != NULL)
2097                 ll_finish_md_op_data(op_data);
2098
2099         if (S_ISREG(inode->i_mode)) {
2100                 inode_lock(inode);
2101                 if ((attr->ia_valid & ATTR_SIZE) && !hsm_import)
2102                         inode_dio_wait(inode);
2103                 /* Once we've got the i_mutex, it's safe to set the S_NOSEC
2104                  * flag.  ll_update_inode (called from ll_md_setattr), clears
2105                  * inode flags, so there is a gap where S_NOSEC is not set.
2106                  * This can cause a writer to take the i_mutex unnecessarily,
2107                  * but this is safe to do and should be rare. */
2108                 inode_has_no_xattr(inode);
2109         }
2110
2111         if (!rc)
2112                 ll_stats_ops_tally(ll_i2sbi(inode), attr->ia_valid & ATTR_SIZE ?
2113                                         LPROC_LL_TRUNC : LPROC_LL_SETATTR,
2114                                    ktime_us_delta(ktime_get(), kstart));
2115
2116         return rc;
2117 }
2118
2119 int ll_setattr(struct dentry *de, struct iattr *attr)
2120 {
2121         int mode = de->d_inode->i_mode;
2122         enum op_xvalid xvalid = 0;
2123         int rc;
2124
2125         rc = llcrypt_prepare_setattr(de, attr);
2126         if (rc)
2127                 return rc;
2128
2129         if ((attr->ia_valid & (ATTR_CTIME|ATTR_SIZE|ATTR_MODE)) ==
2130                               (ATTR_CTIME|ATTR_SIZE|ATTR_MODE))
2131                 xvalid |= OP_XVALID_OWNEROVERRIDE;
2132
2133         if (((attr->ia_valid & (ATTR_MODE|ATTR_FORCE|ATTR_SIZE)) ==
2134                                (ATTR_SIZE|ATTR_MODE)) &&
2135             (((mode & S_ISUID) && !(attr->ia_mode & S_ISUID)) ||
2136              (((mode & (S_ISGID|S_IXGRP)) == (S_ISGID|S_IXGRP)) &&
2137               !(attr->ia_mode & S_ISGID))))
2138                 attr->ia_valid |= ATTR_FORCE;
2139
2140         if ((attr->ia_valid & ATTR_MODE) &&
2141             (mode & S_ISUID) &&
2142             !(attr->ia_mode & S_ISUID) &&
2143             !(attr->ia_valid & ATTR_KILL_SUID))
2144                 attr->ia_valid |= ATTR_KILL_SUID;
2145
2146         if ((attr->ia_valid & ATTR_MODE) &&
2147             ((mode & (S_ISGID|S_IXGRP)) == (S_ISGID|S_IXGRP)) &&
2148             !(attr->ia_mode & S_ISGID) &&
2149             !(attr->ia_valid & ATTR_KILL_SGID))
2150                 attr->ia_valid |= ATTR_KILL_SGID;
2151
2152         return ll_setattr_raw(de, attr, xvalid, false);
2153 }
2154
2155 int ll_statfs_internal(struct ll_sb_info *sbi, struct obd_statfs *osfs,
2156                        u32 flags)
2157 {
2158         struct obd_statfs obd_osfs = { 0 };
2159         time64_t max_age;
2160         int rc;
2161
2162         ENTRY;
2163         max_age = ktime_get_seconds() - sbi->ll_statfs_max_age;
2164
2165         if (sbi->ll_flags & LL_SBI_LAZYSTATFS)
2166                 flags |= OBD_STATFS_NODELAY;
2167
2168         rc = obd_statfs(NULL, sbi->ll_md_exp, osfs, max_age, flags);
2169         if (rc)
2170                 RETURN(rc);
2171
2172         osfs->os_type = LL_SUPER_MAGIC;
2173
2174         CDEBUG(D_SUPER, "MDC blocks %llu/%llu objects %llu/%llu\n",
2175               osfs->os_bavail, osfs->os_blocks, osfs->os_ffree, osfs->os_files);
2176
2177         if (osfs->os_state & OS_STATFS_SUM)
2178                 GOTO(out, rc);
2179
2180         rc = obd_statfs(NULL, sbi->ll_dt_exp, &obd_osfs, max_age, flags);
2181         if (rc) /* Possibly a filesystem with no OSTs.  Report MDT totals. */
2182                 GOTO(out, rc = 0);
2183
2184         CDEBUG(D_SUPER, "OSC blocks %llu/%llu objects %llu/%llu\n",
2185                obd_osfs.os_bavail, obd_osfs.os_blocks, obd_osfs.os_ffree,
2186                obd_osfs.os_files);
2187
2188         osfs->os_bsize = obd_osfs.os_bsize;
2189         osfs->os_blocks = obd_osfs.os_blocks;
2190         osfs->os_bfree = obd_osfs.os_bfree;
2191         osfs->os_bavail = obd_osfs.os_bavail;
2192
2193         /* If we have _some_ OSTs, but don't have as many free objects on the
2194          * OSTs as inodes on the MDTs, reduce the reported number of inodes
2195          * to compensate, so that the "inodes in use" number is correct.
2196          * This should be kept in sync with lod_statfs() behaviour.
2197          */
2198         if (obd_osfs.os_files && obd_osfs.os_ffree < osfs->os_ffree) {
2199                 osfs->os_files = (osfs->os_files - osfs->os_ffree) +
2200                                  obd_osfs.os_ffree;
2201                 osfs->os_ffree = obd_osfs.os_ffree;
2202         }
2203
2204 out:
2205         RETURN(rc);
2206 }
2207
2208 int ll_statfs(struct dentry *de, struct kstatfs *sfs)
2209 {
2210         struct super_block *sb = de->d_sb;
2211         struct obd_statfs osfs;
2212         __u64 fsid = huge_encode_dev(sb->s_dev);
2213         ktime_t kstart = ktime_get();
2214         int rc;
2215
2216         CDEBUG(D_VFSTRACE, "VFS Op:sb=%s (%p)\n", sb->s_id, sb);
2217
2218         /* Some amount of caching on the client is allowed */
2219         rc = ll_statfs_internal(ll_s2sbi(sb), &osfs, OBD_STATFS_SUM);
2220         if (rc)
2221                 return rc;
2222
2223         statfs_unpack(sfs, &osfs);
2224
2225         /* We need to downshift for all 32-bit kernels, because we can't
2226          * tell if the kernel is being called via sys_statfs64() or not.
2227          * Stop before overflowing f_bsize - in which case it is better
2228          * to just risk EOVERFLOW if caller is using old sys_statfs(). */
2229         if (sizeof(long) < 8) {
2230                 while (osfs.os_blocks > ~0UL && sfs->f_bsize < 0x40000000) {
2231                         sfs->f_bsize <<= 1;
2232
2233                         osfs.os_blocks >>= 1;
2234                         osfs.os_bfree >>= 1;
2235                         osfs.os_bavail >>= 1;
2236                 }
2237         }
2238
2239         sfs->f_blocks = osfs.os_blocks;
2240         sfs->f_bfree = osfs.os_bfree;
2241         sfs->f_bavail = osfs.os_bavail;
2242         sfs->f_fsid.val[0] = (__u32)fsid;
2243         sfs->f_fsid.val[1] = (__u32)(fsid >> 32);
2244
2245         ll_stats_ops_tally(ll_s2sbi(sb), LPROC_LL_STATFS,
2246                            ktime_us_delta(ktime_get(), kstart));
2247
2248         return 0;
2249 }
2250
2251 void ll_inode_size_lock(struct inode *inode)
2252 {
2253         struct ll_inode_info *lli;
2254
2255         LASSERT(!S_ISDIR(inode->i_mode));
2256
2257         lli = ll_i2info(inode);
2258         mutex_lock(&lli->lli_size_mutex);
2259 }
2260
2261 void ll_inode_size_unlock(struct inode *inode)
2262 {
2263         struct ll_inode_info *lli;
2264
2265         lli = ll_i2info(inode);
2266         mutex_unlock(&lli->lli_size_mutex);
2267 }
2268
2269 void ll_update_inode_flags(struct inode *inode, int ext_flags)
2270 {
2271         /* do not clear encryption flag */
2272         ext_flags |= ll_inode_to_ext_flags(inode->i_flags) & LUSTRE_ENCRYPT_FL;
2273         inode->i_flags = ll_ext_to_inode_flags(ext_flags);
2274         if (ext_flags & LUSTRE_PROJINHERIT_FL)
2275                 ll_file_set_flag(ll_i2info(inode), LLIF_PROJECT_INHERIT);
2276         else
2277                 ll_file_clear_flag(ll_i2info(inode), LLIF_PROJECT_INHERIT);
2278 }
2279
2280 int ll_update_inode(struct inode *inode, struct lustre_md *md)
2281 {
2282         struct ll_inode_info *lli = ll_i2info(inode);
2283         struct mdt_body *body = md->body;
2284         struct ll_sb_info *sbi = ll_i2sbi(inode);
2285         int rc = 0;
2286
2287         if (body->mbo_valid & OBD_MD_FLEASIZE) {
2288                 rc = cl_file_inode_init(inode, md);
2289                 if (rc)
2290                         return rc;
2291         }
2292
2293         if (S_ISDIR(inode->i_mode)) {
2294                 rc = ll_update_lsm_md(inode, md);
2295                 if (rc != 0)
2296                         return rc;
2297         }
2298
2299 #ifdef CONFIG_LUSTRE_FS_POSIX_ACL
2300         if (body->mbo_valid & OBD_MD_FLACL) {
2301                 spin_lock(&lli->lli_lock);
2302                 if (lli->lli_posix_acl)
2303                         posix_acl_release(lli->lli_posix_acl);
2304                 lli->lli_posix_acl = md->posix_acl;
2305                 spin_unlock(&lli->lli_lock);
2306         }
2307 #endif
2308         inode->i_ino = cl_fid_build_ino(&body->mbo_fid1,
2309                                         sbi->ll_flags & LL_SBI_32BIT_API);
2310         inode->i_generation = cl_fid_build_gen(&body->mbo_fid1);
2311
2312         if (body->mbo_valid & OBD_MD_FLATIME) {
2313                 if (body->mbo_atime > inode->i_atime.tv_sec)
2314                         inode->i_atime.tv_sec = body->mbo_atime;
2315                 lli->lli_atime = body->mbo_atime;
2316         }
2317
2318         if (body->mbo_valid & OBD_MD_FLMTIME) {
2319                 if (body->mbo_mtime > inode->i_mtime.tv_sec) {
2320                         CDEBUG(D_INODE,
2321                                "setting ino %lu mtime from %lld to %llu\n",
2322                                inode->i_ino, (s64)inode->i_mtime.tv_sec,
2323                                body->mbo_mtime);
2324                         inode->i_mtime.tv_sec = body->mbo_mtime;
2325                 }
2326                 lli->lli_mtime = body->mbo_mtime;
2327         }
2328
2329         if (body->mbo_valid & OBD_MD_FLCTIME) {
2330                 if (body->mbo_ctime > inode->i_ctime.tv_sec)
2331                         inode->i_ctime.tv_sec = body->mbo_ctime;
2332                 lli->lli_ctime = body->mbo_ctime;
2333         }
2334
2335         if (body->mbo_valid & OBD_MD_FLBTIME)
2336                 lli->lli_btime = body->mbo_btime;
2337
2338         /* Clear i_flags to remove S_NOSEC before permissions are updated */
2339         if (body->mbo_valid & OBD_MD_FLFLAGS)
2340                 ll_update_inode_flags(inode, body->mbo_flags);
2341         if (body->mbo_valid & OBD_MD_FLMODE)
2342                 inode->i_mode = (inode->i_mode & S_IFMT) |
2343                                 (body->mbo_mode & ~S_IFMT);
2344
2345         if (body->mbo_valid & OBD_MD_FLTYPE)
2346                 inode->i_mode = (inode->i_mode & ~S_IFMT) |
2347                                 (body->mbo_mode & S_IFMT);
2348
2349         LASSERT(inode->i_mode != 0);
2350         if (body->mbo_valid & OBD_MD_FLUID)
2351                 inode->i_uid = make_kuid(&init_user_ns, body->mbo_uid);
2352         if (body->mbo_valid & OBD_MD_FLGID)
2353                 inode->i_gid = make_kgid(&init_user_ns, body->mbo_gid);
2354         if (body->mbo_valid & OBD_MD_FLPROJID)
2355                 lli->lli_projid = body->mbo_projid;
2356         if (body->mbo_valid & OBD_MD_FLNLINK)
2357                 set_nlink(inode, body->mbo_nlink);
2358         if (body->mbo_valid & OBD_MD_FLRDEV)
2359                 inode->i_rdev = old_decode_dev(body->mbo_rdev);
2360
2361         if (body->mbo_valid & OBD_MD_FLID) {
2362                 /* FID shouldn't be changed! */
2363                 if (fid_is_sane(&lli->lli_fid)) {
2364                         LASSERTF(lu_fid_eq(&lli->lli_fid, &body->mbo_fid1),
2365                                  "Trying to change FID "DFID
2366                                  " to the "DFID", inode "DFID"(%p)\n",
2367                                  PFID(&lli->lli_fid), PFID(&body->mbo_fid1),
2368                                  PFID(ll_inode2fid(inode)), inode);
2369                 } else {
2370                         lli->lli_fid = body->mbo_fid1;
2371                 }
2372         }
2373
2374         LASSERT(fid_seq(&lli->lli_fid) != 0);
2375
2376         lli->lli_attr_valid = body->mbo_valid;
2377         if (body->mbo_valid & OBD_MD_FLSIZE) {
2378                 i_size_write(inode, body->mbo_size);
2379
2380                 CDEBUG(D_VFSTRACE, "inode="DFID", updating i_size %llu\n",
2381                        PFID(ll_inode2fid(inode)),
2382                        (unsigned long long)body->mbo_size);
2383
2384                 if (body->mbo_valid & OBD_MD_FLBLOCKS)
2385                         inode->i_blocks = body->mbo_blocks;
2386         } else {
2387                 if (body->mbo_valid & OBD_MD_FLLAZYSIZE)
2388                         lli->lli_lazysize = body->mbo_size;
2389                 if (body->mbo_valid & OBD_MD_FLLAZYBLOCKS)
2390                         lli->lli_lazyblocks = body->mbo_blocks;
2391         }
2392
2393         if (body->mbo_valid & OBD_MD_TSTATE) {
2394                 /* Set LLIF_FILE_RESTORING if restore ongoing and
2395                  * clear it when done to ensure to start again
2396                  * glimpsing updated attrs
2397                  */
2398                 if (body->mbo_t_state & MS_RESTORE)
2399                         ll_file_set_flag(lli, LLIF_FILE_RESTORING);
2400                 else
2401                         ll_file_clear_flag(lli, LLIF_FILE_RESTORING);
2402         }
2403
2404         return 0;
2405 }
2406
2407 int ll_read_inode2(struct inode *inode, void *opaque)
2408 {
2409         struct lustre_md *md = opaque;
2410         struct ll_inode_info *lli = ll_i2info(inode);
2411         int     rc;
2412         ENTRY;
2413
2414         CDEBUG(D_VFSTRACE, "VFS Op:inode="DFID"(%p)\n",
2415                PFID(&lli->lli_fid), inode);
2416
2417         /* Core attributes from the MDS first.  This is a new inode, and
2418          * the VFS doesn't zero times in the core inode so we have to do
2419          * it ourselves.  They will be overwritten by either MDS or OST
2420          * attributes - we just need to make sure they aren't newer.
2421          */
2422         inode->i_mtime.tv_sec = 0;
2423         inode->i_atime.tv_sec = 0;
2424         inode->i_ctime.tv_sec = 0;
2425         inode->i_rdev = 0;
2426         rc = ll_update_inode(inode, md);
2427         if (rc != 0)
2428                 RETURN(rc);
2429
2430         /* OIDEBUG(inode); */
2431
2432 #ifdef HAVE_BACKING_DEV_INFO
2433         /* initializing backing dev info. */
2434         inode->i_mapping->backing_dev_info = &s2lsi(inode->i_sb)->lsi_bdi;
2435 #endif
2436         if (S_ISREG(inode->i_mode)) {
2437                 struct ll_sb_info *sbi = ll_i2sbi(inode);
2438                 inode->i_op = &ll_file_inode_operations;
2439                 inode->i_fop = sbi->ll_fop;
2440                 inode->i_mapping->a_ops = (struct address_space_operations *)&ll_aops;
2441                 EXIT;
2442         } else if (S_ISDIR(inode->i_mode)) {
2443                 inode->i_op = &ll_dir_inode_operations;
2444                 inode->i_fop = &ll_dir_operations;
2445                 EXIT;
2446         } else if (S_ISLNK(inode->i_mode)) {
2447                 inode->i_op = &ll_fast_symlink_inode_operations;
2448                 EXIT;
2449         } else {
2450                 inode->i_op = &ll_special_inode_operations;
2451
2452                 init_special_inode(inode, inode->i_mode,
2453                                    inode->i_rdev);
2454
2455                 EXIT;
2456         }
2457
2458         return 0;
2459 }
2460
2461 void ll_delete_inode(struct inode *inode)
2462 {
2463         struct ll_inode_info *lli = ll_i2info(inode);
2464         struct address_space *mapping = &inode->i_data;
2465         unsigned long nrpages;
2466         unsigned long flags;
2467
2468         ENTRY;
2469
2470         if (S_ISREG(inode->i_mode) && lli->lli_clob != NULL) {
2471                 /* It is last chance to write out dirty pages,
2472                  * otherwise we may lose data while umount.
2473                  *
2474                  * If i_nlink is 0 then just discard data. This is safe because
2475                  * local inode gets i_nlink 0 from server only for the last
2476                  * unlink, so that file is not opened somewhere else
2477                  */
2478                 cl_sync_file_range(inode, 0, OBD_OBJECT_EOF, inode->i_nlink ?
2479                                    CL_FSYNC_LOCAL : CL_FSYNC_DISCARD, 1);
2480         }
2481         truncate_inode_pages_final(mapping);
2482
2483         /* Workaround for LU-118: Note nrpages may not be totally updated when
2484          * truncate_inode_pages() returns, as there can be a page in the process
2485          * of deletion (inside __delete_from_page_cache()) in the specified
2486          * range. Thus mapping->nrpages can be non-zero when this function
2487          * returns even after truncation of the whole mapping.  Only do this if
2488          * npages isn't already zero.
2489          */
2490         nrpages = mapping->nrpages;
2491         if (nrpages) {
2492                 ll_xa_lock_irqsave(&mapping->i_pages, flags);
2493                 nrpages = mapping->nrpages;
2494                 ll_xa_unlock_irqrestore(&mapping->i_pages, flags);
2495         } /* Workaround end */
2496
2497         LASSERTF(nrpages == 0, "%s: inode="DFID"(%p) nrpages=%lu, "
2498                  "see https://jira.whamcloud.com/browse/LU-118\n",
2499                  ll_i2sbi(inode)->ll_fsname,
2500                  PFID(ll_inode2fid(inode)), inode, nrpages);
2501
2502         ll_clear_inode(inode);
2503         clear_inode(inode);
2504
2505         EXIT;
2506 }
2507
2508 int ll_iocontrol(struct inode *inode, struct file *file,
2509                  unsigned int cmd, unsigned long arg)
2510 {
2511         struct ll_sb_info *sbi = ll_i2sbi(inode);
2512         struct ptlrpc_request *req = NULL;
2513         int rc, flags = 0;
2514         ENTRY;
2515
2516         switch (cmd) {
2517         case FS_IOC_GETFLAGS: {
2518                 struct mdt_body *body;
2519                 struct md_op_data *op_data;
2520
2521                 op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL,
2522                                              0, 0, LUSTRE_OPC_ANY,
2523                                              NULL);
2524                 if (IS_ERR(op_data))
2525                         RETURN(PTR_ERR(op_data));
2526
2527                 op_data->op_valid = OBD_MD_FLFLAGS;
2528                 rc = md_getattr(sbi->ll_md_exp, op_data, &req);
2529                 ll_finish_md_op_data(op_data);
2530                 if (rc) {
2531                         CERROR("%s: failure inode "DFID": rc = %d\n",
2532                                sbi->ll_md_exp->exp_obd->obd_name,
2533                                PFID(ll_inode2fid(inode)), rc);
2534                         RETURN(-abs(rc));
2535                 }
2536
2537                 body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
2538
2539                 flags = body->mbo_flags;
2540
2541                 ptlrpc_req_finished(req);
2542
2543                 RETURN(put_user(flags, (int __user *)arg));
2544         }
2545         case FS_IOC_SETFLAGS: {
2546                 struct iattr *attr;
2547                 struct md_op_data *op_data;
2548                 struct cl_object *obj;
2549                 struct fsxattr fa = { 0 };
2550
2551                 if (get_user(flags, (int __user *)arg))
2552                         RETURN(-EFAULT);
2553
2554                 fa.fsx_projid = ll_i2info(inode)->lli_projid;
2555                 if (flags & LUSTRE_PROJINHERIT_FL)
2556                         fa.fsx_xflags = FS_XFLAG_PROJINHERIT;
2557
2558                 rc = ll_ioctl_check_project(inode, &fa);
2559                 if (rc)
2560                         RETURN(rc);
2561
2562                 op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL, 0, 0,
2563                                              LUSTRE_OPC_ANY, NULL);
2564                 if (IS_ERR(op_data))
2565                         RETURN(PTR_ERR(op_data));
2566
2567                 op_data->op_attr_flags = flags;
2568                 op_data->op_xvalid |= OP_XVALID_FLAGS;
2569                 rc = md_setattr(sbi->ll_md_exp, op_data, NULL, 0, &req);
2570                 ll_finish_md_op_data(op_data);
2571                 ptlrpc_req_finished(req);
2572                 if (rc)
2573                         RETURN(rc);
2574
2575                 ll_update_inode_flags(inode, flags);
2576
2577                 obj = ll_i2info(inode)->lli_clob;
2578                 if (obj == NULL)
2579                         RETURN(0);
2580
2581                 OBD_ALLOC_PTR(attr);
2582                 if (attr == NULL)
2583                         RETURN(-ENOMEM);
2584
2585                 rc = cl_setattr_ost(obj, attr, OP_XVALID_FLAGS, flags);
2586
2587                 OBD_FREE_PTR(attr);
2588                 RETURN(rc);
2589         }
2590         default:
2591                 RETURN(-ENOSYS);
2592         }
2593
2594         RETURN(0);
2595 }
2596
2597 int ll_flush_ctx(struct inode *inode)
2598 {
2599         struct ll_sb_info  *sbi = ll_i2sbi(inode);
2600
2601         CDEBUG(D_SEC, "flush context for user %d\n",
2602                from_kuid(&init_user_ns, current_uid()));
2603
2604         obd_set_info_async(NULL, sbi->ll_md_exp,
2605                            sizeof(KEY_FLUSH_CTX), KEY_FLUSH_CTX,
2606                            0, NULL, NULL);
2607         obd_set_info_async(NULL, sbi->ll_dt_exp,
2608                            sizeof(KEY_FLUSH_CTX), KEY_FLUSH_CTX,
2609                            0, NULL, NULL);
2610         return 0;
2611 }
2612
2613 /* umount -f client means force down, don't save state */
2614 void ll_umount_begin(struct super_block *sb)
2615 {
2616         struct ll_sb_info *sbi = ll_s2sbi(sb);
2617         struct obd_device *obd;
2618         struct obd_ioctl_data *ioc_data;
2619         int cnt;
2620         ENTRY;
2621
2622         CDEBUG(D_VFSTRACE, "VFS Op: superblock %p count %d active %d\n", sb,
2623                sb->s_count, atomic_read(&sb->s_active));
2624
2625         obd = class_exp2obd(sbi->ll_md_exp);
2626         if (obd == NULL) {
2627                 CERROR("Invalid MDC connection handle %#llx\n",
2628                        sbi->ll_md_exp->exp_handle.h_cookie);
2629                 EXIT;
2630                 return;
2631         }
2632         obd->obd_force = 1;
2633
2634         obd = class_exp2obd(sbi->ll_dt_exp);
2635         if (obd == NULL) {
2636                 CERROR("Invalid LOV connection handle %#llx\n",
2637                        sbi->ll_dt_exp->exp_handle.h_cookie);
2638                 EXIT;
2639                 return;
2640         }
2641         obd->obd_force = 1;
2642
2643         OBD_ALLOC_PTR(ioc_data);
2644         if (ioc_data) {
2645                 obd_iocontrol(IOC_OSC_SET_ACTIVE, sbi->ll_md_exp,
2646                               sizeof *ioc_data, ioc_data, NULL);
2647
2648                 obd_iocontrol(IOC_OSC_SET_ACTIVE, sbi->ll_dt_exp,
2649                               sizeof *ioc_data, ioc_data, NULL);
2650
2651                 OBD_FREE_PTR(ioc_data);
2652         }
2653
2654         /* Really, we'd like to wait until there are no requests outstanding,
2655          * and then continue.  For now, we just periodically checking for vfs
2656          * to decrement mnt_cnt and hope to finish it within 10sec.
2657          */
2658         cnt = 10;
2659         while (cnt > 0 &&
2660                !may_umount(sbi->ll_mnt.mnt)) {
2661                 ssleep(1);
2662                 cnt -= 1;
2663         }
2664
2665         EXIT;
2666 }
2667
2668 int ll_remount_fs(struct super_block *sb, int *flags, char *data)
2669 {
2670         struct ll_sb_info *sbi = ll_s2sbi(sb);
2671         char *profilenm = get_profile_name(sb);
2672         int err;
2673         __u32 read_only;
2674
2675         if ((*flags & MS_RDONLY) != (sb->s_flags & SB_RDONLY)) {
2676                 read_only = *flags & MS_RDONLY;
2677                 err = obd_set_info_async(NULL, sbi->ll_md_exp,
2678                                          sizeof(KEY_READ_ONLY),
2679                                          KEY_READ_ONLY, sizeof(read_only),
2680                                          &read_only, NULL);
2681                 if (err) {
2682                         LCONSOLE_WARN("Failed to remount %s %s (%d)\n",
2683                                       profilenm, read_only ?
2684                                       "read-only" : "read-write", err);
2685                         return err;
2686                 }
2687
2688                 if (read_only)
2689                         sb->s_flags |= SB_RDONLY;
2690                 else
2691                         sb->s_flags &= ~SB_RDONLY;
2692
2693                 if (sbi->ll_flags & LL_SBI_VERBOSE)
2694                         LCONSOLE_WARN("Remounted %s %s\n", profilenm,
2695                                       read_only ?  "read-only" : "read-write");
2696         }
2697         return 0;
2698 }
2699
2700 /**
2701  * Cleanup the open handle that is cached on MDT-side.
2702  *
2703  * For open case, the client side open handling thread may hit error
2704  * after the MDT grant the open. Under such case, the client should
2705  * send close RPC to the MDT as cleanup; otherwise, the open handle
2706  * on the MDT will be leaked there until the client umount or evicted.
2707  *
2708  * In further, if someone unlinked the file, because the open handle
2709  * holds the reference on such file/object, then it will block the
2710  * subsequent threads that want to locate such object via FID.
2711  *
2712  * \param[in] sb        super block for this file-system
2713  * \param[in] open_req  pointer to the original open request
2714  */
2715 void ll_open_cleanup(struct super_block *sb, struct ptlrpc_request *open_req)
2716 {
2717         struct mdt_body                 *body;
2718         struct md_op_data               *op_data;
2719         struct ptlrpc_request           *close_req = NULL;
2720         struct obd_export               *exp       = ll_s2sbi(sb)->ll_md_exp;
2721         ENTRY;
2722
2723         body = req_capsule_server_get(&open_req->rq_pill, &RMF_MDT_BODY);
2724         OBD_ALLOC_PTR(op_data);
2725         if (op_data == NULL) {
2726                 CWARN("%s: cannot allocate op_data to release open handle for "
2727                       DFID"\n", ll_s2sbi(sb)->ll_fsname, PFID(&body->mbo_fid1));
2728
2729                 RETURN_EXIT;
2730         }
2731
2732         op_data->op_fid1 = body->mbo_fid1;
2733         op_data->op_open_handle = body->mbo_open_handle;
2734         op_data->op_mod_time = ktime_get_real_seconds();
2735         md_close(exp, op_data, NULL, &close_req);
2736         ptlrpc_req_finished(close_req);
2737         ll_finish_md_op_data(op_data);
2738
2739         EXIT;
2740 }
2741
2742 int ll_prep_inode(struct inode **inode, struct ptlrpc_request *req,
2743                   struct super_block *sb, struct lookup_intent *it)
2744 {
2745         struct ll_sb_info *sbi = NULL;
2746         struct lustre_md md = { NULL };
2747         bool default_lmv_deleted = false;
2748         int rc;
2749
2750         ENTRY;
2751
2752         LASSERT(*inode || sb);
2753         sbi = sb ? ll_s2sbi(sb) : ll_i2sbi(*inode);
2754         rc = md_get_lustre_md(sbi->ll_md_exp, req, sbi->ll_dt_exp,
2755                               sbi->ll_md_exp, &md);
2756         if (rc != 0)
2757                 GOTO(out, rc);
2758
2759         /*
2760          * clear default_lmv only if intent_getattr reply doesn't contain it.
2761          * but it needs to be done after iget, check this early because
2762          * ll_update_lsm_md() may change md.
2763          */
2764         if (it && (it->it_op & (IT_LOOKUP | IT_GETATTR)) &&
2765             S_ISDIR(md.body->mbo_mode) && !md.default_lmv)
2766                 default_lmv_deleted = true;
2767
2768         if (*inode) {
2769                 rc = ll_update_inode(*inode, &md);
2770                 if (rc != 0)
2771                         GOTO(out, rc);
2772         } else {
2773                 LASSERT(sb != NULL);
2774
2775                 /*
2776                  * At this point server returns to client's same fid as client
2777                  * generated for creating. So using ->fid1 is okay here.
2778                  */
2779                 if (!fid_is_sane(&md.body->mbo_fid1)) {
2780                         CERROR("%s: Fid is insane "DFID"\n",
2781                                 sbi->ll_fsname,
2782                                 PFID(&md.body->mbo_fid1));
2783                         GOTO(out, rc = -EINVAL);
2784                 }
2785
2786                 *inode = ll_iget(sb, cl_fid_build_ino(&md.body->mbo_fid1,
2787                                              sbi->ll_flags & LL_SBI_32BIT_API),
2788                                  &md);
2789                 if (IS_ERR(*inode)) {
2790 #ifdef CONFIG_LUSTRE_FS_POSIX_ACL
2791                         if (md.posix_acl) {
2792                                 posix_acl_release(md.posix_acl);
2793                                 md.posix_acl = NULL;
2794                         }
2795 #endif
2796                         rc = IS_ERR(*inode) ? PTR_ERR(*inode) : -ENOMEM;
2797                         *inode = NULL;
2798                         CERROR("new_inode -fatal: rc %d\n", rc);
2799                         GOTO(out, rc);
2800                 }
2801         }
2802
2803         /* Handling piggyback layout lock.
2804          * Layout lock can be piggybacked by getattr and open request.
2805          * The lsm can be applied to inode only if it comes with a layout lock
2806          * otherwise correct layout may be overwritten, for example:
2807          * 1. proc1: mdt returns a lsm but not granting layout
2808          * 2. layout was changed by another client
2809          * 3. proc2: refresh layout and layout lock granted
2810          * 4. proc1: to apply a stale layout */
2811         if (it != NULL && it->it_lock_mode != 0) {
2812                 struct lustre_handle lockh;
2813                 struct ldlm_lock *lock;
2814
2815                 lockh.cookie = it->it_lock_handle;
2816                 lock = ldlm_handle2lock(&lockh);
2817                 LASSERT(lock != NULL);
2818                 if (ldlm_has_layout(lock)) {
2819                         struct cl_object_conf conf;
2820
2821                         memset(&conf, 0, sizeof(conf));
2822                         conf.coc_opc = OBJECT_CONF_SET;
2823                         conf.coc_inode = *inode;
2824                         conf.coc_lock = lock;
2825                         conf.u.coc_layout = md.layout;
2826                         (void)ll_layout_conf(*inode, &conf);
2827                 }
2828                 LDLM_LOCK_PUT(lock);
2829         }
2830
2831         if (default_lmv_deleted)
2832                 ll_update_default_lsm_md(*inode, &md);
2833
2834         GOTO(out, rc = 0);
2835
2836 out:
2837         /* cleanup will be done if necessary */
2838         md_free_lustre_md(sbi->ll_md_exp, &md);
2839
2840         if (rc != 0 && it != NULL && it->it_op & IT_OPEN) {
2841                 ll_intent_drop_lock(it);
2842                 ll_open_cleanup(sb != NULL ? sb : (*inode)->i_sb, req);
2843         }
2844
2845         return rc;
2846 }
2847
2848 int ll_obd_statfs(struct inode *inode, void __user *arg)
2849 {
2850         struct ll_sb_info *sbi = NULL;
2851         struct obd_export *exp;
2852         char *buf = NULL;
2853         struct obd_ioctl_data *data = NULL;
2854         __u32 type;
2855         int len = 0, rc;
2856
2857         if (!inode || !(sbi = ll_i2sbi(inode)))
2858                 GOTO(out_statfs, rc = -EINVAL);
2859
2860         rc = obd_ioctl_getdata(&buf, &len, arg);
2861         if (rc)
2862                 GOTO(out_statfs, rc);
2863
2864         data = (void*)buf;
2865         if (!data->ioc_inlbuf1 || !data->ioc_inlbuf2 ||
2866             !data->ioc_pbuf1 || !data->ioc_pbuf2)
2867                 GOTO(out_statfs, rc = -EINVAL);
2868
2869         if (data->ioc_inllen1 != sizeof(__u32) ||
2870             data->ioc_inllen2 != sizeof(__u32) ||
2871             data->ioc_plen1 != sizeof(struct obd_statfs) ||
2872             data->ioc_plen2 != sizeof(struct obd_uuid))
2873                 GOTO(out_statfs, rc = -EINVAL);
2874
2875         memcpy(&type, data->ioc_inlbuf1, sizeof(__u32));
2876         if (type & LL_STATFS_LMV)
2877                 exp = sbi->ll_md_exp;
2878         else if (type & LL_STATFS_LOV)
2879                 exp = sbi->ll_dt_exp;
2880         else
2881                 GOTO(out_statfs, rc = -ENODEV);
2882
2883         rc = obd_iocontrol(IOC_OBD_STATFS, exp, len, buf, NULL);
2884         if (rc)
2885                 GOTO(out_statfs, rc);
2886 out_statfs:
2887         OBD_FREE_LARGE(buf, len);
2888         return rc;
2889 }
2890
2891 /*
2892  * this is normally called in ll_fini_md_op_data(), but sometimes it needs to
2893  * be called early to avoid deadlock.
2894  */
2895 void ll_unlock_md_op_lsm(struct md_op_data *op_data)
2896 {
2897         if (op_data->op_mea2_sem) {
2898                 up_read(op_data->op_mea2_sem);
2899                 op_data->op_mea2_sem = NULL;
2900         }
2901
2902         if (op_data->op_mea1_sem) {
2903                 up_read(op_data->op_mea1_sem);
2904                 op_data->op_mea1_sem = NULL;
2905         }
2906 }
2907
2908 /* this function prepares md_op_data hint for passing it down to MD stack. */
2909 struct md_op_data *ll_prep_md_op_data(struct md_op_data *op_data,
2910                                       struct inode *i1, struct inode *i2,
2911                                       const char *name, size_t namelen,
2912                                       __u32 mode, enum md_op_code opc,
2913                                       void *data)
2914 {
2915         LASSERT(i1 != NULL);
2916
2917         if (name == NULL) {
2918                 /* Do not reuse namelen for something else. */
2919                 if (namelen != 0)
2920                         return ERR_PTR(-EINVAL);
2921         } else {
2922                 if (namelen > ll_i2sbi(i1)->ll_namelen)
2923                         return ERR_PTR(-ENAMETOOLONG);
2924
2925                 if (!lu_name_is_valid_2(name, namelen))
2926                         return ERR_PTR(-EINVAL);
2927         }
2928
2929         if (op_data == NULL)
2930                 OBD_ALLOC_PTR(op_data);
2931
2932         if (op_data == NULL)
2933                 return ERR_PTR(-ENOMEM);
2934
2935         ll_i2gids(op_data->op_suppgids, i1, i2);
2936         op_data->op_fid1 = *ll_inode2fid(i1);
2937         op_data->op_code = opc;
2938
2939         if (S_ISDIR(i1->i_mode)) {
2940                 down_read(&ll_i2info(i1)->lli_lsm_sem);
2941                 op_data->op_mea1_sem = &ll_i2info(i1)->lli_lsm_sem;
2942                 op_data->op_mea1 = ll_i2info(i1)->lli_lsm_md;
2943                 op_data->op_default_mea1 = ll_i2info(i1)->lli_default_lsm_md;
2944         }
2945
2946         if (i2) {
2947                 op_data->op_fid2 = *ll_inode2fid(i2);
2948                 if (S_ISDIR(i2->i_mode)) {
2949                         if (i2 != i1) {
2950                                 down_read(&ll_i2info(i2)->lli_lsm_sem);
2951                                 op_data->op_mea2_sem =
2952                                                 &ll_i2info(i2)->lli_lsm_sem;
2953                         }
2954                         op_data->op_mea2 = ll_i2info(i2)->lli_lsm_md;
2955                 }
2956         } else {
2957                 fid_zero(&op_data->op_fid2);
2958         }
2959
2960         if (ll_i2sbi(i1)->ll_flags & LL_SBI_64BIT_HASH)
2961                 op_data->op_cli_flags |= CLI_HASH64;
2962
2963         if (ll_need_32bit_api(ll_i2sbi(i1)))
2964                 op_data->op_cli_flags |= CLI_API32;
2965
2966         op_data->op_name = name;
2967         op_data->op_namelen = namelen;
2968         op_data->op_mode = mode;
2969         op_data->op_mod_time = ktime_get_real_seconds();
2970         op_data->op_fsuid = from_kuid(&init_user_ns, current_fsuid());
2971         op_data->op_fsgid = from_kgid(&init_user_ns, current_fsgid());
2972         op_data->op_cap = cfs_curproc_cap_pack();
2973         op_data->op_mds = 0;
2974         if ((opc == LUSTRE_OPC_CREATE) && (name != NULL) &&
2975              filename_is_volatile(name, namelen, &op_data->op_mds)) {
2976                 op_data->op_bias |= MDS_CREATE_VOLATILE;
2977         }
2978         op_data->op_data = data;
2979
2980         return op_data;
2981 }
2982
2983 void ll_finish_md_op_data(struct md_op_data *op_data)
2984 {
2985         ll_unlock_md_op_lsm(op_data);
2986         security_release_secctx(op_data->op_file_secctx,
2987                                 op_data->op_file_secctx_size);
2988         OBD_FREE_PTR(op_data);
2989 }
2990
2991 int ll_show_options(struct seq_file *seq, struct dentry *dentry)
2992 {
2993         struct ll_sb_info *sbi;
2994
2995         LASSERT(seq && dentry);
2996         sbi = ll_s2sbi(dentry->d_sb);
2997
2998         if (sbi->ll_flags & LL_SBI_NOLCK)
2999                 seq_puts(seq, ",nolock");
3000
3001         /* "flock" is the default since 2.13, but it wasn't for many years,
3002          * so it is still useful to print this to show it is enabled.
3003          * Start to print "noflock" so it is now clear when flock is disabled.
3004          */
3005         if (sbi->ll_flags & LL_SBI_FLOCK)
3006                 seq_puts(seq, ",flock");
3007         else if (sbi->ll_flags & LL_SBI_LOCALFLOCK)
3008                 seq_puts(seq, ",localflock");
3009         else
3010                 seq_puts(seq, ",noflock");
3011
3012         if (sbi->ll_flags & LL_SBI_USER_XATTR)
3013                 seq_puts(seq, ",user_xattr");
3014
3015         if (sbi->ll_flags & LL_SBI_LAZYSTATFS)
3016                 seq_puts(seq, ",lazystatfs");
3017
3018         if (sbi->ll_flags & LL_SBI_USER_FID2PATH)
3019                 seq_puts(seq, ",user_fid2path");
3020
3021         if (sbi->ll_flags & LL_SBI_ALWAYS_PING)
3022                 seq_puts(seq, ",always_ping");
3023
3024         if (ll_sbi_has_test_dummy_encryption(sbi))
3025                 seq_puts(seq, ",test_dummy_encryption");
3026
3027         if (ll_sbi_has_encrypt(sbi))
3028                 seq_puts(seq, ",encrypt");
3029         else
3030                 seq_puts(seq, ",noencrypt");
3031
3032         RETURN(0);
3033 }
3034
3035 /**
3036  * Get obd name by cmd, and copy out to user space
3037  */
3038 int ll_get_obd_name(struct inode *inode, unsigned int cmd, unsigned long arg)
3039 {
3040         struct ll_sb_info *sbi = ll_i2sbi(inode);
3041         struct obd_device *obd;
3042         ENTRY;
3043
3044         if (cmd == OBD_IOC_GETDTNAME)
3045                 obd = class_exp2obd(sbi->ll_dt_exp);
3046         else if (cmd == OBD_IOC_GETMDNAME)
3047                 obd = class_exp2obd(sbi->ll_md_exp);
3048         else
3049                 RETURN(-EINVAL);
3050
3051         if (!obd)
3052                 RETURN(-ENOENT);
3053
3054         if (copy_to_user((void __user *)arg, obd->obd_name,
3055                          strlen(obd->obd_name) + 1))
3056                 RETURN(-EFAULT);
3057
3058         RETURN(0);
3059 }
3060
3061 static char* ll_d_path(struct dentry *dentry, char *buf, int bufsize)
3062 {
3063         char *path = NULL;
3064
3065         struct path p;
3066
3067         p.dentry = dentry;
3068         p.mnt = current->fs->root.mnt;
3069         path_get(&p);
3070         path = d_path(&p, buf, bufsize);
3071         path_put(&p);
3072         return path;
3073 }
3074
3075 void ll_dirty_page_discard_warn(struct page *page, int ioret)
3076 {
3077         char *buf, *path = NULL;
3078         struct dentry *dentry = NULL;
3079         struct inode *inode = page->mapping->host;
3080
3081         /* this can be called inside spin lock so use GFP_ATOMIC. */
3082         buf = (char *)__get_free_page(GFP_ATOMIC);
3083         if (buf != NULL) {
3084                 dentry = d_find_alias(page->mapping->host);
3085                 if (dentry != NULL)
3086                         path = ll_d_path(dentry, buf, PAGE_SIZE);
3087         }
3088
3089         /* The below message is checked in recovery-small.sh test_24b */
3090         CDEBUG(D_WARNING,
3091                "%s: dirty page discard: %s/fid: "DFID"/%s may get corrupted "
3092                "(rc %d)\n", ll_i2sbi(inode)->ll_fsname,
3093                s2lsi(page->mapping->host->i_sb)->lsi_lmd->lmd_dev,
3094                PFID(ll_inode2fid(inode)),
3095                (path && !IS_ERR(path)) ? path : "", ioret);
3096
3097         if (dentry != NULL)
3098                 dput(dentry);
3099
3100         if (buf != NULL)
3101                 free_page((unsigned long)buf);
3102 }
3103
3104 ssize_t ll_copy_user_md(const struct lov_user_md __user *md,
3105                         struct lov_user_md **kbuf)
3106 {
3107         struct lov_user_md      lum;
3108         ssize_t                 lum_size;
3109         ENTRY;
3110
3111         if (copy_from_user(&lum, md, sizeof(lum)))
3112                 RETURN(-EFAULT);
3113
3114         lum_size = ll_lov_user_md_size(&lum);
3115         if (lum_size < 0)
3116                 RETURN(lum_size);
3117
3118         OBD_ALLOC_LARGE(*kbuf, lum_size);
3119         if (*kbuf == NULL)
3120                 RETURN(-ENOMEM);
3121
3122         if (copy_from_user(*kbuf, md, lum_size) != 0) {
3123                 OBD_FREE_LARGE(*kbuf, lum_size);
3124                 RETURN(-EFAULT);
3125         }
3126
3127         RETURN(lum_size);
3128 }
3129
3130 /*
3131  * Compute llite root squash state after a change of root squash
3132  * configuration setting or add/remove of a lnet nid
3133  */
3134 void ll_compute_rootsquash_state(struct ll_sb_info *sbi)
3135 {
3136         struct root_squash_info *squash = &sbi->ll_squash;
3137         int i;
3138         bool matched;
3139         struct lnet_process_id id;
3140
3141         /* Update norootsquash flag */
3142         spin_lock(&squash->rsi_lock);
3143         if (list_empty(&squash->rsi_nosquash_nids))
3144                 sbi->ll_flags &= ~LL_SBI_NOROOTSQUASH;
3145         else {
3146                 /* Do not apply root squash as soon as one of our NIDs is
3147                  * in the nosquash_nids list */
3148                 matched = false;
3149                 i = 0;
3150                 while (LNetGetId(i++, &id) != -ENOENT) {
3151                         if (id.nid == LNET_NID_LO_0)
3152                                 continue;
3153                         if (cfs_match_nid(id.nid, &squash->rsi_nosquash_nids)) {
3154                                 matched = true;
3155                                 break;
3156                         }
3157                 }
3158                 if (matched)
3159                         sbi->ll_flags |= LL_SBI_NOROOTSQUASH;
3160                 else
3161                         sbi->ll_flags &= ~LL_SBI_NOROOTSQUASH;
3162         }
3163         spin_unlock(&squash->rsi_lock);
3164 }
3165
3166 /**
3167  * Parse linkea content to extract information about a given hardlink
3168  *
3169  * \param[in]   ldata      - Initialized linkea data
3170  * \param[in]   linkno     - Link identifier
3171  * \param[out]  parent_fid - The entry's parent FID
3172  * \param[out]  ln         - Entry name destination buffer
3173  *
3174  * \retval 0 on success
3175  * \retval Appropriate negative error code on failure
3176  */
3177 static int ll_linkea_decode(struct linkea_data *ldata, unsigned int linkno,
3178                             struct lu_fid *parent_fid, struct lu_name *ln)
3179 {
3180         unsigned int    idx;
3181         int             rc;
3182         ENTRY;
3183
3184         rc = linkea_init_with_rec(ldata);
3185         if (rc < 0)
3186                 RETURN(rc);
3187
3188         if (linkno >= ldata->ld_leh->leh_reccount)
3189                 /* beyond last link */
3190                 RETURN(-ENODATA);
3191
3192         linkea_first_entry(ldata);
3193         for (idx = 0; ldata->ld_lee != NULL; idx++) {
3194                 linkea_entry_unpack(ldata->ld_lee, &ldata->ld_reclen, ln,
3195                                     parent_fid);
3196                 if (idx == linkno)
3197                         break;
3198
3199                 linkea_next_entry(ldata);
3200         }
3201
3202         if (idx < linkno)
3203                 RETURN(-ENODATA);
3204
3205         RETURN(0);
3206 }
3207
3208 /**
3209  * Get parent FID and name of an identified link. Operation is performed for
3210  * a given link number, letting the caller iterate over linkno to list one or
3211  * all links of an entry.
3212  *
3213  * \param[in]     file - File descriptor against which to perform the operation
3214  * \param[in,out] arg  - User-filled structure containing the linkno to operate
3215  *                       on and the available size. It is eventually filled with
3216  *                       the requested information or left untouched on error
3217  *
3218  * \retval - 0 on success
3219  * \retval - Appropriate negative error code on failure
3220  */
3221 int ll_getparent(struct file *file, struct getparent __user *arg)
3222 {
3223         struct inode            *inode = file_inode(file);
3224         struct linkea_data      *ldata;
3225         struct lu_buf            buf = LU_BUF_NULL;
3226         struct lu_name           ln;
3227         struct lu_fid            parent_fid;
3228         __u32                    linkno;
3229         __u32                    name_size;
3230         int                      rc;
3231
3232         ENTRY;
3233
3234         if (!cfs_capable(CFS_CAP_DAC_READ_SEARCH) &&
3235             !(ll_i2sbi(inode)->ll_flags & LL_SBI_USER_FID2PATH))
3236                 RETURN(-EPERM);
3237
3238         if (get_user(name_size, &arg->gp_name_size))
3239                 RETURN(-EFAULT);
3240
3241         if (get_user(linkno, &arg->gp_linkno))
3242                 RETURN(-EFAULT);
3243
3244         if (name_size > PATH_MAX)
3245                 RETURN(-EINVAL);
3246
3247         OBD_ALLOC(ldata, sizeof(*ldata));
3248         if (ldata == NULL)
3249                 RETURN(-ENOMEM);
3250
3251         rc = linkea_data_new(ldata, &buf);
3252         if (rc < 0)
3253                 GOTO(ldata_free, rc);
3254
3255         rc = ll_xattr_list(inode, XATTR_NAME_LINK, XATTR_TRUSTED_T, buf.lb_buf,
3256                            buf.lb_len, OBD_MD_FLXATTR);
3257         if (rc < 0)
3258                 GOTO(lb_free, rc);
3259
3260         rc = ll_linkea_decode(ldata, linkno, &parent_fid, &ln);
3261         if (rc < 0)
3262                 GOTO(lb_free, rc);
3263
3264         if (ln.ln_namelen >= name_size)
3265                 GOTO(lb_free, rc = -EOVERFLOW);
3266
3267         if (copy_to_user(&arg->gp_fid, &parent_fid, sizeof(arg->gp_fid)))
3268                 GOTO(lb_free, rc = -EFAULT);
3269
3270         if (copy_to_user(&arg->gp_name, ln.ln_name, ln.ln_namelen))
3271                 GOTO(lb_free, rc = -EFAULT);
3272
3273         if (put_user('\0', arg->gp_name + ln.ln_namelen))
3274                 GOTO(lb_free, rc = -EFAULT);
3275
3276 lb_free:
3277         lu_buf_free(&buf);
3278 ldata_free:
3279         OBD_FREE(ldata, sizeof(*ldata));
3280
3281         RETURN(rc);
3282 }