lustre/llite/llite_lib.c

   1 /*
   2  * GPL HEADER START
   3  *
   4  * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
   5  *
   6  * This program is free software; you can redistribute it and/or modify
   7  * it under the terms of the GNU General Public License version 2 only,
   8  * as published by the Free Software Foundation.
   9  *
  10  * This program is distributed in the hope that it will be useful, but
  11  * WITHOUT ANY WARRANTY; without even the implied warranty of
  12  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  13  * General Public License version 2 for more details (a copy is included
  14  * in the LICENSE file that accompanied this code).
  15  *
  16  * You should have received a copy of the GNU General Public License
  17  * version 2 along with this program; If not, see
  18  * http://www.gnu.org/licenses/gpl-2.0.html
  19  *
  20  * GPL HEADER END
  21  */
  22 /*
  23  * Copyright (c) 2003, 2010, Oracle and/or its affiliates. All rights reserved.
  24  * Use is subject to license terms.
  25  *
  26  * Copyright (c) 2011, 2017, Intel Corporation.
  27  */
  28 /*
  29  * This file is part of Lustre, http://www.lustre.org/
  30  * Lustre is a trademark of Sun Microsystems, Inc.
  31  *
  32  * lustre/llite/llite_lib.c
  33  *
  34  * Lustre Light Super operations
  35  */
  36
  37 #define DEBUG_SUBSYSTEM S_LLITE
  38
  39 #include <linux/cpu.h>
  40 #include <linux/module.h>
  41 #include <linux/random.h>
  42 #include <linux/statfs.h>
  43 #include <linux/time.h>
  44 #include <linux/types.h>
  45 #include <libcfs/linux/linux-uuid.h>
  46 #include <linux/version.h>
  47 #include <linux/mm.h>
  48 #include <linux/user_namespace.h>
  49 #include <linux/delay.h>
  50 #include <linux/uidgid.h>
  51 #include <linux/security.h>
  52 #include <linux/fs_struct.h>
  53
  54 #ifndef HAVE_CPUS_READ_LOCK
  55 #include <libcfs/linux/linux-cpu.h>
  56 #endif
  57 #include <uapi/linux/lustre/lustre_ioctl.h>
  58 #ifdef HAVE_UAPI_LINUX_MOUNT_H
  59 #include <uapi/linux/mount.h>
  60 #endif
  61
  62 #include <lustre_ha.h>
  63 #include <lustre_dlm.h>
  64 #include <lprocfs_status.h>
  65 #include <lustre_disk.h>
  66 #include <uapi/linux/lustre/lustre_param.h>
  67 #include <lustre_log.h>
  68 #include <cl_object.h>
  69 #include <obd_cksum.h>
  70 #include "llite_internal.h"
  71
  72 struct kmem_cache *ll_file_data_slab;
  73
  74 #ifndef log2
  75 #define log2(n) ffz(~(n))
  76 #endif
  77
  78 /**
  79  * If there is only one number of core visible to Lustre,
  80  * async readahead will be disabled, to avoid massive over
  81  * subscription, we use 1/2 of active cores as default max
  82  * async readahead requests.
  83  */
  84 static inline unsigned int ll_get_ra_async_max_active(void)
  85 {
  86         return cfs_cpt_weight(cfs_cpt_tab, CFS_CPT_ANY) >> 1;
  87 }
  88
  89 static struct ll_sb_info *ll_init_sbi(void)
  90 {
  91         struct ll_sb_info *sbi = NULL;
  92         unsigned long pages;
  93         unsigned long lru_page_max;
  94         struct sysinfo si;
  95         int rc;
  96         int i;
  97
  98         ENTRY;
  99
 100         OBD_ALLOC_PTR(sbi);
 101         if (sbi == NULL)
 102                 RETURN(ERR_PTR(-ENOMEM));
 103
 104         rc = pcc_super_init(&sbi->ll_pcc_super);
 105         if (rc < 0)
 106                 GOTO(out_sbi, rc);
 107
 108         spin_lock_init(&sbi->ll_lock);
 109         mutex_init(&sbi->ll_lco.lco_lock);
 110         spin_lock_init(&sbi->ll_pp_extent_lock);
 111         spin_lock_init(&sbi->ll_process_lock);
 112         sbi->ll_rw_stats_on = 0;
 113         sbi->ll_statfs_max_age = OBD_STATFS_CACHE_SECONDS;
 114
 115         si_meminfo(&si);
 116         pages = si.totalram - si.totalhigh;
 117         lru_page_max = pages / 2;
 118
 119         sbi->ll_ra_info.ra_async_max_active = ll_get_ra_async_max_active();
 120         sbi->ll_ra_info.ll_readahead_wq =
 121                 cfs_cpt_bind_workqueue("ll-readahead-wq", cfs_cpt_tab,
 122                                        0, CFS_CPT_ANY,
 123                                        sbi->ll_ra_info.ra_async_max_active);
 124         if (IS_ERR(sbi->ll_ra_info.ll_readahead_wq))
 125                 GOTO(out_pcc, rc = PTR_ERR(sbi->ll_ra_info.ll_readahead_wq));
 126
 127         /* initialize ll_cache data */
 128         sbi->ll_cache = cl_cache_init(lru_page_max);
 129         if (sbi->ll_cache == NULL)
 130                 GOTO(out_destroy_ra, rc = -ENOMEM);
 131
 132         sbi->ll_ra_info.ra_max_pages =
 133                 min(pages / 32, SBI_DEFAULT_READ_AHEAD_MAX);
 134         sbi->ll_ra_info.ra_max_pages_per_file =
 135                 min(sbi->ll_ra_info.ra_max_pages / 4,
 136                     SBI_DEFAULT_READ_AHEAD_PER_FILE_MAX);
 137         sbi->ll_ra_info.ra_async_pages_per_file_threshold =
 138                                 sbi->ll_ra_info.ra_max_pages_per_file;
 139         sbi->ll_ra_info.ra_range_pages = SBI_DEFAULT_RA_RANGE_PAGES;
 140         sbi->ll_ra_info.ra_max_read_ahead_whole_pages = -1;
 141         atomic_set(&sbi->ll_ra_info.ra_async_inflight, 0);
 142
 143         sbi->ll_flags |= LL_SBI_VERBOSE;
 144 #ifdef ENABLE_CHECKSUM
 145         sbi->ll_flags |= LL_SBI_CHECKSUM;
 146 #endif
 147 #ifdef ENABLE_FLOCK
 148         sbi->ll_flags |= LL_SBI_FLOCK;
 149 #endif
 150
 151 #ifdef HAVE_LRU_RESIZE_SUPPORT
 152         sbi->ll_flags |= LL_SBI_LRU_RESIZE;
 153 #endif
 154         sbi->ll_flags |= LL_SBI_LAZYSTATFS;
 155
 156         for (i = 0; i <= LL_PROCESS_HIST_MAX; i++) {
 157                 spin_lock_init(&sbi->ll_rw_extents_info.pp_extents[i].
 158                                pp_r_hist.oh_lock);
 159                 spin_lock_init(&sbi->ll_rw_extents_info.pp_extents[i].
 160                                pp_w_hist.oh_lock);
 161         }
 162
 163         /* metadata statahead is enabled by default */
 164         sbi->ll_sa_running_max = LL_SA_RUNNING_DEF;
 165         sbi->ll_sa_max = LL_SA_RPC_DEF;
 166         atomic_set(&sbi->ll_sa_total, 0);
 167         atomic_set(&sbi->ll_sa_wrong, 0);
 168         atomic_set(&sbi->ll_sa_running, 0);
 169         atomic_set(&sbi->ll_agl_total, 0);
 170         sbi->ll_flags |= LL_SBI_AGL_ENABLED;
 171         sbi->ll_flags |= LL_SBI_FAST_READ;
 172         sbi->ll_flags |= LL_SBI_TINY_WRITE;
 173         ll_sbi_set_encrypt(sbi, true);
 174
 175         /* root squash */
 176         sbi->ll_squash.rsi_uid = 0;
 177         sbi->ll_squash.rsi_gid = 0;
 178         INIT_LIST_HEAD(&sbi->ll_squash.rsi_nosquash_nids);
 179         spin_lock_init(&sbi->ll_squash.rsi_lock);
 180
 181         /* Per-filesystem file heat */
 182         sbi->ll_heat_decay_weight = SBI_DEFAULT_HEAT_DECAY_WEIGHT;
 183         sbi->ll_heat_period_second = SBI_DEFAULT_HEAT_PERIOD_SECOND;
 184         RETURN(sbi);
 185 out_destroy_ra:
 186         destroy_workqueue(sbi->ll_ra_info.ll_readahead_wq);
 187 out_pcc:
 188         pcc_super_fini(&sbi->ll_pcc_super);
 189 out_sbi:
 190         OBD_FREE_PTR(sbi);
 191         RETURN(ERR_PTR(rc));
 192 }
 193
 194 static void ll_free_sbi(struct super_block *sb)
 195 {
 196         struct ll_sb_info *sbi = ll_s2sbi(sb);
 197         ENTRY;
 198
 199         if (sbi != NULL) {
 200                 if (!list_empty(&sbi->ll_squash.rsi_nosquash_nids))
 201                         cfs_free_nidlist(&sbi->ll_squash.rsi_nosquash_nids);
 202                 if (sbi->ll_ra_info.ll_readahead_wq)
 203                         destroy_workqueue(sbi->ll_ra_info.ll_readahead_wq);
 204                 if (sbi->ll_cache != NULL) {
 205                         cl_cache_decref(sbi->ll_cache);
 206                         sbi->ll_cache = NULL;
 207                 }
 208                 pcc_super_fini(&sbi->ll_pcc_super);
 209                 OBD_FREE(sbi, sizeof(*sbi));
 210         }
 211         EXIT;
 212 }
 213
 214 static int client_common_fill_super(struct super_block *sb, char *md, char *dt)
 215 {
 216         struct inode *root = NULL;
 217         struct ll_sb_info *sbi = ll_s2sbi(sb);
 218         struct obd_statfs *osfs = NULL;
 219         struct ptlrpc_request *request = NULL;
 220         struct obd_connect_data *data = NULL;
 221         struct obd_uuid *uuid;
 222         struct md_op_data *op_data;
 223         struct lustre_md lmd;
 224         u64 valid;
 225         int size, err, checksum;
 226
 227         ENTRY;
 228         sbi->ll_md_obd = class_name2obd(md);
 229         if (!sbi->ll_md_obd) {
 230                 CERROR("MD %s: not setup or attached\n", md);
 231                 RETURN(-EINVAL);
 232         }
 233
 234         OBD_ALLOC_PTR(data);
 235         if (data == NULL)
 236                 RETURN(-ENOMEM);
 237
 238         OBD_ALLOC_PTR(osfs);
 239         if (osfs == NULL) {
 240                 OBD_FREE_PTR(data);
 241                 RETURN(-ENOMEM);
 242         }
 243
 244         /* pass client page size via ocd_grant_blkbits, the server should report
 245          * back its backend blocksize for grant calculation purpose */
 246         data->ocd_grant_blkbits = PAGE_SHIFT;
 247
 248         /* indicate MDT features supported by this client */
 249         data->ocd_connect_flags = OBD_CONNECT_IBITS    | OBD_CONNECT_NODEVOH  |
 250                                   OBD_CONNECT_ATTRFID  | OBD_CONNECT_GRANT |
 251                                   OBD_CONNECT_VERSION  | OBD_CONNECT_BRW_SIZE |
 252                                   OBD_CONNECT_SRVLOCK  | OBD_CONNECT_TRUNCLOCK|
 253                                   OBD_CONNECT_MDS_CAPA | OBD_CONNECT_OSS_CAPA |
 254                                   OBD_CONNECT_CANCELSET | OBD_CONNECT_FID     |
 255                                   OBD_CONNECT_AT       | OBD_CONNECT_LOV_V3   |
 256                                   OBD_CONNECT_VBR | OBD_CONNECT_FULL20 |
 257                                   OBD_CONNECT_64BITHASH |
 258                                   OBD_CONNECT_EINPROGRESS |
 259                                   OBD_CONNECT_JOBSTATS | OBD_CONNECT_LVB_TYPE |
 260                                   OBD_CONNECT_LAYOUTLOCK | OBD_CONNECT_PINGLESS|
 261                                   OBD_CONNECT_MAX_EASIZE |
 262                                   OBD_CONNECT_FLOCK_DEAD |
 263                                   OBD_CONNECT_DISP_STRIPE | OBD_CONNECT_LFSCK |
 264                                   OBD_CONNECT_OPEN_BY_FID |
 265                                   OBD_CONNECT_DIR_STRIPE |
 266                                   OBD_CONNECT_BULK_MBITS | OBD_CONNECT_CKSUM |
 267                                   OBD_CONNECT_SUBTREE |
 268                                   OBD_CONNECT_MULTIMODRPCS |
 269                                   OBD_CONNECT_GRANT_PARAM |
 270                                   OBD_CONNECT_SHORTIO | OBD_CONNECT_FLAGS2;
 271
 272         data->ocd_connect_flags2 = OBD_CONNECT2_DIR_MIGRATE |
 273                                    OBD_CONNECT2_SUM_STATFS |
 274                                    OBD_CONNECT2_OVERSTRIPING |
 275                                    OBD_CONNECT2_FLR |
 276                                    OBD_CONNECT2_LOCK_CONVERT |
 277                                    OBD_CONNECT2_ARCHIVE_ID_ARRAY |
 278                                    OBD_CONNECT2_INC_XID |
 279                                    OBD_CONNECT2_LSOM |
 280                                    OBD_CONNECT2_ASYNC_DISCARD |
 281                                    OBD_CONNECT2_PCC |
 282                                    OBD_CONNECT2_CRUSH | OBD_CONNECT2_LSEEK |
 283                                    OBD_CONNECT2_GETATTR_PFID |
 284                                    OBD_CONNECT2_DOM_LVB;
 285
 286 #ifdef HAVE_LRU_RESIZE_SUPPORT
 287         if (sbi->ll_flags & LL_SBI_LRU_RESIZE)
 288                 data->ocd_connect_flags |= OBD_CONNECT_LRU_RESIZE;
 289 #endif
 290         data->ocd_connect_flags |= OBD_CONNECT_ACL_FLAGS;
 291
 292         data->ocd_cksum_types = obd_cksum_types_supported_client();
 293
 294         if (OBD_FAIL_CHECK(OBD_FAIL_MDC_LIGHTWEIGHT))
 295                 /* flag mdc connection as lightweight, only used for test
 296                  * purpose, use with care */
 297                 data->ocd_connect_flags |= OBD_CONNECT_LIGHTWEIGHT;
 298
 299         data->ocd_ibits_known = MDS_INODELOCK_FULL;
 300         data->ocd_version = LUSTRE_VERSION_CODE;
 301
 302         if (sb->s_flags & SB_RDONLY)
 303                 data->ocd_connect_flags |= OBD_CONNECT_RDONLY;
 304         if (sbi->ll_flags & LL_SBI_USER_XATTR)
 305                 data->ocd_connect_flags |= OBD_CONNECT_XATTR;
 306
 307 #ifdef SB_NOSEC
 308         /* Setting this indicates we correctly support S_NOSEC (See kernel
 309          * commit 9e1f1de02c2275d7172e18dc4e7c2065777611bf)
 310          */
 311         sb->s_flags |= SB_NOSEC;
 312 #endif
 313
 314         if (sbi->ll_flags & LL_SBI_FLOCK)
 315                 sbi->ll_fop = &ll_file_operations_flock;
 316         else if (sbi->ll_flags & LL_SBI_LOCALFLOCK)
 317                 sbi->ll_fop = &ll_file_operations;
 318         else
 319                 sbi->ll_fop = &ll_file_operations_noflock;
 320
 321         /* always ping even if server suppress_pings */
 322         if (sbi->ll_flags & LL_SBI_ALWAYS_PING)
 323                 data->ocd_connect_flags &= ~OBD_CONNECT_PINGLESS;
 324
 325         obd_connect_set_secctx(data);
 326         if (ll_sbi_has_encrypt(sbi))
 327                 obd_connect_set_enc(data);
 328
 329 #if defined(CONFIG_SECURITY)
 330         data->ocd_connect_flags2 |= OBD_CONNECT2_SELINUX_POLICY;
 331 #endif
 332
 333         data->ocd_brw_size = MD_MAX_BRW_SIZE;
 334
 335         err = obd_connect(NULL, &sbi->ll_md_exp, sbi->ll_md_obd,
 336                           &sbi->ll_sb_uuid, data, sbi->ll_cache);
 337         if (err == -EBUSY) {
 338                 LCONSOLE_ERROR_MSG(0x14f, "An MDT (md %s) is performing "
 339                                    "recovery, of which this client is not a "
 340                                    "part. Please wait for recovery to complete,"
 341                                    " abort, or time out.\n", md);
 342                 GOTO(out, err);
 343         } else if (err) {
 344                 CERROR("cannot connect to %s: rc = %d\n", md, err);
 345                 GOTO(out, err);
 346         }
 347
 348         sbi->ll_md_exp->exp_connect_data = *data;
 349
 350         err = obd_fid_init(sbi->ll_md_exp->exp_obd, sbi->ll_md_exp,
 351                            LUSTRE_SEQ_METADATA);
 352         if (err) {
 353                 CERROR("%s: Can't init metadata layer FID infrastructure, "
 354                        "rc = %d\n", sbi->ll_md_exp->exp_obd->obd_name, err);
 355                 GOTO(out_md, err);
 356         }
 357
 358         /* For mount, we only need fs info from MDT0, and also in DNE, it
 359          * can make sure the client can be mounted as long as MDT0 is
 360          * avaible */
 361         err = obd_statfs(NULL, sbi->ll_md_exp, osfs,
 362                         ktime_get_seconds() - sbi->ll_statfs_max_age,
 363                         OBD_STATFS_FOR_MDT0);
 364         if (err)
 365                 GOTO(out_md_fid, err);
 366
 367         /* This needs to be after statfs to ensure connect has finished.
 368          * Note that "data" does NOT contain the valid connect reply.
 369          * If connecting to a 1.8 server there will be no LMV device, so
 370          * we can access the MDC export directly and exp_connect_flags will
 371          * be non-zero, but if accessing an upgraded 2.1 server it will
 372          * have the correct flags filled in.
 373          * XXX: fill in the LMV exp_connect_flags from MDC(s). */
 374         valid = exp_connect_flags(sbi->ll_md_exp) & CLIENT_CONNECT_MDT_REQD;
 375         if (exp_connect_flags(sbi->ll_md_exp) != 0 &&
 376             valid != CLIENT_CONNECT_MDT_REQD) {
 377                 char *buf;
 378
 379                 OBD_ALLOC_WAIT(buf, PAGE_SIZE);
 380                 obd_connect_flags2str(buf, PAGE_SIZE,
 381                                       valid ^ CLIENT_CONNECT_MDT_REQD, 0, ",");
 382                 LCONSOLE_ERROR_MSG(0x170, "Server %s does not support "
 383                                    "feature(s) needed for correct operation "
 384                                    "of this client (%s). Please upgrade "
 385                                    "server or downgrade client.\n",
 386                                    sbi->ll_md_exp->exp_obd->obd_name, buf);
 387                 OBD_FREE(buf, PAGE_SIZE);
 388                 GOTO(out_md_fid, err = -EPROTO);
 389         }
 390
 391         size = sizeof(*data);
 392         err = obd_get_info(NULL, sbi->ll_md_exp, sizeof(KEY_CONN_DATA),
 393                            KEY_CONN_DATA,  &size, data);
 394         if (err) {
 395                 CERROR("%s: Get connect data failed: rc = %d\n",
 396                        sbi->ll_md_exp->exp_obd->obd_name, err);
 397                 GOTO(out_md_fid, err);
 398         }
 399
 400         LASSERT(osfs->os_bsize);
 401         sb->s_blocksize = osfs->os_bsize;
 402         sb->s_blocksize_bits = log2(osfs->os_bsize);
 403         sb->s_magic = LL_SUPER_MAGIC;
 404         sb->s_maxbytes = MAX_LFS_FILESIZE;
 405         sbi->ll_namelen = osfs->os_namelen;
 406         sbi->ll_mnt.mnt = current->fs->root.mnt;
 407
 408         if ((sbi->ll_flags & LL_SBI_USER_XATTR) &&
 409             !(data->ocd_connect_flags & OBD_CONNECT_XATTR)) {
 410                 LCONSOLE_INFO("Disabling user_xattr feature because "
 411                               "it is not supported on the server\n");
 412                 sbi->ll_flags &= ~LL_SBI_USER_XATTR;
 413         }
 414
 415         if (data->ocd_connect_flags & OBD_CONNECT_ACL) {
 416 #ifdef SB_POSIXACL
 417                 sb->s_flags |= SB_POSIXACL;
 418 #endif
 419                 sbi->ll_flags |= LL_SBI_ACL;
 420         } else {
 421                 LCONSOLE_INFO("client wants to enable acl, but mdt not!\n");
 422 #ifdef SB_POSIXACL
 423                 sb->s_flags &= ~SB_POSIXACL;
 424 #endif
 425                 sbi->ll_flags &= ~LL_SBI_ACL;
 426         }
 427
 428         if (data->ocd_connect_flags & OBD_CONNECT_64BITHASH)
 429                 sbi->ll_flags |= LL_SBI_64BIT_HASH;
 430
 431         if (data->ocd_connect_flags & OBD_CONNECT_LAYOUTLOCK)
 432                 sbi->ll_flags |= LL_SBI_LAYOUT_LOCK;
 433
 434         if (obd_connect_has_secctx(data))
 435                 sbi->ll_flags |= LL_SBI_FILE_SECCTX;
 436
 437         if (ll_sbi_has_encrypt(sbi) && !obd_connect_has_enc(data)) {
 438                 if (ll_sbi_has_test_dummy_encryption(sbi))
 439                         LCONSOLE_WARN("%s: server %s does not support encryption feature, encryption deactivated.\n",
 440                                       sbi->ll_fsname,
 441                                       sbi->ll_md_exp->exp_obd->obd_name);
 442                 ll_sbi_set_encrypt(sbi, false);
 443         }
 444
 445         if (data->ocd_ibits_known & MDS_INODELOCK_XATTR) {
 446                 if (!(data->ocd_connect_flags & OBD_CONNECT_MAX_EASIZE)) {
 447                         LCONSOLE_INFO("%s: disabling xattr cache due to "
 448                                       "unknown maximum xattr size.\n", dt);
 449                 } else if (!sbi->ll_xattr_cache_set) {
 450                         /* If xattr_cache is already set (no matter 0 or 1)
 451                          * during processing llog, it won't be enabled here. */
 452                         sbi->ll_flags |= LL_SBI_XATTR_CACHE;
 453                         sbi->ll_xattr_cache_enabled = 1;
 454                 }
 455         }
 456
 457         sbi->ll_dt_obd = class_name2obd(dt);
 458         if (!sbi->ll_dt_obd) {
 459                 CERROR("DT %s: not setup or attached\n", dt);
 460                 GOTO(out_md_fid, err = -ENODEV);
 461         }
 462
 463         /* pass client page size via ocd_grant_blkbits, the server should report
 464          * back its backend blocksize for grant calculation purpose */
 465         data->ocd_grant_blkbits = PAGE_SHIFT;
 466
 467         /* indicate OST features supported by this client */
 468         data->ocd_connect_flags = OBD_CONNECT_GRANT | OBD_CONNECT_VERSION |
 469                                   OBD_CONNECT_REQPORTAL | OBD_CONNECT_BRW_SIZE |
 470                                   OBD_CONNECT_CANCELSET | OBD_CONNECT_FID |
 471                                   OBD_CONNECT_SRVLOCK | OBD_CONNECT_TRUNCLOCK|
 472                                   OBD_CONNECT_AT | OBD_CONNECT_OSS_CAPA |
 473                                   OBD_CONNECT_VBR | OBD_CONNECT_FULL20 |
 474                                   OBD_CONNECT_64BITHASH | OBD_CONNECT_MAXBYTES |
 475                                   OBD_CONNECT_EINPROGRESS |
 476                                   OBD_CONNECT_JOBSTATS | OBD_CONNECT_LVB_TYPE |
 477                                   OBD_CONNECT_LAYOUTLOCK |
 478                                   OBD_CONNECT_PINGLESS | OBD_CONNECT_LFSCK |
 479                                   OBD_CONNECT_BULK_MBITS | OBD_CONNECT_SHORTIO |
 480                                   OBD_CONNECT_FLAGS2 | OBD_CONNECT_GRANT_SHRINK;
 481         data->ocd_connect_flags2 = OBD_CONNECT2_LOCKAHEAD |
 482                                    OBD_CONNECT2_INC_XID | OBD_CONNECT2_LSEEK;
 483
 484         if (!OBD_FAIL_CHECK(OBD_FAIL_OSC_CONNECT_GRANT_PARAM))
 485                 data->ocd_connect_flags |= OBD_CONNECT_GRANT_PARAM;
 486
 487         /* OBD_CONNECT_CKSUM should always be set, even if checksums are
 488          * disabled by default, because it can still be enabled on the
 489          * fly via /sys. As a consequence, we still need to come to an
 490          * agreement on the supported algorithms at connect time
 491          */
 492         data->ocd_connect_flags |= OBD_CONNECT_CKSUM;
 493
 494         if (OBD_FAIL_CHECK(OBD_FAIL_OSC_CKSUM_ADLER_ONLY))
 495                 data->ocd_cksum_types = OBD_CKSUM_ADLER;
 496         else
 497                 data->ocd_cksum_types = obd_cksum_types_supported_client();
 498
 499 #ifdef HAVE_LRU_RESIZE_SUPPORT
 500         data->ocd_connect_flags |= OBD_CONNECT_LRU_RESIZE;
 501 #endif
 502         /* always ping even if server suppress_pings */
 503         if (sbi->ll_flags & LL_SBI_ALWAYS_PING)
 504                 data->ocd_connect_flags &= ~OBD_CONNECT_PINGLESS;
 505
 506         if (ll_sbi_has_encrypt(sbi))
 507                 obd_connect_set_enc(data);
 508
 509         CDEBUG(D_RPCTRACE, "ocd_connect_flags: %#llx ocd_version: %d "
 510                "ocd_grant: %d\n", data->ocd_connect_flags,
 511                data->ocd_version, data->ocd_grant);
 512
 513         sbi->ll_dt_obd->obd_upcall.onu_owner = &sbi->ll_lco;
 514         sbi->ll_dt_obd->obd_upcall.onu_upcall = cl_ocd_update;
 515
 516         data->ocd_brw_size = DT_MAX_BRW_SIZE;
 517
 518         err = obd_connect(NULL, &sbi->ll_dt_exp, sbi->ll_dt_obd,
 519                           &sbi->ll_sb_uuid, data, sbi->ll_cache);
 520         if (err == -EBUSY) {
 521                 LCONSOLE_ERROR_MSG(0x150, "An OST (dt %s) is performing "
 522                                    "recovery, of which this client is not a "
 523                                    "part.  Please wait for recovery to "
 524                                    "complete, abort, or time out.\n", dt);
 525                 GOTO(out_md, err);
 526         } else if (err) {
 527                 CERROR("%s: Cannot connect to %s: rc = %d\n",
 528                        sbi->ll_dt_exp->exp_obd->obd_name, dt, err);
 529                 GOTO(out_md, err);
 530         }
 531
 532         if (ll_sbi_has_encrypt(sbi) &&
 533             !obd_connect_has_enc(&sbi->ll_dt_obd->u.lov.lov_ocd)) {
 534                 if (ll_sbi_has_test_dummy_encryption(sbi))
 535                         LCONSOLE_WARN("%s: server %s does not support encryption feature, encryption deactivated.\n",
 536                                       sbi->ll_fsname, dt);
 537                 ll_sbi_set_encrypt(sbi, false);
 538         } else if (ll_sbi_has_test_dummy_encryption(sbi)) {
 539                 LCONSOLE_WARN("Test dummy encryption mode enabled\n");
 540         }
 541
 542         sbi->ll_dt_exp->exp_connect_data = *data;
 543
 544         /* Don't change value if it was specified in the config log */
 545         if (sbi->ll_ra_info.ra_max_read_ahead_whole_pages == -1) {
 546                 sbi->ll_ra_info.ra_max_read_ahead_whole_pages =
 547                         max_t(unsigned long, SBI_DEFAULT_READ_AHEAD_WHOLE_MAX,
 548                               (data->ocd_brw_size >> PAGE_SHIFT));
 549                 if (sbi->ll_ra_info.ra_max_read_ahead_whole_pages >
 550                     sbi->ll_ra_info.ra_max_pages_per_file)
 551                         sbi->ll_ra_info.ra_max_read_ahead_whole_pages =
 552                                 sbi->ll_ra_info.ra_max_pages_per_file;
 553         }
 554
 555         err = obd_fid_init(sbi->ll_dt_exp->exp_obd, sbi->ll_dt_exp,
 556                            LUSTRE_SEQ_METADATA);
 557         if (err) {
 558                 CERROR("%s: Can't init data layer FID infrastructure, "
 559                        "rc = %d\n", sbi->ll_dt_exp->exp_obd->obd_name, err);
 560                 GOTO(out_dt, err);
 561         }
 562
 563         mutex_lock(&sbi->ll_lco.lco_lock);
 564         sbi->ll_lco.lco_flags = data->ocd_connect_flags;
 565         sbi->ll_lco.lco_md_exp = sbi->ll_md_exp;
 566         sbi->ll_lco.lco_dt_exp = sbi->ll_dt_exp;
 567         mutex_unlock(&sbi->ll_lco.lco_lock);
 568
 569         fid_zero(&sbi->ll_root_fid);
 570         err = md_get_root(sbi->ll_md_exp, get_mount_fileset(sb),
 571                            &sbi->ll_root_fid);
 572         if (err) {
 573                 CERROR("cannot mds_connect: rc = %d\n", err);
 574                 GOTO(out_lock_cn_cb, err);
 575         }
 576         if (!fid_is_sane(&sbi->ll_root_fid)) {
 577                 CERROR("%s: Invalid root fid "DFID" during mount\n",
 578                        sbi->ll_md_exp->exp_obd->obd_name,
 579                        PFID(&sbi->ll_root_fid));
 580                 GOTO(out_lock_cn_cb, err = -EINVAL);
 581         }
 582         CDEBUG(D_SUPER, "rootfid "DFID"\n", PFID(&sbi->ll_root_fid));
 583
 584         sb->s_op = &lustre_super_operations;
 585         sb->s_xattr = ll_xattr_handlers;
 586 #if THREAD_SIZE >= 8192 /*b=17630*/
 587         sb->s_export_op = &lustre_export_operations;
 588 #endif
 589 #ifdef HAVE_LUSTRE_CRYPTO
 590         llcrypt_set_ops(sb, &lustre_cryptops);
 591 #endif
 592
 593         /* make root inode
 594          * XXX: move this to after cbd setup? */
 595         valid = OBD_MD_FLGETATTR | OBD_MD_FLBLOCKS | OBD_MD_FLMODEASIZE;
 596         if (sbi->ll_flags & LL_SBI_ACL)
 597                 valid |= OBD_MD_FLACL;
 598
 599         OBD_ALLOC_PTR(op_data);
 600         if (op_data == NULL)
 601                 GOTO(out_lock_cn_cb, err = -ENOMEM);
 602
 603         op_data->op_fid1 = sbi->ll_root_fid;
 604         op_data->op_mode = 0;
 605         op_data->op_valid = valid;
 606
 607         err = md_getattr(sbi->ll_md_exp, op_data, &request);
 608
 609         OBD_FREE_PTR(op_data);
 610         if (err) {
 611                 CERROR("%s: md_getattr failed for root: rc = %d\n",
 612                        sbi->ll_md_exp->exp_obd->obd_name, err);
 613                 GOTO(out_lock_cn_cb, err);
 614         }
 615
 616         err = md_get_lustre_md(sbi->ll_md_exp, request, sbi->ll_dt_exp,
 617                                sbi->ll_md_exp, &lmd);
 618         if (err) {
 619                 CERROR("failed to understand root inode md: rc = %d\n", err);
 620                 ptlrpc_req_finished(request);
 621                 GOTO(out_lock_cn_cb, err);
 622         }
 623
 624         LASSERT(fid_is_sane(&sbi->ll_root_fid));
 625         root = ll_iget(sb, cl_fid_build_ino(&sbi->ll_root_fid,
 626                                             sbi->ll_flags & LL_SBI_32BIT_API),
 627                        &lmd);
 628         md_free_lustre_md(sbi->ll_md_exp, &lmd);
 629         ptlrpc_req_finished(request);
 630
 631         if (IS_ERR(root)) {
 632                 lmd_clear_acl(&lmd);
 633                 err = IS_ERR(root) ? PTR_ERR(root) : -EBADF;
 634                 root = NULL;
 635                 CERROR("%s: bad ll_iget() for root: rc = %d\n",
 636                        sbi->ll_fsname, err);
 637                 GOTO(out_root, err);
 638         }
 639
 640         checksum = sbi->ll_flags & LL_SBI_CHECKSUM;
 641         if (sbi->ll_checksum_set) {
 642                 err = obd_set_info_async(NULL, sbi->ll_dt_exp,
 643                                          sizeof(KEY_CHECKSUM), KEY_CHECKSUM,
 644                                          sizeof(checksum), &checksum, NULL);
 645                 if (err) {
 646                         CERROR("%s: Set checksum failed: rc = %d\n",
 647                                sbi->ll_dt_exp->exp_obd->obd_name, err);
 648                         GOTO(out_root, err);
 649                 }
 650         }
 651         cl_sb_init(sb);
 652
 653         sb->s_root = d_make_root(root);
 654         if (sb->s_root == NULL) {
 655                 err = -ENOMEM;
 656                 CERROR("%s: can't make root dentry: rc = %d\n",
 657                        sbi->ll_fsname, err);
 658                 GOTO(out_root, err);
 659         }
 660
 661         sbi->ll_sdev_orig = sb->s_dev;
 662
 663         /* We set sb->s_dev equal on all lustre clients in order to support
 664          * NFS export clustering.  NFSD requires that the FSID be the same
 665          * on all clients. */
 666         /* s_dev is also used in lt_compare() to compare two fs, but that is
 667          * only a node-local comparison. */
 668         uuid = obd_get_uuid(sbi->ll_md_exp);
 669         if (uuid != NULL)
 670                 sb->s_dev = get_uuid2int(uuid->uuid, strlen(uuid->uuid));
 671
 672         if (data != NULL)
 673                 OBD_FREE_PTR(data);
 674         if (osfs != NULL)
 675                 OBD_FREE_PTR(osfs);
 676
 677         if (sbi->ll_dt_obd) {
 678                 err = sysfs_create_link(&sbi->ll_kset.kobj,
 679                                         &sbi->ll_dt_obd->obd_kset.kobj,
 680                                         sbi->ll_dt_obd->obd_type->typ_name);
 681                 if (err < 0) {
 682                         CERROR("%s: could not register %s in llite: rc = %d\n",
 683                                dt, sbi->ll_fsname, err);
 684                         err = 0;
 685                 }
 686         }
 687
 688         if (sbi->ll_md_obd) {
 689                 err = sysfs_create_link(&sbi->ll_kset.kobj,
 690                                         &sbi->ll_md_obd->obd_kset.kobj,
 691                                         sbi->ll_md_obd->obd_type->typ_name);
 692                 if (err < 0) {
 693                         CERROR("%s: could not register %s in llite: rc = %d\n",
 694                                md, sbi->ll_fsname, err);
 695                         err = 0;
 696                 }
 697         }
 698
 699         RETURN(err);
 700 out_root:
 701         if (root)
 702                 iput(root);
 703 out_lock_cn_cb:
 704         obd_fid_fini(sbi->ll_dt_exp->exp_obd);
 705 out_dt:
 706         obd_disconnect(sbi->ll_dt_exp);
 707         sbi->ll_dt_exp = NULL;
 708         sbi->ll_dt_obd = NULL;
 709 out_md_fid:
 710         obd_fid_fini(sbi->ll_md_exp->exp_obd);
 711 out_md:
 712         obd_disconnect(sbi->ll_md_exp);
 713         sbi->ll_md_exp = NULL;
 714         sbi->ll_md_obd = NULL;
 715 out:
 716         if (data != NULL)
 717                 OBD_FREE_PTR(data);
 718         if (osfs != NULL)
 719                 OBD_FREE_PTR(osfs);
 720         return err;
 721 }
 722
 723 int ll_get_max_mdsize(struct ll_sb_info *sbi, int *lmmsize)
 724 {
 725         int size, rc;
 726
 727         size = sizeof(*lmmsize);
 728         rc = obd_get_info(NULL, sbi->ll_dt_exp, sizeof(KEY_MAX_EASIZE),
 729                           KEY_MAX_EASIZE, &size, lmmsize);
 730         if (rc != 0) {
 731                 CERROR("%s: cannot get max LOV EA size: rc = %d\n",
 732                        sbi->ll_dt_exp->exp_obd->obd_name, rc);
 733                 RETURN(rc);
 734         }
 735
 736         CDEBUG(D_INFO, "max LOV ea size: %d\n", *lmmsize);
 737
 738         size = sizeof(int);
 739         rc = obd_get_info(NULL, sbi->ll_md_exp, sizeof(KEY_MAX_EASIZE),
 740                           KEY_MAX_EASIZE, &size, lmmsize);
 741         if (rc)
 742                 CERROR("Get max mdsize error rc %d\n", rc);
 743
 744         CDEBUG(D_INFO, "max LMV ea size: %d\n", *lmmsize);
 745
 746         RETURN(rc);
 747 }
 748
 749 /**
 750  * Get the value of the default_easize parameter.
 751  *
 752  * \see client_obd::cl_default_mds_easize
 753  *
 754  * \param[in] sbi       superblock info for this filesystem
 755  * \param[out] lmmsize  pointer to storage location for value
 756  *
 757  * \retval 0            on success
 758  * \retval negative     negated errno on failure
 759  */
 760 int ll_get_default_mdsize(struct ll_sb_info *sbi, int *lmmsize)
 761 {
 762         int size, rc;
 763
 764         size = sizeof(int);
 765         rc = obd_get_info(NULL, sbi->ll_md_exp, sizeof(KEY_DEFAULT_EASIZE),
 766                          KEY_DEFAULT_EASIZE, &size, lmmsize);
 767         if (rc)
 768                 CERROR("Get default mdsize error rc %d\n", rc);
 769
 770         RETURN(rc);
 771 }
 772
 773 /**
 774  * Set the default_easize parameter to the given value.
 775  *
 776  * \see client_obd::cl_default_mds_easize
 777  *
 778  * \param[in] sbi       superblock info for this filesystem
 779  * \param[in] lmmsize   the size to set
 780  *
 781  * \retval 0            on success
 782  * \retval negative     negated errno on failure
 783  */
 784 int ll_set_default_mdsize(struct ll_sb_info *sbi, int lmmsize)
 785 {
 786         int rc;
 787
 788         if (lmmsize < sizeof(struct lov_mds_md) ||
 789             lmmsize > OBD_MAX_DEFAULT_EA_SIZE)
 790                 return -EINVAL;
 791
 792         rc = obd_set_info_async(NULL, sbi->ll_md_exp,
 793                                 sizeof(KEY_DEFAULT_EASIZE), KEY_DEFAULT_EASIZE,
 794                                 sizeof(int), &lmmsize, NULL);
 795
 796         RETURN(rc);
 797 }
 798
 799 static void client_common_put_super(struct super_block *sb)
 800 {
 801         struct ll_sb_info *sbi = ll_s2sbi(sb);
 802         ENTRY;
 803
 804         cl_sb_fini(sb);
 805
 806         obd_fid_fini(sbi->ll_dt_exp->exp_obd);
 807         obd_disconnect(sbi->ll_dt_exp);
 808         sbi->ll_dt_exp = NULL;
 809
 810         ll_debugfs_unregister_super(sb);
 811
 812         obd_fid_fini(sbi->ll_md_exp->exp_obd);
 813         obd_disconnect(sbi->ll_md_exp);
 814         sbi->ll_md_exp = NULL;
 815
 816         EXIT;
 817 }
 818
 819 void ll_kill_super(struct super_block *sb)
 820 {
 821         struct ll_sb_info *sbi;
 822         ENTRY;
 823
 824         /* not init sb ?*/
 825         if (!(sb->s_flags & SB_ACTIVE))
 826                 return;
 827
 828         sbi = ll_s2sbi(sb);
 829         /* we need restore s_dev from changed for clustred NFS before put_super
 830          * because new kernels have cached s_dev and change sb->s_dev in
 831          * put_super not affected real removing devices */
 832         if (sbi) {
 833                 sb->s_dev = sbi->ll_sdev_orig;
 834
 835                 /* wait running statahead threads to quit */
 836                 while (atomic_read(&sbi->ll_sa_running) > 0)
 837                         schedule_timeout_uninterruptible(
 838                                 cfs_time_seconds(1) >> 3);
 839         }
 840
 841         EXIT;
 842 }
 843
 844 static inline int ll_set_opt(const char *opt, char *data, int fl)
 845 {
 846         if (strncmp(opt, data, strlen(opt)) != 0)
 847                 return 0;
 848         else
 849                 return fl;
 850 }
 851
 852 /* non-client-specific mount options are parsed in lmd_parse */
 853 static int ll_options(char *options, struct ll_sb_info *sbi)
 854 {
 855         int tmp;
 856         char *s1 = options, *s2;
 857         int *flags = &sbi->ll_flags;
 858         ENTRY;
 859
 860         if (!options)
 861                 RETURN(0);
 862
 863         CDEBUG(D_CONFIG, "Parsing opts %s\n", options);
 864
 865         while (*s1) {
 866                 CDEBUG(D_SUPER, "next opt=%s\n", s1);
 867                 tmp = ll_set_opt("nolock", s1, LL_SBI_NOLCK);
 868                 if (tmp) {
 869                         *flags |= tmp;
 870                         goto next;
 871                 }
 872                 tmp = ll_set_opt("flock", s1, LL_SBI_FLOCK);
 873                 if (tmp) {
 874                         *flags = (*flags & ~LL_SBI_LOCALFLOCK) | tmp;
 875                         goto next;
 876                 }
 877                 tmp = ll_set_opt("localflock", s1, LL_SBI_LOCALFLOCK);
 878                 if (tmp) {
 879                         *flags = (*flags & ~LL_SBI_FLOCK) | tmp;
 880                         goto next;
 881                 }
 882                 tmp = ll_set_opt("noflock", s1, LL_SBI_FLOCK|LL_SBI_LOCALFLOCK);
 883                 if (tmp) {
 884                         *flags &= ~tmp;
 885                         goto next;
 886                 }
 887                 tmp = ll_set_opt("user_xattr", s1, LL_SBI_USER_XATTR);
 888                 if (tmp) {
 889                         *flags |= tmp;
 890                         goto next;
 891                 }
 892                 tmp = ll_set_opt("nouser_xattr", s1, LL_SBI_USER_XATTR);
 893                 if (tmp) {
 894                         *flags &= ~tmp;
 895                         goto next;
 896                 }
 897                 tmp = ll_set_opt("context", s1, 1);
 898                 if (tmp)
 899                         goto next;
 900                 tmp = ll_set_opt("fscontext", s1, 1);
 901                 if (tmp)
 902                         goto next;
 903                 tmp = ll_set_opt("defcontext", s1, 1);
 904                 if (tmp)
 905                         goto next;
 906                 tmp = ll_set_opt("rootcontext", s1, 1);
 907                 if (tmp)
 908                         goto next;
 909                 tmp = ll_set_opt("user_fid2path", s1, LL_SBI_USER_FID2PATH);
 910                 if (tmp) {
 911                         *flags |= tmp;
 912                         goto next;
 913                 }
 914                 tmp = ll_set_opt("nouser_fid2path", s1, LL_SBI_USER_FID2PATH);
 915                 if (tmp) {
 916                         *flags &= ~tmp;
 917                         goto next;
 918                 }
 919
 920                 tmp = ll_set_opt("checksum", s1, LL_SBI_CHECKSUM);
 921                 if (tmp) {
 922                         *flags |= tmp;
 923                         sbi->ll_checksum_set = 1;
 924                         goto next;
 925                 }
 926                 tmp = ll_set_opt("nochecksum", s1, LL_SBI_CHECKSUM);
 927                 if (tmp) {
 928                         *flags &= ~tmp;
 929                         sbi->ll_checksum_set = 1;
 930                         goto next;
 931                 }
 932                 tmp = ll_set_opt("lruresize", s1, LL_SBI_LRU_RESIZE);
 933                 if (tmp) {
 934                         *flags |= tmp;
 935                         goto next;
 936                 }
 937                 tmp = ll_set_opt("nolruresize", s1, LL_SBI_LRU_RESIZE);
 938                 if (tmp) {
 939                         *flags &= ~tmp;
 940                         goto next;
 941                 }
 942                 tmp = ll_set_opt("lazystatfs", s1, LL_SBI_LAZYSTATFS);
 943                 if (tmp) {
 944                         *flags |= tmp;
 945                         goto next;
 946                 }
 947                 tmp = ll_set_opt("nolazystatfs", s1, LL_SBI_LAZYSTATFS);
 948                 if (tmp) {
 949                         *flags &= ~tmp;
 950                         goto next;
 951                 }
 952                 tmp = ll_set_opt("32bitapi", s1, LL_SBI_32BIT_API);
 953                 if (tmp) {
 954                         *flags |= tmp;
 955                         goto next;
 956                 }
 957                 tmp = ll_set_opt("verbose", s1, LL_SBI_VERBOSE);
 958                 if (tmp) {
 959                         *flags |= tmp;
 960                         goto next;
 961                 }
 962                 tmp = ll_set_opt("noverbose", s1, LL_SBI_VERBOSE);
 963                 if (tmp) {
 964                         *flags &= ~tmp;
 965                         goto next;
 966                 }
 967                 tmp = ll_set_opt("always_ping", s1, LL_SBI_ALWAYS_PING);
 968                 if (tmp) {
 969                         *flags |= tmp;
 970                         goto next;
 971                 }
 972                 tmp = ll_set_opt("test_dummy_encryption", s1,
 973                                  LL_SBI_TEST_DUMMY_ENCRYPTION);
 974                 if (tmp) {
 975 #ifdef HAVE_LUSTRE_CRYPTO
 976                         *flags |= tmp;
 977 #else
 978                         LCONSOLE_WARN("Test dummy encryption mount option ignored: encryption not supported\n");
 979 #endif
 980                         goto next;
 981                 }
 982                 tmp = ll_set_opt("noencrypt", s1, LL_SBI_ENCRYPT);
 983                 if (tmp) {
 984 #ifdef HAVE_LUSTRE_CRYPTO
 985                         *flags &= ~tmp;
 986 #else
 987                         LCONSOLE_WARN("noencrypt mount option ignored: encryption not supported\n");
 988 #endif
 989                         goto next;
 990                 }
 991                 LCONSOLE_ERROR_MSG(0x152, "Unknown option '%s', won't mount.\n",
 992                                    s1);
 993                 RETURN(-EINVAL);
 994
 995 next:
 996                 /* Find next opt */
 997                 s2 = strchr(s1, ',');
 998                 if (s2 == NULL)
 999                         break;
1000                 s1 = s2 + 1;
1001         }
1002         RETURN(0);
1003 }
1004
1005 void ll_lli_init(struct ll_inode_info *lli)
1006 {
1007         lli->lli_inode_magic = LLI_INODE_MAGIC;
1008         lli->lli_flags = 0;
1009         spin_lock_init(&lli->lli_lock);
1010         lli->lli_posix_acl = NULL;
1011         /* Do not set lli_fid, it has been initialized already. */
1012         fid_zero(&lli->lli_pfid);
1013         lli->lli_mds_read_och = NULL;
1014         lli->lli_mds_write_och = NULL;
1015         lli->lli_mds_exec_och = NULL;
1016         lli->lli_open_fd_read_count = 0;
1017         lli->lli_open_fd_write_count = 0;
1018         lli->lli_open_fd_exec_count = 0;
1019         mutex_init(&lli->lli_och_mutex);
1020         spin_lock_init(&lli->lli_agl_lock);
1021         spin_lock_init(&lli->lli_layout_lock);
1022         ll_layout_version_set(lli, CL_LAYOUT_GEN_NONE);
1023         lli->lli_clob = NULL;
1024
1025         init_rwsem(&lli->lli_xattrs_list_rwsem);
1026         mutex_init(&lli->lli_xattrs_enq_lock);
1027
1028         LASSERT(lli->lli_vfs_inode.i_mode != 0);
1029         if (S_ISDIR(lli->lli_vfs_inode.i_mode)) {
1030                 lli->lli_opendir_key = NULL;
1031                 lli->lli_sai = NULL;
1032                 spin_lock_init(&lli->lli_sa_lock);
1033                 lli->lli_opendir_pid = 0;
1034                 lli->lli_sa_enabled = 0;
1035                 init_rwsem(&lli->lli_lsm_sem);
1036         } else {
1037                 mutex_init(&lli->lli_size_mutex);
1038                 mutex_init(&lli->lli_setattr_mutex);
1039                 lli->lli_symlink_name = NULL;
1040                 ll_trunc_sem_init(&lli->lli_trunc_sem);
1041                 range_lock_tree_init(&lli->lli_write_tree);
1042                 init_rwsem(&lli->lli_glimpse_sem);
1043                 lli->lli_glimpse_time = ktime_set(0, 0);
1044                 INIT_LIST_HEAD(&lli->lli_agl_list);
1045                 lli->lli_agl_index = 0;
1046                 lli->lli_async_rc = 0;
1047                 spin_lock_init(&lli->lli_heat_lock);
1048                 obd_heat_clear(lli->lli_heat_instances, OBD_HEAT_COUNT);
1049                 lli->lli_heat_flags = 0;
1050                 mutex_init(&lli->lli_pcc_lock);
1051                 lli->lli_pcc_state = PCC_STATE_FL_NONE;
1052                 lli->lli_pcc_inode = NULL;
1053                 lli->lli_pcc_dsflags = PCC_DATASET_INVALID;
1054                 lli->lli_pcc_generation = 0;
1055                 mutex_init(&lli->lli_group_mutex);
1056                 lli->lli_group_users = 0;
1057                 lli->lli_group_gid = 0;
1058         }
1059         mutex_init(&lli->lli_layout_mutex);
1060         memset(lli->lli_jobid, 0, sizeof(lli->lli_jobid));
1061 }
1062
1063 #define MAX_STRING_SIZE 128
1064
1065 #ifndef HAVE_SUPER_SETUP_BDI_NAME
1066
1067 #define LSI_BDI_INITIALIZED     0x00400000
1068
1069 #ifndef HAVE_BDI_CAP_MAP_COPY
1070 # define BDI_CAP_MAP_COPY       0
1071 #endif
1072
1073 static int super_setup_bdi_name(struct super_block *sb, char *fmt, ...)
1074 {
1075         struct  lustre_sb_info *lsi = s2lsi(sb);
1076         char buf[MAX_STRING_SIZE];
1077         va_list args;
1078         int err;
1079
1080         err = bdi_init(&lsi->lsi_bdi);
1081         if (err)
1082                 return err;
1083
1084         lsi->lsi_flags |= LSI_BDI_INITIALIZED;
1085         lsi->lsi_bdi.capabilities = BDI_CAP_MAP_COPY;
1086         lsi->lsi_bdi.name = "lustre";
1087         va_start(args, fmt);
1088         vsnprintf(buf, MAX_STRING_SIZE, fmt, args);
1089         va_end(args);
1090         err = bdi_register(&lsi->lsi_bdi, NULL, "%s", buf);
1091         va_end(args);
1092         if (!err)
1093                 sb->s_bdi = &lsi->lsi_bdi;
1094
1095         return err;
1096 }
1097 #endif /* !HAVE_SUPER_SETUP_BDI_NAME */
1098
1099 int ll_fill_super(struct super_block *sb)
1100 {
1101         struct  lustre_profile *lprof = NULL;
1102         struct  lustre_sb_info *lsi = s2lsi(sb);
1103         struct  ll_sb_info *sbi = NULL;
1104         char    *dt = NULL, *md = NULL;
1105         char    *profilenm = get_profile_name(sb);
1106         struct config_llog_instance *cfg;
1107         /* %p for void* in printf needs 16+2 characters: 0xffffffffffffffff */
1108         const int instlen = LUSTRE_MAXINSTANCE + 2;
1109         unsigned long cfg_instance = ll_get_cfg_instance(sb);
1110         char name[MAX_STRING_SIZE];
1111         int md_len = 0;
1112         int dt_len = 0;
1113         uuid_t uuid;
1114         char *ptr;
1115         int len;
1116         int err;
1117
1118         ENTRY;
1119         /* for ASLR, to map between cfg_instance and hashed ptr */
1120         CDEBUG(D_VFSTRACE, "VFS Op: cfg_instance %s-%016lx (sb %p)\n",
1121                profilenm, cfg_instance, sb);
1122
1123         OBD_ALLOC_PTR(cfg);
1124         if (cfg == NULL)
1125                 GOTO(out_free_cfg, err = -ENOMEM);
1126
1127         /* client additional sb info */
1128         lsi->lsi_llsbi = sbi = ll_init_sbi();
1129         if (IS_ERR(sbi))
1130                 GOTO(out_free_cfg, err = PTR_ERR(sbi));
1131
1132         err = ll_options(lsi->lsi_lmd->lmd_opts, sbi);
1133         if (err)
1134                 GOTO(out_free_cfg, err);
1135
1136         /* kernel >= 2.6.38 store dentry operations in sb->s_d_op. */
1137         sb->s_d_op = &ll_d_ops;
1138
1139         /* UUID handling */
1140         generate_random_uuid(uuid.b);
1141         snprintf(sbi->ll_sb_uuid.uuid, sizeof(sbi->ll_sb_uuid), "%pU", uuid.b);
1142
1143         CDEBUG(D_CONFIG, "llite sb uuid: %s\n", sbi->ll_sb_uuid.uuid);
1144
1145         /* Get fsname */
1146         len = strlen(profilenm);
1147         ptr = strrchr(profilenm, '-');
1148         if (ptr && (strcmp(ptr, "-client") == 0))
1149                 len -= 7;
1150
1151         if (len > LUSTRE_MAXFSNAME) {
1152                 if (unlikely(len >= MAX_STRING_SIZE))
1153                         len = MAX_STRING_SIZE - 1;
1154                 strncpy(name, profilenm, len);
1155                 name[len] = '\0';
1156                 err = -ENAMETOOLONG;
1157                 CERROR("%s: fsname longer than %u characters: rc = %d\n",
1158                        name, LUSTRE_MAXFSNAME, err);
1159                 GOTO(out_free_cfg, err);
1160         }
1161         strncpy(sbi->ll_fsname, profilenm, len);
1162         sbi->ll_fsname[len] = '\0';
1163
1164         /* Mount info */
1165         snprintf(name, sizeof(name), "%.*s-%016lx", len,
1166                  profilenm, cfg_instance);
1167
1168         err = super_setup_bdi_name(sb, "%s", name);
1169         if (err)
1170                 GOTO(out_free_cfg, err);
1171
1172         /* Call ll_debugfs_register_super() before lustre_process_log()
1173          * so that "llite.*.*" params can be processed correctly.
1174          */
1175         err = ll_debugfs_register_super(sb, name);
1176         if (err < 0) {
1177                 CERROR("%s: could not register mountpoint in llite: rc = %d\n",
1178                        sbi->ll_fsname, err);
1179                 err = 0;
1180         }
1181
1182         /* The cfg_instance is a value unique to this super, in case some
1183          * joker tries to mount the same fs at two mount points.
1184          */
1185         cfg->cfg_instance = cfg_instance;
1186         cfg->cfg_uuid = lsi->lsi_llsbi->ll_sb_uuid;
1187         cfg->cfg_callback = class_config_llog_handler;
1188         cfg->cfg_sub_clds = CONFIG_SUB_CLIENT;
1189         /* set up client obds */
1190         err = lustre_process_log(sb, profilenm, cfg);
1191         if (err < 0)
1192                 GOTO(out_debugfs, err);
1193
1194         /* Profile set with LCFG_MOUNTOPT so we can find our mdc and osc obds */
1195         lprof = class_get_profile(profilenm);
1196         if (lprof == NULL) {
1197                 LCONSOLE_ERROR_MSG(0x156, "The client profile '%s' could not be"
1198                                    " read from the MGS.  Does that filesystem "
1199                                    "exist?\n", profilenm);
1200                 GOTO(out_debugfs, err = -EINVAL);
1201         }
1202         CDEBUG(D_CONFIG, "Found profile %s: mdc=%s osc=%s\n", profilenm,
1203                lprof->lp_md, lprof->lp_dt);
1204
1205         dt_len = strlen(lprof->lp_dt) + instlen + 2;
1206         OBD_ALLOC(dt, dt_len);
1207         if (!dt)
1208                 GOTO(out_profile, err = -ENOMEM);
1209         snprintf(dt, dt_len - 1, "%s-%016lx", lprof->lp_dt, cfg_instance);
1210
1211         md_len = strlen(lprof->lp_md) + instlen + 2;
1212         OBD_ALLOC(md, md_len);
1213         if (!md)
1214                 GOTO(out_free_dt, err = -ENOMEM);
1215         snprintf(md, md_len - 1, "%s-%016lx", lprof->lp_md, cfg_instance);
1216
1217         /* connections, registrations, sb setup */
1218         err = client_common_fill_super(sb, md, dt);
1219         if (err < 0)
1220                 GOTO(out_free_md, err);
1221
1222         sbi->ll_client_common_fill_super_succeeded = 1;
1223
1224 out_free_md:
1225         if (md)
1226                 OBD_FREE(md, md_len);
1227 out_free_dt:
1228         if (dt)
1229                 OBD_FREE(dt, dt_len);
1230 out_profile:
1231         if (lprof)
1232                 class_put_profile(lprof);
1233 out_debugfs:
1234         if (err < 0)
1235                 ll_debugfs_unregister_super(sb);
1236 out_free_cfg:
1237         if (cfg)
1238                 OBD_FREE_PTR(cfg);
1239
1240         if (err)
1241                 ll_put_super(sb);
1242         else if (sbi->ll_flags & LL_SBI_VERBOSE)
1243                 LCONSOLE_WARN("Mounted %s\n", profilenm);
1244         RETURN(err);
1245 } /* ll_fill_super */
1246
1247 void ll_put_super(struct super_block *sb)
1248 {
1249         struct config_llog_instance cfg, params_cfg;
1250         struct obd_device *obd;
1251         struct lustre_sb_info *lsi = s2lsi(sb);
1252         struct ll_sb_info *sbi = ll_s2sbi(sb);
1253         char *profilenm = get_profile_name(sb);
1254         unsigned long cfg_instance = ll_get_cfg_instance(sb);
1255         long ccc_count;
1256         int next, force = 1, rc = 0;
1257         ENTRY;
1258
1259         if (IS_ERR(sbi))
1260                 GOTO(out_no_sbi, 0);
1261
1262         /* Should replace instance_id with something better for ASLR */
1263         CDEBUG(D_VFSTRACE, "VFS Op: cfg_instance %s-%016lx (sb %p)\n",
1264                profilenm, cfg_instance, sb);
1265
1266         cfg.cfg_instance = cfg_instance;
1267         lustre_end_log(sb, profilenm, &cfg);
1268
1269         params_cfg.cfg_instance = cfg_instance;
1270         lustre_end_log(sb, PARAMS_FILENAME, &params_cfg);
1271
1272         if (sbi->ll_md_exp) {
1273                 obd = class_exp2obd(sbi->ll_md_exp);
1274                 if (obd)
1275                         force = obd->obd_force;
1276         }
1277
1278         /* Wait for unstable pages to be committed to stable storage */
1279         if (force == 0) {
1280                 rc = l_wait_event_abortable(
1281                         sbi->ll_cache->ccc_unstable_waitq,
1282                         atomic_long_read(&sbi->ll_cache->ccc_unstable_nr) == 0);
1283         }
1284
1285         ccc_count = atomic_long_read(&sbi->ll_cache->ccc_unstable_nr);
1286         if (force == 0 && rc != -ERESTARTSYS)
1287                 LASSERTF(ccc_count == 0, "count: %li\n", ccc_count);
1288
1289         /* We need to set force before the lov_disconnect in
1290          * lustre_common_put_super, since l_d cleans up osc's as well.
1291          */
1292         if (force) {
1293                 next = 0;
1294                 while ((obd = class_devices_in_group(&sbi->ll_sb_uuid,
1295                                                      &next)) != NULL) {
1296                         obd->obd_force = force;
1297                 }
1298         }
1299
1300         if (sbi->ll_client_common_fill_super_succeeded) {
1301                 /* Only if client_common_fill_super succeeded */
1302                 client_common_put_super(sb);
1303         }
1304
1305         next = 0;
1306         while ((obd = class_devices_in_group(&sbi->ll_sb_uuid, &next)))
1307                 class_manual_cleanup(obd);
1308
1309         if (sbi->ll_flags & LL_SBI_VERBOSE)
1310                 LCONSOLE_WARN("Unmounted %s\n", profilenm ? profilenm : "");
1311
1312         if (profilenm)
1313                 class_del_profile(profilenm);
1314
1315 #ifndef HAVE_SUPER_SETUP_BDI_NAME
1316         if (lsi->lsi_flags & LSI_BDI_INITIALIZED) {
1317                 bdi_destroy(&lsi->lsi_bdi);
1318                 lsi->lsi_flags &= ~LSI_BDI_INITIALIZED;
1319         }
1320 #endif
1321
1322         ll_free_sbi(sb);
1323         lsi->lsi_llsbi = NULL;
1324 out_no_sbi:
1325         lustre_common_put_super(sb);
1326
1327         cl_env_cache_purge(~0);
1328
1329         module_put(THIS_MODULE);
1330
1331         EXIT;
1332 } /* client_put_super */
1333
1334 struct inode *ll_inode_from_resource_lock(struct ldlm_lock *lock)
1335 {
1336         struct inode *inode = NULL;
1337
1338         /* NOTE: we depend on atomic igrab() -bzzz */
1339         lock_res_and_lock(lock);
1340         if (lock->l_resource->lr_lvb_inode) {
1341                 struct ll_inode_info * lli;
1342                 lli = ll_i2info(lock->l_resource->lr_lvb_inode);
1343                 if (lli->lli_inode_magic == LLI_INODE_MAGIC) {
1344                         inode = igrab(lock->l_resource->lr_lvb_inode);
1345                 } else {
1346                         inode = lock->l_resource->lr_lvb_inode;
1347                         LDLM_DEBUG_LIMIT(inode->i_state & I_FREEING ?  D_INFO :
1348                                          D_WARNING, lock, "lr_lvb_inode %p is "
1349                                          "bogus: magic %08x",
1350                                          lock->l_resource->lr_lvb_inode,
1351                                          lli->lli_inode_magic);
1352                         inode = NULL;
1353                 }
1354         }
1355         unlock_res_and_lock(lock);
1356         return inode;
1357 }
1358
1359 void ll_dir_clear_lsm_md(struct inode *inode)
1360 {
1361         struct ll_inode_info *lli = ll_i2info(inode);
1362
1363         LASSERT(S_ISDIR(inode->i_mode));
1364
1365         if (lli->lli_lsm_md) {
1366                 lmv_free_memmd(lli->lli_lsm_md);
1367                 lli->lli_lsm_md = NULL;
1368         }
1369
1370         if (lli->lli_default_lsm_md) {
1371                 lmv_free_memmd(lli->lli_default_lsm_md);
1372                 lli->lli_default_lsm_md = NULL;
1373         }
1374 }
1375
1376 static struct inode *ll_iget_anon_dir(struct super_block *sb,
1377                                       const struct lu_fid *fid,
1378                                       struct lustre_md *md)
1379 {
1380         struct ll_sb_info       *sbi = ll_s2sbi(sb);
1381         struct mdt_body         *body = md->body;
1382         struct inode            *inode;
1383         ino_t                   ino;
1384         ENTRY;
1385
1386         ino = cl_fid_build_ino(fid, sbi->ll_flags & LL_SBI_32BIT_API);
1387         inode = iget_locked(sb, ino);
1388         if (inode == NULL) {
1389                 CERROR("%s: failed get simple inode "DFID": rc = -ENOENT\n",
1390                        sbi->ll_fsname, PFID(fid));
1391                 RETURN(ERR_PTR(-ENOENT));
1392         }
1393
1394         if (inode->i_state & I_NEW) {
1395                 struct ll_inode_info *lli = ll_i2info(inode);
1396                 struct lmv_stripe_md *lsm = md->lmv;
1397
1398                 inode->i_mode = (inode->i_mode & ~S_IFMT) |
1399                                 (body->mbo_mode & S_IFMT);
1400                 LASSERTF(S_ISDIR(inode->i_mode), "Not slave inode "DFID"\n",
1401                          PFID(fid));
1402
1403                 inode->i_mtime.tv_sec = 0;
1404                 inode->i_atime.tv_sec = 0;
1405                 inode->i_ctime.tv_sec = 0;
1406                 inode->i_rdev = 0;
1407
1408 #ifdef HAVE_BACKING_DEV_INFO
1409                 /* initializing backing dev info. */
1410                 inode->i_mapping->backing_dev_info =
1411                                                 &s2lsi(inode->i_sb)->lsi_bdi;
1412 #endif
1413                 inode->i_op = &ll_dir_inode_operations;
1414                 inode->i_fop = &ll_dir_operations;
1415                 lli->lli_fid = *fid;
1416                 ll_lli_init(lli);
1417
1418                 LASSERT(lsm != NULL);
1419                 /* master object FID */
1420                 lli->lli_pfid = body->mbo_fid1;
1421                 CDEBUG(D_INODE, "lli %p slave "DFID" master "DFID"\n",
1422                        lli, PFID(fid), PFID(&lli->lli_pfid));
1423                 unlock_new_inode(inode);
1424         }
1425
1426         RETURN(inode);
1427 }
1428
1429 static int ll_init_lsm_md(struct inode *inode, struct lustre_md *md)
1430 {
1431         struct lu_fid *fid;
1432         struct lmv_stripe_md *lsm = md->lmv;
1433         struct ll_inode_info *lli = ll_i2info(inode);
1434         int i;
1435
1436         LASSERT(lsm != NULL);
1437
1438         CDEBUG(D_INODE, "%s: "DFID" set dir layout:\n",
1439                ll_i2sbi(inode)->ll_fsname, PFID(&lli->lli_fid));
1440         lsm_md_dump(D_INODE, lsm);
1441
1442         if (!lmv_dir_striped(lsm))
1443                 goto out;
1444
1445         /* XXX sigh, this lsm_root initialization should be in
1446          * LMV layer, but it needs ll_iget right now, so we
1447          * put this here right now. */
1448         for (i = 0; i < lsm->lsm_md_stripe_count; i++) {
1449                 fid = &lsm->lsm_md_oinfo[i].lmo_fid;
1450                 LASSERT(lsm->lsm_md_oinfo[i].lmo_root == NULL);
1451
1452                 if (!fid_is_sane(fid))
1453                         continue;
1454
1455                 /* Unfortunately ll_iget will call ll_update_inode,
1456                  * where the initialization of slave inode is slightly
1457                  * different, so it reset lsm_md to NULL to avoid
1458                  * initializing lsm for slave inode. */
1459                 lsm->lsm_md_oinfo[i].lmo_root =
1460                                 ll_iget_anon_dir(inode->i_sb, fid, md);
1461                 if (IS_ERR(lsm->lsm_md_oinfo[i].lmo_root)) {
1462                         int rc = PTR_ERR(lsm->lsm_md_oinfo[i].lmo_root);
1463
1464                         lsm->lsm_md_oinfo[i].lmo_root = NULL;
1465                         while (i-- > 0) {
1466                                 iput(lsm->lsm_md_oinfo[i].lmo_root);
1467                                 lsm->lsm_md_oinfo[i].lmo_root = NULL;
1468                         }
1469                         return rc;
1470                 }
1471         }
1472 out:
1473         lli->lli_lsm_md = lsm;
1474
1475         return 0;
1476 }
1477
1478 static void ll_update_default_lsm_md(struct inode *inode, struct lustre_md *md)
1479 {
1480         struct ll_inode_info *lli = ll_i2info(inode);
1481
1482         if (!md->default_lmv) {
1483                 /* clear default lsm */
1484                 if (lli->lli_default_lsm_md) {
1485                         down_write(&lli->lli_lsm_sem);
1486                         if (lli->lli_default_lsm_md) {
1487                                 lmv_free_memmd(lli->lli_default_lsm_md);
1488                                 lli->lli_default_lsm_md = NULL;
1489                         }
1490                         up_write(&lli->lli_lsm_sem);
1491                 }
1492         } else if (lli->lli_default_lsm_md) {
1493                 /* update default lsm if it changes */
1494                 down_read(&lli->lli_lsm_sem);
1495                 if (lli->lli_default_lsm_md &&
1496                     !lsm_md_eq(lli->lli_default_lsm_md, md->default_lmv)) {
1497                         up_read(&lli->lli_lsm_sem);
1498                         down_write(&lli->lli_lsm_sem);
1499                         if (lli->lli_default_lsm_md)
1500                                 lmv_free_memmd(lli->lli_default_lsm_md);
1501                         lli->lli_default_lsm_md = md->default_lmv;
1502                         lsm_md_dump(D_INODE, md->default_lmv);
1503                         md->default_lmv = NULL;
1504                         up_write(&lli->lli_lsm_sem);
1505                 } else {
1506                         up_read(&lli->lli_lsm_sem);
1507                 }
1508         } else {
1509                 /* init default lsm */
1510                 down_write(&lli->lli_lsm_sem);
1511                 lli->lli_default_lsm_md = md->default_lmv;
1512                 lsm_md_dump(D_INODE, md->default_lmv);
1513                 md->default_lmv = NULL;
1514                 up_write(&lli->lli_lsm_sem);
1515         }
1516 }
1517
1518 static int ll_update_lsm_md(struct inode *inode, struct lustre_md *md)
1519 {
1520         struct ll_inode_info *lli = ll_i2info(inode);
1521         struct lmv_stripe_md *lsm = md->lmv;
1522         struct cl_attr  *attr;
1523         int rc = 0;
1524
1525         ENTRY;
1526
1527         LASSERT(S_ISDIR(inode->i_mode));
1528         CDEBUG(D_INODE, "update lsm %p of "DFID"\n", lli->lli_lsm_md,
1529                PFID(ll_inode2fid(inode)));
1530
1531         /* update default LMV */
1532         if (md->default_lmv)
1533                 ll_update_default_lsm_md(inode, md);
1534
1535         /*
1536          * no striped information from request, lustre_md from req does not
1537          * include stripeEA, see ll_md_setattr()
1538          */
1539         if (!lsm)
1540                 RETURN(0);
1541
1542         /*
1543          * normally dir layout doesn't change, only take read lock to check
1544          * that to avoid blocking other MD operations.
1545          */
1546         down_read(&lli->lli_lsm_sem);
1547
1548         /* some current lookup initialized lsm, and unchanged */
1549         if (lli->lli_lsm_md && lsm_md_eq(lli->lli_lsm_md, lsm))
1550                 GOTO(unlock, rc = 0);
1551
1552         /* if dir layout doesn't match, check whether version is increased,
1553          * which means layout is changed, this happens in dir split/merge and
1554          * lfsck.
1555          *
1556          * foreign LMV should not change.
1557          */
1558         if (lli->lli_lsm_md && lmv_dir_striped(lli->lli_lsm_md) &&
1559             lsm->lsm_md_layout_version <=
1560             lli->lli_lsm_md->lsm_md_layout_version) {
1561                 CERROR("%s: "DFID" dir layout mismatch:\n",
1562                        ll_i2sbi(inode)->ll_fsname, PFID(&lli->lli_fid));
1563                 lsm_md_dump(D_ERROR, lli->lli_lsm_md);
1564                 lsm_md_dump(D_ERROR, lsm);
1565                 GOTO(unlock, rc = -EINVAL);
1566         }
1567
1568         up_read(&lli->lli_lsm_sem);
1569         down_write(&lli->lli_lsm_sem);
1570         /* clear existing lsm */
1571         if (lli->lli_lsm_md) {
1572                 lmv_free_memmd(lli->lli_lsm_md);
1573                 lli->lli_lsm_md = NULL;
1574         }
1575
1576         rc = ll_init_lsm_md(inode, md);
1577         up_write(&lli->lli_lsm_sem);
1578
1579         if (rc)
1580                 RETURN(rc);
1581
1582         /* set md->lmv to NULL, so the following free lustre_md will not free
1583          * this lsm.
1584          */
1585         md->lmv = NULL;
1586
1587         /* md_merge_attr() may take long, since lsm is already set, switch to
1588          * read lock.
1589          */
1590         down_read(&lli->lli_lsm_sem);
1591
1592         if (!lmv_dir_striped(lli->lli_lsm_md))
1593                 GOTO(unlock, rc = 0);
1594
1595         OBD_ALLOC_PTR(attr);
1596         if (!attr)
1597                 GOTO(unlock, rc = -ENOMEM);
1598
1599         /* validate the lsm */
1600         rc = md_merge_attr(ll_i2mdexp(inode), lli->lli_lsm_md, attr,
1601                            ll_md_blocking_ast);
1602         if (!rc) {
1603                 if (md->body->mbo_valid & OBD_MD_FLNLINK)
1604                         md->body->mbo_nlink = attr->cat_nlink;
1605                 if (md->body->mbo_valid & OBD_MD_FLSIZE)
1606                         md->body->mbo_size = attr->cat_size;
1607                 if (md->body->mbo_valid & OBD_MD_FLATIME)
1608                         md->body->mbo_atime = attr->cat_atime;
1609                 if (md->body->mbo_valid & OBD_MD_FLCTIME)
1610                         md->body->mbo_ctime = attr->cat_ctime;
1611                 if (md->body->mbo_valid & OBD_MD_FLMTIME)
1612                         md->body->mbo_mtime = attr->cat_mtime;
1613         }
1614
1615         OBD_FREE_PTR(attr);
1616         GOTO(unlock, rc);
1617 unlock:
1618         up_read(&lli->lli_lsm_sem);
1619
1620         return rc;
1621 }
1622
1623 void ll_clear_inode(struct inode *inode)
1624 {
1625         struct ll_inode_info *lli = ll_i2info(inode);
1626         struct ll_sb_info *sbi = ll_i2sbi(inode);
1627
1628         ENTRY;
1629
1630         CDEBUG(D_VFSTRACE, "VFS Op:inode="DFID"(%p)\n",
1631                PFID(ll_inode2fid(inode)), inode);
1632
1633         if (S_ISDIR(inode->i_mode)) {
1634                 /* these should have been cleared in ll_file_release */
1635                 LASSERT(lli->lli_opendir_key == NULL);
1636                 LASSERT(lli->lli_sai == NULL);
1637                 LASSERT(lli->lli_opendir_pid == 0);
1638         } else {
1639                 pcc_inode_free(inode);
1640         }
1641
1642         md_null_inode(sbi->ll_md_exp, ll_inode2fid(inode));
1643
1644         LASSERT(!lli->lli_open_fd_write_count);
1645         LASSERT(!lli->lli_open_fd_read_count);
1646         LASSERT(!lli->lli_open_fd_exec_count);
1647
1648         if (lli->lli_mds_write_och)
1649                 ll_md_real_close(inode, FMODE_WRITE);
1650         if (lli->lli_mds_exec_och)
1651                 ll_md_real_close(inode, FMODE_EXEC);
1652         if (lli->lli_mds_read_och)
1653                 ll_md_real_close(inode, FMODE_READ);
1654
1655         if (S_ISLNK(inode->i_mode) && lli->lli_symlink_name) {
1656                 OBD_FREE(lli->lli_symlink_name,
1657                          strlen(lli->lli_symlink_name) + 1);
1658                 lli->lli_symlink_name = NULL;
1659         }
1660
1661         ll_xattr_cache_destroy(inode);
1662
1663         forget_all_cached_acls(inode);
1664         lli_clear_acl(lli);
1665         lli->lli_inode_magic = LLI_INODE_DEAD;
1666
1667         if (S_ISDIR(inode->i_mode))
1668                 ll_dir_clear_lsm_md(inode);
1669         else if (S_ISREG(inode->i_mode) && !is_bad_inode(inode))
1670                 LASSERT(list_empty(&lli->lli_agl_list));
1671
1672         /*
1673          * XXX This has to be done before lsm is freed below, because
1674          * cl_object still uses inode lsm.
1675          */
1676         cl_inode_fini(inode);
1677
1678         llcrypt_put_encryption_info(inode);
1679
1680         EXIT;
1681 }
1682
1683 static int ll_md_setattr(struct dentry *dentry, struct md_op_data *op_data)
1684 {
1685         struct lustre_md md;
1686         struct inode *inode = dentry->d_inode;
1687         struct ll_sb_info *sbi = ll_i2sbi(inode);
1688         struct ptlrpc_request *request = NULL;
1689         int rc, ia_valid;
1690         ENTRY;
1691
1692         op_data = ll_prep_md_op_data(op_data, inode, NULL, NULL, 0, 0,
1693                                      LUSTRE_OPC_ANY, NULL);
1694         if (IS_ERR(op_data))
1695                 RETURN(PTR_ERR(op_data));
1696
1697         rc = md_setattr(sbi->ll_md_exp, op_data, NULL, 0, &request);
1698         if (rc) {
1699                 ptlrpc_req_finished(request);
1700                 if (rc == -ENOENT) {
1701                         clear_nlink(inode);
1702                         /* Unlinked special device node? Or just a race?
1703                          * Pretend we done everything. */
1704                         if (!S_ISREG(inode->i_mode) &&
1705                             !S_ISDIR(inode->i_mode)) {
1706                                 ia_valid = op_data->op_attr.ia_valid;
1707                                 op_data->op_attr.ia_valid &= ~TIMES_SET_FLAGS;
1708                                 rc = simple_setattr(dentry, &op_data->op_attr);
1709                                 op_data->op_attr.ia_valid = ia_valid;
1710                         }
1711                 } else if (rc != -EPERM && rc != -EACCES && rc != -ETXTBSY) {
1712                         CERROR("md_setattr fails: rc = %d\n", rc);
1713                 }
1714                 RETURN(rc);
1715         }
1716
1717         rc = md_get_lustre_md(sbi->ll_md_exp, request, sbi->ll_dt_exp,
1718                               sbi->ll_md_exp, &md);
1719         if (rc) {
1720                 ptlrpc_req_finished(request);
1721                 RETURN(rc);
1722         }
1723
1724         ia_valid = op_data->op_attr.ia_valid;
1725         /* inode size will be in ll_setattr_ost, can't do it now since dirty
1726          * cache is not cleared yet. */
1727         op_data->op_attr.ia_valid &= ~(TIMES_SET_FLAGS | ATTR_SIZE);
1728         if (S_ISREG(inode->i_mode))
1729                 inode_lock(inode);
1730         rc = simple_setattr(dentry, &op_data->op_attr);
1731         if (S_ISREG(inode->i_mode))
1732                 inode_unlock(inode);
1733         op_data->op_attr.ia_valid = ia_valid;
1734
1735         rc = ll_update_inode(inode, &md);
1736         ptlrpc_req_finished(request);
1737
1738         RETURN(rc);
1739 }
1740
1741 /**
1742  * Zero portion of page that is part of @inode.
1743  * This implies, if necessary:
1744  * - taking cl_lock on range corresponding to concerned page
1745  * - grabbing vm page
1746  * - associating cl_page
1747  * - proceeding to clio read
1748  * - zeroing range in page
1749  * - proceeding to cl_page flush
1750  * - releasing cl_lock
1751  *
1752  * \param[in] inode     inode
1753  * \param[in] index     page index
1754  * \param[in] offset    offset in page to start zero from
1755  * \param[in] len       len to zero
1756  *
1757  * \retval 0            on success
1758  * \retval negative     errno on failure
1759  */
1760 int ll_io_zero_page(struct inode *inode, pgoff_t index, pgoff_t offset,
1761                     unsigned len)
1762 {
1763         struct ll_inode_info *lli = ll_i2info(inode);
1764         struct cl_object *clob = lli->lli_clob;
1765         __u16 refcheck;
1766         struct lu_env *env = NULL;
1767         struct cl_io *io = NULL;
1768         struct cl_page *clpage = NULL;
1769         struct page *vmpage = NULL;
1770         unsigned from = index << PAGE_SHIFT;
1771         struct cl_lock *lock = NULL;
1772         struct cl_lock_descr *descr = NULL;
1773         struct cl_2queue *queue = NULL;
1774         struct cl_sync_io *anchor = NULL;
1775         bool holdinglock = false;
1776         bool lockedbymyself = true;
1777         int rc;
1778
1779         ENTRY;
1780
1781         env = cl_env_get(&refcheck);
1782         if (IS_ERR(env))
1783                 RETURN(PTR_ERR(env));
1784
1785         io = vvp_env_thread_io(env);
1786         io->ci_obj = clob;
1787         rc = cl_io_rw_init(env, io, CIT_WRITE, from, PAGE_SIZE);
1788         if (rc)
1789                 GOTO(putenv, rc);
1790
1791         lock = vvp_env_lock(env);
1792         descr = &lock->cll_descr;
1793         descr->cld_obj   = io->ci_obj;
1794         descr->cld_start = cl_index(io->ci_obj, from);
1795         descr->cld_end   = cl_index(io->ci_obj, from + PAGE_SIZE - 1);
1796         descr->cld_mode  = CLM_WRITE;
1797         descr->cld_enq_flags = CEF_MUST | CEF_NONBLOCK;
1798
1799         /* request lock for page */
1800         rc = cl_lock_request(env, io, lock);
1801         /* -ECANCELED indicates a matching lock with a different extent
1802          * was already present, and -EEXIST indicates a matching lock
1803          * on exactly the same extent was already present.
1804          * In both cases it means we are covered.
1805          */
1806         if (rc == -ECANCELED || rc == -EEXIST)
1807                 rc = 0;
1808         else if (rc < 0)
1809                 GOTO(iofini, rc);
1810         else
1811                 holdinglock = true;
1812
1813         /* grab page */
1814         vmpage = grab_cache_page_nowait(inode->i_mapping, index);
1815         if (vmpage == NULL)
1816                 GOTO(rellock, rc = -EOPNOTSUPP);
1817
1818         if (!PageDirty(vmpage)) {
1819                 /* associate cl_page */
1820                 clpage = cl_page_find(env, clob, vmpage->index,
1821                                       vmpage, CPT_CACHEABLE);
1822                 if (IS_ERR(clpage))
1823                         GOTO(pagefini, rc = PTR_ERR(clpage));
1824
1825                 cl_page_assume(env, io, clpage);
1826         }
1827
1828         if (!PageUptodate(vmpage) && !PageDirty(vmpage) &&
1829             !PageWriteback(vmpage)) {
1830                 /* read page */
1831                 /* set PagePrivate2 to detect special case of empty page
1832                  * in osc_brw_fini_request()
1833                  */
1834                 SetPagePrivate2(vmpage);
1835                 rc = ll_io_read_page(env, io, clpage, NULL);
1836                 if (!PagePrivate2(vmpage))
1837                         /* PagePrivate2 was cleared in osc_brw_fini_request()
1838                          * meaning we read an empty page. In this case, in order
1839                          * to avoid allocating unnecessary block in truncated
1840                          * file, we must not zero and write as below. Subsequent
1841                          * server-side truncate will handle things correctly.
1842                          */
1843                         GOTO(clpfini, rc = 0);
1844                 ClearPagePrivate2(vmpage);
1845                 if (rc)
1846                         GOTO(clpfini, rc);
1847                 lockedbymyself = trylock_page(vmpage);
1848                 cl_page_assume(env, io, clpage);
1849         }
1850
1851         /* zero range in page */
1852         zero_user(vmpage, offset, len);
1853
1854         if (holdinglock && clpage) {
1855                 /* explicitly write newly modified page */
1856                 queue = &io->ci_queue;
1857                 cl_2queue_init(queue);
1858                 anchor = &vvp_env_info(env)->vti_anchor;
1859                 cl_sync_io_init(anchor, 1);
1860                 clpage->cp_sync_io = anchor;
1861                 cl_2queue_add(queue, clpage);
1862                 rc = cl_io_submit_rw(env, io, CRT_WRITE, queue);
1863                 if (rc)
1864                         GOTO(queuefini1, rc);
1865                 rc = cl_sync_io_wait(env, anchor, 0);
1866                 if (rc)
1867                         GOTO(queuefini2, rc);
1868                 cl_page_assume(env, io, clpage);
1869
1870 queuefini2:
1871                 cl_2queue_discard(env, io, queue);
1872 queuefini1:
1873                 cl_2queue_disown(env, io, queue);
1874                 cl_2queue_fini(env, queue);
1875         }
1876
1877 clpfini:
1878         if (clpage)
1879                 cl_page_put(env, clpage);
1880 pagefini:
1881         if (lockedbymyself) {
1882                 unlock_page(vmpage);
1883                 put_page(vmpage);
1884         }
1885 rellock:
1886         if (holdinglock)
1887                 cl_lock_release(env, lock);
1888 iofini:
1889         cl_io_fini(env, io);
1890 putenv:
1891         if (env)
1892                 cl_env_put(env, &refcheck);
1893
1894         RETURN(rc);
1895 }
1896
1897 /* If this inode has objects allocated to it (lsm != NULL), then the OST
1898  * object(s) determine the file size and mtime.  Otherwise, the MDS will
1899  * keep these values until such a time that objects are allocated for it.
1900  * We do the MDS operations first, as it is checking permissions for us.
1901  * We don't to the MDS RPC if there is nothing that we want to store there,
1902  * otherwise there is no harm in updating mtime/atime on the MDS if we are
1903  * going to do an RPC anyways.
1904  *
1905  * If we are doing a truncate, we will send the mtime and ctime updates
1906  * to the OST with the punch RPC, otherwise we do an explicit setattr RPC.
1907  * I don't believe it is possible to get e.g. ATTR_MTIME_SET and ATTR_SIZE
1908  * at the same time.
1909  *
1910  * In case of HSMimport, we only set attr on MDS.
1911  */
1912 int ll_setattr_raw(struct dentry *dentry, struct iattr *attr,
1913                    enum op_xvalid xvalid, bool hsm_import)
1914 {
1915         struct inode *inode = dentry->d_inode;
1916         struct ll_inode_info *lli = ll_i2info(inode);
1917         struct md_op_data *op_data = NULL;
1918         ktime_t kstart = ktime_get();
1919         int rc = 0;
1920
1921         ENTRY;
1922
1923         CDEBUG(D_VFSTRACE, "%s: setattr inode "DFID"(%p) from %llu to %llu, "
1924                "valid %x, hsm_import %d\n",
1925                ll_i2sbi(inode)->ll_fsname, PFID(&lli->lli_fid),
1926                inode, i_size_read(inode), attr->ia_size, attr->ia_valid,
1927                hsm_import);
1928
1929         if (attr->ia_valid & ATTR_SIZE) {
1930                 /* Check new size against VFS/VM file size limit and rlimit */
1931                 rc = inode_newsize_ok(inode, attr->ia_size);
1932                 if (rc)
1933                         RETURN(rc);
1934
1935                 /* The maximum Lustre file size is variable, based on the
1936                  * OST maximum object size and number of stripes.  This
1937                  * needs another check in addition to the VFS check above. */
1938                 if (attr->ia_size > ll_file_maxbytes(inode)) {
1939                         CDEBUG(D_INODE,"file "DFID" too large %llu > %llu\n",
1940                                PFID(&lli->lli_fid), attr->ia_size,
1941                                ll_file_maxbytes(inode));
1942                         RETURN(-EFBIG);
1943                 }
1944
1945                 attr->ia_valid |= ATTR_MTIME | ATTR_CTIME;
1946         }
1947
1948         /* POSIX: check before ATTR_*TIME_SET set (from inode_change_ok) */
1949         if (attr->ia_valid & TIMES_SET_FLAGS) {
1950                 if ((!uid_eq(current_fsuid(), inode->i_uid)) &&
1951                     !cfs_capable(CFS_CAP_FOWNER))
1952                         RETURN(-EPERM);
1953         }
1954
1955         /* We mark all of the fields "set" so MDS/OST does not re-set them */
1956         if (!(xvalid & OP_XVALID_CTIME_SET) &&
1957              (attr->ia_valid & ATTR_CTIME)) {
1958                 attr->ia_ctime = current_time(inode);
1959                 xvalid |= OP_XVALID_CTIME_SET;
1960         }
1961         if (!(attr->ia_valid & ATTR_ATIME_SET) &&
1962             (attr->ia_valid & ATTR_ATIME)) {
1963                 attr->ia_atime = current_time(inode);
1964                 attr->ia_valid |= ATTR_ATIME_SET;
1965         }
1966         if (!(attr->ia_valid & ATTR_MTIME_SET) &&
1967             (attr->ia_valid & ATTR_MTIME)) {
1968                 attr->ia_mtime = current_time(inode);
1969                 attr->ia_valid |= ATTR_MTIME_SET;
1970         }
1971
1972         if (attr->ia_valid & (ATTR_MTIME | ATTR_CTIME))
1973                 CDEBUG(D_INODE, "setting mtime %lld, ctime %lld, now = %lld\n",
1974                        (s64)attr->ia_mtime.tv_sec, (s64)attr->ia_ctime.tv_sec,
1975                        ktime_get_real_seconds());
1976
1977         if (S_ISREG(inode->i_mode))
1978                 inode_unlock(inode);
1979
1980         /* We always do an MDS RPC, even if we're only changing the size;
1981          * only the MDS knows whether truncate() should fail with -ETXTBUSY */
1982
1983         OBD_ALLOC_PTR(op_data);
1984         if (op_data == NULL)
1985                 GOTO(out, rc = -ENOMEM);
1986
1987         if (!hsm_import && attr->ia_valid & ATTR_SIZE) {
1988                 /* If we are changing file size, file content is
1989                  * modified, flag it.
1990                  */
1991                 xvalid |= OP_XVALID_OWNEROVERRIDE;
1992                 op_data->op_bias |= MDS_DATA_MODIFIED;
1993                 ll_file_clear_flag(lli, LLIF_DATA_MODIFIED);
1994         }
1995
1996         if (attr->ia_valid & ATTR_FILE) {
1997                 struct ll_file_data *fd = attr->ia_file->private_data;
1998
1999                 if (fd->fd_lease_och)
2000                         op_data->op_bias |= MDS_TRUNC_KEEP_LEASE;
2001         }
2002
2003         op_data->op_attr = *attr;
2004         op_data->op_xvalid = xvalid;
2005
2006         rc = ll_md_setattr(dentry, op_data);
2007         if (rc)
2008                 GOTO(out, rc);
2009
2010         if (!S_ISREG(inode->i_mode) || hsm_import)
2011                 GOTO(out, rc = 0);
2012
2013         if (attr->ia_valid & (ATTR_SIZE | ATTR_ATIME | ATTR_ATIME_SET |
2014                               ATTR_MTIME | ATTR_MTIME_SET | ATTR_CTIME) ||
2015             xvalid & OP_XVALID_CTIME_SET) {
2016                 bool cached = false;
2017
2018                 rc = pcc_inode_setattr(inode, attr, &cached);
2019                 if (cached) {
2020                         if (rc) {
2021                                 CERROR("%s: PCC inode "DFID" setattr failed: "
2022                                        "rc = %d\n",
2023                                        ll_i2sbi(inode)->ll_fsname,
2024                                        PFID(&lli->lli_fid), rc);
2025                                 GOTO(out, rc);
2026                         }
2027                 } else {
2028                         unsigned int flags = 0;
2029
2030                         /* For truncate and utimes sending attributes to OSTs,
2031                          * setting mtime/atime to the past will be performed
2032                          * under PW [0:EOF] extent lock (new_size:EOF for
2033                          * truncate). It may seem excessive to send mtime/atime
2034                          * updates to OSTs when not setting times to past, but
2035                          * it is necessary due to possible time
2036                          * de-synchronization between MDT inode and OST objects
2037                          */
2038                         if (S_ISREG(inode->i_mode) && IS_ENCRYPTED(inode) &&
2039                             attr->ia_valid & ATTR_SIZE) {
2040                                 xvalid |= OP_XVALID_FLAGS;
2041                                 flags = LUSTRE_ENCRYPT_FL;
2042                                 /* Call to ll_io_zero_page is not necessary if
2043                                  * truncating on PAGE_SIZE boundary, because
2044                                  * whole pages will be wiped.
2045                                  * In case of Direct IO, all we need is to set
2046                                  * new size.
2047                                  */
2048                                 if (attr->ia_size & ~PAGE_MASK &&
2049                                     !(attr->ia_valid & ATTR_FILE &&
2050                                       attr->ia_file->f_flags & O_DIRECT)) {
2051                                         pgoff_t offset =
2052                                                 attr->ia_size & (PAGE_SIZE - 1);
2053
2054                                         rc = ll_io_zero_page(inode,
2055                                                     attr->ia_size >> PAGE_SHIFT,
2056                                                     offset, PAGE_SIZE - offset);
2057                                         if (rc)
2058                                                 GOTO(out, rc);
2059                                 }
2060                         }
2061                         rc = cl_setattr_ost(lli->lli_clob, attr, xvalid, flags);
2062                 }
2063         }
2064
2065         /* If the file was restored, it needs to set dirty flag.
2066          *
2067          * We've already sent MDS_DATA_MODIFIED flag in
2068          * ll_md_setattr() for truncate. However, the MDT refuses to
2069          * set the HS_DIRTY flag on released files, so we have to set
2070          * it again if the file has been restored. Please check how
2071          * LLIF_DATA_MODIFIED is set in vvp_io_setattr_fini().
2072          *
2073          * Please notice that if the file is not released, the previous
2074          * MDS_DATA_MODIFIED has taken effect and usually
2075          * LLIF_DATA_MODIFIED is not set(see vvp_io_setattr_fini()).
2076          * This way we can save an RPC for common open + trunc
2077          * operation. */
2078         if (ll_file_test_and_clear_flag(lli, LLIF_DATA_MODIFIED)) {
2079                 struct hsm_state_set hss = {
2080                         .hss_valid = HSS_SETMASK,
2081                         .hss_setmask = HS_DIRTY,
2082                 };
2083                 int rc2;
2084
2085                 rc2 = ll_hsm_state_set(inode, &hss);
2086                 /* truncate and write can happen at the same time, so that
2087                  * the file can be set modified even though the file is not
2088                  * restored from released state, and ll_hsm_state_set() is
2089                  * not applicable for the file, and rc2 < 0 is normal in this
2090                  * case. */
2091                 if (rc2 < 0)
2092                         CDEBUG(D_INFO, DFID "HSM set dirty failed: rc2 = %d\n",
2093                                PFID(ll_inode2fid(inode)), rc2);
2094         }
2095
2096         EXIT;
2097 out:
2098         if (op_data != NULL)
2099                 ll_finish_md_op_data(op_data);
2100
2101         if (S_ISREG(inode->i_mode)) {
2102                 inode_lock(inode);
2103                 if ((attr->ia_valid & ATTR_SIZE) && !hsm_import)
2104                         inode_dio_wait(inode);
2105                 /* Once we've got the i_mutex, it's safe to set the S_NOSEC
2106                  * flag.  ll_update_inode (called from ll_md_setattr), clears
2107                  * inode flags, so there is a gap where S_NOSEC is not set.
2108                  * This can cause a writer to take the i_mutex unnecessarily,
2109                  * but this is safe to do and should be rare. */
2110                 inode_has_no_xattr(inode);
2111         }
2112
2113         if (!rc)
2114                 ll_stats_ops_tally(ll_i2sbi(inode), attr->ia_valid & ATTR_SIZE ?
2115                                         LPROC_LL_TRUNC : LPROC_LL_SETATTR,
2116                                    ktime_us_delta(ktime_get(), kstart));
2117
2118         return rc;
2119 }
2120
2121 int ll_setattr(struct dentry *de, struct iattr *attr)
2122 {
2123         int mode = de->d_inode->i_mode;
2124         enum op_xvalid xvalid = 0;
2125         int rc;
2126
2127         rc = llcrypt_prepare_setattr(de, attr);
2128         if (rc)
2129                 return rc;
2130
2131         if ((attr->ia_valid & (ATTR_CTIME|ATTR_SIZE|ATTR_MODE)) ==
2132                               (ATTR_CTIME|ATTR_SIZE|ATTR_MODE))
2133                 xvalid |= OP_XVALID_OWNEROVERRIDE;
2134
2135         if (((attr->ia_valid & (ATTR_MODE|ATTR_FORCE|ATTR_SIZE)) ==
2136                                (ATTR_SIZE|ATTR_MODE)) &&
2137             (((mode & S_ISUID) && !(attr->ia_mode & S_ISUID)) ||
2138              (((mode & (S_ISGID|S_IXGRP)) == (S_ISGID|S_IXGRP)) &&
2139               !(attr->ia_mode & S_ISGID))))
2140                 attr->ia_valid |= ATTR_FORCE;
2141
2142         if ((attr->ia_valid & ATTR_MODE) &&
2143             (mode & S_ISUID) &&
2144             !(attr->ia_mode & S_ISUID) &&
2145             !(attr->ia_valid & ATTR_KILL_SUID))
2146                 attr->ia_valid |= ATTR_KILL_SUID;
2147
2148         if ((attr->ia_valid & ATTR_MODE) &&
2149             ((mode & (S_ISGID|S_IXGRP)) == (S_ISGID|S_IXGRP)) &&
2150             !(attr->ia_mode & S_ISGID) &&
2151             !(attr->ia_valid & ATTR_KILL_SGID))
2152                 attr->ia_valid |= ATTR_KILL_SGID;
2153
2154         return ll_setattr_raw(de, attr, xvalid, false);
2155 }
2156
2157 int ll_statfs_internal(struct ll_sb_info *sbi, struct obd_statfs *osfs,
2158                        u32 flags)
2159 {
2160         struct obd_statfs obd_osfs = { 0 };
2161         time64_t max_age;
2162         int rc;
2163
2164         ENTRY;
2165         max_age = ktime_get_seconds() - sbi->ll_statfs_max_age;
2166
2167         if (sbi->ll_flags & LL_SBI_LAZYSTATFS)
2168                 flags |= OBD_STATFS_NODELAY;
2169
2170         rc = obd_statfs(NULL, sbi->ll_md_exp, osfs, max_age, flags);
2171         if (rc)
2172                 RETURN(rc);
2173
2174         osfs->os_type = LL_SUPER_MAGIC;
2175
2176         CDEBUG(D_SUPER, "MDC blocks %llu/%llu objects %llu/%llu\n",
2177               osfs->os_bavail, osfs->os_blocks, osfs->os_ffree, osfs->os_files);
2178
2179         if (osfs->os_state & OS_STATFS_SUM)
2180                 GOTO(out, rc);
2181
2182         rc = obd_statfs(NULL, sbi->ll_dt_exp, &obd_osfs, max_age, flags);
2183         if (rc) /* Possibly a filesystem with no OSTs.  Report MDT totals. */
2184                 GOTO(out, rc = 0);
2185
2186         CDEBUG(D_SUPER, "OSC blocks %llu/%llu objects %llu/%llu\n",
2187                obd_osfs.os_bavail, obd_osfs.os_blocks, obd_osfs.os_ffree,
2188                obd_osfs.os_files);
2189
2190         osfs->os_bsize = obd_osfs.os_bsize;
2191         osfs->os_blocks = obd_osfs.os_blocks;
2192         osfs->os_bfree = obd_osfs.os_bfree;
2193         osfs->os_bavail = obd_osfs.os_bavail;
2194
2195         /* If we have _some_ OSTs, but don't have as many free objects on the
2196          * OSTs as inodes on the MDTs, reduce the reported number of inodes
2197          * to compensate, so that the "inodes in use" number is correct.
2198          * This should be kept in sync with lod_statfs() behaviour.
2199          */
2200         if (obd_osfs.os_files && obd_osfs.os_ffree < osfs->os_ffree) {
2201                 osfs->os_files = (osfs->os_files - osfs->os_ffree) +
2202                                  obd_osfs.os_ffree;
2203                 osfs->os_ffree = obd_osfs.os_ffree;
2204         }
2205
2206 out:
2207         RETURN(rc);
2208 }
2209
2210 static int ll_statfs_project(struct inode *inode, struct kstatfs *sfs)
2211 {
2212         struct if_quotactl qctl = {
2213                 .qc_cmd = LUSTRE_Q_GETQUOTA,
2214                 .qc_type = PRJQUOTA,
2215                 .qc_valid = QC_GENERAL,
2216         };
2217         u64 limit, curblock;
2218         int ret;
2219
2220         qctl.qc_id = ll_i2info(inode)->lli_projid;
2221         ret = quotactl_ioctl(ll_i2sbi(inode), &qctl);
2222         if (ret) {
2223                 /* ignore errors if project ID does not have
2224                  * a quota limit or feature unsupported.
2225                  */
2226                 if (ret == -ESRCH || ret == -EOPNOTSUPP)
2227                         ret = 0;
2228                 return ret;
2229         }
2230
2231         limit = ((qctl.qc_dqblk.dqb_bsoftlimit ?
2232                  qctl.qc_dqblk.dqb_bsoftlimit :
2233                  qctl.qc_dqblk.dqb_bhardlimit) * 1024) / sfs->f_bsize;
2234         if (limit && sfs->f_blocks > limit) {
2235                 curblock = (qctl.qc_dqblk.dqb_curspace +
2236                                 sfs->f_bsize - 1) / sfs->f_bsize;
2237                 sfs->f_blocks = limit;
2238                 sfs->f_bfree = sfs->f_bavail =
2239                         (sfs->f_blocks > curblock) ?
2240                         (sfs->f_blocks - curblock) : 0;
2241         }
2242
2243         limit = qctl.qc_dqblk.dqb_isoftlimit ?
2244                 qctl.qc_dqblk.dqb_isoftlimit :
2245                 qctl.qc_dqblk.dqb_ihardlimit;
2246         if (limit && sfs->f_files > limit) {
2247                 sfs->f_files = limit;
2248                 sfs->f_ffree = (sfs->f_files >
2249                         qctl.qc_dqblk.dqb_curinodes) ?
2250                         (sfs->f_files - qctl.qc_dqblk.dqb_curinodes) : 0;
2251         }
2252
2253         return 0;
2254 }
2255
2256 int ll_statfs(struct dentry *de, struct kstatfs *sfs)
2257 {
2258         struct super_block *sb = de->d_sb;
2259         struct obd_statfs osfs;
2260         __u64 fsid = huge_encode_dev(sb->s_dev);
2261         ktime_t kstart = ktime_get();
2262         int rc;
2263
2264         CDEBUG(D_VFSTRACE, "VFS Op:sb=%s (%p)\n", sb->s_id, sb);
2265
2266         /* Some amount of caching on the client is allowed */
2267         rc = ll_statfs_internal(ll_s2sbi(sb), &osfs, OBD_STATFS_SUM);
2268         if (rc)
2269                 return rc;
2270
2271         statfs_unpack(sfs, &osfs);
2272
2273         /* We need to downshift for all 32-bit kernels, because we can't
2274          * tell if the kernel is being called via sys_statfs64() or not.
2275          * Stop before overflowing f_bsize - in which case it is better
2276          * to just risk EOVERFLOW if caller is using old sys_statfs(). */
2277         if (sizeof(long) < 8) {
2278                 while (osfs.os_blocks > ~0UL && sfs->f_bsize < 0x40000000) {
2279                         sfs->f_bsize <<= 1;
2280
2281                         osfs.os_blocks >>= 1;
2282                         osfs.os_bfree >>= 1;
2283                         osfs.os_bavail >>= 1;
2284                 }
2285         }
2286
2287         sfs->f_blocks = osfs.os_blocks;
2288         sfs->f_bfree = osfs.os_bfree;
2289         sfs->f_bavail = osfs.os_bavail;
2290         sfs->f_fsid.val[0] = (__u32)fsid;
2291         sfs->f_fsid.val[1] = (__u32)(fsid >> 32);
2292         if (ll_i2info(de->d_inode)->lli_projid)
2293                 return ll_statfs_project(de->d_inode, sfs);
2294
2295         ll_stats_ops_tally(ll_s2sbi(sb), LPROC_LL_STATFS,
2296                            ktime_us_delta(ktime_get(), kstart));
2297
2298         return 0;
2299 }
2300
2301 void ll_inode_size_lock(struct inode *inode)
2302 {
2303         struct ll_inode_info *lli;
2304
2305         LASSERT(!S_ISDIR(inode->i_mode));
2306
2307         lli = ll_i2info(inode);
2308         mutex_lock(&lli->lli_size_mutex);
2309 }
2310
2311 void ll_inode_size_unlock(struct inode *inode)
2312 {
2313         struct ll_inode_info *lli;
2314
2315         lli = ll_i2info(inode);
2316         mutex_unlock(&lli->lli_size_mutex);
2317 }
2318
2319 void ll_update_inode_flags(struct inode *inode, int ext_flags)
2320 {
2321         /* do not clear encryption flag */
2322         ext_flags |= ll_inode_to_ext_flags(inode->i_flags) & LUSTRE_ENCRYPT_FL;
2323         inode->i_flags = ll_ext_to_inode_flags(ext_flags);
2324         if (ext_flags & LUSTRE_PROJINHERIT_FL)
2325                 ll_file_set_flag(ll_i2info(inode), LLIF_PROJECT_INHERIT);
2326         else
2327                 ll_file_clear_flag(ll_i2info(inode), LLIF_PROJECT_INHERIT);
2328 }
2329
2330 int ll_update_inode(struct inode *inode, struct lustre_md *md)
2331 {
2332         struct ll_inode_info *lli = ll_i2info(inode);
2333         struct mdt_body *body = md->body;
2334         struct ll_sb_info *sbi = ll_i2sbi(inode);
2335         int rc = 0;
2336
2337         if (body->mbo_valid & OBD_MD_FLEASIZE) {
2338                 rc = cl_file_inode_init(inode, md);
2339                 if (rc)
2340                         return rc;
2341         }
2342
2343         if (S_ISDIR(inode->i_mode)) {
2344                 rc = ll_update_lsm_md(inode, md);
2345                 if (rc != 0)
2346                         return rc;
2347         }
2348
2349         if (body->mbo_valid & OBD_MD_FLACL)
2350                 lli_replace_acl(lli, md);
2351
2352         inode->i_ino = cl_fid_build_ino(&body->mbo_fid1,
2353                                         sbi->ll_flags & LL_SBI_32BIT_API);
2354         inode->i_generation = cl_fid_build_gen(&body->mbo_fid1);
2355
2356         if (body->mbo_valid & OBD_MD_FLATIME) {
2357                 if (body->mbo_atime > inode->i_atime.tv_sec)
2358                         inode->i_atime.tv_sec = body->mbo_atime;
2359                 lli->lli_atime = body->mbo_atime;
2360         }
2361
2362         if (body->mbo_valid & OBD_MD_FLMTIME) {
2363                 if (body->mbo_mtime > inode->i_mtime.tv_sec) {
2364                         CDEBUG(D_INODE,
2365                                "setting ino %lu mtime from %lld to %llu\n",
2366                                inode->i_ino, (s64)inode->i_mtime.tv_sec,
2367                                body->mbo_mtime);
2368                         inode->i_mtime.tv_sec = body->mbo_mtime;
2369                 }
2370                 lli->lli_mtime = body->mbo_mtime;
2371         }
2372
2373         if (body->mbo_valid & OBD_MD_FLCTIME) {
2374                 if (body->mbo_ctime > inode->i_ctime.tv_sec)
2375                         inode->i_ctime.tv_sec = body->mbo_ctime;
2376                 lli->lli_ctime = body->mbo_ctime;
2377         }
2378
2379         if (body->mbo_valid & OBD_MD_FLBTIME)
2380                 lli->lli_btime = body->mbo_btime;
2381
2382         /* Clear i_flags to remove S_NOSEC before permissions are updated */
2383         if (body->mbo_valid & OBD_MD_FLFLAGS)
2384                 ll_update_inode_flags(inode, body->mbo_flags);
2385         if (body->mbo_valid & OBD_MD_FLMODE)
2386                 inode->i_mode = (inode->i_mode & S_IFMT) |
2387                                 (body->mbo_mode & ~S_IFMT);
2388
2389         if (body->mbo_valid & OBD_MD_FLTYPE)
2390                 inode->i_mode = (inode->i_mode & ~S_IFMT) |
2391                                 (body->mbo_mode & S_IFMT);
2392
2393         LASSERT(inode->i_mode != 0);
2394         if (body->mbo_valid & OBD_MD_FLUID)
2395                 inode->i_uid = make_kuid(&init_user_ns, body->mbo_uid);
2396         if (body->mbo_valid & OBD_MD_FLGID)
2397                 inode->i_gid = make_kgid(&init_user_ns, body->mbo_gid);
2398         if (body->mbo_valid & OBD_MD_FLPROJID)
2399                 lli->lli_projid = body->mbo_projid;
2400         if (body->mbo_valid & OBD_MD_FLNLINK)
2401                 set_nlink(inode, body->mbo_nlink);
2402         if (body->mbo_valid & OBD_MD_FLRDEV)
2403                 inode->i_rdev = old_decode_dev(body->mbo_rdev);
2404
2405         if (body->mbo_valid & OBD_MD_FLID) {
2406                 /* FID shouldn't be changed! */
2407                 if (fid_is_sane(&lli->lli_fid)) {
2408                         LASSERTF(lu_fid_eq(&lli->lli_fid, &body->mbo_fid1),
2409                                  "Trying to change FID "DFID
2410                                  " to the "DFID", inode "DFID"(%p)\n",
2411                                  PFID(&lli->lli_fid), PFID(&body->mbo_fid1),
2412                                  PFID(ll_inode2fid(inode)), inode);
2413                 } else {
2414                         lli->lli_fid = body->mbo_fid1;
2415                 }
2416         }
2417
2418         LASSERT(fid_seq(&lli->lli_fid) != 0);
2419
2420         lli->lli_attr_valid = body->mbo_valid;
2421         if (body->mbo_valid & OBD_MD_FLSIZE) {
2422                 i_size_write(inode, body->mbo_size);
2423
2424                 CDEBUG(D_VFSTRACE, "inode="DFID", updating i_size %llu\n",
2425                        PFID(ll_inode2fid(inode)),
2426                        (unsigned long long)body->mbo_size);
2427
2428                 if (body->mbo_valid & OBD_MD_FLBLOCKS)
2429                         inode->i_blocks = body->mbo_blocks;
2430         } else {
2431                 if (body->mbo_valid & OBD_MD_FLLAZYSIZE)
2432                         lli->lli_lazysize = body->mbo_size;
2433                 if (body->mbo_valid & OBD_MD_FLLAZYBLOCKS)
2434                         lli->lli_lazyblocks = body->mbo_blocks;
2435         }
2436
2437         if (body->mbo_valid & OBD_MD_TSTATE) {
2438                 /* Set LLIF_FILE_RESTORING if restore ongoing and
2439                  * clear it when done to ensure to start again
2440                  * glimpsing updated attrs
2441                  */
2442                 if (body->mbo_t_state & MS_RESTORE)
2443                         ll_file_set_flag(lli, LLIF_FILE_RESTORING);
2444                 else
2445                         ll_file_clear_flag(lli, LLIF_FILE_RESTORING);
2446         }
2447
2448         return 0;
2449 }
2450
2451 int ll_read_inode2(struct inode *inode, void *opaque)
2452 {
2453         struct lustre_md *md = opaque;
2454         struct ll_inode_info *lli = ll_i2info(inode);
2455         int     rc;
2456         ENTRY;
2457
2458         CDEBUG(D_VFSTRACE, "VFS Op:inode="DFID"(%p)\n",
2459                PFID(&lli->lli_fid), inode);
2460
2461         /* Core attributes from the MDS first.  This is a new inode, and
2462          * the VFS doesn't zero times in the core inode so we have to do
2463          * it ourselves.  They will be overwritten by either MDS or OST
2464          * attributes - we just need to make sure they aren't newer.
2465          */
2466         inode->i_mtime.tv_sec = 0;
2467         inode->i_atime.tv_sec = 0;
2468         inode->i_ctime.tv_sec = 0;
2469         inode->i_rdev = 0;
2470         rc = ll_update_inode(inode, md);
2471         if (rc != 0)
2472                 RETURN(rc);
2473
2474         /* OIDEBUG(inode); */
2475
2476 #ifdef HAVE_BACKING_DEV_INFO
2477         /* initializing backing dev info. */
2478         inode->i_mapping->backing_dev_info = &s2lsi(inode->i_sb)->lsi_bdi;
2479 #endif
2480         if (S_ISREG(inode->i_mode)) {
2481                 struct ll_sb_info *sbi = ll_i2sbi(inode);
2482                 inode->i_op = &ll_file_inode_operations;
2483                 inode->i_fop = sbi->ll_fop;
2484                 inode->i_mapping->a_ops = (struct address_space_operations *)&ll_aops;
2485                 EXIT;
2486         } else if (S_ISDIR(inode->i_mode)) {
2487                 inode->i_op = &ll_dir_inode_operations;
2488                 inode->i_fop = &ll_dir_operations;
2489                 EXIT;
2490         } else if (S_ISLNK(inode->i_mode)) {
2491                 inode->i_op = &ll_fast_symlink_inode_operations;
2492                 EXIT;
2493         } else {
2494                 inode->i_op = &ll_special_inode_operations;
2495
2496                 init_special_inode(inode, inode->i_mode,
2497                                    inode->i_rdev);
2498
2499                 EXIT;
2500         }
2501
2502         return 0;
2503 }
2504
2505 void ll_delete_inode(struct inode *inode)
2506 {
2507         struct ll_inode_info *lli = ll_i2info(inode);
2508         struct address_space *mapping = &inode->i_data;
2509         unsigned long nrpages;
2510         unsigned long flags;
2511
2512         ENTRY;
2513
2514         if (S_ISREG(inode->i_mode) && lli->lli_clob != NULL) {
2515                 /* It is last chance to write out dirty pages,
2516                  * otherwise we may lose data while umount.
2517                  *
2518                  * If i_nlink is 0 then just discard data. This is safe because
2519                  * local inode gets i_nlink 0 from server only for the last
2520                  * unlink, so that file is not opened somewhere else
2521                  */
2522                 cl_sync_file_range(inode, 0, OBD_OBJECT_EOF, inode->i_nlink ?
2523                                    CL_FSYNC_LOCAL : CL_FSYNC_DISCARD, 1);
2524         }
2525         truncate_inode_pages_final(mapping);
2526
2527         /* Workaround for LU-118: Note nrpages may not be totally updated when
2528          * truncate_inode_pages() returns, as there can be a page in the process
2529          * of deletion (inside __delete_from_page_cache()) in the specified
2530          * range. Thus mapping->nrpages can be non-zero when this function
2531          * returns even after truncation of the whole mapping.  Only do this if
2532          * npages isn't already zero.
2533          */
2534         nrpages = mapping->nrpages;
2535         if (nrpages) {
2536                 ll_xa_lock_irqsave(&mapping->i_pages, flags);
2537                 nrpages = mapping->nrpages;
2538                 ll_xa_unlock_irqrestore(&mapping->i_pages, flags);
2539         } /* Workaround end */
2540
2541         LASSERTF(nrpages == 0, "%s: inode="DFID"(%p) nrpages=%lu, "
2542                  "see https://jira.whamcloud.com/browse/LU-118\n",
2543                  ll_i2sbi(inode)->ll_fsname,
2544                  PFID(ll_inode2fid(inode)), inode, nrpages);
2545
2546         ll_clear_inode(inode);
2547         clear_inode(inode);
2548
2549         EXIT;
2550 }
2551
2552 int ll_iocontrol(struct inode *inode, struct file *file,
2553                  unsigned int cmd, unsigned long arg)
2554 {
2555         struct ll_sb_info *sbi = ll_i2sbi(inode);
2556         struct ptlrpc_request *req = NULL;
2557         int rc, flags = 0;
2558         ENTRY;
2559
2560         switch (cmd) {
2561         case FS_IOC_GETFLAGS: {
2562                 struct mdt_body *body;
2563                 struct md_op_data *op_data;
2564
2565                 op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL,
2566                                              0, 0, LUSTRE_OPC_ANY,
2567                                              NULL);
2568                 if (IS_ERR(op_data))
2569                         RETURN(PTR_ERR(op_data));
2570
2571                 op_data->op_valid = OBD_MD_FLFLAGS;
2572                 rc = md_getattr(sbi->ll_md_exp, op_data, &req);
2573                 ll_finish_md_op_data(op_data);
2574                 if (rc) {
2575                         CERROR("%s: failure inode "DFID": rc = %d\n",
2576                                sbi->ll_md_exp->exp_obd->obd_name,
2577                                PFID(ll_inode2fid(inode)), rc);
2578                         RETURN(-abs(rc));
2579                 }
2580
2581                 body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
2582
2583                 flags = body->mbo_flags;
2584
2585                 ptlrpc_req_finished(req);
2586
2587                 RETURN(put_user(flags, (int __user *)arg));
2588         }
2589         case FS_IOC_SETFLAGS: {
2590                 struct iattr *attr;
2591                 struct md_op_data *op_data;
2592                 struct cl_object *obj;
2593                 struct fsxattr fa = { 0 };
2594
2595                 if (get_user(flags, (int __user *)arg))
2596                         RETURN(-EFAULT);
2597
2598                 fa.fsx_projid = ll_i2info(inode)->lli_projid;
2599                 if (flags & LUSTRE_PROJINHERIT_FL)
2600                         fa.fsx_xflags = FS_XFLAG_PROJINHERIT;
2601
2602                 rc = ll_ioctl_check_project(inode, &fa);
2603                 if (rc)
2604                         RETURN(rc);
2605
2606                 op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL, 0, 0,
2607                                              LUSTRE_OPC_ANY, NULL);
2608                 if (IS_ERR(op_data))
2609                         RETURN(PTR_ERR(op_data));
2610
2611                 op_data->op_attr_flags = flags;
2612                 op_data->op_xvalid |= OP_XVALID_FLAGS;
2613                 rc = md_setattr(sbi->ll_md_exp, op_data, NULL, 0, &req);
2614                 ll_finish_md_op_data(op_data);
2615                 ptlrpc_req_finished(req);
2616                 if (rc)
2617                         RETURN(rc);
2618
2619                 ll_update_inode_flags(inode, flags);
2620
2621                 obj = ll_i2info(inode)->lli_clob;
2622                 if (obj == NULL)
2623                         RETURN(0);
2624
2625                 OBD_ALLOC_PTR(attr);
2626                 if (attr == NULL)
2627                         RETURN(-ENOMEM);
2628
2629                 rc = cl_setattr_ost(obj, attr, OP_XVALID_FLAGS, flags);
2630
2631                 OBD_FREE_PTR(attr);
2632                 RETURN(rc);
2633         }
2634         default:
2635                 RETURN(-ENOSYS);
2636         }
2637
2638         RETURN(0);
2639 }
2640
2641 int ll_flush_ctx(struct inode *inode)
2642 {
2643         struct ll_sb_info  *sbi = ll_i2sbi(inode);
2644
2645         CDEBUG(D_SEC, "flush context for user %d\n",
2646                from_kuid(&init_user_ns, current_uid()));
2647
2648         obd_set_info_async(NULL, sbi->ll_md_exp,
2649                            sizeof(KEY_FLUSH_CTX), KEY_FLUSH_CTX,
2650                            0, NULL, NULL);
2651         obd_set_info_async(NULL, sbi->ll_dt_exp,
2652                            sizeof(KEY_FLUSH_CTX), KEY_FLUSH_CTX,
2653                            0, NULL, NULL);
2654         return 0;
2655 }
2656
2657 /* umount -f client means force down, don't save state */
2658 void ll_umount_begin(struct super_block *sb)
2659 {
2660         struct ll_sb_info *sbi = ll_s2sbi(sb);
2661         struct obd_device *obd;
2662         struct obd_ioctl_data *ioc_data;
2663         int cnt;
2664         ENTRY;
2665
2666         CDEBUG(D_VFSTRACE, "VFS Op: superblock %p count %d active %d\n", sb,
2667                sb->s_count, atomic_read(&sb->s_active));
2668
2669         obd = class_exp2obd(sbi->ll_md_exp);
2670         if (obd == NULL) {
2671                 CERROR("Invalid MDC connection handle %#llx\n",
2672                        sbi->ll_md_exp->exp_handle.h_cookie);
2673                 EXIT;
2674                 return;
2675         }
2676         obd->obd_force = 1;
2677
2678         obd = class_exp2obd(sbi->ll_dt_exp);
2679         if (obd == NULL) {
2680                 CERROR("Invalid LOV connection handle %#llx\n",
2681                        sbi->ll_dt_exp->exp_handle.h_cookie);
2682                 EXIT;
2683                 return;
2684         }
2685         obd->obd_force = 1;
2686
2687         OBD_ALLOC_PTR(ioc_data);
2688         if (ioc_data) {
2689                 obd_iocontrol(IOC_OSC_SET_ACTIVE, sbi->ll_md_exp,
2690                               sizeof *ioc_data, ioc_data, NULL);
2691
2692                 obd_iocontrol(IOC_OSC_SET_ACTIVE, sbi->ll_dt_exp,
2693                               sizeof *ioc_data, ioc_data, NULL);
2694
2695                 OBD_FREE_PTR(ioc_data);
2696         }
2697
2698         /* Really, we'd like to wait until there are no requests outstanding,
2699          * and then continue.  For now, we just periodically checking for vfs
2700          * to decrement mnt_cnt and hope to finish it within 10sec.
2701          */
2702         cnt = 10;
2703         while (cnt > 0 &&
2704                !may_umount(sbi->ll_mnt.mnt)) {
2705                 ssleep(1);
2706                 cnt -= 1;
2707         }
2708
2709         EXIT;
2710 }
2711
2712 int ll_remount_fs(struct super_block *sb, int *flags, char *data)
2713 {
2714         struct ll_sb_info *sbi = ll_s2sbi(sb);
2715         char *profilenm = get_profile_name(sb);
2716         int err;
2717         __u32 read_only;
2718
2719         if ((*flags & MS_RDONLY) != (sb->s_flags & SB_RDONLY)) {
2720                 read_only = *flags & MS_RDONLY;
2721                 err = obd_set_info_async(NULL, sbi->ll_md_exp,
2722                                          sizeof(KEY_READ_ONLY),
2723                                          KEY_READ_ONLY, sizeof(read_only),
2724                                          &read_only, NULL);
2725                 if (err) {
2726                         LCONSOLE_WARN("Failed to remount %s %s (%d)\n",
2727                                       profilenm, read_only ?
2728                                       "read-only" : "read-write", err);
2729                         return err;
2730                 }
2731
2732                 if (read_only)
2733                         sb->s_flags |= SB_RDONLY;
2734                 else
2735                         sb->s_flags &= ~SB_RDONLY;
2736
2737                 if (sbi->ll_flags & LL_SBI_VERBOSE)
2738                         LCONSOLE_WARN("Remounted %s %s\n", profilenm,
2739                                       read_only ?  "read-only" : "read-write");
2740         }
2741         return 0;
2742 }
2743
2744 /**
2745  * Cleanup the open handle that is cached on MDT-side.
2746  *
2747  * For open case, the client side open handling thread may hit error
2748  * after the MDT grant the open. Under such case, the client should
2749  * send close RPC to the MDT as cleanup; otherwise, the open handle
2750  * on the MDT will be leaked there until the client umount or evicted.
2751  *
2752  * In further, if someone unlinked the file, because the open handle
2753  * holds the reference on such file/object, then it will block the
2754  * subsequent threads that want to locate such object via FID.
2755  *
2756  * \param[in] sb        super block for this file-system
2757  * \param[in] open_req  pointer to the original open request
2758  */
2759 void ll_open_cleanup(struct super_block *sb, struct ptlrpc_request *open_req)
2760 {
2761         struct mdt_body                 *body;
2762         struct md_op_data               *op_data;
2763         struct ptlrpc_request           *close_req = NULL;
2764         struct obd_export               *exp       = ll_s2sbi(sb)->ll_md_exp;
2765         ENTRY;
2766
2767         body = req_capsule_server_get(&open_req->rq_pill, &RMF_MDT_BODY);
2768         OBD_ALLOC_PTR(op_data);
2769         if (op_data == NULL) {
2770                 CWARN("%s: cannot allocate op_data to release open handle for "
2771                       DFID"\n", ll_s2sbi(sb)->ll_fsname, PFID(&body->mbo_fid1));
2772
2773                 RETURN_EXIT;
2774         }
2775
2776         op_data->op_fid1 = body->mbo_fid1;
2777         op_data->op_open_handle = body->mbo_open_handle;
2778         op_data->op_mod_time = ktime_get_real_seconds();
2779         md_close(exp, op_data, NULL, &close_req);
2780         ptlrpc_req_finished(close_req);
2781         ll_finish_md_op_data(op_data);
2782
2783         EXIT;
2784 }
2785
2786 int ll_prep_inode(struct inode **inode, struct ptlrpc_request *req,
2787                   struct super_block *sb, struct lookup_intent *it)
2788 {
2789         struct ll_sb_info *sbi = NULL;
2790         struct lustre_md md = { NULL };
2791         bool default_lmv_deleted = false;
2792         int rc;
2793
2794         ENTRY;
2795
2796         LASSERT(*inode || sb);
2797         sbi = sb ? ll_s2sbi(sb) : ll_i2sbi(*inode);
2798         rc = md_get_lustre_md(sbi->ll_md_exp, req, sbi->ll_dt_exp,
2799                               sbi->ll_md_exp, &md);
2800         if (rc != 0)
2801                 GOTO(out, rc);
2802
2803         /*
2804          * clear default_lmv only if intent_getattr reply doesn't contain it.
2805          * but it needs to be done after iget, check this early because
2806          * ll_update_lsm_md() may change md.
2807          */
2808         if (it && (it->it_op & (IT_LOOKUP | IT_GETATTR)) &&
2809             S_ISDIR(md.body->mbo_mode) && !md.default_lmv)
2810                 default_lmv_deleted = true;
2811
2812         if (*inode) {
2813                 rc = ll_update_inode(*inode, &md);
2814                 if (rc != 0)
2815                         GOTO(out, rc);
2816         } else {
2817                 LASSERT(sb != NULL);
2818
2819                 /*
2820                  * At this point server returns to client's same fid as client
2821                  * generated for creating. So using ->fid1 is okay here.
2822                  */
2823                 if (!fid_is_sane(&md.body->mbo_fid1)) {
2824                         CERROR("%s: Fid is insane "DFID"\n",
2825                                 sbi->ll_fsname,
2826                                 PFID(&md.body->mbo_fid1));
2827                         GOTO(out, rc = -EINVAL);
2828                 }
2829
2830                 *inode = ll_iget(sb, cl_fid_build_ino(&md.body->mbo_fid1,
2831                                              sbi->ll_flags & LL_SBI_32BIT_API),
2832                                  &md);
2833                 if (IS_ERR(*inode)) {
2834                         lmd_clear_acl(&md);
2835                         rc = IS_ERR(*inode) ? PTR_ERR(*inode) : -ENOMEM;
2836                         *inode = NULL;
2837                         CERROR("new_inode -fatal: rc %d\n", rc);
2838                         GOTO(out, rc);
2839                 }
2840         }
2841
2842         /* Handling piggyback layout lock.
2843          * Layout lock can be piggybacked by getattr and open request.
2844          * The lsm can be applied to inode only if it comes with a layout lock
2845          * otherwise correct layout may be overwritten, for example:
2846          * 1. proc1: mdt returns a lsm but not granting layout
2847          * 2. layout was changed by another client
2848          * 3. proc2: refresh layout and layout lock granted
2849          * 4. proc1: to apply a stale layout */
2850         if (it != NULL && it->it_lock_mode != 0) {
2851                 struct lustre_handle lockh;
2852                 struct ldlm_lock *lock;
2853
2854                 lockh.cookie = it->it_lock_handle;
2855                 lock = ldlm_handle2lock(&lockh);
2856                 LASSERT(lock != NULL);
2857                 if (ldlm_has_layout(lock)) {
2858                         struct cl_object_conf conf;
2859
2860                         memset(&conf, 0, sizeof(conf));
2861                         conf.coc_opc = OBJECT_CONF_SET;
2862                         conf.coc_inode = *inode;
2863                         conf.coc_lock = lock;
2864                         conf.u.coc_layout = md.layout;
2865                         (void)ll_layout_conf(*inode, &conf);
2866                 }
2867                 LDLM_LOCK_PUT(lock);
2868         }
2869
2870         if (default_lmv_deleted)
2871                 ll_update_default_lsm_md(*inode, &md);
2872
2873         GOTO(out, rc = 0);
2874
2875 out:
2876         /* cleanup will be done if necessary */
2877         md_free_lustre_md(sbi->ll_md_exp, &md);
2878
2879         if (rc != 0 && it != NULL && it->it_op & IT_OPEN) {
2880                 ll_intent_drop_lock(it);
2881                 ll_open_cleanup(sb != NULL ? sb : (*inode)->i_sb, req);
2882         }
2883
2884         return rc;
2885 }
2886
2887 int ll_obd_statfs(struct inode *inode, void __user *arg)
2888 {
2889         struct ll_sb_info *sbi = NULL;
2890         struct obd_export *exp;
2891         char *buf = NULL;
2892         struct obd_ioctl_data *data = NULL;
2893         __u32 type;
2894         int len = 0, rc;
2895
2896         if (!inode || !(sbi = ll_i2sbi(inode)))
2897                 GOTO(out_statfs, rc = -EINVAL);
2898
2899         rc = obd_ioctl_getdata(&buf, &len, arg);
2900         if (rc)
2901                 GOTO(out_statfs, rc);
2902
2903         data = (void*)buf;
2904         if (!data->ioc_inlbuf1 || !data->ioc_inlbuf2 ||
2905             !data->ioc_pbuf1 || !data->ioc_pbuf2)
2906                 GOTO(out_statfs, rc = -EINVAL);
2907
2908         if (data->ioc_inllen1 != sizeof(__u32) ||
2909             data->ioc_inllen2 != sizeof(__u32) ||
2910             data->ioc_plen1 != sizeof(struct obd_statfs) ||
2911             data->ioc_plen2 != sizeof(struct obd_uuid))
2912                 GOTO(out_statfs, rc = -EINVAL);
2913
2914         memcpy(&type, data->ioc_inlbuf1, sizeof(__u32));
2915         if (type & LL_STATFS_LMV)
2916                 exp = sbi->ll_md_exp;
2917         else if (type & LL_STATFS_LOV)
2918                 exp = sbi->ll_dt_exp;
2919         else
2920                 GOTO(out_statfs, rc = -ENODEV);
2921
2922         rc = obd_iocontrol(IOC_OBD_STATFS, exp, len, buf, NULL);
2923         if (rc)
2924                 GOTO(out_statfs, rc);
2925 out_statfs:
2926         OBD_FREE_LARGE(buf, len);
2927         return rc;
2928 }
2929
2930 /*
2931  * this is normally called in ll_fini_md_op_data(), but sometimes it needs to
2932  * be called early to avoid deadlock.
2933  */
2934 void ll_unlock_md_op_lsm(struct md_op_data *op_data)
2935 {
2936         if (op_data->op_mea2_sem) {
2937                 up_read_non_owner(op_data->op_mea2_sem);
2938                 op_data->op_mea2_sem = NULL;
2939         }
2940
2941         if (op_data->op_mea1_sem) {
2942                 up_read_non_owner(op_data->op_mea1_sem);
2943                 op_data->op_mea1_sem = NULL;
2944         }
2945 }
2946
2947 /* this function prepares md_op_data hint for passing it down to MD stack. */
2948 struct md_op_data *ll_prep_md_op_data(struct md_op_data *op_data,
2949                                       struct inode *i1, struct inode *i2,
2950                                       const char *name, size_t namelen,
2951                                       __u32 mode, enum md_op_code opc,
2952                                       void *data)
2953 {
2954         LASSERT(i1 != NULL);
2955
2956         if (name == NULL) {
2957                 /* Do not reuse namelen for something else. */
2958                 if (namelen != 0)
2959                         return ERR_PTR(-EINVAL);
2960         } else {
2961                 if (namelen > ll_i2sbi(i1)->ll_namelen)
2962                         return ERR_PTR(-ENAMETOOLONG);
2963
2964                 /* "/" is not valid name, but it's allowed */
2965                 if (!lu_name_is_valid_2(name, namelen) &&
2966                     strncmp("/", name, namelen) != 0)
2967                         return ERR_PTR(-EINVAL);
2968         }
2969
2970         if (op_data == NULL)
2971                 OBD_ALLOC_PTR(op_data);
2972
2973         if (op_data == NULL)
2974                 return ERR_PTR(-ENOMEM);
2975
2976         ll_i2gids(op_data->op_suppgids, i1, i2);
2977         op_data->op_fid1 = *ll_inode2fid(i1);
2978         op_data->op_code = opc;
2979
2980         if (S_ISDIR(i1->i_mode)) {
2981                 down_read_non_owner(&ll_i2info(i1)->lli_lsm_sem);
2982                 op_data->op_mea1_sem = &ll_i2info(i1)->lli_lsm_sem;
2983                 op_data->op_mea1 = ll_i2info(i1)->lli_lsm_md;
2984                 op_data->op_default_mea1 = ll_i2info(i1)->lli_default_lsm_md;
2985         }
2986
2987         if (i2) {
2988                 op_data->op_fid2 = *ll_inode2fid(i2);
2989                 if (S_ISDIR(i2->i_mode)) {
2990                         if (i2 != i1) {
2991                                 /* i2 is typically a child of i1, and MUST be
2992                                  * further from the root to avoid deadlocks.
2993                                  */
2994                                 down_read_non_owner(&ll_i2info(i2)->lli_lsm_sem);
2995                                 op_data->op_mea2_sem =
2996                                                 &ll_i2info(i2)->lli_lsm_sem;
2997                         }
2998                         op_data->op_mea2 = ll_i2info(i2)->lli_lsm_md;
2999                 }
3000         } else {
3001                 fid_zero(&op_data->op_fid2);
3002         }
3003
3004         if (ll_i2sbi(i1)->ll_flags & LL_SBI_64BIT_HASH)
3005                 op_data->op_cli_flags |= CLI_HASH64;
3006
3007         if (ll_need_32bit_api(ll_i2sbi(i1)))
3008                 op_data->op_cli_flags |= CLI_API32;
3009
3010         op_data->op_name = name;
3011         op_data->op_namelen = namelen;
3012         op_data->op_mode = mode;
3013         op_data->op_mod_time = ktime_get_real_seconds();
3014         op_data->op_fsuid = from_kuid(&init_user_ns, current_fsuid());
3015         op_data->op_fsgid = from_kgid(&init_user_ns, current_fsgid());
3016         op_data->op_cap = cfs_curproc_cap_pack();
3017         op_data->op_mds = 0;
3018         if ((opc == LUSTRE_OPC_CREATE) && (name != NULL) &&
3019              filename_is_volatile(name, namelen, &op_data->op_mds)) {
3020                 op_data->op_bias |= MDS_CREATE_VOLATILE;
3021         }
3022         op_data->op_data = data;
3023
3024         return op_data;
3025 }
3026
3027 void ll_finish_md_op_data(struct md_op_data *op_data)
3028 {
3029         ll_unlock_md_op_lsm(op_data);
3030         security_release_secctx(op_data->op_file_secctx,
3031                                 op_data->op_file_secctx_size);
3032         llcrypt_free_ctx(op_data->op_file_encctx, op_data->op_file_encctx_size);
3033         OBD_FREE_PTR(op_data);
3034 }
3035
3036 int ll_show_options(struct seq_file *seq, struct dentry *dentry)
3037 {
3038         struct ll_sb_info *sbi;
3039
3040         LASSERT(seq && dentry);
3041         sbi = ll_s2sbi(dentry->d_sb);
3042
3043         if (sbi->ll_flags & LL_SBI_NOLCK)
3044                 seq_puts(seq, ",nolock");
3045
3046         /* "flock" is the default since 2.13, but it wasn't for many years,
3047          * so it is still useful to print this to show it is enabled.
3048          * Start to print "noflock" so it is now clear when flock is disabled.
3049          */
3050         if (sbi->ll_flags & LL_SBI_FLOCK)
3051                 seq_puts(seq, ",flock");
3052         else if (sbi->ll_flags & LL_SBI_LOCALFLOCK)
3053                 seq_puts(seq, ",localflock");
3054         else
3055                 seq_puts(seq, ",noflock");
3056
3057         if (sbi->ll_flags & LL_SBI_USER_XATTR)
3058                 seq_puts(seq, ",user_xattr");
3059
3060         if (sbi->ll_flags & LL_SBI_LAZYSTATFS)
3061                 seq_puts(seq, ",lazystatfs");
3062
3063         if (sbi->ll_flags & LL_SBI_USER_FID2PATH)
3064                 seq_puts(seq, ",user_fid2path");
3065
3066         if (sbi->ll_flags & LL_SBI_ALWAYS_PING)
3067                 seq_puts(seq, ",always_ping");
3068
3069         if (ll_sbi_has_test_dummy_encryption(sbi))
3070                 seq_puts(seq, ",test_dummy_encryption");
3071
3072         if (ll_sbi_has_encrypt(sbi))
3073                 seq_puts(seq, ",encrypt");
3074         else
3075                 seq_puts(seq, ",noencrypt");
3076
3077         RETURN(0);
3078 }
3079
3080 /**
3081  * Get obd name by cmd, and copy out to user space
3082  */
3083 int ll_get_obd_name(struct inode *inode, unsigned int cmd, unsigned long arg)
3084 {
3085         struct ll_sb_info *sbi = ll_i2sbi(inode);
3086         struct obd_device *obd;
3087         ENTRY;
3088
3089         if (cmd == OBD_IOC_GETDTNAME)
3090                 obd = class_exp2obd(sbi->ll_dt_exp);
3091         else if (cmd == OBD_IOC_GETMDNAME)
3092                 obd = class_exp2obd(sbi->ll_md_exp);
3093         else
3094                 RETURN(-EINVAL);
3095
3096         if (!obd)
3097                 RETURN(-ENOENT);
3098
3099         if (copy_to_user((void __user *)arg, obd->obd_name,
3100                          strlen(obd->obd_name) + 1))
3101                 RETURN(-EFAULT);
3102
3103         RETURN(0);
3104 }
3105
3106 static char* ll_d_path(struct dentry *dentry, char *buf, int bufsize)
3107 {
3108         char *path = NULL;
3109
3110         struct path p;
3111
3112         p.dentry = dentry;
3113         p.mnt = current->fs->root.mnt;
3114         path_get(&p);
3115         path = d_path(&p, buf, bufsize);
3116         path_put(&p);
3117         return path;
3118 }
3119
3120 void ll_dirty_page_discard_warn(struct page *page, int ioret)
3121 {
3122         char *buf, *path = NULL;
3123         struct dentry *dentry = NULL;
3124         struct inode *inode = page->mapping->host;
3125
3126         /* this can be called inside spin lock so use GFP_ATOMIC. */
3127         buf = (char *)__get_free_page(GFP_ATOMIC);
3128         if (buf != NULL) {
3129                 dentry = d_find_alias(page->mapping->host);
3130                 if (dentry != NULL)
3131                         path = ll_d_path(dentry, buf, PAGE_SIZE);
3132         }
3133
3134         /* The below message is checked in recovery-small.sh test_24b */
3135         CDEBUG(D_WARNING,
3136                "%s: dirty page discard: %s/fid: "DFID"/%s may get corrupted "
3137                "(rc %d)\n", ll_i2sbi(inode)->ll_fsname,
3138                s2lsi(page->mapping->host->i_sb)->lsi_lmd->lmd_dev,
3139                PFID(ll_inode2fid(inode)),
3140                (path && !IS_ERR(path)) ? path : "", ioret);
3141
3142         if (dentry != NULL)
3143                 dput(dentry);
3144
3145         if (buf != NULL)
3146                 free_page((unsigned long)buf);
3147 }
3148
3149 ssize_t ll_copy_user_md(const struct lov_user_md __user *md,
3150                         struct lov_user_md **kbuf)
3151 {
3152         struct lov_user_md      lum;
3153         ssize_t                 lum_size;
3154         ENTRY;
3155
3156         if (copy_from_user(&lum, md, sizeof(lum)))
3157                 RETURN(-EFAULT);
3158
3159         lum_size = ll_lov_user_md_size(&lum);
3160         if (lum_size < 0)
3161                 RETURN(lum_size);
3162
3163         OBD_ALLOC_LARGE(*kbuf, lum_size);
3164         if (*kbuf == NULL)
3165                 RETURN(-ENOMEM);
3166
3167         if (copy_from_user(*kbuf, md, lum_size) != 0) {
3168                 OBD_FREE_LARGE(*kbuf, lum_size);
3169                 RETURN(-EFAULT);
3170         }
3171
3172         RETURN(lum_size);
3173 }
3174
3175 /*
3176  * Compute llite root squash state after a change of root squash
3177  * configuration setting or add/remove of a lnet nid
3178  */
3179 void ll_compute_rootsquash_state(struct ll_sb_info *sbi)
3180 {
3181         struct root_squash_info *squash = &sbi->ll_squash;
3182         int i;
3183         bool matched;
3184         struct lnet_process_id id;
3185
3186         /* Update norootsquash flag */
3187         spin_lock(&squash->rsi_lock);
3188         if (list_empty(&squash->rsi_nosquash_nids))
3189                 sbi->ll_flags &= ~LL_SBI_NOROOTSQUASH;
3190         else {
3191                 /* Do not apply root squash as soon as one of our NIDs is
3192                  * in the nosquash_nids list */
3193                 matched = false;
3194                 i = 0;
3195                 while (LNetGetId(i++, &id) != -ENOENT) {
3196                         if (id.nid == LNET_NID_LO_0)
3197                                 continue;
3198                         if (cfs_match_nid(id.nid, &squash->rsi_nosquash_nids)) {
3199                                 matched = true;
3200                                 break;
3201                         }
3202                 }
3203                 if (matched)
3204                         sbi->ll_flags |= LL_SBI_NOROOTSQUASH;
3205                 else
3206                         sbi->ll_flags &= ~LL_SBI_NOROOTSQUASH;
3207         }
3208         spin_unlock(&squash->rsi_lock);
3209 }
3210
3211 /**
3212  * Parse linkea content to extract information about a given hardlink
3213  *
3214  * \param[in]   ldata      - Initialized linkea data
3215  * \param[in]   linkno     - Link identifier
3216  * \param[out]  parent_fid - The entry's parent FID
3217  * \param[out]  ln         - Entry name destination buffer
3218  *
3219  * \retval 0 on success
3220  * \retval Appropriate negative error code on failure
3221  */
3222 static int ll_linkea_decode(struct linkea_data *ldata, unsigned int linkno,
3223                             struct lu_fid *parent_fid, struct lu_name *ln)
3224 {
3225         unsigned int    idx;
3226         int             rc;
3227         ENTRY;
3228
3229         rc = linkea_init_with_rec(ldata);
3230         if (rc < 0)
3231                 RETURN(rc);
3232
3233         if (linkno >= ldata->ld_leh->leh_reccount)
3234                 /* beyond last link */
3235                 RETURN(-ENODATA);
3236
3237         linkea_first_entry(ldata);
3238         for (idx = 0; ldata->ld_lee != NULL; idx++) {
3239                 linkea_entry_unpack(ldata->ld_lee, &ldata->ld_reclen, ln,
3240                                     parent_fid);
3241                 if (idx == linkno)
3242                         break;
3243
3244                 linkea_next_entry(ldata);
3245         }
3246
3247         if (idx < linkno)
3248                 RETURN(-ENODATA);
3249
3250         RETURN(0);
3251 }
3252
3253 /**
3254  * Get parent FID and name of an identified link. Operation is performed for
3255  * a given link number, letting the caller iterate over linkno to list one or
3256  * all links of an entry.
3257  *
3258  * \param[in]     file - File descriptor against which to perform the operation
3259  * \param[in,out] arg  - User-filled structure containing the linkno to operate
3260  *                       on and the available size. It is eventually filled with
3261  *                       the requested information or left untouched on error
3262  *
3263  * \retval - 0 on success
3264  * \retval - Appropriate negative error code on failure
3265  */
3266 int ll_getparent(struct file *file, struct getparent __user *arg)
3267 {
3268         struct inode            *inode = file_inode(file);
3269         struct linkea_data      *ldata;
3270         struct lu_buf            buf = LU_BUF_NULL;
3271         struct lu_name           ln;
3272         struct lu_fid            parent_fid;
3273         __u32                    linkno;
3274         __u32                    name_size;
3275         int                      rc;
3276
3277         ENTRY;
3278
3279         if (!cfs_capable(CFS_CAP_DAC_READ_SEARCH) &&
3280             !(ll_i2sbi(inode)->ll_flags & LL_SBI_USER_FID2PATH))
3281                 RETURN(-EPERM);
3282
3283         if (get_user(name_size, &arg->gp_name_size))
3284                 RETURN(-EFAULT);
3285
3286         if (get_user(linkno, &arg->gp_linkno))
3287                 RETURN(-EFAULT);
3288
3289         if (name_size > PATH_MAX)
3290                 RETURN(-EINVAL);
3291
3292         OBD_ALLOC(ldata, sizeof(*ldata));
3293         if (ldata == NULL)
3294                 RETURN(-ENOMEM);
3295
3296         rc = linkea_data_new(ldata, &buf);
3297         if (rc < 0)
3298                 GOTO(ldata_free, rc);
3299
3300         rc = ll_xattr_list(inode, XATTR_NAME_LINK, XATTR_TRUSTED_T, buf.lb_buf,
3301                            buf.lb_len, OBD_MD_FLXATTR);
3302         if (rc < 0)
3303                 GOTO(lb_free, rc);
3304
3305         rc = ll_linkea_decode(ldata, linkno, &parent_fid, &ln);
3306         if (rc < 0)
3307                 GOTO(lb_free, rc);
3308
3309         if (ln.ln_namelen >= name_size)
3310                 GOTO(lb_free, rc = -EOVERFLOW);
3311
3312         if (copy_to_user(&arg->gp_fid, &parent_fid, sizeof(arg->gp_fid)))
3313                 GOTO(lb_free, rc = -EFAULT);
3314
3315         if (copy_to_user(&arg->gp_name, ln.ln_name, ln.ln_namelen))
3316                 GOTO(lb_free, rc = -EFAULT);
3317
3318         if (put_user('\0', arg->gp_name + ln.ln_namelen))
3319                 GOTO(lb_free, rc = -EFAULT);
3320
3321 lb_free:
3322         lu_buf_free(&buf);
3323 ldata_free:
3324         OBD_FREE(ldata, sizeof(*ldata));
3325
3326         RETURN(rc);
3327 }