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