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 #ifdef HAVE_UIDGID_HEADER
  49 # include <linux/uidgid.h>
  50 #endif
  51 #include <linux/security.h>
  52
  53 #include <uapi/linux/lustre/lustre_ioctl.h>
  54 #include <lustre_ha.h>
  55 #include <lustre_dlm.h>
  56 #include <lprocfs_status.h>
  57 #include <lustre_disk.h>
  58 #include <uapi/linux/lustre/lustre_param.h>
  59 #include <lustre_log.h>
  60 #include <cl_object.h>
  61 #include <obd_cksum.h>
  62 #include "llite_internal.h"
  63
  64 struct kmem_cache *ll_file_data_slab;
  65
  66 #ifndef log2
  67 #define log2(n) ffz(~(n))
  68 #endif
  69
  70 static struct ll_sb_info *ll_init_sbi(void)
  71 {
  72         struct ll_sb_info *sbi = NULL;
  73         unsigned long pages;
  74         unsigned long lru_page_max;
  75         struct sysinfo si;
  76         int rc;
  77         int i;
  78
  79         ENTRY;
  80
  81         OBD_ALLOC_PTR(sbi);
  82         if (sbi == NULL)
  83                 RETURN(ERR_PTR(-ENOMEM));
  84
  85         rc = pcc_super_init(&sbi->ll_pcc_super);
  86         if (rc < 0)
  87                 GOTO(out_sbi, rc);
  88
  89         spin_lock_init(&sbi->ll_lock);
  90         mutex_init(&sbi->ll_lco.lco_lock);
  91         spin_lock_init(&sbi->ll_pp_extent_lock);
  92         spin_lock_init(&sbi->ll_process_lock);
  93         sbi->ll_rw_stats_on = 0;
  94
  95         si_meminfo(&si);
  96         pages = si.totalram - si.totalhigh;
  97         lru_page_max = pages / 2;
  98
  99         /* initialize ll_cache data */
 100         sbi->ll_cache = cl_cache_init(lru_page_max);
 101         if (sbi->ll_cache == NULL)
 102                 GOTO(out_pcc, rc = -ENOMEM);
 103
 104         sbi->ll_ra_info.ra_max_pages_per_file = min(pages / 32,
 105                                            SBI_DEFAULT_READAHEAD_MAX);
 106         sbi->ll_ra_info.ra_max_pages = sbi->ll_ra_info.ra_max_pages_per_file;
 107         sbi->ll_ra_info.ra_max_read_ahead_whole_pages = -1;
 108
 109         sbi->ll_flags |= LL_SBI_VERBOSE;
 110 #ifdef ENABLE_CHECKSUM
 111         sbi->ll_flags |= LL_SBI_CHECKSUM;
 112 #endif
 113 #ifdef ENABLE_FLOCK
 114         sbi->ll_flags |= LL_SBI_FLOCK;
 115 #endif
 116
 117 #ifdef HAVE_LRU_RESIZE_SUPPORT
 118         sbi->ll_flags |= LL_SBI_LRU_RESIZE;
 119 #endif
 120         sbi->ll_flags |= LL_SBI_LAZYSTATFS;
 121
 122         for (i = 0; i <= LL_PROCESS_HIST_MAX; i++) {
 123                 spin_lock_init(&sbi->ll_rw_extents_info.pp_extents[i].
 124                                pp_r_hist.oh_lock);
 125                 spin_lock_init(&sbi->ll_rw_extents_info.pp_extents[i].
 126                                pp_w_hist.oh_lock);
 127         }
 128
 129         /* metadata statahead is enabled by default */
 130         sbi->ll_sa_running_max = LL_SA_RUNNING_DEF;
 131         sbi->ll_sa_max = LL_SA_RPC_DEF;
 132         atomic_set(&sbi->ll_sa_total, 0);
 133         atomic_set(&sbi->ll_sa_wrong, 0);
 134         atomic_set(&sbi->ll_sa_running, 0);
 135         atomic_set(&sbi->ll_agl_total, 0);
 136         sbi->ll_flags |= LL_SBI_AGL_ENABLED;
 137         sbi->ll_flags |= LL_SBI_FAST_READ;
 138         sbi->ll_flags |= LL_SBI_TINY_WRITE;
 139
 140         /* root squash */
 141         sbi->ll_squash.rsi_uid = 0;
 142         sbi->ll_squash.rsi_gid = 0;
 143         INIT_LIST_HEAD(&sbi->ll_squash.rsi_nosquash_nids);
 144         init_rwsem(&sbi->ll_squash.rsi_sem);
 145
 146         /* Per-filesystem file heat */
 147         sbi->ll_heat_decay_weight = SBI_DEFAULT_HEAT_DECAY_WEIGHT;
 148         sbi->ll_heat_period_second = SBI_DEFAULT_HEAT_PERIOD_SECOND;
 149         RETURN(sbi);
 150 out_pcc:
 151         pcc_super_fini(&sbi->ll_pcc_super);
 152 out_sbi:
 153         OBD_FREE_PTR(sbi);
 154         RETURN(ERR_PTR(rc));
 155 }
 156
 157 static void ll_free_sbi(struct super_block *sb)
 158 {
 159         struct ll_sb_info *sbi = ll_s2sbi(sb);
 160         ENTRY;
 161
 162         if (sbi != NULL) {
 163                 if (!list_empty(&sbi->ll_squash.rsi_nosquash_nids))
 164                         cfs_free_nidlist(&sbi->ll_squash.rsi_nosquash_nids);
 165                 if (sbi->ll_cache != NULL) {
 166                         cl_cache_decref(sbi->ll_cache);
 167                         sbi->ll_cache = NULL;
 168                 }
 169                 pcc_super_fini(&sbi->ll_pcc_super);
 170                 OBD_FREE(sbi, sizeof(*sbi));
 171         }
 172         EXIT;
 173 }
 174
 175 static int client_common_fill_super(struct super_block *sb, char *md, char *dt,
 176                                     struct vfsmount *mnt)
 177 {
 178         struct inode *root = NULL;
 179         struct ll_sb_info *sbi = ll_s2sbi(sb);
 180         struct obd_statfs *osfs = NULL;
 181         struct ptlrpc_request *request = NULL;
 182         struct obd_connect_data *data = NULL;
 183         struct obd_uuid *uuid;
 184         struct md_op_data *op_data;
 185         struct lustre_md lmd;
 186         u64 valid;
 187         int size, err, checksum;
 188
 189         ENTRY;
 190         sbi->ll_md_obd = class_name2obd(md);
 191         if (!sbi->ll_md_obd) {
 192                 CERROR("MD %s: not setup or attached\n", md);
 193                 RETURN(-EINVAL);
 194         }
 195
 196         OBD_ALLOC_PTR(data);
 197         if (data == NULL)
 198                 RETURN(-ENOMEM);
 199
 200         OBD_ALLOC_PTR(osfs);
 201         if (osfs == NULL) {
 202                 OBD_FREE_PTR(data);
 203                 RETURN(-ENOMEM);
 204         }
 205
 206         /* pass client page size via ocd_grant_blkbits, the server should report
 207          * back its backend blocksize for grant calculation purpose */
 208         data->ocd_grant_blkbits = PAGE_SHIFT;
 209
 210         /* indicate MDT features supported by this client */
 211         data->ocd_connect_flags = OBD_CONNECT_IBITS    | OBD_CONNECT_NODEVOH  |
 212                                   OBD_CONNECT_ATTRFID  | OBD_CONNECT_GRANT |
 213                                   OBD_CONNECT_VERSION  | OBD_CONNECT_BRW_SIZE |
 214                                   OBD_CONNECT_SRVLOCK  | OBD_CONNECT_TRUNCLOCK|
 215                                   OBD_CONNECT_MDS_CAPA | OBD_CONNECT_OSS_CAPA |
 216                                   OBD_CONNECT_CANCELSET | OBD_CONNECT_FID     |
 217                                   OBD_CONNECT_AT       | OBD_CONNECT_LOV_V3   |
 218                                   OBD_CONNECT_VBR | OBD_CONNECT_FULL20 |
 219                                   OBD_CONNECT_64BITHASH |
 220                                   OBD_CONNECT_EINPROGRESS |
 221                                   OBD_CONNECT_JOBSTATS | OBD_CONNECT_LVB_TYPE |
 222                                   OBD_CONNECT_LAYOUTLOCK | OBD_CONNECT_PINGLESS|
 223                                   OBD_CONNECT_MAX_EASIZE |
 224                                   OBD_CONNECT_FLOCK_DEAD |
 225                                   OBD_CONNECT_DISP_STRIPE | OBD_CONNECT_LFSCK |
 226                                   OBD_CONNECT_OPEN_BY_FID |
 227                                   OBD_CONNECT_DIR_STRIPE |
 228                                   OBD_CONNECT_BULK_MBITS | OBD_CONNECT_CKSUM |
 229                                   OBD_CONNECT_SUBTREE |
 230                                   OBD_CONNECT_MULTIMODRPCS |
 231                                   OBD_CONNECT_GRANT_PARAM |
 232                                   OBD_CONNECT_SHORTIO | OBD_CONNECT_FLAGS2;
 233
 234         data->ocd_connect_flags2 = OBD_CONNECT2_DIR_MIGRATE |
 235                                    OBD_CONNECT2_SUM_STATFS |
 236                                    OBD_CONNECT2_OVERSTRIPING |
 237                                    OBD_CONNECT2_FLR |
 238                                    OBD_CONNECT2_LOCK_CONVERT |
 239                                    OBD_CONNECT2_ARCHIVE_ID_ARRAY |
 240                                    OBD_CONNECT2_LSOM |
 241                                    OBD_CONNECT2_ASYNC_DISCARD |
 242                                    OBD_CONNECT2_PCC;
 243
 244 #ifdef HAVE_LRU_RESIZE_SUPPORT
 245         if (sbi->ll_flags & LL_SBI_LRU_RESIZE)
 246                 data->ocd_connect_flags |= OBD_CONNECT_LRU_RESIZE;
 247 #endif
 248 #ifdef CONFIG_FS_POSIX_ACL
 249         data->ocd_connect_flags |= OBD_CONNECT_ACL | OBD_CONNECT_UMASK |
 250                                    OBD_CONNECT_LARGE_ACL;
 251 #endif
 252
 253         data->ocd_cksum_types = obd_cksum_types_supported_client();
 254
 255         if (OBD_FAIL_CHECK(OBD_FAIL_MDC_LIGHTWEIGHT))
 256                 /* flag mdc connection as lightweight, only used for test
 257                  * purpose, use with care */
 258                 data->ocd_connect_flags |= OBD_CONNECT_LIGHTWEIGHT;
 259
 260         data->ocd_ibits_known = MDS_INODELOCK_FULL;
 261         data->ocd_version = LUSTRE_VERSION_CODE;
 262
 263         if (sb->s_flags & MS_RDONLY)
 264                 data->ocd_connect_flags |= OBD_CONNECT_RDONLY;
 265         if (sbi->ll_flags & LL_SBI_USER_XATTR)
 266                 data->ocd_connect_flags |= OBD_CONNECT_XATTR;
 267
 268 #ifdef MS_NOSEC
 269         /* Setting this indicates we correctly support S_NOSEC (See kernel
 270          * commit 9e1f1de02c2275d7172e18dc4e7c2065777611bf)
 271          */
 272         sb->s_flags |= MS_NOSEC;
 273 #endif
 274
 275         if (sbi->ll_flags & LL_SBI_FLOCK)
 276                 sbi->ll_fop = &ll_file_operations_flock;
 277         else if (sbi->ll_flags & LL_SBI_LOCALFLOCK)
 278                 sbi->ll_fop = &ll_file_operations;
 279         else
 280                 sbi->ll_fop = &ll_file_operations_noflock;
 281
 282         /* always ping even if server suppress_pings */
 283         if (sbi->ll_flags & LL_SBI_ALWAYS_PING)
 284                 data->ocd_connect_flags &= ~OBD_CONNECT_PINGLESS;
 285
 286         obd_connect_set_secctx(data);
 287
 288 #if defined(CONFIG_SECURITY)
 289         data->ocd_connect_flags2 |= OBD_CONNECT2_SELINUX_POLICY;
 290 #endif
 291
 292         data->ocd_brw_size = MD_MAX_BRW_SIZE;
 293
 294         err = obd_connect(NULL, &sbi->ll_md_exp, sbi->ll_md_obd,
 295                           &sbi->ll_sb_uuid, data, sbi->ll_cache);
 296         if (err == -EBUSY) {
 297                 LCONSOLE_ERROR_MSG(0x14f, "An MDT (md %s) is performing "
 298                                    "recovery, of which this client is not a "
 299                                    "part. Please wait for recovery to complete,"
 300                                    " abort, or time out.\n", md);
 301                 GOTO(out, err);
 302         } else if (err) {
 303                 CERROR("cannot connect to %s: rc = %d\n", md, err);
 304                 GOTO(out, err);
 305         }
 306
 307         sbi->ll_md_exp->exp_connect_data = *data;
 308
 309         err = obd_fid_init(sbi->ll_md_exp->exp_obd, sbi->ll_md_exp,
 310                            LUSTRE_SEQ_METADATA);
 311         if (err) {
 312                 CERROR("%s: Can't init metadata layer FID infrastructure, "
 313                        "rc = %d\n", sbi->ll_md_exp->exp_obd->obd_name, err);
 314                 GOTO(out_md, err);
 315         }
 316
 317         /* For mount, we only need fs info from MDT0, and also in DNE, it
 318          * can make sure the client can be mounted as long as MDT0 is
 319          * avaible */
 320         err = obd_statfs(NULL, sbi->ll_md_exp, osfs,
 321                         ktime_get_seconds() - OBD_STATFS_CACHE_SECONDS,
 322                         OBD_STATFS_FOR_MDT0);
 323         if (err)
 324                 GOTO(out_md_fid, err);
 325
 326         /* This needs to be after statfs to ensure connect has finished.
 327          * Note that "data" does NOT contain the valid connect reply.
 328          * If connecting to a 1.8 server there will be no LMV device, so
 329          * we can access the MDC export directly and exp_connect_flags will
 330          * be non-zero, but if accessing an upgraded 2.1 server it will
 331          * have the correct flags filled in.
 332          * XXX: fill in the LMV exp_connect_flags from MDC(s). */
 333         valid = exp_connect_flags(sbi->ll_md_exp) & CLIENT_CONNECT_MDT_REQD;
 334         if (exp_connect_flags(sbi->ll_md_exp) != 0 &&
 335             valid != CLIENT_CONNECT_MDT_REQD) {
 336                 char *buf;
 337
 338                 OBD_ALLOC_WAIT(buf, PAGE_SIZE);
 339                 obd_connect_flags2str(buf, PAGE_SIZE,
 340                                       valid ^ CLIENT_CONNECT_MDT_REQD, 0, ",");
 341                 LCONSOLE_ERROR_MSG(0x170, "Server %s does not support "
 342                                    "feature(s) needed for correct operation "
 343                                    "of this client (%s). Please upgrade "
 344                                    "server or downgrade client.\n",
 345                                    sbi->ll_md_exp->exp_obd->obd_name, buf);
 346                 OBD_FREE(buf, PAGE_SIZE);
 347                 GOTO(out_md_fid, err = -EPROTO);
 348         }
 349
 350         size = sizeof(*data);
 351         err = obd_get_info(NULL, sbi->ll_md_exp, sizeof(KEY_CONN_DATA),
 352                            KEY_CONN_DATA,  &size, data);
 353         if (err) {
 354                 CERROR("%s: Get connect data failed: rc = %d\n",
 355                        sbi->ll_md_exp->exp_obd->obd_name, err);
 356                 GOTO(out_md_fid, err);
 357         }
 358
 359         LASSERT(osfs->os_bsize);
 360         sb->s_blocksize = osfs->os_bsize;
 361         sb->s_blocksize_bits = log2(osfs->os_bsize);
 362         sb->s_magic = LL_SUPER_MAGIC;
 363         sb->s_maxbytes = MAX_LFS_FILESIZE;
 364         sbi->ll_namelen = osfs->os_namelen;
 365         sbi->ll_mnt.mnt = current->fs->root.mnt;
 366
 367         if ((sbi->ll_flags & LL_SBI_USER_XATTR) &&
 368             !(data->ocd_connect_flags & OBD_CONNECT_XATTR)) {
 369                 LCONSOLE_INFO("Disabling user_xattr feature because "
 370                               "it is not supported on the server\n");
 371                 sbi->ll_flags &= ~LL_SBI_USER_XATTR;
 372         }
 373
 374         if (data->ocd_connect_flags & OBD_CONNECT_ACL) {
 375 #ifdef MS_POSIXACL
 376                 sb->s_flags |= MS_POSIXACL;
 377 #endif
 378                 sbi->ll_flags |= LL_SBI_ACL;
 379         } else {
 380                 LCONSOLE_INFO("client wants to enable acl, but mdt not!\n");
 381 #ifdef MS_POSIXACL
 382                 sb->s_flags &= ~MS_POSIXACL;
 383 #endif
 384                 sbi->ll_flags &= ~LL_SBI_ACL;
 385         }
 386
 387         if (data->ocd_connect_flags & OBD_CONNECT_64BITHASH)
 388                 sbi->ll_flags |= LL_SBI_64BIT_HASH;
 389
 390         if (data->ocd_connect_flags & OBD_CONNECT_LAYOUTLOCK)
 391                 sbi->ll_flags |= LL_SBI_LAYOUT_LOCK;
 392
 393         if (obd_connect_has_secctx(data))
 394                 sbi->ll_flags |= LL_SBI_FILE_SECCTX;
 395
 396         if (data->ocd_ibits_known & MDS_INODELOCK_XATTR) {
 397                 if (!(data->ocd_connect_flags & OBD_CONNECT_MAX_EASIZE)) {
 398                         LCONSOLE_INFO("%s: disabling xattr cache due to "
 399                                       "unknown maximum xattr size.\n", dt);
 400                 } else if (!sbi->ll_xattr_cache_set) {
 401                         /* If xattr_cache is already set (no matter 0 or 1)
 402                          * during processing llog, it won't be enabled here. */
 403                         sbi->ll_flags |= LL_SBI_XATTR_CACHE;
 404                         sbi->ll_xattr_cache_enabled = 1;
 405                 }
 406         }
 407
 408         sbi->ll_dt_obd = class_name2obd(dt);
 409         if (!sbi->ll_dt_obd) {
 410                 CERROR("DT %s: not setup or attached\n", dt);
 411                 GOTO(out_md_fid, err = -ENODEV);
 412         }
 413
 414         /* pass client page size via ocd_grant_blkbits, the server should report
 415          * back its backend blocksize for grant calculation purpose */
 416         data->ocd_grant_blkbits = PAGE_SHIFT;
 417
 418         /* indicate OST features supported by this client */
 419         data->ocd_connect_flags = OBD_CONNECT_GRANT | OBD_CONNECT_VERSION |
 420                                   OBD_CONNECT_REQPORTAL | OBD_CONNECT_BRW_SIZE |
 421                                   OBD_CONNECT_CANCELSET | OBD_CONNECT_FID |
 422                                   OBD_CONNECT_SRVLOCK | OBD_CONNECT_TRUNCLOCK|
 423                                   OBD_CONNECT_AT | OBD_CONNECT_OSS_CAPA |
 424                                   OBD_CONNECT_VBR | OBD_CONNECT_FULL20 |
 425                                   OBD_CONNECT_64BITHASH | OBD_CONNECT_MAXBYTES |
 426                                   OBD_CONNECT_EINPROGRESS |
 427                                   OBD_CONNECT_JOBSTATS | OBD_CONNECT_LVB_TYPE |
 428                                   OBD_CONNECT_LAYOUTLOCK |
 429                                   OBD_CONNECT_PINGLESS | OBD_CONNECT_LFSCK |
 430                                   OBD_CONNECT_BULK_MBITS | OBD_CONNECT_SHORTIO |
 431                                   OBD_CONNECT_FLAGS2 | OBD_CONNECT_GRANT_SHRINK;
 432
 433 /* The client currently advertises support for OBD_CONNECT_LOCKAHEAD_OLD so it
 434  * can interoperate with an older version of lockahead which was released prior
 435  * to landing in master. This support will be dropped when 2.13 development
 436  * starts.  At the point, we should not just drop the connect flag (below), we
 437  * should also remove the support in the code.
 438  *
 439  * Removing it means a few things:
 440  * 1. Remove this section here
 441  * 2. Remove CEF_NONBLOCK in ll_file_lockahead()
 442  * 3. Remove function exp_connect_lockahead_old
 443  * 4. Remove LDLM_FL_LOCKAHEAD_OLD_RESERVED in lustre_dlm_flags.h
 444  * */
 445 #if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(2, 12, 50, 0)
 446         data->ocd_connect_flags |= OBD_CONNECT_LOCKAHEAD_OLD;
 447 #endif
 448
 449         data->ocd_connect_flags2 = OBD_CONNECT2_LOCKAHEAD;
 450
 451         if (!OBD_FAIL_CHECK(OBD_FAIL_OSC_CONNECT_GRANT_PARAM))
 452                 data->ocd_connect_flags |= OBD_CONNECT_GRANT_PARAM;
 453
 454         /* OBD_CONNECT_CKSUM should always be set, even if checksums are
 455          * disabled by default, because it can still be enabled on the
 456          * fly via /sys. As a consequence, we still need to come to an
 457          * agreement on the supported algorithms at connect time
 458          */
 459         data->ocd_connect_flags |= OBD_CONNECT_CKSUM;
 460
 461         if (OBD_FAIL_CHECK(OBD_FAIL_OSC_CKSUM_ADLER_ONLY))
 462                 data->ocd_cksum_types = OBD_CKSUM_ADLER;
 463         else
 464                 data->ocd_cksum_types = obd_cksum_types_supported_client();
 465
 466 #ifdef HAVE_LRU_RESIZE_SUPPORT
 467         data->ocd_connect_flags |= OBD_CONNECT_LRU_RESIZE;
 468 #endif
 469         /* always ping even if server suppress_pings */
 470         if (sbi->ll_flags & LL_SBI_ALWAYS_PING)
 471                 data->ocd_connect_flags &= ~OBD_CONNECT_PINGLESS;
 472
 473         CDEBUG(D_RPCTRACE, "ocd_connect_flags: %#llx ocd_version: %d "
 474                "ocd_grant: %d\n", data->ocd_connect_flags,
 475                data->ocd_version, data->ocd_grant);
 476
 477         sbi->ll_dt_obd->obd_upcall.onu_owner = &sbi->ll_lco;
 478         sbi->ll_dt_obd->obd_upcall.onu_upcall = cl_ocd_update;
 479
 480         data->ocd_brw_size = DT_MAX_BRW_SIZE;
 481
 482         err = obd_connect(NULL, &sbi->ll_dt_exp, sbi->ll_dt_obd,
 483                           &sbi->ll_sb_uuid, data, sbi->ll_cache);
 484         if (err == -EBUSY) {
 485                 LCONSOLE_ERROR_MSG(0x150, "An OST (dt %s) is performing "
 486                                    "recovery, of which this client is not a "
 487                                    "part.  Please wait for recovery to "
 488                                    "complete, abort, or time out.\n", dt);
 489                 GOTO(out_md, err);
 490         } else if (err) {
 491                 CERROR("%s: Cannot connect to %s: rc = %d\n",
 492                        sbi->ll_dt_exp->exp_obd->obd_name, dt, err);
 493                 GOTO(out_md, err);
 494         }
 495
 496         sbi->ll_dt_exp->exp_connect_data = *data;
 497
 498         /* Don't change value if it was specified in the config log */
 499         if (sbi->ll_ra_info.ra_max_read_ahead_whole_pages == -1) {
 500                 sbi->ll_ra_info.ra_max_read_ahead_whole_pages =
 501                         max_t(unsigned long, SBI_DEFAULT_READAHEAD_WHOLE_MAX,
 502                               (data->ocd_brw_size >> PAGE_SHIFT));
 503                 if (sbi->ll_ra_info.ra_max_read_ahead_whole_pages >
 504                     sbi->ll_ra_info.ra_max_pages_per_file)
 505                         sbi->ll_ra_info.ra_max_read_ahead_whole_pages =
 506                                 sbi->ll_ra_info.ra_max_pages_per_file;
 507         }
 508
 509         err = obd_fid_init(sbi->ll_dt_exp->exp_obd, sbi->ll_dt_exp,
 510                            LUSTRE_SEQ_METADATA);
 511         if (err) {
 512                 CERROR("%s: Can't init data layer FID infrastructure, "
 513                        "rc = %d\n", sbi->ll_dt_exp->exp_obd->obd_name, err);
 514                 GOTO(out_dt, err);
 515         }
 516
 517         mutex_lock(&sbi->ll_lco.lco_lock);
 518         sbi->ll_lco.lco_flags = data->ocd_connect_flags;
 519         sbi->ll_lco.lco_md_exp = sbi->ll_md_exp;
 520         sbi->ll_lco.lco_dt_exp = sbi->ll_dt_exp;
 521         mutex_unlock(&sbi->ll_lco.lco_lock);
 522
 523         fid_zero(&sbi->ll_root_fid);
 524         err = md_get_root(sbi->ll_md_exp, get_mount_fileset(sb),
 525                            &sbi->ll_root_fid);
 526         if (err) {
 527                 CERROR("cannot mds_connect: rc = %d\n", err);
 528                 GOTO(out_lock_cn_cb, err);
 529         }
 530         if (!fid_is_sane(&sbi->ll_root_fid)) {
 531                 CERROR("%s: Invalid root fid "DFID" during mount\n",
 532                        sbi->ll_md_exp->exp_obd->obd_name,
 533                        PFID(&sbi->ll_root_fid));
 534                 GOTO(out_lock_cn_cb, err = -EINVAL);
 535         }
 536         CDEBUG(D_SUPER, "rootfid "DFID"\n", PFID(&sbi->ll_root_fid));
 537
 538         sb->s_op = &lustre_super_operations;
 539 #ifdef HAVE_XATTR_HANDLER_FLAGS
 540         sb->s_xattr = ll_xattr_handlers;
 541 #endif
 542 #if THREAD_SIZE >= 8192 /*b=17630*/
 543         sb->s_export_op = &lustre_export_operations;
 544 #endif
 545
 546         /* make root inode
 547          * XXX: move this to after cbd setup? */
 548         valid = OBD_MD_FLGETATTR | OBD_MD_FLBLOCKS | OBD_MD_FLMODEASIZE;
 549         if (sbi->ll_flags & LL_SBI_ACL)
 550                 valid |= OBD_MD_FLACL;
 551
 552         OBD_ALLOC_PTR(op_data);
 553         if (op_data == NULL)
 554                 GOTO(out_lock_cn_cb, err = -ENOMEM);
 555
 556         op_data->op_fid1 = sbi->ll_root_fid;
 557         op_data->op_mode = 0;
 558         op_data->op_valid = valid;
 559
 560         err = md_getattr(sbi->ll_md_exp, op_data, &request);
 561
 562         OBD_FREE_PTR(op_data);
 563         if (err) {
 564                 CERROR("%s: md_getattr failed for root: rc = %d\n",
 565                        sbi->ll_md_exp->exp_obd->obd_name, err);
 566                 GOTO(out_lock_cn_cb, err);
 567         }
 568
 569         err = md_get_lustre_md(sbi->ll_md_exp, request, sbi->ll_dt_exp,
 570                                sbi->ll_md_exp, &lmd);
 571         if (err) {
 572                 CERROR("failed to understand root inode md: rc = %d\n", err);
 573                 ptlrpc_req_finished(request);
 574                 GOTO(out_lock_cn_cb, err);
 575         }
 576
 577         LASSERT(fid_is_sane(&sbi->ll_root_fid));
 578         root = ll_iget(sb, cl_fid_build_ino(&sbi->ll_root_fid,
 579                                             sbi->ll_flags & LL_SBI_32BIT_API),
 580                        &lmd);
 581         md_free_lustre_md(sbi->ll_md_exp, &lmd);
 582         ptlrpc_req_finished(request);
 583
 584         if (IS_ERR(root)) {
 585 #ifdef CONFIG_FS_POSIX_ACL
 586                 if (lmd.posix_acl) {
 587                         posix_acl_release(lmd.posix_acl);
 588                         lmd.posix_acl = NULL;
 589                 }
 590 #endif
 591                 err = IS_ERR(root) ? PTR_ERR(root) : -EBADF;
 592                 root = NULL;
 593                 CERROR("lustre_lite: bad iget4 for root\n");
 594                 GOTO(out_root, err);
 595         }
 596
 597         checksum = sbi->ll_flags & LL_SBI_CHECKSUM;
 598         if (sbi->ll_checksum_set) {
 599                 err = obd_set_info_async(NULL, sbi->ll_dt_exp,
 600                                          sizeof(KEY_CHECKSUM), KEY_CHECKSUM,
 601                                          sizeof(checksum), &checksum, NULL);
 602                 if (err) {
 603                         CERROR("%s: Set checksum failed: rc = %d\n",
 604                                sbi->ll_dt_exp->exp_obd->obd_name, err);
 605                         GOTO(out_root, err);
 606                 }
 607         }
 608         cl_sb_init(sb);
 609
 610         sb->s_root = d_make_root(root);
 611         if (sb->s_root == NULL) {
 612                 err = -ENOMEM;
 613                 CERROR("%s: can't make root dentry: rc = %d\n",
 614                        sbi->ll_fsname, err);
 615                 GOTO(out_root, err);
 616         }
 617 #ifdef HAVE_DCACHE_LOCK
 618         sb->s_root->d_op = &ll_d_ops;
 619 #endif
 620
 621         sbi->ll_sdev_orig = sb->s_dev;
 622
 623         /* We set sb->s_dev equal on all lustre clients in order to support
 624          * NFS export clustering.  NFSD requires that the FSID be the same
 625          * on all clients. */
 626         /* s_dev is also used in lt_compare() to compare two fs, but that is
 627          * only a node-local comparison. */
 628         uuid = obd_get_uuid(sbi->ll_md_exp);
 629         if (uuid != NULL)
 630                 sb->s_dev = get_uuid2int(uuid->uuid, strlen(uuid->uuid));
 631
 632         if (data != NULL)
 633                 OBD_FREE_PTR(data);
 634         if (osfs != NULL)
 635                 OBD_FREE_PTR(osfs);
 636
 637         if (sbi->ll_dt_obd) {
 638                 err = sysfs_create_link(&sbi->ll_kset.kobj,
 639                                         &sbi->ll_dt_obd->obd_kset.kobj,
 640                                         sbi->ll_dt_obd->obd_type->typ_name);
 641                 if (err < 0) {
 642                         CERROR("%s: could not register %s in llite: rc = %d\n",
 643                                dt, sbi->ll_fsname, err);
 644                         err = 0;
 645                 }
 646         }
 647
 648         if (sbi->ll_md_obd) {
 649                 err = sysfs_create_link(&sbi->ll_kset.kobj,
 650                                         &sbi->ll_md_obd->obd_kset.kobj,
 651                                         sbi->ll_md_obd->obd_type->typ_name);
 652                 if (err < 0) {
 653                         CERROR("%s: could not register %s in llite: rc = %d\n",
 654                                md, sbi->ll_fsname, err);
 655                         err = 0;
 656                 }
 657         }
 658
 659         RETURN(err);
 660 out_root:
 661         if (root)
 662                 iput(root);
 663 out_lock_cn_cb:
 664         obd_fid_fini(sbi->ll_dt_exp->exp_obd);
 665 out_dt:
 666         obd_disconnect(sbi->ll_dt_exp);
 667         sbi->ll_dt_exp = NULL;
 668         sbi->ll_dt_obd = NULL;
 669 out_md_fid:
 670         obd_fid_fini(sbi->ll_md_exp->exp_obd);
 671 out_md:
 672         obd_disconnect(sbi->ll_md_exp);
 673         sbi->ll_md_exp = NULL;
 674         sbi->ll_md_obd = NULL;
 675 out:
 676         if (data != NULL)
 677                 OBD_FREE_PTR(data);
 678         if (osfs != NULL)
 679                 OBD_FREE_PTR(osfs);
 680         return err;
 681 }
 682
 683 int ll_get_max_mdsize(struct ll_sb_info *sbi, int *lmmsize)
 684 {
 685         int size, rc;
 686
 687         size = sizeof(*lmmsize);
 688         rc = obd_get_info(NULL, sbi->ll_dt_exp, sizeof(KEY_MAX_EASIZE),
 689                           KEY_MAX_EASIZE, &size, lmmsize);
 690         if (rc != 0) {
 691                 CERROR("%s: cannot get max LOV EA size: rc = %d\n",
 692                        sbi->ll_dt_exp->exp_obd->obd_name, rc);
 693                 RETURN(rc);
 694         }
 695
 696         CDEBUG(D_INFO, "max LOV ea size: %d\n", *lmmsize);
 697
 698         size = sizeof(int);
 699         rc = obd_get_info(NULL, sbi->ll_md_exp, sizeof(KEY_MAX_EASIZE),
 700                           KEY_MAX_EASIZE, &size, lmmsize);
 701         if (rc)
 702                 CERROR("Get max mdsize error rc %d\n", rc);
 703
 704         CDEBUG(D_INFO, "max LMV ea size: %d\n", *lmmsize);
 705
 706         RETURN(rc);
 707 }
 708
 709 /**
 710  * Get the value of the default_easize parameter.
 711  *
 712  * \see client_obd::cl_default_mds_easize
 713  *
 714  * \param[in] sbi       superblock info for this filesystem
 715  * \param[out] lmmsize  pointer to storage location for value
 716  *
 717  * \retval 0            on success
 718  * \retval negative     negated errno on failure
 719  */
 720 int ll_get_default_mdsize(struct ll_sb_info *sbi, int *lmmsize)
 721 {
 722         int size, rc;
 723
 724         size = sizeof(int);
 725         rc = obd_get_info(NULL, sbi->ll_md_exp, sizeof(KEY_DEFAULT_EASIZE),
 726                          KEY_DEFAULT_EASIZE, &size, lmmsize);
 727         if (rc)
 728                 CERROR("Get default mdsize error rc %d\n", rc);
 729
 730         RETURN(rc);
 731 }
 732
 733 /**
 734  * Set the default_easize parameter to the given value.
 735  *
 736  * \see client_obd::cl_default_mds_easize
 737  *
 738  * \param[in] sbi       superblock info for this filesystem
 739  * \param[in] lmmsize   the size to set
 740  *
 741  * \retval 0            on success
 742  * \retval negative     negated errno on failure
 743  */
 744 int ll_set_default_mdsize(struct ll_sb_info *sbi, int lmmsize)
 745 {
 746         int rc;
 747
 748         if (lmmsize < sizeof(struct lov_mds_md) ||
 749             lmmsize > OBD_MAX_DEFAULT_EA_SIZE)
 750                 return -EINVAL;
 751
 752         rc = obd_set_info_async(NULL, sbi->ll_md_exp,
 753                                 sizeof(KEY_DEFAULT_EASIZE), KEY_DEFAULT_EASIZE,
 754                                 sizeof(int), &lmmsize, NULL);
 755
 756         RETURN(rc);
 757 }
 758
 759 static void client_common_put_super(struct super_block *sb)
 760 {
 761         struct ll_sb_info *sbi = ll_s2sbi(sb);
 762         ENTRY;
 763
 764         cl_sb_fini(sb);
 765
 766         obd_fid_fini(sbi->ll_dt_exp->exp_obd);
 767         obd_disconnect(sbi->ll_dt_exp);
 768         sbi->ll_dt_exp = NULL;
 769
 770         ll_debugfs_unregister_super(sb);
 771
 772         obd_fid_fini(sbi->ll_md_exp->exp_obd);
 773         obd_disconnect(sbi->ll_md_exp);
 774         sbi->ll_md_exp = NULL;
 775
 776         EXIT;
 777 }
 778
 779 void ll_kill_super(struct super_block *sb)
 780 {
 781         struct ll_sb_info *sbi;
 782         ENTRY;
 783
 784         /* not init sb ?*/
 785         if (!(sb->s_flags & MS_ACTIVE))
 786                 return;
 787
 788         sbi = ll_s2sbi(sb);
 789         /* we need restore s_dev from changed for clustred NFS before put_super
 790          * because new kernels have cached s_dev and change sb->s_dev in
 791          * put_super not affected real removing devices */
 792         if (sbi) {
 793                 sb->s_dev = sbi->ll_sdev_orig;
 794                 sbi->ll_umounting = 1;
 795
 796                 /* wait running statahead threads to quit */
 797                 while (atomic_read(&sbi->ll_sa_running) > 0) {
 798                         set_current_state(TASK_UNINTERRUPTIBLE);
 799                         schedule_timeout(msecs_to_jiffies(MSEC_PER_SEC >> 3));
 800                 }
 801         }
 802
 803         EXIT;
 804 }
 805
 806 static inline int ll_set_opt(const char *opt, char *data, int fl)
 807 {
 808         if (strncmp(opt, data, strlen(opt)) != 0)
 809                 return 0;
 810         else
 811                 return fl;
 812 }
 813
 814 /* non-client-specific mount options are parsed in lmd_parse */
 815 static int ll_options(char *options, struct ll_sb_info *sbi)
 816 {
 817         int tmp;
 818         char *s1 = options, *s2;
 819         int *flags = &sbi->ll_flags;
 820         ENTRY;
 821
 822         if (!options)
 823                 RETURN(0);
 824
 825         CDEBUG(D_CONFIG, "Parsing opts %s\n", options);
 826
 827         while (*s1) {
 828                 CDEBUG(D_SUPER, "next opt=%s\n", s1);
 829                 tmp = ll_set_opt("nolock", s1, LL_SBI_NOLCK);
 830                 if (tmp) {
 831                         *flags |= tmp;
 832                         goto next;
 833                 }
 834                 tmp = ll_set_opt("flock", s1, LL_SBI_FLOCK);
 835                 if (tmp) {
 836                         *flags |= tmp;
 837                         goto next;
 838                 }
 839                 tmp = ll_set_opt("localflock", s1, LL_SBI_LOCALFLOCK);
 840                 if (tmp) {
 841                         *flags |= tmp;
 842                         goto next;
 843                 }
 844                 tmp = ll_set_opt("noflock", s1, LL_SBI_FLOCK|LL_SBI_LOCALFLOCK);
 845                 if (tmp) {
 846                         *flags &= ~tmp;
 847                         goto next;
 848                 }
 849                 tmp = ll_set_opt("user_xattr", s1, LL_SBI_USER_XATTR);
 850                 if (tmp) {
 851                         *flags |= tmp;
 852                         goto next;
 853                 }
 854                 tmp = ll_set_opt("nouser_xattr", s1, LL_SBI_USER_XATTR);
 855                 if (tmp) {
 856                         *flags &= ~tmp;
 857                         goto next;
 858                 }
 859                 tmp = ll_set_opt("context", s1, 1);
 860                 if (tmp)
 861                         goto next;
 862                 tmp = ll_set_opt("fscontext", s1, 1);
 863                 if (tmp)
 864                         goto next;
 865                 tmp = ll_set_opt("defcontext", s1, 1);
 866                 if (tmp)
 867                         goto next;
 868                 tmp = ll_set_opt("rootcontext", s1, 1);
 869                 if (tmp)
 870                         goto next;
 871                 tmp = ll_set_opt("user_fid2path", s1, LL_SBI_USER_FID2PATH);
 872                 if (tmp) {
 873                         *flags |= tmp;
 874                         goto next;
 875                 }
 876                 tmp = ll_set_opt("nouser_fid2path", s1, LL_SBI_USER_FID2PATH);
 877                 if (tmp) {
 878                         *flags &= ~tmp;
 879                         goto next;
 880                 }
 881
 882                 tmp = ll_set_opt("checksum", s1, LL_SBI_CHECKSUM);
 883                 if (tmp) {
 884                         *flags |= tmp;
 885                         sbi->ll_checksum_set = 1;
 886                         goto next;
 887                 }
 888                 tmp = ll_set_opt("nochecksum", s1, LL_SBI_CHECKSUM);
 889                 if (tmp) {
 890                         *flags &= ~tmp;
 891                         sbi->ll_checksum_set = 1;
 892                         goto next;
 893                 }
 894                 tmp = ll_set_opt("lruresize", s1, LL_SBI_LRU_RESIZE);
 895                 if (tmp) {
 896                         *flags |= tmp;
 897                         goto next;
 898                 }
 899                 tmp = ll_set_opt("nolruresize", s1, LL_SBI_LRU_RESIZE);
 900                 if (tmp) {
 901                         *flags &= ~tmp;
 902                         goto next;
 903                 }
 904                 tmp = ll_set_opt("lazystatfs", s1, LL_SBI_LAZYSTATFS);
 905                 if (tmp) {
 906                         *flags |= tmp;
 907                         goto next;
 908                 }
 909                 tmp = ll_set_opt("nolazystatfs", s1, LL_SBI_LAZYSTATFS);
 910                 if (tmp) {
 911                         *flags &= ~tmp;
 912                         goto next;
 913                 }
 914                 tmp = ll_set_opt("32bitapi", s1, LL_SBI_32BIT_API);
 915                 if (tmp) {
 916                         *flags |= tmp;
 917                         goto next;
 918                 }
 919                 tmp = ll_set_opt("verbose", s1, LL_SBI_VERBOSE);
 920                 if (tmp) {
 921                         *flags |= tmp;
 922                         goto next;
 923                 }
 924                 tmp = ll_set_opt("noverbose", s1, LL_SBI_VERBOSE);
 925                 if (tmp) {
 926                         *flags &= ~tmp;
 927                         goto next;
 928                 }
 929                 tmp = ll_set_opt("always_ping", s1, LL_SBI_ALWAYS_PING);
 930                 if (tmp) {
 931                         *flags |= tmp;
 932                         goto next;
 933                 }
 934                 LCONSOLE_ERROR_MSG(0x152, "Unknown option '%s', won't mount.\n",
 935                                    s1);
 936                 RETURN(-EINVAL);
 937
 938 next:
 939                 /* Find next opt */
 940                 s2 = strchr(s1, ',');
 941                 if (s2 == NULL)
 942                         break;
 943                 s1 = s2 + 1;
 944         }
 945         RETURN(0);
 946 }
 947
 948 void ll_lli_init(struct ll_inode_info *lli)
 949 {
 950         lli->lli_inode_magic = LLI_INODE_MAGIC;
 951         lli->lli_flags = 0;
 952         spin_lock_init(&lli->lli_lock);
 953         lli->lli_posix_acl = NULL;
 954         /* Do not set lli_fid, it has been initialized already. */
 955         fid_zero(&lli->lli_pfid);
 956         lli->lli_mds_read_och = NULL;
 957         lli->lli_mds_write_och = NULL;
 958         lli->lli_mds_exec_och = NULL;
 959         lli->lli_open_fd_read_count = 0;
 960         lli->lli_open_fd_write_count = 0;
 961         lli->lli_open_fd_exec_count = 0;
 962         mutex_init(&lli->lli_och_mutex);
 963         spin_lock_init(&lli->lli_agl_lock);
 964         spin_lock_init(&lli->lli_layout_lock);
 965         ll_layout_version_set(lli, CL_LAYOUT_GEN_NONE);
 966         lli->lli_clob = NULL;
 967
 968         init_rwsem(&lli->lli_xattrs_list_rwsem);
 969         mutex_init(&lli->lli_xattrs_enq_lock);
 970
 971         LASSERT(lli->lli_vfs_inode.i_mode != 0);
 972         if (S_ISDIR(lli->lli_vfs_inode.i_mode)) {
 973                 mutex_init(&lli->lli_readdir_mutex);
 974                 lli->lli_opendir_key = NULL;
 975                 lli->lli_sai = NULL;
 976                 spin_lock_init(&lli->lli_sa_lock);
 977                 lli->lli_opendir_pid = 0;
 978                 lli->lli_sa_enabled = 0;
 979                 init_rwsem(&lli->lli_lsm_sem);
 980         } else {
 981                 mutex_init(&lli->lli_size_mutex);
 982                 lli->lli_symlink_name = NULL;
 983                 init_rwsem(&lli->lli_trunc_sem);
 984                 range_lock_tree_init(&lli->lli_write_tree);
 985                 init_rwsem(&lli->lli_glimpse_sem);
 986                 lli->lli_glimpse_time = ktime_set(0, 0);
 987                 INIT_LIST_HEAD(&lli->lli_agl_list);
 988                 lli->lli_agl_index = 0;
 989                 lli->lli_async_rc = 0;
 990                 spin_lock_init(&lli->lli_heat_lock);
 991                 obd_heat_clear(lli->lli_heat_instances, OBD_HEAT_COUNT);
 992                 lli->lli_heat_flags = 0;
 993                 mutex_init(&lli->lli_pcc_lock);
 994                 lli->lli_pcc_state = PCC_STATE_FL_NONE;
 995                 lli->lli_pcc_inode = NULL;
 996         }
 997         mutex_init(&lli->lli_layout_mutex);
 998         memset(lli->lli_jobid, 0, sizeof(lli->lli_jobid));
 999 }
1000
1001 #define MAX_STRING_SIZE 128
1002
1003 #ifndef HAVE_SUPER_SETUP_BDI_NAME
1004
1005 #define LSI_BDI_INITIALIZED     0x00400000
1006
1007 #ifndef HAVE_BDI_CAP_MAP_COPY
1008 # define BDI_CAP_MAP_COPY       0
1009 #endif
1010
1011 static int super_setup_bdi_name(struct super_block *sb, char *fmt, ...)
1012 {
1013         struct  lustre_sb_info *lsi = s2lsi(sb);
1014         char buf[MAX_STRING_SIZE];
1015         va_list args;
1016         int err;
1017
1018         err = bdi_init(&lsi->lsi_bdi);
1019         if (err)
1020                 return err;
1021
1022         lsi->lsi_flags |= LSI_BDI_INITIALIZED;
1023         lsi->lsi_bdi.capabilities = BDI_CAP_MAP_COPY;
1024         lsi->lsi_bdi.name = "lustre";
1025         va_start(args, fmt);
1026         vsnprintf(buf, MAX_STRING_SIZE, fmt, args);
1027         va_end(args);
1028         err = bdi_register(&lsi->lsi_bdi, NULL, "%s", buf);
1029         va_end(args);
1030         if (!err)
1031                 sb->s_bdi = &lsi->lsi_bdi;
1032
1033         return err;
1034 }
1035 #endif /* !HAVE_SUPER_SETUP_BDI_NAME */
1036
1037 int ll_fill_super(struct super_block *sb, struct vfsmount *mnt)
1038 {
1039         struct  lustre_profile *lprof = NULL;
1040         struct  lustre_sb_info *lsi = s2lsi(sb);
1041         struct  ll_sb_info *sbi = NULL;
1042         char    *dt = NULL, *md = NULL;
1043         char    *profilenm = get_profile_name(sb);
1044         struct config_llog_instance *cfg;
1045         /* %p for void* in printf needs 16+2 characters: 0xffffffffffffffff */
1046         const int instlen = 16 + 2;
1047         unsigned long cfg_instance = ll_get_cfg_instance(sb);
1048         char name[MAX_STRING_SIZE];
1049         int md_len = 0;
1050         int dt_len = 0;
1051         uuid_t uuid;
1052         char *ptr;
1053         int len;
1054         int err;
1055
1056         ENTRY;
1057         /* for ASLR, to map between cfg_instance and hashed ptr */
1058         CDEBUG(D_VFSTRACE, "VFS Op: cfg_instance %s-%016lx (sb %p)\n",
1059                profilenm, cfg_instance, sb);
1060
1061         try_module_get(THIS_MODULE);
1062
1063         OBD_ALLOC_PTR(cfg);
1064         if (cfg == NULL)
1065                 GOTO(out_free_cfg, err = -ENOMEM);
1066
1067         /* client additional sb info */
1068         lsi->lsi_llsbi = sbi = ll_init_sbi();
1069         if (IS_ERR(sbi))
1070                 GOTO(out_free_cfg, err = PTR_ERR(sbi));
1071
1072         err = ll_options(lsi->lsi_lmd->lmd_opts, sbi);
1073         if (err)
1074                 GOTO(out_free_cfg, err);
1075
1076 #ifndef HAVE_DCACHE_LOCK
1077         /* kernel >= 2.6.38 store dentry operations in sb->s_d_op. */
1078         sb->s_d_op = &ll_d_ops;
1079 #endif
1080         /* UUID handling */
1081         generate_random_uuid(uuid.b);
1082         snprintf(sbi->ll_sb_uuid.uuid, UUID_SIZE, "%pU", uuid.b);
1083
1084         CDEBUG(D_CONFIG, "llite sb uuid: %s\n", sbi->ll_sb_uuid.uuid);
1085
1086         /* Get fsname */
1087         len = strlen(profilenm);
1088         ptr = strrchr(profilenm, '-');
1089         if (ptr && (strcmp(ptr, "-client") == 0))
1090                 len -= 7;
1091
1092         if (len > LUSTRE_MAXFSNAME) {
1093                 if (unlikely(len >= MAX_STRING_SIZE))
1094                         len = MAX_STRING_SIZE - 1;
1095                 strncpy(name, profilenm, len);
1096                 name[len] = '\0';
1097                 err = -ENAMETOOLONG;
1098                 CERROR("%s: fsname longer than %u characters: rc = %d\n",
1099                        name, LUSTRE_MAXFSNAME, err);
1100                 GOTO(out_free_cfg, err);
1101         }
1102         strncpy(sbi->ll_fsname, profilenm, len);
1103         sbi->ll_fsname[len] = '\0';
1104
1105         /* Mount info */
1106         snprintf(name, sizeof(name), "%.*s-%016lx", len,
1107                  profilenm, cfg_instance);
1108
1109         err = super_setup_bdi_name(sb, "%s", name);
1110         if (err)
1111                 GOTO(out_free_cfg, err);
1112
1113         /* Call ll_debugfs_register_super() before lustre_process_log()
1114          * so that "llite.*.*" params can be processed correctly.
1115          */
1116         err = ll_debugfs_register_super(sb, name);
1117         if (err < 0) {
1118                 CERROR("%s: could not register mountpoint in llite: rc = %d\n",
1119                        sbi->ll_fsname, err);
1120                 err = 0;
1121         }
1122
1123         /* The cfg_instance is a value unique to this super, in case some
1124          * joker tries to mount the same fs at two mount points.
1125          */
1126         cfg->cfg_instance = cfg_instance;
1127         cfg->cfg_uuid = lsi->lsi_llsbi->ll_sb_uuid;
1128         cfg->cfg_callback = class_config_llog_handler;
1129         cfg->cfg_sub_clds = CONFIG_SUB_CLIENT;
1130         /* set up client obds */
1131         err = lustre_process_log(sb, profilenm, cfg);
1132         if (err < 0)
1133                 GOTO(out_debugfs, err);
1134
1135         /* Profile set with LCFG_MOUNTOPT so we can find our mdc and osc obds */
1136         lprof = class_get_profile(profilenm);
1137         if (lprof == NULL) {
1138                 LCONSOLE_ERROR_MSG(0x156, "The client profile '%s' could not be"
1139                                    " read from the MGS.  Does that filesystem "
1140                                    "exist?\n", profilenm);
1141                 GOTO(out_debugfs, err = -EINVAL);
1142         }
1143         CDEBUG(D_CONFIG, "Found profile %s: mdc=%s osc=%s\n", profilenm,
1144                lprof->lp_md, lprof->lp_dt);
1145
1146         dt_len = strlen(lprof->lp_dt) + instlen + 2;
1147         OBD_ALLOC(dt, dt_len);
1148         if (!dt)
1149                 GOTO(out_profile, err = -ENOMEM);
1150         snprintf(dt, dt_len - 1, "%s-%016lx", lprof->lp_dt, cfg_instance);
1151
1152         md_len = strlen(lprof->lp_md) + instlen + 2;
1153         OBD_ALLOC(md, md_len);
1154         if (!md)
1155                 GOTO(out_free_dt, err = -ENOMEM);
1156         snprintf(md, md_len - 1, "%s-%016lx", lprof->lp_md, cfg_instance);
1157
1158         /* connections, registrations, sb setup */
1159         err = client_common_fill_super(sb, md, dt, mnt);
1160         if (err < 0)
1161                 GOTO(out_free_md, err);
1162
1163         sbi->ll_client_common_fill_super_succeeded = 1;
1164
1165 out_free_md:
1166         if (md)
1167                 OBD_FREE(md, md_len);
1168 out_free_dt:
1169         if (dt)
1170                 OBD_FREE(dt, dt_len);
1171 out_profile:
1172         if (lprof)
1173                 class_put_profile(lprof);
1174 out_debugfs:
1175         if (err < 0)
1176                 ll_debugfs_unregister_super(sb);
1177 out_free_cfg:
1178         if (cfg)
1179                 OBD_FREE_PTR(cfg);
1180
1181         if (err)
1182                 ll_put_super(sb);
1183         else if (sbi->ll_flags & LL_SBI_VERBOSE)
1184                 LCONSOLE_WARN("Mounted %s\n", profilenm);
1185         RETURN(err);
1186 } /* ll_fill_super */
1187
1188 void ll_put_super(struct super_block *sb)
1189 {
1190         struct config_llog_instance cfg, params_cfg;
1191         struct obd_device *obd;
1192         struct lustre_sb_info *lsi = s2lsi(sb);
1193         struct ll_sb_info *sbi = ll_s2sbi(sb);
1194         char *profilenm = get_profile_name(sb);
1195         unsigned long cfg_instance = ll_get_cfg_instance(sb);
1196         long ccc_count;
1197         int next, force = 1, rc = 0;
1198         ENTRY;
1199
1200         if (IS_ERR(sbi))
1201                 GOTO(out_no_sbi, 0);
1202
1203         /* Should replace instance_id with something better for ASLR */
1204         CDEBUG(D_VFSTRACE, "VFS Op: cfg_instance %s-%016lx (sb %p)\n",
1205                profilenm, cfg_instance, sb);
1206
1207         cfg.cfg_instance = cfg_instance;
1208         lustre_end_log(sb, profilenm, &cfg);
1209
1210         params_cfg.cfg_instance = cfg_instance;
1211         lustre_end_log(sb, PARAMS_FILENAME, &params_cfg);
1212
1213         if (sbi->ll_md_exp) {
1214                 obd = class_exp2obd(sbi->ll_md_exp);
1215                 if (obd)
1216                         force = obd->obd_force;
1217         }
1218
1219         /* Wait for unstable pages to be committed to stable storage */
1220         if (force == 0) {
1221                 struct l_wait_info lwi = LWI_INTR(LWI_ON_SIGNAL_NOOP, NULL);
1222                 rc = l_wait_event(sbi->ll_cache->ccc_unstable_waitq,
1223                         atomic_long_read(&sbi->ll_cache->ccc_unstable_nr) == 0,
1224                         &lwi);
1225         }
1226
1227         ccc_count = atomic_long_read(&sbi->ll_cache->ccc_unstable_nr);
1228         if (force == 0 && rc != -EINTR)
1229                 LASSERTF(ccc_count == 0, "count: %li\n", ccc_count);
1230
1231         /* We need to set force before the lov_disconnect in
1232            lustre_common_put_super, since l_d cleans up osc's as well. */
1233         if (force) {
1234                 next = 0;
1235                 while ((obd = class_devices_in_group(&sbi->ll_sb_uuid,
1236                                                      &next)) != NULL) {
1237                         obd->obd_force = force;
1238                 }
1239         }
1240
1241         if (sbi->ll_client_common_fill_super_succeeded) {
1242                 /* Only if client_common_fill_super succeeded */
1243                 client_common_put_super(sb);
1244         }
1245
1246         next = 0;
1247         while ((obd = class_devices_in_group(&sbi->ll_sb_uuid, &next)) !=NULL) {
1248                 class_manual_cleanup(obd);
1249         }
1250
1251         if (sbi->ll_flags & LL_SBI_VERBOSE)
1252                 LCONSOLE_WARN("Unmounted %s\n", profilenm ? profilenm : "");
1253
1254         if (profilenm)
1255                 class_del_profile(profilenm);
1256
1257 #ifndef HAVE_SUPER_SETUP_BDI_NAME
1258         if (lsi->lsi_flags & LSI_BDI_INITIALIZED) {
1259                 bdi_destroy(&lsi->lsi_bdi);
1260                 lsi->lsi_flags &= ~LSI_BDI_INITIALIZED;
1261         }
1262 #endif
1263
1264         ll_free_sbi(sb);
1265         lsi->lsi_llsbi = NULL;
1266 out_no_sbi:
1267         lustre_common_put_super(sb);
1268
1269         cl_env_cache_purge(~0);
1270
1271         module_put(THIS_MODULE);
1272
1273         EXIT;
1274 } /* client_put_super */
1275
1276 struct inode *ll_inode_from_resource_lock(struct ldlm_lock *lock)
1277 {
1278         struct inode *inode = NULL;
1279
1280         /* NOTE: we depend on atomic igrab() -bzzz */
1281         lock_res_and_lock(lock);
1282         if (lock->l_resource->lr_lvb_inode) {
1283                 struct ll_inode_info * lli;
1284                 lli = ll_i2info(lock->l_resource->lr_lvb_inode);
1285                 if (lli->lli_inode_magic == LLI_INODE_MAGIC) {
1286                         inode = igrab(lock->l_resource->lr_lvb_inode);
1287                 } else {
1288                         inode = lock->l_resource->lr_lvb_inode;
1289                         LDLM_DEBUG_LIMIT(inode->i_state & I_FREEING ?  D_INFO :
1290                                          D_WARNING, lock, "lr_lvb_inode %p is "
1291                                          "bogus: magic %08x",
1292                                          lock->l_resource->lr_lvb_inode,
1293                                          lli->lli_inode_magic);
1294                         inode = NULL;
1295                 }
1296         }
1297         unlock_res_and_lock(lock);
1298         return inode;
1299 }
1300
1301 void ll_dir_clear_lsm_md(struct inode *inode)
1302 {
1303         struct ll_inode_info *lli = ll_i2info(inode);
1304
1305         LASSERT(S_ISDIR(inode->i_mode));
1306
1307         if (lli->lli_lsm_md) {
1308                 lmv_free_memmd(lli->lli_lsm_md);
1309                 lli->lli_lsm_md = NULL;
1310         }
1311
1312         if (lli->lli_default_lsm_md) {
1313                 lmv_free_memmd(lli->lli_default_lsm_md);
1314                 lli->lli_default_lsm_md = NULL;
1315         }
1316 }
1317
1318 static struct inode *ll_iget_anon_dir(struct super_block *sb,
1319                                       const struct lu_fid *fid,
1320                                       struct lustre_md *md)
1321 {
1322         struct ll_sb_info       *sbi = ll_s2sbi(sb);
1323         struct mdt_body         *body = md->body;
1324         struct inode            *inode;
1325         ino_t                   ino;
1326         ENTRY;
1327
1328         ino = cl_fid_build_ino(fid, sbi->ll_flags & LL_SBI_32BIT_API);
1329         inode = iget_locked(sb, ino);
1330         if (inode == NULL) {
1331                 CERROR("%s: failed get simple inode "DFID": rc = -ENOENT\n",
1332                        sbi->ll_fsname, PFID(fid));
1333                 RETURN(ERR_PTR(-ENOENT));
1334         }
1335
1336         if (inode->i_state & I_NEW) {
1337                 struct ll_inode_info *lli = ll_i2info(inode);
1338                 struct lmv_stripe_md *lsm = md->lmv;
1339
1340                 inode->i_mode = (inode->i_mode & ~S_IFMT) |
1341                                 (body->mbo_mode & S_IFMT);
1342                 LASSERTF(S_ISDIR(inode->i_mode), "Not slave inode "DFID"\n",
1343                          PFID(fid));
1344
1345                 inode->i_mtime.tv_sec = 0;
1346                 inode->i_atime.tv_sec = 0;
1347                 inode->i_ctime.tv_sec = 0;
1348                 inode->i_rdev = 0;
1349
1350 #ifdef HAVE_BACKING_DEV_INFO
1351                 /* initializing backing dev info. */
1352                 inode->i_mapping->backing_dev_info =
1353                                                 &s2lsi(inode->i_sb)->lsi_bdi;
1354 #endif
1355                 inode->i_op = &ll_dir_inode_operations;
1356                 inode->i_fop = &ll_dir_operations;
1357                 lli->lli_fid = *fid;
1358                 ll_lli_init(lli);
1359
1360                 LASSERT(lsm != NULL);
1361                 /* master object FID */
1362                 lli->lli_pfid = body->mbo_fid1;
1363                 CDEBUG(D_INODE, "lli %p slave "DFID" master "DFID"\n",
1364                        lli, PFID(fid), PFID(&lli->lli_pfid));
1365                 unlock_new_inode(inode);
1366         }
1367
1368         RETURN(inode);
1369 }
1370
1371 static int ll_init_lsm_md(struct inode *inode, struct lustre_md *md)
1372 {
1373         struct lu_fid *fid;
1374         struct lmv_stripe_md *lsm = md->lmv;
1375         struct ll_inode_info *lli = ll_i2info(inode);
1376         int i;
1377
1378         LASSERT(lsm != NULL);
1379
1380         CDEBUG(D_INODE, "%s: "DFID" set dir layout:\n",
1381                ll_i2sbi(inode)->ll_fsname, PFID(&lli->lli_fid));
1382         lsm_md_dump(D_INODE, lsm);
1383
1384         if (!lmv_dir_striped(lsm))
1385                 goto out;
1386
1387         /* XXX sigh, this lsm_root initialization should be in
1388          * LMV layer, but it needs ll_iget right now, so we
1389          * put this here right now. */
1390         for (i = 0; i < lsm->lsm_md_stripe_count; i++) {
1391                 fid = &lsm->lsm_md_oinfo[i].lmo_fid;
1392                 LASSERT(lsm->lsm_md_oinfo[i].lmo_root == NULL);
1393
1394                 if (!fid_is_sane(fid))
1395                         continue;
1396
1397                 /* Unfortunately ll_iget will call ll_update_inode,
1398                  * where the initialization of slave inode is slightly
1399                  * different, so it reset lsm_md to NULL to avoid
1400                  * initializing lsm for slave inode. */
1401                 lsm->lsm_md_oinfo[i].lmo_root =
1402                                 ll_iget_anon_dir(inode->i_sb, fid, md);
1403                 if (IS_ERR(lsm->lsm_md_oinfo[i].lmo_root)) {
1404                         int rc = PTR_ERR(lsm->lsm_md_oinfo[i].lmo_root);
1405
1406                         lsm->lsm_md_oinfo[i].lmo_root = NULL;
1407                         while (i-- > 0) {
1408                                 iput(lsm->lsm_md_oinfo[i].lmo_root);
1409                                 lsm->lsm_md_oinfo[i].lmo_root = NULL;
1410                         }
1411                         return rc;
1412                 }
1413         }
1414 out:
1415         lli->lli_lsm_md = lsm;
1416
1417         return 0;
1418 }
1419
1420 static void ll_update_default_lsm_md(struct inode *inode, struct lustre_md *md)
1421 {
1422         struct ll_inode_info *lli = ll_i2info(inode);
1423
1424         if (!md->default_lmv) {
1425                 /* clear default lsm */
1426                 if (lli->lli_default_lsm_md) {
1427                         down_write(&lli->lli_lsm_sem);
1428                         if (lli->lli_default_lsm_md) {
1429                                 lmv_free_memmd(lli->lli_default_lsm_md);
1430                                 lli->lli_default_lsm_md = NULL;
1431                         }
1432                         up_write(&lli->lli_lsm_sem);
1433                 }
1434         } else if (lli->lli_default_lsm_md) {
1435                 /* update default lsm if it changes */
1436                 down_read(&lli->lli_lsm_sem);
1437                 if (lli->lli_default_lsm_md &&
1438                     !lsm_md_eq(lli->lli_default_lsm_md, md->default_lmv)) {
1439                         up_read(&lli->lli_lsm_sem);
1440                         down_write(&lli->lli_lsm_sem);
1441                         if (lli->lli_default_lsm_md)
1442                                 lmv_free_memmd(lli->lli_default_lsm_md);
1443                         lli->lli_default_lsm_md = md->default_lmv;
1444                         lsm_md_dump(D_INODE, md->default_lmv);
1445                         md->default_lmv = NULL;
1446                         up_write(&lli->lli_lsm_sem);
1447                 } else {
1448                         up_read(&lli->lli_lsm_sem);
1449                 }
1450         } else {
1451                 /* init default lsm */
1452                 down_write(&lli->lli_lsm_sem);
1453                 lli->lli_default_lsm_md = md->default_lmv;
1454                 lsm_md_dump(D_INODE, md->default_lmv);
1455                 md->default_lmv = NULL;
1456                 up_write(&lli->lli_lsm_sem);
1457         }
1458 }
1459
1460 static int ll_update_lsm_md(struct inode *inode, struct lustre_md *md)
1461 {
1462         struct ll_inode_info *lli = ll_i2info(inode);
1463         struct lmv_stripe_md *lsm = md->lmv;
1464         int rc = 0;
1465
1466         ENTRY;
1467
1468         LASSERT(S_ISDIR(inode->i_mode));
1469         CDEBUG(D_INODE, "update lsm %p of "DFID"\n", lli->lli_lsm_md,
1470                PFID(ll_inode2fid(inode)));
1471
1472         /* update default LMV */
1473         if (md->default_lmv)
1474                 ll_update_default_lsm_md(inode, md);
1475
1476         /*
1477          * no striped information from request, lustre_md from req does not
1478          * include stripeEA, see ll_md_setattr()
1479          */
1480         if (!lsm)
1481                 RETURN(0);
1482
1483         /*
1484          * normally dir layout doesn't change, only take read lock to check
1485          * that to avoid blocking other MD operations.
1486          */
1487         if (lli->lli_lsm_md)
1488                 down_read(&lli->lli_lsm_sem);
1489         else
1490                 down_write(&lli->lli_lsm_sem);
1491
1492         /*
1493          * if dir layout mismatch, check whether version is increased, which
1494          * means layout is changed, this happens in dir migration and lfsck.
1495          *
1496          * foreign LMV should not change.
1497          */
1498         if (lli->lli_lsm_md && !lsm_md_eq(lli->lli_lsm_md, lsm)) {
1499                 if (lmv_dir_striped(lli->lli_lsm_md) &&
1500                     lsm->lsm_md_layout_version <=
1501                     lli->lli_lsm_md->lsm_md_layout_version) {
1502                         CERROR("%s: "DFID" dir layout mismatch:\n",
1503                                ll_i2sbi(inode)->ll_fsname,
1504                                PFID(&lli->lli_fid));
1505                         lsm_md_dump(D_ERROR, lli->lli_lsm_md);
1506                         lsm_md_dump(D_ERROR, lsm);
1507                         GOTO(unlock, rc = -EINVAL);
1508                 }
1509
1510                 /* layout changed, switch to write lock */
1511                 up_read(&lli->lli_lsm_sem);
1512                 down_write(&lli->lli_lsm_sem);
1513                 ll_dir_clear_lsm_md(inode);
1514         }
1515
1516         /* set directory layout */
1517         if (!lli->lli_lsm_md) {
1518                 struct cl_attr  *attr;
1519
1520                 rc = ll_init_lsm_md(inode, md);
1521                 up_write(&lli->lli_lsm_sem);
1522                 if (rc != 0)
1523                         RETURN(rc);
1524
1525                 /* set md->lmv to NULL, so the following free lustre_md
1526                  * will not free this lsm */
1527                 md->lmv = NULL;
1528
1529                 /*
1530                  * md_merge_attr() may take long, since lsm is already set,
1531                  * switch to read lock.
1532                  */
1533                 down_read(&lli->lli_lsm_sem);
1534
1535                 if (!lmv_dir_striped(lli->lli_lsm_md))
1536                         GOTO(unlock, rc);
1537
1538                 OBD_ALLOC_PTR(attr);
1539                 if (attr == NULL)
1540                         GOTO(unlock, rc = -ENOMEM);
1541
1542                 /* validate the lsm */
1543                 rc = md_merge_attr(ll_i2mdexp(inode), lsm, attr,
1544                                    ll_md_blocking_ast);
1545                 if (rc != 0) {
1546                         OBD_FREE_PTR(attr);
1547                         GOTO(unlock, rc);
1548                 }
1549
1550                 if (md->body->mbo_valid & OBD_MD_FLNLINK)
1551                         md->body->mbo_nlink = attr->cat_nlink;
1552                 if (md->body->mbo_valid & OBD_MD_FLSIZE)
1553                         md->body->mbo_size = attr->cat_size;
1554                 if (md->body->mbo_valid & OBD_MD_FLATIME)
1555                         md->body->mbo_atime = attr->cat_atime;
1556                 if (md->body->mbo_valid & OBD_MD_FLCTIME)
1557                         md->body->mbo_ctime = attr->cat_ctime;
1558                 if (md->body->mbo_valid & OBD_MD_FLMTIME)
1559                         md->body->mbo_mtime = attr->cat_mtime;
1560
1561                 OBD_FREE_PTR(attr);
1562         }
1563 unlock:
1564         up_read(&lli->lli_lsm_sem);
1565
1566         RETURN(rc);
1567 }
1568
1569 void ll_clear_inode(struct inode *inode)
1570 {
1571         struct ll_inode_info *lli = ll_i2info(inode);
1572         struct ll_sb_info *sbi = ll_i2sbi(inode);
1573
1574         ENTRY;
1575
1576         CDEBUG(D_VFSTRACE, "VFS Op:inode="DFID"(%p)\n",
1577                PFID(ll_inode2fid(inode)), inode);
1578
1579         if (S_ISDIR(inode->i_mode)) {
1580                 /* these should have been cleared in ll_file_release */
1581                 LASSERT(lli->lli_opendir_key == NULL);
1582                 LASSERT(lli->lli_sai == NULL);
1583                 LASSERT(lli->lli_opendir_pid == 0);
1584         } else {
1585                 pcc_inode_free(inode);
1586         }
1587
1588         md_null_inode(sbi->ll_md_exp, ll_inode2fid(inode));
1589
1590         LASSERT(!lli->lli_open_fd_write_count);
1591         LASSERT(!lli->lli_open_fd_read_count);
1592         LASSERT(!lli->lli_open_fd_exec_count);
1593
1594         if (lli->lli_mds_write_och)
1595                 ll_md_real_close(inode, FMODE_WRITE);
1596         if (lli->lli_mds_exec_och)
1597                 ll_md_real_close(inode, FMODE_EXEC);
1598         if (lli->lli_mds_read_och)
1599                 ll_md_real_close(inode, FMODE_READ);
1600
1601         if (S_ISLNK(inode->i_mode) && lli->lli_symlink_name) {
1602                 OBD_FREE(lli->lli_symlink_name,
1603                          strlen(lli->lli_symlink_name) + 1);
1604                 lli->lli_symlink_name = NULL;
1605         }
1606
1607         ll_xattr_cache_destroy(inode);
1608
1609 #ifdef CONFIG_FS_POSIX_ACL
1610         forget_all_cached_acls(inode);
1611         if (lli->lli_posix_acl) {
1612                 posix_acl_release(lli->lli_posix_acl);
1613                 lli->lli_posix_acl = NULL;
1614         }
1615 #endif
1616         lli->lli_inode_magic = LLI_INODE_DEAD;
1617
1618         if (S_ISDIR(inode->i_mode))
1619                 ll_dir_clear_lsm_md(inode);
1620         else if (S_ISREG(inode->i_mode) && !is_bad_inode(inode))
1621                 LASSERT(list_empty(&lli->lli_agl_list));
1622
1623         /*
1624          * XXX This has to be done before lsm is freed below, because
1625          * cl_object still uses inode lsm.
1626          */
1627         cl_inode_fini(inode);
1628
1629         EXIT;
1630 }
1631
1632 static int ll_md_setattr(struct dentry *dentry, struct md_op_data *op_data)
1633 {
1634         struct lustre_md md;
1635         struct inode *inode = dentry->d_inode;
1636         struct ll_sb_info *sbi = ll_i2sbi(inode);
1637         struct ptlrpc_request *request = NULL;
1638         int rc, ia_valid;
1639         ENTRY;
1640
1641         op_data = ll_prep_md_op_data(op_data, inode, NULL, NULL, 0, 0,
1642                                      LUSTRE_OPC_ANY, NULL);
1643         if (IS_ERR(op_data))
1644                 RETURN(PTR_ERR(op_data));
1645
1646         rc = md_setattr(sbi->ll_md_exp, op_data, NULL, 0, &request);
1647         if (rc) {
1648                 ptlrpc_req_finished(request);
1649                 if (rc == -ENOENT) {
1650                         clear_nlink(inode);
1651                         /* Unlinked special device node? Or just a race?
1652                          * Pretend we done everything. */
1653                         if (!S_ISREG(inode->i_mode) &&
1654                             !S_ISDIR(inode->i_mode)) {
1655                                 ia_valid = op_data->op_attr.ia_valid;
1656                                 op_data->op_attr.ia_valid &= ~TIMES_SET_FLAGS;
1657                                 rc = simple_setattr(dentry, &op_data->op_attr);
1658                                 op_data->op_attr.ia_valid = ia_valid;
1659                         }
1660                 } else if (rc != -EPERM && rc != -EACCES && rc != -ETXTBSY) {
1661                         CERROR("md_setattr fails: rc = %d\n", rc);
1662                 }
1663                 RETURN(rc);
1664         }
1665
1666         rc = md_get_lustre_md(sbi->ll_md_exp, request, sbi->ll_dt_exp,
1667                               sbi->ll_md_exp, &md);
1668         if (rc) {
1669                 ptlrpc_req_finished(request);
1670                 RETURN(rc);
1671         }
1672
1673         ia_valid = op_data->op_attr.ia_valid;
1674         /* inode size will be in ll_setattr_ost, can't do it now since dirty
1675          * cache is not cleared yet. */
1676         op_data->op_attr.ia_valid &= ~(TIMES_SET_FLAGS | ATTR_SIZE);
1677         if (S_ISREG(inode->i_mode))
1678                 inode_lock(inode);
1679         rc = simple_setattr(dentry, &op_data->op_attr);
1680         if (S_ISREG(inode->i_mode))
1681                 inode_unlock(inode);
1682         op_data->op_attr.ia_valid = ia_valid;
1683
1684         rc = ll_update_inode(inode, &md);
1685         ptlrpc_req_finished(request);
1686
1687         RETURN(rc);
1688 }
1689
1690 /* If this inode has objects allocated to it (lsm != NULL), then the OST
1691  * object(s) determine the file size and mtime.  Otherwise, the MDS will
1692  * keep these values until such a time that objects are allocated for it.
1693  * We do the MDS operations first, as it is checking permissions for us.
1694  * We don't to the MDS RPC if there is nothing that we want to store there,
1695  * otherwise there is no harm in updating mtime/atime on the MDS if we are
1696  * going to do an RPC anyways.
1697  *
1698  * If we are doing a truncate, we will send the mtime and ctime updates
1699  * to the OST with the punch RPC, otherwise we do an explicit setattr RPC.
1700  * I don't believe it is possible to get e.g. ATTR_MTIME_SET and ATTR_SIZE
1701  * at the same time.
1702  *
1703  * In case of HSMimport, we only set attr on MDS.
1704  */
1705 int ll_setattr_raw(struct dentry *dentry, struct iattr *attr,
1706                    enum op_xvalid xvalid, bool hsm_import)
1707 {
1708         struct inode *inode = dentry->d_inode;
1709         struct ll_inode_info *lli = ll_i2info(inode);
1710         struct md_op_data *op_data = NULL;
1711         int rc = 0;
1712
1713         ENTRY;
1714
1715         CDEBUG(D_VFSTRACE, "%s: setattr inode "DFID"(%p) from %llu to %llu, "
1716                "valid %x, hsm_import %d\n",
1717                ll_i2sbi(inode)->ll_fsname, PFID(&lli->lli_fid),
1718                inode, i_size_read(inode), attr->ia_size, attr->ia_valid,
1719                hsm_import);
1720
1721         if (attr->ia_valid & ATTR_SIZE) {
1722                 /* Check new size against VFS/VM file size limit and rlimit */
1723                 rc = inode_newsize_ok(inode, attr->ia_size);
1724                 if (rc)
1725                         RETURN(rc);
1726
1727                 /* The maximum Lustre file size is variable, based on the
1728                  * OST maximum object size and number of stripes.  This
1729                  * needs another check in addition to the VFS check above. */
1730                 if (attr->ia_size > ll_file_maxbytes(inode)) {
1731                         CDEBUG(D_INODE,"file "DFID" too large %llu > %llu\n",
1732                                PFID(&lli->lli_fid), attr->ia_size,
1733                                ll_file_maxbytes(inode));
1734                         RETURN(-EFBIG);
1735                 }
1736
1737                 attr->ia_valid |= ATTR_MTIME | ATTR_CTIME;
1738         }
1739
1740         /* POSIX: check before ATTR_*TIME_SET set (from inode_change_ok) */
1741         if (attr->ia_valid & TIMES_SET_FLAGS) {
1742                 if ((!uid_eq(current_fsuid(), inode->i_uid)) &&
1743                     !cfs_capable(CFS_CAP_FOWNER))
1744                         RETURN(-EPERM);
1745         }
1746
1747         /* We mark all of the fields "set" so MDS/OST does not re-set them */
1748         if (!(xvalid & OP_XVALID_CTIME_SET) &&
1749              (attr->ia_valid & ATTR_CTIME)) {
1750                 attr->ia_ctime = current_time(inode);
1751                 xvalid |= OP_XVALID_CTIME_SET;
1752         }
1753         if (!(attr->ia_valid & ATTR_ATIME_SET) &&
1754             (attr->ia_valid & ATTR_ATIME)) {
1755                 attr->ia_atime = current_time(inode);
1756                 attr->ia_valid |= ATTR_ATIME_SET;
1757         }
1758         if (!(attr->ia_valid & ATTR_MTIME_SET) &&
1759             (attr->ia_valid & ATTR_MTIME)) {
1760                 attr->ia_mtime = current_time(inode);
1761                 attr->ia_valid |= ATTR_MTIME_SET;
1762         }
1763
1764         if (attr->ia_valid & (ATTR_MTIME | ATTR_CTIME))
1765                 CDEBUG(D_INODE, "setting mtime %lld, ctime %lld, now = %lld\n",
1766                        (s64)attr->ia_mtime.tv_sec, (s64)attr->ia_ctime.tv_sec,
1767                        ktime_get_real_seconds());
1768
1769         if (S_ISREG(inode->i_mode)) {
1770                 if (attr->ia_valid & ATTR_SIZE)
1771                         inode_dio_write_done(inode);
1772                 inode_unlock(inode);
1773         }
1774
1775         /* We always do an MDS RPC, even if we're only changing the size;
1776          * only the MDS knows whether truncate() should fail with -ETXTBUSY */
1777
1778         OBD_ALLOC_PTR(op_data);
1779         if (op_data == NULL)
1780                 GOTO(out, rc = -ENOMEM);
1781
1782         if (!hsm_import && attr->ia_valid & ATTR_SIZE) {
1783                 /* If we are changing file size, file content is
1784                  * modified, flag it.
1785                  */
1786                 xvalid |= OP_XVALID_OWNEROVERRIDE;
1787                 op_data->op_bias |= MDS_DATA_MODIFIED;
1788                 ll_file_clear_flag(lli, LLIF_DATA_MODIFIED);
1789         }
1790
1791         if (attr->ia_valid & ATTR_FILE) {
1792                 struct ll_file_data *fd = LUSTRE_FPRIVATE(attr->ia_file);
1793
1794                 if (fd->fd_lease_och)
1795                         op_data->op_bias |= MDS_TRUNC_KEEP_LEASE;
1796         }
1797
1798         op_data->op_attr = *attr;
1799         op_data->op_xvalid = xvalid;
1800
1801         rc = ll_md_setattr(dentry, op_data);
1802         if (rc)
1803                 GOTO(out, rc);
1804
1805         if (!S_ISREG(inode->i_mode) || hsm_import)
1806                 GOTO(out, rc = 0);
1807
1808         if (attr->ia_valid & (ATTR_SIZE | ATTR_ATIME | ATTR_ATIME_SET |
1809                               ATTR_MTIME | ATTR_MTIME_SET | ATTR_CTIME) ||
1810             xvalid & OP_XVALID_CTIME_SET) {
1811                 bool cached = false;
1812
1813                 rc = pcc_inode_setattr(inode, attr, &cached);
1814                 if (cached) {
1815                         if (rc) {
1816                                 CERROR("%s: PCC inode "DFID" setattr failed: "
1817                                        "rc = %d\n",
1818                                        ll_i2sbi(inode)->ll_fsname,
1819                                        PFID(&lli->lli_fid), rc);
1820                                 GOTO(out, rc);
1821                         }
1822                 } else {
1823                         /* For truncate and utimes sending attributes to OSTs,
1824                          * setting mtime/atime to the past will be performed
1825                          * under PW [0:EOF] extent lock (new_size:EOF for
1826                          * truncate). It may seem excessive to send mtime/atime
1827                          * updates to OSTs when not setting times to past, but
1828                          * it is necessary due to possible time
1829                          * de-synchronization between MDT inode and OST objects
1830                          */
1831                         rc = cl_setattr_ost(lli->lli_clob, attr, xvalid, 0);
1832                 }
1833         }
1834
1835         /* If the file was restored, it needs to set dirty flag.
1836          *
1837          * We've already sent MDS_DATA_MODIFIED flag in
1838          * ll_md_setattr() for truncate. However, the MDT refuses to
1839          * set the HS_DIRTY flag on released files, so we have to set
1840          * it again if the file has been restored. Please check how
1841          * LLIF_DATA_MODIFIED is set in vvp_io_setattr_fini().
1842          *
1843          * Please notice that if the file is not released, the previous
1844          * MDS_DATA_MODIFIED has taken effect and usually
1845          * LLIF_DATA_MODIFIED is not set(see vvp_io_setattr_fini()).
1846          * This way we can save an RPC for common open + trunc
1847          * operation. */
1848         if (ll_file_test_and_clear_flag(lli, LLIF_DATA_MODIFIED)) {
1849                 struct hsm_state_set hss = {
1850                         .hss_valid = HSS_SETMASK,
1851                         .hss_setmask = HS_DIRTY,
1852                 };
1853                 int rc2;
1854
1855                 rc2 = ll_hsm_state_set(inode, &hss);
1856                 /* truncate and write can happen at the same time, so that
1857                  * the file can be set modified even though the file is not
1858                  * restored from released state, and ll_hsm_state_set() is
1859                  * not applicable for the file, and rc2 < 0 is normal in this
1860                  * case. */
1861                 if (rc2 < 0)
1862                         CDEBUG(D_INFO, DFID "HSM set dirty failed: rc2 = %d\n",
1863                                PFID(ll_inode2fid(inode)), rc2);
1864         }
1865
1866         EXIT;
1867 out:
1868         if (op_data != NULL)
1869                 ll_finish_md_op_data(op_data);
1870
1871         if (S_ISREG(inode->i_mode)) {
1872                 inode_lock(inode);
1873                 if ((attr->ia_valid & ATTR_SIZE) && !hsm_import)
1874                         inode_dio_wait(inode);
1875                 /* Once we've got the i_mutex, it's safe to set the S_NOSEC
1876                  * flag.  ll_update_inode (called from ll_md_setattr), clears
1877                  * inode flags, so there is a gap where S_NOSEC is not set.
1878                  * This can cause a writer to take the i_mutex unnecessarily,
1879                  * but this is safe to do and should be rare. */
1880                 inode_has_no_xattr(inode);
1881         }
1882
1883         ll_stats_ops_tally(ll_i2sbi(inode), (attr->ia_valid & ATTR_SIZE) ?
1884                         LPROC_LL_TRUNC : LPROC_LL_SETATTR, 1);
1885
1886         return rc;
1887 }
1888
1889 int ll_setattr(struct dentry *de, struct iattr *attr)
1890 {
1891         int mode = de->d_inode->i_mode;
1892         enum op_xvalid xvalid = 0;
1893
1894         if ((attr->ia_valid & (ATTR_CTIME|ATTR_SIZE|ATTR_MODE)) ==
1895                               (ATTR_CTIME|ATTR_SIZE|ATTR_MODE))
1896                 xvalid |= OP_XVALID_OWNEROVERRIDE;
1897
1898         if (((attr->ia_valid & (ATTR_MODE|ATTR_FORCE|ATTR_SIZE)) ==
1899                                (ATTR_SIZE|ATTR_MODE)) &&
1900             (((mode & S_ISUID) && !(attr->ia_mode & S_ISUID)) ||
1901              (((mode & (S_ISGID|S_IXGRP)) == (S_ISGID|S_IXGRP)) &&
1902               !(attr->ia_mode & S_ISGID))))
1903                 attr->ia_valid |= ATTR_FORCE;
1904
1905         if ((attr->ia_valid & ATTR_MODE) &&
1906             (mode & S_ISUID) &&
1907             !(attr->ia_mode & S_ISUID) &&
1908             !(attr->ia_valid & ATTR_KILL_SUID))
1909                 attr->ia_valid |= ATTR_KILL_SUID;
1910
1911         if ((attr->ia_valid & ATTR_MODE) &&
1912             ((mode & (S_ISGID|S_IXGRP)) == (S_ISGID|S_IXGRP)) &&
1913             !(attr->ia_mode & S_ISGID) &&
1914             !(attr->ia_valid & ATTR_KILL_SGID))
1915                 attr->ia_valid |= ATTR_KILL_SGID;
1916
1917         return ll_setattr_raw(de, attr, xvalid, false);
1918 }
1919
1920 int ll_statfs_internal(struct ll_sb_info *sbi, struct obd_statfs *osfs,
1921                        u32 flags)
1922 {
1923         struct obd_statfs obd_osfs = { 0 };
1924         time64_t max_age;
1925         int rc;
1926
1927         ENTRY;
1928         max_age = ktime_get_seconds() - OBD_STATFS_CACHE_SECONDS;
1929
1930         rc = obd_statfs(NULL, sbi->ll_md_exp, osfs, max_age, flags);
1931         if (rc)
1932                 RETURN(rc);
1933
1934         osfs->os_type = LL_SUPER_MAGIC;
1935
1936         CDEBUG(D_SUPER, "MDC blocks %llu/%llu objects %llu/%llu\n",
1937               osfs->os_bavail, osfs->os_blocks, osfs->os_ffree, osfs->os_files);
1938
1939         if (osfs->os_state & OS_STATE_SUM)
1940                 GOTO(out, rc);
1941
1942         if (sbi->ll_flags & LL_SBI_LAZYSTATFS)
1943                 flags |= OBD_STATFS_NODELAY;
1944
1945         rc = obd_statfs(NULL, sbi->ll_dt_exp, &obd_osfs, max_age, flags);
1946         if (rc) /* Possibly a filesystem with no OSTs.  Report MDT totals. */
1947                 GOTO(out, rc = 0);
1948
1949         CDEBUG(D_SUPER, "OSC blocks %llu/%llu objects %llu/%llu\n",
1950                obd_osfs.os_bavail, obd_osfs.os_blocks, obd_osfs.os_ffree,
1951                obd_osfs.os_files);
1952
1953         osfs->os_bsize = obd_osfs.os_bsize;
1954         osfs->os_blocks = obd_osfs.os_blocks;
1955         osfs->os_bfree = obd_osfs.os_bfree;
1956         osfs->os_bavail = obd_osfs.os_bavail;
1957
1958         /* If we have _some_ OSTs, but don't have as many free objects on the
1959          * OSTs as inodes on the MDTs, reduce the reported number of inodes
1960          * to compensate, so that the "inodes in use" number is correct.
1961          * This should be kept in sync with lod_statfs() behaviour.
1962          */
1963         if (obd_osfs.os_files && obd_osfs.os_ffree < osfs->os_ffree) {
1964                 osfs->os_files = (osfs->os_files - osfs->os_ffree) +
1965                                  obd_osfs.os_ffree;
1966                 osfs->os_ffree = obd_osfs.os_ffree;
1967         }
1968
1969 out:
1970         RETURN(rc);
1971 }
1972 int ll_statfs(struct dentry *de, struct kstatfs *sfs)
1973 {
1974         struct super_block *sb = de->d_sb;
1975         struct obd_statfs osfs;
1976         __u64 fsid = huge_encode_dev(sb->s_dev);
1977         int rc;
1978
1979         CDEBUG(D_VFSTRACE, "VFS Op: at %llu jiffies\n", get_jiffies_64());
1980         ll_stats_ops_tally(ll_s2sbi(sb), LPROC_LL_STAFS, 1);
1981
1982         /* Some amount of caching on the client is allowed */
1983         rc = ll_statfs_internal(ll_s2sbi(sb), &osfs, OBD_STATFS_SUM);
1984         if (rc)
1985                 return rc;
1986
1987         statfs_unpack(sfs, &osfs);
1988
1989         /* We need to downshift for all 32-bit kernels, because we can't
1990          * tell if the kernel is being called via sys_statfs64() or not.
1991          * Stop before overflowing f_bsize - in which case it is better
1992          * to just risk EOVERFLOW if caller is using old sys_statfs(). */
1993         if (sizeof(long) < 8) {
1994                 while (osfs.os_blocks > ~0UL && sfs->f_bsize < 0x40000000) {
1995                         sfs->f_bsize <<= 1;
1996
1997                         osfs.os_blocks >>= 1;
1998                         osfs.os_bfree >>= 1;
1999                         osfs.os_bavail >>= 1;
2000                 }
2001         }
2002
2003         sfs->f_blocks = osfs.os_blocks;
2004         sfs->f_bfree = osfs.os_bfree;
2005         sfs->f_bavail = osfs.os_bavail;
2006         sfs->f_fsid.val[0] = (__u32)fsid;
2007         sfs->f_fsid.val[1] = (__u32)(fsid >> 32);
2008         return 0;
2009 }
2010
2011 void ll_inode_size_lock(struct inode *inode)
2012 {
2013         struct ll_inode_info *lli;
2014
2015         LASSERT(!S_ISDIR(inode->i_mode));
2016
2017         lli = ll_i2info(inode);
2018         mutex_lock(&lli->lli_size_mutex);
2019 }
2020
2021 void ll_inode_size_unlock(struct inode *inode)
2022 {
2023         struct ll_inode_info *lli;
2024
2025         lli = ll_i2info(inode);
2026         mutex_unlock(&lli->lli_size_mutex);
2027 }
2028
2029 void ll_update_inode_flags(struct inode *inode, int ext_flags)
2030 {
2031         inode->i_flags = ll_ext_to_inode_flags(ext_flags);
2032         if (ext_flags & LUSTRE_PROJINHERIT_FL)
2033                 ll_file_set_flag(ll_i2info(inode), LLIF_PROJECT_INHERIT);
2034         else
2035                 ll_file_clear_flag(ll_i2info(inode), LLIF_PROJECT_INHERIT);
2036 }
2037
2038 int ll_update_inode(struct inode *inode, struct lustre_md *md)
2039 {
2040         struct ll_inode_info *lli = ll_i2info(inode);
2041         struct mdt_body *body = md->body;
2042         struct ll_sb_info *sbi = ll_i2sbi(inode);
2043         int rc = 0;
2044
2045         if (body->mbo_valid & OBD_MD_FLEASIZE) {
2046                 rc = cl_file_inode_init(inode, md);
2047                 if (rc)
2048                         return rc;
2049         }
2050
2051         if (S_ISDIR(inode->i_mode)) {
2052                 rc = ll_update_lsm_md(inode, md);
2053                 if (rc != 0)
2054                         return rc;
2055         }
2056
2057 #ifdef CONFIG_FS_POSIX_ACL
2058         if (body->mbo_valid & OBD_MD_FLACL) {
2059                 spin_lock(&lli->lli_lock);
2060                 if (lli->lli_posix_acl)
2061                         posix_acl_release(lli->lli_posix_acl);
2062                 lli->lli_posix_acl = md->posix_acl;
2063                 spin_unlock(&lli->lli_lock);
2064         }
2065 #endif
2066         inode->i_ino = cl_fid_build_ino(&body->mbo_fid1,
2067                                         sbi->ll_flags & LL_SBI_32BIT_API);
2068         inode->i_generation = cl_fid_build_gen(&body->mbo_fid1);
2069
2070         if (body->mbo_valid & OBD_MD_FLATIME) {
2071                 if (body->mbo_atime > inode->i_atime.tv_sec)
2072                         inode->i_atime.tv_sec = body->mbo_atime;
2073                 lli->lli_atime = body->mbo_atime;
2074         }
2075
2076         if (body->mbo_valid & OBD_MD_FLMTIME) {
2077                 if (body->mbo_mtime > inode->i_mtime.tv_sec) {
2078                         CDEBUG(D_INODE,
2079                                "setting ino %lu mtime from %lld to %llu\n",
2080                                inode->i_ino, (s64)inode->i_mtime.tv_sec,
2081                                body->mbo_mtime);
2082                         inode->i_mtime.tv_sec = body->mbo_mtime;
2083                 }
2084                 lli->lli_mtime = body->mbo_mtime;
2085         }
2086
2087         if (body->mbo_valid & OBD_MD_FLCTIME) {
2088                 if (body->mbo_ctime > inode->i_ctime.tv_sec)
2089                         inode->i_ctime.tv_sec = body->mbo_ctime;
2090                 lli->lli_ctime = body->mbo_ctime;
2091         }
2092
2093         /* Clear i_flags to remove S_NOSEC before permissions are updated */
2094         if (body->mbo_valid & OBD_MD_FLFLAGS)
2095                 ll_update_inode_flags(inode, body->mbo_flags);
2096         if (body->mbo_valid & OBD_MD_FLMODE)
2097                 inode->i_mode = (inode->i_mode & S_IFMT) |
2098                                 (body->mbo_mode & ~S_IFMT);
2099
2100         if (body->mbo_valid & OBD_MD_FLTYPE)
2101                 inode->i_mode = (inode->i_mode & ~S_IFMT) |
2102                                 (body->mbo_mode & S_IFMT);
2103
2104         LASSERT(inode->i_mode != 0);
2105         if (S_ISREG(inode->i_mode))
2106                 inode->i_blkbits = min(PTLRPC_MAX_BRW_BITS + 1,
2107                                        LL_MAX_BLKSIZE_BITS);
2108         else
2109                 inode->i_blkbits = inode->i_sb->s_blocksize_bits;
2110
2111         if (body->mbo_valid & OBD_MD_FLUID)
2112                 inode->i_uid = make_kuid(&init_user_ns, body->mbo_uid);
2113         if (body->mbo_valid & OBD_MD_FLGID)
2114                 inode->i_gid = make_kgid(&init_user_ns, body->mbo_gid);
2115         if (body->mbo_valid & OBD_MD_FLPROJID)
2116                 lli->lli_projid = body->mbo_projid;
2117         if (body->mbo_valid & OBD_MD_FLNLINK)
2118                 set_nlink(inode, body->mbo_nlink);
2119         if (body->mbo_valid & OBD_MD_FLRDEV)
2120                 inode->i_rdev = old_decode_dev(body->mbo_rdev);
2121
2122         if (body->mbo_valid & OBD_MD_FLID) {
2123                 /* FID shouldn't be changed! */
2124                 if (fid_is_sane(&lli->lli_fid)) {
2125                         LASSERTF(lu_fid_eq(&lli->lli_fid, &body->mbo_fid1),
2126                                  "Trying to change FID "DFID
2127                                  " to the "DFID", inode "DFID"(%p)\n",
2128                                  PFID(&lli->lli_fid), PFID(&body->mbo_fid1),
2129                                  PFID(ll_inode2fid(inode)), inode);
2130                 } else {
2131                         lli->lli_fid = body->mbo_fid1;
2132                 }
2133         }
2134
2135         LASSERT(fid_seq(&lli->lli_fid) != 0);
2136
2137         if (body->mbo_valid & OBD_MD_FLSIZE) {
2138                 i_size_write(inode, body->mbo_size);
2139
2140                 CDEBUG(D_VFSTRACE, "inode="DFID", updating i_size %llu\n",
2141                        PFID(ll_inode2fid(inode)),
2142                        (unsigned long long)body->mbo_size);
2143
2144                 if (body->mbo_valid & OBD_MD_FLBLOCKS)
2145                         inode->i_blocks = body->mbo_blocks;
2146         }
2147
2148         if (body->mbo_valid & OBD_MD_TSTATE) {
2149                 /* Set LLIF_FILE_RESTORING if restore ongoing and
2150                  * clear it when done to ensure to start again
2151                  * glimpsing updated attrs
2152                  */
2153                 if (body->mbo_t_state & MS_RESTORE)
2154                         ll_file_set_flag(lli, LLIF_FILE_RESTORING);
2155                 else
2156                         ll_file_clear_flag(lli, LLIF_FILE_RESTORING);
2157         }
2158
2159         return 0;
2160 }
2161
2162 int ll_read_inode2(struct inode *inode, void *opaque)
2163 {
2164         struct lustre_md *md = opaque;
2165         struct ll_inode_info *lli = ll_i2info(inode);
2166         int     rc;
2167         ENTRY;
2168
2169         CDEBUG(D_VFSTRACE, "VFS Op:inode="DFID"(%p)\n",
2170                PFID(&lli->lli_fid), inode);
2171
2172         /* Core attributes from the MDS first.  This is a new inode, and
2173          * the VFS doesn't zero times in the core inode so we have to do
2174          * it ourselves.  They will be overwritten by either MDS or OST
2175          * attributes - we just need to make sure they aren't newer.
2176          */
2177         inode->i_mtime.tv_sec = 0;
2178         inode->i_atime.tv_sec = 0;
2179         inode->i_ctime.tv_sec = 0;
2180         inode->i_rdev = 0;
2181         rc = ll_update_inode(inode, md);
2182         if (rc != 0)
2183                 RETURN(rc);
2184
2185         /* OIDEBUG(inode); */
2186
2187 #ifdef HAVE_BACKING_DEV_INFO
2188         /* initializing backing dev info. */
2189         inode->i_mapping->backing_dev_info = &s2lsi(inode->i_sb)->lsi_bdi;
2190 #endif
2191         if (S_ISREG(inode->i_mode)) {
2192                 struct ll_sb_info *sbi = ll_i2sbi(inode);
2193                 inode->i_op = &ll_file_inode_operations;
2194                 inode->i_fop = sbi->ll_fop;
2195                 inode->i_mapping->a_ops = (struct address_space_operations *)&ll_aops;
2196                 EXIT;
2197         } else if (S_ISDIR(inode->i_mode)) {
2198                 inode->i_op = &ll_dir_inode_operations;
2199                 inode->i_fop = &ll_dir_operations;
2200                 EXIT;
2201         } else if (S_ISLNK(inode->i_mode)) {
2202                 inode->i_op = &ll_fast_symlink_inode_operations;
2203                 EXIT;
2204         } else {
2205                 inode->i_op = &ll_special_inode_operations;
2206
2207                 init_special_inode(inode, inode->i_mode,
2208                                    inode->i_rdev);
2209
2210                 EXIT;
2211         }
2212
2213         return 0;
2214 }
2215
2216 void ll_delete_inode(struct inode *inode)
2217 {
2218         struct ll_inode_info *lli = ll_i2info(inode);
2219         struct address_space *mapping = &inode->i_data;
2220         unsigned long nrpages;
2221         ENTRY;
2222
2223         if (S_ISREG(inode->i_mode) && lli->lli_clob != NULL) {
2224                 /* It is last chance to write out dirty pages,
2225                  * otherwise we may lose data while umount.
2226                  *
2227                  * If i_nlink is 0 then just discard data. This is safe because
2228                  * local inode gets i_nlink 0 from server only for the last
2229                  * unlink, so that file is not opened somewhere else
2230                  */
2231                 cl_sync_file_range(inode, 0, OBD_OBJECT_EOF, inode->i_nlink ?
2232                                    CL_FSYNC_LOCAL : CL_FSYNC_DISCARD, 1);
2233         }
2234         truncate_inode_pages_final(mapping);
2235
2236         /* Workaround for LU-118: Note nrpages may not be totally updated when
2237          * truncate_inode_pages() returns, as there can be a page in the process
2238          * of deletion (inside __delete_from_page_cache()) in the specified
2239          * range. Thus mapping->nrpages can be non-zero when this function
2240          * returns even after truncation of the whole mapping.  Only do this if
2241          * npages isn't already zero.
2242          */
2243         nrpages = mapping->nrpages;
2244         if (nrpages) {
2245                 xa_lock_irq(&mapping->i_pages);
2246                 nrpages = mapping->nrpages;
2247                 xa_unlock_irq(&mapping->i_pages);
2248         } /* Workaround end */
2249
2250         LASSERTF(nrpages == 0, "%s: inode="DFID"(%p) nrpages=%lu, "
2251                  "see https://jira.whamcloud.com/browse/LU-118\n",
2252                  ll_i2sbi(inode)->ll_fsname,
2253                  PFID(ll_inode2fid(inode)), inode, nrpages);
2254
2255 #ifdef HAVE_SBOPS_EVICT_INODE
2256         ll_clear_inode(inode);
2257 #endif
2258         clear_inode(inode);
2259
2260         EXIT;
2261 }
2262
2263 int ll_iocontrol(struct inode *inode, struct file *file,
2264                  unsigned int cmd, unsigned long arg)
2265 {
2266         struct ll_sb_info *sbi = ll_i2sbi(inode);
2267         struct ptlrpc_request *req = NULL;
2268         int rc, flags = 0;
2269         ENTRY;
2270
2271         switch (cmd) {
2272         case FS_IOC_GETFLAGS: {
2273                 struct mdt_body *body;
2274                 struct md_op_data *op_data;
2275
2276                 op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL,
2277                                              0, 0, LUSTRE_OPC_ANY,
2278                                              NULL);
2279                 if (IS_ERR(op_data))
2280                         RETURN(PTR_ERR(op_data));
2281
2282                 op_data->op_valid = OBD_MD_FLFLAGS;
2283                 rc = md_getattr(sbi->ll_md_exp, op_data, &req);
2284                 ll_finish_md_op_data(op_data);
2285                 if (rc) {
2286                         CERROR("%s: failure inode "DFID": rc = %d\n",
2287                                sbi->ll_md_exp->exp_obd->obd_name,
2288                                PFID(ll_inode2fid(inode)), rc);
2289                         RETURN(-abs(rc));
2290                 }
2291
2292                 body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
2293
2294                 flags = body->mbo_flags;
2295
2296                 ptlrpc_req_finished(req);
2297
2298                 RETURN(put_user(flags, (int __user *)arg));
2299         }
2300         case FS_IOC_SETFLAGS: {
2301                 struct iattr *attr;
2302                 struct md_op_data *op_data;
2303                 struct cl_object *obj;
2304                 struct fsxattr fa = { 0 };
2305
2306                 if (get_user(flags, (int __user *)arg))
2307                         RETURN(-EFAULT);
2308
2309                 fa.fsx_projid = ll_i2info(inode)->lli_projid;
2310                 if (flags & LUSTRE_PROJINHERIT_FL)
2311                         fa.fsx_xflags = FS_XFLAG_PROJINHERIT;
2312
2313                 rc = ll_ioctl_check_project(inode, &fa);
2314                 if (rc)
2315                         RETURN(rc);
2316
2317                 op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL, 0, 0,
2318                                              LUSTRE_OPC_ANY, NULL);
2319                 if (IS_ERR(op_data))
2320                         RETURN(PTR_ERR(op_data));
2321
2322                 op_data->op_attr_flags = flags;
2323                 op_data->op_xvalid |= OP_XVALID_FLAGS;
2324                 rc = md_setattr(sbi->ll_md_exp, op_data, NULL, 0, &req);
2325                 ll_finish_md_op_data(op_data);
2326                 ptlrpc_req_finished(req);
2327                 if (rc)
2328                         RETURN(rc);
2329
2330                 ll_update_inode_flags(inode, flags);
2331
2332                 obj = ll_i2info(inode)->lli_clob;
2333                 if (obj == NULL)
2334                         RETURN(0);
2335
2336                 OBD_ALLOC_PTR(attr);
2337                 if (attr == NULL)
2338                         RETURN(-ENOMEM);
2339
2340                 rc = cl_setattr_ost(obj, attr, OP_XVALID_FLAGS, flags);
2341
2342                 OBD_FREE_PTR(attr);
2343                 RETURN(rc);
2344         }
2345         default:
2346                 RETURN(-ENOSYS);
2347         }
2348
2349         RETURN(0);
2350 }
2351
2352 int ll_flush_ctx(struct inode *inode)
2353 {
2354         struct ll_sb_info  *sbi = ll_i2sbi(inode);
2355
2356         CDEBUG(D_SEC, "flush context for user %d\n",
2357                from_kuid(&init_user_ns, current_uid()));
2358
2359         obd_set_info_async(NULL, sbi->ll_md_exp,
2360                            sizeof(KEY_FLUSH_CTX), KEY_FLUSH_CTX,
2361                            0, NULL, NULL);
2362         obd_set_info_async(NULL, sbi->ll_dt_exp,
2363                            sizeof(KEY_FLUSH_CTX), KEY_FLUSH_CTX,
2364                            0, NULL, NULL);
2365         return 0;
2366 }
2367
2368 /* umount -f client means force down, don't save state */
2369 void ll_umount_begin(struct super_block *sb)
2370 {
2371         struct ll_sb_info *sbi = ll_s2sbi(sb);
2372         struct obd_device *obd;
2373         struct obd_ioctl_data *ioc_data;
2374         struct l_wait_info lwi;
2375         wait_queue_head_t waitq;
2376         ENTRY;
2377
2378         CDEBUG(D_VFSTRACE, "VFS Op: superblock %p count %d active %d\n", sb,
2379                sb->s_count, atomic_read(&sb->s_active));
2380
2381         obd = class_exp2obd(sbi->ll_md_exp);
2382         if (obd == NULL) {
2383                 CERROR("Invalid MDC connection handle %#llx\n",
2384                        sbi->ll_md_exp->exp_handle.h_cookie);
2385                 EXIT;
2386                 return;
2387         }
2388         obd->obd_force = 1;
2389
2390         obd = class_exp2obd(sbi->ll_dt_exp);
2391         if (obd == NULL) {
2392                 CERROR("Invalid LOV connection handle %#llx\n",
2393                        sbi->ll_dt_exp->exp_handle.h_cookie);
2394                 EXIT;
2395                 return;
2396         }
2397         obd->obd_force = 1;
2398
2399         OBD_ALLOC_PTR(ioc_data);
2400         if (ioc_data) {
2401                 obd_iocontrol(IOC_OSC_SET_ACTIVE, sbi->ll_md_exp,
2402                               sizeof *ioc_data, ioc_data, NULL);
2403
2404                 obd_iocontrol(IOC_OSC_SET_ACTIVE, sbi->ll_dt_exp,
2405                               sizeof *ioc_data, ioc_data, NULL);
2406
2407                 OBD_FREE_PTR(ioc_data);
2408         }
2409
2410         /* Really, we'd like to wait until there are no requests outstanding,
2411          * and then continue.  For now, we just periodically checking for vfs
2412          * to decrement mnt_cnt and hope to finish it within 10sec.
2413          */
2414         init_waitqueue_head(&waitq);
2415         lwi = LWI_TIMEOUT_INTERVAL(cfs_time_seconds(10),
2416                                    cfs_time_seconds(1), NULL, NULL);
2417         l_wait_event(waitq, may_umount(sbi->ll_mnt.mnt), &lwi);
2418
2419         EXIT;
2420 }
2421
2422 int ll_remount_fs(struct super_block *sb, int *flags, char *data)
2423 {
2424         struct ll_sb_info *sbi = ll_s2sbi(sb);
2425         char *profilenm = get_profile_name(sb);
2426         int err;
2427         __u32 read_only;
2428
2429         if ((*flags & MS_RDONLY) != (sb->s_flags & MS_RDONLY)) {
2430                 read_only = *flags & MS_RDONLY;
2431                 err = obd_set_info_async(NULL, sbi->ll_md_exp,
2432                                          sizeof(KEY_READ_ONLY),
2433                                          KEY_READ_ONLY, sizeof(read_only),
2434                                          &read_only, NULL);
2435                 if (err) {
2436                         LCONSOLE_WARN("Failed to remount %s %s (%d)\n",
2437                                       profilenm, read_only ?
2438                                       "read-only" : "read-write", err);
2439                         return err;
2440                 }
2441
2442                 if (read_only)
2443                         sb->s_flags |= MS_RDONLY;
2444                 else
2445                         sb->s_flags &= ~MS_RDONLY;
2446
2447                 if (sbi->ll_flags & LL_SBI_VERBOSE)
2448                         LCONSOLE_WARN("Remounted %s %s\n", profilenm,
2449                                       read_only ?  "read-only" : "read-write");
2450         }
2451         return 0;
2452 }
2453
2454 /**
2455  * Cleanup the open handle that is cached on MDT-side.
2456  *
2457  * For open case, the client side open handling thread may hit error
2458  * after the MDT grant the open. Under such case, the client should
2459  * send close RPC to the MDT as cleanup; otherwise, the open handle
2460  * on the MDT will be leaked there until the client umount or evicted.
2461  *
2462  * In further, if someone unlinked the file, because the open handle
2463  * holds the reference on such file/object, then it will block the
2464  * subsequent threads that want to locate such object via FID.
2465  *
2466  * \param[in] sb        super block for this file-system
2467  * \param[in] open_req  pointer to the original open request
2468  */
2469 void ll_open_cleanup(struct super_block *sb, struct ptlrpc_request *open_req)
2470 {
2471         struct mdt_body                 *body;
2472         struct md_op_data               *op_data;
2473         struct ptlrpc_request           *close_req = NULL;
2474         struct obd_export               *exp       = ll_s2sbi(sb)->ll_md_exp;
2475         ENTRY;
2476
2477         body = req_capsule_server_get(&open_req->rq_pill, &RMF_MDT_BODY);
2478         OBD_ALLOC_PTR(op_data);
2479         if (op_data == NULL) {
2480                 CWARN("%s: cannot allocate op_data to release open handle for "
2481                       DFID"\n", ll_s2sbi(sb)->ll_fsname, PFID(&body->mbo_fid1));
2482
2483                 RETURN_EXIT;
2484         }
2485
2486         op_data->op_fid1 = body->mbo_fid1;
2487         op_data->op_open_handle = body->mbo_open_handle;
2488         op_data->op_mod_time = ktime_get_real_seconds();
2489         md_close(exp, op_data, NULL, &close_req);
2490         ptlrpc_req_finished(close_req);
2491         ll_finish_md_op_data(op_data);
2492
2493         EXIT;
2494 }
2495
2496 int ll_prep_inode(struct inode **inode, struct ptlrpc_request *req,
2497                   struct super_block *sb, struct lookup_intent *it)
2498 {
2499         struct ll_sb_info *sbi = NULL;
2500         struct lustre_md md = { NULL };
2501         bool default_lmv_deleted = false;
2502         int rc;
2503
2504         ENTRY;
2505
2506         LASSERT(*inode || sb);
2507         sbi = sb ? ll_s2sbi(sb) : ll_i2sbi(*inode);
2508         rc = md_get_lustre_md(sbi->ll_md_exp, req, sbi->ll_dt_exp,
2509                               sbi->ll_md_exp, &md);
2510         if (rc != 0)
2511                 GOTO(out, rc);
2512
2513         /*
2514          * clear default_lmv only if intent_getattr reply doesn't contain it.
2515          * but it needs to be done after iget, check this early because
2516          * ll_update_lsm_md() may change md.
2517          */
2518         if (it && (it->it_op & (IT_LOOKUP | IT_GETATTR)) &&
2519             S_ISDIR(md.body->mbo_mode) && !md.default_lmv)
2520                 default_lmv_deleted = true;
2521
2522         if (*inode) {
2523                 rc = ll_update_inode(*inode, &md);
2524                 if (rc != 0)
2525                         GOTO(out, rc);
2526         } else {
2527                 LASSERT(sb != NULL);
2528
2529                 /*
2530                  * At this point server returns to client's same fid as client
2531                  * generated for creating. So using ->fid1 is okay here.
2532                  */
2533                 if (!fid_is_sane(&md.body->mbo_fid1)) {
2534                         CERROR("%s: Fid is insane "DFID"\n",
2535                                 sbi->ll_fsname,
2536                                 PFID(&md.body->mbo_fid1));
2537                         GOTO(out, rc = -EINVAL);
2538                 }
2539
2540                 *inode = ll_iget(sb, cl_fid_build_ino(&md.body->mbo_fid1,
2541                                              sbi->ll_flags & LL_SBI_32BIT_API),
2542                                  &md);
2543                 if (IS_ERR(*inode)) {
2544 #ifdef CONFIG_FS_POSIX_ACL
2545                         if (md.posix_acl) {
2546                                 posix_acl_release(md.posix_acl);
2547                                 md.posix_acl = NULL;
2548                         }
2549 #endif
2550                         rc = IS_ERR(*inode) ? PTR_ERR(*inode) : -ENOMEM;
2551                         *inode = NULL;
2552                         CERROR("new_inode -fatal: rc %d\n", rc);
2553                         GOTO(out, rc);
2554                 }
2555         }
2556
2557         /* Handling piggyback layout lock.
2558          * Layout lock can be piggybacked by getattr and open request.
2559          * The lsm can be applied to inode only if it comes with a layout lock
2560          * otherwise correct layout may be overwritten, for example:
2561          * 1. proc1: mdt returns a lsm but not granting layout
2562          * 2. layout was changed by another client
2563          * 3. proc2: refresh layout and layout lock granted
2564          * 4. proc1: to apply a stale layout */
2565         if (it != NULL && it->it_lock_mode != 0) {
2566                 struct lustre_handle lockh;
2567                 struct ldlm_lock *lock;
2568
2569                 lockh.cookie = it->it_lock_handle;
2570                 lock = ldlm_handle2lock(&lockh);
2571                 LASSERT(lock != NULL);
2572                 if (ldlm_has_layout(lock)) {
2573                         struct cl_object_conf conf;
2574
2575                         memset(&conf, 0, sizeof(conf));
2576                         conf.coc_opc = OBJECT_CONF_SET;
2577                         conf.coc_inode = *inode;
2578                         conf.coc_lock = lock;
2579                         conf.u.coc_layout = md.layout;
2580                         (void)ll_layout_conf(*inode, &conf);
2581                 }
2582                 LDLM_LOCK_PUT(lock);
2583         }
2584
2585         if (default_lmv_deleted)
2586                 ll_update_default_lsm_md(*inode, &md);
2587
2588         GOTO(out, rc = 0);
2589
2590 out:
2591         /* cleanup will be done if necessary */
2592         md_free_lustre_md(sbi->ll_md_exp, &md);
2593
2594         if (rc != 0 && it != NULL && it->it_op & IT_OPEN)
2595                 ll_open_cleanup(sb != NULL ? sb : (*inode)->i_sb, req);
2596
2597         return rc;
2598 }
2599
2600 int ll_obd_statfs(struct inode *inode, void __user *arg)
2601 {
2602         struct ll_sb_info *sbi = NULL;
2603         struct obd_export *exp;
2604         char *buf = NULL;
2605         struct obd_ioctl_data *data = NULL;
2606         __u32 type;
2607         int len = 0, rc;
2608
2609         if (!inode || !(sbi = ll_i2sbi(inode)))
2610                 GOTO(out_statfs, rc = -EINVAL);
2611
2612         rc = obd_ioctl_getdata(&buf, &len, arg);
2613         if (rc)
2614                 GOTO(out_statfs, rc);
2615
2616         data = (void*)buf;
2617         if (!data->ioc_inlbuf1 || !data->ioc_inlbuf2 ||
2618             !data->ioc_pbuf1 || !data->ioc_pbuf2)
2619                 GOTO(out_statfs, rc = -EINVAL);
2620
2621         if (data->ioc_inllen1 != sizeof(__u32) ||
2622             data->ioc_inllen2 != sizeof(__u32) ||
2623             data->ioc_plen1 != sizeof(struct obd_statfs) ||
2624             data->ioc_plen2 != sizeof(struct obd_uuid))
2625                 GOTO(out_statfs, rc = -EINVAL);
2626
2627         memcpy(&type, data->ioc_inlbuf1, sizeof(__u32));
2628         if (type & LL_STATFS_LMV)
2629                 exp = sbi->ll_md_exp;
2630         else if (type & LL_STATFS_LOV)
2631                 exp = sbi->ll_dt_exp;
2632         else
2633                 GOTO(out_statfs, rc = -ENODEV);
2634
2635         rc = obd_iocontrol(IOC_OBD_STATFS, exp, len, buf, NULL);
2636         if (rc)
2637                 GOTO(out_statfs, rc);
2638 out_statfs:
2639         OBD_FREE_LARGE(buf, len);
2640         return rc;
2641 }
2642
2643 /*
2644  * this is normally called in ll_fini_md_op_data(), but sometimes it needs to
2645  * be called early to avoid deadlock.
2646  */
2647 void ll_unlock_md_op_lsm(struct md_op_data *op_data)
2648 {
2649         if (op_data->op_mea2_sem) {
2650                 up_read(op_data->op_mea2_sem);
2651                 op_data->op_mea2_sem = NULL;
2652         }
2653
2654         if (op_data->op_mea1_sem) {
2655                 up_read(op_data->op_mea1_sem);
2656                 op_data->op_mea1_sem = NULL;
2657         }
2658 }
2659
2660 /* this function prepares md_op_data hint for passing it down to MD stack. */
2661 struct md_op_data *ll_prep_md_op_data(struct md_op_data *op_data,
2662                                       struct inode *i1, struct inode *i2,
2663                                       const char *name, size_t namelen,
2664                                       __u32 mode, enum md_op_code opc,
2665                                       void *data)
2666 {
2667         LASSERT(i1 != NULL);
2668
2669         if (name == NULL) {
2670                 /* Do not reuse namelen for something else. */
2671                 if (namelen != 0)
2672                         return ERR_PTR(-EINVAL);
2673         } else {
2674                 if (namelen > ll_i2sbi(i1)->ll_namelen)
2675                         return ERR_PTR(-ENAMETOOLONG);
2676
2677                 if (!lu_name_is_valid_2(name, namelen))
2678                         return ERR_PTR(-EINVAL);
2679         }
2680
2681         if (op_data == NULL)
2682                 OBD_ALLOC_PTR(op_data);
2683
2684         if (op_data == NULL)
2685                 return ERR_PTR(-ENOMEM);
2686
2687         ll_i2gids(op_data->op_suppgids, i1, i2);
2688         op_data->op_fid1 = *ll_inode2fid(i1);
2689         op_data->op_code = opc;
2690
2691         if (S_ISDIR(i1->i_mode)) {
2692                 down_read(&ll_i2info(i1)->lli_lsm_sem);
2693                 op_data->op_mea1_sem = &ll_i2info(i1)->lli_lsm_sem;
2694                 op_data->op_mea1 = ll_i2info(i1)->lli_lsm_md;
2695                 op_data->op_default_mea1 = ll_i2info(i1)->lli_default_lsm_md;
2696         }
2697
2698         if (i2) {
2699                 op_data->op_fid2 = *ll_inode2fid(i2);
2700                 if (S_ISDIR(i2->i_mode)) {
2701                         if (i2 != i1) {
2702                                 down_read(&ll_i2info(i2)->lli_lsm_sem);
2703                                 op_data->op_mea2_sem =
2704                                                 &ll_i2info(i2)->lli_lsm_sem;
2705                         }
2706                         op_data->op_mea2 = ll_i2info(i2)->lli_lsm_md;
2707                 }
2708         } else {
2709                 fid_zero(&op_data->op_fid2);
2710         }
2711
2712         if (ll_i2sbi(i1)->ll_flags & LL_SBI_64BIT_HASH)
2713                 op_data->op_cli_flags |= CLI_HASH64;
2714
2715         if (ll_need_32bit_api(ll_i2sbi(i1)))
2716                 op_data->op_cli_flags |= CLI_API32;
2717
2718         op_data->op_name = name;
2719         op_data->op_namelen = namelen;
2720         op_data->op_mode = mode;
2721         op_data->op_mod_time = ktime_get_real_seconds();
2722         op_data->op_fsuid = from_kuid(&init_user_ns, current_fsuid());
2723         op_data->op_fsgid = from_kgid(&init_user_ns, current_fsgid());
2724         op_data->op_cap = cfs_curproc_cap_pack();
2725         op_data->op_mds = 0;
2726         if ((opc == LUSTRE_OPC_CREATE) && (name != NULL) &&
2727              filename_is_volatile(name, namelen, &op_data->op_mds)) {
2728                 op_data->op_bias |= MDS_CREATE_VOLATILE;
2729         }
2730         op_data->op_data = data;
2731
2732         return op_data;
2733 }
2734
2735 void ll_finish_md_op_data(struct md_op_data *op_data)
2736 {
2737         ll_unlock_md_op_lsm(op_data);
2738         security_release_secctx(op_data->op_file_secctx,
2739                                 op_data->op_file_secctx_size);
2740         OBD_FREE_PTR(op_data);
2741 }
2742
2743 #ifdef HAVE_SUPEROPS_USE_DENTRY
2744 int ll_show_options(struct seq_file *seq, struct dentry *dentry)
2745 #else
2746 int ll_show_options(struct seq_file *seq, struct vfsmount *vfs)
2747 #endif
2748 {
2749         struct ll_sb_info *sbi;
2750
2751 #ifdef HAVE_SUPEROPS_USE_DENTRY
2752         LASSERT((seq != NULL) && (dentry != NULL));
2753         sbi = ll_s2sbi(dentry->d_sb);
2754 #else
2755         LASSERT((seq != NULL) && (vfs != NULL));
2756         sbi = ll_s2sbi(vfs->mnt_sb);
2757 #endif
2758
2759         if (sbi->ll_flags & LL_SBI_NOLCK)
2760                 seq_puts(seq, ",nolock");
2761
2762         if (sbi->ll_flags & LL_SBI_FLOCK)
2763                 seq_puts(seq, ",flock");
2764
2765         if (sbi->ll_flags & LL_SBI_LOCALFLOCK)
2766                 seq_puts(seq, ",localflock");
2767
2768         if (sbi->ll_flags & LL_SBI_USER_XATTR)
2769                 seq_puts(seq, ",user_xattr");
2770
2771         if (sbi->ll_flags & LL_SBI_LAZYSTATFS)
2772                 seq_puts(seq, ",lazystatfs");
2773
2774         if (sbi->ll_flags & LL_SBI_USER_FID2PATH)
2775                 seq_puts(seq, ",user_fid2path");
2776
2777         if (sbi->ll_flags & LL_SBI_ALWAYS_PING)
2778                 seq_puts(seq, ",always_ping");
2779
2780         RETURN(0);
2781 }
2782
2783 /**
2784  * Get obd name by cmd, and copy out to user space
2785  */
2786 int ll_get_obd_name(struct inode *inode, unsigned int cmd, unsigned long arg)
2787 {
2788         struct ll_sb_info *sbi = ll_i2sbi(inode);
2789         struct obd_device *obd;
2790         ENTRY;
2791
2792         if (cmd == OBD_IOC_GETDTNAME)
2793                 obd = class_exp2obd(sbi->ll_dt_exp);
2794         else if (cmd == OBD_IOC_GETMDNAME)
2795                 obd = class_exp2obd(sbi->ll_md_exp);
2796         else
2797                 RETURN(-EINVAL);
2798
2799         if (!obd)
2800                 RETURN(-ENOENT);
2801
2802         if (copy_to_user((void __user *)arg, obd->obd_name,
2803                          strlen(obd->obd_name) + 1))
2804                 RETURN(-EFAULT);
2805
2806         RETURN(0);
2807 }
2808
2809 static char* ll_d_path(struct dentry *dentry, char *buf, int bufsize)
2810 {
2811         char *path = NULL;
2812
2813         struct path p;
2814
2815         p.dentry = dentry;
2816         p.mnt = current->fs->root.mnt;
2817         path_get(&p);
2818         path = d_path(&p, buf, bufsize);
2819         path_put(&p);
2820         return path;
2821 }
2822
2823 void ll_dirty_page_discard_warn(struct page *page, int ioret)
2824 {
2825         char *buf, *path = NULL;
2826         struct dentry *dentry = NULL;
2827         struct inode *inode = page->mapping->host;
2828
2829         /* this can be called inside spin lock so use GFP_ATOMIC. */
2830         buf = (char *)__get_free_page(GFP_ATOMIC);
2831         if (buf != NULL) {
2832                 dentry = d_find_alias(page->mapping->host);
2833                 if (dentry != NULL)
2834                         path = ll_d_path(dentry, buf, PAGE_SIZE);
2835         }
2836
2837         CDEBUG(D_WARNING,
2838                "%s: dirty page discard: %s/fid: "DFID"/%s may get corrupted "
2839                "(rc %d)\n", ll_i2sbi(inode)->ll_fsname,
2840                s2lsi(page->mapping->host->i_sb)->lsi_lmd->lmd_dev,
2841                PFID(ll_inode2fid(inode)),
2842                (path && !IS_ERR(path)) ? path : "", ioret);
2843
2844         if (dentry != NULL)
2845                 dput(dentry);
2846
2847         if (buf != NULL)
2848                 free_page((unsigned long)buf);
2849 }
2850
2851 ssize_t ll_copy_user_md(const struct lov_user_md __user *md,
2852                         struct lov_user_md **kbuf)
2853 {
2854         struct lov_user_md      lum;
2855         ssize_t                 lum_size;
2856         ENTRY;
2857
2858         if (copy_from_user(&lum, md, sizeof(lum)))
2859                 RETURN(-EFAULT);
2860
2861         lum_size = ll_lov_user_md_size(&lum);
2862         if (lum_size < 0)
2863                 RETURN(lum_size);
2864
2865         OBD_ALLOC(*kbuf, lum_size);
2866         if (*kbuf == NULL)
2867                 RETURN(-ENOMEM);
2868
2869         if (copy_from_user(*kbuf, md, lum_size) != 0) {
2870                 OBD_FREE(*kbuf, lum_size);
2871                 RETURN(-EFAULT);
2872         }
2873
2874         RETURN(lum_size);
2875 }
2876
2877 /*
2878  * Compute llite root squash state after a change of root squash
2879  * configuration setting or add/remove of a lnet nid
2880  */
2881 void ll_compute_rootsquash_state(struct ll_sb_info *sbi)
2882 {
2883         struct root_squash_info *squash = &sbi->ll_squash;
2884         int i;
2885         bool matched;
2886         struct lnet_process_id id;
2887
2888         /* Update norootsquash flag */
2889         down_write(&squash->rsi_sem);
2890         if (list_empty(&squash->rsi_nosquash_nids))
2891                 sbi->ll_flags &= ~LL_SBI_NOROOTSQUASH;
2892         else {
2893                 /* Do not apply root squash as soon as one of our NIDs is
2894                  * in the nosquash_nids list */
2895                 matched = false;
2896                 i = 0;
2897                 while (LNetGetId(i++, &id) != -ENOENT) {
2898                         if (LNET_NETTYP(LNET_NIDNET(id.nid)) == LOLND)
2899                                 continue;
2900                         if (cfs_match_nid(id.nid, &squash->rsi_nosquash_nids)) {
2901                                 matched = true;
2902                                 break;
2903                         }
2904                 }
2905                 if (matched)
2906                         sbi->ll_flags |= LL_SBI_NOROOTSQUASH;
2907                 else
2908                         sbi->ll_flags &= ~LL_SBI_NOROOTSQUASH;
2909         }
2910         up_write(&squash->rsi_sem);
2911 }
2912
2913 /**
2914  * Parse linkea content to extract information about a given hardlink
2915  *
2916  * \param[in]   ldata      - Initialized linkea data
2917  * \param[in]   linkno     - Link identifier
2918  * \param[out]  parent_fid - The entry's parent FID
2919  * \param[out]  ln         - Entry name destination buffer
2920  *
2921  * \retval 0 on success
2922  * \retval Appropriate negative error code on failure
2923  */
2924 static int ll_linkea_decode(struct linkea_data *ldata, unsigned int linkno,
2925                             struct lu_fid *parent_fid, struct lu_name *ln)
2926 {
2927         unsigned int    idx;
2928         int             rc;
2929         ENTRY;
2930
2931         rc = linkea_init_with_rec(ldata);
2932         if (rc < 0)
2933                 RETURN(rc);
2934
2935         if (linkno >= ldata->ld_leh->leh_reccount)
2936                 /* beyond last link */
2937                 RETURN(-ENODATA);
2938
2939         linkea_first_entry(ldata);
2940         for (idx = 0; ldata->ld_lee != NULL; idx++) {
2941                 linkea_entry_unpack(ldata->ld_lee, &ldata->ld_reclen, ln,
2942                                     parent_fid);
2943                 if (idx == linkno)
2944                         break;
2945
2946                 linkea_next_entry(ldata);
2947         }
2948
2949         if (idx < linkno)
2950                 RETURN(-ENODATA);
2951
2952         RETURN(0);
2953 }
2954
2955 /**
2956  * Get parent FID and name of an identified link. Operation is performed for
2957  * a given link number, letting the caller iterate over linkno to list one or
2958  * all links of an entry.
2959  *
2960  * \param[in]     file - File descriptor against which to perform the operation
2961  * \param[in,out] arg  - User-filled structure containing the linkno to operate
2962  *                       on and the available size. It is eventually filled with
2963  *                       the requested information or left untouched on error
2964  *
2965  * \retval - 0 on success
2966  * \retval - Appropriate negative error code on failure
2967  */
2968 int ll_getparent(struct file *file, struct getparent __user *arg)
2969 {
2970         struct inode            *inode = file_inode(file);
2971         struct linkea_data      *ldata;
2972         struct lu_buf            buf = LU_BUF_NULL;
2973         struct lu_name           ln;
2974         struct lu_fid            parent_fid;
2975         __u32                    linkno;
2976         __u32                    name_size;
2977         int                      rc;
2978
2979         ENTRY;
2980
2981         if (!cfs_capable(CFS_CAP_DAC_READ_SEARCH) &&
2982             !(ll_i2sbi(inode)->ll_flags & LL_SBI_USER_FID2PATH))
2983                 RETURN(-EPERM);
2984
2985         if (get_user(name_size, &arg->gp_name_size))
2986                 RETURN(-EFAULT);
2987
2988         if (get_user(linkno, &arg->gp_linkno))
2989                 RETURN(-EFAULT);
2990
2991         if (name_size > PATH_MAX)
2992                 RETURN(-EINVAL);
2993
2994         OBD_ALLOC(ldata, sizeof(*ldata));
2995         if (ldata == NULL)
2996                 RETURN(-ENOMEM);
2997
2998         rc = linkea_data_new(ldata, &buf);
2999         if (rc < 0)
3000                 GOTO(ldata_free, rc);
3001
3002 #ifdef HAVE_XATTR_HANDLER_FLAGS
3003         rc = ll_xattr_list(inode, XATTR_NAME_LINK, XATTR_TRUSTED_T, buf.lb_buf,
3004                            buf.lb_len, OBD_MD_FLXATTR);
3005 #else
3006         rc = ll_getxattr(file_dentry(file), XATTR_NAME_LINK, buf.lb_buf,
3007                          buf.lb_len);
3008 #endif /* HAVE_XATTR_HANDLER_FLAGS */
3009         if (rc < 0)
3010                 GOTO(lb_free, rc);
3011
3012         rc = ll_linkea_decode(ldata, linkno, &parent_fid, &ln);
3013         if (rc < 0)
3014                 GOTO(lb_free, rc);
3015
3016         if (ln.ln_namelen >= name_size)
3017                 GOTO(lb_free, rc = -EOVERFLOW);
3018
3019         if (copy_to_user(&arg->gp_fid, &parent_fid, sizeof(arg->gp_fid)))
3020                 GOTO(lb_free, rc = -EFAULT);
3021
3022         if (copy_to_user(&arg->gp_name, ln.ln_name, ln.ln_namelen))
3023                 GOTO(lb_free, rc = -EFAULT);
3024
3025         if (put_user('\0', arg->gp_name + ln.ln_namelen))
3026                 GOTO(lb_free, rc = -EFAULT);
3027
3028 lb_free:
3029         lu_buf_free(&buf);
3030 ldata_free:
3031         OBD_FREE(ldata, sizeof(*ldata));
3032
3033         RETURN(rc);
3034 }