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                 lli->lli_def_stripe_offset = -1;
 980                 init_rwsem(&lli->lli_lsm_sem);
 981         } else {
 982                 mutex_init(&lli->lli_size_mutex);
 983                 lli->lli_symlink_name = NULL;
 984                 init_rwsem(&lli->lli_trunc_sem);
 985                 range_lock_tree_init(&lli->lli_write_tree);
 986                 init_rwsem(&lli->lli_glimpse_sem);
 987                 lli->lli_glimpse_time = ktime_set(0, 0);
 988                 INIT_LIST_HEAD(&lli->lli_agl_list);
 989                 lli->lli_agl_index = 0;
 990                 lli->lli_async_rc = 0;
 991                 spin_lock_init(&lli->lli_heat_lock);
 992                 obd_heat_clear(lli->lli_heat_instances, OBD_HEAT_COUNT);
 993                 lli->lli_heat_flags = 0;
 994                 mutex_init(&lli->lli_pcc_lock);
 995                 lli->lli_pcc_state = PCC_STATE_FL_NONE;
 996                 lli->lli_pcc_inode = NULL;
 997         }
 998         mutex_init(&lli->lli_layout_mutex);
 999         memset(lli->lli_jobid, 0, sizeof(lli->lli_jobid));
1000 }
1001
1002 #define MAX_STRING_SIZE 128
1003
1004 #ifndef HAVE_SUPER_SETUP_BDI_NAME
1005
1006 #define LSI_BDI_INITIALIZED     0x00400000
1007
1008 #ifndef HAVE_BDI_CAP_MAP_COPY
1009 # define BDI_CAP_MAP_COPY       0
1010 #endif
1011
1012 static int super_setup_bdi_name(struct super_block *sb, char *fmt, ...)
1013 {
1014         struct  lustre_sb_info *lsi = s2lsi(sb);
1015         char buf[MAX_STRING_SIZE];
1016         va_list args;
1017         int err;
1018
1019         err = bdi_init(&lsi->lsi_bdi);
1020         if (err)
1021                 return err;
1022
1023         lsi->lsi_flags |= LSI_BDI_INITIALIZED;
1024         lsi->lsi_bdi.capabilities = BDI_CAP_MAP_COPY;
1025         lsi->lsi_bdi.name = "lustre";
1026         va_start(args, fmt);
1027         vsnprintf(buf, MAX_STRING_SIZE, fmt, args);
1028         va_end(args);
1029         err = bdi_register(&lsi->lsi_bdi, NULL, "%s", buf);
1030         va_end(args);
1031         if (!err)
1032                 sb->s_bdi = &lsi->lsi_bdi;
1033
1034         return err;
1035 }
1036 #endif /* !HAVE_SUPER_SETUP_BDI_NAME */
1037
1038 int ll_fill_super(struct super_block *sb, struct vfsmount *mnt)
1039 {
1040         struct  lustre_profile *lprof = NULL;
1041         struct  lustre_sb_info *lsi = s2lsi(sb);
1042         struct  ll_sb_info *sbi = NULL;
1043         char    *dt = NULL, *md = NULL;
1044         char    *profilenm = get_profile_name(sb);
1045         struct config_llog_instance *cfg;
1046         /* %p for void* in printf needs 16+2 characters: 0xffffffffffffffff */
1047         const int instlen = 16 + 2;
1048         unsigned long cfg_instance = ll_get_cfg_instance(sb);
1049         char name[MAX_STRING_SIZE];
1050         int md_len = 0;
1051         int dt_len = 0;
1052         uuid_t uuid;
1053         char *ptr;
1054         int len;
1055         int err;
1056
1057         ENTRY;
1058         /* for ASLR, to map between cfg_instance and hashed ptr */
1059         CDEBUG(D_VFSTRACE, "VFS Op: cfg_instance %s-%016lx (sb %p)\n",
1060                profilenm, cfg_instance, sb);
1061
1062         try_module_get(THIS_MODULE);
1063
1064         OBD_ALLOC_PTR(cfg);
1065         if (cfg == NULL)
1066                 GOTO(out_free_cfg, err = -ENOMEM);
1067
1068         /* client additional sb info */
1069         lsi->lsi_llsbi = sbi = ll_init_sbi();
1070         if (IS_ERR(sbi))
1071                 GOTO(out_free_cfg, err = PTR_ERR(sbi));
1072
1073         err = ll_options(lsi->lsi_lmd->lmd_opts, sbi);
1074         if (err)
1075                 GOTO(out_free_cfg, err);
1076
1077 #ifndef HAVE_DCACHE_LOCK
1078         /* kernel >= 2.6.38 store dentry operations in sb->s_d_op. */
1079         sb->s_d_op = &ll_d_ops;
1080 #endif
1081         /* UUID handling */
1082         generate_random_uuid(uuid.b);
1083         snprintf(sbi->ll_sb_uuid.uuid, UUID_SIZE, "%pU", uuid.b);
1084
1085         CDEBUG(D_CONFIG, "llite sb uuid: %s\n", sbi->ll_sb_uuid.uuid);
1086
1087         /* Get fsname */
1088         len = strlen(profilenm);
1089         ptr = strrchr(profilenm, '-');
1090         if (ptr && (strcmp(ptr, "-client") == 0))
1091                 len -= 7;
1092
1093         if (len > LUSTRE_MAXFSNAME) {
1094                 if (unlikely(len >= MAX_STRING_SIZE))
1095                         len = MAX_STRING_SIZE - 1;
1096                 strncpy(name, profilenm, len);
1097                 name[len] = '\0';
1098                 err = -ENAMETOOLONG;
1099                 CERROR("%s: fsname longer than %u characters: rc = %d\n",
1100                        name, LUSTRE_MAXFSNAME, err);
1101                 GOTO(out_free_cfg, err);
1102         }
1103         strncpy(sbi->ll_fsname, profilenm, len);
1104         sbi->ll_fsname[len] = '\0';
1105
1106         /* Mount info */
1107         snprintf(name, sizeof(name), "%.*s-%016lx", len,
1108                  profilenm, cfg_instance);
1109
1110         err = super_setup_bdi_name(sb, "%s", name);
1111         if (err)
1112                 GOTO(out_free_cfg, err);
1113
1114         /* Call ll_debugfs_register_super() before lustre_process_log()
1115          * so that "llite.*.*" params can be processed correctly.
1116          */
1117         err = ll_debugfs_register_super(sb, name);
1118         if (err < 0) {
1119                 CERROR("%s: could not register mountpoint in llite: rc = %d\n",
1120                        sbi->ll_fsname, err);
1121                 err = 0;
1122         }
1123
1124         /* The cfg_instance is a value unique to this super, in case some
1125          * joker tries to mount the same fs at two mount points.
1126          */
1127         cfg->cfg_instance = cfg_instance;
1128         cfg->cfg_uuid = lsi->lsi_llsbi->ll_sb_uuid;
1129         cfg->cfg_callback = class_config_llog_handler;
1130         cfg->cfg_sub_clds = CONFIG_SUB_CLIENT;
1131         /* set up client obds */
1132         err = lustre_process_log(sb, profilenm, cfg);
1133         if (err < 0)
1134                 GOTO(out_debugfs, err);
1135
1136         /* Profile set with LCFG_MOUNTOPT so we can find our mdc and osc obds */
1137         lprof = class_get_profile(profilenm);
1138         if (lprof == NULL) {
1139                 LCONSOLE_ERROR_MSG(0x156, "The client profile '%s' could not be"
1140                                    " read from the MGS.  Does that filesystem "
1141                                    "exist?\n", profilenm);
1142                 GOTO(out_debugfs, err = -EINVAL);
1143         }
1144         CDEBUG(D_CONFIG, "Found profile %s: mdc=%s osc=%s\n", profilenm,
1145                lprof->lp_md, lprof->lp_dt);
1146
1147         dt_len = strlen(lprof->lp_dt) + instlen + 2;
1148         OBD_ALLOC(dt, dt_len);
1149         if (!dt)
1150                 GOTO(out_profile, err = -ENOMEM);
1151         snprintf(dt, dt_len - 1, "%s-%016lx", lprof->lp_dt, cfg_instance);
1152
1153         md_len = strlen(lprof->lp_md) + instlen + 2;
1154         OBD_ALLOC(md, md_len);
1155         if (!md)
1156                 GOTO(out_free_dt, err = -ENOMEM);
1157         snprintf(md, md_len - 1, "%s-%016lx", lprof->lp_md, cfg_instance);
1158
1159         /* connections, registrations, sb setup */
1160         err = client_common_fill_super(sb, md, dt, mnt);
1161         if (err < 0)
1162                 GOTO(out_free_md, err);
1163
1164         sbi->ll_client_common_fill_super_succeeded = 1;
1165
1166 out_free_md:
1167         if (md)
1168                 OBD_FREE(md, md_len);
1169 out_free_dt:
1170         if (dt)
1171                 OBD_FREE(dt, dt_len);
1172 out_profile:
1173         if (lprof)
1174                 class_put_profile(lprof);
1175 out_debugfs:
1176         if (err < 0)
1177                 ll_debugfs_unregister_super(sb);
1178 out_free_cfg:
1179         if (cfg)
1180                 OBD_FREE_PTR(cfg);
1181
1182         if (err)
1183                 ll_put_super(sb);
1184         else if (sbi->ll_flags & LL_SBI_VERBOSE)
1185                 LCONSOLE_WARN("Mounted %s\n", profilenm);
1186         RETURN(err);
1187 } /* ll_fill_super */
1188
1189 void ll_put_super(struct super_block *sb)
1190 {
1191         struct config_llog_instance cfg, params_cfg;
1192         struct obd_device *obd;
1193         struct lustre_sb_info *lsi = s2lsi(sb);
1194         struct ll_sb_info *sbi = ll_s2sbi(sb);
1195         char *profilenm = get_profile_name(sb);
1196         unsigned long cfg_instance = ll_get_cfg_instance(sb);
1197         long ccc_count;
1198         int next, force = 1, rc = 0;
1199         ENTRY;
1200
1201         if (IS_ERR(sbi))
1202                 GOTO(out_no_sbi, 0);
1203
1204         /* Should replace instance_id with something better for ASLR */
1205         CDEBUG(D_VFSTRACE, "VFS Op: cfg_instance %s-%016lx (sb %p)\n",
1206                profilenm, cfg_instance, sb);
1207
1208         cfg.cfg_instance = cfg_instance;
1209         lustre_end_log(sb, profilenm, &cfg);
1210
1211         params_cfg.cfg_instance = cfg_instance;
1212         lustre_end_log(sb, PARAMS_FILENAME, &params_cfg);
1213
1214         if (sbi->ll_md_exp) {
1215                 obd = class_exp2obd(sbi->ll_md_exp);
1216                 if (obd)
1217                         force = obd->obd_force;
1218         }
1219
1220         /* Wait for unstable pages to be committed to stable storage */
1221         if (force == 0) {
1222                 struct l_wait_info lwi = LWI_INTR(LWI_ON_SIGNAL_NOOP, NULL);
1223                 rc = l_wait_event(sbi->ll_cache->ccc_unstable_waitq,
1224                         atomic_long_read(&sbi->ll_cache->ccc_unstable_nr) == 0,
1225                         &lwi);
1226         }
1227
1228         ccc_count = atomic_long_read(&sbi->ll_cache->ccc_unstable_nr);
1229         if (force == 0 && rc != -EINTR)
1230                 LASSERTF(ccc_count == 0, "count: %li\n", ccc_count);
1231
1232         /* We need to set force before the lov_disconnect in
1233            lustre_common_put_super, since l_d cleans up osc's as well. */
1234         if (force) {
1235                 next = 0;
1236                 while ((obd = class_devices_in_group(&sbi->ll_sb_uuid,
1237                                                      &next)) != NULL) {
1238                         obd->obd_force = force;
1239                 }
1240         }
1241
1242         if (sbi->ll_client_common_fill_super_succeeded) {
1243                 /* Only if client_common_fill_super succeeded */
1244                 client_common_put_super(sb);
1245         }
1246
1247         next = 0;
1248         while ((obd = class_devices_in_group(&sbi->ll_sb_uuid, &next)) !=NULL) {
1249                 class_manual_cleanup(obd);
1250         }
1251
1252         if (sbi->ll_flags & LL_SBI_VERBOSE)
1253                 LCONSOLE_WARN("Unmounted %s\n", profilenm ? profilenm : "");
1254
1255         if (profilenm)
1256                 class_del_profile(profilenm);
1257
1258 #ifndef HAVE_SUPER_SETUP_BDI_NAME
1259         if (lsi->lsi_flags & LSI_BDI_INITIALIZED) {
1260                 bdi_destroy(&lsi->lsi_bdi);
1261                 lsi->lsi_flags &= ~LSI_BDI_INITIALIZED;
1262         }
1263 #endif
1264
1265         ll_free_sbi(sb);
1266         lsi->lsi_llsbi = NULL;
1267 out_no_sbi:
1268         lustre_common_put_super(sb);
1269
1270         cl_env_cache_purge(~0);
1271
1272         module_put(THIS_MODULE);
1273
1274         EXIT;
1275 } /* client_put_super */
1276
1277 struct inode *ll_inode_from_resource_lock(struct ldlm_lock *lock)
1278 {
1279         struct inode *inode = NULL;
1280
1281         /* NOTE: we depend on atomic igrab() -bzzz */
1282         lock_res_and_lock(lock);
1283         if (lock->l_resource->lr_lvb_inode) {
1284                 struct ll_inode_info * lli;
1285                 lli = ll_i2info(lock->l_resource->lr_lvb_inode);
1286                 if (lli->lli_inode_magic == LLI_INODE_MAGIC) {
1287                         inode = igrab(lock->l_resource->lr_lvb_inode);
1288                 } else {
1289                         inode = lock->l_resource->lr_lvb_inode;
1290                         LDLM_DEBUG_LIMIT(inode->i_state & I_FREEING ?  D_INFO :
1291                                          D_WARNING, lock, "lr_lvb_inode %p is "
1292                                          "bogus: magic %08x",
1293                                          lock->l_resource->lr_lvb_inode,
1294                                          lli->lli_inode_magic);
1295                         inode = NULL;
1296                 }
1297         }
1298         unlock_res_and_lock(lock);
1299         return inode;
1300 }
1301
1302 void ll_dir_clear_lsm_md(struct inode *inode)
1303 {
1304         struct ll_inode_info *lli = ll_i2info(inode);
1305
1306         LASSERT(S_ISDIR(inode->i_mode));
1307
1308         if (lli->lli_lsm_md != NULL) {
1309                 lmv_free_memmd(lli->lli_lsm_md);
1310                 lli->lli_lsm_md = NULL;
1311         }
1312 }
1313
1314 static struct inode *ll_iget_anon_dir(struct super_block *sb,
1315                                       const struct lu_fid *fid,
1316                                       struct lustre_md *md)
1317 {
1318         struct ll_sb_info       *sbi = ll_s2sbi(sb);
1319         struct mdt_body         *body = md->body;
1320         struct inode            *inode;
1321         ino_t                   ino;
1322         ENTRY;
1323
1324         ino = cl_fid_build_ino(fid, sbi->ll_flags & LL_SBI_32BIT_API);
1325         inode = iget_locked(sb, ino);
1326         if (inode == NULL) {
1327                 CERROR("%s: failed get simple inode "DFID": rc = -ENOENT\n",
1328                        sbi->ll_fsname, PFID(fid));
1329                 RETURN(ERR_PTR(-ENOENT));
1330         }
1331
1332         if (inode->i_state & I_NEW) {
1333                 struct ll_inode_info *lli = ll_i2info(inode);
1334                 struct lmv_stripe_md *lsm = md->lmv;
1335
1336                 inode->i_mode = (inode->i_mode & ~S_IFMT) |
1337                                 (body->mbo_mode & S_IFMT);
1338                 LASSERTF(S_ISDIR(inode->i_mode), "Not slave inode "DFID"\n",
1339                          PFID(fid));
1340
1341                 inode->i_mtime.tv_sec = 0;
1342                 inode->i_atime.tv_sec = 0;
1343                 inode->i_ctime.tv_sec = 0;
1344                 inode->i_rdev = 0;
1345
1346 #ifdef HAVE_BACKING_DEV_INFO
1347                 /* initializing backing dev info. */
1348                 inode->i_mapping->backing_dev_info =
1349                                                 &s2lsi(inode->i_sb)->lsi_bdi;
1350 #endif
1351                 inode->i_op = &ll_dir_inode_operations;
1352                 inode->i_fop = &ll_dir_operations;
1353                 lli->lli_fid = *fid;
1354                 ll_lli_init(lli);
1355
1356                 LASSERT(lsm != NULL);
1357                 /* master object FID */
1358                 lli->lli_pfid = body->mbo_fid1;
1359                 CDEBUG(D_INODE, "lli %p slave "DFID" master "DFID"\n",
1360                        lli, PFID(fid), PFID(&lli->lli_pfid));
1361                 unlock_new_inode(inode);
1362         }
1363
1364         RETURN(inode);
1365 }
1366
1367 static int ll_init_lsm_md(struct inode *inode, struct lustre_md *md)
1368 {
1369         struct lu_fid *fid;
1370         struct lmv_stripe_md *lsm = md->lmv;
1371         struct ll_inode_info *lli = ll_i2info(inode);
1372         int i;
1373
1374         LASSERT(lsm != NULL);
1375
1376         CDEBUG(D_INODE, "%s: "DFID" set dir layout:\n",
1377                ll_i2sbi(inode)->ll_fsname, PFID(&lli->lli_fid));
1378         lsm_md_dump(D_INODE, lsm);
1379
1380         /* XXX sigh, this lsm_root initialization should be in
1381          * LMV layer, but it needs ll_iget right now, so we
1382          * put this here right now. */
1383         for (i = 0; i < lsm->lsm_md_stripe_count; i++) {
1384                 fid = &lsm->lsm_md_oinfo[i].lmo_fid;
1385                 LASSERT(lsm->lsm_md_oinfo[i].lmo_root == NULL);
1386
1387                 if (!fid_is_sane(fid))
1388                         continue;
1389
1390                 /* Unfortunately ll_iget will call ll_update_inode,
1391                  * where the initialization of slave inode is slightly
1392                  * different, so it reset lsm_md to NULL to avoid
1393                  * initializing lsm for slave inode. */
1394                 lsm->lsm_md_oinfo[i].lmo_root =
1395                                 ll_iget_anon_dir(inode->i_sb, fid, md);
1396                 if (IS_ERR(lsm->lsm_md_oinfo[i].lmo_root)) {
1397                         int rc = PTR_ERR(lsm->lsm_md_oinfo[i].lmo_root);
1398
1399                         lsm->lsm_md_oinfo[i].lmo_root = NULL;
1400                         while (i-- > 0) {
1401                                 iput(lsm->lsm_md_oinfo[i].lmo_root);
1402                                 lsm->lsm_md_oinfo[i].lmo_root = NULL;
1403                         }
1404                         return rc;
1405                 }
1406         }
1407
1408         lli->lli_lsm_md = lsm;
1409
1410         return 0;
1411 }
1412
1413 static int ll_update_lsm_md(struct inode *inode, struct lustre_md *md)
1414 {
1415         struct ll_inode_info *lli = ll_i2info(inode);
1416         struct lmv_stripe_md *lsm = md->lmv;
1417         int rc = 0;
1418
1419         ENTRY;
1420
1421         LASSERT(S_ISDIR(inode->i_mode));
1422         CDEBUG(D_INODE, "update lsm %p of "DFID"\n", lli->lli_lsm_md,
1423                PFID(ll_inode2fid(inode)));
1424
1425         /*
1426          * no striped information from request, lustre_md from req does not
1427          * include stripeEA, see ll_md_setattr()
1428          */
1429         if (!lsm)
1430                 RETURN(0);
1431
1432         /*
1433          * normally dir layout doesn't change, only take read lock to check
1434          * that to avoid blocking other MD operations.
1435          */
1436         if (lli->lli_lsm_md)
1437                 down_read(&lli->lli_lsm_sem);
1438         else
1439                 down_write(&lli->lli_lsm_sem);
1440
1441         /*
1442          * if dir layout mismatch, check whether version is increased, which
1443          * means layout is changed, this happens in dir migration and lfsck.
1444          *
1445          * foreign LMV should not change.
1446          */
1447         if (lli->lli_lsm_md &&
1448             lli->lli_lsm_md->lsm_md_magic != LMV_MAGIC_FOREIGN &&
1449            !lsm_md_eq(lli->lli_lsm_md, lsm)) {
1450                 if (lsm->lsm_md_layout_version <=
1451                     lli->lli_lsm_md->lsm_md_layout_version) {
1452                         CERROR("%s: "DFID" dir layout mismatch:\n",
1453                                ll_i2sbi(inode)->ll_fsname,
1454                                PFID(&lli->lli_fid));
1455                         lsm_md_dump(D_ERROR, lli->lli_lsm_md);
1456                         lsm_md_dump(D_ERROR, lsm);
1457                         GOTO(unlock, rc = -EINVAL);
1458                 }
1459
1460                 /* layout changed, switch to write lock */
1461                 up_read(&lli->lli_lsm_sem);
1462                 down_write(&lli->lli_lsm_sem);
1463                 ll_dir_clear_lsm_md(inode);
1464         }
1465
1466         /* set directory layout */
1467         if (!lli->lli_lsm_md) {
1468                 struct cl_attr  *attr;
1469
1470                 if (lsm->lsm_md_magic == LMV_MAGIC_FOREIGN) {
1471                         /* set md->lmv to NULL, so the following free lustre_md
1472                          * will not free this lsm */
1473                         md->lmv = NULL;
1474                         lli->lli_lsm_md = lsm;
1475                         up_write(&lli->lli_lsm_sem);
1476                         RETURN(0);
1477                 }
1478
1479                 rc = ll_init_lsm_md(inode, md);
1480                 up_write(&lli->lli_lsm_sem);
1481                 if (rc != 0)
1482                         RETURN(rc);
1483
1484                 /* set md->lmv to NULL, so the following free lustre_md
1485                  * will not free this lsm */
1486                 md->lmv = NULL;
1487
1488                 /*
1489                  * md_merge_attr() may take long, since lsm is already set,
1490                  * switch to read lock.
1491                  */
1492                 down_read(&lli->lli_lsm_sem);
1493
1494                 OBD_ALLOC_PTR(attr);
1495                 if (attr == NULL)
1496                         GOTO(unlock, rc = -ENOMEM);
1497
1498                 /* validate the lsm */
1499                 rc = md_merge_attr(ll_i2mdexp(inode), lsm, attr,
1500                                    ll_md_blocking_ast);
1501                 if (rc != 0) {
1502                         OBD_FREE_PTR(attr);
1503                         GOTO(unlock, rc);
1504                 }
1505
1506                 if (md->body->mbo_valid & OBD_MD_FLNLINK)
1507                         md->body->mbo_nlink = attr->cat_nlink;
1508                 if (md->body->mbo_valid & OBD_MD_FLSIZE)
1509                         md->body->mbo_size = attr->cat_size;
1510                 if (md->body->mbo_valid & OBD_MD_FLATIME)
1511                         md->body->mbo_atime = attr->cat_atime;
1512                 if (md->body->mbo_valid & OBD_MD_FLCTIME)
1513                         md->body->mbo_ctime = attr->cat_ctime;
1514                 if (md->body->mbo_valid & OBD_MD_FLMTIME)
1515                         md->body->mbo_mtime = attr->cat_mtime;
1516
1517                 OBD_FREE_PTR(attr);
1518         }
1519 unlock:
1520         up_read(&lli->lli_lsm_sem);
1521
1522         RETURN(rc);
1523 }
1524
1525 void ll_clear_inode(struct inode *inode)
1526 {
1527         struct ll_inode_info *lli = ll_i2info(inode);
1528         struct ll_sb_info *sbi = ll_i2sbi(inode);
1529
1530         ENTRY;
1531
1532         CDEBUG(D_VFSTRACE, "VFS Op:inode="DFID"(%p)\n",
1533                PFID(ll_inode2fid(inode)), inode);
1534
1535         if (S_ISDIR(inode->i_mode)) {
1536                 /* these should have been cleared in ll_file_release */
1537                 LASSERT(lli->lli_opendir_key == NULL);
1538                 LASSERT(lli->lli_sai == NULL);
1539                 LASSERT(lli->lli_opendir_pid == 0);
1540         } else {
1541                 pcc_inode_free(inode);
1542         }
1543
1544         md_null_inode(sbi->ll_md_exp, ll_inode2fid(inode));
1545
1546         LASSERT(!lli->lli_open_fd_write_count);
1547         LASSERT(!lli->lli_open_fd_read_count);
1548         LASSERT(!lli->lli_open_fd_exec_count);
1549
1550         if (lli->lli_mds_write_och)
1551                 ll_md_real_close(inode, FMODE_WRITE);
1552         if (lli->lli_mds_exec_och)
1553                 ll_md_real_close(inode, FMODE_EXEC);
1554         if (lli->lli_mds_read_och)
1555                 ll_md_real_close(inode, FMODE_READ);
1556
1557         if (S_ISLNK(inode->i_mode) && lli->lli_symlink_name) {
1558                 OBD_FREE(lli->lli_symlink_name,
1559                          strlen(lli->lli_symlink_name) + 1);
1560                 lli->lli_symlink_name = NULL;
1561         }
1562
1563         ll_xattr_cache_destroy(inode);
1564
1565 #ifdef CONFIG_FS_POSIX_ACL
1566         forget_all_cached_acls(inode);
1567         if (lli->lli_posix_acl) {
1568                 posix_acl_release(lli->lli_posix_acl);
1569                 lli->lli_posix_acl = NULL;
1570         }
1571 #endif
1572         lli->lli_inode_magic = LLI_INODE_DEAD;
1573
1574         if (S_ISDIR(inode->i_mode))
1575                 ll_dir_clear_lsm_md(inode);
1576         else if (S_ISREG(inode->i_mode) && !is_bad_inode(inode))
1577                 LASSERT(list_empty(&lli->lli_agl_list));
1578
1579         /*
1580          * XXX This has to be done before lsm is freed below, because
1581          * cl_object still uses inode lsm.
1582          */
1583         cl_inode_fini(inode);
1584
1585         EXIT;
1586 }
1587
1588 static int ll_md_setattr(struct dentry *dentry, struct md_op_data *op_data)
1589 {
1590         struct lustre_md md;
1591         struct inode *inode = dentry->d_inode;
1592         struct ll_sb_info *sbi = ll_i2sbi(inode);
1593         struct ptlrpc_request *request = NULL;
1594         int rc, ia_valid;
1595         ENTRY;
1596
1597         op_data = ll_prep_md_op_data(op_data, inode, NULL, NULL, 0, 0,
1598                                      LUSTRE_OPC_ANY, NULL);
1599         if (IS_ERR(op_data))
1600                 RETURN(PTR_ERR(op_data));
1601
1602         rc = md_setattr(sbi->ll_md_exp, op_data, NULL, 0, &request);
1603         if (rc) {
1604                 ptlrpc_req_finished(request);
1605                 if (rc == -ENOENT) {
1606                         clear_nlink(inode);
1607                         /* Unlinked special device node? Or just a race?
1608                          * Pretend we done everything. */
1609                         if (!S_ISREG(inode->i_mode) &&
1610                             !S_ISDIR(inode->i_mode)) {
1611                                 ia_valid = op_data->op_attr.ia_valid;
1612                                 op_data->op_attr.ia_valid &= ~TIMES_SET_FLAGS;
1613                                 rc = simple_setattr(dentry, &op_data->op_attr);
1614                                 op_data->op_attr.ia_valid = ia_valid;
1615                         }
1616                 } else if (rc != -EPERM && rc != -EACCES && rc != -ETXTBSY) {
1617                         CERROR("md_setattr fails: rc = %d\n", rc);
1618                 }
1619                 RETURN(rc);
1620         }
1621
1622         rc = md_get_lustre_md(sbi->ll_md_exp, request, sbi->ll_dt_exp,
1623                               sbi->ll_md_exp, &md);
1624         if (rc) {
1625                 ptlrpc_req_finished(request);
1626                 RETURN(rc);
1627         }
1628
1629         ia_valid = op_data->op_attr.ia_valid;
1630         /* inode size will be in ll_setattr_ost, can't do it now since dirty
1631          * cache is not cleared yet. */
1632         op_data->op_attr.ia_valid &= ~(TIMES_SET_FLAGS | ATTR_SIZE);
1633         if (S_ISREG(inode->i_mode))
1634                 inode_lock(inode);
1635         rc = simple_setattr(dentry, &op_data->op_attr);
1636         if (S_ISREG(inode->i_mode))
1637                 inode_unlock(inode);
1638         op_data->op_attr.ia_valid = ia_valid;
1639
1640         rc = ll_update_inode(inode, &md);
1641         ptlrpc_req_finished(request);
1642
1643         RETURN(rc);
1644 }
1645
1646 /* If this inode has objects allocated to it (lsm != NULL), then the OST
1647  * object(s) determine the file size and mtime.  Otherwise, the MDS will
1648  * keep these values until such a time that objects are allocated for it.
1649  * We do the MDS operations first, as it is checking permissions for us.
1650  * We don't to the MDS RPC if there is nothing that we want to store there,
1651  * otherwise there is no harm in updating mtime/atime on the MDS if we are
1652  * going to do an RPC anyways.
1653  *
1654  * If we are doing a truncate, we will send the mtime and ctime updates
1655  * to the OST with the punch RPC, otherwise we do an explicit setattr RPC.
1656  * I don't believe it is possible to get e.g. ATTR_MTIME_SET and ATTR_SIZE
1657  * at the same time.
1658  *
1659  * In case of HSMimport, we only set attr on MDS.
1660  */
1661 int ll_setattr_raw(struct dentry *dentry, struct iattr *attr,
1662                    enum op_xvalid xvalid, bool hsm_import)
1663 {
1664         struct inode *inode = dentry->d_inode;
1665         struct ll_inode_info *lli = ll_i2info(inode);
1666         struct md_op_data *op_data = NULL;
1667         int rc = 0;
1668
1669         ENTRY;
1670
1671         CDEBUG(D_VFSTRACE, "%s: setattr inode "DFID"(%p) from %llu to %llu, "
1672                "valid %x, hsm_import %d\n",
1673                ll_i2sbi(inode)->ll_fsname, PFID(&lli->lli_fid),
1674                inode, i_size_read(inode), attr->ia_size, attr->ia_valid,
1675                hsm_import);
1676
1677         if (attr->ia_valid & ATTR_SIZE) {
1678                 /* Check new size against VFS/VM file size limit and rlimit */
1679                 rc = inode_newsize_ok(inode, attr->ia_size);
1680                 if (rc)
1681                         RETURN(rc);
1682
1683                 /* The maximum Lustre file size is variable, based on the
1684                  * OST maximum object size and number of stripes.  This
1685                  * needs another check in addition to the VFS check above. */
1686                 if (attr->ia_size > ll_file_maxbytes(inode)) {
1687                         CDEBUG(D_INODE,"file "DFID" too large %llu > %llu\n",
1688                                PFID(&lli->lli_fid), attr->ia_size,
1689                                ll_file_maxbytes(inode));
1690                         RETURN(-EFBIG);
1691                 }
1692
1693                 attr->ia_valid |= ATTR_MTIME | ATTR_CTIME;
1694         }
1695
1696         /* POSIX: check before ATTR_*TIME_SET set (from inode_change_ok) */
1697         if (attr->ia_valid & TIMES_SET_FLAGS) {
1698                 if ((!uid_eq(current_fsuid(), inode->i_uid)) &&
1699                     !cfs_capable(CFS_CAP_FOWNER))
1700                         RETURN(-EPERM);
1701         }
1702
1703         /* We mark all of the fields "set" so MDS/OST does not re-set them */
1704         if (!(xvalid & OP_XVALID_CTIME_SET) &&
1705              (attr->ia_valid & ATTR_CTIME)) {
1706                 attr->ia_ctime = current_time(inode);
1707                 xvalid |= OP_XVALID_CTIME_SET;
1708         }
1709         if (!(attr->ia_valid & ATTR_ATIME_SET) &&
1710             (attr->ia_valid & ATTR_ATIME)) {
1711                 attr->ia_atime = current_time(inode);
1712                 attr->ia_valid |= ATTR_ATIME_SET;
1713         }
1714         if (!(attr->ia_valid & ATTR_MTIME_SET) &&
1715             (attr->ia_valid & ATTR_MTIME)) {
1716                 attr->ia_mtime = current_time(inode);
1717                 attr->ia_valid |= ATTR_MTIME_SET;
1718         }
1719
1720         if (attr->ia_valid & (ATTR_MTIME | ATTR_CTIME))
1721                 CDEBUG(D_INODE, "setting mtime %lld, ctime %lld, now = %lld\n",
1722                        (s64)attr->ia_mtime.tv_sec, (s64)attr->ia_ctime.tv_sec,
1723                        ktime_get_real_seconds());
1724
1725         if (S_ISREG(inode->i_mode)) {
1726                 if (attr->ia_valid & ATTR_SIZE)
1727                         inode_dio_write_done(inode);
1728                 inode_unlock(inode);
1729         }
1730
1731         /* We always do an MDS RPC, even if we're only changing the size;
1732          * only the MDS knows whether truncate() should fail with -ETXTBUSY */
1733
1734         OBD_ALLOC_PTR(op_data);
1735         if (op_data == NULL)
1736                 GOTO(out, rc = -ENOMEM);
1737
1738         if (!hsm_import && attr->ia_valid & ATTR_SIZE) {
1739                 /* If we are changing file size, file content is
1740                  * modified, flag it.
1741                  */
1742                 xvalid |= OP_XVALID_OWNEROVERRIDE;
1743                 op_data->op_bias |= MDS_DATA_MODIFIED;
1744                 ll_file_clear_flag(lli, LLIF_DATA_MODIFIED);
1745         }
1746
1747         if (attr->ia_valid & ATTR_FILE) {
1748                 struct ll_file_data *fd = LUSTRE_FPRIVATE(attr->ia_file);
1749
1750                 if (fd->fd_lease_och)
1751                         op_data->op_bias |= MDS_TRUNC_KEEP_LEASE;
1752         }
1753
1754         op_data->op_attr = *attr;
1755         op_data->op_xvalid = xvalid;
1756
1757         rc = ll_md_setattr(dentry, op_data);
1758         if (rc)
1759                 GOTO(out, rc);
1760
1761         if (!S_ISREG(inode->i_mode) || hsm_import)
1762                 GOTO(out, rc = 0);
1763
1764         if (attr->ia_valid & (ATTR_SIZE | ATTR_ATIME | ATTR_ATIME_SET |
1765                               ATTR_MTIME | ATTR_MTIME_SET | ATTR_CTIME) ||
1766             xvalid & OP_XVALID_CTIME_SET) {
1767                 bool cached = false;
1768
1769                 rc = pcc_inode_setattr(inode, attr, &cached);
1770                 if (cached) {
1771                         if (rc) {
1772                                 CERROR("%s: PCC inode "DFID" setattr failed: "
1773                                        "rc = %d\n",
1774                                        ll_i2sbi(inode)->ll_fsname,
1775                                        PFID(&lli->lli_fid), rc);
1776                                 GOTO(out, rc);
1777                         }
1778                 } else {
1779                         /* For truncate and utimes sending attributes to OSTs,
1780                          * setting mtime/atime to the past will be performed
1781                          * under PW [0:EOF] extent lock (new_size:EOF for
1782                          * truncate). It may seem excessive to send mtime/atime
1783                          * updates to OSTs when not setting times to past, but
1784                          * it is necessary due to possible time
1785                          * de-synchronization between MDT inode and OST objects
1786                          */
1787                         rc = cl_setattr_ost(lli->lli_clob, attr, xvalid, 0);
1788                 }
1789         }
1790
1791         /* If the file was restored, it needs to set dirty flag.
1792          *
1793          * We've already sent MDS_DATA_MODIFIED flag in
1794          * ll_md_setattr() for truncate. However, the MDT refuses to
1795          * set the HS_DIRTY flag on released files, so we have to set
1796          * it again if the file has been restored. Please check how
1797          * LLIF_DATA_MODIFIED is set in vvp_io_setattr_fini().
1798          *
1799          * Please notice that if the file is not released, the previous
1800          * MDS_DATA_MODIFIED has taken effect and usually
1801          * LLIF_DATA_MODIFIED is not set(see vvp_io_setattr_fini()).
1802          * This way we can save an RPC for common open + trunc
1803          * operation. */
1804         if (ll_file_test_and_clear_flag(lli, LLIF_DATA_MODIFIED)) {
1805                 struct hsm_state_set hss = {
1806                         .hss_valid = HSS_SETMASK,
1807                         .hss_setmask = HS_DIRTY,
1808                 };
1809                 int rc2;
1810
1811                 rc2 = ll_hsm_state_set(inode, &hss);
1812                 /* truncate and write can happen at the same time, so that
1813                  * the file can be set modified even though the file is not
1814                  * restored from released state, and ll_hsm_state_set() is
1815                  * not applicable for the file, and rc2 < 0 is normal in this
1816                  * case. */
1817                 if (rc2 < 0)
1818                         CDEBUG(D_INFO, DFID "HSM set dirty failed: rc2 = %d\n",
1819                                PFID(ll_inode2fid(inode)), rc2);
1820         }
1821
1822         EXIT;
1823 out:
1824         if (op_data != NULL)
1825                 ll_finish_md_op_data(op_data);
1826
1827         if (S_ISREG(inode->i_mode)) {
1828                 inode_lock(inode);
1829                 if ((attr->ia_valid & ATTR_SIZE) && !hsm_import)
1830                         inode_dio_wait(inode);
1831                 /* Once we've got the i_mutex, it's safe to set the S_NOSEC
1832                  * flag.  ll_update_inode (called from ll_md_setattr), clears
1833                  * inode flags, so there is a gap where S_NOSEC is not set.
1834                  * This can cause a writer to take the i_mutex unnecessarily,
1835                  * but this is safe to do and should be rare. */
1836                 inode_has_no_xattr(inode);
1837         }
1838
1839         ll_stats_ops_tally(ll_i2sbi(inode), (attr->ia_valid & ATTR_SIZE) ?
1840                         LPROC_LL_TRUNC : LPROC_LL_SETATTR, 1);
1841
1842         return rc;
1843 }
1844
1845 int ll_setattr(struct dentry *de, struct iattr *attr)
1846 {
1847         int mode = de->d_inode->i_mode;
1848         enum op_xvalid xvalid = 0;
1849
1850         if ((attr->ia_valid & (ATTR_CTIME|ATTR_SIZE|ATTR_MODE)) ==
1851                               (ATTR_CTIME|ATTR_SIZE|ATTR_MODE))
1852                 xvalid |= OP_XVALID_OWNEROVERRIDE;
1853
1854         if (((attr->ia_valid & (ATTR_MODE|ATTR_FORCE|ATTR_SIZE)) ==
1855                                (ATTR_SIZE|ATTR_MODE)) &&
1856             (((mode & S_ISUID) && !(attr->ia_mode & S_ISUID)) ||
1857              (((mode & (S_ISGID|S_IXGRP)) == (S_ISGID|S_IXGRP)) &&
1858               !(attr->ia_mode & S_ISGID))))
1859                 attr->ia_valid |= ATTR_FORCE;
1860
1861         if ((attr->ia_valid & ATTR_MODE) &&
1862             (mode & S_ISUID) &&
1863             !(attr->ia_mode & S_ISUID) &&
1864             !(attr->ia_valid & ATTR_KILL_SUID))
1865                 attr->ia_valid |= ATTR_KILL_SUID;
1866
1867         if ((attr->ia_valid & ATTR_MODE) &&
1868             ((mode & (S_ISGID|S_IXGRP)) == (S_ISGID|S_IXGRP)) &&
1869             !(attr->ia_mode & S_ISGID) &&
1870             !(attr->ia_valid & ATTR_KILL_SGID))
1871                 attr->ia_valid |= ATTR_KILL_SGID;
1872
1873         return ll_setattr_raw(de, attr, xvalid, false);
1874 }
1875
1876 int ll_statfs_internal(struct ll_sb_info *sbi, struct obd_statfs *osfs,
1877                        u32 flags)
1878 {
1879         struct obd_statfs obd_osfs = { 0 };
1880         time64_t max_age;
1881         int rc;
1882
1883         ENTRY;
1884         max_age = ktime_get_seconds() - OBD_STATFS_CACHE_SECONDS;
1885
1886         rc = obd_statfs(NULL, sbi->ll_md_exp, osfs, max_age, flags);
1887         if (rc)
1888                 RETURN(rc);
1889
1890         osfs->os_type = LL_SUPER_MAGIC;
1891
1892         CDEBUG(D_SUPER, "MDC blocks %llu/%llu objects %llu/%llu\n",
1893               osfs->os_bavail, osfs->os_blocks, osfs->os_ffree, osfs->os_files);
1894
1895         if (osfs->os_state & OS_STATE_SUM)
1896                 GOTO(out, rc);
1897
1898         if (sbi->ll_flags & LL_SBI_LAZYSTATFS)
1899                 flags |= OBD_STATFS_NODELAY;
1900
1901         rc = obd_statfs(NULL, sbi->ll_dt_exp, &obd_osfs, max_age, flags);
1902         if (rc) /* Possibly a filesystem with no OSTs.  Report MDT totals. */
1903                 GOTO(out, rc = 0);
1904
1905         CDEBUG(D_SUPER, "OSC blocks %llu/%llu objects %llu/%llu\n",
1906                obd_osfs.os_bavail, obd_osfs.os_blocks, obd_osfs.os_ffree,
1907                obd_osfs.os_files);
1908
1909         osfs->os_bsize = obd_osfs.os_bsize;
1910         osfs->os_blocks = obd_osfs.os_blocks;
1911         osfs->os_bfree = obd_osfs.os_bfree;
1912         osfs->os_bavail = obd_osfs.os_bavail;
1913
1914         /* If we have _some_ OSTs, but don't have as many free objects on the
1915          * OSTs as inodes on the MDTs, reduce the reported number of inodes
1916          * to compensate, so that the "inodes in use" number is correct.
1917          * This should be kept in sync with lod_statfs() behaviour.
1918          */
1919         if (obd_osfs.os_files && obd_osfs.os_ffree < osfs->os_ffree) {
1920                 osfs->os_files = (osfs->os_files - osfs->os_ffree) +
1921                                  obd_osfs.os_ffree;
1922                 osfs->os_ffree = obd_osfs.os_ffree;
1923         }
1924
1925 out:
1926         RETURN(rc);
1927 }
1928 int ll_statfs(struct dentry *de, struct kstatfs *sfs)
1929 {
1930         struct super_block *sb = de->d_sb;
1931         struct obd_statfs osfs;
1932         __u64 fsid = huge_encode_dev(sb->s_dev);
1933         int rc;
1934
1935         CDEBUG(D_VFSTRACE, "VFS Op: at %llu jiffies\n", get_jiffies_64());
1936         ll_stats_ops_tally(ll_s2sbi(sb), LPROC_LL_STAFS, 1);
1937
1938         /* Some amount of caching on the client is allowed */
1939         rc = ll_statfs_internal(ll_s2sbi(sb), &osfs, OBD_STATFS_SUM);
1940         if (rc)
1941                 return rc;
1942
1943         statfs_unpack(sfs, &osfs);
1944
1945         /* We need to downshift for all 32-bit kernels, because we can't
1946          * tell if the kernel is being called via sys_statfs64() or not.
1947          * Stop before overflowing f_bsize - in which case it is better
1948          * to just risk EOVERFLOW if caller is using old sys_statfs(). */
1949         if (sizeof(long) < 8) {
1950                 while (osfs.os_blocks > ~0UL && sfs->f_bsize < 0x40000000) {
1951                         sfs->f_bsize <<= 1;
1952
1953                         osfs.os_blocks >>= 1;
1954                         osfs.os_bfree >>= 1;
1955                         osfs.os_bavail >>= 1;
1956                 }
1957         }
1958
1959         sfs->f_blocks = osfs.os_blocks;
1960         sfs->f_bfree = osfs.os_bfree;
1961         sfs->f_bavail = osfs.os_bavail;
1962         sfs->f_fsid.val[0] = (__u32)fsid;
1963         sfs->f_fsid.val[1] = (__u32)(fsid >> 32);
1964         return 0;
1965 }
1966
1967 void ll_inode_size_lock(struct inode *inode)
1968 {
1969         struct ll_inode_info *lli;
1970
1971         LASSERT(!S_ISDIR(inode->i_mode));
1972
1973         lli = ll_i2info(inode);
1974         mutex_lock(&lli->lli_size_mutex);
1975 }
1976
1977 void ll_inode_size_unlock(struct inode *inode)
1978 {
1979         struct ll_inode_info *lli;
1980
1981         lli = ll_i2info(inode);
1982         mutex_unlock(&lli->lli_size_mutex);
1983 }
1984
1985 void ll_update_inode_flags(struct inode *inode, int ext_flags)
1986 {
1987         inode->i_flags = ll_ext_to_inode_flags(ext_flags);
1988         if (ext_flags & LUSTRE_PROJINHERIT_FL)
1989                 ll_file_set_flag(ll_i2info(inode), LLIF_PROJECT_INHERIT);
1990         else
1991                 ll_file_clear_flag(ll_i2info(inode), LLIF_PROJECT_INHERIT);
1992 }
1993
1994 int ll_update_inode(struct inode *inode, struct lustre_md *md)
1995 {
1996         struct ll_inode_info *lli = ll_i2info(inode);
1997         struct mdt_body *body = md->body;
1998         struct ll_sb_info *sbi = ll_i2sbi(inode);
1999         int rc = 0;
2000
2001         if (body->mbo_valid & OBD_MD_FLEASIZE) {
2002                 rc = cl_file_inode_init(inode, md);
2003                 if (rc)
2004                         return rc;
2005         }
2006
2007         if (S_ISDIR(inode->i_mode)) {
2008                 rc = ll_update_lsm_md(inode, md);
2009                 if (rc != 0)
2010                         return rc;
2011         }
2012
2013 #ifdef CONFIG_FS_POSIX_ACL
2014         if (body->mbo_valid & OBD_MD_FLACL) {
2015                 spin_lock(&lli->lli_lock);
2016                 if (lli->lli_posix_acl)
2017                         posix_acl_release(lli->lli_posix_acl);
2018                 lli->lli_posix_acl = md->posix_acl;
2019                 spin_unlock(&lli->lli_lock);
2020         }
2021 #endif
2022         inode->i_ino = cl_fid_build_ino(&body->mbo_fid1,
2023                                         sbi->ll_flags & LL_SBI_32BIT_API);
2024         inode->i_generation = cl_fid_build_gen(&body->mbo_fid1);
2025
2026         if (body->mbo_valid & OBD_MD_FLATIME) {
2027                 if (body->mbo_atime > inode->i_atime.tv_sec)
2028                         inode->i_atime.tv_sec = body->mbo_atime;
2029                 lli->lli_atime = body->mbo_atime;
2030         }
2031
2032         if (body->mbo_valid & OBD_MD_FLMTIME) {
2033                 if (body->mbo_mtime > inode->i_mtime.tv_sec) {
2034                         CDEBUG(D_INODE,
2035                                "setting ino %lu mtime from %lld to %llu\n",
2036                                inode->i_ino, (s64)inode->i_mtime.tv_sec,
2037                                body->mbo_mtime);
2038                         inode->i_mtime.tv_sec = body->mbo_mtime;
2039                 }
2040                 lli->lli_mtime = body->mbo_mtime;
2041         }
2042
2043         if (body->mbo_valid & OBD_MD_FLCTIME) {
2044                 if (body->mbo_ctime > inode->i_ctime.tv_sec)
2045                         inode->i_ctime.tv_sec = body->mbo_ctime;
2046                 lli->lli_ctime = body->mbo_ctime;
2047         }
2048
2049         /* Clear i_flags to remove S_NOSEC before permissions are updated */
2050         if (body->mbo_valid & OBD_MD_FLFLAGS)
2051                 ll_update_inode_flags(inode, body->mbo_flags);
2052         if (body->mbo_valid & OBD_MD_FLMODE)
2053                 inode->i_mode = (inode->i_mode & S_IFMT) |
2054                                 (body->mbo_mode & ~S_IFMT);
2055
2056         if (body->mbo_valid & OBD_MD_FLTYPE)
2057                 inode->i_mode = (inode->i_mode & ~S_IFMT) |
2058                                 (body->mbo_mode & S_IFMT);
2059
2060         LASSERT(inode->i_mode != 0);
2061         if (S_ISREG(inode->i_mode))
2062                 inode->i_blkbits = min(PTLRPC_MAX_BRW_BITS + 1,
2063                                        LL_MAX_BLKSIZE_BITS);
2064         else
2065                 inode->i_blkbits = inode->i_sb->s_blocksize_bits;
2066
2067         if (body->mbo_valid & OBD_MD_FLUID)
2068                 inode->i_uid = make_kuid(&init_user_ns, body->mbo_uid);
2069         if (body->mbo_valid & OBD_MD_FLGID)
2070                 inode->i_gid = make_kgid(&init_user_ns, body->mbo_gid);
2071         if (body->mbo_valid & OBD_MD_FLPROJID)
2072                 lli->lli_projid = body->mbo_projid;
2073         if (body->mbo_valid & OBD_MD_FLNLINK)
2074                 set_nlink(inode, body->mbo_nlink);
2075         if (body->mbo_valid & OBD_MD_FLRDEV)
2076                 inode->i_rdev = old_decode_dev(body->mbo_rdev);
2077
2078         if (body->mbo_valid & OBD_MD_FLID) {
2079                 /* FID shouldn't be changed! */
2080                 if (fid_is_sane(&lli->lli_fid)) {
2081                         LASSERTF(lu_fid_eq(&lli->lli_fid, &body->mbo_fid1),
2082                                  "Trying to change FID "DFID
2083                                  " to the "DFID", inode "DFID"(%p)\n",
2084                                  PFID(&lli->lli_fid), PFID(&body->mbo_fid1),
2085                                  PFID(ll_inode2fid(inode)), inode);
2086                 } else {
2087                         lli->lli_fid = body->mbo_fid1;
2088                 }
2089         }
2090
2091         LASSERT(fid_seq(&lli->lli_fid) != 0);
2092
2093         if (body->mbo_valid & OBD_MD_FLSIZE) {
2094                 i_size_write(inode, body->mbo_size);
2095
2096                 CDEBUG(D_VFSTRACE, "inode="DFID", updating i_size %llu\n",
2097                        PFID(ll_inode2fid(inode)),
2098                        (unsigned long long)body->mbo_size);
2099
2100                 if (body->mbo_valid & OBD_MD_FLBLOCKS)
2101                         inode->i_blocks = body->mbo_blocks;
2102         }
2103
2104         if (body->mbo_valid & OBD_MD_TSTATE) {
2105                 /* Set LLIF_FILE_RESTORING if restore ongoing and
2106                  * clear it when done to ensure to start again
2107                  * glimpsing updated attrs
2108                  */
2109                 if (body->mbo_t_state & MS_RESTORE)
2110                         ll_file_set_flag(lli, LLIF_FILE_RESTORING);
2111                 else
2112                         ll_file_clear_flag(lli, LLIF_FILE_RESTORING);
2113         }
2114
2115         return 0;
2116 }
2117
2118 int ll_read_inode2(struct inode *inode, void *opaque)
2119 {
2120         struct lustre_md *md = opaque;
2121         struct ll_inode_info *lli = ll_i2info(inode);
2122         int     rc;
2123         ENTRY;
2124
2125         CDEBUG(D_VFSTRACE, "VFS Op:inode="DFID"(%p)\n",
2126                PFID(&lli->lli_fid), inode);
2127
2128         /* Core attributes from the MDS first.  This is a new inode, and
2129          * the VFS doesn't zero times in the core inode so we have to do
2130          * it ourselves.  They will be overwritten by either MDS or OST
2131          * attributes - we just need to make sure they aren't newer.
2132          */
2133         inode->i_mtime.tv_sec = 0;
2134         inode->i_atime.tv_sec = 0;
2135         inode->i_ctime.tv_sec = 0;
2136         inode->i_rdev = 0;
2137         rc = ll_update_inode(inode, md);
2138         if (rc != 0)
2139                 RETURN(rc);
2140
2141         /* OIDEBUG(inode); */
2142
2143 #ifdef HAVE_BACKING_DEV_INFO
2144         /* initializing backing dev info. */
2145         inode->i_mapping->backing_dev_info = &s2lsi(inode->i_sb)->lsi_bdi;
2146 #endif
2147         if (S_ISREG(inode->i_mode)) {
2148                 struct ll_sb_info *sbi = ll_i2sbi(inode);
2149                 inode->i_op = &ll_file_inode_operations;
2150                 inode->i_fop = sbi->ll_fop;
2151                 inode->i_mapping->a_ops = (struct address_space_operations *)&ll_aops;
2152                 EXIT;
2153         } else if (S_ISDIR(inode->i_mode)) {
2154                 inode->i_op = &ll_dir_inode_operations;
2155                 inode->i_fop = &ll_dir_operations;
2156                 EXIT;
2157         } else if (S_ISLNK(inode->i_mode)) {
2158                 inode->i_op = &ll_fast_symlink_inode_operations;
2159                 EXIT;
2160         } else {
2161                 inode->i_op = &ll_special_inode_operations;
2162
2163                 init_special_inode(inode, inode->i_mode,
2164                                    inode->i_rdev);
2165
2166                 EXIT;
2167         }
2168
2169         return 0;
2170 }
2171
2172 void ll_delete_inode(struct inode *inode)
2173 {
2174         struct ll_inode_info *lli = ll_i2info(inode);
2175         struct address_space *mapping = &inode->i_data;
2176         unsigned long nrpages;
2177         ENTRY;
2178
2179         if (S_ISREG(inode->i_mode) && lli->lli_clob != NULL) {
2180                 /* It is last chance to write out dirty pages,
2181                  * otherwise we may lose data while umount.
2182                  *
2183                  * If i_nlink is 0 then just discard data. This is safe because
2184                  * local inode gets i_nlink 0 from server only for the last
2185                  * unlink, so that file is not opened somewhere else
2186                  */
2187                 cl_sync_file_range(inode, 0, OBD_OBJECT_EOF, inode->i_nlink ?
2188                                    CL_FSYNC_LOCAL : CL_FSYNC_DISCARD, 1);
2189         }
2190         truncate_inode_pages_final(mapping);
2191
2192         /* Workaround for LU-118: Note nrpages may not be totally updated when
2193          * truncate_inode_pages() returns, as there can be a page in the process
2194          * of deletion (inside __delete_from_page_cache()) in the specified
2195          * range. Thus mapping->nrpages can be non-zero when this function
2196          * returns even after truncation of the whole mapping.  Only do this if
2197          * npages isn't already zero.
2198          */
2199         nrpages = mapping->nrpages;
2200         if (nrpages) {
2201                 xa_lock_irq(&mapping->i_pages);
2202                 nrpages = mapping->nrpages;
2203                 xa_unlock_irq(&mapping->i_pages);
2204         } /* Workaround end */
2205
2206         LASSERTF(nrpages == 0, "%s: inode="DFID"(%p) nrpages=%lu, "
2207                  "see https://jira.whamcloud.com/browse/LU-118\n",
2208                  ll_i2sbi(inode)->ll_fsname,
2209                  PFID(ll_inode2fid(inode)), inode, nrpages);
2210
2211 #ifdef HAVE_SBOPS_EVICT_INODE
2212         ll_clear_inode(inode);
2213 #endif
2214         clear_inode(inode);
2215
2216         EXIT;
2217 }
2218
2219 int ll_iocontrol(struct inode *inode, struct file *file,
2220                  unsigned int cmd, unsigned long arg)
2221 {
2222         struct ll_sb_info *sbi = ll_i2sbi(inode);
2223         struct ptlrpc_request *req = NULL;
2224         int rc, flags = 0;
2225         ENTRY;
2226
2227         switch (cmd) {
2228         case FS_IOC_GETFLAGS: {
2229                 struct mdt_body *body;
2230                 struct md_op_data *op_data;
2231
2232                 op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL,
2233                                              0, 0, LUSTRE_OPC_ANY,
2234                                              NULL);
2235                 if (IS_ERR(op_data))
2236                         RETURN(PTR_ERR(op_data));
2237
2238                 op_data->op_valid = OBD_MD_FLFLAGS;
2239                 rc = md_getattr(sbi->ll_md_exp, op_data, &req);
2240                 ll_finish_md_op_data(op_data);
2241                 if (rc) {
2242                         CERROR("%s: failure inode "DFID": rc = %d\n",
2243                                sbi->ll_md_exp->exp_obd->obd_name,
2244                                PFID(ll_inode2fid(inode)), rc);
2245                         RETURN(-abs(rc));
2246                 }
2247
2248                 body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
2249
2250                 flags = body->mbo_flags;
2251
2252                 ptlrpc_req_finished(req);
2253
2254                 RETURN(put_user(flags, (int __user *)arg));
2255         }
2256         case FS_IOC_SETFLAGS: {
2257                 struct iattr *attr;
2258                 struct md_op_data *op_data;
2259                 struct cl_object *obj;
2260                 struct fsxattr fa = { 0 };
2261
2262                 if (get_user(flags, (int __user *)arg))
2263                         RETURN(-EFAULT);
2264
2265                 fa.fsx_projid = ll_i2info(inode)->lli_projid;
2266                 if (flags & LUSTRE_PROJINHERIT_FL)
2267                         fa.fsx_xflags = FS_XFLAG_PROJINHERIT;
2268
2269                 rc = ll_ioctl_check_project(inode, &fa);
2270                 if (rc)
2271                         RETURN(rc);
2272
2273                 op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL, 0, 0,
2274                                              LUSTRE_OPC_ANY, NULL);
2275                 if (IS_ERR(op_data))
2276                         RETURN(PTR_ERR(op_data));
2277
2278                 op_data->op_attr_flags = flags;
2279                 op_data->op_xvalid |= OP_XVALID_FLAGS;
2280                 rc = md_setattr(sbi->ll_md_exp, op_data, NULL, 0, &req);
2281                 ll_finish_md_op_data(op_data);
2282                 ptlrpc_req_finished(req);
2283                 if (rc)
2284                         RETURN(rc);
2285
2286                 ll_update_inode_flags(inode, flags);
2287
2288                 obj = ll_i2info(inode)->lli_clob;
2289                 if (obj == NULL)
2290                         RETURN(0);
2291
2292                 OBD_ALLOC_PTR(attr);
2293                 if (attr == NULL)
2294                         RETURN(-ENOMEM);
2295
2296                 rc = cl_setattr_ost(obj, attr, OP_XVALID_FLAGS, flags);
2297
2298                 OBD_FREE_PTR(attr);
2299                 RETURN(rc);
2300         }
2301         default:
2302                 RETURN(-ENOSYS);
2303         }
2304
2305         RETURN(0);
2306 }
2307
2308 int ll_flush_ctx(struct inode *inode)
2309 {
2310         struct ll_sb_info  *sbi = ll_i2sbi(inode);
2311
2312         CDEBUG(D_SEC, "flush context for user %d\n",
2313                from_kuid(&init_user_ns, current_uid()));
2314
2315         obd_set_info_async(NULL, sbi->ll_md_exp,
2316                            sizeof(KEY_FLUSH_CTX), KEY_FLUSH_CTX,
2317                            0, NULL, NULL);
2318         obd_set_info_async(NULL, sbi->ll_dt_exp,
2319                            sizeof(KEY_FLUSH_CTX), KEY_FLUSH_CTX,
2320                            0, NULL, NULL);
2321         return 0;
2322 }
2323
2324 /* umount -f client means force down, don't save state */
2325 void ll_umount_begin(struct super_block *sb)
2326 {
2327         struct ll_sb_info *sbi = ll_s2sbi(sb);
2328         struct obd_device *obd;
2329         struct obd_ioctl_data *ioc_data;
2330         struct l_wait_info lwi;
2331         wait_queue_head_t waitq;
2332         ENTRY;
2333
2334         CDEBUG(D_VFSTRACE, "VFS Op: superblock %p count %d active %d\n", sb,
2335                sb->s_count, atomic_read(&sb->s_active));
2336
2337         obd = class_exp2obd(sbi->ll_md_exp);
2338         if (obd == NULL) {
2339                 CERROR("Invalid MDC connection handle %#llx\n",
2340                        sbi->ll_md_exp->exp_handle.h_cookie);
2341                 EXIT;
2342                 return;
2343         }
2344         obd->obd_force = 1;
2345
2346         obd = class_exp2obd(sbi->ll_dt_exp);
2347         if (obd == NULL) {
2348                 CERROR("Invalid LOV connection handle %#llx\n",
2349                        sbi->ll_dt_exp->exp_handle.h_cookie);
2350                 EXIT;
2351                 return;
2352         }
2353         obd->obd_force = 1;
2354
2355         OBD_ALLOC_PTR(ioc_data);
2356         if (ioc_data) {
2357                 obd_iocontrol(IOC_OSC_SET_ACTIVE, sbi->ll_md_exp,
2358                               sizeof *ioc_data, ioc_data, NULL);
2359
2360                 obd_iocontrol(IOC_OSC_SET_ACTIVE, sbi->ll_dt_exp,
2361                               sizeof *ioc_data, ioc_data, NULL);
2362
2363                 OBD_FREE_PTR(ioc_data);
2364         }
2365
2366         /* Really, we'd like to wait until there are no requests outstanding,
2367          * and then continue.  For now, we just periodically checking for vfs
2368          * to decrement mnt_cnt and hope to finish it within 10sec.
2369          */
2370         init_waitqueue_head(&waitq);
2371         lwi = LWI_TIMEOUT_INTERVAL(cfs_time_seconds(10),
2372                                    cfs_time_seconds(1), NULL, NULL);
2373         l_wait_event(waitq, may_umount(sbi->ll_mnt.mnt), &lwi);
2374
2375         EXIT;
2376 }
2377
2378 int ll_remount_fs(struct super_block *sb, int *flags, char *data)
2379 {
2380         struct ll_sb_info *sbi = ll_s2sbi(sb);
2381         char *profilenm = get_profile_name(sb);
2382         int err;
2383         __u32 read_only;
2384
2385         if ((*flags & MS_RDONLY) != (sb->s_flags & MS_RDONLY)) {
2386                 read_only = *flags & MS_RDONLY;
2387                 err = obd_set_info_async(NULL, sbi->ll_md_exp,
2388                                          sizeof(KEY_READ_ONLY),
2389                                          KEY_READ_ONLY, sizeof(read_only),
2390                                          &read_only, NULL);
2391                 if (err) {
2392                         LCONSOLE_WARN("Failed to remount %s %s (%d)\n",
2393                                       profilenm, read_only ?
2394                                       "read-only" : "read-write", err);
2395                         return err;
2396                 }
2397
2398                 if (read_only)
2399                         sb->s_flags |= MS_RDONLY;
2400                 else
2401                         sb->s_flags &= ~MS_RDONLY;
2402
2403                 if (sbi->ll_flags & LL_SBI_VERBOSE)
2404                         LCONSOLE_WARN("Remounted %s %s\n", profilenm,
2405                                       read_only ?  "read-only" : "read-write");
2406         }
2407         return 0;
2408 }
2409
2410 /**
2411  * Cleanup the open handle that is cached on MDT-side.
2412  *
2413  * For open case, the client side open handling thread may hit error
2414  * after the MDT grant the open. Under such case, the client should
2415  * send close RPC to the MDT as cleanup; otherwise, the open handle
2416  * on the MDT will be leaked there until the client umount or evicted.
2417  *
2418  * In further, if someone unlinked the file, because the open handle
2419  * holds the reference on such file/object, then it will block the
2420  * subsequent threads that want to locate such object via FID.
2421  *
2422  * \param[in] sb        super block for this file-system
2423  * \param[in] open_req  pointer to the original open request
2424  */
2425 void ll_open_cleanup(struct super_block *sb, struct ptlrpc_request *open_req)
2426 {
2427         struct mdt_body                 *body;
2428         struct md_op_data               *op_data;
2429         struct ptlrpc_request           *close_req = NULL;
2430         struct obd_export               *exp       = ll_s2sbi(sb)->ll_md_exp;
2431         ENTRY;
2432
2433         body = req_capsule_server_get(&open_req->rq_pill, &RMF_MDT_BODY);
2434         OBD_ALLOC_PTR(op_data);
2435         if (op_data == NULL) {
2436                 CWARN("%s: cannot allocate op_data to release open handle for "
2437                       DFID"\n", ll_s2sbi(sb)->ll_fsname, PFID(&body->mbo_fid1));
2438
2439                 RETURN_EXIT;
2440         }
2441
2442         op_data->op_fid1 = body->mbo_fid1;
2443         op_data->op_open_handle = body->mbo_open_handle;
2444         op_data->op_mod_time = ktime_get_real_seconds();
2445         md_close(exp, op_data, NULL, &close_req);
2446         ptlrpc_req_finished(close_req);
2447         ll_finish_md_op_data(op_data);
2448
2449         EXIT;
2450 }
2451
2452 int ll_prep_inode(struct inode **inode, struct ptlrpc_request *req,
2453                   struct super_block *sb, struct lookup_intent *it)
2454 {
2455         struct ll_sb_info *sbi = NULL;
2456         struct lustre_md md = { NULL };
2457         int rc;
2458         ENTRY;
2459
2460         LASSERT(*inode || sb);
2461         sbi = sb ? ll_s2sbi(sb) : ll_i2sbi(*inode);
2462         rc = md_get_lustre_md(sbi->ll_md_exp, req, sbi->ll_dt_exp,
2463                               sbi->ll_md_exp, &md);
2464         if (rc != 0)
2465                 GOTO(out, rc);
2466
2467         if (*inode) {
2468                 rc = ll_update_inode(*inode, &md);
2469                 if (rc != 0)
2470                         GOTO(out, rc);
2471         } else {
2472                 LASSERT(sb != NULL);
2473
2474                 /*
2475                  * At this point server returns to client's same fid as client
2476                  * generated for creating. So using ->fid1 is okay here.
2477                  */
2478                 if (!fid_is_sane(&md.body->mbo_fid1)) {
2479                         CERROR("%s: Fid is insane "DFID"\n",
2480                                 sbi->ll_fsname,
2481                                 PFID(&md.body->mbo_fid1));
2482                         GOTO(out, rc = -EINVAL);
2483                 }
2484
2485                 *inode = ll_iget(sb, cl_fid_build_ino(&md.body->mbo_fid1,
2486                                              sbi->ll_flags & LL_SBI_32BIT_API),
2487                                  &md);
2488                 if (IS_ERR(*inode)) {
2489 #ifdef CONFIG_FS_POSIX_ACL
2490                         if (md.posix_acl) {
2491                                 posix_acl_release(md.posix_acl);
2492                                 md.posix_acl = NULL;
2493                         }
2494 #endif
2495                         rc = IS_ERR(*inode) ? PTR_ERR(*inode) : -ENOMEM;
2496                         *inode = NULL;
2497                         CERROR("new_inode -fatal: rc %d\n", rc);
2498                         GOTO(out, rc);
2499                 }
2500         }
2501
2502         /* Handling piggyback layout lock.
2503          * Layout lock can be piggybacked by getattr and open request.
2504          * The lsm can be applied to inode only if it comes with a layout lock
2505          * otherwise correct layout may be overwritten, for example:
2506          * 1. proc1: mdt returns a lsm but not granting layout
2507          * 2. layout was changed by another client
2508          * 3. proc2: refresh layout and layout lock granted
2509          * 4. proc1: to apply a stale layout */
2510         if (it != NULL && it->it_lock_mode != 0) {
2511                 struct lustre_handle lockh;
2512                 struct ldlm_lock *lock;
2513
2514                 lockh.cookie = it->it_lock_handle;
2515                 lock = ldlm_handle2lock(&lockh);
2516                 LASSERT(lock != NULL);
2517                 if (ldlm_has_layout(lock)) {
2518                         struct cl_object_conf conf;
2519
2520                         memset(&conf, 0, sizeof(conf));
2521                         conf.coc_opc = OBJECT_CONF_SET;
2522                         conf.coc_inode = *inode;
2523                         conf.coc_lock = lock;
2524                         conf.u.coc_layout = md.layout;
2525                         (void)ll_layout_conf(*inode, &conf);
2526                 }
2527                 LDLM_LOCK_PUT(lock);
2528         }
2529
2530         GOTO(out, rc = 0);
2531
2532 out:
2533         /* cleanup will be done if necessary */
2534         md_free_lustre_md(sbi->ll_md_exp, &md);
2535
2536         if (rc != 0 && it != NULL && it->it_op & IT_OPEN)
2537                 ll_open_cleanup(sb != NULL ? sb : (*inode)->i_sb, req);
2538
2539         return rc;
2540 }
2541
2542 int ll_obd_statfs(struct inode *inode, void __user *arg)
2543 {
2544         struct ll_sb_info *sbi = NULL;
2545         struct obd_export *exp;
2546         char *buf = NULL;
2547         struct obd_ioctl_data *data = NULL;
2548         __u32 type;
2549         int len = 0, rc;
2550
2551         if (!inode || !(sbi = ll_i2sbi(inode)))
2552                 GOTO(out_statfs, rc = -EINVAL);
2553
2554         rc = obd_ioctl_getdata(&buf, &len, arg);
2555         if (rc)
2556                 GOTO(out_statfs, rc);
2557
2558         data = (void*)buf;
2559         if (!data->ioc_inlbuf1 || !data->ioc_inlbuf2 ||
2560             !data->ioc_pbuf1 || !data->ioc_pbuf2)
2561                 GOTO(out_statfs, rc = -EINVAL);
2562
2563         if (data->ioc_inllen1 != sizeof(__u32) ||
2564             data->ioc_inllen2 != sizeof(__u32) ||
2565             data->ioc_plen1 != sizeof(struct obd_statfs) ||
2566             data->ioc_plen2 != sizeof(struct obd_uuid))
2567                 GOTO(out_statfs, rc = -EINVAL);
2568
2569         memcpy(&type, data->ioc_inlbuf1, sizeof(__u32));
2570         if (type & LL_STATFS_LMV)
2571                 exp = sbi->ll_md_exp;
2572         else if (type & LL_STATFS_LOV)
2573                 exp = sbi->ll_dt_exp;
2574         else
2575                 GOTO(out_statfs, rc = -ENODEV);
2576
2577         rc = obd_iocontrol(IOC_OBD_STATFS, exp, len, buf, NULL);
2578         if (rc)
2579                 GOTO(out_statfs, rc);
2580 out_statfs:
2581         OBD_FREE_LARGE(buf, len);
2582         return rc;
2583 }
2584
2585 /*
2586  * this is normally called in ll_fini_md_op_data(), but sometimes it needs to
2587  * be called early to avoid deadlock.
2588  */
2589 void ll_unlock_md_op_lsm(struct md_op_data *op_data)
2590 {
2591         if (op_data->op_mea2_sem) {
2592                 up_read(op_data->op_mea2_sem);
2593                 op_data->op_mea2_sem = NULL;
2594         }
2595
2596         if (op_data->op_mea1_sem) {
2597                 up_read(op_data->op_mea1_sem);
2598                 op_data->op_mea1_sem = NULL;
2599         }
2600 }
2601
2602 /* this function prepares md_op_data hint for passing it down to MD stack. */
2603 struct md_op_data *ll_prep_md_op_data(struct md_op_data *op_data,
2604                                       struct inode *i1, struct inode *i2,
2605                                       const char *name, size_t namelen,
2606                                       __u32 mode, __u32 opc, void *data)
2607 {
2608         LASSERT(i1 != NULL);
2609
2610         if (name == NULL) {
2611                 /* Do not reuse namelen for something else. */
2612                 if (namelen != 0)
2613                         return ERR_PTR(-EINVAL);
2614         } else {
2615                 if (namelen > ll_i2sbi(i1)->ll_namelen)
2616                         return ERR_PTR(-ENAMETOOLONG);
2617
2618                 if (!lu_name_is_valid_2(name, namelen))
2619                         return ERR_PTR(-EINVAL);
2620         }
2621
2622         if (op_data == NULL)
2623                 OBD_ALLOC_PTR(op_data);
2624
2625         if (op_data == NULL)
2626                 return ERR_PTR(-ENOMEM);
2627
2628         ll_i2gids(op_data->op_suppgids, i1, i2);
2629         op_data->op_fid1 = *ll_inode2fid(i1);
2630         op_data->op_default_stripe_offset = -1;
2631
2632         if (S_ISDIR(i1->i_mode)) {
2633                 down_read(&ll_i2info(i1)->lli_lsm_sem);
2634                 op_data->op_mea1_sem = &ll_i2info(i1)->lli_lsm_sem;
2635                 op_data->op_mea1 = ll_i2info(i1)->lli_lsm_md;
2636                 if (opc == LUSTRE_OPC_MKDIR)
2637                         op_data->op_default_stripe_offset =
2638                                    ll_i2info(i1)->lli_def_stripe_offset;
2639         }
2640
2641         if (i2) {
2642                 op_data->op_fid2 = *ll_inode2fid(i2);
2643                 if (S_ISDIR(i2->i_mode)) {
2644                         if (i2 != i1) {
2645                                 down_read(&ll_i2info(i2)->lli_lsm_sem);
2646                                 op_data->op_mea2_sem =
2647                                                 &ll_i2info(i2)->lli_lsm_sem;
2648                         }
2649                         op_data->op_mea2 = ll_i2info(i2)->lli_lsm_md;
2650                 }
2651         } else {
2652                 fid_zero(&op_data->op_fid2);
2653         }
2654
2655         if (ll_i2sbi(i1)->ll_flags & LL_SBI_64BIT_HASH)
2656                 op_data->op_cli_flags |= CLI_HASH64;
2657
2658         if (ll_need_32bit_api(ll_i2sbi(i1)))
2659                 op_data->op_cli_flags |= CLI_API32;
2660
2661         op_data->op_name = name;
2662         op_data->op_namelen = namelen;
2663         op_data->op_mode = mode;
2664         op_data->op_mod_time = ktime_get_real_seconds();
2665         op_data->op_fsuid = from_kuid(&init_user_ns, current_fsuid());
2666         op_data->op_fsgid = from_kgid(&init_user_ns, current_fsgid());
2667         op_data->op_cap = cfs_curproc_cap_pack();
2668         op_data->op_mds = 0;
2669         if ((opc == LUSTRE_OPC_CREATE) && (name != NULL) &&
2670              filename_is_volatile(name, namelen, &op_data->op_mds)) {
2671                 op_data->op_bias |= MDS_CREATE_VOLATILE;
2672         }
2673         op_data->op_data = data;
2674
2675         return op_data;
2676 }
2677
2678 void ll_finish_md_op_data(struct md_op_data *op_data)
2679 {
2680         ll_unlock_md_op_lsm(op_data);
2681         security_release_secctx(op_data->op_file_secctx,
2682                                 op_data->op_file_secctx_size);
2683         OBD_FREE_PTR(op_data);
2684 }
2685
2686 #ifdef HAVE_SUPEROPS_USE_DENTRY
2687 int ll_show_options(struct seq_file *seq, struct dentry *dentry)
2688 #else
2689 int ll_show_options(struct seq_file *seq, struct vfsmount *vfs)
2690 #endif
2691 {
2692         struct ll_sb_info *sbi;
2693
2694 #ifdef HAVE_SUPEROPS_USE_DENTRY
2695         LASSERT((seq != NULL) && (dentry != NULL));
2696         sbi = ll_s2sbi(dentry->d_sb);
2697 #else
2698         LASSERT((seq != NULL) && (vfs != NULL));
2699         sbi = ll_s2sbi(vfs->mnt_sb);
2700 #endif
2701
2702         if (sbi->ll_flags & LL_SBI_NOLCK)
2703                 seq_puts(seq, ",nolock");
2704
2705         if (sbi->ll_flags & LL_SBI_FLOCK)
2706                 seq_puts(seq, ",flock");
2707
2708         if (sbi->ll_flags & LL_SBI_LOCALFLOCK)
2709                 seq_puts(seq, ",localflock");
2710
2711         if (sbi->ll_flags & LL_SBI_USER_XATTR)
2712                 seq_puts(seq, ",user_xattr");
2713
2714         if (sbi->ll_flags & LL_SBI_LAZYSTATFS)
2715                 seq_puts(seq, ",lazystatfs");
2716
2717         if (sbi->ll_flags & LL_SBI_USER_FID2PATH)
2718                 seq_puts(seq, ",user_fid2path");
2719
2720         if (sbi->ll_flags & LL_SBI_ALWAYS_PING)
2721                 seq_puts(seq, ",always_ping");
2722
2723         RETURN(0);
2724 }
2725
2726 /**
2727  * Get obd name by cmd, and copy out to user space
2728  */
2729 int ll_get_obd_name(struct inode *inode, unsigned int cmd, unsigned long arg)
2730 {
2731         struct ll_sb_info *sbi = ll_i2sbi(inode);
2732         struct obd_device *obd;
2733         ENTRY;
2734
2735         if (cmd == OBD_IOC_GETDTNAME)
2736                 obd = class_exp2obd(sbi->ll_dt_exp);
2737         else if (cmd == OBD_IOC_GETMDNAME)
2738                 obd = class_exp2obd(sbi->ll_md_exp);
2739         else
2740                 RETURN(-EINVAL);
2741
2742         if (!obd)
2743                 RETURN(-ENOENT);
2744
2745         if (copy_to_user((void __user *)arg, obd->obd_name,
2746                          strlen(obd->obd_name) + 1))
2747                 RETURN(-EFAULT);
2748
2749         RETURN(0);
2750 }
2751
2752 static char* ll_d_path(struct dentry *dentry, char *buf, int bufsize)
2753 {
2754         char *path = NULL;
2755
2756         struct path p;
2757
2758         p.dentry = dentry;
2759         p.mnt = current->fs->root.mnt;
2760         path_get(&p);
2761         path = d_path(&p, buf, bufsize);
2762         path_put(&p);
2763         return path;
2764 }
2765
2766 void ll_dirty_page_discard_warn(struct page *page, int ioret)
2767 {
2768         char *buf, *path = NULL;
2769         struct dentry *dentry = NULL;
2770         struct inode *inode = page->mapping->host;
2771
2772         /* this can be called inside spin lock so use GFP_ATOMIC. */
2773         buf = (char *)__get_free_page(GFP_ATOMIC);
2774         if (buf != NULL) {
2775                 dentry = d_find_alias(page->mapping->host);
2776                 if (dentry != NULL)
2777                         path = ll_d_path(dentry, buf, PAGE_SIZE);
2778         }
2779
2780         CDEBUG(D_WARNING,
2781                "%s: dirty page discard: %s/fid: "DFID"/%s may get corrupted "
2782                "(rc %d)\n", ll_i2sbi(inode)->ll_fsname,
2783                s2lsi(page->mapping->host->i_sb)->lsi_lmd->lmd_dev,
2784                PFID(ll_inode2fid(inode)),
2785                (path && !IS_ERR(path)) ? path : "", ioret);
2786
2787         if (dentry != NULL)
2788                 dput(dentry);
2789
2790         if (buf != NULL)
2791                 free_page((unsigned long)buf);
2792 }
2793
2794 ssize_t ll_copy_user_md(const struct lov_user_md __user *md,
2795                         struct lov_user_md **kbuf)
2796 {
2797         struct lov_user_md      lum;
2798         ssize_t                 lum_size;
2799         ENTRY;
2800
2801         if (copy_from_user(&lum, md, sizeof(lum)))
2802                 RETURN(-EFAULT);
2803
2804         lum_size = ll_lov_user_md_size(&lum);
2805         if (lum_size < 0)
2806                 RETURN(lum_size);
2807
2808         OBD_ALLOC(*kbuf, lum_size);
2809         if (*kbuf == NULL)
2810                 RETURN(-ENOMEM);
2811
2812         if (copy_from_user(*kbuf, md, lum_size) != 0) {
2813                 OBD_FREE(*kbuf, lum_size);
2814                 RETURN(-EFAULT);
2815         }
2816
2817         RETURN(lum_size);
2818 }
2819
2820 /*
2821  * Compute llite root squash state after a change of root squash
2822  * configuration setting or add/remove of a lnet nid
2823  */
2824 void ll_compute_rootsquash_state(struct ll_sb_info *sbi)
2825 {
2826         struct root_squash_info *squash = &sbi->ll_squash;
2827         int i;
2828         bool matched;
2829         struct lnet_process_id id;
2830
2831         /* Update norootsquash flag */
2832         down_write(&squash->rsi_sem);
2833         if (list_empty(&squash->rsi_nosquash_nids))
2834                 sbi->ll_flags &= ~LL_SBI_NOROOTSQUASH;
2835         else {
2836                 /* Do not apply root squash as soon as one of our NIDs is
2837                  * in the nosquash_nids list */
2838                 matched = false;
2839                 i = 0;
2840                 while (LNetGetId(i++, &id) != -ENOENT) {
2841                         if (LNET_NETTYP(LNET_NIDNET(id.nid)) == LOLND)
2842                                 continue;
2843                         if (cfs_match_nid(id.nid, &squash->rsi_nosquash_nids)) {
2844                                 matched = true;
2845                                 break;
2846                         }
2847                 }
2848                 if (matched)
2849                         sbi->ll_flags |= LL_SBI_NOROOTSQUASH;
2850                 else
2851                         sbi->ll_flags &= ~LL_SBI_NOROOTSQUASH;
2852         }
2853         up_write(&squash->rsi_sem);
2854 }
2855
2856 /**
2857  * Parse linkea content to extract information about a given hardlink
2858  *
2859  * \param[in]   ldata      - Initialized linkea data
2860  * \param[in]   linkno     - Link identifier
2861  * \param[out]  parent_fid - The entry's parent FID
2862  * \param[out]  ln         - Entry name destination buffer
2863  *
2864  * \retval 0 on success
2865  * \retval Appropriate negative error code on failure
2866  */
2867 static int ll_linkea_decode(struct linkea_data *ldata, unsigned int linkno,
2868                             struct lu_fid *parent_fid, struct lu_name *ln)
2869 {
2870         unsigned int    idx;
2871         int             rc;
2872         ENTRY;
2873
2874         rc = linkea_init_with_rec(ldata);
2875         if (rc < 0)
2876                 RETURN(rc);
2877
2878         if (linkno >= ldata->ld_leh->leh_reccount)
2879                 /* beyond last link */
2880                 RETURN(-ENODATA);
2881
2882         linkea_first_entry(ldata);
2883         for (idx = 0; ldata->ld_lee != NULL; idx++) {
2884                 linkea_entry_unpack(ldata->ld_lee, &ldata->ld_reclen, ln,
2885                                     parent_fid);
2886                 if (idx == linkno)
2887                         break;
2888
2889                 linkea_next_entry(ldata);
2890         }
2891
2892         if (idx < linkno)
2893                 RETURN(-ENODATA);
2894
2895         RETURN(0);
2896 }
2897
2898 /**
2899  * Get parent FID and name of an identified link. Operation is performed for
2900  * a given link number, letting the caller iterate over linkno to list one or
2901  * all links of an entry.
2902  *
2903  * \param[in]     file - File descriptor against which to perform the operation
2904  * \param[in,out] arg  - User-filled structure containing the linkno to operate
2905  *                       on and the available size. It is eventually filled with
2906  *                       the requested information or left untouched on error
2907  *
2908  * \retval - 0 on success
2909  * \retval - Appropriate negative error code on failure
2910  */
2911 int ll_getparent(struct file *file, struct getparent __user *arg)
2912 {
2913         struct inode            *inode = file_inode(file);
2914         struct linkea_data      *ldata;
2915         struct lu_buf            buf = LU_BUF_NULL;
2916         struct lu_name           ln;
2917         struct lu_fid            parent_fid;
2918         __u32                    linkno;
2919         __u32                    name_size;
2920         int                      rc;
2921
2922         ENTRY;
2923
2924         if (!cfs_capable(CFS_CAP_DAC_READ_SEARCH) &&
2925             !(ll_i2sbi(inode)->ll_flags & LL_SBI_USER_FID2PATH))
2926                 RETURN(-EPERM);
2927
2928         if (get_user(name_size, &arg->gp_name_size))
2929                 RETURN(-EFAULT);
2930
2931         if (get_user(linkno, &arg->gp_linkno))
2932                 RETURN(-EFAULT);
2933
2934         if (name_size > PATH_MAX)
2935                 RETURN(-EINVAL);
2936
2937         OBD_ALLOC(ldata, sizeof(*ldata));
2938         if (ldata == NULL)
2939                 RETURN(-ENOMEM);
2940
2941         rc = linkea_data_new(ldata, &buf);
2942         if (rc < 0)
2943                 GOTO(ldata_free, rc);
2944
2945 #ifdef HAVE_XATTR_HANDLER_FLAGS
2946         rc = ll_xattr_list(inode, XATTR_NAME_LINK, XATTR_TRUSTED_T, buf.lb_buf,
2947                            buf.lb_len, OBD_MD_FLXATTR);
2948 #else
2949         rc = ll_getxattr(file_dentry(file), XATTR_NAME_LINK, buf.lb_buf,
2950                          buf.lb_len);
2951 #endif /* HAVE_XATTR_HANDLER_FLAGS */
2952         if (rc < 0)
2953                 GOTO(lb_free, rc);
2954
2955         rc = ll_linkea_decode(ldata, linkno, &parent_fid, &ln);
2956         if (rc < 0)
2957                 GOTO(lb_free, rc);
2958
2959         if (ln.ln_namelen >= name_size)
2960                 GOTO(lb_free, rc = -EOVERFLOW);
2961
2962         if (copy_to_user(&arg->gp_fid, &parent_fid, sizeof(arg->gp_fid)))
2963                 GOTO(lb_free, rc = -EFAULT);
2964
2965         if (copy_to_user(&arg->gp_name, ln.ln_name, ln.ln_namelen))
2966                 GOTO(lb_free, rc = -EFAULT);
2967
2968         if (put_user('\0', arg->gp_name + ln.ln_namelen))
2969                 GOTO(lb_free, rc = -EFAULT);
2970
2971 lb_free:
2972         lu_buf_free(&buf);
2973 ldata_free:
2974         OBD_FREE(ldata, sizeof(*ldata));
2975
2976         RETURN(rc);
2977 }