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/statfs.h>
  41 #include <linux/time.h>
  42 #include <linux/types.h>
  43 #include <linux/version.h>
  44 #include <linux/mm.h>
  45 #include <linux/user_namespace.h>
  46 #ifdef HAVE_UIDGID_HEADER
  47 # include <linux/uidgid.h>
  48 #endif
  49 #include <linux/security.h>
  50
  51 #include <uapi/linux/lustre/lustre_ioctl.h>
  52 #ifdef HAVE_UAPI_LINUX_MOUNT_H
  53 #include <uapi/linux/mount.h>
  54 #endif
  55
  56 #include <lustre_ha.h>
  57 #include <lustre_dlm.h>
  58 #include <lprocfs_status.h>
  59 #include <lustre_disk.h>
  60 #include <uapi/linux/lustre/lustre_param.h>
  61 #include <lustre_log.h>
  62 #include <cl_object.h>
  63 #include <obd_cksum.h>
  64 #include "llite_internal.h"
  65
  66 struct kmem_cache *ll_file_data_slab;
  67
  68 #ifndef log2
  69 #define log2(n) ffz(~(n))
  70 #endif
  71
  72 static struct ll_sb_info *ll_init_sbi(void)
  73 {
  74         struct ll_sb_info *sbi = NULL;
  75         unsigned long pages;
  76         unsigned long lru_page_max;
  77         struct sysinfo si;
  78         class_uuid_t uuid;
  79         int i;
  80         ENTRY;
  81
  82         OBD_ALLOC_PTR(sbi);
  83         if (sbi == NULL)
  84                 RETURN(NULL);
  85
  86         spin_lock_init(&sbi->ll_lock);
  87         mutex_init(&sbi->ll_lco.lco_lock);
  88         spin_lock_init(&sbi->ll_pp_extent_lock);
  89         spin_lock_init(&sbi->ll_process_lock);
  90         sbi->ll_rw_stats_on = 0;
  91         sbi->ll_statfs_max_age = OBD_STATFS_CACHE_SECONDS;
  92
  93         si_meminfo(&si);
  94         pages = si.totalram - si.totalhigh;
  95         lru_page_max = pages / 2;
  96
  97         /* initialize ll_cache data */
  98         sbi->ll_cache = cl_cache_init(lru_page_max);
  99         if (sbi->ll_cache == NULL) {
 100                 OBD_FREE(sbi, sizeof(*sbi));
 101                 RETURN(NULL);
 102         }
 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         ll_generate_random_uuid(uuid);
 110         class_uuid_unparse(uuid, &sbi->ll_sb_uuid);
 111         CDEBUG(D_CONFIG, "generated uuid: %s\n", sbi->ll_sb_uuid.uuid);
 112
 113         sbi->ll_flags |= LL_SBI_VERBOSE;
 114 #ifdef ENABLE_CHECKSUM
 115         sbi->ll_flags |= LL_SBI_CHECKSUM;
 116 #endif
 117 #ifdef ENABLE_FLOCK
 118         sbi->ll_flags |= LL_SBI_FLOCK;
 119 #endif
 120
 121 #ifdef HAVE_LRU_RESIZE_SUPPORT
 122         sbi->ll_flags |= LL_SBI_LRU_RESIZE;
 123 #endif
 124         sbi->ll_flags |= LL_SBI_LAZYSTATFS;
 125
 126         for (i = 0; i <= LL_PROCESS_HIST_MAX; i++) {
 127                 spin_lock_init(&sbi->ll_rw_extents_info.pp_extents[i].
 128                                pp_r_hist.oh_lock);
 129                 spin_lock_init(&sbi->ll_rw_extents_info.pp_extents[i].
 130                                pp_w_hist.oh_lock);
 131         }
 132
 133         /* metadata statahead is enabled by default */
 134         sbi->ll_sa_running_max = LL_SA_RUNNING_DEF;
 135         sbi->ll_sa_max = LL_SA_RPC_DEF;
 136         atomic_set(&sbi->ll_sa_total, 0);
 137         atomic_set(&sbi->ll_sa_wrong, 0);
 138         atomic_set(&sbi->ll_sa_running, 0);
 139         atomic_set(&sbi->ll_agl_total, 0);
 140         sbi->ll_flags |= LL_SBI_AGL_ENABLED;
 141         sbi->ll_flags |= LL_SBI_FAST_READ;
 142         sbi->ll_flags |= LL_SBI_TINY_WRITE;
 143
 144         /* root squash */
 145         sbi->ll_squash.rsi_uid = 0;
 146         sbi->ll_squash.rsi_gid = 0;
 147         INIT_LIST_HEAD(&sbi->ll_squash.rsi_nosquash_nids);
 148         init_rwsem(&sbi->ll_squash.rsi_sem);
 149
 150         RETURN(sbi);
 151 }
 152
 153 static void ll_free_sbi(struct super_block *sb)
 154 {
 155         struct ll_sb_info *sbi = ll_s2sbi(sb);
 156         ENTRY;
 157
 158         if (sbi != NULL) {
 159                 if (!list_empty(&sbi->ll_squash.rsi_nosquash_nids))
 160                         cfs_free_nidlist(&sbi->ll_squash.rsi_nosquash_nids);
 161                 if (sbi->ll_cache != NULL) {
 162                         cl_cache_decref(sbi->ll_cache);
 163                         sbi->ll_cache = NULL;
 164                 }
 165                 OBD_FREE(sbi, sizeof(*sbi));
 166         }
 167         EXIT;
 168 }
 169
 170 static int client_common_fill_super(struct super_block *sb, char *md, char *dt,
 171                                     struct vfsmount *mnt)
 172 {
 173         struct inode *root = NULL;
 174         struct ll_sb_info *sbi = ll_s2sbi(sb);
 175         struct obd_statfs *osfs = NULL;
 176         struct ptlrpc_request *request = NULL;
 177         struct obd_connect_data *data = NULL;
 178         struct obd_uuid *uuid;
 179         struct md_op_data *op_data;
 180         struct lustre_md lmd;
 181         u64 valid;
 182         int size, err, checksum;
 183
 184         ENTRY;
 185         sbi->ll_md_obd = class_name2obd(md);
 186         if (!sbi->ll_md_obd) {
 187                 CERROR("MD %s: not setup or attached\n", md);
 188                 RETURN(-EINVAL);
 189         }
 190
 191         OBD_ALLOC_PTR(data);
 192         if (data == NULL)
 193                 RETURN(-ENOMEM);
 194
 195         OBD_ALLOC_PTR(osfs);
 196         if (osfs == NULL) {
 197                 OBD_FREE_PTR(data);
 198                 RETURN(-ENOMEM);
 199         }
 200
 201         /* pass client page size via ocd_grant_blkbits, the server should report
 202          * back its backend blocksize for grant calculation purpose */
 203         data->ocd_grant_blkbits = PAGE_SHIFT;
 204
 205         /* indicate MDT features supported by this client */
 206         data->ocd_connect_flags = OBD_CONNECT_IBITS    | OBD_CONNECT_NODEVOH  |
 207                                   OBD_CONNECT_ATTRFID  | OBD_CONNECT_GRANT |
 208                                   OBD_CONNECT_VERSION  | OBD_CONNECT_BRW_SIZE |
 209                                   OBD_CONNECT_SRVLOCK  | OBD_CONNECT_TRUNCLOCK|
 210                                   OBD_CONNECT_MDS_CAPA | OBD_CONNECT_OSS_CAPA |
 211                                   OBD_CONNECT_CANCELSET | OBD_CONNECT_FID     |
 212                                   OBD_CONNECT_AT       | OBD_CONNECT_LOV_V3   |
 213                                   OBD_CONNECT_VBR | OBD_CONNECT_FULL20 |
 214                                   OBD_CONNECT_64BITHASH |
 215                                   OBD_CONNECT_EINPROGRESS |
 216                                   OBD_CONNECT_JOBSTATS | OBD_CONNECT_LVB_TYPE |
 217                                   OBD_CONNECT_LAYOUTLOCK | OBD_CONNECT_PINGLESS|
 218                                   OBD_CONNECT_MAX_EASIZE |
 219                                   OBD_CONNECT_FLOCK_DEAD |
 220                                   OBD_CONNECT_DISP_STRIPE | OBD_CONNECT_LFSCK |
 221                                   OBD_CONNECT_OPEN_BY_FID |
 222                                   OBD_CONNECT_DIR_STRIPE |
 223                                   OBD_CONNECT_BULK_MBITS | OBD_CONNECT_CKSUM |
 224                                   OBD_CONNECT_SUBTREE |
 225                                   OBD_CONNECT_MULTIMODRPCS |
 226                                   OBD_CONNECT_GRANT_PARAM |
 227                                   OBD_CONNECT_SHORTIO | OBD_CONNECT_FLAGS2;
 228
 229         data->ocd_connect_flags2 = OBD_CONNECT2_FLR |
 230                                    OBD_CONNECT2_LOCK_CONVERT |
 231                                    OBD_CONNECT2_DIR_MIGRATE |
 232                                    OBD_CONNECT2_SUM_STATFS |
 233                                    OBD_CONNECT2_ARCHIVE_ID_ARRAY |
 234                                    OBD_CONNECT2_LSOM |
 235                                    OBD_CONNECT2_ASYNC_DISCARD |
 236                                    OBD_CONNECT2_GETATTR_PFID;
 237
 238 #ifdef HAVE_LRU_RESIZE_SUPPORT
 239         if (sbi->ll_flags & LL_SBI_LRU_RESIZE)
 240                 data->ocd_connect_flags |= OBD_CONNECT_LRU_RESIZE;
 241 #endif
 242 #ifdef CONFIG_FS_POSIX_ACL
 243         data->ocd_connect_flags |= OBD_CONNECT_ACL | OBD_CONNECT_UMASK |
 244                                    OBD_CONNECT_LARGE_ACL;
 245 #endif
 246
 247         data->ocd_cksum_types = obd_cksum_types_supported_client();
 248
 249         if (OBD_FAIL_CHECK(OBD_FAIL_MDC_LIGHTWEIGHT))
 250                 /* flag mdc connection as lightweight, only used for test
 251                  * purpose, use with care */
 252                 data->ocd_connect_flags |= OBD_CONNECT_LIGHTWEIGHT;
 253
 254         data->ocd_ibits_known = MDS_INODELOCK_FULL;
 255         data->ocd_version = LUSTRE_VERSION_CODE;
 256
 257         if (sb->s_flags & SB_RDONLY)
 258                 data->ocd_connect_flags |= OBD_CONNECT_RDONLY;
 259         if (sbi->ll_flags & LL_SBI_USER_XATTR)
 260                 data->ocd_connect_flags |= OBD_CONNECT_XATTR;
 261
 262 #ifdef SB_NOSEC
 263         /* Setting this indicates we correctly support S_NOSEC (See kernel
 264          * commit 9e1f1de02c2275d7172e18dc4e7c2065777611bf)
 265          */
 266         sb->s_flags |= SB_NOSEC;
 267 #endif
 268
 269         if (sbi->ll_flags & LL_SBI_FLOCK)
 270                 sbi->ll_fop = &ll_file_operations_flock;
 271         else if (sbi->ll_flags & LL_SBI_LOCALFLOCK)
 272                 sbi->ll_fop = &ll_file_operations;
 273         else
 274                 sbi->ll_fop = &ll_file_operations_noflock;
 275
 276         /* always ping even if server suppress_pings */
 277         if (sbi->ll_flags & LL_SBI_ALWAYS_PING)
 278                 data->ocd_connect_flags &= ~OBD_CONNECT_PINGLESS;
 279
 280         obd_connect_set_secctx(data);
 281
 282 #if defined(CONFIG_SECURITY)
 283         data->ocd_connect_flags2 |= OBD_CONNECT2_SELINUX_POLICY;
 284 #endif
 285
 286         data->ocd_brw_size = MD_MAX_BRW_SIZE;
 287
 288         err = obd_connect(NULL, &sbi->ll_md_exp, sbi->ll_md_obd,
 289                           &sbi->ll_sb_uuid, data, NULL);
 290         if (err == -EBUSY) {
 291                 LCONSOLE_ERROR_MSG(0x14f, "An MDT (md %s) is performing "
 292                                    "recovery, of which this client is not a "
 293                                    "part. Please wait for recovery to complete,"
 294                                    " abort, or time out.\n", md);
 295                 GOTO(out, err);
 296         } else if (err) {
 297                 CERROR("cannot connect to %s: rc = %d\n", md, err);
 298                 GOTO(out, err);
 299         }
 300
 301         sbi->ll_md_exp->exp_connect_data = *data;
 302
 303         err = obd_fid_init(sbi->ll_md_exp->exp_obd, sbi->ll_md_exp,
 304                            LUSTRE_SEQ_METADATA);
 305         if (err) {
 306                 CERROR("%s: Can't init metadata layer FID infrastructure, "
 307                        "rc = %d\n", sbi->ll_md_exp->exp_obd->obd_name, err);
 308                 GOTO(out_md, err);
 309         }
 310
 311         /* For mount, we only need fs info from MDT0, and also in DNE, it
 312          * can make sure the client can be mounted as long as MDT0 is
 313          * avaible */
 314         err = obd_statfs(NULL, sbi->ll_md_exp, osfs,
 315                         ktime_get_seconds() - sbi->ll_statfs_max_age,
 316                         OBD_STATFS_FOR_MDT0);
 317         if (err)
 318                 GOTO(out_md_fid, err);
 319
 320         /* This needs to be after statfs to ensure connect has finished.
 321          * Note that "data" does NOT contain the valid connect reply.
 322          * If connecting to a 1.8 server there will be no LMV device, so
 323          * we can access the MDC export directly and exp_connect_flags will
 324          * be non-zero, but if accessing an upgraded 2.1 server it will
 325          * have the correct flags filled in.
 326          * XXX: fill in the LMV exp_connect_flags from MDC(s). */
 327         valid = exp_connect_flags(sbi->ll_md_exp) & CLIENT_CONNECT_MDT_REQD;
 328         if (exp_connect_flags(sbi->ll_md_exp) != 0 &&
 329             valid != CLIENT_CONNECT_MDT_REQD) {
 330                 char *buf;
 331
 332                 OBD_ALLOC_WAIT(buf, PAGE_SIZE);
 333                 obd_connect_flags2str(buf, PAGE_SIZE,
 334                                       valid ^ CLIENT_CONNECT_MDT_REQD, 0, ",");
 335                 LCONSOLE_ERROR_MSG(0x170, "Server %s does not support "
 336                                    "feature(s) needed for correct operation "
 337                                    "of this client (%s). Please upgrade "
 338                                    "server or downgrade client.\n",
 339                                    sbi->ll_md_exp->exp_obd->obd_name, buf);
 340                 OBD_FREE(buf, PAGE_SIZE);
 341                 GOTO(out_md_fid, err = -EPROTO);
 342         }
 343
 344         size = sizeof(*data);
 345         err = obd_get_info(NULL, sbi->ll_md_exp, sizeof(KEY_CONN_DATA),
 346                            KEY_CONN_DATA,  &size, data);
 347         if (err) {
 348                 CERROR("%s: Get connect data failed: rc = %d\n",
 349                        sbi->ll_md_exp->exp_obd->obd_name, err);
 350                 GOTO(out_md_fid, err);
 351         }
 352
 353         LASSERT(osfs->os_bsize);
 354         sb->s_blocksize = osfs->os_bsize;
 355         sb->s_blocksize_bits = log2(osfs->os_bsize);
 356         sb->s_magic = LL_SUPER_MAGIC;
 357         sb->s_maxbytes = MAX_LFS_FILESIZE;
 358         sbi->ll_namelen = osfs->os_namelen;
 359         sbi->ll_mnt.mnt = current->fs->root.mnt;
 360
 361         if ((sbi->ll_flags & LL_SBI_USER_XATTR) &&
 362             !(data->ocd_connect_flags & OBD_CONNECT_XATTR)) {
 363                 LCONSOLE_INFO("Disabling user_xattr feature because "
 364                               "it is not supported on the server\n");
 365                 sbi->ll_flags &= ~LL_SBI_USER_XATTR;
 366         }
 367
 368         if (data->ocd_connect_flags & OBD_CONNECT_ACL) {
 369 #ifdef SB_POSIXACL
 370                 sb->s_flags |= SB_POSIXACL;
 371 #endif
 372                 sbi->ll_flags |= LL_SBI_ACL;
 373         } else {
 374                 LCONSOLE_INFO("client wants to enable acl, but mdt not!\n");
 375 #ifdef SB_POSIXACL
 376                 sb->s_flags &= ~SB_POSIXACL;
 377 #endif
 378                 sbi->ll_flags &= ~LL_SBI_ACL;
 379         }
 380
 381         if (data->ocd_connect_flags & OBD_CONNECT_64BITHASH)
 382                 sbi->ll_flags |= LL_SBI_64BIT_HASH;
 383
 384         if (data->ocd_connect_flags & OBD_CONNECT_LAYOUTLOCK)
 385                 sbi->ll_flags |= LL_SBI_LAYOUT_LOCK;
 386
 387         if (obd_connect_has_secctx(data))
 388                 sbi->ll_flags |= LL_SBI_FILE_SECCTX;
 389
 390         if (data->ocd_ibits_known & MDS_INODELOCK_XATTR) {
 391                 if (!(data->ocd_connect_flags & OBD_CONNECT_MAX_EASIZE)) {
 392                         LCONSOLE_INFO("%s: disabling xattr cache due to "
 393                                       "unknown maximum xattr size.\n", dt);
 394                 } else if (!sbi->ll_xattr_cache_set) {
 395                         /* If xattr_cache is already set (no matter 0 or 1)
 396                          * during processing llog, it won't be enabled here. */
 397                         sbi->ll_flags |= LL_SBI_XATTR_CACHE;
 398                         sbi->ll_xattr_cache_enabled = 1;
 399                 }
 400         }
 401
 402         sbi->ll_dt_obd = class_name2obd(dt);
 403         if (!sbi->ll_dt_obd) {
 404                 CERROR("DT %s: not setup or attached\n", dt);
 405                 GOTO(out_md_fid, err = -ENODEV);
 406         }
 407
 408         /* pass client page size via ocd_grant_blkbits, the server should report
 409          * back its backend blocksize for grant calculation purpose */
 410         data->ocd_grant_blkbits = PAGE_SHIFT;
 411
 412         /* indicate OST features supported by this client */
 413         data->ocd_connect_flags = OBD_CONNECT_GRANT | OBD_CONNECT_VERSION |
 414                                   OBD_CONNECT_REQPORTAL | OBD_CONNECT_BRW_SIZE |
 415                                   OBD_CONNECT_CANCELSET | OBD_CONNECT_FID |
 416                                   OBD_CONNECT_SRVLOCK | OBD_CONNECT_TRUNCLOCK|
 417                                   OBD_CONNECT_AT | OBD_CONNECT_OSS_CAPA |
 418                                   OBD_CONNECT_VBR | OBD_CONNECT_FULL20 |
 419                                   OBD_CONNECT_64BITHASH | OBD_CONNECT_MAXBYTES |
 420                                   OBD_CONNECT_EINPROGRESS |
 421                                   OBD_CONNECT_JOBSTATS | OBD_CONNECT_LVB_TYPE |
 422                                   OBD_CONNECT_LAYOUTLOCK |
 423                                   OBD_CONNECT_PINGLESS | OBD_CONNECT_LFSCK |
 424                                   OBD_CONNECT_BULK_MBITS | OBD_CONNECT_SHORTIO |
 425                                   OBD_CONNECT_FLAGS2 | OBD_CONNECT_GRANT_SHRINK;
 426
 427 /* The client currently advertises support for OBD_CONNECT_LOCKAHEAD_OLD so it
 428  * can interoperate with an older version of lockahead which was released prior
 429  * to landing in master. This support will be dropped when 2.13 development
 430  * starts.  At the point, we should not just drop the connect flag (below), we
 431  * should also remove the support in the code.
 432  *
 433  * Removing it means a few things:
 434  * 1. Remove this section here
 435  * 2. Remove CEF_NONBLOCK in ll_file_lockahead()
 436  * 3. Remove function exp_connect_lockahead_old
 437  * 4. Remove LDLM_FL_LOCKAHEAD_OLD_RESERVED in lustre_dlm_flags.h
 438  * */
 439 #if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(2, 12, 50, 0)
 440         data->ocd_connect_flags |= OBD_CONNECT_LOCKAHEAD_OLD;
 441 #endif
 442
 443         data->ocd_connect_flags2 = OBD_CONNECT2_LOCKAHEAD;
 444
 445         if (!OBD_FAIL_CHECK(OBD_FAIL_OSC_CONNECT_GRANT_PARAM))
 446                 data->ocd_connect_flags |= OBD_CONNECT_GRANT_PARAM;
 447
 448         /* OBD_CONNECT_CKSUM should always be set, even if checksums are
 449          * disabled by default, because it can still be enabled on the
 450          * fly via /sys. As a consequence, we still need to come to an
 451          * agreement on the supported algorithms at connect time
 452          */
 453         data->ocd_connect_flags |= OBD_CONNECT_CKSUM;
 454
 455         if (OBD_FAIL_CHECK(OBD_FAIL_OSC_CKSUM_ADLER_ONLY))
 456                 data->ocd_cksum_types = OBD_CKSUM_ADLER;
 457         else
 458                 data->ocd_cksum_types = obd_cksum_types_supported_client();
 459
 460 #ifdef HAVE_LRU_RESIZE_SUPPORT
 461         data->ocd_connect_flags |= OBD_CONNECT_LRU_RESIZE;
 462 #endif
 463         /* always ping even if server suppress_pings */
 464         if (sbi->ll_flags & LL_SBI_ALWAYS_PING)
 465                 data->ocd_connect_flags &= ~OBD_CONNECT_PINGLESS;
 466
 467         CDEBUG(D_RPCTRACE, "ocd_connect_flags: %#llx ocd_version: %d "
 468                "ocd_grant: %d\n", data->ocd_connect_flags,
 469                data->ocd_version, data->ocd_grant);
 470
 471         sbi->ll_dt_obd->obd_upcall.onu_owner = &sbi->ll_lco;
 472         sbi->ll_dt_obd->obd_upcall.onu_upcall = cl_ocd_update;
 473
 474         data->ocd_brw_size = DT_MAX_BRW_SIZE;
 475
 476         err = obd_connect(NULL, &sbi->ll_dt_exp, sbi->ll_dt_obd,
 477                           &sbi->ll_sb_uuid, data, NULL);
 478         if (err == -EBUSY) {
 479                 LCONSOLE_ERROR_MSG(0x150, "An OST (dt %s) is performing "
 480                                    "recovery, of which this client is not a "
 481                                    "part.  Please wait for recovery to "
 482                                    "complete, abort, or time out.\n", dt);
 483                 GOTO(out_md, err);
 484         } else if (err) {
 485                 CERROR("%s: Cannot connect to %s: rc = %d\n",
 486                        sbi->ll_dt_exp->exp_obd->obd_name, dt, err);
 487                 GOTO(out_md, err);
 488         }
 489
 490         sbi->ll_dt_exp->exp_connect_data = *data;
 491
 492         /* Don't change value if it was specified in the config log */
 493         if (sbi->ll_ra_info.ra_max_read_ahead_whole_pages == -1) {
 494                 sbi->ll_ra_info.ra_max_read_ahead_whole_pages =
 495                         max_t(unsigned long, SBI_DEFAULT_READAHEAD_WHOLE_MAX,
 496                               (data->ocd_brw_size >> PAGE_SHIFT));
 497                 if (sbi->ll_ra_info.ra_max_read_ahead_whole_pages >
 498                     sbi->ll_ra_info.ra_max_pages_per_file)
 499                         sbi->ll_ra_info.ra_max_read_ahead_whole_pages =
 500                                 sbi->ll_ra_info.ra_max_pages_per_file;
 501         }
 502
 503         err = obd_fid_init(sbi->ll_dt_exp->exp_obd, sbi->ll_dt_exp,
 504                            LUSTRE_SEQ_METADATA);
 505         if (err) {
 506                 CERROR("%s: Can't init data layer FID infrastructure, "
 507                        "rc = %d\n", sbi->ll_dt_exp->exp_obd->obd_name, err);
 508                 GOTO(out_dt, err);
 509         }
 510
 511         mutex_lock(&sbi->ll_lco.lco_lock);
 512         sbi->ll_lco.lco_flags = data->ocd_connect_flags;
 513         sbi->ll_lco.lco_md_exp = sbi->ll_md_exp;
 514         sbi->ll_lco.lco_dt_exp = sbi->ll_dt_exp;
 515         mutex_unlock(&sbi->ll_lco.lco_lock);
 516
 517         fid_zero(&sbi->ll_root_fid);
 518         err = md_get_root(sbi->ll_md_exp, get_mount_fileset(sb),
 519                            &sbi->ll_root_fid);
 520         if (err) {
 521                 CERROR("cannot mds_connect: rc = %d\n", err);
 522                 GOTO(out_lock_cn_cb, err);
 523         }
 524         if (!fid_is_sane(&sbi->ll_root_fid)) {
 525                 CERROR("%s: Invalid root fid "DFID" during mount\n",
 526                        sbi->ll_md_exp->exp_obd->obd_name,
 527                        PFID(&sbi->ll_root_fid));
 528                 GOTO(out_lock_cn_cb, err = -EINVAL);
 529         }
 530         CDEBUG(D_SUPER, "rootfid "DFID"\n", PFID(&sbi->ll_root_fid));
 531
 532         sb->s_op = &lustre_super_operations;
 533 #ifdef HAVE_XATTR_HANDLER_FLAGS
 534         sb->s_xattr = ll_xattr_handlers;
 535 #endif
 536 #if THREAD_SIZE >= 8192 /*b=17630*/
 537         sb->s_export_op = &lustre_export_operations;
 538 #endif
 539
 540         /* make root inode
 541          * XXX: move this to after cbd setup? */
 542         valid = OBD_MD_FLGETATTR | OBD_MD_FLBLOCKS | OBD_MD_FLMODEASIZE;
 543         if (sbi->ll_flags & LL_SBI_ACL)
 544                 valid |= OBD_MD_FLACL;
 545
 546         OBD_ALLOC_PTR(op_data);
 547         if (op_data == NULL)
 548                 GOTO(out_lock_cn_cb, err = -ENOMEM);
 549
 550         op_data->op_fid1 = sbi->ll_root_fid;
 551         op_data->op_mode = 0;
 552         op_data->op_valid = valid;
 553
 554         err = md_getattr(sbi->ll_md_exp, op_data, &request);
 555
 556         OBD_FREE_PTR(op_data);
 557         if (err) {
 558                 CERROR("%s: md_getattr failed for root: rc = %d\n",
 559                        sbi->ll_md_exp->exp_obd->obd_name, err);
 560                 GOTO(out_lock_cn_cb, err);
 561         }
 562
 563         err = md_get_lustre_md(sbi->ll_md_exp, request, sbi->ll_dt_exp,
 564                                sbi->ll_md_exp, &lmd);
 565         if (err) {
 566                 CERROR("failed to understand root inode md: rc = %d\n", err);
 567                 ptlrpc_req_finished(request);
 568                 GOTO(out_lock_cn_cb, err);
 569         }
 570
 571         LASSERT(fid_is_sane(&sbi->ll_root_fid));
 572         root = ll_iget(sb, cl_fid_build_ino(&sbi->ll_root_fid,
 573                                             sbi->ll_flags & LL_SBI_32BIT_API),
 574                        &lmd);
 575         md_free_lustre_md(sbi->ll_md_exp, &lmd);
 576         ptlrpc_req_finished(request);
 577
 578         if (IS_ERR(root)) {
 579 #ifdef CONFIG_FS_POSIX_ACL
 580                 if (lmd.posix_acl) {
 581                         posix_acl_release(lmd.posix_acl);
 582                         lmd.posix_acl = NULL;
 583                 }
 584 #endif
 585                 err = IS_ERR(root) ? PTR_ERR(root) : -EBADF;
 586                 root = NULL;
 587                 CERROR("lustre_lite: bad iget4 for root\n");
 588                 GOTO(out_root, err);
 589         }
 590
 591         checksum = sbi->ll_flags & LL_SBI_CHECKSUM;
 592         if (sbi->ll_checksum_set) {
 593                 err = obd_set_info_async(NULL, sbi->ll_dt_exp,
 594                                          sizeof(KEY_CHECKSUM), KEY_CHECKSUM,
 595                                          sizeof(checksum), &checksum, NULL);
 596                 if (err) {
 597                         CERROR("%s: Set checksum failed: rc = %d\n",
 598                                sbi->ll_dt_exp->exp_obd->obd_name, err);
 599                         GOTO(out_root, err);
 600                 }
 601         }
 602         cl_sb_init(sb);
 603
 604         err = obd_set_info_async(NULL, sbi->ll_dt_exp, sizeof(KEY_CACHE_SET),
 605                                  KEY_CACHE_SET, sizeof(*sbi->ll_cache),
 606                                  sbi->ll_cache, NULL);
 607         if (err) {
 608                 CERROR("%s: Set cache_set failed: rc = %d\n",
 609                        sbi->ll_dt_exp->exp_obd->obd_name, err);
 610                 GOTO(out_root, err);
 611         }
 612
 613         sb->s_root = d_make_root(root);
 614         if (sb->s_root == NULL) {
 615                 CERROR("%s: can't make root dentry\n",
 616                         ll_get_fsname(sb, NULL, 0));
 617                 GOTO(out_root, err = -ENOMEM);
 618         }
 619 #ifdef HAVE_DCACHE_LOCK
 620         sb->s_root->d_op = &ll_d_ops;
 621 #endif
 622
 623         sbi->ll_sdev_orig = sb->s_dev;
 624
 625         /* We set sb->s_dev equal on all lustre clients in order to support
 626          * NFS export clustering.  NFSD requires that the FSID be the same
 627          * on all clients. */
 628         /* s_dev is also used in lt_compare() to compare two fs, but that is
 629          * only a node-local comparison. */
 630         uuid = obd_get_uuid(sbi->ll_md_exp);
 631         if (uuid != NULL)
 632                 sb->s_dev = get_uuid2int(uuid->uuid, strlen(uuid->uuid));
 633
 634         if (data != NULL)
 635                 OBD_FREE_PTR(data);
 636         if (osfs != NULL)
 637                 OBD_FREE_PTR(osfs);
 638
 639         if (sbi->ll_dt_obd) {
 640                 err = sysfs_create_link(&sbi->ll_kset.kobj,
 641                                         &sbi->ll_dt_obd->obd_kset.kobj,
 642                                         sbi->ll_dt_obd->obd_type->typ_name);
 643                 if (err < 0) {
 644                         CERROR("%s: could not register %s in llite: rc = %d\n",
 645                                dt, ll_get_fsname(sb, NULL, 0), err);
 646                         err = 0;
 647                 }
 648         }
 649
 650         if (sbi->ll_md_obd) {
 651                 err = sysfs_create_link(&sbi->ll_kset.kobj,
 652                                         &sbi->ll_md_obd->obd_kset.kobj,
 653                                         sbi->ll_md_obd->obd_type->typ_name);
 654                 if (err < 0) {
 655                         CERROR("%s: could not register %s in llite: rc = %d\n",
 656                                md, ll_get_fsname(sb, NULL, 0), err);
 657                         err = 0;
 658                 }
 659         }
 660
 661         RETURN(err);
 662 out_root:
 663         if (root)
 664                 iput(root);
 665 out_lock_cn_cb:
 666         obd_fid_fini(sbi->ll_dt_exp->exp_obd);
 667 out_dt:
 668         obd_disconnect(sbi->ll_dt_exp);
 669         sbi->ll_dt_exp = NULL;
 670         sbi->ll_dt_obd = NULL;
 671 out_md_fid:
 672         obd_fid_fini(sbi->ll_md_exp->exp_obd);
 673 out_md:
 674         obd_disconnect(sbi->ll_md_exp);
 675         sbi->ll_md_exp = NULL;
 676         sbi->ll_md_obd = NULL;
 677 out:
 678         if (data != NULL)
 679                 OBD_FREE_PTR(data);
 680         if (osfs != NULL)
 681                 OBD_FREE_PTR(osfs);
 682         return err;
 683 }
 684
 685 int ll_get_max_mdsize(struct ll_sb_info *sbi, int *lmmsize)
 686 {
 687         int size, rc;
 688
 689         size = sizeof(*lmmsize);
 690         rc = obd_get_info(NULL, sbi->ll_dt_exp, sizeof(KEY_MAX_EASIZE),
 691                           KEY_MAX_EASIZE, &size, lmmsize);
 692         if (rc != 0) {
 693                 CERROR("%s: cannot get max LOV EA size: rc = %d\n",
 694                        sbi->ll_dt_exp->exp_obd->obd_name, rc);
 695                 RETURN(rc);
 696         }
 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         RETURN(rc);
 705 }
 706
 707 /**
 708  * Get the value of the default_easize parameter.
 709  *
 710  * \see client_obd::cl_default_mds_easize
 711  *
 712  * \param[in] sbi       superblock info for this filesystem
 713  * \param[out] lmmsize  pointer to storage location for value
 714  *
 715  * \retval 0            on success
 716  * \retval negative     negated errno on failure
 717  */
 718 int ll_get_default_mdsize(struct ll_sb_info *sbi, int *lmmsize)
 719 {
 720         int size, rc;
 721
 722         size = sizeof(int);
 723         rc = obd_get_info(NULL, sbi->ll_md_exp, sizeof(KEY_DEFAULT_EASIZE),
 724                          KEY_DEFAULT_EASIZE, &size, lmmsize);
 725         if (rc)
 726                 CERROR("Get default mdsize error rc %d\n", rc);
 727
 728         RETURN(rc);
 729 }
 730
 731 /**
 732  * Set the default_easize parameter to the given value.
 733  *
 734  * \see client_obd::cl_default_mds_easize
 735  *
 736  * \param[in] sbi       superblock info for this filesystem
 737  * \param[in] lmmsize   the size to set
 738  *
 739  * \retval 0            on success
 740  * \retval negative     negated errno on failure
 741  */
 742 int ll_set_default_mdsize(struct ll_sb_info *sbi, int lmmsize)
 743 {
 744         int rc;
 745
 746         if (lmmsize < sizeof(struct lov_mds_md) ||
 747             lmmsize > OBD_MAX_DEFAULT_EA_SIZE)
 748                 return -EINVAL;
 749
 750         rc = obd_set_info_async(NULL, sbi->ll_md_exp,
 751                                 sizeof(KEY_DEFAULT_EASIZE), KEY_DEFAULT_EASIZE,
 752                                 sizeof(int), &lmmsize, NULL);
 753
 754         RETURN(rc);
 755 }
 756
 757 static void client_common_put_super(struct super_block *sb)
 758 {
 759         struct ll_sb_info *sbi = ll_s2sbi(sb);
 760         ENTRY;
 761
 762         cl_sb_fini(sb);
 763
 764         obd_fid_fini(sbi->ll_dt_exp->exp_obd);
 765         obd_disconnect(sbi->ll_dt_exp);
 766         sbi->ll_dt_exp = NULL;
 767
 768         ll_debugfs_unregister_super(sb);
 769
 770         obd_fid_fini(sbi->ll_md_exp->exp_obd);
 771         obd_disconnect(sbi->ll_md_exp);
 772         sbi->ll_md_exp = NULL;
 773
 774         EXIT;
 775 }
 776
 777 void ll_kill_super(struct super_block *sb)
 778 {
 779         struct ll_sb_info *sbi;
 780         ENTRY;
 781
 782         /* not init sb ?*/
 783         if (!(sb->s_flags & SB_ACTIVE))
 784                 return;
 785
 786         sbi = ll_s2sbi(sb);
 787         /* we need restore s_dev from changed for clustred NFS before put_super
 788          * because new kernels have cached s_dev and change sb->s_dev in
 789          * put_super not affected real removing devices */
 790         if (sbi) {
 791                 sb->s_dev = sbi->ll_sdev_orig;
 792                 sbi->ll_umounting = 1;
 793
 794                 /* wait running statahead threads to quit */
 795                 while (atomic_read(&sbi->ll_sa_running) > 0) {
 796                         set_current_state(TASK_UNINTERRUPTIBLE);
 797                         schedule_timeout(msecs_to_jiffies(MSEC_PER_SEC >> 3));
 798                 }
 799         }
 800
 801         EXIT;
 802 }
 803
 804 static inline int ll_set_opt(const char *opt, char *data, int fl)
 805 {
 806         if (strncmp(opt, data, strlen(opt)) != 0)
 807                 return 0;
 808         else
 809                 return fl;
 810 }
 811
 812 /* non-client-specific mount options are parsed in lmd_parse */
 813 static int ll_options(char *options, struct ll_sb_info *sbi)
 814 {
 815         int tmp;
 816         char *s1 = options, *s2;
 817         int *flags = &sbi->ll_flags;
 818         ENTRY;
 819
 820         if (!options)
 821                 RETURN(0);
 822
 823         CDEBUG(D_CONFIG, "Parsing opts %s\n", options);
 824
 825         while (*s1) {
 826                 CDEBUG(D_SUPER, "next opt=%s\n", s1);
 827                 tmp = ll_set_opt("nolock", s1, LL_SBI_NOLCK);
 828                 if (tmp) {
 829                         *flags |= tmp;
 830                         goto next;
 831                 }
 832                 tmp = ll_set_opt("flock", s1, LL_SBI_FLOCK);
 833                 if (tmp) {
 834                         *flags = (*flags & ~LL_SBI_LOCALFLOCK) | tmp;
 835                         goto next;
 836                 }
 837                 tmp = ll_set_opt("localflock", s1, LL_SBI_LOCALFLOCK);
 838                 if (tmp) {
 839                         *flags = (*flags & ~LL_SBI_FLOCK) | tmp;
 840                         goto next;
 841                 }
 842                 tmp = ll_set_opt("noflock", s1, LL_SBI_FLOCK|LL_SBI_LOCALFLOCK);
 843                 if (tmp) {
 844                         *flags &= ~tmp;
 845                         goto next;
 846                 }
 847                 tmp = ll_set_opt("user_xattr", s1, LL_SBI_USER_XATTR);
 848                 if (tmp) {
 849                         *flags |= tmp;
 850                         goto next;
 851                 }
 852                 tmp = ll_set_opt("nouser_xattr", s1, LL_SBI_USER_XATTR);
 853                 if (tmp) {
 854                         *flags &= ~tmp;
 855                         goto next;
 856                 }
 857                 tmp = ll_set_opt("context", s1, 1);
 858                 if (tmp)
 859                         goto next;
 860                 tmp = ll_set_opt("fscontext", s1, 1);
 861                 if (tmp)
 862                         goto next;
 863                 tmp = ll_set_opt("defcontext", s1, 1);
 864                 if (tmp)
 865                         goto next;
 866                 tmp = ll_set_opt("rootcontext", s1, 1);
 867                 if (tmp)
 868                         goto next;
 869                 tmp = ll_set_opt("user_fid2path", s1, LL_SBI_USER_FID2PATH);
 870                 if (tmp) {
 871                         *flags |= tmp;
 872                         goto next;
 873                 }
 874                 tmp = ll_set_opt("nouser_fid2path", s1, LL_SBI_USER_FID2PATH);
 875                 if (tmp) {
 876                         *flags &= ~tmp;
 877                         goto next;
 878                 }
 879
 880                 tmp = ll_set_opt("checksum", s1, LL_SBI_CHECKSUM);
 881                 if (tmp) {
 882                         *flags |= tmp;
 883                         sbi->ll_checksum_set = 1;
 884                         goto next;
 885                 }
 886                 tmp = ll_set_opt("nochecksum", s1, LL_SBI_CHECKSUM);
 887                 if (tmp) {
 888                         *flags &= ~tmp;
 889                         sbi->ll_checksum_set = 1;
 890                         goto next;
 891                 }
 892                 tmp = ll_set_opt("lruresize", s1, LL_SBI_LRU_RESIZE);
 893                 if (tmp) {
 894                         *flags |= tmp;
 895                         goto next;
 896                 }
 897                 tmp = ll_set_opt("nolruresize", s1, LL_SBI_LRU_RESIZE);
 898                 if (tmp) {
 899                         *flags &= ~tmp;
 900                         goto next;
 901                 }
 902                 tmp = ll_set_opt("lazystatfs", s1, LL_SBI_LAZYSTATFS);
 903                 if (tmp) {
 904                         *flags |= tmp;
 905                         goto next;
 906                 }
 907                 tmp = ll_set_opt("nolazystatfs", s1, LL_SBI_LAZYSTATFS);
 908                 if (tmp) {
 909                         *flags &= ~tmp;
 910                         goto next;
 911                 }
 912                 tmp = ll_set_opt("32bitapi", s1, LL_SBI_32BIT_API);
 913                 if (tmp) {
 914                         *flags |= tmp;
 915                         goto next;
 916                 }
 917                 tmp = ll_set_opt("verbose", s1, LL_SBI_VERBOSE);
 918                 if (tmp) {
 919                         *flags |= tmp;
 920                         goto next;
 921                 }
 922                 tmp = ll_set_opt("noverbose", s1, LL_SBI_VERBOSE);
 923                 if (tmp) {
 924                         *flags &= ~tmp;
 925                         goto next;
 926                 }
 927                 tmp = ll_set_opt("always_ping", s1, LL_SBI_ALWAYS_PING);
 928                 if (tmp) {
 929                         *flags |= tmp;
 930                         goto next;
 931                 }
 932                 LCONSOLE_ERROR_MSG(0x152, "Unknown option '%s', won't mount.\n",
 933                                    s1);
 934                 RETURN(-EINVAL);
 935
 936 next:
 937                 /* Find next opt */
 938                 s2 = strchr(s1, ',');
 939                 if (s2 == NULL)
 940                         break;
 941                 s1 = s2 + 1;
 942         }
 943         RETURN(0);
 944 }
 945
 946 void ll_lli_init(struct ll_inode_info *lli)
 947 {
 948         lli->lli_inode_magic = LLI_INODE_MAGIC;
 949         lli->lli_flags = 0;
 950         spin_lock_init(&lli->lli_lock);
 951         lli->lli_posix_acl = NULL;
 952         /* Do not set lli_fid, it has been initialized already. */
 953         fid_zero(&lli->lli_pfid);
 954         lli->lli_mds_read_och = NULL;
 955         lli->lli_mds_write_och = NULL;
 956         lli->lli_mds_exec_och = NULL;
 957         lli->lli_open_fd_read_count = 0;
 958         lli->lli_open_fd_write_count = 0;
 959         lli->lli_open_fd_exec_count = 0;
 960         mutex_init(&lli->lli_och_mutex);
 961         spin_lock_init(&lli->lli_agl_lock);
 962         spin_lock_init(&lli->lli_layout_lock);
 963         ll_layout_version_set(lli, CL_LAYOUT_GEN_NONE);
 964         lli->lli_clob = NULL;
 965
 966         init_rwsem(&lli->lli_xattrs_list_rwsem);
 967         mutex_init(&lli->lli_xattrs_enq_lock);
 968
 969         LASSERT(lli->lli_vfs_inode.i_mode != 0);
 970         if (S_ISDIR(lli->lli_vfs_inode.i_mode)) {
 971                 mutex_init(&lli->lli_readdir_mutex);
 972                 lli->lli_opendir_key = NULL;
 973                 lli->lli_sai = NULL;
 974                 spin_lock_init(&lli->lli_sa_lock);
 975                 lli->lli_opendir_pid = 0;
 976                 lli->lli_sa_enabled = 0;
 977                 lli->lli_def_stripe_offset = -1;
 978                 init_rwsem(&lli->lli_lsm_sem);
 979         } else {
 980                 mutex_init(&lli->lli_size_mutex);
 981                 lli->lli_symlink_name = NULL;
 982                 init_rwsem(&lli->lli_trunc_sem);
 983                 range_lock_tree_init(&lli->lli_write_tree);
 984                 init_rwsem(&lli->lli_glimpse_sem);
 985                 lli->lli_glimpse_time = ktime_set(0, 0);
 986                 INIT_LIST_HEAD(&lli->lli_agl_list);
 987                 lli->lli_agl_index = 0;
 988                 lli->lli_async_rc = 0;
 989         }
 990         mutex_init(&lli->lli_layout_mutex);
 991         memset(lli->lli_jobid, 0, sizeof(lli->lli_jobid));
 992 }
 993
 994 #define MAX_STRING_SIZE 128
 995
 996 #ifndef HAVE_SUPER_SETUP_BDI_NAME
 997
 998 #define LSI_BDI_INITIALIZED     0x00400000
 999
1000 #ifndef HAVE_BDI_CAP_MAP_COPY
1001 # define BDI_CAP_MAP_COPY       0
1002 #endif
1003
1004 static int super_setup_bdi_name(struct super_block *sb, char *fmt, ...)
1005 {
1006         struct  lustre_sb_info *lsi = s2lsi(sb);
1007         char buf[MAX_STRING_SIZE];
1008         va_list args;
1009         int err;
1010
1011         err = bdi_init(&lsi->lsi_bdi);
1012         if (err)
1013                 return err;
1014
1015         lsi->lsi_flags |= LSI_BDI_INITIALIZED;
1016         lsi->lsi_bdi.capabilities = BDI_CAP_MAP_COPY;
1017         lsi->lsi_bdi.name = "lustre";
1018         va_start(args, fmt);
1019         vsnprintf(buf, MAX_STRING_SIZE, fmt, args);
1020         va_end(args);
1021         err = bdi_register(&lsi->lsi_bdi, NULL, "%s", buf);
1022         va_end(args);
1023         if (!err)
1024                 sb->s_bdi = &lsi->lsi_bdi;
1025
1026         return err;
1027 }
1028 #endif /* !HAVE_SUPER_SETUP_BDI_NAME */
1029
1030 int ll_fill_super(struct super_block *sb, struct vfsmount *mnt)
1031 {
1032         struct  lustre_profile *lprof = NULL;
1033         struct  lustre_sb_info *lsi = s2lsi(sb);
1034         struct  ll_sb_info *sbi = NULL;
1035         char    *dt = NULL, *md = NULL;
1036         char    *profilenm = get_profile_name(sb);
1037         struct config_llog_instance *cfg;
1038         /* %p for void* in printf needs 16+2 characters: 0xffffffffffffffff */
1039         const int instlen = 16 + 2;
1040         unsigned long cfg_instance = ll_get_cfg_instance(sb);
1041         char name[MAX_STRING_SIZE];
1042         int md_len = 0;
1043         int dt_len = 0;
1044         char *ptr;
1045         int len;
1046         int err;
1047
1048         ENTRY;
1049         /* for ASLR, to map between cfg_instance and hashed ptr */
1050         CDEBUG(D_VFSTRACE, "VFS Op: cfg_instance %s-%016lx (sb %p)\n",
1051                profilenm, cfg_instance, sb);
1052
1053         try_module_get(THIS_MODULE);
1054
1055         OBD_ALLOC_PTR(cfg);
1056         if (cfg == NULL)
1057                 GOTO(out_free_cfg, err = -ENOMEM);
1058
1059         /* client additional sb info */
1060         lsi->lsi_llsbi = sbi = ll_init_sbi();
1061         if (!sbi)
1062                 GOTO(out_free_cfg, err = -ENOMEM);
1063
1064         err = ll_options(lsi->lsi_lmd->lmd_opts, sbi);
1065         if (err)
1066                 GOTO(out_free_cfg, err);
1067
1068         err = super_setup_bdi_name(sb, "lustre-%016lx", cfg_instance);
1069         if (err)
1070                 GOTO(out_free_cfg, err);
1071
1072 #ifndef HAVE_DCACHE_LOCK
1073         /* kernel >= 2.6.38 store dentry operations in sb->s_d_op. */
1074         sb->s_d_op = &ll_d_ops;
1075 #endif
1076         /* Get fsname */
1077         len = strlen(profilenm);
1078         ptr = strrchr(profilenm, '-');
1079         if (ptr && (strcmp(ptr, "-client") == 0))
1080                 len -= 7;
1081
1082         /* Mount info */
1083         snprintf(name, MAX_STRING_SIZE, "%.*s-%016lx", len,
1084                  profilenm, cfg_instance);
1085
1086         /* Call ll_debugfs_register_super() before lustre_process_log()
1087          * so that "llite.*.*" params can be processed correctly.
1088          */
1089         err = ll_debugfs_register_super(sb, name);
1090         if (err < 0) {
1091                 CERROR("%s: could not register mountpoint in llite: rc = %d\n",
1092                        ll_get_fsname(sb, NULL, 0), err);
1093                 err = 0;
1094         }
1095
1096         /* The cfg_instance is a value unique to this super, in case some
1097          * joker tries to mount the same fs at two mount points.
1098          */
1099         cfg->cfg_instance = cfg_instance;
1100         cfg->cfg_uuid = lsi->lsi_llsbi->ll_sb_uuid;
1101         cfg->cfg_callback = class_config_llog_handler;
1102         cfg->cfg_sub_clds = CONFIG_SUB_CLIENT;
1103         /* set up client obds */
1104         err = lustre_process_log(sb, profilenm, cfg);
1105         if (err < 0)
1106                 GOTO(out_debugfs, err);
1107
1108         /* Profile set with LCFG_MOUNTOPT so we can find our mdc and osc obds */
1109         lprof = class_get_profile(profilenm);
1110         if (lprof == NULL) {
1111                 LCONSOLE_ERROR_MSG(0x156, "The client profile '%s' could not be"
1112                                    " read from the MGS.  Does that filesystem "
1113                                    "exist?\n", profilenm);
1114                 GOTO(out_debugfs, err = -EINVAL);
1115         }
1116         CDEBUG(D_CONFIG, "Found profile %s: mdc=%s osc=%s\n", profilenm,
1117                lprof->lp_md, lprof->lp_dt);
1118
1119         dt_len = strlen(lprof->lp_dt) + instlen + 2;
1120         OBD_ALLOC(dt, dt_len);
1121         if (!dt)
1122                 GOTO(out_profile, err = -ENOMEM);
1123         snprintf(dt, dt_len - 1, "%s-%016lx", lprof->lp_dt, cfg_instance);
1124
1125         md_len = strlen(lprof->lp_md) + instlen + 2;
1126         OBD_ALLOC(md, md_len);
1127         if (!md)
1128                 GOTO(out_free_dt, err = -ENOMEM);
1129         snprintf(md, md_len - 1, "%s-%016lx", lprof->lp_md, cfg_instance);
1130
1131         /* connections, registrations, sb setup */
1132         err = client_common_fill_super(sb, md, dt, mnt);
1133         if (err < 0)
1134                 GOTO(out_free_md, err);
1135
1136         sbi->ll_client_common_fill_super_succeeded = 1;
1137
1138 out_free_md:
1139         if (md)
1140                 OBD_FREE(md, md_len);
1141 out_free_dt:
1142         if (dt)
1143                 OBD_FREE(dt, dt_len);
1144 out_profile:
1145         if (lprof)
1146                 class_put_profile(lprof);
1147 out_debugfs:
1148         if (err < 0)
1149                 ll_debugfs_unregister_super(sb);
1150 out_free_cfg:
1151         if (cfg)
1152                 OBD_FREE_PTR(cfg);
1153
1154         if (err)
1155                 ll_put_super(sb);
1156         else if (sbi->ll_flags & LL_SBI_VERBOSE)
1157                 LCONSOLE_WARN("Mounted %s\n", profilenm);
1158         RETURN(err);
1159 } /* ll_fill_super */
1160
1161 void ll_put_super(struct super_block *sb)
1162 {
1163         struct config_llog_instance cfg, params_cfg;
1164         struct obd_device *obd;
1165         struct lustre_sb_info *lsi = s2lsi(sb);
1166         struct ll_sb_info *sbi = ll_s2sbi(sb);
1167         char *profilenm = get_profile_name(sb);
1168         unsigned long cfg_instance = ll_get_cfg_instance(sb);
1169         long ccc_count;
1170         int next, force = 1, rc = 0;
1171         ENTRY;
1172
1173         if (!sbi)
1174                 GOTO(out_no_sbi, 0);
1175
1176         /* Should replace instance_id with something better for ASLR */
1177         CDEBUG(D_VFSTRACE, "VFS Op: cfg_instance %s-%016lx (sb %p)\n",
1178                profilenm, cfg_instance, sb);
1179
1180         cfg.cfg_instance = cfg_instance;
1181         lustre_end_log(sb, profilenm, &cfg);
1182
1183         params_cfg.cfg_instance = cfg_instance;
1184         lustre_end_log(sb, PARAMS_FILENAME, &params_cfg);
1185
1186         if (sbi->ll_md_exp) {
1187                 obd = class_exp2obd(sbi->ll_md_exp);
1188                 if (obd)
1189                         force = obd->obd_force;
1190         }
1191
1192         /* Wait for unstable pages to be committed to stable storage */
1193         if (force == 0) {
1194                 struct l_wait_info lwi = LWI_INTR(LWI_ON_SIGNAL_NOOP, NULL);
1195                 rc = l_wait_event(sbi->ll_cache->ccc_unstable_waitq,
1196                         atomic_long_read(&sbi->ll_cache->ccc_unstable_nr) == 0,
1197                         &lwi);
1198         }
1199
1200         ccc_count = atomic_long_read(&sbi->ll_cache->ccc_unstable_nr);
1201         if (force == 0 && rc != -EINTR)
1202                 LASSERTF(ccc_count == 0, "count: %li\n", ccc_count);
1203
1204         /* We need to set force before the lov_disconnect in
1205            lustre_common_put_super, since l_d cleans up osc's as well. */
1206         if (force) {
1207                 next = 0;
1208                 while ((obd = class_devices_in_group(&sbi->ll_sb_uuid,
1209                                                      &next)) != NULL) {
1210                         obd->obd_force = force;
1211                 }
1212         }
1213
1214         if (sbi->ll_client_common_fill_super_succeeded) {
1215                 /* Only if client_common_fill_super succeeded */
1216                 client_common_put_super(sb);
1217         }
1218
1219         next = 0;
1220         while ((obd = class_devices_in_group(&sbi->ll_sb_uuid, &next)) !=NULL) {
1221                 class_manual_cleanup(obd);
1222         }
1223
1224         if (sbi->ll_flags & LL_SBI_VERBOSE)
1225                 LCONSOLE_WARN("Unmounted %s\n", profilenm ? profilenm : "");
1226
1227         if (profilenm)
1228                 class_del_profile(profilenm);
1229
1230 #ifndef HAVE_SUPER_SETUP_BDI_NAME
1231         if (lsi->lsi_flags & LSI_BDI_INITIALIZED) {
1232                 bdi_destroy(&lsi->lsi_bdi);
1233                 lsi->lsi_flags &= ~LSI_BDI_INITIALIZED;
1234         }
1235 #endif
1236
1237         ll_free_sbi(sb);
1238         lsi->lsi_llsbi = NULL;
1239 out_no_sbi:
1240         lustre_common_put_super(sb);
1241
1242         cl_env_cache_purge(~0);
1243
1244         module_put(THIS_MODULE);
1245
1246         EXIT;
1247 } /* client_put_super */
1248
1249 struct inode *ll_inode_from_resource_lock(struct ldlm_lock *lock)
1250 {
1251         struct inode *inode = NULL;
1252
1253         /* NOTE: we depend on atomic igrab() -bzzz */
1254         lock_res_and_lock(lock);
1255         if (lock->l_resource->lr_lvb_inode) {
1256                 struct ll_inode_info * lli;
1257                 lli = ll_i2info(lock->l_resource->lr_lvb_inode);
1258                 if (lli->lli_inode_magic == LLI_INODE_MAGIC) {
1259                         inode = igrab(lock->l_resource->lr_lvb_inode);
1260                 } else {
1261                         inode = lock->l_resource->lr_lvb_inode;
1262                         LDLM_DEBUG_LIMIT(inode->i_state & I_FREEING ?  D_INFO :
1263                                          D_WARNING, lock, "lr_lvb_inode %p is "
1264                                          "bogus: magic %08x",
1265                                          lock->l_resource->lr_lvb_inode,
1266                                          lli->lli_inode_magic);
1267                         inode = NULL;
1268                 }
1269         }
1270         unlock_res_and_lock(lock);
1271         return inode;
1272 }
1273
1274 void ll_dir_clear_lsm_md(struct inode *inode)
1275 {
1276         struct ll_inode_info *lli = ll_i2info(inode);
1277
1278         LASSERT(S_ISDIR(inode->i_mode));
1279
1280         if (lli->lli_lsm_md != NULL) {
1281                 lmv_free_memmd(lli->lli_lsm_md);
1282                 lli->lli_lsm_md = NULL;
1283         }
1284 }
1285
1286 static struct inode *ll_iget_anon_dir(struct super_block *sb,
1287                                       const struct lu_fid *fid,
1288                                       struct lustre_md *md)
1289 {
1290         struct ll_sb_info       *sbi = ll_s2sbi(sb);
1291         struct mdt_body         *body = md->body;
1292         struct inode            *inode;
1293         ino_t                   ino;
1294         ENTRY;
1295
1296         ino = cl_fid_build_ino(fid, sbi->ll_flags & LL_SBI_32BIT_API);
1297         inode = iget_locked(sb, ino);
1298         if (inode == NULL) {
1299                 CERROR("%s: failed get simple inode "DFID": rc = -ENOENT\n",
1300                        ll_get_fsname(sb, NULL, 0), PFID(fid));
1301                 RETURN(ERR_PTR(-ENOENT));
1302         }
1303
1304         if (inode->i_state & I_NEW) {
1305                 struct ll_inode_info *lli = ll_i2info(inode);
1306                 struct lmv_stripe_md *lsm = md->lmv;
1307
1308                 inode->i_mode = (inode->i_mode & ~S_IFMT) |
1309                                 (body->mbo_mode & S_IFMT);
1310                 LASSERTF(S_ISDIR(inode->i_mode), "Not slave inode "DFID"\n",
1311                          PFID(fid));
1312
1313                 inode->i_mtime.tv_sec = 0;
1314                 inode->i_atime.tv_sec = 0;
1315                 inode->i_ctime.tv_sec = 0;
1316                 inode->i_rdev = 0;
1317
1318 #ifdef HAVE_BACKING_DEV_INFO
1319                 /* initializing backing dev info. */
1320                 inode->i_mapping->backing_dev_info =
1321                                                 &s2lsi(inode->i_sb)->lsi_bdi;
1322 #endif
1323                 inode->i_op = &ll_dir_inode_operations;
1324                 inode->i_fop = &ll_dir_operations;
1325                 lli->lli_fid = *fid;
1326                 ll_lli_init(lli);
1327
1328                 LASSERT(lsm != NULL);
1329                 /* master object FID */
1330                 lli->lli_pfid = body->mbo_fid1;
1331                 CDEBUG(D_INODE, "lli %p slave "DFID" master "DFID"\n",
1332                        lli, PFID(fid), PFID(&lli->lli_pfid));
1333                 unlock_new_inode(inode);
1334         }
1335
1336         RETURN(inode);
1337 }
1338
1339 static int ll_init_lsm_md(struct inode *inode, struct lustre_md *md)
1340 {
1341         struct lu_fid *fid;
1342         struct lmv_stripe_md *lsm = md->lmv;
1343         struct ll_inode_info *lli = ll_i2info(inode);
1344         int i;
1345
1346         LASSERT(lsm != NULL);
1347
1348         CDEBUG(D_INODE, "%s: "DFID" set dir layout:\n",
1349                 ll_get_fsname(inode->i_sb, NULL, 0),
1350                 PFID(&lli->lli_fid));
1351         lsm_md_dump(D_INODE, lsm);
1352
1353         /* XXX sigh, this lsm_root initialization should be in
1354          * LMV layer, but it needs ll_iget right now, so we
1355          * put this here right now. */
1356         for (i = 0; i < lsm->lsm_md_stripe_count; i++) {
1357                 fid = &lsm->lsm_md_oinfo[i].lmo_fid;
1358                 LASSERT(lsm->lsm_md_oinfo[i].lmo_root == NULL);
1359
1360                 if (!fid_is_sane(fid))
1361                         continue;
1362
1363                 /* Unfortunately ll_iget will call ll_update_inode,
1364                  * where the initialization of slave inode is slightly
1365                  * different, so it reset lsm_md to NULL to avoid
1366                  * initializing lsm for slave inode. */
1367                 lsm->lsm_md_oinfo[i].lmo_root =
1368                                 ll_iget_anon_dir(inode->i_sb, fid, md);
1369                 if (IS_ERR(lsm->lsm_md_oinfo[i].lmo_root)) {
1370                         int rc = PTR_ERR(lsm->lsm_md_oinfo[i].lmo_root);
1371
1372                         lsm->lsm_md_oinfo[i].lmo_root = NULL;
1373                         while (i-- > 0) {
1374                                 iput(lsm->lsm_md_oinfo[i].lmo_root);
1375                                 lsm->lsm_md_oinfo[i].lmo_root = NULL;
1376                         }
1377                         return rc;
1378                 }
1379         }
1380
1381         lli->lli_lsm_md = lsm;
1382
1383         return 0;
1384 }
1385
1386 static int ll_update_lsm_md(struct inode *inode, struct lustre_md *md)
1387 {
1388         struct ll_inode_info *lli = ll_i2info(inode);
1389         struct lmv_stripe_md *lsm = md->lmv;
1390         struct cl_attr  *attr;
1391         int rc = 0;
1392
1393         ENTRY;
1394
1395         LASSERT(S_ISDIR(inode->i_mode));
1396         CDEBUG(D_INODE, "update lsm %p of "DFID"\n", lli->lli_lsm_md,
1397                PFID(ll_inode2fid(inode)));
1398
1399         /*
1400          * no striped information from request, lustre_md from req does not
1401          * include stripeEA, see ll_md_setattr()
1402          */
1403         if (!lsm)
1404                 RETURN(0);
1405
1406         /*
1407          * normally dir layout doesn't change, only take read lock to check
1408          * that to avoid blocking other MD operations.
1409          */
1410         down_read(&lli->lli_lsm_sem);
1411
1412         /* some concurrent lookup initialized lsm, and unchanged */
1413         if (lli->lli_lsm_md && lsm_md_eq(lli->lli_lsm_md, lsm))
1414                 GOTO(unlock, rc = 0);
1415
1416         /* if dir layout doesn't match, check whether version is increased,
1417          * which means layout is changed, this happens in dir split/merge and
1418          * lfsck.
1419          */
1420         if (lli->lli_lsm_md &&
1421             lsm->lsm_md_layout_version <=
1422             lli->lli_lsm_md->lsm_md_layout_version) {
1423                 CERROR("%s: "DFID" dir layout mismatch:\n",
1424                        ll_get_fsname(inode->i_sb, NULL, 0),
1425                        PFID(&lli->lli_fid));
1426                 lsm_md_dump(D_ERROR, lli->lli_lsm_md);
1427                 lsm_md_dump(D_ERROR, lsm);
1428                 GOTO(unlock, rc = -EINVAL);
1429         }
1430
1431         up_read(&lli->lli_lsm_sem);
1432         down_write(&lli->lli_lsm_sem);
1433         /* clear existing lsm */
1434         if (lli->lli_lsm_md) {
1435                 lmv_free_memmd(lli->lli_lsm_md);
1436                 lli->lli_lsm_md = NULL;
1437         }
1438
1439         rc = ll_init_lsm_md(inode, md);
1440         up_write(&lli->lli_lsm_sem);
1441         if (rc)
1442                 RETURN(rc);
1443
1444         /* set md->lmv to NULL, so the following free lustre_md will not free
1445          * this lsm.
1446          */
1447         md->lmv = NULL;
1448
1449         /* md_merge_attr() may take long, since lsm is already set, switch to
1450          * read lock.
1451          */
1452         down_read(&lli->lli_lsm_sem);
1453         OBD_ALLOC_PTR(attr);
1454         if (!attr)
1455                 GOTO(unlock, rc = -ENOMEM);
1456
1457         /* validate the lsm */
1458         rc = md_merge_attr(ll_i2mdexp(inode), lli->lli_lsm_md, attr,
1459                            ll_md_blocking_ast);
1460         if (!rc) {
1461                 if (md->body->mbo_valid & OBD_MD_FLNLINK)
1462                         md->body->mbo_nlink = attr->cat_nlink;
1463                 if (md->body->mbo_valid & OBD_MD_FLSIZE)
1464                         md->body->mbo_size = attr->cat_size;
1465                 if (md->body->mbo_valid & OBD_MD_FLATIME)
1466                         md->body->mbo_atime = attr->cat_atime;
1467                 if (md->body->mbo_valid & OBD_MD_FLCTIME)
1468                         md->body->mbo_ctime = attr->cat_ctime;
1469                 if (md->body->mbo_valid & OBD_MD_FLMTIME)
1470                         md->body->mbo_mtime = attr->cat_mtime;
1471         }
1472
1473         OBD_FREE_PTR(attr);
1474         GOTO(unlock, rc);
1475 unlock:
1476         up_read(&lli->lli_lsm_sem);
1477
1478         return rc;
1479 }
1480
1481 void ll_clear_inode(struct inode *inode)
1482 {
1483         struct ll_inode_info *lli = ll_i2info(inode);
1484         struct ll_sb_info *sbi = ll_i2sbi(inode);
1485         ENTRY;
1486
1487         CDEBUG(D_VFSTRACE, "VFS Op:inode="DFID"(%p)\n",
1488                PFID(ll_inode2fid(inode)), inode);
1489
1490         if (S_ISDIR(inode->i_mode)) {
1491                 /* these should have been cleared in ll_file_release */
1492                 LASSERT(lli->lli_opendir_key == NULL);
1493                 LASSERT(lli->lli_sai == NULL);
1494                 LASSERT(lli->lli_opendir_pid == 0);
1495         }
1496
1497         md_null_inode(sbi->ll_md_exp, ll_inode2fid(inode));
1498
1499         LASSERT(!lli->lli_open_fd_write_count);
1500         LASSERT(!lli->lli_open_fd_read_count);
1501         LASSERT(!lli->lli_open_fd_exec_count);
1502
1503         if (lli->lli_mds_write_och)
1504                 ll_md_real_close(inode, FMODE_WRITE);
1505         if (lli->lli_mds_exec_och)
1506                 ll_md_real_close(inode, FMODE_EXEC);
1507         if (lli->lli_mds_read_och)
1508                 ll_md_real_close(inode, FMODE_READ);
1509
1510         if (S_ISLNK(inode->i_mode) && lli->lli_symlink_name) {
1511                 OBD_FREE(lli->lli_symlink_name,
1512                          strlen(lli->lli_symlink_name) + 1);
1513                 lli->lli_symlink_name = NULL;
1514         }
1515
1516         ll_xattr_cache_destroy(inode);
1517
1518 #ifdef CONFIG_FS_POSIX_ACL
1519         forget_all_cached_acls(inode);
1520         if (lli->lli_posix_acl) {
1521                 posix_acl_release(lli->lli_posix_acl);
1522                 lli->lli_posix_acl = NULL;
1523         }
1524 #endif
1525         lli->lli_inode_magic = LLI_INODE_DEAD;
1526
1527         if (S_ISDIR(inode->i_mode))
1528                 ll_dir_clear_lsm_md(inode);
1529         else if (S_ISREG(inode->i_mode) && !is_bad_inode(inode))
1530                 LASSERT(list_empty(&lli->lli_agl_list));
1531
1532         /*
1533          * XXX This has to be done before lsm is freed below, because
1534          * cl_object still uses inode lsm.
1535          */
1536         cl_inode_fini(inode);
1537
1538         EXIT;
1539 }
1540
1541 static int ll_md_setattr(struct dentry *dentry, struct md_op_data *op_data)
1542 {
1543         struct lustre_md md;
1544         struct inode *inode = dentry->d_inode;
1545         struct ll_sb_info *sbi = ll_i2sbi(inode);
1546         struct ptlrpc_request *request = NULL;
1547         int rc, ia_valid;
1548         ENTRY;
1549
1550         op_data = ll_prep_md_op_data(op_data, inode, NULL, NULL, 0, 0,
1551                                      LUSTRE_OPC_ANY, NULL);
1552         if (IS_ERR(op_data))
1553                 RETURN(PTR_ERR(op_data));
1554
1555         rc = md_setattr(sbi->ll_md_exp, op_data, NULL, 0, &request);
1556         if (rc) {
1557                 ptlrpc_req_finished(request);
1558                 if (rc == -ENOENT) {
1559                         clear_nlink(inode);
1560                         /* Unlinked special device node? Or just a race?
1561                          * Pretend we done everything. */
1562                         if (!S_ISREG(inode->i_mode) &&
1563                             !S_ISDIR(inode->i_mode)) {
1564                                 ia_valid = op_data->op_attr.ia_valid;
1565                                 op_data->op_attr.ia_valid &= ~TIMES_SET_FLAGS;
1566                                 rc = simple_setattr(dentry, &op_data->op_attr);
1567                                 op_data->op_attr.ia_valid = ia_valid;
1568                         }
1569                 } else if (rc != -EPERM && rc != -EACCES && rc != -ETXTBSY) {
1570                         CERROR("md_setattr fails: rc = %d\n", rc);
1571                 }
1572                 RETURN(rc);
1573         }
1574
1575         rc = md_get_lustre_md(sbi->ll_md_exp, request, sbi->ll_dt_exp,
1576                               sbi->ll_md_exp, &md);
1577         if (rc) {
1578                 ptlrpc_req_finished(request);
1579                 RETURN(rc);
1580         }
1581
1582         ia_valid = op_data->op_attr.ia_valid;
1583         /* inode size will be in ll_setattr_ost, can't do it now since dirty
1584          * cache is not cleared yet. */
1585         op_data->op_attr.ia_valid &= ~(TIMES_SET_FLAGS | ATTR_SIZE);
1586         if (S_ISREG(inode->i_mode))
1587                 inode_lock(inode);
1588         rc = simple_setattr(dentry, &op_data->op_attr);
1589         if (S_ISREG(inode->i_mode))
1590                 inode_unlock(inode);
1591         op_data->op_attr.ia_valid = ia_valid;
1592
1593         rc = ll_update_inode(inode, &md);
1594         ptlrpc_req_finished(request);
1595
1596         RETURN(rc);
1597 }
1598
1599 /* If this inode has objects allocated to it (lsm != NULL), then the OST
1600  * object(s) determine the file size and mtime.  Otherwise, the MDS will
1601  * keep these values until such a time that objects are allocated for it.
1602  * We do the MDS operations first, as it is checking permissions for us.
1603  * We don't to the MDS RPC if there is nothing that we want to store there,
1604  * otherwise there is no harm in updating mtime/atime on the MDS if we are
1605  * going to do an RPC anyways.
1606  *
1607  * If we are doing a truncate, we will send the mtime and ctime updates
1608  * to the OST with the punch RPC, otherwise we do an explicit setattr RPC.
1609  * I don't believe it is possible to get e.g. ATTR_MTIME_SET and ATTR_SIZE
1610  * at the same time.
1611  *
1612  * In case of HSMimport, we only set attr on MDS.
1613  */
1614 int ll_setattr_raw(struct dentry *dentry, struct iattr *attr,
1615                    enum op_xvalid xvalid, bool hsm_import)
1616 {
1617         struct inode *inode = dentry->d_inode;
1618         struct ll_inode_info *lli = ll_i2info(inode);
1619         struct md_op_data *op_data = NULL;
1620         int rc = 0;
1621         ENTRY;
1622
1623         CDEBUG(D_VFSTRACE, "%s: setattr inode "DFID"(%p) from %llu to %llu, "
1624                "valid %x, hsm_import %d\n",
1625                ll_get_fsname(inode->i_sb, NULL, 0), PFID(&lli->lli_fid),
1626                inode, i_size_read(inode), attr->ia_size, attr->ia_valid,
1627                hsm_import);
1628
1629         if (attr->ia_valid & ATTR_SIZE) {
1630                 /* Check new size against VFS/VM file size limit and rlimit */
1631                 rc = inode_newsize_ok(inode, attr->ia_size);
1632                 if (rc)
1633                         RETURN(rc);
1634
1635                 /* The maximum Lustre file size is variable, based on the
1636                  * OST maximum object size and number of stripes.  This
1637                  * needs another check in addition to the VFS check above. */
1638                 if (attr->ia_size > ll_file_maxbytes(inode)) {
1639                         CDEBUG(D_INODE,"file "DFID" too large %llu > %llu\n",
1640                                PFID(&lli->lli_fid), attr->ia_size,
1641                                ll_file_maxbytes(inode));
1642                         RETURN(-EFBIG);
1643                 }
1644
1645                 attr->ia_valid |= ATTR_MTIME | ATTR_CTIME;
1646         }
1647
1648         /* POSIX: check before ATTR_*TIME_SET set (from inode_change_ok) */
1649         if (attr->ia_valid & TIMES_SET_FLAGS) {
1650                 if ((!uid_eq(current_fsuid(), inode->i_uid)) &&
1651                     !cfs_capable(CFS_CAP_FOWNER))
1652                         RETURN(-EPERM);
1653         }
1654
1655         /* We mark all of the fields "set" so MDS/OST does not re-set them */
1656         if (!(xvalid & OP_XVALID_CTIME_SET) &&
1657              (attr->ia_valid & ATTR_CTIME)) {
1658                 attr->ia_ctime = current_time(inode);
1659                 xvalid |= OP_XVALID_CTIME_SET;
1660         }
1661         if (!(attr->ia_valid & ATTR_ATIME_SET) &&
1662             (attr->ia_valid & ATTR_ATIME)) {
1663                 attr->ia_atime = current_time(inode);
1664                 attr->ia_valid |= ATTR_ATIME_SET;
1665         }
1666         if (!(attr->ia_valid & ATTR_MTIME_SET) &&
1667             (attr->ia_valid & ATTR_MTIME)) {
1668                 attr->ia_mtime = current_time(inode);
1669                 attr->ia_valid |= ATTR_MTIME_SET;
1670         }
1671
1672         if (attr->ia_valid & (ATTR_MTIME | ATTR_CTIME))
1673                 CDEBUG(D_INODE, "setting mtime %lld, ctime %lld, now = %lld\n",
1674                        (s64)attr->ia_mtime.tv_sec, (s64)attr->ia_ctime.tv_sec,
1675                        ktime_get_real_seconds());
1676
1677         if (S_ISREG(inode->i_mode)) {
1678                 if (attr->ia_valid & ATTR_SIZE)
1679                         inode_dio_write_done(inode);
1680                 inode_unlock(inode);
1681         }
1682
1683         /* We always do an MDS RPC, even if we're only changing the size;
1684          * only the MDS knows whether truncate() should fail with -ETXTBUSY */
1685
1686         OBD_ALLOC_PTR(op_data);
1687         if (op_data == NULL)
1688                 GOTO(out, rc = -ENOMEM);
1689
1690         if (!hsm_import && attr->ia_valid & ATTR_SIZE) {
1691                 /* If we are changing file size, file content is
1692                  * modified, flag it.
1693                  */
1694                 xvalid |= OP_XVALID_OWNEROVERRIDE;
1695                 op_data->op_bias |= MDS_DATA_MODIFIED;
1696                 ll_file_clear_flag(lli, LLIF_DATA_MODIFIED);
1697         }
1698
1699         if (attr->ia_valid & ATTR_FILE) {
1700                 struct ll_file_data *fd = LUSTRE_FPRIVATE(attr->ia_file);
1701
1702                 if (fd->fd_lease_och)
1703                         op_data->op_bias |= MDS_TRUNC_KEEP_LEASE;
1704         }
1705
1706         op_data->op_attr = *attr;
1707         op_data->op_xvalid = xvalid;
1708
1709         rc = ll_md_setattr(dentry, op_data);
1710         if (rc)
1711                 GOTO(out, rc);
1712
1713         if (!S_ISREG(inode->i_mode) || hsm_import)
1714                 GOTO(out, rc = 0);
1715
1716         if (attr->ia_valid & (ATTR_SIZE | ATTR_ATIME | ATTR_ATIME_SET |
1717                               ATTR_MTIME | ATTR_MTIME_SET | ATTR_CTIME) ||
1718             xvalid & OP_XVALID_CTIME_SET) {
1719                 /* For truncate and utimes sending attributes to OSTs, setting
1720                  * mtime/atime to the past will be performed under PW [0:EOF]
1721                  * extent lock (new_size:EOF for truncate).  It may seem
1722                  * excessive to send mtime/atime updates to OSTs when not
1723                  * setting times to past, but it is necessary due to possible
1724                  * time de-synchronization between MDT inode and OST objects
1725                  */
1726                 rc = cl_setattr_ost(lli->lli_clob, attr, xvalid, 0);
1727         }
1728
1729         /* If the file was restored, it needs to set dirty flag.
1730          *
1731          * We've already sent MDS_DATA_MODIFIED flag in
1732          * ll_md_setattr() for truncate. However, the MDT refuses to
1733          * set the HS_DIRTY flag on released files, so we have to set
1734          * it again if the file has been restored. Please check how
1735          * LLIF_DATA_MODIFIED is set in vvp_io_setattr_fini().
1736          *
1737          * Please notice that if the file is not released, the previous
1738          * MDS_DATA_MODIFIED has taken effect and usually
1739          * LLIF_DATA_MODIFIED is not set(see vvp_io_setattr_fini()).
1740          * This way we can save an RPC for common open + trunc
1741          * operation. */
1742         if (ll_file_test_and_clear_flag(lli, LLIF_DATA_MODIFIED)) {
1743                 struct hsm_state_set hss = {
1744                         .hss_valid = HSS_SETMASK,
1745                         .hss_setmask = HS_DIRTY,
1746                 };
1747                 int rc2;
1748
1749                 rc2 = ll_hsm_state_set(inode, &hss);
1750                 /* truncate and write can happen at the same time, so that
1751                  * the file can be set modified even though the file is not
1752                  * restored from released state, and ll_hsm_state_set() is
1753                  * not applicable for the file, and rc2 < 0 is normal in this
1754                  * case. */
1755                 if (rc2 < 0)
1756                         CDEBUG(D_INFO, DFID "HSM set dirty failed: rc2 = %d\n",
1757                                PFID(ll_inode2fid(inode)), rc2);
1758         }
1759
1760         EXIT;
1761 out:
1762         if (op_data != NULL)
1763                 ll_finish_md_op_data(op_data);
1764
1765         if (S_ISREG(inode->i_mode)) {
1766                 inode_lock(inode);
1767                 if ((attr->ia_valid & ATTR_SIZE) && !hsm_import)
1768                         inode_dio_wait(inode);
1769                 /* Once we've got the i_mutex, it's safe to set the S_NOSEC
1770                  * flag.  ll_update_inode (called from ll_md_setattr), clears
1771                  * inode flags, so there is a gap where S_NOSEC is not set.
1772                  * This can cause a writer to take the i_mutex unnecessarily,
1773                  * but this is safe to do and should be rare. */
1774                 inode_has_no_xattr(inode);
1775         }
1776
1777         ll_stats_ops_tally(ll_i2sbi(inode), (attr->ia_valid & ATTR_SIZE) ?
1778                         LPROC_LL_TRUNC : LPROC_LL_SETATTR, 1);
1779
1780         return rc;
1781 }
1782
1783 int ll_setattr(struct dentry *de, struct iattr *attr)
1784 {
1785         int mode = de->d_inode->i_mode;
1786         enum op_xvalid xvalid = 0;
1787
1788         if ((attr->ia_valid & (ATTR_CTIME|ATTR_SIZE|ATTR_MODE)) ==
1789                               (ATTR_CTIME|ATTR_SIZE|ATTR_MODE))
1790                 xvalid |= OP_XVALID_OWNEROVERRIDE;
1791
1792         if (((attr->ia_valid & (ATTR_MODE|ATTR_FORCE|ATTR_SIZE)) ==
1793                                (ATTR_SIZE|ATTR_MODE)) &&
1794             (((mode & S_ISUID) && !(attr->ia_mode & S_ISUID)) ||
1795              (((mode & (S_ISGID|S_IXGRP)) == (S_ISGID|S_IXGRP)) &&
1796               !(attr->ia_mode & S_ISGID))))
1797                 attr->ia_valid |= ATTR_FORCE;
1798
1799         if ((attr->ia_valid & ATTR_MODE) &&
1800             (mode & S_ISUID) &&
1801             !(attr->ia_mode & S_ISUID) &&
1802             !(attr->ia_valid & ATTR_KILL_SUID))
1803                 attr->ia_valid |= ATTR_KILL_SUID;
1804
1805         if ((attr->ia_valid & ATTR_MODE) &&
1806             ((mode & (S_ISGID|S_IXGRP)) == (S_ISGID|S_IXGRP)) &&
1807             !(attr->ia_mode & S_ISGID) &&
1808             !(attr->ia_valid & ATTR_KILL_SGID))
1809                 attr->ia_valid |= ATTR_KILL_SGID;
1810
1811         return ll_setattr_raw(de, attr, xvalid, false);
1812 }
1813
1814 int ll_statfs_internal(struct ll_sb_info *sbi, struct obd_statfs *osfs,
1815                        u32 flags)
1816 {
1817         struct obd_statfs obd_osfs = { 0 };
1818         time64_t max_age;
1819         int rc;
1820
1821         ENTRY;
1822         max_age = ktime_get_seconds() - sbi->ll_statfs_max_age;
1823
1824         rc = obd_statfs(NULL, sbi->ll_md_exp, osfs, max_age, flags);
1825         if (rc)
1826                 RETURN(rc);
1827
1828         osfs->os_type = LL_SUPER_MAGIC;
1829
1830         CDEBUG(D_SUPER, "MDC blocks %llu/%llu objects %llu/%llu\n",
1831               osfs->os_bavail, osfs->os_blocks, osfs->os_ffree, osfs->os_files);
1832
1833         if (osfs->os_state & OS_STATE_SUM)
1834                 GOTO(out, rc);
1835
1836         if (sbi->ll_flags & LL_SBI_LAZYSTATFS)
1837                 flags |= OBD_STATFS_NODELAY;
1838
1839         rc = obd_statfs(NULL, sbi->ll_dt_exp, &obd_osfs, max_age, flags);
1840         if (rc) /* Possibly a filesystem with no OSTs.  Report MDT totals. */
1841                 GOTO(out, rc = 0);
1842
1843         CDEBUG(D_SUPER, "OSC blocks %llu/%llu objects %llu/%llu\n",
1844                obd_osfs.os_bavail, obd_osfs.os_blocks, obd_osfs.os_ffree,
1845                obd_osfs.os_files);
1846
1847         osfs->os_bsize = obd_osfs.os_bsize;
1848         osfs->os_blocks = obd_osfs.os_blocks;
1849         osfs->os_bfree = obd_osfs.os_bfree;
1850         osfs->os_bavail = obd_osfs.os_bavail;
1851
1852         /* If we have _some_ OSTs, but don't have as many free objects on the
1853          * OSTs as inodes on the MDTs, reduce the reported number of inodes
1854          * to compensate, so that the "inodes in use" number is correct.
1855          * This should be kept in sync with lod_statfs() behaviour.
1856          */
1857         if (obd_osfs.os_files && obd_osfs.os_ffree < osfs->os_ffree) {
1858                 osfs->os_files = (osfs->os_files - osfs->os_ffree) +
1859                                  obd_osfs.os_ffree;
1860                 osfs->os_ffree = obd_osfs.os_ffree;
1861         }
1862
1863 out:
1864         RETURN(rc);
1865 }
1866 int ll_statfs(struct dentry *de, struct kstatfs *sfs)
1867 {
1868         struct super_block *sb = de->d_sb;
1869         struct obd_statfs osfs;
1870         __u64 fsid = huge_encode_dev(sb->s_dev);
1871         int rc;
1872
1873         CDEBUG(D_VFSTRACE, "VFS Op: at %llu jiffies\n", get_jiffies_64());
1874         ll_stats_ops_tally(ll_s2sbi(sb), LPROC_LL_STAFS, 1);
1875
1876         /* Some amount of caching on the client is allowed */
1877         rc = ll_statfs_internal(ll_s2sbi(sb), &osfs, OBD_STATFS_SUM);
1878         if (rc)
1879                 return rc;
1880
1881         statfs_unpack(sfs, &osfs);
1882
1883         /* We need to downshift for all 32-bit kernels, because we can't
1884          * tell if the kernel is being called via sys_statfs64() or not.
1885          * Stop before overflowing f_bsize - in which case it is better
1886          * to just risk EOVERFLOW if caller is using old sys_statfs(). */
1887         if (sizeof(long) < 8) {
1888                 while (osfs.os_blocks > ~0UL && sfs->f_bsize < 0x40000000) {
1889                         sfs->f_bsize <<= 1;
1890
1891                         osfs.os_blocks >>= 1;
1892                         osfs.os_bfree >>= 1;
1893                         osfs.os_bavail >>= 1;
1894                 }
1895         }
1896
1897         sfs->f_blocks = osfs.os_blocks;
1898         sfs->f_bfree = osfs.os_bfree;
1899         sfs->f_bavail = osfs.os_bavail;
1900         sfs->f_fsid.val[0] = (__u32)fsid;
1901         sfs->f_fsid.val[1] = (__u32)(fsid >> 32);
1902         return 0;
1903 }
1904
1905 void ll_inode_size_lock(struct inode *inode)
1906 {
1907         struct ll_inode_info *lli;
1908
1909         LASSERT(!S_ISDIR(inode->i_mode));
1910
1911         lli = ll_i2info(inode);
1912         mutex_lock(&lli->lli_size_mutex);
1913 }
1914
1915 void ll_inode_size_unlock(struct inode *inode)
1916 {
1917         struct ll_inode_info *lli;
1918
1919         lli = ll_i2info(inode);
1920         mutex_unlock(&lli->lli_size_mutex);
1921 }
1922
1923 void ll_update_inode_flags(struct inode *inode, int ext_flags)
1924 {
1925         inode->i_flags = ll_ext_to_inode_flags(ext_flags);
1926         if (ext_flags & LUSTRE_PROJINHERIT_FL)
1927                 ll_file_set_flag(ll_i2info(inode), LLIF_PROJECT_INHERIT);
1928         else
1929                 ll_file_clear_flag(ll_i2info(inode), LLIF_PROJECT_INHERIT);
1930 }
1931
1932 int ll_update_inode(struct inode *inode, struct lustre_md *md)
1933 {
1934         struct ll_inode_info *lli = ll_i2info(inode);
1935         struct mdt_body *body = md->body;
1936         struct ll_sb_info *sbi = ll_i2sbi(inode);
1937         int rc = 0;
1938
1939         if (body->mbo_valid & OBD_MD_FLEASIZE) {
1940                 rc = cl_file_inode_init(inode, md);
1941                 if (rc)
1942                         return rc;
1943         }
1944
1945         if (S_ISDIR(inode->i_mode)) {
1946                 rc = ll_update_lsm_md(inode, md);
1947                 if (rc != 0)
1948                         return rc;
1949         }
1950
1951 #ifdef CONFIG_FS_POSIX_ACL
1952         if (body->mbo_valid & OBD_MD_FLACL) {
1953                 spin_lock(&lli->lli_lock);
1954                 if (lli->lli_posix_acl)
1955                         posix_acl_release(lli->lli_posix_acl);
1956                 lli->lli_posix_acl = md->posix_acl;
1957                 spin_unlock(&lli->lli_lock);
1958         }
1959 #endif
1960         inode->i_ino = cl_fid_build_ino(&body->mbo_fid1,
1961                                         sbi->ll_flags & LL_SBI_32BIT_API);
1962         inode->i_generation = cl_fid_build_gen(&body->mbo_fid1);
1963
1964         if (body->mbo_valid & OBD_MD_FLATIME) {
1965                 if (body->mbo_atime > inode->i_atime.tv_sec)
1966                         inode->i_atime.tv_sec = body->mbo_atime;
1967                 lli->lli_atime = body->mbo_atime;
1968         }
1969
1970         if (body->mbo_valid & OBD_MD_FLMTIME) {
1971                 if (body->mbo_mtime > inode->i_mtime.tv_sec) {
1972                         CDEBUG(D_INODE,
1973                                "setting ino %lu mtime from %lld to %llu\n",
1974                                inode->i_ino, (s64)inode->i_mtime.tv_sec,
1975                                body->mbo_mtime);
1976                         inode->i_mtime.tv_sec = body->mbo_mtime;
1977                 }
1978                 lli->lli_mtime = body->mbo_mtime;
1979         }
1980
1981         if (body->mbo_valid & OBD_MD_FLCTIME) {
1982                 if (body->mbo_ctime > inode->i_ctime.tv_sec)
1983                         inode->i_ctime.tv_sec = body->mbo_ctime;
1984                 lli->lli_ctime = body->mbo_ctime;
1985         }
1986
1987         /* Clear i_flags to remove S_NOSEC before permissions are updated */
1988         if (body->mbo_valid & OBD_MD_FLFLAGS)
1989                 ll_update_inode_flags(inode, body->mbo_flags);
1990         if (body->mbo_valid & OBD_MD_FLMODE)
1991                 inode->i_mode = (inode->i_mode & S_IFMT) |
1992                                 (body->mbo_mode & ~S_IFMT);
1993
1994         if (body->mbo_valid & OBD_MD_FLTYPE)
1995                 inode->i_mode = (inode->i_mode & ~S_IFMT) |
1996                                 (body->mbo_mode & S_IFMT);
1997
1998         LASSERT(inode->i_mode != 0);
1999         if (S_ISREG(inode->i_mode))
2000                 inode->i_blkbits = min(PTLRPC_MAX_BRW_BITS + 1,
2001                                        LL_MAX_BLKSIZE_BITS);
2002         else
2003                 inode->i_blkbits = inode->i_sb->s_blocksize_bits;
2004
2005         if (body->mbo_valid & OBD_MD_FLUID)
2006                 inode->i_uid = make_kuid(&init_user_ns, body->mbo_uid);
2007         if (body->mbo_valid & OBD_MD_FLGID)
2008                 inode->i_gid = make_kgid(&init_user_ns, body->mbo_gid);
2009         if (body->mbo_valid & OBD_MD_FLPROJID)
2010                 lli->lli_projid = body->mbo_projid;
2011         if (body->mbo_valid & OBD_MD_FLNLINK)
2012                 set_nlink(inode, body->mbo_nlink);
2013         if (body->mbo_valid & OBD_MD_FLRDEV)
2014                 inode->i_rdev = old_decode_dev(body->mbo_rdev);
2015
2016         if (body->mbo_valid & OBD_MD_FLID) {
2017                 /* FID shouldn't be changed! */
2018                 if (fid_is_sane(&lli->lli_fid)) {
2019                         LASSERTF(lu_fid_eq(&lli->lli_fid, &body->mbo_fid1),
2020                                  "Trying to change FID "DFID
2021                                  " to the "DFID", inode "DFID"(%p)\n",
2022                                  PFID(&lli->lli_fid), PFID(&body->mbo_fid1),
2023                                  PFID(ll_inode2fid(inode)), inode);
2024                 } else {
2025                         lli->lli_fid = body->mbo_fid1;
2026                 }
2027         }
2028
2029         LASSERT(fid_seq(&lli->lli_fid) != 0);
2030
2031         if (body->mbo_valid & OBD_MD_FLSIZE) {
2032                 i_size_write(inode, body->mbo_size);
2033
2034                 CDEBUG(D_VFSTRACE, "inode="DFID", updating i_size %llu\n",
2035                        PFID(ll_inode2fid(inode)),
2036                        (unsigned long long)body->mbo_size);
2037
2038                 if (body->mbo_valid & OBD_MD_FLBLOCKS)
2039                         inode->i_blocks = body->mbo_blocks;
2040         }
2041
2042         if (body->mbo_valid & OBD_MD_TSTATE) {
2043                 /* Set LLIF_FILE_RESTORING if restore ongoing and
2044                  * clear it when done to ensure to start again
2045                  * glimpsing updated attrs
2046                  */
2047                 if (body->mbo_t_state & MS_RESTORE)
2048                         ll_file_set_flag(lli, LLIF_FILE_RESTORING);
2049                 else
2050                         ll_file_clear_flag(lli, LLIF_FILE_RESTORING);
2051         }
2052
2053         return 0;
2054 }
2055
2056 void ll_truncate_inode_pages_final(struct inode *inode)
2057 {
2058         struct address_space *mapping = &inode->i_data;
2059         unsigned long nrpages;
2060         unsigned long flags;
2061
2062         truncate_inode_pages_final(mapping);
2063
2064         /* Workaround for LU-118: Note nrpages may not be totally updated when
2065          * truncate_inode_pages() returns, as there can be a page in the process
2066          * of deletion (inside __delete_from_page_cache()) in the specified
2067          * range. Thus mapping->nrpages can be non-zero when this function
2068          * returns even after truncation of the whole mapping.  Only do this if
2069          * npages isn't already zero.
2070          */
2071         nrpages = mapping->nrpages;
2072         if (nrpages) {
2073                 xa_lock_irqsave(&mapping->i_pages, flags);
2074                 nrpages = mapping->nrpages;
2075                 xa_unlock_irqrestore(&mapping->i_pages, flags);
2076         } /* Workaround end */
2077
2078         LASSERTF(nrpages == 0, "%s: inode="DFID"(%p) nrpages=%lu, "
2079                  "see https://jira.whamcloud.com/browse/LU-118\n",
2080                  ll_get_fsname(inode->i_sb, NULL, 0),
2081                  PFID(ll_inode2fid(inode)), inode, nrpages);
2082 }
2083
2084 int ll_read_inode2(struct inode *inode, void *opaque)
2085 {
2086         struct lustre_md *md = opaque;
2087         struct ll_inode_info *lli = ll_i2info(inode);
2088         int     rc;
2089         ENTRY;
2090
2091         CDEBUG(D_VFSTRACE, "VFS Op:inode="DFID"(%p)\n",
2092                PFID(&lli->lli_fid), inode);
2093
2094         /* Core attributes from the MDS first.  This is a new inode, and
2095          * the VFS doesn't zero times in the core inode so we have to do
2096          * it ourselves.  They will be overwritten by either MDS or OST
2097          * attributes - we just need to make sure they aren't newer.
2098          */
2099         inode->i_mtime.tv_sec = 0;
2100         inode->i_atime.tv_sec = 0;
2101         inode->i_ctime.tv_sec = 0;
2102         inode->i_rdev = 0;
2103         rc = ll_update_inode(inode, md);
2104         if (rc != 0)
2105                 RETURN(rc);
2106
2107         /* OIDEBUG(inode); */
2108
2109 #ifdef HAVE_BACKING_DEV_INFO
2110         /* initializing backing dev info. */
2111         inode->i_mapping->backing_dev_info = &s2lsi(inode->i_sb)->lsi_bdi;
2112 #endif
2113         if (S_ISREG(inode->i_mode)) {
2114                 struct ll_sb_info *sbi = ll_i2sbi(inode);
2115                 inode->i_op = &ll_file_inode_operations;
2116                 inode->i_fop = sbi->ll_fop;
2117                 inode->i_mapping->a_ops = (struct address_space_operations *)&ll_aops;
2118                 EXIT;
2119         } else if (S_ISDIR(inode->i_mode)) {
2120                 inode->i_op = &ll_dir_inode_operations;
2121                 inode->i_fop = &ll_dir_operations;
2122                 EXIT;
2123         } else if (S_ISLNK(inode->i_mode)) {
2124                 inode->i_op = &ll_fast_symlink_inode_operations;
2125                 EXIT;
2126         } else {
2127                 inode->i_op = &ll_special_inode_operations;
2128
2129                 init_special_inode(inode, inode->i_mode,
2130                                    inode->i_rdev);
2131
2132                 EXIT;
2133         }
2134
2135         return 0;
2136 }
2137
2138 void ll_delete_inode(struct inode *inode)
2139 {
2140         struct ll_inode_info *lli = ll_i2info(inode);
2141         ENTRY;
2142
2143         if (S_ISREG(inode->i_mode) && lli->lli_clob != NULL) {
2144                 /* It is last chance to write out dirty pages,
2145                  * otherwise we may lose data while umount.
2146                  *
2147                  * If i_nlink is 0 then just discard data. This is safe because
2148                  * local inode gets i_nlink 0 from server only for the last
2149                  * unlink, so that file is not opened somewhere else
2150                  */
2151                 cl_sync_file_range(inode, 0, OBD_OBJECT_EOF, inode->i_nlink ?
2152                                    CL_FSYNC_LOCAL : CL_FSYNC_DISCARD, 1);
2153         }
2154
2155         ll_truncate_inode_pages_final(inode);
2156
2157 #ifdef HAVE_SBOPS_EVICT_INODE
2158         ll_clear_inode(inode);
2159 #endif
2160         clear_inode(inode);
2161
2162         EXIT;
2163 }
2164
2165 int ll_iocontrol(struct inode *inode, struct file *file,
2166                  unsigned int cmd, unsigned long arg)
2167 {
2168         struct ll_sb_info *sbi = ll_i2sbi(inode);
2169         struct ptlrpc_request *req = NULL;
2170         int rc, flags = 0;
2171         ENTRY;
2172
2173         switch (cmd) {
2174         case FS_IOC_GETFLAGS: {
2175                 struct mdt_body *body;
2176                 struct md_op_data *op_data;
2177
2178                 op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL,
2179                                              0, 0, LUSTRE_OPC_ANY,
2180                                              NULL);
2181                 if (IS_ERR(op_data))
2182                         RETURN(PTR_ERR(op_data));
2183
2184                 op_data->op_valid = OBD_MD_FLFLAGS;
2185                 rc = md_getattr(sbi->ll_md_exp, op_data, &req);
2186                 ll_finish_md_op_data(op_data);
2187                 if (rc) {
2188                         CERROR("%s: failure inode "DFID": rc = %d\n",
2189                                sbi->ll_md_exp->exp_obd->obd_name,
2190                                PFID(ll_inode2fid(inode)), rc);
2191                         RETURN(-abs(rc));
2192                 }
2193
2194                 body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
2195
2196                 flags = body->mbo_flags;
2197
2198                 ptlrpc_req_finished(req);
2199
2200                 RETURN(put_user(flags, (int __user *)arg));
2201         }
2202         case FS_IOC_SETFLAGS: {
2203                 struct iattr *attr;
2204                 struct md_op_data *op_data;
2205                 struct cl_object *obj;
2206                 struct fsxattr fa = { 0 };
2207
2208                 if (get_user(flags, (int __user *)arg))
2209                         RETURN(-EFAULT);
2210
2211                 fa.fsx_projid = ll_i2info(inode)->lli_projid;
2212                 if (flags & LUSTRE_PROJINHERIT_FL)
2213                         fa.fsx_xflags = FS_XFLAG_PROJINHERIT;
2214
2215                 rc = ll_ioctl_check_project(inode, &fa);
2216                 if (rc)
2217                         RETURN(rc);
2218
2219                 op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL, 0, 0,
2220                                              LUSTRE_OPC_ANY, NULL);
2221                 if (IS_ERR(op_data))
2222                         RETURN(PTR_ERR(op_data));
2223
2224                 op_data->op_attr_flags = flags;
2225                 op_data->op_xvalid |= OP_XVALID_FLAGS;
2226                 rc = md_setattr(sbi->ll_md_exp, op_data, NULL, 0, &req);
2227                 ll_finish_md_op_data(op_data);
2228                 ptlrpc_req_finished(req);
2229                 if (rc)
2230                         RETURN(rc);
2231
2232                 ll_update_inode_flags(inode, flags);
2233
2234                 obj = ll_i2info(inode)->lli_clob;
2235                 if (obj == NULL)
2236                         RETURN(0);
2237
2238                 OBD_ALLOC_PTR(attr);
2239                 if (attr == NULL)
2240                         RETURN(-ENOMEM);
2241
2242                 rc = cl_setattr_ost(obj, attr, OP_XVALID_FLAGS, flags);
2243
2244                 OBD_FREE_PTR(attr);
2245                 RETURN(rc);
2246         }
2247         default:
2248                 RETURN(-ENOSYS);
2249         }
2250
2251         RETURN(0);
2252 }
2253
2254 int ll_flush_ctx(struct inode *inode)
2255 {
2256         struct ll_sb_info  *sbi = ll_i2sbi(inode);
2257
2258         CDEBUG(D_SEC, "flush context for user %d\n",
2259                from_kuid(&init_user_ns, current_uid()));
2260
2261         obd_set_info_async(NULL, sbi->ll_md_exp,
2262                            sizeof(KEY_FLUSH_CTX), KEY_FLUSH_CTX,
2263                            0, NULL, NULL);
2264         obd_set_info_async(NULL, sbi->ll_dt_exp,
2265                            sizeof(KEY_FLUSH_CTX), KEY_FLUSH_CTX,
2266                            0, NULL, NULL);
2267         return 0;
2268 }
2269
2270 /* umount -f client means force down, don't save state */
2271 void ll_umount_begin(struct super_block *sb)
2272 {
2273         struct ll_sb_info *sbi = ll_s2sbi(sb);
2274         struct obd_device *obd;
2275         struct obd_ioctl_data *ioc_data;
2276         struct l_wait_info lwi;
2277         wait_queue_head_t waitq;
2278         ENTRY;
2279
2280         CDEBUG(D_VFSTRACE, "VFS Op: superblock %p count %d active %d\n", sb,
2281                sb->s_count, atomic_read(&sb->s_active));
2282
2283         obd = class_exp2obd(sbi->ll_md_exp);
2284         if (obd == NULL) {
2285                 CERROR("Invalid MDC connection handle %#llx\n",
2286                        sbi->ll_md_exp->exp_handle.h_cookie);
2287                 EXIT;
2288                 return;
2289         }
2290         obd->obd_force = 1;
2291
2292         obd = class_exp2obd(sbi->ll_dt_exp);
2293         if (obd == NULL) {
2294                 CERROR("Invalid LOV connection handle %#llx\n",
2295                        sbi->ll_dt_exp->exp_handle.h_cookie);
2296                 EXIT;
2297                 return;
2298         }
2299         obd->obd_force = 1;
2300
2301         OBD_ALLOC_PTR(ioc_data);
2302         if (ioc_data) {
2303                 obd_iocontrol(IOC_OSC_SET_ACTIVE, sbi->ll_md_exp,
2304                               sizeof *ioc_data, ioc_data, NULL);
2305
2306                 obd_iocontrol(IOC_OSC_SET_ACTIVE, sbi->ll_dt_exp,
2307                               sizeof *ioc_data, ioc_data, NULL);
2308
2309                 OBD_FREE_PTR(ioc_data);
2310         }
2311
2312         /* Really, we'd like to wait until there are no requests outstanding,
2313          * and then continue.  For now, we just periodically checking for vfs
2314          * to decrement mnt_cnt and hope to finish it within 10sec.
2315          */
2316         init_waitqueue_head(&waitq);
2317         lwi = LWI_TIMEOUT_INTERVAL(cfs_time_seconds(10),
2318                                    cfs_time_seconds(1), NULL, NULL);
2319         l_wait_event(waitq, may_umount(sbi->ll_mnt.mnt), &lwi);
2320
2321         EXIT;
2322 }
2323
2324 int ll_remount_fs(struct super_block *sb, int *flags, char *data)
2325 {
2326         struct ll_sb_info *sbi = ll_s2sbi(sb);
2327         char *profilenm = get_profile_name(sb);
2328         int err;
2329         __u32 read_only;
2330
2331         if ((*flags & MS_RDONLY) != (sb->s_flags & SB_RDONLY)) {
2332                 read_only = *flags & MS_RDONLY;
2333                 err = obd_set_info_async(NULL, sbi->ll_md_exp,
2334                                          sizeof(KEY_READ_ONLY),
2335                                          KEY_READ_ONLY, sizeof(read_only),
2336                                          &read_only, NULL);
2337                 if (err) {
2338                         LCONSOLE_WARN("Failed to remount %s %s (%d)\n",
2339                                       profilenm, read_only ?
2340                                       "read-only" : "read-write", err);
2341                         return err;
2342                 }
2343
2344                 if (read_only)
2345                         sb->s_flags |= SB_RDONLY;
2346                 else
2347                         sb->s_flags &= ~SB_RDONLY;
2348
2349                 if (sbi->ll_flags & LL_SBI_VERBOSE)
2350                         LCONSOLE_WARN("Remounted %s %s\n", profilenm,
2351                                       read_only ?  "read-only" : "read-write");
2352         }
2353         return 0;
2354 }
2355
2356 /**
2357  * Cleanup the open handle that is cached on MDT-side.
2358  *
2359  * For open case, the client side open handling thread may hit error
2360  * after the MDT grant the open. Under such case, the client should
2361  * send close RPC to the MDT as cleanup; otherwise, the open handle
2362  * on the MDT will be leaked there until the client umount or evicted.
2363  *
2364  * In further, if someone unlinked the file, because the open handle
2365  * holds the reference on such file/object, then it will block the
2366  * subsequent threads that want to locate such object via FID.
2367  *
2368  * \param[in] sb        super block for this file-system
2369  * \param[in] open_req  pointer to the original open request
2370  */
2371 void ll_open_cleanup(struct super_block *sb, struct ptlrpc_request *open_req)
2372 {
2373         struct mdt_body                 *body;
2374         struct md_op_data               *op_data;
2375         struct ptlrpc_request           *close_req = NULL;
2376         struct obd_export               *exp       = ll_s2sbi(sb)->ll_md_exp;
2377         ENTRY;
2378
2379         body = req_capsule_server_get(&open_req->rq_pill, &RMF_MDT_BODY);
2380         OBD_ALLOC_PTR(op_data);
2381         if (op_data == NULL) {
2382                 CWARN("%s: cannot allocate op_data to release open handle for "
2383                       DFID"\n",
2384                       ll_get_fsname(sb, NULL, 0), PFID(&body->mbo_fid1));
2385
2386                 RETURN_EXIT;
2387         }
2388
2389         op_data->op_fid1 = body->mbo_fid1;
2390         op_data->op_open_handle = body->mbo_open_handle;
2391         op_data->op_mod_time = ktime_get_real_seconds();
2392         md_close(exp, op_data, NULL, &close_req);
2393         ptlrpc_req_finished(close_req);
2394         ll_finish_md_op_data(op_data);
2395
2396         EXIT;
2397 }
2398
2399 int ll_prep_inode(struct inode **inode, struct ptlrpc_request *req,
2400                   struct super_block *sb, struct lookup_intent *it)
2401 {
2402         struct ll_sb_info *sbi = NULL;
2403         struct lustre_md md = { NULL };
2404         int rc;
2405         ENTRY;
2406
2407         LASSERT(*inode || sb);
2408         sbi = sb ? ll_s2sbi(sb) : ll_i2sbi(*inode);
2409         rc = md_get_lustre_md(sbi->ll_md_exp, req, sbi->ll_dt_exp,
2410                               sbi->ll_md_exp, &md);
2411         if (rc != 0)
2412                 GOTO(cleanup, rc);
2413
2414         if (*inode) {
2415                 rc = ll_update_inode(*inode, &md);
2416                 if (rc != 0)
2417                         GOTO(out, rc);
2418         } else {
2419                 LASSERT(sb != NULL);
2420
2421                 /*
2422                  * At this point server returns to client's same fid as client
2423                  * generated for creating. So using ->fid1 is okay here.
2424                  */
2425                 if (!fid_is_sane(&md.body->mbo_fid1)) {
2426                         CERROR("%s: Fid is insane "DFID"\n",
2427                                 ll_get_fsname(sb, NULL, 0),
2428                                 PFID(&md.body->mbo_fid1));
2429                         GOTO(out, rc = -EINVAL);
2430                 }
2431
2432                 *inode = ll_iget(sb, cl_fid_build_ino(&md.body->mbo_fid1,
2433                                              sbi->ll_flags & LL_SBI_32BIT_API),
2434                                  &md);
2435                 if (IS_ERR(*inode)) {
2436 #ifdef CONFIG_FS_POSIX_ACL
2437                         if (md.posix_acl) {
2438                                 posix_acl_release(md.posix_acl);
2439                                 md.posix_acl = NULL;
2440                         }
2441 #endif
2442                         rc = IS_ERR(*inode) ? PTR_ERR(*inode) : -ENOMEM;
2443                         *inode = NULL;
2444                         CERROR("new_inode -fatal: rc %d\n", rc);
2445                         GOTO(out, rc);
2446                 }
2447         }
2448
2449         /* Handling piggyback layout lock.
2450          * Layout lock can be piggybacked by getattr and open request.
2451          * The lsm can be applied to inode only if it comes with a layout lock
2452          * otherwise correct layout may be overwritten, for example:
2453          * 1. proc1: mdt returns a lsm but not granting layout
2454          * 2. layout was changed by another client
2455          * 3. proc2: refresh layout and layout lock granted
2456          * 4. proc1: to apply a stale layout */
2457         if (it != NULL && it->it_lock_mode != 0) {
2458                 struct lustre_handle lockh;
2459                 struct ldlm_lock *lock;
2460
2461                 lockh.cookie = it->it_lock_handle;
2462                 lock = ldlm_handle2lock(&lockh);
2463                 LASSERT(lock != NULL);
2464                 if (ldlm_has_layout(lock)) {
2465                         struct cl_object_conf conf;
2466
2467                         memset(&conf, 0, sizeof(conf));
2468                         conf.coc_opc = OBJECT_CONF_SET;
2469                         conf.coc_inode = *inode;
2470                         conf.coc_lock = lock;
2471                         conf.u.coc_layout = md.layout;
2472                         (void)ll_layout_conf(*inode, &conf);
2473                 }
2474                 LDLM_LOCK_PUT(lock);
2475         }
2476
2477         GOTO(out, rc = 0);
2478
2479 out:
2480         md_free_lustre_md(sbi->ll_md_exp, &md);
2481
2482 cleanup:
2483         if (rc != 0 && it != NULL && it->it_op & IT_OPEN) {
2484                 ll_intent_drop_lock(it);
2485                 ll_open_cleanup(sb != NULL ? sb : (*inode)->i_sb, req);
2486         }
2487
2488         return rc;
2489 }
2490
2491 int ll_obd_statfs(struct inode *inode, void __user *arg)
2492 {
2493         struct ll_sb_info *sbi = NULL;
2494         struct obd_export *exp;
2495         char *buf = NULL;
2496         struct obd_ioctl_data *data = NULL;
2497         __u32 type;
2498         int len = 0, rc;
2499
2500         if (!inode || !(sbi = ll_i2sbi(inode)))
2501                 GOTO(out_statfs, rc = -EINVAL);
2502
2503         rc = obd_ioctl_getdata(&buf, &len, arg);
2504         if (rc)
2505                 GOTO(out_statfs, rc);
2506
2507         data = (void*)buf;
2508         if (!data->ioc_inlbuf1 || !data->ioc_inlbuf2 ||
2509             !data->ioc_pbuf1 || !data->ioc_pbuf2)
2510                 GOTO(out_statfs, rc = -EINVAL);
2511
2512         if (data->ioc_inllen1 != sizeof(__u32) ||
2513             data->ioc_inllen2 != sizeof(__u32) ||
2514             data->ioc_plen1 != sizeof(struct obd_statfs) ||
2515             data->ioc_plen2 != sizeof(struct obd_uuid))
2516                 GOTO(out_statfs, rc = -EINVAL);
2517
2518         memcpy(&type, data->ioc_inlbuf1, sizeof(__u32));
2519         if (type & LL_STATFS_LMV)
2520                 exp = sbi->ll_md_exp;
2521         else if (type & LL_STATFS_LOV)
2522                 exp = sbi->ll_dt_exp;
2523         else
2524                 GOTO(out_statfs, rc = -ENODEV);
2525
2526         rc = obd_iocontrol(IOC_OBD_STATFS, exp, len, buf, NULL);
2527         if (rc)
2528                 GOTO(out_statfs, rc);
2529 out_statfs:
2530         OBD_FREE_LARGE(buf, len);
2531         return rc;
2532 }
2533
2534 /*
2535  * this is normally called in ll_fini_md_op_data(), but sometimes it needs to
2536  * be called early to avoid deadlock.
2537  */
2538 void ll_unlock_md_op_lsm(struct md_op_data *op_data)
2539 {
2540         if (op_data->op_mea2_sem) {
2541                 up_read(op_data->op_mea2_sem);
2542                 op_data->op_mea2_sem = NULL;
2543         }
2544
2545         if (op_data->op_mea1_sem) {
2546                 up_read(op_data->op_mea1_sem);
2547                 op_data->op_mea1_sem = NULL;
2548         }
2549 }
2550
2551 /* this function prepares md_op_data hint for passing it down to MD stack. */
2552 struct md_op_data *ll_prep_md_op_data(struct md_op_data *op_data,
2553                                       struct inode *i1, struct inode *i2,
2554                                       const char *name, size_t namelen,
2555                                       __u32 mode, __u32 opc, void *data)
2556 {
2557         LASSERT(i1 != NULL);
2558
2559         if (name == NULL) {
2560                 /* Do not reuse namelen for something else. */
2561                 if (namelen != 0)
2562                         return ERR_PTR(-EINVAL);
2563         } else {
2564                 if (namelen > ll_i2sbi(i1)->ll_namelen)
2565                         return ERR_PTR(-ENAMETOOLONG);
2566
2567                 /* "/" is not valid name, but it's allowed */
2568                 if (!lu_name_is_valid_2(name, namelen) &&
2569                     strncmp("/", name, namelen) != 0)
2570                         return ERR_PTR(-EINVAL);
2571         }
2572
2573         if (op_data == NULL)
2574                 OBD_ALLOC_PTR(op_data);
2575
2576         if (op_data == NULL)
2577                 return ERR_PTR(-ENOMEM);
2578
2579         ll_i2gids(op_data->op_suppgids, i1, i2);
2580         op_data->op_fid1 = *ll_inode2fid(i1);
2581         op_data->op_default_stripe_offset = -1;
2582
2583         if (S_ISDIR(i1->i_mode)) {
2584                 down_read(&ll_i2info(i1)->lli_lsm_sem);
2585                 op_data->op_mea1_sem = &ll_i2info(i1)->lli_lsm_sem;
2586                 op_data->op_mea1 = ll_i2info(i1)->lli_lsm_md;
2587                 if (opc == LUSTRE_OPC_MKDIR)
2588                         op_data->op_default_stripe_offset =
2589                                    ll_i2info(i1)->lli_def_stripe_offset;
2590         }
2591
2592         if (i2) {
2593                 op_data->op_fid2 = *ll_inode2fid(i2);
2594                 if (S_ISDIR(i2->i_mode)) {
2595                         if (i2 != i1) {
2596                                 down_read(&ll_i2info(i2)->lli_lsm_sem);
2597                                 op_data->op_mea2_sem =
2598                                                 &ll_i2info(i2)->lli_lsm_sem;
2599                         }
2600                         op_data->op_mea2 = ll_i2info(i2)->lli_lsm_md;
2601                 }
2602         } else {
2603                 fid_zero(&op_data->op_fid2);
2604         }
2605
2606         if (ll_i2sbi(i1)->ll_flags & LL_SBI_64BIT_HASH)
2607                 op_data->op_cli_flags |= CLI_HASH64;
2608
2609         if (ll_need_32bit_api(ll_i2sbi(i1)))
2610                 op_data->op_cli_flags |= CLI_API32;
2611
2612         op_data->op_name = name;
2613         op_data->op_namelen = namelen;
2614         op_data->op_mode = mode;
2615         op_data->op_mod_time = ktime_get_real_seconds();
2616         op_data->op_fsuid = from_kuid(&init_user_ns, current_fsuid());
2617         op_data->op_fsgid = from_kgid(&init_user_ns, current_fsgid());
2618         op_data->op_cap = cfs_curproc_cap_pack();
2619         op_data->op_mds = 0;
2620         if ((opc == LUSTRE_OPC_CREATE) && (name != NULL) &&
2621              filename_is_volatile(name, namelen, &op_data->op_mds)) {
2622                 op_data->op_bias |= MDS_CREATE_VOLATILE;
2623         }
2624         op_data->op_data = data;
2625
2626         return op_data;
2627 }
2628
2629 void ll_finish_md_op_data(struct md_op_data *op_data)
2630 {
2631         ll_unlock_md_op_lsm(op_data);
2632         security_release_secctx(op_data->op_file_secctx,
2633                                 op_data->op_file_secctx_size);
2634         OBD_FREE_PTR(op_data);
2635 }
2636
2637 #ifdef HAVE_SUPEROPS_USE_DENTRY
2638 int ll_show_options(struct seq_file *seq, struct dentry *dentry)
2639 #else
2640 int ll_show_options(struct seq_file *seq, struct vfsmount *vfs)
2641 #endif
2642 {
2643         struct ll_sb_info *sbi;
2644
2645 #ifdef HAVE_SUPEROPS_USE_DENTRY
2646         LASSERT((seq != NULL) && (dentry != NULL));
2647         sbi = ll_s2sbi(dentry->d_sb);
2648 #else
2649         LASSERT((seq != NULL) && (vfs != NULL));
2650         sbi = ll_s2sbi(vfs->mnt_sb);
2651 #endif
2652
2653         if (sbi->ll_flags & LL_SBI_NOLCK)
2654                 seq_puts(seq, ",nolock");
2655
2656         /* "flock" is the default since 2.13, but it wasn't for many years,
2657          * so it is still useful to print this to show it is enabled.
2658          * Start to print "noflock" so it is now clear when flock is disabled.
2659          */
2660         if (sbi->ll_flags & LL_SBI_FLOCK)
2661                 seq_puts(seq, ",flock");
2662         else if (sbi->ll_flags & LL_SBI_LOCALFLOCK)
2663                 seq_puts(seq, ",localflock");
2664         else
2665                 seq_puts(seq, ",noflock");
2666
2667         if (sbi->ll_flags & LL_SBI_USER_XATTR)
2668                 seq_puts(seq, ",user_xattr");
2669
2670         if (sbi->ll_flags & LL_SBI_LAZYSTATFS)
2671                 seq_puts(seq, ",lazystatfs");
2672
2673         if (sbi->ll_flags & LL_SBI_USER_FID2PATH)
2674                 seq_puts(seq, ",user_fid2path");
2675
2676         if (sbi->ll_flags & LL_SBI_ALWAYS_PING)
2677                 seq_puts(seq, ",always_ping");
2678
2679         RETURN(0);
2680 }
2681
2682 /**
2683  * Get obd name by cmd, and copy out to user space
2684  */
2685 int ll_get_obd_name(struct inode *inode, unsigned int cmd, unsigned long arg)
2686 {
2687         struct ll_sb_info *sbi = ll_i2sbi(inode);
2688         struct obd_device *obd;
2689         ENTRY;
2690
2691         if (cmd == OBD_IOC_GETDTNAME)
2692                 obd = class_exp2obd(sbi->ll_dt_exp);
2693         else if (cmd == OBD_IOC_GETMDNAME)
2694                 obd = class_exp2obd(sbi->ll_md_exp);
2695         else
2696                 RETURN(-EINVAL);
2697
2698         if (!obd)
2699                 RETURN(-ENOENT);
2700
2701         if (copy_to_user((void __user *)arg, obd->obd_name,
2702                          strlen(obd->obd_name) + 1))
2703                 RETURN(-EFAULT);
2704
2705         RETURN(0);
2706 }
2707
2708 /**
2709  * Get lustre file system name by \a sbi. If \a buf is provided(non-NULL), the
2710  * fsname will be returned in this buffer; otherwise, a static buffer will be
2711  * used to store the fsname and returned to caller.
2712  */
2713 char *ll_get_fsname(struct super_block *sb, char *buf, int buflen)
2714 {
2715         static char fsname_static[MTI_NAME_MAXLEN];
2716         struct lustre_sb_info *lsi = s2lsi(sb);
2717         char *ptr;
2718         int len;
2719
2720         if (buf == NULL) {
2721                 /* this means the caller wants to use static buffer
2722                  * and it doesn't care about race. Usually this is
2723                  * in error reporting path */
2724                 buf = fsname_static;
2725                 buflen = sizeof(fsname_static);
2726         }
2727
2728         len = strlen(lsi->lsi_lmd->lmd_profile);
2729         ptr = strrchr(lsi->lsi_lmd->lmd_profile, '-');
2730         if (ptr && (strcmp(ptr, "-client") == 0))
2731                 len -= 7;
2732
2733         if (unlikely(len >= buflen))
2734                 len = buflen - 1;
2735         strncpy(buf, lsi->lsi_lmd->lmd_profile, len);
2736         buf[len] = '\0';
2737
2738         return buf;
2739 }
2740
2741 static char* ll_d_path(struct dentry *dentry, char *buf, int bufsize)
2742 {
2743         char *path = NULL;
2744
2745         struct path p;
2746
2747         p.dentry = dentry;
2748         p.mnt = current->fs->root.mnt;
2749         path_get(&p);
2750         path = d_path(&p, buf, bufsize);
2751         path_put(&p);
2752         return path;
2753 }
2754
2755 void ll_dirty_page_discard_warn(struct page *page, int ioret)
2756 {
2757         char *buf, *path = NULL;
2758         struct dentry *dentry = NULL;
2759         struct inode *inode = page->mapping->host;
2760
2761         /* this can be called inside spin lock so use GFP_ATOMIC. */
2762         buf = (char *)__get_free_page(GFP_ATOMIC);
2763         if (buf != NULL) {
2764                 dentry = d_find_alias(page->mapping->host);
2765                 if (dentry != NULL)
2766                         path = ll_d_path(dentry, buf, PAGE_SIZE);
2767         }
2768
2769         CDEBUG(D_WARNING,
2770                "%s: dirty page discard: %s/fid: "DFID"/%s may get corrupted "
2771                "(rc %d)\n", ll_get_fsname(page->mapping->host->i_sb, NULL, 0),
2772                s2lsi(page->mapping->host->i_sb)->lsi_lmd->lmd_dev,
2773                PFID(ll_inode2fid(inode)),
2774                (path && !IS_ERR(path)) ? path : "", ioret);
2775
2776         if (dentry != NULL)
2777                 dput(dentry);
2778
2779         if (buf != NULL)
2780                 free_page((unsigned long)buf);
2781 }
2782
2783 ssize_t ll_copy_user_md(const struct lov_user_md __user *md,
2784                         struct lov_user_md **kbuf)
2785 {
2786         struct lov_user_md      lum;
2787         ssize_t                 lum_size;
2788         ENTRY;
2789
2790         if (copy_from_user(&lum, md, sizeof(lum)))
2791                 RETURN(-EFAULT);
2792
2793         lum_size = ll_lov_user_md_size(&lum);
2794         if (lum_size < 0)
2795                 RETURN(lum_size);
2796
2797         OBD_ALLOC_LARGE(*kbuf, lum_size);
2798         if (*kbuf == NULL)
2799                 RETURN(-ENOMEM);
2800
2801         if (copy_from_user(*kbuf, md, lum_size) != 0) {
2802                 OBD_FREE_LARGE(*kbuf, lum_size);
2803                 RETURN(-EFAULT);
2804         }
2805
2806         RETURN(lum_size);
2807 }
2808
2809 /*
2810  * Compute llite root squash state after a change of root squash
2811  * configuration setting or add/remove of a lnet nid
2812  */
2813 void ll_compute_rootsquash_state(struct ll_sb_info *sbi)
2814 {
2815         struct root_squash_info *squash = &sbi->ll_squash;
2816         int i;
2817         bool matched;
2818         struct lnet_process_id id;
2819
2820         /* Update norootsquash flag */
2821         down_write(&squash->rsi_sem);
2822         if (list_empty(&squash->rsi_nosquash_nids))
2823                 sbi->ll_flags &= ~LL_SBI_NOROOTSQUASH;
2824         else {
2825                 /* Do not apply root squash as soon as one of our NIDs is
2826                  * in the nosquash_nids list */
2827                 matched = false;
2828                 i = 0;
2829                 while (LNetGetId(i++, &id) != -ENOENT) {
2830                         if (id.nid == LNET_NID_LO_0)
2831                                 continue;
2832                         if (cfs_match_nid(id.nid, &squash->rsi_nosquash_nids)) {
2833                                 matched = true;
2834                                 break;
2835                         }
2836                 }
2837                 if (matched)
2838                         sbi->ll_flags |= LL_SBI_NOROOTSQUASH;
2839                 else
2840                         sbi->ll_flags &= ~LL_SBI_NOROOTSQUASH;
2841         }
2842         up_write(&squash->rsi_sem);
2843 }
2844
2845 /**
2846  * Parse linkea content to extract information about a given hardlink
2847  *
2848  * \param[in]   ldata      - Initialized linkea data
2849  * \param[in]   linkno     - Link identifier
2850  * \param[out]  parent_fid - The entry's parent FID
2851  * \param[out]  ln         - Entry name destination buffer
2852  *
2853  * \retval 0 on success
2854  * \retval Appropriate negative error code on failure
2855  */
2856 static int ll_linkea_decode(struct linkea_data *ldata, unsigned int linkno,
2857                             struct lu_fid *parent_fid, struct lu_name *ln)
2858 {
2859         unsigned int    idx;
2860         int             rc;
2861         ENTRY;
2862
2863         rc = linkea_init_with_rec(ldata);
2864         if (rc < 0)
2865                 RETURN(rc);
2866
2867         if (linkno >= ldata->ld_leh->leh_reccount)
2868                 /* beyond last link */
2869                 RETURN(-ENODATA);
2870
2871         linkea_first_entry(ldata);
2872         for (idx = 0; ldata->ld_lee != NULL; idx++) {
2873                 linkea_entry_unpack(ldata->ld_lee, &ldata->ld_reclen, ln,
2874                                     parent_fid);
2875                 if (idx == linkno)
2876                         break;
2877
2878                 linkea_next_entry(ldata);
2879         }
2880
2881         if (idx < linkno)
2882                 RETURN(-ENODATA);
2883
2884         RETURN(0);
2885 }
2886
2887 /**
2888  * Get parent FID and name of an identified link. Operation is performed for
2889  * a given link number, letting the caller iterate over linkno to list one or
2890  * all links of an entry.
2891  *
2892  * \param[in]     file - File descriptor against which to perform the operation
2893  * \param[in,out] arg  - User-filled structure containing the linkno to operate
2894  *                       on and the available size. It is eventually filled with
2895  *                       the requested information or left untouched on error
2896  *
2897  * \retval - 0 on success
2898  * \retval - Appropriate negative error code on failure
2899  */
2900 int ll_getparent(struct file *file, struct getparent __user *arg)
2901 {
2902         struct inode            *inode = file_inode(file);
2903         struct linkea_data      *ldata;
2904         struct lu_buf            buf = LU_BUF_NULL;
2905         struct lu_name           ln;
2906         struct lu_fid            parent_fid;
2907         __u32                    linkno;
2908         __u32                    name_size;
2909         int                      rc;
2910
2911         ENTRY;
2912
2913         if (!cfs_capable(CFS_CAP_DAC_READ_SEARCH) &&
2914             !(ll_i2sbi(inode)->ll_flags & LL_SBI_USER_FID2PATH))
2915                 RETURN(-EPERM);
2916
2917         if (get_user(name_size, &arg->gp_name_size))
2918                 RETURN(-EFAULT);
2919
2920         if (get_user(linkno, &arg->gp_linkno))
2921                 RETURN(-EFAULT);
2922
2923         if (name_size > PATH_MAX)
2924                 RETURN(-EINVAL);
2925
2926         OBD_ALLOC(ldata, sizeof(*ldata));
2927         if (ldata == NULL)
2928                 RETURN(-ENOMEM);
2929
2930         rc = linkea_data_new(ldata, &buf);
2931         if (rc < 0)
2932                 GOTO(ldata_free, rc);
2933
2934 #ifdef HAVE_XATTR_HANDLER_FLAGS
2935         rc = ll_xattr_list(inode, XATTR_NAME_LINK, XATTR_TRUSTED_T, buf.lb_buf,
2936                            buf.lb_len, OBD_MD_FLXATTR);
2937 #else
2938         rc = ll_getxattr(file_dentry(file), XATTR_NAME_LINK, buf.lb_buf,
2939                          buf.lb_len);
2940 #endif /* HAVE_XATTR_HANDLER_FLAGS */
2941         if (rc < 0)
2942                 GOTO(lb_free, rc);
2943
2944         rc = ll_linkea_decode(ldata, linkno, &parent_fid, &ln);
2945         if (rc < 0)
2946                 GOTO(lb_free, rc);
2947
2948         if (ln.ln_namelen >= name_size)
2949                 GOTO(lb_free, rc = -EOVERFLOW);
2950
2951         if (copy_to_user(&arg->gp_fid, &parent_fid, sizeof(arg->gp_fid)))
2952                 GOTO(lb_free, rc = -EFAULT);
2953
2954         if (copy_to_user(&arg->gp_name, ln.ln_name, ln.ln_namelen))
2955                 GOTO(lb_free, rc = -EFAULT);
2956
2957         if (put_user('\0', arg->gp_name + ln.ln_namelen))
2958                 GOTO(lb_free, rc = -EFAULT);
2959
2960 lb_free:
2961         lu_buf_free(&buf);
2962 ldata_free:
2963         OBD_FREE(ldata, sizeof(*ldata));
2964
2965         RETURN(rc);
2966 }