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, 2016, 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 #include <lustre_ha.h>
  53 #include <lustre_dlm.h>
  54 #include <lprocfs_status.h>
  55 #include <lustre_disk.h>
  56 #include <uapi/linux/lustre/lustre_param.h>
  57 #include <lustre_log.h>
  58 #include <cl_object.h>
  59 #include <obd_cksum.h>
  60 #include "llite_internal.h"
  61
  62 struct kmem_cache *ll_file_data_slab;
  63
  64 #ifndef log2
  65 #define log2(n) ffz(~(n))
  66 #endif
  67
  68 static struct ll_sb_info *ll_init_sbi(void)
  69 {
  70         struct ll_sb_info *sbi = NULL;
  71         unsigned long pages;
  72         unsigned long lru_page_max;
  73         struct sysinfo si;
  74         class_uuid_t uuid;
  75         int i;
  76         ENTRY;
  77
  78         OBD_ALLOC_PTR(sbi);
  79         if (sbi == NULL)
  80                 RETURN(NULL);
  81
  82         spin_lock_init(&sbi->ll_lock);
  83         mutex_init(&sbi->ll_lco.lco_lock);
  84         spin_lock_init(&sbi->ll_pp_extent_lock);
  85         spin_lock_init(&sbi->ll_process_lock);
  86         sbi->ll_rw_stats_on = 0;
  87
  88         si_meminfo(&si);
  89         pages = si.totalram - si.totalhigh;
  90         lru_page_max = pages / 2;
  91
  92         /* initialize ll_cache data */
  93         sbi->ll_cache = cl_cache_init(lru_page_max);
  94         if (sbi->ll_cache == NULL) {
  95                 OBD_FREE(sbi, sizeof(*sbi));
  96                 RETURN(NULL);
  97         }
  98
  99         sbi->ll_ra_info.ra_max_pages_per_file = min(pages / 32,
 100                                            SBI_DEFAULT_READAHEAD_MAX);
 101         sbi->ll_ra_info.ra_max_pages = sbi->ll_ra_info.ra_max_pages_per_file;
 102         sbi->ll_ra_info.ra_max_read_ahead_whole_pages =
 103                                            SBI_DEFAULT_READAHEAD_WHOLE_MAX;
 104
 105         ll_generate_random_uuid(uuid);
 106         class_uuid_unparse(uuid, &sbi->ll_sb_uuid);
 107         CDEBUG(D_CONFIG, "generated uuid: %s\n", sbi->ll_sb_uuid.uuid);
 108
 109         sbi->ll_flags |= LL_SBI_VERBOSE;
 110 #ifdef ENABLE_CHECKSUM
 111         sbi->ll_flags |= LL_SBI_CHECKSUM;
 112 #endif
 113
 114 #ifdef HAVE_LRU_RESIZE_SUPPORT
 115         sbi->ll_flags |= LL_SBI_LRU_RESIZE;
 116 #endif
 117         sbi->ll_flags |= LL_SBI_LAZYSTATFS;
 118
 119         for (i = 0; i <= LL_PROCESS_HIST_MAX; i++) {
 120                 spin_lock_init(&sbi->ll_rw_extents_info.pp_extents[i].
 121                                pp_r_hist.oh_lock);
 122                 spin_lock_init(&sbi->ll_rw_extents_info.pp_extents[i].
 123                                pp_w_hist.oh_lock);
 124         }
 125
 126         /* metadata statahead is enabled by default */
 127         sbi->ll_sa_max = LL_SA_RPC_DEF;
 128         atomic_set(&sbi->ll_sa_total, 0);
 129         atomic_set(&sbi->ll_sa_wrong, 0);
 130         atomic_set(&sbi->ll_sa_running, 0);
 131         atomic_set(&sbi->ll_agl_total, 0);
 132         sbi->ll_flags |= LL_SBI_AGL_ENABLED;
 133         sbi->ll_flags |= LL_SBI_FAST_READ;
 134
 135         /* root squash */
 136         sbi->ll_squash.rsi_uid = 0;
 137         sbi->ll_squash.rsi_gid = 0;
 138         INIT_LIST_HEAD(&sbi->ll_squash.rsi_nosquash_nids);
 139         init_rwsem(&sbi->ll_squash.rsi_sem);
 140
 141         RETURN(sbi);
 142 }
 143
 144 static void ll_free_sbi(struct super_block *sb)
 145 {
 146         struct ll_sb_info *sbi = ll_s2sbi(sb);
 147         ENTRY;
 148
 149         if (sbi != NULL) {
 150                 if (!list_empty(&sbi->ll_squash.rsi_nosquash_nids))
 151                         cfs_free_nidlist(&sbi->ll_squash.rsi_nosquash_nids);
 152                 if (sbi->ll_cache != NULL) {
 153                         cl_cache_decref(sbi->ll_cache);
 154                         sbi->ll_cache = NULL;
 155                 }
 156                 OBD_FREE(sbi, sizeof(*sbi));
 157         }
 158         EXIT;
 159 }
 160
 161 static inline int obd_connect_has_secctx(struct obd_connect_data *data)
 162 {
 163         return data->ocd_connect_flags & OBD_CONNECT_FLAGS2 &&
 164                data->ocd_connect_flags2 & OBD_CONNECT2_FILE_SECCTX;
 165 }
 166
 167 static int client_common_fill_super(struct super_block *sb, char *md, char *dt,
 168                                     struct vfsmount *mnt)
 169 {
 170         struct inode *root = NULL;
 171         struct ll_sb_info *sbi = ll_s2sbi(sb);
 172         struct obd_device *obd;
 173         struct obd_statfs *osfs = NULL;
 174         struct ptlrpc_request *request = NULL;
 175         struct obd_connect_data *data = NULL;
 176         struct obd_uuid *uuid;
 177         struct md_op_data *op_data;
 178         struct lustre_md lmd;
 179         u64 valid;
 180         int size, err, checksum;
 181         ENTRY;
 182
 183         obd = class_name2obd(md);
 184         if (!obd) {
 185                 CERROR("MD %s: not setup or attached\n", md);
 186                 RETURN(-EINVAL);
 187         }
 188
 189         OBD_ALLOC_PTR(data);
 190         if (data == NULL)
 191                 RETURN(-ENOMEM);
 192
 193         OBD_ALLOC_PTR(osfs);
 194         if (osfs == NULL) {
 195                 OBD_FREE_PTR(data);
 196                 RETURN(-ENOMEM);
 197         }
 198
 199         /* indicate MDT features supported by this client */
 200         data->ocd_connect_flags = OBD_CONNECT_IBITS    | OBD_CONNECT_NODEVOH  |
 201                                   OBD_CONNECT_ATTRFID  |
 202                                   OBD_CONNECT_VERSION  | OBD_CONNECT_BRW_SIZE |
 203                                   OBD_CONNECT_MDS_CAPA | OBD_CONNECT_OSS_CAPA |
 204                                   OBD_CONNECT_CANCELSET | OBD_CONNECT_FID     |
 205                                   OBD_CONNECT_AT       | OBD_CONNECT_LOV_V3   |
 206                                   OBD_CONNECT_VBR | OBD_CONNECT_FULL20 |
 207                                   OBD_CONNECT_64BITHASH |
 208                                   OBD_CONNECT_EINPROGRESS |
 209                                   OBD_CONNECT_JOBSTATS | OBD_CONNECT_LVB_TYPE |
 210                                   OBD_CONNECT_LAYOUTLOCK | OBD_CONNECT_PINGLESS|
 211                                   OBD_CONNECT_MAX_EASIZE |
 212                                   OBD_CONNECT_FLOCK_DEAD |
 213                                   OBD_CONNECT_DISP_STRIPE | OBD_CONNECT_LFSCK |
 214                                   OBD_CONNECT_OPEN_BY_FID |
 215                                   OBD_CONNECT_DIR_STRIPE |
 216                                   OBD_CONNECT_BULK_MBITS |
 217                                   OBD_CONNECT_SUBTREE |
 218                                   OBD_CONNECT_FLAGS2 | OBD_CONNECT_MULTIMODRPCS;
 219
 220         data->ocd_connect_flags2 = 0;
 221
 222 #ifdef HAVE_LRU_RESIZE_SUPPORT
 223         if (sbi->ll_flags & LL_SBI_LRU_RESIZE)
 224                 data->ocd_connect_flags |= OBD_CONNECT_LRU_RESIZE;
 225 #endif
 226 #ifdef CONFIG_FS_POSIX_ACL
 227         data->ocd_connect_flags |= OBD_CONNECT_ACL | OBD_CONNECT_UMASK |
 228                                    OBD_CONNECT_LARGE_ACL;
 229 #endif
 230
 231         if (OBD_FAIL_CHECK(OBD_FAIL_MDC_LIGHTWEIGHT))
 232                 /* flag mdc connection as lightweight, only used for test
 233                  * purpose, use with care */
 234                 data->ocd_connect_flags |= OBD_CONNECT_LIGHTWEIGHT;
 235
 236         data->ocd_ibits_known = MDS_INODELOCK_FULL;
 237         data->ocd_version = LUSTRE_VERSION_CODE;
 238
 239         if (sb->s_flags & MS_RDONLY)
 240                 data->ocd_connect_flags |= OBD_CONNECT_RDONLY;
 241         if (sbi->ll_flags & LL_SBI_USER_XATTR)
 242                 data->ocd_connect_flags |= OBD_CONNECT_XATTR;
 243
 244 #ifdef MS_NOSEC
 245         /* Setting this indicates we correctly support S_NOSEC (See kernel
 246          * commit 9e1f1de02c2275d7172e18dc4e7c2065777611bf)
 247          */
 248         sb->s_flags |= MS_NOSEC;
 249 #endif
 250
 251         if (sbi->ll_flags & LL_SBI_FLOCK)
 252                 sbi->ll_fop = &ll_file_operations_flock;
 253         else if (sbi->ll_flags & LL_SBI_LOCALFLOCK)
 254                 sbi->ll_fop = &ll_file_operations;
 255         else
 256                 sbi->ll_fop = &ll_file_operations_noflock;
 257
 258         /* always ping even if server suppress_pings */
 259         if (sbi->ll_flags & LL_SBI_ALWAYS_PING)
 260                 data->ocd_connect_flags &= ~OBD_CONNECT_PINGLESS;
 261
 262 #ifdef HAVE_SECURITY_DENTRY_INIT_SECURITY
 263         data->ocd_connect_flags2 |= OBD_CONNECT2_FILE_SECCTX;
 264 #endif /* HAVE_SECURITY_DENTRY_INIT_SECURITY */
 265
 266         data->ocd_brw_size = MD_MAX_BRW_SIZE;
 267
 268         err = obd_connect(NULL, &sbi->ll_md_exp, obd, &sbi->ll_sb_uuid, data, NULL);
 269         if (err == -EBUSY) {
 270                 LCONSOLE_ERROR_MSG(0x14f, "An MDT (md %s) is performing "
 271                                    "recovery, of which this client is not a "
 272                                    "part. Please wait for recovery to complete,"
 273                                    " abort, or time out.\n", md);
 274                 GOTO(out, err);
 275         } else if (err) {
 276                 CERROR("cannot connect to %s: rc = %d\n", md, err);
 277                 GOTO(out, err);
 278         }
 279
 280         sbi->ll_md_exp->exp_connect_data = *data;
 281
 282         err = obd_fid_init(sbi->ll_md_exp->exp_obd, sbi->ll_md_exp,
 283                            LUSTRE_SEQ_METADATA);
 284         if (err) {
 285                 CERROR("%s: Can't init metadata layer FID infrastructure, "
 286                        "rc = %d\n", sbi->ll_md_exp->exp_obd->obd_name, err);
 287                 GOTO(out_md, err);
 288         }
 289
 290         /* For mount, we only need fs info from MDT0, and also in DNE, it
 291          * can make sure the client can be mounted as long as MDT0 is
 292          * avaible */
 293         err = obd_statfs(NULL, sbi->ll_md_exp, osfs,
 294                         cfs_time_shift_64(-OBD_STATFS_CACHE_SECONDS),
 295                         OBD_STATFS_FOR_MDT0);
 296         if (err)
 297                 GOTO(out_md_fid, err);
 298
 299         /* This needs to be after statfs to ensure connect has finished.
 300          * Note that "data" does NOT contain the valid connect reply.
 301          * If connecting to a 1.8 server there will be no LMV device, so
 302          * we can access the MDC export directly and exp_connect_flags will
 303          * be non-zero, but if accessing an upgraded 2.1 server it will
 304          * have the correct flags filled in.
 305          * XXX: fill in the LMV exp_connect_flags from MDC(s). */
 306         valid = exp_connect_flags(sbi->ll_md_exp) & CLIENT_CONNECT_MDT_REQD;
 307         if (exp_connect_flags(sbi->ll_md_exp) != 0 &&
 308             valid != CLIENT_CONNECT_MDT_REQD) {
 309                 char *buf;
 310
 311                 OBD_ALLOC_WAIT(buf, PAGE_SIZE);
 312                 obd_connect_flags2str(buf, PAGE_SIZE,
 313                                       valid ^ CLIENT_CONNECT_MDT_REQD, 0, ",");
 314                 LCONSOLE_ERROR_MSG(0x170, "Server %s does not support "
 315                                    "feature(s) needed for correct operation "
 316                                    "of this client (%s). Please upgrade "
 317                                    "server or downgrade client.\n",
 318                                    sbi->ll_md_exp->exp_obd->obd_name, buf);
 319                 OBD_FREE(buf, PAGE_SIZE);
 320                 GOTO(out_md_fid, err = -EPROTO);
 321         }
 322
 323         size = sizeof(*data);
 324         err = obd_get_info(NULL, sbi->ll_md_exp, sizeof(KEY_CONN_DATA),
 325                            KEY_CONN_DATA,  &size, data);
 326         if (err) {
 327                 CERROR("%s: Get connect data failed: rc = %d\n",
 328                        sbi->ll_md_exp->exp_obd->obd_name, err);
 329                 GOTO(out_md_fid, err);
 330         }
 331
 332         LASSERT(osfs->os_bsize);
 333         sb->s_blocksize = osfs->os_bsize;
 334         sb->s_blocksize_bits = log2(osfs->os_bsize);
 335         sb->s_magic = LL_SUPER_MAGIC;
 336         sb->s_maxbytes = MAX_LFS_FILESIZE;
 337         sbi->ll_namelen = osfs->os_namelen;
 338         sbi->ll_mnt.mnt = current->fs->root.mnt;
 339
 340         if ((sbi->ll_flags & LL_SBI_USER_XATTR) &&
 341             !(data->ocd_connect_flags & OBD_CONNECT_XATTR)) {
 342                 LCONSOLE_INFO("Disabling user_xattr feature because "
 343                               "it is not supported on the server\n");
 344                 sbi->ll_flags &= ~LL_SBI_USER_XATTR;
 345         }
 346
 347         if (data->ocd_connect_flags & OBD_CONNECT_ACL) {
 348 #ifdef MS_POSIXACL
 349                 sb->s_flags |= MS_POSIXACL;
 350 #endif
 351                 sbi->ll_flags |= LL_SBI_ACL;
 352         } else {
 353                 LCONSOLE_INFO("client wants to enable acl, but mdt not!\n");
 354 #ifdef MS_POSIXACL
 355                 sb->s_flags &= ~MS_POSIXACL;
 356 #endif
 357                 sbi->ll_flags &= ~LL_SBI_ACL;
 358         }
 359
 360         if (data->ocd_connect_flags & OBD_CONNECT_64BITHASH)
 361                 sbi->ll_flags |= LL_SBI_64BIT_HASH;
 362
 363         if (data->ocd_connect_flags & OBD_CONNECT_LAYOUTLOCK)
 364                 sbi->ll_flags |= LL_SBI_LAYOUT_LOCK;
 365
 366         if (obd_connect_has_secctx(data))
 367                 sbi->ll_flags |= LL_SBI_FILE_SECCTX;
 368
 369         if (data->ocd_ibits_known & MDS_INODELOCK_XATTR) {
 370                 if (!(data->ocd_connect_flags & OBD_CONNECT_MAX_EASIZE)) {
 371                         LCONSOLE_INFO("%s: disabling xattr cache due to "
 372                                       "unknown maximum xattr size.\n", dt);
 373                 } else if (!sbi->ll_xattr_cache_set) {
 374                         /* If xattr_cache is already set (no matter 0 or 1)
 375                          * during processing llog, it won't be enabled here. */
 376                         sbi->ll_flags |= LL_SBI_XATTR_CACHE;
 377                         sbi->ll_xattr_cache_enabled = 1;
 378                 }
 379         }
 380
 381         obd = class_name2obd(dt);
 382         if (!obd) {
 383                 CERROR("DT %s: not setup or attached\n", dt);
 384                 GOTO(out_md_fid, err = -ENODEV);
 385         }
 386
 387         /* pass client page size via ocd_grant_blkbits, the server should report
 388          * back its backend blocksize for grant calculation purpose */
 389         data->ocd_grant_blkbits = PAGE_SHIFT;
 390
 391         /* indicate OST features supported by this client */
 392         data->ocd_connect_flags = OBD_CONNECT_GRANT | OBD_CONNECT_VERSION |
 393                                   OBD_CONNECT_REQPORTAL | OBD_CONNECT_BRW_SIZE |
 394                                   OBD_CONNECT_CANCELSET | OBD_CONNECT_FID |
 395                                   OBD_CONNECT_SRVLOCK | OBD_CONNECT_TRUNCLOCK|
 396                                   OBD_CONNECT_AT | OBD_CONNECT_OSS_CAPA |
 397                                   OBD_CONNECT_VBR | OBD_CONNECT_FULL20 |
 398                                   OBD_CONNECT_64BITHASH | OBD_CONNECT_MAXBYTES |
 399                                   OBD_CONNECT_EINPROGRESS |
 400                                   OBD_CONNECT_JOBSTATS | OBD_CONNECT_LVB_TYPE |
 401                                   OBD_CONNECT_LAYOUTLOCK |
 402                                   OBD_CONNECT_PINGLESS | OBD_CONNECT_LFSCK |
 403                                   OBD_CONNECT_BULK_MBITS |
 404                                   OBD_CONNECT_FLAGS2;
 405
 406 /* The client currently advertises support for OBD_CONNECT_LOCKAHEAD_OLD so it
 407  * can interoperate with an older version of lockahead which was released prior
 408  * to landing in master. This support will be dropped when 2.13 development
 409  * starts.  At the point, we should not just drop the connect flag (below), we
 410  * should also remove the support in the code.
 411  *
 412  * Removing it means a few things:
 413  * 1. Remove this section here
 414  * 2. Remove CEF_NONBLOCK in ll_file_lockahead()
 415  * 3. Remove function exp_connect_lockahead_old
 416  * 4. Remove LDLM_FL_LOCKAHEAD_OLD_RESERVED in lustre_dlm_flags.h
 417  * */
 418 #if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(2, 12, 50, 0)
 419         data->ocd_connect_flags |= OBD_CONNECT_LOCKAHEAD_OLD;
 420 #endif
 421
 422         data->ocd_connect_flags2 = OBD_CONNECT2_LOCKAHEAD;
 423
 424         if (!OBD_FAIL_CHECK(OBD_FAIL_OSC_CONNECT_GRANT_PARAM))
 425                 data->ocd_connect_flags |= OBD_CONNECT_GRANT_PARAM;
 426
 427         /* OBD_CONNECT_CKSUM should always be set, even if checksums are
 428          * disabled by default, because it can still be enabled on the
 429          * fly via /proc. As a consequence, we still need to come to an
 430          * agreement on the supported algorithms at connect time */
 431         data->ocd_connect_flags |= OBD_CONNECT_CKSUM;
 432
 433         if (OBD_FAIL_CHECK(OBD_FAIL_OSC_CKSUM_ADLER_ONLY))
 434                 data->ocd_cksum_types = OBD_CKSUM_ADLER;
 435         else
 436                 data->ocd_cksum_types = cksum_types_supported_client();
 437
 438 #ifdef HAVE_LRU_RESIZE_SUPPORT
 439         data->ocd_connect_flags |= OBD_CONNECT_LRU_RESIZE;
 440 #endif
 441         /* always ping even if server suppress_pings */
 442         if (sbi->ll_flags & LL_SBI_ALWAYS_PING)
 443                 data->ocd_connect_flags &= ~OBD_CONNECT_PINGLESS;
 444
 445         CDEBUG(D_RPCTRACE, "ocd_connect_flags: %#llx ocd_version: %d "
 446                "ocd_grant: %d\n", data->ocd_connect_flags,
 447                data->ocd_version, data->ocd_grant);
 448
 449         obd->obd_upcall.onu_owner = &sbi->ll_lco;
 450         obd->obd_upcall.onu_upcall = cl_ocd_update;
 451
 452         data->ocd_brw_size = DT_MAX_BRW_SIZE;
 453
 454         err = obd_connect(NULL, &sbi->ll_dt_exp, obd, &sbi->ll_sb_uuid, data,
 455                           NULL);
 456         if (err == -EBUSY) {
 457                 LCONSOLE_ERROR_MSG(0x150, "An OST (dt %s) is performing "
 458                                    "recovery, of which this client is not a "
 459                                    "part.  Please wait for recovery to "
 460                                    "complete, abort, or time out.\n", dt);
 461                 GOTO(out_md, err);
 462         } else if (err) {
 463                 CERROR("%s: Cannot connect to %s: rc = %d\n",
 464                        sbi->ll_dt_exp->exp_obd->obd_name, dt, err);
 465                 GOTO(out_md, err);
 466         }
 467
 468         sbi->ll_dt_exp->exp_connect_data = *data;
 469
 470         err = obd_fid_init(sbi->ll_dt_exp->exp_obd, sbi->ll_dt_exp,
 471                            LUSTRE_SEQ_METADATA);
 472         if (err) {
 473                 CERROR("%s: Can't init data layer FID infrastructure, "
 474                        "rc = %d\n", sbi->ll_dt_exp->exp_obd->obd_name, err);
 475                 GOTO(out_dt, err);
 476         }
 477
 478         mutex_lock(&sbi->ll_lco.lco_lock);
 479         sbi->ll_lco.lco_flags = data->ocd_connect_flags;
 480         sbi->ll_lco.lco_md_exp = sbi->ll_md_exp;
 481         sbi->ll_lco.lco_dt_exp = sbi->ll_dt_exp;
 482         mutex_unlock(&sbi->ll_lco.lco_lock);
 483
 484         fid_zero(&sbi->ll_root_fid);
 485         err = md_get_root(sbi->ll_md_exp, get_mount_fileset(sb),
 486                            &sbi->ll_root_fid);
 487         if (err) {
 488                 CERROR("cannot mds_connect: rc = %d\n", err);
 489                 GOTO(out_lock_cn_cb, err);
 490         }
 491         if (!fid_is_sane(&sbi->ll_root_fid)) {
 492                 CERROR("%s: Invalid root fid "DFID" during mount\n",
 493                        sbi->ll_md_exp->exp_obd->obd_name,
 494                        PFID(&sbi->ll_root_fid));
 495                 GOTO(out_lock_cn_cb, err = -EINVAL);
 496         }
 497         CDEBUG(D_SUPER, "rootfid "DFID"\n", PFID(&sbi->ll_root_fid));
 498
 499         sb->s_op = &lustre_super_operations;
 500 #ifdef HAVE_XATTR_HANDLER_FLAGS
 501         sb->s_xattr = ll_xattr_handlers;
 502 #endif
 503 #if THREAD_SIZE >= 8192 /*b=17630*/
 504         sb->s_export_op = &lustre_export_operations;
 505 #endif
 506
 507         /* make root inode
 508          * XXX: move this to after cbd setup? */
 509         valid = OBD_MD_FLGETATTR | OBD_MD_FLBLOCKS | OBD_MD_FLMODEASIZE;
 510         if (sbi->ll_flags & LL_SBI_ACL)
 511                 valid |= OBD_MD_FLACL;
 512
 513         OBD_ALLOC_PTR(op_data);
 514         if (op_data == NULL)
 515                 GOTO(out_lock_cn_cb, err = -ENOMEM);
 516
 517         op_data->op_fid1 = sbi->ll_root_fid;
 518         op_data->op_mode = 0;
 519         op_data->op_valid = valid;
 520
 521         err = md_getattr(sbi->ll_md_exp, op_data, &request);
 522
 523         OBD_FREE_PTR(op_data);
 524         if (err) {
 525                 CERROR("%s: md_getattr failed for root: rc = %d\n",
 526                        sbi->ll_md_exp->exp_obd->obd_name, err);
 527                 GOTO(out_lock_cn_cb, err);
 528         }
 529
 530         err = md_get_lustre_md(sbi->ll_md_exp, request, sbi->ll_dt_exp,
 531                                sbi->ll_md_exp, &lmd);
 532         if (err) {
 533                 CERROR("failed to understand root inode md: rc = %d\n", err);
 534                 ptlrpc_req_finished(request);
 535                 GOTO(out_lock_cn_cb, err);
 536         }
 537
 538         LASSERT(fid_is_sane(&sbi->ll_root_fid));
 539         root = ll_iget(sb, cl_fid_build_ino(&sbi->ll_root_fid,
 540                                             sbi->ll_flags & LL_SBI_32BIT_API),
 541                        &lmd);
 542         md_free_lustre_md(sbi->ll_md_exp, &lmd);
 543         ptlrpc_req_finished(request);
 544
 545         if (IS_ERR(root)) {
 546 #ifdef CONFIG_FS_POSIX_ACL
 547                 if (lmd.posix_acl) {
 548                         posix_acl_release(lmd.posix_acl);
 549                         lmd.posix_acl = NULL;
 550                 }
 551 #endif
 552                 err = IS_ERR(root) ? PTR_ERR(root) : -EBADF;
 553                 root = NULL;
 554                 CERROR("lustre_lite: bad iget4 for root\n");
 555                 GOTO(out_root, err);
 556         }
 557
 558         checksum = sbi->ll_flags & LL_SBI_CHECKSUM;
 559         err = obd_set_info_async(NULL, sbi->ll_dt_exp, sizeof(KEY_CHECKSUM),
 560                                  KEY_CHECKSUM, sizeof(checksum), &checksum,
 561                                  NULL);
 562         if (err) {
 563                 CERROR("%s: Set checksum failed: rc = %d\n",
 564                        sbi->ll_dt_exp->exp_obd->obd_name, err);
 565                 GOTO(out_root, err);
 566         }
 567         cl_sb_init(sb);
 568
 569         err = obd_set_info_async(NULL, sbi->ll_dt_exp, sizeof(KEY_CACHE_SET),
 570                                  KEY_CACHE_SET, sizeof(*sbi->ll_cache),
 571                                  sbi->ll_cache, NULL);
 572         if (err) {
 573                 CERROR("%s: Set cache_set failed: rc = %d\n",
 574                        sbi->ll_dt_exp->exp_obd->obd_name, err);
 575                 GOTO(out_root, err);
 576         }
 577
 578         sb->s_root = d_make_root(root);
 579         if (sb->s_root == NULL) {
 580                 CERROR("%s: can't make root dentry\n",
 581                         ll_get_fsname(sb, NULL, 0));
 582                 GOTO(out_root, err = -ENOMEM);
 583         }
 584 #ifdef HAVE_DCACHE_LOCK
 585         sb->s_root->d_op = &ll_d_ops;
 586 #endif
 587
 588         sbi->ll_sdev_orig = sb->s_dev;
 589
 590         /* We set sb->s_dev equal on all lustre clients in order to support
 591          * NFS export clustering.  NFSD requires that the FSID be the same
 592          * on all clients. */
 593         /* s_dev is also used in lt_compare() to compare two fs, but that is
 594          * only a node-local comparison. */
 595         uuid = obd_get_uuid(sbi->ll_md_exp);
 596         if (uuid != NULL)
 597                 sb->s_dev = get_uuid2int(uuid->uuid, strlen(uuid->uuid));
 598
 599         if (data != NULL)
 600                 OBD_FREE_PTR(data);
 601         if (osfs != NULL)
 602                 OBD_FREE_PTR(osfs);
 603
 604         if (sbi->ll_proc_root != NULL) {
 605                 err = lprocfs_ll_register_obd(sb, dt);
 606                 if (err < 0) {
 607                         CERROR("%s: could not register %s in llite: rc = %d\n",
 608                                dt, ll_get_fsname(sb, NULL, 0), err);
 609                         err = 0;
 610                 }
 611                 err = lprocfs_ll_register_obd(sb, md);
 612                 if (err < 0) {
 613                         CERROR("%s: could not register %s in llite: rc = %d\n",
 614                                md, ll_get_fsname(sb, NULL, 0), err);
 615                         err = 0;
 616                 }
 617         }
 618
 619         RETURN(err);
 620 out_root:
 621         if (root)
 622                 iput(root);
 623 out_lock_cn_cb:
 624         obd_fid_fini(sbi->ll_dt_exp->exp_obd);
 625 out_dt:
 626         obd_disconnect(sbi->ll_dt_exp);
 627         sbi->ll_dt_exp = NULL;
 628 out_md_fid:
 629         obd_fid_fini(sbi->ll_md_exp->exp_obd);
 630 out_md:
 631         obd_disconnect(sbi->ll_md_exp);
 632         sbi->ll_md_exp = NULL;
 633 out:
 634         if (data != NULL)
 635                 OBD_FREE_PTR(data);
 636         if (osfs != NULL)
 637                 OBD_FREE_PTR(osfs);
 638         return err;
 639 }
 640
 641 int ll_get_max_mdsize(struct ll_sb_info *sbi, int *lmmsize)
 642 {
 643         int size, rc;
 644
 645         size = sizeof(*lmmsize);
 646         rc = obd_get_info(NULL, sbi->ll_dt_exp, sizeof(KEY_MAX_EASIZE),
 647                           KEY_MAX_EASIZE, &size, lmmsize);
 648         if (rc != 0) {
 649                 CERROR("%s: cannot get max LOV EA size: rc = %d\n",
 650                        sbi->ll_dt_exp->exp_obd->obd_name, rc);
 651                 RETURN(rc);
 652         }
 653
 654         size = sizeof(int);
 655         rc = obd_get_info(NULL, sbi->ll_md_exp, sizeof(KEY_MAX_EASIZE),
 656                           KEY_MAX_EASIZE, &size, lmmsize);
 657         if (rc)
 658                 CERROR("Get max mdsize error rc %d\n", rc);
 659
 660         RETURN(rc);
 661 }
 662
 663 /**
 664  * Get the value of the default_easize parameter.
 665  *
 666  * \see client_obd::cl_default_mds_easize
 667  *
 668  * \param[in] sbi       superblock info for this filesystem
 669  * \param[out] lmmsize  pointer to storage location for value
 670  *
 671  * \retval 0            on success
 672  * \retval negative     negated errno on failure
 673  */
 674 int ll_get_default_mdsize(struct ll_sb_info *sbi, int *lmmsize)
 675 {
 676         int size, rc;
 677
 678         size = sizeof(int);
 679         rc = obd_get_info(NULL, sbi->ll_md_exp, sizeof(KEY_DEFAULT_EASIZE),
 680                          KEY_DEFAULT_EASIZE, &size, lmmsize);
 681         if (rc)
 682                 CERROR("Get default mdsize error rc %d\n", rc);
 683
 684         RETURN(rc);
 685 }
 686
 687 /**
 688  * Set the default_easize parameter to the given value.
 689  *
 690  * \see client_obd::cl_default_mds_easize
 691  *
 692  * \param[in] sbi       superblock info for this filesystem
 693  * \param[in] lmmsize   the size to set
 694  *
 695  * \retval 0            on success
 696  * \retval negative     negated errno on failure
 697  */
 698 int ll_set_default_mdsize(struct ll_sb_info *sbi, int lmmsize)
 699 {
 700         int rc;
 701
 702         if (lmmsize < sizeof(struct lov_mds_md) ||
 703             lmmsize > OBD_MAX_DEFAULT_EA_SIZE)
 704                 return -EINVAL;
 705
 706         rc = obd_set_info_async(NULL, sbi->ll_md_exp,
 707                                 sizeof(KEY_DEFAULT_EASIZE), KEY_DEFAULT_EASIZE,
 708                                 sizeof(int), &lmmsize, NULL);
 709
 710         RETURN(rc);
 711 }
 712
 713 static void client_common_put_super(struct super_block *sb)
 714 {
 715         struct ll_sb_info *sbi = ll_s2sbi(sb);
 716         ENTRY;
 717
 718         cl_sb_fini(sb);
 719
 720         obd_fid_fini(sbi->ll_dt_exp->exp_obd);
 721         obd_disconnect(sbi->ll_dt_exp);
 722         sbi->ll_dt_exp = NULL;
 723
 724         lprocfs_ll_unregister_mountpoint(sbi);
 725
 726         obd_fid_fini(sbi->ll_md_exp->exp_obd);
 727         obd_disconnect(sbi->ll_md_exp);
 728         sbi->ll_md_exp = NULL;
 729
 730         EXIT;
 731 }
 732
 733 void ll_kill_super(struct super_block *sb)
 734 {
 735         struct ll_sb_info *sbi;
 736         ENTRY;
 737
 738         /* not init sb ?*/
 739         if (!(sb->s_flags & MS_ACTIVE))
 740                 return;
 741
 742         sbi = ll_s2sbi(sb);
 743         /* we need restore s_dev from changed for clustred NFS before put_super
 744          * because new kernels have cached s_dev and change sb->s_dev in
 745          * put_super not affected real removing devices */
 746         if (sbi) {
 747                 sb->s_dev = sbi->ll_sdev_orig;
 748                 sbi->ll_umounting = 1;
 749
 750                 /* wait running statahead threads to quit */
 751                 while (atomic_read(&sbi->ll_sa_running) > 0) {
 752                         set_current_state(TASK_UNINTERRUPTIBLE);
 753                         schedule_timeout(msecs_to_jiffies(MSEC_PER_SEC >> 3));
 754                 }
 755         }
 756
 757         EXIT;
 758 }
 759
 760 static inline int ll_set_opt(const char *opt, char *data, int fl)
 761 {
 762         if (strncmp(opt, data, strlen(opt)) != 0)
 763                 return(0);
 764         else
 765                 return(fl);
 766 }
 767
 768 /* non-client-specific mount options are parsed in lmd_parse */
 769 static int ll_options(char *options, int *flags)
 770 {
 771         int tmp;
 772         char *s1 = options, *s2;
 773         ENTRY;
 774
 775         if (!options)
 776                 RETURN(0);
 777
 778         CDEBUG(D_CONFIG, "Parsing opts %s\n", options);
 779
 780         while (*s1) {
 781                 CDEBUG(D_SUPER, "next opt=%s\n", s1);
 782                 tmp = ll_set_opt("nolock", s1, LL_SBI_NOLCK);
 783                 if (tmp) {
 784                         *flags |= tmp;
 785                         goto next;
 786                 }
 787                 tmp = ll_set_opt("flock", s1, LL_SBI_FLOCK);
 788                 if (tmp) {
 789                         *flags |= tmp;
 790                         goto next;
 791                 }
 792                 tmp = ll_set_opt("localflock", s1, LL_SBI_LOCALFLOCK);
 793                 if (tmp) {
 794                         *flags |= tmp;
 795                         goto next;
 796                 }
 797                 tmp = ll_set_opt("noflock", s1, LL_SBI_FLOCK|LL_SBI_LOCALFLOCK);
 798                 if (tmp) {
 799                         *flags &= ~tmp;
 800                         goto next;
 801                 }
 802                 tmp = ll_set_opt("user_xattr", s1, LL_SBI_USER_XATTR);
 803                 if (tmp) {
 804                         *flags |= tmp;
 805                         goto next;
 806                 }
 807                 tmp = ll_set_opt("nouser_xattr", s1, LL_SBI_USER_XATTR);
 808                 if (tmp) {
 809                         *flags &= ~tmp;
 810                         goto next;
 811                 }
 812                 tmp = ll_set_opt("context", s1, 1);
 813                 if (tmp)
 814                         goto next;
 815                 tmp = ll_set_opt("fscontext", s1, 1);
 816                 if (tmp)
 817                         goto next;
 818                 tmp = ll_set_opt("defcontext", s1, 1);
 819                 if (tmp)
 820                         goto next;
 821                 tmp = ll_set_opt("rootcontext", s1, 1);
 822                 if (tmp)
 823                         goto next;
 824                 tmp = ll_set_opt("user_fid2path", s1, LL_SBI_USER_FID2PATH);
 825                 if (tmp) {
 826                         *flags |= tmp;
 827                         goto next;
 828                 }
 829                 tmp = ll_set_opt("nouser_fid2path", s1, LL_SBI_USER_FID2PATH);
 830                 if (tmp) {
 831                         *flags &= ~tmp;
 832                         goto next;
 833                 }
 834
 835                 tmp = ll_set_opt("checksum", s1, LL_SBI_CHECKSUM);
 836                 if (tmp) {
 837                         *flags |= tmp;
 838                         goto next;
 839                 }
 840                 tmp = ll_set_opt("nochecksum", s1, LL_SBI_CHECKSUM);
 841                 if (tmp) {
 842                         *flags &= ~tmp;
 843                         goto next;
 844                 }
 845                 tmp = ll_set_opt("lruresize", s1, LL_SBI_LRU_RESIZE);
 846                 if (tmp) {
 847                         *flags |= tmp;
 848                         goto next;
 849                 }
 850                 tmp = ll_set_opt("nolruresize", s1, LL_SBI_LRU_RESIZE);
 851                 if (tmp) {
 852                         *flags &= ~tmp;
 853                         goto next;
 854                 }
 855                 tmp = ll_set_opt("lazystatfs", s1, LL_SBI_LAZYSTATFS);
 856                 if (tmp) {
 857                         *flags |= tmp;
 858                         goto next;
 859                 }
 860                 tmp = ll_set_opt("nolazystatfs", s1, LL_SBI_LAZYSTATFS);
 861                 if (tmp) {
 862                         *flags &= ~tmp;
 863                         goto next;
 864                 }
 865                 tmp = ll_set_opt("32bitapi", s1, LL_SBI_32BIT_API);
 866                 if (tmp) {
 867                         *flags |= tmp;
 868                         goto next;
 869                 }
 870                 tmp = ll_set_opt("verbose", s1, LL_SBI_VERBOSE);
 871                 if (tmp) {
 872                         *flags |= tmp;
 873                         goto next;
 874                 }
 875                 tmp = ll_set_opt("noverbose", s1, LL_SBI_VERBOSE);
 876                 if (tmp) {
 877                         *flags &= ~tmp;
 878                         goto next;
 879                 }
 880                 tmp = ll_set_opt("always_ping", s1, LL_SBI_ALWAYS_PING);
 881                 if (tmp) {
 882                         *flags |= tmp;
 883                         goto next;
 884                 }
 885                 LCONSOLE_ERROR_MSG(0x152, "Unknown option '%s', won't mount.\n",
 886                                    s1);
 887                 RETURN(-EINVAL);
 888
 889 next:
 890                 /* Find next opt */
 891                 s2 = strchr(s1, ',');
 892                 if (s2 == NULL)
 893                         break;
 894                 s1 = s2 + 1;
 895         }
 896         RETURN(0);
 897 }
 898
 899 void ll_lli_init(struct ll_inode_info *lli)
 900 {
 901         lli->lli_inode_magic = LLI_INODE_MAGIC;
 902         lli->lli_flags = 0;
 903         spin_lock_init(&lli->lli_lock);
 904         lli->lli_posix_acl = NULL;
 905         /* Do not set lli_fid, it has been initialized already. */
 906         fid_zero(&lli->lli_pfid);
 907         lli->lli_mds_read_och = NULL;
 908         lli->lli_mds_write_och = NULL;
 909         lli->lli_mds_exec_och = NULL;
 910         lli->lli_open_fd_read_count = 0;
 911         lli->lli_open_fd_write_count = 0;
 912         lli->lli_open_fd_exec_count = 0;
 913         mutex_init(&lli->lli_och_mutex);
 914         spin_lock_init(&lli->lli_agl_lock);
 915         spin_lock_init(&lli->lli_layout_lock);
 916         ll_layout_version_set(lli, CL_LAYOUT_GEN_NONE);
 917         lli->lli_clob = NULL;
 918
 919         init_rwsem(&lli->lli_xattrs_list_rwsem);
 920         mutex_init(&lli->lli_xattrs_enq_lock);
 921
 922         LASSERT(lli->lli_vfs_inode.i_mode != 0);
 923         if (S_ISDIR(lli->lli_vfs_inode.i_mode)) {
 924                 mutex_init(&lli->lli_readdir_mutex);
 925                 lli->lli_opendir_key = NULL;
 926                 lli->lli_sai = NULL;
 927                 spin_lock_init(&lli->lli_sa_lock);
 928                 lli->lli_opendir_pid = 0;
 929                 lli->lli_sa_enabled = 0;
 930                 lli->lli_def_stripe_offset = -1;
 931         } else {
 932                 mutex_init(&lli->lli_size_mutex);
 933                 lli->lli_symlink_name = NULL;
 934                 init_rwsem(&lli->lli_trunc_sem);
 935                 range_lock_tree_init(&lli->lli_write_tree);
 936                 init_rwsem(&lli->lli_glimpse_sem);
 937                 lli->lli_glimpse_time = 0;
 938                 INIT_LIST_HEAD(&lli->lli_agl_list);
 939                 lli->lli_agl_index = 0;
 940                 lli->lli_async_rc = 0;
 941         }
 942         mutex_init(&lli->lli_layout_mutex);
 943         memset(lli->lli_jobid, 0, LUSTRE_JOBID_SIZE);
 944 }
 945
 946 static inline int ll_bdi_register(struct backing_dev_info *bdi)
 947 {
 948         static atomic_t ll_bdi_num = ATOMIC_INIT(0);
 949
 950         bdi->name = "lustre";
 951         return bdi_register(bdi, NULL, "lustre-%d",
 952                             atomic_inc_return(&ll_bdi_num));
 953 }
 954
 955 int ll_fill_super(struct super_block *sb, struct vfsmount *mnt)
 956 {
 957         struct  lustre_profile *lprof = NULL;
 958         struct  lustre_sb_info *lsi = s2lsi(sb);
 959         struct  ll_sb_info *sbi;
 960         char    *dt = NULL, *md = NULL;
 961         char    *profilenm = get_profile_name(sb);
 962         struct config_llog_instance *cfg;
 963         /* %p for void* in printf needs 16+2 characters: 0xffffffffffffffff */
 964         const int instlen = sizeof(cfg->cfg_instance) * 2 + 2;
 965         int     md_len = 0;
 966         int     dt_len = 0;
 967         int     err;
 968         ENTRY;
 969
 970         CDEBUG(D_VFSTRACE, "VFS Op: sb %p\n", sb);
 971
 972         OBD_ALLOC_PTR(cfg);
 973         if (cfg == NULL)
 974                 RETURN(-ENOMEM);
 975
 976         try_module_get(THIS_MODULE);
 977
 978         /* client additional sb info */
 979         lsi->lsi_llsbi = sbi = ll_init_sbi();
 980         if (!sbi) {
 981                 module_put(THIS_MODULE);
 982                 OBD_FREE_PTR(cfg);
 983                 RETURN(-ENOMEM);
 984         }
 985
 986         err = ll_options(lsi->lsi_lmd->lmd_opts, &sbi->ll_flags);
 987         if (err)
 988                 GOTO(out_free, err);
 989
 990         err = bdi_init(&lsi->lsi_bdi);
 991         if (err)
 992                 GOTO(out_free, err);
 993         lsi->lsi_flags |= LSI_BDI_INITIALIZED;
 994 #ifdef HAVE_BDI_CAP_MAP_COPY
 995         lsi->lsi_bdi.capabilities = BDI_CAP_MAP_COPY;
 996 #else
 997         lsi->lsi_bdi.capabilities = 0;
 998 #endif
 999         err = ll_bdi_register(&lsi->lsi_bdi);
1000         if (err)
1001                 GOTO(out_free, err);
1002
1003         sb->s_bdi = &lsi->lsi_bdi;
1004 #ifndef HAVE_DCACHE_LOCK
1005         /* kernel >= 2.6.38 store dentry operations in sb->s_d_op. */
1006         sb->s_d_op = &ll_d_ops;
1007 #endif
1008
1009         /* Call lprocfs_ll_register_mountpoint() before lustre_process_log()
1010          * so that "llite.*.*" params can be processed correctly. */
1011         if (proc_lustre_fs_root != NULL) {
1012                 err = lprocfs_ll_register_mountpoint(proc_lustre_fs_root, sb);
1013                 if (err < 0) {
1014                         CERROR("%s: could not register mountpoint in llite: "
1015                                "rc = %d\n", ll_get_fsname(sb, NULL, 0), err);
1016                         err = 0;
1017                 }
1018         }
1019
1020         /* Generate a string unique to this super, in case some joker tries
1021            to mount the same fs at two mount points.
1022            Use the address of the super itself.*/
1023         cfg->cfg_instance = sb;
1024         cfg->cfg_uuid = lsi->lsi_llsbi->ll_sb_uuid;
1025         cfg->cfg_callback = class_config_llog_handler;
1026         cfg->cfg_sub_clds = CONFIG_SUB_CLIENT;
1027         /* set up client obds */
1028         err = lustre_process_log(sb, profilenm, cfg);
1029         if (err < 0)
1030                 GOTO(out_proc, err);
1031
1032         /* Profile set with LCFG_MOUNTOPT so we can find our mdc and osc obds */
1033         lprof = class_get_profile(profilenm);
1034         if (lprof == NULL) {
1035                 LCONSOLE_ERROR_MSG(0x156, "The client profile '%s' could not be"
1036                                    " read from the MGS.  Does that filesystem "
1037                                    "exist?\n", profilenm);
1038                 GOTO(out_proc, err = -EINVAL);
1039         }
1040         CDEBUG(D_CONFIG, "Found profile %s: mdc=%s osc=%s\n", profilenm,
1041                lprof->lp_md, lprof->lp_dt);
1042
1043         dt_len = strlen(lprof->lp_dt) + instlen + 2;
1044         OBD_ALLOC(dt, dt_len);
1045         if (!dt)
1046                 GOTO(out_proc, err = -ENOMEM);
1047         snprintf(dt, dt_len - 1, "%s-%p", lprof->lp_dt, cfg->cfg_instance);
1048
1049         md_len = strlen(lprof->lp_md) + instlen + 2;
1050         OBD_ALLOC(md, md_len);
1051         if (!md)
1052                 GOTO(out_proc, err = -ENOMEM);
1053         snprintf(md, md_len - 1, "%s-%p", lprof->lp_md, cfg->cfg_instance);
1054
1055         /* connections, registrations, sb setup */
1056         err = client_common_fill_super(sb, md, dt, mnt);
1057         if (err < 0)
1058                 GOTO(out_proc, err);
1059
1060         sbi->ll_client_common_fill_super_succeeded = 1;
1061
1062 out_proc:
1063         if (err < 0)
1064                 lprocfs_ll_unregister_mountpoint(sbi);
1065 out_free:
1066         if (md)
1067                 OBD_FREE(md, md_len);
1068         if (dt)
1069                 OBD_FREE(dt, dt_len);
1070         if (lprof != NULL)
1071                 class_put_profile(lprof);
1072         if (err)
1073                 ll_put_super(sb);
1074         else if (sbi->ll_flags & LL_SBI_VERBOSE)
1075                 LCONSOLE_WARN("Mounted %s\n", profilenm);
1076
1077         OBD_FREE_PTR(cfg);
1078         RETURN(err);
1079 } /* ll_fill_super */
1080
1081 void ll_put_super(struct super_block *sb)
1082 {
1083         struct config_llog_instance cfg, params_cfg;
1084         struct obd_device *obd;
1085         struct lustre_sb_info *lsi = s2lsi(sb);
1086         struct ll_sb_info *sbi = ll_s2sbi(sb);
1087         char *profilenm = get_profile_name(sb);
1088         long ccc_count;
1089         int next, force = 1, rc = 0;
1090         ENTRY;
1091
1092         CDEBUG(D_VFSTRACE, "VFS Op: sb %p - %s\n", sb, profilenm);
1093
1094         cfg.cfg_instance = sb;
1095         lustre_end_log(sb, profilenm, &cfg);
1096
1097         params_cfg.cfg_instance = sb;
1098         lustre_end_log(sb, PARAMS_FILENAME, &params_cfg);
1099
1100         if (sbi->ll_md_exp) {
1101                 obd = class_exp2obd(sbi->ll_md_exp);
1102                 if (obd)
1103                         force = obd->obd_force;
1104         }
1105
1106         /* Wait for unstable pages to be committed to stable storage */
1107         if (force == 0) {
1108                 struct l_wait_info lwi = LWI_INTR(LWI_ON_SIGNAL_NOOP, NULL);
1109                 rc = l_wait_event(sbi->ll_cache->ccc_unstable_waitq,
1110                         atomic_long_read(&sbi->ll_cache->ccc_unstable_nr) == 0,
1111                         &lwi);
1112         }
1113
1114         ccc_count = atomic_long_read(&sbi->ll_cache->ccc_unstable_nr);
1115         if (force == 0 && rc != -EINTR)
1116                 LASSERTF(ccc_count == 0, "count: %li\n", ccc_count);
1117
1118
1119         /* We need to set force before the lov_disconnect in
1120            lustre_common_put_super, since l_d cleans up osc's as well. */
1121         if (force) {
1122                 next = 0;
1123                 while ((obd = class_devices_in_group(&sbi->ll_sb_uuid,
1124                                                      &next)) != NULL) {
1125                         obd->obd_force = force;
1126                 }
1127         }
1128
1129         if (sbi->ll_client_common_fill_super_succeeded) {
1130                 /* Only if client_common_fill_super succeeded */
1131                 client_common_put_super(sb);
1132         }
1133
1134         next = 0;
1135         while ((obd = class_devices_in_group(&sbi->ll_sb_uuid, &next)) !=NULL) {
1136                 class_manual_cleanup(obd);
1137         }
1138
1139         if (sbi->ll_flags & LL_SBI_VERBOSE)
1140                 LCONSOLE_WARN("Unmounted %s\n", profilenm ? profilenm : "");
1141
1142         if (profilenm)
1143                 class_del_profile(profilenm);
1144
1145         if (lsi->lsi_flags & LSI_BDI_INITIALIZED) {
1146                 bdi_destroy(&lsi->lsi_bdi);
1147                 lsi->lsi_flags &= ~LSI_BDI_INITIALIZED;
1148         }
1149
1150         ll_free_sbi(sb);
1151         lsi->lsi_llsbi = NULL;
1152
1153         lustre_common_put_super(sb);
1154
1155         cl_env_cache_purge(~0);
1156
1157         module_put(THIS_MODULE);
1158
1159         EXIT;
1160 } /* client_put_super */
1161
1162 struct inode *ll_inode_from_resource_lock(struct ldlm_lock *lock)
1163 {
1164         struct inode *inode = NULL;
1165
1166         /* NOTE: we depend on atomic igrab() -bzzz */
1167         lock_res_and_lock(lock);
1168         if (lock->l_resource->lr_lvb_inode) {
1169                 struct ll_inode_info * lli;
1170                 lli = ll_i2info(lock->l_resource->lr_lvb_inode);
1171                 if (lli->lli_inode_magic == LLI_INODE_MAGIC) {
1172                         inode = igrab(lock->l_resource->lr_lvb_inode);
1173                 } else {
1174                         inode = lock->l_resource->lr_lvb_inode;
1175                         LDLM_DEBUG_LIMIT(inode->i_state & I_FREEING ?  D_INFO :
1176                                          D_WARNING, lock, "lr_lvb_inode %p is "
1177                                          "bogus: magic %08x",
1178                                          lock->l_resource->lr_lvb_inode,
1179                                          lli->lli_inode_magic);
1180                         inode = NULL;
1181                 }
1182         }
1183         unlock_res_and_lock(lock);
1184         return inode;
1185 }
1186
1187 void ll_dir_clear_lsm_md(struct inode *inode)
1188 {
1189         struct ll_inode_info *lli = ll_i2info(inode);
1190
1191         LASSERT(S_ISDIR(inode->i_mode));
1192
1193         if (lli->lli_lsm_md != NULL) {
1194                 lmv_free_memmd(lli->lli_lsm_md);
1195                 lli->lli_lsm_md = NULL;
1196         }
1197 }
1198
1199 static struct inode *ll_iget_anon_dir(struct super_block *sb,
1200                                       const struct lu_fid *fid,
1201                                       struct lustre_md *md)
1202 {
1203         struct ll_sb_info       *sbi = ll_s2sbi(sb);
1204         struct mdt_body         *body = md->body;
1205         struct inode            *inode;
1206         ino_t                   ino;
1207         ENTRY;
1208
1209         ino = cl_fid_build_ino(fid, sbi->ll_flags & LL_SBI_32BIT_API);
1210         inode = iget_locked(sb, ino);
1211         if (inode == NULL) {
1212                 CERROR("%s: failed get simple inode "DFID": rc = -ENOENT\n",
1213                        ll_get_fsname(sb, NULL, 0), PFID(fid));
1214                 RETURN(ERR_PTR(-ENOENT));
1215         }
1216
1217         if (inode->i_state & I_NEW) {
1218                 struct ll_inode_info *lli = ll_i2info(inode);
1219                 struct lmv_stripe_md *lsm = md->lmv;
1220
1221                 inode->i_mode = (inode->i_mode & ~S_IFMT) |
1222                                 (body->mbo_mode & S_IFMT);
1223                 LASSERTF(S_ISDIR(inode->i_mode), "Not slave inode "DFID"\n",
1224                          PFID(fid));
1225
1226                 LTIME_S(inode->i_mtime) = 0;
1227                 LTIME_S(inode->i_atime) = 0;
1228                 LTIME_S(inode->i_ctime) = 0;
1229                 inode->i_rdev = 0;
1230
1231 #ifdef HAVE_BACKING_DEV_INFO
1232                 /* initializing backing dev info. */
1233                 inode->i_mapping->backing_dev_info =
1234                                                 &s2lsi(inode->i_sb)->lsi_bdi;
1235 #endif
1236                 inode->i_op = &ll_dir_inode_operations;
1237                 inode->i_fop = &ll_dir_operations;
1238                 lli->lli_fid = *fid;
1239                 ll_lli_init(lli);
1240
1241                 LASSERT(lsm != NULL);
1242                 /* master object FID */
1243                 lli->lli_pfid = body->mbo_fid1;
1244                 CDEBUG(D_INODE, "lli %p slave "DFID" master "DFID"\n",
1245                        lli, PFID(fid), PFID(&lli->lli_pfid));
1246                 unlock_new_inode(inode);
1247         }
1248
1249         RETURN(inode);
1250 }
1251
1252 static int ll_init_lsm_md(struct inode *inode, struct lustre_md *md)
1253 {
1254         struct lu_fid *fid;
1255         struct lmv_stripe_md *lsm = md->lmv;
1256         int i;
1257
1258         LASSERT(lsm != NULL);
1259         /* XXX sigh, this lsm_root initialization should be in
1260          * LMV layer, but it needs ll_iget right now, so we
1261          * put this here right now. */
1262         for (i = 0; i < lsm->lsm_md_stripe_count; i++) {
1263                 fid = &lsm->lsm_md_oinfo[i].lmo_fid;
1264                 LASSERT(lsm->lsm_md_oinfo[i].lmo_root == NULL);
1265                 /* Unfortunately ll_iget will call ll_update_inode,
1266                  * where the initialization of slave inode is slightly
1267                  * different, so it reset lsm_md to NULL to avoid
1268                  * initializing lsm for slave inode. */
1269                 /* For migrating inode, master stripe and master object will
1270                  * be same, so we only need assign this inode */
1271                 if (lsm->lsm_md_hash_type & LMV_HASH_FLAG_MIGRATION && i == 0)
1272                         lsm->lsm_md_oinfo[i].lmo_root = inode;
1273                 else
1274                         lsm->lsm_md_oinfo[i].lmo_root =
1275                                 ll_iget_anon_dir(inode->i_sb, fid, md);
1276
1277                 if (IS_ERR(lsm->lsm_md_oinfo[i].lmo_root)) {
1278                         int rc = PTR_ERR(lsm->lsm_md_oinfo[i].lmo_root);
1279
1280                         lsm->lsm_md_oinfo[i].lmo_root = NULL;
1281                         return rc;
1282                 }
1283         }
1284
1285         return 0;
1286 }
1287
1288 static inline int lli_lsm_md_eq(const struct lmv_stripe_md *lsm_md1,
1289                                 const struct lmv_stripe_md *lsm_md2)
1290 {
1291         return lsm_md1->lsm_md_magic == lsm_md2->lsm_md_magic &&
1292                lsm_md1->lsm_md_stripe_count == lsm_md2->lsm_md_stripe_count &&
1293                lsm_md1->lsm_md_master_mdt_index ==
1294                                         lsm_md2->lsm_md_master_mdt_index &&
1295                lsm_md1->lsm_md_hash_type == lsm_md2->lsm_md_hash_type &&
1296                lsm_md1->lsm_md_layout_version ==
1297                                         lsm_md2->lsm_md_layout_version &&
1298                strcmp(lsm_md1->lsm_md_pool_name,
1299                       lsm_md2->lsm_md_pool_name) == 0;
1300 }
1301
1302 static int ll_update_lsm_md(struct inode *inode, struct lustre_md *md)
1303 {
1304         struct ll_inode_info *lli = ll_i2info(inode);
1305         struct lmv_stripe_md *lsm = md->lmv;
1306         int     rc;
1307         ENTRY;
1308
1309         LASSERT(S_ISDIR(inode->i_mode));
1310         CDEBUG(D_INODE, "update lsm %p of "DFID"\n", lli->lli_lsm_md,
1311                PFID(ll_inode2fid(inode)));
1312
1313         /* no striped information from request. */
1314         if (lsm == NULL) {
1315                 if (lli->lli_lsm_md == NULL) {
1316                         RETURN(0);
1317                 } else if (lli->lli_lsm_md->lsm_md_hash_type &
1318                                                 LMV_HASH_FLAG_MIGRATION) {
1319                         /* migration is done, the temporay MIGRATE layout has
1320                          * been removed */
1321                         CDEBUG(D_INODE, DFID" finish migration.\n",
1322                                PFID(ll_inode2fid(inode)));
1323                         lmv_free_memmd(lli->lli_lsm_md);
1324                         lli->lli_lsm_md = NULL;
1325                         RETURN(0);
1326                 } else {
1327                         /* The lustre_md from req does not include stripeEA,
1328                          * see ll_md_setattr */
1329                         RETURN(0);
1330                 }
1331         }
1332
1333         /* set the directory layout */
1334         if (lli->lli_lsm_md == NULL) {
1335                 struct cl_attr  *attr;
1336
1337                 rc = ll_init_lsm_md(inode, md);
1338                 if (rc != 0)
1339                         RETURN(rc);
1340
1341                 /* set md->lmv to NULL, so the following free lustre_md
1342                  * will not free this lsm */
1343                 md->lmv = NULL;
1344                 lli->lli_lsm_md = lsm;
1345
1346                 OBD_ALLOC_PTR(attr);
1347                 if (attr == NULL)
1348                         RETURN(-ENOMEM);
1349
1350                 /* validate the lsm */
1351                 rc = md_merge_attr(ll_i2mdexp(inode), lsm, attr,
1352                                    ll_md_blocking_ast);
1353                 if (rc != 0) {
1354                         OBD_FREE_PTR(attr);
1355                         RETURN(rc);
1356                 }
1357
1358                 if (md->body->mbo_valid & OBD_MD_FLNLINK)
1359                         md->body->mbo_nlink = attr->cat_nlink;
1360                 if (md->body->mbo_valid & OBD_MD_FLSIZE)
1361                         md->body->mbo_size = attr->cat_size;
1362                 if (md->body->mbo_valid & OBD_MD_FLATIME)
1363                         md->body->mbo_atime = attr->cat_atime;
1364                 if (md->body->mbo_valid & OBD_MD_FLCTIME)
1365                         md->body->mbo_ctime = attr->cat_ctime;
1366                 if (md->body->mbo_valid & OBD_MD_FLMTIME)
1367                         md->body->mbo_mtime = attr->cat_mtime;
1368
1369                 OBD_FREE_PTR(attr);
1370
1371                 CDEBUG(D_INODE, "Set lsm %p magic %x to "DFID"\n", lsm,
1372                        lsm->lsm_md_magic, PFID(ll_inode2fid(inode)));
1373                 RETURN(0);
1374         }
1375
1376         /* Compare the old and new stripe information */
1377         if (!lsm_md_eq(lli->lli_lsm_md, lsm)) {
1378                 struct lmv_stripe_md    *old_lsm = lli->lli_lsm_md;
1379                 int                     idx;
1380
1381                 CERROR("%s: inode "DFID"(%p)'s lmv layout mismatch (%p)/(%p)"
1382                        "magic:0x%x/0x%x stripe count: %d/%d master_mdt: %d/%d"
1383                        "hash_type:0x%x/0x%x layout: 0x%x/0x%x pool:%s/%s\n",
1384                        ll_get_fsname(inode->i_sb, NULL, 0), PFID(&lli->lli_fid),
1385                        inode, lsm, old_lsm,
1386                        lsm->lsm_md_magic, old_lsm->lsm_md_magic,
1387                        lsm->lsm_md_stripe_count,
1388                        old_lsm->lsm_md_stripe_count,
1389                        lsm->lsm_md_master_mdt_index,
1390                        old_lsm->lsm_md_master_mdt_index,
1391                        lsm->lsm_md_hash_type, old_lsm->lsm_md_hash_type,
1392                        lsm->lsm_md_layout_version,
1393                        old_lsm->lsm_md_layout_version,
1394                        lsm->lsm_md_pool_name,
1395                        old_lsm->lsm_md_pool_name);
1396
1397                 for (idx = 0; idx < old_lsm->lsm_md_stripe_count; idx++) {
1398                         CERROR("%s: sub FIDs in old lsm idx %d, old: "DFID"\n",
1399                                ll_get_fsname(inode->i_sb, NULL, 0), idx,
1400                                PFID(&old_lsm->lsm_md_oinfo[idx].lmo_fid));
1401                 }
1402
1403                 for (idx = 0; idx < lsm->lsm_md_stripe_count; idx++) {
1404                         CERROR("%s: sub FIDs in new lsm idx %d, new: "DFID"\n",
1405                                ll_get_fsname(inode->i_sb, NULL, 0), idx,
1406                                PFID(&lsm->lsm_md_oinfo[idx].lmo_fid));
1407                 }
1408
1409                 RETURN(-EIO);
1410         }
1411
1412         RETURN(0);
1413 }
1414
1415 void ll_clear_inode(struct inode *inode)
1416 {
1417         struct ll_inode_info *lli = ll_i2info(inode);
1418         struct ll_sb_info *sbi = ll_i2sbi(inode);
1419         ENTRY;
1420
1421         CDEBUG(D_VFSTRACE, "VFS Op:inode="DFID"(%p)\n",
1422                PFID(ll_inode2fid(inode)), inode);
1423
1424         if (S_ISDIR(inode->i_mode)) {
1425                 /* these should have been cleared in ll_file_release */
1426                 LASSERT(lli->lli_opendir_key == NULL);
1427                 LASSERT(lli->lli_sai == NULL);
1428                 LASSERT(lli->lli_opendir_pid == 0);
1429         }
1430
1431         md_null_inode(sbi->ll_md_exp, ll_inode2fid(inode));
1432
1433         LASSERT(!lli->lli_open_fd_write_count);
1434         LASSERT(!lli->lli_open_fd_read_count);
1435         LASSERT(!lli->lli_open_fd_exec_count);
1436
1437         if (lli->lli_mds_write_och)
1438                 ll_md_real_close(inode, FMODE_WRITE);
1439         if (lli->lli_mds_exec_och)
1440                 ll_md_real_close(inode, FMODE_EXEC);
1441         if (lli->lli_mds_read_och)
1442                 ll_md_real_close(inode, FMODE_READ);
1443
1444         if (S_ISLNK(inode->i_mode) && lli->lli_symlink_name) {
1445                 OBD_FREE(lli->lli_symlink_name,
1446                          strlen(lli->lli_symlink_name) + 1);
1447                 lli->lli_symlink_name = NULL;
1448         }
1449
1450         ll_xattr_cache_destroy(inode);
1451
1452 #ifdef CONFIG_FS_POSIX_ACL
1453         forget_all_cached_acls(inode);
1454         if (lli->lli_posix_acl) {
1455                 LASSERT(atomic_read(&lli->lli_posix_acl->a_refcount) == 1);
1456                 posix_acl_release(lli->lli_posix_acl);
1457                 lli->lli_posix_acl = NULL;
1458         }
1459 #endif
1460         lli->lli_inode_magic = LLI_INODE_DEAD;
1461
1462         if (S_ISDIR(inode->i_mode))
1463                 ll_dir_clear_lsm_md(inode);
1464         else if (S_ISREG(inode->i_mode) && !is_bad_inode(inode))
1465                 LASSERT(list_empty(&lli->lli_agl_list));
1466
1467         /*
1468          * XXX This has to be done before lsm is freed below, because
1469          * cl_object still uses inode lsm.
1470          */
1471         cl_inode_fini(inode);
1472
1473         EXIT;
1474 }
1475
1476 static int ll_md_setattr(struct dentry *dentry, struct md_op_data *op_data)
1477 {
1478         struct lustre_md md;
1479         struct inode *inode = dentry->d_inode;
1480         struct ll_sb_info *sbi = ll_i2sbi(inode);
1481         struct ptlrpc_request *request = NULL;
1482         int rc, ia_valid;
1483         ENTRY;
1484
1485         op_data = ll_prep_md_op_data(op_data, inode, NULL, NULL, 0, 0,
1486                                      LUSTRE_OPC_ANY, NULL);
1487         if (IS_ERR(op_data))
1488                 RETURN(PTR_ERR(op_data));
1489
1490         rc = md_setattr(sbi->ll_md_exp, op_data, NULL, 0, &request);
1491         if (rc) {
1492                 ptlrpc_req_finished(request);
1493                 if (rc == -ENOENT) {
1494                         clear_nlink(inode);
1495                         /* Unlinked special device node? Or just a race?
1496                          * Pretend we done everything. */
1497                         if (!S_ISREG(inode->i_mode) &&
1498                             !S_ISDIR(inode->i_mode)) {
1499                                 ia_valid = op_data->op_attr.ia_valid;
1500                                 op_data->op_attr.ia_valid &= ~TIMES_SET_FLAGS;
1501                                 rc = simple_setattr(dentry, &op_data->op_attr);
1502                                 op_data->op_attr.ia_valid = ia_valid;
1503                         }
1504                 } else if (rc != -EPERM && rc != -EACCES && rc != -ETXTBSY) {
1505                         CERROR("md_setattr fails: rc = %d\n", rc);
1506                 }
1507                 RETURN(rc);
1508         }
1509
1510         rc = md_get_lustre_md(sbi->ll_md_exp, request, sbi->ll_dt_exp,
1511                               sbi->ll_md_exp, &md);
1512         if (rc) {
1513                 ptlrpc_req_finished(request);
1514                 RETURN(rc);
1515         }
1516
1517         ia_valid = op_data->op_attr.ia_valid;
1518         /* inode size will be in ll_setattr_ost, can't do it now since dirty
1519          * cache is not cleared yet. */
1520         op_data->op_attr.ia_valid &= ~(TIMES_SET_FLAGS | ATTR_SIZE);
1521         if (S_ISREG(inode->i_mode))
1522                 inode_lock(inode);
1523         rc = simple_setattr(dentry, &op_data->op_attr);
1524         if (S_ISREG(inode->i_mode))
1525                 inode_unlock(inode);
1526         op_data->op_attr.ia_valid = ia_valid;
1527
1528         rc = ll_update_inode(inode, &md);
1529         ptlrpc_req_finished(request);
1530
1531         RETURN(rc);
1532 }
1533
1534 /* If this inode has objects allocated to it (lsm != NULL), then the OST
1535  * object(s) determine the file size and mtime.  Otherwise, the MDS will
1536  * keep these values until such a time that objects are allocated for it.
1537  * We do the MDS operations first, as it is checking permissions for us.
1538  * We don't to the MDS RPC if there is nothing that we want to store there,
1539  * otherwise there is no harm in updating mtime/atime on the MDS if we are
1540  * going to do an RPC anyways.
1541  *
1542  * If we are doing a truncate, we will send the mtime and ctime updates
1543  * to the OST with the punch RPC, otherwise we do an explicit setattr RPC.
1544  * I don't believe it is possible to get e.g. ATTR_MTIME_SET and ATTR_SIZE
1545  * at the same time.
1546  *
1547  * In case of HSMimport, we only set attr on MDS.
1548  */
1549 int ll_setattr_raw(struct dentry *dentry, struct iattr *attr, bool hsm_import)
1550 {
1551         struct inode *inode = dentry->d_inode;
1552         struct ll_inode_info *lli = ll_i2info(inode);
1553         struct md_op_data *op_data = NULL;
1554         int rc = 0;
1555         ENTRY;
1556
1557         CDEBUG(D_VFSTRACE, "%s: setattr inode "DFID"(%p) from %llu to %llu, "
1558                "valid %x, hsm_import %d\n",
1559                ll_get_fsname(inode->i_sb, NULL, 0), PFID(&lli->lli_fid),
1560                inode, i_size_read(inode), attr->ia_size, attr->ia_valid,
1561                hsm_import);
1562
1563         if (attr->ia_valid & ATTR_SIZE) {
1564                 /* Check new size against VFS/VM file size limit and rlimit */
1565                 rc = inode_newsize_ok(inode, attr->ia_size);
1566                 if (rc)
1567                         RETURN(rc);
1568
1569                 /* The maximum Lustre file size is variable, based on the
1570                  * OST maximum object size and number of stripes.  This
1571                  * needs another check in addition to the VFS check above. */
1572                 if (attr->ia_size > ll_file_maxbytes(inode)) {
1573                         CDEBUG(D_INODE,"file "DFID" too large %llu > %llu\n",
1574                                PFID(&lli->lli_fid), attr->ia_size,
1575                                ll_file_maxbytes(inode));
1576                         RETURN(-EFBIG);
1577                 }
1578
1579                 attr->ia_valid |= ATTR_MTIME | ATTR_CTIME;
1580         }
1581
1582         /* POSIX: check before ATTR_*TIME_SET set (from inode_change_ok) */
1583         if (attr->ia_valid & TIMES_SET_FLAGS) {
1584                 if ((!uid_eq(current_fsuid(), inode->i_uid)) &&
1585                     !cfs_capable(CFS_CAP_FOWNER))
1586                         RETURN(-EPERM);
1587         }
1588
1589         /* We mark all of the fields "set" so MDS/OST does not re-set them */
1590         if (!(attr->ia_valid & ATTR_CTIME_SET) &&
1591             (attr->ia_valid & ATTR_CTIME)) {
1592                 attr->ia_ctime = current_time(inode);
1593                 attr->ia_valid |= ATTR_CTIME_SET;
1594         }
1595         if (!(attr->ia_valid & ATTR_ATIME_SET) &&
1596             (attr->ia_valid & ATTR_ATIME)) {
1597                 attr->ia_atime = current_time(inode);
1598                 attr->ia_valid |= ATTR_ATIME_SET;
1599         }
1600         if (!(attr->ia_valid & ATTR_MTIME_SET) &&
1601             (attr->ia_valid & ATTR_MTIME)) {
1602                 attr->ia_mtime = current_time(inode);
1603                 attr->ia_valid |= ATTR_MTIME_SET;
1604         }
1605
1606         if (attr->ia_valid & (ATTR_MTIME | ATTR_CTIME))
1607                 CDEBUG(D_INODE, "setting mtime %lu, ctime %lu, now = %llu\n",
1608                        LTIME_S(attr->ia_mtime), LTIME_S(attr->ia_ctime),
1609                        (s64)ktime_get_real_seconds());
1610
1611         if (S_ISREG(inode->i_mode)) {
1612                 if (attr->ia_valid & ATTR_SIZE)
1613                         inode_dio_write_done(inode);
1614                 inode_unlock(inode);
1615         }
1616
1617         /* We always do an MDS RPC, even if we're only changing the size;
1618          * only the MDS knows whether truncate() should fail with -ETXTBUSY */
1619
1620         OBD_ALLOC_PTR(op_data);
1621         if (op_data == NULL)
1622                 GOTO(out, rc = -ENOMEM);
1623
1624         if (!hsm_import && attr->ia_valid & ATTR_SIZE) {
1625                 /* If we are changing file size, file content is
1626                  * modified, flag it. */
1627                 attr->ia_valid |= MDS_OPEN_OWNEROVERRIDE;
1628                 op_data->op_bias |= MDS_DATA_MODIFIED;
1629                 ll_file_clear_flag(lli, LLIF_DATA_MODIFIED);
1630         }
1631
1632         op_data->op_attr = *attr;
1633
1634         rc = ll_md_setattr(dentry, op_data);
1635         if (rc)
1636                 GOTO(out, rc);
1637
1638         if (!S_ISREG(inode->i_mode) || hsm_import)
1639                 GOTO(out, rc = 0);
1640
1641         if (attr->ia_valid & (ATTR_SIZE |
1642                               ATTR_ATIME | ATTR_ATIME_SET |
1643                               ATTR_MTIME | ATTR_MTIME_SET |
1644                               ATTR_CTIME | ATTR_CTIME_SET)) {
1645                 /* For truncate and utimes sending attributes to OSTs, setting
1646                  * mtime/atime to the past will be performed under PW [0:EOF]
1647                  * extent lock (new_size:EOF for truncate).  It may seem
1648                  * excessive to send mtime/atime updates to OSTs when not
1649                  * setting times to past, but it is necessary due to possible
1650                  * time de-synchronization between MDT inode and OST objects */
1651                 rc = cl_setattr_ost(lli->lli_clob, attr, 0);
1652         }
1653
1654         /* If the file was restored, it needs to set dirty flag.
1655          *
1656          * We've already sent MDS_DATA_MODIFIED flag in
1657          * ll_md_setattr() for truncate. However, the MDT refuses to
1658          * set the HS_DIRTY flag on released files, so we have to set
1659          * it again if the file has been restored. Please check how
1660          * LLIF_DATA_MODIFIED is set in vvp_io_setattr_fini().
1661          *
1662          * Please notice that if the file is not released, the previous
1663          * MDS_DATA_MODIFIED has taken effect and usually
1664          * LLIF_DATA_MODIFIED is not set(see vvp_io_setattr_fini()).
1665          * This way we can save an RPC for common open + trunc
1666          * operation. */
1667         if (ll_file_test_and_clear_flag(lli, LLIF_DATA_MODIFIED)) {
1668                 struct hsm_state_set hss = {
1669                         .hss_valid = HSS_SETMASK,
1670                         .hss_setmask = HS_DIRTY,
1671                 };
1672                 int rc2;
1673
1674                 rc2 = ll_hsm_state_set(inode, &hss);
1675                 /* truncate and write can happen at the same time, so that
1676                  * the file can be set modified even though the file is not
1677                  * restored from released state, and ll_hsm_state_set() is
1678                  * not applicable for the file, and rc2 < 0 is normal in this
1679                  * case. */
1680                 if (rc2 < 0)
1681                         CDEBUG(D_INFO, DFID "HSM set dirty failed: rc2 = %d\n",
1682                                PFID(ll_inode2fid(inode)), rc2);
1683         }
1684
1685         EXIT;
1686 out:
1687         if (op_data != NULL)
1688                 ll_finish_md_op_data(op_data);
1689
1690         if (S_ISREG(inode->i_mode)) {
1691                 inode_lock(inode);
1692                 if ((attr->ia_valid & ATTR_SIZE) && !hsm_import)
1693                         inode_dio_wait(inode);
1694                 /* Once we've got the i_mutex, it's safe to set the S_NOSEC
1695                  * flag.  ll_update_inode (called from ll_md_setattr), clears
1696                  * inode flags, so there is a gap where S_NOSEC is not set.
1697                  * This can cause a writer to take the i_mutex unnecessarily,
1698                  * but this is safe to do and should be rare. */
1699                 inode_has_no_xattr(inode);
1700         }
1701
1702         ll_stats_ops_tally(ll_i2sbi(inode), (attr->ia_valid & ATTR_SIZE) ?
1703                         LPROC_LL_TRUNC : LPROC_LL_SETATTR, 1);
1704
1705         return rc;
1706 }
1707
1708 int ll_setattr(struct dentry *de, struct iattr *attr)
1709 {
1710         int mode = de->d_inode->i_mode;
1711
1712         if ((attr->ia_valid & (ATTR_CTIME|ATTR_SIZE|ATTR_MODE)) ==
1713                               (ATTR_CTIME|ATTR_SIZE|ATTR_MODE))
1714                 attr->ia_valid |= MDS_OPEN_OWNEROVERRIDE;
1715
1716         if (((attr->ia_valid & (ATTR_MODE|ATTR_FORCE|ATTR_SIZE)) ==
1717                                (ATTR_SIZE|ATTR_MODE)) &&
1718             (((mode & S_ISUID) && !(attr->ia_mode & S_ISUID)) ||
1719              (((mode & (S_ISGID|S_IXGRP)) == (S_ISGID|S_IXGRP)) &&
1720               !(attr->ia_mode & S_ISGID))))
1721                 attr->ia_valid |= ATTR_FORCE;
1722
1723         if ((attr->ia_valid & ATTR_MODE) &&
1724             (mode & S_ISUID) &&
1725             !(attr->ia_mode & S_ISUID) &&
1726             !(attr->ia_valid & ATTR_KILL_SUID))
1727                 attr->ia_valid |= ATTR_KILL_SUID;
1728
1729         if ((attr->ia_valid & ATTR_MODE) &&
1730             ((mode & (S_ISGID|S_IXGRP)) == (S_ISGID|S_IXGRP)) &&
1731             !(attr->ia_mode & S_ISGID) &&
1732             !(attr->ia_valid & ATTR_KILL_SGID))
1733                 attr->ia_valid |= ATTR_KILL_SGID;
1734
1735         /* avoid polluted from ATTR_TIMES_SET,
1736          * projid is not expected to be set here */
1737         attr->ia_valid &= ~MDS_ATTR_PROJID;
1738
1739         return ll_setattr_raw(de, attr, false);
1740 }
1741
1742 int ll_statfs_internal(struct super_block *sb, struct obd_statfs *osfs,
1743                        __u64 max_age, __u32 flags)
1744 {
1745         struct ll_sb_info *sbi = ll_s2sbi(sb);
1746         struct obd_statfs obd_osfs;
1747         int rc;
1748         ENTRY;
1749
1750         rc = obd_statfs(NULL, sbi->ll_md_exp, osfs, max_age, flags);
1751         if (rc) {
1752                 CERROR("md_statfs fails: rc = %d\n", rc);
1753                 RETURN(rc);
1754         }
1755
1756         osfs->os_type = sb->s_magic;
1757
1758         CDEBUG(D_SUPER, "MDC blocks %llu/%llu objects %llu/%llu\n",
1759                osfs->os_bavail, osfs->os_blocks, osfs->os_ffree,osfs->os_files);
1760
1761         if (sbi->ll_flags & LL_SBI_LAZYSTATFS)
1762                 flags |= OBD_STATFS_NODELAY;
1763
1764         rc = obd_statfs_rqset(sbi->ll_dt_exp, &obd_osfs, max_age, flags);
1765         if (rc) {
1766                 CERROR("obd_statfs fails: rc = %d\n", rc);
1767                 RETURN(rc);
1768         }
1769
1770         CDEBUG(D_SUPER, "OSC blocks %llu/%llu objects %llu/%llu\n",
1771                obd_osfs.os_bavail, obd_osfs.os_blocks, obd_osfs.os_ffree,
1772                obd_osfs.os_files);
1773
1774         osfs->os_bsize = obd_osfs.os_bsize;
1775         osfs->os_blocks = obd_osfs.os_blocks;
1776         osfs->os_bfree = obd_osfs.os_bfree;
1777         osfs->os_bavail = obd_osfs.os_bavail;
1778
1779         /* If we don't have as many objects free on the OST as inodes
1780          * on the MDS, we reduce the total number of inodes to
1781          * compensate, so that the "inodes in use" number is correct.
1782          */
1783         if (obd_osfs.os_ffree < osfs->os_ffree) {
1784                 osfs->os_files = (osfs->os_files - osfs->os_ffree) +
1785                         obd_osfs.os_ffree;
1786                 osfs->os_ffree = obd_osfs.os_ffree;
1787         }
1788
1789         RETURN(rc);
1790 }
1791 int ll_statfs(struct dentry *de, struct kstatfs *sfs)
1792 {
1793         struct super_block *sb = de->d_sb;
1794         struct obd_statfs osfs;
1795         __u64 fsid = huge_encode_dev(sb->s_dev);
1796         int rc;
1797
1798         CDEBUG(D_VFSTRACE, "VFS Op: at %llu jiffies\n", get_jiffies_64());
1799         ll_stats_ops_tally(ll_s2sbi(sb), LPROC_LL_STAFS, 1);
1800
1801         /* Some amount of caching on the client is allowed */
1802         rc = ll_statfs_internal(sb, &osfs,
1803                                 cfs_time_shift_64(-OBD_STATFS_CACHE_SECONDS),
1804                                 0);
1805         if (rc)
1806                 return rc;
1807
1808         statfs_unpack(sfs, &osfs);
1809
1810         /* We need to downshift for all 32-bit kernels, because we can't
1811          * tell if the kernel is being called via sys_statfs64() or not.
1812          * Stop before overflowing f_bsize - in which case it is better
1813          * to just risk EOVERFLOW if caller is using old sys_statfs(). */
1814         if (sizeof(long) < 8) {
1815                 while (osfs.os_blocks > ~0UL && sfs->f_bsize < 0x40000000) {
1816                         sfs->f_bsize <<= 1;
1817
1818                         osfs.os_blocks >>= 1;
1819                         osfs.os_bfree >>= 1;
1820                         osfs.os_bavail >>= 1;
1821                 }
1822         }
1823
1824         sfs->f_blocks = osfs.os_blocks;
1825         sfs->f_bfree = osfs.os_bfree;
1826         sfs->f_bavail = osfs.os_bavail;
1827         sfs->f_fsid.val[0] = (__u32)fsid;
1828         sfs->f_fsid.val[1] = (__u32)(fsid >> 32);
1829         return 0;
1830 }
1831
1832 void ll_inode_size_lock(struct inode *inode)
1833 {
1834         struct ll_inode_info *lli;
1835
1836         LASSERT(!S_ISDIR(inode->i_mode));
1837
1838         lli = ll_i2info(inode);
1839         mutex_lock(&lli->lli_size_mutex);
1840 }
1841
1842 void ll_inode_size_unlock(struct inode *inode)
1843 {
1844         struct ll_inode_info *lli;
1845
1846         lli = ll_i2info(inode);
1847         mutex_unlock(&lli->lli_size_mutex);
1848 }
1849
1850 int ll_update_inode(struct inode *inode, struct lustre_md *md)
1851 {
1852         struct ll_inode_info *lli = ll_i2info(inode);
1853         struct mdt_body *body = md->body;
1854         struct ll_sb_info *sbi = ll_i2sbi(inode);
1855         int rc = 0;
1856
1857         if (body->mbo_valid & OBD_MD_FLEASIZE) {
1858                 rc = cl_file_inode_init(inode, md);
1859                 if (rc)
1860                         return rc;
1861         }
1862
1863         if (S_ISDIR(inode->i_mode)) {
1864                 rc = ll_update_lsm_md(inode, md);
1865                 if (rc != 0)
1866                         return rc;
1867         }
1868
1869 #ifdef CONFIG_FS_POSIX_ACL
1870         if (body->mbo_valid & OBD_MD_FLACL) {
1871                 spin_lock(&lli->lli_lock);
1872                 if (lli->lli_posix_acl)
1873                         posix_acl_release(lli->lli_posix_acl);
1874                 lli->lli_posix_acl = md->posix_acl;
1875                 spin_unlock(&lli->lli_lock);
1876         }
1877 #endif
1878         inode->i_ino = cl_fid_build_ino(&body->mbo_fid1,
1879                                         sbi->ll_flags & LL_SBI_32BIT_API);
1880         inode->i_generation = cl_fid_build_gen(&body->mbo_fid1);
1881
1882         if (body->mbo_valid & OBD_MD_FLATIME) {
1883                 if (body->mbo_atime > LTIME_S(inode->i_atime))
1884                         LTIME_S(inode->i_atime) = body->mbo_atime;
1885                 lli->lli_atime = body->mbo_atime;
1886         }
1887
1888         if (body->mbo_valid & OBD_MD_FLMTIME) {
1889                 if (body->mbo_mtime > LTIME_S(inode->i_mtime)) {
1890                         CDEBUG(D_INODE, "setting ino %lu mtime from %lu "
1891                                "to %llu\n", inode->i_ino,
1892                                LTIME_S(inode->i_mtime), body->mbo_mtime);
1893                         LTIME_S(inode->i_mtime) = body->mbo_mtime;
1894                 }
1895                 lli->lli_mtime = body->mbo_mtime;
1896         }
1897
1898         if (body->mbo_valid & OBD_MD_FLCTIME) {
1899                 if (body->mbo_ctime > LTIME_S(inode->i_ctime))
1900                         LTIME_S(inode->i_ctime) = body->mbo_ctime;
1901                 lli->lli_ctime = body->mbo_ctime;
1902         }
1903
1904         /* Clear i_flags to remove S_NOSEC before permissions are updated */
1905         if (body->mbo_valid & OBD_MD_FLFLAGS)
1906                 inode->i_flags = ll_ext_to_inode_flags(body->mbo_flags);
1907         if (body->mbo_valid & OBD_MD_FLMODE)
1908                 inode->i_mode = (inode->i_mode & S_IFMT) |
1909                                 (body->mbo_mode & ~S_IFMT);
1910
1911         if (body->mbo_valid & OBD_MD_FLTYPE)
1912                 inode->i_mode = (inode->i_mode & ~S_IFMT) |
1913                                 (body->mbo_mode & S_IFMT);
1914
1915         LASSERT(inode->i_mode != 0);
1916         if (S_ISREG(inode->i_mode))
1917                 inode->i_blkbits = min(PTLRPC_MAX_BRW_BITS + 1,
1918                                        LL_MAX_BLKSIZE_BITS);
1919         else
1920                 inode->i_blkbits = inode->i_sb->s_blocksize_bits;
1921
1922         if (body->mbo_valid & OBD_MD_FLUID)
1923                 inode->i_uid = make_kuid(&init_user_ns, body->mbo_uid);
1924         if (body->mbo_valid & OBD_MD_FLGID)
1925                 inode->i_gid = make_kgid(&init_user_ns, body->mbo_gid);
1926         if (body->mbo_valid & OBD_MD_FLPROJID)
1927                 lli->lli_projid = body->mbo_projid;
1928         if (body->mbo_valid & OBD_MD_FLNLINK)
1929                 set_nlink(inode, body->mbo_nlink);
1930         if (body->mbo_valid & OBD_MD_FLRDEV)
1931                 inode->i_rdev = old_decode_dev(body->mbo_rdev);
1932
1933         if (body->mbo_valid & OBD_MD_FLID) {
1934                 /* FID shouldn't be changed! */
1935                 if (fid_is_sane(&lli->lli_fid)) {
1936                         LASSERTF(lu_fid_eq(&lli->lli_fid, &body->mbo_fid1),
1937                                  "Trying to change FID "DFID
1938                                  " to the "DFID", inode "DFID"(%p)\n",
1939                                  PFID(&lli->lli_fid), PFID(&body->mbo_fid1),
1940                                  PFID(ll_inode2fid(inode)), inode);
1941                 } else {
1942                         lli->lli_fid = body->mbo_fid1;
1943                 }
1944         }
1945
1946         LASSERT(fid_seq(&lli->lli_fid) != 0);
1947
1948         if (body->mbo_valid & OBD_MD_FLSIZE) {
1949                 i_size_write(inode, body->mbo_size);
1950
1951                 CDEBUG(D_VFSTRACE, "inode="DFID", updating i_size %llu\n",
1952                        PFID(ll_inode2fid(inode)),
1953                        (unsigned long long)body->mbo_size);
1954
1955                 if (body->mbo_valid & OBD_MD_FLBLOCKS)
1956                         inode->i_blocks = body->mbo_blocks;
1957         }
1958
1959         if (body->mbo_valid & OBD_MD_TSTATE) {
1960                 /* Set LLIF_FILE_RESTORING if restore ongoing and
1961                  * clear it when done to ensure to start again
1962                  * glimpsing updated attrs
1963                  */
1964                 if (body->mbo_t_state & MS_RESTORE)
1965                         ll_file_set_flag(lli, LLIF_FILE_RESTORING);
1966                 else
1967                         ll_file_clear_flag(lli, LLIF_FILE_RESTORING);
1968         }
1969
1970         return 0;
1971 }
1972
1973 int ll_read_inode2(struct inode *inode, void *opaque)
1974 {
1975         struct lustre_md *md = opaque;
1976         struct ll_inode_info *lli = ll_i2info(inode);
1977         int     rc;
1978         ENTRY;
1979
1980         CDEBUG(D_VFSTRACE, "VFS Op:inode="DFID"(%p)\n",
1981                PFID(&lli->lli_fid), inode);
1982
1983         /* Core attributes from the MDS first.  This is a new inode, and
1984          * the VFS doesn't zero times in the core inode so we have to do
1985          * it ourselves.  They will be overwritten by either MDS or OST
1986          * attributes - we just need to make sure they aren't newer. */
1987         LTIME_S(inode->i_mtime) = 0;
1988         LTIME_S(inode->i_atime) = 0;
1989         LTIME_S(inode->i_ctime) = 0;
1990         inode->i_rdev = 0;
1991         rc = ll_update_inode(inode, md);
1992         if (rc != 0)
1993                 RETURN(rc);
1994
1995         /* OIDEBUG(inode); */
1996
1997 #ifdef HAVE_BACKING_DEV_INFO
1998         /* initializing backing dev info. */
1999         inode->i_mapping->backing_dev_info = &s2lsi(inode->i_sb)->lsi_bdi;
2000 #endif
2001         if (S_ISREG(inode->i_mode)) {
2002                 struct ll_sb_info *sbi = ll_i2sbi(inode);
2003                 inode->i_op = &ll_file_inode_operations;
2004                 inode->i_fop = sbi->ll_fop;
2005                 inode->i_mapping->a_ops = (struct address_space_operations *)&ll_aops;
2006                 EXIT;
2007         } else if (S_ISDIR(inode->i_mode)) {
2008                 inode->i_op = &ll_dir_inode_operations;
2009                 inode->i_fop = &ll_dir_operations;
2010                 EXIT;
2011         } else if (S_ISLNK(inode->i_mode)) {
2012                 inode->i_op = &ll_fast_symlink_inode_operations;
2013                 EXIT;
2014         } else {
2015                 inode->i_op = &ll_special_inode_operations;
2016
2017                 init_special_inode(inode, inode->i_mode,
2018                                    inode->i_rdev);
2019
2020                 EXIT;
2021         }
2022
2023         return 0;
2024 }
2025
2026 void ll_delete_inode(struct inode *inode)
2027 {
2028         struct ll_inode_info *lli = ll_i2info(inode);
2029         ENTRY;
2030
2031         if (S_ISREG(inode->i_mode) && lli->lli_clob != NULL)
2032                 /* It is last chance to write out dirty pages,
2033                  * otherwise we may lose data while umount */
2034                 cl_sync_file_range(inode, 0, OBD_OBJECT_EOF, CL_FSYNC_LOCAL, 1);
2035
2036         truncate_inode_pages_final(&inode->i_data);
2037
2038         LASSERTF(inode->i_data.nrpages == 0, "inode="DFID"(%p) nrpages=%lu, "
2039                  "see https://jira.hpdd.intel.com/browse/LU-118\n",
2040                  PFID(ll_inode2fid(inode)), inode, inode->i_data.nrpages);
2041
2042 #ifdef HAVE_SBOPS_EVICT_INODE
2043         ll_clear_inode(inode);
2044 #endif
2045         clear_inode(inode);
2046
2047         EXIT;
2048 }
2049
2050 int ll_iocontrol(struct inode *inode, struct file *file,
2051                  unsigned int cmd, unsigned long arg)
2052 {
2053         struct ll_sb_info *sbi = ll_i2sbi(inode);
2054         struct ptlrpc_request *req = NULL;
2055         int rc, flags = 0;
2056         ENTRY;
2057
2058         switch(cmd) {
2059         case FSFILT_IOC_GETFLAGS: {
2060                 struct mdt_body *body;
2061                 struct md_op_data *op_data;
2062
2063                 op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL,
2064                                              0, 0, LUSTRE_OPC_ANY,
2065                                              NULL);
2066                 if (IS_ERR(op_data))
2067                         RETURN(PTR_ERR(op_data));
2068
2069                 op_data->op_valid = OBD_MD_FLFLAGS;
2070                 rc = md_getattr(sbi->ll_md_exp, op_data, &req);
2071                 ll_finish_md_op_data(op_data);
2072                 if (rc) {
2073                         CERROR("%s: failure inode "DFID": rc = %d\n",
2074                                sbi->ll_md_exp->exp_obd->obd_name,
2075                                PFID(ll_inode2fid(inode)), rc);
2076                         RETURN(-abs(rc));
2077                 }
2078
2079                 body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
2080
2081                 flags = body->mbo_flags;
2082
2083                 ptlrpc_req_finished(req);
2084
2085                 RETURN(put_user(flags, (int __user *)arg));
2086         }
2087         case FSFILT_IOC_SETFLAGS: {
2088                 struct iattr *attr;
2089                 struct md_op_data *op_data;
2090                 struct cl_object *obj;
2091
2092                 if (get_user(flags, (int __user *)arg))
2093                         RETURN(-EFAULT);
2094
2095                 op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL, 0, 0,
2096                                              LUSTRE_OPC_ANY, NULL);
2097                 if (IS_ERR(op_data))
2098                         RETURN(PTR_ERR(op_data));
2099
2100                 op_data->op_attr_flags = flags;
2101                 op_data->op_attr.ia_valid |= ATTR_ATTR_FLAG;
2102                 rc = md_setattr(sbi->ll_md_exp, op_data, NULL, 0, &req);
2103                 ll_finish_md_op_data(op_data);
2104                 ptlrpc_req_finished(req);
2105                 if (rc)
2106                         RETURN(rc);
2107
2108                 inode->i_flags = ll_ext_to_inode_flags(flags);
2109
2110                 obj = ll_i2info(inode)->lli_clob;
2111                 if (obj == NULL)
2112                         RETURN(0);
2113
2114                 OBD_ALLOC_PTR(attr);
2115                 if (attr == NULL)
2116                         RETURN(-ENOMEM);
2117
2118                 attr->ia_valid = ATTR_ATTR_FLAG;
2119                 rc = cl_setattr_ost(obj, attr, flags);
2120
2121                 OBD_FREE_PTR(attr);
2122                 RETURN(rc);
2123         }
2124         default:
2125                 RETURN(-ENOSYS);
2126         }
2127
2128         RETURN(0);
2129 }
2130
2131 int ll_flush_ctx(struct inode *inode)
2132 {
2133         struct ll_sb_info  *sbi = ll_i2sbi(inode);
2134
2135         CDEBUG(D_SEC, "flush context for user %d\n",
2136                from_kuid(&init_user_ns, current_uid()));
2137
2138         obd_set_info_async(NULL, sbi->ll_md_exp,
2139                            sizeof(KEY_FLUSH_CTX), KEY_FLUSH_CTX,
2140                            0, NULL, NULL);
2141         obd_set_info_async(NULL, sbi->ll_dt_exp,
2142                            sizeof(KEY_FLUSH_CTX), KEY_FLUSH_CTX,
2143                            0, NULL, NULL);
2144         return 0;
2145 }
2146
2147 /* umount -f client means force down, don't save state */
2148 void ll_umount_begin(struct super_block *sb)
2149 {
2150         struct ll_sb_info *sbi = ll_s2sbi(sb);
2151         struct obd_device *obd;
2152         struct obd_ioctl_data *ioc_data;
2153         struct l_wait_info lwi;
2154         wait_queue_head_t waitq;
2155         ENTRY;
2156
2157         CDEBUG(D_VFSTRACE, "VFS Op: superblock %p count %d active %d\n", sb,
2158                sb->s_count, atomic_read(&sb->s_active));
2159
2160         obd = class_exp2obd(sbi->ll_md_exp);
2161         if (obd == NULL) {
2162                 CERROR("Invalid MDC connection handle %#llx\n",
2163                        sbi->ll_md_exp->exp_handle.h_cookie);
2164                 EXIT;
2165                 return;
2166         }
2167         obd->obd_force = 1;
2168
2169         obd = class_exp2obd(sbi->ll_dt_exp);
2170         if (obd == NULL) {
2171                 CERROR("Invalid LOV connection handle %#llx\n",
2172                        sbi->ll_dt_exp->exp_handle.h_cookie);
2173                 EXIT;
2174                 return;
2175         }
2176         obd->obd_force = 1;
2177
2178         OBD_ALLOC_PTR(ioc_data);
2179         if (ioc_data) {
2180                 obd_iocontrol(IOC_OSC_SET_ACTIVE, sbi->ll_md_exp,
2181                               sizeof *ioc_data, ioc_data, NULL);
2182
2183                 obd_iocontrol(IOC_OSC_SET_ACTIVE, sbi->ll_dt_exp,
2184                               sizeof *ioc_data, ioc_data, NULL);
2185
2186                 OBD_FREE_PTR(ioc_data);
2187         }
2188
2189         /* Really, we'd like to wait until there are no requests outstanding,
2190          * and then continue.  For now, we just periodically checking for vfs
2191          * to decrement mnt_cnt and hope to finish it within 10sec.
2192          */
2193         init_waitqueue_head(&waitq);
2194         lwi = LWI_TIMEOUT_INTERVAL(cfs_time_seconds(10),
2195                                    cfs_time_seconds(1), NULL, NULL);
2196         l_wait_event(waitq, may_umount(sbi->ll_mnt.mnt), &lwi);
2197
2198         EXIT;
2199 }
2200
2201 int ll_remount_fs(struct super_block *sb, int *flags, char *data)
2202 {
2203         struct ll_sb_info *sbi = ll_s2sbi(sb);
2204         char *profilenm = get_profile_name(sb);
2205         int err;
2206         __u32 read_only;
2207
2208         if ((*flags & MS_RDONLY) != (sb->s_flags & MS_RDONLY)) {
2209                 read_only = *flags & MS_RDONLY;
2210                 err = obd_set_info_async(NULL, sbi->ll_md_exp,
2211                                          sizeof(KEY_READ_ONLY),
2212                                          KEY_READ_ONLY, sizeof(read_only),
2213                                          &read_only, NULL);
2214                 if (err) {
2215                         LCONSOLE_WARN("Failed to remount %s %s (%d)\n",
2216                                       profilenm, read_only ?
2217                                       "read-only" : "read-write", err);
2218                         return err;
2219                 }
2220
2221                 if (read_only)
2222                         sb->s_flags |= MS_RDONLY;
2223                 else
2224                         sb->s_flags &= ~MS_RDONLY;
2225
2226                 if (sbi->ll_flags & LL_SBI_VERBOSE)
2227                         LCONSOLE_WARN("Remounted %s %s\n", profilenm,
2228                                       read_only ?  "read-only" : "read-write");
2229         }
2230         return 0;
2231 }
2232
2233 /**
2234  * Cleanup the open handle that is cached on MDT-side.
2235  *
2236  * For open case, the client side open handling thread may hit error
2237  * after the MDT grant the open. Under such case, the client should
2238  * send close RPC to the MDT as cleanup; otherwise, the open handle
2239  * on the MDT will be leaked there until the client umount or evicted.
2240  *
2241  * In further, if someone unlinked the file, because the open handle
2242  * holds the reference on such file/object, then it will block the
2243  * subsequent threads that want to locate such object via FID.
2244  *
2245  * \param[in] sb        super block for this file-system
2246  * \param[in] open_req  pointer to the original open request
2247  */
2248 void ll_open_cleanup(struct super_block *sb, struct ptlrpc_request *open_req)
2249 {
2250         struct mdt_body                 *body;
2251         struct md_op_data               *op_data;
2252         struct ptlrpc_request           *close_req = NULL;
2253         struct obd_export               *exp       = ll_s2sbi(sb)->ll_md_exp;
2254         ENTRY;
2255
2256         body = req_capsule_server_get(&open_req->rq_pill, &RMF_MDT_BODY);
2257         OBD_ALLOC_PTR(op_data);
2258         if (op_data == NULL) {
2259                 CWARN("%s: cannot allocate op_data to release open handle for "
2260                       DFID"\n",
2261                       ll_get_fsname(sb, NULL, 0), PFID(&body->mbo_fid1));
2262
2263                 RETURN_EXIT;
2264         }
2265
2266         op_data->op_fid1 = body->mbo_fid1;
2267         op_data->op_handle = body->mbo_handle;
2268         op_data->op_mod_time = ktime_get_real_seconds();
2269         md_close(exp, op_data, NULL, &close_req);
2270         ptlrpc_req_finished(close_req);
2271         ll_finish_md_op_data(op_data);
2272
2273         EXIT;
2274 }
2275
2276 int ll_prep_inode(struct inode **inode, struct ptlrpc_request *req,
2277                   struct super_block *sb, struct lookup_intent *it)
2278 {
2279         struct ll_sb_info *sbi = NULL;
2280         struct lustre_md md = { NULL };
2281         int rc;
2282         ENTRY;
2283
2284         LASSERT(*inode || sb);
2285         sbi = sb ? ll_s2sbi(sb) : ll_i2sbi(*inode);
2286         rc = md_get_lustre_md(sbi->ll_md_exp, req, sbi->ll_dt_exp,
2287                               sbi->ll_md_exp, &md);
2288         if (rc != 0)
2289                 GOTO(cleanup, rc);
2290
2291         if (*inode) {
2292                 rc = ll_update_inode(*inode, &md);
2293                 if (rc != 0)
2294                         GOTO(out, rc);
2295         } else {
2296                 LASSERT(sb != NULL);
2297
2298                 /*
2299                  * At this point server returns to client's same fid as client
2300                  * generated for creating. So using ->fid1 is okay here.
2301                  */
2302                 if (!fid_is_sane(&md.body->mbo_fid1)) {
2303                         CERROR("%s: Fid is insane "DFID"\n",
2304                                 ll_get_fsname(sb, NULL, 0),
2305                                 PFID(&md.body->mbo_fid1));
2306                         GOTO(out, rc = -EINVAL);
2307                 }
2308
2309                 *inode = ll_iget(sb, cl_fid_build_ino(&md.body->mbo_fid1,
2310                                              sbi->ll_flags & LL_SBI_32BIT_API),
2311                                  &md);
2312                 if (IS_ERR(*inode)) {
2313 #ifdef CONFIG_FS_POSIX_ACL
2314                         if (md.posix_acl) {
2315                                 posix_acl_release(md.posix_acl);
2316                                 md.posix_acl = NULL;
2317                         }
2318 #endif
2319                         rc = IS_ERR(*inode) ? PTR_ERR(*inode) : -ENOMEM;
2320                         *inode = NULL;
2321                         CERROR("new_inode -fatal: rc %d\n", rc);
2322                         GOTO(out, rc);
2323                 }
2324         }
2325
2326         /* Handling piggyback layout lock.
2327          * Layout lock can be piggybacked by getattr and open request.
2328          * The lsm can be applied to inode only if it comes with a layout lock
2329          * otherwise correct layout may be overwritten, for example:
2330          * 1. proc1: mdt returns a lsm but not granting layout
2331          * 2. layout was changed by another client
2332          * 3. proc2: refresh layout and layout lock granted
2333          * 4. proc1: to apply a stale layout */
2334         if (it != NULL && it->it_lock_mode != 0) {
2335                 struct lustre_handle lockh;
2336                 struct ldlm_lock *lock;
2337
2338                 lockh.cookie = it->it_lock_handle;
2339                 lock = ldlm_handle2lock(&lockh);
2340                 LASSERT(lock != NULL);
2341                 if (ldlm_has_layout(lock)) {
2342                         struct cl_object_conf conf;
2343
2344                         memset(&conf, 0, sizeof(conf));
2345                         conf.coc_opc = OBJECT_CONF_SET;
2346                         conf.coc_inode = *inode;
2347                         conf.coc_lock = lock;
2348                         conf.u.coc_layout = md.layout;
2349                         (void)ll_layout_conf(*inode, &conf);
2350                 }
2351                 LDLM_LOCK_PUT(lock);
2352         }
2353
2354         GOTO(out, rc = 0);
2355
2356 out:
2357         md_free_lustre_md(sbi->ll_md_exp, &md);
2358
2359 cleanup:
2360         if (rc != 0 && it != NULL && it->it_op & IT_OPEN)
2361                 ll_open_cleanup(sb != NULL ? sb : (*inode)->i_sb, req);
2362
2363         return rc;
2364 }
2365
2366 int ll_obd_statfs(struct inode *inode, void __user *arg)
2367 {
2368         struct ll_sb_info *sbi = NULL;
2369         struct obd_export *exp;
2370         char *buf = NULL;
2371         struct obd_ioctl_data *data = NULL;
2372         __u32 type;
2373         int len = 0, rc;
2374
2375         if (!inode || !(sbi = ll_i2sbi(inode)))
2376                 GOTO(out_statfs, rc = -EINVAL);
2377
2378         rc = obd_ioctl_getdata(&buf, &len, arg);
2379         if (rc)
2380                 GOTO(out_statfs, rc);
2381
2382         data = (void*)buf;
2383         if (!data->ioc_inlbuf1 || !data->ioc_inlbuf2 ||
2384             !data->ioc_pbuf1 || !data->ioc_pbuf2)
2385                 GOTO(out_statfs, rc = -EINVAL);
2386
2387         if (data->ioc_inllen1 != sizeof(__u32) ||
2388             data->ioc_inllen2 != sizeof(__u32) ||
2389             data->ioc_plen1 != sizeof(struct obd_statfs) ||
2390             data->ioc_plen2 != sizeof(struct obd_uuid))
2391                 GOTO(out_statfs, rc = -EINVAL);
2392
2393         memcpy(&type, data->ioc_inlbuf1, sizeof(__u32));
2394         if (type & LL_STATFS_LMV)
2395                 exp = sbi->ll_md_exp;
2396         else if (type & LL_STATFS_LOV)
2397                 exp = sbi->ll_dt_exp;
2398         else
2399                 GOTO(out_statfs, rc = -ENODEV);
2400
2401         rc = obd_iocontrol(IOC_OBD_STATFS, exp, len, buf, NULL);
2402         if (rc)
2403                 GOTO(out_statfs, rc);
2404 out_statfs:
2405         OBD_FREE_LARGE(buf, len);
2406         return rc;
2407 }
2408
2409 int ll_process_config(struct lustre_cfg *lcfg)
2410 {
2411         struct super_block *sb;
2412         unsigned long x;
2413         int rc = 0;
2414         char *ptr;
2415
2416         /* The instance name contains the sb: lustre-client-aacfe000 */
2417         ptr = strrchr(lustre_cfg_string(lcfg, 0), '-');
2418         if (!ptr || !*(++ptr))
2419                 return -EINVAL;
2420         if (sscanf(ptr, "%lx", &x) != 1)
2421                 return -EINVAL;
2422         sb = (struct super_block *)x;
2423         /* This better be a real Lustre superblock! */
2424         LASSERT(s2lsi(sb)->lsi_lmd->lmd_magic == LMD_MAGIC);
2425
2426         /* Note we have not called client_common_fill_super yet, so
2427            proc fns must be able to handle that! */
2428         rc = class_process_proc_param(PARAM_LLITE, lprocfs_llite_obd_vars,
2429                                       lcfg, sb);
2430         if (rc > 0)
2431                 rc = 0;
2432         return rc;
2433 }
2434
2435 /* this function prepares md_op_data hint for passing it down to MD stack. */
2436 struct md_op_data *ll_prep_md_op_data(struct md_op_data *op_data,
2437                                       struct inode *i1, struct inode *i2,
2438                                       const char *name, size_t namelen,
2439                                       __u32 mode, __u32 opc, void *data)
2440 {
2441         LASSERT(i1 != NULL);
2442
2443         if (name == NULL) {
2444                 /* Do not reuse namelen for something else. */
2445                 if (namelen != 0)
2446                         return ERR_PTR(-EINVAL);
2447         } else {
2448                 if (namelen > ll_i2sbi(i1)->ll_namelen)
2449                         return ERR_PTR(-ENAMETOOLONG);
2450
2451                 if (!lu_name_is_valid_2(name, namelen))
2452                         return ERR_PTR(-EINVAL);
2453         }
2454
2455         if (op_data == NULL)
2456                 OBD_ALLOC_PTR(op_data);
2457
2458         if (op_data == NULL)
2459                 return ERR_PTR(-ENOMEM);
2460
2461         ll_i2gids(op_data->op_suppgids, i1, i2);
2462         op_data->op_fid1 = *ll_inode2fid(i1);
2463         op_data->op_default_stripe_offset = -1;
2464         if (S_ISDIR(i1->i_mode)) {
2465                 op_data->op_mea1 = ll_i2info(i1)->lli_lsm_md;
2466                 if (opc == LUSTRE_OPC_MKDIR)
2467                         op_data->op_default_stripe_offset =
2468                                    ll_i2info(i1)->lli_def_stripe_offset;
2469         }
2470
2471         if (i2) {
2472                 op_data->op_fid2 = *ll_inode2fid(i2);
2473                 if (S_ISDIR(i2->i_mode))
2474                         op_data->op_mea2 = ll_i2info(i2)->lli_lsm_md;
2475         } else {
2476                 fid_zero(&op_data->op_fid2);
2477         }
2478
2479         if (ll_i2sbi(i1)->ll_flags & LL_SBI_64BIT_HASH)
2480                 op_data->op_cli_flags |= CLI_HASH64;
2481
2482         if (ll_need_32bit_api(ll_i2sbi(i1)))
2483                 op_data->op_cli_flags |= CLI_API32;
2484
2485         op_data->op_name = name;
2486         op_data->op_namelen = namelen;
2487         op_data->op_mode = mode;
2488         op_data->op_mod_time = cfs_time_current_sec();
2489         op_data->op_fsuid = from_kuid(&init_user_ns, current_fsuid());
2490         op_data->op_fsgid = from_kgid(&init_user_ns, current_fsgid());
2491         op_data->op_cap = cfs_curproc_cap_pack();
2492         if ((opc == LUSTRE_OPC_CREATE) && (name != NULL) &&
2493              filename_is_volatile(name, namelen, &op_data->op_mds)) {
2494                 op_data->op_bias |= MDS_CREATE_VOLATILE;
2495         } else {
2496                 op_data->op_mds = 0;
2497         }
2498         op_data->op_data = data;
2499
2500         return op_data;
2501 }
2502
2503 void ll_finish_md_op_data(struct md_op_data *op_data)
2504 {
2505         security_release_secctx(op_data->op_file_secctx,
2506                                 op_data->op_file_secctx_size);
2507         OBD_FREE_PTR(op_data);
2508 }
2509
2510 #ifdef HAVE_SUPEROPS_USE_DENTRY
2511 int ll_show_options(struct seq_file *seq, struct dentry *dentry)
2512 #else
2513 int ll_show_options(struct seq_file *seq, struct vfsmount *vfs)
2514 #endif
2515 {
2516         struct ll_sb_info *sbi;
2517
2518 #ifdef HAVE_SUPEROPS_USE_DENTRY
2519         LASSERT((seq != NULL) && (dentry != NULL));
2520         sbi = ll_s2sbi(dentry->d_sb);
2521 #else
2522         LASSERT((seq != NULL) && (vfs != NULL));
2523         sbi = ll_s2sbi(vfs->mnt_sb);
2524 #endif
2525
2526         if (sbi->ll_flags & LL_SBI_NOLCK)
2527                 seq_puts(seq, ",nolock");
2528
2529         if (sbi->ll_flags & LL_SBI_FLOCK)
2530                 seq_puts(seq, ",flock");
2531
2532         if (sbi->ll_flags & LL_SBI_LOCALFLOCK)
2533                 seq_puts(seq, ",localflock");
2534
2535         if (sbi->ll_flags & LL_SBI_USER_XATTR)
2536                 seq_puts(seq, ",user_xattr");
2537
2538         if (sbi->ll_flags & LL_SBI_LAZYSTATFS)
2539                 seq_puts(seq, ",lazystatfs");
2540
2541         if (sbi->ll_flags & LL_SBI_USER_FID2PATH)
2542                 seq_puts(seq, ",user_fid2path");
2543
2544         if (sbi->ll_flags & LL_SBI_ALWAYS_PING)
2545                 seq_puts(seq, ",always_ping");
2546
2547         RETURN(0);
2548 }
2549
2550 /**
2551  * Get obd name by cmd, and copy out to user space
2552  */
2553 int ll_get_obd_name(struct inode *inode, unsigned int cmd, unsigned long arg)
2554 {
2555         struct ll_sb_info *sbi = ll_i2sbi(inode);
2556         struct obd_device *obd;
2557         ENTRY;
2558
2559         if (cmd == OBD_IOC_GETDTNAME)
2560                 obd = class_exp2obd(sbi->ll_dt_exp);
2561         else if (cmd == OBD_IOC_GETMDNAME)
2562                 obd = class_exp2obd(sbi->ll_md_exp);
2563         else
2564                 RETURN(-EINVAL);
2565
2566         if (!obd)
2567                 RETURN(-ENOENT);
2568
2569         if (copy_to_user((void __user *)arg, obd->obd_name,
2570                          strlen(obd->obd_name) + 1))
2571                 RETURN(-EFAULT);
2572
2573         RETURN(0);
2574 }
2575
2576 /**
2577  * Get lustre file system name by \a sbi. If \a buf is provided(non-NULL), the
2578  * fsname will be returned in this buffer; otherwise, a static buffer will be
2579  * used to store the fsname and returned to caller.
2580  */
2581 char *ll_get_fsname(struct super_block *sb, char *buf, int buflen)
2582 {
2583         static char fsname_static[MTI_NAME_MAXLEN];
2584         struct lustre_sb_info *lsi = s2lsi(sb);
2585         char *ptr;
2586         int len;
2587
2588         if (buf == NULL) {
2589                 /* this means the caller wants to use static buffer
2590                  * and it doesn't care about race. Usually this is
2591                  * in error reporting path */
2592                 buf = fsname_static;
2593                 buflen = sizeof(fsname_static);
2594         }
2595
2596         len = strlen(lsi->lsi_lmd->lmd_profile);
2597         ptr = strrchr(lsi->lsi_lmd->lmd_profile, '-');
2598         if (ptr && (strcmp(ptr, "-client") == 0))
2599                 len -= 7;
2600
2601         if (unlikely(len >= buflen))
2602                 len = buflen - 1;
2603         strncpy(buf, lsi->lsi_lmd->lmd_profile, len);
2604         buf[len] = '\0';
2605
2606         return buf;
2607 }
2608
2609 static char* ll_d_path(struct dentry *dentry, char *buf, int bufsize)
2610 {
2611         char *path = NULL;
2612
2613         struct path p;
2614
2615         p.dentry = dentry;
2616         p.mnt = current->fs->root.mnt;
2617         path_get(&p);
2618         path = d_path(&p, buf, bufsize);
2619         path_put(&p);
2620         return path;
2621 }
2622
2623 void ll_dirty_page_discard_warn(struct page *page, int ioret)
2624 {
2625         char *buf, *path = NULL;
2626         struct dentry *dentry = NULL;
2627         struct inode *inode = page->mapping->host;
2628
2629         /* this can be called inside spin lock so use GFP_ATOMIC. */
2630         buf = (char *)__get_free_page(GFP_ATOMIC);
2631         if (buf != NULL) {
2632                 dentry = d_find_alias(page->mapping->host);
2633                 if (dentry != NULL)
2634                         path = ll_d_path(dentry, buf, PAGE_SIZE);
2635         }
2636
2637         CDEBUG(D_WARNING,
2638                "%s: dirty page discard: %s/fid: "DFID"/%s may get corrupted "
2639                "(rc %d)\n", ll_get_fsname(page->mapping->host->i_sb, NULL, 0),
2640                s2lsi(page->mapping->host->i_sb)->lsi_lmd->lmd_dev,
2641                PFID(ll_inode2fid(inode)),
2642                (path && !IS_ERR(path)) ? path : "", ioret);
2643
2644         if (dentry != NULL)
2645                 dput(dentry);
2646
2647         if (buf != NULL)
2648                 free_page((unsigned long)buf);
2649 }
2650
2651 ssize_t ll_copy_user_md(const struct lov_user_md __user *md,
2652                         struct lov_user_md **kbuf)
2653 {
2654         struct lov_user_md      lum;
2655         ssize_t                 lum_size;
2656         ENTRY;
2657
2658         if (copy_from_user(&lum, md, sizeof(lum)))
2659                 RETURN(-EFAULT);
2660
2661         lum_size = ll_lov_user_md_size(&lum);
2662         if (lum_size < 0)
2663                 RETURN(lum_size);
2664
2665         OBD_ALLOC(*kbuf, lum_size);
2666         if (*kbuf == NULL)
2667                 RETURN(-ENOMEM);
2668
2669         if (copy_from_user(*kbuf, md, lum_size) != 0) {
2670                 OBD_FREE(*kbuf, lum_size);
2671                 RETURN(-EFAULT);
2672         }
2673
2674         RETURN(lum_size);
2675 }
2676
2677 /*
2678  * Compute llite root squash state after a change of root squash
2679  * configuration setting or add/remove of a lnet nid
2680  */
2681 void ll_compute_rootsquash_state(struct ll_sb_info *sbi)
2682 {
2683         struct root_squash_info *squash = &sbi->ll_squash;
2684         int i;
2685         bool matched;
2686         struct lnet_process_id id;
2687
2688         /* Update norootsquash flag */
2689         down_write(&squash->rsi_sem);
2690         if (list_empty(&squash->rsi_nosquash_nids))
2691                 sbi->ll_flags &= ~LL_SBI_NOROOTSQUASH;
2692         else {
2693                 /* Do not apply root squash as soon as one of our NIDs is
2694                  * in the nosquash_nids list */
2695                 matched = false;
2696                 i = 0;
2697                 while (LNetGetId(i++, &id) != -ENOENT) {
2698                         if (LNET_NETTYP(LNET_NIDNET(id.nid)) == LOLND)
2699                                 continue;
2700                         if (cfs_match_nid(id.nid, &squash->rsi_nosquash_nids)) {
2701                                 matched = true;
2702                                 break;
2703                         }
2704                 }
2705                 if (matched)
2706                         sbi->ll_flags |= LL_SBI_NOROOTSQUASH;
2707                 else
2708                         sbi->ll_flags &= ~LL_SBI_NOROOTSQUASH;
2709         }
2710         up_write(&squash->rsi_sem);
2711 }
2712
2713 /**
2714  * Parse linkea content to extract information about a given hardlink
2715  *
2716  * \param[in]   ldata      - Initialized linkea data
2717  * \param[in]   linkno     - Link identifier
2718  * \param[out]  parent_fid - The entry's parent FID
2719  * \param[out]  ln         - Entry name destination buffer
2720  *
2721  * \retval 0 on success
2722  * \retval Appropriate negative error code on failure
2723  */
2724 static int ll_linkea_decode(struct linkea_data *ldata, unsigned int linkno,
2725                             struct lu_fid *parent_fid, struct lu_name *ln)
2726 {
2727         unsigned int    idx;
2728         int             rc;
2729         ENTRY;
2730
2731         rc = linkea_init_with_rec(ldata);
2732         if (rc < 0)
2733                 RETURN(rc);
2734
2735         if (linkno >= ldata->ld_leh->leh_reccount)
2736                 /* beyond last link */
2737                 RETURN(-ENODATA);
2738
2739         linkea_first_entry(ldata);
2740         for (idx = 0; ldata->ld_lee != NULL; idx++) {
2741                 linkea_entry_unpack(ldata->ld_lee, &ldata->ld_reclen, ln,
2742                                     parent_fid);
2743                 if (idx == linkno)
2744                         break;
2745
2746                 linkea_next_entry(ldata);
2747         }
2748
2749         if (idx < linkno)
2750                 RETURN(-ENODATA);
2751
2752         RETURN(0);
2753 }
2754
2755 /**
2756  * Get parent FID and name of an identified link. Operation is performed for
2757  * a given link number, letting the caller iterate over linkno to list one or
2758  * all links of an entry.
2759  *
2760  * \param[in]     file - File descriptor against which to perform the operation
2761  * \param[in,out] arg  - User-filled structure containing the linkno to operate
2762  *                       on and the available size. It is eventually filled with
2763  *                       the requested information or left untouched on error
2764  *
2765  * \retval - 0 on success
2766  * \retval - Appropriate negative error code on failure
2767  */
2768 int ll_getparent(struct file *file, struct getparent __user *arg)
2769 {
2770         struct inode            *inode = file_inode(file);
2771         struct linkea_data      *ldata;
2772         struct lu_buf            buf = LU_BUF_NULL;
2773         struct lu_name           ln;
2774         struct lu_fid            parent_fid;
2775         __u32                    linkno;
2776         __u32                    name_size;
2777         int                      rc;
2778
2779         ENTRY;
2780
2781         if (!cfs_capable(CFS_CAP_DAC_READ_SEARCH) &&
2782             !(ll_i2sbi(inode)->ll_flags & LL_SBI_USER_FID2PATH))
2783                 RETURN(-EPERM);
2784
2785         if (get_user(name_size, &arg->gp_name_size))
2786                 RETURN(-EFAULT);
2787
2788         if (get_user(linkno, &arg->gp_linkno))
2789                 RETURN(-EFAULT);
2790
2791         if (name_size > PATH_MAX)
2792                 RETURN(-EINVAL);
2793
2794         OBD_ALLOC(ldata, sizeof(*ldata));
2795         if (ldata == NULL)
2796                 RETURN(-ENOMEM);
2797
2798         rc = linkea_data_new(ldata, &buf);
2799         if (rc < 0)
2800                 GOTO(ldata_free, rc);
2801
2802 #ifdef HAVE_XATTR_HANDLER_FLAGS
2803         rc = ll_xattr_list(inode, XATTR_NAME_LINK, XATTR_TRUSTED_T, buf.lb_buf,
2804                            buf.lb_len, OBD_MD_FLXATTR);
2805 #else
2806         rc = ll_getxattr(file_dentry(file), XATTR_NAME_LINK, buf.lb_buf,
2807                          buf.lb_len);
2808 #endif /* HAVE_XATTR_HANDLER_FLAGS */
2809         if (rc < 0)
2810                 GOTO(lb_free, rc);
2811
2812         rc = ll_linkea_decode(ldata, linkno, &parent_fid, &ln);
2813         if (rc < 0)
2814                 GOTO(lb_free, rc);
2815
2816         if (ln.ln_namelen >= name_size)
2817                 GOTO(lb_free, rc = -EOVERFLOW);
2818
2819         if (copy_to_user(&arg->gp_fid, &parent_fid, sizeof(arg->gp_fid)))
2820                 GOTO(lb_free, rc = -EFAULT);
2821
2822         if (copy_to_user(&arg->gp_name, ln.ln_name, ln.ln_namelen))
2823                 GOTO(lb_free, rc = -EFAULT);
2824
2825         if (put_user('\0', arg->gp_name + ln.ln_namelen))
2826                 GOTO(lb_free, rc = -EFAULT);
2827
2828 lb_free:
2829         lu_buf_free(&buf);
2830 ldata_free:
2831         OBD_FREE(ldata, sizeof(*ldata));
2832
2833         RETURN(rc);
2834 }