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  | OBD_CONNECT_GRANT |
 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 | OBD_CONNECT_CKSUM |
 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         data->ocd_cksum_types = cksum_types_supported_client();
 232
 233         if (OBD_FAIL_CHECK(OBD_FAIL_MDC_LIGHTWEIGHT))
 234                 /* flag mdc connection as lightweight, only used for test
 235                  * purpose, use with care */
 236                 data->ocd_connect_flags |= OBD_CONNECT_LIGHTWEIGHT;
 237
 238         data->ocd_ibits_known = MDS_INODELOCK_FULL;
 239         data->ocd_version = LUSTRE_VERSION_CODE;
 240
 241         if (sb->s_flags & MS_RDONLY)
 242                 data->ocd_connect_flags |= OBD_CONNECT_RDONLY;
 243         if (sbi->ll_flags & LL_SBI_USER_XATTR)
 244                 data->ocd_connect_flags |= OBD_CONNECT_XATTR;
 245
 246 #ifdef MS_NOSEC
 247         /* Setting this indicates we correctly support S_NOSEC (See kernel
 248          * commit 9e1f1de02c2275d7172e18dc4e7c2065777611bf)
 249          */
 250         sb->s_flags |= MS_NOSEC;
 251 #endif
 252
 253         if (sbi->ll_flags & LL_SBI_FLOCK)
 254                 sbi->ll_fop = &ll_file_operations_flock;
 255         else if (sbi->ll_flags & LL_SBI_LOCALFLOCK)
 256                 sbi->ll_fop = &ll_file_operations;
 257         else
 258                 sbi->ll_fop = &ll_file_operations_noflock;
 259
 260         /* always ping even if server suppress_pings */
 261         if (sbi->ll_flags & LL_SBI_ALWAYS_PING)
 262                 data->ocd_connect_flags &= ~OBD_CONNECT_PINGLESS;
 263
 264 #ifdef HAVE_SECURITY_DENTRY_INIT_SECURITY
 265         data->ocd_connect_flags2 |= OBD_CONNECT2_FILE_SECCTX;
 266 #endif /* HAVE_SECURITY_DENTRY_INIT_SECURITY */
 267
 268         data->ocd_brw_size = MD_MAX_BRW_SIZE;
 269
 270         err = obd_connect(NULL, &sbi->ll_md_exp, obd, &sbi->ll_sb_uuid, data, NULL);
 271         if (err == -EBUSY) {
 272                 LCONSOLE_ERROR_MSG(0x14f, "An MDT (md %s) is performing "
 273                                    "recovery, of which this client is not a "
 274                                    "part. Please wait for recovery to complete,"
 275                                    " abort, or time out.\n", md);
 276                 GOTO(out, err);
 277         } else if (err) {
 278                 CERROR("cannot connect to %s: rc = %d\n", md, err);
 279                 GOTO(out, err);
 280         }
 281
 282         sbi->ll_md_exp->exp_connect_data = *data;
 283
 284         err = obd_fid_init(sbi->ll_md_exp->exp_obd, sbi->ll_md_exp,
 285                            LUSTRE_SEQ_METADATA);
 286         if (err) {
 287                 CERROR("%s: Can't init metadata layer FID infrastructure, "
 288                        "rc = %d\n", sbi->ll_md_exp->exp_obd->obd_name, err);
 289                 GOTO(out_md, err);
 290         }
 291
 292         /* For mount, we only need fs info from MDT0, and also in DNE, it
 293          * can make sure the client can be mounted as long as MDT0 is
 294          * avaible */
 295         err = obd_statfs(NULL, sbi->ll_md_exp, osfs,
 296                         cfs_time_shift_64(-OBD_STATFS_CACHE_SECONDS),
 297                         OBD_STATFS_FOR_MDT0);
 298         if (err)
 299                 GOTO(out_md_fid, err);
 300
 301         /* This needs to be after statfs to ensure connect has finished.
 302          * Note that "data" does NOT contain the valid connect reply.
 303          * If connecting to a 1.8 server there will be no LMV device, so
 304          * we can access the MDC export directly and exp_connect_flags will
 305          * be non-zero, but if accessing an upgraded 2.1 server it will
 306          * have the correct flags filled in.
 307          * XXX: fill in the LMV exp_connect_flags from MDC(s). */
 308         valid = exp_connect_flags(sbi->ll_md_exp) & CLIENT_CONNECT_MDT_REQD;
 309         if (exp_connect_flags(sbi->ll_md_exp) != 0 &&
 310             valid != CLIENT_CONNECT_MDT_REQD) {
 311                 char *buf;
 312
 313                 OBD_ALLOC_WAIT(buf, PAGE_SIZE);
 314                 obd_connect_flags2str(buf, PAGE_SIZE,
 315                                       valid ^ CLIENT_CONNECT_MDT_REQD, 0, ",");
 316                 LCONSOLE_ERROR_MSG(0x170, "Server %s does not support "
 317                                    "feature(s) needed for correct operation "
 318                                    "of this client (%s). Please upgrade "
 319                                    "server or downgrade client.\n",
 320                                    sbi->ll_md_exp->exp_obd->obd_name, buf);
 321                 OBD_FREE(buf, PAGE_SIZE);
 322                 GOTO(out_md_fid, err = -EPROTO);
 323         }
 324
 325         size = sizeof(*data);
 326         err = obd_get_info(NULL, sbi->ll_md_exp, sizeof(KEY_CONN_DATA),
 327                            KEY_CONN_DATA,  &size, data);
 328         if (err) {
 329                 CERROR("%s: Get connect data failed: rc = %d\n",
 330                        sbi->ll_md_exp->exp_obd->obd_name, err);
 331                 GOTO(out_md_fid, err);
 332         }
 333
 334         LASSERT(osfs->os_bsize);
 335         sb->s_blocksize = osfs->os_bsize;
 336         sb->s_blocksize_bits = log2(osfs->os_bsize);
 337         sb->s_magic = LL_SUPER_MAGIC;
 338         sb->s_maxbytes = MAX_LFS_FILESIZE;
 339         sbi->ll_namelen = osfs->os_namelen;
 340         sbi->ll_mnt.mnt = current->fs->root.mnt;
 341
 342         if ((sbi->ll_flags & LL_SBI_USER_XATTR) &&
 343             !(data->ocd_connect_flags & OBD_CONNECT_XATTR)) {
 344                 LCONSOLE_INFO("Disabling user_xattr feature because "
 345                               "it is not supported on the server\n");
 346                 sbi->ll_flags &= ~LL_SBI_USER_XATTR;
 347         }
 348
 349         if (data->ocd_connect_flags & OBD_CONNECT_ACL) {
 350 #ifdef MS_POSIXACL
 351                 sb->s_flags |= MS_POSIXACL;
 352 #endif
 353                 sbi->ll_flags |= LL_SBI_ACL;
 354         } else {
 355                 LCONSOLE_INFO("client wants to enable acl, but mdt not!\n");
 356 #ifdef MS_POSIXACL
 357                 sb->s_flags &= ~MS_POSIXACL;
 358 #endif
 359                 sbi->ll_flags &= ~LL_SBI_ACL;
 360         }
 361
 362         if (data->ocd_connect_flags & OBD_CONNECT_64BITHASH)
 363                 sbi->ll_flags |= LL_SBI_64BIT_HASH;
 364
 365         if (data->ocd_connect_flags & OBD_CONNECT_LAYOUTLOCK)
 366                 sbi->ll_flags |= LL_SBI_LAYOUT_LOCK;
 367
 368         if (obd_connect_has_secctx(data))
 369                 sbi->ll_flags |= LL_SBI_FILE_SECCTX;
 370
 371         if (data->ocd_ibits_known & MDS_INODELOCK_XATTR) {
 372                 if (!(data->ocd_connect_flags & OBD_CONNECT_MAX_EASIZE)) {
 373                         LCONSOLE_INFO("%s: disabling xattr cache due to "
 374                                       "unknown maximum xattr size.\n", dt);
 375                 } else if (!sbi->ll_xattr_cache_set) {
 376                         /* If xattr_cache is already set (no matter 0 or 1)
 377                          * during processing llog, it won't be enabled here. */
 378                         sbi->ll_flags |= LL_SBI_XATTR_CACHE;
 379                         sbi->ll_xattr_cache_enabled = 1;
 380                 }
 381         }
 382
 383         obd = class_name2obd(dt);
 384         if (!obd) {
 385                 CERROR("DT %s: not setup or attached\n", dt);
 386                 GOTO(out_md_fid, err = -ENODEV);
 387         }
 388
 389         /* pass client page size via ocd_grant_blkbits, the server should report
 390          * back its backend blocksize for grant calculation purpose */
 391         data->ocd_grant_blkbits = PAGE_SHIFT;
 392
 393         /* indicate OST features supported by this client */
 394         data->ocd_connect_flags = OBD_CONNECT_GRANT | OBD_CONNECT_VERSION |
 395                                   OBD_CONNECT_REQPORTAL | OBD_CONNECT_BRW_SIZE |
 396                                   OBD_CONNECT_CANCELSET | OBD_CONNECT_FID |
 397                                   OBD_CONNECT_SRVLOCK | OBD_CONNECT_TRUNCLOCK|
 398                                   OBD_CONNECT_AT | OBD_CONNECT_OSS_CAPA |
 399                                   OBD_CONNECT_VBR | OBD_CONNECT_FULL20 |
 400                                   OBD_CONNECT_64BITHASH | OBD_CONNECT_MAXBYTES |
 401                                   OBD_CONNECT_EINPROGRESS |
 402                                   OBD_CONNECT_JOBSTATS | OBD_CONNECT_LVB_TYPE |
 403                                   OBD_CONNECT_LAYOUTLOCK |
 404                                   OBD_CONNECT_PINGLESS | OBD_CONNECT_LFSCK |
 405                                   OBD_CONNECT_BULK_MBITS |
 406                                   OBD_CONNECT_FLAGS2;
 407
 408 /* The client currently advertises support for OBD_CONNECT_LOCKAHEAD_OLD so it
 409  * can interoperate with an older version of lockahead which was released prior
 410  * to landing in master. This support will be dropped when 2.13 development
 411  * starts.  At the point, we should not just drop the connect flag (below), we
 412  * should also remove the support in the code.
 413  *
 414  * Removing it means a few things:
 415  * 1. Remove this section here
 416  * 2. Remove CEF_NONBLOCK in ll_file_lockahead()
 417  * 3. Remove function exp_connect_lockahead_old
 418  * 4. Remove LDLM_FL_LOCKAHEAD_OLD_RESERVED in lustre_dlm_flags.h
 419  * */
 420 #if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(2, 12, 50, 0)
 421         data->ocd_connect_flags |= OBD_CONNECT_LOCKAHEAD_OLD;
 422 #endif
 423
 424         data->ocd_connect_flags2 = OBD_CONNECT2_LOCKAHEAD;
 425
 426         if (!OBD_FAIL_CHECK(OBD_FAIL_OSC_CONNECT_GRANT_PARAM))
 427                 data->ocd_connect_flags |= OBD_CONNECT_GRANT_PARAM;
 428
 429         /* OBD_CONNECT_CKSUM should always be set, even if checksums are
 430          * disabled by default, because it can still be enabled on the
 431          * fly via /proc. As a consequence, we still need to come to an
 432          * agreement on the supported algorithms at connect time */
 433         data->ocd_connect_flags |= OBD_CONNECT_CKSUM;
 434
 435         if (OBD_FAIL_CHECK(OBD_FAIL_OSC_CKSUM_ADLER_ONLY))
 436                 data->ocd_cksum_types = OBD_CKSUM_ADLER;
 437         else
 438                 data->ocd_cksum_types = cksum_types_supported_client();
 439
 440 #ifdef HAVE_LRU_RESIZE_SUPPORT
 441         data->ocd_connect_flags |= OBD_CONNECT_LRU_RESIZE;
 442 #endif
 443         /* always ping even if server suppress_pings */
 444         if (sbi->ll_flags & LL_SBI_ALWAYS_PING)
 445                 data->ocd_connect_flags &= ~OBD_CONNECT_PINGLESS;
 446
 447         CDEBUG(D_RPCTRACE, "ocd_connect_flags: %#llx ocd_version: %d "
 448                "ocd_grant: %d\n", data->ocd_connect_flags,
 449                data->ocd_version, data->ocd_grant);
 450
 451         obd->obd_upcall.onu_owner = &sbi->ll_lco;
 452         obd->obd_upcall.onu_upcall = cl_ocd_update;
 453
 454         data->ocd_brw_size = DT_MAX_BRW_SIZE;
 455
 456         err = obd_connect(NULL, &sbi->ll_dt_exp, obd, &sbi->ll_sb_uuid, data,
 457                           NULL);
 458         if (err == -EBUSY) {
 459                 LCONSOLE_ERROR_MSG(0x150, "An OST (dt %s) is performing "
 460                                    "recovery, of which this client is not a "
 461                                    "part.  Please wait for recovery to "
 462                                    "complete, abort, or time out.\n", dt);
 463                 GOTO(out_md, err);
 464         } else if (err) {
 465                 CERROR("%s: Cannot connect to %s: rc = %d\n",
 466                        sbi->ll_dt_exp->exp_obd->obd_name, dt, err);
 467                 GOTO(out_md, err);
 468         }
 469
 470         sbi->ll_dt_exp->exp_connect_data = *data;
 471
 472         err = obd_fid_init(sbi->ll_dt_exp->exp_obd, sbi->ll_dt_exp,
 473                            LUSTRE_SEQ_METADATA);
 474         if (err) {
 475                 CERROR("%s: Can't init data layer FID infrastructure, "
 476                        "rc = %d\n", sbi->ll_dt_exp->exp_obd->obd_name, err);
 477                 GOTO(out_dt, err);
 478         }
 479
 480         mutex_lock(&sbi->ll_lco.lco_lock);
 481         sbi->ll_lco.lco_flags = data->ocd_connect_flags;
 482         sbi->ll_lco.lco_md_exp = sbi->ll_md_exp;
 483         sbi->ll_lco.lco_dt_exp = sbi->ll_dt_exp;
 484         mutex_unlock(&sbi->ll_lco.lco_lock);
 485
 486         fid_zero(&sbi->ll_root_fid);
 487         err = md_get_root(sbi->ll_md_exp, get_mount_fileset(sb),
 488                            &sbi->ll_root_fid);
 489         if (err) {
 490                 CERROR("cannot mds_connect: rc = %d\n", err);
 491                 GOTO(out_lock_cn_cb, err);
 492         }
 493         if (!fid_is_sane(&sbi->ll_root_fid)) {
 494                 CERROR("%s: Invalid root fid "DFID" during mount\n",
 495                        sbi->ll_md_exp->exp_obd->obd_name,
 496                        PFID(&sbi->ll_root_fid));
 497                 GOTO(out_lock_cn_cb, err = -EINVAL);
 498         }
 499         CDEBUG(D_SUPER, "rootfid "DFID"\n", PFID(&sbi->ll_root_fid));
 500
 501         sb->s_op = &lustre_super_operations;
 502 #ifdef HAVE_XATTR_HANDLER_FLAGS
 503         sb->s_xattr = ll_xattr_handlers;
 504 #endif
 505 #if THREAD_SIZE >= 8192 /*b=17630*/
 506         sb->s_export_op = &lustre_export_operations;
 507 #endif
 508
 509         /* make root inode
 510          * XXX: move this to after cbd setup? */
 511         valid = OBD_MD_FLGETATTR | OBD_MD_FLBLOCKS | OBD_MD_FLMODEASIZE;
 512         if (sbi->ll_flags & LL_SBI_ACL)
 513                 valid |= OBD_MD_FLACL;
 514
 515         OBD_ALLOC_PTR(op_data);
 516         if (op_data == NULL)
 517                 GOTO(out_lock_cn_cb, err = -ENOMEM);
 518
 519         op_data->op_fid1 = sbi->ll_root_fid;
 520         op_data->op_mode = 0;
 521         op_data->op_valid = valid;
 522
 523         err = md_getattr(sbi->ll_md_exp, op_data, &request);
 524
 525         OBD_FREE_PTR(op_data);
 526         if (err) {
 527                 CERROR("%s: md_getattr failed for root: rc = %d\n",
 528                        sbi->ll_md_exp->exp_obd->obd_name, err);
 529                 GOTO(out_lock_cn_cb, err);
 530         }
 531
 532         err = md_get_lustre_md(sbi->ll_md_exp, request, sbi->ll_dt_exp,
 533                                sbi->ll_md_exp, &lmd);
 534         if (err) {
 535                 CERROR("failed to understand root inode md: rc = %d\n", err);
 536                 ptlrpc_req_finished(request);
 537                 GOTO(out_lock_cn_cb, err);
 538         }
 539
 540         LASSERT(fid_is_sane(&sbi->ll_root_fid));
 541         root = ll_iget(sb, cl_fid_build_ino(&sbi->ll_root_fid,
 542                                             sbi->ll_flags & LL_SBI_32BIT_API),
 543                        &lmd);
 544         md_free_lustre_md(sbi->ll_md_exp, &lmd);
 545         ptlrpc_req_finished(request);
 546
 547         if (IS_ERR(root)) {
 548 #ifdef CONFIG_FS_POSIX_ACL
 549                 if (lmd.posix_acl) {
 550                         posix_acl_release(lmd.posix_acl);
 551                         lmd.posix_acl = NULL;
 552                 }
 553 #endif
 554                 err = IS_ERR(root) ? PTR_ERR(root) : -EBADF;
 555                 root = NULL;
 556                 CERROR("lustre_lite: bad iget4 for root\n");
 557                 GOTO(out_root, err);
 558         }
 559
 560         checksum = sbi->ll_flags & LL_SBI_CHECKSUM;
 561         err = obd_set_info_async(NULL, sbi->ll_dt_exp, sizeof(KEY_CHECKSUM),
 562                                  KEY_CHECKSUM, sizeof(checksum), &checksum,
 563                                  NULL);
 564         if (err) {
 565                 CERROR("%s: Set checksum failed: rc = %d\n",
 566                        sbi->ll_dt_exp->exp_obd->obd_name, err);
 567                 GOTO(out_root, err);
 568         }
 569         cl_sb_init(sb);
 570
 571         err = obd_set_info_async(NULL, sbi->ll_dt_exp, sizeof(KEY_CACHE_SET),
 572                                  KEY_CACHE_SET, sizeof(*sbi->ll_cache),
 573                                  sbi->ll_cache, NULL);
 574         if (err) {
 575                 CERROR("%s: Set cache_set failed: rc = %d\n",
 576                        sbi->ll_dt_exp->exp_obd->obd_name, err);
 577                 GOTO(out_root, err);
 578         }
 579
 580         sb->s_root = d_make_root(root);
 581         if (sb->s_root == NULL) {
 582                 CERROR("%s: can't make root dentry\n",
 583                         ll_get_fsname(sb, NULL, 0));
 584                 GOTO(out_root, err = -ENOMEM);
 585         }
 586 #ifdef HAVE_DCACHE_LOCK
 587         sb->s_root->d_op = &ll_d_ops;
 588 #endif
 589
 590         sbi->ll_sdev_orig = sb->s_dev;
 591
 592         /* We set sb->s_dev equal on all lustre clients in order to support
 593          * NFS export clustering.  NFSD requires that the FSID be the same
 594          * on all clients. */
 595         /* s_dev is also used in lt_compare() to compare two fs, but that is
 596          * only a node-local comparison. */
 597         uuid = obd_get_uuid(sbi->ll_md_exp);
 598         if (uuid != NULL)
 599                 sb->s_dev = get_uuid2int(uuid->uuid, strlen(uuid->uuid));
 600
 601         if (data != NULL)
 602                 OBD_FREE_PTR(data);
 603         if (osfs != NULL)
 604                 OBD_FREE_PTR(osfs);
 605
 606         if (sbi->ll_proc_root != NULL) {
 607                 err = lprocfs_ll_register_obd(sb, dt);
 608                 if (err < 0) {
 609                         CERROR("%s: could not register %s in llite: rc = %d\n",
 610                                dt, ll_get_fsname(sb, NULL, 0), err);
 611                         err = 0;
 612                 }
 613                 err = lprocfs_ll_register_obd(sb, md);
 614                 if (err < 0) {
 615                         CERROR("%s: could not register %s in llite: rc = %d\n",
 616                                md, ll_get_fsname(sb, NULL, 0), err);
 617                         err = 0;
 618                 }
 619         }
 620
 621         RETURN(err);
 622 out_root:
 623         if (root)
 624                 iput(root);
 625 out_lock_cn_cb:
 626         obd_fid_fini(sbi->ll_dt_exp->exp_obd);
 627 out_dt:
 628         obd_disconnect(sbi->ll_dt_exp);
 629         sbi->ll_dt_exp = NULL;
 630 out_md_fid:
 631         obd_fid_fini(sbi->ll_md_exp->exp_obd);
 632 out_md:
 633         obd_disconnect(sbi->ll_md_exp);
 634         sbi->ll_md_exp = NULL;
 635 out:
 636         if (data != NULL)
 637                 OBD_FREE_PTR(data);
 638         if (osfs != NULL)
 639                 OBD_FREE_PTR(osfs);
 640         return err;
 641 }
 642
 643 int ll_get_max_mdsize(struct ll_sb_info *sbi, int *lmmsize)
 644 {
 645         int size, rc;
 646
 647         size = sizeof(*lmmsize);
 648         rc = obd_get_info(NULL, sbi->ll_dt_exp, sizeof(KEY_MAX_EASIZE),
 649                           KEY_MAX_EASIZE, &size, lmmsize);
 650         if (rc != 0) {
 651                 CERROR("%s: cannot get max LOV EA size: rc = %d\n",
 652                        sbi->ll_dt_exp->exp_obd->obd_name, rc);
 653                 RETURN(rc);
 654         }
 655
 656         size = sizeof(int);
 657         rc = obd_get_info(NULL, sbi->ll_md_exp, sizeof(KEY_MAX_EASIZE),
 658                           KEY_MAX_EASIZE, &size, lmmsize);
 659         if (rc)
 660                 CERROR("Get max mdsize error rc %d\n", rc);
 661
 662         RETURN(rc);
 663 }
 664
 665 /**
 666  * Get the value of the default_easize parameter.
 667  *
 668  * \see client_obd::cl_default_mds_easize
 669  *
 670  * \param[in] sbi       superblock info for this filesystem
 671  * \param[out] lmmsize  pointer to storage location for value
 672  *
 673  * \retval 0            on success
 674  * \retval negative     negated errno on failure
 675  */
 676 int ll_get_default_mdsize(struct ll_sb_info *sbi, int *lmmsize)
 677 {
 678         int size, rc;
 679
 680         size = sizeof(int);
 681         rc = obd_get_info(NULL, sbi->ll_md_exp, sizeof(KEY_DEFAULT_EASIZE),
 682                          KEY_DEFAULT_EASIZE, &size, lmmsize);
 683         if (rc)
 684                 CERROR("Get default mdsize error rc %d\n", rc);
 685
 686         RETURN(rc);
 687 }
 688
 689 /**
 690  * Set the default_easize parameter to the given value.
 691  *
 692  * \see client_obd::cl_default_mds_easize
 693  *
 694  * \param[in] sbi       superblock info for this filesystem
 695  * \param[in] lmmsize   the size to set
 696  *
 697  * \retval 0            on success
 698  * \retval negative     negated errno on failure
 699  */
 700 int ll_set_default_mdsize(struct ll_sb_info *sbi, int lmmsize)
 701 {
 702         int rc;
 703
 704         if (lmmsize < sizeof(struct lov_mds_md) ||
 705             lmmsize > OBD_MAX_DEFAULT_EA_SIZE)
 706                 return -EINVAL;
 707
 708         rc = obd_set_info_async(NULL, sbi->ll_md_exp,
 709                                 sizeof(KEY_DEFAULT_EASIZE), KEY_DEFAULT_EASIZE,
 710                                 sizeof(int), &lmmsize, NULL);
 711
 712         RETURN(rc);
 713 }
 714
 715 static void client_common_put_super(struct super_block *sb)
 716 {
 717         struct ll_sb_info *sbi = ll_s2sbi(sb);
 718         ENTRY;
 719
 720         cl_sb_fini(sb);
 721
 722         obd_fid_fini(sbi->ll_dt_exp->exp_obd);
 723         obd_disconnect(sbi->ll_dt_exp);
 724         sbi->ll_dt_exp = NULL;
 725
 726         lprocfs_ll_unregister_mountpoint(sbi);
 727
 728         obd_fid_fini(sbi->ll_md_exp->exp_obd);
 729         obd_disconnect(sbi->ll_md_exp);
 730         sbi->ll_md_exp = NULL;
 731
 732         EXIT;
 733 }
 734
 735 void ll_kill_super(struct super_block *sb)
 736 {
 737         struct ll_sb_info *sbi;
 738         ENTRY;
 739
 740         /* not init sb ?*/
 741         if (!(sb->s_flags & MS_ACTIVE))
 742                 return;
 743
 744         sbi = ll_s2sbi(sb);
 745         /* we need restore s_dev from changed for clustred NFS before put_super
 746          * because new kernels have cached s_dev and change sb->s_dev in
 747          * put_super not affected real removing devices */
 748         if (sbi) {
 749                 sb->s_dev = sbi->ll_sdev_orig;
 750                 sbi->ll_umounting = 1;
 751
 752                 /* wait running statahead threads to quit */
 753                 while (atomic_read(&sbi->ll_sa_running) > 0) {
 754                         set_current_state(TASK_UNINTERRUPTIBLE);
 755                         schedule_timeout(msecs_to_jiffies(MSEC_PER_SEC >> 3));
 756                 }
 757         }
 758
 759         EXIT;
 760 }
 761
 762 static inline int ll_set_opt(const char *opt, char *data, int fl)
 763 {
 764         if (strncmp(opt, data, strlen(opt)) != 0)
 765                 return(0);
 766         else
 767                 return(fl);
 768 }
 769
 770 /* non-client-specific mount options are parsed in lmd_parse */
 771 static int ll_options(char *options, int *flags)
 772 {
 773         int tmp;
 774         char *s1 = options, *s2;
 775         ENTRY;
 776
 777         if (!options)
 778                 RETURN(0);
 779
 780         CDEBUG(D_CONFIG, "Parsing opts %s\n", options);
 781
 782         while (*s1) {
 783                 CDEBUG(D_SUPER, "next opt=%s\n", s1);
 784                 tmp = ll_set_opt("nolock", s1, LL_SBI_NOLCK);
 785                 if (tmp) {
 786                         *flags |= tmp;
 787                         goto next;
 788                 }
 789                 tmp = ll_set_opt("flock", s1, LL_SBI_FLOCK);
 790                 if (tmp) {
 791                         *flags |= tmp;
 792                         goto next;
 793                 }
 794                 tmp = ll_set_opt("localflock", s1, LL_SBI_LOCALFLOCK);
 795                 if (tmp) {
 796                         *flags |= tmp;
 797                         goto next;
 798                 }
 799                 tmp = ll_set_opt("noflock", s1, LL_SBI_FLOCK|LL_SBI_LOCALFLOCK);
 800                 if (tmp) {
 801                         *flags &= ~tmp;
 802                         goto next;
 803                 }
 804                 tmp = ll_set_opt("user_xattr", s1, LL_SBI_USER_XATTR);
 805                 if (tmp) {
 806                         *flags |= tmp;
 807                         goto next;
 808                 }
 809                 tmp = ll_set_opt("nouser_xattr", s1, LL_SBI_USER_XATTR);
 810                 if (tmp) {
 811                         *flags &= ~tmp;
 812                         goto next;
 813                 }
 814                 tmp = ll_set_opt("context", s1, 1);
 815                 if (tmp)
 816                         goto next;
 817                 tmp = ll_set_opt("fscontext", s1, 1);
 818                 if (tmp)
 819                         goto next;
 820                 tmp = ll_set_opt("defcontext", s1, 1);
 821                 if (tmp)
 822                         goto next;
 823                 tmp = ll_set_opt("rootcontext", s1, 1);
 824                 if (tmp)
 825                         goto next;
 826                 tmp = ll_set_opt("user_fid2path", s1, LL_SBI_USER_FID2PATH);
 827                 if (tmp) {
 828                         *flags |= tmp;
 829                         goto next;
 830                 }
 831                 tmp = ll_set_opt("nouser_fid2path", s1, LL_SBI_USER_FID2PATH);
 832                 if (tmp) {
 833                         *flags &= ~tmp;
 834                         goto next;
 835                 }
 836
 837                 tmp = ll_set_opt("checksum", s1, LL_SBI_CHECKSUM);
 838                 if (tmp) {
 839                         *flags |= tmp;
 840                         goto next;
 841                 }
 842                 tmp = ll_set_opt("nochecksum", s1, LL_SBI_CHECKSUM);
 843                 if (tmp) {
 844                         *flags &= ~tmp;
 845                         goto next;
 846                 }
 847                 tmp = ll_set_opt("lruresize", s1, LL_SBI_LRU_RESIZE);
 848                 if (tmp) {
 849                         *flags |= tmp;
 850                         goto next;
 851                 }
 852                 tmp = ll_set_opt("nolruresize", s1, LL_SBI_LRU_RESIZE);
 853                 if (tmp) {
 854                         *flags &= ~tmp;
 855                         goto next;
 856                 }
 857                 tmp = ll_set_opt("lazystatfs", s1, LL_SBI_LAZYSTATFS);
 858                 if (tmp) {
 859                         *flags |= tmp;
 860                         goto next;
 861                 }
 862                 tmp = ll_set_opt("nolazystatfs", s1, LL_SBI_LAZYSTATFS);
 863                 if (tmp) {
 864                         *flags &= ~tmp;
 865                         goto next;
 866                 }
 867                 tmp = ll_set_opt("32bitapi", s1, LL_SBI_32BIT_API);
 868                 if (tmp) {
 869                         *flags |= tmp;
 870                         goto next;
 871                 }
 872                 tmp = ll_set_opt("verbose", s1, LL_SBI_VERBOSE);
 873                 if (tmp) {
 874                         *flags |= tmp;
 875                         goto next;
 876                 }
 877                 tmp = ll_set_opt("noverbose", s1, LL_SBI_VERBOSE);
 878                 if (tmp) {
 879                         *flags &= ~tmp;
 880                         goto next;
 881                 }
 882                 tmp = ll_set_opt("always_ping", s1, LL_SBI_ALWAYS_PING);
 883                 if (tmp) {
 884                         *flags |= tmp;
 885                         goto next;
 886                 }
 887                 LCONSOLE_ERROR_MSG(0x152, "Unknown option '%s', won't mount.\n",
 888                                    s1);
 889                 RETURN(-EINVAL);
 890
 891 next:
 892                 /* Find next opt */
 893                 s2 = strchr(s1, ',');
 894                 if (s2 == NULL)
 895                         break;
 896                 s1 = s2 + 1;
 897         }
 898         RETURN(0);
 899 }
 900
 901 void ll_lli_init(struct ll_inode_info *lli)
 902 {
 903         lli->lli_inode_magic = LLI_INODE_MAGIC;
 904         lli->lli_flags = 0;
 905         spin_lock_init(&lli->lli_lock);
 906         lli->lli_posix_acl = NULL;
 907         /* Do not set lli_fid, it has been initialized already. */
 908         fid_zero(&lli->lli_pfid);
 909         lli->lli_mds_read_och = NULL;
 910         lli->lli_mds_write_och = NULL;
 911         lli->lli_mds_exec_och = NULL;
 912         lli->lli_open_fd_read_count = 0;
 913         lli->lli_open_fd_write_count = 0;
 914         lli->lli_open_fd_exec_count = 0;
 915         mutex_init(&lli->lli_och_mutex);
 916         spin_lock_init(&lli->lli_agl_lock);
 917         spin_lock_init(&lli->lli_layout_lock);
 918         ll_layout_version_set(lli, CL_LAYOUT_GEN_NONE);
 919         lli->lli_clob = NULL;
 920
 921         init_rwsem(&lli->lli_xattrs_list_rwsem);
 922         mutex_init(&lli->lli_xattrs_enq_lock);
 923
 924         LASSERT(lli->lli_vfs_inode.i_mode != 0);
 925         if (S_ISDIR(lli->lli_vfs_inode.i_mode)) {
 926                 mutex_init(&lli->lli_readdir_mutex);
 927                 lli->lli_opendir_key = NULL;
 928                 lli->lli_sai = NULL;
 929                 spin_lock_init(&lli->lli_sa_lock);
 930                 lli->lli_opendir_pid = 0;
 931                 lli->lli_sa_enabled = 0;
 932                 lli->lli_def_stripe_offset = -1;
 933         } else {
 934                 mutex_init(&lli->lli_size_mutex);
 935                 lli->lli_symlink_name = NULL;
 936                 init_rwsem(&lli->lli_trunc_sem);
 937                 range_lock_tree_init(&lli->lli_write_tree);
 938                 init_rwsem(&lli->lli_glimpse_sem);
 939                 lli->lli_glimpse_time = 0;
 940                 INIT_LIST_HEAD(&lli->lli_agl_list);
 941                 lli->lli_agl_index = 0;
 942                 lli->lli_async_rc = 0;
 943         }
 944         mutex_init(&lli->lli_layout_mutex);
 945         memset(lli->lli_jobid, 0, LUSTRE_JOBID_SIZE);
 946 }
 947
 948 static inline int ll_bdi_register(struct backing_dev_info *bdi)
 949 {
 950         static atomic_t ll_bdi_num = ATOMIC_INIT(0);
 951
 952         bdi->name = "lustre";
 953         return bdi_register(bdi, NULL, "lustre-%d",
 954                             atomic_inc_return(&ll_bdi_num));
 955 }
 956
 957 int ll_fill_super(struct super_block *sb, struct vfsmount *mnt)
 958 {
 959         struct  lustre_profile *lprof = NULL;
 960         struct  lustre_sb_info *lsi = s2lsi(sb);
 961         struct  ll_sb_info *sbi;
 962         char    *dt = NULL, *md = NULL;
 963         char    *profilenm = get_profile_name(sb);
 964         struct config_llog_instance *cfg;
 965         /* %p for void* in printf needs 16+2 characters: 0xffffffffffffffff */
 966         const int instlen = sizeof(cfg->cfg_instance) * 2 + 2;
 967         int     md_len = 0;
 968         int     dt_len = 0;
 969         int     err;
 970         ENTRY;
 971
 972         CDEBUG(D_VFSTRACE, "VFS Op: sb %p\n", sb);
 973
 974         OBD_ALLOC_PTR(cfg);
 975         if (cfg == NULL)
 976                 RETURN(-ENOMEM);
 977
 978         try_module_get(THIS_MODULE);
 979
 980         /* client additional sb info */
 981         lsi->lsi_llsbi = sbi = ll_init_sbi();
 982         if (!sbi) {
 983                 module_put(THIS_MODULE);
 984                 OBD_FREE_PTR(cfg);
 985                 RETURN(-ENOMEM);
 986         }
 987
 988         err = ll_options(lsi->lsi_lmd->lmd_opts, &sbi->ll_flags);
 989         if (err)
 990                 GOTO(out_free, err);
 991
 992         err = bdi_init(&lsi->lsi_bdi);
 993         if (err)
 994                 GOTO(out_free, err);
 995         lsi->lsi_flags |= LSI_BDI_INITIALIZED;
 996 #ifdef HAVE_BDI_CAP_MAP_COPY
 997         lsi->lsi_bdi.capabilities = BDI_CAP_MAP_COPY;
 998 #else
 999         lsi->lsi_bdi.capabilities = 0;
1000 #endif
1001         err = ll_bdi_register(&lsi->lsi_bdi);
1002         if (err)
1003                 GOTO(out_free, err);
1004
1005         sb->s_bdi = &lsi->lsi_bdi;
1006 #ifndef HAVE_DCACHE_LOCK
1007         /* kernel >= 2.6.38 store dentry operations in sb->s_d_op. */
1008         sb->s_d_op = &ll_d_ops;
1009 #endif
1010
1011         /* Call lprocfs_ll_register_mountpoint() before lustre_process_log()
1012          * so that "llite.*.*" params can be processed correctly. */
1013         if (proc_lustre_fs_root != NULL) {
1014                 err = lprocfs_ll_register_mountpoint(proc_lustre_fs_root, sb);
1015                 if (err < 0) {
1016                         CERROR("%s: could not register mountpoint in llite: "
1017                                "rc = %d\n", ll_get_fsname(sb, NULL, 0), err);
1018                         err = 0;
1019                 }
1020         }
1021
1022         /* Generate a string unique to this super, in case some joker tries
1023            to mount the same fs at two mount points.
1024            Use the address of the super itself.*/
1025         cfg->cfg_instance = sb;
1026         cfg->cfg_uuid = lsi->lsi_llsbi->ll_sb_uuid;
1027         cfg->cfg_callback = class_config_llog_handler;
1028         cfg->cfg_sub_clds = CONFIG_SUB_CLIENT;
1029         /* set up client obds */
1030         err = lustre_process_log(sb, profilenm, cfg);
1031         if (err < 0)
1032                 GOTO(out_proc, err);
1033
1034         /* Profile set with LCFG_MOUNTOPT so we can find our mdc and osc obds */
1035         lprof = class_get_profile(profilenm);
1036         if (lprof == NULL) {
1037                 LCONSOLE_ERROR_MSG(0x156, "The client profile '%s' could not be"
1038                                    " read from the MGS.  Does that filesystem "
1039                                    "exist?\n", profilenm);
1040                 GOTO(out_proc, err = -EINVAL);
1041         }
1042         CDEBUG(D_CONFIG, "Found profile %s: mdc=%s osc=%s\n", profilenm,
1043                lprof->lp_md, lprof->lp_dt);
1044
1045         dt_len = strlen(lprof->lp_dt) + instlen + 2;
1046         OBD_ALLOC(dt, dt_len);
1047         if (!dt)
1048                 GOTO(out_proc, err = -ENOMEM);
1049         snprintf(dt, dt_len - 1, "%s-%p", lprof->lp_dt, cfg->cfg_instance);
1050
1051         md_len = strlen(lprof->lp_md) + instlen + 2;
1052         OBD_ALLOC(md, md_len);
1053         if (!md)
1054                 GOTO(out_proc, err = -ENOMEM);
1055         snprintf(md, md_len - 1, "%s-%p", lprof->lp_md, cfg->cfg_instance);
1056
1057         /* connections, registrations, sb setup */
1058         err = client_common_fill_super(sb, md, dt, mnt);
1059         if (err < 0)
1060                 GOTO(out_proc, err);
1061
1062         sbi->ll_client_common_fill_super_succeeded = 1;
1063
1064 out_proc:
1065         if (err < 0)
1066                 lprocfs_ll_unregister_mountpoint(sbi);
1067 out_free:
1068         if (md)
1069                 OBD_FREE(md, md_len);
1070         if (dt)
1071                 OBD_FREE(dt, dt_len);
1072         if (lprof != NULL)
1073                 class_put_profile(lprof);
1074         if (err)
1075                 ll_put_super(sb);
1076         else if (sbi->ll_flags & LL_SBI_VERBOSE)
1077                 LCONSOLE_WARN("Mounted %s\n", profilenm);
1078
1079         OBD_FREE_PTR(cfg);
1080         RETURN(err);
1081 } /* ll_fill_super */
1082
1083 void ll_put_super(struct super_block *sb)
1084 {
1085         struct config_llog_instance cfg, params_cfg;
1086         struct obd_device *obd;
1087         struct lustre_sb_info *lsi = s2lsi(sb);
1088         struct ll_sb_info *sbi = ll_s2sbi(sb);
1089         char *profilenm = get_profile_name(sb);
1090         long ccc_count;
1091         int next, force = 1, rc = 0;
1092         ENTRY;
1093
1094         CDEBUG(D_VFSTRACE, "VFS Op: sb %p - %s\n", sb, profilenm);
1095
1096         cfg.cfg_instance = sb;
1097         lustre_end_log(sb, profilenm, &cfg);
1098
1099         params_cfg.cfg_instance = sb;
1100         lustre_end_log(sb, PARAMS_FILENAME, &params_cfg);
1101
1102         if (sbi->ll_md_exp) {
1103                 obd = class_exp2obd(sbi->ll_md_exp);
1104                 if (obd)
1105                         force = obd->obd_force;
1106         }
1107
1108         /* Wait for unstable pages to be committed to stable storage */
1109         if (force == 0) {
1110                 struct l_wait_info lwi = LWI_INTR(LWI_ON_SIGNAL_NOOP, NULL);
1111                 rc = l_wait_event(sbi->ll_cache->ccc_unstable_waitq,
1112                         atomic_long_read(&sbi->ll_cache->ccc_unstable_nr) == 0,
1113                         &lwi);
1114         }
1115
1116         ccc_count = atomic_long_read(&sbi->ll_cache->ccc_unstable_nr);
1117         if (force == 0 && rc != -EINTR)
1118                 LASSERTF(ccc_count == 0, "count: %li\n", ccc_count);
1119
1120
1121         /* We need to set force before the lov_disconnect in
1122            lustre_common_put_super, since l_d cleans up osc's as well. */
1123         if (force) {
1124                 next = 0;
1125                 while ((obd = class_devices_in_group(&sbi->ll_sb_uuid,
1126                                                      &next)) != NULL) {
1127                         obd->obd_force = force;
1128                 }
1129         }
1130
1131         if (sbi->ll_client_common_fill_super_succeeded) {
1132                 /* Only if client_common_fill_super succeeded */
1133                 client_common_put_super(sb);
1134         }
1135
1136         next = 0;
1137         while ((obd = class_devices_in_group(&sbi->ll_sb_uuid, &next)) !=NULL) {
1138                 class_manual_cleanup(obd);
1139         }
1140
1141         if (sbi->ll_flags & LL_SBI_VERBOSE)
1142                 LCONSOLE_WARN("Unmounted %s\n", profilenm ? profilenm : "");
1143
1144         if (profilenm)
1145                 class_del_profile(profilenm);
1146
1147         if (lsi->lsi_flags & LSI_BDI_INITIALIZED) {
1148                 bdi_destroy(&lsi->lsi_bdi);
1149                 lsi->lsi_flags &= ~LSI_BDI_INITIALIZED;
1150         }
1151
1152         ll_free_sbi(sb);
1153         lsi->lsi_llsbi = NULL;
1154
1155         lustre_common_put_super(sb);
1156
1157         cl_env_cache_purge(~0);
1158
1159         module_put(THIS_MODULE);
1160
1161         EXIT;
1162 } /* client_put_super */
1163
1164 struct inode *ll_inode_from_resource_lock(struct ldlm_lock *lock)
1165 {
1166         struct inode *inode = NULL;
1167
1168         /* NOTE: we depend on atomic igrab() -bzzz */
1169         lock_res_and_lock(lock);
1170         if (lock->l_resource->lr_lvb_inode) {
1171                 struct ll_inode_info * lli;
1172                 lli = ll_i2info(lock->l_resource->lr_lvb_inode);
1173                 if (lli->lli_inode_magic == LLI_INODE_MAGIC) {
1174                         inode = igrab(lock->l_resource->lr_lvb_inode);
1175                 } else {
1176                         inode = lock->l_resource->lr_lvb_inode;
1177                         LDLM_DEBUG_LIMIT(inode->i_state & I_FREEING ?  D_INFO :
1178                                          D_WARNING, lock, "lr_lvb_inode %p is "
1179                                          "bogus: magic %08x",
1180                                          lock->l_resource->lr_lvb_inode,
1181                                          lli->lli_inode_magic);
1182                         inode = NULL;
1183                 }
1184         }
1185         unlock_res_and_lock(lock);
1186         return inode;
1187 }
1188
1189 void ll_dir_clear_lsm_md(struct inode *inode)
1190 {
1191         struct ll_inode_info *lli = ll_i2info(inode);
1192
1193         LASSERT(S_ISDIR(inode->i_mode));
1194
1195         if (lli->lli_lsm_md != NULL) {
1196                 lmv_free_memmd(lli->lli_lsm_md);
1197                 lli->lli_lsm_md = NULL;
1198         }
1199 }
1200
1201 static struct inode *ll_iget_anon_dir(struct super_block *sb,
1202                                       const struct lu_fid *fid,
1203                                       struct lustre_md *md)
1204 {
1205         struct ll_sb_info       *sbi = ll_s2sbi(sb);
1206         struct mdt_body         *body = md->body;
1207         struct inode            *inode;
1208         ino_t                   ino;
1209         ENTRY;
1210
1211         ino = cl_fid_build_ino(fid, sbi->ll_flags & LL_SBI_32BIT_API);
1212         inode = iget_locked(sb, ino);
1213         if (inode == NULL) {
1214                 CERROR("%s: failed get simple inode "DFID": rc = -ENOENT\n",
1215                        ll_get_fsname(sb, NULL, 0), PFID(fid));
1216                 RETURN(ERR_PTR(-ENOENT));
1217         }
1218
1219         if (inode->i_state & I_NEW) {
1220                 struct ll_inode_info *lli = ll_i2info(inode);
1221                 struct lmv_stripe_md *lsm = md->lmv;
1222
1223                 inode->i_mode = (inode->i_mode & ~S_IFMT) |
1224                                 (body->mbo_mode & S_IFMT);
1225                 LASSERTF(S_ISDIR(inode->i_mode), "Not slave inode "DFID"\n",
1226                          PFID(fid));
1227
1228                 LTIME_S(inode->i_mtime) = 0;
1229                 LTIME_S(inode->i_atime) = 0;
1230                 LTIME_S(inode->i_ctime) = 0;
1231                 inode->i_rdev = 0;
1232
1233 #ifdef HAVE_BACKING_DEV_INFO
1234                 /* initializing backing dev info. */
1235                 inode->i_mapping->backing_dev_info =
1236                                                 &s2lsi(inode->i_sb)->lsi_bdi;
1237 #endif
1238                 inode->i_op = &ll_dir_inode_operations;
1239                 inode->i_fop = &ll_dir_operations;
1240                 lli->lli_fid = *fid;
1241                 ll_lli_init(lli);
1242
1243                 LASSERT(lsm != NULL);
1244                 /* master object FID */
1245                 lli->lli_pfid = body->mbo_fid1;
1246                 CDEBUG(D_INODE, "lli %p slave "DFID" master "DFID"\n",
1247                        lli, PFID(fid), PFID(&lli->lli_pfid));
1248                 unlock_new_inode(inode);
1249         }
1250
1251         RETURN(inode);
1252 }
1253
1254 static int ll_init_lsm_md(struct inode *inode, struct lustre_md *md)
1255 {
1256         struct lu_fid *fid;
1257         struct lmv_stripe_md *lsm = md->lmv;
1258         int i;
1259
1260         LASSERT(lsm != NULL);
1261         /* XXX sigh, this lsm_root initialization should be in
1262          * LMV layer, but it needs ll_iget right now, so we
1263          * put this here right now. */
1264         for (i = 0; i < lsm->lsm_md_stripe_count; i++) {
1265                 fid = &lsm->lsm_md_oinfo[i].lmo_fid;
1266                 LASSERT(lsm->lsm_md_oinfo[i].lmo_root == NULL);
1267                 /* Unfortunately ll_iget will call ll_update_inode,
1268                  * where the initialization of slave inode is slightly
1269                  * different, so it reset lsm_md to NULL to avoid
1270                  * initializing lsm for slave inode. */
1271                 /* For migrating inode, master stripe and master object will
1272                  * be same, so we only need assign this inode */
1273                 if (lsm->lsm_md_hash_type & LMV_HASH_FLAG_MIGRATION && i == 0)
1274                         lsm->lsm_md_oinfo[i].lmo_root = inode;
1275                 else
1276                         lsm->lsm_md_oinfo[i].lmo_root =
1277                                 ll_iget_anon_dir(inode->i_sb, fid, md);
1278
1279                 if (IS_ERR(lsm->lsm_md_oinfo[i].lmo_root)) {
1280                         int rc = PTR_ERR(lsm->lsm_md_oinfo[i].lmo_root);
1281
1282                         lsm->lsm_md_oinfo[i].lmo_root = NULL;
1283                         return rc;
1284                 }
1285         }
1286
1287         return 0;
1288 }
1289
1290 static inline int lli_lsm_md_eq(const struct lmv_stripe_md *lsm_md1,
1291                                 const struct lmv_stripe_md *lsm_md2)
1292 {
1293         return lsm_md1->lsm_md_magic == lsm_md2->lsm_md_magic &&
1294                lsm_md1->lsm_md_stripe_count == lsm_md2->lsm_md_stripe_count &&
1295                lsm_md1->lsm_md_master_mdt_index ==
1296                                         lsm_md2->lsm_md_master_mdt_index &&
1297                lsm_md1->lsm_md_hash_type == lsm_md2->lsm_md_hash_type &&
1298                lsm_md1->lsm_md_layout_version ==
1299                                         lsm_md2->lsm_md_layout_version &&
1300                strcmp(lsm_md1->lsm_md_pool_name,
1301                       lsm_md2->lsm_md_pool_name) == 0;
1302 }
1303
1304 static int ll_update_lsm_md(struct inode *inode, struct lustre_md *md)
1305 {
1306         struct ll_inode_info *lli = ll_i2info(inode);
1307         struct lmv_stripe_md *lsm = md->lmv;
1308         int     rc;
1309         ENTRY;
1310
1311         LASSERT(S_ISDIR(inode->i_mode));
1312         CDEBUG(D_INODE, "update lsm %p of "DFID"\n", lli->lli_lsm_md,
1313                PFID(ll_inode2fid(inode)));
1314
1315         /* no striped information from request. */
1316         if (lsm == NULL) {
1317                 if (lli->lli_lsm_md == NULL) {
1318                         RETURN(0);
1319                 } else if (lli->lli_lsm_md->lsm_md_hash_type &
1320                                                 LMV_HASH_FLAG_MIGRATION) {
1321                         /* migration is done, the temporay MIGRATE layout has
1322                          * been removed */
1323                         CDEBUG(D_INODE, DFID" finish migration.\n",
1324                                PFID(ll_inode2fid(inode)));
1325                         lmv_free_memmd(lli->lli_lsm_md);
1326                         lli->lli_lsm_md = NULL;
1327                         RETURN(0);
1328                 } else {
1329                         /* The lustre_md from req does not include stripeEA,
1330                          * see ll_md_setattr */
1331                         RETURN(0);
1332                 }
1333         }
1334
1335         /* set the directory layout */
1336         if (lli->lli_lsm_md == NULL) {
1337                 struct cl_attr  *attr;
1338
1339                 rc = ll_init_lsm_md(inode, md);
1340                 if (rc != 0)
1341                         RETURN(rc);
1342
1343                 /* set md->lmv to NULL, so the following free lustre_md
1344                  * will not free this lsm */
1345                 md->lmv = NULL;
1346                 lli->lli_lsm_md = lsm;
1347
1348                 OBD_ALLOC_PTR(attr);
1349                 if (attr == NULL)
1350                         RETURN(-ENOMEM);
1351
1352                 /* validate the lsm */
1353                 rc = md_merge_attr(ll_i2mdexp(inode), lsm, attr,
1354                                    ll_md_blocking_ast);
1355                 if (rc != 0) {
1356                         OBD_FREE_PTR(attr);
1357                         RETURN(rc);
1358                 }
1359
1360                 if (md->body->mbo_valid & OBD_MD_FLNLINK)
1361                         md->body->mbo_nlink = attr->cat_nlink;
1362                 if (md->body->mbo_valid & OBD_MD_FLSIZE)
1363                         md->body->mbo_size = attr->cat_size;
1364                 if (md->body->mbo_valid & OBD_MD_FLATIME)
1365                         md->body->mbo_atime = attr->cat_atime;
1366                 if (md->body->mbo_valid & OBD_MD_FLCTIME)
1367                         md->body->mbo_ctime = attr->cat_ctime;
1368                 if (md->body->mbo_valid & OBD_MD_FLMTIME)
1369                         md->body->mbo_mtime = attr->cat_mtime;
1370
1371                 OBD_FREE_PTR(attr);
1372
1373                 CDEBUG(D_INODE, "Set lsm %p magic %x to "DFID"\n", lsm,
1374                        lsm->lsm_md_magic, PFID(ll_inode2fid(inode)));
1375                 RETURN(0);
1376         }
1377
1378         /* Compare the old and new stripe information */
1379         if (!lsm_md_eq(lli->lli_lsm_md, lsm)) {
1380                 struct lmv_stripe_md    *old_lsm = lli->lli_lsm_md;
1381                 int                     idx;
1382
1383                 CERROR("%s: inode "DFID"(%p)'s lmv layout mismatch (%p)/(%p)"
1384                        "magic:0x%x/0x%x stripe count: %d/%d master_mdt: %d/%d"
1385                        "hash_type:0x%x/0x%x layout: 0x%x/0x%x pool:%s/%s\n",
1386                        ll_get_fsname(inode->i_sb, NULL, 0), PFID(&lli->lli_fid),
1387                        inode, lsm, old_lsm,
1388                        lsm->lsm_md_magic, old_lsm->lsm_md_magic,
1389                        lsm->lsm_md_stripe_count,
1390                        old_lsm->lsm_md_stripe_count,
1391                        lsm->lsm_md_master_mdt_index,
1392                        old_lsm->lsm_md_master_mdt_index,
1393                        lsm->lsm_md_hash_type, old_lsm->lsm_md_hash_type,
1394                        lsm->lsm_md_layout_version,
1395                        old_lsm->lsm_md_layout_version,
1396                        lsm->lsm_md_pool_name,
1397                        old_lsm->lsm_md_pool_name);
1398
1399                 for (idx = 0; idx < old_lsm->lsm_md_stripe_count; idx++) {
1400                         CERROR("%s: sub FIDs in old lsm idx %d, old: "DFID"\n",
1401                                ll_get_fsname(inode->i_sb, NULL, 0), idx,
1402                                PFID(&old_lsm->lsm_md_oinfo[idx].lmo_fid));
1403                 }
1404
1405                 for (idx = 0; idx < lsm->lsm_md_stripe_count; idx++) {
1406                         CERROR("%s: sub FIDs in new lsm idx %d, new: "DFID"\n",
1407                                ll_get_fsname(inode->i_sb, NULL, 0), idx,
1408                                PFID(&lsm->lsm_md_oinfo[idx].lmo_fid));
1409                 }
1410
1411                 RETURN(-EIO);
1412         }
1413
1414         RETURN(0);
1415 }
1416
1417 void ll_clear_inode(struct inode *inode)
1418 {
1419         struct ll_inode_info *lli = ll_i2info(inode);
1420         struct ll_sb_info *sbi = ll_i2sbi(inode);
1421         ENTRY;
1422
1423         CDEBUG(D_VFSTRACE, "VFS Op:inode="DFID"(%p)\n",
1424                PFID(ll_inode2fid(inode)), inode);
1425
1426         if (S_ISDIR(inode->i_mode)) {
1427                 /* these should have been cleared in ll_file_release */
1428                 LASSERT(lli->lli_opendir_key == NULL);
1429                 LASSERT(lli->lli_sai == NULL);
1430                 LASSERT(lli->lli_opendir_pid == 0);
1431         }
1432
1433         md_null_inode(sbi->ll_md_exp, ll_inode2fid(inode));
1434
1435         LASSERT(!lli->lli_open_fd_write_count);
1436         LASSERT(!lli->lli_open_fd_read_count);
1437         LASSERT(!lli->lli_open_fd_exec_count);
1438
1439         if (lli->lli_mds_write_och)
1440                 ll_md_real_close(inode, FMODE_WRITE);
1441         if (lli->lli_mds_exec_och)
1442                 ll_md_real_close(inode, FMODE_EXEC);
1443         if (lli->lli_mds_read_och)
1444                 ll_md_real_close(inode, FMODE_READ);
1445
1446         if (S_ISLNK(inode->i_mode) && lli->lli_symlink_name) {
1447                 OBD_FREE(lli->lli_symlink_name,
1448                          strlen(lli->lli_symlink_name) + 1);
1449                 lli->lli_symlink_name = NULL;
1450         }
1451
1452         ll_xattr_cache_destroy(inode);
1453
1454 #ifdef CONFIG_FS_POSIX_ACL
1455         forget_all_cached_acls(inode);
1456         if (lli->lli_posix_acl) {
1457                 LASSERT(atomic_read(&lli->lli_posix_acl->a_refcount) == 1);
1458                 posix_acl_release(lli->lli_posix_acl);
1459                 lli->lli_posix_acl = NULL;
1460         }
1461 #endif
1462         lli->lli_inode_magic = LLI_INODE_DEAD;
1463
1464         if (S_ISDIR(inode->i_mode))
1465                 ll_dir_clear_lsm_md(inode);
1466         else if (S_ISREG(inode->i_mode) && !is_bad_inode(inode))
1467                 LASSERT(list_empty(&lli->lli_agl_list));
1468
1469         /*
1470          * XXX This has to be done before lsm is freed below, because
1471          * cl_object still uses inode lsm.
1472          */
1473         cl_inode_fini(inode);
1474
1475         EXIT;
1476 }
1477
1478 static int ll_md_setattr(struct dentry *dentry, struct md_op_data *op_data)
1479 {
1480         struct lustre_md md;
1481         struct inode *inode = dentry->d_inode;
1482         struct ll_sb_info *sbi = ll_i2sbi(inode);
1483         struct ptlrpc_request *request = NULL;
1484         int rc, ia_valid;
1485         ENTRY;
1486
1487         op_data = ll_prep_md_op_data(op_data, inode, NULL, NULL, 0, 0,
1488                                      LUSTRE_OPC_ANY, NULL);
1489         if (IS_ERR(op_data))
1490                 RETURN(PTR_ERR(op_data));
1491
1492         rc = md_setattr(sbi->ll_md_exp, op_data, NULL, 0, &request);
1493         if (rc) {
1494                 ptlrpc_req_finished(request);
1495                 if (rc == -ENOENT) {
1496                         clear_nlink(inode);
1497                         /* Unlinked special device node? Or just a race?
1498                          * Pretend we done everything. */
1499                         if (!S_ISREG(inode->i_mode) &&
1500                             !S_ISDIR(inode->i_mode)) {
1501                                 ia_valid = op_data->op_attr.ia_valid;
1502                                 op_data->op_attr.ia_valid &= ~TIMES_SET_FLAGS;
1503                                 rc = simple_setattr(dentry, &op_data->op_attr);
1504                                 op_data->op_attr.ia_valid = ia_valid;
1505                         }
1506                 } else if (rc != -EPERM && rc != -EACCES && rc != -ETXTBSY) {
1507                         CERROR("md_setattr fails: rc = %d\n", rc);
1508                 }
1509                 RETURN(rc);
1510         }
1511
1512         rc = md_get_lustre_md(sbi->ll_md_exp, request, sbi->ll_dt_exp,
1513                               sbi->ll_md_exp, &md);
1514         if (rc) {
1515                 ptlrpc_req_finished(request);
1516                 RETURN(rc);
1517         }
1518
1519         ia_valid = op_data->op_attr.ia_valid;
1520         /* inode size will be in ll_setattr_ost, can't do it now since dirty
1521          * cache is not cleared yet. */
1522         op_data->op_attr.ia_valid &= ~(TIMES_SET_FLAGS | ATTR_SIZE);
1523         if (S_ISREG(inode->i_mode))
1524                 inode_lock(inode);
1525         rc = simple_setattr(dentry, &op_data->op_attr);
1526         if (S_ISREG(inode->i_mode))
1527                 inode_unlock(inode);
1528         op_data->op_attr.ia_valid = ia_valid;
1529
1530         rc = ll_update_inode(inode, &md);
1531         ptlrpc_req_finished(request);
1532
1533         RETURN(rc);
1534 }
1535
1536 /* If this inode has objects allocated to it (lsm != NULL), then the OST
1537  * object(s) determine the file size and mtime.  Otherwise, the MDS will
1538  * keep these values until such a time that objects are allocated for it.
1539  * We do the MDS operations first, as it is checking permissions for us.
1540  * We don't to the MDS RPC if there is nothing that we want to store there,
1541  * otherwise there is no harm in updating mtime/atime on the MDS if we are
1542  * going to do an RPC anyways.
1543  *
1544  * If we are doing a truncate, we will send the mtime and ctime updates
1545  * to the OST with the punch RPC, otherwise we do an explicit setattr RPC.
1546  * I don't believe it is possible to get e.g. ATTR_MTIME_SET and ATTR_SIZE
1547  * at the same time.
1548  *
1549  * In case of HSMimport, we only set attr on MDS.
1550  */
1551 int ll_setattr_raw(struct dentry *dentry, struct iattr *attr, bool hsm_import)
1552 {
1553         struct inode *inode = dentry->d_inode;
1554         struct ll_inode_info *lli = ll_i2info(inode);
1555         struct md_op_data *op_data = NULL;
1556         int rc = 0;
1557         ENTRY;
1558
1559         CDEBUG(D_VFSTRACE, "%s: setattr inode "DFID"(%p) from %llu to %llu, "
1560                "valid %x, hsm_import %d\n",
1561                ll_get_fsname(inode->i_sb, NULL, 0), PFID(&lli->lli_fid),
1562                inode, i_size_read(inode), attr->ia_size, attr->ia_valid,
1563                hsm_import);
1564
1565         if (attr->ia_valid & ATTR_SIZE) {
1566                 /* Check new size against VFS/VM file size limit and rlimit */
1567                 rc = inode_newsize_ok(inode, attr->ia_size);
1568                 if (rc)
1569                         RETURN(rc);
1570
1571                 /* The maximum Lustre file size is variable, based on the
1572                  * OST maximum object size and number of stripes.  This
1573                  * needs another check in addition to the VFS check above. */
1574                 if (attr->ia_size > ll_file_maxbytes(inode)) {
1575                         CDEBUG(D_INODE,"file "DFID" too large %llu > %llu\n",
1576                                PFID(&lli->lli_fid), attr->ia_size,
1577                                ll_file_maxbytes(inode));
1578                         RETURN(-EFBIG);
1579                 }
1580
1581                 attr->ia_valid |= ATTR_MTIME | ATTR_CTIME;
1582         }
1583
1584         /* POSIX: check before ATTR_*TIME_SET set (from inode_change_ok) */
1585         if (attr->ia_valid & TIMES_SET_FLAGS) {
1586                 if ((!uid_eq(current_fsuid(), inode->i_uid)) &&
1587                     !cfs_capable(CFS_CAP_FOWNER))
1588                         RETURN(-EPERM);
1589         }
1590
1591         /* We mark all of the fields "set" so MDS/OST does not re-set them */
1592         if (!(attr->ia_valid & ATTR_CTIME_SET) &&
1593             (attr->ia_valid & ATTR_CTIME)) {
1594                 attr->ia_ctime = current_time(inode);
1595                 attr->ia_valid |= ATTR_CTIME_SET;
1596         }
1597         if (!(attr->ia_valid & ATTR_ATIME_SET) &&
1598             (attr->ia_valid & ATTR_ATIME)) {
1599                 attr->ia_atime = current_time(inode);
1600                 attr->ia_valid |= ATTR_ATIME_SET;
1601         }
1602         if (!(attr->ia_valid & ATTR_MTIME_SET) &&
1603             (attr->ia_valid & ATTR_MTIME)) {
1604                 attr->ia_mtime = current_time(inode);
1605                 attr->ia_valid |= ATTR_MTIME_SET;
1606         }
1607
1608         if (attr->ia_valid & (ATTR_MTIME | ATTR_CTIME))
1609                 CDEBUG(D_INODE, "setting mtime %lu, ctime %lu, now = %llu\n",
1610                        LTIME_S(attr->ia_mtime), LTIME_S(attr->ia_ctime),
1611                        (s64)ktime_get_real_seconds());
1612
1613         if (S_ISREG(inode->i_mode)) {
1614                 if (attr->ia_valid & ATTR_SIZE)
1615                         inode_dio_write_done(inode);
1616                 inode_unlock(inode);
1617         }
1618
1619         /* We always do an MDS RPC, even if we're only changing the size;
1620          * only the MDS knows whether truncate() should fail with -ETXTBUSY */
1621
1622         OBD_ALLOC_PTR(op_data);
1623         if (op_data == NULL)
1624                 GOTO(out, rc = -ENOMEM);
1625
1626         if (!hsm_import && attr->ia_valid & ATTR_SIZE) {
1627                 /* If we are changing file size, file content is
1628                  * modified, flag it. */
1629                 attr->ia_valid |= MDS_OPEN_OWNEROVERRIDE;
1630                 op_data->op_bias |= MDS_DATA_MODIFIED;
1631                 ll_file_clear_flag(lli, LLIF_DATA_MODIFIED);
1632         }
1633
1634         op_data->op_attr = *attr;
1635
1636         rc = ll_md_setattr(dentry, op_data);
1637         if (rc)
1638                 GOTO(out, rc);
1639
1640         if (!S_ISREG(inode->i_mode) || hsm_import)
1641                 GOTO(out, rc = 0);
1642
1643         if (attr->ia_valid & (ATTR_SIZE |
1644                               ATTR_ATIME | ATTR_ATIME_SET |
1645                               ATTR_MTIME | ATTR_MTIME_SET |
1646                               ATTR_CTIME | ATTR_CTIME_SET)) {
1647                 /* For truncate and utimes sending attributes to OSTs, setting
1648                  * mtime/atime to the past will be performed under PW [0:EOF]
1649                  * extent lock (new_size:EOF for truncate).  It may seem
1650                  * excessive to send mtime/atime updates to OSTs when not
1651                  * setting times to past, but it is necessary due to possible
1652                  * time de-synchronization between MDT inode and OST objects */
1653                 rc = cl_setattr_ost(lli->lli_clob, attr, 0);
1654         }
1655
1656         /* If the file was restored, it needs to set dirty flag.
1657          *
1658          * We've already sent MDS_DATA_MODIFIED flag in
1659          * ll_md_setattr() for truncate. However, the MDT refuses to
1660          * set the HS_DIRTY flag on released files, so we have to set
1661          * it again if the file has been restored. Please check how
1662          * LLIF_DATA_MODIFIED is set in vvp_io_setattr_fini().
1663          *
1664          * Please notice that if the file is not released, the previous
1665          * MDS_DATA_MODIFIED has taken effect and usually
1666          * LLIF_DATA_MODIFIED is not set(see vvp_io_setattr_fini()).
1667          * This way we can save an RPC for common open + trunc
1668          * operation. */
1669         if (ll_file_test_and_clear_flag(lli, LLIF_DATA_MODIFIED)) {
1670                 struct hsm_state_set hss = {
1671                         .hss_valid = HSS_SETMASK,
1672                         .hss_setmask = HS_DIRTY,
1673                 };
1674                 int rc2;
1675
1676                 rc2 = ll_hsm_state_set(inode, &hss);
1677                 /* truncate and write can happen at the same time, so that
1678                  * the file can be set modified even though the file is not
1679                  * restored from released state, and ll_hsm_state_set() is
1680                  * not applicable for the file, and rc2 < 0 is normal in this
1681                  * case. */
1682                 if (rc2 < 0)
1683                         CDEBUG(D_INFO, DFID "HSM set dirty failed: rc2 = %d\n",
1684                                PFID(ll_inode2fid(inode)), rc2);
1685         }
1686
1687         EXIT;
1688 out:
1689         if (op_data != NULL)
1690                 ll_finish_md_op_data(op_data);
1691
1692         if (S_ISREG(inode->i_mode)) {
1693                 inode_lock(inode);
1694                 if ((attr->ia_valid & ATTR_SIZE) && !hsm_import)
1695                         inode_dio_wait(inode);
1696                 /* Once we've got the i_mutex, it's safe to set the S_NOSEC
1697                  * flag.  ll_update_inode (called from ll_md_setattr), clears
1698                  * inode flags, so there is a gap where S_NOSEC is not set.
1699                  * This can cause a writer to take the i_mutex unnecessarily,
1700                  * but this is safe to do and should be rare. */
1701                 inode_has_no_xattr(inode);
1702         }
1703
1704         ll_stats_ops_tally(ll_i2sbi(inode), (attr->ia_valid & ATTR_SIZE) ?
1705                         LPROC_LL_TRUNC : LPROC_LL_SETATTR, 1);
1706
1707         return rc;
1708 }
1709
1710 int ll_setattr(struct dentry *de, struct iattr *attr)
1711 {
1712         int mode = de->d_inode->i_mode;
1713
1714         if ((attr->ia_valid & (ATTR_CTIME|ATTR_SIZE|ATTR_MODE)) ==
1715                               (ATTR_CTIME|ATTR_SIZE|ATTR_MODE))
1716                 attr->ia_valid |= MDS_OPEN_OWNEROVERRIDE;
1717
1718         if (((attr->ia_valid & (ATTR_MODE|ATTR_FORCE|ATTR_SIZE)) ==
1719                                (ATTR_SIZE|ATTR_MODE)) &&
1720             (((mode & S_ISUID) && !(attr->ia_mode & S_ISUID)) ||
1721              (((mode & (S_ISGID|S_IXGRP)) == (S_ISGID|S_IXGRP)) &&
1722               !(attr->ia_mode & S_ISGID))))
1723                 attr->ia_valid |= ATTR_FORCE;
1724
1725         if ((attr->ia_valid & ATTR_MODE) &&
1726             (mode & S_ISUID) &&
1727             !(attr->ia_mode & S_ISUID) &&
1728             !(attr->ia_valid & ATTR_KILL_SUID))
1729                 attr->ia_valid |= ATTR_KILL_SUID;
1730
1731         if ((attr->ia_valid & ATTR_MODE) &&
1732             ((mode & (S_ISGID|S_IXGRP)) == (S_ISGID|S_IXGRP)) &&
1733             !(attr->ia_mode & S_ISGID) &&
1734             !(attr->ia_valid & ATTR_KILL_SGID))
1735                 attr->ia_valid |= ATTR_KILL_SGID;
1736
1737         /* avoid polluted from ATTR_TIMES_SET,
1738          * projid is not expected to be set here */
1739         attr->ia_valid &= ~MDS_ATTR_PROJID;
1740
1741         return ll_setattr_raw(de, attr, false);
1742 }
1743
1744 int ll_statfs_internal(struct super_block *sb, struct obd_statfs *osfs,
1745                        __u64 max_age, __u32 flags)
1746 {
1747         struct ll_sb_info *sbi = ll_s2sbi(sb);
1748         struct obd_statfs obd_osfs;
1749         int rc;
1750         ENTRY;
1751
1752         rc = obd_statfs(NULL, sbi->ll_md_exp, osfs, max_age, flags);
1753         if (rc) {
1754                 CERROR("md_statfs fails: rc = %d\n", rc);
1755                 RETURN(rc);
1756         }
1757
1758         osfs->os_type = sb->s_magic;
1759
1760         CDEBUG(D_SUPER, "MDC blocks %llu/%llu objects %llu/%llu\n",
1761                osfs->os_bavail, osfs->os_blocks, osfs->os_ffree,osfs->os_files);
1762
1763         if (sbi->ll_flags & LL_SBI_LAZYSTATFS)
1764                 flags |= OBD_STATFS_NODELAY;
1765
1766         rc = obd_statfs_rqset(sbi->ll_dt_exp, &obd_osfs, max_age, flags);
1767         if (rc) {
1768                 CERROR("obd_statfs fails: rc = %d\n", rc);
1769                 RETURN(rc);
1770         }
1771
1772         CDEBUG(D_SUPER, "OSC blocks %llu/%llu objects %llu/%llu\n",
1773                obd_osfs.os_bavail, obd_osfs.os_blocks, obd_osfs.os_ffree,
1774                obd_osfs.os_files);
1775
1776         osfs->os_bsize = obd_osfs.os_bsize;
1777         osfs->os_blocks = obd_osfs.os_blocks;
1778         osfs->os_bfree = obd_osfs.os_bfree;
1779         osfs->os_bavail = obd_osfs.os_bavail;
1780
1781         /* If we don't have as many objects free on the OST as inodes
1782          * on the MDS, we reduce the total number of inodes to
1783          * compensate, so that the "inodes in use" number is correct.
1784          */
1785         if (obd_osfs.os_ffree < osfs->os_ffree) {
1786                 osfs->os_files = (osfs->os_files - osfs->os_ffree) +
1787                         obd_osfs.os_ffree;
1788                 osfs->os_ffree = obd_osfs.os_ffree;
1789         }
1790
1791         RETURN(rc);
1792 }
1793 int ll_statfs(struct dentry *de, struct kstatfs *sfs)
1794 {
1795         struct super_block *sb = de->d_sb;
1796         struct obd_statfs osfs;
1797         __u64 fsid = huge_encode_dev(sb->s_dev);
1798         int rc;
1799
1800         CDEBUG(D_VFSTRACE, "VFS Op: at %llu jiffies\n", get_jiffies_64());
1801         ll_stats_ops_tally(ll_s2sbi(sb), LPROC_LL_STAFS, 1);
1802
1803         /* Some amount of caching on the client is allowed */
1804         rc = ll_statfs_internal(sb, &osfs,
1805                                 cfs_time_shift_64(-OBD_STATFS_CACHE_SECONDS),
1806                                 0);
1807         if (rc)
1808                 return rc;
1809
1810         statfs_unpack(sfs, &osfs);
1811
1812         /* We need to downshift for all 32-bit kernels, because we can't
1813          * tell if the kernel is being called via sys_statfs64() or not.
1814          * Stop before overflowing f_bsize - in which case it is better
1815          * to just risk EOVERFLOW if caller is using old sys_statfs(). */
1816         if (sizeof(long) < 8) {
1817                 while (osfs.os_blocks > ~0UL && sfs->f_bsize < 0x40000000) {
1818                         sfs->f_bsize <<= 1;
1819
1820                         osfs.os_blocks >>= 1;
1821                         osfs.os_bfree >>= 1;
1822                         osfs.os_bavail >>= 1;
1823                 }
1824         }
1825
1826         sfs->f_blocks = osfs.os_blocks;
1827         sfs->f_bfree = osfs.os_bfree;
1828         sfs->f_bavail = osfs.os_bavail;
1829         sfs->f_fsid.val[0] = (__u32)fsid;
1830         sfs->f_fsid.val[1] = (__u32)(fsid >> 32);
1831         return 0;
1832 }
1833
1834 void ll_inode_size_lock(struct inode *inode)
1835 {
1836         struct ll_inode_info *lli;
1837
1838         LASSERT(!S_ISDIR(inode->i_mode));
1839
1840         lli = ll_i2info(inode);
1841         mutex_lock(&lli->lli_size_mutex);
1842 }
1843
1844 void ll_inode_size_unlock(struct inode *inode)
1845 {
1846         struct ll_inode_info *lli;
1847
1848         lli = ll_i2info(inode);
1849         mutex_unlock(&lli->lli_size_mutex);
1850 }
1851
1852 int ll_update_inode(struct inode *inode, struct lustre_md *md)
1853 {
1854         struct ll_inode_info *lli = ll_i2info(inode);
1855         struct mdt_body *body = md->body;
1856         struct ll_sb_info *sbi = ll_i2sbi(inode);
1857         int rc = 0;
1858
1859         if (body->mbo_valid & OBD_MD_FLEASIZE) {
1860                 rc = cl_file_inode_init(inode, md);
1861                 if (rc)
1862                         return rc;
1863         }
1864
1865         if (S_ISDIR(inode->i_mode)) {
1866                 rc = ll_update_lsm_md(inode, md);
1867                 if (rc != 0)
1868                         return rc;
1869         }
1870
1871 #ifdef CONFIG_FS_POSIX_ACL
1872         if (body->mbo_valid & OBD_MD_FLACL) {
1873                 spin_lock(&lli->lli_lock);
1874                 if (lli->lli_posix_acl)
1875                         posix_acl_release(lli->lli_posix_acl);
1876                 lli->lli_posix_acl = md->posix_acl;
1877                 spin_unlock(&lli->lli_lock);
1878         }
1879 #endif
1880         inode->i_ino = cl_fid_build_ino(&body->mbo_fid1,
1881                                         sbi->ll_flags & LL_SBI_32BIT_API);
1882         inode->i_generation = cl_fid_build_gen(&body->mbo_fid1);
1883
1884         if (body->mbo_valid & OBD_MD_FLATIME) {
1885                 if (body->mbo_atime > LTIME_S(inode->i_atime))
1886                         LTIME_S(inode->i_atime) = body->mbo_atime;
1887                 lli->lli_atime = body->mbo_atime;
1888         }
1889
1890         if (body->mbo_valid & OBD_MD_FLMTIME) {
1891                 if (body->mbo_mtime > LTIME_S(inode->i_mtime)) {
1892                         CDEBUG(D_INODE, "setting ino %lu mtime from %lu "
1893                                "to %llu\n", inode->i_ino,
1894                                LTIME_S(inode->i_mtime), body->mbo_mtime);
1895                         LTIME_S(inode->i_mtime) = body->mbo_mtime;
1896                 }
1897                 lli->lli_mtime = body->mbo_mtime;
1898         }
1899
1900         if (body->mbo_valid & OBD_MD_FLCTIME) {
1901                 if (body->mbo_ctime > LTIME_S(inode->i_ctime))
1902                         LTIME_S(inode->i_ctime) = body->mbo_ctime;
1903                 lli->lli_ctime = body->mbo_ctime;
1904         }
1905
1906         /* Clear i_flags to remove S_NOSEC before permissions are updated */
1907         if (body->mbo_valid & OBD_MD_FLFLAGS)
1908                 inode->i_flags = ll_ext_to_inode_flags(body->mbo_flags);
1909         if (body->mbo_valid & OBD_MD_FLMODE)
1910                 inode->i_mode = (inode->i_mode & S_IFMT) |
1911                                 (body->mbo_mode & ~S_IFMT);
1912
1913         if (body->mbo_valid & OBD_MD_FLTYPE)
1914                 inode->i_mode = (inode->i_mode & ~S_IFMT) |
1915                                 (body->mbo_mode & S_IFMT);
1916
1917         LASSERT(inode->i_mode != 0);
1918         if (S_ISREG(inode->i_mode))
1919                 inode->i_blkbits = min(PTLRPC_MAX_BRW_BITS + 1,
1920                                        LL_MAX_BLKSIZE_BITS);
1921         else
1922                 inode->i_blkbits = inode->i_sb->s_blocksize_bits;
1923
1924         if (body->mbo_valid & OBD_MD_FLUID)
1925                 inode->i_uid = make_kuid(&init_user_ns, body->mbo_uid);
1926         if (body->mbo_valid & OBD_MD_FLGID)
1927                 inode->i_gid = make_kgid(&init_user_ns, body->mbo_gid);
1928         if (body->mbo_valid & OBD_MD_FLPROJID)
1929                 lli->lli_projid = body->mbo_projid;
1930         if (body->mbo_valid & OBD_MD_FLNLINK)
1931                 set_nlink(inode, body->mbo_nlink);
1932         if (body->mbo_valid & OBD_MD_FLRDEV)
1933                 inode->i_rdev = old_decode_dev(body->mbo_rdev);
1934
1935         if (body->mbo_valid & OBD_MD_FLID) {
1936                 /* FID shouldn't be changed! */
1937                 if (fid_is_sane(&lli->lli_fid)) {
1938                         LASSERTF(lu_fid_eq(&lli->lli_fid, &body->mbo_fid1),
1939                                  "Trying to change FID "DFID
1940                                  " to the "DFID", inode "DFID"(%p)\n",
1941                                  PFID(&lli->lli_fid), PFID(&body->mbo_fid1),
1942                                  PFID(ll_inode2fid(inode)), inode);
1943                 } else {
1944                         lli->lli_fid = body->mbo_fid1;
1945                 }
1946         }
1947
1948         LASSERT(fid_seq(&lli->lli_fid) != 0);
1949
1950         if (body->mbo_valid & OBD_MD_FLSIZE) {
1951                 i_size_write(inode, body->mbo_size);
1952
1953                 CDEBUG(D_VFSTRACE, "inode="DFID", updating i_size %llu\n",
1954                        PFID(ll_inode2fid(inode)),
1955                        (unsigned long long)body->mbo_size);
1956
1957                 if (body->mbo_valid & OBD_MD_FLBLOCKS)
1958                         inode->i_blocks = body->mbo_blocks;
1959         }
1960
1961         if (body->mbo_valid & OBD_MD_TSTATE) {
1962                 /* Set LLIF_FILE_RESTORING if restore ongoing and
1963                  * clear it when done to ensure to start again
1964                  * glimpsing updated attrs
1965                  */
1966                 if (body->mbo_t_state & MS_RESTORE)
1967                         ll_file_set_flag(lli, LLIF_FILE_RESTORING);
1968                 else
1969                         ll_file_clear_flag(lli, LLIF_FILE_RESTORING);
1970         }
1971
1972         return 0;
1973 }
1974
1975 int ll_read_inode2(struct inode *inode, void *opaque)
1976 {
1977         struct lustre_md *md = opaque;
1978         struct ll_inode_info *lli = ll_i2info(inode);
1979         int     rc;
1980         ENTRY;
1981
1982         CDEBUG(D_VFSTRACE, "VFS Op:inode="DFID"(%p)\n",
1983                PFID(&lli->lli_fid), inode);
1984
1985         /* Core attributes from the MDS first.  This is a new inode, and
1986          * the VFS doesn't zero times in the core inode so we have to do
1987          * it ourselves.  They will be overwritten by either MDS or OST
1988          * attributes - we just need to make sure they aren't newer. */
1989         LTIME_S(inode->i_mtime) = 0;
1990         LTIME_S(inode->i_atime) = 0;
1991         LTIME_S(inode->i_ctime) = 0;
1992         inode->i_rdev = 0;
1993         rc = ll_update_inode(inode, md);
1994         if (rc != 0)
1995                 RETURN(rc);
1996
1997         /* OIDEBUG(inode); */
1998
1999 #ifdef HAVE_BACKING_DEV_INFO
2000         /* initializing backing dev info. */
2001         inode->i_mapping->backing_dev_info = &s2lsi(inode->i_sb)->lsi_bdi;
2002 #endif
2003         if (S_ISREG(inode->i_mode)) {
2004                 struct ll_sb_info *sbi = ll_i2sbi(inode);
2005                 inode->i_op = &ll_file_inode_operations;
2006                 inode->i_fop = sbi->ll_fop;
2007                 inode->i_mapping->a_ops = (struct address_space_operations *)&ll_aops;
2008                 EXIT;
2009         } else if (S_ISDIR(inode->i_mode)) {
2010                 inode->i_op = &ll_dir_inode_operations;
2011                 inode->i_fop = &ll_dir_operations;
2012                 EXIT;
2013         } else if (S_ISLNK(inode->i_mode)) {
2014                 inode->i_op = &ll_fast_symlink_inode_operations;
2015                 EXIT;
2016         } else {
2017                 inode->i_op = &ll_special_inode_operations;
2018
2019                 init_special_inode(inode, inode->i_mode,
2020                                    inode->i_rdev);
2021
2022                 EXIT;
2023         }
2024
2025         return 0;
2026 }
2027
2028 void ll_delete_inode(struct inode *inode)
2029 {
2030         struct ll_inode_info *lli = ll_i2info(inode);
2031         ENTRY;
2032
2033         if (S_ISREG(inode->i_mode) && lli->lli_clob != NULL)
2034                 /* It is last chance to write out dirty pages,
2035                  * otherwise we may lose data while umount */
2036                 cl_sync_file_range(inode, 0, OBD_OBJECT_EOF, CL_FSYNC_LOCAL, 1);
2037
2038         truncate_inode_pages_final(&inode->i_data);
2039
2040         LASSERTF(inode->i_data.nrpages == 0, "inode="DFID"(%p) nrpages=%lu, "
2041                  "see https://jira.hpdd.intel.com/browse/LU-118\n",
2042                  PFID(ll_inode2fid(inode)), inode, inode->i_data.nrpages);
2043
2044 #ifdef HAVE_SBOPS_EVICT_INODE
2045         ll_clear_inode(inode);
2046 #endif
2047         clear_inode(inode);
2048
2049         EXIT;
2050 }
2051
2052 int ll_iocontrol(struct inode *inode, struct file *file,
2053                  unsigned int cmd, unsigned long arg)
2054 {
2055         struct ll_sb_info *sbi = ll_i2sbi(inode);
2056         struct ptlrpc_request *req = NULL;
2057         int rc, flags = 0;
2058         ENTRY;
2059
2060         switch(cmd) {
2061         case FSFILT_IOC_GETFLAGS: {
2062                 struct mdt_body *body;
2063                 struct md_op_data *op_data;
2064
2065                 op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL,
2066                                              0, 0, LUSTRE_OPC_ANY,
2067                                              NULL);
2068                 if (IS_ERR(op_data))
2069                         RETURN(PTR_ERR(op_data));
2070
2071                 op_data->op_valid = OBD_MD_FLFLAGS;
2072                 rc = md_getattr(sbi->ll_md_exp, op_data, &req);
2073                 ll_finish_md_op_data(op_data);
2074                 if (rc) {
2075                         CERROR("%s: failure inode "DFID": rc = %d\n",
2076                                sbi->ll_md_exp->exp_obd->obd_name,
2077                                PFID(ll_inode2fid(inode)), rc);
2078                         RETURN(-abs(rc));
2079                 }
2080
2081                 body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
2082
2083                 flags = body->mbo_flags;
2084
2085                 ptlrpc_req_finished(req);
2086
2087                 RETURN(put_user(flags, (int __user *)arg));
2088         }
2089         case FSFILT_IOC_SETFLAGS: {
2090                 struct iattr *attr;
2091                 struct md_op_data *op_data;
2092                 struct cl_object *obj;
2093
2094                 if (get_user(flags, (int __user *)arg))
2095                         RETURN(-EFAULT);
2096
2097                 op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL, 0, 0,
2098                                              LUSTRE_OPC_ANY, NULL);
2099                 if (IS_ERR(op_data))
2100                         RETURN(PTR_ERR(op_data));
2101
2102                 op_data->op_attr_flags = flags;
2103                 op_data->op_attr.ia_valid |= ATTR_ATTR_FLAG;
2104                 rc = md_setattr(sbi->ll_md_exp, op_data, NULL, 0, &req);
2105                 ll_finish_md_op_data(op_data);
2106                 ptlrpc_req_finished(req);
2107                 if (rc)
2108                         RETURN(rc);
2109
2110                 inode->i_flags = ll_ext_to_inode_flags(flags);
2111
2112                 obj = ll_i2info(inode)->lli_clob;
2113                 if (obj == NULL)
2114                         RETURN(0);
2115
2116                 OBD_ALLOC_PTR(attr);
2117                 if (attr == NULL)
2118                         RETURN(-ENOMEM);
2119
2120                 attr->ia_valid = ATTR_ATTR_FLAG;
2121                 rc = cl_setattr_ost(obj, attr, flags);
2122
2123                 OBD_FREE_PTR(attr);
2124                 RETURN(rc);
2125         }
2126         default:
2127                 RETURN(-ENOSYS);
2128         }
2129
2130         RETURN(0);
2131 }
2132
2133 int ll_flush_ctx(struct inode *inode)
2134 {
2135         struct ll_sb_info  *sbi = ll_i2sbi(inode);
2136
2137         CDEBUG(D_SEC, "flush context for user %d\n",
2138                from_kuid(&init_user_ns, current_uid()));
2139
2140         obd_set_info_async(NULL, sbi->ll_md_exp,
2141                            sizeof(KEY_FLUSH_CTX), KEY_FLUSH_CTX,
2142                            0, NULL, NULL);
2143         obd_set_info_async(NULL, sbi->ll_dt_exp,
2144                            sizeof(KEY_FLUSH_CTX), KEY_FLUSH_CTX,
2145                            0, NULL, NULL);
2146         return 0;
2147 }
2148
2149 /* umount -f client means force down, don't save state */
2150 void ll_umount_begin(struct super_block *sb)
2151 {
2152         struct ll_sb_info *sbi = ll_s2sbi(sb);
2153         struct obd_device *obd;
2154         struct obd_ioctl_data *ioc_data;
2155         struct l_wait_info lwi;
2156         wait_queue_head_t waitq;
2157         ENTRY;
2158
2159         CDEBUG(D_VFSTRACE, "VFS Op: superblock %p count %d active %d\n", sb,
2160                sb->s_count, atomic_read(&sb->s_active));
2161
2162         obd = class_exp2obd(sbi->ll_md_exp);
2163         if (obd == NULL) {
2164                 CERROR("Invalid MDC connection handle %#llx\n",
2165                        sbi->ll_md_exp->exp_handle.h_cookie);
2166                 EXIT;
2167                 return;
2168         }
2169         obd->obd_force = 1;
2170
2171         obd = class_exp2obd(sbi->ll_dt_exp);
2172         if (obd == NULL) {
2173                 CERROR("Invalid LOV connection handle %#llx\n",
2174                        sbi->ll_dt_exp->exp_handle.h_cookie);
2175                 EXIT;
2176                 return;
2177         }
2178         obd->obd_force = 1;
2179
2180         OBD_ALLOC_PTR(ioc_data);
2181         if (ioc_data) {
2182                 obd_iocontrol(IOC_OSC_SET_ACTIVE, sbi->ll_md_exp,
2183                               sizeof *ioc_data, ioc_data, NULL);
2184
2185                 obd_iocontrol(IOC_OSC_SET_ACTIVE, sbi->ll_dt_exp,
2186                               sizeof *ioc_data, ioc_data, NULL);
2187
2188                 OBD_FREE_PTR(ioc_data);
2189         }
2190
2191         /* Really, we'd like to wait until there are no requests outstanding,
2192          * and then continue.  For now, we just periodically checking for vfs
2193          * to decrement mnt_cnt and hope to finish it within 10sec.
2194          */
2195         init_waitqueue_head(&waitq);
2196         lwi = LWI_TIMEOUT_INTERVAL(cfs_time_seconds(10),
2197                                    cfs_time_seconds(1), NULL, NULL);
2198         l_wait_event(waitq, may_umount(sbi->ll_mnt.mnt), &lwi);
2199
2200         EXIT;
2201 }
2202
2203 int ll_remount_fs(struct super_block *sb, int *flags, char *data)
2204 {
2205         struct ll_sb_info *sbi = ll_s2sbi(sb);
2206         char *profilenm = get_profile_name(sb);
2207         int err;
2208         __u32 read_only;
2209
2210         if ((*flags & MS_RDONLY) != (sb->s_flags & MS_RDONLY)) {
2211                 read_only = *flags & MS_RDONLY;
2212                 err = obd_set_info_async(NULL, sbi->ll_md_exp,
2213                                          sizeof(KEY_READ_ONLY),
2214                                          KEY_READ_ONLY, sizeof(read_only),
2215                                          &read_only, NULL);
2216                 if (err) {
2217                         LCONSOLE_WARN("Failed to remount %s %s (%d)\n",
2218                                       profilenm, read_only ?
2219                                       "read-only" : "read-write", err);
2220                         return err;
2221                 }
2222
2223                 if (read_only)
2224                         sb->s_flags |= MS_RDONLY;
2225                 else
2226                         sb->s_flags &= ~MS_RDONLY;
2227
2228                 if (sbi->ll_flags & LL_SBI_VERBOSE)
2229                         LCONSOLE_WARN("Remounted %s %s\n", profilenm,
2230                                       read_only ?  "read-only" : "read-write");
2231         }
2232         return 0;
2233 }
2234
2235 /**
2236  * Cleanup the open handle that is cached on MDT-side.
2237  *
2238  * For open case, the client side open handling thread may hit error
2239  * after the MDT grant the open. Under such case, the client should
2240  * send close RPC to the MDT as cleanup; otherwise, the open handle
2241  * on the MDT will be leaked there until the client umount or evicted.
2242  *
2243  * In further, if someone unlinked the file, because the open handle
2244  * holds the reference on such file/object, then it will block the
2245  * subsequent threads that want to locate such object via FID.
2246  *
2247  * \param[in] sb        super block for this file-system
2248  * \param[in] open_req  pointer to the original open request
2249  */
2250 void ll_open_cleanup(struct super_block *sb, struct ptlrpc_request *open_req)
2251 {
2252         struct mdt_body                 *body;
2253         struct md_op_data               *op_data;
2254         struct ptlrpc_request           *close_req = NULL;
2255         struct obd_export               *exp       = ll_s2sbi(sb)->ll_md_exp;
2256         ENTRY;
2257
2258         body = req_capsule_server_get(&open_req->rq_pill, &RMF_MDT_BODY);
2259         OBD_ALLOC_PTR(op_data);
2260         if (op_data == NULL) {
2261                 CWARN("%s: cannot allocate op_data to release open handle for "
2262                       DFID"\n",
2263                       ll_get_fsname(sb, NULL, 0), PFID(&body->mbo_fid1));
2264
2265                 RETURN_EXIT;
2266         }
2267
2268         op_data->op_fid1 = body->mbo_fid1;
2269         op_data->op_handle = body->mbo_handle;
2270         op_data->op_mod_time = ktime_get_real_seconds();
2271         md_close(exp, op_data, NULL, &close_req);
2272         ptlrpc_req_finished(close_req);
2273         ll_finish_md_op_data(op_data);
2274
2275         EXIT;
2276 }
2277
2278 int ll_prep_inode(struct inode **inode, struct ptlrpc_request *req,
2279                   struct super_block *sb, struct lookup_intent *it)
2280 {
2281         struct ll_sb_info *sbi = NULL;
2282         struct lustre_md md = { NULL };
2283         int rc;
2284         ENTRY;
2285
2286         LASSERT(*inode || sb);
2287         sbi = sb ? ll_s2sbi(sb) : ll_i2sbi(*inode);
2288         rc = md_get_lustre_md(sbi->ll_md_exp, req, sbi->ll_dt_exp,
2289                               sbi->ll_md_exp, &md);
2290         if (rc != 0)
2291                 GOTO(cleanup, rc);
2292
2293         if (*inode) {
2294                 rc = ll_update_inode(*inode, &md);
2295                 if (rc != 0)
2296                         GOTO(out, rc);
2297         } else {
2298                 LASSERT(sb != NULL);
2299
2300                 /*
2301                  * At this point server returns to client's same fid as client
2302                  * generated for creating. So using ->fid1 is okay here.
2303                  */
2304                 if (!fid_is_sane(&md.body->mbo_fid1)) {
2305                         CERROR("%s: Fid is insane "DFID"\n",
2306                                 ll_get_fsname(sb, NULL, 0),
2307                                 PFID(&md.body->mbo_fid1));
2308                         GOTO(out, rc = -EINVAL);
2309                 }
2310
2311                 *inode = ll_iget(sb, cl_fid_build_ino(&md.body->mbo_fid1,
2312                                              sbi->ll_flags & LL_SBI_32BIT_API),
2313                                  &md);
2314                 if (IS_ERR(*inode)) {
2315 #ifdef CONFIG_FS_POSIX_ACL
2316                         if (md.posix_acl) {
2317                                 posix_acl_release(md.posix_acl);
2318                                 md.posix_acl = NULL;
2319                         }
2320 #endif
2321                         rc = IS_ERR(*inode) ? PTR_ERR(*inode) : -ENOMEM;
2322                         *inode = NULL;
2323                         CERROR("new_inode -fatal: rc %d\n", rc);
2324                         GOTO(out, rc);
2325                 }
2326         }
2327
2328         /* Handling piggyback layout lock.
2329          * Layout lock can be piggybacked by getattr and open request.
2330          * The lsm can be applied to inode only if it comes with a layout lock
2331          * otherwise correct layout may be overwritten, for example:
2332          * 1. proc1: mdt returns a lsm but not granting layout
2333          * 2. layout was changed by another client
2334          * 3. proc2: refresh layout and layout lock granted
2335          * 4. proc1: to apply a stale layout */
2336         if (it != NULL && it->it_lock_mode != 0) {
2337                 struct lustre_handle lockh;
2338                 struct ldlm_lock *lock;
2339
2340                 lockh.cookie = it->it_lock_handle;
2341                 lock = ldlm_handle2lock(&lockh);
2342                 LASSERT(lock != NULL);
2343                 if (ldlm_has_layout(lock)) {
2344                         struct cl_object_conf conf;
2345
2346                         memset(&conf, 0, sizeof(conf));
2347                         conf.coc_opc = OBJECT_CONF_SET;
2348                         conf.coc_inode = *inode;
2349                         conf.coc_lock = lock;
2350                         conf.u.coc_layout = md.layout;
2351                         (void)ll_layout_conf(*inode, &conf);
2352                 }
2353                 LDLM_LOCK_PUT(lock);
2354         }
2355
2356         GOTO(out, rc = 0);
2357
2358 out:
2359         md_free_lustre_md(sbi->ll_md_exp, &md);
2360
2361 cleanup:
2362         if (rc != 0 && it != NULL && it->it_op & IT_OPEN)
2363                 ll_open_cleanup(sb != NULL ? sb : (*inode)->i_sb, req);
2364
2365         return rc;
2366 }
2367
2368 int ll_obd_statfs(struct inode *inode, void __user *arg)
2369 {
2370         struct ll_sb_info *sbi = NULL;
2371         struct obd_export *exp;
2372         char *buf = NULL;
2373         struct obd_ioctl_data *data = NULL;
2374         __u32 type;
2375         int len = 0, rc;
2376
2377         if (!inode || !(sbi = ll_i2sbi(inode)))
2378                 GOTO(out_statfs, rc = -EINVAL);
2379
2380         rc = obd_ioctl_getdata(&buf, &len, arg);
2381         if (rc)
2382                 GOTO(out_statfs, rc);
2383
2384         data = (void*)buf;
2385         if (!data->ioc_inlbuf1 || !data->ioc_inlbuf2 ||
2386             !data->ioc_pbuf1 || !data->ioc_pbuf2)
2387                 GOTO(out_statfs, rc = -EINVAL);
2388
2389         if (data->ioc_inllen1 != sizeof(__u32) ||
2390             data->ioc_inllen2 != sizeof(__u32) ||
2391             data->ioc_plen1 != sizeof(struct obd_statfs) ||
2392             data->ioc_plen2 != sizeof(struct obd_uuid))
2393                 GOTO(out_statfs, rc = -EINVAL);
2394
2395         memcpy(&type, data->ioc_inlbuf1, sizeof(__u32));
2396         if (type & LL_STATFS_LMV)
2397                 exp = sbi->ll_md_exp;
2398         else if (type & LL_STATFS_LOV)
2399                 exp = sbi->ll_dt_exp;
2400         else
2401                 GOTO(out_statfs, rc = -ENODEV);
2402
2403         rc = obd_iocontrol(IOC_OBD_STATFS, exp, len, buf, NULL);
2404         if (rc)
2405                 GOTO(out_statfs, rc);
2406 out_statfs:
2407         OBD_FREE_LARGE(buf, len);
2408         return rc;
2409 }
2410
2411 int ll_process_config(struct lustre_cfg *lcfg)
2412 {
2413         struct super_block *sb;
2414         unsigned long x;
2415         int rc = 0;
2416         char *ptr;
2417
2418         /* The instance name contains the sb: lustre-client-aacfe000 */
2419         ptr = strrchr(lustre_cfg_string(lcfg, 0), '-');
2420         if (!ptr || !*(++ptr))
2421                 return -EINVAL;
2422         if (sscanf(ptr, "%lx", &x) != 1)
2423                 return -EINVAL;
2424         sb = (struct super_block *)x;
2425         /* This better be a real Lustre superblock! */
2426         LASSERT(s2lsi(sb)->lsi_lmd->lmd_magic == LMD_MAGIC);
2427
2428         /* Note we have not called client_common_fill_super yet, so
2429            proc fns must be able to handle that! */
2430         rc = class_process_proc_param(PARAM_LLITE, lprocfs_llite_obd_vars,
2431                                       lcfg, sb);
2432         if (rc > 0)
2433                 rc = 0;
2434         return rc;
2435 }
2436
2437 /* this function prepares md_op_data hint for passing it down to MD stack. */
2438 struct md_op_data *ll_prep_md_op_data(struct md_op_data *op_data,
2439                                       struct inode *i1, struct inode *i2,
2440                                       const char *name, size_t namelen,
2441                                       __u32 mode, __u32 opc, void *data)
2442 {
2443         LASSERT(i1 != NULL);
2444
2445         if (name == NULL) {
2446                 /* Do not reuse namelen for something else. */
2447                 if (namelen != 0)
2448                         return ERR_PTR(-EINVAL);
2449         } else {
2450                 if (namelen > ll_i2sbi(i1)->ll_namelen)
2451                         return ERR_PTR(-ENAMETOOLONG);
2452
2453                 if (!lu_name_is_valid_2(name, namelen))
2454                         return ERR_PTR(-EINVAL);
2455         }
2456
2457         if (op_data == NULL)
2458                 OBD_ALLOC_PTR(op_data);
2459
2460         if (op_data == NULL)
2461                 return ERR_PTR(-ENOMEM);
2462
2463         ll_i2gids(op_data->op_suppgids, i1, i2);
2464         op_data->op_fid1 = *ll_inode2fid(i1);
2465         op_data->op_default_stripe_offset = -1;
2466         if (S_ISDIR(i1->i_mode)) {
2467                 op_data->op_mea1 = ll_i2info(i1)->lli_lsm_md;
2468                 if (opc == LUSTRE_OPC_MKDIR)
2469                         op_data->op_default_stripe_offset =
2470                                    ll_i2info(i1)->lli_def_stripe_offset;
2471         }
2472
2473         if (i2) {
2474                 op_data->op_fid2 = *ll_inode2fid(i2);
2475                 if (S_ISDIR(i2->i_mode))
2476                         op_data->op_mea2 = ll_i2info(i2)->lli_lsm_md;
2477         } else {
2478                 fid_zero(&op_data->op_fid2);
2479         }
2480
2481         if (ll_i2sbi(i1)->ll_flags & LL_SBI_64BIT_HASH)
2482                 op_data->op_cli_flags |= CLI_HASH64;
2483
2484         if (ll_need_32bit_api(ll_i2sbi(i1)))
2485                 op_data->op_cli_flags |= CLI_API32;
2486
2487         op_data->op_name = name;
2488         op_data->op_namelen = namelen;
2489         op_data->op_mode = mode;
2490         op_data->op_mod_time = cfs_time_current_sec();
2491         op_data->op_fsuid = from_kuid(&init_user_ns, current_fsuid());
2492         op_data->op_fsgid = from_kgid(&init_user_ns, current_fsgid());
2493         op_data->op_cap = cfs_curproc_cap_pack();
2494         if ((opc == LUSTRE_OPC_CREATE) && (name != NULL) &&
2495              filename_is_volatile(name, namelen, &op_data->op_mds)) {
2496                 op_data->op_bias |= MDS_CREATE_VOLATILE;
2497         } else {
2498                 op_data->op_mds = 0;
2499         }
2500         op_data->op_data = data;
2501
2502         return op_data;
2503 }
2504
2505 void ll_finish_md_op_data(struct md_op_data *op_data)
2506 {
2507         security_release_secctx(op_data->op_file_secctx,
2508                                 op_data->op_file_secctx_size);
2509         OBD_FREE_PTR(op_data);
2510 }
2511
2512 #ifdef HAVE_SUPEROPS_USE_DENTRY
2513 int ll_show_options(struct seq_file *seq, struct dentry *dentry)
2514 #else
2515 int ll_show_options(struct seq_file *seq, struct vfsmount *vfs)
2516 #endif
2517 {
2518         struct ll_sb_info *sbi;
2519
2520 #ifdef HAVE_SUPEROPS_USE_DENTRY
2521         LASSERT((seq != NULL) && (dentry != NULL));
2522         sbi = ll_s2sbi(dentry->d_sb);
2523 #else
2524         LASSERT((seq != NULL) && (vfs != NULL));
2525         sbi = ll_s2sbi(vfs->mnt_sb);
2526 #endif
2527
2528         if (sbi->ll_flags & LL_SBI_NOLCK)
2529                 seq_puts(seq, ",nolock");
2530
2531         if (sbi->ll_flags & LL_SBI_FLOCK)
2532                 seq_puts(seq, ",flock");
2533
2534         if (sbi->ll_flags & LL_SBI_LOCALFLOCK)
2535                 seq_puts(seq, ",localflock");
2536
2537         if (sbi->ll_flags & LL_SBI_USER_XATTR)
2538                 seq_puts(seq, ",user_xattr");
2539
2540         if (sbi->ll_flags & LL_SBI_LAZYSTATFS)
2541                 seq_puts(seq, ",lazystatfs");
2542
2543         if (sbi->ll_flags & LL_SBI_USER_FID2PATH)
2544                 seq_puts(seq, ",user_fid2path");
2545
2546         if (sbi->ll_flags & LL_SBI_ALWAYS_PING)
2547                 seq_puts(seq, ",always_ping");
2548
2549         RETURN(0);
2550 }
2551
2552 /**
2553  * Get obd name by cmd, and copy out to user space
2554  */
2555 int ll_get_obd_name(struct inode *inode, unsigned int cmd, unsigned long arg)
2556 {
2557         struct ll_sb_info *sbi = ll_i2sbi(inode);
2558         struct obd_device *obd;
2559         ENTRY;
2560
2561         if (cmd == OBD_IOC_GETDTNAME)
2562                 obd = class_exp2obd(sbi->ll_dt_exp);
2563         else if (cmd == OBD_IOC_GETMDNAME)
2564                 obd = class_exp2obd(sbi->ll_md_exp);
2565         else
2566                 RETURN(-EINVAL);
2567
2568         if (!obd)
2569                 RETURN(-ENOENT);
2570
2571         if (copy_to_user((void __user *)arg, obd->obd_name,
2572                          strlen(obd->obd_name) + 1))
2573                 RETURN(-EFAULT);
2574
2575         RETURN(0);
2576 }
2577
2578 /**
2579  * Get lustre file system name by \a sbi. If \a buf is provided(non-NULL), the
2580  * fsname will be returned in this buffer; otherwise, a static buffer will be
2581  * used to store the fsname and returned to caller.
2582  */
2583 char *ll_get_fsname(struct super_block *sb, char *buf, int buflen)
2584 {
2585         static char fsname_static[MTI_NAME_MAXLEN];
2586         struct lustre_sb_info *lsi = s2lsi(sb);
2587         char *ptr;
2588         int len;
2589
2590         if (buf == NULL) {
2591                 /* this means the caller wants to use static buffer
2592                  * and it doesn't care about race. Usually this is
2593                  * in error reporting path */
2594                 buf = fsname_static;
2595                 buflen = sizeof(fsname_static);
2596         }
2597
2598         len = strlen(lsi->lsi_lmd->lmd_profile);
2599         ptr = strrchr(lsi->lsi_lmd->lmd_profile, '-');
2600         if (ptr && (strcmp(ptr, "-client") == 0))
2601                 len -= 7;
2602
2603         if (unlikely(len >= buflen))
2604                 len = buflen - 1;
2605         strncpy(buf, lsi->lsi_lmd->lmd_profile, len);
2606         buf[len] = '\0';
2607
2608         return buf;
2609 }
2610
2611 static char* ll_d_path(struct dentry *dentry, char *buf, int bufsize)
2612 {
2613         char *path = NULL;
2614
2615         struct path p;
2616
2617         p.dentry = dentry;
2618         p.mnt = current->fs->root.mnt;
2619         path_get(&p);
2620         path = d_path(&p, buf, bufsize);
2621         path_put(&p);
2622         return path;
2623 }
2624
2625 void ll_dirty_page_discard_warn(struct page *page, int ioret)
2626 {
2627         char *buf, *path = NULL;
2628         struct dentry *dentry = NULL;
2629         struct inode *inode = page->mapping->host;
2630
2631         /* this can be called inside spin lock so use GFP_ATOMIC. */
2632         buf = (char *)__get_free_page(GFP_ATOMIC);
2633         if (buf != NULL) {
2634                 dentry = d_find_alias(page->mapping->host);
2635                 if (dentry != NULL)
2636                         path = ll_d_path(dentry, buf, PAGE_SIZE);
2637         }
2638
2639         CDEBUG(D_WARNING,
2640                "%s: dirty page discard: %s/fid: "DFID"/%s may get corrupted "
2641                "(rc %d)\n", ll_get_fsname(page->mapping->host->i_sb, NULL, 0),
2642                s2lsi(page->mapping->host->i_sb)->lsi_lmd->lmd_dev,
2643                PFID(ll_inode2fid(inode)),
2644                (path && !IS_ERR(path)) ? path : "", ioret);
2645
2646         if (dentry != NULL)
2647                 dput(dentry);
2648
2649         if (buf != NULL)
2650                 free_page((unsigned long)buf);
2651 }
2652
2653 ssize_t ll_copy_user_md(const struct lov_user_md __user *md,
2654                         struct lov_user_md **kbuf)
2655 {
2656         struct lov_user_md      lum;
2657         ssize_t                 lum_size;
2658         ENTRY;
2659
2660         if (copy_from_user(&lum, md, sizeof(lum)))
2661                 RETURN(-EFAULT);
2662
2663         lum_size = ll_lov_user_md_size(&lum);
2664         if (lum_size < 0)
2665                 RETURN(lum_size);
2666
2667         OBD_ALLOC(*kbuf, lum_size);
2668         if (*kbuf == NULL)
2669                 RETURN(-ENOMEM);
2670
2671         if (copy_from_user(*kbuf, md, lum_size) != 0) {
2672                 OBD_FREE(*kbuf, lum_size);
2673                 RETURN(-EFAULT);
2674         }
2675
2676         RETURN(lum_size);
2677 }
2678
2679 /*
2680  * Compute llite root squash state after a change of root squash
2681  * configuration setting or add/remove of a lnet nid
2682  */
2683 void ll_compute_rootsquash_state(struct ll_sb_info *sbi)
2684 {
2685         struct root_squash_info *squash = &sbi->ll_squash;
2686         int i;
2687         bool matched;
2688         struct lnet_process_id id;
2689
2690         /* Update norootsquash flag */
2691         down_write(&squash->rsi_sem);
2692         if (list_empty(&squash->rsi_nosquash_nids))
2693                 sbi->ll_flags &= ~LL_SBI_NOROOTSQUASH;
2694         else {
2695                 /* Do not apply root squash as soon as one of our NIDs is
2696                  * in the nosquash_nids list */
2697                 matched = false;
2698                 i = 0;
2699                 while (LNetGetId(i++, &id) != -ENOENT) {
2700                         if (LNET_NETTYP(LNET_NIDNET(id.nid)) == LOLND)
2701                                 continue;
2702                         if (cfs_match_nid(id.nid, &squash->rsi_nosquash_nids)) {
2703                                 matched = true;
2704                                 break;
2705                         }
2706                 }
2707                 if (matched)
2708                         sbi->ll_flags |= LL_SBI_NOROOTSQUASH;
2709                 else
2710                         sbi->ll_flags &= ~LL_SBI_NOROOTSQUASH;
2711         }
2712         up_write(&squash->rsi_sem);
2713 }
2714
2715 /**
2716  * Parse linkea content to extract information about a given hardlink
2717  *
2718  * \param[in]   ldata      - Initialized linkea data
2719  * \param[in]   linkno     - Link identifier
2720  * \param[out]  parent_fid - The entry's parent FID
2721  * \param[out]  ln         - Entry name destination buffer
2722  *
2723  * \retval 0 on success
2724  * \retval Appropriate negative error code on failure
2725  */
2726 static int ll_linkea_decode(struct linkea_data *ldata, unsigned int linkno,
2727                             struct lu_fid *parent_fid, struct lu_name *ln)
2728 {
2729         unsigned int    idx;
2730         int             rc;
2731         ENTRY;
2732
2733         rc = linkea_init_with_rec(ldata);
2734         if (rc < 0)
2735                 RETURN(rc);
2736
2737         if (linkno >= ldata->ld_leh->leh_reccount)
2738                 /* beyond last link */
2739                 RETURN(-ENODATA);
2740
2741         linkea_first_entry(ldata);
2742         for (idx = 0; ldata->ld_lee != NULL; idx++) {
2743                 linkea_entry_unpack(ldata->ld_lee, &ldata->ld_reclen, ln,
2744                                     parent_fid);
2745                 if (idx == linkno)
2746                         break;
2747
2748                 linkea_next_entry(ldata);
2749         }
2750
2751         if (idx < linkno)
2752                 RETURN(-ENODATA);
2753
2754         RETURN(0);
2755 }
2756
2757 /**
2758  * Get parent FID and name of an identified link. Operation is performed for
2759  * a given link number, letting the caller iterate over linkno to list one or
2760  * all links of an entry.
2761  *
2762  * \param[in]     file - File descriptor against which to perform the operation
2763  * \param[in,out] arg  - User-filled structure containing the linkno to operate
2764  *                       on and the available size. It is eventually filled with
2765  *                       the requested information or left untouched on error
2766  *
2767  * \retval - 0 on success
2768  * \retval - Appropriate negative error code on failure
2769  */
2770 int ll_getparent(struct file *file, struct getparent __user *arg)
2771 {
2772         struct inode            *inode = file_inode(file);
2773         struct linkea_data      *ldata;
2774         struct lu_buf            buf = LU_BUF_NULL;
2775         struct lu_name           ln;
2776         struct lu_fid            parent_fid;
2777         __u32                    linkno;
2778         __u32                    name_size;
2779         int                      rc;
2780
2781         ENTRY;
2782
2783         if (!cfs_capable(CFS_CAP_DAC_READ_SEARCH) &&
2784             !(ll_i2sbi(inode)->ll_flags & LL_SBI_USER_FID2PATH))
2785                 RETURN(-EPERM);
2786
2787         if (get_user(name_size, &arg->gp_name_size))
2788                 RETURN(-EFAULT);
2789
2790         if (get_user(linkno, &arg->gp_linkno))
2791                 RETURN(-EFAULT);
2792
2793         if (name_size > PATH_MAX)
2794                 RETURN(-EINVAL);
2795
2796         OBD_ALLOC(ldata, sizeof(*ldata));
2797         if (ldata == NULL)
2798                 RETURN(-ENOMEM);
2799
2800         rc = linkea_data_new(ldata, &buf);
2801         if (rc < 0)
2802                 GOTO(ldata_free, rc);
2803
2804 #ifdef HAVE_XATTR_HANDLER_FLAGS
2805         rc = ll_xattr_list(inode, XATTR_NAME_LINK, XATTR_TRUSTED_T, buf.lb_buf,
2806                            buf.lb_len, OBD_MD_FLXATTR);
2807 #else
2808         rc = ll_getxattr(file_dentry(file), XATTR_NAME_LINK, buf.lb_buf,
2809                          buf.lb_len);
2810 #endif /* HAVE_XATTR_HANDLER_FLAGS */
2811         if (rc < 0)
2812                 GOTO(lb_free, rc);
2813
2814         rc = ll_linkea_decode(ldata, linkno, &parent_fid, &ln);
2815         if (rc < 0)
2816                 GOTO(lb_free, rc);
2817
2818         if (ln.ln_namelen >= name_size)
2819                 GOTO(lb_free, rc = -EOVERFLOW);
2820
2821         if (copy_to_user(&arg->gp_fid, &parent_fid, sizeof(arg->gp_fid)))
2822                 GOTO(lb_free, rc = -EFAULT);
2823
2824         if (copy_to_user(&arg->gp_name, ln.ln_name, ln.ln_namelen))
2825                 GOTO(lb_free, rc = -EFAULT);
2826
2827         if (put_user('\0', arg->gp_name + ln.ln_namelen))
2828                 GOTO(lb_free, rc = -EFAULT);
2829
2830 lb_free:
2831         lu_buf_free(&buf);
2832 ldata_free:
2833         OBD_FREE(ldata, sizeof(*ldata));
2834
2835         RETURN(rc);
2836 }