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         /* pass client page size via ocd_grant_blkbits, the server should report
 200          * back its backend blocksize for grant calculation purpose */
 201         data->ocd_grant_blkbits = PAGE_SHIFT;
 202
 203         /* indicate MDT features supported by this client */
 204         data->ocd_connect_flags = OBD_CONNECT_IBITS    | OBD_CONNECT_NODEVOH  |
 205                                   OBD_CONNECT_ATTRFID  | OBD_CONNECT_GRANT |
 206                                   OBD_CONNECT_VERSION  | OBD_CONNECT_BRW_SIZE |
 207                                   OBD_CONNECT_SRVLOCK  | OBD_CONNECT_TRUNCLOCK|
 208                                   OBD_CONNECT_MDS_CAPA | OBD_CONNECT_OSS_CAPA |
 209                                   OBD_CONNECT_CANCELSET | OBD_CONNECT_FID     |
 210                                   OBD_CONNECT_AT       | OBD_CONNECT_LOV_V3   |
 211                                   OBD_CONNECT_VBR | OBD_CONNECT_FULL20 |
 212                                   OBD_CONNECT_64BITHASH |
 213                                   OBD_CONNECT_EINPROGRESS |
 214                                   OBD_CONNECT_JOBSTATS | OBD_CONNECT_LVB_TYPE |
 215                                   OBD_CONNECT_LAYOUTLOCK | OBD_CONNECT_PINGLESS|
 216                                   OBD_CONNECT_MAX_EASIZE |
 217                                   OBD_CONNECT_FLOCK_DEAD |
 218                                   OBD_CONNECT_DISP_STRIPE | OBD_CONNECT_LFSCK |
 219                                   OBD_CONNECT_OPEN_BY_FID |
 220                                   OBD_CONNECT_DIR_STRIPE |
 221                                   OBD_CONNECT_BULK_MBITS | OBD_CONNECT_CKSUM |
 222                                   OBD_CONNECT_SUBTREE |
 223                                   OBD_CONNECT_MULTIMODRPCS |
 224                                   OBD_CONNECT_GRANT_PARAM | OBD_CONNECT_FLAGS2;
 225
 226         data->ocd_connect_flags2 = 0;
 227
 228 #ifdef HAVE_LRU_RESIZE_SUPPORT
 229         if (sbi->ll_flags & LL_SBI_LRU_RESIZE)
 230                 data->ocd_connect_flags |= OBD_CONNECT_LRU_RESIZE;
 231 #endif
 232 #ifdef CONFIG_FS_POSIX_ACL
 233         data->ocd_connect_flags |= OBD_CONNECT_ACL | OBD_CONNECT_UMASK |
 234                                    OBD_CONNECT_LARGE_ACL;
 235 #endif
 236
 237         data->ocd_cksum_types = cksum_types_supported_client();
 238
 239         if (OBD_FAIL_CHECK(OBD_FAIL_MDC_LIGHTWEIGHT))
 240                 /* flag mdc connection as lightweight, only used for test
 241                  * purpose, use with care */
 242                 data->ocd_connect_flags |= OBD_CONNECT_LIGHTWEIGHT;
 243
 244         data->ocd_ibits_known = MDS_INODELOCK_FULL;
 245         data->ocd_version = LUSTRE_VERSION_CODE;
 246
 247         if (sb->s_flags & MS_RDONLY)
 248                 data->ocd_connect_flags |= OBD_CONNECT_RDONLY;
 249         if (sbi->ll_flags & LL_SBI_USER_XATTR)
 250                 data->ocd_connect_flags |= OBD_CONNECT_XATTR;
 251
 252 #ifdef MS_NOSEC
 253         /* Setting this indicates we correctly support S_NOSEC (See kernel
 254          * commit 9e1f1de02c2275d7172e18dc4e7c2065777611bf)
 255          */
 256         sb->s_flags |= MS_NOSEC;
 257 #endif
 258
 259         if (sbi->ll_flags & LL_SBI_FLOCK)
 260                 sbi->ll_fop = &ll_file_operations_flock;
 261         else if (sbi->ll_flags & LL_SBI_LOCALFLOCK)
 262                 sbi->ll_fop = &ll_file_operations;
 263         else
 264                 sbi->ll_fop = &ll_file_operations_noflock;
 265
 266         /* always ping even if server suppress_pings */
 267         if (sbi->ll_flags & LL_SBI_ALWAYS_PING)
 268                 data->ocd_connect_flags &= ~OBD_CONNECT_PINGLESS;
 269
 270 #ifdef HAVE_SECURITY_DENTRY_INIT_SECURITY
 271         data->ocd_connect_flags2 |= OBD_CONNECT2_FILE_SECCTX;
 272 #endif /* HAVE_SECURITY_DENTRY_INIT_SECURITY */
 273
 274         data->ocd_brw_size = MD_MAX_BRW_SIZE;
 275
 276         err = obd_connect(NULL, &sbi->ll_md_exp, obd, &sbi->ll_sb_uuid, data, NULL);
 277         if (err == -EBUSY) {
 278                 LCONSOLE_ERROR_MSG(0x14f, "An MDT (md %s) is performing "
 279                                    "recovery, of which this client is not a "
 280                                    "part. Please wait for recovery to complete,"
 281                                    " abort, or time out.\n", md);
 282                 GOTO(out, err);
 283         } else if (err) {
 284                 CERROR("cannot connect to %s: rc = %d\n", md, err);
 285                 GOTO(out, err);
 286         }
 287
 288         sbi->ll_md_exp->exp_connect_data = *data;
 289
 290         err = obd_fid_init(sbi->ll_md_exp->exp_obd, sbi->ll_md_exp,
 291                            LUSTRE_SEQ_METADATA);
 292         if (err) {
 293                 CERROR("%s: Can't init metadata layer FID infrastructure, "
 294                        "rc = %d\n", sbi->ll_md_exp->exp_obd->obd_name, err);
 295                 GOTO(out_md, err);
 296         }
 297
 298         /* For mount, we only need fs info from MDT0, and also in DNE, it
 299          * can make sure the client can be mounted as long as MDT0 is
 300          * avaible */
 301         err = obd_statfs(NULL, sbi->ll_md_exp, osfs,
 302                         cfs_time_shift_64(-OBD_STATFS_CACHE_SECONDS),
 303                         OBD_STATFS_FOR_MDT0);
 304         if (err)
 305                 GOTO(out_md_fid, err);
 306
 307         /* This needs to be after statfs to ensure connect has finished.
 308          * Note that "data" does NOT contain the valid connect reply.
 309          * If connecting to a 1.8 server there will be no LMV device, so
 310          * we can access the MDC export directly and exp_connect_flags will
 311          * be non-zero, but if accessing an upgraded 2.1 server it will
 312          * have the correct flags filled in.
 313          * XXX: fill in the LMV exp_connect_flags from MDC(s). */
 314         valid = exp_connect_flags(sbi->ll_md_exp) & CLIENT_CONNECT_MDT_REQD;
 315         if (exp_connect_flags(sbi->ll_md_exp) != 0 &&
 316             valid != CLIENT_CONNECT_MDT_REQD) {
 317                 char *buf;
 318
 319                 OBD_ALLOC_WAIT(buf, PAGE_SIZE);
 320                 obd_connect_flags2str(buf, PAGE_SIZE,
 321                                       valid ^ CLIENT_CONNECT_MDT_REQD, 0, ",");
 322                 LCONSOLE_ERROR_MSG(0x170, "Server %s does not support "
 323                                    "feature(s) needed for correct operation "
 324                                    "of this client (%s). Please upgrade "
 325                                    "server or downgrade client.\n",
 326                                    sbi->ll_md_exp->exp_obd->obd_name, buf);
 327                 OBD_FREE(buf, PAGE_SIZE);
 328                 GOTO(out_md_fid, err = -EPROTO);
 329         }
 330
 331         size = sizeof(*data);
 332         err = obd_get_info(NULL, sbi->ll_md_exp, sizeof(KEY_CONN_DATA),
 333                            KEY_CONN_DATA,  &size, data);
 334         if (err) {
 335                 CERROR("%s: Get connect data failed: rc = %d\n",
 336                        sbi->ll_md_exp->exp_obd->obd_name, err);
 337                 GOTO(out_md_fid, err);
 338         }
 339
 340         LASSERT(osfs->os_bsize);
 341         sb->s_blocksize = osfs->os_bsize;
 342         sb->s_blocksize_bits = log2(osfs->os_bsize);
 343         sb->s_magic = LL_SUPER_MAGIC;
 344         sb->s_maxbytes = MAX_LFS_FILESIZE;
 345         sbi->ll_namelen = osfs->os_namelen;
 346         sbi->ll_mnt.mnt = current->fs->root.mnt;
 347
 348         if ((sbi->ll_flags & LL_SBI_USER_XATTR) &&
 349             !(data->ocd_connect_flags & OBD_CONNECT_XATTR)) {
 350                 LCONSOLE_INFO("Disabling user_xattr feature because "
 351                               "it is not supported on the server\n");
 352                 sbi->ll_flags &= ~LL_SBI_USER_XATTR;
 353         }
 354
 355         if (data->ocd_connect_flags & OBD_CONNECT_ACL) {
 356 #ifdef MS_POSIXACL
 357                 sb->s_flags |= MS_POSIXACL;
 358 #endif
 359                 sbi->ll_flags |= LL_SBI_ACL;
 360         } else {
 361                 LCONSOLE_INFO("client wants to enable acl, but mdt not!\n");
 362 #ifdef MS_POSIXACL
 363                 sb->s_flags &= ~MS_POSIXACL;
 364 #endif
 365                 sbi->ll_flags &= ~LL_SBI_ACL;
 366         }
 367
 368         if (data->ocd_connect_flags & OBD_CONNECT_64BITHASH)
 369                 sbi->ll_flags |= LL_SBI_64BIT_HASH;
 370
 371         if (data->ocd_connect_flags & OBD_CONNECT_LAYOUTLOCK)
 372                 sbi->ll_flags |= LL_SBI_LAYOUT_LOCK;
 373
 374         if (obd_connect_has_secctx(data))
 375                 sbi->ll_flags |= LL_SBI_FILE_SECCTX;
 376
 377         if (data->ocd_ibits_known & MDS_INODELOCK_XATTR) {
 378                 if (!(data->ocd_connect_flags & OBD_CONNECT_MAX_EASIZE)) {
 379                         LCONSOLE_INFO("%s: disabling xattr cache due to "
 380                                       "unknown maximum xattr size.\n", dt);
 381                 } else if (!sbi->ll_xattr_cache_set) {
 382                         /* If xattr_cache is already set (no matter 0 or 1)
 383                          * during processing llog, it won't be enabled here. */
 384                         sbi->ll_flags |= LL_SBI_XATTR_CACHE;
 385                         sbi->ll_xattr_cache_enabled = 1;
 386                 }
 387         }
 388
 389         obd = class_name2obd(dt);
 390         if (!obd) {
 391                 CERROR("DT %s: not setup or attached\n", dt);
 392                 GOTO(out_md_fid, err = -ENODEV);
 393         }
 394
 395         /* pass client page size via ocd_grant_blkbits, the server should report
 396          * back its backend blocksize for grant calculation purpose */
 397         data->ocd_grant_blkbits = PAGE_SHIFT;
 398
 399         /* indicate OST features supported by this client */
 400         data->ocd_connect_flags = OBD_CONNECT_GRANT | OBD_CONNECT_VERSION |
 401                                   OBD_CONNECT_REQPORTAL | OBD_CONNECT_BRW_SIZE |
 402                                   OBD_CONNECT_CANCELSET | OBD_CONNECT_FID |
 403                                   OBD_CONNECT_SRVLOCK | OBD_CONNECT_TRUNCLOCK|
 404                                   OBD_CONNECT_AT | OBD_CONNECT_OSS_CAPA |
 405                                   OBD_CONNECT_VBR | OBD_CONNECT_FULL20 |
 406                                   OBD_CONNECT_64BITHASH | OBD_CONNECT_MAXBYTES |
 407                                   OBD_CONNECT_EINPROGRESS |
 408                                   OBD_CONNECT_JOBSTATS | OBD_CONNECT_LVB_TYPE |
 409                                   OBD_CONNECT_LAYOUTLOCK |
 410                                   OBD_CONNECT_PINGLESS | OBD_CONNECT_LFSCK |
 411                                   OBD_CONNECT_BULK_MBITS |
 412                                   OBD_CONNECT_FLAGS2;
 413
 414 /* The client currently advertises support for OBD_CONNECT_LOCKAHEAD_OLD so it
 415  * can interoperate with an older version of lockahead which was released prior
 416  * to landing in master. This support will be dropped when 2.13 development
 417  * starts.  At the point, we should not just drop the connect flag (below), we
 418  * should also remove the support in the code.
 419  *
 420  * Removing it means a few things:
 421  * 1. Remove this section here
 422  * 2. Remove CEF_NONBLOCK in ll_file_lockahead()
 423  * 3. Remove function exp_connect_lockahead_old
 424  * 4. Remove LDLM_FL_LOCKAHEAD_OLD_RESERVED in lustre_dlm_flags.h
 425  * */
 426 #if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(2, 12, 50, 0)
 427         data->ocd_connect_flags |= OBD_CONNECT_LOCKAHEAD_OLD;
 428 #endif
 429
 430         data->ocd_connect_flags2 = OBD_CONNECT2_LOCKAHEAD;
 431
 432         if (!OBD_FAIL_CHECK(OBD_FAIL_OSC_CONNECT_GRANT_PARAM))
 433                 data->ocd_connect_flags |= OBD_CONNECT_GRANT_PARAM;
 434
 435         /* OBD_CONNECT_CKSUM should always be set, even if checksums are
 436          * disabled by default, because it can still be enabled on the
 437          * fly via /proc. As a consequence, we still need to come to an
 438          * agreement on the supported algorithms at connect time */
 439         data->ocd_connect_flags |= OBD_CONNECT_CKSUM;
 440
 441         if (OBD_FAIL_CHECK(OBD_FAIL_OSC_CKSUM_ADLER_ONLY))
 442                 data->ocd_cksum_types = OBD_CKSUM_ADLER;
 443         else
 444                 data->ocd_cksum_types = cksum_types_supported_client();
 445
 446 #ifdef HAVE_LRU_RESIZE_SUPPORT
 447         data->ocd_connect_flags |= OBD_CONNECT_LRU_RESIZE;
 448 #endif
 449         /* always ping even if server suppress_pings */
 450         if (sbi->ll_flags & LL_SBI_ALWAYS_PING)
 451                 data->ocd_connect_flags &= ~OBD_CONNECT_PINGLESS;
 452
 453         CDEBUG(D_RPCTRACE, "ocd_connect_flags: %#llx ocd_version: %d "
 454                "ocd_grant: %d\n", data->ocd_connect_flags,
 455                data->ocd_version, data->ocd_grant);
 456
 457         obd->obd_upcall.onu_owner = &sbi->ll_lco;
 458         obd->obd_upcall.onu_upcall = cl_ocd_update;
 459
 460         data->ocd_brw_size = DT_MAX_BRW_SIZE;
 461
 462         err = obd_connect(NULL, &sbi->ll_dt_exp, obd, &sbi->ll_sb_uuid, data,
 463                           NULL);
 464         if (err == -EBUSY) {
 465                 LCONSOLE_ERROR_MSG(0x150, "An OST (dt %s) is performing "
 466                                    "recovery, of which this client is not a "
 467                                    "part.  Please wait for recovery to "
 468                                    "complete, abort, or time out.\n", dt);
 469                 GOTO(out_md, err);
 470         } else if (err) {
 471                 CERROR("%s: Cannot connect to %s: rc = %d\n",
 472                        sbi->ll_dt_exp->exp_obd->obd_name, dt, err);
 473                 GOTO(out_md, err);
 474         }
 475
 476         sbi->ll_dt_exp->exp_connect_data = *data;
 477
 478         err = obd_fid_init(sbi->ll_dt_exp->exp_obd, sbi->ll_dt_exp,
 479                            LUSTRE_SEQ_METADATA);
 480         if (err) {
 481                 CERROR("%s: Can't init data layer FID infrastructure, "
 482                        "rc = %d\n", sbi->ll_dt_exp->exp_obd->obd_name, err);
 483                 GOTO(out_dt, err);
 484         }
 485
 486         mutex_lock(&sbi->ll_lco.lco_lock);
 487         sbi->ll_lco.lco_flags = data->ocd_connect_flags;
 488         sbi->ll_lco.lco_md_exp = sbi->ll_md_exp;
 489         sbi->ll_lco.lco_dt_exp = sbi->ll_dt_exp;
 490         mutex_unlock(&sbi->ll_lco.lco_lock);
 491
 492         fid_zero(&sbi->ll_root_fid);
 493         err = md_get_root(sbi->ll_md_exp, get_mount_fileset(sb),
 494                            &sbi->ll_root_fid);
 495         if (err) {
 496                 CERROR("cannot mds_connect: rc = %d\n", err);
 497                 GOTO(out_lock_cn_cb, err);
 498         }
 499         if (!fid_is_sane(&sbi->ll_root_fid)) {
 500                 CERROR("%s: Invalid root fid "DFID" during mount\n",
 501                        sbi->ll_md_exp->exp_obd->obd_name,
 502                        PFID(&sbi->ll_root_fid));
 503                 GOTO(out_lock_cn_cb, err = -EINVAL);
 504         }
 505         CDEBUG(D_SUPER, "rootfid "DFID"\n", PFID(&sbi->ll_root_fid));
 506
 507         sb->s_op = &lustre_super_operations;
 508 #ifdef HAVE_XATTR_HANDLER_FLAGS
 509         sb->s_xattr = ll_xattr_handlers;
 510 #endif
 511 #if THREAD_SIZE >= 8192 /*b=17630*/
 512         sb->s_export_op = &lustre_export_operations;
 513 #endif
 514
 515         /* make root inode
 516          * XXX: move this to after cbd setup? */
 517         valid = OBD_MD_FLGETATTR | OBD_MD_FLBLOCKS | OBD_MD_FLMODEASIZE;
 518         if (sbi->ll_flags & LL_SBI_ACL)
 519                 valid |= OBD_MD_FLACL;
 520
 521         OBD_ALLOC_PTR(op_data);
 522         if (op_data == NULL)
 523                 GOTO(out_lock_cn_cb, err = -ENOMEM);
 524
 525         op_data->op_fid1 = sbi->ll_root_fid;
 526         op_data->op_mode = 0;
 527         op_data->op_valid = valid;
 528
 529         err = md_getattr(sbi->ll_md_exp, op_data, &request);
 530
 531         OBD_FREE_PTR(op_data);
 532         if (err) {
 533                 CERROR("%s: md_getattr failed for root: rc = %d\n",
 534                        sbi->ll_md_exp->exp_obd->obd_name, err);
 535                 GOTO(out_lock_cn_cb, err);
 536         }
 537
 538         err = md_get_lustre_md(sbi->ll_md_exp, request, sbi->ll_dt_exp,
 539                                sbi->ll_md_exp, &lmd);
 540         if (err) {
 541                 CERROR("failed to understand root inode md: rc = %d\n", err);
 542                 ptlrpc_req_finished(request);
 543                 GOTO(out_lock_cn_cb, err);
 544         }
 545
 546         LASSERT(fid_is_sane(&sbi->ll_root_fid));
 547         root = ll_iget(sb, cl_fid_build_ino(&sbi->ll_root_fid,
 548                                             sbi->ll_flags & LL_SBI_32BIT_API),
 549                        &lmd);
 550         md_free_lustre_md(sbi->ll_md_exp, &lmd);
 551         ptlrpc_req_finished(request);
 552
 553         if (IS_ERR(root)) {
 554 #ifdef CONFIG_FS_POSIX_ACL
 555                 if (lmd.posix_acl) {
 556                         posix_acl_release(lmd.posix_acl);
 557                         lmd.posix_acl = NULL;
 558                 }
 559 #endif
 560                 err = IS_ERR(root) ? PTR_ERR(root) : -EBADF;
 561                 root = NULL;
 562                 CERROR("lustre_lite: bad iget4 for root\n");
 563                 GOTO(out_root, err);
 564         }
 565
 566         checksum = sbi->ll_flags & LL_SBI_CHECKSUM;
 567         err = obd_set_info_async(NULL, sbi->ll_dt_exp, sizeof(KEY_CHECKSUM),
 568                                  KEY_CHECKSUM, sizeof(checksum), &checksum,
 569                                  NULL);
 570         if (err) {
 571                 CERROR("%s: Set checksum failed: rc = %d\n",
 572                        sbi->ll_dt_exp->exp_obd->obd_name, err);
 573                 GOTO(out_root, err);
 574         }
 575         cl_sb_init(sb);
 576
 577         err = obd_set_info_async(NULL, sbi->ll_dt_exp, sizeof(KEY_CACHE_SET),
 578                                  KEY_CACHE_SET, sizeof(*sbi->ll_cache),
 579                                  sbi->ll_cache, NULL);
 580         if (err) {
 581                 CERROR("%s: Set cache_set failed: rc = %d\n",
 582                        sbi->ll_dt_exp->exp_obd->obd_name, err);
 583                 GOTO(out_root, err);
 584         }
 585
 586         sb->s_root = d_make_root(root);
 587         if (sb->s_root == NULL) {
 588                 CERROR("%s: can't make root dentry\n",
 589                         ll_get_fsname(sb, NULL, 0));
 590                 GOTO(out_root, err = -ENOMEM);
 591         }
 592 #ifdef HAVE_DCACHE_LOCK
 593         sb->s_root->d_op = &ll_d_ops;
 594 #endif
 595
 596         sbi->ll_sdev_orig = sb->s_dev;
 597
 598         /* We set sb->s_dev equal on all lustre clients in order to support
 599          * NFS export clustering.  NFSD requires that the FSID be the same
 600          * on all clients. */
 601         /* s_dev is also used in lt_compare() to compare two fs, but that is
 602          * only a node-local comparison. */
 603         uuid = obd_get_uuid(sbi->ll_md_exp);
 604         if (uuid != NULL)
 605                 sb->s_dev = get_uuid2int(uuid->uuid, strlen(uuid->uuid));
 606
 607         if (data != NULL)
 608                 OBD_FREE_PTR(data);
 609         if (osfs != NULL)
 610                 OBD_FREE_PTR(osfs);
 611
 612         if (sbi->ll_proc_root != NULL) {
 613                 err = lprocfs_ll_register_obd(sb, dt);
 614                 if (err < 0) {
 615                         CERROR("%s: could not register %s in llite: rc = %d\n",
 616                                dt, ll_get_fsname(sb, NULL, 0), err);
 617                         err = 0;
 618                 }
 619                 err = lprocfs_ll_register_obd(sb, md);
 620                 if (err < 0) {
 621                         CERROR("%s: could not register %s in llite: rc = %d\n",
 622                                md, ll_get_fsname(sb, NULL, 0), err);
 623                         err = 0;
 624                 }
 625         }
 626
 627         RETURN(err);
 628 out_root:
 629         if (root)
 630                 iput(root);
 631 out_lock_cn_cb:
 632         obd_fid_fini(sbi->ll_dt_exp->exp_obd);
 633 out_dt:
 634         obd_disconnect(sbi->ll_dt_exp);
 635         sbi->ll_dt_exp = NULL;
 636 out_md_fid:
 637         obd_fid_fini(sbi->ll_md_exp->exp_obd);
 638 out_md:
 639         obd_disconnect(sbi->ll_md_exp);
 640         sbi->ll_md_exp = NULL;
 641 out:
 642         if (data != NULL)
 643                 OBD_FREE_PTR(data);
 644         if (osfs != NULL)
 645                 OBD_FREE_PTR(osfs);
 646         return err;
 647 }
 648
 649 int ll_get_max_mdsize(struct ll_sb_info *sbi, int *lmmsize)
 650 {
 651         int size, rc;
 652
 653         size = sizeof(*lmmsize);
 654         rc = obd_get_info(NULL, sbi->ll_dt_exp, sizeof(KEY_MAX_EASIZE),
 655                           KEY_MAX_EASIZE, &size, lmmsize);
 656         if (rc != 0) {
 657                 CERROR("%s: cannot get max LOV EA size: rc = %d\n",
 658                        sbi->ll_dt_exp->exp_obd->obd_name, rc);
 659                 RETURN(rc);
 660         }
 661
 662         size = sizeof(int);
 663         rc = obd_get_info(NULL, sbi->ll_md_exp, sizeof(KEY_MAX_EASIZE),
 664                           KEY_MAX_EASIZE, &size, lmmsize);
 665         if (rc)
 666                 CERROR("Get max mdsize error rc %d\n", rc);
 667
 668         RETURN(rc);
 669 }
 670
 671 /**
 672  * Get the value of the default_easize parameter.
 673  *
 674  * \see client_obd::cl_default_mds_easize
 675  *
 676  * \param[in] sbi       superblock info for this filesystem
 677  * \param[out] lmmsize  pointer to storage location for value
 678  *
 679  * \retval 0            on success
 680  * \retval negative     negated errno on failure
 681  */
 682 int ll_get_default_mdsize(struct ll_sb_info *sbi, int *lmmsize)
 683 {
 684         int size, rc;
 685
 686         size = sizeof(int);
 687         rc = obd_get_info(NULL, sbi->ll_md_exp, sizeof(KEY_DEFAULT_EASIZE),
 688                          KEY_DEFAULT_EASIZE, &size, lmmsize);
 689         if (rc)
 690                 CERROR("Get default mdsize error rc %d\n", rc);
 691
 692         RETURN(rc);
 693 }
 694
 695 /**
 696  * Set the default_easize parameter to the given value.
 697  *
 698  * \see client_obd::cl_default_mds_easize
 699  *
 700  * \param[in] sbi       superblock info for this filesystem
 701  * \param[in] lmmsize   the size to set
 702  *
 703  * \retval 0            on success
 704  * \retval negative     negated errno on failure
 705  */
 706 int ll_set_default_mdsize(struct ll_sb_info *sbi, int lmmsize)
 707 {
 708         int rc;
 709
 710         if (lmmsize < sizeof(struct lov_mds_md) ||
 711             lmmsize > OBD_MAX_DEFAULT_EA_SIZE)
 712                 return -EINVAL;
 713
 714         rc = obd_set_info_async(NULL, sbi->ll_md_exp,
 715                                 sizeof(KEY_DEFAULT_EASIZE), KEY_DEFAULT_EASIZE,
 716                                 sizeof(int), &lmmsize, NULL);
 717
 718         RETURN(rc);
 719 }
 720
 721 static void client_common_put_super(struct super_block *sb)
 722 {
 723         struct ll_sb_info *sbi = ll_s2sbi(sb);
 724         ENTRY;
 725
 726         cl_sb_fini(sb);
 727
 728         obd_fid_fini(sbi->ll_dt_exp->exp_obd);
 729         obd_disconnect(sbi->ll_dt_exp);
 730         sbi->ll_dt_exp = NULL;
 731
 732         lprocfs_ll_unregister_mountpoint(sbi);
 733
 734         obd_fid_fini(sbi->ll_md_exp->exp_obd);
 735         obd_disconnect(sbi->ll_md_exp);
 736         sbi->ll_md_exp = NULL;
 737
 738         EXIT;
 739 }
 740
 741 void ll_kill_super(struct super_block *sb)
 742 {
 743         struct ll_sb_info *sbi;
 744         ENTRY;
 745
 746         /* not init sb ?*/
 747         if (!(sb->s_flags & MS_ACTIVE))
 748                 return;
 749
 750         sbi = ll_s2sbi(sb);
 751         /* we need restore s_dev from changed for clustred NFS before put_super
 752          * because new kernels have cached s_dev and change sb->s_dev in
 753          * put_super not affected real removing devices */
 754         if (sbi) {
 755                 sb->s_dev = sbi->ll_sdev_orig;
 756                 sbi->ll_umounting = 1;
 757
 758                 /* wait running statahead threads to quit */
 759                 while (atomic_read(&sbi->ll_sa_running) > 0) {
 760                         set_current_state(TASK_UNINTERRUPTIBLE);
 761                         schedule_timeout(msecs_to_jiffies(MSEC_PER_SEC >> 3));
 762                 }
 763         }
 764
 765         EXIT;
 766 }
 767
 768 static inline int ll_set_opt(const char *opt, char *data, int fl)
 769 {
 770         if (strncmp(opt, data, strlen(opt)) != 0)
 771                 return(0);
 772         else
 773                 return(fl);
 774 }
 775
 776 /* non-client-specific mount options are parsed in lmd_parse */
 777 static int ll_options(char *options, int *flags)
 778 {
 779         int tmp;
 780         char *s1 = options, *s2;
 781         ENTRY;
 782
 783         if (!options)
 784                 RETURN(0);
 785
 786         CDEBUG(D_CONFIG, "Parsing opts %s\n", options);
 787
 788         while (*s1) {
 789                 CDEBUG(D_SUPER, "next opt=%s\n", s1);
 790                 tmp = ll_set_opt("nolock", s1, LL_SBI_NOLCK);
 791                 if (tmp) {
 792                         *flags |= tmp;
 793                         goto next;
 794                 }
 795                 tmp = ll_set_opt("flock", s1, LL_SBI_FLOCK);
 796                 if (tmp) {
 797                         *flags |= tmp;
 798                         goto next;
 799                 }
 800                 tmp = ll_set_opt("localflock", s1, LL_SBI_LOCALFLOCK);
 801                 if (tmp) {
 802                         *flags |= tmp;
 803                         goto next;
 804                 }
 805                 tmp = ll_set_opt("noflock", s1, LL_SBI_FLOCK|LL_SBI_LOCALFLOCK);
 806                 if (tmp) {
 807                         *flags &= ~tmp;
 808                         goto next;
 809                 }
 810                 tmp = ll_set_opt("user_xattr", s1, LL_SBI_USER_XATTR);
 811                 if (tmp) {
 812                         *flags |= tmp;
 813                         goto next;
 814                 }
 815                 tmp = ll_set_opt("nouser_xattr", s1, LL_SBI_USER_XATTR);
 816                 if (tmp) {
 817                         *flags &= ~tmp;
 818                         goto next;
 819                 }
 820                 tmp = ll_set_opt("context", s1, 1);
 821                 if (tmp)
 822                         goto next;
 823                 tmp = ll_set_opt("fscontext", s1, 1);
 824                 if (tmp)
 825                         goto next;
 826                 tmp = ll_set_opt("defcontext", s1, 1);
 827                 if (tmp)
 828                         goto next;
 829                 tmp = ll_set_opt("rootcontext", s1, 1);
 830                 if (tmp)
 831                         goto next;
 832                 tmp = ll_set_opt("user_fid2path", s1, LL_SBI_USER_FID2PATH);
 833                 if (tmp) {
 834                         *flags |= tmp;
 835                         goto next;
 836                 }
 837                 tmp = ll_set_opt("nouser_fid2path", s1, LL_SBI_USER_FID2PATH);
 838                 if (tmp) {
 839                         *flags &= ~tmp;
 840                         goto next;
 841                 }
 842
 843                 tmp = ll_set_opt("checksum", s1, LL_SBI_CHECKSUM);
 844                 if (tmp) {
 845                         *flags |= tmp;
 846                         goto next;
 847                 }
 848                 tmp = ll_set_opt("nochecksum", s1, LL_SBI_CHECKSUM);
 849                 if (tmp) {
 850                         *flags &= ~tmp;
 851                         goto next;
 852                 }
 853                 tmp = ll_set_opt("lruresize", s1, LL_SBI_LRU_RESIZE);
 854                 if (tmp) {
 855                         *flags |= tmp;
 856                         goto next;
 857                 }
 858                 tmp = ll_set_opt("nolruresize", s1, LL_SBI_LRU_RESIZE);
 859                 if (tmp) {
 860                         *flags &= ~tmp;
 861                         goto next;
 862                 }
 863                 tmp = ll_set_opt("lazystatfs", s1, LL_SBI_LAZYSTATFS);
 864                 if (tmp) {
 865                         *flags |= tmp;
 866                         goto next;
 867                 }
 868                 tmp = ll_set_opt("nolazystatfs", s1, LL_SBI_LAZYSTATFS);
 869                 if (tmp) {
 870                         *flags &= ~tmp;
 871                         goto next;
 872                 }
 873                 tmp = ll_set_opt("32bitapi", s1, LL_SBI_32BIT_API);
 874                 if (tmp) {
 875                         *flags |= tmp;
 876                         goto next;
 877                 }
 878                 tmp = ll_set_opt("verbose", s1, LL_SBI_VERBOSE);
 879                 if (tmp) {
 880                         *flags |= tmp;
 881                         goto next;
 882                 }
 883                 tmp = ll_set_opt("noverbose", s1, LL_SBI_VERBOSE);
 884                 if (tmp) {
 885                         *flags &= ~tmp;
 886                         goto next;
 887                 }
 888                 tmp = ll_set_opt("always_ping", s1, LL_SBI_ALWAYS_PING);
 889                 if (tmp) {
 890                         *flags |= tmp;
 891                         goto next;
 892                 }
 893                 LCONSOLE_ERROR_MSG(0x152, "Unknown option '%s', won't mount.\n",
 894                                    s1);
 895                 RETURN(-EINVAL);
 896
 897 next:
 898                 /* Find next opt */
 899                 s2 = strchr(s1, ',');
 900                 if (s2 == NULL)
 901                         break;
 902                 s1 = s2 + 1;
 903         }
 904         RETURN(0);
 905 }
 906
 907 void ll_lli_init(struct ll_inode_info *lli)
 908 {
 909         lli->lli_inode_magic = LLI_INODE_MAGIC;
 910         lli->lli_flags = 0;
 911         spin_lock_init(&lli->lli_lock);
 912         lli->lli_posix_acl = NULL;
 913         /* Do not set lli_fid, it has been initialized already. */
 914         fid_zero(&lli->lli_pfid);
 915         lli->lli_mds_read_och = NULL;
 916         lli->lli_mds_write_och = NULL;
 917         lli->lli_mds_exec_och = NULL;
 918         lli->lli_open_fd_read_count = 0;
 919         lli->lli_open_fd_write_count = 0;
 920         lli->lli_open_fd_exec_count = 0;
 921         mutex_init(&lli->lli_och_mutex);
 922         spin_lock_init(&lli->lli_agl_lock);
 923         spin_lock_init(&lli->lli_layout_lock);
 924         ll_layout_version_set(lli, CL_LAYOUT_GEN_NONE);
 925         lli->lli_clob = NULL;
 926
 927         init_rwsem(&lli->lli_xattrs_list_rwsem);
 928         mutex_init(&lli->lli_xattrs_enq_lock);
 929
 930         LASSERT(lli->lli_vfs_inode.i_mode != 0);
 931         if (S_ISDIR(lli->lli_vfs_inode.i_mode)) {
 932                 mutex_init(&lli->lli_readdir_mutex);
 933                 lli->lli_opendir_key = NULL;
 934                 lli->lli_sai = NULL;
 935                 spin_lock_init(&lli->lli_sa_lock);
 936                 lli->lli_opendir_pid = 0;
 937                 lli->lli_sa_enabled = 0;
 938                 lli->lli_def_stripe_offset = -1;
 939         } else {
 940                 mutex_init(&lli->lli_size_mutex);
 941                 lli->lli_symlink_name = NULL;
 942                 init_rwsem(&lli->lli_trunc_sem);
 943                 range_lock_tree_init(&lli->lli_write_tree);
 944                 init_rwsem(&lli->lli_glimpse_sem);
 945                 lli->lli_glimpse_time = 0;
 946                 INIT_LIST_HEAD(&lli->lli_agl_list);
 947                 lli->lli_agl_index = 0;
 948                 lli->lli_async_rc = 0;
 949         }
 950         mutex_init(&lli->lli_layout_mutex);
 951         memset(lli->lli_jobid, 0, LUSTRE_JOBID_SIZE);
 952 }
 953
 954 static inline int ll_bdi_register(struct backing_dev_info *bdi)
 955 {
 956         static atomic_t ll_bdi_num = ATOMIC_INIT(0);
 957
 958         bdi->name = "lustre";
 959         return bdi_register(bdi, NULL, "lustre-%d",
 960                             atomic_inc_return(&ll_bdi_num));
 961 }
 962
 963 int ll_fill_super(struct super_block *sb, struct vfsmount *mnt)
 964 {
 965         struct  lustre_profile *lprof = NULL;
 966         struct  lustre_sb_info *lsi = s2lsi(sb);
 967         struct  ll_sb_info *sbi;
 968         char    *dt = NULL, *md = NULL;
 969         char    *profilenm = get_profile_name(sb);
 970         struct config_llog_instance *cfg;
 971         /* %p for void* in printf needs 16+2 characters: 0xffffffffffffffff */
 972         const int instlen = sizeof(cfg->cfg_instance) * 2 + 2;
 973         int     md_len = 0;
 974         int     dt_len = 0;
 975         int     err;
 976         ENTRY;
 977
 978         CDEBUG(D_VFSTRACE, "VFS Op: sb %p\n", sb);
 979
 980         OBD_ALLOC_PTR(cfg);
 981         if (cfg == NULL)
 982                 RETURN(-ENOMEM);
 983
 984         try_module_get(THIS_MODULE);
 985
 986         /* client additional sb info */
 987         lsi->lsi_llsbi = sbi = ll_init_sbi();
 988         if (!sbi) {
 989                 module_put(THIS_MODULE);
 990                 OBD_FREE_PTR(cfg);
 991                 RETURN(-ENOMEM);
 992         }
 993
 994         err = ll_options(lsi->lsi_lmd->lmd_opts, &sbi->ll_flags);
 995         if (err)
 996                 GOTO(out_free, err);
 997
 998         err = bdi_init(&lsi->lsi_bdi);
 999         if (err)
1000                 GOTO(out_free, err);
1001         lsi->lsi_flags |= LSI_BDI_INITIALIZED;
1002 #ifdef HAVE_BDI_CAP_MAP_COPY
1003         lsi->lsi_bdi.capabilities = BDI_CAP_MAP_COPY;
1004 #else
1005         lsi->lsi_bdi.capabilities = 0;
1006 #endif
1007         err = ll_bdi_register(&lsi->lsi_bdi);
1008         if (err)
1009                 GOTO(out_free, err);
1010
1011         sb->s_bdi = &lsi->lsi_bdi;
1012 #ifndef HAVE_DCACHE_LOCK
1013         /* kernel >= 2.6.38 store dentry operations in sb->s_d_op. */
1014         sb->s_d_op = &ll_d_ops;
1015 #endif
1016
1017         /* Call lprocfs_ll_register_mountpoint() before lustre_process_log()
1018          * so that "llite.*.*" params can be processed correctly. */
1019         if (proc_lustre_fs_root != NULL) {
1020                 err = lprocfs_ll_register_mountpoint(proc_lustre_fs_root, sb);
1021                 if (err < 0) {
1022                         CERROR("%s: could not register mountpoint in llite: "
1023                                "rc = %d\n", ll_get_fsname(sb, NULL, 0), err);
1024                         err = 0;
1025                 }
1026         }
1027
1028         /* Generate a string unique to this super, in case some joker tries
1029            to mount the same fs at two mount points.
1030            Use the address of the super itself.*/
1031         cfg->cfg_instance = sb;
1032         cfg->cfg_uuid = lsi->lsi_llsbi->ll_sb_uuid;
1033         cfg->cfg_callback = class_config_llog_handler;
1034         cfg->cfg_sub_clds = CONFIG_SUB_CLIENT;
1035         /* set up client obds */
1036         err = lustre_process_log(sb, profilenm, cfg);
1037         if (err < 0)
1038                 GOTO(out_proc, err);
1039
1040         /* Profile set with LCFG_MOUNTOPT so we can find our mdc and osc obds */
1041         lprof = class_get_profile(profilenm);
1042         if (lprof == NULL) {
1043                 LCONSOLE_ERROR_MSG(0x156, "The client profile '%s' could not be"
1044                                    " read from the MGS.  Does that filesystem "
1045                                    "exist?\n", profilenm);
1046                 GOTO(out_proc, err = -EINVAL);
1047         }
1048         CDEBUG(D_CONFIG, "Found profile %s: mdc=%s osc=%s\n", profilenm,
1049                lprof->lp_md, lprof->lp_dt);
1050
1051         dt_len = strlen(lprof->lp_dt) + instlen + 2;
1052         OBD_ALLOC(dt, dt_len);
1053         if (!dt)
1054                 GOTO(out_proc, err = -ENOMEM);
1055         snprintf(dt, dt_len - 1, "%s-%p", lprof->lp_dt, cfg->cfg_instance);
1056
1057         md_len = strlen(lprof->lp_md) + instlen + 2;
1058         OBD_ALLOC(md, md_len);
1059         if (!md)
1060                 GOTO(out_proc, err = -ENOMEM);
1061         snprintf(md, md_len - 1, "%s-%p", lprof->lp_md, cfg->cfg_instance);
1062
1063         /* connections, registrations, sb setup */
1064         err = client_common_fill_super(sb, md, dt, mnt);
1065         if (err < 0)
1066                 GOTO(out_proc, err);
1067
1068         sbi->ll_client_common_fill_super_succeeded = 1;
1069
1070 out_proc:
1071         if (err < 0)
1072                 lprocfs_ll_unregister_mountpoint(sbi);
1073 out_free:
1074         if (md)
1075                 OBD_FREE(md, md_len);
1076         if (dt)
1077                 OBD_FREE(dt, dt_len);
1078         if (lprof != NULL)
1079                 class_put_profile(lprof);
1080         if (err)
1081                 ll_put_super(sb);
1082         else if (sbi->ll_flags & LL_SBI_VERBOSE)
1083                 LCONSOLE_WARN("Mounted %s\n", profilenm);
1084
1085         OBD_FREE_PTR(cfg);
1086         RETURN(err);
1087 } /* ll_fill_super */
1088
1089 void ll_put_super(struct super_block *sb)
1090 {
1091         struct config_llog_instance cfg, params_cfg;
1092         struct obd_device *obd;
1093         struct lustre_sb_info *lsi = s2lsi(sb);
1094         struct ll_sb_info *sbi = ll_s2sbi(sb);
1095         char *profilenm = get_profile_name(sb);
1096         long ccc_count;
1097         int next, force = 1, rc = 0;
1098         ENTRY;
1099
1100         CDEBUG(D_VFSTRACE, "VFS Op: sb %p - %s\n", sb, profilenm);
1101
1102         cfg.cfg_instance = sb;
1103         lustre_end_log(sb, profilenm, &cfg);
1104
1105         params_cfg.cfg_instance = sb;
1106         lustre_end_log(sb, PARAMS_FILENAME, &params_cfg);
1107
1108         if (sbi->ll_md_exp) {
1109                 obd = class_exp2obd(sbi->ll_md_exp);
1110                 if (obd)
1111                         force = obd->obd_force;
1112         }
1113
1114         /* Wait for unstable pages to be committed to stable storage */
1115         if (force == 0) {
1116                 struct l_wait_info lwi = LWI_INTR(LWI_ON_SIGNAL_NOOP, NULL);
1117                 rc = l_wait_event(sbi->ll_cache->ccc_unstable_waitq,
1118                         atomic_long_read(&sbi->ll_cache->ccc_unstable_nr) == 0,
1119                         &lwi);
1120         }
1121
1122         ccc_count = atomic_long_read(&sbi->ll_cache->ccc_unstable_nr);
1123         if (force == 0 && rc != -EINTR)
1124                 LASSERTF(ccc_count == 0, "count: %li\n", ccc_count);
1125
1126
1127         /* We need to set force before the lov_disconnect in
1128            lustre_common_put_super, since l_d cleans up osc's as well. */
1129         if (force) {
1130                 next = 0;
1131                 while ((obd = class_devices_in_group(&sbi->ll_sb_uuid,
1132                                                      &next)) != NULL) {
1133                         obd->obd_force = force;
1134                 }
1135         }
1136
1137         if (sbi->ll_client_common_fill_super_succeeded) {
1138                 /* Only if client_common_fill_super succeeded */
1139                 client_common_put_super(sb);
1140         }
1141
1142         next = 0;
1143         while ((obd = class_devices_in_group(&sbi->ll_sb_uuid, &next)) !=NULL) {
1144                 class_manual_cleanup(obd);
1145         }
1146
1147         if (sbi->ll_flags & LL_SBI_VERBOSE)
1148                 LCONSOLE_WARN("Unmounted %s\n", profilenm ? profilenm : "");
1149
1150         if (profilenm)
1151                 class_del_profile(profilenm);
1152
1153         if (lsi->lsi_flags & LSI_BDI_INITIALIZED) {
1154                 bdi_destroy(&lsi->lsi_bdi);
1155                 lsi->lsi_flags &= ~LSI_BDI_INITIALIZED;
1156         }
1157
1158         ll_free_sbi(sb);
1159         lsi->lsi_llsbi = NULL;
1160
1161         lustre_common_put_super(sb);
1162
1163         cl_env_cache_purge(~0);
1164
1165         module_put(THIS_MODULE);
1166
1167         EXIT;
1168 } /* client_put_super */
1169
1170 struct inode *ll_inode_from_resource_lock(struct ldlm_lock *lock)
1171 {
1172         struct inode *inode = NULL;
1173
1174         /* NOTE: we depend on atomic igrab() -bzzz */
1175         lock_res_and_lock(lock);
1176         if (lock->l_resource->lr_lvb_inode) {
1177                 struct ll_inode_info * lli;
1178                 lli = ll_i2info(lock->l_resource->lr_lvb_inode);
1179                 if (lli->lli_inode_magic == LLI_INODE_MAGIC) {
1180                         inode = igrab(lock->l_resource->lr_lvb_inode);
1181                 } else {
1182                         inode = lock->l_resource->lr_lvb_inode;
1183                         LDLM_DEBUG_LIMIT(inode->i_state & I_FREEING ?  D_INFO :
1184                                          D_WARNING, lock, "lr_lvb_inode %p is "
1185                                          "bogus: magic %08x",
1186                                          lock->l_resource->lr_lvb_inode,
1187                                          lli->lli_inode_magic);
1188                         inode = NULL;
1189                 }
1190         }
1191         unlock_res_and_lock(lock);
1192         return inode;
1193 }
1194
1195 void ll_dir_clear_lsm_md(struct inode *inode)
1196 {
1197         struct ll_inode_info *lli = ll_i2info(inode);
1198
1199         LASSERT(S_ISDIR(inode->i_mode));
1200
1201         if (lli->lli_lsm_md != NULL) {
1202                 lmv_free_memmd(lli->lli_lsm_md);
1203                 lli->lli_lsm_md = NULL;
1204         }
1205 }
1206
1207 static struct inode *ll_iget_anon_dir(struct super_block *sb,
1208                                       const struct lu_fid *fid,
1209                                       struct lustre_md *md)
1210 {
1211         struct ll_sb_info       *sbi = ll_s2sbi(sb);
1212         struct mdt_body         *body = md->body;
1213         struct inode            *inode;
1214         ino_t                   ino;
1215         ENTRY;
1216
1217         ino = cl_fid_build_ino(fid, sbi->ll_flags & LL_SBI_32BIT_API);
1218         inode = iget_locked(sb, ino);
1219         if (inode == NULL) {
1220                 CERROR("%s: failed get simple inode "DFID": rc = -ENOENT\n",
1221                        ll_get_fsname(sb, NULL, 0), PFID(fid));
1222                 RETURN(ERR_PTR(-ENOENT));
1223         }
1224
1225         if (inode->i_state & I_NEW) {
1226                 struct ll_inode_info *lli = ll_i2info(inode);
1227                 struct lmv_stripe_md *lsm = md->lmv;
1228
1229                 inode->i_mode = (inode->i_mode & ~S_IFMT) |
1230                                 (body->mbo_mode & S_IFMT);
1231                 LASSERTF(S_ISDIR(inode->i_mode), "Not slave inode "DFID"\n",
1232                          PFID(fid));
1233
1234                 LTIME_S(inode->i_mtime) = 0;
1235                 LTIME_S(inode->i_atime) = 0;
1236                 LTIME_S(inode->i_ctime) = 0;
1237                 inode->i_rdev = 0;
1238
1239 #ifdef HAVE_BACKING_DEV_INFO
1240                 /* initializing backing dev info. */
1241                 inode->i_mapping->backing_dev_info =
1242                                                 &s2lsi(inode->i_sb)->lsi_bdi;
1243 #endif
1244                 inode->i_op = &ll_dir_inode_operations;
1245                 inode->i_fop = &ll_dir_operations;
1246                 lli->lli_fid = *fid;
1247                 ll_lli_init(lli);
1248
1249                 LASSERT(lsm != NULL);
1250                 /* master object FID */
1251                 lli->lli_pfid = body->mbo_fid1;
1252                 CDEBUG(D_INODE, "lli %p slave "DFID" master "DFID"\n",
1253                        lli, PFID(fid), PFID(&lli->lli_pfid));
1254                 unlock_new_inode(inode);
1255         }
1256
1257         RETURN(inode);
1258 }
1259
1260 static int ll_init_lsm_md(struct inode *inode, struct lustre_md *md)
1261 {
1262         struct lu_fid *fid;
1263         struct lmv_stripe_md *lsm = md->lmv;
1264         int i;
1265
1266         LASSERT(lsm != NULL);
1267         /* XXX sigh, this lsm_root initialization should be in
1268          * LMV layer, but it needs ll_iget right now, so we
1269          * put this here right now. */
1270         for (i = 0; i < lsm->lsm_md_stripe_count; i++) {
1271                 fid = &lsm->lsm_md_oinfo[i].lmo_fid;
1272                 LASSERT(lsm->lsm_md_oinfo[i].lmo_root == NULL);
1273                 /* Unfortunately ll_iget will call ll_update_inode,
1274                  * where the initialization of slave inode is slightly
1275                  * different, so it reset lsm_md to NULL to avoid
1276                  * initializing lsm for slave inode. */
1277                 /* For migrating inode, master stripe and master object will
1278                  * be same, so we only need assign this inode */
1279                 if (lsm->lsm_md_hash_type & LMV_HASH_FLAG_MIGRATION && i == 0)
1280                         lsm->lsm_md_oinfo[i].lmo_root = inode;
1281                 else
1282                         lsm->lsm_md_oinfo[i].lmo_root =
1283                                 ll_iget_anon_dir(inode->i_sb, fid, md);
1284
1285                 if (IS_ERR(lsm->lsm_md_oinfo[i].lmo_root)) {
1286                         int rc = PTR_ERR(lsm->lsm_md_oinfo[i].lmo_root);
1287
1288                         lsm->lsm_md_oinfo[i].lmo_root = NULL;
1289                         return rc;
1290                 }
1291         }
1292
1293         return 0;
1294 }
1295
1296 static inline int lli_lsm_md_eq(const struct lmv_stripe_md *lsm_md1,
1297                                 const struct lmv_stripe_md *lsm_md2)
1298 {
1299         return lsm_md1->lsm_md_magic == lsm_md2->lsm_md_magic &&
1300                lsm_md1->lsm_md_stripe_count == lsm_md2->lsm_md_stripe_count &&
1301                lsm_md1->lsm_md_master_mdt_index ==
1302                                         lsm_md2->lsm_md_master_mdt_index &&
1303                lsm_md1->lsm_md_hash_type == lsm_md2->lsm_md_hash_type &&
1304                lsm_md1->lsm_md_layout_version ==
1305                                         lsm_md2->lsm_md_layout_version &&
1306                strcmp(lsm_md1->lsm_md_pool_name,
1307                       lsm_md2->lsm_md_pool_name) == 0;
1308 }
1309
1310 static int ll_update_lsm_md(struct inode *inode, struct lustre_md *md)
1311 {
1312         struct ll_inode_info *lli = ll_i2info(inode);
1313         struct lmv_stripe_md *lsm = md->lmv;
1314         int     rc;
1315         ENTRY;
1316
1317         LASSERT(S_ISDIR(inode->i_mode));
1318         CDEBUG(D_INODE, "update lsm %p of "DFID"\n", lli->lli_lsm_md,
1319                PFID(ll_inode2fid(inode)));
1320
1321         /* no striped information from request. */
1322         if (lsm == NULL) {
1323                 if (lli->lli_lsm_md == NULL) {
1324                         RETURN(0);
1325                 } else if (lli->lli_lsm_md->lsm_md_hash_type &
1326                                                 LMV_HASH_FLAG_MIGRATION) {
1327                         /* migration is done, the temporay MIGRATE layout has
1328                          * been removed */
1329                         CDEBUG(D_INODE, DFID" finish migration.\n",
1330                                PFID(ll_inode2fid(inode)));
1331                         lmv_free_memmd(lli->lli_lsm_md);
1332                         lli->lli_lsm_md = NULL;
1333                         RETURN(0);
1334                 } else {
1335                         /* The lustre_md from req does not include stripeEA,
1336                          * see ll_md_setattr */
1337                         RETURN(0);
1338                 }
1339         }
1340
1341         /* set the directory layout */
1342         if (lli->lli_lsm_md == NULL) {
1343                 struct cl_attr  *attr;
1344
1345                 rc = ll_init_lsm_md(inode, md);
1346                 if (rc != 0)
1347                         RETURN(rc);
1348
1349                 /* set md->lmv to NULL, so the following free lustre_md
1350                  * will not free this lsm */
1351                 md->lmv = NULL;
1352                 lli->lli_lsm_md = lsm;
1353
1354                 OBD_ALLOC_PTR(attr);
1355                 if (attr == NULL)
1356                         RETURN(-ENOMEM);
1357
1358                 /* validate the lsm */
1359                 rc = md_merge_attr(ll_i2mdexp(inode), lsm, attr,
1360                                    ll_md_blocking_ast);
1361                 if (rc != 0) {
1362                         OBD_FREE_PTR(attr);
1363                         RETURN(rc);
1364                 }
1365
1366                 if (md->body->mbo_valid & OBD_MD_FLNLINK)
1367                         md->body->mbo_nlink = attr->cat_nlink;
1368                 if (md->body->mbo_valid & OBD_MD_FLSIZE)
1369                         md->body->mbo_size = attr->cat_size;
1370                 if (md->body->mbo_valid & OBD_MD_FLATIME)
1371                         md->body->mbo_atime = attr->cat_atime;
1372                 if (md->body->mbo_valid & OBD_MD_FLCTIME)
1373                         md->body->mbo_ctime = attr->cat_ctime;
1374                 if (md->body->mbo_valid & OBD_MD_FLMTIME)
1375                         md->body->mbo_mtime = attr->cat_mtime;
1376
1377                 OBD_FREE_PTR(attr);
1378
1379                 CDEBUG(D_INODE, "Set lsm %p magic %x to "DFID"\n", lsm,
1380                        lsm->lsm_md_magic, PFID(ll_inode2fid(inode)));
1381                 RETURN(0);
1382         }
1383
1384         /* Compare the old and new stripe information */
1385         if (!lsm_md_eq(lli->lli_lsm_md, lsm)) {
1386                 struct lmv_stripe_md    *old_lsm = lli->lli_lsm_md;
1387                 int                     idx;
1388
1389                 CERROR("%s: inode "DFID"(%p)'s lmv layout mismatch (%p)/(%p)"
1390                        "magic:0x%x/0x%x stripe count: %d/%d master_mdt: %d/%d"
1391                        "hash_type:0x%x/0x%x layout: 0x%x/0x%x pool:%s/%s\n",
1392                        ll_get_fsname(inode->i_sb, NULL, 0), PFID(&lli->lli_fid),
1393                        inode, lsm, old_lsm,
1394                        lsm->lsm_md_magic, old_lsm->lsm_md_magic,
1395                        lsm->lsm_md_stripe_count,
1396                        old_lsm->lsm_md_stripe_count,
1397                        lsm->lsm_md_master_mdt_index,
1398                        old_lsm->lsm_md_master_mdt_index,
1399                        lsm->lsm_md_hash_type, old_lsm->lsm_md_hash_type,
1400                        lsm->lsm_md_layout_version,
1401                        old_lsm->lsm_md_layout_version,
1402                        lsm->lsm_md_pool_name,
1403                        old_lsm->lsm_md_pool_name);
1404
1405                 for (idx = 0; idx < old_lsm->lsm_md_stripe_count; idx++) {
1406                         CERROR("%s: sub FIDs in old lsm idx %d, old: "DFID"\n",
1407                                ll_get_fsname(inode->i_sb, NULL, 0), idx,
1408                                PFID(&old_lsm->lsm_md_oinfo[idx].lmo_fid));
1409                 }
1410
1411                 for (idx = 0; idx < lsm->lsm_md_stripe_count; idx++) {
1412                         CERROR("%s: sub FIDs in new lsm idx %d, new: "DFID"\n",
1413                                ll_get_fsname(inode->i_sb, NULL, 0), idx,
1414                                PFID(&lsm->lsm_md_oinfo[idx].lmo_fid));
1415                 }
1416
1417                 RETURN(-EIO);
1418         }
1419
1420         RETURN(0);
1421 }
1422
1423 void ll_clear_inode(struct inode *inode)
1424 {
1425         struct ll_inode_info *lli = ll_i2info(inode);
1426         struct ll_sb_info *sbi = ll_i2sbi(inode);
1427         ENTRY;
1428
1429         CDEBUG(D_VFSTRACE, "VFS Op:inode="DFID"(%p)\n",
1430                PFID(ll_inode2fid(inode)), inode);
1431
1432         if (S_ISDIR(inode->i_mode)) {
1433                 /* these should have been cleared in ll_file_release */
1434                 LASSERT(lli->lli_opendir_key == NULL);
1435                 LASSERT(lli->lli_sai == NULL);
1436                 LASSERT(lli->lli_opendir_pid == 0);
1437         }
1438
1439         md_null_inode(sbi->ll_md_exp, ll_inode2fid(inode));
1440
1441         LASSERT(!lli->lli_open_fd_write_count);
1442         LASSERT(!lli->lli_open_fd_read_count);
1443         LASSERT(!lli->lli_open_fd_exec_count);
1444
1445         if (lli->lli_mds_write_och)
1446                 ll_md_real_close(inode, FMODE_WRITE);
1447         if (lli->lli_mds_exec_och)
1448                 ll_md_real_close(inode, FMODE_EXEC);
1449         if (lli->lli_mds_read_och)
1450                 ll_md_real_close(inode, FMODE_READ);
1451
1452         if (S_ISLNK(inode->i_mode) && lli->lli_symlink_name) {
1453                 OBD_FREE(lli->lli_symlink_name,
1454                          strlen(lli->lli_symlink_name) + 1);
1455                 lli->lli_symlink_name = NULL;
1456         }
1457
1458         ll_xattr_cache_destroy(inode);
1459
1460 #ifdef CONFIG_FS_POSIX_ACL
1461         forget_all_cached_acls(inode);
1462         if (lli->lli_posix_acl) {
1463                 LASSERT(atomic_read(&lli->lli_posix_acl->a_refcount) == 1);
1464                 posix_acl_release(lli->lli_posix_acl);
1465                 lli->lli_posix_acl = NULL;
1466         }
1467 #endif
1468         lli->lli_inode_magic = LLI_INODE_DEAD;
1469
1470         if (S_ISDIR(inode->i_mode))
1471                 ll_dir_clear_lsm_md(inode);
1472         else if (S_ISREG(inode->i_mode) && !is_bad_inode(inode))
1473                 LASSERT(list_empty(&lli->lli_agl_list));
1474
1475         /*
1476          * XXX This has to be done before lsm is freed below, because
1477          * cl_object still uses inode lsm.
1478          */
1479         cl_inode_fini(inode);
1480
1481         EXIT;
1482 }
1483
1484 static int ll_md_setattr(struct dentry *dentry, struct md_op_data *op_data)
1485 {
1486         struct lustre_md md;
1487         struct inode *inode = dentry->d_inode;
1488         struct ll_sb_info *sbi = ll_i2sbi(inode);
1489         struct ptlrpc_request *request = NULL;
1490         int rc, ia_valid;
1491         ENTRY;
1492
1493         op_data = ll_prep_md_op_data(op_data, inode, NULL, NULL, 0, 0,
1494                                      LUSTRE_OPC_ANY, NULL);
1495         if (IS_ERR(op_data))
1496                 RETURN(PTR_ERR(op_data));
1497
1498         rc = md_setattr(sbi->ll_md_exp, op_data, NULL, 0, &request);
1499         if (rc) {
1500                 ptlrpc_req_finished(request);
1501                 if (rc == -ENOENT) {
1502                         clear_nlink(inode);
1503                         /* Unlinked special device node? Or just a race?
1504                          * Pretend we done everything. */
1505                         if (!S_ISREG(inode->i_mode) &&
1506                             !S_ISDIR(inode->i_mode)) {
1507                                 ia_valid = op_data->op_attr.ia_valid;
1508                                 op_data->op_attr.ia_valid &= ~TIMES_SET_FLAGS;
1509                                 rc = simple_setattr(dentry, &op_data->op_attr);
1510                                 op_data->op_attr.ia_valid = ia_valid;
1511                         }
1512                 } else if (rc != -EPERM && rc != -EACCES && rc != -ETXTBSY) {
1513                         CERROR("md_setattr fails: rc = %d\n", rc);
1514                 }
1515                 RETURN(rc);
1516         }
1517
1518         rc = md_get_lustre_md(sbi->ll_md_exp, request, sbi->ll_dt_exp,
1519                               sbi->ll_md_exp, &md);
1520         if (rc) {
1521                 ptlrpc_req_finished(request);
1522                 RETURN(rc);
1523         }
1524
1525         ia_valid = op_data->op_attr.ia_valid;
1526         /* inode size will be in ll_setattr_ost, can't do it now since dirty
1527          * cache is not cleared yet. */
1528         op_data->op_attr.ia_valid &= ~(TIMES_SET_FLAGS | ATTR_SIZE);
1529         if (S_ISREG(inode->i_mode))
1530                 inode_lock(inode);
1531         rc = simple_setattr(dentry, &op_data->op_attr);
1532         if (S_ISREG(inode->i_mode))
1533                 inode_unlock(inode);
1534         op_data->op_attr.ia_valid = ia_valid;
1535
1536         rc = ll_update_inode(inode, &md);
1537         ptlrpc_req_finished(request);
1538
1539         RETURN(rc);
1540 }
1541
1542 /* If this inode has objects allocated to it (lsm != NULL), then the OST
1543  * object(s) determine the file size and mtime.  Otherwise, the MDS will
1544  * keep these values until such a time that objects are allocated for it.
1545  * We do the MDS operations first, as it is checking permissions for us.
1546  * We don't to the MDS RPC if there is nothing that we want to store there,
1547  * otherwise there is no harm in updating mtime/atime on the MDS if we are
1548  * going to do an RPC anyways.
1549  *
1550  * If we are doing a truncate, we will send the mtime and ctime updates
1551  * to the OST with the punch RPC, otherwise we do an explicit setattr RPC.
1552  * I don't believe it is possible to get e.g. ATTR_MTIME_SET and ATTR_SIZE
1553  * at the same time.
1554  *
1555  * In case of HSMimport, we only set attr on MDS.
1556  */
1557 int ll_setattr_raw(struct dentry *dentry, struct iattr *attr, bool hsm_import)
1558 {
1559         struct inode *inode = dentry->d_inode;
1560         struct ll_inode_info *lli = ll_i2info(inode);
1561         struct md_op_data *op_data = NULL;
1562         int rc = 0;
1563         ENTRY;
1564
1565         CDEBUG(D_VFSTRACE, "%s: setattr inode "DFID"(%p) from %llu to %llu, "
1566                "valid %x, hsm_import %d\n",
1567                ll_get_fsname(inode->i_sb, NULL, 0), PFID(&lli->lli_fid),
1568                inode, i_size_read(inode), attr->ia_size, attr->ia_valid,
1569                hsm_import);
1570
1571         if (attr->ia_valid & ATTR_SIZE) {
1572                 /* Check new size against VFS/VM file size limit and rlimit */
1573                 rc = inode_newsize_ok(inode, attr->ia_size);
1574                 if (rc)
1575                         RETURN(rc);
1576
1577                 /* The maximum Lustre file size is variable, based on the
1578                  * OST maximum object size and number of stripes.  This
1579                  * needs another check in addition to the VFS check above. */
1580                 if (attr->ia_size > ll_file_maxbytes(inode)) {
1581                         CDEBUG(D_INODE,"file "DFID" too large %llu > %llu\n",
1582                                PFID(&lli->lli_fid), attr->ia_size,
1583                                ll_file_maxbytes(inode));
1584                         RETURN(-EFBIG);
1585                 }
1586
1587                 attr->ia_valid |= ATTR_MTIME | ATTR_CTIME;
1588         }
1589
1590         /* POSIX: check before ATTR_*TIME_SET set (from inode_change_ok) */
1591         if (attr->ia_valid & TIMES_SET_FLAGS) {
1592                 if ((!uid_eq(current_fsuid(), inode->i_uid)) &&
1593                     !cfs_capable(CFS_CAP_FOWNER))
1594                         RETURN(-EPERM);
1595         }
1596
1597         /* We mark all of the fields "set" so MDS/OST does not re-set them */
1598         if (!(attr->ia_valid & ATTR_CTIME_SET) &&
1599             (attr->ia_valid & ATTR_CTIME)) {
1600                 attr->ia_ctime = current_time(inode);
1601                 attr->ia_valid |= ATTR_CTIME_SET;
1602         }
1603         if (!(attr->ia_valid & ATTR_ATIME_SET) &&
1604             (attr->ia_valid & ATTR_ATIME)) {
1605                 attr->ia_atime = current_time(inode);
1606                 attr->ia_valid |= ATTR_ATIME_SET;
1607         }
1608         if (!(attr->ia_valid & ATTR_MTIME_SET) &&
1609             (attr->ia_valid & ATTR_MTIME)) {
1610                 attr->ia_mtime = current_time(inode);
1611                 attr->ia_valid |= ATTR_MTIME_SET;
1612         }
1613
1614         if (attr->ia_valid & (ATTR_MTIME | ATTR_CTIME))
1615                 CDEBUG(D_INODE, "setting mtime %lu, ctime %lu, now = %llu\n",
1616                        LTIME_S(attr->ia_mtime), LTIME_S(attr->ia_ctime),
1617                        (s64)ktime_get_real_seconds());
1618
1619         if (S_ISREG(inode->i_mode)) {
1620                 if (attr->ia_valid & ATTR_SIZE)
1621                         inode_dio_write_done(inode);
1622                 inode_unlock(inode);
1623         }
1624
1625         /* We always do an MDS RPC, even if we're only changing the size;
1626          * only the MDS knows whether truncate() should fail with -ETXTBUSY */
1627
1628         OBD_ALLOC_PTR(op_data);
1629         if (op_data == NULL)
1630                 GOTO(out, rc = -ENOMEM);
1631
1632         if (!hsm_import && attr->ia_valid & ATTR_SIZE) {
1633                 /* If we are changing file size, file content is
1634                  * modified, flag it. */
1635                 attr->ia_valid |= MDS_OPEN_OWNEROVERRIDE;
1636                 op_data->op_bias |= MDS_DATA_MODIFIED;
1637                 ll_file_clear_flag(lli, LLIF_DATA_MODIFIED);
1638         }
1639
1640         op_data->op_attr = *attr;
1641
1642         rc = ll_md_setattr(dentry, op_data);
1643         if (rc)
1644                 GOTO(out, rc);
1645
1646         if (!S_ISREG(inode->i_mode) || hsm_import)
1647                 GOTO(out, rc = 0);
1648
1649         if (attr->ia_valid & (ATTR_SIZE |
1650                               ATTR_ATIME | ATTR_ATIME_SET |
1651                               ATTR_MTIME | ATTR_MTIME_SET |
1652                               ATTR_CTIME | ATTR_CTIME_SET)) {
1653                 /* For truncate and utimes sending attributes to OSTs, setting
1654                  * mtime/atime to the past will be performed under PW [0:EOF]
1655                  * extent lock (new_size:EOF for truncate).  It may seem
1656                  * excessive to send mtime/atime updates to OSTs when not
1657                  * setting times to past, but it is necessary due to possible
1658                  * time de-synchronization between MDT inode and OST objects */
1659                 rc = cl_setattr_ost(lli->lli_clob, attr, 0);
1660         }
1661
1662         /* If the file was restored, it needs to set dirty flag.
1663          *
1664          * We've already sent MDS_DATA_MODIFIED flag in
1665          * ll_md_setattr() for truncate. However, the MDT refuses to
1666          * set the HS_DIRTY flag on released files, so we have to set
1667          * it again if the file has been restored. Please check how
1668          * LLIF_DATA_MODIFIED is set in vvp_io_setattr_fini().
1669          *
1670          * Please notice that if the file is not released, the previous
1671          * MDS_DATA_MODIFIED has taken effect and usually
1672          * LLIF_DATA_MODIFIED is not set(see vvp_io_setattr_fini()).
1673          * This way we can save an RPC for common open + trunc
1674          * operation. */
1675         if (ll_file_test_and_clear_flag(lli, LLIF_DATA_MODIFIED)) {
1676                 struct hsm_state_set hss = {
1677                         .hss_valid = HSS_SETMASK,
1678                         .hss_setmask = HS_DIRTY,
1679                 };
1680                 int rc2;
1681
1682                 rc2 = ll_hsm_state_set(inode, &hss);
1683                 /* truncate and write can happen at the same time, so that
1684                  * the file can be set modified even though the file is not
1685                  * restored from released state, and ll_hsm_state_set() is
1686                  * not applicable for the file, and rc2 < 0 is normal in this
1687                  * case. */
1688                 if (rc2 < 0)
1689                         CDEBUG(D_INFO, DFID "HSM set dirty failed: rc2 = %d\n",
1690                                PFID(ll_inode2fid(inode)), rc2);
1691         }
1692
1693         EXIT;
1694 out:
1695         if (op_data != NULL)
1696                 ll_finish_md_op_data(op_data);
1697
1698         if (S_ISREG(inode->i_mode)) {
1699                 inode_lock(inode);
1700                 if ((attr->ia_valid & ATTR_SIZE) && !hsm_import)
1701                         inode_dio_wait(inode);
1702                 /* Once we've got the i_mutex, it's safe to set the S_NOSEC
1703                  * flag.  ll_update_inode (called from ll_md_setattr), clears
1704                  * inode flags, so there is a gap where S_NOSEC is not set.
1705                  * This can cause a writer to take the i_mutex unnecessarily,
1706                  * but this is safe to do and should be rare. */
1707                 inode_has_no_xattr(inode);
1708         }
1709
1710         ll_stats_ops_tally(ll_i2sbi(inode), (attr->ia_valid & ATTR_SIZE) ?
1711                         LPROC_LL_TRUNC : LPROC_LL_SETATTR, 1);
1712
1713         return rc;
1714 }
1715
1716 int ll_setattr(struct dentry *de, struct iattr *attr)
1717 {
1718         int mode = de->d_inode->i_mode;
1719
1720         if ((attr->ia_valid & (ATTR_CTIME|ATTR_SIZE|ATTR_MODE)) ==
1721                               (ATTR_CTIME|ATTR_SIZE|ATTR_MODE))
1722                 attr->ia_valid |= MDS_OPEN_OWNEROVERRIDE;
1723
1724         if (((attr->ia_valid & (ATTR_MODE|ATTR_FORCE|ATTR_SIZE)) ==
1725                                (ATTR_SIZE|ATTR_MODE)) &&
1726             (((mode & S_ISUID) && !(attr->ia_mode & S_ISUID)) ||
1727              (((mode & (S_ISGID|S_IXGRP)) == (S_ISGID|S_IXGRP)) &&
1728               !(attr->ia_mode & S_ISGID))))
1729                 attr->ia_valid |= ATTR_FORCE;
1730
1731         if ((attr->ia_valid & ATTR_MODE) &&
1732             (mode & S_ISUID) &&
1733             !(attr->ia_mode & S_ISUID) &&
1734             !(attr->ia_valid & ATTR_KILL_SUID))
1735                 attr->ia_valid |= ATTR_KILL_SUID;
1736
1737         if ((attr->ia_valid & ATTR_MODE) &&
1738             ((mode & (S_ISGID|S_IXGRP)) == (S_ISGID|S_IXGRP)) &&
1739             !(attr->ia_mode & S_ISGID) &&
1740             !(attr->ia_valid & ATTR_KILL_SGID))
1741                 attr->ia_valid |= ATTR_KILL_SGID;
1742
1743         /* avoid polluted from ATTR_TIMES_SET,
1744          * projid is not expected to be set here */
1745         attr->ia_valid &= ~MDS_ATTR_PROJID;
1746
1747         return ll_setattr_raw(de, attr, false);
1748 }
1749
1750 int ll_statfs_internal(struct super_block *sb, struct obd_statfs *osfs,
1751                        __u64 max_age, __u32 flags)
1752 {
1753         struct ll_sb_info *sbi = ll_s2sbi(sb);
1754         struct obd_statfs obd_osfs;
1755         int rc;
1756         ENTRY;
1757
1758         rc = obd_statfs(NULL, sbi->ll_md_exp, osfs, max_age, flags);
1759         if (rc) {
1760                 CERROR("md_statfs fails: rc = %d\n", rc);
1761                 RETURN(rc);
1762         }
1763
1764         osfs->os_type = sb->s_magic;
1765
1766         CDEBUG(D_SUPER, "MDC blocks %llu/%llu objects %llu/%llu\n",
1767                osfs->os_bavail, osfs->os_blocks, osfs->os_ffree,osfs->os_files);
1768
1769         if (sbi->ll_flags & LL_SBI_LAZYSTATFS)
1770                 flags |= OBD_STATFS_NODELAY;
1771
1772         rc = obd_statfs_rqset(sbi->ll_dt_exp, &obd_osfs, max_age, flags);
1773         if (rc) {
1774                 CERROR("obd_statfs fails: rc = %d\n", rc);
1775                 RETURN(rc);
1776         }
1777
1778         CDEBUG(D_SUPER, "OSC blocks %llu/%llu objects %llu/%llu\n",
1779                obd_osfs.os_bavail, obd_osfs.os_blocks, obd_osfs.os_ffree,
1780                obd_osfs.os_files);
1781
1782         osfs->os_bsize = obd_osfs.os_bsize;
1783         osfs->os_blocks = obd_osfs.os_blocks;
1784         osfs->os_bfree = obd_osfs.os_bfree;
1785         osfs->os_bavail = obd_osfs.os_bavail;
1786
1787         /* If we don't have as many objects free on the OST as inodes
1788          * on the MDS, we reduce the total number of inodes to
1789          * compensate, so that the "inodes in use" number is correct.
1790          */
1791         if (obd_osfs.os_ffree < osfs->os_ffree) {
1792                 osfs->os_files = (osfs->os_files - osfs->os_ffree) +
1793                         obd_osfs.os_ffree;
1794                 osfs->os_ffree = obd_osfs.os_ffree;
1795         }
1796
1797         RETURN(rc);
1798 }
1799 int ll_statfs(struct dentry *de, struct kstatfs *sfs)
1800 {
1801         struct super_block *sb = de->d_sb;
1802         struct obd_statfs osfs;
1803         __u64 fsid = huge_encode_dev(sb->s_dev);
1804         int rc;
1805
1806         CDEBUG(D_VFSTRACE, "VFS Op: at %llu jiffies\n", get_jiffies_64());
1807         ll_stats_ops_tally(ll_s2sbi(sb), LPROC_LL_STAFS, 1);
1808
1809         /* Some amount of caching on the client is allowed */
1810         rc = ll_statfs_internal(sb, &osfs,
1811                                 cfs_time_shift_64(-OBD_STATFS_CACHE_SECONDS),
1812                                 0);
1813         if (rc)
1814                 return rc;
1815
1816         statfs_unpack(sfs, &osfs);
1817
1818         /* We need to downshift for all 32-bit kernels, because we can't
1819          * tell if the kernel is being called via sys_statfs64() or not.
1820          * Stop before overflowing f_bsize - in which case it is better
1821          * to just risk EOVERFLOW if caller is using old sys_statfs(). */
1822         if (sizeof(long) < 8) {
1823                 while (osfs.os_blocks > ~0UL && sfs->f_bsize < 0x40000000) {
1824                         sfs->f_bsize <<= 1;
1825
1826                         osfs.os_blocks >>= 1;
1827                         osfs.os_bfree >>= 1;
1828                         osfs.os_bavail >>= 1;
1829                 }
1830         }
1831
1832         sfs->f_blocks = osfs.os_blocks;
1833         sfs->f_bfree = osfs.os_bfree;
1834         sfs->f_bavail = osfs.os_bavail;
1835         sfs->f_fsid.val[0] = (__u32)fsid;
1836         sfs->f_fsid.val[1] = (__u32)(fsid >> 32);
1837         return 0;
1838 }
1839
1840 void ll_inode_size_lock(struct inode *inode)
1841 {
1842         struct ll_inode_info *lli;
1843
1844         LASSERT(!S_ISDIR(inode->i_mode));
1845
1846         lli = ll_i2info(inode);
1847         mutex_lock(&lli->lli_size_mutex);
1848 }
1849
1850 void ll_inode_size_unlock(struct inode *inode)
1851 {
1852         struct ll_inode_info *lli;
1853
1854         lli = ll_i2info(inode);
1855         mutex_unlock(&lli->lli_size_mutex);
1856 }
1857
1858 int ll_update_inode(struct inode *inode, struct lustre_md *md)
1859 {
1860         struct ll_inode_info *lli = ll_i2info(inode);
1861         struct mdt_body *body = md->body;
1862         struct ll_sb_info *sbi = ll_i2sbi(inode);
1863         int rc = 0;
1864
1865         if (body->mbo_valid & OBD_MD_FLEASIZE) {
1866                 rc = cl_file_inode_init(inode, md);
1867                 if (rc)
1868                         return rc;
1869         }
1870
1871         if (S_ISDIR(inode->i_mode)) {
1872                 rc = ll_update_lsm_md(inode, md);
1873                 if (rc != 0)
1874                         return rc;
1875         }
1876
1877 #ifdef CONFIG_FS_POSIX_ACL
1878         if (body->mbo_valid & OBD_MD_FLACL) {
1879                 spin_lock(&lli->lli_lock);
1880                 if (lli->lli_posix_acl)
1881                         posix_acl_release(lli->lli_posix_acl);
1882                 lli->lli_posix_acl = md->posix_acl;
1883                 spin_unlock(&lli->lli_lock);
1884         }
1885 #endif
1886         inode->i_ino = cl_fid_build_ino(&body->mbo_fid1,
1887                                         sbi->ll_flags & LL_SBI_32BIT_API);
1888         inode->i_generation = cl_fid_build_gen(&body->mbo_fid1);
1889
1890         if (body->mbo_valid & OBD_MD_FLATIME) {
1891                 if (body->mbo_atime > LTIME_S(inode->i_atime))
1892                         LTIME_S(inode->i_atime) = body->mbo_atime;
1893                 lli->lli_atime = body->mbo_atime;
1894         }
1895
1896         if (body->mbo_valid & OBD_MD_FLMTIME) {
1897                 if (body->mbo_mtime > LTIME_S(inode->i_mtime)) {
1898                         CDEBUG(D_INODE, "setting ino %lu mtime from %lu "
1899                                "to %llu\n", inode->i_ino,
1900                                LTIME_S(inode->i_mtime), body->mbo_mtime);
1901                         LTIME_S(inode->i_mtime) = body->mbo_mtime;
1902                 }
1903                 lli->lli_mtime = body->mbo_mtime;
1904         }
1905
1906         if (body->mbo_valid & OBD_MD_FLCTIME) {
1907                 if (body->mbo_ctime > LTIME_S(inode->i_ctime))
1908                         LTIME_S(inode->i_ctime) = body->mbo_ctime;
1909                 lli->lli_ctime = body->mbo_ctime;
1910         }
1911
1912         /* Clear i_flags to remove S_NOSEC before permissions are updated */
1913         if (body->mbo_valid & OBD_MD_FLFLAGS)
1914                 inode->i_flags = ll_ext_to_inode_flags(body->mbo_flags);
1915         if (body->mbo_valid & OBD_MD_FLMODE)
1916                 inode->i_mode = (inode->i_mode & S_IFMT) |
1917                                 (body->mbo_mode & ~S_IFMT);
1918
1919         if (body->mbo_valid & OBD_MD_FLTYPE)
1920                 inode->i_mode = (inode->i_mode & ~S_IFMT) |
1921                                 (body->mbo_mode & S_IFMT);
1922
1923         LASSERT(inode->i_mode != 0);
1924         if (S_ISREG(inode->i_mode))
1925                 inode->i_blkbits = min(PTLRPC_MAX_BRW_BITS + 1,
1926                                        LL_MAX_BLKSIZE_BITS);
1927         else
1928                 inode->i_blkbits = inode->i_sb->s_blocksize_bits;
1929
1930         if (body->mbo_valid & OBD_MD_FLUID)
1931                 inode->i_uid = make_kuid(&init_user_ns, body->mbo_uid);
1932         if (body->mbo_valid & OBD_MD_FLGID)
1933                 inode->i_gid = make_kgid(&init_user_ns, body->mbo_gid);
1934         if (body->mbo_valid & OBD_MD_FLPROJID)
1935                 lli->lli_projid = body->mbo_projid;
1936         if (body->mbo_valid & OBD_MD_FLNLINK)
1937                 set_nlink(inode, body->mbo_nlink);
1938         if (body->mbo_valid & OBD_MD_FLRDEV)
1939                 inode->i_rdev = old_decode_dev(body->mbo_rdev);
1940
1941         if (body->mbo_valid & OBD_MD_FLID) {
1942                 /* FID shouldn't be changed! */
1943                 if (fid_is_sane(&lli->lli_fid)) {
1944                         LASSERTF(lu_fid_eq(&lli->lli_fid, &body->mbo_fid1),
1945                                  "Trying to change FID "DFID
1946                                  " to the "DFID", inode "DFID"(%p)\n",
1947                                  PFID(&lli->lli_fid), PFID(&body->mbo_fid1),
1948                                  PFID(ll_inode2fid(inode)), inode);
1949                 } else {
1950                         lli->lli_fid = body->mbo_fid1;
1951                 }
1952         }
1953
1954         LASSERT(fid_seq(&lli->lli_fid) != 0);
1955
1956         if (body->mbo_valid & OBD_MD_FLSIZE) {
1957                 i_size_write(inode, body->mbo_size);
1958
1959                 CDEBUG(D_VFSTRACE, "inode="DFID", updating i_size %llu\n",
1960                        PFID(ll_inode2fid(inode)),
1961                        (unsigned long long)body->mbo_size);
1962
1963                 if (body->mbo_valid & OBD_MD_FLBLOCKS)
1964                         inode->i_blocks = body->mbo_blocks;
1965         }
1966
1967         if (body->mbo_valid & OBD_MD_TSTATE) {
1968                 /* Set LLIF_FILE_RESTORING if restore ongoing and
1969                  * clear it when done to ensure to start again
1970                  * glimpsing updated attrs
1971                  */
1972                 if (body->mbo_t_state & MS_RESTORE)
1973                         ll_file_set_flag(lli, LLIF_FILE_RESTORING);
1974                 else
1975                         ll_file_clear_flag(lli, LLIF_FILE_RESTORING);
1976         }
1977
1978         return 0;
1979 }
1980
1981 int ll_read_inode2(struct inode *inode, void *opaque)
1982 {
1983         struct lustre_md *md = opaque;
1984         struct ll_inode_info *lli = ll_i2info(inode);
1985         int     rc;
1986         ENTRY;
1987
1988         CDEBUG(D_VFSTRACE, "VFS Op:inode="DFID"(%p)\n",
1989                PFID(&lli->lli_fid), inode);
1990
1991         /* Core attributes from the MDS first.  This is a new inode, and
1992          * the VFS doesn't zero times in the core inode so we have to do
1993          * it ourselves.  They will be overwritten by either MDS or OST
1994          * attributes - we just need to make sure they aren't newer. */
1995         LTIME_S(inode->i_mtime) = 0;
1996         LTIME_S(inode->i_atime) = 0;
1997         LTIME_S(inode->i_ctime) = 0;
1998         inode->i_rdev = 0;
1999         rc = ll_update_inode(inode, md);
2000         if (rc != 0)
2001                 RETURN(rc);
2002
2003         /* OIDEBUG(inode); */
2004
2005 #ifdef HAVE_BACKING_DEV_INFO
2006         /* initializing backing dev info. */
2007         inode->i_mapping->backing_dev_info = &s2lsi(inode->i_sb)->lsi_bdi;
2008 #endif
2009         if (S_ISREG(inode->i_mode)) {
2010                 struct ll_sb_info *sbi = ll_i2sbi(inode);
2011                 inode->i_op = &ll_file_inode_operations;
2012                 inode->i_fop = sbi->ll_fop;
2013                 inode->i_mapping->a_ops = (struct address_space_operations *)&ll_aops;
2014                 EXIT;
2015         } else if (S_ISDIR(inode->i_mode)) {
2016                 inode->i_op = &ll_dir_inode_operations;
2017                 inode->i_fop = &ll_dir_operations;
2018                 EXIT;
2019         } else if (S_ISLNK(inode->i_mode)) {
2020                 inode->i_op = &ll_fast_symlink_inode_operations;
2021                 EXIT;
2022         } else {
2023                 inode->i_op = &ll_special_inode_operations;
2024
2025                 init_special_inode(inode, inode->i_mode,
2026                                    inode->i_rdev);
2027
2028                 EXIT;
2029         }
2030
2031         return 0;
2032 }
2033
2034 void ll_delete_inode(struct inode *inode)
2035 {
2036         struct ll_inode_info *lli = ll_i2info(inode);
2037         ENTRY;
2038
2039         if (S_ISREG(inode->i_mode) && lli->lli_clob != NULL)
2040                 /* It is last chance to write out dirty pages,
2041                  * otherwise we may lose data while umount */
2042                 cl_sync_file_range(inode, 0, OBD_OBJECT_EOF, CL_FSYNC_LOCAL, 1);
2043
2044         truncate_inode_pages_final(&inode->i_data);
2045
2046         LASSERTF(inode->i_data.nrpages == 0, "inode="DFID"(%p) nrpages=%lu, "
2047                  "see https://jira.hpdd.intel.com/browse/LU-118\n",
2048                  PFID(ll_inode2fid(inode)), inode, inode->i_data.nrpages);
2049
2050 #ifdef HAVE_SBOPS_EVICT_INODE
2051         ll_clear_inode(inode);
2052 #endif
2053         clear_inode(inode);
2054
2055         EXIT;
2056 }
2057
2058 int ll_iocontrol(struct inode *inode, struct file *file,
2059                  unsigned int cmd, unsigned long arg)
2060 {
2061         struct ll_sb_info *sbi = ll_i2sbi(inode);
2062         struct ptlrpc_request *req = NULL;
2063         int rc, flags = 0;
2064         ENTRY;
2065
2066         switch(cmd) {
2067         case FSFILT_IOC_GETFLAGS: {
2068                 struct mdt_body *body;
2069                 struct md_op_data *op_data;
2070
2071                 op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL,
2072                                              0, 0, LUSTRE_OPC_ANY,
2073                                              NULL);
2074                 if (IS_ERR(op_data))
2075                         RETURN(PTR_ERR(op_data));
2076
2077                 op_data->op_valid = OBD_MD_FLFLAGS;
2078                 rc = md_getattr(sbi->ll_md_exp, op_data, &req);
2079                 ll_finish_md_op_data(op_data);
2080                 if (rc) {
2081                         CERROR("%s: failure inode "DFID": rc = %d\n",
2082                                sbi->ll_md_exp->exp_obd->obd_name,
2083                                PFID(ll_inode2fid(inode)), rc);
2084                         RETURN(-abs(rc));
2085                 }
2086
2087                 body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
2088
2089                 flags = body->mbo_flags;
2090
2091                 ptlrpc_req_finished(req);
2092
2093                 RETURN(put_user(flags, (int __user *)arg));
2094         }
2095         case FSFILT_IOC_SETFLAGS: {
2096                 struct iattr *attr;
2097                 struct md_op_data *op_data;
2098                 struct cl_object *obj;
2099
2100                 if (get_user(flags, (int __user *)arg))
2101                         RETURN(-EFAULT);
2102
2103                 op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL, 0, 0,
2104                                              LUSTRE_OPC_ANY, NULL);
2105                 if (IS_ERR(op_data))
2106                         RETURN(PTR_ERR(op_data));
2107
2108                 op_data->op_attr_flags = flags;
2109                 op_data->op_attr.ia_valid |= ATTR_ATTR_FLAG;
2110                 rc = md_setattr(sbi->ll_md_exp, op_data, NULL, 0, &req);
2111                 ll_finish_md_op_data(op_data);
2112                 ptlrpc_req_finished(req);
2113                 if (rc)
2114                         RETURN(rc);
2115
2116                 inode->i_flags = ll_ext_to_inode_flags(flags);
2117
2118                 obj = ll_i2info(inode)->lli_clob;
2119                 if (obj == NULL)
2120                         RETURN(0);
2121
2122                 OBD_ALLOC_PTR(attr);
2123                 if (attr == NULL)
2124                         RETURN(-ENOMEM);
2125
2126                 attr->ia_valid = ATTR_ATTR_FLAG;
2127                 rc = cl_setattr_ost(obj, attr, flags);
2128
2129                 OBD_FREE_PTR(attr);
2130                 RETURN(rc);
2131         }
2132         default:
2133                 RETURN(-ENOSYS);
2134         }
2135
2136         RETURN(0);
2137 }
2138
2139 int ll_flush_ctx(struct inode *inode)
2140 {
2141         struct ll_sb_info  *sbi = ll_i2sbi(inode);
2142
2143         CDEBUG(D_SEC, "flush context for user %d\n",
2144                from_kuid(&init_user_ns, current_uid()));
2145
2146         obd_set_info_async(NULL, sbi->ll_md_exp,
2147                            sizeof(KEY_FLUSH_CTX), KEY_FLUSH_CTX,
2148                            0, NULL, NULL);
2149         obd_set_info_async(NULL, sbi->ll_dt_exp,
2150                            sizeof(KEY_FLUSH_CTX), KEY_FLUSH_CTX,
2151                            0, NULL, NULL);
2152         return 0;
2153 }
2154
2155 /* umount -f client means force down, don't save state */
2156 void ll_umount_begin(struct super_block *sb)
2157 {
2158         struct ll_sb_info *sbi = ll_s2sbi(sb);
2159         struct obd_device *obd;
2160         struct obd_ioctl_data *ioc_data;
2161         struct l_wait_info lwi;
2162         wait_queue_head_t waitq;
2163         ENTRY;
2164
2165         CDEBUG(D_VFSTRACE, "VFS Op: superblock %p count %d active %d\n", sb,
2166                sb->s_count, atomic_read(&sb->s_active));
2167
2168         obd = class_exp2obd(sbi->ll_md_exp);
2169         if (obd == NULL) {
2170                 CERROR("Invalid MDC connection handle %#llx\n",
2171                        sbi->ll_md_exp->exp_handle.h_cookie);
2172                 EXIT;
2173                 return;
2174         }
2175         obd->obd_force = 1;
2176
2177         obd = class_exp2obd(sbi->ll_dt_exp);
2178         if (obd == NULL) {
2179                 CERROR("Invalid LOV connection handle %#llx\n",
2180                        sbi->ll_dt_exp->exp_handle.h_cookie);
2181                 EXIT;
2182                 return;
2183         }
2184         obd->obd_force = 1;
2185
2186         OBD_ALLOC_PTR(ioc_data);
2187         if (ioc_data) {
2188                 obd_iocontrol(IOC_OSC_SET_ACTIVE, sbi->ll_md_exp,
2189                               sizeof *ioc_data, ioc_data, NULL);
2190
2191                 obd_iocontrol(IOC_OSC_SET_ACTIVE, sbi->ll_dt_exp,
2192                               sizeof *ioc_data, ioc_data, NULL);
2193
2194                 OBD_FREE_PTR(ioc_data);
2195         }
2196
2197         /* Really, we'd like to wait until there are no requests outstanding,
2198          * and then continue.  For now, we just periodically checking for vfs
2199          * to decrement mnt_cnt and hope to finish it within 10sec.
2200          */
2201         init_waitqueue_head(&waitq);
2202         lwi = LWI_TIMEOUT_INTERVAL(cfs_time_seconds(10),
2203                                    cfs_time_seconds(1), NULL, NULL);
2204         l_wait_event(waitq, may_umount(sbi->ll_mnt.mnt), &lwi);
2205
2206         EXIT;
2207 }
2208
2209 int ll_remount_fs(struct super_block *sb, int *flags, char *data)
2210 {
2211         struct ll_sb_info *sbi = ll_s2sbi(sb);
2212         char *profilenm = get_profile_name(sb);
2213         int err;
2214         __u32 read_only;
2215
2216         if ((*flags & MS_RDONLY) != (sb->s_flags & MS_RDONLY)) {
2217                 read_only = *flags & MS_RDONLY;
2218                 err = obd_set_info_async(NULL, sbi->ll_md_exp,
2219                                          sizeof(KEY_READ_ONLY),
2220                                          KEY_READ_ONLY, sizeof(read_only),
2221                                          &read_only, NULL);
2222                 if (err) {
2223                         LCONSOLE_WARN("Failed to remount %s %s (%d)\n",
2224                                       profilenm, read_only ?
2225                                       "read-only" : "read-write", err);
2226                         return err;
2227                 }
2228
2229                 if (read_only)
2230                         sb->s_flags |= MS_RDONLY;
2231                 else
2232                         sb->s_flags &= ~MS_RDONLY;
2233
2234                 if (sbi->ll_flags & LL_SBI_VERBOSE)
2235                         LCONSOLE_WARN("Remounted %s %s\n", profilenm,
2236                                       read_only ?  "read-only" : "read-write");
2237         }
2238         return 0;
2239 }
2240
2241 /**
2242  * Cleanup the open handle that is cached on MDT-side.
2243  *
2244  * For open case, the client side open handling thread may hit error
2245  * after the MDT grant the open. Under such case, the client should
2246  * send close RPC to the MDT as cleanup; otherwise, the open handle
2247  * on the MDT will be leaked there until the client umount or evicted.
2248  *
2249  * In further, if someone unlinked the file, because the open handle
2250  * holds the reference on such file/object, then it will block the
2251  * subsequent threads that want to locate such object via FID.
2252  *
2253  * \param[in] sb        super block for this file-system
2254  * \param[in] open_req  pointer to the original open request
2255  */
2256 void ll_open_cleanup(struct super_block *sb, struct ptlrpc_request *open_req)
2257 {
2258         struct mdt_body                 *body;
2259         struct md_op_data               *op_data;
2260         struct ptlrpc_request           *close_req = NULL;
2261         struct obd_export               *exp       = ll_s2sbi(sb)->ll_md_exp;
2262         ENTRY;
2263
2264         body = req_capsule_server_get(&open_req->rq_pill, &RMF_MDT_BODY);
2265         OBD_ALLOC_PTR(op_data);
2266         if (op_data == NULL) {
2267                 CWARN("%s: cannot allocate op_data to release open handle for "
2268                       DFID"\n",
2269                       ll_get_fsname(sb, NULL, 0), PFID(&body->mbo_fid1));
2270
2271                 RETURN_EXIT;
2272         }
2273
2274         op_data->op_fid1 = body->mbo_fid1;
2275         op_data->op_handle = body->mbo_handle;
2276         op_data->op_mod_time = ktime_get_real_seconds();
2277         md_close(exp, op_data, NULL, &close_req);
2278         ptlrpc_req_finished(close_req);
2279         ll_finish_md_op_data(op_data);
2280
2281         EXIT;
2282 }
2283
2284 int ll_prep_inode(struct inode **inode, struct ptlrpc_request *req,
2285                   struct super_block *sb, struct lookup_intent *it)
2286 {
2287         struct ll_sb_info *sbi = NULL;
2288         struct lustre_md md = { NULL };
2289         int rc;
2290         ENTRY;
2291
2292         LASSERT(*inode || sb);
2293         sbi = sb ? ll_s2sbi(sb) : ll_i2sbi(*inode);
2294         rc = md_get_lustre_md(sbi->ll_md_exp, req, sbi->ll_dt_exp,
2295                               sbi->ll_md_exp, &md);
2296         if (rc != 0)
2297                 GOTO(cleanup, rc);
2298
2299         if (*inode) {
2300                 rc = ll_update_inode(*inode, &md);
2301                 if (rc != 0)
2302                         GOTO(out, rc);
2303         } else {
2304                 LASSERT(sb != NULL);
2305
2306                 /*
2307                  * At this point server returns to client's same fid as client
2308                  * generated for creating. So using ->fid1 is okay here.
2309                  */
2310                 if (!fid_is_sane(&md.body->mbo_fid1)) {
2311                         CERROR("%s: Fid is insane "DFID"\n",
2312                                 ll_get_fsname(sb, NULL, 0),
2313                                 PFID(&md.body->mbo_fid1));
2314                         GOTO(out, rc = -EINVAL);
2315                 }
2316
2317                 *inode = ll_iget(sb, cl_fid_build_ino(&md.body->mbo_fid1,
2318                                              sbi->ll_flags & LL_SBI_32BIT_API),
2319                                  &md);
2320                 if (IS_ERR(*inode)) {
2321 #ifdef CONFIG_FS_POSIX_ACL
2322                         if (md.posix_acl) {
2323                                 posix_acl_release(md.posix_acl);
2324                                 md.posix_acl = NULL;
2325                         }
2326 #endif
2327                         rc = IS_ERR(*inode) ? PTR_ERR(*inode) : -ENOMEM;
2328                         *inode = NULL;
2329                         CERROR("new_inode -fatal: rc %d\n", rc);
2330                         GOTO(out, rc);
2331                 }
2332         }
2333
2334         /* Handling piggyback layout lock.
2335          * Layout lock can be piggybacked by getattr and open request.
2336          * The lsm can be applied to inode only if it comes with a layout lock
2337          * otherwise correct layout may be overwritten, for example:
2338          * 1. proc1: mdt returns a lsm but not granting layout
2339          * 2. layout was changed by another client
2340          * 3. proc2: refresh layout and layout lock granted
2341          * 4. proc1: to apply a stale layout */
2342         if (it != NULL && it->it_lock_mode != 0) {
2343                 struct lustre_handle lockh;
2344                 struct ldlm_lock *lock;
2345
2346                 lockh.cookie = it->it_lock_handle;
2347                 lock = ldlm_handle2lock(&lockh);
2348                 LASSERT(lock != NULL);
2349                 if (ldlm_has_layout(lock)) {
2350                         struct cl_object_conf conf;
2351
2352                         memset(&conf, 0, sizeof(conf));
2353                         conf.coc_opc = OBJECT_CONF_SET;
2354                         conf.coc_inode = *inode;
2355                         conf.coc_lock = lock;
2356                         conf.u.coc_layout = md.layout;
2357                         (void)ll_layout_conf(*inode, &conf);
2358                 }
2359                 LDLM_LOCK_PUT(lock);
2360         }
2361
2362         GOTO(out, rc = 0);
2363
2364 out:
2365         md_free_lustre_md(sbi->ll_md_exp, &md);
2366
2367 cleanup:
2368         if (rc != 0 && it != NULL && it->it_op & IT_OPEN)
2369                 ll_open_cleanup(sb != NULL ? sb : (*inode)->i_sb, req);
2370
2371         return rc;
2372 }
2373
2374 int ll_obd_statfs(struct inode *inode, void __user *arg)
2375 {
2376         struct ll_sb_info *sbi = NULL;
2377         struct obd_export *exp;
2378         char *buf = NULL;
2379         struct obd_ioctl_data *data = NULL;
2380         __u32 type;
2381         int len = 0, rc;
2382
2383         if (!inode || !(sbi = ll_i2sbi(inode)))
2384                 GOTO(out_statfs, rc = -EINVAL);
2385
2386         rc = obd_ioctl_getdata(&buf, &len, arg);
2387         if (rc)
2388                 GOTO(out_statfs, rc);
2389
2390         data = (void*)buf;
2391         if (!data->ioc_inlbuf1 || !data->ioc_inlbuf2 ||
2392             !data->ioc_pbuf1 || !data->ioc_pbuf2)
2393                 GOTO(out_statfs, rc = -EINVAL);
2394
2395         if (data->ioc_inllen1 != sizeof(__u32) ||
2396             data->ioc_inllen2 != sizeof(__u32) ||
2397             data->ioc_plen1 != sizeof(struct obd_statfs) ||
2398             data->ioc_plen2 != sizeof(struct obd_uuid))
2399                 GOTO(out_statfs, rc = -EINVAL);
2400
2401         memcpy(&type, data->ioc_inlbuf1, sizeof(__u32));
2402         if (type & LL_STATFS_LMV)
2403                 exp = sbi->ll_md_exp;
2404         else if (type & LL_STATFS_LOV)
2405                 exp = sbi->ll_dt_exp;
2406         else
2407                 GOTO(out_statfs, rc = -ENODEV);
2408
2409         rc = obd_iocontrol(IOC_OBD_STATFS, exp, len, buf, NULL);
2410         if (rc)
2411                 GOTO(out_statfs, rc);
2412 out_statfs:
2413         OBD_FREE_LARGE(buf, len);
2414         return rc;
2415 }
2416
2417 int ll_process_config(struct lustre_cfg *lcfg)
2418 {
2419         struct super_block *sb;
2420         unsigned long x;
2421         int rc = 0;
2422         char *ptr;
2423
2424         /* The instance name contains the sb: lustre-client-aacfe000 */
2425         ptr = strrchr(lustre_cfg_string(lcfg, 0), '-');
2426         if (!ptr || !*(++ptr))
2427                 return -EINVAL;
2428         if (sscanf(ptr, "%lx", &x) != 1)
2429                 return -EINVAL;
2430         sb = (struct super_block *)x;
2431         /* This better be a real Lustre superblock! */
2432         LASSERT(s2lsi(sb)->lsi_lmd->lmd_magic == LMD_MAGIC);
2433
2434         /* Note we have not called client_common_fill_super yet, so
2435            proc fns must be able to handle that! */
2436         rc = class_process_proc_param(PARAM_LLITE, lprocfs_llite_obd_vars,
2437                                       lcfg, sb);
2438         if (rc > 0)
2439                 rc = 0;
2440         return rc;
2441 }
2442
2443 /* this function prepares md_op_data hint for passing it down to MD stack. */
2444 struct md_op_data *ll_prep_md_op_data(struct md_op_data *op_data,
2445                                       struct inode *i1, struct inode *i2,
2446                                       const char *name, size_t namelen,
2447                                       __u32 mode, __u32 opc, void *data)
2448 {
2449         LASSERT(i1 != NULL);
2450
2451         if (name == NULL) {
2452                 /* Do not reuse namelen for something else. */
2453                 if (namelen != 0)
2454                         return ERR_PTR(-EINVAL);
2455         } else {
2456                 if (namelen > ll_i2sbi(i1)->ll_namelen)
2457                         return ERR_PTR(-ENAMETOOLONG);
2458
2459                 if (!lu_name_is_valid_2(name, namelen))
2460                         return ERR_PTR(-EINVAL);
2461         }
2462
2463         if (op_data == NULL)
2464                 OBD_ALLOC_PTR(op_data);
2465
2466         if (op_data == NULL)
2467                 return ERR_PTR(-ENOMEM);
2468
2469         ll_i2gids(op_data->op_suppgids, i1, i2);
2470         op_data->op_fid1 = *ll_inode2fid(i1);
2471         op_data->op_default_stripe_offset = -1;
2472         if (S_ISDIR(i1->i_mode)) {
2473                 op_data->op_mea1 = ll_i2info(i1)->lli_lsm_md;
2474                 if (opc == LUSTRE_OPC_MKDIR)
2475                         op_data->op_default_stripe_offset =
2476                                    ll_i2info(i1)->lli_def_stripe_offset;
2477         }
2478
2479         if (i2) {
2480                 op_data->op_fid2 = *ll_inode2fid(i2);
2481                 if (S_ISDIR(i2->i_mode))
2482                         op_data->op_mea2 = ll_i2info(i2)->lli_lsm_md;
2483         } else {
2484                 fid_zero(&op_data->op_fid2);
2485         }
2486
2487         if (ll_i2sbi(i1)->ll_flags & LL_SBI_64BIT_HASH)
2488                 op_data->op_cli_flags |= CLI_HASH64;
2489
2490         if (ll_need_32bit_api(ll_i2sbi(i1)))
2491                 op_data->op_cli_flags |= CLI_API32;
2492
2493         op_data->op_name = name;
2494         op_data->op_namelen = namelen;
2495         op_data->op_mode = mode;
2496         op_data->op_mod_time = cfs_time_current_sec();
2497         op_data->op_fsuid = from_kuid(&init_user_ns, current_fsuid());
2498         op_data->op_fsgid = from_kgid(&init_user_ns, current_fsgid());
2499         op_data->op_cap = cfs_curproc_cap_pack();
2500         if ((opc == LUSTRE_OPC_CREATE) && (name != NULL) &&
2501              filename_is_volatile(name, namelen, &op_data->op_mds)) {
2502                 op_data->op_bias |= MDS_CREATE_VOLATILE;
2503         } else {
2504                 op_data->op_mds = 0;
2505         }
2506         op_data->op_data = data;
2507
2508         return op_data;
2509 }
2510
2511 void ll_finish_md_op_data(struct md_op_data *op_data)
2512 {
2513         security_release_secctx(op_data->op_file_secctx,
2514                                 op_data->op_file_secctx_size);
2515         OBD_FREE_PTR(op_data);
2516 }
2517
2518 #ifdef HAVE_SUPEROPS_USE_DENTRY
2519 int ll_show_options(struct seq_file *seq, struct dentry *dentry)
2520 #else
2521 int ll_show_options(struct seq_file *seq, struct vfsmount *vfs)
2522 #endif
2523 {
2524         struct ll_sb_info *sbi;
2525
2526 #ifdef HAVE_SUPEROPS_USE_DENTRY
2527         LASSERT((seq != NULL) && (dentry != NULL));
2528         sbi = ll_s2sbi(dentry->d_sb);
2529 #else
2530         LASSERT((seq != NULL) && (vfs != NULL));
2531         sbi = ll_s2sbi(vfs->mnt_sb);
2532 #endif
2533
2534         if (sbi->ll_flags & LL_SBI_NOLCK)
2535                 seq_puts(seq, ",nolock");
2536
2537         if (sbi->ll_flags & LL_SBI_FLOCK)
2538                 seq_puts(seq, ",flock");
2539
2540         if (sbi->ll_flags & LL_SBI_LOCALFLOCK)
2541                 seq_puts(seq, ",localflock");
2542
2543         if (sbi->ll_flags & LL_SBI_USER_XATTR)
2544                 seq_puts(seq, ",user_xattr");
2545
2546         if (sbi->ll_flags & LL_SBI_LAZYSTATFS)
2547                 seq_puts(seq, ",lazystatfs");
2548
2549         if (sbi->ll_flags & LL_SBI_USER_FID2PATH)
2550                 seq_puts(seq, ",user_fid2path");
2551
2552         if (sbi->ll_flags & LL_SBI_ALWAYS_PING)
2553                 seq_puts(seq, ",always_ping");
2554
2555         RETURN(0);
2556 }
2557
2558 /**
2559  * Get obd name by cmd, and copy out to user space
2560  */
2561 int ll_get_obd_name(struct inode *inode, unsigned int cmd, unsigned long arg)
2562 {
2563         struct ll_sb_info *sbi = ll_i2sbi(inode);
2564         struct obd_device *obd;
2565         ENTRY;
2566
2567         if (cmd == OBD_IOC_GETDTNAME)
2568                 obd = class_exp2obd(sbi->ll_dt_exp);
2569         else if (cmd == OBD_IOC_GETMDNAME)
2570                 obd = class_exp2obd(sbi->ll_md_exp);
2571         else
2572                 RETURN(-EINVAL);
2573
2574         if (!obd)
2575                 RETURN(-ENOENT);
2576
2577         if (copy_to_user((void __user *)arg, obd->obd_name,
2578                          strlen(obd->obd_name) + 1))
2579                 RETURN(-EFAULT);
2580
2581         RETURN(0);
2582 }
2583
2584 /**
2585  * Get lustre file system name by \a sbi. If \a buf is provided(non-NULL), the
2586  * fsname will be returned in this buffer; otherwise, a static buffer will be
2587  * used to store the fsname and returned to caller.
2588  */
2589 char *ll_get_fsname(struct super_block *sb, char *buf, int buflen)
2590 {
2591         static char fsname_static[MTI_NAME_MAXLEN];
2592         struct lustre_sb_info *lsi = s2lsi(sb);
2593         char *ptr;
2594         int len;
2595
2596         if (buf == NULL) {
2597                 /* this means the caller wants to use static buffer
2598                  * and it doesn't care about race. Usually this is
2599                  * in error reporting path */
2600                 buf = fsname_static;
2601                 buflen = sizeof(fsname_static);
2602         }
2603
2604         len = strlen(lsi->lsi_lmd->lmd_profile);
2605         ptr = strrchr(lsi->lsi_lmd->lmd_profile, '-');
2606         if (ptr && (strcmp(ptr, "-client") == 0))
2607                 len -= 7;
2608
2609         if (unlikely(len >= buflen))
2610                 len = buflen - 1;
2611         strncpy(buf, lsi->lsi_lmd->lmd_profile, len);
2612         buf[len] = '\0';
2613
2614         return buf;
2615 }
2616
2617 static char* ll_d_path(struct dentry *dentry, char *buf, int bufsize)
2618 {
2619         char *path = NULL;
2620
2621         struct path p;
2622
2623         p.dentry = dentry;
2624         p.mnt = current->fs->root.mnt;
2625         path_get(&p);
2626         path = d_path(&p, buf, bufsize);
2627         path_put(&p);
2628         return path;
2629 }
2630
2631 void ll_dirty_page_discard_warn(struct page *page, int ioret)
2632 {
2633         char *buf, *path = NULL;
2634         struct dentry *dentry = NULL;
2635         struct inode *inode = page->mapping->host;
2636
2637         /* this can be called inside spin lock so use GFP_ATOMIC. */
2638         buf = (char *)__get_free_page(GFP_ATOMIC);
2639         if (buf != NULL) {
2640                 dentry = d_find_alias(page->mapping->host);
2641                 if (dentry != NULL)
2642                         path = ll_d_path(dentry, buf, PAGE_SIZE);
2643         }
2644
2645         CDEBUG(D_WARNING,
2646                "%s: dirty page discard: %s/fid: "DFID"/%s may get corrupted "
2647                "(rc %d)\n", ll_get_fsname(page->mapping->host->i_sb, NULL, 0),
2648                s2lsi(page->mapping->host->i_sb)->lsi_lmd->lmd_dev,
2649                PFID(ll_inode2fid(inode)),
2650                (path && !IS_ERR(path)) ? path : "", ioret);
2651
2652         if (dentry != NULL)
2653                 dput(dentry);
2654
2655         if (buf != NULL)
2656                 free_page((unsigned long)buf);
2657 }
2658
2659 ssize_t ll_copy_user_md(const struct lov_user_md __user *md,
2660                         struct lov_user_md **kbuf)
2661 {
2662         struct lov_user_md      lum;
2663         ssize_t                 lum_size;
2664         ENTRY;
2665
2666         if (copy_from_user(&lum, md, sizeof(lum)))
2667                 RETURN(-EFAULT);
2668
2669         lum_size = ll_lov_user_md_size(&lum);
2670         if (lum_size < 0)
2671                 RETURN(lum_size);
2672
2673         OBD_ALLOC(*kbuf, lum_size);
2674         if (*kbuf == NULL)
2675                 RETURN(-ENOMEM);
2676
2677         if (copy_from_user(*kbuf, md, lum_size) != 0) {
2678                 OBD_FREE(*kbuf, lum_size);
2679                 RETURN(-EFAULT);
2680         }
2681
2682         RETURN(lum_size);
2683 }
2684
2685 /*
2686  * Compute llite root squash state after a change of root squash
2687  * configuration setting or add/remove of a lnet nid
2688  */
2689 void ll_compute_rootsquash_state(struct ll_sb_info *sbi)
2690 {
2691         struct root_squash_info *squash = &sbi->ll_squash;
2692         int i;
2693         bool matched;
2694         struct lnet_process_id id;
2695
2696         /* Update norootsquash flag */
2697         down_write(&squash->rsi_sem);
2698         if (list_empty(&squash->rsi_nosquash_nids))
2699                 sbi->ll_flags &= ~LL_SBI_NOROOTSQUASH;
2700         else {
2701                 /* Do not apply root squash as soon as one of our NIDs is
2702                  * in the nosquash_nids list */
2703                 matched = false;
2704                 i = 0;
2705                 while (LNetGetId(i++, &id) != -ENOENT) {
2706                         if (LNET_NETTYP(LNET_NIDNET(id.nid)) == LOLND)
2707                                 continue;
2708                         if (cfs_match_nid(id.nid, &squash->rsi_nosquash_nids)) {
2709                                 matched = true;
2710                                 break;
2711                         }
2712                 }
2713                 if (matched)
2714                         sbi->ll_flags |= LL_SBI_NOROOTSQUASH;
2715                 else
2716                         sbi->ll_flags &= ~LL_SBI_NOROOTSQUASH;
2717         }
2718         up_write(&squash->rsi_sem);
2719 }
2720
2721 /**
2722  * Parse linkea content to extract information about a given hardlink
2723  *
2724  * \param[in]   ldata      - Initialized linkea data
2725  * \param[in]   linkno     - Link identifier
2726  * \param[out]  parent_fid - The entry's parent FID
2727  * \param[out]  ln         - Entry name destination buffer
2728  *
2729  * \retval 0 on success
2730  * \retval Appropriate negative error code on failure
2731  */
2732 static int ll_linkea_decode(struct linkea_data *ldata, unsigned int linkno,
2733                             struct lu_fid *parent_fid, struct lu_name *ln)
2734 {
2735         unsigned int    idx;
2736         int             rc;
2737         ENTRY;
2738
2739         rc = linkea_init_with_rec(ldata);
2740         if (rc < 0)
2741                 RETURN(rc);
2742
2743         if (linkno >= ldata->ld_leh->leh_reccount)
2744                 /* beyond last link */
2745                 RETURN(-ENODATA);
2746
2747         linkea_first_entry(ldata);
2748         for (idx = 0; ldata->ld_lee != NULL; idx++) {
2749                 linkea_entry_unpack(ldata->ld_lee, &ldata->ld_reclen, ln,
2750                                     parent_fid);
2751                 if (idx == linkno)
2752                         break;
2753
2754                 linkea_next_entry(ldata);
2755         }
2756
2757         if (idx < linkno)
2758                 RETURN(-ENODATA);
2759
2760         RETURN(0);
2761 }
2762
2763 /**
2764  * Get parent FID and name of an identified link. Operation is performed for
2765  * a given link number, letting the caller iterate over linkno to list one or
2766  * all links of an entry.
2767  *
2768  * \param[in]     file - File descriptor against which to perform the operation
2769  * \param[in,out] arg  - User-filled structure containing the linkno to operate
2770  *                       on and the available size. It is eventually filled with
2771  *                       the requested information or left untouched on error
2772  *
2773  * \retval - 0 on success
2774  * \retval - Appropriate negative error code on failure
2775  */
2776 int ll_getparent(struct file *file, struct getparent __user *arg)
2777 {
2778         struct inode            *inode = file_inode(file);
2779         struct linkea_data      *ldata;
2780         struct lu_buf            buf = LU_BUF_NULL;
2781         struct lu_name           ln;
2782         struct lu_fid            parent_fid;
2783         __u32                    linkno;
2784         __u32                    name_size;
2785         int                      rc;
2786
2787         ENTRY;
2788
2789         if (!cfs_capable(CFS_CAP_DAC_READ_SEARCH) &&
2790             !(ll_i2sbi(inode)->ll_flags & LL_SBI_USER_FID2PATH))
2791                 RETURN(-EPERM);
2792
2793         if (get_user(name_size, &arg->gp_name_size))
2794                 RETURN(-EFAULT);
2795
2796         if (get_user(linkno, &arg->gp_linkno))
2797                 RETURN(-EFAULT);
2798
2799         if (name_size > PATH_MAX)
2800                 RETURN(-EINVAL);
2801
2802         OBD_ALLOC(ldata, sizeof(*ldata));
2803         if (ldata == NULL)
2804                 RETURN(-ENOMEM);
2805
2806         rc = linkea_data_new(ldata, &buf);
2807         if (rc < 0)
2808                 GOTO(ldata_free, rc);
2809
2810 #ifdef HAVE_XATTR_HANDLER_FLAGS
2811         rc = ll_xattr_list(inode, XATTR_NAME_LINK, XATTR_TRUSTED_T, buf.lb_buf,
2812                            buf.lb_len, OBD_MD_FLXATTR);
2813 #else
2814         rc = ll_getxattr(file_dentry(file), XATTR_NAME_LINK, buf.lb_buf,
2815                          buf.lb_len);
2816 #endif /* HAVE_XATTR_HANDLER_FLAGS */
2817         if (rc < 0)
2818                 GOTO(lb_free, rc);
2819
2820         rc = ll_linkea_decode(ldata, linkno, &parent_fid, &ln);
2821         if (rc < 0)
2822                 GOTO(lb_free, rc);
2823
2824         if (ln.ln_namelen >= name_size)
2825                 GOTO(lb_free, rc = -EOVERFLOW);
2826
2827         if (copy_to_user(&arg->gp_fid, &parent_fid, sizeof(arg->gp_fid)))
2828                 GOTO(lb_free, rc = -EFAULT);
2829
2830         if (copy_to_user(&arg->gp_name, ln.ln_name, ln.ln_namelen))
2831                 GOTO(lb_free, rc = -EFAULT);
2832
2833         if (put_user('\0', arg->gp_name + ln.ln_namelen))
2834                 GOTO(lb_free, rc = -EFAULT);
2835
2836 lb_free:
2837         lu_buf_free(&buf);
2838 ldata_free:
2839         OBD_FREE(ldata, sizeof(*ldata));
2840
2841         RETURN(rc);
2842 }