lustre/llite/llite_lib.c

   1 /*
   2  * GPL HEADER START
   3  *
   4  * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
   5  *
   6  * This program is free software; you can redistribute it and/or modify
   7  * it under the terms of the GNU General Public License version 2 only,
   8  * as published by the Free Software Foundation.
   9  *
  10  * This program is distributed in the hope that it will be useful, but
  11  * WITHOUT ANY WARRANTY; without even the implied warranty of
  12  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  13  * General Public License version 2 for more details (a copy is included
  14  * in the LICENSE file that accompanied this code).
  15  *
  16  * You should have received a copy of the GNU General Public License
  17  * version 2 along with this program; If not, see
  18  * http://www.gnu.org/licenses/gpl-2.0.html
  19  *
  20  * GPL HEADER END
  21  */
  22 /*
  23  * Copyright (c) 2003, 2010, Oracle and/or its affiliates. All rights reserved.
  24  * Use is subject to license terms.
  25  *
  26  * Copyright (c) 2011, 2017, Intel Corporation.
  27  */
  28 /*
  29  * This file is part of Lustre, http://www.lustre.org/
  30  * Lustre is a trademark of Sun Microsystems, Inc.
  31  *
  32  * lustre/llite/llite_lib.c
  33  *
  34  * Lustre Light Super operations
  35  */
  36
  37 #define DEBUG_SUBSYSTEM S_LLITE
  38
  39 #include <linux/module.h>
  40 #include <linux/statfs.h>
  41 #include <linux/time.h>
  42 #include <linux/types.h>
  43 #include <linux/version.h>
  44 #include <linux/mm.h>
  45 #include <linux/user_namespace.h>
  46 #ifdef HAVE_UIDGID_HEADER
  47 # include <linux/uidgid.h>
  48 #endif
  49 #include <linux/security.h>
  50
  51 #include <uapi/linux/lustre/lustre_ioctl.h>
  52 #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 = -1;
 103
 104         ll_generate_random_uuid(uuid);
 105         class_uuid_unparse(uuid, &sbi->ll_sb_uuid);
 106         CDEBUG(D_CONFIG, "generated uuid: %s\n", sbi->ll_sb_uuid.uuid);
 107
 108         sbi->ll_flags |= LL_SBI_VERBOSE;
 109 #ifdef ENABLE_CHECKSUM
 110         sbi->ll_flags |= LL_SBI_CHECKSUM;
 111 #endif
 112
 113 #ifdef HAVE_LRU_RESIZE_SUPPORT
 114         sbi->ll_flags |= LL_SBI_LRU_RESIZE;
 115 #endif
 116         sbi->ll_flags |= LL_SBI_LAZYSTATFS;
 117
 118         for (i = 0; i <= LL_PROCESS_HIST_MAX; i++) {
 119                 spin_lock_init(&sbi->ll_rw_extents_info.pp_extents[i].
 120                                pp_r_hist.oh_lock);
 121                 spin_lock_init(&sbi->ll_rw_extents_info.pp_extents[i].
 122                                pp_w_hist.oh_lock);
 123         }
 124
 125         /* metadata statahead is enabled by default */
 126         sbi->ll_sa_max = LL_SA_RPC_DEF;
 127         atomic_set(&sbi->ll_sa_total, 0);
 128         atomic_set(&sbi->ll_sa_wrong, 0);
 129         atomic_set(&sbi->ll_sa_running, 0);
 130         atomic_set(&sbi->ll_agl_total, 0);
 131         sbi->ll_flags |= LL_SBI_AGL_ENABLED;
 132         sbi->ll_flags |= LL_SBI_FAST_READ;
 133
 134         /* root squash */
 135         sbi->ll_squash.rsi_uid = 0;
 136         sbi->ll_squash.rsi_gid = 0;
 137         INIT_LIST_HEAD(&sbi->ll_squash.rsi_nosquash_nids);
 138         init_rwsem(&sbi->ll_squash.rsi_sem);
 139
 140         RETURN(sbi);
 141 }
 142
 143 static void ll_free_sbi(struct super_block *sb)
 144 {
 145         struct ll_sb_info *sbi = ll_s2sbi(sb);
 146         ENTRY;
 147
 148         if (sbi != NULL) {
 149                 if (!list_empty(&sbi->ll_squash.rsi_nosquash_nids))
 150                         cfs_free_nidlist(&sbi->ll_squash.rsi_nosquash_nids);
 151                 if (sbi->ll_cache != NULL) {
 152                         cl_cache_decref(sbi->ll_cache);
 153                         sbi->ll_cache = NULL;
 154                 }
 155                 OBD_FREE(sbi, sizeof(*sbi));
 156         }
 157         EXIT;
 158 }
 159
 160 static inline int obd_connect_has_secctx(struct obd_connect_data *data)
 161 {
 162         return data->ocd_connect_flags & OBD_CONNECT_FLAGS2 &&
 163                data->ocd_connect_flags2 & OBD_CONNECT2_FILE_SECCTX;
 164 }
 165
 166 static int client_common_fill_super(struct super_block *sb, char *md, char *dt,
 167                                     struct vfsmount *mnt)
 168 {
 169         struct inode *root = NULL;
 170         struct ll_sb_info *sbi = ll_s2sbi(sb);
 171         struct obd_device *obd;
 172         struct obd_statfs *osfs = NULL;
 173         struct ptlrpc_request *request = NULL;
 174         struct obd_connect_data *data = NULL;
 175         struct obd_uuid *uuid;
 176         struct md_op_data *op_data;
 177         struct lustre_md lmd;
 178         u64 valid;
 179         int size, err, checksum;
 180         ENTRY;
 181
 182         obd = class_name2obd(md);
 183         if (!obd) {
 184                 CERROR("MD %s: not setup or attached\n", md);
 185                 RETURN(-EINVAL);
 186         }
 187
 188         OBD_ALLOC_PTR(data);
 189         if (data == NULL)
 190                 RETURN(-ENOMEM);
 191
 192         OBD_ALLOC_PTR(osfs);
 193         if (osfs == NULL) {
 194                 OBD_FREE_PTR(data);
 195                 RETURN(-ENOMEM);
 196         }
 197
 198         /* pass client page size via ocd_grant_blkbits, the server should report
 199          * back its backend blocksize for grant calculation purpose */
 200         data->ocd_grant_blkbits = PAGE_SHIFT;
 201
 202         /* indicate MDT features supported by this client */
 203         data->ocd_connect_flags = OBD_CONNECT_IBITS    | OBD_CONNECT_NODEVOH  |
 204                                   OBD_CONNECT_ATTRFID  | OBD_CONNECT_GRANT |
 205                                   OBD_CONNECT_VERSION  | OBD_CONNECT_BRW_SIZE |
 206                                   OBD_CONNECT_SRVLOCK  | OBD_CONNECT_TRUNCLOCK|
 207                                   OBD_CONNECT_MDS_CAPA | OBD_CONNECT_OSS_CAPA |
 208                                   OBD_CONNECT_CANCELSET | OBD_CONNECT_FID     |
 209                                   OBD_CONNECT_AT       | OBD_CONNECT_LOV_V3   |
 210                                   OBD_CONNECT_VBR | OBD_CONNECT_FULL20 |
 211                                   OBD_CONNECT_64BITHASH |
 212                                   OBD_CONNECT_EINPROGRESS |
 213                                   OBD_CONNECT_JOBSTATS | OBD_CONNECT_LVB_TYPE |
 214                                   OBD_CONNECT_LAYOUTLOCK | OBD_CONNECT_PINGLESS|
 215                                   OBD_CONNECT_MAX_EASIZE |
 216                                   OBD_CONNECT_FLOCK_DEAD |
 217                                   OBD_CONNECT_DISP_STRIPE | OBD_CONNECT_LFSCK |
 218                                   OBD_CONNECT_OPEN_BY_FID |
 219                                   OBD_CONNECT_DIR_STRIPE |
 220                                   OBD_CONNECT_BULK_MBITS | OBD_CONNECT_CKSUM |
 221                                   OBD_CONNECT_SUBTREE |
 222                                   OBD_CONNECT_MULTIMODRPCS |
 223                                   OBD_CONNECT_GRANT_PARAM |
 224                                   OBD_CONNECT_SHORTIO | OBD_CONNECT_FLAGS2;
 225
 226         data->ocd_connect_flags2 = OBD_CONNECT2_FLR;
 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 #if defined(HAVE_SECURITY_DENTRY_INIT_SECURITY) && defined(CONFIG_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                         ktime_get_seconds() -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 | OBD_CONNECT_SHORTIO |
 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         /* Don't change value if it was specified in the config log */
 479         if (sbi->ll_ra_info.ra_max_read_ahead_whole_pages == -1)
 480                 sbi->ll_ra_info.ra_max_read_ahead_whole_pages =
 481                         max_t(unsigned long, SBI_DEFAULT_READAHEAD_WHOLE_MAX,
 482                               (data->ocd_brw_size >> PAGE_SHIFT));
 483
 484         err = obd_fid_init(sbi->ll_dt_exp->exp_obd, sbi->ll_dt_exp,
 485                            LUSTRE_SEQ_METADATA);
 486         if (err) {
 487                 CERROR("%s: Can't init data layer FID infrastructure, "
 488                        "rc = %d\n", sbi->ll_dt_exp->exp_obd->obd_name, err);
 489                 GOTO(out_dt, err);
 490         }
 491
 492         mutex_lock(&sbi->ll_lco.lco_lock);
 493         sbi->ll_lco.lco_flags = data->ocd_connect_flags;
 494         sbi->ll_lco.lco_md_exp = sbi->ll_md_exp;
 495         sbi->ll_lco.lco_dt_exp = sbi->ll_dt_exp;
 496         mutex_unlock(&sbi->ll_lco.lco_lock);
 497
 498         fid_zero(&sbi->ll_root_fid);
 499         err = md_get_root(sbi->ll_md_exp, get_mount_fileset(sb),
 500                            &sbi->ll_root_fid);
 501         if (err) {
 502                 CERROR("cannot mds_connect: rc = %d\n", err);
 503                 GOTO(out_lock_cn_cb, err);
 504         }
 505         if (!fid_is_sane(&sbi->ll_root_fid)) {
 506                 CERROR("%s: Invalid root fid "DFID" during mount\n",
 507                        sbi->ll_md_exp->exp_obd->obd_name,
 508                        PFID(&sbi->ll_root_fid));
 509                 GOTO(out_lock_cn_cb, err = -EINVAL);
 510         }
 511         CDEBUG(D_SUPER, "rootfid "DFID"\n", PFID(&sbi->ll_root_fid));
 512
 513         sb->s_op = &lustre_super_operations;
 514 #ifdef HAVE_XATTR_HANDLER_FLAGS
 515         sb->s_xattr = ll_xattr_handlers;
 516 #endif
 517 #if THREAD_SIZE >= 8192 /*b=17630*/
 518         sb->s_export_op = &lustre_export_operations;
 519 #endif
 520
 521         /* make root inode
 522          * XXX: move this to after cbd setup? */
 523         valid = OBD_MD_FLGETATTR | OBD_MD_FLBLOCKS | OBD_MD_FLMODEASIZE;
 524         if (sbi->ll_flags & LL_SBI_ACL)
 525                 valid |= OBD_MD_FLACL;
 526
 527         OBD_ALLOC_PTR(op_data);
 528         if (op_data == NULL)
 529                 GOTO(out_lock_cn_cb, err = -ENOMEM);
 530
 531         op_data->op_fid1 = sbi->ll_root_fid;
 532         op_data->op_mode = 0;
 533         op_data->op_valid = valid;
 534
 535         err = md_getattr(sbi->ll_md_exp, op_data, &request);
 536
 537         OBD_FREE_PTR(op_data);
 538         if (err) {
 539                 CERROR("%s: md_getattr failed for root: rc = %d\n",
 540                        sbi->ll_md_exp->exp_obd->obd_name, err);
 541                 GOTO(out_lock_cn_cb, err);
 542         }
 543
 544         err = md_get_lustre_md(sbi->ll_md_exp, request, sbi->ll_dt_exp,
 545                                sbi->ll_md_exp, &lmd);
 546         if (err) {
 547                 CERROR("failed to understand root inode md: rc = %d\n", err);
 548                 ptlrpc_req_finished(request);
 549                 GOTO(out_lock_cn_cb, err);
 550         }
 551
 552         LASSERT(fid_is_sane(&sbi->ll_root_fid));
 553         root = ll_iget(sb, cl_fid_build_ino(&sbi->ll_root_fid,
 554                                             sbi->ll_flags & LL_SBI_32BIT_API),
 555                        &lmd);
 556         md_free_lustre_md(sbi->ll_md_exp, &lmd);
 557         ptlrpc_req_finished(request);
 558
 559         if (IS_ERR(root)) {
 560 #ifdef CONFIG_FS_POSIX_ACL
 561                 if (lmd.posix_acl) {
 562                         posix_acl_release(lmd.posix_acl);
 563                         lmd.posix_acl = NULL;
 564                 }
 565 #endif
 566                 err = IS_ERR(root) ? PTR_ERR(root) : -EBADF;
 567                 root = NULL;
 568                 CERROR("lustre_lite: bad iget4 for root\n");
 569                 GOTO(out_root, err);
 570         }
 571
 572         checksum = sbi->ll_flags & LL_SBI_CHECKSUM;
 573         err = obd_set_info_async(NULL, sbi->ll_dt_exp, sizeof(KEY_CHECKSUM),
 574                                  KEY_CHECKSUM, sizeof(checksum), &checksum,
 575                                  NULL);
 576         if (err) {
 577                 CERROR("%s: Set checksum failed: rc = %d\n",
 578                        sbi->ll_dt_exp->exp_obd->obd_name, err);
 579                 GOTO(out_root, err);
 580         }
 581         cl_sb_init(sb);
 582
 583         err = obd_set_info_async(NULL, sbi->ll_dt_exp, sizeof(KEY_CACHE_SET),
 584                                  KEY_CACHE_SET, sizeof(*sbi->ll_cache),
 585                                  sbi->ll_cache, NULL);
 586         if (err) {
 587                 CERROR("%s: Set cache_set failed: rc = %d\n",
 588                        sbi->ll_dt_exp->exp_obd->obd_name, err);
 589                 GOTO(out_root, err);
 590         }
 591
 592         sb->s_root = d_make_root(root);
 593         if (sb->s_root == NULL) {
 594                 CERROR("%s: can't make root dentry\n",
 595                         ll_get_fsname(sb, NULL, 0));
 596                 GOTO(out_root, err = -ENOMEM);
 597         }
 598 #ifdef HAVE_DCACHE_LOCK
 599         sb->s_root->d_op = &ll_d_ops;
 600 #endif
 601
 602         sbi->ll_sdev_orig = sb->s_dev;
 603
 604         /* We set sb->s_dev equal on all lustre clients in order to support
 605          * NFS export clustering.  NFSD requires that the FSID be the same
 606          * on all clients. */
 607         /* s_dev is also used in lt_compare() to compare two fs, but that is
 608          * only a node-local comparison. */
 609         uuid = obd_get_uuid(sbi->ll_md_exp);
 610         if (uuid != NULL)
 611                 sb->s_dev = get_uuid2int(uuid->uuid, strlen(uuid->uuid));
 612
 613         if (data != NULL)
 614                 OBD_FREE_PTR(data);
 615         if (osfs != NULL)
 616                 OBD_FREE_PTR(osfs);
 617
 618         if (sbi->ll_proc_root != NULL) {
 619                 err = lprocfs_ll_register_obd(sb, dt);
 620                 if (err < 0) {
 621                         CERROR("%s: could not register %s in llite: rc = %d\n",
 622                                dt, ll_get_fsname(sb, NULL, 0), err);
 623                         err = 0;
 624                 }
 625                 err = lprocfs_ll_register_obd(sb, md);
 626                 if (err < 0) {
 627                         CERROR("%s: could not register %s in llite: rc = %d\n",
 628                                md, ll_get_fsname(sb, NULL, 0), err);
 629                         err = 0;
 630                 }
 631         }
 632
 633         RETURN(err);
 634 out_root:
 635         if (root)
 636                 iput(root);
 637 out_lock_cn_cb:
 638         obd_fid_fini(sbi->ll_dt_exp->exp_obd);
 639 out_dt:
 640         obd_disconnect(sbi->ll_dt_exp);
 641         sbi->ll_dt_exp = NULL;
 642 out_md_fid:
 643         obd_fid_fini(sbi->ll_md_exp->exp_obd);
 644 out_md:
 645         obd_disconnect(sbi->ll_md_exp);
 646         sbi->ll_md_exp = NULL;
 647 out:
 648         if (data != NULL)
 649                 OBD_FREE_PTR(data);
 650         if (osfs != NULL)
 651                 OBD_FREE_PTR(osfs);
 652         return err;
 653 }
 654
 655 int ll_get_max_mdsize(struct ll_sb_info *sbi, int *lmmsize)
 656 {
 657         int size, rc;
 658
 659         size = sizeof(*lmmsize);
 660         rc = obd_get_info(NULL, sbi->ll_dt_exp, sizeof(KEY_MAX_EASIZE),
 661                           KEY_MAX_EASIZE, &size, lmmsize);
 662         if (rc != 0) {
 663                 CERROR("%s: cannot get max LOV EA size: rc = %d\n",
 664                        sbi->ll_dt_exp->exp_obd->obd_name, rc);
 665                 RETURN(rc);
 666         }
 667
 668         size = sizeof(int);
 669         rc = obd_get_info(NULL, sbi->ll_md_exp, sizeof(KEY_MAX_EASIZE),
 670                           KEY_MAX_EASIZE, &size, lmmsize);
 671         if (rc)
 672                 CERROR("Get max mdsize error rc %d\n", rc);
 673
 674         RETURN(rc);
 675 }
 676
 677 /**
 678  * Get the value of the default_easize parameter.
 679  *
 680  * \see client_obd::cl_default_mds_easize
 681  *
 682  * \param[in] sbi       superblock info for this filesystem
 683  * \param[out] lmmsize  pointer to storage location for value
 684  *
 685  * \retval 0            on success
 686  * \retval negative     negated errno on failure
 687  */
 688 int ll_get_default_mdsize(struct ll_sb_info *sbi, int *lmmsize)
 689 {
 690         int size, rc;
 691
 692         size = sizeof(int);
 693         rc = obd_get_info(NULL, sbi->ll_md_exp, sizeof(KEY_DEFAULT_EASIZE),
 694                          KEY_DEFAULT_EASIZE, &size, lmmsize);
 695         if (rc)
 696                 CERROR("Get default mdsize error rc %d\n", rc);
 697
 698         RETURN(rc);
 699 }
 700
 701 /**
 702  * Set the default_easize parameter to the given value.
 703  *
 704  * \see client_obd::cl_default_mds_easize
 705  *
 706  * \param[in] sbi       superblock info for this filesystem
 707  * \param[in] lmmsize   the size to set
 708  *
 709  * \retval 0            on success
 710  * \retval negative     negated errno on failure
 711  */
 712 int ll_set_default_mdsize(struct ll_sb_info *sbi, int lmmsize)
 713 {
 714         int rc;
 715
 716         if (lmmsize < sizeof(struct lov_mds_md) ||
 717             lmmsize > OBD_MAX_DEFAULT_EA_SIZE)
 718                 return -EINVAL;
 719
 720         rc = obd_set_info_async(NULL, sbi->ll_md_exp,
 721                                 sizeof(KEY_DEFAULT_EASIZE), KEY_DEFAULT_EASIZE,
 722                                 sizeof(int), &lmmsize, NULL);
 723
 724         RETURN(rc);
 725 }
 726
 727 static void client_common_put_super(struct super_block *sb)
 728 {
 729         struct ll_sb_info *sbi = ll_s2sbi(sb);
 730         ENTRY;
 731
 732         cl_sb_fini(sb);
 733
 734         obd_fid_fini(sbi->ll_dt_exp->exp_obd);
 735         obd_disconnect(sbi->ll_dt_exp);
 736         sbi->ll_dt_exp = NULL;
 737
 738         lprocfs_ll_unregister_mountpoint(sbi);
 739
 740         obd_fid_fini(sbi->ll_md_exp->exp_obd);
 741         obd_disconnect(sbi->ll_md_exp);
 742         sbi->ll_md_exp = NULL;
 743
 744         EXIT;
 745 }
 746
 747 void ll_kill_super(struct super_block *sb)
 748 {
 749         struct ll_sb_info *sbi;
 750         ENTRY;
 751
 752         /* not init sb ?*/
 753         if (!(sb->s_flags & MS_ACTIVE))
 754                 return;
 755
 756         sbi = ll_s2sbi(sb);
 757         /* we need restore s_dev from changed for clustred NFS before put_super
 758          * because new kernels have cached s_dev and change sb->s_dev in
 759          * put_super not affected real removing devices */
 760         if (sbi) {
 761                 sb->s_dev = sbi->ll_sdev_orig;
 762                 sbi->ll_umounting = 1;
 763
 764                 /* wait running statahead threads to quit */
 765                 while (atomic_read(&sbi->ll_sa_running) > 0) {
 766                         set_current_state(TASK_UNINTERRUPTIBLE);
 767                         schedule_timeout(msecs_to_jiffies(MSEC_PER_SEC >> 3));
 768                 }
 769         }
 770
 771         EXIT;
 772 }
 773
 774 static inline int ll_set_opt(const char *opt, char *data, int fl)
 775 {
 776         if (strncmp(opt, data, strlen(opt)) != 0)
 777                 return(0);
 778         else
 779                 return(fl);
 780 }
 781
 782 /* non-client-specific mount options are parsed in lmd_parse */
 783 static int ll_options(char *options, int *flags)
 784 {
 785         int tmp;
 786         char *s1 = options, *s2;
 787         ENTRY;
 788
 789         if (!options)
 790                 RETURN(0);
 791
 792         CDEBUG(D_CONFIG, "Parsing opts %s\n", options);
 793
 794         while (*s1) {
 795                 CDEBUG(D_SUPER, "next opt=%s\n", s1);
 796                 tmp = ll_set_opt("nolock", s1, LL_SBI_NOLCK);
 797                 if (tmp) {
 798                         *flags |= tmp;
 799                         goto next;
 800                 }
 801                 tmp = ll_set_opt("flock", s1, LL_SBI_FLOCK);
 802                 if (tmp) {
 803                         *flags |= tmp;
 804                         goto next;
 805                 }
 806                 tmp = ll_set_opt("localflock", s1, LL_SBI_LOCALFLOCK);
 807                 if (tmp) {
 808                         *flags |= tmp;
 809                         goto next;
 810                 }
 811                 tmp = ll_set_opt("noflock", s1, LL_SBI_FLOCK|LL_SBI_LOCALFLOCK);
 812                 if (tmp) {
 813                         *flags &= ~tmp;
 814                         goto next;
 815                 }
 816                 tmp = ll_set_opt("user_xattr", s1, LL_SBI_USER_XATTR);
 817                 if (tmp) {
 818                         *flags |= tmp;
 819                         goto next;
 820                 }
 821                 tmp = ll_set_opt("nouser_xattr", s1, LL_SBI_USER_XATTR);
 822                 if (tmp) {
 823                         *flags &= ~tmp;
 824                         goto next;
 825                 }
 826                 tmp = ll_set_opt("context", s1, 1);
 827                 if (tmp)
 828                         goto next;
 829                 tmp = ll_set_opt("fscontext", s1, 1);
 830                 if (tmp)
 831                         goto next;
 832                 tmp = ll_set_opt("defcontext", s1, 1);
 833                 if (tmp)
 834                         goto next;
 835                 tmp = ll_set_opt("rootcontext", s1, 1);
 836                 if (tmp)
 837                         goto next;
 838                 tmp = ll_set_opt("user_fid2path", s1, LL_SBI_USER_FID2PATH);
 839                 if (tmp) {
 840                         *flags |= tmp;
 841                         goto next;
 842                 }
 843                 tmp = ll_set_opt("nouser_fid2path", s1, LL_SBI_USER_FID2PATH);
 844                 if (tmp) {
 845                         *flags &= ~tmp;
 846                         goto next;
 847                 }
 848
 849                 tmp = ll_set_opt("checksum", s1, LL_SBI_CHECKSUM);
 850                 if (tmp) {
 851                         *flags |= tmp;
 852                         goto next;
 853                 }
 854                 tmp = ll_set_opt("nochecksum", s1, LL_SBI_CHECKSUM);
 855                 if (tmp) {
 856                         *flags &= ~tmp;
 857                         goto next;
 858                 }
 859                 tmp = ll_set_opt("lruresize", s1, LL_SBI_LRU_RESIZE);
 860                 if (tmp) {
 861                         *flags |= tmp;
 862                         goto next;
 863                 }
 864                 tmp = ll_set_opt("nolruresize", s1, LL_SBI_LRU_RESIZE);
 865                 if (tmp) {
 866                         *flags &= ~tmp;
 867                         goto next;
 868                 }
 869                 tmp = ll_set_opt("lazystatfs", s1, LL_SBI_LAZYSTATFS);
 870                 if (tmp) {
 871                         *flags |= tmp;
 872                         goto next;
 873                 }
 874                 tmp = ll_set_opt("nolazystatfs", s1, LL_SBI_LAZYSTATFS);
 875                 if (tmp) {
 876                         *flags &= ~tmp;
 877                         goto next;
 878                 }
 879                 tmp = ll_set_opt("32bitapi", s1, LL_SBI_32BIT_API);
 880                 if (tmp) {
 881                         *flags |= tmp;
 882                         goto next;
 883                 }
 884                 tmp = ll_set_opt("verbose", s1, LL_SBI_VERBOSE);
 885                 if (tmp) {
 886                         *flags |= tmp;
 887                         goto next;
 888                 }
 889                 tmp = ll_set_opt("noverbose", s1, LL_SBI_VERBOSE);
 890                 if (tmp) {
 891                         *flags &= ~tmp;
 892                         goto next;
 893                 }
 894                 tmp = ll_set_opt("always_ping", s1, LL_SBI_ALWAYS_PING);
 895                 if (tmp) {
 896                         *flags |= tmp;
 897                         goto next;
 898                 }
 899                 LCONSOLE_ERROR_MSG(0x152, "Unknown option '%s', won't mount.\n",
 900                                    s1);
 901                 RETURN(-EINVAL);
 902
 903 next:
 904                 /* Find next opt */
 905                 s2 = strchr(s1, ',');
 906                 if (s2 == NULL)
 907                         break;
 908                 s1 = s2 + 1;
 909         }
 910         RETURN(0);
 911 }
 912
 913 void ll_lli_init(struct ll_inode_info *lli)
 914 {
 915         lli->lli_inode_magic = LLI_INODE_MAGIC;
 916         lli->lli_flags = 0;
 917         spin_lock_init(&lli->lli_lock);
 918         lli->lli_posix_acl = NULL;
 919         /* Do not set lli_fid, it has been initialized already. */
 920         fid_zero(&lli->lli_pfid);
 921         lli->lli_mds_read_och = NULL;
 922         lli->lli_mds_write_och = NULL;
 923         lli->lli_mds_exec_och = NULL;
 924         lli->lli_open_fd_read_count = 0;
 925         lli->lli_open_fd_write_count = 0;
 926         lli->lli_open_fd_exec_count = 0;
 927         mutex_init(&lli->lli_och_mutex);
 928         spin_lock_init(&lli->lli_agl_lock);
 929         spin_lock_init(&lli->lli_layout_lock);
 930         ll_layout_version_set(lli, CL_LAYOUT_GEN_NONE);
 931         lli->lli_clob = NULL;
 932
 933         init_rwsem(&lli->lli_xattrs_list_rwsem);
 934         mutex_init(&lli->lli_xattrs_enq_lock);
 935
 936         LASSERT(lli->lli_vfs_inode.i_mode != 0);
 937         if (S_ISDIR(lli->lli_vfs_inode.i_mode)) {
 938                 mutex_init(&lli->lli_readdir_mutex);
 939                 lli->lli_opendir_key = NULL;
 940                 lli->lli_sai = NULL;
 941                 spin_lock_init(&lli->lli_sa_lock);
 942                 lli->lli_opendir_pid = 0;
 943                 lli->lli_sa_enabled = 0;
 944                 lli->lli_def_stripe_offset = -1;
 945         } else {
 946                 mutex_init(&lli->lli_size_mutex);
 947                 lli->lli_symlink_name = NULL;
 948                 init_rwsem(&lli->lli_trunc_sem);
 949                 range_lock_tree_init(&lli->lli_write_tree);
 950                 init_rwsem(&lli->lli_glimpse_sem);
 951                 lli->lli_glimpse_time = ktime_set(0, 0);
 952                 INIT_LIST_HEAD(&lli->lli_agl_list);
 953                 lli->lli_agl_index = 0;
 954                 lli->lli_async_rc = 0;
 955         }
 956         mutex_init(&lli->lli_layout_mutex);
 957         memset(lli->lli_jobid, 0, LUSTRE_JOBID_SIZE);
 958 }
 959
 960 #ifndef HAVE_SUPER_SETUP_BDI_NAME
 961
 962 #define LSI_BDI_INITIALIZED     0x00400000
 963
 964 #ifndef HAVE_BDI_CAP_MAP_COPY
 965 # define BDI_CAP_MAP_COPY       0
 966 #endif
 967
 968 #define MAX_STRING_SIZE 128
 969
 970 static int super_setup_bdi_name(struct super_block *sb, char *fmt, ...)
 971 {
 972         struct  lustre_sb_info *lsi = s2lsi(sb);
 973         char buf[MAX_STRING_SIZE];
 974         va_list args;
 975         int err;
 976
 977         err = bdi_init(&lsi->lsi_bdi);
 978         if (err)
 979                 return err;
 980
 981         lsi->lsi_flags |= LSI_BDI_INITIALIZED;
 982         lsi->lsi_bdi.capabilities = BDI_CAP_MAP_COPY;
 983         lsi->lsi_bdi.name = "lustre";
 984         va_start(args, fmt);
 985         vsnprintf(buf, MAX_STRING_SIZE, fmt, args);
 986         va_end(args);
 987         err = bdi_register(&lsi->lsi_bdi, NULL, "%s", buf);
 988         va_end(args);
 989         if (!err)
 990                 sb->s_bdi = &lsi->lsi_bdi;
 991
 992         return err;
 993 }
 994 #endif /* !HAVE_SUPER_SETUP_BDI_NAME */
 995
 996 int ll_fill_super(struct super_block *sb, struct vfsmount *mnt)
 997 {
 998         struct  lustre_profile *lprof = NULL;
 999         struct  lustre_sb_info *lsi = s2lsi(sb);
1000         struct  ll_sb_info *sbi = NULL;
1001         char    *dt = NULL, *md = NULL;
1002         char    *profilenm = get_profile_name(sb);
1003         struct config_llog_instance *cfg;
1004         /* %p for void* in printf needs 16+2 characters: 0xffffffffffffffff */
1005         const int instlen = sizeof(cfg->cfg_instance) * 2 + 2;
1006         int     md_len = 0;
1007         int     dt_len = 0;
1008         int     err;
1009         ENTRY;
1010
1011         CDEBUG(D_VFSTRACE, "VFS Op: sb %p\n", sb);
1012
1013         try_module_get(THIS_MODULE);
1014
1015         OBD_ALLOC_PTR(cfg);
1016         if (cfg == NULL)
1017                 GOTO(out_free, err = -ENOMEM);
1018
1019         /* client additional sb info */
1020         lsi->lsi_llsbi = sbi = ll_init_sbi();
1021         if (!sbi)
1022                 GOTO(out_free, err = -ENOMEM);
1023
1024         err = ll_options(lsi->lsi_lmd->lmd_opts, &sbi->ll_flags);
1025         if (err)
1026                 GOTO(out_free, err);
1027
1028         err = super_setup_bdi_name(sb, "lustre-%p", sb);
1029         if (err)
1030                 GOTO(out_free, err);
1031
1032 #ifndef HAVE_DCACHE_LOCK
1033         /* kernel >= 2.6.38 store dentry operations in sb->s_d_op. */
1034         sb->s_d_op = &ll_d_ops;
1035 #endif
1036
1037         /* Call lprocfs_ll_register_mountpoint() before lustre_process_log()
1038          * so that "llite.*.*" params can be processed correctly. */
1039         if (proc_lustre_fs_root != NULL) {
1040                 err = lprocfs_ll_register_mountpoint(proc_lustre_fs_root, sb);
1041                 if (err < 0) {
1042                         CERROR("%s: could not register mountpoint in llite: "
1043                                "rc = %d\n", ll_get_fsname(sb, NULL, 0), err);
1044                         err = 0;
1045                 }
1046         }
1047
1048         /* Generate a string unique to this super, in case some joker tries
1049            to mount the same fs at two mount points.
1050            Use the address of the super itself.*/
1051         cfg->cfg_instance = sb;
1052         cfg->cfg_uuid = lsi->lsi_llsbi->ll_sb_uuid;
1053         cfg->cfg_callback = class_config_llog_handler;
1054         cfg->cfg_sub_clds = CONFIG_SUB_CLIENT;
1055         /* set up client obds */
1056         err = lustre_process_log(sb, profilenm, cfg);
1057         if (err < 0)
1058                 GOTO(out_proc, err);
1059
1060         /* Profile set with LCFG_MOUNTOPT so we can find our mdc and osc obds */
1061         lprof = class_get_profile(profilenm);
1062         if (lprof == NULL) {
1063                 LCONSOLE_ERROR_MSG(0x156, "The client profile '%s' could not be"
1064                                    " read from the MGS.  Does that filesystem "
1065                                    "exist?\n", profilenm);
1066                 GOTO(out_proc, err = -EINVAL);
1067         }
1068         CDEBUG(D_CONFIG, "Found profile %s: mdc=%s osc=%s\n", profilenm,
1069                lprof->lp_md, lprof->lp_dt);
1070
1071         dt_len = strlen(lprof->lp_dt) + instlen + 2;
1072         OBD_ALLOC(dt, dt_len);
1073         if (!dt)
1074                 GOTO(out_proc, err = -ENOMEM);
1075         snprintf(dt, dt_len - 1, "%s-%p", lprof->lp_dt, cfg->cfg_instance);
1076
1077         md_len = strlen(lprof->lp_md) + instlen + 2;
1078         OBD_ALLOC(md, md_len);
1079         if (!md)
1080                 GOTO(out_proc, err = -ENOMEM);
1081         snprintf(md, md_len - 1, "%s-%p", lprof->lp_md, cfg->cfg_instance);
1082
1083         /* connections, registrations, sb setup */
1084         err = client_common_fill_super(sb, md, dt, mnt);
1085         if (err < 0)
1086                 GOTO(out_proc, err);
1087
1088         sbi->ll_client_common_fill_super_succeeded = 1;
1089
1090 out_proc:
1091         if (err < 0)
1092                 lprocfs_ll_unregister_mountpoint(sbi);
1093 out_free:
1094         if (md)
1095                 OBD_FREE(md, md_len);
1096         if (dt)
1097                 OBD_FREE(dt, dt_len);
1098         if (lprof != NULL)
1099                 class_put_profile(lprof);
1100         if (cfg)
1101                 OBD_FREE_PTR(cfg);
1102         if (err)
1103                 ll_put_super(sb);
1104         else if (sbi->ll_flags & LL_SBI_VERBOSE)
1105                 LCONSOLE_WARN("Mounted %s\n", profilenm);
1106         RETURN(err);
1107 } /* ll_fill_super */
1108
1109 void ll_put_super(struct super_block *sb)
1110 {
1111         struct config_llog_instance cfg, params_cfg;
1112         struct obd_device *obd;
1113         struct lustre_sb_info *lsi = s2lsi(sb);
1114         struct ll_sb_info *sbi = ll_s2sbi(sb);
1115         char *profilenm = get_profile_name(sb);
1116         long ccc_count;
1117         int next, force = 1, rc = 0;
1118         ENTRY;
1119
1120         if (!sbi)
1121                 GOTO(out_no_sbi, 0);
1122
1123         CDEBUG(D_VFSTRACE, "VFS Op: sb %p - %s\n", sb, profilenm);
1124
1125         cfg.cfg_instance = sb;
1126         lustre_end_log(sb, profilenm, &cfg);
1127
1128         params_cfg.cfg_instance = sb;
1129         lustre_end_log(sb, PARAMS_FILENAME, &params_cfg);
1130
1131         if (sbi->ll_md_exp) {
1132                 obd = class_exp2obd(sbi->ll_md_exp);
1133                 if (obd)
1134                         force = obd->obd_force;
1135         }
1136
1137         /* Wait for unstable pages to be committed to stable storage */
1138         if (force == 0) {
1139                 struct l_wait_info lwi = LWI_INTR(LWI_ON_SIGNAL_NOOP, NULL);
1140                 rc = l_wait_event(sbi->ll_cache->ccc_unstable_waitq,
1141                         atomic_long_read(&sbi->ll_cache->ccc_unstable_nr) == 0,
1142                         &lwi);
1143         }
1144
1145         ccc_count = atomic_long_read(&sbi->ll_cache->ccc_unstable_nr);
1146         if (force == 0 && rc != -EINTR)
1147                 LASSERTF(ccc_count == 0, "count: %li\n", ccc_count);
1148
1149
1150         /* We need to set force before the lov_disconnect in
1151            lustre_common_put_super, since l_d cleans up osc's as well. */
1152         if (force) {
1153                 next = 0;
1154                 while ((obd = class_devices_in_group(&sbi->ll_sb_uuid,
1155                                                      &next)) != NULL) {
1156                         obd->obd_force = force;
1157                 }
1158         }
1159
1160         if (sbi->ll_client_common_fill_super_succeeded) {
1161                 /* Only if client_common_fill_super succeeded */
1162                 client_common_put_super(sb);
1163         }
1164
1165         next = 0;
1166         while ((obd = class_devices_in_group(&sbi->ll_sb_uuid, &next)) !=NULL) {
1167                 class_manual_cleanup(obd);
1168         }
1169
1170         if (sbi->ll_flags & LL_SBI_VERBOSE)
1171                 LCONSOLE_WARN("Unmounted %s\n", profilenm ? profilenm : "");
1172
1173         if (profilenm)
1174                 class_del_profile(profilenm);
1175
1176 #ifndef HAVE_SUPER_SETUP_BDI_NAME
1177         if (lsi->lsi_flags & LSI_BDI_INITIALIZED) {
1178                 bdi_destroy(&lsi->lsi_bdi);
1179                 lsi->lsi_flags &= ~LSI_BDI_INITIALIZED;
1180         }
1181 #endif
1182
1183         ll_free_sbi(sb);
1184         lsi->lsi_llsbi = NULL;
1185 out_no_sbi:
1186         lustre_common_put_super(sb);
1187
1188         cl_env_cache_purge(~0);
1189
1190         module_put(THIS_MODULE);
1191
1192         EXIT;
1193 } /* client_put_super */
1194
1195 struct inode *ll_inode_from_resource_lock(struct ldlm_lock *lock)
1196 {
1197         struct inode *inode = NULL;
1198
1199         /* NOTE: we depend on atomic igrab() -bzzz */
1200         lock_res_and_lock(lock);
1201         if (lock->l_resource->lr_lvb_inode) {
1202                 struct ll_inode_info * lli;
1203                 lli = ll_i2info(lock->l_resource->lr_lvb_inode);
1204                 if (lli->lli_inode_magic == LLI_INODE_MAGIC) {
1205                         inode = igrab(lock->l_resource->lr_lvb_inode);
1206                 } else {
1207                         inode = lock->l_resource->lr_lvb_inode;
1208                         LDLM_DEBUG_LIMIT(inode->i_state & I_FREEING ?  D_INFO :
1209                                          D_WARNING, lock, "lr_lvb_inode %p is "
1210                                          "bogus: magic %08x",
1211                                          lock->l_resource->lr_lvb_inode,
1212                                          lli->lli_inode_magic);
1213                         inode = NULL;
1214                 }
1215         }
1216         unlock_res_and_lock(lock);
1217         return inode;
1218 }
1219
1220 void ll_dir_clear_lsm_md(struct inode *inode)
1221 {
1222         struct ll_inode_info *lli = ll_i2info(inode);
1223
1224         LASSERT(S_ISDIR(inode->i_mode));
1225
1226         if (lli->lli_lsm_md != NULL) {
1227                 lmv_free_memmd(lli->lli_lsm_md);
1228                 lli->lli_lsm_md = NULL;
1229         }
1230 }
1231
1232 static struct inode *ll_iget_anon_dir(struct super_block *sb,
1233                                       const struct lu_fid *fid,
1234                                       struct lustre_md *md)
1235 {
1236         struct ll_sb_info       *sbi = ll_s2sbi(sb);
1237         struct mdt_body         *body = md->body;
1238         struct inode            *inode;
1239         ino_t                   ino;
1240         ENTRY;
1241
1242         ino = cl_fid_build_ino(fid, sbi->ll_flags & LL_SBI_32BIT_API);
1243         inode = iget_locked(sb, ino);
1244         if (inode == NULL) {
1245                 CERROR("%s: failed get simple inode "DFID": rc = -ENOENT\n",
1246                        ll_get_fsname(sb, NULL, 0), PFID(fid));
1247                 RETURN(ERR_PTR(-ENOENT));
1248         }
1249
1250         if (inode->i_state & I_NEW) {
1251                 struct ll_inode_info *lli = ll_i2info(inode);
1252                 struct lmv_stripe_md *lsm = md->lmv;
1253
1254                 inode->i_mode = (inode->i_mode & ~S_IFMT) |
1255                                 (body->mbo_mode & S_IFMT);
1256                 LASSERTF(S_ISDIR(inode->i_mode), "Not slave inode "DFID"\n",
1257                          PFID(fid));
1258
1259                 LTIME_S(inode->i_mtime) = 0;
1260                 LTIME_S(inode->i_atime) = 0;
1261                 LTIME_S(inode->i_ctime) = 0;
1262                 inode->i_rdev = 0;
1263
1264 #ifdef HAVE_BACKING_DEV_INFO
1265                 /* initializing backing dev info. */
1266                 inode->i_mapping->backing_dev_info =
1267                                                 &s2lsi(inode->i_sb)->lsi_bdi;
1268 #endif
1269                 inode->i_op = &ll_dir_inode_operations;
1270                 inode->i_fop = &ll_dir_operations;
1271                 lli->lli_fid = *fid;
1272                 ll_lli_init(lli);
1273
1274                 LASSERT(lsm != NULL);
1275                 /* master object FID */
1276                 lli->lli_pfid = body->mbo_fid1;
1277                 CDEBUG(D_INODE, "lli %p slave "DFID" master "DFID"\n",
1278                        lli, PFID(fid), PFID(&lli->lli_pfid));
1279                 unlock_new_inode(inode);
1280         }
1281
1282         RETURN(inode);
1283 }
1284
1285 static int ll_init_lsm_md(struct inode *inode, struct lustre_md *md)
1286 {
1287         struct lu_fid *fid;
1288         struct lmv_stripe_md *lsm = md->lmv;
1289         int i;
1290
1291         LASSERT(lsm != NULL);
1292         /* XXX sigh, this lsm_root initialization should be in
1293          * LMV layer, but it needs ll_iget right now, so we
1294          * put this here right now. */
1295         for (i = 0; i < lsm->lsm_md_stripe_count; i++) {
1296                 fid = &lsm->lsm_md_oinfo[i].lmo_fid;
1297                 LASSERT(lsm->lsm_md_oinfo[i].lmo_root == NULL);
1298                 /* Unfortunately ll_iget will call ll_update_inode,
1299                  * where the initialization of slave inode is slightly
1300                  * different, so it reset lsm_md to NULL to avoid
1301                  * initializing lsm for slave inode. */
1302                 /* For migrating inode, master stripe and master object will
1303                  * be same, so we only need assign this inode */
1304                 if (lsm->lsm_md_hash_type & LMV_HASH_FLAG_MIGRATION && i == 0)
1305                         lsm->lsm_md_oinfo[i].lmo_root = inode;
1306                 else
1307                         lsm->lsm_md_oinfo[i].lmo_root =
1308                                 ll_iget_anon_dir(inode->i_sb, fid, md);
1309
1310                 if (IS_ERR(lsm->lsm_md_oinfo[i].lmo_root)) {
1311                         int rc = PTR_ERR(lsm->lsm_md_oinfo[i].lmo_root);
1312
1313                         lsm->lsm_md_oinfo[i].lmo_root = NULL;
1314                         return rc;
1315                 }
1316         }
1317
1318         return 0;
1319 }
1320
1321 static inline int lli_lsm_md_eq(const struct lmv_stripe_md *lsm_md1,
1322                                 const struct lmv_stripe_md *lsm_md2)
1323 {
1324         return lsm_md1->lsm_md_magic == lsm_md2->lsm_md_magic &&
1325                lsm_md1->lsm_md_stripe_count == lsm_md2->lsm_md_stripe_count &&
1326                lsm_md1->lsm_md_master_mdt_index ==
1327                                         lsm_md2->lsm_md_master_mdt_index &&
1328                lsm_md1->lsm_md_hash_type == lsm_md2->lsm_md_hash_type &&
1329                lsm_md1->lsm_md_layout_version ==
1330                                         lsm_md2->lsm_md_layout_version &&
1331                strcmp(lsm_md1->lsm_md_pool_name,
1332                       lsm_md2->lsm_md_pool_name) == 0;
1333 }
1334
1335 static int ll_update_lsm_md(struct inode *inode, struct lustre_md *md)
1336 {
1337         struct ll_inode_info *lli = ll_i2info(inode);
1338         struct lmv_stripe_md *lsm = md->lmv;
1339         int     rc;
1340         ENTRY;
1341
1342         LASSERT(S_ISDIR(inode->i_mode));
1343         CDEBUG(D_INODE, "update lsm %p of "DFID"\n", lli->lli_lsm_md,
1344                PFID(ll_inode2fid(inode)));
1345
1346         /* no striped information from request. */
1347         if (lsm == NULL) {
1348                 if (lli->lli_lsm_md == NULL) {
1349                         RETURN(0);
1350                 } else if (lli->lli_lsm_md->lsm_md_hash_type &
1351                                                 LMV_HASH_FLAG_MIGRATION) {
1352                         /* migration is done, the temporay MIGRATE layout has
1353                          * been removed */
1354                         CDEBUG(D_INODE, DFID" finish migration.\n",
1355                                PFID(ll_inode2fid(inode)));
1356                         lmv_free_memmd(lli->lli_lsm_md);
1357                         lli->lli_lsm_md = NULL;
1358                         RETURN(0);
1359                 } else {
1360                         /* The lustre_md from req does not include stripeEA,
1361                          * see ll_md_setattr */
1362                         RETURN(0);
1363                 }
1364         }
1365
1366         /* set the directory layout */
1367         if (lli->lli_lsm_md == NULL) {
1368                 struct cl_attr  *attr;
1369
1370                 rc = ll_init_lsm_md(inode, md);
1371                 if (rc != 0)
1372                         RETURN(rc);
1373
1374                 /* set md->lmv to NULL, so the following free lustre_md
1375                  * will not free this lsm */
1376                 md->lmv = NULL;
1377                 lli->lli_lsm_md = lsm;
1378
1379                 OBD_ALLOC_PTR(attr);
1380                 if (attr == NULL)
1381                         RETURN(-ENOMEM);
1382
1383                 /* validate the lsm */
1384                 rc = md_merge_attr(ll_i2mdexp(inode), lsm, attr,
1385                                    ll_md_blocking_ast);
1386                 if (rc != 0) {
1387                         OBD_FREE_PTR(attr);
1388                         RETURN(rc);
1389                 }
1390
1391                 if (md->body->mbo_valid & OBD_MD_FLNLINK)
1392                         md->body->mbo_nlink = attr->cat_nlink;
1393                 if (md->body->mbo_valid & OBD_MD_FLSIZE)
1394                         md->body->mbo_size = attr->cat_size;
1395                 if (md->body->mbo_valid & OBD_MD_FLATIME)
1396                         md->body->mbo_atime = attr->cat_atime;
1397                 if (md->body->mbo_valid & OBD_MD_FLCTIME)
1398                         md->body->mbo_ctime = attr->cat_ctime;
1399                 if (md->body->mbo_valid & OBD_MD_FLMTIME)
1400                         md->body->mbo_mtime = attr->cat_mtime;
1401
1402                 OBD_FREE_PTR(attr);
1403
1404                 CDEBUG(D_INODE, "Set lsm %p magic %x to "DFID"\n", lsm,
1405                        lsm->lsm_md_magic, PFID(ll_inode2fid(inode)));
1406                 RETURN(0);
1407         }
1408
1409         /* Compare the old and new stripe information */
1410         if (!lsm_md_eq(lli->lli_lsm_md, lsm)) {
1411                 struct lmv_stripe_md    *old_lsm = lli->lli_lsm_md;
1412                 int                     idx;
1413
1414                 CERROR("%s: inode "DFID"(%p)'s lmv layout mismatch (%p)/(%p)"
1415                        "magic:0x%x/0x%x stripe count: %d/%d master_mdt: %d/%d"
1416                        "hash_type:0x%x/0x%x layout: 0x%x/0x%x pool:%s/%s\n",
1417                        ll_get_fsname(inode->i_sb, NULL, 0), PFID(&lli->lli_fid),
1418                        inode, lsm, old_lsm,
1419                        lsm->lsm_md_magic, old_lsm->lsm_md_magic,
1420                        lsm->lsm_md_stripe_count,
1421                        old_lsm->lsm_md_stripe_count,
1422                        lsm->lsm_md_master_mdt_index,
1423                        old_lsm->lsm_md_master_mdt_index,
1424                        lsm->lsm_md_hash_type, old_lsm->lsm_md_hash_type,
1425                        lsm->lsm_md_layout_version,
1426                        old_lsm->lsm_md_layout_version,
1427                        lsm->lsm_md_pool_name,
1428                        old_lsm->lsm_md_pool_name);
1429
1430                 for (idx = 0; idx < old_lsm->lsm_md_stripe_count; idx++) {
1431                         CERROR("%s: sub FIDs in old lsm idx %d, old: "DFID"\n",
1432                                ll_get_fsname(inode->i_sb, NULL, 0), idx,
1433                                PFID(&old_lsm->lsm_md_oinfo[idx].lmo_fid));
1434                 }
1435
1436                 for (idx = 0; idx < lsm->lsm_md_stripe_count; idx++) {
1437                         CERROR("%s: sub FIDs in new lsm idx %d, new: "DFID"\n",
1438                                ll_get_fsname(inode->i_sb, NULL, 0), idx,
1439                                PFID(&lsm->lsm_md_oinfo[idx].lmo_fid));
1440                 }
1441
1442                 RETURN(-EIO);
1443         }
1444
1445         RETURN(0);
1446 }
1447
1448 void ll_clear_inode(struct inode *inode)
1449 {
1450         struct ll_inode_info *lli = ll_i2info(inode);
1451         struct ll_sb_info *sbi = ll_i2sbi(inode);
1452         ENTRY;
1453
1454         CDEBUG(D_VFSTRACE, "VFS Op:inode="DFID"(%p)\n",
1455                PFID(ll_inode2fid(inode)), inode);
1456
1457         if (S_ISDIR(inode->i_mode)) {
1458                 /* these should have been cleared in ll_file_release */
1459                 LASSERT(lli->lli_opendir_key == NULL);
1460                 LASSERT(lli->lli_sai == NULL);
1461                 LASSERT(lli->lli_opendir_pid == 0);
1462         }
1463
1464         md_null_inode(sbi->ll_md_exp, ll_inode2fid(inode));
1465
1466         LASSERT(!lli->lli_open_fd_write_count);
1467         LASSERT(!lli->lli_open_fd_read_count);
1468         LASSERT(!lli->lli_open_fd_exec_count);
1469
1470         if (lli->lli_mds_write_och)
1471                 ll_md_real_close(inode, FMODE_WRITE);
1472         if (lli->lli_mds_exec_och)
1473                 ll_md_real_close(inode, FMODE_EXEC);
1474         if (lli->lli_mds_read_och)
1475                 ll_md_real_close(inode, FMODE_READ);
1476
1477         if (S_ISLNK(inode->i_mode) && lli->lli_symlink_name) {
1478                 OBD_FREE(lli->lli_symlink_name,
1479                          strlen(lli->lli_symlink_name) + 1);
1480                 lli->lli_symlink_name = NULL;
1481         }
1482
1483         ll_xattr_cache_destroy(inode);
1484
1485 #ifdef CONFIG_FS_POSIX_ACL
1486         forget_all_cached_acls(inode);
1487         if (lli->lli_posix_acl) {
1488                 LASSERT(atomic_read(&lli->lli_posix_acl->a_refcount) == 1);
1489                 posix_acl_release(lli->lli_posix_acl);
1490                 lli->lli_posix_acl = NULL;
1491         }
1492 #endif
1493         lli->lli_inode_magic = LLI_INODE_DEAD;
1494
1495         if (S_ISDIR(inode->i_mode))
1496                 ll_dir_clear_lsm_md(inode);
1497         else if (S_ISREG(inode->i_mode) && !is_bad_inode(inode))
1498                 LASSERT(list_empty(&lli->lli_agl_list));
1499
1500         /*
1501          * XXX This has to be done before lsm is freed below, because
1502          * cl_object still uses inode lsm.
1503          */
1504         cl_inode_fini(inode);
1505
1506         EXIT;
1507 }
1508
1509 static int ll_md_setattr(struct dentry *dentry, struct md_op_data *op_data)
1510 {
1511         struct lustre_md md;
1512         struct inode *inode = dentry->d_inode;
1513         struct ll_sb_info *sbi = ll_i2sbi(inode);
1514         struct ptlrpc_request *request = NULL;
1515         int rc, ia_valid;
1516         ENTRY;
1517
1518         op_data = ll_prep_md_op_data(op_data, inode, NULL, NULL, 0, 0,
1519                                      LUSTRE_OPC_ANY, NULL);
1520         if (IS_ERR(op_data))
1521                 RETURN(PTR_ERR(op_data));
1522
1523         rc = md_setattr(sbi->ll_md_exp, op_data, NULL, 0, &request);
1524         if (rc) {
1525                 ptlrpc_req_finished(request);
1526                 if (rc == -ENOENT) {
1527                         clear_nlink(inode);
1528                         /* Unlinked special device node? Or just a race?
1529                          * Pretend we done everything. */
1530                         if (!S_ISREG(inode->i_mode) &&
1531                             !S_ISDIR(inode->i_mode)) {
1532                                 ia_valid = op_data->op_attr.ia_valid;
1533                                 op_data->op_attr.ia_valid &= ~TIMES_SET_FLAGS;
1534                                 rc = simple_setattr(dentry, &op_data->op_attr);
1535                                 op_data->op_attr.ia_valid = ia_valid;
1536                         }
1537                 } else if (rc != -EPERM && rc != -EACCES && rc != -ETXTBSY) {
1538                         CERROR("md_setattr fails: rc = %d\n", rc);
1539                 }
1540                 RETURN(rc);
1541         }
1542
1543         rc = md_get_lustre_md(sbi->ll_md_exp, request, sbi->ll_dt_exp,
1544                               sbi->ll_md_exp, &md);
1545         if (rc) {
1546                 ptlrpc_req_finished(request);
1547                 RETURN(rc);
1548         }
1549
1550         ia_valid = op_data->op_attr.ia_valid;
1551         /* inode size will be in ll_setattr_ost, can't do it now since dirty
1552          * cache is not cleared yet. */
1553         op_data->op_attr.ia_valid &= ~(TIMES_SET_FLAGS | ATTR_SIZE);
1554         if (S_ISREG(inode->i_mode))
1555                 inode_lock(inode);
1556         rc = simple_setattr(dentry, &op_data->op_attr);
1557         if (S_ISREG(inode->i_mode))
1558                 inode_unlock(inode);
1559         op_data->op_attr.ia_valid = ia_valid;
1560
1561         rc = ll_update_inode(inode, &md);
1562         ptlrpc_req_finished(request);
1563
1564         RETURN(rc);
1565 }
1566
1567 /* If this inode has objects allocated to it (lsm != NULL), then the OST
1568  * object(s) determine the file size and mtime.  Otherwise, the MDS will
1569  * keep these values until such a time that objects are allocated for it.
1570  * We do the MDS operations first, as it is checking permissions for us.
1571  * We don't to the MDS RPC if there is nothing that we want to store there,
1572  * otherwise there is no harm in updating mtime/atime on the MDS if we are
1573  * going to do an RPC anyways.
1574  *
1575  * If we are doing a truncate, we will send the mtime and ctime updates
1576  * to the OST with the punch RPC, otherwise we do an explicit setattr RPC.
1577  * I don't believe it is possible to get e.g. ATTR_MTIME_SET and ATTR_SIZE
1578  * at the same time.
1579  *
1580  * In case of HSMimport, we only set attr on MDS.
1581  */
1582 int ll_setattr_raw(struct dentry *dentry, struct iattr *attr, bool hsm_import)
1583 {
1584         struct inode *inode = dentry->d_inode;
1585         struct ll_inode_info *lli = ll_i2info(inode);
1586         struct md_op_data *op_data = NULL;
1587         int rc = 0;
1588         ENTRY;
1589
1590         CDEBUG(D_VFSTRACE, "%s: setattr inode "DFID"(%p) from %llu to %llu, "
1591                "valid %x, hsm_import %d\n",
1592                ll_get_fsname(inode->i_sb, NULL, 0), PFID(&lli->lli_fid),
1593                inode, i_size_read(inode), attr->ia_size, attr->ia_valid,
1594                hsm_import);
1595
1596         if (attr->ia_valid & ATTR_SIZE) {
1597                 /* Check new size against VFS/VM file size limit and rlimit */
1598                 rc = inode_newsize_ok(inode, attr->ia_size);
1599                 if (rc)
1600                         RETURN(rc);
1601
1602                 /* The maximum Lustre file size is variable, based on the
1603                  * OST maximum object size and number of stripes.  This
1604                  * needs another check in addition to the VFS check above. */
1605                 if (attr->ia_size > ll_file_maxbytes(inode)) {
1606                         CDEBUG(D_INODE,"file "DFID" too large %llu > %llu\n",
1607                                PFID(&lli->lli_fid), attr->ia_size,
1608                                ll_file_maxbytes(inode));
1609                         RETURN(-EFBIG);
1610                 }
1611
1612                 attr->ia_valid |= ATTR_MTIME | ATTR_CTIME;
1613         }
1614
1615         /* POSIX: check before ATTR_*TIME_SET set (from inode_change_ok) */
1616         if (attr->ia_valid & TIMES_SET_FLAGS) {
1617                 if ((!uid_eq(current_fsuid(), inode->i_uid)) &&
1618                     !cfs_capable(CFS_CAP_FOWNER))
1619                         RETURN(-EPERM);
1620         }
1621
1622         /* We mark all of the fields "set" so MDS/OST does not re-set them */
1623         if (!(attr->ia_valid & ATTR_CTIME_SET) &&
1624             (attr->ia_valid & ATTR_CTIME)) {
1625                 attr->ia_ctime = current_time(inode);
1626                 attr->ia_valid |= ATTR_CTIME_SET;
1627         }
1628         if (!(attr->ia_valid & ATTR_ATIME_SET) &&
1629             (attr->ia_valid & ATTR_ATIME)) {
1630                 attr->ia_atime = current_time(inode);
1631                 attr->ia_valid |= ATTR_ATIME_SET;
1632         }
1633         if (!(attr->ia_valid & ATTR_MTIME_SET) &&
1634             (attr->ia_valid & ATTR_MTIME)) {
1635                 attr->ia_mtime = current_time(inode);
1636                 attr->ia_valid |= ATTR_MTIME_SET;
1637         }
1638
1639         if (attr->ia_valid & (ATTR_MTIME | ATTR_CTIME))
1640                 CDEBUG(D_INODE, "setting mtime %lu, ctime %lu, now = %llu\n",
1641                        LTIME_S(attr->ia_mtime), LTIME_S(attr->ia_ctime),
1642                        (s64)ktime_get_real_seconds());
1643
1644         if (S_ISREG(inode->i_mode)) {
1645                 if (attr->ia_valid & ATTR_SIZE)
1646                         inode_dio_write_done(inode);
1647                 inode_unlock(inode);
1648         }
1649
1650         /* We always do an MDS RPC, even if we're only changing the size;
1651          * only the MDS knows whether truncate() should fail with -ETXTBUSY */
1652
1653         OBD_ALLOC_PTR(op_data);
1654         if (op_data == NULL)
1655                 GOTO(out, rc = -ENOMEM);
1656
1657         if (!hsm_import && attr->ia_valid & ATTR_SIZE) {
1658                 /* If we are changing file size, file content is
1659                  * modified, flag it. */
1660                 attr->ia_valid |= MDS_OPEN_OWNEROVERRIDE;
1661                 op_data->op_bias |= MDS_DATA_MODIFIED;
1662                 ll_file_clear_flag(lli, LLIF_DATA_MODIFIED);
1663         }
1664
1665         op_data->op_attr = *attr;
1666
1667         rc = ll_md_setattr(dentry, op_data);
1668         if (rc)
1669                 GOTO(out, rc);
1670
1671         if (!S_ISREG(inode->i_mode) || hsm_import)
1672                 GOTO(out, rc = 0);
1673
1674         if (attr->ia_valid & (ATTR_SIZE |
1675                               ATTR_ATIME | ATTR_ATIME_SET |
1676                               ATTR_MTIME | ATTR_MTIME_SET |
1677                               ATTR_CTIME | ATTR_CTIME_SET)) {
1678                 /* For truncate and utimes sending attributes to OSTs, setting
1679                  * mtime/atime to the past will be performed under PW [0:EOF]
1680                  * extent lock (new_size:EOF for truncate).  It may seem
1681                  * excessive to send mtime/atime updates to OSTs when not
1682                  * setting times to past, but it is necessary due to possible
1683                  * time de-synchronization between MDT inode and OST objects */
1684                 rc = cl_setattr_ost(lli->lli_clob, attr, 0);
1685         }
1686
1687         /* If the file was restored, it needs to set dirty flag.
1688          *
1689          * We've already sent MDS_DATA_MODIFIED flag in
1690          * ll_md_setattr() for truncate. However, the MDT refuses to
1691          * set the HS_DIRTY flag on released files, so we have to set
1692          * it again if the file has been restored. Please check how
1693          * LLIF_DATA_MODIFIED is set in vvp_io_setattr_fini().
1694          *
1695          * Please notice that if the file is not released, the previous
1696          * MDS_DATA_MODIFIED has taken effect and usually
1697          * LLIF_DATA_MODIFIED is not set(see vvp_io_setattr_fini()).
1698          * This way we can save an RPC for common open + trunc
1699          * operation. */
1700         if (ll_file_test_and_clear_flag(lli, LLIF_DATA_MODIFIED)) {
1701                 struct hsm_state_set hss = {
1702                         .hss_valid = HSS_SETMASK,
1703                         .hss_setmask = HS_DIRTY,
1704                 };
1705                 int rc2;
1706
1707                 rc2 = ll_hsm_state_set(inode, &hss);
1708                 /* truncate and write can happen at the same time, so that
1709                  * the file can be set modified even though the file is not
1710                  * restored from released state, and ll_hsm_state_set() is
1711                  * not applicable for the file, and rc2 < 0 is normal in this
1712                  * case. */
1713                 if (rc2 < 0)
1714                         CDEBUG(D_INFO, DFID "HSM set dirty failed: rc2 = %d\n",
1715                                PFID(ll_inode2fid(inode)), rc2);
1716         }
1717
1718         EXIT;
1719 out:
1720         if (op_data != NULL)
1721                 ll_finish_md_op_data(op_data);
1722
1723         if (S_ISREG(inode->i_mode)) {
1724                 inode_lock(inode);
1725                 if ((attr->ia_valid & ATTR_SIZE) && !hsm_import)
1726                         inode_dio_wait(inode);
1727                 /* Once we've got the i_mutex, it's safe to set the S_NOSEC
1728                  * flag.  ll_update_inode (called from ll_md_setattr), clears
1729                  * inode flags, so there is a gap where S_NOSEC is not set.
1730                  * This can cause a writer to take the i_mutex unnecessarily,
1731                  * but this is safe to do and should be rare. */
1732                 inode_has_no_xattr(inode);
1733         }
1734
1735         ll_stats_ops_tally(ll_i2sbi(inode), (attr->ia_valid & ATTR_SIZE) ?
1736                         LPROC_LL_TRUNC : LPROC_LL_SETATTR, 1);
1737
1738         return rc;
1739 }
1740
1741 int ll_setattr(struct dentry *de, struct iattr *attr)
1742 {
1743         int mode = de->d_inode->i_mode;
1744
1745         if ((attr->ia_valid & (ATTR_CTIME|ATTR_SIZE|ATTR_MODE)) ==
1746                               (ATTR_CTIME|ATTR_SIZE|ATTR_MODE))
1747                 attr->ia_valid |= MDS_OPEN_OWNEROVERRIDE;
1748
1749         if (((attr->ia_valid & (ATTR_MODE|ATTR_FORCE|ATTR_SIZE)) ==
1750                                (ATTR_SIZE|ATTR_MODE)) &&
1751             (((mode & S_ISUID) && !(attr->ia_mode & S_ISUID)) ||
1752              (((mode & (S_ISGID|S_IXGRP)) == (S_ISGID|S_IXGRP)) &&
1753               !(attr->ia_mode & S_ISGID))))
1754                 attr->ia_valid |= ATTR_FORCE;
1755
1756         if ((attr->ia_valid & ATTR_MODE) &&
1757             (mode & S_ISUID) &&
1758             !(attr->ia_mode & S_ISUID) &&
1759             !(attr->ia_valid & ATTR_KILL_SUID))
1760                 attr->ia_valid |= ATTR_KILL_SUID;
1761
1762         if ((attr->ia_valid & ATTR_MODE) &&
1763             ((mode & (S_ISGID|S_IXGRP)) == (S_ISGID|S_IXGRP)) &&
1764             !(attr->ia_mode & S_ISGID) &&
1765             !(attr->ia_valid & ATTR_KILL_SGID))
1766                 attr->ia_valid |= ATTR_KILL_SGID;
1767
1768         /* avoid polluted from ATTR_TIMES_SET,
1769          * projid is not expected to be set here */
1770         attr->ia_valid &= ~MDS_ATTR_PROJID;
1771
1772         return ll_setattr_raw(de, attr, false);
1773 }
1774
1775 int ll_statfs_internal(struct super_block *sb, struct obd_statfs *osfs,
1776                        u32 flags)
1777 {
1778         struct ll_sb_info *sbi = ll_s2sbi(sb);
1779         struct obd_statfs obd_osfs;
1780         time64_t max_age;
1781         int rc;
1782
1783         ENTRY;
1784         max_age = ktime_get_seconds() - OBD_STATFS_CACHE_SECONDS;
1785
1786         rc = obd_statfs(NULL, sbi->ll_md_exp, osfs, max_age, flags);
1787         if (rc) {
1788                 CERROR("md_statfs fails: rc = %d\n", rc);
1789                 RETURN(rc);
1790         }
1791
1792         osfs->os_type = sb->s_magic;
1793
1794         CDEBUG(D_SUPER, "MDC blocks %llu/%llu objects %llu/%llu\n",
1795                osfs->os_bavail, osfs->os_blocks, osfs->os_ffree,osfs->os_files);
1796
1797         if (sbi->ll_flags & LL_SBI_LAZYSTATFS)
1798                 flags |= OBD_STATFS_NODELAY;
1799
1800         rc = obd_statfs(NULL, sbi->ll_dt_exp, &obd_osfs, max_age, flags);
1801         if (rc) {
1802                 CERROR("obd_statfs fails: rc = %d\n", rc);
1803                 RETURN(rc);
1804         }
1805
1806         CDEBUG(D_SUPER, "OSC blocks %llu/%llu objects %llu/%llu\n",
1807                obd_osfs.os_bavail, obd_osfs.os_blocks, obd_osfs.os_ffree,
1808                obd_osfs.os_files);
1809
1810         osfs->os_bsize = obd_osfs.os_bsize;
1811         osfs->os_blocks = obd_osfs.os_blocks;
1812         osfs->os_bfree = obd_osfs.os_bfree;
1813         osfs->os_bavail = obd_osfs.os_bavail;
1814
1815         /* If we don't have as many objects free on the OST as inodes
1816          * on the MDS, we reduce the total number of inodes to
1817          * compensate, so that the "inodes in use" number is correct.
1818          */
1819         if (obd_osfs.os_ffree < osfs->os_ffree) {
1820                 osfs->os_files = (osfs->os_files - osfs->os_ffree) +
1821                         obd_osfs.os_ffree;
1822                 osfs->os_ffree = obd_osfs.os_ffree;
1823         }
1824
1825         RETURN(rc);
1826 }
1827 int ll_statfs(struct dentry *de, struct kstatfs *sfs)
1828 {
1829         struct super_block *sb = de->d_sb;
1830         struct obd_statfs osfs;
1831         __u64 fsid = huge_encode_dev(sb->s_dev);
1832         int rc;
1833
1834         CDEBUG(D_VFSTRACE, "VFS Op: at %llu jiffies\n", get_jiffies_64());
1835         ll_stats_ops_tally(ll_s2sbi(sb), LPROC_LL_STAFS, 1);
1836
1837         /* Some amount of caching on the client is allowed */
1838         rc = ll_statfs_internal(sb, &osfs, 0);
1839         if (rc)
1840                 return rc;
1841
1842         statfs_unpack(sfs, &osfs);
1843
1844         /* We need to downshift for all 32-bit kernels, because we can't
1845          * tell if the kernel is being called via sys_statfs64() or not.
1846          * Stop before overflowing f_bsize - in which case it is better
1847          * to just risk EOVERFLOW if caller is using old sys_statfs(). */
1848         if (sizeof(long) < 8) {
1849                 while (osfs.os_blocks > ~0UL && sfs->f_bsize < 0x40000000) {
1850                         sfs->f_bsize <<= 1;
1851
1852                         osfs.os_blocks >>= 1;
1853                         osfs.os_bfree >>= 1;
1854                         osfs.os_bavail >>= 1;
1855                 }
1856         }
1857
1858         sfs->f_blocks = osfs.os_blocks;
1859         sfs->f_bfree = osfs.os_bfree;
1860         sfs->f_bavail = osfs.os_bavail;
1861         sfs->f_fsid.val[0] = (__u32)fsid;
1862         sfs->f_fsid.val[1] = (__u32)(fsid >> 32);
1863         return 0;
1864 }
1865
1866 void ll_inode_size_lock(struct inode *inode)
1867 {
1868         struct ll_inode_info *lli;
1869
1870         LASSERT(!S_ISDIR(inode->i_mode));
1871
1872         lli = ll_i2info(inode);
1873         mutex_lock(&lli->lli_size_mutex);
1874 }
1875
1876 void ll_inode_size_unlock(struct inode *inode)
1877 {
1878         struct ll_inode_info *lli;
1879
1880         lli = ll_i2info(inode);
1881         mutex_unlock(&lli->lli_size_mutex);
1882 }
1883
1884 int ll_update_inode(struct inode *inode, struct lustre_md *md)
1885 {
1886         struct ll_inode_info *lli = ll_i2info(inode);
1887         struct mdt_body *body = md->body;
1888         struct ll_sb_info *sbi = ll_i2sbi(inode);
1889         int rc = 0;
1890
1891         if (body->mbo_valid & OBD_MD_FLEASIZE) {
1892                 rc = cl_file_inode_init(inode, md);
1893                 if (rc)
1894                         return rc;
1895         }
1896
1897         if (S_ISDIR(inode->i_mode)) {
1898                 rc = ll_update_lsm_md(inode, md);
1899                 if (rc != 0)
1900                         return rc;
1901         }
1902
1903 #ifdef CONFIG_FS_POSIX_ACL
1904         if (body->mbo_valid & OBD_MD_FLACL) {
1905                 spin_lock(&lli->lli_lock);
1906                 if (lli->lli_posix_acl)
1907                         posix_acl_release(lli->lli_posix_acl);
1908                 lli->lli_posix_acl = md->posix_acl;
1909                 spin_unlock(&lli->lli_lock);
1910         }
1911 #endif
1912         inode->i_ino = cl_fid_build_ino(&body->mbo_fid1,
1913                                         sbi->ll_flags & LL_SBI_32BIT_API);
1914         inode->i_generation = cl_fid_build_gen(&body->mbo_fid1);
1915
1916         if (body->mbo_valid & OBD_MD_FLATIME) {
1917                 if (body->mbo_atime > LTIME_S(inode->i_atime))
1918                         LTIME_S(inode->i_atime) = body->mbo_atime;
1919                 lli->lli_atime = body->mbo_atime;
1920         }
1921
1922         if (body->mbo_valid & OBD_MD_FLMTIME) {
1923                 if (body->mbo_mtime > LTIME_S(inode->i_mtime)) {
1924                         CDEBUG(D_INODE, "setting ino %lu mtime from %lu "
1925                                "to %llu\n", inode->i_ino,
1926                                LTIME_S(inode->i_mtime), body->mbo_mtime);
1927                         LTIME_S(inode->i_mtime) = body->mbo_mtime;
1928                 }
1929                 lli->lli_mtime = body->mbo_mtime;
1930         }
1931
1932         if (body->mbo_valid & OBD_MD_FLCTIME) {
1933                 if (body->mbo_ctime > LTIME_S(inode->i_ctime))
1934                         LTIME_S(inode->i_ctime) = body->mbo_ctime;
1935                 lli->lli_ctime = body->mbo_ctime;
1936         }
1937
1938         /* Clear i_flags to remove S_NOSEC before permissions are updated */
1939         if (body->mbo_valid & OBD_MD_FLFLAGS)
1940                 inode->i_flags = ll_ext_to_inode_flags(body->mbo_flags);
1941         if (body->mbo_valid & OBD_MD_FLMODE)
1942                 inode->i_mode = (inode->i_mode & S_IFMT) |
1943                                 (body->mbo_mode & ~S_IFMT);
1944
1945         if (body->mbo_valid & OBD_MD_FLTYPE)
1946                 inode->i_mode = (inode->i_mode & ~S_IFMT) |
1947                                 (body->mbo_mode & S_IFMT);
1948
1949         LASSERT(inode->i_mode != 0);
1950         if (S_ISREG(inode->i_mode))
1951                 inode->i_blkbits = min(PTLRPC_MAX_BRW_BITS + 1,
1952                                        LL_MAX_BLKSIZE_BITS);
1953         else
1954                 inode->i_blkbits = inode->i_sb->s_blocksize_bits;
1955
1956         if (body->mbo_valid & OBD_MD_FLUID)
1957                 inode->i_uid = make_kuid(&init_user_ns, body->mbo_uid);
1958         if (body->mbo_valid & OBD_MD_FLGID)
1959                 inode->i_gid = make_kgid(&init_user_ns, body->mbo_gid);
1960         if (body->mbo_valid & OBD_MD_FLPROJID)
1961                 lli->lli_projid = body->mbo_projid;
1962         if (body->mbo_valid & OBD_MD_FLNLINK)
1963                 set_nlink(inode, body->mbo_nlink);
1964         if (body->mbo_valid & OBD_MD_FLRDEV)
1965                 inode->i_rdev = old_decode_dev(body->mbo_rdev);
1966
1967         if (body->mbo_valid & OBD_MD_FLID) {
1968                 /* FID shouldn't be changed! */
1969                 if (fid_is_sane(&lli->lli_fid)) {
1970                         LASSERTF(lu_fid_eq(&lli->lli_fid, &body->mbo_fid1),
1971                                  "Trying to change FID "DFID
1972                                  " to the "DFID", inode "DFID"(%p)\n",
1973                                  PFID(&lli->lli_fid), PFID(&body->mbo_fid1),
1974                                  PFID(ll_inode2fid(inode)), inode);
1975                 } else {
1976                         lli->lli_fid = body->mbo_fid1;
1977                 }
1978         }
1979
1980         LASSERT(fid_seq(&lli->lli_fid) != 0);
1981
1982         if (body->mbo_valid & OBD_MD_FLSIZE) {
1983                 i_size_write(inode, body->mbo_size);
1984
1985                 CDEBUG(D_VFSTRACE, "inode="DFID", updating i_size %llu\n",
1986                        PFID(ll_inode2fid(inode)),
1987                        (unsigned long long)body->mbo_size);
1988
1989                 if (body->mbo_valid & OBD_MD_FLBLOCKS)
1990                         inode->i_blocks = body->mbo_blocks;
1991         }
1992
1993         if (body->mbo_valid & OBD_MD_TSTATE) {
1994                 /* Set LLIF_FILE_RESTORING if restore ongoing and
1995                  * clear it when done to ensure to start again
1996                  * glimpsing updated attrs
1997                  */
1998                 if (body->mbo_t_state & MS_RESTORE)
1999                         ll_file_set_flag(lli, LLIF_FILE_RESTORING);
2000                 else
2001                         ll_file_clear_flag(lli, LLIF_FILE_RESTORING);
2002         }
2003
2004         return 0;
2005 }
2006
2007 int ll_read_inode2(struct inode *inode, void *opaque)
2008 {
2009         struct lustre_md *md = opaque;
2010         struct ll_inode_info *lli = ll_i2info(inode);
2011         int     rc;
2012         ENTRY;
2013
2014         CDEBUG(D_VFSTRACE, "VFS Op:inode="DFID"(%p)\n",
2015                PFID(&lli->lli_fid), inode);
2016
2017         /* Core attributes from the MDS first.  This is a new inode, and
2018          * the VFS doesn't zero times in the core inode so we have to do
2019          * it ourselves.  They will be overwritten by either MDS or OST
2020          * attributes - we just need to make sure they aren't newer. */
2021         LTIME_S(inode->i_mtime) = 0;
2022         LTIME_S(inode->i_atime) = 0;
2023         LTIME_S(inode->i_ctime) = 0;
2024         inode->i_rdev = 0;
2025         rc = ll_update_inode(inode, md);
2026         if (rc != 0)
2027                 RETURN(rc);
2028
2029         /* OIDEBUG(inode); */
2030
2031 #ifdef HAVE_BACKING_DEV_INFO
2032         /* initializing backing dev info. */
2033         inode->i_mapping->backing_dev_info = &s2lsi(inode->i_sb)->lsi_bdi;
2034 #endif
2035         if (S_ISREG(inode->i_mode)) {
2036                 struct ll_sb_info *sbi = ll_i2sbi(inode);
2037                 inode->i_op = &ll_file_inode_operations;
2038                 inode->i_fop = sbi->ll_fop;
2039                 inode->i_mapping->a_ops = (struct address_space_operations *)&ll_aops;
2040                 EXIT;
2041         } else if (S_ISDIR(inode->i_mode)) {
2042                 inode->i_op = &ll_dir_inode_operations;
2043                 inode->i_fop = &ll_dir_operations;
2044                 EXIT;
2045         } else if (S_ISLNK(inode->i_mode)) {
2046                 inode->i_op = &ll_fast_symlink_inode_operations;
2047                 EXIT;
2048         } else {
2049                 inode->i_op = &ll_special_inode_operations;
2050
2051                 init_special_inode(inode, inode->i_mode,
2052                                    inode->i_rdev);
2053
2054                 EXIT;
2055         }
2056
2057         return 0;
2058 }
2059
2060 void ll_delete_inode(struct inode *inode)
2061 {
2062         struct ll_inode_info *lli = ll_i2info(inode);
2063         ENTRY;
2064
2065         if (S_ISREG(inode->i_mode) && lli->lli_clob != NULL)
2066                 /* It is last chance to write out dirty pages,
2067                  * otherwise we may lose data while umount */
2068                 cl_sync_file_range(inode, 0, OBD_OBJECT_EOF, CL_FSYNC_LOCAL, 1);
2069
2070         truncate_inode_pages_final(&inode->i_data);
2071
2072         LASSERTF(inode->i_data.nrpages == 0, "inode="DFID"(%p) nrpages=%lu, "
2073                  "see https://jira.hpdd.intel.com/browse/LU-118\n",
2074                  PFID(ll_inode2fid(inode)), inode, inode->i_data.nrpages);
2075
2076 #ifdef HAVE_SBOPS_EVICT_INODE
2077         ll_clear_inode(inode);
2078 #endif
2079         clear_inode(inode);
2080
2081         EXIT;
2082 }
2083
2084 int ll_iocontrol(struct inode *inode, struct file *file,
2085                  unsigned int cmd, unsigned long arg)
2086 {
2087         struct ll_sb_info *sbi = ll_i2sbi(inode);
2088         struct ptlrpc_request *req = NULL;
2089         int rc, flags = 0;
2090         ENTRY;
2091
2092         switch(cmd) {
2093         case FSFILT_IOC_GETFLAGS: {
2094                 struct mdt_body *body;
2095                 struct md_op_data *op_data;
2096
2097                 op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL,
2098                                              0, 0, LUSTRE_OPC_ANY,
2099                                              NULL);
2100                 if (IS_ERR(op_data))
2101                         RETURN(PTR_ERR(op_data));
2102
2103                 op_data->op_valid = OBD_MD_FLFLAGS;
2104                 rc = md_getattr(sbi->ll_md_exp, op_data, &req);
2105                 ll_finish_md_op_data(op_data);
2106                 if (rc) {
2107                         CERROR("%s: failure inode "DFID": rc = %d\n",
2108                                sbi->ll_md_exp->exp_obd->obd_name,
2109                                PFID(ll_inode2fid(inode)), rc);
2110                         RETURN(-abs(rc));
2111                 }
2112
2113                 body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
2114
2115                 flags = body->mbo_flags;
2116
2117                 ptlrpc_req_finished(req);
2118
2119                 RETURN(put_user(flags, (int __user *)arg));
2120         }
2121         case FSFILT_IOC_SETFLAGS: {
2122                 struct iattr *attr;
2123                 struct md_op_data *op_data;
2124                 struct cl_object *obj;
2125
2126                 if (get_user(flags, (int __user *)arg))
2127                         RETURN(-EFAULT);
2128
2129                 op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL, 0, 0,
2130                                              LUSTRE_OPC_ANY, NULL);
2131                 if (IS_ERR(op_data))
2132                         RETURN(PTR_ERR(op_data));
2133
2134                 op_data->op_attr_flags = flags;
2135                 op_data->op_attr.ia_valid |= ATTR_ATTR_FLAG;
2136                 rc = md_setattr(sbi->ll_md_exp, op_data, NULL, 0, &req);
2137                 ll_finish_md_op_data(op_data);
2138                 ptlrpc_req_finished(req);
2139                 if (rc)
2140                         RETURN(rc);
2141
2142                 inode->i_flags = ll_ext_to_inode_flags(flags);
2143
2144                 obj = ll_i2info(inode)->lli_clob;
2145                 if (obj == NULL)
2146                         RETURN(0);
2147
2148                 OBD_ALLOC_PTR(attr);
2149                 if (attr == NULL)
2150                         RETURN(-ENOMEM);
2151
2152                 attr->ia_valid = ATTR_ATTR_FLAG;
2153                 rc = cl_setattr_ost(obj, attr, flags);
2154
2155                 OBD_FREE_PTR(attr);
2156                 RETURN(rc);
2157         }
2158         default:
2159                 RETURN(-ENOSYS);
2160         }
2161
2162         RETURN(0);
2163 }
2164
2165 int ll_flush_ctx(struct inode *inode)
2166 {
2167         struct ll_sb_info  *sbi = ll_i2sbi(inode);
2168
2169         CDEBUG(D_SEC, "flush context for user %d\n",
2170                from_kuid(&init_user_ns, current_uid()));
2171
2172         obd_set_info_async(NULL, sbi->ll_md_exp,
2173                            sizeof(KEY_FLUSH_CTX), KEY_FLUSH_CTX,
2174                            0, NULL, NULL);
2175         obd_set_info_async(NULL, sbi->ll_dt_exp,
2176                            sizeof(KEY_FLUSH_CTX), KEY_FLUSH_CTX,
2177                            0, NULL, NULL);
2178         return 0;
2179 }
2180
2181 /* umount -f client means force down, don't save state */
2182 void ll_umount_begin(struct super_block *sb)
2183 {
2184         struct ll_sb_info *sbi = ll_s2sbi(sb);
2185         struct obd_device *obd;
2186         struct obd_ioctl_data *ioc_data;
2187         struct l_wait_info lwi;
2188         wait_queue_head_t waitq;
2189         ENTRY;
2190
2191         CDEBUG(D_VFSTRACE, "VFS Op: superblock %p count %d active %d\n", sb,
2192                sb->s_count, atomic_read(&sb->s_active));
2193
2194         obd = class_exp2obd(sbi->ll_md_exp);
2195         if (obd == NULL) {
2196                 CERROR("Invalid MDC connection handle %#llx\n",
2197                        sbi->ll_md_exp->exp_handle.h_cookie);
2198                 EXIT;
2199                 return;
2200         }
2201         obd->obd_force = 1;
2202
2203         obd = class_exp2obd(sbi->ll_dt_exp);
2204         if (obd == NULL) {
2205                 CERROR("Invalid LOV connection handle %#llx\n",
2206                        sbi->ll_dt_exp->exp_handle.h_cookie);
2207                 EXIT;
2208                 return;
2209         }
2210         obd->obd_force = 1;
2211
2212         OBD_ALLOC_PTR(ioc_data);
2213         if (ioc_data) {
2214                 obd_iocontrol(IOC_OSC_SET_ACTIVE, sbi->ll_md_exp,
2215                               sizeof *ioc_data, ioc_data, NULL);
2216
2217                 obd_iocontrol(IOC_OSC_SET_ACTIVE, sbi->ll_dt_exp,
2218                               sizeof *ioc_data, ioc_data, NULL);
2219
2220                 OBD_FREE_PTR(ioc_data);
2221         }
2222
2223         /* Really, we'd like to wait until there are no requests outstanding,
2224          * and then continue.  For now, we just periodically checking for vfs
2225          * to decrement mnt_cnt and hope to finish it within 10sec.
2226          */
2227         init_waitqueue_head(&waitq);
2228         lwi = LWI_TIMEOUT_INTERVAL(cfs_time_seconds(10),
2229                                    cfs_time_seconds(1), NULL, NULL);
2230         l_wait_event(waitq, may_umount(sbi->ll_mnt.mnt), &lwi);
2231
2232         EXIT;
2233 }
2234
2235 int ll_remount_fs(struct super_block *sb, int *flags, char *data)
2236 {
2237         struct ll_sb_info *sbi = ll_s2sbi(sb);
2238         char *profilenm = get_profile_name(sb);
2239         int err;
2240         __u32 read_only;
2241
2242         if ((*flags & MS_RDONLY) != (sb->s_flags & MS_RDONLY)) {
2243                 read_only = *flags & MS_RDONLY;
2244                 err = obd_set_info_async(NULL, sbi->ll_md_exp,
2245                                          sizeof(KEY_READ_ONLY),
2246                                          KEY_READ_ONLY, sizeof(read_only),
2247                                          &read_only, NULL);
2248                 if (err) {
2249                         LCONSOLE_WARN("Failed to remount %s %s (%d)\n",
2250                                       profilenm, read_only ?
2251                                       "read-only" : "read-write", err);
2252                         return err;
2253                 }
2254
2255                 if (read_only)
2256                         sb->s_flags |= MS_RDONLY;
2257                 else
2258                         sb->s_flags &= ~MS_RDONLY;
2259
2260                 if (sbi->ll_flags & LL_SBI_VERBOSE)
2261                         LCONSOLE_WARN("Remounted %s %s\n", profilenm,
2262                                       read_only ?  "read-only" : "read-write");
2263         }
2264         return 0;
2265 }
2266
2267 /**
2268  * Cleanup the open handle that is cached on MDT-side.
2269  *
2270  * For open case, the client side open handling thread may hit error
2271  * after the MDT grant the open. Under such case, the client should
2272  * send close RPC to the MDT as cleanup; otherwise, the open handle
2273  * on the MDT will be leaked there until the client umount or evicted.
2274  *
2275  * In further, if someone unlinked the file, because the open handle
2276  * holds the reference on such file/object, then it will block the
2277  * subsequent threads that want to locate such object via FID.
2278  *
2279  * \param[in] sb        super block for this file-system
2280  * \param[in] open_req  pointer to the original open request
2281  */
2282 void ll_open_cleanup(struct super_block *sb, struct ptlrpc_request *open_req)
2283 {
2284         struct mdt_body                 *body;
2285         struct md_op_data               *op_data;
2286         struct ptlrpc_request           *close_req = NULL;
2287         struct obd_export               *exp       = ll_s2sbi(sb)->ll_md_exp;
2288         ENTRY;
2289
2290         body = req_capsule_server_get(&open_req->rq_pill, &RMF_MDT_BODY);
2291         OBD_ALLOC_PTR(op_data);
2292         if (op_data == NULL) {
2293                 CWARN("%s: cannot allocate op_data to release open handle for "
2294                       DFID"\n",
2295                       ll_get_fsname(sb, NULL, 0), PFID(&body->mbo_fid1));
2296
2297                 RETURN_EXIT;
2298         }
2299
2300         op_data->op_fid1 = body->mbo_fid1;
2301         op_data->op_handle = body->mbo_handle;
2302         op_data->op_mod_time = ktime_get_real_seconds();
2303         md_close(exp, op_data, NULL, &close_req);
2304         ptlrpc_req_finished(close_req);
2305         ll_finish_md_op_data(op_data);
2306
2307         EXIT;
2308 }
2309
2310 int ll_prep_inode(struct inode **inode, struct ptlrpc_request *req,
2311                   struct super_block *sb, struct lookup_intent *it)
2312 {
2313         struct ll_sb_info *sbi = NULL;
2314         struct lustre_md md = { NULL };
2315         int rc;
2316         ENTRY;
2317
2318         LASSERT(*inode || sb);
2319         sbi = sb ? ll_s2sbi(sb) : ll_i2sbi(*inode);
2320         rc = md_get_lustre_md(sbi->ll_md_exp, req, sbi->ll_dt_exp,
2321                               sbi->ll_md_exp, &md);
2322         if (rc != 0)
2323                 GOTO(cleanup, rc);
2324
2325         if (*inode) {
2326                 rc = ll_update_inode(*inode, &md);
2327                 if (rc != 0)
2328                         GOTO(out, rc);
2329         } else {
2330                 LASSERT(sb != NULL);
2331
2332                 /*
2333                  * At this point server returns to client's same fid as client
2334                  * generated for creating. So using ->fid1 is okay here.
2335                  */
2336                 if (!fid_is_sane(&md.body->mbo_fid1)) {
2337                         CERROR("%s: Fid is insane "DFID"\n",
2338                                 ll_get_fsname(sb, NULL, 0),
2339                                 PFID(&md.body->mbo_fid1));
2340                         GOTO(out, rc = -EINVAL);
2341                 }
2342
2343                 *inode = ll_iget(sb, cl_fid_build_ino(&md.body->mbo_fid1,
2344                                              sbi->ll_flags & LL_SBI_32BIT_API),
2345                                  &md);
2346                 if (IS_ERR(*inode)) {
2347 #ifdef CONFIG_FS_POSIX_ACL
2348                         if (md.posix_acl) {
2349                                 posix_acl_release(md.posix_acl);
2350                                 md.posix_acl = NULL;
2351                         }
2352 #endif
2353                         rc = IS_ERR(*inode) ? PTR_ERR(*inode) : -ENOMEM;
2354                         *inode = NULL;
2355                         CERROR("new_inode -fatal: rc %d\n", rc);
2356                         GOTO(out, rc);
2357                 }
2358         }
2359
2360         /* Handling piggyback layout lock.
2361          * Layout lock can be piggybacked by getattr and open request.
2362          * The lsm can be applied to inode only if it comes with a layout lock
2363          * otherwise correct layout may be overwritten, for example:
2364          * 1. proc1: mdt returns a lsm but not granting layout
2365          * 2. layout was changed by another client
2366          * 3. proc2: refresh layout and layout lock granted
2367          * 4. proc1: to apply a stale layout */
2368         if (it != NULL && it->it_lock_mode != 0) {
2369                 struct lustre_handle lockh;
2370                 struct ldlm_lock *lock;
2371
2372                 lockh.cookie = it->it_lock_handle;
2373                 lock = ldlm_handle2lock(&lockh);
2374                 LASSERT(lock != NULL);
2375                 if (ldlm_has_layout(lock)) {
2376                         struct cl_object_conf conf;
2377
2378                         memset(&conf, 0, sizeof(conf));
2379                         conf.coc_opc = OBJECT_CONF_SET;
2380                         conf.coc_inode = *inode;
2381                         conf.coc_lock = lock;
2382                         conf.u.coc_layout = md.layout;
2383                         (void)ll_layout_conf(*inode, &conf);
2384                 }
2385                 LDLM_LOCK_PUT(lock);
2386         }
2387
2388         GOTO(out, rc = 0);
2389
2390 out:
2391         md_free_lustre_md(sbi->ll_md_exp, &md);
2392
2393 cleanup:
2394         if (rc != 0 && it != NULL && it->it_op & IT_OPEN)
2395                 ll_open_cleanup(sb != NULL ? sb : (*inode)->i_sb, req);
2396
2397         return rc;
2398 }
2399
2400 int ll_obd_statfs(struct inode *inode, void __user *arg)
2401 {
2402         struct ll_sb_info *sbi = NULL;
2403         struct obd_export *exp;
2404         char *buf = NULL;
2405         struct obd_ioctl_data *data = NULL;
2406         __u32 type;
2407         int len = 0, rc;
2408
2409         if (!inode || !(sbi = ll_i2sbi(inode)))
2410                 GOTO(out_statfs, rc = -EINVAL);
2411
2412         rc = obd_ioctl_getdata(&buf, &len, arg);
2413         if (rc)
2414                 GOTO(out_statfs, rc);
2415
2416         data = (void*)buf;
2417         if (!data->ioc_inlbuf1 || !data->ioc_inlbuf2 ||
2418             !data->ioc_pbuf1 || !data->ioc_pbuf2)
2419                 GOTO(out_statfs, rc = -EINVAL);
2420
2421         if (data->ioc_inllen1 != sizeof(__u32) ||
2422             data->ioc_inllen2 != sizeof(__u32) ||
2423             data->ioc_plen1 != sizeof(struct obd_statfs) ||
2424             data->ioc_plen2 != sizeof(struct obd_uuid))
2425                 GOTO(out_statfs, rc = -EINVAL);
2426
2427         memcpy(&type, data->ioc_inlbuf1, sizeof(__u32));
2428         if (type & LL_STATFS_LMV)
2429                 exp = sbi->ll_md_exp;
2430         else if (type & LL_STATFS_LOV)
2431                 exp = sbi->ll_dt_exp;
2432         else
2433                 GOTO(out_statfs, rc = -ENODEV);
2434
2435         rc = obd_iocontrol(IOC_OBD_STATFS, exp, len, buf, NULL);
2436         if (rc)
2437                 GOTO(out_statfs, rc);
2438 out_statfs:
2439         OBD_FREE_LARGE(buf, len);
2440         return rc;
2441 }
2442
2443 int ll_process_config(struct lustre_cfg *lcfg)
2444 {
2445         struct super_block *sb;
2446         unsigned long x;
2447         int rc = 0;
2448         char *ptr;
2449
2450         /* The instance name contains the sb: lustre-client-aacfe000 */
2451         ptr = strrchr(lustre_cfg_string(lcfg, 0), '-');
2452         if (!ptr || !*(++ptr))
2453                 return -EINVAL;
2454         if (sscanf(ptr, "%lx", &x) != 1)
2455                 return -EINVAL;
2456         sb = (struct super_block *)x;
2457         /* This better be a real Lustre superblock! */
2458         LASSERT(s2lsi(sb)->lsi_lmd->lmd_magic == LMD_MAGIC);
2459
2460         /* Note we have not called client_common_fill_super yet, so
2461            proc fns must be able to handle that! */
2462         rc = class_process_proc_param(PARAM_LLITE, lprocfs_llite_obd_vars,
2463                                       lcfg, sb);
2464         if (rc > 0)
2465                 rc = 0;
2466         return rc;
2467 }
2468
2469 /* this function prepares md_op_data hint for passing it down to MD stack. */
2470 struct md_op_data *ll_prep_md_op_data(struct md_op_data *op_data,
2471                                       struct inode *i1, struct inode *i2,
2472                                       const char *name, size_t namelen,
2473                                       __u32 mode, __u32 opc, void *data)
2474 {
2475         LASSERT(i1 != NULL);
2476
2477         if (name == NULL) {
2478                 /* Do not reuse namelen for something else. */
2479                 if (namelen != 0)
2480                         return ERR_PTR(-EINVAL);
2481         } else {
2482                 if (namelen > ll_i2sbi(i1)->ll_namelen)
2483                         return ERR_PTR(-ENAMETOOLONG);
2484
2485                 if (!lu_name_is_valid_2(name, namelen))
2486                         return ERR_PTR(-EINVAL);
2487         }
2488
2489         if (op_data == NULL)
2490                 OBD_ALLOC_PTR(op_data);
2491
2492         if (op_data == NULL)
2493                 return ERR_PTR(-ENOMEM);
2494
2495         ll_i2gids(op_data->op_suppgids, i1, i2);
2496         op_data->op_fid1 = *ll_inode2fid(i1);
2497         op_data->op_default_stripe_offset = -1;
2498         if (S_ISDIR(i1->i_mode)) {
2499                 op_data->op_mea1 = ll_i2info(i1)->lli_lsm_md;
2500                 if (opc == LUSTRE_OPC_MKDIR)
2501                         op_data->op_default_stripe_offset =
2502                                    ll_i2info(i1)->lli_def_stripe_offset;
2503         }
2504
2505         if (i2) {
2506                 op_data->op_fid2 = *ll_inode2fid(i2);
2507                 if (S_ISDIR(i2->i_mode))
2508                         op_data->op_mea2 = ll_i2info(i2)->lli_lsm_md;
2509         } else {
2510                 fid_zero(&op_data->op_fid2);
2511         }
2512
2513         if (ll_i2sbi(i1)->ll_flags & LL_SBI_64BIT_HASH)
2514                 op_data->op_cli_flags |= CLI_HASH64;
2515
2516         if (ll_need_32bit_api(ll_i2sbi(i1)))
2517                 op_data->op_cli_flags |= CLI_API32;
2518
2519         op_data->op_name = name;
2520         op_data->op_namelen = namelen;
2521         op_data->op_mode = mode;
2522         op_data->op_mod_time = ktime_get_real_seconds();
2523         op_data->op_fsuid = from_kuid(&init_user_ns, current_fsuid());
2524         op_data->op_fsgid = from_kgid(&init_user_ns, current_fsgid());
2525         op_data->op_cap = cfs_curproc_cap_pack();
2526         op_data->op_mds = 0;
2527         if ((opc == LUSTRE_OPC_CREATE) && (name != NULL) &&
2528              filename_is_volatile(name, namelen, &op_data->op_mds)) {
2529                 op_data->op_bias |= MDS_CREATE_VOLATILE;
2530         }
2531         op_data->op_data = data;
2532
2533         return op_data;
2534 }
2535
2536 void ll_finish_md_op_data(struct md_op_data *op_data)
2537 {
2538         security_release_secctx(op_data->op_file_secctx,
2539                                 op_data->op_file_secctx_size);
2540         OBD_FREE_PTR(op_data);
2541 }
2542
2543 #ifdef HAVE_SUPEROPS_USE_DENTRY
2544 int ll_show_options(struct seq_file *seq, struct dentry *dentry)
2545 #else
2546 int ll_show_options(struct seq_file *seq, struct vfsmount *vfs)
2547 #endif
2548 {
2549         struct ll_sb_info *sbi;
2550
2551 #ifdef HAVE_SUPEROPS_USE_DENTRY
2552         LASSERT((seq != NULL) && (dentry != NULL));
2553         sbi = ll_s2sbi(dentry->d_sb);
2554 #else
2555         LASSERT((seq != NULL) && (vfs != NULL));
2556         sbi = ll_s2sbi(vfs->mnt_sb);
2557 #endif
2558
2559         if (sbi->ll_flags & LL_SBI_NOLCK)
2560                 seq_puts(seq, ",nolock");
2561
2562         if (sbi->ll_flags & LL_SBI_FLOCK)
2563                 seq_puts(seq, ",flock");
2564
2565         if (sbi->ll_flags & LL_SBI_LOCALFLOCK)
2566                 seq_puts(seq, ",localflock");
2567
2568         if (sbi->ll_flags & LL_SBI_USER_XATTR)
2569                 seq_puts(seq, ",user_xattr");
2570
2571         if (sbi->ll_flags & LL_SBI_LAZYSTATFS)
2572                 seq_puts(seq, ",lazystatfs");
2573
2574         if (sbi->ll_flags & LL_SBI_USER_FID2PATH)
2575                 seq_puts(seq, ",user_fid2path");
2576
2577         if (sbi->ll_flags & LL_SBI_ALWAYS_PING)
2578                 seq_puts(seq, ",always_ping");
2579
2580         RETURN(0);
2581 }
2582
2583 /**
2584  * Get obd name by cmd, and copy out to user space
2585  */
2586 int ll_get_obd_name(struct inode *inode, unsigned int cmd, unsigned long arg)
2587 {
2588         struct ll_sb_info *sbi = ll_i2sbi(inode);
2589         struct obd_device *obd;
2590         ENTRY;
2591
2592         if (cmd == OBD_IOC_GETDTNAME)
2593                 obd = class_exp2obd(sbi->ll_dt_exp);
2594         else if (cmd == OBD_IOC_GETMDNAME)
2595                 obd = class_exp2obd(sbi->ll_md_exp);
2596         else
2597                 RETURN(-EINVAL);
2598
2599         if (!obd)
2600                 RETURN(-ENOENT);
2601
2602         if (copy_to_user((void __user *)arg, obd->obd_name,
2603                          strlen(obd->obd_name) + 1))
2604                 RETURN(-EFAULT);
2605
2606         RETURN(0);
2607 }
2608
2609 /**
2610  * Get lustre file system name by \a sbi. If \a buf is provided(non-NULL), the
2611  * fsname will be returned in this buffer; otherwise, a static buffer will be
2612  * used to store the fsname and returned to caller.
2613  */
2614 char *ll_get_fsname(struct super_block *sb, char *buf, int buflen)
2615 {
2616         static char fsname_static[MTI_NAME_MAXLEN];
2617         struct lustre_sb_info *lsi = s2lsi(sb);
2618         char *ptr;
2619         int len;
2620
2621         if (buf == NULL) {
2622                 /* this means the caller wants to use static buffer
2623                  * and it doesn't care about race. Usually this is
2624                  * in error reporting path */
2625                 buf = fsname_static;
2626                 buflen = sizeof(fsname_static);
2627         }
2628
2629         len = strlen(lsi->lsi_lmd->lmd_profile);
2630         ptr = strrchr(lsi->lsi_lmd->lmd_profile, '-');
2631         if (ptr && (strcmp(ptr, "-client") == 0))
2632                 len -= 7;
2633
2634         if (unlikely(len >= buflen))
2635                 len = buflen - 1;
2636         strncpy(buf, lsi->lsi_lmd->lmd_profile, len);
2637         buf[len] = '\0';
2638
2639         return buf;
2640 }
2641
2642 static char* ll_d_path(struct dentry *dentry, char *buf, int bufsize)
2643 {
2644         char *path = NULL;
2645
2646         struct path p;
2647
2648         p.dentry = dentry;
2649         p.mnt = current->fs->root.mnt;
2650         path_get(&p);
2651         path = d_path(&p, buf, bufsize);
2652         path_put(&p);
2653         return path;
2654 }
2655
2656 void ll_dirty_page_discard_warn(struct page *page, int ioret)
2657 {
2658         char *buf, *path = NULL;
2659         struct dentry *dentry = NULL;
2660         struct inode *inode = page->mapping->host;
2661
2662         /* this can be called inside spin lock so use GFP_ATOMIC. */
2663         buf = (char *)__get_free_page(GFP_ATOMIC);
2664         if (buf != NULL) {
2665                 dentry = d_find_alias(page->mapping->host);
2666                 if (dentry != NULL)
2667                         path = ll_d_path(dentry, buf, PAGE_SIZE);
2668         }
2669
2670         CDEBUG(D_WARNING,
2671                "%s: dirty page discard: %s/fid: "DFID"/%s may get corrupted "
2672                "(rc %d)\n", ll_get_fsname(page->mapping->host->i_sb, NULL, 0),
2673                s2lsi(page->mapping->host->i_sb)->lsi_lmd->lmd_dev,
2674                PFID(ll_inode2fid(inode)),
2675                (path && !IS_ERR(path)) ? path : "", ioret);
2676
2677         if (dentry != NULL)
2678                 dput(dentry);
2679
2680         if (buf != NULL)
2681                 free_page((unsigned long)buf);
2682 }
2683
2684 ssize_t ll_copy_user_md(const struct lov_user_md __user *md,
2685                         struct lov_user_md **kbuf)
2686 {
2687         struct lov_user_md      lum;
2688         ssize_t                 lum_size;
2689         ENTRY;
2690
2691         if (copy_from_user(&lum, md, sizeof(lum)))
2692                 RETURN(-EFAULT);
2693
2694         lum_size = ll_lov_user_md_size(&lum);
2695         if (lum_size < 0)
2696                 RETURN(lum_size);
2697
2698         OBD_ALLOC(*kbuf, lum_size);
2699         if (*kbuf == NULL)
2700                 RETURN(-ENOMEM);
2701
2702         if (copy_from_user(*kbuf, md, lum_size) != 0) {
2703                 OBD_FREE(*kbuf, lum_size);
2704                 RETURN(-EFAULT);
2705         }
2706
2707         RETURN(lum_size);
2708 }
2709
2710 /*
2711  * Compute llite root squash state after a change of root squash
2712  * configuration setting or add/remove of a lnet nid
2713  */
2714 void ll_compute_rootsquash_state(struct ll_sb_info *sbi)
2715 {
2716         struct root_squash_info *squash = &sbi->ll_squash;
2717         int i;
2718         bool matched;
2719         struct lnet_process_id id;
2720
2721         /* Update norootsquash flag */
2722         down_write(&squash->rsi_sem);
2723         if (list_empty(&squash->rsi_nosquash_nids))
2724                 sbi->ll_flags &= ~LL_SBI_NOROOTSQUASH;
2725         else {
2726                 /* Do not apply root squash as soon as one of our NIDs is
2727                  * in the nosquash_nids list */
2728                 matched = false;
2729                 i = 0;
2730                 while (LNetGetId(i++, &id) != -ENOENT) {
2731                         if (LNET_NETTYP(LNET_NIDNET(id.nid)) == LOLND)
2732                                 continue;
2733                         if (cfs_match_nid(id.nid, &squash->rsi_nosquash_nids)) {
2734                                 matched = true;
2735                                 break;
2736                         }
2737                 }
2738                 if (matched)
2739                         sbi->ll_flags |= LL_SBI_NOROOTSQUASH;
2740                 else
2741                         sbi->ll_flags &= ~LL_SBI_NOROOTSQUASH;
2742         }
2743         up_write(&squash->rsi_sem);
2744 }
2745
2746 /**
2747  * Parse linkea content to extract information about a given hardlink
2748  *
2749  * \param[in]   ldata      - Initialized linkea data
2750  * \param[in]   linkno     - Link identifier
2751  * \param[out]  parent_fid - The entry's parent FID
2752  * \param[out]  ln         - Entry name destination buffer
2753  *
2754  * \retval 0 on success
2755  * \retval Appropriate negative error code on failure
2756  */
2757 static int ll_linkea_decode(struct linkea_data *ldata, unsigned int linkno,
2758                             struct lu_fid *parent_fid, struct lu_name *ln)
2759 {
2760         unsigned int    idx;
2761         int             rc;
2762         ENTRY;
2763
2764         rc = linkea_init_with_rec(ldata);
2765         if (rc < 0)
2766                 RETURN(rc);
2767
2768         if (linkno >= ldata->ld_leh->leh_reccount)
2769                 /* beyond last link */
2770                 RETURN(-ENODATA);
2771
2772         linkea_first_entry(ldata);
2773         for (idx = 0; ldata->ld_lee != NULL; idx++) {
2774                 linkea_entry_unpack(ldata->ld_lee, &ldata->ld_reclen, ln,
2775                                     parent_fid);
2776                 if (idx == linkno)
2777                         break;
2778
2779                 linkea_next_entry(ldata);
2780         }
2781
2782         if (idx < linkno)
2783                 RETURN(-ENODATA);
2784
2785         RETURN(0);
2786 }
2787
2788 /**
2789  * Get parent FID and name of an identified link. Operation is performed for
2790  * a given link number, letting the caller iterate over linkno to list one or
2791  * all links of an entry.
2792  *
2793  * \param[in]     file - File descriptor against which to perform the operation
2794  * \param[in,out] arg  - User-filled structure containing the linkno to operate
2795  *                       on and the available size. It is eventually filled with
2796  *                       the requested information or left untouched on error
2797  *
2798  * \retval - 0 on success
2799  * \retval - Appropriate negative error code on failure
2800  */
2801 int ll_getparent(struct file *file, struct getparent __user *arg)
2802 {
2803         struct inode            *inode = file_inode(file);
2804         struct linkea_data      *ldata;
2805         struct lu_buf            buf = LU_BUF_NULL;
2806         struct lu_name           ln;
2807         struct lu_fid            parent_fid;
2808         __u32                    linkno;
2809         __u32                    name_size;
2810         int                      rc;
2811
2812         ENTRY;
2813
2814         if (!cfs_capable(CFS_CAP_DAC_READ_SEARCH) &&
2815             !(ll_i2sbi(inode)->ll_flags & LL_SBI_USER_FID2PATH))
2816                 RETURN(-EPERM);
2817
2818         if (get_user(name_size, &arg->gp_name_size))
2819                 RETURN(-EFAULT);
2820
2821         if (get_user(linkno, &arg->gp_linkno))
2822                 RETURN(-EFAULT);
2823
2824         if (name_size > PATH_MAX)
2825                 RETURN(-EINVAL);
2826
2827         OBD_ALLOC(ldata, sizeof(*ldata));
2828         if (ldata == NULL)
2829                 RETURN(-ENOMEM);
2830
2831         rc = linkea_data_new(ldata, &buf);
2832         if (rc < 0)
2833                 GOTO(ldata_free, rc);
2834
2835 #ifdef HAVE_XATTR_HANDLER_FLAGS
2836         rc = ll_xattr_list(inode, XATTR_NAME_LINK, XATTR_TRUSTED_T, buf.lb_buf,
2837                            buf.lb_len, OBD_MD_FLXATTR);
2838 #else
2839         rc = ll_getxattr(file_dentry(file), XATTR_NAME_LINK, buf.lb_buf,
2840                          buf.lb_len);
2841 #endif /* HAVE_XATTR_HANDLER_FLAGS */
2842         if (rc < 0)
2843                 GOTO(lb_free, rc);
2844
2845         rc = ll_linkea_decode(ldata, linkno, &parent_fid, &ln);
2846         if (rc < 0)
2847                 GOTO(lb_free, rc);
2848
2849         if (ln.ln_namelen >= name_size)
2850                 GOTO(lb_free, rc = -EOVERFLOW);
2851
2852         if (copy_to_user(&arg->gp_fid, &parent_fid, sizeof(arg->gp_fid)))
2853                 GOTO(lb_free, rc = -EFAULT);
2854
2855         if (copy_to_user(&arg->gp_name, ln.ln_name, ln.ln_namelen))
2856                 GOTO(lb_free, rc = -EFAULT);
2857
2858         if (put_user('\0', arg->gp_name + ln.ln_namelen))
2859                 GOTO(lb_free, rc = -EFAULT);
2860
2861 lb_free:
2862         lu_buf_free(&buf);
2863 ldata_free:
2864         OBD_FREE(ldata, sizeof(*ldata));
2865
2866         RETURN(rc);
2867 }