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         sbi->ll_flags |= LL_SBI_TINY_WRITE;
 134
 135         /* root squash */
 136         sbi->ll_squash.rsi_uid = 0;
 137         sbi->ll_squash.rsi_gid = 0;
 138         INIT_LIST_HEAD(&sbi->ll_squash.rsi_nosquash_nids);
 139         init_rwsem(&sbi->ll_squash.rsi_sem);
 140
 141         RETURN(sbi);
 142 }
 143
 144 static void ll_free_sbi(struct super_block *sb)
 145 {
 146         struct ll_sb_info *sbi = ll_s2sbi(sb);
 147         ENTRY;
 148
 149         if (sbi != NULL) {
 150                 if (!list_empty(&sbi->ll_squash.rsi_nosquash_nids))
 151                         cfs_free_nidlist(&sbi->ll_squash.rsi_nosquash_nids);
 152                 if (sbi->ll_cache != NULL) {
 153                         cl_cache_decref(sbi->ll_cache);
 154                         sbi->ll_cache = NULL;
 155                 }
 156                 OBD_FREE(sbi, sizeof(*sbi));
 157         }
 158         EXIT;
 159 }
 160
 161 static inline int obd_connect_has_secctx(struct obd_connect_data *data)
 162 {
 163         return data->ocd_connect_flags & OBD_CONNECT_FLAGS2 &&
 164                data->ocd_connect_flags2 & OBD_CONNECT2_FILE_SECCTX;
 165 }
 166
 167 static int client_common_fill_super(struct super_block *sb, char *md, char *dt,
 168                                     struct vfsmount *mnt)
 169 {
 170         struct inode *root = NULL;
 171         struct ll_sb_info *sbi = ll_s2sbi(sb);
 172         struct obd_device *obd;
 173         struct obd_statfs *osfs = NULL;
 174         struct ptlrpc_request *request = NULL;
 175         struct obd_connect_data *data = NULL;
 176         struct obd_uuid *uuid;
 177         struct md_op_data *op_data;
 178         struct lustre_md lmd;
 179         u64 valid;
 180         int size, err, checksum;
 181         ENTRY;
 182
 183         obd = class_name2obd(md);
 184         if (!obd) {
 185                 CERROR("MD %s: not setup or attached\n", md);
 186                 RETURN(-EINVAL);
 187         }
 188
 189         OBD_ALLOC_PTR(data);
 190         if (data == NULL)
 191                 RETURN(-ENOMEM);
 192
 193         OBD_ALLOC_PTR(osfs);
 194         if (osfs == NULL) {
 195                 OBD_FREE_PTR(data);
 196                 RETURN(-ENOMEM);
 197         }
 198
 199         /* pass client page size via ocd_grant_blkbits, the server should report
 200          * back its backend blocksize for grant calculation purpose */
 201         data->ocd_grant_blkbits = PAGE_SHIFT;
 202
 203         /* indicate MDT features supported by this client */
 204         data->ocd_connect_flags = OBD_CONNECT_IBITS    | OBD_CONNECT_NODEVOH  |
 205                                   OBD_CONNECT_ATTRFID  | OBD_CONNECT_GRANT |
 206                                   OBD_CONNECT_VERSION  | OBD_CONNECT_BRW_SIZE |
 207                                   OBD_CONNECT_SRVLOCK  | OBD_CONNECT_TRUNCLOCK|
 208                                   OBD_CONNECT_MDS_CAPA | OBD_CONNECT_OSS_CAPA |
 209                                   OBD_CONNECT_CANCELSET | OBD_CONNECT_FID     |
 210                                   OBD_CONNECT_AT       | OBD_CONNECT_LOV_V3   |
 211                                   OBD_CONNECT_VBR | OBD_CONNECT_FULL20 |
 212                                   OBD_CONNECT_64BITHASH |
 213                                   OBD_CONNECT_EINPROGRESS |
 214                                   OBD_CONNECT_JOBSTATS | OBD_CONNECT_LVB_TYPE |
 215                                   OBD_CONNECT_LAYOUTLOCK | OBD_CONNECT_PINGLESS|
 216                                   OBD_CONNECT_MAX_EASIZE |
 217                                   OBD_CONNECT_FLOCK_DEAD |
 218                                   OBD_CONNECT_DISP_STRIPE | OBD_CONNECT_LFSCK |
 219                                   OBD_CONNECT_OPEN_BY_FID |
 220                                   OBD_CONNECT_DIR_STRIPE |
 221                                   OBD_CONNECT_BULK_MBITS | OBD_CONNECT_CKSUM |
 222                                   OBD_CONNECT_SUBTREE |
 223                                   OBD_CONNECT_MULTIMODRPCS |
 224                                   OBD_CONNECT_GRANT_PARAM |
 225                                   OBD_CONNECT_SHORTIO | OBD_CONNECT_FLAGS2;
 226
 227         data->ocd_connect_flags2 = OBD_CONNECT2_FLR;
 228
 229 #ifdef HAVE_LRU_RESIZE_SUPPORT
 230         if (sbi->ll_flags & LL_SBI_LRU_RESIZE)
 231                 data->ocd_connect_flags |= OBD_CONNECT_LRU_RESIZE;
 232 #endif
 233 #ifdef CONFIG_FS_POSIX_ACL
 234         data->ocd_connect_flags |= OBD_CONNECT_ACL | OBD_CONNECT_UMASK |
 235                                    OBD_CONNECT_LARGE_ACL;
 236 #endif
 237
 238         data->ocd_cksum_types = cksum_types_supported_client();
 239
 240         if (OBD_FAIL_CHECK(OBD_FAIL_MDC_LIGHTWEIGHT))
 241                 /* flag mdc connection as lightweight, only used for test
 242                  * purpose, use with care */
 243                 data->ocd_connect_flags |= OBD_CONNECT_LIGHTWEIGHT;
 244
 245         data->ocd_ibits_known = MDS_INODELOCK_FULL;
 246         data->ocd_version = LUSTRE_VERSION_CODE;
 247
 248         if (sb->s_flags & MS_RDONLY)
 249                 data->ocd_connect_flags |= OBD_CONNECT_RDONLY;
 250         if (sbi->ll_flags & LL_SBI_USER_XATTR)
 251                 data->ocd_connect_flags |= OBD_CONNECT_XATTR;
 252
 253 #ifdef MS_NOSEC
 254         /* Setting this indicates we correctly support S_NOSEC (See kernel
 255          * commit 9e1f1de02c2275d7172e18dc4e7c2065777611bf)
 256          */
 257         sb->s_flags |= MS_NOSEC;
 258 #endif
 259
 260         if (sbi->ll_flags & LL_SBI_FLOCK)
 261                 sbi->ll_fop = &ll_file_operations_flock;
 262         else if (sbi->ll_flags & LL_SBI_LOCALFLOCK)
 263                 sbi->ll_fop = &ll_file_operations;
 264         else
 265                 sbi->ll_fop = &ll_file_operations_noflock;
 266
 267         /* always ping even if server suppress_pings */
 268         if (sbi->ll_flags & LL_SBI_ALWAYS_PING)
 269                 data->ocd_connect_flags &= ~OBD_CONNECT_PINGLESS;
 270
 271 #if defined(HAVE_SECURITY_DENTRY_INIT_SECURITY) && defined(CONFIG_SECURITY)
 272         data->ocd_connect_flags2 |= OBD_CONNECT2_FILE_SECCTX;
 273 #endif /* HAVE_SECURITY_DENTRY_INIT_SECURITY */
 274
 275         data->ocd_brw_size = MD_MAX_BRW_SIZE;
 276
 277         err = obd_connect(NULL, &sbi->ll_md_exp, obd, &sbi->ll_sb_uuid, data, NULL);
 278         if (err == -EBUSY) {
 279                 LCONSOLE_ERROR_MSG(0x14f, "An MDT (md %s) is performing "
 280                                    "recovery, of which this client is not a "
 281                                    "part. Please wait for recovery to complete,"
 282                                    " abort, or time out.\n", md);
 283                 GOTO(out, err);
 284         } else if (err) {
 285                 CERROR("cannot connect to %s: rc = %d\n", md, err);
 286                 GOTO(out, err);
 287         }
 288
 289         sbi->ll_md_exp->exp_connect_data = *data;
 290
 291         err = obd_fid_init(sbi->ll_md_exp->exp_obd, sbi->ll_md_exp,
 292                            LUSTRE_SEQ_METADATA);
 293         if (err) {
 294                 CERROR("%s: Can't init metadata layer FID infrastructure, "
 295                        "rc = %d\n", sbi->ll_md_exp->exp_obd->obd_name, err);
 296                 GOTO(out_md, err);
 297         }
 298
 299         /* For mount, we only need fs info from MDT0, and also in DNE, it
 300          * can make sure the client can be mounted as long as MDT0 is
 301          * avaible */
 302         err = obd_statfs(NULL, sbi->ll_md_exp, osfs,
 303                         ktime_get_seconds() -OBD_STATFS_CACHE_SECONDS,
 304                         OBD_STATFS_FOR_MDT0);
 305         if (err)
 306                 GOTO(out_md_fid, err);
 307
 308         /* This needs to be after statfs to ensure connect has finished.
 309          * Note that "data" does NOT contain the valid connect reply.
 310          * If connecting to a 1.8 server there will be no LMV device, so
 311          * we can access the MDC export directly and exp_connect_flags will
 312          * be non-zero, but if accessing an upgraded 2.1 server it will
 313          * have the correct flags filled in.
 314          * XXX: fill in the LMV exp_connect_flags from MDC(s). */
 315         valid = exp_connect_flags(sbi->ll_md_exp) & CLIENT_CONNECT_MDT_REQD;
 316         if (exp_connect_flags(sbi->ll_md_exp) != 0 &&
 317             valid != CLIENT_CONNECT_MDT_REQD) {
 318                 char *buf;
 319
 320                 OBD_ALLOC_WAIT(buf, PAGE_SIZE);
 321                 obd_connect_flags2str(buf, PAGE_SIZE,
 322                                       valid ^ CLIENT_CONNECT_MDT_REQD, 0, ",");
 323                 LCONSOLE_ERROR_MSG(0x170, "Server %s does not support "
 324                                    "feature(s) needed for correct operation "
 325                                    "of this client (%s). Please upgrade "
 326                                    "server or downgrade client.\n",
 327                                    sbi->ll_md_exp->exp_obd->obd_name, buf);
 328                 OBD_FREE(buf, PAGE_SIZE);
 329                 GOTO(out_md_fid, err = -EPROTO);
 330         }
 331
 332         size = sizeof(*data);
 333         err = obd_get_info(NULL, sbi->ll_md_exp, sizeof(KEY_CONN_DATA),
 334                            KEY_CONN_DATA,  &size, data);
 335         if (err) {
 336                 CERROR("%s: Get connect data failed: rc = %d\n",
 337                        sbi->ll_md_exp->exp_obd->obd_name, err);
 338                 GOTO(out_md_fid, err);
 339         }
 340
 341         LASSERT(osfs->os_bsize);
 342         sb->s_blocksize = osfs->os_bsize;
 343         sb->s_blocksize_bits = log2(osfs->os_bsize);
 344         sb->s_magic = LL_SUPER_MAGIC;
 345         sb->s_maxbytes = MAX_LFS_FILESIZE;
 346         sbi->ll_namelen = osfs->os_namelen;
 347         sbi->ll_mnt.mnt = current->fs->root.mnt;
 348
 349         if ((sbi->ll_flags & LL_SBI_USER_XATTR) &&
 350             !(data->ocd_connect_flags & OBD_CONNECT_XATTR)) {
 351                 LCONSOLE_INFO("Disabling user_xattr feature because "
 352                               "it is not supported on the server\n");
 353                 sbi->ll_flags &= ~LL_SBI_USER_XATTR;
 354         }
 355
 356         if (data->ocd_connect_flags & OBD_CONNECT_ACL) {
 357 #ifdef MS_POSIXACL
 358                 sb->s_flags |= MS_POSIXACL;
 359 #endif
 360                 sbi->ll_flags |= LL_SBI_ACL;
 361         } else {
 362                 LCONSOLE_INFO("client wants to enable acl, but mdt not!\n");
 363 #ifdef MS_POSIXACL
 364                 sb->s_flags &= ~MS_POSIXACL;
 365 #endif
 366                 sbi->ll_flags &= ~LL_SBI_ACL;
 367         }
 368
 369         if (data->ocd_connect_flags & OBD_CONNECT_64BITHASH)
 370                 sbi->ll_flags |= LL_SBI_64BIT_HASH;
 371
 372         if (data->ocd_connect_flags & OBD_CONNECT_LAYOUTLOCK)
 373                 sbi->ll_flags |= LL_SBI_LAYOUT_LOCK;
 374
 375         if (obd_connect_has_secctx(data))
 376                 sbi->ll_flags |= LL_SBI_FILE_SECCTX;
 377
 378         if (data->ocd_ibits_known & MDS_INODELOCK_XATTR) {
 379                 if (!(data->ocd_connect_flags & OBD_CONNECT_MAX_EASIZE)) {
 380                         LCONSOLE_INFO("%s: disabling xattr cache due to "
 381                                       "unknown maximum xattr size.\n", dt);
 382                 } else if (!sbi->ll_xattr_cache_set) {
 383                         /* If xattr_cache is already set (no matter 0 or 1)
 384                          * during processing llog, it won't be enabled here. */
 385                         sbi->ll_flags |= LL_SBI_XATTR_CACHE;
 386                         sbi->ll_xattr_cache_enabled = 1;
 387                 }
 388         }
 389
 390         obd = class_name2obd(dt);
 391         if (!obd) {
 392                 CERROR("DT %s: not setup or attached\n", dt);
 393                 GOTO(out_md_fid, err = -ENODEV);
 394         }
 395
 396         /* pass client page size via ocd_grant_blkbits, the server should report
 397          * back its backend blocksize for grant calculation purpose */
 398         data->ocd_grant_blkbits = PAGE_SHIFT;
 399
 400         /* indicate OST features supported by this client */
 401         data->ocd_connect_flags = OBD_CONNECT_GRANT | OBD_CONNECT_VERSION |
 402                                   OBD_CONNECT_REQPORTAL | OBD_CONNECT_BRW_SIZE |
 403                                   OBD_CONNECT_CANCELSET | OBD_CONNECT_FID |
 404                                   OBD_CONNECT_SRVLOCK | OBD_CONNECT_TRUNCLOCK|
 405                                   OBD_CONNECT_AT | OBD_CONNECT_OSS_CAPA |
 406                                   OBD_CONNECT_VBR | OBD_CONNECT_FULL20 |
 407                                   OBD_CONNECT_64BITHASH | OBD_CONNECT_MAXBYTES |
 408                                   OBD_CONNECT_EINPROGRESS |
 409                                   OBD_CONNECT_JOBSTATS | OBD_CONNECT_LVB_TYPE |
 410                                   OBD_CONNECT_LAYOUTLOCK |
 411                                   OBD_CONNECT_PINGLESS | OBD_CONNECT_LFSCK |
 412                                   OBD_CONNECT_BULK_MBITS | OBD_CONNECT_SHORTIO |
 413                                   OBD_CONNECT_FLAGS2;
 414
 415 /* The client currently advertises support for OBD_CONNECT_LOCKAHEAD_OLD so it
 416  * can interoperate with an older version of lockahead which was released prior
 417  * to landing in master. This support will be dropped when 2.13 development
 418  * starts.  At the point, we should not just drop the connect flag (below), we
 419  * should also remove the support in the code.
 420  *
 421  * Removing it means a few things:
 422  * 1. Remove this section here
 423  * 2. Remove CEF_NONBLOCK in ll_file_lockahead()
 424  * 3. Remove function exp_connect_lockahead_old
 425  * 4. Remove LDLM_FL_LOCKAHEAD_OLD_RESERVED in lustre_dlm_flags.h
 426  * */
 427 #if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(2, 12, 50, 0)
 428         data->ocd_connect_flags |= OBD_CONNECT_LOCKAHEAD_OLD;
 429 #endif
 430
 431         data->ocd_connect_flags2 = OBD_CONNECT2_LOCKAHEAD;
 432
 433         if (!OBD_FAIL_CHECK(OBD_FAIL_OSC_CONNECT_GRANT_PARAM))
 434                 data->ocd_connect_flags |= OBD_CONNECT_GRANT_PARAM;
 435
 436         /* OBD_CONNECT_CKSUM should always be set, even if checksums are
 437          * disabled by default, because it can still be enabled on the
 438          * fly via /proc. As a consequence, we still need to come to an
 439          * agreement on the supported algorithms at connect time */
 440         data->ocd_connect_flags |= OBD_CONNECT_CKSUM;
 441
 442         if (OBD_FAIL_CHECK(OBD_FAIL_OSC_CKSUM_ADLER_ONLY))
 443                 data->ocd_cksum_types = OBD_CKSUM_ADLER;
 444         else
 445                 data->ocd_cksum_types = cksum_types_supported_client();
 446
 447 #ifdef HAVE_LRU_RESIZE_SUPPORT
 448         data->ocd_connect_flags |= OBD_CONNECT_LRU_RESIZE;
 449 #endif
 450         /* always ping even if server suppress_pings */
 451         if (sbi->ll_flags & LL_SBI_ALWAYS_PING)
 452                 data->ocd_connect_flags &= ~OBD_CONNECT_PINGLESS;
 453
 454         CDEBUG(D_RPCTRACE, "ocd_connect_flags: %#llx ocd_version: %d "
 455                "ocd_grant: %d\n", data->ocd_connect_flags,
 456                data->ocd_version, data->ocd_grant);
 457
 458         obd->obd_upcall.onu_owner = &sbi->ll_lco;
 459         obd->obd_upcall.onu_upcall = cl_ocd_update;
 460
 461         data->ocd_brw_size = DT_MAX_BRW_SIZE;
 462
 463         err = obd_connect(NULL, &sbi->ll_dt_exp, obd, &sbi->ll_sb_uuid, data,
 464                           NULL);
 465         if (err == -EBUSY) {
 466                 LCONSOLE_ERROR_MSG(0x150, "An OST (dt %s) is performing "
 467                                    "recovery, of which this client is not a "
 468                                    "part.  Please wait for recovery to "
 469                                    "complete, abort, or time out.\n", dt);
 470                 GOTO(out_md, err);
 471         } else if (err) {
 472                 CERROR("%s: Cannot connect to %s: rc = %d\n",
 473                        sbi->ll_dt_exp->exp_obd->obd_name, dt, err);
 474                 GOTO(out_md, err);
 475         }
 476
 477         sbi->ll_dt_exp->exp_connect_data = *data;
 478
 479         /* Don't change value if it was specified in the config log */
 480         if (sbi->ll_ra_info.ra_max_read_ahead_whole_pages == -1)
 481                 sbi->ll_ra_info.ra_max_read_ahead_whole_pages =
 482                         max_t(unsigned long, SBI_DEFAULT_READAHEAD_WHOLE_MAX,
 483                               (data->ocd_brw_size >> PAGE_SHIFT));
 484
 485         err = obd_fid_init(sbi->ll_dt_exp->exp_obd, sbi->ll_dt_exp,
 486                            LUSTRE_SEQ_METADATA);
 487         if (err) {
 488                 CERROR("%s: Can't init data layer FID infrastructure, "
 489                        "rc = %d\n", sbi->ll_dt_exp->exp_obd->obd_name, err);
 490                 GOTO(out_dt, err);
 491         }
 492
 493         mutex_lock(&sbi->ll_lco.lco_lock);
 494         sbi->ll_lco.lco_flags = data->ocd_connect_flags;
 495         sbi->ll_lco.lco_md_exp = sbi->ll_md_exp;
 496         sbi->ll_lco.lco_dt_exp = sbi->ll_dt_exp;
 497         mutex_unlock(&sbi->ll_lco.lco_lock);
 498
 499         fid_zero(&sbi->ll_root_fid);
 500         err = md_get_root(sbi->ll_md_exp, get_mount_fileset(sb),
 501                            &sbi->ll_root_fid);
 502         if (err) {
 503                 CERROR("cannot mds_connect: rc = %d\n", err);
 504                 GOTO(out_lock_cn_cb, err);
 505         }
 506         if (!fid_is_sane(&sbi->ll_root_fid)) {
 507                 CERROR("%s: Invalid root fid "DFID" during mount\n",
 508                        sbi->ll_md_exp->exp_obd->obd_name,
 509                        PFID(&sbi->ll_root_fid));
 510                 GOTO(out_lock_cn_cb, err = -EINVAL);
 511         }
 512         CDEBUG(D_SUPER, "rootfid "DFID"\n", PFID(&sbi->ll_root_fid));
 513
 514         sb->s_op = &lustre_super_operations;
 515 #ifdef HAVE_XATTR_HANDLER_FLAGS
 516         sb->s_xattr = ll_xattr_handlers;
 517 #endif
 518 #if THREAD_SIZE >= 8192 /*b=17630*/
 519         sb->s_export_op = &lustre_export_operations;
 520 #endif
 521
 522         /* make root inode
 523          * XXX: move this to after cbd setup? */
 524         valid = OBD_MD_FLGETATTR | OBD_MD_FLBLOCKS | OBD_MD_FLMODEASIZE;
 525         if (sbi->ll_flags & LL_SBI_ACL)
 526                 valid |= OBD_MD_FLACL;
 527
 528         OBD_ALLOC_PTR(op_data);
 529         if (op_data == NULL)
 530                 GOTO(out_lock_cn_cb, err = -ENOMEM);
 531
 532         op_data->op_fid1 = sbi->ll_root_fid;
 533         op_data->op_mode = 0;
 534         op_data->op_valid = valid;
 535
 536         err = md_getattr(sbi->ll_md_exp, op_data, &request);
 537
 538         OBD_FREE_PTR(op_data);
 539         if (err) {
 540                 CERROR("%s: md_getattr failed for root: rc = %d\n",
 541                        sbi->ll_md_exp->exp_obd->obd_name, err);
 542                 GOTO(out_lock_cn_cb, err);
 543         }
 544
 545         err = md_get_lustre_md(sbi->ll_md_exp, request, sbi->ll_dt_exp,
 546                                sbi->ll_md_exp, &lmd);
 547         if (err) {
 548                 CERROR("failed to understand root inode md: rc = %d\n", err);
 549                 ptlrpc_req_finished(request);
 550                 GOTO(out_lock_cn_cb, err);
 551         }
 552
 553         LASSERT(fid_is_sane(&sbi->ll_root_fid));
 554         root = ll_iget(sb, cl_fid_build_ino(&sbi->ll_root_fid,
 555                                             sbi->ll_flags & LL_SBI_32BIT_API),
 556                        &lmd);
 557         md_free_lustre_md(sbi->ll_md_exp, &lmd);
 558         ptlrpc_req_finished(request);
 559
 560         if (IS_ERR(root)) {
 561 #ifdef CONFIG_FS_POSIX_ACL
 562                 if (lmd.posix_acl) {
 563                         posix_acl_release(lmd.posix_acl);
 564                         lmd.posix_acl = NULL;
 565                 }
 566 #endif
 567                 err = IS_ERR(root) ? PTR_ERR(root) : -EBADF;
 568                 root = NULL;
 569                 CERROR("lustre_lite: bad iget4 for root\n");
 570                 GOTO(out_root, err);
 571         }
 572
 573         checksum = sbi->ll_flags & LL_SBI_CHECKSUM;
 574         if (sbi->ll_checksum_set) {
 575                 err = obd_set_info_async(NULL, sbi->ll_dt_exp,
 576                                          sizeof(KEY_CHECKSUM), KEY_CHECKSUM,
 577                                          sizeof(checksum), &checksum, NULL);
 578                 if (err) {
 579                         CERROR("%s: Set checksum failed: rc = %d\n",
 580                                sbi->ll_dt_exp->exp_obd->obd_name, err);
 581                         GOTO(out_root, err);
 582                 }
 583         }
 584         cl_sb_init(sb);
 585
 586         err = obd_set_info_async(NULL, sbi->ll_dt_exp, sizeof(KEY_CACHE_SET),
 587                                  KEY_CACHE_SET, sizeof(*sbi->ll_cache),
 588                                  sbi->ll_cache, NULL);
 589         if (err) {
 590                 CERROR("%s: Set cache_set failed: rc = %d\n",
 591                        sbi->ll_dt_exp->exp_obd->obd_name, err);
 592                 GOTO(out_root, err);
 593         }
 594
 595         sb->s_root = d_make_root(root);
 596         if (sb->s_root == NULL) {
 597                 CERROR("%s: can't make root dentry\n",
 598                         ll_get_fsname(sb, NULL, 0));
 599                 GOTO(out_root, err = -ENOMEM);
 600         }
 601 #ifdef HAVE_DCACHE_LOCK
 602         sb->s_root->d_op = &ll_d_ops;
 603 #endif
 604
 605         sbi->ll_sdev_orig = sb->s_dev;
 606
 607         /* We set sb->s_dev equal on all lustre clients in order to support
 608          * NFS export clustering.  NFSD requires that the FSID be the same
 609          * on all clients. */
 610         /* s_dev is also used in lt_compare() to compare two fs, but that is
 611          * only a node-local comparison. */
 612         uuid = obd_get_uuid(sbi->ll_md_exp);
 613         if (uuid != NULL)
 614                 sb->s_dev = get_uuid2int(uuid->uuid, strlen(uuid->uuid));
 615
 616         if (data != NULL)
 617                 OBD_FREE_PTR(data);
 618         if (osfs != NULL)
 619                 OBD_FREE_PTR(osfs);
 620
 621         if (sbi->ll_proc_root != NULL) {
 622                 err = lprocfs_ll_register_obd(sb, dt);
 623                 if (err < 0) {
 624                         CERROR("%s: could not register %s in llite: rc = %d\n",
 625                                dt, ll_get_fsname(sb, NULL, 0), err);
 626                         err = 0;
 627                 }
 628                 err = lprocfs_ll_register_obd(sb, md);
 629                 if (err < 0) {
 630                         CERROR("%s: could not register %s in llite: rc = %d\n",
 631                                md, ll_get_fsname(sb, NULL, 0), err);
 632                         err = 0;
 633                 }
 634         }
 635
 636         RETURN(err);
 637 out_root:
 638         if (root)
 639                 iput(root);
 640 out_lock_cn_cb:
 641         obd_fid_fini(sbi->ll_dt_exp->exp_obd);
 642 out_dt:
 643         obd_disconnect(sbi->ll_dt_exp);
 644         sbi->ll_dt_exp = NULL;
 645 out_md_fid:
 646         obd_fid_fini(sbi->ll_md_exp->exp_obd);
 647 out_md:
 648         obd_disconnect(sbi->ll_md_exp);
 649         sbi->ll_md_exp = NULL;
 650 out:
 651         if (data != NULL)
 652                 OBD_FREE_PTR(data);
 653         if (osfs != NULL)
 654                 OBD_FREE_PTR(osfs);
 655         return err;
 656 }
 657
 658 int ll_get_max_mdsize(struct ll_sb_info *sbi, int *lmmsize)
 659 {
 660         int size, rc;
 661
 662         size = sizeof(*lmmsize);
 663         rc = obd_get_info(NULL, sbi->ll_dt_exp, sizeof(KEY_MAX_EASIZE),
 664                           KEY_MAX_EASIZE, &size, lmmsize);
 665         if (rc != 0) {
 666                 CERROR("%s: cannot get max LOV EA size: rc = %d\n",
 667                        sbi->ll_dt_exp->exp_obd->obd_name, rc);
 668                 RETURN(rc);
 669         }
 670
 671         size = sizeof(int);
 672         rc = obd_get_info(NULL, sbi->ll_md_exp, sizeof(KEY_MAX_EASIZE),
 673                           KEY_MAX_EASIZE, &size, lmmsize);
 674         if (rc)
 675                 CERROR("Get max mdsize error rc %d\n", rc);
 676
 677         RETURN(rc);
 678 }
 679
 680 /**
 681  * Get the value of the default_easize parameter.
 682  *
 683  * \see client_obd::cl_default_mds_easize
 684  *
 685  * \param[in] sbi       superblock info for this filesystem
 686  * \param[out] lmmsize  pointer to storage location for value
 687  *
 688  * \retval 0            on success
 689  * \retval negative     negated errno on failure
 690  */
 691 int ll_get_default_mdsize(struct ll_sb_info *sbi, int *lmmsize)
 692 {
 693         int size, rc;
 694
 695         size = sizeof(int);
 696         rc = obd_get_info(NULL, sbi->ll_md_exp, sizeof(KEY_DEFAULT_EASIZE),
 697                          KEY_DEFAULT_EASIZE, &size, lmmsize);
 698         if (rc)
 699                 CERROR("Get default mdsize error rc %d\n", rc);
 700
 701         RETURN(rc);
 702 }
 703
 704 /**
 705  * Set the default_easize parameter to the given value.
 706  *
 707  * \see client_obd::cl_default_mds_easize
 708  *
 709  * \param[in] sbi       superblock info for this filesystem
 710  * \param[in] lmmsize   the size to set
 711  *
 712  * \retval 0            on success
 713  * \retval negative     negated errno on failure
 714  */
 715 int ll_set_default_mdsize(struct ll_sb_info *sbi, int lmmsize)
 716 {
 717         int rc;
 718
 719         if (lmmsize < sizeof(struct lov_mds_md) ||
 720             lmmsize > OBD_MAX_DEFAULT_EA_SIZE)
 721                 return -EINVAL;
 722
 723         rc = obd_set_info_async(NULL, sbi->ll_md_exp,
 724                                 sizeof(KEY_DEFAULT_EASIZE), KEY_DEFAULT_EASIZE,
 725                                 sizeof(int), &lmmsize, NULL);
 726
 727         RETURN(rc);
 728 }
 729
 730 static void client_common_put_super(struct super_block *sb)
 731 {
 732         struct ll_sb_info *sbi = ll_s2sbi(sb);
 733         ENTRY;
 734
 735         cl_sb_fini(sb);
 736
 737         obd_fid_fini(sbi->ll_dt_exp->exp_obd);
 738         obd_disconnect(sbi->ll_dt_exp);
 739         sbi->ll_dt_exp = NULL;
 740
 741         lprocfs_ll_unregister_mountpoint(sbi);
 742
 743         obd_fid_fini(sbi->ll_md_exp->exp_obd);
 744         obd_disconnect(sbi->ll_md_exp);
 745         sbi->ll_md_exp = NULL;
 746
 747         EXIT;
 748 }
 749
 750 void ll_kill_super(struct super_block *sb)
 751 {
 752         struct ll_sb_info *sbi;
 753         ENTRY;
 754
 755         /* not init sb ?*/
 756         if (!(sb->s_flags & MS_ACTIVE))
 757                 return;
 758
 759         sbi = ll_s2sbi(sb);
 760         /* we need restore s_dev from changed for clustred NFS before put_super
 761          * because new kernels have cached s_dev and change sb->s_dev in
 762          * put_super not affected real removing devices */
 763         if (sbi) {
 764                 sb->s_dev = sbi->ll_sdev_orig;
 765                 sbi->ll_umounting = 1;
 766
 767                 /* wait running statahead threads to quit */
 768                 while (atomic_read(&sbi->ll_sa_running) > 0) {
 769                         set_current_state(TASK_UNINTERRUPTIBLE);
 770                         schedule_timeout(msecs_to_jiffies(MSEC_PER_SEC >> 3));
 771                 }
 772         }
 773
 774         EXIT;
 775 }
 776
 777 static inline int ll_set_opt(const char *opt, char *data, int fl)
 778 {
 779         if (strncmp(opt, data, strlen(opt)) != 0)
 780                 return 0;
 781         else
 782                 return fl;
 783 }
 784
 785 /* non-client-specific mount options are parsed in lmd_parse */
 786 static int ll_options(char *options, struct ll_sb_info *sbi)
 787 {
 788         int tmp;
 789         char *s1 = options, *s2;
 790         int *flags = &sbi->ll_flags;
 791         ENTRY;
 792
 793         if (!options)
 794                 RETURN(0);
 795
 796         CDEBUG(D_CONFIG, "Parsing opts %s\n", options);
 797
 798         while (*s1) {
 799                 CDEBUG(D_SUPER, "next opt=%s\n", s1);
 800                 tmp = ll_set_opt("nolock", s1, LL_SBI_NOLCK);
 801                 if (tmp) {
 802                         *flags |= tmp;
 803                         goto next;
 804                 }
 805                 tmp = ll_set_opt("flock", s1, LL_SBI_FLOCK);
 806                 if (tmp) {
 807                         *flags |= tmp;
 808                         goto next;
 809                 }
 810                 tmp = ll_set_opt("localflock", s1, LL_SBI_LOCALFLOCK);
 811                 if (tmp) {
 812                         *flags |= tmp;
 813                         goto next;
 814                 }
 815                 tmp = ll_set_opt("noflock", s1, LL_SBI_FLOCK|LL_SBI_LOCALFLOCK);
 816                 if (tmp) {
 817                         *flags &= ~tmp;
 818                         goto next;
 819                 }
 820                 tmp = ll_set_opt("user_xattr", s1, LL_SBI_USER_XATTR);
 821                 if (tmp) {
 822                         *flags |= tmp;
 823                         goto next;
 824                 }
 825                 tmp = ll_set_opt("nouser_xattr", s1, LL_SBI_USER_XATTR);
 826                 if (tmp) {
 827                         *flags &= ~tmp;
 828                         goto next;
 829                 }
 830                 tmp = ll_set_opt("context", s1, 1);
 831                 if (tmp)
 832                         goto next;
 833                 tmp = ll_set_opt("fscontext", s1, 1);
 834                 if (tmp)
 835                         goto next;
 836                 tmp = ll_set_opt("defcontext", s1, 1);
 837                 if (tmp)
 838                         goto next;
 839                 tmp = ll_set_opt("rootcontext", s1, 1);
 840                 if (tmp)
 841                         goto next;
 842                 tmp = ll_set_opt("user_fid2path", s1, LL_SBI_USER_FID2PATH);
 843                 if (tmp) {
 844                         *flags |= tmp;
 845                         goto next;
 846                 }
 847                 tmp = ll_set_opt("nouser_fid2path", s1, LL_SBI_USER_FID2PATH);
 848                 if (tmp) {
 849                         *flags &= ~tmp;
 850                         goto next;
 851                 }
 852
 853                 tmp = ll_set_opt("checksum", s1, LL_SBI_CHECKSUM);
 854                 if (tmp) {
 855                         *flags |= tmp;
 856                         sbi->ll_checksum_set = 1;
 857                         goto next;
 858                 }
 859                 tmp = ll_set_opt("nochecksum", s1, LL_SBI_CHECKSUM);
 860                 if (tmp) {
 861                         *flags &= ~tmp;
 862                         sbi->ll_checksum_set = 1;
 863                         goto next;
 864                 }
 865                 tmp = ll_set_opt("lruresize", s1, LL_SBI_LRU_RESIZE);
 866                 if (tmp) {
 867                         *flags |= tmp;
 868                         goto next;
 869                 }
 870                 tmp = ll_set_opt("nolruresize", s1, LL_SBI_LRU_RESIZE);
 871                 if (tmp) {
 872                         *flags &= ~tmp;
 873                         goto next;
 874                 }
 875                 tmp = ll_set_opt("lazystatfs", s1, LL_SBI_LAZYSTATFS);
 876                 if (tmp) {
 877                         *flags |= tmp;
 878                         goto next;
 879                 }
 880                 tmp = ll_set_opt("nolazystatfs", s1, LL_SBI_LAZYSTATFS);
 881                 if (tmp) {
 882                         *flags &= ~tmp;
 883                         goto next;
 884                 }
 885                 tmp = ll_set_opt("32bitapi", s1, LL_SBI_32BIT_API);
 886                 if (tmp) {
 887                         *flags |= tmp;
 888                         goto next;
 889                 }
 890                 tmp = ll_set_opt("verbose", s1, LL_SBI_VERBOSE);
 891                 if (tmp) {
 892                         *flags |= tmp;
 893                         goto next;
 894                 }
 895                 tmp = ll_set_opt("noverbose", s1, LL_SBI_VERBOSE);
 896                 if (tmp) {
 897                         *flags &= ~tmp;
 898                         goto next;
 899                 }
 900                 tmp = ll_set_opt("always_ping", s1, LL_SBI_ALWAYS_PING);
 901                 if (tmp) {
 902                         *flags |= tmp;
 903                         goto next;
 904                 }
 905                 LCONSOLE_ERROR_MSG(0x152, "Unknown option '%s', won't mount.\n",
 906                                    s1);
 907                 RETURN(-EINVAL);
 908
 909 next:
 910                 /* Find next opt */
 911                 s2 = strchr(s1, ',');
 912                 if (s2 == NULL)
 913                         break;
 914                 s1 = s2 + 1;
 915         }
 916         RETURN(0);
 917 }
 918
 919 void ll_lli_init(struct ll_inode_info *lli)
 920 {
 921         lli->lli_inode_magic = LLI_INODE_MAGIC;
 922         lli->lli_flags = 0;
 923         spin_lock_init(&lli->lli_lock);
 924         lli->lli_posix_acl = NULL;
 925         /* Do not set lli_fid, it has been initialized already. */
 926         fid_zero(&lli->lli_pfid);
 927         lli->lli_mds_read_och = NULL;
 928         lli->lli_mds_write_och = NULL;
 929         lli->lli_mds_exec_och = NULL;
 930         lli->lli_open_fd_read_count = 0;
 931         lli->lli_open_fd_write_count = 0;
 932         lli->lli_open_fd_exec_count = 0;
 933         mutex_init(&lli->lli_och_mutex);
 934         spin_lock_init(&lli->lli_agl_lock);
 935         spin_lock_init(&lli->lli_layout_lock);
 936         ll_layout_version_set(lli, CL_LAYOUT_GEN_NONE);
 937         lli->lli_clob = NULL;
 938
 939         init_rwsem(&lli->lli_xattrs_list_rwsem);
 940         mutex_init(&lli->lli_xattrs_enq_lock);
 941
 942         LASSERT(lli->lli_vfs_inode.i_mode != 0);
 943         if (S_ISDIR(lli->lli_vfs_inode.i_mode)) {
 944                 mutex_init(&lli->lli_readdir_mutex);
 945                 lli->lli_opendir_key = NULL;
 946                 lli->lli_sai = NULL;
 947                 spin_lock_init(&lli->lli_sa_lock);
 948                 lli->lli_opendir_pid = 0;
 949                 lli->lli_sa_enabled = 0;
 950                 lli->lli_def_stripe_offset = -1;
 951         } else {
 952                 mutex_init(&lli->lli_size_mutex);
 953                 lli->lli_symlink_name = NULL;
 954                 init_rwsem(&lli->lli_trunc_sem);
 955                 range_lock_tree_init(&lli->lli_write_tree);
 956                 init_rwsem(&lli->lli_glimpse_sem);
 957                 lli->lli_glimpse_time = ktime_set(0, 0);
 958                 INIT_LIST_HEAD(&lli->lli_agl_list);
 959                 lli->lli_agl_index = 0;
 960                 lli->lli_async_rc = 0;
 961         }
 962         mutex_init(&lli->lli_layout_mutex);
 963         memset(lli->lli_jobid, 0, sizeof(lli->lli_jobid));
 964 }
 965
 966 #ifndef HAVE_SUPER_SETUP_BDI_NAME
 967
 968 #define LSI_BDI_INITIALIZED     0x00400000
 969
 970 #ifndef HAVE_BDI_CAP_MAP_COPY
 971 # define BDI_CAP_MAP_COPY       0
 972 #endif
 973
 974 #define MAX_STRING_SIZE 128
 975
 976 static int super_setup_bdi_name(struct super_block *sb, char *fmt, ...)
 977 {
 978         struct  lustre_sb_info *lsi = s2lsi(sb);
 979         char buf[MAX_STRING_SIZE];
 980         va_list args;
 981         int err;
 982
 983         err = bdi_init(&lsi->lsi_bdi);
 984         if (err)
 985                 return err;
 986
 987         lsi->lsi_flags |= LSI_BDI_INITIALIZED;
 988         lsi->lsi_bdi.capabilities = BDI_CAP_MAP_COPY;
 989         lsi->lsi_bdi.name = "lustre";
 990         va_start(args, fmt);
 991         vsnprintf(buf, MAX_STRING_SIZE, fmt, args);
 992         va_end(args);
 993         err = bdi_register(&lsi->lsi_bdi, NULL, "%s", buf);
 994         va_end(args);
 995         if (!err)
 996                 sb->s_bdi = &lsi->lsi_bdi;
 997
 998         return err;
 999 }
1000 #endif /* !HAVE_SUPER_SETUP_BDI_NAME */
1001
1002 int ll_fill_super(struct super_block *sb, struct vfsmount *mnt)
1003 {
1004         struct  lustre_profile *lprof = NULL;
1005         struct  lustre_sb_info *lsi = s2lsi(sb);
1006         struct  ll_sb_info *sbi = NULL;
1007         char    *dt = NULL, *md = NULL;
1008         char    *profilenm = get_profile_name(sb);
1009         struct config_llog_instance *cfg;
1010         /* %p for void* in printf needs 16+2 characters: 0xffffffffffffffff */
1011         const int instlen = sizeof(cfg->cfg_instance) * 2 + 2;
1012         int     md_len = 0;
1013         int     dt_len = 0;
1014         int     err;
1015         ENTRY;
1016
1017         CDEBUG(D_VFSTRACE, "VFS Op: sb %p\n", sb);
1018
1019         try_module_get(THIS_MODULE);
1020
1021         OBD_ALLOC_PTR(cfg);
1022         if (cfg == NULL)
1023                 GOTO(out_free, err = -ENOMEM);
1024
1025         /* client additional sb info */
1026         lsi->lsi_llsbi = sbi = ll_init_sbi();
1027         if (!sbi)
1028                 GOTO(out_free, err = -ENOMEM);
1029
1030         err = ll_options(lsi->lsi_lmd->lmd_opts, sbi);
1031         if (err)
1032                 GOTO(out_free, err);
1033
1034         err = super_setup_bdi_name(sb, "lustre-%p", sb);
1035         if (err)
1036                 GOTO(out_free, err);
1037
1038 #ifndef HAVE_DCACHE_LOCK
1039         /* kernel >= 2.6.38 store dentry operations in sb->s_d_op. */
1040         sb->s_d_op = &ll_d_ops;
1041 #endif
1042
1043         /* Call lprocfs_ll_register_mountpoint() before lustre_process_log()
1044          * so that "llite.*.*" params can be processed correctly. */
1045         if (proc_lustre_fs_root != NULL) {
1046                 err = lprocfs_ll_register_mountpoint(proc_lustre_fs_root, sb);
1047                 if (err < 0) {
1048                         CERROR("%s: could not register mountpoint in llite: "
1049                                "rc = %d\n", ll_get_fsname(sb, NULL, 0), err);
1050                         err = 0;
1051                 }
1052         }
1053
1054         /* Generate a string unique to this super, in case some joker tries
1055            to mount the same fs at two mount points.
1056            Use the address of the super itself.*/
1057         cfg->cfg_instance = sb;
1058         cfg->cfg_uuid = lsi->lsi_llsbi->ll_sb_uuid;
1059         cfg->cfg_callback = class_config_llog_handler;
1060         cfg->cfg_sub_clds = CONFIG_SUB_CLIENT;
1061         /* set up client obds */
1062         err = lustre_process_log(sb, profilenm, cfg);
1063         if (err < 0)
1064                 GOTO(out_proc, err);
1065
1066         /* Profile set with LCFG_MOUNTOPT so we can find our mdc and osc obds */
1067         lprof = class_get_profile(profilenm);
1068         if (lprof == NULL) {
1069                 LCONSOLE_ERROR_MSG(0x156, "The client profile '%s' could not be"
1070                                    " read from the MGS.  Does that filesystem "
1071                                    "exist?\n", profilenm);
1072                 GOTO(out_proc, err = -EINVAL);
1073         }
1074         CDEBUG(D_CONFIG, "Found profile %s: mdc=%s osc=%s\n", profilenm,
1075                lprof->lp_md, lprof->lp_dt);
1076
1077         dt_len = strlen(lprof->lp_dt) + instlen + 2;
1078         OBD_ALLOC(dt, dt_len);
1079         if (!dt)
1080                 GOTO(out_proc, err = -ENOMEM);
1081         snprintf(dt, dt_len - 1, "%s-%p", lprof->lp_dt, cfg->cfg_instance);
1082
1083         md_len = strlen(lprof->lp_md) + instlen + 2;
1084         OBD_ALLOC(md, md_len);
1085         if (!md)
1086                 GOTO(out_proc, err = -ENOMEM);
1087         snprintf(md, md_len - 1, "%s-%p", lprof->lp_md, cfg->cfg_instance);
1088
1089         /* connections, registrations, sb setup */
1090         err = client_common_fill_super(sb, md, dt, mnt);
1091         if (err < 0)
1092                 GOTO(out_proc, err);
1093
1094         sbi->ll_client_common_fill_super_succeeded = 1;
1095
1096 out_proc:
1097         if (err < 0)
1098                 lprocfs_ll_unregister_mountpoint(sbi);
1099 out_free:
1100         if (md)
1101                 OBD_FREE(md, md_len);
1102         if (dt)
1103                 OBD_FREE(dt, dt_len);
1104         if (lprof != NULL)
1105                 class_put_profile(lprof);
1106         if (cfg)
1107                 OBD_FREE_PTR(cfg);
1108         if (err)
1109                 ll_put_super(sb);
1110         else if (sbi->ll_flags & LL_SBI_VERBOSE)
1111                 LCONSOLE_WARN("Mounted %s\n", profilenm);
1112         RETURN(err);
1113 } /* ll_fill_super */
1114
1115 void ll_put_super(struct super_block *sb)
1116 {
1117         struct config_llog_instance cfg, params_cfg;
1118         struct obd_device *obd;
1119         struct lustre_sb_info *lsi = s2lsi(sb);
1120         struct ll_sb_info *sbi = ll_s2sbi(sb);
1121         char *profilenm = get_profile_name(sb);
1122         long ccc_count;
1123         int next, force = 1, rc = 0;
1124         ENTRY;
1125
1126         if (!sbi)
1127                 GOTO(out_no_sbi, 0);
1128
1129         CDEBUG(D_VFSTRACE, "VFS Op: sb %p - %s\n", sb, profilenm);
1130
1131         cfg.cfg_instance = sb;
1132         lustre_end_log(sb, profilenm, &cfg);
1133
1134         params_cfg.cfg_instance = sb;
1135         lustre_end_log(sb, PARAMS_FILENAME, &params_cfg);
1136
1137         if (sbi->ll_md_exp) {
1138                 obd = class_exp2obd(sbi->ll_md_exp);
1139                 if (obd)
1140                         force = obd->obd_force;
1141         }
1142
1143         /* Wait for unstable pages to be committed to stable storage */
1144         if (force == 0) {
1145                 struct l_wait_info lwi = LWI_INTR(LWI_ON_SIGNAL_NOOP, NULL);
1146                 rc = l_wait_event(sbi->ll_cache->ccc_unstable_waitq,
1147                         atomic_long_read(&sbi->ll_cache->ccc_unstable_nr) == 0,
1148                         &lwi);
1149         }
1150
1151         ccc_count = atomic_long_read(&sbi->ll_cache->ccc_unstable_nr);
1152         if (force == 0 && rc != -EINTR)
1153                 LASSERTF(ccc_count == 0, "count: %li\n", ccc_count);
1154
1155
1156         /* We need to set force before the lov_disconnect in
1157            lustre_common_put_super, since l_d cleans up osc's as well. */
1158         if (force) {
1159                 next = 0;
1160                 while ((obd = class_devices_in_group(&sbi->ll_sb_uuid,
1161                                                      &next)) != NULL) {
1162                         obd->obd_force = force;
1163                 }
1164         }
1165
1166         if (sbi->ll_client_common_fill_super_succeeded) {
1167                 /* Only if client_common_fill_super succeeded */
1168                 client_common_put_super(sb);
1169         }
1170
1171         next = 0;
1172         while ((obd = class_devices_in_group(&sbi->ll_sb_uuid, &next)) !=NULL) {
1173                 class_manual_cleanup(obd);
1174         }
1175
1176         if (sbi->ll_flags & LL_SBI_VERBOSE)
1177                 LCONSOLE_WARN("Unmounted %s\n", profilenm ? profilenm : "");
1178
1179         if (profilenm)
1180                 class_del_profile(profilenm);
1181
1182 #ifndef HAVE_SUPER_SETUP_BDI_NAME
1183         if (lsi->lsi_flags & LSI_BDI_INITIALIZED) {
1184                 bdi_destroy(&lsi->lsi_bdi);
1185                 lsi->lsi_flags &= ~LSI_BDI_INITIALIZED;
1186         }
1187 #endif
1188
1189         ll_free_sbi(sb);
1190         lsi->lsi_llsbi = NULL;
1191 out_no_sbi:
1192         lustre_common_put_super(sb);
1193
1194         cl_env_cache_purge(~0);
1195
1196         module_put(THIS_MODULE);
1197
1198         EXIT;
1199 } /* client_put_super */
1200
1201 struct inode *ll_inode_from_resource_lock(struct ldlm_lock *lock)
1202 {
1203         struct inode *inode = NULL;
1204
1205         /* NOTE: we depend on atomic igrab() -bzzz */
1206         lock_res_and_lock(lock);
1207         if (lock->l_resource->lr_lvb_inode) {
1208                 struct ll_inode_info * lli;
1209                 lli = ll_i2info(lock->l_resource->lr_lvb_inode);
1210                 if (lli->lli_inode_magic == LLI_INODE_MAGIC) {
1211                         inode = igrab(lock->l_resource->lr_lvb_inode);
1212                 } else {
1213                         inode = lock->l_resource->lr_lvb_inode;
1214                         LDLM_DEBUG_LIMIT(inode->i_state & I_FREEING ?  D_INFO :
1215                                          D_WARNING, lock, "lr_lvb_inode %p is "
1216                                          "bogus: magic %08x",
1217                                          lock->l_resource->lr_lvb_inode,
1218                                          lli->lli_inode_magic);
1219                         inode = NULL;
1220                 }
1221         }
1222         unlock_res_and_lock(lock);
1223         return inode;
1224 }
1225
1226 void ll_dir_clear_lsm_md(struct inode *inode)
1227 {
1228         struct ll_inode_info *lli = ll_i2info(inode);
1229
1230         LASSERT(S_ISDIR(inode->i_mode));
1231
1232         if (lli->lli_lsm_md != NULL) {
1233                 lmv_free_memmd(lli->lli_lsm_md);
1234                 lli->lli_lsm_md = NULL;
1235         }
1236 }
1237
1238 static struct inode *ll_iget_anon_dir(struct super_block *sb,
1239                                       const struct lu_fid *fid,
1240                                       struct lustre_md *md)
1241 {
1242         struct ll_sb_info       *sbi = ll_s2sbi(sb);
1243         struct mdt_body         *body = md->body;
1244         struct inode            *inode;
1245         ino_t                   ino;
1246         ENTRY;
1247
1248         ino = cl_fid_build_ino(fid, sbi->ll_flags & LL_SBI_32BIT_API);
1249         inode = iget_locked(sb, ino);
1250         if (inode == NULL) {
1251                 CERROR("%s: failed get simple inode "DFID": rc = -ENOENT\n",
1252                        ll_get_fsname(sb, NULL, 0), PFID(fid));
1253                 RETURN(ERR_PTR(-ENOENT));
1254         }
1255
1256         if (inode->i_state & I_NEW) {
1257                 struct ll_inode_info *lli = ll_i2info(inode);
1258                 struct lmv_stripe_md *lsm = md->lmv;
1259
1260                 inode->i_mode = (inode->i_mode & ~S_IFMT) |
1261                                 (body->mbo_mode & S_IFMT);
1262                 LASSERTF(S_ISDIR(inode->i_mode), "Not slave inode "DFID"\n",
1263                          PFID(fid));
1264
1265                 LTIME_S(inode->i_mtime) = 0;
1266                 LTIME_S(inode->i_atime) = 0;
1267                 LTIME_S(inode->i_ctime) = 0;
1268                 inode->i_rdev = 0;
1269
1270 #ifdef HAVE_BACKING_DEV_INFO
1271                 /* initializing backing dev info. */
1272                 inode->i_mapping->backing_dev_info =
1273                                                 &s2lsi(inode->i_sb)->lsi_bdi;
1274 #endif
1275                 inode->i_op = &ll_dir_inode_operations;
1276                 inode->i_fop = &ll_dir_operations;
1277                 lli->lli_fid = *fid;
1278                 ll_lli_init(lli);
1279
1280                 LASSERT(lsm != NULL);
1281                 /* master object FID */
1282                 lli->lli_pfid = body->mbo_fid1;
1283                 CDEBUG(D_INODE, "lli %p slave "DFID" master "DFID"\n",
1284                        lli, PFID(fid), PFID(&lli->lli_pfid));
1285                 unlock_new_inode(inode);
1286         }
1287
1288         RETURN(inode);
1289 }
1290
1291 static int ll_init_lsm_md(struct inode *inode, struct lustre_md *md)
1292 {
1293         struct lu_fid *fid;
1294         struct lmv_stripe_md *lsm = md->lmv;
1295         int i;
1296
1297         LASSERT(lsm != NULL);
1298         /* XXX sigh, this lsm_root initialization should be in
1299          * LMV layer, but it needs ll_iget right now, so we
1300          * put this here right now. */
1301         for (i = 0; i < lsm->lsm_md_stripe_count; i++) {
1302                 fid = &lsm->lsm_md_oinfo[i].lmo_fid;
1303                 LASSERT(lsm->lsm_md_oinfo[i].lmo_root == NULL);
1304                 /* Unfortunately ll_iget will call ll_update_inode,
1305                  * where the initialization of slave inode is slightly
1306                  * different, so it reset lsm_md to NULL to avoid
1307                  * initializing lsm for slave inode. */
1308                 /* For migrating inode, master stripe and master object will
1309                  * be same, so we only need assign this inode */
1310                 if (lsm->lsm_md_hash_type & LMV_HASH_FLAG_MIGRATION && i == 0)
1311                         lsm->lsm_md_oinfo[i].lmo_root = inode;
1312                 else
1313                         lsm->lsm_md_oinfo[i].lmo_root =
1314                                 ll_iget_anon_dir(inode->i_sb, fid, md);
1315
1316                 if (IS_ERR(lsm->lsm_md_oinfo[i].lmo_root)) {
1317                         int rc = PTR_ERR(lsm->lsm_md_oinfo[i].lmo_root);
1318
1319                         lsm->lsm_md_oinfo[i].lmo_root = NULL;
1320                         return rc;
1321                 }
1322         }
1323
1324         return 0;
1325 }
1326
1327 static inline int lli_lsm_md_eq(const struct lmv_stripe_md *lsm_md1,
1328                                 const struct lmv_stripe_md *lsm_md2)
1329 {
1330         return lsm_md1->lsm_md_magic == lsm_md2->lsm_md_magic &&
1331                lsm_md1->lsm_md_stripe_count == lsm_md2->lsm_md_stripe_count &&
1332                lsm_md1->lsm_md_master_mdt_index ==
1333                                         lsm_md2->lsm_md_master_mdt_index &&
1334                lsm_md1->lsm_md_hash_type == lsm_md2->lsm_md_hash_type &&
1335                lsm_md1->lsm_md_layout_version ==
1336                                         lsm_md2->lsm_md_layout_version &&
1337                strcmp(lsm_md1->lsm_md_pool_name,
1338                       lsm_md2->lsm_md_pool_name) == 0;
1339 }
1340
1341 static int ll_update_lsm_md(struct inode *inode, struct lustre_md *md)
1342 {
1343         struct ll_inode_info *lli = ll_i2info(inode);
1344         struct lmv_stripe_md *lsm = md->lmv;
1345         int     rc;
1346         ENTRY;
1347
1348         LASSERT(S_ISDIR(inode->i_mode));
1349         CDEBUG(D_INODE, "update lsm %p of "DFID"\n", lli->lli_lsm_md,
1350                PFID(ll_inode2fid(inode)));
1351
1352         /* no striped information from request. */
1353         if (lsm == NULL) {
1354                 if (lli->lli_lsm_md == NULL) {
1355                         RETURN(0);
1356                 } else if (lli->lli_lsm_md->lsm_md_hash_type &
1357                                                 LMV_HASH_FLAG_MIGRATION) {
1358                         /* migration is done, the temporay MIGRATE layout has
1359                          * been removed */
1360                         CDEBUG(D_INODE, DFID" finish migration.\n",
1361                                PFID(ll_inode2fid(inode)));
1362                         lmv_free_memmd(lli->lli_lsm_md);
1363                         lli->lli_lsm_md = NULL;
1364                         RETURN(0);
1365                 } else {
1366                         /* The lustre_md from req does not include stripeEA,
1367                          * see ll_md_setattr */
1368                         RETURN(0);
1369                 }
1370         }
1371
1372         /* set the directory layout */
1373         if (lli->lli_lsm_md == NULL) {
1374                 struct cl_attr  *attr;
1375
1376                 rc = ll_init_lsm_md(inode, md);
1377                 if (rc != 0)
1378                         RETURN(rc);
1379
1380                 /* set md->lmv to NULL, so the following free lustre_md
1381                  * will not free this lsm */
1382                 md->lmv = NULL;
1383                 lli->lli_lsm_md = lsm;
1384
1385                 OBD_ALLOC_PTR(attr);
1386                 if (attr == NULL)
1387                         RETURN(-ENOMEM);
1388
1389                 /* validate the lsm */
1390                 rc = md_merge_attr(ll_i2mdexp(inode), lsm, attr,
1391                                    ll_md_blocking_ast);
1392                 if (rc != 0) {
1393                         OBD_FREE_PTR(attr);
1394                         RETURN(rc);
1395                 }
1396
1397                 if (md->body->mbo_valid & OBD_MD_FLNLINK)
1398                         md->body->mbo_nlink = attr->cat_nlink;
1399                 if (md->body->mbo_valid & OBD_MD_FLSIZE)
1400                         md->body->mbo_size = attr->cat_size;
1401                 if (md->body->mbo_valid & OBD_MD_FLATIME)
1402                         md->body->mbo_atime = attr->cat_atime;
1403                 if (md->body->mbo_valid & OBD_MD_FLCTIME)
1404                         md->body->mbo_ctime = attr->cat_ctime;
1405                 if (md->body->mbo_valid & OBD_MD_FLMTIME)
1406                         md->body->mbo_mtime = attr->cat_mtime;
1407
1408                 OBD_FREE_PTR(attr);
1409
1410                 CDEBUG(D_INODE, "Set lsm %p magic %x to "DFID"\n", lsm,
1411                        lsm->lsm_md_magic, PFID(ll_inode2fid(inode)));
1412                 RETURN(0);
1413         }
1414
1415         /* Compare the old and new stripe information */
1416         if (!lsm_md_eq(lli->lli_lsm_md, lsm)) {
1417                 struct lmv_stripe_md    *old_lsm = lli->lli_lsm_md;
1418                 int                     idx;
1419
1420                 CERROR("%s: inode "DFID"(%p)'s lmv layout mismatch (%p)/(%p)"
1421                        "magic:0x%x/0x%x stripe count: %d/%d master_mdt: %d/%d"
1422                        "hash_type:0x%x/0x%x layout: 0x%x/0x%x pool:%s/%s\n",
1423                        ll_get_fsname(inode->i_sb, NULL, 0), PFID(&lli->lli_fid),
1424                        inode, lsm, old_lsm,
1425                        lsm->lsm_md_magic, old_lsm->lsm_md_magic,
1426                        lsm->lsm_md_stripe_count,
1427                        old_lsm->lsm_md_stripe_count,
1428                        lsm->lsm_md_master_mdt_index,
1429                        old_lsm->lsm_md_master_mdt_index,
1430                        lsm->lsm_md_hash_type, old_lsm->lsm_md_hash_type,
1431                        lsm->lsm_md_layout_version,
1432                        old_lsm->lsm_md_layout_version,
1433                        lsm->lsm_md_pool_name,
1434                        old_lsm->lsm_md_pool_name);
1435
1436                 for (idx = 0; idx < old_lsm->lsm_md_stripe_count; idx++) {
1437                         CERROR("%s: sub FIDs in old lsm idx %d, old: "DFID"\n",
1438                                ll_get_fsname(inode->i_sb, NULL, 0), idx,
1439                                PFID(&old_lsm->lsm_md_oinfo[idx].lmo_fid));
1440                 }
1441
1442                 for (idx = 0; idx < lsm->lsm_md_stripe_count; idx++) {
1443                         CERROR("%s: sub FIDs in new lsm idx %d, new: "DFID"\n",
1444                                ll_get_fsname(inode->i_sb, NULL, 0), idx,
1445                                PFID(&lsm->lsm_md_oinfo[idx].lmo_fid));
1446                 }
1447
1448                 RETURN(-EIO);
1449         }
1450
1451         RETURN(0);
1452 }
1453
1454 void ll_clear_inode(struct inode *inode)
1455 {
1456         struct ll_inode_info *lli = ll_i2info(inode);
1457         struct ll_sb_info *sbi = ll_i2sbi(inode);
1458         ENTRY;
1459
1460         CDEBUG(D_VFSTRACE, "VFS Op:inode="DFID"(%p)\n",
1461                PFID(ll_inode2fid(inode)), inode);
1462
1463         if (S_ISDIR(inode->i_mode)) {
1464                 /* these should have been cleared in ll_file_release */
1465                 LASSERT(lli->lli_opendir_key == NULL);
1466                 LASSERT(lli->lli_sai == NULL);
1467                 LASSERT(lli->lli_opendir_pid == 0);
1468         }
1469
1470         md_null_inode(sbi->ll_md_exp, ll_inode2fid(inode));
1471
1472         LASSERT(!lli->lli_open_fd_write_count);
1473         LASSERT(!lli->lli_open_fd_read_count);
1474         LASSERT(!lli->lli_open_fd_exec_count);
1475
1476         if (lli->lli_mds_write_och)
1477                 ll_md_real_close(inode, FMODE_WRITE);
1478         if (lli->lli_mds_exec_och)
1479                 ll_md_real_close(inode, FMODE_EXEC);
1480         if (lli->lli_mds_read_och)
1481                 ll_md_real_close(inode, FMODE_READ);
1482
1483         if (S_ISLNK(inode->i_mode) && lli->lli_symlink_name) {
1484                 OBD_FREE(lli->lli_symlink_name,
1485                          strlen(lli->lli_symlink_name) + 1);
1486                 lli->lli_symlink_name = NULL;
1487         }
1488
1489         ll_xattr_cache_destroy(inode);
1490
1491 #ifdef CONFIG_FS_POSIX_ACL
1492         forget_all_cached_acls(inode);
1493         if (lli->lli_posix_acl) {
1494                 LASSERT(atomic_read(&lli->lli_posix_acl->a_refcount) == 1);
1495                 posix_acl_release(lli->lli_posix_acl);
1496                 lli->lli_posix_acl = NULL;
1497         }
1498 #endif
1499         lli->lli_inode_magic = LLI_INODE_DEAD;
1500
1501         if (S_ISDIR(inode->i_mode))
1502                 ll_dir_clear_lsm_md(inode);
1503         else if (S_ISREG(inode->i_mode) && !is_bad_inode(inode))
1504                 LASSERT(list_empty(&lli->lli_agl_list));
1505
1506         /*
1507          * XXX This has to be done before lsm is freed below, because
1508          * cl_object still uses inode lsm.
1509          */
1510         cl_inode_fini(inode);
1511
1512         EXIT;
1513 }
1514
1515 static int ll_md_setattr(struct dentry *dentry, struct md_op_data *op_data)
1516 {
1517         struct lustre_md md;
1518         struct inode *inode = dentry->d_inode;
1519         struct ll_sb_info *sbi = ll_i2sbi(inode);
1520         struct ptlrpc_request *request = NULL;
1521         int rc, ia_valid;
1522         ENTRY;
1523
1524         op_data = ll_prep_md_op_data(op_data, inode, NULL, NULL, 0, 0,
1525                                      LUSTRE_OPC_ANY, NULL);
1526         if (IS_ERR(op_data))
1527                 RETURN(PTR_ERR(op_data));
1528
1529         rc = md_setattr(sbi->ll_md_exp, op_data, NULL, 0, &request);
1530         if (rc) {
1531                 ptlrpc_req_finished(request);
1532                 if (rc == -ENOENT) {
1533                         clear_nlink(inode);
1534                         /* Unlinked special device node? Or just a race?
1535                          * Pretend we done everything. */
1536                         if (!S_ISREG(inode->i_mode) &&
1537                             !S_ISDIR(inode->i_mode)) {
1538                                 ia_valid = op_data->op_attr.ia_valid;
1539                                 op_data->op_attr.ia_valid &= ~TIMES_SET_FLAGS;
1540                                 rc = simple_setattr(dentry, &op_data->op_attr);
1541                                 op_data->op_attr.ia_valid = ia_valid;
1542                         }
1543                 } else if (rc != -EPERM && rc != -EACCES && rc != -ETXTBSY) {
1544                         CERROR("md_setattr fails: rc = %d\n", rc);
1545                 }
1546                 RETURN(rc);
1547         }
1548
1549         rc = md_get_lustre_md(sbi->ll_md_exp, request, sbi->ll_dt_exp,
1550                               sbi->ll_md_exp, &md);
1551         if (rc) {
1552                 ptlrpc_req_finished(request);
1553                 RETURN(rc);
1554         }
1555
1556         ia_valid = op_data->op_attr.ia_valid;
1557         /* inode size will be in ll_setattr_ost, can't do it now since dirty
1558          * cache is not cleared yet. */
1559         op_data->op_attr.ia_valid &= ~(TIMES_SET_FLAGS | ATTR_SIZE);
1560         if (S_ISREG(inode->i_mode))
1561                 inode_lock(inode);
1562         rc = simple_setattr(dentry, &op_data->op_attr);
1563         if (S_ISREG(inode->i_mode))
1564                 inode_unlock(inode);
1565         op_data->op_attr.ia_valid = ia_valid;
1566
1567         rc = ll_update_inode(inode, &md);
1568         ptlrpc_req_finished(request);
1569
1570         RETURN(rc);
1571 }
1572
1573 /* If this inode has objects allocated to it (lsm != NULL), then the OST
1574  * object(s) determine the file size and mtime.  Otherwise, the MDS will
1575  * keep these values until such a time that objects are allocated for it.
1576  * We do the MDS operations first, as it is checking permissions for us.
1577  * We don't to the MDS RPC if there is nothing that we want to store there,
1578  * otherwise there is no harm in updating mtime/atime on the MDS if we are
1579  * going to do an RPC anyways.
1580  *
1581  * If we are doing a truncate, we will send the mtime and ctime updates
1582  * to the OST with the punch RPC, otherwise we do an explicit setattr RPC.
1583  * I don't believe it is possible to get e.g. ATTR_MTIME_SET and ATTR_SIZE
1584  * at the same time.
1585  *
1586  * In case of HSMimport, we only set attr on MDS.
1587  */
1588 int ll_setattr_raw(struct dentry *dentry, struct iattr *attr, bool hsm_import)
1589 {
1590         struct inode *inode = dentry->d_inode;
1591         struct ll_inode_info *lli = ll_i2info(inode);
1592         struct md_op_data *op_data = NULL;
1593         int rc = 0;
1594         ENTRY;
1595
1596         CDEBUG(D_VFSTRACE, "%s: setattr inode "DFID"(%p) from %llu to %llu, "
1597                "valid %x, hsm_import %d\n",
1598                ll_get_fsname(inode->i_sb, NULL, 0), PFID(&lli->lli_fid),
1599                inode, i_size_read(inode), attr->ia_size, attr->ia_valid,
1600                hsm_import);
1601
1602         if (attr->ia_valid & ATTR_SIZE) {
1603                 /* Check new size against VFS/VM file size limit and rlimit */
1604                 rc = inode_newsize_ok(inode, attr->ia_size);
1605                 if (rc)
1606                         RETURN(rc);
1607
1608                 /* The maximum Lustre file size is variable, based on the
1609                  * OST maximum object size and number of stripes.  This
1610                  * needs another check in addition to the VFS check above. */
1611                 if (attr->ia_size > ll_file_maxbytes(inode)) {
1612                         CDEBUG(D_INODE,"file "DFID" too large %llu > %llu\n",
1613                                PFID(&lli->lli_fid), attr->ia_size,
1614                                ll_file_maxbytes(inode));
1615                         RETURN(-EFBIG);
1616                 }
1617
1618                 attr->ia_valid |= ATTR_MTIME | ATTR_CTIME;
1619         }
1620
1621         /* POSIX: check before ATTR_*TIME_SET set (from inode_change_ok) */
1622         if (attr->ia_valid & TIMES_SET_FLAGS) {
1623                 if ((!uid_eq(current_fsuid(), inode->i_uid)) &&
1624                     !cfs_capable(CFS_CAP_FOWNER))
1625                         RETURN(-EPERM);
1626         }
1627
1628         /* We mark all of the fields "set" so MDS/OST does not re-set them */
1629         if (!(attr->ia_valid & ATTR_CTIME_SET) &&
1630             (attr->ia_valid & ATTR_CTIME)) {
1631                 attr->ia_ctime = current_time(inode);
1632                 attr->ia_valid |= ATTR_CTIME_SET;
1633         }
1634         if (!(attr->ia_valid & ATTR_ATIME_SET) &&
1635             (attr->ia_valid & ATTR_ATIME)) {
1636                 attr->ia_atime = current_time(inode);
1637                 attr->ia_valid |= ATTR_ATIME_SET;
1638         }
1639         if (!(attr->ia_valid & ATTR_MTIME_SET) &&
1640             (attr->ia_valid & ATTR_MTIME)) {
1641                 attr->ia_mtime = current_time(inode);
1642                 attr->ia_valid |= ATTR_MTIME_SET;
1643         }
1644
1645         if (attr->ia_valid & (ATTR_MTIME | ATTR_CTIME))
1646                 CDEBUG(D_INODE, "setting mtime %lu, ctime %lu, now = %llu\n",
1647                        LTIME_S(attr->ia_mtime), LTIME_S(attr->ia_ctime),
1648                        (s64)ktime_get_real_seconds());
1649
1650         if (S_ISREG(inode->i_mode)) {
1651                 if (attr->ia_valid & ATTR_SIZE)
1652                         inode_dio_write_done(inode);
1653                 inode_unlock(inode);
1654         }
1655
1656         /* We always do an MDS RPC, even if we're only changing the size;
1657          * only the MDS knows whether truncate() should fail with -ETXTBUSY */
1658
1659         OBD_ALLOC_PTR(op_data);
1660         if (op_data == NULL)
1661                 GOTO(out, rc = -ENOMEM);
1662
1663         if (!hsm_import && attr->ia_valid & ATTR_SIZE) {
1664                 /* If we are changing file size, file content is
1665                  * modified, flag it. */
1666                 attr->ia_valid |= MDS_OPEN_OWNEROVERRIDE;
1667                 op_data->op_bias |= MDS_DATA_MODIFIED;
1668                 ll_file_clear_flag(lli, LLIF_DATA_MODIFIED);
1669         }
1670
1671         op_data->op_attr = *attr;
1672
1673         rc = ll_md_setattr(dentry, op_data);
1674         if (rc)
1675                 GOTO(out, rc);
1676
1677         if (!S_ISREG(inode->i_mode) || hsm_import)
1678                 GOTO(out, rc = 0);
1679
1680         if (attr->ia_valid & (ATTR_SIZE |
1681                               ATTR_ATIME | ATTR_ATIME_SET |
1682                               ATTR_MTIME | ATTR_MTIME_SET |
1683                               ATTR_CTIME | ATTR_CTIME_SET)) {
1684                 /* For truncate and utimes sending attributes to OSTs, setting
1685                  * mtime/atime to the past will be performed under PW [0:EOF]
1686                  * extent lock (new_size:EOF for truncate).  It may seem
1687                  * excessive to send mtime/atime updates to OSTs when not
1688                  * setting times to past, but it is necessary due to possible
1689                  * time de-synchronization between MDT inode and OST objects */
1690                 rc = cl_setattr_ost(lli->lli_clob, attr, 0);
1691         }
1692
1693         /* If the file was restored, it needs to set dirty flag.
1694          *
1695          * We've already sent MDS_DATA_MODIFIED flag in
1696          * ll_md_setattr() for truncate. However, the MDT refuses to
1697          * set the HS_DIRTY flag on released files, so we have to set
1698          * it again if the file has been restored. Please check how
1699          * LLIF_DATA_MODIFIED is set in vvp_io_setattr_fini().
1700          *
1701          * Please notice that if the file is not released, the previous
1702          * MDS_DATA_MODIFIED has taken effect and usually
1703          * LLIF_DATA_MODIFIED is not set(see vvp_io_setattr_fini()).
1704          * This way we can save an RPC for common open + trunc
1705          * operation. */
1706         if (ll_file_test_and_clear_flag(lli, LLIF_DATA_MODIFIED)) {
1707                 struct hsm_state_set hss = {
1708                         .hss_valid = HSS_SETMASK,
1709                         .hss_setmask = HS_DIRTY,
1710                 };
1711                 int rc2;
1712
1713                 rc2 = ll_hsm_state_set(inode, &hss);
1714                 /* truncate and write can happen at the same time, so that
1715                  * the file can be set modified even though the file is not
1716                  * restored from released state, and ll_hsm_state_set() is
1717                  * not applicable for the file, and rc2 < 0 is normal in this
1718                  * case. */
1719                 if (rc2 < 0)
1720                         CDEBUG(D_INFO, DFID "HSM set dirty failed: rc2 = %d\n",
1721                                PFID(ll_inode2fid(inode)), rc2);
1722         }
1723
1724         EXIT;
1725 out:
1726         if (op_data != NULL)
1727                 ll_finish_md_op_data(op_data);
1728
1729         if (S_ISREG(inode->i_mode)) {
1730                 inode_lock(inode);
1731                 if ((attr->ia_valid & ATTR_SIZE) && !hsm_import)
1732                         inode_dio_wait(inode);
1733                 /* Once we've got the i_mutex, it's safe to set the S_NOSEC
1734                  * flag.  ll_update_inode (called from ll_md_setattr), clears
1735                  * inode flags, so there is a gap where S_NOSEC is not set.
1736                  * This can cause a writer to take the i_mutex unnecessarily,
1737                  * but this is safe to do and should be rare. */
1738                 inode_has_no_xattr(inode);
1739         }
1740
1741         ll_stats_ops_tally(ll_i2sbi(inode), (attr->ia_valid & ATTR_SIZE) ?
1742                         LPROC_LL_TRUNC : LPROC_LL_SETATTR, 1);
1743
1744         return rc;
1745 }
1746
1747 int ll_setattr(struct dentry *de, struct iattr *attr)
1748 {
1749         int mode = de->d_inode->i_mode;
1750
1751         if ((attr->ia_valid & (ATTR_CTIME|ATTR_SIZE|ATTR_MODE)) ==
1752                               (ATTR_CTIME|ATTR_SIZE|ATTR_MODE))
1753                 attr->ia_valid |= MDS_OPEN_OWNEROVERRIDE;
1754
1755         if (((attr->ia_valid & (ATTR_MODE|ATTR_FORCE|ATTR_SIZE)) ==
1756                                (ATTR_SIZE|ATTR_MODE)) &&
1757             (((mode & S_ISUID) && !(attr->ia_mode & S_ISUID)) ||
1758              (((mode & (S_ISGID|S_IXGRP)) == (S_ISGID|S_IXGRP)) &&
1759               !(attr->ia_mode & S_ISGID))))
1760                 attr->ia_valid |= ATTR_FORCE;
1761
1762         if ((attr->ia_valid & ATTR_MODE) &&
1763             (mode & S_ISUID) &&
1764             !(attr->ia_mode & S_ISUID) &&
1765             !(attr->ia_valid & ATTR_KILL_SUID))
1766                 attr->ia_valid |= ATTR_KILL_SUID;
1767
1768         if ((attr->ia_valid & ATTR_MODE) &&
1769             ((mode & (S_ISGID|S_IXGRP)) == (S_ISGID|S_IXGRP)) &&
1770             !(attr->ia_mode & S_ISGID) &&
1771             !(attr->ia_valid & ATTR_KILL_SGID))
1772                 attr->ia_valid |= ATTR_KILL_SGID;
1773
1774         /* avoid polluted from ATTR_TIMES_SET,
1775          * projid is not expected to be set here */
1776         attr->ia_valid &= ~MDS_ATTR_PROJID;
1777
1778         return ll_setattr_raw(de, attr, false);
1779 }
1780
1781 int ll_statfs_internal(struct super_block *sb, struct obd_statfs *osfs,
1782                        u32 flags)
1783 {
1784         struct ll_sb_info *sbi = ll_s2sbi(sb);
1785         struct obd_statfs obd_osfs;
1786         time64_t max_age;
1787         int rc;
1788
1789         ENTRY;
1790         max_age = ktime_get_seconds() - OBD_STATFS_CACHE_SECONDS;
1791
1792         rc = obd_statfs(NULL, sbi->ll_md_exp, osfs, max_age, flags);
1793         if (rc) {
1794                 CERROR("md_statfs fails: rc = %d\n", rc);
1795                 RETURN(rc);
1796         }
1797
1798         osfs->os_type = sb->s_magic;
1799
1800         CDEBUG(D_SUPER, "MDC blocks %llu/%llu objects %llu/%llu\n",
1801                osfs->os_bavail, osfs->os_blocks, osfs->os_ffree,osfs->os_files);
1802
1803         if (sbi->ll_flags & LL_SBI_LAZYSTATFS)
1804                 flags |= OBD_STATFS_NODELAY;
1805
1806         rc = obd_statfs(NULL, sbi->ll_dt_exp, &obd_osfs, max_age, flags);
1807         if (rc) {
1808                 CERROR("obd_statfs fails: rc = %d\n", rc);
1809                 RETURN(rc);
1810         }
1811
1812         CDEBUG(D_SUPER, "OSC blocks %llu/%llu objects %llu/%llu\n",
1813                obd_osfs.os_bavail, obd_osfs.os_blocks, obd_osfs.os_ffree,
1814                obd_osfs.os_files);
1815
1816         osfs->os_bsize = obd_osfs.os_bsize;
1817         osfs->os_blocks = obd_osfs.os_blocks;
1818         osfs->os_bfree = obd_osfs.os_bfree;
1819         osfs->os_bavail = obd_osfs.os_bavail;
1820
1821         /* If we don't have as many objects free on the OST as inodes
1822          * on the MDS, we reduce the total number of inodes to
1823          * compensate, so that the "inodes in use" number is correct.
1824          */
1825         if (obd_osfs.os_ffree < osfs->os_ffree) {
1826                 osfs->os_files = (osfs->os_files - osfs->os_ffree) +
1827                         obd_osfs.os_ffree;
1828                 osfs->os_ffree = obd_osfs.os_ffree;
1829         }
1830
1831         RETURN(rc);
1832 }
1833 int ll_statfs(struct dentry *de, struct kstatfs *sfs)
1834 {
1835         struct super_block *sb = de->d_sb;
1836         struct obd_statfs osfs;
1837         __u64 fsid = huge_encode_dev(sb->s_dev);
1838         int rc;
1839
1840         CDEBUG(D_VFSTRACE, "VFS Op: at %llu jiffies\n", get_jiffies_64());
1841         ll_stats_ops_tally(ll_s2sbi(sb), LPROC_LL_STAFS, 1);
1842
1843         /* Some amount of caching on the client is allowed */
1844         rc = ll_statfs_internal(sb, &osfs, 0);
1845         if (rc)
1846                 return rc;
1847
1848         statfs_unpack(sfs, &osfs);
1849
1850         /* We need to downshift for all 32-bit kernels, because we can't
1851          * tell if the kernel is being called via sys_statfs64() or not.
1852          * Stop before overflowing f_bsize - in which case it is better
1853          * to just risk EOVERFLOW if caller is using old sys_statfs(). */
1854         if (sizeof(long) < 8) {
1855                 while (osfs.os_blocks > ~0UL && sfs->f_bsize < 0x40000000) {
1856                         sfs->f_bsize <<= 1;
1857
1858                         osfs.os_blocks >>= 1;
1859                         osfs.os_bfree >>= 1;
1860                         osfs.os_bavail >>= 1;
1861                 }
1862         }
1863
1864         sfs->f_blocks = osfs.os_blocks;
1865         sfs->f_bfree = osfs.os_bfree;
1866         sfs->f_bavail = osfs.os_bavail;
1867         sfs->f_fsid.val[0] = (__u32)fsid;
1868         sfs->f_fsid.val[1] = (__u32)(fsid >> 32);
1869         return 0;
1870 }
1871
1872 void ll_inode_size_lock(struct inode *inode)
1873 {
1874         struct ll_inode_info *lli;
1875
1876         LASSERT(!S_ISDIR(inode->i_mode));
1877
1878         lli = ll_i2info(inode);
1879         mutex_lock(&lli->lli_size_mutex);
1880 }
1881
1882 void ll_inode_size_unlock(struct inode *inode)
1883 {
1884         struct ll_inode_info *lli;
1885
1886         lli = ll_i2info(inode);
1887         mutex_unlock(&lli->lli_size_mutex);
1888 }
1889
1890 int ll_update_inode(struct inode *inode, struct lustre_md *md)
1891 {
1892         struct ll_inode_info *lli = ll_i2info(inode);
1893         struct mdt_body *body = md->body;
1894         struct ll_sb_info *sbi = ll_i2sbi(inode);
1895         int rc = 0;
1896
1897         if (body->mbo_valid & OBD_MD_FLEASIZE) {
1898                 rc = cl_file_inode_init(inode, md);
1899                 if (rc)
1900                         return rc;
1901         }
1902
1903         if (S_ISDIR(inode->i_mode)) {
1904                 rc = ll_update_lsm_md(inode, md);
1905                 if (rc != 0)
1906                         return rc;
1907         }
1908
1909 #ifdef CONFIG_FS_POSIX_ACL
1910         if (body->mbo_valid & OBD_MD_FLACL) {
1911                 spin_lock(&lli->lli_lock);
1912                 if (lli->lli_posix_acl)
1913                         posix_acl_release(lli->lli_posix_acl);
1914                 lli->lli_posix_acl = md->posix_acl;
1915                 spin_unlock(&lli->lli_lock);
1916         }
1917 #endif
1918         inode->i_ino = cl_fid_build_ino(&body->mbo_fid1,
1919                                         sbi->ll_flags & LL_SBI_32BIT_API);
1920         inode->i_generation = cl_fid_build_gen(&body->mbo_fid1);
1921
1922         if (body->mbo_valid & OBD_MD_FLATIME) {
1923                 if (body->mbo_atime > LTIME_S(inode->i_atime))
1924                         LTIME_S(inode->i_atime) = body->mbo_atime;
1925                 lli->lli_atime = body->mbo_atime;
1926         }
1927
1928         if (body->mbo_valid & OBD_MD_FLMTIME) {
1929                 if (body->mbo_mtime > LTIME_S(inode->i_mtime)) {
1930                         CDEBUG(D_INODE, "setting ino %lu mtime from %lu "
1931                                "to %llu\n", inode->i_ino,
1932                                LTIME_S(inode->i_mtime), body->mbo_mtime);
1933                         LTIME_S(inode->i_mtime) = body->mbo_mtime;
1934                 }
1935                 lli->lli_mtime = body->mbo_mtime;
1936         }
1937
1938         if (body->mbo_valid & OBD_MD_FLCTIME) {
1939                 if (body->mbo_ctime > LTIME_S(inode->i_ctime))
1940                         LTIME_S(inode->i_ctime) = body->mbo_ctime;
1941                 lli->lli_ctime = body->mbo_ctime;
1942         }
1943
1944         /* Clear i_flags to remove S_NOSEC before permissions are updated */
1945         if (body->mbo_valid & OBD_MD_FLFLAGS)
1946                 inode->i_flags = ll_ext_to_inode_flags(body->mbo_flags);
1947         if (body->mbo_valid & OBD_MD_FLMODE)
1948                 inode->i_mode = (inode->i_mode & S_IFMT) |
1949                                 (body->mbo_mode & ~S_IFMT);
1950
1951         if (body->mbo_valid & OBD_MD_FLTYPE)
1952                 inode->i_mode = (inode->i_mode & ~S_IFMT) |
1953                                 (body->mbo_mode & S_IFMT);
1954
1955         LASSERT(inode->i_mode != 0);
1956         if (S_ISREG(inode->i_mode))
1957                 inode->i_blkbits = min(PTLRPC_MAX_BRW_BITS + 1,
1958                                        LL_MAX_BLKSIZE_BITS);
1959         else
1960                 inode->i_blkbits = inode->i_sb->s_blocksize_bits;
1961
1962         if (body->mbo_valid & OBD_MD_FLUID)
1963                 inode->i_uid = make_kuid(&init_user_ns, body->mbo_uid);
1964         if (body->mbo_valid & OBD_MD_FLGID)
1965                 inode->i_gid = make_kgid(&init_user_ns, body->mbo_gid);
1966         if (body->mbo_valid & OBD_MD_FLPROJID)
1967                 lli->lli_projid = body->mbo_projid;
1968         if (body->mbo_valid & OBD_MD_FLNLINK)
1969                 set_nlink(inode, body->mbo_nlink);
1970         if (body->mbo_valid & OBD_MD_FLRDEV)
1971                 inode->i_rdev = old_decode_dev(body->mbo_rdev);
1972
1973         if (body->mbo_valid & OBD_MD_FLID) {
1974                 /* FID shouldn't be changed! */
1975                 if (fid_is_sane(&lli->lli_fid)) {
1976                         LASSERTF(lu_fid_eq(&lli->lli_fid, &body->mbo_fid1),
1977                                  "Trying to change FID "DFID
1978                                  " to the "DFID", inode "DFID"(%p)\n",
1979                                  PFID(&lli->lli_fid), PFID(&body->mbo_fid1),
1980                                  PFID(ll_inode2fid(inode)), inode);
1981                 } else {
1982                         lli->lli_fid = body->mbo_fid1;
1983                 }
1984         }
1985
1986         LASSERT(fid_seq(&lli->lli_fid) != 0);
1987
1988         if (body->mbo_valid & OBD_MD_FLSIZE) {
1989                 i_size_write(inode, body->mbo_size);
1990
1991                 CDEBUG(D_VFSTRACE, "inode="DFID", updating i_size %llu\n",
1992                        PFID(ll_inode2fid(inode)),
1993                        (unsigned long long)body->mbo_size);
1994
1995                 if (body->mbo_valid & OBD_MD_FLBLOCKS)
1996                         inode->i_blocks = body->mbo_blocks;
1997         }
1998
1999         if (body->mbo_valid & OBD_MD_TSTATE) {
2000                 /* Set LLIF_FILE_RESTORING if restore ongoing and
2001                  * clear it when done to ensure to start again
2002                  * glimpsing updated attrs
2003                  */
2004                 if (body->mbo_t_state & MS_RESTORE)
2005                         ll_file_set_flag(lli, LLIF_FILE_RESTORING);
2006                 else
2007                         ll_file_clear_flag(lli, LLIF_FILE_RESTORING);
2008         }
2009
2010         return 0;
2011 }
2012
2013 int ll_read_inode2(struct inode *inode, void *opaque)
2014 {
2015         struct lustre_md *md = opaque;
2016         struct ll_inode_info *lli = ll_i2info(inode);
2017         int     rc;
2018         ENTRY;
2019
2020         CDEBUG(D_VFSTRACE, "VFS Op:inode="DFID"(%p)\n",
2021                PFID(&lli->lli_fid), inode);
2022
2023         /* Core attributes from the MDS first.  This is a new inode, and
2024          * the VFS doesn't zero times in the core inode so we have to do
2025          * it ourselves.  They will be overwritten by either MDS or OST
2026          * attributes - we just need to make sure they aren't newer. */
2027         LTIME_S(inode->i_mtime) = 0;
2028         LTIME_S(inode->i_atime) = 0;
2029         LTIME_S(inode->i_ctime) = 0;
2030         inode->i_rdev = 0;
2031         rc = ll_update_inode(inode, md);
2032         if (rc != 0)
2033                 RETURN(rc);
2034
2035         /* OIDEBUG(inode); */
2036
2037 #ifdef HAVE_BACKING_DEV_INFO
2038         /* initializing backing dev info. */
2039         inode->i_mapping->backing_dev_info = &s2lsi(inode->i_sb)->lsi_bdi;
2040 #endif
2041         if (S_ISREG(inode->i_mode)) {
2042                 struct ll_sb_info *sbi = ll_i2sbi(inode);
2043                 inode->i_op = &ll_file_inode_operations;
2044                 inode->i_fop = sbi->ll_fop;
2045                 inode->i_mapping->a_ops = (struct address_space_operations *)&ll_aops;
2046                 EXIT;
2047         } else if (S_ISDIR(inode->i_mode)) {
2048                 inode->i_op = &ll_dir_inode_operations;
2049                 inode->i_fop = &ll_dir_operations;
2050                 EXIT;
2051         } else if (S_ISLNK(inode->i_mode)) {
2052                 inode->i_op = &ll_fast_symlink_inode_operations;
2053                 EXIT;
2054         } else {
2055                 inode->i_op = &ll_special_inode_operations;
2056
2057                 init_special_inode(inode, inode->i_mode,
2058                                    inode->i_rdev);
2059
2060                 EXIT;
2061         }
2062
2063         return 0;
2064 }
2065
2066 void ll_delete_inode(struct inode *inode)
2067 {
2068         struct ll_inode_info *lli = ll_i2info(inode);
2069         ENTRY;
2070
2071         if (S_ISREG(inode->i_mode) && lli->lli_clob != NULL)
2072                 /* It is last chance to write out dirty pages,
2073                  * otherwise we may lose data while umount */
2074                 cl_sync_file_range(inode, 0, OBD_OBJECT_EOF, CL_FSYNC_LOCAL, 1);
2075
2076         truncate_inode_pages_final(&inode->i_data);
2077
2078         LASSERTF(inode->i_data.nrpages == 0, "inode="DFID"(%p) nrpages=%lu, "
2079                  "see https://jira.hpdd.intel.com/browse/LU-118\n",
2080                  PFID(ll_inode2fid(inode)), inode, inode->i_data.nrpages);
2081
2082 #ifdef HAVE_SBOPS_EVICT_INODE
2083         ll_clear_inode(inode);
2084 #endif
2085         clear_inode(inode);
2086
2087         EXIT;
2088 }
2089
2090 int ll_iocontrol(struct inode *inode, struct file *file,
2091                  unsigned int cmd, unsigned long arg)
2092 {
2093         struct ll_sb_info *sbi = ll_i2sbi(inode);
2094         struct ptlrpc_request *req = NULL;
2095         int rc, flags = 0;
2096         ENTRY;
2097
2098         switch (cmd) {
2099         case FS_IOC_GETFLAGS: {
2100                 struct mdt_body *body;
2101                 struct md_op_data *op_data;
2102
2103                 op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL,
2104                                              0, 0, LUSTRE_OPC_ANY,
2105                                              NULL);
2106                 if (IS_ERR(op_data))
2107                         RETURN(PTR_ERR(op_data));
2108
2109                 op_data->op_valid = OBD_MD_FLFLAGS;
2110                 rc = md_getattr(sbi->ll_md_exp, op_data, &req);
2111                 ll_finish_md_op_data(op_data);
2112                 if (rc) {
2113                         CERROR("%s: failure inode "DFID": rc = %d\n",
2114                                sbi->ll_md_exp->exp_obd->obd_name,
2115                                PFID(ll_inode2fid(inode)), rc);
2116                         RETURN(-abs(rc));
2117                 }
2118
2119                 body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
2120
2121                 flags = body->mbo_flags;
2122
2123                 ptlrpc_req_finished(req);
2124
2125                 RETURN(put_user(flags, (int __user *)arg));
2126         }
2127         case FS_IOC_SETFLAGS: {
2128                 struct iattr *attr;
2129                 struct md_op_data *op_data;
2130                 struct cl_object *obj;
2131
2132                 if (get_user(flags, (int __user *)arg))
2133                         RETURN(-EFAULT);
2134
2135                 op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL, 0, 0,
2136                                              LUSTRE_OPC_ANY, NULL);
2137                 if (IS_ERR(op_data))
2138                         RETURN(PTR_ERR(op_data));
2139
2140                 op_data->op_attr_flags = flags;
2141                 op_data->op_attr.ia_valid |= ATTR_ATTR_FLAG;
2142                 rc = md_setattr(sbi->ll_md_exp, op_data, NULL, 0, &req);
2143                 ll_finish_md_op_data(op_data);
2144                 ptlrpc_req_finished(req);
2145                 if (rc)
2146                         RETURN(rc);
2147
2148                 inode->i_flags = ll_ext_to_inode_flags(flags);
2149
2150                 obj = ll_i2info(inode)->lli_clob;
2151                 if (obj == NULL)
2152                         RETURN(0);
2153
2154                 OBD_ALLOC_PTR(attr);
2155                 if (attr == NULL)
2156                         RETURN(-ENOMEM);
2157
2158                 attr->ia_valid = ATTR_ATTR_FLAG;
2159                 rc = cl_setattr_ost(obj, attr, flags);
2160
2161                 OBD_FREE_PTR(attr);
2162                 RETURN(rc);
2163         }
2164         default:
2165                 RETURN(-ENOSYS);
2166         }
2167
2168         RETURN(0);
2169 }
2170
2171 int ll_flush_ctx(struct inode *inode)
2172 {
2173         struct ll_sb_info  *sbi = ll_i2sbi(inode);
2174
2175         CDEBUG(D_SEC, "flush context for user %d\n",
2176                from_kuid(&init_user_ns, current_uid()));
2177
2178         obd_set_info_async(NULL, sbi->ll_md_exp,
2179                            sizeof(KEY_FLUSH_CTX), KEY_FLUSH_CTX,
2180                            0, NULL, NULL);
2181         obd_set_info_async(NULL, sbi->ll_dt_exp,
2182                            sizeof(KEY_FLUSH_CTX), KEY_FLUSH_CTX,
2183                            0, NULL, NULL);
2184         return 0;
2185 }
2186
2187 /* umount -f client means force down, don't save state */
2188 void ll_umount_begin(struct super_block *sb)
2189 {
2190         struct ll_sb_info *sbi = ll_s2sbi(sb);
2191         struct obd_device *obd;
2192         struct obd_ioctl_data *ioc_data;
2193         struct l_wait_info lwi;
2194         wait_queue_head_t waitq;
2195         ENTRY;
2196
2197         CDEBUG(D_VFSTRACE, "VFS Op: superblock %p count %d active %d\n", sb,
2198                sb->s_count, atomic_read(&sb->s_active));
2199
2200         obd = class_exp2obd(sbi->ll_md_exp);
2201         if (obd == NULL) {
2202                 CERROR("Invalid MDC connection handle %#llx\n",
2203                        sbi->ll_md_exp->exp_handle.h_cookie);
2204                 EXIT;
2205                 return;
2206         }
2207         obd->obd_force = 1;
2208
2209         obd = class_exp2obd(sbi->ll_dt_exp);
2210         if (obd == NULL) {
2211                 CERROR("Invalid LOV connection handle %#llx\n",
2212                        sbi->ll_dt_exp->exp_handle.h_cookie);
2213                 EXIT;
2214                 return;
2215         }
2216         obd->obd_force = 1;
2217
2218         OBD_ALLOC_PTR(ioc_data);
2219         if (ioc_data) {
2220                 obd_iocontrol(IOC_OSC_SET_ACTIVE, sbi->ll_md_exp,
2221                               sizeof *ioc_data, ioc_data, NULL);
2222
2223                 obd_iocontrol(IOC_OSC_SET_ACTIVE, sbi->ll_dt_exp,
2224                               sizeof *ioc_data, ioc_data, NULL);
2225
2226                 OBD_FREE_PTR(ioc_data);
2227         }
2228
2229         /* Really, we'd like to wait until there are no requests outstanding,
2230          * and then continue.  For now, we just periodically checking for vfs
2231          * to decrement mnt_cnt and hope to finish it within 10sec.
2232          */
2233         init_waitqueue_head(&waitq);
2234         lwi = LWI_TIMEOUT_INTERVAL(cfs_time_seconds(10),
2235                                    cfs_time_seconds(1), NULL, NULL);
2236         l_wait_event(waitq, may_umount(sbi->ll_mnt.mnt), &lwi);
2237
2238         EXIT;
2239 }
2240
2241 int ll_remount_fs(struct super_block *sb, int *flags, char *data)
2242 {
2243         struct ll_sb_info *sbi = ll_s2sbi(sb);
2244         char *profilenm = get_profile_name(sb);
2245         int err;
2246         __u32 read_only;
2247
2248         if ((*flags & MS_RDONLY) != (sb->s_flags & MS_RDONLY)) {
2249                 read_only = *flags & MS_RDONLY;
2250                 err = obd_set_info_async(NULL, sbi->ll_md_exp,
2251                                          sizeof(KEY_READ_ONLY),
2252                                          KEY_READ_ONLY, sizeof(read_only),
2253                                          &read_only, NULL);
2254                 if (err) {
2255                         LCONSOLE_WARN("Failed to remount %s %s (%d)\n",
2256                                       profilenm, read_only ?
2257                                       "read-only" : "read-write", err);
2258                         return err;
2259                 }
2260
2261                 if (read_only)
2262                         sb->s_flags |= MS_RDONLY;
2263                 else
2264                         sb->s_flags &= ~MS_RDONLY;
2265
2266                 if (sbi->ll_flags & LL_SBI_VERBOSE)
2267                         LCONSOLE_WARN("Remounted %s %s\n", profilenm,
2268                                       read_only ?  "read-only" : "read-write");
2269         }
2270         return 0;
2271 }
2272
2273 /**
2274  * Cleanup the open handle that is cached on MDT-side.
2275  *
2276  * For open case, the client side open handling thread may hit error
2277  * after the MDT grant the open. Under such case, the client should
2278  * send close RPC to the MDT as cleanup; otherwise, the open handle
2279  * on the MDT will be leaked there until the client umount or evicted.
2280  *
2281  * In further, if someone unlinked the file, because the open handle
2282  * holds the reference on such file/object, then it will block the
2283  * subsequent threads that want to locate such object via FID.
2284  *
2285  * \param[in] sb        super block for this file-system
2286  * \param[in] open_req  pointer to the original open request
2287  */
2288 void ll_open_cleanup(struct super_block *sb, struct ptlrpc_request *open_req)
2289 {
2290         struct mdt_body                 *body;
2291         struct md_op_data               *op_data;
2292         struct ptlrpc_request           *close_req = NULL;
2293         struct obd_export               *exp       = ll_s2sbi(sb)->ll_md_exp;
2294         ENTRY;
2295
2296         body = req_capsule_server_get(&open_req->rq_pill, &RMF_MDT_BODY);
2297         OBD_ALLOC_PTR(op_data);
2298         if (op_data == NULL) {
2299                 CWARN("%s: cannot allocate op_data to release open handle for "
2300                       DFID"\n",
2301                       ll_get_fsname(sb, NULL, 0), PFID(&body->mbo_fid1));
2302
2303                 RETURN_EXIT;
2304         }
2305
2306         op_data->op_fid1 = body->mbo_fid1;
2307         op_data->op_handle = body->mbo_handle;
2308         op_data->op_mod_time = ktime_get_real_seconds();
2309         md_close(exp, op_data, NULL, &close_req);
2310         ptlrpc_req_finished(close_req);
2311         ll_finish_md_op_data(op_data);
2312
2313         EXIT;
2314 }
2315
2316 int ll_prep_inode(struct inode **inode, struct ptlrpc_request *req,
2317                   struct super_block *sb, struct lookup_intent *it)
2318 {
2319         struct ll_sb_info *sbi = NULL;
2320         struct lustre_md md = { NULL };
2321         int rc;
2322         ENTRY;
2323
2324         LASSERT(*inode || sb);
2325         sbi = sb ? ll_s2sbi(sb) : ll_i2sbi(*inode);
2326         rc = md_get_lustre_md(sbi->ll_md_exp, req, sbi->ll_dt_exp,
2327                               sbi->ll_md_exp, &md);
2328         if (rc != 0)
2329                 GOTO(cleanup, rc);
2330
2331         if (*inode) {
2332                 rc = ll_update_inode(*inode, &md);
2333                 if (rc != 0)
2334                         GOTO(out, rc);
2335         } else {
2336                 LASSERT(sb != NULL);
2337
2338                 /*
2339                  * At this point server returns to client's same fid as client
2340                  * generated for creating. So using ->fid1 is okay here.
2341                  */
2342                 if (!fid_is_sane(&md.body->mbo_fid1)) {
2343                         CERROR("%s: Fid is insane "DFID"\n",
2344                                 ll_get_fsname(sb, NULL, 0),
2345                                 PFID(&md.body->mbo_fid1));
2346                         GOTO(out, rc = -EINVAL);
2347                 }
2348
2349                 *inode = ll_iget(sb, cl_fid_build_ino(&md.body->mbo_fid1,
2350                                              sbi->ll_flags & LL_SBI_32BIT_API),
2351                                  &md);
2352                 if (IS_ERR(*inode)) {
2353 #ifdef CONFIG_FS_POSIX_ACL
2354                         if (md.posix_acl) {
2355                                 posix_acl_release(md.posix_acl);
2356                                 md.posix_acl = NULL;
2357                         }
2358 #endif
2359                         rc = IS_ERR(*inode) ? PTR_ERR(*inode) : -ENOMEM;
2360                         *inode = NULL;
2361                         CERROR("new_inode -fatal: rc %d\n", rc);
2362                         GOTO(out, rc);
2363                 }
2364         }
2365
2366         /* Handling piggyback layout lock.
2367          * Layout lock can be piggybacked by getattr and open request.
2368          * The lsm can be applied to inode only if it comes with a layout lock
2369          * otherwise correct layout may be overwritten, for example:
2370          * 1. proc1: mdt returns a lsm but not granting layout
2371          * 2. layout was changed by another client
2372          * 3. proc2: refresh layout and layout lock granted
2373          * 4. proc1: to apply a stale layout */
2374         if (it != NULL && it->it_lock_mode != 0) {
2375                 struct lustre_handle lockh;
2376                 struct ldlm_lock *lock;
2377
2378                 lockh.cookie = it->it_lock_handle;
2379                 lock = ldlm_handle2lock(&lockh);
2380                 LASSERT(lock != NULL);
2381                 if (ldlm_has_layout(lock)) {
2382                         struct cl_object_conf conf;
2383
2384                         memset(&conf, 0, sizeof(conf));
2385                         conf.coc_opc = OBJECT_CONF_SET;
2386                         conf.coc_inode = *inode;
2387                         conf.coc_lock = lock;
2388                         conf.u.coc_layout = md.layout;
2389                         (void)ll_layout_conf(*inode, &conf);
2390                 }
2391                 LDLM_LOCK_PUT(lock);
2392         }
2393
2394         GOTO(out, rc = 0);
2395
2396 out:
2397         md_free_lustre_md(sbi->ll_md_exp, &md);
2398
2399 cleanup:
2400         if (rc != 0 && it != NULL && it->it_op & IT_OPEN)
2401                 ll_open_cleanup(sb != NULL ? sb : (*inode)->i_sb, req);
2402
2403         return rc;
2404 }
2405
2406 int ll_obd_statfs(struct inode *inode, void __user *arg)
2407 {
2408         struct ll_sb_info *sbi = NULL;
2409         struct obd_export *exp;
2410         char *buf = NULL;
2411         struct obd_ioctl_data *data = NULL;
2412         __u32 type;
2413         int len = 0, rc;
2414
2415         if (!inode || !(sbi = ll_i2sbi(inode)))
2416                 GOTO(out_statfs, rc = -EINVAL);
2417
2418         rc = obd_ioctl_getdata(&buf, &len, arg);
2419         if (rc)
2420                 GOTO(out_statfs, rc);
2421
2422         data = (void*)buf;
2423         if (!data->ioc_inlbuf1 || !data->ioc_inlbuf2 ||
2424             !data->ioc_pbuf1 || !data->ioc_pbuf2)
2425                 GOTO(out_statfs, rc = -EINVAL);
2426
2427         if (data->ioc_inllen1 != sizeof(__u32) ||
2428             data->ioc_inllen2 != sizeof(__u32) ||
2429             data->ioc_plen1 != sizeof(struct obd_statfs) ||
2430             data->ioc_plen2 != sizeof(struct obd_uuid))
2431                 GOTO(out_statfs, rc = -EINVAL);
2432
2433         memcpy(&type, data->ioc_inlbuf1, sizeof(__u32));
2434         if (type & LL_STATFS_LMV)
2435                 exp = sbi->ll_md_exp;
2436         else if (type & LL_STATFS_LOV)
2437                 exp = sbi->ll_dt_exp;
2438         else
2439                 GOTO(out_statfs, rc = -ENODEV);
2440
2441         rc = obd_iocontrol(IOC_OBD_STATFS, exp, len, buf, NULL);
2442         if (rc)
2443                 GOTO(out_statfs, rc);
2444 out_statfs:
2445         OBD_FREE_LARGE(buf, len);
2446         return rc;
2447 }
2448
2449 int ll_process_config(struct lustre_cfg *lcfg)
2450 {
2451         struct super_block *sb;
2452         unsigned long x;
2453         int rc = 0;
2454         char *ptr;
2455
2456         /* The instance name contains the sb: lustre-client-aacfe000 */
2457         ptr = strrchr(lustre_cfg_string(lcfg, 0), '-');
2458         if (!ptr || !*(++ptr))
2459                 return -EINVAL;
2460         if (sscanf(ptr, "%lx", &x) != 1)
2461                 return -EINVAL;
2462         sb = (struct super_block *)x;
2463         /* This better be a real Lustre superblock! */
2464         LASSERT(s2lsi(sb)->lsi_lmd->lmd_magic == LMD_MAGIC);
2465
2466         /* Note we have not called client_common_fill_super yet, so
2467            proc fns must be able to handle that! */
2468         rc = class_process_proc_param(PARAM_LLITE, lprocfs_llite_obd_vars,
2469                                       lcfg, sb);
2470         if (rc > 0)
2471                 rc = 0;
2472         return rc;
2473 }
2474
2475 /* this function prepares md_op_data hint for passing it down to MD stack. */
2476 struct md_op_data *ll_prep_md_op_data(struct md_op_data *op_data,
2477                                       struct inode *i1, struct inode *i2,
2478                                       const char *name, size_t namelen,
2479                                       __u32 mode, __u32 opc, void *data)
2480 {
2481         LASSERT(i1 != NULL);
2482
2483         if (name == NULL) {
2484                 /* Do not reuse namelen for something else. */
2485                 if (namelen != 0)
2486                         return ERR_PTR(-EINVAL);
2487         } else {
2488                 if (namelen > ll_i2sbi(i1)->ll_namelen)
2489                         return ERR_PTR(-ENAMETOOLONG);
2490
2491                 if (!lu_name_is_valid_2(name, namelen))
2492                         return ERR_PTR(-EINVAL);
2493         }
2494
2495         if (op_data == NULL)
2496                 OBD_ALLOC_PTR(op_data);
2497
2498         if (op_data == NULL)
2499                 return ERR_PTR(-ENOMEM);
2500
2501         ll_i2gids(op_data->op_suppgids, i1, i2);
2502         op_data->op_fid1 = *ll_inode2fid(i1);
2503         op_data->op_default_stripe_offset = -1;
2504         if (S_ISDIR(i1->i_mode)) {
2505                 op_data->op_mea1 = ll_i2info(i1)->lli_lsm_md;
2506                 if (opc == LUSTRE_OPC_MKDIR)
2507                         op_data->op_default_stripe_offset =
2508                                    ll_i2info(i1)->lli_def_stripe_offset;
2509         }
2510
2511         if (i2) {
2512                 op_data->op_fid2 = *ll_inode2fid(i2);
2513                 if (S_ISDIR(i2->i_mode))
2514                         op_data->op_mea2 = ll_i2info(i2)->lli_lsm_md;
2515         } else {
2516                 fid_zero(&op_data->op_fid2);
2517         }
2518
2519         if (ll_i2sbi(i1)->ll_flags & LL_SBI_64BIT_HASH)
2520                 op_data->op_cli_flags |= CLI_HASH64;
2521
2522         if (ll_need_32bit_api(ll_i2sbi(i1)))
2523                 op_data->op_cli_flags |= CLI_API32;
2524
2525         op_data->op_name = name;
2526         op_data->op_namelen = namelen;
2527         op_data->op_mode = mode;
2528         op_data->op_mod_time = ktime_get_real_seconds();
2529         op_data->op_fsuid = from_kuid(&init_user_ns, current_fsuid());
2530         op_data->op_fsgid = from_kgid(&init_user_ns, current_fsgid());
2531         op_data->op_cap = cfs_curproc_cap_pack();
2532         op_data->op_mds = 0;
2533         if ((opc == LUSTRE_OPC_CREATE) && (name != NULL) &&
2534              filename_is_volatile(name, namelen, &op_data->op_mds)) {
2535                 op_data->op_bias |= MDS_CREATE_VOLATILE;
2536         }
2537         op_data->op_data = data;
2538
2539         return op_data;
2540 }
2541
2542 void ll_finish_md_op_data(struct md_op_data *op_data)
2543 {
2544         security_release_secctx(op_data->op_file_secctx,
2545                                 op_data->op_file_secctx_size);
2546         OBD_FREE_PTR(op_data);
2547 }
2548
2549 #ifdef HAVE_SUPEROPS_USE_DENTRY
2550 int ll_show_options(struct seq_file *seq, struct dentry *dentry)
2551 #else
2552 int ll_show_options(struct seq_file *seq, struct vfsmount *vfs)
2553 #endif
2554 {
2555         struct ll_sb_info *sbi;
2556
2557 #ifdef HAVE_SUPEROPS_USE_DENTRY
2558         LASSERT((seq != NULL) && (dentry != NULL));
2559         sbi = ll_s2sbi(dentry->d_sb);
2560 #else
2561         LASSERT((seq != NULL) && (vfs != NULL));
2562         sbi = ll_s2sbi(vfs->mnt_sb);
2563 #endif
2564
2565         if (sbi->ll_flags & LL_SBI_NOLCK)
2566                 seq_puts(seq, ",nolock");
2567
2568         if (sbi->ll_flags & LL_SBI_FLOCK)
2569                 seq_puts(seq, ",flock");
2570
2571         if (sbi->ll_flags & LL_SBI_LOCALFLOCK)
2572                 seq_puts(seq, ",localflock");
2573
2574         if (sbi->ll_flags & LL_SBI_USER_XATTR)
2575                 seq_puts(seq, ",user_xattr");
2576
2577         if (sbi->ll_flags & LL_SBI_LAZYSTATFS)
2578                 seq_puts(seq, ",lazystatfs");
2579
2580         if (sbi->ll_flags & LL_SBI_USER_FID2PATH)
2581                 seq_puts(seq, ",user_fid2path");
2582
2583         if (sbi->ll_flags & LL_SBI_ALWAYS_PING)
2584                 seq_puts(seq, ",always_ping");
2585
2586         RETURN(0);
2587 }
2588
2589 /**
2590  * Get obd name by cmd, and copy out to user space
2591  */
2592 int ll_get_obd_name(struct inode *inode, unsigned int cmd, unsigned long arg)
2593 {
2594         struct ll_sb_info *sbi = ll_i2sbi(inode);
2595         struct obd_device *obd;
2596         ENTRY;
2597
2598         if (cmd == OBD_IOC_GETDTNAME)
2599                 obd = class_exp2obd(sbi->ll_dt_exp);
2600         else if (cmd == OBD_IOC_GETMDNAME)
2601                 obd = class_exp2obd(sbi->ll_md_exp);
2602         else
2603                 RETURN(-EINVAL);
2604
2605         if (!obd)
2606                 RETURN(-ENOENT);
2607
2608         if (copy_to_user((void __user *)arg, obd->obd_name,
2609                          strlen(obd->obd_name) + 1))
2610                 RETURN(-EFAULT);
2611
2612         RETURN(0);
2613 }
2614
2615 /**
2616  * Get lustre file system name by \a sbi. If \a buf is provided(non-NULL), the
2617  * fsname will be returned in this buffer; otherwise, a static buffer will be
2618  * used to store the fsname and returned to caller.
2619  */
2620 char *ll_get_fsname(struct super_block *sb, char *buf, int buflen)
2621 {
2622         static char fsname_static[MTI_NAME_MAXLEN];
2623         struct lustre_sb_info *lsi = s2lsi(sb);
2624         char *ptr;
2625         int len;
2626
2627         if (buf == NULL) {
2628                 /* this means the caller wants to use static buffer
2629                  * and it doesn't care about race. Usually this is
2630                  * in error reporting path */
2631                 buf = fsname_static;
2632                 buflen = sizeof(fsname_static);
2633         }
2634
2635         len = strlen(lsi->lsi_lmd->lmd_profile);
2636         ptr = strrchr(lsi->lsi_lmd->lmd_profile, '-');
2637         if (ptr && (strcmp(ptr, "-client") == 0))
2638                 len -= 7;
2639
2640         if (unlikely(len >= buflen))
2641                 len = buflen - 1;
2642         strncpy(buf, lsi->lsi_lmd->lmd_profile, len);
2643         buf[len] = '\0';
2644
2645         return buf;
2646 }
2647
2648 static char* ll_d_path(struct dentry *dentry, char *buf, int bufsize)
2649 {
2650         char *path = NULL;
2651
2652         struct path p;
2653
2654         p.dentry = dentry;
2655         p.mnt = current->fs->root.mnt;
2656         path_get(&p);
2657         path = d_path(&p, buf, bufsize);
2658         path_put(&p);
2659         return path;
2660 }
2661
2662 void ll_dirty_page_discard_warn(struct page *page, int ioret)
2663 {
2664         char *buf, *path = NULL;
2665         struct dentry *dentry = NULL;
2666         struct inode *inode = page->mapping->host;
2667
2668         /* this can be called inside spin lock so use GFP_ATOMIC. */
2669         buf = (char *)__get_free_page(GFP_ATOMIC);
2670         if (buf != NULL) {
2671                 dentry = d_find_alias(page->mapping->host);
2672                 if (dentry != NULL)
2673                         path = ll_d_path(dentry, buf, PAGE_SIZE);
2674         }
2675
2676         CDEBUG(D_WARNING,
2677                "%s: dirty page discard: %s/fid: "DFID"/%s may get corrupted "
2678                "(rc %d)\n", ll_get_fsname(page->mapping->host->i_sb, NULL, 0),
2679                s2lsi(page->mapping->host->i_sb)->lsi_lmd->lmd_dev,
2680                PFID(ll_inode2fid(inode)),
2681                (path && !IS_ERR(path)) ? path : "", ioret);
2682
2683         if (dentry != NULL)
2684                 dput(dentry);
2685
2686         if (buf != NULL)
2687                 free_page((unsigned long)buf);
2688 }
2689
2690 ssize_t ll_copy_user_md(const struct lov_user_md __user *md,
2691                         struct lov_user_md **kbuf)
2692 {
2693         struct lov_user_md      lum;
2694         ssize_t                 lum_size;
2695         ENTRY;
2696
2697         if (copy_from_user(&lum, md, sizeof(lum)))
2698                 RETURN(-EFAULT);
2699
2700         lum_size = ll_lov_user_md_size(&lum);
2701         if (lum_size < 0)
2702                 RETURN(lum_size);
2703
2704         OBD_ALLOC(*kbuf, lum_size);
2705         if (*kbuf == NULL)
2706                 RETURN(-ENOMEM);
2707
2708         if (copy_from_user(*kbuf, md, lum_size) != 0) {
2709                 OBD_FREE(*kbuf, lum_size);
2710                 RETURN(-EFAULT);
2711         }
2712
2713         RETURN(lum_size);
2714 }
2715
2716 /*
2717  * Compute llite root squash state after a change of root squash
2718  * configuration setting or add/remove of a lnet nid
2719  */
2720 void ll_compute_rootsquash_state(struct ll_sb_info *sbi)
2721 {
2722         struct root_squash_info *squash = &sbi->ll_squash;
2723         int i;
2724         bool matched;
2725         struct lnet_process_id id;
2726
2727         /* Update norootsquash flag */
2728         down_write(&squash->rsi_sem);
2729         if (list_empty(&squash->rsi_nosquash_nids))
2730                 sbi->ll_flags &= ~LL_SBI_NOROOTSQUASH;
2731         else {
2732                 /* Do not apply root squash as soon as one of our NIDs is
2733                  * in the nosquash_nids list */
2734                 matched = false;
2735                 i = 0;
2736                 while (LNetGetId(i++, &id) != -ENOENT) {
2737                         if (LNET_NETTYP(LNET_NIDNET(id.nid)) == LOLND)
2738                                 continue;
2739                         if (cfs_match_nid(id.nid, &squash->rsi_nosquash_nids)) {
2740                                 matched = true;
2741                                 break;
2742                         }
2743                 }
2744                 if (matched)
2745                         sbi->ll_flags |= LL_SBI_NOROOTSQUASH;
2746                 else
2747                         sbi->ll_flags &= ~LL_SBI_NOROOTSQUASH;
2748         }
2749         up_write(&squash->rsi_sem);
2750 }
2751
2752 /**
2753  * Parse linkea content to extract information about a given hardlink
2754  *
2755  * \param[in]   ldata      - Initialized linkea data
2756  * \param[in]   linkno     - Link identifier
2757  * \param[out]  parent_fid - The entry's parent FID
2758  * \param[out]  ln         - Entry name destination buffer
2759  *
2760  * \retval 0 on success
2761  * \retval Appropriate negative error code on failure
2762  */
2763 static int ll_linkea_decode(struct linkea_data *ldata, unsigned int linkno,
2764                             struct lu_fid *parent_fid, struct lu_name *ln)
2765 {
2766         unsigned int    idx;
2767         int             rc;
2768         ENTRY;
2769
2770         rc = linkea_init_with_rec(ldata);
2771         if (rc < 0)
2772                 RETURN(rc);
2773
2774         if (linkno >= ldata->ld_leh->leh_reccount)
2775                 /* beyond last link */
2776                 RETURN(-ENODATA);
2777
2778         linkea_first_entry(ldata);
2779         for (idx = 0; ldata->ld_lee != NULL; idx++) {
2780                 linkea_entry_unpack(ldata->ld_lee, &ldata->ld_reclen, ln,
2781                                     parent_fid);
2782                 if (idx == linkno)
2783                         break;
2784
2785                 linkea_next_entry(ldata);
2786         }
2787
2788         if (idx < linkno)
2789                 RETURN(-ENODATA);
2790
2791         RETURN(0);
2792 }
2793
2794 /**
2795  * Get parent FID and name of an identified link. Operation is performed for
2796  * a given link number, letting the caller iterate over linkno to list one or
2797  * all links of an entry.
2798  *
2799  * \param[in]     file - File descriptor against which to perform the operation
2800  * \param[in,out] arg  - User-filled structure containing the linkno to operate
2801  *                       on and the available size. It is eventually filled with
2802  *                       the requested information or left untouched on error
2803  *
2804  * \retval - 0 on success
2805  * \retval - Appropriate negative error code on failure
2806  */
2807 int ll_getparent(struct file *file, struct getparent __user *arg)
2808 {
2809         struct inode            *inode = file_inode(file);
2810         struct linkea_data      *ldata;
2811         struct lu_buf            buf = LU_BUF_NULL;
2812         struct lu_name           ln;
2813         struct lu_fid            parent_fid;
2814         __u32                    linkno;
2815         __u32                    name_size;
2816         int                      rc;
2817
2818         ENTRY;
2819
2820         if (!cfs_capable(CFS_CAP_DAC_READ_SEARCH) &&
2821             !(ll_i2sbi(inode)->ll_flags & LL_SBI_USER_FID2PATH))
2822                 RETURN(-EPERM);
2823
2824         if (get_user(name_size, &arg->gp_name_size))
2825                 RETURN(-EFAULT);
2826
2827         if (get_user(linkno, &arg->gp_linkno))
2828                 RETURN(-EFAULT);
2829
2830         if (name_size > PATH_MAX)
2831                 RETURN(-EINVAL);
2832
2833         OBD_ALLOC(ldata, sizeof(*ldata));
2834         if (ldata == NULL)
2835                 RETURN(-ENOMEM);
2836
2837         rc = linkea_data_new(ldata, &buf);
2838         if (rc < 0)
2839                 GOTO(ldata_free, rc);
2840
2841 #ifdef HAVE_XATTR_HANDLER_FLAGS
2842         rc = ll_xattr_list(inode, XATTR_NAME_LINK, XATTR_TRUSTED_T, buf.lb_buf,
2843                            buf.lb_len, OBD_MD_FLXATTR);
2844 #else
2845         rc = ll_getxattr(file_dentry(file), XATTR_NAME_LINK, buf.lb_buf,
2846                          buf.lb_len);
2847 #endif /* HAVE_XATTR_HANDLER_FLAGS */
2848         if (rc < 0)
2849                 GOTO(lb_free, rc);
2850
2851         rc = ll_linkea_decode(ldata, linkno, &parent_fid, &ln);
2852         if (rc < 0)
2853                 GOTO(lb_free, rc);
2854
2855         if (ln.ln_namelen >= name_size)
2856                 GOTO(lb_free, rc = -EOVERFLOW);
2857
2858         if (copy_to_user(&arg->gp_fid, &parent_fid, sizeof(arg->gp_fid)))
2859                 GOTO(lb_free, rc = -EFAULT);
2860
2861         if (copy_to_user(&arg->gp_name, ln.ln_name, ln.ln_namelen))
2862                 GOTO(lb_free, rc = -EFAULT);
2863
2864         if (put_user('\0', arg->gp_name + ln.ln_namelen))
2865                 GOTO(lb_free, rc = -EFAULT);
2866
2867 lb_free:
2868         lu_buf_free(&buf);
2869 ldata_free:
2870         OBD_FREE(ldata, sizeof(*ldata));
2871
2872         RETURN(rc);
2873 }