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         ll_debugfs_unregister_super(sb);
 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 #define MAX_STRING_SIZE 128
 967
 968 #ifndef HAVE_SUPER_SETUP_BDI_NAME
 969
 970 #define LSI_BDI_INITIALIZED     0x00400000
 971
 972 #ifndef HAVE_BDI_CAP_MAP_COPY
 973 # define BDI_CAP_MAP_COPY       0
 974 #endif
 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         char name[MAX_STRING_SIZE];
1013         int md_len = 0;
1014         int dt_len = 0;
1015         char *ptr;
1016         int len;
1017         int err;
1018
1019         ENTRY;
1020         CDEBUG(D_VFSTRACE, "VFS Op: sb %p\n", sb);
1021
1022         try_module_get(THIS_MODULE);
1023
1024         OBD_ALLOC_PTR(cfg);
1025         if (cfg == NULL)
1026                 GOTO(out_free, err = -ENOMEM);
1027
1028         /* client additional sb info */
1029         lsi->lsi_llsbi = sbi = ll_init_sbi();
1030         if (!sbi)
1031                 GOTO(out_free, err = -ENOMEM);
1032
1033         err = ll_options(lsi->lsi_lmd->lmd_opts, sbi);
1034         if (err)
1035                 GOTO(out_free, err);
1036
1037         err = super_setup_bdi_name(sb, "lustre-%p", sb);
1038         if (err)
1039                 GOTO(out_free, err);
1040
1041 #ifndef HAVE_DCACHE_LOCK
1042         /* kernel >= 2.6.38 store dentry operations in sb->s_d_op. */
1043         sb->s_d_op = &ll_d_ops;
1044 #endif
1045         /* Get fsname */
1046         len = strlen(lsi->lsi_lmd->lmd_profile);
1047         ptr = strrchr(lsi->lsi_lmd->lmd_profile, '-');
1048         if (ptr && (strcmp(ptr, "-client") == 0))
1049                 len -= 7;
1050
1051         /* Mount info */
1052         snprintf(name, MAX_STRING_SIZE, "%.*s-%p", len,
1053                  lsi->lsi_lmd->lmd_profile, sb);
1054
1055         /* Call ll_debugfs_register_super() before lustre_process_log()
1056          * so that "llite.*.*" params can be processed correctly.
1057          */
1058         err = ll_debugfs_register_super(sb, name);
1059         if (err < 0) {
1060                 CERROR("%s: could not register mountpoint in llite: rc = %d\n",
1061                        ll_get_fsname(sb, NULL, 0), err);
1062                 err = 0;
1063         }
1064
1065         /* Generate a string unique to this super, in case some joker tries
1066            to mount the same fs at two mount points.
1067            Use the address of the super itself.*/
1068         cfg->cfg_instance = sb;
1069         cfg->cfg_uuid = lsi->lsi_llsbi->ll_sb_uuid;
1070         cfg->cfg_callback = class_config_llog_handler;
1071         cfg->cfg_sub_clds = CONFIG_SUB_CLIENT;
1072         /* set up client obds */
1073         err = lustre_process_log(sb, profilenm, cfg);
1074         if (err < 0)
1075                 GOTO(out_proc, err);
1076
1077         /* Profile set with LCFG_MOUNTOPT so we can find our mdc and osc obds */
1078         lprof = class_get_profile(profilenm);
1079         if (lprof == NULL) {
1080                 LCONSOLE_ERROR_MSG(0x156, "The client profile '%s' could not be"
1081                                    " read from the MGS.  Does that filesystem "
1082                                    "exist?\n", profilenm);
1083                 GOTO(out_proc, err = -EINVAL);
1084         }
1085         CDEBUG(D_CONFIG, "Found profile %s: mdc=%s osc=%s\n", profilenm,
1086                lprof->lp_md, lprof->lp_dt);
1087
1088         dt_len = strlen(lprof->lp_dt) + instlen + 2;
1089         OBD_ALLOC(dt, dt_len);
1090         if (!dt)
1091                 GOTO(out_proc, err = -ENOMEM);
1092         snprintf(dt, dt_len - 1, "%s-%p", lprof->lp_dt, cfg->cfg_instance);
1093
1094         md_len = strlen(lprof->lp_md) + instlen + 2;
1095         OBD_ALLOC(md, md_len);
1096         if (!md)
1097                 GOTO(out_proc, err = -ENOMEM);
1098         snprintf(md, md_len - 1, "%s-%p", lprof->lp_md, cfg->cfg_instance);
1099
1100         /* connections, registrations, sb setup */
1101         err = client_common_fill_super(sb, md, dt, mnt);
1102         if (err < 0)
1103                 GOTO(out_proc, err);
1104
1105         sbi->ll_client_common_fill_super_succeeded = 1;
1106
1107 out_proc:
1108         if (err < 0)
1109                 ll_debugfs_unregister_super(sb);
1110 out_free:
1111         if (md)
1112                 OBD_FREE(md, md_len);
1113         if (dt)
1114                 OBD_FREE(dt, dt_len);
1115         if (lprof != NULL)
1116                 class_put_profile(lprof);
1117         if (cfg)
1118                 OBD_FREE_PTR(cfg);
1119         if (err)
1120                 ll_put_super(sb);
1121         else if (sbi->ll_flags & LL_SBI_VERBOSE)
1122                 LCONSOLE_WARN("Mounted %s\n", profilenm);
1123         RETURN(err);
1124 } /* ll_fill_super */
1125
1126 void ll_put_super(struct super_block *sb)
1127 {
1128         struct config_llog_instance cfg, params_cfg;
1129         struct obd_device *obd;
1130         struct lustre_sb_info *lsi = s2lsi(sb);
1131         struct ll_sb_info *sbi = ll_s2sbi(sb);
1132         char *profilenm = get_profile_name(sb);
1133         long ccc_count;
1134         int next, force = 1, rc = 0;
1135         ENTRY;
1136
1137         if (!sbi)
1138                 GOTO(out_no_sbi, 0);
1139
1140         CDEBUG(D_VFSTRACE, "VFS Op: sb %p - %s\n", sb, profilenm);
1141
1142         cfg.cfg_instance = sb;
1143         lustre_end_log(sb, profilenm, &cfg);
1144
1145         params_cfg.cfg_instance = sb;
1146         lustre_end_log(sb, PARAMS_FILENAME, &params_cfg);
1147
1148         if (sbi->ll_md_exp) {
1149                 obd = class_exp2obd(sbi->ll_md_exp);
1150                 if (obd)
1151                         force = obd->obd_force;
1152         }
1153
1154         /* Wait for unstable pages to be committed to stable storage */
1155         if (force == 0) {
1156                 struct l_wait_info lwi = LWI_INTR(LWI_ON_SIGNAL_NOOP, NULL);
1157                 rc = l_wait_event(sbi->ll_cache->ccc_unstable_waitq,
1158                         atomic_long_read(&sbi->ll_cache->ccc_unstable_nr) == 0,
1159                         &lwi);
1160         }
1161
1162         ccc_count = atomic_long_read(&sbi->ll_cache->ccc_unstable_nr);
1163         if (force == 0 && rc != -EINTR)
1164                 LASSERTF(ccc_count == 0, "count: %li\n", ccc_count);
1165
1166
1167         /* We need to set force before the lov_disconnect in
1168            lustre_common_put_super, since l_d cleans up osc's as well. */
1169         if (force) {
1170                 next = 0;
1171                 while ((obd = class_devices_in_group(&sbi->ll_sb_uuid,
1172                                                      &next)) != NULL) {
1173                         obd->obd_force = force;
1174                 }
1175         }
1176
1177         if (sbi->ll_client_common_fill_super_succeeded) {
1178                 /* Only if client_common_fill_super succeeded */
1179                 client_common_put_super(sb);
1180         }
1181
1182         next = 0;
1183         while ((obd = class_devices_in_group(&sbi->ll_sb_uuid, &next)) !=NULL) {
1184                 class_manual_cleanup(obd);
1185         }
1186
1187         if (sbi->ll_flags & LL_SBI_VERBOSE)
1188                 LCONSOLE_WARN("Unmounted %s\n", profilenm ? profilenm : "");
1189
1190         if (profilenm)
1191                 class_del_profile(profilenm);
1192
1193 #ifndef HAVE_SUPER_SETUP_BDI_NAME
1194         if (lsi->lsi_flags & LSI_BDI_INITIALIZED) {
1195                 bdi_destroy(&lsi->lsi_bdi);
1196                 lsi->lsi_flags &= ~LSI_BDI_INITIALIZED;
1197         }
1198 #endif
1199
1200         ll_free_sbi(sb);
1201         lsi->lsi_llsbi = NULL;
1202 out_no_sbi:
1203         lustre_common_put_super(sb);
1204
1205         cl_env_cache_purge(~0);
1206
1207         module_put(THIS_MODULE);
1208
1209         EXIT;
1210 } /* client_put_super */
1211
1212 struct inode *ll_inode_from_resource_lock(struct ldlm_lock *lock)
1213 {
1214         struct inode *inode = NULL;
1215
1216         /* NOTE: we depend on atomic igrab() -bzzz */
1217         lock_res_and_lock(lock);
1218         if (lock->l_resource->lr_lvb_inode) {
1219                 struct ll_inode_info * lli;
1220                 lli = ll_i2info(lock->l_resource->lr_lvb_inode);
1221                 if (lli->lli_inode_magic == LLI_INODE_MAGIC) {
1222                         inode = igrab(lock->l_resource->lr_lvb_inode);
1223                 } else {
1224                         inode = lock->l_resource->lr_lvb_inode;
1225                         LDLM_DEBUG_LIMIT(inode->i_state & I_FREEING ?  D_INFO :
1226                                          D_WARNING, lock, "lr_lvb_inode %p is "
1227                                          "bogus: magic %08x",
1228                                          lock->l_resource->lr_lvb_inode,
1229                                          lli->lli_inode_magic);
1230                         inode = NULL;
1231                 }
1232         }
1233         unlock_res_and_lock(lock);
1234         return inode;
1235 }
1236
1237 void ll_dir_clear_lsm_md(struct inode *inode)
1238 {
1239         struct ll_inode_info *lli = ll_i2info(inode);
1240
1241         LASSERT(S_ISDIR(inode->i_mode));
1242
1243         if (lli->lli_lsm_md != NULL) {
1244                 lmv_free_memmd(lli->lli_lsm_md);
1245                 lli->lli_lsm_md = NULL;
1246         }
1247 }
1248
1249 static struct inode *ll_iget_anon_dir(struct super_block *sb,
1250                                       const struct lu_fid *fid,
1251                                       struct lustre_md *md)
1252 {
1253         struct ll_sb_info       *sbi = ll_s2sbi(sb);
1254         struct mdt_body         *body = md->body;
1255         struct inode            *inode;
1256         ino_t                   ino;
1257         ENTRY;
1258
1259         ino = cl_fid_build_ino(fid, sbi->ll_flags & LL_SBI_32BIT_API);
1260         inode = iget_locked(sb, ino);
1261         if (inode == NULL) {
1262                 CERROR("%s: failed get simple inode "DFID": rc = -ENOENT\n",
1263                        ll_get_fsname(sb, NULL, 0), PFID(fid));
1264                 RETURN(ERR_PTR(-ENOENT));
1265         }
1266
1267         if (inode->i_state & I_NEW) {
1268                 struct ll_inode_info *lli = ll_i2info(inode);
1269                 struct lmv_stripe_md *lsm = md->lmv;
1270
1271                 inode->i_mode = (inode->i_mode & ~S_IFMT) |
1272                                 (body->mbo_mode & S_IFMT);
1273                 LASSERTF(S_ISDIR(inode->i_mode), "Not slave inode "DFID"\n",
1274                          PFID(fid));
1275
1276                 LTIME_S(inode->i_mtime) = 0;
1277                 LTIME_S(inode->i_atime) = 0;
1278                 LTIME_S(inode->i_ctime) = 0;
1279                 inode->i_rdev = 0;
1280
1281 #ifdef HAVE_BACKING_DEV_INFO
1282                 /* initializing backing dev info. */
1283                 inode->i_mapping->backing_dev_info =
1284                                                 &s2lsi(inode->i_sb)->lsi_bdi;
1285 #endif
1286                 inode->i_op = &ll_dir_inode_operations;
1287                 inode->i_fop = &ll_dir_operations;
1288                 lli->lli_fid = *fid;
1289                 ll_lli_init(lli);
1290
1291                 LASSERT(lsm != NULL);
1292                 /* master object FID */
1293                 lli->lli_pfid = body->mbo_fid1;
1294                 CDEBUG(D_INODE, "lli %p slave "DFID" master "DFID"\n",
1295                        lli, PFID(fid), PFID(&lli->lli_pfid));
1296                 unlock_new_inode(inode);
1297         }
1298
1299         RETURN(inode);
1300 }
1301
1302 static int ll_init_lsm_md(struct inode *inode, struct lustre_md *md)
1303 {
1304         struct lu_fid *fid;
1305         struct lmv_stripe_md *lsm = md->lmv;
1306         int i;
1307
1308         LASSERT(lsm != NULL);
1309         /* XXX sigh, this lsm_root initialization should be in
1310          * LMV layer, but it needs ll_iget right now, so we
1311          * put this here right now. */
1312         for (i = 0; i < lsm->lsm_md_stripe_count; i++) {
1313                 fid = &lsm->lsm_md_oinfo[i].lmo_fid;
1314                 LASSERT(lsm->lsm_md_oinfo[i].lmo_root == NULL);
1315                 /* Unfortunately ll_iget will call ll_update_inode,
1316                  * where the initialization of slave inode is slightly
1317                  * different, so it reset lsm_md to NULL to avoid
1318                  * initializing lsm for slave inode. */
1319                 /* For migrating inode, master stripe and master object will
1320                  * be same, so we only need assign this inode */
1321                 if (lsm->lsm_md_hash_type & LMV_HASH_FLAG_MIGRATION && i == 0)
1322                         lsm->lsm_md_oinfo[i].lmo_root = inode;
1323                 else
1324                         lsm->lsm_md_oinfo[i].lmo_root =
1325                                 ll_iget_anon_dir(inode->i_sb, fid, md);
1326
1327                 if (IS_ERR(lsm->lsm_md_oinfo[i].lmo_root)) {
1328                         int rc = PTR_ERR(lsm->lsm_md_oinfo[i].lmo_root);
1329
1330                         lsm->lsm_md_oinfo[i].lmo_root = NULL;
1331                         return rc;
1332                 }
1333         }
1334
1335         return 0;
1336 }
1337
1338 static inline int lli_lsm_md_eq(const struct lmv_stripe_md *lsm_md1,
1339                                 const struct lmv_stripe_md *lsm_md2)
1340 {
1341         return lsm_md1->lsm_md_magic == lsm_md2->lsm_md_magic &&
1342                lsm_md1->lsm_md_stripe_count == lsm_md2->lsm_md_stripe_count &&
1343                lsm_md1->lsm_md_master_mdt_index ==
1344                                         lsm_md2->lsm_md_master_mdt_index &&
1345                lsm_md1->lsm_md_hash_type == lsm_md2->lsm_md_hash_type &&
1346                lsm_md1->lsm_md_layout_version ==
1347                                         lsm_md2->lsm_md_layout_version &&
1348                strcmp(lsm_md1->lsm_md_pool_name,
1349                       lsm_md2->lsm_md_pool_name) == 0;
1350 }
1351
1352 static int ll_update_lsm_md(struct inode *inode, struct lustre_md *md)
1353 {
1354         struct ll_inode_info *lli = ll_i2info(inode);
1355         struct lmv_stripe_md *lsm = md->lmv;
1356         int     rc;
1357         ENTRY;
1358
1359         LASSERT(S_ISDIR(inode->i_mode));
1360         CDEBUG(D_INODE, "update lsm %p of "DFID"\n", lli->lli_lsm_md,
1361                PFID(ll_inode2fid(inode)));
1362
1363         /* no striped information from request. */
1364         if (lsm == NULL) {
1365                 if (lli->lli_lsm_md == NULL) {
1366                         RETURN(0);
1367                 } else if (lli->lli_lsm_md->lsm_md_hash_type &
1368                                                 LMV_HASH_FLAG_MIGRATION) {
1369                         /* migration is done, the temporay MIGRATE layout has
1370                          * been removed */
1371                         CDEBUG(D_INODE, DFID" finish migration.\n",
1372                                PFID(ll_inode2fid(inode)));
1373                         lmv_free_memmd(lli->lli_lsm_md);
1374                         lli->lli_lsm_md = NULL;
1375                         RETURN(0);
1376                 } else {
1377                         /* The lustre_md from req does not include stripeEA,
1378                          * see ll_md_setattr */
1379                         RETURN(0);
1380                 }
1381         }
1382
1383         /* set the directory layout */
1384         if (lli->lli_lsm_md == NULL) {
1385                 struct cl_attr  *attr;
1386
1387                 rc = ll_init_lsm_md(inode, md);
1388                 if (rc != 0)
1389                         RETURN(rc);
1390
1391                 /* set md->lmv to NULL, so the following free lustre_md
1392                  * will not free this lsm */
1393                 md->lmv = NULL;
1394                 lli->lli_lsm_md = lsm;
1395
1396                 OBD_ALLOC_PTR(attr);
1397                 if (attr == NULL)
1398                         RETURN(-ENOMEM);
1399
1400                 /* validate the lsm */
1401                 rc = md_merge_attr(ll_i2mdexp(inode), lsm, attr,
1402                                    ll_md_blocking_ast);
1403                 if (rc != 0) {
1404                         OBD_FREE_PTR(attr);
1405                         RETURN(rc);
1406                 }
1407
1408                 if (md->body->mbo_valid & OBD_MD_FLNLINK)
1409                         md->body->mbo_nlink = attr->cat_nlink;
1410                 if (md->body->mbo_valid & OBD_MD_FLSIZE)
1411                         md->body->mbo_size = attr->cat_size;
1412                 if (md->body->mbo_valid & OBD_MD_FLATIME)
1413                         md->body->mbo_atime = attr->cat_atime;
1414                 if (md->body->mbo_valid & OBD_MD_FLCTIME)
1415                         md->body->mbo_ctime = attr->cat_ctime;
1416                 if (md->body->mbo_valid & OBD_MD_FLMTIME)
1417                         md->body->mbo_mtime = attr->cat_mtime;
1418
1419                 OBD_FREE_PTR(attr);
1420
1421                 CDEBUG(D_INODE, "Set lsm %p magic %x to "DFID"\n", lsm,
1422                        lsm->lsm_md_magic, PFID(ll_inode2fid(inode)));
1423                 RETURN(0);
1424         }
1425
1426         /* Compare the old and new stripe information */
1427         if (!lsm_md_eq(lli->lli_lsm_md, lsm)) {
1428                 struct lmv_stripe_md    *old_lsm = lli->lli_lsm_md;
1429                 int                     idx;
1430
1431                 CERROR("%s: inode "DFID"(%p)'s lmv layout mismatch (%p)/(%p)"
1432                        "magic:0x%x/0x%x stripe count: %d/%d master_mdt: %d/%d"
1433                        "hash_type:0x%x/0x%x layout: 0x%x/0x%x pool:%s/%s\n",
1434                        ll_get_fsname(inode->i_sb, NULL, 0), PFID(&lli->lli_fid),
1435                        inode, lsm, old_lsm,
1436                        lsm->lsm_md_magic, old_lsm->lsm_md_magic,
1437                        lsm->lsm_md_stripe_count,
1438                        old_lsm->lsm_md_stripe_count,
1439                        lsm->lsm_md_master_mdt_index,
1440                        old_lsm->lsm_md_master_mdt_index,
1441                        lsm->lsm_md_hash_type, old_lsm->lsm_md_hash_type,
1442                        lsm->lsm_md_layout_version,
1443                        old_lsm->lsm_md_layout_version,
1444                        lsm->lsm_md_pool_name,
1445                        old_lsm->lsm_md_pool_name);
1446
1447                 for (idx = 0; idx < old_lsm->lsm_md_stripe_count; idx++) {
1448                         CERROR("%s: sub FIDs in old lsm idx %d, old: "DFID"\n",
1449                                ll_get_fsname(inode->i_sb, NULL, 0), idx,
1450                                PFID(&old_lsm->lsm_md_oinfo[idx].lmo_fid));
1451                 }
1452
1453                 for (idx = 0; idx < lsm->lsm_md_stripe_count; idx++) {
1454                         CERROR("%s: sub FIDs in new lsm idx %d, new: "DFID"\n",
1455                                ll_get_fsname(inode->i_sb, NULL, 0), idx,
1456                                PFID(&lsm->lsm_md_oinfo[idx].lmo_fid));
1457                 }
1458
1459                 RETURN(-EIO);
1460         }
1461
1462         RETURN(0);
1463 }
1464
1465 void ll_clear_inode(struct inode *inode)
1466 {
1467         struct ll_inode_info *lli = ll_i2info(inode);
1468         struct ll_sb_info *sbi = ll_i2sbi(inode);
1469         ENTRY;
1470
1471         CDEBUG(D_VFSTRACE, "VFS Op:inode="DFID"(%p)\n",
1472                PFID(ll_inode2fid(inode)), inode);
1473
1474         if (S_ISDIR(inode->i_mode)) {
1475                 /* these should have been cleared in ll_file_release */
1476                 LASSERT(lli->lli_opendir_key == NULL);
1477                 LASSERT(lli->lli_sai == NULL);
1478                 LASSERT(lli->lli_opendir_pid == 0);
1479         }
1480
1481         md_null_inode(sbi->ll_md_exp, ll_inode2fid(inode));
1482
1483         LASSERT(!lli->lli_open_fd_write_count);
1484         LASSERT(!lli->lli_open_fd_read_count);
1485         LASSERT(!lli->lli_open_fd_exec_count);
1486
1487         if (lli->lli_mds_write_och)
1488                 ll_md_real_close(inode, FMODE_WRITE);
1489         if (lli->lli_mds_exec_och)
1490                 ll_md_real_close(inode, FMODE_EXEC);
1491         if (lli->lli_mds_read_och)
1492                 ll_md_real_close(inode, FMODE_READ);
1493
1494         if (S_ISLNK(inode->i_mode) && lli->lli_symlink_name) {
1495                 OBD_FREE(lli->lli_symlink_name,
1496                          strlen(lli->lli_symlink_name) + 1);
1497                 lli->lli_symlink_name = NULL;
1498         }
1499
1500         ll_xattr_cache_destroy(inode);
1501
1502 #ifdef CONFIG_FS_POSIX_ACL
1503         forget_all_cached_acls(inode);
1504         if (lli->lli_posix_acl) {
1505                 LASSERT(atomic_read(&lli->lli_posix_acl->a_refcount) == 1);
1506                 posix_acl_release(lli->lli_posix_acl);
1507                 lli->lli_posix_acl = NULL;
1508         }
1509 #endif
1510         lli->lli_inode_magic = LLI_INODE_DEAD;
1511
1512         if (S_ISDIR(inode->i_mode))
1513                 ll_dir_clear_lsm_md(inode);
1514         else if (S_ISREG(inode->i_mode) && !is_bad_inode(inode))
1515                 LASSERT(list_empty(&lli->lli_agl_list));
1516
1517         /*
1518          * XXX This has to be done before lsm is freed below, because
1519          * cl_object still uses inode lsm.
1520          */
1521         cl_inode_fini(inode);
1522
1523         EXIT;
1524 }
1525
1526 static int ll_md_setattr(struct dentry *dentry, struct md_op_data *op_data)
1527 {
1528         struct lustre_md md;
1529         struct inode *inode = dentry->d_inode;
1530         struct ll_sb_info *sbi = ll_i2sbi(inode);
1531         struct ptlrpc_request *request = NULL;
1532         int rc, ia_valid;
1533         ENTRY;
1534
1535         op_data = ll_prep_md_op_data(op_data, inode, NULL, NULL, 0, 0,
1536                                      LUSTRE_OPC_ANY, NULL);
1537         if (IS_ERR(op_data))
1538                 RETURN(PTR_ERR(op_data));
1539
1540         rc = md_setattr(sbi->ll_md_exp, op_data, NULL, 0, &request);
1541         if (rc) {
1542                 ptlrpc_req_finished(request);
1543                 if (rc == -ENOENT) {
1544                         clear_nlink(inode);
1545                         /* Unlinked special device node? Or just a race?
1546                          * Pretend we done everything. */
1547                         if (!S_ISREG(inode->i_mode) &&
1548                             !S_ISDIR(inode->i_mode)) {
1549                                 ia_valid = op_data->op_attr.ia_valid;
1550                                 op_data->op_attr.ia_valid &= ~TIMES_SET_FLAGS;
1551                                 rc = simple_setattr(dentry, &op_data->op_attr);
1552                                 op_data->op_attr.ia_valid = ia_valid;
1553                         }
1554                 } else if (rc != -EPERM && rc != -EACCES && rc != -ETXTBSY) {
1555                         CERROR("md_setattr fails: rc = %d\n", rc);
1556                 }
1557                 RETURN(rc);
1558         }
1559
1560         rc = md_get_lustre_md(sbi->ll_md_exp, request, sbi->ll_dt_exp,
1561                               sbi->ll_md_exp, &md);
1562         if (rc) {
1563                 ptlrpc_req_finished(request);
1564                 RETURN(rc);
1565         }
1566
1567         ia_valid = op_data->op_attr.ia_valid;
1568         /* inode size will be in ll_setattr_ost, can't do it now since dirty
1569          * cache is not cleared yet. */
1570         op_data->op_attr.ia_valid &= ~(TIMES_SET_FLAGS | ATTR_SIZE);
1571         if (S_ISREG(inode->i_mode))
1572                 inode_lock(inode);
1573         rc = simple_setattr(dentry, &op_data->op_attr);
1574         if (S_ISREG(inode->i_mode))
1575                 inode_unlock(inode);
1576         op_data->op_attr.ia_valid = ia_valid;
1577
1578         rc = ll_update_inode(inode, &md);
1579         ptlrpc_req_finished(request);
1580
1581         RETURN(rc);
1582 }
1583
1584 /* If this inode has objects allocated to it (lsm != NULL), then the OST
1585  * object(s) determine the file size and mtime.  Otherwise, the MDS will
1586  * keep these values until such a time that objects are allocated for it.
1587  * We do the MDS operations first, as it is checking permissions for us.
1588  * We don't to the MDS RPC if there is nothing that we want to store there,
1589  * otherwise there is no harm in updating mtime/atime on the MDS if we are
1590  * going to do an RPC anyways.
1591  *
1592  * If we are doing a truncate, we will send the mtime and ctime updates
1593  * to the OST with the punch RPC, otherwise we do an explicit setattr RPC.
1594  * I don't believe it is possible to get e.g. ATTR_MTIME_SET and ATTR_SIZE
1595  * at the same time.
1596  *
1597  * In case of HSMimport, we only set attr on MDS.
1598  */
1599 int ll_setattr_raw(struct dentry *dentry, struct iattr *attr, bool hsm_import)
1600 {
1601         struct inode *inode = dentry->d_inode;
1602         struct ll_inode_info *lli = ll_i2info(inode);
1603         struct md_op_data *op_data = NULL;
1604         int rc = 0;
1605         ENTRY;
1606
1607         CDEBUG(D_VFSTRACE, "%s: setattr inode "DFID"(%p) from %llu to %llu, "
1608                "valid %x, hsm_import %d\n",
1609                ll_get_fsname(inode->i_sb, NULL, 0), PFID(&lli->lli_fid),
1610                inode, i_size_read(inode), attr->ia_size, attr->ia_valid,
1611                hsm_import);
1612
1613         if (attr->ia_valid & ATTR_SIZE) {
1614                 /* Check new size against VFS/VM file size limit and rlimit */
1615                 rc = inode_newsize_ok(inode, attr->ia_size);
1616                 if (rc)
1617                         RETURN(rc);
1618
1619                 /* The maximum Lustre file size is variable, based on the
1620                  * OST maximum object size and number of stripes.  This
1621                  * needs another check in addition to the VFS check above. */
1622                 if (attr->ia_size > ll_file_maxbytes(inode)) {
1623                         CDEBUG(D_INODE,"file "DFID" too large %llu > %llu\n",
1624                                PFID(&lli->lli_fid), attr->ia_size,
1625                                ll_file_maxbytes(inode));
1626                         RETURN(-EFBIG);
1627                 }
1628
1629                 attr->ia_valid |= ATTR_MTIME | ATTR_CTIME;
1630         }
1631
1632         /* POSIX: check before ATTR_*TIME_SET set (from inode_change_ok) */
1633         if (attr->ia_valid & TIMES_SET_FLAGS) {
1634                 if ((!uid_eq(current_fsuid(), inode->i_uid)) &&
1635                     !cfs_capable(CFS_CAP_FOWNER))
1636                         RETURN(-EPERM);
1637         }
1638
1639         /* We mark all of the fields "set" so MDS/OST does not re-set them */
1640         if (!(attr->ia_valid & ATTR_CTIME_SET) &&
1641             (attr->ia_valid & ATTR_CTIME)) {
1642                 attr->ia_ctime = current_time(inode);
1643                 attr->ia_valid |= ATTR_CTIME_SET;
1644         }
1645         if (!(attr->ia_valid & ATTR_ATIME_SET) &&
1646             (attr->ia_valid & ATTR_ATIME)) {
1647                 attr->ia_atime = current_time(inode);
1648                 attr->ia_valid |= ATTR_ATIME_SET;
1649         }
1650         if (!(attr->ia_valid & ATTR_MTIME_SET) &&
1651             (attr->ia_valid & ATTR_MTIME)) {
1652                 attr->ia_mtime = current_time(inode);
1653                 attr->ia_valid |= ATTR_MTIME_SET;
1654         }
1655
1656         if (attr->ia_valid & (ATTR_MTIME | ATTR_CTIME))
1657                 CDEBUG(D_INODE, "setting mtime %lu, ctime %lu, now = %llu\n",
1658                        LTIME_S(attr->ia_mtime), LTIME_S(attr->ia_ctime),
1659                        (s64)ktime_get_real_seconds());
1660
1661         if (S_ISREG(inode->i_mode)) {
1662                 if (attr->ia_valid & ATTR_SIZE)
1663                         inode_dio_write_done(inode);
1664                 inode_unlock(inode);
1665         }
1666
1667         /* We always do an MDS RPC, even if we're only changing the size;
1668          * only the MDS knows whether truncate() should fail with -ETXTBUSY */
1669
1670         OBD_ALLOC_PTR(op_data);
1671         if (op_data == NULL)
1672                 GOTO(out, rc = -ENOMEM);
1673
1674         if (!hsm_import && attr->ia_valid & ATTR_SIZE) {
1675                 /* If we are changing file size, file content is
1676                  * modified, flag it. */
1677                 attr->ia_valid |= MDS_OPEN_OWNEROVERRIDE;
1678                 op_data->op_bias |= MDS_DATA_MODIFIED;
1679                 ll_file_clear_flag(lli, LLIF_DATA_MODIFIED);
1680         }
1681
1682         op_data->op_attr = *attr;
1683
1684         rc = ll_md_setattr(dentry, op_data);
1685         if (rc)
1686                 GOTO(out, rc);
1687
1688         if (!S_ISREG(inode->i_mode) || hsm_import)
1689                 GOTO(out, rc = 0);
1690
1691         if (attr->ia_valid & (ATTR_SIZE |
1692                               ATTR_ATIME | ATTR_ATIME_SET |
1693                               ATTR_MTIME | ATTR_MTIME_SET |
1694                               ATTR_CTIME | ATTR_CTIME_SET)) {
1695                 /* For truncate and utimes sending attributes to OSTs, setting
1696                  * mtime/atime to the past will be performed under PW [0:EOF]
1697                  * extent lock (new_size:EOF for truncate).  It may seem
1698                  * excessive to send mtime/atime updates to OSTs when not
1699                  * setting times to past, but it is necessary due to possible
1700                  * time de-synchronization between MDT inode and OST objects */
1701                 rc = cl_setattr_ost(lli->lli_clob, attr, 0);
1702         }
1703
1704         /* If the file was restored, it needs to set dirty flag.
1705          *
1706          * We've already sent MDS_DATA_MODIFIED flag in
1707          * ll_md_setattr() for truncate. However, the MDT refuses to
1708          * set the HS_DIRTY flag on released files, so we have to set
1709          * it again if the file has been restored. Please check how
1710          * LLIF_DATA_MODIFIED is set in vvp_io_setattr_fini().
1711          *
1712          * Please notice that if the file is not released, the previous
1713          * MDS_DATA_MODIFIED has taken effect and usually
1714          * LLIF_DATA_MODIFIED is not set(see vvp_io_setattr_fini()).
1715          * This way we can save an RPC for common open + trunc
1716          * operation. */
1717         if (ll_file_test_and_clear_flag(lli, LLIF_DATA_MODIFIED)) {
1718                 struct hsm_state_set hss = {
1719                         .hss_valid = HSS_SETMASK,
1720                         .hss_setmask = HS_DIRTY,
1721                 };
1722                 int rc2;
1723
1724                 rc2 = ll_hsm_state_set(inode, &hss);
1725                 /* truncate and write can happen at the same time, so that
1726                  * the file can be set modified even though the file is not
1727                  * restored from released state, and ll_hsm_state_set() is
1728                  * not applicable for the file, and rc2 < 0 is normal in this
1729                  * case. */
1730                 if (rc2 < 0)
1731                         CDEBUG(D_INFO, DFID "HSM set dirty failed: rc2 = %d\n",
1732                                PFID(ll_inode2fid(inode)), rc2);
1733         }
1734
1735         EXIT;
1736 out:
1737         if (op_data != NULL)
1738                 ll_finish_md_op_data(op_data);
1739
1740         if (S_ISREG(inode->i_mode)) {
1741                 inode_lock(inode);
1742                 if ((attr->ia_valid & ATTR_SIZE) && !hsm_import)
1743                         inode_dio_wait(inode);
1744                 /* Once we've got the i_mutex, it's safe to set the S_NOSEC
1745                  * flag.  ll_update_inode (called from ll_md_setattr), clears
1746                  * inode flags, so there is a gap where S_NOSEC is not set.
1747                  * This can cause a writer to take the i_mutex unnecessarily,
1748                  * but this is safe to do and should be rare. */
1749                 inode_has_no_xattr(inode);
1750         }
1751
1752         ll_stats_ops_tally(ll_i2sbi(inode), (attr->ia_valid & ATTR_SIZE) ?
1753                         LPROC_LL_TRUNC : LPROC_LL_SETATTR, 1);
1754
1755         return rc;
1756 }
1757
1758 int ll_setattr(struct dentry *de, struct iattr *attr)
1759 {
1760         int mode = de->d_inode->i_mode;
1761
1762         if ((attr->ia_valid & (ATTR_CTIME|ATTR_SIZE|ATTR_MODE)) ==
1763                               (ATTR_CTIME|ATTR_SIZE|ATTR_MODE))
1764                 attr->ia_valid |= MDS_OPEN_OWNEROVERRIDE;
1765
1766         if (((attr->ia_valid & (ATTR_MODE|ATTR_FORCE|ATTR_SIZE)) ==
1767                                (ATTR_SIZE|ATTR_MODE)) &&
1768             (((mode & S_ISUID) && !(attr->ia_mode & S_ISUID)) ||
1769              (((mode & (S_ISGID|S_IXGRP)) == (S_ISGID|S_IXGRP)) &&
1770               !(attr->ia_mode & S_ISGID))))
1771                 attr->ia_valid |= ATTR_FORCE;
1772
1773         if ((attr->ia_valid & ATTR_MODE) &&
1774             (mode & S_ISUID) &&
1775             !(attr->ia_mode & S_ISUID) &&
1776             !(attr->ia_valid & ATTR_KILL_SUID))
1777                 attr->ia_valid |= ATTR_KILL_SUID;
1778
1779         if ((attr->ia_valid & ATTR_MODE) &&
1780             ((mode & (S_ISGID|S_IXGRP)) == (S_ISGID|S_IXGRP)) &&
1781             !(attr->ia_mode & S_ISGID) &&
1782             !(attr->ia_valid & ATTR_KILL_SGID))
1783                 attr->ia_valid |= ATTR_KILL_SGID;
1784
1785         /* avoid polluted from ATTR_TIMES_SET,
1786          * projid is not expected to be set here */
1787         attr->ia_valid &= ~MDS_ATTR_PROJID;
1788
1789         return ll_setattr_raw(de, attr, false);
1790 }
1791
1792 int ll_statfs_internal(struct ll_sb_info *sbi, struct obd_statfs *osfs,
1793                        u32 flags)
1794 {
1795         struct obd_statfs obd_osfs;
1796         time64_t max_age;
1797         int rc;
1798
1799         ENTRY;
1800         max_age = ktime_get_seconds() - OBD_STATFS_CACHE_SECONDS;
1801
1802         rc = obd_statfs(NULL, sbi->ll_md_exp, osfs, max_age, flags);
1803         if (rc) {
1804                 CERROR("md_statfs fails: rc = %d\n", rc);
1805                 RETURN(rc);
1806         }
1807
1808         osfs->os_type = LL_SUPER_MAGIC;
1809
1810         CDEBUG(D_SUPER, "MDC blocks %llu/%llu objects %llu/%llu\n",
1811                osfs->os_bavail, osfs->os_blocks, osfs->os_ffree,osfs->os_files);
1812
1813         if (sbi->ll_flags & LL_SBI_LAZYSTATFS)
1814                 flags |= OBD_STATFS_NODELAY;
1815
1816         rc = obd_statfs(NULL, sbi->ll_dt_exp, &obd_osfs, max_age, flags);
1817         if (rc) {
1818                 CERROR("obd_statfs fails: rc = %d\n", rc);
1819                 RETURN(rc);
1820         }
1821
1822         CDEBUG(D_SUPER, "OSC blocks %llu/%llu objects %llu/%llu\n",
1823                obd_osfs.os_bavail, obd_osfs.os_blocks, obd_osfs.os_ffree,
1824                obd_osfs.os_files);
1825
1826         osfs->os_bsize = obd_osfs.os_bsize;
1827         osfs->os_blocks = obd_osfs.os_blocks;
1828         osfs->os_bfree = obd_osfs.os_bfree;
1829         osfs->os_bavail = obd_osfs.os_bavail;
1830
1831         /* If we don't have as many objects free on the OST as inodes
1832          * on the MDS, we reduce the total number of inodes to
1833          * compensate, so that the "inodes in use" number is correct.
1834          */
1835         if (obd_osfs.os_ffree < osfs->os_ffree) {
1836                 osfs->os_files = (osfs->os_files - osfs->os_ffree) +
1837                         obd_osfs.os_ffree;
1838                 osfs->os_ffree = obd_osfs.os_ffree;
1839         }
1840
1841         RETURN(rc);
1842 }
1843 int ll_statfs(struct dentry *de, struct kstatfs *sfs)
1844 {
1845         struct super_block *sb = de->d_sb;
1846         struct obd_statfs osfs;
1847         __u64 fsid = huge_encode_dev(sb->s_dev);
1848         int rc;
1849
1850         CDEBUG(D_VFSTRACE, "VFS Op: at %llu jiffies\n", get_jiffies_64());
1851         ll_stats_ops_tally(ll_s2sbi(sb), LPROC_LL_STAFS, 1);
1852
1853         /* Some amount of caching on the client is allowed */
1854         rc = ll_statfs_internal(ll_s2sbi(sb), &osfs, 0);
1855         if (rc)
1856                 return rc;
1857
1858         statfs_unpack(sfs, &osfs);
1859
1860         /* We need to downshift for all 32-bit kernels, because we can't
1861          * tell if the kernel is being called via sys_statfs64() or not.
1862          * Stop before overflowing f_bsize - in which case it is better
1863          * to just risk EOVERFLOW if caller is using old sys_statfs(). */
1864         if (sizeof(long) < 8) {
1865                 while (osfs.os_blocks > ~0UL && sfs->f_bsize < 0x40000000) {
1866                         sfs->f_bsize <<= 1;
1867
1868                         osfs.os_blocks >>= 1;
1869                         osfs.os_bfree >>= 1;
1870                         osfs.os_bavail >>= 1;
1871                 }
1872         }
1873
1874         sfs->f_blocks = osfs.os_blocks;
1875         sfs->f_bfree = osfs.os_bfree;
1876         sfs->f_bavail = osfs.os_bavail;
1877         sfs->f_fsid.val[0] = (__u32)fsid;
1878         sfs->f_fsid.val[1] = (__u32)(fsid >> 32);
1879         return 0;
1880 }
1881
1882 void ll_inode_size_lock(struct inode *inode)
1883 {
1884         struct ll_inode_info *lli;
1885
1886         LASSERT(!S_ISDIR(inode->i_mode));
1887
1888         lli = ll_i2info(inode);
1889         mutex_lock(&lli->lli_size_mutex);
1890 }
1891
1892 void ll_inode_size_unlock(struct inode *inode)
1893 {
1894         struct ll_inode_info *lli;
1895
1896         lli = ll_i2info(inode);
1897         mutex_unlock(&lli->lli_size_mutex);
1898 }
1899
1900 int ll_update_inode(struct inode *inode, struct lustre_md *md)
1901 {
1902         struct ll_inode_info *lli = ll_i2info(inode);
1903         struct mdt_body *body = md->body;
1904         struct ll_sb_info *sbi = ll_i2sbi(inode);
1905         int rc = 0;
1906
1907         if (body->mbo_valid & OBD_MD_FLEASIZE) {
1908                 rc = cl_file_inode_init(inode, md);
1909                 if (rc)
1910                         return rc;
1911         }
1912
1913         if (S_ISDIR(inode->i_mode)) {
1914                 rc = ll_update_lsm_md(inode, md);
1915                 if (rc != 0)
1916                         return rc;
1917         }
1918
1919 #ifdef CONFIG_FS_POSIX_ACL
1920         if (body->mbo_valid & OBD_MD_FLACL) {
1921                 spin_lock(&lli->lli_lock);
1922                 if (lli->lli_posix_acl)
1923                         posix_acl_release(lli->lli_posix_acl);
1924                 lli->lli_posix_acl = md->posix_acl;
1925                 spin_unlock(&lli->lli_lock);
1926         }
1927 #endif
1928         inode->i_ino = cl_fid_build_ino(&body->mbo_fid1,
1929                                         sbi->ll_flags & LL_SBI_32BIT_API);
1930         inode->i_generation = cl_fid_build_gen(&body->mbo_fid1);
1931
1932         if (body->mbo_valid & OBD_MD_FLATIME) {
1933                 if (body->mbo_atime > LTIME_S(inode->i_atime))
1934                         LTIME_S(inode->i_atime) = body->mbo_atime;
1935                 lli->lli_atime = body->mbo_atime;
1936         }
1937
1938         if (body->mbo_valid & OBD_MD_FLMTIME) {
1939                 if (body->mbo_mtime > LTIME_S(inode->i_mtime)) {
1940                         CDEBUG(D_INODE, "setting ino %lu mtime from %lu "
1941                                "to %llu\n", inode->i_ino,
1942                                LTIME_S(inode->i_mtime), body->mbo_mtime);
1943                         LTIME_S(inode->i_mtime) = body->mbo_mtime;
1944                 }
1945                 lli->lli_mtime = body->mbo_mtime;
1946         }
1947
1948         if (body->mbo_valid & OBD_MD_FLCTIME) {
1949                 if (body->mbo_ctime > LTIME_S(inode->i_ctime))
1950                         LTIME_S(inode->i_ctime) = body->mbo_ctime;
1951                 lli->lli_ctime = body->mbo_ctime;
1952         }
1953
1954         /* Clear i_flags to remove S_NOSEC before permissions are updated */
1955         if (body->mbo_valid & OBD_MD_FLFLAGS)
1956                 inode->i_flags = ll_ext_to_inode_flags(body->mbo_flags);
1957         if (body->mbo_valid & OBD_MD_FLMODE)
1958                 inode->i_mode = (inode->i_mode & S_IFMT) |
1959                                 (body->mbo_mode & ~S_IFMT);
1960
1961         if (body->mbo_valid & OBD_MD_FLTYPE)
1962                 inode->i_mode = (inode->i_mode & ~S_IFMT) |
1963                                 (body->mbo_mode & S_IFMT);
1964
1965         LASSERT(inode->i_mode != 0);
1966         if (S_ISREG(inode->i_mode))
1967                 inode->i_blkbits = min(PTLRPC_MAX_BRW_BITS + 1,
1968                                        LL_MAX_BLKSIZE_BITS);
1969         else
1970                 inode->i_blkbits = inode->i_sb->s_blocksize_bits;
1971
1972         if (body->mbo_valid & OBD_MD_FLUID)
1973                 inode->i_uid = make_kuid(&init_user_ns, body->mbo_uid);
1974         if (body->mbo_valid & OBD_MD_FLGID)
1975                 inode->i_gid = make_kgid(&init_user_ns, body->mbo_gid);
1976         if (body->mbo_valid & OBD_MD_FLPROJID)
1977                 lli->lli_projid = body->mbo_projid;
1978         if (body->mbo_valid & OBD_MD_FLNLINK)
1979                 set_nlink(inode, body->mbo_nlink);
1980         if (body->mbo_valid & OBD_MD_FLRDEV)
1981                 inode->i_rdev = old_decode_dev(body->mbo_rdev);
1982
1983         if (body->mbo_valid & OBD_MD_FLID) {
1984                 /* FID shouldn't be changed! */
1985                 if (fid_is_sane(&lli->lli_fid)) {
1986                         LASSERTF(lu_fid_eq(&lli->lli_fid, &body->mbo_fid1),
1987                                  "Trying to change FID "DFID
1988                                  " to the "DFID", inode "DFID"(%p)\n",
1989                                  PFID(&lli->lli_fid), PFID(&body->mbo_fid1),
1990                                  PFID(ll_inode2fid(inode)), inode);
1991                 } else {
1992                         lli->lli_fid = body->mbo_fid1;
1993                 }
1994         }
1995
1996         LASSERT(fid_seq(&lli->lli_fid) != 0);
1997
1998         if (body->mbo_valid & OBD_MD_FLSIZE) {
1999                 i_size_write(inode, body->mbo_size);
2000
2001                 CDEBUG(D_VFSTRACE, "inode="DFID", updating i_size %llu\n",
2002                        PFID(ll_inode2fid(inode)),
2003                        (unsigned long long)body->mbo_size);
2004
2005                 if (body->mbo_valid & OBD_MD_FLBLOCKS)
2006                         inode->i_blocks = body->mbo_blocks;
2007         }
2008
2009         if (body->mbo_valid & OBD_MD_TSTATE) {
2010                 /* Set LLIF_FILE_RESTORING if restore ongoing and
2011                  * clear it when done to ensure to start again
2012                  * glimpsing updated attrs
2013                  */
2014                 if (body->mbo_t_state & MS_RESTORE)
2015                         ll_file_set_flag(lli, LLIF_FILE_RESTORING);
2016                 else
2017                         ll_file_clear_flag(lli, LLIF_FILE_RESTORING);
2018         }
2019
2020         return 0;
2021 }
2022
2023 int ll_read_inode2(struct inode *inode, void *opaque)
2024 {
2025         struct lustre_md *md = opaque;
2026         struct ll_inode_info *lli = ll_i2info(inode);
2027         int     rc;
2028         ENTRY;
2029
2030         CDEBUG(D_VFSTRACE, "VFS Op:inode="DFID"(%p)\n",
2031                PFID(&lli->lli_fid), inode);
2032
2033         /* Core attributes from the MDS first.  This is a new inode, and
2034          * the VFS doesn't zero times in the core inode so we have to do
2035          * it ourselves.  They will be overwritten by either MDS or OST
2036          * attributes - we just need to make sure they aren't newer. */
2037         LTIME_S(inode->i_mtime) = 0;
2038         LTIME_S(inode->i_atime) = 0;
2039         LTIME_S(inode->i_ctime) = 0;
2040         inode->i_rdev = 0;
2041         rc = ll_update_inode(inode, md);
2042         if (rc != 0)
2043                 RETURN(rc);
2044
2045         /* OIDEBUG(inode); */
2046
2047 #ifdef HAVE_BACKING_DEV_INFO
2048         /* initializing backing dev info. */
2049         inode->i_mapping->backing_dev_info = &s2lsi(inode->i_sb)->lsi_bdi;
2050 #endif
2051         if (S_ISREG(inode->i_mode)) {
2052                 struct ll_sb_info *sbi = ll_i2sbi(inode);
2053                 inode->i_op = &ll_file_inode_operations;
2054                 inode->i_fop = sbi->ll_fop;
2055                 inode->i_mapping->a_ops = (struct address_space_operations *)&ll_aops;
2056                 EXIT;
2057         } else if (S_ISDIR(inode->i_mode)) {
2058                 inode->i_op = &ll_dir_inode_operations;
2059                 inode->i_fop = &ll_dir_operations;
2060                 EXIT;
2061         } else if (S_ISLNK(inode->i_mode)) {
2062                 inode->i_op = &ll_fast_symlink_inode_operations;
2063                 EXIT;
2064         } else {
2065                 inode->i_op = &ll_special_inode_operations;
2066
2067                 init_special_inode(inode, inode->i_mode,
2068                                    inode->i_rdev);
2069
2070                 EXIT;
2071         }
2072
2073         return 0;
2074 }
2075
2076 void ll_delete_inode(struct inode *inode)
2077 {
2078         struct ll_inode_info *lli = ll_i2info(inode);
2079         ENTRY;
2080
2081         if (S_ISREG(inode->i_mode) && lli->lli_clob != NULL)
2082                 /* It is last chance to write out dirty pages,
2083                  * otherwise we may lose data while umount */
2084                 cl_sync_file_range(inode, 0, OBD_OBJECT_EOF, CL_FSYNC_LOCAL, 1);
2085
2086         truncate_inode_pages_final(&inode->i_data);
2087
2088         LASSERTF(inode->i_data.nrpages == 0, "inode="DFID"(%p) nrpages=%lu, "
2089                  "see https://jira.hpdd.intel.com/browse/LU-118\n",
2090                  PFID(ll_inode2fid(inode)), inode, inode->i_data.nrpages);
2091
2092 #ifdef HAVE_SBOPS_EVICT_INODE
2093         ll_clear_inode(inode);
2094 #endif
2095         clear_inode(inode);
2096
2097         EXIT;
2098 }
2099
2100 int ll_iocontrol(struct inode *inode, struct file *file,
2101                  unsigned int cmd, unsigned long arg)
2102 {
2103         struct ll_sb_info *sbi = ll_i2sbi(inode);
2104         struct ptlrpc_request *req = NULL;
2105         int rc, flags = 0;
2106         ENTRY;
2107
2108         switch (cmd) {
2109         case FS_IOC_GETFLAGS: {
2110                 struct mdt_body *body;
2111                 struct md_op_data *op_data;
2112
2113                 op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL,
2114                                              0, 0, LUSTRE_OPC_ANY,
2115                                              NULL);
2116                 if (IS_ERR(op_data))
2117                         RETURN(PTR_ERR(op_data));
2118
2119                 op_data->op_valid = OBD_MD_FLFLAGS;
2120                 rc = md_getattr(sbi->ll_md_exp, op_data, &req);
2121                 ll_finish_md_op_data(op_data);
2122                 if (rc) {
2123                         CERROR("%s: failure inode "DFID": rc = %d\n",
2124                                sbi->ll_md_exp->exp_obd->obd_name,
2125                                PFID(ll_inode2fid(inode)), rc);
2126                         RETURN(-abs(rc));
2127                 }
2128
2129                 body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
2130
2131                 flags = body->mbo_flags;
2132
2133                 ptlrpc_req_finished(req);
2134
2135                 RETURN(put_user(flags, (int __user *)arg));
2136         }
2137         case FS_IOC_SETFLAGS: {
2138                 struct iattr *attr;
2139                 struct md_op_data *op_data;
2140                 struct cl_object *obj;
2141
2142                 if (get_user(flags, (int __user *)arg))
2143                         RETURN(-EFAULT);
2144
2145                 op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL, 0, 0,
2146                                              LUSTRE_OPC_ANY, NULL);
2147                 if (IS_ERR(op_data))
2148                         RETURN(PTR_ERR(op_data));
2149
2150                 op_data->op_attr_flags = flags;
2151                 op_data->op_attr.ia_valid |= ATTR_ATTR_FLAG;
2152                 rc = md_setattr(sbi->ll_md_exp, op_data, NULL, 0, &req);
2153                 ll_finish_md_op_data(op_data);
2154                 ptlrpc_req_finished(req);
2155                 if (rc)
2156                         RETURN(rc);
2157
2158                 inode->i_flags = ll_ext_to_inode_flags(flags);
2159
2160                 obj = ll_i2info(inode)->lli_clob;
2161                 if (obj == NULL)
2162                         RETURN(0);
2163
2164                 OBD_ALLOC_PTR(attr);
2165                 if (attr == NULL)
2166                         RETURN(-ENOMEM);
2167
2168                 attr->ia_valid = ATTR_ATTR_FLAG;
2169                 rc = cl_setattr_ost(obj, attr, flags);
2170
2171                 OBD_FREE_PTR(attr);
2172                 RETURN(rc);
2173         }
2174         default:
2175                 RETURN(-ENOSYS);
2176         }
2177
2178         RETURN(0);
2179 }
2180
2181 int ll_flush_ctx(struct inode *inode)
2182 {
2183         struct ll_sb_info  *sbi = ll_i2sbi(inode);
2184
2185         CDEBUG(D_SEC, "flush context for user %d\n",
2186                from_kuid(&init_user_ns, current_uid()));
2187
2188         obd_set_info_async(NULL, sbi->ll_md_exp,
2189                            sizeof(KEY_FLUSH_CTX), KEY_FLUSH_CTX,
2190                            0, NULL, NULL);
2191         obd_set_info_async(NULL, sbi->ll_dt_exp,
2192                            sizeof(KEY_FLUSH_CTX), KEY_FLUSH_CTX,
2193                            0, NULL, NULL);
2194         return 0;
2195 }
2196
2197 /* umount -f client means force down, don't save state */
2198 void ll_umount_begin(struct super_block *sb)
2199 {
2200         struct ll_sb_info *sbi = ll_s2sbi(sb);
2201         struct obd_device *obd;
2202         struct obd_ioctl_data *ioc_data;
2203         struct l_wait_info lwi;
2204         wait_queue_head_t waitq;
2205         ENTRY;
2206
2207         CDEBUG(D_VFSTRACE, "VFS Op: superblock %p count %d active %d\n", sb,
2208                sb->s_count, atomic_read(&sb->s_active));
2209
2210         obd = class_exp2obd(sbi->ll_md_exp);
2211         if (obd == NULL) {
2212                 CERROR("Invalid MDC connection handle %#llx\n",
2213                        sbi->ll_md_exp->exp_handle.h_cookie);
2214                 EXIT;
2215                 return;
2216         }
2217         obd->obd_force = 1;
2218
2219         obd = class_exp2obd(sbi->ll_dt_exp);
2220         if (obd == NULL) {
2221                 CERROR("Invalid LOV connection handle %#llx\n",
2222                        sbi->ll_dt_exp->exp_handle.h_cookie);
2223                 EXIT;
2224                 return;
2225         }
2226         obd->obd_force = 1;
2227
2228         OBD_ALLOC_PTR(ioc_data);
2229         if (ioc_data) {
2230                 obd_iocontrol(IOC_OSC_SET_ACTIVE, sbi->ll_md_exp,
2231                               sizeof *ioc_data, ioc_data, NULL);
2232
2233                 obd_iocontrol(IOC_OSC_SET_ACTIVE, sbi->ll_dt_exp,
2234                               sizeof *ioc_data, ioc_data, NULL);
2235
2236                 OBD_FREE_PTR(ioc_data);
2237         }
2238
2239         /* Really, we'd like to wait until there are no requests outstanding,
2240          * and then continue.  For now, we just periodically checking for vfs
2241          * to decrement mnt_cnt and hope to finish it within 10sec.
2242          */
2243         init_waitqueue_head(&waitq);
2244         lwi = LWI_TIMEOUT_INTERVAL(cfs_time_seconds(10),
2245                                    cfs_time_seconds(1), NULL, NULL);
2246         l_wait_event(waitq, may_umount(sbi->ll_mnt.mnt), &lwi);
2247
2248         EXIT;
2249 }
2250
2251 int ll_remount_fs(struct super_block *sb, int *flags, char *data)
2252 {
2253         struct ll_sb_info *sbi = ll_s2sbi(sb);
2254         char *profilenm = get_profile_name(sb);
2255         int err;
2256         __u32 read_only;
2257
2258         if ((*flags & MS_RDONLY) != (sb->s_flags & MS_RDONLY)) {
2259                 read_only = *flags & MS_RDONLY;
2260                 err = obd_set_info_async(NULL, sbi->ll_md_exp,
2261                                          sizeof(KEY_READ_ONLY),
2262                                          KEY_READ_ONLY, sizeof(read_only),
2263                                          &read_only, NULL);
2264                 if (err) {
2265                         LCONSOLE_WARN("Failed to remount %s %s (%d)\n",
2266                                       profilenm, read_only ?
2267                                       "read-only" : "read-write", err);
2268                         return err;
2269                 }
2270
2271                 if (read_only)
2272                         sb->s_flags |= MS_RDONLY;
2273                 else
2274                         sb->s_flags &= ~MS_RDONLY;
2275
2276                 if (sbi->ll_flags & LL_SBI_VERBOSE)
2277                         LCONSOLE_WARN("Remounted %s %s\n", profilenm,
2278                                       read_only ?  "read-only" : "read-write");
2279         }
2280         return 0;
2281 }
2282
2283 /**
2284  * Cleanup the open handle that is cached on MDT-side.
2285  *
2286  * For open case, the client side open handling thread may hit error
2287  * after the MDT grant the open. Under such case, the client should
2288  * send close RPC to the MDT as cleanup; otherwise, the open handle
2289  * on the MDT will be leaked there until the client umount or evicted.
2290  *
2291  * In further, if someone unlinked the file, because the open handle
2292  * holds the reference on such file/object, then it will block the
2293  * subsequent threads that want to locate such object via FID.
2294  *
2295  * \param[in] sb        super block for this file-system
2296  * \param[in] open_req  pointer to the original open request
2297  */
2298 void ll_open_cleanup(struct super_block *sb, struct ptlrpc_request *open_req)
2299 {
2300         struct mdt_body                 *body;
2301         struct md_op_data               *op_data;
2302         struct ptlrpc_request           *close_req = NULL;
2303         struct obd_export               *exp       = ll_s2sbi(sb)->ll_md_exp;
2304         ENTRY;
2305
2306         body = req_capsule_server_get(&open_req->rq_pill, &RMF_MDT_BODY);
2307         OBD_ALLOC_PTR(op_data);
2308         if (op_data == NULL) {
2309                 CWARN("%s: cannot allocate op_data to release open handle for "
2310                       DFID"\n",
2311                       ll_get_fsname(sb, NULL, 0), PFID(&body->mbo_fid1));
2312
2313                 RETURN_EXIT;
2314         }
2315
2316         op_data->op_fid1 = body->mbo_fid1;
2317         op_data->op_handle = body->mbo_handle;
2318         op_data->op_mod_time = ktime_get_real_seconds();
2319         md_close(exp, op_data, NULL, &close_req);
2320         ptlrpc_req_finished(close_req);
2321         ll_finish_md_op_data(op_data);
2322
2323         EXIT;
2324 }
2325
2326 int ll_prep_inode(struct inode **inode, struct ptlrpc_request *req,
2327                   struct super_block *sb, struct lookup_intent *it)
2328 {
2329         struct ll_sb_info *sbi = NULL;
2330         struct lustre_md md = { NULL };
2331         int rc;
2332         ENTRY;
2333
2334         LASSERT(*inode || sb);
2335         sbi = sb ? ll_s2sbi(sb) : ll_i2sbi(*inode);
2336         rc = md_get_lustre_md(sbi->ll_md_exp, req, sbi->ll_dt_exp,
2337                               sbi->ll_md_exp, &md);
2338         if (rc != 0)
2339                 GOTO(cleanup, rc);
2340
2341         if (*inode) {
2342                 rc = ll_update_inode(*inode, &md);
2343                 if (rc != 0)
2344                         GOTO(out, rc);
2345         } else {
2346                 LASSERT(sb != NULL);
2347
2348                 /*
2349                  * At this point server returns to client's same fid as client
2350                  * generated for creating. So using ->fid1 is okay here.
2351                  */
2352                 if (!fid_is_sane(&md.body->mbo_fid1)) {
2353                         CERROR("%s: Fid is insane "DFID"\n",
2354                                 ll_get_fsname(sb, NULL, 0),
2355                                 PFID(&md.body->mbo_fid1));
2356                         GOTO(out, rc = -EINVAL);
2357                 }
2358
2359                 *inode = ll_iget(sb, cl_fid_build_ino(&md.body->mbo_fid1,
2360                                              sbi->ll_flags & LL_SBI_32BIT_API),
2361                                  &md);
2362                 if (IS_ERR(*inode)) {
2363 #ifdef CONFIG_FS_POSIX_ACL
2364                         if (md.posix_acl) {
2365                                 posix_acl_release(md.posix_acl);
2366                                 md.posix_acl = NULL;
2367                         }
2368 #endif
2369                         rc = IS_ERR(*inode) ? PTR_ERR(*inode) : -ENOMEM;
2370                         *inode = NULL;
2371                         CERROR("new_inode -fatal: rc %d\n", rc);
2372                         GOTO(out, rc);
2373                 }
2374         }
2375
2376         /* Handling piggyback layout lock.
2377          * Layout lock can be piggybacked by getattr and open request.
2378          * The lsm can be applied to inode only if it comes with a layout lock
2379          * otherwise correct layout may be overwritten, for example:
2380          * 1. proc1: mdt returns a lsm but not granting layout
2381          * 2. layout was changed by another client
2382          * 3. proc2: refresh layout and layout lock granted
2383          * 4. proc1: to apply a stale layout */
2384         if (it != NULL && it->it_lock_mode != 0) {
2385                 struct lustre_handle lockh;
2386                 struct ldlm_lock *lock;
2387
2388                 lockh.cookie = it->it_lock_handle;
2389                 lock = ldlm_handle2lock(&lockh);
2390                 LASSERT(lock != NULL);
2391                 if (ldlm_has_layout(lock)) {
2392                         struct cl_object_conf conf;
2393
2394                         memset(&conf, 0, sizeof(conf));
2395                         conf.coc_opc = OBJECT_CONF_SET;
2396                         conf.coc_inode = *inode;
2397                         conf.coc_lock = lock;
2398                         conf.u.coc_layout = md.layout;
2399                         (void)ll_layout_conf(*inode, &conf);
2400                 }
2401                 LDLM_LOCK_PUT(lock);
2402         }
2403
2404         GOTO(out, rc = 0);
2405
2406 out:
2407         md_free_lustre_md(sbi->ll_md_exp, &md);
2408
2409 cleanup:
2410         if (rc != 0 && it != NULL && it->it_op & IT_OPEN)
2411                 ll_open_cleanup(sb != NULL ? sb : (*inode)->i_sb, req);
2412
2413         return rc;
2414 }
2415
2416 int ll_obd_statfs(struct inode *inode, void __user *arg)
2417 {
2418         struct ll_sb_info *sbi = NULL;
2419         struct obd_export *exp;
2420         char *buf = NULL;
2421         struct obd_ioctl_data *data = NULL;
2422         __u32 type;
2423         int len = 0, rc;
2424
2425         if (!inode || !(sbi = ll_i2sbi(inode)))
2426                 GOTO(out_statfs, rc = -EINVAL);
2427
2428         rc = obd_ioctl_getdata(&buf, &len, arg);
2429         if (rc)
2430                 GOTO(out_statfs, rc);
2431
2432         data = (void*)buf;
2433         if (!data->ioc_inlbuf1 || !data->ioc_inlbuf2 ||
2434             !data->ioc_pbuf1 || !data->ioc_pbuf2)
2435                 GOTO(out_statfs, rc = -EINVAL);
2436
2437         if (data->ioc_inllen1 != sizeof(__u32) ||
2438             data->ioc_inllen2 != sizeof(__u32) ||
2439             data->ioc_plen1 != sizeof(struct obd_statfs) ||
2440             data->ioc_plen2 != sizeof(struct obd_uuid))
2441                 GOTO(out_statfs, rc = -EINVAL);
2442
2443         memcpy(&type, data->ioc_inlbuf1, sizeof(__u32));
2444         if (type & LL_STATFS_LMV)
2445                 exp = sbi->ll_md_exp;
2446         else if (type & LL_STATFS_LOV)
2447                 exp = sbi->ll_dt_exp;
2448         else
2449                 GOTO(out_statfs, rc = -ENODEV);
2450
2451         rc = obd_iocontrol(IOC_OBD_STATFS, exp, len, buf, NULL);
2452         if (rc)
2453                 GOTO(out_statfs, rc);
2454 out_statfs:
2455         OBD_FREE_LARGE(buf, len);
2456         return rc;
2457 }
2458
2459 int ll_process_config(struct lustre_cfg *lcfg)
2460 {
2461         struct super_block *sb;
2462         unsigned long x;
2463         int rc = 0;
2464         char *ptr;
2465
2466         /* The instance name contains the sb: lustre-client-aacfe000 */
2467         ptr = strrchr(lustre_cfg_string(lcfg, 0), '-');
2468         if (!ptr || !*(++ptr))
2469                 return -EINVAL;
2470         if (sscanf(ptr, "%lx", &x) != 1)
2471                 return -EINVAL;
2472         sb = (struct super_block *)x;
2473         /* This better be a real Lustre superblock! */
2474         LASSERT(s2lsi(sb)->lsi_lmd->lmd_magic == LMD_MAGIC);
2475
2476         /* Note we have not called client_common_fill_super yet, so
2477            proc fns must be able to handle that! */
2478         rc = class_process_proc_param(PARAM_LLITE, lprocfs_llite_obd_vars,
2479                                       lcfg, sb);
2480         if (rc > 0)
2481                 rc = 0;
2482         return rc;
2483 }
2484
2485 /* this function prepares md_op_data hint for passing it down to MD stack. */
2486 struct md_op_data *ll_prep_md_op_data(struct md_op_data *op_data,
2487                                       struct inode *i1, struct inode *i2,
2488                                       const char *name, size_t namelen,
2489                                       __u32 mode, __u32 opc, void *data)
2490 {
2491         LASSERT(i1 != NULL);
2492
2493         if (name == NULL) {
2494                 /* Do not reuse namelen for something else. */
2495                 if (namelen != 0)
2496                         return ERR_PTR(-EINVAL);
2497         } else {
2498                 if (namelen > ll_i2sbi(i1)->ll_namelen)
2499                         return ERR_PTR(-ENAMETOOLONG);
2500
2501                 if (!lu_name_is_valid_2(name, namelen))
2502                         return ERR_PTR(-EINVAL);
2503         }
2504
2505         if (op_data == NULL)
2506                 OBD_ALLOC_PTR(op_data);
2507
2508         if (op_data == NULL)
2509                 return ERR_PTR(-ENOMEM);
2510
2511         ll_i2gids(op_data->op_suppgids, i1, i2);
2512         op_data->op_fid1 = *ll_inode2fid(i1);
2513         op_data->op_default_stripe_offset = -1;
2514         if (S_ISDIR(i1->i_mode)) {
2515                 op_data->op_mea1 = ll_i2info(i1)->lli_lsm_md;
2516                 if (opc == LUSTRE_OPC_MKDIR)
2517                         op_data->op_default_stripe_offset =
2518                                    ll_i2info(i1)->lli_def_stripe_offset;
2519         }
2520
2521         if (i2) {
2522                 op_data->op_fid2 = *ll_inode2fid(i2);
2523                 if (S_ISDIR(i2->i_mode))
2524                         op_data->op_mea2 = ll_i2info(i2)->lli_lsm_md;
2525         } else {
2526                 fid_zero(&op_data->op_fid2);
2527         }
2528
2529         if (ll_i2sbi(i1)->ll_flags & LL_SBI_64BIT_HASH)
2530                 op_data->op_cli_flags |= CLI_HASH64;
2531
2532         if (ll_need_32bit_api(ll_i2sbi(i1)))
2533                 op_data->op_cli_flags |= CLI_API32;
2534
2535         op_data->op_name = name;
2536         op_data->op_namelen = namelen;
2537         op_data->op_mode = mode;
2538         op_data->op_mod_time = ktime_get_real_seconds();
2539         op_data->op_fsuid = from_kuid(&init_user_ns, current_fsuid());
2540         op_data->op_fsgid = from_kgid(&init_user_ns, current_fsgid());
2541         op_data->op_cap = cfs_curproc_cap_pack();
2542         op_data->op_mds = 0;
2543         if ((opc == LUSTRE_OPC_CREATE) && (name != NULL) &&
2544              filename_is_volatile(name, namelen, &op_data->op_mds)) {
2545                 op_data->op_bias |= MDS_CREATE_VOLATILE;
2546         }
2547         op_data->op_data = data;
2548
2549         return op_data;
2550 }
2551
2552 void ll_finish_md_op_data(struct md_op_data *op_data)
2553 {
2554         security_release_secctx(op_data->op_file_secctx,
2555                                 op_data->op_file_secctx_size);
2556         OBD_FREE_PTR(op_data);
2557 }
2558
2559 #ifdef HAVE_SUPEROPS_USE_DENTRY
2560 int ll_show_options(struct seq_file *seq, struct dentry *dentry)
2561 #else
2562 int ll_show_options(struct seq_file *seq, struct vfsmount *vfs)
2563 #endif
2564 {
2565         struct ll_sb_info *sbi;
2566
2567 #ifdef HAVE_SUPEROPS_USE_DENTRY
2568         LASSERT((seq != NULL) && (dentry != NULL));
2569         sbi = ll_s2sbi(dentry->d_sb);
2570 #else
2571         LASSERT((seq != NULL) && (vfs != NULL));
2572         sbi = ll_s2sbi(vfs->mnt_sb);
2573 #endif
2574
2575         if (sbi->ll_flags & LL_SBI_NOLCK)
2576                 seq_puts(seq, ",nolock");
2577
2578         if (sbi->ll_flags & LL_SBI_FLOCK)
2579                 seq_puts(seq, ",flock");
2580
2581         if (sbi->ll_flags & LL_SBI_LOCALFLOCK)
2582                 seq_puts(seq, ",localflock");
2583
2584         if (sbi->ll_flags & LL_SBI_USER_XATTR)
2585                 seq_puts(seq, ",user_xattr");
2586
2587         if (sbi->ll_flags & LL_SBI_LAZYSTATFS)
2588                 seq_puts(seq, ",lazystatfs");
2589
2590         if (sbi->ll_flags & LL_SBI_USER_FID2PATH)
2591                 seq_puts(seq, ",user_fid2path");
2592
2593         if (sbi->ll_flags & LL_SBI_ALWAYS_PING)
2594                 seq_puts(seq, ",always_ping");
2595
2596         RETURN(0);
2597 }
2598
2599 /**
2600  * Get obd name by cmd, and copy out to user space
2601  */
2602 int ll_get_obd_name(struct inode *inode, unsigned int cmd, unsigned long arg)
2603 {
2604         struct ll_sb_info *sbi = ll_i2sbi(inode);
2605         struct obd_device *obd;
2606         ENTRY;
2607
2608         if (cmd == OBD_IOC_GETDTNAME)
2609                 obd = class_exp2obd(sbi->ll_dt_exp);
2610         else if (cmd == OBD_IOC_GETMDNAME)
2611                 obd = class_exp2obd(sbi->ll_md_exp);
2612         else
2613                 RETURN(-EINVAL);
2614
2615         if (!obd)
2616                 RETURN(-ENOENT);
2617
2618         if (copy_to_user((void __user *)arg, obd->obd_name,
2619                          strlen(obd->obd_name) + 1))
2620                 RETURN(-EFAULT);
2621
2622         RETURN(0);
2623 }
2624
2625 /**
2626  * Get lustre file system name by \a sbi. If \a buf is provided(non-NULL), the
2627  * fsname will be returned in this buffer; otherwise, a static buffer will be
2628  * used to store the fsname and returned to caller.
2629  */
2630 char *ll_get_fsname(struct super_block *sb, char *buf, int buflen)
2631 {
2632         static char fsname_static[MTI_NAME_MAXLEN];
2633         struct lustre_sb_info *lsi = s2lsi(sb);
2634         char *ptr;
2635         int len;
2636
2637         if (buf == NULL) {
2638                 /* this means the caller wants to use static buffer
2639                  * and it doesn't care about race. Usually this is
2640                  * in error reporting path */
2641                 buf = fsname_static;
2642                 buflen = sizeof(fsname_static);
2643         }
2644
2645         len = strlen(lsi->lsi_lmd->lmd_profile);
2646         ptr = strrchr(lsi->lsi_lmd->lmd_profile, '-');
2647         if (ptr && (strcmp(ptr, "-client") == 0))
2648                 len -= 7;
2649
2650         if (unlikely(len >= buflen))
2651                 len = buflen - 1;
2652         strncpy(buf, lsi->lsi_lmd->lmd_profile, len);
2653         buf[len] = '\0';
2654
2655         return buf;
2656 }
2657
2658 static char* ll_d_path(struct dentry *dentry, char *buf, int bufsize)
2659 {
2660         char *path = NULL;
2661
2662         struct path p;
2663
2664         p.dentry = dentry;
2665         p.mnt = current->fs->root.mnt;
2666         path_get(&p);
2667         path = d_path(&p, buf, bufsize);
2668         path_put(&p);
2669         return path;
2670 }
2671
2672 void ll_dirty_page_discard_warn(struct page *page, int ioret)
2673 {
2674         char *buf, *path = NULL;
2675         struct dentry *dentry = NULL;
2676         struct inode *inode = page->mapping->host;
2677
2678         /* this can be called inside spin lock so use GFP_ATOMIC. */
2679         buf = (char *)__get_free_page(GFP_ATOMIC);
2680         if (buf != NULL) {
2681                 dentry = d_find_alias(page->mapping->host);
2682                 if (dentry != NULL)
2683                         path = ll_d_path(dentry, buf, PAGE_SIZE);
2684         }
2685
2686         CDEBUG(D_WARNING,
2687                "%s: dirty page discard: %s/fid: "DFID"/%s may get corrupted "
2688                "(rc %d)\n", ll_get_fsname(page->mapping->host->i_sb, NULL, 0),
2689                s2lsi(page->mapping->host->i_sb)->lsi_lmd->lmd_dev,
2690                PFID(ll_inode2fid(inode)),
2691                (path && !IS_ERR(path)) ? path : "", ioret);
2692
2693         if (dentry != NULL)
2694                 dput(dentry);
2695
2696         if (buf != NULL)
2697                 free_page((unsigned long)buf);
2698 }
2699
2700 ssize_t ll_copy_user_md(const struct lov_user_md __user *md,
2701                         struct lov_user_md **kbuf)
2702 {
2703         struct lov_user_md      lum;
2704         ssize_t                 lum_size;
2705         ENTRY;
2706
2707         if (copy_from_user(&lum, md, sizeof(lum)))
2708                 RETURN(-EFAULT);
2709
2710         lum_size = ll_lov_user_md_size(&lum);
2711         if (lum_size < 0)
2712                 RETURN(lum_size);
2713
2714         OBD_ALLOC(*kbuf, lum_size);
2715         if (*kbuf == NULL)
2716                 RETURN(-ENOMEM);
2717
2718         if (copy_from_user(*kbuf, md, lum_size) != 0) {
2719                 OBD_FREE(*kbuf, lum_size);
2720                 RETURN(-EFAULT);
2721         }
2722
2723         RETURN(lum_size);
2724 }
2725
2726 /*
2727  * Compute llite root squash state after a change of root squash
2728  * configuration setting or add/remove of a lnet nid
2729  */
2730 void ll_compute_rootsquash_state(struct ll_sb_info *sbi)
2731 {
2732         struct root_squash_info *squash = &sbi->ll_squash;
2733         int i;
2734         bool matched;
2735         struct lnet_process_id id;
2736
2737         /* Update norootsquash flag */
2738         down_write(&squash->rsi_sem);
2739         if (list_empty(&squash->rsi_nosquash_nids))
2740                 sbi->ll_flags &= ~LL_SBI_NOROOTSQUASH;
2741         else {
2742                 /* Do not apply root squash as soon as one of our NIDs is
2743                  * in the nosquash_nids list */
2744                 matched = false;
2745                 i = 0;
2746                 while (LNetGetId(i++, &id) != -ENOENT) {
2747                         if (LNET_NETTYP(LNET_NIDNET(id.nid)) == LOLND)
2748                                 continue;
2749                         if (cfs_match_nid(id.nid, &squash->rsi_nosquash_nids)) {
2750                                 matched = true;
2751                                 break;
2752                         }
2753                 }
2754                 if (matched)
2755                         sbi->ll_flags |= LL_SBI_NOROOTSQUASH;
2756                 else
2757                         sbi->ll_flags &= ~LL_SBI_NOROOTSQUASH;
2758         }
2759         up_write(&squash->rsi_sem);
2760 }
2761
2762 /**
2763  * Parse linkea content to extract information about a given hardlink
2764  *
2765  * \param[in]   ldata      - Initialized linkea data
2766  * \param[in]   linkno     - Link identifier
2767  * \param[out]  parent_fid - The entry's parent FID
2768  * \param[out]  ln         - Entry name destination buffer
2769  *
2770  * \retval 0 on success
2771  * \retval Appropriate negative error code on failure
2772  */
2773 static int ll_linkea_decode(struct linkea_data *ldata, unsigned int linkno,
2774                             struct lu_fid *parent_fid, struct lu_name *ln)
2775 {
2776         unsigned int    idx;
2777         int             rc;
2778         ENTRY;
2779
2780         rc = linkea_init_with_rec(ldata);
2781         if (rc < 0)
2782                 RETURN(rc);
2783
2784         if (linkno >= ldata->ld_leh->leh_reccount)
2785                 /* beyond last link */
2786                 RETURN(-ENODATA);
2787
2788         linkea_first_entry(ldata);
2789         for (idx = 0; ldata->ld_lee != NULL; idx++) {
2790                 linkea_entry_unpack(ldata->ld_lee, &ldata->ld_reclen, ln,
2791                                     parent_fid);
2792                 if (idx == linkno)
2793                         break;
2794
2795                 linkea_next_entry(ldata);
2796         }
2797
2798         if (idx < linkno)
2799                 RETURN(-ENODATA);
2800
2801         RETURN(0);
2802 }
2803
2804 /**
2805  * Get parent FID and name of an identified link. Operation is performed for
2806  * a given link number, letting the caller iterate over linkno to list one or
2807  * all links of an entry.
2808  *
2809  * \param[in]     file - File descriptor against which to perform the operation
2810  * \param[in,out] arg  - User-filled structure containing the linkno to operate
2811  *                       on and the available size. It is eventually filled with
2812  *                       the requested information or left untouched on error
2813  *
2814  * \retval - 0 on success
2815  * \retval - Appropriate negative error code on failure
2816  */
2817 int ll_getparent(struct file *file, struct getparent __user *arg)
2818 {
2819         struct inode            *inode = file_inode(file);
2820         struct linkea_data      *ldata;
2821         struct lu_buf            buf = LU_BUF_NULL;
2822         struct lu_name           ln;
2823         struct lu_fid            parent_fid;
2824         __u32                    linkno;
2825         __u32                    name_size;
2826         int                      rc;
2827
2828         ENTRY;
2829
2830         if (!cfs_capable(CFS_CAP_DAC_READ_SEARCH) &&
2831             !(ll_i2sbi(inode)->ll_flags & LL_SBI_USER_FID2PATH))
2832                 RETURN(-EPERM);
2833
2834         if (get_user(name_size, &arg->gp_name_size))
2835                 RETURN(-EFAULT);
2836
2837         if (get_user(linkno, &arg->gp_linkno))
2838                 RETURN(-EFAULT);
2839
2840         if (name_size > PATH_MAX)
2841                 RETURN(-EINVAL);
2842
2843         OBD_ALLOC(ldata, sizeof(*ldata));
2844         if (ldata == NULL)
2845                 RETURN(-ENOMEM);
2846
2847         rc = linkea_data_new(ldata, &buf);
2848         if (rc < 0)
2849                 GOTO(ldata_free, rc);
2850
2851 #ifdef HAVE_XATTR_HANDLER_FLAGS
2852         rc = ll_xattr_list(inode, XATTR_NAME_LINK, XATTR_TRUSTED_T, buf.lb_buf,
2853                            buf.lb_len, OBD_MD_FLXATTR);
2854 #else
2855         rc = ll_getxattr(file_dentry(file), XATTR_NAME_LINK, buf.lb_buf,
2856                          buf.lb_len);
2857 #endif /* HAVE_XATTR_HANDLER_FLAGS */
2858         if (rc < 0)
2859                 GOTO(lb_free, rc);
2860
2861         rc = ll_linkea_decode(ldata, linkno, &parent_fid, &ln);
2862         if (rc < 0)
2863                 GOTO(lb_free, rc);
2864
2865         if (ln.ln_namelen >= name_size)
2866                 GOTO(lb_free, rc = -EOVERFLOW);
2867
2868         if (copy_to_user(&arg->gp_fid, &parent_fid, sizeof(arg->gp_fid)))
2869                 GOTO(lb_free, rc = -EFAULT);
2870
2871         if (copy_to_user(&arg->gp_name, ln.ln_name, ln.ln_namelen))
2872                 GOTO(lb_free, rc = -EFAULT);
2873
2874         if (put_user('\0', arg->gp_name + ln.ln_namelen))
2875                 GOTO(lb_free, rc = -EFAULT);
2876
2877 lb_free:
2878         lu_buf_free(&buf);
2879 ldata_free:
2880         OBD_FREE(ldata, sizeof(*ldata));
2881
2882         RETURN(rc);
2883 }