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_running_max = LL_SA_RUNNING_DEF;
 127         sbi->ll_sa_max = LL_SA_RPC_DEF;
 128         atomic_set(&sbi->ll_sa_total, 0);
 129         atomic_set(&sbi->ll_sa_wrong, 0);
 130         atomic_set(&sbi->ll_sa_running, 0);
 131         atomic_set(&sbi->ll_agl_total, 0);
 132         sbi->ll_flags |= LL_SBI_AGL_ENABLED;
 133         sbi->ll_flags |= LL_SBI_FAST_READ;
 134         sbi->ll_flags |= LL_SBI_TINY_WRITE;
 135
 136         /* root squash */
 137         sbi->ll_squash.rsi_uid = 0;
 138         sbi->ll_squash.rsi_gid = 0;
 139         INIT_LIST_HEAD(&sbi->ll_squash.rsi_nosquash_nids);
 140         init_rwsem(&sbi->ll_squash.rsi_sem);
 141
 142         RETURN(sbi);
 143 }
 144
 145 static void ll_free_sbi(struct super_block *sb)
 146 {
 147         struct ll_sb_info *sbi = ll_s2sbi(sb);
 148         ENTRY;
 149
 150         if (sbi != NULL) {
 151                 if (!list_empty(&sbi->ll_squash.rsi_nosquash_nids))
 152                         cfs_free_nidlist(&sbi->ll_squash.rsi_nosquash_nids);
 153                 if (sbi->ll_cache != NULL) {
 154                         cl_cache_decref(sbi->ll_cache);
 155                         sbi->ll_cache = NULL;
 156                 }
 157                 OBD_FREE(sbi, sizeof(*sbi));
 158         }
 159         EXIT;
 160 }
 161
 162 static inline int obd_connect_has_secctx(struct obd_connect_data *data)
 163 {
 164         return data->ocd_connect_flags & OBD_CONNECT_FLAGS2 &&
 165                data->ocd_connect_flags2 & OBD_CONNECT2_FILE_SECCTX;
 166 }
 167
 168 static int client_common_fill_super(struct super_block *sb, char *md, char *dt,
 169                                     struct vfsmount *mnt)
 170 {
 171         struct inode *root = NULL;
 172         struct ll_sb_info *sbi = ll_s2sbi(sb);
 173         struct obd_device *obd;
 174         struct obd_statfs *osfs = NULL;
 175         struct ptlrpc_request *request = NULL;
 176         struct obd_connect_data *data = NULL;
 177         struct obd_uuid *uuid;
 178         struct md_op_data *op_data;
 179         struct lustre_md lmd;
 180         u64 valid;
 181         int size, err, checksum;
 182         ENTRY;
 183
 184         obd = class_name2obd(md);
 185         if (!obd) {
 186                 CERROR("MD %s: not setup or attached\n", md);
 187                 RETURN(-EINVAL);
 188         }
 189
 190         OBD_ALLOC_PTR(data);
 191         if (data == NULL)
 192                 RETURN(-ENOMEM);
 193
 194         OBD_ALLOC_PTR(osfs);
 195         if (osfs == NULL) {
 196                 OBD_FREE_PTR(data);
 197                 RETURN(-ENOMEM);
 198         }
 199
 200         /* pass client page size via ocd_grant_blkbits, the server should report
 201          * back its backend blocksize for grant calculation purpose */
 202         data->ocd_grant_blkbits = PAGE_SHIFT;
 203
 204         /* indicate MDT features supported by this client */
 205         data->ocd_connect_flags = OBD_CONNECT_IBITS    | OBD_CONNECT_NODEVOH  |
 206                                   OBD_CONNECT_ATTRFID  | OBD_CONNECT_GRANT |
 207                                   OBD_CONNECT_VERSION  | OBD_CONNECT_BRW_SIZE |
 208                                   OBD_CONNECT_SRVLOCK  | OBD_CONNECT_TRUNCLOCK|
 209                                   OBD_CONNECT_MDS_CAPA | OBD_CONNECT_OSS_CAPA |
 210                                   OBD_CONNECT_CANCELSET | OBD_CONNECT_FID     |
 211                                   OBD_CONNECT_AT       | OBD_CONNECT_LOV_V3   |
 212                                   OBD_CONNECT_VBR | OBD_CONNECT_FULL20 |
 213                                   OBD_CONNECT_64BITHASH |
 214                                   OBD_CONNECT_EINPROGRESS |
 215                                   OBD_CONNECT_JOBSTATS | OBD_CONNECT_LVB_TYPE |
 216                                   OBD_CONNECT_LAYOUTLOCK | OBD_CONNECT_PINGLESS|
 217                                   OBD_CONNECT_MAX_EASIZE |
 218                                   OBD_CONNECT_FLOCK_DEAD |
 219                                   OBD_CONNECT_DISP_STRIPE | OBD_CONNECT_LFSCK |
 220                                   OBD_CONNECT_OPEN_BY_FID |
 221                                   OBD_CONNECT_DIR_STRIPE |
 222                                   OBD_CONNECT_BULK_MBITS | OBD_CONNECT_CKSUM |
 223                                   OBD_CONNECT_SUBTREE |
 224                                   OBD_CONNECT_MULTIMODRPCS |
 225                                   OBD_CONNECT_GRANT_PARAM |
 226                                   OBD_CONNECT_SHORTIO | OBD_CONNECT_FLAGS2;
 227
 228         data->ocd_connect_flags2 = OBD_CONNECT2_FLR |
 229                                    OBD_CONNECT2_LOCK_CONVERT |
 230                                    OBD_CONNECT2_DIR_MIGRATE |
 231                                    OBD_CONNECT2_SUM_STATFS;
 232
 233 #ifdef HAVE_LRU_RESIZE_SUPPORT
 234         if (sbi->ll_flags & LL_SBI_LRU_RESIZE)
 235                 data->ocd_connect_flags |= OBD_CONNECT_LRU_RESIZE;
 236 #endif
 237 #ifdef CONFIG_FS_POSIX_ACL
 238         data->ocd_connect_flags |= OBD_CONNECT_ACL | OBD_CONNECT_UMASK |
 239                                    OBD_CONNECT_LARGE_ACL;
 240 #endif
 241
 242         data->ocd_cksum_types = obd_cksum_types_supported_client();
 243
 244         if (OBD_FAIL_CHECK(OBD_FAIL_MDC_LIGHTWEIGHT))
 245                 /* flag mdc connection as lightweight, only used for test
 246                  * purpose, use with care */
 247                 data->ocd_connect_flags |= OBD_CONNECT_LIGHTWEIGHT;
 248
 249         data->ocd_ibits_known = MDS_INODELOCK_FULL;
 250         data->ocd_version = LUSTRE_VERSION_CODE;
 251
 252         if (sb->s_flags & MS_RDONLY)
 253                 data->ocd_connect_flags |= OBD_CONNECT_RDONLY;
 254         if (sbi->ll_flags & LL_SBI_USER_XATTR)
 255                 data->ocd_connect_flags |= OBD_CONNECT_XATTR;
 256
 257 #ifdef MS_NOSEC
 258         /* Setting this indicates we correctly support S_NOSEC (See kernel
 259          * commit 9e1f1de02c2275d7172e18dc4e7c2065777611bf)
 260          */
 261         sb->s_flags |= MS_NOSEC;
 262 #endif
 263
 264         if (sbi->ll_flags & LL_SBI_FLOCK)
 265                 sbi->ll_fop = &ll_file_operations_flock;
 266         else if (sbi->ll_flags & LL_SBI_LOCALFLOCK)
 267                 sbi->ll_fop = &ll_file_operations;
 268         else
 269                 sbi->ll_fop = &ll_file_operations_noflock;
 270
 271         /* always ping even if server suppress_pings */
 272         if (sbi->ll_flags & LL_SBI_ALWAYS_PING)
 273                 data->ocd_connect_flags &= ~OBD_CONNECT_PINGLESS;
 274
 275 #if defined(HAVE_SECURITY_DENTRY_INIT_SECURITY) && defined(CONFIG_SECURITY)
 276         data->ocd_connect_flags2 |= OBD_CONNECT2_FILE_SECCTX;
 277 #endif /* HAVE_SECURITY_DENTRY_INIT_SECURITY */
 278
 279         data->ocd_brw_size = MD_MAX_BRW_SIZE;
 280
 281         err = obd_connect(NULL, &sbi->ll_md_exp, obd, &sbi->ll_sb_uuid, data, NULL);
 282         if (err == -EBUSY) {
 283                 LCONSOLE_ERROR_MSG(0x14f, "An MDT (md %s) is performing "
 284                                    "recovery, of which this client is not a "
 285                                    "part. Please wait for recovery to complete,"
 286                                    " abort, or time out.\n", md);
 287                 GOTO(out, err);
 288         } else if (err) {
 289                 CERROR("cannot connect to %s: rc = %d\n", md, err);
 290                 GOTO(out, err);
 291         }
 292
 293         sbi->ll_md_exp->exp_connect_data = *data;
 294
 295         err = obd_fid_init(sbi->ll_md_exp->exp_obd, sbi->ll_md_exp,
 296                            LUSTRE_SEQ_METADATA);
 297         if (err) {
 298                 CERROR("%s: Can't init metadata layer FID infrastructure, "
 299                        "rc = %d\n", sbi->ll_md_exp->exp_obd->obd_name, err);
 300                 GOTO(out_md, err);
 301         }
 302
 303         /* For mount, we only need fs info from MDT0, and also in DNE, it
 304          * can make sure the client can be mounted as long as MDT0 is
 305          * avaible */
 306         err = obd_statfs(NULL, sbi->ll_md_exp, osfs,
 307                         ktime_get_seconds() -OBD_STATFS_CACHE_SECONDS,
 308                         OBD_STATFS_FOR_MDT0);
 309         if (err)
 310                 GOTO(out_md_fid, err);
 311
 312         /* This needs to be after statfs to ensure connect has finished.
 313          * Note that "data" does NOT contain the valid connect reply.
 314          * If connecting to a 1.8 server there will be no LMV device, so
 315          * we can access the MDC export directly and exp_connect_flags will
 316          * be non-zero, but if accessing an upgraded 2.1 server it will
 317          * have the correct flags filled in.
 318          * XXX: fill in the LMV exp_connect_flags from MDC(s). */
 319         valid = exp_connect_flags(sbi->ll_md_exp) & CLIENT_CONNECT_MDT_REQD;
 320         if (exp_connect_flags(sbi->ll_md_exp) != 0 &&
 321             valid != CLIENT_CONNECT_MDT_REQD) {
 322                 char *buf;
 323
 324                 OBD_ALLOC_WAIT(buf, PAGE_SIZE);
 325                 obd_connect_flags2str(buf, PAGE_SIZE,
 326                                       valid ^ CLIENT_CONNECT_MDT_REQD, 0, ",");
 327                 LCONSOLE_ERROR_MSG(0x170, "Server %s does not support "
 328                                    "feature(s) needed for correct operation "
 329                                    "of this client (%s). Please upgrade "
 330                                    "server or downgrade client.\n",
 331                                    sbi->ll_md_exp->exp_obd->obd_name, buf);
 332                 OBD_FREE(buf, PAGE_SIZE);
 333                 GOTO(out_md_fid, err = -EPROTO);
 334         }
 335
 336         size = sizeof(*data);
 337         err = obd_get_info(NULL, sbi->ll_md_exp, sizeof(KEY_CONN_DATA),
 338                            KEY_CONN_DATA,  &size, data);
 339         if (err) {
 340                 CERROR("%s: Get connect data failed: rc = %d\n",
 341                        sbi->ll_md_exp->exp_obd->obd_name, err);
 342                 GOTO(out_md_fid, err);
 343         }
 344
 345         LASSERT(osfs->os_bsize);
 346         sb->s_blocksize = osfs->os_bsize;
 347         sb->s_blocksize_bits = log2(osfs->os_bsize);
 348         sb->s_magic = LL_SUPER_MAGIC;
 349         sb->s_maxbytes = MAX_LFS_FILESIZE;
 350         sbi->ll_namelen = osfs->os_namelen;
 351         sbi->ll_mnt.mnt = current->fs->root.mnt;
 352
 353         if ((sbi->ll_flags & LL_SBI_USER_XATTR) &&
 354             !(data->ocd_connect_flags & OBD_CONNECT_XATTR)) {
 355                 LCONSOLE_INFO("Disabling user_xattr feature because "
 356                               "it is not supported on the server\n");
 357                 sbi->ll_flags &= ~LL_SBI_USER_XATTR;
 358         }
 359
 360         if (data->ocd_connect_flags & OBD_CONNECT_ACL) {
 361 #ifdef MS_POSIXACL
 362                 sb->s_flags |= MS_POSIXACL;
 363 #endif
 364                 sbi->ll_flags |= LL_SBI_ACL;
 365         } else {
 366                 LCONSOLE_INFO("client wants to enable acl, but mdt not!\n");
 367 #ifdef MS_POSIXACL
 368                 sb->s_flags &= ~MS_POSIXACL;
 369 #endif
 370                 sbi->ll_flags &= ~LL_SBI_ACL;
 371         }
 372
 373         if (data->ocd_connect_flags & OBD_CONNECT_64BITHASH)
 374                 sbi->ll_flags |= LL_SBI_64BIT_HASH;
 375
 376         if (data->ocd_connect_flags & OBD_CONNECT_LAYOUTLOCK)
 377                 sbi->ll_flags |= LL_SBI_LAYOUT_LOCK;
 378
 379         if (obd_connect_has_secctx(data))
 380                 sbi->ll_flags |= LL_SBI_FILE_SECCTX;
 381
 382         if (data->ocd_ibits_known & MDS_INODELOCK_XATTR) {
 383                 if (!(data->ocd_connect_flags & OBD_CONNECT_MAX_EASIZE)) {
 384                         LCONSOLE_INFO("%s: disabling xattr cache due to "
 385                                       "unknown maximum xattr size.\n", dt);
 386                 } else if (!sbi->ll_xattr_cache_set) {
 387                         /* If xattr_cache is already set (no matter 0 or 1)
 388                          * during processing llog, it won't be enabled here. */
 389                         sbi->ll_flags |= LL_SBI_XATTR_CACHE;
 390                         sbi->ll_xattr_cache_enabled = 1;
 391                 }
 392         }
 393
 394         obd = class_name2obd(dt);
 395         if (!obd) {
 396                 CERROR("DT %s: not setup or attached\n", dt);
 397                 GOTO(out_md_fid, err = -ENODEV);
 398         }
 399
 400         /* pass client page size via ocd_grant_blkbits, the server should report
 401          * back its backend blocksize for grant calculation purpose */
 402         data->ocd_grant_blkbits = PAGE_SHIFT;
 403
 404         /* indicate OST features supported by this client */
 405         data->ocd_connect_flags = OBD_CONNECT_GRANT | OBD_CONNECT_VERSION |
 406                                   OBD_CONNECT_REQPORTAL | OBD_CONNECT_BRW_SIZE |
 407                                   OBD_CONNECT_CANCELSET | OBD_CONNECT_FID |
 408                                   OBD_CONNECT_SRVLOCK | OBD_CONNECT_TRUNCLOCK|
 409                                   OBD_CONNECT_AT | OBD_CONNECT_OSS_CAPA |
 410                                   OBD_CONNECT_VBR | OBD_CONNECT_FULL20 |
 411                                   OBD_CONNECT_64BITHASH | OBD_CONNECT_MAXBYTES |
 412                                   OBD_CONNECT_EINPROGRESS |
 413                                   OBD_CONNECT_JOBSTATS | OBD_CONNECT_LVB_TYPE |
 414                                   OBD_CONNECT_LAYOUTLOCK |
 415                                   OBD_CONNECT_PINGLESS | OBD_CONNECT_LFSCK |
 416                                   OBD_CONNECT_BULK_MBITS | OBD_CONNECT_SHORTIO |
 417                                   OBD_CONNECT_FLAGS2 | OBD_CONNECT_GRANT_SHRINK;
 418
 419 /* The client currently advertises support for OBD_CONNECT_LOCKAHEAD_OLD so it
 420  * can interoperate with an older version of lockahead which was released prior
 421  * to landing in master. This support will be dropped when 2.13 development
 422  * starts.  At the point, we should not just drop the connect flag (below), we
 423  * should also remove the support in the code.
 424  *
 425  * Removing it means a few things:
 426  * 1. Remove this section here
 427  * 2. Remove CEF_NONBLOCK in ll_file_lockahead()
 428  * 3. Remove function exp_connect_lockahead_old
 429  * 4. Remove LDLM_FL_LOCKAHEAD_OLD_RESERVED in lustre_dlm_flags.h
 430  * */
 431 #if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(2, 12, 50, 0)
 432         data->ocd_connect_flags |= OBD_CONNECT_LOCKAHEAD_OLD;
 433 #endif
 434
 435         data->ocd_connect_flags2 = OBD_CONNECT2_LOCKAHEAD;
 436
 437         if (!OBD_FAIL_CHECK(OBD_FAIL_OSC_CONNECT_GRANT_PARAM))
 438                 data->ocd_connect_flags |= OBD_CONNECT_GRANT_PARAM;
 439
 440         /* OBD_CONNECT_CKSUM should always be set, even if checksums are
 441          * disabled by default, because it can still be enabled on the
 442          * fly via /proc. As a consequence, we still need to come to an
 443          * agreement on the supported algorithms at connect time */
 444         data->ocd_connect_flags |= OBD_CONNECT_CKSUM;
 445
 446         if (OBD_FAIL_CHECK(OBD_FAIL_OSC_CKSUM_ADLER_ONLY))
 447                 data->ocd_cksum_types = OBD_CKSUM_ADLER;
 448         else
 449                 data->ocd_cksum_types = obd_cksum_types_supported_client();
 450
 451 #ifdef HAVE_LRU_RESIZE_SUPPORT
 452         data->ocd_connect_flags |= OBD_CONNECT_LRU_RESIZE;
 453 #endif
 454         /* always ping even if server suppress_pings */
 455         if (sbi->ll_flags & LL_SBI_ALWAYS_PING)
 456                 data->ocd_connect_flags &= ~OBD_CONNECT_PINGLESS;
 457
 458         CDEBUG(D_RPCTRACE, "ocd_connect_flags: %#llx ocd_version: %d "
 459                "ocd_grant: %d\n", data->ocd_connect_flags,
 460                data->ocd_version, data->ocd_grant);
 461
 462         obd->obd_upcall.onu_owner = &sbi->ll_lco;
 463         obd->obd_upcall.onu_upcall = cl_ocd_update;
 464
 465         data->ocd_brw_size = DT_MAX_BRW_SIZE;
 466
 467         err = obd_connect(NULL, &sbi->ll_dt_exp, obd, &sbi->ll_sb_uuid, data,
 468                           NULL);
 469         if (err == -EBUSY) {
 470                 LCONSOLE_ERROR_MSG(0x150, "An OST (dt %s) is performing "
 471                                    "recovery, of which this client is not a "
 472                                    "part.  Please wait for recovery to "
 473                                    "complete, abort, or time out.\n", dt);
 474                 GOTO(out_md, err);
 475         } else if (err) {
 476                 CERROR("%s: Cannot connect to %s: rc = %d\n",
 477                        sbi->ll_dt_exp->exp_obd->obd_name, dt, err);
 478                 GOTO(out_md, err);
 479         }
 480
 481         sbi->ll_dt_exp->exp_connect_data = *data;
 482
 483         /* Don't change value if it was specified in the config log */
 484         if (sbi->ll_ra_info.ra_max_read_ahead_whole_pages == -1)
 485                 sbi->ll_ra_info.ra_max_read_ahead_whole_pages =
 486                         max_t(unsigned long, SBI_DEFAULT_READAHEAD_WHOLE_MAX,
 487                               (data->ocd_brw_size >> PAGE_SHIFT));
 488
 489         err = obd_fid_init(sbi->ll_dt_exp->exp_obd, sbi->ll_dt_exp,
 490                            LUSTRE_SEQ_METADATA);
 491         if (err) {
 492                 CERROR("%s: Can't init data layer FID infrastructure, "
 493                        "rc = %d\n", sbi->ll_dt_exp->exp_obd->obd_name, err);
 494                 GOTO(out_dt, err);
 495         }
 496
 497         mutex_lock(&sbi->ll_lco.lco_lock);
 498         sbi->ll_lco.lco_flags = data->ocd_connect_flags;
 499         sbi->ll_lco.lco_md_exp = sbi->ll_md_exp;
 500         sbi->ll_lco.lco_dt_exp = sbi->ll_dt_exp;
 501         mutex_unlock(&sbi->ll_lco.lco_lock);
 502
 503         fid_zero(&sbi->ll_root_fid);
 504         err = md_get_root(sbi->ll_md_exp, get_mount_fileset(sb),
 505                            &sbi->ll_root_fid);
 506         if (err) {
 507                 CERROR("cannot mds_connect: rc = %d\n", err);
 508                 GOTO(out_lock_cn_cb, err);
 509         }
 510         if (!fid_is_sane(&sbi->ll_root_fid)) {
 511                 CERROR("%s: Invalid root fid "DFID" during mount\n",
 512                        sbi->ll_md_exp->exp_obd->obd_name,
 513                        PFID(&sbi->ll_root_fid));
 514                 GOTO(out_lock_cn_cb, err = -EINVAL);
 515         }
 516         CDEBUG(D_SUPER, "rootfid "DFID"\n", PFID(&sbi->ll_root_fid));
 517
 518         sb->s_op = &lustre_super_operations;
 519 #ifdef HAVE_XATTR_HANDLER_FLAGS
 520         sb->s_xattr = ll_xattr_handlers;
 521 #endif
 522 #if THREAD_SIZE >= 8192 /*b=17630*/
 523         sb->s_export_op = &lustre_export_operations;
 524 #endif
 525
 526         /* make root inode
 527          * XXX: move this to after cbd setup? */
 528         valid = OBD_MD_FLGETATTR | OBD_MD_FLBLOCKS | OBD_MD_FLMODEASIZE;
 529         if (sbi->ll_flags & LL_SBI_ACL)
 530                 valid |= OBD_MD_FLACL;
 531
 532         OBD_ALLOC_PTR(op_data);
 533         if (op_data == NULL)
 534                 GOTO(out_lock_cn_cb, err = -ENOMEM);
 535
 536         op_data->op_fid1 = sbi->ll_root_fid;
 537         op_data->op_mode = 0;
 538         op_data->op_valid = valid;
 539
 540         err = md_getattr(sbi->ll_md_exp, op_data, &request);
 541
 542         OBD_FREE_PTR(op_data);
 543         if (err) {
 544                 CERROR("%s: md_getattr failed for root: rc = %d\n",
 545                        sbi->ll_md_exp->exp_obd->obd_name, err);
 546                 GOTO(out_lock_cn_cb, err);
 547         }
 548
 549         err = md_get_lustre_md(sbi->ll_md_exp, request, sbi->ll_dt_exp,
 550                                sbi->ll_md_exp, &lmd);
 551         if (err) {
 552                 CERROR("failed to understand root inode md: rc = %d\n", err);
 553                 ptlrpc_req_finished(request);
 554                 GOTO(out_lock_cn_cb, err);
 555         }
 556
 557         LASSERT(fid_is_sane(&sbi->ll_root_fid));
 558         root = ll_iget(sb, cl_fid_build_ino(&sbi->ll_root_fid,
 559                                             sbi->ll_flags & LL_SBI_32BIT_API),
 560                        &lmd);
 561         md_free_lustre_md(sbi->ll_md_exp, &lmd);
 562         ptlrpc_req_finished(request);
 563
 564         if (IS_ERR(root)) {
 565 #ifdef CONFIG_FS_POSIX_ACL
 566                 if (lmd.posix_acl) {
 567                         posix_acl_release(lmd.posix_acl);
 568                         lmd.posix_acl = NULL;
 569                 }
 570 #endif
 571                 err = IS_ERR(root) ? PTR_ERR(root) : -EBADF;
 572                 root = NULL;
 573                 CERROR("lustre_lite: bad iget4 for root\n");
 574                 GOTO(out_root, err);
 575         }
 576
 577         checksum = sbi->ll_flags & LL_SBI_CHECKSUM;
 578         if (sbi->ll_checksum_set) {
 579                 err = obd_set_info_async(NULL, sbi->ll_dt_exp,
 580                                          sizeof(KEY_CHECKSUM), KEY_CHECKSUM,
 581                                          sizeof(checksum), &checksum, NULL);
 582                 if (err) {
 583                         CERROR("%s: Set checksum failed: rc = %d\n",
 584                                sbi->ll_dt_exp->exp_obd->obd_name, err);
 585                         GOTO(out_root, err);
 586                 }
 587         }
 588         cl_sb_init(sb);
 589
 590         err = obd_set_info_async(NULL, sbi->ll_dt_exp, sizeof(KEY_CACHE_SET),
 591                                  KEY_CACHE_SET, sizeof(*sbi->ll_cache),
 592                                  sbi->ll_cache, NULL);
 593         if (err) {
 594                 CERROR("%s: Set cache_set failed: rc = %d\n",
 595                        sbi->ll_dt_exp->exp_obd->obd_name, err);
 596                 GOTO(out_root, err);
 597         }
 598
 599         sb->s_root = d_make_root(root);
 600         if (sb->s_root == NULL) {
 601                 CERROR("%s: can't make root dentry\n",
 602                         ll_get_fsname(sb, NULL, 0));
 603                 GOTO(out_root, err = -ENOMEM);
 604         }
 605 #ifdef HAVE_DCACHE_LOCK
 606         sb->s_root->d_op = &ll_d_ops;
 607 #endif
 608
 609         sbi->ll_sdev_orig = sb->s_dev;
 610
 611         /* We set sb->s_dev equal on all lustre clients in order to support
 612          * NFS export clustering.  NFSD requires that the FSID be the same
 613          * on all clients. */
 614         /* s_dev is also used in lt_compare() to compare two fs, but that is
 615          * only a node-local comparison. */
 616         uuid = obd_get_uuid(sbi->ll_md_exp);
 617         if (uuid != NULL)
 618                 sb->s_dev = get_uuid2int(uuid->uuid, strlen(uuid->uuid));
 619
 620         if (data != NULL)
 621                 OBD_FREE_PTR(data);
 622         if (osfs != NULL)
 623                 OBD_FREE_PTR(osfs);
 624
 625         if (sbi->ll_proc_root != NULL) {
 626                 err = lprocfs_ll_register_obd(sb, dt);
 627                 if (err < 0) {
 628                         CERROR("%s: could not register %s in llite: rc = %d\n",
 629                                dt, ll_get_fsname(sb, NULL, 0), err);
 630                         err = 0;
 631                 }
 632                 err = lprocfs_ll_register_obd(sb, md);
 633                 if (err < 0) {
 634                         CERROR("%s: could not register %s in llite: rc = %d\n",
 635                                md, ll_get_fsname(sb, NULL, 0), err);
 636                         err = 0;
 637                 }
 638         }
 639
 640         RETURN(err);
 641 out_root:
 642         if (root)
 643                 iput(root);
 644 out_lock_cn_cb:
 645         obd_fid_fini(sbi->ll_dt_exp->exp_obd);
 646 out_dt:
 647         obd_disconnect(sbi->ll_dt_exp);
 648         sbi->ll_dt_exp = NULL;
 649 out_md_fid:
 650         obd_fid_fini(sbi->ll_md_exp->exp_obd);
 651 out_md:
 652         obd_disconnect(sbi->ll_md_exp);
 653         sbi->ll_md_exp = NULL;
 654 out:
 655         if (data != NULL)
 656                 OBD_FREE_PTR(data);
 657         if (osfs != NULL)
 658                 OBD_FREE_PTR(osfs);
 659         return err;
 660 }
 661
 662 int ll_get_max_mdsize(struct ll_sb_info *sbi, int *lmmsize)
 663 {
 664         int size, rc;
 665
 666         size = sizeof(*lmmsize);
 667         rc = obd_get_info(NULL, sbi->ll_dt_exp, sizeof(KEY_MAX_EASIZE),
 668                           KEY_MAX_EASIZE, &size, lmmsize);
 669         if (rc != 0) {
 670                 CERROR("%s: cannot get max LOV EA size: rc = %d\n",
 671                        sbi->ll_dt_exp->exp_obd->obd_name, rc);
 672                 RETURN(rc);
 673         }
 674
 675         size = sizeof(int);
 676         rc = obd_get_info(NULL, sbi->ll_md_exp, sizeof(KEY_MAX_EASIZE),
 677                           KEY_MAX_EASIZE, &size, lmmsize);
 678         if (rc)
 679                 CERROR("Get max mdsize error rc %d\n", rc);
 680
 681         RETURN(rc);
 682 }
 683
 684 /**
 685  * Get the value of the default_easize parameter.
 686  *
 687  * \see client_obd::cl_default_mds_easize
 688  *
 689  * \param[in] sbi       superblock info for this filesystem
 690  * \param[out] lmmsize  pointer to storage location for value
 691  *
 692  * \retval 0            on success
 693  * \retval negative     negated errno on failure
 694  */
 695 int ll_get_default_mdsize(struct ll_sb_info *sbi, int *lmmsize)
 696 {
 697         int size, rc;
 698
 699         size = sizeof(int);
 700         rc = obd_get_info(NULL, sbi->ll_md_exp, sizeof(KEY_DEFAULT_EASIZE),
 701                          KEY_DEFAULT_EASIZE, &size, lmmsize);
 702         if (rc)
 703                 CERROR("Get default mdsize error rc %d\n", rc);
 704
 705         RETURN(rc);
 706 }
 707
 708 /**
 709  * Set the default_easize parameter to the given value.
 710  *
 711  * \see client_obd::cl_default_mds_easize
 712  *
 713  * \param[in] sbi       superblock info for this filesystem
 714  * \param[in] lmmsize   the size to set
 715  *
 716  * \retval 0            on success
 717  * \retval negative     negated errno on failure
 718  */
 719 int ll_set_default_mdsize(struct ll_sb_info *sbi, int lmmsize)
 720 {
 721         int rc;
 722
 723         if (lmmsize < sizeof(struct lov_mds_md) ||
 724             lmmsize > OBD_MAX_DEFAULT_EA_SIZE)
 725                 return -EINVAL;
 726
 727         rc = obd_set_info_async(NULL, sbi->ll_md_exp,
 728                                 sizeof(KEY_DEFAULT_EASIZE), KEY_DEFAULT_EASIZE,
 729                                 sizeof(int), &lmmsize, NULL);
 730
 731         RETURN(rc);
 732 }
 733
 734 static void client_common_put_super(struct super_block *sb)
 735 {
 736         struct ll_sb_info *sbi = ll_s2sbi(sb);
 737         ENTRY;
 738
 739         cl_sb_fini(sb);
 740
 741         obd_fid_fini(sbi->ll_dt_exp->exp_obd);
 742         obd_disconnect(sbi->ll_dt_exp);
 743         sbi->ll_dt_exp = NULL;
 744
 745         ll_debugfs_unregister_super(sb);
 746
 747         obd_fid_fini(sbi->ll_md_exp->exp_obd);
 748         obd_disconnect(sbi->ll_md_exp);
 749         sbi->ll_md_exp = NULL;
 750
 751         EXIT;
 752 }
 753
 754 void ll_kill_super(struct super_block *sb)
 755 {
 756         struct ll_sb_info *sbi;
 757         ENTRY;
 758
 759         /* not init sb ?*/
 760         if (!(sb->s_flags & MS_ACTIVE))
 761                 return;
 762
 763         sbi = ll_s2sbi(sb);
 764         /* we need restore s_dev from changed for clustred NFS before put_super
 765          * because new kernels have cached s_dev and change sb->s_dev in
 766          * put_super not affected real removing devices */
 767         if (sbi) {
 768                 sb->s_dev = sbi->ll_sdev_orig;
 769                 sbi->ll_umounting = 1;
 770
 771                 /* wait running statahead threads to quit */
 772                 while (atomic_read(&sbi->ll_sa_running) > 0) {
 773                         set_current_state(TASK_UNINTERRUPTIBLE);
 774                         schedule_timeout(msecs_to_jiffies(MSEC_PER_SEC >> 3));
 775                 }
 776         }
 777
 778         EXIT;
 779 }
 780
 781 static inline int ll_set_opt(const char *opt, char *data, int fl)
 782 {
 783         if (strncmp(opt, data, strlen(opt)) != 0)
 784                 return 0;
 785         else
 786                 return fl;
 787 }
 788
 789 /* non-client-specific mount options are parsed in lmd_parse */
 790 static int ll_options(char *options, struct ll_sb_info *sbi)
 791 {
 792         int tmp;
 793         char *s1 = options, *s2;
 794         int *flags = &sbi->ll_flags;
 795         ENTRY;
 796
 797         if (!options)
 798                 RETURN(0);
 799
 800         CDEBUG(D_CONFIG, "Parsing opts %s\n", options);
 801
 802         while (*s1) {
 803                 CDEBUG(D_SUPER, "next opt=%s\n", s1);
 804                 tmp = ll_set_opt("nolock", s1, LL_SBI_NOLCK);
 805                 if (tmp) {
 806                         *flags |= tmp;
 807                         goto next;
 808                 }
 809                 tmp = ll_set_opt("flock", s1, LL_SBI_FLOCK);
 810                 if (tmp) {
 811                         *flags |= tmp;
 812                         goto next;
 813                 }
 814                 tmp = ll_set_opt("localflock", s1, LL_SBI_LOCALFLOCK);
 815                 if (tmp) {
 816                         *flags |= tmp;
 817                         goto next;
 818                 }
 819                 tmp = ll_set_opt("noflock", s1, LL_SBI_FLOCK|LL_SBI_LOCALFLOCK);
 820                 if (tmp) {
 821                         *flags &= ~tmp;
 822                         goto next;
 823                 }
 824                 tmp = ll_set_opt("user_xattr", s1, LL_SBI_USER_XATTR);
 825                 if (tmp) {
 826                         *flags |= tmp;
 827                         goto next;
 828                 }
 829                 tmp = ll_set_opt("nouser_xattr", s1, LL_SBI_USER_XATTR);
 830                 if (tmp) {
 831                         *flags &= ~tmp;
 832                         goto next;
 833                 }
 834                 tmp = ll_set_opt("context", s1, 1);
 835                 if (tmp)
 836                         goto next;
 837                 tmp = ll_set_opt("fscontext", s1, 1);
 838                 if (tmp)
 839                         goto next;
 840                 tmp = ll_set_opt("defcontext", s1, 1);
 841                 if (tmp)
 842                         goto next;
 843                 tmp = ll_set_opt("rootcontext", s1, 1);
 844                 if (tmp)
 845                         goto next;
 846                 tmp = ll_set_opt("user_fid2path", s1, LL_SBI_USER_FID2PATH);
 847                 if (tmp) {
 848                         *flags |= tmp;
 849                         goto next;
 850                 }
 851                 tmp = ll_set_opt("nouser_fid2path", s1, LL_SBI_USER_FID2PATH);
 852                 if (tmp) {
 853                         *flags &= ~tmp;
 854                         goto next;
 855                 }
 856
 857                 tmp = ll_set_opt("checksum", s1, LL_SBI_CHECKSUM);
 858                 if (tmp) {
 859                         *flags |= tmp;
 860                         sbi->ll_checksum_set = 1;
 861                         goto next;
 862                 }
 863                 tmp = ll_set_opt("nochecksum", s1, LL_SBI_CHECKSUM);
 864                 if (tmp) {
 865                         *flags &= ~tmp;
 866                         sbi->ll_checksum_set = 1;
 867                         goto next;
 868                 }
 869                 tmp = ll_set_opt("lruresize", s1, LL_SBI_LRU_RESIZE);
 870                 if (tmp) {
 871                         *flags |= tmp;
 872                         goto next;
 873                 }
 874                 tmp = ll_set_opt("nolruresize", s1, LL_SBI_LRU_RESIZE);
 875                 if (tmp) {
 876                         *flags &= ~tmp;
 877                         goto next;
 878                 }
 879                 tmp = ll_set_opt("lazystatfs", s1, LL_SBI_LAZYSTATFS);
 880                 if (tmp) {
 881                         *flags |= tmp;
 882                         goto next;
 883                 }
 884                 tmp = ll_set_opt("nolazystatfs", s1, LL_SBI_LAZYSTATFS);
 885                 if (tmp) {
 886                         *flags &= ~tmp;
 887                         goto next;
 888                 }
 889                 tmp = ll_set_opt("32bitapi", s1, LL_SBI_32BIT_API);
 890                 if (tmp) {
 891                         *flags |= tmp;
 892                         goto next;
 893                 }
 894                 tmp = ll_set_opt("verbose", s1, LL_SBI_VERBOSE);
 895                 if (tmp) {
 896                         *flags |= tmp;
 897                         goto next;
 898                 }
 899                 tmp = ll_set_opt("noverbose", s1, LL_SBI_VERBOSE);
 900                 if (tmp) {
 901                         *flags &= ~tmp;
 902                         goto next;
 903                 }
 904                 tmp = ll_set_opt("always_ping", s1, LL_SBI_ALWAYS_PING);
 905                 if (tmp) {
 906                         *flags |= tmp;
 907                         goto next;
 908                 }
 909                 LCONSOLE_ERROR_MSG(0x152, "Unknown option '%s', won't mount.\n",
 910                                    s1);
 911                 RETURN(-EINVAL);
 912
 913 next:
 914                 /* Find next opt */
 915                 s2 = strchr(s1, ',');
 916                 if (s2 == NULL)
 917                         break;
 918                 s1 = s2 + 1;
 919         }
 920         RETURN(0);
 921 }
 922
 923 void ll_lli_init(struct ll_inode_info *lli)
 924 {
 925         lli->lli_inode_magic = LLI_INODE_MAGIC;
 926         lli->lli_flags = 0;
 927         spin_lock_init(&lli->lli_lock);
 928         lli->lli_posix_acl = NULL;
 929         /* Do not set lli_fid, it has been initialized already. */
 930         fid_zero(&lli->lli_pfid);
 931         lli->lli_mds_read_och = NULL;
 932         lli->lli_mds_write_och = NULL;
 933         lli->lli_mds_exec_och = NULL;
 934         lli->lli_open_fd_read_count = 0;
 935         lli->lli_open_fd_write_count = 0;
 936         lli->lli_open_fd_exec_count = 0;
 937         mutex_init(&lli->lli_och_mutex);
 938         spin_lock_init(&lli->lli_agl_lock);
 939         spin_lock_init(&lli->lli_layout_lock);
 940         ll_layout_version_set(lli, CL_LAYOUT_GEN_NONE);
 941         lli->lli_clob = NULL;
 942
 943         init_rwsem(&lli->lli_xattrs_list_rwsem);
 944         mutex_init(&lli->lli_xattrs_enq_lock);
 945
 946         LASSERT(lli->lli_vfs_inode.i_mode != 0);
 947         if (S_ISDIR(lli->lli_vfs_inode.i_mode)) {
 948                 mutex_init(&lli->lli_readdir_mutex);
 949                 lli->lli_opendir_key = NULL;
 950                 lli->lli_sai = NULL;
 951                 spin_lock_init(&lli->lli_sa_lock);
 952                 lli->lli_opendir_pid = 0;
 953                 lli->lli_sa_enabled = 0;
 954                 lli->lli_def_stripe_offset = -1;
 955         } else {
 956                 mutex_init(&lli->lli_size_mutex);
 957                 lli->lli_symlink_name = NULL;
 958                 init_rwsem(&lli->lli_trunc_sem);
 959                 range_lock_tree_init(&lli->lli_write_tree);
 960                 init_rwsem(&lli->lli_glimpse_sem);
 961                 lli->lli_glimpse_time = ktime_set(0, 0);
 962                 INIT_LIST_HEAD(&lli->lli_agl_list);
 963                 lli->lli_agl_index = 0;
 964                 lli->lli_async_rc = 0;
 965         }
 966         mutex_init(&lli->lli_layout_mutex);
 967         memset(lli->lli_jobid, 0, sizeof(lli->lli_jobid));
 968 }
 969
 970 #define MAX_STRING_SIZE 128
 971
 972 #ifndef HAVE_SUPER_SETUP_BDI_NAME
 973
 974 #define LSI_BDI_INITIALIZED     0x00400000
 975
 976 #ifndef HAVE_BDI_CAP_MAP_COPY
 977 # define BDI_CAP_MAP_COPY       0
 978 #endif
 979
 980 static int super_setup_bdi_name(struct super_block *sb, char *fmt, ...)
 981 {
 982         struct  lustre_sb_info *lsi = s2lsi(sb);
 983         char buf[MAX_STRING_SIZE];
 984         va_list args;
 985         int err;
 986
 987         err = bdi_init(&lsi->lsi_bdi);
 988         if (err)
 989                 return err;
 990
 991         lsi->lsi_flags |= LSI_BDI_INITIALIZED;
 992         lsi->lsi_bdi.capabilities = BDI_CAP_MAP_COPY;
 993         lsi->lsi_bdi.name = "lustre";
 994         va_start(args, fmt);
 995         vsnprintf(buf, MAX_STRING_SIZE, fmt, args);
 996         va_end(args);
 997         err = bdi_register(&lsi->lsi_bdi, NULL, "%s", buf);
 998         va_end(args);
 999         if (!err)
1000                 sb->s_bdi = &lsi->lsi_bdi;
1001
1002         return err;
1003 }
1004 #endif /* !HAVE_SUPER_SETUP_BDI_NAME */
1005
1006 int ll_fill_super(struct super_block *sb, struct vfsmount *mnt)
1007 {
1008         struct  lustre_profile *lprof = NULL;
1009         struct  lustre_sb_info *lsi = s2lsi(sb);
1010         struct  ll_sb_info *sbi = NULL;
1011         char    *dt = NULL, *md = NULL;
1012         char    *profilenm = get_profile_name(sb);
1013         struct config_llog_instance *cfg;
1014         /* %p for void* in printf needs 16+2 characters: 0xffffffffffffffff */
1015         const int instlen = sizeof(cfg->cfg_instance) * 2 + 2;
1016         char name[MAX_STRING_SIZE];
1017         int md_len = 0;
1018         int dt_len = 0;
1019         char *ptr;
1020         int len;
1021         int err;
1022
1023         ENTRY;
1024         CDEBUG(D_VFSTRACE, "VFS Op: sb %p\n", sb);
1025
1026         try_module_get(THIS_MODULE);
1027
1028         OBD_ALLOC_PTR(cfg);
1029         if (cfg == NULL)
1030                 GOTO(out_free, err = -ENOMEM);
1031
1032         /* client additional sb info */
1033         lsi->lsi_llsbi = sbi = ll_init_sbi();
1034         if (!sbi)
1035                 GOTO(out_free, err = -ENOMEM);
1036
1037         err = ll_options(lsi->lsi_lmd->lmd_opts, sbi);
1038         if (err)
1039                 GOTO(out_free, err);
1040
1041         err = super_setup_bdi_name(sb, "lustre-%p", sb);
1042         if (err)
1043                 GOTO(out_free, err);
1044
1045 #ifndef HAVE_DCACHE_LOCK
1046         /* kernel >= 2.6.38 store dentry operations in sb->s_d_op. */
1047         sb->s_d_op = &ll_d_ops;
1048 #endif
1049         /* Get fsname */
1050         len = strlen(lsi->lsi_lmd->lmd_profile);
1051         ptr = strrchr(lsi->lsi_lmd->lmd_profile, '-');
1052         if (ptr && (strcmp(ptr, "-client") == 0))
1053                 len -= 7;
1054
1055         /* Mount info */
1056         snprintf(name, MAX_STRING_SIZE, "%.*s-%p", len,
1057                  lsi->lsi_lmd->lmd_profile, sb);
1058
1059         /* Call ll_debugfs_register_super() before lustre_process_log()
1060          * so that "llite.*.*" params can be processed correctly.
1061          */
1062         err = ll_debugfs_register_super(sb, name);
1063         if (err < 0) {
1064                 CERROR("%s: could not register mountpoint in llite: rc = %d\n",
1065                        ll_get_fsname(sb, NULL, 0), err);
1066                 err = 0;
1067         }
1068
1069         /* Generate a string unique to this super, in case some joker tries
1070            to mount the same fs at two mount points.
1071            Use the address of the super itself.*/
1072         cfg->cfg_instance = sb;
1073         cfg->cfg_uuid = lsi->lsi_llsbi->ll_sb_uuid;
1074         cfg->cfg_callback = class_config_llog_handler;
1075         cfg->cfg_sub_clds = CONFIG_SUB_CLIENT;
1076         /* set up client obds */
1077         err = lustre_process_log(sb, profilenm, cfg);
1078         if (err < 0)
1079                 GOTO(out_proc, err);
1080
1081         /* Profile set with LCFG_MOUNTOPT so we can find our mdc and osc obds */
1082         lprof = class_get_profile(profilenm);
1083         if (lprof == NULL) {
1084                 LCONSOLE_ERROR_MSG(0x156, "The client profile '%s' could not be"
1085                                    " read from the MGS.  Does that filesystem "
1086                                    "exist?\n", profilenm);
1087                 GOTO(out_proc, err = -EINVAL);
1088         }
1089         CDEBUG(D_CONFIG, "Found profile %s: mdc=%s osc=%s\n", profilenm,
1090                lprof->lp_md, lprof->lp_dt);
1091
1092         dt_len = strlen(lprof->lp_dt) + instlen + 2;
1093         OBD_ALLOC(dt, dt_len);
1094         if (!dt)
1095                 GOTO(out_proc, err = -ENOMEM);
1096         snprintf(dt, dt_len - 1, "%s-%p", lprof->lp_dt, cfg->cfg_instance);
1097
1098         md_len = strlen(lprof->lp_md) + instlen + 2;
1099         OBD_ALLOC(md, md_len);
1100         if (!md)
1101                 GOTO(out_proc, err = -ENOMEM);
1102         snprintf(md, md_len - 1, "%s-%p", lprof->lp_md, cfg->cfg_instance);
1103
1104         /* connections, registrations, sb setup */
1105         err = client_common_fill_super(sb, md, dt, mnt);
1106         if (err < 0)
1107                 GOTO(out_proc, err);
1108
1109         sbi->ll_client_common_fill_super_succeeded = 1;
1110
1111 out_proc:
1112         if (err < 0)
1113                 ll_debugfs_unregister_super(sb);
1114 out_free:
1115         if (md)
1116                 OBD_FREE(md, md_len);
1117         if (dt)
1118                 OBD_FREE(dt, dt_len);
1119         if (lprof != NULL)
1120                 class_put_profile(lprof);
1121         if (cfg)
1122                 OBD_FREE_PTR(cfg);
1123         if (err)
1124                 ll_put_super(sb);
1125         else if (sbi->ll_flags & LL_SBI_VERBOSE)
1126                 LCONSOLE_WARN("Mounted %s\n", profilenm);
1127         RETURN(err);
1128 } /* ll_fill_super */
1129
1130 void ll_put_super(struct super_block *sb)
1131 {
1132         struct config_llog_instance cfg, params_cfg;
1133         struct obd_device *obd;
1134         struct lustre_sb_info *lsi = s2lsi(sb);
1135         struct ll_sb_info *sbi = ll_s2sbi(sb);
1136         char *profilenm = get_profile_name(sb);
1137         long ccc_count;
1138         int next, force = 1, rc = 0;
1139         ENTRY;
1140
1141         if (!sbi)
1142                 GOTO(out_no_sbi, 0);
1143
1144         CDEBUG(D_VFSTRACE, "VFS Op: sb %p - %s\n", sb, profilenm);
1145
1146         cfg.cfg_instance = sb;
1147         lustre_end_log(sb, profilenm, &cfg);
1148
1149         params_cfg.cfg_instance = sb;
1150         lustre_end_log(sb, PARAMS_FILENAME, &params_cfg);
1151
1152         if (sbi->ll_md_exp) {
1153                 obd = class_exp2obd(sbi->ll_md_exp);
1154                 if (obd)
1155                         force = obd->obd_force;
1156         }
1157
1158         /* Wait for unstable pages to be committed to stable storage */
1159         if (force == 0) {
1160                 struct l_wait_info lwi = LWI_INTR(LWI_ON_SIGNAL_NOOP, NULL);
1161                 rc = l_wait_event(sbi->ll_cache->ccc_unstable_waitq,
1162                         atomic_long_read(&sbi->ll_cache->ccc_unstable_nr) == 0,
1163                         &lwi);
1164         }
1165
1166         ccc_count = atomic_long_read(&sbi->ll_cache->ccc_unstable_nr);
1167         if (force == 0 && rc != -EINTR)
1168                 LASSERTF(ccc_count == 0, "count: %li\n", ccc_count);
1169
1170
1171         /* We need to set force before the lov_disconnect in
1172            lustre_common_put_super, since l_d cleans up osc's as well. */
1173         if (force) {
1174                 next = 0;
1175                 while ((obd = class_devices_in_group(&sbi->ll_sb_uuid,
1176                                                      &next)) != NULL) {
1177                         obd->obd_force = force;
1178                 }
1179         }
1180
1181         if (sbi->ll_client_common_fill_super_succeeded) {
1182                 /* Only if client_common_fill_super succeeded */
1183                 client_common_put_super(sb);
1184         }
1185
1186         next = 0;
1187         while ((obd = class_devices_in_group(&sbi->ll_sb_uuid, &next)) !=NULL) {
1188                 class_manual_cleanup(obd);
1189         }
1190
1191         if (sbi->ll_flags & LL_SBI_VERBOSE)
1192                 LCONSOLE_WARN("Unmounted %s\n", profilenm ? profilenm : "");
1193
1194         if (profilenm)
1195                 class_del_profile(profilenm);
1196
1197 #ifndef HAVE_SUPER_SETUP_BDI_NAME
1198         if (lsi->lsi_flags & LSI_BDI_INITIALIZED) {
1199                 bdi_destroy(&lsi->lsi_bdi);
1200                 lsi->lsi_flags &= ~LSI_BDI_INITIALIZED;
1201         }
1202 #endif
1203
1204         ll_free_sbi(sb);
1205         lsi->lsi_llsbi = NULL;
1206 out_no_sbi:
1207         lustre_common_put_super(sb);
1208
1209         cl_env_cache_purge(~0);
1210
1211         module_put(THIS_MODULE);
1212
1213         EXIT;
1214 } /* client_put_super */
1215
1216 struct inode *ll_inode_from_resource_lock(struct ldlm_lock *lock)
1217 {
1218         struct inode *inode = NULL;
1219
1220         /* NOTE: we depend on atomic igrab() -bzzz */
1221         lock_res_and_lock(lock);
1222         if (lock->l_resource->lr_lvb_inode) {
1223                 struct ll_inode_info * lli;
1224                 lli = ll_i2info(lock->l_resource->lr_lvb_inode);
1225                 if (lli->lli_inode_magic == LLI_INODE_MAGIC) {
1226                         inode = igrab(lock->l_resource->lr_lvb_inode);
1227                 } else {
1228                         inode = lock->l_resource->lr_lvb_inode;
1229                         LDLM_DEBUG_LIMIT(inode->i_state & I_FREEING ?  D_INFO :
1230                                          D_WARNING, lock, "lr_lvb_inode %p is "
1231                                          "bogus: magic %08x",
1232                                          lock->l_resource->lr_lvb_inode,
1233                                          lli->lli_inode_magic);
1234                         inode = NULL;
1235                 }
1236         }
1237         unlock_res_and_lock(lock);
1238         return inode;
1239 }
1240
1241 void ll_dir_clear_lsm_md(struct inode *inode)
1242 {
1243         struct ll_inode_info *lli = ll_i2info(inode);
1244
1245         LASSERT(S_ISDIR(inode->i_mode));
1246
1247         if (lli->lli_lsm_md != NULL) {
1248                 lmv_free_memmd(lli->lli_lsm_md);
1249                 lli->lli_lsm_md = NULL;
1250         }
1251 }
1252
1253 static struct inode *ll_iget_anon_dir(struct super_block *sb,
1254                                       const struct lu_fid *fid,
1255                                       struct lustre_md *md)
1256 {
1257         struct ll_sb_info       *sbi = ll_s2sbi(sb);
1258         struct mdt_body         *body = md->body;
1259         struct inode            *inode;
1260         ino_t                   ino;
1261         ENTRY;
1262
1263         ino = cl_fid_build_ino(fid, sbi->ll_flags & LL_SBI_32BIT_API);
1264         inode = iget_locked(sb, ino);
1265         if (inode == NULL) {
1266                 CERROR("%s: failed get simple inode "DFID": rc = -ENOENT\n",
1267                        ll_get_fsname(sb, NULL, 0), PFID(fid));
1268                 RETURN(ERR_PTR(-ENOENT));
1269         }
1270
1271         if (inode->i_state & I_NEW) {
1272                 struct ll_inode_info *lli = ll_i2info(inode);
1273                 struct lmv_stripe_md *lsm = md->lmv;
1274
1275                 inode->i_mode = (inode->i_mode & ~S_IFMT) |
1276                                 (body->mbo_mode & S_IFMT);
1277                 LASSERTF(S_ISDIR(inode->i_mode), "Not slave inode "DFID"\n",
1278                          PFID(fid));
1279
1280                 LTIME_S(inode->i_mtime) = 0;
1281                 LTIME_S(inode->i_atime) = 0;
1282                 LTIME_S(inode->i_ctime) = 0;
1283                 inode->i_rdev = 0;
1284
1285 #ifdef HAVE_BACKING_DEV_INFO
1286                 /* initializing backing dev info. */
1287                 inode->i_mapping->backing_dev_info =
1288                                                 &s2lsi(inode->i_sb)->lsi_bdi;
1289 #endif
1290                 inode->i_op = &ll_dir_inode_operations;
1291                 inode->i_fop = &ll_dir_operations;
1292                 lli->lli_fid = *fid;
1293                 ll_lli_init(lli);
1294
1295                 LASSERT(lsm != NULL);
1296                 /* master object FID */
1297                 lli->lli_pfid = body->mbo_fid1;
1298                 CDEBUG(D_INODE, "lli %p slave "DFID" master "DFID"\n",
1299                        lli, PFID(fid), PFID(&lli->lli_pfid));
1300                 unlock_new_inode(inode);
1301         }
1302
1303         RETURN(inode);
1304 }
1305
1306 static int ll_init_lsm_md(struct inode *inode, struct lustre_md *md)
1307 {
1308         struct lu_fid *fid;
1309         struct lmv_stripe_md *lsm = md->lmv;
1310         int i;
1311
1312         LASSERT(lsm != NULL);
1313         /* XXX sigh, this lsm_root initialization should be in
1314          * LMV layer, but it needs ll_iget right now, so we
1315          * put this here right now. */
1316         for (i = 0; i < lsm->lsm_md_stripe_count; i++) {
1317                 fid = &lsm->lsm_md_oinfo[i].lmo_fid;
1318                 LASSERT(lsm->lsm_md_oinfo[i].lmo_root == NULL);
1319                 /* Unfortunately ll_iget will call ll_update_inode,
1320                  * where the initialization of slave inode is slightly
1321                  * different, so it reset lsm_md to NULL to avoid
1322                  * initializing lsm for slave inode. */
1323                 /* For migrating inode, master stripe and master object will
1324                  * be same, so we only need assign this inode */
1325                 if (lsm->lsm_md_hash_type & LMV_HASH_FLAG_MIGRATION && i == 0)
1326                         lsm->lsm_md_oinfo[i].lmo_root = inode;
1327                 else
1328                         lsm->lsm_md_oinfo[i].lmo_root =
1329                                 ll_iget_anon_dir(inode->i_sb, fid, md);
1330
1331                 if (IS_ERR(lsm->lsm_md_oinfo[i].lmo_root)) {
1332                         int rc = PTR_ERR(lsm->lsm_md_oinfo[i].lmo_root);
1333
1334                         lsm->lsm_md_oinfo[i].lmo_root = NULL;
1335                         return rc;
1336                 }
1337         }
1338
1339         return 0;
1340 }
1341
1342 static inline int lli_lsm_md_eq(const struct lmv_stripe_md *lsm_md1,
1343                                 const struct lmv_stripe_md *lsm_md2)
1344 {
1345         return lsm_md1->lsm_md_magic == lsm_md2->lsm_md_magic &&
1346                lsm_md1->lsm_md_stripe_count == lsm_md2->lsm_md_stripe_count &&
1347                lsm_md1->lsm_md_master_mdt_index ==
1348                                         lsm_md2->lsm_md_master_mdt_index &&
1349                lsm_md1->lsm_md_hash_type == lsm_md2->lsm_md_hash_type &&
1350                lsm_md1->lsm_md_layout_version ==
1351                                         lsm_md2->lsm_md_layout_version &&
1352                strcmp(lsm_md1->lsm_md_pool_name,
1353                       lsm_md2->lsm_md_pool_name) == 0;
1354 }
1355
1356 static int ll_update_lsm_md(struct inode *inode, struct lustre_md *md)
1357 {
1358         struct ll_inode_info *lli = ll_i2info(inode);
1359         struct lmv_stripe_md *lsm = md->lmv;
1360         int     rc;
1361         ENTRY;
1362
1363         LASSERT(S_ISDIR(inode->i_mode));
1364         CDEBUG(D_INODE, "update lsm %p of "DFID"\n", lli->lli_lsm_md,
1365                PFID(ll_inode2fid(inode)));
1366
1367         /* no striped information from request. */
1368         if (lsm == NULL) {
1369                 if (lli->lli_lsm_md == NULL) {
1370                         RETURN(0);
1371                 } else if (lli->lli_lsm_md->lsm_md_hash_type &
1372                                                 LMV_HASH_FLAG_MIGRATION) {
1373                         /* migration is done, the temporay MIGRATE layout has
1374                          * been removed */
1375                         CDEBUG(D_INODE, DFID" finish migration.\n",
1376                                PFID(ll_inode2fid(inode)));
1377                         lmv_free_memmd(lli->lli_lsm_md);
1378                         lli->lli_lsm_md = NULL;
1379                         RETURN(0);
1380                 } else {
1381                         /* The lustre_md from req does not include stripeEA,
1382                          * see ll_md_setattr */
1383                         RETURN(0);
1384                 }
1385         }
1386
1387         /* set the directory layout */
1388         if (lli->lli_lsm_md == NULL) {
1389                 struct cl_attr  *attr;
1390
1391                 rc = ll_init_lsm_md(inode, md);
1392                 if (rc != 0)
1393                         RETURN(rc);
1394
1395                 /* set md->lmv to NULL, so the following free lustre_md
1396                  * will not free this lsm */
1397                 md->lmv = NULL;
1398                 lli->lli_lsm_md = lsm;
1399
1400                 OBD_ALLOC_PTR(attr);
1401                 if (attr == NULL)
1402                         RETURN(-ENOMEM);
1403
1404                 /* validate the lsm */
1405                 rc = md_merge_attr(ll_i2mdexp(inode), lsm, attr,
1406                                    ll_md_blocking_ast);
1407                 if (rc != 0) {
1408                         OBD_FREE_PTR(attr);
1409                         RETURN(rc);
1410                 }
1411
1412                 if (md->body->mbo_valid & OBD_MD_FLNLINK)
1413                         md->body->mbo_nlink = attr->cat_nlink;
1414                 if (md->body->mbo_valid & OBD_MD_FLSIZE)
1415                         md->body->mbo_size = attr->cat_size;
1416                 if (md->body->mbo_valid & OBD_MD_FLATIME)
1417                         md->body->mbo_atime = attr->cat_atime;
1418                 if (md->body->mbo_valid & OBD_MD_FLCTIME)
1419                         md->body->mbo_ctime = attr->cat_ctime;
1420                 if (md->body->mbo_valid & OBD_MD_FLMTIME)
1421                         md->body->mbo_mtime = attr->cat_mtime;
1422
1423                 OBD_FREE_PTR(attr);
1424
1425                 CDEBUG(D_INODE, "Set lsm %p magic %x to "DFID"\n", lsm,
1426                        lsm->lsm_md_magic, PFID(ll_inode2fid(inode)));
1427                 RETURN(0);
1428         }
1429
1430         /* Compare the old and new stripe information */
1431         if (!lsm_md_eq(lli->lli_lsm_md, lsm)) {
1432                 struct lmv_stripe_md    *old_lsm = lli->lli_lsm_md;
1433                 int                     idx;
1434
1435                 CERROR("%s: inode "DFID"(%p)'s lmv layout mismatch (%p)/(%p)"
1436                        "magic:0x%x/0x%x stripe count: %d/%d master_mdt: %d/%d"
1437                        "hash_type:0x%x/0x%x layout: 0x%x/0x%x pool:%s/%s\n",
1438                        ll_get_fsname(inode->i_sb, NULL, 0), PFID(&lli->lli_fid),
1439                        inode, lsm, old_lsm,
1440                        lsm->lsm_md_magic, old_lsm->lsm_md_magic,
1441                        lsm->lsm_md_stripe_count,
1442                        old_lsm->lsm_md_stripe_count,
1443                        lsm->lsm_md_master_mdt_index,
1444                        old_lsm->lsm_md_master_mdt_index,
1445                        lsm->lsm_md_hash_type, old_lsm->lsm_md_hash_type,
1446                        lsm->lsm_md_layout_version,
1447                        old_lsm->lsm_md_layout_version,
1448                        lsm->lsm_md_pool_name,
1449                        old_lsm->lsm_md_pool_name);
1450
1451                 for (idx = 0; idx < old_lsm->lsm_md_stripe_count; idx++) {
1452                         CERROR("%s: sub FIDs in old lsm idx %d, old: "DFID"\n",
1453                                ll_get_fsname(inode->i_sb, NULL, 0), idx,
1454                                PFID(&old_lsm->lsm_md_oinfo[idx].lmo_fid));
1455                 }
1456
1457                 for (idx = 0; idx < lsm->lsm_md_stripe_count; idx++) {
1458                         CERROR("%s: sub FIDs in new lsm idx %d, new: "DFID"\n",
1459                                ll_get_fsname(inode->i_sb, NULL, 0), idx,
1460                                PFID(&lsm->lsm_md_oinfo[idx].lmo_fid));
1461                 }
1462
1463                 RETURN(-EIO);
1464         }
1465
1466         RETURN(0);
1467 }
1468
1469 void ll_clear_inode(struct inode *inode)
1470 {
1471         struct ll_inode_info *lli = ll_i2info(inode);
1472         struct ll_sb_info *sbi = ll_i2sbi(inode);
1473         ENTRY;
1474
1475         CDEBUG(D_VFSTRACE, "VFS Op:inode="DFID"(%p)\n",
1476                PFID(ll_inode2fid(inode)), inode);
1477
1478         if (S_ISDIR(inode->i_mode)) {
1479                 /* these should have been cleared in ll_file_release */
1480                 LASSERT(lli->lli_opendir_key == NULL);
1481                 LASSERT(lli->lli_sai == NULL);
1482                 LASSERT(lli->lli_opendir_pid == 0);
1483         }
1484
1485         md_null_inode(sbi->ll_md_exp, ll_inode2fid(inode));
1486
1487         LASSERT(!lli->lli_open_fd_write_count);
1488         LASSERT(!lli->lli_open_fd_read_count);
1489         LASSERT(!lli->lli_open_fd_exec_count);
1490
1491         if (lli->lli_mds_write_och)
1492                 ll_md_real_close(inode, FMODE_WRITE);
1493         if (lli->lli_mds_exec_och)
1494                 ll_md_real_close(inode, FMODE_EXEC);
1495         if (lli->lli_mds_read_och)
1496                 ll_md_real_close(inode, FMODE_READ);
1497
1498         if (S_ISLNK(inode->i_mode) && lli->lli_symlink_name) {
1499                 OBD_FREE(lli->lli_symlink_name,
1500                          strlen(lli->lli_symlink_name) + 1);
1501                 lli->lli_symlink_name = NULL;
1502         }
1503
1504         ll_xattr_cache_destroy(inode);
1505
1506 #ifdef CONFIG_FS_POSIX_ACL
1507         forget_all_cached_acls(inode);
1508         if (lli->lli_posix_acl) {
1509                 LASSERT(atomic_read(&lli->lli_posix_acl->a_refcount) == 1);
1510                 posix_acl_release(lli->lli_posix_acl);
1511                 lli->lli_posix_acl = NULL;
1512         }
1513 #endif
1514         lli->lli_inode_magic = LLI_INODE_DEAD;
1515
1516         if (S_ISDIR(inode->i_mode))
1517                 ll_dir_clear_lsm_md(inode);
1518         else if (S_ISREG(inode->i_mode) && !is_bad_inode(inode))
1519                 LASSERT(list_empty(&lli->lli_agl_list));
1520
1521         /*
1522          * XXX This has to be done before lsm is freed below, because
1523          * cl_object still uses inode lsm.
1524          */
1525         cl_inode_fini(inode);
1526
1527         EXIT;
1528 }
1529
1530 static int ll_md_setattr(struct dentry *dentry, struct md_op_data *op_data)
1531 {
1532         struct lustre_md md;
1533         struct inode *inode = dentry->d_inode;
1534         struct ll_sb_info *sbi = ll_i2sbi(inode);
1535         struct ptlrpc_request *request = NULL;
1536         int rc, ia_valid;
1537         ENTRY;
1538
1539         op_data = ll_prep_md_op_data(op_data, inode, NULL, NULL, 0, 0,
1540                                      LUSTRE_OPC_ANY, NULL);
1541         if (IS_ERR(op_data))
1542                 RETURN(PTR_ERR(op_data));
1543
1544         rc = md_setattr(sbi->ll_md_exp, op_data, NULL, 0, &request);
1545         if (rc) {
1546                 ptlrpc_req_finished(request);
1547                 if (rc == -ENOENT) {
1548                         clear_nlink(inode);
1549                         /* Unlinked special device node? Or just a race?
1550                          * Pretend we done everything. */
1551                         if (!S_ISREG(inode->i_mode) &&
1552                             !S_ISDIR(inode->i_mode)) {
1553                                 ia_valid = op_data->op_attr.ia_valid;
1554                                 op_data->op_attr.ia_valid &= ~TIMES_SET_FLAGS;
1555                                 rc = simple_setattr(dentry, &op_data->op_attr);
1556                                 op_data->op_attr.ia_valid = ia_valid;
1557                         }
1558                 } else if (rc != -EPERM && rc != -EACCES && rc != -ETXTBSY) {
1559                         CERROR("md_setattr fails: rc = %d\n", rc);
1560                 }
1561                 RETURN(rc);
1562         }
1563
1564         rc = md_get_lustre_md(sbi->ll_md_exp, request, sbi->ll_dt_exp,
1565                               sbi->ll_md_exp, &md);
1566         if (rc) {
1567                 ptlrpc_req_finished(request);
1568                 RETURN(rc);
1569         }
1570
1571         ia_valid = op_data->op_attr.ia_valid;
1572         /* inode size will be in ll_setattr_ost, can't do it now since dirty
1573          * cache is not cleared yet. */
1574         op_data->op_attr.ia_valid &= ~(TIMES_SET_FLAGS | ATTR_SIZE);
1575         if (S_ISREG(inode->i_mode))
1576                 inode_lock(inode);
1577         rc = simple_setattr(dentry, &op_data->op_attr);
1578         if (S_ISREG(inode->i_mode))
1579                 inode_unlock(inode);
1580         op_data->op_attr.ia_valid = ia_valid;
1581
1582         rc = ll_update_inode(inode, &md);
1583         ptlrpc_req_finished(request);
1584
1585         RETURN(rc);
1586 }
1587
1588 /* If this inode has objects allocated to it (lsm != NULL), then the OST
1589  * object(s) determine the file size and mtime.  Otherwise, the MDS will
1590  * keep these values until such a time that objects are allocated for it.
1591  * We do the MDS operations first, as it is checking permissions for us.
1592  * We don't to the MDS RPC if there is nothing that we want to store there,
1593  * otherwise there is no harm in updating mtime/atime on the MDS if we are
1594  * going to do an RPC anyways.
1595  *
1596  * If we are doing a truncate, we will send the mtime and ctime updates
1597  * to the OST with the punch RPC, otherwise we do an explicit setattr RPC.
1598  * I don't believe it is possible to get e.g. ATTR_MTIME_SET and ATTR_SIZE
1599  * at the same time.
1600  *
1601  * In case of HSMimport, we only set attr on MDS.
1602  */
1603 int ll_setattr_raw(struct dentry *dentry, struct iattr *attr, bool hsm_import)
1604 {
1605         struct inode *inode = dentry->d_inode;
1606         struct ll_inode_info *lli = ll_i2info(inode);
1607         struct md_op_data *op_data = NULL;
1608         int rc = 0;
1609         ENTRY;
1610
1611         CDEBUG(D_VFSTRACE, "%s: setattr inode "DFID"(%p) from %llu to %llu, "
1612                "valid %x, hsm_import %d\n",
1613                ll_get_fsname(inode->i_sb, NULL, 0), PFID(&lli->lli_fid),
1614                inode, i_size_read(inode), attr->ia_size, attr->ia_valid,
1615                hsm_import);
1616
1617         if (attr->ia_valid & ATTR_SIZE) {
1618                 /* Check new size against VFS/VM file size limit and rlimit */
1619                 rc = inode_newsize_ok(inode, attr->ia_size);
1620                 if (rc)
1621                         RETURN(rc);
1622
1623                 /* The maximum Lustre file size is variable, based on the
1624                  * OST maximum object size and number of stripes.  This
1625                  * needs another check in addition to the VFS check above. */
1626                 if (attr->ia_size > ll_file_maxbytes(inode)) {
1627                         CDEBUG(D_INODE,"file "DFID" too large %llu > %llu\n",
1628                                PFID(&lli->lli_fid), attr->ia_size,
1629                                ll_file_maxbytes(inode));
1630                         RETURN(-EFBIG);
1631                 }
1632
1633                 attr->ia_valid |= ATTR_MTIME | ATTR_CTIME;
1634         }
1635
1636         /* POSIX: check before ATTR_*TIME_SET set (from inode_change_ok) */
1637         if (attr->ia_valid & TIMES_SET_FLAGS) {
1638                 if ((!uid_eq(current_fsuid(), inode->i_uid)) &&
1639                     !cfs_capable(CFS_CAP_FOWNER))
1640                         RETURN(-EPERM);
1641         }
1642
1643         /* We mark all of the fields "set" so MDS/OST does not re-set them */
1644         if (!(attr->ia_valid & ATTR_CTIME_SET) &&
1645             (attr->ia_valid & ATTR_CTIME)) {
1646                 attr->ia_ctime = current_time(inode);
1647                 attr->ia_valid |= ATTR_CTIME_SET;
1648         }
1649         if (!(attr->ia_valid & ATTR_ATIME_SET) &&
1650             (attr->ia_valid & ATTR_ATIME)) {
1651                 attr->ia_atime = current_time(inode);
1652                 attr->ia_valid |= ATTR_ATIME_SET;
1653         }
1654         if (!(attr->ia_valid & ATTR_MTIME_SET) &&
1655             (attr->ia_valid & ATTR_MTIME)) {
1656                 attr->ia_mtime = current_time(inode);
1657                 attr->ia_valid |= ATTR_MTIME_SET;
1658         }
1659
1660         if (attr->ia_valid & (ATTR_MTIME | ATTR_CTIME))
1661                 CDEBUG(D_INODE, "setting mtime %lu, ctime %lu, now = %llu\n",
1662                        LTIME_S(attr->ia_mtime), LTIME_S(attr->ia_ctime),
1663                        (s64)ktime_get_real_seconds());
1664
1665         if (S_ISREG(inode->i_mode)) {
1666                 if (attr->ia_valid & ATTR_SIZE)
1667                         inode_dio_write_done(inode);
1668                 inode_unlock(inode);
1669         }
1670
1671         /* We always do an MDS RPC, even if we're only changing the size;
1672          * only the MDS knows whether truncate() should fail with -ETXTBUSY */
1673
1674         OBD_ALLOC_PTR(op_data);
1675         if (op_data == NULL)
1676                 GOTO(out, rc = -ENOMEM);
1677
1678         if (!hsm_import && attr->ia_valid & ATTR_SIZE) {
1679                 /* If we are changing file size, file content is
1680                  * modified, flag it. */
1681                 attr->ia_valid |= MDS_OPEN_OWNEROVERRIDE;
1682                 op_data->op_bias |= MDS_DATA_MODIFIED;
1683                 ll_file_clear_flag(lli, LLIF_DATA_MODIFIED);
1684         }
1685
1686         if (attr->ia_valid & ATTR_FILE) {
1687                 struct ll_file_data *fd = LUSTRE_FPRIVATE(attr->ia_file);
1688
1689                 if (fd->fd_lease_och)
1690                         op_data->op_bias |= MDS_TRUNC_KEEP_LEASE;
1691         }
1692
1693         op_data->op_attr = *attr;
1694
1695         rc = ll_md_setattr(dentry, op_data);
1696         if (rc)
1697                 GOTO(out, rc);
1698
1699         if (!S_ISREG(inode->i_mode) || hsm_import)
1700                 GOTO(out, rc = 0);
1701
1702         if (attr->ia_valid & (ATTR_SIZE |
1703                               ATTR_ATIME | ATTR_ATIME_SET |
1704                               ATTR_MTIME | ATTR_MTIME_SET |
1705                               ATTR_CTIME | ATTR_CTIME_SET)) {
1706                 /* For truncate and utimes sending attributes to OSTs, setting
1707                  * mtime/atime to the past will be performed under PW [0:EOF]
1708                  * extent lock (new_size:EOF for truncate).  It may seem
1709                  * excessive to send mtime/atime updates to OSTs when not
1710                  * setting times to past, but it is necessary due to possible
1711                  * time de-synchronization between MDT inode and OST objects */
1712                 rc = cl_setattr_ost(lli->lli_clob, attr, 0);
1713         }
1714
1715         /* If the file was restored, it needs to set dirty flag.
1716          *
1717          * We've already sent MDS_DATA_MODIFIED flag in
1718          * ll_md_setattr() for truncate. However, the MDT refuses to
1719          * set the HS_DIRTY flag on released files, so we have to set
1720          * it again if the file has been restored. Please check how
1721          * LLIF_DATA_MODIFIED is set in vvp_io_setattr_fini().
1722          *
1723          * Please notice that if the file is not released, the previous
1724          * MDS_DATA_MODIFIED has taken effect and usually
1725          * LLIF_DATA_MODIFIED is not set(see vvp_io_setattr_fini()).
1726          * This way we can save an RPC for common open + trunc
1727          * operation. */
1728         if (ll_file_test_and_clear_flag(lli, LLIF_DATA_MODIFIED)) {
1729                 struct hsm_state_set hss = {
1730                         .hss_valid = HSS_SETMASK,
1731                         .hss_setmask = HS_DIRTY,
1732                 };
1733                 int rc2;
1734
1735                 rc2 = ll_hsm_state_set(inode, &hss);
1736                 /* truncate and write can happen at the same time, so that
1737                  * the file can be set modified even though the file is not
1738                  * restored from released state, and ll_hsm_state_set() is
1739                  * not applicable for the file, and rc2 < 0 is normal in this
1740                  * case. */
1741                 if (rc2 < 0)
1742                         CDEBUG(D_INFO, DFID "HSM set dirty failed: rc2 = %d\n",
1743                                PFID(ll_inode2fid(inode)), rc2);
1744         }
1745
1746         EXIT;
1747 out:
1748         if (op_data != NULL)
1749                 ll_finish_md_op_data(op_data);
1750
1751         if (S_ISREG(inode->i_mode)) {
1752                 inode_lock(inode);
1753                 if ((attr->ia_valid & ATTR_SIZE) && !hsm_import)
1754                         inode_dio_wait(inode);
1755                 /* Once we've got the i_mutex, it's safe to set the S_NOSEC
1756                  * flag.  ll_update_inode (called from ll_md_setattr), clears
1757                  * inode flags, so there is a gap where S_NOSEC is not set.
1758                  * This can cause a writer to take the i_mutex unnecessarily,
1759                  * but this is safe to do and should be rare. */
1760                 inode_has_no_xattr(inode);
1761         }
1762
1763         ll_stats_ops_tally(ll_i2sbi(inode), (attr->ia_valid & ATTR_SIZE) ?
1764                         LPROC_LL_TRUNC : LPROC_LL_SETATTR, 1);
1765
1766         return rc;
1767 }
1768
1769 int ll_setattr(struct dentry *de, struct iattr *attr)
1770 {
1771         int mode = de->d_inode->i_mode;
1772
1773         if ((attr->ia_valid & (ATTR_CTIME|ATTR_SIZE|ATTR_MODE)) ==
1774                               (ATTR_CTIME|ATTR_SIZE|ATTR_MODE))
1775                 attr->ia_valid |= MDS_OPEN_OWNEROVERRIDE;
1776
1777         if (((attr->ia_valid & (ATTR_MODE|ATTR_FORCE|ATTR_SIZE)) ==
1778                                (ATTR_SIZE|ATTR_MODE)) &&
1779             (((mode & S_ISUID) && !(attr->ia_mode & S_ISUID)) ||
1780              (((mode & (S_ISGID|S_IXGRP)) == (S_ISGID|S_IXGRP)) &&
1781               !(attr->ia_mode & S_ISGID))))
1782                 attr->ia_valid |= ATTR_FORCE;
1783
1784         if ((attr->ia_valid & ATTR_MODE) &&
1785             (mode & S_ISUID) &&
1786             !(attr->ia_mode & S_ISUID) &&
1787             !(attr->ia_valid & ATTR_KILL_SUID))
1788                 attr->ia_valid |= ATTR_KILL_SUID;
1789
1790         if ((attr->ia_valid & ATTR_MODE) &&
1791             ((mode & (S_ISGID|S_IXGRP)) == (S_ISGID|S_IXGRP)) &&
1792             !(attr->ia_mode & S_ISGID) &&
1793             !(attr->ia_valid & ATTR_KILL_SGID))
1794                 attr->ia_valid |= ATTR_KILL_SGID;
1795
1796         /* avoid polluted from ATTR_TIMES_SET,
1797          * projid is not expected to be set here */
1798         attr->ia_valid &= ~MDS_ATTR_PROJID;
1799
1800         return ll_setattr_raw(de, attr, false);
1801 }
1802
1803 int ll_statfs_internal(struct ll_sb_info *sbi, struct obd_statfs *osfs,
1804                        u32 flags)
1805 {
1806         struct obd_statfs obd_osfs;
1807         time64_t max_age;
1808         int rc;
1809
1810         ENTRY;
1811         max_age = ktime_get_seconds() - OBD_STATFS_CACHE_SECONDS;
1812
1813         rc = obd_statfs(NULL, sbi->ll_md_exp, osfs, max_age, flags);
1814         if (rc) {
1815                 CERROR("md_statfs fails: rc = %d\n", rc);
1816                 RETURN(rc);
1817         }
1818
1819         osfs->os_type = LL_SUPER_MAGIC;
1820
1821         CDEBUG(D_SUPER, "MDC blocks %llu/%llu objects %llu/%llu\n",
1822                osfs->os_bavail, osfs->os_blocks, osfs->os_ffree,osfs->os_files);
1823
1824         if (osfs->os_state & OS_STATE_SUM)
1825                 GOTO(out, rc);
1826
1827         if (sbi->ll_flags & LL_SBI_LAZYSTATFS)
1828                 flags |= OBD_STATFS_NODELAY;
1829
1830         rc = obd_statfs(NULL, sbi->ll_dt_exp, &obd_osfs, max_age, flags);
1831         if (rc) {
1832                 CERROR("obd_statfs fails: rc = %d\n", rc);
1833                 RETURN(rc);
1834         }
1835
1836         CDEBUG(D_SUPER, "OSC blocks %llu/%llu objects %llu/%llu\n",
1837                obd_osfs.os_bavail, obd_osfs.os_blocks, obd_osfs.os_ffree,
1838                obd_osfs.os_files);
1839
1840         osfs->os_bsize = obd_osfs.os_bsize;
1841         osfs->os_blocks = obd_osfs.os_blocks;
1842         osfs->os_bfree = obd_osfs.os_bfree;
1843         osfs->os_bavail = obd_osfs.os_bavail;
1844
1845         /* If we don't have as many objects free on the OST as inodes
1846          * on the MDS, we reduce the total number of inodes to
1847          * compensate, so that the "inodes in use" number is correct.
1848          */
1849         if (obd_osfs.os_ffree < osfs->os_ffree) {
1850                 osfs->os_files = (osfs->os_files - osfs->os_ffree) +
1851                         obd_osfs.os_ffree;
1852                 osfs->os_ffree = obd_osfs.os_ffree;
1853         }
1854
1855 out:
1856         RETURN(rc);
1857 }
1858 int ll_statfs(struct dentry *de, struct kstatfs *sfs)
1859 {
1860         struct super_block *sb = de->d_sb;
1861         struct obd_statfs osfs;
1862         __u64 fsid = huge_encode_dev(sb->s_dev);
1863         int rc;
1864
1865         CDEBUG(D_VFSTRACE, "VFS Op: at %llu jiffies\n", get_jiffies_64());
1866         ll_stats_ops_tally(ll_s2sbi(sb), LPROC_LL_STAFS, 1);
1867
1868         /* Some amount of caching on the client is allowed */
1869         rc = ll_statfs_internal(ll_s2sbi(sb), &osfs, OBD_STATFS_SUM);
1870         if (rc)
1871                 return rc;
1872
1873         statfs_unpack(sfs, &osfs);
1874
1875         /* We need to downshift for all 32-bit kernels, because we can't
1876          * tell if the kernel is being called via sys_statfs64() or not.
1877          * Stop before overflowing f_bsize - in which case it is better
1878          * to just risk EOVERFLOW if caller is using old sys_statfs(). */
1879         if (sizeof(long) < 8) {
1880                 while (osfs.os_blocks > ~0UL && sfs->f_bsize < 0x40000000) {
1881                         sfs->f_bsize <<= 1;
1882
1883                         osfs.os_blocks >>= 1;
1884                         osfs.os_bfree >>= 1;
1885                         osfs.os_bavail >>= 1;
1886                 }
1887         }
1888
1889         sfs->f_blocks = osfs.os_blocks;
1890         sfs->f_bfree = osfs.os_bfree;
1891         sfs->f_bavail = osfs.os_bavail;
1892         sfs->f_fsid.val[0] = (__u32)fsid;
1893         sfs->f_fsid.val[1] = (__u32)(fsid >> 32);
1894         return 0;
1895 }
1896
1897 void ll_inode_size_lock(struct inode *inode)
1898 {
1899         struct ll_inode_info *lli;
1900
1901         LASSERT(!S_ISDIR(inode->i_mode));
1902
1903         lli = ll_i2info(inode);
1904         mutex_lock(&lli->lli_size_mutex);
1905 }
1906
1907 void ll_inode_size_unlock(struct inode *inode)
1908 {
1909         struct ll_inode_info *lli;
1910
1911         lli = ll_i2info(inode);
1912         mutex_unlock(&lli->lli_size_mutex);
1913 }
1914
1915 void ll_update_inode_flags(struct inode *inode, int ext_flags)
1916 {
1917         inode->i_flags = ll_ext_to_inode_flags(ext_flags);
1918         if (ext_flags & LUSTRE_PROJINHERIT_FL)
1919                 ll_file_set_flag(ll_i2info(inode), LLIF_PROJECT_INHERIT);
1920         else
1921                 ll_file_clear_flag(ll_i2info(inode), LLIF_PROJECT_INHERIT);
1922 }
1923
1924 int ll_update_inode(struct inode *inode, struct lustre_md *md)
1925 {
1926         struct ll_inode_info *lli = ll_i2info(inode);
1927         struct mdt_body *body = md->body;
1928         struct ll_sb_info *sbi = ll_i2sbi(inode);
1929         int rc = 0;
1930
1931         if (body->mbo_valid & OBD_MD_FLEASIZE) {
1932                 rc = cl_file_inode_init(inode, md);
1933                 if (rc)
1934                         return rc;
1935         }
1936
1937         if (S_ISDIR(inode->i_mode)) {
1938                 rc = ll_update_lsm_md(inode, md);
1939                 if (rc != 0)
1940                         return rc;
1941         }
1942
1943 #ifdef CONFIG_FS_POSIX_ACL
1944         if (body->mbo_valid & OBD_MD_FLACL) {
1945                 spin_lock(&lli->lli_lock);
1946                 if (lli->lli_posix_acl)
1947                         posix_acl_release(lli->lli_posix_acl);
1948                 lli->lli_posix_acl = md->posix_acl;
1949                 spin_unlock(&lli->lli_lock);
1950         }
1951 #endif
1952         inode->i_ino = cl_fid_build_ino(&body->mbo_fid1,
1953                                         sbi->ll_flags & LL_SBI_32BIT_API);
1954         inode->i_generation = cl_fid_build_gen(&body->mbo_fid1);
1955
1956         if (body->mbo_valid & OBD_MD_FLATIME) {
1957                 if (body->mbo_atime > LTIME_S(inode->i_atime))
1958                         LTIME_S(inode->i_atime) = body->mbo_atime;
1959                 lli->lli_atime = body->mbo_atime;
1960         }
1961
1962         if (body->mbo_valid & OBD_MD_FLMTIME) {
1963                 if (body->mbo_mtime > LTIME_S(inode->i_mtime)) {
1964                         CDEBUG(D_INODE, "setting ino %lu mtime from %lu "
1965                                "to %llu\n", inode->i_ino,
1966                                LTIME_S(inode->i_mtime), body->mbo_mtime);
1967                         LTIME_S(inode->i_mtime) = body->mbo_mtime;
1968                 }
1969                 lli->lli_mtime = body->mbo_mtime;
1970         }
1971
1972         if (body->mbo_valid & OBD_MD_FLCTIME) {
1973                 if (body->mbo_ctime > LTIME_S(inode->i_ctime))
1974                         LTIME_S(inode->i_ctime) = body->mbo_ctime;
1975                 lli->lli_ctime = body->mbo_ctime;
1976         }
1977
1978         /* Clear i_flags to remove S_NOSEC before permissions are updated */
1979         if (body->mbo_valid & OBD_MD_FLFLAGS)
1980                 ll_update_inode_flags(inode, body->mbo_flags);
1981         if (body->mbo_valid & OBD_MD_FLMODE)
1982                 inode->i_mode = (inode->i_mode & S_IFMT) |
1983                                 (body->mbo_mode & ~S_IFMT);
1984
1985         if (body->mbo_valid & OBD_MD_FLTYPE)
1986                 inode->i_mode = (inode->i_mode & ~S_IFMT) |
1987                                 (body->mbo_mode & S_IFMT);
1988
1989         LASSERT(inode->i_mode != 0);
1990         if (S_ISREG(inode->i_mode))
1991                 inode->i_blkbits = min(PTLRPC_MAX_BRW_BITS + 1,
1992                                        LL_MAX_BLKSIZE_BITS);
1993         else
1994                 inode->i_blkbits = inode->i_sb->s_blocksize_bits;
1995
1996         if (body->mbo_valid & OBD_MD_FLUID)
1997                 inode->i_uid = make_kuid(&init_user_ns, body->mbo_uid);
1998         if (body->mbo_valid & OBD_MD_FLGID)
1999                 inode->i_gid = make_kgid(&init_user_ns, body->mbo_gid);
2000         if (body->mbo_valid & OBD_MD_FLPROJID)
2001                 lli->lli_projid = body->mbo_projid;
2002         if (body->mbo_valid & OBD_MD_FLNLINK)
2003                 set_nlink(inode, body->mbo_nlink);
2004         if (body->mbo_valid & OBD_MD_FLRDEV)
2005                 inode->i_rdev = old_decode_dev(body->mbo_rdev);
2006
2007         if (body->mbo_valid & OBD_MD_FLID) {
2008                 /* FID shouldn't be changed! */
2009                 if (fid_is_sane(&lli->lli_fid)) {
2010                         LASSERTF(lu_fid_eq(&lli->lli_fid, &body->mbo_fid1),
2011                                  "Trying to change FID "DFID
2012                                  " to the "DFID", inode "DFID"(%p)\n",
2013                                  PFID(&lli->lli_fid), PFID(&body->mbo_fid1),
2014                                  PFID(ll_inode2fid(inode)), inode);
2015                 } else {
2016                         lli->lli_fid = body->mbo_fid1;
2017                 }
2018         }
2019
2020         LASSERT(fid_seq(&lli->lli_fid) != 0);
2021
2022         if (body->mbo_valid & OBD_MD_FLSIZE) {
2023                 i_size_write(inode, body->mbo_size);
2024
2025                 CDEBUG(D_VFSTRACE, "inode="DFID", updating i_size %llu\n",
2026                        PFID(ll_inode2fid(inode)),
2027                        (unsigned long long)body->mbo_size);
2028
2029                 if (body->mbo_valid & OBD_MD_FLBLOCKS)
2030                         inode->i_blocks = body->mbo_blocks;
2031         }
2032
2033         if (body->mbo_valid & OBD_MD_TSTATE) {
2034                 /* Set LLIF_FILE_RESTORING if restore ongoing and
2035                  * clear it when done to ensure to start again
2036                  * glimpsing updated attrs
2037                  */
2038                 if (body->mbo_t_state & MS_RESTORE)
2039                         ll_file_set_flag(lli, LLIF_FILE_RESTORING);
2040                 else
2041                         ll_file_clear_flag(lli, LLIF_FILE_RESTORING);
2042         }
2043
2044         return 0;
2045 }
2046
2047 int ll_read_inode2(struct inode *inode, void *opaque)
2048 {
2049         struct lustre_md *md = opaque;
2050         struct ll_inode_info *lli = ll_i2info(inode);
2051         int     rc;
2052         ENTRY;
2053
2054         CDEBUG(D_VFSTRACE, "VFS Op:inode="DFID"(%p)\n",
2055                PFID(&lli->lli_fid), inode);
2056
2057         /* Core attributes from the MDS first.  This is a new inode, and
2058          * the VFS doesn't zero times in the core inode so we have to do
2059          * it ourselves.  They will be overwritten by either MDS or OST
2060          * attributes - we just need to make sure they aren't newer. */
2061         LTIME_S(inode->i_mtime) = 0;
2062         LTIME_S(inode->i_atime) = 0;
2063         LTIME_S(inode->i_ctime) = 0;
2064         inode->i_rdev = 0;
2065         rc = ll_update_inode(inode, md);
2066         if (rc != 0)
2067                 RETURN(rc);
2068
2069         /* OIDEBUG(inode); */
2070
2071 #ifdef HAVE_BACKING_DEV_INFO
2072         /* initializing backing dev info. */
2073         inode->i_mapping->backing_dev_info = &s2lsi(inode->i_sb)->lsi_bdi;
2074 #endif
2075         if (S_ISREG(inode->i_mode)) {
2076                 struct ll_sb_info *sbi = ll_i2sbi(inode);
2077                 inode->i_op = &ll_file_inode_operations;
2078                 inode->i_fop = sbi->ll_fop;
2079                 inode->i_mapping->a_ops = (struct address_space_operations *)&ll_aops;
2080                 EXIT;
2081         } else if (S_ISDIR(inode->i_mode)) {
2082                 inode->i_op = &ll_dir_inode_operations;
2083                 inode->i_fop = &ll_dir_operations;
2084                 EXIT;
2085         } else if (S_ISLNK(inode->i_mode)) {
2086                 inode->i_op = &ll_fast_symlink_inode_operations;
2087                 EXIT;
2088         } else {
2089                 inode->i_op = &ll_special_inode_operations;
2090
2091                 init_special_inode(inode, inode->i_mode,
2092                                    inode->i_rdev);
2093
2094                 EXIT;
2095         }
2096
2097         return 0;
2098 }
2099
2100 void ll_delete_inode(struct inode *inode)
2101 {
2102         struct ll_inode_info *lli = ll_i2info(inode);
2103         ENTRY;
2104
2105         if (S_ISREG(inode->i_mode) && lli->lli_clob != NULL)
2106                 /* It is last chance to write out dirty pages,
2107                  * otherwise we may lose data while umount */
2108                 cl_sync_file_range(inode, 0, OBD_OBJECT_EOF, CL_FSYNC_LOCAL, 1);
2109
2110         truncate_inode_pages_final(&inode->i_data);
2111
2112         LASSERTF(inode->i_data.nrpages == 0, "inode="DFID"(%p) nrpages=%lu, "
2113                  "see https://jira.whamcloud.com/browse/LU-118\n",
2114                  PFID(ll_inode2fid(inode)), inode, inode->i_data.nrpages);
2115
2116 #ifdef HAVE_SBOPS_EVICT_INODE
2117         ll_clear_inode(inode);
2118 #endif
2119         clear_inode(inode);
2120
2121         EXIT;
2122 }
2123
2124 int ll_iocontrol(struct inode *inode, struct file *file,
2125                  unsigned int cmd, unsigned long arg)
2126 {
2127         struct ll_sb_info *sbi = ll_i2sbi(inode);
2128         struct ptlrpc_request *req = NULL;
2129         int rc, flags = 0;
2130         ENTRY;
2131
2132         switch (cmd) {
2133         case FS_IOC_GETFLAGS: {
2134                 struct mdt_body *body;
2135                 struct md_op_data *op_data;
2136
2137                 op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL,
2138                                              0, 0, LUSTRE_OPC_ANY,
2139                                              NULL);
2140                 if (IS_ERR(op_data))
2141                         RETURN(PTR_ERR(op_data));
2142
2143                 op_data->op_valid = OBD_MD_FLFLAGS;
2144                 rc = md_getattr(sbi->ll_md_exp, op_data, &req);
2145                 ll_finish_md_op_data(op_data);
2146                 if (rc) {
2147                         CERROR("%s: failure inode "DFID": rc = %d\n",
2148                                sbi->ll_md_exp->exp_obd->obd_name,
2149                                PFID(ll_inode2fid(inode)), rc);
2150                         RETURN(-abs(rc));
2151                 }
2152
2153                 body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
2154
2155                 flags = body->mbo_flags;
2156
2157                 ptlrpc_req_finished(req);
2158
2159                 RETURN(put_user(flags, (int __user *)arg));
2160         }
2161         case FS_IOC_SETFLAGS: {
2162                 struct iattr *attr;
2163                 struct md_op_data *op_data;
2164                 struct cl_object *obj;
2165
2166                 if (get_user(flags, (int __user *)arg))
2167                         RETURN(-EFAULT);
2168
2169                 op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL, 0, 0,
2170                                              LUSTRE_OPC_ANY, NULL);
2171                 if (IS_ERR(op_data))
2172                         RETURN(PTR_ERR(op_data));
2173
2174                 op_data->op_attr_flags = flags;
2175                 op_data->op_attr.ia_valid |= ATTR_ATTR_FLAG;
2176                 rc = md_setattr(sbi->ll_md_exp, op_data, NULL, 0, &req);
2177                 ll_finish_md_op_data(op_data);
2178                 ptlrpc_req_finished(req);
2179                 if (rc)
2180                         RETURN(rc);
2181
2182                 ll_update_inode_flags(inode, flags);
2183
2184                 obj = ll_i2info(inode)->lli_clob;
2185                 if (obj == NULL)
2186                         RETURN(0);
2187
2188                 OBD_ALLOC_PTR(attr);
2189                 if (attr == NULL)
2190                         RETURN(-ENOMEM);
2191
2192                 attr->ia_valid = ATTR_ATTR_FLAG;
2193                 rc = cl_setattr_ost(obj, attr, flags);
2194
2195                 OBD_FREE_PTR(attr);
2196                 RETURN(rc);
2197         }
2198         default:
2199                 RETURN(-ENOSYS);
2200         }
2201
2202         RETURN(0);
2203 }
2204
2205 int ll_flush_ctx(struct inode *inode)
2206 {
2207         struct ll_sb_info  *sbi = ll_i2sbi(inode);
2208
2209         CDEBUG(D_SEC, "flush context for user %d\n",
2210                from_kuid(&init_user_ns, current_uid()));
2211
2212         obd_set_info_async(NULL, sbi->ll_md_exp,
2213                            sizeof(KEY_FLUSH_CTX), KEY_FLUSH_CTX,
2214                            0, NULL, NULL);
2215         obd_set_info_async(NULL, sbi->ll_dt_exp,
2216                            sizeof(KEY_FLUSH_CTX), KEY_FLUSH_CTX,
2217                            0, NULL, NULL);
2218         return 0;
2219 }
2220
2221 /* umount -f client means force down, don't save state */
2222 void ll_umount_begin(struct super_block *sb)
2223 {
2224         struct ll_sb_info *sbi = ll_s2sbi(sb);
2225         struct obd_device *obd;
2226         struct obd_ioctl_data *ioc_data;
2227         struct l_wait_info lwi;
2228         wait_queue_head_t waitq;
2229         ENTRY;
2230
2231         CDEBUG(D_VFSTRACE, "VFS Op: superblock %p count %d active %d\n", sb,
2232                sb->s_count, atomic_read(&sb->s_active));
2233
2234         obd = class_exp2obd(sbi->ll_md_exp);
2235         if (obd == NULL) {
2236                 CERROR("Invalid MDC connection handle %#llx\n",
2237                        sbi->ll_md_exp->exp_handle.h_cookie);
2238                 EXIT;
2239                 return;
2240         }
2241         obd->obd_force = 1;
2242
2243         obd = class_exp2obd(sbi->ll_dt_exp);
2244         if (obd == NULL) {
2245                 CERROR("Invalid LOV connection handle %#llx\n",
2246                        sbi->ll_dt_exp->exp_handle.h_cookie);
2247                 EXIT;
2248                 return;
2249         }
2250         obd->obd_force = 1;
2251
2252         OBD_ALLOC_PTR(ioc_data);
2253         if (ioc_data) {
2254                 obd_iocontrol(IOC_OSC_SET_ACTIVE, sbi->ll_md_exp,
2255                               sizeof *ioc_data, ioc_data, NULL);
2256
2257                 obd_iocontrol(IOC_OSC_SET_ACTIVE, sbi->ll_dt_exp,
2258                               sizeof *ioc_data, ioc_data, NULL);
2259
2260                 OBD_FREE_PTR(ioc_data);
2261         }
2262
2263         /* Really, we'd like to wait until there are no requests outstanding,
2264          * and then continue.  For now, we just periodically checking for vfs
2265          * to decrement mnt_cnt and hope to finish it within 10sec.
2266          */
2267         init_waitqueue_head(&waitq);
2268         lwi = LWI_TIMEOUT_INTERVAL(cfs_time_seconds(10),
2269                                    cfs_time_seconds(1), NULL, NULL);
2270         l_wait_event(waitq, may_umount(sbi->ll_mnt.mnt), &lwi);
2271
2272         EXIT;
2273 }
2274
2275 int ll_remount_fs(struct super_block *sb, int *flags, char *data)
2276 {
2277         struct ll_sb_info *sbi = ll_s2sbi(sb);
2278         char *profilenm = get_profile_name(sb);
2279         int err;
2280         __u32 read_only;
2281
2282         if ((*flags & MS_RDONLY) != (sb->s_flags & MS_RDONLY)) {
2283                 read_only = *flags & MS_RDONLY;
2284                 err = obd_set_info_async(NULL, sbi->ll_md_exp,
2285                                          sizeof(KEY_READ_ONLY),
2286                                          KEY_READ_ONLY, sizeof(read_only),
2287                                          &read_only, NULL);
2288                 if (err) {
2289                         LCONSOLE_WARN("Failed to remount %s %s (%d)\n",
2290                                       profilenm, read_only ?
2291                                       "read-only" : "read-write", err);
2292                         return err;
2293                 }
2294
2295                 if (read_only)
2296                         sb->s_flags |= MS_RDONLY;
2297                 else
2298                         sb->s_flags &= ~MS_RDONLY;
2299
2300                 if (sbi->ll_flags & LL_SBI_VERBOSE)
2301                         LCONSOLE_WARN("Remounted %s %s\n", profilenm,
2302                                       read_only ?  "read-only" : "read-write");
2303         }
2304         return 0;
2305 }
2306
2307 /**
2308  * Cleanup the open handle that is cached on MDT-side.
2309  *
2310  * For open case, the client side open handling thread may hit error
2311  * after the MDT grant the open. Under such case, the client should
2312  * send close RPC to the MDT as cleanup; otherwise, the open handle
2313  * on the MDT will be leaked there until the client umount or evicted.
2314  *
2315  * In further, if someone unlinked the file, because the open handle
2316  * holds the reference on such file/object, then it will block the
2317  * subsequent threads that want to locate such object via FID.
2318  *
2319  * \param[in] sb        super block for this file-system
2320  * \param[in] open_req  pointer to the original open request
2321  */
2322 void ll_open_cleanup(struct super_block *sb, struct ptlrpc_request *open_req)
2323 {
2324         struct mdt_body                 *body;
2325         struct md_op_data               *op_data;
2326         struct ptlrpc_request           *close_req = NULL;
2327         struct obd_export               *exp       = ll_s2sbi(sb)->ll_md_exp;
2328         ENTRY;
2329
2330         body = req_capsule_server_get(&open_req->rq_pill, &RMF_MDT_BODY);
2331         OBD_ALLOC_PTR(op_data);
2332         if (op_data == NULL) {
2333                 CWARN("%s: cannot allocate op_data to release open handle for "
2334                       DFID"\n",
2335                       ll_get_fsname(sb, NULL, 0), PFID(&body->mbo_fid1));
2336
2337                 RETURN_EXIT;
2338         }
2339
2340         op_data->op_fid1 = body->mbo_fid1;
2341         op_data->op_handle = body->mbo_handle;
2342         op_data->op_mod_time = ktime_get_real_seconds();
2343         md_close(exp, op_data, NULL, &close_req);
2344         ptlrpc_req_finished(close_req);
2345         ll_finish_md_op_data(op_data);
2346
2347         EXIT;
2348 }
2349
2350 int ll_prep_inode(struct inode **inode, struct ptlrpc_request *req,
2351                   struct super_block *sb, struct lookup_intent *it)
2352 {
2353         struct ll_sb_info *sbi = NULL;
2354         struct lustre_md md = { NULL };
2355         int rc;
2356         ENTRY;
2357
2358         LASSERT(*inode || sb);
2359         sbi = sb ? ll_s2sbi(sb) : ll_i2sbi(*inode);
2360         rc = md_get_lustre_md(sbi->ll_md_exp, req, sbi->ll_dt_exp,
2361                               sbi->ll_md_exp, &md);
2362         if (rc != 0)
2363                 GOTO(cleanup, rc);
2364
2365         if (*inode) {
2366                 rc = ll_update_inode(*inode, &md);
2367                 if (rc != 0)
2368                         GOTO(out, rc);
2369         } else {
2370                 LASSERT(sb != NULL);
2371
2372                 /*
2373                  * At this point server returns to client's same fid as client
2374                  * generated for creating. So using ->fid1 is okay here.
2375                  */
2376                 if (!fid_is_sane(&md.body->mbo_fid1)) {
2377                         CERROR("%s: Fid is insane "DFID"\n",
2378                                 ll_get_fsname(sb, NULL, 0),
2379                                 PFID(&md.body->mbo_fid1));
2380                         GOTO(out, rc = -EINVAL);
2381                 }
2382
2383                 *inode = ll_iget(sb, cl_fid_build_ino(&md.body->mbo_fid1,
2384                                              sbi->ll_flags & LL_SBI_32BIT_API),
2385                                  &md);
2386                 if (IS_ERR(*inode)) {
2387 #ifdef CONFIG_FS_POSIX_ACL
2388                         if (md.posix_acl) {
2389                                 posix_acl_release(md.posix_acl);
2390                                 md.posix_acl = NULL;
2391                         }
2392 #endif
2393                         rc = IS_ERR(*inode) ? PTR_ERR(*inode) : -ENOMEM;
2394                         *inode = NULL;
2395                         CERROR("new_inode -fatal: rc %d\n", rc);
2396                         GOTO(out, rc);
2397                 }
2398         }
2399
2400         /* Handling piggyback layout lock.
2401          * Layout lock can be piggybacked by getattr and open request.
2402          * The lsm can be applied to inode only if it comes with a layout lock
2403          * otherwise correct layout may be overwritten, for example:
2404          * 1. proc1: mdt returns a lsm but not granting layout
2405          * 2. layout was changed by another client
2406          * 3. proc2: refresh layout and layout lock granted
2407          * 4. proc1: to apply a stale layout */
2408         if (it != NULL && it->it_lock_mode != 0) {
2409                 struct lustre_handle lockh;
2410                 struct ldlm_lock *lock;
2411
2412                 lockh.cookie = it->it_lock_handle;
2413                 lock = ldlm_handle2lock(&lockh);
2414                 LASSERT(lock != NULL);
2415                 if (ldlm_has_layout(lock)) {
2416                         struct cl_object_conf conf;
2417
2418                         memset(&conf, 0, sizeof(conf));
2419                         conf.coc_opc = OBJECT_CONF_SET;
2420                         conf.coc_inode = *inode;
2421                         conf.coc_lock = lock;
2422                         conf.u.coc_layout = md.layout;
2423                         (void)ll_layout_conf(*inode, &conf);
2424                 }
2425                 LDLM_LOCK_PUT(lock);
2426         }
2427
2428         GOTO(out, rc = 0);
2429
2430 out:
2431         md_free_lustre_md(sbi->ll_md_exp, &md);
2432
2433 cleanup:
2434         if (rc != 0 && it != NULL && it->it_op & IT_OPEN)
2435                 ll_open_cleanup(sb != NULL ? sb : (*inode)->i_sb, req);
2436
2437         return rc;
2438 }
2439
2440 int ll_obd_statfs(struct inode *inode, void __user *arg)
2441 {
2442         struct ll_sb_info *sbi = NULL;
2443         struct obd_export *exp;
2444         char *buf = NULL;
2445         struct obd_ioctl_data *data = NULL;
2446         __u32 type;
2447         int len = 0, rc;
2448
2449         if (!inode || !(sbi = ll_i2sbi(inode)))
2450                 GOTO(out_statfs, rc = -EINVAL);
2451
2452         rc = obd_ioctl_getdata(&buf, &len, arg);
2453         if (rc)
2454                 GOTO(out_statfs, rc);
2455
2456         data = (void*)buf;
2457         if (!data->ioc_inlbuf1 || !data->ioc_inlbuf2 ||
2458             !data->ioc_pbuf1 || !data->ioc_pbuf2)
2459                 GOTO(out_statfs, rc = -EINVAL);
2460
2461         if (data->ioc_inllen1 != sizeof(__u32) ||
2462             data->ioc_inllen2 != sizeof(__u32) ||
2463             data->ioc_plen1 != sizeof(struct obd_statfs) ||
2464             data->ioc_plen2 != sizeof(struct obd_uuid))
2465                 GOTO(out_statfs, rc = -EINVAL);
2466
2467         memcpy(&type, data->ioc_inlbuf1, sizeof(__u32));
2468         if (type & LL_STATFS_LMV)
2469                 exp = sbi->ll_md_exp;
2470         else if (type & LL_STATFS_LOV)
2471                 exp = sbi->ll_dt_exp;
2472         else
2473                 GOTO(out_statfs, rc = -ENODEV);
2474
2475         rc = obd_iocontrol(IOC_OBD_STATFS, exp, len, buf, NULL);
2476         if (rc)
2477                 GOTO(out_statfs, rc);
2478 out_statfs:
2479         OBD_FREE_LARGE(buf, len);
2480         return rc;
2481 }
2482
2483 /* this function prepares md_op_data hint for passing it down to MD stack. */
2484 struct md_op_data *ll_prep_md_op_data(struct md_op_data *op_data,
2485                                       struct inode *i1, struct inode *i2,
2486                                       const char *name, size_t namelen,
2487                                       __u32 mode, __u32 opc, void *data)
2488 {
2489         LASSERT(i1 != NULL);
2490
2491         if (name == NULL) {
2492                 /* Do not reuse namelen for something else. */
2493                 if (namelen != 0)
2494                         return ERR_PTR(-EINVAL);
2495         } else {
2496                 if (namelen > ll_i2sbi(i1)->ll_namelen)
2497                         return ERR_PTR(-ENAMETOOLONG);
2498
2499                 if (!lu_name_is_valid_2(name, namelen))
2500                         return ERR_PTR(-EINVAL);
2501         }
2502
2503         if (op_data == NULL)
2504                 OBD_ALLOC_PTR(op_data);
2505
2506         if (op_data == NULL)
2507                 return ERR_PTR(-ENOMEM);
2508
2509         ll_i2gids(op_data->op_suppgids, i1, i2);
2510         op_data->op_fid1 = *ll_inode2fid(i1);
2511         op_data->op_default_stripe_offset = -1;
2512         if (S_ISDIR(i1->i_mode)) {
2513                 op_data->op_mea1 = ll_i2info(i1)->lli_lsm_md;
2514                 if (opc == LUSTRE_OPC_MKDIR)
2515                         op_data->op_default_stripe_offset =
2516                                    ll_i2info(i1)->lli_def_stripe_offset;
2517         }
2518
2519         if (i2) {
2520                 op_data->op_fid2 = *ll_inode2fid(i2);
2521                 if (S_ISDIR(i2->i_mode))
2522                         op_data->op_mea2 = ll_i2info(i2)->lli_lsm_md;
2523         } else {
2524                 fid_zero(&op_data->op_fid2);
2525         }
2526
2527         if (ll_i2sbi(i1)->ll_flags & LL_SBI_64BIT_HASH)
2528                 op_data->op_cli_flags |= CLI_HASH64;
2529
2530         if (ll_need_32bit_api(ll_i2sbi(i1)))
2531                 op_data->op_cli_flags |= CLI_API32;
2532
2533         op_data->op_name = name;
2534         op_data->op_namelen = namelen;
2535         op_data->op_mode = mode;
2536         op_data->op_mod_time = ktime_get_real_seconds();
2537         op_data->op_fsuid = from_kuid(&init_user_ns, current_fsuid());
2538         op_data->op_fsgid = from_kgid(&init_user_ns, current_fsgid());
2539         op_data->op_cap = cfs_curproc_cap_pack();
2540         op_data->op_mds = 0;
2541         if ((opc == LUSTRE_OPC_CREATE) && (name != NULL) &&
2542              filename_is_volatile(name, namelen, &op_data->op_mds)) {
2543                 op_data->op_bias |= MDS_CREATE_VOLATILE;
2544         }
2545         op_data->op_data = data;
2546
2547         return op_data;
2548 }
2549
2550 void ll_finish_md_op_data(struct md_op_data *op_data)
2551 {
2552         security_release_secctx(op_data->op_file_secctx,
2553                                 op_data->op_file_secctx_size);
2554         OBD_FREE_PTR(op_data);
2555 }
2556
2557 #ifdef HAVE_SUPEROPS_USE_DENTRY
2558 int ll_show_options(struct seq_file *seq, struct dentry *dentry)
2559 #else
2560 int ll_show_options(struct seq_file *seq, struct vfsmount *vfs)
2561 #endif
2562 {
2563         struct ll_sb_info *sbi;
2564
2565 #ifdef HAVE_SUPEROPS_USE_DENTRY
2566         LASSERT((seq != NULL) && (dentry != NULL));
2567         sbi = ll_s2sbi(dentry->d_sb);
2568 #else
2569         LASSERT((seq != NULL) && (vfs != NULL));
2570         sbi = ll_s2sbi(vfs->mnt_sb);
2571 #endif
2572
2573         if (sbi->ll_flags & LL_SBI_NOLCK)
2574                 seq_puts(seq, ",nolock");
2575
2576         if (sbi->ll_flags & LL_SBI_FLOCK)
2577                 seq_puts(seq, ",flock");
2578
2579         if (sbi->ll_flags & LL_SBI_LOCALFLOCK)
2580                 seq_puts(seq, ",localflock");
2581
2582         if (sbi->ll_flags & LL_SBI_USER_XATTR)
2583                 seq_puts(seq, ",user_xattr");
2584
2585         if (sbi->ll_flags & LL_SBI_LAZYSTATFS)
2586                 seq_puts(seq, ",lazystatfs");
2587
2588         if (sbi->ll_flags & LL_SBI_USER_FID2PATH)
2589                 seq_puts(seq, ",user_fid2path");
2590
2591         if (sbi->ll_flags & LL_SBI_ALWAYS_PING)
2592                 seq_puts(seq, ",always_ping");
2593
2594         RETURN(0);
2595 }
2596
2597 /**
2598  * Get obd name by cmd, and copy out to user space
2599  */
2600 int ll_get_obd_name(struct inode *inode, unsigned int cmd, unsigned long arg)
2601 {
2602         struct ll_sb_info *sbi = ll_i2sbi(inode);
2603         struct obd_device *obd;
2604         ENTRY;
2605
2606         if (cmd == OBD_IOC_GETDTNAME)
2607                 obd = class_exp2obd(sbi->ll_dt_exp);
2608         else if (cmd == OBD_IOC_GETMDNAME)
2609                 obd = class_exp2obd(sbi->ll_md_exp);
2610         else
2611                 RETURN(-EINVAL);
2612
2613         if (!obd)
2614                 RETURN(-ENOENT);
2615
2616         if (copy_to_user((void __user *)arg, obd->obd_name,
2617                          strlen(obd->obd_name) + 1))
2618                 RETURN(-EFAULT);
2619
2620         RETURN(0);
2621 }
2622
2623 /**
2624  * Get lustre file system name by \a sbi. If \a buf is provided(non-NULL), the
2625  * fsname will be returned in this buffer; otherwise, a static buffer will be
2626  * used to store the fsname and returned to caller.
2627  */
2628 char *ll_get_fsname(struct super_block *sb, char *buf, int buflen)
2629 {
2630         static char fsname_static[MTI_NAME_MAXLEN];
2631         struct lustre_sb_info *lsi = s2lsi(sb);
2632         char *ptr;
2633         int len;
2634
2635         if (buf == NULL) {
2636                 /* this means the caller wants to use static buffer
2637                  * and it doesn't care about race. Usually this is
2638                  * in error reporting path */
2639                 buf = fsname_static;
2640                 buflen = sizeof(fsname_static);
2641         }
2642
2643         len = strlen(lsi->lsi_lmd->lmd_profile);
2644         ptr = strrchr(lsi->lsi_lmd->lmd_profile, '-');
2645         if (ptr && (strcmp(ptr, "-client") == 0))
2646                 len -= 7;
2647
2648         if (unlikely(len >= buflen))
2649                 len = buflen - 1;
2650         strncpy(buf, lsi->lsi_lmd->lmd_profile, len);
2651         buf[len] = '\0';
2652
2653         return buf;
2654 }
2655
2656 static char* ll_d_path(struct dentry *dentry, char *buf, int bufsize)
2657 {
2658         char *path = NULL;
2659
2660         struct path p;
2661
2662         p.dentry = dentry;
2663         p.mnt = current->fs->root.mnt;
2664         path_get(&p);
2665         path = d_path(&p, buf, bufsize);
2666         path_put(&p);
2667         return path;
2668 }
2669
2670 void ll_dirty_page_discard_warn(struct page *page, int ioret)
2671 {
2672         char *buf, *path = NULL;
2673         struct dentry *dentry = NULL;
2674         struct inode *inode = page->mapping->host;
2675
2676         /* this can be called inside spin lock so use GFP_ATOMIC. */
2677         buf = (char *)__get_free_page(GFP_ATOMIC);
2678         if (buf != NULL) {
2679                 dentry = d_find_alias(page->mapping->host);
2680                 if (dentry != NULL)
2681                         path = ll_d_path(dentry, buf, PAGE_SIZE);
2682         }
2683
2684         CDEBUG(D_WARNING,
2685                "%s: dirty page discard: %s/fid: "DFID"/%s may get corrupted "
2686                "(rc %d)\n", ll_get_fsname(page->mapping->host->i_sb, NULL, 0),
2687                s2lsi(page->mapping->host->i_sb)->lsi_lmd->lmd_dev,
2688                PFID(ll_inode2fid(inode)),
2689                (path && !IS_ERR(path)) ? path : "", ioret);
2690
2691         if (dentry != NULL)
2692                 dput(dentry);
2693
2694         if (buf != NULL)
2695                 free_page((unsigned long)buf);
2696 }
2697
2698 ssize_t ll_copy_user_md(const struct lov_user_md __user *md,
2699                         struct lov_user_md **kbuf)
2700 {
2701         struct lov_user_md      lum;
2702         ssize_t                 lum_size;
2703         ENTRY;
2704
2705         if (copy_from_user(&lum, md, sizeof(lum)))
2706                 RETURN(-EFAULT);
2707
2708         lum_size = ll_lov_user_md_size(&lum);
2709         if (lum_size < 0)
2710                 RETURN(lum_size);
2711
2712         OBD_ALLOC(*kbuf, lum_size);
2713         if (*kbuf == NULL)
2714                 RETURN(-ENOMEM);
2715
2716         if (copy_from_user(*kbuf, md, lum_size) != 0) {
2717                 OBD_FREE(*kbuf, lum_size);
2718                 RETURN(-EFAULT);
2719         }
2720
2721         RETURN(lum_size);
2722 }
2723
2724 /*
2725  * Compute llite root squash state after a change of root squash
2726  * configuration setting or add/remove of a lnet nid
2727  */
2728 void ll_compute_rootsquash_state(struct ll_sb_info *sbi)
2729 {
2730         struct root_squash_info *squash = &sbi->ll_squash;
2731         int i;
2732         bool matched;
2733         struct lnet_process_id id;
2734
2735         /* Update norootsquash flag */
2736         down_write(&squash->rsi_sem);
2737         if (list_empty(&squash->rsi_nosquash_nids))
2738                 sbi->ll_flags &= ~LL_SBI_NOROOTSQUASH;
2739         else {
2740                 /* Do not apply root squash as soon as one of our NIDs is
2741                  * in the nosquash_nids list */
2742                 matched = false;
2743                 i = 0;
2744                 while (LNetGetId(i++, &id) != -ENOENT) {
2745                         if (LNET_NETTYP(LNET_NIDNET(id.nid)) == LOLND)
2746                                 continue;
2747                         if (cfs_match_nid(id.nid, &squash->rsi_nosquash_nids)) {
2748                                 matched = true;
2749                                 break;
2750                         }
2751                 }
2752                 if (matched)
2753                         sbi->ll_flags |= LL_SBI_NOROOTSQUASH;
2754                 else
2755                         sbi->ll_flags &= ~LL_SBI_NOROOTSQUASH;
2756         }
2757         up_write(&squash->rsi_sem);
2758 }
2759
2760 /**
2761  * Parse linkea content to extract information about a given hardlink
2762  *
2763  * \param[in]   ldata      - Initialized linkea data
2764  * \param[in]   linkno     - Link identifier
2765  * \param[out]  parent_fid - The entry's parent FID
2766  * \param[out]  ln         - Entry name destination buffer
2767  *
2768  * \retval 0 on success
2769  * \retval Appropriate negative error code on failure
2770  */
2771 static int ll_linkea_decode(struct linkea_data *ldata, unsigned int linkno,
2772                             struct lu_fid *parent_fid, struct lu_name *ln)
2773 {
2774         unsigned int    idx;
2775         int             rc;
2776         ENTRY;
2777
2778         rc = linkea_init_with_rec(ldata);
2779         if (rc < 0)
2780                 RETURN(rc);
2781
2782         if (linkno >= ldata->ld_leh->leh_reccount)
2783                 /* beyond last link */
2784                 RETURN(-ENODATA);
2785
2786         linkea_first_entry(ldata);
2787         for (idx = 0; ldata->ld_lee != NULL; idx++) {
2788                 linkea_entry_unpack(ldata->ld_lee, &ldata->ld_reclen, ln,
2789                                     parent_fid);
2790                 if (idx == linkno)
2791                         break;
2792
2793                 linkea_next_entry(ldata);
2794         }
2795
2796         if (idx < linkno)
2797                 RETURN(-ENODATA);
2798
2799         RETURN(0);
2800 }
2801
2802 /**
2803  * Get parent FID and name of an identified link. Operation is performed for
2804  * a given link number, letting the caller iterate over linkno to list one or
2805  * all links of an entry.
2806  *
2807  * \param[in]     file - File descriptor against which to perform the operation
2808  * \param[in,out] arg  - User-filled structure containing the linkno to operate
2809  *                       on and the available size. It is eventually filled with
2810  *                       the requested information or left untouched on error
2811  *
2812  * \retval - 0 on success
2813  * \retval - Appropriate negative error code on failure
2814  */
2815 int ll_getparent(struct file *file, struct getparent __user *arg)
2816 {
2817         struct inode            *inode = file_inode(file);
2818         struct linkea_data      *ldata;
2819         struct lu_buf            buf = LU_BUF_NULL;
2820         struct lu_name           ln;
2821         struct lu_fid            parent_fid;
2822         __u32                    linkno;
2823         __u32                    name_size;
2824         int                      rc;
2825
2826         ENTRY;
2827
2828         if (!cfs_capable(CFS_CAP_DAC_READ_SEARCH) &&
2829             !(ll_i2sbi(inode)->ll_flags & LL_SBI_USER_FID2PATH))
2830                 RETURN(-EPERM);
2831
2832         if (get_user(name_size, &arg->gp_name_size))
2833                 RETURN(-EFAULT);
2834
2835         if (get_user(linkno, &arg->gp_linkno))
2836                 RETURN(-EFAULT);
2837
2838         if (name_size > PATH_MAX)
2839                 RETURN(-EINVAL);
2840
2841         OBD_ALLOC(ldata, sizeof(*ldata));
2842         if (ldata == NULL)
2843                 RETURN(-ENOMEM);
2844
2845         rc = linkea_data_new(ldata, &buf);
2846         if (rc < 0)
2847                 GOTO(ldata_free, rc);
2848
2849 #ifdef HAVE_XATTR_HANDLER_FLAGS
2850         rc = ll_xattr_list(inode, XATTR_NAME_LINK, XATTR_TRUSTED_T, buf.lb_buf,
2851                            buf.lb_len, OBD_MD_FLXATTR);
2852 #else
2853         rc = ll_getxattr(file_dentry(file), XATTR_NAME_LINK, buf.lb_buf,
2854                          buf.lb_len);
2855 #endif /* HAVE_XATTR_HANDLER_FLAGS */
2856         if (rc < 0)
2857                 GOTO(lb_free, rc);
2858
2859         rc = ll_linkea_decode(ldata, linkno, &parent_fid, &ln);
2860         if (rc < 0)
2861                 GOTO(lb_free, rc);
2862
2863         if (ln.ln_namelen >= name_size)
2864                 GOTO(lb_free, rc = -EOVERFLOW);
2865
2866         if (copy_to_user(&arg->gp_fid, &parent_fid, sizeof(arg->gp_fid)))
2867                 GOTO(lb_free, rc = -EFAULT);
2868
2869         if (copy_to_user(&arg->gp_name, ln.ln_name, ln.ln_namelen))
2870                 GOTO(lb_free, rc = -EFAULT);
2871
2872         if (put_user('\0', arg->gp_name + ln.ln_namelen))
2873                 GOTO(lb_free, rc = -EFAULT);
2874
2875 lb_free:
2876         lu_buf_free(&buf);
2877 ldata_free:
2878         OBD_FREE(ldata, sizeof(*ldata));
2879
2880         RETURN(rc);
2881 }