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 #ifdef ENABLE_FLOCK
 113         sbi->ll_flags |= LL_SBI_FLOCK;
 114 #endif
 115
 116 #ifdef HAVE_LRU_RESIZE_SUPPORT
 117         sbi->ll_flags |= LL_SBI_LRU_RESIZE;
 118 #endif
 119         sbi->ll_flags |= LL_SBI_LAZYSTATFS;
 120
 121         for (i = 0; i <= LL_PROCESS_HIST_MAX; i++) {
 122                 spin_lock_init(&sbi->ll_rw_extents_info.pp_extents[i].
 123                                pp_r_hist.oh_lock);
 124                 spin_lock_init(&sbi->ll_rw_extents_info.pp_extents[i].
 125                                pp_w_hist.oh_lock);
 126         }
 127
 128         /* metadata statahead is enabled by default */
 129         sbi->ll_sa_running_max = LL_SA_RUNNING_DEF;
 130         sbi->ll_sa_max = LL_SA_RPC_DEF;
 131         atomic_set(&sbi->ll_sa_total, 0);
 132         atomic_set(&sbi->ll_sa_wrong, 0);
 133         atomic_set(&sbi->ll_sa_running, 0);
 134         atomic_set(&sbi->ll_agl_total, 0);
 135         sbi->ll_flags |= LL_SBI_AGL_ENABLED;
 136         sbi->ll_flags |= LL_SBI_FAST_READ;
 137         sbi->ll_flags |= LL_SBI_TINY_WRITE;
 138
 139         /* root squash */
 140         sbi->ll_squash.rsi_uid = 0;
 141         sbi->ll_squash.rsi_gid = 0;
 142         INIT_LIST_HEAD(&sbi->ll_squash.rsi_nosquash_nids);
 143         init_rwsem(&sbi->ll_squash.rsi_sem);
 144
 145         RETURN(sbi);
 146 }
 147
 148 static void ll_free_sbi(struct super_block *sb)
 149 {
 150         struct ll_sb_info *sbi = ll_s2sbi(sb);
 151         ENTRY;
 152
 153         if (sbi != NULL) {
 154                 if (!list_empty(&sbi->ll_squash.rsi_nosquash_nids))
 155                         cfs_free_nidlist(&sbi->ll_squash.rsi_nosquash_nids);
 156                 if (sbi->ll_cache != NULL) {
 157                         cl_cache_decref(sbi->ll_cache);
 158                         sbi->ll_cache = NULL;
 159                 }
 160                 OBD_FREE(sbi, sizeof(*sbi));
 161         }
 162         EXIT;
 163 }
 164
 165 static int client_common_fill_super(struct super_block *sb, char *md, char *dt,
 166                                     struct vfsmount *mnt)
 167 {
 168         struct inode *root = NULL;
 169         struct ll_sb_info *sbi = ll_s2sbi(sb);
 170         struct obd_statfs *osfs = NULL;
 171         struct ptlrpc_request *request = NULL;
 172         struct obd_connect_data *data = NULL;
 173         struct obd_uuid *uuid;
 174         struct md_op_data *op_data;
 175         struct lustre_md lmd;
 176         u64 valid;
 177         int size, err, checksum;
 178
 179         ENTRY;
 180         sbi->ll_md_obd = class_name2obd(md);
 181         if (!sbi->ll_md_obd) {
 182                 CERROR("MD %s: not setup or attached\n", md);
 183                 RETURN(-EINVAL);
 184         }
 185
 186         OBD_ALLOC_PTR(data);
 187         if (data == NULL)
 188                 RETURN(-ENOMEM);
 189
 190         OBD_ALLOC_PTR(osfs);
 191         if (osfs == NULL) {
 192                 OBD_FREE_PTR(data);
 193                 RETURN(-ENOMEM);
 194         }
 195
 196         /* pass client page size via ocd_grant_blkbits, the server should report
 197          * back its backend blocksize for grant calculation purpose */
 198         data->ocd_grant_blkbits = PAGE_SHIFT;
 199
 200         /* indicate MDT features supported by this client */
 201         data->ocd_connect_flags = OBD_CONNECT_IBITS    | OBD_CONNECT_NODEVOH  |
 202                                   OBD_CONNECT_ATTRFID  | OBD_CONNECT_GRANT |
 203                                   OBD_CONNECT_VERSION  | OBD_CONNECT_BRW_SIZE |
 204                                   OBD_CONNECT_SRVLOCK  | OBD_CONNECT_TRUNCLOCK|
 205                                   OBD_CONNECT_MDS_CAPA | OBD_CONNECT_OSS_CAPA |
 206                                   OBD_CONNECT_CANCELSET | OBD_CONNECT_FID     |
 207                                   OBD_CONNECT_AT       | OBD_CONNECT_LOV_V3   |
 208                                   OBD_CONNECT_VBR | OBD_CONNECT_FULL20 |
 209                                   OBD_CONNECT_64BITHASH |
 210                                   OBD_CONNECT_EINPROGRESS |
 211                                   OBD_CONNECT_JOBSTATS | OBD_CONNECT_LVB_TYPE |
 212                                   OBD_CONNECT_LAYOUTLOCK | OBD_CONNECT_PINGLESS|
 213                                   OBD_CONNECT_MAX_EASIZE |
 214                                   OBD_CONNECT_FLOCK_DEAD |
 215                                   OBD_CONNECT_DISP_STRIPE | OBD_CONNECT_LFSCK |
 216                                   OBD_CONNECT_OPEN_BY_FID |
 217                                   OBD_CONNECT_DIR_STRIPE |
 218                                   OBD_CONNECT_BULK_MBITS | OBD_CONNECT_CKSUM |
 219                                   OBD_CONNECT_SUBTREE |
 220                                   OBD_CONNECT_MULTIMODRPCS |
 221                                   OBD_CONNECT_GRANT_PARAM |
 222                                   OBD_CONNECT_SHORTIO | OBD_CONNECT_FLAGS2;
 223
 224         data->ocd_connect_flags2 = OBD_CONNECT2_FLR |
 225                                    OBD_CONNECT2_LOCK_CONVERT |
 226                                    OBD_CONNECT2_DIR_MIGRATE |
 227                                    OBD_CONNECT2_SUM_STATFS |
 228                                    OBD_CONNECT2_ARCHIVE_ID_ARRAY;
 229
 230 #ifdef HAVE_LRU_RESIZE_SUPPORT
 231         if (sbi->ll_flags & LL_SBI_LRU_RESIZE)
 232                 data->ocd_connect_flags |= OBD_CONNECT_LRU_RESIZE;
 233 #endif
 234 #ifdef CONFIG_FS_POSIX_ACL
 235         data->ocd_connect_flags |= OBD_CONNECT_ACL | OBD_CONNECT_UMASK |
 236                                    OBD_CONNECT_LARGE_ACL;
 237 #endif
 238
 239         data->ocd_cksum_types = obd_cksum_types_supported_client();
 240
 241         if (OBD_FAIL_CHECK(OBD_FAIL_MDC_LIGHTWEIGHT))
 242                 /* flag mdc connection as lightweight, only used for test
 243                  * purpose, use with care */
 244                 data->ocd_connect_flags |= OBD_CONNECT_LIGHTWEIGHT;
 245
 246         data->ocd_ibits_known = MDS_INODELOCK_FULL;
 247         data->ocd_version = LUSTRE_VERSION_CODE;
 248
 249         if (sb->s_flags & MS_RDONLY)
 250                 data->ocd_connect_flags |= OBD_CONNECT_RDONLY;
 251         if (sbi->ll_flags & LL_SBI_USER_XATTR)
 252                 data->ocd_connect_flags |= OBD_CONNECT_XATTR;
 253
 254 #ifdef MS_NOSEC
 255         /* Setting this indicates we correctly support S_NOSEC (See kernel
 256          * commit 9e1f1de02c2275d7172e18dc4e7c2065777611bf)
 257          */
 258         sb->s_flags |= MS_NOSEC;
 259 #endif
 260
 261         if (sbi->ll_flags & LL_SBI_FLOCK)
 262                 sbi->ll_fop = &ll_file_operations_flock;
 263         else if (sbi->ll_flags & LL_SBI_LOCALFLOCK)
 264                 sbi->ll_fop = &ll_file_operations;
 265         else
 266                 sbi->ll_fop = &ll_file_operations_noflock;
 267
 268         /* always ping even if server suppress_pings */
 269         if (sbi->ll_flags & LL_SBI_ALWAYS_PING)
 270                 data->ocd_connect_flags &= ~OBD_CONNECT_PINGLESS;
 271
 272         obd_connect_set_secctx(data);
 273
 274 #if defined(CONFIG_SECURITY)
 275         data->ocd_connect_flags2 |= OBD_CONNECT2_SELINUX_POLICY;
 276 #endif
 277
 278         data->ocd_brw_size = MD_MAX_BRW_SIZE;
 279
 280         err = obd_connect(NULL, &sbi->ll_md_exp, sbi->ll_md_obd,
 281                           &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         sbi->ll_dt_obd = class_name2obd(dt);
 395         if (!sbi->ll_dt_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 /sys. As a consequence, we still need to come to an
 443          * agreement on the supported algorithms at connect time
 444          */
 445         data->ocd_connect_flags |= OBD_CONNECT_CKSUM;
 446
 447         if (OBD_FAIL_CHECK(OBD_FAIL_OSC_CKSUM_ADLER_ONLY))
 448                 data->ocd_cksum_types = OBD_CKSUM_ADLER;
 449         else
 450                 data->ocd_cksum_types = obd_cksum_types_supported_client();
 451
 452 #ifdef HAVE_LRU_RESIZE_SUPPORT
 453         data->ocd_connect_flags |= OBD_CONNECT_LRU_RESIZE;
 454 #endif
 455         /* always ping even if server suppress_pings */
 456         if (sbi->ll_flags & LL_SBI_ALWAYS_PING)
 457                 data->ocd_connect_flags &= ~OBD_CONNECT_PINGLESS;
 458
 459         CDEBUG(D_RPCTRACE, "ocd_connect_flags: %#llx ocd_version: %d "
 460                "ocd_grant: %d\n", data->ocd_connect_flags,
 461                data->ocd_version, data->ocd_grant);
 462
 463         sbi->ll_dt_obd->obd_upcall.onu_owner = &sbi->ll_lco;
 464         sbi->ll_dt_obd->obd_upcall.onu_upcall = cl_ocd_update;
 465
 466         data->ocd_brw_size = DT_MAX_BRW_SIZE;
 467
 468         err = obd_connect(NULL, &sbi->ll_dt_exp, sbi->ll_dt_obd,
 469                           &sbi->ll_sb_uuid, data, NULL);
 470         if (err == -EBUSY) {
 471                 LCONSOLE_ERROR_MSG(0x150, "An OST (dt %s) is performing "
 472                                    "recovery, of which this client is not a "
 473                                    "part.  Please wait for recovery to "
 474                                    "complete, abort, or time out.\n", dt);
 475                 GOTO(out_md, err);
 476         } else if (err) {
 477                 CERROR("%s: Cannot connect to %s: rc = %d\n",
 478                        sbi->ll_dt_exp->exp_obd->obd_name, dt, err);
 479                 GOTO(out_md, err);
 480         }
 481
 482         sbi->ll_dt_exp->exp_connect_data = *data;
 483
 484         /* Don't change value if it was specified in the config log */
 485         if (sbi->ll_ra_info.ra_max_read_ahead_whole_pages == -1)
 486                 sbi->ll_ra_info.ra_max_read_ahead_whole_pages =
 487                         max_t(unsigned long, SBI_DEFAULT_READAHEAD_WHOLE_MAX,
 488                               (data->ocd_brw_size >> PAGE_SHIFT));
 489
 490         err = obd_fid_init(sbi->ll_dt_exp->exp_obd, sbi->ll_dt_exp,
 491                            LUSTRE_SEQ_METADATA);
 492         if (err) {
 493                 CERROR("%s: Can't init data layer FID infrastructure, "
 494                        "rc = %d\n", sbi->ll_dt_exp->exp_obd->obd_name, err);
 495                 GOTO(out_dt, err);
 496         }
 497
 498         mutex_lock(&sbi->ll_lco.lco_lock);
 499         sbi->ll_lco.lco_flags = data->ocd_connect_flags;
 500         sbi->ll_lco.lco_md_exp = sbi->ll_md_exp;
 501         sbi->ll_lco.lco_dt_exp = sbi->ll_dt_exp;
 502         mutex_unlock(&sbi->ll_lco.lco_lock);
 503
 504         fid_zero(&sbi->ll_root_fid);
 505         err = md_get_root(sbi->ll_md_exp, get_mount_fileset(sb),
 506                            &sbi->ll_root_fid);
 507         if (err) {
 508                 CERROR("cannot mds_connect: rc = %d\n", err);
 509                 GOTO(out_lock_cn_cb, err);
 510         }
 511         if (!fid_is_sane(&sbi->ll_root_fid)) {
 512                 CERROR("%s: Invalid root fid "DFID" during mount\n",
 513                        sbi->ll_md_exp->exp_obd->obd_name,
 514                        PFID(&sbi->ll_root_fid));
 515                 GOTO(out_lock_cn_cb, err = -EINVAL);
 516         }
 517         CDEBUG(D_SUPER, "rootfid "DFID"\n", PFID(&sbi->ll_root_fid));
 518
 519         sb->s_op = &lustre_super_operations;
 520 #ifdef HAVE_XATTR_HANDLER_FLAGS
 521         sb->s_xattr = ll_xattr_handlers;
 522 #endif
 523 #if THREAD_SIZE >= 8192 /*b=17630*/
 524         sb->s_export_op = &lustre_export_operations;
 525 #endif
 526
 527         /* make root inode
 528          * XXX: move this to after cbd setup? */
 529         valid = OBD_MD_FLGETATTR | OBD_MD_FLBLOCKS | OBD_MD_FLMODEASIZE;
 530         if (sbi->ll_flags & LL_SBI_ACL)
 531                 valid |= OBD_MD_FLACL;
 532
 533         OBD_ALLOC_PTR(op_data);
 534         if (op_data == NULL)
 535                 GOTO(out_lock_cn_cb, err = -ENOMEM);
 536
 537         op_data->op_fid1 = sbi->ll_root_fid;
 538         op_data->op_mode = 0;
 539         op_data->op_valid = valid;
 540
 541         err = md_getattr(sbi->ll_md_exp, op_data, &request);
 542
 543         OBD_FREE_PTR(op_data);
 544         if (err) {
 545                 CERROR("%s: md_getattr failed for root: rc = %d\n",
 546                        sbi->ll_md_exp->exp_obd->obd_name, err);
 547                 GOTO(out_lock_cn_cb, err);
 548         }
 549
 550         err = md_get_lustre_md(sbi->ll_md_exp, request, sbi->ll_dt_exp,
 551                                sbi->ll_md_exp, &lmd);
 552         if (err) {
 553                 CERROR("failed to understand root inode md: rc = %d\n", err);
 554                 ptlrpc_req_finished(request);
 555                 GOTO(out_lock_cn_cb, err);
 556         }
 557
 558         LASSERT(fid_is_sane(&sbi->ll_root_fid));
 559         root = ll_iget(sb, cl_fid_build_ino(&sbi->ll_root_fid,
 560                                             sbi->ll_flags & LL_SBI_32BIT_API),
 561                        &lmd);
 562         md_free_lustre_md(sbi->ll_md_exp, &lmd);
 563         ptlrpc_req_finished(request);
 564
 565         if (IS_ERR(root)) {
 566 #ifdef CONFIG_FS_POSIX_ACL
 567                 if (lmd.posix_acl) {
 568                         posix_acl_release(lmd.posix_acl);
 569                         lmd.posix_acl = NULL;
 570                 }
 571 #endif
 572                 err = IS_ERR(root) ? PTR_ERR(root) : -EBADF;
 573                 root = NULL;
 574                 CERROR("lustre_lite: bad iget4 for root\n");
 575                 GOTO(out_root, err);
 576         }
 577
 578         checksum = sbi->ll_flags & LL_SBI_CHECKSUM;
 579         if (sbi->ll_checksum_set) {
 580                 err = obd_set_info_async(NULL, sbi->ll_dt_exp,
 581                                          sizeof(KEY_CHECKSUM), KEY_CHECKSUM,
 582                                          sizeof(checksum), &checksum, NULL);
 583                 if (err) {
 584                         CERROR("%s: Set checksum failed: rc = %d\n",
 585                                sbi->ll_dt_exp->exp_obd->obd_name, err);
 586                         GOTO(out_root, err);
 587                 }
 588         }
 589         cl_sb_init(sb);
 590
 591         err = obd_set_info_async(NULL, sbi->ll_dt_exp, sizeof(KEY_CACHE_SET),
 592                                  KEY_CACHE_SET, sizeof(*sbi->ll_cache),
 593                                  sbi->ll_cache, NULL);
 594         if (err) {
 595                 CERROR("%s: Set cache_set failed: rc = %d\n",
 596                        sbi->ll_dt_exp->exp_obd->obd_name, err);
 597                 GOTO(out_root, err);
 598         }
 599
 600         sb->s_root = d_make_root(root);
 601         if (sb->s_root == NULL) {
 602                 CERROR("%s: can't make root dentry\n",
 603                         ll_get_fsname(sb, NULL, 0));
 604                 GOTO(out_root, err = -ENOMEM);
 605         }
 606 #ifdef HAVE_DCACHE_LOCK
 607         sb->s_root->d_op = &ll_d_ops;
 608 #endif
 609
 610         sbi->ll_sdev_orig = sb->s_dev;
 611
 612         /* We set sb->s_dev equal on all lustre clients in order to support
 613          * NFS export clustering.  NFSD requires that the FSID be the same
 614          * on all clients. */
 615         /* s_dev is also used in lt_compare() to compare two fs, but that is
 616          * only a node-local comparison. */
 617         uuid = obd_get_uuid(sbi->ll_md_exp);
 618         if (uuid != NULL)
 619                 sb->s_dev = get_uuid2int(uuid->uuid, strlen(uuid->uuid));
 620
 621         if (data != NULL)
 622                 OBD_FREE_PTR(data);
 623         if (osfs != NULL)
 624                 OBD_FREE_PTR(osfs);
 625
 626         if (sbi->ll_dt_obd) {
 627                 err = sysfs_create_link(&sbi->ll_kset.kobj,
 628                                         &sbi->ll_dt_obd->obd_kset.kobj,
 629                                         sbi->ll_dt_obd->obd_type->typ_name);
 630                 if (err < 0) {
 631                         CERROR("%s: could not register %s in llite: rc = %d\n",
 632                                dt, ll_get_fsname(sb, NULL, 0), err);
 633                         err = 0;
 634                 }
 635         }
 636
 637         if (sbi->ll_md_obd) {
 638                 err = sysfs_create_link(&sbi->ll_kset.kobj,
 639                                         &sbi->ll_md_obd->obd_kset.kobj,
 640                                         sbi->ll_md_obd->obd_type->typ_name);
 641                 if (err < 0) {
 642                         CERROR("%s: could not register %s in llite: rc = %d\n",
 643                                md, ll_get_fsname(sb, NULL, 0), err);
 644                         err = 0;
 645                 }
 646         }
 647
 648         RETURN(err);
 649 out_root:
 650         if (root)
 651                 iput(root);
 652 out_lock_cn_cb:
 653         obd_fid_fini(sbi->ll_dt_exp->exp_obd);
 654 out_dt:
 655         obd_disconnect(sbi->ll_dt_exp);
 656         sbi->ll_dt_exp = NULL;
 657         sbi->ll_dt_obd = NULL;
 658 out_md_fid:
 659         obd_fid_fini(sbi->ll_md_exp->exp_obd);
 660 out_md:
 661         obd_disconnect(sbi->ll_md_exp);
 662         sbi->ll_md_exp = NULL;
 663         sbi->ll_md_obd = NULL;
 664 out:
 665         if (data != NULL)
 666                 OBD_FREE_PTR(data);
 667         if (osfs != NULL)
 668                 OBD_FREE_PTR(osfs);
 669         return err;
 670 }
 671
 672 int ll_get_max_mdsize(struct ll_sb_info *sbi, int *lmmsize)
 673 {
 674         int size, rc;
 675
 676         size = sizeof(*lmmsize);
 677         rc = obd_get_info(NULL, sbi->ll_dt_exp, sizeof(KEY_MAX_EASIZE),
 678                           KEY_MAX_EASIZE, &size, lmmsize);
 679         if (rc != 0) {
 680                 CERROR("%s: cannot get max LOV EA size: rc = %d\n",
 681                        sbi->ll_dt_exp->exp_obd->obd_name, rc);
 682                 RETURN(rc);
 683         }
 684
 685         size = sizeof(int);
 686         rc = obd_get_info(NULL, sbi->ll_md_exp, sizeof(KEY_MAX_EASIZE),
 687                           KEY_MAX_EASIZE, &size, lmmsize);
 688         if (rc)
 689                 CERROR("Get max mdsize error rc %d\n", rc);
 690
 691         RETURN(rc);
 692 }
 693
 694 /**
 695  * Get the value of the default_easize parameter.
 696  *
 697  * \see client_obd::cl_default_mds_easize
 698  *
 699  * \param[in] sbi       superblock info for this filesystem
 700  * \param[out] lmmsize  pointer to storage location for value
 701  *
 702  * \retval 0            on success
 703  * \retval negative     negated errno on failure
 704  */
 705 int ll_get_default_mdsize(struct ll_sb_info *sbi, int *lmmsize)
 706 {
 707         int size, rc;
 708
 709         size = sizeof(int);
 710         rc = obd_get_info(NULL, sbi->ll_md_exp, sizeof(KEY_DEFAULT_EASIZE),
 711                          KEY_DEFAULT_EASIZE, &size, lmmsize);
 712         if (rc)
 713                 CERROR("Get default mdsize error rc %d\n", rc);
 714
 715         RETURN(rc);
 716 }
 717
 718 /**
 719  * Set the default_easize parameter to the given value.
 720  *
 721  * \see client_obd::cl_default_mds_easize
 722  *
 723  * \param[in] sbi       superblock info for this filesystem
 724  * \param[in] lmmsize   the size to set
 725  *
 726  * \retval 0            on success
 727  * \retval negative     negated errno on failure
 728  */
 729 int ll_set_default_mdsize(struct ll_sb_info *sbi, int lmmsize)
 730 {
 731         int rc;
 732
 733         if (lmmsize < sizeof(struct lov_mds_md) ||
 734             lmmsize > OBD_MAX_DEFAULT_EA_SIZE)
 735                 return -EINVAL;
 736
 737         rc = obd_set_info_async(NULL, sbi->ll_md_exp,
 738                                 sizeof(KEY_DEFAULT_EASIZE), KEY_DEFAULT_EASIZE,
 739                                 sizeof(int), &lmmsize, NULL);
 740
 741         RETURN(rc);
 742 }
 743
 744 static void client_common_put_super(struct super_block *sb)
 745 {
 746         struct ll_sb_info *sbi = ll_s2sbi(sb);
 747         ENTRY;
 748
 749         cl_sb_fini(sb);
 750
 751         obd_fid_fini(sbi->ll_dt_exp->exp_obd);
 752         obd_disconnect(sbi->ll_dt_exp);
 753         sbi->ll_dt_exp = NULL;
 754
 755         ll_debugfs_unregister_super(sb);
 756
 757         obd_fid_fini(sbi->ll_md_exp->exp_obd);
 758         obd_disconnect(sbi->ll_md_exp);
 759         sbi->ll_md_exp = NULL;
 760
 761         EXIT;
 762 }
 763
 764 void ll_kill_super(struct super_block *sb)
 765 {
 766         struct ll_sb_info *sbi;
 767         ENTRY;
 768
 769         /* not init sb ?*/
 770         if (!(sb->s_flags & MS_ACTIVE))
 771                 return;
 772
 773         sbi = ll_s2sbi(sb);
 774         /* we need restore s_dev from changed for clustred NFS before put_super
 775          * because new kernels have cached s_dev and change sb->s_dev in
 776          * put_super not affected real removing devices */
 777         if (sbi) {
 778                 sb->s_dev = sbi->ll_sdev_orig;
 779                 sbi->ll_umounting = 1;
 780
 781                 /* wait running statahead threads to quit */
 782                 while (atomic_read(&sbi->ll_sa_running) > 0) {
 783                         set_current_state(TASK_UNINTERRUPTIBLE);
 784                         schedule_timeout(msecs_to_jiffies(MSEC_PER_SEC >> 3));
 785                 }
 786         }
 787
 788         EXIT;
 789 }
 790
 791 static inline int ll_set_opt(const char *opt, char *data, int fl)
 792 {
 793         if (strncmp(opt, data, strlen(opt)) != 0)
 794                 return 0;
 795         else
 796                 return fl;
 797 }
 798
 799 /* non-client-specific mount options are parsed in lmd_parse */
 800 static int ll_options(char *options, struct ll_sb_info *sbi)
 801 {
 802         int tmp;
 803         char *s1 = options, *s2;
 804         int *flags = &sbi->ll_flags;
 805         ENTRY;
 806
 807         if (!options)
 808                 RETURN(0);
 809
 810         CDEBUG(D_CONFIG, "Parsing opts %s\n", options);
 811
 812         while (*s1) {
 813                 CDEBUG(D_SUPER, "next opt=%s\n", s1);
 814                 tmp = ll_set_opt("nolock", s1, LL_SBI_NOLCK);
 815                 if (tmp) {
 816                         *flags |= tmp;
 817                         goto next;
 818                 }
 819                 tmp = ll_set_opt("flock", s1, LL_SBI_FLOCK);
 820                 if (tmp) {
 821                         *flags |= tmp;
 822                         goto next;
 823                 }
 824                 tmp = ll_set_opt("localflock", s1, LL_SBI_LOCALFLOCK);
 825                 if (tmp) {
 826                         *flags |= tmp;
 827                         goto next;
 828                 }
 829                 tmp = ll_set_opt("noflock", s1, LL_SBI_FLOCK|LL_SBI_LOCALFLOCK);
 830                 if (tmp) {
 831                         *flags &= ~tmp;
 832                         goto next;
 833                 }
 834                 tmp = ll_set_opt("user_xattr", s1, LL_SBI_USER_XATTR);
 835                 if (tmp) {
 836                         *flags |= tmp;
 837                         goto next;
 838                 }
 839                 tmp = ll_set_opt("nouser_xattr", s1, LL_SBI_USER_XATTR);
 840                 if (tmp) {
 841                         *flags &= ~tmp;
 842                         goto next;
 843                 }
 844                 tmp = ll_set_opt("context", s1, 1);
 845                 if (tmp)
 846                         goto next;
 847                 tmp = ll_set_opt("fscontext", s1, 1);
 848                 if (tmp)
 849                         goto next;
 850                 tmp = ll_set_opt("defcontext", s1, 1);
 851                 if (tmp)
 852                         goto next;
 853                 tmp = ll_set_opt("rootcontext", s1, 1);
 854                 if (tmp)
 855                         goto next;
 856                 tmp = ll_set_opt("user_fid2path", s1, LL_SBI_USER_FID2PATH);
 857                 if (tmp) {
 858                         *flags |= tmp;
 859                         goto next;
 860                 }
 861                 tmp = ll_set_opt("nouser_fid2path", s1, LL_SBI_USER_FID2PATH);
 862                 if (tmp) {
 863                         *flags &= ~tmp;
 864                         goto next;
 865                 }
 866
 867                 tmp = ll_set_opt("checksum", s1, LL_SBI_CHECKSUM);
 868                 if (tmp) {
 869                         *flags |= tmp;
 870                         sbi->ll_checksum_set = 1;
 871                         goto next;
 872                 }
 873                 tmp = ll_set_opt("nochecksum", s1, LL_SBI_CHECKSUM);
 874                 if (tmp) {
 875                         *flags &= ~tmp;
 876                         sbi->ll_checksum_set = 1;
 877                         goto next;
 878                 }
 879                 tmp = ll_set_opt("lruresize", s1, LL_SBI_LRU_RESIZE);
 880                 if (tmp) {
 881                         *flags |= tmp;
 882                         goto next;
 883                 }
 884                 tmp = ll_set_opt("nolruresize", s1, LL_SBI_LRU_RESIZE);
 885                 if (tmp) {
 886                         *flags &= ~tmp;
 887                         goto next;
 888                 }
 889                 tmp = ll_set_opt("lazystatfs", s1, LL_SBI_LAZYSTATFS);
 890                 if (tmp) {
 891                         *flags |= tmp;
 892                         goto next;
 893                 }
 894                 tmp = ll_set_opt("nolazystatfs", s1, LL_SBI_LAZYSTATFS);
 895                 if (tmp) {
 896                         *flags &= ~tmp;
 897                         goto next;
 898                 }
 899                 tmp = ll_set_opt("32bitapi", s1, LL_SBI_32BIT_API);
 900                 if (tmp) {
 901                         *flags |= tmp;
 902                         goto next;
 903                 }
 904                 tmp = ll_set_opt("verbose", s1, LL_SBI_VERBOSE);
 905                 if (tmp) {
 906                         *flags |= tmp;
 907                         goto next;
 908                 }
 909                 tmp = ll_set_opt("noverbose", s1, LL_SBI_VERBOSE);
 910                 if (tmp) {
 911                         *flags &= ~tmp;
 912                         goto next;
 913                 }
 914                 tmp = ll_set_opt("always_ping", s1, LL_SBI_ALWAYS_PING);
 915                 if (tmp) {
 916                         *flags |= tmp;
 917                         goto next;
 918                 }
 919                 LCONSOLE_ERROR_MSG(0x152, "Unknown option '%s', won't mount.\n",
 920                                    s1);
 921                 RETURN(-EINVAL);
 922
 923 next:
 924                 /* Find next opt */
 925                 s2 = strchr(s1, ',');
 926                 if (s2 == NULL)
 927                         break;
 928                 s1 = s2 + 1;
 929         }
 930         RETURN(0);
 931 }
 932
 933 void ll_lli_init(struct ll_inode_info *lli)
 934 {
 935         lli->lli_inode_magic = LLI_INODE_MAGIC;
 936         lli->lli_flags = 0;
 937         spin_lock_init(&lli->lli_lock);
 938         lli->lli_posix_acl = NULL;
 939         /* Do not set lli_fid, it has been initialized already. */
 940         fid_zero(&lli->lli_pfid);
 941         lli->lli_mds_read_och = NULL;
 942         lli->lli_mds_write_och = NULL;
 943         lli->lli_mds_exec_och = NULL;
 944         lli->lli_open_fd_read_count = 0;
 945         lli->lli_open_fd_write_count = 0;
 946         lli->lli_open_fd_exec_count = 0;
 947         mutex_init(&lli->lli_och_mutex);
 948         spin_lock_init(&lli->lli_agl_lock);
 949         spin_lock_init(&lli->lli_layout_lock);
 950         ll_layout_version_set(lli, CL_LAYOUT_GEN_NONE);
 951         lli->lli_clob = NULL;
 952
 953         init_rwsem(&lli->lli_xattrs_list_rwsem);
 954         mutex_init(&lli->lli_xattrs_enq_lock);
 955
 956         LASSERT(lli->lli_vfs_inode.i_mode != 0);
 957         if (S_ISDIR(lli->lli_vfs_inode.i_mode)) {
 958                 mutex_init(&lli->lli_readdir_mutex);
 959                 lli->lli_opendir_key = NULL;
 960                 lli->lli_sai = NULL;
 961                 spin_lock_init(&lli->lli_sa_lock);
 962                 lli->lli_opendir_pid = 0;
 963                 lli->lli_sa_enabled = 0;
 964                 lli->lli_def_stripe_offset = -1;
 965                 init_rwsem(&lli->lli_lsm_sem);
 966         } else {
 967                 mutex_init(&lli->lli_size_mutex);
 968                 lli->lli_symlink_name = NULL;
 969                 init_rwsem(&lli->lli_trunc_sem);
 970                 range_lock_tree_init(&lli->lli_write_tree);
 971                 init_rwsem(&lli->lli_glimpse_sem);
 972                 lli->lli_glimpse_time = ktime_set(0, 0);
 973                 INIT_LIST_HEAD(&lli->lli_agl_list);
 974                 lli->lli_agl_index = 0;
 975                 lli->lli_async_rc = 0;
 976         }
 977         mutex_init(&lli->lli_layout_mutex);
 978         memset(lli->lli_jobid, 0, sizeof(lli->lli_jobid));
 979 }
 980
 981 #define MAX_STRING_SIZE 128
 982
 983 #ifndef HAVE_SUPER_SETUP_BDI_NAME
 984
 985 #define LSI_BDI_INITIALIZED     0x00400000
 986
 987 #ifndef HAVE_BDI_CAP_MAP_COPY
 988 # define BDI_CAP_MAP_COPY       0
 989 #endif
 990
 991 static int super_setup_bdi_name(struct super_block *sb, char *fmt, ...)
 992 {
 993         struct  lustre_sb_info *lsi = s2lsi(sb);
 994         char buf[MAX_STRING_SIZE];
 995         va_list args;
 996         int err;
 997
 998         err = bdi_init(&lsi->lsi_bdi);
 999         if (err)
1000                 return err;
1001
1002         lsi->lsi_flags |= LSI_BDI_INITIALIZED;
1003         lsi->lsi_bdi.capabilities = BDI_CAP_MAP_COPY;
1004         lsi->lsi_bdi.name = "lustre";
1005         va_start(args, fmt);
1006         vsnprintf(buf, MAX_STRING_SIZE, fmt, args);
1007         va_end(args);
1008         err = bdi_register(&lsi->lsi_bdi, NULL, "%s", buf);
1009         va_end(args);
1010         if (!err)
1011                 sb->s_bdi = &lsi->lsi_bdi;
1012
1013         return err;
1014 }
1015 #endif /* !HAVE_SUPER_SETUP_BDI_NAME */
1016
1017 int ll_fill_super(struct super_block *sb, struct vfsmount *mnt)
1018 {
1019         struct  lustre_profile *lprof = NULL;
1020         struct  lustre_sb_info *lsi = s2lsi(sb);
1021         struct  ll_sb_info *sbi = NULL;
1022         char    *dt = NULL, *md = NULL;
1023         char    *profilenm = get_profile_name(sb);
1024         struct config_llog_instance *cfg;
1025         /* %p for void* in printf needs 16+2 characters: 0xffffffffffffffff */
1026         const int instlen = 16 + 2;
1027         unsigned long cfg_instance = ll_get_cfg_instance(sb);
1028         char name[MAX_STRING_SIZE];
1029         int md_len = 0;
1030         int dt_len = 0;
1031         char *ptr;
1032         int len;
1033         int err;
1034
1035         ENTRY;
1036         /* for ASLR, to map between cfg_instance and hashed ptr */
1037         CDEBUG(D_VFSTRACE, "VFS Op: cfg_instance %s-%016lx (sb %p)\n",
1038                profilenm, cfg_instance, sb);
1039
1040         try_module_get(THIS_MODULE);
1041
1042         OBD_ALLOC_PTR(cfg);
1043         if (cfg == NULL)
1044                 GOTO(out_free_cfg, err = -ENOMEM);
1045
1046         /* client additional sb info */
1047         lsi->lsi_llsbi = sbi = ll_init_sbi();
1048         if (!sbi)
1049                 GOTO(out_free_cfg, err = -ENOMEM);
1050
1051         err = ll_options(lsi->lsi_lmd->lmd_opts, sbi);
1052         if (err)
1053                 GOTO(out_free_cfg, err);
1054
1055         err = super_setup_bdi_name(sb, "lustre-%016lx", cfg_instance);
1056         if (err)
1057                 GOTO(out_free_cfg, err);
1058
1059 #ifndef HAVE_DCACHE_LOCK
1060         /* kernel >= 2.6.38 store dentry operations in sb->s_d_op. */
1061         sb->s_d_op = &ll_d_ops;
1062 #endif
1063         /* Get fsname */
1064         len = strlen(profilenm);
1065         ptr = strrchr(profilenm, '-');
1066         if (ptr && (strcmp(ptr, "-client") == 0))
1067                 len -= 7;
1068
1069         /* Mount info */
1070         snprintf(name, MAX_STRING_SIZE, "%.*s-%016lx", len,
1071                  profilenm, cfg_instance);
1072
1073         /* Call ll_debugfs_register_super() before lustre_process_log()
1074          * so that "llite.*.*" params can be processed correctly.
1075          */
1076         err = ll_debugfs_register_super(sb, name);
1077         if (err < 0) {
1078                 CERROR("%s: could not register mountpoint in llite: rc = %d\n",
1079                        ll_get_fsname(sb, NULL, 0), err);
1080                 err = 0;
1081         }
1082
1083         /* The cfg_instance is a value unique to this super, in case some
1084          * joker tries to mount the same fs at two mount points.
1085          */
1086         cfg->cfg_instance = cfg_instance;
1087         cfg->cfg_uuid = lsi->lsi_llsbi->ll_sb_uuid;
1088         cfg->cfg_callback = class_config_llog_handler;
1089         cfg->cfg_sub_clds = CONFIG_SUB_CLIENT;
1090         /* set up client obds */
1091         err = lustre_process_log(sb, profilenm, cfg);
1092         if (err < 0)
1093                 GOTO(out_debugfs, err);
1094
1095         /* Profile set with LCFG_MOUNTOPT so we can find our mdc and osc obds */
1096         lprof = class_get_profile(profilenm);
1097         if (lprof == NULL) {
1098                 LCONSOLE_ERROR_MSG(0x156, "The client profile '%s' could not be"
1099                                    " read from the MGS.  Does that filesystem "
1100                                    "exist?\n", profilenm);
1101                 GOTO(out_debugfs, err = -EINVAL);
1102         }
1103         CDEBUG(D_CONFIG, "Found profile %s: mdc=%s osc=%s\n", profilenm,
1104                lprof->lp_md, lprof->lp_dt);
1105
1106         dt_len = strlen(lprof->lp_dt) + instlen + 2;
1107         OBD_ALLOC(dt, dt_len);
1108         if (!dt)
1109                 GOTO(out_profile, err = -ENOMEM);
1110         snprintf(dt, dt_len - 1, "%s-%016lx", lprof->lp_dt, cfg_instance);
1111
1112         md_len = strlen(lprof->lp_md) + instlen + 2;
1113         OBD_ALLOC(md, md_len);
1114         if (!md)
1115                 GOTO(out_free_dt, err = -ENOMEM);
1116         snprintf(md, md_len - 1, "%s-%016lx", lprof->lp_md, cfg_instance);
1117
1118         /* connections, registrations, sb setup */
1119         err = client_common_fill_super(sb, md, dt, mnt);
1120         if (err < 0)
1121                 GOTO(out_free_md, err);
1122
1123         sbi->ll_client_common_fill_super_succeeded = 1;
1124
1125 out_free_md:
1126         if (md)
1127                 OBD_FREE(md, md_len);
1128 out_free_dt:
1129         if (dt)
1130                 OBD_FREE(dt, dt_len);
1131 out_profile:
1132         if (lprof)
1133                 class_put_profile(lprof);
1134 out_debugfs:
1135         if (err < 0)
1136                 ll_debugfs_unregister_super(sb);
1137 out_free_cfg:
1138         if (cfg)
1139                 OBD_FREE_PTR(cfg);
1140
1141         if (err)
1142                 ll_put_super(sb);
1143         else if (sbi->ll_flags & LL_SBI_VERBOSE)
1144                 LCONSOLE_WARN("Mounted %s\n", profilenm);
1145         RETURN(err);
1146 } /* ll_fill_super */
1147
1148 void ll_put_super(struct super_block *sb)
1149 {
1150         struct config_llog_instance cfg, params_cfg;
1151         struct obd_device *obd;
1152         struct lustre_sb_info *lsi = s2lsi(sb);
1153         struct ll_sb_info *sbi = ll_s2sbi(sb);
1154         char *profilenm = get_profile_name(sb);
1155         unsigned long cfg_instance = ll_get_cfg_instance(sb);
1156         long ccc_count;
1157         int next, force = 1, rc = 0;
1158         ENTRY;
1159
1160         if (!sbi)
1161                 GOTO(out_no_sbi, 0);
1162
1163         /* Should replace instance_id with something better for ASLR */
1164         CDEBUG(D_VFSTRACE, "VFS Op: cfg_instance %s-%016lx (sb %p)\n",
1165                profilenm, cfg_instance, sb);
1166
1167         cfg.cfg_instance = cfg_instance;
1168         lustre_end_log(sb, profilenm, &cfg);
1169
1170         params_cfg.cfg_instance = cfg_instance;
1171         lustre_end_log(sb, PARAMS_FILENAME, &params_cfg);
1172
1173         if (sbi->ll_md_exp) {
1174                 obd = class_exp2obd(sbi->ll_md_exp);
1175                 if (obd)
1176                         force = obd->obd_force;
1177         }
1178
1179         /* Wait for unstable pages to be committed to stable storage */
1180         if (force == 0) {
1181                 struct l_wait_info lwi = LWI_INTR(LWI_ON_SIGNAL_NOOP, NULL);
1182                 rc = l_wait_event(sbi->ll_cache->ccc_unstable_waitq,
1183                         atomic_long_read(&sbi->ll_cache->ccc_unstable_nr) == 0,
1184                         &lwi);
1185         }
1186
1187         ccc_count = atomic_long_read(&sbi->ll_cache->ccc_unstable_nr);
1188         if (force == 0 && rc != -EINTR)
1189                 LASSERTF(ccc_count == 0, "count: %li\n", ccc_count);
1190
1191         /* We need to set force before the lov_disconnect in
1192            lustre_common_put_super, since l_d cleans up osc's as well. */
1193         if (force) {
1194                 next = 0;
1195                 while ((obd = class_devices_in_group(&sbi->ll_sb_uuid,
1196                                                      &next)) != NULL) {
1197                         obd->obd_force = force;
1198                 }
1199         }
1200
1201         if (sbi->ll_client_common_fill_super_succeeded) {
1202                 /* Only if client_common_fill_super succeeded */
1203                 client_common_put_super(sb);
1204         }
1205
1206         next = 0;
1207         while ((obd = class_devices_in_group(&sbi->ll_sb_uuid, &next)) !=NULL) {
1208                 class_manual_cleanup(obd);
1209         }
1210
1211         if (sbi->ll_flags & LL_SBI_VERBOSE)
1212                 LCONSOLE_WARN("Unmounted %s\n", profilenm ? profilenm : "");
1213
1214         if (profilenm)
1215                 class_del_profile(profilenm);
1216
1217 #ifndef HAVE_SUPER_SETUP_BDI_NAME
1218         if (lsi->lsi_flags & LSI_BDI_INITIALIZED) {
1219                 bdi_destroy(&lsi->lsi_bdi);
1220                 lsi->lsi_flags &= ~LSI_BDI_INITIALIZED;
1221         }
1222 #endif
1223
1224         ll_free_sbi(sb);
1225         lsi->lsi_llsbi = NULL;
1226 out_no_sbi:
1227         lustre_common_put_super(sb);
1228
1229         cl_env_cache_purge(~0);
1230
1231         module_put(THIS_MODULE);
1232
1233         EXIT;
1234 } /* client_put_super */
1235
1236 struct inode *ll_inode_from_resource_lock(struct ldlm_lock *lock)
1237 {
1238         struct inode *inode = NULL;
1239
1240         /* NOTE: we depend on atomic igrab() -bzzz */
1241         lock_res_and_lock(lock);
1242         if (lock->l_resource->lr_lvb_inode) {
1243                 struct ll_inode_info * lli;
1244                 lli = ll_i2info(lock->l_resource->lr_lvb_inode);
1245                 if (lli->lli_inode_magic == LLI_INODE_MAGIC) {
1246                         inode = igrab(lock->l_resource->lr_lvb_inode);
1247                 } else {
1248                         inode = lock->l_resource->lr_lvb_inode;
1249                         LDLM_DEBUG_LIMIT(inode->i_state & I_FREEING ?  D_INFO :
1250                                          D_WARNING, lock, "lr_lvb_inode %p is "
1251                                          "bogus: magic %08x",
1252                                          lock->l_resource->lr_lvb_inode,
1253                                          lli->lli_inode_magic);
1254                         inode = NULL;
1255                 }
1256         }
1257         unlock_res_and_lock(lock);
1258         return inode;
1259 }
1260
1261 void ll_dir_clear_lsm_md(struct inode *inode)
1262 {
1263         struct ll_inode_info *lli = ll_i2info(inode);
1264
1265         LASSERT(S_ISDIR(inode->i_mode));
1266
1267         if (lli->lli_lsm_md != NULL) {
1268                 lmv_free_memmd(lli->lli_lsm_md);
1269                 lli->lli_lsm_md = NULL;
1270         }
1271 }
1272
1273 static struct inode *ll_iget_anon_dir(struct super_block *sb,
1274                                       const struct lu_fid *fid,
1275                                       struct lustre_md *md)
1276 {
1277         struct ll_sb_info       *sbi = ll_s2sbi(sb);
1278         struct mdt_body         *body = md->body;
1279         struct inode            *inode;
1280         ino_t                   ino;
1281         ENTRY;
1282
1283         ino = cl_fid_build_ino(fid, sbi->ll_flags & LL_SBI_32BIT_API);
1284         inode = iget_locked(sb, ino);
1285         if (inode == NULL) {
1286                 CERROR("%s: failed get simple inode "DFID": rc = -ENOENT\n",
1287                        ll_get_fsname(sb, NULL, 0), PFID(fid));
1288                 RETURN(ERR_PTR(-ENOENT));
1289         }
1290
1291         if (inode->i_state & I_NEW) {
1292                 struct ll_inode_info *lli = ll_i2info(inode);
1293                 struct lmv_stripe_md *lsm = md->lmv;
1294
1295                 inode->i_mode = (inode->i_mode & ~S_IFMT) |
1296                                 (body->mbo_mode & S_IFMT);
1297                 LASSERTF(S_ISDIR(inode->i_mode), "Not slave inode "DFID"\n",
1298                          PFID(fid));
1299
1300                 inode->i_mtime.tv_sec = 0;
1301                 inode->i_atime.tv_sec = 0;
1302                 inode->i_ctime.tv_sec = 0;
1303                 inode->i_rdev = 0;
1304
1305 #ifdef HAVE_BACKING_DEV_INFO
1306                 /* initializing backing dev info. */
1307                 inode->i_mapping->backing_dev_info =
1308                                                 &s2lsi(inode->i_sb)->lsi_bdi;
1309 #endif
1310                 inode->i_op = &ll_dir_inode_operations;
1311                 inode->i_fop = &ll_dir_operations;
1312                 lli->lli_fid = *fid;
1313                 ll_lli_init(lli);
1314
1315                 LASSERT(lsm != NULL);
1316                 /* master object FID */
1317                 lli->lli_pfid = body->mbo_fid1;
1318                 CDEBUG(D_INODE, "lli %p slave "DFID" master "DFID"\n",
1319                        lli, PFID(fid), PFID(&lli->lli_pfid));
1320                 unlock_new_inode(inode);
1321         }
1322
1323         RETURN(inode);
1324 }
1325
1326 static int ll_init_lsm_md(struct inode *inode, struct lustre_md *md)
1327 {
1328         struct lu_fid *fid;
1329         struct lmv_stripe_md *lsm = md->lmv;
1330         struct ll_inode_info *lli = ll_i2info(inode);
1331         int i;
1332
1333         LASSERT(lsm != NULL);
1334
1335         CDEBUG(D_INODE, "%s: "DFID" set dir layout:\n",
1336                 ll_get_fsname(inode->i_sb, NULL, 0),
1337                 PFID(&lli->lli_fid));
1338         lsm_md_dump(D_INODE, lsm);
1339
1340         /* XXX sigh, this lsm_root initialization should be in
1341          * LMV layer, but it needs ll_iget right now, so we
1342          * put this here right now. */
1343         for (i = 0; i < lsm->lsm_md_stripe_count; i++) {
1344                 fid = &lsm->lsm_md_oinfo[i].lmo_fid;
1345                 LASSERT(lsm->lsm_md_oinfo[i].lmo_root == NULL);
1346                 /* Unfortunately ll_iget will call ll_update_inode,
1347                  * where the initialization of slave inode is slightly
1348                  * different, so it reset lsm_md to NULL to avoid
1349                  * initializing lsm for slave inode. */
1350                 lsm->lsm_md_oinfo[i].lmo_root =
1351                                 ll_iget_anon_dir(inode->i_sb, fid, md);
1352                 if (IS_ERR(lsm->lsm_md_oinfo[i].lmo_root)) {
1353                         int rc = PTR_ERR(lsm->lsm_md_oinfo[i].lmo_root);
1354
1355                         lsm->lsm_md_oinfo[i].lmo_root = NULL;
1356                         while (i-- > 0) {
1357                                 iput(lsm->lsm_md_oinfo[i].lmo_root);
1358                                 lsm->lsm_md_oinfo[i].lmo_root = NULL;
1359                         }
1360                         return rc;
1361                 }
1362         }
1363
1364         lli->lli_lsm_md = lsm;
1365
1366         return 0;
1367 }
1368
1369 static int ll_update_lsm_md(struct inode *inode, struct lustre_md *md)
1370 {
1371         struct ll_inode_info *lli = ll_i2info(inode);
1372         struct lmv_stripe_md *lsm = md->lmv;
1373         int rc = 0;
1374
1375         ENTRY;
1376
1377         LASSERT(S_ISDIR(inode->i_mode));
1378         CDEBUG(D_INODE, "update lsm %p of "DFID"\n", lli->lli_lsm_md,
1379                PFID(ll_inode2fid(inode)));
1380
1381         /*
1382          * no striped information from request, lustre_md from req does not
1383          * include stripeEA, see ll_md_setattr()
1384          */
1385         if (!lsm)
1386                 RETURN(0);
1387
1388         /*
1389          * normally dir layout doesn't change, only take read lock to check
1390          * that to avoid blocking other MD operations.
1391          */
1392         if (lli->lli_lsm_md)
1393                 down_read(&lli->lli_lsm_sem);
1394         else
1395                 down_write(&lli->lli_lsm_sem);
1396
1397         /*
1398          * if dir layout mismatch, check whether version is increased, which
1399          * means layout is changed, this happens in dir migration and lfsck.
1400          */
1401         if (lli->lli_lsm_md && !lsm_md_eq(lli->lli_lsm_md, lsm)) {
1402                 if (lsm->lsm_md_layout_version <=
1403                     lli->lli_lsm_md->lsm_md_layout_version) {
1404                         CERROR("%s: "DFID" dir layout mismatch:\n",
1405                                 ll_get_fsname(inode->i_sb, NULL, 0),
1406                                 PFID(&lli->lli_fid));
1407                         lsm_md_dump(D_ERROR, lli->lli_lsm_md);
1408                         lsm_md_dump(D_ERROR, lsm);
1409                         GOTO(unlock, rc = -EINVAL);
1410                 }
1411
1412                 /* layout changed, switch to write lock */
1413                 up_read(&lli->lli_lsm_sem);
1414                 down_write(&lli->lli_lsm_sem);
1415                 ll_dir_clear_lsm_md(inode);
1416         }
1417
1418         /* set directory layout */
1419         if (!lli->lli_lsm_md) {
1420                 struct cl_attr  *attr;
1421
1422                 rc = ll_init_lsm_md(inode, md);
1423                 up_write(&lli->lli_lsm_sem);
1424                 if (rc != 0)
1425                         RETURN(rc);
1426
1427                 /* set md->lmv to NULL, so the following free lustre_md
1428                  * will not free this lsm */
1429                 md->lmv = NULL;
1430
1431                 /*
1432                  * md_merge_attr() may take long, since lsm is already set,
1433                  * switch to read lock.
1434                  */
1435                 down_read(&lli->lli_lsm_sem);
1436
1437                 OBD_ALLOC_PTR(attr);
1438                 if (attr == NULL)
1439                         GOTO(unlock, rc = -ENOMEM);
1440
1441                 /* validate the lsm */
1442                 rc = md_merge_attr(ll_i2mdexp(inode), lsm, attr,
1443                                    ll_md_blocking_ast);
1444                 if (rc != 0) {
1445                         OBD_FREE_PTR(attr);
1446                         GOTO(unlock, rc);
1447                 }
1448
1449                 if (md->body->mbo_valid & OBD_MD_FLNLINK)
1450                         md->body->mbo_nlink = attr->cat_nlink;
1451                 if (md->body->mbo_valid & OBD_MD_FLSIZE)
1452                         md->body->mbo_size = attr->cat_size;
1453                 if (md->body->mbo_valid & OBD_MD_FLATIME)
1454                         md->body->mbo_atime = attr->cat_atime;
1455                 if (md->body->mbo_valid & OBD_MD_FLCTIME)
1456                         md->body->mbo_ctime = attr->cat_ctime;
1457                 if (md->body->mbo_valid & OBD_MD_FLMTIME)
1458                         md->body->mbo_mtime = attr->cat_mtime;
1459
1460                 OBD_FREE_PTR(attr);
1461         }
1462 unlock:
1463         up_read(&lli->lli_lsm_sem);
1464
1465         RETURN(rc);
1466 }
1467
1468 void ll_clear_inode(struct inode *inode)
1469 {
1470         struct ll_inode_info *lli = ll_i2info(inode);
1471         struct ll_sb_info *sbi = ll_i2sbi(inode);
1472         ENTRY;
1473
1474         CDEBUG(D_VFSTRACE, "VFS Op:inode="DFID"(%p)\n",
1475                PFID(ll_inode2fid(inode)), inode);
1476
1477         if (S_ISDIR(inode->i_mode)) {
1478                 /* these should have been cleared in ll_file_release */
1479                 LASSERT(lli->lli_opendir_key == NULL);
1480                 LASSERT(lli->lli_sai == NULL);
1481                 LASSERT(lli->lli_opendir_pid == 0);
1482         }
1483
1484         md_null_inode(sbi->ll_md_exp, ll_inode2fid(inode));
1485
1486         LASSERT(!lli->lli_open_fd_write_count);
1487         LASSERT(!lli->lli_open_fd_read_count);
1488         LASSERT(!lli->lli_open_fd_exec_count);
1489
1490         if (lli->lli_mds_write_och)
1491                 ll_md_real_close(inode, FMODE_WRITE);
1492         if (lli->lli_mds_exec_och)
1493                 ll_md_real_close(inode, FMODE_EXEC);
1494         if (lli->lli_mds_read_och)
1495                 ll_md_real_close(inode, FMODE_READ);
1496
1497         if (S_ISLNK(inode->i_mode) && lli->lli_symlink_name) {
1498                 OBD_FREE(lli->lli_symlink_name,
1499                          strlen(lli->lli_symlink_name) + 1);
1500                 lli->lli_symlink_name = NULL;
1501         }
1502
1503         ll_xattr_cache_destroy(inode);
1504
1505 #ifdef CONFIG_FS_POSIX_ACL
1506         forget_all_cached_acls(inode);
1507         if (lli->lli_posix_acl) {
1508                 posix_acl_release(lli->lli_posix_acl);
1509                 lli->lli_posix_acl = NULL;
1510         }
1511 #endif
1512         lli->lli_inode_magic = LLI_INODE_DEAD;
1513
1514         if (S_ISDIR(inode->i_mode))
1515                 ll_dir_clear_lsm_md(inode);
1516         else if (S_ISREG(inode->i_mode) && !is_bad_inode(inode))
1517                 LASSERT(list_empty(&lli->lli_agl_list));
1518
1519         /*
1520          * XXX This has to be done before lsm is freed below, because
1521          * cl_object still uses inode lsm.
1522          */
1523         cl_inode_fini(inode);
1524
1525         EXIT;
1526 }
1527
1528 static int ll_md_setattr(struct dentry *dentry, struct md_op_data *op_data)
1529 {
1530         struct lustre_md md;
1531         struct inode *inode = dentry->d_inode;
1532         struct ll_sb_info *sbi = ll_i2sbi(inode);
1533         struct ptlrpc_request *request = NULL;
1534         int rc, ia_valid;
1535         ENTRY;
1536
1537         op_data = ll_prep_md_op_data(op_data, inode, NULL, NULL, 0, 0,
1538                                      LUSTRE_OPC_ANY, NULL);
1539         if (IS_ERR(op_data))
1540                 RETURN(PTR_ERR(op_data));
1541
1542         rc = md_setattr(sbi->ll_md_exp, op_data, NULL, 0, &request);
1543         if (rc) {
1544                 ptlrpc_req_finished(request);
1545                 if (rc == -ENOENT) {
1546                         clear_nlink(inode);
1547                         /* Unlinked special device node? Or just a race?
1548                          * Pretend we done everything. */
1549                         if (!S_ISREG(inode->i_mode) &&
1550                             !S_ISDIR(inode->i_mode)) {
1551                                 ia_valid = op_data->op_attr.ia_valid;
1552                                 op_data->op_attr.ia_valid &= ~TIMES_SET_FLAGS;
1553                                 rc = simple_setattr(dentry, &op_data->op_attr);
1554                                 op_data->op_attr.ia_valid = ia_valid;
1555                         }
1556                 } else if (rc != -EPERM && rc != -EACCES && rc != -ETXTBSY) {
1557                         CERROR("md_setattr fails: rc = %d\n", rc);
1558                 }
1559                 RETURN(rc);
1560         }
1561
1562         rc = md_get_lustre_md(sbi->ll_md_exp, request, sbi->ll_dt_exp,
1563                               sbi->ll_md_exp, &md);
1564         if (rc) {
1565                 ptlrpc_req_finished(request);
1566                 RETURN(rc);
1567         }
1568
1569         ia_valid = op_data->op_attr.ia_valid;
1570         /* inode size will be in ll_setattr_ost, can't do it now since dirty
1571          * cache is not cleared yet. */
1572         op_data->op_attr.ia_valid &= ~(TIMES_SET_FLAGS | ATTR_SIZE);
1573         if (S_ISREG(inode->i_mode))
1574                 inode_lock(inode);
1575         rc = simple_setattr(dentry, &op_data->op_attr);
1576         if (S_ISREG(inode->i_mode))
1577                 inode_unlock(inode);
1578         op_data->op_attr.ia_valid = ia_valid;
1579
1580         rc = ll_update_inode(inode, &md);
1581         ptlrpc_req_finished(request);
1582
1583         RETURN(rc);
1584 }
1585
1586 /* If this inode has objects allocated to it (lsm != NULL), then the OST
1587  * object(s) determine the file size and mtime.  Otherwise, the MDS will
1588  * keep these values until such a time that objects are allocated for it.
1589  * We do the MDS operations first, as it is checking permissions for us.
1590  * We don't to the MDS RPC if there is nothing that we want to store there,
1591  * otherwise there is no harm in updating mtime/atime on the MDS if we are
1592  * going to do an RPC anyways.
1593  *
1594  * If we are doing a truncate, we will send the mtime and ctime updates
1595  * to the OST with the punch RPC, otherwise we do an explicit setattr RPC.
1596  * I don't believe it is possible to get e.g. ATTR_MTIME_SET and ATTR_SIZE
1597  * at the same time.
1598  *
1599  * In case of HSMimport, we only set attr on MDS.
1600  */
1601 int ll_setattr_raw(struct dentry *dentry, struct iattr *attr,
1602                    enum op_xvalid xvalid, bool hsm_import)
1603 {
1604         struct inode *inode = dentry->d_inode;
1605         struct ll_inode_info *lli = ll_i2info(inode);
1606         struct md_op_data *op_data = NULL;
1607         int rc = 0;
1608         ENTRY;
1609
1610         CDEBUG(D_VFSTRACE, "%s: setattr inode "DFID"(%p) from %llu to %llu, "
1611                "valid %x, hsm_import %d\n",
1612                ll_get_fsname(inode->i_sb, NULL, 0), PFID(&lli->lli_fid),
1613                inode, i_size_read(inode), attr->ia_size, attr->ia_valid,
1614                hsm_import);
1615
1616         if (attr->ia_valid & ATTR_SIZE) {
1617                 /* Check new size against VFS/VM file size limit and rlimit */
1618                 rc = inode_newsize_ok(inode, attr->ia_size);
1619                 if (rc)
1620                         RETURN(rc);
1621
1622                 /* The maximum Lustre file size is variable, based on the
1623                  * OST maximum object size and number of stripes.  This
1624                  * needs another check in addition to the VFS check above. */
1625                 if (attr->ia_size > ll_file_maxbytes(inode)) {
1626                         CDEBUG(D_INODE,"file "DFID" too large %llu > %llu\n",
1627                                PFID(&lli->lli_fid), attr->ia_size,
1628                                ll_file_maxbytes(inode));
1629                         RETURN(-EFBIG);
1630                 }
1631
1632                 attr->ia_valid |= ATTR_MTIME | ATTR_CTIME;
1633         }
1634
1635         /* POSIX: check before ATTR_*TIME_SET set (from inode_change_ok) */
1636         if (attr->ia_valid & TIMES_SET_FLAGS) {
1637                 if ((!uid_eq(current_fsuid(), inode->i_uid)) &&
1638                     !cfs_capable(CFS_CAP_FOWNER))
1639                         RETURN(-EPERM);
1640         }
1641
1642         /* We mark all of the fields "set" so MDS/OST does not re-set them */
1643         if (!(xvalid & OP_XVALID_CTIME_SET) &&
1644              (attr->ia_valid & ATTR_CTIME)) {
1645                 attr->ia_ctime = current_time(inode);
1646                 xvalid |= OP_XVALID_CTIME_SET;
1647         }
1648         if (!(attr->ia_valid & ATTR_ATIME_SET) &&
1649             (attr->ia_valid & ATTR_ATIME)) {
1650                 attr->ia_atime = current_time(inode);
1651                 attr->ia_valid |= ATTR_ATIME_SET;
1652         }
1653         if (!(attr->ia_valid & ATTR_MTIME_SET) &&
1654             (attr->ia_valid & ATTR_MTIME)) {
1655                 attr->ia_mtime = current_time(inode);
1656                 attr->ia_valid |= ATTR_MTIME_SET;
1657         }
1658
1659         if (attr->ia_valid & (ATTR_MTIME | ATTR_CTIME))
1660                 CDEBUG(D_INODE, "setting mtime %lld, ctime %lld, now = %lld\n",
1661                        (s64)attr->ia_mtime.tv_sec, (s64)attr->ia_ctime.tv_sec,
1662                        ktime_get_real_seconds());
1663
1664         if (S_ISREG(inode->i_mode)) {
1665                 if (attr->ia_valid & ATTR_SIZE)
1666                         inode_dio_write_done(inode);
1667                 inode_unlock(inode);
1668         }
1669
1670         /* We always do an MDS RPC, even if we're only changing the size;
1671          * only the MDS knows whether truncate() should fail with -ETXTBUSY */
1672
1673         OBD_ALLOC_PTR(op_data);
1674         if (op_data == NULL)
1675                 GOTO(out, rc = -ENOMEM);
1676
1677         if (!hsm_import && attr->ia_valid & ATTR_SIZE) {
1678                 /* If we are changing file size, file content is
1679                  * modified, flag it.
1680                  */
1681                 xvalid |= OP_XVALID_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         op_data->op_xvalid = xvalid;
1695
1696         rc = ll_md_setattr(dentry, op_data);
1697         if (rc)
1698                 GOTO(out, rc);
1699
1700         if (!S_ISREG(inode->i_mode) || hsm_import)
1701                 GOTO(out, rc = 0);
1702
1703         if (attr->ia_valid & (ATTR_SIZE | ATTR_ATIME | ATTR_ATIME_SET |
1704                               ATTR_MTIME | ATTR_MTIME_SET | ATTR_CTIME) ||
1705             xvalid & OP_XVALID_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                  */
1713                 rc = cl_setattr_ost(lli->lli_clob, attr, xvalid, 0);
1714         }
1715
1716         /* If the file was restored, it needs to set dirty flag.
1717          *
1718          * We've already sent MDS_DATA_MODIFIED flag in
1719          * ll_md_setattr() for truncate. However, the MDT refuses to
1720          * set the HS_DIRTY flag on released files, so we have to set
1721          * it again if the file has been restored. Please check how
1722          * LLIF_DATA_MODIFIED is set in vvp_io_setattr_fini().
1723          *
1724          * Please notice that if the file is not released, the previous
1725          * MDS_DATA_MODIFIED has taken effect and usually
1726          * LLIF_DATA_MODIFIED is not set(see vvp_io_setattr_fini()).
1727          * This way we can save an RPC for common open + trunc
1728          * operation. */
1729         if (ll_file_test_and_clear_flag(lli, LLIF_DATA_MODIFIED)) {
1730                 struct hsm_state_set hss = {
1731                         .hss_valid = HSS_SETMASK,
1732                         .hss_setmask = HS_DIRTY,
1733                 };
1734                 int rc2;
1735
1736                 rc2 = ll_hsm_state_set(inode, &hss);
1737                 /* truncate and write can happen at the same time, so that
1738                  * the file can be set modified even though the file is not
1739                  * restored from released state, and ll_hsm_state_set() is
1740                  * not applicable for the file, and rc2 < 0 is normal in this
1741                  * case. */
1742                 if (rc2 < 0)
1743                         CDEBUG(D_INFO, DFID "HSM set dirty failed: rc2 = %d\n",
1744                                PFID(ll_inode2fid(inode)), rc2);
1745         }
1746
1747         EXIT;
1748 out:
1749         if (op_data != NULL)
1750                 ll_finish_md_op_data(op_data);
1751
1752         if (S_ISREG(inode->i_mode)) {
1753                 inode_lock(inode);
1754                 if ((attr->ia_valid & ATTR_SIZE) && !hsm_import)
1755                         inode_dio_wait(inode);
1756                 /* Once we've got the i_mutex, it's safe to set the S_NOSEC
1757                  * flag.  ll_update_inode (called from ll_md_setattr), clears
1758                  * inode flags, so there is a gap where S_NOSEC is not set.
1759                  * This can cause a writer to take the i_mutex unnecessarily,
1760                  * but this is safe to do and should be rare. */
1761                 inode_has_no_xattr(inode);
1762         }
1763
1764         ll_stats_ops_tally(ll_i2sbi(inode), (attr->ia_valid & ATTR_SIZE) ?
1765                         LPROC_LL_TRUNC : LPROC_LL_SETATTR, 1);
1766
1767         return rc;
1768 }
1769
1770 int ll_setattr(struct dentry *de, struct iattr *attr)
1771 {
1772         int mode = de->d_inode->i_mode;
1773         enum op_xvalid xvalid = 0;
1774
1775         if ((attr->ia_valid & (ATTR_CTIME|ATTR_SIZE|ATTR_MODE)) ==
1776                               (ATTR_CTIME|ATTR_SIZE|ATTR_MODE))
1777                 xvalid |= OP_XVALID_OWNEROVERRIDE;
1778
1779         if (((attr->ia_valid & (ATTR_MODE|ATTR_FORCE|ATTR_SIZE)) ==
1780                                (ATTR_SIZE|ATTR_MODE)) &&
1781             (((mode & S_ISUID) && !(attr->ia_mode & S_ISUID)) ||
1782              (((mode & (S_ISGID|S_IXGRP)) == (S_ISGID|S_IXGRP)) &&
1783               !(attr->ia_mode & S_ISGID))))
1784                 attr->ia_valid |= ATTR_FORCE;
1785
1786         if ((attr->ia_valid & ATTR_MODE) &&
1787             (mode & S_ISUID) &&
1788             !(attr->ia_mode & S_ISUID) &&
1789             !(attr->ia_valid & ATTR_KILL_SUID))
1790                 attr->ia_valid |= ATTR_KILL_SUID;
1791
1792         if ((attr->ia_valid & ATTR_MODE) &&
1793             ((mode & (S_ISGID|S_IXGRP)) == (S_ISGID|S_IXGRP)) &&
1794             !(attr->ia_mode & S_ISGID) &&
1795             !(attr->ia_valid & ATTR_KILL_SGID))
1796                 attr->ia_valid |= ATTR_KILL_SGID;
1797
1798         return ll_setattr_raw(de, attr, xvalid, false);
1799 }
1800
1801 int ll_statfs_internal(struct ll_sb_info *sbi, struct obd_statfs *osfs,
1802                        u32 flags)
1803 {
1804         struct obd_statfs obd_osfs = { 0 };
1805         time64_t max_age;
1806         int rc;
1807
1808         ENTRY;
1809         max_age = ktime_get_seconds() - OBD_STATFS_CACHE_SECONDS;
1810
1811         rc = obd_statfs(NULL, sbi->ll_md_exp, osfs, max_age, flags);
1812         if (rc)
1813                 RETURN(rc);
1814
1815         osfs->os_type = LL_SUPER_MAGIC;
1816
1817         CDEBUG(D_SUPER, "MDC blocks %llu/%llu objects %llu/%llu\n",
1818               osfs->os_bavail, osfs->os_blocks, osfs->os_ffree, osfs->os_files);
1819
1820         if (osfs->os_state & OS_STATE_SUM)
1821                 GOTO(out, rc);
1822
1823         if (sbi->ll_flags & LL_SBI_LAZYSTATFS)
1824                 flags |= OBD_STATFS_NODELAY;
1825
1826         rc = obd_statfs(NULL, sbi->ll_dt_exp, &obd_osfs, max_age, flags);
1827         if (rc) /* Possibly a filesystem with no OSTs.  Report MDT totals. */
1828                 GOTO(out, rc = 0);
1829
1830         CDEBUG(D_SUPER, "OSC blocks %llu/%llu objects %llu/%llu\n",
1831                obd_osfs.os_bavail, obd_osfs.os_blocks, obd_osfs.os_ffree,
1832                obd_osfs.os_files);
1833
1834         osfs->os_bsize = obd_osfs.os_bsize;
1835         osfs->os_blocks = obd_osfs.os_blocks;
1836         osfs->os_bfree = obd_osfs.os_bfree;
1837         osfs->os_bavail = obd_osfs.os_bavail;
1838
1839         /* If we have _some_ OSTs, but don't have as many free objects on the
1840          * OSTs as inodes on the MDTs, reduce the reported number of inodes
1841          * to compensate, so that the "inodes in use" number is correct.
1842          * This should be kept in sync with lod_statfs() behaviour.
1843          */
1844         if (obd_osfs.os_files && obd_osfs.os_ffree < osfs->os_ffree) {
1845                 osfs->os_files = (osfs->os_files - osfs->os_ffree) +
1846                                  obd_osfs.os_ffree;
1847                 osfs->os_ffree = obd_osfs.os_ffree;
1848         }
1849
1850 out:
1851         RETURN(rc);
1852 }
1853 int ll_statfs(struct dentry *de, struct kstatfs *sfs)
1854 {
1855         struct super_block *sb = de->d_sb;
1856         struct obd_statfs osfs;
1857         __u64 fsid = huge_encode_dev(sb->s_dev);
1858         int rc;
1859
1860         CDEBUG(D_VFSTRACE, "VFS Op: at %llu jiffies\n", get_jiffies_64());
1861         ll_stats_ops_tally(ll_s2sbi(sb), LPROC_LL_STAFS, 1);
1862
1863         /* Some amount of caching on the client is allowed */
1864         rc = ll_statfs_internal(ll_s2sbi(sb), &osfs, OBD_STATFS_SUM);
1865         if (rc)
1866                 return rc;
1867
1868         statfs_unpack(sfs, &osfs);
1869
1870         /* We need to downshift for all 32-bit kernels, because we can't
1871          * tell if the kernel is being called via sys_statfs64() or not.
1872          * Stop before overflowing f_bsize - in which case it is better
1873          * to just risk EOVERFLOW if caller is using old sys_statfs(). */
1874         if (sizeof(long) < 8) {
1875                 while (osfs.os_blocks > ~0UL && sfs->f_bsize < 0x40000000) {
1876                         sfs->f_bsize <<= 1;
1877
1878                         osfs.os_blocks >>= 1;
1879                         osfs.os_bfree >>= 1;
1880                         osfs.os_bavail >>= 1;
1881                 }
1882         }
1883
1884         sfs->f_blocks = osfs.os_blocks;
1885         sfs->f_bfree = osfs.os_bfree;
1886         sfs->f_bavail = osfs.os_bavail;
1887         sfs->f_fsid.val[0] = (__u32)fsid;
1888         sfs->f_fsid.val[1] = (__u32)(fsid >> 32);
1889         return 0;
1890 }
1891
1892 void ll_inode_size_lock(struct inode *inode)
1893 {
1894         struct ll_inode_info *lli;
1895
1896         LASSERT(!S_ISDIR(inode->i_mode));
1897
1898         lli = ll_i2info(inode);
1899         mutex_lock(&lli->lli_size_mutex);
1900 }
1901
1902 void ll_inode_size_unlock(struct inode *inode)
1903 {
1904         struct ll_inode_info *lli;
1905
1906         lli = ll_i2info(inode);
1907         mutex_unlock(&lli->lli_size_mutex);
1908 }
1909
1910 void ll_update_inode_flags(struct inode *inode, int ext_flags)
1911 {
1912         inode->i_flags = ll_ext_to_inode_flags(ext_flags);
1913         if (ext_flags & LUSTRE_PROJINHERIT_FL)
1914                 ll_file_set_flag(ll_i2info(inode), LLIF_PROJECT_INHERIT);
1915         else
1916                 ll_file_clear_flag(ll_i2info(inode), LLIF_PROJECT_INHERIT);
1917 }
1918
1919 int ll_update_inode(struct inode *inode, struct lustre_md *md)
1920 {
1921         struct ll_inode_info *lli = ll_i2info(inode);
1922         struct mdt_body *body = md->body;
1923         struct ll_sb_info *sbi = ll_i2sbi(inode);
1924         int rc = 0;
1925
1926         if (body->mbo_valid & OBD_MD_FLEASIZE) {
1927                 rc = cl_file_inode_init(inode, md);
1928                 if (rc)
1929                         return rc;
1930         }
1931
1932         if (S_ISDIR(inode->i_mode)) {
1933                 rc = ll_update_lsm_md(inode, md);
1934                 if (rc != 0)
1935                         return rc;
1936         }
1937
1938 #ifdef CONFIG_FS_POSIX_ACL
1939         if (body->mbo_valid & OBD_MD_FLACL) {
1940                 spin_lock(&lli->lli_lock);
1941                 if (lli->lli_posix_acl)
1942                         posix_acl_release(lli->lli_posix_acl);
1943                 lli->lli_posix_acl = md->posix_acl;
1944                 spin_unlock(&lli->lli_lock);
1945         }
1946 #endif
1947         inode->i_ino = cl_fid_build_ino(&body->mbo_fid1,
1948                                         sbi->ll_flags & LL_SBI_32BIT_API);
1949         inode->i_generation = cl_fid_build_gen(&body->mbo_fid1);
1950
1951         if (body->mbo_valid & OBD_MD_FLATIME) {
1952                 if (body->mbo_atime > inode->i_atime.tv_sec)
1953                         inode->i_atime.tv_sec = body->mbo_atime;
1954                 lli->lli_atime = body->mbo_atime;
1955         }
1956
1957         if (body->mbo_valid & OBD_MD_FLMTIME) {
1958                 if (body->mbo_mtime > inode->i_mtime.tv_sec) {
1959                         CDEBUG(D_INODE,
1960                                "setting ino %lu mtime from %lld to %llu\n",
1961                                inode->i_ino, (s64)inode->i_mtime.tv_sec,
1962                                body->mbo_mtime);
1963                         inode->i_mtime.tv_sec = body->mbo_mtime;
1964                 }
1965                 lli->lli_mtime = body->mbo_mtime;
1966         }
1967
1968         if (body->mbo_valid & OBD_MD_FLCTIME) {
1969                 if (body->mbo_ctime > inode->i_ctime.tv_sec)
1970                         inode->i_ctime.tv_sec = body->mbo_ctime;
1971                 lli->lli_ctime = body->mbo_ctime;
1972         }
1973
1974         /* Clear i_flags to remove S_NOSEC before permissions are updated */
1975         if (body->mbo_valid & OBD_MD_FLFLAGS)
1976                 ll_update_inode_flags(inode, body->mbo_flags);
1977         if (body->mbo_valid & OBD_MD_FLMODE)
1978                 inode->i_mode = (inode->i_mode & S_IFMT) |
1979                                 (body->mbo_mode & ~S_IFMT);
1980
1981         if (body->mbo_valid & OBD_MD_FLTYPE)
1982                 inode->i_mode = (inode->i_mode & ~S_IFMT) |
1983                                 (body->mbo_mode & S_IFMT);
1984
1985         LASSERT(inode->i_mode != 0);
1986         if (S_ISREG(inode->i_mode))
1987                 inode->i_blkbits = min(PTLRPC_MAX_BRW_BITS + 1,
1988                                        LL_MAX_BLKSIZE_BITS);
1989         else
1990                 inode->i_blkbits = inode->i_sb->s_blocksize_bits;
1991
1992         if (body->mbo_valid & OBD_MD_FLUID)
1993                 inode->i_uid = make_kuid(&init_user_ns, body->mbo_uid);
1994         if (body->mbo_valid & OBD_MD_FLGID)
1995                 inode->i_gid = make_kgid(&init_user_ns, body->mbo_gid);
1996         if (body->mbo_valid & OBD_MD_FLPROJID)
1997                 lli->lli_projid = body->mbo_projid;
1998         if (body->mbo_valid & OBD_MD_FLNLINK)
1999                 set_nlink(inode, body->mbo_nlink);
2000         if (body->mbo_valid & OBD_MD_FLRDEV)
2001                 inode->i_rdev = old_decode_dev(body->mbo_rdev);
2002
2003         if (body->mbo_valid & OBD_MD_FLID) {
2004                 /* FID shouldn't be changed! */
2005                 if (fid_is_sane(&lli->lli_fid)) {
2006                         LASSERTF(lu_fid_eq(&lli->lli_fid, &body->mbo_fid1),
2007                                  "Trying to change FID "DFID
2008                                  " to the "DFID", inode "DFID"(%p)\n",
2009                                  PFID(&lli->lli_fid), PFID(&body->mbo_fid1),
2010                                  PFID(ll_inode2fid(inode)), inode);
2011                 } else {
2012                         lli->lli_fid = body->mbo_fid1;
2013                 }
2014         }
2015
2016         LASSERT(fid_seq(&lli->lli_fid) != 0);
2017
2018         if (body->mbo_valid & OBD_MD_FLSIZE) {
2019                 i_size_write(inode, body->mbo_size);
2020
2021                 CDEBUG(D_VFSTRACE, "inode="DFID", updating i_size %llu\n",
2022                        PFID(ll_inode2fid(inode)),
2023                        (unsigned long long)body->mbo_size);
2024
2025                 if (body->mbo_valid & OBD_MD_FLBLOCKS)
2026                         inode->i_blocks = body->mbo_blocks;
2027         }
2028
2029         if (body->mbo_valid & OBD_MD_TSTATE) {
2030                 /* Set LLIF_FILE_RESTORING if restore ongoing and
2031                  * clear it when done to ensure to start again
2032                  * glimpsing updated attrs
2033                  */
2034                 if (body->mbo_t_state & MS_RESTORE)
2035                         ll_file_set_flag(lli, LLIF_FILE_RESTORING);
2036                 else
2037                         ll_file_clear_flag(lli, LLIF_FILE_RESTORING);
2038         }
2039
2040         return 0;
2041 }
2042
2043 int ll_read_inode2(struct inode *inode, void *opaque)
2044 {
2045         struct lustre_md *md = opaque;
2046         struct ll_inode_info *lli = ll_i2info(inode);
2047         int     rc;
2048         ENTRY;
2049
2050         CDEBUG(D_VFSTRACE, "VFS Op:inode="DFID"(%p)\n",
2051                PFID(&lli->lli_fid), inode);
2052
2053         /* Core attributes from the MDS first.  This is a new inode, and
2054          * the VFS doesn't zero times in the core inode so we have to do
2055          * it ourselves.  They will be overwritten by either MDS or OST
2056          * attributes - we just need to make sure they aren't newer.
2057          */
2058         inode->i_mtime.tv_sec = 0;
2059         inode->i_atime.tv_sec = 0;
2060         inode->i_ctime.tv_sec = 0;
2061         inode->i_rdev = 0;
2062         rc = ll_update_inode(inode, md);
2063         if (rc != 0)
2064                 RETURN(rc);
2065
2066         /* OIDEBUG(inode); */
2067
2068 #ifdef HAVE_BACKING_DEV_INFO
2069         /* initializing backing dev info. */
2070         inode->i_mapping->backing_dev_info = &s2lsi(inode->i_sb)->lsi_bdi;
2071 #endif
2072         if (S_ISREG(inode->i_mode)) {
2073                 struct ll_sb_info *sbi = ll_i2sbi(inode);
2074                 inode->i_op = &ll_file_inode_operations;
2075                 inode->i_fop = sbi->ll_fop;
2076                 inode->i_mapping->a_ops = (struct address_space_operations *)&ll_aops;
2077                 EXIT;
2078         } else if (S_ISDIR(inode->i_mode)) {
2079                 inode->i_op = &ll_dir_inode_operations;
2080                 inode->i_fop = &ll_dir_operations;
2081                 EXIT;
2082         } else if (S_ISLNK(inode->i_mode)) {
2083                 inode->i_op = &ll_fast_symlink_inode_operations;
2084                 EXIT;
2085         } else {
2086                 inode->i_op = &ll_special_inode_operations;
2087
2088                 init_special_inode(inode, inode->i_mode,
2089                                    inode->i_rdev);
2090
2091                 EXIT;
2092         }
2093
2094         return 0;
2095 }
2096
2097 void ll_delete_inode(struct inode *inode)
2098 {
2099         struct ll_inode_info *lli = ll_i2info(inode);
2100         struct address_space *mapping = &inode->i_data;
2101         unsigned long nrpages;
2102         ENTRY;
2103
2104         if (S_ISREG(inode->i_mode) && lli->lli_clob != NULL)
2105                 /* It is last chance to write out dirty pages,
2106                  * otherwise we may lose data while umount */
2107                 cl_sync_file_range(inode, 0, OBD_OBJECT_EOF, CL_FSYNC_LOCAL, 1);
2108
2109         truncate_inode_pages_final(mapping);
2110
2111         /* Workaround for LU-118: Note nrpages may not be totally updated when
2112          * truncate_inode_pages() returns, as there can be a page in the process
2113          * of deletion (inside __delete_from_page_cache()) in the specified
2114          * range. Thus mapping->nrpages can be non-zero when this function
2115          * returns even after truncation of the whole mapping.  Only do this if
2116          * npages isn't already zero.
2117          */
2118         nrpages = mapping->nrpages;
2119         if (nrpages) {
2120                 spin_lock_irq(&mapping->tree_lock);
2121                 nrpages = mapping->nrpages;
2122                 spin_unlock_irq(&mapping->tree_lock);
2123         } /* Workaround end */
2124
2125         LASSERTF(nrpages == 0, "%s: inode="DFID"(%p) nrpages=%lu, "
2126                  "see https://jira.whamcloud.com/browse/LU-118\n",
2127                  ll_get_fsname(inode->i_sb, NULL, 0),
2128                  PFID(ll_inode2fid(inode)), inode, nrpages);
2129
2130 #ifdef HAVE_SBOPS_EVICT_INODE
2131         ll_clear_inode(inode);
2132 #endif
2133         clear_inode(inode);
2134
2135         EXIT;
2136 }
2137
2138 int ll_iocontrol(struct inode *inode, struct file *file,
2139                  unsigned int cmd, unsigned long arg)
2140 {
2141         struct ll_sb_info *sbi = ll_i2sbi(inode);
2142         struct ptlrpc_request *req = NULL;
2143         int rc, flags = 0;
2144         ENTRY;
2145
2146         switch (cmd) {
2147         case FS_IOC_GETFLAGS: {
2148                 struct mdt_body *body;
2149                 struct md_op_data *op_data;
2150
2151                 op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL,
2152                                              0, 0, LUSTRE_OPC_ANY,
2153                                              NULL);
2154                 if (IS_ERR(op_data))
2155                         RETURN(PTR_ERR(op_data));
2156
2157                 op_data->op_valid = OBD_MD_FLFLAGS;
2158                 rc = md_getattr(sbi->ll_md_exp, op_data, &req);
2159                 ll_finish_md_op_data(op_data);
2160                 if (rc) {
2161                         CERROR("%s: failure inode "DFID": rc = %d\n",
2162                                sbi->ll_md_exp->exp_obd->obd_name,
2163                                PFID(ll_inode2fid(inode)), rc);
2164                         RETURN(-abs(rc));
2165                 }
2166
2167                 body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
2168
2169                 flags = body->mbo_flags;
2170
2171                 ptlrpc_req_finished(req);
2172
2173                 RETURN(put_user(flags, (int __user *)arg));
2174         }
2175         case FS_IOC_SETFLAGS: {
2176                 struct iattr *attr;
2177                 struct md_op_data *op_data;
2178                 struct cl_object *obj;
2179                 struct fsxattr fa = { 0 };
2180
2181                 if (get_user(flags, (int __user *)arg))
2182                         RETURN(-EFAULT);
2183
2184                 fa.fsx_projid = ll_i2info(inode)->lli_projid;
2185                 if (flags & LUSTRE_PROJINHERIT_FL)
2186                         fa.fsx_xflags = FS_XFLAG_PROJINHERIT;
2187
2188                 rc = ll_ioctl_check_project(inode, &fa);
2189                 if (rc)
2190                         RETURN(rc);
2191
2192                 op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL, 0, 0,
2193                                              LUSTRE_OPC_ANY, NULL);
2194                 if (IS_ERR(op_data))
2195                         RETURN(PTR_ERR(op_data));
2196
2197                 op_data->op_attr_flags = flags;
2198                 op_data->op_xvalid |= OP_XVALID_FLAGS;
2199                 rc = md_setattr(sbi->ll_md_exp, op_data, NULL, 0, &req);
2200                 ll_finish_md_op_data(op_data);
2201                 ptlrpc_req_finished(req);
2202                 if (rc)
2203                         RETURN(rc);
2204
2205                 ll_update_inode_flags(inode, flags);
2206
2207                 obj = ll_i2info(inode)->lli_clob;
2208                 if (obj == NULL)
2209                         RETURN(0);
2210
2211                 OBD_ALLOC_PTR(attr);
2212                 if (attr == NULL)
2213                         RETURN(-ENOMEM);
2214
2215                 rc = cl_setattr_ost(obj, attr, OP_XVALID_FLAGS, flags);
2216
2217                 OBD_FREE_PTR(attr);
2218                 RETURN(rc);
2219         }
2220         default:
2221                 RETURN(-ENOSYS);
2222         }
2223
2224         RETURN(0);
2225 }
2226
2227 int ll_flush_ctx(struct inode *inode)
2228 {
2229         struct ll_sb_info  *sbi = ll_i2sbi(inode);
2230
2231         CDEBUG(D_SEC, "flush context for user %d\n",
2232                from_kuid(&init_user_ns, current_uid()));
2233
2234         obd_set_info_async(NULL, sbi->ll_md_exp,
2235                            sizeof(KEY_FLUSH_CTX), KEY_FLUSH_CTX,
2236                            0, NULL, NULL);
2237         obd_set_info_async(NULL, sbi->ll_dt_exp,
2238                            sizeof(KEY_FLUSH_CTX), KEY_FLUSH_CTX,
2239                            0, NULL, NULL);
2240         return 0;
2241 }
2242
2243 /* umount -f client means force down, don't save state */
2244 void ll_umount_begin(struct super_block *sb)
2245 {
2246         struct ll_sb_info *sbi = ll_s2sbi(sb);
2247         struct obd_device *obd;
2248         struct obd_ioctl_data *ioc_data;
2249         struct l_wait_info lwi;
2250         wait_queue_head_t waitq;
2251         ENTRY;
2252
2253         CDEBUG(D_VFSTRACE, "VFS Op: superblock %p count %d active %d\n", sb,
2254                sb->s_count, atomic_read(&sb->s_active));
2255
2256         obd = class_exp2obd(sbi->ll_md_exp);
2257         if (obd == NULL) {
2258                 CERROR("Invalid MDC connection handle %#llx\n",
2259                        sbi->ll_md_exp->exp_handle.h_cookie);
2260                 EXIT;
2261                 return;
2262         }
2263         obd->obd_force = 1;
2264
2265         obd = class_exp2obd(sbi->ll_dt_exp);
2266         if (obd == NULL) {
2267                 CERROR("Invalid LOV connection handle %#llx\n",
2268                        sbi->ll_dt_exp->exp_handle.h_cookie);
2269                 EXIT;
2270                 return;
2271         }
2272         obd->obd_force = 1;
2273
2274         OBD_ALLOC_PTR(ioc_data);
2275         if (ioc_data) {
2276                 obd_iocontrol(IOC_OSC_SET_ACTIVE, sbi->ll_md_exp,
2277                               sizeof *ioc_data, ioc_data, NULL);
2278
2279                 obd_iocontrol(IOC_OSC_SET_ACTIVE, sbi->ll_dt_exp,
2280                               sizeof *ioc_data, ioc_data, NULL);
2281
2282                 OBD_FREE_PTR(ioc_data);
2283         }
2284
2285         /* Really, we'd like to wait until there are no requests outstanding,
2286          * and then continue.  For now, we just periodically checking for vfs
2287          * to decrement mnt_cnt and hope to finish it within 10sec.
2288          */
2289         init_waitqueue_head(&waitq);
2290         lwi = LWI_TIMEOUT_INTERVAL(cfs_time_seconds(10),
2291                                    cfs_time_seconds(1), NULL, NULL);
2292         l_wait_event(waitq, may_umount(sbi->ll_mnt.mnt), &lwi);
2293
2294         EXIT;
2295 }
2296
2297 int ll_remount_fs(struct super_block *sb, int *flags, char *data)
2298 {
2299         struct ll_sb_info *sbi = ll_s2sbi(sb);
2300         char *profilenm = get_profile_name(sb);
2301         int err;
2302         __u32 read_only;
2303
2304         if ((*flags & MS_RDONLY) != (sb->s_flags & MS_RDONLY)) {
2305                 read_only = *flags & MS_RDONLY;
2306                 err = obd_set_info_async(NULL, sbi->ll_md_exp,
2307                                          sizeof(KEY_READ_ONLY),
2308                                          KEY_READ_ONLY, sizeof(read_only),
2309                                          &read_only, NULL);
2310                 if (err) {
2311                         LCONSOLE_WARN("Failed to remount %s %s (%d)\n",
2312                                       profilenm, read_only ?
2313                                       "read-only" : "read-write", err);
2314                         return err;
2315                 }
2316
2317                 if (read_only)
2318                         sb->s_flags |= MS_RDONLY;
2319                 else
2320                         sb->s_flags &= ~MS_RDONLY;
2321
2322                 if (sbi->ll_flags & LL_SBI_VERBOSE)
2323                         LCONSOLE_WARN("Remounted %s %s\n", profilenm,
2324                                       read_only ?  "read-only" : "read-write");
2325         }
2326         return 0;
2327 }
2328
2329 /**
2330  * Cleanup the open handle that is cached on MDT-side.
2331  *
2332  * For open case, the client side open handling thread may hit error
2333  * after the MDT grant the open. Under such case, the client should
2334  * send close RPC to the MDT as cleanup; otherwise, the open handle
2335  * on the MDT will be leaked there until the client umount or evicted.
2336  *
2337  * In further, if someone unlinked the file, because the open handle
2338  * holds the reference on such file/object, then it will block the
2339  * subsequent threads that want to locate such object via FID.
2340  *
2341  * \param[in] sb        super block for this file-system
2342  * \param[in] open_req  pointer to the original open request
2343  */
2344 void ll_open_cleanup(struct super_block *sb, struct ptlrpc_request *open_req)
2345 {
2346         struct mdt_body                 *body;
2347         struct md_op_data               *op_data;
2348         struct ptlrpc_request           *close_req = NULL;
2349         struct obd_export               *exp       = ll_s2sbi(sb)->ll_md_exp;
2350         ENTRY;
2351
2352         body = req_capsule_server_get(&open_req->rq_pill, &RMF_MDT_BODY);
2353         OBD_ALLOC_PTR(op_data);
2354         if (op_data == NULL) {
2355                 CWARN("%s: cannot allocate op_data to release open handle for "
2356                       DFID"\n",
2357                       ll_get_fsname(sb, NULL, 0), PFID(&body->mbo_fid1));
2358
2359                 RETURN_EXIT;
2360         }
2361
2362         op_data->op_fid1 = body->mbo_fid1;
2363         op_data->op_open_handle = body->mbo_open_handle;
2364         op_data->op_mod_time = ktime_get_real_seconds();
2365         md_close(exp, op_data, NULL, &close_req);
2366         ptlrpc_req_finished(close_req);
2367         ll_finish_md_op_data(op_data);
2368
2369         EXIT;
2370 }
2371
2372 int ll_prep_inode(struct inode **inode, struct ptlrpc_request *req,
2373                   struct super_block *sb, struct lookup_intent *it)
2374 {
2375         struct ll_sb_info *sbi = NULL;
2376         struct lustre_md md = { NULL };
2377         int rc;
2378         ENTRY;
2379
2380         LASSERT(*inode || sb);
2381         sbi = sb ? ll_s2sbi(sb) : ll_i2sbi(*inode);
2382         rc = md_get_lustre_md(sbi->ll_md_exp, req, sbi->ll_dt_exp,
2383                               sbi->ll_md_exp, &md);
2384         if (rc != 0)
2385                 GOTO(cleanup, rc);
2386
2387         if (*inode) {
2388                 rc = ll_update_inode(*inode, &md);
2389                 if (rc != 0)
2390                         GOTO(out, rc);
2391         } else {
2392                 LASSERT(sb != NULL);
2393
2394                 /*
2395                  * At this point server returns to client's same fid as client
2396                  * generated for creating. So using ->fid1 is okay here.
2397                  */
2398                 if (!fid_is_sane(&md.body->mbo_fid1)) {
2399                         CERROR("%s: Fid is insane "DFID"\n",
2400                                 ll_get_fsname(sb, NULL, 0),
2401                                 PFID(&md.body->mbo_fid1));
2402                         GOTO(out, rc = -EINVAL);
2403                 }
2404
2405                 *inode = ll_iget(sb, cl_fid_build_ino(&md.body->mbo_fid1,
2406                                              sbi->ll_flags & LL_SBI_32BIT_API),
2407                                  &md);
2408                 if (IS_ERR(*inode)) {
2409 #ifdef CONFIG_FS_POSIX_ACL
2410                         if (md.posix_acl) {
2411                                 posix_acl_release(md.posix_acl);
2412                                 md.posix_acl = NULL;
2413                         }
2414 #endif
2415                         rc = IS_ERR(*inode) ? PTR_ERR(*inode) : -ENOMEM;
2416                         *inode = NULL;
2417                         CERROR("new_inode -fatal: rc %d\n", rc);
2418                         GOTO(out, rc);
2419                 }
2420         }
2421
2422         /* Handling piggyback layout lock.
2423          * Layout lock can be piggybacked by getattr and open request.
2424          * The lsm can be applied to inode only if it comes with a layout lock
2425          * otherwise correct layout may be overwritten, for example:
2426          * 1. proc1: mdt returns a lsm but not granting layout
2427          * 2. layout was changed by another client
2428          * 3. proc2: refresh layout and layout lock granted
2429          * 4. proc1: to apply a stale layout */
2430         if (it != NULL && it->it_lock_mode != 0) {
2431                 struct lustre_handle lockh;
2432                 struct ldlm_lock *lock;
2433
2434                 lockh.cookie = it->it_lock_handle;
2435                 lock = ldlm_handle2lock(&lockh);
2436                 LASSERT(lock != NULL);
2437                 if (ldlm_has_layout(lock)) {
2438                         struct cl_object_conf conf;
2439
2440                         memset(&conf, 0, sizeof(conf));
2441                         conf.coc_opc = OBJECT_CONF_SET;
2442                         conf.coc_inode = *inode;
2443                         conf.coc_lock = lock;
2444                         conf.u.coc_layout = md.layout;
2445                         (void)ll_layout_conf(*inode, &conf);
2446                 }
2447                 LDLM_LOCK_PUT(lock);
2448         }
2449
2450         GOTO(out, rc = 0);
2451
2452 out:
2453         md_free_lustre_md(sbi->ll_md_exp, &md);
2454
2455 cleanup:
2456         if (rc != 0 && it != NULL && it->it_op & IT_OPEN)
2457                 ll_open_cleanup(sb != NULL ? sb : (*inode)->i_sb, req);
2458
2459         return rc;
2460 }
2461
2462 int ll_obd_statfs(struct inode *inode, void __user *arg)
2463 {
2464         struct ll_sb_info *sbi = NULL;
2465         struct obd_export *exp;
2466         char *buf = NULL;
2467         struct obd_ioctl_data *data = NULL;
2468         __u32 type;
2469         int len = 0, rc;
2470
2471         if (!inode || !(sbi = ll_i2sbi(inode)))
2472                 GOTO(out_statfs, rc = -EINVAL);
2473
2474         rc = obd_ioctl_getdata(&buf, &len, arg);
2475         if (rc)
2476                 GOTO(out_statfs, rc);
2477
2478         data = (void*)buf;
2479         if (!data->ioc_inlbuf1 || !data->ioc_inlbuf2 ||
2480             !data->ioc_pbuf1 || !data->ioc_pbuf2)
2481                 GOTO(out_statfs, rc = -EINVAL);
2482
2483         if (data->ioc_inllen1 != sizeof(__u32) ||
2484             data->ioc_inllen2 != sizeof(__u32) ||
2485             data->ioc_plen1 != sizeof(struct obd_statfs) ||
2486             data->ioc_plen2 != sizeof(struct obd_uuid))
2487                 GOTO(out_statfs, rc = -EINVAL);
2488
2489         memcpy(&type, data->ioc_inlbuf1, sizeof(__u32));
2490         if (type & LL_STATFS_LMV)
2491                 exp = sbi->ll_md_exp;
2492         else if (type & LL_STATFS_LOV)
2493                 exp = sbi->ll_dt_exp;
2494         else
2495                 GOTO(out_statfs, rc = -ENODEV);
2496
2497         rc = obd_iocontrol(IOC_OBD_STATFS, exp, len, buf, NULL);
2498         if (rc)
2499                 GOTO(out_statfs, rc);
2500 out_statfs:
2501         OBD_FREE_LARGE(buf, len);
2502         return rc;
2503 }
2504
2505 /*
2506  * this is normally called in ll_fini_md_op_data(), but sometimes it needs to
2507  * be called early to avoid deadlock.
2508  */
2509 void ll_unlock_md_op_lsm(struct md_op_data *op_data)
2510 {
2511         if (op_data->op_mea2_sem) {
2512                 up_read(op_data->op_mea2_sem);
2513                 op_data->op_mea2_sem = NULL;
2514         }
2515
2516         if (op_data->op_mea1_sem) {
2517                 up_read(op_data->op_mea1_sem);
2518                 op_data->op_mea1_sem = NULL;
2519         }
2520 }
2521
2522 /* this function prepares md_op_data hint for passing it down to MD stack. */
2523 struct md_op_data *ll_prep_md_op_data(struct md_op_data *op_data,
2524                                       struct inode *i1, struct inode *i2,
2525                                       const char *name, size_t namelen,
2526                                       __u32 mode, __u32 opc, void *data)
2527 {
2528         LASSERT(i1 != NULL);
2529
2530         if (name == NULL) {
2531                 /* Do not reuse namelen for something else. */
2532                 if (namelen != 0)
2533                         return ERR_PTR(-EINVAL);
2534         } else {
2535                 if (namelen > ll_i2sbi(i1)->ll_namelen)
2536                         return ERR_PTR(-ENAMETOOLONG);
2537
2538                 if (!lu_name_is_valid_2(name, namelen))
2539                         return ERR_PTR(-EINVAL);
2540         }
2541
2542         if (op_data == NULL)
2543                 OBD_ALLOC_PTR(op_data);
2544
2545         if (op_data == NULL)
2546                 return ERR_PTR(-ENOMEM);
2547
2548         ll_i2gids(op_data->op_suppgids, i1, i2);
2549         op_data->op_fid1 = *ll_inode2fid(i1);
2550         op_data->op_default_stripe_offset = -1;
2551
2552         if (S_ISDIR(i1->i_mode)) {
2553                 down_read(&ll_i2info(i1)->lli_lsm_sem);
2554                 op_data->op_mea1_sem = &ll_i2info(i1)->lli_lsm_sem;
2555                 op_data->op_mea1 = ll_i2info(i1)->lli_lsm_md;
2556                 if (opc == LUSTRE_OPC_MKDIR)
2557                         op_data->op_default_stripe_offset =
2558                                    ll_i2info(i1)->lli_def_stripe_offset;
2559         }
2560
2561         if (i2) {
2562                 op_data->op_fid2 = *ll_inode2fid(i2);
2563                 if (S_ISDIR(i2->i_mode)) {
2564                         if (i2 != i1) {
2565                                 down_read(&ll_i2info(i2)->lli_lsm_sem);
2566                                 op_data->op_mea2_sem =
2567                                                 &ll_i2info(i2)->lli_lsm_sem;
2568                         }
2569                         op_data->op_mea2 = ll_i2info(i2)->lli_lsm_md;
2570                 }
2571         } else {
2572                 fid_zero(&op_data->op_fid2);
2573         }
2574
2575         if (ll_i2sbi(i1)->ll_flags & LL_SBI_64BIT_HASH)
2576                 op_data->op_cli_flags |= CLI_HASH64;
2577
2578         if (ll_need_32bit_api(ll_i2sbi(i1)))
2579                 op_data->op_cli_flags |= CLI_API32;
2580
2581         op_data->op_name = name;
2582         op_data->op_namelen = namelen;
2583         op_data->op_mode = mode;
2584         op_data->op_mod_time = ktime_get_real_seconds();
2585         op_data->op_fsuid = from_kuid(&init_user_ns, current_fsuid());
2586         op_data->op_fsgid = from_kgid(&init_user_ns, current_fsgid());
2587         op_data->op_cap = cfs_curproc_cap_pack();
2588         op_data->op_mds = 0;
2589         if ((opc == LUSTRE_OPC_CREATE) && (name != NULL) &&
2590              filename_is_volatile(name, namelen, &op_data->op_mds)) {
2591                 op_data->op_bias |= MDS_CREATE_VOLATILE;
2592         }
2593         op_data->op_data = data;
2594
2595         return op_data;
2596 }
2597
2598 void ll_finish_md_op_data(struct md_op_data *op_data)
2599 {
2600         ll_unlock_md_op_lsm(op_data);
2601         security_release_secctx(op_data->op_file_secctx,
2602                                 op_data->op_file_secctx_size);
2603         OBD_FREE_PTR(op_data);
2604 }
2605
2606 #ifdef HAVE_SUPEROPS_USE_DENTRY
2607 int ll_show_options(struct seq_file *seq, struct dentry *dentry)
2608 #else
2609 int ll_show_options(struct seq_file *seq, struct vfsmount *vfs)
2610 #endif
2611 {
2612         struct ll_sb_info *sbi;
2613
2614 #ifdef HAVE_SUPEROPS_USE_DENTRY
2615         LASSERT((seq != NULL) && (dentry != NULL));
2616         sbi = ll_s2sbi(dentry->d_sb);
2617 #else
2618         LASSERT((seq != NULL) && (vfs != NULL));
2619         sbi = ll_s2sbi(vfs->mnt_sb);
2620 #endif
2621
2622         if (sbi->ll_flags & LL_SBI_NOLCK)
2623                 seq_puts(seq, ",nolock");
2624
2625         if (sbi->ll_flags & LL_SBI_FLOCK)
2626                 seq_puts(seq, ",flock");
2627
2628         if (sbi->ll_flags & LL_SBI_LOCALFLOCK)
2629                 seq_puts(seq, ",localflock");
2630
2631         if (sbi->ll_flags & LL_SBI_USER_XATTR)
2632                 seq_puts(seq, ",user_xattr");
2633
2634         if (sbi->ll_flags & LL_SBI_LAZYSTATFS)
2635                 seq_puts(seq, ",lazystatfs");
2636
2637         if (sbi->ll_flags & LL_SBI_USER_FID2PATH)
2638                 seq_puts(seq, ",user_fid2path");
2639
2640         if (sbi->ll_flags & LL_SBI_ALWAYS_PING)
2641                 seq_puts(seq, ",always_ping");
2642
2643         RETURN(0);
2644 }
2645
2646 /**
2647  * Get obd name by cmd, and copy out to user space
2648  */
2649 int ll_get_obd_name(struct inode *inode, unsigned int cmd, unsigned long arg)
2650 {
2651         struct ll_sb_info *sbi = ll_i2sbi(inode);
2652         struct obd_device *obd;
2653         ENTRY;
2654
2655         if (cmd == OBD_IOC_GETDTNAME)
2656                 obd = class_exp2obd(sbi->ll_dt_exp);
2657         else if (cmd == OBD_IOC_GETMDNAME)
2658                 obd = class_exp2obd(sbi->ll_md_exp);
2659         else
2660                 RETURN(-EINVAL);
2661
2662         if (!obd)
2663                 RETURN(-ENOENT);
2664
2665         if (copy_to_user((void __user *)arg, obd->obd_name,
2666                          strlen(obd->obd_name) + 1))
2667                 RETURN(-EFAULT);
2668
2669         RETURN(0);
2670 }
2671
2672 /**
2673  * Get lustre file system name by \a sbi. If \a buf is provided(non-NULL), the
2674  * fsname will be returned in this buffer; otherwise, a static buffer will be
2675  * used to store the fsname and returned to caller.
2676  */
2677 char *ll_get_fsname(struct super_block *sb, char *buf, int buflen)
2678 {
2679         static char fsname_static[MTI_NAME_MAXLEN];
2680         struct lustre_sb_info *lsi = s2lsi(sb);
2681         char *ptr;
2682         int len;
2683
2684         if (buf == NULL) {
2685                 /* this means the caller wants to use static buffer
2686                  * and it doesn't care about race. Usually this is
2687                  * in error reporting path */
2688                 buf = fsname_static;
2689                 buflen = sizeof(fsname_static);
2690         }
2691
2692         len = strlen(lsi->lsi_lmd->lmd_profile);
2693         ptr = strrchr(lsi->lsi_lmd->lmd_profile, '-');
2694         if (ptr && (strcmp(ptr, "-client") == 0))
2695                 len -= 7;
2696
2697         if (unlikely(len >= buflen))
2698                 len = buflen - 1;
2699         strncpy(buf, lsi->lsi_lmd->lmd_profile, len);
2700         buf[len] = '\0';
2701
2702         return buf;
2703 }
2704
2705 static char* ll_d_path(struct dentry *dentry, char *buf, int bufsize)
2706 {
2707         char *path = NULL;
2708
2709         struct path p;
2710
2711         p.dentry = dentry;
2712         p.mnt = current->fs->root.mnt;
2713         path_get(&p);
2714         path = d_path(&p, buf, bufsize);
2715         path_put(&p);
2716         return path;
2717 }
2718
2719 void ll_dirty_page_discard_warn(struct page *page, int ioret)
2720 {
2721         char *buf, *path = NULL;
2722         struct dentry *dentry = NULL;
2723         struct inode *inode = page->mapping->host;
2724
2725         /* this can be called inside spin lock so use GFP_ATOMIC. */
2726         buf = (char *)__get_free_page(GFP_ATOMIC);
2727         if (buf != NULL) {
2728                 dentry = d_find_alias(page->mapping->host);
2729                 if (dentry != NULL)
2730                         path = ll_d_path(dentry, buf, PAGE_SIZE);
2731         }
2732
2733         CDEBUG(D_WARNING,
2734                "%s: dirty page discard: %s/fid: "DFID"/%s may get corrupted "
2735                "(rc %d)\n", ll_get_fsname(page->mapping->host->i_sb, NULL, 0),
2736                s2lsi(page->mapping->host->i_sb)->lsi_lmd->lmd_dev,
2737                PFID(ll_inode2fid(inode)),
2738                (path && !IS_ERR(path)) ? path : "", ioret);
2739
2740         if (dentry != NULL)
2741                 dput(dentry);
2742
2743         if (buf != NULL)
2744                 free_page((unsigned long)buf);
2745 }
2746
2747 ssize_t ll_copy_user_md(const struct lov_user_md __user *md,
2748                         struct lov_user_md **kbuf)
2749 {
2750         struct lov_user_md      lum;
2751         ssize_t                 lum_size;
2752         ENTRY;
2753
2754         if (copy_from_user(&lum, md, sizeof(lum)))
2755                 RETURN(-EFAULT);
2756
2757         lum_size = ll_lov_user_md_size(&lum);
2758         if (lum_size < 0)
2759                 RETURN(lum_size);
2760
2761         OBD_ALLOC(*kbuf, lum_size);
2762         if (*kbuf == NULL)
2763                 RETURN(-ENOMEM);
2764
2765         if (copy_from_user(*kbuf, md, lum_size) != 0) {
2766                 OBD_FREE(*kbuf, lum_size);
2767                 RETURN(-EFAULT);
2768         }
2769
2770         RETURN(lum_size);
2771 }
2772
2773 /*
2774  * Compute llite root squash state after a change of root squash
2775  * configuration setting or add/remove of a lnet nid
2776  */
2777 void ll_compute_rootsquash_state(struct ll_sb_info *sbi)
2778 {
2779         struct root_squash_info *squash = &sbi->ll_squash;
2780         int i;
2781         bool matched;
2782         struct lnet_process_id id;
2783
2784         /* Update norootsquash flag */
2785         down_write(&squash->rsi_sem);
2786         if (list_empty(&squash->rsi_nosquash_nids))
2787                 sbi->ll_flags &= ~LL_SBI_NOROOTSQUASH;
2788         else {
2789                 /* Do not apply root squash as soon as one of our NIDs is
2790                  * in the nosquash_nids list */
2791                 matched = false;
2792                 i = 0;
2793                 while (LNetGetId(i++, &id) != -ENOENT) {
2794                         if (LNET_NETTYP(LNET_NIDNET(id.nid)) == LOLND)
2795                                 continue;
2796                         if (cfs_match_nid(id.nid, &squash->rsi_nosquash_nids)) {
2797                                 matched = true;
2798                                 break;
2799                         }
2800                 }
2801                 if (matched)
2802                         sbi->ll_flags |= LL_SBI_NOROOTSQUASH;
2803                 else
2804                         sbi->ll_flags &= ~LL_SBI_NOROOTSQUASH;
2805         }
2806         up_write(&squash->rsi_sem);
2807 }
2808
2809 /**
2810  * Parse linkea content to extract information about a given hardlink
2811  *
2812  * \param[in]   ldata      - Initialized linkea data
2813  * \param[in]   linkno     - Link identifier
2814  * \param[out]  parent_fid - The entry's parent FID
2815  * \param[out]  ln         - Entry name destination buffer
2816  *
2817  * \retval 0 on success
2818  * \retval Appropriate negative error code on failure
2819  */
2820 static int ll_linkea_decode(struct linkea_data *ldata, unsigned int linkno,
2821                             struct lu_fid *parent_fid, struct lu_name *ln)
2822 {
2823         unsigned int    idx;
2824         int             rc;
2825         ENTRY;
2826
2827         rc = linkea_init_with_rec(ldata);
2828         if (rc < 0)
2829                 RETURN(rc);
2830
2831         if (linkno >= ldata->ld_leh->leh_reccount)
2832                 /* beyond last link */
2833                 RETURN(-ENODATA);
2834
2835         linkea_first_entry(ldata);
2836         for (idx = 0; ldata->ld_lee != NULL; idx++) {
2837                 linkea_entry_unpack(ldata->ld_lee, &ldata->ld_reclen, ln,
2838                                     parent_fid);
2839                 if (idx == linkno)
2840                         break;
2841
2842                 linkea_next_entry(ldata);
2843         }
2844
2845         if (idx < linkno)
2846                 RETURN(-ENODATA);
2847
2848         RETURN(0);
2849 }
2850
2851 /**
2852  * Get parent FID and name of an identified link. Operation is performed for
2853  * a given link number, letting the caller iterate over linkno to list one or
2854  * all links of an entry.
2855  *
2856  * \param[in]     file - File descriptor against which to perform the operation
2857  * \param[in,out] arg  - User-filled structure containing the linkno to operate
2858  *                       on and the available size. It is eventually filled with
2859  *                       the requested information or left untouched on error
2860  *
2861  * \retval - 0 on success
2862  * \retval - Appropriate negative error code on failure
2863  */
2864 int ll_getparent(struct file *file, struct getparent __user *arg)
2865 {
2866         struct inode            *inode = file_inode(file);
2867         struct linkea_data      *ldata;
2868         struct lu_buf            buf = LU_BUF_NULL;
2869         struct lu_name           ln;
2870         struct lu_fid            parent_fid;
2871         __u32                    linkno;
2872         __u32                    name_size;
2873         int                      rc;
2874
2875         ENTRY;
2876
2877         if (!cfs_capable(CFS_CAP_DAC_READ_SEARCH) &&
2878             !(ll_i2sbi(inode)->ll_flags & LL_SBI_USER_FID2PATH))
2879                 RETURN(-EPERM);
2880
2881         if (get_user(name_size, &arg->gp_name_size))
2882                 RETURN(-EFAULT);
2883
2884         if (get_user(linkno, &arg->gp_linkno))
2885                 RETURN(-EFAULT);
2886
2887         if (name_size > PATH_MAX)
2888                 RETURN(-EINVAL);
2889
2890         OBD_ALLOC(ldata, sizeof(*ldata));
2891         if (ldata == NULL)
2892                 RETURN(-ENOMEM);
2893
2894         rc = linkea_data_new(ldata, &buf);
2895         if (rc < 0)
2896                 GOTO(ldata_free, rc);
2897
2898 #ifdef HAVE_XATTR_HANDLER_FLAGS
2899         rc = ll_xattr_list(inode, XATTR_NAME_LINK, XATTR_TRUSTED_T, buf.lb_buf,
2900                            buf.lb_len, OBD_MD_FLXATTR);
2901 #else
2902         rc = ll_getxattr(file_dentry(file), XATTR_NAME_LINK, buf.lb_buf,
2903                          buf.lb_len);
2904 #endif /* HAVE_XATTR_HANDLER_FLAGS */
2905         if (rc < 0)
2906                 GOTO(lb_free, rc);
2907
2908         rc = ll_linkea_decode(ldata, linkno, &parent_fid, &ln);
2909         if (rc < 0)
2910                 GOTO(lb_free, rc);
2911
2912         if (ln.ln_namelen >= name_size)
2913                 GOTO(lb_free, rc = -EOVERFLOW);
2914
2915         if (copy_to_user(&arg->gp_fid, &parent_fid, sizeof(arg->gp_fid)))
2916                 GOTO(lb_free, rc = -EFAULT);
2917
2918         if (copy_to_user(&arg->gp_name, ln.ln_name, ln.ln_namelen))
2919                 GOTO(lb_free, rc = -EFAULT);
2920
2921         if (put_user('\0', arg->gp_name + ln.ln_namelen))
2922                 GOTO(lb_free, rc = -EFAULT);
2923
2924 lb_free:
2925         lu_buf_free(&buf);
2926 ldata_free:
2927         OBD_FREE(ldata, sizeof(*ldata));
2928
2929         RETURN(rc);
2930 }