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/random.h>
  41 #include <linux/statfs.h>
  42 #include <linux/time.h>
  43 #include <linux/types.h>
  44 #include <libcfs/linux/linux-uuid.h>
  45 #include <linux/version.h>
  46 #include <linux/mm.h>
  47 #include <linux/user_namespace.h>
  48 #ifdef HAVE_UIDGID_HEADER
  49 # include <linux/uidgid.h>
  50 #endif
  51 #include <linux/security.h>
  52
  53 #include <uapi/linux/lustre/lustre_ioctl.h>
  54 #ifdef HAVE_UAPI_LINUX_MOUNT_H
  55 #include <uapi/linux/mount.h>
  56 #endif
  57
  58 #include <lustre_ha.h>
  59 #include <lustre_dlm.h>
  60 #include <lprocfs_status.h>
  61 #include <lustre_disk.h>
  62 #include <uapi/linux/lustre/lustre_param.h>
  63 #include <lustre_log.h>
  64 #include <cl_object.h>
  65 #include <obd_cksum.h>
  66 #include "llite_internal.h"
  67
  68 struct kmem_cache *ll_file_data_slab;
  69
  70 #ifndef log2
  71 #define log2(n) ffz(~(n))
  72 #endif
  73
  74 static struct ll_sb_info *ll_init_sbi(void)
  75 {
  76         struct ll_sb_info *sbi = NULL;
  77         unsigned long pages;
  78         unsigned long lru_page_max;
  79         struct sysinfo si;
  80         int rc;
  81         int i;
  82
  83         ENTRY;
  84
  85         OBD_ALLOC_PTR(sbi);
  86         if (sbi == NULL)
  87                 RETURN(ERR_PTR(-ENOMEM));
  88
  89         rc = pcc_super_init(&sbi->ll_pcc_super);
  90         if (rc < 0)
  91                 GOTO(out_sbi, rc);
  92
  93         spin_lock_init(&sbi->ll_lock);
  94         mutex_init(&sbi->ll_lco.lco_lock);
  95         spin_lock_init(&sbi->ll_pp_extent_lock);
  96         spin_lock_init(&sbi->ll_process_lock);
  97         sbi->ll_rw_stats_on = 0;
  98         sbi->ll_statfs_max_age = OBD_STATFS_CACHE_SECONDS;
  99
 100         si_meminfo(&si);
 101         pages = si.totalram - si.totalhigh;
 102         lru_page_max = pages / 2;
 103
 104         sbi->ll_ra_info.ra_async_max_active = 0;
 105         sbi->ll_ra_info.ll_readahead_wq =
 106                 alloc_workqueue("ll-readahead-wq", WQ_UNBOUND,
 107                                 sbi->ll_ra_info.ra_async_max_active);
 108         if (!sbi->ll_ra_info.ll_readahead_wq)
 109                 GOTO(out_pcc, rc = -ENOMEM);
 110
 111         /* initialize ll_cache data */
 112         sbi->ll_cache = cl_cache_init(lru_page_max);
 113         if (sbi->ll_cache == NULL)
 114                 GOTO(out_destroy_ra, rc = -ENOMEM);
 115
 116         sbi->ll_ra_info.ra_max_pages_per_file = min(pages / 32,
 117                                            SBI_DEFAULT_READAHEAD_MAX);
 118         sbi->ll_ra_info.ra_async_pages_per_file_threshold =
 119                                 sbi->ll_ra_info.ra_max_pages_per_file;
 120         sbi->ll_ra_info.ra_max_pages = sbi->ll_ra_info.ra_max_pages_per_file;
 121         sbi->ll_ra_info.ra_max_read_ahead_whole_pages = -1;
 122
 123         sbi->ll_flags |= LL_SBI_VERBOSE;
 124 #ifdef ENABLE_CHECKSUM
 125         sbi->ll_flags |= LL_SBI_CHECKSUM;
 126 #endif
 127 #ifdef ENABLE_FLOCK
 128         sbi->ll_flags |= LL_SBI_FLOCK;
 129 #endif
 130
 131 #ifdef HAVE_LRU_RESIZE_SUPPORT
 132         sbi->ll_flags |= LL_SBI_LRU_RESIZE;
 133 #endif
 134         sbi->ll_flags |= LL_SBI_LAZYSTATFS;
 135
 136         for (i = 0; i <= LL_PROCESS_HIST_MAX; i++) {
 137                 spin_lock_init(&sbi->ll_rw_extents_info.pp_extents[i].
 138                                pp_r_hist.oh_lock);
 139                 spin_lock_init(&sbi->ll_rw_extents_info.pp_extents[i].
 140                                pp_w_hist.oh_lock);
 141         }
 142
 143         /* metadata statahead is enabled by default */
 144         sbi->ll_sa_running_max = LL_SA_RUNNING_DEF;
 145         sbi->ll_sa_max = LL_SA_RPC_DEF;
 146         atomic_set(&sbi->ll_sa_total, 0);
 147         atomic_set(&sbi->ll_sa_wrong, 0);
 148         atomic_set(&sbi->ll_sa_running, 0);
 149         atomic_set(&sbi->ll_agl_total, 0);
 150         sbi->ll_flags |= LL_SBI_AGL_ENABLED;
 151         sbi->ll_flags |= LL_SBI_FAST_READ;
 152         sbi->ll_flags |= LL_SBI_TINY_WRITE;
 153
 154         /* root squash */
 155         sbi->ll_squash.rsi_uid = 0;
 156         sbi->ll_squash.rsi_gid = 0;
 157         INIT_LIST_HEAD(&sbi->ll_squash.rsi_nosquash_nids);
 158         spin_lock_init(&sbi->ll_squash.rsi_lock);
 159
 160         /* Per-filesystem file heat */
 161         sbi->ll_heat_decay_weight = SBI_DEFAULT_HEAT_DECAY_WEIGHT;
 162         sbi->ll_heat_period_second = SBI_DEFAULT_HEAT_PERIOD_SECOND;
 163         RETURN(sbi);
 164 out_destroy_ra:
 165         destroy_workqueue(sbi->ll_ra_info.ll_readahead_wq);
 166 out_pcc:
 167         pcc_super_fini(&sbi->ll_pcc_super);
 168 out_sbi:
 169         OBD_FREE_PTR(sbi);
 170         RETURN(ERR_PTR(rc));
 171 }
 172
 173 static void ll_free_sbi(struct super_block *sb)
 174 {
 175         struct ll_sb_info *sbi = ll_s2sbi(sb);
 176         ENTRY;
 177
 178         if (sbi != NULL) {
 179                 if (!list_empty(&sbi->ll_squash.rsi_nosquash_nids))
 180                         cfs_free_nidlist(&sbi->ll_squash.rsi_nosquash_nids);
 181                 if (sbi->ll_ra_info.ll_readahead_wq)
 182                         destroy_workqueue(sbi->ll_ra_info.ll_readahead_wq);
 183                 if (sbi->ll_cache != NULL) {
 184                         cl_cache_decref(sbi->ll_cache);
 185                         sbi->ll_cache = NULL;
 186                 }
 187                 pcc_super_fini(&sbi->ll_pcc_super);
 188                 OBD_FREE(sbi, sizeof(*sbi));
 189         }
 190         EXIT;
 191 }
 192
 193 static int client_common_fill_super(struct super_block *sb, char *md, char *dt,
 194                                     struct vfsmount *mnt)
 195 {
 196         struct inode *root = NULL;
 197         struct ll_sb_info *sbi = ll_s2sbi(sb);
 198         struct obd_statfs *osfs = NULL;
 199         struct ptlrpc_request *request = NULL;
 200         struct obd_connect_data *data = NULL;
 201         struct obd_uuid *uuid;
 202         struct md_op_data *op_data;
 203         struct lustre_md lmd;
 204         u64 valid;
 205         int size, err, checksum;
 206
 207         ENTRY;
 208         sbi->ll_md_obd = class_name2obd(md);
 209         if (!sbi->ll_md_obd) {
 210                 CERROR("MD %s: not setup or attached\n", md);
 211                 RETURN(-EINVAL);
 212         }
 213
 214         OBD_ALLOC_PTR(data);
 215         if (data == NULL)
 216                 RETURN(-ENOMEM);
 217
 218         OBD_ALLOC_PTR(osfs);
 219         if (osfs == NULL) {
 220                 OBD_FREE_PTR(data);
 221                 RETURN(-ENOMEM);
 222         }
 223
 224         /* pass client page size via ocd_grant_blkbits, the server should report
 225          * back its backend blocksize for grant calculation purpose */
 226         data->ocd_grant_blkbits = PAGE_SHIFT;
 227
 228         /* indicate MDT features supported by this client */
 229         data->ocd_connect_flags = OBD_CONNECT_IBITS    | OBD_CONNECT_NODEVOH  |
 230                                   OBD_CONNECT_ATTRFID  | OBD_CONNECT_GRANT |
 231                                   OBD_CONNECT_VERSION  | OBD_CONNECT_BRW_SIZE |
 232                                   OBD_CONNECT_SRVLOCK  | OBD_CONNECT_TRUNCLOCK|
 233                                   OBD_CONNECT_MDS_CAPA | OBD_CONNECT_OSS_CAPA |
 234                                   OBD_CONNECT_CANCELSET | OBD_CONNECT_FID     |
 235                                   OBD_CONNECT_AT       | OBD_CONNECT_LOV_V3   |
 236                                   OBD_CONNECT_VBR | OBD_CONNECT_FULL20 |
 237                                   OBD_CONNECT_64BITHASH |
 238                                   OBD_CONNECT_EINPROGRESS |
 239                                   OBD_CONNECT_JOBSTATS | OBD_CONNECT_LVB_TYPE |
 240                                   OBD_CONNECT_LAYOUTLOCK | OBD_CONNECT_PINGLESS|
 241                                   OBD_CONNECT_MAX_EASIZE |
 242                                   OBD_CONNECT_FLOCK_DEAD |
 243                                   OBD_CONNECT_DISP_STRIPE | OBD_CONNECT_LFSCK |
 244                                   OBD_CONNECT_OPEN_BY_FID |
 245                                   OBD_CONNECT_DIR_STRIPE |
 246                                   OBD_CONNECT_BULK_MBITS | OBD_CONNECT_CKSUM |
 247                                   OBD_CONNECT_SUBTREE |
 248                                   OBD_CONNECT_MULTIMODRPCS |
 249                                   OBD_CONNECT_GRANT_PARAM |
 250                                   OBD_CONNECT_SHORTIO | OBD_CONNECT_FLAGS2;
 251
 252         data->ocd_connect_flags2 = OBD_CONNECT2_DIR_MIGRATE |
 253                                    OBD_CONNECT2_SUM_STATFS |
 254                                    OBD_CONNECT2_OVERSTRIPING |
 255                                    OBD_CONNECT2_FLR |
 256                                    OBD_CONNECT2_LOCK_CONVERT |
 257                                    OBD_CONNECT2_ARCHIVE_ID_ARRAY |
 258                                    OBD_CONNECT2_INC_XID |
 259                                    OBD_CONNECT2_LSOM |
 260                                    OBD_CONNECT2_ASYNC_DISCARD |
 261                                    OBD_CONNECT2_PCC;
 262
 263 #ifdef HAVE_LRU_RESIZE_SUPPORT
 264         if (sbi->ll_flags & LL_SBI_LRU_RESIZE)
 265                 data->ocd_connect_flags |= OBD_CONNECT_LRU_RESIZE;
 266 #endif
 267 #ifdef CONFIG_LUSTRE_FS_POSIX_ACL
 268         data->ocd_connect_flags |= OBD_CONNECT_ACL | OBD_CONNECT_UMASK |
 269                                    OBD_CONNECT_LARGE_ACL;
 270 #endif
 271
 272         data->ocd_cksum_types = obd_cksum_types_supported_client();
 273
 274         if (OBD_FAIL_CHECK(OBD_FAIL_MDC_LIGHTWEIGHT))
 275                 /* flag mdc connection as lightweight, only used for test
 276                  * purpose, use with care */
 277                 data->ocd_connect_flags |= OBD_CONNECT_LIGHTWEIGHT;
 278
 279         data->ocd_ibits_known = MDS_INODELOCK_FULL;
 280         data->ocd_version = LUSTRE_VERSION_CODE;
 281
 282         if (sb->s_flags & SB_RDONLY)
 283                 data->ocd_connect_flags |= OBD_CONNECT_RDONLY;
 284         if (sbi->ll_flags & LL_SBI_USER_XATTR)
 285                 data->ocd_connect_flags |= OBD_CONNECT_XATTR;
 286
 287 #ifdef SB_NOSEC
 288         /* Setting this indicates we correctly support S_NOSEC (See kernel
 289          * commit 9e1f1de02c2275d7172e18dc4e7c2065777611bf)
 290          */
 291         sb->s_flags |= SB_NOSEC;
 292 #endif
 293
 294         if (sbi->ll_flags & LL_SBI_FLOCK)
 295                 sbi->ll_fop = &ll_file_operations_flock;
 296         else if (sbi->ll_flags & LL_SBI_LOCALFLOCK)
 297                 sbi->ll_fop = &ll_file_operations;
 298         else
 299                 sbi->ll_fop = &ll_file_operations_noflock;
 300
 301         /* always ping even if server suppress_pings */
 302         if (sbi->ll_flags & LL_SBI_ALWAYS_PING)
 303                 data->ocd_connect_flags &= ~OBD_CONNECT_PINGLESS;
 304
 305         obd_connect_set_secctx(data);
 306
 307 #if defined(CONFIG_SECURITY)
 308         data->ocd_connect_flags2 |= OBD_CONNECT2_SELINUX_POLICY;
 309 #endif
 310
 311         data->ocd_brw_size = MD_MAX_BRW_SIZE;
 312
 313         err = obd_connect(NULL, &sbi->ll_md_exp, sbi->ll_md_obd,
 314                           &sbi->ll_sb_uuid, data, sbi->ll_cache);
 315         if (err == -EBUSY) {
 316                 LCONSOLE_ERROR_MSG(0x14f, "An MDT (md %s) is performing "
 317                                    "recovery, of which this client is not a "
 318                                    "part. Please wait for recovery to complete,"
 319                                    " abort, or time out.\n", md);
 320                 GOTO(out, err);
 321         } else if (err) {
 322                 CERROR("cannot connect to %s: rc = %d\n", md, err);
 323                 GOTO(out, err);
 324         }
 325
 326         sbi->ll_md_exp->exp_connect_data = *data;
 327
 328         err = obd_fid_init(sbi->ll_md_exp->exp_obd, sbi->ll_md_exp,
 329                            LUSTRE_SEQ_METADATA);
 330         if (err) {
 331                 CERROR("%s: Can't init metadata layer FID infrastructure, "
 332                        "rc = %d\n", sbi->ll_md_exp->exp_obd->obd_name, err);
 333                 GOTO(out_md, err);
 334         }
 335
 336         /* For mount, we only need fs info from MDT0, and also in DNE, it
 337          * can make sure the client can be mounted as long as MDT0 is
 338          * avaible */
 339         err = obd_statfs(NULL, sbi->ll_md_exp, osfs,
 340                         ktime_get_seconds() - sbi->ll_statfs_max_age,
 341                         OBD_STATFS_FOR_MDT0);
 342         if (err)
 343                 GOTO(out_md_fid, err);
 344
 345         /* This needs to be after statfs to ensure connect has finished.
 346          * Note that "data" does NOT contain the valid connect reply.
 347          * If connecting to a 1.8 server there will be no LMV device, so
 348          * we can access the MDC export directly and exp_connect_flags will
 349          * be non-zero, but if accessing an upgraded 2.1 server it will
 350          * have the correct flags filled in.
 351          * XXX: fill in the LMV exp_connect_flags from MDC(s). */
 352         valid = exp_connect_flags(sbi->ll_md_exp) & CLIENT_CONNECT_MDT_REQD;
 353         if (exp_connect_flags(sbi->ll_md_exp) != 0 &&
 354             valid != CLIENT_CONNECT_MDT_REQD) {
 355                 char *buf;
 356
 357                 OBD_ALLOC_WAIT(buf, PAGE_SIZE);
 358                 obd_connect_flags2str(buf, PAGE_SIZE,
 359                                       valid ^ CLIENT_CONNECT_MDT_REQD, 0, ",");
 360                 LCONSOLE_ERROR_MSG(0x170, "Server %s does not support "
 361                                    "feature(s) needed for correct operation "
 362                                    "of this client (%s). Please upgrade "
 363                                    "server or downgrade client.\n",
 364                                    sbi->ll_md_exp->exp_obd->obd_name, buf);
 365                 OBD_FREE(buf, PAGE_SIZE);
 366                 GOTO(out_md_fid, err = -EPROTO);
 367         }
 368
 369         size = sizeof(*data);
 370         err = obd_get_info(NULL, sbi->ll_md_exp, sizeof(KEY_CONN_DATA),
 371                            KEY_CONN_DATA,  &size, data);
 372         if (err) {
 373                 CERROR("%s: Get connect data failed: rc = %d\n",
 374                        sbi->ll_md_exp->exp_obd->obd_name, err);
 375                 GOTO(out_md_fid, err);
 376         }
 377
 378         LASSERT(osfs->os_bsize);
 379         sb->s_blocksize = osfs->os_bsize;
 380         sb->s_blocksize_bits = log2(osfs->os_bsize);
 381         sb->s_magic = LL_SUPER_MAGIC;
 382         sb->s_maxbytes = MAX_LFS_FILESIZE;
 383         sbi->ll_namelen = osfs->os_namelen;
 384         sbi->ll_mnt.mnt = current->fs->root.mnt;
 385
 386         if ((sbi->ll_flags & LL_SBI_USER_XATTR) &&
 387             !(data->ocd_connect_flags & OBD_CONNECT_XATTR)) {
 388                 LCONSOLE_INFO("Disabling user_xattr feature because "
 389                               "it is not supported on the server\n");
 390                 sbi->ll_flags &= ~LL_SBI_USER_XATTR;
 391         }
 392
 393         if (data->ocd_connect_flags & OBD_CONNECT_ACL) {
 394 #ifdef SB_POSIXACL
 395                 sb->s_flags |= SB_POSIXACL;
 396 #endif
 397                 sbi->ll_flags |= LL_SBI_ACL;
 398         } else {
 399                 LCONSOLE_INFO("client wants to enable acl, but mdt not!\n");
 400 #ifdef SB_POSIXACL
 401                 sb->s_flags &= ~SB_POSIXACL;
 402 #endif
 403                 sbi->ll_flags &= ~LL_SBI_ACL;
 404         }
 405
 406         if (data->ocd_connect_flags & OBD_CONNECT_64BITHASH)
 407                 sbi->ll_flags |= LL_SBI_64BIT_HASH;
 408
 409         if (data->ocd_connect_flags & OBD_CONNECT_LAYOUTLOCK)
 410                 sbi->ll_flags |= LL_SBI_LAYOUT_LOCK;
 411
 412         if (obd_connect_has_secctx(data))
 413                 sbi->ll_flags |= LL_SBI_FILE_SECCTX;
 414
 415         if (data->ocd_ibits_known & MDS_INODELOCK_XATTR) {
 416                 if (!(data->ocd_connect_flags & OBD_CONNECT_MAX_EASIZE)) {
 417                         LCONSOLE_INFO("%s: disabling xattr cache due to "
 418                                       "unknown maximum xattr size.\n", dt);
 419                 } else if (!sbi->ll_xattr_cache_set) {
 420                         /* If xattr_cache is already set (no matter 0 or 1)
 421                          * during processing llog, it won't be enabled here. */
 422                         sbi->ll_flags |= LL_SBI_XATTR_CACHE;
 423                         sbi->ll_xattr_cache_enabled = 1;
 424                 }
 425         }
 426
 427         sbi->ll_dt_obd = class_name2obd(dt);
 428         if (!sbi->ll_dt_obd) {
 429                 CERROR("DT %s: not setup or attached\n", dt);
 430                 GOTO(out_md_fid, err = -ENODEV);
 431         }
 432
 433         /* pass client page size via ocd_grant_blkbits, the server should report
 434          * back its backend blocksize for grant calculation purpose */
 435         data->ocd_grant_blkbits = PAGE_SHIFT;
 436
 437         /* indicate OST features supported by this client */
 438         data->ocd_connect_flags = OBD_CONNECT_GRANT | OBD_CONNECT_VERSION |
 439                                   OBD_CONNECT_REQPORTAL | OBD_CONNECT_BRW_SIZE |
 440                                   OBD_CONNECT_CANCELSET | OBD_CONNECT_FID |
 441                                   OBD_CONNECT_SRVLOCK | OBD_CONNECT_TRUNCLOCK|
 442                                   OBD_CONNECT_AT | OBD_CONNECT_OSS_CAPA |
 443                                   OBD_CONNECT_VBR | OBD_CONNECT_FULL20 |
 444                                   OBD_CONNECT_64BITHASH | OBD_CONNECT_MAXBYTES |
 445                                   OBD_CONNECT_EINPROGRESS |
 446                                   OBD_CONNECT_JOBSTATS | OBD_CONNECT_LVB_TYPE |
 447                                   OBD_CONNECT_LAYOUTLOCK |
 448                                   OBD_CONNECT_PINGLESS | OBD_CONNECT_LFSCK |
 449                                   OBD_CONNECT_BULK_MBITS | OBD_CONNECT_SHORTIO |
 450                                   OBD_CONNECT_FLAGS2 | OBD_CONNECT_GRANT_SHRINK;
 451
 452 /* The client currently advertises support for OBD_CONNECT_LOCKAHEAD_OLD so it
 453  * can interoperate with an older version of lockahead which was released prior
 454  * to landing in master. This support will be dropped when 2.13 development
 455  * starts.  At the point, we should not just drop the connect flag (below), we
 456  * should also remove the support in the code.
 457  *
 458  * Removing it means a few things:
 459  * 1. Remove this section here
 460  * 2. Remove CEF_NONBLOCK in ll_file_lockahead()
 461  * 3. Remove function exp_connect_lockahead_old
 462  * 4. Remove LDLM_FL_LOCKAHEAD_OLD_RESERVED in lustre_dlm_flags.h
 463  * */
 464 #if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(2, 12, 50, 0)
 465         data->ocd_connect_flags |= OBD_CONNECT_LOCKAHEAD_OLD;
 466 #endif
 467
 468         data->ocd_connect_flags2 = OBD_CONNECT2_LOCKAHEAD |
 469                                    OBD_CONNECT2_INC_XID;
 470
 471         if (!OBD_FAIL_CHECK(OBD_FAIL_OSC_CONNECT_GRANT_PARAM))
 472                 data->ocd_connect_flags |= OBD_CONNECT_GRANT_PARAM;
 473
 474         /* OBD_CONNECT_CKSUM should always be set, even if checksums are
 475          * disabled by default, because it can still be enabled on the
 476          * fly via /sys. As a consequence, we still need to come to an
 477          * agreement on the supported algorithms at connect time
 478          */
 479         data->ocd_connect_flags |= OBD_CONNECT_CKSUM;
 480
 481         if (OBD_FAIL_CHECK(OBD_FAIL_OSC_CKSUM_ADLER_ONLY))
 482                 data->ocd_cksum_types = OBD_CKSUM_ADLER;
 483         else
 484                 data->ocd_cksum_types = obd_cksum_types_supported_client();
 485
 486 #ifdef HAVE_LRU_RESIZE_SUPPORT
 487         data->ocd_connect_flags |= OBD_CONNECT_LRU_RESIZE;
 488 #endif
 489         /* always ping even if server suppress_pings */
 490         if (sbi->ll_flags & LL_SBI_ALWAYS_PING)
 491                 data->ocd_connect_flags &= ~OBD_CONNECT_PINGLESS;
 492
 493         CDEBUG(D_RPCTRACE, "ocd_connect_flags: %#llx ocd_version: %d "
 494                "ocd_grant: %d\n", data->ocd_connect_flags,
 495                data->ocd_version, data->ocd_grant);
 496
 497         sbi->ll_dt_obd->obd_upcall.onu_owner = &sbi->ll_lco;
 498         sbi->ll_dt_obd->obd_upcall.onu_upcall = cl_ocd_update;
 499
 500         data->ocd_brw_size = DT_MAX_BRW_SIZE;
 501
 502         err = obd_connect(NULL, &sbi->ll_dt_exp, sbi->ll_dt_obd,
 503                           &sbi->ll_sb_uuid, data, sbi->ll_cache);
 504         if (err == -EBUSY) {
 505                 LCONSOLE_ERROR_MSG(0x150, "An OST (dt %s) is performing "
 506                                    "recovery, of which this client is not a "
 507                                    "part.  Please wait for recovery to "
 508                                    "complete, abort, or time out.\n", dt);
 509                 GOTO(out_md, err);
 510         } else if (err) {
 511                 CERROR("%s: Cannot connect to %s: rc = %d\n",
 512                        sbi->ll_dt_exp->exp_obd->obd_name, dt, err);
 513                 GOTO(out_md, err);
 514         }
 515
 516         sbi->ll_dt_exp->exp_connect_data = *data;
 517
 518         /* Don't change value if it was specified in the config log */
 519         if (sbi->ll_ra_info.ra_max_read_ahead_whole_pages == -1) {
 520                 sbi->ll_ra_info.ra_max_read_ahead_whole_pages =
 521                         max_t(unsigned long, SBI_DEFAULT_READAHEAD_WHOLE_MAX,
 522                               (data->ocd_brw_size >> PAGE_SHIFT));
 523                 if (sbi->ll_ra_info.ra_max_read_ahead_whole_pages >
 524                     sbi->ll_ra_info.ra_max_pages_per_file)
 525                         sbi->ll_ra_info.ra_max_read_ahead_whole_pages =
 526                                 sbi->ll_ra_info.ra_max_pages_per_file;
 527         }
 528
 529         err = obd_fid_init(sbi->ll_dt_exp->exp_obd, sbi->ll_dt_exp,
 530                            LUSTRE_SEQ_METADATA);
 531         if (err) {
 532                 CERROR("%s: Can't init data layer FID infrastructure, "
 533                        "rc = %d\n", sbi->ll_dt_exp->exp_obd->obd_name, err);
 534                 GOTO(out_dt, err);
 535         }
 536
 537         mutex_lock(&sbi->ll_lco.lco_lock);
 538         sbi->ll_lco.lco_flags = data->ocd_connect_flags;
 539         sbi->ll_lco.lco_md_exp = sbi->ll_md_exp;
 540         sbi->ll_lco.lco_dt_exp = sbi->ll_dt_exp;
 541         mutex_unlock(&sbi->ll_lco.lco_lock);
 542
 543         fid_zero(&sbi->ll_root_fid);
 544         err = md_get_root(sbi->ll_md_exp, get_mount_fileset(sb),
 545                            &sbi->ll_root_fid);
 546         if (err) {
 547                 CERROR("cannot mds_connect: rc = %d\n", err);
 548                 GOTO(out_lock_cn_cb, err);
 549         }
 550         if (!fid_is_sane(&sbi->ll_root_fid)) {
 551                 CERROR("%s: Invalid root fid "DFID" during mount\n",
 552                        sbi->ll_md_exp->exp_obd->obd_name,
 553                        PFID(&sbi->ll_root_fid));
 554                 GOTO(out_lock_cn_cb, err = -EINVAL);
 555         }
 556         CDEBUG(D_SUPER, "rootfid "DFID"\n", PFID(&sbi->ll_root_fid));
 557
 558         sb->s_op = &lustre_super_operations;
 559         sb->s_xattr = ll_xattr_handlers;
 560 #if THREAD_SIZE >= 8192 /*b=17630*/
 561         sb->s_export_op = &lustre_export_operations;
 562 #endif
 563
 564         /* make root inode
 565          * XXX: move this to after cbd setup? */
 566         valid = OBD_MD_FLGETATTR | OBD_MD_FLBLOCKS | OBD_MD_FLMODEASIZE;
 567         if (sbi->ll_flags & LL_SBI_ACL)
 568                 valid |= OBD_MD_FLACL;
 569
 570         OBD_ALLOC_PTR(op_data);
 571         if (op_data == NULL)
 572                 GOTO(out_lock_cn_cb, err = -ENOMEM);
 573
 574         op_data->op_fid1 = sbi->ll_root_fid;
 575         op_data->op_mode = 0;
 576         op_data->op_valid = valid;
 577
 578         err = md_getattr(sbi->ll_md_exp, op_data, &request);
 579
 580         OBD_FREE_PTR(op_data);
 581         if (err) {
 582                 CERROR("%s: md_getattr failed for root: rc = %d\n",
 583                        sbi->ll_md_exp->exp_obd->obd_name, err);
 584                 GOTO(out_lock_cn_cb, err);
 585         }
 586
 587         err = md_get_lustre_md(sbi->ll_md_exp, request, sbi->ll_dt_exp,
 588                                sbi->ll_md_exp, &lmd);
 589         if (err) {
 590                 CERROR("failed to understand root inode md: rc = %d\n", err);
 591                 ptlrpc_req_finished(request);
 592                 GOTO(out_lock_cn_cb, err);
 593         }
 594
 595         LASSERT(fid_is_sane(&sbi->ll_root_fid));
 596         root = ll_iget(sb, cl_fid_build_ino(&sbi->ll_root_fid,
 597                                             sbi->ll_flags & LL_SBI_32BIT_API),
 598                        &lmd);
 599         md_free_lustre_md(sbi->ll_md_exp, &lmd);
 600         ptlrpc_req_finished(request);
 601
 602         if (IS_ERR(root)) {
 603 #ifdef CONFIG_LUSTRE_FS_POSIX_ACL
 604                 if (lmd.posix_acl) {
 605                         posix_acl_release(lmd.posix_acl);
 606                         lmd.posix_acl = NULL;
 607                 }
 608 #endif
 609                 err = IS_ERR(root) ? PTR_ERR(root) : -EBADF;
 610                 root = NULL;
 611                 CERROR("%s: bad ll_iget() for root: rc = %d\n",
 612                        sbi->ll_fsname, err);
 613                 GOTO(out_root, err);
 614         }
 615
 616         checksum = sbi->ll_flags & LL_SBI_CHECKSUM;
 617         if (sbi->ll_checksum_set) {
 618                 err = obd_set_info_async(NULL, sbi->ll_dt_exp,
 619                                          sizeof(KEY_CHECKSUM), KEY_CHECKSUM,
 620                                          sizeof(checksum), &checksum, NULL);
 621                 if (err) {
 622                         CERROR("%s: Set checksum failed: rc = %d\n",
 623                                sbi->ll_dt_exp->exp_obd->obd_name, err);
 624                         GOTO(out_root, err);
 625                 }
 626         }
 627         cl_sb_init(sb);
 628
 629         sb->s_root = d_make_root(root);
 630         if (sb->s_root == NULL) {
 631                 err = -ENOMEM;
 632                 CERROR("%s: can't make root dentry: rc = %d\n",
 633                        sbi->ll_fsname, err);
 634                 GOTO(out_root, err);
 635         }
 636
 637         sbi->ll_sdev_orig = sb->s_dev;
 638
 639         /* We set sb->s_dev equal on all lustre clients in order to support
 640          * NFS export clustering.  NFSD requires that the FSID be the same
 641          * on all clients. */
 642         /* s_dev is also used in lt_compare() to compare two fs, but that is
 643          * only a node-local comparison. */
 644         uuid = obd_get_uuid(sbi->ll_md_exp);
 645         if (uuid != NULL)
 646                 sb->s_dev = get_uuid2int(uuid->uuid, strlen(uuid->uuid));
 647
 648         if (data != NULL)
 649                 OBD_FREE_PTR(data);
 650         if (osfs != NULL)
 651                 OBD_FREE_PTR(osfs);
 652
 653         if (sbi->ll_dt_obd) {
 654                 err = sysfs_create_link(&sbi->ll_kset.kobj,
 655                                         &sbi->ll_dt_obd->obd_kset.kobj,
 656                                         sbi->ll_dt_obd->obd_type->typ_name);
 657                 if (err < 0) {
 658                         CERROR("%s: could not register %s in llite: rc = %d\n",
 659                                dt, sbi->ll_fsname, err);
 660                         err = 0;
 661                 }
 662         }
 663
 664         if (sbi->ll_md_obd) {
 665                 err = sysfs_create_link(&sbi->ll_kset.kobj,
 666                                         &sbi->ll_md_obd->obd_kset.kobj,
 667                                         sbi->ll_md_obd->obd_type->typ_name);
 668                 if (err < 0) {
 669                         CERROR("%s: could not register %s in llite: rc = %d\n",
 670                                md, sbi->ll_fsname, err);
 671                         err = 0;
 672                 }
 673         }
 674
 675         RETURN(err);
 676 out_root:
 677         if (root)
 678                 iput(root);
 679 out_lock_cn_cb:
 680         obd_fid_fini(sbi->ll_dt_exp->exp_obd);
 681 out_dt:
 682         obd_disconnect(sbi->ll_dt_exp);
 683         sbi->ll_dt_exp = NULL;
 684         sbi->ll_dt_obd = NULL;
 685 out_md_fid:
 686         obd_fid_fini(sbi->ll_md_exp->exp_obd);
 687 out_md:
 688         obd_disconnect(sbi->ll_md_exp);
 689         sbi->ll_md_exp = NULL;
 690         sbi->ll_md_obd = NULL;
 691 out:
 692         if (data != NULL)
 693                 OBD_FREE_PTR(data);
 694         if (osfs != NULL)
 695                 OBD_FREE_PTR(osfs);
 696         return err;
 697 }
 698
 699 int ll_get_max_mdsize(struct ll_sb_info *sbi, int *lmmsize)
 700 {
 701         int size, rc;
 702
 703         size = sizeof(*lmmsize);
 704         rc = obd_get_info(NULL, sbi->ll_dt_exp, sizeof(KEY_MAX_EASIZE),
 705                           KEY_MAX_EASIZE, &size, lmmsize);
 706         if (rc != 0) {
 707                 CERROR("%s: cannot get max LOV EA size: rc = %d\n",
 708                        sbi->ll_dt_exp->exp_obd->obd_name, rc);
 709                 RETURN(rc);
 710         }
 711
 712         CDEBUG(D_INFO, "max LOV ea size: %d\n", *lmmsize);
 713
 714         size = sizeof(int);
 715         rc = obd_get_info(NULL, sbi->ll_md_exp, sizeof(KEY_MAX_EASIZE),
 716                           KEY_MAX_EASIZE, &size, lmmsize);
 717         if (rc)
 718                 CERROR("Get max mdsize error rc %d\n", rc);
 719
 720         CDEBUG(D_INFO, "max LMV ea size: %d\n", *lmmsize);
 721
 722         RETURN(rc);
 723 }
 724
 725 /**
 726  * Get the value of the default_easize parameter.
 727  *
 728  * \see client_obd::cl_default_mds_easize
 729  *
 730  * \param[in] sbi       superblock info for this filesystem
 731  * \param[out] lmmsize  pointer to storage location for value
 732  *
 733  * \retval 0            on success
 734  * \retval negative     negated errno on failure
 735  */
 736 int ll_get_default_mdsize(struct ll_sb_info *sbi, int *lmmsize)
 737 {
 738         int size, rc;
 739
 740         size = sizeof(int);
 741         rc = obd_get_info(NULL, sbi->ll_md_exp, sizeof(KEY_DEFAULT_EASIZE),
 742                          KEY_DEFAULT_EASIZE, &size, lmmsize);
 743         if (rc)
 744                 CERROR("Get default mdsize error rc %d\n", rc);
 745
 746         RETURN(rc);
 747 }
 748
 749 /**
 750  * Set the default_easize parameter to the given value.
 751  *
 752  * \see client_obd::cl_default_mds_easize
 753  *
 754  * \param[in] sbi       superblock info for this filesystem
 755  * \param[in] lmmsize   the size to set
 756  *
 757  * \retval 0            on success
 758  * \retval negative     negated errno on failure
 759  */
 760 int ll_set_default_mdsize(struct ll_sb_info *sbi, int lmmsize)
 761 {
 762         int rc;
 763
 764         if (lmmsize < sizeof(struct lov_mds_md) ||
 765             lmmsize > OBD_MAX_DEFAULT_EA_SIZE)
 766                 return -EINVAL;
 767
 768         rc = obd_set_info_async(NULL, sbi->ll_md_exp,
 769                                 sizeof(KEY_DEFAULT_EASIZE), KEY_DEFAULT_EASIZE,
 770                                 sizeof(int), &lmmsize, NULL);
 771
 772         RETURN(rc);
 773 }
 774
 775 static void client_common_put_super(struct super_block *sb)
 776 {
 777         struct ll_sb_info *sbi = ll_s2sbi(sb);
 778         ENTRY;
 779
 780         cl_sb_fini(sb);
 781
 782         obd_fid_fini(sbi->ll_dt_exp->exp_obd);
 783         obd_disconnect(sbi->ll_dt_exp);
 784         sbi->ll_dt_exp = NULL;
 785
 786         ll_debugfs_unregister_super(sb);
 787
 788         obd_fid_fini(sbi->ll_md_exp->exp_obd);
 789         obd_disconnect(sbi->ll_md_exp);
 790         sbi->ll_md_exp = NULL;
 791
 792         EXIT;
 793 }
 794
 795 void ll_kill_super(struct super_block *sb)
 796 {
 797         struct ll_sb_info *sbi;
 798         ENTRY;
 799
 800         /* not init sb ?*/
 801         if (!(sb->s_flags & SB_ACTIVE))
 802                 return;
 803
 804         sbi = ll_s2sbi(sb);
 805         /* we need restore s_dev from changed for clustred NFS before put_super
 806          * because new kernels have cached s_dev and change sb->s_dev in
 807          * put_super not affected real removing devices */
 808         if (sbi) {
 809                 sb->s_dev = sbi->ll_sdev_orig;
 810
 811                 /* wait running statahead threads to quit */
 812                 while (atomic_read(&sbi->ll_sa_running) > 0) {
 813                         set_current_state(TASK_UNINTERRUPTIBLE);
 814                         schedule_timeout(cfs_time_seconds(1) >> 3);
 815                 }
 816         }
 817
 818         EXIT;
 819 }
 820
 821 static inline int ll_set_opt(const char *opt, char *data, int fl)
 822 {
 823         if (strncmp(opt, data, strlen(opt)) != 0)
 824                 return 0;
 825         else
 826                 return fl;
 827 }
 828
 829 /* non-client-specific mount options are parsed in lmd_parse */
 830 static int ll_options(char *options, struct ll_sb_info *sbi)
 831 {
 832         int tmp;
 833         char *s1 = options, *s2;
 834         int *flags = &sbi->ll_flags;
 835         ENTRY;
 836
 837         if (!options)
 838                 RETURN(0);
 839
 840         CDEBUG(D_CONFIG, "Parsing opts %s\n", options);
 841
 842         while (*s1) {
 843                 CDEBUG(D_SUPER, "next opt=%s\n", s1);
 844                 tmp = ll_set_opt("nolock", s1, LL_SBI_NOLCK);
 845                 if (tmp) {
 846                         *flags |= tmp;
 847                         goto next;
 848                 }
 849                 tmp = ll_set_opt("flock", s1, LL_SBI_FLOCK);
 850                 if (tmp) {
 851                         *flags = (*flags & ~LL_SBI_LOCALFLOCK) | tmp;
 852                         goto next;
 853                 }
 854                 tmp = ll_set_opt("localflock", s1, LL_SBI_LOCALFLOCK);
 855                 if (tmp) {
 856                         *flags = (*flags & ~LL_SBI_FLOCK) | tmp;
 857                         goto next;
 858                 }
 859                 tmp = ll_set_opt("noflock", s1, LL_SBI_FLOCK|LL_SBI_LOCALFLOCK);
 860                 if (tmp) {
 861                         *flags &= ~tmp;
 862                         goto next;
 863                 }
 864                 tmp = ll_set_opt("user_xattr", s1, LL_SBI_USER_XATTR);
 865                 if (tmp) {
 866                         *flags |= tmp;
 867                         goto next;
 868                 }
 869                 tmp = ll_set_opt("nouser_xattr", s1, LL_SBI_USER_XATTR);
 870                 if (tmp) {
 871                         *flags &= ~tmp;
 872                         goto next;
 873                 }
 874                 tmp = ll_set_opt("context", s1, 1);
 875                 if (tmp)
 876                         goto next;
 877                 tmp = ll_set_opt("fscontext", s1, 1);
 878                 if (tmp)
 879                         goto next;
 880                 tmp = ll_set_opt("defcontext", s1, 1);
 881                 if (tmp)
 882                         goto next;
 883                 tmp = ll_set_opt("rootcontext", s1, 1);
 884                 if (tmp)
 885                         goto next;
 886                 tmp = ll_set_opt("user_fid2path", s1, LL_SBI_USER_FID2PATH);
 887                 if (tmp) {
 888                         *flags |= tmp;
 889                         goto next;
 890                 }
 891                 tmp = ll_set_opt("nouser_fid2path", s1, LL_SBI_USER_FID2PATH);
 892                 if (tmp) {
 893                         *flags &= ~tmp;
 894                         goto next;
 895                 }
 896
 897                 tmp = ll_set_opt("checksum", s1, LL_SBI_CHECKSUM);
 898                 if (tmp) {
 899                         *flags |= tmp;
 900                         sbi->ll_checksum_set = 1;
 901                         goto next;
 902                 }
 903                 tmp = ll_set_opt("nochecksum", s1, LL_SBI_CHECKSUM);
 904                 if (tmp) {
 905                         *flags &= ~tmp;
 906                         sbi->ll_checksum_set = 1;
 907                         goto next;
 908                 }
 909                 tmp = ll_set_opt("lruresize", s1, LL_SBI_LRU_RESIZE);
 910                 if (tmp) {
 911                         *flags |= tmp;
 912                         goto next;
 913                 }
 914                 tmp = ll_set_opt("nolruresize", s1, LL_SBI_LRU_RESIZE);
 915                 if (tmp) {
 916                         *flags &= ~tmp;
 917                         goto next;
 918                 }
 919                 tmp = ll_set_opt("lazystatfs", s1, LL_SBI_LAZYSTATFS);
 920                 if (tmp) {
 921                         *flags |= tmp;
 922                         goto next;
 923                 }
 924                 tmp = ll_set_opt("nolazystatfs", s1, LL_SBI_LAZYSTATFS);
 925                 if (tmp) {
 926                         *flags &= ~tmp;
 927                         goto next;
 928                 }
 929                 tmp = ll_set_opt("32bitapi", s1, LL_SBI_32BIT_API);
 930                 if (tmp) {
 931                         *flags |= tmp;
 932                         goto next;
 933                 }
 934                 tmp = ll_set_opt("verbose", s1, LL_SBI_VERBOSE);
 935                 if (tmp) {
 936                         *flags |= tmp;
 937                         goto next;
 938                 }
 939                 tmp = ll_set_opt("noverbose", s1, LL_SBI_VERBOSE);
 940                 if (tmp) {
 941                         *flags &= ~tmp;
 942                         goto next;
 943                 }
 944                 tmp = ll_set_opt("always_ping", s1, LL_SBI_ALWAYS_PING);
 945                 if (tmp) {
 946                         *flags |= tmp;
 947                         goto next;
 948                 }
 949                 LCONSOLE_ERROR_MSG(0x152, "Unknown option '%s', won't mount.\n",
 950                                    s1);
 951                 RETURN(-EINVAL);
 952
 953 next:
 954                 /* Find next opt */
 955                 s2 = strchr(s1, ',');
 956                 if (s2 == NULL)
 957                         break;
 958                 s1 = s2 + 1;
 959         }
 960         RETURN(0);
 961 }
 962
 963 void ll_lli_init(struct ll_inode_info *lli)
 964 {
 965         lli->lli_inode_magic = LLI_INODE_MAGIC;
 966         lli->lli_flags = 0;
 967         spin_lock_init(&lli->lli_lock);
 968         lli->lli_posix_acl = NULL;
 969         /* Do not set lli_fid, it has been initialized already. */
 970         fid_zero(&lli->lli_pfid);
 971         lli->lli_mds_read_och = NULL;
 972         lli->lli_mds_write_och = NULL;
 973         lli->lli_mds_exec_och = NULL;
 974         lli->lli_open_fd_read_count = 0;
 975         lli->lli_open_fd_write_count = 0;
 976         lli->lli_open_fd_exec_count = 0;
 977         mutex_init(&lli->lli_och_mutex);
 978         spin_lock_init(&lli->lli_agl_lock);
 979         spin_lock_init(&lli->lli_layout_lock);
 980         ll_layout_version_set(lli, CL_LAYOUT_GEN_NONE);
 981         lli->lli_clob = NULL;
 982
 983         init_rwsem(&lli->lli_xattrs_list_rwsem);
 984         mutex_init(&lli->lli_xattrs_enq_lock);
 985
 986         LASSERT(lli->lli_vfs_inode.i_mode != 0);
 987         if (S_ISDIR(lli->lli_vfs_inode.i_mode)) {
 988                 lli->lli_opendir_key = NULL;
 989                 lli->lli_sai = NULL;
 990                 spin_lock_init(&lli->lli_sa_lock);
 991                 lli->lli_opendir_pid = 0;
 992                 lli->lli_sa_enabled = 0;
 993                 init_rwsem(&lli->lli_lsm_sem);
 994         } else {
 995                 mutex_init(&lli->lli_size_mutex);
 996                 lli->lli_symlink_name = NULL;
 997                 init_rwsem(&lli->lli_trunc_sem);
 998                 range_lock_tree_init(&lli->lli_write_tree);
 999                 init_rwsem(&lli->lli_glimpse_sem);
1000                 lli->lli_glimpse_time = ktime_set(0, 0);
1001                 INIT_LIST_HEAD(&lli->lli_agl_list);
1002                 lli->lli_agl_index = 0;
1003                 lli->lli_async_rc = 0;
1004                 spin_lock_init(&lli->lli_heat_lock);
1005                 obd_heat_clear(lli->lli_heat_instances, OBD_HEAT_COUNT);
1006                 lli->lli_heat_flags = 0;
1007                 mutex_init(&lli->lli_pcc_lock);
1008                 lli->lli_pcc_state = PCC_STATE_FL_NONE;
1009                 lli->lli_pcc_inode = NULL;
1010                 lli->lli_pcc_dsflags = PCC_DATASET_NONE;
1011                 lli->lli_pcc_generation = 0;
1012                 mutex_init(&lli->lli_group_mutex);
1013                 lli->lli_group_users = 0;
1014                 lli->lli_group_gid = 0;
1015         }
1016         mutex_init(&lli->lli_layout_mutex);
1017         memset(lli->lli_jobid, 0, sizeof(lli->lli_jobid));
1018 }
1019
1020 #define MAX_STRING_SIZE 128
1021
1022 #ifndef HAVE_SUPER_SETUP_BDI_NAME
1023
1024 #define LSI_BDI_INITIALIZED     0x00400000
1025
1026 #ifndef HAVE_BDI_CAP_MAP_COPY
1027 # define BDI_CAP_MAP_COPY       0
1028 #endif
1029
1030 static int super_setup_bdi_name(struct super_block *sb, char *fmt, ...)
1031 {
1032         struct  lustre_sb_info *lsi = s2lsi(sb);
1033         char buf[MAX_STRING_SIZE];
1034         va_list args;
1035         int err;
1036
1037         err = bdi_init(&lsi->lsi_bdi);
1038         if (err)
1039                 return err;
1040
1041         lsi->lsi_flags |= LSI_BDI_INITIALIZED;
1042         lsi->lsi_bdi.capabilities = BDI_CAP_MAP_COPY;
1043         lsi->lsi_bdi.name = "lustre";
1044         va_start(args, fmt);
1045         vsnprintf(buf, MAX_STRING_SIZE, fmt, args);
1046         va_end(args);
1047         err = bdi_register(&lsi->lsi_bdi, NULL, "%s", buf);
1048         va_end(args);
1049         if (!err)
1050                 sb->s_bdi = &lsi->lsi_bdi;
1051
1052         return err;
1053 }
1054 #endif /* !HAVE_SUPER_SETUP_BDI_NAME */
1055
1056 int ll_fill_super(struct super_block *sb, struct vfsmount *mnt)
1057 {
1058         struct  lustre_profile *lprof = NULL;
1059         struct  lustre_sb_info *lsi = s2lsi(sb);
1060         struct  ll_sb_info *sbi = NULL;
1061         char    *dt = NULL, *md = NULL;
1062         char    *profilenm = get_profile_name(sb);
1063         struct config_llog_instance *cfg;
1064         /* %p for void* in printf needs 16+2 characters: 0xffffffffffffffff */
1065         const int instlen = 16 + 2;
1066         unsigned long cfg_instance = ll_get_cfg_instance(sb);
1067         char name[MAX_STRING_SIZE];
1068         int md_len = 0;
1069         int dt_len = 0;
1070         uuid_t uuid;
1071         char *ptr;
1072         int len;
1073         int err;
1074
1075         ENTRY;
1076         /* for ASLR, to map between cfg_instance and hashed ptr */
1077         CDEBUG(D_VFSTRACE, "VFS Op: cfg_instance %s-%016lx (sb %p)\n",
1078                profilenm, cfg_instance, sb);
1079
1080         try_module_get(THIS_MODULE);
1081
1082         OBD_ALLOC_PTR(cfg);
1083         if (cfg == NULL)
1084                 GOTO(out_free_cfg, err = -ENOMEM);
1085
1086         /* client additional sb info */
1087         lsi->lsi_llsbi = sbi = ll_init_sbi();
1088         if (IS_ERR(sbi))
1089                 GOTO(out_free_cfg, err = PTR_ERR(sbi));
1090
1091         err = ll_options(lsi->lsi_lmd->lmd_opts, sbi);
1092         if (err)
1093                 GOTO(out_free_cfg, err);
1094
1095         /* kernel >= 2.6.38 store dentry operations in sb->s_d_op. */
1096         sb->s_d_op = &ll_d_ops;
1097
1098         /* UUID handling */
1099         generate_random_uuid(uuid.b);
1100         snprintf(sbi->ll_sb_uuid.uuid, UUID_SIZE, "%pU", uuid.b);
1101
1102         CDEBUG(D_CONFIG, "llite sb uuid: %s\n", sbi->ll_sb_uuid.uuid);
1103
1104         /* Get fsname */
1105         len = strlen(profilenm);
1106         ptr = strrchr(profilenm, '-');
1107         if (ptr && (strcmp(ptr, "-client") == 0))
1108                 len -= 7;
1109
1110         if (len > LUSTRE_MAXFSNAME) {
1111                 if (unlikely(len >= MAX_STRING_SIZE))
1112                         len = MAX_STRING_SIZE - 1;
1113                 strncpy(name, profilenm, len);
1114                 name[len] = '\0';
1115                 err = -ENAMETOOLONG;
1116                 CERROR("%s: fsname longer than %u characters: rc = %d\n",
1117                        name, LUSTRE_MAXFSNAME, err);
1118                 GOTO(out_free_cfg, err);
1119         }
1120         strncpy(sbi->ll_fsname, profilenm, len);
1121         sbi->ll_fsname[len] = '\0';
1122
1123         /* Mount info */
1124         snprintf(name, sizeof(name), "%.*s-%016lx", len,
1125                  profilenm, cfg_instance);
1126
1127         err = super_setup_bdi_name(sb, "%s", name);
1128         if (err)
1129                 GOTO(out_free_cfg, err);
1130
1131         /* Call ll_debugfs_register_super() before lustre_process_log()
1132          * so that "llite.*.*" params can be processed correctly.
1133          */
1134         err = ll_debugfs_register_super(sb, name);
1135         if (err < 0) {
1136                 CERROR("%s: could not register mountpoint in llite: rc = %d\n",
1137                        sbi->ll_fsname, err);
1138                 err = 0;
1139         }
1140
1141         /* The cfg_instance is a value unique to this super, in case some
1142          * joker tries to mount the same fs at two mount points.
1143          */
1144         cfg->cfg_instance = cfg_instance;
1145         cfg->cfg_uuid = lsi->lsi_llsbi->ll_sb_uuid;
1146         cfg->cfg_callback = class_config_llog_handler;
1147         cfg->cfg_sub_clds = CONFIG_SUB_CLIENT;
1148         /* set up client obds */
1149         err = lustre_process_log(sb, profilenm, cfg);
1150         if (err < 0)
1151                 GOTO(out_debugfs, err);
1152
1153         /* Profile set with LCFG_MOUNTOPT so we can find our mdc and osc obds */
1154         lprof = class_get_profile(profilenm);
1155         if (lprof == NULL) {
1156                 LCONSOLE_ERROR_MSG(0x156, "The client profile '%s' could not be"
1157                                    " read from the MGS.  Does that filesystem "
1158                                    "exist?\n", profilenm);
1159                 GOTO(out_debugfs, err = -EINVAL);
1160         }
1161         CDEBUG(D_CONFIG, "Found profile %s: mdc=%s osc=%s\n", profilenm,
1162                lprof->lp_md, lprof->lp_dt);
1163
1164         dt_len = strlen(lprof->lp_dt) + instlen + 2;
1165         OBD_ALLOC(dt, dt_len);
1166         if (!dt)
1167                 GOTO(out_profile, err = -ENOMEM);
1168         snprintf(dt, dt_len - 1, "%s-%016lx", lprof->lp_dt, cfg_instance);
1169
1170         md_len = strlen(lprof->lp_md) + instlen + 2;
1171         OBD_ALLOC(md, md_len);
1172         if (!md)
1173                 GOTO(out_free_dt, err = -ENOMEM);
1174         snprintf(md, md_len - 1, "%s-%016lx", lprof->lp_md, cfg_instance);
1175
1176         /* connections, registrations, sb setup */
1177         err = client_common_fill_super(sb, md, dt, mnt);
1178         if (err < 0)
1179                 GOTO(out_free_md, err);
1180
1181         sbi->ll_client_common_fill_super_succeeded = 1;
1182
1183 out_free_md:
1184         if (md)
1185                 OBD_FREE(md, md_len);
1186 out_free_dt:
1187         if (dt)
1188                 OBD_FREE(dt, dt_len);
1189 out_profile:
1190         if (lprof)
1191                 class_put_profile(lprof);
1192 out_debugfs:
1193         if (err < 0)
1194                 ll_debugfs_unregister_super(sb);
1195 out_free_cfg:
1196         if (cfg)
1197                 OBD_FREE_PTR(cfg);
1198
1199         if (err)
1200                 ll_put_super(sb);
1201         else if (sbi->ll_flags & LL_SBI_VERBOSE)
1202                 LCONSOLE_WARN("Mounted %s\n", profilenm);
1203         RETURN(err);
1204 } /* ll_fill_super */
1205
1206 void ll_put_super(struct super_block *sb)
1207 {
1208         struct config_llog_instance cfg, params_cfg;
1209         struct obd_device *obd;
1210         struct lustre_sb_info *lsi = s2lsi(sb);
1211         struct ll_sb_info *sbi = ll_s2sbi(sb);
1212         char *profilenm = get_profile_name(sb);
1213         unsigned long cfg_instance = ll_get_cfg_instance(sb);
1214         long ccc_count;
1215         int next, force = 1, rc = 0;
1216         ENTRY;
1217
1218         if (IS_ERR(sbi))
1219                 GOTO(out_no_sbi, 0);
1220
1221         /* Should replace instance_id with something better for ASLR */
1222         CDEBUG(D_VFSTRACE, "VFS Op: cfg_instance %s-%016lx (sb %p)\n",
1223                profilenm, cfg_instance, sb);
1224
1225         cfg.cfg_instance = cfg_instance;
1226         lustre_end_log(sb, profilenm, &cfg);
1227
1228         params_cfg.cfg_instance = cfg_instance;
1229         lustre_end_log(sb, PARAMS_FILENAME, &params_cfg);
1230
1231         if (sbi->ll_md_exp) {
1232                 obd = class_exp2obd(sbi->ll_md_exp);
1233                 if (obd)
1234                         force = obd->obd_force;
1235         }
1236
1237         /* Wait for unstable pages to be committed to stable storage */
1238         if (force == 0) {
1239                 struct l_wait_info lwi = LWI_INTR(LWI_ON_SIGNAL_NOOP, NULL);
1240                 rc = l_wait_event(sbi->ll_cache->ccc_unstable_waitq,
1241                         atomic_long_read(&sbi->ll_cache->ccc_unstable_nr) == 0,
1242                         &lwi);
1243         }
1244
1245         ccc_count = atomic_long_read(&sbi->ll_cache->ccc_unstable_nr);
1246         if (force == 0 && rc != -EINTR)
1247                 LASSERTF(ccc_count == 0, "count: %li\n", ccc_count);
1248
1249         /* We need to set force before the lov_disconnect in
1250            lustre_common_put_super, since l_d cleans up osc's as well. */
1251         if (force) {
1252                 next = 0;
1253                 while ((obd = class_devices_in_group(&sbi->ll_sb_uuid,
1254                                                      &next)) != NULL) {
1255                         obd->obd_force = force;
1256                 }
1257         }
1258
1259         if (sbi->ll_client_common_fill_super_succeeded) {
1260                 /* Only if client_common_fill_super succeeded */
1261                 client_common_put_super(sb);
1262         }
1263
1264         next = 0;
1265         while ((obd = class_devices_in_group(&sbi->ll_sb_uuid, &next)) !=NULL) {
1266                 class_manual_cleanup(obd);
1267         }
1268
1269         if (sbi->ll_flags & LL_SBI_VERBOSE)
1270                 LCONSOLE_WARN("Unmounted %s\n", profilenm ? profilenm : "");
1271
1272         if (profilenm)
1273                 class_del_profile(profilenm);
1274
1275 #ifndef HAVE_SUPER_SETUP_BDI_NAME
1276         if (lsi->lsi_flags & LSI_BDI_INITIALIZED) {
1277                 bdi_destroy(&lsi->lsi_bdi);
1278                 lsi->lsi_flags &= ~LSI_BDI_INITIALIZED;
1279         }
1280 #endif
1281
1282         ll_free_sbi(sb);
1283         lsi->lsi_llsbi = NULL;
1284 out_no_sbi:
1285         lustre_common_put_super(sb);
1286
1287         cl_env_cache_purge(~0);
1288
1289         module_put(THIS_MODULE);
1290
1291         EXIT;
1292 } /* client_put_super */
1293
1294 struct inode *ll_inode_from_resource_lock(struct ldlm_lock *lock)
1295 {
1296         struct inode *inode = NULL;
1297
1298         /* NOTE: we depend on atomic igrab() -bzzz */
1299         lock_res_and_lock(lock);
1300         if (lock->l_resource->lr_lvb_inode) {
1301                 struct ll_inode_info * lli;
1302                 lli = ll_i2info(lock->l_resource->lr_lvb_inode);
1303                 if (lli->lli_inode_magic == LLI_INODE_MAGIC) {
1304                         inode = igrab(lock->l_resource->lr_lvb_inode);
1305                 } else {
1306                         inode = lock->l_resource->lr_lvb_inode;
1307                         LDLM_DEBUG_LIMIT(inode->i_state & I_FREEING ?  D_INFO :
1308                                          D_WARNING, lock, "lr_lvb_inode %p is "
1309                                          "bogus: magic %08x",
1310                                          lock->l_resource->lr_lvb_inode,
1311                                          lli->lli_inode_magic);
1312                         inode = NULL;
1313                 }
1314         }
1315         unlock_res_and_lock(lock);
1316         return inode;
1317 }
1318
1319 void ll_dir_clear_lsm_md(struct inode *inode)
1320 {
1321         struct ll_inode_info *lli = ll_i2info(inode);
1322
1323         LASSERT(S_ISDIR(inode->i_mode));
1324
1325         if (lli->lli_lsm_md) {
1326                 lmv_free_memmd(lli->lli_lsm_md);
1327                 lli->lli_lsm_md = NULL;
1328         }
1329
1330         if (lli->lli_default_lsm_md) {
1331                 lmv_free_memmd(lli->lli_default_lsm_md);
1332                 lli->lli_default_lsm_md = NULL;
1333         }
1334 }
1335
1336 static struct inode *ll_iget_anon_dir(struct super_block *sb,
1337                                       const struct lu_fid *fid,
1338                                       struct lustre_md *md)
1339 {
1340         struct ll_sb_info       *sbi = ll_s2sbi(sb);
1341         struct mdt_body         *body = md->body;
1342         struct inode            *inode;
1343         ino_t                   ino;
1344         ENTRY;
1345
1346         ino = cl_fid_build_ino(fid, sbi->ll_flags & LL_SBI_32BIT_API);
1347         inode = iget_locked(sb, ino);
1348         if (inode == NULL) {
1349                 CERROR("%s: failed get simple inode "DFID": rc = -ENOENT\n",
1350                        sbi->ll_fsname, PFID(fid));
1351                 RETURN(ERR_PTR(-ENOENT));
1352         }
1353
1354         if (inode->i_state & I_NEW) {
1355                 struct ll_inode_info *lli = ll_i2info(inode);
1356                 struct lmv_stripe_md *lsm = md->lmv;
1357
1358                 inode->i_mode = (inode->i_mode & ~S_IFMT) |
1359                                 (body->mbo_mode & S_IFMT);
1360                 LASSERTF(S_ISDIR(inode->i_mode), "Not slave inode "DFID"\n",
1361                          PFID(fid));
1362
1363                 inode->i_mtime.tv_sec = 0;
1364                 inode->i_atime.tv_sec = 0;
1365                 inode->i_ctime.tv_sec = 0;
1366                 inode->i_rdev = 0;
1367
1368 #ifdef HAVE_BACKING_DEV_INFO
1369                 /* initializing backing dev info. */
1370                 inode->i_mapping->backing_dev_info =
1371                                                 &s2lsi(inode->i_sb)->lsi_bdi;
1372 #endif
1373                 inode->i_op = &ll_dir_inode_operations;
1374                 inode->i_fop = &ll_dir_operations;
1375                 lli->lli_fid = *fid;
1376                 ll_lli_init(lli);
1377
1378                 LASSERT(lsm != NULL);
1379                 /* master object FID */
1380                 lli->lli_pfid = body->mbo_fid1;
1381                 CDEBUG(D_INODE, "lli %p slave "DFID" master "DFID"\n",
1382                        lli, PFID(fid), PFID(&lli->lli_pfid));
1383                 unlock_new_inode(inode);
1384         }
1385
1386         RETURN(inode);
1387 }
1388
1389 static int ll_init_lsm_md(struct inode *inode, struct lustre_md *md)
1390 {
1391         struct lu_fid *fid;
1392         struct lmv_stripe_md *lsm = md->lmv;
1393         struct ll_inode_info *lli = ll_i2info(inode);
1394         int i;
1395
1396         LASSERT(lsm != NULL);
1397
1398         CDEBUG(D_INODE, "%s: "DFID" set dir layout:\n",
1399                ll_i2sbi(inode)->ll_fsname, PFID(&lli->lli_fid));
1400         lsm_md_dump(D_INODE, lsm);
1401
1402         if (!lmv_dir_striped(lsm))
1403                 goto out;
1404
1405         /* XXX sigh, this lsm_root initialization should be in
1406          * LMV layer, but it needs ll_iget right now, so we
1407          * put this here right now. */
1408         for (i = 0; i < lsm->lsm_md_stripe_count; i++) {
1409                 fid = &lsm->lsm_md_oinfo[i].lmo_fid;
1410                 LASSERT(lsm->lsm_md_oinfo[i].lmo_root == NULL);
1411
1412                 if (!fid_is_sane(fid))
1413                         continue;
1414
1415                 /* Unfortunately ll_iget will call ll_update_inode,
1416                  * where the initialization of slave inode is slightly
1417                  * different, so it reset lsm_md to NULL to avoid
1418                  * initializing lsm for slave inode. */
1419                 lsm->lsm_md_oinfo[i].lmo_root =
1420                                 ll_iget_anon_dir(inode->i_sb, fid, md);
1421                 if (IS_ERR(lsm->lsm_md_oinfo[i].lmo_root)) {
1422                         int rc = PTR_ERR(lsm->lsm_md_oinfo[i].lmo_root);
1423
1424                         lsm->lsm_md_oinfo[i].lmo_root = NULL;
1425                         while (i-- > 0) {
1426                                 iput(lsm->lsm_md_oinfo[i].lmo_root);
1427                                 lsm->lsm_md_oinfo[i].lmo_root = NULL;
1428                         }
1429                         return rc;
1430                 }
1431         }
1432 out:
1433         lli->lli_lsm_md = lsm;
1434
1435         return 0;
1436 }
1437
1438 static void ll_update_default_lsm_md(struct inode *inode, struct lustre_md *md)
1439 {
1440         struct ll_inode_info *lli = ll_i2info(inode);
1441
1442         if (!md->default_lmv) {
1443                 /* clear default lsm */
1444                 if (lli->lli_default_lsm_md) {
1445                         down_write(&lli->lli_lsm_sem);
1446                         if (lli->lli_default_lsm_md) {
1447                                 lmv_free_memmd(lli->lli_default_lsm_md);
1448                                 lli->lli_default_lsm_md = NULL;
1449                         }
1450                         up_write(&lli->lli_lsm_sem);
1451                 }
1452         } else if (lli->lli_default_lsm_md) {
1453                 /* update default lsm if it changes */
1454                 down_read(&lli->lli_lsm_sem);
1455                 if (lli->lli_default_lsm_md &&
1456                     !lsm_md_eq(lli->lli_default_lsm_md, md->default_lmv)) {
1457                         up_read(&lli->lli_lsm_sem);
1458                         down_write(&lli->lli_lsm_sem);
1459                         if (lli->lli_default_lsm_md)
1460                                 lmv_free_memmd(lli->lli_default_lsm_md);
1461                         lli->lli_default_lsm_md = md->default_lmv;
1462                         lsm_md_dump(D_INODE, md->default_lmv);
1463                         md->default_lmv = NULL;
1464                         up_write(&lli->lli_lsm_sem);
1465                 } else {
1466                         up_read(&lli->lli_lsm_sem);
1467                 }
1468         } else {
1469                 /* init default lsm */
1470                 down_write(&lli->lli_lsm_sem);
1471                 lli->lli_default_lsm_md = md->default_lmv;
1472                 lsm_md_dump(D_INODE, md->default_lmv);
1473                 md->default_lmv = NULL;
1474                 up_write(&lli->lli_lsm_sem);
1475         }
1476 }
1477
1478 static int ll_update_lsm_md(struct inode *inode, struct lustre_md *md)
1479 {
1480         struct ll_inode_info *lli = ll_i2info(inode);
1481         struct lmv_stripe_md *lsm = md->lmv;
1482         int rc = 0;
1483
1484         ENTRY;
1485
1486         LASSERT(S_ISDIR(inode->i_mode));
1487         CDEBUG(D_INODE, "update lsm %p of "DFID"\n", lli->lli_lsm_md,
1488                PFID(ll_inode2fid(inode)));
1489
1490         /* update default LMV */
1491         if (md->default_lmv)
1492                 ll_update_default_lsm_md(inode, md);
1493
1494         /*
1495          * no striped information from request, lustre_md from req does not
1496          * include stripeEA, see ll_md_setattr()
1497          */
1498         if (!lsm)
1499                 RETURN(0);
1500
1501         /*
1502          * normally dir layout doesn't change, only take read lock to check
1503          * that to avoid blocking other MD operations.
1504          */
1505         if (lli->lli_lsm_md)
1506                 down_read(&lli->lli_lsm_sem);
1507         else
1508                 down_write(&lli->lli_lsm_sem);
1509
1510         /*
1511          * if dir layout mismatch, check whether version is increased, which
1512          * means layout is changed, this happens in dir migration and lfsck.
1513          *
1514          * foreign LMV should not change.
1515          */
1516         if (lli->lli_lsm_md && !lsm_md_eq(lli->lli_lsm_md, lsm)) {
1517                 if (lmv_dir_striped(lli->lli_lsm_md) &&
1518                     lsm->lsm_md_layout_version <=
1519                     lli->lli_lsm_md->lsm_md_layout_version) {
1520                         CERROR("%s: "DFID" dir layout mismatch:\n",
1521                                ll_i2sbi(inode)->ll_fsname,
1522                                PFID(&lli->lli_fid));
1523                         lsm_md_dump(D_ERROR, lli->lli_lsm_md);
1524                         lsm_md_dump(D_ERROR, lsm);
1525                         GOTO(unlock, rc = -EINVAL);
1526                 }
1527
1528                 /* layout changed, switch to write lock */
1529                 up_read(&lli->lli_lsm_sem);
1530                 down_write(&lli->lli_lsm_sem);
1531                 ll_dir_clear_lsm_md(inode);
1532         }
1533
1534         /* set directory layout */
1535         if (!lli->lli_lsm_md) {
1536                 struct cl_attr  *attr;
1537
1538                 rc = ll_init_lsm_md(inode, md);
1539                 up_write(&lli->lli_lsm_sem);
1540                 if (rc != 0)
1541                         RETURN(rc);
1542
1543                 /* set md->lmv to NULL, so the following free lustre_md
1544                  * will not free this lsm */
1545                 md->lmv = NULL;
1546
1547                 /*
1548                  * md_merge_attr() may take long, since lsm is already set,
1549                  * switch to read lock.
1550                  */
1551                 down_read(&lli->lli_lsm_sem);
1552
1553                 if (!lmv_dir_striped(lli->lli_lsm_md))
1554                         GOTO(unlock, rc);
1555
1556                 OBD_ALLOC_PTR(attr);
1557                 if (attr == NULL)
1558                         GOTO(unlock, rc = -ENOMEM);
1559
1560                 /* validate the lsm */
1561                 rc = md_merge_attr(ll_i2mdexp(inode), lsm, attr,
1562                                    ll_md_blocking_ast);
1563                 if (rc != 0) {
1564                         OBD_FREE_PTR(attr);
1565                         GOTO(unlock, rc);
1566                 }
1567
1568                 if (md->body->mbo_valid & OBD_MD_FLNLINK)
1569                         md->body->mbo_nlink = attr->cat_nlink;
1570                 if (md->body->mbo_valid & OBD_MD_FLSIZE)
1571                         md->body->mbo_size = attr->cat_size;
1572                 if (md->body->mbo_valid & OBD_MD_FLATIME)
1573                         md->body->mbo_atime = attr->cat_atime;
1574                 if (md->body->mbo_valid & OBD_MD_FLCTIME)
1575                         md->body->mbo_ctime = attr->cat_ctime;
1576                 if (md->body->mbo_valid & OBD_MD_FLMTIME)
1577                         md->body->mbo_mtime = attr->cat_mtime;
1578
1579                 OBD_FREE_PTR(attr);
1580         }
1581 unlock:
1582         up_read(&lli->lli_lsm_sem);
1583
1584         RETURN(rc);
1585 }
1586
1587 void ll_clear_inode(struct inode *inode)
1588 {
1589         struct ll_inode_info *lli = ll_i2info(inode);
1590         struct ll_sb_info *sbi = ll_i2sbi(inode);
1591
1592         ENTRY;
1593
1594         CDEBUG(D_VFSTRACE, "VFS Op:inode="DFID"(%p)\n",
1595                PFID(ll_inode2fid(inode)), inode);
1596
1597         if (S_ISDIR(inode->i_mode)) {
1598                 /* these should have been cleared in ll_file_release */
1599                 LASSERT(lli->lli_opendir_key == NULL);
1600                 LASSERT(lli->lli_sai == NULL);
1601                 LASSERT(lli->lli_opendir_pid == 0);
1602         } else {
1603                 pcc_inode_free(inode);
1604         }
1605
1606         md_null_inode(sbi->ll_md_exp, ll_inode2fid(inode));
1607
1608         LASSERT(!lli->lli_open_fd_write_count);
1609         LASSERT(!lli->lli_open_fd_read_count);
1610         LASSERT(!lli->lli_open_fd_exec_count);
1611
1612         if (lli->lli_mds_write_och)
1613                 ll_md_real_close(inode, FMODE_WRITE);
1614         if (lli->lli_mds_exec_och)
1615                 ll_md_real_close(inode, FMODE_EXEC);
1616         if (lli->lli_mds_read_och)
1617                 ll_md_real_close(inode, FMODE_READ);
1618
1619         if (S_ISLNK(inode->i_mode) && lli->lli_symlink_name) {
1620                 OBD_FREE(lli->lli_symlink_name,
1621                          strlen(lli->lli_symlink_name) + 1);
1622                 lli->lli_symlink_name = NULL;
1623         }
1624
1625         ll_xattr_cache_destroy(inode);
1626
1627 #ifdef CONFIG_LUSTRE_FS_POSIX_ACL
1628         forget_all_cached_acls(inode);
1629         if (lli->lli_posix_acl) {
1630                 posix_acl_release(lli->lli_posix_acl);
1631                 lli->lli_posix_acl = NULL;
1632         }
1633 #endif
1634         lli->lli_inode_magic = LLI_INODE_DEAD;
1635
1636         if (S_ISDIR(inode->i_mode))
1637                 ll_dir_clear_lsm_md(inode);
1638         else if (S_ISREG(inode->i_mode) && !is_bad_inode(inode))
1639                 LASSERT(list_empty(&lli->lli_agl_list));
1640
1641         /*
1642          * XXX This has to be done before lsm is freed below, because
1643          * cl_object still uses inode lsm.
1644          */
1645         cl_inode_fini(inode);
1646
1647         EXIT;
1648 }
1649
1650 static int ll_md_setattr(struct dentry *dentry, struct md_op_data *op_data)
1651 {
1652         struct lustre_md md;
1653         struct inode *inode = dentry->d_inode;
1654         struct ll_sb_info *sbi = ll_i2sbi(inode);
1655         struct ptlrpc_request *request = NULL;
1656         int rc, ia_valid;
1657         ENTRY;
1658
1659         op_data = ll_prep_md_op_data(op_data, inode, NULL, NULL, 0, 0,
1660                                      LUSTRE_OPC_ANY, NULL);
1661         if (IS_ERR(op_data))
1662                 RETURN(PTR_ERR(op_data));
1663
1664         rc = md_setattr(sbi->ll_md_exp, op_data, NULL, 0, &request);
1665         if (rc) {
1666                 ptlrpc_req_finished(request);
1667                 if (rc == -ENOENT) {
1668                         clear_nlink(inode);
1669                         /* Unlinked special device node? Or just a race?
1670                          * Pretend we done everything. */
1671                         if (!S_ISREG(inode->i_mode) &&
1672                             !S_ISDIR(inode->i_mode)) {
1673                                 ia_valid = op_data->op_attr.ia_valid;
1674                                 op_data->op_attr.ia_valid &= ~TIMES_SET_FLAGS;
1675                                 rc = simple_setattr(dentry, &op_data->op_attr);
1676                                 op_data->op_attr.ia_valid = ia_valid;
1677                         }
1678                 } else if (rc != -EPERM && rc != -EACCES && rc != -ETXTBSY) {
1679                         CERROR("md_setattr fails: rc = %d\n", rc);
1680                 }
1681                 RETURN(rc);
1682         }
1683
1684         rc = md_get_lustre_md(sbi->ll_md_exp, request, sbi->ll_dt_exp,
1685                               sbi->ll_md_exp, &md);
1686         if (rc) {
1687                 ptlrpc_req_finished(request);
1688                 RETURN(rc);
1689         }
1690
1691         ia_valid = op_data->op_attr.ia_valid;
1692         /* inode size will be in ll_setattr_ost, can't do it now since dirty
1693          * cache is not cleared yet. */
1694         op_data->op_attr.ia_valid &= ~(TIMES_SET_FLAGS | ATTR_SIZE);
1695         if (S_ISREG(inode->i_mode))
1696                 inode_lock(inode);
1697         rc = simple_setattr(dentry, &op_data->op_attr);
1698         if (S_ISREG(inode->i_mode))
1699                 inode_unlock(inode);
1700         op_data->op_attr.ia_valid = ia_valid;
1701
1702         rc = ll_update_inode(inode, &md);
1703         ptlrpc_req_finished(request);
1704
1705         RETURN(rc);
1706 }
1707
1708 /* If this inode has objects allocated to it (lsm != NULL), then the OST
1709  * object(s) determine the file size and mtime.  Otherwise, the MDS will
1710  * keep these values until such a time that objects are allocated for it.
1711  * We do the MDS operations first, as it is checking permissions for us.
1712  * We don't to the MDS RPC if there is nothing that we want to store there,
1713  * otherwise there is no harm in updating mtime/atime on the MDS if we are
1714  * going to do an RPC anyways.
1715  *
1716  * If we are doing a truncate, we will send the mtime and ctime updates
1717  * to the OST with the punch RPC, otherwise we do an explicit setattr RPC.
1718  * I don't believe it is possible to get e.g. ATTR_MTIME_SET and ATTR_SIZE
1719  * at the same time.
1720  *
1721  * In case of HSMimport, we only set attr on MDS.
1722  */
1723 int ll_setattr_raw(struct dentry *dentry, struct iattr *attr,
1724                    enum op_xvalid xvalid, bool hsm_import)
1725 {
1726         struct inode *inode = dentry->d_inode;
1727         struct ll_inode_info *lli = ll_i2info(inode);
1728         struct md_op_data *op_data = NULL;
1729         ktime_t kstart = ktime_get();
1730         int rc = 0;
1731
1732         ENTRY;
1733
1734         CDEBUG(D_VFSTRACE, "%s: setattr inode "DFID"(%p) from %llu to %llu, "
1735                "valid %x, hsm_import %d\n",
1736                ll_i2sbi(inode)->ll_fsname, PFID(&lli->lli_fid),
1737                inode, i_size_read(inode), attr->ia_size, attr->ia_valid,
1738                hsm_import);
1739
1740         if (attr->ia_valid & ATTR_SIZE) {
1741                 /* Check new size against VFS/VM file size limit and rlimit */
1742                 rc = inode_newsize_ok(inode, attr->ia_size);
1743                 if (rc)
1744                         RETURN(rc);
1745
1746                 /* The maximum Lustre file size is variable, based on the
1747                  * OST maximum object size and number of stripes.  This
1748                  * needs another check in addition to the VFS check above. */
1749                 if (attr->ia_size > ll_file_maxbytes(inode)) {
1750                         CDEBUG(D_INODE,"file "DFID" too large %llu > %llu\n",
1751                                PFID(&lli->lli_fid), attr->ia_size,
1752                                ll_file_maxbytes(inode));
1753                         RETURN(-EFBIG);
1754                 }
1755
1756                 attr->ia_valid |= ATTR_MTIME | ATTR_CTIME;
1757         }
1758
1759         /* POSIX: check before ATTR_*TIME_SET set (from inode_change_ok) */
1760         if (attr->ia_valid & TIMES_SET_FLAGS) {
1761                 if ((!uid_eq(current_fsuid(), inode->i_uid)) &&
1762                     !cfs_capable(CFS_CAP_FOWNER))
1763                         RETURN(-EPERM);
1764         }
1765
1766         /* We mark all of the fields "set" so MDS/OST does not re-set them */
1767         if (!(xvalid & OP_XVALID_CTIME_SET) &&
1768              (attr->ia_valid & ATTR_CTIME)) {
1769                 attr->ia_ctime = current_time(inode);
1770                 xvalid |= OP_XVALID_CTIME_SET;
1771         }
1772         if (!(attr->ia_valid & ATTR_ATIME_SET) &&
1773             (attr->ia_valid & ATTR_ATIME)) {
1774                 attr->ia_atime = current_time(inode);
1775                 attr->ia_valid |= ATTR_ATIME_SET;
1776         }
1777         if (!(attr->ia_valid & ATTR_MTIME_SET) &&
1778             (attr->ia_valid & ATTR_MTIME)) {
1779                 attr->ia_mtime = current_time(inode);
1780                 attr->ia_valid |= ATTR_MTIME_SET;
1781         }
1782
1783         if (attr->ia_valid & (ATTR_MTIME | ATTR_CTIME))
1784                 CDEBUG(D_INODE, "setting mtime %lld, ctime %lld, now = %lld\n",
1785                        (s64)attr->ia_mtime.tv_sec, (s64)attr->ia_ctime.tv_sec,
1786                        ktime_get_real_seconds());
1787
1788         if (S_ISREG(inode->i_mode)) {
1789                 if (attr->ia_valid & ATTR_SIZE)
1790                         inode_dio_write_done(inode);
1791                 inode_unlock(inode);
1792         }
1793
1794         /* We always do an MDS RPC, even if we're only changing the size;
1795          * only the MDS knows whether truncate() should fail with -ETXTBUSY */
1796
1797         OBD_ALLOC_PTR(op_data);
1798         if (op_data == NULL)
1799                 GOTO(out, rc = -ENOMEM);
1800
1801         if (!hsm_import && attr->ia_valid & ATTR_SIZE) {
1802                 /* If we are changing file size, file content is
1803                  * modified, flag it.
1804                  */
1805                 xvalid |= OP_XVALID_OWNEROVERRIDE;
1806                 op_data->op_bias |= MDS_DATA_MODIFIED;
1807                 ll_file_clear_flag(lli, LLIF_DATA_MODIFIED);
1808         }
1809
1810         if (attr->ia_valid & ATTR_FILE) {
1811                 struct ll_file_data *fd = LUSTRE_FPRIVATE(attr->ia_file);
1812
1813                 if (fd->fd_lease_och)
1814                         op_data->op_bias |= MDS_TRUNC_KEEP_LEASE;
1815         }
1816
1817         op_data->op_attr = *attr;
1818         op_data->op_xvalid = xvalid;
1819
1820         rc = ll_md_setattr(dentry, op_data);
1821         if (rc)
1822                 GOTO(out, rc);
1823
1824         if (!S_ISREG(inode->i_mode) || hsm_import)
1825                 GOTO(out, rc = 0);
1826
1827         if (attr->ia_valid & (ATTR_SIZE | ATTR_ATIME | ATTR_ATIME_SET |
1828                               ATTR_MTIME | ATTR_MTIME_SET | ATTR_CTIME) ||
1829             xvalid & OP_XVALID_CTIME_SET) {
1830                 bool cached = false;
1831
1832                 rc = pcc_inode_setattr(inode, attr, &cached);
1833                 if (cached) {
1834                         if (rc) {
1835                                 CERROR("%s: PCC inode "DFID" setattr failed: "
1836                                        "rc = %d\n",
1837                                        ll_i2sbi(inode)->ll_fsname,
1838                                        PFID(&lli->lli_fid), rc);
1839                                 GOTO(out, rc);
1840                         }
1841                 } else {
1842                         /* For truncate and utimes sending attributes to OSTs,
1843                          * setting mtime/atime to the past will be performed
1844                          * under PW [0:EOF] extent lock (new_size:EOF for
1845                          * truncate). It may seem excessive to send mtime/atime
1846                          * updates to OSTs when not setting times to past, but
1847                          * it is necessary due to possible time
1848                          * de-synchronization between MDT inode and OST objects
1849                          */
1850                         rc = cl_setattr_ost(lli->lli_clob, attr, xvalid, 0);
1851                 }
1852         }
1853
1854         /* If the file was restored, it needs to set dirty flag.
1855          *
1856          * We've already sent MDS_DATA_MODIFIED flag in
1857          * ll_md_setattr() for truncate. However, the MDT refuses to
1858          * set the HS_DIRTY flag on released files, so we have to set
1859          * it again if the file has been restored. Please check how
1860          * LLIF_DATA_MODIFIED is set in vvp_io_setattr_fini().
1861          *
1862          * Please notice that if the file is not released, the previous
1863          * MDS_DATA_MODIFIED has taken effect and usually
1864          * LLIF_DATA_MODIFIED is not set(see vvp_io_setattr_fini()).
1865          * This way we can save an RPC for common open + trunc
1866          * operation. */
1867         if (ll_file_test_and_clear_flag(lli, LLIF_DATA_MODIFIED)) {
1868                 struct hsm_state_set hss = {
1869                         .hss_valid = HSS_SETMASK,
1870                         .hss_setmask = HS_DIRTY,
1871                 };
1872                 int rc2;
1873
1874                 rc2 = ll_hsm_state_set(inode, &hss);
1875                 /* truncate and write can happen at the same time, so that
1876                  * the file can be set modified even though the file is not
1877                  * restored from released state, and ll_hsm_state_set() is
1878                  * not applicable for the file, and rc2 < 0 is normal in this
1879                  * case. */
1880                 if (rc2 < 0)
1881                         CDEBUG(D_INFO, DFID "HSM set dirty failed: rc2 = %d\n",
1882                                PFID(ll_inode2fid(inode)), rc2);
1883         }
1884
1885         EXIT;
1886 out:
1887         if (op_data != NULL)
1888                 ll_finish_md_op_data(op_data);
1889
1890         if (S_ISREG(inode->i_mode)) {
1891                 inode_lock(inode);
1892                 if ((attr->ia_valid & ATTR_SIZE) && !hsm_import)
1893                         inode_dio_wait(inode);
1894                 /* Once we've got the i_mutex, it's safe to set the S_NOSEC
1895                  * flag.  ll_update_inode (called from ll_md_setattr), clears
1896                  * inode flags, so there is a gap where S_NOSEC is not set.
1897                  * This can cause a writer to take the i_mutex unnecessarily,
1898                  * but this is safe to do and should be rare. */
1899                 inode_has_no_xattr(inode);
1900         }
1901
1902         if (!rc)
1903                 ll_stats_ops_tally(ll_i2sbi(inode), attr->ia_valid & ATTR_SIZE ?
1904                                         LPROC_LL_TRUNC : LPROC_LL_SETATTR,
1905                                    ktime_us_delta(ktime_get(), kstart));
1906
1907         return rc;
1908 }
1909
1910 int ll_setattr(struct dentry *de, struct iattr *attr)
1911 {
1912         int mode = de->d_inode->i_mode;
1913         enum op_xvalid xvalid = 0;
1914
1915         if ((attr->ia_valid & (ATTR_CTIME|ATTR_SIZE|ATTR_MODE)) ==
1916                               (ATTR_CTIME|ATTR_SIZE|ATTR_MODE))
1917                 xvalid |= OP_XVALID_OWNEROVERRIDE;
1918
1919         if (((attr->ia_valid & (ATTR_MODE|ATTR_FORCE|ATTR_SIZE)) ==
1920                                (ATTR_SIZE|ATTR_MODE)) &&
1921             (((mode & S_ISUID) && !(attr->ia_mode & S_ISUID)) ||
1922              (((mode & (S_ISGID|S_IXGRP)) == (S_ISGID|S_IXGRP)) &&
1923               !(attr->ia_mode & S_ISGID))))
1924                 attr->ia_valid |= ATTR_FORCE;
1925
1926         if ((attr->ia_valid & ATTR_MODE) &&
1927             (mode & S_ISUID) &&
1928             !(attr->ia_mode & S_ISUID) &&
1929             !(attr->ia_valid & ATTR_KILL_SUID))
1930                 attr->ia_valid |= ATTR_KILL_SUID;
1931
1932         if ((attr->ia_valid & ATTR_MODE) &&
1933             ((mode & (S_ISGID|S_IXGRP)) == (S_ISGID|S_IXGRP)) &&
1934             !(attr->ia_mode & S_ISGID) &&
1935             !(attr->ia_valid & ATTR_KILL_SGID))
1936                 attr->ia_valid |= ATTR_KILL_SGID;
1937
1938         return ll_setattr_raw(de, attr, xvalid, false);
1939 }
1940
1941 int ll_statfs_internal(struct ll_sb_info *sbi, struct obd_statfs *osfs,
1942                        u32 flags)
1943 {
1944         struct obd_statfs obd_osfs = { 0 };
1945         time64_t max_age;
1946         int rc;
1947
1948         ENTRY;
1949         max_age = ktime_get_seconds() - sbi->ll_statfs_max_age;
1950
1951         if (sbi->ll_flags & LL_SBI_LAZYSTATFS)
1952                 flags |= OBD_STATFS_NODELAY;
1953
1954         rc = obd_statfs(NULL, sbi->ll_md_exp, osfs, max_age, flags);
1955         if (rc)
1956                 RETURN(rc);
1957
1958         osfs->os_type = LL_SUPER_MAGIC;
1959
1960         CDEBUG(D_SUPER, "MDC blocks %llu/%llu objects %llu/%llu\n",
1961               osfs->os_bavail, osfs->os_blocks, osfs->os_ffree, osfs->os_files);
1962
1963         if (osfs->os_state & OS_STATE_SUM)
1964                 GOTO(out, rc);
1965
1966         rc = obd_statfs(NULL, sbi->ll_dt_exp, &obd_osfs, max_age, flags);
1967         if (rc) /* Possibly a filesystem with no OSTs.  Report MDT totals. */
1968                 GOTO(out, rc = 0);
1969
1970         CDEBUG(D_SUPER, "OSC blocks %llu/%llu objects %llu/%llu\n",
1971                obd_osfs.os_bavail, obd_osfs.os_blocks, obd_osfs.os_ffree,
1972                obd_osfs.os_files);
1973
1974         osfs->os_bsize = obd_osfs.os_bsize;
1975         osfs->os_blocks = obd_osfs.os_blocks;
1976         osfs->os_bfree = obd_osfs.os_bfree;
1977         osfs->os_bavail = obd_osfs.os_bavail;
1978
1979         /* If we have _some_ OSTs, but don't have as many free objects on the
1980          * OSTs as inodes on the MDTs, reduce the reported number of inodes
1981          * to compensate, so that the "inodes in use" number is correct.
1982          * This should be kept in sync with lod_statfs() behaviour.
1983          */
1984         if (obd_osfs.os_files && obd_osfs.os_ffree < osfs->os_ffree) {
1985                 osfs->os_files = (osfs->os_files - osfs->os_ffree) +
1986                                  obd_osfs.os_ffree;
1987                 osfs->os_ffree = obd_osfs.os_ffree;
1988         }
1989
1990 out:
1991         RETURN(rc);
1992 }
1993
1994 int ll_statfs(struct dentry *de, struct kstatfs *sfs)
1995 {
1996         struct super_block *sb = de->d_sb;
1997         struct obd_statfs osfs;
1998         __u64 fsid = huge_encode_dev(sb->s_dev);
1999         ktime_t kstart = ktime_get();
2000         int rc;
2001
2002         CDEBUG(D_VFSTRACE, "VFS Op:sb=%s (%p)\n", sb->s_id, sb);
2003
2004         /* Some amount of caching on the client is allowed */
2005         rc = ll_statfs_internal(ll_s2sbi(sb), &osfs, OBD_STATFS_SUM);
2006         if (rc)
2007                 return rc;
2008
2009         statfs_unpack(sfs, &osfs);
2010
2011         /* We need to downshift for all 32-bit kernels, because we can't
2012          * tell if the kernel is being called via sys_statfs64() or not.
2013          * Stop before overflowing f_bsize - in which case it is better
2014          * to just risk EOVERFLOW if caller is using old sys_statfs(). */
2015         if (sizeof(long) < 8) {
2016                 while (osfs.os_blocks > ~0UL && sfs->f_bsize < 0x40000000) {
2017                         sfs->f_bsize <<= 1;
2018
2019                         osfs.os_blocks >>= 1;
2020                         osfs.os_bfree >>= 1;
2021                         osfs.os_bavail >>= 1;
2022                 }
2023         }
2024
2025         sfs->f_blocks = osfs.os_blocks;
2026         sfs->f_bfree = osfs.os_bfree;
2027         sfs->f_bavail = osfs.os_bavail;
2028         sfs->f_fsid.val[0] = (__u32)fsid;
2029         sfs->f_fsid.val[1] = (__u32)(fsid >> 32);
2030
2031         ll_stats_ops_tally(ll_s2sbi(sb), LPROC_LL_STATFS,
2032                            ktime_us_delta(ktime_get(), kstart));
2033
2034         return 0;
2035 }
2036
2037 void ll_inode_size_lock(struct inode *inode)
2038 {
2039         struct ll_inode_info *lli;
2040
2041         LASSERT(!S_ISDIR(inode->i_mode));
2042
2043         lli = ll_i2info(inode);
2044         mutex_lock(&lli->lli_size_mutex);
2045 }
2046
2047 void ll_inode_size_unlock(struct inode *inode)
2048 {
2049         struct ll_inode_info *lli;
2050
2051         lli = ll_i2info(inode);
2052         mutex_unlock(&lli->lli_size_mutex);
2053 }
2054
2055 void ll_update_inode_flags(struct inode *inode, int ext_flags)
2056 {
2057         inode->i_flags = ll_ext_to_inode_flags(ext_flags);
2058         if (ext_flags & LUSTRE_PROJINHERIT_FL)
2059                 ll_file_set_flag(ll_i2info(inode), LLIF_PROJECT_INHERIT);
2060         else
2061                 ll_file_clear_flag(ll_i2info(inode), LLIF_PROJECT_INHERIT);
2062 }
2063
2064 int ll_update_inode(struct inode *inode, struct lustre_md *md)
2065 {
2066         struct ll_inode_info *lli = ll_i2info(inode);
2067         struct mdt_body *body = md->body;
2068         struct ll_sb_info *sbi = ll_i2sbi(inode);
2069         int rc = 0;
2070
2071         if (body->mbo_valid & OBD_MD_FLEASIZE) {
2072                 rc = cl_file_inode_init(inode, md);
2073                 if (rc)
2074                         return rc;
2075         }
2076
2077         if (S_ISDIR(inode->i_mode)) {
2078                 rc = ll_update_lsm_md(inode, md);
2079                 if (rc != 0)
2080                         return rc;
2081         }
2082
2083 #ifdef CONFIG_LUSTRE_FS_POSIX_ACL
2084         if (body->mbo_valid & OBD_MD_FLACL) {
2085                 spin_lock(&lli->lli_lock);
2086                 if (lli->lli_posix_acl)
2087                         posix_acl_release(lli->lli_posix_acl);
2088                 lli->lli_posix_acl = md->posix_acl;
2089                 spin_unlock(&lli->lli_lock);
2090         }
2091 #endif
2092         inode->i_ino = cl_fid_build_ino(&body->mbo_fid1,
2093                                         sbi->ll_flags & LL_SBI_32BIT_API);
2094         inode->i_generation = cl_fid_build_gen(&body->mbo_fid1);
2095
2096         if (body->mbo_valid & OBD_MD_FLATIME) {
2097                 if (body->mbo_atime > inode->i_atime.tv_sec)
2098                         inode->i_atime.tv_sec = body->mbo_atime;
2099                 lli->lli_atime = body->mbo_atime;
2100         }
2101
2102         if (body->mbo_valid & OBD_MD_FLMTIME) {
2103                 if (body->mbo_mtime > inode->i_mtime.tv_sec) {
2104                         CDEBUG(D_INODE,
2105                                "setting ino %lu mtime from %lld to %llu\n",
2106                                inode->i_ino, (s64)inode->i_mtime.tv_sec,
2107                                body->mbo_mtime);
2108                         inode->i_mtime.tv_sec = body->mbo_mtime;
2109                 }
2110                 lli->lli_mtime = body->mbo_mtime;
2111         }
2112
2113         if (body->mbo_valid & OBD_MD_FLCTIME) {
2114                 if (body->mbo_ctime > inode->i_ctime.tv_sec)
2115                         inode->i_ctime.tv_sec = body->mbo_ctime;
2116                 lli->lli_ctime = body->mbo_ctime;
2117         }
2118
2119         /* Clear i_flags to remove S_NOSEC before permissions are updated */
2120         if (body->mbo_valid & OBD_MD_FLFLAGS)
2121                 ll_update_inode_flags(inode, body->mbo_flags);
2122         if (body->mbo_valid & OBD_MD_FLMODE)
2123                 inode->i_mode = (inode->i_mode & S_IFMT) |
2124                                 (body->mbo_mode & ~S_IFMT);
2125
2126         if (body->mbo_valid & OBD_MD_FLTYPE)
2127                 inode->i_mode = (inode->i_mode & ~S_IFMT) |
2128                                 (body->mbo_mode & S_IFMT);
2129
2130         LASSERT(inode->i_mode != 0);
2131         if (S_ISREG(inode->i_mode))
2132                 inode->i_blkbits = min(PTLRPC_MAX_BRW_BITS + 1,
2133                                        LL_MAX_BLKSIZE_BITS);
2134         else
2135                 inode->i_blkbits = inode->i_sb->s_blocksize_bits;
2136
2137         if (body->mbo_valid & OBD_MD_FLUID)
2138                 inode->i_uid = make_kuid(&init_user_ns, body->mbo_uid);
2139         if (body->mbo_valid & OBD_MD_FLGID)
2140                 inode->i_gid = make_kgid(&init_user_ns, body->mbo_gid);
2141         if (body->mbo_valid & OBD_MD_FLPROJID)
2142                 lli->lli_projid = body->mbo_projid;
2143         if (body->mbo_valid & OBD_MD_FLNLINK)
2144                 set_nlink(inode, body->mbo_nlink);
2145         if (body->mbo_valid & OBD_MD_FLRDEV)
2146                 inode->i_rdev = old_decode_dev(body->mbo_rdev);
2147
2148         if (body->mbo_valid & OBD_MD_FLID) {
2149                 /* FID shouldn't be changed! */
2150                 if (fid_is_sane(&lli->lli_fid)) {
2151                         LASSERTF(lu_fid_eq(&lli->lli_fid, &body->mbo_fid1),
2152                                  "Trying to change FID "DFID
2153                                  " to the "DFID", inode "DFID"(%p)\n",
2154                                  PFID(&lli->lli_fid), PFID(&body->mbo_fid1),
2155                                  PFID(ll_inode2fid(inode)), inode);
2156                 } else {
2157                         lli->lli_fid = body->mbo_fid1;
2158                 }
2159         }
2160
2161         LASSERT(fid_seq(&lli->lli_fid) != 0);
2162
2163         if (body->mbo_valid & OBD_MD_FLSIZE) {
2164                 i_size_write(inode, body->mbo_size);
2165
2166                 CDEBUG(D_VFSTRACE, "inode="DFID", updating i_size %llu\n",
2167                        PFID(ll_inode2fid(inode)),
2168                        (unsigned long long)body->mbo_size);
2169
2170                 if (body->mbo_valid & OBD_MD_FLBLOCKS)
2171                         inode->i_blocks = body->mbo_blocks;
2172         }
2173
2174         if (body->mbo_valid & OBD_MD_TSTATE) {
2175                 /* Set LLIF_FILE_RESTORING if restore ongoing and
2176                  * clear it when done to ensure to start again
2177                  * glimpsing updated attrs
2178                  */
2179                 if (body->mbo_t_state & MS_RESTORE)
2180                         ll_file_set_flag(lli, LLIF_FILE_RESTORING);
2181                 else
2182                         ll_file_clear_flag(lli, LLIF_FILE_RESTORING);
2183         }
2184
2185         return 0;
2186 }
2187
2188 int ll_read_inode2(struct inode *inode, void *opaque)
2189 {
2190         struct lustre_md *md = opaque;
2191         struct ll_inode_info *lli = ll_i2info(inode);
2192         int     rc;
2193         ENTRY;
2194
2195         CDEBUG(D_VFSTRACE, "VFS Op:inode="DFID"(%p)\n",
2196                PFID(&lli->lli_fid), inode);
2197
2198         /* Core attributes from the MDS first.  This is a new inode, and
2199          * the VFS doesn't zero times in the core inode so we have to do
2200          * it ourselves.  They will be overwritten by either MDS or OST
2201          * attributes - we just need to make sure they aren't newer.
2202          */
2203         inode->i_mtime.tv_sec = 0;
2204         inode->i_atime.tv_sec = 0;
2205         inode->i_ctime.tv_sec = 0;
2206         inode->i_rdev = 0;
2207         rc = ll_update_inode(inode, md);
2208         if (rc != 0)
2209                 RETURN(rc);
2210
2211         /* OIDEBUG(inode); */
2212
2213 #ifdef HAVE_BACKING_DEV_INFO
2214         /* initializing backing dev info. */
2215         inode->i_mapping->backing_dev_info = &s2lsi(inode->i_sb)->lsi_bdi;
2216 #endif
2217         if (S_ISREG(inode->i_mode)) {
2218                 struct ll_sb_info *sbi = ll_i2sbi(inode);
2219                 inode->i_op = &ll_file_inode_operations;
2220                 inode->i_fop = sbi->ll_fop;
2221                 inode->i_mapping->a_ops = (struct address_space_operations *)&ll_aops;
2222                 EXIT;
2223         } else if (S_ISDIR(inode->i_mode)) {
2224                 inode->i_op = &ll_dir_inode_operations;
2225                 inode->i_fop = &ll_dir_operations;
2226                 EXIT;
2227         } else if (S_ISLNK(inode->i_mode)) {
2228                 inode->i_op = &ll_fast_symlink_inode_operations;
2229                 EXIT;
2230         } else {
2231                 inode->i_op = &ll_special_inode_operations;
2232
2233                 init_special_inode(inode, inode->i_mode,
2234                                    inode->i_rdev);
2235
2236                 EXIT;
2237         }
2238
2239         return 0;
2240 }
2241
2242 void ll_delete_inode(struct inode *inode)
2243 {
2244         struct ll_inode_info *lli = ll_i2info(inode);
2245         struct address_space *mapping = &inode->i_data;
2246         unsigned long nrpages;
2247         unsigned long flags;
2248
2249         ENTRY;
2250
2251         if (S_ISREG(inode->i_mode) && lli->lli_clob != NULL) {
2252                 /* It is last chance to write out dirty pages,
2253                  * otherwise we may lose data while umount.
2254                  *
2255                  * If i_nlink is 0 then just discard data. This is safe because
2256                  * local inode gets i_nlink 0 from server only for the last
2257                  * unlink, so that file is not opened somewhere else
2258                  */
2259                 cl_sync_file_range(inode, 0, OBD_OBJECT_EOF, inode->i_nlink ?
2260                                    CL_FSYNC_LOCAL : CL_FSYNC_DISCARD, 1);
2261         }
2262         truncate_inode_pages_final(mapping);
2263
2264         /* Workaround for LU-118: Note nrpages may not be totally updated when
2265          * truncate_inode_pages() returns, as there can be a page in the process
2266          * of deletion (inside __delete_from_page_cache()) in the specified
2267          * range. Thus mapping->nrpages can be non-zero when this function
2268          * returns even after truncation of the whole mapping.  Only do this if
2269          * npages isn't already zero.
2270          */
2271         nrpages = mapping->nrpages;
2272         if (nrpages) {
2273                 xa_lock_irqsave(&mapping->i_pages, flags);
2274                 nrpages = mapping->nrpages;
2275                 xa_unlock_irqrestore(&mapping->i_pages, flags);
2276         } /* Workaround end */
2277
2278         LASSERTF(nrpages == 0, "%s: inode="DFID"(%p) nrpages=%lu, "
2279                  "see https://jira.whamcloud.com/browse/LU-118\n",
2280                  ll_i2sbi(inode)->ll_fsname,
2281                  PFID(ll_inode2fid(inode)), inode, nrpages);
2282
2283         ll_clear_inode(inode);
2284         clear_inode(inode);
2285
2286         EXIT;
2287 }
2288
2289 int ll_iocontrol(struct inode *inode, struct file *file,
2290                  unsigned int cmd, unsigned long arg)
2291 {
2292         struct ll_sb_info *sbi = ll_i2sbi(inode);
2293         struct ptlrpc_request *req = NULL;
2294         int rc, flags = 0;
2295         ENTRY;
2296
2297         switch (cmd) {
2298         case FS_IOC_GETFLAGS: {
2299                 struct mdt_body *body;
2300                 struct md_op_data *op_data;
2301
2302                 op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL,
2303                                              0, 0, LUSTRE_OPC_ANY,
2304                                              NULL);
2305                 if (IS_ERR(op_data))
2306                         RETURN(PTR_ERR(op_data));
2307
2308                 op_data->op_valid = OBD_MD_FLFLAGS;
2309                 rc = md_getattr(sbi->ll_md_exp, op_data, &req);
2310                 ll_finish_md_op_data(op_data);
2311                 if (rc) {
2312                         CERROR("%s: failure inode "DFID": rc = %d\n",
2313                                sbi->ll_md_exp->exp_obd->obd_name,
2314                                PFID(ll_inode2fid(inode)), rc);
2315                         RETURN(-abs(rc));
2316                 }
2317
2318                 body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
2319
2320                 flags = body->mbo_flags;
2321
2322                 ptlrpc_req_finished(req);
2323
2324                 RETURN(put_user(flags, (int __user *)arg));
2325         }
2326         case FS_IOC_SETFLAGS: {
2327                 struct iattr *attr;
2328                 struct md_op_data *op_data;
2329                 struct cl_object *obj;
2330                 struct fsxattr fa = { 0 };
2331
2332                 if (get_user(flags, (int __user *)arg))
2333                         RETURN(-EFAULT);
2334
2335                 fa.fsx_projid = ll_i2info(inode)->lli_projid;
2336                 if (flags & LUSTRE_PROJINHERIT_FL)
2337                         fa.fsx_xflags = FS_XFLAG_PROJINHERIT;
2338
2339                 rc = ll_ioctl_check_project(inode, &fa);
2340                 if (rc)
2341                         RETURN(rc);
2342
2343                 op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL, 0, 0,
2344                                              LUSTRE_OPC_ANY, NULL);
2345                 if (IS_ERR(op_data))
2346                         RETURN(PTR_ERR(op_data));
2347
2348                 op_data->op_attr_flags = flags;
2349                 op_data->op_xvalid |= OP_XVALID_FLAGS;
2350                 rc = md_setattr(sbi->ll_md_exp, op_data, NULL, 0, &req);
2351                 ll_finish_md_op_data(op_data);
2352                 ptlrpc_req_finished(req);
2353                 if (rc)
2354                         RETURN(rc);
2355
2356                 ll_update_inode_flags(inode, flags);
2357
2358                 obj = ll_i2info(inode)->lli_clob;
2359                 if (obj == NULL)
2360                         RETURN(0);
2361
2362                 OBD_ALLOC_PTR(attr);
2363                 if (attr == NULL)
2364                         RETURN(-ENOMEM);
2365
2366                 rc = cl_setattr_ost(obj, attr, OP_XVALID_FLAGS, flags);
2367
2368                 OBD_FREE_PTR(attr);
2369                 RETURN(rc);
2370         }
2371         default:
2372                 RETURN(-ENOSYS);
2373         }
2374
2375         RETURN(0);
2376 }
2377
2378 int ll_flush_ctx(struct inode *inode)
2379 {
2380         struct ll_sb_info  *sbi = ll_i2sbi(inode);
2381
2382         CDEBUG(D_SEC, "flush context for user %d\n",
2383                from_kuid(&init_user_ns, current_uid()));
2384
2385         obd_set_info_async(NULL, sbi->ll_md_exp,
2386                            sizeof(KEY_FLUSH_CTX), KEY_FLUSH_CTX,
2387                            0, NULL, NULL);
2388         obd_set_info_async(NULL, sbi->ll_dt_exp,
2389                            sizeof(KEY_FLUSH_CTX), KEY_FLUSH_CTX,
2390                            0, NULL, NULL);
2391         return 0;
2392 }
2393
2394 /* umount -f client means force down, don't save state */
2395 void ll_umount_begin(struct super_block *sb)
2396 {
2397         struct ll_sb_info *sbi = ll_s2sbi(sb);
2398         struct obd_device *obd;
2399         struct obd_ioctl_data *ioc_data;
2400         struct l_wait_info lwi;
2401         wait_queue_head_t waitq;
2402         ENTRY;
2403
2404         CDEBUG(D_VFSTRACE, "VFS Op: superblock %p count %d active %d\n", sb,
2405                sb->s_count, atomic_read(&sb->s_active));
2406
2407         obd = class_exp2obd(sbi->ll_md_exp);
2408         if (obd == NULL) {
2409                 CERROR("Invalid MDC connection handle %#llx\n",
2410                        sbi->ll_md_exp->exp_handle.h_cookie);
2411                 EXIT;
2412                 return;
2413         }
2414         obd->obd_force = 1;
2415
2416         obd = class_exp2obd(sbi->ll_dt_exp);
2417         if (obd == NULL) {
2418                 CERROR("Invalid LOV connection handle %#llx\n",
2419                        sbi->ll_dt_exp->exp_handle.h_cookie);
2420                 EXIT;
2421                 return;
2422         }
2423         obd->obd_force = 1;
2424
2425         OBD_ALLOC_PTR(ioc_data);
2426         if (ioc_data) {
2427                 obd_iocontrol(IOC_OSC_SET_ACTIVE, sbi->ll_md_exp,
2428                               sizeof *ioc_data, ioc_data, NULL);
2429
2430                 obd_iocontrol(IOC_OSC_SET_ACTIVE, sbi->ll_dt_exp,
2431                               sizeof *ioc_data, ioc_data, NULL);
2432
2433                 OBD_FREE_PTR(ioc_data);
2434         }
2435
2436         /* Really, we'd like to wait until there are no requests outstanding,
2437          * and then continue.  For now, we just periodically checking for vfs
2438          * to decrement mnt_cnt and hope to finish it within 10sec.
2439          */
2440         init_waitqueue_head(&waitq);
2441         lwi = LWI_TIMEOUT_INTERVAL(cfs_time_seconds(10),
2442                                    cfs_time_seconds(1), NULL, NULL);
2443         l_wait_event(waitq, may_umount(sbi->ll_mnt.mnt), &lwi);
2444
2445         EXIT;
2446 }
2447
2448 int ll_remount_fs(struct super_block *sb, int *flags, char *data)
2449 {
2450         struct ll_sb_info *sbi = ll_s2sbi(sb);
2451         char *profilenm = get_profile_name(sb);
2452         int err;
2453         __u32 read_only;
2454
2455         if ((*flags & MS_RDONLY) != (sb->s_flags & SB_RDONLY)) {
2456                 read_only = *flags & MS_RDONLY;
2457                 err = obd_set_info_async(NULL, sbi->ll_md_exp,
2458                                          sizeof(KEY_READ_ONLY),
2459                                          KEY_READ_ONLY, sizeof(read_only),
2460                                          &read_only, NULL);
2461                 if (err) {
2462                         LCONSOLE_WARN("Failed to remount %s %s (%d)\n",
2463                                       profilenm, read_only ?
2464                                       "read-only" : "read-write", err);
2465                         return err;
2466                 }
2467
2468                 if (read_only)
2469                         sb->s_flags |= SB_RDONLY;
2470                 else
2471                         sb->s_flags &= ~SB_RDONLY;
2472
2473                 if (sbi->ll_flags & LL_SBI_VERBOSE)
2474                         LCONSOLE_WARN("Remounted %s %s\n", profilenm,
2475                                       read_only ?  "read-only" : "read-write");
2476         }
2477         return 0;
2478 }
2479
2480 /**
2481  * Cleanup the open handle that is cached on MDT-side.
2482  *
2483  * For open case, the client side open handling thread may hit error
2484  * after the MDT grant the open. Under such case, the client should
2485  * send close RPC to the MDT as cleanup; otherwise, the open handle
2486  * on the MDT will be leaked there until the client umount or evicted.
2487  *
2488  * In further, if someone unlinked the file, because the open handle
2489  * holds the reference on such file/object, then it will block the
2490  * subsequent threads that want to locate such object via FID.
2491  *
2492  * \param[in] sb        super block for this file-system
2493  * \param[in] open_req  pointer to the original open request
2494  */
2495 void ll_open_cleanup(struct super_block *sb, struct ptlrpc_request *open_req)
2496 {
2497         struct mdt_body                 *body;
2498         struct md_op_data               *op_data;
2499         struct ptlrpc_request           *close_req = NULL;
2500         struct obd_export               *exp       = ll_s2sbi(sb)->ll_md_exp;
2501         ENTRY;
2502
2503         body = req_capsule_server_get(&open_req->rq_pill, &RMF_MDT_BODY);
2504         OBD_ALLOC_PTR(op_data);
2505         if (op_data == NULL) {
2506                 CWARN("%s: cannot allocate op_data to release open handle for "
2507                       DFID"\n", ll_s2sbi(sb)->ll_fsname, PFID(&body->mbo_fid1));
2508
2509                 RETURN_EXIT;
2510         }
2511
2512         op_data->op_fid1 = body->mbo_fid1;
2513         op_data->op_open_handle = body->mbo_open_handle;
2514         op_data->op_mod_time = ktime_get_real_seconds();
2515         md_close(exp, op_data, NULL, &close_req);
2516         ptlrpc_req_finished(close_req);
2517         ll_finish_md_op_data(op_data);
2518
2519         EXIT;
2520 }
2521
2522 int ll_prep_inode(struct inode **inode, struct ptlrpc_request *req,
2523                   struct super_block *sb, struct lookup_intent *it)
2524 {
2525         struct ll_sb_info *sbi = NULL;
2526         struct lustre_md md = { NULL };
2527         bool default_lmv_deleted = false;
2528         int rc;
2529
2530         ENTRY;
2531
2532         LASSERT(*inode || sb);
2533         sbi = sb ? ll_s2sbi(sb) : ll_i2sbi(*inode);
2534         rc = md_get_lustre_md(sbi->ll_md_exp, req, sbi->ll_dt_exp,
2535                               sbi->ll_md_exp, &md);
2536         if (rc != 0)
2537                 GOTO(out, rc);
2538
2539         /*
2540          * clear default_lmv only if intent_getattr reply doesn't contain it.
2541          * but it needs to be done after iget, check this early because
2542          * ll_update_lsm_md() may change md.
2543          */
2544         if (it && (it->it_op & (IT_LOOKUP | IT_GETATTR)) &&
2545             S_ISDIR(md.body->mbo_mode) && !md.default_lmv)
2546                 default_lmv_deleted = true;
2547
2548         if (*inode) {
2549                 rc = ll_update_inode(*inode, &md);
2550                 if (rc != 0)
2551                         GOTO(out, rc);
2552         } else {
2553                 LASSERT(sb != NULL);
2554
2555                 /*
2556                  * At this point server returns to client's same fid as client
2557                  * generated for creating. So using ->fid1 is okay here.
2558                  */
2559                 if (!fid_is_sane(&md.body->mbo_fid1)) {
2560                         CERROR("%s: Fid is insane "DFID"\n",
2561                                 sbi->ll_fsname,
2562                                 PFID(&md.body->mbo_fid1));
2563                         GOTO(out, rc = -EINVAL);
2564                 }
2565
2566                 *inode = ll_iget(sb, cl_fid_build_ino(&md.body->mbo_fid1,
2567                                              sbi->ll_flags & LL_SBI_32BIT_API),
2568                                  &md);
2569                 if (IS_ERR(*inode)) {
2570 #ifdef CONFIG_LUSTRE_FS_POSIX_ACL
2571                         if (md.posix_acl) {
2572                                 posix_acl_release(md.posix_acl);
2573                                 md.posix_acl = NULL;
2574                         }
2575 #endif
2576                         rc = IS_ERR(*inode) ? PTR_ERR(*inode) : -ENOMEM;
2577                         *inode = NULL;
2578                         CERROR("new_inode -fatal: rc %d\n", rc);
2579                         GOTO(out, rc);
2580                 }
2581         }
2582
2583         /* Handling piggyback layout lock.
2584          * Layout lock can be piggybacked by getattr and open request.
2585          * The lsm can be applied to inode only if it comes with a layout lock
2586          * otherwise correct layout may be overwritten, for example:
2587          * 1. proc1: mdt returns a lsm but not granting layout
2588          * 2. layout was changed by another client
2589          * 3. proc2: refresh layout and layout lock granted
2590          * 4. proc1: to apply a stale layout */
2591         if (it != NULL && it->it_lock_mode != 0) {
2592                 struct lustre_handle lockh;
2593                 struct ldlm_lock *lock;
2594
2595                 lockh.cookie = it->it_lock_handle;
2596                 lock = ldlm_handle2lock(&lockh);
2597                 LASSERT(lock != NULL);
2598                 if (ldlm_has_layout(lock)) {
2599                         struct cl_object_conf conf;
2600
2601                         memset(&conf, 0, sizeof(conf));
2602                         conf.coc_opc = OBJECT_CONF_SET;
2603                         conf.coc_inode = *inode;
2604                         conf.coc_lock = lock;
2605                         conf.u.coc_layout = md.layout;
2606                         (void)ll_layout_conf(*inode, &conf);
2607                 }
2608                 LDLM_LOCK_PUT(lock);
2609         }
2610
2611         if (default_lmv_deleted)
2612                 ll_update_default_lsm_md(*inode, &md);
2613
2614         GOTO(out, rc = 0);
2615
2616 out:
2617         /* cleanup will be done if necessary */
2618         md_free_lustre_md(sbi->ll_md_exp, &md);
2619
2620         if (rc != 0 && it != NULL && it->it_op & IT_OPEN)
2621                 ll_open_cleanup(sb != NULL ? sb : (*inode)->i_sb, req);
2622
2623         return rc;
2624 }
2625
2626 int ll_obd_statfs(struct inode *inode, void __user *arg)
2627 {
2628         struct ll_sb_info *sbi = NULL;
2629         struct obd_export *exp;
2630         char *buf = NULL;
2631         struct obd_ioctl_data *data = NULL;
2632         __u32 type;
2633         int len = 0, rc;
2634
2635         if (!inode || !(sbi = ll_i2sbi(inode)))
2636                 GOTO(out_statfs, rc = -EINVAL);
2637
2638         rc = obd_ioctl_getdata(&buf, &len, arg);
2639         if (rc)
2640                 GOTO(out_statfs, rc);
2641
2642         data = (void*)buf;
2643         if (!data->ioc_inlbuf1 || !data->ioc_inlbuf2 ||
2644             !data->ioc_pbuf1 || !data->ioc_pbuf2)
2645                 GOTO(out_statfs, rc = -EINVAL);
2646
2647         if (data->ioc_inllen1 != sizeof(__u32) ||
2648             data->ioc_inllen2 != sizeof(__u32) ||
2649             data->ioc_plen1 != sizeof(struct obd_statfs) ||
2650             data->ioc_plen2 != sizeof(struct obd_uuid))
2651                 GOTO(out_statfs, rc = -EINVAL);
2652
2653         memcpy(&type, data->ioc_inlbuf1, sizeof(__u32));
2654         if (type & LL_STATFS_LMV)
2655                 exp = sbi->ll_md_exp;
2656         else if (type & LL_STATFS_LOV)
2657                 exp = sbi->ll_dt_exp;
2658         else
2659                 GOTO(out_statfs, rc = -ENODEV);
2660
2661         rc = obd_iocontrol(IOC_OBD_STATFS, exp, len, buf, NULL);
2662         if (rc)
2663                 GOTO(out_statfs, rc);
2664 out_statfs:
2665         OBD_FREE_LARGE(buf, len);
2666         return rc;
2667 }
2668
2669 /*
2670  * this is normally called in ll_fini_md_op_data(), but sometimes it needs to
2671  * be called early to avoid deadlock.
2672  */
2673 void ll_unlock_md_op_lsm(struct md_op_data *op_data)
2674 {
2675         if (op_data->op_mea2_sem) {
2676                 up_read(op_data->op_mea2_sem);
2677                 op_data->op_mea2_sem = NULL;
2678         }
2679
2680         if (op_data->op_mea1_sem) {
2681                 up_read(op_data->op_mea1_sem);
2682                 op_data->op_mea1_sem = NULL;
2683         }
2684 }
2685
2686 /* this function prepares md_op_data hint for passing it down to MD stack. */
2687 struct md_op_data *ll_prep_md_op_data(struct md_op_data *op_data,
2688                                       struct inode *i1, struct inode *i2,
2689                                       const char *name, size_t namelen,
2690                                       __u32 mode, enum md_op_code opc,
2691                                       void *data)
2692 {
2693         LASSERT(i1 != NULL);
2694
2695         if (name == NULL) {
2696                 /* Do not reuse namelen for something else. */
2697                 if (namelen != 0)
2698                         return ERR_PTR(-EINVAL);
2699         } else {
2700                 if (namelen > ll_i2sbi(i1)->ll_namelen)
2701                         return ERR_PTR(-ENAMETOOLONG);
2702
2703                 if (!lu_name_is_valid_2(name, namelen))
2704                         return ERR_PTR(-EINVAL);
2705         }
2706
2707         if (op_data == NULL)
2708                 OBD_ALLOC_PTR(op_data);
2709
2710         if (op_data == NULL)
2711                 return ERR_PTR(-ENOMEM);
2712
2713         ll_i2gids(op_data->op_suppgids, i1, i2);
2714         op_data->op_fid1 = *ll_inode2fid(i1);
2715         op_data->op_code = opc;
2716
2717         if (S_ISDIR(i1->i_mode)) {
2718                 down_read(&ll_i2info(i1)->lli_lsm_sem);
2719                 op_data->op_mea1_sem = &ll_i2info(i1)->lli_lsm_sem;
2720                 op_data->op_mea1 = ll_i2info(i1)->lli_lsm_md;
2721                 op_data->op_default_mea1 = ll_i2info(i1)->lli_default_lsm_md;
2722         }
2723
2724         if (i2) {
2725                 op_data->op_fid2 = *ll_inode2fid(i2);
2726                 if (S_ISDIR(i2->i_mode)) {
2727                         if (i2 != i1) {
2728                                 down_read(&ll_i2info(i2)->lli_lsm_sem);
2729                                 op_data->op_mea2_sem =
2730                                                 &ll_i2info(i2)->lli_lsm_sem;
2731                         }
2732                         op_data->op_mea2 = ll_i2info(i2)->lli_lsm_md;
2733                 }
2734         } else {
2735                 fid_zero(&op_data->op_fid2);
2736         }
2737
2738         if (ll_i2sbi(i1)->ll_flags & LL_SBI_64BIT_HASH)
2739                 op_data->op_cli_flags |= CLI_HASH64;
2740
2741         if (ll_need_32bit_api(ll_i2sbi(i1)))
2742                 op_data->op_cli_flags |= CLI_API32;
2743
2744         op_data->op_name = name;
2745         op_data->op_namelen = namelen;
2746         op_data->op_mode = mode;
2747         op_data->op_mod_time = ktime_get_real_seconds();
2748         op_data->op_fsuid = from_kuid(&init_user_ns, current_fsuid());
2749         op_data->op_fsgid = from_kgid(&init_user_ns, current_fsgid());
2750         op_data->op_cap = cfs_curproc_cap_pack();
2751         op_data->op_mds = 0;
2752         if ((opc == LUSTRE_OPC_CREATE) && (name != NULL) &&
2753              filename_is_volatile(name, namelen, &op_data->op_mds)) {
2754                 op_data->op_bias |= MDS_CREATE_VOLATILE;
2755         }
2756         op_data->op_data = data;
2757
2758         return op_data;
2759 }
2760
2761 void ll_finish_md_op_data(struct md_op_data *op_data)
2762 {
2763         ll_unlock_md_op_lsm(op_data);
2764         security_release_secctx(op_data->op_file_secctx,
2765                                 op_data->op_file_secctx_size);
2766         OBD_FREE_PTR(op_data);
2767 }
2768
2769 #ifdef HAVE_SUPEROPS_USE_DENTRY
2770 int ll_show_options(struct seq_file *seq, struct dentry *dentry)
2771 #else
2772 int ll_show_options(struct seq_file *seq, struct vfsmount *vfs)
2773 #endif
2774 {
2775         struct ll_sb_info *sbi;
2776
2777 #ifdef HAVE_SUPEROPS_USE_DENTRY
2778         LASSERT((seq != NULL) && (dentry != NULL));
2779         sbi = ll_s2sbi(dentry->d_sb);
2780 #else
2781         LASSERT((seq != NULL) && (vfs != NULL));
2782         sbi = ll_s2sbi(vfs->mnt_sb);
2783 #endif
2784
2785         if (sbi->ll_flags & LL_SBI_NOLCK)
2786                 seq_puts(seq, ",nolock");
2787
2788         /* "flock" is the default since 2.13, but it wasn't for many years,
2789          * so it is still useful to print this to show it is enabled.
2790          * Start to print "noflock" so it is now clear when flock is disabled.
2791          */
2792         if (sbi->ll_flags & LL_SBI_FLOCK)
2793                 seq_puts(seq, ",flock");
2794         else if (sbi->ll_flags & LL_SBI_LOCALFLOCK)
2795                 seq_puts(seq, ",localflock");
2796         else
2797                 seq_puts(seq, ",noflock");
2798
2799         if (sbi->ll_flags & LL_SBI_USER_XATTR)
2800                 seq_puts(seq, ",user_xattr");
2801
2802         if (sbi->ll_flags & LL_SBI_LAZYSTATFS)
2803                 seq_puts(seq, ",lazystatfs");
2804
2805         if (sbi->ll_flags & LL_SBI_USER_FID2PATH)
2806                 seq_puts(seq, ",user_fid2path");
2807
2808         if (sbi->ll_flags & LL_SBI_ALWAYS_PING)
2809                 seq_puts(seq, ",always_ping");
2810
2811         RETURN(0);
2812 }
2813
2814 /**
2815  * Get obd name by cmd, and copy out to user space
2816  */
2817 int ll_get_obd_name(struct inode *inode, unsigned int cmd, unsigned long arg)
2818 {
2819         struct ll_sb_info *sbi = ll_i2sbi(inode);
2820         struct obd_device *obd;
2821         ENTRY;
2822
2823         if (cmd == OBD_IOC_GETDTNAME)
2824                 obd = class_exp2obd(sbi->ll_dt_exp);
2825         else if (cmd == OBD_IOC_GETMDNAME)
2826                 obd = class_exp2obd(sbi->ll_md_exp);
2827         else
2828                 RETURN(-EINVAL);
2829
2830         if (!obd)
2831                 RETURN(-ENOENT);
2832
2833         if (copy_to_user((void __user *)arg, obd->obd_name,
2834                          strlen(obd->obd_name) + 1))
2835                 RETURN(-EFAULT);
2836
2837         RETURN(0);
2838 }
2839
2840 static char* ll_d_path(struct dentry *dentry, char *buf, int bufsize)
2841 {
2842         char *path = NULL;
2843
2844         struct path p;
2845
2846         p.dentry = dentry;
2847         p.mnt = current->fs->root.mnt;
2848         path_get(&p);
2849         path = d_path(&p, buf, bufsize);
2850         path_put(&p);
2851         return path;
2852 }
2853
2854 void ll_dirty_page_discard_warn(struct page *page, int ioret)
2855 {
2856         char *buf, *path = NULL;
2857         struct dentry *dentry = NULL;
2858         struct inode *inode = page->mapping->host;
2859
2860         /* this can be called inside spin lock so use GFP_ATOMIC. */
2861         buf = (char *)__get_free_page(GFP_ATOMIC);
2862         if (buf != NULL) {
2863                 dentry = d_find_alias(page->mapping->host);
2864                 if (dentry != NULL)
2865                         path = ll_d_path(dentry, buf, PAGE_SIZE);
2866         }
2867
2868         /* The below message is checked in recovery-small.sh test_24b */
2869         CDEBUG(D_WARNING,
2870                "%s: dirty page discard: %s/fid: "DFID"/%s may get corrupted "
2871                "(rc %d)\n", ll_i2sbi(inode)->ll_fsname,
2872                s2lsi(page->mapping->host->i_sb)->lsi_lmd->lmd_dev,
2873                PFID(ll_inode2fid(inode)),
2874                (path && !IS_ERR(path)) ? path : "", ioret);
2875
2876         if (dentry != NULL)
2877                 dput(dentry);
2878
2879         if (buf != NULL)
2880                 free_page((unsigned long)buf);
2881 }
2882
2883 ssize_t ll_copy_user_md(const struct lov_user_md __user *md,
2884                         struct lov_user_md **kbuf)
2885 {
2886         struct lov_user_md      lum;
2887         ssize_t                 lum_size;
2888         ENTRY;
2889
2890         if (copy_from_user(&lum, md, sizeof(lum)))
2891                 RETURN(-EFAULT);
2892
2893         lum_size = ll_lov_user_md_size(&lum);
2894         if (lum_size < 0)
2895                 RETURN(lum_size);
2896
2897         OBD_ALLOC_LARGE(*kbuf, lum_size);
2898         if (*kbuf == NULL)
2899                 RETURN(-ENOMEM);
2900
2901         if (copy_from_user(*kbuf, md, lum_size) != 0) {
2902                 OBD_FREE_LARGE(*kbuf, lum_size);
2903                 RETURN(-EFAULT);
2904         }
2905
2906         RETURN(lum_size);
2907 }
2908
2909 /*
2910  * Compute llite root squash state after a change of root squash
2911  * configuration setting or add/remove of a lnet nid
2912  */
2913 void ll_compute_rootsquash_state(struct ll_sb_info *sbi)
2914 {
2915         struct root_squash_info *squash = &sbi->ll_squash;
2916         int i;
2917         bool matched;
2918         struct lnet_process_id id;
2919
2920         /* Update norootsquash flag */
2921         spin_lock(&squash->rsi_lock);
2922         if (list_empty(&squash->rsi_nosquash_nids))
2923                 sbi->ll_flags &= ~LL_SBI_NOROOTSQUASH;
2924         else {
2925                 /* Do not apply root squash as soon as one of our NIDs is
2926                  * in the nosquash_nids list */
2927                 matched = false;
2928                 i = 0;
2929                 while (LNetGetId(i++, &id) != -ENOENT) {
2930                         if (LNET_NETTYP(LNET_NIDNET(id.nid)) == LOLND)
2931                                 continue;
2932                         if (cfs_match_nid(id.nid, &squash->rsi_nosquash_nids)) {
2933                                 matched = true;
2934                                 break;
2935                         }
2936                 }
2937                 if (matched)
2938                         sbi->ll_flags |= LL_SBI_NOROOTSQUASH;
2939                 else
2940                         sbi->ll_flags &= ~LL_SBI_NOROOTSQUASH;
2941         }
2942         spin_unlock(&squash->rsi_lock);
2943 }
2944
2945 /**
2946  * Parse linkea content to extract information about a given hardlink
2947  *
2948  * \param[in]   ldata      - Initialized linkea data
2949  * \param[in]   linkno     - Link identifier
2950  * \param[out]  parent_fid - The entry's parent FID
2951  * \param[out]  ln         - Entry name destination buffer
2952  *
2953  * \retval 0 on success
2954  * \retval Appropriate negative error code on failure
2955  */
2956 static int ll_linkea_decode(struct linkea_data *ldata, unsigned int linkno,
2957                             struct lu_fid *parent_fid, struct lu_name *ln)
2958 {
2959         unsigned int    idx;
2960         int             rc;
2961         ENTRY;
2962
2963         rc = linkea_init_with_rec(ldata);
2964         if (rc < 0)
2965                 RETURN(rc);
2966
2967         if (linkno >= ldata->ld_leh->leh_reccount)
2968                 /* beyond last link */
2969                 RETURN(-ENODATA);
2970
2971         linkea_first_entry(ldata);
2972         for (idx = 0; ldata->ld_lee != NULL; idx++) {
2973                 linkea_entry_unpack(ldata->ld_lee, &ldata->ld_reclen, ln,
2974                                     parent_fid);
2975                 if (idx == linkno)
2976                         break;
2977
2978                 linkea_next_entry(ldata);
2979         }
2980
2981         if (idx < linkno)
2982                 RETURN(-ENODATA);
2983
2984         RETURN(0);
2985 }
2986
2987 /**
2988  * Get parent FID and name of an identified link. Operation is performed for
2989  * a given link number, letting the caller iterate over linkno to list one or
2990  * all links of an entry.
2991  *
2992  * \param[in]     file - File descriptor against which to perform the operation
2993  * \param[in,out] arg  - User-filled structure containing the linkno to operate
2994  *                       on and the available size. It is eventually filled with
2995  *                       the requested information or left untouched on error
2996  *
2997  * \retval - 0 on success
2998  * \retval - Appropriate negative error code on failure
2999  */
3000 int ll_getparent(struct file *file, struct getparent __user *arg)
3001 {
3002         struct inode            *inode = file_inode(file);
3003         struct linkea_data      *ldata;
3004         struct lu_buf            buf = LU_BUF_NULL;
3005         struct lu_name           ln;
3006         struct lu_fid            parent_fid;
3007         __u32                    linkno;
3008         __u32                    name_size;
3009         int                      rc;
3010
3011         ENTRY;
3012
3013         if (!cfs_capable(CFS_CAP_DAC_READ_SEARCH) &&
3014             !(ll_i2sbi(inode)->ll_flags & LL_SBI_USER_FID2PATH))
3015                 RETURN(-EPERM);
3016
3017         if (get_user(name_size, &arg->gp_name_size))
3018                 RETURN(-EFAULT);
3019
3020         if (get_user(linkno, &arg->gp_linkno))
3021                 RETURN(-EFAULT);
3022
3023         if (name_size > PATH_MAX)
3024                 RETURN(-EINVAL);
3025
3026         OBD_ALLOC(ldata, sizeof(*ldata));
3027         if (ldata == NULL)
3028                 RETURN(-ENOMEM);
3029
3030         rc = linkea_data_new(ldata, &buf);
3031         if (rc < 0)
3032                 GOTO(ldata_free, rc);
3033
3034         rc = ll_xattr_list(inode, XATTR_NAME_LINK, XATTR_TRUSTED_T, buf.lb_buf,
3035                            buf.lb_len, OBD_MD_FLXATTR);
3036         if (rc < 0)
3037                 GOTO(lb_free, rc);
3038
3039         rc = ll_linkea_decode(ldata, linkno, &parent_fid, &ln);
3040         if (rc < 0)
3041                 GOTO(lb_free, rc);
3042
3043         if (ln.ln_namelen >= name_size)
3044                 GOTO(lb_free, rc = -EOVERFLOW);
3045
3046         if (copy_to_user(&arg->gp_fid, &parent_fid, sizeof(arg->gp_fid)))
3047                 GOTO(lb_free, rc = -EFAULT);
3048
3049         if (copy_to_user(&arg->gp_name, ln.ln_name, ln.ln_namelen))
3050                 GOTO(lb_free, rc = -EFAULT);
3051
3052         if (put_user('\0', arg->gp_name + ln.ln_namelen))
3053                 GOTO(lb_free, rc = -EFAULT);
3054
3055 lb_free:
3056         lu_buf_free(&buf);
3057 ldata_free:
3058         OBD_FREE(ldata, sizeof(*ldata));
3059
3060         RETURN(rc);
3061 }