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 #include <lustre_ha.h>
  55 #include <lustre_dlm.h>
  56 #include <lprocfs_status.h>
  57 #include <lustre_disk.h>
  58 #include <uapi/linux/lustre/lustre_param.h>
  59 #include <lustre_log.h>
  60 #include <cl_object.h>
  61 #include <obd_cksum.h>
  62 #include "llite_internal.h"
  63
  64 struct kmem_cache *ll_file_data_slab;
  65
  66 #ifndef log2
  67 #define log2(n) ffz(~(n))
  68 #endif
  69
  70 static struct ll_sb_info *ll_init_sbi(void)
  71 {
  72         struct ll_sb_info *sbi = NULL;
  73         unsigned long pages;
  74         unsigned long lru_page_max;
  75         struct sysinfo si;
  76         int rc;
  77         int i;
  78
  79         ENTRY;
  80
  81         OBD_ALLOC_PTR(sbi);
  82         if (sbi == NULL)
  83                 RETURN(ERR_PTR(-ENOMEM));
  84
  85         rc = pcc_super_init(&sbi->ll_pcc_super);
  86         if (rc < 0)
  87                 GOTO(out_sbi, rc);
  88
  89         spin_lock_init(&sbi->ll_lock);
  90         mutex_init(&sbi->ll_lco.lco_lock);
  91         spin_lock_init(&sbi->ll_pp_extent_lock);
  92         spin_lock_init(&sbi->ll_process_lock);
  93         sbi->ll_rw_stats_on = 0;
  94         sbi->ll_statfs_max_age = OBD_STATFS_CACHE_SECONDS;
  95
  96         si_meminfo(&si);
  97         pages = si.totalram - si.totalhigh;
  98         lru_page_max = pages / 2;
  99
 100         sbi->ll_ra_info.ra_async_max_active = 0;
 101         sbi->ll_ra_info.ll_readahead_wq =
 102                 alloc_workqueue("ll-readahead-wq", WQ_UNBOUND,
 103                                 sbi->ll_ra_info.ra_async_max_active);
 104         if (!sbi->ll_ra_info.ll_readahead_wq)
 105                 GOTO(out_pcc, rc = -ENOMEM);
 106
 107         /* initialize ll_cache data */
 108         sbi->ll_cache = cl_cache_init(lru_page_max);
 109         if (sbi->ll_cache == NULL)
 110                 GOTO(out_destroy_ra, rc = -ENOMEM);
 111
 112         sbi->ll_ra_info.ra_max_pages_per_file = min(pages / 32,
 113                                            SBI_DEFAULT_READAHEAD_MAX);
 114         sbi->ll_ra_info.ra_async_pages_per_file_threshold =
 115                                 sbi->ll_ra_info.ra_max_pages_per_file;
 116         sbi->ll_ra_info.ra_max_pages = sbi->ll_ra_info.ra_max_pages_per_file;
 117         sbi->ll_ra_info.ra_max_read_ahead_whole_pages = -1;
 118
 119         sbi->ll_flags |= LL_SBI_VERBOSE;
 120 #ifdef ENABLE_CHECKSUM
 121         sbi->ll_flags |= LL_SBI_CHECKSUM;
 122 #endif
 123 #ifdef ENABLE_FLOCK
 124         sbi->ll_flags |= LL_SBI_FLOCK;
 125 #endif
 126
 127 #ifdef HAVE_LRU_RESIZE_SUPPORT
 128         sbi->ll_flags |= LL_SBI_LRU_RESIZE;
 129 #endif
 130         sbi->ll_flags |= LL_SBI_LAZYSTATFS;
 131
 132         for (i = 0; i <= LL_PROCESS_HIST_MAX; i++) {
 133                 spin_lock_init(&sbi->ll_rw_extents_info.pp_extents[i].
 134                                pp_r_hist.oh_lock);
 135                 spin_lock_init(&sbi->ll_rw_extents_info.pp_extents[i].
 136                                pp_w_hist.oh_lock);
 137         }
 138
 139         /* metadata statahead is enabled by default */
 140         sbi->ll_sa_running_max = LL_SA_RUNNING_DEF;
 141         sbi->ll_sa_max = LL_SA_RPC_DEF;
 142         atomic_set(&sbi->ll_sa_total, 0);
 143         atomic_set(&sbi->ll_sa_wrong, 0);
 144         atomic_set(&sbi->ll_sa_running, 0);
 145         atomic_set(&sbi->ll_agl_total, 0);
 146         sbi->ll_flags |= LL_SBI_AGL_ENABLED;
 147         sbi->ll_flags |= LL_SBI_FAST_READ;
 148         sbi->ll_flags |= LL_SBI_TINY_WRITE;
 149
 150         /* root squash */
 151         sbi->ll_squash.rsi_uid = 0;
 152         sbi->ll_squash.rsi_gid = 0;
 153         INIT_LIST_HEAD(&sbi->ll_squash.rsi_nosquash_nids);
 154         spin_lock_init(&sbi->ll_squash.rsi_lock);
 155
 156         /* Per-filesystem file heat */
 157         sbi->ll_heat_decay_weight = SBI_DEFAULT_HEAT_DECAY_WEIGHT;
 158         sbi->ll_heat_period_second = SBI_DEFAULT_HEAT_PERIOD_SECOND;
 159         RETURN(sbi);
 160 out_destroy_ra:
 161         destroy_workqueue(sbi->ll_ra_info.ll_readahead_wq);
 162 out_pcc:
 163         pcc_super_fini(&sbi->ll_pcc_super);
 164 out_sbi:
 165         OBD_FREE_PTR(sbi);
 166         RETURN(ERR_PTR(rc));
 167 }
 168
 169 static void ll_free_sbi(struct super_block *sb)
 170 {
 171         struct ll_sb_info *sbi = ll_s2sbi(sb);
 172         ENTRY;
 173
 174         if (sbi != NULL) {
 175                 if (!list_empty(&sbi->ll_squash.rsi_nosquash_nids))
 176                         cfs_free_nidlist(&sbi->ll_squash.rsi_nosquash_nids);
 177                 if (sbi->ll_ra_info.ll_readahead_wq)
 178                         destroy_workqueue(sbi->ll_ra_info.ll_readahead_wq);
 179                 if (sbi->ll_cache != NULL) {
 180                         cl_cache_decref(sbi->ll_cache);
 181                         sbi->ll_cache = NULL;
 182                 }
 183                 pcc_super_fini(&sbi->ll_pcc_super);
 184                 OBD_FREE(sbi, sizeof(*sbi));
 185         }
 186         EXIT;
 187 }
 188
 189 static int client_common_fill_super(struct super_block *sb, char *md, char *dt,
 190                                     struct vfsmount *mnt)
 191 {
 192         struct inode *root = NULL;
 193         struct ll_sb_info *sbi = ll_s2sbi(sb);
 194         struct obd_statfs *osfs = NULL;
 195         struct ptlrpc_request *request = NULL;
 196         struct obd_connect_data *data = NULL;
 197         struct obd_uuid *uuid;
 198         struct md_op_data *op_data;
 199         struct lustre_md lmd;
 200         u64 valid;
 201         int size, err, checksum;
 202
 203         ENTRY;
 204         sbi->ll_md_obd = class_name2obd(md);
 205         if (!sbi->ll_md_obd) {
 206                 CERROR("MD %s: not setup or attached\n", md);
 207                 RETURN(-EINVAL);
 208         }
 209
 210         OBD_ALLOC_PTR(data);
 211         if (data == NULL)
 212                 RETURN(-ENOMEM);
 213
 214         OBD_ALLOC_PTR(osfs);
 215         if (osfs == NULL) {
 216                 OBD_FREE_PTR(data);
 217                 RETURN(-ENOMEM);
 218         }
 219
 220         /* pass client page size via ocd_grant_blkbits, the server should report
 221          * back its backend blocksize for grant calculation purpose */
 222         data->ocd_grant_blkbits = PAGE_SHIFT;
 223
 224         /* indicate MDT features supported by this client */
 225         data->ocd_connect_flags = OBD_CONNECT_IBITS    | OBD_CONNECT_NODEVOH  |
 226                                   OBD_CONNECT_ATTRFID  | OBD_CONNECT_GRANT |
 227                                   OBD_CONNECT_VERSION  | OBD_CONNECT_BRW_SIZE |
 228                                   OBD_CONNECT_SRVLOCK  | OBD_CONNECT_TRUNCLOCK|
 229                                   OBD_CONNECT_MDS_CAPA | OBD_CONNECT_OSS_CAPA |
 230                                   OBD_CONNECT_CANCELSET | OBD_CONNECT_FID     |
 231                                   OBD_CONNECT_AT       | OBD_CONNECT_LOV_V3   |
 232                                   OBD_CONNECT_VBR | OBD_CONNECT_FULL20 |
 233                                   OBD_CONNECT_64BITHASH |
 234                                   OBD_CONNECT_EINPROGRESS |
 235                                   OBD_CONNECT_JOBSTATS | OBD_CONNECT_LVB_TYPE |
 236                                   OBD_CONNECT_LAYOUTLOCK | OBD_CONNECT_PINGLESS|
 237                                   OBD_CONNECT_MAX_EASIZE |
 238                                   OBD_CONNECT_FLOCK_DEAD |
 239                                   OBD_CONNECT_DISP_STRIPE | OBD_CONNECT_LFSCK |
 240                                   OBD_CONNECT_OPEN_BY_FID |
 241                                   OBD_CONNECT_DIR_STRIPE |
 242                                   OBD_CONNECT_BULK_MBITS | OBD_CONNECT_CKSUM |
 243                                   OBD_CONNECT_SUBTREE |
 244                                   OBD_CONNECT_MULTIMODRPCS |
 245                                   OBD_CONNECT_GRANT_PARAM |
 246                                   OBD_CONNECT_SHORTIO | OBD_CONNECT_FLAGS2;
 247
 248         data->ocd_connect_flags2 = OBD_CONNECT2_DIR_MIGRATE |
 249                                    OBD_CONNECT2_SUM_STATFS |
 250                                    OBD_CONNECT2_OVERSTRIPING |
 251                                    OBD_CONNECT2_FLR |
 252                                    OBD_CONNECT2_LOCK_CONVERT |
 253                                    OBD_CONNECT2_ARCHIVE_ID_ARRAY |
 254                                    OBD_CONNECT2_LSOM |
 255                                    OBD_CONNECT2_ASYNC_DISCARD |
 256                                    OBD_CONNECT2_PCC;
 257
 258 #ifdef HAVE_LRU_RESIZE_SUPPORT
 259         if (sbi->ll_flags & LL_SBI_LRU_RESIZE)
 260                 data->ocd_connect_flags |= OBD_CONNECT_LRU_RESIZE;
 261 #endif
 262 #ifdef CONFIG_FS_POSIX_ACL
 263         data->ocd_connect_flags |= OBD_CONNECT_ACL | OBD_CONNECT_UMASK |
 264                                    OBD_CONNECT_LARGE_ACL;
 265 #endif
 266
 267         data->ocd_cksum_types = obd_cksum_types_supported_client();
 268
 269         if (OBD_FAIL_CHECK(OBD_FAIL_MDC_LIGHTWEIGHT))
 270                 /* flag mdc connection as lightweight, only used for test
 271                  * purpose, use with care */
 272                 data->ocd_connect_flags |= OBD_CONNECT_LIGHTWEIGHT;
 273
 274         data->ocd_ibits_known = MDS_INODELOCK_FULL;
 275         data->ocd_version = LUSTRE_VERSION_CODE;
 276
 277         if (sb->s_flags & MS_RDONLY)
 278                 data->ocd_connect_flags |= OBD_CONNECT_RDONLY;
 279         if (sbi->ll_flags & LL_SBI_USER_XATTR)
 280                 data->ocd_connect_flags |= OBD_CONNECT_XATTR;
 281
 282 #ifdef MS_NOSEC
 283         /* Setting this indicates we correctly support S_NOSEC (See kernel
 284          * commit 9e1f1de02c2275d7172e18dc4e7c2065777611bf)
 285          */
 286         sb->s_flags |= MS_NOSEC;
 287 #endif
 288
 289         if (sbi->ll_flags & LL_SBI_FLOCK)
 290                 sbi->ll_fop = &ll_file_operations_flock;
 291         else if (sbi->ll_flags & LL_SBI_LOCALFLOCK)
 292                 sbi->ll_fop = &ll_file_operations;
 293         else
 294                 sbi->ll_fop = &ll_file_operations_noflock;
 295
 296         /* always ping even if server suppress_pings */
 297         if (sbi->ll_flags & LL_SBI_ALWAYS_PING)
 298                 data->ocd_connect_flags &= ~OBD_CONNECT_PINGLESS;
 299
 300         obd_connect_set_secctx(data);
 301
 302 #if defined(CONFIG_SECURITY)
 303         data->ocd_connect_flags2 |= OBD_CONNECT2_SELINUX_POLICY;
 304 #endif
 305
 306         data->ocd_brw_size = MD_MAX_BRW_SIZE;
 307
 308         err = obd_connect(NULL, &sbi->ll_md_exp, sbi->ll_md_obd,
 309                           &sbi->ll_sb_uuid, data, sbi->ll_cache);
 310         if (err == -EBUSY) {
 311                 LCONSOLE_ERROR_MSG(0x14f, "An MDT (md %s) is performing "
 312                                    "recovery, of which this client is not a "
 313                                    "part. Please wait for recovery to complete,"
 314                                    " abort, or time out.\n", md);
 315                 GOTO(out, err);
 316         } else if (err) {
 317                 CERROR("cannot connect to %s: rc = %d\n", md, err);
 318                 GOTO(out, err);
 319         }
 320
 321         sbi->ll_md_exp->exp_connect_data = *data;
 322
 323         err = obd_fid_init(sbi->ll_md_exp->exp_obd, sbi->ll_md_exp,
 324                            LUSTRE_SEQ_METADATA);
 325         if (err) {
 326                 CERROR("%s: Can't init metadata layer FID infrastructure, "
 327                        "rc = %d\n", sbi->ll_md_exp->exp_obd->obd_name, err);
 328                 GOTO(out_md, err);
 329         }
 330
 331         /* For mount, we only need fs info from MDT0, and also in DNE, it
 332          * can make sure the client can be mounted as long as MDT0 is
 333          * avaible */
 334         err = obd_statfs(NULL, sbi->ll_md_exp, osfs,
 335                         ktime_get_seconds() - sbi->ll_statfs_max_age,
 336                         OBD_STATFS_FOR_MDT0);
 337         if (err)
 338                 GOTO(out_md_fid, err);
 339
 340         /* This needs to be after statfs to ensure connect has finished.
 341          * Note that "data" does NOT contain the valid connect reply.
 342          * If connecting to a 1.8 server there will be no LMV device, so
 343          * we can access the MDC export directly and exp_connect_flags will
 344          * be non-zero, but if accessing an upgraded 2.1 server it will
 345          * have the correct flags filled in.
 346          * XXX: fill in the LMV exp_connect_flags from MDC(s). */
 347         valid = exp_connect_flags(sbi->ll_md_exp) & CLIENT_CONNECT_MDT_REQD;
 348         if (exp_connect_flags(sbi->ll_md_exp) != 0 &&
 349             valid != CLIENT_CONNECT_MDT_REQD) {
 350                 char *buf;
 351
 352                 OBD_ALLOC_WAIT(buf, PAGE_SIZE);
 353                 obd_connect_flags2str(buf, PAGE_SIZE,
 354                                       valid ^ CLIENT_CONNECT_MDT_REQD, 0, ",");
 355                 LCONSOLE_ERROR_MSG(0x170, "Server %s does not support "
 356                                    "feature(s) needed for correct operation "
 357                                    "of this client (%s). Please upgrade "
 358                                    "server or downgrade client.\n",
 359                                    sbi->ll_md_exp->exp_obd->obd_name, buf);
 360                 OBD_FREE(buf, PAGE_SIZE);
 361                 GOTO(out_md_fid, err = -EPROTO);
 362         }
 363
 364         size = sizeof(*data);
 365         err = obd_get_info(NULL, sbi->ll_md_exp, sizeof(KEY_CONN_DATA),
 366                            KEY_CONN_DATA,  &size, data);
 367         if (err) {
 368                 CERROR("%s: Get connect data failed: rc = %d\n",
 369                        sbi->ll_md_exp->exp_obd->obd_name, err);
 370                 GOTO(out_md_fid, err);
 371         }
 372
 373         LASSERT(osfs->os_bsize);
 374         sb->s_blocksize = osfs->os_bsize;
 375         sb->s_blocksize_bits = log2(osfs->os_bsize);
 376         sb->s_magic = LL_SUPER_MAGIC;
 377         sb->s_maxbytes = MAX_LFS_FILESIZE;
 378         sbi->ll_namelen = osfs->os_namelen;
 379         sbi->ll_mnt.mnt = current->fs->root.mnt;
 380
 381         if ((sbi->ll_flags & LL_SBI_USER_XATTR) &&
 382             !(data->ocd_connect_flags & OBD_CONNECT_XATTR)) {
 383                 LCONSOLE_INFO("Disabling user_xattr feature because "
 384                               "it is not supported on the server\n");
 385                 sbi->ll_flags &= ~LL_SBI_USER_XATTR;
 386         }
 387
 388         if (data->ocd_connect_flags & OBD_CONNECT_ACL) {
 389 #ifdef MS_POSIXACL
 390                 sb->s_flags |= MS_POSIXACL;
 391 #endif
 392                 sbi->ll_flags |= LL_SBI_ACL;
 393         } else {
 394                 LCONSOLE_INFO("client wants to enable acl, but mdt not!\n");
 395 #ifdef MS_POSIXACL
 396                 sb->s_flags &= ~MS_POSIXACL;
 397 #endif
 398                 sbi->ll_flags &= ~LL_SBI_ACL;
 399         }
 400
 401         if (data->ocd_connect_flags & OBD_CONNECT_64BITHASH)
 402                 sbi->ll_flags |= LL_SBI_64BIT_HASH;
 403
 404         if (data->ocd_connect_flags & OBD_CONNECT_LAYOUTLOCK)
 405                 sbi->ll_flags |= LL_SBI_LAYOUT_LOCK;
 406
 407         if (obd_connect_has_secctx(data))
 408                 sbi->ll_flags |= LL_SBI_FILE_SECCTX;
 409
 410         if (data->ocd_ibits_known & MDS_INODELOCK_XATTR) {
 411                 if (!(data->ocd_connect_flags & OBD_CONNECT_MAX_EASIZE)) {
 412                         LCONSOLE_INFO("%s: disabling xattr cache due to "
 413                                       "unknown maximum xattr size.\n", dt);
 414                 } else if (!sbi->ll_xattr_cache_set) {
 415                         /* If xattr_cache is already set (no matter 0 or 1)
 416                          * during processing llog, it won't be enabled here. */
 417                         sbi->ll_flags |= LL_SBI_XATTR_CACHE;
 418                         sbi->ll_xattr_cache_enabled = 1;
 419                 }
 420         }
 421
 422         sbi->ll_dt_obd = class_name2obd(dt);
 423         if (!sbi->ll_dt_obd) {
 424                 CERROR("DT %s: not setup or attached\n", dt);
 425                 GOTO(out_md_fid, err = -ENODEV);
 426         }
 427
 428         /* pass client page size via ocd_grant_blkbits, the server should report
 429          * back its backend blocksize for grant calculation purpose */
 430         data->ocd_grant_blkbits = PAGE_SHIFT;
 431
 432         /* indicate OST features supported by this client */
 433         data->ocd_connect_flags = OBD_CONNECT_GRANT | OBD_CONNECT_VERSION |
 434                                   OBD_CONNECT_REQPORTAL | OBD_CONNECT_BRW_SIZE |
 435                                   OBD_CONNECT_CANCELSET | OBD_CONNECT_FID |
 436                                   OBD_CONNECT_SRVLOCK | OBD_CONNECT_TRUNCLOCK|
 437                                   OBD_CONNECT_AT | OBD_CONNECT_OSS_CAPA |
 438                                   OBD_CONNECT_VBR | OBD_CONNECT_FULL20 |
 439                                   OBD_CONNECT_64BITHASH | OBD_CONNECT_MAXBYTES |
 440                                   OBD_CONNECT_EINPROGRESS |
 441                                   OBD_CONNECT_JOBSTATS | OBD_CONNECT_LVB_TYPE |
 442                                   OBD_CONNECT_LAYOUTLOCK |
 443                                   OBD_CONNECT_PINGLESS | OBD_CONNECT_LFSCK |
 444                                   OBD_CONNECT_BULK_MBITS | OBD_CONNECT_SHORTIO |
 445                                   OBD_CONNECT_FLAGS2 | OBD_CONNECT_GRANT_SHRINK;
 446
 447 /* The client currently advertises support for OBD_CONNECT_LOCKAHEAD_OLD so it
 448  * can interoperate with an older version of lockahead which was released prior
 449  * to landing in master. This support will be dropped when 2.13 development
 450  * starts.  At the point, we should not just drop the connect flag (below), we
 451  * should also remove the support in the code.
 452  *
 453  * Removing it means a few things:
 454  * 1. Remove this section here
 455  * 2. Remove CEF_NONBLOCK in ll_file_lockahead()
 456  * 3. Remove function exp_connect_lockahead_old
 457  * 4. Remove LDLM_FL_LOCKAHEAD_OLD_RESERVED in lustre_dlm_flags.h
 458  * */
 459 #if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(2, 12, 50, 0)
 460         data->ocd_connect_flags |= OBD_CONNECT_LOCKAHEAD_OLD;
 461 #endif
 462
 463         data->ocd_connect_flags2 = OBD_CONNECT2_LOCKAHEAD;
 464
 465         if (!OBD_FAIL_CHECK(OBD_FAIL_OSC_CONNECT_GRANT_PARAM))
 466                 data->ocd_connect_flags |= OBD_CONNECT_GRANT_PARAM;
 467
 468         /* OBD_CONNECT_CKSUM should always be set, even if checksums are
 469          * disabled by default, because it can still be enabled on the
 470          * fly via /sys. As a consequence, we still need to come to an
 471          * agreement on the supported algorithms at connect time
 472          */
 473         data->ocd_connect_flags |= OBD_CONNECT_CKSUM;
 474
 475         if (OBD_FAIL_CHECK(OBD_FAIL_OSC_CKSUM_ADLER_ONLY))
 476                 data->ocd_cksum_types = OBD_CKSUM_ADLER;
 477         else
 478                 data->ocd_cksum_types = obd_cksum_types_supported_client();
 479
 480 #ifdef HAVE_LRU_RESIZE_SUPPORT
 481         data->ocd_connect_flags |= OBD_CONNECT_LRU_RESIZE;
 482 #endif
 483         /* always ping even if server suppress_pings */
 484         if (sbi->ll_flags & LL_SBI_ALWAYS_PING)
 485                 data->ocd_connect_flags &= ~OBD_CONNECT_PINGLESS;
 486
 487         CDEBUG(D_RPCTRACE, "ocd_connect_flags: %#llx ocd_version: %d "
 488                "ocd_grant: %d\n", data->ocd_connect_flags,
 489                data->ocd_version, data->ocd_grant);
 490
 491         sbi->ll_dt_obd->obd_upcall.onu_owner = &sbi->ll_lco;
 492         sbi->ll_dt_obd->obd_upcall.onu_upcall = cl_ocd_update;
 493
 494         data->ocd_brw_size = DT_MAX_BRW_SIZE;
 495
 496         err = obd_connect(NULL, &sbi->ll_dt_exp, sbi->ll_dt_obd,
 497                           &sbi->ll_sb_uuid, data, sbi->ll_cache);
 498         if (err == -EBUSY) {
 499                 LCONSOLE_ERROR_MSG(0x150, "An OST (dt %s) is performing "
 500                                    "recovery, of which this client is not a "
 501                                    "part.  Please wait for recovery to "
 502                                    "complete, abort, or time out.\n", dt);
 503                 GOTO(out_md, err);
 504         } else if (err) {
 505                 CERROR("%s: Cannot connect to %s: rc = %d\n",
 506                        sbi->ll_dt_exp->exp_obd->obd_name, dt, err);
 507                 GOTO(out_md, err);
 508         }
 509
 510         sbi->ll_dt_exp->exp_connect_data = *data;
 511
 512         /* Don't change value if it was specified in the config log */
 513         if (sbi->ll_ra_info.ra_max_read_ahead_whole_pages == -1) {
 514                 sbi->ll_ra_info.ra_max_read_ahead_whole_pages =
 515                         max_t(unsigned long, SBI_DEFAULT_READAHEAD_WHOLE_MAX,
 516                               (data->ocd_brw_size >> PAGE_SHIFT));
 517                 if (sbi->ll_ra_info.ra_max_read_ahead_whole_pages >
 518                     sbi->ll_ra_info.ra_max_pages_per_file)
 519                         sbi->ll_ra_info.ra_max_read_ahead_whole_pages =
 520                                 sbi->ll_ra_info.ra_max_pages_per_file;
 521         }
 522
 523         err = obd_fid_init(sbi->ll_dt_exp->exp_obd, sbi->ll_dt_exp,
 524                            LUSTRE_SEQ_METADATA);
 525         if (err) {
 526                 CERROR("%s: Can't init data layer FID infrastructure, "
 527                        "rc = %d\n", sbi->ll_dt_exp->exp_obd->obd_name, err);
 528                 GOTO(out_dt, err);
 529         }
 530
 531         mutex_lock(&sbi->ll_lco.lco_lock);
 532         sbi->ll_lco.lco_flags = data->ocd_connect_flags;
 533         sbi->ll_lco.lco_md_exp = sbi->ll_md_exp;
 534         sbi->ll_lco.lco_dt_exp = sbi->ll_dt_exp;
 535         mutex_unlock(&sbi->ll_lco.lco_lock);
 536
 537         fid_zero(&sbi->ll_root_fid);
 538         err = md_get_root(sbi->ll_md_exp, get_mount_fileset(sb),
 539                            &sbi->ll_root_fid);
 540         if (err) {
 541                 CERROR("cannot mds_connect: rc = %d\n", err);
 542                 GOTO(out_lock_cn_cb, err);
 543         }
 544         if (!fid_is_sane(&sbi->ll_root_fid)) {
 545                 CERROR("%s: Invalid root fid "DFID" during mount\n",
 546                        sbi->ll_md_exp->exp_obd->obd_name,
 547                        PFID(&sbi->ll_root_fid));
 548                 GOTO(out_lock_cn_cb, err = -EINVAL);
 549         }
 550         CDEBUG(D_SUPER, "rootfid "DFID"\n", PFID(&sbi->ll_root_fid));
 551
 552         sb->s_op = &lustre_super_operations;
 553 #ifdef HAVE_XATTR_HANDLER_FLAGS
 554         sb->s_xattr = ll_xattr_handlers;
 555 #endif
 556 #if THREAD_SIZE >= 8192 /*b=17630*/
 557         sb->s_export_op = &lustre_export_operations;
 558 #endif
 559
 560         /* make root inode
 561          * XXX: move this to after cbd setup? */
 562         valid = OBD_MD_FLGETATTR | OBD_MD_FLBLOCKS | OBD_MD_FLMODEASIZE;
 563         if (sbi->ll_flags & LL_SBI_ACL)
 564                 valid |= OBD_MD_FLACL;
 565
 566         OBD_ALLOC_PTR(op_data);
 567         if (op_data == NULL)
 568                 GOTO(out_lock_cn_cb, err = -ENOMEM);
 569
 570         op_data->op_fid1 = sbi->ll_root_fid;
 571         op_data->op_mode = 0;
 572         op_data->op_valid = valid;
 573
 574         err = md_getattr(sbi->ll_md_exp, op_data, &request);
 575
 576         OBD_FREE_PTR(op_data);
 577         if (err) {
 578                 CERROR("%s: md_getattr failed for root: rc = %d\n",
 579                        sbi->ll_md_exp->exp_obd->obd_name, err);
 580                 GOTO(out_lock_cn_cb, err);
 581         }
 582
 583         err = md_get_lustre_md(sbi->ll_md_exp, request, sbi->ll_dt_exp,
 584                                sbi->ll_md_exp, &lmd);
 585         if (err) {
 586                 CERROR("failed to understand root inode md: rc = %d\n", err);
 587                 ptlrpc_req_finished(request);
 588                 GOTO(out_lock_cn_cb, err);
 589         }
 590
 591         LASSERT(fid_is_sane(&sbi->ll_root_fid));
 592         root = ll_iget(sb, cl_fid_build_ino(&sbi->ll_root_fid,
 593                                             sbi->ll_flags & LL_SBI_32BIT_API),
 594                        &lmd);
 595         md_free_lustre_md(sbi->ll_md_exp, &lmd);
 596         ptlrpc_req_finished(request);
 597
 598         if (IS_ERR(root)) {
 599 #ifdef CONFIG_FS_POSIX_ACL
 600                 if (lmd.posix_acl) {
 601                         posix_acl_release(lmd.posix_acl);
 602                         lmd.posix_acl = NULL;
 603                 }
 604 #endif
 605                 err = IS_ERR(root) ? PTR_ERR(root) : -EBADF;
 606                 root = NULL;
 607                 CERROR("lustre_lite: bad iget4 for root\n");
 608                 GOTO(out_root, err);
 609         }
 610
 611         checksum = sbi->ll_flags & LL_SBI_CHECKSUM;
 612         if (sbi->ll_checksum_set) {
 613                 err = obd_set_info_async(NULL, sbi->ll_dt_exp,
 614                                          sizeof(KEY_CHECKSUM), KEY_CHECKSUM,
 615                                          sizeof(checksum), &checksum, NULL);
 616                 if (err) {
 617                         CERROR("%s: Set checksum failed: rc = %d\n",
 618                                sbi->ll_dt_exp->exp_obd->obd_name, err);
 619                         GOTO(out_root, err);
 620                 }
 621         }
 622         cl_sb_init(sb);
 623
 624         sb->s_root = d_make_root(root);
 625         if (sb->s_root == NULL) {
 626                 err = -ENOMEM;
 627                 CERROR("%s: can't make root dentry: rc = %d\n",
 628                        sbi->ll_fsname, err);
 629                 GOTO(out_root, err);
 630         }
 631 #ifdef HAVE_DCACHE_LOCK
 632         sb->s_root->d_op = &ll_d_ops;
 633 #endif
 634
 635         sbi->ll_sdev_orig = sb->s_dev;
 636
 637         /* We set sb->s_dev equal on all lustre clients in order to support
 638          * NFS export clustering.  NFSD requires that the FSID be the same
 639          * on all clients. */
 640         /* s_dev is also used in lt_compare() to compare two fs, but that is
 641          * only a node-local comparison. */
 642         uuid = obd_get_uuid(sbi->ll_md_exp);
 643         if (uuid != NULL)
 644                 sb->s_dev = get_uuid2int(uuid->uuid, strlen(uuid->uuid));
 645
 646         if (data != NULL)
 647                 OBD_FREE_PTR(data);
 648         if (osfs != NULL)
 649                 OBD_FREE_PTR(osfs);
 650
 651         if (sbi->ll_dt_obd) {
 652                 err = sysfs_create_link(&sbi->ll_kset.kobj,
 653                                         &sbi->ll_dt_obd->obd_kset.kobj,
 654                                         sbi->ll_dt_obd->obd_type->typ_name);
 655                 if (err < 0) {
 656                         CERROR("%s: could not register %s in llite: rc = %d\n",
 657                                dt, sbi->ll_fsname, err);
 658                         err = 0;
 659                 }
 660         }
 661
 662         if (sbi->ll_md_obd) {
 663                 err = sysfs_create_link(&sbi->ll_kset.kobj,
 664                                         &sbi->ll_md_obd->obd_kset.kobj,
 665                                         sbi->ll_md_obd->obd_type->typ_name);
 666                 if (err < 0) {
 667                         CERROR("%s: could not register %s in llite: rc = %d\n",
 668                                md, sbi->ll_fsname, err);
 669                         err = 0;
 670                 }
 671         }
 672
 673         RETURN(err);
 674 out_root:
 675         if (root)
 676                 iput(root);
 677 out_lock_cn_cb:
 678         obd_fid_fini(sbi->ll_dt_exp->exp_obd);
 679 out_dt:
 680         obd_disconnect(sbi->ll_dt_exp);
 681         sbi->ll_dt_exp = NULL;
 682         sbi->ll_dt_obd = NULL;
 683 out_md_fid:
 684         obd_fid_fini(sbi->ll_md_exp->exp_obd);
 685 out_md:
 686         obd_disconnect(sbi->ll_md_exp);
 687         sbi->ll_md_exp = NULL;
 688         sbi->ll_md_obd = NULL;
 689 out:
 690         if (data != NULL)
 691                 OBD_FREE_PTR(data);
 692         if (osfs != NULL)
 693                 OBD_FREE_PTR(osfs);
 694         return err;
 695 }
 696
 697 int ll_get_max_mdsize(struct ll_sb_info *sbi, int *lmmsize)
 698 {
 699         int size, rc;
 700
 701         size = sizeof(*lmmsize);
 702         rc = obd_get_info(NULL, sbi->ll_dt_exp, sizeof(KEY_MAX_EASIZE),
 703                           KEY_MAX_EASIZE, &size, lmmsize);
 704         if (rc != 0) {
 705                 CERROR("%s: cannot get max LOV EA size: rc = %d\n",
 706                        sbi->ll_dt_exp->exp_obd->obd_name, rc);
 707                 RETURN(rc);
 708         }
 709
 710         CDEBUG(D_INFO, "max LOV ea size: %d\n", *lmmsize);
 711
 712         size = sizeof(int);
 713         rc = obd_get_info(NULL, sbi->ll_md_exp, sizeof(KEY_MAX_EASIZE),
 714                           KEY_MAX_EASIZE, &size, lmmsize);
 715         if (rc)
 716                 CERROR("Get max mdsize error rc %d\n", rc);
 717
 718         CDEBUG(D_INFO, "max LMV ea size: %d\n", *lmmsize);
 719
 720         RETURN(rc);
 721 }
 722
 723 /**
 724  * Get the value of the default_easize parameter.
 725  *
 726  * \see client_obd::cl_default_mds_easize
 727  *
 728  * \param[in] sbi       superblock info for this filesystem
 729  * \param[out] lmmsize  pointer to storage location for value
 730  *
 731  * \retval 0            on success
 732  * \retval negative     negated errno on failure
 733  */
 734 int ll_get_default_mdsize(struct ll_sb_info *sbi, int *lmmsize)
 735 {
 736         int size, rc;
 737
 738         size = sizeof(int);
 739         rc = obd_get_info(NULL, sbi->ll_md_exp, sizeof(KEY_DEFAULT_EASIZE),
 740                          KEY_DEFAULT_EASIZE, &size, lmmsize);
 741         if (rc)
 742                 CERROR("Get default mdsize error rc %d\n", rc);
 743
 744         RETURN(rc);
 745 }
 746
 747 /**
 748  * Set the default_easize parameter to the given value.
 749  *
 750  * \see client_obd::cl_default_mds_easize
 751  *
 752  * \param[in] sbi       superblock info for this filesystem
 753  * \param[in] lmmsize   the size to set
 754  *
 755  * \retval 0            on success
 756  * \retval negative     negated errno on failure
 757  */
 758 int ll_set_default_mdsize(struct ll_sb_info *sbi, int lmmsize)
 759 {
 760         int rc;
 761
 762         if (lmmsize < sizeof(struct lov_mds_md) ||
 763             lmmsize > OBD_MAX_DEFAULT_EA_SIZE)
 764                 return -EINVAL;
 765
 766         rc = obd_set_info_async(NULL, sbi->ll_md_exp,
 767                                 sizeof(KEY_DEFAULT_EASIZE), KEY_DEFAULT_EASIZE,
 768                                 sizeof(int), &lmmsize, NULL);
 769
 770         RETURN(rc);
 771 }
 772
 773 static void client_common_put_super(struct super_block *sb)
 774 {
 775         struct ll_sb_info *sbi = ll_s2sbi(sb);
 776         ENTRY;
 777
 778         cl_sb_fini(sb);
 779
 780         obd_fid_fini(sbi->ll_dt_exp->exp_obd);
 781         obd_disconnect(sbi->ll_dt_exp);
 782         sbi->ll_dt_exp = NULL;
 783
 784         ll_debugfs_unregister_super(sb);
 785
 786         obd_fid_fini(sbi->ll_md_exp->exp_obd);
 787         obd_disconnect(sbi->ll_md_exp);
 788         sbi->ll_md_exp = NULL;
 789
 790         EXIT;
 791 }
 792
 793 void ll_kill_super(struct super_block *sb)
 794 {
 795         struct ll_sb_info *sbi;
 796         ENTRY;
 797
 798         /* not init sb ?*/
 799         if (!(sb->s_flags & MS_ACTIVE))
 800                 return;
 801
 802         sbi = ll_s2sbi(sb);
 803         /* we need restore s_dev from changed for clustred NFS before put_super
 804          * because new kernels have cached s_dev and change sb->s_dev in
 805          * put_super not affected real removing devices */
 806         if (sbi) {
 807                 sb->s_dev = sbi->ll_sdev_orig;
 808                 sbi->ll_umounting = 1;
 809
 810                 /* wait running statahead threads to quit */
 811                 while (atomic_read(&sbi->ll_sa_running) > 0) {
 812                         set_current_state(TASK_UNINTERRUPTIBLE);
 813                         schedule_timeout(cfs_time_seconds(1 >> 3));
 814                 }
 815         }
 816
 817         EXIT;
 818 }
 819
 820 static inline int ll_set_opt(const char *opt, char *data, int fl)
 821 {
 822         if (strncmp(opt, data, strlen(opt)) != 0)
 823                 return 0;
 824         else
 825                 return fl;
 826 }
 827
 828 /* non-client-specific mount options are parsed in lmd_parse */
 829 static int ll_options(char *options, struct ll_sb_info *sbi)
 830 {
 831         int tmp;
 832         char *s1 = options, *s2;
 833         int *flags = &sbi->ll_flags;
 834         ENTRY;
 835
 836         if (!options)
 837                 RETURN(0);
 838
 839         CDEBUG(D_CONFIG, "Parsing opts %s\n", options);
 840
 841         while (*s1) {
 842                 CDEBUG(D_SUPER, "next opt=%s\n", s1);
 843                 tmp = ll_set_opt("nolock", s1, LL_SBI_NOLCK);
 844                 if (tmp) {
 845                         *flags |= tmp;
 846                         goto next;
 847                 }
 848                 tmp = ll_set_opt("flock", s1, LL_SBI_FLOCK);
 849                 if (tmp) {
 850                         *flags |= tmp;
 851                         goto next;
 852                 }
 853                 tmp = ll_set_opt("localflock", s1, LL_SBI_LOCALFLOCK);
 854                 if (tmp) {
 855                         *flags |= tmp;
 856                         goto next;
 857                 }
 858                 tmp = ll_set_opt("noflock", s1, LL_SBI_FLOCK|LL_SBI_LOCALFLOCK);
 859                 if (tmp) {
 860                         *flags &= ~tmp;
 861                         goto next;
 862                 }
 863                 tmp = ll_set_opt("user_xattr", s1, LL_SBI_USER_XATTR);
 864                 if (tmp) {
 865                         *flags |= tmp;
 866                         goto next;
 867                 }
 868                 tmp = ll_set_opt("nouser_xattr", s1, LL_SBI_USER_XATTR);
 869                 if (tmp) {
 870                         *flags &= ~tmp;
 871                         goto next;
 872                 }
 873                 tmp = ll_set_opt("context", s1, 1);
 874                 if (tmp)
 875                         goto next;
 876                 tmp = ll_set_opt("fscontext", s1, 1);
 877                 if (tmp)
 878                         goto next;
 879                 tmp = ll_set_opt("defcontext", s1, 1);
 880                 if (tmp)
 881                         goto next;
 882                 tmp = ll_set_opt("rootcontext", s1, 1);
 883                 if (tmp)
 884                         goto next;
 885                 tmp = ll_set_opt("user_fid2path", s1, LL_SBI_USER_FID2PATH);
 886                 if (tmp) {
 887                         *flags |= tmp;
 888                         goto next;
 889                 }
 890                 tmp = ll_set_opt("nouser_fid2path", s1, LL_SBI_USER_FID2PATH);
 891                 if (tmp) {
 892                         *flags &= ~tmp;
 893                         goto next;
 894                 }
 895
 896                 tmp = ll_set_opt("checksum", s1, LL_SBI_CHECKSUM);
 897                 if (tmp) {
 898                         *flags |= tmp;
 899                         sbi->ll_checksum_set = 1;
 900                         goto next;
 901                 }
 902                 tmp = ll_set_opt("nochecksum", s1, LL_SBI_CHECKSUM);
 903                 if (tmp) {
 904                         *flags &= ~tmp;
 905                         sbi->ll_checksum_set = 1;
 906                         goto next;
 907                 }
 908                 tmp = ll_set_opt("lruresize", s1, LL_SBI_LRU_RESIZE);
 909                 if (tmp) {
 910                         *flags |= tmp;
 911                         goto next;
 912                 }
 913                 tmp = ll_set_opt("nolruresize", s1, LL_SBI_LRU_RESIZE);
 914                 if (tmp) {
 915                         *flags &= ~tmp;
 916                         goto next;
 917                 }
 918                 tmp = ll_set_opt("lazystatfs", s1, LL_SBI_LAZYSTATFS);
 919                 if (tmp) {
 920                         *flags |= tmp;
 921                         goto next;
 922                 }
 923                 tmp = ll_set_opt("nolazystatfs", s1, LL_SBI_LAZYSTATFS);
 924                 if (tmp) {
 925                         *flags &= ~tmp;
 926                         goto next;
 927                 }
 928                 tmp = ll_set_opt("32bitapi", s1, LL_SBI_32BIT_API);
 929                 if (tmp) {
 930                         *flags |= tmp;
 931                         goto next;
 932                 }
 933                 tmp = ll_set_opt("verbose", s1, LL_SBI_VERBOSE);
 934                 if (tmp) {
 935                         *flags |= tmp;
 936                         goto next;
 937                 }
 938                 tmp = ll_set_opt("noverbose", s1, LL_SBI_VERBOSE);
 939                 if (tmp) {
 940                         *flags &= ~tmp;
 941                         goto next;
 942                 }
 943                 tmp = ll_set_opt("always_ping", s1, LL_SBI_ALWAYS_PING);
 944                 if (tmp) {
 945                         *flags |= tmp;
 946                         goto next;
 947                 }
 948                 LCONSOLE_ERROR_MSG(0x152, "Unknown option '%s', won't mount.\n",
 949                                    s1);
 950                 RETURN(-EINVAL);
 951
 952 next:
 953                 /* Find next opt */
 954                 s2 = strchr(s1, ',');
 955                 if (s2 == NULL)
 956                         break;
 957                 s1 = s2 + 1;
 958         }
 959         RETURN(0);
 960 }
 961
 962 void ll_lli_init(struct ll_inode_info *lli)
 963 {
 964         lli->lli_inode_magic = LLI_INODE_MAGIC;
 965         lli->lli_flags = 0;
 966         spin_lock_init(&lli->lli_lock);
 967         lli->lli_posix_acl = NULL;
 968         /* Do not set lli_fid, it has been initialized already. */
 969         fid_zero(&lli->lli_pfid);
 970         lli->lli_mds_read_och = NULL;
 971         lli->lli_mds_write_och = NULL;
 972         lli->lli_mds_exec_och = NULL;
 973         lli->lli_open_fd_read_count = 0;
 974         lli->lli_open_fd_write_count = 0;
 975         lli->lli_open_fd_exec_count = 0;
 976         mutex_init(&lli->lli_och_mutex);
 977         spin_lock_init(&lli->lli_agl_lock);
 978         spin_lock_init(&lli->lli_layout_lock);
 979         ll_layout_version_set(lli, CL_LAYOUT_GEN_NONE);
 980         lli->lli_clob = NULL;
 981
 982         init_rwsem(&lli->lli_xattrs_list_rwsem);
 983         mutex_init(&lli->lli_xattrs_enq_lock);
 984
 985         LASSERT(lli->lli_vfs_inode.i_mode != 0);
 986         if (S_ISDIR(lli->lli_vfs_inode.i_mode)) {
 987                 mutex_init(&lli->lli_readdir_mutex);
 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         }
1011         mutex_init(&lli->lli_layout_mutex);
1012         memset(lli->lli_jobid, 0, sizeof(lli->lli_jobid));
1013 }
1014
1015 #define MAX_STRING_SIZE 128
1016
1017 #ifndef HAVE_SUPER_SETUP_BDI_NAME
1018
1019 #define LSI_BDI_INITIALIZED     0x00400000
1020
1021 #ifndef HAVE_BDI_CAP_MAP_COPY
1022 # define BDI_CAP_MAP_COPY       0
1023 #endif
1024
1025 static int super_setup_bdi_name(struct super_block *sb, char *fmt, ...)
1026 {
1027         struct  lustre_sb_info *lsi = s2lsi(sb);
1028         char buf[MAX_STRING_SIZE];
1029         va_list args;
1030         int err;
1031
1032         err = bdi_init(&lsi->lsi_bdi);
1033         if (err)
1034                 return err;
1035
1036         lsi->lsi_flags |= LSI_BDI_INITIALIZED;
1037         lsi->lsi_bdi.capabilities = BDI_CAP_MAP_COPY;
1038         lsi->lsi_bdi.name = "lustre";
1039         va_start(args, fmt);
1040         vsnprintf(buf, MAX_STRING_SIZE, fmt, args);
1041         va_end(args);
1042         err = bdi_register(&lsi->lsi_bdi, NULL, "%s", buf);
1043         va_end(args);
1044         if (!err)
1045                 sb->s_bdi = &lsi->lsi_bdi;
1046
1047         return err;
1048 }
1049 #endif /* !HAVE_SUPER_SETUP_BDI_NAME */
1050
1051 int ll_fill_super(struct super_block *sb, struct vfsmount *mnt)
1052 {
1053         struct  lustre_profile *lprof = NULL;
1054         struct  lustre_sb_info *lsi = s2lsi(sb);
1055         struct  ll_sb_info *sbi = NULL;
1056         char    *dt = NULL, *md = NULL;
1057         char    *profilenm = get_profile_name(sb);
1058         struct config_llog_instance *cfg;
1059         /* %p for void* in printf needs 16+2 characters: 0xffffffffffffffff */
1060         const int instlen = 16 + 2;
1061         unsigned long cfg_instance = ll_get_cfg_instance(sb);
1062         char name[MAX_STRING_SIZE];
1063         int md_len = 0;
1064         int dt_len = 0;
1065         uuid_t uuid;
1066         char *ptr;
1067         int len;
1068         int err;
1069
1070         ENTRY;
1071         /* for ASLR, to map between cfg_instance and hashed ptr */
1072         CDEBUG(D_VFSTRACE, "VFS Op: cfg_instance %s-%016lx (sb %p)\n",
1073                profilenm, cfg_instance, sb);
1074
1075         try_module_get(THIS_MODULE);
1076
1077         OBD_ALLOC_PTR(cfg);
1078         if (cfg == NULL)
1079                 GOTO(out_free_cfg, err = -ENOMEM);
1080
1081         /* client additional sb info */
1082         lsi->lsi_llsbi = sbi = ll_init_sbi();
1083         if (IS_ERR(sbi))
1084                 GOTO(out_free_cfg, err = PTR_ERR(sbi));
1085
1086         err = ll_options(lsi->lsi_lmd->lmd_opts, sbi);
1087         if (err)
1088                 GOTO(out_free_cfg, err);
1089
1090 #ifndef HAVE_DCACHE_LOCK
1091         /* kernel >= 2.6.38 store dentry operations in sb->s_d_op. */
1092         sb->s_d_op = &ll_d_ops;
1093 #endif
1094         /* UUID handling */
1095         generate_random_uuid(uuid.b);
1096         snprintf(sbi->ll_sb_uuid.uuid, UUID_SIZE, "%pU", uuid.b);
1097
1098         CDEBUG(D_CONFIG, "llite sb uuid: %s\n", sbi->ll_sb_uuid.uuid);
1099
1100         /* Get fsname */
1101         len = strlen(profilenm);
1102         ptr = strrchr(profilenm, '-');
1103         if (ptr && (strcmp(ptr, "-client") == 0))
1104                 len -= 7;
1105
1106         if (len > LUSTRE_MAXFSNAME) {
1107                 if (unlikely(len >= MAX_STRING_SIZE))
1108                         len = MAX_STRING_SIZE - 1;
1109                 strncpy(name, profilenm, len);
1110                 name[len] = '\0';
1111                 err = -ENAMETOOLONG;
1112                 CERROR("%s: fsname longer than %u characters: rc = %d\n",
1113                        name, LUSTRE_MAXFSNAME, err);
1114                 GOTO(out_free_cfg, err);
1115         }
1116         strncpy(sbi->ll_fsname, profilenm, len);
1117         sbi->ll_fsname[len] = '\0';
1118
1119         /* Mount info */
1120         snprintf(name, sizeof(name), "%.*s-%016lx", len,
1121                  profilenm, cfg_instance);
1122
1123         err = super_setup_bdi_name(sb, "%s", name);
1124         if (err)
1125                 GOTO(out_free_cfg, err);
1126
1127         /* Call ll_debugfs_register_super() before lustre_process_log()
1128          * so that "llite.*.*" params can be processed correctly.
1129          */
1130         err = ll_debugfs_register_super(sb, name);
1131         if (err < 0) {
1132                 CERROR("%s: could not register mountpoint in llite: rc = %d\n",
1133                        sbi->ll_fsname, err);
1134                 err = 0;
1135         }
1136
1137         /* The cfg_instance is a value unique to this super, in case some
1138          * joker tries to mount the same fs at two mount points.
1139          */
1140         cfg->cfg_instance = cfg_instance;
1141         cfg->cfg_uuid = lsi->lsi_llsbi->ll_sb_uuid;
1142         cfg->cfg_callback = class_config_llog_handler;
1143         cfg->cfg_sub_clds = CONFIG_SUB_CLIENT;
1144         /* set up client obds */
1145         err = lustre_process_log(sb, profilenm, cfg);
1146         if (err < 0)
1147                 GOTO(out_debugfs, err);
1148
1149         /* Profile set with LCFG_MOUNTOPT so we can find our mdc and osc obds */
1150         lprof = class_get_profile(profilenm);
1151         if (lprof == NULL) {
1152                 LCONSOLE_ERROR_MSG(0x156, "The client profile '%s' could not be"
1153                                    " read from the MGS.  Does that filesystem "
1154                                    "exist?\n", profilenm);
1155                 GOTO(out_debugfs, err = -EINVAL);
1156         }
1157         CDEBUG(D_CONFIG, "Found profile %s: mdc=%s osc=%s\n", profilenm,
1158                lprof->lp_md, lprof->lp_dt);
1159
1160         dt_len = strlen(lprof->lp_dt) + instlen + 2;
1161         OBD_ALLOC(dt, dt_len);
1162         if (!dt)
1163                 GOTO(out_profile, err = -ENOMEM);
1164         snprintf(dt, dt_len - 1, "%s-%016lx", lprof->lp_dt, cfg_instance);
1165
1166         md_len = strlen(lprof->lp_md) + instlen + 2;
1167         OBD_ALLOC(md, md_len);
1168         if (!md)
1169                 GOTO(out_free_dt, err = -ENOMEM);
1170         snprintf(md, md_len - 1, "%s-%016lx", lprof->lp_md, cfg_instance);
1171
1172         /* connections, registrations, sb setup */
1173         err = client_common_fill_super(sb, md, dt, mnt);
1174         if (err < 0)
1175                 GOTO(out_free_md, err);
1176
1177         sbi->ll_client_common_fill_super_succeeded = 1;
1178
1179 out_free_md:
1180         if (md)
1181                 OBD_FREE(md, md_len);
1182 out_free_dt:
1183         if (dt)
1184                 OBD_FREE(dt, dt_len);
1185 out_profile:
1186         if (lprof)
1187                 class_put_profile(lprof);
1188 out_debugfs:
1189         if (err < 0)
1190                 ll_debugfs_unregister_super(sb);
1191 out_free_cfg:
1192         if (cfg)
1193                 OBD_FREE_PTR(cfg);
1194
1195         if (err)
1196                 ll_put_super(sb);
1197         else if (sbi->ll_flags & LL_SBI_VERBOSE)
1198                 LCONSOLE_WARN("Mounted %s\n", profilenm);
1199         RETURN(err);
1200 } /* ll_fill_super */
1201
1202 void ll_put_super(struct super_block *sb)
1203 {
1204         struct config_llog_instance cfg, params_cfg;
1205         struct obd_device *obd;
1206         struct lustre_sb_info *lsi = s2lsi(sb);
1207         struct ll_sb_info *sbi = ll_s2sbi(sb);
1208         char *profilenm = get_profile_name(sb);
1209         unsigned long cfg_instance = ll_get_cfg_instance(sb);
1210         long ccc_count;
1211         int next, force = 1, rc = 0;
1212         ENTRY;
1213
1214         if (IS_ERR(sbi))
1215                 GOTO(out_no_sbi, 0);
1216
1217         /* Should replace instance_id with something better for ASLR */
1218         CDEBUG(D_VFSTRACE, "VFS Op: cfg_instance %s-%016lx (sb %p)\n",
1219                profilenm, cfg_instance, sb);
1220
1221         cfg.cfg_instance = cfg_instance;
1222         lustre_end_log(sb, profilenm, &cfg);
1223
1224         params_cfg.cfg_instance = cfg_instance;
1225         lustre_end_log(sb, PARAMS_FILENAME, &params_cfg);
1226
1227         if (sbi->ll_md_exp) {
1228                 obd = class_exp2obd(sbi->ll_md_exp);
1229                 if (obd)
1230                         force = obd->obd_force;
1231         }
1232
1233         /* Wait for unstable pages to be committed to stable storage */
1234         if (force == 0) {
1235                 struct l_wait_info lwi = LWI_INTR(LWI_ON_SIGNAL_NOOP, NULL);
1236                 rc = l_wait_event(sbi->ll_cache->ccc_unstable_waitq,
1237                         atomic_long_read(&sbi->ll_cache->ccc_unstable_nr) == 0,
1238                         &lwi);
1239         }
1240
1241         ccc_count = atomic_long_read(&sbi->ll_cache->ccc_unstable_nr);
1242         if (force == 0 && rc != -EINTR)
1243                 LASSERTF(ccc_count == 0, "count: %li\n", ccc_count);
1244
1245         /* We need to set force before the lov_disconnect in
1246            lustre_common_put_super, since l_d cleans up osc's as well. */
1247         if (force) {
1248                 next = 0;
1249                 while ((obd = class_devices_in_group(&sbi->ll_sb_uuid,
1250                                                      &next)) != NULL) {
1251                         obd->obd_force = force;
1252                 }
1253         }
1254
1255         if (sbi->ll_client_common_fill_super_succeeded) {
1256                 /* Only if client_common_fill_super succeeded */
1257                 client_common_put_super(sb);
1258         }
1259
1260         next = 0;
1261         while ((obd = class_devices_in_group(&sbi->ll_sb_uuid, &next)) !=NULL) {
1262                 class_manual_cleanup(obd);
1263         }
1264
1265         if (sbi->ll_flags & LL_SBI_VERBOSE)
1266                 LCONSOLE_WARN("Unmounted %s\n", profilenm ? profilenm : "");
1267
1268         if (profilenm)
1269                 class_del_profile(profilenm);
1270
1271 #ifndef HAVE_SUPER_SETUP_BDI_NAME
1272         if (lsi->lsi_flags & LSI_BDI_INITIALIZED) {
1273                 bdi_destroy(&lsi->lsi_bdi);
1274                 lsi->lsi_flags &= ~LSI_BDI_INITIALIZED;
1275         }
1276 #endif
1277
1278         ll_free_sbi(sb);
1279         lsi->lsi_llsbi = NULL;
1280 out_no_sbi:
1281         lustre_common_put_super(sb);
1282
1283         cl_env_cache_purge(~0);
1284
1285         module_put(THIS_MODULE);
1286
1287         EXIT;
1288 } /* client_put_super */
1289
1290 struct inode *ll_inode_from_resource_lock(struct ldlm_lock *lock)
1291 {
1292         struct inode *inode = NULL;
1293
1294         /* NOTE: we depend on atomic igrab() -bzzz */
1295         lock_res_and_lock(lock);
1296         if (lock->l_resource->lr_lvb_inode) {
1297                 struct ll_inode_info * lli;
1298                 lli = ll_i2info(lock->l_resource->lr_lvb_inode);
1299                 if (lli->lli_inode_magic == LLI_INODE_MAGIC) {
1300                         inode = igrab(lock->l_resource->lr_lvb_inode);
1301                 } else {
1302                         inode = lock->l_resource->lr_lvb_inode;
1303                         LDLM_DEBUG_LIMIT(inode->i_state & I_FREEING ?  D_INFO :
1304                                          D_WARNING, lock, "lr_lvb_inode %p is "
1305                                          "bogus: magic %08x",
1306                                          lock->l_resource->lr_lvb_inode,
1307                                          lli->lli_inode_magic);
1308                         inode = NULL;
1309                 }
1310         }
1311         unlock_res_and_lock(lock);
1312         return inode;
1313 }
1314
1315 void ll_dir_clear_lsm_md(struct inode *inode)
1316 {
1317         struct ll_inode_info *lli = ll_i2info(inode);
1318
1319         LASSERT(S_ISDIR(inode->i_mode));
1320
1321         if (lli->lli_lsm_md) {
1322                 lmv_free_memmd(lli->lli_lsm_md);
1323                 lli->lli_lsm_md = NULL;
1324         }
1325
1326         if (lli->lli_default_lsm_md) {
1327                 lmv_free_memmd(lli->lli_default_lsm_md);
1328                 lli->lli_default_lsm_md = NULL;
1329         }
1330 }
1331
1332 static struct inode *ll_iget_anon_dir(struct super_block *sb,
1333                                       const struct lu_fid *fid,
1334                                       struct lustre_md *md)
1335 {
1336         struct ll_sb_info       *sbi = ll_s2sbi(sb);
1337         struct mdt_body         *body = md->body;
1338         struct inode            *inode;
1339         ino_t                   ino;
1340         ENTRY;
1341
1342         ino = cl_fid_build_ino(fid, sbi->ll_flags & LL_SBI_32BIT_API);
1343         inode = iget_locked(sb, ino);
1344         if (inode == NULL) {
1345                 CERROR("%s: failed get simple inode "DFID": rc = -ENOENT\n",
1346                        sbi->ll_fsname, PFID(fid));
1347                 RETURN(ERR_PTR(-ENOENT));
1348         }
1349
1350         if (inode->i_state & I_NEW) {
1351                 struct ll_inode_info *lli = ll_i2info(inode);
1352                 struct lmv_stripe_md *lsm = md->lmv;
1353
1354                 inode->i_mode = (inode->i_mode & ~S_IFMT) |
1355                                 (body->mbo_mode & S_IFMT);
1356                 LASSERTF(S_ISDIR(inode->i_mode), "Not slave inode "DFID"\n",
1357                          PFID(fid));
1358
1359                 inode->i_mtime.tv_sec = 0;
1360                 inode->i_atime.tv_sec = 0;
1361                 inode->i_ctime.tv_sec = 0;
1362                 inode->i_rdev = 0;
1363
1364 #ifdef HAVE_BACKING_DEV_INFO
1365                 /* initializing backing dev info. */
1366                 inode->i_mapping->backing_dev_info =
1367                                                 &s2lsi(inode->i_sb)->lsi_bdi;
1368 #endif
1369                 inode->i_op = &ll_dir_inode_operations;
1370                 inode->i_fop = &ll_dir_operations;
1371                 lli->lli_fid = *fid;
1372                 ll_lli_init(lli);
1373
1374                 LASSERT(lsm != NULL);
1375                 /* master object FID */
1376                 lli->lli_pfid = body->mbo_fid1;
1377                 CDEBUG(D_INODE, "lli %p slave "DFID" master "DFID"\n",
1378                        lli, PFID(fid), PFID(&lli->lli_pfid));
1379                 unlock_new_inode(inode);
1380         }
1381
1382         RETURN(inode);
1383 }
1384
1385 static int ll_init_lsm_md(struct inode *inode, struct lustre_md *md)
1386 {
1387         struct lu_fid *fid;
1388         struct lmv_stripe_md *lsm = md->lmv;
1389         struct ll_inode_info *lli = ll_i2info(inode);
1390         int i;
1391
1392         LASSERT(lsm != NULL);
1393
1394         CDEBUG(D_INODE, "%s: "DFID" set dir layout:\n",
1395                ll_i2sbi(inode)->ll_fsname, PFID(&lli->lli_fid));
1396         lsm_md_dump(D_INODE, lsm);
1397
1398         if (!lmv_dir_striped(lsm))
1399                 goto out;
1400
1401         /* XXX sigh, this lsm_root initialization should be in
1402          * LMV layer, but it needs ll_iget right now, so we
1403          * put this here right now. */
1404         for (i = 0; i < lsm->lsm_md_stripe_count; i++) {
1405                 fid = &lsm->lsm_md_oinfo[i].lmo_fid;
1406                 LASSERT(lsm->lsm_md_oinfo[i].lmo_root == NULL);
1407
1408                 if (!fid_is_sane(fid))
1409                         continue;
1410
1411                 /* Unfortunately ll_iget will call ll_update_inode,
1412                  * where the initialization of slave inode is slightly
1413                  * different, so it reset lsm_md to NULL to avoid
1414                  * initializing lsm for slave inode. */
1415                 lsm->lsm_md_oinfo[i].lmo_root =
1416                                 ll_iget_anon_dir(inode->i_sb, fid, md);
1417                 if (IS_ERR(lsm->lsm_md_oinfo[i].lmo_root)) {
1418                         int rc = PTR_ERR(lsm->lsm_md_oinfo[i].lmo_root);
1419
1420                         lsm->lsm_md_oinfo[i].lmo_root = NULL;
1421                         while (i-- > 0) {
1422                                 iput(lsm->lsm_md_oinfo[i].lmo_root);
1423                                 lsm->lsm_md_oinfo[i].lmo_root = NULL;
1424                         }
1425                         return rc;
1426                 }
1427         }
1428 out:
1429         lli->lli_lsm_md = lsm;
1430
1431         return 0;
1432 }
1433
1434 static void ll_update_default_lsm_md(struct inode *inode, struct lustre_md *md)
1435 {
1436         struct ll_inode_info *lli = ll_i2info(inode);
1437
1438         if (!md->default_lmv) {
1439                 /* clear default lsm */
1440                 if (lli->lli_default_lsm_md) {
1441                         down_write(&lli->lli_lsm_sem);
1442                         if (lli->lli_default_lsm_md) {
1443                                 lmv_free_memmd(lli->lli_default_lsm_md);
1444                                 lli->lli_default_lsm_md = NULL;
1445                         }
1446                         up_write(&lli->lli_lsm_sem);
1447                 }
1448         } else if (lli->lli_default_lsm_md) {
1449                 /* update default lsm if it changes */
1450                 down_read(&lli->lli_lsm_sem);
1451                 if (lli->lli_default_lsm_md &&
1452                     !lsm_md_eq(lli->lli_default_lsm_md, md->default_lmv)) {
1453                         up_read(&lli->lli_lsm_sem);
1454                         down_write(&lli->lli_lsm_sem);
1455                         if (lli->lli_default_lsm_md)
1456                                 lmv_free_memmd(lli->lli_default_lsm_md);
1457                         lli->lli_default_lsm_md = md->default_lmv;
1458                         lsm_md_dump(D_INODE, md->default_lmv);
1459                         md->default_lmv = NULL;
1460                         up_write(&lli->lli_lsm_sem);
1461                 } else {
1462                         up_read(&lli->lli_lsm_sem);
1463                 }
1464         } else {
1465                 /* init default lsm */
1466                 down_write(&lli->lli_lsm_sem);
1467                 lli->lli_default_lsm_md = md->default_lmv;
1468                 lsm_md_dump(D_INODE, md->default_lmv);
1469                 md->default_lmv = NULL;
1470                 up_write(&lli->lli_lsm_sem);
1471         }
1472 }
1473
1474 static int ll_update_lsm_md(struct inode *inode, struct lustre_md *md)
1475 {
1476         struct ll_inode_info *lli = ll_i2info(inode);
1477         struct lmv_stripe_md *lsm = md->lmv;
1478         int rc = 0;
1479
1480         ENTRY;
1481
1482         LASSERT(S_ISDIR(inode->i_mode));
1483         CDEBUG(D_INODE, "update lsm %p of "DFID"\n", lli->lli_lsm_md,
1484                PFID(ll_inode2fid(inode)));
1485
1486         /* update default LMV */
1487         if (md->default_lmv)
1488                 ll_update_default_lsm_md(inode, md);
1489
1490         /*
1491          * no striped information from request, lustre_md from req does not
1492          * include stripeEA, see ll_md_setattr()
1493          */
1494         if (!lsm)
1495                 RETURN(0);
1496
1497         /*
1498          * normally dir layout doesn't change, only take read lock to check
1499          * that to avoid blocking other MD operations.
1500          */
1501         if (lli->lli_lsm_md)
1502                 down_read(&lli->lli_lsm_sem);
1503         else
1504                 down_write(&lli->lli_lsm_sem);
1505
1506         /*
1507          * if dir layout mismatch, check whether version is increased, which
1508          * means layout is changed, this happens in dir migration and lfsck.
1509          *
1510          * foreign LMV should not change.
1511          */
1512         if (lli->lli_lsm_md && !lsm_md_eq(lli->lli_lsm_md, lsm)) {
1513                 if (lmv_dir_striped(lli->lli_lsm_md) &&
1514                     lsm->lsm_md_layout_version <=
1515                     lli->lli_lsm_md->lsm_md_layout_version) {
1516                         CERROR("%s: "DFID" dir layout mismatch:\n",
1517                                ll_i2sbi(inode)->ll_fsname,
1518                                PFID(&lli->lli_fid));
1519                         lsm_md_dump(D_ERROR, lli->lli_lsm_md);
1520                         lsm_md_dump(D_ERROR, lsm);
1521                         GOTO(unlock, rc = -EINVAL);
1522                 }
1523
1524                 /* layout changed, switch to write lock */
1525                 up_read(&lli->lli_lsm_sem);
1526                 down_write(&lli->lli_lsm_sem);
1527                 ll_dir_clear_lsm_md(inode);
1528         }
1529
1530         /* set directory layout */
1531         if (!lli->lli_lsm_md) {
1532                 struct cl_attr  *attr;
1533
1534                 rc = ll_init_lsm_md(inode, md);
1535                 up_write(&lli->lli_lsm_sem);
1536                 if (rc != 0)
1537                         RETURN(rc);
1538
1539                 /* set md->lmv to NULL, so the following free lustre_md
1540                  * will not free this lsm */
1541                 md->lmv = NULL;
1542
1543                 /*
1544                  * md_merge_attr() may take long, since lsm is already set,
1545                  * switch to read lock.
1546                  */
1547                 down_read(&lli->lli_lsm_sem);
1548
1549                 if (!lmv_dir_striped(lli->lli_lsm_md))
1550                         GOTO(unlock, rc);
1551
1552                 OBD_ALLOC_PTR(attr);
1553                 if (attr == NULL)
1554                         GOTO(unlock, rc = -ENOMEM);
1555
1556                 /* validate the lsm */
1557                 rc = md_merge_attr(ll_i2mdexp(inode), lsm, attr,
1558                                    ll_md_blocking_ast);
1559                 if (rc != 0) {
1560                         OBD_FREE_PTR(attr);
1561                         GOTO(unlock, rc);
1562                 }
1563
1564                 if (md->body->mbo_valid & OBD_MD_FLNLINK)
1565                         md->body->mbo_nlink = attr->cat_nlink;
1566                 if (md->body->mbo_valid & OBD_MD_FLSIZE)
1567                         md->body->mbo_size = attr->cat_size;
1568                 if (md->body->mbo_valid & OBD_MD_FLATIME)
1569                         md->body->mbo_atime = attr->cat_atime;
1570                 if (md->body->mbo_valid & OBD_MD_FLCTIME)
1571                         md->body->mbo_ctime = attr->cat_ctime;
1572                 if (md->body->mbo_valid & OBD_MD_FLMTIME)
1573                         md->body->mbo_mtime = attr->cat_mtime;
1574
1575                 OBD_FREE_PTR(attr);
1576         }
1577 unlock:
1578         up_read(&lli->lli_lsm_sem);
1579
1580         RETURN(rc);
1581 }
1582
1583 void ll_clear_inode(struct inode *inode)
1584 {
1585         struct ll_inode_info *lli = ll_i2info(inode);
1586         struct ll_sb_info *sbi = ll_i2sbi(inode);
1587
1588         ENTRY;
1589
1590         CDEBUG(D_VFSTRACE, "VFS Op:inode="DFID"(%p)\n",
1591                PFID(ll_inode2fid(inode)), inode);
1592
1593         if (S_ISDIR(inode->i_mode)) {
1594                 /* these should have been cleared in ll_file_release */
1595                 LASSERT(lli->lli_opendir_key == NULL);
1596                 LASSERT(lli->lli_sai == NULL);
1597                 LASSERT(lli->lli_opendir_pid == 0);
1598         } else {
1599                 pcc_inode_free(inode);
1600         }
1601
1602         md_null_inode(sbi->ll_md_exp, ll_inode2fid(inode));
1603
1604         LASSERT(!lli->lli_open_fd_write_count);
1605         LASSERT(!lli->lli_open_fd_read_count);
1606         LASSERT(!lli->lli_open_fd_exec_count);
1607
1608         if (lli->lli_mds_write_och)
1609                 ll_md_real_close(inode, FMODE_WRITE);
1610         if (lli->lli_mds_exec_och)
1611                 ll_md_real_close(inode, FMODE_EXEC);
1612         if (lli->lli_mds_read_och)
1613                 ll_md_real_close(inode, FMODE_READ);
1614
1615         if (S_ISLNK(inode->i_mode) && lli->lli_symlink_name) {
1616                 OBD_FREE(lli->lli_symlink_name,
1617                          strlen(lli->lli_symlink_name) + 1);
1618                 lli->lli_symlink_name = NULL;
1619         }
1620
1621         ll_xattr_cache_destroy(inode);
1622
1623 #ifdef CONFIG_FS_POSIX_ACL
1624         forget_all_cached_acls(inode);
1625         if (lli->lli_posix_acl) {
1626                 posix_acl_release(lli->lli_posix_acl);
1627                 lli->lli_posix_acl = NULL;
1628         }
1629 #endif
1630         lli->lli_inode_magic = LLI_INODE_DEAD;
1631
1632         if (S_ISDIR(inode->i_mode))
1633                 ll_dir_clear_lsm_md(inode);
1634         else if (S_ISREG(inode->i_mode) && !is_bad_inode(inode))
1635                 LASSERT(list_empty(&lli->lli_agl_list));
1636
1637         /*
1638          * XXX This has to be done before lsm is freed below, because
1639          * cl_object still uses inode lsm.
1640          */
1641         cl_inode_fini(inode);
1642
1643         EXIT;
1644 }
1645
1646 static int ll_md_setattr(struct dentry *dentry, struct md_op_data *op_data)
1647 {
1648         struct lustre_md md;
1649         struct inode *inode = dentry->d_inode;
1650         struct ll_sb_info *sbi = ll_i2sbi(inode);
1651         struct ptlrpc_request *request = NULL;
1652         int rc, ia_valid;
1653         ENTRY;
1654
1655         op_data = ll_prep_md_op_data(op_data, inode, NULL, NULL, 0, 0,
1656                                      LUSTRE_OPC_ANY, NULL);
1657         if (IS_ERR(op_data))
1658                 RETURN(PTR_ERR(op_data));
1659
1660         rc = md_setattr(sbi->ll_md_exp, op_data, NULL, 0, &request);
1661         if (rc) {
1662                 ptlrpc_req_finished(request);
1663                 if (rc == -ENOENT) {
1664                         clear_nlink(inode);
1665                         /* Unlinked special device node? Or just a race?
1666                          * Pretend we done everything. */
1667                         if (!S_ISREG(inode->i_mode) &&
1668                             !S_ISDIR(inode->i_mode)) {
1669                                 ia_valid = op_data->op_attr.ia_valid;
1670                                 op_data->op_attr.ia_valid &= ~TIMES_SET_FLAGS;
1671                                 rc = simple_setattr(dentry, &op_data->op_attr);
1672                                 op_data->op_attr.ia_valid = ia_valid;
1673                         }
1674                 } else if (rc != -EPERM && rc != -EACCES && rc != -ETXTBSY) {
1675                         CERROR("md_setattr fails: rc = %d\n", rc);
1676                 }
1677                 RETURN(rc);
1678         }
1679
1680         rc = md_get_lustre_md(sbi->ll_md_exp, request, sbi->ll_dt_exp,
1681                               sbi->ll_md_exp, &md);
1682         if (rc) {
1683                 ptlrpc_req_finished(request);
1684                 RETURN(rc);
1685         }
1686
1687         ia_valid = op_data->op_attr.ia_valid;
1688         /* inode size will be in ll_setattr_ost, can't do it now since dirty
1689          * cache is not cleared yet. */
1690         op_data->op_attr.ia_valid &= ~(TIMES_SET_FLAGS | ATTR_SIZE);
1691         if (S_ISREG(inode->i_mode))
1692                 inode_lock(inode);
1693         rc = simple_setattr(dentry, &op_data->op_attr);
1694         if (S_ISREG(inode->i_mode))
1695                 inode_unlock(inode);
1696         op_data->op_attr.ia_valid = ia_valid;
1697
1698         rc = ll_update_inode(inode, &md);
1699         ptlrpc_req_finished(request);
1700
1701         RETURN(rc);
1702 }
1703
1704 /* If this inode has objects allocated to it (lsm != NULL), then the OST
1705  * object(s) determine the file size and mtime.  Otherwise, the MDS will
1706  * keep these values until such a time that objects are allocated for it.
1707  * We do the MDS operations first, as it is checking permissions for us.
1708  * We don't to the MDS RPC if there is nothing that we want to store there,
1709  * otherwise there is no harm in updating mtime/atime on the MDS if we are
1710  * going to do an RPC anyways.
1711  *
1712  * If we are doing a truncate, we will send the mtime and ctime updates
1713  * to the OST with the punch RPC, otherwise we do an explicit setattr RPC.
1714  * I don't believe it is possible to get e.g. ATTR_MTIME_SET and ATTR_SIZE
1715  * at the same time.
1716  *
1717  * In case of HSMimport, we only set attr on MDS.
1718  */
1719 int ll_setattr_raw(struct dentry *dentry, struct iattr *attr,
1720                    enum op_xvalid xvalid, bool hsm_import)
1721 {
1722         struct inode *inode = dentry->d_inode;
1723         struct ll_inode_info *lli = ll_i2info(inode);
1724         struct md_op_data *op_data = NULL;
1725         int rc = 0;
1726
1727         ENTRY;
1728
1729         CDEBUG(D_VFSTRACE, "%s: setattr inode "DFID"(%p) from %llu to %llu, "
1730                "valid %x, hsm_import %d\n",
1731                ll_i2sbi(inode)->ll_fsname, PFID(&lli->lli_fid),
1732                inode, i_size_read(inode), attr->ia_size, attr->ia_valid,
1733                hsm_import);
1734
1735         if (attr->ia_valid & ATTR_SIZE) {
1736                 /* Check new size against VFS/VM file size limit and rlimit */
1737                 rc = inode_newsize_ok(inode, attr->ia_size);
1738                 if (rc)
1739                         RETURN(rc);
1740
1741                 /* The maximum Lustre file size is variable, based on the
1742                  * OST maximum object size and number of stripes.  This
1743                  * needs another check in addition to the VFS check above. */
1744                 if (attr->ia_size > ll_file_maxbytes(inode)) {
1745                         CDEBUG(D_INODE,"file "DFID" too large %llu > %llu\n",
1746                                PFID(&lli->lli_fid), attr->ia_size,
1747                                ll_file_maxbytes(inode));
1748                         RETURN(-EFBIG);
1749                 }
1750
1751                 attr->ia_valid |= ATTR_MTIME | ATTR_CTIME;
1752         }
1753
1754         /* POSIX: check before ATTR_*TIME_SET set (from inode_change_ok) */
1755         if (attr->ia_valid & TIMES_SET_FLAGS) {
1756                 if ((!uid_eq(current_fsuid(), inode->i_uid)) &&
1757                     !cfs_capable(CFS_CAP_FOWNER))
1758                         RETURN(-EPERM);
1759         }
1760
1761         /* We mark all of the fields "set" so MDS/OST does not re-set them */
1762         if (!(xvalid & OP_XVALID_CTIME_SET) &&
1763              (attr->ia_valid & ATTR_CTIME)) {
1764                 attr->ia_ctime = current_time(inode);
1765                 xvalid |= OP_XVALID_CTIME_SET;
1766         }
1767         if (!(attr->ia_valid & ATTR_ATIME_SET) &&
1768             (attr->ia_valid & ATTR_ATIME)) {
1769                 attr->ia_atime = current_time(inode);
1770                 attr->ia_valid |= ATTR_ATIME_SET;
1771         }
1772         if (!(attr->ia_valid & ATTR_MTIME_SET) &&
1773             (attr->ia_valid & ATTR_MTIME)) {
1774                 attr->ia_mtime = current_time(inode);
1775                 attr->ia_valid |= ATTR_MTIME_SET;
1776         }
1777
1778         if (attr->ia_valid & (ATTR_MTIME | ATTR_CTIME))
1779                 CDEBUG(D_INODE, "setting mtime %lld, ctime %lld, now = %lld\n",
1780                        (s64)attr->ia_mtime.tv_sec, (s64)attr->ia_ctime.tv_sec,
1781                        ktime_get_real_seconds());
1782
1783         if (S_ISREG(inode->i_mode)) {
1784                 if (attr->ia_valid & ATTR_SIZE)
1785                         inode_dio_write_done(inode);
1786                 inode_unlock(inode);
1787         }
1788
1789         /* We always do an MDS RPC, even if we're only changing the size;
1790          * only the MDS knows whether truncate() should fail with -ETXTBUSY */
1791
1792         OBD_ALLOC_PTR(op_data);
1793         if (op_data == NULL)
1794                 GOTO(out, rc = -ENOMEM);
1795
1796         if (!hsm_import && attr->ia_valid & ATTR_SIZE) {
1797                 /* If we are changing file size, file content is
1798                  * modified, flag it.
1799                  */
1800                 xvalid |= OP_XVALID_OWNEROVERRIDE;
1801                 op_data->op_bias |= MDS_DATA_MODIFIED;
1802                 ll_file_clear_flag(lli, LLIF_DATA_MODIFIED);
1803         }
1804
1805         if (attr->ia_valid & ATTR_FILE) {
1806                 struct ll_file_data *fd = LUSTRE_FPRIVATE(attr->ia_file);
1807
1808                 if (fd->fd_lease_och)
1809                         op_data->op_bias |= MDS_TRUNC_KEEP_LEASE;
1810         }
1811
1812         op_data->op_attr = *attr;
1813         op_data->op_xvalid = xvalid;
1814
1815         rc = ll_md_setattr(dentry, op_data);
1816         if (rc)
1817                 GOTO(out, rc);
1818
1819         if (!S_ISREG(inode->i_mode) || hsm_import)
1820                 GOTO(out, rc = 0);
1821
1822         if (attr->ia_valid & (ATTR_SIZE | ATTR_ATIME | ATTR_ATIME_SET |
1823                               ATTR_MTIME | ATTR_MTIME_SET | ATTR_CTIME) ||
1824             xvalid & OP_XVALID_CTIME_SET) {
1825                 bool cached = false;
1826
1827                 rc = pcc_inode_setattr(inode, attr, &cached);
1828                 if (cached) {
1829                         if (rc) {
1830                                 CERROR("%s: PCC inode "DFID" setattr failed: "
1831                                        "rc = %d\n",
1832                                        ll_i2sbi(inode)->ll_fsname,
1833                                        PFID(&lli->lli_fid), rc);
1834                                 GOTO(out, rc);
1835                         }
1836                 } else {
1837                         /* For truncate and utimes sending attributes to OSTs,
1838                          * setting mtime/atime to the past will be performed
1839                          * under PW [0:EOF] extent lock (new_size:EOF for
1840                          * truncate). It may seem excessive to send mtime/atime
1841                          * updates to OSTs when not setting times to past, but
1842                          * it is necessary due to possible time
1843                          * de-synchronization between MDT inode and OST objects
1844                          */
1845                         rc = cl_setattr_ost(lli->lli_clob, attr, xvalid, 0);
1846                 }
1847         }
1848
1849         /* If the file was restored, it needs to set dirty flag.
1850          *
1851          * We've already sent MDS_DATA_MODIFIED flag in
1852          * ll_md_setattr() for truncate. However, the MDT refuses to
1853          * set the HS_DIRTY flag on released files, so we have to set
1854          * it again if the file has been restored. Please check how
1855          * LLIF_DATA_MODIFIED is set in vvp_io_setattr_fini().
1856          *
1857          * Please notice that if the file is not released, the previous
1858          * MDS_DATA_MODIFIED has taken effect and usually
1859          * LLIF_DATA_MODIFIED is not set(see vvp_io_setattr_fini()).
1860          * This way we can save an RPC for common open + trunc
1861          * operation. */
1862         if (ll_file_test_and_clear_flag(lli, LLIF_DATA_MODIFIED)) {
1863                 struct hsm_state_set hss = {
1864                         .hss_valid = HSS_SETMASK,
1865                         .hss_setmask = HS_DIRTY,
1866                 };
1867                 int rc2;
1868
1869                 rc2 = ll_hsm_state_set(inode, &hss);
1870                 /* truncate and write can happen at the same time, so that
1871                  * the file can be set modified even though the file is not
1872                  * restored from released state, and ll_hsm_state_set() is
1873                  * not applicable for the file, and rc2 < 0 is normal in this
1874                  * case. */
1875                 if (rc2 < 0)
1876                         CDEBUG(D_INFO, DFID "HSM set dirty failed: rc2 = %d\n",
1877                                PFID(ll_inode2fid(inode)), rc2);
1878         }
1879
1880         EXIT;
1881 out:
1882         if (op_data != NULL)
1883                 ll_finish_md_op_data(op_data);
1884
1885         if (S_ISREG(inode->i_mode)) {
1886                 inode_lock(inode);
1887                 if ((attr->ia_valid & ATTR_SIZE) && !hsm_import)
1888                         inode_dio_wait(inode);
1889                 /* Once we've got the i_mutex, it's safe to set the S_NOSEC
1890                  * flag.  ll_update_inode (called from ll_md_setattr), clears
1891                  * inode flags, so there is a gap where S_NOSEC is not set.
1892                  * This can cause a writer to take the i_mutex unnecessarily,
1893                  * but this is safe to do and should be rare. */
1894                 inode_has_no_xattr(inode);
1895         }
1896
1897         ll_stats_ops_tally(ll_i2sbi(inode), (attr->ia_valid & ATTR_SIZE) ?
1898                         LPROC_LL_TRUNC : LPROC_LL_SETATTR, 1);
1899
1900         return rc;
1901 }
1902
1903 int ll_setattr(struct dentry *de, struct iattr *attr)
1904 {
1905         int mode = de->d_inode->i_mode;
1906         enum op_xvalid xvalid = 0;
1907
1908         if ((attr->ia_valid & (ATTR_CTIME|ATTR_SIZE|ATTR_MODE)) ==
1909                               (ATTR_CTIME|ATTR_SIZE|ATTR_MODE))
1910                 xvalid |= OP_XVALID_OWNEROVERRIDE;
1911
1912         if (((attr->ia_valid & (ATTR_MODE|ATTR_FORCE|ATTR_SIZE)) ==
1913                                (ATTR_SIZE|ATTR_MODE)) &&
1914             (((mode & S_ISUID) && !(attr->ia_mode & S_ISUID)) ||
1915              (((mode & (S_ISGID|S_IXGRP)) == (S_ISGID|S_IXGRP)) &&
1916               !(attr->ia_mode & S_ISGID))))
1917                 attr->ia_valid |= ATTR_FORCE;
1918
1919         if ((attr->ia_valid & ATTR_MODE) &&
1920             (mode & S_ISUID) &&
1921             !(attr->ia_mode & S_ISUID) &&
1922             !(attr->ia_valid & ATTR_KILL_SUID))
1923                 attr->ia_valid |= ATTR_KILL_SUID;
1924
1925         if ((attr->ia_valid & ATTR_MODE) &&
1926             ((mode & (S_ISGID|S_IXGRP)) == (S_ISGID|S_IXGRP)) &&
1927             !(attr->ia_mode & S_ISGID) &&
1928             !(attr->ia_valid & ATTR_KILL_SGID))
1929                 attr->ia_valid |= ATTR_KILL_SGID;
1930
1931         return ll_setattr_raw(de, attr, xvalid, false);
1932 }
1933
1934 int ll_statfs_internal(struct ll_sb_info *sbi, struct obd_statfs *osfs,
1935                        u32 flags)
1936 {
1937         struct obd_statfs obd_osfs = { 0 };
1938         time64_t max_age;
1939         int rc;
1940
1941         ENTRY;
1942         max_age = ktime_get_seconds() - sbi->ll_statfs_max_age;
1943
1944         rc = obd_statfs(NULL, sbi->ll_md_exp, osfs, max_age, flags);
1945         if (rc)
1946                 RETURN(rc);
1947
1948         osfs->os_type = LL_SUPER_MAGIC;
1949
1950         CDEBUG(D_SUPER, "MDC blocks %llu/%llu objects %llu/%llu\n",
1951               osfs->os_bavail, osfs->os_blocks, osfs->os_ffree, osfs->os_files);
1952
1953         if (osfs->os_state & OS_STATE_SUM)
1954                 GOTO(out, rc);
1955
1956         if (sbi->ll_flags & LL_SBI_LAZYSTATFS)
1957                 flags |= OBD_STATFS_NODELAY;
1958
1959         rc = obd_statfs(NULL, sbi->ll_dt_exp, &obd_osfs, max_age, flags);
1960         if (rc) /* Possibly a filesystem with no OSTs.  Report MDT totals. */
1961                 GOTO(out, rc = 0);
1962
1963         CDEBUG(D_SUPER, "OSC blocks %llu/%llu objects %llu/%llu\n",
1964                obd_osfs.os_bavail, obd_osfs.os_blocks, obd_osfs.os_ffree,
1965                obd_osfs.os_files);
1966
1967         osfs->os_bsize = obd_osfs.os_bsize;
1968         osfs->os_blocks = obd_osfs.os_blocks;
1969         osfs->os_bfree = obd_osfs.os_bfree;
1970         osfs->os_bavail = obd_osfs.os_bavail;
1971
1972         /* If we have _some_ OSTs, but don't have as many free objects on the
1973          * OSTs as inodes on the MDTs, reduce the reported number of inodes
1974          * to compensate, so that the "inodes in use" number is correct.
1975          * This should be kept in sync with lod_statfs() behaviour.
1976          */
1977         if (obd_osfs.os_files && obd_osfs.os_ffree < osfs->os_ffree) {
1978                 osfs->os_files = (osfs->os_files - osfs->os_ffree) +
1979                                  obd_osfs.os_ffree;
1980                 osfs->os_ffree = obd_osfs.os_ffree;
1981         }
1982
1983 out:
1984         RETURN(rc);
1985 }
1986 int ll_statfs(struct dentry *de, struct kstatfs *sfs)
1987 {
1988         struct super_block *sb = de->d_sb;
1989         struct obd_statfs osfs;
1990         __u64 fsid = huge_encode_dev(sb->s_dev);
1991         int rc;
1992
1993         CDEBUG(D_VFSTRACE, "VFS Op: at %llu jiffies\n", get_jiffies_64());
1994         ll_stats_ops_tally(ll_s2sbi(sb), LPROC_LL_STATFS, 1);
1995
1996         /* Some amount of caching on the client is allowed */
1997         rc = ll_statfs_internal(ll_s2sbi(sb), &osfs, OBD_STATFS_SUM);
1998         if (rc)
1999                 return rc;
2000
2001         statfs_unpack(sfs, &osfs);
2002
2003         /* We need to downshift for all 32-bit kernels, because we can't
2004          * tell if the kernel is being called via sys_statfs64() or not.
2005          * Stop before overflowing f_bsize - in which case it is better
2006          * to just risk EOVERFLOW if caller is using old sys_statfs(). */
2007         if (sizeof(long) < 8) {
2008                 while (osfs.os_blocks > ~0UL && sfs->f_bsize < 0x40000000) {
2009                         sfs->f_bsize <<= 1;
2010
2011                         osfs.os_blocks >>= 1;
2012                         osfs.os_bfree >>= 1;
2013                         osfs.os_bavail >>= 1;
2014                 }
2015         }
2016
2017         sfs->f_blocks = osfs.os_blocks;
2018         sfs->f_bfree = osfs.os_bfree;
2019         sfs->f_bavail = osfs.os_bavail;
2020         sfs->f_fsid.val[0] = (__u32)fsid;
2021         sfs->f_fsid.val[1] = (__u32)(fsid >> 32);
2022         return 0;
2023 }
2024
2025 void ll_inode_size_lock(struct inode *inode)
2026 {
2027         struct ll_inode_info *lli;
2028
2029         LASSERT(!S_ISDIR(inode->i_mode));
2030
2031         lli = ll_i2info(inode);
2032         mutex_lock(&lli->lli_size_mutex);
2033 }
2034
2035 void ll_inode_size_unlock(struct inode *inode)
2036 {
2037         struct ll_inode_info *lli;
2038
2039         lli = ll_i2info(inode);
2040         mutex_unlock(&lli->lli_size_mutex);
2041 }
2042
2043 void ll_update_inode_flags(struct inode *inode, int ext_flags)
2044 {
2045         inode->i_flags = ll_ext_to_inode_flags(ext_flags);
2046         if (ext_flags & LUSTRE_PROJINHERIT_FL)
2047                 ll_file_set_flag(ll_i2info(inode), LLIF_PROJECT_INHERIT);
2048         else
2049                 ll_file_clear_flag(ll_i2info(inode), LLIF_PROJECT_INHERIT);
2050 }
2051
2052 int ll_update_inode(struct inode *inode, struct lustre_md *md)
2053 {
2054         struct ll_inode_info *lli = ll_i2info(inode);
2055         struct mdt_body *body = md->body;
2056         struct ll_sb_info *sbi = ll_i2sbi(inode);
2057         int rc = 0;
2058
2059         if (body->mbo_valid & OBD_MD_FLEASIZE) {
2060                 rc = cl_file_inode_init(inode, md);
2061                 if (rc)
2062                         return rc;
2063         }
2064
2065         if (S_ISDIR(inode->i_mode)) {
2066                 rc = ll_update_lsm_md(inode, md);
2067                 if (rc != 0)
2068                         return rc;
2069         }
2070
2071 #ifdef CONFIG_FS_POSIX_ACL
2072         if (body->mbo_valid & OBD_MD_FLACL) {
2073                 spin_lock(&lli->lli_lock);
2074                 if (lli->lli_posix_acl)
2075                         posix_acl_release(lli->lli_posix_acl);
2076                 lli->lli_posix_acl = md->posix_acl;
2077                 spin_unlock(&lli->lli_lock);
2078         }
2079 #endif
2080         inode->i_ino = cl_fid_build_ino(&body->mbo_fid1,
2081                                         sbi->ll_flags & LL_SBI_32BIT_API);
2082         inode->i_generation = cl_fid_build_gen(&body->mbo_fid1);
2083
2084         if (body->mbo_valid & OBD_MD_FLATIME) {
2085                 if (body->mbo_atime > inode->i_atime.tv_sec)
2086                         inode->i_atime.tv_sec = body->mbo_atime;
2087                 lli->lli_atime = body->mbo_atime;
2088         }
2089
2090         if (body->mbo_valid & OBD_MD_FLMTIME) {
2091                 if (body->mbo_mtime > inode->i_mtime.tv_sec) {
2092                         CDEBUG(D_INODE,
2093                                "setting ino %lu mtime from %lld to %llu\n",
2094                                inode->i_ino, (s64)inode->i_mtime.tv_sec,
2095                                body->mbo_mtime);
2096                         inode->i_mtime.tv_sec = body->mbo_mtime;
2097                 }
2098                 lli->lli_mtime = body->mbo_mtime;
2099         }
2100
2101         if (body->mbo_valid & OBD_MD_FLCTIME) {
2102                 if (body->mbo_ctime > inode->i_ctime.tv_sec)
2103                         inode->i_ctime.tv_sec = body->mbo_ctime;
2104                 lli->lli_ctime = body->mbo_ctime;
2105         }
2106
2107         /* Clear i_flags to remove S_NOSEC before permissions are updated */
2108         if (body->mbo_valid & OBD_MD_FLFLAGS)
2109                 ll_update_inode_flags(inode, body->mbo_flags);
2110         if (body->mbo_valid & OBD_MD_FLMODE)
2111                 inode->i_mode = (inode->i_mode & S_IFMT) |
2112                                 (body->mbo_mode & ~S_IFMT);
2113
2114         if (body->mbo_valid & OBD_MD_FLTYPE)
2115                 inode->i_mode = (inode->i_mode & ~S_IFMT) |
2116                                 (body->mbo_mode & S_IFMT);
2117
2118         LASSERT(inode->i_mode != 0);
2119         if (S_ISREG(inode->i_mode))
2120                 inode->i_blkbits = min(PTLRPC_MAX_BRW_BITS + 1,
2121                                        LL_MAX_BLKSIZE_BITS);
2122         else
2123                 inode->i_blkbits = inode->i_sb->s_blocksize_bits;
2124
2125         if (body->mbo_valid & OBD_MD_FLUID)
2126                 inode->i_uid = make_kuid(&init_user_ns, body->mbo_uid);
2127         if (body->mbo_valid & OBD_MD_FLGID)
2128                 inode->i_gid = make_kgid(&init_user_ns, body->mbo_gid);
2129         if (body->mbo_valid & OBD_MD_FLPROJID)
2130                 lli->lli_projid = body->mbo_projid;
2131         if (body->mbo_valid & OBD_MD_FLNLINK)
2132                 set_nlink(inode, body->mbo_nlink);
2133         if (body->mbo_valid & OBD_MD_FLRDEV)
2134                 inode->i_rdev = old_decode_dev(body->mbo_rdev);
2135
2136         if (body->mbo_valid & OBD_MD_FLID) {
2137                 /* FID shouldn't be changed! */
2138                 if (fid_is_sane(&lli->lli_fid)) {
2139                         LASSERTF(lu_fid_eq(&lli->lli_fid, &body->mbo_fid1),
2140                                  "Trying to change FID "DFID
2141                                  " to the "DFID", inode "DFID"(%p)\n",
2142                                  PFID(&lli->lli_fid), PFID(&body->mbo_fid1),
2143                                  PFID(ll_inode2fid(inode)), inode);
2144                 } else {
2145                         lli->lli_fid = body->mbo_fid1;
2146                 }
2147         }
2148
2149         LASSERT(fid_seq(&lli->lli_fid) != 0);
2150
2151         if (body->mbo_valid & OBD_MD_FLSIZE) {
2152                 i_size_write(inode, body->mbo_size);
2153
2154                 CDEBUG(D_VFSTRACE, "inode="DFID", updating i_size %llu\n",
2155                        PFID(ll_inode2fid(inode)),
2156                        (unsigned long long)body->mbo_size);
2157
2158                 if (body->mbo_valid & OBD_MD_FLBLOCKS)
2159                         inode->i_blocks = body->mbo_blocks;
2160         }
2161
2162         if (body->mbo_valid & OBD_MD_TSTATE) {
2163                 /* Set LLIF_FILE_RESTORING if restore ongoing and
2164                  * clear it when done to ensure to start again
2165                  * glimpsing updated attrs
2166                  */
2167                 if (body->mbo_t_state & MS_RESTORE)
2168                         ll_file_set_flag(lli, LLIF_FILE_RESTORING);
2169                 else
2170                         ll_file_clear_flag(lli, LLIF_FILE_RESTORING);
2171         }
2172
2173         return 0;
2174 }
2175
2176 int ll_read_inode2(struct inode *inode, void *opaque)
2177 {
2178         struct lustre_md *md = opaque;
2179         struct ll_inode_info *lli = ll_i2info(inode);
2180         int     rc;
2181         ENTRY;
2182
2183         CDEBUG(D_VFSTRACE, "VFS Op:inode="DFID"(%p)\n",
2184                PFID(&lli->lli_fid), inode);
2185
2186         /* Core attributes from the MDS first.  This is a new inode, and
2187          * the VFS doesn't zero times in the core inode so we have to do
2188          * it ourselves.  They will be overwritten by either MDS or OST
2189          * attributes - we just need to make sure they aren't newer.
2190          */
2191         inode->i_mtime.tv_sec = 0;
2192         inode->i_atime.tv_sec = 0;
2193         inode->i_ctime.tv_sec = 0;
2194         inode->i_rdev = 0;
2195         rc = ll_update_inode(inode, md);
2196         if (rc != 0)
2197                 RETURN(rc);
2198
2199         /* OIDEBUG(inode); */
2200
2201 #ifdef HAVE_BACKING_DEV_INFO
2202         /* initializing backing dev info. */
2203         inode->i_mapping->backing_dev_info = &s2lsi(inode->i_sb)->lsi_bdi;
2204 #endif
2205         if (S_ISREG(inode->i_mode)) {
2206                 struct ll_sb_info *sbi = ll_i2sbi(inode);
2207                 inode->i_op = &ll_file_inode_operations;
2208                 inode->i_fop = sbi->ll_fop;
2209                 inode->i_mapping->a_ops = (struct address_space_operations *)&ll_aops;
2210                 EXIT;
2211         } else if (S_ISDIR(inode->i_mode)) {
2212                 inode->i_op = &ll_dir_inode_operations;
2213                 inode->i_fop = &ll_dir_operations;
2214                 EXIT;
2215         } else if (S_ISLNK(inode->i_mode)) {
2216                 inode->i_op = &ll_fast_symlink_inode_operations;
2217                 EXIT;
2218         } else {
2219                 inode->i_op = &ll_special_inode_operations;
2220
2221                 init_special_inode(inode, inode->i_mode,
2222                                    inode->i_rdev);
2223
2224                 EXIT;
2225         }
2226
2227         return 0;
2228 }
2229
2230 void ll_delete_inode(struct inode *inode)
2231 {
2232         struct ll_inode_info *lli = ll_i2info(inode);
2233         struct address_space *mapping = &inode->i_data;
2234         unsigned long nrpages;
2235         ENTRY;
2236
2237         if (S_ISREG(inode->i_mode) && lli->lli_clob != NULL) {
2238                 /* It is last chance to write out dirty pages,
2239                  * otherwise we may lose data while umount.
2240                  *
2241                  * If i_nlink is 0 then just discard data. This is safe because
2242                  * local inode gets i_nlink 0 from server only for the last
2243                  * unlink, so that file is not opened somewhere else
2244                  */
2245                 cl_sync_file_range(inode, 0, OBD_OBJECT_EOF, inode->i_nlink ?
2246                                    CL_FSYNC_LOCAL : CL_FSYNC_DISCARD, 1);
2247         }
2248         truncate_inode_pages_final(mapping);
2249
2250         /* Workaround for LU-118: Note nrpages may not be totally updated when
2251          * truncate_inode_pages() returns, as there can be a page in the process
2252          * of deletion (inside __delete_from_page_cache()) in the specified
2253          * range. Thus mapping->nrpages can be non-zero when this function
2254          * returns even after truncation of the whole mapping.  Only do this if
2255          * npages isn't already zero.
2256          */
2257         nrpages = mapping->nrpages;
2258         if (nrpages) {
2259                 xa_lock_irq(&mapping->i_pages);
2260                 nrpages = mapping->nrpages;
2261                 xa_unlock_irq(&mapping->i_pages);
2262         } /* Workaround end */
2263
2264         LASSERTF(nrpages == 0, "%s: inode="DFID"(%p) nrpages=%lu, "
2265                  "see https://jira.whamcloud.com/browse/LU-118\n",
2266                  ll_i2sbi(inode)->ll_fsname,
2267                  PFID(ll_inode2fid(inode)), inode, nrpages);
2268
2269 #ifdef HAVE_SBOPS_EVICT_INODE
2270         ll_clear_inode(inode);
2271 #endif
2272         clear_inode(inode);
2273
2274         EXIT;
2275 }
2276
2277 int ll_iocontrol(struct inode *inode, struct file *file,
2278                  unsigned int cmd, unsigned long arg)
2279 {
2280         struct ll_sb_info *sbi = ll_i2sbi(inode);
2281         struct ptlrpc_request *req = NULL;
2282         int rc, flags = 0;
2283         ENTRY;
2284
2285         switch (cmd) {
2286         case FS_IOC_GETFLAGS: {
2287                 struct mdt_body *body;
2288                 struct md_op_data *op_data;
2289
2290                 op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL,
2291                                              0, 0, LUSTRE_OPC_ANY,
2292                                              NULL);
2293                 if (IS_ERR(op_data))
2294                         RETURN(PTR_ERR(op_data));
2295
2296                 op_data->op_valid = OBD_MD_FLFLAGS;
2297                 rc = md_getattr(sbi->ll_md_exp, op_data, &req);
2298                 ll_finish_md_op_data(op_data);
2299                 if (rc) {
2300                         CERROR("%s: failure inode "DFID": rc = %d\n",
2301                                sbi->ll_md_exp->exp_obd->obd_name,
2302                                PFID(ll_inode2fid(inode)), rc);
2303                         RETURN(-abs(rc));
2304                 }
2305
2306                 body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
2307
2308                 flags = body->mbo_flags;
2309
2310                 ptlrpc_req_finished(req);
2311
2312                 RETURN(put_user(flags, (int __user *)arg));
2313         }
2314         case FS_IOC_SETFLAGS: {
2315                 struct iattr *attr;
2316                 struct md_op_data *op_data;
2317                 struct cl_object *obj;
2318                 struct fsxattr fa = { 0 };
2319
2320                 if (get_user(flags, (int __user *)arg))
2321                         RETURN(-EFAULT);
2322
2323                 fa.fsx_projid = ll_i2info(inode)->lli_projid;
2324                 if (flags & LUSTRE_PROJINHERIT_FL)
2325                         fa.fsx_xflags = FS_XFLAG_PROJINHERIT;
2326
2327                 rc = ll_ioctl_check_project(inode, &fa);
2328                 if (rc)
2329                         RETURN(rc);
2330
2331                 op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL, 0, 0,
2332                                              LUSTRE_OPC_ANY, NULL);
2333                 if (IS_ERR(op_data))
2334                         RETURN(PTR_ERR(op_data));
2335
2336                 op_data->op_attr_flags = flags;
2337                 op_data->op_xvalid |= OP_XVALID_FLAGS;
2338                 rc = md_setattr(sbi->ll_md_exp, op_data, NULL, 0, &req);
2339                 ll_finish_md_op_data(op_data);
2340                 ptlrpc_req_finished(req);
2341                 if (rc)
2342                         RETURN(rc);
2343
2344                 ll_update_inode_flags(inode, flags);
2345
2346                 obj = ll_i2info(inode)->lli_clob;
2347                 if (obj == NULL)
2348                         RETURN(0);
2349
2350                 OBD_ALLOC_PTR(attr);
2351                 if (attr == NULL)
2352                         RETURN(-ENOMEM);
2353
2354                 rc = cl_setattr_ost(obj, attr, OP_XVALID_FLAGS, flags);
2355
2356                 OBD_FREE_PTR(attr);
2357                 RETURN(rc);
2358         }
2359         default:
2360                 RETURN(-ENOSYS);
2361         }
2362
2363         RETURN(0);
2364 }
2365
2366 int ll_flush_ctx(struct inode *inode)
2367 {
2368         struct ll_sb_info  *sbi = ll_i2sbi(inode);
2369
2370         CDEBUG(D_SEC, "flush context for user %d\n",
2371                from_kuid(&init_user_ns, current_uid()));
2372
2373         obd_set_info_async(NULL, sbi->ll_md_exp,
2374                            sizeof(KEY_FLUSH_CTX), KEY_FLUSH_CTX,
2375                            0, NULL, NULL);
2376         obd_set_info_async(NULL, sbi->ll_dt_exp,
2377                            sizeof(KEY_FLUSH_CTX), KEY_FLUSH_CTX,
2378                            0, NULL, NULL);
2379         return 0;
2380 }
2381
2382 /* umount -f client means force down, don't save state */
2383 void ll_umount_begin(struct super_block *sb)
2384 {
2385         struct ll_sb_info *sbi = ll_s2sbi(sb);
2386         struct obd_device *obd;
2387         struct obd_ioctl_data *ioc_data;
2388         struct l_wait_info lwi;
2389         wait_queue_head_t waitq;
2390         ENTRY;
2391
2392         CDEBUG(D_VFSTRACE, "VFS Op: superblock %p count %d active %d\n", sb,
2393                sb->s_count, atomic_read(&sb->s_active));
2394
2395         obd = class_exp2obd(sbi->ll_md_exp);
2396         if (obd == NULL) {
2397                 CERROR("Invalid MDC connection handle %#llx\n",
2398                        sbi->ll_md_exp->exp_handle.h_cookie);
2399                 EXIT;
2400                 return;
2401         }
2402         obd->obd_force = 1;
2403
2404         obd = class_exp2obd(sbi->ll_dt_exp);
2405         if (obd == NULL) {
2406                 CERROR("Invalid LOV connection handle %#llx\n",
2407                        sbi->ll_dt_exp->exp_handle.h_cookie);
2408                 EXIT;
2409                 return;
2410         }
2411         obd->obd_force = 1;
2412
2413         OBD_ALLOC_PTR(ioc_data);
2414         if (ioc_data) {
2415                 obd_iocontrol(IOC_OSC_SET_ACTIVE, sbi->ll_md_exp,
2416                               sizeof *ioc_data, ioc_data, NULL);
2417
2418                 obd_iocontrol(IOC_OSC_SET_ACTIVE, sbi->ll_dt_exp,
2419                               sizeof *ioc_data, ioc_data, NULL);
2420
2421                 OBD_FREE_PTR(ioc_data);
2422         }
2423
2424         /* Really, we'd like to wait until there are no requests outstanding,
2425          * and then continue.  For now, we just periodically checking for vfs
2426          * to decrement mnt_cnt and hope to finish it within 10sec.
2427          */
2428         init_waitqueue_head(&waitq);
2429         lwi = LWI_TIMEOUT_INTERVAL(cfs_time_seconds(10),
2430                                    cfs_time_seconds(1), NULL, NULL);
2431         l_wait_event(waitq, may_umount(sbi->ll_mnt.mnt), &lwi);
2432
2433         EXIT;
2434 }
2435
2436 int ll_remount_fs(struct super_block *sb, int *flags, char *data)
2437 {
2438         struct ll_sb_info *sbi = ll_s2sbi(sb);
2439         char *profilenm = get_profile_name(sb);
2440         int err;
2441         __u32 read_only;
2442
2443         if ((*flags & MS_RDONLY) != (sb->s_flags & MS_RDONLY)) {
2444                 read_only = *flags & MS_RDONLY;
2445                 err = obd_set_info_async(NULL, sbi->ll_md_exp,
2446                                          sizeof(KEY_READ_ONLY),
2447                                          KEY_READ_ONLY, sizeof(read_only),
2448                                          &read_only, NULL);
2449                 if (err) {
2450                         LCONSOLE_WARN("Failed to remount %s %s (%d)\n",
2451                                       profilenm, read_only ?
2452                                       "read-only" : "read-write", err);
2453                         return err;
2454                 }
2455
2456                 if (read_only)
2457                         sb->s_flags |= MS_RDONLY;
2458                 else
2459                         sb->s_flags &= ~MS_RDONLY;
2460
2461                 if (sbi->ll_flags & LL_SBI_VERBOSE)
2462                         LCONSOLE_WARN("Remounted %s %s\n", profilenm,
2463                                       read_only ?  "read-only" : "read-write");
2464         }
2465         return 0;
2466 }
2467
2468 /**
2469  * Cleanup the open handle that is cached on MDT-side.
2470  *
2471  * For open case, the client side open handling thread may hit error
2472  * after the MDT grant the open. Under such case, the client should
2473  * send close RPC to the MDT as cleanup; otherwise, the open handle
2474  * on the MDT will be leaked there until the client umount or evicted.
2475  *
2476  * In further, if someone unlinked the file, because the open handle
2477  * holds the reference on such file/object, then it will block the
2478  * subsequent threads that want to locate such object via FID.
2479  *
2480  * \param[in] sb        super block for this file-system
2481  * \param[in] open_req  pointer to the original open request
2482  */
2483 void ll_open_cleanup(struct super_block *sb, struct ptlrpc_request *open_req)
2484 {
2485         struct mdt_body                 *body;
2486         struct md_op_data               *op_data;
2487         struct ptlrpc_request           *close_req = NULL;
2488         struct obd_export               *exp       = ll_s2sbi(sb)->ll_md_exp;
2489         ENTRY;
2490
2491         body = req_capsule_server_get(&open_req->rq_pill, &RMF_MDT_BODY);
2492         OBD_ALLOC_PTR(op_data);
2493         if (op_data == NULL) {
2494                 CWARN("%s: cannot allocate op_data to release open handle for "
2495                       DFID"\n", ll_s2sbi(sb)->ll_fsname, PFID(&body->mbo_fid1));
2496
2497                 RETURN_EXIT;
2498         }
2499
2500         op_data->op_fid1 = body->mbo_fid1;
2501         op_data->op_open_handle = body->mbo_open_handle;
2502         op_data->op_mod_time = ktime_get_real_seconds();
2503         md_close(exp, op_data, NULL, &close_req);
2504         ptlrpc_req_finished(close_req);
2505         ll_finish_md_op_data(op_data);
2506
2507         EXIT;
2508 }
2509
2510 int ll_prep_inode(struct inode **inode, struct ptlrpc_request *req,
2511                   struct super_block *sb, struct lookup_intent *it)
2512 {
2513         struct ll_sb_info *sbi = NULL;
2514         struct lustre_md md = { NULL };
2515         bool default_lmv_deleted = false;
2516         int rc;
2517
2518         ENTRY;
2519
2520         LASSERT(*inode || sb);
2521         sbi = sb ? ll_s2sbi(sb) : ll_i2sbi(*inode);
2522         rc = md_get_lustre_md(sbi->ll_md_exp, req, sbi->ll_dt_exp,
2523                               sbi->ll_md_exp, &md);
2524         if (rc != 0)
2525                 GOTO(out, rc);
2526
2527         /*
2528          * clear default_lmv only if intent_getattr reply doesn't contain it.
2529          * but it needs to be done after iget, check this early because
2530          * ll_update_lsm_md() may change md.
2531          */
2532         if (it && (it->it_op & (IT_LOOKUP | IT_GETATTR)) &&
2533             S_ISDIR(md.body->mbo_mode) && !md.default_lmv)
2534                 default_lmv_deleted = true;
2535
2536         if (*inode) {
2537                 rc = ll_update_inode(*inode, &md);
2538                 if (rc != 0)
2539                         GOTO(out, rc);
2540         } else {
2541                 LASSERT(sb != NULL);
2542
2543                 /*
2544                  * At this point server returns to client's same fid as client
2545                  * generated for creating. So using ->fid1 is okay here.
2546                  */
2547                 if (!fid_is_sane(&md.body->mbo_fid1)) {
2548                         CERROR("%s: Fid is insane "DFID"\n",
2549                                 sbi->ll_fsname,
2550                                 PFID(&md.body->mbo_fid1));
2551                         GOTO(out, rc = -EINVAL);
2552                 }
2553
2554                 *inode = ll_iget(sb, cl_fid_build_ino(&md.body->mbo_fid1,
2555                                              sbi->ll_flags & LL_SBI_32BIT_API),
2556                                  &md);
2557                 if (IS_ERR(*inode)) {
2558 #ifdef CONFIG_FS_POSIX_ACL
2559                         if (md.posix_acl) {
2560                                 posix_acl_release(md.posix_acl);
2561                                 md.posix_acl = NULL;
2562                         }
2563 #endif
2564                         rc = IS_ERR(*inode) ? PTR_ERR(*inode) : -ENOMEM;
2565                         *inode = NULL;
2566                         CERROR("new_inode -fatal: rc %d\n", rc);
2567                         GOTO(out, rc);
2568                 }
2569         }
2570
2571         /* Handling piggyback layout lock.
2572          * Layout lock can be piggybacked by getattr and open request.
2573          * The lsm can be applied to inode only if it comes with a layout lock
2574          * otherwise correct layout may be overwritten, for example:
2575          * 1. proc1: mdt returns a lsm but not granting layout
2576          * 2. layout was changed by another client
2577          * 3. proc2: refresh layout and layout lock granted
2578          * 4. proc1: to apply a stale layout */
2579         if (it != NULL && it->it_lock_mode != 0) {
2580                 struct lustre_handle lockh;
2581                 struct ldlm_lock *lock;
2582
2583                 lockh.cookie = it->it_lock_handle;
2584                 lock = ldlm_handle2lock(&lockh);
2585                 LASSERT(lock != NULL);
2586                 if (ldlm_has_layout(lock)) {
2587                         struct cl_object_conf conf;
2588
2589                         memset(&conf, 0, sizeof(conf));
2590                         conf.coc_opc = OBJECT_CONF_SET;
2591                         conf.coc_inode = *inode;
2592                         conf.coc_lock = lock;
2593                         conf.u.coc_layout = md.layout;
2594                         (void)ll_layout_conf(*inode, &conf);
2595                 }
2596                 LDLM_LOCK_PUT(lock);
2597         }
2598
2599         if (default_lmv_deleted)
2600                 ll_update_default_lsm_md(*inode, &md);
2601
2602         GOTO(out, rc = 0);
2603
2604 out:
2605         /* cleanup will be done if necessary */
2606         md_free_lustre_md(sbi->ll_md_exp, &md);
2607
2608         if (rc != 0 && it != NULL && it->it_op & IT_OPEN)
2609                 ll_open_cleanup(sb != NULL ? sb : (*inode)->i_sb, req);
2610
2611         return rc;
2612 }
2613
2614 int ll_obd_statfs(struct inode *inode, void __user *arg)
2615 {
2616         struct ll_sb_info *sbi = NULL;
2617         struct obd_export *exp;
2618         char *buf = NULL;
2619         struct obd_ioctl_data *data = NULL;
2620         __u32 type;
2621         int len = 0, rc;
2622
2623         if (!inode || !(sbi = ll_i2sbi(inode)))
2624                 GOTO(out_statfs, rc = -EINVAL);
2625
2626         rc = obd_ioctl_getdata(&buf, &len, arg);
2627         if (rc)
2628                 GOTO(out_statfs, rc);
2629
2630         data = (void*)buf;
2631         if (!data->ioc_inlbuf1 || !data->ioc_inlbuf2 ||
2632             !data->ioc_pbuf1 || !data->ioc_pbuf2)
2633                 GOTO(out_statfs, rc = -EINVAL);
2634
2635         if (data->ioc_inllen1 != sizeof(__u32) ||
2636             data->ioc_inllen2 != sizeof(__u32) ||
2637             data->ioc_plen1 != sizeof(struct obd_statfs) ||
2638             data->ioc_plen2 != sizeof(struct obd_uuid))
2639                 GOTO(out_statfs, rc = -EINVAL);
2640
2641         memcpy(&type, data->ioc_inlbuf1, sizeof(__u32));
2642         if (type & LL_STATFS_LMV)
2643                 exp = sbi->ll_md_exp;
2644         else if (type & LL_STATFS_LOV)
2645                 exp = sbi->ll_dt_exp;
2646         else
2647                 GOTO(out_statfs, rc = -ENODEV);
2648
2649         rc = obd_iocontrol(IOC_OBD_STATFS, exp, len, buf, NULL);
2650         if (rc)
2651                 GOTO(out_statfs, rc);
2652 out_statfs:
2653         OBD_FREE_LARGE(buf, len);
2654         return rc;
2655 }
2656
2657 /*
2658  * this is normally called in ll_fini_md_op_data(), but sometimes it needs to
2659  * be called early to avoid deadlock.
2660  */
2661 void ll_unlock_md_op_lsm(struct md_op_data *op_data)
2662 {
2663         if (op_data->op_mea2_sem) {
2664                 up_read(op_data->op_mea2_sem);
2665                 op_data->op_mea2_sem = NULL;
2666         }
2667
2668         if (op_data->op_mea1_sem) {
2669                 up_read(op_data->op_mea1_sem);
2670                 op_data->op_mea1_sem = NULL;
2671         }
2672 }
2673
2674 /* this function prepares md_op_data hint for passing it down to MD stack. */
2675 struct md_op_data *ll_prep_md_op_data(struct md_op_data *op_data,
2676                                       struct inode *i1, struct inode *i2,
2677                                       const char *name, size_t namelen,
2678                                       __u32 mode, enum md_op_code opc,
2679                                       void *data)
2680 {
2681         LASSERT(i1 != NULL);
2682
2683         if (name == NULL) {
2684                 /* Do not reuse namelen for something else. */
2685                 if (namelen != 0)
2686                         return ERR_PTR(-EINVAL);
2687         } else {
2688                 if (namelen > ll_i2sbi(i1)->ll_namelen)
2689                         return ERR_PTR(-ENAMETOOLONG);
2690
2691                 if (!lu_name_is_valid_2(name, namelen))
2692                         return ERR_PTR(-EINVAL);
2693         }
2694
2695         if (op_data == NULL)
2696                 OBD_ALLOC_PTR(op_data);
2697
2698         if (op_data == NULL)
2699                 return ERR_PTR(-ENOMEM);
2700
2701         ll_i2gids(op_data->op_suppgids, i1, i2);
2702         op_data->op_fid1 = *ll_inode2fid(i1);
2703         op_data->op_code = opc;
2704
2705         if (S_ISDIR(i1->i_mode)) {
2706                 down_read(&ll_i2info(i1)->lli_lsm_sem);
2707                 op_data->op_mea1_sem = &ll_i2info(i1)->lli_lsm_sem;
2708                 op_data->op_mea1 = ll_i2info(i1)->lli_lsm_md;
2709                 op_data->op_default_mea1 = ll_i2info(i1)->lli_default_lsm_md;
2710         }
2711
2712         if (i2) {
2713                 op_data->op_fid2 = *ll_inode2fid(i2);
2714                 if (S_ISDIR(i2->i_mode)) {
2715                         if (i2 != i1) {
2716                                 down_read(&ll_i2info(i2)->lli_lsm_sem);
2717                                 op_data->op_mea2_sem =
2718                                                 &ll_i2info(i2)->lli_lsm_sem;
2719                         }
2720                         op_data->op_mea2 = ll_i2info(i2)->lli_lsm_md;
2721                 }
2722         } else {
2723                 fid_zero(&op_data->op_fid2);
2724         }
2725
2726         if (ll_i2sbi(i1)->ll_flags & LL_SBI_64BIT_HASH)
2727                 op_data->op_cli_flags |= CLI_HASH64;
2728
2729         if (ll_need_32bit_api(ll_i2sbi(i1)))
2730                 op_data->op_cli_flags |= CLI_API32;
2731
2732         op_data->op_name = name;
2733         op_data->op_namelen = namelen;
2734         op_data->op_mode = mode;
2735         op_data->op_mod_time = ktime_get_real_seconds();
2736         op_data->op_fsuid = from_kuid(&init_user_ns, current_fsuid());
2737         op_data->op_fsgid = from_kgid(&init_user_ns, current_fsgid());
2738         op_data->op_cap = cfs_curproc_cap_pack();
2739         op_data->op_mds = 0;
2740         if ((opc == LUSTRE_OPC_CREATE) && (name != NULL) &&
2741              filename_is_volatile(name, namelen, &op_data->op_mds)) {
2742                 op_data->op_bias |= MDS_CREATE_VOLATILE;
2743         }
2744         op_data->op_data = data;
2745
2746         return op_data;
2747 }
2748
2749 void ll_finish_md_op_data(struct md_op_data *op_data)
2750 {
2751         ll_unlock_md_op_lsm(op_data);
2752         security_release_secctx(op_data->op_file_secctx,
2753                                 op_data->op_file_secctx_size);
2754         OBD_FREE_PTR(op_data);
2755 }
2756
2757 #ifdef HAVE_SUPEROPS_USE_DENTRY
2758 int ll_show_options(struct seq_file *seq, struct dentry *dentry)
2759 #else
2760 int ll_show_options(struct seq_file *seq, struct vfsmount *vfs)
2761 #endif
2762 {
2763         struct ll_sb_info *sbi;
2764
2765 #ifdef HAVE_SUPEROPS_USE_DENTRY
2766         LASSERT((seq != NULL) && (dentry != NULL));
2767         sbi = ll_s2sbi(dentry->d_sb);
2768 #else
2769         LASSERT((seq != NULL) && (vfs != NULL));
2770         sbi = ll_s2sbi(vfs->mnt_sb);
2771 #endif
2772
2773         if (sbi->ll_flags & LL_SBI_NOLCK)
2774                 seq_puts(seq, ",nolock");
2775
2776         if (sbi->ll_flags & LL_SBI_FLOCK)
2777                 seq_puts(seq, ",flock");
2778
2779         if (sbi->ll_flags & LL_SBI_LOCALFLOCK)
2780                 seq_puts(seq, ",localflock");
2781
2782         if (sbi->ll_flags & LL_SBI_USER_XATTR)
2783                 seq_puts(seq, ",user_xattr");
2784
2785         if (sbi->ll_flags & LL_SBI_LAZYSTATFS)
2786                 seq_puts(seq, ",lazystatfs");
2787
2788         if (sbi->ll_flags & LL_SBI_USER_FID2PATH)
2789                 seq_puts(seq, ",user_fid2path");
2790
2791         if (sbi->ll_flags & LL_SBI_ALWAYS_PING)
2792                 seq_puts(seq, ",always_ping");
2793
2794         RETURN(0);
2795 }
2796
2797 /**
2798  * Get obd name by cmd, and copy out to user space
2799  */
2800 int ll_get_obd_name(struct inode *inode, unsigned int cmd, unsigned long arg)
2801 {
2802         struct ll_sb_info *sbi = ll_i2sbi(inode);
2803         struct obd_device *obd;
2804         ENTRY;
2805
2806         if (cmd == OBD_IOC_GETDTNAME)
2807                 obd = class_exp2obd(sbi->ll_dt_exp);
2808         else if (cmd == OBD_IOC_GETMDNAME)
2809                 obd = class_exp2obd(sbi->ll_md_exp);
2810         else
2811                 RETURN(-EINVAL);
2812
2813         if (!obd)
2814                 RETURN(-ENOENT);
2815
2816         if (copy_to_user((void __user *)arg, obd->obd_name,
2817                          strlen(obd->obd_name) + 1))
2818                 RETURN(-EFAULT);
2819
2820         RETURN(0);
2821 }
2822
2823 static char* ll_d_path(struct dentry *dentry, char *buf, int bufsize)
2824 {
2825         char *path = NULL;
2826
2827         struct path p;
2828
2829         p.dentry = dentry;
2830         p.mnt = current->fs->root.mnt;
2831         path_get(&p);
2832         path = d_path(&p, buf, bufsize);
2833         path_put(&p);
2834         return path;
2835 }
2836
2837 void ll_dirty_page_discard_warn(struct page *page, int ioret)
2838 {
2839         char *buf, *path = NULL;
2840         struct dentry *dentry = NULL;
2841         struct inode *inode = page->mapping->host;
2842
2843         /* this can be called inside spin lock so use GFP_ATOMIC. */
2844         buf = (char *)__get_free_page(GFP_ATOMIC);
2845         if (buf != NULL) {
2846                 dentry = d_find_alias(page->mapping->host);
2847                 if (dentry != NULL)
2848                         path = ll_d_path(dentry, buf, PAGE_SIZE);
2849         }
2850
2851         CDEBUG(D_WARNING,
2852                "%s: dirty page discard: %s/fid: "DFID"/%s may get corrupted "
2853                "(rc %d)\n", ll_i2sbi(inode)->ll_fsname,
2854                s2lsi(page->mapping->host->i_sb)->lsi_lmd->lmd_dev,
2855                PFID(ll_inode2fid(inode)),
2856                (path && !IS_ERR(path)) ? path : "", ioret);
2857
2858         if (dentry != NULL)
2859                 dput(dentry);
2860
2861         if (buf != NULL)
2862                 free_page((unsigned long)buf);
2863 }
2864
2865 ssize_t ll_copy_user_md(const struct lov_user_md __user *md,
2866                         struct lov_user_md **kbuf)
2867 {
2868         struct lov_user_md      lum;
2869         ssize_t                 lum_size;
2870         ENTRY;
2871
2872         if (copy_from_user(&lum, md, sizeof(lum)))
2873                 RETURN(-EFAULT);
2874
2875         lum_size = ll_lov_user_md_size(&lum);
2876         if (lum_size < 0)
2877                 RETURN(lum_size);
2878
2879         OBD_ALLOC(*kbuf, lum_size);
2880         if (*kbuf == NULL)
2881                 RETURN(-ENOMEM);
2882
2883         if (copy_from_user(*kbuf, md, lum_size) != 0) {
2884                 OBD_FREE(*kbuf, lum_size);
2885                 RETURN(-EFAULT);
2886         }
2887
2888         RETURN(lum_size);
2889 }
2890
2891 /*
2892  * Compute llite root squash state after a change of root squash
2893  * configuration setting or add/remove of a lnet nid
2894  */
2895 void ll_compute_rootsquash_state(struct ll_sb_info *sbi)
2896 {
2897         struct root_squash_info *squash = &sbi->ll_squash;
2898         int i;
2899         bool matched;
2900         struct lnet_process_id id;
2901
2902         /* Update norootsquash flag */
2903         spin_lock(&squash->rsi_lock);
2904         if (list_empty(&squash->rsi_nosquash_nids))
2905                 sbi->ll_flags &= ~LL_SBI_NOROOTSQUASH;
2906         else {
2907                 /* Do not apply root squash as soon as one of our NIDs is
2908                  * in the nosquash_nids list */
2909                 matched = false;
2910                 i = 0;
2911                 while (LNetGetId(i++, &id) != -ENOENT) {
2912                         if (LNET_NETTYP(LNET_NIDNET(id.nid)) == LOLND)
2913                                 continue;
2914                         if (cfs_match_nid(id.nid, &squash->rsi_nosquash_nids)) {
2915                                 matched = true;
2916                                 break;
2917                         }
2918                 }
2919                 if (matched)
2920                         sbi->ll_flags |= LL_SBI_NOROOTSQUASH;
2921                 else
2922                         sbi->ll_flags &= ~LL_SBI_NOROOTSQUASH;
2923         }
2924         spin_unlock(&squash->rsi_lock);
2925 }
2926
2927 /**
2928  * Parse linkea content to extract information about a given hardlink
2929  *
2930  * \param[in]   ldata      - Initialized linkea data
2931  * \param[in]   linkno     - Link identifier
2932  * \param[out]  parent_fid - The entry's parent FID
2933  * \param[out]  ln         - Entry name destination buffer
2934  *
2935  * \retval 0 on success
2936  * \retval Appropriate negative error code on failure
2937  */
2938 static int ll_linkea_decode(struct linkea_data *ldata, unsigned int linkno,
2939                             struct lu_fid *parent_fid, struct lu_name *ln)
2940 {
2941         unsigned int    idx;
2942         int             rc;
2943         ENTRY;
2944
2945         rc = linkea_init_with_rec(ldata);
2946         if (rc < 0)
2947                 RETURN(rc);
2948
2949         if (linkno >= ldata->ld_leh->leh_reccount)
2950                 /* beyond last link */
2951                 RETURN(-ENODATA);
2952
2953         linkea_first_entry(ldata);
2954         for (idx = 0; ldata->ld_lee != NULL; idx++) {
2955                 linkea_entry_unpack(ldata->ld_lee, &ldata->ld_reclen, ln,
2956                                     parent_fid);
2957                 if (idx == linkno)
2958                         break;
2959
2960                 linkea_next_entry(ldata);
2961         }
2962
2963         if (idx < linkno)
2964                 RETURN(-ENODATA);
2965
2966         RETURN(0);
2967 }
2968
2969 /**
2970  * Get parent FID and name of an identified link. Operation is performed for
2971  * a given link number, letting the caller iterate over linkno to list one or
2972  * all links of an entry.
2973  *
2974  * \param[in]     file - File descriptor against which to perform the operation
2975  * \param[in,out] arg  - User-filled structure containing the linkno to operate
2976  *                       on and the available size. It is eventually filled with
2977  *                       the requested information or left untouched on error
2978  *
2979  * \retval - 0 on success
2980  * \retval - Appropriate negative error code on failure
2981  */
2982 int ll_getparent(struct file *file, struct getparent __user *arg)
2983 {
2984         struct inode            *inode = file_inode(file);
2985         struct linkea_data      *ldata;
2986         struct lu_buf            buf = LU_BUF_NULL;
2987         struct lu_name           ln;
2988         struct lu_fid            parent_fid;
2989         __u32                    linkno;
2990         __u32                    name_size;
2991         int                      rc;
2992
2993         ENTRY;
2994
2995         if (!cfs_capable(CFS_CAP_DAC_READ_SEARCH) &&
2996             !(ll_i2sbi(inode)->ll_flags & LL_SBI_USER_FID2PATH))
2997                 RETURN(-EPERM);
2998
2999         if (get_user(name_size, &arg->gp_name_size))
3000                 RETURN(-EFAULT);
3001
3002         if (get_user(linkno, &arg->gp_linkno))
3003                 RETURN(-EFAULT);
3004
3005         if (name_size > PATH_MAX)
3006                 RETURN(-EINVAL);
3007
3008         OBD_ALLOC(ldata, sizeof(*ldata));
3009         if (ldata == NULL)
3010                 RETURN(-ENOMEM);
3011
3012         rc = linkea_data_new(ldata, &buf);
3013         if (rc < 0)
3014                 GOTO(ldata_free, rc);
3015
3016 #ifdef HAVE_XATTR_HANDLER_FLAGS
3017         rc = ll_xattr_list(inode, XATTR_NAME_LINK, XATTR_TRUSTED_T, buf.lb_buf,
3018                            buf.lb_len, OBD_MD_FLXATTR);
3019 #else
3020         rc = ll_getxattr(file_dentry(file), XATTR_NAME_LINK, buf.lb_buf,
3021                          buf.lb_len);
3022 #endif /* HAVE_XATTR_HANDLER_FLAGS */
3023         if (rc < 0)
3024                 GOTO(lb_free, rc);
3025
3026         rc = ll_linkea_decode(ldata, linkno, &parent_fid, &ln);
3027         if (rc < 0)
3028                 GOTO(lb_free, rc);
3029
3030         if (ln.ln_namelen >= name_size)
3031                 GOTO(lb_free, rc = -EOVERFLOW);
3032
3033         if (copy_to_user(&arg->gp_fid, &parent_fid, sizeof(arg->gp_fid)))
3034                 GOTO(lb_free, rc = -EFAULT);
3035
3036         if (copy_to_user(&arg->gp_name, ln.ln_name, ln.ln_namelen))
3037                 GOTO(lb_free, rc = -EFAULT);
3038
3039         if (put_user('\0', arg->gp_name + ln.ln_namelen))
3040                 GOTO(lb_free, rc = -EFAULT);
3041
3042 lb_free:
3043         lu_buf_free(&buf);
3044 ldata_free:
3045         OBD_FREE(ldata, sizeof(*ldata));
3046
3047         RETURN(rc);
3048 }