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 i;
  77         ENTRY;
  78
  79         OBD_ALLOC_PTR(sbi);
  80         if (sbi == NULL)
  81                 RETURN(NULL);
  82
  83         spin_lock_init(&sbi->ll_lock);
  84         mutex_init(&sbi->ll_lco.lco_lock);
  85         spin_lock_init(&sbi->ll_pp_extent_lock);
  86         spin_lock_init(&sbi->ll_process_lock);
  87         sbi->ll_rw_stats_on = 0;
  88
  89         si_meminfo(&si);
  90         pages = si.totalram - si.totalhigh;
  91         lru_page_max = pages / 2;
  92
  93         /* initialize ll_cache data */
  94         sbi->ll_cache = cl_cache_init(lru_page_max);
  95         if (sbi->ll_cache == NULL) {
  96                 OBD_FREE(sbi, sizeof(*sbi));
  97                 RETURN(NULL);
  98         }
  99
 100         sbi->ll_ra_info.ra_max_pages_per_file = min(pages / 32,
 101                                            SBI_DEFAULT_READAHEAD_MAX);
 102         sbi->ll_ra_info.ra_max_pages = sbi->ll_ra_info.ra_max_pages_per_file;
 103         sbi->ll_ra_info.ra_max_read_ahead_whole_pages = -1;
 104
 105         sbi->ll_flags |= LL_SBI_VERBOSE;
 106 #ifdef ENABLE_CHECKSUM
 107         sbi->ll_flags |= LL_SBI_CHECKSUM;
 108 #endif
 109 #ifdef ENABLE_FLOCK
 110         sbi->ll_flags |= LL_SBI_FLOCK;
 111 #endif
 112
 113 #ifdef HAVE_LRU_RESIZE_SUPPORT
 114         sbi->ll_flags |= LL_SBI_LRU_RESIZE;
 115 #endif
 116         sbi->ll_flags |= LL_SBI_LAZYSTATFS;
 117
 118         for (i = 0; i <= LL_PROCESS_HIST_MAX; i++) {
 119                 spin_lock_init(&sbi->ll_rw_extents_info.pp_extents[i].
 120                                pp_r_hist.oh_lock);
 121                 spin_lock_init(&sbi->ll_rw_extents_info.pp_extents[i].
 122                                pp_w_hist.oh_lock);
 123         }
 124
 125         /* metadata statahead is enabled by default */
 126         sbi->ll_sa_running_max = LL_SA_RUNNING_DEF;
 127         sbi->ll_sa_max = LL_SA_RPC_DEF;
 128         atomic_set(&sbi->ll_sa_total, 0);
 129         atomic_set(&sbi->ll_sa_wrong, 0);
 130         atomic_set(&sbi->ll_sa_running, 0);
 131         atomic_set(&sbi->ll_agl_total, 0);
 132         sbi->ll_flags |= LL_SBI_AGL_ENABLED;
 133         sbi->ll_flags |= LL_SBI_FAST_READ;
 134         sbi->ll_flags |= LL_SBI_TINY_WRITE;
 135
 136         /* root squash */
 137         sbi->ll_squash.rsi_uid = 0;
 138         sbi->ll_squash.rsi_gid = 0;
 139         INIT_LIST_HEAD(&sbi->ll_squash.rsi_nosquash_nids);
 140         init_rwsem(&sbi->ll_squash.rsi_sem);
 141
 142         /* Per-filesystem file heat */
 143         sbi->ll_heat_decay_weight = SBI_DEFAULT_HEAT_DECAY_WEIGHT;
 144         sbi->ll_heat_period_second = SBI_DEFAULT_HEAT_PERIOD_SECOND;
 145         RETURN(sbi);
 146 }
 147
 148 static void ll_free_sbi(struct super_block *sb)
 149 {
 150         struct ll_sb_info *sbi = ll_s2sbi(sb);
 151         ENTRY;
 152
 153         if (sbi != NULL) {
 154                 if (!list_empty(&sbi->ll_squash.rsi_nosquash_nids))
 155                         cfs_free_nidlist(&sbi->ll_squash.rsi_nosquash_nids);
 156                 if (sbi->ll_cache != NULL) {
 157                         cl_cache_decref(sbi->ll_cache);
 158                         sbi->ll_cache = NULL;
 159                 }
 160                 OBD_FREE(sbi, sizeof(*sbi));
 161         }
 162         EXIT;
 163 }
 164
 165 static int client_common_fill_super(struct super_block *sb, char *md, char *dt,
 166                                     struct vfsmount *mnt)
 167 {
 168         struct inode *root = NULL;
 169         struct ll_sb_info *sbi = ll_s2sbi(sb);
 170         struct obd_statfs *osfs = NULL;
 171         struct ptlrpc_request *request = NULL;
 172         struct obd_connect_data *data = NULL;
 173         struct obd_uuid *uuid;
 174         struct md_op_data *op_data;
 175         struct lustre_md lmd;
 176         u64 valid;
 177         int size, err, checksum;
 178
 179         ENTRY;
 180         sbi->ll_md_obd = class_name2obd(md);
 181         if (!sbi->ll_md_obd) {
 182                 CERROR("MD %s: not setup or attached\n", md);
 183                 RETURN(-EINVAL);
 184         }
 185
 186         OBD_ALLOC_PTR(data);
 187         if (data == NULL)
 188                 RETURN(-ENOMEM);
 189
 190         OBD_ALLOC_PTR(osfs);
 191         if (osfs == NULL) {
 192                 OBD_FREE_PTR(data);
 193                 RETURN(-ENOMEM);
 194         }
 195
 196         /* pass client page size via ocd_grant_blkbits, the server should report
 197          * back its backend blocksize for grant calculation purpose */
 198         data->ocd_grant_blkbits = PAGE_SHIFT;
 199
 200         /* indicate MDT features supported by this client */
 201         data->ocd_connect_flags = OBD_CONNECT_IBITS    | OBD_CONNECT_NODEVOH  |
 202                                   OBD_CONNECT_ATTRFID  | OBD_CONNECT_GRANT |
 203                                   OBD_CONNECT_VERSION  | OBD_CONNECT_BRW_SIZE |
 204                                   OBD_CONNECT_SRVLOCK  | OBD_CONNECT_TRUNCLOCK|
 205                                   OBD_CONNECT_MDS_CAPA | OBD_CONNECT_OSS_CAPA |
 206                                   OBD_CONNECT_CANCELSET | OBD_CONNECT_FID     |
 207                                   OBD_CONNECT_AT       | OBD_CONNECT_LOV_V3   |
 208                                   OBD_CONNECT_VBR | OBD_CONNECT_FULL20 |
 209                                   OBD_CONNECT_64BITHASH |
 210                                   OBD_CONNECT_EINPROGRESS |
 211                                   OBD_CONNECT_JOBSTATS | OBD_CONNECT_LVB_TYPE |
 212                                   OBD_CONNECT_LAYOUTLOCK | OBD_CONNECT_PINGLESS|
 213                                   OBD_CONNECT_MAX_EASIZE |
 214                                   OBD_CONNECT_FLOCK_DEAD |
 215                                   OBD_CONNECT_DISP_STRIPE | OBD_CONNECT_LFSCK |
 216                                   OBD_CONNECT_OPEN_BY_FID |
 217                                   OBD_CONNECT_DIR_STRIPE |
 218                                   OBD_CONNECT_BULK_MBITS | OBD_CONNECT_CKSUM |
 219                                   OBD_CONNECT_SUBTREE |
 220                                   OBD_CONNECT_MULTIMODRPCS |
 221                                   OBD_CONNECT_GRANT_PARAM |
 222                                   OBD_CONNECT_SHORTIO | OBD_CONNECT_FLAGS2;
 223
 224         data->ocd_connect_flags2 = OBD_CONNECT2_DIR_MIGRATE |
 225                                    OBD_CONNECT2_SUM_STATFS |
 226                                    OBD_CONNECT2_OVERSTRIPING |
 227                                    OBD_CONNECT2_FLR |
 228                                    OBD_CONNECT2_LOCK_CONVERT |
 229                                    OBD_CONNECT2_ARCHIVE_ID_ARRAY |
 230                                    OBD_CONNECT2_LSOM |
 231                                    OBD_CONNECT2_ASYNC_DISCARD;
 232
 233 #ifdef HAVE_LRU_RESIZE_SUPPORT
 234         if (sbi->ll_flags & LL_SBI_LRU_RESIZE)
 235                 data->ocd_connect_flags |= OBD_CONNECT_LRU_RESIZE;
 236 #endif
 237 #ifdef CONFIG_FS_POSIX_ACL
 238         data->ocd_connect_flags |= OBD_CONNECT_ACL | OBD_CONNECT_UMASK |
 239                                    OBD_CONNECT_LARGE_ACL;
 240 #endif
 241
 242         data->ocd_cksum_types = obd_cksum_types_supported_client();
 243
 244         if (OBD_FAIL_CHECK(OBD_FAIL_MDC_LIGHTWEIGHT))
 245                 /* flag mdc connection as lightweight, only used for test
 246                  * purpose, use with care */
 247                 data->ocd_connect_flags |= OBD_CONNECT_LIGHTWEIGHT;
 248
 249         data->ocd_ibits_known = MDS_INODELOCK_FULL;
 250         data->ocd_version = LUSTRE_VERSION_CODE;
 251
 252         if (sb->s_flags & MS_RDONLY)
 253                 data->ocd_connect_flags |= OBD_CONNECT_RDONLY;
 254         if (sbi->ll_flags & LL_SBI_USER_XATTR)
 255                 data->ocd_connect_flags |= OBD_CONNECT_XATTR;
 256
 257 #ifdef MS_NOSEC
 258         /* Setting this indicates we correctly support S_NOSEC (See kernel
 259          * commit 9e1f1de02c2275d7172e18dc4e7c2065777611bf)
 260          */
 261         sb->s_flags |= MS_NOSEC;
 262 #endif
 263
 264         if (sbi->ll_flags & LL_SBI_FLOCK)
 265                 sbi->ll_fop = &ll_file_operations_flock;
 266         else if (sbi->ll_flags & LL_SBI_LOCALFLOCK)
 267                 sbi->ll_fop = &ll_file_operations;
 268         else
 269                 sbi->ll_fop = &ll_file_operations_noflock;
 270
 271         /* always ping even if server suppress_pings */
 272         if (sbi->ll_flags & LL_SBI_ALWAYS_PING)
 273                 data->ocd_connect_flags &= ~OBD_CONNECT_PINGLESS;
 274
 275         obd_connect_set_secctx(data);
 276
 277 #if defined(CONFIG_SECURITY)
 278         data->ocd_connect_flags2 |= OBD_CONNECT2_SELINUX_POLICY;
 279 #endif
 280
 281         data->ocd_brw_size = MD_MAX_BRW_SIZE;
 282
 283         err = obd_connect(NULL, &sbi->ll_md_exp, sbi->ll_md_obd,
 284                           &sbi->ll_sb_uuid, data, sbi->ll_cache);
 285         if (err == -EBUSY) {
 286                 LCONSOLE_ERROR_MSG(0x14f, "An MDT (md %s) is performing "
 287                                    "recovery, of which this client is not a "
 288                                    "part. Please wait for recovery to complete,"
 289                                    " abort, or time out.\n", md);
 290                 GOTO(out, err);
 291         } else if (err) {
 292                 CERROR("cannot connect to %s: rc = %d\n", md, err);
 293                 GOTO(out, err);
 294         }
 295
 296         sbi->ll_md_exp->exp_connect_data = *data;
 297
 298         err = obd_fid_init(sbi->ll_md_exp->exp_obd, sbi->ll_md_exp,
 299                            LUSTRE_SEQ_METADATA);
 300         if (err) {
 301                 CERROR("%s: Can't init metadata layer FID infrastructure, "
 302                        "rc = %d\n", sbi->ll_md_exp->exp_obd->obd_name, err);
 303                 GOTO(out_md, err);
 304         }
 305
 306         /* For mount, we only need fs info from MDT0, and also in DNE, it
 307          * can make sure the client can be mounted as long as MDT0 is
 308          * avaible */
 309         err = obd_statfs(NULL, sbi->ll_md_exp, osfs,
 310                         ktime_get_seconds() - OBD_STATFS_CACHE_SECONDS,
 311                         OBD_STATFS_FOR_MDT0);
 312         if (err)
 313                 GOTO(out_md_fid, err);
 314
 315         /* This needs to be after statfs to ensure connect has finished.
 316          * Note that "data" does NOT contain the valid connect reply.
 317          * If connecting to a 1.8 server there will be no LMV device, so
 318          * we can access the MDC export directly and exp_connect_flags will
 319          * be non-zero, but if accessing an upgraded 2.1 server it will
 320          * have the correct flags filled in.
 321          * XXX: fill in the LMV exp_connect_flags from MDC(s). */
 322         valid = exp_connect_flags(sbi->ll_md_exp) & CLIENT_CONNECT_MDT_REQD;
 323         if (exp_connect_flags(sbi->ll_md_exp) != 0 &&
 324             valid != CLIENT_CONNECT_MDT_REQD) {
 325                 char *buf;
 326
 327                 OBD_ALLOC_WAIT(buf, PAGE_SIZE);
 328                 obd_connect_flags2str(buf, PAGE_SIZE,
 329                                       valid ^ CLIENT_CONNECT_MDT_REQD, 0, ",");
 330                 LCONSOLE_ERROR_MSG(0x170, "Server %s does not support "
 331                                    "feature(s) needed for correct operation "
 332                                    "of this client (%s). Please upgrade "
 333                                    "server or downgrade client.\n",
 334                                    sbi->ll_md_exp->exp_obd->obd_name, buf);
 335                 OBD_FREE(buf, PAGE_SIZE);
 336                 GOTO(out_md_fid, err = -EPROTO);
 337         }
 338
 339         size = sizeof(*data);
 340         err = obd_get_info(NULL, sbi->ll_md_exp, sizeof(KEY_CONN_DATA),
 341                            KEY_CONN_DATA,  &size, data);
 342         if (err) {
 343                 CERROR("%s: Get connect data failed: rc = %d\n",
 344                        sbi->ll_md_exp->exp_obd->obd_name, err);
 345                 GOTO(out_md_fid, err);
 346         }
 347
 348         LASSERT(osfs->os_bsize);
 349         sb->s_blocksize = osfs->os_bsize;
 350         sb->s_blocksize_bits = log2(osfs->os_bsize);
 351         sb->s_magic = LL_SUPER_MAGIC;
 352         sb->s_maxbytes = MAX_LFS_FILESIZE;
 353         sbi->ll_namelen = osfs->os_namelen;
 354         sbi->ll_mnt.mnt = current->fs->root.mnt;
 355
 356         if ((sbi->ll_flags & LL_SBI_USER_XATTR) &&
 357             !(data->ocd_connect_flags & OBD_CONNECT_XATTR)) {
 358                 LCONSOLE_INFO("Disabling user_xattr feature because "
 359                               "it is not supported on the server\n");
 360                 sbi->ll_flags &= ~LL_SBI_USER_XATTR;
 361         }
 362
 363         if (data->ocd_connect_flags & OBD_CONNECT_ACL) {
 364 #ifdef MS_POSIXACL
 365                 sb->s_flags |= MS_POSIXACL;
 366 #endif
 367                 sbi->ll_flags |= LL_SBI_ACL;
 368         } else {
 369                 LCONSOLE_INFO("client wants to enable acl, but mdt not!\n");
 370 #ifdef MS_POSIXACL
 371                 sb->s_flags &= ~MS_POSIXACL;
 372 #endif
 373                 sbi->ll_flags &= ~LL_SBI_ACL;
 374         }
 375
 376         if (data->ocd_connect_flags & OBD_CONNECT_64BITHASH)
 377                 sbi->ll_flags |= LL_SBI_64BIT_HASH;
 378
 379         if (data->ocd_connect_flags & OBD_CONNECT_LAYOUTLOCK)
 380                 sbi->ll_flags |= LL_SBI_LAYOUT_LOCK;
 381
 382         if (obd_connect_has_secctx(data))
 383                 sbi->ll_flags |= LL_SBI_FILE_SECCTX;
 384
 385         if (data->ocd_ibits_known & MDS_INODELOCK_XATTR) {
 386                 if (!(data->ocd_connect_flags & OBD_CONNECT_MAX_EASIZE)) {
 387                         LCONSOLE_INFO("%s: disabling xattr cache due to "
 388                                       "unknown maximum xattr size.\n", dt);
 389                 } else if (!sbi->ll_xattr_cache_set) {
 390                         /* If xattr_cache is already set (no matter 0 or 1)
 391                          * during processing llog, it won't be enabled here. */
 392                         sbi->ll_flags |= LL_SBI_XATTR_CACHE;
 393                         sbi->ll_xattr_cache_enabled = 1;
 394                 }
 395         }
 396
 397         sbi->ll_dt_obd = class_name2obd(dt);
 398         if (!sbi->ll_dt_obd) {
 399                 CERROR("DT %s: not setup or attached\n", dt);
 400                 GOTO(out_md_fid, err = -ENODEV);
 401         }
 402
 403         /* pass client page size via ocd_grant_blkbits, the server should report
 404          * back its backend blocksize for grant calculation purpose */
 405         data->ocd_grant_blkbits = PAGE_SHIFT;
 406
 407         /* indicate OST features supported by this client */
 408         data->ocd_connect_flags = OBD_CONNECT_GRANT | OBD_CONNECT_VERSION |
 409                                   OBD_CONNECT_REQPORTAL | OBD_CONNECT_BRW_SIZE |
 410                                   OBD_CONNECT_CANCELSET | OBD_CONNECT_FID |
 411                                   OBD_CONNECT_SRVLOCK | OBD_CONNECT_TRUNCLOCK|
 412                                   OBD_CONNECT_AT | OBD_CONNECT_OSS_CAPA |
 413                                   OBD_CONNECT_VBR | OBD_CONNECT_FULL20 |
 414                                   OBD_CONNECT_64BITHASH | OBD_CONNECT_MAXBYTES |
 415                                   OBD_CONNECT_EINPROGRESS |
 416                                   OBD_CONNECT_JOBSTATS | OBD_CONNECT_LVB_TYPE |
 417                                   OBD_CONNECT_LAYOUTLOCK |
 418                                   OBD_CONNECT_PINGLESS | OBD_CONNECT_LFSCK |
 419                                   OBD_CONNECT_BULK_MBITS | OBD_CONNECT_SHORTIO |
 420                                   OBD_CONNECT_FLAGS2 | OBD_CONNECT_GRANT_SHRINK;
 421
 422 /* The client currently advertises support for OBD_CONNECT_LOCKAHEAD_OLD so it
 423  * can interoperate with an older version of lockahead which was released prior
 424  * to landing in master. This support will be dropped when 2.13 development
 425  * starts.  At the point, we should not just drop the connect flag (below), we
 426  * should also remove the support in the code.
 427  *
 428  * Removing it means a few things:
 429  * 1. Remove this section here
 430  * 2. Remove CEF_NONBLOCK in ll_file_lockahead()
 431  * 3. Remove function exp_connect_lockahead_old
 432  * 4. Remove LDLM_FL_LOCKAHEAD_OLD_RESERVED in lustre_dlm_flags.h
 433  * */
 434 #if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(2, 12, 50, 0)
 435         data->ocd_connect_flags |= OBD_CONNECT_LOCKAHEAD_OLD;
 436 #endif
 437
 438         data->ocd_connect_flags2 = OBD_CONNECT2_LOCKAHEAD;
 439
 440         if (!OBD_FAIL_CHECK(OBD_FAIL_OSC_CONNECT_GRANT_PARAM))
 441                 data->ocd_connect_flags |= OBD_CONNECT_GRANT_PARAM;
 442
 443         /* OBD_CONNECT_CKSUM should always be set, even if checksums are
 444          * disabled by default, because it can still be enabled on the
 445          * fly via /sys. As a consequence, we still need to come to an
 446          * agreement on the supported algorithms at connect time
 447          */
 448         data->ocd_connect_flags |= OBD_CONNECT_CKSUM;
 449
 450         if (OBD_FAIL_CHECK(OBD_FAIL_OSC_CKSUM_ADLER_ONLY))
 451                 data->ocd_cksum_types = OBD_CKSUM_ADLER;
 452         else
 453                 data->ocd_cksum_types = obd_cksum_types_supported_client();
 454
 455 #ifdef HAVE_LRU_RESIZE_SUPPORT
 456         data->ocd_connect_flags |= OBD_CONNECT_LRU_RESIZE;
 457 #endif
 458         /* always ping even if server suppress_pings */
 459         if (sbi->ll_flags & LL_SBI_ALWAYS_PING)
 460                 data->ocd_connect_flags &= ~OBD_CONNECT_PINGLESS;
 461
 462         CDEBUG(D_RPCTRACE, "ocd_connect_flags: %#llx ocd_version: %d "
 463                "ocd_grant: %d\n", data->ocd_connect_flags,
 464                data->ocd_version, data->ocd_grant);
 465
 466         sbi->ll_dt_obd->obd_upcall.onu_owner = &sbi->ll_lco;
 467         sbi->ll_dt_obd->obd_upcall.onu_upcall = cl_ocd_update;
 468
 469         data->ocd_brw_size = DT_MAX_BRW_SIZE;
 470
 471         err = obd_connect(NULL, &sbi->ll_dt_exp, sbi->ll_dt_obd,
 472                           &sbi->ll_sb_uuid, data, sbi->ll_cache);
 473         if (err == -EBUSY) {
 474                 LCONSOLE_ERROR_MSG(0x150, "An OST (dt %s) is performing "
 475                                    "recovery, of which this client is not a "
 476                                    "part.  Please wait for recovery to "
 477                                    "complete, abort, or time out.\n", dt);
 478                 GOTO(out_md, err);
 479         } else if (err) {
 480                 CERROR("%s: Cannot connect to %s: rc = %d\n",
 481                        sbi->ll_dt_exp->exp_obd->obd_name, dt, err);
 482                 GOTO(out_md, err);
 483         }
 484
 485         sbi->ll_dt_exp->exp_connect_data = *data;
 486
 487         /* Don't change value if it was specified in the config log */
 488         if (sbi->ll_ra_info.ra_max_read_ahead_whole_pages == -1)
 489                 sbi->ll_ra_info.ra_max_read_ahead_whole_pages =
 490                         max_t(unsigned long, SBI_DEFAULT_READAHEAD_WHOLE_MAX,
 491                               (data->ocd_brw_size >> PAGE_SHIFT));
 492
 493         err = obd_fid_init(sbi->ll_dt_exp->exp_obd, sbi->ll_dt_exp,
 494                            LUSTRE_SEQ_METADATA);
 495         if (err) {
 496                 CERROR("%s: Can't init data layer FID infrastructure, "
 497                        "rc = %d\n", sbi->ll_dt_exp->exp_obd->obd_name, err);
 498                 GOTO(out_dt, err);
 499         }
 500
 501         mutex_lock(&sbi->ll_lco.lco_lock);
 502         sbi->ll_lco.lco_flags = data->ocd_connect_flags;
 503         sbi->ll_lco.lco_md_exp = sbi->ll_md_exp;
 504         sbi->ll_lco.lco_dt_exp = sbi->ll_dt_exp;
 505         mutex_unlock(&sbi->ll_lco.lco_lock);
 506
 507         fid_zero(&sbi->ll_root_fid);
 508         err = md_get_root(sbi->ll_md_exp, get_mount_fileset(sb),
 509                            &sbi->ll_root_fid);
 510         if (err) {
 511                 CERROR("cannot mds_connect: rc = %d\n", err);
 512                 GOTO(out_lock_cn_cb, err);
 513         }
 514         if (!fid_is_sane(&sbi->ll_root_fid)) {
 515                 CERROR("%s: Invalid root fid "DFID" during mount\n",
 516                        sbi->ll_md_exp->exp_obd->obd_name,
 517                        PFID(&sbi->ll_root_fid));
 518                 GOTO(out_lock_cn_cb, err = -EINVAL);
 519         }
 520         CDEBUG(D_SUPER, "rootfid "DFID"\n", PFID(&sbi->ll_root_fid));
 521
 522         sb->s_op = &lustre_super_operations;
 523 #ifdef HAVE_XATTR_HANDLER_FLAGS
 524         sb->s_xattr = ll_xattr_handlers;
 525 #endif
 526 #if THREAD_SIZE >= 8192 /*b=17630*/
 527         sb->s_export_op = &lustre_export_operations;
 528 #endif
 529
 530         /* make root inode
 531          * XXX: move this to after cbd setup? */
 532         valid = OBD_MD_FLGETATTR | OBD_MD_FLBLOCKS | OBD_MD_FLMODEASIZE;
 533         if (sbi->ll_flags & LL_SBI_ACL)
 534                 valid |= OBD_MD_FLACL;
 535
 536         OBD_ALLOC_PTR(op_data);
 537         if (op_data == NULL)
 538                 GOTO(out_lock_cn_cb, err = -ENOMEM);
 539
 540         op_data->op_fid1 = sbi->ll_root_fid;
 541         op_data->op_mode = 0;
 542         op_data->op_valid = valid;
 543
 544         err = md_getattr(sbi->ll_md_exp, op_data, &request);
 545
 546         OBD_FREE_PTR(op_data);
 547         if (err) {
 548                 CERROR("%s: md_getattr failed for root: rc = %d\n",
 549                        sbi->ll_md_exp->exp_obd->obd_name, err);
 550                 GOTO(out_lock_cn_cb, err);
 551         }
 552
 553         err = md_get_lustre_md(sbi->ll_md_exp, request, sbi->ll_dt_exp,
 554                                sbi->ll_md_exp, &lmd);
 555         if (err) {
 556                 CERROR("failed to understand root inode md: rc = %d\n", err);
 557                 ptlrpc_req_finished(request);
 558                 GOTO(out_lock_cn_cb, err);
 559         }
 560
 561         LASSERT(fid_is_sane(&sbi->ll_root_fid));
 562         root = ll_iget(sb, cl_fid_build_ino(&sbi->ll_root_fid,
 563                                             sbi->ll_flags & LL_SBI_32BIT_API),
 564                        &lmd);
 565         md_free_lustre_md(sbi->ll_md_exp, &lmd);
 566         ptlrpc_req_finished(request);
 567
 568         if (IS_ERR(root)) {
 569 #ifdef CONFIG_FS_POSIX_ACL
 570                 if (lmd.posix_acl) {
 571                         posix_acl_release(lmd.posix_acl);
 572                         lmd.posix_acl = NULL;
 573                 }
 574 #endif
 575                 err = IS_ERR(root) ? PTR_ERR(root) : -EBADF;
 576                 root = NULL;
 577                 CERROR("lustre_lite: bad iget4 for root\n");
 578                 GOTO(out_root, err);
 579         }
 580
 581         checksum = sbi->ll_flags & LL_SBI_CHECKSUM;
 582         if (sbi->ll_checksum_set) {
 583                 err = obd_set_info_async(NULL, sbi->ll_dt_exp,
 584                                          sizeof(KEY_CHECKSUM), KEY_CHECKSUM,
 585                                          sizeof(checksum), &checksum, NULL);
 586                 if (err) {
 587                         CERROR("%s: Set checksum failed: rc = %d\n",
 588                                sbi->ll_dt_exp->exp_obd->obd_name, err);
 589                         GOTO(out_root, err);
 590                 }
 591         }
 592         cl_sb_init(sb);
 593
 594         sb->s_root = d_make_root(root);
 595         if (sb->s_root == NULL) {
 596                 err = -ENOMEM;
 597                 CERROR("%s: can't make root dentry: rc = %d\n",
 598                        sbi->ll_fsname, err);
 599                 GOTO(out_root, err);
 600         }
 601 #ifdef HAVE_DCACHE_LOCK
 602         sb->s_root->d_op = &ll_d_ops;
 603 #endif
 604
 605         sbi->ll_sdev_orig = sb->s_dev;
 606
 607         /* We set sb->s_dev equal on all lustre clients in order to support
 608          * NFS export clustering.  NFSD requires that the FSID be the same
 609          * on all clients. */
 610         /* s_dev is also used in lt_compare() to compare two fs, but that is
 611          * only a node-local comparison. */
 612         uuid = obd_get_uuid(sbi->ll_md_exp);
 613         if (uuid != NULL)
 614                 sb->s_dev = get_uuid2int(uuid->uuid, strlen(uuid->uuid));
 615
 616         if (data != NULL)
 617                 OBD_FREE_PTR(data);
 618         if (osfs != NULL)
 619                 OBD_FREE_PTR(osfs);
 620
 621         if (sbi->ll_dt_obd) {
 622                 err = sysfs_create_link(&sbi->ll_kset.kobj,
 623                                         &sbi->ll_dt_obd->obd_kset.kobj,
 624                                         sbi->ll_dt_obd->obd_type->typ_name);
 625                 if (err < 0) {
 626                         CERROR("%s: could not register %s in llite: rc = %d\n",
 627                                dt, sbi->ll_fsname, err);
 628                         err = 0;
 629                 }
 630         }
 631
 632         if (sbi->ll_md_obd) {
 633                 err = sysfs_create_link(&sbi->ll_kset.kobj,
 634                                         &sbi->ll_md_obd->obd_kset.kobj,
 635                                         sbi->ll_md_obd->obd_type->typ_name);
 636                 if (err < 0) {
 637                         CERROR("%s: could not register %s in llite: rc = %d\n",
 638                                md, sbi->ll_fsname, err);
 639                         err = 0;
 640                 }
 641         }
 642
 643         RETURN(err);
 644 out_root:
 645         if (root)
 646                 iput(root);
 647 out_lock_cn_cb:
 648         obd_fid_fini(sbi->ll_dt_exp->exp_obd);
 649 out_dt:
 650         obd_disconnect(sbi->ll_dt_exp);
 651         sbi->ll_dt_exp = NULL;
 652         sbi->ll_dt_obd = NULL;
 653 out_md_fid:
 654         obd_fid_fini(sbi->ll_md_exp->exp_obd);
 655 out_md:
 656         obd_disconnect(sbi->ll_md_exp);
 657         sbi->ll_md_exp = NULL;
 658         sbi->ll_md_obd = NULL;
 659 out:
 660         if (data != NULL)
 661                 OBD_FREE_PTR(data);
 662         if (osfs != NULL)
 663                 OBD_FREE_PTR(osfs);
 664         return err;
 665 }
 666
 667 int ll_get_max_mdsize(struct ll_sb_info *sbi, int *lmmsize)
 668 {
 669         int size, rc;
 670
 671         size = sizeof(*lmmsize);
 672         rc = obd_get_info(NULL, sbi->ll_dt_exp, sizeof(KEY_MAX_EASIZE),
 673                           KEY_MAX_EASIZE, &size, lmmsize);
 674         if (rc != 0) {
 675                 CERROR("%s: cannot get max LOV EA size: rc = %d\n",
 676                        sbi->ll_dt_exp->exp_obd->obd_name, rc);
 677                 RETURN(rc);
 678         }
 679
 680         CDEBUG(D_INFO, "max LOV ea size: %d\n", *lmmsize);
 681
 682         size = sizeof(int);
 683         rc = obd_get_info(NULL, sbi->ll_md_exp, sizeof(KEY_MAX_EASIZE),
 684                           KEY_MAX_EASIZE, &size, lmmsize);
 685         if (rc)
 686                 CERROR("Get max mdsize error rc %d\n", rc);
 687
 688         CDEBUG(D_INFO, "max LMV ea size: %d\n", *lmmsize);
 689
 690         RETURN(rc);
 691 }
 692
 693 /**
 694  * Get the value of the default_easize parameter.
 695  *
 696  * \see client_obd::cl_default_mds_easize
 697  *
 698  * \param[in] sbi       superblock info for this filesystem
 699  * \param[out] lmmsize  pointer to storage location for value
 700  *
 701  * \retval 0            on success
 702  * \retval negative     negated errno on failure
 703  */
 704 int ll_get_default_mdsize(struct ll_sb_info *sbi, int *lmmsize)
 705 {
 706         int size, rc;
 707
 708         size = sizeof(int);
 709         rc = obd_get_info(NULL, sbi->ll_md_exp, sizeof(KEY_DEFAULT_EASIZE),
 710                          KEY_DEFAULT_EASIZE, &size, lmmsize);
 711         if (rc)
 712                 CERROR("Get default mdsize error rc %d\n", rc);
 713
 714         RETURN(rc);
 715 }
 716
 717 /**
 718  * Set the default_easize parameter to the given value.
 719  *
 720  * \see client_obd::cl_default_mds_easize
 721  *
 722  * \param[in] sbi       superblock info for this filesystem
 723  * \param[in] lmmsize   the size to set
 724  *
 725  * \retval 0            on success
 726  * \retval negative     negated errno on failure
 727  */
 728 int ll_set_default_mdsize(struct ll_sb_info *sbi, int lmmsize)
 729 {
 730         int rc;
 731
 732         if (lmmsize < sizeof(struct lov_mds_md) ||
 733             lmmsize > OBD_MAX_DEFAULT_EA_SIZE)
 734                 return -EINVAL;
 735
 736         rc = obd_set_info_async(NULL, sbi->ll_md_exp,
 737                                 sizeof(KEY_DEFAULT_EASIZE), KEY_DEFAULT_EASIZE,
 738                                 sizeof(int), &lmmsize, NULL);
 739
 740         RETURN(rc);
 741 }
 742
 743 static void client_common_put_super(struct super_block *sb)
 744 {
 745         struct ll_sb_info *sbi = ll_s2sbi(sb);
 746         ENTRY;
 747
 748         cl_sb_fini(sb);
 749
 750         obd_fid_fini(sbi->ll_dt_exp->exp_obd);
 751         obd_disconnect(sbi->ll_dt_exp);
 752         sbi->ll_dt_exp = NULL;
 753
 754         ll_debugfs_unregister_super(sb);
 755
 756         obd_fid_fini(sbi->ll_md_exp->exp_obd);
 757         obd_disconnect(sbi->ll_md_exp);
 758         sbi->ll_md_exp = NULL;
 759
 760         EXIT;
 761 }
 762
 763 void ll_kill_super(struct super_block *sb)
 764 {
 765         struct ll_sb_info *sbi;
 766         ENTRY;
 767
 768         /* not init sb ?*/
 769         if (!(sb->s_flags & MS_ACTIVE))
 770                 return;
 771
 772         sbi = ll_s2sbi(sb);
 773         /* we need restore s_dev from changed for clustred NFS before put_super
 774          * because new kernels have cached s_dev and change sb->s_dev in
 775          * put_super not affected real removing devices */
 776         if (sbi) {
 777                 sb->s_dev = sbi->ll_sdev_orig;
 778                 sbi->ll_umounting = 1;
 779
 780                 /* wait running statahead threads to quit */
 781                 while (atomic_read(&sbi->ll_sa_running) > 0) {
 782                         set_current_state(TASK_UNINTERRUPTIBLE);
 783                         schedule_timeout(msecs_to_jiffies(MSEC_PER_SEC >> 3));
 784                 }
 785         }
 786
 787         EXIT;
 788 }
 789
 790 static inline int ll_set_opt(const char *opt, char *data, int fl)
 791 {
 792         if (strncmp(opt, data, strlen(opt)) != 0)
 793                 return 0;
 794         else
 795                 return fl;
 796 }
 797
 798 /* non-client-specific mount options are parsed in lmd_parse */
 799 static int ll_options(char *options, struct ll_sb_info *sbi)
 800 {
 801         int tmp;
 802         char *s1 = options, *s2;
 803         int *flags = &sbi->ll_flags;
 804         ENTRY;
 805
 806         if (!options)
 807                 RETURN(0);
 808
 809         CDEBUG(D_CONFIG, "Parsing opts %s\n", options);
 810
 811         while (*s1) {
 812                 CDEBUG(D_SUPER, "next opt=%s\n", s1);
 813                 tmp = ll_set_opt("nolock", s1, LL_SBI_NOLCK);
 814                 if (tmp) {
 815                         *flags |= tmp;
 816                         goto next;
 817                 }
 818                 tmp = ll_set_opt("flock", s1, LL_SBI_FLOCK);
 819                 if (tmp) {
 820                         *flags |= tmp;
 821                         goto next;
 822                 }
 823                 tmp = ll_set_opt("localflock", s1, LL_SBI_LOCALFLOCK);
 824                 if (tmp) {
 825                         *flags |= tmp;
 826                         goto next;
 827                 }
 828                 tmp = ll_set_opt("noflock", s1, LL_SBI_FLOCK|LL_SBI_LOCALFLOCK);
 829                 if (tmp) {
 830                         *flags &= ~tmp;
 831                         goto next;
 832                 }
 833                 tmp = ll_set_opt("user_xattr", s1, LL_SBI_USER_XATTR);
 834                 if (tmp) {
 835                         *flags |= tmp;
 836                         goto next;
 837                 }
 838                 tmp = ll_set_opt("nouser_xattr", s1, LL_SBI_USER_XATTR);
 839                 if (tmp) {
 840                         *flags &= ~tmp;
 841                         goto next;
 842                 }
 843                 tmp = ll_set_opt("context", s1, 1);
 844                 if (tmp)
 845                         goto next;
 846                 tmp = ll_set_opt("fscontext", s1, 1);
 847                 if (tmp)
 848                         goto next;
 849                 tmp = ll_set_opt("defcontext", s1, 1);
 850                 if (tmp)
 851                         goto next;
 852                 tmp = ll_set_opt("rootcontext", s1, 1);
 853                 if (tmp)
 854                         goto next;
 855                 tmp = ll_set_opt("user_fid2path", s1, LL_SBI_USER_FID2PATH);
 856                 if (tmp) {
 857                         *flags |= tmp;
 858                         goto next;
 859                 }
 860                 tmp = ll_set_opt("nouser_fid2path", s1, LL_SBI_USER_FID2PATH);
 861                 if (tmp) {
 862                         *flags &= ~tmp;
 863                         goto next;
 864                 }
 865
 866                 tmp = ll_set_opt("checksum", s1, LL_SBI_CHECKSUM);
 867                 if (tmp) {
 868                         *flags |= tmp;
 869                         sbi->ll_checksum_set = 1;
 870                         goto next;
 871                 }
 872                 tmp = ll_set_opt("nochecksum", s1, LL_SBI_CHECKSUM);
 873                 if (tmp) {
 874                         *flags &= ~tmp;
 875                         sbi->ll_checksum_set = 1;
 876                         goto next;
 877                 }
 878                 tmp = ll_set_opt("lruresize", s1, LL_SBI_LRU_RESIZE);
 879                 if (tmp) {
 880                         *flags |= tmp;
 881                         goto next;
 882                 }
 883                 tmp = ll_set_opt("nolruresize", s1, LL_SBI_LRU_RESIZE);
 884                 if (tmp) {
 885                         *flags &= ~tmp;
 886                         goto next;
 887                 }
 888                 tmp = ll_set_opt("lazystatfs", s1, LL_SBI_LAZYSTATFS);
 889                 if (tmp) {
 890                         *flags |= tmp;
 891                         goto next;
 892                 }
 893                 tmp = ll_set_opt("nolazystatfs", s1, LL_SBI_LAZYSTATFS);
 894                 if (tmp) {
 895                         *flags &= ~tmp;
 896                         goto next;
 897                 }
 898                 tmp = ll_set_opt("32bitapi", s1, LL_SBI_32BIT_API);
 899                 if (tmp) {
 900                         *flags |= tmp;
 901                         goto next;
 902                 }
 903                 tmp = ll_set_opt("verbose", s1, LL_SBI_VERBOSE);
 904                 if (tmp) {
 905                         *flags |= tmp;
 906                         goto next;
 907                 }
 908                 tmp = ll_set_opt("noverbose", s1, LL_SBI_VERBOSE);
 909                 if (tmp) {
 910                         *flags &= ~tmp;
 911                         goto next;
 912                 }
 913                 tmp = ll_set_opt("always_ping", s1, LL_SBI_ALWAYS_PING);
 914                 if (tmp) {
 915                         *flags |= tmp;
 916                         goto next;
 917                 }
 918                 LCONSOLE_ERROR_MSG(0x152, "Unknown option '%s', won't mount.\n",
 919                                    s1);
 920                 RETURN(-EINVAL);
 921
 922 next:
 923                 /* Find next opt */
 924                 s2 = strchr(s1, ',');
 925                 if (s2 == NULL)
 926                         break;
 927                 s1 = s2 + 1;
 928         }
 929         RETURN(0);
 930 }
 931
 932 void ll_lli_init(struct ll_inode_info *lli)
 933 {
 934         lli->lli_inode_magic = LLI_INODE_MAGIC;
 935         lli->lli_flags = 0;
 936         spin_lock_init(&lli->lli_lock);
 937         lli->lli_posix_acl = NULL;
 938         /* Do not set lli_fid, it has been initialized already. */
 939         fid_zero(&lli->lli_pfid);
 940         lli->lli_mds_read_och = NULL;
 941         lli->lli_mds_write_och = NULL;
 942         lli->lli_mds_exec_och = NULL;
 943         lli->lli_open_fd_read_count = 0;
 944         lli->lli_open_fd_write_count = 0;
 945         lli->lli_open_fd_exec_count = 0;
 946         mutex_init(&lli->lli_och_mutex);
 947         spin_lock_init(&lli->lli_agl_lock);
 948         spin_lock_init(&lli->lli_layout_lock);
 949         ll_layout_version_set(lli, CL_LAYOUT_GEN_NONE);
 950         lli->lli_clob = NULL;
 951
 952         init_rwsem(&lli->lli_xattrs_list_rwsem);
 953         mutex_init(&lli->lli_xattrs_enq_lock);
 954
 955         LASSERT(lli->lli_vfs_inode.i_mode != 0);
 956         if (S_ISDIR(lli->lli_vfs_inode.i_mode)) {
 957                 mutex_init(&lli->lli_readdir_mutex);
 958                 lli->lli_opendir_key = NULL;
 959                 lli->lli_sai = NULL;
 960                 spin_lock_init(&lli->lli_sa_lock);
 961                 lli->lli_opendir_pid = 0;
 962                 lli->lli_sa_enabled = 0;
 963                 lli->lli_def_stripe_offset = -1;
 964                 init_rwsem(&lli->lli_lsm_sem);
 965         } else {
 966                 mutex_init(&lli->lli_size_mutex);
 967                 lli->lli_symlink_name = NULL;
 968                 init_rwsem(&lli->lli_trunc_sem);
 969                 range_lock_tree_init(&lli->lli_write_tree);
 970                 init_rwsem(&lli->lli_glimpse_sem);
 971                 lli->lli_glimpse_time = ktime_set(0, 0);
 972                 INIT_LIST_HEAD(&lli->lli_agl_list);
 973                 lli->lli_agl_index = 0;
 974                 lli->lli_async_rc = 0;
 975                 spin_lock_init(&lli->lli_heat_lock);
 976                 obd_heat_clear(lli->lli_heat_instances, OBD_HEAT_COUNT);
 977                 lli->lli_heat_flags = 0;
 978         }
 979         mutex_init(&lli->lli_layout_mutex);
 980         memset(lli->lli_jobid, 0, sizeof(lli->lli_jobid));
 981 }
 982
 983 #define MAX_STRING_SIZE 128
 984
 985 #ifndef HAVE_SUPER_SETUP_BDI_NAME
 986
 987 #define LSI_BDI_INITIALIZED     0x00400000
 988
 989 #ifndef HAVE_BDI_CAP_MAP_COPY
 990 # define BDI_CAP_MAP_COPY       0
 991 #endif
 992
 993 static int super_setup_bdi_name(struct super_block *sb, char *fmt, ...)
 994 {
 995         struct  lustre_sb_info *lsi = s2lsi(sb);
 996         char buf[MAX_STRING_SIZE];
 997         va_list args;
 998         int err;
 999
1000         err = bdi_init(&lsi->lsi_bdi);
1001         if (err)
1002                 return err;
1003
1004         lsi->lsi_flags |= LSI_BDI_INITIALIZED;
1005         lsi->lsi_bdi.capabilities = BDI_CAP_MAP_COPY;
1006         lsi->lsi_bdi.name = "lustre";
1007         va_start(args, fmt);
1008         vsnprintf(buf, MAX_STRING_SIZE, fmt, args);
1009         va_end(args);
1010         err = bdi_register(&lsi->lsi_bdi, NULL, "%s", buf);
1011         va_end(args);
1012         if (!err)
1013                 sb->s_bdi = &lsi->lsi_bdi;
1014
1015         return err;
1016 }
1017 #endif /* !HAVE_SUPER_SETUP_BDI_NAME */
1018
1019 int ll_fill_super(struct super_block *sb, struct vfsmount *mnt)
1020 {
1021         struct  lustre_profile *lprof = NULL;
1022         struct  lustre_sb_info *lsi = s2lsi(sb);
1023         struct  ll_sb_info *sbi = NULL;
1024         char    *dt = NULL, *md = NULL;
1025         char    *profilenm = get_profile_name(sb);
1026         struct config_llog_instance *cfg;
1027         /* %p for void* in printf needs 16+2 characters: 0xffffffffffffffff */
1028         const int instlen = 16 + 2;
1029         unsigned long cfg_instance = ll_get_cfg_instance(sb);
1030         char name[MAX_STRING_SIZE];
1031         int md_len = 0;
1032         int dt_len = 0;
1033         uuid_t uuid;
1034         char *ptr;
1035         int len;
1036         int err;
1037
1038         ENTRY;
1039         /* for ASLR, to map between cfg_instance and hashed ptr */
1040         CDEBUG(D_VFSTRACE, "VFS Op: cfg_instance %s-%016lx (sb %p)\n",
1041                profilenm, cfg_instance, sb);
1042
1043         try_module_get(THIS_MODULE);
1044
1045         OBD_ALLOC_PTR(cfg);
1046         if (cfg == NULL)
1047                 GOTO(out_free_cfg, err = -ENOMEM);
1048
1049         /* client additional sb info */
1050         lsi->lsi_llsbi = sbi = ll_init_sbi();
1051         if (!sbi)
1052                 GOTO(out_free_cfg, err = -ENOMEM);
1053
1054         err = ll_options(lsi->lsi_lmd->lmd_opts, sbi);
1055         if (err)
1056                 GOTO(out_free_cfg, err);
1057
1058 #ifndef HAVE_DCACHE_LOCK
1059         /* kernel >= 2.6.38 store dentry operations in sb->s_d_op. */
1060         sb->s_d_op = &ll_d_ops;
1061 #endif
1062         /* UUID handling */
1063         generate_random_uuid(uuid.b);
1064         snprintf(sbi->ll_sb_uuid.uuid, UUID_SIZE, "%pU", uuid.b);
1065
1066         CDEBUG(D_CONFIG, "llite sb uuid: %s\n", sbi->ll_sb_uuid.uuid);
1067
1068         /* Get fsname */
1069         len = strlen(profilenm);
1070         ptr = strrchr(profilenm, '-');
1071         if (ptr && (strcmp(ptr, "-client") == 0))
1072                 len -= 7;
1073
1074         if (len > LUSTRE_MAXFSNAME) {
1075                 if (unlikely(len >= MAX_STRING_SIZE))
1076                         len = MAX_STRING_SIZE - 1;
1077                 strncpy(name, profilenm, len);
1078                 name[len] = '\0';
1079                 err = -ENAMETOOLONG;
1080                 CERROR("%s: fsname longer than %u characters: rc = %d\n",
1081                        name, LUSTRE_MAXFSNAME, err);
1082                 GOTO(out_free_cfg, err);
1083         }
1084         strncpy(sbi->ll_fsname, profilenm, len);
1085         sbi->ll_fsname[len] = '\0';
1086
1087         /* Mount info */
1088         snprintf(name, sizeof(name), "%.*s-%016lx", len,
1089                  profilenm, cfg_instance);
1090
1091         err = super_setup_bdi_name(sb, "%s", name);
1092         if (err)
1093                 GOTO(out_free_cfg, err);
1094
1095         /* Call ll_debugfs_register_super() before lustre_process_log()
1096          * so that "llite.*.*" params can be processed correctly.
1097          */
1098         err = ll_debugfs_register_super(sb, name);
1099         if (err < 0) {
1100                 CERROR("%s: could not register mountpoint in llite: rc = %d\n",
1101                        sbi->ll_fsname, err);
1102                 err = 0;
1103         }
1104
1105         /* The cfg_instance is a value unique to this super, in case some
1106          * joker tries to mount the same fs at two mount points.
1107          */
1108         cfg->cfg_instance = cfg_instance;
1109         cfg->cfg_uuid = lsi->lsi_llsbi->ll_sb_uuid;
1110         cfg->cfg_callback = class_config_llog_handler;
1111         cfg->cfg_sub_clds = CONFIG_SUB_CLIENT;
1112         /* set up client obds */
1113         err = lustre_process_log(sb, profilenm, cfg);
1114         if (err < 0)
1115                 GOTO(out_debugfs, err);
1116
1117         /* Profile set with LCFG_MOUNTOPT so we can find our mdc and osc obds */
1118         lprof = class_get_profile(profilenm);
1119         if (lprof == NULL) {
1120                 LCONSOLE_ERROR_MSG(0x156, "The client profile '%s' could not be"
1121                                    " read from the MGS.  Does that filesystem "
1122                                    "exist?\n", profilenm);
1123                 GOTO(out_debugfs, err = -EINVAL);
1124         }
1125         CDEBUG(D_CONFIG, "Found profile %s: mdc=%s osc=%s\n", profilenm,
1126                lprof->lp_md, lprof->lp_dt);
1127
1128         dt_len = strlen(lprof->lp_dt) + instlen + 2;
1129         OBD_ALLOC(dt, dt_len);
1130         if (!dt)
1131                 GOTO(out_profile, err = -ENOMEM);
1132         snprintf(dt, dt_len - 1, "%s-%016lx", lprof->lp_dt, cfg_instance);
1133
1134         md_len = strlen(lprof->lp_md) + instlen + 2;
1135         OBD_ALLOC(md, md_len);
1136         if (!md)
1137                 GOTO(out_free_dt, err = -ENOMEM);
1138         snprintf(md, md_len - 1, "%s-%016lx", lprof->lp_md, cfg_instance);
1139
1140         /* connections, registrations, sb setup */
1141         err = client_common_fill_super(sb, md, dt, mnt);
1142         if (err < 0)
1143                 GOTO(out_free_md, err);
1144
1145         sbi->ll_client_common_fill_super_succeeded = 1;
1146
1147 out_free_md:
1148         if (md)
1149                 OBD_FREE(md, md_len);
1150 out_free_dt:
1151         if (dt)
1152                 OBD_FREE(dt, dt_len);
1153 out_profile:
1154         if (lprof)
1155                 class_put_profile(lprof);
1156 out_debugfs:
1157         if (err < 0)
1158                 ll_debugfs_unregister_super(sb);
1159 out_free_cfg:
1160         if (cfg)
1161                 OBD_FREE_PTR(cfg);
1162
1163         if (err)
1164                 ll_put_super(sb);
1165         else if (sbi->ll_flags & LL_SBI_VERBOSE)
1166                 LCONSOLE_WARN("Mounted %s\n", profilenm);
1167         RETURN(err);
1168 } /* ll_fill_super */
1169
1170 void ll_put_super(struct super_block *sb)
1171 {
1172         struct config_llog_instance cfg, params_cfg;
1173         struct obd_device *obd;
1174         struct lustre_sb_info *lsi = s2lsi(sb);
1175         struct ll_sb_info *sbi = ll_s2sbi(sb);
1176         char *profilenm = get_profile_name(sb);
1177         unsigned long cfg_instance = ll_get_cfg_instance(sb);
1178         long ccc_count;
1179         int next, force = 1, rc = 0;
1180         ENTRY;
1181
1182         if (!sbi)
1183                 GOTO(out_no_sbi, 0);
1184
1185         /* Should replace instance_id with something better for ASLR */
1186         CDEBUG(D_VFSTRACE, "VFS Op: cfg_instance %s-%016lx (sb %p)\n",
1187                profilenm, cfg_instance, sb);
1188
1189         cfg.cfg_instance = cfg_instance;
1190         lustre_end_log(sb, profilenm, &cfg);
1191
1192         params_cfg.cfg_instance = cfg_instance;
1193         lustre_end_log(sb, PARAMS_FILENAME, &params_cfg);
1194
1195         if (sbi->ll_md_exp) {
1196                 obd = class_exp2obd(sbi->ll_md_exp);
1197                 if (obd)
1198                         force = obd->obd_force;
1199         }
1200
1201         /* Wait for unstable pages to be committed to stable storage */
1202         if (force == 0) {
1203                 struct l_wait_info lwi = LWI_INTR(LWI_ON_SIGNAL_NOOP, NULL);
1204                 rc = l_wait_event(sbi->ll_cache->ccc_unstable_waitq,
1205                         atomic_long_read(&sbi->ll_cache->ccc_unstable_nr) == 0,
1206                         &lwi);
1207         }
1208
1209         ccc_count = atomic_long_read(&sbi->ll_cache->ccc_unstable_nr);
1210         if (force == 0 && rc != -EINTR)
1211                 LASSERTF(ccc_count == 0, "count: %li\n", ccc_count);
1212
1213         /* We need to set force before the lov_disconnect in
1214            lustre_common_put_super, since l_d cleans up osc's as well. */
1215         if (force) {
1216                 next = 0;
1217                 while ((obd = class_devices_in_group(&sbi->ll_sb_uuid,
1218                                                      &next)) != NULL) {
1219                         obd->obd_force = force;
1220                 }
1221         }
1222
1223         if (sbi->ll_client_common_fill_super_succeeded) {
1224                 /* Only if client_common_fill_super succeeded */
1225                 client_common_put_super(sb);
1226         }
1227
1228         next = 0;
1229         while ((obd = class_devices_in_group(&sbi->ll_sb_uuid, &next)) !=NULL) {
1230                 class_manual_cleanup(obd);
1231         }
1232
1233         if (sbi->ll_flags & LL_SBI_VERBOSE)
1234                 LCONSOLE_WARN("Unmounted %s\n", profilenm ? profilenm : "");
1235
1236         if (profilenm)
1237                 class_del_profile(profilenm);
1238
1239 #ifndef HAVE_SUPER_SETUP_BDI_NAME
1240         if (lsi->lsi_flags & LSI_BDI_INITIALIZED) {
1241                 bdi_destroy(&lsi->lsi_bdi);
1242                 lsi->lsi_flags &= ~LSI_BDI_INITIALIZED;
1243         }
1244 #endif
1245
1246         ll_free_sbi(sb);
1247         lsi->lsi_llsbi = NULL;
1248 out_no_sbi:
1249         lustre_common_put_super(sb);
1250
1251         cl_env_cache_purge(~0);
1252
1253         module_put(THIS_MODULE);
1254
1255         EXIT;
1256 } /* client_put_super */
1257
1258 struct inode *ll_inode_from_resource_lock(struct ldlm_lock *lock)
1259 {
1260         struct inode *inode = NULL;
1261
1262         /* NOTE: we depend on atomic igrab() -bzzz */
1263         lock_res_and_lock(lock);
1264         if (lock->l_resource->lr_lvb_inode) {
1265                 struct ll_inode_info * lli;
1266                 lli = ll_i2info(lock->l_resource->lr_lvb_inode);
1267                 if (lli->lli_inode_magic == LLI_INODE_MAGIC) {
1268                         inode = igrab(lock->l_resource->lr_lvb_inode);
1269                 } else {
1270                         inode = lock->l_resource->lr_lvb_inode;
1271                         LDLM_DEBUG_LIMIT(inode->i_state & I_FREEING ?  D_INFO :
1272                                          D_WARNING, lock, "lr_lvb_inode %p is "
1273                                          "bogus: magic %08x",
1274                                          lock->l_resource->lr_lvb_inode,
1275                                          lli->lli_inode_magic);
1276                         inode = NULL;
1277                 }
1278         }
1279         unlock_res_and_lock(lock);
1280         return inode;
1281 }
1282
1283 void ll_dir_clear_lsm_md(struct inode *inode)
1284 {
1285         struct ll_inode_info *lli = ll_i2info(inode);
1286
1287         LASSERT(S_ISDIR(inode->i_mode));
1288
1289         if (lli->lli_lsm_md != NULL) {
1290                 lmv_free_memmd(lli->lli_lsm_md);
1291                 lli->lli_lsm_md = NULL;
1292         }
1293 }
1294
1295 static struct inode *ll_iget_anon_dir(struct super_block *sb,
1296                                       const struct lu_fid *fid,
1297                                       struct lustre_md *md)
1298 {
1299         struct ll_sb_info       *sbi = ll_s2sbi(sb);
1300         struct mdt_body         *body = md->body;
1301         struct inode            *inode;
1302         ino_t                   ino;
1303         ENTRY;
1304
1305         ino = cl_fid_build_ino(fid, sbi->ll_flags & LL_SBI_32BIT_API);
1306         inode = iget_locked(sb, ino);
1307         if (inode == NULL) {
1308                 CERROR("%s: failed get simple inode "DFID": rc = -ENOENT\n",
1309                        sbi->ll_fsname, PFID(fid));
1310                 RETURN(ERR_PTR(-ENOENT));
1311         }
1312
1313         if (inode->i_state & I_NEW) {
1314                 struct ll_inode_info *lli = ll_i2info(inode);
1315                 struct lmv_stripe_md *lsm = md->lmv;
1316
1317                 inode->i_mode = (inode->i_mode & ~S_IFMT) |
1318                                 (body->mbo_mode & S_IFMT);
1319                 LASSERTF(S_ISDIR(inode->i_mode), "Not slave inode "DFID"\n",
1320                          PFID(fid));
1321
1322                 inode->i_mtime.tv_sec = 0;
1323                 inode->i_atime.tv_sec = 0;
1324                 inode->i_ctime.tv_sec = 0;
1325                 inode->i_rdev = 0;
1326
1327 #ifdef HAVE_BACKING_DEV_INFO
1328                 /* initializing backing dev info. */
1329                 inode->i_mapping->backing_dev_info =
1330                                                 &s2lsi(inode->i_sb)->lsi_bdi;
1331 #endif
1332                 inode->i_op = &ll_dir_inode_operations;
1333                 inode->i_fop = &ll_dir_operations;
1334                 lli->lli_fid = *fid;
1335                 ll_lli_init(lli);
1336
1337                 LASSERT(lsm != NULL);
1338                 /* master object FID */
1339                 lli->lli_pfid = body->mbo_fid1;
1340                 CDEBUG(D_INODE, "lli %p slave "DFID" master "DFID"\n",
1341                        lli, PFID(fid), PFID(&lli->lli_pfid));
1342                 unlock_new_inode(inode);
1343         }
1344
1345         RETURN(inode);
1346 }
1347
1348 static int ll_init_lsm_md(struct inode *inode, struct lustre_md *md)
1349 {
1350         struct lu_fid *fid;
1351         struct lmv_stripe_md *lsm = md->lmv;
1352         struct ll_inode_info *lli = ll_i2info(inode);
1353         int i;
1354
1355         LASSERT(lsm != NULL);
1356
1357         CDEBUG(D_INODE, "%s: "DFID" set dir layout:\n",
1358                ll_i2sbi(inode)->ll_fsname, PFID(&lli->lli_fid));
1359         lsm_md_dump(D_INODE, lsm);
1360
1361         /* XXX sigh, this lsm_root initialization should be in
1362          * LMV layer, but it needs ll_iget right now, so we
1363          * put this here right now. */
1364         for (i = 0; i < lsm->lsm_md_stripe_count; i++) {
1365                 fid = &lsm->lsm_md_oinfo[i].lmo_fid;
1366                 LASSERT(lsm->lsm_md_oinfo[i].lmo_root == NULL);
1367
1368                 if (!fid_is_sane(fid))
1369                         continue;
1370
1371                 /* Unfortunately ll_iget will call ll_update_inode,
1372                  * where the initialization of slave inode is slightly
1373                  * different, so it reset lsm_md to NULL to avoid
1374                  * initializing lsm for slave inode. */
1375                 lsm->lsm_md_oinfo[i].lmo_root =
1376                                 ll_iget_anon_dir(inode->i_sb, fid, md);
1377                 if (IS_ERR(lsm->lsm_md_oinfo[i].lmo_root)) {
1378                         int rc = PTR_ERR(lsm->lsm_md_oinfo[i].lmo_root);
1379
1380                         lsm->lsm_md_oinfo[i].lmo_root = NULL;
1381                         while (i-- > 0) {
1382                                 iput(lsm->lsm_md_oinfo[i].lmo_root);
1383                                 lsm->lsm_md_oinfo[i].lmo_root = NULL;
1384                         }
1385                         return rc;
1386                 }
1387         }
1388
1389         lli->lli_lsm_md = lsm;
1390
1391         return 0;
1392 }
1393
1394 static int ll_update_lsm_md(struct inode *inode, struct lustre_md *md)
1395 {
1396         struct ll_inode_info *lli = ll_i2info(inode);
1397         struct lmv_stripe_md *lsm = md->lmv;
1398         int rc = 0;
1399
1400         ENTRY;
1401
1402         LASSERT(S_ISDIR(inode->i_mode));
1403         CDEBUG(D_INODE, "update lsm %p of "DFID"\n", lli->lli_lsm_md,
1404                PFID(ll_inode2fid(inode)));
1405
1406         /*
1407          * no striped information from request, lustre_md from req does not
1408          * include stripeEA, see ll_md_setattr()
1409          */
1410         if (!lsm)
1411                 RETURN(0);
1412
1413         /*
1414          * normally dir layout doesn't change, only take read lock to check
1415          * that to avoid blocking other MD operations.
1416          */
1417         if (lli->lli_lsm_md)
1418                 down_read(&lli->lli_lsm_sem);
1419         else
1420                 down_write(&lli->lli_lsm_sem);
1421
1422         /*
1423          * if dir layout mismatch, check whether version is increased, which
1424          * means layout is changed, this happens in dir migration and lfsck.
1425          *
1426          * foreign LMV should not change.
1427          */
1428         if (lli->lli_lsm_md &&
1429             lli->lli_lsm_md->lsm_md_magic != LMV_MAGIC_FOREIGN &&
1430            !lsm_md_eq(lli->lli_lsm_md, lsm)) {
1431                 if (lsm->lsm_md_layout_version <=
1432                     lli->lli_lsm_md->lsm_md_layout_version) {
1433                         CERROR("%s: "DFID" dir layout mismatch:\n",
1434                                ll_i2sbi(inode)->ll_fsname,
1435                                PFID(&lli->lli_fid));
1436                         lsm_md_dump(D_ERROR, lli->lli_lsm_md);
1437                         lsm_md_dump(D_ERROR, lsm);
1438                         GOTO(unlock, rc = -EINVAL);
1439                 }
1440
1441                 /* layout changed, switch to write lock */
1442                 up_read(&lli->lli_lsm_sem);
1443                 down_write(&lli->lli_lsm_sem);
1444                 ll_dir_clear_lsm_md(inode);
1445         }
1446
1447         /* set directory layout */
1448         if (!lli->lli_lsm_md) {
1449                 struct cl_attr  *attr;
1450
1451                 if (lsm->lsm_md_magic == LMV_MAGIC_FOREIGN) {
1452                         /* set md->lmv to NULL, so the following free lustre_md
1453                          * will not free this lsm */
1454                         md->lmv = NULL;
1455                         lli->lli_lsm_md = lsm;
1456                         up_write(&lli->lli_lsm_sem);
1457                         RETURN(0);
1458                 }
1459
1460                 rc = ll_init_lsm_md(inode, md);
1461                 up_write(&lli->lli_lsm_sem);
1462                 if (rc != 0)
1463                         RETURN(rc);
1464
1465                 /* set md->lmv to NULL, so the following free lustre_md
1466                  * will not free this lsm */
1467                 md->lmv = NULL;
1468
1469                 /*
1470                  * md_merge_attr() may take long, since lsm is already set,
1471                  * switch to read lock.
1472                  */
1473                 down_read(&lli->lli_lsm_sem);
1474
1475                 OBD_ALLOC_PTR(attr);
1476                 if (attr == NULL)
1477                         GOTO(unlock, rc = -ENOMEM);
1478
1479                 /* validate the lsm */
1480                 rc = md_merge_attr(ll_i2mdexp(inode), lsm, attr,
1481                                    ll_md_blocking_ast);
1482                 if (rc != 0) {
1483                         OBD_FREE_PTR(attr);
1484                         GOTO(unlock, rc);
1485                 }
1486
1487                 if (md->body->mbo_valid & OBD_MD_FLNLINK)
1488                         md->body->mbo_nlink = attr->cat_nlink;
1489                 if (md->body->mbo_valid & OBD_MD_FLSIZE)
1490                         md->body->mbo_size = attr->cat_size;
1491                 if (md->body->mbo_valid & OBD_MD_FLATIME)
1492                         md->body->mbo_atime = attr->cat_atime;
1493                 if (md->body->mbo_valid & OBD_MD_FLCTIME)
1494                         md->body->mbo_ctime = attr->cat_ctime;
1495                 if (md->body->mbo_valid & OBD_MD_FLMTIME)
1496                         md->body->mbo_mtime = attr->cat_mtime;
1497
1498                 OBD_FREE_PTR(attr);
1499         }
1500 unlock:
1501         up_read(&lli->lli_lsm_sem);
1502
1503         RETURN(rc);
1504 }
1505
1506 void ll_clear_inode(struct inode *inode)
1507 {
1508         struct ll_inode_info *lli = ll_i2info(inode);
1509         struct ll_sb_info *sbi = ll_i2sbi(inode);
1510         ENTRY;
1511
1512         CDEBUG(D_VFSTRACE, "VFS Op:inode="DFID"(%p)\n",
1513                PFID(ll_inode2fid(inode)), inode);
1514
1515         if (S_ISDIR(inode->i_mode)) {
1516                 /* these should have been cleared in ll_file_release */
1517                 LASSERT(lli->lli_opendir_key == NULL);
1518                 LASSERT(lli->lli_sai == NULL);
1519                 LASSERT(lli->lli_opendir_pid == 0);
1520         }
1521
1522         md_null_inode(sbi->ll_md_exp, ll_inode2fid(inode));
1523
1524         LASSERT(!lli->lli_open_fd_write_count);
1525         LASSERT(!lli->lli_open_fd_read_count);
1526         LASSERT(!lli->lli_open_fd_exec_count);
1527
1528         if (lli->lli_mds_write_och)
1529                 ll_md_real_close(inode, FMODE_WRITE);
1530         if (lli->lli_mds_exec_och)
1531                 ll_md_real_close(inode, FMODE_EXEC);
1532         if (lli->lli_mds_read_och)
1533                 ll_md_real_close(inode, FMODE_READ);
1534
1535         if (S_ISLNK(inode->i_mode) && lli->lli_symlink_name) {
1536                 OBD_FREE(lli->lli_symlink_name,
1537                          strlen(lli->lli_symlink_name) + 1);
1538                 lli->lli_symlink_name = NULL;
1539         }
1540
1541         ll_xattr_cache_destroy(inode);
1542
1543 #ifdef CONFIG_FS_POSIX_ACL
1544         forget_all_cached_acls(inode);
1545         if (lli->lli_posix_acl) {
1546                 posix_acl_release(lli->lli_posix_acl);
1547                 lli->lli_posix_acl = NULL;
1548         }
1549 #endif
1550         lli->lli_inode_magic = LLI_INODE_DEAD;
1551
1552         if (S_ISDIR(inode->i_mode))
1553                 ll_dir_clear_lsm_md(inode);
1554         else if (S_ISREG(inode->i_mode) && !is_bad_inode(inode))
1555                 LASSERT(list_empty(&lli->lli_agl_list));
1556
1557         /*
1558          * XXX This has to be done before lsm is freed below, because
1559          * cl_object still uses inode lsm.
1560          */
1561         cl_inode_fini(inode);
1562
1563         EXIT;
1564 }
1565
1566 static int ll_md_setattr(struct dentry *dentry, struct md_op_data *op_data)
1567 {
1568         struct lustre_md md;
1569         struct inode *inode = dentry->d_inode;
1570         struct ll_sb_info *sbi = ll_i2sbi(inode);
1571         struct ptlrpc_request *request = NULL;
1572         int rc, ia_valid;
1573         ENTRY;
1574
1575         op_data = ll_prep_md_op_data(op_data, inode, NULL, NULL, 0, 0,
1576                                      LUSTRE_OPC_ANY, NULL);
1577         if (IS_ERR(op_data))
1578                 RETURN(PTR_ERR(op_data));
1579
1580         rc = md_setattr(sbi->ll_md_exp, op_data, NULL, 0, &request);
1581         if (rc) {
1582                 ptlrpc_req_finished(request);
1583                 if (rc == -ENOENT) {
1584                         clear_nlink(inode);
1585                         /* Unlinked special device node? Or just a race?
1586                          * Pretend we done everything. */
1587                         if (!S_ISREG(inode->i_mode) &&
1588                             !S_ISDIR(inode->i_mode)) {
1589                                 ia_valid = op_data->op_attr.ia_valid;
1590                                 op_data->op_attr.ia_valid &= ~TIMES_SET_FLAGS;
1591                                 rc = simple_setattr(dentry, &op_data->op_attr);
1592                                 op_data->op_attr.ia_valid = ia_valid;
1593                         }
1594                 } else if (rc != -EPERM && rc != -EACCES && rc != -ETXTBSY) {
1595                         CERROR("md_setattr fails: rc = %d\n", rc);
1596                 }
1597                 RETURN(rc);
1598         }
1599
1600         rc = md_get_lustre_md(sbi->ll_md_exp, request, sbi->ll_dt_exp,
1601                               sbi->ll_md_exp, &md);
1602         if (rc) {
1603                 ptlrpc_req_finished(request);
1604                 RETURN(rc);
1605         }
1606
1607         ia_valid = op_data->op_attr.ia_valid;
1608         /* inode size will be in ll_setattr_ost, can't do it now since dirty
1609          * cache is not cleared yet. */
1610         op_data->op_attr.ia_valid &= ~(TIMES_SET_FLAGS | ATTR_SIZE);
1611         if (S_ISREG(inode->i_mode))
1612                 inode_lock(inode);
1613         rc = simple_setattr(dentry, &op_data->op_attr);
1614         if (S_ISREG(inode->i_mode))
1615                 inode_unlock(inode);
1616         op_data->op_attr.ia_valid = ia_valid;
1617
1618         rc = ll_update_inode(inode, &md);
1619         ptlrpc_req_finished(request);
1620
1621         RETURN(rc);
1622 }
1623
1624 /* If this inode has objects allocated to it (lsm != NULL), then the OST
1625  * object(s) determine the file size and mtime.  Otherwise, the MDS will
1626  * keep these values until such a time that objects are allocated for it.
1627  * We do the MDS operations first, as it is checking permissions for us.
1628  * We don't to the MDS RPC if there is nothing that we want to store there,
1629  * otherwise there is no harm in updating mtime/atime on the MDS if we are
1630  * going to do an RPC anyways.
1631  *
1632  * If we are doing a truncate, we will send the mtime and ctime updates
1633  * to the OST with the punch RPC, otherwise we do an explicit setattr RPC.
1634  * I don't believe it is possible to get e.g. ATTR_MTIME_SET and ATTR_SIZE
1635  * at the same time.
1636  *
1637  * In case of HSMimport, we only set attr on MDS.
1638  */
1639 int ll_setattr_raw(struct dentry *dentry, struct iattr *attr,
1640                    enum op_xvalid xvalid, bool hsm_import)
1641 {
1642         struct inode *inode = dentry->d_inode;
1643         struct ll_inode_info *lli = ll_i2info(inode);
1644         struct md_op_data *op_data = NULL;
1645         int rc = 0;
1646         ENTRY;
1647
1648         CDEBUG(D_VFSTRACE, "%s: setattr inode "DFID"(%p) from %llu to %llu, "
1649                "valid %x, hsm_import %d\n",
1650                ll_i2sbi(inode)->ll_fsname, PFID(&lli->lli_fid),
1651                inode, i_size_read(inode), attr->ia_size, attr->ia_valid,
1652                hsm_import);
1653
1654         if (attr->ia_valid & ATTR_SIZE) {
1655                 /* Check new size against VFS/VM file size limit and rlimit */
1656                 rc = inode_newsize_ok(inode, attr->ia_size);
1657                 if (rc)
1658                         RETURN(rc);
1659
1660                 /* The maximum Lustre file size is variable, based on the
1661                  * OST maximum object size and number of stripes.  This
1662                  * needs another check in addition to the VFS check above. */
1663                 if (attr->ia_size > ll_file_maxbytes(inode)) {
1664                         CDEBUG(D_INODE,"file "DFID" too large %llu > %llu\n",
1665                                PFID(&lli->lli_fid), attr->ia_size,
1666                                ll_file_maxbytes(inode));
1667                         RETURN(-EFBIG);
1668                 }
1669
1670                 attr->ia_valid |= ATTR_MTIME | ATTR_CTIME;
1671         }
1672
1673         /* POSIX: check before ATTR_*TIME_SET set (from inode_change_ok) */
1674         if (attr->ia_valid & TIMES_SET_FLAGS) {
1675                 if ((!uid_eq(current_fsuid(), inode->i_uid)) &&
1676                     !cfs_capable(CFS_CAP_FOWNER))
1677                         RETURN(-EPERM);
1678         }
1679
1680         /* We mark all of the fields "set" so MDS/OST does not re-set them */
1681         if (!(xvalid & OP_XVALID_CTIME_SET) &&
1682              (attr->ia_valid & ATTR_CTIME)) {
1683                 attr->ia_ctime = current_time(inode);
1684                 xvalid |= OP_XVALID_CTIME_SET;
1685         }
1686         if (!(attr->ia_valid & ATTR_ATIME_SET) &&
1687             (attr->ia_valid & ATTR_ATIME)) {
1688                 attr->ia_atime = current_time(inode);
1689                 attr->ia_valid |= ATTR_ATIME_SET;
1690         }
1691         if (!(attr->ia_valid & ATTR_MTIME_SET) &&
1692             (attr->ia_valid & ATTR_MTIME)) {
1693                 attr->ia_mtime = current_time(inode);
1694                 attr->ia_valid |= ATTR_MTIME_SET;
1695         }
1696
1697         if (attr->ia_valid & (ATTR_MTIME | ATTR_CTIME))
1698                 CDEBUG(D_INODE, "setting mtime %lld, ctime %lld, now = %lld\n",
1699                        (s64)attr->ia_mtime.tv_sec, (s64)attr->ia_ctime.tv_sec,
1700                        ktime_get_real_seconds());
1701
1702         if (S_ISREG(inode->i_mode)) {
1703                 if (attr->ia_valid & ATTR_SIZE)
1704                         inode_dio_write_done(inode);
1705                 inode_unlock(inode);
1706         }
1707
1708         /* We always do an MDS RPC, even if we're only changing the size;
1709          * only the MDS knows whether truncate() should fail with -ETXTBUSY */
1710
1711         OBD_ALLOC_PTR(op_data);
1712         if (op_data == NULL)
1713                 GOTO(out, rc = -ENOMEM);
1714
1715         if (!hsm_import && attr->ia_valid & ATTR_SIZE) {
1716                 /* If we are changing file size, file content is
1717                  * modified, flag it.
1718                  */
1719                 xvalid |= OP_XVALID_OWNEROVERRIDE;
1720                 op_data->op_bias |= MDS_DATA_MODIFIED;
1721                 ll_file_clear_flag(lli, LLIF_DATA_MODIFIED);
1722         }
1723
1724         if (attr->ia_valid & ATTR_FILE) {
1725                 struct ll_file_data *fd = LUSTRE_FPRIVATE(attr->ia_file);
1726
1727                 if (fd->fd_lease_och)
1728                         op_data->op_bias |= MDS_TRUNC_KEEP_LEASE;
1729         }
1730
1731         op_data->op_attr = *attr;
1732         op_data->op_xvalid = xvalid;
1733
1734         rc = ll_md_setattr(dentry, op_data);
1735         if (rc)
1736                 GOTO(out, rc);
1737
1738         if (!S_ISREG(inode->i_mode) || hsm_import)
1739                 GOTO(out, rc = 0);
1740
1741         if (attr->ia_valid & (ATTR_SIZE | ATTR_ATIME | ATTR_ATIME_SET |
1742                               ATTR_MTIME | ATTR_MTIME_SET | ATTR_CTIME) ||
1743             xvalid & OP_XVALID_CTIME_SET) {
1744                 /* For truncate and utimes sending attributes to OSTs, setting
1745                  * mtime/atime to the past will be performed under PW [0:EOF]
1746                  * extent lock (new_size:EOF for truncate).  It may seem
1747                  * excessive to send mtime/atime updates to OSTs when not
1748                  * setting times to past, but it is necessary due to possible
1749                  * time de-synchronization between MDT inode and OST objects
1750                  */
1751                 rc = cl_setattr_ost(lli->lli_clob, attr, xvalid, 0);
1752         }
1753
1754         /* If the file was restored, it needs to set dirty flag.
1755          *
1756          * We've already sent MDS_DATA_MODIFIED flag in
1757          * ll_md_setattr() for truncate. However, the MDT refuses to
1758          * set the HS_DIRTY flag on released files, so we have to set
1759          * it again if the file has been restored. Please check how
1760          * LLIF_DATA_MODIFIED is set in vvp_io_setattr_fini().
1761          *
1762          * Please notice that if the file is not released, the previous
1763          * MDS_DATA_MODIFIED has taken effect and usually
1764          * LLIF_DATA_MODIFIED is not set(see vvp_io_setattr_fini()).
1765          * This way we can save an RPC for common open + trunc
1766          * operation. */
1767         if (ll_file_test_and_clear_flag(lli, LLIF_DATA_MODIFIED)) {
1768                 struct hsm_state_set hss = {
1769                         .hss_valid = HSS_SETMASK,
1770                         .hss_setmask = HS_DIRTY,
1771                 };
1772                 int rc2;
1773
1774                 rc2 = ll_hsm_state_set(inode, &hss);
1775                 /* truncate and write can happen at the same time, so that
1776                  * the file can be set modified even though the file is not
1777                  * restored from released state, and ll_hsm_state_set() is
1778                  * not applicable for the file, and rc2 < 0 is normal in this
1779                  * case. */
1780                 if (rc2 < 0)
1781                         CDEBUG(D_INFO, DFID "HSM set dirty failed: rc2 = %d\n",
1782                                PFID(ll_inode2fid(inode)), rc2);
1783         }
1784
1785         EXIT;
1786 out:
1787         if (op_data != NULL)
1788                 ll_finish_md_op_data(op_data);
1789
1790         if (S_ISREG(inode->i_mode)) {
1791                 inode_lock(inode);
1792                 if ((attr->ia_valid & ATTR_SIZE) && !hsm_import)
1793                         inode_dio_wait(inode);
1794                 /* Once we've got the i_mutex, it's safe to set the S_NOSEC
1795                  * flag.  ll_update_inode (called from ll_md_setattr), clears
1796                  * inode flags, so there is a gap where S_NOSEC is not set.
1797                  * This can cause a writer to take the i_mutex unnecessarily,
1798                  * but this is safe to do and should be rare. */
1799                 inode_has_no_xattr(inode);
1800         }
1801
1802         ll_stats_ops_tally(ll_i2sbi(inode), (attr->ia_valid & ATTR_SIZE) ?
1803                         LPROC_LL_TRUNC : LPROC_LL_SETATTR, 1);
1804
1805         return rc;
1806 }
1807
1808 int ll_setattr(struct dentry *de, struct iattr *attr)
1809 {
1810         int mode = de->d_inode->i_mode;
1811         enum op_xvalid xvalid = 0;
1812
1813         if ((attr->ia_valid & (ATTR_CTIME|ATTR_SIZE|ATTR_MODE)) ==
1814                               (ATTR_CTIME|ATTR_SIZE|ATTR_MODE))
1815                 xvalid |= OP_XVALID_OWNEROVERRIDE;
1816
1817         if (((attr->ia_valid & (ATTR_MODE|ATTR_FORCE|ATTR_SIZE)) ==
1818                                (ATTR_SIZE|ATTR_MODE)) &&
1819             (((mode & S_ISUID) && !(attr->ia_mode & S_ISUID)) ||
1820              (((mode & (S_ISGID|S_IXGRP)) == (S_ISGID|S_IXGRP)) &&
1821               !(attr->ia_mode & S_ISGID))))
1822                 attr->ia_valid |= ATTR_FORCE;
1823
1824         if ((attr->ia_valid & ATTR_MODE) &&
1825             (mode & S_ISUID) &&
1826             !(attr->ia_mode & S_ISUID) &&
1827             !(attr->ia_valid & ATTR_KILL_SUID))
1828                 attr->ia_valid |= ATTR_KILL_SUID;
1829
1830         if ((attr->ia_valid & ATTR_MODE) &&
1831             ((mode & (S_ISGID|S_IXGRP)) == (S_ISGID|S_IXGRP)) &&
1832             !(attr->ia_mode & S_ISGID) &&
1833             !(attr->ia_valid & ATTR_KILL_SGID))
1834                 attr->ia_valid |= ATTR_KILL_SGID;
1835
1836         return ll_setattr_raw(de, attr, xvalid, false);
1837 }
1838
1839 int ll_statfs_internal(struct ll_sb_info *sbi, struct obd_statfs *osfs,
1840                        u32 flags)
1841 {
1842         struct obd_statfs obd_osfs = { 0 };
1843         time64_t max_age;
1844         int rc;
1845
1846         ENTRY;
1847         max_age = ktime_get_seconds() - OBD_STATFS_CACHE_SECONDS;
1848
1849         rc = obd_statfs(NULL, sbi->ll_md_exp, osfs, max_age, flags);
1850         if (rc)
1851                 RETURN(rc);
1852
1853         osfs->os_type = LL_SUPER_MAGIC;
1854
1855         CDEBUG(D_SUPER, "MDC blocks %llu/%llu objects %llu/%llu\n",
1856               osfs->os_bavail, osfs->os_blocks, osfs->os_ffree, osfs->os_files);
1857
1858         if (osfs->os_state & OS_STATE_SUM)
1859                 GOTO(out, rc);
1860
1861         if (sbi->ll_flags & LL_SBI_LAZYSTATFS)
1862                 flags |= OBD_STATFS_NODELAY;
1863
1864         rc = obd_statfs(NULL, sbi->ll_dt_exp, &obd_osfs, max_age, flags);
1865         if (rc) /* Possibly a filesystem with no OSTs.  Report MDT totals. */
1866                 GOTO(out, rc = 0);
1867
1868         CDEBUG(D_SUPER, "OSC blocks %llu/%llu objects %llu/%llu\n",
1869                obd_osfs.os_bavail, obd_osfs.os_blocks, obd_osfs.os_ffree,
1870                obd_osfs.os_files);
1871
1872         osfs->os_bsize = obd_osfs.os_bsize;
1873         osfs->os_blocks = obd_osfs.os_blocks;
1874         osfs->os_bfree = obd_osfs.os_bfree;
1875         osfs->os_bavail = obd_osfs.os_bavail;
1876
1877         /* If we have _some_ OSTs, but don't have as many free objects on the
1878          * OSTs as inodes on the MDTs, reduce the reported number of inodes
1879          * to compensate, so that the "inodes in use" number is correct.
1880          * This should be kept in sync with lod_statfs() behaviour.
1881          */
1882         if (obd_osfs.os_files && obd_osfs.os_ffree < osfs->os_ffree) {
1883                 osfs->os_files = (osfs->os_files - osfs->os_ffree) +
1884                                  obd_osfs.os_ffree;
1885                 osfs->os_ffree = obd_osfs.os_ffree;
1886         }
1887
1888 out:
1889         RETURN(rc);
1890 }
1891 int ll_statfs(struct dentry *de, struct kstatfs *sfs)
1892 {
1893         struct super_block *sb = de->d_sb;
1894         struct obd_statfs osfs;
1895         __u64 fsid = huge_encode_dev(sb->s_dev);
1896         int rc;
1897
1898         CDEBUG(D_VFSTRACE, "VFS Op: at %llu jiffies\n", get_jiffies_64());
1899         ll_stats_ops_tally(ll_s2sbi(sb), LPROC_LL_STAFS, 1);
1900
1901         /* Some amount of caching on the client is allowed */
1902         rc = ll_statfs_internal(ll_s2sbi(sb), &osfs, OBD_STATFS_SUM);
1903         if (rc)
1904                 return rc;
1905
1906         statfs_unpack(sfs, &osfs);
1907
1908         /* We need to downshift for all 32-bit kernels, because we can't
1909          * tell if the kernel is being called via sys_statfs64() or not.
1910          * Stop before overflowing f_bsize - in which case it is better
1911          * to just risk EOVERFLOW if caller is using old sys_statfs(). */
1912         if (sizeof(long) < 8) {
1913                 while (osfs.os_blocks > ~0UL && sfs->f_bsize < 0x40000000) {
1914                         sfs->f_bsize <<= 1;
1915
1916                         osfs.os_blocks >>= 1;
1917                         osfs.os_bfree >>= 1;
1918                         osfs.os_bavail >>= 1;
1919                 }
1920         }
1921
1922         sfs->f_blocks = osfs.os_blocks;
1923         sfs->f_bfree = osfs.os_bfree;
1924         sfs->f_bavail = osfs.os_bavail;
1925         sfs->f_fsid.val[0] = (__u32)fsid;
1926         sfs->f_fsid.val[1] = (__u32)(fsid >> 32);
1927         return 0;
1928 }
1929
1930 void ll_inode_size_lock(struct inode *inode)
1931 {
1932         struct ll_inode_info *lli;
1933
1934         LASSERT(!S_ISDIR(inode->i_mode));
1935
1936         lli = ll_i2info(inode);
1937         mutex_lock(&lli->lli_size_mutex);
1938 }
1939
1940 void ll_inode_size_unlock(struct inode *inode)
1941 {
1942         struct ll_inode_info *lli;
1943
1944         lli = ll_i2info(inode);
1945         mutex_unlock(&lli->lli_size_mutex);
1946 }
1947
1948 void ll_update_inode_flags(struct inode *inode, int ext_flags)
1949 {
1950         inode->i_flags = ll_ext_to_inode_flags(ext_flags);
1951         if (ext_flags & LUSTRE_PROJINHERIT_FL)
1952                 ll_file_set_flag(ll_i2info(inode), LLIF_PROJECT_INHERIT);
1953         else
1954                 ll_file_clear_flag(ll_i2info(inode), LLIF_PROJECT_INHERIT);
1955 }
1956
1957 int ll_update_inode(struct inode *inode, struct lustre_md *md)
1958 {
1959         struct ll_inode_info *lli = ll_i2info(inode);
1960         struct mdt_body *body = md->body;
1961         struct ll_sb_info *sbi = ll_i2sbi(inode);
1962         int rc = 0;
1963
1964         if (body->mbo_valid & OBD_MD_FLEASIZE) {
1965                 rc = cl_file_inode_init(inode, md);
1966                 if (rc)
1967                         return rc;
1968         }
1969
1970         if (S_ISDIR(inode->i_mode)) {
1971                 rc = ll_update_lsm_md(inode, md);
1972                 if (rc != 0)
1973                         return rc;
1974         }
1975
1976 #ifdef CONFIG_FS_POSIX_ACL
1977         if (body->mbo_valid & OBD_MD_FLACL) {
1978                 spin_lock(&lli->lli_lock);
1979                 if (lli->lli_posix_acl)
1980                         posix_acl_release(lli->lli_posix_acl);
1981                 lli->lli_posix_acl = md->posix_acl;
1982                 spin_unlock(&lli->lli_lock);
1983         }
1984 #endif
1985         inode->i_ino = cl_fid_build_ino(&body->mbo_fid1,
1986                                         sbi->ll_flags & LL_SBI_32BIT_API);
1987         inode->i_generation = cl_fid_build_gen(&body->mbo_fid1);
1988
1989         if (body->mbo_valid & OBD_MD_FLATIME) {
1990                 if (body->mbo_atime > inode->i_atime.tv_sec)
1991                         inode->i_atime.tv_sec = body->mbo_atime;
1992                 lli->lli_atime = body->mbo_atime;
1993         }
1994
1995         if (body->mbo_valid & OBD_MD_FLMTIME) {
1996                 if (body->mbo_mtime > inode->i_mtime.tv_sec) {
1997                         CDEBUG(D_INODE,
1998                                "setting ino %lu mtime from %lld to %llu\n",
1999                                inode->i_ino, (s64)inode->i_mtime.tv_sec,
2000                                body->mbo_mtime);
2001                         inode->i_mtime.tv_sec = body->mbo_mtime;
2002                 }
2003                 lli->lli_mtime = body->mbo_mtime;
2004         }
2005
2006         if (body->mbo_valid & OBD_MD_FLCTIME) {
2007                 if (body->mbo_ctime > inode->i_ctime.tv_sec)
2008                         inode->i_ctime.tv_sec = body->mbo_ctime;
2009                 lli->lli_ctime = body->mbo_ctime;
2010         }
2011
2012         /* Clear i_flags to remove S_NOSEC before permissions are updated */
2013         if (body->mbo_valid & OBD_MD_FLFLAGS)
2014                 ll_update_inode_flags(inode, body->mbo_flags);
2015         if (body->mbo_valid & OBD_MD_FLMODE)
2016                 inode->i_mode = (inode->i_mode & S_IFMT) |
2017                                 (body->mbo_mode & ~S_IFMT);
2018
2019         if (body->mbo_valid & OBD_MD_FLTYPE)
2020                 inode->i_mode = (inode->i_mode & ~S_IFMT) |
2021                                 (body->mbo_mode & S_IFMT);
2022
2023         LASSERT(inode->i_mode != 0);
2024         if (S_ISREG(inode->i_mode))
2025                 inode->i_blkbits = min(PTLRPC_MAX_BRW_BITS + 1,
2026                                        LL_MAX_BLKSIZE_BITS);
2027         else
2028                 inode->i_blkbits = inode->i_sb->s_blocksize_bits;
2029
2030         if (body->mbo_valid & OBD_MD_FLUID)
2031                 inode->i_uid = make_kuid(&init_user_ns, body->mbo_uid);
2032         if (body->mbo_valid & OBD_MD_FLGID)
2033                 inode->i_gid = make_kgid(&init_user_ns, body->mbo_gid);
2034         if (body->mbo_valid & OBD_MD_FLPROJID)
2035                 lli->lli_projid = body->mbo_projid;
2036         if (body->mbo_valid & OBD_MD_FLNLINK)
2037                 set_nlink(inode, body->mbo_nlink);
2038         if (body->mbo_valid & OBD_MD_FLRDEV)
2039                 inode->i_rdev = old_decode_dev(body->mbo_rdev);
2040
2041         if (body->mbo_valid & OBD_MD_FLID) {
2042                 /* FID shouldn't be changed! */
2043                 if (fid_is_sane(&lli->lli_fid)) {
2044                         LASSERTF(lu_fid_eq(&lli->lli_fid, &body->mbo_fid1),
2045                                  "Trying to change FID "DFID
2046                                  " to the "DFID", inode "DFID"(%p)\n",
2047                                  PFID(&lli->lli_fid), PFID(&body->mbo_fid1),
2048                                  PFID(ll_inode2fid(inode)), inode);
2049                 } else {
2050                         lli->lli_fid = body->mbo_fid1;
2051                 }
2052         }
2053
2054         LASSERT(fid_seq(&lli->lli_fid) != 0);
2055
2056         if (body->mbo_valid & OBD_MD_FLSIZE) {
2057                 i_size_write(inode, body->mbo_size);
2058
2059                 CDEBUG(D_VFSTRACE, "inode="DFID", updating i_size %llu\n",
2060                        PFID(ll_inode2fid(inode)),
2061                        (unsigned long long)body->mbo_size);
2062
2063                 if (body->mbo_valid & OBD_MD_FLBLOCKS)
2064                         inode->i_blocks = body->mbo_blocks;
2065         }
2066
2067         if (body->mbo_valid & OBD_MD_TSTATE) {
2068                 /* Set LLIF_FILE_RESTORING if restore ongoing and
2069                  * clear it when done to ensure to start again
2070                  * glimpsing updated attrs
2071                  */
2072                 if (body->mbo_t_state & MS_RESTORE)
2073                         ll_file_set_flag(lli, LLIF_FILE_RESTORING);
2074                 else
2075                         ll_file_clear_flag(lli, LLIF_FILE_RESTORING);
2076         }
2077
2078         return 0;
2079 }
2080
2081 int ll_read_inode2(struct inode *inode, void *opaque)
2082 {
2083         struct lustre_md *md = opaque;
2084         struct ll_inode_info *lli = ll_i2info(inode);
2085         int     rc;
2086         ENTRY;
2087
2088         CDEBUG(D_VFSTRACE, "VFS Op:inode="DFID"(%p)\n",
2089                PFID(&lli->lli_fid), inode);
2090
2091         /* Core attributes from the MDS first.  This is a new inode, and
2092          * the VFS doesn't zero times in the core inode so we have to do
2093          * it ourselves.  They will be overwritten by either MDS or OST
2094          * attributes - we just need to make sure they aren't newer.
2095          */
2096         inode->i_mtime.tv_sec = 0;
2097         inode->i_atime.tv_sec = 0;
2098         inode->i_ctime.tv_sec = 0;
2099         inode->i_rdev = 0;
2100         rc = ll_update_inode(inode, md);
2101         if (rc != 0)
2102                 RETURN(rc);
2103
2104         /* OIDEBUG(inode); */
2105
2106 #ifdef HAVE_BACKING_DEV_INFO
2107         /* initializing backing dev info. */
2108         inode->i_mapping->backing_dev_info = &s2lsi(inode->i_sb)->lsi_bdi;
2109 #endif
2110         if (S_ISREG(inode->i_mode)) {
2111                 struct ll_sb_info *sbi = ll_i2sbi(inode);
2112                 inode->i_op = &ll_file_inode_operations;
2113                 inode->i_fop = sbi->ll_fop;
2114                 inode->i_mapping->a_ops = (struct address_space_operations *)&ll_aops;
2115                 EXIT;
2116         } else if (S_ISDIR(inode->i_mode)) {
2117                 inode->i_op = &ll_dir_inode_operations;
2118                 inode->i_fop = &ll_dir_operations;
2119                 EXIT;
2120         } else if (S_ISLNK(inode->i_mode)) {
2121                 inode->i_op = &ll_fast_symlink_inode_operations;
2122                 EXIT;
2123         } else {
2124                 inode->i_op = &ll_special_inode_operations;
2125
2126                 init_special_inode(inode, inode->i_mode,
2127                                    inode->i_rdev);
2128
2129                 EXIT;
2130         }
2131
2132         return 0;
2133 }
2134
2135 void ll_delete_inode(struct inode *inode)
2136 {
2137         struct ll_inode_info *lli = ll_i2info(inode);
2138         struct address_space *mapping = &inode->i_data;
2139         unsigned long nrpages;
2140         ENTRY;
2141
2142         if (S_ISREG(inode->i_mode) && lli->lli_clob != NULL) {
2143                 /* It is last chance to write out dirty pages,
2144                  * otherwise we may lose data while umount.
2145                  *
2146                  * If i_nlink is 0 then just discard data. This is safe because
2147                  * local inode gets i_nlink 0 from server only for the last
2148                  * unlink, so that file is not opened somewhere else
2149                  */
2150                 cl_sync_file_range(inode, 0, OBD_OBJECT_EOF, inode->i_nlink ?
2151                                    CL_FSYNC_LOCAL : CL_FSYNC_DISCARD, 1);
2152         }
2153         truncate_inode_pages_final(mapping);
2154
2155         /* Workaround for LU-118: Note nrpages may not be totally updated when
2156          * truncate_inode_pages() returns, as there can be a page in the process
2157          * of deletion (inside __delete_from_page_cache()) in the specified
2158          * range. Thus mapping->nrpages can be non-zero when this function
2159          * returns even after truncation of the whole mapping.  Only do this if
2160          * npages isn't already zero.
2161          */
2162         nrpages = mapping->nrpages;
2163         if (nrpages) {
2164                 xa_lock_irq(&mapping->i_pages);
2165                 nrpages = mapping->nrpages;
2166                 xa_unlock_irq(&mapping->i_pages);
2167         } /* Workaround end */
2168
2169         LASSERTF(nrpages == 0, "%s: inode="DFID"(%p) nrpages=%lu, "
2170                  "see https://jira.whamcloud.com/browse/LU-118\n",
2171                  ll_i2sbi(inode)->ll_fsname,
2172                  PFID(ll_inode2fid(inode)), inode, nrpages);
2173
2174 #ifdef HAVE_SBOPS_EVICT_INODE
2175         ll_clear_inode(inode);
2176 #endif
2177         clear_inode(inode);
2178
2179         EXIT;
2180 }
2181
2182 int ll_iocontrol(struct inode *inode, struct file *file,
2183                  unsigned int cmd, unsigned long arg)
2184 {
2185         struct ll_sb_info *sbi = ll_i2sbi(inode);
2186         struct ptlrpc_request *req = NULL;
2187         int rc, flags = 0;
2188         ENTRY;
2189
2190         switch (cmd) {
2191         case FS_IOC_GETFLAGS: {
2192                 struct mdt_body *body;
2193                 struct md_op_data *op_data;
2194
2195                 op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL,
2196                                              0, 0, LUSTRE_OPC_ANY,
2197                                              NULL);
2198                 if (IS_ERR(op_data))
2199                         RETURN(PTR_ERR(op_data));
2200
2201                 op_data->op_valid = OBD_MD_FLFLAGS;
2202                 rc = md_getattr(sbi->ll_md_exp, op_data, &req);
2203                 ll_finish_md_op_data(op_data);
2204                 if (rc) {
2205                         CERROR("%s: failure inode "DFID": rc = %d\n",
2206                                sbi->ll_md_exp->exp_obd->obd_name,
2207                                PFID(ll_inode2fid(inode)), rc);
2208                         RETURN(-abs(rc));
2209                 }
2210
2211                 body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
2212
2213                 flags = body->mbo_flags;
2214
2215                 ptlrpc_req_finished(req);
2216
2217                 RETURN(put_user(flags, (int __user *)arg));
2218         }
2219         case FS_IOC_SETFLAGS: {
2220                 struct iattr *attr;
2221                 struct md_op_data *op_data;
2222                 struct cl_object *obj;
2223                 struct fsxattr fa = { 0 };
2224
2225                 if (get_user(flags, (int __user *)arg))
2226                         RETURN(-EFAULT);
2227
2228                 fa.fsx_projid = ll_i2info(inode)->lli_projid;
2229                 if (flags & LUSTRE_PROJINHERIT_FL)
2230                         fa.fsx_xflags = FS_XFLAG_PROJINHERIT;
2231
2232                 rc = ll_ioctl_check_project(inode, &fa);
2233                 if (rc)
2234                         RETURN(rc);
2235
2236                 op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL, 0, 0,
2237                                              LUSTRE_OPC_ANY, NULL);
2238                 if (IS_ERR(op_data))
2239                         RETURN(PTR_ERR(op_data));
2240
2241                 op_data->op_attr_flags = flags;
2242                 op_data->op_xvalid |= OP_XVALID_FLAGS;
2243                 rc = md_setattr(sbi->ll_md_exp, op_data, NULL, 0, &req);
2244                 ll_finish_md_op_data(op_data);
2245                 ptlrpc_req_finished(req);
2246                 if (rc)
2247                         RETURN(rc);
2248
2249                 ll_update_inode_flags(inode, flags);
2250
2251                 obj = ll_i2info(inode)->lli_clob;
2252                 if (obj == NULL)
2253                         RETURN(0);
2254
2255                 OBD_ALLOC_PTR(attr);
2256                 if (attr == NULL)
2257                         RETURN(-ENOMEM);
2258
2259                 rc = cl_setattr_ost(obj, attr, OP_XVALID_FLAGS, flags);
2260
2261                 OBD_FREE_PTR(attr);
2262                 RETURN(rc);
2263         }
2264         default:
2265                 RETURN(-ENOSYS);
2266         }
2267
2268         RETURN(0);
2269 }
2270
2271 int ll_flush_ctx(struct inode *inode)
2272 {
2273         struct ll_sb_info  *sbi = ll_i2sbi(inode);
2274
2275         CDEBUG(D_SEC, "flush context for user %d\n",
2276                from_kuid(&init_user_ns, current_uid()));
2277
2278         obd_set_info_async(NULL, sbi->ll_md_exp,
2279                            sizeof(KEY_FLUSH_CTX), KEY_FLUSH_CTX,
2280                            0, NULL, NULL);
2281         obd_set_info_async(NULL, sbi->ll_dt_exp,
2282                            sizeof(KEY_FLUSH_CTX), KEY_FLUSH_CTX,
2283                            0, NULL, NULL);
2284         return 0;
2285 }
2286
2287 /* umount -f client means force down, don't save state */
2288 void ll_umount_begin(struct super_block *sb)
2289 {
2290         struct ll_sb_info *sbi = ll_s2sbi(sb);
2291         struct obd_device *obd;
2292         struct obd_ioctl_data *ioc_data;
2293         struct l_wait_info lwi;
2294         wait_queue_head_t waitq;
2295         ENTRY;
2296
2297         CDEBUG(D_VFSTRACE, "VFS Op: superblock %p count %d active %d\n", sb,
2298                sb->s_count, atomic_read(&sb->s_active));
2299
2300         obd = class_exp2obd(sbi->ll_md_exp);
2301         if (obd == NULL) {
2302                 CERROR("Invalid MDC connection handle %#llx\n",
2303                        sbi->ll_md_exp->exp_handle.h_cookie);
2304                 EXIT;
2305                 return;
2306         }
2307         obd->obd_force = 1;
2308
2309         obd = class_exp2obd(sbi->ll_dt_exp);
2310         if (obd == NULL) {
2311                 CERROR("Invalid LOV connection handle %#llx\n",
2312                        sbi->ll_dt_exp->exp_handle.h_cookie);
2313                 EXIT;
2314                 return;
2315         }
2316         obd->obd_force = 1;
2317
2318         OBD_ALLOC_PTR(ioc_data);
2319         if (ioc_data) {
2320                 obd_iocontrol(IOC_OSC_SET_ACTIVE, sbi->ll_md_exp,
2321                               sizeof *ioc_data, ioc_data, NULL);
2322
2323                 obd_iocontrol(IOC_OSC_SET_ACTIVE, sbi->ll_dt_exp,
2324                               sizeof *ioc_data, ioc_data, NULL);
2325
2326                 OBD_FREE_PTR(ioc_data);
2327         }
2328
2329         /* Really, we'd like to wait until there are no requests outstanding,
2330          * and then continue.  For now, we just periodically checking for vfs
2331          * to decrement mnt_cnt and hope to finish it within 10sec.
2332          */
2333         init_waitqueue_head(&waitq);
2334         lwi = LWI_TIMEOUT_INTERVAL(cfs_time_seconds(10),
2335                                    cfs_time_seconds(1), NULL, NULL);
2336         l_wait_event(waitq, may_umount(sbi->ll_mnt.mnt), &lwi);
2337
2338         EXIT;
2339 }
2340
2341 int ll_remount_fs(struct super_block *sb, int *flags, char *data)
2342 {
2343         struct ll_sb_info *sbi = ll_s2sbi(sb);
2344         char *profilenm = get_profile_name(sb);
2345         int err;
2346         __u32 read_only;
2347
2348         if ((*flags & MS_RDONLY) != (sb->s_flags & MS_RDONLY)) {
2349                 read_only = *flags & MS_RDONLY;
2350                 err = obd_set_info_async(NULL, sbi->ll_md_exp,
2351                                          sizeof(KEY_READ_ONLY),
2352                                          KEY_READ_ONLY, sizeof(read_only),
2353                                          &read_only, NULL);
2354                 if (err) {
2355                         LCONSOLE_WARN("Failed to remount %s %s (%d)\n",
2356                                       profilenm, read_only ?
2357                                       "read-only" : "read-write", err);
2358                         return err;
2359                 }
2360
2361                 if (read_only)
2362                         sb->s_flags |= MS_RDONLY;
2363                 else
2364                         sb->s_flags &= ~MS_RDONLY;
2365
2366                 if (sbi->ll_flags & LL_SBI_VERBOSE)
2367                         LCONSOLE_WARN("Remounted %s %s\n", profilenm,
2368                                       read_only ?  "read-only" : "read-write");
2369         }
2370         return 0;
2371 }
2372
2373 /**
2374  * Cleanup the open handle that is cached on MDT-side.
2375  *
2376  * For open case, the client side open handling thread may hit error
2377  * after the MDT grant the open. Under such case, the client should
2378  * send close RPC to the MDT as cleanup; otherwise, the open handle
2379  * on the MDT will be leaked there until the client umount or evicted.
2380  *
2381  * In further, if someone unlinked the file, because the open handle
2382  * holds the reference on such file/object, then it will block the
2383  * subsequent threads that want to locate such object via FID.
2384  *
2385  * \param[in] sb        super block for this file-system
2386  * \param[in] open_req  pointer to the original open request
2387  */
2388 void ll_open_cleanup(struct super_block *sb, struct ptlrpc_request *open_req)
2389 {
2390         struct mdt_body                 *body;
2391         struct md_op_data               *op_data;
2392         struct ptlrpc_request           *close_req = NULL;
2393         struct obd_export               *exp       = ll_s2sbi(sb)->ll_md_exp;
2394         ENTRY;
2395
2396         body = req_capsule_server_get(&open_req->rq_pill, &RMF_MDT_BODY);
2397         OBD_ALLOC_PTR(op_data);
2398         if (op_data == NULL) {
2399                 CWARN("%s: cannot allocate op_data to release open handle for "
2400                       DFID"\n", ll_s2sbi(sb)->ll_fsname, PFID(&body->mbo_fid1));
2401
2402                 RETURN_EXIT;
2403         }
2404
2405         op_data->op_fid1 = body->mbo_fid1;
2406         op_data->op_open_handle = body->mbo_open_handle;
2407         op_data->op_mod_time = ktime_get_real_seconds();
2408         md_close(exp, op_data, NULL, &close_req);
2409         ptlrpc_req_finished(close_req);
2410         ll_finish_md_op_data(op_data);
2411
2412         EXIT;
2413 }
2414
2415 int ll_prep_inode(struct inode **inode, struct ptlrpc_request *req,
2416                   struct super_block *sb, struct lookup_intent *it)
2417 {
2418         struct ll_sb_info *sbi = NULL;
2419         struct lustre_md md = { NULL };
2420         int rc;
2421         ENTRY;
2422
2423         LASSERT(*inode || sb);
2424         sbi = sb ? ll_s2sbi(sb) : ll_i2sbi(*inode);
2425         rc = md_get_lustre_md(sbi->ll_md_exp, req, sbi->ll_dt_exp,
2426                               sbi->ll_md_exp, &md);
2427         if (rc != 0)
2428                 GOTO(out, rc);
2429
2430         if (*inode) {
2431                 rc = ll_update_inode(*inode, &md);
2432                 if (rc != 0)
2433                         GOTO(out, rc);
2434         } else {
2435                 LASSERT(sb != NULL);
2436
2437                 /*
2438                  * At this point server returns to client's same fid as client
2439                  * generated for creating. So using ->fid1 is okay here.
2440                  */
2441                 if (!fid_is_sane(&md.body->mbo_fid1)) {
2442                         CERROR("%s: Fid is insane "DFID"\n",
2443                                 sbi->ll_fsname,
2444                                 PFID(&md.body->mbo_fid1));
2445                         GOTO(out, rc = -EINVAL);
2446                 }
2447
2448                 *inode = ll_iget(sb, cl_fid_build_ino(&md.body->mbo_fid1,
2449                                              sbi->ll_flags & LL_SBI_32BIT_API),
2450                                  &md);
2451                 if (IS_ERR(*inode)) {
2452 #ifdef CONFIG_FS_POSIX_ACL
2453                         if (md.posix_acl) {
2454                                 posix_acl_release(md.posix_acl);
2455                                 md.posix_acl = NULL;
2456                         }
2457 #endif
2458                         rc = IS_ERR(*inode) ? PTR_ERR(*inode) : -ENOMEM;
2459                         *inode = NULL;
2460                         CERROR("new_inode -fatal: rc %d\n", rc);
2461                         GOTO(out, rc);
2462                 }
2463         }
2464
2465         /* Handling piggyback layout lock.
2466          * Layout lock can be piggybacked by getattr and open request.
2467          * The lsm can be applied to inode only if it comes with a layout lock
2468          * otherwise correct layout may be overwritten, for example:
2469          * 1. proc1: mdt returns a lsm but not granting layout
2470          * 2. layout was changed by another client
2471          * 3. proc2: refresh layout and layout lock granted
2472          * 4. proc1: to apply a stale layout */
2473         if (it != NULL && it->it_lock_mode != 0) {
2474                 struct lustre_handle lockh;
2475                 struct ldlm_lock *lock;
2476
2477                 lockh.cookie = it->it_lock_handle;
2478                 lock = ldlm_handle2lock(&lockh);
2479                 LASSERT(lock != NULL);
2480                 if (ldlm_has_layout(lock)) {
2481                         struct cl_object_conf conf;
2482
2483                         memset(&conf, 0, sizeof(conf));
2484                         conf.coc_opc = OBJECT_CONF_SET;
2485                         conf.coc_inode = *inode;
2486                         conf.coc_lock = lock;
2487                         conf.u.coc_layout = md.layout;
2488                         (void)ll_layout_conf(*inode, &conf);
2489                 }
2490                 LDLM_LOCK_PUT(lock);
2491         }
2492
2493         GOTO(out, rc = 0);
2494
2495 out:
2496         /* cleanup will be done if necessary */
2497         md_free_lustre_md(sbi->ll_md_exp, &md);
2498
2499         if (rc != 0 && it != NULL && it->it_op & IT_OPEN)
2500                 ll_open_cleanup(sb != NULL ? sb : (*inode)->i_sb, req);
2501
2502         return rc;
2503 }
2504
2505 int ll_obd_statfs(struct inode *inode, void __user *arg)
2506 {
2507         struct ll_sb_info *sbi = NULL;
2508         struct obd_export *exp;
2509         char *buf = NULL;
2510         struct obd_ioctl_data *data = NULL;
2511         __u32 type;
2512         int len = 0, rc;
2513
2514         if (!inode || !(sbi = ll_i2sbi(inode)))
2515                 GOTO(out_statfs, rc = -EINVAL);
2516
2517         rc = obd_ioctl_getdata(&buf, &len, arg);
2518         if (rc)
2519                 GOTO(out_statfs, rc);
2520
2521         data = (void*)buf;
2522         if (!data->ioc_inlbuf1 || !data->ioc_inlbuf2 ||
2523             !data->ioc_pbuf1 || !data->ioc_pbuf2)
2524                 GOTO(out_statfs, rc = -EINVAL);
2525
2526         if (data->ioc_inllen1 != sizeof(__u32) ||
2527             data->ioc_inllen2 != sizeof(__u32) ||
2528             data->ioc_plen1 != sizeof(struct obd_statfs) ||
2529             data->ioc_plen2 != sizeof(struct obd_uuid))
2530                 GOTO(out_statfs, rc = -EINVAL);
2531
2532         memcpy(&type, data->ioc_inlbuf1, sizeof(__u32));
2533         if (type & LL_STATFS_LMV)
2534                 exp = sbi->ll_md_exp;
2535         else if (type & LL_STATFS_LOV)
2536                 exp = sbi->ll_dt_exp;
2537         else
2538                 GOTO(out_statfs, rc = -ENODEV);
2539
2540         rc = obd_iocontrol(IOC_OBD_STATFS, exp, len, buf, NULL);
2541         if (rc)
2542                 GOTO(out_statfs, rc);
2543 out_statfs:
2544         OBD_FREE_LARGE(buf, len);
2545         return rc;
2546 }
2547
2548 /*
2549  * this is normally called in ll_fini_md_op_data(), but sometimes it needs to
2550  * be called early to avoid deadlock.
2551  */
2552 void ll_unlock_md_op_lsm(struct md_op_data *op_data)
2553 {
2554         if (op_data->op_mea2_sem) {
2555                 up_read(op_data->op_mea2_sem);
2556                 op_data->op_mea2_sem = NULL;
2557         }
2558
2559         if (op_data->op_mea1_sem) {
2560                 up_read(op_data->op_mea1_sem);
2561                 op_data->op_mea1_sem = NULL;
2562         }
2563 }
2564
2565 /* this function prepares md_op_data hint for passing it down to MD stack. */
2566 struct md_op_data *ll_prep_md_op_data(struct md_op_data *op_data,
2567                                       struct inode *i1, struct inode *i2,
2568                                       const char *name, size_t namelen,
2569                                       __u32 mode, __u32 opc, void *data)
2570 {
2571         LASSERT(i1 != NULL);
2572
2573         if (name == NULL) {
2574                 /* Do not reuse namelen for something else. */
2575                 if (namelen != 0)
2576                         return ERR_PTR(-EINVAL);
2577         } else {
2578                 if (namelen > ll_i2sbi(i1)->ll_namelen)
2579                         return ERR_PTR(-ENAMETOOLONG);
2580
2581                 if (!lu_name_is_valid_2(name, namelen))
2582                         return ERR_PTR(-EINVAL);
2583         }
2584
2585         if (op_data == NULL)
2586                 OBD_ALLOC_PTR(op_data);
2587
2588         if (op_data == NULL)
2589                 return ERR_PTR(-ENOMEM);
2590
2591         ll_i2gids(op_data->op_suppgids, i1, i2);
2592         op_data->op_fid1 = *ll_inode2fid(i1);
2593         op_data->op_default_stripe_offset = -1;
2594
2595         if (S_ISDIR(i1->i_mode)) {
2596                 down_read(&ll_i2info(i1)->lli_lsm_sem);
2597                 op_data->op_mea1_sem = &ll_i2info(i1)->lli_lsm_sem;
2598                 op_data->op_mea1 = ll_i2info(i1)->lli_lsm_md;
2599                 if (opc == LUSTRE_OPC_MKDIR)
2600                         op_data->op_default_stripe_offset =
2601                                    ll_i2info(i1)->lli_def_stripe_offset;
2602         }
2603
2604         if (i2) {
2605                 op_data->op_fid2 = *ll_inode2fid(i2);
2606                 if (S_ISDIR(i2->i_mode)) {
2607                         if (i2 != i1) {
2608                                 down_read(&ll_i2info(i2)->lli_lsm_sem);
2609                                 op_data->op_mea2_sem =
2610                                                 &ll_i2info(i2)->lli_lsm_sem;
2611                         }
2612                         op_data->op_mea2 = ll_i2info(i2)->lli_lsm_md;
2613                 }
2614         } else {
2615                 fid_zero(&op_data->op_fid2);
2616         }
2617
2618         if (ll_i2sbi(i1)->ll_flags & LL_SBI_64BIT_HASH)
2619                 op_data->op_cli_flags |= CLI_HASH64;
2620
2621         if (ll_need_32bit_api(ll_i2sbi(i1)))
2622                 op_data->op_cli_flags |= CLI_API32;
2623
2624         op_data->op_name = name;
2625         op_data->op_namelen = namelen;
2626         op_data->op_mode = mode;
2627         op_data->op_mod_time = ktime_get_real_seconds();
2628         op_data->op_fsuid = from_kuid(&init_user_ns, current_fsuid());
2629         op_data->op_fsgid = from_kgid(&init_user_ns, current_fsgid());
2630         op_data->op_cap = cfs_curproc_cap_pack();
2631         op_data->op_mds = 0;
2632         if ((opc == LUSTRE_OPC_CREATE) && (name != NULL) &&
2633              filename_is_volatile(name, namelen, &op_data->op_mds)) {
2634                 op_data->op_bias |= MDS_CREATE_VOLATILE;
2635         }
2636         op_data->op_data = data;
2637
2638         return op_data;
2639 }
2640
2641 void ll_finish_md_op_data(struct md_op_data *op_data)
2642 {
2643         ll_unlock_md_op_lsm(op_data);
2644         security_release_secctx(op_data->op_file_secctx,
2645                                 op_data->op_file_secctx_size);
2646         OBD_FREE_PTR(op_data);
2647 }
2648
2649 #ifdef HAVE_SUPEROPS_USE_DENTRY
2650 int ll_show_options(struct seq_file *seq, struct dentry *dentry)
2651 #else
2652 int ll_show_options(struct seq_file *seq, struct vfsmount *vfs)
2653 #endif
2654 {
2655         struct ll_sb_info *sbi;
2656
2657 #ifdef HAVE_SUPEROPS_USE_DENTRY
2658         LASSERT((seq != NULL) && (dentry != NULL));
2659         sbi = ll_s2sbi(dentry->d_sb);
2660 #else
2661         LASSERT((seq != NULL) && (vfs != NULL));
2662         sbi = ll_s2sbi(vfs->mnt_sb);
2663 #endif
2664
2665         if (sbi->ll_flags & LL_SBI_NOLCK)
2666                 seq_puts(seq, ",nolock");
2667
2668         if (sbi->ll_flags & LL_SBI_FLOCK)
2669                 seq_puts(seq, ",flock");
2670
2671         if (sbi->ll_flags & LL_SBI_LOCALFLOCK)
2672                 seq_puts(seq, ",localflock");
2673
2674         if (sbi->ll_flags & LL_SBI_USER_XATTR)
2675                 seq_puts(seq, ",user_xattr");
2676
2677         if (sbi->ll_flags & LL_SBI_LAZYSTATFS)
2678                 seq_puts(seq, ",lazystatfs");
2679
2680         if (sbi->ll_flags & LL_SBI_USER_FID2PATH)
2681                 seq_puts(seq, ",user_fid2path");
2682
2683         if (sbi->ll_flags & LL_SBI_ALWAYS_PING)
2684                 seq_puts(seq, ",always_ping");
2685
2686         RETURN(0);
2687 }
2688
2689 /**
2690  * Get obd name by cmd, and copy out to user space
2691  */
2692 int ll_get_obd_name(struct inode *inode, unsigned int cmd, unsigned long arg)
2693 {
2694         struct ll_sb_info *sbi = ll_i2sbi(inode);
2695         struct obd_device *obd;
2696         ENTRY;
2697
2698         if (cmd == OBD_IOC_GETDTNAME)
2699                 obd = class_exp2obd(sbi->ll_dt_exp);
2700         else if (cmd == OBD_IOC_GETMDNAME)
2701                 obd = class_exp2obd(sbi->ll_md_exp);
2702         else
2703                 RETURN(-EINVAL);
2704
2705         if (!obd)
2706                 RETURN(-ENOENT);
2707
2708         if (copy_to_user((void __user *)arg, obd->obd_name,
2709                          strlen(obd->obd_name) + 1))
2710                 RETURN(-EFAULT);
2711
2712         RETURN(0);
2713 }
2714
2715 static char* ll_d_path(struct dentry *dentry, char *buf, int bufsize)
2716 {
2717         char *path = NULL;
2718
2719         struct path p;
2720
2721         p.dentry = dentry;
2722         p.mnt = current->fs->root.mnt;
2723         path_get(&p);
2724         path = d_path(&p, buf, bufsize);
2725         path_put(&p);
2726         return path;
2727 }
2728
2729 void ll_dirty_page_discard_warn(struct page *page, int ioret)
2730 {
2731         char *buf, *path = NULL;
2732         struct dentry *dentry = NULL;
2733         struct inode *inode = page->mapping->host;
2734
2735         /* this can be called inside spin lock so use GFP_ATOMIC. */
2736         buf = (char *)__get_free_page(GFP_ATOMIC);
2737         if (buf != NULL) {
2738                 dentry = d_find_alias(page->mapping->host);
2739                 if (dentry != NULL)
2740                         path = ll_d_path(dentry, buf, PAGE_SIZE);
2741         }
2742
2743         CDEBUG(D_WARNING,
2744                "%s: dirty page discard: %s/fid: "DFID"/%s may get corrupted "
2745                "(rc %d)\n", ll_i2sbi(inode)->ll_fsname,
2746                s2lsi(page->mapping->host->i_sb)->lsi_lmd->lmd_dev,
2747                PFID(ll_inode2fid(inode)),
2748                (path && !IS_ERR(path)) ? path : "", ioret);
2749
2750         if (dentry != NULL)
2751                 dput(dentry);
2752
2753         if (buf != NULL)
2754                 free_page((unsigned long)buf);
2755 }
2756
2757 ssize_t ll_copy_user_md(const struct lov_user_md __user *md,
2758                         struct lov_user_md **kbuf)
2759 {
2760         struct lov_user_md      lum;
2761         ssize_t                 lum_size;
2762         ENTRY;
2763
2764         if (copy_from_user(&lum, md, sizeof(lum)))
2765                 RETURN(-EFAULT);
2766
2767         lum_size = ll_lov_user_md_size(&lum);
2768         if (lum_size < 0)
2769                 RETURN(lum_size);
2770
2771         OBD_ALLOC(*kbuf, lum_size);
2772         if (*kbuf == NULL)
2773                 RETURN(-ENOMEM);
2774
2775         if (copy_from_user(*kbuf, md, lum_size) != 0) {
2776                 OBD_FREE(*kbuf, lum_size);
2777                 RETURN(-EFAULT);
2778         }
2779
2780         RETURN(lum_size);
2781 }
2782
2783 /*
2784  * Compute llite root squash state after a change of root squash
2785  * configuration setting or add/remove of a lnet nid
2786  */
2787 void ll_compute_rootsquash_state(struct ll_sb_info *sbi)
2788 {
2789         struct root_squash_info *squash = &sbi->ll_squash;
2790         int i;
2791         bool matched;
2792         struct lnet_process_id id;
2793
2794         /* Update norootsquash flag */
2795         down_write(&squash->rsi_sem);
2796         if (list_empty(&squash->rsi_nosquash_nids))
2797                 sbi->ll_flags &= ~LL_SBI_NOROOTSQUASH;
2798         else {
2799                 /* Do not apply root squash as soon as one of our NIDs is
2800                  * in the nosquash_nids list */
2801                 matched = false;
2802                 i = 0;
2803                 while (LNetGetId(i++, &id) != -ENOENT) {
2804                         if (LNET_NETTYP(LNET_NIDNET(id.nid)) == LOLND)
2805                                 continue;
2806                         if (cfs_match_nid(id.nid, &squash->rsi_nosquash_nids)) {
2807                                 matched = true;
2808                                 break;
2809                         }
2810                 }
2811                 if (matched)
2812                         sbi->ll_flags |= LL_SBI_NOROOTSQUASH;
2813                 else
2814                         sbi->ll_flags &= ~LL_SBI_NOROOTSQUASH;
2815         }
2816         up_write(&squash->rsi_sem);
2817 }
2818
2819 /**
2820  * Parse linkea content to extract information about a given hardlink
2821  *
2822  * \param[in]   ldata      - Initialized linkea data
2823  * \param[in]   linkno     - Link identifier
2824  * \param[out]  parent_fid - The entry's parent FID
2825  * \param[out]  ln         - Entry name destination buffer
2826  *
2827  * \retval 0 on success
2828  * \retval Appropriate negative error code on failure
2829  */
2830 static int ll_linkea_decode(struct linkea_data *ldata, unsigned int linkno,
2831                             struct lu_fid *parent_fid, struct lu_name *ln)
2832 {
2833         unsigned int    idx;
2834         int             rc;
2835         ENTRY;
2836
2837         rc = linkea_init_with_rec(ldata);
2838         if (rc < 0)
2839                 RETURN(rc);
2840
2841         if (linkno >= ldata->ld_leh->leh_reccount)
2842                 /* beyond last link */
2843                 RETURN(-ENODATA);
2844
2845         linkea_first_entry(ldata);
2846         for (idx = 0; ldata->ld_lee != NULL; idx++) {
2847                 linkea_entry_unpack(ldata->ld_lee, &ldata->ld_reclen, ln,
2848                                     parent_fid);
2849                 if (idx == linkno)
2850                         break;
2851
2852                 linkea_next_entry(ldata);
2853         }
2854
2855         if (idx < linkno)
2856                 RETURN(-ENODATA);
2857
2858         RETURN(0);
2859 }
2860
2861 /**
2862  * Get parent FID and name of an identified link. Operation is performed for
2863  * a given link number, letting the caller iterate over linkno to list one or
2864  * all links of an entry.
2865  *
2866  * \param[in]     file - File descriptor against which to perform the operation
2867  * \param[in,out] arg  - User-filled structure containing the linkno to operate
2868  *                       on and the available size. It is eventually filled with
2869  *                       the requested information or left untouched on error
2870  *
2871  * \retval - 0 on success
2872  * \retval - Appropriate negative error code on failure
2873  */
2874 int ll_getparent(struct file *file, struct getparent __user *arg)
2875 {
2876         struct inode            *inode = file_inode(file);
2877         struct linkea_data      *ldata;
2878         struct lu_buf            buf = LU_BUF_NULL;
2879         struct lu_name           ln;
2880         struct lu_fid            parent_fid;
2881         __u32                    linkno;
2882         __u32                    name_size;
2883         int                      rc;
2884
2885         ENTRY;
2886
2887         if (!cfs_capable(CFS_CAP_DAC_READ_SEARCH) &&
2888             !(ll_i2sbi(inode)->ll_flags & LL_SBI_USER_FID2PATH))
2889                 RETURN(-EPERM);
2890
2891         if (get_user(name_size, &arg->gp_name_size))
2892                 RETURN(-EFAULT);
2893
2894         if (get_user(linkno, &arg->gp_linkno))
2895                 RETURN(-EFAULT);
2896
2897         if (name_size > PATH_MAX)
2898                 RETURN(-EINVAL);
2899
2900         OBD_ALLOC(ldata, sizeof(*ldata));
2901         if (ldata == NULL)
2902                 RETURN(-ENOMEM);
2903
2904         rc = linkea_data_new(ldata, &buf);
2905         if (rc < 0)
2906                 GOTO(ldata_free, rc);
2907
2908 #ifdef HAVE_XATTR_HANDLER_FLAGS
2909         rc = ll_xattr_list(inode, XATTR_NAME_LINK, XATTR_TRUSTED_T, buf.lb_buf,
2910                            buf.lb_len, OBD_MD_FLXATTR);
2911 #else
2912         rc = ll_getxattr(file_dentry(file), XATTR_NAME_LINK, buf.lb_buf,
2913                          buf.lb_len);
2914 #endif /* HAVE_XATTR_HANDLER_FLAGS */
2915         if (rc < 0)
2916                 GOTO(lb_free, rc);
2917
2918         rc = ll_linkea_decode(ldata, linkno, &parent_fid, &ln);
2919         if (rc < 0)
2920                 GOTO(lb_free, rc);
2921
2922         if (ln.ln_namelen >= name_size)
2923                 GOTO(lb_free, rc = -EOVERFLOW);
2924
2925         if (copy_to_user(&arg->gp_fid, &parent_fid, sizeof(arg->gp_fid)))
2926                 GOTO(lb_free, rc = -EFAULT);
2927
2928         if (copy_to_user(&arg->gp_name, ln.ln_name, ln.ln_namelen))
2929                 GOTO(lb_free, rc = -EFAULT);
2930
2931         if (put_user('\0', arg->gp_name + ln.ln_namelen))
2932                 GOTO(lb_free, rc = -EFAULT);
2933
2934 lb_free:
2935         lu_buf_free(&buf);
2936 ldata_free:
2937         OBD_FREE(ldata, sizeof(*ldata));
2938
2939         RETURN(rc);
2940 }