lustre/llite/llite_lib.c

   1 /*
   2  * GPL HEADER START
   3  *
   4  * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
   5  *
   6  * This program is free software; you can redistribute it and/or modify
   7  * it under the terms of the GNU General Public License version 2 only,
   8  * as published by the Free Software Foundation.
   9  *
  10  * This program is distributed in the hope that it will be useful, but
  11  * WITHOUT ANY WARRANTY; without even the implied warranty of
  12  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  13  * General Public License version 2 for more details (a copy is included
  14  * in the LICENSE file that accompanied this code).
  15  *
  16  * You should have received a copy of the GNU General Public License
  17  * version 2 along with this program; If not, see
  18  * http://www.gnu.org/licenses/gpl-2.0.html
  19  *
  20  * GPL HEADER END
  21  */
  22 /*
  23  * Copyright (c) 2003, 2010, Oracle and/or its affiliates. All rights reserved.
  24  * Use is subject to license terms.
  25  *
  26  * Copyright (c) 2011, 2017, Intel Corporation.
  27  */
  28 /*
  29  * This file is part of Lustre, http://www.lustre.org/
  30  * Lustre is a trademark of Sun Microsystems, Inc.
  31  *
  32  * lustre/llite/llite_lib.c
  33  *
  34  * Lustre Light Super operations
  35  */
  36
  37 #define DEBUG_SUBSYSTEM S_LLITE
  38
  39 #include <linux/module.h>
  40 #include <linux/statfs.h>
  41 #include <linux/time.h>
  42 #include <linux/types.h>
  43 #include <linux/version.h>
  44 #include <linux/mm.h>
  45 #include <linux/user_namespace.h>
  46 #ifdef HAVE_UIDGID_HEADER
  47 # include <linux/uidgid.h>
  48 #endif
  49 #include <linux/security.h>
  50
  51 #include <uapi/linux/lustre/lustre_ioctl.h>
  52 #include <lustre_ha.h>
  53 #include <lustre_dlm.h>
  54 #include <lprocfs_status.h>
  55 #include <lustre_disk.h>
  56 #include <uapi/linux/lustre/lustre_param.h>
  57 #include <lustre_log.h>
  58 #include <cl_object.h>
  59 #include <obd_cksum.h>
  60 #include "llite_internal.h"
  61
  62 struct kmem_cache *ll_file_data_slab;
  63
  64 #ifndef log2
  65 #define log2(n) ffz(~(n))
  66 #endif
  67
  68 static struct ll_sb_info *ll_init_sbi(void)
  69 {
  70         struct ll_sb_info *sbi = NULL;
  71         unsigned long pages;
  72         unsigned long lru_page_max;
  73         struct sysinfo si;
  74         class_uuid_t uuid;
  75         int i;
  76         ENTRY;
  77
  78         OBD_ALLOC_PTR(sbi);
  79         if (sbi == NULL)
  80                 RETURN(NULL);
  81
  82         spin_lock_init(&sbi->ll_lock);
  83         mutex_init(&sbi->ll_lco.lco_lock);
  84         spin_lock_init(&sbi->ll_pp_extent_lock);
  85         spin_lock_init(&sbi->ll_process_lock);
  86         sbi->ll_rw_stats_on = 0;
  87
  88         si_meminfo(&si);
  89         pages = si.totalram - si.totalhigh;
  90         lru_page_max = pages / 2;
  91
  92         /* initialize ll_cache data */
  93         sbi->ll_cache = cl_cache_init(lru_page_max);
  94         if (sbi->ll_cache == NULL) {
  95                 OBD_FREE(sbi, sizeof(*sbi));
  96                 RETURN(NULL);
  97         }
  98
  99         sbi->ll_ra_info.ra_max_pages_per_file = min(pages / 32,
 100                                            SBI_DEFAULT_READAHEAD_MAX);
 101         sbi->ll_ra_info.ra_max_pages = sbi->ll_ra_info.ra_max_pages_per_file;
 102         sbi->ll_ra_info.ra_max_read_ahead_whole_pages = -1;
 103
 104         ll_generate_random_uuid(uuid);
 105         class_uuid_unparse(uuid, &sbi->ll_sb_uuid);
 106         CDEBUG(D_CONFIG, "generated uuid: %s\n", sbi->ll_sb_uuid.uuid);
 107
 108         sbi->ll_flags |= LL_SBI_VERBOSE;
 109 #ifdef ENABLE_CHECKSUM
 110         sbi->ll_flags |= LL_SBI_CHECKSUM;
 111 #endif
 112
 113 #ifdef HAVE_LRU_RESIZE_SUPPORT
 114         sbi->ll_flags |= LL_SBI_LRU_RESIZE;
 115 #endif
 116         sbi->ll_flags |= LL_SBI_LAZYSTATFS;
 117
 118         for (i = 0; i <= LL_PROCESS_HIST_MAX; i++) {
 119                 spin_lock_init(&sbi->ll_rw_extents_info.pp_extents[i].
 120                                pp_r_hist.oh_lock);
 121                 spin_lock_init(&sbi->ll_rw_extents_info.pp_extents[i].
 122                                pp_w_hist.oh_lock);
 123         }
 124
 125         /* metadata statahead is enabled by default */
 126         sbi->ll_sa_running_max = LL_SA_RUNNING_DEF;
 127         sbi->ll_sa_max = LL_SA_RPC_DEF;
 128         atomic_set(&sbi->ll_sa_total, 0);
 129         atomic_set(&sbi->ll_sa_wrong, 0);
 130         atomic_set(&sbi->ll_sa_running, 0);
 131         atomic_set(&sbi->ll_agl_total, 0);
 132         sbi->ll_flags |= LL_SBI_AGL_ENABLED;
 133         sbi->ll_flags |= LL_SBI_FAST_READ;
 134         sbi->ll_flags |= LL_SBI_TINY_WRITE;
 135
 136         /* root squash */
 137         sbi->ll_squash.rsi_uid = 0;
 138         sbi->ll_squash.rsi_gid = 0;
 139         INIT_LIST_HEAD(&sbi->ll_squash.rsi_nosquash_nids);
 140         init_rwsem(&sbi->ll_squash.rsi_sem);
 141
 142         RETURN(sbi);
 143 }
 144
 145 static void ll_free_sbi(struct super_block *sb)
 146 {
 147         struct ll_sb_info *sbi = ll_s2sbi(sb);
 148         ENTRY;
 149
 150         if (sbi != NULL) {
 151                 if (!list_empty(&sbi->ll_squash.rsi_nosquash_nids))
 152                         cfs_free_nidlist(&sbi->ll_squash.rsi_nosquash_nids);
 153                 if (sbi->ll_cache != NULL) {
 154                         cl_cache_decref(sbi->ll_cache);
 155                         sbi->ll_cache = NULL;
 156                 }
 157                 OBD_FREE(sbi, sizeof(*sbi));
 158         }
 159         EXIT;
 160 }
 161
 162 static inline int obd_connect_has_secctx(struct obd_connect_data *data)
 163 {
 164         return data->ocd_connect_flags & OBD_CONNECT_FLAGS2 &&
 165                data->ocd_connect_flags2 & OBD_CONNECT2_FILE_SECCTX;
 166 }
 167
 168 static int client_common_fill_super(struct super_block *sb, char *md, char *dt,
 169                                     struct vfsmount *mnt)
 170 {
 171         struct inode *root = NULL;
 172         struct ll_sb_info *sbi = ll_s2sbi(sb);
 173         struct obd_statfs *osfs = NULL;
 174         struct ptlrpc_request *request = NULL;
 175         struct obd_connect_data *data = NULL;
 176         struct obd_uuid *uuid;
 177         struct md_op_data *op_data;
 178         struct lustre_md lmd;
 179         u64 valid;
 180         int size, err, checksum;
 181
 182         ENTRY;
 183         sbi->ll_md_obd = class_name2obd(md);
 184         if (!sbi->ll_md_obd) {
 185                 CERROR("MD %s: not setup or attached\n", md);
 186                 RETURN(-EINVAL);
 187         }
 188
 189         OBD_ALLOC_PTR(data);
 190         if (data == NULL)
 191                 RETURN(-ENOMEM);
 192
 193         OBD_ALLOC_PTR(osfs);
 194         if (osfs == NULL) {
 195                 OBD_FREE_PTR(data);
 196                 RETURN(-ENOMEM);
 197         }
 198
 199         /* pass client page size via ocd_grant_blkbits, the server should report
 200          * back its backend blocksize for grant calculation purpose */
 201         data->ocd_grant_blkbits = PAGE_SHIFT;
 202
 203         /* indicate MDT features supported by this client */
 204         data->ocd_connect_flags = OBD_CONNECT_IBITS    | OBD_CONNECT_NODEVOH  |
 205                                   OBD_CONNECT_ATTRFID  | OBD_CONNECT_GRANT |
 206                                   OBD_CONNECT_VERSION  | OBD_CONNECT_BRW_SIZE |
 207                                   OBD_CONNECT_SRVLOCK  | OBD_CONNECT_TRUNCLOCK|
 208                                   OBD_CONNECT_MDS_CAPA | OBD_CONNECT_OSS_CAPA |
 209                                   OBD_CONNECT_CANCELSET | OBD_CONNECT_FID     |
 210                                   OBD_CONNECT_AT       | OBD_CONNECT_LOV_V3   |
 211                                   OBD_CONNECT_VBR | OBD_CONNECT_FULL20 |
 212                                   OBD_CONNECT_64BITHASH |
 213                                   OBD_CONNECT_EINPROGRESS |
 214                                   OBD_CONNECT_JOBSTATS | OBD_CONNECT_LVB_TYPE |
 215                                   OBD_CONNECT_LAYOUTLOCK | OBD_CONNECT_PINGLESS|
 216                                   OBD_CONNECT_MAX_EASIZE |
 217                                   OBD_CONNECT_FLOCK_DEAD |
 218                                   OBD_CONNECT_DISP_STRIPE | OBD_CONNECT_LFSCK |
 219                                   OBD_CONNECT_OPEN_BY_FID |
 220                                   OBD_CONNECT_DIR_STRIPE |
 221                                   OBD_CONNECT_BULK_MBITS | OBD_CONNECT_CKSUM |
 222                                   OBD_CONNECT_SUBTREE |
 223                                   OBD_CONNECT_MULTIMODRPCS |
 224                                   OBD_CONNECT_GRANT_PARAM |
 225                                   OBD_CONNECT_SHORTIO | OBD_CONNECT_FLAGS2;
 226
 227         data->ocd_connect_flags2 = OBD_CONNECT2_FLR |
 228                                    OBD_CONNECT2_LOCK_CONVERT |
 229                                    OBD_CONNECT2_DIR_MIGRATE |
 230                                    OBD_CONNECT2_SUM_STATFS;
 231
 232 #ifdef HAVE_LRU_RESIZE_SUPPORT
 233         if (sbi->ll_flags & LL_SBI_LRU_RESIZE)
 234                 data->ocd_connect_flags |= OBD_CONNECT_LRU_RESIZE;
 235 #endif
 236 #ifdef CONFIG_FS_POSIX_ACL
 237         data->ocd_connect_flags |= OBD_CONNECT_ACL | OBD_CONNECT_UMASK |
 238                                    OBD_CONNECT_LARGE_ACL;
 239 #endif
 240
 241         data->ocd_cksum_types = obd_cksum_types_supported_client();
 242
 243         if (OBD_FAIL_CHECK(OBD_FAIL_MDC_LIGHTWEIGHT))
 244                 /* flag mdc connection as lightweight, only used for test
 245                  * purpose, use with care */
 246                 data->ocd_connect_flags |= OBD_CONNECT_LIGHTWEIGHT;
 247
 248         data->ocd_ibits_known = MDS_INODELOCK_FULL;
 249         data->ocd_version = LUSTRE_VERSION_CODE;
 250
 251         if (sb->s_flags & MS_RDONLY)
 252                 data->ocd_connect_flags |= OBD_CONNECT_RDONLY;
 253         if (sbi->ll_flags & LL_SBI_USER_XATTR)
 254                 data->ocd_connect_flags |= OBD_CONNECT_XATTR;
 255
 256 #ifdef MS_NOSEC
 257         /* Setting this indicates we correctly support S_NOSEC (See kernel
 258          * commit 9e1f1de02c2275d7172e18dc4e7c2065777611bf)
 259          */
 260         sb->s_flags |= MS_NOSEC;
 261 #endif
 262
 263         if (sbi->ll_flags & LL_SBI_FLOCK)
 264                 sbi->ll_fop = &ll_file_operations_flock;
 265         else if (sbi->ll_flags & LL_SBI_LOCALFLOCK)
 266                 sbi->ll_fop = &ll_file_operations;
 267         else
 268                 sbi->ll_fop = &ll_file_operations_noflock;
 269
 270         /* always ping even if server suppress_pings */
 271         if (sbi->ll_flags & LL_SBI_ALWAYS_PING)
 272                 data->ocd_connect_flags &= ~OBD_CONNECT_PINGLESS;
 273
 274 #if defined(HAVE_SECURITY_DENTRY_INIT_SECURITY) && defined(CONFIG_SECURITY)
 275         data->ocd_connect_flags2 |= OBD_CONNECT2_FILE_SECCTX;
 276 #endif /* HAVE_SECURITY_DENTRY_INIT_SECURITY */
 277
 278         data->ocd_brw_size = MD_MAX_BRW_SIZE;
 279
 280         err = obd_connect(NULL, &sbi->ll_md_exp, sbi->ll_md_obd,
 281                           &sbi->ll_sb_uuid, data, NULL);
 282         if (err == -EBUSY) {
 283                 LCONSOLE_ERROR_MSG(0x14f, "An MDT (md %s) is performing "
 284                                    "recovery, of which this client is not a "
 285                                    "part. Please wait for recovery to complete,"
 286                                    " abort, or time out.\n", md);
 287                 GOTO(out, err);
 288         } else if (err) {
 289                 CERROR("cannot connect to %s: rc = %d\n", md, err);
 290                 GOTO(out, err);
 291         }
 292
 293         sbi->ll_md_exp->exp_connect_data = *data;
 294
 295         err = obd_fid_init(sbi->ll_md_exp->exp_obd, sbi->ll_md_exp,
 296                            LUSTRE_SEQ_METADATA);
 297         if (err) {
 298                 CERROR("%s: Can't init metadata layer FID infrastructure, "
 299                        "rc = %d\n", sbi->ll_md_exp->exp_obd->obd_name, err);
 300                 GOTO(out_md, err);
 301         }
 302
 303         /* For mount, we only need fs info from MDT0, and also in DNE, it
 304          * can make sure the client can be mounted as long as MDT0 is
 305          * avaible */
 306         err = obd_statfs(NULL, sbi->ll_md_exp, osfs,
 307                         ktime_get_seconds() -OBD_STATFS_CACHE_SECONDS,
 308                         OBD_STATFS_FOR_MDT0);
 309         if (err)
 310                 GOTO(out_md_fid, err);
 311
 312         /* This needs to be after statfs to ensure connect has finished.
 313          * Note that "data" does NOT contain the valid connect reply.
 314          * If connecting to a 1.8 server there will be no LMV device, so
 315          * we can access the MDC export directly and exp_connect_flags will
 316          * be non-zero, but if accessing an upgraded 2.1 server it will
 317          * have the correct flags filled in.
 318          * XXX: fill in the LMV exp_connect_flags from MDC(s). */
 319         valid = exp_connect_flags(sbi->ll_md_exp) & CLIENT_CONNECT_MDT_REQD;
 320         if (exp_connect_flags(sbi->ll_md_exp) != 0 &&
 321             valid != CLIENT_CONNECT_MDT_REQD) {
 322                 char *buf;
 323
 324                 OBD_ALLOC_WAIT(buf, PAGE_SIZE);
 325                 obd_connect_flags2str(buf, PAGE_SIZE,
 326                                       valid ^ CLIENT_CONNECT_MDT_REQD, 0, ",");
 327                 LCONSOLE_ERROR_MSG(0x170, "Server %s does not support "
 328                                    "feature(s) needed for correct operation "
 329                                    "of this client (%s). Please upgrade "
 330                                    "server or downgrade client.\n",
 331                                    sbi->ll_md_exp->exp_obd->obd_name, buf);
 332                 OBD_FREE(buf, PAGE_SIZE);
 333                 GOTO(out_md_fid, err = -EPROTO);
 334         }
 335
 336         size = sizeof(*data);
 337         err = obd_get_info(NULL, sbi->ll_md_exp, sizeof(KEY_CONN_DATA),
 338                            KEY_CONN_DATA,  &size, data);
 339         if (err) {
 340                 CERROR("%s: Get connect data failed: rc = %d\n",
 341                        sbi->ll_md_exp->exp_obd->obd_name, err);
 342                 GOTO(out_md_fid, err);
 343         }
 344
 345         LASSERT(osfs->os_bsize);
 346         sb->s_blocksize = osfs->os_bsize;
 347         sb->s_blocksize_bits = log2(osfs->os_bsize);
 348         sb->s_magic = LL_SUPER_MAGIC;
 349         sb->s_maxbytes = MAX_LFS_FILESIZE;
 350         sbi->ll_namelen = osfs->os_namelen;
 351         sbi->ll_mnt.mnt = current->fs->root.mnt;
 352
 353         if ((sbi->ll_flags & LL_SBI_USER_XATTR) &&
 354             !(data->ocd_connect_flags & OBD_CONNECT_XATTR)) {
 355                 LCONSOLE_INFO("Disabling user_xattr feature because "
 356                               "it is not supported on the server\n");
 357                 sbi->ll_flags &= ~LL_SBI_USER_XATTR;
 358         }
 359
 360         if (data->ocd_connect_flags & OBD_CONNECT_ACL) {
 361 #ifdef MS_POSIXACL
 362                 sb->s_flags |= MS_POSIXACL;
 363 #endif
 364                 sbi->ll_flags |= LL_SBI_ACL;
 365         } else {
 366                 LCONSOLE_INFO("client wants to enable acl, but mdt not!\n");
 367 #ifdef MS_POSIXACL
 368                 sb->s_flags &= ~MS_POSIXACL;
 369 #endif
 370                 sbi->ll_flags &= ~LL_SBI_ACL;
 371         }
 372
 373         if (data->ocd_connect_flags & OBD_CONNECT_64BITHASH)
 374                 sbi->ll_flags |= LL_SBI_64BIT_HASH;
 375
 376         if (data->ocd_connect_flags & OBD_CONNECT_LAYOUTLOCK)
 377                 sbi->ll_flags |= LL_SBI_LAYOUT_LOCK;
 378
 379         if (obd_connect_has_secctx(data))
 380                 sbi->ll_flags |= LL_SBI_FILE_SECCTX;
 381
 382         if (data->ocd_ibits_known & MDS_INODELOCK_XATTR) {
 383                 if (!(data->ocd_connect_flags & OBD_CONNECT_MAX_EASIZE)) {
 384                         LCONSOLE_INFO("%s: disabling xattr cache due to "
 385                                       "unknown maximum xattr size.\n", dt);
 386                 } else if (!sbi->ll_xattr_cache_set) {
 387                         /* If xattr_cache is already set (no matter 0 or 1)
 388                          * during processing llog, it won't be enabled here. */
 389                         sbi->ll_flags |= LL_SBI_XATTR_CACHE;
 390                         sbi->ll_xattr_cache_enabled = 1;
 391                 }
 392         }
 393
 394         sbi->ll_dt_obd = class_name2obd(dt);
 395         if (!sbi->ll_dt_obd) {
 396                 CERROR("DT %s: not setup or attached\n", dt);
 397                 GOTO(out_md_fid, err = -ENODEV);
 398         }
 399
 400         /* pass client page size via ocd_grant_blkbits, the server should report
 401          * back its backend blocksize for grant calculation purpose */
 402         data->ocd_grant_blkbits = PAGE_SHIFT;
 403
 404         /* indicate OST features supported by this client */
 405         data->ocd_connect_flags = OBD_CONNECT_GRANT | OBD_CONNECT_VERSION |
 406                                   OBD_CONNECT_REQPORTAL | OBD_CONNECT_BRW_SIZE |
 407                                   OBD_CONNECT_CANCELSET | OBD_CONNECT_FID |
 408                                   OBD_CONNECT_SRVLOCK | OBD_CONNECT_TRUNCLOCK|
 409                                   OBD_CONNECT_AT | OBD_CONNECT_OSS_CAPA |
 410                                   OBD_CONNECT_VBR | OBD_CONNECT_FULL20 |
 411                                   OBD_CONNECT_64BITHASH | OBD_CONNECT_MAXBYTES |
 412                                   OBD_CONNECT_EINPROGRESS |
 413                                   OBD_CONNECT_JOBSTATS | OBD_CONNECT_LVB_TYPE |
 414                                   OBD_CONNECT_LAYOUTLOCK |
 415                                   OBD_CONNECT_PINGLESS | OBD_CONNECT_LFSCK |
 416                                   OBD_CONNECT_BULK_MBITS | OBD_CONNECT_SHORTIO |
 417                                   OBD_CONNECT_FLAGS2 | OBD_CONNECT_GRANT_SHRINK;
 418
 419 /* The client currently advertises support for OBD_CONNECT_LOCKAHEAD_OLD so it
 420  * can interoperate with an older version of lockahead which was released prior
 421  * to landing in master. This support will be dropped when 2.13 development
 422  * starts.  At the point, we should not just drop the connect flag (below), we
 423  * should also remove the support in the code.
 424  *
 425  * Removing it means a few things:
 426  * 1. Remove this section here
 427  * 2. Remove CEF_NONBLOCK in ll_file_lockahead()
 428  * 3. Remove function exp_connect_lockahead_old
 429  * 4. Remove LDLM_FL_LOCKAHEAD_OLD_RESERVED in lustre_dlm_flags.h
 430  * */
 431 #if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(2, 12, 50, 0)
 432         data->ocd_connect_flags |= OBD_CONNECT_LOCKAHEAD_OLD;
 433 #endif
 434
 435         data->ocd_connect_flags2 = OBD_CONNECT2_LOCKAHEAD;
 436
 437         if (!OBD_FAIL_CHECK(OBD_FAIL_OSC_CONNECT_GRANT_PARAM))
 438                 data->ocd_connect_flags |= OBD_CONNECT_GRANT_PARAM;
 439
 440         /* OBD_CONNECT_CKSUM should always be set, even if checksums are
 441          * disabled by default, because it can still be enabled on the
 442          * fly via /proc. As a consequence, we still need to come to an
 443          * agreement on the supported algorithms at connect time */
 444         data->ocd_connect_flags |= OBD_CONNECT_CKSUM;
 445
 446         if (OBD_FAIL_CHECK(OBD_FAIL_OSC_CKSUM_ADLER_ONLY))
 447                 data->ocd_cksum_types = OBD_CKSUM_ADLER;
 448         else
 449                 data->ocd_cksum_types = obd_cksum_types_supported_client();
 450
 451 #ifdef HAVE_LRU_RESIZE_SUPPORT
 452         data->ocd_connect_flags |= OBD_CONNECT_LRU_RESIZE;
 453 #endif
 454         /* always ping even if server suppress_pings */
 455         if (sbi->ll_flags & LL_SBI_ALWAYS_PING)
 456                 data->ocd_connect_flags &= ~OBD_CONNECT_PINGLESS;
 457
 458         CDEBUG(D_RPCTRACE, "ocd_connect_flags: %#llx ocd_version: %d "
 459                "ocd_grant: %d\n", data->ocd_connect_flags,
 460                data->ocd_version, data->ocd_grant);
 461
 462         sbi->ll_dt_obd->obd_upcall.onu_owner = &sbi->ll_lco;
 463         sbi->ll_dt_obd->obd_upcall.onu_upcall = cl_ocd_update;
 464
 465         data->ocd_brw_size = DT_MAX_BRW_SIZE;
 466
 467         err = obd_connect(NULL, &sbi->ll_dt_exp, sbi->ll_dt_obd,
 468                           &sbi->ll_sb_uuid, data, NULL);
 469         if (err == -EBUSY) {
 470                 LCONSOLE_ERROR_MSG(0x150, "An OST (dt %s) is performing "
 471                                    "recovery, of which this client is not a "
 472                                    "part.  Please wait for recovery to "
 473                                    "complete, abort, or time out.\n", dt);
 474                 GOTO(out_md, err);
 475         } else if (err) {
 476                 CERROR("%s: Cannot connect to %s: rc = %d\n",
 477                        sbi->ll_dt_exp->exp_obd->obd_name, dt, err);
 478                 GOTO(out_md, err);
 479         }
 480
 481         sbi->ll_dt_exp->exp_connect_data = *data;
 482
 483         /* Don't change value if it was specified in the config log */
 484         if (sbi->ll_ra_info.ra_max_read_ahead_whole_pages == -1)
 485                 sbi->ll_ra_info.ra_max_read_ahead_whole_pages =
 486                         max_t(unsigned long, SBI_DEFAULT_READAHEAD_WHOLE_MAX,
 487                               (data->ocd_brw_size >> PAGE_SHIFT));
 488
 489         err = obd_fid_init(sbi->ll_dt_exp->exp_obd, sbi->ll_dt_exp,
 490                            LUSTRE_SEQ_METADATA);
 491         if (err) {
 492                 CERROR("%s: Can't init data layer FID infrastructure, "
 493                        "rc = %d\n", sbi->ll_dt_exp->exp_obd->obd_name, err);
 494                 GOTO(out_dt, err);
 495         }
 496
 497         mutex_lock(&sbi->ll_lco.lco_lock);
 498         sbi->ll_lco.lco_flags = data->ocd_connect_flags;
 499         sbi->ll_lco.lco_md_exp = sbi->ll_md_exp;
 500         sbi->ll_lco.lco_dt_exp = sbi->ll_dt_exp;
 501         mutex_unlock(&sbi->ll_lco.lco_lock);
 502
 503         fid_zero(&sbi->ll_root_fid);
 504         err = md_get_root(sbi->ll_md_exp, get_mount_fileset(sb),
 505                            &sbi->ll_root_fid);
 506         if (err) {
 507                 CERROR("cannot mds_connect: rc = %d\n", err);
 508                 GOTO(out_lock_cn_cb, err);
 509         }
 510         if (!fid_is_sane(&sbi->ll_root_fid)) {
 511                 CERROR("%s: Invalid root fid "DFID" during mount\n",
 512                        sbi->ll_md_exp->exp_obd->obd_name,
 513                        PFID(&sbi->ll_root_fid));
 514                 GOTO(out_lock_cn_cb, err = -EINVAL);
 515         }
 516         CDEBUG(D_SUPER, "rootfid "DFID"\n", PFID(&sbi->ll_root_fid));
 517
 518         sb->s_op = &lustre_super_operations;
 519 #ifdef HAVE_XATTR_HANDLER_FLAGS
 520         sb->s_xattr = ll_xattr_handlers;
 521 #endif
 522 #if THREAD_SIZE >= 8192 /*b=17630*/
 523         sb->s_export_op = &lustre_export_operations;
 524 #endif
 525
 526         /* make root inode
 527          * XXX: move this to after cbd setup? */
 528         valid = OBD_MD_FLGETATTR | OBD_MD_FLBLOCKS | OBD_MD_FLMODEASIZE;
 529         if (sbi->ll_flags & LL_SBI_ACL)
 530                 valid |= OBD_MD_FLACL;
 531
 532         OBD_ALLOC_PTR(op_data);
 533         if (op_data == NULL)
 534                 GOTO(out_lock_cn_cb, err = -ENOMEM);
 535
 536         op_data->op_fid1 = sbi->ll_root_fid;
 537         op_data->op_mode = 0;
 538         op_data->op_valid = valid;
 539
 540         err = md_getattr(sbi->ll_md_exp, op_data, &request);
 541
 542         OBD_FREE_PTR(op_data);
 543         if (err) {
 544                 CERROR("%s: md_getattr failed for root: rc = %d\n",
 545                        sbi->ll_md_exp->exp_obd->obd_name, err);
 546                 GOTO(out_lock_cn_cb, err);
 547         }
 548
 549         err = md_get_lustre_md(sbi->ll_md_exp, request, sbi->ll_dt_exp,
 550                                sbi->ll_md_exp, &lmd);
 551         if (err) {
 552                 CERROR("failed to understand root inode md: rc = %d\n", err);
 553                 ptlrpc_req_finished(request);
 554                 GOTO(out_lock_cn_cb, err);
 555         }
 556
 557         LASSERT(fid_is_sane(&sbi->ll_root_fid));
 558         root = ll_iget(sb, cl_fid_build_ino(&sbi->ll_root_fid,
 559                                             sbi->ll_flags & LL_SBI_32BIT_API),
 560                        &lmd);
 561         md_free_lustre_md(sbi->ll_md_exp, &lmd);
 562         ptlrpc_req_finished(request);
 563
 564         if (IS_ERR(root)) {
 565 #ifdef CONFIG_FS_POSIX_ACL
 566                 if (lmd.posix_acl) {
 567                         posix_acl_release(lmd.posix_acl);
 568                         lmd.posix_acl = NULL;
 569                 }
 570 #endif
 571                 err = IS_ERR(root) ? PTR_ERR(root) : -EBADF;
 572                 root = NULL;
 573                 CERROR("lustre_lite: bad iget4 for root\n");
 574                 GOTO(out_root, err);
 575         }
 576
 577         checksum = sbi->ll_flags & LL_SBI_CHECKSUM;
 578         if (sbi->ll_checksum_set) {
 579                 err = obd_set_info_async(NULL, sbi->ll_dt_exp,
 580                                          sizeof(KEY_CHECKSUM), KEY_CHECKSUM,
 581                                          sizeof(checksum), &checksum, NULL);
 582                 if (err) {
 583                         CERROR("%s: Set checksum failed: rc = %d\n",
 584                                sbi->ll_dt_exp->exp_obd->obd_name, err);
 585                         GOTO(out_root, err);
 586                 }
 587         }
 588         cl_sb_init(sb);
 589
 590         err = obd_set_info_async(NULL, sbi->ll_dt_exp, sizeof(KEY_CACHE_SET),
 591                                  KEY_CACHE_SET, sizeof(*sbi->ll_cache),
 592                                  sbi->ll_cache, NULL);
 593         if (err) {
 594                 CERROR("%s: Set cache_set failed: rc = %d\n",
 595                        sbi->ll_dt_exp->exp_obd->obd_name, err);
 596                 GOTO(out_root, err);
 597         }
 598
 599         sb->s_root = d_make_root(root);
 600         if (sb->s_root == NULL) {
 601                 CERROR("%s: can't make root dentry\n",
 602                         ll_get_fsname(sb, NULL, 0));
 603                 GOTO(out_root, err = -ENOMEM);
 604         }
 605 #ifdef HAVE_DCACHE_LOCK
 606         sb->s_root->d_op = &ll_d_ops;
 607 #endif
 608
 609         sbi->ll_sdev_orig = sb->s_dev;
 610
 611         /* We set sb->s_dev equal on all lustre clients in order to support
 612          * NFS export clustering.  NFSD requires that the FSID be the same
 613          * on all clients. */
 614         /* s_dev is also used in lt_compare() to compare two fs, but that is
 615          * only a node-local comparison. */
 616         uuid = obd_get_uuid(sbi->ll_md_exp);
 617         if (uuid != NULL)
 618                 sb->s_dev = get_uuid2int(uuid->uuid, strlen(uuid->uuid));
 619
 620         if (data != NULL)
 621                 OBD_FREE_PTR(data);
 622         if (osfs != NULL)
 623                 OBD_FREE_PTR(osfs);
 624
 625         if (sbi->ll_dt_obd) {
 626                 err = sysfs_create_link(&sbi->ll_kset.kobj,
 627                                         &sbi->ll_dt_obd->obd_kset.kobj,
 628                                         sbi->ll_dt_obd->obd_type->typ_name);
 629                 if (err < 0) {
 630                         CERROR("%s: could not register %s in llite: rc = %d\n",
 631                                dt, ll_get_fsname(sb, NULL, 0), err);
 632                         err = 0;
 633                 }
 634         }
 635
 636         if (sbi->ll_md_obd) {
 637                 err = sysfs_create_link(&sbi->ll_kset.kobj,
 638                                         &sbi->ll_md_obd->obd_kset.kobj,
 639                                         sbi->ll_md_obd->obd_type->typ_name);
 640                 if (err < 0) {
 641                         CERROR("%s: could not register %s in llite: rc = %d\n",
 642                                md, ll_get_fsname(sb, NULL, 0), err);
 643                         err = 0;
 644                 }
 645         }
 646
 647         RETURN(err);
 648 out_root:
 649         if (root)
 650                 iput(root);
 651 out_lock_cn_cb:
 652         obd_fid_fini(sbi->ll_dt_exp->exp_obd);
 653 out_dt:
 654         obd_disconnect(sbi->ll_dt_exp);
 655         sbi->ll_dt_exp = NULL;
 656         sbi->ll_dt_obd = NULL;
 657 out_md_fid:
 658         obd_fid_fini(sbi->ll_md_exp->exp_obd);
 659 out_md:
 660         obd_disconnect(sbi->ll_md_exp);
 661         sbi->ll_md_exp = NULL;
 662         sbi->ll_md_obd = NULL;
 663 out:
 664         if (data != NULL)
 665                 OBD_FREE_PTR(data);
 666         if (osfs != NULL)
 667                 OBD_FREE_PTR(osfs);
 668         return err;
 669 }
 670
 671 int ll_get_max_mdsize(struct ll_sb_info *sbi, int *lmmsize)
 672 {
 673         int size, rc;
 674
 675         size = sizeof(*lmmsize);
 676         rc = obd_get_info(NULL, sbi->ll_dt_exp, sizeof(KEY_MAX_EASIZE),
 677                           KEY_MAX_EASIZE, &size, lmmsize);
 678         if (rc != 0) {
 679                 CERROR("%s: cannot get max LOV EA size: rc = %d\n",
 680                        sbi->ll_dt_exp->exp_obd->obd_name, rc);
 681                 RETURN(rc);
 682         }
 683
 684         size = sizeof(int);
 685         rc = obd_get_info(NULL, sbi->ll_md_exp, sizeof(KEY_MAX_EASIZE),
 686                           KEY_MAX_EASIZE, &size, lmmsize);
 687         if (rc)
 688                 CERROR("Get max mdsize error rc %d\n", rc);
 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         } else {
 965                 mutex_init(&lli->lli_size_mutex);
 966                 lli->lli_symlink_name = NULL;
 967                 init_rwsem(&lli->lli_trunc_sem);
 968                 range_lock_tree_init(&lli->lli_write_tree);
 969                 init_rwsem(&lli->lli_glimpse_sem);
 970                 lli->lli_glimpse_time = ktime_set(0, 0);
 971                 INIT_LIST_HEAD(&lli->lli_agl_list);
 972                 lli->lli_agl_index = 0;
 973                 lli->lli_async_rc = 0;
 974         }
 975         mutex_init(&lli->lli_layout_mutex);
 976         memset(lli->lli_jobid, 0, sizeof(lli->lli_jobid));
 977 }
 978
 979 #define MAX_STRING_SIZE 128
 980
 981 #ifndef HAVE_SUPER_SETUP_BDI_NAME
 982
 983 #define LSI_BDI_INITIALIZED     0x00400000
 984
 985 #ifndef HAVE_BDI_CAP_MAP_COPY
 986 # define BDI_CAP_MAP_COPY       0
 987 #endif
 988
 989 static int super_setup_bdi_name(struct super_block *sb, char *fmt, ...)
 990 {
 991         struct  lustre_sb_info *lsi = s2lsi(sb);
 992         char buf[MAX_STRING_SIZE];
 993         va_list args;
 994         int err;
 995
 996         err = bdi_init(&lsi->lsi_bdi);
 997         if (err)
 998                 return err;
 999
1000         lsi->lsi_flags |= LSI_BDI_INITIALIZED;
1001         lsi->lsi_bdi.capabilities = BDI_CAP_MAP_COPY;
1002         lsi->lsi_bdi.name = "lustre";
1003         va_start(args, fmt);
1004         vsnprintf(buf, MAX_STRING_SIZE, fmt, args);
1005         va_end(args);
1006         err = bdi_register(&lsi->lsi_bdi, NULL, "%s", buf);
1007         va_end(args);
1008         if (!err)
1009                 sb->s_bdi = &lsi->lsi_bdi;
1010
1011         return err;
1012 }
1013 #endif /* !HAVE_SUPER_SETUP_BDI_NAME */
1014
1015 int ll_fill_super(struct super_block *sb, struct vfsmount *mnt)
1016 {
1017         struct  lustre_profile *lprof = NULL;
1018         struct  lustre_sb_info *lsi = s2lsi(sb);
1019         struct  ll_sb_info *sbi = NULL;
1020         char    *dt = NULL, *md = NULL;
1021         char    *profilenm = get_profile_name(sb);
1022         struct config_llog_instance *cfg;
1023         /* %p for void* in printf needs 16+2 characters: 0xffffffffffffffff */
1024         const int instlen = sizeof(cfg->cfg_instance) * 2 + 2;
1025         char name[MAX_STRING_SIZE];
1026         int md_len = 0;
1027         int dt_len = 0;
1028         char *ptr;
1029         int len;
1030         int err;
1031
1032         ENTRY;
1033         CDEBUG(D_VFSTRACE, "VFS Op: sb %p\n", sb);
1034
1035         try_module_get(THIS_MODULE);
1036
1037         OBD_ALLOC_PTR(cfg);
1038         if (cfg == NULL)
1039                 GOTO(out_free, err = -ENOMEM);
1040
1041         /* client additional sb info */
1042         lsi->lsi_llsbi = sbi = ll_init_sbi();
1043         if (!sbi)
1044                 GOTO(out_free, err = -ENOMEM);
1045
1046         err = ll_options(lsi->lsi_lmd->lmd_opts, sbi);
1047         if (err)
1048                 GOTO(out_free, err);
1049
1050         err = super_setup_bdi_name(sb, "lustre-%p", sb);
1051         if (err)
1052                 GOTO(out_free, err);
1053
1054 #ifndef HAVE_DCACHE_LOCK
1055         /* kernel >= 2.6.38 store dentry operations in sb->s_d_op. */
1056         sb->s_d_op = &ll_d_ops;
1057 #endif
1058         /* Get fsname */
1059         len = strlen(lsi->lsi_lmd->lmd_profile);
1060         ptr = strrchr(lsi->lsi_lmd->lmd_profile, '-');
1061         if (ptr && (strcmp(ptr, "-client") == 0))
1062                 len -= 7;
1063
1064         /* Mount info */
1065         snprintf(name, MAX_STRING_SIZE, "%.*s-%p", len,
1066                  lsi->lsi_lmd->lmd_profile, sb);
1067
1068         /* Call ll_debugfs_register_super() before lustre_process_log()
1069          * so that "llite.*.*" params can be processed correctly.
1070          */
1071         err = ll_debugfs_register_super(sb, name);
1072         if (err < 0) {
1073                 CERROR("%s: could not register mountpoint in llite: rc = %d\n",
1074                        ll_get_fsname(sb, NULL, 0), err);
1075                 err = 0;
1076         }
1077
1078         /* Generate a string unique to this super, in case some joker tries
1079            to mount the same fs at two mount points.
1080            Use the address of the super itself.*/
1081         cfg->cfg_instance = sb;
1082         cfg->cfg_uuid = lsi->lsi_llsbi->ll_sb_uuid;
1083         cfg->cfg_callback = class_config_llog_handler;
1084         cfg->cfg_sub_clds = CONFIG_SUB_CLIENT;
1085         /* set up client obds */
1086         err = lustre_process_log(sb, profilenm, cfg);
1087         if (err < 0)
1088                 GOTO(out_proc, err);
1089
1090         /* Profile set with LCFG_MOUNTOPT so we can find our mdc and osc obds */
1091         lprof = class_get_profile(profilenm);
1092         if (lprof == NULL) {
1093                 LCONSOLE_ERROR_MSG(0x156, "The client profile '%s' could not be"
1094                                    " read from the MGS.  Does that filesystem "
1095                                    "exist?\n", profilenm);
1096                 GOTO(out_proc, err = -EINVAL);
1097         }
1098         CDEBUG(D_CONFIG, "Found profile %s: mdc=%s osc=%s\n", profilenm,
1099                lprof->lp_md, lprof->lp_dt);
1100
1101         dt_len = strlen(lprof->lp_dt) + instlen + 2;
1102         OBD_ALLOC(dt, dt_len);
1103         if (!dt)
1104                 GOTO(out_proc, err = -ENOMEM);
1105         snprintf(dt, dt_len - 1, "%s-%p", lprof->lp_dt, cfg->cfg_instance);
1106
1107         md_len = strlen(lprof->lp_md) + instlen + 2;
1108         OBD_ALLOC(md, md_len);
1109         if (!md)
1110                 GOTO(out_proc, err = -ENOMEM);
1111         snprintf(md, md_len - 1, "%s-%p", lprof->lp_md, cfg->cfg_instance);
1112
1113         /* connections, registrations, sb setup */
1114         err = client_common_fill_super(sb, md, dt, mnt);
1115         if (err < 0)
1116                 GOTO(out_proc, err);
1117
1118         sbi->ll_client_common_fill_super_succeeded = 1;
1119
1120 out_proc:
1121         if (err < 0)
1122                 ll_debugfs_unregister_super(sb);
1123 out_free:
1124         if (md)
1125                 OBD_FREE(md, md_len);
1126         if (dt)
1127                 OBD_FREE(dt, dt_len);
1128         if (lprof != NULL)
1129                 class_put_profile(lprof);
1130         if (cfg)
1131                 OBD_FREE_PTR(cfg);
1132         if (err)
1133                 ll_put_super(sb);
1134         else if (sbi->ll_flags & LL_SBI_VERBOSE)
1135                 LCONSOLE_WARN("Mounted %s\n", profilenm);
1136         RETURN(err);
1137 } /* ll_fill_super */
1138
1139 void ll_put_super(struct super_block *sb)
1140 {
1141         struct config_llog_instance cfg, params_cfg;
1142         struct obd_device *obd;
1143         struct lustre_sb_info *lsi = s2lsi(sb);
1144         struct ll_sb_info *sbi = ll_s2sbi(sb);
1145         char *profilenm = get_profile_name(sb);
1146         long ccc_count;
1147         int next, force = 1, rc = 0;
1148         ENTRY;
1149
1150         if (!sbi)
1151                 GOTO(out_no_sbi, 0);
1152
1153         CDEBUG(D_VFSTRACE, "VFS Op: sb %p - %s\n", sb, profilenm);
1154
1155         cfg.cfg_instance = sb;
1156         lustre_end_log(sb, profilenm, &cfg);
1157
1158         params_cfg.cfg_instance = sb;
1159         lustre_end_log(sb, PARAMS_FILENAME, &params_cfg);
1160
1161         if (sbi->ll_md_exp) {
1162                 obd = class_exp2obd(sbi->ll_md_exp);
1163                 if (obd)
1164                         force = obd->obd_force;
1165         }
1166
1167         /* Wait for unstable pages to be committed to stable storage */
1168         if (force == 0) {
1169                 struct l_wait_info lwi = LWI_INTR(LWI_ON_SIGNAL_NOOP, NULL);
1170                 rc = l_wait_event(sbi->ll_cache->ccc_unstable_waitq,
1171                         atomic_long_read(&sbi->ll_cache->ccc_unstable_nr) == 0,
1172                         &lwi);
1173         }
1174
1175         ccc_count = atomic_long_read(&sbi->ll_cache->ccc_unstable_nr);
1176         if (force == 0 && rc != -EINTR)
1177                 LASSERTF(ccc_count == 0, "count: %li\n", ccc_count);
1178
1179
1180         /* We need to set force before the lov_disconnect in
1181            lustre_common_put_super, since l_d cleans up osc's as well. */
1182         if (force) {
1183                 next = 0;
1184                 while ((obd = class_devices_in_group(&sbi->ll_sb_uuid,
1185                                                      &next)) != NULL) {
1186                         obd->obd_force = force;
1187                 }
1188         }
1189
1190         if (sbi->ll_client_common_fill_super_succeeded) {
1191                 /* Only if client_common_fill_super succeeded */
1192                 client_common_put_super(sb);
1193         }
1194
1195         next = 0;
1196         while ((obd = class_devices_in_group(&sbi->ll_sb_uuid, &next)) !=NULL) {
1197                 class_manual_cleanup(obd);
1198         }
1199
1200         if (sbi->ll_flags & LL_SBI_VERBOSE)
1201                 LCONSOLE_WARN("Unmounted %s\n", profilenm ? profilenm : "");
1202
1203         if (profilenm)
1204                 class_del_profile(profilenm);
1205
1206 #ifndef HAVE_SUPER_SETUP_BDI_NAME
1207         if (lsi->lsi_flags & LSI_BDI_INITIALIZED) {
1208                 bdi_destroy(&lsi->lsi_bdi);
1209                 lsi->lsi_flags &= ~LSI_BDI_INITIALIZED;
1210         }
1211 #endif
1212
1213         ll_free_sbi(sb);
1214         lsi->lsi_llsbi = NULL;
1215 out_no_sbi:
1216         lustre_common_put_super(sb);
1217
1218         cl_env_cache_purge(~0);
1219
1220         module_put(THIS_MODULE);
1221
1222         EXIT;
1223 } /* client_put_super */
1224
1225 struct inode *ll_inode_from_resource_lock(struct ldlm_lock *lock)
1226 {
1227         struct inode *inode = NULL;
1228
1229         /* NOTE: we depend on atomic igrab() -bzzz */
1230         lock_res_and_lock(lock);
1231         if (lock->l_resource->lr_lvb_inode) {
1232                 struct ll_inode_info * lli;
1233                 lli = ll_i2info(lock->l_resource->lr_lvb_inode);
1234                 if (lli->lli_inode_magic == LLI_INODE_MAGIC) {
1235                         inode = igrab(lock->l_resource->lr_lvb_inode);
1236                 } else {
1237                         inode = lock->l_resource->lr_lvb_inode;
1238                         LDLM_DEBUG_LIMIT(inode->i_state & I_FREEING ?  D_INFO :
1239                                          D_WARNING, lock, "lr_lvb_inode %p is "
1240                                          "bogus: magic %08x",
1241                                          lock->l_resource->lr_lvb_inode,
1242                                          lli->lli_inode_magic);
1243                         inode = NULL;
1244                 }
1245         }
1246         unlock_res_and_lock(lock);
1247         return inode;
1248 }
1249
1250 void ll_dir_clear_lsm_md(struct inode *inode)
1251 {
1252         struct ll_inode_info *lli = ll_i2info(inode);
1253
1254         LASSERT(S_ISDIR(inode->i_mode));
1255
1256         if (lli->lli_lsm_md != NULL) {
1257                 lmv_free_memmd(lli->lli_lsm_md);
1258                 lli->lli_lsm_md = NULL;
1259         }
1260 }
1261
1262 static struct inode *ll_iget_anon_dir(struct super_block *sb,
1263                                       const struct lu_fid *fid,
1264                                       struct lustre_md *md)
1265 {
1266         struct ll_sb_info       *sbi = ll_s2sbi(sb);
1267         struct mdt_body         *body = md->body;
1268         struct inode            *inode;
1269         ino_t                   ino;
1270         ENTRY;
1271
1272         ino = cl_fid_build_ino(fid, sbi->ll_flags & LL_SBI_32BIT_API);
1273         inode = iget_locked(sb, ino);
1274         if (inode == NULL) {
1275                 CERROR("%s: failed get simple inode "DFID": rc = -ENOENT\n",
1276                        ll_get_fsname(sb, NULL, 0), PFID(fid));
1277                 RETURN(ERR_PTR(-ENOENT));
1278         }
1279
1280         if (inode->i_state & I_NEW) {
1281                 struct ll_inode_info *lli = ll_i2info(inode);
1282                 struct lmv_stripe_md *lsm = md->lmv;
1283
1284                 inode->i_mode = (inode->i_mode & ~S_IFMT) |
1285                                 (body->mbo_mode & S_IFMT);
1286                 LASSERTF(S_ISDIR(inode->i_mode), "Not slave inode "DFID"\n",
1287                          PFID(fid));
1288
1289                 LTIME_S(inode->i_mtime) = 0;
1290                 LTIME_S(inode->i_atime) = 0;
1291                 LTIME_S(inode->i_ctime) = 0;
1292                 inode->i_rdev = 0;
1293
1294 #ifdef HAVE_BACKING_DEV_INFO
1295                 /* initializing backing dev info. */
1296                 inode->i_mapping->backing_dev_info =
1297                                                 &s2lsi(inode->i_sb)->lsi_bdi;
1298 #endif
1299                 inode->i_op = &ll_dir_inode_operations;
1300                 inode->i_fop = &ll_dir_operations;
1301                 lli->lli_fid = *fid;
1302                 ll_lli_init(lli);
1303
1304                 LASSERT(lsm != NULL);
1305                 /* master object FID */
1306                 lli->lli_pfid = body->mbo_fid1;
1307                 CDEBUG(D_INODE, "lli %p slave "DFID" master "DFID"\n",
1308                        lli, PFID(fid), PFID(&lli->lli_pfid));
1309                 unlock_new_inode(inode);
1310         }
1311
1312         RETURN(inode);
1313 }
1314
1315 static int ll_init_lsm_md(struct inode *inode, struct lustre_md *md)
1316 {
1317         struct lu_fid *fid;
1318         struct lmv_stripe_md *lsm = md->lmv;
1319         int i;
1320
1321         LASSERT(lsm != NULL);
1322         /* XXX sigh, this lsm_root initialization should be in
1323          * LMV layer, but it needs ll_iget right now, so we
1324          * put this here right now. */
1325         for (i = 0; i < lsm->lsm_md_stripe_count; i++) {
1326                 fid = &lsm->lsm_md_oinfo[i].lmo_fid;
1327                 LASSERT(lsm->lsm_md_oinfo[i].lmo_root == NULL);
1328                 /* Unfortunately ll_iget will call ll_update_inode,
1329                  * where the initialization of slave inode is slightly
1330                  * different, so it reset lsm_md to NULL to avoid
1331                  * initializing lsm for slave inode. */
1332                 lsm->lsm_md_oinfo[i].lmo_root =
1333                                 ll_iget_anon_dir(inode->i_sb, fid, md);
1334                 if (IS_ERR(lsm->lsm_md_oinfo[i].lmo_root)) {
1335                         int rc = PTR_ERR(lsm->lsm_md_oinfo[i].lmo_root);
1336
1337                         lsm->lsm_md_oinfo[i].lmo_root = NULL;
1338                         return rc;
1339                 }
1340         }
1341
1342         return 0;
1343 }
1344
1345 static int ll_update_lsm_md(struct inode *inode, struct lustre_md *md)
1346 {
1347         struct ll_inode_info *lli = ll_i2info(inode);
1348         struct lmv_stripe_md *lsm = md->lmv;
1349         int     rc;
1350         ENTRY;
1351
1352         LASSERT(S_ISDIR(inode->i_mode));
1353         CDEBUG(D_INODE, "update lsm %p of "DFID"\n", lli->lli_lsm_md,
1354                PFID(ll_inode2fid(inode)));
1355
1356         /*
1357          * no striped information from request, lustre_md from req does not
1358          * include stripeEA, see ll_md_setattr()
1359          */
1360         if (!lsm)
1361                 RETURN(0);
1362
1363         /* Compare the old and new stripe information */
1364         if (lli->lli_lsm_md && !lsm_md_eq(lli->lli_lsm_md, lsm)) {
1365                 struct lmv_stripe_md *old_lsm = lli->lli_lsm_md;
1366                 int idx;
1367                 bool layout_changed = lsm->lsm_md_layout_version >
1368                                       old_lsm->lsm_md_layout_version;
1369
1370                 int mask = layout_changed ? D_INODE : D_ERROR;
1371
1372                 CDEBUG(mask,
1373                         "%s: inode@%p "DFID" lmv layout %s magic %#x/%#x "
1374                         "stripe count %d/%d master_mdt %d/%d "
1375                         "hash_type %#x/%#x version %d/%d migrate offset %d/%d "
1376                         "migrate hash %#x/%#x pool %s/%s\n",
1377                        ll_get_fsname(inode->i_sb, NULL, 0), inode,
1378                        PFID(&lli->lli_fid),
1379                        layout_changed ? "changed" : "mismatch",
1380                        lsm->lsm_md_magic, old_lsm->lsm_md_magic,
1381                        lsm->lsm_md_stripe_count,
1382                        old_lsm->lsm_md_stripe_count,
1383                        lsm->lsm_md_master_mdt_index,
1384                        old_lsm->lsm_md_master_mdt_index,
1385                        lsm->lsm_md_hash_type, old_lsm->lsm_md_hash_type,
1386                        lsm->lsm_md_layout_version,
1387                        old_lsm->lsm_md_layout_version,
1388                        lsm->lsm_md_migrate_offset,
1389                        old_lsm->lsm_md_migrate_offset,
1390                        lsm->lsm_md_migrate_hash,
1391                        old_lsm->lsm_md_migrate_hash,
1392                        lsm->lsm_md_pool_name,
1393                        old_lsm->lsm_md_pool_name);
1394
1395                 for (idx = 0; idx < old_lsm->lsm_md_stripe_count; idx++)
1396                         CDEBUG(mask, "old stripe[%d] "DFID"\n",
1397                                idx, PFID(&old_lsm->lsm_md_oinfo[idx].lmo_fid));
1398
1399                 for (idx = 0; idx < lsm->lsm_md_stripe_count; idx++)
1400                         CDEBUG(mask, "new stripe[%d] "DFID"\n",
1401                                idx, PFID(&lsm->lsm_md_oinfo[idx].lmo_fid));
1402
1403                 if (!layout_changed)
1404                         RETURN(-EINVAL);
1405
1406                 ll_dir_clear_lsm_md(inode);
1407         }
1408
1409         /* set the directory layout */
1410         if (!lli->lli_lsm_md) {
1411                 struct cl_attr  *attr;
1412
1413                 rc = ll_init_lsm_md(inode, md);
1414                 if (rc != 0)
1415                         RETURN(rc);
1416
1417                 /* set md->lmv to NULL, so the following free lustre_md
1418                  * will not free this lsm */
1419                 md->lmv = NULL;
1420                 lli->lli_lsm_md = lsm;
1421
1422                 OBD_ALLOC_PTR(attr);
1423                 if (attr == NULL)
1424                         RETURN(-ENOMEM);
1425
1426                 /* validate the lsm */
1427                 rc = md_merge_attr(ll_i2mdexp(inode), lsm, attr,
1428                                    ll_md_blocking_ast);
1429                 if (rc != 0) {
1430                         OBD_FREE_PTR(attr);
1431                         RETURN(rc);
1432                 }
1433
1434                 if (md->body->mbo_valid & OBD_MD_FLNLINK)
1435                         md->body->mbo_nlink = attr->cat_nlink;
1436                 if (md->body->mbo_valid & OBD_MD_FLSIZE)
1437                         md->body->mbo_size = attr->cat_size;
1438                 if (md->body->mbo_valid & OBD_MD_FLATIME)
1439                         md->body->mbo_atime = attr->cat_atime;
1440                 if (md->body->mbo_valid & OBD_MD_FLCTIME)
1441                         md->body->mbo_ctime = attr->cat_ctime;
1442                 if (md->body->mbo_valid & OBD_MD_FLMTIME)
1443                         md->body->mbo_mtime = attr->cat_mtime;
1444
1445                 OBD_FREE_PTR(attr);
1446
1447                 CDEBUG(D_INODE, "Set lsm %p magic %x to "DFID"\n", lsm,
1448                        lsm->lsm_md_magic, PFID(ll_inode2fid(inode)));
1449                 RETURN(0);
1450         }
1451
1452         /* Compare the old and new stripe information */
1453         if (!lsm_md_eq(lli->lli_lsm_md, lsm)) {
1454                 struct lmv_stripe_md    *old_lsm = lli->lli_lsm_md;
1455                 int                     idx;
1456
1457                 CERROR("%s: inode "DFID"(%p)'s lmv layout mismatch (%p)/(%p)"
1458                        "magic:0x%x/0x%x stripe count: %d/%d master_mdt: %d/%d"
1459                        "hash_type:0x%x/0x%x layout: 0x%x/0x%x pool:%s/%s\n",
1460                        ll_get_fsname(inode->i_sb, NULL, 0), PFID(&lli->lli_fid),
1461                        inode, lsm, old_lsm,
1462                        lsm->lsm_md_magic, old_lsm->lsm_md_magic,
1463                        lsm->lsm_md_stripe_count,
1464                        old_lsm->lsm_md_stripe_count,
1465                        lsm->lsm_md_master_mdt_index,
1466                        old_lsm->lsm_md_master_mdt_index,
1467                        lsm->lsm_md_hash_type, old_lsm->lsm_md_hash_type,
1468                        lsm->lsm_md_layout_version,
1469                        old_lsm->lsm_md_layout_version,
1470                        lsm->lsm_md_pool_name,
1471                        old_lsm->lsm_md_pool_name);
1472
1473                 for (idx = 0; idx < old_lsm->lsm_md_stripe_count; idx++) {
1474                         CERROR("%s: sub FIDs in old lsm idx %d, old: "DFID"\n",
1475                                ll_get_fsname(inode->i_sb, NULL, 0), idx,
1476                                PFID(&old_lsm->lsm_md_oinfo[idx].lmo_fid));
1477                 }
1478
1479                 for (idx = 0; idx < lsm->lsm_md_stripe_count; idx++) {
1480                         CERROR("%s: sub FIDs in new lsm idx %d, new: "DFID"\n",
1481                                ll_get_fsname(inode->i_sb, NULL, 0), idx,
1482                                PFID(&lsm->lsm_md_oinfo[idx].lmo_fid));
1483                 }
1484
1485                 RETURN(-EIO);
1486         }
1487
1488         RETURN(0);
1489 }
1490
1491 void ll_clear_inode(struct inode *inode)
1492 {
1493         struct ll_inode_info *lli = ll_i2info(inode);
1494         struct ll_sb_info *sbi = ll_i2sbi(inode);
1495         ENTRY;
1496
1497         CDEBUG(D_VFSTRACE, "VFS Op:inode="DFID"(%p)\n",
1498                PFID(ll_inode2fid(inode)), inode);
1499
1500         if (S_ISDIR(inode->i_mode)) {
1501                 /* these should have been cleared in ll_file_release */
1502                 LASSERT(lli->lli_opendir_key == NULL);
1503                 LASSERT(lli->lli_sai == NULL);
1504                 LASSERT(lli->lli_opendir_pid == 0);
1505         }
1506
1507         md_null_inode(sbi->ll_md_exp, ll_inode2fid(inode));
1508
1509         LASSERT(!lli->lli_open_fd_write_count);
1510         LASSERT(!lli->lli_open_fd_read_count);
1511         LASSERT(!lli->lli_open_fd_exec_count);
1512
1513         if (lli->lli_mds_write_och)
1514                 ll_md_real_close(inode, FMODE_WRITE);
1515         if (lli->lli_mds_exec_och)
1516                 ll_md_real_close(inode, FMODE_EXEC);
1517         if (lli->lli_mds_read_och)
1518                 ll_md_real_close(inode, FMODE_READ);
1519
1520         if (S_ISLNK(inode->i_mode) && lli->lli_symlink_name) {
1521                 OBD_FREE(lli->lli_symlink_name,
1522                          strlen(lli->lli_symlink_name) + 1);
1523                 lli->lli_symlink_name = NULL;
1524         }
1525
1526         ll_xattr_cache_destroy(inode);
1527
1528 #ifdef CONFIG_FS_POSIX_ACL
1529         forget_all_cached_acls(inode);
1530         if (lli->lli_posix_acl) {
1531                 LASSERT(atomic_read(&lli->lli_posix_acl->a_refcount) == 1);
1532                 posix_acl_release(lli->lli_posix_acl);
1533                 lli->lli_posix_acl = NULL;
1534         }
1535 #endif
1536         lli->lli_inode_magic = LLI_INODE_DEAD;
1537
1538         if (S_ISDIR(inode->i_mode))
1539                 ll_dir_clear_lsm_md(inode);
1540         else if (S_ISREG(inode->i_mode) && !is_bad_inode(inode))
1541                 LASSERT(list_empty(&lli->lli_agl_list));
1542
1543         /*
1544          * XXX This has to be done before lsm is freed below, because
1545          * cl_object still uses inode lsm.
1546          */
1547         cl_inode_fini(inode);
1548
1549         EXIT;
1550 }
1551
1552 static int ll_md_setattr(struct dentry *dentry, struct md_op_data *op_data)
1553 {
1554         struct lustre_md md;
1555         struct inode *inode = dentry->d_inode;
1556         struct ll_sb_info *sbi = ll_i2sbi(inode);
1557         struct ptlrpc_request *request = NULL;
1558         int rc, ia_valid;
1559         ENTRY;
1560
1561         op_data = ll_prep_md_op_data(op_data, inode, NULL, NULL, 0, 0,
1562                                      LUSTRE_OPC_ANY, NULL);
1563         if (IS_ERR(op_data))
1564                 RETURN(PTR_ERR(op_data));
1565
1566         rc = md_setattr(sbi->ll_md_exp, op_data, NULL, 0, &request);
1567         if (rc) {
1568                 ptlrpc_req_finished(request);
1569                 if (rc == -ENOENT) {
1570                         clear_nlink(inode);
1571                         /* Unlinked special device node? Or just a race?
1572                          * Pretend we done everything. */
1573                         if (!S_ISREG(inode->i_mode) &&
1574                             !S_ISDIR(inode->i_mode)) {
1575                                 ia_valid = op_data->op_attr.ia_valid;
1576                                 op_data->op_attr.ia_valid &= ~TIMES_SET_FLAGS;
1577                                 rc = simple_setattr(dentry, &op_data->op_attr);
1578                                 op_data->op_attr.ia_valid = ia_valid;
1579                         }
1580                 } else if (rc != -EPERM && rc != -EACCES && rc != -ETXTBSY) {
1581                         CERROR("md_setattr fails: rc = %d\n", rc);
1582                 }
1583                 RETURN(rc);
1584         }
1585
1586         rc = md_get_lustre_md(sbi->ll_md_exp, request, sbi->ll_dt_exp,
1587                               sbi->ll_md_exp, &md);
1588         if (rc) {
1589                 ptlrpc_req_finished(request);
1590                 RETURN(rc);
1591         }
1592
1593         ia_valid = op_data->op_attr.ia_valid;
1594         /* inode size will be in ll_setattr_ost, can't do it now since dirty
1595          * cache is not cleared yet. */
1596         op_data->op_attr.ia_valid &= ~(TIMES_SET_FLAGS | ATTR_SIZE);
1597         if (S_ISREG(inode->i_mode))
1598                 inode_lock(inode);
1599         rc = simple_setattr(dentry, &op_data->op_attr);
1600         if (S_ISREG(inode->i_mode))
1601                 inode_unlock(inode);
1602         op_data->op_attr.ia_valid = ia_valid;
1603
1604         rc = ll_update_inode(inode, &md);
1605         ptlrpc_req_finished(request);
1606
1607         RETURN(rc);
1608 }
1609
1610 /* If this inode has objects allocated to it (lsm != NULL), then the OST
1611  * object(s) determine the file size and mtime.  Otherwise, the MDS will
1612  * keep these values until such a time that objects are allocated for it.
1613  * We do the MDS operations first, as it is checking permissions for us.
1614  * We don't to the MDS RPC if there is nothing that we want to store there,
1615  * otherwise there is no harm in updating mtime/atime on the MDS if we are
1616  * going to do an RPC anyways.
1617  *
1618  * If we are doing a truncate, we will send the mtime and ctime updates
1619  * to the OST with the punch RPC, otherwise we do an explicit setattr RPC.
1620  * I don't believe it is possible to get e.g. ATTR_MTIME_SET and ATTR_SIZE
1621  * at the same time.
1622  *
1623  * In case of HSMimport, we only set attr on MDS.
1624  */
1625 int ll_setattr_raw(struct dentry *dentry, struct iattr *attr,
1626                    enum op_xvalid xvalid, bool hsm_import)
1627 {
1628         struct inode *inode = dentry->d_inode;
1629         struct ll_inode_info *lli = ll_i2info(inode);
1630         struct md_op_data *op_data = NULL;
1631         int rc = 0;
1632         ENTRY;
1633
1634         CDEBUG(D_VFSTRACE, "%s: setattr inode "DFID"(%p) from %llu to %llu, "
1635                "valid %x, hsm_import %d\n",
1636                ll_get_fsname(inode->i_sb, NULL, 0), PFID(&lli->lli_fid),
1637                inode, i_size_read(inode), attr->ia_size, attr->ia_valid,
1638                hsm_import);
1639
1640         if (attr->ia_valid & ATTR_SIZE) {
1641                 /* Check new size against VFS/VM file size limit and rlimit */
1642                 rc = inode_newsize_ok(inode, attr->ia_size);
1643                 if (rc)
1644                         RETURN(rc);
1645
1646                 /* The maximum Lustre file size is variable, based on the
1647                  * OST maximum object size and number of stripes.  This
1648                  * needs another check in addition to the VFS check above. */
1649                 if (attr->ia_size > ll_file_maxbytes(inode)) {
1650                         CDEBUG(D_INODE,"file "DFID" too large %llu > %llu\n",
1651                                PFID(&lli->lli_fid), attr->ia_size,
1652                                ll_file_maxbytes(inode));
1653                         RETURN(-EFBIG);
1654                 }
1655
1656                 attr->ia_valid |= ATTR_MTIME | ATTR_CTIME;
1657         }
1658
1659         /* POSIX: check before ATTR_*TIME_SET set (from inode_change_ok) */
1660         if (attr->ia_valid & TIMES_SET_FLAGS) {
1661                 if ((!uid_eq(current_fsuid(), inode->i_uid)) &&
1662                     !cfs_capable(CFS_CAP_FOWNER))
1663                         RETURN(-EPERM);
1664         }
1665
1666         /* We mark all of the fields "set" so MDS/OST does not re-set them */
1667         if (!(xvalid & OP_XVALID_CTIME_SET) &&
1668              (attr->ia_valid & ATTR_CTIME)) {
1669                 attr->ia_ctime = current_time(inode);
1670                 xvalid |= OP_XVALID_CTIME_SET;
1671         }
1672         if (!(attr->ia_valid & ATTR_ATIME_SET) &&
1673             (attr->ia_valid & ATTR_ATIME)) {
1674                 attr->ia_atime = current_time(inode);
1675                 attr->ia_valid |= ATTR_ATIME_SET;
1676         }
1677         if (!(attr->ia_valid & ATTR_MTIME_SET) &&
1678             (attr->ia_valid & ATTR_MTIME)) {
1679                 attr->ia_mtime = current_time(inode);
1680                 attr->ia_valid |= ATTR_MTIME_SET;
1681         }
1682
1683         if (attr->ia_valid & (ATTR_MTIME | ATTR_CTIME))
1684                 CDEBUG(D_INODE, "setting mtime %lu, ctime %lu, now = %llu\n",
1685                        LTIME_S(attr->ia_mtime), LTIME_S(attr->ia_ctime),
1686                        (s64)ktime_get_real_seconds());
1687
1688         if (S_ISREG(inode->i_mode)) {
1689                 if (attr->ia_valid & ATTR_SIZE)
1690                         inode_dio_write_done(inode);
1691                 inode_unlock(inode);
1692         }
1693
1694         /* We always do an MDS RPC, even if we're only changing the size;
1695          * only the MDS knows whether truncate() should fail with -ETXTBUSY */
1696
1697         OBD_ALLOC_PTR(op_data);
1698         if (op_data == NULL)
1699                 GOTO(out, rc = -ENOMEM);
1700
1701         if (!hsm_import && attr->ia_valid & ATTR_SIZE) {
1702                 /* If we are changing file size, file content is
1703                  * modified, flag it.
1704                  */
1705                 xvalid |= OP_XVALID_OWNEROVERRIDE;
1706                 op_data->op_bias |= MDS_DATA_MODIFIED;
1707                 ll_file_clear_flag(lli, LLIF_DATA_MODIFIED);
1708         }
1709
1710         if (attr->ia_valid & ATTR_FILE) {
1711                 struct ll_file_data *fd = LUSTRE_FPRIVATE(attr->ia_file);
1712
1713                 if (fd->fd_lease_och)
1714                         op_data->op_bias |= MDS_TRUNC_KEEP_LEASE;
1715         }
1716
1717         op_data->op_attr = *attr;
1718         op_data->op_xvalid = xvalid;
1719
1720         rc = ll_md_setattr(dentry, op_data);
1721         if (rc)
1722                 GOTO(out, rc);
1723
1724         if (!S_ISREG(inode->i_mode) || hsm_import)
1725                 GOTO(out, rc = 0);
1726
1727         if (attr->ia_valid & (ATTR_SIZE | ATTR_ATIME | ATTR_ATIME_SET |
1728                               ATTR_MTIME | ATTR_MTIME_SET | ATTR_CTIME) ||
1729             xvalid & OP_XVALID_CTIME_SET) {
1730                 /* For truncate and utimes sending attributes to OSTs, setting
1731                  * mtime/atime to the past will be performed under PW [0:EOF]
1732                  * extent lock (new_size:EOF for truncate).  It may seem
1733                  * excessive to send mtime/atime updates to OSTs when not
1734                  * setting times to past, but it is necessary due to possible
1735                  * time de-synchronization between MDT inode and OST objects
1736                  */
1737                 rc = cl_setattr_ost(lli->lli_clob, attr, xvalid, 0);
1738         }
1739
1740         /* If the file was restored, it needs to set dirty flag.
1741          *
1742          * We've already sent MDS_DATA_MODIFIED flag in
1743          * ll_md_setattr() for truncate. However, the MDT refuses to
1744          * set the HS_DIRTY flag on released files, so we have to set
1745          * it again if the file has been restored. Please check how
1746          * LLIF_DATA_MODIFIED is set in vvp_io_setattr_fini().
1747          *
1748          * Please notice that if the file is not released, the previous
1749          * MDS_DATA_MODIFIED has taken effect and usually
1750          * LLIF_DATA_MODIFIED is not set(see vvp_io_setattr_fini()).
1751          * This way we can save an RPC for common open + trunc
1752          * operation. */
1753         if (ll_file_test_and_clear_flag(lli, LLIF_DATA_MODIFIED)) {
1754                 struct hsm_state_set hss = {
1755                         .hss_valid = HSS_SETMASK,
1756                         .hss_setmask = HS_DIRTY,
1757                 };
1758                 int rc2;
1759
1760                 rc2 = ll_hsm_state_set(inode, &hss);
1761                 /* truncate and write can happen at the same time, so that
1762                  * the file can be set modified even though the file is not
1763                  * restored from released state, and ll_hsm_state_set() is
1764                  * not applicable for the file, and rc2 < 0 is normal in this
1765                  * case. */
1766                 if (rc2 < 0)
1767                         CDEBUG(D_INFO, DFID "HSM set dirty failed: rc2 = %d\n",
1768                                PFID(ll_inode2fid(inode)), rc2);
1769         }
1770
1771         EXIT;
1772 out:
1773         if (op_data != NULL)
1774                 ll_finish_md_op_data(op_data);
1775
1776         if (S_ISREG(inode->i_mode)) {
1777                 inode_lock(inode);
1778                 if ((attr->ia_valid & ATTR_SIZE) && !hsm_import)
1779                         inode_dio_wait(inode);
1780                 /* Once we've got the i_mutex, it's safe to set the S_NOSEC
1781                  * flag.  ll_update_inode (called from ll_md_setattr), clears
1782                  * inode flags, so there is a gap where S_NOSEC is not set.
1783                  * This can cause a writer to take the i_mutex unnecessarily,
1784                  * but this is safe to do and should be rare. */
1785                 inode_has_no_xattr(inode);
1786         }
1787
1788         ll_stats_ops_tally(ll_i2sbi(inode), (attr->ia_valid & ATTR_SIZE) ?
1789                         LPROC_LL_TRUNC : LPROC_LL_SETATTR, 1);
1790
1791         return rc;
1792 }
1793
1794 int ll_setattr(struct dentry *de, struct iattr *attr)
1795 {
1796         int mode = de->d_inode->i_mode;
1797         enum op_xvalid xvalid = 0;
1798
1799         if ((attr->ia_valid & (ATTR_CTIME|ATTR_SIZE|ATTR_MODE)) ==
1800                               (ATTR_CTIME|ATTR_SIZE|ATTR_MODE))
1801                 xvalid |= OP_XVALID_OWNEROVERRIDE;
1802
1803         if (((attr->ia_valid & (ATTR_MODE|ATTR_FORCE|ATTR_SIZE)) ==
1804                                (ATTR_SIZE|ATTR_MODE)) &&
1805             (((mode & S_ISUID) && !(attr->ia_mode & S_ISUID)) ||
1806              (((mode & (S_ISGID|S_IXGRP)) == (S_ISGID|S_IXGRP)) &&
1807               !(attr->ia_mode & S_ISGID))))
1808                 attr->ia_valid |= ATTR_FORCE;
1809
1810         if ((attr->ia_valid & ATTR_MODE) &&
1811             (mode & S_ISUID) &&
1812             !(attr->ia_mode & S_ISUID) &&
1813             !(attr->ia_valid & ATTR_KILL_SUID))
1814                 attr->ia_valid |= ATTR_KILL_SUID;
1815
1816         if ((attr->ia_valid & ATTR_MODE) &&
1817             ((mode & (S_ISGID|S_IXGRP)) == (S_ISGID|S_IXGRP)) &&
1818             !(attr->ia_mode & S_ISGID) &&
1819             !(attr->ia_valid & ATTR_KILL_SGID))
1820                 attr->ia_valid |= ATTR_KILL_SGID;
1821
1822         return ll_setattr_raw(de, attr, xvalid, false);
1823 }
1824
1825 int ll_statfs_internal(struct ll_sb_info *sbi, struct obd_statfs *osfs,
1826                        u32 flags)
1827 {
1828         struct obd_statfs obd_osfs;
1829         time64_t max_age;
1830         int rc;
1831
1832         ENTRY;
1833         max_age = ktime_get_seconds() - OBD_STATFS_CACHE_SECONDS;
1834
1835         rc = obd_statfs(NULL, sbi->ll_md_exp, osfs, max_age, flags);
1836         if (rc) {
1837                 CERROR("md_statfs fails: rc = %d\n", rc);
1838                 RETURN(rc);
1839         }
1840
1841         osfs->os_type = LL_SUPER_MAGIC;
1842
1843         CDEBUG(D_SUPER, "MDC blocks %llu/%llu objects %llu/%llu\n",
1844                osfs->os_bavail, osfs->os_blocks, osfs->os_ffree,osfs->os_files);
1845
1846         if (osfs->os_state & OS_STATE_SUM)
1847                 GOTO(out, rc);
1848
1849         if (sbi->ll_flags & LL_SBI_LAZYSTATFS)
1850                 flags |= OBD_STATFS_NODELAY;
1851
1852         rc = obd_statfs(NULL, sbi->ll_dt_exp, &obd_osfs, max_age, flags);
1853         if (rc) {
1854                 CERROR("obd_statfs fails: rc = %d\n", rc);
1855                 RETURN(rc);
1856         }
1857
1858         CDEBUG(D_SUPER, "OSC blocks %llu/%llu objects %llu/%llu\n",
1859                obd_osfs.os_bavail, obd_osfs.os_blocks, obd_osfs.os_ffree,
1860                obd_osfs.os_files);
1861
1862         osfs->os_bsize = obd_osfs.os_bsize;
1863         osfs->os_blocks = obd_osfs.os_blocks;
1864         osfs->os_bfree = obd_osfs.os_bfree;
1865         osfs->os_bavail = obd_osfs.os_bavail;
1866
1867         /* If we don't have as many objects free on the OST as inodes
1868          * on the MDS, we reduce the total number of inodes to
1869          * compensate, so that the "inodes in use" number is correct.
1870          */
1871         if (obd_osfs.os_ffree < osfs->os_ffree) {
1872                 osfs->os_files = (osfs->os_files - osfs->os_ffree) +
1873                         obd_osfs.os_ffree;
1874                 osfs->os_ffree = obd_osfs.os_ffree;
1875         }
1876
1877 out:
1878         RETURN(rc);
1879 }
1880 int ll_statfs(struct dentry *de, struct kstatfs *sfs)
1881 {
1882         struct super_block *sb = de->d_sb;
1883         struct obd_statfs osfs;
1884         __u64 fsid = huge_encode_dev(sb->s_dev);
1885         int rc;
1886
1887         CDEBUG(D_VFSTRACE, "VFS Op: at %llu jiffies\n", get_jiffies_64());
1888         ll_stats_ops_tally(ll_s2sbi(sb), LPROC_LL_STAFS, 1);
1889
1890         /* Some amount of caching on the client is allowed */
1891         rc = ll_statfs_internal(ll_s2sbi(sb), &osfs, OBD_STATFS_SUM);
1892         if (rc)
1893                 return rc;
1894
1895         statfs_unpack(sfs, &osfs);
1896
1897         /* We need to downshift for all 32-bit kernels, because we can't
1898          * tell if the kernel is being called via sys_statfs64() or not.
1899          * Stop before overflowing f_bsize - in which case it is better
1900          * to just risk EOVERFLOW if caller is using old sys_statfs(). */
1901         if (sizeof(long) < 8) {
1902                 while (osfs.os_blocks > ~0UL && sfs->f_bsize < 0x40000000) {
1903                         sfs->f_bsize <<= 1;
1904
1905                         osfs.os_blocks >>= 1;
1906                         osfs.os_bfree >>= 1;
1907                         osfs.os_bavail >>= 1;
1908                 }
1909         }
1910
1911         sfs->f_blocks = osfs.os_blocks;
1912         sfs->f_bfree = osfs.os_bfree;
1913         sfs->f_bavail = osfs.os_bavail;
1914         sfs->f_fsid.val[0] = (__u32)fsid;
1915         sfs->f_fsid.val[1] = (__u32)(fsid >> 32);
1916         return 0;
1917 }
1918
1919 void ll_inode_size_lock(struct inode *inode)
1920 {
1921         struct ll_inode_info *lli;
1922
1923         LASSERT(!S_ISDIR(inode->i_mode));
1924
1925         lli = ll_i2info(inode);
1926         mutex_lock(&lli->lli_size_mutex);
1927 }
1928
1929 void ll_inode_size_unlock(struct inode *inode)
1930 {
1931         struct ll_inode_info *lli;
1932
1933         lli = ll_i2info(inode);
1934         mutex_unlock(&lli->lli_size_mutex);
1935 }
1936
1937 void ll_update_inode_flags(struct inode *inode, int ext_flags)
1938 {
1939         inode->i_flags = ll_ext_to_inode_flags(ext_flags);
1940         if (ext_flags & LUSTRE_PROJINHERIT_FL)
1941                 ll_file_set_flag(ll_i2info(inode), LLIF_PROJECT_INHERIT);
1942         else
1943                 ll_file_clear_flag(ll_i2info(inode), LLIF_PROJECT_INHERIT);
1944 }
1945
1946 int ll_update_inode(struct inode *inode, struct lustre_md *md)
1947 {
1948         struct ll_inode_info *lli = ll_i2info(inode);
1949         struct mdt_body *body = md->body;
1950         struct ll_sb_info *sbi = ll_i2sbi(inode);
1951         int rc = 0;
1952
1953         if (body->mbo_valid & OBD_MD_FLEASIZE) {
1954                 rc = cl_file_inode_init(inode, md);
1955                 if (rc)
1956                         return rc;
1957         }
1958
1959         if (S_ISDIR(inode->i_mode)) {
1960                 rc = ll_update_lsm_md(inode, md);
1961                 if (rc != 0)
1962                         return rc;
1963         }
1964
1965 #ifdef CONFIG_FS_POSIX_ACL
1966         if (body->mbo_valid & OBD_MD_FLACL) {
1967                 spin_lock(&lli->lli_lock);
1968                 if (lli->lli_posix_acl)
1969                         posix_acl_release(lli->lli_posix_acl);
1970                 lli->lli_posix_acl = md->posix_acl;
1971                 spin_unlock(&lli->lli_lock);
1972         }
1973 #endif
1974         inode->i_ino = cl_fid_build_ino(&body->mbo_fid1,
1975                                         sbi->ll_flags & LL_SBI_32BIT_API);
1976         inode->i_generation = cl_fid_build_gen(&body->mbo_fid1);
1977
1978         if (body->mbo_valid & OBD_MD_FLATIME) {
1979                 if (body->mbo_atime > LTIME_S(inode->i_atime))
1980                         LTIME_S(inode->i_atime) = body->mbo_atime;
1981                 lli->lli_atime = body->mbo_atime;
1982         }
1983
1984         if (body->mbo_valid & OBD_MD_FLMTIME) {
1985                 if (body->mbo_mtime > LTIME_S(inode->i_mtime)) {
1986                         CDEBUG(D_INODE, "setting ino %lu mtime from %lu "
1987                                "to %llu\n", inode->i_ino,
1988                                LTIME_S(inode->i_mtime), body->mbo_mtime);
1989                         LTIME_S(inode->i_mtime) = body->mbo_mtime;
1990                 }
1991                 lli->lli_mtime = body->mbo_mtime;
1992         }
1993
1994         if (body->mbo_valid & OBD_MD_FLCTIME) {
1995                 if (body->mbo_ctime > LTIME_S(inode->i_ctime))
1996                         LTIME_S(inode->i_ctime) = body->mbo_ctime;
1997                 lli->lli_ctime = body->mbo_ctime;
1998         }
1999
2000         /* Clear i_flags to remove S_NOSEC before permissions are updated */
2001         if (body->mbo_valid & OBD_MD_FLFLAGS)
2002                 ll_update_inode_flags(inode, body->mbo_flags);
2003         if (body->mbo_valid & OBD_MD_FLMODE)
2004                 inode->i_mode = (inode->i_mode & S_IFMT) |
2005                                 (body->mbo_mode & ~S_IFMT);
2006
2007         if (body->mbo_valid & OBD_MD_FLTYPE)
2008                 inode->i_mode = (inode->i_mode & ~S_IFMT) |
2009                                 (body->mbo_mode & S_IFMT);
2010
2011         LASSERT(inode->i_mode != 0);
2012         if (S_ISREG(inode->i_mode))
2013                 inode->i_blkbits = min(PTLRPC_MAX_BRW_BITS + 1,
2014                                        LL_MAX_BLKSIZE_BITS);
2015         else
2016                 inode->i_blkbits = inode->i_sb->s_blocksize_bits;
2017
2018         if (body->mbo_valid & OBD_MD_FLUID)
2019                 inode->i_uid = make_kuid(&init_user_ns, body->mbo_uid);
2020         if (body->mbo_valid & OBD_MD_FLGID)
2021                 inode->i_gid = make_kgid(&init_user_ns, body->mbo_gid);
2022         if (body->mbo_valid & OBD_MD_FLPROJID)
2023                 lli->lli_projid = body->mbo_projid;
2024         if (body->mbo_valid & OBD_MD_FLNLINK)
2025                 set_nlink(inode, body->mbo_nlink);
2026         if (body->mbo_valid & OBD_MD_FLRDEV)
2027                 inode->i_rdev = old_decode_dev(body->mbo_rdev);
2028
2029         if (body->mbo_valid & OBD_MD_FLID) {
2030                 /* FID shouldn't be changed! */
2031                 if (fid_is_sane(&lli->lli_fid)) {
2032                         LASSERTF(lu_fid_eq(&lli->lli_fid, &body->mbo_fid1),
2033                                  "Trying to change FID "DFID
2034                                  " to the "DFID", inode "DFID"(%p)\n",
2035                                  PFID(&lli->lli_fid), PFID(&body->mbo_fid1),
2036                                  PFID(ll_inode2fid(inode)), inode);
2037                 } else {
2038                         lli->lli_fid = body->mbo_fid1;
2039                 }
2040         }
2041
2042         LASSERT(fid_seq(&lli->lli_fid) != 0);
2043
2044         if (body->mbo_valid & OBD_MD_FLSIZE) {
2045                 i_size_write(inode, body->mbo_size);
2046
2047                 CDEBUG(D_VFSTRACE, "inode="DFID", updating i_size %llu\n",
2048                        PFID(ll_inode2fid(inode)),
2049                        (unsigned long long)body->mbo_size);
2050
2051                 if (body->mbo_valid & OBD_MD_FLBLOCKS)
2052                         inode->i_blocks = body->mbo_blocks;
2053         }
2054
2055         if (body->mbo_valid & OBD_MD_TSTATE) {
2056                 /* Set LLIF_FILE_RESTORING if restore ongoing and
2057                  * clear it when done to ensure to start again
2058                  * glimpsing updated attrs
2059                  */
2060                 if (body->mbo_t_state & MS_RESTORE)
2061                         ll_file_set_flag(lli, LLIF_FILE_RESTORING);
2062                 else
2063                         ll_file_clear_flag(lli, LLIF_FILE_RESTORING);
2064         }
2065
2066         return 0;
2067 }
2068
2069 int ll_read_inode2(struct inode *inode, void *opaque)
2070 {
2071         struct lustre_md *md = opaque;
2072         struct ll_inode_info *lli = ll_i2info(inode);
2073         int     rc;
2074         ENTRY;
2075
2076         CDEBUG(D_VFSTRACE, "VFS Op:inode="DFID"(%p)\n",
2077                PFID(&lli->lli_fid), inode);
2078
2079         /* Core attributes from the MDS first.  This is a new inode, and
2080          * the VFS doesn't zero times in the core inode so we have to do
2081          * it ourselves.  They will be overwritten by either MDS or OST
2082          * attributes - we just need to make sure they aren't newer. */
2083         LTIME_S(inode->i_mtime) = 0;
2084         LTIME_S(inode->i_atime) = 0;
2085         LTIME_S(inode->i_ctime) = 0;
2086         inode->i_rdev = 0;
2087         rc = ll_update_inode(inode, md);
2088         if (rc != 0)
2089                 RETURN(rc);
2090
2091         /* OIDEBUG(inode); */
2092
2093 #ifdef HAVE_BACKING_DEV_INFO
2094         /* initializing backing dev info. */
2095         inode->i_mapping->backing_dev_info = &s2lsi(inode->i_sb)->lsi_bdi;
2096 #endif
2097         if (S_ISREG(inode->i_mode)) {
2098                 struct ll_sb_info *sbi = ll_i2sbi(inode);
2099                 inode->i_op = &ll_file_inode_operations;
2100                 inode->i_fop = sbi->ll_fop;
2101                 inode->i_mapping->a_ops = (struct address_space_operations *)&ll_aops;
2102                 EXIT;
2103         } else if (S_ISDIR(inode->i_mode)) {
2104                 inode->i_op = &ll_dir_inode_operations;
2105                 inode->i_fop = &ll_dir_operations;
2106                 EXIT;
2107         } else if (S_ISLNK(inode->i_mode)) {
2108                 inode->i_op = &ll_fast_symlink_inode_operations;
2109                 EXIT;
2110         } else {
2111                 inode->i_op = &ll_special_inode_operations;
2112
2113                 init_special_inode(inode, inode->i_mode,
2114                                    inode->i_rdev);
2115
2116                 EXIT;
2117         }
2118
2119         return 0;
2120 }
2121
2122 void ll_delete_inode(struct inode *inode)
2123 {
2124         struct ll_inode_info *lli = ll_i2info(inode);
2125         ENTRY;
2126
2127         if (S_ISREG(inode->i_mode) && lli->lli_clob != NULL)
2128                 /* It is last chance to write out dirty pages,
2129                  * otherwise we may lose data while umount */
2130                 cl_sync_file_range(inode, 0, OBD_OBJECT_EOF, CL_FSYNC_LOCAL, 1);
2131
2132         truncate_inode_pages_final(&inode->i_data);
2133
2134         LASSERTF(inode->i_data.nrpages == 0, "inode="DFID"(%p) nrpages=%lu, "
2135                  "see https://jira.whamcloud.com/browse/LU-118\n",
2136                  PFID(ll_inode2fid(inode)), inode, inode->i_data.nrpages);
2137
2138 #ifdef HAVE_SBOPS_EVICT_INODE
2139         ll_clear_inode(inode);
2140 #endif
2141         clear_inode(inode);
2142
2143         EXIT;
2144 }
2145
2146 int ll_iocontrol(struct inode *inode, struct file *file,
2147                  unsigned int cmd, unsigned long arg)
2148 {
2149         struct ll_sb_info *sbi = ll_i2sbi(inode);
2150         struct ptlrpc_request *req = NULL;
2151         int rc, flags = 0;
2152         ENTRY;
2153
2154         switch (cmd) {
2155         case FS_IOC_GETFLAGS: {
2156                 struct mdt_body *body;
2157                 struct md_op_data *op_data;
2158
2159                 op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL,
2160                                              0, 0, LUSTRE_OPC_ANY,
2161                                              NULL);
2162                 if (IS_ERR(op_data))
2163                         RETURN(PTR_ERR(op_data));
2164
2165                 op_data->op_valid = OBD_MD_FLFLAGS;
2166                 rc = md_getattr(sbi->ll_md_exp, op_data, &req);
2167                 ll_finish_md_op_data(op_data);
2168                 if (rc) {
2169                         CERROR("%s: failure inode "DFID": rc = %d\n",
2170                                sbi->ll_md_exp->exp_obd->obd_name,
2171                                PFID(ll_inode2fid(inode)), rc);
2172                         RETURN(-abs(rc));
2173                 }
2174
2175                 body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
2176
2177                 flags = body->mbo_flags;
2178
2179                 ptlrpc_req_finished(req);
2180
2181                 RETURN(put_user(flags, (int __user *)arg));
2182         }
2183         case FS_IOC_SETFLAGS: {
2184                 struct iattr *attr;
2185                 struct md_op_data *op_data;
2186                 struct cl_object *obj;
2187
2188                 if (get_user(flags, (int __user *)arg))
2189                         RETURN(-EFAULT);
2190
2191                 op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL, 0, 0,
2192                                              LUSTRE_OPC_ANY, NULL);
2193                 if (IS_ERR(op_data))
2194                         RETURN(PTR_ERR(op_data));
2195
2196                 op_data->op_attr_flags = flags;
2197                 op_data->op_xvalid |= OP_XVALID_FLAGS;
2198                 rc = md_setattr(sbi->ll_md_exp, op_data, NULL, 0, &req);
2199                 ll_finish_md_op_data(op_data);
2200                 ptlrpc_req_finished(req);
2201                 if (rc)
2202                         RETURN(rc);
2203
2204                 ll_update_inode_flags(inode, flags);
2205
2206                 obj = ll_i2info(inode)->lli_clob;
2207                 if (obj == NULL)
2208                         RETURN(0);
2209
2210                 OBD_ALLOC_PTR(attr);
2211                 if (attr == NULL)
2212                         RETURN(-ENOMEM);
2213
2214                 rc = cl_setattr_ost(obj, attr, OP_XVALID_FLAGS, flags);
2215
2216                 OBD_FREE_PTR(attr);
2217                 RETURN(rc);
2218         }
2219         default:
2220                 RETURN(-ENOSYS);
2221         }
2222
2223         RETURN(0);
2224 }
2225
2226 int ll_flush_ctx(struct inode *inode)
2227 {
2228         struct ll_sb_info  *sbi = ll_i2sbi(inode);
2229
2230         CDEBUG(D_SEC, "flush context for user %d\n",
2231                from_kuid(&init_user_ns, current_uid()));
2232
2233         obd_set_info_async(NULL, sbi->ll_md_exp,
2234                            sizeof(KEY_FLUSH_CTX), KEY_FLUSH_CTX,
2235                            0, NULL, NULL);
2236         obd_set_info_async(NULL, sbi->ll_dt_exp,
2237                            sizeof(KEY_FLUSH_CTX), KEY_FLUSH_CTX,
2238                            0, NULL, NULL);
2239         return 0;
2240 }
2241
2242 /* umount -f client means force down, don't save state */
2243 void ll_umount_begin(struct super_block *sb)
2244 {
2245         struct ll_sb_info *sbi = ll_s2sbi(sb);
2246         struct obd_device *obd;
2247         struct obd_ioctl_data *ioc_data;
2248         struct l_wait_info lwi;
2249         wait_queue_head_t waitq;
2250         ENTRY;
2251
2252         CDEBUG(D_VFSTRACE, "VFS Op: superblock %p count %d active %d\n", sb,
2253                sb->s_count, atomic_read(&sb->s_active));
2254
2255         obd = class_exp2obd(sbi->ll_md_exp);
2256         if (obd == NULL) {
2257                 CERROR("Invalid MDC connection handle %#llx\n",
2258                        sbi->ll_md_exp->exp_handle.h_cookie);
2259                 EXIT;
2260                 return;
2261         }
2262         obd->obd_force = 1;
2263
2264         obd = class_exp2obd(sbi->ll_dt_exp);
2265         if (obd == NULL) {
2266                 CERROR("Invalid LOV connection handle %#llx\n",
2267                        sbi->ll_dt_exp->exp_handle.h_cookie);
2268                 EXIT;
2269                 return;
2270         }
2271         obd->obd_force = 1;
2272
2273         OBD_ALLOC_PTR(ioc_data);
2274         if (ioc_data) {
2275                 obd_iocontrol(IOC_OSC_SET_ACTIVE, sbi->ll_md_exp,
2276                               sizeof *ioc_data, ioc_data, NULL);
2277
2278                 obd_iocontrol(IOC_OSC_SET_ACTIVE, sbi->ll_dt_exp,
2279                               sizeof *ioc_data, ioc_data, NULL);
2280
2281                 OBD_FREE_PTR(ioc_data);
2282         }
2283
2284         /* Really, we'd like to wait until there are no requests outstanding,
2285          * and then continue.  For now, we just periodically checking for vfs
2286          * to decrement mnt_cnt and hope to finish it within 10sec.
2287          */
2288         init_waitqueue_head(&waitq);
2289         lwi = LWI_TIMEOUT_INTERVAL(cfs_time_seconds(10),
2290                                    cfs_time_seconds(1), NULL, NULL);
2291         l_wait_event(waitq, may_umount(sbi->ll_mnt.mnt), &lwi);
2292
2293         EXIT;
2294 }
2295
2296 int ll_remount_fs(struct super_block *sb, int *flags, char *data)
2297 {
2298         struct ll_sb_info *sbi = ll_s2sbi(sb);
2299         char *profilenm = get_profile_name(sb);
2300         int err;
2301         __u32 read_only;
2302
2303         if ((*flags & MS_RDONLY) != (sb->s_flags & MS_RDONLY)) {
2304                 read_only = *flags & MS_RDONLY;
2305                 err = obd_set_info_async(NULL, sbi->ll_md_exp,
2306                                          sizeof(KEY_READ_ONLY),
2307                                          KEY_READ_ONLY, sizeof(read_only),
2308                                          &read_only, NULL);
2309                 if (err) {
2310                         LCONSOLE_WARN("Failed to remount %s %s (%d)\n",
2311                                       profilenm, read_only ?
2312                                       "read-only" : "read-write", err);
2313                         return err;
2314                 }
2315
2316                 if (read_only)
2317                         sb->s_flags |= MS_RDONLY;
2318                 else
2319                         sb->s_flags &= ~MS_RDONLY;
2320
2321                 if (sbi->ll_flags & LL_SBI_VERBOSE)
2322                         LCONSOLE_WARN("Remounted %s %s\n", profilenm,
2323                                       read_only ?  "read-only" : "read-write");
2324         }
2325         return 0;
2326 }
2327
2328 /**
2329  * Cleanup the open handle that is cached on MDT-side.
2330  *
2331  * For open case, the client side open handling thread may hit error
2332  * after the MDT grant the open. Under such case, the client should
2333  * send close RPC to the MDT as cleanup; otherwise, the open handle
2334  * on the MDT will be leaked there until the client umount or evicted.
2335  *
2336  * In further, if someone unlinked the file, because the open handle
2337  * holds the reference on such file/object, then it will block the
2338  * subsequent threads that want to locate such object via FID.
2339  *
2340  * \param[in] sb        super block for this file-system
2341  * \param[in] open_req  pointer to the original open request
2342  */
2343 void ll_open_cleanup(struct super_block *sb, struct ptlrpc_request *open_req)
2344 {
2345         struct mdt_body                 *body;
2346         struct md_op_data               *op_data;
2347         struct ptlrpc_request           *close_req = NULL;
2348         struct obd_export               *exp       = ll_s2sbi(sb)->ll_md_exp;
2349         ENTRY;
2350
2351         body = req_capsule_server_get(&open_req->rq_pill, &RMF_MDT_BODY);
2352         OBD_ALLOC_PTR(op_data);
2353         if (op_data == NULL) {
2354                 CWARN("%s: cannot allocate op_data to release open handle for "
2355                       DFID"\n",
2356                       ll_get_fsname(sb, NULL, 0), PFID(&body->mbo_fid1));
2357
2358                 RETURN_EXIT;
2359         }
2360
2361         op_data->op_fid1 = body->mbo_fid1;
2362         op_data->op_open_handle = body->mbo_open_handle;
2363         op_data->op_mod_time = ktime_get_real_seconds();
2364         md_close(exp, op_data, NULL, &close_req);
2365         ptlrpc_req_finished(close_req);
2366         ll_finish_md_op_data(op_data);
2367
2368         EXIT;
2369 }
2370
2371 int ll_prep_inode(struct inode **inode, struct ptlrpc_request *req,
2372                   struct super_block *sb, struct lookup_intent *it)
2373 {
2374         struct ll_sb_info *sbi = NULL;
2375         struct lustre_md md = { NULL };
2376         int rc;
2377         ENTRY;
2378
2379         LASSERT(*inode || sb);
2380         sbi = sb ? ll_s2sbi(sb) : ll_i2sbi(*inode);
2381         rc = md_get_lustre_md(sbi->ll_md_exp, req, sbi->ll_dt_exp,
2382                               sbi->ll_md_exp, &md);
2383         if (rc != 0)
2384                 GOTO(cleanup, rc);
2385
2386         if (*inode) {
2387                 rc = ll_update_inode(*inode, &md);
2388                 if (rc != 0)
2389                         GOTO(out, rc);
2390         } else {
2391                 LASSERT(sb != NULL);
2392
2393                 /*
2394                  * At this point server returns to client's same fid as client
2395                  * generated for creating. So using ->fid1 is okay here.
2396                  */
2397                 if (!fid_is_sane(&md.body->mbo_fid1)) {
2398                         CERROR("%s: Fid is insane "DFID"\n",
2399                                 ll_get_fsname(sb, NULL, 0),
2400                                 PFID(&md.body->mbo_fid1));
2401                         GOTO(out, rc = -EINVAL);
2402                 }
2403
2404                 *inode = ll_iget(sb, cl_fid_build_ino(&md.body->mbo_fid1,
2405                                              sbi->ll_flags & LL_SBI_32BIT_API),
2406                                  &md);
2407                 if (IS_ERR(*inode)) {
2408 #ifdef CONFIG_FS_POSIX_ACL
2409                         if (md.posix_acl) {
2410                                 posix_acl_release(md.posix_acl);
2411                                 md.posix_acl = NULL;
2412                         }
2413 #endif
2414                         rc = IS_ERR(*inode) ? PTR_ERR(*inode) : -ENOMEM;
2415                         *inode = NULL;
2416                         CERROR("new_inode -fatal: rc %d\n", rc);
2417                         GOTO(out, rc);
2418                 }
2419         }
2420
2421         /* Handling piggyback layout lock.
2422          * Layout lock can be piggybacked by getattr and open request.
2423          * The lsm can be applied to inode only if it comes with a layout lock
2424          * otherwise correct layout may be overwritten, for example:
2425          * 1. proc1: mdt returns a lsm but not granting layout
2426          * 2. layout was changed by another client
2427          * 3. proc2: refresh layout and layout lock granted
2428          * 4. proc1: to apply a stale layout */
2429         if (it != NULL && it->it_lock_mode != 0) {
2430                 struct lustre_handle lockh;
2431                 struct ldlm_lock *lock;
2432
2433                 lockh.cookie = it->it_lock_handle;
2434                 lock = ldlm_handle2lock(&lockh);
2435                 LASSERT(lock != NULL);
2436                 if (ldlm_has_layout(lock)) {
2437                         struct cl_object_conf conf;
2438
2439                         memset(&conf, 0, sizeof(conf));
2440                         conf.coc_opc = OBJECT_CONF_SET;
2441                         conf.coc_inode = *inode;
2442                         conf.coc_lock = lock;
2443                         conf.u.coc_layout = md.layout;
2444                         (void)ll_layout_conf(*inode, &conf);
2445                 }
2446                 LDLM_LOCK_PUT(lock);
2447         }
2448
2449         GOTO(out, rc = 0);
2450
2451 out:
2452         md_free_lustre_md(sbi->ll_md_exp, &md);
2453
2454 cleanup:
2455         if (rc != 0 && it != NULL && it->it_op & IT_OPEN)
2456                 ll_open_cleanup(sb != NULL ? sb : (*inode)->i_sb, req);
2457
2458         return rc;
2459 }
2460
2461 int ll_obd_statfs(struct inode *inode, void __user *arg)
2462 {
2463         struct ll_sb_info *sbi = NULL;
2464         struct obd_export *exp;
2465         char *buf = NULL;
2466         struct obd_ioctl_data *data = NULL;
2467         __u32 type;
2468         int len = 0, rc;
2469
2470         if (!inode || !(sbi = ll_i2sbi(inode)))
2471                 GOTO(out_statfs, rc = -EINVAL);
2472
2473         rc = obd_ioctl_getdata(&buf, &len, arg);
2474         if (rc)
2475                 GOTO(out_statfs, rc);
2476
2477         data = (void*)buf;
2478         if (!data->ioc_inlbuf1 || !data->ioc_inlbuf2 ||
2479             !data->ioc_pbuf1 || !data->ioc_pbuf2)
2480                 GOTO(out_statfs, rc = -EINVAL);
2481
2482         if (data->ioc_inllen1 != sizeof(__u32) ||
2483             data->ioc_inllen2 != sizeof(__u32) ||
2484             data->ioc_plen1 != sizeof(struct obd_statfs) ||
2485             data->ioc_plen2 != sizeof(struct obd_uuid))
2486                 GOTO(out_statfs, rc = -EINVAL);
2487
2488         memcpy(&type, data->ioc_inlbuf1, sizeof(__u32));
2489         if (type & LL_STATFS_LMV)
2490                 exp = sbi->ll_md_exp;
2491         else if (type & LL_STATFS_LOV)
2492                 exp = sbi->ll_dt_exp;
2493         else
2494                 GOTO(out_statfs, rc = -ENODEV);
2495
2496         rc = obd_iocontrol(IOC_OBD_STATFS, exp, len, buf, NULL);
2497         if (rc)
2498                 GOTO(out_statfs, rc);
2499 out_statfs:
2500         OBD_FREE_LARGE(buf, len);
2501         return rc;
2502 }
2503
2504 /* this function prepares md_op_data hint for passing it down to MD stack. */
2505 struct md_op_data *ll_prep_md_op_data(struct md_op_data *op_data,
2506                                       struct inode *i1, struct inode *i2,
2507                                       const char *name, size_t namelen,
2508                                       __u32 mode, __u32 opc, void *data)
2509 {
2510         LASSERT(i1 != NULL);
2511
2512         if (name == NULL) {
2513                 /* Do not reuse namelen for something else. */
2514                 if (namelen != 0)
2515                         return ERR_PTR(-EINVAL);
2516         } else {
2517                 if (namelen > ll_i2sbi(i1)->ll_namelen)
2518                         return ERR_PTR(-ENAMETOOLONG);
2519
2520                 if (!lu_name_is_valid_2(name, namelen))
2521                         return ERR_PTR(-EINVAL);
2522         }
2523
2524         if (op_data == NULL)
2525                 OBD_ALLOC_PTR(op_data);
2526
2527         if (op_data == NULL)
2528                 return ERR_PTR(-ENOMEM);
2529
2530         ll_i2gids(op_data->op_suppgids, i1, i2);
2531         op_data->op_fid1 = *ll_inode2fid(i1);
2532         op_data->op_default_stripe_offset = -1;
2533         if (S_ISDIR(i1->i_mode)) {
2534                 op_data->op_mea1 = ll_i2info(i1)->lli_lsm_md;
2535                 if (opc == LUSTRE_OPC_MKDIR)
2536                         op_data->op_default_stripe_offset =
2537                                    ll_i2info(i1)->lli_def_stripe_offset;
2538         }
2539
2540         if (i2) {
2541                 op_data->op_fid2 = *ll_inode2fid(i2);
2542                 if (S_ISDIR(i2->i_mode))
2543                         op_data->op_mea2 = ll_i2info(i2)->lli_lsm_md;
2544         } else {
2545                 fid_zero(&op_data->op_fid2);
2546         }
2547
2548         if (ll_i2sbi(i1)->ll_flags & LL_SBI_64BIT_HASH)
2549                 op_data->op_cli_flags |= CLI_HASH64;
2550
2551         if (ll_need_32bit_api(ll_i2sbi(i1)))
2552                 op_data->op_cli_flags |= CLI_API32;
2553
2554         op_data->op_name = name;
2555         op_data->op_namelen = namelen;
2556         op_data->op_mode = mode;
2557         op_data->op_mod_time = ktime_get_real_seconds();
2558         op_data->op_fsuid = from_kuid(&init_user_ns, current_fsuid());
2559         op_data->op_fsgid = from_kgid(&init_user_ns, current_fsgid());
2560         op_data->op_cap = cfs_curproc_cap_pack();
2561         op_data->op_mds = 0;
2562         if ((opc == LUSTRE_OPC_CREATE) && (name != NULL) &&
2563              filename_is_volatile(name, namelen, &op_data->op_mds)) {
2564                 op_data->op_bias |= MDS_CREATE_VOLATILE;
2565         }
2566         op_data->op_data = data;
2567
2568         return op_data;
2569 }
2570
2571 void ll_finish_md_op_data(struct md_op_data *op_data)
2572 {
2573         security_release_secctx(op_data->op_file_secctx,
2574                                 op_data->op_file_secctx_size);
2575         OBD_FREE_PTR(op_data);
2576 }
2577
2578 #ifdef HAVE_SUPEROPS_USE_DENTRY
2579 int ll_show_options(struct seq_file *seq, struct dentry *dentry)
2580 #else
2581 int ll_show_options(struct seq_file *seq, struct vfsmount *vfs)
2582 #endif
2583 {
2584         struct ll_sb_info *sbi;
2585
2586 #ifdef HAVE_SUPEROPS_USE_DENTRY
2587         LASSERT((seq != NULL) && (dentry != NULL));
2588         sbi = ll_s2sbi(dentry->d_sb);
2589 #else
2590         LASSERT((seq != NULL) && (vfs != NULL));
2591         sbi = ll_s2sbi(vfs->mnt_sb);
2592 #endif
2593
2594         if (sbi->ll_flags & LL_SBI_NOLCK)
2595                 seq_puts(seq, ",nolock");
2596
2597         if (sbi->ll_flags & LL_SBI_FLOCK)
2598                 seq_puts(seq, ",flock");
2599
2600         if (sbi->ll_flags & LL_SBI_LOCALFLOCK)
2601                 seq_puts(seq, ",localflock");
2602
2603         if (sbi->ll_flags & LL_SBI_USER_XATTR)
2604                 seq_puts(seq, ",user_xattr");
2605
2606         if (sbi->ll_flags & LL_SBI_LAZYSTATFS)
2607                 seq_puts(seq, ",lazystatfs");
2608
2609         if (sbi->ll_flags & LL_SBI_USER_FID2PATH)
2610                 seq_puts(seq, ",user_fid2path");
2611
2612         if (sbi->ll_flags & LL_SBI_ALWAYS_PING)
2613                 seq_puts(seq, ",always_ping");
2614
2615         RETURN(0);
2616 }
2617
2618 /**
2619  * Get obd name by cmd, and copy out to user space
2620  */
2621 int ll_get_obd_name(struct inode *inode, unsigned int cmd, unsigned long arg)
2622 {
2623         struct ll_sb_info *sbi = ll_i2sbi(inode);
2624         struct obd_device *obd;
2625         ENTRY;
2626
2627         if (cmd == OBD_IOC_GETDTNAME)
2628                 obd = class_exp2obd(sbi->ll_dt_exp);
2629         else if (cmd == OBD_IOC_GETMDNAME)
2630                 obd = class_exp2obd(sbi->ll_md_exp);
2631         else
2632                 RETURN(-EINVAL);
2633
2634         if (!obd)
2635                 RETURN(-ENOENT);
2636
2637         if (copy_to_user((void __user *)arg, obd->obd_name,
2638                          strlen(obd->obd_name) + 1))
2639                 RETURN(-EFAULT);
2640
2641         RETURN(0);
2642 }
2643
2644 /**
2645  * Get lustre file system name by \a sbi. If \a buf is provided(non-NULL), the
2646  * fsname will be returned in this buffer; otherwise, a static buffer will be
2647  * used to store the fsname and returned to caller.
2648  */
2649 char *ll_get_fsname(struct super_block *sb, char *buf, int buflen)
2650 {
2651         static char fsname_static[MTI_NAME_MAXLEN];
2652         struct lustre_sb_info *lsi = s2lsi(sb);
2653         char *ptr;
2654         int len;
2655
2656         if (buf == NULL) {
2657                 /* this means the caller wants to use static buffer
2658                  * and it doesn't care about race. Usually this is
2659                  * in error reporting path */
2660                 buf = fsname_static;
2661                 buflen = sizeof(fsname_static);
2662         }
2663
2664         len = strlen(lsi->lsi_lmd->lmd_profile);
2665         ptr = strrchr(lsi->lsi_lmd->lmd_profile, '-');
2666         if (ptr && (strcmp(ptr, "-client") == 0))
2667                 len -= 7;
2668
2669         if (unlikely(len >= buflen))
2670                 len = buflen - 1;
2671         strncpy(buf, lsi->lsi_lmd->lmd_profile, len);
2672         buf[len] = '\0';
2673
2674         return buf;
2675 }
2676
2677 static char* ll_d_path(struct dentry *dentry, char *buf, int bufsize)
2678 {
2679         char *path = NULL;
2680
2681         struct path p;
2682
2683         p.dentry = dentry;
2684         p.mnt = current->fs->root.mnt;
2685         path_get(&p);
2686         path = d_path(&p, buf, bufsize);
2687         path_put(&p);
2688         return path;
2689 }
2690
2691 void ll_dirty_page_discard_warn(struct page *page, int ioret)
2692 {
2693         char *buf, *path = NULL;
2694         struct dentry *dentry = NULL;
2695         struct inode *inode = page->mapping->host;
2696
2697         /* this can be called inside spin lock so use GFP_ATOMIC. */
2698         buf = (char *)__get_free_page(GFP_ATOMIC);
2699         if (buf != NULL) {
2700                 dentry = d_find_alias(page->mapping->host);
2701                 if (dentry != NULL)
2702                         path = ll_d_path(dentry, buf, PAGE_SIZE);
2703         }
2704
2705         CDEBUG(D_WARNING,
2706                "%s: dirty page discard: %s/fid: "DFID"/%s may get corrupted "
2707                "(rc %d)\n", ll_get_fsname(page->mapping->host->i_sb, NULL, 0),
2708                s2lsi(page->mapping->host->i_sb)->lsi_lmd->lmd_dev,
2709                PFID(ll_inode2fid(inode)),
2710                (path && !IS_ERR(path)) ? path : "", ioret);
2711
2712         if (dentry != NULL)
2713                 dput(dentry);
2714
2715         if (buf != NULL)
2716                 free_page((unsigned long)buf);
2717 }
2718
2719 ssize_t ll_copy_user_md(const struct lov_user_md __user *md,
2720                         struct lov_user_md **kbuf)
2721 {
2722         struct lov_user_md      lum;
2723         ssize_t                 lum_size;
2724         ENTRY;
2725
2726         if (copy_from_user(&lum, md, sizeof(lum)))
2727                 RETURN(-EFAULT);
2728
2729         lum_size = ll_lov_user_md_size(&lum);
2730         if (lum_size < 0)
2731                 RETURN(lum_size);
2732
2733         OBD_ALLOC(*kbuf, lum_size);
2734         if (*kbuf == NULL)
2735                 RETURN(-ENOMEM);
2736
2737         if (copy_from_user(*kbuf, md, lum_size) != 0) {
2738                 OBD_FREE(*kbuf, lum_size);
2739                 RETURN(-EFAULT);
2740         }
2741
2742         RETURN(lum_size);
2743 }
2744
2745 /*
2746  * Compute llite root squash state after a change of root squash
2747  * configuration setting or add/remove of a lnet nid
2748  */
2749 void ll_compute_rootsquash_state(struct ll_sb_info *sbi)
2750 {
2751         struct root_squash_info *squash = &sbi->ll_squash;
2752         int i;
2753         bool matched;
2754         struct lnet_process_id id;
2755
2756         /* Update norootsquash flag */
2757         down_write(&squash->rsi_sem);
2758         if (list_empty(&squash->rsi_nosquash_nids))
2759                 sbi->ll_flags &= ~LL_SBI_NOROOTSQUASH;
2760         else {
2761                 /* Do not apply root squash as soon as one of our NIDs is
2762                  * in the nosquash_nids list */
2763                 matched = false;
2764                 i = 0;
2765                 while (LNetGetId(i++, &id) != -ENOENT) {
2766                         if (LNET_NETTYP(LNET_NIDNET(id.nid)) == LOLND)
2767                                 continue;
2768                         if (cfs_match_nid(id.nid, &squash->rsi_nosquash_nids)) {
2769                                 matched = true;
2770                                 break;
2771                         }
2772                 }
2773                 if (matched)
2774                         sbi->ll_flags |= LL_SBI_NOROOTSQUASH;
2775                 else
2776                         sbi->ll_flags &= ~LL_SBI_NOROOTSQUASH;
2777         }
2778         up_write(&squash->rsi_sem);
2779 }
2780
2781 /**
2782  * Parse linkea content to extract information about a given hardlink
2783  *
2784  * \param[in]   ldata      - Initialized linkea data
2785  * \param[in]   linkno     - Link identifier
2786  * \param[out]  parent_fid - The entry's parent FID
2787  * \param[out]  ln         - Entry name destination buffer
2788  *
2789  * \retval 0 on success
2790  * \retval Appropriate negative error code on failure
2791  */
2792 static int ll_linkea_decode(struct linkea_data *ldata, unsigned int linkno,
2793                             struct lu_fid *parent_fid, struct lu_name *ln)
2794 {
2795         unsigned int    idx;
2796         int             rc;
2797         ENTRY;
2798
2799         rc = linkea_init_with_rec(ldata);
2800         if (rc < 0)
2801                 RETURN(rc);
2802
2803         if (linkno >= ldata->ld_leh->leh_reccount)
2804                 /* beyond last link */
2805                 RETURN(-ENODATA);
2806
2807         linkea_first_entry(ldata);
2808         for (idx = 0; ldata->ld_lee != NULL; idx++) {
2809                 linkea_entry_unpack(ldata->ld_lee, &ldata->ld_reclen, ln,
2810                                     parent_fid);
2811                 if (idx == linkno)
2812                         break;
2813
2814                 linkea_next_entry(ldata);
2815         }
2816
2817         if (idx < linkno)
2818                 RETURN(-ENODATA);
2819
2820         RETURN(0);
2821 }
2822
2823 /**
2824  * Get parent FID and name of an identified link. Operation is performed for
2825  * a given link number, letting the caller iterate over linkno to list one or
2826  * all links of an entry.
2827  *
2828  * \param[in]     file - File descriptor against which to perform the operation
2829  * \param[in,out] arg  - User-filled structure containing the linkno to operate
2830  *                       on and the available size. It is eventually filled with
2831  *                       the requested information or left untouched on error
2832  *
2833  * \retval - 0 on success
2834  * \retval - Appropriate negative error code on failure
2835  */
2836 int ll_getparent(struct file *file, struct getparent __user *arg)
2837 {
2838         struct inode            *inode = file_inode(file);
2839         struct linkea_data      *ldata;
2840         struct lu_buf            buf = LU_BUF_NULL;
2841         struct lu_name           ln;
2842         struct lu_fid            parent_fid;
2843         __u32                    linkno;
2844         __u32                    name_size;
2845         int                      rc;
2846
2847         ENTRY;
2848
2849         if (!cfs_capable(CFS_CAP_DAC_READ_SEARCH) &&
2850             !(ll_i2sbi(inode)->ll_flags & LL_SBI_USER_FID2PATH))
2851                 RETURN(-EPERM);
2852
2853         if (get_user(name_size, &arg->gp_name_size))
2854                 RETURN(-EFAULT);
2855
2856         if (get_user(linkno, &arg->gp_linkno))
2857                 RETURN(-EFAULT);
2858
2859         if (name_size > PATH_MAX)
2860                 RETURN(-EINVAL);
2861
2862         OBD_ALLOC(ldata, sizeof(*ldata));
2863         if (ldata == NULL)
2864                 RETURN(-ENOMEM);
2865
2866         rc = linkea_data_new(ldata, &buf);
2867         if (rc < 0)
2868                 GOTO(ldata_free, rc);
2869
2870 #ifdef HAVE_XATTR_HANDLER_FLAGS
2871         rc = ll_xattr_list(inode, XATTR_NAME_LINK, XATTR_TRUSTED_T, buf.lb_buf,
2872                            buf.lb_len, OBD_MD_FLXATTR);
2873 #else
2874         rc = ll_getxattr(file_dentry(file), XATTR_NAME_LINK, buf.lb_buf,
2875                          buf.lb_len);
2876 #endif /* HAVE_XATTR_HANDLER_FLAGS */
2877         if (rc < 0)
2878                 GOTO(lb_free, rc);
2879
2880         rc = ll_linkea_decode(ldata, linkno, &parent_fid, &ln);
2881         if (rc < 0)
2882                 GOTO(lb_free, rc);
2883
2884         if (ln.ln_namelen >= name_size)
2885                 GOTO(lb_free, rc = -EOVERFLOW);
2886
2887         if (copy_to_user(&arg->gp_fid, &parent_fid, sizeof(arg->gp_fid)))
2888                 GOTO(lb_free, rc = -EFAULT);
2889
2890         if (copy_to_user(&arg->gp_name, ln.ln_name, ln.ln_namelen))
2891                 GOTO(lb_free, rc = -EFAULT);
2892
2893         if (put_user('\0', arg->gp_name + ln.ln_namelen))
2894                 GOTO(lb_free, rc = -EFAULT);
2895
2896 lb_free:
2897         lu_buf_free(&buf);
2898 ldata_free:
2899         OBD_FREE(ldata, sizeof(*ldata));
2900
2901         RETURN(rc);
2902 }