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.sun.com/software/products/lustre/docs/GPLv2.pdf
  19  *
  20  * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara,
  21  * CA 95054 USA or visit www.sun.com if you need additional information or
  22  * have any questions.
  23  *
  24  * GPL HEADER END
  25  */
  26 /*
  27  * Copyright (c) 2003, 2010, Oracle and/or its affiliates. All rights reserved.
  28  * Use is subject to license terms.
  29  *
  30  * Copyright (c) 2011, 2013, Intel Corporation.
  31  */
  32 /*
  33  * This file is part of Lustre, http://www.lustre.org/
  34  * Lustre is a trademark of Sun Microsystems, Inc.
  35  *
  36  * lustre/llite/llite_lib.c
  37  *
  38  * Lustre Light Super operations
  39  */
  40
  41 #define DEBUG_SUBSYSTEM S_LLITE
  42
  43 #include <linux/module.h>
  44 #include <linux/types.h>
  45 #include <linux/version.h>
  46 #include <linux/mm.h>
  47
  48 #include <lustre_ioctl.h>
  49 #include <lustre_lite.h>
  50 #include <lustre_ha.h>
  51 #include <lustre_dlm.h>
  52 #include <lprocfs_status.h>
  53 #include <lustre_disk.h>
  54 #include <lustre_param.h>
  55 #include <lustre_log.h>
  56 #include <cl_object.h>
  57 #include <obd_cksum.h>
  58 #include "llite_internal.h"
  59
  60 struct kmem_cache *ll_file_data_slab;
  61
  62 static struct list_head ll_super_blocks = LIST_HEAD_INIT(ll_super_blocks);
  63 static DEFINE_SPINLOCK(ll_sb_lock);
  64
  65 #ifndef log2
  66 #define log2(n) ffz(~(n))
  67 #endif
  68
  69 static struct ll_sb_info *ll_init_sbi(void)
  70 {
  71         struct ll_sb_info *sbi = NULL;
  72         unsigned long pages;
  73         unsigned long lru_page_max;
  74         struct sysinfo si;
  75         class_uuid_t uuid;
  76         int i;
  77         ENTRY;
  78
  79         OBD_ALLOC_PTR(sbi);
  80         if (sbi == NULL)
  81                 RETURN(NULL);
  82
  83         spin_lock_init(&sbi->ll_lock);
  84         mutex_init(&sbi->ll_lco.lco_lock);
  85         spin_lock_init(&sbi->ll_pp_extent_lock);
  86         spin_lock_init(&sbi->ll_process_lock);
  87         sbi->ll_rw_stats_on = 0;
  88
  89         si_meminfo(&si);
  90         pages = si.totalram - si.totalhigh;
  91         lru_page_max = pages / 2;
  92
  93         /* initialize ll_cache data */
  94         atomic_set(&sbi->ll_cache.ccc_users, 0);
  95         sbi->ll_cache.ccc_lru_max = lru_page_max;
  96         atomic_set(&sbi->ll_cache.ccc_lru_left, lru_page_max);
  97         spin_lock_init(&sbi->ll_cache.ccc_lru_lock);
  98         INIT_LIST_HEAD(&sbi->ll_cache.ccc_lru);
  99
 100         atomic_set(&sbi->ll_cache.ccc_unstable_nr, 0);
 101         init_waitqueue_head(&sbi->ll_cache.ccc_unstable_waitq);
 102
 103         sbi->ll_ra_info.ra_max_pages_per_file = min(pages / 32,
 104                                            SBI_DEFAULT_READAHEAD_MAX);
 105         sbi->ll_ra_info.ra_max_pages = sbi->ll_ra_info.ra_max_pages_per_file;
 106         sbi->ll_ra_info.ra_max_read_ahead_whole_pages =
 107                                            SBI_DEFAULT_READAHEAD_WHOLE_MAX;
 108         INIT_LIST_HEAD(&sbi->ll_conn_chain);
 109         INIT_LIST_HEAD(&sbi->ll_orphan_dentry_list);
 110
 111         ll_generate_random_uuid(uuid);
 112         class_uuid_unparse(uuid, &sbi->ll_sb_uuid);
 113         CDEBUG(D_CONFIG, "generated uuid: %s\n", sbi->ll_sb_uuid.uuid);
 114
 115         spin_lock(&ll_sb_lock);
 116         list_add_tail(&sbi->ll_list, &ll_super_blocks);
 117         spin_unlock(&ll_sb_lock);
 118
 119         sbi->ll_flags |= LL_SBI_VERBOSE;
 120 #ifdef ENABLE_CHECKSUM
 121         sbi->ll_flags |= LL_SBI_CHECKSUM;
 122 #endif
 123
 124 #ifdef HAVE_LRU_RESIZE_SUPPORT
 125         sbi->ll_flags |= LL_SBI_LRU_RESIZE;
 126 #endif
 127
 128         for (i = 0; i <= LL_PROCESS_HIST_MAX; i++) {
 129                 spin_lock_init(&sbi->ll_rw_extents_info.pp_extents[i].
 130                                pp_r_hist.oh_lock);
 131                 spin_lock_init(&sbi->ll_rw_extents_info.pp_extents[i].
 132                                pp_w_hist.oh_lock);
 133         }
 134
 135         /* metadata statahead is enabled by default */
 136         sbi->ll_sa_max = LL_SA_RPC_DEF;
 137         atomic_set(&sbi->ll_sa_total, 0);
 138         atomic_set(&sbi->ll_sa_wrong, 0);
 139         atomic_set(&sbi->ll_agl_total, 0);
 140         sbi->ll_flags |= LL_SBI_AGL_ENABLED;
 141
 142         /* root squash */
 143         sbi->ll_squash.rsi_uid = 0;
 144         sbi->ll_squash.rsi_gid = 0;
 145         INIT_LIST_HEAD(&sbi->ll_squash.rsi_nosquash_nids);
 146         init_rwsem(&sbi->ll_squash.rsi_sem);
 147
 148         RETURN(sbi);
 149 }
 150
 151 static void ll_free_sbi(struct super_block *sb)
 152 {
 153         struct ll_sb_info *sbi = ll_s2sbi(sb);
 154         ENTRY;
 155
 156         if (sbi != NULL) {
 157                 spin_lock(&ll_sb_lock);
 158                 list_del(&sbi->ll_list);
 159                 spin_unlock(&ll_sb_lock);
 160                 if (!list_empty(&sbi->ll_squash.rsi_nosquash_nids))
 161                         cfs_free_nidlist(&sbi->ll_squash.rsi_nosquash_nids);
 162                 OBD_FREE(sbi, sizeof(*sbi));
 163         }
 164         EXIT;
 165 }
 166
 167 static int client_common_fill_super(struct super_block *sb, char *md, char *dt,
 168                                     struct vfsmount *mnt)
 169 {
 170         struct inode *root = 0;
 171         struct ll_sb_info *sbi = ll_s2sbi(sb);
 172         struct obd_device *obd;
 173         struct obd_capa *oc = NULL;
 174         struct obd_statfs *osfs = NULL;
 175         struct ptlrpc_request *request = NULL;
 176         struct obd_connect_data *data = NULL;
 177         struct obd_uuid *uuid;
 178         struct md_op_data *op_data;
 179         struct lustre_md lmd;
 180         obd_valid valid;
 181         int size, err, checksum;
 182         ENTRY;
 183
 184         obd = class_name2obd(md);
 185         if (!obd) {
 186                 CERROR("MD %s: not setup or attached\n", md);
 187                 RETURN(-EINVAL);
 188         }
 189
 190         OBD_ALLOC_PTR(data);
 191         if (data == NULL)
 192                 RETURN(-ENOMEM);
 193
 194         OBD_ALLOC_PTR(osfs);
 195         if (osfs == NULL) {
 196                 OBD_FREE_PTR(data);
 197                 RETURN(-ENOMEM);
 198         }
 199
 200         /* indicate the features supported by this client */
 201         data->ocd_connect_flags = OBD_CONNECT_IBITS    | OBD_CONNECT_NODEVOH  |
 202                                   OBD_CONNECT_ATTRFID  |
 203                                   OBD_CONNECT_VERSION  | OBD_CONNECT_BRW_SIZE |
 204                                   OBD_CONNECT_MDS_CAPA | OBD_CONNECT_OSS_CAPA |
 205                                   OBD_CONNECT_CANCELSET | OBD_CONNECT_FID     |
 206                                   OBD_CONNECT_AT       | OBD_CONNECT_LOV_V3   |
 207                                   OBD_CONNECT_RMT_CLIENT | OBD_CONNECT_VBR    |
 208                                   OBD_CONNECT_FULL20   | OBD_CONNECT_64BITHASH|
 209                                   OBD_CONNECT_EINPROGRESS |
 210                                   OBD_CONNECT_JOBSTATS | OBD_CONNECT_LVB_TYPE |
 211                                   OBD_CONNECT_LAYOUTLOCK | OBD_CONNECT_PINGLESS |
 212                                   OBD_CONNECT_MAX_EASIZE |
 213                                   OBD_CONNECT_FLOCK_DEAD |
 214                                   OBD_CONNECT_DISP_STRIPE | OBD_CONNECT_LFSCK;
 215
 216         if (sbi->ll_flags & LL_SBI_SOM_PREVIEW)
 217                 data->ocd_connect_flags |= OBD_CONNECT_SOM;
 218
 219 #ifdef HAVE_LRU_RESIZE_SUPPORT
 220         if (sbi->ll_flags & LL_SBI_LRU_RESIZE)
 221                 data->ocd_connect_flags |= OBD_CONNECT_LRU_RESIZE;
 222 #endif
 223 #ifdef CONFIG_FS_POSIX_ACL
 224         data->ocd_connect_flags |= OBD_CONNECT_ACL | OBD_CONNECT_UMASK;
 225 #endif
 226
 227         if (OBD_FAIL_CHECK(OBD_FAIL_MDC_LIGHTWEIGHT))
 228                 /* flag mdc connection as lightweight, only used for test
 229                  * purpose, use with care */
 230                 data->ocd_connect_flags |= OBD_CONNECT_LIGHTWEIGHT;
 231
 232         data->ocd_ibits_known = MDS_INODELOCK_FULL;
 233         data->ocd_version = LUSTRE_VERSION_CODE;
 234
 235         if (sb->s_flags & MS_RDONLY)
 236                 data->ocd_connect_flags |= OBD_CONNECT_RDONLY;
 237         if (sbi->ll_flags & LL_SBI_USER_XATTR)
 238                 data->ocd_connect_flags |= OBD_CONNECT_XATTR;
 239
 240 #ifdef HAVE_MS_FLOCK_LOCK
 241         /* force vfs to use lustre handler for flock() calls - bug 10743 */
 242         sb->s_flags |= MS_FLOCK_LOCK;
 243 #endif
 244 #ifdef MS_HAS_NEW_AOPS
 245         sb->s_flags |= MS_HAS_NEW_AOPS;
 246 #endif
 247
 248         if (sbi->ll_flags & LL_SBI_FLOCK)
 249                 sbi->ll_fop = &ll_file_operations_flock;
 250         else if (sbi->ll_flags & LL_SBI_LOCALFLOCK)
 251                 sbi->ll_fop = &ll_file_operations;
 252         else
 253                 sbi->ll_fop = &ll_file_operations_noflock;
 254
 255         /* real client */
 256         data->ocd_connect_flags |= OBD_CONNECT_REAL;
 257         if (sbi->ll_flags & LL_SBI_RMT_CLIENT)
 258                 data->ocd_connect_flags |= OBD_CONNECT_RMT_CLIENT_FORCE;
 259
 260         data->ocd_brw_size = MD_MAX_BRW_SIZE;
 261
 262         err = obd_connect(NULL, &sbi->ll_md_exp, obd, &sbi->ll_sb_uuid, data, NULL);
 263         if (err == -EBUSY) {
 264                 LCONSOLE_ERROR_MSG(0x14f, "An MDT (md %s) is performing "
 265                                    "recovery, of which this client is not a "
 266                                    "part. Please wait for recovery to complete,"
 267                                    " abort, or time out.\n", md);
 268                 GOTO(out, err);
 269         } else if (err) {
 270                 CERROR("cannot connect to %s: rc = %d\n", md, err);
 271                 GOTO(out, err);
 272         }
 273
 274         sbi->ll_md_exp->exp_connect_data = *data;
 275
 276         err = obd_fid_init(sbi->ll_md_exp->exp_obd, sbi->ll_md_exp,
 277                            LUSTRE_SEQ_METADATA);
 278         if (err) {
 279                 CERROR("%s: Can't init metadata layer FID infrastructure, "
 280                        "rc = %d\n", sbi->ll_md_exp->exp_obd->obd_name, err);
 281                 GOTO(out_md, err);
 282         }
 283
 284         /* For mount, we only need fs info from MDT0, and also in DNE, it
 285          * can make sure the client can be mounted as long as MDT0 is
 286          * avaible */
 287         err = obd_statfs(NULL, sbi->ll_md_exp, osfs,
 288                         cfs_time_shift_64(-OBD_STATFS_CACHE_SECONDS),
 289                         OBD_STATFS_FOR_MDT0);
 290         if (err)
 291                 GOTO(out_md_fid, err);
 292
 293         /* This needs to be after statfs to ensure connect has finished.
 294          * Note that "data" does NOT contain the valid connect reply.
 295          * If connecting to a 1.8 server there will be no LMV device, so
 296          * we can access the MDC export directly and exp_connect_flags will
 297          * be non-zero, but if accessing an upgraded 2.1 server it will
 298          * have the correct flags filled in.
 299          * XXX: fill in the LMV exp_connect_flags from MDC(s). */
 300         valid = exp_connect_flags(sbi->ll_md_exp) & CLIENT_CONNECT_MDT_REQD;
 301         if (exp_connect_flags(sbi->ll_md_exp) != 0 &&
 302             valid != CLIENT_CONNECT_MDT_REQD) {
 303                 char *buf;
 304
 305                 OBD_ALLOC_WAIT(buf, PAGE_CACHE_SIZE);
 306                 obd_connect_flags2str(buf, PAGE_CACHE_SIZE,
 307                                       valid ^ CLIENT_CONNECT_MDT_REQD, ",");
 308                 LCONSOLE_ERROR_MSG(0x170, "Server %s does not support "
 309                                    "feature(s) needed for correct operation "
 310                                    "of this client (%s). Please upgrade "
 311                                    "server or downgrade client.\n",
 312                                    sbi->ll_md_exp->exp_obd->obd_name, buf);
 313                 OBD_FREE(buf, PAGE_CACHE_SIZE);
 314                 GOTO(out_md_fid, err = -EPROTO);
 315         }
 316
 317         size = sizeof(*data);
 318         err = obd_get_info(NULL, sbi->ll_md_exp, sizeof(KEY_CONN_DATA),
 319                            KEY_CONN_DATA,  &size, data, NULL);
 320         if (err) {
 321                 CERROR("%s: Get connect data failed: rc = %d\n",
 322                        sbi->ll_md_exp->exp_obd->obd_name, err);
 323                 GOTO(out_md_fid, err);
 324         }
 325
 326         LASSERT(osfs->os_bsize);
 327         sb->s_blocksize = osfs->os_bsize;
 328         sb->s_blocksize_bits = log2(osfs->os_bsize);
 329         sb->s_magic = LL_SUPER_MAGIC;
 330         sb->s_maxbytes = MAX_LFS_FILESIZE;
 331         sbi->ll_namelen = osfs->os_namelen;
 332         sbi->ll_max_rw_chunk = LL_DEFAULT_MAX_RW_CHUNK;
 333
 334         if ((sbi->ll_flags & LL_SBI_USER_XATTR) &&
 335             !(data->ocd_connect_flags & OBD_CONNECT_XATTR)) {
 336                 LCONSOLE_INFO("Disabling user_xattr feature because "
 337                               "it is not supported on the server\n");
 338                 sbi->ll_flags &= ~LL_SBI_USER_XATTR;
 339         }
 340
 341         if (data->ocd_connect_flags & OBD_CONNECT_ACL) {
 342 #ifdef MS_POSIXACL
 343                 sb->s_flags |= MS_POSIXACL;
 344 #endif
 345                 sbi->ll_flags |= LL_SBI_ACL;
 346         } else {
 347                 LCONSOLE_INFO("client wants to enable acl, but mdt not!\n");
 348 #ifdef MS_POSIXACL
 349                 sb->s_flags &= ~MS_POSIXACL;
 350 #endif
 351                 sbi->ll_flags &= ~LL_SBI_ACL;
 352         }
 353
 354         if (data->ocd_connect_flags & OBD_CONNECT_RMT_CLIENT) {
 355                 if (!(sbi->ll_flags & LL_SBI_RMT_CLIENT)) {
 356                         sbi->ll_flags |= LL_SBI_RMT_CLIENT;
 357                         LCONSOLE_INFO("client is set as remote by default.\n");
 358                 }
 359         } else {
 360                 if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
 361                         sbi->ll_flags &= ~LL_SBI_RMT_CLIENT;
 362                         LCONSOLE_INFO("client claims to be remote, but server "
 363                                       "rejected, forced to be local.\n");
 364                 }
 365         }
 366
 367         if (data->ocd_connect_flags & OBD_CONNECT_MDS_CAPA) {
 368                 LCONSOLE_INFO("client enabled MDS capability!\n");
 369                 sbi->ll_flags |= LL_SBI_MDS_CAPA;
 370         }
 371
 372         if (data->ocd_connect_flags & OBD_CONNECT_OSS_CAPA) {
 373                 LCONSOLE_INFO("client enabled OSS capability!\n");
 374                 sbi->ll_flags |= LL_SBI_OSS_CAPA;
 375         }
 376
 377         if (data->ocd_connect_flags & OBD_CONNECT_64BITHASH)
 378                 sbi->ll_flags |= LL_SBI_64BIT_HASH;
 379
 380         if (data->ocd_connect_flags & OBD_CONNECT_BRW_SIZE)
 381                 sbi->ll_md_brw_pages = data->ocd_brw_size >> PAGE_CACHE_SHIFT;
 382         else
 383                 sbi->ll_md_brw_pages = 1;
 384
 385         if (data->ocd_connect_flags & OBD_CONNECT_LAYOUTLOCK)
 386                 sbi->ll_flags |= LL_SBI_LAYOUT_LOCK;
 387
 388         if (data->ocd_ibits_known & MDS_INODELOCK_XATTR) {
 389                 if (!(data->ocd_connect_flags & OBD_CONNECT_MAX_EASIZE)) {
 390                         LCONSOLE_INFO("%s: disabling xattr cache due to "
 391                                       "unknown maximum xattr size.\n", dt);
 392                 } else {
 393                         sbi->ll_flags |= LL_SBI_XATTR_CACHE;
 394                         sbi->ll_xattr_cache_enabled = 1;
 395                 }
 396         }
 397
 398         obd = class_name2obd(dt);
 399         if (!obd) {
 400                 CERROR("DT %s: not setup or attached\n", dt);
 401                 GOTO(out_md_fid, err = -ENODEV);
 402         }
 403
 404         data->ocd_connect_flags = OBD_CONNECT_GRANT     | OBD_CONNECT_VERSION  |
 405                                   OBD_CONNECT_REQPORTAL | OBD_CONNECT_BRW_SIZE |
 406                                   OBD_CONNECT_CANCELSET | OBD_CONNECT_FID      |
 407                                   OBD_CONNECT_SRVLOCK   | OBD_CONNECT_TRUNCLOCK|
 408                                   OBD_CONNECT_AT | OBD_CONNECT_RMT_CLIENT |
 409                                   OBD_CONNECT_OSS_CAPA | OBD_CONNECT_VBR|
 410                                   OBD_CONNECT_FULL20 | OBD_CONNECT_64BITHASH |
 411                                   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
 417         if (sbi->ll_flags & LL_SBI_SOM_PREVIEW)
 418                 data->ocd_connect_flags |= OBD_CONNECT_SOM;
 419
 420         if (!OBD_FAIL_CHECK(OBD_FAIL_OSC_CONNECT_CKSUM)) {
 421                 /* OBD_CONNECT_CKSUM should always be set, even if checksums are
 422                  * disabled by default, because it can still be enabled on the
 423                  * fly via /proc. As a consequence, we still need to come to an
 424                  * agreement on the supported algorithms at connect time */
 425                 data->ocd_connect_flags |= OBD_CONNECT_CKSUM;
 426
 427                 if (OBD_FAIL_CHECK(OBD_FAIL_OSC_CKSUM_ADLER_ONLY))
 428                         data->ocd_cksum_types = OBD_CKSUM_ADLER;
 429                 else
 430                         data->ocd_cksum_types = cksum_types_supported_client();
 431         }
 432
 433 #ifdef HAVE_LRU_RESIZE_SUPPORT
 434         data->ocd_connect_flags |= OBD_CONNECT_LRU_RESIZE;
 435 #endif
 436         if (sbi->ll_flags & LL_SBI_RMT_CLIENT)
 437                 data->ocd_connect_flags |= OBD_CONNECT_RMT_CLIENT_FORCE;
 438
 439         CDEBUG(D_RPCTRACE, "ocd_connect_flags: "LPX64" ocd_version: %d "
 440                "ocd_grant: %d\n", data->ocd_connect_flags,
 441                data->ocd_version, data->ocd_grant);
 442
 443         obd->obd_upcall.onu_owner = &sbi->ll_lco;
 444         obd->obd_upcall.onu_upcall = cl_ocd_update;
 445
 446         data->ocd_brw_size = DT_MAX_BRW_SIZE;
 447
 448         err = obd_connect(NULL, &sbi->ll_dt_exp, obd, &sbi->ll_sb_uuid, data,
 449                           NULL);
 450         if (err == -EBUSY) {
 451                 LCONSOLE_ERROR_MSG(0x150, "An OST (dt %s) is performing "
 452                                    "recovery, of which this client is not a "
 453                                    "part.  Please wait for recovery to "
 454                                    "complete, abort, or time out.\n", dt);
 455                 GOTO(out_md, err);
 456         } else if (err) {
 457                 CERROR("%s: Cannot connect to %s: rc = %d\n",
 458                        sbi->ll_dt_exp->exp_obd->obd_name, dt, err);
 459                 GOTO(out_md, err);
 460         }
 461
 462         sbi->ll_dt_exp->exp_connect_data = *data;
 463
 464         err = obd_fid_init(sbi->ll_dt_exp->exp_obd, sbi->ll_dt_exp,
 465                            LUSTRE_SEQ_METADATA);
 466         if (err) {
 467                 CERROR("%s: Can't init data layer FID infrastructure, "
 468                        "rc = %d\n", sbi->ll_dt_exp->exp_obd->obd_name, err);
 469                 GOTO(out_dt, err);
 470         }
 471
 472         mutex_lock(&sbi->ll_lco.lco_lock);
 473         sbi->ll_lco.lco_flags = data->ocd_connect_flags;
 474         sbi->ll_lco.lco_md_exp = sbi->ll_md_exp;
 475         sbi->ll_lco.lco_dt_exp = sbi->ll_dt_exp;
 476         mutex_unlock(&sbi->ll_lco.lco_lock);
 477
 478         fid_zero(&sbi->ll_root_fid);
 479         err = md_getstatus(sbi->ll_md_exp, &sbi->ll_root_fid, &oc);
 480         if (err) {
 481                 CERROR("cannot mds_connect: rc = %d\n", err);
 482                 GOTO(out_lock_cn_cb, err);
 483         }
 484         if (!fid_is_sane(&sbi->ll_root_fid)) {
 485                 CERROR("%s: Invalid root fid "DFID" during mount\n",
 486                        sbi->ll_md_exp->exp_obd->obd_name,
 487                        PFID(&sbi->ll_root_fid));
 488                 GOTO(out_lock_cn_cb, err = -EINVAL);
 489         }
 490         CDEBUG(D_SUPER, "rootfid "DFID"\n", PFID(&sbi->ll_root_fid));
 491
 492         sb->s_op = &lustre_super_operations;
 493 #if THREAD_SIZE >= 8192 /*b=17630*/
 494         sb->s_export_op = &lustre_export_operations;
 495 #endif
 496
 497         /* make root inode
 498          * XXX: move this to after cbd setup? */
 499         valid = OBD_MD_FLGETATTR | OBD_MD_FLBLOCKS | OBD_MD_FLMDSCAPA |
 500                 OBD_MD_FLMODEASIZE;
 501         if (sbi->ll_flags & LL_SBI_RMT_CLIENT)
 502                 valid |= OBD_MD_FLRMTPERM;
 503         else if (sbi->ll_flags & LL_SBI_ACL)
 504                 valid |= OBD_MD_FLACL;
 505
 506         OBD_ALLOC_PTR(op_data);
 507         if (op_data == NULL)
 508                 GOTO(out_lock_cn_cb, err = -ENOMEM);
 509
 510         op_data->op_fid1 = sbi->ll_root_fid;
 511         op_data->op_mode = 0;
 512         op_data->op_capa1 = oc;
 513         op_data->op_valid = valid;
 514
 515         err = md_getattr(sbi->ll_md_exp, op_data, &request);
 516         if (oc)
 517                 capa_put(oc);
 518         OBD_FREE_PTR(op_data);
 519         if (err) {
 520                 CERROR("%s: md_getattr failed for root: rc = %d\n",
 521                        sbi->ll_md_exp->exp_obd->obd_name, err);
 522                 GOTO(out_lock_cn_cb, err);
 523         }
 524
 525         err = md_get_lustre_md(sbi->ll_md_exp, request, sbi->ll_dt_exp,
 526                                sbi->ll_md_exp, &lmd);
 527         if (err) {
 528                 CERROR("failed to understand root inode md: rc = %d\n", err);
 529                 ptlrpc_req_finished(request);
 530                 GOTO(out_lock_cn_cb, err);
 531         }
 532
 533         LASSERT(fid_is_sane(&sbi->ll_root_fid));
 534         root = ll_iget(sb, cl_fid_build_ino(&sbi->ll_root_fid,
 535                                             sbi->ll_flags & LL_SBI_32BIT_API),
 536                        &lmd);
 537         md_free_lustre_md(sbi->ll_md_exp, &lmd);
 538         ptlrpc_req_finished(request);
 539
 540         if (IS_ERR(root)) {
 541                 if (lmd.lsm)
 542                         obd_free_memmd(sbi->ll_dt_exp, &lmd.lsm);
 543 #ifdef CONFIG_FS_POSIX_ACL
 544                 if (lmd.posix_acl) {
 545                         posix_acl_release(lmd.posix_acl);
 546                         lmd.posix_acl = NULL;
 547                 }
 548 #endif
 549                 err = IS_ERR(root) ? PTR_ERR(root) : -EBADF;
 550                 root = NULL;
 551                 CERROR("lustre_lite: bad iget4 for root\n");
 552                 GOTO(out_root, err);
 553         }
 554
 555         err = ll_close_thread_start(&sbi->ll_lcq);
 556         if (err) {
 557                 CERROR("cannot start close thread: rc %d\n", err);
 558                 GOTO(out_root, err);
 559         }
 560
 561 #ifdef CONFIG_FS_POSIX_ACL
 562         if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
 563                 rct_init(&sbi->ll_rct);
 564                 et_init(&sbi->ll_et);
 565         }
 566 #endif
 567
 568         checksum = sbi->ll_flags & LL_SBI_CHECKSUM;
 569         err = obd_set_info_async(NULL, sbi->ll_dt_exp, sizeof(KEY_CHECKSUM),
 570                                  KEY_CHECKSUM, sizeof(checksum), &checksum,
 571                                  NULL);
 572         cl_sb_init(sb);
 573
 574         err = obd_set_info_async(NULL, sbi->ll_dt_exp, sizeof(KEY_CACHE_SET),
 575                                  KEY_CACHE_SET, sizeof(sbi->ll_cache),
 576                                  &sbi->ll_cache, NULL);
 577
 578         sb->s_root = d_make_root(root);
 579         if (sb->s_root == NULL) {
 580                 CERROR("%s: can't make root dentry\n",
 581                         ll_get_fsname(sb, NULL, 0));
 582                 GOTO(out_root, err = -ENOMEM);
 583         }
 584 #ifdef HAVE_DCACHE_LOCK
 585         sb->s_root->d_op = &ll_d_ops;
 586 #endif
 587
 588         sbi->ll_sdev_orig = sb->s_dev;
 589
 590         /* We set sb->s_dev equal on all lustre clients in order to support
 591          * NFS export clustering.  NFSD requires that the FSID be the same
 592          * on all clients. */
 593         /* s_dev is also used in lt_compare() to compare two fs, but that is
 594          * only a node-local comparison. */
 595         uuid = obd_get_uuid(sbi->ll_md_exp);
 596         if (uuid != NULL)
 597                 sb->s_dev = get_uuid2int(uuid->uuid, strlen(uuid->uuid));
 598
 599         if (data != NULL)
 600                 OBD_FREE_PTR(data);
 601         if (osfs != NULL)
 602                 OBD_FREE_PTR(osfs);
 603         if (proc_lustre_fs_root != NULL) {
 604                 err = lprocfs_register_mountpoint(proc_lustre_fs_root, sb,
 605                                                   dt, md);
 606                 if (err < 0) {
 607                         CERROR("%s: could not register mount in lprocfs: "
 608                                "rc = %d\n", ll_get_fsname(sb, NULL, 0), err);
 609                         err = 0;
 610                 }
 611         }
 612
 613         RETURN(err);
 614 out_root:
 615         if (root)
 616                 iput(root);
 617 out_lock_cn_cb:
 618         obd_fid_fini(sbi->ll_dt_exp->exp_obd);
 619 out_dt:
 620         obd_disconnect(sbi->ll_dt_exp);
 621         sbi->ll_dt_exp = NULL;
 622         /* Make sure all OScs are gone, since cl_cache is accessing sbi. */
 623         obd_zombie_barrier();
 624 out_md_fid:
 625         obd_fid_fini(sbi->ll_md_exp->exp_obd);
 626 out_md:
 627         obd_disconnect(sbi->ll_md_exp);
 628         sbi->ll_md_exp = NULL;
 629 out:
 630         if (data != NULL)
 631                 OBD_FREE_PTR(data);
 632         if (osfs != NULL)
 633                 OBD_FREE_PTR(osfs);
 634         return err;
 635 }
 636
 637 int ll_get_max_mdsize(struct ll_sb_info *sbi, int *lmmsize)
 638 {
 639         int size, rc;
 640
 641         *lmmsize = obd_size_diskmd(sbi->ll_dt_exp, NULL);
 642         size = sizeof(int);
 643         rc = obd_get_info(NULL, sbi->ll_md_exp, sizeof(KEY_MAX_EASIZE),
 644                           KEY_MAX_EASIZE, &size, lmmsize, NULL);
 645         if (rc)
 646                 CERROR("Get max mdsize error rc %d\n", rc);
 647
 648         RETURN(rc);
 649 }
 650
 651 int ll_get_default_mdsize(struct ll_sb_info *sbi, int *lmmsize)
 652 {
 653         int size, rc;
 654
 655         size = sizeof(int);
 656         rc = obd_get_info(NULL, sbi->ll_md_exp, sizeof(KEY_DEFAULT_EASIZE),
 657                          KEY_DEFAULT_EASIZE, &size, lmmsize, NULL);
 658         if (rc)
 659                 CERROR("Get default mdsize error rc %d\n", rc);
 660
 661         RETURN(rc);
 662 }
 663
 664 int ll_get_max_cookiesize(struct ll_sb_info *sbi, int *lmmsize)
 665 {
 666         int size, rc;
 667
 668         size = sizeof(int);
 669         rc = obd_get_info(NULL, sbi->ll_md_exp, sizeof(KEY_MAX_COOKIESIZE),
 670                           KEY_MAX_COOKIESIZE, &size, lmmsize, NULL);
 671         if (rc)
 672                 CERROR("Get max cookiesize error rc %d\n", rc);
 673
 674         RETURN(rc);
 675 }
 676
 677 int ll_get_default_cookiesize(struct ll_sb_info *sbi, int *lmmsize)
 678 {
 679         int size, rc;
 680
 681         size = sizeof(int);
 682         rc = obd_get_info(NULL, sbi->ll_md_exp, sizeof(KEY_DEFAULT_COOKIESIZE),
 683                           KEY_DEFAULT_COOKIESIZE, &size, lmmsize, NULL);
 684         if (rc)
 685                 CERROR("Get default cookiesize error rc %d\n", rc);
 686
 687         RETURN(rc);
 688 }
 689
 690 static void ll_dump_inode(struct inode *inode)
 691 {
 692         struct ll_d_hlist_node *tmp;
 693         int dentry_count = 0;
 694
 695         LASSERT(inode != NULL);
 696
 697         ll_d_hlist_for_each(tmp, &inode->i_dentry)
 698                 dentry_count++;
 699
 700         CERROR("%s: inode %p dump: dev=%s fid="DFID
 701                " mode=%o count=%u, %d dentries\n",
 702                ll_get_fsname(inode->i_sb, NULL, 0), inode,
 703                ll_i2mdexp(inode)->exp_obd->obd_name, PFID(ll_inode2fid(inode)),
 704                inode->i_mode, atomic_read(&inode->i_count), dentry_count);
 705 }
 706
 707 void lustre_dump_dentry(struct dentry *dentry, int recur)
 708 {
 709         struct list_head *tmp;
 710         int subdirs = 0;
 711
 712         LASSERT(dentry != NULL);
 713
 714         list_for_each(tmp, &dentry->d_subdirs)
 715                 subdirs++;
 716
 717         CERROR("dentry %p dump: name=%.*s parent=%.*s (%p), inode=%p, count=%u,"
 718                " flags=0x%x, fsdata=%p, %d subdirs\n", dentry,
 719                dentry->d_name.len, dentry->d_name.name,
 720                dentry->d_parent->d_name.len, dentry->d_parent->d_name.name,
 721                dentry->d_parent, dentry->d_inode, d_count(dentry),
 722                dentry->d_flags, dentry->d_fsdata, subdirs);
 723         if (dentry->d_inode != NULL)
 724                 ll_dump_inode(dentry->d_inode);
 725
 726         if (recur == 0)
 727                 return;
 728
 729         list_for_each(tmp, &dentry->d_subdirs) {
 730                 struct dentry *d = list_entry(tmp, struct dentry, d_u.d_child);
 731                 lustre_dump_dentry(d, recur - 1);
 732         }
 733 }
 734
 735 static void client_common_put_super(struct super_block *sb)
 736 {
 737         struct ll_sb_info *sbi = ll_s2sbi(sb);
 738         ENTRY;
 739
 740 #ifdef CONFIG_FS_POSIX_ACL
 741         if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
 742                 et_fini(&sbi->ll_et);
 743                 rct_fini(&sbi->ll_rct);
 744         }
 745 #endif
 746
 747         ll_close_thread_shutdown(sbi->ll_lcq);
 748
 749         cl_sb_fini(sb);
 750
 751         list_del(&sbi->ll_conn_chain);
 752
 753         obd_fid_fini(sbi->ll_dt_exp->exp_obd);
 754         obd_disconnect(sbi->ll_dt_exp);
 755         sbi->ll_dt_exp = NULL;
 756         /* wait till all OSCs are gone, since cl_cache is accessing sbi.
 757          * see LU-2543. */
 758         obd_zombie_barrier();
 759
 760         lprocfs_unregister_mountpoint(sbi);
 761
 762         obd_fid_fini(sbi->ll_md_exp->exp_obd);
 763         obd_disconnect(sbi->ll_md_exp);
 764         sbi->ll_md_exp = NULL;
 765
 766         EXIT;
 767 }
 768
 769 void ll_kill_super(struct super_block *sb)
 770 {
 771         struct ll_sb_info *sbi;
 772
 773         ENTRY;
 774
 775         /* not init sb ?*/
 776         if (!(sb->s_flags & MS_ACTIVE))
 777                 return;
 778
 779         sbi = ll_s2sbi(sb);
 780         /* we need restore s_dev from changed for clustred NFS before put_super
 781          * because new kernels have cached s_dev and change sb->s_dev in
 782          * put_super not affected real removing devices */
 783         if (sbi) {
 784                 sb->s_dev = sbi->ll_sdev_orig;
 785                 sbi->ll_umounting = 1;
 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, int *flags)
 800 {
 801         int tmp;
 802         char *s1 = options, *s2;
 803         ENTRY;
 804
 805         if (!options)
 806                 RETURN(0);
 807
 808         CDEBUG(D_CONFIG, "Parsing opts %s\n", options);
 809
 810         while (*s1) {
 811                 CDEBUG(D_SUPER, "next opt=%s\n", s1);
 812                 tmp = ll_set_opt("nolock", s1, LL_SBI_NOLCK);
 813                 if (tmp) {
 814                         *flags |= tmp;
 815                         goto next;
 816                 }
 817                 tmp = ll_set_opt("flock", s1, LL_SBI_FLOCK);
 818                 if (tmp) {
 819                         *flags |= tmp;
 820                         goto next;
 821                 }
 822                 tmp = ll_set_opt("localflock", s1, LL_SBI_LOCALFLOCK);
 823                 if (tmp) {
 824                         *flags |= tmp;
 825                         goto next;
 826                 }
 827                 tmp = ll_set_opt("noflock", s1, LL_SBI_FLOCK|LL_SBI_LOCALFLOCK);
 828                 if (tmp) {
 829                         *flags &= ~tmp;
 830                         goto next;
 831                 }
 832                 tmp = ll_set_opt("user_xattr", s1, LL_SBI_USER_XATTR);
 833                 if (tmp) {
 834                         *flags |= tmp;
 835                         goto next;
 836                 }
 837                 tmp = ll_set_opt("nouser_xattr", s1, LL_SBI_USER_XATTR);
 838                 if (tmp) {
 839                         *flags &= ~tmp;
 840                         goto next;
 841                 }
 842 #if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(2, 6, 51, 0)
 843                 tmp = ll_set_opt("acl", s1, LL_SBI_ACL);
 844                 if (tmp) {
 845                         /* Ignore deprecated mount option.  The client will
 846                          * always try to mount with ACL support, whether this
 847                          * is used depends on whether server supports it. */
 848                         LCONSOLE_ERROR_MSG(0x152, "Ignoring deprecated "
 849                                                   "mount option 'acl'.\n");
 850                         goto next;
 851                 }
 852                 tmp = ll_set_opt("noacl", s1, LL_SBI_ACL);
 853                 if (tmp) {
 854                         LCONSOLE_ERROR_MSG(0x152, "Ignoring deprecated "
 855                                                   "mount option 'noacl'.\n");
 856                         goto next;
 857                 }
 858 #else
 859 #warning "{no}acl options have been deprecated since 1.8, please remove them"
 860 #endif
 861                 tmp = ll_set_opt("remote_client", s1, LL_SBI_RMT_CLIENT);
 862                 if (tmp) {
 863                         *flags |= tmp;
 864                         goto next;
 865                 }
 866                 tmp = ll_set_opt("user_fid2path", s1, LL_SBI_USER_FID2PATH);
 867                 if (tmp) {
 868                         *flags |= tmp;
 869                         goto next;
 870                 }
 871                 tmp = ll_set_opt("nouser_fid2path", s1, LL_SBI_USER_FID2PATH);
 872                 if (tmp) {
 873                         *flags &= ~tmp;
 874                         goto next;
 875                 }
 876
 877                 tmp = ll_set_opt("checksum", s1, LL_SBI_CHECKSUM);
 878                 if (tmp) {
 879                         *flags |= tmp;
 880                         goto next;
 881                 }
 882                 tmp = ll_set_opt("nochecksum", s1, LL_SBI_CHECKSUM);
 883                 if (tmp) {
 884                         *flags &= ~tmp;
 885                         goto next;
 886                 }
 887                 tmp = ll_set_opt("lruresize", s1, LL_SBI_LRU_RESIZE);
 888                 if (tmp) {
 889                         *flags |= tmp;
 890                         goto next;
 891                 }
 892                 tmp = ll_set_opt("nolruresize", s1, LL_SBI_LRU_RESIZE);
 893                 if (tmp) {
 894                         *flags &= ~tmp;
 895                         goto next;
 896                 }
 897                 tmp = ll_set_opt("lazystatfs", s1, LL_SBI_LAZYSTATFS);
 898                 if (tmp) {
 899                         *flags |= tmp;
 900                         goto next;
 901                 }
 902                 tmp = ll_set_opt("nolazystatfs", s1, LL_SBI_LAZYSTATFS);
 903                 if (tmp) {
 904                         *flags &= ~tmp;
 905                         goto next;
 906                 }
 907                 tmp = ll_set_opt("som_preview", s1, LL_SBI_SOM_PREVIEW);
 908                 if (tmp) {
 909                         *flags |= tmp;
 910                         goto next;
 911                 }
 912                 tmp = ll_set_opt("32bitapi", s1, LL_SBI_32BIT_API);
 913                 if (tmp) {
 914                         *flags |= tmp;
 915                         goto next;
 916                 }
 917                 tmp = ll_set_opt("verbose", s1, LL_SBI_VERBOSE);
 918                 if (tmp) {
 919                         *flags |= tmp;
 920                         goto next;
 921                 }
 922                 tmp = ll_set_opt("noverbose", s1, LL_SBI_VERBOSE);
 923                 if (tmp) {
 924                         *flags &= ~tmp;
 925                         goto next;
 926                 }
 927                 LCONSOLE_ERROR_MSG(0x152, "Unknown option '%s', won't mount.\n",
 928                                    s1);
 929                 RETURN(-EINVAL);
 930
 931 next:
 932                 /* Find next opt */
 933                 s2 = strchr(s1, ',');
 934                 if (s2 == NULL)
 935                         break;
 936                 s1 = s2 + 1;
 937         }
 938         RETURN(0);
 939 }
 940
 941 void ll_lli_init(struct ll_inode_info *lli)
 942 {
 943         lli->lli_inode_magic = LLI_INODE_MAGIC;
 944         lli->lli_flags = 0;
 945         lli->lli_ioepoch = 0;
 946         lli->lli_maxbytes = MAX_LFS_FILESIZE;
 947         spin_lock_init(&lli->lli_lock);
 948         lli->lli_posix_acl = NULL;
 949         lli->lli_remote_perms = NULL;
 950         mutex_init(&lli->lli_rmtperm_mutex);
 951         /* Do not set lli_fid, it has been initialized already. */
 952         fid_zero(&lli->lli_pfid);
 953         INIT_LIST_HEAD(&lli->lli_close_list);
 954         INIT_LIST_HEAD(&lli->lli_oss_capas);
 955         atomic_set(&lli->lli_open_count, 0);
 956         lli->lli_mds_capa = NULL;
 957         lli->lli_rmtperm_time = 0;
 958         lli->lli_pending_och = NULL;
 959         lli->lli_mds_read_och = NULL;
 960         lli->lli_mds_write_och = NULL;
 961         lli->lli_mds_exec_och = NULL;
 962         lli->lli_open_fd_read_count = 0;
 963         lli->lli_open_fd_write_count = 0;
 964         lli->lli_open_fd_exec_count = 0;
 965         mutex_init(&lli->lli_och_mutex);
 966         spin_lock_init(&lli->lli_agl_lock);
 967         lli->lli_has_smd = false;
 968         spin_lock_init(&lli->lli_layout_lock);
 969         ll_layout_version_set(lli, LL_LAYOUT_GEN_NONE);
 970         lli->lli_clob = NULL;
 971
 972         init_rwsem(&lli->lli_xattrs_list_rwsem);
 973         mutex_init(&lli->lli_xattrs_enq_lock);
 974
 975         LASSERT(lli->lli_vfs_inode.i_mode != 0);
 976         if (S_ISDIR(lli->lli_vfs_inode.i_mode)) {
 977                 mutex_init(&lli->lli_readdir_mutex);
 978                 lli->lli_opendir_key = NULL;
 979                 lli->lli_sai = NULL;
 980                 spin_lock_init(&lli->lli_sa_lock);
 981                 lli->lli_opendir_pid = 0;
 982         } else {
 983                 mutex_init(&lli->lli_size_mutex);
 984                 lli->lli_symlink_name = NULL;
 985                 init_rwsem(&lli->lli_trunc_sem);
 986                 mutex_init(&lli->lli_write_mutex);
 987                 init_rwsem(&lli->lli_glimpse_sem);
 988                 lli->lli_glimpse_time = 0;
 989                 INIT_LIST_HEAD(&lli->lli_agl_list);
 990                 lli->lli_agl_index = 0;
 991                 lli->lli_async_rc = 0;
 992         }
 993         mutex_init(&lli->lli_layout_mutex);
 994 }
 995
 996 static inline int ll_bdi_register(struct backing_dev_info *bdi)
 997 {
 998         static atomic_t ll_bdi_num = ATOMIC_INIT(0);
 999
1000         bdi->name = "lustre";
1001         return bdi_register(bdi, NULL, "lustre-%d",
1002                             atomic_inc_return(&ll_bdi_num));
1003 }
1004
1005 int ll_fill_super(struct super_block *sb, struct vfsmount *mnt)
1006 {
1007         struct lustre_profile *lprof = NULL;
1008         struct lustre_sb_info *lsi = s2lsi(sb);
1009         struct ll_sb_info *sbi;
1010         char  *dt = NULL, *md = NULL;
1011         char  *profilenm = get_profile_name(sb);
1012         struct config_llog_instance *cfg;
1013         /* %p for void* in printf needs 16+2 characters: 0xffffffffffffffff */
1014         const int instlen = sizeof(cfg->cfg_instance) * 2 + 2;
1015         int    err;
1016         ENTRY;
1017
1018         CDEBUG(D_VFSTRACE, "VFS Op: sb %p\n", sb);
1019
1020         OBD_ALLOC_PTR(cfg);
1021         if (cfg == NULL)
1022                 RETURN(-ENOMEM);
1023
1024         try_module_get(THIS_MODULE);
1025
1026         /* client additional sb info */
1027         lsi->lsi_llsbi = sbi = ll_init_sbi();
1028         if (!sbi) {
1029                 module_put(THIS_MODULE);
1030                 OBD_FREE_PTR(cfg);
1031                 RETURN(-ENOMEM);
1032         }
1033
1034         err = ll_options(lsi->lsi_lmd->lmd_opts, &sbi->ll_flags);
1035         if (err)
1036                 GOTO(out_free, err);
1037
1038         err = bdi_init(&lsi->lsi_bdi);
1039         if (err)
1040                 GOTO(out_free, err);
1041         lsi->lsi_flags |= LSI_BDI_INITIALIZED;
1042         lsi->lsi_bdi.capabilities = BDI_CAP_MAP_COPY;
1043         err = ll_bdi_register(&lsi->lsi_bdi);
1044         if (err)
1045                 GOTO(out_free, err);
1046
1047         sb->s_bdi = &lsi->lsi_bdi;
1048 #ifndef HAVE_DCACHE_LOCK
1049         /* kernel >= 2.6.38 store dentry operations in sb->s_d_op. */
1050         sb->s_d_op = &ll_d_ops;
1051 #endif
1052
1053         /* Generate a string unique to this super, in case some joker tries
1054            to mount the same fs at two mount points.
1055            Use the address of the super itself.*/
1056         cfg->cfg_instance = sb;
1057         cfg->cfg_uuid = lsi->lsi_llsbi->ll_sb_uuid;
1058         cfg->cfg_callback = class_config_llog_handler;
1059         /* set up client obds */
1060         err = lustre_process_log(sb, profilenm, cfg);
1061         if (err < 0)
1062                 GOTO(out_free, err);
1063
1064         /* Profile set with LCFG_MOUNTOPT so we can find our mdc and osc obds */
1065         lprof = class_get_profile(profilenm);
1066         if (lprof == NULL) {
1067                 LCONSOLE_ERROR_MSG(0x156, "The client profile '%s' could not be"
1068                                    " read from the MGS.  Does that filesystem "
1069                                    "exist?\n", profilenm);
1070                 GOTO(out_free, err = -EINVAL);
1071         }
1072         CDEBUG(D_CONFIG, "Found profile %s: mdc=%s osc=%s\n", profilenm,
1073                lprof->lp_md, lprof->lp_dt);
1074
1075         OBD_ALLOC(dt, strlen(lprof->lp_dt) + instlen + 2);
1076         if (!dt)
1077                 GOTO(out_free, err = -ENOMEM);
1078         sprintf(dt, "%s-%p", lprof->lp_dt, cfg->cfg_instance);
1079
1080         OBD_ALLOC(md, strlen(lprof->lp_md) + instlen + 2);
1081         if (!md)
1082                 GOTO(out_free, err = -ENOMEM);
1083         sprintf(md, "%s-%p", lprof->lp_md, cfg->cfg_instance);
1084
1085         /* connections, registrations, sb setup */
1086         err = client_common_fill_super(sb, md, dt, mnt);
1087
1088 out_free:
1089         if (md)
1090                 OBD_FREE(md, strlen(lprof->lp_md) + instlen + 2);
1091         if (dt)
1092                 OBD_FREE(dt, strlen(lprof->lp_dt) + instlen + 2);
1093         if (err)
1094                 ll_put_super(sb);
1095         else if (sbi->ll_flags & LL_SBI_VERBOSE)
1096                 LCONSOLE_WARN("Mounted %s\n", profilenm);
1097
1098         OBD_FREE_PTR(cfg);
1099         RETURN(err);
1100 } /* ll_fill_super */
1101
1102 void ll_put_super(struct super_block *sb)
1103 {
1104         struct config_llog_instance cfg, params_cfg;
1105         struct obd_device *obd;
1106         struct lustre_sb_info *lsi = s2lsi(sb);
1107         struct ll_sb_info *sbi = ll_s2sbi(sb);
1108         char *profilenm = get_profile_name(sb);
1109         int ccc_count, next, force = 1, rc = 0;
1110         ENTRY;
1111
1112         CDEBUG(D_VFSTRACE, "VFS Op: sb %p - %s\n", sb, profilenm);
1113
1114         ll_print_capa_stat(sbi);
1115
1116         cfg.cfg_instance = sb;
1117         lustre_end_log(sb, profilenm, &cfg);
1118
1119         params_cfg.cfg_instance = sb;
1120         lustre_end_log(sb, PARAMS_FILENAME, &params_cfg);
1121
1122         if (sbi->ll_md_exp) {
1123                 obd = class_exp2obd(sbi->ll_md_exp);
1124                 if (obd)
1125                         force = obd->obd_force;
1126         }
1127
1128         /* Wait for unstable pages to be committed to stable storage */
1129         if (force == 0) {
1130                 struct l_wait_info lwi = LWI_INTR(LWI_ON_SIGNAL_NOOP, NULL);
1131                 rc = l_wait_event(sbi->ll_cache.ccc_unstable_waitq,
1132                         atomic_read(&sbi->ll_cache.ccc_unstable_nr) == 0,
1133                         &lwi);
1134         }
1135
1136         ccc_count = atomic_read(&sbi->ll_cache.ccc_unstable_nr);
1137         if (force == 0 && rc != -EINTR)
1138                 LASSERTF(ccc_count == 0, "count: %i\n", ccc_count);
1139
1140
1141         /* We need to set force before the lov_disconnect in
1142            lustre_common_put_super, since l_d cleans up osc's as well. */
1143         if (force) {
1144                 next = 0;
1145                 while ((obd = class_devices_in_group(&sbi->ll_sb_uuid,
1146                                                      &next)) != NULL) {
1147                         obd->obd_force = force;
1148                 }
1149         }
1150
1151         if (sbi->ll_lcq) {
1152                 /* Only if client_common_fill_super succeeded */
1153                 client_common_put_super(sb);
1154         }
1155
1156         next = 0;
1157         while ((obd = class_devices_in_group(&sbi->ll_sb_uuid, &next)) !=NULL) {
1158                 class_manual_cleanup(obd);
1159         }
1160
1161         if (sbi->ll_flags & LL_SBI_VERBOSE)
1162                 LCONSOLE_WARN("Unmounted %s\n", profilenm ? profilenm : "");
1163
1164         if (profilenm)
1165                 class_del_profile(profilenm);
1166
1167         if (lsi->lsi_flags & LSI_BDI_INITIALIZED) {
1168                 bdi_destroy(&lsi->lsi_bdi);
1169                 lsi->lsi_flags &= ~LSI_BDI_INITIALIZED;
1170         }
1171
1172         ll_free_sbi(sb);
1173         lsi->lsi_llsbi = NULL;
1174
1175         lustre_common_put_super(sb);
1176
1177         cl_env_cache_purge(~0);
1178
1179         module_put(THIS_MODULE);
1180
1181         EXIT;
1182 } /* client_put_super */
1183
1184 struct inode *ll_inode_from_resource_lock(struct ldlm_lock *lock)
1185 {
1186         struct inode *inode = NULL;
1187
1188         /* NOTE: we depend on atomic igrab() -bzzz */
1189         lock_res_and_lock(lock);
1190         if (lock->l_resource->lr_lvb_inode) {
1191                 struct ll_inode_info * lli;
1192                 lli = ll_i2info(lock->l_resource->lr_lvb_inode);
1193                 if (lli->lli_inode_magic == LLI_INODE_MAGIC) {
1194                         inode = igrab(lock->l_resource->lr_lvb_inode);
1195                 } else {
1196                         inode = lock->l_resource->lr_lvb_inode;
1197                         LDLM_DEBUG_LIMIT(inode->i_state & I_FREEING ?  D_INFO :
1198                                          D_WARNING, lock, "lr_lvb_inode %p is "
1199                                          "bogus: magic %08x",
1200                                          lock->l_resource->lr_lvb_inode,
1201                                          lli->lli_inode_magic);
1202                         inode = NULL;
1203                 }
1204         }
1205         unlock_res_and_lock(lock);
1206         return inode;
1207 }
1208
1209 static void ll_dir_clear_lsm_md(struct inode *inode)
1210 {
1211         struct ll_inode_info *lli = ll_i2info(inode);
1212
1213         LASSERT(S_ISDIR(inode->i_mode));
1214
1215         if (lli->lli_lsm_md != NULL) {
1216                 lmv_free_memmd(lli->lli_lsm_md);
1217                 lli->lli_lsm_md = NULL;
1218         }
1219 }
1220
1221 static struct inode *ll_iget_anon_dir(struct super_block *sb,
1222                                       const struct lu_fid *fid,
1223                                       struct lustre_md *md)
1224 {
1225         struct ll_sb_info       *sbi = ll_s2sbi(sb);
1226         struct mdt_body         *body = md->body;
1227         struct inode            *inode;
1228         ino_t                   ino;
1229         ENTRY;
1230
1231         ino = cl_fid_build_ino(fid, sbi->ll_flags & LL_SBI_32BIT_API);
1232         inode = iget_locked(sb, ino);
1233         if (inode == NULL) {
1234                 CERROR("%s: failed get simple inode "DFID": rc = -ENOENT\n",
1235                        ll_get_fsname(sb, NULL, 0), PFID(fid));
1236                 RETURN(ERR_PTR(-ENOENT));
1237         }
1238
1239         if (inode->i_state & I_NEW) {
1240                 struct ll_inode_info *lli = ll_i2info(inode);
1241                 struct lmv_stripe_md *lsm = md->lmv;
1242
1243                 inode->i_mode = (inode->i_mode & ~S_IFMT) |
1244                                 (body->mode & S_IFMT);
1245                 LASSERTF(S_ISDIR(inode->i_mode), "Not slave inode "DFID"\n",
1246                          PFID(fid));
1247
1248                 LTIME_S(inode->i_mtime) = 0;
1249                 LTIME_S(inode->i_atime) = 0;
1250                 LTIME_S(inode->i_ctime) = 0;
1251                 inode->i_rdev = 0;
1252
1253                 /* initializing backing dev info. */
1254                 inode->i_mapping->backing_dev_info =
1255                                                 &s2lsi(inode->i_sb)->lsi_bdi;
1256                 inode->i_op = &ll_dir_inode_operations;
1257                 inode->i_fop = &ll_dir_operations;
1258                 lli->lli_fid = *fid;
1259                 ll_lli_init(lli);
1260
1261                 LASSERT(lsm != NULL);
1262                 /* master object FID */
1263                 lli->lli_pfid = body->fid1;
1264                 CDEBUG(D_INODE, "lli %p slave "DFID" master "DFID"\n",
1265                        lli, PFID(fid), PFID(&lli->lli_pfid));
1266                 unlock_new_inode(inode);
1267         }
1268
1269         RETURN(inode);
1270 }
1271
1272 static int ll_init_lsm_md(struct inode *inode, struct lustre_md *md)
1273 {
1274         struct lu_fid *fid;
1275         struct lmv_stripe_md *lsm = md->lmv;
1276         int i;
1277
1278         LASSERT(lsm != NULL);
1279         /* XXX sigh, this lsm_root initialization should be in
1280          * LMV layer, but it needs ll_iget right now, so we
1281          * put this here right now. */
1282         for (i = 0; i < lsm->lsm_md_stripe_count; i++) {
1283                 fid = &lsm->lsm_md_oinfo[i].lmo_fid;
1284                 LASSERT(lsm->lsm_md_oinfo[i].lmo_root == NULL);
1285                 /* Unfortunately ll_iget will call ll_update_inode,
1286                  * where the initialization of slave inode is slightly
1287                  * different, so it reset lsm_md to NULL to avoid
1288                  * initializing lsm for slave inode. */
1289                 /* For migrating inode, master stripe and master object will
1290                  * be same, so we only need assign this inode */
1291                 if (lsm->lsm_md_hash_type & LMV_HASH_FLAG_MIGRATION && i == 0)
1292                         lsm->lsm_md_oinfo[i].lmo_root = inode;
1293                 else
1294                         lsm->lsm_md_oinfo[i].lmo_root =
1295                                 ll_iget_anon_dir(inode->i_sb, fid, md);
1296
1297                 if (IS_ERR(lsm->lsm_md_oinfo[i].lmo_root)) {
1298                         int rc = PTR_ERR(lsm->lsm_md_oinfo[i].lmo_root);
1299
1300                         lsm->lsm_md_oinfo[i].lmo_root = NULL;
1301                         return rc;
1302                 }
1303         }
1304
1305         /* Here is where the lsm is being initialized(fill lmo_info) after
1306          * client retrieve MD stripe information from MDT. */
1307         return md_update_lsm_md(ll_i2mdexp(inode), lsm, md->body,
1308                                 ll_md_blocking_ast);
1309 }
1310
1311 static inline int lli_lsm_md_eq(const struct lmv_stripe_md *lsm_md1,
1312                                 const struct lmv_stripe_md *lsm_md2)
1313 {
1314         return lsm_md1->lsm_md_magic == lsm_md2->lsm_md_magic &&
1315                lsm_md1->lsm_md_stripe_count == lsm_md2->lsm_md_stripe_count &&
1316                lsm_md1->lsm_md_master_mdt_index ==
1317                                         lsm_md2->lsm_md_master_mdt_index &&
1318                lsm_md1->lsm_md_hash_type == lsm_md2->lsm_md_hash_type &&
1319                lsm_md1->lsm_md_layout_version ==
1320                                         lsm_md2->lsm_md_layout_version &&
1321                strcmp(lsm_md1->lsm_md_pool_name,
1322                       lsm_md2->lsm_md_pool_name) == 0;
1323 }
1324
1325 static int ll_update_lsm_md(struct inode *inode, struct lustre_md *md)
1326 {
1327         struct ll_inode_info *lli = ll_i2info(inode);
1328         struct lmv_stripe_md *lsm = md->lmv;
1329         int     rc;
1330         ENTRY;
1331
1332         LASSERT(S_ISDIR(inode->i_mode));
1333         CDEBUG(D_INODE, "update lsm %p of "DFID"\n", lli->lli_lsm_md,
1334                PFID(ll_inode2fid(inode)));
1335
1336         /* no striped information from request. */
1337         if (lsm == NULL) {
1338                 if (lli->lli_lsm_md == NULL) {
1339                         RETURN(0);
1340                 } else if (lli->lli_lsm_md->lsm_md_hash_type &
1341                                                 LMV_HASH_FLAG_MIGRATION) {
1342                         /* migration is done, the temporay MIGRATE layout has
1343                          * been removed */
1344                         CDEBUG(D_INODE, DFID" finish migration.\n",
1345                                PFID(ll_inode2fid(inode)));
1346                         lmv_free_memmd(lli->lli_lsm_md);
1347                         lli->lli_lsm_md = NULL;
1348                         RETURN(0);
1349                 } else {
1350                         /* The lustre_md from req does not include stripeEA,
1351                          * see ll_md_setattr */
1352                         RETURN(0);
1353                 }
1354         }
1355
1356         /* set the directory layout */
1357         if (lli->lli_lsm_md == NULL) {
1358
1359                 rc = ll_init_lsm_md(inode, md);
1360                 if (rc != 0)
1361                         RETURN(rc);
1362
1363                 lli->lli_lsm_md = lsm;
1364                 /* set lsm_md to NULL, so the following free lustre_md
1365                  * will not free this lsm */
1366                 md->lmv = NULL;
1367                 CDEBUG(D_INODE, "Set lsm %p magic %x to "DFID"\n", lsm,
1368                        lsm->lsm_md_magic, PFID(ll_inode2fid(inode)));
1369                 RETURN(0);
1370         }
1371
1372         /* Compare the old and new stripe information */
1373         if (!lsm_md_eq(lli->lli_lsm_md, lsm)) {
1374                 struct lmv_stripe_md    *old_lsm = lli->lli_lsm_md;
1375                 int                     idx;
1376
1377                 CERROR("%s: lmv layout mismatch "DFID"(%p)/"DFID"(%p)"
1378                        "magic:0x%x/0x%x stripe count: %d/%d master_mdt: %d/%d"
1379                        "hash_type:0x%x/0x%x layout: 0x%x/0x%x pool:%s/%s\n",
1380                        ll_get_fsname(inode->i_sb, NULL, 0),
1381                        PFID(&lsm->lsm_md_master_fid), lsm,
1382                        PFID(&old_lsm->lsm_md_master_fid), old_lsm,
1383                        lsm->lsm_md_magic, old_lsm->lsm_md_magic,
1384                        lsm->lsm_md_stripe_count,
1385                        old_lsm->lsm_md_stripe_count,
1386                        lsm->lsm_md_master_mdt_index,
1387                        old_lsm->lsm_md_master_mdt_index,
1388                        lsm->lsm_md_hash_type, old_lsm->lsm_md_hash_type,
1389                        lsm->lsm_md_layout_version,
1390                        old_lsm->lsm_md_layout_version,
1391                        lsm->lsm_md_pool_name,
1392                        old_lsm->lsm_md_pool_name);
1393
1394                 for (idx = 0; idx < old_lsm->lsm_md_stripe_count; idx++) {
1395                         CERROR("%s: sub FIDs in old lsm idx %d, old: "DFID"\n",
1396                                ll_get_fsname(inode->i_sb, NULL, 0), idx,
1397                                PFID(&old_lsm->lsm_md_oinfo[idx].lmo_fid));
1398                 }
1399
1400                 for (idx = 0; idx < lsm->lsm_md_stripe_count; idx++) {
1401                         CERROR("%s: sub FIDs in new lsm idx %d, new: "DFID"\n",
1402                                ll_get_fsname(inode->i_sb, NULL, 0), idx,
1403                                PFID(&lsm->lsm_md_oinfo[idx].lmo_fid));
1404                 }
1405
1406                 RETURN(-EIO);
1407         }
1408
1409         rc = md_update_lsm_md(ll_i2mdexp(inode), ll_i2info(inode)->lli_lsm_md,
1410                               md->body, ll_md_blocking_ast);
1411
1412         RETURN(rc);
1413 }
1414
1415 void ll_clear_inode(struct inode *inode)
1416 {
1417         struct ll_inode_info *lli = ll_i2info(inode);
1418         struct ll_sb_info *sbi = ll_i2sbi(inode);
1419         ENTRY;
1420
1421         CDEBUG(D_VFSTRACE, "VFS Op:inode="DFID"(%p)\n",
1422                PFID(ll_inode2fid(inode)), inode);
1423
1424         if (S_ISDIR(inode->i_mode)) {
1425                 /* these should have been cleared in ll_file_release */
1426                 LASSERT(lli->lli_opendir_key == NULL);
1427                 LASSERT(lli->lli_sai == NULL);
1428                 LASSERT(lli->lli_opendir_pid == 0);
1429         }
1430
1431         spin_lock(&lli->lli_lock);
1432         ll_i2info(inode)->lli_flags &= ~LLIF_MDS_SIZE_LOCK;
1433         spin_unlock(&lli->lli_lock);
1434         md_null_inode(sbi->ll_md_exp, ll_inode2fid(inode));
1435
1436         LASSERT(!lli->lli_open_fd_write_count);
1437         LASSERT(!lli->lli_open_fd_read_count);
1438         LASSERT(!lli->lli_open_fd_exec_count);
1439
1440         if (lli->lli_mds_write_och)
1441                 ll_md_real_close(inode, FMODE_WRITE);
1442         if (lli->lli_mds_exec_och)
1443                 ll_md_real_close(inode, FMODE_EXEC);
1444         if (lli->lli_mds_read_och)
1445                 ll_md_real_close(inode, FMODE_READ);
1446
1447         if (S_ISLNK(inode->i_mode) && lli->lli_symlink_name) {
1448                 OBD_FREE(lli->lli_symlink_name,
1449                          strlen(lli->lli_symlink_name) + 1);
1450                 lli->lli_symlink_name = NULL;
1451         }
1452
1453         ll_xattr_cache_destroy(inode);
1454
1455         if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
1456                 LASSERT(lli->lli_posix_acl == NULL);
1457                 if (lli->lli_remote_perms) {
1458                         free_rmtperm_hash(lli->lli_remote_perms);
1459                         lli->lli_remote_perms = NULL;
1460                 }
1461         }
1462 #ifdef CONFIG_FS_POSIX_ACL
1463         else if (lli->lli_posix_acl) {
1464                 LASSERT(atomic_read(&lli->lli_posix_acl->a_refcount) == 1);
1465                 LASSERT(lli->lli_remote_perms == NULL);
1466                 posix_acl_release(lli->lli_posix_acl);
1467                 lli->lli_posix_acl = NULL;
1468         }
1469 #endif
1470         lli->lli_inode_magic = LLI_INODE_DEAD;
1471
1472         ll_clear_inode_capas(inode);
1473         if (S_ISDIR(inode->i_mode))
1474                 ll_dir_clear_lsm_md(inode);
1475         else if (S_ISREG(inode->i_mode) && !is_bad_inode(inode))
1476                 LASSERT(list_empty(&lli->lli_agl_list));
1477
1478         /*
1479          * XXX This has to be done before lsm is freed below, because
1480          * cl_object still uses inode lsm.
1481          */
1482         cl_inode_fini(inode);
1483         lli->lli_has_smd = false;
1484
1485         EXIT;
1486 }
1487
1488 int ll_md_setattr(struct dentry *dentry, struct md_op_data *op_data,
1489                   struct md_open_data **mod)
1490 {
1491         struct lustre_md md;
1492         struct inode *inode = dentry->d_inode;
1493         struct ll_sb_info *sbi = ll_i2sbi(inode);
1494         struct ptlrpc_request *request = NULL;
1495         int rc, ia_valid;
1496         ENTRY;
1497
1498         op_data = ll_prep_md_op_data(op_data, inode, NULL, NULL, 0, 0,
1499                                      LUSTRE_OPC_ANY, NULL);
1500         if (IS_ERR(op_data))
1501                 RETURN(PTR_ERR(op_data));
1502
1503         rc = md_setattr(sbi->ll_md_exp, op_data, NULL, 0, NULL, 0,
1504                         &request, mod);
1505         if (rc) {
1506                 ptlrpc_req_finished(request);
1507                 if (rc == -ENOENT) {
1508                         clear_nlink(inode);
1509                         /* Unlinked special device node? Or just a race?
1510                          * Pretend we done everything. */
1511                         if (!S_ISREG(inode->i_mode) &&
1512                             !S_ISDIR(inode->i_mode)) {
1513                                 ia_valid = op_data->op_attr.ia_valid;
1514                                 op_data->op_attr.ia_valid &= ~TIMES_SET_FLAGS;
1515                                 rc = simple_setattr(dentry, &op_data->op_attr);
1516                                 op_data->op_attr.ia_valid = ia_valid;
1517                         }
1518                 } else if (rc != -EPERM && rc != -EACCES && rc != -ETXTBSY) {
1519                         CERROR("md_setattr fails: rc = %d\n", rc);
1520                 }
1521                 RETURN(rc);
1522         }
1523
1524         rc = md_get_lustre_md(sbi->ll_md_exp, request, sbi->ll_dt_exp,
1525                               sbi->ll_md_exp, &md);
1526         if (rc) {
1527                 ptlrpc_req_finished(request);
1528                 RETURN(rc);
1529         }
1530
1531         ia_valid = op_data->op_attr.ia_valid;
1532         /* inode size will be in ll_setattr_ost, can't do it now since dirty
1533          * cache is not cleared yet. */
1534         op_data->op_attr.ia_valid &= ~(TIMES_SET_FLAGS | ATTR_SIZE);
1535         rc = simple_setattr(dentry, &op_data->op_attr);
1536         op_data->op_attr.ia_valid = ia_valid;
1537
1538         /* Extract epoch data if obtained. */
1539         op_data->op_handle = md.body->handle;
1540         op_data->op_ioepoch = md.body->ioepoch;
1541
1542         rc = ll_update_inode(inode, &md);
1543         ptlrpc_req_finished(request);
1544
1545         RETURN(rc);
1546 }
1547
1548 /* Close IO epoch and send Size-on-MDS attribute update. */
1549 static int ll_setattr_done_writing(struct inode *inode,
1550                                    struct md_op_data *op_data,
1551                                    struct md_open_data *mod)
1552 {
1553         struct ll_inode_info *lli = ll_i2info(inode);
1554         int rc = 0;
1555         ENTRY;
1556
1557         LASSERT(op_data != NULL);
1558         if (!S_ISREG(inode->i_mode))
1559                 RETURN(0);
1560
1561         CDEBUG(D_INODE, "Epoch "LPU64" closed on "DFID" for truncate\n",
1562                op_data->op_ioepoch, PFID(&lli->lli_fid));
1563
1564         op_data->op_flags = MF_EPOCH_CLOSE;
1565         ll_done_writing_attr(inode, op_data);
1566         ll_pack_inode2opdata(inode, op_data, NULL);
1567
1568         rc = md_done_writing(ll_i2sbi(inode)->ll_md_exp, op_data, mod);
1569         if (rc == -EAGAIN) {
1570                 /* MDS has instructed us to obtain Size-on-MDS attribute
1571                  * from OSTs and send setattr to back to MDS. */
1572                 rc = ll_som_update(inode, op_data);
1573         } else if (rc) {
1574                 CERROR("%s: inode "DFID" mdc truncate failed: rc = %d\n",
1575                        ll_i2sbi(inode)->ll_md_exp->exp_obd->obd_name,
1576                        PFID(ll_inode2fid(inode)), rc);
1577         }
1578         RETURN(rc);
1579 }
1580
1581 static int ll_setattr_ost(struct inode *inode, struct iattr *attr)
1582 {
1583         struct obd_capa *capa;
1584         int rc;
1585
1586         if (attr->ia_valid & ATTR_SIZE)
1587                 capa = ll_osscapa_get(inode, CAPA_OPC_OSS_TRUNC);
1588         else
1589                 capa = ll_mdscapa_get(inode);
1590
1591         rc = cl_setattr_ost(inode, attr, capa);
1592
1593         if (attr->ia_valid & ATTR_SIZE)
1594                 ll_truncate_free_capa(capa);
1595         else
1596                 capa_put(capa);
1597
1598         return rc;
1599 }
1600
1601 /* If this inode has objects allocated to it (lsm != NULL), then the OST
1602  * object(s) determine the file size and mtime.  Otherwise, the MDS will
1603  * keep these values until such a time that objects are allocated for it.
1604  * We do the MDS operations first, as it is checking permissions for us.
1605  * We don't to the MDS RPC if there is nothing that we want to store there,
1606  * otherwise there is no harm in updating mtime/atime on the MDS if we are
1607  * going to do an RPC anyways.
1608  *
1609  * If we are doing a truncate, we will send the mtime and ctime updates
1610  * to the OST with the punch RPC, otherwise we do an explicit setattr RPC.
1611  * I don't believe it is possible to get e.g. ATTR_MTIME_SET and ATTR_SIZE
1612  * at the same time.
1613  *
1614  * In case of HSMimport, we only set attr on MDS.
1615  */
1616 int ll_setattr_raw(struct dentry *dentry, struct iattr *attr, bool hsm_import)
1617 {
1618         struct inode *inode = dentry->d_inode;
1619         struct ll_inode_info *lli = ll_i2info(inode);
1620         struct md_op_data *op_data = NULL;
1621         struct md_open_data *mod = NULL;
1622         bool file_is_released = false;
1623         int rc = 0, rc1 = 0;
1624         ENTRY;
1625
1626         CDEBUG(D_VFSTRACE, "%s: setattr inode "DFID"(%p) from %llu to %llu, "
1627                "valid %x, hsm_import %d\n",
1628                ll_get_fsname(inode->i_sb, NULL, 0), PFID(&lli->lli_fid),
1629                inode, i_size_read(inode), attr->ia_size, attr->ia_valid,
1630                hsm_import);
1631
1632         if (attr->ia_valid & ATTR_SIZE) {
1633                 /* Check new size against VFS/VM file size limit and rlimit */
1634                 rc = inode_newsize_ok(inode, attr->ia_size);
1635                 if (rc)
1636                         RETURN(rc);
1637
1638                 /* The maximum Lustre file size is variable, based on the
1639                  * OST maximum object size and number of stripes.  This
1640                  * needs another check in addition to the VFS check above. */
1641                 if (attr->ia_size > ll_file_maxbytes(inode)) {
1642                         CDEBUG(D_INODE,"file "DFID" too large %llu > "LPU64"\n",
1643                                PFID(&lli->lli_fid), attr->ia_size,
1644                                ll_file_maxbytes(inode));
1645                         RETURN(-EFBIG);
1646                 }
1647
1648                 attr->ia_valid |= ATTR_MTIME | ATTR_CTIME;
1649         }
1650
1651         /* POSIX: check before ATTR_*TIME_SET set (from inode_change_ok) */
1652         if (attr->ia_valid & TIMES_SET_FLAGS) {
1653                 if ((!uid_eq(current_fsuid(), inode->i_uid)) &&
1654                     !cfs_capable(CFS_CAP_FOWNER))
1655                         RETURN(-EPERM);
1656         }
1657
1658         /* We mark all of the fields "set" so MDS/OST does not re-set them */
1659         if (attr->ia_valid & ATTR_CTIME) {
1660                 attr->ia_ctime = CFS_CURRENT_TIME;
1661                 attr->ia_valid |= ATTR_CTIME_SET;
1662         }
1663         if (!(attr->ia_valid & ATTR_ATIME_SET) &&
1664             (attr->ia_valid & ATTR_ATIME)) {
1665                 attr->ia_atime = CFS_CURRENT_TIME;
1666                 attr->ia_valid |= ATTR_ATIME_SET;
1667         }
1668         if (!(attr->ia_valid & ATTR_MTIME_SET) &&
1669             (attr->ia_valid & ATTR_MTIME)) {
1670                 attr->ia_mtime = CFS_CURRENT_TIME;
1671                 attr->ia_valid |= ATTR_MTIME_SET;
1672         }
1673
1674         if (attr->ia_valid & (ATTR_MTIME | ATTR_CTIME))
1675                 CDEBUG(D_INODE, "setting mtime %lu, ctime %lu, now = %lu\n",
1676                        LTIME_S(attr->ia_mtime), LTIME_S(attr->ia_ctime),
1677                        cfs_time_current_sec());
1678
1679         /* We always do an MDS RPC, even if we're only changing the size;
1680          * only the MDS knows whether truncate() should fail with -ETXTBUSY */
1681
1682         OBD_ALLOC_PTR(op_data);
1683         if (op_data == NULL)
1684                 RETURN(-ENOMEM);
1685
1686         if (!S_ISDIR(inode->i_mode)) {
1687                 if (attr->ia_valid & ATTR_SIZE)
1688                         inode_dio_write_done(inode);
1689                 mutex_unlock(&inode->i_mutex);
1690         }
1691
1692         /* truncate on a released file must failed with -ENODATA,
1693          * so size must not be set on MDS for released file
1694          * but other attributes must be set
1695          */
1696         if (S_ISREG(inode->i_mode)) {
1697                 struct lov_stripe_md *lsm;
1698                 __u32 gen;
1699
1700                 ll_layout_refresh(inode, &gen);
1701                 lsm = ccc_inode_lsm_get(inode);
1702                 if (lsm && lsm->lsm_pattern & LOV_PATTERN_F_RELEASED)
1703                         file_is_released = true;
1704                 ccc_inode_lsm_put(inode, lsm);
1705
1706                 if (!hsm_import && attr->ia_valid & ATTR_SIZE) {
1707                         if (file_is_released) {
1708                                 rc = ll_layout_restore(inode, 0, attr->ia_size);
1709                                 if (rc < 0)
1710                                         GOTO(out, rc);
1711
1712                                 file_is_released = false;
1713                                 ll_layout_refresh(inode, &gen);
1714                         }
1715
1716                         /* If we are changing file size, file content is
1717                          * modified, flag it. */
1718                         attr->ia_valid |= MDS_OPEN_OWNEROVERRIDE;
1719                         spin_lock(&lli->lli_lock);
1720                         lli->lli_flags |= LLIF_DATA_MODIFIED;
1721                         spin_unlock(&lli->lli_lock);
1722                         op_data->op_bias |= MDS_DATA_MODIFIED;
1723                 }
1724         }
1725
1726         memcpy(&op_data->op_attr, attr, sizeof(*attr));
1727
1728         /* Open epoch for truncate. */
1729         if (exp_connect_som(ll_i2mdexp(inode)) && !hsm_import &&
1730             (attr->ia_valid & (ATTR_SIZE | ATTR_MTIME | ATTR_MTIME_SET)))
1731                 op_data->op_flags = MF_EPOCH_OPEN;
1732
1733         rc = ll_md_setattr(dentry, op_data, &mod);
1734         if (rc)
1735                 GOTO(out, rc);
1736
1737         /* RPC to MDT is sent, cancel data modification flag */
1738         if (rc == 0 && (op_data->op_bias & MDS_DATA_MODIFIED)) {
1739                 spin_lock(&lli->lli_lock);
1740                 lli->lli_flags &= ~LLIF_DATA_MODIFIED;
1741                 spin_unlock(&lli->lli_lock);
1742         }
1743
1744         ll_ioepoch_open(lli, op_data->op_ioepoch);
1745         if (!S_ISREG(inode->i_mode) || file_is_released)
1746                 GOTO(out, rc = 0);
1747
1748         if (attr->ia_valid & (ATTR_SIZE |
1749                               ATTR_ATIME | ATTR_ATIME_SET |
1750                               ATTR_MTIME | ATTR_MTIME_SET)) {
1751                 /* For truncate and utimes sending attributes to OSTs, setting
1752                  * mtime/atime to the past will be performed under PW [0:EOF]
1753                  * extent lock (new_size:EOF for truncate).  It may seem
1754                  * excessive to send mtime/atime updates to OSTs when not
1755                  * setting times to past, but it is necessary due to possible
1756                  * time de-synchronization between MDT inode and OST objects */
1757                 if (attr->ia_valid & ATTR_SIZE)
1758                         down_write(&lli->lli_trunc_sem);
1759                 rc = ll_setattr_ost(inode, attr);
1760                 if (attr->ia_valid & ATTR_SIZE)
1761                         up_write(&lli->lli_trunc_sem);
1762         }
1763         EXIT;
1764 out:
1765         if (op_data) {
1766                 if (op_data->op_ioepoch) {
1767                         rc1 = ll_setattr_done_writing(inode, op_data, mod);
1768                         if (!rc)
1769                                 rc = rc1;
1770                 }
1771                 ll_finish_md_op_data(op_data);
1772         }
1773         if (!S_ISDIR(inode->i_mode)) {
1774                 mutex_lock(&inode->i_mutex);
1775                 if ((attr->ia_valid & ATTR_SIZE) && !hsm_import)
1776                         inode_dio_wait(inode);
1777         }
1778
1779         ll_stats_ops_tally(ll_i2sbi(inode), (attr->ia_valid & ATTR_SIZE) ?
1780                         LPROC_LL_TRUNC : LPROC_LL_SETATTR, 1);
1781
1782         return rc;
1783 }
1784
1785 int ll_setattr(struct dentry *de, struct iattr *attr)
1786 {
1787         int mode = de->d_inode->i_mode;
1788
1789         if ((attr->ia_valid & (ATTR_CTIME|ATTR_SIZE|ATTR_MODE)) ==
1790                               (ATTR_CTIME|ATTR_SIZE|ATTR_MODE))
1791                 attr->ia_valid |= MDS_OPEN_OWNEROVERRIDE;
1792
1793         if (((attr->ia_valid & (ATTR_MODE|ATTR_FORCE|ATTR_SIZE)) ==
1794                                (ATTR_SIZE|ATTR_MODE)) &&
1795             (((mode & S_ISUID) && !(attr->ia_mode & S_ISUID)) ||
1796              (((mode & (S_ISGID|S_IXGRP)) == (S_ISGID|S_IXGRP)) &&
1797               !(attr->ia_mode & S_ISGID))))
1798                 attr->ia_valid |= ATTR_FORCE;
1799
1800         if ((attr->ia_valid & ATTR_MODE) &&
1801             (mode & S_ISUID) &&
1802             !(attr->ia_mode & S_ISUID) &&
1803             !(attr->ia_valid & ATTR_KILL_SUID))
1804                 attr->ia_valid |= ATTR_KILL_SUID;
1805
1806         if ((attr->ia_valid & ATTR_MODE) &&
1807             ((mode & (S_ISGID|S_IXGRP)) == (S_ISGID|S_IXGRP)) &&
1808             !(attr->ia_mode & S_ISGID) &&
1809             !(attr->ia_valid & ATTR_KILL_SGID))
1810                 attr->ia_valid |= ATTR_KILL_SGID;
1811
1812         return ll_setattr_raw(de, attr, false);
1813 }
1814
1815 int ll_statfs_internal(struct super_block *sb, struct obd_statfs *osfs,
1816                        __u64 max_age, __u32 flags)
1817 {
1818         struct ll_sb_info *sbi = ll_s2sbi(sb);
1819         struct obd_statfs obd_osfs;
1820         int rc;
1821         ENTRY;
1822
1823         rc = obd_statfs(NULL, sbi->ll_md_exp, osfs, max_age, flags);
1824         if (rc) {
1825                 CERROR("md_statfs fails: rc = %d\n", rc);
1826                 RETURN(rc);
1827         }
1828
1829         osfs->os_type = sb->s_magic;
1830
1831         CDEBUG(D_SUPER, "MDC blocks "LPU64"/"LPU64" objects "LPU64"/"LPU64"\n",
1832                osfs->os_bavail, osfs->os_blocks, osfs->os_ffree,osfs->os_files);
1833
1834         if (sbi->ll_flags & LL_SBI_LAZYSTATFS)
1835                 flags |= OBD_STATFS_NODELAY;
1836
1837         rc = obd_statfs_rqset(sbi->ll_dt_exp, &obd_osfs, max_age, flags);
1838         if (rc) {
1839                 CERROR("obd_statfs fails: rc = %d\n", rc);
1840                 RETURN(rc);
1841         }
1842
1843         CDEBUG(D_SUPER, "OSC blocks "LPU64"/"LPU64" objects "LPU64"/"LPU64"\n",
1844                obd_osfs.os_bavail, obd_osfs.os_blocks, obd_osfs.os_ffree,
1845                obd_osfs.os_files);
1846
1847         osfs->os_bsize = obd_osfs.os_bsize;
1848         osfs->os_blocks = obd_osfs.os_blocks;
1849         osfs->os_bfree = obd_osfs.os_bfree;
1850         osfs->os_bavail = obd_osfs.os_bavail;
1851
1852         /* If we don't have as many objects free on the OST as inodes
1853          * on the MDS, we reduce the total number of inodes to
1854          * compensate, so that the "inodes in use" number is correct.
1855          */
1856         if (obd_osfs.os_ffree < osfs->os_ffree) {
1857                 osfs->os_files = (osfs->os_files - osfs->os_ffree) +
1858                         obd_osfs.os_ffree;
1859                 osfs->os_ffree = obd_osfs.os_ffree;
1860         }
1861
1862         RETURN(rc);
1863 }
1864 int ll_statfs(struct dentry *de, struct kstatfs *sfs)
1865 {
1866         struct super_block *sb = de->d_sb;
1867         struct obd_statfs osfs;
1868         __u64 fsid = huge_encode_dev(sb->s_dev);
1869         int rc;
1870
1871         CDEBUG(D_VFSTRACE, "VFS Op: at "LPU64" jiffies\n", get_jiffies_64());
1872         ll_stats_ops_tally(ll_s2sbi(sb), LPROC_LL_STAFS, 1);
1873
1874         /* Some amount of caching on the client is allowed */
1875         rc = ll_statfs_internal(sb, &osfs,
1876                                 cfs_time_shift_64(-OBD_STATFS_CACHE_SECONDS),
1877                                 0);
1878         if (rc)
1879                 return rc;
1880
1881         statfs_unpack(sfs, &osfs);
1882
1883         /* We need to downshift for all 32-bit kernels, because we can't
1884          * tell if the kernel is being called via sys_statfs64() or not.
1885          * Stop before overflowing f_bsize - in which case it is better
1886          * to just risk EOVERFLOW if caller is using old sys_statfs(). */
1887         if (sizeof(long) < 8) {
1888                 while (osfs.os_blocks > ~0UL && sfs->f_bsize < 0x40000000) {
1889                         sfs->f_bsize <<= 1;
1890
1891                         osfs.os_blocks >>= 1;
1892                         osfs.os_bfree >>= 1;
1893                         osfs.os_bavail >>= 1;
1894                 }
1895         }
1896
1897         sfs->f_blocks = osfs.os_blocks;
1898         sfs->f_bfree = osfs.os_bfree;
1899         sfs->f_bavail = osfs.os_bavail;
1900         sfs->f_fsid.val[0] = (__u32)fsid;
1901         sfs->f_fsid.val[1] = (__u32)(fsid >> 32);
1902         return 0;
1903 }
1904
1905 void ll_inode_size_lock(struct inode *inode)
1906 {
1907         struct ll_inode_info *lli;
1908
1909         LASSERT(!S_ISDIR(inode->i_mode));
1910
1911         lli = ll_i2info(inode);
1912         mutex_lock(&lli->lli_size_mutex);
1913 }
1914
1915 void ll_inode_size_unlock(struct inode *inode)
1916 {
1917         struct ll_inode_info *lli;
1918
1919         lli = ll_i2info(inode);
1920         mutex_unlock(&lli->lli_size_mutex);
1921 }
1922
1923 int ll_update_inode(struct inode *inode, struct lustre_md *md)
1924 {
1925         struct ll_inode_info *lli = ll_i2info(inode);
1926         struct mdt_body *body = md->body;
1927         struct lov_stripe_md *lsm = md->lsm;
1928         struct ll_sb_info *sbi = ll_i2sbi(inode);
1929
1930         LASSERT ((lsm != NULL) == ((body->valid & OBD_MD_FLEASIZE) != 0));
1931         if (lsm != NULL) {
1932                 if (!lli->lli_has_smd &&
1933                     !(sbi->ll_flags & LL_SBI_LAYOUT_LOCK))
1934                         cl_file_inode_init(inode, md);
1935
1936                 lli->lli_maxbytes = lsm->lsm_maxbytes;
1937                 if (lli->lli_maxbytes > MAX_LFS_FILESIZE)
1938                         lli->lli_maxbytes = MAX_LFS_FILESIZE;
1939         }
1940
1941         if (S_ISDIR(inode->i_mode)) {
1942                 int     rc;
1943
1944                 rc = ll_update_lsm_md(inode, md);
1945                 if (rc != 0)
1946                         return rc;
1947         }
1948
1949         if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
1950                 if (body->valid & OBD_MD_FLRMTPERM)
1951                         ll_update_remote_perm(inode, md->remote_perm);
1952         }
1953 #ifdef CONFIG_FS_POSIX_ACL
1954         else if (body->valid & OBD_MD_FLACL) {
1955                 spin_lock(&lli->lli_lock);
1956                 if (lli->lli_posix_acl)
1957                         posix_acl_release(lli->lli_posix_acl);
1958                 lli->lli_posix_acl = md->posix_acl;
1959                 spin_unlock(&lli->lli_lock);
1960         }
1961 #endif
1962         inode->i_ino = cl_fid_build_ino(&body->fid1,
1963                                         sbi->ll_flags & LL_SBI_32BIT_API);
1964         inode->i_generation = cl_fid_build_gen(&body->fid1);
1965
1966         if (body->valid & OBD_MD_FLATIME) {
1967                 if (body->atime > LTIME_S(inode->i_atime))
1968                         LTIME_S(inode->i_atime) = body->atime;
1969                 lli->lli_lvb.lvb_atime = body->atime;
1970         }
1971         if (body->valid & OBD_MD_FLMTIME) {
1972                 if (body->mtime > LTIME_S(inode->i_mtime)) {
1973                         CDEBUG(D_INODE, "setting ino %lu mtime from %lu "
1974                                "to "LPU64"\n", inode->i_ino,
1975                                LTIME_S(inode->i_mtime), body->mtime);
1976                         LTIME_S(inode->i_mtime) = body->mtime;
1977                 }
1978                 lli->lli_lvb.lvb_mtime = body->mtime;
1979         }
1980         if (body->valid & OBD_MD_FLCTIME) {
1981                 if (body->ctime > LTIME_S(inode->i_ctime))
1982                         LTIME_S(inode->i_ctime) = body->ctime;
1983                 lli->lli_lvb.lvb_ctime = body->ctime;
1984         }
1985         if (body->valid & OBD_MD_FLMODE)
1986                 inode->i_mode = (inode->i_mode & S_IFMT)|(body->mode & ~S_IFMT);
1987         if (body->valid & OBD_MD_FLTYPE)
1988                 inode->i_mode = (inode->i_mode & ~S_IFMT)|(body->mode & S_IFMT);
1989         LASSERT(inode->i_mode != 0);
1990         if (S_ISREG(inode->i_mode)) {
1991                 inode->i_blkbits = min(PTLRPC_MAX_BRW_BITS + 1, LL_MAX_BLKSIZE_BITS);
1992         } else {
1993                 inode->i_blkbits = inode->i_sb->s_blocksize_bits;
1994         }
1995         if (body->valid & OBD_MD_FLUID)
1996                 inode->i_uid = make_kuid(&init_user_ns, body->uid);
1997         if (body->valid & OBD_MD_FLGID)
1998                 inode->i_gid = make_kgid(&init_user_ns, body->gid);
1999         if (body->valid & OBD_MD_FLFLAGS)
2000                 inode->i_flags = ll_ext_to_inode_flags(body->flags);
2001         if (body->valid & OBD_MD_FLNLINK)
2002                 set_nlink(inode, body->nlink);
2003         if (body->valid & OBD_MD_FLRDEV)
2004                 inode->i_rdev = old_decode_dev(body->rdev);
2005
2006         if (body->valid & OBD_MD_FLID) {
2007                 /* FID shouldn't be changed! */
2008                 if (fid_is_sane(&lli->lli_fid)) {
2009                         LASSERTF(lu_fid_eq(&lli->lli_fid, &body->fid1),
2010                                  "Trying to change FID "DFID
2011                                  " to the "DFID", inode "DFID"(%p)\n",
2012                                  PFID(&lli->lli_fid), PFID(&body->fid1),
2013                                  PFID(ll_inode2fid(inode)), inode);
2014                 } else {
2015                         lli->lli_fid = body->fid1;
2016                 }
2017         }
2018
2019         LASSERT(fid_seq(&lli->lli_fid) != 0);
2020
2021         if (body->valid & OBD_MD_FLSIZE) {
2022                 if (exp_connect_som(ll_i2mdexp(inode)) &&
2023                     S_ISREG(inode->i_mode)) {
2024                         struct lustre_handle lockh;
2025                         ldlm_mode_t mode;
2026
2027                         /* As it is possible a blocking ast has been processed
2028                          * by this time, we need to check there is an UPDATE
2029                          * lock on the client and set LLIF_MDS_SIZE_LOCK holding
2030                          * it. */
2031                         mode = ll_take_md_lock(inode, MDS_INODELOCK_UPDATE,
2032                                                &lockh, LDLM_FL_CBPENDING,
2033                                                LCK_CR | LCK_CW |
2034                                                LCK_PR | LCK_PW);
2035                         if (mode) {
2036                                 if (lli->lli_flags & (LLIF_DONE_WRITING |
2037                                                       LLIF_EPOCH_PENDING |
2038                                                       LLIF_SOM_DIRTY)) {
2039                                         CERROR("%s: inode "DFID" flags %u still"
2040                                                " has size authority! do not "
2041                                                "trust the size from MDS\n",
2042                                                sbi->ll_md_exp->exp_obd->obd_name,
2043                                                PFID(ll_inode2fid(inode)),
2044                                                lli->lli_flags);
2045                                 } else {
2046                                         /* Use old size assignment to avoid
2047                                          * deadlock bz14138 & bz14326 */
2048                                         i_size_write(inode, body->size);
2049                                         spin_lock(&lli->lli_lock);
2050                                         lli->lli_flags |= LLIF_MDS_SIZE_LOCK;
2051                                         spin_unlock(&lli->lli_lock);
2052                                 }
2053                                 ldlm_lock_decref(&lockh, mode);
2054                         }
2055                 } else {
2056                         /* Use old size assignment to avoid
2057                          * deadlock bz14138 & bz14326 */
2058                         i_size_write(inode, body->size);
2059
2060                         CDEBUG(D_VFSTRACE,
2061                                "inode="DFID", updating i_size %llu\n",
2062                                PFID(ll_inode2fid(inode)),
2063                                (unsigned long long)body->size);
2064                 }
2065
2066                 if (body->valid & OBD_MD_FLBLOCKS)
2067                         inode->i_blocks = body->blocks;
2068         }
2069
2070         if (body->valid & OBD_MD_FLMDSCAPA) {
2071                 LASSERT(md->mds_capa);
2072                 ll_add_capa(inode, md->mds_capa);
2073         }
2074         if (body->valid & OBD_MD_FLOSSCAPA) {
2075                 LASSERT(md->oss_capa);
2076                 ll_add_capa(inode, md->oss_capa);
2077         }
2078
2079         if (body->valid & OBD_MD_TSTATE) {
2080                 if (body->t_state & MS_RESTORE)
2081                         lli->lli_flags |= LLIF_FILE_RESTORING;
2082         }
2083
2084         return 0;
2085 }
2086
2087 int ll_read_inode2(struct inode *inode, void *opaque)
2088 {
2089         struct lustre_md *md = opaque;
2090         struct ll_inode_info *lli = ll_i2info(inode);
2091         int     rc;
2092         ENTRY;
2093
2094         CDEBUG(D_VFSTRACE, "VFS Op:inode="DFID"(%p)\n",
2095                PFID(&lli->lli_fid), inode);
2096
2097         LASSERT(!lli->lli_has_smd);
2098
2099         /* Core attributes from the MDS first.  This is a new inode, and
2100          * the VFS doesn't zero times in the core inode so we have to do
2101          * it ourselves.  They will be overwritten by either MDS or OST
2102          * attributes - we just need to make sure they aren't newer. */
2103         LTIME_S(inode->i_mtime) = 0;
2104         LTIME_S(inode->i_atime) = 0;
2105         LTIME_S(inode->i_ctime) = 0;
2106         inode->i_rdev = 0;
2107         rc = ll_update_inode(inode, md);
2108         if (rc != 0)
2109                 RETURN(rc);
2110
2111         /* OIDEBUG(inode); */
2112
2113         /* initializing backing dev info. */
2114         inode->i_mapping->backing_dev_info = &s2lsi(inode->i_sb)->lsi_bdi;
2115
2116
2117         if (S_ISREG(inode->i_mode)) {
2118                 struct ll_sb_info *sbi = ll_i2sbi(inode);
2119                 inode->i_op = &ll_file_inode_operations;
2120                 inode->i_fop = sbi->ll_fop;
2121                 inode->i_mapping->a_ops = (struct address_space_operations *)&ll_aops;
2122                 EXIT;
2123         } else if (S_ISDIR(inode->i_mode)) {
2124                 inode->i_op = &ll_dir_inode_operations;
2125                 inode->i_fop = &ll_dir_operations;
2126                 EXIT;
2127         } else if (S_ISLNK(inode->i_mode)) {
2128                 inode->i_op = &ll_fast_symlink_inode_operations;
2129                 EXIT;
2130         } else {
2131                 inode->i_op = &ll_special_inode_operations;
2132
2133                 init_special_inode(inode, inode->i_mode,
2134                                    inode->i_rdev);
2135
2136                 EXIT;
2137         }
2138
2139         return 0;
2140 }
2141
2142 void ll_delete_inode(struct inode *inode)
2143 {
2144         struct cl_inode_info *lli = cl_i2info(inode);
2145         ENTRY;
2146
2147         if (S_ISREG(inode->i_mode) && lli->lli_clob != NULL)
2148                 /* discard all dirty pages before truncating them, required by
2149                  * osc_extent implementation at LU-1030. */
2150                 cl_sync_file_range(inode, 0, OBD_OBJECT_EOF,
2151                                    CL_FSYNC_DISCARD, 1);
2152
2153         truncate_inode_pages(&inode->i_data, 0);
2154
2155         /* Workaround for LU-118 */
2156         if (inode->i_data.nrpages) {
2157                 spin_lock_irq(&inode->i_data.tree_lock);
2158                 spin_unlock_irq(&inode->i_data.tree_lock);
2159                 LASSERTF(inode->i_data.nrpages == 0,
2160                          "inode="DFID"(%p) nrpages=%lu, see "
2161                          "http://jira.whamcloud.com/browse/LU-118\n",
2162                          PFID(ll_inode2fid(inode)), inode,
2163                          inode->i_data.nrpages);
2164         }
2165         /* Workaround end */
2166
2167 #ifdef HAVE_SBOPS_EVICT_INODE
2168         ll_clear_inode(inode);
2169 #endif
2170         clear_inode(inode);
2171
2172         EXIT;
2173 }
2174
2175 int ll_iocontrol(struct inode *inode, struct file *file,
2176                  unsigned int cmd, unsigned long arg)
2177 {
2178         struct ll_sb_info *sbi = ll_i2sbi(inode);
2179         struct ptlrpc_request *req = NULL;
2180         int rc, flags = 0;
2181         ENTRY;
2182
2183         switch(cmd) {
2184         case FSFILT_IOC_GETFLAGS: {
2185                 struct mdt_body *body;
2186                 struct md_op_data *op_data;
2187
2188                 op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL,
2189                                              0, 0, LUSTRE_OPC_ANY,
2190                                              NULL);
2191                 if (IS_ERR(op_data))
2192                         RETURN(PTR_ERR(op_data));
2193
2194                 op_data->op_valid = OBD_MD_FLFLAGS;
2195                 rc = md_getattr(sbi->ll_md_exp, op_data, &req);
2196                 ll_finish_md_op_data(op_data);
2197                 if (rc) {
2198                         CERROR("%s: failure inode "DFID": rc = %d\n",
2199                                sbi->ll_md_exp->exp_obd->obd_name,
2200                                PFID(ll_inode2fid(inode)), rc);
2201                         RETURN(-abs(rc));
2202                 }
2203
2204                 body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
2205
2206                 flags = body->flags;
2207
2208                 ptlrpc_req_finished(req);
2209
2210                 RETURN(put_user(flags, (int *)arg));
2211         }
2212         case FSFILT_IOC_SETFLAGS: {
2213                 struct lov_stripe_md *lsm;
2214                 struct obd_info oinfo = { { { 0 } } };
2215                 struct md_op_data *op_data;
2216
2217                 if (get_user(flags, (int *)arg))
2218                         RETURN(-EFAULT);
2219
2220                 op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL, 0, 0,
2221                                              LUSTRE_OPC_ANY, NULL);
2222                 if (IS_ERR(op_data))
2223                         RETURN(PTR_ERR(op_data));
2224
2225                 ((struct ll_iattr *)&op_data->op_attr)->ia_attr_flags = flags;
2226                 op_data->op_attr.ia_valid |= ATTR_ATTR_FLAG;
2227                 rc = md_setattr(sbi->ll_md_exp, op_data,
2228                                 NULL, 0, NULL, 0, &req, NULL);
2229                 ll_finish_md_op_data(op_data);
2230                 ptlrpc_req_finished(req);
2231                 if (rc)
2232                         RETURN(rc);
2233
2234                 inode->i_flags = ll_ext_to_inode_flags(flags);
2235
2236                 lsm = ccc_inode_lsm_get(inode);
2237                 if (!lsm_has_objects(lsm)) {
2238                         ccc_inode_lsm_put(inode, lsm);
2239                         RETURN(0);
2240                 }
2241
2242                 OBDO_ALLOC(oinfo.oi_oa);
2243                 if (!oinfo.oi_oa) {
2244                         ccc_inode_lsm_put(inode, lsm);
2245                         RETURN(-ENOMEM);
2246                 }
2247                 oinfo.oi_md = lsm;
2248                 oinfo.oi_oa->o_oi = lsm->lsm_oi;
2249                 oinfo.oi_oa->o_flags = flags;
2250                 oinfo.oi_oa->o_valid = OBD_MD_FLID | OBD_MD_FLFLAGS |
2251                                        OBD_MD_FLGROUP;
2252                 oinfo.oi_capa = ll_mdscapa_get(inode);
2253                 obdo_set_parent_fid(oinfo.oi_oa, &ll_i2info(inode)->lli_fid);
2254                 rc = obd_setattr_rqset(sbi->ll_dt_exp, &oinfo, NULL);
2255                 capa_put(oinfo.oi_capa);
2256                 OBDO_FREE(oinfo.oi_oa);
2257                 ccc_inode_lsm_put(inode, lsm);
2258
2259                 if (rc && rc != -EPERM && rc != -EACCES)
2260                         CERROR("osc_setattr_async fails: rc = %d\n", rc);
2261
2262                 RETURN(rc);
2263         }
2264         default:
2265                 RETURN(-ENOSYS);
2266         }
2267
2268         RETURN(0);
2269 }
2270
2271 int ll_flush_ctx(struct inode *inode)
2272 {
2273         struct ll_sb_info  *sbi = ll_i2sbi(inode);
2274
2275         CDEBUG(D_SEC, "flush context for user %d\n",
2276                from_kuid(&init_user_ns, current_uid()));
2277
2278         obd_set_info_async(NULL, sbi->ll_md_exp,
2279                            sizeof(KEY_FLUSH_CTX), KEY_FLUSH_CTX,
2280                            0, NULL, NULL);
2281         obd_set_info_async(NULL, sbi->ll_dt_exp,
2282                            sizeof(KEY_FLUSH_CTX), KEY_FLUSH_CTX,
2283                            0, NULL, NULL);
2284         return 0;
2285 }
2286
2287 /* umount -f client means force down, don't save state */
2288 void ll_umount_begin(struct super_block *sb)
2289 {
2290         struct ll_sb_info *sbi = ll_s2sbi(sb);
2291         struct obd_device *obd;
2292         struct obd_ioctl_data *ioc_data;
2293         ENTRY;
2294
2295         CDEBUG(D_VFSTRACE, "VFS Op: superblock %p count %d active %d\n", sb,
2296                sb->s_count, atomic_read(&sb->s_active));
2297
2298         obd = class_exp2obd(sbi->ll_md_exp);
2299         if (obd == NULL) {
2300                 CERROR("Invalid MDC connection handle "LPX64"\n",
2301                        sbi->ll_md_exp->exp_handle.h_cookie);
2302                 EXIT;
2303                 return;
2304         }
2305         obd->obd_force = 1;
2306
2307         obd = class_exp2obd(sbi->ll_dt_exp);
2308         if (obd == NULL) {
2309                 CERROR("Invalid LOV connection handle "LPX64"\n",
2310                        sbi->ll_dt_exp->exp_handle.h_cookie);
2311                 EXIT;
2312                 return;
2313         }
2314         obd->obd_force = 1;
2315
2316         OBD_ALLOC_PTR(ioc_data);
2317         if (ioc_data) {
2318                 obd_iocontrol(IOC_OSC_SET_ACTIVE, sbi->ll_md_exp,
2319                               sizeof *ioc_data, ioc_data, NULL);
2320
2321                 obd_iocontrol(IOC_OSC_SET_ACTIVE, sbi->ll_dt_exp,
2322                               sizeof *ioc_data, ioc_data, NULL);
2323
2324                 OBD_FREE_PTR(ioc_data);
2325         }
2326
2327         /* Really, we'd like to wait until there are no requests outstanding,
2328          * and then continue.  For now, we just invalidate the requests,
2329          * schedule() and sleep one second if needed, and hope.
2330          */
2331         schedule();
2332         EXIT;
2333 }
2334
2335 int ll_remount_fs(struct super_block *sb, int *flags, char *data)
2336 {
2337         struct ll_sb_info *sbi = ll_s2sbi(sb);
2338         char *profilenm = get_profile_name(sb);
2339         int err;
2340         __u32 read_only;
2341
2342         if ((*flags & MS_RDONLY) != (sb->s_flags & MS_RDONLY)) {
2343                 read_only = *flags & MS_RDONLY;
2344                 err = obd_set_info_async(NULL, sbi->ll_md_exp,
2345                                          sizeof(KEY_READ_ONLY),
2346                                          KEY_READ_ONLY, sizeof(read_only),
2347                                          &read_only, NULL);
2348                 if (err) {
2349                         LCONSOLE_WARN("Failed to remount %s %s (%d)\n",
2350                                       profilenm, read_only ?
2351                                       "read-only" : "read-write", err);
2352                         return err;
2353                 }
2354
2355                 if (read_only)
2356                         sb->s_flags |= MS_RDONLY;
2357                 else
2358                         sb->s_flags &= ~MS_RDONLY;
2359
2360                 if (sbi->ll_flags & LL_SBI_VERBOSE)
2361                         LCONSOLE_WARN("Remounted %s %s\n", profilenm,
2362                                       read_only ?  "read-only" : "read-write");
2363         }
2364         return 0;
2365 }
2366
2367 int ll_prep_inode(struct inode **inode, struct ptlrpc_request *req,
2368                   struct super_block *sb, struct lookup_intent *it)
2369 {
2370         struct ll_sb_info *sbi = NULL;
2371         struct lustre_md md = { 0 };
2372         int rc;
2373         ENTRY;
2374
2375         LASSERT(*inode || sb);
2376         sbi = sb ? ll_s2sbi(sb) : ll_i2sbi(*inode);
2377         rc = md_get_lustre_md(sbi->ll_md_exp, req, sbi->ll_dt_exp,
2378                               sbi->ll_md_exp, &md);
2379         if (rc)
2380                 RETURN(rc);
2381
2382         if (*inode) {
2383                 rc = ll_update_inode(*inode, &md);
2384                 if (rc != 0)
2385                         GOTO(out, rc);
2386         } else {
2387                 LASSERT(sb != NULL);
2388
2389                 /*
2390                  * At this point server returns to client's same fid as client
2391                  * generated for creating. So using ->fid1 is okay here.
2392                  */
2393                 LASSERT(fid_is_sane(&md.body->fid1));
2394
2395                 *inode = ll_iget(sb, cl_fid_build_ino(&md.body->fid1,
2396                                              sbi->ll_flags & LL_SBI_32BIT_API),
2397                                  &md);
2398                 if (IS_ERR(*inode)) {
2399 #ifdef CONFIG_FS_POSIX_ACL
2400                         if (md.posix_acl) {
2401                                 posix_acl_release(md.posix_acl);
2402                                 md.posix_acl = NULL;
2403                         }
2404 #endif
2405                         rc = IS_ERR(*inode) ? PTR_ERR(*inode) : -ENOMEM;
2406                         *inode = NULL;
2407                         CERROR("new_inode -fatal: rc %d\n", rc);
2408                         GOTO(out, rc);
2409                 }
2410         }
2411
2412         /* Handling piggyback layout lock.
2413          * Layout lock can be piggybacked by getattr and open request.
2414          * The lsm can be applied to inode only if it comes with a layout lock
2415          * otherwise correct layout may be overwritten, for example:
2416          * 1. proc1: mdt returns a lsm but not granting layout
2417          * 2. layout was changed by another client
2418          * 3. proc2: refresh layout and layout lock granted
2419          * 4. proc1: to apply a stale layout */
2420         if (it != NULL && it->d.lustre.it_lock_mode != 0) {
2421                 struct lustre_handle lockh;
2422                 struct ldlm_lock *lock;
2423
2424                 lockh.cookie = it->d.lustre.it_lock_handle;
2425                 lock = ldlm_handle2lock(&lockh);
2426                 LASSERT(lock != NULL);
2427                 if (ldlm_has_layout(lock)) {
2428                         struct cl_object_conf conf;
2429
2430                         memset(&conf, 0, sizeof(conf));
2431                         conf.coc_opc = OBJECT_CONF_SET;
2432                         conf.coc_inode = *inode;
2433                         conf.coc_lock = lock;
2434                         conf.u.coc_md = &md;
2435                         (void)ll_layout_conf(*inode, &conf);
2436                 }
2437                 LDLM_LOCK_PUT(lock);
2438         }
2439
2440 out:
2441         if (md.lsm != NULL)
2442                 obd_free_memmd(sbi->ll_dt_exp, &md.lsm);
2443         md_free_lustre_md(sbi->ll_md_exp, &md);
2444         RETURN(rc);
2445 }
2446
2447 int ll_obd_statfs(struct inode *inode, void *arg)
2448 {
2449         struct ll_sb_info *sbi = NULL;
2450         struct obd_export *exp;
2451         char *buf = NULL;
2452         struct obd_ioctl_data *data = NULL;
2453         __u32 type;
2454         __u32 flags;
2455         int len = 0, rc;
2456
2457         if (!inode || !(sbi = ll_i2sbi(inode)))
2458                 GOTO(out_statfs, rc = -EINVAL);
2459
2460         rc = obd_ioctl_getdata(&buf, &len, arg);
2461         if (rc)
2462                 GOTO(out_statfs, rc);
2463
2464         data = (void*)buf;
2465         if (!data->ioc_inlbuf1 || !data->ioc_inlbuf2 ||
2466             !data->ioc_pbuf1 || !data->ioc_pbuf2)
2467                 GOTO(out_statfs, rc = -EINVAL);
2468
2469         if (data->ioc_inllen1 != sizeof(__u32) ||
2470             data->ioc_inllen2 != sizeof(__u32) ||
2471             data->ioc_plen1 != sizeof(struct obd_statfs) ||
2472             data->ioc_plen2 != sizeof(struct obd_uuid))
2473                 GOTO(out_statfs, rc = -EINVAL);
2474
2475         memcpy(&type, data->ioc_inlbuf1, sizeof(__u32));
2476         if (type & LL_STATFS_LMV)
2477                 exp = sbi->ll_md_exp;
2478         else if (type & LL_STATFS_LOV)
2479                 exp = sbi->ll_dt_exp;
2480         else
2481                 GOTO(out_statfs, rc = -ENODEV);
2482
2483         flags = (type & LL_STATFS_NODELAY) ? OBD_STATFS_NODELAY : 0;
2484         rc = obd_iocontrol(IOC_OBD_STATFS, exp, len, buf, &flags);
2485         if (rc)
2486                 GOTO(out_statfs, rc);
2487 out_statfs:
2488         if (buf)
2489                 obd_ioctl_freedata(buf, len);
2490         return rc;
2491 }
2492
2493 int ll_process_config(struct lustre_cfg *lcfg)
2494 {
2495         struct super_block *sb;
2496         unsigned long x;
2497         int rc = 0;
2498         char *ptr;
2499
2500         /* The instance name contains the sb: lustre-client-aacfe000 */
2501         ptr = strrchr(lustre_cfg_string(lcfg, 0), '-');
2502         if (!ptr || !*(++ptr))
2503                 return -EINVAL;
2504         if (sscanf(ptr, "%lx", &x) != 1)
2505                 return -EINVAL;
2506         sb = (struct super_block *)x;
2507         /* This better be a real Lustre superblock! */
2508         LASSERT(s2lsi(sb)->lsi_lmd->lmd_magic == LMD_MAGIC);
2509
2510         /* Note we have not called client_common_fill_super yet, so
2511            proc fns must be able to handle that! */
2512         rc = class_process_proc_seq_param(PARAM_LLITE, lprocfs_llite_obd_vars,
2513                                           lcfg, sb);
2514         if (rc > 0)
2515                 rc = 0;
2516         return rc;
2517 }
2518
2519 /* this function prepares md_op_data hint for passing ot down to MD stack. */
2520 struct md_op_data * ll_prep_md_op_data(struct md_op_data *op_data,
2521                                        struct inode *i1, struct inode *i2,
2522                                        const char *name, int namelen,
2523                                        int mode, __u32 opc, void *data)
2524 {
2525         LASSERT(i1 != NULL);
2526
2527         if (name == NULL) {
2528                 /* Do not reuse namelen for something else. */
2529                 if (namelen != 0)
2530                         return ERR_PTR(-EINVAL);
2531         } else {
2532                 if (namelen > ll_i2sbi(i1)->ll_namelen)
2533                         return ERR_PTR(-ENAMETOOLONG);
2534
2535                 if (!lu_name_is_valid_2(name, namelen))
2536                         return ERR_PTR(-EINVAL);
2537         }
2538
2539         if (op_data == NULL)
2540                 OBD_ALLOC_PTR(op_data);
2541
2542         if (op_data == NULL)
2543                 return ERR_PTR(-ENOMEM);
2544
2545         ll_i2gids(op_data->op_suppgids, i1, i2);
2546         op_data->op_fid1 = *ll_inode2fid(i1);
2547         op_data->op_capa1 = ll_mdscapa_get(i1);
2548         if (S_ISDIR(i1->i_mode))
2549                 op_data->op_mea1 = ll_i2info(i1)->lli_lsm_md;
2550
2551         if (i2) {
2552                 op_data->op_fid2 = *ll_inode2fid(i2);
2553                 op_data->op_capa2 = ll_mdscapa_get(i2);
2554                 if (S_ISDIR(i2->i_mode))
2555                         op_data->op_mea2 = ll_i2info(i2)->lli_lsm_md;
2556         } else {
2557                 fid_zero(&op_data->op_fid2);
2558                 op_data->op_capa2 = NULL;
2559         }
2560
2561         if (ll_i2sbi(i1)->ll_flags & LL_SBI_64BIT_HASH)
2562                 op_data->op_cli_flags |= CLI_HASH64;
2563
2564         if (ll_need_32bit_api(ll_i2sbi(i1)))
2565                 op_data->op_cli_flags |= CLI_API32;
2566
2567         op_data->op_name = name;
2568         op_data->op_namelen = namelen;
2569         op_data->op_mode = mode;
2570         op_data->op_mod_time = cfs_time_current_sec();
2571         op_data->op_fsuid = from_kuid(&init_user_ns, current_fsuid());
2572         op_data->op_fsgid = from_kgid(&init_user_ns, current_fsgid());
2573         op_data->op_cap = cfs_curproc_cap_pack();
2574         op_data->op_bias = 0;
2575         op_data->op_cli_flags = 0;
2576         if ((opc == LUSTRE_OPC_CREATE) && (name != NULL) &&
2577              filename_is_volatile(name, namelen, NULL))
2578                 op_data->op_bias |= MDS_CREATE_VOLATILE;
2579         op_data->op_mds = 0;
2580         op_data->op_data = data;
2581
2582         /* If the file is being opened after mknod() (normally due to NFS)
2583          * try to use the default stripe data from parent directory for
2584          * allocating OST objects.  Try to pass the parent FID to MDS. */
2585         if (opc == LUSTRE_OPC_CREATE && i1 == i2 && S_ISREG(i2->i_mode) &&
2586             !ll_i2info(i2)->lli_has_smd) {
2587                 struct ll_inode_info *lli = ll_i2info(i2);
2588
2589                 spin_lock(&lli->lli_lock);
2590                 if (likely(!lli->lli_has_smd && !fid_is_zero(&lli->lli_pfid)))
2591                         op_data->op_fid1 = lli->lli_pfid;
2592                 spin_unlock(&lli->lli_lock);
2593                 /** We ignore parent's capability temporary. */
2594         }
2595
2596         /* When called by ll_setattr_raw, file is i1. */
2597         if (LLIF_DATA_MODIFIED & ll_i2info(i1)->lli_flags)
2598                 op_data->op_bias |= MDS_DATA_MODIFIED;
2599
2600         return op_data;
2601 }
2602
2603 void ll_finish_md_op_data(struct md_op_data *op_data)
2604 {
2605         capa_put(op_data->op_capa1);
2606         capa_put(op_data->op_capa2);
2607         OBD_FREE_PTR(op_data);
2608 }
2609
2610 #ifdef HAVE_SUPEROPS_USE_DENTRY
2611 int ll_show_options(struct seq_file *seq, struct dentry *dentry)
2612 #else
2613 int ll_show_options(struct seq_file *seq, struct vfsmount *vfs)
2614 #endif
2615 {
2616         struct ll_sb_info *sbi;
2617
2618 #ifdef HAVE_SUPEROPS_USE_DENTRY
2619         LASSERT((seq != NULL) && (dentry != NULL));
2620         sbi = ll_s2sbi(dentry->d_sb);
2621 #else
2622         LASSERT((seq != NULL) && (vfs != NULL));
2623         sbi = ll_s2sbi(vfs->mnt_sb);
2624 #endif
2625
2626         if (sbi->ll_flags & LL_SBI_NOLCK)
2627                 seq_puts(seq, ",nolock");
2628
2629         if (sbi->ll_flags & LL_SBI_FLOCK)
2630                 seq_puts(seq, ",flock");
2631
2632         if (sbi->ll_flags & LL_SBI_LOCALFLOCK)
2633                 seq_puts(seq, ",localflock");
2634
2635         if (sbi->ll_flags & LL_SBI_USER_XATTR)
2636                 seq_puts(seq, ",user_xattr");
2637
2638         if (sbi->ll_flags & LL_SBI_LAZYSTATFS)
2639                 seq_puts(seq, ",lazystatfs");
2640
2641         if (sbi->ll_flags & LL_SBI_USER_FID2PATH)
2642                 seq_puts(seq, ",user_fid2path");
2643
2644         RETURN(0);
2645 }
2646
2647 /**
2648  * Get obd name by cmd, and copy out to user space
2649  */
2650 int ll_get_obd_name(struct inode *inode, unsigned int cmd, unsigned long arg)
2651 {
2652         struct ll_sb_info *sbi = ll_i2sbi(inode);
2653         struct obd_device *obd;
2654         ENTRY;
2655
2656         if (cmd == OBD_IOC_GETDTNAME)
2657                 obd = class_exp2obd(sbi->ll_dt_exp);
2658         else if (cmd == OBD_IOC_GETMDNAME)
2659                 obd = class_exp2obd(sbi->ll_md_exp);
2660         else
2661                 RETURN(-EINVAL);
2662
2663         if (!obd)
2664                 RETURN(-ENOENT);
2665
2666         if (copy_to_user((void *)arg, obd->obd_name,
2667                          strlen(obd->obd_name) + 1))
2668                 RETURN(-EFAULT);
2669
2670         RETURN(0);
2671 }
2672
2673 /**
2674  * Get lustre file system name by \a sbi. If \a buf is provided(non-NULL), the
2675  * fsname will be returned in this buffer; otherwise, a static buffer will be
2676  * used to store the fsname and returned to caller.
2677  */
2678 char *ll_get_fsname(struct super_block *sb, char *buf, int buflen)
2679 {
2680         static char fsname_static[MTI_NAME_MAXLEN];
2681         struct lustre_sb_info *lsi = s2lsi(sb);
2682         char *ptr;
2683         int len;
2684
2685         if (buf == NULL) {
2686                 /* this means the caller wants to use static buffer
2687                  * and it doesn't care about race. Usually this is
2688                  * in error reporting path */
2689                 buf = fsname_static;
2690                 buflen = sizeof(fsname_static);
2691         }
2692
2693         len = strlen(lsi->lsi_lmd->lmd_profile);
2694         ptr = strrchr(lsi->lsi_lmd->lmd_profile, '-');
2695         if (ptr && (strcmp(ptr, "-client") == 0))
2696                 len -= 7;
2697
2698         if (unlikely(len >= buflen))
2699                 len = buflen - 1;
2700         strncpy(buf, lsi->lsi_lmd->lmd_profile, len);
2701         buf[len] = '\0';
2702
2703         return buf;
2704 }
2705
2706 static char* ll_d_path(struct dentry *dentry, char *buf, int bufsize)
2707 {
2708         char *path = NULL;
2709
2710         struct path p;
2711
2712         p.dentry = dentry;
2713         p.mnt = current->fs->root.mnt;
2714         path_get(&p);
2715         path = d_path(&p, buf, bufsize);
2716         path_put(&p);
2717         return path;
2718 }
2719
2720 void ll_dirty_page_discard_warn(struct page *page, int ioret)
2721 {
2722         char *buf, *path = NULL;
2723         struct dentry *dentry = NULL;
2724         struct ccc_object *obj = cl_inode2ccc(page->mapping->host);
2725
2726         /* this can be called inside spin lock so use GFP_ATOMIC. */
2727         buf = (char *)__get_free_page(GFP_ATOMIC);
2728         if (buf != NULL) {
2729                 dentry = d_find_alias(page->mapping->host);
2730                 if (dentry != NULL)
2731                         path = ll_d_path(dentry, buf, PAGE_SIZE);
2732         }
2733
2734         CDEBUG(D_WARNING,
2735                "%s: dirty page discard: %s/fid: "DFID"/%s may get corrupted "
2736                "(rc %d)\n", ll_get_fsname(page->mapping->host->i_sb, NULL, 0),
2737                s2lsi(page->mapping->host->i_sb)->lsi_lmd->lmd_dev,
2738                PFID(&obj->cob_header.coh_lu.loh_fid),
2739                (path && !IS_ERR(path)) ? path : "", ioret);
2740
2741         if (dentry != NULL)
2742                 dput(dentry);
2743
2744         if (buf != NULL)
2745                 free_page((unsigned long)buf);
2746 }
2747
2748 /*
2749  * Compute llite root squash state after a change of root squash
2750  * configuration setting or add/remove of a lnet nid
2751  */
2752 void ll_compute_rootsquash_state(struct ll_sb_info *sbi)
2753 {
2754         struct root_squash_info *squash = &sbi->ll_squash;
2755         int i;
2756         bool matched;
2757         lnet_process_id_t id;
2758
2759         /* Update norootsquash flag */
2760         down_write(&squash->rsi_sem);
2761         if (list_empty(&squash->rsi_nosquash_nids))
2762                 sbi->ll_flags &= ~LL_SBI_NOROOTSQUASH;
2763         else {
2764                 /* Do not apply root squash as soon as one of our NIDs is
2765                  * in the nosquash_nids list */
2766                 matched = false;
2767                 i = 0;
2768                 while (LNetGetId(i++, &id) != -ENOENT) {
2769                         if (LNET_NETTYP(LNET_NIDNET(id.nid)) == LOLND)
2770                                 continue;
2771                         if (cfs_match_nid(id.nid, &squash->rsi_nosquash_nids)) {
2772                                 matched = true;
2773                                 break;
2774                         }
2775                 }
2776                 if (matched)
2777                         sbi->ll_flags |= LL_SBI_NOROOTSQUASH;
2778                 else
2779                         sbi->ll_flags &= ~LL_SBI_NOROOTSQUASH;
2780         }
2781         up_write(&squash->rsi_sem);
2782 }
2783
2784