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, 2015, 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/statfs.h>
  45 #include <linux/time.h>
  46 #include <linux/types.h>
  47 #include <linux/version.h>
  48 #include <linux/mm.h>
  49 #include <linux/user_namespace.h>
  50 #ifdef HAVE_UIDGID_HEADER
  51 # include <linux/uidgid.h>
  52 #endif
  53
  54 #include <lustre_ioctl.h>
  55 #include <lustre_ha.h>
  56 #include <lustre_dlm.h>
  57 #include <lprocfs_status.h>
  58 #include <lustre_disk.h>
  59 #include <lustre_param.h>
  60 #include <lustre_log.h>
  61 #include <cl_object.h>
  62 #include <obd_cksum.h>
  63 #include "llite_internal.h"
  64
  65 struct kmem_cache *ll_file_data_slab;
  66
  67 #ifndef log2
  68 #define log2(n) ffz(~(n))
  69 #endif
  70
  71 static struct ll_sb_info *ll_init_sbi(void)
  72 {
  73         struct ll_sb_info *sbi = NULL;
  74         unsigned long pages;
  75         unsigned long lru_page_max;
  76         struct sysinfo si;
  77         class_uuid_t uuid;
  78         int i;
  79         ENTRY;
  80
  81         OBD_ALLOC_PTR(sbi);
  82         if (sbi == NULL)
  83                 RETURN(NULL);
  84
  85         spin_lock_init(&sbi->ll_lock);
  86         mutex_init(&sbi->ll_lco.lco_lock);
  87         spin_lock_init(&sbi->ll_pp_extent_lock);
  88         spin_lock_init(&sbi->ll_process_lock);
  89         sbi->ll_rw_stats_on = 0;
  90
  91         si_meminfo(&si);
  92         pages = si.totalram - si.totalhigh;
  93         lru_page_max = pages / 2;
  94
  95         /* initialize ll_cache data */
  96         sbi->ll_cache = cl_cache_init(lru_page_max);
  97         if (sbi->ll_cache == NULL) {
  98                 OBD_FREE(sbi, sizeof(*sbi));
  99                 RETURN(NULL);
 100         }
 101
 102         sbi->ll_ra_info.ra_max_pages_per_file = min(pages / 32,
 103                                            SBI_DEFAULT_READAHEAD_MAX);
 104         sbi->ll_ra_info.ra_max_pages = sbi->ll_ra_info.ra_max_pages_per_file;
 105         sbi->ll_ra_info.ra_max_read_ahead_whole_pages =
 106                                            SBI_DEFAULT_READAHEAD_WHOLE_MAX;
 107
 108         ll_generate_random_uuid(uuid);
 109         class_uuid_unparse(uuid, &sbi->ll_sb_uuid);
 110         CDEBUG(D_CONFIG, "generated uuid: %s\n", sbi->ll_sb_uuid.uuid);
 111
 112         sbi->ll_flags |= LL_SBI_VERBOSE;
 113 #ifdef ENABLE_CHECKSUM
 114         sbi->ll_flags |= LL_SBI_CHECKSUM;
 115 #endif
 116
 117 #ifdef HAVE_LRU_RESIZE_SUPPORT
 118         sbi->ll_flags |= LL_SBI_LRU_RESIZE;
 119 #endif
 120         sbi->ll_flags |= LL_SBI_LAZYSTATFS;
 121
 122         for (i = 0; i <= LL_PROCESS_HIST_MAX; i++) {
 123                 spin_lock_init(&sbi->ll_rw_extents_info.pp_extents[i].
 124                                pp_r_hist.oh_lock);
 125                 spin_lock_init(&sbi->ll_rw_extents_info.pp_extents[i].
 126                                pp_w_hist.oh_lock);
 127         }
 128
 129         /* metadata statahead is enabled by default */
 130         sbi->ll_sa_max = LL_SA_RPC_DEF;
 131         atomic_set(&sbi->ll_sa_total, 0);
 132         atomic_set(&sbi->ll_sa_wrong, 0);
 133         atomic_set(&sbi->ll_sa_running, 0);
 134         atomic_set(&sbi->ll_agl_total, 0);
 135         sbi->ll_flags |= LL_SBI_AGL_ENABLED;
 136         sbi->ll_flags |= LL_SBI_FAST_READ;
 137
 138         /* root squash */
 139         sbi->ll_squash.rsi_uid = 0;
 140         sbi->ll_squash.rsi_gid = 0;
 141         INIT_LIST_HEAD(&sbi->ll_squash.rsi_nosquash_nids);
 142         init_rwsem(&sbi->ll_squash.rsi_sem);
 143
 144         RETURN(sbi);
 145 }
 146
 147 static void ll_free_sbi(struct super_block *sb)
 148 {
 149         struct ll_sb_info *sbi = ll_s2sbi(sb);
 150         ENTRY;
 151
 152         if (sbi != NULL) {
 153                 if (!list_empty(&sbi->ll_squash.rsi_nosquash_nids))
 154                         cfs_free_nidlist(&sbi->ll_squash.rsi_nosquash_nids);
 155                 if (sbi->ll_cache != NULL) {
 156                         cl_cache_decref(sbi->ll_cache);
 157                         sbi->ll_cache = NULL;
 158                 }
 159                 OBD_FREE(sbi, sizeof(*sbi));
 160         }
 161         EXIT;
 162 }
 163
 164 static int client_common_fill_super(struct super_block *sb, char *md, char *dt,
 165                                     struct vfsmount *mnt)
 166 {
 167         struct inode *root = NULL;
 168         struct ll_sb_info *sbi = ll_s2sbi(sb);
 169         struct obd_device *obd;
 170         struct obd_statfs *osfs = NULL;
 171         struct ptlrpc_request *request = NULL;
 172         struct obd_connect_data *data = NULL;
 173         struct obd_uuid *uuid;
 174         struct md_op_data *op_data;
 175         struct lustre_md lmd;
 176         u64 valid;
 177         int size, err, checksum;
 178         ENTRY;
 179
 180         obd = class_name2obd(md);
 181         if (!obd) {
 182                 CERROR("MD %s: not setup or attached\n", md);
 183                 RETURN(-EINVAL);
 184         }
 185
 186         OBD_ALLOC_PTR(data);
 187         if (data == NULL)
 188                 RETURN(-ENOMEM);
 189
 190         OBD_ALLOC_PTR(osfs);
 191         if (osfs == NULL) {
 192                 OBD_FREE_PTR(data);
 193                 RETURN(-ENOMEM);
 194         }
 195
 196         /* indicate the features supported by this client */
 197         data->ocd_connect_flags = OBD_CONNECT_IBITS    | OBD_CONNECT_NODEVOH  |
 198                                   OBD_CONNECT_ATTRFID  |
 199                                   OBD_CONNECT_VERSION  | OBD_CONNECT_BRW_SIZE |
 200                                   OBD_CONNECT_MDS_CAPA | OBD_CONNECT_OSS_CAPA |
 201                                   OBD_CONNECT_CANCELSET | OBD_CONNECT_FID     |
 202                                   OBD_CONNECT_AT       | OBD_CONNECT_LOV_V3   |
 203                                   OBD_CONNECT_VBR | OBD_CONNECT_FULL20 |
 204                                   OBD_CONNECT_64BITHASH |
 205                                   OBD_CONNECT_EINPROGRESS |
 206                                   OBD_CONNECT_JOBSTATS | OBD_CONNECT_LVB_TYPE |
 207                                   OBD_CONNECT_LAYOUTLOCK | OBD_CONNECT_PINGLESS|
 208                                   OBD_CONNECT_MAX_EASIZE |
 209                                   OBD_CONNECT_FLOCK_DEAD |
 210                                   OBD_CONNECT_DISP_STRIPE | OBD_CONNECT_LFSCK |
 211                                   OBD_CONNECT_OPEN_BY_FID |
 212                                   OBD_CONNECT_DIR_STRIPE |
 213                                   OBD_CONNECT_BULK_MBITS |
 214                                   OBD_CONNECT_SUBTREE |
 215                                   OBD_CONNECT_FLAGS2;
 216
 217         data->ocd_connect_flags2 = 0;
 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         if (sbi->ll_flags & LL_SBI_FLOCK)
 241                 sbi->ll_fop = &ll_file_operations_flock;
 242         else if (sbi->ll_flags & LL_SBI_LOCALFLOCK)
 243                 sbi->ll_fop = &ll_file_operations;
 244         else
 245                 sbi->ll_fop = &ll_file_operations_noflock;
 246
 247         /* real client */
 248         data->ocd_connect_flags |= OBD_CONNECT_REAL;
 249
 250         /* always ping even if server suppress_pings */
 251         if (sbi->ll_flags & LL_SBI_ALWAYS_PING)
 252                 data->ocd_connect_flags &= ~OBD_CONNECT_PINGLESS;
 253
 254         data->ocd_brw_size = MD_MAX_BRW_SIZE;
 255
 256         err = obd_connect(NULL, &sbi->ll_md_exp, obd, &sbi->ll_sb_uuid, data, NULL);
 257         if (err == -EBUSY) {
 258                 LCONSOLE_ERROR_MSG(0x14f, "An MDT (md %s) is performing "
 259                                    "recovery, of which this client is not a "
 260                                    "part. Please wait for recovery to complete,"
 261                                    " abort, or time out.\n", md);
 262                 GOTO(out, err);
 263         } else if (err) {
 264                 CERROR("cannot connect to %s: rc = %d\n", md, err);
 265                 GOTO(out, err);
 266         }
 267
 268         sbi->ll_md_exp->exp_connect_data = *data;
 269
 270         err = obd_fid_init(sbi->ll_md_exp->exp_obd, sbi->ll_md_exp,
 271                            LUSTRE_SEQ_METADATA);
 272         if (err) {
 273                 CERROR("%s: Can't init metadata layer FID infrastructure, "
 274                        "rc = %d\n", sbi->ll_md_exp->exp_obd->obd_name, err);
 275                 GOTO(out_md, err);
 276         }
 277
 278         /* For mount, we only need fs info from MDT0, and also in DNE, it
 279          * can make sure the client can be mounted as long as MDT0 is
 280          * avaible */
 281         err = obd_statfs(NULL, sbi->ll_md_exp, osfs,
 282                         cfs_time_shift_64(-OBD_STATFS_CACHE_SECONDS),
 283                         OBD_STATFS_FOR_MDT0);
 284         if (err)
 285                 GOTO(out_md_fid, err);
 286
 287         /* This needs to be after statfs to ensure connect has finished.
 288          * Note that "data" does NOT contain the valid connect reply.
 289          * If connecting to a 1.8 server there will be no LMV device, so
 290          * we can access the MDC export directly and exp_connect_flags will
 291          * be non-zero, but if accessing an upgraded 2.1 server it will
 292          * have the correct flags filled in.
 293          * XXX: fill in the LMV exp_connect_flags from MDC(s). */
 294         valid = exp_connect_flags(sbi->ll_md_exp) & CLIENT_CONNECT_MDT_REQD;
 295         if (exp_connect_flags(sbi->ll_md_exp) != 0 &&
 296             valid != CLIENT_CONNECT_MDT_REQD) {
 297                 char *buf;
 298
 299                 OBD_ALLOC_WAIT(buf, PAGE_CACHE_SIZE);
 300                 obd_connect_flags2str(buf, PAGE_CACHE_SIZE,
 301                                       valid ^ CLIENT_CONNECT_MDT_REQD, 0, ",");
 302                 LCONSOLE_ERROR_MSG(0x170, "Server %s does not support "
 303                                    "feature(s) needed for correct operation "
 304                                    "of this client (%s). Please upgrade "
 305                                    "server or downgrade client.\n",
 306                                    sbi->ll_md_exp->exp_obd->obd_name, buf);
 307                 OBD_FREE(buf, PAGE_CACHE_SIZE);
 308                 GOTO(out_md_fid, err = -EPROTO);
 309         }
 310
 311         size = sizeof(*data);
 312         err = obd_get_info(NULL, sbi->ll_md_exp, sizeof(KEY_CONN_DATA),
 313                            KEY_CONN_DATA,  &size, data);
 314         if (err) {
 315                 CERROR("%s: Get connect data failed: rc = %d\n",
 316                        sbi->ll_md_exp->exp_obd->obd_name, err);
 317                 GOTO(out_md_fid, err);
 318         }
 319
 320         LASSERT(osfs->os_bsize);
 321         sb->s_blocksize = osfs->os_bsize;
 322         sb->s_blocksize_bits = log2(osfs->os_bsize);
 323         sb->s_magic = LL_SUPER_MAGIC;
 324         sb->s_maxbytes = MAX_LFS_FILESIZE;
 325         sbi->ll_namelen = osfs->os_namelen;
 326
 327         if ((sbi->ll_flags & LL_SBI_USER_XATTR) &&
 328             !(data->ocd_connect_flags & OBD_CONNECT_XATTR)) {
 329                 LCONSOLE_INFO("Disabling user_xattr feature because "
 330                               "it is not supported on the server\n");
 331                 sbi->ll_flags &= ~LL_SBI_USER_XATTR;
 332         }
 333
 334         if (data->ocd_connect_flags & OBD_CONNECT_ACL) {
 335 #ifdef MS_POSIXACL
 336                 sb->s_flags |= MS_POSIXACL;
 337 #endif
 338                 sbi->ll_flags |= LL_SBI_ACL;
 339         } else {
 340                 LCONSOLE_INFO("client wants to enable acl, but mdt not!\n");
 341 #ifdef MS_POSIXACL
 342                 sb->s_flags &= ~MS_POSIXACL;
 343 #endif
 344                 sbi->ll_flags &= ~LL_SBI_ACL;
 345         }
 346
 347         if (data->ocd_connect_flags & OBD_CONNECT_64BITHASH)
 348                 sbi->ll_flags |= LL_SBI_64BIT_HASH;
 349
 350         if (data->ocd_connect_flags & OBD_CONNECT_BRW_SIZE)
 351                 sbi->ll_md_brw_pages = data->ocd_brw_size >> PAGE_CACHE_SHIFT;
 352         else
 353                 sbi->ll_md_brw_pages = 1;
 354
 355         if (data->ocd_connect_flags & OBD_CONNECT_LAYOUTLOCK)
 356                 sbi->ll_flags |= LL_SBI_LAYOUT_LOCK;
 357
 358         if (data->ocd_ibits_known & MDS_INODELOCK_XATTR) {
 359                 if (!(data->ocd_connect_flags & OBD_CONNECT_MAX_EASIZE)) {
 360                         LCONSOLE_INFO("%s: disabling xattr cache due to "
 361                                       "unknown maximum xattr size.\n", dt);
 362                 } else {
 363                         sbi->ll_flags |= LL_SBI_XATTR_CACHE;
 364                         sbi->ll_xattr_cache_enabled = 1;
 365                 }
 366         }
 367
 368         obd = class_name2obd(dt);
 369         if (!obd) {
 370                 CERROR("DT %s: not setup or attached\n", dt);
 371                 GOTO(out_md_fid, err = -ENODEV);
 372         }
 373
 374         /* pass client page size via ocd_grant_blkbits, the server should report
 375          * back its backend blocksize for grant calculation purpose */
 376         data->ocd_grant_blkbits = PAGE_SHIFT;
 377
 378         data->ocd_connect_flags = OBD_CONNECT_GRANT | OBD_CONNECT_VERSION |
 379                                   OBD_CONNECT_REQPORTAL | OBD_CONNECT_BRW_SIZE |
 380                                   OBD_CONNECT_CANCELSET | OBD_CONNECT_FID |
 381                                   OBD_CONNECT_SRVLOCK | OBD_CONNECT_TRUNCLOCK|
 382                                   OBD_CONNECT_AT | OBD_CONNECT_OSS_CAPA |
 383                                   OBD_CONNECT_VBR | OBD_CONNECT_FULL20 |
 384                                   OBD_CONNECT_64BITHASH | OBD_CONNECT_MAXBYTES |
 385                                   OBD_CONNECT_EINPROGRESS |
 386                                   OBD_CONNECT_JOBSTATS | OBD_CONNECT_LVB_TYPE |
 387                                   OBD_CONNECT_LAYOUTLOCK |
 388                                   OBD_CONNECT_PINGLESS | OBD_CONNECT_LFSCK |
 389                                   OBD_CONNECT_BULK_MBITS;
 390
 391         data->ocd_connect_flags2 = 0;
 392
 393         if (!OBD_FAIL_CHECK(OBD_FAIL_OSC_CONNECT_GRANT_PARAM))
 394                 data->ocd_connect_flags |= OBD_CONNECT_GRANT_PARAM;
 395
 396         if (!OBD_FAIL_CHECK(OBD_FAIL_OSC_CONNECT_CKSUM)) {
 397                 /* OBD_CONNECT_CKSUM should always be set, even if checksums are
 398                  * disabled by default, because it can still be enabled on the
 399                  * fly via /proc. As a consequence, we still need to come to an
 400                  * agreement on the supported algorithms at connect time */
 401                 data->ocd_connect_flags |= OBD_CONNECT_CKSUM;
 402
 403                 if (OBD_FAIL_CHECK(OBD_FAIL_OSC_CKSUM_ADLER_ONLY))
 404                         data->ocd_cksum_types = OBD_CKSUM_ADLER;
 405                 else
 406                         data->ocd_cksum_types = cksum_types_supported_client();
 407         }
 408
 409 #ifdef HAVE_LRU_RESIZE_SUPPORT
 410         data->ocd_connect_flags |= OBD_CONNECT_LRU_RESIZE;
 411 #endif
 412         /* always ping even if server suppress_pings */
 413         if (sbi->ll_flags & LL_SBI_ALWAYS_PING)
 414                 data->ocd_connect_flags &= ~OBD_CONNECT_PINGLESS;
 415
 416         CDEBUG(D_RPCTRACE, "ocd_connect_flags: "LPX64" ocd_version: %d "
 417                "ocd_grant: %d\n", data->ocd_connect_flags,
 418                data->ocd_version, data->ocd_grant);
 419
 420         obd->obd_upcall.onu_owner = &sbi->ll_lco;
 421         obd->obd_upcall.onu_upcall = cl_ocd_update;
 422
 423         data->ocd_brw_size = DT_MAX_BRW_SIZE;
 424
 425         err = obd_connect(NULL, &sbi->ll_dt_exp, obd, &sbi->ll_sb_uuid, data,
 426                           NULL);
 427         if (err == -EBUSY) {
 428                 LCONSOLE_ERROR_MSG(0x150, "An OST (dt %s) is performing "
 429                                    "recovery, of which this client is not a "
 430                                    "part.  Please wait for recovery to "
 431                                    "complete, abort, or time out.\n", dt);
 432                 GOTO(out_md, err);
 433         } else if (err) {
 434                 CERROR("%s: Cannot connect to %s: rc = %d\n",
 435                        sbi->ll_dt_exp->exp_obd->obd_name, dt, err);
 436                 GOTO(out_md, err);
 437         }
 438
 439         sbi->ll_dt_exp->exp_connect_data = *data;
 440
 441         err = obd_fid_init(sbi->ll_dt_exp->exp_obd, sbi->ll_dt_exp,
 442                            LUSTRE_SEQ_METADATA);
 443         if (err) {
 444                 CERROR("%s: Can't init data layer FID infrastructure, "
 445                        "rc = %d\n", sbi->ll_dt_exp->exp_obd->obd_name, err);
 446                 GOTO(out_dt, err);
 447         }
 448
 449         mutex_lock(&sbi->ll_lco.lco_lock);
 450         sbi->ll_lco.lco_flags = data->ocd_connect_flags;
 451         sbi->ll_lco.lco_md_exp = sbi->ll_md_exp;
 452         sbi->ll_lco.lco_dt_exp = sbi->ll_dt_exp;
 453         mutex_unlock(&sbi->ll_lco.lco_lock);
 454
 455         fid_zero(&sbi->ll_root_fid);
 456         err = md_get_root(sbi->ll_md_exp, get_mount_fileset(sb),
 457                            &sbi->ll_root_fid);
 458         if (err) {
 459                 CERROR("cannot mds_connect: rc = %d\n", err);
 460                 GOTO(out_lock_cn_cb, err);
 461         }
 462         if (!fid_is_sane(&sbi->ll_root_fid)) {
 463                 CERROR("%s: Invalid root fid "DFID" during mount\n",
 464                        sbi->ll_md_exp->exp_obd->obd_name,
 465                        PFID(&sbi->ll_root_fid));
 466                 GOTO(out_lock_cn_cb, err = -EINVAL);
 467         }
 468         CDEBUG(D_SUPER, "rootfid "DFID"\n", PFID(&sbi->ll_root_fid));
 469
 470         sb->s_op = &lustre_super_operations;
 471 #if THREAD_SIZE >= 8192 /*b=17630*/
 472         sb->s_export_op = &lustre_export_operations;
 473 #endif
 474
 475         /* make root inode
 476          * XXX: move this to after cbd setup? */
 477         valid = OBD_MD_FLGETATTR | OBD_MD_FLBLOCKS | OBD_MD_FLMODEASIZE;
 478         if (sbi->ll_flags & LL_SBI_ACL)
 479                 valid |= OBD_MD_FLACL;
 480
 481         OBD_ALLOC_PTR(op_data);
 482         if (op_data == NULL)
 483                 GOTO(out_lock_cn_cb, err = -ENOMEM);
 484
 485         op_data->op_fid1 = sbi->ll_root_fid;
 486         op_data->op_mode = 0;
 487         op_data->op_valid = valid;
 488
 489         err = md_getattr(sbi->ll_md_exp, op_data, &request);
 490
 491         OBD_FREE_PTR(op_data);
 492         if (err) {
 493                 CERROR("%s: md_getattr failed for root: rc = %d\n",
 494                        sbi->ll_md_exp->exp_obd->obd_name, err);
 495                 GOTO(out_lock_cn_cb, err);
 496         }
 497
 498         err = md_get_lustre_md(sbi->ll_md_exp, request, sbi->ll_dt_exp,
 499                                sbi->ll_md_exp, &lmd);
 500         if (err) {
 501                 CERROR("failed to understand root inode md: rc = %d\n", err);
 502                 ptlrpc_req_finished(request);
 503                 GOTO(out_lock_cn_cb, err);
 504         }
 505
 506         LASSERT(fid_is_sane(&sbi->ll_root_fid));
 507         root = ll_iget(sb, cl_fid_build_ino(&sbi->ll_root_fid,
 508                                             sbi->ll_flags & LL_SBI_32BIT_API),
 509                        &lmd);
 510         md_free_lustre_md(sbi->ll_md_exp, &lmd);
 511         ptlrpc_req_finished(request);
 512
 513         if (IS_ERR(root)) {
 514 #ifdef CONFIG_FS_POSIX_ACL
 515                 if (lmd.posix_acl) {
 516                         posix_acl_release(lmd.posix_acl);
 517                         lmd.posix_acl = NULL;
 518                 }
 519 #endif
 520                 err = IS_ERR(root) ? PTR_ERR(root) : -EBADF;
 521                 root = NULL;
 522                 CERROR("lustre_lite: bad iget4 for root\n");
 523                 GOTO(out_root, err);
 524         }
 525
 526         checksum = sbi->ll_flags & LL_SBI_CHECKSUM;
 527         err = obd_set_info_async(NULL, sbi->ll_dt_exp, sizeof(KEY_CHECKSUM),
 528                                  KEY_CHECKSUM, sizeof(checksum), &checksum,
 529                                  NULL);
 530         cl_sb_init(sb);
 531
 532         err = obd_set_info_async(NULL, sbi->ll_dt_exp, sizeof(KEY_CACHE_SET),
 533                                  KEY_CACHE_SET, sizeof(*sbi->ll_cache),
 534                                  sbi->ll_cache, NULL);
 535
 536         sb->s_root = d_make_root(root);
 537         if (sb->s_root == NULL) {
 538                 CERROR("%s: can't make root dentry\n",
 539                         ll_get_fsname(sb, NULL, 0));
 540                 GOTO(out_root, err = -ENOMEM);
 541         }
 542 #ifdef HAVE_DCACHE_LOCK
 543         sb->s_root->d_op = &ll_d_ops;
 544 #endif
 545
 546         sbi->ll_sdev_orig = sb->s_dev;
 547
 548         /* We set sb->s_dev equal on all lustre clients in order to support
 549          * NFS export clustering.  NFSD requires that the FSID be the same
 550          * on all clients. */
 551         /* s_dev is also used in lt_compare() to compare two fs, but that is
 552          * only a node-local comparison. */
 553         uuid = obd_get_uuid(sbi->ll_md_exp);
 554         if (uuid != NULL)
 555                 sb->s_dev = get_uuid2int(uuid->uuid, strlen(uuid->uuid));
 556
 557         if (data != NULL)
 558                 OBD_FREE_PTR(data);
 559         if (osfs != NULL)
 560                 OBD_FREE_PTR(osfs);
 561         if (proc_lustre_fs_root != NULL) {
 562                 err = lprocfs_register_mountpoint(proc_lustre_fs_root, sb,
 563                                                   dt, md);
 564                 if (err < 0) {
 565                         CERROR("%s: could not register mount in lprocfs: "
 566                                "rc = %d\n", ll_get_fsname(sb, NULL, 0), err);
 567                         err = 0;
 568                 }
 569         }
 570
 571         RETURN(err);
 572 out_root:
 573         if (root)
 574                 iput(root);
 575 out_lock_cn_cb:
 576         obd_fid_fini(sbi->ll_dt_exp->exp_obd);
 577 out_dt:
 578         obd_disconnect(sbi->ll_dt_exp);
 579         sbi->ll_dt_exp = NULL;
 580 out_md_fid:
 581         obd_fid_fini(sbi->ll_md_exp->exp_obd);
 582 out_md:
 583         obd_disconnect(sbi->ll_md_exp);
 584         sbi->ll_md_exp = NULL;
 585 out:
 586         if (data != NULL)
 587                 OBD_FREE_PTR(data);
 588         if (osfs != NULL)
 589                 OBD_FREE_PTR(osfs);
 590         return err;
 591 }
 592
 593 int ll_get_max_mdsize(struct ll_sb_info *sbi, int *lmmsize)
 594 {
 595         int size, rc;
 596
 597         size = sizeof(*lmmsize);
 598         rc = obd_get_info(NULL, sbi->ll_dt_exp, sizeof(KEY_MAX_EASIZE),
 599                           KEY_MAX_EASIZE, &size, lmmsize);
 600         if (rc != 0) {
 601                 CERROR("%s: cannot get max LOV EA size: rc = %d\n",
 602                        sbi->ll_dt_exp->exp_obd->obd_name, rc);
 603                 RETURN(rc);
 604         }
 605
 606         size = sizeof(int);
 607         rc = obd_get_info(NULL, sbi->ll_md_exp, sizeof(KEY_MAX_EASIZE),
 608                           KEY_MAX_EASIZE, &size, lmmsize);
 609         if (rc)
 610                 CERROR("Get max mdsize error rc %d\n", rc);
 611
 612         RETURN(rc);
 613 }
 614
 615 /**
 616  * Get the value of the default_easize parameter.
 617  *
 618  * \see client_obd::cl_default_mds_easize
 619  *
 620  * \param[in] sbi       superblock info for this filesystem
 621  * \param[out] lmmsize  pointer to storage location for value
 622  *
 623  * \retval 0            on success
 624  * \retval negative     negated errno on failure
 625  */
 626 int ll_get_default_mdsize(struct ll_sb_info *sbi, int *lmmsize)
 627 {
 628         int size, rc;
 629
 630         size = sizeof(int);
 631         rc = obd_get_info(NULL, sbi->ll_md_exp, sizeof(KEY_DEFAULT_EASIZE),
 632                          KEY_DEFAULT_EASIZE, &size, lmmsize);
 633         if (rc)
 634                 CERROR("Get default mdsize error rc %d\n", rc);
 635
 636         RETURN(rc);
 637 }
 638
 639 /**
 640  * Set the default_easize parameter to the given value.
 641  *
 642  * \see client_obd::cl_default_mds_easize
 643  *
 644  * \param[in] sbi       superblock info for this filesystem
 645  * \param[in] lmmsize   the size to set
 646  *
 647  * \retval 0            on success
 648  * \retval negative     negated errno on failure
 649  */
 650 int ll_set_default_mdsize(struct ll_sb_info *sbi, int lmmsize)
 651 {
 652         int rc;
 653
 654         if (lmmsize < sizeof(struct lov_mds_md) ||
 655             lmmsize > OBD_MAX_DEFAULT_EA_SIZE)
 656                 return -EINVAL;
 657
 658         rc = obd_set_info_async(NULL, sbi->ll_md_exp,
 659                                 sizeof(KEY_DEFAULT_EASIZE), KEY_DEFAULT_EASIZE,
 660                                 sizeof(int), &lmmsize, NULL);
 661
 662         RETURN(rc);
 663 }
 664
 665 static void ll_dump_inode(struct inode *inode)
 666 {
 667         struct ll_d_hlist_node *tmp;
 668         int dentry_count = 0;
 669
 670         LASSERT(inode != NULL);
 671
 672         ll_d_hlist_for_each(tmp, &inode->i_dentry)
 673                 dentry_count++;
 674
 675         CERROR("%s: inode %p dump: dev=%s fid="DFID
 676                " mode=%o count=%u, %d dentries\n",
 677                ll_get_fsname(inode->i_sb, NULL, 0), inode,
 678                ll_i2mdexp(inode)->exp_obd->obd_name, PFID(ll_inode2fid(inode)),
 679                inode->i_mode, atomic_read(&inode->i_count), dentry_count);
 680 }
 681
 682 void lustre_dump_dentry(struct dentry *dentry, int recur)
 683 {
 684         struct list_head *tmp;
 685         int subdirs = 0;
 686
 687         LASSERT(dentry != NULL);
 688
 689         list_for_each(tmp, &dentry->d_subdirs)
 690                 subdirs++;
 691
 692         CERROR("dentry %p dump: name=%.*s parent=%.*s (%p), inode=%p, count=%u,"
 693                " flags=0x%x, fsdata=%p, %d subdirs\n", dentry,
 694                dentry->d_name.len, dentry->d_name.name,
 695                dentry->d_parent->d_name.len, dentry->d_parent->d_name.name,
 696                dentry->d_parent, dentry->d_inode, ll_d_count(dentry),
 697                dentry->d_flags, dentry->d_fsdata, subdirs);
 698         if (dentry->d_inode != NULL)
 699                 ll_dump_inode(dentry->d_inode);
 700
 701         if (recur == 0)
 702                 return;
 703
 704         list_for_each(tmp, &dentry->d_subdirs) {
 705                 struct dentry *d = list_entry(tmp, struct dentry, d_child);
 706                 lustre_dump_dentry(d, recur - 1);
 707         }
 708 }
 709
 710 static void client_common_put_super(struct super_block *sb)
 711 {
 712         struct ll_sb_info *sbi = ll_s2sbi(sb);
 713         ENTRY;
 714
 715         cl_sb_fini(sb);
 716
 717         obd_fid_fini(sbi->ll_dt_exp->exp_obd);
 718         obd_disconnect(sbi->ll_dt_exp);
 719         sbi->ll_dt_exp = NULL;
 720
 721         lprocfs_unregister_mountpoint(sbi);
 722
 723         obd_fid_fini(sbi->ll_md_exp->exp_obd);
 724         obd_disconnect(sbi->ll_md_exp);
 725         sbi->ll_md_exp = NULL;
 726
 727         EXIT;
 728 }
 729
 730 void ll_kill_super(struct super_block *sb)
 731 {
 732         struct ll_sb_info *sbi;
 733         ENTRY;
 734
 735         /* not init sb ?*/
 736         if (!(sb->s_flags & MS_ACTIVE))
 737                 return;
 738
 739         sbi = ll_s2sbi(sb);
 740         /* we need restore s_dev from changed for clustred NFS before put_super
 741          * because new kernels have cached s_dev and change sb->s_dev in
 742          * put_super not affected real removing devices */
 743         if (sbi) {
 744                 sb->s_dev = sbi->ll_sdev_orig;
 745                 sbi->ll_umounting = 1;
 746
 747                 /* wait running statahead threads to quit */
 748                 while (atomic_read(&sbi->ll_sa_running) > 0) {
 749                         set_current_state(TASK_UNINTERRUPTIBLE);
 750                         schedule_timeout(msecs_to_jiffies(MSEC_PER_SEC >> 3));
 751                 }
 752         }
 753
 754         EXIT;
 755 }
 756
 757 static inline int ll_set_opt(const char *opt, char *data, int fl)
 758 {
 759         if (strncmp(opt, data, strlen(opt)) != 0)
 760                 return(0);
 761         else
 762                 return(fl);
 763 }
 764
 765 /* non-client-specific mount options are parsed in lmd_parse */
 766 static int ll_options(char *options, int *flags)
 767 {
 768         int tmp;
 769         char *s1 = options, *s2;
 770         ENTRY;
 771
 772         if (!options)
 773                 RETURN(0);
 774
 775         CDEBUG(D_CONFIG, "Parsing opts %s\n", options);
 776
 777         while (*s1) {
 778                 CDEBUG(D_SUPER, "next opt=%s\n", s1);
 779                 tmp = ll_set_opt("nolock", s1, LL_SBI_NOLCK);
 780                 if (tmp) {
 781                         *flags |= tmp;
 782                         goto next;
 783                 }
 784                 tmp = ll_set_opt("flock", s1, LL_SBI_FLOCK);
 785                 if (tmp) {
 786                         *flags |= tmp;
 787                         goto next;
 788                 }
 789                 tmp = ll_set_opt("localflock", s1, LL_SBI_LOCALFLOCK);
 790                 if (tmp) {
 791                         *flags |= tmp;
 792                         goto next;
 793                 }
 794                 tmp = ll_set_opt("noflock", s1, LL_SBI_FLOCK|LL_SBI_LOCALFLOCK);
 795                 if (tmp) {
 796                         *flags &= ~tmp;
 797                         goto next;
 798                 }
 799                 tmp = ll_set_opt("user_xattr", s1, LL_SBI_USER_XATTR);
 800                 if (tmp) {
 801                         *flags |= tmp;
 802                         goto next;
 803                 }
 804                 tmp = ll_set_opt("nouser_xattr", s1, LL_SBI_USER_XATTR);
 805                 if (tmp) {
 806                         *flags &= ~tmp;
 807                         goto next;
 808                 }
 809                 tmp = ll_set_opt("context", s1, 1);
 810                 if (tmp)
 811                         goto next;
 812                 tmp = ll_set_opt("fscontext", s1, 1);
 813                 if (tmp)
 814                         goto next;
 815                 tmp = ll_set_opt("defcontext", s1, 1);
 816                 if (tmp)
 817                         goto next;
 818                 tmp = ll_set_opt("rootcontext", s1, 1);
 819                 if (tmp)
 820                         goto next;
 821                 tmp = ll_set_opt("user_fid2path", s1, LL_SBI_USER_FID2PATH);
 822                 if (tmp) {
 823                         *flags |= tmp;
 824                         goto next;
 825                 }
 826                 tmp = ll_set_opt("nouser_fid2path", s1, LL_SBI_USER_FID2PATH);
 827                 if (tmp) {
 828                         *flags &= ~tmp;
 829                         goto next;
 830                 }
 831
 832                 tmp = ll_set_opt("checksum", s1, LL_SBI_CHECKSUM);
 833                 if (tmp) {
 834                         *flags |= tmp;
 835                         goto next;
 836                 }
 837                 tmp = ll_set_opt("nochecksum", s1, LL_SBI_CHECKSUM);
 838                 if (tmp) {
 839                         *flags &= ~tmp;
 840                         goto next;
 841                 }
 842                 tmp = ll_set_opt("lruresize", s1, LL_SBI_LRU_RESIZE);
 843                 if (tmp) {
 844                         *flags |= tmp;
 845                         goto next;
 846                 }
 847                 tmp = ll_set_opt("nolruresize", s1, LL_SBI_LRU_RESIZE);
 848                 if (tmp) {
 849                         *flags &= ~tmp;
 850                         goto next;
 851                 }
 852                 tmp = ll_set_opt("lazystatfs", s1, LL_SBI_LAZYSTATFS);
 853                 if (tmp) {
 854                         *flags |= tmp;
 855                         goto next;
 856                 }
 857                 tmp = ll_set_opt("nolazystatfs", s1, LL_SBI_LAZYSTATFS);
 858                 if (tmp) {
 859                         *flags &= ~tmp;
 860                         goto next;
 861                 }
 862                 tmp = ll_set_opt("32bitapi", s1, LL_SBI_32BIT_API);
 863                 if (tmp) {
 864                         *flags |= tmp;
 865                         goto next;
 866                 }
 867                 tmp = ll_set_opt("verbose", s1, LL_SBI_VERBOSE);
 868                 if (tmp) {
 869                         *flags |= tmp;
 870                         goto next;
 871                 }
 872                 tmp = ll_set_opt("noverbose", s1, LL_SBI_VERBOSE);
 873                 if (tmp) {
 874                         *flags &= ~tmp;
 875                         goto next;
 876                 }
 877                 tmp = ll_set_opt("always_ping", s1, LL_SBI_ALWAYS_PING);
 878                 if (tmp) {
 879                         *flags |= tmp;
 880                         goto next;
 881                 }
 882                 LCONSOLE_ERROR_MSG(0x152, "Unknown option '%s', won't mount.\n",
 883                                    s1);
 884                 RETURN(-EINVAL);
 885
 886 next:
 887                 /* Find next opt */
 888                 s2 = strchr(s1, ',');
 889                 if (s2 == NULL)
 890                         break;
 891                 s1 = s2 + 1;
 892         }
 893         RETURN(0);
 894 }
 895
 896 void ll_lli_init(struct ll_inode_info *lli)
 897 {
 898         lli->lli_inode_magic = LLI_INODE_MAGIC;
 899         lli->lli_flags = 0;
 900         spin_lock_init(&lli->lli_lock);
 901         lli->lli_posix_acl = NULL;
 902         /* Do not set lli_fid, it has been initialized already. */
 903         fid_zero(&lli->lli_pfid);
 904         atomic_set(&lli->lli_open_count, 0);
 905         lli->lli_mds_read_och = NULL;
 906         lli->lli_mds_write_och = NULL;
 907         lli->lli_mds_exec_och = NULL;
 908         lli->lli_open_fd_read_count = 0;
 909         lli->lli_open_fd_write_count = 0;
 910         lli->lli_open_fd_exec_count = 0;
 911         mutex_init(&lli->lli_och_mutex);
 912         spin_lock_init(&lli->lli_agl_lock);
 913         spin_lock_init(&lli->lli_layout_lock);
 914         ll_layout_version_set(lli, CL_LAYOUT_GEN_NONE);
 915         lli->lli_clob = NULL;
 916
 917         init_rwsem(&lli->lli_xattrs_list_rwsem);
 918         mutex_init(&lli->lli_xattrs_enq_lock);
 919
 920         LASSERT(lli->lli_vfs_inode.i_mode != 0);
 921         if (S_ISDIR(lli->lli_vfs_inode.i_mode)) {
 922                 mutex_init(&lli->lli_readdir_mutex);
 923                 lli->lli_opendir_key = NULL;
 924                 lli->lli_sai = NULL;
 925                 spin_lock_init(&lli->lli_sa_lock);
 926                 lli->lli_opendir_pid = 0;
 927                 lli->lli_sa_enabled = 0;
 928                 lli->lli_def_stripe_offset = -1;
 929         } else {
 930                 mutex_init(&lli->lli_size_mutex);
 931                 lli->lli_symlink_name = NULL;
 932                 init_rwsem(&lli->lli_trunc_sem);
 933                 range_lock_tree_init(&lli->lli_write_tree);
 934                 init_rwsem(&lli->lli_glimpse_sem);
 935                 lli->lli_glimpse_time = 0;
 936                 INIT_LIST_HEAD(&lli->lli_agl_list);
 937                 lli->lli_agl_index = 0;
 938                 lli->lli_async_rc = 0;
 939         }
 940         mutex_init(&lli->lli_layout_mutex);
 941 }
 942
 943 static inline int ll_bdi_register(struct backing_dev_info *bdi)
 944 {
 945         static atomic_t ll_bdi_num = ATOMIC_INIT(0);
 946
 947         bdi->name = "lustre";
 948         return bdi_register(bdi, NULL, "lustre-%d",
 949                             atomic_inc_return(&ll_bdi_num));
 950 }
 951
 952 int ll_fill_super(struct super_block *sb, struct vfsmount *mnt)
 953 {
 954         struct lustre_profile *lprof = NULL;
 955         struct lustre_sb_info *lsi = s2lsi(sb);
 956         struct ll_sb_info *sbi;
 957         char  *dt = NULL, *md = NULL;
 958         char  *profilenm = get_profile_name(sb);
 959         struct config_llog_instance *cfg;
 960         /* %p for void* in printf needs 16+2 characters: 0xffffffffffffffff */
 961         const int instlen = sizeof(cfg->cfg_instance) * 2 + 2;
 962         int    err;
 963         ENTRY;
 964
 965         CDEBUG(D_VFSTRACE, "VFS Op: sb %p\n", sb);
 966
 967         OBD_ALLOC_PTR(cfg);
 968         if (cfg == NULL)
 969                 RETURN(-ENOMEM);
 970
 971         try_module_get(THIS_MODULE);
 972
 973         /* client additional sb info */
 974         lsi->lsi_llsbi = sbi = ll_init_sbi();
 975         if (!sbi) {
 976                 module_put(THIS_MODULE);
 977                 OBD_FREE_PTR(cfg);
 978                 RETURN(-ENOMEM);
 979         }
 980
 981         err = ll_options(lsi->lsi_lmd->lmd_opts, &sbi->ll_flags);
 982         if (err)
 983                 GOTO(out_free, err);
 984
 985         err = bdi_init(&lsi->lsi_bdi);
 986         if (err)
 987                 GOTO(out_free, err);
 988         lsi->lsi_flags |= LSI_BDI_INITIALIZED;
 989 #ifdef HAVE_BDI_CAP_MAP_COPY
 990         lsi->lsi_bdi.capabilities = BDI_CAP_MAP_COPY;
 991 #else
 992         lsi->lsi_bdi.capabilities = 0;
 993 #endif
 994         err = ll_bdi_register(&lsi->lsi_bdi);
 995         if (err)
 996                 GOTO(out_free, err);
 997
 998         sb->s_bdi = &lsi->lsi_bdi;
 999 #ifndef HAVE_DCACHE_LOCK
1000         /* kernel >= 2.6.38 store dentry operations in sb->s_d_op. */
1001         sb->s_d_op = &ll_d_ops;
1002 #endif
1003
1004         /* Generate a string unique to this super, in case some joker tries
1005            to mount the same fs at two mount points.
1006            Use the address of the super itself.*/
1007         cfg->cfg_instance = sb;
1008         cfg->cfg_uuid = lsi->lsi_llsbi->ll_sb_uuid;
1009         cfg->cfg_callback = class_config_llog_handler;
1010         /* set up client obds */
1011         err = lustre_process_log(sb, profilenm, cfg);
1012         if (err < 0)
1013                 GOTO(out_free, err);
1014
1015         /* Profile set with LCFG_MOUNTOPT so we can find our mdc and osc obds */
1016         lprof = class_get_profile(profilenm);
1017         if (lprof == NULL) {
1018                 LCONSOLE_ERROR_MSG(0x156, "The client profile '%s' could not be"
1019                                    " read from the MGS.  Does that filesystem "
1020                                    "exist?\n", profilenm);
1021                 GOTO(out_free, err = -EINVAL);
1022         }
1023         CDEBUG(D_CONFIG, "Found profile %s: mdc=%s osc=%s\n", profilenm,
1024                lprof->lp_md, lprof->lp_dt);
1025
1026         OBD_ALLOC(dt, strlen(lprof->lp_dt) + instlen + 2);
1027         if (!dt)
1028                 GOTO(out_free, err = -ENOMEM);
1029         sprintf(dt, "%s-%p", lprof->lp_dt, cfg->cfg_instance);
1030
1031         OBD_ALLOC(md, strlen(lprof->lp_md) + instlen + 2);
1032         if (!md)
1033                 GOTO(out_free, err = -ENOMEM);
1034         sprintf(md, "%s-%p", lprof->lp_md, cfg->cfg_instance);
1035
1036         /* connections, registrations, sb setup */
1037         err = client_common_fill_super(sb, md, dt, mnt);
1038         if (err < 0)
1039                 GOTO(out_free, err);
1040
1041         sbi->ll_client_common_fill_super_succeeded = 1;
1042
1043 out_free:
1044         if (md)
1045                 OBD_FREE(md, strlen(lprof->lp_md) + instlen + 2);
1046         if (dt)
1047                 OBD_FREE(dt, strlen(lprof->lp_dt) + instlen + 2);
1048         if (lprof != NULL)
1049                 class_put_profile(lprof);
1050         if (err)
1051                 ll_put_super(sb);
1052         else if (sbi->ll_flags & LL_SBI_VERBOSE)
1053                 LCONSOLE_WARN("Mounted %s\n", profilenm);
1054
1055         OBD_FREE_PTR(cfg);
1056         RETURN(err);
1057 } /* ll_fill_super */
1058
1059 void ll_put_super(struct super_block *sb)
1060 {
1061         struct config_llog_instance cfg, params_cfg;
1062         struct obd_device *obd;
1063         struct lustre_sb_info *lsi = s2lsi(sb);
1064         struct ll_sb_info *sbi = ll_s2sbi(sb);
1065         char *profilenm = get_profile_name(sb);
1066         long ccc_count;
1067         int next, force = 1, rc = 0;
1068         ENTRY;
1069
1070         CDEBUG(D_VFSTRACE, "VFS Op: sb %p - %s\n", sb, profilenm);
1071
1072         cfg.cfg_instance = sb;
1073         lustre_end_log(sb, profilenm, &cfg);
1074
1075         params_cfg.cfg_instance = sb;
1076         lustre_end_log(sb, PARAMS_FILENAME, &params_cfg);
1077
1078         if (sbi->ll_md_exp) {
1079                 obd = class_exp2obd(sbi->ll_md_exp);
1080                 if (obd)
1081                         force = obd->obd_force;
1082         }
1083
1084         /* Wait for unstable pages to be committed to stable storage */
1085         if (force == 0) {
1086                 struct l_wait_info lwi = LWI_INTR(LWI_ON_SIGNAL_NOOP, NULL);
1087                 rc = l_wait_event(sbi->ll_cache->ccc_unstable_waitq,
1088                         atomic_long_read(&sbi->ll_cache->ccc_unstable_nr) == 0,
1089                         &lwi);
1090         }
1091
1092         ccc_count = atomic_long_read(&sbi->ll_cache->ccc_unstable_nr);
1093         if (force == 0 && rc != -EINTR)
1094                 LASSERTF(ccc_count == 0, "count: %li\n", ccc_count);
1095
1096
1097         /* We need to set force before the lov_disconnect in
1098            lustre_common_put_super, since l_d cleans up osc's as well. */
1099         if (force) {
1100                 next = 0;
1101                 while ((obd = class_devices_in_group(&sbi->ll_sb_uuid,
1102                                                      &next)) != NULL) {
1103                         obd->obd_force = force;
1104                 }
1105         }
1106
1107         if (sbi->ll_client_common_fill_super_succeeded) {
1108                 /* Only if client_common_fill_super succeeded */
1109                 client_common_put_super(sb);
1110         }
1111
1112         next = 0;
1113         while ((obd = class_devices_in_group(&sbi->ll_sb_uuid, &next)) !=NULL) {
1114                 class_manual_cleanup(obd);
1115         }
1116
1117         if (sbi->ll_flags & LL_SBI_VERBOSE)
1118                 LCONSOLE_WARN("Unmounted %s\n", profilenm ? profilenm : "");
1119
1120         if (profilenm)
1121                 class_del_profile(profilenm);
1122
1123         if (lsi->lsi_flags & LSI_BDI_INITIALIZED) {
1124                 bdi_destroy(&lsi->lsi_bdi);
1125                 lsi->lsi_flags &= ~LSI_BDI_INITIALIZED;
1126         }
1127
1128         ll_free_sbi(sb);
1129         lsi->lsi_llsbi = NULL;
1130
1131         lustre_common_put_super(sb);
1132
1133         cl_env_cache_purge(~0);
1134
1135         module_put(THIS_MODULE);
1136
1137         EXIT;
1138 } /* client_put_super */
1139
1140 struct inode *ll_inode_from_resource_lock(struct ldlm_lock *lock)
1141 {
1142         struct inode *inode = NULL;
1143
1144         /* NOTE: we depend on atomic igrab() -bzzz */
1145         lock_res_and_lock(lock);
1146         if (lock->l_resource->lr_lvb_inode) {
1147                 struct ll_inode_info * lli;
1148                 lli = ll_i2info(lock->l_resource->lr_lvb_inode);
1149                 if (lli->lli_inode_magic == LLI_INODE_MAGIC) {
1150                         inode = igrab(lock->l_resource->lr_lvb_inode);
1151                 } else {
1152                         inode = lock->l_resource->lr_lvb_inode;
1153                         LDLM_DEBUG_LIMIT(inode->i_state & I_FREEING ?  D_INFO :
1154                                          D_WARNING, lock, "lr_lvb_inode %p is "
1155                                          "bogus: magic %08x",
1156                                          lock->l_resource->lr_lvb_inode,
1157                                          lli->lli_inode_magic);
1158                         inode = NULL;
1159                 }
1160         }
1161         unlock_res_and_lock(lock);
1162         return inode;
1163 }
1164
1165 void ll_dir_clear_lsm_md(struct inode *inode)
1166 {
1167         struct ll_inode_info *lli = ll_i2info(inode);
1168
1169         LASSERT(S_ISDIR(inode->i_mode));
1170
1171         if (lli->lli_lsm_md != NULL) {
1172                 lmv_free_memmd(lli->lli_lsm_md);
1173                 lli->lli_lsm_md = NULL;
1174         }
1175 }
1176
1177 static struct inode *ll_iget_anon_dir(struct super_block *sb,
1178                                       const struct lu_fid *fid,
1179                                       struct lustre_md *md)
1180 {
1181         struct ll_sb_info       *sbi = ll_s2sbi(sb);
1182         struct mdt_body         *body = md->body;
1183         struct inode            *inode;
1184         ino_t                   ino;
1185         ENTRY;
1186
1187         ino = cl_fid_build_ino(fid, sbi->ll_flags & LL_SBI_32BIT_API);
1188         inode = iget_locked(sb, ino);
1189         if (inode == NULL) {
1190                 CERROR("%s: failed get simple inode "DFID": rc = -ENOENT\n",
1191                        ll_get_fsname(sb, NULL, 0), PFID(fid));
1192                 RETURN(ERR_PTR(-ENOENT));
1193         }
1194
1195         if (inode->i_state & I_NEW) {
1196                 struct ll_inode_info *lli = ll_i2info(inode);
1197                 struct lmv_stripe_md *lsm = md->lmv;
1198
1199                 inode->i_mode = (inode->i_mode & ~S_IFMT) |
1200                                 (body->mbo_mode & S_IFMT);
1201                 LASSERTF(S_ISDIR(inode->i_mode), "Not slave inode "DFID"\n",
1202                          PFID(fid));
1203
1204                 LTIME_S(inode->i_mtime) = 0;
1205                 LTIME_S(inode->i_atime) = 0;
1206                 LTIME_S(inode->i_ctime) = 0;
1207                 inode->i_rdev = 0;
1208
1209 #ifdef HAVE_BACKING_DEV_INFO
1210                 /* initializing backing dev info. */
1211                 inode->i_mapping->backing_dev_info =
1212                                                 &s2lsi(inode->i_sb)->lsi_bdi;
1213 #endif
1214                 inode->i_op = &ll_dir_inode_operations;
1215                 inode->i_fop = &ll_dir_operations;
1216                 lli->lli_fid = *fid;
1217                 ll_lli_init(lli);
1218
1219                 LASSERT(lsm != NULL);
1220                 /* master object FID */
1221                 lli->lli_pfid = body->mbo_fid1;
1222                 CDEBUG(D_INODE, "lli %p slave "DFID" master "DFID"\n",
1223                        lli, PFID(fid), PFID(&lli->lli_pfid));
1224                 unlock_new_inode(inode);
1225         }
1226
1227         RETURN(inode);
1228 }
1229
1230 static int ll_init_lsm_md(struct inode *inode, struct lustre_md *md)
1231 {
1232         struct lu_fid *fid;
1233         struct lmv_stripe_md *lsm = md->lmv;
1234         int i;
1235
1236         LASSERT(lsm != NULL);
1237         /* XXX sigh, this lsm_root initialization should be in
1238          * LMV layer, but it needs ll_iget right now, so we
1239          * put this here right now. */
1240         for (i = 0; i < lsm->lsm_md_stripe_count; i++) {
1241                 fid = &lsm->lsm_md_oinfo[i].lmo_fid;
1242                 LASSERT(lsm->lsm_md_oinfo[i].lmo_root == NULL);
1243                 /* Unfortunately ll_iget will call ll_update_inode,
1244                  * where the initialization of slave inode is slightly
1245                  * different, so it reset lsm_md to NULL to avoid
1246                  * initializing lsm for slave inode. */
1247                 /* For migrating inode, master stripe and master object will
1248                  * be same, so we only need assign this inode */
1249                 if (lsm->lsm_md_hash_type & LMV_HASH_FLAG_MIGRATION && i == 0)
1250                         lsm->lsm_md_oinfo[i].lmo_root = inode;
1251                 else
1252                         lsm->lsm_md_oinfo[i].lmo_root =
1253                                 ll_iget_anon_dir(inode->i_sb, fid, md);
1254
1255                 if (IS_ERR(lsm->lsm_md_oinfo[i].lmo_root)) {
1256                         int rc = PTR_ERR(lsm->lsm_md_oinfo[i].lmo_root);
1257
1258                         lsm->lsm_md_oinfo[i].lmo_root = NULL;
1259                         return rc;
1260                 }
1261         }
1262
1263         return 0;
1264 }
1265
1266 static inline int lli_lsm_md_eq(const struct lmv_stripe_md *lsm_md1,
1267                                 const struct lmv_stripe_md *lsm_md2)
1268 {
1269         return lsm_md1->lsm_md_magic == lsm_md2->lsm_md_magic &&
1270                lsm_md1->lsm_md_stripe_count == lsm_md2->lsm_md_stripe_count &&
1271                lsm_md1->lsm_md_master_mdt_index ==
1272                                         lsm_md2->lsm_md_master_mdt_index &&
1273                lsm_md1->lsm_md_hash_type == lsm_md2->lsm_md_hash_type &&
1274                lsm_md1->lsm_md_layout_version ==
1275                                         lsm_md2->lsm_md_layout_version &&
1276                strcmp(lsm_md1->lsm_md_pool_name,
1277                       lsm_md2->lsm_md_pool_name) == 0;
1278 }
1279
1280 static int ll_update_lsm_md(struct inode *inode, struct lustre_md *md)
1281 {
1282         struct ll_inode_info *lli = ll_i2info(inode);
1283         struct lmv_stripe_md *lsm = md->lmv;
1284         int     rc;
1285         ENTRY;
1286
1287         LASSERT(S_ISDIR(inode->i_mode));
1288         CDEBUG(D_INODE, "update lsm %p of "DFID"\n", lli->lli_lsm_md,
1289                PFID(ll_inode2fid(inode)));
1290
1291         /* no striped information from request. */
1292         if (lsm == NULL) {
1293                 if (lli->lli_lsm_md == NULL) {
1294                         RETURN(0);
1295                 } else if (lli->lli_lsm_md->lsm_md_hash_type &
1296                                                 LMV_HASH_FLAG_MIGRATION) {
1297                         /* migration is done, the temporay MIGRATE layout has
1298                          * been removed */
1299                         CDEBUG(D_INODE, DFID" finish migration.\n",
1300                                PFID(ll_inode2fid(inode)));
1301                         lmv_free_memmd(lli->lli_lsm_md);
1302                         lli->lli_lsm_md = NULL;
1303                         RETURN(0);
1304                 } else {
1305                         /* The lustre_md from req does not include stripeEA,
1306                          * see ll_md_setattr */
1307                         RETURN(0);
1308                 }
1309         }
1310
1311         /* set the directory layout */
1312         if (lli->lli_lsm_md == NULL) {
1313                 struct cl_attr  *attr;
1314
1315                 rc = ll_init_lsm_md(inode, md);
1316                 if (rc != 0)
1317                         RETURN(rc);
1318
1319                 /* set md->lmv to NULL, so the following free lustre_md
1320                  * will not free this lsm */
1321                 md->lmv = NULL;
1322                 lli->lli_lsm_md = lsm;
1323
1324                 OBD_ALLOC_PTR(attr);
1325                 if (attr == NULL)
1326                         RETURN(-ENOMEM);
1327
1328                 /* validate the lsm */
1329                 rc = md_merge_attr(ll_i2mdexp(inode), lsm, attr,
1330                                    ll_md_blocking_ast);
1331                 if (rc != 0) {
1332                         OBD_FREE_PTR(attr);
1333                         RETURN(rc);
1334                 }
1335
1336                 if (md->body->mbo_valid & OBD_MD_FLNLINK)
1337                         md->body->mbo_nlink = attr->cat_nlink;
1338                 if (md->body->mbo_valid & OBD_MD_FLSIZE)
1339                         md->body->mbo_size = attr->cat_size;
1340                 if (md->body->mbo_valid & OBD_MD_FLATIME)
1341                         md->body->mbo_atime = attr->cat_atime;
1342                 if (md->body->mbo_valid & OBD_MD_FLCTIME)
1343                         md->body->mbo_ctime = attr->cat_ctime;
1344                 if (md->body->mbo_valid & OBD_MD_FLMTIME)
1345                         md->body->mbo_mtime = attr->cat_mtime;
1346
1347                 OBD_FREE_PTR(attr);
1348
1349                 CDEBUG(D_INODE, "Set lsm %p magic %x to "DFID"\n", lsm,
1350                        lsm->lsm_md_magic, PFID(ll_inode2fid(inode)));
1351                 RETURN(0);
1352         }
1353
1354         /* Compare the old and new stripe information */
1355         if (!lsm_md_eq(lli->lli_lsm_md, lsm)) {
1356                 struct lmv_stripe_md    *old_lsm = lli->lli_lsm_md;
1357                 int                     idx;
1358
1359                 CERROR("%s: inode "DFID"(%p)'s lmv layout mismatch (%p)/(%p)"
1360                        "magic:0x%x/0x%x stripe count: %d/%d master_mdt: %d/%d"
1361                        "hash_type:0x%x/0x%x layout: 0x%x/0x%x pool:%s/%s\n",
1362                        ll_get_fsname(inode->i_sb, NULL, 0), PFID(&lli->lli_fid),
1363                        inode, lsm, old_lsm,
1364                        lsm->lsm_md_magic, old_lsm->lsm_md_magic,
1365                        lsm->lsm_md_stripe_count,
1366                        old_lsm->lsm_md_stripe_count,
1367                        lsm->lsm_md_master_mdt_index,
1368                        old_lsm->lsm_md_master_mdt_index,
1369                        lsm->lsm_md_hash_type, old_lsm->lsm_md_hash_type,
1370                        lsm->lsm_md_layout_version,
1371                        old_lsm->lsm_md_layout_version,
1372                        lsm->lsm_md_pool_name,
1373                        old_lsm->lsm_md_pool_name);
1374
1375                 for (idx = 0; idx < old_lsm->lsm_md_stripe_count; idx++) {
1376                         CERROR("%s: sub FIDs in old lsm idx %d, old: "DFID"\n",
1377                                ll_get_fsname(inode->i_sb, NULL, 0), idx,
1378                                PFID(&old_lsm->lsm_md_oinfo[idx].lmo_fid));
1379                 }
1380
1381                 for (idx = 0; idx < lsm->lsm_md_stripe_count; idx++) {
1382                         CERROR("%s: sub FIDs in new lsm idx %d, new: "DFID"\n",
1383                                ll_get_fsname(inode->i_sb, NULL, 0), idx,
1384                                PFID(&lsm->lsm_md_oinfo[idx].lmo_fid));
1385                 }
1386
1387                 RETURN(-EIO);
1388         }
1389
1390         RETURN(0);
1391 }
1392
1393 void ll_clear_inode(struct inode *inode)
1394 {
1395         struct ll_inode_info *lli = ll_i2info(inode);
1396         struct ll_sb_info *sbi = ll_i2sbi(inode);
1397         ENTRY;
1398
1399         CDEBUG(D_VFSTRACE, "VFS Op:inode="DFID"(%p)\n",
1400                PFID(ll_inode2fid(inode)), inode);
1401
1402         if (S_ISDIR(inode->i_mode)) {
1403                 /* these should have been cleared in ll_file_release */
1404                 LASSERT(lli->lli_opendir_key == NULL);
1405                 LASSERT(lli->lli_sai == NULL);
1406                 LASSERT(lli->lli_opendir_pid == 0);
1407         }
1408
1409         md_null_inode(sbi->ll_md_exp, ll_inode2fid(inode));
1410
1411         LASSERT(!lli->lli_open_fd_write_count);
1412         LASSERT(!lli->lli_open_fd_read_count);
1413         LASSERT(!lli->lli_open_fd_exec_count);
1414
1415         if (lli->lli_mds_write_och)
1416                 ll_md_real_close(inode, FMODE_WRITE);
1417         if (lli->lli_mds_exec_och)
1418                 ll_md_real_close(inode, FMODE_EXEC);
1419         if (lli->lli_mds_read_och)
1420                 ll_md_real_close(inode, FMODE_READ);
1421
1422         if (S_ISLNK(inode->i_mode) && lli->lli_symlink_name) {
1423                 OBD_FREE(lli->lli_symlink_name,
1424                          strlen(lli->lli_symlink_name) + 1);
1425                 lli->lli_symlink_name = NULL;
1426         }
1427
1428         ll_xattr_cache_destroy(inode);
1429
1430 #ifdef CONFIG_FS_POSIX_ACL
1431         if (lli->lli_posix_acl) {
1432                 LASSERT(atomic_read(&lli->lli_posix_acl->a_refcount) == 1);
1433                 posix_acl_release(lli->lli_posix_acl);
1434                 lli->lli_posix_acl = NULL;
1435         }
1436 #endif
1437         lli->lli_inode_magic = LLI_INODE_DEAD;
1438
1439         if (S_ISDIR(inode->i_mode))
1440                 ll_dir_clear_lsm_md(inode);
1441         else if (S_ISREG(inode->i_mode) && !is_bad_inode(inode))
1442                 LASSERT(list_empty(&lli->lli_agl_list));
1443
1444         /*
1445          * XXX This has to be done before lsm is freed below, because
1446          * cl_object still uses inode lsm.
1447          */
1448         cl_inode_fini(inode);
1449
1450         EXIT;
1451 }
1452
1453 static int ll_md_setattr(struct dentry *dentry, struct md_op_data *op_data)
1454 {
1455         struct lustre_md md;
1456         struct inode *inode = dentry->d_inode;
1457         struct ll_sb_info *sbi = ll_i2sbi(inode);
1458         struct ptlrpc_request *request = NULL;
1459         int rc, ia_valid;
1460         ENTRY;
1461
1462         op_data = ll_prep_md_op_data(op_data, inode, NULL, NULL, 0, 0,
1463                                      LUSTRE_OPC_ANY, NULL);
1464         if (IS_ERR(op_data))
1465                 RETURN(PTR_ERR(op_data));
1466
1467         rc = md_setattr(sbi->ll_md_exp, op_data, NULL, 0, &request);
1468         if (rc) {
1469                 ptlrpc_req_finished(request);
1470                 if (rc == -ENOENT) {
1471                         clear_nlink(inode);
1472                         /* Unlinked special device node? Or just a race?
1473                          * Pretend we done everything. */
1474                         if (!S_ISREG(inode->i_mode) &&
1475                             !S_ISDIR(inode->i_mode)) {
1476                                 ia_valid = op_data->op_attr.ia_valid;
1477                                 op_data->op_attr.ia_valid &= ~TIMES_SET_FLAGS;
1478                                 rc = simple_setattr(dentry, &op_data->op_attr);
1479                                 op_data->op_attr.ia_valid = ia_valid;
1480                         }
1481                 } else if (rc != -EPERM && rc != -EACCES && rc != -ETXTBSY) {
1482                         CERROR("md_setattr fails: rc = %d\n", rc);
1483                 }
1484                 RETURN(rc);
1485         }
1486
1487         rc = md_get_lustre_md(sbi->ll_md_exp, request, sbi->ll_dt_exp,
1488                               sbi->ll_md_exp, &md);
1489         if (rc) {
1490                 ptlrpc_req_finished(request);
1491                 RETURN(rc);
1492         }
1493
1494         ia_valid = op_data->op_attr.ia_valid;
1495         /* inode size will be in ll_setattr_ost, can't do it now since dirty
1496          * cache is not cleared yet. */
1497         op_data->op_attr.ia_valid &= ~(TIMES_SET_FLAGS | ATTR_SIZE);
1498         if (S_ISREG(inode->i_mode))
1499                 mutex_lock(&inode->i_mutex);
1500         rc = simple_setattr(dentry, &op_data->op_attr);
1501         if (S_ISREG(inode->i_mode))
1502                 mutex_unlock(&inode->i_mutex);
1503         op_data->op_attr.ia_valid = ia_valid;
1504
1505         rc = ll_update_inode(inode, &md);
1506         ptlrpc_req_finished(request);
1507
1508         RETURN(rc);
1509 }
1510
1511 /* If this inode has objects allocated to it (lsm != NULL), then the OST
1512  * object(s) determine the file size and mtime.  Otherwise, the MDS will
1513  * keep these values until such a time that objects are allocated for it.
1514  * We do the MDS operations first, as it is checking permissions for us.
1515  * We don't to the MDS RPC if there is nothing that we want to store there,
1516  * otherwise there is no harm in updating mtime/atime on the MDS if we are
1517  * going to do an RPC anyways.
1518  *
1519  * If we are doing a truncate, we will send the mtime and ctime updates
1520  * to the OST with the punch RPC, otherwise we do an explicit setattr RPC.
1521  * I don't believe it is possible to get e.g. ATTR_MTIME_SET and ATTR_SIZE
1522  * at the same time.
1523  *
1524  * In case of HSMimport, we only set attr on MDS.
1525  */
1526 int ll_setattr_raw(struct dentry *dentry, struct iattr *attr, bool hsm_import)
1527 {
1528         struct inode *inode = dentry->d_inode;
1529         struct ll_inode_info *lli = ll_i2info(inode);
1530         struct md_op_data *op_data = NULL;
1531         int rc = 0;
1532         ENTRY;
1533
1534         CDEBUG(D_VFSTRACE, "%s: setattr inode "DFID"(%p) from %llu to %llu, "
1535                "valid %x, hsm_import %d\n",
1536                ll_get_fsname(inode->i_sb, NULL, 0), PFID(&lli->lli_fid),
1537                inode, i_size_read(inode), attr->ia_size, attr->ia_valid,
1538                hsm_import);
1539
1540         if (attr->ia_valid & ATTR_SIZE) {
1541                 /* Check new size against VFS/VM file size limit and rlimit */
1542                 rc = inode_newsize_ok(inode, attr->ia_size);
1543                 if (rc)
1544                         RETURN(rc);
1545
1546                 /* The maximum Lustre file size is variable, based on the
1547                  * OST maximum object size and number of stripes.  This
1548                  * needs another check in addition to the VFS check above. */
1549                 if (attr->ia_size > ll_file_maxbytes(inode)) {
1550                         CDEBUG(D_INODE,"file "DFID" too large %llu > "LPU64"\n",
1551                                PFID(&lli->lli_fid), attr->ia_size,
1552                                ll_file_maxbytes(inode));
1553                         RETURN(-EFBIG);
1554                 }
1555
1556                 attr->ia_valid |= ATTR_MTIME | ATTR_CTIME;
1557         }
1558
1559         /* POSIX: check before ATTR_*TIME_SET set (from inode_change_ok) */
1560         if (attr->ia_valid & TIMES_SET_FLAGS) {
1561                 if ((!uid_eq(current_fsuid(), inode->i_uid)) &&
1562                     !cfs_capable(CFS_CAP_FOWNER))
1563                         RETURN(-EPERM);
1564         }
1565
1566         /* We mark all of the fields "set" so MDS/OST does not re-set them */
1567         if (!(attr->ia_valid & ATTR_CTIME_SET) &&
1568             (attr->ia_valid & ATTR_CTIME)) {
1569                 attr->ia_ctime = CURRENT_TIME;
1570                 attr->ia_valid |= ATTR_CTIME_SET;
1571         }
1572         if (!(attr->ia_valid & ATTR_ATIME_SET) &&
1573             (attr->ia_valid & ATTR_ATIME)) {
1574                 attr->ia_atime = CURRENT_TIME;
1575                 attr->ia_valid |= ATTR_ATIME_SET;
1576         }
1577         if (!(attr->ia_valid & ATTR_MTIME_SET) &&
1578             (attr->ia_valid & ATTR_MTIME)) {
1579                 attr->ia_mtime = CURRENT_TIME;
1580                 attr->ia_valid |= ATTR_MTIME_SET;
1581         }
1582
1583         if (attr->ia_valid & (ATTR_MTIME | ATTR_CTIME))
1584                 CDEBUG(D_INODE, "setting mtime %lu, ctime %lu, now = %lu\n",
1585                        LTIME_S(attr->ia_mtime), LTIME_S(attr->ia_ctime),
1586                        cfs_time_current_sec());
1587
1588         if (S_ISREG(inode->i_mode)) {
1589                 if (attr->ia_valid & ATTR_SIZE)
1590                         inode_dio_write_done(inode);
1591                 mutex_unlock(&inode->i_mutex);
1592         }
1593
1594         /* We always do an MDS RPC, even if we're only changing the size;
1595          * only the MDS knows whether truncate() should fail with -ETXTBUSY */
1596
1597         OBD_ALLOC_PTR(op_data);
1598         if (op_data == NULL)
1599                 GOTO(out, rc = -ENOMEM);
1600
1601         if (!hsm_import && attr->ia_valid & ATTR_SIZE) {
1602                 /* If we are changing file size, file content is
1603                  * modified, flag it. */
1604                 attr->ia_valid |= MDS_OPEN_OWNEROVERRIDE;
1605                 op_data->op_bias |= MDS_DATA_MODIFIED;
1606                 ll_file_clear_flag(lli, LLIF_DATA_MODIFIED);
1607         }
1608
1609         op_data->op_attr = *attr;
1610
1611         rc = ll_md_setattr(dentry, op_data);
1612         if (rc)
1613                 GOTO(out, rc);
1614
1615         if (!S_ISREG(inode->i_mode) || hsm_import)
1616                 GOTO(out, rc = 0);
1617
1618         if (attr->ia_valid & (ATTR_SIZE |
1619                               ATTR_ATIME | ATTR_ATIME_SET |
1620                               ATTR_MTIME | ATTR_MTIME_SET |
1621                               ATTR_CTIME | ATTR_CTIME_SET)) {
1622                 /* For truncate and utimes sending attributes to OSTs, setting
1623                  * mtime/atime to the past will be performed under PW [0:EOF]
1624                  * extent lock (new_size:EOF for truncate).  It may seem
1625                  * excessive to send mtime/atime updates to OSTs when not
1626                  * setting times to past, but it is necessary due to possible
1627                  * time de-synchronization between MDT inode and OST objects */
1628                 rc = cl_setattr_ost(lli->lli_clob, attr, 0);
1629         }
1630
1631         /* If the file was restored, it needs to set dirty flag.
1632          *
1633          * We've already sent MDS_DATA_MODIFIED flag in
1634          * ll_md_setattr() for truncate. However, the MDT refuses to
1635          * set the HS_DIRTY flag on released files, so we have to set
1636          * it again if the file has been restored. Please check how
1637          * LLIF_DATA_MODIFIED is set in vvp_io_setattr_fini().
1638          *
1639          * Please notice that if the file is not released, the previous
1640          * MDS_DATA_MODIFIED has taken effect and usually
1641          * LLIF_DATA_MODIFIED is not set(see vvp_io_setattr_fini()).
1642          * This way we can save an RPC for common open + trunc
1643          * operation. */
1644         if (ll_file_test_and_clear_flag(lli, LLIF_DATA_MODIFIED)) {
1645                 struct hsm_state_set hss = {
1646                         .hss_valid = HSS_SETMASK,
1647                         .hss_setmask = HS_DIRTY,
1648                 };
1649                 int rc2;
1650
1651                 rc2 = ll_hsm_state_set(inode, &hss);
1652                 /* truncate and write can happen at the same time, so that
1653                  * the file can be set modified even though the file is not
1654                  * restored from released state, and ll_hsm_state_set() is
1655                  * not applicable for the file, and rc2 < 0 is normal in this
1656                  * case. */
1657                 if (rc2 < 0)
1658                         CDEBUG(D_INFO, DFID "HSM set dirty failed: rc2 = %d\n",
1659                                PFID(ll_inode2fid(inode)), rc2);
1660         }
1661
1662         EXIT;
1663 out:
1664         if (op_data != NULL)
1665                 ll_finish_md_op_data(op_data);
1666
1667         if (S_ISREG(inode->i_mode)) {
1668                 mutex_lock(&inode->i_mutex);
1669                 if ((attr->ia_valid & ATTR_SIZE) && !hsm_import)
1670                         inode_dio_wait(inode);
1671         }
1672
1673         ll_stats_ops_tally(ll_i2sbi(inode), (attr->ia_valid & ATTR_SIZE) ?
1674                         LPROC_LL_TRUNC : LPROC_LL_SETATTR, 1);
1675
1676         return rc;
1677 }
1678
1679 int ll_setattr(struct dentry *de, struct iattr *attr)
1680 {
1681         int mode = de->d_inode->i_mode;
1682
1683         if ((attr->ia_valid & (ATTR_CTIME|ATTR_SIZE|ATTR_MODE)) ==
1684                               (ATTR_CTIME|ATTR_SIZE|ATTR_MODE))
1685                 attr->ia_valid |= MDS_OPEN_OWNEROVERRIDE;
1686
1687         if (((attr->ia_valid & (ATTR_MODE|ATTR_FORCE|ATTR_SIZE)) ==
1688                                (ATTR_SIZE|ATTR_MODE)) &&
1689             (((mode & S_ISUID) && !(attr->ia_mode & S_ISUID)) ||
1690              (((mode & (S_ISGID|S_IXGRP)) == (S_ISGID|S_IXGRP)) &&
1691               !(attr->ia_mode & S_ISGID))))
1692                 attr->ia_valid |= ATTR_FORCE;
1693
1694         if ((attr->ia_valid & ATTR_MODE) &&
1695             (mode & S_ISUID) &&
1696             !(attr->ia_mode & S_ISUID) &&
1697             !(attr->ia_valid & ATTR_KILL_SUID))
1698                 attr->ia_valid |= ATTR_KILL_SUID;
1699
1700         if ((attr->ia_valid & ATTR_MODE) &&
1701             ((mode & (S_ISGID|S_IXGRP)) == (S_ISGID|S_IXGRP)) &&
1702             !(attr->ia_mode & S_ISGID) &&
1703             !(attr->ia_valid & ATTR_KILL_SGID))
1704                 attr->ia_valid |= ATTR_KILL_SGID;
1705
1706         return ll_setattr_raw(de, attr, false);
1707 }
1708
1709 int ll_statfs_internal(struct super_block *sb, struct obd_statfs *osfs,
1710                        __u64 max_age, __u32 flags)
1711 {
1712         struct ll_sb_info *sbi = ll_s2sbi(sb);
1713         struct obd_statfs obd_osfs;
1714         int rc;
1715         ENTRY;
1716
1717         rc = obd_statfs(NULL, sbi->ll_md_exp, osfs, max_age, flags);
1718         if (rc) {
1719                 CERROR("md_statfs fails: rc = %d\n", rc);
1720                 RETURN(rc);
1721         }
1722
1723         osfs->os_type = sb->s_magic;
1724
1725         CDEBUG(D_SUPER, "MDC blocks "LPU64"/"LPU64" objects "LPU64"/"LPU64"\n",
1726                osfs->os_bavail, osfs->os_blocks, osfs->os_ffree,osfs->os_files);
1727
1728         if (sbi->ll_flags & LL_SBI_LAZYSTATFS)
1729                 flags |= OBD_STATFS_NODELAY;
1730
1731         rc = obd_statfs_rqset(sbi->ll_dt_exp, &obd_osfs, max_age, flags);
1732         if (rc) {
1733                 CERROR("obd_statfs fails: rc = %d\n", rc);
1734                 RETURN(rc);
1735         }
1736
1737         CDEBUG(D_SUPER, "OSC blocks "LPU64"/"LPU64" objects "LPU64"/"LPU64"\n",
1738                obd_osfs.os_bavail, obd_osfs.os_blocks, obd_osfs.os_ffree,
1739                obd_osfs.os_files);
1740
1741         osfs->os_bsize = obd_osfs.os_bsize;
1742         osfs->os_blocks = obd_osfs.os_blocks;
1743         osfs->os_bfree = obd_osfs.os_bfree;
1744         osfs->os_bavail = obd_osfs.os_bavail;
1745
1746         /* If we don't have as many objects free on the OST as inodes
1747          * on the MDS, we reduce the total number of inodes to
1748          * compensate, so that the "inodes in use" number is correct.
1749          */
1750         if (obd_osfs.os_ffree < osfs->os_ffree) {
1751                 osfs->os_files = (osfs->os_files - osfs->os_ffree) +
1752                         obd_osfs.os_ffree;
1753                 osfs->os_ffree = obd_osfs.os_ffree;
1754         }
1755
1756         RETURN(rc);
1757 }
1758 int ll_statfs(struct dentry *de, struct kstatfs *sfs)
1759 {
1760         struct super_block *sb = de->d_sb;
1761         struct obd_statfs osfs;
1762         __u64 fsid = huge_encode_dev(sb->s_dev);
1763         int rc;
1764
1765         CDEBUG(D_VFSTRACE, "VFS Op: at "LPU64" jiffies\n", get_jiffies_64());
1766         ll_stats_ops_tally(ll_s2sbi(sb), LPROC_LL_STAFS, 1);
1767
1768         /* Some amount of caching on the client is allowed */
1769         rc = ll_statfs_internal(sb, &osfs,
1770                                 cfs_time_shift_64(-OBD_STATFS_CACHE_SECONDS),
1771                                 0);
1772         if (rc)
1773                 return rc;
1774
1775         statfs_unpack(sfs, &osfs);
1776
1777         /* We need to downshift for all 32-bit kernels, because we can't
1778          * tell if the kernel is being called via sys_statfs64() or not.
1779          * Stop before overflowing f_bsize - in which case it is better
1780          * to just risk EOVERFLOW if caller is using old sys_statfs(). */
1781         if (sizeof(long) < 8) {
1782                 while (osfs.os_blocks > ~0UL && sfs->f_bsize < 0x40000000) {
1783                         sfs->f_bsize <<= 1;
1784
1785                         osfs.os_blocks >>= 1;
1786                         osfs.os_bfree >>= 1;
1787                         osfs.os_bavail >>= 1;
1788                 }
1789         }
1790
1791         sfs->f_blocks = osfs.os_blocks;
1792         sfs->f_bfree = osfs.os_bfree;
1793         sfs->f_bavail = osfs.os_bavail;
1794         sfs->f_fsid.val[0] = (__u32)fsid;
1795         sfs->f_fsid.val[1] = (__u32)(fsid >> 32);
1796         return 0;
1797 }
1798
1799 void ll_inode_size_lock(struct inode *inode)
1800 {
1801         struct ll_inode_info *lli;
1802
1803         LASSERT(!S_ISDIR(inode->i_mode));
1804
1805         lli = ll_i2info(inode);
1806         mutex_lock(&lli->lli_size_mutex);
1807 }
1808
1809 void ll_inode_size_unlock(struct inode *inode)
1810 {
1811         struct ll_inode_info *lli;
1812
1813         lli = ll_i2info(inode);
1814         mutex_unlock(&lli->lli_size_mutex);
1815 }
1816
1817 int ll_update_inode(struct inode *inode, struct lustre_md *md)
1818 {
1819         struct ll_inode_info *lli = ll_i2info(inode);
1820         struct mdt_body *body = md->body;
1821         struct ll_sb_info *sbi = ll_i2sbi(inode);
1822
1823         if (body->mbo_valid & OBD_MD_FLEASIZE)
1824                 cl_file_inode_init(inode, md);
1825
1826         if (S_ISDIR(inode->i_mode)) {
1827                 int     rc;
1828
1829                 rc = ll_update_lsm_md(inode, md);
1830                 if (rc != 0)
1831                         return rc;
1832         }
1833
1834 #ifdef CONFIG_FS_POSIX_ACL
1835         if (body->mbo_valid & OBD_MD_FLACL) {
1836                 spin_lock(&lli->lli_lock);
1837                 if (lli->lli_posix_acl)
1838                         posix_acl_release(lli->lli_posix_acl);
1839                 lli->lli_posix_acl = md->posix_acl;
1840                 spin_unlock(&lli->lli_lock);
1841         }
1842 #endif
1843         inode->i_ino = cl_fid_build_ino(&body->mbo_fid1,
1844                                         sbi->ll_flags & LL_SBI_32BIT_API);
1845         inode->i_generation = cl_fid_build_gen(&body->mbo_fid1);
1846
1847         if (body->mbo_valid & OBD_MD_FLATIME) {
1848                 if (body->mbo_atime > LTIME_S(inode->i_atime))
1849                         LTIME_S(inode->i_atime) = body->mbo_atime;
1850                 lli->lli_atime = body->mbo_atime;
1851         }
1852
1853         if (body->mbo_valid & OBD_MD_FLMTIME) {
1854                 if (body->mbo_mtime > LTIME_S(inode->i_mtime)) {
1855                         CDEBUG(D_INODE, "setting ino %lu mtime from %lu "
1856                                "to "LPU64"\n", inode->i_ino,
1857                                LTIME_S(inode->i_mtime), body->mbo_mtime);
1858                         LTIME_S(inode->i_mtime) = body->mbo_mtime;
1859                 }
1860                 lli->lli_mtime = body->mbo_mtime;
1861         }
1862
1863         if (body->mbo_valid & OBD_MD_FLCTIME) {
1864                 if (body->mbo_ctime > LTIME_S(inode->i_ctime))
1865                         LTIME_S(inode->i_ctime) = body->mbo_ctime;
1866                 lli->lli_ctime = body->mbo_ctime;
1867         }
1868
1869         if (body->mbo_valid & OBD_MD_FLMODE)
1870                 inode->i_mode = (inode->i_mode & S_IFMT) |
1871                                 (body->mbo_mode & ~S_IFMT);
1872
1873         if (body->mbo_valid & OBD_MD_FLTYPE)
1874                 inode->i_mode = (inode->i_mode & ~S_IFMT) |
1875                                 (body->mbo_mode & S_IFMT);
1876
1877         LASSERT(inode->i_mode != 0);
1878         if (S_ISREG(inode->i_mode))
1879                 inode->i_blkbits = min(PTLRPC_MAX_BRW_BITS + 1,
1880                                        LL_MAX_BLKSIZE_BITS);
1881         else
1882                 inode->i_blkbits = inode->i_sb->s_blocksize_bits;
1883
1884         if (body->mbo_valid & OBD_MD_FLUID)
1885                 inode->i_uid = make_kuid(&init_user_ns, body->mbo_uid);
1886         if (body->mbo_valid & OBD_MD_FLGID)
1887                 inode->i_gid = make_kgid(&init_user_ns, body->mbo_gid);
1888         if (body->mbo_valid & OBD_MD_FLFLAGS)
1889                 inode->i_flags = ll_ext_to_inode_flags(body->mbo_flags);
1890         if (body->mbo_valid & OBD_MD_FLNLINK)
1891                 set_nlink(inode, body->mbo_nlink);
1892         if (body->mbo_valid & OBD_MD_FLRDEV)
1893                 inode->i_rdev = old_decode_dev(body->mbo_rdev);
1894
1895         if (body->mbo_valid & OBD_MD_FLID) {
1896                 /* FID shouldn't be changed! */
1897                 if (fid_is_sane(&lli->lli_fid)) {
1898                         LASSERTF(lu_fid_eq(&lli->lli_fid, &body->mbo_fid1),
1899                                  "Trying to change FID "DFID
1900                                  " to the "DFID", inode "DFID"(%p)\n",
1901                                  PFID(&lli->lli_fid), PFID(&body->mbo_fid1),
1902                                  PFID(ll_inode2fid(inode)), inode);
1903                 } else {
1904                         lli->lli_fid = body->mbo_fid1;
1905                 }
1906         }
1907
1908         LASSERT(fid_seq(&lli->lli_fid) != 0);
1909
1910         if (body->mbo_valid & OBD_MD_FLSIZE) {
1911                 i_size_write(inode, body->mbo_size);
1912
1913                 CDEBUG(D_VFSTRACE, "inode="DFID", updating i_size %llu\n",
1914                        PFID(ll_inode2fid(inode)),
1915                        (unsigned long long)body->mbo_size);
1916
1917                 if (body->mbo_valid & OBD_MD_FLBLOCKS)
1918                         inode->i_blocks = body->mbo_blocks;
1919         }
1920
1921         if (body->mbo_valid & OBD_MD_TSTATE) {
1922                 /* Set LLIF_FILE_RESTORING if restore ongoing and
1923                  * clear it when done to ensure to start again
1924                  * glimpsing updated attrs
1925                  */
1926                 if (body->mbo_t_state & MS_RESTORE)
1927                         ll_file_set_flag(lli, LLIF_FILE_RESTORING);
1928                 else
1929                         ll_file_clear_flag(lli, LLIF_FILE_RESTORING);
1930         }
1931
1932         return 0;
1933 }
1934
1935 int ll_read_inode2(struct inode *inode, void *opaque)
1936 {
1937         struct lustre_md *md = opaque;
1938         struct ll_inode_info *lli = ll_i2info(inode);
1939         int     rc;
1940         ENTRY;
1941
1942         CDEBUG(D_VFSTRACE, "VFS Op:inode="DFID"(%p)\n",
1943                PFID(&lli->lli_fid), inode);
1944
1945         /* Core attributes from the MDS first.  This is a new inode, and
1946          * the VFS doesn't zero times in the core inode so we have to do
1947          * it ourselves.  They will be overwritten by either MDS or OST
1948          * attributes - we just need to make sure they aren't newer. */
1949         LTIME_S(inode->i_mtime) = 0;
1950         LTIME_S(inode->i_atime) = 0;
1951         LTIME_S(inode->i_ctime) = 0;
1952         inode->i_rdev = 0;
1953         rc = ll_update_inode(inode, md);
1954         if (rc != 0)
1955                 RETURN(rc);
1956
1957         /* OIDEBUG(inode); */
1958
1959 #ifdef HAVE_BACKING_DEV_INFO
1960         /* initializing backing dev info. */
1961         inode->i_mapping->backing_dev_info = &s2lsi(inode->i_sb)->lsi_bdi;
1962 #endif
1963         if (S_ISREG(inode->i_mode)) {
1964                 struct ll_sb_info *sbi = ll_i2sbi(inode);
1965                 inode->i_op = &ll_file_inode_operations;
1966                 inode->i_fop = sbi->ll_fop;
1967                 inode->i_mapping->a_ops = (struct address_space_operations *)&ll_aops;
1968                 EXIT;
1969         } else if (S_ISDIR(inode->i_mode)) {
1970                 inode->i_op = &ll_dir_inode_operations;
1971                 inode->i_fop = &ll_dir_operations;
1972                 EXIT;
1973         } else if (S_ISLNK(inode->i_mode)) {
1974                 inode->i_op = &ll_fast_symlink_inode_operations;
1975                 EXIT;
1976         } else {
1977                 inode->i_op = &ll_special_inode_operations;
1978
1979                 init_special_inode(inode, inode->i_mode,
1980                                    inode->i_rdev);
1981
1982                 EXIT;
1983         }
1984
1985         return 0;
1986 }
1987
1988 void ll_delete_inode(struct inode *inode)
1989 {
1990         struct ll_inode_info *lli = ll_i2info(inode);
1991         ENTRY;
1992
1993         if (S_ISREG(inode->i_mode) && lli->lli_clob != NULL)
1994                 /* It is last chance to write out dirty pages,
1995                  * otherwise we may lose data while umount */
1996                 cl_sync_file_range(inode, 0, OBD_OBJECT_EOF, CL_FSYNC_LOCAL, 1);
1997
1998         truncate_inode_pages_final(&inode->i_data);
1999
2000         LASSERTF(inode->i_data.nrpages == 0, "inode="DFID"(%p) nrpages=%lu, "
2001                  "see https://jira.hpdd.intel.com/browse/LU-118\n",
2002                  PFID(ll_inode2fid(inode)), inode, inode->i_data.nrpages);
2003
2004 #ifdef HAVE_SBOPS_EVICT_INODE
2005         ll_clear_inode(inode);
2006 #endif
2007         clear_inode(inode);
2008
2009         EXIT;
2010 }
2011
2012 int ll_iocontrol(struct inode *inode, struct file *file,
2013                  unsigned int cmd, unsigned long arg)
2014 {
2015         struct ll_sb_info *sbi = ll_i2sbi(inode);
2016         struct ptlrpc_request *req = NULL;
2017         int rc, flags = 0;
2018         ENTRY;
2019
2020         switch(cmd) {
2021         case FSFILT_IOC_GETFLAGS: {
2022                 struct mdt_body *body;
2023                 struct md_op_data *op_data;
2024
2025                 op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL,
2026                                              0, 0, LUSTRE_OPC_ANY,
2027                                              NULL);
2028                 if (IS_ERR(op_data))
2029                         RETURN(PTR_ERR(op_data));
2030
2031                 op_data->op_valid = OBD_MD_FLFLAGS;
2032                 rc = md_getattr(sbi->ll_md_exp, op_data, &req);
2033                 ll_finish_md_op_data(op_data);
2034                 if (rc) {
2035                         CERROR("%s: failure inode "DFID": rc = %d\n",
2036                                sbi->ll_md_exp->exp_obd->obd_name,
2037                                PFID(ll_inode2fid(inode)), rc);
2038                         RETURN(-abs(rc));
2039                 }
2040
2041                 body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
2042
2043                 flags = body->mbo_flags;
2044
2045                 ptlrpc_req_finished(req);
2046
2047                 RETURN(put_user(flags, (int __user *)arg));
2048         }
2049         case FSFILT_IOC_SETFLAGS: {
2050                 struct iattr *attr;
2051                 struct md_op_data *op_data;
2052                 struct cl_object *obj;
2053
2054                 if (get_user(flags, (int __user *)arg))
2055                         RETURN(-EFAULT);
2056
2057                 op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL, 0, 0,
2058                                              LUSTRE_OPC_ANY, NULL);
2059                 if (IS_ERR(op_data))
2060                         RETURN(PTR_ERR(op_data));
2061
2062                 op_data->op_attr_flags = flags;
2063                 op_data->op_attr.ia_valid |= ATTR_ATTR_FLAG;
2064                 rc = md_setattr(sbi->ll_md_exp, op_data, NULL, 0, &req);
2065                 ll_finish_md_op_data(op_data);
2066                 ptlrpc_req_finished(req);
2067                 if (rc)
2068                         RETURN(rc);
2069
2070                 inode->i_flags = ll_ext_to_inode_flags(flags);
2071
2072                 obj = ll_i2info(inode)->lli_clob;
2073                 if (obj == NULL)
2074                         RETURN(0);
2075
2076                 OBD_ALLOC_PTR(attr);
2077                 if (attr == NULL)
2078                         RETURN(-ENOMEM);
2079
2080                 attr->ia_valid = ATTR_ATTR_FLAG;
2081                 rc = cl_setattr_ost(obj, attr, flags);
2082
2083                 OBD_FREE_PTR(attr);
2084                 RETURN(rc);
2085         }
2086         default:
2087                 RETURN(-ENOSYS);
2088         }
2089
2090         RETURN(0);
2091 }
2092
2093 int ll_flush_ctx(struct inode *inode)
2094 {
2095         struct ll_sb_info  *sbi = ll_i2sbi(inode);
2096
2097         CDEBUG(D_SEC, "flush context for user %d\n",
2098                from_kuid(&init_user_ns, current_uid()));
2099
2100         obd_set_info_async(NULL, sbi->ll_md_exp,
2101                            sizeof(KEY_FLUSH_CTX), KEY_FLUSH_CTX,
2102                            0, NULL, NULL);
2103         obd_set_info_async(NULL, sbi->ll_dt_exp,
2104                            sizeof(KEY_FLUSH_CTX), KEY_FLUSH_CTX,
2105                            0, NULL, NULL);
2106         return 0;
2107 }
2108
2109 /* umount -f client means force down, don't save state */
2110 void ll_umount_begin(struct super_block *sb)
2111 {
2112         struct ll_sb_info *sbi = ll_s2sbi(sb);
2113         struct obd_device *obd;
2114         struct obd_ioctl_data *ioc_data;
2115         ENTRY;
2116
2117         CDEBUG(D_VFSTRACE, "VFS Op: superblock %p count %d active %d\n", sb,
2118                sb->s_count, atomic_read(&sb->s_active));
2119
2120         obd = class_exp2obd(sbi->ll_md_exp);
2121         if (obd == NULL) {
2122                 CERROR("Invalid MDC connection handle "LPX64"\n",
2123                        sbi->ll_md_exp->exp_handle.h_cookie);
2124                 EXIT;
2125                 return;
2126         }
2127         obd->obd_force = 1;
2128
2129         obd = class_exp2obd(sbi->ll_dt_exp);
2130         if (obd == NULL) {
2131                 CERROR("Invalid LOV connection handle "LPX64"\n",
2132                        sbi->ll_dt_exp->exp_handle.h_cookie);
2133                 EXIT;
2134                 return;
2135         }
2136         obd->obd_force = 1;
2137
2138         OBD_ALLOC_PTR(ioc_data);
2139         if (ioc_data) {
2140                 obd_iocontrol(IOC_OSC_SET_ACTIVE, sbi->ll_md_exp,
2141                               sizeof *ioc_data, ioc_data, NULL);
2142
2143                 obd_iocontrol(IOC_OSC_SET_ACTIVE, sbi->ll_dt_exp,
2144                               sizeof *ioc_data, ioc_data, NULL);
2145
2146                 OBD_FREE_PTR(ioc_data);
2147         }
2148
2149         /* Really, we'd like to wait until there are no requests outstanding,
2150          * and then continue.  For now, we just invalidate the requests,
2151          * schedule() and sleep one second if needed, and hope.
2152          */
2153         schedule();
2154         EXIT;
2155 }
2156
2157 int ll_remount_fs(struct super_block *sb, int *flags, char *data)
2158 {
2159         struct ll_sb_info *sbi = ll_s2sbi(sb);
2160         char *profilenm = get_profile_name(sb);
2161         int err;
2162         __u32 read_only;
2163
2164         if ((*flags & MS_RDONLY) != (sb->s_flags & MS_RDONLY)) {
2165                 read_only = *flags & MS_RDONLY;
2166                 err = obd_set_info_async(NULL, sbi->ll_md_exp,
2167                                          sizeof(KEY_READ_ONLY),
2168                                          KEY_READ_ONLY, sizeof(read_only),
2169                                          &read_only, NULL);
2170                 if (err) {
2171                         LCONSOLE_WARN("Failed to remount %s %s (%d)\n",
2172                                       profilenm, read_only ?
2173                                       "read-only" : "read-write", err);
2174                         return err;
2175                 }
2176
2177                 if (read_only)
2178                         sb->s_flags |= MS_RDONLY;
2179                 else
2180                         sb->s_flags &= ~MS_RDONLY;
2181
2182                 if (sbi->ll_flags & LL_SBI_VERBOSE)
2183                         LCONSOLE_WARN("Remounted %s %s\n", profilenm,
2184                                       read_only ?  "read-only" : "read-write");
2185         }
2186         return 0;
2187 }
2188
2189 /**
2190  * Cleanup the open handle that is cached on MDT-side.
2191  *
2192  * For open case, the client side open handling thread may hit error
2193  * after the MDT grant the open. Under such case, the client should
2194  * send close RPC to the MDT as cleanup; otherwise, the open handle
2195  * on the MDT will be leaked there until the client umount or evicted.
2196  *
2197  * In further, if someone unlinked the file, because the open handle
2198  * holds the reference on such file/object, then it will block the
2199  * subsequent threads that want to locate such object via FID.
2200  *
2201  * \param[in] sb        super block for this file-system
2202  * \param[in] open_req  pointer to the original open request
2203  */
2204 void ll_open_cleanup(struct super_block *sb, struct ptlrpc_request *open_req)
2205 {
2206         struct mdt_body                 *body;
2207         struct md_op_data               *op_data;
2208         struct ptlrpc_request           *close_req = NULL;
2209         struct obd_export               *exp       = ll_s2sbi(sb)->ll_md_exp;
2210         ENTRY;
2211
2212         body = req_capsule_server_get(&open_req->rq_pill, &RMF_MDT_BODY);
2213         OBD_ALLOC_PTR(op_data);
2214         if (op_data == NULL) {
2215                 CWARN("%s: cannot allocate op_data to release open handle for "
2216                       DFID"\n",
2217                       ll_get_fsname(sb, NULL, 0), PFID(&body->mbo_fid1));
2218
2219                 RETURN_EXIT;
2220         }
2221
2222         op_data->op_fid1 = body->mbo_fid1;
2223         op_data->op_handle = body->mbo_handle;
2224         op_data->op_mod_time = cfs_time_current_sec();
2225         md_close(exp, op_data, NULL, &close_req);
2226         ptlrpc_req_finished(close_req);
2227         ll_finish_md_op_data(op_data);
2228
2229         EXIT;
2230 }
2231
2232 int ll_prep_inode(struct inode **inode, struct ptlrpc_request *req,
2233                   struct super_block *sb, struct lookup_intent *it)
2234 {
2235         struct ll_sb_info *sbi = NULL;
2236         struct lustre_md md = { NULL };
2237         int rc;
2238         ENTRY;
2239
2240         LASSERT(*inode || sb);
2241         sbi = sb ? ll_s2sbi(sb) : ll_i2sbi(*inode);
2242         rc = md_get_lustre_md(sbi->ll_md_exp, req, sbi->ll_dt_exp,
2243                               sbi->ll_md_exp, &md);
2244         if (rc != 0)
2245                 GOTO(cleanup, rc);
2246
2247         if (*inode) {
2248                 rc = ll_update_inode(*inode, &md);
2249                 if (rc != 0)
2250                         GOTO(out, rc);
2251         } else {
2252                 LASSERT(sb != NULL);
2253
2254                 /*
2255                  * At this point server returns to client's same fid as client
2256                  * generated for creating. So using ->fid1 is okay here.
2257                  */
2258                 if (!fid_is_sane(&md.body->mbo_fid1)) {
2259                         CERROR("%s: Fid is insane "DFID"\n",
2260                                 ll_get_fsname(sb, NULL, 0),
2261                                 PFID(&md.body->mbo_fid1));
2262                         GOTO(out, rc = -EINVAL);
2263                 }
2264
2265                 *inode = ll_iget(sb, cl_fid_build_ino(&md.body->mbo_fid1,
2266                                              sbi->ll_flags & LL_SBI_32BIT_API),
2267                                  &md);
2268                 if (IS_ERR(*inode)) {
2269 #ifdef CONFIG_FS_POSIX_ACL
2270                         if (md.posix_acl) {
2271                                 posix_acl_release(md.posix_acl);
2272                                 md.posix_acl = NULL;
2273                         }
2274 #endif
2275                         rc = IS_ERR(*inode) ? PTR_ERR(*inode) : -ENOMEM;
2276                         *inode = NULL;
2277                         CERROR("new_inode -fatal: rc %d\n", rc);
2278                         GOTO(out, rc);
2279                 }
2280         }
2281
2282         /* Handling piggyback layout lock.
2283          * Layout lock can be piggybacked by getattr and open request.
2284          * The lsm can be applied to inode only if it comes with a layout lock
2285          * otherwise correct layout may be overwritten, for example:
2286          * 1. proc1: mdt returns a lsm but not granting layout
2287          * 2. layout was changed by another client
2288          * 3. proc2: refresh layout and layout lock granted
2289          * 4. proc1: to apply a stale layout */
2290         if (it != NULL && it->it_lock_mode != 0) {
2291                 struct lustre_handle lockh;
2292                 struct ldlm_lock *lock;
2293
2294                 lockh.cookie = it->it_lock_handle;
2295                 lock = ldlm_handle2lock(&lockh);
2296                 LASSERT(lock != NULL);
2297                 if (ldlm_has_layout(lock)) {
2298                         struct cl_object_conf conf;
2299
2300                         memset(&conf, 0, sizeof(conf));
2301                         conf.coc_opc = OBJECT_CONF_SET;
2302                         conf.coc_inode = *inode;
2303                         conf.coc_lock = lock;
2304                         conf.u.coc_layout = md.layout;
2305                         (void)ll_layout_conf(*inode, &conf);
2306                 }
2307                 LDLM_LOCK_PUT(lock);
2308         }
2309
2310         GOTO(out, rc = 0);
2311
2312 out:
2313         md_free_lustre_md(sbi->ll_md_exp, &md);
2314
2315 cleanup:
2316         if (rc != 0 && it != NULL && it->it_op & IT_OPEN)
2317                 ll_open_cleanup(sb != NULL ? sb : (*inode)->i_sb, req);
2318
2319         return rc;
2320 }
2321
2322 int ll_obd_statfs(struct inode *inode, void __user *arg)
2323 {
2324         struct ll_sb_info *sbi = NULL;
2325         struct obd_export *exp;
2326         char *buf = NULL;
2327         struct obd_ioctl_data *data = NULL;
2328         __u32 type;
2329         int len = 0, rc;
2330
2331         if (!inode || !(sbi = ll_i2sbi(inode)))
2332                 GOTO(out_statfs, rc = -EINVAL);
2333
2334         rc = obd_ioctl_getdata(&buf, &len, arg);
2335         if (rc)
2336                 GOTO(out_statfs, rc);
2337
2338         data = (void*)buf;
2339         if (!data->ioc_inlbuf1 || !data->ioc_inlbuf2 ||
2340             !data->ioc_pbuf1 || !data->ioc_pbuf2)
2341                 GOTO(out_statfs, rc = -EINVAL);
2342
2343         if (data->ioc_inllen1 != sizeof(__u32) ||
2344             data->ioc_inllen2 != sizeof(__u32) ||
2345             data->ioc_plen1 != sizeof(struct obd_statfs) ||
2346             data->ioc_plen2 != sizeof(struct obd_uuid))
2347                 GOTO(out_statfs, rc = -EINVAL);
2348
2349         memcpy(&type, data->ioc_inlbuf1, sizeof(__u32));
2350         if (type & LL_STATFS_LMV)
2351                 exp = sbi->ll_md_exp;
2352         else if (type & LL_STATFS_LOV)
2353                 exp = sbi->ll_dt_exp;
2354         else
2355                 GOTO(out_statfs, rc = -ENODEV);
2356
2357         rc = obd_iocontrol(IOC_OBD_STATFS, exp, len, buf, NULL);
2358         if (rc)
2359                 GOTO(out_statfs, rc);
2360 out_statfs:
2361         if (buf)
2362                 obd_ioctl_freedata(buf, len);
2363         return rc;
2364 }
2365
2366 int ll_process_config(struct lustre_cfg *lcfg)
2367 {
2368         struct super_block *sb;
2369         unsigned long x;
2370         int rc = 0;
2371         char *ptr;
2372
2373         /* The instance name contains the sb: lustre-client-aacfe000 */
2374         ptr = strrchr(lustre_cfg_string(lcfg, 0), '-');
2375         if (!ptr || !*(++ptr))
2376                 return -EINVAL;
2377         if (sscanf(ptr, "%lx", &x) != 1)
2378                 return -EINVAL;
2379         sb = (struct super_block *)x;
2380         /* This better be a real Lustre superblock! */
2381         LASSERT(s2lsi(sb)->lsi_lmd->lmd_magic == LMD_MAGIC);
2382
2383         /* Note we have not called client_common_fill_super yet, so
2384            proc fns must be able to handle that! */
2385         rc = class_process_proc_param(PARAM_LLITE, lprocfs_llite_obd_vars,
2386                                       lcfg, sb);
2387         if (rc > 0)
2388                 rc = 0;
2389         return rc;
2390 }
2391
2392 /* this function prepares md_op_data hint for passing it down to MD stack. */
2393 struct md_op_data *ll_prep_md_op_data(struct md_op_data *op_data,
2394                                       struct inode *i1, struct inode *i2,
2395                                       const char *name, size_t namelen,
2396                                       __u32 mode, __u32 opc, void *data)
2397 {
2398         LASSERT(i1 != NULL);
2399
2400         if (name == NULL) {
2401                 /* Do not reuse namelen for something else. */
2402                 if (namelen != 0)
2403                         return ERR_PTR(-EINVAL);
2404         } else {
2405                 if (namelen > ll_i2sbi(i1)->ll_namelen)
2406                         return ERR_PTR(-ENAMETOOLONG);
2407
2408                 if (!lu_name_is_valid_2(name, namelen))
2409                         return ERR_PTR(-EINVAL);
2410         }
2411
2412         if (op_data == NULL)
2413                 OBD_ALLOC_PTR(op_data);
2414
2415         if (op_data == NULL)
2416                 return ERR_PTR(-ENOMEM);
2417
2418         ll_i2gids(op_data->op_suppgids, i1, i2);
2419         op_data->op_fid1 = *ll_inode2fid(i1);
2420         op_data->op_default_stripe_offset = -1;
2421         if (S_ISDIR(i1->i_mode)) {
2422                 op_data->op_mea1 = ll_i2info(i1)->lli_lsm_md;
2423                 if (opc == LUSTRE_OPC_MKDIR)
2424                         op_data->op_default_stripe_offset =
2425                                    ll_i2info(i1)->lli_def_stripe_offset;
2426         }
2427
2428         if (i2) {
2429                 op_data->op_fid2 = *ll_inode2fid(i2);
2430                 if (S_ISDIR(i2->i_mode))
2431                         op_data->op_mea2 = ll_i2info(i2)->lli_lsm_md;
2432         } else {
2433                 fid_zero(&op_data->op_fid2);
2434         }
2435
2436         if (ll_i2sbi(i1)->ll_flags & LL_SBI_64BIT_HASH)
2437                 op_data->op_cli_flags |= CLI_HASH64;
2438
2439         if (ll_need_32bit_api(ll_i2sbi(i1)))
2440                 op_data->op_cli_flags |= CLI_API32;
2441
2442         op_data->op_name = name;
2443         op_data->op_namelen = namelen;
2444         op_data->op_mode = mode;
2445         op_data->op_mod_time = cfs_time_current_sec();
2446         op_data->op_fsuid = from_kuid(&init_user_ns, current_fsuid());
2447         op_data->op_fsgid = from_kgid(&init_user_ns, current_fsgid());
2448         op_data->op_cap = cfs_curproc_cap_pack();
2449         if ((opc == LUSTRE_OPC_CREATE) && (name != NULL) &&
2450              filename_is_volatile(name, namelen, &op_data->op_mds)) {
2451                 op_data->op_bias |= MDS_CREATE_VOLATILE;
2452         } else {
2453                 op_data->op_mds = 0;
2454         }
2455         op_data->op_data = data;
2456
2457         return op_data;
2458 }
2459
2460 void ll_finish_md_op_data(struct md_op_data *op_data)
2461 {
2462         OBD_FREE_PTR(op_data);
2463 }
2464
2465 #ifdef HAVE_SUPEROPS_USE_DENTRY
2466 int ll_show_options(struct seq_file *seq, struct dentry *dentry)
2467 #else
2468 int ll_show_options(struct seq_file *seq, struct vfsmount *vfs)
2469 #endif
2470 {
2471         struct ll_sb_info *sbi;
2472
2473 #ifdef HAVE_SUPEROPS_USE_DENTRY
2474         LASSERT((seq != NULL) && (dentry != NULL));
2475         sbi = ll_s2sbi(dentry->d_sb);
2476 #else
2477         LASSERT((seq != NULL) && (vfs != NULL));
2478         sbi = ll_s2sbi(vfs->mnt_sb);
2479 #endif
2480
2481         if (sbi->ll_flags & LL_SBI_NOLCK)
2482                 seq_puts(seq, ",nolock");
2483
2484         if (sbi->ll_flags & LL_SBI_FLOCK)
2485                 seq_puts(seq, ",flock");
2486
2487         if (sbi->ll_flags & LL_SBI_LOCALFLOCK)
2488                 seq_puts(seq, ",localflock");
2489
2490         if (sbi->ll_flags & LL_SBI_USER_XATTR)
2491                 seq_puts(seq, ",user_xattr");
2492
2493         if (sbi->ll_flags & LL_SBI_LAZYSTATFS)
2494                 seq_puts(seq, ",lazystatfs");
2495
2496         if (sbi->ll_flags & LL_SBI_USER_FID2PATH)
2497                 seq_puts(seq, ",user_fid2path");
2498
2499         if (sbi->ll_flags & LL_SBI_ALWAYS_PING)
2500                 seq_puts(seq, ",always_ping");
2501
2502         RETURN(0);
2503 }
2504
2505 /**
2506  * Get obd name by cmd, and copy out to user space
2507  */
2508 int ll_get_obd_name(struct inode *inode, unsigned int cmd, unsigned long arg)
2509 {
2510         struct ll_sb_info *sbi = ll_i2sbi(inode);
2511         struct obd_device *obd;
2512         ENTRY;
2513
2514         if (cmd == OBD_IOC_GETDTNAME)
2515                 obd = class_exp2obd(sbi->ll_dt_exp);
2516         else if (cmd == OBD_IOC_GETMDNAME)
2517                 obd = class_exp2obd(sbi->ll_md_exp);
2518         else
2519                 RETURN(-EINVAL);
2520
2521         if (!obd)
2522                 RETURN(-ENOENT);
2523
2524         if (copy_to_user((void __user *)arg, obd->obd_name,
2525                          strlen(obd->obd_name) + 1))
2526                 RETURN(-EFAULT);
2527
2528         RETURN(0);
2529 }
2530
2531 /**
2532  * Get lustre file system name by \a sbi. If \a buf is provided(non-NULL), the
2533  * fsname will be returned in this buffer; otherwise, a static buffer will be
2534  * used to store the fsname and returned to caller.
2535  */
2536 char *ll_get_fsname(struct super_block *sb, char *buf, int buflen)
2537 {
2538         static char fsname_static[MTI_NAME_MAXLEN];
2539         struct lustre_sb_info *lsi = s2lsi(sb);
2540         char *ptr;
2541         int len;
2542
2543         if (buf == NULL) {
2544                 /* this means the caller wants to use static buffer
2545                  * and it doesn't care about race. Usually this is
2546                  * in error reporting path */
2547                 buf = fsname_static;
2548                 buflen = sizeof(fsname_static);
2549         }
2550
2551         len = strlen(lsi->lsi_lmd->lmd_profile);
2552         ptr = strrchr(lsi->lsi_lmd->lmd_profile, '-');
2553         if (ptr && (strcmp(ptr, "-client") == 0))
2554                 len -= 7;
2555
2556         if (unlikely(len >= buflen))
2557                 len = buflen - 1;
2558         strncpy(buf, lsi->lsi_lmd->lmd_profile, len);
2559         buf[len] = '\0';
2560
2561         return buf;
2562 }
2563
2564 static char* ll_d_path(struct dentry *dentry, char *buf, int bufsize)
2565 {
2566         char *path = NULL;
2567
2568         struct path p;
2569
2570         p.dentry = dentry;
2571         p.mnt = current->fs->root.mnt;
2572         path_get(&p);
2573         path = d_path(&p, buf, bufsize);
2574         path_put(&p);
2575         return path;
2576 }
2577
2578 void ll_dirty_page_discard_warn(struct page *page, int ioret)
2579 {
2580         char *buf, *path = NULL;
2581         struct dentry *dentry = NULL;
2582         struct inode *inode = page->mapping->host;
2583
2584         /* this can be called inside spin lock so use GFP_ATOMIC. */
2585         buf = (char *)__get_free_page(GFP_ATOMIC);
2586         if (buf != NULL) {
2587                 dentry = d_find_alias(page->mapping->host);
2588                 if (dentry != NULL)
2589                         path = ll_d_path(dentry, buf, PAGE_SIZE);
2590         }
2591
2592         CDEBUG(D_WARNING,
2593                "%s: dirty page discard: %s/fid: "DFID"/%s may get corrupted "
2594                "(rc %d)\n", ll_get_fsname(page->mapping->host->i_sb, NULL, 0),
2595                s2lsi(page->mapping->host->i_sb)->lsi_lmd->lmd_dev,
2596                PFID(ll_inode2fid(inode)),
2597                (path && !IS_ERR(path)) ? path : "", ioret);
2598
2599         if (dentry != NULL)
2600                 dput(dentry);
2601
2602         if (buf != NULL)
2603                 free_page((unsigned long)buf);
2604 }
2605
2606 ssize_t ll_copy_user_md(const struct lov_user_md __user *md,
2607                         struct lov_user_md **kbuf)
2608 {
2609         struct lov_user_md      lum;
2610         ssize_t                 lum_size;
2611         ENTRY;
2612
2613         if (copy_from_user(&lum, md, sizeof(lum)))
2614                 RETURN(-EFAULT);
2615
2616         lum_size = ll_lov_user_md_size(&lum);
2617         if (lum_size < 0)
2618                 RETURN(lum_size);
2619
2620         OBD_ALLOC(*kbuf, lum_size);
2621         if (*kbuf == NULL)
2622                 RETURN(-ENOMEM);
2623
2624         if (copy_from_user(*kbuf, md, lum_size) != 0) {
2625                 OBD_FREE(*kbuf, lum_size);
2626                 RETURN(-EFAULT);
2627         }
2628
2629         RETURN(lum_size);
2630 }
2631
2632 /*
2633  * Compute llite root squash state after a change of root squash
2634  * configuration setting or add/remove of a lnet nid
2635  */
2636 void ll_compute_rootsquash_state(struct ll_sb_info *sbi)
2637 {
2638         struct root_squash_info *squash = &sbi->ll_squash;
2639         int i;
2640         bool matched;
2641         lnet_process_id_t id;
2642
2643         /* Update norootsquash flag */
2644         down_write(&squash->rsi_sem);
2645         if (list_empty(&squash->rsi_nosquash_nids))
2646                 sbi->ll_flags &= ~LL_SBI_NOROOTSQUASH;
2647         else {
2648                 /* Do not apply root squash as soon as one of our NIDs is
2649                  * in the nosquash_nids list */
2650                 matched = false;
2651                 i = 0;
2652                 while (LNetGetId(i++, &id) != -ENOENT) {
2653                         if (LNET_NETTYP(LNET_NIDNET(id.nid)) == LOLND)
2654                                 continue;
2655                         if (cfs_match_nid(id.nid, &squash->rsi_nosquash_nids)) {
2656                                 matched = true;
2657                                 break;
2658                         }
2659                 }
2660                 if (matched)
2661                         sbi->ll_flags |= LL_SBI_NOROOTSQUASH;
2662                 else
2663                         sbi->ll_flags &= ~LL_SBI_NOROOTSQUASH;
2664         }
2665         up_write(&squash->rsi_sem);
2666 }
2667
2668 /**
2669  * Parse linkea content to extract information about a given hardlink
2670  *
2671  * \param[in]   ldata      - Initialized linkea data
2672  * \param[in]   linkno     - Link identifier
2673  * \param[out]  parent_fid - The entry's parent FID
2674  * \param[out]  ln         - Entry name destination buffer
2675  *
2676  * \retval 0 on success
2677  * \retval Appropriate negative error code on failure
2678  */
2679 static int ll_linkea_decode(struct linkea_data *ldata, unsigned int linkno,
2680                             struct lu_fid *parent_fid, struct lu_name *ln)
2681 {
2682         unsigned int    idx;
2683         int             rc;
2684         ENTRY;
2685
2686         rc = linkea_init(ldata);
2687         if (rc < 0)
2688                 RETURN(rc);
2689
2690         if (linkno >= ldata->ld_leh->leh_reccount)
2691                 /* beyond last link */
2692                 RETURN(-ENODATA);
2693
2694         linkea_first_entry(ldata);
2695         for (idx = 0; ldata->ld_lee != NULL; idx++) {
2696                 linkea_entry_unpack(ldata->ld_lee, &ldata->ld_reclen, ln,
2697                                     parent_fid);
2698                 if (idx == linkno)
2699                         break;
2700
2701                 linkea_next_entry(ldata);
2702         }
2703
2704         if (idx < linkno)
2705                 RETURN(-ENODATA);
2706
2707         RETURN(0);
2708 }
2709
2710 /**
2711  * Get parent FID and name of an identified link. Operation is performed for
2712  * a given link number, letting the caller iterate over linkno to list one or
2713  * all links of an entry.
2714  *
2715  * \param[in]     file - File descriptor against which to perform the operation
2716  * \param[in,out] arg  - User-filled structure containing the linkno to operate
2717  *                       on and the available size. It is eventually filled with
2718  *                       the requested information or left untouched on error
2719  *
2720  * \retval - 0 on success
2721  * \retval - Appropriate negative error code on failure
2722  */
2723 int ll_getparent(struct file *file, struct getparent __user *arg)
2724 {
2725         struct dentry           *dentry = file_dentry(file);
2726         struct inode            *inode = file_inode(file);
2727         struct linkea_data      *ldata;
2728         struct lu_buf            buf = LU_BUF_NULL;
2729         struct lu_name           ln;
2730         struct lu_fid            parent_fid;
2731         __u32                    linkno;
2732         __u32                    name_size;
2733         int                      rc;
2734
2735         ENTRY;
2736
2737         if (!cfs_capable(CFS_CAP_DAC_READ_SEARCH) &&
2738             !(ll_i2sbi(inode)->ll_flags & LL_SBI_USER_FID2PATH))
2739                 RETURN(-EPERM);
2740
2741         if (get_user(name_size, &arg->gp_name_size))
2742                 RETURN(-EFAULT);
2743
2744         if (get_user(linkno, &arg->gp_linkno))
2745                 RETURN(-EFAULT);
2746
2747         if (name_size > PATH_MAX)
2748                 RETURN(-EINVAL);
2749
2750         OBD_ALLOC(ldata, sizeof(*ldata));
2751         if (ldata == NULL)
2752                 RETURN(-ENOMEM);
2753
2754         rc = linkea_data_new(ldata, &buf);
2755         if (rc < 0)
2756                 GOTO(ldata_free, rc);
2757
2758         rc = ll_getxattr(dentry, XATTR_NAME_LINK, buf.lb_buf, buf.lb_len);
2759         if (rc < 0)
2760                 GOTO(lb_free, rc);
2761
2762         rc = ll_linkea_decode(ldata, linkno, &parent_fid, &ln);
2763         if (rc < 0)
2764                 GOTO(lb_free, rc);
2765
2766         if (ln.ln_namelen >= name_size)
2767                 GOTO(lb_free, rc = -EOVERFLOW);
2768
2769         if (copy_to_user(&arg->gp_fid, &parent_fid, sizeof(arg->gp_fid)))
2770                 GOTO(lb_free, rc = -EFAULT);
2771
2772         if (copy_to_user(&arg->gp_name, ln.ln_name, ln.ln_namelen))
2773                 GOTO(lb_free, rc = -EFAULT);
2774
2775         if (put_user('\0', arg->gp_name + ln.ln_namelen))
2776                 GOTO(lb_free, rc = -EFAULT);
2777
2778 lb_free:
2779         lu_buf_free(&buf);
2780 ldata_free:
2781         OBD_FREE(ldata, sizeof(*ldata));
2782
2783         RETURN(rc);
2784 }