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         lli->lli_mds_read_och = NULL;
 905         lli->lli_mds_write_och = NULL;
 906         lli->lli_mds_exec_och = NULL;
 907         lli->lli_open_fd_read_count = 0;
 908         lli->lli_open_fd_write_count = 0;
 909         lli->lli_open_fd_exec_count = 0;
 910         mutex_init(&lli->lli_och_mutex);
 911         spin_lock_init(&lli->lli_agl_lock);
 912         spin_lock_init(&lli->lli_layout_lock);
 913         ll_layout_version_set(lli, CL_LAYOUT_GEN_NONE);
 914         lli->lli_clob = NULL;
 915
 916         init_rwsem(&lli->lli_xattrs_list_rwsem);
 917         mutex_init(&lli->lli_xattrs_enq_lock);
 918
 919         LASSERT(lli->lli_vfs_inode.i_mode != 0);
 920         if (S_ISDIR(lli->lli_vfs_inode.i_mode)) {
 921                 mutex_init(&lli->lli_readdir_mutex);
 922                 lli->lli_opendir_key = NULL;
 923                 lli->lli_sai = NULL;
 924                 spin_lock_init(&lli->lli_sa_lock);
 925                 lli->lli_opendir_pid = 0;
 926                 lli->lli_sa_enabled = 0;
 927                 lli->lli_def_stripe_offset = -1;
 928         } else {
 929                 mutex_init(&lli->lli_size_mutex);
 930                 lli->lli_symlink_name = NULL;
 931                 init_rwsem(&lli->lli_trunc_sem);
 932                 range_lock_tree_init(&lli->lli_write_tree);
 933                 init_rwsem(&lli->lli_glimpse_sem);
 934                 lli->lli_glimpse_time = 0;
 935                 INIT_LIST_HEAD(&lli->lli_agl_list);
 936                 lli->lli_agl_index = 0;
 937                 lli->lli_async_rc = 0;
 938         }
 939         mutex_init(&lli->lli_layout_mutex);
 940 }
 941
 942 static inline int ll_bdi_register(struct backing_dev_info *bdi)
 943 {
 944         static atomic_t ll_bdi_num = ATOMIC_INIT(0);
 945
 946         bdi->name = "lustre";
 947         return bdi_register(bdi, NULL, "lustre-%d",
 948                             atomic_inc_return(&ll_bdi_num));
 949 }
 950
 951 int ll_fill_super(struct super_block *sb, struct vfsmount *mnt)
 952 {
 953         struct lustre_profile *lprof = NULL;
 954         struct lustre_sb_info *lsi = s2lsi(sb);
 955         struct ll_sb_info *sbi;
 956         char  *dt = NULL, *md = NULL;
 957         char  *profilenm = get_profile_name(sb);
 958         struct config_llog_instance *cfg;
 959         /* %p for void* in printf needs 16+2 characters: 0xffffffffffffffff */
 960         const int instlen = sizeof(cfg->cfg_instance) * 2 + 2;
 961         int    err;
 962         ENTRY;
 963
 964         CDEBUG(D_VFSTRACE, "VFS Op: sb %p\n", sb);
 965
 966         OBD_ALLOC_PTR(cfg);
 967         if (cfg == NULL)
 968                 RETURN(-ENOMEM);
 969
 970         try_module_get(THIS_MODULE);
 971
 972         /* client additional sb info */
 973         lsi->lsi_llsbi = sbi = ll_init_sbi();
 974         if (!sbi) {
 975                 module_put(THIS_MODULE);
 976                 OBD_FREE_PTR(cfg);
 977                 RETURN(-ENOMEM);
 978         }
 979
 980         err = ll_options(lsi->lsi_lmd->lmd_opts, &sbi->ll_flags);
 981         if (err)
 982                 GOTO(out_free, err);
 983
 984         err = bdi_init(&lsi->lsi_bdi);
 985         if (err)
 986                 GOTO(out_free, err);
 987         lsi->lsi_flags |= LSI_BDI_INITIALIZED;
 988 #ifdef HAVE_BDI_CAP_MAP_COPY
 989         lsi->lsi_bdi.capabilities = BDI_CAP_MAP_COPY;
 990 #else
 991         lsi->lsi_bdi.capabilities = 0;
 992 #endif
 993         err = ll_bdi_register(&lsi->lsi_bdi);
 994         if (err)
 995                 GOTO(out_free, err);
 996
 997         sb->s_bdi = &lsi->lsi_bdi;
 998 #ifndef HAVE_DCACHE_LOCK
 999         /* kernel >= 2.6.38 store dentry operations in sb->s_d_op. */
1000         sb->s_d_op = &ll_d_ops;
1001 #endif
1002
1003         /* Generate a string unique to this super, in case some joker tries
1004            to mount the same fs at two mount points.
1005            Use the address of the super itself.*/
1006         cfg->cfg_instance = sb;
1007         cfg->cfg_uuid = lsi->lsi_llsbi->ll_sb_uuid;
1008         cfg->cfg_callback = class_config_llog_handler;
1009         /* set up client obds */
1010         err = lustre_process_log(sb, profilenm, cfg);
1011         if (err < 0)
1012                 GOTO(out_free, err);
1013
1014         /* Profile set with LCFG_MOUNTOPT so we can find our mdc and osc obds */
1015         lprof = class_get_profile(profilenm);
1016         if (lprof == NULL) {
1017                 LCONSOLE_ERROR_MSG(0x156, "The client profile '%s' could not be"
1018                                    " read from the MGS.  Does that filesystem "
1019                                    "exist?\n", profilenm);
1020                 GOTO(out_free, err = -EINVAL);
1021         }
1022         CDEBUG(D_CONFIG, "Found profile %s: mdc=%s osc=%s\n", profilenm,
1023                lprof->lp_md, lprof->lp_dt);
1024
1025         OBD_ALLOC(dt, strlen(lprof->lp_dt) + instlen + 2);
1026         if (!dt)
1027                 GOTO(out_free, err = -ENOMEM);
1028         sprintf(dt, "%s-%p", lprof->lp_dt, cfg->cfg_instance);
1029
1030         OBD_ALLOC(md, strlen(lprof->lp_md) + instlen + 2);
1031         if (!md)
1032                 GOTO(out_free, err = -ENOMEM);
1033         sprintf(md, "%s-%p", lprof->lp_md, cfg->cfg_instance);
1034
1035         /* connections, registrations, sb setup */
1036         err = client_common_fill_super(sb, md, dt, mnt);
1037         if (err < 0)
1038                 GOTO(out_free, err);
1039
1040         sbi->ll_client_common_fill_super_succeeded = 1;
1041
1042 out_free:
1043         if (md)
1044                 OBD_FREE(md, strlen(lprof->lp_md) + instlen + 2);
1045         if (dt)
1046                 OBD_FREE(dt, strlen(lprof->lp_dt) + instlen + 2);
1047         if (lprof != NULL)
1048                 class_put_profile(lprof);
1049         if (err)
1050                 ll_put_super(sb);
1051         else if (sbi->ll_flags & LL_SBI_VERBOSE)
1052                 LCONSOLE_WARN("Mounted %s\n", profilenm);
1053
1054         OBD_FREE_PTR(cfg);
1055         RETURN(err);
1056 } /* ll_fill_super */
1057
1058 void ll_put_super(struct super_block *sb)
1059 {
1060         struct config_llog_instance cfg, params_cfg;
1061         struct obd_device *obd;
1062         struct lustre_sb_info *lsi = s2lsi(sb);
1063         struct ll_sb_info *sbi = ll_s2sbi(sb);
1064         char *profilenm = get_profile_name(sb);
1065         long ccc_count;
1066         int next, force = 1, rc = 0;
1067         ENTRY;
1068
1069         CDEBUG(D_VFSTRACE, "VFS Op: sb %p - %s\n", sb, profilenm);
1070
1071         cfg.cfg_instance = sb;
1072         lustre_end_log(sb, profilenm, &cfg);
1073
1074         params_cfg.cfg_instance = sb;
1075         lustre_end_log(sb, PARAMS_FILENAME, &params_cfg);
1076
1077         if (sbi->ll_md_exp) {
1078                 obd = class_exp2obd(sbi->ll_md_exp);
1079                 if (obd)
1080                         force = obd->obd_force;
1081         }
1082
1083         /* Wait for unstable pages to be committed to stable storage */
1084         if (force == 0) {
1085                 struct l_wait_info lwi = LWI_INTR(LWI_ON_SIGNAL_NOOP, NULL);
1086                 rc = l_wait_event(sbi->ll_cache->ccc_unstable_waitq,
1087                         atomic_long_read(&sbi->ll_cache->ccc_unstable_nr) == 0,
1088                         &lwi);
1089         }
1090
1091         ccc_count = atomic_long_read(&sbi->ll_cache->ccc_unstable_nr);
1092         if (force == 0 && rc != -EINTR)
1093                 LASSERTF(ccc_count == 0, "count: %li\n", ccc_count);
1094
1095
1096         /* We need to set force before the lov_disconnect in
1097            lustre_common_put_super, since l_d cleans up osc's as well. */
1098         if (force) {
1099                 next = 0;
1100                 while ((obd = class_devices_in_group(&sbi->ll_sb_uuid,
1101                                                      &next)) != NULL) {
1102                         obd->obd_force = force;
1103                 }
1104         }
1105
1106         if (sbi->ll_client_common_fill_super_succeeded) {
1107                 /* Only if client_common_fill_super succeeded */
1108                 client_common_put_super(sb);
1109         }
1110
1111         next = 0;
1112         while ((obd = class_devices_in_group(&sbi->ll_sb_uuid, &next)) !=NULL) {
1113                 class_manual_cleanup(obd);
1114         }
1115
1116         if (sbi->ll_flags & LL_SBI_VERBOSE)
1117                 LCONSOLE_WARN("Unmounted %s\n", profilenm ? profilenm : "");
1118
1119         if (profilenm)
1120                 class_del_profile(profilenm);
1121
1122         if (lsi->lsi_flags & LSI_BDI_INITIALIZED) {
1123                 bdi_destroy(&lsi->lsi_bdi);
1124                 lsi->lsi_flags &= ~LSI_BDI_INITIALIZED;
1125         }
1126
1127         ll_free_sbi(sb);
1128         lsi->lsi_llsbi = NULL;
1129
1130         lustre_common_put_super(sb);
1131
1132         cl_env_cache_purge(~0);
1133
1134         module_put(THIS_MODULE);
1135
1136         EXIT;
1137 } /* client_put_super */
1138
1139 struct inode *ll_inode_from_resource_lock(struct ldlm_lock *lock)
1140 {
1141         struct inode *inode = NULL;
1142
1143         /* NOTE: we depend on atomic igrab() -bzzz */
1144         lock_res_and_lock(lock);
1145         if (lock->l_resource->lr_lvb_inode) {
1146                 struct ll_inode_info * lli;
1147                 lli = ll_i2info(lock->l_resource->lr_lvb_inode);
1148                 if (lli->lli_inode_magic == LLI_INODE_MAGIC) {
1149                         inode = igrab(lock->l_resource->lr_lvb_inode);
1150                 } else {
1151                         inode = lock->l_resource->lr_lvb_inode;
1152                         LDLM_DEBUG_LIMIT(inode->i_state & I_FREEING ?  D_INFO :
1153                                          D_WARNING, lock, "lr_lvb_inode %p is "
1154                                          "bogus: magic %08x",
1155                                          lock->l_resource->lr_lvb_inode,
1156                                          lli->lli_inode_magic);
1157                         inode = NULL;
1158                 }
1159         }
1160         unlock_res_and_lock(lock);
1161         return inode;
1162 }
1163
1164 void ll_dir_clear_lsm_md(struct inode *inode)
1165 {
1166         struct ll_inode_info *lli = ll_i2info(inode);
1167
1168         LASSERT(S_ISDIR(inode->i_mode));
1169
1170         if (lli->lli_lsm_md != NULL) {
1171                 lmv_free_memmd(lli->lli_lsm_md);
1172                 lli->lli_lsm_md = NULL;
1173         }
1174 }
1175
1176 static struct inode *ll_iget_anon_dir(struct super_block *sb,
1177                                       const struct lu_fid *fid,
1178                                       struct lustre_md *md)
1179 {
1180         struct ll_sb_info       *sbi = ll_s2sbi(sb);
1181         struct mdt_body         *body = md->body;
1182         struct inode            *inode;
1183         ino_t                   ino;
1184         ENTRY;
1185
1186         ino = cl_fid_build_ino(fid, sbi->ll_flags & LL_SBI_32BIT_API);
1187         inode = iget_locked(sb, ino);
1188         if (inode == NULL) {
1189                 CERROR("%s: failed get simple inode "DFID": rc = -ENOENT\n",
1190                        ll_get_fsname(sb, NULL, 0), PFID(fid));
1191                 RETURN(ERR_PTR(-ENOENT));
1192         }
1193
1194         if (inode->i_state & I_NEW) {
1195                 struct ll_inode_info *lli = ll_i2info(inode);
1196                 struct lmv_stripe_md *lsm = md->lmv;
1197
1198                 inode->i_mode = (inode->i_mode & ~S_IFMT) |
1199                                 (body->mbo_mode & S_IFMT);
1200                 LASSERTF(S_ISDIR(inode->i_mode), "Not slave inode "DFID"\n",
1201                          PFID(fid));
1202
1203                 LTIME_S(inode->i_mtime) = 0;
1204                 LTIME_S(inode->i_atime) = 0;
1205                 LTIME_S(inode->i_ctime) = 0;
1206                 inode->i_rdev = 0;
1207
1208 #ifdef HAVE_BACKING_DEV_INFO
1209                 /* initializing backing dev info. */
1210                 inode->i_mapping->backing_dev_info =
1211                                                 &s2lsi(inode->i_sb)->lsi_bdi;
1212 #endif
1213                 inode->i_op = &ll_dir_inode_operations;
1214                 inode->i_fop = &ll_dir_operations;
1215                 lli->lli_fid = *fid;
1216                 ll_lli_init(lli);
1217
1218                 LASSERT(lsm != NULL);
1219                 /* master object FID */
1220                 lli->lli_pfid = body->mbo_fid1;
1221                 CDEBUG(D_INODE, "lli %p slave "DFID" master "DFID"\n",
1222                        lli, PFID(fid), PFID(&lli->lli_pfid));
1223                 unlock_new_inode(inode);
1224         }
1225
1226         RETURN(inode);
1227 }
1228
1229 static int ll_init_lsm_md(struct inode *inode, struct lustre_md *md)
1230 {
1231         struct lu_fid *fid;
1232         struct lmv_stripe_md *lsm = md->lmv;
1233         int i;
1234
1235         LASSERT(lsm != NULL);
1236         /* XXX sigh, this lsm_root initialization should be in
1237          * LMV layer, but it needs ll_iget right now, so we
1238          * put this here right now. */
1239         for (i = 0; i < lsm->lsm_md_stripe_count; i++) {
1240                 fid = &lsm->lsm_md_oinfo[i].lmo_fid;
1241                 LASSERT(lsm->lsm_md_oinfo[i].lmo_root == NULL);
1242                 /* Unfortunately ll_iget will call ll_update_inode,
1243                  * where the initialization of slave inode is slightly
1244                  * different, so it reset lsm_md to NULL to avoid
1245                  * initializing lsm for slave inode. */
1246                 /* For migrating inode, master stripe and master object will
1247                  * be same, so we only need assign this inode */
1248                 if (lsm->lsm_md_hash_type & LMV_HASH_FLAG_MIGRATION && i == 0)
1249                         lsm->lsm_md_oinfo[i].lmo_root = inode;
1250                 else
1251                         lsm->lsm_md_oinfo[i].lmo_root =
1252                                 ll_iget_anon_dir(inode->i_sb, fid, md);
1253
1254                 if (IS_ERR(lsm->lsm_md_oinfo[i].lmo_root)) {
1255                         int rc = PTR_ERR(lsm->lsm_md_oinfo[i].lmo_root);
1256
1257                         lsm->lsm_md_oinfo[i].lmo_root = NULL;
1258                         return rc;
1259                 }
1260         }
1261
1262         return 0;
1263 }
1264
1265 static inline int lli_lsm_md_eq(const struct lmv_stripe_md *lsm_md1,
1266                                 const struct lmv_stripe_md *lsm_md2)
1267 {
1268         return lsm_md1->lsm_md_magic == lsm_md2->lsm_md_magic &&
1269                lsm_md1->lsm_md_stripe_count == lsm_md2->lsm_md_stripe_count &&
1270                lsm_md1->lsm_md_master_mdt_index ==
1271                                         lsm_md2->lsm_md_master_mdt_index &&
1272                lsm_md1->lsm_md_hash_type == lsm_md2->lsm_md_hash_type &&
1273                lsm_md1->lsm_md_layout_version ==
1274                                         lsm_md2->lsm_md_layout_version &&
1275                strcmp(lsm_md1->lsm_md_pool_name,
1276                       lsm_md2->lsm_md_pool_name) == 0;
1277 }
1278
1279 static int ll_update_lsm_md(struct inode *inode, struct lustre_md *md)
1280 {
1281         struct ll_inode_info *lli = ll_i2info(inode);
1282         struct lmv_stripe_md *lsm = md->lmv;
1283         int     rc;
1284         ENTRY;
1285
1286         LASSERT(S_ISDIR(inode->i_mode));
1287         CDEBUG(D_INODE, "update lsm %p of "DFID"\n", lli->lli_lsm_md,
1288                PFID(ll_inode2fid(inode)));
1289
1290         /* no striped information from request. */
1291         if (lsm == NULL) {
1292                 if (lli->lli_lsm_md == NULL) {
1293                         RETURN(0);
1294                 } else if (lli->lli_lsm_md->lsm_md_hash_type &
1295                                                 LMV_HASH_FLAG_MIGRATION) {
1296                         /* migration is done, the temporay MIGRATE layout has
1297                          * been removed */
1298                         CDEBUG(D_INODE, DFID" finish migration.\n",
1299                                PFID(ll_inode2fid(inode)));
1300                         lmv_free_memmd(lli->lli_lsm_md);
1301                         lli->lli_lsm_md = NULL;
1302                         RETURN(0);
1303                 } else {
1304                         /* The lustre_md from req does not include stripeEA,
1305                          * see ll_md_setattr */
1306                         RETURN(0);
1307                 }
1308         }
1309
1310         /* set the directory layout */
1311         if (lli->lli_lsm_md == NULL) {
1312                 struct cl_attr  *attr;
1313
1314                 rc = ll_init_lsm_md(inode, md);
1315                 if (rc != 0)
1316                         RETURN(rc);
1317
1318                 /* set md->lmv to NULL, so the following free lustre_md
1319                  * will not free this lsm */
1320                 md->lmv = NULL;
1321                 lli->lli_lsm_md = lsm;
1322
1323                 OBD_ALLOC_PTR(attr);
1324                 if (attr == NULL)
1325                         RETURN(-ENOMEM);
1326
1327                 /* validate the lsm */
1328                 rc = md_merge_attr(ll_i2mdexp(inode), lsm, attr,
1329                                    ll_md_blocking_ast);
1330                 if (rc != 0) {
1331                         OBD_FREE_PTR(attr);
1332                         RETURN(rc);
1333                 }
1334
1335                 if (md->body->mbo_valid & OBD_MD_FLNLINK)
1336                         md->body->mbo_nlink = attr->cat_nlink;
1337                 if (md->body->mbo_valid & OBD_MD_FLSIZE)
1338                         md->body->mbo_size = attr->cat_size;
1339                 if (md->body->mbo_valid & OBD_MD_FLATIME)
1340                         md->body->mbo_atime = attr->cat_atime;
1341                 if (md->body->mbo_valid & OBD_MD_FLCTIME)
1342                         md->body->mbo_ctime = attr->cat_ctime;
1343                 if (md->body->mbo_valid & OBD_MD_FLMTIME)
1344                         md->body->mbo_mtime = attr->cat_mtime;
1345
1346                 OBD_FREE_PTR(attr);
1347
1348                 CDEBUG(D_INODE, "Set lsm %p magic %x to "DFID"\n", lsm,
1349                        lsm->lsm_md_magic, PFID(ll_inode2fid(inode)));
1350                 RETURN(0);
1351         }
1352
1353         /* Compare the old and new stripe information */
1354         if (!lsm_md_eq(lli->lli_lsm_md, lsm)) {
1355                 struct lmv_stripe_md    *old_lsm = lli->lli_lsm_md;
1356                 int                     idx;
1357
1358                 CERROR("%s: inode "DFID"(%p)'s lmv layout mismatch (%p)/(%p)"
1359                        "magic:0x%x/0x%x stripe count: %d/%d master_mdt: %d/%d"
1360                        "hash_type:0x%x/0x%x layout: 0x%x/0x%x pool:%s/%s\n",
1361                        ll_get_fsname(inode->i_sb, NULL, 0), PFID(&lli->lli_fid),
1362                        inode, lsm, old_lsm,
1363                        lsm->lsm_md_magic, old_lsm->lsm_md_magic,
1364                        lsm->lsm_md_stripe_count,
1365                        old_lsm->lsm_md_stripe_count,
1366                        lsm->lsm_md_master_mdt_index,
1367                        old_lsm->lsm_md_master_mdt_index,
1368                        lsm->lsm_md_hash_type, old_lsm->lsm_md_hash_type,
1369                        lsm->lsm_md_layout_version,
1370                        old_lsm->lsm_md_layout_version,
1371                        lsm->lsm_md_pool_name,
1372                        old_lsm->lsm_md_pool_name);
1373
1374                 for (idx = 0; idx < old_lsm->lsm_md_stripe_count; idx++) {
1375                         CERROR("%s: sub FIDs in old lsm idx %d, old: "DFID"\n",
1376                                ll_get_fsname(inode->i_sb, NULL, 0), idx,
1377                                PFID(&old_lsm->lsm_md_oinfo[idx].lmo_fid));
1378                 }
1379
1380                 for (idx = 0; idx < lsm->lsm_md_stripe_count; idx++) {
1381                         CERROR("%s: sub FIDs in new lsm idx %d, new: "DFID"\n",
1382                                ll_get_fsname(inode->i_sb, NULL, 0), idx,
1383                                PFID(&lsm->lsm_md_oinfo[idx].lmo_fid));
1384                 }
1385
1386                 RETURN(-EIO);
1387         }
1388
1389         RETURN(0);
1390 }
1391
1392 void ll_clear_inode(struct inode *inode)
1393 {
1394         struct ll_inode_info *lli = ll_i2info(inode);
1395         struct ll_sb_info *sbi = ll_i2sbi(inode);
1396         ENTRY;
1397
1398         CDEBUG(D_VFSTRACE, "VFS Op:inode="DFID"(%p)\n",
1399                PFID(ll_inode2fid(inode)), inode);
1400
1401         if (S_ISDIR(inode->i_mode)) {
1402                 /* these should have been cleared in ll_file_release */
1403                 LASSERT(lli->lli_opendir_key == NULL);
1404                 LASSERT(lli->lli_sai == NULL);
1405                 LASSERT(lli->lli_opendir_pid == 0);
1406         }
1407
1408         md_null_inode(sbi->ll_md_exp, ll_inode2fid(inode));
1409
1410         LASSERT(!lli->lli_open_fd_write_count);
1411         LASSERT(!lli->lli_open_fd_read_count);
1412         LASSERT(!lli->lli_open_fd_exec_count);
1413
1414         if (lli->lli_mds_write_och)
1415                 ll_md_real_close(inode, FMODE_WRITE);
1416         if (lli->lli_mds_exec_och)
1417                 ll_md_real_close(inode, FMODE_EXEC);
1418         if (lli->lli_mds_read_och)
1419                 ll_md_real_close(inode, FMODE_READ);
1420
1421         if (S_ISLNK(inode->i_mode) && lli->lli_symlink_name) {
1422                 OBD_FREE(lli->lli_symlink_name,
1423                          strlen(lli->lli_symlink_name) + 1);
1424                 lli->lli_symlink_name = NULL;
1425         }
1426
1427         ll_xattr_cache_destroy(inode);
1428
1429 #ifdef CONFIG_FS_POSIX_ACL
1430         if (lli->lli_posix_acl) {
1431                 LASSERT(atomic_read(&lli->lli_posix_acl->a_refcount) == 1);
1432                 posix_acl_release(lli->lli_posix_acl);
1433                 lli->lli_posix_acl = NULL;
1434         }
1435 #endif
1436         lli->lli_inode_magic = LLI_INODE_DEAD;
1437
1438         if (S_ISDIR(inode->i_mode))
1439                 ll_dir_clear_lsm_md(inode);
1440         else if (S_ISREG(inode->i_mode) && !is_bad_inode(inode))
1441                 LASSERT(list_empty(&lli->lli_agl_list));
1442
1443         /*
1444          * XXX This has to be done before lsm is freed below, because
1445          * cl_object still uses inode lsm.
1446          */
1447         cl_inode_fini(inode);
1448
1449         EXIT;
1450 }
1451
1452 static int ll_md_setattr(struct dentry *dentry, struct md_op_data *op_data)
1453 {
1454         struct lustre_md md;
1455         struct inode *inode = dentry->d_inode;
1456         struct ll_sb_info *sbi = ll_i2sbi(inode);
1457         struct ptlrpc_request *request = NULL;
1458         int rc, ia_valid;
1459         ENTRY;
1460
1461         op_data = ll_prep_md_op_data(op_data, inode, NULL, NULL, 0, 0,
1462                                      LUSTRE_OPC_ANY, NULL);
1463         if (IS_ERR(op_data))
1464                 RETURN(PTR_ERR(op_data));
1465
1466         rc = md_setattr(sbi->ll_md_exp, op_data, NULL, 0, &request);
1467         if (rc) {
1468                 ptlrpc_req_finished(request);
1469                 if (rc == -ENOENT) {
1470                         clear_nlink(inode);
1471                         /* Unlinked special device node? Or just a race?
1472                          * Pretend we done everything. */
1473                         if (!S_ISREG(inode->i_mode) &&
1474                             !S_ISDIR(inode->i_mode)) {
1475                                 ia_valid = op_data->op_attr.ia_valid;
1476                                 op_data->op_attr.ia_valid &= ~TIMES_SET_FLAGS;
1477                                 rc = simple_setattr(dentry, &op_data->op_attr);
1478                                 op_data->op_attr.ia_valid = ia_valid;
1479                         }
1480                 } else if (rc != -EPERM && rc != -EACCES && rc != -ETXTBSY) {
1481                         CERROR("md_setattr fails: rc = %d\n", rc);
1482                 }
1483                 RETURN(rc);
1484         }
1485
1486         rc = md_get_lustre_md(sbi->ll_md_exp, request, sbi->ll_dt_exp,
1487                               sbi->ll_md_exp, &md);
1488         if (rc) {
1489                 ptlrpc_req_finished(request);
1490                 RETURN(rc);
1491         }
1492
1493         ia_valid = op_data->op_attr.ia_valid;
1494         /* inode size will be in ll_setattr_ost, can't do it now since dirty
1495          * cache is not cleared yet. */
1496         op_data->op_attr.ia_valid &= ~(TIMES_SET_FLAGS | ATTR_SIZE);
1497         if (S_ISREG(inode->i_mode))
1498                 mutex_lock(&inode->i_mutex);
1499         rc = simple_setattr(dentry, &op_data->op_attr);
1500         if (S_ISREG(inode->i_mode))
1501                 mutex_unlock(&inode->i_mutex);
1502         op_data->op_attr.ia_valid = ia_valid;
1503
1504         rc = ll_update_inode(inode, &md);
1505         ptlrpc_req_finished(request);
1506
1507         RETURN(rc);
1508 }
1509
1510 /* If this inode has objects allocated to it (lsm != NULL), then the OST
1511  * object(s) determine the file size and mtime.  Otherwise, the MDS will
1512  * keep these values until such a time that objects are allocated for it.
1513  * We do the MDS operations first, as it is checking permissions for us.
1514  * We don't to the MDS RPC if there is nothing that we want to store there,
1515  * otherwise there is no harm in updating mtime/atime on the MDS if we are
1516  * going to do an RPC anyways.
1517  *
1518  * If we are doing a truncate, we will send the mtime and ctime updates
1519  * to the OST with the punch RPC, otherwise we do an explicit setattr RPC.
1520  * I don't believe it is possible to get e.g. ATTR_MTIME_SET and ATTR_SIZE
1521  * at the same time.
1522  *
1523  * In case of HSMimport, we only set attr on MDS.
1524  */
1525 int ll_setattr_raw(struct dentry *dentry, struct iattr *attr, bool hsm_import)
1526 {
1527         struct inode *inode = dentry->d_inode;
1528         struct ll_inode_info *lli = ll_i2info(inode);
1529         struct md_op_data *op_data = NULL;
1530         int rc = 0;
1531         ENTRY;
1532
1533         CDEBUG(D_VFSTRACE, "%s: setattr inode "DFID"(%p) from %llu to %llu, "
1534                "valid %x, hsm_import %d\n",
1535                ll_get_fsname(inode->i_sb, NULL, 0), PFID(&lli->lli_fid),
1536                inode, i_size_read(inode), attr->ia_size, attr->ia_valid,
1537                hsm_import);
1538
1539         if (attr->ia_valid & ATTR_SIZE) {
1540                 /* Check new size against VFS/VM file size limit and rlimit */
1541                 rc = inode_newsize_ok(inode, attr->ia_size);
1542                 if (rc)
1543                         RETURN(rc);
1544
1545                 /* The maximum Lustre file size is variable, based on the
1546                  * OST maximum object size and number of stripes.  This
1547                  * needs another check in addition to the VFS check above. */
1548                 if (attr->ia_size > ll_file_maxbytes(inode)) {
1549                         CDEBUG(D_INODE,"file "DFID" too large %llu > "LPU64"\n",
1550                                PFID(&lli->lli_fid), attr->ia_size,
1551                                ll_file_maxbytes(inode));
1552                         RETURN(-EFBIG);
1553                 }
1554
1555                 attr->ia_valid |= ATTR_MTIME | ATTR_CTIME;
1556         }
1557
1558         /* POSIX: check before ATTR_*TIME_SET set (from inode_change_ok) */
1559         if (attr->ia_valid & TIMES_SET_FLAGS) {
1560                 if ((!uid_eq(current_fsuid(), inode->i_uid)) &&
1561                     !cfs_capable(CFS_CAP_FOWNER))
1562                         RETURN(-EPERM);
1563         }
1564
1565         /* We mark all of the fields "set" so MDS/OST does not re-set them */
1566         if (!(attr->ia_valid & ATTR_CTIME_SET) &&
1567             (attr->ia_valid & ATTR_CTIME)) {
1568                 attr->ia_ctime = CURRENT_TIME;
1569                 attr->ia_valid |= ATTR_CTIME_SET;
1570         }
1571         if (!(attr->ia_valid & ATTR_ATIME_SET) &&
1572             (attr->ia_valid & ATTR_ATIME)) {
1573                 attr->ia_atime = CURRENT_TIME;
1574                 attr->ia_valid |= ATTR_ATIME_SET;
1575         }
1576         if (!(attr->ia_valid & ATTR_MTIME_SET) &&
1577             (attr->ia_valid & ATTR_MTIME)) {
1578                 attr->ia_mtime = CURRENT_TIME;
1579                 attr->ia_valid |= ATTR_MTIME_SET;
1580         }
1581
1582         if (attr->ia_valid & (ATTR_MTIME | ATTR_CTIME))
1583                 CDEBUG(D_INODE, "setting mtime %lu, ctime %lu, now = %lu\n",
1584                        LTIME_S(attr->ia_mtime), LTIME_S(attr->ia_ctime),
1585                        cfs_time_current_sec());
1586
1587         if (S_ISREG(inode->i_mode)) {
1588                 if (attr->ia_valid & ATTR_SIZE)
1589                         inode_dio_write_done(inode);
1590                 mutex_unlock(&inode->i_mutex);
1591         }
1592
1593         /* We always do an MDS RPC, even if we're only changing the size;
1594          * only the MDS knows whether truncate() should fail with -ETXTBUSY */
1595
1596         OBD_ALLOC_PTR(op_data);
1597         if (op_data == NULL)
1598                 GOTO(out, rc = -ENOMEM);
1599
1600         if (!hsm_import && attr->ia_valid & ATTR_SIZE) {
1601                 /* If we are changing file size, file content is
1602                  * modified, flag it. */
1603                 attr->ia_valid |= MDS_OPEN_OWNEROVERRIDE;
1604                 op_data->op_bias |= MDS_DATA_MODIFIED;
1605                 ll_file_clear_flag(lli, LLIF_DATA_MODIFIED);
1606         }
1607
1608         op_data->op_attr = *attr;
1609
1610         rc = ll_md_setattr(dentry, op_data);
1611         if (rc)
1612                 GOTO(out, rc);
1613
1614         if (!S_ISREG(inode->i_mode) || hsm_import)
1615                 GOTO(out, rc = 0);
1616
1617         if (attr->ia_valid & (ATTR_SIZE |
1618                               ATTR_ATIME | ATTR_ATIME_SET |
1619                               ATTR_MTIME | ATTR_MTIME_SET |
1620                               ATTR_CTIME | ATTR_CTIME_SET)) {
1621                 /* For truncate and utimes sending attributes to OSTs, setting
1622                  * mtime/atime to the past will be performed under PW [0:EOF]
1623                  * extent lock (new_size:EOF for truncate).  It may seem
1624                  * excessive to send mtime/atime updates to OSTs when not
1625                  * setting times to past, but it is necessary due to possible
1626                  * time de-synchronization between MDT inode and OST objects */
1627                 rc = cl_setattr_ost(lli->lli_clob, attr, 0);
1628         }
1629
1630         /* If the file was restored, it needs to set dirty flag.
1631          *
1632          * We've already sent MDS_DATA_MODIFIED flag in
1633          * ll_md_setattr() for truncate. However, the MDT refuses to
1634          * set the HS_DIRTY flag on released files, so we have to set
1635          * it again if the file has been restored. Please check how
1636          * LLIF_DATA_MODIFIED is set in vvp_io_setattr_fini().
1637          *
1638          * Please notice that if the file is not released, the previous
1639          * MDS_DATA_MODIFIED has taken effect and usually
1640          * LLIF_DATA_MODIFIED is not set(see vvp_io_setattr_fini()).
1641          * This way we can save an RPC for common open + trunc
1642          * operation. */
1643         if (ll_file_test_and_clear_flag(lli, LLIF_DATA_MODIFIED)) {
1644                 struct hsm_state_set hss = {
1645                         .hss_valid = HSS_SETMASK,
1646                         .hss_setmask = HS_DIRTY,
1647                 };
1648                 int rc2;
1649
1650                 rc2 = ll_hsm_state_set(inode, &hss);
1651                 /* truncate and write can happen at the same time, so that
1652                  * the file can be set modified even though the file is not
1653                  * restored from released state, and ll_hsm_state_set() is
1654                  * not applicable for the file, and rc2 < 0 is normal in this
1655                  * case. */
1656                 if (rc2 < 0)
1657                         CDEBUG(D_INFO, DFID "HSM set dirty failed: rc2 = %d\n",
1658                                PFID(ll_inode2fid(inode)), rc2);
1659         }
1660
1661         EXIT;
1662 out:
1663         if (op_data != NULL)
1664                 ll_finish_md_op_data(op_data);
1665
1666         if (S_ISREG(inode->i_mode)) {
1667                 mutex_lock(&inode->i_mutex);
1668                 if ((attr->ia_valid & ATTR_SIZE) && !hsm_import)
1669                         inode_dio_wait(inode);
1670         }
1671
1672         ll_stats_ops_tally(ll_i2sbi(inode), (attr->ia_valid & ATTR_SIZE) ?
1673                         LPROC_LL_TRUNC : LPROC_LL_SETATTR, 1);
1674
1675         return rc;
1676 }
1677
1678 int ll_setattr(struct dentry *de, struct iattr *attr)
1679 {
1680         int mode = de->d_inode->i_mode;
1681
1682         if ((attr->ia_valid & (ATTR_CTIME|ATTR_SIZE|ATTR_MODE)) ==
1683                               (ATTR_CTIME|ATTR_SIZE|ATTR_MODE))
1684                 attr->ia_valid |= MDS_OPEN_OWNEROVERRIDE;
1685
1686         if (((attr->ia_valid & (ATTR_MODE|ATTR_FORCE|ATTR_SIZE)) ==
1687                                (ATTR_SIZE|ATTR_MODE)) &&
1688             (((mode & S_ISUID) && !(attr->ia_mode & S_ISUID)) ||
1689              (((mode & (S_ISGID|S_IXGRP)) == (S_ISGID|S_IXGRP)) &&
1690               !(attr->ia_mode & S_ISGID))))
1691                 attr->ia_valid |= ATTR_FORCE;
1692
1693         if ((attr->ia_valid & ATTR_MODE) &&
1694             (mode & S_ISUID) &&
1695             !(attr->ia_mode & S_ISUID) &&
1696             !(attr->ia_valid & ATTR_KILL_SUID))
1697                 attr->ia_valid |= ATTR_KILL_SUID;
1698
1699         if ((attr->ia_valid & ATTR_MODE) &&
1700             ((mode & (S_ISGID|S_IXGRP)) == (S_ISGID|S_IXGRP)) &&
1701             !(attr->ia_mode & S_ISGID) &&
1702             !(attr->ia_valid & ATTR_KILL_SGID))
1703                 attr->ia_valid |= ATTR_KILL_SGID;
1704
1705         return ll_setattr_raw(de, attr, false);
1706 }
1707
1708 int ll_statfs_internal(struct super_block *sb, struct obd_statfs *osfs,
1709                        __u64 max_age, __u32 flags)
1710 {
1711         struct ll_sb_info *sbi = ll_s2sbi(sb);
1712         struct obd_statfs obd_osfs;
1713         int rc;
1714         ENTRY;
1715
1716         rc = obd_statfs(NULL, sbi->ll_md_exp, osfs, max_age, flags);
1717         if (rc) {
1718                 CERROR("md_statfs fails: rc = %d\n", rc);
1719                 RETURN(rc);
1720         }
1721
1722         osfs->os_type = sb->s_magic;
1723
1724         CDEBUG(D_SUPER, "MDC blocks "LPU64"/"LPU64" objects "LPU64"/"LPU64"\n",
1725                osfs->os_bavail, osfs->os_blocks, osfs->os_ffree,osfs->os_files);
1726
1727         if (sbi->ll_flags & LL_SBI_LAZYSTATFS)
1728                 flags |= OBD_STATFS_NODELAY;
1729
1730         rc = obd_statfs_rqset(sbi->ll_dt_exp, &obd_osfs, max_age, flags);
1731         if (rc) {
1732                 CERROR("obd_statfs fails: rc = %d\n", rc);
1733                 RETURN(rc);
1734         }
1735
1736         CDEBUG(D_SUPER, "OSC blocks "LPU64"/"LPU64" objects "LPU64"/"LPU64"\n",
1737                obd_osfs.os_bavail, obd_osfs.os_blocks, obd_osfs.os_ffree,
1738                obd_osfs.os_files);
1739
1740         osfs->os_bsize = obd_osfs.os_bsize;
1741         osfs->os_blocks = obd_osfs.os_blocks;
1742         osfs->os_bfree = obd_osfs.os_bfree;
1743         osfs->os_bavail = obd_osfs.os_bavail;
1744
1745         /* If we don't have as many objects free on the OST as inodes
1746          * on the MDS, we reduce the total number of inodes to
1747          * compensate, so that the "inodes in use" number is correct.
1748          */
1749         if (obd_osfs.os_ffree < osfs->os_ffree) {
1750                 osfs->os_files = (osfs->os_files - osfs->os_ffree) +
1751                         obd_osfs.os_ffree;
1752                 osfs->os_ffree = obd_osfs.os_ffree;
1753         }
1754
1755         RETURN(rc);
1756 }
1757 int ll_statfs(struct dentry *de, struct kstatfs *sfs)
1758 {
1759         struct super_block *sb = de->d_sb;
1760         struct obd_statfs osfs;
1761         __u64 fsid = huge_encode_dev(sb->s_dev);
1762         int rc;
1763
1764         CDEBUG(D_VFSTRACE, "VFS Op: at "LPU64" jiffies\n", get_jiffies_64());
1765         ll_stats_ops_tally(ll_s2sbi(sb), LPROC_LL_STAFS, 1);
1766
1767         /* Some amount of caching on the client is allowed */
1768         rc = ll_statfs_internal(sb, &osfs,
1769                                 cfs_time_shift_64(-OBD_STATFS_CACHE_SECONDS),
1770                                 0);
1771         if (rc)
1772                 return rc;
1773
1774         statfs_unpack(sfs, &osfs);
1775
1776         /* We need to downshift for all 32-bit kernels, because we can't
1777          * tell if the kernel is being called via sys_statfs64() or not.
1778          * Stop before overflowing f_bsize - in which case it is better
1779          * to just risk EOVERFLOW if caller is using old sys_statfs(). */
1780         if (sizeof(long) < 8) {
1781                 while (osfs.os_blocks > ~0UL && sfs->f_bsize < 0x40000000) {
1782                         sfs->f_bsize <<= 1;
1783
1784                         osfs.os_blocks >>= 1;
1785                         osfs.os_bfree >>= 1;
1786                         osfs.os_bavail >>= 1;
1787                 }
1788         }
1789
1790         sfs->f_blocks = osfs.os_blocks;
1791         sfs->f_bfree = osfs.os_bfree;
1792         sfs->f_bavail = osfs.os_bavail;
1793         sfs->f_fsid.val[0] = (__u32)fsid;
1794         sfs->f_fsid.val[1] = (__u32)(fsid >> 32);
1795         return 0;
1796 }
1797
1798 void ll_inode_size_lock(struct inode *inode)
1799 {
1800         struct ll_inode_info *lli;
1801
1802         LASSERT(!S_ISDIR(inode->i_mode));
1803
1804         lli = ll_i2info(inode);
1805         mutex_lock(&lli->lli_size_mutex);
1806 }
1807
1808 void ll_inode_size_unlock(struct inode *inode)
1809 {
1810         struct ll_inode_info *lli;
1811
1812         lli = ll_i2info(inode);
1813         mutex_unlock(&lli->lli_size_mutex);
1814 }
1815
1816 int ll_update_inode(struct inode *inode, struct lustre_md *md)
1817 {
1818         struct ll_inode_info *lli = ll_i2info(inode);
1819         struct mdt_body *body = md->body;
1820         struct ll_sb_info *sbi = ll_i2sbi(inode);
1821
1822         if (body->mbo_valid & OBD_MD_FLEASIZE)
1823                 cl_file_inode_init(inode, md);
1824
1825         if (S_ISDIR(inode->i_mode)) {
1826                 int     rc;
1827
1828                 rc = ll_update_lsm_md(inode, md);
1829                 if (rc != 0)
1830                         return rc;
1831         }
1832
1833 #ifdef CONFIG_FS_POSIX_ACL
1834         if (body->mbo_valid & OBD_MD_FLACL) {
1835                 spin_lock(&lli->lli_lock);
1836                 if (lli->lli_posix_acl)
1837                         posix_acl_release(lli->lli_posix_acl);
1838                 lli->lli_posix_acl = md->posix_acl;
1839                 spin_unlock(&lli->lli_lock);
1840         }
1841 #endif
1842         inode->i_ino = cl_fid_build_ino(&body->mbo_fid1,
1843                                         sbi->ll_flags & LL_SBI_32BIT_API);
1844         inode->i_generation = cl_fid_build_gen(&body->mbo_fid1);
1845
1846         if (body->mbo_valid & OBD_MD_FLATIME) {
1847                 if (body->mbo_atime > LTIME_S(inode->i_atime))
1848                         LTIME_S(inode->i_atime) = body->mbo_atime;
1849                 lli->lli_atime = body->mbo_atime;
1850         }
1851
1852         if (body->mbo_valid & OBD_MD_FLMTIME) {
1853                 if (body->mbo_mtime > LTIME_S(inode->i_mtime)) {
1854                         CDEBUG(D_INODE, "setting ino %lu mtime from %lu "
1855                                "to "LPU64"\n", inode->i_ino,
1856                                LTIME_S(inode->i_mtime), body->mbo_mtime);
1857                         LTIME_S(inode->i_mtime) = body->mbo_mtime;
1858                 }
1859                 lli->lli_mtime = body->mbo_mtime;
1860         }
1861
1862         if (body->mbo_valid & OBD_MD_FLCTIME) {
1863                 if (body->mbo_ctime > LTIME_S(inode->i_ctime))
1864                         LTIME_S(inode->i_ctime) = body->mbo_ctime;
1865                 lli->lli_ctime = body->mbo_ctime;
1866         }
1867
1868         if (body->mbo_valid & OBD_MD_FLMODE)
1869                 inode->i_mode = (inode->i_mode & S_IFMT) |
1870                                 (body->mbo_mode & ~S_IFMT);
1871
1872         if (body->mbo_valid & OBD_MD_FLTYPE)
1873                 inode->i_mode = (inode->i_mode & ~S_IFMT) |
1874                                 (body->mbo_mode & S_IFMT);
1875
1876         LASSERT(inode->i_mode != 0);
1877         if (S_ISREG(inode->i_mode))
1878                 inode->i_blkbits = min(PTLRPC_MAX_BRW_BITS + 1,
1879                                        LL_MAX_BLKSIZE_BITS);
1880         else
1881                 inode->i_blkbits = inode->i_sb->s_blocksize_bits;
1882
1883         if (body->mbo_valid & OBD_MD_FLUID)
1884                 inode->i_uid = make_kuid(&init_user_ns, body->mbo_uid);
1885         if (body->mbo_valid & OBD_MD_FLGID)
1886                 inode->i_gid = make_kgid(&init_user_ns, body->mbo_gid);
1887         if (body->mbo_valid & OBD_MD_FLFLAGS)
1888                 inode->i_flags = ll_ext_to_inode_flags(body->mbo_flags);
1889         if (body->mbo_valid & OBD_MD_FLNLINK)
1890                 set_nlink(inode, body->mbo_nlink);
1891         if (body->mbo_valid & OBD_MD_FLRDEV)
1892                 inode->i_rdev = old_decode_dev(body->mbo_rdev);
1893
1894         if (body->mbo_valid & OBD_MD_FLID) {
1895                 /* FID shouldn't be changed! */
1896                 if (fid_is_sane(&lli->lli_fid)) {
1897                         LASSERTF(lu_fid_eq(&lli->lli_fid, &body->mbo_fid1),
1898                                  "Trying to change FID "DFID
1899                                  " to the "DFID", inode "DFID"(%p)\n",
1900                                  PFID(&lli->lli_fid), PFID(&body->mbo_fid1),
1901                                  PFID(ll_inode2fid(inode)), inode);
1902                 } else {
1903                         lli->lli_fid = body->mbo_fid1;
1904                 }
1905         }
1906
1907         LASSERT(fid_seq(&lli->lli_fid) != 0);
1908
1909         if (body->mbo_valid & OBD_MD_FLSIZE) {
1910                 i_size_write(inode, body->mbo_size);
1911
1912                 CDEBUG(D_VFSTRACE, "inode="DFID", updating i_size %llu\n",
1913                        PFID(ll_inode2fid(inode)),
1914                        (unsigned long long)body->mbo_size);
1915
1916                 if (body->mbo_valid & OBD_MD_FLBLOCKS)
1917                         inode->i_blocks = body->mbo_blocks;
1918         }
1919
1920         if (body->mbo_valid & OBD_MD_TSTATE) {
1921                 /* Set LLIF_FILE_RESTORING if restore ongoing and
1922                  * clear it when done to ensure to start again
1923                  * glimpsing updated attrs
1924                  */
1925                 if (body->mbo_t_state & MS_RESTORE)
1926                         ll_file_set_flag(lli, LLIF_FILE_RESTORING);
1927                 else
1928                         ll_file_clear_flag(lli, LLIF_FILE_RESTORING);
1929         }
1930
1931         return 0;
1932 }
1933
1934 int ll_read_inode2(struct inode *inode, void *opaque)
1935 {
1936         struct lustre_md *md = opaque;
1937         struct ll_inode_info *lli = ll_i2info(inode);
1938         int     rc;
1939         ENTRY;
1940
1941         CDEBUG(D_VFSTRACE, "VFS Op:inode="DFID"(%p)\n",
1942                PFID(&lli->lli_fid), inode);
1943
1944         /* Core attributes from the MDS first.  This is a new inode, and
1945          * the VFS doesn't zero times in the core inode so we have to do
1946          * it ourselves.  They will be overwritten by either MDS or OST
1947          * attributes - we just need to make sure they aren't newer. */
1948         LTIME_S(inode->i_mtime) = 0;
1949         LTIME_S(inode->i_atime) = 0;
1950         LTIME_S(inode->i_ctime) = 0;
1951         inode->i_rdev = 0;
1952         rc = ll_update_inode(inode, md);
1953         if (rc != 0)
1954                 RETURN(rc);
1955
1956         /* OIDEBUG(inode); */
1957
1958 #ifdef HAVE_BACKING_DEV_INFO
1959         /* initializing backing dev info. */
1960         inode->i_mapping->backing_dev_info = &s2lsi(inode->i_sb)->lsi_bdi;
1961 #endif
1962         if (S_ISREG(inode->i_mode)) {
1963                 struct ll_sb_info *sbi = ll_i2sbi(inode);
1964                 inode->i_op = &ll_file_inode_operations;
1965                 inode->i_fop = sbi->ll_fop;
1966                 inode->i_mapping->a_ops = (struct address_space_operations *)&ll_aops;
1967                 EXIT;
1968         } else if (S_ISDIR(inode->i_mode)) {
1969                 inode->i_op = &ll_dir_inode_operations;
1970                 inode->i_fop = &ll_dir_operations;
1971                 EXIT;
1972         } else if (S_ISLNK(inode->i_mode)) {
1973                 inode->i_op = &ll_fast_symlink_inode_operations;
1974                 EXIT;
1975         } else {
1976                 inode->i_op = &ll_special_inode_operations;
1977
1978                 init_special_inode(inode, inode->i_mode,
1979                                    inode->i_rdev);
1980
1981                 EXIT;
1982         }
1983
1984         return 0;
1985 }
1986
1987 void ll_delete_inode(struct inode *inode)
1988 {
1989         struct ll_inode_info *lli = ll_i2info(inode);
1990         ENTRY;
1991
1992         if (S_ISREG(inode->i_mode) && lli->lli_clob != NULL)
1993                 /* It is last chance to write out dirty pages,
1994                  * otherwise we may lose data while umount */
1995                 cl_sync_file_range(inode, 0, OBD_OBJECT_EOF, CL_FSYNC_LOCAL, 1);
1996
1997         truncate_inode_pages_final(&inode->i_data);
1998
1999         LASSERTF(inode->i_data.nrpages == 0, "inode="DFID"(%p) nrpages=%lu, "
2000                  "see https://jira.hpdd.intel.com/browse/LU-118\n",
2001                  PFID(ll_inode2fid(inode)), inode, inode->i_data.nrpages);
2002
2003 #ifdef HAVE_SBOPS_EVICT_INODE
2004         ll_clear_inode(inode);
2005 #endif
2006         clear_inode(inode);
2007
2008         EXIT;
2009 }
2010
2011 int ll_iocontrol(struct inode *inode, struct file *file,
2012                  unsigned int cmd, unsigned long arg)
2013 {
2014         struct ll_sb_info *sbi = ll_i2sbi(inode);
2015         struct ptlrpc_request *req = NULL;
2016         int rc, flags = 0;
2017         ENTRY;
2018
2019         switch(cmd) {
2020         case FSFILT_IOC_GETFLAGS: {
2021                 struct mdt_body *body;
2022                 struct md_op_data *op_data;
2023
2024                 op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL,
2025                                              0, 0, LUSTRE_OPC_ANY,
2026                                              NULL);
2027                 if (IS_ERR(op_data))
2028                         RETURN(PTR_ERR(op_data));
2029
2030                 op_data->op_valid = OBD_MD_FLFLAGS;
2031                 rc = md_getattr(sbi->ll_md_exp, op_data, &req);
2032                 ll_finish_md_op_data(op_data);
2033                 if (rc) {
2034                         CERROR("%s: failure inode "DFID": rc = %d\n",
2035                                sbi->ll_md_exp->exp_obd->obd_name,
2036                                PFID(ll_inode2fid(inode)), rc);
2037                         RETURN(-abs(rc));
2038                 }
2039
2040                 body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
2041
2042                 flags = body->mbo_flags;
2043
2044                 ptlrpc_req_finished(req);
2045
2046                 RETURN(put_user(flags, (int __user *)arg));
2047         }
2048         case FSFILT_IOC_SETFLAGS: {
2049                 struct iattr *attr;
2050                 struct md_op_data *op_data;
2051                 struct cl_object *obj;
2052
2053                 if (get_user(flags, (int __user *)arg))
2054                         RETURN(-EFAULT);
2055
2056                 op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL, 0, 0,
2057                                              LUSTRE_OPC_ANY, NULL);
2058                 if (IS_ERR(op_data))
2059                         RETURN(PTR_ERR(op_data));
2060
2061                 op_data->op_attr_flags = flags;
2062                 op_data->op_attr.ia_valid |= ATTR_ATTR_FLAG;
2063                 rc = md_setattr(sbi->ll_md_exp, op_data, NULL, 0, &req);
2064                 ll_finish_md_op_data(op_data);
2065                 ptlrpc_req_finished(req);
2066                 if (rc)
2067                         RETURN(rc);
2068
2069                 inode->i_flags = ll_ext_to_inode_flags(flags);
2070
2071                 obj = ll_i2info(inode)->lli_clob;
2072                 if (obj == NULL)
2073                         RETURN(0);
2074
2075                 OBD_ALLOC_PTR(attr);
2076                 if (attr == NULL)
2077                         RETURN(-ENOMEM);
2078
2079                 attr->ia_valid = ATTR_ATTR_FLAG;
2080                 rc = cl_setattr_ost(obj, attr, flags);
2081
2082                 OBD_FREE_PTR(attr);
2083                 RETURN(rc);
2084         }
2085         default:
2086                 RETURN(-ENOSYS);
2087         }
2088
2089         RETURN(0);
2090 }
2091
2092 int ll_flush_ctx(struct inode *inode)
2093 {
2094         struct ll_sb_info  *sbi = ll_i2sbi(inode);
2095
2096         CDEBUG(D_SEC, "flush context for user %d\n",
2097                from_kuid(&init_user_ns, current_uid()));
2098
2099         obd_set_info_async(NULL, sbi->ll_md_exp,
2100                            sizeof(KEY_FLUSH_CTX), KEY_FLUSH_CTX,
2101                            0, NULL, NULL);
2102         obd_set_info_async(NULL, sbi->ll_dt_exp,
2103                            sizeof(KEY_FLUSH_CTX), KEY_FLUSH_CTX,
2104                            0, NULL, NULL);
2105         return 0;
2106 }
2107
2108 /* umount -f client means force down, don't save state */
2109 void ll_umount_begin(struct super_block *sb)
2110 {
2111         struct ll_sb_info *sbi = ll_s2sbi(sb);
2112         struct obd_device *obd;
2113         struct obd_ioctl_data *ioc_data;
2114         ENTRY;
2115
2116         CDEBUG(D_VFSTRACE, "VFS Op: superblock %p count %d active %d\n", sb,
2117                sb->s_count, atomic_read(&sb->s_active));
2118
2119         obd = class_exp2obd(sbi->ll_md_exp);
2120         if (obd == NULL) {
2121                 CERROR("Invalid MDC connection handle "LPX64"\n",
2122                        sbi->ll_md_exp->exp_handle.h_cookie);
2123                 EXIT;
2124                 return;
2125         }
2126         obd->obd_force = 1;
2127
2128         obd = class_exp2obd(sbi->ll_dt_exp);
2129         if (obd == NULL) {
2130                 CERROR("Invalid LOV connection handle "LPX64"\n",
2131                        sbi->ll_dt_exp->exp_handle.h_cookie);
2132                 EXIT;
2133                 return;
2134         }
2135         obd->obd_force = 1;
2136
2137         OBD_ALLOC_PTR(ioc_data);
2138         if (ioc_data) {
2139                 obd_iocontrol(IOC_OSC_SET_ACTIVE, sbi->ll_md_exp,
2140                               sizeof *ioc_data, ioc_data, NULL);
2141
2142                 obd_iocontrol(IOC_OSC_SET_ACTIVE, sbi->ll_dt_exp,
2143                               sizeof *ioc_data, ioc_data, NULL);
2144
2145                 OBD_FREE_PTR(ioc_data);
2146         }
2147
2148         /* Really, we'd like to wait until there are no requests outstanding,
2149          * and then continue.  For now, we just invalidate the requests,
2150          * schedule() and sleep one second if needed, and hope.
2151          */
2152         schedule();
2153         EXIT;
2154 }
2155
2156 int ll_remount_fs(struct super_block *sb, int *flags, char *data)
2157 {
2158         struct ll_sb_info *sbi = ll_s2sbi(sb);
2159         char *profilenm = get_profile_name(sb);
2160         int err;
2161         __u32 read_only;
2162
2163         if ((*flags & MS_RDONLY) != (sb->s_flags & MS_RDONLY)) {
2164                 read_only = *flags & MS_RDONLY;
2165                 err = obd_set_info_async(NULL, sbi->ll_md_exp,
2166                                          sizeof(KEY_READ_ONLY),
2167                                          KEY_READ_ONLY, sizeof(read_only),
2168                                          &read_only, NULL);
2169                 if (err) {
2170                         LCONSOLE_WARN("Failed to remount %s %s (%d)\n",
2171                                       profilenm, read_only ?
2172                                       "read-only" : "read-write", err);
2173                         return err;
2174                 }
2175
2176                 if (read_only)
2177                         sb->s_flags |= MS_RDONLY;
2178                 else
2179                         sb->s_flags &= ~MS_RDONLY;
2180
2181                 if (sbi->ll_flags & LL_SBI_VERBOSE)
2182                         LCONSOLE_WARN("Remounted %s %s\n", profilenm,
2183                                       read_only ?  "read-only" : "read-write");
2184         }
2185         return 0;
2186 }
2187
2188 /**
2189  * Cleanup the open handle that is cached on MDT-side.
2190  *
2191  * For open case, the client side open handling thread may hit error
2192  * after the MDT grant the open. Under such case, the client should
2193  * send close RPC to the MDT as cleanup; otherwise, the open handle
2194  * on the MDT will be leaked there until the client umount or evicted.
2195  *
2196  * In further, if someone unlinked the file, because the open handle
2197  * holds the reference on such file/object, then it will block the
2198  * subsequent threads that want to locate such object via FID.
2199  *
2200  * \param[in] sb        super block for this file-system
2201  * \param[in] open_req  pointer to the original open request
2202  */
2203 void ll_open_cleanup(struct super_block *sb, struct ptlrpc_request *open_req)
2204 {
2205         struct mdt_body                 *body;
2206         struct md_op_data               *op_data;
2207         struct ptlrpc_request           *close_req = NULL;
2208         struct obd_export               *exp       = ll_s2sbi(sb)->ll_md_exp;
2209         ENTRY;
2210
2211         body = req_capsule_server_get(&open_req->rq_pill, &RMF_MDT_BODY);
2212         OBD_ALLOC_PTR(op_data);
2213         if (op_data == NULL) {
2214                 CWARN("%s: cannot allocate op_data to release open handle for "
2215                       DFID"\n",
2216                       ll_get_fsname(sb, NULL, 0), PFID(&body->mbo_fid1));
2217
2218                 RETURN_EXIT;
2219         }
2220
2221         op_data->op_fid1 = body->mbo_fid1;
2222         op_data->op_handle = body->mbo_handle;
2223         op_data->op_mod_time = cfs_time_current_sec();
2224         md_close(exp, op_data, NULL, &close_req);
2225         ptlrpc_req_finished(close_req);
2226         ll_finish_md_op_data(op_data);
2227
2228         EXIT;
2229 }
2230
2231 int ll_prep_inode(struct inode **inode, struct ptlrpc_request *req,
2232                   struct super_block *sb, struct lookup_intent *it)
2233 {
2234         struct ll_sb_info *sbi = NULL;
2235         struct lustre_md md = { NULL };
2236         int rc;
2237         ENTRY;
2238
2239         LASSERT(*inode || sb);
2240         sbi = sb ? ll_s2sbi(sb) : ll_i2sbi(*inode);
2241         rc = md_get_lustre_md(sbi->ll_md_exp, req, sbi->ll_dt_exp,
2242                               sbi->ll_md_exp, &md);
2243         if (rc != 0)
2244                 GOTO(cleanup, rc);
2245
2246         if (*inode) {
2247                 rc = ll_update_inode(*inode, &md);
2248                 if (rc != 0)
2249                         GOTO(out, rc);
2250         } else {
2251                 LASSERT(sb != NULL);
2252
2253                 /*
2254                  * At this point server returns to client's same fid as client
2255                  * generated for creating. So using ->fid1 is okay here.
2256                  */
2257                 if (!fid_is_sane(&md.body->mbo_fid1)) {
2258                         CERROR("%s: Fid is insane "DFID"\n",
2259                                 ll_get_fsname(sb, NULL, 0),
2260                                 PFID(&md.body->mbo_fid1));
2261                         GOTO(out, rc = -EINVAL);
2262                 }
2263
2264                 *inode = ll_iget(sb, cl_fid_build_ino(&md.body->mbo_fid1,
2265                                              sbi->ll_flags & LL_SBI_32BIT_API),
2266                                  &md);
2267                 if (IS_ERR(*inode)) {
2268 #ifdef CONFIG_FS_POSIX_ACL
2269                         if (md.posix_acl) {
2270                                 posix_acl_release(md.posix_acl);
2271                                 md.posix_acl = NULL;
2272                         }
2273 #endif
2274                         rc = IS_ERR(*inode) ? PTR_ERR(*inode) : -ENOMEM;
2275                         *inode = NULL;
2276                         CERROR("new_inode -fatal: rc %d\n", rc);
2277                         GOTO(out, rc);
2278                 }
2279         }
2280
2281         /* Handling piggyback layout lock.
2282          * Layout lock can be piggybacked by getattr and open request.
2283          * The lsm can be applied to inode only if it comes with a layout lock
2284          * otherwise correct layout may be overwritten, for example:
2285          * 1. proc1: mdt returns a lsm but not granting layout
2286          * 2. layout was changed by another client
2287          * 3. proc2: refresh layout and layout lock granted
2288          * 4. proc1: to apply a stale layout */
2289         if (it != NULL && it->it_lock_mode != 0) {
2290                 struct lustre_handle lockh;
2291                 struct ldlm_lock *lock;
2292
2293                 lockh.cookie = it->it_lock_handle;
2294                 lock = ldlm_handle2lock(&lockh);
2295                 LASSERT(lock != NULL);
2296                 if (ldlm_has_layout(lock)) {
2297                         struct cl_object_conf conf;
2298
2299                         memset(&conf, 0, sizeof(conf));
2300                         conf.coc_opc = OBJECT_CONF_SET;
2301                         conf.coc_inode = *inode;
2302                         conf.coc_lock = lock;
2303                         conf.u.coc_layout = md.layout;
2304                         (void)ll_layout_conf(*inode, &conf);
2305                 }
2306                 LDLM_LOCK_PUT(lock);
2307         }
2308
2309         GOTO(out, rc = 0);
2310
2311 out:
2312         md_free_lustre_md(sbi->ll_md_exp, &md);
2313
2314 cleanup:
2315         if (rc != 0 && it != NULL && it->it_op & IT_OPEN)
2316                 ll_open_cleanup(sb != NULL ? sb : (*inode)->i_sb, req);
2317
2318         return rc;
2319 }
2320
2321 int ll_obd_statfs(struct inode *inode, void __user *arg)
2322 {
2323         struct ll_sb_info *sbi = NULL;
2324         struct obd_export *exp;
2325         char *buf = NULL;
2326         struct obd_ioctl_data *data = NULL;
2327         __u32 type;
2328         int len = 0, rc;
2329
2330         if (!inode || !(sbi = ll_i2sbi(inode)))
2331                 GOTO(out_statfs, rc = -EINVAL);
2332
2333         rc = obd_ioctl_getdata(&buf, &len, arg);
2334         if (rc)
2335                 GOTO(out_statfs, rc);
2336
2337         data = (void*)buf;
2338         if (!data->ioc_inlbuf1 || !data->ioc_inlbuf2 ||
2339             !data->ioc_pbuf1 || !data->ioc_pbuf2)
2340                 GOTO(out_statfs, rc = -EINVAL);
2341
2342         if (data->ioc_inllen1 != sizeof(__u32) ||
2343             data->ioc_inllen2 != sizeof(__u32) ||
2344             data->ioc_plen1 != sizeof(struct obd_statfs) ||
2345             data->ioc_plen2 != sizeof(struct obd_uuid))
2346                 GOTO(out_statfs, rc = -EINVAL);
2347
2348         memcpy(&type, data->ioc_inlbuf1, sizeof(__u32));
2349         if (type & LL_STATFS_LMV)
2350                 exp = sbi->ll_md_exp;
2351         else if (type & LL_STATFS_LOV)
2352                 exp = sbi->ll_dt_exp;
2353         else
2354                 GOTO(out_statfs, rc = -ENODEV);
2355
2356         rc = obd_iocontrol(IOC_OBD_STATFS, exp, len, buf, NULL);
2357         if (rc)
2358                 GOTO(out_statfs, rc);
2359 out_statfs:
2360         if (buf)
2361                 obd_ioctl_freedata(buf, len);
2362         return rc;
2363 }
2364
2365 int ll_process_config(struct lustre_cfg *lcfg)
2366 {
2367         struct super_block *sb;
2368         unsigned long x;
2369         int rc = 0;
2370         char *ptr;
2371
2372         /* The instance name contains the sb: lustre-client-aacfe000 */
2373         ptr = strrchr(lustre_cfg_string(lcfg, 0), '-');
2374         if (!ptr || !*(++ptr))
2375                 return -EINVAL;
2376         if (sscanf(ptr, "%lx", &x) != 1)
2377                 return -EINVAL;
2378         sb = (struct super_block *)x;
2379         /* This better be a real Lustre superblock! */
2380         LASSERT(s2lsi(sb)->lsi_lmd->lmd_magic == LMD_MAGIC);
2381
2382         /* Note we have not called client_common_fill_super yet, so
2383            proc fns must be able to handle that! */
2384         rc = class_process_proc_param(PARAM_LLITE, lprocfs_llite_obd_vars,
2385                                       lcfg, sb);
2386         if (rc > 0)
2387                 rc = 0;
2388         return rc;
2389 }
2390
2391 /* this function prepares md_op_data hint for passing it down to MD stack. */
2392 struct md_op_data *ll_prep_md_op_data(struct md_op_data *op_data,
2393                                       struct inode *i1, struct inode *i2,
2394                                       const char *name, size_t namelen,
2395                                       __u32 mode, __u32 opc, void *data)
2396 {
2397         LASSERT(i1 != NULL);
2398
2399         if (name == NULL) {
2400                 /* Do not reuse namelen for something else. */
2401                 if (namelen != 0)
2402                         return ERR_PTR(-EINVAL);
2403         } else {
2404                 if (namelen > ll_i2sbi(i1)->ll_namelen)
2405                         return ERR_PTR(-ENAMETOOLONG);
2406
2407                 if (!lu_name_is_valid_2(name, namelen))
2408                         return ERR_PTR(-EINVAL);
2409         }
2410
2411         if (op_data == NULL)
2412                 OBD_ALLOC_PTR(op_data);
2413
2414         if (op_data == NULL)
2415                 return ERR_PTR(-ENOMEM);
2416
2417         ll_i2gids(op_data->op_suppgids, i1, i2);
2418         op_data->op_fid1 = *ll_inode2fid(i1);
2419         op_data->op_default_stripe_offset = -1;
2420         if (S_ISDIR(i1->i_mode)) {
2421                 op_data->op_mea1 = ll_i2info(i1)->lli_lsm_md;
2422                 if (opc == LUSTRE_OPC_MKDIR)
2423                         op_data->op_default_stripe_offset =
2424                                    ll_i2info(i1)->lli_def_stripe_offset;
2425         }
2426
2427         if (i2) {
2428                 op_data->op_fid2 = *ll_inode2fid(i2);
2429                 if (S_ISDIR(i2->i_mode))
2430                         op_data->op_mea2 = ll_i2info(i2)->lli_lsm_md;
2431         } else {
2432                 fid_zero(&op_data->op_fid2);
2433         }
2434
2435         if (ll_i2sbi(i1)->ll_flags & LL_SBI_64BIT_HASH)
2436                 op_data->op_cli_flags |= CLI_HASH64;
2437
2438         if (ll_need_32bit_api(ll_i2sbi(i1)))
2439                 op_data->op_cli_flags |= CLI_API32;
2440
2441         op_data->op_name = name;
2442         op_data->op_namelen = namelen;
2443         op_data->op_mode = mode;
2444         op_data->op_mod_time = cfs_time_current_sec();
2445         op_data->op_fsuid = from_kuid(&init_user_ns, current_fsuid());
2446         op_data->op_fsgid = from_kgid(&init_user_ns, current_fsgid());
2447         op_data->op_cap = cfs_curproc_cap_pack();
2448         if ((opc == LUSTRE_OPC_CREATE) && (name != NULL) &&
2449              filename_is_volatile(name, namelen, &op_data->op_mds)) {
2450                 op_data->op_bias |= MDS_CREATE_VOLATILE;
2451         } else {
2452                 op_data->op_mds = 0;
2453         }
2454         op_data->op_data = data;
2455
2456         return op_data;
2457 }
2458
2459 void ll_finish_md_op_data(struct md_op_data *op_data)
2460 {
2461         OBD_FREE_PTR(op_data);
2462 }
2463
2464 #ifdef HAVE_SUPEROPS_USE_DENTRY
2465 int ll_show_options(struct seq_file *seq, struct dentry *dentry)
2466 #else
2467 int ll_show_options(struct seq_file *seq, struct vfsmount *vfs)
2468 #endif
2469 {
2470         struct ll_sb_info *sbi;
2471
2472 #ifdef HAVE_SUPEROPS_USE_DENTRY
2473         LASSERT((seq != NULL) && (dentry != NULL));
2474         sbi = ll_s2sbi(dentry->d_sb);
2475 #else
2476         LASSERT((seq != NULL) && (vfs != NULL));
2477         sbi = ll_s2sbi(vfs->mnt_sb);
2478 #endif
2479
2480         if (sbi->ll_flags & LL_SBI_NOLCK)
2481                 seq_puts(seq, ",nolock");
2482
2483         if (sbi->ll_flags & LL_SBI_FLOCK)
2484                 seq_puts(seq, ",flock");
2485
2486         if (sbi->ll_flags & LL_SBI_LOCALFLOCK)
2487                 seq_puts(seq, ",localflock");
2488
2489         if (sbi->ll_flags & LL_SBI_USER_XATTR)
2490                 seq_puts(seq, ",user_xattr");
2491
2492         if (sbi->ll_flags & LL_SBI_LAZYSTATFS)
2493                 seq_puts(seq, ",lazystatfs");
2494
2495         if (sbi->ll_flags & LL_SBI_USER_FID2PATH)
2496                 seq_puts(seq, ",user_fid2path");
2497
2498         if (sbi->ll_flags & LL_SBI_ALWAYS_PING)
2499                 seq_puts(seq, ",always_ping");
2500
2501         RETURN(0);
2502 }
2503
2504 /**
2505  * Get obd name by cmd, and copy out to user space
2506  */
2507 int ll_get_obd_name(struct inode *inode, unsigned int cmd, unsigned long arg)
2508 {
2509         struct ll_sb_info *sbi = ll_i2sbi(inode);
2510         struct obd_device *obd;
2511         ENTRY;
2512
2513         if (cmd == OBD_IOC_GETDTNAME)
2514                 obd = class_exp2obd(sbi->ll_dt_exp);
2515         else if (cmd == OBD_IOC_GETMDNAME)
2516                 obd = class_exp2obd(sbi->ll_md_exp);
2517         else
2518                 RETURN(-EINVAL);
2519
2520         if (!obd)
2521                 RETURN(-ENOENT);
2522
2523         if (copy_to_user((void __user *)arg, obd->obd_name,
2524                          strlen(obd->obd_name) + 1))
2525                 RETURN(-EFAULT);
2526
2527         RETURN(0);
2528 }
2529
2530 /**
2531  * Get lustre file system name by \a sbi. If \a buf is provided(non-NULL), the
2532  * fsname will be returned in this buffer; otherwise, a static buffer will be
2533  * used to store the fsname and returned to caller.
2534  */
2535 char *ll_get_fsname(struct super_block *sb, char *buf, int buflen)
2536 {
2537         static char fsname_static[MTI_NAME_MAXLEN];
2538         struct lustre_sb_info *lsi = s2lsi(sb);
2539         char *ptr;
2540         int len;
2541
2542         if (buf == NULL) {
2543                 /* this means the caller wants to use static buffer
2544                  * and it doesn't care about race. Usually this is
2545                  * in error reporting path */
2546                 buf = fsname_static;
2547                 buflen = sizeof(fsname_static);
2548         }
2549
2550         len = strlen(lsi->lsi_lmd->lmd_profile);
2551         ptr = strrchr(lsi->lsi_lmd->lmd_profile, '-');
2552         if (ptr && (strcmp(ptr, "-client") == 0))
2553                 len -= 7;
2554
2555         if (unlikely(len >= buflen))
2556                 len = buflen - 1;
2557         strncpy(buf, lsi->lsi_lmd->lmd_profile, len);
2558         buf[len] = '\0';
2559
2560         return buf;
2561 }
2562
2563 static char* ll_d_path(struct dentry *dentry, char *buf, int bufsize)
2564 {
2565         char *path = NULL;
2566
2567         struct path p;
2568
2569         p.dentry = dentry;
2570         p.mnt = current->fs->root.mnt;
2571         path_get(&p);
2572         path = d_path(&p, buf, bufsize);
2573         path_put(&p);
2574         return path;
2575 }
2576
2577 void ll_dirty_page_discard_warn(struct page *page, int ioret)
2578 {
2579         char *buf, *path = NULL;
2580         struct dentry *dentry = NULL;
2581         struct inode *inode = page->mapping->host;
2582
2583         /* this can be called inside spin lock so use GFP_ATOMIC. */
2584         buf = (char *)__get_free_page(GFP_ATOMIC);
2585         if (buf != NULL) {
2586                 dentry = d_find_alias(page->mapping->host);
2587                 if (dentry != NULL)
2588                         path = ll_d_path(dentry, buf, PAGE_SIZE);
2589         }
2590
2591         CDEBUG(D_WARNING,
2592                "%s: dirty page discard: %s/fid: "DFID"/%s may get corrupted "
2593                "(rc %d)\n", ll_get_fsname(page->mapping->host->i_sb, NULL, 0),
2594                s2lsi(page->mapping->host->i_sb)->lsi_lmd->lmd_dev,
2595                PFID(ll_inode2fid(inode)),
2596                (path && !IS_ERR(path)) ? path : "", ioret);
2597
2598         if (dentry != NULL)
2599                 dput(dentry);
2600
2601         if (buf != NULL)
2602                 free_page((unsigned long)buf);
2603 }
2604
2605 ssize_t ll_copy_user_md(const struct lov_user_md __user *md,
2606                         struct lov_user_md **kbuf)
2607 {
2608         struct lov_user_md      lum;
2609         ssize_t                 lum_size;
2610         ENTRY;
2611
2612         if (copy_from_user(&lum, md, sizeof(lum)))
2613                 RETURN(-EFAULT);
2614
2615         lum_size = ll_lov_user_md_size(&lum);
2616         if (lum_size < 0)
2617                 RETURN(lum_size);
2618
2619         OBD_ALLOC(*kbuf, lum_size);
2620         if (*kbuf == NULL)
2621                 RETURN(-ENOMEM);
2622
2623         if (copy_from_user(*kbuf, md, lum_size) != 0) {
2624                 OBD_FREE(*kbuf, lum_size);
2625                 RETURN(-EFAULT);
2626         }
2627
2628         RETURN(lum_size);
2629 }
2630
2631 /*
2632  * Compute llite root squash state after a change of root squash
2633  * configuration setting or add/remove of a lnet nid
2634  */
2635 void ll_compute_rootsquash_state(struct ll_sb_info *sbi)
2636 {
2637         struct root_squash_info *squash = &sbi->ll_squash;
2638         int i;
2639         bool matched;
2640         lnet_process_id_t id;
2641
2642         /* Update norootsquash flag */
2643         down_write(&squash->rsi_sem);
2644         if (list_empty(&squash->rsi_nosquash_nids))
2645                 sbi->ll_flags &= ~LL_SBI_NOROOTSQUASH;
2646         else {
2647                 /* Do not apply root squash as soon as one of our NIDs is
2648                  * in the nosquash_nids list */
2649                 matched = false;
2650                 i = 0;
2651                 while (LNetGetId(i++, &id) != -ENOENT) {
2652                         if (LNET_NETTYP(LNET_NIDNET(id.nid)) == LOLND)
2653                                 continue;
2654                         if (cfs_match_nid(id.nid, &squash->rsi_nosquash_nids)) {
2655                                 matched = true;
2656                                 break;
2657                         }
2658                 }
2659                 if (matched)
2660                         sbi->ll_flags |= LL_SBI_NOROOTSQUASH;
2661                 else
2662                         sbi->ll_flags &= ~LL_SBI_NOROOTSQUASH;
2663         }
2664         up_write(&squash->rsi_sem);
2665 }
2666
2667 /**
2668  * Parse linkea content to extract information about a given hardlink
2669  *
2670  * \param[in]   ldata      - Initialized linkea data
2671  * \param[in]   linkno     - Link identifier
2672  * \param[out]  parent_fid - The entry's parent FID
2673  * \param[out]  ln         - Entry name destination buffer
2674  *
2675  * \retval 0 on success
2676  * \retval Appropriate negative error code on failure
2677  */
2678 static int ll_linkea_decode(struct linkea_data *ldata, unsigned int linkno,
2679                             struct lu_fid *parent_fid, struct lu_name *ln)
2680 {
2681         unsigned int    idx;
2682         int             rc;
2683         ENTRY;
2684
2685         rc = linkea_init(ldata);
2686         if (rc < 0)
2687                 RETURN(rc);
2688
2689         if (linkno >= ldata->ld_leh->leh_reccount)
2690                 /* beyond last link */
2691                 RETURN(-ENODATA);
2692
2693         linkea_first_entry(ldata);
2694         for (idx = 0; ldata->ld_lee != NULL; idx++) {
2695                 linkea_entry_unpack(ldata->ld_lee, &ldata->ld_reclen, ln,
2696                                     parent_fid);
2697                 if (idx == linkno)
2698                         break;
2699
2700                 linkea_next_entry(ldata);
2701         }
2702
2703         if (idx < linkno)
2704                 RETURN(-ENODATA);
2705
2706         RETURN(0);
2707 }
2708
2709 /**
2710  * Get parent FID and name of an identified link. Operation is performed for
2711  * a given link number, letting the caller iterate over linkno to list one or
2712  * all links of an entry.
2713  *
2714  * \param[in]     file - File descriptor against which to perform the operation
2715  * \param[in,out] arg  - User-filled structure containing the linkno to operate
2716  *                       on and the available size. It is eventually filled with
2717  *                       the requested information or left untouched on error
2718  *
2719  * \retval - 0 on success
2720  * \retval - Appropriate negative error code on failure
2721  */
2722 int ll_getparent(struct file *file, struct getparent __user *arg)
2723 {
2724         struct dentry           *dentry = file_dentry(file);
2725         struct inode            *inode = file_inode(file);
2726         struct linkea_data      *ldata;
2727         struct lu_buf            buf = LU_BUF_NULL;
2728         struct lu_name           ln;
2729         struct lu_fid            parent_fid;
2730         __u32                    linkno;
2731         __u32                    name_size;
2732         int                      rc;
2733
2734         ENTRY;
2735
2736         if (!cfs_capable(CFS_CAP_DAC_READ_SEARCH) &&
2737             !(ll_i2sbi(inode)->ll_flags & LL_SBI_USER_FID2PATH))
2738                 RETURN(-EPERM);
2739
2740         if (get_user(name_size, &arg->gp_name_size))
2741                 RETURN(-EFAULT);
2742
2743         if (get_user(linkno, &arg->gp_linkno))
2744                 RETURN(-EFAULT);
2745
2746         if (name_size > PATH_MAX)
2747                 RETURN(-EINVAL);
2748
2749         OBD_ALLOC(ldata, sizeof(*ldata));
2750         if (ldata == NULL)
2751                 RETURN(-ENOMEM);
2752
2753         rc = linkea_data_new(ldata, &buf);
2754         if (rc < 0)
2755                 GOTO(ldata_free, rc);
2756
2757         rc = ll_getxattr(dentry, XATTR_NAME_LINK, buf.lb_buf, buf.lb_len);
2758         if (rc < 0)
2759                 GOTO(lb_free, rc);
2760
2761         rc = ll_linkea_decode(ldata, linkno, &parent_fid, &ln);
2762         if (rc < 0)
2763                 GOTO(lb_free, rc);
2764
2765         if (ln.ln_namelen >= name_size)
2766                 GOTO(lb_free, rc = -EOVERFLOW);
2767
2768         if (copy_to_user(&arg->gp_fid, &parent_fid, sizeof(arg->gp_fid)))
2769                 GOTO(lb_free, rc = -EFAULT);
2770
2771         if (copy_to_user(&arg->gp_name, ln.ln_name, ln.ln_namelen))
2772                 GOTO(lb_free, rc = -EFAULT);
2773
2774         if (put_user('\0', arg->gp_name + ln.ln_namelen))
2775                 GOTO(lb_free, rc = -EFAULT);
2776
2777 lb_free:
2778         lu_buf_free(&buf);
2779 ldata_free:
2780         OBD_FREE(ldata, sizeof(*ldata));
2781
2782         RETURN(rc);
2783 }