lustre/llite/llite_lib.c

   1 /*
   2  * GPL HEADER START
   3  *
   4  * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
   5  *
   6  * This program is free software; you can redistribute it and/or modify
   7  * it under the terms of the GNU General Public License version 2 only,
   8  * as published by the Free Software Foundation.
   9  *
  10  * This program is distributed in the hope that it will be useful, but
  11  * WITHOUT ANY WARRANTY; without even the implied warranty of
  12  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  13  * General Public License version 2 for more details (a copy is included
  14  * in the LICENSE file that accompanied this code).
  15  *
  16  * You should have received a copy of the GNU General Public License
  17  * version 2 along with this program; If not, see
  18  * http://www.gnu.org/licenses/gpl-2.0.html
  19  *
  20  * GPL HEADER END
  21  */
  22 /*
  23  * Copyright (c) 2003, 2010, Oracle and/or its affiliates. All rights reserved.
  24  * Use is subject to license terms.
  25  *
  26  * Copyright (c) 2011, 2017, Intel Corporation.
  27  */
  28 /*
  29  * This file is part of Lustre, http://www.lustre.org/
  30  * Lustre is a trademark of Sun Microsystems, Inc.
  31  *
  32  * lustre/llite/llite_lib.c
  33  *
  34  * Lustre Light Super operations
  35  */
  36
  37 #define DEBUG_SUBSYSTEM S_LLITE
  38
  39 #include <linux/module.h>
  40 #include <linux/random.h>
  41 #include <linux/statfs.h>
  42 #include <linux/time.h>
  43 #include <linux/types.h>
  44 #include <libcfs/linux/linux-uuid.h>
  45 #include <linux/version.h>
  46 #include <linux/mm.h>
  47 #include <linux/user_namespace.h>
  48 #include <linux/delay.h>
  49 #include <linux/uidgid.h>
  50 #include <linux/security.h>
  51
  52 #include <uapi/linux/lustre/lustre_ioctl.h>
  53 #ifdef HAVE_UAPI_LINUX_MOUNT_H
  54 #include <uapi/linux/mount.h>
  55 #endif
  56
  57 #include <lustre_ha.h>
  58 #include <lustre_dlm.h>
  59 #include <lprocfs_status.h>
  60 #include <lustre_disk.h>
  61 #include <uapi/linux/lustre/lustre_param.h>
  62 #include <lustre_log.h>
  63 #include <cl_object.h>
  64 #include <obd_cksum.h>
  65 #include "llite_internal.h"
  66
  67 struct kmem_cache *ll_file_data_slab;
  68
  69 #ifndef log2
  70 #define log2(n) ffz(~(n))
  71 #endif
  72
  73 /**
  74  * If there is only one number of core visible to Lustre,
  75  * async readahead will be disabled, to avoid massive over
  76  * subscription, we use 1/2 of active cores as default max
  77  * async readahead requests.
  78  */
  79 static inline unsigned int ll_get_ra_async_max_active(void)
  80 {
  81         return cfs_cpt_weight(cfs_cpt_tab, CFS_CPT_ANY) >> 1;
  82 }
  83
  84 static struct ll_sb_info *ll_init_sbi(void)
  85 {
  86         struct ll_sb_info *sbi = NULL;
  87         unsigned long pages;
  88         unsigned long lru_page_max;
  89         struct sysinfo si;
  90         int rc;
  91         int i;
  92
  93         ENTRY;
  94
  95         OBD_ALLOC_PTR(sbi);
  96         if (sbi == NULL)
  97                 RETURN(ERR_PTR(-ENOMEM));
  98
  99         rc = pcc_super_init(&sbi->ll_pcc_super);
 100         if (rc < 0)
 101                 GOTO(out_sbi, rc);
 102
 103         spin_lock_init(&sbi->ll_lock);
 104         mutex_init(&sbi->ll_lco.lco_lock);
 105         spin_lock_init(&sbi->ll_pp_extent_lock);
 106         spin_lock_init(&sbi->ll_process_lock);
 107         sbi->ll_rw_stats_on = 0;
 108         sbi->ll_statfs_max_age = OBD_STATFS_CACHE_SECONDS;
 109
 110         si_meminfo(&si);
 111         pages = si.totalram - si.totalhigh;
 112         lru_page_max = pages / 2;
 113
 114         sbi->ll_ra_info.ra_async_max_active = 0;
 115         sbi->ll_ra_info.ll_readahead_wq =
 116                 alloc_workqueue("ll-readahead-wq", WQ_UNBOUND,
 117                                 sbi->ll_ra_info.ra_async_max_active);
 118         if (!sbi->ll_ra_info.ll_readahead_wq)
 119                 GOTO(out_pcc, rc = -ENOMEM);
 120
 121         /* initialize ll_cache data */
 122         sbi->ll_cache = cl_cache_init(lru_page_max);
 123         if (sbi->ll_cache == NULL)
 124                 GOTO(out_destroy_ra, rc = -ENOMEM);
 125
 126         sbi->ll_ra_info.ra_max_pages_per_file = min(pages / 32,
 127                                                     SBI_DEFAULT_READ_AHEAD_MAX);
 128         sbi->ll_ra_info.ra_async_pages_per_file_threshold =
 129                                 sbi->ll_ra_info.ra_max_pages_per_file;
 130         sbi->ll_ra_info.ra_max_pages = sbi->ll_ra_info.ra_max_pages_per_file;
 131         sbi->ll_ra_info.ra_max_read_ahead_whole_pages = -1;
 132         sbi->ll_ra_info.ra_async_max_active = ll_get_ra_async_max_active();
 133         atomic_set(&sbi->ll_ra_info.ra_async_inflight, 0);
 134
 135         sbi->ll_flags |= LL_SBI_VERBOSE;
 136 #ifdef ENABLE_CHECKSUM
 137         sbi->ll_flags |= LL_SBI_CHECKSUM;
 138 #endif
 139 #ifdef ENABLE_FLOCK
 140         sbi->ll_flags |= LL_SBI_FLOCK;
 141 #endif
 142
 143 #ifdef HAVE_LRU_RESIZE_SUPPORT
 144         sbi->ll_flags |= LL_SBI_LRU_RESIZE;
 145 #endif
 146         sbi->ll_flags |= LL_SBI_LAZYSTATFS;
 147
 148         for (i = 0; i <= LL_PROCESS_HIST_MAX; i++) {
 149                 spin_lock_init(&sbi->ll_rw_extents_info.pp_extents[i].
 150                                pp_r_hist.oh_lock);
 151                 spin_lock_init(&sbi->ll_rw_extents_info.pp_extents[i].
 152                                pp_w_hist.oh_lock);
 153         }
 154
 155         /* metadata statahead is enabled by default */
 156         sbi->ll_sa_running_max = LL_SA_RUNNING_DEF;
 157         sbi->ll_sa_max = LL_SA_RPC_DEF;
 158         atomic_set(&sbi->ll_sa_total, 0);
 159         atomic_set(&sbi->ll_sa_wrong, 0);
 160         atomic_set(&sbi->ll_sa_running, 0);
 161         atomic_set(&sbi->ll_agl_total, 0);
 162         sbi->ll_flags |= LL_SBI_AGL_ENABLED;
 163         sbi->ll_flags |= LL_SBI_FAST_READ;
 164         sbi->ll_flags |= LL_SBI_TINY_WRITE;
 165
 166         /* root squash */
 167         sbi->ll_squash.rsi_uid = 0;
 168         sbi->ll_squash.rsi_gid = 0;
 169         INIT_LIST_HEAD(&sbi->ll_squash.rsi_nosquash_nids);
 170         spin_lock_init(&sbi->ll_squash.rsi_lock);
 171
 172         /* Per-filesystem file heat */
 173         sbi->ll_heat_decay_weight = SBI_DEFAULT_HEAT_DECAY_WEIGHT;
 174         sbi->ll_heat_period_second = SBI_DEFAULT_HEAT_PERIOD_SECOND;
 175         RETURN(sbi);
 176 out_destroy_ra:
 177         destroy_workqueue(sbi->ll_ra_info.ll_readahead_wq);
 178 out_pcc:
 179         pcc_super_fini(&sbi->ll_pcc_super);
 180 out_sbi:
 181         OBD_FREE_PTR(sbi);
 182         RETURN(ERR_PTR(rc));
 183 }
 184
 185 static void ll_free_sbi(struct super_block *sb)
 186 {
 187         struct ll_sb_info *sbi = ll_s2sbi(sb);
 188         ENTRY;
 189
 190         if (sbi != NULL) {
 191                 if (!list_empty(&sbi->ll_squash.rsi_nosquash_nids))
 192                         cfs_free_nidlist(&sbi->ll_squash.rsi_nosquash_nids);
 193                 if (sbi->ll_ra_info.ll_readahead_wq)
 194                         destroy_workqueue(sbi->ll_ra_info.ll_readahead_wq);
 195                 if (sbi->ll_cache != NULL) {
 196                         cl_cache_decref(sbi->ll_cache);
 197                         sbi->ll_cache = NULL;
 198                 }
 199                 pcc_super_fini(&sbi->ll_pcc_super);
 200                 OBD_FREE(sbi, sizeof(*sbi));
 201         }
 202         EXIT;
 203 }
 204
 205 static int client_common_fill_super(struct super_block *sb, char *md, char *dt)
 206 {
 207         struct inode *root = NULL;
 208         struct ll_sb_info *sbi = ll_s2sbi(sb);
 209         struct obd_statfs *osfs = NULL;
 210         struct ptlrpc_request *request = NULL;
 211         struct obd_connect_data *data = NULL;
 212         struct obd_uuid *uuid;
 213         struct md_op_data *op_data;
 214         struct lustre_md lmd;
 215         u64 valid;
 216         int size, err, checksum;
 217
 218         ENTRY;
 219         sbi->ll_md_obd = class_name2obd(md);
 220         if (!sbi->ll_md_obd) {
 221                 CERROR("MD %s: not setup or attached\n", md);
 222                 RETURN(-EINVAL);
 223         }
 224
 225         OBD_ALLOC_PTR(data);
 226         if (data == NULL)
 227                 RETURN(-ENOMEM);
 228
 229         OBD_ALLOC_PTR(osfs);
 230         if (osfs == NULL) {
 231                 OBD_FREE_PTR(data);
 232                 RETURN(-ENOMEM);
 233         }
 234
 235         /* pass client page size via ocd_grant_blkbits, the server should report
 236          * back its backend blocksize for grant calculation purpose */
 237         data->ocd_grant_blkbits = PAGE_SHIFT;
 238
 239         /* indicate MDT features supported by this client */
 240         data->ocd_connect_flags = OBD_CONNECT_IBITS    | OBD_CONNECT_NODEVOH  |
 241                                   OBD_CONNECT_ATTRFID  | OBD_CONNECT_GRANT |
 242                                   OBD_CONNECT_VERSION  | OBD_CONNECT_BRW_SIZE |
 243                                   OBD_CONNECT_SRVLOCK  | OBD_CONNECT_TRUNCLOCK|
 244                                   OBD_CONNECT_MDS_CAPA | OBD_CONNECT_OSS_CAPA |
 245                                   OBD_CONNECT_CANCELSET | OBD_CONNECT_FID     |
 246                                   OBD_CONNECT_AT       | OBD_CONNECT_LOV_V3   |
 247                                   OBD_CONNECT_VBR | OBD_CONNECT_FULL20 |
 248                                   OBD_CONNECT_64BITHASH |
 249                                   OBD_CONNECT_EINPROGRESS |
 250                                   OBD_CONNECT_JOBSTATS | OBD_CONNECT_LVB_TYPE |
 251                                   OBD_CONNECT_LAYOUTLOCK | OBD_CONNECT_PINGLESS|
 252                                   OBD_CONNECT_MAX_EASIZE |
 253                                   OBD_CONNECT_FLOCK_DEAD |
 254                                   OBD_CONNECT_DISP_STRIPE | OBD_CONNECT_LFSCK |
 255                                   OBD_CONNECT_OPEN_BY_FID |
 256                                   OBD_CONNECT_DIR_STRIPE |
 257                                   OBD_CONNECT_BULK_MBITS | OBD_CONNECT_CKSUM |
 258                                   OBD_CONNECT_SUBTREE |
 259                                   OBD_CONNECT_MULTIMODRPCS |
 260                                   OBD_CONNECT_GRANT_PARAM |
 261                                   OBD_CONNECT_SHORTIO | OBD_CONNECT_FLAGS2;
 262
 263         data->ocd_connect_flags2 = OBD_CONNECT2_DIR_MIGRATE |
 264                                    OBD_CONNECT2_SUM_STATFS |
 265                                    OBD_CONNECT2_OVERSTRIPING |
 266                                    OBD_CONNECT2_FLR |
 267                                    OBD_CONNECT2_LOCK_CONVERT |
 268                                    OBD_CONNECT2_ARCHIVE_ID_ARRAY |
 269                                    OBD_CONNECT2_INC_XID |
 270                                    OBD_CONNECT2_LSOM |
 271                                    OBD_CONNECT2_ASYNC_DISCARD |
 272                                    OBD_CONNECT2_PCC |
 273                                    OBD_CONNECT2_CRUSH;
 274
 275 #ifdef HAVE_LRU_RESIZE_SUPPORT
 276         if (sbi->ll_flags & LL_SBI_LRU_RESIZE)
 277                 data->ocd_connect_flags |= OBD_CONNECT_LRU_RESIZE;
 278 #endif
 279 #ifdef CONFIG_LUSTRE_FS_POSIX_ACL
 280         data->ocd_connect_flags |= OBD_CONNECT_ACL | OBD_CONNECT_UMASK |
 281                                    OBD_CONNECT_LARGE_ACL;
 282 #endif
 283
 284         data->ocd_cksum_types = obd_cksum_types_supported_client();
 285
 286         if (OBD_FAIL_CHECK(OBD_FAIL_MDC_LIGHTWEIGHT))
 287                 /* flag mdc connection as lightweight, only used for test
 288                  * purpose, use with care */
 289                 data->ocd_connect_flags |= OBD_CONNECT_LIGHTWEIGHT;
 290
 291         data->ocd_ibits_known = MDS_INODELOCK_FULL;
 292         data->ocd_version = LUSTRE_VERSION_CODE;
 293
 294         if (sb->s_flags & SB_RDONLY)
 295                 data->ocd_connect_flags |= OBD_CONNECT_RDONLY;
 296         if (sbi->ll_flags & LL_SBI_USER_XATTR)
 297                 data->ocd_connect_flags |= OBD_CONNECT_XATTR;
 298
 299 #ifdef SB_NOSEC
 300         /* Setting this indicates we correctly support S_NOSEC (See kernel
 301          * commit 9e1f1de02c2275d7172e18dc4e7c2065777611bf)
 302          */
 303         sb->s_flags |= SB_NOSEC;
 304 #endif
 305
 306         if (sbi->ll_flags & LL_SBI_FLOCK)
 307                 sbi->ll_fop = &ll_file_operations_flock;
 308         else if (sbi->ll_flags & LL_SBI_LOCALFLOCK)
 309                 sbi->ll_fop = &ll_file_operations;
 310         else
 311                 sbi->ll_fop = &ll_file_operations_noflock;
 312
 313         /* always ping even if server suppress_pings */
 314         if (sbi->ll_flags & LL_SBI_ALWAYS_PING)
 315                 data->ocd_connect_flags &= ~OBD_CONNECT_PINGLESS;
 316
 317         obd_connect_set_secctx(data);
 318
 319 #if defined(CONFIG_SECURITY)
 320         data->ocd_connect_flags2 |= OBD_CONNECT2_SELINUX_POLICY;
 321 #endif
 322
 323         data->ocd_brw_size = MD_MAX_BRW_SIZE;
 324
 325         err = obd_connect(NULL, &sbi->ll_md_exp, sbi->ll_md_obd,
 326                           &sbi->ll_sb_uuid, data, sbi->ll_cache);
 327         if (err == -EBUSY) {
 328                 LCONSOLE_ERROR_MSG(0x14f, "An MDT (md %s) is performing "
 329                                    "recovery, of which this client is not a "
 330                                    "part. Please wait for recovery to complete,"
 331                                    " abort, or time out.\n", md);
 332                 GOTO(out, err);
 333         } else if (err) {
 334                 CERROR("cannot connect to %s: rc = %d\n", md, err);
 335                 GOTO(out, err);
 336         }
 337
 338         sbi->ll_md_exp->exp_connect_data = *data;
 339
 340         err = obd_fid_init(sbi->ll_md_exp->exp_obd, sbi->ll_md_exp,
 341                            LUSTRE_SEQ_METADATA);
 342         if (err) {
 343                 CERROR("%s: Can't init metadata layer FID infrastructure, "
 344                        "rc = %d\n", sbi->ll_md_exp->exp_obd->obd_name, err);
 345                 GOTO(out_md, err);
 346         }
 347
 348         /* For mount, we only need fs info from MDT0, and also in DNE, it
 349          * can make sure the client can be mounted as long as MDT0 is
 350          * avaible */
 351         err = obd_statfs(NULL, sbi->ll_md_exp, osfs,
 352                         ktime_get_seconds() - sbi->ll_statfs_max_age,
 353                         OBD_STATFS_FOR_MDT0);
 354         if (err)
 355                 GOTO(out_md_fid, err);
 356
 357         /* This needs to be after statfs to ensure connect has finished.
 358          * Note that "data" does NOT contain the valid connect reply.
 359          * If connecting to a 1.8 server there will be no LMV device, so
 360          * we can access the MDC export directly and exp_connect_flags will
 361          * be non-zero, but if accessing an upgraded 2.1 server it will
 362          * have the correct flags filled in.
 363          * XXX: fill in the LMV exp_connect_flags from MDC(s). */
 364         valid = exp_connect_flags(sbi->ll_md_exp) & CLIENT_CONNECT_MDT_REQD;
 365         if (exp_connect_flags(sbi->ll_md_exp) != 0 &&
 366             valid != CLIENT_CONNECT_MDT_REQD) {
 367                 char *buf;
 368
 369                 OBD_ALLOC_WAIT(buf, PAGE_SIZE);
 370                 obd_connect_flags2str(buf, PAGE_SIZE,
 371                                       valid ^ CLIENT_CONNECT_MDT_REQD, 0, ",");
 372                 LCONSOLE_ERROR_MSG(0x170, "Server %s does not support "
 373                                    "feature(s) needed for correct operation "
 374                                    "of this client (%s). Please upgrade "
 375                                    "server or downgrade client.\n",
 376                                    sbi->ll_md_exp->exp_obd->obd_name, buf);
 377                 OBD_FREE(buf, PAGE_SIZE);
 378                 GOTO(out_md_fid, err = -EPROTO);
 379         }
 380
 381         size = sizeof(*data);
 382         err = obd_get_info(NULL, sbi->ll_md_exp, sizeof(KEY_CONN_DATA),
 383                            KEY_CONN_DATA,  &size, data);
 384         if (err) {
 385                 CERROR("%s: Get connect data failed: rc = %d\n",
 386                        sbi->ll_md_exp->exp_obd->obd_name, err);
 387                 GOTO(out_md_fid, err);
 388         }
 389
 390         LASSERT(osfs->os_bsize);
 391         sb->s_blocksize = osfs->os_bsize;
 392         sb->s_blocksize_bits = log2(osfs->os_bsize);
 393         sb->s_magic = LL_SUPER_MAGIC;
 394         sb->s_maxbytes = MAX_LFS_FILESIZE;
 395         sbi->ll_namelen = osfs->os_namelen;
 396         sbi->ll_mnt.mnt = current->fs->root.mnt;
 397
 398         if ((sbi->ll_flags & LL_SBI_USER_XATTR) &&
 399             !(data->ocd_connect_flags & OBD_CONNECT_XATTR)) {
 400                 LCONSOLE_INFO("Disabling user_xattr feature because "
 401                               "it is not supported on the server\n");
 402                 sbi->ll_flags &= ~LL_SBI_USER_XATTR;
 403         }
 404
 405         if (data->ocd_connect_flags & OBD_CONNECT_ACL) {
 406 #ifdef SB_POSIXACL
 407                 sb->s_flags |= SB_POSIXACL;
 408 #endif
 409                 sbi->ll_flags |= LL_SBI_ACL;
 410         } else {
 411                 LCONSOLE_INFO("client wants to enable acl, but mdt not!\n");
 412 #ifdef SB_POSIXACL
 413                 sb->s_flags &= ~SB_POSIXACL;
 414 #endif
 415                 sbi->ll_flags &= ~LL_SBI_ACL;
 416         }
 417
 418         if (data->ocd_connect_flags & OBD_CONNECT_64BITHASH)
 419                 sbi->ll_flags |= LL_SBI_64BIT_HASH;
 420
 421         if (data->ocd_connect_flags & OBD_CONNECT_LAYOUTLOCK)
 422                 sbi->ll_flags |= LL_SBI_LAYOUT_LOCK;
 423
 424         if (obd_connect_has_secctx(data))
 425                 sbi->ll_flags |= LL_SBI_FILE_SECCTX;
 426
 427         if (data->ocd_ibits_known & MDS_INODELOCK_XATTR) {
 428                 if (!(data->ocd_connect_flags & OBD_CONNECT_MAX_EASIZE)) {
 429                         LCONSOLE_INFO("%s: disabling xattr cache due to "
 430                                       "unknown maximum xattr size.\n", dt);
 431                 } else if (!sbi->ll_xattr_cache_set) {
 432                         /* If xattr_cache is already set (no matter 0 or 1)
 433                          * during processing llog, it won't be enabled here. */
 434                         sbi->ll_flags |= LL_SBI_XATTR_CACHE;
 435                         sbi->ll_xattr_cache_enabled = 1;
 436                 }
 437         }
 438
 439         sbi->ll_dt_obd = class_name2obd(dt);
 440         if (!sbi->ll_dt_obd) {
 441                 CERROR("DT %s: not setup or attached\n", dt);
 442                 GOTO(out_md_fid, err = -ENODEV);
 443         }
 444
 445         /* pass client page size via ocd_grant_blkbits, the server should report
 446          * back its backend blocksize for grant calculation purpose */
 447         data->ocd_grant_blkbits = PAGE_SHIFT;
 448
 449         /* indicate OST features supported by this client */
 450         data->ocd_connect_flags = OBD_CONNECT_GRANT | OBD_CONNECT_VERSION |
 451                                   OBD_CONNECT_REQPORTAL | OBD_CONNECT_BRW_SIZE |
 452                                   OBD_CONNECT_CANCELSET | OBD_CONNECT_FID |
 453                                   OBD_CONNECT_SRVLOCK | OBD_CONNECT_TRUNCLOCK|
 454                                   OBD_CONNECT_AT | OBD_CONNECT_OSS_CAPA |
 455                                   OBD_CONNECT_VBR | OBD_CONNECT_FULL20 |
 456                                   OBD_CONNECT_64BITHASH | OBD_CONNECT_MAXBYTES |
 457                                   OBD_CONNECT_EINPROGRESS |
 458                                   OBD_CONNECT_JOBSTATS | OBD_CONNECT_LVB_TYPE |
 459                                   OBD_CONNECT_LAYOUTLOCK |
 460                                   OBD_CONNECT_PINGLESS | OBD_CONNECT_LFSCK |
 461                                   OBD_CONNECT_BULK_MBITS | OBD_CONNECT_SHORTIO |
 462                                   OBD_CONNECT_FLAGS2 | OBD_CONNECT_GRANT_SHRINK;
 463
 464 /* The client currently advertises support for OBD_CONNECT_LOCKAHEAD_OLD so it
 465  * can interoperate with an older version of lockahead which was released prior
 466  * to landing in master. This support will be dropped when 2.13 development
 467  * starts.  At the point, we should not just drop the connect flag (below), we
 468  * should also remove the support in the code.
 469  *
 470  * Removing it means a few things:
 471  * 1. Remove this section here
 472  * 2. Remove CEF_NONBLOCK in ll_file_lockahead()
 473  * 3. Remove function exp_connect_lockahead_old
 474  * 4. Remove LDLM_FL_LOCKAHEAD_OLD_RESERVED in lustre_dlm_flags.h
 475  * */
 476 #if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(2, 12, 50, 0)
 477         data->ocd_connect_flags |= OBD_CONNECT_LOCKAHEAD_OLD;
 478 #endif
 479
 480         data->ocd_connect_flags2 = OBD_CONNECT2_LOCKAHEAD |
 481                                    OBD_CONNECT2_INC_XID;
 482
 483         if (!OBD_FAIL_CHECK(OBD_FAIL_OSC_CONNECT_GRANT_PARAM))
 484                 data->ocd_connect_flags |= OBD_CONNECT_GRANT_PARAM;
 485
 486         /* OBD_CONNECT_CKSUM should always be set, even if checksums are
 487          * disabled by default, because it can still be enabled on the
 488          * fly via /sys. As a consequence, we still need to come to an
 489          * agreement on the supported algorithms at connect time
 490          */
 491         data->ocd_connect_flags |= OBD_CONNECT_CKSUM;
 492
 493         if (OBD_FAIL_CHECK(OBD_FAIL_OSC_CKSUM_ADLER_ONLY))
 494                 data->ocd_cksum_types = OBD_CKSUM_ADLER;
 495         else
 496                 data->ocd_cksum_types = obd_cksum_types_supported_client();
 497
 498 #ifdef HAVE_LRU_RESIZE_SUPPORT
 499         data->ocd_connect_flags |= OBD_CONNECT_LRU_RESIZE;
 500 #endif
 501         /* always ping even if server suppress_pings */
 502         if (sbi->ll_flags & LL_SBI_ALWAYS_PING)
 503                 data->ocd_connect_flags &= ~OBD_CONNECT_PINGLESS;
 504
 505         CDEBUG(D_RPCTRACE, "ocd_connect_flags: %#llx ocd_version: %d "
 506                "ocd_grant: %d\n", data->ocd_connect_flags,
 507                data->ocd_version, data->ocd_grant);
 508
 509         sbi->ll_dt_obd->obd_upcall.onu_owner = &sbi->ll_lco;
 510         sbi->ll_dt_obd->obd_upcall.onu_upcall = cl_ocd_update;
 511
 512         data->ocd_brw_size = DT_MAX_BRW_SIZE;
 513
 514         err = obd_connect(NULL, &sbi->ll_dt_exp, sbi->ll_dt_obd,
 515                           &sbi->ll_sb_uuid, data, sbi->ll_cache);
 516         if (err == -EBUSY) {
 517                 LCONSOLE_ERROR_MSG(0x150, "An OST (dt %s) is performing "
 518                                    "recovery, of which this client is not a "
 519                                    "part.  Please wait for recovery to "
 520                                    "complete, abort, or time out.\n", dt);
 521                 GOTO(out_md, err);
 522         } else if (err) {
 523                 CERROR("%s: Cannot connect to %s: rc = %d\n",
 524                        sbi->ll_dt_exp->exp_obd->obd_name, dt, err);
 525                 GOTO(out_md, err);
 526         }
 527
 528         sbi->ll_dt_exp->exp_connect_data = *data;
 529
 530         /* Don't change value if it was specified in the config log */
 531         if (sbi->ll_ra_info.ra_max_read_ahead_whole_pages == -1) {
 532                 sbi->ll_ra_info.ra_max_read_ahead_whole_pages =
 533                         max_t(unsigned long, SBI_DEFAULT_READ_AHEAD_WHOLE_MAX,
 534                               (data->ocd_brw_size >> PAGE_SHIFT));
 535                 if (sbi->ll_ra_info.ra_max_read_ahead_whole_pages >
 536                     sbi->ll_ra_info.ra_max_pages_per_file)
 537                         sbi->ll_ra_info.ra_max_read_ahead_whole_pages =
 538                                 sbi->ll_ra_info.ra_max_pages_per_file;
 539         }
 540
 541         err = obd_fid_init(sbi->ll_dt_exp->exp_obd, sbi->ll_dt_exp,
 542                            LUSTRE_SEQ_METADATA);
 543         if (err) {
 544                 CERROR("%s: Can't init data layer FID infrastructure, "
 545                        "rc = %d\n", sbi->ll_dt_exp->exp_obd->obd_name, err);
 546                 GOTO(out_dt, err);
 547         }
 548
 549         mutex_lock(&sbi->ll_lco.lco_lock);
 550         sbi->ll_lco.lco_flags = data->ocd_connect_flags;
 551         sbi->ll_lco.lco_md_exp = sbi->ll_md_exp;
 552         sbi->ll_lco.lco_dt_exp = sbi->ll_dt_exp;
 553         mutex_unlock(&sbi->ll_lco.lco_lock);
 554
 555         fid_zero(&sbi->ll_root_fid);
 556         err = md_get_root(sbi->ll_md_exp, get_mount_fileset(sb),
 557                            &sbi->ll_root_fid);
 558         if (err) {
 559                 CERROR("cannot mds_connect: rc = %d\n", err);
 560                 GOTO(out_lock_cn_cb, err);
 561         }
 562         if (!fid_is_sane(&sbi->ll_root_fid)) {
 563                 CERROR("%s: Invalid root fid "DFID" during mount\n",
 564                        sbi->ll_md_exp->exp_obd->obd_name,
 565                        PFID(&sbi->ll_root_fid));
 566                 GOTO(out_lock_cn_cb, err = -EINVAL);
 567         }
 568         CDEBUG(D_SUPER, "rootfid "DFID"\n", PFID(&sbi->ll_root_fid));
 569
 570         sb->s_op = &lustre_super_operations;
 571         sb->s_xattr = ll_xattr_handlers;
 572 #if THREAD_SIZE >= 8192 /*b=17630*/
 573         sb->s_export_op = &lustre_export_operations;
 574 #endif
 575
 576         /* make root inode
 577          * XXX: move this to after cbd setup? */
 578         valid = OBD_MD_FLGETATTR | OBD_MD_FLBLOCKS | OBD_MD_FLMODEASIZE;
 579         if (sbi->ll_flags & LL_SBI_ACL)
 580                 valid |= OBD_MD_FLACL;
 581
 582         OBD_ALLOC_PTR(op_data);
 583         if (op_data == NULL)
 584                 GOTO(out_lock_cn_cb, err = -ENOMEM);
 585
 586         op_data->op_fid1 = sbi->ll_root_fid;
 587         op_data->op_mode = 0;
 588         op_data->op_valid = valid;
 589
 590         err = md_getattr(sbi->ll_md_exp, op_data, &request);
 591
 592         OBD_FREE_PTR(op_data);
 593         if (err) {
 594                 CERROR("%s: md_getattr failed for root: rc = %d\n",
 595                        sbi->ll_md_exp->exp_obd->obd_name, err);
 596                 GOTO(out_lock_cn_cb, err);
 597         }
 598
 599         err = md_get_lustre_md(sbi->ll_md_exp, request, sbi->ll_dt_exp,
 600                                sbi->ll_md_exp, &lmd);
 601         if (err) {
 602                 CERROR("failed to understand root inode md: rc = %d\n", err);
 603                 ptlrpc_req_finished(request);
 604                 GOTO(out_lock_cn_cb, err);
 605         }
 606
 607         LASSERT(fid_is_sane(&sbi->ll_root_fid));
 608         root = ll_iget(sb, cl_fid_build_ino(&sbi->ll_root_fid,
 609                                             sbi->ll_flags & LL_SBI_32BIT_API),
 610                        &lmd);
 611         md_free_lustre_md(sbi->ll_md_exp, &lmd);
 612         ptlrpc_req_finished(request);
 613
 614         if (IS_ERR(root)) {
 615 #ifdef CONFIG_LUSTRE_FS_POSIX_ACL
 616                 if (lmd.posix_acl) {
 617                         posix_acl_release(lmd.posix_acl);
 618                         lmd.posix_acl = NULL;
 619                 }
 620 #endif
 621                 err = IS_ERR(root) ? PTR_ERR(root) : -EBADF;
 622                 root = NULL;
 623                 CERROR("%s: bad ll_iget() for root: rc = %d\n",
 624                        sbi->ll_fsname, err);
 625                 GOTO(out_root, err);
 626         }
 627
 628         checksum = sbi->ll_flags & LL_SBI_CHECKSUM;
 629         if (sbi->ll_checksum_set) {
 630                 err = obd_set_info_async(NULL, sbi->ll_dt_exp,
 631                                          sizeof(KEY_CHECKSUM), KEY_CHECKSUM,
 632                                          sizeof(checksum), &checksum, NULL);
 633                 if (err) {
 634                         CERROR("%s: Set checksum failed: rc = %d\n",
 635                                sbi->ll_dt_exp->exp_obd->obd_name, err);
 636                         GOTO(out_root, err);
 637                 }
 638         }
 639         cl_sb_init(sb);
 640
 641         sb->s_root = d_make_root(root);
 642         if (sb->s_root == NULL) {
 643                 err = -ENOMEM;
 644                 CERROR("%s: can't make root dentry: rc = %d\n",
 645                        sbi->ll_fsname, err);
 646                 GOTO(out_root, err);
 647         }
 648
 649         sbi->ll_sdev_orig = sb->s_dev;
 650
 651         /* We set sb->s_dev equal on all lustre clients in order to support
 652          * NFS export clustering.  NFSD requires that the FSID be the same
 653          * on all clients. */
 654         /* s_dev is also used in lt_compare() to compare two fs, but that is
 655          * only a node-local comparison. */
 656         uuid = obd_get_uuid(sbi->ll_md_exp);
 657         if (uuid != NULL)
 658                 sb->s_dev = get_uuid2int(uuid->uuid, strlen(uuid->uuid));
 659
 660         if (data != NULL)
 661                 OBD_FREE_PTR(data);
 662         if (osfs != NULL)
 663                 OBD_FREE_PTR(osfs);
 664
 665         if (sbi->ll_dt_obd) {
 666                 err = sysfs_create_link(&sbi->ll_kset.kobj,
 667                                         &sbi->ll_dt_obd->obd_kset.kobj,
 668                                         sbi->ll_dt_obd->obd_type->typ_name);
 669                 if (err < 0) {
 670                         CERROR("%s: could not register %s in llite: rc = %d\n",
 671                                dt, sbi->ll_fsname, err);
 672                         err = 0;
 673                 }
 674         }
 675
 676         if (sbi->ll_md_obd) {
 677                 err = sysfs_create_link(&sbi->ll_kset.kobj,
 678                                         &sbi->ll_md_obd->obd_kset.kobj,
 679                                         sbi->ll_md_obd->obd_type->typ_name);
 680                 if (err < 0) {
 681                         CERROR("%s: could not register %s in llite: rc = %d\n",
 682                                md, sbi->ll_fsname, err);
 683                         err = 0;
 684                 }
 685         }
 686
 687         RETURN(err);
 688 out_root:
 689         if (root)
 690                 iput(root);
 691 out_lock_cn_cb:
 692         obd_fid_fini(sbi->ll_dt_exp->exp_obd);
 693 out_dt:
 694         obd_disconnect(sbi->ll_dt_exp);
 695         sbi->ll_dt_exp = NULL;
 696         sbi->ll_dt_obd = NULL;
 697 out_md_fid:
 698         obd_fid_fini(sbi->ll_md_exp->exp_obd);
 699 out_md:
 700         obd_disconnect(sbi->ll_md_exp);
 701         sbi->ll_md_exp = NULL;
 702         sbi->ll_md_obd = NULL;
 703 out:
 704         if (data != NULL)
 705                 OBD_FREE_PTR(data);
 706         if (osfs != NULL)
 707                 OBD_FREE_PTR(osfs);
 708         return err;
 709 }
 710
 711 int ll_get_max_mdsize(struct ll_sb_info *sbi, int *lmmsize)
 712 {
 713         int size, rc;
 714
 715         size = sizeof(*lmmsize);
 716         rc = obd_get_info(NULL, sbi->ll_dt_exp, sizeof(KEY_MAX_EASIZE),
 717                           KEY_MAX_EASIZE, &size, lmmsize);
 718         if (rc != 0) {
 719                 CERROR("%s: cannot get max LOV EA size: rc = %d\n",
 720                        sbi->ll_dt_exp->exp_obd->obd_name, rc);
 721                 RETURN(rc);
 722         }
 723
 724         CDEBUG(D_INFO, "max LOV ea size: %d\n", *lmmsize);
 725
 726         size = sizeof(int);
 727         rc = obd_get_info(NULL, sbi->ll_md_exp, sizeof(KEY_MAX_EASIZE),
 728                           KEY_MAX_EASIZE, &size, lmmsize);
 729         if (rc)
 730                 CERROR("Get max mdsize error rc %d\n", rc);
 731
 732         CDEBUG(D_INFO, "max LMV ea size: %d\n", *lmmsize);
 733
 734         RETURN(rc);
 735 }
 736
 737 /**
 738  * Get the value of the default_easize parameter.
 739  *
 740  * \see client_obd::cl_default_mds_easize
 741  *
 742  * \param[in] sbi       superblock info for this filesystem
 743  * \param[out] lmmsize  pointer to storage location for value
 744  *
 745  * \retval 0            on success
 746  * \retval negative     negated errno on failure
 747  */
 748 int ll_get_default_mdsize(struct ll_sb_info *sbi, int *lmmsize)
 749 {
 750         int size, rc;
 751
 752         size = sizeof(int);
 753         rc = obd_get_info(NULL, sbi->ll_md_exp, sizeof(KEY_DEFAULT_EASIZE),
 754                          KEY_DEFAULT_EASIZE, &size, lmmsize);
 755         if (rc)
 756                 CERROR("Get default mdsize error rc %d\n", rc);
 757
 758         RETURN(rc);
 759 }
 760
 761 /**
 762  * Set the default_easize parameter to the given value.
 763  *
 764  * \see client_obd::cl_default_mds_easize
 765  *
 766  * \param[in] sbi       superblock info for this filesystem
 767  * \param[in] lmmsize   the size to set
 768  *
 769  * \retval 0            on success
 770  * \retval negative     negated errno on failure
 771  */
 772 int ll_set_default_mdsize(struct ll_sb_info *sbi, int lmmsize)
 773 {
 774         int rc;
 775
 776         if (lmmsize < sizeof(struct lov_mds_md) ||
 777             lmmsize > OBD_MAX_DEFAULT_EA_SIZE)
 778                 return -EINVAL;
 779
 780         rc = obd_set_info_async(NULL, sbi->ll_md_exp,
 781                                 sizeof(KEY_DEFAULT_EASIZE), KEY_DEFAULT_EASIZE,
 782                                 sizeof(int), &lmmsize, NULL);
 783
 784         RETURN(rc);
 785 }
 786
 787 static void client_common_put_super(struct super_block *sb)
 788 {
 789         struct ll_sb_info *sbi = ll_s2sbi(sb);
 790         ENTRY;
 791
 792         cl_sb_fini(sb);
 793
 794         obd_fid_fini(sbi->ll_dt_exp->exp_obd);
 795         obd_disconnect(sbi->ll_dt_exp);
 796         sbi->ll_dt_exp = NULL;
 797
 798         ll_debugfs_unregister_super(sb);
 799
 800         obd_fid_fini(sbi->ll_md_exp->exp_obd);
 801         obd_disconnect(sbi->ll_md_exp);
 802         sbi->ll_md_exp = NULL;
 803
 804         EXIT;
 805 }
 806
 807 void ll_kill_super(struct super_block *sb)
 808 {
 809         struct ll_sb_info *sbi;
 810         ENTRY;
 811
 812         /* not init sb ?*/
 813         if (!(sb->s_flags & SB_ACTIVE))
 814                 return;
 815
 816         sbi = ll_s2sbi(sb);
 817         /* we need restore s_dev from changed for clustred NFS before put_super
 818          * because new kernels have cached s_dev and change sb->s_dev in
 819          * put_super not affected real removing devices */
 820         if (sbi) {
 821                 sb->s_dev = sbi->ll_sdev_orig;
 822
 823                 /* wait running statahead threads to quit */
 824                 while (atomic_read(&sbi->ll_sa_running) > 0)
 825                         schedule_timeout_uninterruptible(
 826                                 cfs_time_seconds(1) >> 3);
 827         }
 828
 829         EXIT;
 830 }
 831
 832 static inline int ll_set_opt(const char *opt, char *data, int fl)
 833 {
 834         if (strncmp(opt, data, strlen(opt)) != 0)
 835                 return 0;
 836         else
 837                 return fl;
 838 }
 839
 840 /* non-client-specific mount options are parsed in lmd_parse */
 841 static int ll_options(char *options, struct ll_sb_info *sbi)
 842 {
 843         int tmp;
 844         char *s1 = options, *s2;
 845         int *flags = &sbi->ll_flags;
 846         ENTRY;
 847
 848         if (!options)
 849                 RETURN(0);
 850
 851         CDEBUG(D_CONFIG, "Parsing opts %s\n", options);
 852
 853         while (*s1) {
 854                 CDEBUG(D_SUPER, "next opt=%s\n", s1);
 855                 tmp = ll_set_opt("nolock", s1, LL_SBI_NOLCK);
 856                 if (tmp) {
 857                         *flags |= tmp;
 858                         goto next;
 859                 }
 860                 tmp = ll_set_opt("flock", s1, LL_SBI_FLOCK);
 861                 if (tmp) {
 862                         *flags = (*flags & ~LL_SBI_LOCALFLOCK) | tmp;
 863                         goto next;
 864                 }
 865                 tmp = ll_set_opt("localflock", s1, LL_SBI_LOCALFLOCK);
 866                 if (tmp) {
 867                         *flags = (*flags & ~LL_SBI_FLOCK) | tmp;
 868                         goto next;
 869                 }
 870                 tmp = ll_set_opt("noflock", s1, LL_SBI_FLOCK|LL_SBI_LOCALFLOCK);
 871                 if (tmp) {
 872                         *flags &= ~tmp;
 873                         goto next;
 874                 }
 875                 tmp = ll_set_opt("user_xattr", s1, LL_SBI_USER_XATTR);
 876                 if (tmp) {
 877                         *flags |= tmp;
 878                         goto next;
 879                 }
 880                 tmp = ll_set_opt("nouser_xattr", s1, LL_SBI_USER_XATTR);
 881                 if (tmp) {
 882                         *flags &= ~tmp;
 883                         goto next;
 884                 }
 885                 tmp = ll_set_opt("context", s1, 1);
 886                 if (tmp)
 887                         goto next;
 888                 tmp = ll_set_opt("fscontext", s1, 1);
 889                 if (tmp)
 890                         goto next;
 891                 tmp = ll_set_opt("defcontext", s1, 1);
 892                 if (tmp)
 893                         goto next;
 894                 tmp = ll_set_opt("rootcontext", s1, 1);
 895                 if (tmp)
 896                         goto next;
 897                 tmp = ll_set_opt("user_fid2path", s1, LL_SBI_USER_FID2PATH);
 898                 if (tmp) {
 899                         *flags |= tmp;
 900                         goto next;
 901                 }
 902                 tmp = ll_set_opt("nouser_fid2path", s1, LL_SBI_USER_FID2PATH);
 903                 if (tmp) {
 904                         *flags &= ~tmp;
 905                         goto next;
 906                 }
 907
 908                 tmp = ll_set_opt("checksum", s1, LL_SBI_CHECKSUM);
 909                 if (tmp) {
 910                         *flags |= tmp;
 911                         sbi->ll_checksum_set = 1;
 912                         goto next;
 913                 }
 914                 tmp = ll_set_opt("nochecksum", s1, LL_SBI_CHECKSUM);
 915                 if (tmp) {
 916                         *flags &= ~tmp;
 917                         sbi->ll_checksum_set = 1;
 918                         goto next;
 919                 }
 920                 tmp = ll_set_opt("lruresize", s1, LL_SBI_LRU_RESIZE);
 921                 if (tmp) {
 922                         *flags |= tmp;
 923                         goto next;
 924                 }
 925                 tmp = ll_set_opt("nolruresize", s1, LL_SBI_LRU_RESIZE);
 926                 if (tmp) {
 927                         *flags &= ~tmp;
 928                         goto next;
 929                 }
 930                 tmp = ll_set_opt("lazystatfs", s1, LL_SBI_LAZYSTATFS);
 931                 if (tmp) {
 932                         *flags |= tmp;
 933                         goto next;
 934                 }
 935                 tmp = ll_set_opt("nolazystatfs", s1, LL_SBI_LAZYSTATFS);
 936                 if (tmp) {
 937                         *flags &= ~tmp;
 938                         goto next;
 939                 }
 940                 tmp = ll_set_opt("32bitapi", s1, LL_SBI_32BIT_API);
 941                 if (tmp) {
 942                         *flags |= tmp;
 943                         goto next;
 944                 }
 945                 tmp = ll_set_opt("verbose", s1, LL_SBI_VERBOSE);
 946                 if (tmp) {
 947                         *flags |= tmp;
 948                         goto next;
 949                 }
 950                 tmp = ll_set_opt("noverbose", s1, LL_SBI_VERBOSE);
 951                 if (tmp) {
 952                         *flags &= ~tmp;
 953                         goto next;
 954                 }
 955                 tmp = ll_set_opt("always_ping", s1, LL_SBI_ALWAYS_PING);
 956                 if (tmp) {
 957                         *flags |= tmp;
 958                         goto next;
 959                 }
 960                 LCONSOLE_ERROR_MSG(0x152, "Unknown option '%s', won't mount.\n",
 961                                    s1);
 962                 RETURN(-EINVAL);
 963
 964 next:
 965                 /* Find next opt */
 966                 s2 = strchr(s1, ',');
 967                 if (s2 == NULL)
 968                         break;
 969                 s1 = s2 + 1;
 970         }
 971         RETURN(0);
 972 }
 973
 974 void ll_lli_init(struct ll_inode_info *lli)
 975 {
 976         lli->lli_inode_magic = LLI_INODE_MAGIC;
 977         lli->lli_flags = 0;
 978         spin_lock_init(&lli->lli_lock);
 979         lli->lli_posix_acl = NULL;
 980         /* Do not set lli_fid, it has been initialized already. */
 981         fid_zero(&lli->lli_pfid);
 982         lli->lli_mds_read_och = NULL;
 983         lli->lli_mds_write_och = NULL;
 984         lli->lli_mds_exec_och = NULL;
 985         lli->lli_open_fd_read_count = 0;
 986         lli->lli_open_fd_write_count = 0;
 987         lli->lli_open_fd_exec_count = 0;
 988         mutex_init(&lli->lli_och_mutex);
 989         spin_lock_init(&lli->lli_agl_lock);
 990         spin_lock_init(&lli->lli_layout_lock);
 991         ll_layout_version_set(lli, CL_LAYOUT_GEN_NONE);
 992         lli->lli_clob = NULL;
 993
 994         init_rwsem(&lli->lli_xattrs_list_rwsem);
 995         mutex_init(&lli->lli_xattrs_enq_lock);
 996
 997         LASSERT(lli->lli_vfs_inode.i_mode != 0);
 998         if (S_ISDIR(lli->lli_vfs_inode.i_mode)) {
 999                 lli->lli_opendir_key = NULL;
1000                 lli->lli_sai = NULL;
1001                 spin_lock_init(&lli->lli_sa_lock);
1002                 lli->lli_opendir_pid = 0;
1003                 lli->lli_sa_enabled = 0;
1004                 init_rwsem(&lli->lli_lsm_sem);
1005         } else {
1006                 mutex_init(&lli->lli_size_mutex);
1007                 lli->lli_symlink_name = NULL;
1008                 ll_trunc_sem_init(&lli->lli_trunc_sem);
1009                 range_lock_tree_init(&lli->lli_write_tree);
1010                 init_rwsem(&lli->lli_glimpse_sem);
1011                 lli->lli_glimpse_time = ktime_set(0, 0);
1012                 INIT_LIST_HEAD(&lli->lli_agl_list);
1013                 lli->lli_agl_index = 0;
1014                 lli->lli_async_rc = 0;
1015                 spin_lock_init(&lli->lli_heat_lock);
1016                 obd_heat_clear(lli->lli_heat_instances, OBD_HEAT_COUNT);
1017                 lli->lli_heat_flags = 0;
1018                 mutex_init(&lli->lli_pcc_lock);
1019                 lli->lli_pcc_state = PCC_STATE_FL_NONE;
1020                 lli->lli_pcc_inode = NULL;
1021                 lli->lli_pcc_dsflags = PCC_DATASET_INVALID;
1022                 lli->lli_pcc_generation = 0;
1023                 mutex_init(&lli->lli_group_mutex);
1024                 lli->lli_group_users = 0;
1025                 lli->lli_group_gid = 0;
1026         }
1027         mutex_init(&lli->lli_layout_mutex);
1028         memset(lli->lli_jobid, 0, sizeof(lli->lli_jobid));
1029 }
1030
1031 #define MAX_STRING_SIZE 128
1032
1033 #ifndef HAVE_SUPER_SETUP_BDI_NAME
1034
1035 #define LSI_BDI_INITIALIZED     0x00400000
1036
1037 #ifndef HAVE_BDI_CAP_MAP_COPY
1038 # define BDI_CAP_MAP_COPY       0
1039 #endif
1040
1041 static int super_setup_bdi_name(struct super_block *sb, char *fmt, ...)
1042 {
1043         struct  lustre_sb_info *lsi = s2lsi(sb);
1044         char buf[MAX_STRING_SIZE];
1045         va_list args;
1046         int err;
1047
1048         err = bdi_init(&lsi->lsi_bdi);
1049         if (err)
1050                 return err;
1051
1052         lsi->lsi_flags |= LSI_BDI_INITIALIZED;
1053         lsi->lsi_bdi.capabilities = BDI_CAP_MAP_COPY;
1054         lsi->lsi_bdi.name = "lustre";
1055         va_start(args, fmt);
1056         vsnprintf(buf, MAX_STRING_SIZE, fmt, args);
1057         va_end(args);
1058         err = bdi_register(&lsi->lsi_bdi, NULL, "%s", buf);
1059         va_end(args);
1060         if (!err)
1061                 sb->s_bdi = &lsi->lsi_bdi;
1062
1063         return err;
1064 }
1065 #endif /* !HAVE_SUPER_SETUP_BDI_NAME */
1066
1067 int ll_fill_super(struct super_block *sb)
1068 {
1069         struct  lustre_profile *lprof = NULL;
1070         struct  lustre_sb_info *lsi = s2lsi(sb);
1071         struct  ll_sb_info *sbi = NULL;
1072         char    *dt = NULL, *md = NULL;
1073         char    *profilenm = get_profile_name(sb);
1074         struct config_llog_instance *cfg;
1075         /* %p for void* in printf needs 16+2 characters: 0xffffffffffffffff */
1076         const int instlen = LUSTRE_MAXINSTANCE + 2;
1077         unsigned long cfg_instance = ll_get_cfg_instance(sb);
1078         char name[MAX_STRING_SIZE];
1079         int md_len = 0;
1080         int dt_len = 0;
1081         uuid_t uuid;
1082         char *ptr;
1083         int len;
1084         int err;
1085
1086         ENTRY;
1087         /* for ASLR, to map between cfg_instance and hashed ptr */
1088         CDEBUG(D_VFSTRACE, "VFS Op: cfg_instance %s-%016lx (sb %p)\n",
1089                profilenm, cfg_instance, sb);
1090
1091         OBD_ALLOC_PTR(cfg);
1092         if (cfg == NULL)
1093                 GOTO(out_free_cfg, err = -ENOMEM);
1094
1095         /* client additional sb info */
1096         lsi->lsi_llsbi = sbi = ll_init_sbi();
1097         if (IS_ERR(sbi))
1098                 GOTO(out_free_cfg, err = PTR_ERR(sbi));
1099
1100         err = ll_options(lsi->lsi_lmd->lmd_opts, sbi);
1101         if (err)
1102                 GOTO(out_free_cfg, err);
1103
1104         /* kernel >= 2.6.38 store dentry operations in sb->s_d_op. */
1105         sb->s_d_op = &ll_d_ops;
1106
1107         /* UUID handling */
1108         generate_random_uuid(uuid.b);
1109         snprintf(sbi->ll_sb_uuid.uuid, UUID_SIZE, "%pU", uuid.b);
1110
1111         CDEBUG(D_CONFIG, "llite sb uuid: %s\n", sbi->ll_sb_uuid.uuid);
1112
1113         /* Get fsname */
1114         len = strlen(profilenm);
1115         ptr = strrchr(profilenm, '-');
1116         if (ptr && (strcmp(ptr, "-client") == 0))
1117                 len -= 7;
1118
1119         if (len > LUSTRE_MAXFSNAME) {
1120                 if (unlikely(len >= MAX_STRING_SIZE))
1121                         len = MAX_STRING_SIZE - 1;
1122                 strncpy(name, profilenm, len);
1123                 name[len] = '\0';
1124                 err = -ENAMETOOLONG;
1125                 CERROR("%s: fsname longer than %u characters: rc = %d\n",
1126                        name, LUSTRE_MAXFSNAME, err);
1127                 GOTO(out_free_cfg, err);
1128         }
1129         strncpy(sbi->ll_fsname, profilenm, len);
1130         sbi->ll_fsname[len] = '\0';
1131
1132         /* Mount info */
1133         snprintf(name, sizeof(name), "%.*s-%016lx", len,
1134                  profilenm, cfg_instance);
1135
1136         err = super_setup_bdi_name(sb, "%s", name);
1137         if (err)
1138                 GOTO(out_free_cfg, err);
1139
1140         /* Call ll_debugfs_register_super() before lustre_process_log()
1141          * so that "llite.*.*" params can be processed correctly.
1142          */
1143         err = ll_debugfs_register_super(sb, name);
1144         if (err < 0) {
1145                 CERROR("%s: could not register mountpoint in llite: rc = %d\n",
1146                        sbi->ll_fsname, err);
1147                 err = 0;
1148         }
1149
1150         /* The cfg_instance is a value unique to this super, in case some
1151          * joker tries to mount the same fs at two mount points.
1152          */
1153         cfg->cfg_instance = cfg_instance;
1154         cfg->cfg_uuid = lsi->lsi_llsbi->ll_sb_uuid;
1155         cfg->cfg_callback = class_config_llog_handler;
1156         cfg->cfg_sub_clds = CONFIG_SUB_CLIENT;
1157         /* set up client obds */
1158         err = lustre_process_log(sb, profilenm, cfg);
1159         if (err < 0)
1160                 GOTO(out_debugfs, err);
1161
1162         /* Profile set with LCFG_MOUNTOPT so we can find our mdc and osc obds */
1163         lprof = class_get_profile(profilenm);
1164         if (lprof == NULL) {
1165                 LCONSOLE_ERROR_MSG(0x156, "The client profile '%s' could not be"
1166                                    " read from the MGS.  Does that filesystem "
1167                                    "exist?\n", profilenm);
1168                 GOTO(out_debugfs, err = -EINVAL);
1169         }
1170         CDEBUG(D_CONFIG, "Found profile %s: mdc=%s osc=%s\n", profilenm,
1171                lprof->lp_md, lprof->lp_dt);
1172
1173         dt_len = strlen(lprof->lp_dt) + instlen + 2;
1174         OBD_ALLOC(dt, dt_len);
1175         if (!dt)
1176                 GOTO(out_profile, err = -ENOMEM);
1177         snprintf(dt, dt_len - 1, "%s-%016lx", lprof->lp_dt, cfg_instance);
1178
1179         md_len = strlen(lprof->lp_md) + instlen + 2;
1180         OBD_ALLOC(md, md_len);
1181         if (!md)
1182                 GOTO(out_free_dt, err = -ENOMEM);
1183         snprintf(md, md_len - 1, "%s-%016lx", lprof->lp_md, cfg_instance);
1184
1185         /* connections, registrations, sb setup */
1186         err = client_common_fill_super(sb, md, dt);
1187         if (err < 0)
1188                 GOTO(out_free_md, err);
1189
1190         sbi->ll_client_common_fill_super_succeeded = 1;
1191
1192 out_free_md:
1193         if (md)
1194                 OBD_FREE(md, md_len);
1195 out_free_dt:
1196         if (dt)
1197                 OBD_FREE(dt, dt_len);
1198 out_profile:
1199         if (lprof)
1200                 class_put_profile(lprof);
1201 out_debugfs:
1202         if (err < 0)
1203                 ll_debugfs_unregister_super(sb);
1204 out_free_cfg:
1205         if (cfg)
1206                 OBD_FREE_PTR(cfg);
1207
1208         if (err)
1209                 ll_put_super(sb);
1210         else if (sbi->ll_flags & LL_SBI_VERBOSE)
1211                 LCONSOLE_WARN("Mounted %s\n", profilenm);
1212         RETURN(err);
1213 } /* ll_fill_super */
1214
1215 void ll_put_super(struct super_block *sb)
1216 {
1217         struct config_llog_instance cfg, params_cfg;
1218         struct obd_device *obd;
1219         struct lustre_sb_info *lsi = s2lsi(sb);
1220         struct ll_sb_info *sbi = ll_s2sbi(sb);
1221         char *profilenm = get_profile_name(sb);
1222         unsigned long cfg_instance = ll_get_cfg_instance(sb);
1223         long ccc_count;
1224         int next, force = 1, rc = 0;
1225         ENTRY;
1226
1227         if (IS_ERR(sbi))
1228                 GOTO(out_no_sbi, 0);
1229
1230         /* Should replace instance_id with something better for ASLR */
1231         CDEBUG(D_VFSTRACE, "VFS Op: cfg_instance %s-%016lx (sb %p)\n",
1232                profilenm, cfg_instance, sb);
1233
1234         cfg.cfg_instance = cfg_instance;
1235         lustre_end_log(sb, profilenm, &cfg);
1236
1237         params_cfg.cfg_instance = cfg_instance;
1238         lustre_end_log(sb, PARAMS_FILENAME, &params_cfg);
1239
1240         if (sbi->ll_md_exp) {
1241                 obd = class_exp2obd(sbi->ll_md_exp);
1242                 if (obd)
1243                         force = obd->obd_force;
1244         }
1245
1246         /* Wait for unstable pages to be committed to stable storage */
1247         if (force == 0) {
1248                 rc = l_wait_event_abortable(
1249                         sbi->ll_cache->ccc_unstable_waitq,
1250                         atomic_long_read(&sbi->ll_cache->ccc_unstable_nr) == 0);
1251         }
1252
1253         ccc_count = atomic_long_read(&sbi->ll_cache->ccc_unstable_nr);
1254         if (force == 0 && rc != -ERESTARTSYS)
1255                 LASSERTF(ccc_count == 0, "count: %li\n", ccc_count);
1256
1257         /* We need to set force before the lov_disconnect in
1258          * lustre_common_put_super, since l_d cleans up osc's as well.
1259          */
1260         if (force) {
1261                 next = 0;
1262                 while ((obd = class_devices_in_group(&sbi->ll_sb_uuid,
1263                                                      &next)) != NULL) {
1264                         obd->obd_force = force;
1265                 }
1266         }
1267
1268         if (sbi->ll_client_common_fill_super_succeeded) {
1269                 /* Only if client_common_fill_super succeeded */
1270                 client_common_put_super(sb);
1271         }
1272
1273         next = 0;
1274         while ((obd = class_devices_in_group(&sbi->ll_sb_uuid, &next)))
1275                 class_manual_cleanup(obd);
1276
1277         if (sbi->ll_flags & LL_SBI_VERBOSE)
1278                 LCONSOLE_WARN("Unmounted %s\n", profilenm ? profilenm : "");
1279
1280         if (profilenm)
1281                 class_del_profile(profilenm);
1282
1283 #ifndef HAVE_SUPER_SETUP_BDI_NAME
1284         if (lsi->lsi_flags & LSI_BDI_INITIALIZED) {
1285                 bdi_destroy(&lsi->lsi_bdi);
1286                 lsi->lsi_flags &= ~LSI_BDI_INITIALIZED;
1287         }
1288 #endif
1289
1290         ll_free_sbi(sb);
1291         lsi->lsi_llsbi = NULL;
1292 out_no_sbi:
1293         lustre_common_put_super(sb);
1294
1295         cl_env_cache_purge(~0);
1296
1297         module_put(THIS_MODULE);
1298
1299         EXIT;
1300 } /* client_put_super */
1301
1302 struct inode *ll_inode_from_resource_lock(struct ldlm_lock *lock)
1303 {
1304         struct inode *inode = NULL;
1305
1306         /* NOTE: we depend on atomic igrab() -bzzz */
1307         lock_res_and_lock(lock);
1308         if (lock->l_resource->lr_lvb_inode) {
1309                 struct ll_inode_info * lli;
1310                 lli = ll_i2info(lock->l_resource->lr_lvb_inode);
1311                 if (lli->lli_inode_magic == LLI_INODE_MAGIC) {
1312                         inode = igrab(lock->l_resource->lr_lvb_inode);
1313                 } else {
1314                         inode = lock->l_resource->lr_lvb_inode;
1315                         LDLM_DEBUG_LIMIT(inode->i_state & I_FREEING ?  D_INFO :
1316                                          D_WARNING, lock, "lr_lvb_inode %p is "
1317                                          "bogus: magic %08x",
1318                                          lock->l_resource->lr_lvb_inode,
1319                                          lli->lli_inode_magic);
1320                         inode = NULL;
1321                 }
1322         }
1323         unlock_res_and_lock(lock);
1324         return inode;
1325 }
1326
1327 void ll_dir_clear_lsm_md(struct inode *inode)
1328 {
1329         struct ll_inode_info *lli = ll_i2info(inode);
1330
1331         LASSERT(S_ISDIR(inode->i_mode));
1332
1333         if (lli->lli_lsm_md) {
1334                 lmv_free_memmd(lli->lli_lsm_md);
1335                 lli->lli_lsm_md = NULL;
1336         }
1337
1338         if (lli->lli_default_lsm_md) {
1339                 lmv_free_memmd(lli->lli_default_lsm_md);
1340                 lli->lli_default_lsm_md = NULL;
1341         }
1342 }
1343
1344 static struct inode *ll_iget_anon_dir(struct super_block *sb,
1345                                       const struct lu_fid *fid,
1346                                       struct lustre_md *md)
1347 {
1348         struct ll_sb_info       *sbi = ll_s2sbi(sb);
1349         struct mdt_body         *body = md->body;
1350         struct inode            *inode;
1351         ino_t                   ino;
1352         ENTRY;
1353
1354         ino = cl_fid_build_ino(fid, sbi->ll_flags & LL_SBI_32BIT_API);
1355         inode = iget_locked(sb, ino);
1356         if (inode == NULL) {
1357                 CERROR("%s: failed get simple inode "DFID": rc = -ENOENT\n",
1358                        sbi->ll_fsname, PFID(fid));
1359                 RETURN(ERR_PTR(-ENOENT));
1360         }
1361
1362         if (inode->i_state & I_NEW) {
1363                 struct ll_inode_info *lli = ll_i2info(inode);
1364                 struct lmv_stripe_md *lsm = md->lmv;
1365
1366                 inode->i_mode = (inode->i_mode & ~S_IFMT) |
1367                                 (body->mbo_mode & S_IFMT);
1368                 LASSERTF(S_ISDIR(inode->i_mode), "Not slave inode "DFID"\n",
1369                          PFID(fid));
1370
1371                 inode->i_mtime.tv_sec = 0;
1372                 inode->i_atime.tv_sec = 0;
1373                 inode->i_ctime.tv_sec = 0;
1374                 inode->i_rdev = 0;
1375
1376 #ifdef HAVE_BACKING_DEV_INFO
1377                 /* initializing backing dev info. */
1378                 inode->i_mapping->backing_dev_info =
1379                                                 &s2lsi(inode->i_sb)->lsi_bdi;
1380 #endif
1381                 inode->i_op = &ll_dir_inode_operations;
1382                 inode->i_fop = &ll_dir_operations;
1383                 lli->lli_fid = *fid;
1384                 ll_lli_init(lli);
1385
1386                 LASSERT(lsm != NULL);
1387                 /* master object FID */
1388                 lli->lli_pfid = body->mbo_fid1;
1389                 CDEBUG(D_INODE, "lli %p slave "DFID" master "DFID"\n",
1390                        lli, PFID(fid), PFID(&lli->lli_pfid));
1391                 unlock_new_inode(inode);
1392         }
1393
1394         RETURN(inode);
1395 }
1396
1397 static int ll_init_lsm_md(struct inode *inode, struct lustre_md *md)
1398 {
1399         struct lu_fid *fid;
1400         struct lmv_stripe_md *lsm = md->lmv;
1401         struct ll_inode_info *lli = ll_i2info(inode);
1402         int i;
1403
1404         LASSERT(lsm != NULL);
1405
1406         CDEBUG(D_INODE, "%s: "DFID" set dir layout:\n",
1407                ll_i2sbi(inode)->ll_fsname, PFID(&lli->lli_fid));
1408         lsm_md_dump(D_INODE, lsm);
1409
1410         if (!lmv_dir_striped(lsm))
1411                 goto out;
1412
1413         /* XXX sigh, this lsm_root initialization should be in
1414          * LMV layer, but it needs ll_iget right now, so we
1415          * put this here right now. */
1416         for (i = 0; i < lsm->lsm_md_stripe_count; i++) {
1417                 fid = &lsm->lsm_md_oinfo[i].lmo_fid;
1418                 LASSERT(lsm->lsm_md_oinfo[i].lmo_root == NULL);
1419
1420                 if (!fid_is_sane(fid))
1421                         continue;
1422
1423                 /* Unfortunately ll_iget will call ll_update_inode,
1424                  * where the initialization of slave inode is slightly
1425                  * different, so it reset lsm_md to NULL to avoid
1426                  * initializing lsm for slave inode. */
1427                 lsm->lsm_md_oinfo[i].lmo_root =
1428                                 ll_iget_anon_dir(inode->i_sb, fid, md);
1429                 if (IS_ERR(lsm->lsm_md_oinfo[i].lmo_root)) {
1430                         int rc = PTR_ERR(lsm->lsm_md_oinfo[i].lmo_root);
1431
1432                         lsm->lsm_md_oinfo[i].lmo_root = NULL;
1433                         while (i-- > 0) {
1434                                 iput(lsm->lsm_md_oinfo[i].lmo_root);
1435                                 lsm->lsm_md_oinfo[i].lmo_root = NULL;
1436                         }
1437                         return rc;
1438                 }
1439         }
1440 out:
1441         lli->lli_lsm_md = lsm;
1442
1443         return 0;
1444 }
1445
1446 static void ll_update_default_lsm_md(struct inode *inode, struct lustre_md *md)
1447 {
1448         struct ll_inode_info *lli = ll_i2info(inode);
1449
1450         if (!md->default_lmv) {
1451                 /* clear default lsm */
1452                 if (lli->lli_default_lsm_md) {
1453                         down_write(&lli->lli_lsm_sem);
1454                         if (lli->lli_default_lsm_md) {
1455                                 lmv_free_memmd(lli->lli_default_lsm_md);
1456                                 lli->lli_default_lsm_md = NULL;
1457                         }
1458                         up_write(&lli->lli_lsm_sem);
1459                 }
1460         } else if (lli->lli_default_lsm_md) {
1461                 /* update default lsm if it changes */
1462                 down_read(&lli->lli_lsm_sem);
1463                 if (lli->lli_default_lsm_md &&
1464                     !lsm_md_eq(lli->lli_default_lsm_md, md->default_lmv)) {
1465                         up_read(&lli->lli_lsm_sem);
1466                         down_write(&lli->lli_lsm_sem);
1467                         if (lli->lli_default_lsm_md)
1468                                 lmv_free_memmd(lli->lli_default_lsm_md);
1469                         lli->lli_default_lsm_md = md->default_lmv;
1470                         lsm_md_dump(D_INODE, md->default_lmv);
1471                         md->default_lmv = NULL;
1472                         up_write(&lli->lli_lsm_sem);
1473                 } else {
1474                         up_read(&lli->lli_lsm_sem);
1475                 }
1476         } else {
1477                 /* init default lsm */
1478                 down_write(&lli->lli_lsm_sem);
1479                 lli->lli_default_lsm_md = md->default_lmv;
1480                 lsm_md_dump(D_INODE, md->default_lmv);
1481                 md->default_lmv = NULL;
1482                 up_write(&lli->lli_lsm_sem);
1483         }
1484 }
1485
1486 static int ll_update_lsm_md(struct inode *inode, struct lustre_md *md)
1487 {
1488         struct ll_inode_info *lli = ll_i2info(inode);
1489         struct lmv_stripe_md *lsm = md->lmv;
1490         struct cl_attr  *attr;
1491         int rc = 0;
1492
1493         ENTRY;
1494
1495         LASSERT(S_ISDIR(inode->i_mode));
1496         CDEBUG(D_INODE, "update lsm %p of "DFID"\n", lli->lli_lsm_md,
1497                PFID(ll_inode2fid(inode)));
1498
1499         /* update default LMV */
1500         if (md->default_lmv)
1501                 ll_update_default_lsm_md(inode, md);
1502
1503         /*
1504          * no striped information from request, lustre_md from req does not
1505          * include stripeEA, see ll_md_setattr()
1506          */
1507         if (!lsm)
1508                 RETURN(0);
1509
1510         /*
1511          * normally dir layout doesn't change, only take read lock to check
1512          * that to avoid blocking other MD operations.
1513          */
1514         down_read(&lli->lli_lsm_sem);
1515
1516         /* some current lookup initialized lsm, and unchanged */
1517         if (lli->lli_lsm_md && lsm_md_eq(lli->lli_lsm_md, lsm))
1518                 GOTO(unlock, rc = 0);
1519
1520         /* if dir layout doesn't match, check whether version is increased,
1521          * which means layout is changed, this happens in dir split/merge and
1522          * lfsck.
1523          *
1524          * foreign LMV should not change.
1525          */
1526         if (lli->lli_lsm_md && lmv_dir_striped(lli->lli_lsm_md) &&
1527             lsm->lsm_md_layout_version <=
1528             lli->lli_lsm_md->lsm_md_layout_version) {
1529                 CERROR("%s: "DFID" dir layout mismatch:\n",
1530                        ll_i2sbi(inode)->ll_fsname, PFID(&lli->lli_fid));
1531                 lsm_md_dump(D_ERROR, lli->lli_lsm_md);
1532                 lsm_md_dump(D_ERROR, lsm);
1533                 GOTO(unlock, rc = -EINVAL);
1534         }
1535
1536         up_read(&lli->lli_lsm_sem);
1537         down_write(&lli->lli_lsm_sem);
1538         /* clear existing lsm */
1539         if (lli->lli_lsm_md) {
1540                 lmv_free_memmd(lli->lli_lsm_md);
1541                 lli->lli_lsm_md = NULL;
1542         }
1543
1544         rc = ll_init_lsm_md(inode, md);
1545         up_write(&lli->lli_lsm_sem);
1546
1547         if (rc)
1548                 RETURN(rc);
1549
1550         /* set md->lmv to NULL, so the following free lustre_md will not free
1551          * this lsm.
1552          */
1553         md->lmv = NULL;
1554
1555         /* md_merge_attr() may take long, since lsm is already set, switch to
1556          * read lock.
1557          */
1558         down_read(&lli->lli_lsm_sem);
1559
1560         if (!lmv_dir_striped(lli->lli_lsm_md))
1561                 GOTO(unlock, rc = 0);
1562
1563         OBD_ALLOC_PTR(attr);
1564         if (!attr)
1565                 GOTO(unlock, rc = -ENOMEM);
1566
1567         /* validate the lsm */
1568         rc = md_merge_attr(ll_i2mdexp(inode), lli->lli_lsm_md, attr,
1569                            ll_md_blocking_ast);
1570         if (!rc) {
1571                 if (md->body->mbo_valid & OBD_MD_FLNLINK)
1572                         md->body->mbo_nlink = attr->cat_nlink;
1573                 if (md->body->mbo_valid & OBD_MD_FLSIZE)
1574                         md->body->mbo_size = attr->cat_size;
1575                 if (md->body->mbo_valid & OBD_MD_FLATIME)
1576                         md->body->mbo_atime = attr->cat_atime;
1577                 if (md->body->mbo_valid & OBD_MD_FLCTIME)
1578                         md->body->mbo_ctime = attr->cat_ctime;
1579                 if (md->body->mbo_valid & OBD_MD_FLMTIME)
1580                         md->body->mbo_mtime = attr->cat_mtime;
1581         }
1582
1583         OBD_FREE_PTR(attr);
1584         GOTO(unlock, rc);
1585 unlock:
1586         up_read(&lli->lli_lsm_sem);
1587
1588         return rc;
1589 }
1590
1591 void ll_clear_inode(struct inode *inode)
1592 {
1593         struct ll_inode_info *lli = ll_i2info(inode);
1594         struct ll_sb_info *sbi = ll_i2sbi(inode);
1595
1596         ENTRY;
1597
1598         CDEBUG(D_VFSTRACE, "VFS Op:inode="DFID"(%p)\n",
1599                PFID(ll_inode2fid(inode)), inode);
1600
1601         if (S_ISDIR(inode->i_mode)) {
1602                 /* these should have been cleared in ll_file_release */
1603                 LASSERT(lli->lli_opendir_key == NULL);
1604                 LASSERT(lli->lli_sai == NULL);
1605                 LASSERT(lli->lli_opendir_pid == 0);
1606         } else {
1607                 pcc_inode_free(inode);
1608         }
1609
1610         md_null_inode(sbi->ll_md_exp, ll_inode2fid(inode));
1611
1612         LASSERT(!lli->lli_open_fd_write_count);
1613         LASSERT(!lli->lli_open_fd_read_count);
1614         LASSERT(!lli->lli_open_fd_exec_count);
1615
1616         if (lli->lli_mds_write_och)
1617                 ll_md_real_close(inode, FMODE_WRITE);
1618         if (lli->lli_mds_exec_och)
1619                 ll_md_real_close(inode, FMODE_EXEC);
1620         if (lli->lli_mds_read_och)
1621                 ll_md_real_close(inode, FMODE_READ);
1622
1623         if (S_ISLNK(inode->i_mode) && lli->lli_symlink_name) {
1624                 OBD_FREE(lli->lli_symlink_name,
1625                          strlen(lli->lli_symlink_name) + 1);
1626                 lli->lli_symlink_name = NULL;
1627         }
1628
1629         ll_xattr_cache_destroy(inode);
1630
1631 #ifdef CONFIG_LUSTRE_FS_POSIX_ACL
1632         forget_all_cached_acls(inode);
1633         if (lli->lli_posix_acl) {
1634                 posix_acl_release(lli->lli_posix_acl);
1635                 lli->lli_posix_acl = NULL;
1636         }
1637 #endif
1638         lli->lli_inode_magic = LLI_INODE_DEAD;
1639
1640         if (S_ISDIR(inode->i_mode))
1641                 ll_dir_clear_lsm_md(inode);
1642         else if (S_ISREG(inode->i_mode) && !is_bad_inode(inode))
1643                 LASSERT(list_empty(&lli->lli_agl_list));
1644
1645         /*
1646          * XXX This has to be done before lsm is freed below, because
1647          * cl_object still uses inode lsm.
1648          */
1649         cl_inode_fini(inode);
1650
1651         EXIT;
1652 }
1653
1654 static int ll_md_setattr(struct dentry *dentry, struct md_op_data *op_data)
1655 {
1656         struct lustre_md md;
1657         struct inode *inode = dentry->d_inode;
1658         struct ll_sb_info *sbi = ll_i2sbi(inode);
1659         struct ptlrpc_request *request = NULL;
1660         int rc, ia_valid;
1661         ENTRY;
1662
1663         op_data = ll_prep_md_op_data(op_data, inode, NULL, NULL, 0, 0,
1664                                      LUSTRE_OPC_ANY, NULL);
1665         if (IS_ERR(op_data))
1666                 RETURN(PTR_ERR(op_data));
1667
1668         rc = md_setattr(sbi->ll_md_exp, op_data, NULL, 0, &request);
1669         if (rc) {
1670                 ptlrpc_req_finished(request);
1671                 if (rc == -ENOENT) {
1672                         clear_nlink(inode);
1673                         /* Unlinked special device node? Or just a race?
1674                          * Pretend we done everything. */
1675                         if (!S_ISREG(inode->i_mode) &&
1676                             !S_ISDIR(inode->i_mode)) {
1677                                 ia_valid = op_data->op_attr.ia_valid;
1678                                 op_data->op_attr.ia_valid &= ~TIMES_SET_FLAGS;
1679                                 rc = simple_setattr(dentry, &op_data->op_attr);
1680                                 op_data->op_attr.ia_valid = ia_valid;
1681                         }
1682                 } else if (rc != -EPERM && rc != -EACCES && rc != -ETXTBSY) {
1683                         CERROR("md_setattr fails: rc = %d\n", rc);
1684                 }
1685                 RETURN(rc);
1686         }
1687
1688         rc = md_get_lustre_md(sbi->ll_md_exp, request, sbi->ll_dt_exp,
1689                               sbi->ll_md_exp, &md);
1690         if (rc) {
1691                 ptlrpc_req_finished(request);
1692                 RETURN(rc);
1693         }
1694
1695         ia_valid = op_data->op_attr.ia_valid;
1696         /* inode size will be in ll_setattr_ost, can't do it now since dirty
1697          * cache is not cleared yet. */
1698         op_data->op_attr.ia_valid &= ~(TIMES_SET_FLAGS | ATTR_SIZE);
1699         if (S_ISREG(inode->i_mode))
1700                 inode_lock(inode);
1701         rc = simple_setattr(dentry, &op_data->op_attr);
1702         if (S_ISREG(inode->i_mode))
1703                 inode_unlock(inode);
1704         op_data->op_attr.ia_valid = ia_valid;
1705
1706         rc = ll_update_inode(inode, &md);
1707         ptlrpc_req_finished(request);
1708
1709         RETURN(rc);
1710 }
1711
1712 /* If this inode has objects allocated to it (lsm != NULL), then the OST
1713  * object(s) determine the file size and mtime.  Otherwise, the MDS will
1714  * keep these values until such a time that objects are allocated for it.
1715  * We do the MDS operations first, as it is checking permissions for us.
1716  * We don't to the MDS RPC if there is nothing that we want to store there,
1717  * otherwise there is no harm in updating mtime/atime on the MDS if we are
1718  * going to do an RPC anyways.
1719  *
1720  * If we are doing a truncate, we will send the mtime and ctime updates
1721  * to the OST with the punch RPC, otherwise we do an explicit setattr RPC.
1722  * I don't believe it is possible to get e.g. ATTR_MTIME_SET and ATTR_SIZE
1723  * at the same time.
1724  *
1725  * In case of HSMimport, we only set attr on MDS.
1726  */
1727 int ll_setattr_raw(struct dentry *dentry, struct iattr *attr,
1728                    enum op_xvalid xvalid, bool hsm_import)
1729 {
1730         struct inode *inode = dentry->d_inode;
1731         struct ll_inode_info *lli = ll_i2info(inode);
1732         struct md_op_data *op_data = NULL;
1733         ktime_t kstart = ktime_get();
1734         int rc = 0;
1735
1736         ENTRY;
1737
1738         CDEBUG(D_VFSTRACE, "%s: setattr inode "DFID"(%p) from %llu to %llu, "
1739                "valid %x, hsm_import %d\n",
1740                ll_i2sbi(inode)->ll_fsname, PFID(&lli->lli_fid),
1741                inode, i_size_read(inode), attr->ia_size, attr->ia_valid,
1742                hsm_import);
1743
1744         if (attr->ia_valid & ATTR_SIZE) {
1745                 /* Check new size against VFS/VM file size limit and rlimit */
1746                 rc = inode_newsize_ok(inode, attr->ia_size);
1747                 if (rc)
1748                         RETURN(rc);
1749
1750                 /* The maximum Lustre file size is variable, based on the
1751                  * OST maximum object size and number of stripes.  This
1752                  * needs another check in addition to the VFS check above. */
1753                 if (attr->ia_size > ll_file_maxbytes(inode)) {
1754                         CDEBUG(D_INODE,"file "DFID" too large %llu > %llu\n",
1755                                PFID(&lli->lli_fid), attr->ia_size,
1756                                ll_file_maxbytes(inode));
1757                         RETURN(-EFBIG);
1758                 }
1759
1760                 attr->ia_valid |= ATTR_MTIME | ATTR_CTIME;
1761         }
1762
1763         /* POSIX: check before ATTR_*TIME_SET set (from inode_change_ok) */
1764         if (attr->ia_valid & TIMES_SET_FLAGS) {
1765                 if ((!uid_eq(current_fsuid(), inode->i_uid)) &&
1766                     !cfs_capable(CFS_CAP_FOWNER))
1767                         RETURN(-EPERM);
1768         }
1769
1770         /* We mark all of the fields "set" so MDS/OST does not re-set them */
1771         if (!(xvalid & OP_XVALID_CTIME_SET) &&
1772              (attr->ia_valid & ATTR_CTIME)) {
1773                 attr->ia_ctime = current_time(inode);
1774                 xvalid |= OP_XVALID_CTIME_SET;
1775         }
1776         if (!(attr->ia_valid & ATTR_ATIME_SET) &&
1777             (attr->ia_valid & ATTR_ATIME)) {
1778                 attr->ia_atime = current_time(inode);
1779                 attr->ia_valid |= ATTR_ATIME_SET;
1780         }
1781         if (!(attr->ia_valid & ATTR_MTIME_SET) &&
1782             (attr->ia_valid & ATTR_MTIME)) {
1783                 attr->ia_mtime = current_time(inode);
1784                 attr->ia_valid |= ATTR_MTIME_SET;
1785         }
1786
1787         if (attr->ia_valid & (ATTR_MTIME | ATTR_CTIME))
1788                 CDEBUG(D_INODE, "setting mtime %lld, ctime %lld, now = %lld\n",
1789                        (s64)attr->ia_mtime.tv_sec, (s64)attr->ia_ctime.tv_sec,
1790                        ktime_get_real_seconds());
1791
1792         if (S_ISREG(inode->i_mode))
1793                 inode_unlock(inode);
1794
1795         /* We always do an MDS RPC, even if we're only changing the size;
1796          * only the MDS knows whether truncate() should fail with -ETXTBUSY */
1797
1798         OBD_ALLOC_PTR(op_data);
1799         if (op_data == NULL)
1800                 GOTO(out, rc = -ENOMEM);
1801
1802         if (!hsm_import && attr->ia_valid & ATTR_SIZE) {
1803                 /* If we are changing file size, file content is
1804                  * modified, flag it.
1805                  */
1806                 xvalid |= OP_XVALID_OWNEROVERRIDE;
1807                 op_data->op_bias |= MDS_DATA_MODIFIED;
1808                 ll_file_clear_flag(lli, LLIF_DATA_MODIFIED);
1809         }
1810
1811         if (attr->ia_valid & ATTR_FILE) {
1812                 struct ll_file_data *fd = attr->ia_file->private_data;
1813
1814                 if (fd->fd_lease_och)
1815                         op_data->op_bias |= MDS_TRUNC_KEEP_LEASE;
1816         }
1817
1818         op_data->op_attr = *attr;
1819         op_data->op_xvalid = xvalid;
1820
1821         rc = ll_md_setattr(dentry, op_data);
1822         if (rc)
1823                 GOTO(out, rc);
1824
1825         if (!S_ISREG(inode->i_mode) || hsm_import)
1826                 GOTO(out, rc = 0);
1827
1828         if (attr->ia_valid & (ATTR_SIZE | ATTR_ATIME | ATTR_ATIME_SET |
1829                               ATTR_MTIME | ATTR_MTIME_SET | ATTR_CTIME) ||
1830             xvalid & OP_XVALID_CTIME_SET) {
1831                 bool cached = false;
1832
1833                 rc = pcc_inode_setattr(inode, attr, &cached);
1834                 if (cached) {
1835                         if (rc) {
1836                                 CERROR("%s: PCC inode "DFID" setattr failed: "
1837                                        "rc = %d\n",
1838                                        ll_i2sbi(inode)->ll_fsname,
1839                                        PFID(&lli->lli_fid), rc);
1840                                 GOTO(out, rc);
1841                         }
1842                 } else {
1843                         /* For truncate and utimes sending attributes to OSTs,
1844                          * setting mtime/atime to the past will be performed
1845                          * under PW [0:EOF] extent lock (new_size:EOF for
1846                          * truncate). It may seem excessive to send mtime/atime
1847                          * updates to OSTs when not setting times to past, but
1848                          * it is necessary due to possible time
1849                          * de-synchronization between MDT inode and OST objects
1850                          */
1851                         rc = cl_setattr_ost(lli->lli_clob, attr, xvalid, 0);
1852                 }
1853         }
1854
1855         /* If the file was restored, it needs to set dirty flag.
1856          *
1857          * We've already sent MDS_DATA_MODIFIED flag in
1858          * ll_md_setattr() for truncate. However, the MDT refuses to
1859          * set the HS_DIRTY flag on released files, so we have to set
1860          * it again if the file has been restored. Please check how
1861          * LLIF_DATA_MODIFIED is set in vvp_io_setattr_fini().
1862          *
1863          * Please notice that if the file is not released, the previous
1864          * MDS_DATA_MODIFIED has taken effect and usually
1865          * LLIF_DATA_MODIFIED is not set(see vvp_io_setattr_fini()).
1866          * This way we can save an RPC for common open + trunc
1867          * operation. */
1868         if (ll_file_test_and_clear_flag(lli, LLIF_DATA_MODIFIED)) {
1869                 struct hsm_state_set hss = {
1870                         .hss_valid = HSS_SETMASK,
1871                         .hss_setmask = HS_DIRTY,
1872                 };
1873                 int rc2;
1874
1875                 rc2 = ll_hsm_state_set(inode, &hss);
1876                 /* truncate and write can happen at the same time, so that
1877                  * the file can be set modified even though the file is not
1878                  * restored from released state, and ll_hsm_state_set() is
1879                  * not applicable for the file, and rc2 < 0 is normal in this
1880                  * case. */
1881                 if (rc2 < 0)
1882                         CDEBUG(D_INFO, DFID "HSM set dirty failed: rc2 = %d\n",
1883                                PFID(ll_inode2fid(inode)), rc2);
1884         }
1885
1886         EXIT;
1887 out:
1888         if (op_data != NULL)
1889                 ll_finish_md_op_data(op_data);
1890
1891         if (S_ISREG(inode->i_mode)) {
1892                 inode_lock(inode);
1893                 if ((attr->ia_valid & ATTR_SIZE) && !hsm_import)
1894                         inode_dio_wait(inode);
1895                 /* Once we've got the i_mutex, it's safe to set the S_NOSEC
1896                  * flag.  ll_update_inode (called from ll_md_setattr), clears
1897                  * inode flags, so there is a gap where S_NOSEC is not set.
1898                  * This can cause a writer to take the i_mutex unnecessarily,
1899                  * but this is safe to do and should be rare. */
1900                 inode_has_no_xattr(inode);
1901         }
1902
1903         if (!rc)
1904                 ll_stats_ops_tally(ll_i2sbi(inode), attr->ia_valid & ATTR_SIZE ?
1905                                         LPROC_LL_TRUNC : LPROC_LL_SETATTR,
1906                                    ktime_us_delta(ktime_get(), kstart));
1907
1908         return rc;
1909 }
1910
1911 int ll_setattr(struct dentry *de, struct iattr *attr)
1912 {
1913         int mode = de->d_inode->i_mode;
1914         enum op_xvalid xvalid = 0;
1915
1916         if ((attr->ia_valid & (ATTR_CTIME|ATTR_SIZE|ATTR_MODE)) ==
1917                               (ATTR_CTIME|ATTR_SIZE|ATTR_MODE))
1918                 xvalid |= OP_XVALID_OWNEROVERRIDE;
1919
1920         if (((attr->ia_valid & (ATTR_MODE|ATTR_FORCE|ATTR_SIZE)) ==
1921                                (ATTR_SIZE|ATTR_MODE)) &&
1922             (((mode & S_ISUID) && !(attr->ia_mode & S_ISUID)) ||
1923              (((mode & (S_ISGID|S_IXGRP)) == (S_ISGID|S_IXGRP)) &&
1924               !(attr->ia_mode & S_ISGID))))
1925                 attr->ia_valid |= ATTR_FORCE;
1926
1927         if ((attr->ia_valid & ATTR_MODE) &&
1928             (mode & S_ISUID) &&
1929             !(attr->ia_mode & S_ISUID) &&
1930             !(attr->ia_valid & ATTR_KILL_SUID))
1931                 attr->ia_valid |= ATTR_KILL_SUID;
1932
1933         if ((attr->ia_valid & ATTR_MODE) &&
1934             ((mode & (S_ISGID|S_IXGRP)) == (S_ISGID|S_IXGRP)) &&
1935             !(attr->ia_mode & S_ISGID) &&
1936             !(attr->ia_valid & ATTR_KILL_SGID))
1937                 attr->ia_valid |= ATTR_KILL_SGID;
1938
1939         return ll_setattr_raw(de, attr, xvalid, false);
1940 }
1941
1942 int ll_statfs_internal(struct ll_sb_info *sbi, struct obd_statfs *osfs,
1943                        u32 flags)
1944 {
1945         struct obd_statfs obd_osfs = { 0 };
1946         time64_t max_age;
1947         int rc;
1948
1949         ENTRY;
1950         max_age = ktime_get_seconds() - sbi->ll_statfs_max_age;
1951
1952         if (sbi->ll_flags & LL_SBI_LAZYSTATFS)
1953                 flags |= OBD_STATFS_NODELAY;
1954
1955         rc = obd_statfs(NULL, sbi->ll_md_exp, osfs, max_age, flags);
1956         if (rc)
1957                 RETURN(rc);
1958
1959         osfs->os_type = LL_SUPER_MAGIC;
1960
1961         CDEBUG(D_SUPER, "MDC blocks %llu/%llu objects %llu/%llu\n",
1962               osfs->os_bavail, osfs->os_blocks, osfs->os_ffree, osfs->os_files);
1963
1964         if (osfs->os_state & OS_STATE_SUM)
1965                 GOTO(out, rc);
1966
1967         rc = obd_statfs(NULL, sbi->ll_dt_exp, &obd_osfs, max_age, flags);
1968         if (rc) /* Possibly a filesystem with no OSTs.  Report MDT totals. */
1969                 GOTO(out, rc = 0);
1970
1971         CDEBUG(D_SUPER, "OSC blocks %llu/%llu objects %llu/%llu\n",
1972                obd_osfs.os_bavail, obd_osfs.os_blocks, obd_osfs.os_ffree,
1973                obd_osfs.os_files);
1974
1975         osfs->os_bsize = obd_osfs.os_bsize;
1976         osfs->os_blocks = obd_osfs.os_blocks;
1977         osfs->os_bfree = obd_osfs.os_bfree;
1978         osfs->os_bavail = obd_osfs.os_bavail;
1979
1980         /* If we have _some_ OSTs, but don't have as many free objects on the
1981          * OSTs as inodes on the MDTs, reduce the reported number of inodes
1982          * to compensate, so that the "inodes in use" number is correct.
1983          * This should be kept in sync with lod_statfs() behaviour.
1984          */
1985         if (obd_osfs.os_files && obd_osfs.os_ffree < osfs->os_ffree) {
1986                 osfs->os_files = (osfs->os_files - osfs->os_ffree) +
1987                                  obd_osfs.os_ffree;
1988                 osfs->os_ffree = obd_osfs.os_ffree;
1989         }
1990
1991 out:
1992         RETURN(rc);
1993 }
1994
1995 int ll_statfs(struct dentry *de, struct kstatfs *sfs)
1996 {
1997         struct super_block *sb = de->d_sb;
1998         struct obd_statfs osfs;
1999         __u64 fsid = huge_encode_dev(sb->s_dev);
2000         ktime_t kstart = ktime_get();
2001         int rc;
2002
2003         CDEBUG(D_VFSTRACE, "VFS Op:sb=%s (%p)\n", sb->s_id, sb);
2004
2005         /* Some amount of caching on the client is allowed */
2006         rc = ll_statfs_internal(ll_s2sbi(sb), &osfs, OBD_STATFS_SUM);
2007         if (rc)
2008                 return rc;
2009
2010         statfs_unpack(sfs, &osfs);
2011
2012         /* We need to downshift for all 32-bit kernels, because we can't
2013          * tell if the kernel is being called via sys_statfs64() or not.
2014          * Stop before overflowing f_bsize - in which case it is better
2015          * to just risk EOVERFLOW if caller is using old sys_statfs(). */
2016         if (sizeof(long) < 8) {
2017                 while (osfs.os_blocks > ~0UL && sfs->f_bsize < 0x40000000) {
2018                         sfs->f_bsize <<= 1;
2019
2020                         osfs.os_blocks >>= 1;
2021                         osfs.os_bfree >>= 1;
2022                         osfs.os_bavail >>= 1;
2023                 }
2024         }
2025
2026         sfs->f_blocks = osfs.os_blocks;
2027         sfs->f_bfree = osfs.os_bfree;
2028         sfs->f_bavail = osfs.os_bavail;
2029         sfs->f_fsid.val[0] = (__u32)fsid;
2030         sfs->f_fsid.val[1] = (__u32)(fsid >> 32);
2031
2032         ll_stats_ops_tally(ll_s2sbi(sb), LPROC_LL_STATFS,
2033                            ktime_us_delta(ktime_get(), kstart));
2034
2035         return 0;
2036 }
2037
2038 void ll_inode_size_lock(struct inode *inode)
2039 {
2040         struct ll_inode_info *lli;
2041
2042         LASSERT(!S_ISDIR(inode->i_mode));
2043
2044         lli = ll_i2info(inode);
2045         mutex_lock(&lli->lli_size_mutex);
2046 }
2047
2048 void ll_inode_size_unlock(struct inode *inode)
2049 {
2050         struct ll_inode_info *lli;
2051
2052         lli = ll_i2info(inode);
2053         mutex_unlock(&lli->lli_size_mutex);
2054 }
2055
2056 void ll_update_inode_flags(struct inode *inode, int ext_flags)
2057 {
2058         inode->i_flags = ll_ext_to_inode_flags(ext_flags);
2059         if (ext_flags & LUSTRE_PROJINHERIT_FL)
2060                 ll_file_set_flag(ll_i2info(inode), LLIF_PROJECT_INHERIT);
2061         else
2062                 ll_file_clear_flag(ll_i2info(inode), LLIF_PROJECT_INHERIT);
2063 }
2064
2065 int ll_update_inode(struct inode *inode, struct lustre_md *md)
2066 {
2067         struct ll_inode_info *lli = ll_i2info(inode);
2068         struct mdt_body *body = md->body;
2069         struct ll_sb_info *sbi = ll_i2sbi(inode);
2070         int rc = 0;
2071
2072         if (body->mbo_valid & OBD_MD_FLEASIZE) {
2073                 rc = cl_file_inode_init(inode, md);
2074                 if (rc)
2075                         return rc;
2076         }
2077
2078         if (S_ISDIR(inode->i_mode)) {
2079                 rc = ll_update_lsm_md(inode, md);
2080                 if (rc != 0)
2081                         return rc;
2082         }
2083
2084 #ifdef CONFIG_LUSTRE_FS_POSIX_ACL
2085         if (body->mbo_valid & OBD_MD_FLACL) {
2086                 spin_lock(&lli->lli_lock);
2087                 if (lli->lli_posix_acl)
2088                         posix_acl_release(lli->lli_posix_acl);
2089                 lli->lli_posix_acl = md->posix_acl;
2090                 spin_unlock(&lli->lli_lock);
2091         }
2092 #endif
2093         inode->i_ino = cl_fid_build_ino(&body->mbo_fid1,
2094                                         sbi->ll_flags & LL_SBI_32BIT_API);
2095         inode->i_generation = cl_fid_build_gen(&body->mbo_fid1);
2096
2097         if (body->mbo_valid & OBD_MD_FLATIME) {
2098                 if (body->mbo_atime > inode->i_atime.tv_sec)
2099                         inode->i_atime.tv_sec = body->mbo_atime;
2100                 lli->lli_atime = body->mbo_atime;
2101         }
2102
2103         if (body->mbo_valid & OBD_MD_FLMTIME) {
2104                 if (body->mbo_mtime > inode->i_mtime.tv_sec) {
2105                         CDEBUG(D_INODE,
2106                                "setting ino %lu mtime from %lld to %llu\n",
2107                                inode->i_ino, (s64)inode->i_mtime.tv_sec,
2108                                body->mbo_mtime);
2109                         inode->i_mtime.tv_sec = body->mbo_mtime;
2110                 }
2111                 lli->lli_mtime = body->mbo_mtime;
2112         }
2113
2114         if (body->mbo_valid & OBD_MD_FLCTIME) {
2115                 if (body->mbo_ctime > inode->i_ctime.tv_sec)
2116                         inode->i_ctime.tv_sec = body->mbo_ctime;
2117                 lli->lli_ctime = body->mbo_ctime;
2118         }
2119
2120         /* Clear i_flags to remove S_NOSEC before permissions are updated */
2121         if (body->mbo_valid & OBD_MD_FLFLAGS)
2122                 ll_update_inode_flags(inode, body->mbo_flags);
2123         if (body->mbo_valid & OBD_MD_FLMODE)
2124                 inode->i_mode = (inode->i_mode & S_IFMT) |
2125                                 (body->mbo_mode & ~S_IFMT);
2126
2127         if (body->mbo_valid & OBD_MD_FLTYPE)
2128                 inode->i_mode = (inode->i_mode & ~S_IFMT) |
2129                                 (body->mbo_mode & S_IFMT);
2130
2131         LASSERT(inode->i_mode != 0);
2132         if (S_ISREG(inode->i_mode))
2133                 inode->i_blkbits = min(PTLRPC_MAX_BRW_BITS + 1,
2134                                        LL_MAX_BLKSIZE_BITS);
2135         else
2136                 inode->i_blkbits = inode->i_sb->s_blocksize_bits;
2137
2138         if (body->mbo_valid & OBD_MD_FLUID)
2139                 inode->i_uid = make_kuid(&init_user_ns, body->mbo_uid);
2140         if (body->mbo_valid & OBD_MD_FLGID)
2141                 inode->i_gid = make_kgid(&init_user_ns, body->mbo_gid);
2142         if (body->mbo_valid & OBD_MD_FLPROJID)
2143                 lli->lli_projid = body->mbo_projid;
2144         if (body->mbo_valid & OBD_MD_FLNLINK)
2145                 set_nlink(inode, body->mbo_nlink);
2146         if (body->mbo_valid & OBD_MD_FLRDEV)
2147                 inode->i_rdev = old_decode_dev(body->mbo_rdev);
2148
2149         if (body->mbo_valid & OBD_MD_FLID) {
2150                 /* FID shouldn't be changed! */
2151                 if (fid_is_sane(&lli->lli_fid)) {
2152                         LASSERTF(lu_fid_eq(&lli->lli_fid, &body->mbo_fid1),
2153                                  "Trying to change FID "DFID
2154                                  " to the "DFID", inode "DFID"(%p)\n",
2155                                  PFID(&lli->lli_fid), PFID(&body->mbo_fid1),
2156                                  PFID(ll_inode2fid(inode)), inode);
2157                 } else {
2158                         lli->lli_fid = body->mbo_fid1;
2159                 }
2160         }
2161
2162         LASSERT(fid_seq(&lli->lli_fid) != 0);
2163
2164         if (body->mbo_valid & OBD_MD_FLSIZE) {
2165                 i_size_write(inode, body->mbo_size);
2166
2167                 CDEBUG(D_VFSTRACE, "inode="DFID", updating i_size %llu\n",
2168                        PFID(ll_inode2fid(inode)),
2169                        (unsigned long long)body->mbo_size);
2170
2171                 if (body->mbo_valid & OBD_MD_FLBLOCKS)
2172                         inode->i_blocks = body->mbo_blocks;
2173         }
2174
2175         if (body->mbo_valid & OBD_MD_TSTATE) {
2176                 /* Set LLIF_FILE_RESTORING if restore ongoing and
2177                  * clear it when done to ensure to start again
2178                  * glimpsing updated attrs
2179                  */
2180                 if (body->mbo_t_state & MS_RESTORE)
2181                         ll_file_set_flag(lli, LLIF_FILE_RESTORING);
2182                 else
2183                         ll_file_clear_flag(lli, LLIF_FILE_RESTORING);
2184         }
2185
2186         return 0;
2187 }
2188
2189 int ll_read_inode2(struct inode *inode, void *opaque)
2190 {
2191         struct lustre_md *md = opaque;
2192         struct ll_inode_info *lli = ll_i2info(inode);
2193         int     rc;
2194         ENTRY;
2195
2196         CDEBUG(D_VFSTRACE, "VFS Op:inode="DFID"(%p)\n",
2197                PFID(&lli->lli_fid), inode);
2198
2199         /* Core attributes from the MDS first.  This is a new inode, and
2200          * the VFS doesn't zero times in the core inode so we have to do
2201          * it ourselves.  They will be overwritten by either MDS or OST
2202          * attributes - we just need to make sure they aren't newer.
2203          */
2204         inode->i_mtime.tv_sec = 0;
2205         inode->i_atime.tv_sec = 0;
2206         inode->i_ctime.tv_sec = 0;
2207         inode->i_rdev = 0;
2208         rc = ll_update_inode(inode, md);
2209         if (rc != 0)
2210                 RETURN(rc);
2211
2212         /* OIDEBUG(inode); */
2213
2214 #ifdef HAVE_BACKING_DEV_INFO
2215         /* initializing backing dev info. */
2216         inode->i_mapping->backing_dev_info = &s2lsi(inode->i_sb)->lsi_bdi;
2217 #endif
2218         if (S_ISREG(inode->i_mode)) {
2219                 struct ll_sb_info *sbi = ll_i2sbi(inode);
2220                 inode->i_op = &ll_file_inode_operations;
2221                 inode->i_fop = sbi->ll_fop;
2222                 inode->i_mapping->a_ops = (struct address_space_operations *)&ll_aops;
2223                 EXIT;
2224         } else if (S_ISDIR(inode->i_mode)) {
2225                 inode->i_op = &ll_dir_inode_operations;
2226                 inode->i_fop = &ll_dir_operations;
2227                 EXIT;
2228         } else if (S_ISLNK(inode->i_mode)) {
2229                 inode->i_op = &ll_fast_symlink_inode_operations;
2230                 EXIT;
2231         } else {
2232                 inode->i_op = &ll_special_inode_operations;
2233
2234                 init_special_inode(inode, inode->i_mode,
2235                                    inode->i_rdev);
2236
2237                 EXIT;
2238         }
2239
2240         return 0;
2241 }
2242
2243 void ll_delete_inode(struct inode *inode)
2244 {
2245         struct ll_inode_info *lli = ll_i2info(inode);
2246         struct address_space *mapping = &inode->i_data;
2247         unsigned long nrpages;
2248         unsigned long flags;
2249
2250         ENTRY;
2251
2252         if (S_ISREG(inode->i_mode) && lli->lli_clob != NULL) {
2253                 /* It is last chance to write out dirty pages,
2254                  * otherwise we may lose data while umount.
2255                  *
2256                  * If i_nlink is 0 then just discard data. This is safe because
2257                  * local inode gets i_nlink 0 from server only for the last
2258                  * unlink, so that file is not opened somewhere else
2259                  */
2260                 cl_sync_file_range(inode, 0, OBD_OBJECT_EOF, inode->i_nlink ?
2261                                    CL_FSYNC_LOCAL : CL_FSYNC_DISCARD, 1);
2262         }
2263         truncate_inode_pages_final(mapping);
2264
2265         /* Workaround for LU-118: Note nrpages may not be totally updated when
2266          * truncate_inode_pages() returns, as there can be a page in the process
2267          * of deletion (inside __delete_from_page_cache()) in the specified
2268          * range. Thus mapping->nrpages can be non-zero when this function
2269          * returns even after truncation of the whole mapping.  Only do this if
2270          * npages isn't already zero.
2271          */
2272         nrpages = mapping->nrpages;
2273         if (nrpages) {
2274                 xa_lock_irqsave(&mapping->i_pages, flags);
2275                 nrpages = mapping->nrpages;
2276                 xa_unlock_irqrestore(&mapping->i_pages, flags);
2277         } /* Workaround end */
2278
2279         LASSERTF(nrpages == 0, "%s: inode="DFID"(%p) nrpages=%lu, "
2280                  "see https://jira.whamcloud.com/browse/LU-118\n",
2281                  ll_i2sbi(inode)->ll_fsname,
2282                  PFID(ll_inode2fid(inode)), inode, nrpages);
2283
2284         ll_clear_inode(inode);
2285         clear_inode(inode);
2286
2287         EXIT;
2288 }
2289
2290 int ll_iocontrol(struct inode *inode, struct file *file,
2291                  unsigned int cmd, unsigned long arg)
2292 {
2293         struct ll_sb_info *sbi = ll_i2sbi(inode);
2294         struct ptlrpc_request *req = NULL;
2295         int rc, flags = 0;
2296         ENTRY;
2297
2298         switch (cmd) {
2299         case FS_IOC_GETFLAGS: {
2300                 struct mdt_body *body;
2301                 struct md_op_data *op_data;
2302
2303                 op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL,
2304                                              0, 0, LUSTRE_OPC_ANY,
2305                                              NULL);
2306                 if (IS_ERR(op_data))
2307                         RETURN(PTR_ERR(op_data));
2308
2309                 op_data->op_valid = OBD_MD_FLFLAGS;
2310                 rc = md_getattr(sbi->ll_md_exp, op_data, &req);
2311                 ll_finish_md_op_data(op_data);
2312                 if (rc) {
2313                         CERROR("%s: failure inode "DFID": rc = %d\n",
2314                                sbi->ll_md_exp->exp_obd->obd_name,
2315                                PFID(ll_inode2fid(inode)), rc);
2316                         RETURN(-abs(rc));
2317                 }
2318
2319                 body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
2320
2321                 flags = body->mbo_flags;
2322
2323                 ptlrpc_req_finished(req);
2324
2325                 RETURN(put_user(flags, (int __user *)arg));
2326         }
2327         case FS_IOC_SETFLAGS: {
2328                 struct iattr *attr;
2329                 struct md_op_data *op_data;
2330                 struct cl_object *obj;
2331                 struct fsxattr fa = { 0 };
2332
2333                 if (get_user(flags, (int __user *)arg))
2334                         RETURN(-EFAULT);
2335
2336                 fa.fsx_projid = ll_i2info(inode)->lli_projid;
2337                 if (flags & LUSTRE_PROJINHERIT_FL)
2338                         fa.fsx_xflags = FS_XFLAG_PROJINHERIT;
2339
2340                 rc = ll_ioctl_check_project(inode, &fa);
2341                 if (rc)
2342                         RETURN(rc);
2343
2344                 op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL, 0, 0,
2345                                              LUSTRE_OPC_ANY, NULL);
2346                 if (IS_ERR(op_data))
2347                         RETURN(PTR_ERR(op_data));
2348
2349                 op_data->op_attr_flags = flags;
2350                 op_data->op_xvalid |= OP_XVALID_FLAGS;
2351                 rc = md_setattr(sbi->ll_md_exp, op_data, NULL, 0, &req);
2352                 ll_finish_md_op_data(op_data);
2353                 ptlrpc_req_finished(req);
2354                 if (rc)
2355                         RETURN(rc);
2356
2357                 ll_update_inode_flags(inode, flags);
2358
2359                 obj = ll_i2info(inode)->lli_clob;
2360                 if (obj == NULL)
2361                         RETURN(0);
2362
2363                 OBD_ALLOC_PTR(attr);
2364                 if (attr == NULL)
2365                         RETURN(-ENOMEM);
2366
2367                 rc = cl_setattr_ost(obj, attr, OP_XVALID_FLAGS, flags);
2368
2369                 OBD_FREE_PTR(attr);
2370                 RETURN(rc);
2371         }
2372         default:
2373                 RETURN(-ENOSYS);
2374         }
2375
2376         RETURN(0);
2377 }
2378
2379 int ll_flush_ctx(struct inode *inode)
2380 {
2381         struct ll_sb_info  *sbi = ll_i2sbi(inode);
2382
2383         CDEBUG(D_SEC, "flush context for user %d\n",
2384                from_kuid(&init_user_ns, current_uid()));
2385
2386         obd_set_info_async(NULL, sbi->ll_md_exp,
2387                            sizeof(KEY_FLUSH_CTX), KEY_FLUSH_CTX,
2388                            0, NULL, NULL);
2389         obd_set_info_async(NULL, sbi->ll_dt_exp,
2390                            sizeof(KEY_FLUSH_CTX), KEY_FLUSH_CTX,
2391                            0, NULL, NULL);
2392         return 0;
2393 }
2394
2395 /* umount -f client means force down, don't save state */
2396 void ll_umount_begin(struct super_block *sb)
2397 {
2398         struct ll_sb_info *sbi = ll_s2sbi(sb);
2399         struct obd_device *obd;
2400         struct obd_ioctl_data *ioc_data;
2401         int cnt;
2402         ENTRY;
2403
2404         CDEBUG(D_VFSTRACE, "VFS Op: superblock %p count %d active %d\n", sb,
2405                sb->s_count, atomic_read(&sb->s_active));
2406
2407         obd = class_exp2obd(sbi->ll_md_exp);
2408         if (obd == NULL) {
2409                 CERROR("Invalid MDC connection handle %#llx\n",
2410                        sbi->ll_md_exp->exp_handle.h_cookie);
2411                 EXIT;
2412                 return;
2413         }
2414         obd->obd_force = 1;
2415
2416         obd = class_exp2obd(sbi->ll_dt_exp);
2417         if (obd == NULL) {
2418                 CERROR("Invalid LOV connection handle %#llx\n",
2419                        sbi->ll_dt_exp->exp_handle.h_cookie);
2420                 EXIT;
2421                 return;
2422         }
2423         obd->obd_force = 1;
2424
2425         OBD_ALLOC_PTR(ioc_data);
2426         if (ioc_data) {
2427                 obd_iocontrol(IOC_OSC_SET_ACTIVE, sbi->ll_md_exp,
2428                               sizeof *ioc_data, ioc_data, NULL);
2429
2430                 obd_iocontrol(IOC_OSC_SET_ACTIVE, sbi->ll_dt_exp,
2431                               sizeof *ioc_data, ioc_data, NULL);
2432
2433                 OBD_FREE_PTR(ioc_data);
2434         }
2435
2436         /* Really, we'd like to wait until there are no requests outstanding,
2437          * and then continue.  For now, we just periodically checking for vfs
2438          * to decrement mnt_cnt and hope to finish it within 10sec.
2439          */
2440         cnt = 10;
2441         while (cnt > 0 &&
2442                !may_umount(sbi->ll_mnt.mnt)) {
2443                 ssleep(1);
2444                 cnt -= 1;
2445         }
2446
2447         EXIT;
2448 }
2449
2450 int ll_remount_fs(struct super_block *sb, int *flags, char *data)
2451 {
2452         struct ll_sb_info *sbi = ll_s2sbi(sb);
2453         char *profilenm = get_profile_name(sb);
2454         int err;
2455         __u32 read_only;
2456
2457         if ((*flags & MS_RDONLY) != (sb->s_flags & SB_RDONLY)) {
2458                 read_only = *flags & MS_RDONLY;
2459                 err = obd_set_info_async(NULL, sbi->ll_md_exp,
2460                                          sizeof(KEY_READ_ONLY),
2461                                          KEY_READ_ONLY, sizeof(read_only),
2462                                          &read_only, NULL);
2463                 if (err) {
2464                         LCONSOLE_WARN("Failed to remount %s %s (%d)\n",
2465                                       profilenm, read_only ?
2466                                       "read-only" : "read-write", err);
2467                         return err;
2468                 }
2469
2470                 if (read_only)
2471                         sb->s_flags |= SB_RDONLY;
2472                 else
2473                         sb->s_flags &= ~SB_RDONLY;
2474
2475                 if (sbi->ll_flags & LL_SBI_VERBOSE)
2476                         LCONSOLE_WARN("Remounted %s %s\n", profilenm,
2477                                       read_only ?  "read-only" : "read-write");
2478         }
2479         return 0;
2480 }
2481
2482 /**
2483  * Cleanup the open handle that is cached on MDT-side.
2484  *
2485  * For open case, the client side open handling thread may hit error
2486  * after the MDT grant the open. Under such case, the client should
2487  * send close RPC to the MDT as cleanup; otherwise, the open handle
2488  * on the MDT will be leaked there until the client umount or evicted.
2489  *
2490  * In further, if someone unlinked the file, because the open handle
2491  * holds the reference on such file/object, then it will block the
2492  * subsequent threads that want to locate such object via FID.
2493  *
2494  * \param[in] sb        super block for this file-system
2495  * \param[in] open_req  pointer to the original open request
2496  */
2497 void ll_open_cleanup(struct super_block *sb, struct ptlrpc_request *open_req)
2498 {
2499         struct mdt_body                 *body;
2500         struct md_op_data               *op_data;
2501         struct ptlrpc_request           *close_req = NULL;
2502         struct obd_export               *exp       = ll_s2sbi(sb)->ll_md_exp;
2503         ENTRY;
2504
2505         body = req_capsule_server_get(&open_req->rq_pill, &RMF_MDT_BODY);
2506         OBD_ALLOC_PTR(op_data);
2507         if (op_data == NULL) {
2508                 CWARN("%s: cannot allocate op_data to release open handle for "
2509                       DFID"\n", ll_s2sbi(sb)->ll_fsname, PFID(&body->mbo_fid1));
2510
2511                 RETURN_EXIT;
2512         }
2513
2514         op_data->op_fid1 = body->mbo_fid1;
2515         op_data->op_open_handle = body->mbo_open_handle;
2516         op_data->op_mod_time = ktime_get_real_seconds();
2517         md_close(exp, op_data, NULL, &close_req);
2518         ptlrpc_req_finished(close_req);
2519         ll_finish_md_op_data(op_data);
2520
2521         EXIT;
2522 }
2523
2524 int ll_prep_inode(struct inode **inode, struct ptlrpc_request *req,
2525                   struct super_block *sb, struct lookup_intent *it)
2526 {
2527         struct ll_sb_info *sbi = NULL;
2528         struct lustre_md md = { NULL };
2529         bool default_lmv_deleted = false;
2530         int rc;
2531
2532         ENTRY;
2533
2534         LASSERT(*inode || sb);
2535         sbi = sb ? ll_s2sbi(sb) : ll_i2sbi(*inode);
2536         rc = md_get_lustre_md(sbi->ll_md_exp, req, sbi->ll_dt_exp,
2537                               sbi->ll_md_exp, &md);
2538         if (rc != 0)
2539                 GOTO(out, rc);
2540
2541         /*
2542          * clear default_lmv only if intent_getattr reply doesn't contain it.
2543          * but it needs to be done after iget, check this early because
2544          * ll_update_lsm_md() may change md.
2545          */
2546         if (it && (it->it_op & (IT_LOOKUP | IT_GETATTR)) &&
2547             S_ISDIR(md.body->mbo_mode) && !md.default_lmv)
2548                 default_lmv_deleted = true;
2549
2550         if (*inode) {
2551                 rc = ll_update_inode(*inode, &md);
2552                 if (rc != 0)
2553                         GOTO(out, rc);
2554         } else {
2555                 LASSERT(sb != NULL);
2556
2557                 /*
2558                  * At this point server returns to client's same fid as client
2559                  * generated for creating. So using ->fid1 is okay here.
2560                  */
2561                 if (!fid_is_sane(&md.body->mbo_fid1)) {
2562                         CERROR("%s: Fid is insane "DFID"\n",
2563                                 sbi->ll_fsname,
2564                                 PFID(&md.body->mbo_fid1));
2565                         GOTO(out, rc = -EINVAL);
2566                 }
2567
2568                 *inode = ll_iget(sb, cl_fid_build_ino(&md.body->mbo_fid1,
2569                                              sbi->ll_flags & LL_SBI_32BIT_API),
2570                                  &md);
2571                 if (IS_ERR(*inode)) {
2572 #ifdef CONFIG_LUSTRE_FS_POSIX_ACL
2573                         if (md.posix_acl) {
2574                                 posix_acl_release(md.posix_acl);
2575                                 md.posix_acl = NULL;
2576                         }
2577 #endif
2578                         rc = IS_ERR(*inode) ? PTR_ERR(*inode) : -ENOMEM;
2579                         *inode = NULL;
2580                         CERROR("new_inode -fatal: rc %d\n", rc);
2581                         GOTO(out, rc);
2582                 }
2583         }
2584
2585         /* Handling piggyback layout lock.
2586          * Layout lock can be piggybacked by getattr and open request.
2587          * The lsm can be applied to inode only if it comes with a layout lock
2588          * otherwise correct layout may be overwritten, for example:
2589          * 1. proc1: mdt returns a lsm but not granting layout
2590          * 2. layout was changed by another client
2591          * 3. proc2: refresh layout and layout lock granted
2592          * 4. proc1: to apply a stale layout */
2593         if (it != NULL && it->it_lock_mode != 0) {
2594                 struct lustre_handle lockh;
2595                 struct ldlm_lock *lock;
2596
2597                 lockh.cookie = it->it_lock_handle;
2598                 lock = ldlm_handle2lock(&lockh);
2599                 LASSERT(lock != NULL);
2600                 if (ldlm_has_layout(lock)) {
2601                         struct cl_object_conf conf;
2602
2603                         memset(&conf, 0, sizeof(conf));
2604                         conf.coc_opc = OBJECT_CONF_SET;
2605                         conf.coc_inode = *inode;
2606                         conf.coc_lock = lock;
2607                         conf.u.coc_layout = md.layout;
2608                         (void)ll_layout_conf(*inode, &conf);
2609                 }
2610                 LDLM_LOCK_PUT(lock);
2611         }
2612
2613         if (default_lmv_deleted)
2614                 ll_update_default_lsm_md(*inode, &md);
2615
2616         GOTO(out, rc = 0);
2617
2618 out:
2619         /* cleanup will be done if necessary */
2620         md_free_lustre_md(sbi->ll_md_exp, &md);
2621
2622         if (rc != 0 && it != NULL && it->it_op & IT_OPEN) {
2623                 ll_intent_drop_lock(it);
2624                 ll_open_cleanup(sb != NULL ? sb : (*inode)->i_sb, req);
2625         }
2626
2627         return rc;
2628 }
2629
2630 int ll_obd_statfs(struct inode *inode, void __user *arg)
2631 {
2632         struct ll_sb_info *sbi = NULL;
2633         struct obd_export *exp;
2634         char *buf = NULL;
2635         struct obd_ioctl_data *data = NULL;
2636         __u32 type;
2637         int len = 0, rc;
2638
2639         if (!inode || !(sbi = ll_i2sbi(inode)))
2640                 GOTO(out_statfs, rc = -EINVAL);
2641
2642         rc = obd_ioctl_getdata(&buf, &len, arg);
2643         if (rc)
2644                 GOTO(out_statfs, rc);
2645
2646         data = (void*)buf;
2647         if (!data->ioc_inlbuf1 || !data->ioc_inlbuf2 ||
2648             !data->ioc_pbuf1 || !data->ioc_pbuf2)
2649                 GOTO(out_statfs, rc = -EINVAL);
2650
2651         if (data->ioc_inllen1 != sizeof(__u32) ||
2652             data->ioc_inllen2 != sizeof(__u32) ||
2653             data->ioc_plen1 != sizeof(struct obd_statfs) ||
2654             data->ioc_plen2 != sizeof(struct obd_uuid))
2655                 GOTO(out_statfs, rc = -EINVAL);
2656
2657         memcpy(&type, data->ioc_inlbuf1, sizeof(__u32));
2658         if (type & LL_STATFS_LMV)
2659                 exp = sbi->ll_md_exp;
2660         else if (type & LL_STATFS_LOV)
2661                 exp = sbi->ll_dt_exp;
2662         else
2663                 GOTO(out_statfs, rc = -ENODEV);
2664
2665         rc = obd_iocontrol(IOC_OBD_STATFS, exp, len, buf, NULL);
2666         if (rc)
2667                 GOTO(out_statfs, rc);
2668 out_statfs:
2669         OBD_FREE_LARGE(buf, len);
2670         return rc;
2671 }
2672
2673 /*
2674  * this is normally called in ll_fini_md_op_data(), but sometimes it needs to
2675  * be called early to avoid deadlock.
2676  */
2677 void ll_unlock_md_op_lsm(struct md_op_data *op_data)
2678 {
2679         if (op_data->op_mea2_sem) {
2680                 up_read(op_data->op_mea2_sem);
2681                 op_data->op_mea2_sem = NULL;
2682         }
2683
2684         if (op_data->op_mea1_sem) {
2685                 up_read(op_data->op_mea1_sem);
2686                 op_data->op_mea1_sem = NULL;
2687         }
2688 }
2689
2690 /* this function prepares md_op_data hint for passing it down to MD stack. */
2691 struct md_op_data *ll_prep_md_op_data(struct md_op_data *op_data,
2692                                       struct inode *i1, struct inode *i2,
2693                                       const char *name, size_t namelen,
2694                                       __u32 mode, enum md_op_code opc,
2695                                       void *data)
2696 {
2697         LASSERT(i1 != NULL);
2698
2699         if (name == NULL) {
2700                 /* Do not reuse namelen for something else. */
2701                 if (namelen != 0)
2702                         return ERR_PTR(-EINVAL);
2703         } else {
2704                 if (namelen > ll_i2sbi(i1)->ll_namelen)
2705                         return ERR_PTR(-ENAMETOOLONG);
2706
2707                 if (!lu_name_is_valid_2(name, namelen))
2708                         return ERR_PTR(-EINVAL);
2709         }
2710
2711         if (op_data == NULL)
2712                 OBD_ALLOC_PTR(op_data);
2713
2714         if (op_data == NULL)
2715                 return ERR_PTR(-ENOMEM);
2716
2717         ll_i2gids(op_data->op_suppgids, i1, i2);
2718         op_data->op_fid1 = *ll_inode2fid(i1);
2719         op_data->op_code = opc;
2720
2721         if (S_ISDIR(i1->i_mode)) {
2722                 down_read(&ll_i2info(i1)->lli_lsm_sem);
2723                 op_data->op_mea1_sem = &ll_i2info(i1)->lli_lsm_sem;
2724                 op_data->op_mea1 = ll_i2info(i1)->lli_lsm_md;
2725                 op_data->op_default_mea1 = ll_i2info(i1)->lli_default_lsm_md;
2726         }
2727
2728         if (i2) {
2729                 op_data->op_fid2 = *ll_inode2fid(i2);
2730                 if (S_ISDIR(i2->i_mode)) {
2731                         if (i2 != i1) {
2732                                 down_read(&ll_i2info(i2)->lli_lsm_sem);
2733                                 op_data->op_mea2_sem =
2734                                                 &ll_i2info(i2)->lli_lsm_sem;
2735                         }
2736                         op_data->op_mea2 = ll_i2info(i2)->lli_lsm_md;
2737                 }
2738         } else {
2739                 fid_zero(&op_data->op_fid2);
2740         }
2741
2742         if (ll_i2sbi(i1)->ll_flags & LL_SBI_64BIT_HASH)
2743                 op_data->op_cli_flags |= CLI_HASH64;
2744
2745         if (ll_need_32bit_api(ll_i2sbi(i1)))
2746                 op_data->op_cli_flags |= CLI_API32;
2747
2748         op_data->op_name = name;
2749         op_data->op_namelen = namelen;
2750         op_data->op_mode = mode;
2751         op_data->op_mod_time = ktime_get_real_seconds();
2752         op_data->op_fsuid = from_kuid(&init_user_ns, current_fsuid());
2753         op_data->op_fsgid = from_kgid(&init_user_ns, current_fsgid());
2754         op_data->op_cap = cfs_curproc_cap_pack();
2755         op_data->op_mds = 0;
2756         if ((opc == LUSTRE_OPC_CREATE) && (name != NULL) &&
2757              filename_is_volatile(name, namelen, &op_data->op_mds)) {
2758                 op_data->op_bias |= MDS_CREATE_VOLATILE;
2759         }
2760         op_data->op_data = data;
2761
2762         return op_data;
2763 }
2764
2765 void ll_finish_md_op_data(struct md_op_data *op_data)
2766 {
2767         ll_unlock_md_op_lsm(op_data);
2768         security_release_secctx(op_data->op_file_secctx,
2769                                 op_data->op_file_secctx_size);
2770         OBD_FREE_PTR(op_data);
2771 }
2772
2773 int ll_show_options(struct seq_file *seq, struct dentry *dentry)
2774 {
2775         struct ll_sb_info *sbi;
2776
2777         LASSERT(seq && dentry);
2778         sbi = ll_s2sbi(dentry->d_sb);
2779
2780         if (sbi->ll_flags & LL_SBI_NOLCK)
2781                 seq_puts(seq, ",nolock");
2782
2783         /* "flock" is the default since 2.13, but it wasn't for many years,
2784          * so it is still useful to print this to show it is enabled.
2785          * Start to print "noflock" so it is now clear when flock is disabled.
2786          */
2787         if (sbi->ll_flags & LL_SBI_FLOCK)
2788                 seq_puts(seq, ",flock");
2789         else if (sbi->ll_flags & LL_SBI_LOCALFLOCK)
2790                 seq_puts(seq, ",localflock");
2791         else
2792                 seq_puts(seq, ",noflock");
2793
2794         if (sbi->ll_flags & LL_SBI_USER_XATTR)
2795                 seq_puts(seq, ",user_xattr");
2796
2797         if (sbi->ll_flags & LL_SBI_LAZYSTATFS)
2798                 seq_puts(seq, ",lazystatfs");
2799
2800         if (sbi->ll_flags & LL_SBI_USER_FID2PATH)
2801                 seq_puts(seq, ",user_fid2path");
2802
2803         if (sbi->ll_flags & LL_SBI_ALWAYS_PING)
2804                 seq_puts(seq, ",always_ping");
2805
2806         RETURN(0);
2807 }
2808
2809 /**
2810  * Get obd name by cmd, and copy out to user space
2811  */
2812 int ll_get_obd_name(struct inode *inode, unsigned int cmd, unsigned long arg)
2813 {
2814         struct ll_sb_info *sbi = ll_i2sbi(inode);
2815         struct obd_device *obd;
2816         ENTRY;
2817
2818         if (cmd == OBD_IOC_GETDTNAME)
2819                 obd = class_exp2obd(sbi->ll_dt_exp);
2820         else if (cmd == OBD_IOC_GETMDNAME)
2821                 obd = class_exp2obd(sbi->ll_md_exp);
2822         else
2823                 RETURN(-EINVAL);
2824
2825         if (!obd)
2826                 RETURN(-ENOENT);
2827
2828         if (copy_to_user((void __user *)arg, obd->obd_name,
2829                          strlen(obd->obd_name) + 1))
2830                 RETURN(-EFAULT);
2831
2832         RETURN(0);
2833 }
2834
2835 static char* ll_d_path(struct dentry *dentry, char *buf, int bufsize)
2836 {
2837         char *path = NULL;
2838
2839         struct path p;
2840
2841         p.dentry = dentry;
2842         p.mnt = current->fs->root.mnt;
2843         path_get(&p);
2844         path = d_path(&p, buf, bufsize);
2845         path_put(&p);
2846         return path;
2847 }
2848
2849 void ll_dirty_page_discard_warn(struct page *page, int ioret)
2850 {
2851         char *buf, *path = NULL;
2852         struct dentry *dentry = NULL;
2853         struct inode *inode = page->mapping->host;
2854
2855         /* this can be called inside spin lock so use GFP_ATOMIC. */
2856         buf = (char *)__get_free_page(GFP_ATOMIC);
2857         if (buf != NULL) {
2858                 dentry = d_find_alias(page->mapping->host);
2859                 if (dentry != NULL)
2860                         path = ll_d_path(dentry, buf, PAGE_SIZE);
2861         }
2862
2863         /* The below message is checked in recovery-small.sh test_24b */
2864         CDEBUG(D_WARNING,
2865                "%s: dirty page discard: %s/fid: "DFID"/%s may get corrupted "
2866                "(rc %d)\n", ll_i2sbi(inode)->ll_fsname,
2867                s2lsi(page->mapping->host->i_sb)->lsi_lmd->lmd_dev,
2868                PFID(ll_inode2fid(inode)),
2869                (path && !IS_ERR(path)) ? path : "", ioret);
2870
2871         if (dentry != NULL)
2872                 dput(dentry);
2873
2874         if (buf != NULL)
2875                 free_page((unsigned long)buf);
2876 }
2877
2878 ssize_t ll_copy_user_md(const struct lov_user_md __user *md,
2879                         struct lov_user_md **kbuf)
2880 {
2881         struct lov_user_md      lum;
2882         ssize_t                 lum_size;
2883         ENTRY;
2884
2885         if (copy_from_user(&lum, md, sizeof(lum)))
2886                 RETURN(-EFAULT);
2887
2888         lum_size = ll_lov_user_md_size(&lum);
2889         if (lum_size < 0)
2890                 RETURN(lum_size);
2891
2892         OBD_ALLOC_LARGE(*kbuf, lum_size);
2893         if (*kbuf == NULL)
2894                 RETURN(-ENOMEM);
2895
2896         if (copy_from_user(*kbuf, md, lum_size) != 0) {
2897                 OBD_FREE_LARGE(*kbuf, lum_size);
2898                 RETURN(-EFAULT);
2899         }
2900
2901         RETURN(lum_size);
2902 }
2903
2904 /*
2905  * Compute llite root squash state after a change of root squash
2906  * configuration setting or add/remove of a lnet nid
2907  */
2908 void ll_compute_rootsquash_state(struct ll_sb_info *sbi)
2909 {
2910         struct root_squash_info *squash = &sbi->ll_squash;
2911         int i;
2912         bool matched;
2913         struct lnet_process_id id;
2914
2915         /* Update norootsquash flag */
2916         spin_lock(&squash->rsi_lock);
2917         if (list_empty(&squash->rsi_nosquash_nids))
2918                 sbi->ll_flags &= ~LL_SBI_NOROOTSQUASH;
2919         else {
2920                 /* Do not apply root squash as soon as one of our NIDs is
2921                  * in the nosquash_nids list */
2922                 matched = false;
2923                 i = 0;
2924                 while (LNetGetId(i++, &id) != -ENOENT) {
2925                         if (LNET_NETTYP(LNET_NIDNET(id.nid)) == LOLND)
2926                                 continue;
2927                         if (cfs_match_nid(id.nid, &squash->rsi_nosquash_nids)) {
2928                                 matched = true;
2929                                 break;
2930                         }
2931                 }
2932                 if (matched)
2933                         sbi->ll_flags |= LL_SBI_NOROOTSQUASH;
2934                 else
2935                         sbi->ll_flags &= ~LL_SBI_NOROOTSQUASH;
2936         }
2937         spin_unlock(&squash->rsi_lock);
2938 }
2939
2940 /**
2941  * Parse linkea content to extract information about a given hardlink
2942  *
2943  * \param[in]   ldata      - Initialized linkea data
2944  * \param[in]   linkno     - Link identifier
2945  * \param[out]  parent_fid - The entry's parent FID
2946  * \param[out]  ln         - Entry name destination buffer
2947  *
2948  * \retval 0 on success
2949  * \retval Appropriate negative error code on failure
2950  */
2951 static int ll_linkea_decode(struct linkea_data *ldata, unsigned int linkno,
2952                             struct lu_fid *parent_fid, struct lu_name *ln)
2953 {
2954         unsigned int    idx;
2955         int             rc;
2956         ENTRY;
2957
2958         rc = linkea_init_with_rec(ldata);
2959         if (rc < 0)
2960                 RETURN(rc);
2961
2962         if (linkno >= ldata->ld_leh->leh_reccount)
2963                 /* beyond last link */
2964                 RETURN(-ENODATA);
2965
2966         linkea_first_entry(ldata);
2967         for (idx = 0; ldata->ld_lee != NULL; idx++) {
2968                 linkea_entry_unpack(ldata->ld_lee, &ldata->ld_reclen, ln,
2969                                     parent_fid);
2970                 if (idx == linkno)
2971                         break;
2972
2973                 linkea_next_entry(ldata);
2974         }
2975
2976         if (idx < linkno)
2977                 RETURN(-ENODATA);
2978
2979         RETURN(0);
2980 }
2981
2982 /**
2983  * Get parent FID and name of an identified link. Operation is performed for
2984  * a given link number, letting the caller iterate over linkno to list one or
2985  * all links of an entry.
2986  *
2987  * \param[in]     file - File descriptor against which to perform the operation
2988  * \param[in,out] arg  - User-filled structure containing the linkno to operate
2989  *                       on and the available size. It is eventually filled with
2990  *                       the requested information or left untouched on error
2991  *
2992  * \retval - 0 on success
2993  * \retval - Appropriate negative error code on failure
2994  */
2995 int ll_getparent(struct file *file, struct getparent __user *arg)
2996 {
2997         struct inode            *inode = file_inode(file);
2998         struct linkea_data      *ldata;
2999         struct lu_buf            buf = LU_BUF_NULL;
3000         struct lu_name           ln;
3001         struct lu_fid            parent_fid;
3002         __u32                    linkno;
3003         __u32                    name_size;
3004         int                      rc;
3005
3006         ENTRY;
3007
3008         if (!cfs_capable(CFS_CAP_DAC_READ_SEARCH) &&
3009             !(ll_i2sbi(inode)->ll_flags & LL_SBI_USER_FID2PATH))
3010                 RETURN(-EPERM);
3011
3012         if (get_user(name_size, &arg->gp_name_size))
3013                 RETURN(-EFAULT);
3014
3015         if (get_user(linkno, &arg->gp_linkno))
3016                 RETURN(-EFAULT);
3017
3018         if (name_size > PATH_MAX)
3019                 RETURN(-EINVAL);
3020
3021         OBD_ALLOC(ldata, sizeof(*ldata));
3022         if (ldata == NULL)
3023                 RETURN(-ENOMEM);
3024
3025         rc = linkea_data_new(ldata, &buf);
3026         if (rc < 0)
3027                 GOTO(ldata_free, rc);
3028
3029         rc = ll_xattr_list(inode, XATTR_NAME_LINK, XATTR_TRUSTED_T, buf.lb_buf,
3030                            buf.lb_len, OBD_MD_FLXATTR);
3031         if (rc < 0)
3032                 GOTO(lb_free, rc);
3033
3034         rc = ll_linkea_decode(ldata, linkno, &parent_fid, &ln);
3035         if (rc < 0)
3036                 GOTO(lb_free, rc);
3037
3038         if (ln.ln_namelen >= name_size)
3039                 GOTO(lb_free, rc = -EOVERFLOW);
3040
3041         if (copy_to_user(&arg->gp_fid, &parent_fid, sizeof(arg->gp_fid)))
3042                 GOTO(lb_free, rc = -EFAULT);
3043
3044         if (copy_to_user(&arg->gp_name, ln.ln_name, ln.ln_namelen))
3045                 GOTO(lb_free, rc = -EFAULT);
3046
3047         if (put_user('\0', arg->gp_name + ln.ln_namelen))
3048                 GOTO(lb_free, rc = -EFAULT);
3049
3050 lb_free:
3051         lu_buf_free(&buf);
3052 ldata_free:
3053         OBD_FREE(ldata, sizeof(*ldata));
3054
3055         RETURN(rc);
3056 }