lustre/llite/llite_lib.c

   1 /* -*- mode: c; c-basic-offset: 8; indent-tabs-mode: nil; -*-
   2  * vim:expandtab:shiftwidth=8:tabstop=8:
   3  *
   4  * GPL HEADER START
   5  *
   6  * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
   7  *
   8  * This program is free software; you can redistribute it and/or modify
   9  * it under the terms of the GNU General Public License version 2 only,
  10  * as published by the Free Software Foundation.
  11  *
  12  * This program is distributed in the hope that it will be useful, but
  13  * WITHOUT ANY WARRANTY; without even the implied warranty of
  14  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  15  * General Public License version 2 for more details (a copy is included
  16  * in the LICENSE file that accompanied this code).
  17  *
  18  * You should have received a copy of the GNU General Public License
  19  * version 2 along with this program; If not, see
  20  * http://www.sun.com/software/products/lustre/docs/GPLv2.pdf
  21  *
  22  * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara,
  23  * CA 95054 USA or visit www.sun.com if you need additional information or
  24  * have any questions.
  25  *
  26  * GPL HEADER END
  27  */
  28 /*
  29  * Copyright  2008 Sun Microsystems, Inc. All rights reserved
  30  * Use is subject to license terms.
  31  */
  32 /*
  33  * This file is part of Lustre, http://www.lustre.org/
  34  * Lustre is a trademark of Sun Microsystems, Inc.
  35  *
  36  * lustre/llite/llite_lib.c
  37  *
  38  * Lustre Light Super operations
  39  */
  40
  41 #define DEBUG_SUBSYSTEM S_LLITE
  42
  43 #include <linux/module.h>
  44 #include <linux/types.h>
  45 #include <linux/random.h>
  46 #include <linux/version.h>
  47
  48 #include <lustre_lite.h>
  49 #include <lustre_ha.h>
  50 #include <lustre_dlm.h>
  51 #include <lprocfs_status.h>
  52 #include <lustre_disk.h>
  53 #include <lustre_param.h>
  54 #include <lustre_cache.h>
  55 #include <obd_support.h>
  56 #include "llite_internal.h"
  57
  58 cfs_mem_cache_t *ll_file_data_slab;
  59
  60 LIST_HEAD(ll_super_blocks);
  61 struct rw_semaphore ll_sb_sem;
  62
  63 extern struct address_space_operations ll_aops;
  64 extern struct address_space_operations ll_dir_aops;
  65
  66 #ifndef log2
  67 #define log2(n) ffz(~(n))
  68 #endif
  69
  70 static void ll_pglist_fini(struct ll_sb_info *sbi)
  71 {
  72         struct page *page;
  73         int i;
  74
  75         if (sbi->ll_pglist == NULL)
  76                 return;
  77
  78         for_each_possible_cpu(i) {
  79                 page = sbi->ll_pglist[i]->llpd_page;
  80                 if (page) {
  81                         sbi->ll_pglist[i] = NULL;
  82                         __free_page(page);
  83                 }
  84         }
  85
  86         OBD_FREE(sbi->ll_pglist, sizeof(void *)*num_possible_cpus());
  87         sbi->ll_pglist = NULL;
  88 }
  89
  90 static int ll_pglist_init(struct ll_sb_info *sbi)
  91 {
  92         struct ll_pglist_data *pd;
  93         unsigned long budget;
  94         int i, color = 0;
  95         ENTRY;
  96
  97         OBD_ALLOC(sbi->ll_pglist, sizeof(void *) * num_possible_cpus());
  98         if (sbi->ll_pglist == NULL)
  99                 RETURN(-ENOMEM);
 100
 101         budget = sbi->ll_async_page_max / num_online_cpus();
 102         for_each_possible_cpu(i) {
 103                 struct page *page = alloc_pages_node(cpu_to_node(i),
 104                                                     GFP_KERNEL, 0);
 105                 if (page == NULL) {
 106                         ll_pglist_fini(sbi);
 107                         RETURN(-ENOMEM);
 108                 }
 109
 110                 if (color + L1_CACHE_ALIGN(sizeof(*pd)) > PAGE_SIZE)
 111                         color = 0;
 112
 113                 pd = (struct ll_pglist_data *)(page_address(page) + color);
 114                 memset(pd, 0, sizeof(*pd));
 115                 spin_lock_init(&pd->llpd_lock);
 116                 INIT_LIST_HEAD(&pd->llpd_list);
 117                 if (cpu_online(i))
 118                         pd->llpd_budget = budget;
 119                 pd->llpd_cpu = i;
 120                 pd->llpd_page = page;
 121                 atomic_set(&pd->llpd_sample_count, 0);
 122                 sbi->ll_pglist[i] = pd;
 123                 color += L1_CACHE_ALIGN(sizeof(*pd));
 124         }
 125
 126         RETURN(0);
 127 }
 128
 129 static struct ll_sb_info *ll_init_sbi(void)
 130 {
 131         struct ll_sb_info *sbi = NULL;
 132         unsigned long pages;
 133         struct sysinfo si;
 134         class_uuid_t uuid;
 135         int i;
 136         ENTRY;
 137
 138         OBD_ALLOC(sbi, sizeof(*sbi));
 139         if (!sbi)
 140                 RETURN(NULL);
 141
 142         OBD_ALLOC(sbi->ll_async_page_sample, sizeof(long)*num_possible_cpus());
 143         if (sbi->ll_async_page_sample == NULL)
 144                 GOTO(out, 0);
 145
 146         spin_lock_init(&sbi->ll_lock);
 147         init_mutex(&sbi->ll_lco.lco_lock);
 148         spin_lock_init(&sbi->ll_pp_extent_lock);
 149         spin_lock_init(&sbi->ll_process_lock);
 150         sbi->ll_rw_stats_on = 0;
 151
 152         si_meminfo(&si);
 153         pages = si.totalram - si.totalhigh;
 154         if (pages >> (20 - CFS_PAGE_SHIFT) < 512) {
 155 #ifdef HAVE_BGL_SUPPORT
 156                 sbi->ll_async_page_max = pages / 4;
 157 #else
 158                 sbi->ll_async_page_max = pages / 2;
 159 #endif
 160         } else {
 161                 sbi->ll_async_page_max = (pages / 4) * 3;
 162         }
 163
 164         lcounter_init(&sbi->ll_async_page_count);
 165         spin_lock_init(&sbi->ll_async_page_reblnc_lock);
 166         sbi->ll_async_page_sample_max = 64 * num_online_cpus();
 167         sbi->ll_async_page_reblnc_count = 0;
 168         sbi->ll_async_page_clock_hand = 0;
 169         if (ll_pglist_init(sbi))
 170                 GOTO(out, 0);
 171
 172         sbi->ll_ra_info.ra_max_pages_per_file = min(pages / 32,
 173                                            SBI_DEFAULT_READAHEAD_MAX);
 174         sbi->ll_ra_info.ra_max_pages = sbi->ll_ra_info.ra_max_pages_per_file;
 175         sbi->ll_ra_info.ra_max_read_ahead_whole_pages =
 176                                            SBI_DEFAULT_READAHEAD_WHOLE_MAX;
 177         sbi->ll_contention_time = SBI_DEFAULT_CONTENTION_SECONDS;
 178         sbi->ll_lockless_truncate_enable = SBI_DEFAULT_LOCKLESS_TRUNCATE_ENABLE;
 179         sbi->ll_direct_io_default = SBI_DEFAULT_DIRECT_IO_DEFAULT;
 180         INIT_LIST_HEAD(&sbi->ll_conn_chain);
 181         INIT_LIST_HEAD(&sbi->ll_orphan_dentry_list);
 182
 183         ll_generate_random_uuid(uuid);
 184         class_uuid_unparse(uuid, &sbi->ll_sb_uuid);
 185         CDEBUG(D_CONFIG, "generated uuid: %s\n", sbi->ll_sb_uuid.uuid);
 186
 187         down_write(&ll_sb_sem);
 188         list_add_tail(&sbi->ll_list, &ll_super_blocks);
 189         up_write(&ll_sb_sem);
 190
 191 #ifdef ENABLE_CHECKSUM
 192         sbi->ll_flags |= LL_SBI_DATA_CHECKSUM;
 193 #endif
 194 #ifdef ENABLE_LLITE_CHECKSUM
 195         sbi->ll_flags |= LL_SBI_LLITE_CHECKSUM;
 196 #endif
 197
 198 #ifdef HAVE_LRU_RESIZE_SUPPORT
 199         sbi->ll_flags |= LL_SBI_LRU_RESIZE;
 200 #endif
 201
 202         for (i = 0; i <= LL_PROCESS_HIST_MAX; i++) {
 203                 spin_lock_init(&sbi->ll_rw_extents_info.pp_extents[i].pp_r_hist.oh_lock);
 204                 spin_lock_init(&sbi->ll_rw_extents_info.pp_extents[i].pp_w_hist.oh_lock);
 205         }
 206
 207         /* metadata statahead is enabled by default */
 208         sbi->ll_sa_max = LL_SA_RPC_DEF;
 209
 210         RETURN(sbi);
 211
 212 out:
 213         if (sbi->ll_async_page_sample)
 214                 OBD_FREE(sbi->ll_async_page_sample,
 215                          sizeof(long) * num_possible_cpus());
 216         ll_pglist_fini(sbi);
 217         OBD_FREE(sbi, sizeof(*sbi));
 218         RETURN(NULL);
 219 }
 220
 221 void ll_free_sbi(struct super_block *sb)
 222 {
 223         struct ll_sb_info *sbi = ll_s2sbi(sb);
 224         ENTRY;
 225
 226         if (sbi != NULL) {
 227                 down_write(&ll_sb_sem);
 228                 list_del(&sbi->ll_list);
 229                 up_write(&ll_sb_sem);
 230                 /* dont allow find cache via sb list first */
 231                 ll_pglist_fini(sbi);
 232                 lcounter_destroy(&sbi->ll_async_page_count);
 233                 OBD_FREE(sbi->ll_async_page_sample,
 234                          sizeof(long) * num_possible_cpus());
 235                 OBD_FREE(sbi, sizeof(*sbi));
 236         }
 237         EXIT;
 238 }
 239
 240 static struct dentry_operations ll_d_root_ops = {
 241         .d_compare = ll_dcompare,
 242         .d_revalidate = ll_revalidate_nd,
 243 };
 244
 245 static int client_common_fill_super(struct super_block *sb,
 246                                     char *mdc, char *osc)
 247 {
 248         struct inode *root = 0;
 249         struct ll_sb_info *sbi = ll_s2sbi(sb);
 250         struct obd_device *obd;
 251         struct ll_fid rootfid;
 252         struct obd_statfs osfs;
 253         struct ptlrpc_request *request = NULL;
 254         struct lustre_handle osc_conn = {0, };
 255         struct lustre_handle mdc_conn = {0, };
 256         struct lustre_md md;
 257         struct obd_connect_data *data = NULL;
 258         int err, checksum;
 259         ENTRY;
 260
 261         obd = class_name2obd(mdc);
 262         if (!obd) {
 263                 CERROR("MDC %s: not setup or attached\n", mdc);
 264                 RETURN(-EINVAL);
 265         }
 266
 267         OBD_ALLOC(data, sizeof(*data));
 268         if (data == NULL)
 269                 RETURN(-ENOMEM);
 270
 271         if (proc_lustre_fs_root) {
 272                 err = lprocfs_register_mountpoint(proc_lustre_fs_root, sb,
 273                                                   osc, mdc);
 274                 if (err < 0)
 275                         CERROR("could not register mount in /proc/fs/lustre\n");
 276         }
 277
 278         /* indicate the features supported by this client */
 279         data->ocd_connect_flags = OBD_CONNECT_VERSION | OBD_CONNECT_IBITS      |
 280                                   OBD_CONNECT_JOIN    | OBD_CONNECT_ATTRFID    |
 281                                   OBD_CONNECT_NODEVOH | OBD_CONNECT_CANCELSET  |
 282                                   OBD_CONNECT_AT      | OBD_CONNECT_FID |
 283                                   OBD_CONNECT_VBR     | OBD_CONNECT_LOV_V3;
 284 #ifdef HAVE_LRU_RESIZE_SUPPORT
 285         if (sbi->ll_flags & LL_SBI_LRU_RESIZE)
 286                 data->ocd_connect_flags |= OBD_CONNECT_LRU_RESIZE;
 287 #endif
 288 #ifdef CONFIG_FS_POSIX_ACL
 289         data->ocd_connect_flags |= OBD_CONNECT_ACL;
 290 #endif
 291         data->ocd_ibits_known = MDS_INODELOCK_FULL;
 292         data->ocd_version = LUSTRE_VERSION_CODE;
 293
 294         if (sb->s_flags & MS_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 HAVE_MS_FLOCK_LOCK
 300         /* force vfs to use lustre handler for flock() calls - bug 10743 */
 301         sb->s_flags |= MS_FLOCK_LOCK;
 302 #endif
 303
 304         if (sbi->ll_flags & LL_SBI_FLOCK)
 305                 sbi->ll_fop = &ll_file_operations_flock;
 306         else if (sbi->ll_flags & LL_SBI_LOCALFLOCK)
 307                 sbi->ll_fop = &ll_file_operations;
 308         else
 309                 sbi->ll_fop = &ll_file_operations_noflock;
 310
 311
 312         err = obd_connect(&mdc_conn, obd, &sbi->ll_sb_uuid, data, &sbi->ll_mdc_exp);
 313         if (err == -EBUSY) {
 314                 LCONSOLE_ERROR_MSG(0x14f, "An MDT (mdc %s) is performing "
 315                                    "recovery, of which this client is not a "
 316                                    "part. Please wait for recovery to complete,"
 317                                    " abort, or time out.\n", mdc);
 318                 GOTO(out, err);
 319         } else if (err) {
 320                 CERROR("cannot connect to %s: rc = %d\n", mdc, err);
 321                 GOTO(out, err);
 322         }
 323
 324         err = obd_fid_init(sbi->ll_mdc_exp);
 325         if (err)
 326                 GOTO(out_mdc, err);
 327
 328         err = obd_statfs(obd, &osfs,
 329                          cfs_time_shift_64(-OBD_STATFS_CACHE_SECONDS), 0);
 330         if (err)
 331                 GOTO(out_mdc, err);
 332
 333         /* MDC connect is surely finished by now because we actually sent
 334          * a statfs RPC, otherwise obd_connect() is asynchronous. */
 335         *data = class_exp2cliimp(sbi->ll_mdc_exp)->imp_connect_data;
 336
 337         LASSERT(osfs.os_bsize);
 338         sb->s_blocksize = osfs.os_bsize;
 339         sb->s_blocksize_bits = log2(osfs.os_bsize);
 340         sb->s_magic = LL_SUPER_MAGIC;
 341
 342         /* for bug 11559. in $LINUX/fs/read_write.c, function do_sendfile():
 343          *         retval = in_file->f_op->sendfile(...);
 344          *         if (*ppos > max)
 345          *                 retval = -EOVERFLOW;
 346          *
 347          * it will check if *ppos is greater than max. However, max equals to
 348          * s_maxbytes, which is a negative integer in a x86_64 box since loff_t
 349          * has been defined as a signed long long ineger in linux kernel. */
 350 #if BITS_PER_LONG == 64
 351         sb->s_maxbytes = PAGE_CACHE_MAXBYTES >> 1;
 352 #else
 353         sb->s_maxbytes = PAGE_CACHE_MAXBYTES;
 354 #endif
 355         sbi->ll_namelen = osfs.os_namelen;
 356         sbi->ll_max_rw_chunk = LL_DEFAULT_MAX_RW_CHUNK;
 357
 358         if ((sbi->ll_flags & LL_SBI_USER_XATTR) &&
 359             !(data->ocd_connect_flags & OBD_CONNECT_XATTR)) {
 360                 LCONSOLE_INFO("Disabling user_xattr feature because "
 361                               "it is not supported on the server\n");
 362                 sbi->ll_flags &= ~LL_SBI_USER_XATTR;
 363         }
 364
 365         if (data->ocd_connect_flags & OBD_CONNECT_ACL) {
 366 #ifdef MS_POSIXACL
 367                 sb->s_flags |= MS_POSIXACL;
 368 #endif
 369                 sbi->ll_flags |= LL_SBI_ACL;
 370         } else
 371                 sbi->ll_flags &= ~LL_SBI_ACL;
 372
 373         if (data->ocd_connect_flags & OBD_CONNECT_JOIN)
 374                 sbi->ll_flags |= LL_SBI_JOIN;
 375
 376         obd = class_name2obd(osc);
 377         if (!obd) {
 378                 CERROR("OSC %s: not setup or attached\n", osc);
 379                 GOTO(out_mdc, err = -ENODEV);
 380         }
 381
 382         data->ocd_connect_flags = OBD_CONNECT_VERSION   | OBD_CONNECT_GRANT    |
 383                                   OBD_CONNECT_REQPORTAL | OBD_CONNECT_BRW_SIZE |
 384                                   OBD_CONNECT_SRVLOCK   | OBD_CONNECT_CANCELSET|
 385                                   OBD_CONNECT_AT        | OBD_CONNECT_FID      |
 386                                   OBD_CONNECT_VBR       | OBD_CONNECT_TRUNCLOCK;
 387
 388         if (!OBD_FAIL_CHECK(OBD_FAIL_OSC_CONNECT_CKSUM)) {
 389                 /* OBD_CONNECT_CKSUM should always be set, even if checksums are
 390                  * disabled by default, because it can still be enabled on the
 391                  * fly via /proc. As a consequence, we still need to come to an
 392                  * agreement on the supported algorithms at connect time */
 393                 data->ocd_connect_flags |= OBD_CONNECT_CKSUM;
 394
 395                 if (OBD_FAIL_CHECK(OBD_FAIL_OSC_CKSUM_ADLER_ONLY))
 396                         data->ocd_cksum_types = OBD_CKSUM_ADLER;
 397                 else
 398                         /* send the list of supported checksum types */
 399                         data->ocd_cksum_types = OBD_CKSUM_ALL;
 400         }
 401
 402 #ifdef HAVE_LRU_RESIZE_SUPPORT
 403         if (sbi->ll_flags & LL_SBI_LRU_RESIZE)
 404                 data->ocd_connect_flags |= OBD_CONNECT_LRU_RESIZE;
 405 #endif
 406
 407         CDEBUG(D_RPCTRACE, "ocd_connect_flags: "LPX64" ocd_version: %d "
 408                "ocd_grant: %d\n", data->ocd_connect_flags,
 409                data->ocd_version, data->ocd_grant);
 410
 411         obd->obd_upcall.onu_owner = &sbi->ll_lco;
 412         obd->obd_upcall.onu_upcall = ll_ocd_update;
 413         data->ocd_brw_size = PTLRPC_MAX_BRW_PAGES << CFS_PAGE_SHIFT;
 414
 415         /* ask lov to generate OBD_NOTIFY_CREATE events for already registered
 416          * targets */
 417         obd_notify(obd, NULL, OBD_NOTIFY_CREATE, NULL);
 418
 419         obd_register_lock_cancel_cb(obd, ll_extent_lock_cancel_cb);
 420         obd_register_page_removal_cb(obd, ll_page_removal_cb, ll_pin_extent_cb);
 421
 422         err = obd_connect(&osc_conn, obd, &sbi->ll_sb_uuid, data, &sbi->ll_osc_exp);
 423         if (err == -EBUSY) {
 424                 LCONSOLE_ERROR_MSG(0x150, "An OST (osc %s) is performing "
 425                                    "recovery, of which this client is not a "
 426                                    "part.  Please wait for recovery to "
 427                                    "complete, abort, or time out.\n", osc);
 428                 GOTO(out, err);
 429         } else if (err) {
 430                 CERROR("cannot connect to %s: rc = %d\n", osc, err);
 431                 GOTO(out_cb, err);
 432         }
 433
 434         mutex_down(&sbi->ll_lco.lco_lock);
 435         sbi->ll_lco.lco_flags = data->ocd_connect_flags;
 436         sbi->ll_lco.lco_mdc_exp = sbi->ll_mdc_exp;
 437         sbi->ll_lco.lco_osc_exp = sbi->ll_osc_exp;
 438         mutex_up(&sbi->ll_lco.lco_lock);
 439
 440         err = mdc_init_ea_size(sbi->ll_mdc_exp, sbi->ll_osc_exp);
 441         if (err) {
 442                 CERROR("cannot set max EA and cookie sizes: rc = %d\n", err);
 443                 GOTO(out_osc, err);
 444         }
 445
 446         err = obd_prep_async_page(sbi->ll_osc_exp, NULL, NULL, NULL,
 447                                   0, NULL, NULL, NULL, 0, NULL);
 448         if (err < 0) {
 449                 LCONSOLE_ERROR_MSG(0x151, "There are no OST's in this "
 450                                    "filesystem. There must be at least one "
 451                                    "active OST for a client to start.\n");
 452                 GOTO(out_osc, err);
 453         }
 454
 455         if (!ll_async_page_slab) {
 456                 ll_async_page_slab_size =
 457                         size_round(sizeof(struct ll_async_page)) + err;
 458                 ll_async_page_slab = cfs_mem_cache_create("ll_async_page",
 459                                                           ll_async_page_slab_size,
 460                                                           0, 0);
 461                 if (!ll_async_page_slab)
 462                         GOTO(out_osc, err = -ENOMEM);
 463         }
 464
 465         err = mdc_getstatus(sbi->ll_mdc_exp, &rootfid);
 466         if (err) {
 467                 CERROR("cannot mds_connect: rc = %d\n", err);
 468                 GOTO(out_osc, err);
 469         }
 470         CDEBUG(D_SUPER, "rootfid "LPU64":"DFID"\n", rootfid.id,
 471                         PFID((struct lu_fid*)&rootfid));
 472         sbi->ll_rootino = rootfid.id;
 473
 474         sb->s_op = &lustre_super_operations;
 475 #if THREAD_SIZE >= 8192
 476         /* Disable the NFS export because of stack overflow
 477          * when THREAD_SIZE < 8192. Please refer to 17630. */
 478         sb->s_export_op = &lustre_export_operations;
 479 #endif
 480
 481         /* make root inode
 482          * XXX: move this to after cbd setup? */
 483         err = mdc_getattr(sbi->ll_mdc_exp, &rootfid,
 484                           OBD_MD_FLGETATTR | OBD_MD_FLBLOCKS |
 485                           (sbi->ll_flags & LL_SBI_ACL ? OBD_MD_FLACL : 0),
 486                           0, &request);
 487         if (err) {
 488                 CERROR("mdc_getattr failed for root: rc = %d\n", err);
 489                 GOTO(out_osc, err);
 490         }
 491
 492         err = mdc_req2lustre_md(request, REPLY_REC_OFF, sbi->ll_osc_exp, &md);
 493         if (err) {
 494                 CERROR("failed to understand root inode md: rc = %d\n",err);
 495                 ptlrpc_req_finished (request);
 496                 GOTO(out_osc, err);
 497         }
 498
 499         LASSERT(sbi->ll_rootino != 0);
 500         root = ll_iget(sb, ll_fid_build_ino(sbi, &rootfid), &md);
 501
 502         ptlrpc_req_finished(request);
 503
 504         if (root == NULL || is_bad_inode(root)) {
 505                 mdc_free_lustre_md(sbi->ll_osc_exp, &md);
 506                 CERROR("lustre_lite: bad iget4 for root\n");
 507                 GOTO(out_root, err = -EBADF);
 508         }
 509
 510         err = ll_close_thread_start(&sbi->ll_lcq);
 511         if (err) {
 512                 CERROR("cannot start close thread: rc %d\n", err);
 513                 GOTO(out_root, err);
 514         }
 515
 516         checksum = sbi->ll_flags & LL_SBI_DATA_CHECKSUM;
 517         err = obd_set_info_async(sbi->ll_osc_exp, sizeof(KEY_CHECKSUM),
 518                                  KEY_CHECKSUM, sizeof(checksum),
 519                                  &checksum, NULL);
 520
 521         /* making vm readahead 0 for 2.4.x. In the case of 2.6.x,
 522            backing dev info assigned to inode mapping is used for
 523            determining maximal readahead. */
 524 #if (LINUX_VERSION_CODE < KERNEL_VERSION(2,6,0)) && \
 525     !defined(KERNEL_HAS_AS_MAX_READAHEAD)
 526         /* bug 2805 - set VM readahead to zero */
 527         vm_max_readahead = vm_min_readahead = 0;
 528 #endif
 529
 530         sb->s_root = d_alloc_root(root);
 531         if (data != NULL)
 532                 OBD_FREE(data, sizeof(*data));
 533         sb->s_root->d_op = &ll_d_root_ops;
 534
 535         sbi->ll_sdev_orig = sb->s_dev;
 536         /* We set sb->s_dev equal on all lustre clients in order to support
 537          * NFS export clustering.  NFSD requires that the FSID be the same
 538          * on all clients. */
 539         /* s_dev is also used in lt_compare() to compare two fs, but that is
 540          * only a node-local comparison. */
 541         sb->s_dev = get_uuid2int(sbi2mdc(sbi)->cl_target_uuid.uuid,
 542                                  strlen(sbi2mdc(sbi)->cl_target_uuid.uuid));
 543
 544         RETURN(err);
 545
 546 out_root:
 547         if (root)
 548                 iput(root);
 549 out_osc:
 550         obd_disconnect(sbi->ll_osc_exp);
 551         sbi->ll_osc_exp = NULL;
 552 out_cb:
 553         obd = class_name2obd(osc);
 554         obd_unregister_lock_cancel_cb(obd, ll_extent_lock_cancel_cb);
 555         obd_unregister_page_removal_cb(obd, ll_page_removal_cb);
 556 out_mdc:
 557         obd_fid_fini(sbi->ll_mdc_exp);
 558         obd_disconnect(sbi->ll_mdc_exp);
 559         sbi->ll_mdc_exp = NULL;
 560 out:
 561         if (data != NULL)
 562                 OBD_FREE(data, sizeof(*data));
 563         lprocfs_unregister_mountpoint(sbi);
 564         RETURN(err);
 565 }
 566
 567 int ll_get_max_mdsize(struct ll_sb_info *sbi, int *lmmsize)
 568 {
 569         int size, rc;
 570
 571         *lmmsize = obd_size_diskmd(sbi->ll_osc_exp, NULL);
 572         size = sizeof(int);
 573         rc = obd_get_info(sbi->ll_mdc_exp, sizeof(KEY_MAX_EASIZE),
 574                           KEY_MAX_EASIZE, &size, lmmsize, NULL);
 575         if (rc)
 576                 CERROR("Get max mdsize error rc %d \n", rc);
 577
 578         RETURN(rc);
 579 }
 580
 581 void ll_dump_inode(struct inode *inode)
 582 {
 583         struct list_head *tmp;
 584         int dentry_count = 0;
 585
 586         LASSERT(inode != NULL);
 587
 588         list_for_each(tmp, &inode->i_dentry)
 589                 dentry_count++;
 590
 591         CERROR("inode %p dump: dev=%s ino=%lu mode=%o count=%u, %d dentries\n",
 592                inode, ll_i2mdcexp(inode)->exp_obd->obd_name, inode->i_ino,
 593                inode->i_mode, atomic_read(&inode->i_count), dentry_count);
 594 }
 595
 596 void lustre_dump_dentry(struct dentry *dentry, int recur)
 597 {
 598         struct list_head *tmp;
 599         int subdirs = 0;
 600
 601         LASSERT(dentry != NULL);
 602
 603         list_for_each(tmp, &dentry->d_subdirs)
 604                 subdirs++;
 605
 606         CERROR("dentry %p dump: name=%.*s parent=%.*s (%p), inode=%p, count=%u,"
 607                " flags=0x%x, fsdata=%p, %d subdirs\n", dentry,
 608                dentry->d_name.len, dentry->d_name.name,
 609                dentry->d_parent->d_name.len, dentry->d_parent->d_name.name,
 610                dentry->d_parent, dentry->d_inode, atomic_read(&dentry->d_count),
 611                dentry->d_flags, dentry->d_fsdata, subdirs);
 612         if (dentry->d_inode != NULL)
 613                 ll_dump_inode(dentry->d_inode);
 614
 615         if (recur == 0)
 616                 return;
 617
 618         list_for_each(tmp, &dentry->d_subdirs) {
 619                 struct dentry *d = list_entry(tmp, struct dentry, d_child);
 620                 lustre_dump_dentry(d, recur - 1);
 621         }
 622 }
 623
 624 void client_common_put_super(struct super_block *sb)
 625 {
 626         struct ll_sb_info *sbi = ll_s2sbi(sb);
 627         ENTRY;
 628
 629         ll_close_thread_shutdown(sbi->ll_lcq);
 630
 631         lprocfs_unregister_mountpoint(sbi);
 632
 633         list_del(&sbi->ll_conn_chain);
 634
 635         /* callbacks is cleared after disconnect each target */
 636         obd_disconnect(sbi->ll_osc_exp);
 637         sbi->ll_osc_exp = NULL;
 638
 639         obd_fid_fini(sbi->ll_mdc_exp);
 640         obd_disconnect(sbi->ll_mdc_exp);
 641         sbi->ll_mdc_exp = NULL;
 642
 643         EXIT;
 644 }
 645
 646 void ll_kill_super(struct super_block *sb)
 647 {
 648         struct ll_sb_info *sbi;
 649
 650         ENTRY;
 651
 652         /* not init sb ?*/
 653         if (!(sb->s_flags & MS_ACTIVE))
 654                 return;
 655
 656         sbi = ll_s2sbi(sb);
 657         /* we need restore s_dev from changed for clustred NFS before put_super
 658          * because new kernels have cached s_dev and change sb->s_dev in
 659          * put_super not affected real removing devices */
 660         if (sbi)
 661                 sb->s_dev = sbi->ll_sdev_orig;
 662         EXIT;
 663 }
 664
 665 char *ll_read_opt(const char *opt, char *data)
 666 {
 667         char *value;
 668         char *retval;
 669         ENTRY;
 670
 671         CDEBUG(D_SUPER, "option: %s, data %s\n", opt, data);
 672         if (strncmp(opt, data, strlen(opt)))
 673                 RETURN(NULL);
 674         if ((value = strchr(data, '=')) == NULL)
 675                 RETURN(NULL);
 676
 677         value++;
 678         OBD_ALLOC(retval, strlen(value) + 1);
 679         if (!retval) {
 680                 CERROR("out of memory!\n");
 681                 RETURN(NULL);
 682         }
 683
 684         memcpy(retval, value, strlen(value)+1);
 685         CDEBUG(D_SUPER, "Assigned option: %s, value %s\n", opt, retval);
 686         RETURN(retval);
 687 }
 688
 689 static inline int ll_set_opt(const char *opt, char *data, int fl)
 690 {
 691         if (strncmp(opt, data, strlen(opt)) != 0)
 692                 return(0);
 693         else
 694                 return(fl);
 695 }
 696
 697 /* non-client-specific mount options are parsed in lmd_parse */
 698 static int ll_options(char *options, int *flags)
 699 {
 700         int tmp;
 701         char *s1 = options, *s2;
 702         ENTRY;
 703
 704         if (!options)
 705                 RETURN(0);
 706
 707         CDEBUG(D_CONFIG, "Parsing opts %s\n", options);
 708
 709         while (*s1) {
 710                 CDEBUG(D_SUPER, "next opt=%s\n", s1);
 711                 tmp = ll_set_opt("nolock", s1, LL_SBI_NOLCK);
 712                 if (tmp) {
 713                         *flags |= tmp;
 714                         goto next;
 715                 }
 716                 tmp = ll_set_opt("flock", s1, LL_SBI_FLOCK);
 717                 if (tmp) {
 718                         *flags |= tmp;
 719                         goto next;
 720                 }
 721                 tmp = ll_set_opt("localflock", s1, LL_SBI_LOCALFLOCK);
 722                 if (tmp) {
 723                         *flags |= tmp;
 724                         goto next;
 725                 }
 726                 tmp = ll_set_opt("noflock", s1, LL_SBI_FLOCK|LL_SBI_LOCALFLOCK);
 727                 if (tmp) {
 728                         *flags &= ~tmp;
 729                         goto next;
 730                 }
 731                 tmp = ll_set_opt("user_xattr", s1, LL_SBI_USER_XATTR);
 732                 if (tmp) {
 733                         *flags |= tmp;
 734                         goto next;
 735                 }
 736                 tmp = ll_set_opt("nouser_xattr", s1, LL_SBI_USER_XATTR);
 737                 if (tmp) {
 738                         *flags &= ~tmp;
 739                         goto next;
 740                 }
 741                 tmp = ll_set_opt("acl", s1, LL_SBI_ACL);
 742                 if (tmp) {
 743                         /* Ignore deprecated mount option.  The client will
 744                          * always try to mount with ACL support, whether this
 745                          * is used depends on whether server supports it. */
 746                         LCONSOLE_ERROR_MSG(0x152, "Ignoring deprecated "
 747                                                   "mount option 'acl'.\n");
 748                         goto next;
 749                 }
 750                 tmp = ll_set_opt("noacl", s1, LL_SBI_ACL);
 751                 if (tmp) {
 752                         LCONSOLE_ERROR_MSG(0x152, "Ignoring deprecated "
 753                                                   "mount option 'noacl'.\n");
 754                         goto next;
 755                 }
 756
 757                 tmp = ll_set_opt("checksum", s1, LL_SBI_DATA_CHECKSUM);
 758                 if (tmp) {
 759                         *flags |= tmp;
 760                         goto next;
 761                 }
 762                 tmp = ll_set_opt("nochecksum", s1, LL_SBI_DATA_CHECKSUM);
 763                 if (tmp) {
 764                         *flags &= ~tmp;
 765                         goto next;
 766                 }
 767
 768                 tmp = ll_set_opt("lruresize", s1, LL_SBI_LRU_RESIZE);
 769                 if (tmp) {
 770                         *flags |= tmp;
 771                         goto next;
 772                 }
 773                 tmp = ll_set_opt("nolruresize", s1, LL_SBI_LRU_RESIZE);
 774                 if (tmp) {
 775                         *flags &= ~tmp;
 776                         goto next;
 777                 }
 778                 tmp = ll_set_opt("lazystatfs", s1, LL_SBI_LAZYSTATFS);
 779                 if (tmp) {
 780                         *flags |= tmp;
 781                         goto next;
 782                 }
 783                 tmp = ll_set_opt("nolazystatfs", s1, LL_SBI_LAZYSTATFS);
 784                 if (tmp) {
 785                         *flags &= ~tmp;
 786                         goto next;
 787                 }
 788                 LCONSOLE_ERROR_MSG(0x152, "Unknown option '%s', won't mount.\n",
 789                                    s1);
 790                 RETURN(-EINVAL);
 791
 792 next:
 793                 /* Find next opt */
 794                 s2 = strchr(s1, ',');
 795                 if (s2 == NULL)
 796                         break;
 797                 s1 = s2 + 1;
 798         }
 799         RETURN(0);
 800 }
 801
 802 void ll_lli_init(struct ll_inode_info *lli)
 803 {
 804         lli->lli_inode_magic = LLI_INODE_MAGIC;
 805         sema_init(&lli->lli_size_sem, 1);
 806         sema_init(&lli->lli_write_sem, 1);
 807         lli->lli_flags = 0;
 808         lli->lli_maxbytes = PAGE_CACHE_MAXBYTES;
 809         spin_lock_init(&lli->lli_lock);
 810         sema_init(&lli->lli_och_sem, 1);
 811         lli->lli_mds_read_och = lli->lli_mds_write_och = NULL;
 812         lli->lli_mds_exec_och = NULL;
 813         lli->lli_open_fd_read_count = lli->lli_open_fd_write_count = 0;
 814         lli->lli_open_fd_exec_count = 0;
 815         INIT_LIST_HEAD(&lli->lli_dead_list);
 816 #ifdef HAVE_CLOSE_THREAD
 817         INIT_LIST_HEAD(&lli->lli_pending_write_llaps);
 818 #endif
 819         init_rwsem(&lli->lli_truncate_rwsem);
 820 }
 821
 822 /* COMPAT_146 */
 823 #define MDCDEV "mdc_dev"
 824 static int old_lustre_process_log(struct super_block *sb, char *newprofile,
 825                                   struct config_llog_instance *cfg)
 826 {
 827         struct lustre_sb_info *lsi = s2lsi(sb);
 828         struct obd_device *obd;
 829         struct lustre_handle mdc_conn = {0, };
 830         struct obd_export *exp;
 831         char *ptr, *mdt, *profile;
 832         char niduuid[10] = "mdtnid0";
 833         class_uuid_t uuid;
 834         struct obd_uuid mdc_uuid;
 835         struct llog_ctxt *ctxt;
 836         struct obd_connect_data ocd = { 0 };
 837         lnet_nid_t nid;
 838         int i, rc = 0, recov_bk = 1, failnodes = 0;
 839         ENTRY;
 840
 841         ll_generate_random_uuid(uuid);
 842         class_uuid_unparse(uuid, &mdc_uuid);
 843         CDEBUG(D_HA, "generated uuid: %s\n", mdc_uuid.uuid);
 844
 845         /* Figure out the old mdt and profile name from new-style profile
 846            ("lustre" from "mds/lustre-client") */
 847         mdt = newprofile;
 848         profile = strchr(mdt, '/');
 849         if (profile == NULL) {
 850                 CDEBUG(D_CONFIG, "Can't find MDT name in %s\n", newprofile);
 851                 GOTO(out, rc = -EINVAL);
 852         }
 853         *profile = '\0';
 854         profile++;
 855         ptr = strrchr(profile, '-');
 856         if (ptr == NULL) {
 857                 CDEBUG(D_CONFIG, "Can't find client name in %s\n", newprofile);
 858                 GOTO(out, rc = -EINVAL);
 859         }
 860         *ptr = '\0';
 861
 862         LCONSOLE_WARN("This looks like an old mount command; I will try to "
 863                       "contact MDT '%s' for profile '%s'\n", mdt, profile);
 864
 865         /* Use nids from mount line: uml1,1@elan:uml2,2@elan:/lustre */
 866         i = 0;
 867         ptr = lsi->lsi_lmd->lmd_dev;
 868         while (class_parse_nid(ptr, &nid, &ptr) == 0) {
 869                 rc = do_lcfg(MDCDEV, nid, LCFG_ADD_UUID, niduuid, 0,0,0);
 870                 i++;
 871                 /* Stop at the first failover nid */
 872                 if (*ptr == ':')
 873                         break;
 874         }
 875         if (i == 0) {
 876                 CERROR("No valid MDT nids found.\n");
 877                 GOTO(out, rc = -EINVAL);
 878         }
 879         failnodes++;
 880
 881         rc = do_lcfg(MDCDEV, 0, LCFG_ATTACH, LUSTRE_MDC_NAME,mdc_uuid.uuid,0,0);
 882         if (rc < 0)
 883                 GOTO(out_del_uuid, rc);
 884
 885         rc = do_lcfg(MDCDEV, 0, LCFG_SETUP, mdt, niduuid, 0, 0);
 886         if (rc < 0) {
 887                 LCONSOLE_ERROR_MSG(0x153, "I couldn't establish a connection "
 888                                    "with the MDT. Check that the MDT host NID "
 889                                    "is correct and the networks are up.\n");
 890                 GOTO(out_detach, rc);
 891         }
 892
 893         obd = class_name2obd(MDCDEV);
 894         if (obd == NULL)
 895                 GOTO(out_cleanup, rc = -EINVAL);
 896
 897         /* Add any failover nids */
 898         while (*ptr == ':') {
 899                 /* New failover node */
 900                 sprintf(niduuid, "mdtnid%d", failnodes);
 901                 i = 0;
 902                 while (class_parse_nid(ptr, &nid, &ptr) == 0) {
 903                         i++;
 904                         rc = do_lcfg(MDCDEV, nid, LCFG_ADD_UUID, niduuid,0,0,0);
 905                         if (rc)
 906                                 CERROR("Add uuid for %s failed %d\n",
 907                                        libcfs_nid2str(nid), rc);
 908                         if (*ptr == ':')
 909                                 break;
 910                 }
 911                 if (i > 0) {
 912                         rc = do_lcfg(MDCDEV, 0, LCFG_ADD_CONN, niduuid, 0, 0,0);
 913                         if (rc)
 914                                 CERROR("Add conn for %s failed %d\n",
 915                                        libcfs_nid2str(nid), rc);
 916                         failnodes++;
 917                 } else {
 918                         /* at ":/fsname" */
 919                         break;
 920                 }
 921         }
 922
 923         /* Try all connections, but only once. */
 924         rc = obd_set_info_async(obd->obd_self_export,
 925                                 sizeof(KEY_INIT_RECOV_BACKUP), KEY_INIT_RECOV_BACKUP,
 926                                 sizeof(recov_bk), &recov_bk, NULL);
 927         if (rc)
 928                 GOTO(out_cleanup, rc);
 929
 930         /* If we don't have this then an ACL MDS will refuse the connection */
 931         ocd.ocd_connect_flags = OBD_CONNECT_ACL;
 932
 933         rc = obd_connect(&mdc_conn, obd, &mdc_uuid, &ocd, &exp);
 934         if (rc) {
 935                 CERROR("cannot connect to %s: rc = %d\n", mdt, rc);
 936                 GOTO(out_cleanup, rc);
 937         }
 938
 939         ctxt = llog_get_context(exp->exp_obd, LLOG_CONFIG_REPL_CTXT);
 940
 941         cfg->cfg_flags |= CFG_F_COMPAT146;
 942
 943 #if 1
 944         rc = class_config_parse_llog(ctxt, profile, cfg);
 945 #else
 946         /*
 947          * For debugging, it's useful to just dump the log
 948          */
 949         rc = class_config_dump_llog(ctxt, profile, cfg);
 950 #endif
 951         llog_ctxt_put(ctxt);
 952         switch (rc) {
 953         case 0: {
 954                 /* Set the caller's profile name to the old-style */
 955                 memcpy(newprofile, profile, strlen(profile) + 1);
 956                 break;
 957         }
 958         case -EINVAL:
 959                 LCONSOLE_ERROR_MSG(0x154, "%s: The configuration '%s' could not"
 960                                    " be read from the MDT '%s'.  Make sure this"
 961                                    " client and the MDT are running compatible "
 962                                    "versions of Lustre.\n",
 963                                    obd->obd_name, profile, mdt);
 964                 /* fall through */
 965         default:
 966                 LCONSOLE_ERROR_MSG(0x155, "%s: The configuration '%s' could not"
 967                                    " be read from the MDT '%s'.  This may be "
 968                                    "the result of communication errors between "
 969                                    "the client and the MDT, or if the MDT is "
 970                                    "not running.\n", obd->obd_name, profile,
 971                                    mdt);
 972                 break;
 973         }
 974
 975         /* We don't so much care about errors in cleaning up the config llog
 976          * connection, as we have already read the config by this point. */
 977         obd_disconnect(exp);
 978
 979 out_cleanup:
 980         do_lcfg(MDCDEV, 0, LCFG_CLEANUP, 0, 0, 0, 0);
 981
 982 out_detach:
 983         do_lcfg(MDCDEV, 0, LCFG_DETACH, 0, 0, 0, 0);
 984
 985 out_del_uuid:
 986         /* class_add_uuid adds a nid even if the same uuid exists; we might
 987            delete any copy here.  So they all better match. */
 988         for (i = 0; i < failnodes; i++) {
 989                 sprintf(niduuid, "mdtnid%d", i);
 990                 do_lcfg(MDCDEV, 0, LCFG_DEL_UUID, niduuid, 0, 0, 0);
 991         }
 992         /* class_import_put will get rid of the additional connections */
 993 out:
 994         RETURN(rc);
 995 }
 996 /* end COMPAT_146 */
 997
 998 int ll_fill_super(struct super_block *sb)
 999 {
1000         struct lustre_profile *lprof = NULL;
1001         struct lustre_sb_info *lsi = s2lsi(sb);
1002         struct ll_sb_info *sbi;
1003         char  *osc = NULL, *mdc = NULL;
1004         char  *profilenm = get_profile_name(sb);
1005         struct config_llog_instance cfg = {0, };
1006         char   ll_instance[sizeof(sb) * 2 + 1];
1007         int    err;
1008         char  *save = NULL;
1009         char  pseudo[32] = { 0 };
1010         ENTRY;
1011
1012         CDEBUG(D_VFSTRACE, "VFS Op: sb %p\n", sb);
1013
1014         cfs_module_get();
1015
1016         /* client additional sb info */
1017         lsi->lsi_llsbi = sbi = ll_init_sbi();
1018         if (!sbi) {
1019                 cfs_module_put();
1020                 RETURN(-ENOMEM);
1021         }
1022
1023         err = ll_options(lsi->lsi_lmd->lmd_opts, &sbi->ll_flags);
1024         if (err)
1025                 GOTO(out_free, err);
1026
1027         /* Generate a string unique to this super, in case some joker tries
1028            to mount the same fs at two mount points.
1029            Use the address of the super itself.*/
1030         sprintf(ll_instance, "%p", sb);
1031         cfg.cfg_instance = ll_instance;
1032         cfg.cfg_uuid = lsi->lsi_llsbi->ll_sb_uuid;
1033         cfg.cfg_sb = sb;
1034
1035         /* set up client obds */
1036         if (strchr(profilenm, '/') != NULL) /* COMPAT_146 */
1037                 err = -EINVAL; /* skip error messages, use old config code */
1038         else
1039                 err = lustre_process_log(sb, profilenm, &cfg);
1040         /* COMPAT_146 */
1041         if (err < 0) {
1042                 char *oldname;
1043                 int rc, oldnamelen;
1044                 oldnamelen = strlen(profilenm) + 1;
1045                 /* Temp storage for 1.4.6 profile name */
1046                 OBD_ALLOC(oldname, oldnamelen);
1047                 if (oldname) {
1048                         memcpy(oldname, profilenm, oldnamelen);
1049                         rc = old_lustre_process_log(sb, oldname, &cfg);
1050                         if (rc >= 0) {
1051                                 /* That worked - update the profile name
1052                                    permanently */
1053                                 err = rc;
1054                                 OBD_FREE(lsi->lsi_lmd->lmd_profile,
1055                                          strlen(lsi->lsi_lmd->lmd_profile) + 1);
1056                                 OBD_ALLOC(lsi->lsi_lmd->lmd_profile,
1057                                          strlen(oldname) + 1);
1058                                 if (!lsi->lsi_lmd->lmd_profile) {
1059                                         OBD_FREE(oldname, oldnamelen);
1060                                         GOTO(out_free, err = -ENOMEM);
1061                                 }
1062                                 memcpy(lsi->lsi_lmd->lmd_profile, oldname,
1063                                        strlen(oldname) + 1);
1064                                 profilenm = get_profile_name(sb);
1065                                 /* Don't ever try to recover the MGS */
1066                                 rc = ptlrpc_set_import_active(
1067                                         lsi->lsi_mgc->u.cli.cl_import, 0);
1068                         }
1069                         OBD_FREE(oldname, oldnamelen);
1070                 }
1071         }
1072         /* end COMPAT_146 */
1073         if (err < 0) {
1074                 CERROR("Unable to process log: %d\n", err);
1075                 GOTO(out_free, err);
1076         }
1077
1078         lprof = class_get_profile(profilenm);
1079         if (lprof == NULL) {
1080                 LCONSOLE_ERROR_MSG(0x156, "The client profile '%s' could not be"
1081                                    " read from the MGS.  Does that filesystem "
1082                                    "exist?\n", profilenm);
1083                 GOTO(out_free, err = -EINVAL);
1084         }
1085
1086         /*
1087          * The configuration for 1.8 client and 2.0 client are different.
1088          * 2.0 introduces lmv, but 1.8 directly uses mdc.
1089          * Here, we will hack to use proper name for mdc if needed.
1090          */
1091         {
1092                 char *fsname_end;
1093                 int namelen;
1094
1095                 save = lprof->lp_mdc;
1096                 fsname_end = strrchr(save, '-');
1097                 if (fsname_end) {
1098                         namelen = fsname_end - save;
1099                         if (strcmp(fsname_end, "-clilmv") == 0) {
1100                                 strncpy(pseudo, save, namelen);
1101                                 strcat(pseudo, "-MDT0000-mdc");
1102                                 lprof->lp_mdc = pseudo;
1103                                 CDEBUG(D_INFO, "1.8.x connecting to 2.0: lmv=%s"
1104                                        " new mdc=%s\n", save, pseudo);
1105                         }
1106                 }
1107         }
1108
1109         CDEBUG(D_CONFIG, "Found profile %s: mdc=%s osc=%s\n", profilenm,
1110                lprof->lp_mdc, lprof->lp_osc);
1111
1112         OBD_ALLOC(osc, strlen(lprof->lp_osc) +
1113                   strlen(ll_instance) + 2);
1114         if (!osc)
1115                 GOTO(out_free, err = -ENOMEM);
1116         sprintf(osc, "%s-%s", lprof->lp_osc, ll_instance);
1117
1118         OBD_ALLOC(mdc, strlen(lprof->lp_mdc) +
1119                   strlen(ll_instance) + 2);
1120         if (!mdc)
1121                 GOTO(out_free, err = -ENOMEM);
1122         sprintf(mdc, "%s-%s", lprof->lp_mdc, ll_instance);
1123
1124         /* connections, registrations, sb setup */
1125         err = client_common_fill_super(sb, mdc, osc);
1126
1127 out_free:
1128         if (save && lprof)
1129                 lprof->lp_mdc = save;
1130         if (mdc)
1131                 OBD_FREE(mdc, strlen(mdc) + 1);
1132         if (osc)
1133                 OBD_FREE(osc, strlen(osc) + 1);
1134         if (err)
1135                 ll_put_super(sb);
1136         else
1137                 LCONSOLE_WARN("Client %s has started\n", profilenm);
1138
1139         RETURN(err);
1140 } /* ll_fill_super */
1141
1142
1143 void ll_put_super(struct super_block *sb)
1144 {
1145         struct config_llog_instance cfg;
1146         char   ll_instance[sizeof(sb) * 2 + 1];
1147         struct obd_device *obd;
1148         struct lustre_sb_info *lsi = s2lsi(sb);
1149         struct ll_sb_info *sbi = ll_s2sbi(sb);
1150         char *profilenm = get_profile_name(sb);
1151         int force = 1, next;
1152         ENTRY;
1153
1154         CDEBUG(D_VFSTRACE, "VFS Op: sb %p - %s\n", sb, profilenm);
1155
1156         sprintf(ll_instance, "%p", sb);
1157         cfg.cfg_instance = ll_instance;
1158         lustre_end_log(sb, NULL, &cfg);
1159
1160         if (sbi->ll_mdc_exp) {
1161                 obd = class_exp2obd(sbi->ll_mdc_exp);
1162                 if (obd)
1163                         force = obd->obd_force;
1164         }
1165
1166         /* We need to set force before the lov_disconnect in
1167            lustre_common_put_super, since l_d cleans up osc's as well. */
1168         if (force) {
1169                 next = 0;
1170                 while ((obd = class_devices_in_group(&sbi->ll_sb_uuid,
1171                                                      &next)) != NULL) {
1172                         obd->obd_force = force;
1173                 }
1174         }
1175
1176         if (sbi->ll_lcq) {
1177                 /* Only if client_common_fill_super succeeded */
1178                 client_common_put_super(sb);
1179         }
1180
1181         next = 0;
1182         while ((obd = class_devices_in_group(&sbi->ll_sb_uuid, &next)) !=NULL) {
1183                 class_manual_cleanup(obd);
1184         }
1185
1186         if (profilenm)
1187                 class_del_profile(profilenm);
1188
1189         ll_free_sbi(sb);
1190         lsi->lsi_llsbi = NULL;
1191
1192         lustre_common_put_super(sb);
1193
1194         LCONSOLE_WARN("client %s umount complete\n", ll_instance);
1195
1196         cfs_module_put();
1197
1198         EXIT;
1199 } /* client_put_super */
1200
1201 int ll_shrink_cache(int nr_to_scan, SHRINKER_MASK_T gfp_mask)
1202 {
1203         struct ll_sb_info *sbi;
1204         int count = 0;
1205
1206         if ((nr_to_scan != 0) && (gfp_mask & __GFP_FS))
1207                 return -1;
1208
1209         /* don't race with umount */
1210         down_read(&ll_sb_sem);
1211         list_for_each_entry(sbi, &ll_super_blocks, ll_list)
1212                 count += llap_shrink_cache(sbi, nr_to_scan);
1213         up_read(&ll_sb_sem);
1214
1215         return count;
1216 }
1217
1218 struct inode *ll_inode_from_lock(struct ldlm_lock *lock)
1219 {
1220         struct inode *inode = NULL;
1221         /* NOTE: we depend on atomic igrab() -bzzz */
1222         lock_res_and_lock(lock);
1223         if (lock->l_ast_data) {
1224                 struct ll_inode_info *lli = ll_i2info(lock->l_ast_data);
1225                 if (lli->lli_inode_magic == LLI_INODE_MAGIC) {
1226                         inode = igrab(lock->l_ast_data);
1227                 } else {
1228                         inode = lock->l_ast_data;
1229                         ldlm_lock_debug(NULL, inode->i_state & I_FREEING ?
1230                                                 D_INFO : D_WARNING,
1231                                         lock, __FILE__, __func__, __LINE__,
1232                                         "l_ast_data %p is bogus: magic %08x",
1233                                         lock->l_ast_data, lli->lli_inode_magic);
1234                         inode = NULL;
1235                 }
1236         }
1237         unlock_res_and_lock(lock);
1238         return inode;
1239 }
1240
1241 static int null_if_equal(struct ldlm_lock *lock, void *data)
1242 {
1243         if (data == lock->l_ast_data) {
1244                 lock->l_ast_data = NULL;
1245
1246                 if (lock->l_req_mode != lock->l_granted_mode)
1247                         LDLM_ERROR(lock,"clearing inode with ungranted lock");
1248         }
1249
1250         return LDLM_ITER_CONTINUE;
1251 }
1252
1253 void ll_clear_inode(struct inode *inode)
1254 {
1255         struct ll_fid fid;
1256         struct ll_inode_info *lli = ll_i2info(inode);
1257         struct ll_sb_info *sbi = ll_i2sbi(inode);
1258         ENTRY;
1259
1260         CDEBUG(D_VFSTRACE, "VFS Op:inode=%lu/%u(%p)\n", inode->i_ino,
1261                inode->i_generation, inode);
1262
1263         if (S_ISDIR(inode->i_mode)) {
1264                 /* these should have been cleared in ll_file_release */
1265                 LASSERT(lli->lli_sai == NULL);
1266                 LASSERT(lli->lli_opendir_key == NULL);
1267                 LASSERT(lli->lli_opendir_pid == 0);
1268         }
1269
1270         ll_inode2fid(&fid, inode);
1271         clear_bit(LLI_F_HAVE_MDS_SIZE_LOCK, &lli->lli_flags);
1272         mdc_change_cbdata(sbi->ll_mdc_exp, &fid, null_if_equal, inode);
1273
1274         LASSERT(!lli->lli_open_fd_write_count);
1275         LASSERT(!lli->lli_open_fd_read_count);
1276         LASSERT(!lli->lli_open_fd_exec_count);
1277
1278         if (lli->lli_mds_write_och)
1279                 ll_mdc_real_close(inode, FMODE_WRITE);
1280         if (lli->lli_mds_exec_och) {
1281                 if (!FMODE_EXEC)
1282                         CERROR("No FMODE exec, bug exec och is present for "
1283                                "inode %ld\n", inode->i_ino);
1284                 ll_mdc_real_close(inode, FMODE_EXEC);
1285         }
1286         if (lli->lli_mds_read_och)
1287                 ll_mdc_real_close(inode, FMODE_READ);
1288
1289
1290         if (lli->lli_smd) {
1291                 obd_change_cbdata(sbi->ll_osc_exp, lli->lli_smd,
1292                                   null_if_equal, inode);
1293
1294                 obd_free_memmd(sbi->ll_osc_exp, &lli->lli_smd);
1295                 lli->lli_smd = NULL;
1296         }
1297
1298         if (lli->lli_symlink_name) {
1299                 OBD_FREE(lli->lli_symlink_name,
1300                          strlen(lli->lli_symlink_name) + 1);
1301                 lli->lli_symlink_name = NULL;
1302         }
1303
1304 #ifdef CONFIG_FS_POSIX_ACL
1305         if (lli->lli_posix_acl) {
1306                 LASSERT(atomic_read(&lli->lli_posix_acl->a_refcount) == 1);
1307                 posix_acl_release(lli->lli_posix_acl);
1308                 lli->lli_posix_acl = NULL;
1309         }
1310 #endif
1311
1312         lli->lli_inode_magic = LLI_INODE_DEAD;
1313
1314         EXIT;
1315 }
1316 static int ll_setattr_do_truncate(struct inode *inode, loff_t new_size)
1317 {
1318         struct ll_sb_info *sbi = ll_i2sbi(inode);
1319         struct ll_inode_info *lli = ll_i2info(inode);
1320         struct lov_stripe_md *lsm = lli->lli_smd;
1321         int rc;
1322         ldlm_policy_data_t policy = { .l_extent = {new_size,
1323                                                    OBD_OBJECT_EOF } };
1324         struct lustre_handle lockh = { 0 };
1325         int local_lock = 1; /* 0 - no local lock;
1326                              * 1 - lock taken by lock_extent;
1327                              * 2 - by obd_match*/
1328         int ast_flags;
1329         ENTRY;
1330
1331         UNLOCK_INODE_MUTEX(inode);
1332         UP_WRITE_I_ALLOC_SEM(inode);
1333
1334         down_write(&lli->lli_truncate_rwsem);
1335         if (sbi->ll_lockless_truncate_enable &&
1336             (sbi->ll_lco.lco_flags & OBD_CONNECT_TRUNCLOCK)) {
1337                 int n_matches = 0;
1338
1339                 ast_flags = LDLM_FL_BLOCK_GRANTED;
1340                 rc = obd_match(sbi->ll_osc_exp, lsm, LDLM_EXTENT,
1341                                &policy, LCK_PW, &ast_flags, inode, &lockh,
1342                                &n_matches);
1343                 if (rc > 0) {
1344                         local_lock = 2;
1345                         rc = 0;
1346                 } else {
1347                         /* clear the lock handle as it not cleared
1348                          * by obd_match if no matched lock found */
1349                         lockh.cookie = 0;
1350                         if (rc == 0 && n_matches == 0) {
1351                                 local_lock = 0;
1352                                 rc = ll_file_punch(inode, new_size, 1);
1353                         }
1354                 }
1355         }
1356         if (local_lock == 1) {
1357                 /* XXX when we fix the AST intents to pass the discard-range
1358                  * XXX extent, make ast_flags always LDLM_AST_DISCARD_DATA
1359                  * XXX here. */
1360                 ast_flags = (new_size == 0) ? LDLM_AST_DISCARD_DATA : 0;
1361                 rc = ll_extent_lock(NULL, inode, lsm, LCK_PW, &policy,
1362                                     &lockh, ast_flags);
1363                 if (unlikely(rc != 0))
1364                         local_lock = 0;
1365         }
1366
1367         LOCK_INODE_MUTEX(inode);
1368         DOWN_WRITE_I_ALLOC_SEM(inode);
1369         if (likely(rc == 0)) {
1370                 /* Only ll_inode_size_lock is taken at this level.
1371                  * lov_stripe_lock() is grabbed by ll_truncate() only over
1372                  * call to obd_adjust_kms().  If vmtruncate returns 0, then
1373                  * ll_truncate dropped ll_inode_size_lock() */
1374                 ll_inode_size_lock(inode, 0);
1375                 if (!local_lock)
1376                         set_bit(LLI_F_SRVLOCK, &lli->lli_flags);
1377                 rc = vmtruncate(inode, new_size);
1378                 clear_bit(LLI_F_SRVLOCK, &lli->lli_flags);
1379                 if (rc != 0) {
1380                         LASSERT(SEM_COUNT(&lli->lli_size_sem) <= 0);
1381                         ll_inode_size_unlock(inode, 0);
1382                 }
1383         }
1384         if (local_lock) {
1385                 int err;
1386
1387                 err = (local_lock == 2) ?
1388                         obd_cancel(sbi->ll_osc_exp, lsm, LCK_PW, &lockh, 0, 0):
1389                         ll_extent_unlock(NULL, inode, lsm, LCK_PW, &lockh);
1390                 if (unlikely(err != 0)){
1391                         CERROR("extent unlock failed: err=%d,"
1392                                " unlock method =%d\n", err, local_lock);
1393                         if (rc == 0)
1394                                 rc = err;
1395                 }
1396         }
1397         up_write(&lli->lli_truncate_rwsem);
1398         RETURN(rc);
1399 }
1400
1401 /* If this inode has objects allocated to it (lsm != NULL), then the OST
1402  * object(s) determine the file size and mtime.  Otherwise, the MDS will
1403  * keep these values until such a time that objects are allocated for it.
1404  * We do the MDS operations first, as it is checking permissions for us.
1405  * We don't to the MDS RPC if there is nothing that we want to store there,
1406  * otherwise there is no harm in updating mtime/atime on the MDS if we are
1407  * going to do an RPC anyways.
1408  *
1409  * If we are doing a truncate, we will send the mtime and ctime updates
1410  * to the OST with the punch RPC, otherwise we do an explicit setattr RPC.
1411  * I don't believe it is possible to get e.g. ATTR_MTIME_SET and ATTR_SIZE
1412  * at the same time.
1413  */
1414 int ll_setattr_raw(struct inode *inode, struct iattr *attr)
1415 {
1416         struct ll_inode_info *lli = ll_i2info(inode);
1417         struct lov_stripe_md *lsm = lli->lli_smd;
1418         struct ll_sb_info *sbi = ll_i2sbi(inode);
1419         struct ptlrpc_request *request = NULL;
1420         struct mdc_op_data *op_data;
1421         struct lustre_md md;
1422         int ia_valid = attr->ia_valid;
1423         int rc = 0;
1424         ENTRY;
1425
1426         CDEBUG(D_VFSTRACE, "VFS Op:inode=%lu valid %x\n", inode->i_ino,
1427                attr->ia_valid);
1428         ll_stats_ops_tally(ll_i2sbi(inode), LPROC_LL_SETATTR, 1);
1429
1430         if (ia_valid & ATTR_SIZE) {
1431                 if (attr->ia_size > ll_file_maxbytes(inode)) {
1432                         CDEBUG(D_INODE, "file too large %llu > "LPU64"\n",
1433                                attr->ia_size, ll_file_maxbytes(inode));
1434                         RETURN(-EFBIG);
1435                 }
1436
1437                 attr->ia_valid |= ATTR_MTIME | ATTR_CTIME;
1438         }
1439
1440         /* POSIX: check before ATTR_*TIME_SET set (from inode_change_ok) */
1441         if (ia_valid & (ATTR_MTIME_SET | ATTR_ATIME_SET)) {
1442                 if (cfs_curproc_fsuid() != inode->i_uid &&
1443                     !cfs_capable(CFS_CAP_FOWNER))
1444                         RETURN(-EPERM);
1445         }
1446
1447         /* We mark all of the fields "set" so MDS/OST does not re-set them */
1448         if (attr->ia_valid & ATTR_CTIME) {
1449                 attr->ia_ctime = CURRENT_TIME;
1450                 attr->ia_valid |= ATTR_CTIME_SET;
1451         }
1452         if (!(ia_valid & ATTR_ATIME_SET) && (attr->ia_valid & ATTR_ATIME)) {
1453                 attr->ia_atime = CURRENT_TIME;
1454                 attr->ia_valid |= ATTR_ATIME_SET;
1455         }
1456         if (!(ia_valid & ATTR_MTIME_SET) && (attr->ia_valid & ATTR_MTIME)) {
1457                 attr->ia_mtime = CURRENT_TIME;
1458                 attr->ia_valid |= ATTR_MTIME_SET;
1459         }
1460
1461         if (attr->ia_valid & (ATTR_MTIME | ATTR_CTIME))
1462                 CDEBUG(D_INODE, "setting mtime %lu, ctime %lu, now = %lu\n",
1463                        LTIME_S(attr->ia_mtime), LTIME_S(attr->ia_ctime),
1464                        CURRENT_SECONDS);
1465
1466         /* NB: ATTR_SIZE will only be set after this point if the size
1467          * resides on the MDS, ie, this file has no objects. */
1468         if (lsm)
1469                 attr->ia_valid &= ~ATTR_SIZE;
1470
1471         OBD_ALLOC_PTR(op_data);
1472         if (NULL == op_data)
1473                 RETURN(-ENOMEM);
1474         /* We always do an MDS RPC, even if we're only changing the size;
1475          * only the MDS knows whether truncate() should fail with -ETXTBUSY */
1476         ll_prepare_mdc_op_data(op_data, inode, NULL, NULL, 0, 0, NULL);
1477
1478         rc = mdc_setattr(sbi->ll_mdc_exp, op_data,
1479                          attr, NULL, 0, NULL, 0, &request);
1480
1481         if (rc) {
1482                 ptlrpc_req_finished(request);
1483                 if (rc == -ENOENT) {
1484                         inode->i_nlink = 0;
1485                         /* Unlinked special device node?  Or just a race?
1486                          * Pretend we done everything. */
1487                         if (!S_ISREG(inode->i_mode) &&
1488                             !S_ISDIR(inode->i_mode))
1489                                 rc = inode_setattr(inode, attr);
1490                 } else if (rc != -EPERM && rc != -EACCES && rc != -ETXTBSY)
1491                         CERROR("mdc_setattr fails: rc = %d\n", rc);
1492                 OBD_FREE_PTR(op_data);
1493                 RETURN(rc);
1494         }
1495         OBD_FREE_PTR(op_data);
1496
1497         rc = mdc_req2lustre_md(request, REPLY_REC_OFF, sbi->ll_osc_exp, &md);
1498         if (rc) {
1499                 ptlrpc_req_finished(request);
1500                 RETURN(rc);
1501         }
1502
1503         /* We call inode_setattr to adjust timestamps.
1504          * If there is at least some data in file, we cleared ATTR_SIZE above to
1505          * avoid invoking vmtruncate, otherwise it is important to call
1506          * vmtruncate in inode_setattr to update inode->i_size (bug 6196) */
1507         rc = inode_setattr(inode, attr);
1508
1509         ll_update_inode(inode, &md);
1510         ptlrpc_req_finished(request);
1511
1512         if (!lsm || !S_ISREG(inode->i_mode)) {
1513                 CDEBUG(D_INODE, "no lsm: not setting attrs on OST\n");
1514                 RETURN(rc);
1515         }
1516
1517         /* We really need to get our PW lock before we change inode->i_size.
1518          * If we don't we can race with other i_size updaters on our node, like
1519          * ll_file_read.  We can also race with i_size propogation to other
1520          * nodes through dirtying and writeback of final cached pages.  This
1521          * last one is especially bad for racing o_append users on other
1522          * nodes. */
1523         if (ia_valid & ATTR_SIZE) {
1524                 rc = ll_setattr_do_truncate(inode, attr->ia_size);
1525         } else if (ia_valid & (ATTR_MTIME | ATTR_MTIME_SET)) {
1526                 struct obd_info *oinfo;
1527                 struct obdo *oa;
1528                 struct lustre_handle lockh = { 0 };
1529                 obd_valid valid;
1530
1531                 OBD_ALLOC_PTR(oinfo);
1532                 if (NULL == oinfo)
1533                         RETURN(-ENOMEM);
1534
1535                 CDEBUG(D_INODE, "set mtime on OST inode %lu to %lu\n",
1536                        inode->i_ino, LTIME_S(attr->ia_mtime));
1537
1538                 OBDO_ALLOC(oa);
1539                 if (oa) {
1540                         oa->o_id = lsm->lsm_object_id;
1541                         oa->o_gr = lsm->lsm_object_gr;
1542                         oa->o_valid = OBD_MD_FLID | OBD_MD_FLGROUP;
1543
1544                         valid = OBD_MD_FLTYPE | OBD_MD_FLFID | OBD_MD_FLGENER;
1545
1546                         if (LTIME_S(attr->ia_mtime) < LTIME_S(attr->ia_ctime)){
1547                                 struct ost_lvb xtimes;
1548
1549                                 /* setting mtime to past is performed under PW
1550                                  * EOF extent lock */
1551                                 oinfo->oi_policy.l_extent.start = 0;
1552                                 oinfo->oi_policy.l_extent.end = OBD_OBJECT_EOF;
1553                                 rc = ll_extent_lock(NULL, inode, lsm, LCK_PW,
1554                                                     &oinfo->oi_policy,
1555                                                     &lockh, 0);
1556                                 if (rc) {
1557                                         OBD_FREE_PTR(oinfo);
1558                                         RETURN(rc);
1559                                 }
1560
1561                                 /* setattr under locks
1562                                  *
1563                                  * 1. restore inode's timestamps which
1564                                  * are about to be set as long as
1565                                  * concurrent stat (via
1566                                  * ll_glimpse_size) might bring
1567                                  * out-of-date ones
1568                                  *
1569                                  * 2. update lsm so that next stat
1570                                  * (via ll_glimpse_size) could get
1571                                  * correct values in lsm */
1572                                 lov_stripe_lock(lli->lli_smd);
1573                                 if (ia_valid & ATTR_ATIME) {
1574                                         LTIME_S(inode->i_atime) =
1575                                                 xtimes.lvb_atime =
1576                                                 LTIME_S(attr->ia_atime);
1577                                         valid |= OBD_MD_FLATIME;
1578                                 }
1579                                 if (ia_valid & ATTR_MTIME) {
1580                                         LTIME_S(inode->i_mtime) =
1581                                                 xtimes.lvb_mtime =
1582                                                 LTIME_S(attr->ia_mtime);
1583                                         valid |= OBD_MD_FLMTIME;
1584                                 }
1585                                 if (ia_valid & ATTR_CTIME) {
1586                                         LTIME_S(inode->i_ctime) =
1587                                                 xtimes.lvb_ctime =
1588                                                 LTIME_S(attr->ia_ctime);
1589                                         valid |= OBD_MD_FLCTIME;
1590                                 }
1591
1592                                 obd_update_lvb(ll_i2obdexp(inode), lli->lli_smd,
1593                                                &xtimes, valid);
1594                                 lov_stripe_unlock(lli->lli_smd);
1595                         } else {
1596                                 /* lockless setattr
1597                                  *
1598                                  * 1. do not use inode's timestamps
1599                                  * because concurrent stat might fill
1600                                  * the inode with out-of-date times,
1601                                  * send values from attr instead
1602                                  *
1603                                  * 2.do no update lsm, as long as stat
1604                                  * (via ll_glimpse_size) will bring
1605                                  * attributes from osts anyway */
1606                                 if (ia_valid & ATTR_ATIME) {
1607                                         oa->o_atime = LTIME_S(attr->ia_atime);
1608                                         oa->o_valid |= OBD_MD_FLATIME;
1609                                 }
1610                                 if (ia_valid & ATTR_MTIME) {
1611                                         oa->o_mtime = LTIME_S(attr->ia_mtime);
1612                                         oa->o_valid |= OBD_MD_FLMTIME;
1613                                 }
1614                                 if (ia_valid & ATTR_CTIME) {
1615                                         oa->o_ctime = LTIME_S(attr->ia_ctime);
1616                                         oa->o_valid |= OBD_MD_FLCTIME;
1617                                 }
1618                         }
1619
1620                         obdo_from_inode(oa, inode, valid);
1621
1622                         oinfo->oi_oa = oa;
1623                         oinfo->oi_md = lsm;
1624
1625                         rc = obd_setattr_rqset(sbi->ll_osc_exp, oinfo, NULL);
1626                         if (rc)
1627                                 CERROR("obd_setattr_async fails: rc=%d\n", rc);
1628
1629                         if (LTIME_S(attr->ia_mtime) < LTIME_S(attr->ia_ctime)){
1630                                 int err;
1631
1632                                 err = ll_extent_unlock(NULL, inode, lsm,
1633                                                        LCK_PW, &lockh);
1634                                 if (unlikely(err != 0)) {
1635                                         CERROR("extent unlock failed: "
1636                                                "err=%d\n", err);
1637                                         if (rc == 0)
1638                                                 rc = err;
1639                                 }
1640                         }
1641                         OBDO_FREE(oa);
1642                 } else {
1643                         rc = -ENOMEM;
1644                 }
1645                 OBD_FREE_PTR(oinfo);
1646         }
1647         RETURN(rc);
1648 }
1649
1650 int ll_setattr(struct dentry *de, struct iattr *attr)
1651 {
1652         int mode = de->d_inode->i_mode;
1653
1654         if ((attr->ia_valid & (ATTR_CTIME|ATTR_SIZE|ATTR_MODE)) ==
1655             (ATTR_CTIME|ATTR_SIZE|ATTR_MODE))
1656                 attr->ia_valid |= MDS_OPEN_OWNEROVERRIDE;
1657         if ((attr->ia_valid & (ATTR_MODE|ATTR_FORCE|ATTR_SIZE)) ==
1658             (ATTR_SIZE|ATTR_MODE)) {
1659                 if (((mode & S_ISUID) && (!(attr->ia_mode & S_ISUID))) ||
1660                     ((mode & S_ISGID) && (mode & S_IXGRP) &&
1661                     (!(attr->ia_mode & S_ISGID))))
1662                         attr->ia_valid |= ATTR_FORCE;
1663         }
1664
1665         if ((mode & S_ISUID) && S_ISREG(mode) &&
1666             !(attr->ia_mode & S_ISUID))
1667                 attr->ia_valid |= ATTR_KILL_SUID;
1668
1669         if (((mode & (S_ISGID|S_IXGRP)) == (S_ISGID|S_IXGRP)) &&
1670             S_ISREG(mode) && !(attr->ia_mode & S_ISGID))
1671                 attr->ia_valid |= ATTR_KILL_SGID;
1672
1673         return ll_setattr_raw(de->d_inode, attr);
1674 }
1675
1676 int ll_statfs_internal(struct super_block *sb, struct obd_statfs *osfs,
1677                        __u64 max_age, __u32 flags)
1678 {
1679         struct ll_sb_info *sbi = ll_s2sbi(sb);
1680         struct obd_statfs obd_osfs;
1681         int rc;
1682         ENTRY;
1683
1684         rc = obd_statfs(class_exp2obd(sbi->ll_mdc_exp), osfs, max_age, flags);
1685         if (rc) {
1686                 CERROR("mdc_statfs fails: rc = %d\n", rc);
1687                 RETURN(rc);
1688         }
1689
1690         osfs->os_type = sb->s_magic;
1691
1692         CDEBUG(D_SUPER, "MDC blocks "LPU64"/"LPU64" objects "LPU64"/"LPU64"\n",
1693                osfs->os_bavail, osfs->os_blocks, osfs->os_ffree,osfs->os_files);
1694
1695         if (sbi->ll_flags & LL_SBI_LAZYSTATFS)
1696                 flags |= OBD_STATFS_NODELAY;
1697
1698         rc = obd_statfs_rqset(class_exp2obd(sbi->ll_osc_exp),
1699                               &obd_osfs, max_age, flags);
1700         if (rc) {
1701                 CERROR("obd_statfs fails: rc = %d\n", rc);
1702                 RETURN(rc);
1703         }
1704
1705         CDEBUG(D_SUPER, "OSC blocks "LPU64"/"LPU64" objects "LPU64"/"LPU64"\n",
1706                obd_osfs.os_bavail, obd_osfs.os_blocks, obd_osfs.os_ffree,
1707                obd_osfs.os_files);
1708
1709         osfs->os_bsize = obd_osfs.os_bsize;
1710         osfs->os_blocks = obd_osfs.os_blocks;
1711         osfs->os_bfree = obd_osfs.os_bfree;
1712         osfs->os_bavail = obd_osfs.os_bavail;
1713
1714         /* If we don't have as many objects free on the OST as inodes
1715          * on the MDS, we reduce the total number of inodes to
1716          * compensate, so that the "inodes in use" number is correct.
1717          */
1718         if (obd_osfs.os_ffree < osfs->os_ffree) {
1719                 osfs->os_files = (osfs->os_files - osfs->os_ffree) +
1720                         obd_osfs.os_ffree;
1721                 osfs->os_ffree = obd_osfs.os_ffree;
1722         }
1723
1724         RETURN(rc);
1725 }
1726 #ifndef HAVE_STATFS_DENTRY_PARAM
1727 int ll_statfs(struct super_block *sb, struct kstatfs *sfs)
1728 {
1729 #else
1730 int ll_statfs(struct dentry *de, struct kstatfs *sfs)
1731 {
1732         struct super_block *sb = de->d_sb;
1733 #endif
1734         struct obd_statfs osfs;
1735         int rc;
1736
1737         CDEBUG(D_VFSTRACE, "VFS Op: at "LPU64" jiffies\n", get_jiffies_64());
1738         ll_stats_ops_tally(ll_s2sbi(sb), LPROC_LL_STAFS, 1);
1739
1740         /* Some amount of caching on the client is allowed */
1741         rc = ll_statfs_internal(sb, &osfs,
1742                                 cfs_time_shift_64(-OBD_STATFS_CACHE_SECONDS),
1743                                 0);
1744         if (rc)
1745                 return rc;
1746
1747         statfs_unpack(sfs, &osfs);
1748
1749         /* We need to downshift for all 32-bit kernels, because we can't
1750          * tell if the kernel is being called via sys_statfs64() or not.
1751          * Stop before overflowing f_bsize - in which case it is better
1752          * to just risk EOVERFLOW if caller is using old sys_statfs(). */
1753         if (sizeof(long) < 8) {
1754                 while (osfs.os_blocks > ~0UL && sfs->f_bsize < 0x40000000) {
1755                         sfs->f_bsize <<= 1;
1756
1757                         osfs.os_blocks >>= 1;
1758                         osfs.os_bfree >>= 1;
1759                         osfs.os_bavail >>= 1;
1760                 }
1761         }
1762
1763         sfs->f_blocks = osfs.os_blocks;
1764         sfs->f_bfree = osfs.os_bfree;
1765         sfs->f_bavail = osfs.os_bavail;
1766
1767         return 0;
1768 }
1769
1770 void ll_inode_size_lock(struct inode *inode, int lock_lsm)
1771 {
1772         struct ll_inode_info *lli;
1773         struct lov_stripe_md *lsm;
1774
1775         lli = ll_i2info(inode);
1776         LASSERT(lli->lli_size_sem_owner != current);
1777         down(&lli->lli_size_sem);
1778         LASSERT(lli->lli_size_sem_owner == NULL);
1779         lli->lli_size_sem_owner = current;
1780         lsm = lli->lli_smd;
1781         LASSERTF(lsm != NULL || lock_lsm == 0, "lsm %p, lock_lsm %d\n",
1782                  lsm, lock_lsm);
1783         if (lock_lsm)
1784                 lov_stripe_lock(lsm);
1785 }
1786
1787 void ll_inode_size_unlock(struct inode *inode, int unlock_lsm)
1788 {
1789         struct ll_inode_info *lli;
1790         struct lov_stripe_md *lsm;
1791
1792         lli = ll_i2info(inode);
1793         lsm = lli->lli_smd;
1794         LASSERTF(lsm != NULL || unlock_lsm == 0, "lsm %p, lock_lsm %d\n",
1795                  lsm, unlock_lsm);
1796         if (unlock_lsm)
1797                 lov_stripe_unlock(lsm);
1798         LASSERT(lli->lli_size_sem_owner == current);
1799         lli->lli_size_sem_owner = NULL;
1800         up(&lli->lli_size_sem);
1801 }
1802
1803 static void ll_replace_lsm(struct inode *inode, struct lov_stripe_md *lsm)
1804 {
1805         struct ll_inode_info *lli = ll_i2info(inode);
1806
1807         dump_lsm(D_INODE, lsm);
1808         dump_lsm(D_INODE, lli->lli_smd);
1809         LASSERTF(lsm->lsm_magic == LOV_MAGIC_JOIN,
1810                  "lsm must be joined lsm %p\n", lsm);
1811         obd_free_memmd(ll_i2obdexp(inode), &lli->lli_smd);
1812         CDEBUG(D_INODE, "replace lsm %p to lli_smd %p for inode %lu%u(%p)\n",
1813                lsm, lli->lli_smd, inode->i_ino, inode->i_generation, inode);
1814         lli->lli_smd = lsm;
1815         lli->lli_maxbytes = lsm->lsm_maxbytes;
1816         if (lli->lli_maxbytes > PAGE_CACHE_MAXBYTES)
1817                 lli->lli_maxbytes = PAGE_CACHE_MAXBYTES;
1818 }
1819
1820 void ll_update_inode(struct inode *inode, struct lustre_md *md)
1821 {
1822         struct ll_inode_info *lli = ll_i2info(inode);
1823         struct mds_body *body = md->body;
1824         struct lov_stripe_md *lsm = md->lsm;
1825         struct ll_sb_info *sbi = ll_i2sbi(inode);
1826         ENTRY;
1827
1828         CDEBUG(D_INODE, "body->valid = "LPX64"\n", body->valid);
1829
1830         LASSERT ((lsm != NULL) == ((body->valid & OBD_MD_FLEASIZE) != 0));
1831         if (lsm != NULL) {
1832                 if (lli->lli_smd == NULL) {
1833                         if (lsm->lsm_magic != LOV_MAGIC_V1 &&
1834                             lsm->lsm_magic != LOV_MAGIC_V3 &&
1835                             lsm->lsm_magic != LOV_MAGIC_JOIN) {
1836                                 dump_lsm(D_ERROR, lsm);
1837                                 LBUG();
1838                         }
1839                         CDEBUG(D_INODE, "adding lsm %p to inode %lu/%u(%p)\n",
1840                                lsm, inode->i_ino, inode->i_generation, inode);
1841                         /* ll_inode_size_lock() requires it is only called
1842                          * with lli_smd != NULL or lock_lsm == 0 or we can
1843                          * race between lock/unlock.  bug 9547 */
1844                         lli->lli_smd = lsm;
1845                         lli->lli_maxbytes = lsm->lsm_maxbytes;
1846                         if (lli->lli_maxbytes > PAGE_CACHE_MAXBYTES)
1847                                 lli->lli_maxbytes = PAGE_CACHE_MAXBYTES;
1848                 } else {
1849                         if ((lli->lli_smd->lsm_magic == lsm->lsm_magic ||
1850                              (lli->lli_smd->lsm_magic == LOV_MAGIC_V3 &&
1851                               lsm->lsm_magic == LOV_MAGIC_V1) ||
1852                              (lli->lli_smd->lsm_magic == LOV_MAGIC_V1 &&
1853                               lsm->lsm_magic == LOV_MAGIC_V3)) &&
1854                             lli->lli_smd->lsm_stripe_count ==
1855                                         lsm->lsm_stripe_count) {
1856                                 /* The MDS can suddenly change the magic to
1857                                  * v1/v3 if it has been upgraded/downgraded to a
1858                                  * version that does/doesn't support OST pools.
1859                                  * In this case, we consider that the cached lsm
1860                                  * is equivalent to the new one and keep it in
1861                                  * place until the inode is evicted from the
1862                                  * cache */
1863                                 if (lov_stripe_md_cmp(lli->lli_smd, lsm)) {
1864                                         CERROR("lsm mismatch for inode %ld\n",
1865                                                 inode->i_ino);
1866                                         CERROR("lli_smd:\n");
1867                                         dump_lsm(D_ERROR, lli->lli_smd);
1868                                         CERROR("lsm:\n");
1869                                         dump_lsm(D_ERROR, lsm);
1870                                         LBUG();
1871                                 }
1872                         } else
1873                                 ll_replace_lsm(inode, lsm);
1874                 }
1875                 if (lli->lli_smd != lsm)
1876                         obd_free_memmd(ll_i2obdexp(inode), &lsm);
1877         }
1878
1879 #ifdef CONFIG_FS_POSIX_ACL
1880         LASSERT(!md->posix_acl || (body->valid & OBD_MD_FLACL));
1881         if (body->valid & OBD_MD_FLACL) {
1882                 spin_lock(&lli->lli_lock);
1883                 if (lli->lli_posix_acl)
1884                         posix_acl_release(lli->lli_posix_acl);
1885                 lli->lli_posix_acl = md->posix_acl;
1886                 spin_unlock(&lli->lli_lock);
1887         }
1888 #endif
1889
1890         inode->i_ino = ll_fid_build_ino(sbi, &body->fid1);
1891         inode->i_generation = ll_fid_build_gen(sbi, &body->fid1);
1892         *ll_inode_lu_fid(inode) = *((struct lu_fid*)&md->body->fid1);
1893
1894         if (body->valid & OBD_MD_FLATIME) {
1895                 if (body->atime > LTIME_S(inode->i_atime))
1896                         LTIME_S(inode->i_atime) = body->atime;
1897                 lli->lli_lvb.lvb_atime = body->atime;
1898         }
1899         if (body->valid & OBD_MD_FLMTIME) {
1900                 if (body->mtime > LTIME_S(inode->i_mtime)) {
1901                         CDEBUG(D_INODE, "setting ino %lu mtime from %lu "
1902                                "to "LPU64"\n", inode->i_ino,
1903                                LTIME_S(inode->i_mtime), body->mtime);
1904                         LTIME_S(inode->i_mtime) = body->mtime;
1905                 }
1906                 lli->lli_lvb.lvb_mtime = body->mtime;
1907         }
1908         if (body->valid & OBD_MD_FLCTIME) {
1909                 if (body->ctime > LTIME_S(inode->i_ctime))
1910                         LTIME_S(inode->i_ctime) = body->ctime;
1911                 lli->lli_lvb.lvb_ctime = body->ctime;
1912         }
1913         if (body->valid & OBD_MD_FLMODE)
1914                 inode->i_mode = (inode->i_mode & S_IFMT)|(body->mode & ~S_IFMT);
1915         if (body->valid & OBD_MD_FLTYPE)
1916                 inode->i_mode = (inode->i_mode & ~S_IFMT)|(body->mode & S_IFMT);
1917         if (S_ISREG(inode->i_mode)) {
1918                 inode->i_blkbits = min(PTLRPC_MAX_BRW_BITS+1, LL_MAX_BLKSIZE_BITS);
1919         } else {
1920                 inode->i_blkbits = inode->i_sb->s_blocksize_bits;
1921         }
1922 #ifdef HAVE_INODE_BLKSIZE
1923         inode->i_blksize = 1<<inode->i_blkbits;
1924 #endif
1925         if (body->valid & OBD_MD_FLUID)
1926                 inode->i_uid = body->uid;
1927         if (body->valid & OBD_MD_FLGID)
1928                 inode->i_gid = body->gid;
1929         if (body->valid & OBD_MD_FLFLAGS)
1930                 inode->i_flags = ll_ext_to_inode_flags(body->flags |
1931                                                        MDS_BFLAG_EXT_FLAGS);
1932         if (body->valid & OBD_MD_FLNLINK)
1933                 inode->i_nlink = body->nlink;
1934
1935         if (body->valid & OBD_MD_FLRDEV)
1936                 inode->i_rdev = old_decode_dev(body->rdev);
1937         if (body->valid & OBD_MD_FLSIZE) {
1938 #if 0           /* Can't block ll_test_inode->ll_update_inode, b=14326*/
1939                 ll_inode_size_lock(inode, 0);
1940                 i_size_write(inode, body->size);
1941                 ll_inode_size_unlock(inode, 0);
1942 #else
1943                 inode->i_size = body->size;
1944 #endif
1945         }
1946         if (body->valid & OBD_MD_FLBLOCKS)
1947                 inode->i_blocks = body->blocks;
1948
1949         if (body->valid & OBD_MD_FLSIZE)
1950                 set_bit(LLI_F_HAVE_MDS_SIZE_LOCK, &lli->lli_flags);
1951         EXIT;
1952 }
1953
1954 static struct backing_dev_info ll_backing_dev_info = {
1955         .ra_pages       = 0,    /* No readahead */
1956 #if (LINUX_VERSION_CODE >= KERNEL_VERSION(2,6,12))
1957         .capabilities   = 0,    /* Does contribute to dirty memory */
1958 #else
1959         .memory_backed  = 0,    /* Does contribute to dirty memory */
1960 #endif
1961 };
1962
1963 void ll_read_inode2(struct inode *inode, void *opaque)
1964 {
1965         struct lustre_md *md = opaque;
1966         struct ll_inode_info *lli = ll_i2info(inode);
1967         ENTRY;
1968
1969         CDEBUG(D_VFSTRACE, "VFS Op:inode=%lu/%u(%p)\n", inode->i_ino,
1970                inode->i_generation, inode);
1971
1972         ll_lli_init(lli);
1973
1974         LASSERT(!lli->lli_smd);
1975
1976         /* Core attributes from the MDS first.  This is a new inode, and
1977          * the VFS doesn't zero times in the core inode so we have to do
1978          * it ourselves.  They will be overwritten by either MDS or OST
1979          * attributes - we just need to make sure they aren't newer. */
1980         LTIME_S(inode->i_mtime) = 0;
1981         LTIME_S(inode->i_atime) = 0;
1982         LTIME_S(inode->i_ctime) = 0;
1983         inode->i_rdev = 0;
1984         ll_update_inode(inode, md);
1985
1986         /* OIDEBUG(inode); */
1987
1988         if (S_ISREG(inode->i_mode)) {
1989                 struct ll_sb_info *sbi = ll_i2sbi(inode);
1990                 inode->i_op = &ll_file_inode_operations;
1991                 inode->i_fop = sbi->ll_fop;
1992                 inode->i_mapping->a_ops = &ll_aops;
1993                 EXIT;
1994         } else if (S_ISDIR(inode->i_mode)) {
1995                 inode->i_op = &ll_dir_inode_operations;
1996                 inode->i_fop = &ll_dir_operations;
1997                 inode->i_mapping->a_ops = &ll_dir_aops;
1998                 EXIT;
1999         } else if (S_ISLNK(inode->i_mode)) {
2000                 inode->i_op = &ll_fast_symlink_inode_operations;
2001                 EXIT;
2002         } else {
2003                 inode->i_op = &ll_special_inode_operations;
2004                 init_special_inode(inode, inode->i_mode,
2005                                    kdev_t_to_nr(inode->i_rdev));
2006                 /* initializing backing dev info. */
2007                 inode->i_mapping->backing_dev_info = &ll_backing_dev_info;
2008                 EXIT;
2009         }
2010 }
2011
2012 int ll_iocontrol(struct inode *inode, struct file *file,
2013                  unsigned int cmd, unsigned long arg)
2014 {
2015         struct ll_sb_info *sbi = ll_i2sbi(inode);
2016         struct ptlrpc_request *req = NULL;
2017         int rc, flags = 0;
2018         ENTRY;
2019
2020         switch(cmd) {
2021         case FSFILT_IOC_GETFLAGS: {
2022                 struct ll_fid fid;
2023                 struct mds_body *body;
2024
2025                 ll_inode2fid(&fid, inode);
2026                 rc = mdc_getattr(sbi->ll_mdc_exp, &fid, OBD_MD_FLFLAGS,0,&req);
2027                 if (rc) {
2028                         CERROR("failure %d inode %lu\n", rc, inode->i_ino);
2029                         RETURN(-abs(rc));
2030                 }
2031
2032                 body = lustre_msg_buf(req->rq_repmsg, REPLY_REC_OFF,
2033                                       sizeof(*body));
2034
2035                 /* We want to return EXT3_*_FL flags to the caller via this
2036                  * ioctl.  An older MDS may be sending S_* flags, fix it up. */
2037                 flags = ll_inode_to_ext_flags(body->flags,
2038                                               MDS_BFLAG_EXT_FLAGS);
2039                 ptlrpc_req_finished (req);
2040
2041                 RETURN(put_user(flags, (int *)arg));
2042         }
2043         case FSFILT_IOC_SETFLAGS: {
2044                 struct mdc_op_data op_data = { { 0 } };
2045                 struct ll_iattr_struct attr;
2046                 struct obd_info oinfo = { { { 0 } } };
2047                 struct lov_stripe_md *lsm = ll_i2info(inode)->lli_smd;
2048
2049                 if (get_user(flags, (int *)arg))
2050                         RETURN(-EFAULT);
2051
2052                 oinfo.oi_md = lsm;
2053                 OBDO_ALLOC(oinfo.oi_oa);
2054                 if (!oinfo.oi_oa)
2055                         RETURN(-ENOMEM);
2056
2057                 ll_prepare_mdc_op_data(&op_data, inode, NULL, NULL, 0, 0, NULL);
2058
2059                 memset(&attr, 0, sizeof(attr));
2060                 attr.ia_attr_flags = flags;
2061                 ((struct iattr *)&attr)->ia_valid |= ATTR_ATTR_FLAG;
2062
2063                 rc = mdc_setattr(sbi->ll_mdc_exp, &op_data,
2064                                  (struct iattr *)&attr, NULL, 0, NULL, 0, &req);
2065                 ptlrpc_req_finished(req);
2066                 if (rc) {
2067                         OBDO_FREE(oinfo.oi_oa);
2068                         RETURN(rc);
2069                 }
2070
2071                 if (lsm == NULL) {
2072                         OBDO_FREE(oinfo.oi_oa);
2073                         GOTO(update_cache, rc);
2074                 }
2075
2076                 oinfo.oi_oa->o_id = lsm->lsm_object_id;
2077                 oinfo.oi_oa->o_gr = lsm->lsm_object_gr;
2078                 oinfo.oi_oa->o_flags = flags;
2079                 oinfo.oi_oa->o_valid = OBD_MD_FLID | OBD_MD_FLGROUP |
2080                                        OBD_MD_FLFLAGS;
2081
2082                 obdo_from_inode(oinfo.oi_oa, inode,
2083                                 OBD_MD_FLFID | OBD_MD_FLGENER);
2084                 rc = obd_setattr_rqset(sbi->ll_osc_exp, &oinfo, NULL);
2085                 OBDO_FREE(oinfo.oi_oa);
2086                 if (rc) {
2087                         if (rc != -EPERM && rc != -EACCES)
2088                                 CERROR("osc_setattr_async fails: rc = %d\n",rc);
2089                         RETURN(rc);
2090                 }
2091
2092                 EXIT;
2093 update_cache:
2094                 inode->i_flags = ll_ext_to_inode_flags(flags |
2095                                                        MDS_BFLAG_EXT_FLAGS);
2096                 return 0;
2097         }
2098         default:
2099                 RETURN(-ENOSYS);
2100         }
2101
2102         RETURN(0);
2103 }
2104
2105 /* umount -f client means force down, don't save state */
2106 #ifdef HAVE_UMOUNTBEGIN_VFSMOUNT
2107 void ll_umount_begin(struct vfsmount *vfsmnt, int flags)
2108 {
2109         struct super_block *sb = vfsmnt->mnt_sb;
2110 #else
2111 void ll_umount_begin(struct super_block *sb)
2112 {
2113 #endif
2114         struct lustre_sb_info *lsi = s2lsi(sb);
2115         struct ll_sb_info *sbi = ll_s2sbi(sb);
2116         struct obd_device *obd;
2117         struct obd_ioctl_data ioc_data = { 0 };
2118         ENTRY;
2119
2120 #ifdef HAVE_UMOUNTBEGIN_VFSMOUNT
2121         if (!(flags & MNT_FORCE)) {
2122                 EXIT;
2123                 return;
2124         }
2125 #endif
2126
2127         /* Tell the MGC we got umount -f */
2128         lsi->lsi_flags |= LSI_UMOUNT_FORCE;
2129
2130         CDEBUG(D_VFSTRACE, "VFS Op: superblock %p count %d active %d\n", sb,
2131                sb->s_count, atomic_read(&sb->s_active));
2132
2133         obd = class_exp2obd(sbi->ll_mdc_exp);
2134         if (obd == NULL) {
2135                 CERROR("Invalid MDC connection handle "LPX64"\n",
2136                        sbi->ll_mdc_exp->exp_handle.h_cookie);
2137                 EXIT;
2138                 return;
2139         }
2140         obd->obd_force = 1;
2141         obd_iocontrol(IOC_OSC_SET_ACTIVE, sbi->ll_mdc_exp, sizeof ioc_data,
2142                       &ioc_data, NULL);
2143
2144         obd = class_exp2obd(sbi->ll_osc_exp);
2145         if (obd == NULL) {
2146                 CERROR("Invalid LOV connection handle "LPX64"\n",
2147                        sbi->ll_osc_exp->exp_handle.h_cookie);
2148                 EXIT;
2149                 return;
2150         }
2151
2152         obd->obd_force = 1;
2153         obd_iocontrol(IOC_OSC_SET_ACTIVE, sbi->ll_osc_exp, sizeof ioc_data,
2154                       &ioc_data, NULL);
2155
2156         /* Really, we'd like to wait until there are no requests outstanding,
2157          * and then continue.  For now, we just invalidate the requests,
2158          * schedule() and sleep one second if needed, and hope.
2159          */
2160         schedule();
2161 #ifdef HAVE_UMOUNTBEGIN_VFSMOUNT
2162         if (atomic_read(&vfsmnt->mnt_count) > 2) {
2163                 cfs_schedule_timeout(CFS_TASK_INTERRUPTIBLE,
2164                                      cfs_time_seconds(1));
2165                 if (atomic_read(&vfsmnt->mnt_count) > 2)
2166                         LCONSOLE_WARN("Mount still busy with %d refs! You "
2167                                       "may try to umount it a bit later\n",
2168                                       atomic_read(&vfsmnt->mnt_count));
2169         }
2170 #endif
2171
2172         EXIT;
2173 }
2174
2175 int ll_remount_fs(struct super_block *sb, int *flags, char *data)
2176 {
2177         struct ll_sb_info *sbi = ll_s2sbi(sb);
2178         int err;
2179         __u32 read_only;
2180
2181         if ((*flags & MS_RDONLY) != (sb->s_flags & MS_RDONLY)) {
2182                 read_only = *flags & MS_RDONLY;
2183                 err = obd_set_info_async(sbi->ll_mdc_exp, sizeof(KEY_READONLY),
2184                                          KEY_READONLY, sizeof(read_only),
2185                                          &read_only, NULL);
2186
2187                 /* MDS might have expected a different ro key value, b=17493 */
2188                 if (err == -EINVAL) {
2189                         CDEBUG(D_CONFIG, "Retrying remount with 1.6.6 ro key\n");
2190                         err = obd_set_info_async(sbi->ll_mdc_exp,
2191                                                  sizeof(KEY_READONLY_166COMPAT),
2192                                                  KEY_READONLY_166COMPAT,
2193                                                  sizeof(read_only),
2194                                                  &read_only, NULL);
2195                 }
2196
2197                 if (err) {
2198                         CERROR("Failed to change the read-only flag during "
2199                                "remount: %d\n", err);
2200                         return err;
2201                 }
2202
2203                 if (read_only)
2204                         sb->s_flags |= MS_RDONLY;
2205                 else
2206                         sb->s_flags &= ~MS_RDONLY;
2207         }
2208         return 0;
2209 }
2210
2211 int ll_prep_inode(struct obd_export *exp, struct inode **inode,
2212                   struct ptlrpc_request *req, int offset,struct super_block *sb)
2213 {
2214         struct lustre_md md;
2215         struct ll_sb_info *sbi = NULL;
2216         int rc = 0;
2217         ENTRY;
2218
2219         LASSERT(*inode || sb);
2220         sbi = sb ? ll_s2sbi(sb) : ll_i2sbi(*inode);
2221
2222         rc = mdc_req2lustre_md(req, offset, exp, &md);
2223         if (rc)
2224                 RETURN(rc);
2225
2226         if (*inode) {
2227                 ll_update_inode(*inode, &md);
2228         } else {
2229                 LASSERT(sb);
2230                 /** hashing VFS inode by FIDs.
2231                  * IGIF will be used for for compatibility if needed.
2232                  */
2233                 *inode =ll_iget(sb, ll_fid_build_ino(sbi, &md.body->fid1), &md);
2234                 if (*inode == NULL || is_bad_inode(*inode)) {
2235                         mdc_free_lustre_md(exp, &md);
2236                         rc = -ENOMEM;
2237                         CERROR("new_inode -fatal: rc %d\n", rc);
2238                         GOTO(out, rc);
2239                 }
2240         }
2241
2242         rc = obd_checkmd(exp, ll_i2mdcexp(*inode),
2243                          ll_i2info(*inode)->lli_smd);
2244 out:
2245         RETURN(rc);
2246 }
2247
2248 char *llap_origins[] = {
2249         [LLAP_ORIGIN_UNKNOWN] = "--",
2250         [LLAP_ORIGIN_READPAGE] = "rp",
2251         [LLAP_ORIGIN_READAHEAD] = "ra",
2252         [LLAP_ORIGIN_COMMIT_WRITE] = "cw",
2253         [LLAP_ORIGIN_WRITEPAGE] = "wp"
2254 };
2255
2256 struct ll_async_page *llite_pglist_next_llap(struct list_head *head,
2257                                              struct list_head *list)
2258 {
2259         struct ll_async_page *llap;
2260         struct list_head *pos;
2261
2262         list_for_each(pos, list) {
2263                 if (pos == head)
2264                         return NULL;
2265                 llap = list_entry(pos, struct ll_async_page, llap_pglist_item);
2266                 if (llap->llap_page == NULL)
2267                         continue;
2268                 return llap;
2269         }
2270         LBUG();
2271         return NULL;
2272 }
2273
2274 int ll_obd_statfs(struct inode *inode, void *arg)
2275 {
2276         struct ll_sb_info *sbi = NULL;
2277         struct obd_device *client_obd = NULL, *lov_obd = NULL;
2278         struct lov_obd *lov = NULL;
2279         struct obd_statfs stat_buf = {0};
2280         char *buf = NULL;
2281         struct obd_ioctl_data *data = NULL;
2282         __u32 type, index;
2283         int len = 0, rc;
2284
2285         if (!inode || !(sbi = ll_i2sbi(inode)))
2286                 GOTO(out_statfs, rc = -EINVAL);
2287
2288         rc = obd_ioctl_getdata(&buf, &len, arg);
2289         if (rc)
2290                 GOTO(out_statfs, rc);
2291
2292         data = (void*)buf;
2293         if (!data->ioc_inlbuf1 || !data->ioc_inlbuf2 ||
2294             !data->ioc_pbuf1 || !data->ioc_pbuf2)
2295                 GOTO(out_statfs, rc = -EINVAL);
2296
2297         memcpy(&type, data->ioc_inlbuf1, sizeof(__u32));
2298         memcpy(&index, data->ioc_inlbuf2, sizeof(__u32));
2299
2300         if (type == LL_STATFS_MDC) {
2301                 if (index > 0)
2302                         GOTO(out_statfs, rc = -ENODEV);
2303                 client_obd = class_exp2obd(sbi->ll_mdc_exp);
2304         } else if (type == LL_STATFS_LOV) {
2305                 lov_obd = class_exp2obd(sbi->ll_osc_exp);
2306                 lov = &lov_obd->u.lov;
2307
2308                 if (index >= lov->desc.ld_tgt_count)
2309                         GOTO(out_statfs, rc = -ENODEV);
2310
2311                 if (!lov->lov_tgts[index])
2312                         /* Try again with the next index */
2313                         GOTO(out_statfs, rc = -EAGAIN);
2314
2315                 client_obd = class_exp2obd(lov->lov_tgts[index]->ltd_exp);
2316                 if (!lov->lov_tgts[index]->ltd_active)
2317                         GOTO(out_uuid, rc = -ENODATA);
2318         }
2319
2320         if (!client_obd)
2321                 GOTO(out_statfs, rc = -EINVAL);
2322
2323         rc = obd_statfs(client_obd, &stat_buf,
2324                         cfs_time_shift_64(-OBD_STATFS_CACHE_SECONDS), 1);
2325         if (rc)
2326                 GOTO(out_statfs, rc);
2327
2328         if (copy_to_user(data->ioc_pbuf1, &stat_buf, data->ioc_plen1))
2329                 GOTO(out_statfs, rc = -EFAULT);
2330
2331 out_uuid:
2332         if (client_obd) {
2333                 if (copy_to_user(data->ioc_pbuf2, obd2cli_tgt(client_obd),
2334                                  data->ioc_plen2))
2335                         rc = -EFAULT;
2336         } else {
2337                 if (copy_to_user(data->ioc_pbuf2, "Unknown UUID",
2338                                  sizeof("Unknown UUID") + 1))
2339                         rc = -EFAULT;
2340         }
2341
2342 out_statfs:
2343         if (buf)
2344                 obd_ioctl_freedata(buf, len);
2345         return rc;
2346 }
2347
2348 int ll_process_config(struct lustre_cfg *lcfg)
2349 {
2350         char *ptr;
2351         void *sb;
2352         struct lprocfs_static_vars lvars;
2353         unsigned long x;
2354         int rc = 0;
2355
2356         lprocfs_llite_init_vars(&lvars);
2357
2358         /* The instance name contains the sb: lustre-client-aacfe000 */
2359         ptr = strrchr(lustre_cfg_string(lcfg, 0), '-');
2360         if (!ptr || !*(++ptr))
2361                 return -EINVAL;
2362         if (sscanf(ptr, "%lx", &x) != 1)
2363                 return -EINVAL;
2364         sb = (void *)x;
2365         /* This better be a real Lustre superblock! */
2366         LASSERT(s2lsi((struct super_block *)sb)->lsi_lmd->lmd_magic == LMD_MAGIC);
2367
2368         /* Note we have not called client_common_fill_super yet, so
2369            proc fns must be able to handle that! */
2370         rc = class_process_proc_param(PARAM_LLITE, lvars.obd_vars,
2371                                       lcfg, sb);
2372         return(rc);
2373 }
2374
2375 int ll_show_options(struct seq_file *seq, struct vfsmount *vfs)
2376 {
2377         struct ll_sb_info *sbi;
2378
2379         LASSERT((seq != NULL) && (vfs != NULL));
2380         sbi = ll_s2sbi(vfs->mnt_sb);
2381
2382         if (sbi->ll_flags & LL_SBI_NOLCK)
2383                 seq_puts(seq, ",nolock");
2384
2385         if (sbi->ll_flags & LL_SBI_FLOCK)
2386                 seq_puts(seq, ",flock");
2387
2388         if (sbi->ll_flags & LL_SBI_LOCALFLOCK)
2389                 seq_puts(seq, ",localflock");
2390
2391         if (sbi->ll_flags & LL_SBI_USER_XATTR)
2392                 seq_puts(seq, ",user_xattr");
2393
2394         if (sbi->ll_flags & LL_SBI_ACL)
2395                 seq_puts(seq, ",acl");
2396
2397         if (sbi->ll_flags & LL_SBI_LAZYSTATFS)
2398                 seq_puts(seq, ",lazystatfs");
2399
2400         RETURN(0);
2401 }