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 (c) 2003, 2010, Oracle and/or its affiliates. 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         obd_register_lock_cancel_cb(obd, ll_extent_lock_cancel_cb);
 416         obd_register_page_removal_cb(obd, ll_page_removal_cb, ll_pin_extent_cb);
 417
 418         err = obd_connect(&osc_conn, obd, &sbi->ll_sb_uuid, data, &sbi->ll_osc_exp);
 419         if (err == -EBUSY) {
 420                 LCONSOLE_ERROR_MSG(0x150, "An OST (osc %s) is performing "
 421                                    "recovery, of which this client is not a "
 422                                    "part.  Please wait for recovery to "
 423                                    "complete, abort, or time out.\n", osc);
 424                 GOTO(out, err);
 425         } else if (err) {
 426                 CERROR("cannot connect to %s: rc = %d\n", osc, err);
 427                 GOTO(out_cb, err);
 428         }
 429
 430         mutex_down(&sbi->ll_lco.lco_lock);
 431         sbi->ll_lco.lco_flags = data->ocd_connect_flags;
 432         sbi->ll_lco.lco_mdc_exp = sbi->ll_mdc_exp;
 433         sbi->ll_lco.lco_osc_exp = sbi->ll_osc_exp;
 434         mutex_up(&sbi->ll_lco.lco_lock);
 435
 436         err = mdc_init_ea_size(sbi->ll_mdc_exp, sbi->ll_osc_exp);
 437         if (err) {
 438                 CERROR("cannot set max EA and cookie sizes: rc = %d\n", err);
 439                 GOTO(out_osc, err);
 440         }
 441
 442         err = obd_prep_async_page(sbi->ll_osc_exp, NULL, NULL, NULL,
 443                                   0, NULL, NULL, NULL, 0, NULL);
 444         if (err < 0) {
 445                 LCONSOLE_ERROR_MSG(0x151, "There are no OST's in this "
 446                                    "filesystem. There must be at least one "
 447                                    "active OST for a client to start.\n");
 448                 GOTO(out_osc, err);
 449         }
 450
 451         if (!ll_async_page_slab) {
 452                 ll_async_page_slab_size =
 453                         size_round(sizeof(struct ll_async_page)) + err;
 454                 ll_async_page_slab = cfs_mem_cache_create("ll_async_page",
 455                                                           ll_async_page_slab_size,
 456                                                           0, 0);
 457                 if (!ll_async_page_slab)
 458                         GOTO(out_osc, err = -ENOMEM);
 459         }
 460
 461         err = mdc_getstatus(sbi->ll_mdc_exp, &rootfid);
 462         if (err) {
 463                 CERROR("cannot mds_connect: rc = %d\n", err);
 464                 GOTO(out_osc, err);
 465         }
 466         CDEBUG(D_SUPER, "rootfid "LPU64":"DFID"\n", rootfid.id,
 467                         PFID((struct lu_fid*)&rootfid));
 468         sbi->ll_rootino = rootfid.id;
 469
 470         sb->s_op = &lustre_super_operations;
 471 #if THREAD_SIZE >= 8192
 472         /* Disable the NFS export because of stack overflow
 473          * when THREAD_SIZE < 8192. Please refer to 17630. */
 474         sb->s_export_op = &lustre_export_operations;
 475 #endif
 476
 477         /* make root inode
 478          * XXX: move this to after cbd setup? */
 479         err = mdc_getattr(sbi->ll_mdc_exp, &rootfid,
 480                           OBD_MD_FLGETATTR | OBD_MD_FLBLOCKS |
 481                           (sbi->ll_flags & LL_SBI_ACL ? OBD_MD_FLACL : 0),
 482                           0, &request);
 483         if (err) {
 484                 CERROR("mdc_getattr failed for root: rc = %d\n", err);
 485                 GOTO(out_osc, err);
 486         }
 487
 488         err = mdc_req2lustre_md(request, REPLY_REC_OFF, sbi->ll_osc_exp, &md);
 489         if (err) {
 490                 CERROR("failed to understand root inode md: rc = %d\n",err);
 491                 ptlrpc_req_finished (request);
 492                 GOTO(out_osc, err);
 493         }
 494
 495         LASSERT(sbi->ll_rootino != 0);
 496         root = ll_iget(sb, ll_fid_build_ino(&rootfid), &md);
 497
 498         ptlrpc_req_finished(request);
 499
 500         if (root == NULL || is_bad_inode(root)) {
 501                 mdc_free_lustre_md(sbi->ll_osc_exp, &md);
 502                 CERROR("lustre_lite: bad iget4 for root\n");
 503                 GOTO(out_root, err = -EBADF);
 504         }
 505
 506         err = ll_close_thread_start(&sbi->ll_lcq);
 507         if (err) {
 508                 CERROR("cannot start close thread: rc %d\n", err);
 509                 GOTO(out_root, err);
 510         }
 511
 512         checksum = sbi->ll_flags & LL_SBI_DATA_CHECKSUM;
 513         err = obd_set_info_async(sbi->ll_osc_exp, sizeof(KEY_CHECKSUM),
 514                                  KEY_CHECKSUM, sizeof(checksum),
 515                                  &checksum, NULL);
 516
 517         /* making vm readahead 0 for 2.4.x. In the case of 2.6.x,
 518            backing dev info assigned to inode mapping is used for
 519            determining maximal readahead. */
 520 #if (LINUX_VERSION_CODE < KERNEL_VERSION(2,6,0)) && \
 521     !defined(KERNEL_HAS_AS_MAX_READAHEAD)
 522         /* bug 2805 - set VM readahead to zero */
 523         vm_max_readahead = vm_min_readahead = 0;
 524 #endif
 525
 526         sb->s_root = d_alloc_root(root);
 527         if (data != NULL)
 528                 OBD_FREE(data, sizeof(*data));
 529         sb->s_root->d_op = &ll_d_root_ops;
 530
 531         sbi->ll_sdev_orig = sb->s_dev;
 532         /* We set sb->s_dev equal on all lustre clients in order to support
 533          * NFS export clustering.  NFSD requires that the FSID be the same
 534          * on all clients. */
 535         /* s_dev is also used in lt_compare() to compare two fs, but that is
 536          * only a node-local comparison. */
 537         sb->s_dev = get_uuid2int(sbi2mdc(sbi)->cl_target_uuid.uuid,
 538                                  strlen(sbi2mdc(sbi)->cl_target_uuid.uuid));
 539
 540         RETURN(err);
 541
 542 out_root:
 543         if (root)
 544                 iput(root);
 545 out_osc:
 546         obd_disconnect(sbi->ll_osc_exp);
 547         sbi->ll_osc_exp = NULL;
 548 out_cb:
 549         obd = class_name2obd(osc);
 550         obd_unregister_lock_cancel_cb(obd, ll_extent_lock_cancel_cb);
 551         obd_unregister_page_removal_cb(obd, ll_page_removal_cb);
 552 out_mdc:
 553         obd_fid_fini(sbi->ll_mdc_exp);
 554         obd_disconnect(sbi->ll_mdc_exp);
 555         sbi->ll_mdc_exp = NULL;
 556 out:
 557         if (data != NULL)
 558                 OBD_FREE(data, sizeof(*data));
 559         lprocfs_unregister_mountpoint(sbi);
 560         RETURN(err);
 561 }
 562
 563 int ll_get_max_mdsize(struct ll_sb_info *sbi, int *lmmsize)
 564 {
 565         int size, rc;
 566
 567         *lmmsize = obd_size_diskmd(sbi->ll_osc_exp, NULL);
 568         size = sizeof(int);
 569         rc = obd_get_info(sbi->ll_mdc_exp, sizeof(KEY_MAX_EASIZE),
 570                           KEY_MAX_EASIZE, &size, lmmsize, NULL);
 571         if (rc)
 572                 CERROR("Get max mdsize error rc %d \n", rc);
 573
 574         RETURN(rc);
 575 }
 576
 577 void ll_dump_inode(struct inode *inode)
 578 {
 579         struct list_head *tmp;
 580         int dentry_count = 0;
 581
 582         LASSERT(inode != NULL);
 583
 584         list_for_each(tmp, &inode->i_dentry)
 585                 dentry_count++;
 586
 587         CERROR("inode %p dump: dev=%s ino=%lu mode=%o count=%u, %d dentries\n",
 588                inode, ll_i2mdcexp(inode)->exp_obd->obd_name, inode->i_ino,
 589                inode->i_mode, atomic_read(&inode->i_count), dentry_count);
 590 }
 591
 592 void lustre_dump_dentry(struct dentry *dentry, int recur)
 593 {
 594         struct list_head *tmp;
 595         int subdirs = 0;
 596
 597         LASSERT(dentry != NULL);
 598
 599         list_for_each(tmp, &dentry->d_subdirs)
 600                 subdirs++;
 601
 602         CERROR("dentry %p dump: name=%.*s parent=%.*s (%p), inode=%p, count=%u,"
 603                " flags=0x%x, fsdata=%p, %d subdirs\n", dentry,
 604                dentry->d_name.len, dentry->d_name.name,
 605                dentry->d_parent->d_name.len, dentry->d_parent->d_name.name,
 606                dentry->d_parent, dentry->d_inode, atomic_read(&dentry->d_count),
 607                dentry->d_flags, dentry->d_fsdata, subdirs);
 608         if (dentry->d_inode != NULL)
 609                 ll_dump_inode(dentry->d_inode);
 610
 611         if (recur == 0)
 612                 return;
 613
 614         list_for_each(tmp, &dentry->d_subdirs) {
 615                 struct dentry *d = list_entry(tmp, struct dentry, d_child);
 616                 lustre_dump_dentry(d, recur - 1);
 617         }
 618 }
 619
 620 void client_common_put_super(struct super_block *sb)
 621 {
 622         struct ll_sb_info *sbi = ll_s2sbi(sb);
 623         ENTRY;
 624
 625         ll_close_thread_shutdown(sbi->ll_lcq);
 626
 627         lprocfs_unregister_mountpoint(sbi);
 628
 629         list_del(&sbi->ll_conn_chain);
 630
 631         /* callbacks is cleared after disconnect each target */
 632         obd_disconnect(sbi->ll_osc_exp);
 633         sbi->ll_osc_exp = NULL;
 634
 635         obd_fid_fini(sbi->ll_mdc_exp);
 636         obd_disconnect(sbi->ll_mdc_exp);
 637         sbi->ll_mdc_exp = NULL;
 638
 639         EXIT;
 640 }
 641
 642 void ll_kill_super(struct super_block *sb)
 643 {
 644         struct ll_sb_info *sbi;
 645
 646         ENTRY;
 647
 648         /* not init sb ?*/
 649         if (!(sb->s_flags & MS_ACTIVE))
 650                 return;
 651
 652         sbi = ll_s2sbi(sb);
 653         /* we need restore s_dev from changed for clustred NFS before put_super
 654          * because new kernels have cached s_dev and change sb->s_dev in
 655          * put_super not affected real removing devices */
 656         if (sbi)
 657                 sb->s_dev = sbi->ll_sdev_orig;
 658         EXIT;
 659 }
 660
 661 char *ll_read_opt(const char *opt, char *data)
 662 {
 663         char *value;
 664         char *retval;
 665         ENTRY;
 666
 667         CDEBUG(D_SUPER, "option: %s, data %s\n", opt, data);
 668         if (strncmp(opt, data, strlen(opt)))
 669                 RETURN(NULL);
 670         if ((value = strchr(data, '=')) == NULL)
 671                 RETURN(NULL);
 672
 673         value++;
 674         OBD_ALLOC(retval, strlen(value) + 1);
 675         if (!retval) {
 676                 CERROR("out of memory!\n");
 677                 RETURN(NULL);
 678         }
 679
 680         memcpy(retval, value, strlen(value)+1);
 681         CDEBUG(D_SUPER, "Assigned option: %s, value %s\n", opt, retval);
 682         RETURN(retval);
 683 }
 684
 685 static inline int ll_set_opt(const char *opt, char *data, int fl)
 686 {
 687         if (strncmp(opt, data, strlen(opt)) != 0)
 688                 return(0);
 689         else
 690                 return(fl);
 691 }
 692
 693 /* non-client-specific mount options are parsed in lmd_parse */
 694 static int ll_options(char *options, int *flags)
 695 {
 696         int tmp;
 697         char *s1 = options, *s2;
 698         ENTRY;
 699
 700         if (!options)
 701                 RETURN(0);
 702
 703         CDEBUG(D_CONFIG, "Parsing opts %s\n", options);
 704
 705         while (*s1) {
 706                 CDEBUG(D_SUPER, "next opt=%s\n", s1);
 707                 tmp = ll_set_opt("nolock", s1, LL_SBI_NOLCK);
 708                 if (tmp) {
 709                         *flags |= tmp;
 710                         goto next;
 711                 }
 712                 tmp = ll_set_opt("flock", s1, LL_SBI_FLOCK);
 713                 if (tmp) {
 714                         *flags |= tmp;
 715                         goto next;
 716                 }
 717                 tmp = ll_set_opt("localflock", s1, LL_SBI_LOCALFLOCK);
 718                 if (tmp) {
 719                         *flags |= tmp;
 720                         goto next;
 721                 }
 722                 tmp = ll_set_opt("noflock", s1, LL_SBI_FLOCK|LL_SBI_LOCALFLOCK);
 723                 if (tmp) {
 724                         *flags &= ~tmp;
 725                         goto next;
 726                 }
 727                 tmp = ll_set_opt("user_xattr", s1, LL_SBI_USER_XATTR);
 728                 if (tmp) {
 729                         *flags |= tmp;
 730                         goto next;
 731                 }
 732                 tmp = ll_set_opt("nouser_xattr", s1, LL_SBI_USER_XATTR);
 733                 if (tmp) {
 734                         *flags &= ~tmp;
 735                         goto next;
 736                 }
 737                 tmp = ll_set_opt("acl", s1, LL_SBI_ACL);
 738                 if (tmp) {
 739                         /* Ignore deprecated mount option.  The client will
 740                          * always try to mount with ACL support, whether this
 741                          * is used depends on whether server supports it. */
 742                         LCONSOLE_ERROR_MSG(0x152, "Ignoring deprecated "
 743                                                   "mount option 'acl'.\n");
 744                         goto next;
 745                 }
 746                 tmp = ll_set_opt("noacl", s1, LL_SBI_ACL);
 747                 if (tmp) {
 748                         LCONSOLE_ERROR_MSG(0x152, "Ignoring deprecated "
 749                                                   "mount option 'noacl'.\n");
 750                         goto next;
 751                 }
 752
 753                 tmp = ll_set_opt("checksum", s1, LL_SBI_DATA_CHECKSUM);
 754                 if (tmp) {
 755                         *flags |= tmp;
 756                         goto next;
 757                 }
 758                 tmp = ll_set_opt("nochecksum", s1, LL_SBI_DATA_CHECKSUM);
 759                 if (tmp) {
 760                         *flags &= ~tmp;
 761                         goto next;
 762                 }
 763
 764                 tmp = ll_set_opt("lruresize", s1, LL_SBI_LRU_RESIZE);
 765                 if (tmp) {
 766                         *flags |= tmp;
 767                         goto next;
 768                 }
 769                 tmp = ll_set_opt("nolruresize", s1, LL_SBI_LRU_RESIZE);
 770                 if (tmp) {
 771                         *flags &= ~tmp;
 772                         goto next;
 773                 }
 774                 tmp = ll_set_opt("lazystatfs", s1, LL_SBI_LAZYSTATFS);
 775                 if (tmp) {
 776                         *flags |= tmp;
 777                         goto next;
 778                 }
 779                 tmp = ll_set_opt("nolazystatfs", s1, LL_SBI_LAZYSTATFS);
 780                 if (tmp) {
 781                         *flags &= ~tmp;
 782                         goto next;
 783                 }
 784                 tmp = ll_set_opt("32bitapi", s1, LL_SBI_32BIT_API);
 785                 if (tmp) {
 786                         *flags |= tmp;
 787                         goto next;
 788                 }
 789                 LCONSOLE_ERROR_MSG(0x152, "Unknown option '%s', won't mount.\n",
 790                                    s1);
 791                 RETURN(-EINVAL);
 792
 793 next:
 794                 /* Find next opt */
 795                 s2 = strchr(s1, ',');
 796                 if (s2 == NULL)
 797                         break;
 798                 s1 = s2 + 1;
 799         }
 800         RETURN(0);
 801 }
 802
 803 void ll_lli_init(struct ll_inode_info *lli)
 804 {
 805         lli->lli_inode_magic = LLI_INODE_MAGIC;
 806         sema_init(&lli->lli_size_sem, 1);
 807         sema_init(&lli->lli_write_sem, 1);
 808         lli->lli_flags = 0;
 809         lli->lli_maxbytes = PAGE_CACHE_MAXBYTES;
 810         spin_lock_init(&lli->lli_lock);
 811         sema_init(&lli->lli_och_sem, 1);
 812         lli->lli_mds_read_och = lli->lli_mds_write_och = NULL;
 813         lli->lli_mds_exec_och = NULL;
 814         lli->lli_open_fd_read_count = lli->lli_open_fd_write_count = 0;
 815         lli->lli_open_fd_exec_count = 0;
 816         INIT_LIST_HEAD(&lli->lli_dead_list);
 817 #ifdef HAVE_CLOSE_THREAD
 818         INIT_LIST_HEAD(&lli->lli_pending_write_llaps);
 819 #endif
 820         init_rwsem(&lli->lli_truncate_rwsem);
 821 }
 822
 823 /* COMPAT_146 */
 824 #define MDCDEV "mdc_dev"
 825 static int old_lustre_process_log(struct super_block *sb, char *newprofile,
 826                                   struct config_llog_instance *cfg)
 827 {
 828         struct lustre_sb_info *lsi = s2lsi(sb);
 829         struct obd_device *obd;
 830         struct lustre_handle mdc_conn = {0, };
 831         struct obd_export *exp;
 832         char *ptr, *mdt, *profile;
 833         char niduuid[10] = "mdtnid0";
 834         class_uuid_t uuid;
 835         struct obd_uuid mdc_uuid;
 836         struct llog_ctxt *ctxt;
 837         struct obd_connect_data ocd = { 0 };
 838         lnet_nid_t nid;
 839         int i, rc = 0, recov_bk = 1, failnodes = 0;
 840         ENTRY;
 841
 842         ll_generate_random_uuid(uuid);
 843         class_uuid_unparse(uuid, &mdc_uuid);
 844         CDEBUG(D_HA, "generated uuid: %s\n", mdc_uuid.uuid);
 845
 846         /* Figure out the old mdt and profile name from new-style profile
 847            ("lustre" from "mds/lustre-client") */
 848         mdt = newprofile;
 849         profile = strchr(mdt, '/');
 850         if (profile == NULL) {
 851                 CDEBUG(D_CONFIG, "Can't find MDT name in %s\n", newprofile);
 852                 GOTO(out, rc = -EINVAL);
 853         }
 854         *profile = '\0';
 855         profile++;
 856         ptr = strrchr(profile, '-');
 857         if (ptr == NULL) {
 858                 CDEBUG(D_CONFIG, "Can't find client name in %s\n", newprofile);
 859                 GOTO(out, rc = -EINVAL);
 860         }
 861         *ptr = '\0';
 862
 863         LCONSOLE_WARN("This looks like an old mount command; I will try to "
 864                       "contact MDT '%s' for profile '%s'\n", mdt, profile);
 865
 866         /* Use nids from mount line: uml1,1@elan:uml2,2@elan:/lustre */
 867         i = 0;
 868         ptr = lsi->lsi_lmd->lmd_dev;
 869         while (class_parse_nid(ptr, &nid, &ptr) == 0) {
 870                 rc = do_lcfg(MDCDEV, nid, LCFG_ADD_UUID, niduuid, 0,0,0);
 871                 i++;
 872                 /* Stop at the first failover nid */
 873                 if (*ptr == ':')
 874                         break;
 875         }
 876         if (i == 0) {
 877                 CERROR("No valid MDT nids found.\n");
 878                 GOTO(out, rc = -EINVAL);
 879         }
 880         failnodes++;
 881
 882         rc = do_lcfg(MDCDEV, 0, LCFG_ATTACH, LUSTRE_MDC_NAME,mdc_uuid.uuid,0,0);
 883         if (rc < 0)
 884                 GOTO(out_del_uuid, rc);
 885
 886         rc = do_lcfg(MDCDEV, 0, LCFG_SETUP, mdt, niduuid, 0, 0);
 887         if (rc < 0) {
 888                 LCONSOLE_ERROR_MSG(0x153, "I couldn't establish a connection "
 889                                    "with the MDT. Check that the MDT host NID "
 890                                    "is correct and the networks are up.\n");
 891                 GOTO(out_detach, rc);
 892         }
 893
 894         obd = class_name2obd(MDCDEV);
 895         if (obd == NULL)
 896                 GOTO(out_cleanup, rc = -EINVAL);
 897
 898         /* Add any failover nids */
 899         while (*ptr == ':') {
 900                 /* New failover node */
 901                 sprintf(niduuid, "mdtnid%d", failnodes);
 902                 i = 0;
 903                 while (class_parse_nid(ptr, &nid, &ptr) == 0) {
 904                         i++;
 905                         rc = do_lcfg(MDCDEV, nid, LCFG_ADD_UUID, niduuid,0,0,0);
 906                         if (rc)
 907                                 CERROR("Add uuid for %s failed %d\n",
 908                                        libcfs_nid2str(nid), rc);
 909                         if (*ptr == ':')
 910                                 break;
 911                 }
 912                 if (i > 0) {
 913                         rc = do_lcfg(MDCDEV, 0, LCFG_ADD_CONN, niduuid, 0, 0,0);
 914                         if (rc)
 915                                 CERROR("Add conn for %s failed %d\n",
 916                                        libcfs_nid2str(nid), rc);
 917                         failnodes++;
 918                 } else {
 919                         /* at ":/fsname" */
 920                         break;
 921                 }
 922         }
 923
 924         /* Try all connections, but only once. */
 925         rc = obd_set_info_async(obd->obd_self_export,
 926                                 sizeof(KEY_INIT_RECOV_BACKUP), KEY_INIT_RECOV_BACKUP,
 927                                 sizeof(recov_bk), &recov_bk, NULL);
 928         if (rc)
 929                 GOTO(out_cleanup, rc);
 930
 931         /* If we don't have this then an ACL MDS will refuse the connection */
 932         ocd.ocd_connect_flags = OBD_CONNECT_ACL;
 933
 934         rc = obd_connect(&mdc_conn, obd, &mdc_uuid, &ocd, &exp);
 935         if (rc) {
 936                 CERROR("cannot connect to %s: rc = %d\n", mdt, rc);
 937                 GOTO(out_cleanup, rc);
 938         }
 939
 940         ctxt = llog_get_context(exp->exp_obd, LLOG_CONFIG_REPL_CTXT);
 941
 942         cfg->cfg_flags |= CFG_F_COMPAT146;
 943
 944 #if 1
 945         rc = class_config_parse_llog(ctxt, profile, cfg);
 946 #else
 947         /*
 948          * For debugging, it's useful to just dump the log
 949          */
 950         rc = class_config_dump_llog(ctxt, profile, cfg);
 951 #endif
 952         llog_ctxt_put(ctxt);
 953         switch (rc) {
 954         case 0: {
 955                 /* Set the caller's profile name to the old-style */
 956                 memcpy(newprofile, profile, strlen(profile) + 1);
 957                 break;
 958         }
 959         case -EINVAL:
 960                 LCONSOLE_ERROR_MSG(0x154, "%s: The configuration '%s' could not"
 961                                    " be read from the MDT '%s'.  Make sure this"
 962                                    " client and the MDT are running compatible "
 963                                    "versions of Lustre.\n",
 964                                    obd->obd_name, profile, mdt);
 965                 /* fall through */
 966         default:
 967                 LCONSOLE_ERROR_MSG(0x155, "%s: The configuration '%s' could not"
 968                                    " be read from the MDT '%s'.  This may be "
 969                                    "the result of communication errors between "
 970                                    "the client and the MDT, or if the MDT is "
 971                                    "not running.\n", obd->obd_name, profile,
 972                                    mdt);
 973                 break;
 974         }
 975
 976         /* We don't so much care about errors in cleaning up the config llog
 977          * connection, as we have already read the config by this point. */
 978         obd_disconnect(exp);
 979
 980 out_cleanup:
 981         do_lcfg(MDCDEV, 0, LCFG_CLEANUP, 0, 0, 0, 0);
 982
 983 out_detach:
 984         do_lcfg(MDCDEV, 0, LCFG_DETACH, 0, 0, 0, 0);
 985
 986 out_del_uuid:
 987         /* class_add_uuid adds a nid even if the same uuid exists; we might
 988            delete any copy here.  So they all better match. */
 989         for (i = 0; i < failnodes; i++) {
 990                 sprintf(niduuid, "mdtnid%d", i);
 991                 do_lcfg(MDCDEV, 0, LCFG_DEL_UUID, niduuid, 0, 0, 0);
 992         }
 993         /* class_import_put will get rid of the additional connections */
 994 out:
 995         RETURN(rc);
 996 }
 997 /* end COMPAT_146 */
 998
 999 #ifdef HAVE_NEW_BACKING_DEV_INFO
1000 static atomic_t ll_bdi_num = ATOMIC_INIT(0);
1001 #endif
1002
1003 int ll_fill_super(struct super_block *sb)
1004 {
1005         struct lustre_profile *lprof = NULL;
1006         struct lustre_sb_info *lsi = s2lsi(sb);
1007         struct ll_sb_info *sbi;
1008         char  *osc = NULL, *mdc = NULL;
1009         char  *profilenm = get_profile_name(sb);
1010         struct config_llog_instance cfg = {0, };
1011         char   ll_instance[sizeof(sb) * 2 + 1];
1012         int    err;
1013         char  *save = NULL;
1014         char  pseudo[32] = { 0 };
1015         ENTRY;
1016
1017         CDEBUG(D_VFSTRACE, "VFS Op: sb %p\n", sb);
1018
1019         cfs_module_get();
1020
1021         /* client additional sb info */
1022         lsi->lsi_llsbi = sbi = ll_init_sbi();
1023         if (!sbi) {
1024                 cfs_module_put();
1025                 RETURN(-ENOMEM);
1026         }
1027
1028         err = ll_options(lsi->lsi_lmd->lmd_opts, &sbi->ll_flags);
1029         if (err)
1030                 GOTO(out_free, err);
1031
1032         /* Generate a string unique to this super, in case some joker tries
1033            to mount the same fs at two mount points.
1034            Use the address of the super itself.*/
1035         sprintf(ll_instance, "%p", sb);
1036         cfg.cfg_instance = ll_instance;
1037         cfg.cfg_uuid = lsi->lsi_llsbi->ll_sb_uuid;
1038         cfg.cfg_sb = sb;
1039
1040         /* set up client obds */
1041         if (strchr(profilenm, '/') != NULL) /* COMPAT_146 */
1042                 err = -EINVAL; /* skip error messages, use old config code */
1043         else
1044                 err = lustre_process_log(sb, profilenm, &cfg);
1045         /* COMPAT_146 */
1046         if (err < 0) {
1047                 char *oldname;
1048                 int rc, oldnamelen;
1049                 oldnamelen = strlen(profilenm) + 1;
1050                 /* Temp storage for 1.4.6 profile name */
1051                 OBD_ALLOC(oldname, oldnamelen);
1052                 if (oldname) {
1053                         memcpy(oldname, profilenm, oldnamelen);
1054                         rc = old_lustre_process_log(sb, oldname, &cfg);
1055                         if (rc >= 0) {
1056                                 /* That worked - update the profile name
1057                                    permanently */
1058                                 err = rc;
1059                                 OBD_FREE(lsi->lsi_lmd->lmd_profile,
1060                                          strlen(lsi->lsi_lmd->lmd_profile) + 1);
1061                                 OBD_ALLOC(lsi->lsi_lmd->lmd_profile,
1062                                          strlen(oldname) + 1);
1063                                 if (!lsi->lsi_lmd->lmd_profile) {
1064                                         OBD_FREE(oldname, oldnamelen);
1065                                         GOTO(out_free, err = -ENOMEM);
1066                                 }
1067                                 memcpy(lsi->lsi_lmd->lmd_profile, oldname,
1068                                        strlen(oldname) + 1);
1069                                 profilenm = get_profile_name(sb);
1070                                 /* Don't ever try to recover the MGS */
1071                                 rc = ptlrpc_set_import_active(
1072                                         lsi->lsi_mgc->u.cli.cl_import, 0);
1073                         }
1074                         OBD_FREE(oldname, oldnamelen);
1075                 }
1076         }
1077         /* end COMPAT_146 */
1078         if (err < 0) {
1079                 CERROR("Unable to process log: %d\n", err);
1080                 GOTO(out_free, err);
1081         }
1082
1083         lprof = class_get_profile(profilenm);
1084         if (lprof == NULL) {
1085                 LCONSOLE_ERROR_MSG(0x156, "The client profile '%s' could not be"
1086                                    " read from the MGS.  Does that filesystem "
1087                                    "exist?\n", profilenm);
1088                 GOTO(out_free, err = -EINVAL);
1089         }
1090
1091         /*
1092          * The configuration for 1.8 client and 2.0 client are different.
1093          * 2.0 introduces lmv, but 1.8 directly uses mdc.
1094          * Here, we will hack to use proper name for mdc if needed.
1095          */
1096         {
1097                 char *fsname_end;
1098                 int namelen;
1099
1100                 save = lprof->lp_mdc;
1101                 fsname_end = strrchr(save, '-');
1102                 if (fsname_end) {
1103                         namelen = fsname_end - save;
1104                         if (strcmp(fsname_end, "-clilmv") == 0) {
1105                                 strncpy(pseudo, save, namelen);
1106                                 strcat(pseudo, "-MDT0000-mdc");
1107                                 lprof->lp_mdc = pseudo;
1108                                 CDEBUG(D_INFO, "1.8.x connecting to 2.0: lmv=%s"
1109                                        " new mdc=%s\n", save, pseudo);
1110                         }
1111                 }
1112         }
1113
1114         CDEBUG(D_CONFIG, "Found profile %s: mdc=%s osc=%s\n", profilenm,
1115                lprof->lp_mdc, lprof->lp_osc);
1116
1117         OBD_ALLOC(osc, strlen(lprof->lp_osc) +
1118                   strlen(ll_instance) + 2);
1119         if (!osc)
1120                 GOTO(out_free, err = -ENOMEM);
1121         sprintf(osc, "%s-%s", lprof->lp_osc, ll_instance);
1122
1123         OBD_ALLOC(mdc, strlen(lprof->lp_mdc) +
1124                   strlen(ll_instance) + 2);
1125         if (!mdc)
1126                 GOTO(out_free, err = -ENOMEM);
1127         sprintf(mdc, "%s-%s", lprof->lp_mdc, ll_instance);
1128
1129         /* connections, registrations, sb setup */
1130         err = client_common_fill_super(sb, mdc, osc);
1131         if (err)
1132                 GOTO(out_free, err);
1133
1134 #ifdef HAVE_BDI_INIT
1135         err = bdi_init(&lsi->bdi);
1136         if (err)
1137                 GOTO(out_free, err);
1138 #endif
1139 #ifdef HAVE_NEW_BACKING_DEV_INFO
1140         lsi->bdi.name = "lustre";
1141         lsi->bdi.capabilities = BDI_CAP_MAP_COPY;
1142         err = bdi_register(&lsi->bdi, NULL, "lustre-%d",
1143                                 atomic_inc_return(&ll_bdi_num));
1144         sb->s_bdi = &lsi->bdi;
1145 #endif
1146
1147 out_free:
1148         if (save && lprof)
1149                 lprof->lp_mdc = save;
1150         if (mdc)
1151                 OBD_FREE(mdc, strlen(mdc) + 1);
1152         if (osc)
1153                 OBD_FREE(osc, strlen(osc) + 1);
1154         if (err)
1155                 ll_put_super(sb);
1156         else
1157                 LCONSOLE_WARN("Client %s has started\n", profilenm);
1158
1159         RETURN(err);
1160 } /* ll_fill_super */
1161
1162
1163 void ll_put_super(struct super_block *sb)
1164 {
1165         struct config_llog_instance cfg;
1166         char   ll_instance[sizeof(sb) * 2 + 1];
1167         struct obd_device *obd;
1168         struct lustre_sb_info *lsi = s2lsi(sb);
1169         struct ll_sb_info *sbi = ll_s2sbi(sb);
1170         char *profilenm = get_profile_name(sb);
1171         int force = 1, next;
1172         ENTRY;
1173
1174         CDEBUG(D_VFSTRACE, "VFS Op: sb %p - %s\n", sb, profilenm);
1175
1176         sprintf(ll_instance, "%p", sb);
1177         cfg.cfg_instance = ll_instance;
1178         lustre_end_log(sb, NULL, &cfg);
1179
1180         if (sbi->ll_mdc_exp) {
1181                 obd = class_exp2obd(sbi->ll_mdc_exp);
1182                 if (obd)
1183                         force = obd->obd_force;
1184         }
1185
1186         /* We need to set force before the lov_disconnect in
1187            lustre_common_put_super, since l_d cleans up osc's as well. */
1188         if (force) {
1189                 next = 0;
1190                 while ((obd = class_devices_in_group(&sbi->ll_sb_uuid,
1191                                                      &next)) != NULL) {
1192                         obd->obd_force = force;
1193                 }
1194         }
1195
1196         if (sbi->ll_lcq) {
1197                 /* Only if client_common_fill_super succeeded */
1198                 client_common_put_super(sb);
1199         }
1200
1201         next = 0;
1202         while ((obd = class_devices_in_group(&sbi->ll_sb_uuid, &next)) !=NULL) {
1203                 class_manual_cleanup(obd);
1204         }
1205
1206         if (profilenm)
1207                 class_del_profile(profilenm);
1208
1209 #ifdef HAVE_NEW_BACKING_DEV_INFO
1210         if (lsi->bdi.wb_cnt > 0) {
1211 #endif
1212 #ifdef HAVE_BDI_INIT
1213                 bdi_destroy(&lsi->bdi);
1214 #endif
1215 #ifdef HAVE_NEW_BACKING_DEV_INFO
1216         }
1217 #endif
1218
1219         ll_free_sbi(sb);
1220         lsi->lsi_llsbi = NULL;
1221
1222         lustre_common_put_super(sb);
1223
1224         LCONSOLE_WARN("client %s umount complete\n", ll_instance);
1225
1226         cfs_module_put();
1227
1228         EXIT;
1229 } /* client_put_super */
1230
1231 int ll_shrink_cache(int nr_to_scan, SHRINKER_MASK_T gfp_mask)
1232 {
1233         struct ll_sb_info *sbi;
1234         int count = 0;
1235
1236         if ((nr_to_scan != 0) && !(gfp_mask & __GFP_FS))
1237                 return -1;
1238
1239         /* don't race with umount */
1240         down_read(&ll_sb_sem);
1241         list_for_each_entry(sbi, &ll_super_blocks, ll_list)
1242                 count += llap_shrink_cache(sbi, nr_to_scan);
1243         up_read(&ll_sb_sem);
1244
1245         return count;
1246 }
1247
1248 struct inode *ll_inode_from_lock(struct ldlm_lock *lock)
1249 {
1250         struct inode *inode = NULL;
1251         /* NOTE: we depend on atomic igrab() -bzzz */
1252         lock_res_and_lock(lock);
1253         if (lock->l_ast_data) {
1254                 struct ll_inode_info *lli = ll_i2info(lock->l_ast_data);
1255                 if (lli->lli_inode_magic == LLI_INODE_MAGIC) {
1256                         inode = igrab(lock->l_ast_data);
1257                 } else {
1258                         inode = lock->l_ast_data;
1259                         LDLM_DEBUG_LIMIT(inode->i_state & I_FREEING ?  D_INFO :
1260                                          D_WARNING, lock, "l_ast_data %p is "
1261                                          "bogus: magic %08x", lock->l_ast_data,
1262                                          lli->lli_inode_magic);
1263                         inode = NULL;
1264                 }
1265         }
1266         unlock_res_and_lock(lock);
1267         return inode;
1268 }
1269
1270 static int null_if_equal(struct ldlm_lock *lock, void *data)
1271 {
1272         if (data == lock->l_ast_data) {
1273                 lock->l_ast_data = NULL;
1274
1275                 if (lock->l_req_mode != lock->l_granted_mode)
1276                         LDLM_ERROR(lock,"clearing inode with ungranted lock");
1277         }
1278
1279         return LDLM_ITER_CONTINUE;
1280 }
1281
1282 void ll_clear_inode(struct inode *inode)
1283 {
1284         struct ll_fid fid;
1285         struct ll_inode_info *lli = ll_i2info(inode);
1286         struct ll_sb_info *sbi = ll_i2sbi(inode);
1287         ENTRY;
1288
1289         CDEBUG(D_VFSTRACE, "VFS Op:inode=%lu/%u(%p)\n", inode->i_ino,
1290                inode->i_generation, inode);
1291
1292         if (S_ISDIR(inode->i_mode)) {
1293                 /* these should have been cleared in ll_file_release */
1294                 LASSERT(lli->lli_sai == NULL);
1295                 LASSERT(lli->lli_opendir_key == NULL);
1296                 LASSERT(lli->lli_opendir_pid == 0);
1297         }
1298
1299         ll_inode2fid(&fid, inode);
1300         clear_bit(LLI_F_HAVE_MDS_SIZE_LOCK, &lli->lli_flags);
1301         mdc_change_cbdata(sbi->ll_mdc_exp, &fid, null_if_equal, inode);
1302
1303         LASSERT(!lli->lli_open_fd_write_count);
1304         LASSERT(!lli->lli_open_fd_read_count);
1305         LASSERT(!lli->lli_open_fd_exec_count);
1306
1307         if (lli->lli_mds_write_och)
1308                 ll_mdc_real_close(inode, FMODE_WRITE);
1309         if (lli->lli_mds_exec_och) {
1310                 if (!FMODE_EXEC)
1311                         CERROR("No FMODE exec, bug exec och is present for "
1312                                "inode %ld\n", inode->i_ino);
1313                 ll_mdc_real_close(inode, FMODE_EXEC);
1314         }
1315         if (lli->lli_mds_read_och)
1316                 ll_mdc_real_close(inode, FMODE_READ);
1317
1318
1319         if (lli->lli_smd) {
1320                 obd_change_cbdata(sbi->ll_osc_exp, lli->lli_smd,
1321                                   null_if_equal, inode);
1322
1323                 obd_free_memmd(sbi->ll_osc_exp, &lli->lli_smd);
1324                 lli->lli_smd = NULL;
1325         }
1326
1327         if (lli->lli_symlink_name) {
1328                 OBD_FREE(lli->lli_symlink_name,
1329                          strlen(lli->lli_symlink_name) + 1);
1330                 lli->lli_symlink_name = NULL;
1331         }
1332
1333 #ifdef CONFIG_FS_POSIX_ACL
1334         if (lli->lli_posix_acl) {
1335                 LASSERT(atomic_read(&lli->lli_posix_acl->a_refcount) == 1);
1336                 posix_acl_release(lli->lli_posix_acl);
1337                 lli->lli_posix_acl = NULL;
1338         }
1339 #endif
1340
1341         lli->lli_inode_magic = LLI_INODE_DEAD;
1342
1343         EXIT;
1344 }
1345 static int ll_setattr_do_truncate(struct inode *inode, loff_t new_size)
1346 {
1347         struct ll_sb_info *sbi = ll_i2sbi(inode);
1348         struct ll_inode_info *lli = ll_i2info(inode);
1349         struct lov_stripe_md *lsm = lli->lli_smd;
1350         int rc;
1351         ldlm_policy_data_t policy = { .l_extent = {new_size,
1352                                                    OBD_OBJECT_EOF } };
1353         struct lustre_handle lockh = { 0 };
1354         int local_lock = 1; /* 0 - no local lock;
1355                              * 1 - lock taken by lock_extent;
1356                              * 2 - by obd_match*/
1357         int ast_flags;
1358         ENTRY;
1359
1360         UNLOCK_INODE_MUTEX(inode);
1361         UP_WRITE_I_ALLOC_SEM(inode);
1362
1363         down_write(&lli->lli_truncate_rwsem);
1364         if (sbi->ll_lockless_truncate_enable &&
1365             (sbi->ll_lco.lco_flags & OBD_CONNECT_TRUNCLOCK)) {
1366                 int n_matches = 0;
1367
1368                 ast_flags = LDLM_FL_BLOCK_GRANTED;
1369                 rc = obd_match(sbi->ll_osc_exp, lsm, LDLM_EXTENT,
1370                                &policy, LCK_PW, &ast_flags, inode, &lockh,
1371                                &n_matches);
1372                 if (rc > 0) {
1373                         local_lock = 2;
1374                         rc = 0;
1375                 } else {
1376                         /* clear the lock handle as it not cleared
1377                          * by obd_match if no matched lock found */
1378                         lockh.cookie = 0;
1379                         if (rc == 0 && n_matches == 0) {
1380                                 local_lock = 0;
1381                                 rc = ll_file_punch(inode, new_size, 1);
1382                         }
1383                 }
1384         }
1385         if (local_lock == 1) {
1386                 /* XXX when we fix the AST intents to pass the discard-range
1387                  * XXX extent, make ast_flags always LDLM_AST_DISCARD_DATA
1388                  * XXX here. */
1389                 ast_flags = (new_size == 0) ? LDLM_AST_DISCARD_DATA : 0;
1390                 rc = ll_extent_lock(NULL, inode, lsm, LCK_PW, &policy,
1391                                     &lockh, ast_flags);
1392                 if (unlikely(rc != 0))
1393                         local_lock = 0;
1394         }
1395
1396         LOCK_INODE_MUTEX(inode);
1397         DOWN_WRITE_I_ALLOC_SEM(inode);
1398         if (likely(rc == 0)) {
1399                 /* Only ll_inode_size_lock is taken at this level.
1400                  * lov_stripe_lock() is grabbed by ll_truncate() only over
1401                  * call to obd_adjust_kms().  If vmtruncate returns 0, then
1402                  * ll_truncate dropped ll_inode_size_lock() */
1403                 ll_inode_size_lock(inode, 0);
1404                 if (!local_lock)
1405                         set_bit(LLI_F_SRVLOCK, &lli->lli_flags);
1406                 rc = vmtruncate(inode, new_size);
1407                 clear_bit(LLI_F_SRVLOCK, &lli->lli_flags);
1408                 if (rc != 0) {
1409                         LASSERT(SEM_COUNT(&lli->lli_size_sem) <= 0);
1410                         ll_inode_size_unlock(inode, 0);
1411                 }
1412         }
1413         if (local_lock) {
1414                 int err;
1415
1416                 err = (local_lock == 2) ?
1417                         obd_cancel(sbi->ll_osc_exp, lsm, LCK_PW, &lockh, 0, 0):
1418                         ll_extent_unlock(NULL, inode, lsm, LCK_PW, &lockh);
1419                 if (unlikely(err != 0)){
1420                         CERROR("extent unlock failed: err=%d,"
1421                                " unlock method =%d\n", err, local_lock);
1422                         if (rc == 0)
1423                                 rc = err;
1424                 }
1425         }
1426         up_write(&lli->lli_truncate_rwsem);
1427         RETURN(rc);
1428 }
1429
1430 /* If this inode has objects allocated to it (lsm != NULL), then the OST
1431  * object(s) determine the file size and mtime.  Otherwise, the MDS will
1432  * keep these values until such a time that objects are allocated for it.
1433  * We do the MDS operations first, as it is checking permissions for us.
1434  * We don't to the MDS RPC if there is nothing that we want to store there,
1435  * otherwise there is no harm in updating mtime/atime on the MDS if we are
1436  * going to do an RPC anyways.
1437  *
1438  * If we are doing a truncate, we will send the mtime and ctime updates
1439  * to the OST with the punch RPC, otherwise we do an explicit setattr RPC.
1440  * I don't believe it is possible to get e.g. ATTR_MTIME_SET and ATTR_SIZE
1441  * at the same time.
1442  */
1443 int ll_setattr_raw(struct inode *inode, struct iattr *attr)
1444 {
1445         struct ll_inode_info *lli = ll_i2info(inode);
1446         struct lov_stripe_md *lsm = lli->lli_smd;
1447         struct ll_sb_info *sbi = ll_i2sbi(inode);
1448         struct ptlrpc_request *request = NULL;
1449         struct mdc_op_data *op_data;
1450         struct lustre_md md;
1451         int ia_valid = attr->ia_valid;
1452         int rc = 0;
1453         ENTRY;
1454
1455         CDEBUG(D_VFSTRACE, "VFS Op:inode=%lu valid %x\n", inode->i_ino,
1456                attr->ia_valid);
1457         ll_stats_ops_tally(ll_i2sbi(inode), LPROC_LL_SETATTR, 1);
1458
1459         if (ia_valid & ATTR_SIZE) {
1460                 if (attr->ia_size > ll_file_maxbytes(inode)) {
1461                         CDEBUG(D_INODE, "file too large %llu > "LPU64"\n",
1462                                attr->ia_size, ll_file_maxbytes(inode));
1463                         RETURN(-EFBIG);
1464                 }
1465
1466                 attr->ia_valid |= ATTR_MTIME | ATTR_CTIME;
1467         }
1468
1469         /* POSIX: check before ATTR_*TIME_SET set (from inode_change_ok) */
1470         if (ia_valid & (ATTR_MTIME_SET | ATTR_ATIME_SET)) {
1471                 if (cfs_curproc_fsuid() != inode->i_uid &&
1472                     !cfs_capable(CFS_CAP_FOWNER))
1473                         RETURN(-EPERM);
1474         }
1475
1476         /* We mark all of the fields "set" so MDS/OST does not re-set them */
1477         if (attr->ia_valid & ATTR_CTIME) {
1478                 attr->ia_ctime = CURRENT_TIME;
1479                 attr->ia_valid |= ATTR_CTIME_SET;
1480         }
1481         if (!(ia_valid & ATTR_ATIME_SET) && (attr->ia_valid & ATTR_ATIME)) {
1482                 attr->ia_atime = CURRENT_TIME;
1483                 attr->ia_valid |= ATTR_ATIME_SET;
1484         }
1485         if (!(ia_valid & ATTR_MTIME_SET) && (attr->ia_valid & ATTR_MTIME)) {
1486                 attr->ia_mtime = CURRENT_TIME;
1487                 attr->ia_valid |= ATTR_MTIME_SET;
1488         }
1489
1490         if (attr->ia_valid & (ATTR_MTIME | ATTR_CTIME))
1491                 CDEBUG(D_INODE, "setting mtime %lu, ctime %lu, now = %lu\n",
1492                        LTIME_S(attr->ia_mtime), LTIME_S(attr->ia_ctime),
1493                        CURRENT_SECONDS);
1494
1495         /* NB: ATTR_SIZE will only be set after this point if the size
1496          * resides on the MDS, ie, this file has no objects. */
1497         if (lsm)
1498                 attr->ia_valid &= ~ATTR_SIZE;
1499
1500         OBD_ALLOC_PTR(op_data);
1501         if (NULL == op_data)
1502                 RETURN(-ENOMEM);
1503         /* We always do an MDS RPC, even if we're only changing the size;
1504          * only the MDS knows whether truncate() should fail with -ETXTBUSY */
1505         ll_prepare_mdc_op_data(op_data, inode, NULL, NULL, 0, 0, NULL);
1506
1507         rc = mdc_setattr(sbi->ll_mdc_exp, op_data,
1508                          attr, NULL, 0, NULL, 0, &request);
1509
1510         if (rc) {
1511                 ptlrpc_req_finished(request);
1512                 if (rc == -ENOENT) {
1513                         inode->i_nlink = 0;
1514                         /* Unlinked special device node?  Or just a race?
1515                          * Pretend we done everything. */
1516                         if (!S_ISREG(inode->i_mode) &&
1517                             !S_ISDIR(inode->i_mode))
1518                                 rc = inode_setattr(inode, attr);
1519                 } else if (rc != -EPERM && rc != -EACCES && rc != -ETXTBSY)
1520                         CERROR("mdc_setattr fails: rc = %d\n", rc);
1521                 OBD_FREE_PTR(op_data);
1522                 RETURN(rc);
1523         }
1524         OBD_FREE_PTR(op_data);
1525
1526         rc = mdc_req2lustre_md(request, REPLY_REC_OFF, sbi->ll_osc_exp, &md);
1527         if (rc) {
1528                 ptlrpc_req_finished(request);
1529                 RETURN(rc);
1530         }
1531
1532         /* We call inode_setattr to adjust timestamps.
1533          * If there is at least some data in file, we cleared ATTR_SIZE above to
1534          * avoid invoking vmtruncate, otherwise it is important to call
1535          * vmtruncate in inode_setattr to update inode->i_size (bug 6196) */
1536         rc = inode_setattr(inode, attr);
1537
1538         ll_update_inode(inode, &md);
1539         ptlrpc_req_finished(request);
1540
1541         if (!lsm || !S_ISREG(inode->i_mode)) {
1542                 CDEBUG(D_INODE, "no lsm: not setting attrs on OST\n");
1543                 RETURN(rc);
1544         }
1545
1546         /* We really need to get our PW lock before we change inode->i_size.
1547          * If we don't we can race with other i_size updaters on our node, like
1548          * ll_file_read.  We can also race with i_size propogation to other
1549          * nodes through dirtying and writeback of final cached pages.  This
1550          * last one is especially bad for racing o_append users on other
1551          * nodes. */
1552         if (ia_valid & ATTR_SIZE) {
1553                 rc = ll_setattr_do_truncate(inode, attr->ia_size);
1554         } else if (ia_valid & (ATTR_MTIME | ATTR_MTIME_SET)) {
1555                 struct obd_info *oinfo;
1556                 struct obdo *oa;
1557                 struct lustre_handle lockh = { 0 };
1558                 obd_valid valid;
1559
1560                 OBD_ALLOC_PTR(oinfo);
1561                 if (NULL == oinfo)
1562                         RETURN(-ENOMEM);
1563
1564                 CDEBUG(D_INODE, "set mtime on OST inode %lu to %lu\n",
1565                        inode->i_ino, LTIME_S(attr->ia_mtime));
1566
1567                 OBDO_ALLOC(oa);
1568                 if (oa) {
1569                         oa->o_id = lsm->lsm_object_id;
1570                         oa->o_gr = lsm->lsm_object_gr;
1571                         oa->o_valid = OBD_MD_FLID | OBD_MD_FLGROUP;
1572
1573                         valid = OBD_MD_FLTYPE | OBD_MD_FLFID | OBD_MD_FLGENER;
1574
1575                         if (LTIME_S(attr->ia_mtime) < LTIME_S(attr->ia_ctime)){
1576                                 struct ost_lvb xtimes;
1577
1578                                 UNLOCK_INODE_MUTEX(inode);
1579                                 /* setting mtime to past is performed under PW
1580                                  * EOF extent lock */
1581                                 oinfo->oi_policy.l_extent.start = 0;
1582                                 oinfo->oi_policy.l_extent.end = OBD_OBJECT_EOF;
1583                                 rc = ll_extent_lock(NULL, inode, lsm, LCK_PW,
1584                                                     &oinfo->oi_policy,
1585                                                     &lockh, 0);
1586                                 LOCK_INODE_MUTEX(inode);
1587                                 if (rc) {
1588                                         OBDO_FREE(oa);
1589                                         OBD_FREE_PTR(oinfo);
1590                                         RETURN(rc);
1591                                 }
1592
1593                                 /* setattr under locks
1594                                  *
1595                                  * 1. restore inode's timestamps which
1596                                  * are about to be set as long as
1597                                  * concurrent stat (via
1598                                  * ll_glimpse_size) might bring
1599                                  * out-of-date ones
1600                                  *
1601                                  * 2. update lsm so that next stat
1602                                  * (via ll_glimpse_size) could get
1603                                  * correct values in lsm */
1604                                 lov_stripe_lock(lli->lli_smd);
1605                                 if (ia_valid & ATTR_ATIME) {
1606                                         LTIME_S(inode->i_atime) =
1607                                                 xtimes.lvb_atime =
1608                                                 LTIME_S(attr->ia_atime);
1609                                         valid |= OBD_MD_FLATIME;
1610                                 }
1611                                 if (ia_valid & ATTR_MTIME) {
1612                                         LTIME_S(inode->i_mtime) =
1613                                                 xtimes.lvb_mtime =
1614                                                 LTIME_S(attr->ia_mtime);
1615                                         valid |= OBD_MD_FLMTIME;
1616                                 }
1617                                 if (ia_valid & ATTR_CTIME) {
1618                                         LTIME_S(inode->i_ctime) =
1619                                                 xtimes.lvb_ctime =
1620                                                 LTIME_S(attr->ia_ctime);
1621                                         valid |= OBD_MD_FLCTIME;
1622                                 }
1623
1624                                 obd_update_lvb(ll_i2obdexp(inode), lli->lli_smd,
1625                                                &xtimes, valid);
1626                                 lov_stripe_unlock(lli->lli_smd);
1627                         } else {
1628                                 /* lockless setattr
1629                                  *
1630                                  * 1. do not use inode's timestamps
1631                                  * because concurrent stat might fill
1632                                  * the inode with out-of-date times,
1633                                  * send values from attr instead
1634                                  *
1635                                  * 2.do no update lsm, as long as stat
1636                                  * (via ll_glimpse_size) will bring
1637                                  * attributes from osts anyway */
1638                                 if (ia_valid & ATTR_ATIME) {
1639                                         oa->o_atime = LTIME_S(attr->ia_atime);
1640                                         oa->o_valid |= OBD_MD_FLATIME;
1641                                 }
1642                                 if (ia_valid & ATTR_MTIME) {
1643                                         oa->o_mtime = LTIME_S(attr->ia_mtime);
1644                                         oa->o_valid |= OBD_MD_FLMTIME;
1645                                 }
1646                                 if (ia_valid & ATTR_CTIME) {
1647                                         oa->o_ctime = LTIME_S(attr->ia_ctime);
1648                                         oa->o_valid |= OBD_MD_FLCTIME;
1649                                 }
1650                         }
1651
1652                         obdo_from_inode(oa, inode, valid);
1653
1654                         oinfo->oi_oa = oa;
1655                         oinfo->oi_md = lsm;
1656
1657                         rc = obd_setattr_rqset(sbi->ll_osc_exp, oinfo, NULL);
1658                         if (rc)
1659                                 CERROR("obd_setattr_async fails: rc=%d\n", rc);
1660
1661                         if (LTIME_S(attr->ia_mtime) < LTIME_S(attr->ia_ctime)){
1662                                 int err;
1663
1664                                 err = ll_extent_unlock(NULL, inode, lsm,
1665                                                        LCK_PW, &lockh);
1666                                 if (unlikely(err != 0)) {
1667                                         CERROR("extent unlock failed: "
1668                                                "err=%d\n", err);
1669                                         if (rc == 0)
1670                                                 rc = err;
1671                                 }
1672                         }
1673                         OBDO_FREE(oa);
1674                 } else {
1675                         rc = -ENOMEM;
1676                 }
1677                 OBD_FREE_PTR(oinfo);
1678         }
1679         RETURN(rc);
1680 }
1681
1682 int ll_setattr(struct dentry *de, struct iattr *attr)
1683 {
1684         int mode = de->d_inode->i_mode;
1685
1686         if ((attr->ia_valid & (ATTR_CTIME|ATTR_SIZE|ATTR_MODE)) ==
1687             (ATTR_CTIME|ATTR_SIZE|ATTR_MODE))
1688                 attr->ia_valid |= MDS_OPEN_OWNEROVERRIDE;
1689         if ((attr->ia_valid & (ATTR_MODE|ATTR_FORCE|ATTR_SIZE)) ==
1690             (ATTR_SIZE|ATTR_MODE)) {
1691                 if (((mode & S_ISUID) && (!(attr->ia_mode & S_ISUID))) ||
1692                     ((mode & S_ISGID) && (mode & S_IXGRP) &&
1693                     (!(attr->ia_mode & S_ISGID))))
1694                         attr->ia_valid |= ATTR_FORCE;
1695         }
1696
1697         if ((mode & S_ISUID) && S_ISREG(mode) &&
1698             !(attr->ia_mode & S_ISUID))
1699                 attr->ia_valid |= ATTR_KILL_SUID;
1700
1701         if (((mode & (S_ISGID|S_IXGRP)) == (S_ISGID|S_IXGRP)) &&
1702             S_ISREG(mode) && !(attr->ia_mode & S_ISGID))
1703                 attr->ia_valid |= ATTR_KILL_SGID;
1704
1705         return ll_setattr_raw(de->d_inode, attr);
1706 }
1707
1708 int ll_statfs_internal(struct super_block *sb, struct obd_statfs *osfs,
1709                        __u64 max_age, __u32 flags)
1710 {
1711         struct ll_sb_info *sbi = ll_s2sbi(sb);
1712         struct obd_statfs obd_osfs;
1713         int rc;
1714         ENTRY;
1715
1716         rc = obd_statfs(class_exp2obd(sbi->ll_mdc_exp), osfs, max_age, flags);
1717         if (rc) {
1718                 CERROR("mdc_statfs fails: rc = %d\n", rc);
1719                 RETURN(rc);
1720         }
1721
1722         osfs->os_type = sb->s_magic;
1723
1724         CDEBUG(D_SUPER, "MDC blocks "LPU64"/"LPU64" objects "LPU64"/"LPU64"\n",
1725                osfs->os_bavail, osfs->os_blocks, osfs->os_ffree,osfs->os_files);
1726
1727         if (sbi->ll_flags & LL_SBI_LAZYSTATFS)
1728                 flags |= OBD_STATFS_NODELAY;
1729
1730         rc = obd_statfs_rqset(class_exp2obd(sbi->ll_osc_exp),
1731                               &obd_osfs, max_age, flags);
1732         if (rc) {
1733                 CERROR("obd_statfs fails: rc = %d\n", rc);
1734                 RETURN(rc);
1735         }
1736
1737         CDEBUG(D_SUPER, "OSC blocks "LPU64"/"LPU64" objects "LPU64"/"LPU64"\n",
1738                obd_osfs.os_bavail, obd_osfs.os_blocks, obd_osfs.os_ffree,
1739                obd_osfs.os_files);
1740
1741         osfs->os_bsize = obd_osfs.os_bsize;
1742         osfs->os_blocks = obd_osfs.os_blocks;
1743         osfs->os_bfree = obd_osfs.os_bfree;
1744         osfs->os_bavail = obd_osfs.os_bavail;
1745
1746         /* If we don't have as many objects free on the OST as inodes
1747          * on the MDS, we reduce the total number of inodes to
1748          * compensate, so that the "inodes in use" number is correct.
1749          */
1750         if (obd_osfs.os_ffree < osfs->os_ffree) {
1751                 osfs->os_files = (osfs->os_files - osfs->os_ffree) +
1752                         obd_osfs.os_ffree;
1753                 osfs->os_ffree = obd_osfs.os_ffree;
1754         }
1755
1756         RETURN(rc);
1757 }
1758 #ifndef HAVE_STATFS_DENTRY_PARAM
1759 int ll_statfs(struct super_block *sb, struct kstatfs *sfs)
1760 {
1761 #else
1762 int ll_statfs(struct dentry *de, struct kstatfs *sfs)
1763 {
1764         struct super_block *sb = de->d_sb;
1765 #endif
1766         struct obd_statfs osfs;
1767         int rc;
1768
1769         CDEBUG(D_VFSTRACE, "VFS Op: at "LPU64" jiffies\n", get_jiffies_64());
1770         ll_stats_ops_tally(ll_s2sbi(sb), LPROC_LL_STAFS, 1);
1771
1772         /* Some amount of caching on the client is allowed */
1773         rc = ll_statfs_internal(sb, &osfs,
1774                                 cfs_time_shift_64(-OBD_STATFS_CACHE_SECONDS),
1775                                 0);
1776         if (rc)
1777                 return rc;
1778
1779         statfs_unpack(sfs, &osfs);
1780
1781         /* We need to downshift for all 32-bit kernels, because we can't
1782          * tell if the kernel is being called via sys_statfs64() or not.
1783          * Stop before overflowing f_bsize - in which case it is better
1784          * to just risk EOVERFLOW if caller is using old sys_statfs(). */
1785         if (sizeof(long) < 8) {
1786                 while (osfs.os_blocks > ~0UL && sfs->f_bsize < 0x40000000) {
1787                         sfs->f_bsize <<= 1;
1788
1789                         osfs.os_blocks >>= 1;
1790                         osfs.os_bfree >>= 1;
1791                         osfs.os_bavail >>= 1;
1792                 }
1793         }
1794
1795         sfs->f_blocks = osfs.os_blocks;
1796         sfs->f_bfree = osfs.os_bfree;
1797         sfs->f_bavail = osfs.os_bavail;
1798
1799         return 0;
1800 }
1801
1802 void ll_inode_size_lock(struct inode *inode, int lock_lsm)
1803 {
1804         struct ll_inode_info *lli;
1805         struct lov_stripe_md *lsm;
1806
1807         lli = ll_i2info(inode);
1808         LASSERT(lli->lli_size_sem_owner != current);
1809         down(&lli->lli_size_sem);
1810         LASSERT(lli->lli_size_sem_owner == NULL);
1811         lli->lli_size_sem_owner = current;
1812         lsm = lli->lli_smd;
1813         LASSERTF(lsm != NULL || lock_lsm == 0, "lsm %p, lock_lsm %d\n",
1814                  lsm, lock_lsm);
1815         if (lock_lsm)
1816                 lov_stripe_lock(lsm);
1817 }
1818
1819 void ll_inode_size_unlock(struct inode *inode, int unlock_lsm)
1820 {
1821         struct ll_inode_info *lli;
1822         struct lov_stripe_md *lsm;
1823
1824         lli = ll_i2info(inode);
1825         lsm = lli->lli_smd;
1826         LASSERTF(lsm != NULL || unlock_lsm == 0, "lsm %p, lock_lsm %d\n",
1827                  lsm, unlock_lsm);
1828         if (unlock_lsm)
1829                 lov_stripe_unlock(lsm);
1830         LASSERT(lli->lli_size_sem_owner == current);
1831         lli->lli_size_sem_owner = NULL;
1832         up(&lli->lli_size_sem);
1833 }
1834
1835 static void ll_replace_lsm(struct inode *inode, struct lov_stripe_md *lsm)
1836 {
1837         struct ll_inode_info *lli = ll_i2info(inode);
1838
1839         dump_lsm(D_INODE, lsm);
1840         dump_lsm(D_INODE, lli->lli_smd);
1841         LASSERTF(lsm->lsm_magic == LOV_MAGIC_JOIN,
1842                  "lsm must be joined lsm %p\n", lsm);
1843         obd_free_memmd(ll_i2obdexp(inode), &lli->lli_smd);
1844         CDEBUG(D_INODE, "replace lsm %p to lli_smd %p for inode %lu%u(%p)\n",
1845                lsm, lli->lli_smd, inode->i_ino, inode->i_generation, inode);
1846         lli->lli_smd = lsm;
1847         lli->lli_maxbytes = lsm->lsm_maxbytes;
1848         if (lli->lli_maxbytes > PAGE_CACHE_MAXBYTES)
1849                 lli->lli_maxbytes = PAGE_CACHE_MAXBYTES;
1850 }
1851
1852 void ll_update_inode(struct inode *inode, struct lustre_md *md)
1853 {
1854         struct ll_inode_info *lli = ll_i2info(inode);
1855         struct mds_body *body = md->body;
1856         struct lov_stripe_md *lsm = md->lsm;
1857         struct ll_sb_info *sbi = ll_i2sbi(inode);
1858         ENTRY;
1859
1860         CDEBUG(D_INODE, "body->valid = "LPX64"\n", body->valid);
1861
1862         LASSERT ((lsm != NULL) == ((body->valid & OBD_MD_FLEASIZE) != 0));
1863         if (lsm != NULL) {
1864                 spin_lock(&lli->lli_lock);
1865                 if (lli->lli_smd == NULL) {
1866                         if (lsm->lsm_magic != LOV_MAGIC_V1 &&
1867                             lsm->lsm_magic != LOV_MAGIC_V3 &&
1868                             lsm->lsm_magic != LOV_MAGIC_JOIN) {
1869                                 dump_lsm(D_ERROR, lsm);
1870                                 LBUG();
1871                         }
1872                         CDEBUG(D_INODE, "adding lsm %p to inode %lu/%u(%p)\n",
1873                                lsm, inode->i_ino, inode->i_generation, inode);
1874                         /* ll_inode_size_lock() requires it is only called
1875                          * with lli_smd != NULL or lock_lsm == 0 or we can
1876                          * race between lock/unlock.  bug 9547 */
1877                         lli->lli_smd = lsm;
1878                         spin_unlock(&lli->lli_lock);
1879                         lli->lli_maxbytes = lsm->lsm_maxbytes;
1880                         if (lli->lli_maxbytes > PAGE_CACHE_MAXBYTES)
1881                                 lli->lli_maxbytes = PAGE_CACHE_MAXBYTES;
1882                 } else {
1883                         spin_unlock(&lli->lli_lock);
1884                         if ((lli->lli_smd->lsm_magic == lsm->lsm_magic ||
1885                              (lli->lli_smd->lsm_magic == LOV_MAGIC_V3 &&
1886                               lsm->lsm_magic == LOV_MAGIC_V1) ||
1887                              (lli->lli_smd->lsm_magic == LOV_MAGIC_V1 &&
1888                               lsm->lsm_magic == LOV_MAGIC_V3)) &&
1889                             lli->lli_smd->lsm_stripe_count ==
1890                                         lsm->lsm_stripe_count) {
1891                                 /* The MDS can suddenly change the magic to
1892                                  * v1/v3 if it has been upgraded/downgraded to a
1893                                  * version that does/doesn't support OST pools.
1894                                  * In this case, we consider that the cached lsm
1895                                  * is equivalent to the new one and keep it in
1896                                  * place until the inode is evicted from the
1897                                  * cache */
1898                                 if (lov_stripe_md_cmp(lli->lli_smd, lsm)) {
1899                                         CERROR("lsm mismatch for inode %ld\n",
1900                                                 inode->i_ino);
1901                                         CERROR("lli_smd:\n");
1902                                         dump_lsm(D_ERROR, lli->lli_smd);
1903                                         CERROR("lsm:\n");
1904                                         dump_lsm(D_ERROR, lsm);
1905                                         LBUG();
1906                                 }
1907                         } else
1908                                 ll_replace_lsm(inode, lsm);
1909                 }
1910                 if (lli->lli_smd != lsm)
1911                         obd_free_memmd(ll_i2obdexp(inode), &lsm);
1912         }
1913
1914 #ifdef CONFIG_FS_POSIX_ACL
1915         LASSERT(!md->posix_acl || (body->valid & OBD_MD_FLACL));
1916         if (body->valid & OBD_MD_FLACL) {
1917                 spin_lock(&lli->lli_lock);
1918                 if (lli->lli_posix_acl)
1919                         posix_acl_release(lli->lli_posix_acl);
1920                 lli->lli_posix_acl = md->posix_acl;
1921                 spin_unlock(&lli->lli_lock);
1922         }
1923 #endif
1924
1925         inode->i_ino = ll_fid_build_ino(&body->fid1);
1926         inode->i_generation = ll_fid_build_gen(sbi, &body->fid1);
1927         *ll_inode_lu_fid(inode) = *((struct lu_fid*)&md->body->fid1);
1928
1929         if (body->valid & OBD_MD_FLATIME) {
1930                 if (body->atime > LTIME_S(inode->i_atime))
1931                         LTIME_S(inode->i_atime) = body->atime;
1932                 lli->lli_lvb.lvb_atime = body->atime;
1933         }
1934         if (body->valid & OBD_MD_FLMTIME) {
1935                 if (body->mtime > LTIME_S(inode->i_mtime)) {
1936                         CDEBUG(D_INODE, "setting ino %lu mtime from %lu "
1937                                "to "LPU64"\n", inode->i_ino,
1938                                LTIME_S(inode->i_mtime), body->mtime);
1939                         LTIME_S(inode->i_mtime) = body->mtime;
1940                 }
1941                 lli->lli_lvb.lvb_mtime = body->mtime;
1942         }
1943         if (body->valid & OBD_MD_FLCTIME) {
1944                 if (body->ctime > LTIME_S(inode->i_ctime))
1945                         LTIME_S(inode->i_ctime) = body->ctime;
1946                 lli->lli_lvb.lvb_ctime = body->ctime;
1947         }
1948         if (body->valid & OBD_MD_FLMODE)
1949                 inode->i_mode = (inode->i_mode & S_IFMT)|(body->mode & ~S_IFMT);
1950         if (body->valid & OBD_MD_FLTYPE)
1951                 inode->i_mode = (inode->i_mode & ~S_IFMT)|(body->mode & S_IFMT);
1952         if (S_ISREG(inode->i_mode)) {
1953                 inode->i_blkbits = min(PTLRPC_MAX_BRW_BITS+1, LL_MAX_BLKSIZE_BITS);
1954         } else {
1955                 inode->i_blkbits = inode->i_sb->s_blocksize_bits;
1956         }
1957 #ifdef HAVE_INODE_BLKSIZE
1958         inode->i_blksize = 1<<inode->i_blkbits;
1959 #endif
1960         if (body->valid & OBD_MD_FLUID)
1961                 inode->i_uid = body->uid;
1962         if (body->valid & OBD_MD_FLGID)
1963                 inode->i_gid = body->gid;
1964         if (body->valid & OBD_MD_FLFLAGS)
1965                 inode->i_flags = ll_ext_to_inode_flags(body->flags |
1966                                                        MDS_BFLAG_EXT_FLAGS);
1967         if (body->valid & OBD_MD_FLNLINK)
1968                 inode->i_nlink = body->nlink;
1969
1970         if (body->valid & OBD_MD_FLRDEV)
1971                 inode->i_rdev = old_decode_dev(body->rdev);
1972         if (body->valid & OBD_MD_FLSIZE) {
1973 #if 0           /* Can't block ll_test_inode->ll_update_inode, b=14326*/
1974                 ll_inode_size_lock(inode, 0);
1975                 i_size_write(inode, body->size);
1976                 ll_inode_size_unlock(inode, 0);
1977 #else
1978                 inode->i_size = body->size;
1979 #endif
1980         }
1981         if (body->valid & OBD_MD_FLBLOCKS)
1982                 inode->i_blocks = body->blocks;
1983
1984         if (body->valid & OBD_MD_FLSIZE)
1985                 set_bit(LLI_F_HAVE_MDS_SIZE_LOCK, &lli->lli_flags);
1986         EXIT;
1987 }
1988
1989 void ll_read_inode2(struct inode *inode, void *opaque)
1990 {
1991         struct lustre_md *md = opaque;
1992         struct ll_inode_info *lli = ll_i2info(inode);
1993         ENTRY;
1994
1995         CDEBUG(D_VFSTRACE, "VFS Op:inode=%lu/%u(%p)\n", inode->i_ino,
1996                inode->i_generation, inode);
1997
1998         ll_lli_init(lli);
1999
2000         LASSERT(!lli->lli_smd);
2001
2002         /* Core attributes from the MDS first.  This is a new inode, and
2003          * the VFS doesn't zero times in the core inode so we have to do
2004          * it ourselves.  They will be overwritten by either MDS or OST
2005          * attributes - we just need to make sure they aren't newer. */
2006         LTIME_S(inode->i_mtime) = 0;
2007         LTIME_S(inode->i_atime) = 0;
2008         LTIME_S(inode->i_ctime) = 0;
2009         inode->i_rdev = 0;
2010         ll_update_inode(inode, md);
2011
2012         /* OIDEBUG(inode); */
2013
2014         /* initializing backing dev info. */
2015         inode->i_mapping->backing_dev_info = &(s2lsi(inode->i_sb)->bdi);
2016
2017         if (S_ISREG(inode->i_mode)) {
2018                 struct ll_sb_info *sbi = ll_i2sbi(inode);
2019                 inode->i_op = &ll_file_inode_operations;
2020                 inode->i_fop = sbi->ll_fop;
2021                 inode->i_mapping->a_ops = &ll_aops;
2022                 EXIT;
2023         } else if (S_ISDIR(inode->i_mode)) {
2024                 inode->i_op = &ll_dir_inode_operations;
2025                 inode->i_fop = &ll_dir_operations;
2026                 inode->i_mapping->a_ops = &ll_dir_aops;
2027                 EXIT;
2028         } else if (S_ISLNK(inode->i_mode)) {
2029                 inode->i_op = &ll_fast_symlink_inode_operations;
2030                 EXIT;
2031         } else {
2032                 inode->i_op = &ll_special_inode_operations;
2033                 init_special_inode(inode, inode->i_mode,
2034                                    kdev_t_to_nr(inode->i_rdev));
2035                 EXIT;
2036         }
2037 }
2038
2039 int ll_iocontrol(struct inode *inode, struct file *file,
2040                  unsigned int cmd, unsigned long arg)
2041 {
2042         struct ll_sb_info *sbi = ll_i2sbi(inode);
2043         struct ptlrpc_request *req = NULL;
2044         int rc, flags = 0;
2045         ENTRY;
2046
2047         switch(cmd) {
2048         case FSFILT_IOC_GETFLAGS: {
2049                 struct ll_fid fid;
2050                 struct mds_body *body;
2051
2052                 ll_inode2fid(&fid, inode);
2053                 rc = mdc_getattr(sbi->ll_mdc_exp, &fid, OBD_MD_FLFLAGS,0,&req);
2054                 if (rc) {
2055                         CERROR("failure %d inode %lu\n", rc, inode->i_ino);
2056                         RETURN(-abs(rc));
2057                 }
2058
2059                 body = lustre_msg_buf(req->rq_repmsg, REPLY_REC_OFF,
2060                                       sizeof(*body));
2061
2062                 /* We want to return EXT3_*_FL flags to the caller via this
2063                  * ioctl.  An older MDS may be sending S_* flags, fix it up. */
2064                 flags = ll_inode_to_ext_flags(body->flags,
2065                                               MDS_BFLAG_EXT_FLAGS);
2066                 ptlrpc_req_finished (req);
2067
2068                 RETURN(put_user(flags, (int *)arg));
2069         }
2070         case FSFILT_IOC_SETFLAGS: {
2071                 struct mdc_op_data op_data = { { 0 } };
2072                 struct ll_iattr_struct attr;
2073                 struct obd_info oinfo = { { { 0 } } };
2074                 struct lov_stripe_md *lsm = ll_i2info(inode)->lli_smd;
2075
2076                 if (get_user(flags, (int *)arg))
2077                         RETURN(-EFAULT);
2078
2079                 oinfo.oi_md = lsm;
2080                 OBDO_ALLOC(oinfo.oi_oa);
2081                 if (!oinfo.oi_oa)
2082                         RETURN(-ENOMEM);
2083
2084                 ll_prepare_mdc_op_data(&op_data, inode, NULL, NULL, 0, 0, NULL);
2085
2086                 memset(&attr, 0, sizeof(attr));
2087                 attr.ia_attr_flags = flags;
2088                 ((struct iattr *)&attr)->ia_valid |= ATTR_ATTR_FLAG;
2089
2090                 rc = mdc_setattr(sbi->ll_mdc_exp, &op_data,
2091                                  (struct iattr *)&attr, NULL, 0, NULL, 0, &req);
2092                 ptlrpc_req_finished(req);
2093                 if (rc) {
2094                         OBDO_FREE(oinfo.oi_oa);
2095                         RETURN(rc);
2096                 }
2097
2098                 if (lsm == NULL) {
2099                         OBDO_FREE(oinfo.oi_oa);
2100                         GOTO(update_cache, rc);
2101                 }
2102
2103                 oinfo.oi_oa->o_id = lsm->lsm_object_id;
2104                 oinfo.oi_oa->o_gr = lsm->lsm_object_gr;
2105                 oinfo.oi_oa->o_flags = flags;
2106                 oinfo.oi_oa->o_valid = OBD_MD_FLID | OBD_MD_FLGROUP |
2107                                        OBD_MD_FLFLAGS;
2108
2109                 obdo_from_inode(oinfo.oi_oa, inode,
2110                                 OBD_MD_FLFID | OBD_MD_FLGENER);
2111                 rc = obd_setattr_rqset(sbi->ll_osc_exp, &oinfo, NULL);
2112                 OBDO_FREE(oinfo.oi_oa);
2113                 if (rc) {
2114                         if (rc != -EPERM && rc != -EACCES)
2115                                 CERROR("osc_setattr_async fails: rc = %d\n",rc);
2116                         RETURN(rc);
2117                 }
2118
2119                 EXIT;
2120 update_cache:
2121                 inode->i_flags = ll_ext_to_inode_flags(flags |
2122                                                        MDS_BFLAG_EXT_FLAGS);
2123                 return 0;
2124         }
2125         default:
2126                 RETURN(-ENOSYS);
2127         }
2128
2129         RETURN(0);
2130 }
2131
2132 /* umount -f client means force down, don't save state */
2133 #ifdef HAVE_UMOUNTBEGIN_VFSMOUNT
2134 void ll_umount_begin(struct vfsmount *vfsmnt, int flags)
2135 {
2136         struct super_block *sb = vfsmnt->mnt_sb;
2137 #else
2138 void ll_umount_begin(struct super_block *sb)
2139 {
2140 #endif
2141         struct lustre_sb_info *lsi = s2lsi(sb);
2142         struct ll_sb_info *sbi = ll_s2sbi(sb);
2143         struct obd_device *obd;
2144         struct obd_ioctl_data ioc_data = { 0 };
2145         ENTRY;
2146
2147 #ifdef HAVE_UMOUNTBEGIN_VFSMOUNT
2148         if (!(flags & MNT_FORCE)) {
2149                 EXIT;
2150                 return;
2151         }
2152 #endif
2153
2154         /* Tell the MGC we got umount -f */
2155         lsi->lsi_flags |= LSI_UMOUNT_FORCE;
2156
2157         CDEBUG(D_VFSTRACE, "VFS Op: superblock %p count %d active %d\n", sb,
2158                sb->s_count, atomic_read(&sb->s_active));
2159
2160         obd = class_exp2obd(sbi->ll_mdc_exp);
2161         if (obd == NULL) {
2162                 CERROR("Invalid MDC connection handle "LPX64"\n",
2163                        sbi->ll_mdc_exp->exp_handle.h_cookie);
2164                 EXIT;
2165                 return;
2166         }
2167         obd->obd_force = 1;
2168         obd_iocontrol(IOC_OSC_SET_ACTIVE, sbi->ll_mdc_exp, sizeof ioc_data,
2169                       &ioc_data, NULL);
2170
2171         obd = class_exp2obd(sbi->ll_osc_exp);
2172         if (obd == NULL) {
2173                 CERROR("Invalid LOV connection handle "LPX64"\n",
2174                        sbi->ll_osc_exp->exp_handle.h_cookie);
2175                 EXIT;
2176                 return;
2177         }
2178
2179         obd->obd_force = 1;
2180         obd_iocontrol(IOC_OSC_SET_ACTIVE, sbi->ll_osc_exp, sizeof ioc_data,
2181                       &ioc_data, NULL);
2182
2183         /* Really, we'd like to wait until there are no requests outstanding,
2184          * and then continue.  For now, we just invalidate the requests,
2185          * schedule() and sleep one second if needed, and hope.
2186          */
2187         schedule();
2188 #ifdef HAVE_UMOUNTBEGIN_VFSMOUNT
2189         if (atomic_read(&vfsmnt->mnt_count) > 2) {
2190                 cfs_schedule_timeout(CFS_TASK_INTERRUPTIBLE,
2191                                      cfs_time_seconds(1));
2192                 if (atomic_read(&vfsmnt->mnt_count) > 2)
2193                         LCONSOLE_WARN("Mount still busy with %d refs! You "
2194                                       "may try to umount it a bit later\n",
2195                                       atomic_read(&vfsmnt->mnt_count));
2196         }
2197 #endif
2198
2199         EXIT;
2200 }
2201
2202 int ll_remount_fs(struct super_block *sb, int *flags, char *data)
2203 {
2204         struct ll_sb_info *sbi = ll_s2sbi(sb);
2205         int err;
2206         __u32 read_only;
2207
2208         if ((*flags & MS_RDONLY) != (sb->s_flags & MS_RDONLY)) {
2209                 read_only = *flags & MS_RDONLY;
2210                 err = obd_set_info_async(sbi->ll_mdc_exp, sizeof(KEY_READONLY),
2211                                          KEY_READONLY, sizeof(read_only),
2212                                          &read_only, NULL);
2213
2214                 /* MDS might have expected a different ro key value, b=17493 */
2215                 if (err == -EINVAL) {
2216                         CDEBUG(D_CONFIG, "Retrying remount with 1.6.6 ro key\n");
2217                         err = obd_set_info_async(sbi->ll_mdc_exp,
2218                                                  sizeof(KEY_READONLY_166COMPAT),
2219                                                  KEY_READONLY_166COMPAT,
2220                                                  sizeof(read_only),
2221                                                  &read_only, NULL);
2222                 }
2223
2224                 if (err) {
2225                         CERROR("Failed to change the read-only flag during "
2226                                "remount: %d\n", err);
2227                         return err;
2228                 }
2229
2230                 if (read_only)
2231                         sb->s_flags |= MS_RDONLY;
2232                 else
2233                         sb->s_flags &= ~MS_RDONLY;
2234         }
2235         return 0;
2236 }
2237
2238 int ll_prep_inode(struct obd_export *exp, struct inode **inode,
2239                   struct ptlrpc_request *req, int offset,struct super_block *sb)
2240 {
2241         struct lustre_md md;
2242         struct ll_sb_info *sbi = NULL;
2243         int rc = 0;
2244         ENTRY;
2245
2246         LASSERT(*inode || sb);
2247         sbi = sb ? ll_s2sbi(sb) : ll_i2sbi(*inode);
2248
2249         rc = mdc_req2lustre_md(req, offset, exp, &md);
2250         if (rc)
2251                 RETURN(rc);
2252
2253         if (*inode) {
2254                 ll_update_inode(*inode, &md);
2255         } else {
2256                 LASSERT(sb);
2257                 /** hashing VFS inode by FIDs.
2258                  * IGIF will be used for for compatibility if needed.
2259                  */
2260                 *inode =ll_iget(sb, ll_fid_build_ino(&md.body->fid1), &md);
2261                 if (*inode == NULL || is_bad_inode(*inode)) {
2262                         mdc_free_lustre_md(exp, &md);
2263                         rc = -ENOMEM;
2264                         CERROR("new_inode -fatal: rc %d\n", rc);
2265                         GOTO(out, rc);
2266                 }
2267         }
2268
2269         rc = obd_checkmd(exp, ll_i2mdcexp(*inode),
2270                          ll_i2info(*inode)->lli_smd);
2271 out:
2272         RETURN(rc);
2273 }
2274
2275 char *llap_origins[] = {
2276         [LLAP_ORIGIN_UNKNOWN] = "--",
2277         [LLAP_ORIGIN_READPAGE] = "rp",
2278         [LLAP_ORIGIN_READAHEAD] = "ra",
2279         [LLAP_ORIGIN_COMMIT_WRITE] = "cw",
2280         [LLAP_ORIGIN_WRITEPAGE] = "wp"
2281 };
2282
2283 struct ll_async_page *llite_pglist_next_llap(struct list_head *head,
2284                                              struct list_head *list)
2285 {
2286         struct ll_async_page *llap;
2287         struct list_head *pos;
2288
2289         list_for_each(pos, list) {
2290                 if (pos == head)
2291                         return NULL;
2292                 llap = list_entry(pos, struct ll_async_page, llap_pglist_item);
2293                 if (llap->llap_page == NULL)
2294                         continue;
2295                 return llap;
2296         }
2297         LBUG();
2298         return NULL;
2299 }
2300
2301 int ll_obd_statfs(struct inode *inode, void *arg)
2302 {
2303         struct ll_sb_info *sbi = NULL;
2304         struct obd_device *client_obd = NULL, *lov_obd = NULL;
2305         struct lov_obd *lov = NULL;
2306         struct obd_statfs stat_buf = {0};
2307         char *buf = NULL;
2308         struct obd_ioctl_data *data = NULL;
2309         __u32 type, index;
2310         int len = 0, rc;
2311
2312         if (!inode || !(sbi = ll_i2sbi(inode)))
2313                 GOTO(out_statfs, rc = -EINVAL);
2314
2315         rc = obd_ioctl_getdata(&buf, &len, arg);
2316         if (rc)
2317                 GOTO(out_statfs, rc);
2318
2319         data = (void*)buf;
2320         if (!data->ioc_inlbuf1 || !data->ioc_inlbuf2 ||
2321             !data->ioc_pbuf1 || !data->ioc_pbuf2)
2322                 GOTO(out_statfs, rc = -EINVAL);
2323
2324         memcpy(&type, data->ioc_inlbuf1, sizeof(__u32));
2325         memcpy(&index, data->ioc_inlbuf2, sizeof(__u32));
2326
2327         if (type == LL_STATFS_MDC) {
2328                 if (index > 0)
2329                         GOTO(out_statfs, rc = -ENODEV);
2330                 client_obd = class_exp2obd(sbi->ll_mdc_exp);
2331         } else if (type == LL_STATFS_LOV) {
2332                 lov_obd = class_exp2obd(sbi->ll_osc_exp);
2333                 lov = &lov_obd->u.lov;
2334
2335                 if (index >= lov->desc.ld_tgt_count)
2336                         GOTO(out_statfs, rc = -ENODEV);
2337
2338                 if (!lov->lov_tgts[index])
2339                         /* Try again with the next index */
2340                         GOTO(out_statfs, rc = -EAGAIN);
2341
2342                 client_obd = class_exp2obd(lov->lov_tgts[index]->ltd_exp);
2343                 if (!lov->lov_tgts[index]->ltd_active)
2344                         GOTO(out_uuid, rc = -ENODATA);
2345         }
2346
2347         if (!client_obd)
2348                 GOTO(out_statfs, rc = -EINVAL);
2349
2350         rc = obd_statfs(client_obd, &stat_buf,
2351                         cfs_time_shift_64(-OBD_STATFS_CACHE_SECONDS), 1);
2352         if (rc)
2353                 GOTO(out_uuid, rc);
2354
2355         if (copy_to_user(data->ioc_pbuf1, &stat_buf, data->ioc_plen1))
2356                 GOTO(out_statfs, rc = -EFAULT);
2357
2358 out_uuid:
2359         if (client_obd) {
2360                 if (copy_to_user(data->ioc_pbuf2, obd2cli_tgt(client_obd),
2361                                  data->ioc_plen2))
2362                         rc = -EFAULT;
2363         } else {
2364                 if (copy_to_user(data->ioc_pbuf2, "Unknown UUID",
2365                                  sizeof("Unknown UUID") + 1))
2366                         rc = -EFAULT;
2367         }
2368
2369 out_statfs:
2370         if (buf)
2371                 obd_ioctl_freedata(buf, len);
2372         return rc;
2373 }
2374
2375 int ll_process_config(struct lustre_cfg *lcfg)
2376 {
2377         char *ptr;
2378         void *sb;
2379         struct lprocfs_static_vars lvars;
2380         unsigned long x;
2381         int rc = 0;
2382
2383         lprocfs_llite_init_vars(&lvars);
2384
2385         /* The instance name contains the sb: lustre-client-aacfe000 */
2386         ptr = strrchr(lustre_cfg_string(lcfg, 0), '-');
2387         if (!ptr || !*(++ptr))
2388                 return -EINVAL;
2389         if (sscanf(ptr, "%lx", &x) != 1)
2390                 return -EINVAL;
2391         sb = (void *)x;
2392         /* This better be a real Lustre superblock! */
2393         LASSERT(s2lsi((struct super_block *)sb)->lsi_lmd->lmd_magic == LMD_MAGIC);
2394
2395         /* Note we have not called client_common_fill_super yet, so
2396            proc fns must be able to handle that! */
2397         rc = class_process_proc_param(PARAM_LLITE, lvars.obd_vars,
2398                                       lcfg, sb);
2399         return(rc);
2400 }
2401
2402 int ll_show_options(struct seq_file *seq, struct vfsmount *vfs)
2403 {
2404         struct ll_sb_info *sbi;
2405
2406         LASSERT((seq != NULL) && (vfs != NULL));
2407         sbi = ll_s2sbi(vfs->mnt_sb);
2408
2409         if (sbi->ll_flags & LL_SBI_NOLCK)
2410                 seq_puts(seq, ",nolock");
2411
2412         if (sbi->ll_flags & LL_SBI_FLOCK)
2413                 seq_puts(seq, ",flock");
2414
2415         if (sbi->ll_flags & LL_SBI_LOCALFLOCK)
2416                 seq_puts(seq, ",localflock");
2417
2418         if (sbi->ll_flags & LL_SBI_USER_XATTR)
2419                 seq_puts(seq, ",user_xattr");
2420
2421         if (sbi->ll_flags & LL_SBI_ACL)
2422                 seq_puts(seq, ",acl");
2423
2424         if (sbi->ll_flags & LL_SBI_LAZYSTATFS)
2425                 seq_puts(seq, ",lazystatfs");
2426
2427         RETURN(0);
2428 }