lustre/llite/super.c

   1 /* -*- mode: c; c-basic-offset: 8; indent-tabs-mode: nil; -*-
   2  * vim:expandtab:shiftwidth=8:tabstop=8:
   3  *
   4  * Lustre Light Super operations
   5  *
   6  *  Copyright (c) 2002, 2003 Cluster File Systems, Inc.
   7  *
   8  *   This file is part of Lustre, http://www.lustre.org.
   9  *
  10  *   Lustre is free software; you can redistribute it and/or
  11  *   modify it under the terms of version 2 of the GNU General Public
  12  *   License as published by the Free Software Foundation.
  13  *
  14  *   Lustre is distributed in the hope that it will be useful,
  15  *   but WITHOUT ANY WARRANTY; without even the implied warranty of
  16  *   MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
  17  *   GNU General Public License for more details.
  18  *
  19  *   You should have received a copy of the GNU General Public License
  20  *   along with Lustre; if not, write to the Free Software
  21  *   Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.
  22  */
  23
  24 #define DEBUG_SUBSYSTEM S_LLITE
  25
  26 #include <linux/module.h>
  27 #include <linux/random.h>
  28 #include <linux/version.h>
  29 #include <linux/lustre_lite.h>
  30 #include <linux/lustre_ha.h>
  31 #include <linux/lustre_dlm.h>
  32 #include <linux/init.h>
  33 #include <linux/fs.h>
  34 #include <linux/lprocfs_status.h>
  35
  36 #if (LINUX_VERSION_CODE < KERNEL_VERSION(2,5,0))
  37 kmem_cache_t *ll_file_data_slab;
  38 extern struct address_space_operations ll_aops;
  39 extern struct address_space_operations ll_dir_aops;
  40 struct super_operations ll_super_operations;
  41
  42 /* /proc/lustre/llite root that tracks llite mount points */
  43 struct proc_dir_entry *proc_lustre_fs_root = NULL;
  44 /* lproc_llite.c */
  45 extern int lprocfs_register_mountpoint(struct proc_dir_entry *parent,
  46                                        struct super_block *sb,
  47                                        char *osc, char *mdc);
  48
  49 extern int ll_recover(struct recovd_data *, int);
  50 extern int ll_commitcbd_setup(struct ll_sb_info *);
  51 extern int ll_commitcbd_cleanup(struct ll_sb_info *);
  52
  53 static char *ll_read_opt(const char *opt, char *data)
  54 {
  55         char *value;
  56         char *retval;
  57         ENTRY;
  58
  59         CDEBUG(D_SUPER, "option: %s, data %s\n", opt, data);
  60         if (strncmp(opt, data, strlen(opt)))
  61                 RETURN(NULL);
  62         if ((value = strchr(data, '=')) == NULL)
  63                 RETURN(NULL);
  64
  65         value++;
  66         OBD_ALLOC(retval, strlen(value) + 1);
  67         if (!retval) {
  68                 CERROR("out of memory!\n");
  69                 RETURN(NULL);
  70         }
  71
  72         memcpy(retval, value, strlen(value)+1);
  73         CDEBUG(D_SUPER, "Assigned option: %s, value %s\n", opt, retval);
  74         RETURN(retval);
  75 }
  76
  77 static int ll_set_opt(const char *opt, char *data, int fl)
  78 {
  79         ENTRY;
  80
  81         CDEBUG(D_SUPER, "option: %s, data %s\n", opt, data);
  82         if (strncmp(opt, data, strlen(opt)))
  83                 RETURN(0);
  84         else
  85                 RETURN(fl);
  86 }
  87
  88 static void ll_options(char *options, char **ost, char **mds, int *flags)
  89 {
  90         char *this_char;
  91         ENTRY;
  92
  93         if (!options) {
  94                 EXIT;
  95                 return;
  96         }
  97
  98         for (this_char = strtok (options, ",");
  99              this_char != NULL;
 100              this_char = strtok (NULL, ",")) {
 101                 CDEBUG(D_SUPER, "this_char %s\n", this_char);
 102                 if ((!*ost && (*ost = ll_read_opt("osc", this_char)))||
 103                     (!*mds && (*mds = ll_read_opt("mdc", this_char)))||
 104                     (!(*flags & LL_SBI_NOLCK) &&
 105                      ((*flags) = (*flags) |
 106                       ll_set_opt("nolock", this_char, LL_SBI_NOLCK))))
 107                         continue;
 108         }
 109         EXIT;
 110 }
 111
 112 #ifndef log2
 113 #define log2(n) ffz(~(n))
 114 #endif
 115
 116 static struct super_block *ll_read_super(struct super_block *sb,
 117                                          void *data, int silent)
 118 {
 119         struct inode *root = 0;
 120         struct obd_device *obd;
 121         struct ll_sb_info *sbi;
 122         char *osc = NULL;
 123         char *mdc = NULL;
 124         int err;
 125         struct ll_fid rootfid;
 126         struct obd_statfs osfs;
 127         struct ptlrpc_request *request = NULL;
 128         struct ptlrpc_connection *mdc_conn;
 129         struct ll_read_inode2_cookie lic;
 130         class_uuid_t uuid;
 131         struct obd_uuid param_uuid;
 132
 133         ENTRY;
 134
 135         CDEBUG(D_VFSTRACE, "VFS Op\n");
 136         OBD_ALLOC(sbi, sizeof(*sbi));
 137         if (!sbi)
 138                 RETURN(NULL);
 139
 140         INIT_LIST_HEAD(&sbi->ll_conn_chain);
 141         INIT_LIST_HEAD(&sbi->ll_orphan_dentry_list);
 142         generate_random_uuid(uuid);
 143         class_uuid_unparse(uuid, &sbi->ll_sb_uuid);
 144
 145         sb->u.generic_sbp = sbi;
 146
 147         ll_options(data, &osc, &mdc, &sbi->ll_flags);
 148
 149         if (!osc) {
 150                 CERROR("no osc\n");
 151                 GOTO(out_free, sb = NULL);
 152         }
 153
 154         if (!mdc) {
 155                 CERROR("no mdc\n");
 156                 GOTO(out_free, sb = NULL);
 157         }
 158
 159         strncpy(param_uuid.uuid, mdc, sizeof(param_uuid.uuid));
 160         obd = class_uuid2obd(&param_uuid);
 161         if (!obd) {
 162                 CERROR("MDC %s: not setup or attached\n", mdc);
 163                 GOTO(out_free, sb = NULL);
 164         }
 165
 166         err = obd_connect(&sbi->ll_mdc_conn, obd, &sbi->ll_sb_uuid,
 167                           ptlrpc_recovd, ll_recover);
 168         if (err) {
 169                 CERROR("cannot connect to %s: rc = %d\n", mdc, err);
 170                 GOTO(out_free, sb = NULL);
 171         }
 172
 173         mdc_conn = sbi2mdc(sbi)->cl_import.imp_connection;
 174         list_add(&mdc_conn->c_sb_chain, &sbi->ll_conn_chain);
 175
 176         strncpy(param_uuid.uuid, osc, sizeof(param_uuid.uuid));
 177         obd = class_uuid2obd(&param_uuid);
 178         if (!obd) {
 179                 CERROR("OSC %s: not setup or attached\n", osc);
 180                 GOTO(out_mdc, sb = NULL);
 181         }
 182
 183         err = obd_connect(&sbi->ll_osc_conn, obd, &sbi->ll_sb_uuid,
 184                           ptlrpc_recovd, ll_recover);
 185         if (err) {
 186                 CERROR("cannot connect to %s: rc = %d\n", osc, err);
 187                 GOTO(out_mdc, sb = NULL);
 188         }
 189
 190         err = mdc_getstatus(&sbi->ll_mdc_conn, &rootfid);
 191         if (err) {
 192                 CERROR("cannot mds_connect: rc = %d\n", err);
 193                 GOTO(out_mdc, sb = NULL);
 194         }
 195         CDEBUG(D_SUPER, "rootfid "LPU64"\n", rootfid.id);
 196         sbi->ll_rootino = rootfid.id;
 197
 198         memset(&osfs, 0, sizeof(osfs));
 199         err = obd_statfs(&sbi->ll_mdc_conn, &osfs);
 200         sb->s_blocksize = osfs.os_bsize;
 201         sb->s_blocksize_bits = log2(osfs.os_bsize);
 202         sb->s_magic = LL_SUPER_MAGIC;
 203         sb->s_maxbytes = (1ULL << (32 + 9)) - osfs.os_bsize;
 204
 205         sb->s_op = &ll_super_operations;
 206
 207         /* make root inode */
 208         err = mdc_getattr(&sbi->ll_mdc_conn, sbi->ll_rootino, S_IFDIR,
 209                           OBD_MD_FLNOTOBD|OBD_MD_FLBLOCKS, 0, &request);
 210         if (err) {
 211                 CERROR("mdc_getattr failed for root: rc = %d\n", err);
 212                 GOTO(out_request, sb = NULL);
 213         }
 214
 215         /* initialize committed transaction callback daemon */
 216         spin_lock_init(&sbi->ll_commitcbd_lock);
 217         init_waitqueue_head(&sbi->ll_commitcbd_waitq);
 218         init_waitqueue_head(&sbi->ll_commitcbd_ctl_waitq);
 219         sbi->ll_commitcbd_flags = 0;
 220         err = ll_commitcbd_setup(sbi);
 221         if (err) {
 222                 CERROR("failed to start commit callback daemon: rc = %d\n",err);
 223                 GOTO(out_request, sb = NULL);
 224         }
 225
 226         lic.lic_body = lustre_msg_buf(request->rq_repmsg, 0);
 227         lic.lic_lmm = NULL;
 228         LASSERT(sbi->ll_rootino != 0);
 229         root = iget4(sb, sbi->ll_rootino, NULL, &lic);
 230
 231         if (root) {
 232                 sb->s_root = d_alloc_root(root);
 233         } else {
 234                 CERROR("lustre_lite: bad iget4 for root\n");
 235                 GOTO(out_cdb, sb = NULL);
 236         }
 237
 238         ptlrpc_req_finished(request);
 239         request = NULL;
 240
 241         if (proc_lustre_fs_root) {
 242                 err = lprocfs_register_mountpoint(proc_lustre_fs_root, sb,
 243                                                   osc, mdc);
 244                 if (err < 0)
 245                         CERROR("could not register mount in /proc/lustre");
 246         }
 247
 248 out_dev:
 249         if (mdc)
 250                 OBD_FREE(mdc, strlen(mdc) + 1);
 251         if (osc)
 252                 OBD_FREE(osc, strlen(osc) + 1);
 253
 254         RETURN(sb);
 255
 256 out_cdb:
 257         ll_commitcbd_cleanup(sbi);
 258 out_request:
 259         ptlrpc_req_finished(request);
 260         obd_disconnect(&sbi->ll_osc_conn);
 261 out_mdc:
 262         obd_disconnect(&sbi->ll_mdc_conn);
 263 out_free:
 264         OBD_FREE(sbi, sizeof(*sbi));
 265
 266         goto out_dev;
 267 } /* ll_read_super */
 268
 269 static void ll_put_super(struct super_block *sb)
 270 {
 271         struct ll_sb_info *sbi = ll_s2sbi(sb);
 272         struct list_head *tmp, *next;
 273         struct ll_fid rootfid;
 274         ENTRY;
 275
 276         CDEBUG(D_VFSTRACE, "VFS Op\n");
 277         list_del(&sbi->ll_conn_chain);
 278         ll_commitcbd_cleanup(sbi);
 279         obd_disconnect(&sbi->ll_osc_conn);
 280
 281         /* NULL request to force sync on the MDS, and get the last_committed
 282          * value to flush remaining RPCs from the sending queue on client.
 283          *
 284          * XXX This should be an mdc_sync() call to sync the whole MDS fs,
 285          *     which we can call for other reasons as well.
 286          */
 287         mdc_getstatus(&sbi->ll_mdc_conn, &rootfid);
 288
 289         if (sbi->ll_proc_root) {
 290                 lprocfs_remove(sbi->ll_proc_root);
 291                 sbi->ll_proc_root = NULL;
 292         }
 293
 294         obd_disconnect(&sbi->ll_mdc_conn);
 295
 296         spin_lock(&dcache_lock);
 297         list_for_each_safe(tmp, next, &sbi->ll_orphan_dentry_list) {
 298                 struct dentry *dentry = list_entry(tmp, struct dentry, d_hash);
 299                 shrink_dcache_parent(dentry);
 300         }
 301         spin_unlock(&dcache_lock);
 302
 303         OBD_FREE(sbi, sizeof(*sbi));
 304
 305         EXIT;
 306 } /* ll_put_super */
 307
 308 static void ll_clear_inode(struct inode *inode)
 309 {
 310         struct ll_sb_info *sbi = ll_i2sbi(inode);
 311         struct ll_inode_info *lli = ll_i2info(inode);
 312         int rc;
 313         ENTRY;
 314
 315         CDEBUG(D_VFSTRACE, "VFS Op\n");
 316         rc = mdc_cancel_unused(&sbi->ll_mdc_conn, inode, LDLM_FL_NO_CALLBACK);
 317         if (rc < 0) {
 318                 CERROR("mdc_cancel_unused: %d\n", rc);
 319                 /* XXX FIXME do something dramatic */
 320         }
 321
 322         if (lli->lli_smd) {
 323                 rc = obd_cancel_unused(&sbi->ll_osc_conn, lli->lli_smd, 0);
 324                 if (rc < 0) {
 325                         CERROR("obd_cancel_unused: %d\n", rc);
 326                         /* XXX FIXME do something dramatic */
 327                 }
 328         }
 329
 330         if (atomic_read(&inode->i_count) != 0)
 331                 CERROR("clearing in-use inode %lu: count = %d\n",
 332                        inode->i_ino, atomic_read(&inode->i_count));
 333
 334         if (lli->lli_smd)
 335                 obd_free_memmd(&sbi->ll_osc_conn, &lli->lli_smd);
 336
 337         if (lli->lli_symlink_name) {
 338                 OBD_FREE(lli->lli_symlink_name,
 339                          strlen(lli->lli_symlink_name) + 1);
 340                 lli->lli_symlink_name = NULL;
 341         }
 342
 343         EXIT;
 344 }
 345
 346 #if 0
 347 static void ll_delete_inode(struct inode *inode)
 348 {
 349         ENTRY;
 350         CDEBUG(D_VFSTRACE, "VFS Op\n");
 351         if (S_ISREG(inode->i_mode)) {
 352                 int err;
 353                 struct obdo *oa;
 354                 struct lov_stripe_md *lsm = ll_i2info(inode)->lli_smd;
 355
 356                 /* mcreate with no open */
 357                 if (!lsm)
 358                         GOTO(out, 0);
 359
 360                 if (lsm->lsm_object_id == 0) {
 361                         CERROR("This really happens\n");
 362                         /* No obdo was ever created */
 363                         GOTO(out, 0);
 364                 }
 365
 366                 oa = obdo_alloc();
 367                 if (oa == NULL)
 368                         GOTO(out, -ENOMEM);
 369
 370                 oa->o_id = lsm->lsm_object_id;
 371                 obdo_from_inode(oa, inode, OBD_MD_FLID | OBD_MD_FLTYPE);
 372
 373                 err = obd_destroy(ll_i2obdconn(inode), oa, lsm, NULL);
 374                 obdo_free(oa);
 375                 if (err)
 376                         CDEBUG(D_INODE,
 377                                "inode %lu obd_destroy objid "LPX64" error %d\n",
 378                                inode->i_ino, lsm->lsm_object_id, err);
 379         }
 380 out:
 381         clear_inode(inode);
 382         EXIT;
 383 }
 384 #endif
 385
 386 /* like inode_setattr, but doesn't mark the inode dirty */
 387 static int ll_attr2inode(struct inode *inode, struct iattr *attr, int trunc)
 388 {
 389         unsigned int ia_valid = attr->ia_valid;
 390         int error = 0;
 391
 392         if ((ia_valid & ATTR_SIZE) && trunc) {
 393                 error = vmtruncate(inode, attr->ia_size);
 394                 if (error)
 395                         goto out;
 396         } else if (ia_valid & ATTR_SIZE)
 397                 inode->i_size = attr->ia_size;
 398
 399         if (ia_valid & ATTR_UID)
 400                 inode->i_uid = attr->ia_uid;
 401         if (ia_valid & ATTR_GID)
 402                 inode->i_gid = attr->ia_gid;
 403         if (ia_valid & ATTR_ATIME)
 404                 inode->i_atime = attr->ia_atime;
 405         if (ia_valid & ATTR_MTIME)
 406                 inode->i_mtime = attr->ia_mtime;
 407         if (ia_valid & ATTR_CTIME)
 408                 inode->i_ctime = attr->ia_ctime;
 409         if (ia_valid & ATTR_MODE) {
 410                 inode->i_mode = attr->ia_mode;
 411                 if (!in_group_p(inode->i_gid) && !capable(CAP_FSETID))
 412                         inode->i_mode &= ~S_ISGID;
 413         }
 414 out:
 415         return error;
 416 }
 417
 418 int ll_inode_setattr(struct inode *inode, struct iattr *attr, int do_trunc)
 419 {
 420         struct ptlrpc_request *request = NULL;
 421         struct ll_sb_info *sbi = ll_i2sbi(inode);
 422         int err = 0;
 423         ENTRY;
 424
 425         /* change incore inode */
 426         ll_attr2inode(inode, attr, do_trunc);
 427
 428         /* Don't send size changes to MDS to avoid "fast EA" problems, and
 429          * also avoid a pointless RPC (we get file size from OST anyways).
 430          */
 431         attr->ia_valid &= ~ATTR_SIZE;
 432         if (attr->ia_valid) {
 433                 err = mdc_setattr(&sbi->ll_mdc_conn, inode, attr, NULL, 0,
 434                                   &request);
 435                 if (err)
 436                         CERROR("mdc_setattr fails: err = %d\n", err);
 437
 438                 ptlrpc_req_finished(request);
 439                 if (S_ISREG(inode->i_mode) && attr->ia_valid & ATTR_MTIME_SET) {
 440                         struct lov_stripe_md *lsm = ll_i2info(inode)->lli_smd;
 441                         struct obdo oa;
 442                         int err2;
 443
 444                         CDEBUG(D_INODE, "set mtime on OST inode %lu to %lu\n",
 445                                inode->i_ino, attr->ia_mtime);
 446                         oa.o_id = lsm->lsm_object_id;
 447                         oa.o_mode = S_IFREG;
 448                         oa.o_valid = OBD_MD_FLID |OBD_MD_FLTYPE |OBD_MD_FLMTIME;
 449                         oa.o_mtime = attr->ia_mtime;
 450                         err2 = obd_setattr(&sbi->ll_osc_conn, &oa, lsm, NULL);
 451                         if (err2) {
 452                                 CERROR("obd_setattr fails: rc=%d\n", err);
 453                                 if (!err)
 454                                         err = err2;
 455                         }
 456                 }
 457         }
 458
 459         RETURN(err);
 460 }
 461
 462 int ll_setattr_raw(struct inode *inode, struct iattr *attr)
 463 {
 464         struct ptlrpc_request *request = NULL;
 465         struct ll_sb_info *sbi = ll_i2sbi(inode);
 466         int err = 0;
 467         ENTRY;
 468
 469         if ((attr->ia_valid & ATTR_SIZE)) {
 470                 /* writeback uses inode->i_size to determine how far out
 471                  * its cached pages go.  ll_truncate gets a PW lock, canceling
 472                  * our lock, _after_ it has updated i_size.  this can confuse
 473                  * us into zero extending the file to the newly truncated
 474                  * size, and this has bad implications for a racing o_append.
 475                  * if we're extending our size we need to flush the pages
 476                  * with the correct i_size before vmtruncate stomps on
 477                  * the new i_size.  again, this can only find pages to
 478                  * purge if the PW lock that generated them is still held.
 479                  */
 480                 if ( attr->ia_size > inode->i_size ) {
 481                         filemap_fdatasync(inode->i_mapping);
 482                         filemap_fdatawait(inode->i_mapping);
 483                 }
 484                 err = vmtruncate(inode, attr->ia_size);
 485                 if (err)
 486                         RETURN(err);
 487         }
 488
 489         /* Don't send size changes to MDS to avoid "fast EA" problems, and
 490          * also avoid a pointless RPC (we get file size from OST anyways).
 491          */
 492         attr->ia_valid &= ~ATTR_SIZE;
 493         if (!attr->ia_valid)
 494                 RETURN(0);
 495
 496         err = mdc_setattr(&sbi->ll_mdc_conn, inode, attr, NULL, 0,
 497                           &request);
 498         if (err)
 499                 CERROR("mdc_setattr fails: err = %d\n", err);
 500
 501         ptlrpc_req_finished(request);
 502
 503         if (S_ISREG(inode->i_mode) && attr->ia_valid & ATTR_MTIME_SET) {
 504                 struct lov_stripe_md *lsm = ll_i2info(inode)->lli_smd;
 505                 struct obdo oa;
 506                 int err2;
 507
 508                 CDEBUG(D_INODE, "set mtime on OST inode %lu to %lu\n",
 509                        inode->i_ino, attr->ia_mtime);
 510                 oa.o_id = lsm->lsm_object_id;
 511                 oa.o_mode = S_IFREG;
 512                 oa.o_valid = OBD_MD_FLID | OBD_MD_FLTYPE | OBD_MD_FLMTIME;
 513                 oa.o_mtime = attr->ia_mtime;
 514                 err2 = obd_setattr(&sbi->ll_osc_conn, &oa, lsm, NULL);
 515                 if (err2) {
 516                         CERROR("obd_setattr fails: rc=%d\n", err);
 517                         if (!err)
 518                                 err = err2;
 519                 }
 520         }
 521         RETURN(err);
 522 }
 523
 524 int ll_setattr(struct dentry *de, struct iattr *attr)
 525 {
 526         int rc = inode_change_ok(de->d_inode, attr);
 527
 528         CDEBUG(D_VFSTRACE, "VFS Op\n");
 529         if (rc)
 530                 return rc;
 531
 532         return ll_inode_setattr(de->d_inode, attr, 1);
 533 }
 534
 535 static int ll_statfs(struct super_block *sb, struct statfs *sfs)
 536 {
 537         struct ll_sb_info *sbi = ll_s2sbi(sb);
 538         struct obd_statfs osfs;
 539         int rc;
 540         ENTRY;
 541
 542         CDEBUG(D_VFSTRACE, "VFS Op\n");
 543         memset(sfs, 0, sizeof(*sfs));
 544         rc = obd_statfs(&sbi->ll_mdc_conn, &osfs);
 545         statfs_unpack(sfs, &osfs);
 546         if (rc)
 547                 CERROR("mdc_statfs fails: rc = %d\n", rc);
 548         else
 549                 CDEBUG(D_SUPER, "mdc_statfs shows blocks "LPU64"/"LPU64
 550                        " objects "LPU64"/"LPU64"\n",
 551                        osfs.os_bavail, osfs.os_blocks,
 552                        osfs.os_ffree, osfs.os_files);
 553
 554         /* temporary until mds_statfs returns statfs info for all OSTs */
 555         if (!rc) {
 556                 rc = obd_statfs(&sbi->ll_osc_conn, &osfs);
 557                 if (rc) {
 558                         CERROR("obd_statfs fails: rc = %d\n", rc);
 559                         GOTO(out, rc);
 560                 }
 561                 CDEBUG(D_SUPER, "obd_statfs shows blocks "LPU64"/"LPU64
 562                        " objects "LPU64"/"LPU64"\n",
 563                        osfs.os_bavail, osfs.os_blocks,
 564                        osfs.os_ffree, osfs.os_files);
 565
 566                 while (osfs.os_blocks > ~0UL) {
 567                         sfs->f_bsize <<= 1;
 568
 569                         osfs.os_blocks >>= 1;
 570                         osfs.os_bfree >>= 1;
 571                         osfs.os_bavail >>= 1;
 572                 }
 573                 sfs->f_blocks = osfs.os_blocks;
 574                 sfs->f_bfree = osfs.os_bfree;
 575                 sfs->f_bavail = osfs.os_bavail;
 576                 if (osfs.os_ffree < (__u64)sfs->f_ffree)
 577                         sfs->f_ffree = osfs.os_ffree;
 578         }
 579
 580 out:
 581         RETURN(rc);
 582 }
 583
 584 void ll_update_inode(struct inode *inode, struct mds_body *body,
 585                      struct lov_mds_md *lmm)
 586 {
 587         struct ll_inode_info *lli = ll_i2info(inode);
 588
 589         if (lmm != NULL)
 590                 obd_unpackmd(ll_i2obdconn(inode), &lli->lli_smd, lmm);
 591
 592         if (body->valid & OBD_MD_FLID)
 593                 inode->i_ino = body->ino;
 594         if (body->valid & OBD_MD_FLATIME)
 595                 inode->i_atime = body->atime;
 596         if (body->valid & OBD_MD_FLMTIME)
 597                 inode->i_mtime = body->mtime;
 598         if (body->valid & OBD_MD_FLCTIME)
 599                 inode->i_ctime = body->ctime;
 600         if (body->valid & OBD_MD_FLMODE)
 601                 inode->i_mode = (inode->i_mode & S_IFMT)|(body->mode & ~S_IFMT);
 602         if (body->valid & OBD_MD_FLTYPE)
 603                 inode->i_mode = (inode->i_mode & ~S_IFMT)|(body->mode & S_IFMT);
 604         if (body->valid & OBD_MD_FLUID)
 605                 inode->i_uid = body->uid;
 606         if (body->valid & OBD_MD_FLGID)
 607                 inode->i_gid = body->gid;
 608         if (body->valid & OBD_MD_FLFLAGS)
 609                 inode->i_flags = body->flags;
 610         if (body->valid & OBD_MD_FLNLINK)
 611                 inode->i_nlink = body->nlink;
 612         if (body->valid & OBD_MD_FLGENER)
 613                 inode->i_generation = body->generation;
 614         if (body->valid & OBD_MD_FLRDEV)
 615                 inode->i_rdev = body->rdev;
 616         if (body->valid & OBD_MD_FLSIZE)
 617                 inode->i_size = body->size;
 618         if (body->valid & OBD_MD_FLBLOCKS)
 619                 inode->i_blocks = body->blocks;
 620 }
 621
 622 static void ll_read_inode2(struct inode *inode, void *opaque)
 623 {
 624         struct ll_read_inode2_cookie *lic = opaque;
 625         struct mds_body *body = lic->lic_body;
 626         struct ll_inode_info *lli = ll_i2info(inode);
 627         ENTRY;
 628
 629         CDEBUG(D_VFSTRACE, "VFS Op\n");
 630         sema_init(&lli->lli_open_sem, 1);
 631         atomic_set(&lli->lli_open_count, 0);
 632         lli->lli_flags = 0;
 633         init_MUTEX(&lli->lli_getattr_sem);
 634         spin_lock_init(&lli->lli_read_extent_lock);
 635         INIT_LIST_HEAD(&lli->lli_read_extents);
 636
 637         LASSERT(!lli->lli_smd);
 638
 639         /* core attributes from the MDS first */
 640         ll_update_inode(inode, body, lic ? lic->lic_lmm : NULL);
 641
 642         /* Get the authoritative file size */
 643         if (lli->lli_smd && (inode->i_mode & S_IFREG)) {
 644                 struct ldlm_extent extent = {0, OBD_OBJECT_EOF};
 645                 struct lustre_handle lockh = {0, 0};
 646                 struct lov_stripe_md *lsm = ll_i2info(inode)->lli_smd;
 647                 ldlm_error_t rc;
 648
 649                 LASSERT(lli->lli_smd->lsm_object_id != 0);
 650
 651                 rc = ll_extent_lock(NULL, inode, lsm, LCK_PR, &extent, &lockh);
 652                 if (rc != ELDLM_OK && rc != ELDLM_LOCK_MATCHED) {
 653                         ll_clear_inode(inode);
 654                         make_bad_inode(inode);
 655                 } else {
 656                         ll_extent_unlock(NULL, inode, lsm, LCK_PR, &lockh);
 657                 }
 658         }
 659
 660         /* OIDEBUG(inode); */
 661
 662         if (S_ISREG(inode->i_mode)) {
 663                 inode->i_op = &ll_file_inode_operations;
 664                 inode->i_fop = &ll_file_operations;
 665                 inode->i_mapping->a_ops = &ll_aops;
 666                 EXIT;
 667         } else if (S_ISDIR(inode->i_mode)) {
 668                 inode->i_op = &ll_dir_inode_operations;
 669                 inode->i_fop = &ll_dir_operations;
 670                 inode->i_mapping->a_ops = &ll_dir_aops;
 671                 EXIT;
 672         } else if (S_ISLNK(inode->i_mode)) {
 673                 inode->i_op = &ll_fast_symlink_inode_operations;
 674                 EXIT;
 675         } else {
 676                 inode->i_op = &ll_special_inode_operations;
 677                 init_special_inode(inode, inode->i_mode, inode->i_rdev);
 678                 EXIT;
 679         }
 680 }
 681
 682 static inline void invalidate_request_list(struct list_head *req_list)
 683 {
 684         struct list_head *tmp, *n;
 685         list_for_each_safe(tmp, n, req_list) {
 686                 struct ptlrpc_request *req =
 687                         list_entry(tmp, struct ptlrpc_request, rq_list);
 688                 CERROR("invalidating req xid "LPU64" op %d to %s:%d\n",
 689                        req->rq_xid, req->rq_reqmsg->opc,
 690                        req->rq_connection->c_remote_uuid.uuid,
 691                        req->rq_import->imp_client->cli_request_portal);
 692                 req->rq_flags |= PTL_RPC_FL_ERR;
 693                 wake_up(&req->rq_wait_for_rep);
 694         }
 695 }
 696
 697 void ll_umount_begin(struct super_block *sb)
 698 {
 699         struct ll_sb_info *sbi = ll_s2sbi(sb);
 700         struct list_head *ctmp;
 701
 702         ENTRY;
 703         CDEBUG(D_VFSTRACE, "VFS Op\n");
 704
 705         list_for_each(ctmp, &sbi->ll_conn_chain) {
 706                 struct ptlrpc_connection *conn;
 707                 conn = list_entry(ctmp, struct ptlrpc_connection, c_sb_chain);
 708
 709                 spin_lock(&conn->c_lock);
 710                 /* XXX should just be dealing with imports, probably through
 711                  * XXX iocontrol, need next-gen recovery! */
 712                 conn->c_flags |= CONN_INVALID;
 713                 /* invalidate_request_list(&conn->c_sending_head); */
 714                 invalidate_request_list(&conn->c_delayed_head);
 715                 spin_unlock(&conn->c_lock);
 716         }
 717
 718         EXIT;
 719 }
 720
 721 /* exported operations */
 722 struct super_operations ll_super_operations =
 723 {
 724         read_inode2: ll_read_inode2,
 725         clear_inode: ll_clear_inode,
 726         //        delete_inode: ll_delete_inode,
 727         put_super: ll_put_super,
 728         statfs: ll_statfs,
 729         umount_begin: ll_umount_begin
 730 };
 731
 732 static struct file_system_type lustre_lite_fs_type = {
 733         name:           "lustre_lite",
 734         fs_flags:       0,
 735         read_super:     ll_read_super,
 736         owner:          THIS_MODULE,
 737 };
 738
 739 static int __init init_lustre_lite(void)
 740 {
 741         printk(KERN_INFO "Lustre Lite Client File System; "
 742                "info@clusterfs.com\n");
 743         ll_file_data_slab = kmem_cache_create("ll_file_data",
 744                                               sizeof(struct ll_file_data), 0,
 745                                               SLAB_HWCACHE_ALIGN, NULL, NULL);
 746         if (ll_file_data_slab == NULL)
 747                 return -ENOMEM;
 748
 749         proc_lustre_fs_root = proc_lustre_root ? proc_mkdir("llite", proc_lustre_root) : NULL;
 750
 751         return register_filesystem(&lustre_lite_fs_type);
 752 }
 753
 754 static void __exit exit_lustre_lite(void)
 755 {
 756         unregister_filesystem(&lustre_lite_fs_type);
 757         kmem_cache_destroy(ll_file_data_slab);
 758
 759         if (proc_lustre_fs_root) {
 760                 lprocfs_remove(proc_lustre_fs_root);
 761                 proc_lustre_fs_root = NULL;
 762         }
 763 }
 764
 765 MODULE_AUTHOR("Cluster File Systems, Inc. <info@clusterfs.com>");
 766 MODULE_DESCRIPTION("Lustre Lite Client File System");
 767 MODULE_LICENSE("GPL");
 768
 769 module_init(init_lustre_lite);
 770 module_exit(exit_lustre_lite);
 771 #endif