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 #include "llite_internal.h"
  36
  37 #if (LINUX_VERSION_CODE < KERNEL_VERSION(2,5,0))
  38 kmem_cache_t *ll_file_data_slab;
  39 extern struct address_space_operations ll_aops;
  40 extern struct address_space_operations ll_dir_aops;
  41 struct super_operations ll_super_operations;
  42
  43 /* /proc/lustre/llite root that tracks llite mount points */
  44 struct proc_dir_entry *proc_lustre_fs_root = NULL;
  45 /* lproc_llite.c */
  46 extern int lprocfs_register_mountpoint(struct proc_dir_entry *parent,
  47                                        struct super_block *sb,
  48                                        char *osc, char *mdc);
  49
  50 extern int ll_recover(struct recovd_data *, int);
  51 extern int ll_commitcbd_setup(struct ll_sb_info *);
  52 extern int ll_commitcbd_cleanup(struct ll_sb_info *);
  53
  54 static char *ll_read_opt(const char *opt, char *data)
  55 {
  56         char *value;
  57         char *retval;
  58         ENTRY;
  59
  60         CDEBUG(D_SUPER, "option: %s, data %s\n", opt, data);
  61         if (strncmp(opt, data, strlen(opt)))
  62                 RETURN(NULL);
  63         if ((value = strchr(data, '=')) == NULL)
  64                 RETURN(NULL);
  65
  66         value++;
  67         OBD_ALLOC(retval, strlen(value) + 1);
  68         if (!retval) {
  69                 CERROR("out of memory!\n");
  70                 RETURN(NULL);
  71         }
  72
  73         memcpy(retval, value, strlen(value)+1);
  74         CDEBUG(D_SUPER, "Assigned option: %s, value %s\n", opt, retval);
  75         RETURN(retval);
  76 }
  77
  78 static int ll_set_opt(const char *opt, char *data, int fl)
  79 {
  80         ENTRY;
  81
  82         CDEBUG(D_SUPER, "option: %s, data %s\n", opt, data);
  83         if (strncmp(opt, data, strlen(opt)))
  84                 RETURN(0);
  85         else
  86                 RETURN(fl);
  87 }
  88
  89 static void ll_options(char *options, char **ost, char **mds, int *flags)
  90 {
  91         char *this_char;
  92         ENTRY;
  93
  94         if (!options) {
  95                 EXIT;
  96                 return;
  97         }
  98
  99         for (this_char = strtok (options, ",");
 100              this_char != NULL;
 101              this_char = strtok (NULL, ",")) {
 102                 CDEBUG(D_SUPER, "this_char %s\n", this_char);
 103                 if ((!*ost && (*ost = ll_read_opt("osc", this_char)))||
 104                     (!*mds && (*mds = ll_read_opt("mdc", this_char)))||
 105                     (!(*flags & LL_SBI_NOLCK) &&
 106                      ((*flags) = (*flags) |
 107                       ll_set_opt("nolock", this_char, LL_SBI_NOLCK))))
 108                         continue;
 109         }
 110         EXIT;
 111 }
 112
 113 #ifndef log2
 114 #define log2(n) ffz(~(n))
 115 #endif
 116
 117 static struct super_block *ll_read_super(struct super_block *sb,
 118                                          void *data, int silent)
 119 {
 120         struct inode *root = 0;
 121         struct obd_device *obd;
 122         struct ll_sb_info *sbi;
 123         char *osc = NULL;
 124         char *mdc = NULL;
 125         int err;
 126         struct ll_fid rootfid;
 127         struct obd_statfs osfs;
 128         struct ptlrpc_request *request = NULL;
 129         struct ptlrpc_connection *mdc_conn;
 130         struct ll_read_inode2_cookie lic;
 131         class_uuid_t uuid;
 132         struct obd_uuid param_uuid;
 133
 134         ENTRY;
 135
 136         CDEBUG(D_VFSTRACE, "VFS Op: sb %p\n", sb);
 137         OBD_ALLOC(sbi, sizeof(*sbi));
 138         if (!sbi)
 139                 RETURN(NULL);
 140
 141         INIT_LIST_HEAD(&sbi->ll_conn_chain);
 142         INIT_LIST_HEAD(&sbi->ll_orphan_dentry_list);
 143         generate_random_uuid(uuid);
 144         spin_lock_init(&sbi->ll_iostats.fis_lock);
 145         class_uuid_unparse(uuid, &sbi->ll_sb_uuid);
 146
 147         sb->u.generic_sbp = sbi;
 148
 149         ll_options(data, &osc, &mdc, &sbi->ll_flags);
 150
 151         if (!osc) {
 152                 CERROR("no osc\n");
 153                 GOTO(out_free, sb = NULL);
 154         }
 155
 156         if (!mdc) {
 157                 CERROR("no mdc\n");
 158                 GOTO(out_free, sb = NULL);
 159         }
 160
 161         strncpy(param_uuid.uuid, mdc, sizeof(param_uuid.uuid));
 162         obd = class_uuid2obd(&param_uuid);
 163         if (!obd) {
 164                 CERROR("MDC %s: not setup or attached\n", mdc);
 165                 GOTO(out_free, sb = NULL);
 166         }
 167
 168         err = obd_connect(&sbi->ll_mdc_conn, obd, &sbi->ll_sb_uuid);
 169         if (err) {
 170                 CERROR("cannot connect to %s: rc = %d\n", mdc, err);
 171                 GOTO(out_free, sb = NULL);
 172         }
 173
 174         mdc_conn = sbi2mdc(sbi)->cl_import->imp_connection;
 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         if (err) {
 185                 CERROR("cannot connect to %s: rc = %d\n", osc, err);
 186                 GOTO(out_mdc, sb = NULL);
 187         }
 188
 189         err = mdc_getstatus(&sbi->ll_mdc_conn, &rootfid);
 190         if (err) {
 191                 CERROR("cannot mds_connect: rc = %d\n", err);
 192                 GOTO(out_osc, sb = NULL);
 193         }
 194         CDEBUG(D_SUPER, "rootfid "LPU64"\n", rootfid.id);
 195         sbi->ll_rootino = rootfid.id;
 196
 197         memset(&osfs, 0, sizeof(osfs));
 198         err = obd_statfs(&sbi->ll_mdc_conn, &osfs);
 199         sb->s_blocksize = osfs.os_bsize;
 200         sb->s_blocksize_bits = log2(osfs.os_bsize);
 201         sb->s_magic = LL_SUPER_MAGIC;
 202         sb->s_maxbytes = PAGE_CACHE_MAXBYTES;
 203
 204         sb->s_op = &ll_super_operations;
 205
 206         /* make root inode
 207          * XXX: move this to after cbd setup? */
 208         err = mdc_getattr(&sbi->ll_mdc_conn, &rootfid,
 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_osc, 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                 ptlrpc_req_finished (request);
 224                 GOTO(out_osc, sb = NULL);
 225         }
 226
 227         lic.lic_body = lustre_msg_buf(request->rq_repmsg, 0,
 228                                       sizeof(*lic.lic_body));
 229         LASSERT (lic.lic_body != NULL);         /* checked by mdc_getattr() */
 230         LASSERT_REPSWABBED (request, 0);        /* swabbed by mdc_getattr() */
 231
 232         lic.lic_lsm = NULL;
 233
 234         LASSERT(sbi->ll_rootino != 0);
 235         root = iget4(sb, sbi->ll_rootino, NULL, &lic);
 236
 237         ptlrpc_req_finished(request);
 238
 239         if (root == NULL || is_bad_inode(root)) {
 240                 /* XXX might need iput() for bad inode */
 241                 CERROR("lustre_lite: bad iget4 for root\n");
 242                 GOTO(out_cbd, sb = NULL);
 243         }
 244
 245         sb->s_root = d_alloc_root(root);
 246
 247         if (proc_lustre_fs_root) {
 248                 err = lprocfs_register_mountpoint(proc_lustre_fs_root, sb,
 249                                                   osc, mdc);
 250                 if (err < 0)
 251                         CERROR("could not register mount in /proc/lustre");
 252         }
 253
 254 out_dev:
 255         if (mdc)
 256                 OBD_FREE(mdc, strlen(mdc) + 1);
 257         if (osc)
 258                 OBD_FREE(osc, strlen(osc) + 1);
 259
 260         RETURN(sb);
 261
 262 out_cbd:
 263         ll_commitcbd_cleanup(sbi);
 264 out_osc:
 265         obd_disconnect(&sbi->ll_osc_conn, 0);
 266 out_mdc:
 267         obd_disconnect(&sbi->ll_mdc_conn, 0);
 268 out_free:
 269         OBD_FREE(sbi, sizeof(*sbi));
 270
 271         goto out_dev;
 272 } /* ll_read_super */
 273
 274 static void ll_put_super(struct super_block *sb)
 275 {
 276         struct ll_sb_info *sbi = ll_s2sbi(sb);
 277         struct list_head *tmp, *next;
 278         struct ll_fid rootfid;
 279         struct obd_device *obd = class_conn2obd(&sbi->ll_mdc_conn);
 280         ENTRY;
 281
 282         CDEBUG(D_VFSTRACE, "VFS Op: sb %p\n", sb);
 283         list_del(&sbi->ll_conn_chain);
 284         ll_commitcbd_cleanup(sbi);
 285         obd_disconnect(&sbi->ll_osc_conn, 0);
 286
 287         /* NULL request to force sync on the MDS, and get the last_committed
 288          * value to flush remaining RPCs from the sending queue on client.
 289          *
 290          * XXX This should be an mdc_sync() call to sync the whole MDS fs,
 291          *     which we can call for other reasons as well.
 292          */
 293         if (!obd->obd_no_recov)
 294                 mdc_getstatus(&sbi->ll_mdc_conn, &rootfid);
 295
 296         if (sbi->ll_proc_root) {
 297                 lprocfs_remove(sbi->ll_proc_root);
 298                 sbi->ll_proc_root = NULL;
 299         }
 300
 301         obd_disconnect(&sbi->ll_mdc_conn, 0);
 302
 303         spin_lock(&dcache_lock);
 304         list_for_each_safe(tmp, next, &sbi->ll_orphan_dentry_list) {
 305                 struct dentry *dentry = list_entry(tmp, struct dentry, d_hash);
 306                 shrink_dcache_parent(dentry);
 307         }
 308         spin_unlock(&dcache_lock);
 309
 310         OBD_FREE(sbi, sizeof(*sbi));
 311
 312         EXIT;
 313 } /* ll_put_super */
 314
 315 static void ll_clear_inode(struct inode *inode)
 316 {
 317         struct ll_sb_info *sbi = ll_i2sbi(inode);
 318         struct ll_inode_info *lli = ll_i2info(inode);
 319         int rc;
 320         ENTRY;
 321
 322         CDEBUG(D_VFSTRACE, "VFS Op:inode=%lu/%u(%p)\n", inode->i_ino,
 323                inode->i_generation, inode);
 324         rc = ll_mdc_cancel_unused(&sbi->ll_mdc_conn, inode,
 325                                   LDLM_FL_NO_CALLBACK, inode);
 326         if (rc < 0) {
 327                 CERROR("ll_mdc_cancel_unused: %d\n", rc);
 328                 /* XXX FIXME do something dramatic */
 329         }
 330
 331         if (atomic_read(&inode->i_count) != 0)
 332                 CERROR("clearing in-use inode %lu: count = %d\n",
 333                        inode->i_ino, atomic_read(&inode->i_count));
 334
 335         if (lli->lli_smd) {
 336                 rc = obd_cancel_unused(&sbi->ll_osc_conn, lli->lli_smd,
 337                                        LDLM_FL_WARN, inode);
 338                 if (rc < 0) {
 339                         CERROR("obd_cancel_unused: %d\n", rc);
 340                         /* XXX FIXME do something dramatic */
 341                 }
 342                 obd_free_memmd(&sbi->ll_osc_conn, &lli->lli_smd);
 343                 lli->lli_smd = NULL;
 344         }
 345
 346         if (lli->lli_symlink_name) {
 347                 OBD_FREE(lli->lli_symlink_name,
 348                          strlen(lli->lli_symlink_name) + 1);
 349                 lli->lli_symlink_name = NULL;
 350         }
 351
 352         EXIT;
 353 }
 354
 355 #if 0
 356 static void ll_delete_inode(struct inode *inode)
 357 {
 358         ENTRY;
 359         CDEBUG(D_VFSTRACE, "VFS Op:inode=%lu/%u(%p)\n", inode->i_ino,
 360                inode->i_generation, inode);
 361         if (S_ISREG(inode->i_mode)) {
 362                 int err;
 363                 struct obdo *oa;
 364                 struct lov_stripe_md *lsm = ll_i2info(inode)->lli_smd;
 365
 366                 /* mcreate with no open */
 367                 if (!lsm)
 368                         GOTO(out, 0);
 369
 370                 if (lsm->lsm_object_id == 0) {
 371                         CERROR("This really happens\n");
 372                         /* No obdo was ever created */
 373                         GOTO(out, 0);
 374                 }
 375
 376                 oa = obdo_alloc();
 377                 if (oa == NULL)
 378                         GOTO(out, -ENOMEM);
 379
 380                 oa->o_id = lsm->lsm_object_id;
 381                 obdo_from_inode(oa, inode, OBD_MD_FLID | OBD_MD_FLTYPE);
 382
 383                 err = obd_destroy(ll_i2obdconn(inode), oa, lsm, NULL);
 384                 obdo_free(oa);
 385                 if (err)
 386                         CDEBUG(D_INODE,
 387                                "inode %lu obd_destroy objid "LPX64" error %d\n",
 388                                inode->i_ino, lsm->lsm_object_id, err);
 389         }
 390 out:
 391         clear_inode(inode);
 392         EXIT;
 393 }
 394 #endif
 395
 396 /* like inode_setattr, but doesn't mark the inode dirty */
 397 static int ll_attr2inode(struct inode *inode, struct iattr *attr, int trunc)
 398 {
 399         unsigned int ia_valid = attr->ia_valid;
 400         int error = 0;
 401
 402         if ((ia_valid & ATTR_SIZE) && trunc) {
 403                 if (attr->ia_size > ll_file_maxbytes(inode)) {
 404                         error = -EFBIG;
 405                         goto out;
 406                 }
 407                 error = vmtruncate(inode, attr->ia_size);
 408                 if (error)
 409                         goto out;
 410         } else if (ia_valid & ATTR_SIZE)
 411                 inode->i_size = attr->ia_size;
 412
 413         if (ia_valid & ATTR_UID)
 414                 inode->i_uid = attr->ia_uid;
 415         if (ia_valid & ATTR_GID)
 416                 inode->i_gid = attr->ia_gid;
 417         if (ia_valid & ATTR_ATIME)
 418                 inode->i_atime = attr->ia_atime;
 419         if (ia_valid & ATTR_MTIME)
 420                 inode->i_mtime = attr->ia_mtime;
 421         if (ia_valid & ATTR_CTIME)
 422                 inode->i_ctime = attr->ia_ctime;
 423         if (ia_valid & ATTR_MODE) {
 424                 inode->i_mode = attr->ia_mode;
 425                 if (!in_group_p(inode->i_gid) && !capable(CAP_FSETID))
 426                         inode->i_mode &= ~S_ISGID;
 427         }
 428 out:
 429         return error;
 430 }
 431
 432 int ll_inode_setattr(struct inode *inode, struct iattr *attr, int do_trunc)
 433 {
 434         struct ptlrpc_request *request = NULL;
 435         struct ll_sb_info *sbi = ll_i2sbi(inode);
 436         int err = 0;
 437         ENTRY;
 438
 439         /* change incore inode */
 440         err = ll_attr2inode(inode, attr, do_trunc);
 441         if (err)
 442                 RETURN(err);
 443
 444         /* Don't send size changes to MDS to avoid "fast EA" problems, and
 445          * also avoid a pointless RPC (we get file size from OST anyways).
 446          */
 447         attr->ia_valid &= ~ATTR_SIZE;
 448         if (attr->ia_valid) {
 449                 struct mdc_op_data op_data;
 450
 451                 ll_prepare_mdc_op_data(&op_data, inode, NULL, NULL, 0, 0);
 452                 err = mdc_setattr(&sbi->ll_mdc_conn, &op_data,
 453                                   attr, NULL, 0, &request);
 454                 if (err)
 455                         CERROR("mdc_setattr fails: err = %d\n", err);
 456
 457                 ptlrpc_req_finished(request);
 458                 if (S_ISREG(inode->i_mode) && attr->ia_valid & ATTR_MTIME_SET) {
 459                         struct lov_stripe_md *lsm = ll_i2info(inode)->lli_smd;
 460                         struct obdo oa;
 461                         int err2;
 462
 463                         CDEBUG(D_INODE, "set mtime on OST inode %lu to %lu\n",
 464                                inode->i_ino, attr->ia_mtime);
 465                         oa.o_id = lsm->lsm_object_id;
 466                         oa.o_mode = S_IFREG;
 467                         oa.o_valid = OBD_MD_FLID |OBD_MD_FLTYPE |OBD_MD_FLMTIME;
 468                         oa.o_mtime = attr->ia_mtime;
 469                         err2 = obd_setattr(&sbi->ll_osc_conn, &oa, lsm, NULL);
 470                         if (err2) {
 471                                 CERROR("obd_setattr fails: rc=%d\n", err);
 472                                 if (!err)
 473                                         err = err2;
 474                         }
 475                 }
 476         }
 477
 478         RETURN(err);
 479 }
 480
 481 int ll_setattr_raw(struct inode *inode, struct iattr *attr)
 482 {
 483         struct lov_stripe_md *lsm = ll_i2info(inode)->lli_smd;
 484         struct ll_sb_info *sbi = ll_i2sbi(inode);
 485         struct ptlrpc_request *request = NULL;
 486         struct mdc_op_data op_data;
 487         int rc = 0, err;
 488         ENTRY;
 489         CDEBUG(D_VFSTRACE, "VFS Op:inode=%lu/%u(%p)\n", inode->i_ino,
 490                inode->i_generation, inode);
 491
 492         if ((attr->ia_valid & ATTR_SIZE)) {
 493                 struct ldlm_extent extent = {attr->ia_size, OBD_OBJECT_EOF};
 494                 struct lustre_handle lockh = { 0 };
 495
 496                 if (attr->ia_size > ll_file_maxbytes(inode))
 497                         RETURN(-EFBIG);
 498
 499                 /* writeback uses inode->i_size to determine how far out
 500                  * its cached pages go.  ll_truncate gets a PW lock, canceling
 501                  * our lock, _after_ it has updated i_size.  this can confuse
 502                  *
 503                  * If this file doesn't have stripes yet, it is already,
 504                  * by definition, truncated. */
 505                 if ((attr->ia_valid & ATTR_FROM_OPEN) && lsm == NULL) {
 506                         LASSERT(attr->ia_size == 0);
 507                         GOTO(skip_extent_lock, rc = 0);
 508                 }
 509
 510                 /* we really need to get our PW lock before we change
 511                  * inode->i_size.  if we don't we can race with other
 512                  * i_size updaters on our node, like ll_file_read.  we
 513                  * can also race with i_size propogation to other
 514                  * nodes through dirtying and writeback of final cached
 515                  * pages.  this last one is especially bad for racing
 516                  * o_append users on other nodes. */
 517                 rc = ll_extent_lock_no_validate(NULL, inode, lsm, LCK_PW,
 518                                                 &extent, &lockh);
 519                 if (rc != ELDLM_OK) {
 520                         if (rc > 0)
 521                                 RETURN(-ENOLCK);
 522                         RETURN(rc);
 523                 }
 524
 525                 rc = vmtruncate(inode, attr->ia_size);
 526                 if (rc == 0)
 527                         set_bit(LLI_F_HAVE_SIZE_LOCK,
 528                                 &ll_i2info(inode)->lli_flags);
 529
 530                 /* unlock now as we don't mind others file lockers racing with
 531                  * the mds updates below? */
 532                 err = ll_extent_unlock(NULL, inode, lsm, LCK_PW, &lockh);
 533                 if (err)
 534                         CERROR("ll_extent_unlock failed: %d\n", err);
 535                 if (rc)
 536                         RETURN(rc);
 537         }
 538
 539 skip_extent_lock:
 540         /* Don't send size changes to MDS to avoid "fast EA" problems, and
 541          * also avoid a pointless RPC (we get file size from OST anyways).
 542          */
 543         attr->ia_valid &= ~ATTR_SIZE;
 544         if (!attr->ia_valid)
 545                 RETURN(0);
 546
 547         ll_prepare_mdc_op_data(&op_data, inode, NULL, NULL, 0, 0);
 548
 549         err = mdc_setattr(&sbi->ll_mdc_conn, &op_data,
 550                           attr, NULL, 0, &request);
 551         if (err)
 552                 CERROR("mdc_setattr fails: err = %d\n", err);
 553
 554         ptlrpc_req_finished(request);
 555
 556         if (S_ISREG(inode->i_mode) && (attr->ia_valid & ATTR_MTIME_SET)) {
 557                 struct lov_stripe_md *lsm = ll_i2info(inode)->lli_smd;
 558                 struct obdo oa;
 559                 int err2;
 560
 561                 if (lsm == NULL) {
 562                         CDEBUG(D_INODE, "no lsm: not setting mtime on OSTs\n");
 563                         RETURN(err);
 564                 }
 565
 566                 CDEBUG(D_INODE, "set mtime on OST inode %lu to %lu\n",
 567                        inode->i_ino, attr->ia_mtime);
 568                 oa.o_id = lsm->lsm_object_id;
 569                 oa.o_mode = S_IFREG;
 570                 oa.o_valid = OBD_MD_FLID | OBD_MD_FLTYPE | OBD_MD_FLMTIME;
 571                 oa.o_mtime = attr->ia_mtime;
 572                 err2 = obd_setattr(&sbi->ll_osc_conn, &oa, lsm, NULL);
 573                 if (err2) {
 574                         CERROR("obd_setattr fails: rc=%d\n", err);
 575                         if (!err)
 576                                 err = err2;
 577                 }
 578         }
 579         RETURN(err);
 580 }
 581
 582 int ll_setattr(struct dentry *de, struct iattr *attr)
 583 {
 584         int rc = inode_change_ok(de->d_inode, attr);
 585         CDEBUG(D_VFSTRACE, "VFS Op:name=%s\n", de->d_name.name);
 586         if (rc)
 587                 return rc;
 588
 589         return ll_inode_setattr(de->d_inode, attr, 1);
 590 }
 591
 592 static int ll_statfs(struct super_block *sb, struct statfs *sfs)
 593 {
 594         struct ll_sb_info *sbi = ll_s2sbi(sb);
 595         struct obd_statfs osfs;
 596         int rc;
 597         ENTRY;
 598
 599         CDEBUG(D_VFSTRACE, "VFS Op:\n");
 600         memset(sfs, 0, sizeof(*sfs));
 601         rc = obd_statfs(&sbi->ll_mdc_conn, &osfs);
 602         statfs_unpack(sfs, &osfs);
 603         if (rc)
 604                 CERROR("mdc_statfs fails: rc = %d\n", rc);
 605         else
 606                 CDEBUG(D_SUPER, "mdc_statfs shows blocks "LPU64"/"LPU64
 607                        " objects "LPU64"/"LPU64"\n",
 608                        osfs.os_bavail, osfs.os_blocks,
 609                        osfs.os_ffree, osfs.os_files);
 610
 611         /* temporary until mds_statfs returns statfs info for all OSTs */
 612         if (!rc) {
 613                 rc = obd_statfs(&sbi->ll_osc_conn, &osfs);
 614                 if (rc) {
 615                         CERROR("obd_statfs fails: rc = %d\n", rc);
 616                         GOTO(out, rc);
 617                 }
 618                 CDEBUG(D_SUPER, "obd_statfs shows blocks "LPU64"/"LPU64
 619                        " objects "LPU64"/"LPU64"\n",
 620                        osfs.os_bavail, osfs.os_blocks,
 621                        osfs.os_ffree, osfs.os_files);
 622
 623                 while (osfs.os_blocks > ~0UL) {
 624                         sfs->f_bsize <<= 1;
 625
 626                         osfs.os_blocks >>= 1;
 627                         osfs.os_bfree >>= 1;
 628                         osfs.os_bavail >>= 1;
 629                 }
 630
 631                 sfs->f_blocks = osfs.os_blocks;
 632                 sfs->f_bfree = osfs.os_bfree;
 633                 sfs->f_bavail = osfs.os_bavail;
 634
 635                 /* If we don't have as many objects free on the OST as inodes
 636                  * on the MDS, we reduce the total number of inodes to
 637                  * compensate, so that the "inodes in use" number is correct.
 638                  */
 639                 if (osfs.os_ffree < (__u64)sfs->f_ffree) {
 640                         sfs->f_files = (sfs->f_files - sfs->f_ffree) +
 641                                        osfs.os_ffree;
 642                         sfs->f_ffree = osfs.os_ffree;
 643                 }
 644         }
 645
 646 out:
 647         RETURN(rc);
 648 }
 649
 650 void dump_lsm(int level, struct lov_stripe_md *lsm)
 651 {
 652         CDEBUG(level, "objid "LPX64", maxbytes "LPX64", magic %#08x, "
 653                "stripe_size %#08x, offset %u, stripe_count %u\n",
 654                lsm->lsm_object_id, lsm->lsm_maxbytes, lsm->lsm_magic,
 655                lsm->lsm_stripe_size, lsm->lsm_stripe_offset,
 656                lsm->lsm_stripe_count);
 657 }
 658
 659 void ll_update_inode(struct inode *inode, struct mds_body *body,
 660                      struct lov_stripe_md *lsm)
 661 {
 662         struct ll_inode_info *lli = ll_i2info(inode);
 663
 664         LASSERT ((lsm != NULL) == ((body->valid & OBD_MD_FLEASIZE) != 0));
 665         if (lsm != NULL) {
 666                 if (lli->lli_smd == NULL) {
 667                         lli->lli_maxbytes = lsm->lsm_maxbytes;
 668                         if (lli->lli_maxbytes > PAGE_CACHE_MAXBYTES)
 669                                 lli->lli_maxbytes = PAGE_CACHE_MAXBYTES;
 670                         lli->lli_smd = lsm;
 671                 } else {
 672                         if (memcmp(lli->lli_smd, lsm, sizeof(*lsm))) {
 673                                 CERROR("lsm mismatch for inode %ld\n",
 674                                        inode->i_ino);
 675                                 CERROR("lli_smd:\n");
 676                                 dump_lsm(D_ERROR, lli->lli_smd);
 677                                 CERROR("lsm:\n");
 678                                 dump_lsm(D_ERROR, lsm);
 679                                 LBUG();
 680                         }
 681                 }
 682         }
 683
 684         if (body->valid & OBD_MD_FLID)
 685                 inode->i_ino = body->ino;
 686         if (body->valid & OBD_MD_FLATIME)
 687                 LTIME_S(inode->i_atime) = body->atime;
 688         if (body->valid & OBD_MD_FLMTIME)
 689                 LTIME_S(inode->i_mtime) = body->mtime;
 690         if (body->valid & OBD_MD_FLCTIME)
 691                 LTIME_S(inode->i_ctime) = body->ctime;
 692         if (body->valid & OBD_MD_FLMODE)
 693                 inode->i_mode = (inode->i_mode & S_IFMT)|(body->mode & ~S_IFMT);
 694         if (body->valid & OBD_MD_FLTYPE)
 695                 inode->i_mode = (inode->i_mode & ~S_IFMT)|(body->mode & S_IFMT);
 696         if (body->valid & OBD_MD_FLUID)
 697                 inode->i_uid = body->uid;
 698         if (body->valid & OBD_MD_FLGID)
 699                 inode->i_gid = body->gid;
 700         if (body->valid & OBD_MD_FLFLAGS)
 701                 inode->i_flags = body->flags;
 702         if (body->valid & OBD_MD_FLNLINK)
 703                 inode->i_nlink = body->nlink;
 704         if (body->valid & OBD_MD_FLGENER)
 705                 inode->i_generation = body->generation;
 706         if (body->valid & OBD_MD_FLRDEV)
 707                 inode->i_rdev = body->rdev;
 708         if (body->valid & OBD_MD_FLSIZE)
 709                 inode->i_size = body->size;
 710         if (body->valid & OBD_MD_FLBLOCKS)
 711                 inode->i_blocks = body->blocks;
 712 }
 713
 714 static void ll_read_inode2(struct inode *inode, void *opaque)
 715 {
 716         struct ll_read_inode2_cookie *lic = opaque;
 717         struct mds_body *body = lic->lic_body;
 718         struct ll_inode_info *lli = ll_i2info(inode);
 719         ENTRY;
 720         CDEBUG(D_VFSTRACE, "VFS Op:inode=%lu/%u(%p)\n", inode->i_ino,
 721                inode->i_generation, inode);
 722
 723         sema_init(&lli->lli_open_sem, 1);
 724         spin_lock_init(&lli->lli_read_extent_lock);
 725         INIT_LIST_HEAD(&lli->lli_read_extents);
 726         ll_lldo_init(&lli->lli_dirty);
 727         lli->lli_flags = 0;
 728         /* We default to 2T-4k until the LSM is created/read, at which point
 729          * it'll be updated. */
 730         lli->lli_maxbytes = LUSTRE_STRIPE_MAXBYTES;
 731
 732         LASSERT(!lli->lli_smd);
 733
 734         /* core attributes from the MDS first */
 735         ll_update_inode(inode, body, lic->lic_lsm);
 736
 737         /* OIDEBUG(inode); */
 738
 739         if (S_ISREG(inode->i_mode)) {
 740                 inode->i_op = &ll_file_inode_operations;
 741                 inode->i_fop = &ll_file_operations;
 742                 inode->i_mapping->a_ops = &ll_aops;
 743                 EXIT;
 744         } else if (S_ISDIR(inode->i_mode)) {
 745                 inode->i_op = &ll_dir_inode_operations;
 746                 inode->i_fop = &ll_dir_operations;
 747                 inode->i_mapping->a_ops = &ll_dir_aops;
 748                 EXIT;
 749         } else if (S_ISLNK(inode->i_mode)) {
 750                 inode->i_op = &ll_fast_symlink_inode_operations;
 751                 EXIT;
 752         } else {
 753                 inode->i_op = &ll_special_inode_operations;
 754                 init_special_inode(inode, inode->i_mode, inode->i_rdev);
 755                 EXIT;
 756         }
 757 }
 758
 759 void ll_umount_begin(struct super_block *sb)
 760 {
 761         struct ll_sb_info *sbi = ll_s2sbi(sb);
 762         struct obd_device *obd;
 763         struct obd_ioctl_data ioc_data = { 0 };
 764
 765         ENTRY;
 766         CDEBUG(D_VFSTRACE, "VFS Op:\n");
 767
 768         obd = class_conn2obd(&sbi->ll_mdc_conn);
 769         obd->obd_no_recov = 1;
 770         obd_iocontrol(IOC_OSC_SET_ACTIVE, &sbi->ll_mdc_conn, sizeof ioc_data,
 771                       &ioc_data, NULL);
 772
 773         obd = class_conn2obd(&sbi->ll_osc_conn);
 774         obd->obd_no_recov = 1;
 775         obd_iocontrol(IOC_OSC_SET_ACTIVE, &sbi->ll_osc_conn, sizeof ioc_data,
 776                       &ioc_data, NULL);
 777
 778         /* Really, we'd like to wait until there are no requests outstanding,
 779          * and then continue.  For now, we just invalidate the requests,
 780          * schedule, and hope.
 781          */
 782         schedule();
 783
 784         EXIT;
 785 }
 786
 787 /* exported operations */
 788 struct super_operations ll_super_operations =
 789 {
 790         read_inode2: ll_read_inode2,
 791         clear_inode: ll_clear_inode,
 792         //        delete_inode: ll_delete_inode,
 793         put_super: ll_put_super,
 794         statfs: ll_statfs,
 795         umount_begin: ll_umount_begin
 796 };
 797
 798 static struct file_system_type lustre_lite_fs_type = {
 799         name:           "lustre_lite",
 800         fs_flags:       0,
 801         read_super:     ll_read_super,
 802         owner:          THIS_MODULE,
 803 };
 804
 805 static int __init init_lustre_lite(void)
 806 {
 807         printk(KERN_INFO "Lustre Lite Client File System; "
 808                "info@clusterfs.com\n");
 809         ll_file_data_slab = kmem_cache_create("ll_file_data",
 810                                               sizeof(struct ll_file_data), 0,
 811                                               SLAB_HWCACHE_ALIGN, NULL, NULL);
 812         if (ll_file_data_slab == NULL)
 813                 return -ENOMEM;
 814
 815         proc_lustre_fs_root = proc_lustre_root ? proc_mkdir("llite", proc_lustre_root) : NULL;
 816
 817         return register_filesystem(&lustre_lite_fs_type);
 818 }
 819
 820 static void __exit exit_lustre_lite(void)
 821 {
 822         unregister_filesystem(&lustre_lite_fs_type);
 823         kmem_cache_destroy(ll_file_data_slab);
 824
 825         if (proc_lustre_fs_root) {
 826                 lprocfs_remove(proc_lustre_fs_root);
 827                 proc_lustre_fs_root = NULL;
 828         }
 829 }
 830
 831 MODULE_AUTHOR("Cluster File Systems, Inc. <info@clusterfs.com>");
 832 MODULE_DESCRIPTION("Lustre Lite Client File System");
 833 MODULE_LICENSE("GPL");
 834
 835 module_init(init_lustre_lite);
 836 module_exit(exit_lustre_lite);
 837 #endif