lustre/liblustre/super.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/liblustre/super.c
  37  *
  38  * Lustre Light Super operations
  39  */
  40
  41 #define DEBUG_SUBSYSTEM S_LLITE
  42
  43 #include <stdlib.h>
  44 #include <string.h>
  45 #include <assert.h>
  46 #include <time.h>
  47 #include <sys/types.h>
  48 #include <sys/stat.h>
  49 #include <fcntl.h>
  50 #include <sys/queue.h>
  51 #ifndef __CYGWIN__
  52 # include <sys/statvfs.h>
  53 #else
  54 # include <sys/statfs.h>
  55 #endif
  56
  57 #include "llite_lib.h"
  58
  59 #ifndef MAY_EXEC
  60 #define MAY_EXEC        1
  61 #define MAY_WRITE       2
  62 #define MAY_READ        4
  63 #endif
  64
  65 #define S_IXUGO (S_IXUSR|S_IXGRP|S_IXOTH)
  66
  67 static int ll_permission(struct inode *inode, int mask)
  68 {
  69         struct intnl_stat *st = llu_i2stat(inode);
  70         mode_t mode = st->st_mode;
  71
  72         if (current->fsuid == st->st_uid)
  73                 mode >>= 6;
  74         else if (cfs_curproc_is_in_groups(st->st_gid))
  75                 mode >>= 3;
  76
  77         if ((mode & mask & (MAY_READ|MAY_WRITE|MAY_EXEC)) == mask)
  78                 return 0;
  79
  80         if ((mask & (MAY_READ|MAY_WRITE)) ||
  81             (st->st_mode & S_IXUGO))
  82                 if (cfs_capable(CFS_CAP_DAC_OVERRIDE))
  83                         return 0;
  84
  85         if (mask == MAY_READ ||
  86             (S_ISDIR(st->st_mode) && !(mask & MAY_WRITE))) {
  87                 if (cfs_capable(CFS_CAP_DAC_READ_SEARCH))
  88                         return 0;
  89         }
  90
  91         return -EACCES;
  92 }
  93
  94 static void llu_fsop_gone(struct filesys *fs)
  95 {
  96         struct llu_sb_info *sbi = (struct llu_sb_info *) fs->fs_private;
  97         struct obd_device *obd = class_exp2obd(sbi->ll_md_exp);
  98         int next = 0;
  99         ENTRY;
 100
 101         cfs_list_del(&sbi->ll_conn_chain);
 102         cl_sb_fini(sbi);
 103         obd_disconnect(sbi->ll_dt_exp);
 104         obd_disconnect(sbi->ll_md_exp);
 105
 106         while ((obd = class_devices_in_group(&sbi->ll_sb_uuid, &next)) != NULL)
 107                 class_manual_cleanup(obd);
 108
 109         OBD_FREE(sbi, sizeof(*sbi));
 110
 111         liblustre_wait_idle();
 112         EXIT;
 113 }
 114
 115 static struct inode_ops llu_inode_ops;
 116
 117 static ldlm_mode_t llu_take_md_lock(struct inode *inode, __u64 bits,
 118                                     struct lustre_handle *lockh)
 119 {
 120         ldlm_policy_data_t policy = { .l_inodebits = {bits}};
 121         struct lu_fid *fid;
 122         ldlm_mode_t rc;
 123         int flags;
 124         ENTRY;
 125
 126         fid = &llu_i2info(inode)->lli_fid;
 127         CDEBUG(D_INFO, "trying to match res "DFID"\n", PFID(fid));
 128
 129         flags = LDLM_FL_BLOCK_GRANTED | LDLM_FL_CBPENDING;
 130         rc = md_lock_match(llu_i2mdexp(inode), flags, fid, LDLM_IBITS, &policy,
 131                            LCK_CR|LCK_CW|LCK_PR|LCK_PW, lockh);
 132         RETURN(rc);
 133 }
 134
 135 void llu_update_inode(struct inode *inode, struct lustre_md *md)
 136 {
 137         struct llu_inode_info *lli = llu_i2info(inode);
 138         struct mdt_body *body = md->body;
 139         struct lov_stripe_md *lsm = md->lsm;
 140         struct intnl_stat *st = llu_i2stat(inode);
 141
 142         LASSERT ((lsm != NULL) == ((body->valid & OBD_MD_FLEASIZE) != 0));
 143
 144         if (body->valid & OBD_MD_FLMODE)
 145                 st->st_mode = (st->st_mode & S_IFMT)|(body->mode & ~S_IFMT);
 146         if (body->valid & OBD_MD_FLTYPE)
 147                 st->st_mode = (st->st_mode & ~S_IFMT)|(body->mode & S_IFMT);
 148
 149         if (lsm != NULL) {
 150                 if (lli->lli_smd == NULL) {
 151                         cl_inode_init(inode, md);
 152                         lli->lli_smd = lsm;
 153                         lli->lli_maxbytes = lsm->lsm_maxbytes;
 154                         if (lli->lli_maxbytes > PAGE_CACHE_MAXBYTES)
 155                                 lli->lli_maxbytes = PAGE_CACHE_MAXBYTES;
 156                 } else {
 157                         if (lov_stripe_md_cmp(lli->lli_smd, lsm)) {
 158                                 CERROR("lsm mismatch for inode %lld\n",
 159                                        (long long)st->st_ino);
 160                                 LBUG();
 161                         }
 162                 }
 163         }
 164
 165         if (body->valid & OBD_MD_FLATIME) {
 166                 if (body->atime > LTIME_S(st->st_atime))
 167                         LTIME_S(st->st_atime) = body->atime;
 168                 lli->lli_lvb.lvb_atime = body->atime;
 169         }
 170         if (body->valid & OBD_MD_FLMTIME) {
 171                 if (body->mtime > LTIME_S(st->st_mtime))
 172                         LTIME_S(st->st_mtime) = body->mtime;
 173                 lli->lli_lvb.lvb_mtime = body->mtime;
 174         }
 175         if (body->valid & OBD_MD_FLCTIME) {
 176                 if (body->ctime > LTIME_S(st->st_ctime))
 177                         LTIME_S(st->st_ctime) = body->ctime;
 178                 lli->lli_lvb.lvb_ctime = body->ctime;
 179         }
 180         if (S_ISREG(st->st_mode))
 181                 st->st_blksize = min(2UL * PTLRPC_MAX_BRW_SIZE, LL_MAX_BLKSIZE);
 182         else
 183                 st->st_blksize = 4096;
 184         if (body->valid & OBD_MD_FLUID)
 185                 st->st_uid = body->uid;
 186         if (body->valid & OBD_MD_FLGID)
 187                 st->st_gid = body->gid;
 188         if (body->valid & OBD_MD_FLNLINK)
 189                 st->st_nlink = body->nlink;
 190         if (body->valid & OBD_MD_FLRDEV)
 191                 st->st_rdev = body->rdev;
 192         if (body->valid & OBD_MD_FLFLAGS)
 193                 lli->lli_st_flags = body->flags;
 194         if (body->valid & OBD_MD_FLSIZE) {
 195                 if ((llu_i2sbi(inode)->ll_lco.lco_flags & OBD_CONNECT_SOM) &&
 196                     S_ISREG(st->st_mode) && lli->lli_smd) {
 197                         struct lustre_handle lockh;
 198                         ldlm_mode_t mode;
 199
 200                         /* As it is possible a blocking ast has been processed
 201                          * by this time, we need to check there is an UPDATE
 202                          * lock on the client and set LLIF_MDS_SIZE_LOCK holding
 203                          * it. */
 204                         mode = llu_take_md_lock(inode, MDS_INODELOCK_UPDATE,
 205                                                 &lockh);
 206                         if (mode) {
 207                                 st->st_size = body->size;
 208                                 lli->lli_flags |= LLIF_MDS_SIZE_LOCK;
 209                                 ldlm_lock_decref(&lockh, mode);
 210                         }
 211                 } else {
 212                     st->st_size = body->size;
 213                 }
 214
 215                 if (body->valid & OBD_MD_FLBLOCKS)
 216                         st->st_blocks = body->blocks;
 217         }
 218 }
 219
 220 void obdo_to_inode(struct inode *dst, struct obdo *src, obd_flag valid)
 221 {
 222         struct llu_inode_info *lli = llu_i2info(dst);
 223         struct intnl_stat *st = llu_i2stat(dst);
 224
 225         valid &= src->o_valid;
 226
 227         LASSERTF(!(valid & (OBD_MD_FLTYPE | OBD_MD_FLGENER | OBD_MD_FLFID |
 228                             OBD_MD_FLID | OBD_MD_FLGROUP)),
 229                  "object "LPU64"/"LPU64", valid %x\n",
 230                  src->o_id, src->o_seq, valid);
 231
 232         if (valid & (OBD_MD_FLCTIME | OBD_MD_FLMTIME))
 233                 CDEBUG(D_INODE,"valid "LPX64", cur time "CFS_TIME_T"/"CFS_TIME_T
 234                        ", new %lu/%lu\n",
 235                        src->o_valid,
 236                        LTIME_S(st->st_mtime), LTIME_S(st->st_ctime),
 237                        (long)src->o_mtime, (long)src->o_ctime);
 238
 239         if (valid & OBD_MD_FLATIME)
 240                 LTIME_S(st->st_atime) = src->o_atime;
 241         if (valid & OBD_MD_FLMTIME)
 242                 LTIME_S(st->st_mtime) = src->o_mtime;
 243         if (valid & OBD_MD_FLCTIME && src->o_ctime > LTIME_S(st->st_ctime))
 244                 LTIME_S(st->st_ctime) = src->o_ctime;
 245         if (valid & OBD_MD_FLSIZE)
 246                 st->st_size = src->o_size;
 247         if (valid & OBD_MD_FLBLOCKS) /* allocation of space */
 248                 st->st_blocks = src->o_blocks;
 249         if (valid & OBD_MD_FLBLKSZ)
 250                 st->st_blksize = src->o_blksize;
 251         if (valid & OBD_MD_FLTYPE)
 252                 st->st_mode = (st->st_mode & ~S_IFMT) | (src->o_mode & S_IFMT);
 253         if (valid & OBD_MD_FLMODE)
 254                 st->st_mode = (st->st_mode & S_IFMT) | (src->o_mode & ~S_IFMT);
 255         if (valid & OBD_MD_FLUID)
 256                 st->st_uid = src->o_uid;
 257         if (valid & OBD_MD_FLGID)
 258                 st->st_gid = src->o_gid;
 259         if (valid & OBD_MD_FLFLAGS)
 260                 lli->lli_st_flags = src->o_flags;
 261 }
 262
 263 /**
 264  * Performs the getattr on the inode and updates its fields.
 265  * If @sync != 0, perform the getattr under the server-side lock.
 266  */
 267 int llu_inode_getattr(struct inode *inode, struct obdo *obdo,
 268                       __u64 ioepoch, int sync)
 269 {
 270         struct llu_inode_info *lli = llu_i2info(inode);
 271         struct ptlrpc_request_set *set;
 272         struct lov_stripe_md *lsm = lli->lli_smd;
 273         struct obd_info oinfo = { { { 0 } } };
 274         int rc;
 275         ENTRY;
 276
 277         LASSERT(lsm);
 278
 279         oinfo.oi_md = lsm;
 280         oinfo.oi_oa = obdo;
 281         oinfo.oi_oa->o_id = lsm->lsm_object_id;
 282         oinfo.oi_oa->o_seq = lsm->lsm_object_seq;
 283         oinfo.oi_oa->o_mode = S_IFREG;
 284         oinfo.oi_oa->o_ioepoch = ioepoch;
 285         oinfo.oi_oa->o_valid = OBD_MD_FLID | OBD_MD_FLTYPE |
 286                                OBD_MD_FLSIZE | OBD_MD_FLBLOCKS |
 287                                OBD_MD_FLBLKSZ | OBD_MD_FLMTIME |
 288                                OBD_MD_FLCTIME | OBD_MD_FLGROUP |
 289                                OBD_MD_FLATIME | OBD_MD_FLEPOCH;
 290         obdo_from_inode(oinfo.oi_oa, NULL, &llu_i2info(inode)->lli_fid, 0);
 291         if (sync) {
 292                 oinfo.oi_oa->o_valid |= OBD_MD_FLFLAGS;
 293                 oinfo.oi_oa->o_flags |= OBD_FL_SRVLOCK;
 294         }
 295
 296         set = ptlrpc_prep_set();
 297         if (set == NULL) {
 298                 CERROR ("ENOMEM allocing request set\n");
 299                 rc = -ENOMEM;
 300         } else {
 301                 rc = obd_getattr_async(llu_i2obdexp(inode), &oinfo, set);
 302                 if (rc == 0)
 303                         rc = ptlrpc_set_wait(set);
 304                 ptlrpc_set_destroy(set);
 305         }
 306         if (rc)
 307                 RETURN(rc);
 308
 309         oinfo.oi_oa->o_valid = OBD_MD_FLBLOCKS | OBD_MD_FLBLKSZ |
 310                                OBD_MD_FLMTIME | OBD_MD_FLCTIME |
 311                                OBD_MD_FLSIZE;
 312
 313         obdo_refresh_inode(inode, oinfo.oi_oa, oinfo.oi_oa->o_valid);
 314         CDEBUG(D_INODE, "objid "LPX64" size %Lu, blocks %Lu, "
 315                "blksize %Lu\n", lli->lli_smd->lsm_object_id,
 316                (long long unsigned)llu_i2stat(inode)->st_size,
 317                (long long unsigned)llu_i2stat(inode)->st_blocks,
 318                (long long unsigned)llu_i2stat(inode)->st_blksize);
 319         RETURN(0);
 320 }
 321
 322 static struct inode* llu_new_inode(struct filesys *fs,
 323                                    struct lu_fid *fid)
 324 {
 325         struct inode *inode;
 326         struct llu_inode_info *lli;
 327         struct intnl_stat st = {
 328                 .st_dev  = 0,
 329 #if 0
 330 #ifndef AUTOMOUNT_FILE_NAME
 331                 .st_mode = fid->f_type & S_IFMT,
 332 #else
 333                 .st_mode = fid->f_type /* all of the bits! */
 334 #endif
 335 #endif
 336                 /* FIXME: fix this later */
 337                 .st_mode = 0,
 338
 339                 .st_uid  = geteuid(),
 340                 .st_gid  = getegid(),
 341         };
 342
 343         OBD_ALLOC(lli, sizeof(*lli));
 344         if (!lli)
 345                 return NULL;
 346
 347         /* initialize lli here */
 348         lli->lli_sbi = llu_fs2sbi(fs);
 349         lli->lli_smd = NULL;
 350         lli->lli_symlink_name = NULL;
 351         lli->lli_flags = 0;
 352         lli->lli_maxbytes = (__u64)(~0UL);
 353         lli->lli_file_data = NULL;
 354
 355         lli->lli_sysio_fid.fid_data = &lli->lli_fid;
 356         lli->lli_sysio_fid.fid_len = sizeof(lli->lli_fid);
 357         lli->lli_fid = *fid;
 358
 359         /* file identifier is needed by functions like _sysio_i_find() */
 360         inode = _sysio_i_new(fs, &lli->lli_sysio_fid,
 361                              &st, 0, &llu_inode_ops, lli);
 362
 363         if (!inode)
 364                 OBD_FREE(lli, sizeof(*lli));
 365
 366         return inode;
 367 }
 368
 369 static int llu_have_md_lock(struct inode *inode, __u64 lockpart)
 370 {
 371         struct lustre_handle lockh;
 372         ldlm_policy_data_t policy = { .l_inodebits = { lockpart } };
 373         struct lu_fid *fid;
 374         int flags;
 375         ENTRY;
 376
 377         LASSERT(inode);
 378
 379         fid = &llu_i2info(inode)->lli_fid;
 380         CDEBUG(D_INFO, "trying to match res "DFID"\n", PFID(fid));
 381
 382         flags = LDLM_FL_BLOCK_GRANTED | LDLM_FL_CBPENDING | LDLM_FL_TEST_LOCK;
 383         if (md_lock_match(llu_i2mdexp(inode), flags, fid, LDLM_IBITS, &policy,
 384                           LCK_CR|LCK_CW|LCK_PR|LCK_PW, &lockh)) {
 385                 RETURN(1);
 386         }
 387         RETURN(0);
 388 }
 389
 390 static int llu_inode_revalidate(struct inode *inode)
 391 {
 392         struct llu_inode_info *lli = llu_i2info(inode);
 393         struct intnl_stat *st = llu_i2stat(inode);
 394         ENTRY;
 395
 396         if (!inode) {
 397                 CERROR("REPORT THIS LINE TO PETER\n");
 398                 RETURN(0);
 399         }
 400
 401         if (!llu_have_md_lock(inode, MDS_INODELOCK_UPDATE)) {
 402                 struct lustre_md md;
 403                 struct ptlrpc_request *req = NULL;
 404                 struct llu_sb_info *sbi = llu_i2sbi(inode);
 405                 struct md_op_data op_data = { { 0 } };
 406                 unsigned long valid = OBD_MD_FLGETATTR;
 407                 int rc, ealen = 0;
 408
 409                 /* Why don't we update all valid MDS fields here, if we're
 410                  * doing an RPC anyways?  -phil */
 411                 if (S_ISREG(st->st_mode)) {
 412                         ealen = obd_size_diskmd(sbi->ll_dt_exp, NULL);
 413                         valid |= OBD_MD_FLEASIZE;
 414                 }
 415
 416                 llu_prep_md_op_data(&op_data, inode, NULL, NULL, 0, ealen,
 417                                     LUSTRE_OPC_ANY);
 418                 op_data.op_valid = valid;
 419
 420                 rc = md_getattr(sbi->ll_md_exp, &op_data, &req);
 421                 if (rc) {
 422                         CERROR("failure %d inode %llu\n", rc,
 423                                (long long)st->st_ino);
 424                         RETURN(-abs(rc));
 425                 }
 426                 rc = md_get_lustre_md(sbi->ll_md_exp, req,
 427                                       sbi->ll_dt_exp, sbi->ll_md_exp, &md);
 428
 429                 /* XXX Too paranoid? */
 430                 if (((md.body->valid ^ valid) & OBD_MD_FLEASIZE) &&
 431                     !((md.body->valid & OBD_MD_FLNLINK) &&
 432                       (md.body->nlink == 0))) {
 433                         CERROR("Asked for %s eadata but got %s (%d)\n",
 434                                (valid & OBD_MD_FLEASIZE) ? "some" : "no",
 435                                (md.body->valid & OBD_MD_FLEASIZE) ? "some":"none",
 436                                 md.body->eadatasize);
 437                 }
 438                 if (rc) {
 439                         ptlrpc_req_finished(req);
 440                         RETURN(rc);
 441                 }
 442
 443
 444                 llu_update_inode(inode, &md);
 445                 if (md.lsm != NULL && lli->lli_smd != md.lsm)
 446                         obd_free_memmd(sbi->ll_dt_exp, &md.lsm);
 447                 ptlrpc_req_finished(req);
 448         }
 449
 450         if (!lli->lli_smd) {
 451                 /* object not yet allocated, don't validate size */
 452                 st->st_atime = lli->lli_lvb.lvb_atime;
 453                 st->st_mtime = lli->lli_lvb.lvb_mtime;
 454                 st->st_ctime = lli->lli_lvb.lvb_ctime;
 455                 RETURN(0);
 456         }
 457
 458         /* ll_glimpse_size will prefer locally cached writes if they extend
 459          * the file */
 460         RETURN(cl_glimpse_size(inode));
 461 }
 462
 463 static void copy_stat_buf(struct inode *ino, struct intnl_stat *b)
 464 {
 465         *b = *llu_i2stat(ino);
 466 }
 467
 468 static int llu_iop_getattr(struct pnode *pno,
 469                            struct inode *ino,
 470                            struct intnl_stat *b)
 471 {
 472         int rc;
 473         ENTRY;
 474
 475         liblustre_wait_event(0);
 476
 477         if (!ino) {
 478                 LASSERT(pno);
 479                 LASSERT(pno->p_base->pb_ino);
 480                 ino = pno->p_base->pb_ino;
 481         } else {
 482                 LASSERT(!pno || pno->p_base->pb_ino == ino);
 483         }
 484
 485         /* libsysio might call us directly without intent lock,
 486          * we must re-fetch the attrs here
 487          */
 488         rc = llu_inode_revalidate(ino);
 489         if (!rc) {
 490                 copy_stat_buf(ino, b);
 491                 LASSERT(!llu_i2info(ino)->lli_it);
 492         }
 493
 494         liblustre_wait_event(0);
 495         RETURN(rc);
 496 }
 497
 498 static int null_if_equal(struct ldlm_lock *lock, void *data)
 499 {
 500         if (data == lock->l_ast_data) {
 501                 lock->l_ast_data = NULL;
 502
 503                 if (lock->l_req_mode != lock->l_granted_mode)
 504                         LDLM_ERROR(lock,"clearing inode with ungranted lock\n");
 505         }
 506
 507         return LDLM_ITER_CONTINUE;
 508 }
 509
 510 void llu_clear_inode(struct inode *inode)
 511 {
 512         struct llu_inode_info *lli = llu_i2info(inode);
 513         struct llu_sb_info *sbi = llu_i2sbi(inode);
 514         ENTRY;
 515
 516         CDEBUG(D_VFSTRACE, "VFS Op:inode=%llu/%lu(%p)\n",
 517                (long long)llu_i2stat(inode)->st_ino, lli->lli_st_generation,
 518                inode);
 519
 520         lli->lli_flags &= ~LLIF_MDS_SIZE_LOCK;
 521         md_change_cbdata(sbi->ll_md_exp, ll_inode2fid(inode),
 522                          null_if_equal, inode);
 523
 524         if (lli->lli_smd)
 525                 obd_change_cbdata(sbi->ll_dt_exp, lli->lli_smd,
 526                                   null_if_equal, inode);
 527
 528         cl_inode_fini(inode);
 529
 530         if (lli->lli_smd) {
 531                 obd_free_memmd(sbi->ll_dt_exp, &lli->lli_smd);
 532                 lli->lli_smd = NULL;
 533         }
 534
 535         if (lli->lli_symlink_name) {
 536                 OBD_FREE(lli->lli_symlink_name,
 537                          strlen(lli->lli_symlink_name) + 1);
 538                 lli->lli_symlink_name = NULL;
 539         }
 540
 541         EXIT;
 542 }
 543
 544 void llu_iop_gone(struct inode *inode)
 545 {
 546         struct llu_inode_info *lli = llu_i2info(inode);
 547         ENTRY;
 548
 549         liblustre_wait_event(0);
 550         llu_clear_inode(inode);
 551
 552         OBD_FREE(lli, sizeof(*lli));
 553         EXIT;
 554 }
 555
 556 static int inode_setattr(struct inode * inode, struct iattr * attr)
 557 {
 558         unsigned int ia_valid = attr->ia_valid;
 559         struct intnl_stat *st = llu_i2stat(inode);
 560         int error = 0;
 561
 562         /*
 563          * inode_setattr() is only ever invoked with ATTR_SIZE (by
 564          * llu_setattr_raw()) when file has no bodies. Check this.
 565          */
 566         LASSERT(ergo(ia_valid & ATTR_SIZE, llu_i2info(inode)->lli_smd == NULL));
 567
 568         if (ia_valid & ATTR_SIZE)
 569                 st->st_size = attr->ia_size;
 570         if (ia_valid & ATTR_UID)
 571                 st->st_uid = attr->ia_uid;
 572         if (ia_valid & ATTR_GID)
 573                 st->st_gid = attr->ia_gid;
 574         if (ia_valid & ATTR_ATIME)
 575                 st->st_atime = attr->ia_atime;
 576         if (ia_valid & ATTR_MTIME)
 577                 st->st_mtime = attr->ia_mtime;
 578         if (ia_valid & ATTR_CTIME)
 579                 st->st_ctime = attr->ia_ctime;
 580         if (ia_valid & ATTR_MODE) {
 581                 st->st_mode = attr->ia_mode;
 582                 if (!cfs_curproc_is_in_groups(st->st_gid) &&
 583                     !cfs_capable(CFS_CAP_FSETID))
 584                         st->st_mode &= ~S_ISGID;
 585         }
 586         /* mark_inode_dirty(inode); */
 587         return error;
 588 }
 589
 590 int llu_md_setattr(struct inode *inode, struct md_op_data *op_data,
 591                    struct md_open_data **mod)
 592 {
 593         struct lustre_md md;
 594         struct llu_sb_info *sbi = llu_i2sbi(inode);
 595         struct ptlrpc_request *request = NULL;
 596         int rc;
 597         ENTRY;
 598
 599         llu_prep_md_op_data(op_data, inode, NULL, NULL, 0, 0, LUSTRE_OPC_ANY);
 600         rc = md_setattr(sbi->ll_md_exp, op_data, NULL, 0, NULL,
 601                         0, &request, mod);
 602
 603         if (rc) {
 604                 ptlrpc_req_finished(request);
 605                 if (rc != -EPERM && rc != -EACCES)
 606                         CERROR("md_setattr fails: rc = %d\n", rc);
 607                 RETURN(rc);
 608         }
 609
 610         rc = md_get_lustre_md(sbi->ll_md_exp, request,
 611                               sbi->ll_dt_exp, sbi->ll_md_exp, &md);
 612         if (rc) {
 613                 ptlrpc_req_finished(request);
 614                 RETURN(rc);
 615         }
 616
 617         /* We call inode_setattr to adjust timestamps.
 618          * If there is at least some data in file, we cleared ATTR_SIZE
 619          * above to avoid invoking vmtruncate, otherwise it is important
 620          * to call vmtruncate in inode_setattr to update inode->i_size
 621          * (bug 6196) */
 622         inode_setattr(inode, &op_data->op_attr);
 623         llu_update_inode(inode, &md);
 624         ptlrpc_req_finished(request);
 625
 626         RETURN(rc);
 627 }
 628
 629 /* Close IO epoch and send Size-on-MDS attribute update. */
 630 static int llu_setattr_done_writing(struct inode *inode,
 631                                     struct md_op_data *op_data,
 632                                     struct md_open_data *mod)
 633 {
 634         struct llu_inode_info *lli = llu_i2info(inode);
 635         struct intnl_stat *st = llu_i2stat(inode);
 636         int rc = 0;
 637         ENTRY;
 638
 639         LASSERT(op_data != NULL);
 640         if (!S_ISREG(st->st_mode))
 641                 RETURN(0);
 642
 643         /* XXX: pass och here for the recovery purpose. */
 644         CDEBUG(D_INODE, "Epoch "LPU64" closed on "DFID" for truncate\n",
 645                op_data->op_ioepoch, PFID(&lli->lli_fid));
 646
 647         op_data->op_flags = MF_EPOCH_CLOSE;
 648         llu_done_writing_attr(inode, op_data);
 649         llu_pack_inode2opdata(inode, op_data, NULL);
 650
 651         rc = md_done_writing(llu_i2sbi(inode)->ll_md_exp, op_data, mod);
 652         if (rc == -EAGAIN) {
 653                 /* MDS has instructed us to obtain Size-on-MDS attribute
 654                  * from OSTs and send setattr to back to MDS. */
 655                 rc = llu_som_update(inode, op_data);
 656         } else if (rc) {
 657                 CERROR("inode %llu mdc truncate failed: rc = %d\n",
 658                        (unsigned long long)st->st_ino, rc);
 659         }
 660         RETURN(rc);
 661 }
 662
 663 /* If this inode has objects allocated to it (lsm != NULL), then the OST
 664  * object(s) determine the file size and mtime.  Otherwise, the MDS will
 665  * keep these values until such a time that objects are allocated for it.
 666  * We do the MDS operations first, as it is checking permissions for us.
 667  * We don't to the MDS RPC if there is nothing that we want to store there,
 668  * otherwise there is no harm in updating mtime/atime on the MDS if we are
 669  * going to do an RPC anyways.
 670  *
 671  * If we are doing a truncate, we will send the mtime and ctime updates
 672  * to the OST with the punch RPC, otherwise we do an explicit setattr RPC.
 673  * I don't believe it is possible to get e.g. ATTR_MTIME_SET and ATTR_SIZE
 674  * at the same time.
 675  */
 676 int llu_setattr_raw(struct inode *inode, struct iattr *attr)
 677 {
 678         struct lov_stripe_md *lsm = llu_i2info(inode)->lli_smd;
 679         struct intnl_stat *st = llu_i2stat(inode);
 680         int ia_valid = attr->ia_valid;
 681         struct md_op_data op_data = { { 0 } };
 682         struct md_open_data *mod = NULL;
 683         int rc = 0, rc1 = 0;
 684         ENTRY;
 685
 686         CDEBUG(D_VFSTRACE, "VFS Op:inode=%llu\n", (long long)st->st_ino);
 687
 688         if (ia_valid & ATTR_SIZE) {
 689                 if (attr->ia_size > ll_file_maxbytes(inode)) {
 690                         CDEBUG(D_INODE, "file too large %llu > "LPU64"\n",
 691                                (long long)attr->ia_size,
 692                                ll_file_maxbytes(inode));
 693                         RETURN(-EFBIG);
 694                 }
 695
 696                 attr->ia_valid |= ATTR_MTIME | ATTR_CTIME;
 697         }
 698
 699         /* We mark all of the fields "set" so MDS/OST does not re-set them */
 700         if (attr->ia_valid & ATTR_CTIME) {
 701                 attr->ia_ctime = CFS_CURRENT_TIME;
 702                 attr->ia_valid |= ATTR_CTIME_SET;
 703         }
 704         if (!(ia_valid & ATTR_ATIME_SET) && (attr->ia_valid & ATTR_ATIME)) {
 705                 attr->ia_atime = CFS_CURRENT_TIME;
 706                 attr->ia_valid |= ATTR_ATIME_SET;
 707         }
 708         if (!(ia_valid & ATTR_MTIME_SET) && (attr->ia_valid & ATTR_MTIME)) {
 709                 attr->ia_mtime = CFS_CURRENT_TIME;
 710                 attr->ia_valid |= ATTR_MTIME_SET;
 711         }
 712
 713         if (attr->ia_valid & (ATTR_MTIME | ATTR_CTIME))
 714                 CDEBUG(D_INODE, "setting mtime "CFS_TIME_T", ctime "CFS_TIME_T
 715                        ", now = "CFS_TIME_T"\n",
 716                        LTIME_S(attr->ia_mtime), LTIME_S(attr->ia_ctime),
 717                        LTIME_S(CFS_CURRENT_TIME));
 718
 719         /* NB: ATTR_SIZE will only be set after this point if the size
 720          * resides on the MDS, ie, this file has no objects. */
 721         if (lsm)
 722                 attr->ia_valid &= ~ATTR_SIZE;
 723
 724         /* If only OST attributes being set on objects, don't do MDS RPC.
 725          * In that case, we need to check permissions and update the local
 726          * inode ourselves so we can call obdo_from_inode() always. */
 727         if (ia_valid & (lsm ? ~(ATTR_FROM_OPEN | ATTR_RAW) : ~0)) {
 728                 memcpy(&op_data.op_attr, attr, sizeof(*attr));
 729
 730                 /* Open epoch for truncate. */
 731                 if (exp_connect_som(llu_i2mdexp(inode)) &&
 732                     (ia_valid & ATTR_SIZE))
 733                         op_data.op_flags = MF_EPOCH_OPEN;
 734                 rc = llu_md_setattr(inode, &op_data, &mod);
 735                 if (rc)
 736                         RETURN(rc);
 737
 738                 llu_ioepoch_open(llu_i2info(inode), op_data.op_ioepoch);
 739                 if (!lsm || !S_ISREG(st->st_mode)) {
 740                         CDEBUG(D_INODE, "no lsm: not setting attrs on OST\n");
 741                         GOTO(out, rc);
 742                 }
 743         } else {
 744                 /* The OST doesn't check permissions, but the alternative is
 745                  * a gratuitous RPC to the MDS.  We already rely on the client
 746                  * to do read/write/truncate permission checks, so is mtime OK?
 747                  */
 748                 if (ia_valid & (ATTR_MTIME | ATTR_ATIME)) {
 749                         /* from sys_utime() */
 750                         if (!(ia_valid & (ATTR_MTIME_SET | ATTR_ATIME_SET))) {
 751                                 if (current->fsuid != st->st_uid &&
 752                                     (rc = ll_permission(inode, MAY_WRITE)) != 0)
 753                                         RETURN(rc);
 754                         } else {
 755                                 /* from inode_change_ok() */
 756                                 if (current->fsuid != st->st_uid &&
 757                                     !cfs_capable(CFS_CAP_FOWNER))
 758                                         RETURN(-EPERM);
 759                         }
 760                 }
 761
 762
 763                 /* Won't invoke llu_vmtruncate(), as we already cleared
 764                  * ATTR_SIZE */
 765                 inode_setattr(inode, attr);
 766         }
 767
 768         if (ia_valid & ATTR_SIZE)
 769                 attr->ia_valid |= ATTR_SIZE;
 770         if (ia_valid & (ATTR_SIZE |
 771                         ATTR_ATIME | ATTR_ATIME_SET |
 772                         ATTR_MTIME | ATTR_MTIME_SET))
 773                 /* on truncate and utimes send attributes to osts, setting
 774                  * mtime/atime to past will be performed under PW 0:EOF extent
 775                  * lock (new_size:EOF for truncate)
 776                  * it may seem excessive to send mtime/atime updates to osts
 777                  * when not setting times to past, but it is necessary due to
 778                  * possible time de-synchronization */
 779                 rc = cl_setattr_ost(inode, attr, NULL);
 780         EXIT;
 781 out:
 782         if (op_data.op_ioepoch)
 783                 rc1 = llu_setattr_done_writing(inode, &op_data, mod);
 784         return rc ? rc : rc1;
 785 }
 786
 787 /* here we simply act as a thin layer to glue it with
 788  * llu_setattr_raw(), which is copy from kernel
 789  */
 790 static int llu_iop_setattr(struct pnode *pno,
 791                            struct inode *ino,
 792                            unsigned mask,
 793                            struct intnl_stat *stbuf)
 794 {
 795         struct iattr iattr;
 796         int rc;
 797         ENTRY;
 798
 799         liblustre_wait_event(0);
 800
 801         LASSERT(!(mask & ~(SETATTR_MTIME | SETATTR_ATIME |
 802                            SETATTR_UID | SETATTR_GID |
 803                            SETATTR_LEN | SETATTR_MODE)));
 804         memset(&iattr, 0, sizeof(iattr));
 805
 806         if (mask & SETATTR_MODE) {
 807                 iattr.ia_mode = stbuf->st_mode;
 808                 iattr.ia_valid |= ATTR_MODE;
 809         }
 810         if (mask & SETATTR_MTIME) {
 811                 iattr.ia_mtime = stbuf->st_mtime;
 812                 iattr.ia_valid |= ATTR_MTIME | ATTR_MTIME_SET;
 813         }
 814         if (mask & SETATTR_ATIME) {
 815                 iattr.ia_atime = stbuf->st_atime;
 816                 iattr.ia_valid |= ATTR_ATIME | ATTR_ATIME_SET;
 817         }
 818         if (mask & SETATTR_UID) {
 819                 iattr.ia_uid = stbuf->st_uid;
 820                 iattr.ia_valid |= ATTR_UID;
 821         }
 822         if (mask & SETATTR_GID) {
 823                 iattr.ia_gid = stbuf->st_gid;
 824                 iattr.ia_valid |= ATTR_GID;
 825         }
 826         if (mask & SETATTR_LEN) {
 827                 iattr.ia_size = stbuf->st_size; /* XXX signed expansion problem */
 828                 iattr.ia_valid |= ATTR_SIZE;
 829         }
 830
 831         iattr.ia_valid |= ATTR_RAW | ATTR_CTIME;
 832         iattr.ia_ctime = CFS_CURRENT_TIME;
 833
 834         rc = llu_setattr_raw(ino, &iattr);
 835         liblustre_wait_idle();
 836         RETURN(rc);
 837 }
 838
 839 #define EXT2_LINK_MAX           32000
 840
 841 static int llu_iop_symlink_raw(struct pnode *pno, const char *tgt)
 842 {
 843         struct inode *dir = pno->p_base->pb_parent->pb_ino;
 844         struct qstr *qstr = &pno->p_base->pb_name;
 845         const char *name = qstr->name;
 846         int len = qstr->len;
 847         struct ptlrpc_request *request = NULL;
 848         struct llu_sb_info *sbi = llu_i2sbi(dir);
 849         struct md_op_data op_data = {{ 0 }};
 850         int err = -EMLINK;
 851         ENTRY;
 852
 853         liblustre_wait_event(0);
 854         if (llu_i2stat(dir)->st_nlink >= EXT2_LINK_MAX)
 855                 RETURN(err);
 856
 857         llu_prep_md_op_data(&op_data, dir, NULL, name, len, 0,
 858                             LUSTRE_OPC_SYMLINK);
 859
 860         err = md_create(sbi->ll_md_exp, &op_data, tgt, strlen(tgt) + 1,
 861                         S_IFLNK | S_IRWXUGO, current->fsuid, current->fsgid,
 862                         cfs_curproc_cap_pack(), 0, &request);
 863         ptlrpc_req_finished(request);
 864         liblustre_wait_event(0);
 865         RETURN(err);
 866 }
 867
 868 static int llu_readlink_internal(struct inode *inode,
 869                                  struct ptlrpc_request **request,
 870                                  char **symname)
 871 {
 872         struct llu_inode_info *lli = llu_i2info(inode);
 873         struct llu_sb_info *sbi = llu_i2sbi(inode);
 874         struct mdt_body *body;
 875         struct intnl_stat *st = llu_i2stat(inode);
 876         struct md_op_data op_data = {{ 0 }};
 877         int rc, symlen = st->st_size + 1;
 878         ENTRY;
 879
 880         *request = NULL;
 881         *symname = NULL;
 882
 883         if (lli->lli_symlink_name) {
 884                 *symname = lli->lli_symlink_name;
 885                 CDEBUG(D_INODE, "using cached symlink %s\n", *symname);
 886                 RETURN(0);
 887         }
 888
 889         llu_prep_md_op_data(&op_data, inode, NULL, NULL, 0, symlen,
 890                             LUSTRE_OPC_ANY);
 891         op_data.op_valid = OBD_MD_LINKNAME;
 892
 893         rc = md_getattr(sbi->ll_md_exp, &op_data, request);
 894         if (rc) {
 895                 CERROR("inode %llu: rc = %d\n", (long long)st->st_ino, rc);
 896                 RETURN(rc);
 897         }
 898
 899         body = req_capsule_server_get(&(*request)->rq_pill, &RMF_MDT_BODY);
 900         LASSERT(body != NULL);
 901
 902         if ((body->valid & OBD_MD_LINKNAME) == 0) {
 903                 CERROR ("OBD_MD_LINKNAME not set on reply\n");
 904                 GOTO (failed, rc = -EPROTO);
 905         }
 906
 907         LASSERT(symlen != 0);
 908         if (body->eadatasize != symlen) {
 909                 CERROR("inode %llu: symlink length %d not expected %d\n",
 910                        (long long)st->st_ino, body->eadatasize - 1, symlen - 1);
 911                 GOTO(failed, rc = -EPROTO);
 912         }
 913
 914         *symname = req_capsule_server_get(&(*request)->rq_pill, &RMF_MDT_MD);
 915         if (*symname == NULL ||
 916             strnlen(*symname, symlen) != symlen - 1) {
 917                 /* not full/NULL terminated */
 918                 CERROR("inode %llu: symlink not NULL terminated string"
 919                        "of length %d\n", (long long)st->st_ino, symlen - 1);
 920                 GOTO(failed, rc = -EPROTO);
 921         }
 922
 923         OBD_ALLOC(lli->lli_symlink_name, symlen);
 924         /* do not return an error if we cannot cache the symlink locally */
 925         if (lli->lli_symlink_name)
 926                 memcpy(lli->lli_symlink_name, *symname, symlen);
 927
 928         RETURN(0);
 929
 930  failed:
 931         ptlrpc_req_finished (*request);
 932         RETURN (-EPROTO);
 933 }
 934
 935 static int llu_iop_readlink(struct pnode *pno, char *data, size_t bufsize)
 936 {
 937         struct inode *inode = pno->p_base->pb_ino;
 938         struct ptlrpc_request *request;
 939         char *symname;
 940         int rc;
 941         ENTRY;
 942
 943         liblustre_wait_event(0);
 944         rc = llu_readlink_internal(inode, &request, &symname);
 945         if (rc)
 946                 GOTO(out, rc);
 947
 948         LASSERT(symname);
 949         strncpy(data, symname, bufsize);
 950         rc = strlen(symname);
 951
 952         ptlrpc_req_finished(request);
 953  out:
 954         liblustre_wait_event(0);
 955         RETURN(rc);
 956 }
 957
 958 static int llu_iop_mknod_raw(struct pnode *pno,
 959                              mode_t mode,
 960                              dev_t dev)
 961 {
 962         struct ptlrpc_request *request = NULL;
 963         struct inode *dir = pno->p_parent->p_base->pb_ino;
 964         struct llu_sb_info *sbi = llu_i2sbi(dir);
 965         struct md_op_data op_data = {{ 0 }};
 966         int err = -EMLINK;
 967         ENTRY;
 968
 969         liblustre_wait_event(0);
 970         CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s,dir=%llu\n",
 971                (int)pno->p_base->pb_name.len, pno->p_base->pb_name.name,
 972                (long long)llu_i2stat(dir)->st_ino);
 973
 974         if (llu_i2stat(dir)->st_nlink >= EXT2_LINK_MAX)
 975                 RETURN(err);
 976
 977         switch (mode & S_IFMT) {
 978         case 0:
 979         case S_IFREG:
 980                 mode |= S_IFREG; /* for mode = 0 case, fallthrough */
 981         case S_IFCHR:
 982         case S_IFBLK:
 983         case S_IFIFO:
 984         case S_IFSOCK:
 985                 llu_prep_md_op_data(&op_data, dir, NULL,
 986                                     pno->p_base->pb_name.name,
 987                                     pno->p_base->pb_name.len, 0,
 988                                     LUSTRE_OPC_MKNOD);
 989
 990                 err = md_create(sbi->ll_md_exp, &op_data, NULL, 0, mode,
 991                                 current->fsuid, current->fsgid,
 992                                 cfs_curproc_cap_pack(), dev, &request);
 993                 ptlrpc_req_finished(request);
 994                 break;
 995         case S_IFDIR:
 996                 err = -EPERM;
 997                 break;
 998         default:
 999                 err = -EINVAL;
1000         }
1001         liblustre_wait_event(0);
1002         RETURN(err);
1003 }
1004
1005 static int llu_iop_link_raw(struct pnode *old, struct pnode *new)
1006 {
1007         struct inode *src = old->p_base->pb_ino;
1008         struct inode *dir = new->p_parent->p_base->pb_ino;
1009         const char *name = new->p_base->pb_name.name;
1010         int namelen = new->p_base->pb_name.len;
1011         struct ptlrpc_request *request = NULL;
1012         struct md_op_data op_data = {{ 0 }};
1013         int rc;
1014         ENTRY;
1015
1016         LASSERT(src);
1017         LASSERT(dir);
1018
1019         liblustre_wait_event(0);
1020         llu_prep_md_op_data(&op_data, src, dir, name, namelen, 0,
1021                             LUSTRE_OPC_ANY);
1022         rc = md_link(llu_i2sbi(src)->ll_md_exp, &op_data, &request);
1023         ptlrpc_req_finished(request);
1024         liblustre_wait_event(0);
1025
1026         RETURN(rc);
1027 }
1028
1029 /*
1030  * libsysio will clear the inode immediately after return
1031  */
1032 static int llu_iop_unlink_raw(struct pnode *pno)
1033 {
1034         struct inode *dir = pno->p_base->pb_parent->pb_ino;
1035         struct qstr *qstr = &pno->p_base->pb_name;
1036         const char *name = qstr->name;
1037         int len = qstr->len;
1038         struct inode *target = pno->p_base->pb_ino;
1039         struct ptlrpc_request *request = NULL;
1040         struct md_op_data op_data = { { 0 } };
1041         int rc;
1042         ENTRY;
1043
1044         LASSERT(target);
1045
1046         liblustre_wait_event(0);
1047         llu_prep_md_op_data(&op_data, dir, NULL, name, len, 0,
1048                             LUSTRE_OPC_ANY);
1049         rc = md_unlink(llu_i2sbi(dir)->ll_md_exp, &op_data, &request);
1050         if (!rc)
1051                 rc = llu_objects_destroy(request, dir);
1052         ptlrpc_req_finished(request);
1053         liblustre_wait_idle();
1054
1055         RETURN(rc);
1056 }
1057
1058 static int llu_iop_rename_raw(struct pnode *old, struct pnode *new)
1059 {
1060         struct inode *src = old->p_parent->p_base->pb_ino;
1061         struct inode *tgt = new->p_parent->p_base->pb_ino;
1062         const char *oldname = old->p_base->pb_name.name;
1063         int oldnamelen = old->p_base->pb_name.len;
1064         const char *newname = new->p_base->pb_name.name;
1065         int newnamelen = new->p_base->pb_name.len;
1066         struct ptlrpc_request *request = NULL;
1067         struct md_op_data op_data = { { 0 } };
1068         int rc;
1069         ENTRY;
1070
1071         LASSERT(src);
1072         LASSERT(tgt);
1073
1074         liblustre_wait_event(0);
1075         llu_prep_md_op_data(&op_data, src, tgt, NULL, 0, 0,
1076                             LUSTRE_OPC_ANY);
1077         rc = md_rename(llu_i2sbi(src)->ll_md_exp, &op_data,
1078                        oldname, oldnamelen, newname, newnamelen,
1079                        &request);
1080         if (!rc) {
1081                 rc = llu_objects_destroy(request, src);
1082         }
1083
1084         ptlrpc_req_finished(request);
1085         liblustre_wait_idle();
1086
1087         RETURN(rc);
1088 }
1089
1090 #ifdef _HAVE_STATVFS
1091 static int llu_statfs_internal(struct llu_sb_info *sbi,
1092                                struct obd_statfs *osfs, __u64 max_age)
1093 {
1094         struct obd_statfs obd_osfs;
1095         int rc;
1096         ENTRY;
1097
1098         rc = obd_statfs(class_exp2obd(sbi->ll_md_exp), osfs, max_age, 0);
1099         if (rc) {
1100                 CERROR("md_statfs fails: rc = %d\n", rc);
1101                 RETURN(rc);
1102         }
1103
1104         CDEBUG(D_SUPER, "MDC blocks "LPU64"/"LPU64" objects "LPU64"/"LPU64"\n",
1105                osfs->os_bavail, osfs->os_blocks, osfs->os_ffree,osfs->os_files);
1106
1107         rc = obd_statfs_rqset(class_exp2obd(sbi->ll_dt_exp),
1108                               &obd_statfs, max_age, 0);
1109         if (rc) {
1110                 CERROR("obd_statfs fails: rc = %d\n", rc);
1111                 RETURN(rc);
1112         }
1113
1114         CDEBUG(D_SUPER, "OSC blocks "LPU64"/"LPU64" objects "LPU64"/"LPU64"\n",
1115                obd_osfs.os_bavail, obd_osfs.os_blocks, obd_osfs.os_ffree,
1116                obd_osfs.os_files);
1117
1118         osfs->os_blocks = obd_osfs.os_blocks;
1119         osfs->os_bfree = obd_osfs.os_bfree;
1120         osfs->os_bavail = obd_osfs.os_bavail;
1121
1122         /* If we don't have as many objects free on the OST as inodes
1123          * on the MDS, we reduce the total number of inodes to
1124          * compensate, so that the "inodes in use" number is correct.
1125          */
1126         if (obd_osfs.os_ffree < osfs->os_ffree) {
1127                 osfs->os_files = (osfs->os_files - osfs->os_ffree) +
1128                         obd_osfs.os_ffree;
1129                 osfs->os_ffree = obd_osfs.os_ffree;
1130         }
1131
1132         RETURN(rc);
1133 }
1134
1135 static int llu_statfs(struct llu_sb_info *sbi, struct statfs *sfs)
1136 {
1137         struct obd_statfs osfs;
1138         int rc;
1139
1140         CDEBUG(D_VFSTRACE, "VFS Op:\n");
1141
1142         /* For now we will always get up-to-date statfs values, but in the
1143          * future we may allow some amount of caching on the client (e.g.
1144          * from QOS or lprocfs updates). */
1145         rc = llu_statfs_internal(sbi, &osfs,
1146                                  cfs_time_shift_64(-OBD_STATFS_CACHE_SECONDS));
1147         if (rc)
1148                 return rc;
1149
1150         statfs_unpack(sfs, &osfs);
1151
1152         if (sizeof(sfs->f_blocks) == 4) {
1153                 while (osfs.os_blocks > ~0UL) {
1154                         sfs->f_bsize <<= 1;
1155
1156                         osfs.os_blocks >>= 1;
1157                         osfs.os_bfree >>= 1;
1158                         osfs.os_bavail >>= 1;
1159                 }
1160         }
1161
1162         sfs->f_blocks = osfs.os_blocks;
1163         sfs->f_bfree = osfs.os_bfree;
1164         sfs->f_bavail = osfs.os_bavail;
1165
1166         return 0;
1167 }
1168
1169 static int llu_iop_statvfs(struct pnode *pno,
1170                            struct inode *ino,
1171                            struct intnl_statvfs *buf)
1172 {
1173         struct statfs fs;
1174         int rc;
1175         ENTRY;
1176
1177         liblustre_wait_event(0);
1178
1179 #ifndef __CYGWIN__
1180         LASSERT(pno->p_base->pb_ino);
1181         rc = llu_statfs(llu_i2sbi(pno->p_base->pb_ino), &fs);
1182         if (rc)
1183                 RETURN(rc);
1184
1185         /* from native driver */
1186         buf->f_bsize = fs.f_bsize;  /* file system block size */
1187         buf->f_frsize = fs.f_bsize; /* file system fundamental block size */
1188         buf->f_blocks = fs.f_blocks;
1189         buf->f_bfree = fs.f_bfree;
1190         buf->f_bavail = fs.f_bavail;
1191         buf->f_files = fs.f_files;  /* Total number serial numbers */
1192         buf->f_ffree = fs.f_ffree;  /* Number free serial numbers */
1193         buf->f_favail = fs.f_ffree; /* Number free ser num for non-privileged*/
1194         buf->f_fsid = fs.f_fsid.__val[1];
1195         buf->f_flag = 0;            /* No equiv in statfs; maybe use type? */
1196         buf->f_namemax = fs.f_namelen;
1197 #endif
1198
1199         liblustre_wait_event(0);
1200         RETURN(0);
1201 }
1202 #endif /* _HAVE_STATVFS */
1203
1204 static int llu_iop_mkdir_raw(struct pnode *pno, mode_t mode)
1205 {
1206         struct inode *dir = pno->p_base->pb_parent->pb_ino;
1207         struct qstr *qstr = &pno->p_base->pb_name;
1208         const char *name = qstr->name;
1209         int len = qstr->len;
1210         struct ptlrpc_request *request = NULL;
1211         struct intnl_stat *st = llu_i2stat(dir);
1212         struct md_op_data op_data = {{ 0 }};
1213         int err = -EMLINK;
1214         ENTRY;
1215
1216         liblustre_wait_event(0);
1217         CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s,dir=%llu/%lu(%p)\n", len, name,
1218                (long long)st->st_ino, llu_i2info(dir)->lli_st_generation, dir);
1219
1220         if (st->st_nlink >= EXT2_LINK_MAX)
1221                 RETURN(err);
1222
1223         llu_prep_md_op_data(&op_data, dir, NULL, name, len, 0,
1224                             LUSTRE_OPC_MKDIR);
1225
1226         err = md_create(llu_i2sbi(dir)->ll_md_exp, &op_data, NULL, 0,
1227                         mode | S_IFDIR, current->fsuid, current->fsgid,
1228                         cfs_curproc_cap_pack(), 0, &request);
1229         ptlrpc_req_finished(request);
1230         liblustre_wait_event(0);
1231         RETURN(err);
1232 }
1233
1234 static int llu_iop_rmdir_raw(struct pnode *pno)
1235 {
1236         struct inode *dir = pno->p_base->pb_parent->pb_ino;
1237         struct qstr *qstr = &pno->p_base->pb_name;
1238         const char *name = qstr->name;
1239         int len = qstr->len;
1240         struct ptlrpc_request *request = NULL;
1241         struct md_op_data op_data = {{ 0 }};
1242         int rc;
1243         ENTRY;
1244
1245         liblustre_wait_event(0);
1246         CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s,dir=%llu/%lu(%p)\n", len, name,
1247                (long long)llu_i2stat(dir)->st_ino,
1248                llu_i2info(dir)->lli_st_generation, dir);
1249
1250         llu_prep_md_op_data(&op_data, dir, NULL, name, len, S_IFDIR,
1251                             LUSTRE_OPC_ANY);
1252         rc = md_unlink(llu_i2sbi(dir)->ll_md_exp, &op_data, &request);
1253         ptlrpc_req_finished(request);
1254
1255         liblustre_wait_event(0);
1256         RETURN(rc);
1257 }
1258
1259 #ifdef O_DIRECT
1260 #define FCNTL_FLMASK (O_APPEND|O_NONBLOCK|O_ASYNC|O_DIRECT)
1261 #else
1262 #define FCNTL_FLMASK (O_APPEND|O_NONBLOCK|O_ASYNC)
1263 #endif
1264 #define FCNTL_FLMASK_INVALID (O_NONBLOCK|O_ASYNC)
1265
1266 /* refer to ll_file_flock() for details */
1267 static int llu_file_flock(struct inode *ino,
1268                           int cmd,
1269                           struct file_lock *file_lock)
1270 {
1271         struct llu_inode_info *lli = llu_i2info(ino);
1272         struct intnl_stat *st = llu_i2stat(ino);
1273         struct ldlm_res_id res_id =
1274                 { .name = {fid_seq(&lli->lli_fid),
1275                            fid_oid(&lli->lli_fid),
1276                            fid_ver(&lli->lli_fid),
1277                            LDLM_FLOCK} };
1278         struct ldlm_enqueue_info einfo = { LDLM_FLOCK, 0, NULL,
1279                 ldlm_flock_completion_ast, NULL, NULL, file_lock };
1280
1281         struct lustre_handle lockh = {0};
1282         ldlm_policy_data_t flock;
1283         int flags = 0;
1284         int rc;
1285
1286         CDEBUG(D_VFSTRACE, "VFS Op:inode=%llu file_lock=%p\n",
1287                (unsigned long long)st->st_ino, file_lock);
1288
1289         flock.l_flock.pid = file_lock->fl_pid;
1290         flock.l_flock.start = file_lock->fl_start;
1291         flock.l_flock.end = file_lock->fl_end;
1292
1293         switch (file_lock->fl_type) {
1294         case F_RDLCK:
1295                 einfo.ei_mode = LCK_PR;
1296                 break;
1297         case F_UNLCK:
1298                 einfo.ei_mode = LCK_NL;
1299                 break;
1300         case F_WRLCK:
1301                 einfo.ei_mode = LCK_PW;
1302                 break;
1303         default:
1304                 CERROR("unknown fcntl lock type: %d\n", file_lock->fl_type);
1305                 LBUG();
1306         }
1307
1308         switch (cmd) {
1309         case F_SETLKW:
1310 #ifdef F_SETLKW64
1311 #if F_SETLKW64 != F_SETLKW
1312         case F_SETLKW64:
1313 #endif
1314 #endif
1315                 flags = 0;
1316                 break;
1317         case F_SETLK:
1318 #ifdef F_SETLK64
1319 #if F_SETLK64 != F_SETLK
1320         case F_SETLK64:
1321 #endif
1322 #endif
1323                 flags = LDLM_FL_BLOCK_NOWAIT;
1324                 break;
1325         case F_GETLK:
1326 #ifdef F_GETLK64
1327 #if F_GETLK64 != F_GETLK
1328         case F_GETLK64:
1329 #endif
1330 #endif
1331                 flags = LDLM_FL_TEST_LOCK;
1332                 file_lock->fl_type = einfo.ei_mode;
1333                 break;
1334         default:
1335                 CERROR("unknown fcntl cmd: %d\n", cmd);
1336                 LBUG();
1337         }
1338
1339         CDEBUG(D_DLMTRACE, "inode=%llu, pid=%u, cmd=%d, flags=%#x, mode=%u, "
1340                "start="LPX64", end="LPX64"\n", (unsigned long long)st->st_ino,
1341                flock.l_flock.pid, cmd, flags, einfo.ei_mode, flock.l_flock.start,
1342                flock.l_flock.end);
1343
1344         {
1345                 struct lmv_obd *lmv;
1346                 struct obd_device *lmv_obd;
1347                 lmv_obd = class_exp2obd(llu_i2mdexp(ino));
1348                 lmv = &lmv_obd->u.lmv;
1349
1350                 if (lmv->desc.ld_tgt_count < 1)
1351                         RETURN(rc = -ENODEV);
1352
1353                 if (lmv->tgts[0].ltd_exp != NULL)
1354                         rc = ldlm_cli_enqueue(lmv->tgts[0].ltd_exp, NULL, &einfo, &res_id,
1355                                               &flock, &flags, NULL, 0, &lockh, 0);
1356                 else
1357                         rc = -ENODEV;
1358         }
1359         RETURN(rc);
1360 }
1361
1362 static int assign_type(struct file_lock *fl, int type)
1363 {
1364         switch (type) {
1365         case F_RDLCK:
1366         case F_WRLCK:
1367         case F_UNLCK:
1368                 fl->fl_type = type;
1369                 return 0;
1370         default:
1371                 return -EINVAL;
1372         }
1373 }
1374
1375 static int flock_to_posix_lock(struct inode *ino,
1376                                struct file_lock *fl,
1377                                struct flock *l)
1378 {
1379         switch (l->l_whence) {
1380         /* XXX: only SEEK_SET is supported in lustre */
1381         case SEEK_SET:
1382                 fl->fl_start = 0;
1383                 break;
1384         default:
1385                 return -EINVAL;
1386         }
1387
1388         fl->fl_end = l->l_len - 1;
1389         if (l->l_len < 0)
1390                 return -EINVAL;
1391         if (l->l_len == 0)
1392                 fl->fl_end = OFFSET_MAX;
1393
1394         fl->fl_pid = getpid();
1395         fl->fl_flags = FL_POSIX;
1396         fl->fl_notify = NULL;
1397         fl->fl_insert = NULL;
1398         fl->fl_remove = NULL;
1399         /* XXX: these fields can't be filled with suitable values,
1400                 but I think lustre doesn't use them.
1401          */
1402         fl->fl_owner = NULL;
1403         fl->fl_file = NULL;
1404
1405         return assign_type(fl, l->l_type);
1406 }
1407
1408 static int llu_fcntl_getlk(struct inode *ino, struct flock *flock)
1409 {
1410         struct file_lock fl;
1411         int error;
1412
1413         error = EINVAL;
1414         if ((flock->l_type != F_RDLCK) && (flock->l_type != F_WRLCK))
1415                 goto out;
1416
1417         error = flock_to_posix_lock(ino, &fl, flock);
1418         if (error)
1419                 goto out;
1420
1421         error = llu_file_flock(ino, F_GETLK, &fl);
1422         if (error)
1423                 goto out;
1424
1425         flock->l_type = F_UNLCK;
1426         if (fl.fl_type != F_UNLCK) {
1427                 flock->l_pid = fl.fl_pid;
1428                 flock->l_start = fl.fl_start;
1429                 flock->l_len = fl.fl_end == OFFSET_MAX ? 0:
1430                         fl.fl_end - fl.fl_start + 1;
1431                 flock->l_whence = SEEK_SET;
1432                 flock->l_type = fl.fl_type;
1433         }
1434
1435 out:
1436         return error;
1437 }
1438
1439 static int llu_fcntl_setlk(struct inode *ino, int cmd, struct flock *flock)
1440 {
1441         struct file_lock fl;
1442         int flags = llu_i2info(ino)->lli_open_flags + 1;
1443         int error;
1444
1445         error = flock_to_posix_lock(ino, &fl, flock);
1446         if (error)
1447                 goto out;
1448         if (cmd == F_SETLKW)
1449                 fl.fl_flags |= FL_SLEEP;
1450
1451         error = -EBADF;
1452         switch (flock->l_type) {
1453         case F_RDLCK:
1454                 if (!(flags & FMODE_READ))
1455                         goto out;
1456                 break;
1457         case F_WRLCK:
1458                 if (!(flags & FMODE_WRITE))
1459                         goto out;
1460                 break;
1461         case F_UNLCK:
1462                 break;
1463         default:
1464                 error = -EINVAL;
1465                 goto out;
1466         }
1467
1468         error = llu_file_flock(ino, cmd, &fl);
1469         if (error)
1470                 goto out;
1471
1472 out:
1473         return error;
1474 }
1475
1476 static int llu_iop_fcntl(struct inode *ino, int cmd, va_list ap, int *rtn)
1477 {
1478         struct llu_inode_info *lli = llu_i2info(ino);
1479         long flags;
1480         struct flock *flock;
1481         long err = 0;
1482
1483         liblustre_wait_event(0);
1484         switch (cmd) {
1485         case F_GETFL:
1486                 *rtn = lli->lli_open_flags;
1487                 break;
1488         case F_SETFL:
1489                 flags = va_arg(ap, long);
1490                 flags &= FCNTL_FLMASK;
1491                 if (flags & FCNTL_FLMASK_INVALID) {
1492                         LCONSOLE_ERROR_MSG(0x010, "liblustre does not support "
1493                                            "the O_NONBLOCK or O_ASYNC flags. "
1494                                            "Please fix your application.\n");
1495                         *rtn = -EINVAL;
1496                         err = EINVAL;
1497                         break;
1498                 }
1499                 lli->lli_open_flags = (int)(flags & FCNTL_FLMASK) |
1500                                       (lli->lli_open_flags & ~FCNTL_FLMASK);
1501                 *rtn = 0;
1502                 break;
1503         case F_GETLK:
1504 #ifdef F_GETLK64
1505 #if F_GETLK64 != F_GETLK
1506         case F_GETLK64:
1507 #endif
1508 #endif
1509                 flock = va_arg(ap, struct flock *);
1510                 err = llu_fcntl_getlk(ino, flock);
1511                 *rtn = err? -1: 0;
1512                 break;
1513         case F_SETLK:
1514 #ifdef F_SETLKW64
1515 #if F_SETLKW64 != F_SETLKW
1516         case F_SETLKW64:
1517 #endif
1518 #endif
1519         case F_SETLKW:
1520 #ifdef F_SETLK64
1521 #if F_SETLK64 != F_SETLK
1522         case F_SETLK64:
1523 #endif
1524 #endif
1525                 flock = va_arg(ap, struct flock *);
1526                 err = llu_fcntl_setlk(ino, cmd, flock);
1527                 *rtn = err? -1: 0;
1528                 break;
1529         default:
1530                 CERROR("unsupported fcntl cmd %x\n", cmd);
1531                 *rtn = -ENOSYS;
1532                 err = ENOSYS;
1533                 break;
1534         }
1535
1536         liblustre_wait_event(0);
1537         return err;
1538 }
1539
1540 static int llu_get_grouplock(struct inode *inode, unsigned long arg)
1541 {
1542         struct llu_inode_info *lli = llu_i2info(inode);
1543         struct ll_file_data *fd = lli->lli_file_data;
1544         int rc;
1545         struct ccc_grouplock grouplock;
1546         ENTRY;
1547
1548         if (fd->fd_flags & LL_FILE_IGNORE_LOCK) {
1549                 RETURN(-ENOTSUPP);
1550         }
1551         if (fd->fd_flags & LL_FILE_GROUP_LOCKED) {
1552                 RETURN(-EINVAL);
1553         }
1554         LASSERT(fd->fd_grouplock.cg_lock == NULL);
1555
1556         rc = cl_get_grouplock(cl_i2info(inode)->lli_clob,
1557                               arg, (lli->lli_open_flags & O_NONBLOCK),
1558                               &grouplock);
1559
1560         if (rc)
1561                 RETURN(rc);
1562
1563         fd->fd_flags |= LL_FILE_GROUP_LOCKED;
1564         fd->fd_grouplock = grouplock;
1565
1566         RETURN(0);
1567 }
1568
1569 int llu_put_grouplock(struct inode *inode, unsigned long arg)
1570 {
1571         struct llu_inode_info *lli = llu_i2info(inode);
1572         struct ll_file_data *fd = lli->lli_file_data;
1573         struct ccc_grouplock grouplock;
1574         ENTRY;
1575
1576         if (!(fd->fd_flags & LL_FILE_GROUP_LOCKED))
1577                 RETURN(-EINVAL);
1578
1579         LASSERT(fd->fd_grouplock.cg_lock != NULL);
1580
1581         if (fd->fd_grouplock.cg_gid != arg)
1582                 RETURN(-EINVAL);
1583
1584         grouplock = fd->fd_grouplock;
1585         memset(&fd->fd_grouplock, 0, sizeof(fd->fd_grouplock));
1586         fd->fd_flags &= ~LL_FILE_GROUP_LOCKED;
1587
1588         cl_put_grouplock(&grouplock);
1589
1590         RETURN(0);
1591 }
1592
1593 static int llu_lov_dir_setstripe(struct inode *ino, unsigned long arg)
1594 {
1595         struct llu_sb_info *sbi = llu_i2sbi(ino);
1596         struct ptlrpc_request *request = NULL;
1597         struct md_op_data op_data = {{ 0 }};
1598         struct lov_user_md lum, *lump = (struct lov_user_md *)arg;
1599         int rc = 0;
1600
1601         llu_prep_md_op_data(&op_data, ino, NULL, NULL, 0, 0,
1602                             LUSTRE_OPC_ANY);
1603
1604         LASSERT(sizeof(lum) == sizeof(*lump));
1605         LASSERT(sizeof(lum.lmm_objects[0]) ==
1606                 sizeof(lump->lmm_objects[0]));
1607         if (cfs_copy_from_user(&lum, lump, sizeof(lum)))
1608                 return(-EFAULT);
1609
1610         switch (lum.lmm_magic) {
1611         case LOV_USER_MAGIC_V1: {
1612                 if (lum.lmm_magic != cpu_to_le32(LOV_USER_MAGIC_V1))
1613                         lustre_swab_lov_user_md_v1(&lum);
1614                 break;
1615                 }
1616         case LOV_USER_MAGIC_V3: {
1617                 if (lum.lmm_magic != cpu_to_le32(LOV_USER_MAGIC_V3))
1618                         lustre_swab_lov_user_md_v3((struct lov_user_md_v3 *)&lum);
1619                 break;
1620                 }
1621         default: {
1622                 CDEBUG(D_IOCTL, "bad userland LOV MAGIC:"
1623                                 " %#08x != %#08x nor %#08x\n",
1624                                 lum.lmm_magic, LOV_USER_MAGIC_V1,
1625                                 LOV_USER_MAGIC_V3);
1626                 RETURN(-EINVAL);
1627         }
1628         }
1629
1630         /* swabbing is done in lov_setstripe() on server side */
1631         rc = md_setattr(sbi->ll_md_exp, &op_data, &lum,
1632                         sizeof(lum), NULL, 0, &request, NULL);
1633         if (rc) {
1634                 ptlrpc_req_finished(request);
1635                 if (rc != -EPERM && rc != -EACCES)
1636                         CERROR("md_setattr fails: rc = %d\n", rc);
1637                 return rc;
1638         }
1639         ptlrpc_req_finished(request);
1640
1641         return rc;
1642 }
1643
1644 static int llu_lov_setstripe_ea_info(struct inode *ino, int flags,
1645                                      struct lov_user_md *lum, int lum_size)
1646 {
1647         struct llu_sb_info *sbi = llu_i2sbi(ino);
1648         struct llu_inode_info *lli = llu_i2info(ino);
1649         struct lookup_intent oit = {.it_op = IT_OPEN, .it_flags = flags};
1650         struct ldlm_enqueue_info einfo = { LDLM_IBITS, LCK_CR,
1651                 llu_md_blocking_ast, ldlm_completion_ast, NULL, NULL, NULL };
1652         struct ptlrpc_request *req = NULL;
1653         struct lustre_md md;
1654         struct md_op_data data = {{ 0 }};
1655         struct lustre_handle lockh;
1656         int rc = 0;
1657         ENTRY;
1658
1659         if (lli->lli_smd) {
1660                 CDEBUG(D_IOCTL, "stripe already exists for ino "DFID"\n",
1661                        PFID(&lli->lli_fid));
1662                 return -EEXIST;
1663         }
1664
1665         llu_prep_md_op_data(&data, NULL, ino, NULL, 0, O_RDWR,
1666                             LUSTRE_OPC_ANY);
1667         rc = md_enqueue(sbi->ll_md_exp, &einfo, &oit, &data,
1668                         &lockh, lum, lum_size, NULL, LDLM_FL_INTENT_ONLY);
1669         if (rc)
1670                 GOTO(out, rc);
1671
1672         req = oit.d.lustre.it_data;
1673         rc = it_open_error(DISP_IT_EXECD, &oit);
1674         if (rc) {
1675                 req->rq_replay = 0;
1676                 GOTO(out, rc);
1677         }
1678
1679         rc = it_open_error(DISP_OPEN_OPEN, &oit);
1680         if (rc) {
1681                 req->rq_replay = 0;
1682                 GOTO(out, rc);
1683         }
1684
1685         rc = md_get_lustre_md(sbi->ll_md_exp, req,
1686                               sbi->ll_dt_exp, sbi->ll_md_exp, &md);
1687         if (rc)
1688                 GOTO(out, rc);
1689
1690         llu_update_inode(ino, &md);
1691         llu_local_open(lli, &oit);
1692         /* release intent */
1693         if (lustre_handle_is_used(&lockh))
1694                 ldlm_lock_decref(&lockh, LCK_CR);
1695         ptlrpc_req_finished(req);
1696         req = NULL;
1697         rc = llu_file_release(ino);
1698         EXIT;
1699
1700 out:
1701         if (req != NULL)
1702                 ptlrpc_req_finished(req);
1703         return rc;
1704 }
1705
1706 static int llu_lov_file_setstripe(struct inode *ino, unsigned long arg)
1707 {
1708         struct lov_user_md lum, *lump = (struct lov_user_md *)arg;
1709         int rc;
1710         int flags = FMODE_WRITE;
1711         ENTRY;
1712
1713         LASSERT(sizeof(lum) == sizeof(*lump));
1714         LASSERT(sizeof(lum.lmm_objects[0]) == sizeof(lump->lmm_objects[0]));
1715         if (cfs_copy_from_user(&lum, lump, sizeof(lum)))
1716                 RETURN(-EFAULT);
1717
1718         rc = llu_lov_setstripe_ea_info(ino, flags, &lum, sizeof(lum));
1719         RETURN(rc);
1720 }
1721
1722 static int llu_lov_setstripe(struct inode *ino, unsigned long arg)
1723 {
1724         struct intnl_stat *st = llu_i2stat(ino);
1725         if (S_ISREG(st->st_mode))
1726                 return llu_lov_file_setstripe(ino, arg);
1727         if (S_ISDIR(st->st_mode))
1728                 return llu_lov_dir_setstripe(ino, arg);
1729
1730         return -EINVAL;
1731 }
1732
1733 static int llu_lov_getstripe(struct inode *ino, unsigned long arg)
1734 {
1735         struct lov_stripe_md *lsm = llu_i2info(ino)->lli_smd;
1736
1737         if (!lsm)
1738                 RETURN(-ENODATA);
1739
1740         return obd_iocontrol(LL_IOC_LOV_GETSTRIPE, llu_i2obdexp(ino), 0, lsm,
1741                             (void *)arg);
1742 }
1743
1744 static int llu_iop_ioctl(struct inode *ino, unsigned long int request,
1745                          va_list ap)
1746 {
1747         unsigned long arg;
1748         int rc;
1749
1750         liblustre_wait_event(0);
1751
1752         switch (request) {
1753         case LL_IOC_GROUP_LOCK:
1754                 arg = va_arg(ap, unsigned long);
1755                 rc = llu_get_grouplock(ino, arg);
1756                 break;
1757         case LL_IOC_GROUP_UNLOCK:
1758                 arg = va_arg(ap, unsigned long);
1759                 rc = llu_put_grouplock(ino, arg);
1760                 break;
1761         case LL_IOC_LOV_SETSTRIPE:
1762                 arg = va_arg(ap, unsigned long);
1763                 rc = llu_lov_setstripe(ino, arg);
1764                 break;
1765         case LL_IOC_LOV_GETSTRIPE:
1766                 arg = va_arg(ap, unsigned long);
1767                 rc = llu_lov_getstripe(ino, arg);
1768                 break;
1769         default:
1770                 CERROR("did not support ioctl cmd %lx\n", request);
1771                 rc = -ENOSYS;
1772                 break;
1773         }
1774
1775         liblustre_wait_event(0);
1776         return rc;
1777 }
1778
1779 /*
1780  * we already do syncronous read/write
1781  */
1782 static int llu_iop_sync(struct inode *inode)
1783 {
1784         liblustre_wait_event(0);
1785         return 0;
1786 }
1787
1788 static int llu_iop_datasync(struct inode *inode)
1789 {
1790         liblustre_wait_event(0);
1791         return 0;
1792 }
1793
1794 struct filesys_ops llu_filesys_ops =
1795 {
1796         fsop_gone: llu_fsop_gone,
1797 };
1798
1799 struct inode *llu_iget(struct filesys *fs, struct lustre_md *md)
1800 {
1801         struct inode *inode;
1802         struct lu_fid fid;
1803         struct file_identifier fileid = {&fid, sizeof(fid)};
1804
1805         if ((md->body->valid & (OBD_MD_FLID | OBD_MD_FLTYPE)) !=
1806             (OBD_MD_FLID | OBD_MD_FLTYPE)) {
1807                 CERROR("bad md body valid mask "LPX64"\n", md->body->valid);
1808                 LBUG();
1809                 return ERR_PTR(-EPERM);
1810         }
1811
1812         /* try to find existing inode */
1813         fid = md->body->fid1;
1814
1815         inode = _sysio_i_find(fs, &fileid);
1816         if (inode) {
1817                 if (inode->i_zombie/* ||
1818                     lli->lli_st_generation != md->body->generation*/) {
1819                         I_RELE(inode);
1820                 }
1821                 else {
1822                         llu_update_inode(inode, md);
1823                         return inode;
1824                 }
1825         }
1826
1827         inode = llu_new_inode(fs, &fid);
1828         if (inode)
1829                 llu_update_inode(inode, md);
1830
1831         return inode;
1832 }
1833
1834 static int
1835 llu_fsswop_mount(const char *source,
1836                  unsigned flags,
1837                  const void *data __IS_UNUSED,
1838                  struct pnode *tocover,
1839                  struct mount **mntp)
1840 {
1841         struct filesys *fs;
1842         struct inode *root;
1843         struct pnode_base *rootpb;
1844         struct obd_device *obd;
1845         struct llu_sb_info *sbi;
1846         struct obd_statfs osfs;
1847         static struct qstr noname = { NULL, 0, 0 };
1848         struct ptlrpc_request *request = NULL;
1849         struct lustre_md md;
1850         class_uuid_t uuid;
1851         struct config_llog_instance cfg = {0, };
1852         char ll_instance[sizeof(sbi) * 2 + 3];
1853         struct lustre_profile *lprof;
1854         char *zconf_mgsnid, *zconf_profile;
1855         char *osc = NULL, *mdc = NULL;
1856         int async = 1, err = -EINVAL;
1857         struct obd_connect_data ocd = {0,};
1858         struct md_op_data op_data = {{0}};
1859
1860         ENTRY;
1861
1862         if (ll_parse_mount_target(source,
1863                                   &zconf_mgsnid,
1864                                   &zconf_profile)) {
1865                 CERROR("mal-formed target %s\n", source);
1866                 RETURN(err);
1867         }
1868         if (!zconf_mgsnid || !zconf_profile) {
1869                 printf("Liblustre: invalid target %s\n", source);
1870                 RETURN(err);
1871         }
1872         /* allocate & initialize sbi */
1873         OBD_ALLOC(sbi, sizeof(*sbi));
1874         if (!sbi)
1875                 RETURN(-ENOMEM);
1876
1877         CFS_INIT_LIST_HEAD(&sbi->ll_conn_chain);
1878         ll_generate_random_uuid(uuid);
1879         class_uuid_unparse(uuid, &sbi->ll_sb_uuid);
1880
1881         /* generate a string unique to this super, let's try
1882          the address of the super itself.*/
1883         snprintf(ll_instance, sizeof(ll_instance), "%p", sbi);
1884
1885         /* retrive & parse config log */
1886         cfg.cfg_instance = ll_instance;
1887         cfg.cfg_uuid = sbi->ll_sb_uuid;
1888         err = liblustre_process_log(&cfg, zconf_mgsnid, zconf_profile, 1);
1889         if (err < 0) {
1890                 CERROR("Unable to process log: %s\n", zconf_profile);
1891                 GOTO(out_free, err);
1892         }
1893
1894         lprof = class_get_profile(zconf_profile);
1895         if (lprof == NULL) {
1896                 CERROR("No profile found: %s\n", zconf_profile);
1897                 GOTO(out_free, err = -EINVAL);
1898         }
1899         OBD_ALLOC(osc, strlen(lprof->lp_dt) + strlen(ll_instance) + 2);
1900         sprintf(osc, "%s-%s", lprof->lp_dt, ll_instance);
1901
1902         OBD_ALLOC(mdc, strlen(lprof->lp_md) + strlen(ll_instance) + 2);
1903         sprintf(mdc, "%s-%s", lprof->lp_md, ll_instance);
1904
1905         if (!osc) {
1906                 CERROR("no osc\n");
1907                 GOTO(out_free, err = -EINVAL);
1908         }
1909         if (!mdc) {
1910                 CERROR("no mdc\n");
1911                 GOTO(out_free, err = -EINVAL);
1912         }
1913
1914         fs = _sysio_fs_new(&llu_filesys_ops, flags, sbi);
1915         if (!fs) {
1916                 err = -ENOMEM;
1917                 goto out_free;
1918         }
1919
1920         obd = class_name2obd(mdc);
1921         if (!obd) {
1922                 CERROR("MDC %s: not setup or attached\n", mdc);
1923                 GOTO(out_free, err = -EINVAL);
1924         }
1925         obd_set_info_async(obd->obd_self_export, sizeof(KEY_ASYNC), KEY_ASYNC,
1926                            sizeof(async), &async, NULL);
1927
1928         ocd.ocd_connect_flags = OBD_CONNECT_IBITS | OBD_CONNECT_VERSION |
1929                                 OBD_CONNECT_FID | OBD_CONNECT_AT |
1930                                 OBD_CONNECT_VBR | OBD_CONNECT_FULL20;
1931
1932 #ifdef LIBLUSTRE_POSIX_ACL
1933         ocd.ocd_connect_flags |= OBD_CONNECT_ACL;
1934 #endif
1935         ocd.ocd_ibits_known = MDS_INODELOCK_FULL;
1936         ocd.ocd_version = LUSTRE_VERSION_CODE;
1937
1938         /* setup mdc */
1939         err = obd_connect(NULL, &sbi->ll_md_exp, obd, &sbi->ll_sb_uuid, &ocd, NULL);
1940         if (err) {
1941                 CERROR("cannot connect to %s: rc = %d\n", mdc, err);
1942                 GOTO(out_free, err);
1943         }
1944
1945         err = obd_statfs(obd, &osfs, 100000000, 0);
1946         if (err)
1947                 GOTO(out_md, err);
1948
1949         /*
1950          * FIXME fill fs stat data into sbi here!!! FIXME
1951          */
1952
1953         /* setup osc */
1954         obd = class_name2obd(osc);
1955         if (!obd) {
1956                 CERROR("OSC %s: not setup or attached\n", osc);
1957                 GOTO(out_md, err = -EINVAL);
1958         }
1959         obd_set_info_async(obd->obd_self_export, sizeof(KEY_ASYNC), KEY_ASYNC,
1960                            sizeof(async), &async, NULL);
1961
1962         obd->obd_upcall.onu_owner = &sbi->ll_lco;
1963         obd->obd_upcall.onu_upcall = cl_ocd_update;
1964
1965         ocd.ocd_connect_flags = OBD_CONNECT_SRVLOCK | OBD_CONNECT_REQPORTAL |
1966                                 OBD_CONNECT_VERSION | OBD_CONNECT_TRUNCLOCK |
1967                                 OBD_CONNECT_FID | OBD_CONNECT_AT |
1968                                 OBD_CONNECT_FULL20;
1969
1970         ocd.ocd_version = LUSTRE_VERSION_CODE;
1971         err = obd_connect(NULL, &sbi->ll_dt_exp, obd, &sbi->ll_sb_uuid, &ocd, NULL);
1972         if (err) {
1973                 CERROR("cannot connect to %s: rc = %d\n", osc, err);
1974                 GOTO(out_md, err);
1975         }
1976         sbi->ll_lco.lco_flags = ocd.ocd_connect_flags;
1977         sbi->ll_lco.lco_md_exp = sbi->ll_md_exp;
1978         sbi->ll_lco.lco_dt_exp = sbi->ll_dt_exp;
1979
1980         fid_zero(&sbi->ll_root_fid);
1981         err = md_getstatus(sbi->ll_md_exp, &sbi->ll_root_fid, NULL);
1982         if (err) {
1983                 CERROR("cannot mds_connect: rc = %d\n", err);
1984                 GOTO(out_lock_cn_cb, err);
1985         }
1986         if (!fid_is_sane(&sbi->ll_root_fid)) {
1987                 CERROR("Invalid root fid during mount\n");
1988                 GOTO(out_lock_cn_cb, err = -EINVAL);
1989         }
1990         CDEBUG(D_SUPER, "rootfid "DFID"\n", PFID(&sbi->ll_root_fid));
1991
1992         op_data.op_fid1 = sbi->ll_root_fid;
1993         op_data.op_valid = OBD_MD_FLGETATTR | OBD_MD_FLBLOCKS;
1994         /* fetch attr of root inode */
1995         err = md_getattr(sbi->ll_md_exp, &op_data, &request);
1996         if (err) {
1997                 CERROR("md_getattr failed for root: rc = %d\n", err);
1998                 GOTO(out_lock_cn_cb, err);
1999         }
2000
2001         err = md_get_lustre_md(sbi->ll_md_exp, request,
2002                                sbi->ll_dt_exp, sbi->ll_md_exp, &md);
2003         if (err) {
2004                 CERROR("failed to understand root inode md: rc = %d\n",err);
2005                 GOTO(out_request, err);
2006         }
2007
2008         LASSERT(fid_is_sane(&sbi->ll_root_fid));
2009
2010         root = llu_iget(fs, &md);
2011         if (!root || IS_ERR(root)) {
2012                 CERROR("fail to generate root inode\n");
2013                 GOTO(out_request, err = -EBADF);
2014         }
2015
2016         /*
2017          * Generate base path-node for root.
2018          */
2019         rootpb = _sysio_pb_new(&noname, NULL, root);
2020         if (!rootpb) {
2021                 err = -ENOMEM;
2022                 goto out_inode;
2023         }
2024
2025         err = _sysio_do_mount(fs, rootpb, flags, tocover, mntp);
2026         if (err) {
2027                 _sysio_pb_gone(rootpb);
2028                 goto out_inode;
2029         }
2030
2031         cl_sb_init(sbi);
2032
2033         ptlrpc_req_finished(request);
2034
2035         CDEBUG(D_SUPER, "LibLustre: %s mounted successfully!\n", source);
2036         liblustre_wait_idle();
2037
2038         return 0;
2039
2040 out_inode:
2041         _sysio_i_gone(root);
2042 out_request:
2043         ptlrpc_req_finished(request);
2044 out_lock_cn_cb:
2045         obd_disconnect(sbi->ll_dt_exp);
2046 out_md:
2047         obd_disconnect(sbi->ll_md_exp);
2048 out_free:
2049         if (osc)
2050                 OBD_FREE(osc, strlen(osc) + 1);
2051         if (mdc)
2052                 OBD_FREE(mdc, strlen(mdc) + 1);
2053         OBD_FREE(sbi, sizeof(*sbi));
2054         liblustre_wait_idle();
2055         return err;
2056 }
2057
2058 struct fssw_ops llu_fssw_ops = {
2059         llu_fsswop_mount
2060 };
2061
2062 static struct inode_ops llu_inode_ops = {
2063         inop_lookup:    llu_iop_lookup,
2064         inop_getattr:   llu_iop_getattr,
2065         inop_setattr:   llu_iop_setattr,
2066         inop_filldirentries:     llu_iop_filldirentries,
2067         inop_mkdir:     llu_iop_mkdir_raw,
2068         inop_rmdir:     llu_iop_rmdir_raw,
2069         inop_symlink:   llu_iop_symlink_raw,
2070         inop_readlink:  llu_iop_readlink,
2071         inop_open:      llu_iop_open,
2072         inop_close:     llu_iop_close,
2073         inop_link:      llu_iop_link_raw,
2074         inop_unlink:    llu_iop_unlink_raw,
2075         inop_rename:    llu_iop_rename_raw,
2076         inop_pos:       llu_iop_pos,
2077         inop_read:      llu_iop_read,
2078         inop_write:     llu_iop_write,
2079         inop_iodone:    llu_iop_iodone,
2080         inop_fcntl:     llu_iop_fcntl,
2081         inop_sync:      llu_iop_sync,
2082         inop_datasync:  llu_iop_datasync,
2083         inop_ioctl:     llu_iop_ioctl,
2084         inop_mknod:     llu_iop_mknod_raw,
2085 #ifdef _HAVE_STATVFS
2086         inop_statvfs:   llu_iop_statvfs,
2087 #endif
2088         inop_gone:      llu_iop_gone,
2089 };