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  * Copyright (c) 2011 Whamcloud, Inc.
  33  *
  34  */
  35 /*
  36  * This file is part of Lustre, http://www.lustre.org/
  37  * Lustre is a trademark of Sun Microsystems, Inc.
  38  *
  39  * lustre/liblustre/super.c
  40  *
  41  * Lustre Light Super operations
  42  */
  43
  44 #define DEBUG_SUBSYSTEM S_LLITE
  45
  46 #include <stdlib.h>
  47 #include <string.h>
  48 #include <assert.h>
  49 #include <time.h>
  50 #include <sys/types.h>
  51 #include <sys/stat.h>
  52 #include <fcntl.h>
  53 #include <sys/queue.h>
  54 #ifndef __CYGWIN__
  55 # include <sys/statvfs.h>
  56 #else
  57 # include <sys/statfs.h>
  58 #endif
  59
  60 #include "llite_lib.h"
  61
  62 #ifndef MAY_EXEC
  63 #define MAY_EXEC        1
  64 #define MAY_WRITE       2
  65 #define MAY_READ        4
  66 #endif
  67
  68 #define S_IXUGO (S_IXUSR|S_IXGRP|S_IXOTH)
  69
  70 static int ll_permission(struct inode *inode, int mask)
  71 {
  72         struct intnl_stat *st = llu_i2stat(inode);
  73         mode_t mode = st->st_mode;
  74
  75         if (current->fsuid == st->st_uid)
  76                 mode >>= 6;
  77         else if (cfs_curproc_is_in_groups(st->st_gid))
  78                 mode >>= 3;
  79
  80         if ((mode & mask & (MAY_READ|MAY_WRITE|MAY_EXEC)) == mask)
  81                 return 0;
  82
  83         if ((mask & (MAY_READ|MAY_WRITE)) ||
  84             (st->st_mode & S_IXUGO))
  85                 if (cfs_capable(CFS_CAP_DAC_OVERRIDE))
  86                         return 0;
  87
  88         if (mask == MAY_READ ||
  89             (S_ISDIR(st->st_mode) && !(mask & MAY_WRITE))) {
  90                 if (cfs_capable(CFS_CAP_DAC_READ_SEARCH))
  91                         return 0;
  92         }
  93
  94         return -EACCES;
  95 }
  96
  97 static void llu_fsop_gone(struct filesys *fs)
  98 {
  99         struct llu_sb_info *sbi = (struct llu_sb_info *) fs->fs_private;
 100         struct obd_device *obd = class_exp2obd(sbi->ll_md_exp);
 101         int next = 0;
 102         ENTRY;
 103
 104         cfs_list_del(&sbi->ll_conn_chain);
 105         cl_sb_fini(sbi);
 106         obd_disconnect(sbi->ll_dt_exp);
 107         obd_disconnect(sbi->ll_md_exp);
 108
 109         while ((obd = class_devices_in_group(&sbi->ll_sb_uuid, &next)) != NULL)
 110                 class_manual_cleanup(obd);
 111
 112         OBD_FREE(sbi, sizeof(*sbi));
 113
 114         liblustre_wait_idle();
 115         EXIT;
 116 }
 117
 118 static struct inode_ops llu_inode_ops;
 119
 120 static ldlm_mode_t llu_take_md_lock(struct inode *inode, __u64 bits,
 121                                     struct lustre_handle *lockh)
 122 {
 123         ldlm_policy_data_t policy = { .l_inodebits = {bits}};
 124         struct lu_fid *fid;
 125         ldlm_mode_t rc;
 126         int flags;
 127         ENTRY;
 128
 129         fid = &llu_i2info(inode)->lli_fid;
 130         CDEBUG(D_INFO, "trying to match res "DFID"\n", PFID(fid));
 131
 132         flags = LDLM_FL_BLOCK_GRANTED | LDLM_FL_CBPENDING;
 133         rc = md_lock_match(llu_i2mdexp(inode), flags, fid, LDLM_IBITS, &policy,
 134                            LCK_CR|LCK_CW|LCK_PR|LCK_PW, lockh);
 135         RETURN(rc);
 136 }
 137
 138 void llu_update_inode(struct inode *inode, struct lustre_md *md)
 139 {
 140         struct llu_inode_info *lli = llu_i2info(inode);
 141         struct mdt_body *body = md->body;
 142         struct lov_stripe_md *lsm = md->lsm;
 143         struct intnl_stat *st = llu_i2stat(inode);
 144
 145         LASSERT ((lsm != NULL) == ((body->valid & OBD_MD_FLEASIZE) != 0));
 146
 147         if (body->valid & OBD_MD_FLMODE)
 148                 st->st_mode = (st->st_mode & S_IFMT)|(body->mode & ~S_IFMT);
 149         if (body->valid & OBD_MD_FLTYPE)
 150                 st->st_mode = (st->st_mode & ~S_IFMT)|(body->mode & S_IFMT);
 151
 152         if (lsm != NULL) {
 153                 if (lli->lli_smd == NULL) {
 154                         cl_inode_init(inode, md);
 155                         lli->lli_smd = lsm;
 156                         lli->lli_maxbytes = lsm->lsm_maxbytes;
 157                         if (lli->lli_maxbytes > PAGE_CACHE_MAXBYTES)
 158                                 lli->lli_maxbytes = PAGE_CACHE_MAXBYTES;
 159                 } else {
 160                         if (lov_stripe_md_cmp(lli->lli_smd, lsm)) {
 161                                 CERROR("lsm mismatch for inode %lld\n",
 162                                        (long long)st->st_ino);
 163                                 LBUG();
 164                         }
 165                 }
 166         }
 167
 168         if (body->valid & OBD_MD_FLATIME) {
 169                 if (body->atime > LTIME_S(st->st_atime))
 170                         LTIME_S(st->st_atime) = body->atime;
 171                 lli->lli_lvb.lvb_atime = body->atime;
 172         }
 173         if (body->valid & OBD_MD_FLMTIME) {
 174                 if (body->mtime > LTIME_S(st->st_mtime))
 175                         LTIME_S(st->st_mtime) = body->mtime;
 176                 lli->lli_lvb.lvb_mtime = body->mtime;
 177         }
 178         if (body->valid & OBD_MD_FLCTIME) {
 179                 if (body->ctime > LTIME_S(st->st_ctime))
 180                         LTIME_S(st->st_ctime) = body->ctime;
 181                 lli->lli_lvb.lvb_ctime = body->ctime;
 182         }
 183         if (S_ISREG(st->st_mode))
 184                 st->st_blksize = min(2UL * PTLRPC_MAX_BRW_SIZE, LL_MAX_BLKSIZE);
 185         else
 186                 st->st_blksize = 4096;
 187         if (body->valid & OBD_MD_FLUID)
 188                 st->st_uid = body->uid;
 189         if (body->valid & OBD_MD_FLGID)
 190                 st->st_gid = body->gid;
 191         if (body->valid & OBD_MD_FLNLINK)
 192                 st->st_nlink = body->nlink;
 193         if (body->valid & OBD_MD_FLRDEV)
 194                 st->st_rdev = body->rdev;
 195         if (body->valid & OBD_MD_FLFLAGS)
 196                 lli->lli_st_flags = body->flags;
 197         if (body->valid & OBD_MD_FLSIZE) {
 198                 if ((llu_i2sbi(inode)->ll_lco.lco_flags & OBD_CONNECT_SOM) &&
 199                     S_ISREG(st->st_mode) && lli->lli_smd) {
 200                         struct lustre_handle lockh;
 201                         ldlm_mode_t mode;
 202
 203                         /* As it is possible a blocking ast has been processed
 204                          * by this time, we need to check there is an UPDATE
 205                          * lock on the client and set LLIF_MDS_SIZE_LOCK holding
 206                          * it. */
 207                         mode = llu_take_md_lock(inode, MDS_INODELOCK_UPDATE,
 208                                                 &lockh);
 209                         if (mode) {
 210                                 st->st_size = body->size;
 211                                 lli->lli_flags |= LLIF_MDS_SIZE_LOCK;
 212                                 ldlm_lock_decref(&lockh, mode);
 213                         }
 214                 } else {
 215                     st->st_size = body->size;
 216                 }
 217
 218                 if (body->valid & OBD_MD_FLBLOCKS)
 219                         st->st_blocks = body->blocks;
 220         }
 221 }
 222
 223 void obdo_to_inode(struct inode *dst, struct obdo *src, obd_flag valid)
 224 {
 225         struct llu_inode_info *lli = llu_i2info(dst);
 226         struct intnl_stat *st = llu_i2stat(dst);
 227
 228         valid &= src->o_valid;
 229
 230         LASSERTF(!(valid & (OBD_MD_FLTYPE | OBD_MD_FLGENER | OBD_MD_FLFID |
 231                             OBD_MD_FLID | OBD_MD_FLGROUP)),
 232                  "object "LPU64"/"LPU64", valid %x\n",
 233                  src->o_id, src->o_seq, valid);
 234
 235         if (valid & (OBD_MD_FLCTIME | OBD_MD_FLMTIME))
 236                 CDEBUG(D_INODE,"valid "LPX64", cur time "CFS_TIME_T"/"CFS_TIME_T
 237                        ", new %lu/%lu\n",
 238                        src->o_valid,
 239                        LTIME_S(st->st_mtime), LTIME_S(st->st_ctime),
 240                        (long)src->o_mtime, (long)src->o_ctime);
 241
 242         if (valid & OBD_MD_FLATIME)
 243                 LTIME_S(st->st_atime) = src->o_atime;
 244         if (valid & OBD_MD_FLMTIME)
 245                 LTIME_S(st->st_mtime) = src->o_mtime;
 246         if (valid & OBD_MD_FLCTIME && src->o_ctime > LTIME_S(st->st_ctime))
 247                 LTIME_S(st->st_ctime) = src->o_ctime;
 248         if (valid & OBD_MD_FLSIZE)
 249                 st->st_size = src->o_size;
 250         if (valid & OBD_MD_FLBLOCKS) /* allocation of space */
 251                 st->st_blocks = src->o_blocks;
 252         if (valid & OBD_MD_FLBLKSZ)
 253                 st->st_blksize = src->o_blksize;
 254         if (valid & OBD_MD_FLTYPE)
 255                 st->st_mode = (st->st_mode & ~S_IFMT) | (src->o_mode & S_IFMT);
 256         if (valid & OBD_MD_FLMODE)
 257                 st->st_mode = (st->st_mode & S_IFMT) | (src->o_mode & ~S_IFMT);
 258         if (valid & OBD_MD_FLUID)
 259                 st->st_uid = src->o_uid;
 260         if (valid & OBD_MD_FLGID)
 261                 st->st_gid = src->o_gid;
 262         if (valid & OBD_MD_FLFLAGS)
 263                 lli->lli_st_flags = src->o_flags;
 264 }
 265
 266 /**
 267  * Performs the getattr on the inode and updates its fields.
 268  * If @sync != 0, perform the getattr under the server-side lock.
 269  */
 270 int llu_inode_getattr(struct inode *inode, struct obdo *obdo,
 271                       __u64 ioepoch, int sync)
 272 {
 273         struct llu_inode_info *lli = llu_i2info(inode);
 274         struct ptlrpc_request_set *set;
 275         struct lov_stripe_md *lsm = lli->lli_smd;
 276         struct obd_info oinfo = { { { 0 } } };
 277         int rc;
 278         ENTRY;
 279
 280         LASSERT(lsm);
 281
 282         oinfo.oi_md = lsm;
 283         oinfo.oi_oa = obdo;
 284         oinfo.oi_oa->o_id = lsm->lsm_object_id;
 285         oinfo.oi_oa->o_seq = lsm->lsm_object_seq;
 286         oinfo.oi_oa->o_mode = S_IFREG;
 287         oinfo.oi_oa->o_ioepoch = ioepoch;
 288         oinfo.oi_oa->o_valid = OBD_MD_FLID | OBD_MD_FLTYPE |
 289                                OBD_MD_FLSIZE | OBD_MD_FLBLOCKS |
 290                                OBD_MD_FLBLKSZ | OBD_MD_FLMTIME |
 291                                OBD_MD_FLCTIME | OBD_MD_FLGROUP |
 292                                OBD_MD_FLATIME | OBD_MD_FLEPOCH;
 293         obdo_from_inode(oinfo.oi_oa, NULL, &llu_i2info(inode)->lli_fid, 0);
 294         if (sync) {
 295                 oinfo.oi_oa->o_valid |= OBD_MD_FLFLAGS;
 296                 oinfo.oi_oa->o_flags |= OBD_FL_SRVLOCK;
 297         }
 298
 299         set = ptlrpc_prep_set();
 300         if (set == NULL) {
 301                 CERROR ("ENOMEM allocing request set\n");
 302                 rc = -ENOMEM;
 303         } else {
 304                 rc = obd_getattr_async(llu_i2obdexp(inode), &oinfo, set);
 305                 if (rc == 0)
 306                         rc = ptlrpc_set_wait(set);
 307                 ptlrpc_set_destroy(set);
 308         }
 309         if (rc)
 310                 RETURN(rc);
 311
 312         oinfo.oi_oa->o_valid = OBD_MD_FLBLOCKS | OBD_MD_FLBLKSZ |
 313                                OBD_MD_FLMTIME | OBD_MD_FLCTIME |
 314                                OBD_MD_FLSIZE;
 315
 316         obdo_refresh_inode(inode, oinfo.oi_oa, oinfo.oi_oa->o_valid);
 317         CDEBUG(D_INODE, "objid "LPX64" size %llu, blocks %llu, "
 318                "blksize %llu\n", lli->lli_smd->lsm_object_id,
 319                (long long unsigned)llu_i2stat(inode)->st_size,
 320                (long long unsigned)llu_i2stat(inode)->st_blocks,
 321                (long long unsigned)llu_i2stat(inode)->st_blksize);
 322         RETURN(0);
 323 }
 324
 325 static struct inode* llu_new_inode(struct filesys *fs,
 326                                    struct lu_fid *fid)
 327 {
 328         struct inode *inode;
 329         struct llu_inode_info *lli;
 330         struct intnl_stat st = {
 331                 .st_dev  = 0,
 332 #if 0
 333 #ifndef AUTOMOUNT_FILE_NAME
 334                 .st_mode = fid->f_type & S_IFMT,
 335 #else
 336                 .st_mode = fid->f_type /* all of the bits! */
 337 #endif
 338 #endif
 339                 /* FIXME: fix this later */
 340                 .st_mode = 0,
 341
 342                 .st_uid  = geteuid(),
 343                 .st_gid  = getegid(),
 344         };
 345
 346         OBD_ALLOC(lli, sizeof(*lli));
 347         if (!lli)
 348                 return NULL;
 349
 350         /* initialize lli here */
 351         lli->lli_sbi = llu_fs2sbi(fs);
 352         lli->lli_smd = NULL;
 353         lli->lli_symlink_name = NULL;
 354         lli->lli_flags = 0;
 355         lli->lli_maxbytes = (__u64)(~0UL);
 356         lli->lli_file_data = NULL;
 357
 358         lli->lli_sysio_fid.fid_data = &lli->lli_fid;
 359         lli->lli_sysio_fid.fid_len = sizeof(lli->lli_fid);
 360         lli->lli_fid = *fid;
 361
 362         /* file identifier is needed by functions like _sysio_i_find() */
 363         inode = _sysio_i_new(fs, &lli->lli_sysio_fid,
 364                              &st, 0, &llu_inode_ops, lli);
 365
 366         if (!inode)
 367                 OBD_FREE(lli, sizeof(*lli));
 368
 369         return inode;
 370 }
 371
 372 static int llu_have_md_lock(struct inode *inode, __u64 lockpart)
 373 {
 374         struct lustre_handle lockh;
 375         ldlm_policy_data_t policy = { .l_inodebits = { lockpart } };
 376         struct lu_fid *fid;
 377         int flags;
 378         ENTRY;
 379
 380         LASSERT(inode);
 381
 382         fid = &llu_i2info(inode)->lli_fid;
 383         CDEBUG(D_INFO, "trying to match res "DFID"\n", PFID(fid));
 384
 385         flags = LDLM_FL_BLOCK_GRANTED | LDLM_FL_CBPENDING | LDLM_FL_TEST_LOCK;
 386         if (md_lock_match(llu_i2mdexp(inode), flags, fid, LDLM_IBITS, &policy,
 387                           LCK_CR|LCK_CW|LCK_PR|LCK_PW, &lockh)) {
 388                 RETURN(1);
 389         }
 390         RETURN(0);
 391 }
 392
 393 static int llu_inode_revalidate(struct inode *inode)
 394 {
 395         struct llu_inode_info *lli = llu_i2info(inode);
 396         struct intnl_stat *st = llu_i2stat(inode);
 397         ENTRY;
 398
 399         if (!inode) {
 400                 CERROR("REPORT THIS LINE TO PETER\n");
 401                 RETURN(0);
 402         }
 403
 404         if (!llu_have_md_lock(inode, MDS_INODELOCK_UPDATE)) {
 405                 struct lustre_md md;
 406                 struct ptlrpc_request *req = NULL;
 407                 struct llu_sb_info *sbi = llu_i2sbi(inode);
 408                 struct md_op_data op_data = { { 0 } };
 409                 unsigned long valid = OBD_MD_FLGETATTR;
 410                 int rc, ealen = 0;
 411
 412                 /* Why don't we update all valid MDS fields here, if we're
 413                  * doing an RPC anyways?  -phil */
 414                 if (S_ISREG(st->st_mode)) {
 415                         ealen = obd_size_diskmd(sbi->ll_dt_exp, NULL);
 416                         valid |= OBD_MD_FLEASIZE;
 417                 }
 418
 419                 llu_prep_md_op_data(&op_data, inode, NULL, NULL, 0, ealen,
 420                                     LUSTRE_OPC_ANY);
 421                 op_data.op_valid = valid;
 422
 423                 rc = md_getattr(sbi->ll_md_exp, &op_data, &req);
 424                 if (rc) {
 425                         CERROR("failure %d inode %llu\n", rc,
 426                                (long long)st->st_ino);
 427                         RETURN(-abs(rc));
 428                 }
 429                 rc = md_get_lustre_md(sbi->ll_md_exp, req,
 430                                       sbi->ll_dt_exp, sbi->ll_md_exp, &md);
 431
 432                 /* XXX Too paranoid? */
 433                 if (((md.body->valid ^ valid) & OBD_MD_FLEASIZE) &&
 434                     !((md.body->valid & OBD_MD_FLNLINK) &&
 435                       (md.body->nlink == 0))) {
 436                         CERROR("Asked for %s eadata but got %s (%d)\n",
 437                                (valid & OBD_MD_FLEASIZE) ? "some" : "no",
 438                                (md.body->valid & OBD_MD_FLEASIZE) ? "some":"none",
 439                                 md.body->eadatasize);
 440                 }
 441                 if (rc) {
 442                         ptlrpc_req_finished(req);
 443                         RETURN(rc);
 444                 }
 445
 446
 447                 llu_update_inode(inode, &md);
 448                 if (md.lsm != NULL && lli->lli_smd != md.lsm)
 449                         obd_free_memmd(sbi->ll_dt_exp, &md.lsm);
 450                 ptlrpc_req_finished(req);
 451         }
 452
 453         if (!lli->lli_smd) {
 454                 /* object not yet allocated, don't validate size */
 455                 st->st_atime = lli->lli_lvb.lvb_atime;
 456                 st->st_mtime = lli->lli_lvb.lvb_mtime;
 457                 st->st_ctime = lli->lli_lvb.lvb_ctime;
 458                 RETURN(0);
 459         }
 460
 461         /* ll_glimpse_size will prefer locally cached writes if they extend
 462          * the file */
 463         RETURN(cl_glimpse_size(inode));
 464 }
 465
 466 static void copy_stat_buf(struct inode *ino, struct intnl_stat *b)
 467 {
 468         *b = *llu_i2stat(ino);
 469 }
 470
 471 static int llu_iop_getattr(struct pnode *pno,
 472                            struct inode *ino,
 473                            struct intnl_stat *b)
 474 {
 475         int rc;
 476         ENTRY;
 477
 478         liblustre_wait_event(0);
 479
 480         if (!ino) {
 481                 LASSERT(pno);
 482                 LASSERT(pno->p_base->pb_ino);
 483                 ino = pno->p_base->pb_ino;
 484         } else {
 485                 LASSERT(!pno || pno->p_base->pb_ino == ino);
 486         }
 487
 488         /* libsysio might call us directly without intent lock,
 489          * we must re-fetch the attrs here
 490          */
 491         rc = llu_inode_revalidate(ino);
 492         if (!rc) {
 493                 copy_stat_buf(ino, b);
 494                 LASSERT(!llu_i2info(ino)->lli_it);
 495         }
 496
 497         liblustre_wait_event(0);
 498         RETURN(rc);
 499 }
 500
 501 static int null_if_equal(struct ldlm_lock *lock, void *data)
 502 {
 503         if (data == lock->l_ast_data) {
 504                 lock->l_ast_data = NULL;
 505
 506                 if (lock->l_req_mode != lock->l_granted_mode)
 507                         LDLM_ERROR(lock,"clearing inode with ungranted lock\n");
 508         }
 509
 510         return LDLM_ITER_CONTINUE;
 511 }
 512
 513 void llu_clear_inode(struct inode *inode)
 514 {
 515         struct llu_inode_info *lli = llu_i2info(inode);
 516         struct llu_sb_info *sbi = llu_i2sbi(inode);
 517         ENTRY;
 518
 519         CDEBUG(D_VFSTRACE, "VFS Op:inode=%llu/%lu(%p)\n",
 520                (long long)llu_i2stat(inode)->st_ino, lli->lli_st_generation,
 521                inode);
 522
 523         lli->lli_flags &= ~LLIF_MDS_SIZE_LOCK;
 524         md_change_cbdata(sbi->ll_md_exp, ll_inode2fid(inode),
 525                          null_if_equal, inode);
 526
 527         if (lli->lli_smd)
 528                 obd_change_cbdata(sbi->ll_dt_exp, lli->lli_smd,
 529                                   null_if_equal, inode);
 530
 531         cl_inode_fini(inode);
 532
 533         if (lli->lli_smd) {
 534                 obd_free_memmd(sbi->ll_dt_exp, &lli->lli_smd);
 535                 lli->lli_smd = NULL;
 536         }
 537
 538         if (lli->lli_symlink_name) {
 539                 OBD_FREE(lli->lli_symlink_name,
 540                          strlen(lli->lli_symlink_name) + 1);
 541                 lli->lli_symlink_name = NULL;
 542         }
 543
 544         EXIT;
 545 }
 546
 547 void llu_iop_gone(struct inode *inode)
 548 {
 549         struct llu_inode_info *lli = llu_i2info(inode);
 550         ENTRY;
 551
 552         liblustre_wait_event(0);
 553         llu_clear_inode(inode);
 554
 555         OBD_FREE(lli, sizeof(*lli));
 556         EXIT;
 557 }
 558
 559 static int inode_setattr(struct inode * inode, struct iattr * attr)
 560 {
 561         unsigned int ia_valid = attr->ia_valid;
 562         struct intnl_stat *st = llu_i2stat(inode);
 563         int error = 0;
 564
 565         /*
 566          * inode_setattr() is only ever invoked with ATTR_SIZE (by
 567          * llu_setattr_raw()) when file has no bodies. Check this.
 568          */
 569         LASSERT(ergo(ia_valid & ATTR_SIZE, llu_i2info(inode)->lli_smd == NULL));
 570
 571         if (ia_valid & ATTR_SIZE)
 572                 st->st_size = attr->ia_size;
 573         if (ia_valid & ATTR_UID)
 574                 st->st_uid = attr->ia_uid;
 575         if (ia_valid & ATTR_GID)
 576                 st->st_gid = attr->ia_gid;
 577         if (ia_valid & ATTR_ATIME)
 578                 st->st_atime = attr->ia_atime;
 579         if (ia_valid & ATTR_MTIME)
 580                 st->st_mtime = attr->ia_mtime;
 581         if (ia_valid & ATTR_CTIME)
 582                 st->st_ctime = attr->ia_ctime;
 583         if (ia_valid & ATTR_MODE) {
 584                 st->st_mode = attr->ia_mode;
 585                 if (!cfs_curproc_is_in_groups(st->st_gid) &&
 586                     !cfs_capable(CFS_CAP_FSETID))
 587                         st->st_mode &= ~S_ISGID;
 588         }
 589         /* mark_inode_dirty(inode); */
 590         return error;
 591 }
 592
 593 int llu_md_setattr(struct inode *inode, struct md_op_data *op_data,
 594                    struct md_open_data **mod)
 595 {
 596         struct lustre_md md;
 597         struct llu_sb_info *sbi = llu_i2sbi(inode);
 598         struct ptlrpc_request *request = NULL;
 599         int rc;
 600         ENTRY;
 601
 602         llu_prep_md_op_data(op_data, inode, NULL, NULL, 0, 0, LUSTRE_OPC_ANY);
 603         rc = md_setattr(sbi->ll_md_exp, op_data, NULL, 0, NULL,
 604                         0, &request, mod);
 605
 606         if (rc) {
 607                 ptlrpc_req_finished(request);
 608                 if (rc != -EPERM && rc != -EACCES)
 609                         CERROR("md_setattr fails: rc = %d\n", rc);
 610                 RETURN(rc);
 611         }
 612
 613         rc = md_get_lustre_md(sbi->ll_md_exp, request,
 614                               sbi->ll_dt_exp, sbi->ll_md_exp, &md);
 615         if (rc) {
 616                 ptlrpc_req_finished(request);
 617                 RETURN(rc);
 618         }
 619
 620         /* We call inode_setattr to adjust timestamps.
 621          * If there is at least some data in file, we cleared ATTR_SIZE
 622          * above to avoid invoking vmtruncate, otherwise it is important
 623          * to call vmtruncate in inode_setattr to update inode->i_size
 624          * (bug 6196) */
 625         inode_setattr(inode, &op_data->op_attr);
 626         llu_update_inode(inode, &md);
 627         ptlrpc_req_finished(request);
 628
 629         RETURN(rc);
 630 }
 631
 632 /* Close IO epoch and send Size-on-MDS attribute update. */
 633 static int llu_setattr_done_writing(struct inode *inode,
 634                                     struct md_op_data *op_data,
 635                                     struct md_open_data *mod)
 636 {
 637         struct llu_inode_info *lli = llu_i2info(inode);
 638         struct intnl_stat *st = llu_i2stat(inode);
 639         int rc = 0;
 640         ENTRY;
 641
 642         LASSERT(op_data != NULL);
 643         if (!S_ISREG(st->st_mode))
 644                 RETURN(0);
 645
 646         /* XXX: pass och here for the recovery purpose. */
 647         CDEBUG(D_INODE, "Epoch "LPU64" closed on "DFID" for truncate\n",
 648                op_data->op_ioepoch, PFID(&lli->lli_fid));
 649
 650         op_data->op_flags = MF_EPOCH_CLOSE;
 651         llu_done_writing_attr(inode, op_data);
 652         llu_pack_inode2opdata(inode, op_data, NULL);
 653
 654         rc = md_done_writing(llu_i2sbi(inode)->ll_md_exp, op_data, mod);
 655         if (rc == -EAGAIN) {
 656                 /* MDS has instructed us to obtain Size-on-MDS attribute
 657                  * from OSTs and send setattr to back to MDS. */
 658                 rc = llu_som_update(inode, op_data);
 659         } else if (rc) {
 660                 CERROR("inode %llu mdc truncate failed: rc = %d\n",
 661                        (unsigned long long)st->st_ino, rc);
 662         }
 663         RETURN(rc);
 664 }
 665
 666 /* If this inode has objects allocated to it (lsm != NULL), then the OST
 667  * object(s) determine the file size and mtime.  Otherwise, the MDS will
 668  * keep these values until such a time that objects are allocated for it.
 669  * We do the MDS operations first, as it is checking permissions for us.
 670  * We don't to the MDS RPC if there is nothing that we want to store there,
 671  * otherwise there is no harm in updating mtime/atime on the MDS if we are
 672  * going to do an RPC anyways.
 673  *
 674  * If we are doing a truncate, we will send the mtime and ctime updates
 675  * to the OST with the punch RPC, otherwise we do an explicit setattr RPC.
 676  * I don't believe it is possible to get e.g. ATTR_MTIME_SET and ATTR_SIZE
 677  * at the same time.
 678  */
 679 int llu_setattr_raw(struct inode *inode, struct iattr *attr)
 680 {
 681         struct lov_stripe_md *lsm = llu_i2info(inode)->lli_smd;
 682         struct intnl_stat *st = llu_i2stat(inode);
 683         int ia_valid = attr->ia_valid;
 684         struct md_op_data op_data = { { 0 } };
 685         struct md_open_data *mod = NULL;
 686         int rc = 0, rc1 = 0;
 687         ENTRY;
 688
 689         CDEBUG(D_VFSTRACE, "VFS Op:inode=%llu\n", (long long)st->st_ino);
 690
 691         if (ia_valid & ATTR_SIZE) {
 692                 if (attr->ia_size > ll_file_maxbytes(inode)) {
 693                         CDEBUG(D_INODE, "file too large %llu > "LPU64"\n",
 694                                (long long)attr->ia_size,
 695                                ll_file_maxbytes(inode));
 696                         RETURN(-EFBIG);
 697                 }
 698
 699                 attr->ia_valid |= ATTR_MTIME | ATTR_CTIME;
 700         }
 701
 702         /* We mark all of the fields "set" so MDS/OST does not re-set them */
 703         if (attr->ia_valid & ATTR_CTIME) {
 704                 attr->ia_ctime = CFS_CURRENT_TIME;
 705                 attr->ia_valid |= ATTR_CTIME_SET;
 706         }
 707         if (!(ia_valid & ATTR_ATIME_SET) && (attr->ia_valid & ATTR_ATIME)) {
 708                 attr->ia_atime = CFS_CURRENT_TIME;
 709                 attr->ia_valid |= ATTR_ATIME_SET;
 710         }
 711         if (!(ia_valid & ATTR_MTIME_SET) && (attr->ia_valid & ATTR_MTIME)) {
 712                 attr->ia_mtime = CFS_CURRENT_TIME;
 713                 attr->ia_valid |= ATTR_MTIME_SET;
 714         }
 715
 716         if (attr->ia_valid & (ATTR_MTIME | ATTR_CTIME))
 717                 CDEBUG(D_INODE, "setting mtime "CFS_TIME_T", ctime "CFS_TIME_T
 718                        ", now = "CFS_TIME_T"\n",
 719                        LTIME_S(attr->ia_mtime), LTIME_S(attr->ia_ctime),
 720                        LTIME_S(CFS_CURRENT_TIME));
 721
 722         /* NB: ATTR_SIZE will only be set after this point if the size
 723          * resides on the MDS, ie, this file has no objects. */
 724         if (lsm)
 725                 attr->ia_valid &= ~ATTR_SIZE;
 726
 727         /* If only OST attributes being set on objects, don't do MDS RPC.
 728          * In that case, we need to check permissions and update the local
 729          * inode ourselves so we can call obdo_from_inode() always. */
 730         if (ia_valid & (lsm ? ~(ATTR_FROM_OPEN | ATTR_RAW) : ~0)) {
 731                 memcpy(&op_data.op_attr, attr, sizeof(*attr));
 732
 733                 /* Open epoch for truncate. */
 734                 if (exp_connect_som(llu_i2mdexp(inode)) &&
 735                     (ia_valid & ATTR_SIZE))
 736                         op_data.op_flags = MF_EPOCH_OPEN;
 737                 rc = llu_md_setattr(inode, &op_data, &mod);
 738                 if (rc)
 739                         RETURN(rc);
 740
 741                 llu_ioepoch_open(llu_i2info(inode), op_data.op_ioepoch);
 742                 if (!lsm || !S_ISREG(st->st_mode)) {
 743                         CDEBUG(D_INODE, "no lsm: not setting attrs on OST\n");
 744                         GOTO(out, rc);
 745                 }
 746         } else {
 747                 /* The OST doesn't check permissions, but the alternative is
 748                  * a gratuitous RPC to the MDS.  We already rely on the client
 749                  * to do read/write/truncate permission checks, so is mtime OK?
 750                  */
 751                 if (ia_valid & (ATTR_MTIME | ATTR_ATIME)) {
 752                         /* from sys_utime() */
 753                         if (!(ia_valid & (ATTR_MTIME_SET | ATTR_ATIME_SET))) {
 754                                 if (current->fsuid != st->st_uid &&
 755                                     (rc = ll_permission(inode, MAY_WRITE)) != 0)
 756                                         RETURN(rc);
 757                         } else {
 758                                 /* from inode_change_ok() */
 759                                 if (current->fsuid != st->st_uid &&
 760                                     !cfs_capable(CFS_CAP_FOWNER))
 761                                         RETURN(-EPERM);
 762                         }
 763                 }
 764
 765
 766                 /* Won't invoke llu_vmtruncate(), as we already cleared
 767                  * ATTR_SIZE */
 768                 inode_setattr(inode, attr);
 769         }
 770
 771         if (ia_valid & ATTR_SIZE)
 772                 attr->ia_valid |= ATTR_SIZE;
 773         if (ia_valid & (ATTR_SIZE |
 774                         ATTR_ATIME | ATTR_ATIME_SET |
 775                         ATTR_MTIME | ATTR_MTIME_SET))
 776                 /* on truncate and utimes send attributes to osts, setting
 777                  * mtime/atime to past will be performed under PW 0:EOF extent
 778                  * lock (new_size:EOF for truncate)
 779                  * it may seem excessive to send mtime/atime updates to osts
 780                  * when not setting times to past, but it is necessary due to
 781                  * possible time de-synchronization */
 782                 rc = cl_setattr_ost(inode, attr, NULL);
 783         EXIT;
 784 out:
 785         if (op_data.op_ioepoch)
 786                 rc1 = llu_setattr_done_writing(inode, &op_data, mod);
 787         return rc ? rc : rc1;
 788 }
 789
 790 /* here we simply act as a thin layer to glue it with
 791  * llu_setattr_raw(), which is copy from kernel
 792  */
 793 static int llu_iop_setattr(struct pnode *pno,
 794                            struct inode *ino,
 795                            unsigned mask,
 796                            struct intnl_stat *stbuf)
 797 {
 798         struct iattr iattr;
 799         int rc;
 800         ENTRY;
 801
 802         liblustre_wait_event(0);
 803
 804         LASSERT(!(mask & ~(SETATTR_MTIME | SETATTR_ATIME |
 805                            SETATTR_UID | SETATTR_GID |
 806                            SETATTR_LEN | SETATTR_MODE)));
 807         memset(&iattr, 0, sizeof(iattr));
 808
 809         if (mask & SETATTR_MODE) {
 810                 iattr.ia_mode = stbuf->st_mode;
 811                 iattr.ia_valid |= ATTR_MODE;
 812         }
 813         if (mask & SETATTR_MTIME) {
 814                 iattr.ia_mtime = stbuf->st_mtime;
 815                 iattr.ia_valid |= ATTR_MTIME | ATTR_MTIME_SET;
 816         }
 817         if (mask & SETATTR_ATIME) {
 818                 iattr.ia_atime = stbuf->st_atime;
 819                 iattr.ia_valid |= ATTR_ATIME | ATTR_ATIME_SET;
 820         }
 821         if (mask & SETATTR_UID) {
 822                 iattr.ia_uid = stbuf->st_uid;
 823                 iattr.ia_valid |= ATTR_UID;
 824         }
 825         if (mask & SETATTR_GID) {
 826                 iattr.ia_gid = stbuf->st_gid;
 827                 iattr.ia_valid |= ATTR_GID;
 828         }
 829         if (mask & SETATTR_LEN) {
 830                 iattr.ia_size = stbuf->st_size; /* XXX signed expansion problem */
 831                 iattr.ia_valid |= ATTR_SIZE;
 832         }
 833
 834         iattr.ia_valid |= ATTR_RAW | ATTR_CTIME;
 835         iattr.ia_ctime = CFS_CURRENT_TIME;
 836
 837         rc = llu_setattr_raw(ino, &iattr);
 838         liblustre_wait_idle();
 839         RETURN(rc);
 840 }
 841
 842 #define EXT2_LINK_MAX           32000
 843
 844 static int llu_iop_symlink_raw(struct pnode *pno, const char *tgt)
 845 {
 846         struct inode *dir = pno->p_base->pb_parent->pb_ino;
 847         struct qstr *qstr = &pno->p_base->pb_name;
 848         const char *name = qstr->name;
 849         int len = qstr->len;
 850         struct ptlrpc_request *request = NULL;
 851         struct llu_sb_info *sbi = llu_i2sbi(dir);
 852         struct md_op_data op_data = {{ 0 }};
 853         int err = -EMLINK;
 854         ENTRY;
 855
 856         liblustre_wait_event(0);
 857         if (llu_i2stat(dir)->st_nlink >= EXT2_LINK_MAX)
 858                 RETURN(err);
 859
 860         llu_prep_md_op_data(&op_data, dir, NULL, name, len, 0,
 861                             LUSTRE_OPC_SYMLINK);
 862
 863         err = md_create(sbi->ll_md_exp, &op_data, tgt, strlen(tgt) + 1,
 864                         S_IFLNK | S_IRWXUGO, current->fsuid, current->fsgid,
 865                         cfs_curproc_cap_pack(), 0, &request);
 866         ptlrpc_req_finished(request);
 867         liblustre_wait_event(0);
 868         RETURN(err);
 869 }
 870
 871 static int llu_readlink_internal(struct inode *inode,
 872                                  struct ptlrpc_request **request,
 873                                  char **symname)
 874 {
 875         struct llu_inode_info *lli = llu_i2info(inode);
 876         struct llu_sb_info *sbi = llu_i2sbi(inode);
 877         struct mdt_body *body;
 878         struct intnl_stat *st = llu_i2stat(inode);
 879         struct md_op_data op_data = {{ 0 }};
 880         int rc, symlen = st->st_size + 1;
 881         ENTRY;
 882
 883         *request = NULL;
 884         *symname = NULL;
 885
 886         if (lli->lli_symlink_name) {
 887                 *symname = lli->lli_symlink_name;
 888                 CDEBUG(D_INODE, "using cached symlink %s\n", *symname);
 889                 RETURN(0);
 890         }
 891
 892         llu_prep_md_op_data(&op_data, inode, NULL, NULL, 0, symlen,
 893                             LUSTRE_OPC_ANY);
 894         op_data.op_valid = OBD_MD_LINKNAME;
 895
 896         rc = md_getattr(sbi->ll_md_exp, &op_data, request);
 897         if (rc) {
 898                 CERROR("inode %llu: rc = %d\n", (long long)st->st_ino, rc);
 899                 RETURN(rc);
 900         }
 901
 902         body = req_capsule_server_get(&(*request)->rq_pill, &RMF_MDT_BODY);
 903         LASSERT(body != NULL);
 904
 905         if ((body->valid & OBD_MD_LINKNAME) == 0) {
 906                 CERROR ("OBD_MD_LINKNAME not set on reply\n");
 907                 GOTO (failed, rc = -EPROTO);
 908         }
 909
 910         LASSERT(symlen != 0);
 911         if (body->eadatasize != symlen) {
 912                 CERROR("inode %llu: symlink length %d not expected %d\n",
 913                        (long long)st->st_ino, body->eadatasize - 1, symlen - 1);
 914                 GOTO(failed, rc = -EPROTO);
 915         }
 916
 917         *symname = req_capsule_server_get(&(*request)->rq_pill, &RMF_MDT_MD);
 918         if (*symname == NULL ||
 919             strnlen(*symname, symlen) != symlen - 1) {
 920                 /* not full/NULL terminated */
 921                 CERROR("inode %llu: symlink not NULL terminated string"
 922                        "of length %d\n", (long long)st->st_ino, symlen - 1);
 923                 GOTO(failed, rc = -EPROTO);
 924         }
 925
 926         OBD_ALLOC(lli->lli_symlink_name, symlen);
 927         /* do not return an error if we cannot cache the symlink locally */
 928         if (lli->lli_symlink_name)
 929                 memcpy(lli->lli_symlink_name, *symname, symlen);
 930
 931         RETURN(0);
 932
 933  failed:
 934         ptlrpc_req_finished (*request);
 935         RETURN (-EPROTO);
 936 }
 937
 938 static int llu_iop_readlink(struct pnode *pno, char *data, size_t bufsize)
 939 {
 940         struct inode *inode = pno->p_base->pb_ino;
 941         struct ptlrpc_request *request;
 942         char *symname;
 943         int rc;
 944         ENTRY;
 945
 946         liblustre_wait_event(0);
 947         rc = llu_readlink_internal(inode, &request, &symname);
 948         if (rc)
 949                 GOTO(out, rc);
 950
 951         LASSERT(symname);
 952         strncpy(data, symname, bufsize);
 953         rc = strlen(symname);
 954
 955         ptlrpc_req_finished(request);
 956  out:
 957         liblustre_wait_event(0);
 958         RETURN(rc);
 959 }
 960
 961 static int llu_iop_mknod_raw(struct pnode *pno,
 962                              mode_t mode,
 963                              dev_t dev)
 964 {
 965         struct ptlrpc_request *request = NULL;
 966         struct inode *dir = pno->p_parent->p_base->pb_ino;
 967         struct llu_sb_info *sbi = llu_i2sbi(dir);
 968         struct md_op_data op_data = {{ 0 }};
 969         int err = -EMLINK;
 970         ENTRY;
 971
 972         liblustre_wait_event(0);
 973         CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s,dir=%llu\n",
 974                (int)pno->p_base->pb_name.len, pno->p_base->pb_name.name,
 975                (long long)llu_i2stat(dir)->st_ino);
 976
 977         if (llu_i2stat(dir)->st_nlink >= EXT2_LINK_MAX)
 978                 RETURN(err);
 979
 980         switch (mode & S_IFMT) {
 981         case 0:
 982         case S_IFREG:
 983                 mode |= S_IFREG; /* for mode = 0 case, fallthrough */
 984         case S_IFCHR:
 985         case S_IFBLK:
 986         case S_IFIFO:
 987         case S_IFSOCK:
 988                 llu_prep_md_op_data(&op_data, dir, NULL,
 989                                     pno->p_base->pb_name.name,
 990                                     pno->p_base->pb_name.len, 0,
 991                                     LUSTRE_OPC_MKNOD);
 992
 993                 err = md_create(sbi->ll_md_exp, &op_data, NULL, 0, mode,
 994                                 current->fsuid, current->fsgid,
 995                                 cfs_curproc_cap_pack(), dev, &request);
 996                 ptlrpc_req_finished(request);
 997                 break;
 998         case S_IFDIR:
 999                 err = -EPERM;
1000                 break;
1001         default:
1002                 err = -EINVAL;
1003         }
1004         liblustre_wait_event(0);
1005         RETURN(err);
1006 }
1007
1008 static int llu_iop_link_raw(struct pnode *old, struct pnode *new)
1009 {
1010         struct inode *src = old->p_base->pb_ino;
1011         struct inode *dir = new->p_parent->p_base->pb_ino;
1012         const char *name = new->p_base->pb_name.name;
1013         int namelen = new->p_base->pb_name.len;
1014         struct ptlrpc_request *request = NULL;
1015         struct md_op_data op_data = {{ 0 }};
1016         int rc;
1017         ENTRY;
1018
1019         LASSERT(src);
1020         LASSERT(dir);
1021
1022         liblustre_wait_event(0);
1023         llu_prep_md_op_data(&op_data, src, dir, name, namelen, 0,
1024                             LUSTRE_OPC_ANY);
1025         rc = md_link(llu_i2sbi(src)->ll_md_exp, &op_data, &request);
1026         ptlrpc_req_finished(request);
1027         liblustre_wait_event(0);
1028
1029         RETURN(rc);
1030 }
1031
1032 /*
1033  * libsysio will clear the inode immediately after return
1034  */
1035 static int llu_iop_unlink_raw(struct pnode *pno)
1036 {
1037         struct inode *dir = pno->p_base->pb_parent->pb_ino;
1038         struct qstr *qstr = &pno->p_base->pb_name;
1039         const char *name = qstr->name;
1040         int len = qstr->len;
1041         struct inode *target = pno->p_base->pb_ino;
1042         struct ptlrpc_request *request = NULL;
1043         struct md_op_data op_data = { { 0 } };
1044         int rc;
1045         ENTRY;
1046
1047         LASSERT(target);
1048
1049         liblustre_wait_event(0);
1050         llu_prep_md_op_data(&op_data, dir, NULL, name, len, 0,
1051                             LUSTRE_OPC_ANY);
1052         rc = md_unlink(llu_i2sbi(dir)->ll_md_exp, &op_data, &request);
1053         if (!rc)
1054                 rc = llu_objects_destroy(request, dir);
1055         ptlrpc_req_finished(request);
1056         liblustre_wait_idle();
1057
1058         RETURN(rc);
1059 }
1060
1061 static int llu_iop_rename_raw(struct pnode *old, struct pnode *new)
1062 {
1063         struct inode *src = old->p_parent->p_base->pb_ino;
1064         struct inode *tgt = new->p_parent->p_base->pb_ino;
1065         const char *oldname = old->p_base->pb_name.name;
1066         int oldnamelen = old->p_base->pb_name.len;
1067         const char *newname = new->p_base->pb_name.name;
1068         int newnamelen = new->p_base->pb_name.len;
1069         struct ptlrpc_request *request = NULL;
1070         struct md_op_data op_data = { { 0 } };
1071         int rc;
1072         ENTRY;
1073
1074         LASSERT(src);
1075         LASSERT(tgt);
1076
1077         liblustre_wait_event(0);
1078         llu_prep_md_op_data(&op_data, src, tgt, NULL, 0, 0,
1079                             LUSTRE_OPC_ANY);
1080         rc = md_rename(llu_i2sbi(src)->ll_md_exp, &op_data,
1081                        oldname, oldnamelen, newname, newnamelen,
1082                        &request);
1083         if (!rc) {
1084                 rc = llu_objects_destroy(request, src);
1085         }
1086
1087         ptlrpc_req_finished(request);
1088         liblustre_wait_idle();
1089
1090         RETURN(rc);
1091 }
1092
1093 #ifdef _HAVE_STATVFS
1094 static int llu_statfs_internal(struct llu_sb_info *sbi,
1095                                struct obd_statfs *osfs, __u64 max_age)
1096 {
1097         struct obd_statfs obd_osfs;
1098         int rc;
1099         ENTRY;
1100
1101         rc = obd_statfs(class_exp2obd(sbi->ll_md_exp), osfs, max_age, 0);
1102         if (rc) {
1103                 CERROR("md_statfs fails: rc = %d\n", rc);
1104                 RETURN(rc);
1105         }
1106
1107         CDEBUG(D_SUPER, "MDC blocks "LPU64"/"LPU64" objects "LPU64"/"LPU64"\n",
1108                osfs->os_bavail, osfs->os_blocks, osfs->os_ffree,osfs->os_files);
1109
1110         rc = obd_statfs_rqset(class_exp2obd(sbi->ll_dt_exp),
1111                               &obd_statfs, max_age, 0);
1112         if (rc) {
1113                 CERROR("obd_statfs fails: rc = %d\n", rc);
1114                 RETURN(rc);
1115         }
1116
1117         CDEBUG(D_SUPER, "OSC blocks "LPU64"/"LPU64" objects "LPU64"/"LPU64"\n",
1118                obd_osfs.os_bavail, obd_osfs.os_blocks, obd_osfs.os_ffree,
1119                obd_osfs.os_files);
1120
1121         osfs->os_blocks = obd_osfs.os_blocks;
1122         osfs->os_bfree = obd_osfs.os_bfree;
1123         osfs->os_bavail = obd_osfs.os_bavail;
1124
1125         /* If we don't have as many objects free on the OST as inodes
1126          * on the MDS, we reduce the total number of inodes to
1127          * compensate, so that the "inodes in use" number is correct.
1128          */
1129         if (obd_osfs.os_ffree < osfs->os_ffree) {
1130                 osfs->os_files = (osfs->os_files - osfs->os_ffree) +
1131                         obd_osfs.os_ffree;
1132                 osfs->os_ffree = obd_osfs.os_ffree;
1133         }
1134
1135         RETURN(rc);
1136 }
1137
1138 static int llu_statfs(struct llu_sb_info *sbi, struct statfs *sfs)
1139 {
1140         struct obd_statfs osfs;
1141         int rc;
1142
1143         CDEBUG(D_VFSTRACE, "VFS Op:\n");
1144
1145         /* For now we will always get up-to-date statfs values, but in the
1146          * future we may allow some amount of caching on the client (e.g.
1147          * from QOS or lprocfs updates). */
1148         rc = llu_statfs_internal(sbi, &osfs,
1149                                  cfs_time_shift_64(-OBD_STATFS_CACHE_SECONDS));
1150         if (rc)
1151                 return rc;
1152
1153         statfs_unpack(sfs, &osfs);
1154
1155         if (sizeof(sfs->f_blocks) == 4) {
1156                 while (osfs.os_blocks > ~0UL) {
1157                         sfs->f_bsize <<= 1;
1158
1159                         osfs.os_blocks >>= 1;
1160                         osfs.os_bfree >>= 1;
1161                         osfs.os_bavail >>= 1;
1162                 }
1163         }
1164
1165         sfs->f_blocks = osfs.os_blocks;
1166         sfs->f_bfree = osfs.os_bfree;
1167         sfs->f_bavail = osfs.os_bavail;
1168
1169         return 0;
1170 }
1171
1172 static int llu_iop_statvfs(struct pnode *pno,
1173                            struct inode *ino,
1174                            struct intnl_statvfs *buf)
1175 {
1176         struct statfs fs;
1177         int rc;
1178         ENTRY;
1179
1180         liblustre_wait_event(0);
1181
1182 #ifndef __CYGWIN__
1183         LASSERT(pno->p_base->pb_ino);
1184         rc = llu_statfs(llu_i2sbi(pno->p_base->pb_ino), &fs);
1185         if (rc)
1186                 RETURN(rc);
1187
1188         /* from native driver */
1189         buf->f_bsize = fs.f_bsize;  /* file system block size */
1190         buf->f_frsize = fs.f_bsize; /* file system fundamental block size */
1191         buf->f_blocks = fs.f_blocks;
1192         buf->f_bfree = fs.f_bfree;
1193         buf->f_bavail = fs.f_bavail;
1194         buf->f_files = fs.f_files;  /* Total number serial numbers */
1195         buf->f_ffree = fs.f_ffree;  /* Number free serial numbers */
1196         buf->f_favail = fs.f_ffree; /* Number free ser num for non-privileged*/
1197         buf->f_fsid = fs.f_fsid.__val[1];
1198         buf->f_flag = 0;            /* No equiv in statfs; maybe use type? */
1199         buf->f_namemax = fs.f_namelen;
1200 #endif
1201
1202         liblustre_wait_event(0);
1203         RETURN(0);
1204 }
1205 #endif /* _HAVE_STATVFS */
1206
1207 static int llu_iop_mkdir_raw(struct pnode *pno, mode_t mode)
1208 {
1209         struct inode *dir = pno->p_base->pb_parent->pb_ino;
1210         struct qstr *qstr = &pno->p_base->pb_name;
1211         const char *name = qstr->name;
1212         int len = qstr->len;
1213         struct ptlrpc_request *request = NULL;
1214         struct intnl_stat *st = llu_i2stat(dir);
1215         struct md_op_data op_data = {{ 0 }};
1216         int err = -EMLINK;
1217         ENTRY;
1218
1219         liblustre_wait_event(0);
1220         CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s,dir=%llu/%lu(%p)\n", len, name,
1221                (long long)st->st_ino, llu_i2info(dir)->lli_st_generation, dir);
1222
1223         if (st->st_nlink >= EXT2_LINK_MAX)
1224                 RETURN(err);
1225
1226         llu_prep_md_op_data(&op_data, dir, NULL, name, len, 0,
1227                             LUSTRE_OPC_MKDIR);
1228
1229         err = md_create(llu_i2sbi(dir)->ll_md_exp, &op_data, NULL, 0,
1230                         mode | S_IFDIR, current->fsuid, current->fsgid,
1231                         cfs_curproc_cap_pack(), 0, &request);
1232         ptlrpc_req_finished(request);
1233         liblustre_wait_event(0);
1234         RETURN(err);
1235 }
1236
1237 static int llu_iop_rmdir_raw(struct pnode *pno)
1238 {
1239         struct inode *dir = pno->p_base->pb_parent->pb_ino;
1240         struct qstr *qstr = &pno->p_base->pb_name;
1241         const char *name = qstr->name;
1242         int len = qstr->len;
1243         struct ptlrpc_request *request = NULL;
1244         struct md_op_data op_data = {{ 0 }};
1245         int rc;
1246         ENTRY;
1247
1248         liblustre_wait_event(0);
1249         CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s,dir=%llu/%lu(%p)\n", len, name,
1250                (long long)llu_i2stat(dir)->st_ino,
1251                llu_i2info(dir)->lli_st_generation, dir);
1252
1253         llu_prep_md_op_data(&op_data, dir, NULL, name, len, S_IFDIR,
1254                             LUSTRE_OPC_ANY);
1255         rc = md_unlink(llu_i2sbi(dir)->ll_md_exp, &op_data, &request);
1256         ptlrpc_req_finished(request);
1257
1258         liblustre_wait_event(0);
1259         RETURN(rc);
1260 }
1261
1262 #ifdef O_DIRECT
1263 #define FCNTL_FLMASK (O_APPEND|O_NONBLOCK|O_ASYNC|O_DIRECT)
1264 #else
1265 #define FCNTL_FLMASK (O_APPEND|O_NONBLOCK|O_ASYNC)
1266 #endif
1267 #define FCNTL_FLMASK_INVALID (O_NONBLOCK|O_ASYNC)
1268
1269 /* refer to ll_file_flock() for details */
1270 static int llu_file_flock(struct inode *ino,
1271                           int cmd,
1272                           struct file_lock *file_lock)
1273 {
1274         struct llu_inode_info *lli = llu_i2info(ino);
1275         struct intnl_stat *st = llu_i2stat(ino);
1276         struct ldlm_res_id res_id =
1277                 { .name = {fid_seq(&lli->lli_fid),
1278                            fid_oid(&lli->lli_fid),
1279                            fid_ver(&lli->lli_fid),
1280                            LDLM_FLOCK} };
1281         struct ldlm_enqueue_info einfo = { LDLM_FLOCK, 0, NULL,
1282                 ldlm_flock_completion_ast, NULL, NULL, file_lock };
1283
1284         struct lustre_handle lockh = {0};
1285         ldlm_policy_data_t flock;
1286         int flags = 0;
1287         int rc;
1288
1289         CDEBUG(D_VFSTRACE, "VFS Op:inode=%llu file_lock=%p\n",
1290                (unsigned long long)st->st_ino, file_lock);
1291
1292         flock.l_flock.pid = file_lock->fl_pid;
1293         flock.l_flock.start = file_lock->fl_start;
1294         flock.l_flock.end = file_lock->fl_end;
1295
1296         switch (file_lock->fl_type) {
1297         case F_RDLCK:
1298                 einfo.ei_mode = LCK_PR;
1299                 break;
1300         case F_UNLCK:
1301                 einfo.ei_mode = LCK_NL;
1302                 break;
1303         case F_WRLCK:
1304                 einfo.ei_mode = LCK_PW;
1305                 break;
1306         default:
1307                 CERROR("unknown fcntl lock type: %d\n", file_lock->fl_type);
1308                 LBUG();
1309         }
1310
1311         switch (cmd) {
1312         case F_SETLKW:
1313 #ifdef F_SETLKW64
1314 #if F_SETLKW64 != F_SETLKW
1315         case F_SETLKW64:
1316 #endif
1317 #endif
1318                 flags = 0;
1319                 break;
1320         case F_SETLK:
1321 #ifdef F_SETLK64
1322 #if F_SETLK64 != F_SETLK
1323         case F_SETLK64:
1324 #endif
1325 #endif
1326                 flags = LDLM_FL_BLOCK_NOWAIT;
1327                 break;
1328         case F_GETLK:
1329 #ifdef F_GETLK64
1330 #if F_GETLK64 != F_GETLK
1331         case F_GETLK64:
1332 #endif
1333 #endif
1334                 flags = LDLM_FL_TEST_LOCK;
1335                 file_lock->fl_type = einfo.ei_mode;
1336                 break;
1337         default:
1338                 CERROR("unknown fcntl cmd: %d\n", cmd);
1339                 LBUG();
1340         }
1341
1342         CDEBUG(D_DLMTRACE, "inode=%llu, pid=%u, cmd=%d, flags=%#x, mode=%u, "
1343                "start="LPX64", end="LPX64"\n", (unsigned long long)st->st_ino,
1344                flock.l_flock.pid, cmd, flags, einfo.ei_mode, flock.l_flock.start,
1345                flock.l_flock.end);
1346
1347         {
1348                 struct lmv_obd *lmv;
1349                 struct obd_device *lmv_obd;
1350                 lmv_obd = class_exp2obd(llu_i2mdexp(ino));
1351                 lmv = &lmv_obd->u.lmv;
1352
1353                 if (lmv->desc.ld_tgt_count < 1)
1354                         RETURN(rc = -ENODEV);
1355
1356                 if (lmv->tgts[0].ltd_exp != NULL)
1357                         rc = ldlm_cli_enqueue(lmv->tgts[0].ltd_exp, NULL, &einfo, &res_id,
1358                                               &flock, &flags, NULL, 0, &lockh, 0);
1359                 else
1360                         rc = -ENODEV;
1361         }
1362         RETURN(rc);
1363 }
1364
1365 static int assign_type(struct file_lock *fl, int type)
1366 {
1367         switch (type) {
1368         case F_RDLCK:
1369         case F_WRLCK:
1370         case F_UNLCK:
1371                 fl->fl_type = type;
1372                 return 0;
1373         default:
1374                 return -EINVAL;
1375         }
1376 }
1377
1378 static int flock_to_posix_lock(struct inode *ino,
1379                                struct file_lock *fl,
1380                                struct flock *l)
1381 {
1382         switch (l->l_whence) {
1383         /* XXX: only SEEK_SET is supported in lustre */
1384         case SEEK_SET:
1385                 fl->fl_start = 0;
1386                 break;
1387         default:
1388                 return -EINVAL;
1389         }
1390
1391         fl->fl_end = l->l_len - 1;
1392         if (l->l_len < 0)
1393                 return -EINVAL;
1394         if (l->l_len == 0)
1395                 fl->fl_end = OFFSET_MAX;
1396
1397         fl->fl_pid = getpid();
1398         fl->fl_flags = FL_POSIX;
1399         fl->fl_notify = NULL;
1400         fl->fl_insert = NULL;
1401         fl->fl_remove = NULL;
1402         /* XXX: these fields can't be filled with suitable values,
1403                 but I think lustre doesn't use them.
1404          */
1405         fl->fl_owner = NULL;
1406         fl->fl_file = NULL;
1407
1408         return assign_type(fl, l->l_type);
1409 }
1410
1411 static int llu_fcntl_getlk(struct inode *ino, struct flock *flock)
1412 {
1413         struct file_lock fl;
1414         int error;
1415
1416         error = EINVAL;
1417         if ((flock->l_type != F_RDLCK) && (flock->l_type != F_WRLCK))
1418                 goto out;
1419
1420         error = flock_to_posix_lock(ino, &fl, flock);
1421         if (error)
1422                 goto out;
1423
1424         error = llu_file_flock(ino, F_GETLK, &fl);
1425         if (error)
1426                 goto out;
1427
1428         flock->l_type = F_UNLCK;
1429         if (fl.fl_type != F_UNLCK) {
1430                 flock->l_pid = fl.fl_pid;
1431                 flock->l_start = fl.fl_start;
1432                 flock->l_len = fl.fl_end == OFFSET_MAX ? 0:
1433                         fl.fl_end - fl.fl_start + 1;
1434                 flock->l_whence = SEEK_SET;
1435                 flock->l_type = fl.fl_type;
1436         }
1437
1438 out:
1439         return error;
1440 }
1441
1442 static int llu_fcntl_setlk(struct inode *ino, int cmd, struct flock *flock)
1443 {
1444         struct file_lock fl;
1445         int flags = llu_i2info(ino)->lli_open_flags + 1;
1446         int error;
1447
1448         error = flock_to_posix_lock(ino, &fl, flock);
1449         if (error)
1450                 goto out;
1451         if (cmd == F_SETLKW)
1452                 fl.fl_flags |= FL_SLEEP;
1453
1454         error = -EBADF;
1455         switch (flock->l_type) {
1456         case F_RDLCK:
1457                 if (!(flags & FMODE_READ))
1458                         goto out;
1459                 break;
1460         case F_WRLCK:
1461                 if (!(flags & FMODE_WRITE))
1462                         goto out;
1463                 break;
1464         case F_UNLCK:
1465                 break;
1466         default:
1467                 error = -EINVAL;
1468                 goto out;
1469         }
1470
1471         error = llu_file_flock(ino, cmd, &fl);
1472         if (error)
1473                 goto out;
1474
1475 out:
1476         return error;
1477 }
1478
1479 static int llu_iop_fcntl(struct inode *ino, int cmd, va_list ap, int *rtn)
1480 {
1481         struct llu_inode_info *lli = llu_i2info(ino);
1482         long flags;
1483         struct flock *flock;
1484         long err = 0;
1485
1486         liblustre_wait_event(0);
1487         switch (cmd) {
1488         case F_GETFL:
1489                 *rtn = lli->lli_open_flags;
1490                 break;
1491         case F_SETFL:
1492                 flags = va_arg(ap, long);
1493                 flags &= FCNTL_FLMASK;
1494                 if (flags & FCNTL_FLMASK_INVALID) {
1495                         LCONSOLE_ERROR_MSG(0x010, "liblustre does not support "
1496                                            "the O_NONBLOCK or O_ASYNC flags. "
1497                                            "Please fix your application.\n");
1498                         *rtn = -EINVAL;
1499                         err = EINVAL;
1500                         break;
1501                 }
1502                 lli->lli_open_flags = (int)(flags & FCNTL_FLMASK) |
1503                                       (lli->lli_open_flags & ~FCNTL_FLMASK);
1504                 *rtn = 0;
1505                 break;
1506         case F_GETLK:
1507 #ifdef F_GETLK64
1508 #if F_GETLK64 != F_GETLK
1509         case F_GETLK64:
1510 #endif
1511 #endif
1512                 flock = va_arg(ap, struct flock *);
1513                 err = llu_fcntl_getlk(ino, flock);
1514                 *rtn = err? -1: 0;
1515                 break;
1516         case F_SETLK:
1517 #ifdef F_SETLKW64
1518 #if F_SETLKW64 != F_SETLKW
1519         case F_SETLKW64:
1520 #endif
1521 #endif
1522         case F_SETLKW:
1523 #ifdef F_SETLK64
1524 #if F_SETLK64 != F_SETLK
1525         case F_SETLK64:
1526 #endif
1527 #endif
1528                 flock = va_arg(ap, struct flock *);
1529                 err = llu_fcntl_setlk(ino, cmd, flock);
1530                 *rtn = err? -1: 0;
1531                 break;
1532         default:
1533                 CERROR("unsupported fcntl cmd %x\n", cmd);
1534                 *rtn = -ENOSYS;
1535                 err = ENOSYS;
1536                 break;
1537         }
1538
1539         liblustre_wait_event(0);
1540         return err;
1541 }
1542
1543 static int llu_get_grouplock(struct inode *inode, unsigned long arg)
1544 {
1545         struct llu_inode_info *lli = llu_i2info(inode);
1546         struct ll_file_data *fd = lli->lli_file_data;
1547         int rc;
1548         struct ccc_grouplock grouplock;
1549         ENTRY;
1550
1551         if (fd->fd_flags & LL_FILE_IGNORE_LOCK) {
1552                 RETURN(-ENOTSUPP);
1553         }
1554         if (fd->fd_flags & LL_FILE_GROUP_LOCKED) {
1555                 RETURN(-EINVAL);
1556         }
1557         LASSERT(fd->fd_grouplock.cg_lock == NULL);
1558
1559         rc = cl_get_grouplock(cl_i2info(inode)->lli_clob,
1560                               arg, (lli->lli_open_flags & O_NONBLOCK),
1561                               &grouplock);
1562
1563         if (rc)
1564                 RETURN(rc);
1565
1566         fd->fd_flags |= LL_FILE_GROUP_LOCKED;
1567         fd->fd_grouplock = grouplock;
1568
1569         RETURN(0);
1570 }
1571
1572 int llu_put_grouplock(struct inode *inode, unsigned long arg)
1573 {
1574         struct llu_inode_info *lli = llu_i2info(inode);
1575         struct ll_file_data *fd = lli->lli_file_data;
1576         struct ccc_grouplock grouplock;
1577         ENTRY;
1578
1579         if (!(fd->fd_flags & LL_FILE_GROUP_LOCKED))
1580                 RETURN(-EINVAL);
1581
1582         LASSERT(fd->fd_grouplock.cg_lock != NULL);
1583
1584         if (fd->fd_grouplock.cg_gid != arg)
1585                 RETURN(-EINVAL);
1586
1587         grouplock = fd->fd_grouplock;
1588         memset(&fd->fd_grouplock, 0, sizeof(fd->fd_grouplock));
1589         fd->fd_flags &= ~LL_FILE_GROUP_LOCKED;
1590
1591         cl_put_grouplock(&grouplock);
1592
1593         RETURN(0);
1594 }
1595
1596 static int llu_lov_dir_setstripe(struct inode *ino, unsigned long arg)
1597 {
1598         struct llu_sb_info *sbi = llu_i2sbi(ino);
1599         struct ptlrpc_request *request = NULL;
1600         struct md_op_data op_data = {{ 0 }};
1601         struct lov_user_md lum, *lump = (struct lov_user_md *)arg;
1602         int rc = 0;
1603
1604         llu_prep_md_op_data(&op_data, ino, NULL, NULL, 0, 0,
1605                             LUSTRE_OPC_ANY);
1606
1607         LASSERT(sizeof(lum) == sizeof(*lump));
1608         LASSERT(sizeof(lum.lmm_objects[0]) ==
1609                 sizeof(lump->lmm_objects[0]));
1610         if (cfs_copy_from_user(&lum, lump, sizeof(lum)))
1611                 return(-EFAULT);
1612
1613         switch (lum.lmm_magic) {
1614         case LOV_USER_MAGIC_V1: {
1615                 if (lum.lmm_magic != cpu_to_le32(LOV_USER_MAGIC_V1))
1616                         lustre_swab_lov_user_md_v1(&lum);
1617                 break;
1618                 }
1619         case LOV_USER_MAGIC_V3: {
1620                 if (lum.lmm_magic != cpu_to_le32(LOV_USER_MAGIC_V3))
1621                         lustre_swab_lov_user_md_v3((struct lov_user_md_v3 *)&lum);
1622                 break;
1623                 }
1624         default: {
1625                 CDEBUG(D_IOCTL, "bad userland LOV MAGIC:"
1626                                 " %#08x != %#08x nor %#08x\n",
1627                                 lum.lmm_magic, LOV_USER_MAGIC_V1,
1628                                 LOV_USER_MAGIC_V3);
1629                 RETURN(-EINVAL);
1630         }
1631         }
1632
1633         /* swabbing is done in lov_setstripe() on server side */
1634         rc = md_setattr(sbi->ll_md_exp, &op_data, &lum,
1635                         sizeof(lum), NULL, 0, &request, NULL);
1636         if (rc) {
1637                 ptlrpc_req_finished(request);
1638                 if (rc != -EPERM && rc != -EACCES)
1639                         CERROR("md_setattr fails: rc = %d\n", rc);
1640                 return rc;
1641         }
1642         ptlrpc_req_finished(request);
1643
1644         return rc;
1645 }
1646
1647 static int llu_lov_setstripe_ea_info(struct inode *ino, int flags,
1648                                      struct lov_user_md *lum, int lum_size)
1649 {
1650         struct llu_sb_info *sbi = llu_i2sbi(ino);
1651         struct llu_inode_info *lli = llu_i2info(ino);
1652         struct lookup_intent oit = {.it_op = IT_OPEN, .it_flags = flags};
1653         struct ldlm_enqueue_info einfo = { LDLM_IBITS, LCK_CR,
1654                 llu_md_blocking_ast, ldlm_completion_ast, NULL, NULL, NULL };
1655         struct ptlrpc_request *req = NULL;
1656         struct lustre_md md;
1657         struct md_op_data data = {{ 0 }};
1658         struct lustre_handle lockh;
1659         int rc = 0;
1660         ENTRY;
1661
1662         if (lli->lli_smd) {
1663                 CDEBUG(D_IOCTL, "stripe already exists for ino "DFID"\n",
1664                        PFID(&lli->lli_fid));
1665                 return -EEXIST;
1666         }
1667
1668         llu_prep_md_op_data(&data, NULL, ino, NULL, 0, O_RDWR,
1669                             LUSTRE_OPC_ANY);
1670         rc = md_enqueue(sbi->ll_md_exp, &einfo, &oit, &data,
1671                         &lockh, lum, lum_size, NULL, LDLM_FL_INTENT_ONLY);
1672         if (rc)
1673                 GOTO(out, rc);
1674
1675         req = oit.d.lustre.it_data;
1676         rc = it_open_error(DISP_IT_EXECD, &oit);
1677         if (rc) {
1678                 req->rq_replay = 0;
1679                 GOTO(out, rc);
1680         }
1681
1682         rc = it_open_error(DISP_OPEN_OPEN, &oit);
1683         if (rc) {
1684                 req->rq_replay = 0;
1685                 GOTO(out, rc);
1686         }
1687
1688         rc = md_get_lustre_md(sbi->ll_md_exp, req,
1689                               sbi->ll_dt_exp, sbi->ll_md_exp, &md);
1690         if (rc)
1691                 GOTO(out, rc);
1692
1693         llu_update_inode(ino, &md);
1694         llu_local_open(lli, &oit);
1695         /* release intent */
1696         if (lustre_handle_is_used(&lockh))
1697                 ldlm_lock_decref(&lockh, LCK_CR);
1698         ptlrpc_req_finished(req);
1699         req = NULL;
1700         rc = llu_file_release(ino);
1701         EXIT;
1702
1703 out:
1704         if (req != NULL)
1705                 ptlrpc_req_finished(req);
1706         return rc;
1707 }
1708
1709 static int llu_lov_file_setstripe(struct inode *ino, unsigned long arg)
1710 {
1711         struct lov_user_md lum, *lump = (struct lov_user_md *)arg;
1712         int rc;
1713         int flags = FMODE_WRITE;
1714         ENTRY;
1715
1716         LASSERT(sizeof(lum) == sizeof(*lump));
1717         LASSERT(sizeof(lum.lmm_objects[0]) == sizeof(lump->lmm_objects[0]));
1718         if (cfs_copy_from_user(&lum, lump, sizeof(lum)))
1719                 RETURN(-EFAULT);
1720
1721         rc = llu_lov_setstripe_ea_info(ino, flags, &lum, sizeof(lum));
1722         RETURN(rc);
1723 }
1724
1725 static int llu_lov_setstripe(struct inode *ino, unsigned long arg)
1726 {
1727         struct intnl_stat *st = llu_i2stat(ino);
1728         if (S_ISREG(st->st_mode))
1729                 return llu_lov_file_setstripe(ino, arg);
1730         if (S_ISDIR(st->st_mode))
1731                 return llu_lov_dir_setstripe(ino, arg);
1732
1733         return -EINVAL;
1734 }
1735
1736 static int llu_lov_getstripe(struct inode *ino, unsigned long arg)
1737 {
1738         struct lov_stripe_md *lsm = llu_i2info(ino)->lli_smd;
1739
1740         if (!lsm)
1741                 RETURN(-ENODATA);
1742
1743         return obd_iocontrol(LL_IOC_LOV_GETSTRIPE, llu_i2obdexp(ino), 0, lsm,
1744                             (void *)arg);
1745 }
1746
1747 static int llu_iop_ioctl(struct inode *ino, unsigned long int request,
1748                          va_list ap)
1749 {
1750         unsigned long arg;
1751         int rc;
1752
1753         liblustre_wait_event(0);
1754
1755         switch (request) {
1756         case LL_IOC_GROUP_LOCK:
1757                 arg = va_arg(ap, unsigned long);
1758                 rc = llu_get_grouplock(ino, arg);
1759                 break;
1760         case LL_IOC_GROUP_UNLOCK:
1761                 arg = va_arg(ap, unsigned long);
1762                 rc = llu_put_grouplock(ino, arg);
1763                 break;
1764         case LL_IOC_LOV_SETSTRIPE:
1765                 arg = va_arg(ap, unsigned long);
1766                 rc = llu_lov_setstripe(ino, arg);
1767                 break;
1768         case LL_IOC_LOV_GETSTRIPE:
1769                 arg = va_arg(ap, unsigned long);
1770                 rc = llu_lov_getstripe(ino, arg);
1771                 break;
1772         default:
1773                 CERROR("did not support ioctl cmd %lx\n", request);
1774                 rc = -ENOSYS;
1775                 break;
1776         }
1777
1778         liblustre_wait_event(0);
1779         return rc;
1780 }
1781
1782 /*
1783  * we already do syncronous read/write
1784  */
1785 static int llu_iop_sync(struct inode *inode)
1786 {
1787         liblustre_wait_event(0);
1788         return 0;
1789 }
1790
1791 static int llu_iop_datasync(struct inode *inode)
1792 {
1793         liblustre_wait_event(0);
1794         return 0;
1795 }
1796
1797 struct filesys_ops llu_filesys_ops =
1798 {
1799         fsop_gone: llu_fsop_gone,
1800 };
1801
1802 struct inode *llu_iget(struct filesys *fs, struct lustre_md *md)
1803 {
1804         struct inode *inode;
1805         struct lu_fid fid;
1806         struct file_identifier fileid = {&fid, sizeof(fid)};
1807
1808         if ((md->body->valid & (OBD_MD_FLID | OBD_MD_FLTYPE)) !=
1809             (OBD_MD_FLID | OBD_MD_FLTYPE)) {
1810                 CERROR("bad md body valid mask "LPX64"\n", md->body->valid);
1811                 LBUG();
1812                 return ERR_PTR(-EPERM);
1813         }
1814
1815         /* try to find existing inode */
1816         fid = md->body->fid1;
1817
1818         inode = _sysio_i_find(fs, &fileid);
1819         if (inode) {
1820                 if (inode->i_zombie/* ||
1821                     lli->lli_st_generation != md->body->generation*/) {
1822                         I_RELE(inode);
1823                 }
1824                 else {
1825                         llu_update_inode(inode, md);
1826                         return inode;
1827                 }
1828         }
1829
1830         inode = llu_new_inode(fs, &fid);
1831         if (inode)
1832                 llu_update_inode(inode, md);
1833
1834         return inode;
1835 }
1836
1837 static int
1838 llu_fsswop_mount(const char *source,
1839                  unsigned flags,
1840                  const void *data __IS_UNUSED,
1841                  struct pnode *tocover,
1842                  struct mount **mntp)
1843 {
1844         struct filesys *fs;
1845         struct inode *root;
1846         struct pnode_base *rootpb;
1847         struct obd_device *obd;
1848         struct llu_sb_info *sbi;
1849         struct obd_statfs osfs;
1850         static struct qstr noname = { NULL, 0, 0 };
1851         struct ptlrpc_request *request = NULL;
1852         struct lustre_md md;
1853         class_uuid_t uuid;
1854         struct config_llog_instance cfg = {0, };
1855         struct lustre_profile *lprof;
1856         char *zconf_mgsnid, *zconf_profile;
1857         char *osc = NULL, *mdc = NULL;
1858         int async = 1, err = -EINVAL;
1859         struct obd_connect_data ocd = {0,};
1860         struct md_op_data op_data = {{0}};
1861         /* %p for void* in printf needs 16+2 characters: 0xffffffffffffffff */
1862         const int instlen = sizeof(cfg.cfg_instance) * 2 + 2;
1863
1864         ENTRY;
1865
1866         if (ll_parse_mount_target(source,
1867                                   &zconf_mgsnid,
1868                                   &zconf_profile)) {
1869                 CERROR("mal-formed target %s\n", source);
1870                 RETURN(err);
1871         }
1872         if (!zconf_mgsnid || !zconf_profile) {
1873                 printf("Liblustre: invalid target %s\n", source);
1874                 RETURN(err);
1875         }
1876         /* allocate & initialize sbi */
1877         OBD_ALLOC(sbi, sizeof(*sbi));
1878         if (!sbi)
1879                 RETURN(-ENOMEM);
1880
1881         CFS_INIT_LIST_HEAD(&sbi->ll_conn_chain);
1882         ll_generate_random_uuid(uuid);
1883         class_uuid_unparse(uuid, &sbi->ll_sb_uuid);
1884
1885         /* generate a string unique to this super, let's try
1886          the address of the super itself.*/
1887         cfg.cfg_instance = sbi;
1888
1889         /* retrive & parse config log */
1890         cfg.cfg_uuid = sbi->ll_sb_uuid;
1891         err = liblustre_process_log(&cfg, zconf_mgsnid, zconf_profile, 1);
1892         if (err < 0) {
1893                 CERROR("Unable to process log: %s\n", zconf_profile);
1894                 GOTO(out_free, err);
1895         }
1896
1897         lprof = class_get_profile(zconf_profile);
1898         if (lprof == NULL) {
1899                 CERROR("No profile found: %s\n", zconf_profile);
1900                 GOTO(out_free, err = -EINVAL);
1901         }
1902         OBD_ALLOC(osc, strlen(lprof->lp_dt) + instlen + 2);
1903         sprintf(osc, "%s-%p", lprof->lp_dt, cfg.cfg_instance);
1904
1905         OBD_ALLOC(mdc, strlen(lprof->lp_md) + instlen + 2);
1906         sprintf(mdc, "%s-%p", lprof->lp_md, cfg.cfg_instance);
1907
1908         if (!osc) {
1909                 CERROR("no osc\n");
1910                 GOTO(out_free, err = -EINVAL);
1911         }
1912         if (!mdc) {
1913                 CERROR("no mdc\n");
1914                 GOTO(out_free, err = -EINVAL);
1915         }
1916
1917         fs = _sysio_fs_new(&llu_filesys_ops, flags, sbi);
1918         if (!fs) {
1919                 err = -ENOMEM;
1920                 goto out_free;
1921         }
1922
1923         obd = class_name2obd(mdc);
1924         if (!obd) {
1925                 CERROR("MDC %s: not setup or attached\n", mdc);
1926                 GOTO(out_free, err = -EINVAL);
1927         }
1928         obd_set_info_async(obd->obd_self_export, sizeof(KEY_ASYNC), KEY_ASYNC,
1929                            sizeof(async), &async, NULL);
1930
1931         ocd.ocd_connect_flags = OBD_CONNECT_IBITS | OBD_CONNECT_VERSION |
1932                                 OBD_CONNECT_FID | OBD_CONNECT_AT |
1933                                 OBD_CONNECT_VBR | OBD_CONNECT_FULL20;
1934
1935 #ifdef LIBLUSTRE_POSIX_ACL
1936         ocd.ocd_connect_flags |= OBD_CONNECT_ACL;
1937 #endif
1938         ocd.ocd_ibits_known = MDS_INODELOCK_FULL;
1939         ocd.ocd_version = LUSTRE_VERSION_CODE;
1940
1941         /* setup mdc */
1942         err = obd_connect(NULL, &sbi->ll_md_exp, obd, &sbi->ll_sb_uuid, &ocd, NULL);
1943         if (err) {
1944                 CERROR("cannot connect to %s: rc = %d\n", mdc, err);
1945                 GOTO(out_free, err);
1946         }
1947
1948         err = obd_statfs(obd, &osfs, 100000000, 0);
1949         if (err)
1950                 GOTO(out_md, err);
1951
1952         /*
1953          * FIXME fill fs stat data into sbi here!!! FIXME
1954          */
1955
1956         /* setup osc */
1957         obd = class_name2obd(osc);
1958         if (!obd) {
1959                 CERROR("OSC %s: not setup or attached\n", osc);
1960                 GOTO(out_md, err = -EINVAL);
1961         }
1962         obd_set_info_async(obd->obd_self_export, sizeof(KEY_ASYNC), KEY_ASYNC,
1963                            sizeof(async), &async, NULL);
1964
1965         obd->obd_upcall.onu_owner = &sbi->ll_lco;
1966         obd->obd_upcall.onu_upcall = cl_ocd_update;
1967
1968         ocd.ocd_connect_flags = OBD_CONNECT_SRVLOCK | OBD_CONNECT_REQPORTAL |
1969                                 OBD_CONNECT_VERSION | OBD_CONNECT_TRUNCLOCK |
1970                                 OBD_CONNECT_FID | OBD_CONNECT_AT |
1971                                 OBD_CONNECT_FULL20;
1972
1973         ocd.ocd_version = LUSTRE_VERSION_CODE;
1974         err = obd_connect(NULL, &sbi->ll_dt_exp, obd, &sbi->ll_sb_uuid, &ocd, NULL);
1975         if (err) {
1976                 CERROR("cannot connect to %s: rc = %d\n", osc, err);
1977                 GOTO(out_md, err);
1978         }
1979         sbi->ll_lco.lco_flags = ocd.ocd_connect_flags;
1980         sbi->ll_lco.lco_md_exp = sbi->ll_md_exp;
1981         sbi->ll_lco.lco_dt_exp = sbi->ll_dt_exp;
1982
1983         fid_zero(&sbi->ll_root_fid);
1984         err = md_getstatus(sbi->ll_md_exp, &sbi->ll_root_fid, NULL);
1985         if (err) {
1986                 CERROR("cannot mds_connect: rc = %d\n", err);
1987                 GOTO(out_lock_cn_cb, err);
1988         }
1989         if (!fid_is_sane(&sbi->ll_root_fid)) {
1990                 CERROR("Invalid root fid during mount\n");
1991                 GOTO(out_lock_cn_cb, err = -EINVAL);
1992         }
1993         CDEBUG(D_SUPER, "rootfid "DFID"\n", PFID(&sbi->ll_root_fid));
1994
1995         op_data.op_fid1 = sbi->ll_root_fid;
1996         op_data.op_valid = OBD_MD_FLGETATTR | OBD_MD_FLBLOCKS;
1997         /* fetch attr of root inode */
1998         err = md_getattr(sbi->ll_md_exp, &op_data, &request);
1999         if (err) {
2000                 CERROR("md_getattr failed for root: rc = %d\n", err);
2001                 GOTO(out_lock_cn_cb, err);
2002         }
2003
2004         err = md_get_lustre_md(sbi->ll_md_exp, request,
2005                                sbi->ll_dt_exp, sbi->ll_md_exp, &md);
2006         if (err) {
2007                 CERROR("failed to understand root inode md: rc = %d\n",err);
2008                 GOTO(out_request, err);
2009         }
2010
2011         LASSERT(fid_is_sane(&sbi->ll_root_fid));
2012
2013         root = llu_iget(fs, &md);
2014         if (!root || IS_ERR(root)) {
2015                 CERROR("fail to generate root inode\n");
2016                 GOTO(out_request, err = -EBADF);
2017         }
2018
2019         /*
2020          * Generate base path-node for root.
2021          */
2022         rootpb = _sysio_pb_new(&noname, NULL, root);
2023         if (!rootpb) {
2024                 err = -ENOMEM;
2025                 goto out_inode;
2026         }
2027
2028         err = _sysio_do_mount(fs, rootpb, flags, tocover, mntp);
2029         if (err) {
2030                 _sysio_pb_gone(rootpb);
2031                 goto out_inode;
2032         }
2033
2034         cl_sb_init(sbi);
2035
2036         ptlrpc_req_finished(request);
2037
2038         CDEBUG(D_SUPER, "LibLustre: %s mounted successfully!\n", source);
2039         err = 0;
2040         goto out_free;
2041
2042 out_inode:
2043         _sysio_i_gone(root);
2044 out_request:
2045         ptlrpc_req_finished(request);
2046 out_lock_cn_cb:
2047         obd_disconnect(sbi->ll_dt_exp);
2048 out_md:
2049         obd_disconnect(sbi->ll_md_exp);
2050 out_free:
2051         if (osc)
2052                 OBD_FREE(osc, strlen(lprof->lp_dt) + instlen + 2);
2053         if (mdc)
2054                 OBD_FREE(mdc, strlen(lprof->lp_md) + instlen + 2);
2055         OBD_FREE(sbi, sizeof(*sbi));
2056         liblustre_wait_idle();
2057         return err;
2058 }
2059
2060 struct fssw_ops llu_fssw_ops = {
2061         llu_fsswop_mount
2062 };
2063
2064 static struct inode_ops llu_inode_ops = {
2065         inop_lookup:    llu_iop_lookup,
2066         inop_getattr:   llu_iop_getattr,
2067         inop_setattr:   llu_iop_setattr,
2068         inop_filldirentries:     llu_iop_filldirentries,
2069         inop_mkdir:     llu_iop_mkdir_raw,
2070         inop_rmdir:     llu_iop_rmdir_raw,
2071         inop_symlink:   llu_iop_symlink_raw,
2072         inop_readlink:  llu_iop_readlink,
2073         inop_open:      llu_iop_open,
2074         inop_close:     llu_iop_close,
2075         inop_link:      llu_iop_link_raw,
2076         inop_unlink:    llu_iop_unlink_raw,
2077         inop_rename:    llu_iop_rename_raw,
2078         inop_pos:       llu_iop_pos,
2079         inop_read:      llu_iop_read,
2080         inop_write:     llu_iop_write,
2081         inop_iodone:    llu_iop_iodone,
2082         inop_fcntl:     llu_iop_fcntl,
2083         inop_sync:      llu_iop_sync,
2084         inop_datasync:  llu_iop_datasync,
2085         inop_ioctl:     llu_iop_ioctl,
2086         inop_mknod:     llu_iop_mknod_raw,
2087 #ifdef _HAVE_STATVFS
2088         inop_statvfs:   llu_iop_statvfs,
2089 #endif
2090         inop_gone:      llu_iop_gone,
2091 };