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  2008 Sun Microsystems, Inc. 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 <sysio.h>
  58 #ifdef HAVE_XTIO_H
  59 #include <xtio.h>
  60 #endif
  61 #include <fs.h>
  62 #include <mount.h>
  63 #include <inode.h>
  64 #ifdef HAVE_FILE_H
  65 #include <file.h>
  66 #endif
  67
  68 #undef LIST_HEAD
  69
  70 #include "llite_lib.h"
  71
  72 #ifndef MAY_EXEC
  73 #define MAY_EXEC        1
  74 #define MAY_WRITE       2
  75 #define MAY_READ        4
  76 #endif
  77
  78 #define S_IXUGO (S_IXUSR|S_IXGRP|S_IXOTH)
  79
  80 static int ll_permission(struct inode *inode, int mask)
  81 {
  82         struct intnl_stat *st = llu_i2stat(inode);
  83         mode_t mode = st->st_mode;
  84
  85         if (current->fsuid == st->st_uid)
  86                 mode >>= 6;
  87         else if (in_group_p(st->st_gid))
  88                 mode >>= 3;
  89
  90         if ((mode & mask & (MAY_READ|MAY_WRITE|MAY_EXEC)) == mask)
  91                 return 0;
  92
  93         if ((mask & (MAY_READ|MAY_WRITE)) ||
  94             (st->st_mode & S_IXUGO))
  95                 if (cfs_capable(CFS_CAP_DAC_OVERRIDE))
  96                         return 0;
  97
  98         if (mask == MAY_READ ||
  99             (S_ISDIR(st->st_mode) && !(mask & MAY_WRITE))) {
 100                 if (cfs_capable(CFS_CAP_DAC_READ_SEARCH))
 101                         return 0;
 102         }
 103
 104         return -EACCES;
 105 }
 106
 107 static void llu_fsop_gone(struct filesys *fs)
 108 {
 109         struct llu_sb_info *sbi = (struct llu_sb_info *) fs->fs_private;
 110         struct obd_device *obd = class_exp2obd(sbi->ll_md_exp);
 111         int next = 0;
 112         ENTRY;
 113
 114         list_del(&sbi->ll_conn_chain);
 115         cl_sb_fini(sbi);
 116         obd_disconnect(sbi->ll_dt_exp);
 117         obd_disconnect(sbi->ll_md_exp);
 118
 119         while ((obd = class_devices_in_group(&sbi->ll_sb_uuid, &next)) != NULL)
 120                 class_manual_cleanup(obd);
 121
 122         OBD_FREE(sbi, sizeof(*sbi));
 123
 124         liblustre_wait_idle();
 125         EXIT;
 126 }
 127
 128 static struct inode_ops llu_inode_ops;
 129
 130 static ldlm_mode_t llu_take_md_lock(struct inode *inode, __u64 bits,
 131                                     struct lustre_handle *lockh)
 132 {
 133         ldlm_policy_data_t policy = { .l_inodebits = {bits}};
 134         struct lu_fid *fid;
 135         ldlm_mode_t rc;
 136         int flags;
 137         ENTRY;
 138
 139         fid = &llu_i2info(inode)->lli_fid;
 140         CDEBUG(D_INFO, "trying to match res "DFID"\n", PFID(fid));
 141
 142         flags = LDLM_FL_BLOCK_GRANTED | LDLM_FL_CBPENDING;
 143         rc = md_lock_match(llu_i2mdexp(inode), flags, fid, LDLM_IBITS, &policy,
 144                            LCK_CR|LCK_CW|LCK_PR|LCK_PW, lockh);
 145         RETURN(rc);
 146 }
 147
 148 void llu_update_inode(struct inode *inode, struct lustre_md *md)
 149 {
 150         struct llu_inode_info *lli = llu_i2info(inode);
 151         struct mdt_body *body = md->body;
 152         struct lov_stripe_md *lsm = md->lsm;
 153         struct intnl_stat *st = llu_i2stat(inode);
 154
 155         LASSERT ((lsm != NULL) == ((body->valid & OBD_MD_FLEASIZE) != 0));
 156
 157         if (body->valid & OBD_MD_FLMODE)
 158                 st->st_mode = (st->st_mode & S_IFMT)|(body->mode & ~S_IFMT);
 159         if (body->valid & OBD_MD_FLTYPE)
 160                 st->st_mode = (st->st_mode & ~S_IFMT)|(body->mode & S_IFMT);
 161
 162         if (lsm != NULL) {
 163                 if (lli->lli_smd == NULL) {
 164                         cl_inode_init(inode, md);
 165                         lli->lli_smd = lsm;
 166                         lli->lli_maxbytes = lsm->lsm_maxbytes;
 167                         if (lli->lli_maxbytes > PAGE_CACHE_MAXBYTES)
 168                                 lli->lli_maxbytes = PAGE_CACHE_MAXBYTES;
 169                 } else {
 170                         if (lov_stripe_md_cmp(lli->lli_smd, lsm)) {
 171                                 CERROR("lsm mismatch for inode %lld\n",
 172                                        (long long)st->st_ino);
 173                                 LBUG();
 174                         }
 175                 }
 176         }
 177
 178         if (body->valid & OBD_MD_FLMTIME &&
 179             body->mtime > LTIME_S(st->st_mtime))
 180                 LTIME_S(st->st_mtime) = body->mtime;
 181         if (body->valid & OBD_MD_FLATIME &&
 182             body->atime > LTIME_S(st->st_atime))
 183                 LTIME_S(st->st_atime) = body->atime;
 184
 185         /* mtime is always updated with ctime, but can be set in past.
 186            As write and utime(2) may happen within 1 second, and utime's
 187            mtime has a priority over write's one, so take mtime from mds
 188            for the same ctimes. */
 189         if (body->valid & OBD_MD_FLCTIME &&
 190             body->ctime >= LTIME_S(st->st_ctime)) {
 191                 LTIME_S(st->st_ctime) = body->ctime;
 192                 if (body->valid & OBD_MD_FLMTIME)
 193                         LTIME_S(st->st_mtime) = body->mtime;
 194         }
 195        if (S_ISREG(st->st_mode))
 196                 st->st_blksize = min(2UL * PTLRPC_MAX_BRW_SIZE, LL_MAX_BLKSIZE);
 197         else
 198                 st->st_blksize = 4096;
 199         if (body->valid & OBD_MD_FLUID)
 200                 st->st_uid = body->uid;
 201         if (body->valid & OBD_MD_FLGID)
 202                 st->st_gid = body->gid;
 203         if (body->valid & OBD_MD_FLNLINK)
 204                 st->st_nlink = body->nlink;
 205         if (body->valid & OBD_MD_FLRDEV)
 206                 st->st_rdev = body->rdev;
 207         if (body->valid & OBD_MD_FLFLAGS)
 208                 lli->lli_st_flags = body->flags;
 209         if (body->valid & OBD_MD_FLSIZE) {
 210                 if ((llu_i2sbi(inode)->ll_lco.lco_flags & OBD_CONNECT_SOM) &&
 211                     S_ISREG(st->st_mode) && lli->lli_smd) {
 212                         struct lustre_handle lockh;
 213                         ldlm_mode_t mode;
 214
 215                         /* As it is possible a blocking ast has been processed
 216                          * by this time, we need to check there is an UPDATE
 217                          * lock on the client and set LLIF_MDS_SIZE_LOCK holding
 218                          * it. */
 219                         mode = llu_take_md_lock(inode, MDS_INODELOCK_UPDATE,
 220                                                 &lockh);
 221                         if (mode) {
 222                                 st->st_size = body->size;
 223                                 lli->lli_flags |= LLIF_MDS_SIZE_LOCK;
 224                                 ldlm_lock_decref(&lockh, mode);
 225                         }
 226                 } else {
 227                     st->st_size = body->size;
 228                 }
 229
 230                 if (body->valid & OBD_MD_FLBLOCKS)
 231                         st->st_blocks = body->blocks;
 232         }
 233 }
 234
 235 void obdo_to_inode(struct inode *dst, struct obdo *src, obd_flag valid)
 236 {
 237         struct llu_inode_info *lli = llu_i2info(dst);
 238         struct intnl_stat *st = llu_i2stat(dst);
 239
 240         valid &= src->o_valid;
 241
 242         if (valid & (OBD_MD_FLCTIME | OBD_MD_FLMTIME))
 243                 CDEBUG(D_INODE,"valid "LPX64", cur time "CFS_TIME_T"/"CFS_TIME_T
 244                        ", new %lu/%lu\n",
 245                        src->o_valid,
 246                        LTIME_S(st->st_mtime), LTIME_S(st->st_ctime),
 247                        (long)src->o_mtime, (long)src->o_ctime);
 248
 249         if (valid & OBD_MD_FLATIME)
 250                 LTIME_S(st->st_atime) = src->o_atime;
 251         if (valid & OBD_MD_FLMTIME)
 252                 LTIME_S(st->st_mtime) = src->o_mtime;
 253         if (valid & OBD_MD_FLCTIME && src->o_ctime > LTIME_S(st->st_ctime))
 254                 LTIME_S(st->st_ctime) = src->o_ctime;
 255         if (valid & OBD_MD_FLSIZE)
 256                 st->st_size = src->o_size;
 257         if (valid & OBD_MD_FLBLOCKS) /* allocation of space */
 258                 st->st_blocks = src->o_blocks;
 259         if (valid & OBD_MD_FLBLKSZ)
 260                 st->st_blksize = src->o_blksize;
 261         if (valid & OBD_MD_FLTYPE)
 262                 st->st_mode = (st->st_mode & ~S_IFMT) | (src->o_mode & S_IFMT);
 263         if (valid & OBD_MD_FLMODE)
 264                 st->st_mode = (st->st_mode & S_IFMT) | (src->o_mode & ~S_IFMT);
 265         if (valid & OBD_MD_FLUID)
 266                 st->st_uid = src->o_uid;
 267         if (valid & OBD_MD_FLGID)
 268                 st->st_gid = src->o_gid;
 269         if (valid & OBD_MD_FLFLAGS)
 270                 lli->lli_st_flags = src->o_flags;
 271 }
 272
 273 #define S_IRWXUGO       (S_IRWXU|S_IRWXG|S_IRWXO)
 274 #define S_IALLUGO       (S_ISUID|S_ISGID|S_ISVTX|S_IRWXUGO)
 275
 276 void obdo_from_inode(struct obdo *dst, struct inode *src, obd_flag valid)
 277 {
 278         struct llu_inode_info *lli = llu_i2info(src);
 279         struct intnl_stat *st = llu_i2stat(src);
 280         obd_flag newvalid = 0;
 281
 282         if (valid & (OBD_MD_FLCTIME | OBD_MD_FLMTIME))
 283                 CDEBUG(D_INODE, "valid %x, new time "CFS_TIME_T"/"CFS_TIME_T"\n",
 284                        valid, LTIME_S(st->st_mtime),
 285                        LTIME_S(st->st_ctime));
 286
 287         if (valid & OBD_MD_FLATIME) {
 288                 dst->o_atime = LTIME_S(st->st_atime);
 289                 newvalid |= OBD_MD_FLATIME;
 290         }
 291         if (valid & OBD_MD_FLMTIME) {
 292                 dst->o_mtime = LTIME_S(st->st_mtime);
 293                 newvalid |= OBD_MD_FLMTIME;
 294         }
 295         if (valid & OBD_MD_FLCTIME) {
 296                 dst->o_ctime = LTIME_S(st->st_ctime);
 297                 newvalid |= OBD_MD_FLCTIME;
 298         }
 299         if (valid & OBD_MD_FLSIZE) {
 300                 dst->o_size = st->st_size;
 301                 newvalid |= OBD_MD_FLSIZE;
 302         }
 303         if (valid & OBD_MD_FLBLOCKS) {  /* allocation of space (x512 bytes) */
 304                 dst->o_blocks = st->st_blocks;
 305                 newvalid |= OBD_MD_FLBLOCKS;
 306         }
 307         if (valid & OBD_MD_FLBLKSZ) {   /* optimal block size */
 308                 dst->o_blksize = st->st_blksize;
 309                 newvalid |= OBD_MD_FLBLKSZ;
 310         }
 311         if (valid & OBD_MD_FLTYPE) {
 312                 dst->o_mode = (dst->o_mode & S_IALLUGO)|(st->st_mode & S_IFMT);
 313                 newvalid |= OBD_MD_FLTYPE;
 314         }
 315         if (valid & OBD_MD_FLMODE) {
 316                 dst->o_mode = (dst->o_mode & S_IFMT)|(st->st_mode & S_IALLUGO);
 317                 newvalid |= OBD_MD_FLMODE;
 318         }
 319         if (valid & OBD_MD_FLUID) {
 320                 dst->o_uid = st->st_uid;
 321                 newvalid |= OBD_MD_FLUID;
 322         }
 323         if (valid & OBD_MD_FLGID) {
 324                 dst->o_gid = st->st_gid;
 325                 newvalid |= OBD_MD_FLGID;
 326         }
 327         if (valid & OBD_MD_FLFLAGS) {
 328                 dst->o_flags = lli->lli_st_flags;
 329                 newvalid |= OBD_MD_FLFLAGS;
 330         }
 331         if (valid & OBD_MD_FLGENER) {
 332                 dst->o_generation = lli->lli_st_generation;
 333                 newvalid |= OBD_MD_FLGENER;
 334         }
 335         if (valid & OBD_MD_FLFID) {
 336                 dst->o_fid = st->st_ino;
 337                 newvalid |= OBD_MD_FLFID;
 338         }
 339
 340         dst->o_valid |= newvalid;
 341 }
 342
 343 /*
 344  * really does the getattr on the inode and updates its fields
 345  */
 346 int llu_inode_getattr(struct inode *inode, struct obdo *obdo)
 347 {
 348         struct llu_inode_info *lli = llu_i2info(inode);
 349         struct ptlrpc_request_set *set;
 350         struct lov_stripe_md *lsm = lli->lli_smd;
 351         struct obd_info oinfo = { { { 0 } } };
 352         int rc;
 353         ENTRY;
 354
 355         LASSERT(lsm);
 356
 357         oinfo.oi_md = lsm;
 358         oinfo.oi_oa = obdo;
 359         oinfo.oi_oa->o_id = lsm->lsm_object_id;
 360         oinfo.oi_oa->o_gr = lsm->lsm_object_gr;
 361         oinfo.oi_oa->o_mode = S_IFREG;
 362         oinfo.oi_oa->o_valid = OBD_MD_FLID | OBD_MD_FLTYPE |
 363                                OBD_MD_FLSIZE | OBD_MD_FLBLOCKS |
 364                                OBD_MD_FLBLKSZ | OBD_MD_FLMTIME |
 365                                OBD_MD_FLCTIME | OBD_MD_FLGROUP;
 366
 367         set = ptlrpc_prep_set();
 368         if (set == NULL) {
 369                 CERROR ("ENOMEM allocing request set\n");
 370                 rc = -ENOMEM;
 371         } else {
 372                 rc = obd_getattr_async(llu_i2obdexp(inode), &oinfo, set);
 373                 if (rc == 0)
 374                         rc = ptlrpc_set_wait(set);
 375                 ptlrpc_set_destroy(set);
 376         }
 377         if (rc)
 378                 RETURN(rc);
 379
 380         oinfo.oi_oa->o_valid = OBD_MD_FLBLOCKS | OBD_MD_FLBLKSZ |
 381                                OBD_MD_FLMTIME | OBD_MD_FLCTIME |
 382                                OBD_MD_FLSIZE;
 383
 384         obdo_refresh_inode(inode, oinfo.oi_oa, oinfo.oi_oa->o_valid);
 385         CDEBUG(D_INODE, "objid "LPX64" size %Lu, blocks %Lu, "
 386                "blksize %Lu\n", lli->lli_smd->lsm_object_id,
 387                (long long unsigned)llu_i2stat(inode)->st_size,
 388                (long long unsigned)llu_i2stat(inode)->st_blocks,
 389                (long long unsigned)llu_i2stat(inode)->st_blksize);
 390         RETURN(0);
 391 }
 392
 393 static struct inode* llu_new_inode(struct filesys *fs,
 394                                    struct lu_fid *fid)
 395 {
 396         struct inode *inode;
 397         struct llu_inode_info *lli;
 398         struct intnl_stat st = {
 399                 .st_dev  = 0,
 400 #if 0
 401 #ifndef AUTOMOUNT_FILE_NAME
 402                 .st_mode = fid->f_type & S_IFMT,
 403 #else
 404                 .st_mode = fid->f_type /* all of the bits! */
 405 #endif
 406 #endif
 407                 /* FIXME: fix this later */
 408                 .st_mode = 0,
 409
 410                 .st_uid  = geteuid(),
 411                 .st_gid  = getegid(),
 412         };
 413
 414         OBD_ALLOC(lli, sizeof(*lli));
 415         if (!lli)
 416                 return NULL;
 417
 418         /* initialize lli here */
 419         lli->lli_sbi = llu_fs2sbi(fs);
 420         lli->lli_smd = NULL;
 421         lli->lli_symlink_name = NULL;
 422         lli->lli_flags = 0;
 423         lli->lli_maxbytes = (__u64)(~0UL);
 424         lli->lli_file_data = NULL;
 425
 426         lli->lli_sysio_fid.fid_data = &lli->lli_fid;
 427         lli->lli_sysio_fid.fid_len = sizeof(lli->lli_fid);
 428         lli->lli_fid = *fid;
 429
 430         /* file identifier is needed by functions like _sysio_i_find() */
 431         inode = _sysio_i_new(fs, &lli->lli_sysio_fid,
 432                              &st, 0, &llu_inode_ops, lli);
 433
 434         if (!inode)
 435                 OBD_FREE(lli, sizeof(*lli));
 436
 437         return inode;
 438 }
 439
 440 static int llu_have_md_lock(struct inode *inode, __u64 lockpart)
 441 {
 442         struct lustre_handle lockh;
 443         ldlm_policy_data_t policy = { .l_inodebits = { lockpart } };
 444         struct lu_fid *fid;
 445         int flags;
 446         ENTRY;
 447
 448         LASSERT(inode);
 449
 450         fid = &llu_i2info(inode)->lli_fid;
 451         CDEBUG(D_INFO, "trying to match res "DFID"\n", PFID(fid));
 452
 453         flags = LDLM_FL_BLOCK_GRANTED | LDLM_FL_CBPENDING | LDLM_FL_TEST_LOCK;
 454         if (md_lock_match(llu_i2mdexp(inode), flags, fid, LDLM_IBITS, &policy,
 455                           LCK_CR|LCK_CW|LCK_PR|LCK_PW, &lockh)) {
 456                 RETURN(1);
 457         }
 458         RETURN(0);
 459 }
 460
 461 static int llu_inode_revalidate(struct inode *inode)
 462 {
 463         struct lov_stripe_md *lsm = NULL;
 464         ENTRY;
 465
 466         if (!inode) {
 467                 CERROR("REPORT THIS LINE TO PETER\n");
 468                 RETURN(0);
 469         }
 470
 471         if (!llu_have_md_lock(inode, MDS_INODELOCK_UPDATE)) {
 472                 struct lustre_md md;
 473                 struct ptlrpc_request *req = NULL;
 474                 struct llu_sb_info *sbi = llu_i2sbi(inode);
 475                 unsigned long valid = OBD_MD_FLGETATTR;
 476                 int rc, ealen = 0;
 477
 478                 /* Why don't we update all valid MDS fields here, if we're
 479                  * doing an RPC anyways?  -phil */
 480                 if (S_ISREG(llu_i2stat(inode)->st_mode)) {
 481                         ealen = obd_size_diskmd(sbi->ll_dt_exp, NULL);
 482                         valid |= OBD_MD_FLEASIZE;
 483                 }
 484                 rc = md_getattr(sbi->ll_md_exp, ll_inode2fid(inode),
 485                                 NULL, valid, ealen, &req);
 486                 if (rc) {
 487                         CERROR("failure %d inode %llu\n", rc,
 488                                (long long)llu_i2stat(inode)->st_ino);
 489                         RETURN(-abs(rc));
 490                 }
 491                 rc = md_get_lustre_md(sbi->ll_md_exp, req,
 492                                       sbi->ll_dt_exp, sbi->ll_md_exp, &md);
 493
 494                 /* XXX Too paranoid? */
 495                 if (((md.body->valid ^ valid) & OBD_MD_FLEASIZE) &&
 496                     !((md.body->valid & OBD_MD_FLNLINK) &&
 497                       (md.body->nlink == 0))) {
 498                         CERROR("Asked for %s eadata but got %s (%d)\n",
 499                                (valid & OBD_MD_FLEASIZE) ? "some" : "no",
 500                                (md.body->valid & OBD_MD_FLEASIZE) ? "some":"none",
 501                                 md.body->eadatasize);
 502                 }
 503                 if (rc) {
 504                         ptlrpc_req_finished(req);
 505                         RETURN(rc);
 506                 }
 507
 508
 509                 llu_update_inode(inode, &md);
 510                 if (md.lsm != NULL && llu_i2info(inode)->lli_smd != md.lsm)
 511                         obd_free_memmd(sbi->ll_dt_exp, &md.lsm);
 512                 ptlrpc_req_finished(req);
 513         }
 514
 515         lsm = llu_i2info(inode)->lli_smd;
 516         if (!lsm)       /* object not yet allocated, don't validate size */
 517                 RETURN(0);
 518
 519         /* ll_glimpse_size will prefer locally cached writes if they extend
 520          * the file */
 521         RETURN(cl_glimpse_size(inode));
 522 }
 523
 524 static void copy_stat_buf(struct inode *ino, struct intnl_stat *b)
 525 {
 526         *b = *llu_i2stat(ino);
 527 }
 528
 529 static int llu_iop_getattr(struct pnode *pno,
 530                            struct inode *ino,
 531                            struct intnl_stat *b)
 532 {
 533         int rc;
 534         ENTRY;
 535
 536         liblustre_wait_event(0);
 537
 538         if (!ino) {
 539                 LASSERT(pno);
 540                 LASSERT(pno->p_base->pb_ino);
 541                 ino = pno->p_base->pb_ino;
 542         } else {
 543                 LASSERT(!pno || pno->p_base->pb_ino == ino);
 544         }
 545
 546         /* libsysio might call us directly without intent lock,
 547          * we must re-fetch the attrs here
 548          */
 549         rc = llu_inode_revalidate(ino);
 550         if (!rc) {
 551                 copy_stat_buf(ino, b);
 552                 LASSERT(!llu_i2info(ino)->lli_it);
 553         }
 554
 555         liblustre_wait_event(0);
 556         RETURN(rc);
 557 }
 558
 559 static int null_if_equal(struct ldlm_lock *lock, void *data)
 560 {
 561         if (data == lock->l_ast_data) {
 562                 lock->l_ast_data = NULL;
 563
 564                 if (lock->l_req_mode != lock->l_granted_mode)
 565                         LDLM_ERROR(lock,"clearing inode with ungranted lock\n");
 566         }
 567
 568         return LDLM_ITER_CONTINUE;
 569 }
 570
 571 void llu_clear_inode(struct inode *inode)
 572 {
 573         struct llu_inode_info *lli = llu_i2info(inode);
 574         struct llu_sb_info *sbi = llu_i2sbi(inode);
 575         ENTRY;
 576
 577         CDEBUG(D_VFSTRACE, "VFS Op:inode=%llu/%lu(%p)\n",
 578                (long long)llu_i2stat(inode)->st_ino, lli->lli_st_generation,
 579                inode);
 580
 581         lli->lli_flags &= ~LLIF_MDS_SIZE_LOCK;
 582         md_change_cbdata(sbi->ll_md_exp, ll_inode2fid(inode),
 583                          null_if_equal, inode);
 584
 585         if (lli->lli_smd)
 586                 obd_change_cbdata(sbi->ll_dt_exp, lli->lli_smd,
 587                                   null_if_equal, inode);
 588
 589         cl_inode_fini(inode);
 590
 591         if (lli->lli_smd) {
 592                 obd_free_memmd(sbi->ll_dt_exp, &lli->lli_smd);
 593                 lli->lli_smd = NULL;
 594         }
 595
 596         if (lli->lli_symlink_name) {
 597                 OBD_FREE(lli->lli_symlink_name,
 598                          strlen(lli->lli_symlink_name) + 1);
 599                 lli->lli_symlink_name = NULL;
 600         }
 601
 602         EXIT;
 603 }
 604
 605 void llu_iop_gone(struct inode *inode)
 606 {
 607         struct llu_inode_info *lli = llu_i2info(inode);
 608         ENTRY;
 609
 610         liblustre_wait_event(0);
 611         llu_clear_inode(inode);
 612
 613         OBD_FREE(lli, sizeof(*lli));
 614         EXIT;
 615 }
 616
 617 static int inode_setattr(struct inode * inode, struct iattr * attr)
 618 {
 619         unsigned int ia_valid = attr->ia_valid;
 620         struct intnl_stat *st = llu_i2stat(inode);
 621         int error = 0;
 622
 623         /*
 624          * inode_setattr() is only ever invoked with ATTR_SIZE (by
 625          * llu_setattr_raw()) when file has no bodies. Check this.
 626          */
 627         LASSERT(ergo(ia_valid & ATTR_SIZE, llu_i2info(inode)->lli_smd == NULL));
 628
 629         if (ia_valid & ATTR_SIZE)
 630                 st->st_size = attr->ia_size;
 631         if (ia_valid & ATTR_UID)
 632                 st->st_uid = attr->ia_uid;
 633         if (ia_valid & ATTR_GID)
 634                 st->st_gid = attr->ia_gid;
 635         if (ia_valid & ATTR_ATIME)
 636                 st->st_atime = attr->ia_atime;
 637         if (ia_valid & ATTR_MTIME)
 638                 st->st_mtime = attr->ia_mtime;
 639         if (ia_valid & ATTR_CTIME)
 640                 st->st_ctime = attr->ia_ctime;
 641         if (ia_valid & ATTR_MODE) {
 642                 st->st_mode = attr->ia_mode;
 643                 if (!in_group_p(st->st_gid) && !cfs_capable(CFS_CAP_FSETID))
 644                         st->st_mode &= ~S_ISGID;
 645         }
 646         /* mark_inode_dirty(inode); */
 647         return error;
 648 }
 649
 650 int llu_md_setattr(struct inode *inode, struct md_op_data *op_data,
 651                    struct md_open_data **mod)
 652 {
 653         struct lustre_md md;
 654         struct llu_sb_info *sbi = llu_i2sbi(inode);
 655         struct ptlrpc_request *request = NULL;
 656         int rc;
 657         ENTRY;
 658
 659         llu_prep_md_op_data(op_data, inode, NULL, NULL, 0, 0, LUSTRE_OPC_ANY);
 660         rc = md_setattr(sbi->ll_md_exp, op_data, NULL, 0, NULL,
 661                         0, &request, mod);
 662
 663         if (rc) {
 664                 ptlrpc_req_finished(request);
 665                 if (rc != -EPERM && rc != -EACCES)
 666                         CERROR("md_setattr fails: rc = %d\n", rc);
 667                 RETURN(rc);
 668         }
 669
 670         rc = md_get_lustre_md(sbi->ll_md_exp, request,
 671                               sbi->ll_dt_exp, sbi->ll_md_exp, &md);
 672         if (rc) {
 673                 ptlrpc_req_finished(request);
 674                 RETURN(rc);
 675         }
 676
 677         /* We call inode_setattr to adjust timestamps.
 678          * If there is at least some data in file, we cleared ATTR_SIZE
 679          * above to avoid invoking vmtruncate, otherwise it is important
 680          * to call vmtruncate in inode_setattr to update inode->i_size
 681          * (bug 6196) */
 682         inode_setattr(inode, &op_data->op_attr);
 683         llu_update_inode(inode, &md);
 684         ptlrpc_req_finished(request);
 685
 686         RETURN(rc);
 687 }
 688
 689 /* Close IO epoch and send Size-on-MDS attribute update. */
 690 static int llu_setattr_done_writing(struct inode *inode,
 691                                     struct md_op_data *op_data,
 692                                     struct md_open_data *mod)
 693 {
 694         struct llu_inode_info *lli = llu_i2info(inode);
 695         struct intnl_stat *st = llu_i2stat(inode);
 696         int rc = 0;
 697         ENTRY;
 698
 699         LASSERT(op_data != NULL);
 700         if (!S_ISREG(st->st_mode))
 701                 RETURN(0);
 702
 703         /* XXX: pass och here for the recovery purpose. */
 704         CDEBUG(D_INODE, "Epoch "LPU64" closed on "DFID" for truncate\n",
 705                op_data->op_ioepoch, PFID(&lli->lli_fid));
 706
 707         op_data->op_flags = MF_EPOCH_CLOSE | MF_SOM_CHANGE;
 708         rc = md_done_writing(llu_i2sbi(inode)->ll_md_exp, op_data, mod);
 709         if (rc == -EAGAIN) {
 710                 /* MDS has instructed us to obtain Size-on-MDS attribute
 711                  * from OSTs and send setattr to back to MDS. */
 712                 rc = llu_sizeonmds_update(inode, mod, &op_data->op_handle,
 713                                           op_data->op_ioepoch);
 714         } else if (rc) {
 715                 CERROR("inode %llu mdc truncate failed: rc = %d\n",
 716                        (unsigned long long)st->st_ino, rc);
 717         }
 718         RETURN(rc);
 719 }
 720
 721 /* If this inode has objects allocated to it (lsm != NULL), then the OST
 722  * object(s) determine the file size and mtime.  Otherwise, the MDS will
 723  * keep these values until such a time that objects are allocated for it.
 724  * We do the MDS operations first, as it is checking permissions for us.
 725  * We don't to the MDS RPC if there is nothing that we want to store there,
 726  * otherwise there is no harm in updating mtime/atime on the MDS if we are
 727  * going to do an RPC anyways.
 728  *
 729  * If we are doing a truncate, we will send the mtime and ctime updates
 730  * to the OST with the punch RPC, otherwise we do an explicit setattr RPC.
 731  * I don't believe it is possible to get e.g. ATTR_MTIME_SET and ATTR_SIZE
 732  * at the same time.
 733  */
 734 int llu_setattr_raw(struct inode *inode, struct iattr *attr)
 735 {
 736         struct lov_stripe_md *lsm = llu_i2info(inode)->lli_smd;
 737         struct intnl_stat *st = llu_i2stat(inode);
 738         int ia_valid = attr->ia_valid;
 739         struct md_op_data op_data = { { 0 } };
 740         struct md_open_data *mod = NULL;
 741         int rc = 0, rc1 = 0;
 742         ENTRY;
 743
 744         CDEBUG(D_VFSTRACE, "VFS Op:inode=%llu\n", (long long)st->st_ino);
 745
 746         if (ia_valid & ATTR_SIZE) {
 747                 if (attr->ia_size > ll_file_maxbytes(inode)) {
 748                         CDEBUG(D_INODE, "file too large %llu > "LPU64"\n",
 749                                (long long)attr->ia_size,
 750                                ll_file_maxbytes(inode));
 751                         RETURN(-EFBIG);
 752                 }
 753
 754                 attr->ia_valid |= ATTR_MTIME | ATTR_CTIME;
 755         }
 756
 757         /* We mark all of the fields "set" so MDS/OST does not re-set them */
 758         if (attr->ia_valid & ATTR_CTIME) {
 759                 attr->ia_ctime = CURRENT_TIME;
 760                 attr->ia_valid |= ATTR_CTIME_SET;
 761         }
 762         if (!(ia_valid & ATTR_ATIME_SET) && (attr->ia_valid & ATTR_ATIME)) {
 763                 attr->ia_atime = CURRENT_TIME;
 764                 attr->ia_valid |= ATTR_ATIME_SET;
 765         }
 766         if (!(ia_valid & ATTR_MTIME_SET) && (attr->ia_valid & ATTR_MTIME)) {
 767                 attr->ia_mtime = CURRENT_TIME;
 768                 attr->ia_valid |= ATTR_MTIME_SET;
 769         }
 770         if ((attr->ia_valid & ATTR_CTIME) && !(attr->ia_valid & ATTR_MTIME)) {
 771                 /* To avoid stale mtime on mds, obtain it from ost and send
 772                    to mds. */
 773                 rc = cl_glimpse_size(inode);
 774                 if (rc)
 775                         RETURN(rc);
 776
 777                 attr->ia_valid |= ATTR_MTIME_SET | ATTR_MTIME;
 778                 attr->ia_mtime = inode->i_stbuf.st_mtime;
 779         }
 780
 781         if (attr->ia_valid & (ATTR_MTIME | ATTR_CTIME))
 782                 CDEBUG(D_INODE, "setting mtime "CFS_TIME_T", ctime "CFS_TIME_T
 783                        ", now = "CFS_TIME_T"\n",
 784                        LTIME_S(attr->ia_mtime), LTIME_S(attr->ia_ctime),
 785                        LTIME_S(CURRENT_TIME));
 786
 787         /* NB: ATTR_SIZE will only be set after this point if the size
 788          * resides on the MDS, ie, this file has no objects. */
 789         if (lsm)
 790                 attr->ia_valid &= ~ATTR_SIZE;
 791
 792         /* If only OST attributes being set on objects, don't do MDS RPC.
 793          * In that case, we need to check permissions and update the local
 794          * inode ourselves so we can call obdo_from_inode() always. */
 795         if (ia_valid & (lsm ? ~(ATTR_FROM_OPEN | ATTR_RAW) : ~0)) {
 796                 memcpy(&op_data.op_attr, attr, sizeof(*attr));
 797
 798                 /* Open epoch for truncate. */
 799                 if (ia_valid & ATTR_SIZE)
 800                         op_data.op_flags = MF_EPOCH_OPEN;
 801                 rc = llu_md_setattr(inode, &op_data, &mod);
 802                 if (rc)
 803                         RETURN(rc);
 804
 805                 if (op_data.op_ioepoch)
 806                         CDEBUG(D_INODE, "Epoch "LPU64" opened on "DFID" for "
 807                                "truncate\n", op_data.op_ioepoch,
 808                                PFID(&llu_i2info(inode)->lli_fid));
 809
 810                 if (!lsm || !S_ISREG(st->st_mode)) {
 811                         CDEBUG(D_INODE, "no lsm: not setting attrs on OST\n");
 812                         GOTO(out, rc);
 813                 }
 814         } else {
 815                 /* The OST doesn't check permissions, but the alternative is
 816                  * a gratuitous RPC to the MDS.  We already rely on the client
 817                  * to do read/write/truncate permission checks, so is mtime OK?
 818                  */
 819                 if (ia_valid & (ATTR_MTIME | ATTR_ATIME)) {
 820                         /* from sys_utime() */
 821                         if (!(ia_valid & (ATTR_MTIME_SET | ATTR_ATIME_SET))) {
 822                                 if (current->fsuid != st->st_uid &&
 823                                     (rc = ll_permission(inode, MAY_WRITE)) != 0)
 824                                         RETURN(rc);
 825                         } else {
 826                                 /* from inode_change_ok() */
 827                                 if (current->fsuid != st->st_uid &&
 828                                     !cfs_capable(CFS_CAP_FOWNER))
 829                                         RETURN(-EPERM);
 830                         }
 831                 }
 832
 833
 834                 /* Won't invoke llu_vmtruncate(), as we already cleared
 835                  * ATTR_SIZE */
 836                 inode_setattr(inode, attr);
 837         }
 838
 839         if (ia_valid & ATTR_SIZE) {
 840                 rc = cl_setattr_do_truncate(inode, attr->ia_size, NULL);
 841         } else if (ia_valid & (ATTR_MTIME | ATTR_MTIME_SET)) {
 842                 CDEBUG(D_INODE, "set mtime on OST inode %llu to %lu\n",
 843                        (long long unsigned)st->st_ino, LTIME_S(attr->ia_mtime));
 844                 rc = cl_setattr_ost(inode, NULL);
 845         }
 846         EXIT;
 847 out:
 848         if (op_data.op_ioepoch)
 849                 rc1 = llu_setattr_done_writing(inode, &op_data, mod);
 850         return rc ? rc : rc1;
 851 }
 852
 853 /* here we simply act as a thin layer to glue it with
 854  * llu_setattr_raw(), which is copy from kernel
 855  */
 856 static int llu_iop_setattr(struct pnode *pno,
 857                            struct inode *ino,
 858                            unsigned mask,
 859                            struct intnl_stat *stbuf)
 860 {
 861         struct iattr iattr;
 862         int rc;
 863         ENTRY;
 864
 865         liblustre_wait_event(0);
 866
 867         LASSERT(!(mask & ~(SETATTR_MTIME | SETATTR_ATIME |
 868                            SETATTR_UID | SETATTR_GID |
 869                            SETATTR_LEN | SETATTR_MODE)));
 870         memset(&iattr, 0, sizeof(iattr));
 871
 872         if (mask & SETATTR_MODE) {
 873                 iattr.ia_mode = stbuf->st_mode;
 874                 iattr.ia_valid |= ATTR_MODE;
 875         }
 876         if (mask & SETATTR_MTIME) {
 877                 iattr.ia_mtime = stbuf->st_mtime;
 878                 iattr.ia_valid |= ATTR_MTIME | ATTR_MTIME_SET;
 879         }
 880         if (mask & SETATTR_ATIME) {
 881                 iattr.ia_atime = stbuf->st_atime;
 882                 iattr.ia_valid |= ATTR_ATIME | ATTR_ATIME_SET;
 883         }
 884         if (mask & SETATTR_UID) {
 885                 iattr.ia_uid = stbuf->st_uid;
 886                 iattr.ia_valid |= ATTR_UID;
 887         }
 888         if (mask & SETATTR_GID) {
 889                 iattr.ia_gid = stbuf->st_gid;
 890                 iattr.ia_valid |= ATTR_GID;
 891         }
 892         if (mask & SETATTR_LEN) {
 893                 iattr.ia_size = stbuf->st_size; /* XXX signed expansion problem */
 894                 iattr.ia_valid |= ATTR_SIZE;
 895         }
 896
 897         iattr.ia_valid |= ATTR_RAW | ATTR_CTIME;
 898         iattr.ia_ctime = CURRENT_TIME;
 899
 900         rc = llu_setattr_raw(ino, &iattr);
 901         liblustre_wait_idle();
 902         RETURN(rc);
 903 }
 904
 905 #define EXT2_LINK_MAX           32000
 906
 907 static int llu_iop_symlink_raw(struct pnode *pno, const char *tgt)
 908 {
 909         struct inode *dir = pno->p_base->pb_parent->pb_ino;
 910         struct qstr *qstr = &pno->p_base->pb_name;
 911         const char *name = qstr->name;
 912         int len = qstr->len;
 913         struct ptlrpc_request *request = NULL;
 914         struct llu_sb_info *sbi = llu_i2sbi(dir);
 915         struct md_op_data op_data = {{ 0 }};
 916         int err = -EMLINK;
 917         ENTRY;
 918
 919         liblustre_wait_event(0);
 920         if (llu_i2stat(dir)->st_nlink >= EXT2_LINK_MAX)
 921                 RETURN(err);
 922
 923         llu_prep_md_op_data(&op_data, dir, NULL, name, len, 0,
 924                             LUSTRE_OPC_SYMLINK);
 925
 926         err = md_create(sbi->ll_md_exp, &op_data, tgt, strlen(tgt) + 1,
 927                         S_IFLNK | S_IRWXUGO, current->fsuid, current->fsgid,
 928                         cfs_curproc_cap_pack(), 0, &request);
 929         ptlrpc_req_finished(request);
 930         liblustre_wait_event(0);
 931         RETURN(err);
 932 }
 933
 934 static int llu_readlink_internal(struct inode *inode,
 935                                  struct ptlrpc_request **request,
 936                                  char **symname)
 937 {
 938         struct llu_inode_info *lli = llu_i2info(inode);
 939         struct llu_sb_info *sbi = llu_i2sbi(inode);
 940         struct mdt_body *body;
 941         struct intnl_stat *st = llu_i2stat(inode);
 942         int rc, symlen = st->st_size + 1;
 943         ENTRY;
 944
 945         *request = NULL;
 946         *symname = NULL;
 947
 948         if (lli->lli_symlink_name) {
 949                 *symname = lli->lli_symlink_name;
 950                 CDEBUG(D_INODE, "using cached symlink %s\n", *symname);
 951                 RETURN(0);
 952         }
 953
 954         rc = md_getattr(sbi->ll_md_exp, ll_inode2fid(inode), NULL,
 955                         OBD_MD_LINKNAME, symlen, request);
 956         if (rc) {
 957                 CERROR("inode %llu: rc = %d\n", (long long)st->st_ino, rc);
 958                 RETURN(rc);
 959         }
 960
 961         body = req_capsule_server_get(&(*request)->rq_pill, &RMF_MDT_BODY);
 962         LASSERT(body != NULL);
 963
 964         if ((body->valid & OBD_MD_LINKNAME) == 0) {
 965                 CERROR ("OBD_MD_LINKNAME not set on reply\n");
 966                 GOTO (failed, rc = -EPROTO);
 967         }
 968
 969         LASSERT(symlen != 0);
 970         if (body->eadatasize != symlen) {
 971                 CERROR("inode %llu: symlink length %d not expected %d\n",
 972                        (long long)st->st_ino, body->eadatasize - 1, symlen - 1);
 973                 GOTO(failed, rc = -EPROTO);
 974         }
 975
 976         *symname = req_capsule_server_get(&(*request)->rq_pill, &RMF_MDT_MD);
 977         if (*symname == NULL ||
 978             strnlen(*symname, symlen) != symlen - 1) {
 979                 /* not full/NULL terminated */
 980                 CERROR("inode %llu: symlink not NULL terminated string"
 981                        "of length %d\n", (long long)st->st_ino, symlen - 1);
 982                 GOTO(failed, rc = -EPROTO);
 983         }
 984
 985         OBD_ALLOC(lli->lli_symlink_name, symlen);
 986         /* do not return an error if we cannot cache the symlink locally */
 987         if (lli->lli_symlink_name)
 988                 memcpy(lli->lli_symlink_name, *symname, symlen);
 989
 990         RETURN(0);
 991
 992  failed:
 993         ptlrpc_req_finished (*request);
 994         RETURN (-EPROTO);
 995 }
 996
 997 static int llu_iop_readlink(struct pnode *pno, char *data, size_t bufsize)
 998 {
 999         struct inode *inode = pno->p_base->pb_ino;
1000         struct ptlrpc_request *request;
1001         char *symname;
1002         int rc;
1003         ENTRY;
1004
1005         liblustre_wait_event(0);
1006         rc = llu_readlink_internal(inode, &request, &symname);
1007         if (rc)
1008                 GOTO(out, rc);
1009
1010         LASSERT(symname);
1011         strncpy(data, symname, bufsize);
1012         rc = strlen(symname);
1013
1014         ptlrpc_req_finished(request);
1015  out:
1016         liblustre_wait_event(0);
1017         RETURN(rc);
1018 }
1019
1020 static int llu_iop_mknod_raw(struct pnode *pno,
1021                              mode_t mode,
1022                              dev_t dev)
1023 {
1024         struct ptlrpc_request *request = NULL;
1025         struct inode *dir = pno->p_parent->p_base->pb_ino;
1026         struct llu_sb_info *sbi = llu_i2sbi(dir);
1027         struct md_op_data op_data = {{ 0 }};
1028         int err = -EMLINK;
1029         ENTRY;
1030
1031         liblustre_wait_event(0);
1032         CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s,dir=%llu\n",
1033                (int)pno->p_base->pb_name.len, pno->p_base->pb_name.name,
1034                (long long)llu_i2stat(dir)->st_ino);
1035
1036         if (llu_i2stat(dir)->st_nlink >= EXT2_LINK_MAX)
1037                 RETURN(err);
1038
1039         switch (mode & S_IFMT) {
1040         case 0:
1041         case S_IFREG:
1042                 mode |= S_IFREG; /* for mode = 0 case, fallthrough */
1043         case S_IFCHR:
1044         case S_IFBLK:
1045         case S_IFIFO:
1046         case S_IFSOCK:
1047                 llu_prep_md_op_data(&op_data, dir, NULL,
1048                                     pno->p_base->pb_name.name,
1049                                     pno->p_base->pb_name.len, 0,
1050                                     LUSTRE_OPC_MKNOD);
1051
1052                 err = md_create(sbi->ll_md_exp, &op_data, NULL, 0, mode,
1053                                 current->fsuid, current->fsgid,
1054                                 cfs_curproc_cap_pack(), dev, &request);
1055                 ptlrpc_req_finished(request);
1056                 break;
1057         case S_IFDIR:
1058                 err = -EPERM;
1059                 break;
1060         default:
1061                 err = -EINVAL;
1062         }
1063         liblustre_wait_event(0);
1064         RETURN(err);
1065 }
1066
1067 static int llu_iop_link_raw(struct pnode *old, struct pnode *new)
1068 {
1069         struct inode *src = old->p_base->pb_ino;
1070         struct inode *dir = new->p_parent->p_base->pb_ino;
1071         const char *name = new->p_base->pb_name.name;
1072         int namelen = new->p_base->pb_name.len;
1073         struct ptlrpc_request *request = NULL;
1074         struct md_op_data op_data = {{ 0 }};
1075         int rc;
1076         ENTRY;
1077
1078         LASSERT(src);
1079         LASSERT(dir);
1080
1081         liblustre_wait_event(0);
1082         llu_prep_md_op_data(&op_data, src, dir, name, namelen, 0,
1083                             LUSTRE_OPC_ANY);
1084         rc = md_link(llu_i2sbi(src)->ll_md_exp, &op_data, &request);
1085         ptlrpc_req_finished(request);
1086         liblustre_wait_event(0);
1087
1088         RETURN(rc);
1089 }
1090
1091 /*
1092  * libsysio will clear the inode immediately after return
1093  */
1094 static int llu_iop_unlink_raw(struct pnode *pno)
1095 {
1096         struct inode *dir = pno->p_base->pb_parent->pb_ino;
1097         struct qstr *qstr = &pno->p_base->pb_name;
1098         const char *name = qstr->name;
1099         int len = qstr->len;
1100         struct inode *target = pno->p_base->pb_ino;
1101         struct ptlrpc_request *request = NULL;
1102         struct md_op_data op_data = { { 0 } };
1103         int rc;
1104         ENTRY;
1105
1106         LASSERT(target);
1107
1108         liblustre_wait_event(0);
1109         llu_prep_md_op_data(&op_data, dir, NULL, name, len, 0,
1110                             LUSTRE_OPC_ANY);
1111         rc = md_unlink(llu_i2sbi(dir)->ll_md_exp, &op_data, &request);
1112         if (!rc)
1113                 rc = llu_objects_destroy(request, dir);
1114         ptlrpc_req_finished(request);
1115         liblustre_wait_idle();
1116
1117         RETURN(rc);
1118 }
1119
1120 static int llu_iop_rename_raw(struct pnode *old, struct pnode *new)
1121 {
1122         struct inode *src = old->p_parent->p_base->pb_ino;
1123         struct inode *tgt = new->p_parent->p_base->pb_ino;
1124         const char *oldname = old->p_base->pb_name.name;
1125         int oldnamelen = old->p_base->pb_name.len;
1126         const char *newname = new->p_base->pb_name.name;
1127         int newnamelen = new->p_base->pb_name.len;
1128         struct ptlrpc_request *request = NULL;
1129         struct md_op_data op_data = { { 0 } };
1130         int rc;
1131         ENTRY;
1132
1133         LASSERT(src);
1134         LASSERT(tgt);
1135
1136         liblustre_wait_event(0);
1137         llu_prep_md_op_data(&op_data, src, tgt, NULL, 0, 0,
1138                             LUSTRE_OPC_ANY);
1139         rc = md_rename(llu_i2sbi(src)->ll_md_exp, &op_data,
1140                        oldname, oldnamelen, newname, newnamelen,
1141                        &request);
1142         if (!rc) {
1143                 rc = llu_objects_destroy(request, src);
1144         }
1145
1146         ptlrpc_req_finished(request);
1147         liblustre_wait_idle();
1148
1149         RETURN(rc);
1150 }
1151
1152 #ifdef _HAVE_STATVFS
1153 static int llu_statfs_internal(struct llu_sb_info *sbi,
1154                                struct obd_statfs *osfs, __u64 max_age)
1155 {
1156         struct obd_statfs obd_osfs;
1157         int rc;
1158         ENTRY;
1159
1160         rc = obd_statfs(class_exp2obd(sbi->ll_md_exp), osfs, max_age, 0);
1161         if (rc) {
1162                 CERROR("md_statfs fails: rc = %d\n", rc);
1163                 RETURN(rc);
1164         }
1165
1166         CDEBUG(D_SUPER, "MDC blocks "LPU64"/"LPU64" objects "LPU64"/"LPU64"\n",
1167                osfs->os_bavail, osfs->os_blocks, osfs->os_ffree,osfs->os_files);
1168
1169         rc = obd_statfs_rqset(class_exp2obd(sbi->ll_dt_exp),
1170                               &obd_statfs, max_age, 0);
1171         if (rc) {
1172                 CERROR("obd_statfs fails: rc = %d\n", rc);
1173                 RETURN(rc);
1174         }
1175
1176         CDEBUG(D_SUPER, "OSC blocks "LPU64"/"LPU64" objects "LPU64"/"LPU64"\n",
1177                obd_osfs.os_bavail, obd_osfs.os_blocks, obd_osfs.os_ffree,
1178                obd_osfs.os_files);
1179
1180         osfs->os_blocks = obd_osfs.os_blocks;
1181         osfs->os_bfree = obd_osfs.os_bfree;
1182         osfs->os_bavail = obd_osfs.os_bavail;
1183
1184         /* If we don't have as many objects free on the OST as inodes
1185          * on the MDS, we reduce the total number of inodes to
1186          * compensate, so that the "inodes in use" number is correct.
1187          */
1188         if (obd_osfs.os_ffree < osfs->os_ffree) {
1189                 osfs->os_files = (osfs->os_files - osfs->os_ffree) +
1190                         obd_osfs.os_ffree;
1191                 osfs->os_ffree = obd_osfs.os_ffree;
1192         }
1193
1194         RETURN(rc);
1195 }
1196
1197 static int llu_statfs(struct llu_sb_info *sbi, struct statfs *sfs)
1198 {
1199         struct obd_statfs osfs;
1200         int rc;
1201
1202         CDEBUG(D_VFSTRACE, "VFS Op:\n");
1203
1204         /* For now we will always get up-to-date statfs values, but in the
1205          * future we may allow some amount of caching on the client (e.g.
1206          * from QOS or lprocfs updates). */
1207         rc = llu_statfs_internal(sbi, &osfs, cfs_time_current_64() - HZ);
1208         if (rc)
1209                 return rc;
1210
1211         statfs_unpack(sfs, &osfs);
1212
1213         if (sizeof(sfs->f_blocks) == 4) {
1214                 while (osfs.os_blocks > ~0UL) {
1215                         sfs->f_bsize <<= 1;
1216
1217                         osfs.os_blocks >>= 1;
1218                         osfs.os_bfree >>= 1;
1219                         osfs.os_bavail >>= 1;
1220                 }
1221         }
1222
1223         sfs->f_blocks = osfs.os_blocks;
1224         sfs->f_bfree = osfs.os_bfree;
1225         sfs->f_bavail = osfs.os_bavail;
1226
1227         return 0;
1228 }
1229
1230 static int llu_iop_statvfs(struct pnode *pno,
1231                            struct inode *ino,
1232                            struct intnl_statvfs *buf)
1233 {
1234         struct statfs fs;
1235         int rc;
1236         ENTRY;
1237
1238         liblustre_wait_event(0);
1239
1240 #ifndef __CYGWIN__
1241         LASSERT(pno->p_base->pb_ino);
1242         rc = llu_statfs(llu_i2sbi(pno->p_base->pb_ino), &fs);
1243         if (rc)
1244                 RETURN(rc);
1245
1246         /* from native driver */
1247         buf->f_bsize = fs.f_bsize;  /* file system block size */
1248         buf->f_frsize = fs.f_bsize; /* file system fundamental block size */
1249         buf->f_blocks = fs.f_blocks;
1250         buf->f_bfree = fs.f_bfree;
1251         buf->f_bavail = fs.f_bavail;
1252         buf->f_files = fs.f_files;  /* Total number serial numbers */
1253         buf->f_ffree = fs.f_ffree;  /* Number free serial numbers */
1254         buf->f_favail = fs.f_ffree; /* Number free ser num for non-privileged*/
1255         buf->f_fsid = fs.f_fsid.__val[1];
1256         buf->f_flag = 0;            /* No equiv in statfs; maybe use type? */
1257         buf->f_namemax = fs.f_namelen;
1258 #endif
1259
1260         liblustre_wait_event(0);
1261         RETURN(0);
1262 }
1263 #endif /* _HAVE_STATVFS */
1264
1265 static int llu_iop_mkdir_raw(struct pnode *pno, mode_t mode)
1266 {
1267         struct inode *dir = pno->p_base->pb_parent->pb_ino;
1268         struct qstr *qstr = &pno->p_base->pb_name;
1269         const char *name = qstr->name;
1270         int len = qstr->len;
1271         struct ptlrpc_request *request = NULL;
1272         struct intnl_stat *st = llu_i2stat(dir);
1273         struct md_op_data op_data = {{ 0 }};
1274         int err = -EMLINK;
1275         ENTRY;
1276
1277         liblustre_wait_event(0);
1278         CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s,dir=%llu/%lu(%p)\n", len, name,
1279                (long long)st->st_ino, llu_i2info(dir)->lli_st_generation, dir);
1280
1281         if (st->st_nlink >= EXT2_LINK_MAX)
1282                 RETURN(err);
1283
1284         llu_prep_md_op_data(&op_data, dir, NULL, name, len, 0,
1285                             LUSTRE_OPC_MKDIR);
1286
1287         err = md_create(llu_i2sbi(dir)->ll_md_exp, &op_data, NULL, 0,
1288                         mode | S_IFDIR, current->fsuid, current->fsgid,
1289                         cfs_curproc_cap_pack(), 0, &request);
1290         ptlrpc_req_finished(request);
1291         liblustre_wait_event(0);
1292         RETURN(err);
1293 }
1294
1295 static int llu_iop_rmdir_raw(struct pnode *pno)
1296 {
1297         struct inode *dir = pno->p_base->pb_parent->pb_ino;
1298         struct qstr *qstr = &pno->p_base->pb_name;
1299         const char *name = qstr->name;
1300         int len = qstr->len;
1301         struct ptlrpc_request *request = NULL;
1302         struct md_op_data op_data = {{ 0 }};
1303         int rc;
1304         ENTRY;
1305
1306         liblustre_wait_event(0);
1307         CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s,dir=%llu/%lu(%p)\n", len, name,
1308                (long long)llu_i2stat(dir)->st_ino,
1309                llu_i2info(dir)->lli_st_generation, dir);
1310
1311         llu_prep_md_op_data(&op_data, dir, NULL, name, len, S_IFDIR,
1312                             LUSTRE_OPC_ANY);
1313         rc = md_unlink(llu_i2sbi(dir)->ll_md_exp, &op_data, &request);
1314         ptlrpc_req_finished(request);
1315
1316         liblustre_wait_event(0);
1317         RETURN(rc);
1318 }
1319
1320 #ifdef O_DIRECT
1321 #define FCNTL_FLMASK (O_APPEND|O_NONBLOCK|O_ASYNC|O_DIRECT)
1322 #else
1323 #define FCNTL_FLMASK (O_APPEND|O_NONBLOCK|O_ASYNC)
1324 #endif
1325 #define FCNTL_FLMASK_INVALID (O_NONBLOCK|O_ASYNC)
1326
1327 /* refer to ll_file_flock() for details */
1328 static int llu_file_flock(struct inode *ino,
1329                           int cmd,
1330                           struct file_lock *file_lock)
1331 {
1332         struct llu_inode_info *lli = llu_i2info(ino);
1333         struct intnl_stat *st = llu_i2stat(ino);
1334         struct ldlm_res_id res_id =
1335                 { .name = {fid_seq(&lli->lli_fid),
1336                            fid_oid(&lli->lli_fid),
1337                            fid_ver(&lli->lli_fid),
1338                            LDLM_FLOCK} };
1339         struct ldlm_enqueue_info einfo = { LDLM_FLOCK, 0, NULL,
1340                 ldlm_flock_completion_ast, NULL, NULL, file_lock };
1341
1342         struct lustre_handle lockh = {0};
1343         ldlm_policy_data_t flock;
1344         int flags = 0;
1345         int rc;
1346
1347         CDEBUG(D_VFSTRACE, "VFS Op:inode=%llu file_lock=%p\n",
1348                (unsigned long long)st->st_ino, file_lock);
1349
1350         flock.l_flock.pid = file_lock->fl_pid;
1351         flock.l_flock.start = file_lock->fl_start;
1352         flock.l_flock.end = file_lock->fl_end;
1353
1354         switch (file_lock->fl_type) {
1355         case F_RDLCK:
1356                 einfo.ei_mode = LCK_PR;
1357                 break;
1358         case F_UNLCK:
1359                 einfo.ei_mode = LCK_NL;
1360                 break;
1361         case F_WRLCK:
1362                 einfo.ei_mode = LCK_PW;
1363                 break;
1364         default:
1365                 CERROR("unknown fcntl lock type: %d\n", file_lock->fl_type);
1366                 LBUG();
1367         }
1368
1369         switch (cmd) {
1370         case F_SETLKW:
1371 #ifdef F_SETLKW64
1372 #if F_SETLKW64 != F_SETLKW
1373         case F_SETLKW64:
1374 #endif
1375 #endif
1376                 flags = 0;
1377                 break;
1378         case F_SETLK:
1379 #ifdef F_SETLK64
1380 #if F_SETLK64 != F_SETLK
1381         case F_SETLK64:
1382 #endif
1383 #endif
1384                 flags = LDLM_FL_BLOCK_NOWAIT;
1385                 break;
1386         case F_GETLK:
1387 #ifdef F_GETLK64
1388 #if F_GETLK64 != F_GETLK
1389         case F_GETLK64:
1390 #endif
1391 #endif
1392                 flags = LDLM_FL_TEST_LOCK;
1393                 file_lock->fl_type = einfo.ei_mode;
1394                 break;
1395         default:
1396                 CERROR("unknown fcntl cmd: %d\n", cmd);
1397                 LBUG();
1398         }
1399
1400         CDEBUG(D_DLMTRACE, "inode=%llu, pid=%u, cmd=%d, flags=%#x, mode=%u, "
1401                "start="LPX64", end="LPX64"\n", (unsigned long long)st->st_ino,
1402                flock.l_flock.pid, cmd, flags, einfo.ei_mode, flock.l_flock.start,
1403                flock.l_flock.end);
1404
1405         {
1406                 struct lmv_obd *lmv;
1407                 struct obd_device *lmv_obd;
1408                 lmv_obd = class_exp2obd(llu_i2mdexp(ino));
1409                 lmv = &lmv_obd->u.lmv;
1410
1411                 if (lmv->desc.ld_tgt_count < 1)
1412                         RETURN(rc = -ENODEV);
1413
1414                 if (lmv->tgts[0].ltd_exp != NULL)
1415                         rc = ldlm_cli_enqueue(lmv->tgts[0].ltd_exp, NULL, &einfo, &res_id,
1416                                               &flock, &flags, NULL, 0, NULL, &lockh, 0);
1417                 else
1418                         rc = -ENODEV;
1419         }
1420         RETURN(rc);
1421 }
1422
1423 static int assign_type(struct file_lock *fl, int type)
1424 {
1425         switch (type) {
1426         case F_RDLCK:
1427         case F_WRLCK:
1428         case F_UNLCK:
1429                 fl->fl_type = type;
1430                 return 0;
1431         default:
1432                 return -EINVAL;
1433         }
1434 }
1435
1436 static int flock_to_posix_lock(struct inode *ino,
1437                                struct file_lock *fl,
1438                                struct flock *l)
1439 {
1440         switch (l->l_whence) {
1441         /* XXX: only SEEK_SET is supported in lustre */
1442         case SEEK_SET:
1443                 fl->fl_start = 0;
1444                 break;
1445         default:
1446                 return -EINVAL;
1447         }
1448
1449         fl->fl_end = l->l_len - 1;
1450         if (l->l_len < 0)
1451                 return -EINVAL;
1452         if (l->l_len == 0)
1453                 fl->fl_end = OFFSET_MAX;
1454
1455         fl->fl_pid = getpid();
1456         fl->fl_flags = FL_POSIX;
1457         fl->fl_notify = NULL;
1458         fl->fl_insert = NULL;
1459         fl->fl_remove = NULL;
1460         /* XXX: these fields can't be filled with suitable values,
1461                 but I think lustre doesn't use them.
1462          */
1463         fl->fl_owner = NULL;
1464         fl->fl_file = NULL;
1465
1466         return assign_type(fl, l->l_type);
1467 }
1468
1469 static int llu_fcntl_getlk(struct inode *ino, struct flock *flock)
1470 {
1471         struct file_lock fl;
1472         int error;
1473
1474         error = EINVAL;
1475         if ((flock->l_type != F_RDLCK) && (flock->l_type != F_WRLCK))
1476                 goto out;
1477
1478         error = flock_to_posix_lock(ino, &fl, flock);
1479         if (error)
1480                 goto out;
1481
1482         error = llu_file_flock(ino, F_GETLK, &fl);
1483         if (error)
1484                 goto out;
1485
1486         flock->l_type = F_UNLCK;
1487         if (fl.fl_type != F_UNLCK) {
1488                 flock->l_pid = fl.fl_pid;
1489                 flock->l_start = fl.fl_start;
1490                 flock->l_len = fl.fl_end == OFFSET_MAX ? 0:
1491                         fl.fl_end - fl.fl_start + 1;
1492                 flock->l_whence = SEEK_SET;
1493                 flock->l_type = fl.fl_type;
1494         }
1495
1496 out:
1497         return error;
1498 }
1499
1500 static int llu_fcntl_setlk(struct inode *ino, int cmd, struct flock *flock)
1501 {
1502         struct file_lock fl;
1503         int flags = llu_i2info(ino)->lli_open_flags + 1;
1504         int error;
1505
1506         error = flock_to_posix_lock(ino, &fl, flock);
1507         if (error)
1508                 goto out;
1509         if (cmd == F_SETLKW)
1510                 fl.fl_flags |= FL_SLEEP;
1511
1512         error = -EBADF;
1513         switch (flock->l_type) {
1514         case F_RDLCK:
1515                 if (!(flags & FMODE_READ))
1516                         goto out;
1517                 break;
1518         case F_WRLCK:
1519                 if (!(flags & FMODE_WRITE))
1520                         goto out;
1521                 break;
1522         case F_UNLCK:
1523                 break;
1524         default:
1525                 error = -EINVAL;
1526                 goto out;
1527         }
1528
1529         error = llu_file_flock(ino, cmd, &fl);
1530         if (error)
1531                 goto out;
1532
1533 out:
1534         return error;
1535 }
1536
1537 static int llu_iop_fcntl(struct inode *ino, int cmd, va_list ap, int *rtn)
1538 {
1539         struct llu_inode_info *lli = llu_i2info(ino);
1540         long flags;
1541         struct flock *flock;
1542         long err = 0;
1543
1544         liblustre_wait_event(0);
1545         switch (cmd) {
1546         case F_GETFL:
1547                 *rtn = lli->lli_open_flags;
1548                 break;
1549         case F_SETFL:
1550                 flags = va_arg(ap, long);
1551                 flags &= FCNTL_FLMASK;
1552                 if (flags & FCNTL_FLMASK_INVALID) {
1553                         LCONSOLE_ERROR_MSG(0x010, "liblustre does not support "
1554                                            "the O_NONBLOCK or O_ASYNC flags. "
1555                                            "Please fix your application.\n");
1556                         *rtn = -EINVAL;
1557                         err = EINVAL;
1558                         break;
1559                 }
1560                 lli->lli_open_flags = (int)(flags & FCNTL_FLMASK) |
1561                                       (lli->lli_open_flags & ~FCNTL_FLMASK);
1562                 *rtn = 0;
1563                 break;
1564         case F_GETLK:
1565 #ifdef F_GETLK64
1566 #if F_GETLK64 != F_GETLK
1567         case F_GETLK64:
1568 #endif
1569 #endif
1570                 flock = va_arg(ap, struct flock *);
1571                 err = llu_fcntl_getlk(ino, flock);
1572                 *rtn = err? -1: 0;
1573                 break;
1574         case F_SETLK:
1575 #ifdef F_SETLKW64
1576 #if F_SETLKW64 != F_SETLKW
1577         case F_SETLKW64:
1578 #endif
1579 #endif
1580         case F_SETLKW:
1581 #ifdef F_SETLK64
1582 #if F_SETLK64 != F_SETLK
1583         case F_SETLK64:
1584 #endif
1585 #endif
1586                 flock = va_arg(ap, struct flock *);
1587                 err = llu_fcntl_setlk(ino, cmd, flock);
1588                 *rtn = err? -1: 0;
1589                 break;
1590         default:
1591                 CERROR("unsupported fcntl cmd %x\n", cmd);
1592                 *rtn = -ENOSYS;
1593                 err = ENOSYS;
1594                 break;
1595         }
1596
1597         liblustre_wait_event(0);
1598         return err;
1599 }
1600
1601 static int llu_get_grouplock(struct inode *inode, unsigned long arg)
1602 {
1603         struct llu_inode_info *lli = llu_i2info(inode);
1604         struct ll_file_data *fd = lli->lli_file_data;
1605         ldlm_policy_data_t policy = { .l_extent = { .start = 0,
1606                                                     .end = OBD_OBJECT_EOF}};
1607         struct lustre_handle lockh = { 0 };
1608         struct lov_stripe_md *lsm = lli->lli_smd;
1609         ldlm_error_t err;
1610         int flags = 0;
1611         ENTRY;
1612
1613         if (fd->fd_flags & LL_FILE_GROUP_LOCKED) {
1614                 RETURN(-EINVAL);
1615         }
1616
1617         policy.l_extent.gid = arg;
1618         if (lli->lli_open_flags & O_NONBLOCK)
1619                 flags = LDLM_FL_BLOCK_NOWAIT;
1620
1621         err = llu_extent_lock(fd, inode, lsm, LCK_GROUP, &policy, &lockh,
1622                               flags);
1623         if (err)
1624                 RETURN(err);
1625
1626         fd->fd_flags |= LL_FILE_GROUP_LOCKED|LL_FILE_IGNORE_LOCK;
1627         fd->fd_gid = arg;
1628         memcpy(&fd->fd_cwlockh, &lockh, sizeof(lockh));
1629
1630         RETURN(0);
1631 }
1632
1633 static int llu_put_grouplock(struct inode *inode, unsigned long arg)
1634 {
1635         struct llu_inode_info *lli = llu_i2info(inode);
1636         struct ll_file_data *fd = lli->lli_file_data;
1637         struct lov_stripe_md *lsm = lli->lli_smd;
1638         ldlm_error_t err;
1639         ENTRY;
1640
1641         if (!(fd->fd_flags & LL_FILE_GROUP_LOCKED))
1642                 RETURN(-EINVAL);
1643
1644         if (fd->fd_gid != arg)
1645                 RETURN(-EINVAL);
1646
1647         fd->fd_flags &= ~(LL_FILE_GROUP_LOCKED|LL_FILE_IGNORE_LOCK);
1648
1649         err = llu_extent_unlock(fd, inode, lsm, LCK_GROUP, &fd->fd_cwlockh);
1650         if (err)
1651                 RETURN(err);
1652
1653         fd->fd_gid = 0;
1654         memset(&fd->fd_cwlockh, 0, sizeof(fd->fd_cwlockh));
1655
1656         RETURN(0);
1657 }
1658
1659 static int llu_lov_dir_setstripe(struct inode *ino, unsigned long arg)
1660 {
1661         struct llu_sb_info *sbi = llu_i2sbi(ino);
1662         struct ptlrpc_request *request = NULL;
1663         struct md_op_data op_data = {{ 0 }};
1664         struct lov_user_md lum, *lump = (struct lov_user_md *)arg;
1665         int rc = 0;
1666
1667         llu_prep_md_op_data(&op_data, ino, NULL, NULL, 0, 0,
1668                             LUSTRE_OPC_ANY);
1669
1670         LASSERT(sizeof(lum) == sizeof(*lump));
1671         LASSERT(sizeof(lum.lmm_objects[0]) ==
1672                 sizeof(lump->lmm_objects[0]));
1673         if (copy_from_user(&lum, lump, sizeof(lum)))
1674                 return(-EFAULT);
1675
1676         switch (lum.lmm_magic) {
1677         case LOV_USER_MAGIC_V1: {
1678                 if (lum.lmm_magic != cpu_to_le32(LOV_USER_MAGIC_V1))
1679                         lustre_swab_lov_user_md_v1(&lum);
1680                 break;
1681                 }
1682         case LOV_USER_MAGIC_V3: {
1683                 if (lum.lmm_magic != cpu_to_le32(LOV_USER_MAGIC_V3))
1684                         lustre_swab_lov_user_md_v3((struct lov_user_md_v3 *)&lum);
1685                 break;
1686                 }
1687         default: {
1688                 CDEBUG(D_IOCTL, "bad userland LOV MAGIC:"
1689                                 " %#08x != %#08x nor %#08x\n",
1690                                 lum.lmm_magic, LOV_USER_MAGIC_V1,
1691                                 LOV_USER_MAGIC_V3);
1692                 RETURN(-EINVAL);
1693         }
1694         }
1695
1696         /* swabbing is done in lov_setstripe() on server side */
1697         rc = md_setattr(sbi->ll_md_exp, &op_data, &lum,
1698                         sizeof(lum), NULL, 0, &request, NULL);
1699         if (rc) {
1700                 ptlrpc_req_finished(request);
1701                 if (rc != -EPERM && rc != -EACCES)
1702                         CERROR("md_setattr fails: rc = %d\n", rc);
1703                 return rc;
1704         }
1705         ptlrpc_req_finished(request);
1706
1707         return rc;
1708 }
1709
1710 static int llu_lov_setstripe_ea_info(struct inode *ino, int flags,
1711                                      struct lov_user_md *lum, int lum_size)
1712 {
1713         struct llu_sb_info *sbi = llu_i2sbi(ino);
1714         struct llu_inode_info *lli = llu_i2info(ino);
1715         struct llu_inode_info *lli2 = NULL;
1716         struct lov_stripe_md *lsm;
1717         struct lookup_intent oit = {.it_op = IT_OPEN, .it_flags = flags};
1718         struct ldlm_enqueue_info einfo = { LDLM_IBITS, LCK_CR,
1719                 llu_md_blocking_ast, ldlm_completion_ast, NULL, NULL, NULL };
1720
1721         struct ptlrpc_request *req = NULL;
1722         struct lustre_md md;
1723         struct md_op_data data = {{ 0 }};
1724         struct lustre_handle lockh;
1725         int rc = 0;
1726         ENTRY;
1727
1728         lsm = lli->lli_smd;
1729         if (lsm) {
1730                 CDEBUG(D_IOCTL, "stripe already exists for ino "DFID"\n",
1731                        PFID(&lli->lli_fid));
1732                 return -EEXIST;
1733         }
1734
1735         OBD_ALLOC(lli2, sizeof(struct llu_inode_info));
1736         if (!lli2)
1737                 return -ENOMEM;
1738
1739         memcpy(lli2, lli, sizeof(struct llu_inode_info));
1740         lli2->lli_open_count = 0;
1741         lli2->lli_it = NULL;
1742         lli2->lli_file_data = NULL;
1743         lli2->lli_smd = NULL;
1744         lli2->lli_symlink_name = NULL;
1745         ino->i_private = lli2;
1746
1747         llu_prep_md_op_data(&data, NULL, ino, NULL, 0, O_RDWR,
1748                             LUSTRE_OPC_ANY);
1749
1750         rc = md_enqueue(sbi->ll_md_exp, &einfo, &oit, &data,
1751                         &lockh, lum, lum_size, NULL, LDLM_FL_INTENT_ONLY);
1752         if (rc)
1753                 GOTO(out, rc);
1754
1755         req = oit.d.lustre.it_data;
1756         rc = it_open_error(DISP_IT_EXECD, &oit);
1757         if (rc) {
1758                 req->rq_replay = 0;
1759                 GOTO(out, rc);
1760         }
1761
1762         rc = it_open_error(DISP_OPEN_OPEN, &oit);
1763         if (rc) {
1764                 req->rq_replay = 0;
1765                 GOTO(out, rc);
1766         }
1767
1768         rc = md_get_lustre_md(sbi->ll_md_exp, req,
1769                               sbi->ll_dt_exp, sbi->ll_md_exp, &md);
1770         if (rc)
1771                 GOTO(out, rc);
1772
1773         lli->lli_smd = lli2->lli_smd;
1774         lli2->lli_smd = NULL;
1775
1776         llu_local_open(lli2, &oit);
1777
1778         /* release intent */
1779         if (lustre_handle_is_used(&lockh))
1780                 ldlm_lock_decref(&lockh, LCK_CR);
1781
1782         ptlrpc_req_finished(req);
1783         req = NULL;
1784
1785         rc = llu_file_release(ino);
1786  out:
1787         ino->i_private = lli;
1788         if (!rc)
1789                 llu_update_inode(ino, &md);
1790         if (lli2)
1791                 OBD_FREE(lli2, sizeof(struct llu_inode_info));
1792         if (req != NULL)
1793                 ptlrpc_req_finished(req);
1794         RETURN(rc);
1795 }
1796
1797 static int llu_lov_file_setstripe(struct inode *ino, unsigned long arg)
1798 {
1799         struct lov_user_md lum, *lump = (struct lov_user_md *)arg;
1800         int rc;
1801         int flags = FMODE_WRITE;
1802         ENTRY;
1803
1804         LASSERT(sizeof(lum) == sizeof(*lump));
1805         LASSERT(sizeof(lum.lmm_objects[0]) == sizeof(lump->lmm_objects[0]));
1806         if (copy_from_user(&lum, lump, sizeof(lum)))
1807                 RETURN(-EFAULT);
1808
1809         rc = llu_lov_setstripe_ea_info(ino, flags, &lum, sizeof(lum));
1810         RETURN(rc);
1811 }
1812
1813 static int llu_lov_setstripe(struct inode *ino, unsigned long arg)
1814 {
1815         struct intnl_stat *st = llu_i2stat(ino);
1816         if (S_ISREG(st->st_mode))
1817                 return llu_lov_file_setstripe(ino, arg);
1818         if (S_ISDIR(st->st_mode))
1819                 return llu_lov_dir_setstripe(ino, arg);
1820
1821         return -EINVAL;
1822 }
1823
1824 static int llu_lov_getstripe(struct inode *ino, unsigned long arg)
1825 {
1826         struct lov_stripe_md *lsm = llu_i2info(ino)->lli_smd;
1827
1828         if (!lsm)
1829                 RETURN(-ENODATA);
1830
1831         return obd_iocontrol(LL_IOC_LOV_GETSTRIPE, llu_i2obdexp(ino), 0, lsm,
1832                             (void *)arg);
1833 }
1834
1835 static int llu_iop_ioctl(struct inode *ino, unsigned long int request,
1836                          va_list ap)
1837 {
1838         unsigned long arg;
1839         int rc;
1840
1841         liblustre_wait_event(0);
1842
1843         switch (request) {
1844         case LL_IOC_GROUP_LOCK:
1845                 arg = va_arg(ap, unsigned long);
1846                 rc = llu_get_grouplock(ino, arg);
1847                 break;
1848         case LL_IOC_GROUP_UNLOCK:
1849                 arg = va_arg(ap, unsigned long);
1850                 rc = llu_put_grouplock(ino, arg);
1851                 break;
1852         case LL_IOC_LOV_SETSTRIPE:
1853                 arg = va_arg(ap, unsigned long);
1854                 rc = llu_lov_setstripe(ino, arg);
1855                 break;
1856         case LL_IOC_LOV_GETSTRIPE:
1857                 arg = va_arg(ap, unsigned long);
1858                 rc = llu_lov_getstripe(ino, arg);
1859                 break;
1860         default:
1861                 CERROR("did not support ioctl cmd %lx\n", request);
1862                 rc = -ENOSYS;
1863                 break;
1864         }
1865
1866         liblustre_wait_event(0);
1867         return rc;
1868 }
1869
1870 /*
1871  * we already do syncronous read/write
1872  */
1873 static int llu_iop_sync(struct inode *inode)
1874 {
1875         liblustre_wait_event(0);
1876         return 0;
1877 }
1878
1879 static int llu_iop_datasync(struct inode *inode)
1880 {
1881         liblustre_wait_event(0);
1882         return 0;
1883 }
1884
1885 struct filesys_ops llu_filesys_ops =
1886 {
1887         fsop_gone: llu_fsop_gone,
1888 };
1889
1890 struct inode *llu_iget(struct filesys *fs, struct lustre_md *md)
1891 {
1892         struct inode *inode;
1893         struct lu_fid fid;
1894         struct file_identifier fileid = {&fid, sizeof(fid)};
1895
1896         if ((md->body->valid & (OBD_MD_FLID | OBD_MD_FLTYPE)) !=
1897             (OBD_MD_FLID | OBD_MD_FLTYPE)) {
1898                 CERROR("bad md body valid mask "LPX64"\n", md->body->valid);
1899                 LBUG();
1900                 return ERR_PTR(-EPERM);
1901         }
1902
1903         /* try to find existing inode */
1904         fid = md->body->fid1;
1905
1906         inode = _sysio_i_find(fs, &fileid);
1907         if (inode) {
1908                 if (inode->i_zombie/* ||
1909                     lli->lli_st_generation != md->body->generation*/) {
1910                         I_RELE(inode);
1911                 }
1912                 else {
1913                         llu_update_inode(inode, md);
1914                         return inode;
1915                 }
1916         }
1917
1918         inode = llu_new_inode(fs, &fid);
1919         if (inode)
1920                 llu_update_inode(inode, md);
1921
1922         return inode;
1923 }
1924
1925 static int
1926 llu_init_ea_size(struct obd_export *md_exp, struct obd_export *dt_exp)
1927 {
1928         /* even if default lov is LOV_MAGIC_V1 we use LOV_MAGIC_V3
1929          * to be sure buffer are large enough */
1930         struct lov_stripe_md lsm = { .lsm_magic = LOV_MAGIC_V3 };
1931         __u32 valsize = sizeof(struct lov_desc);
1932         int rc, easize, def_easize, cookiesize;
1933         struct lov_desc desc;
1934         __u32 stripes;
1935         ENTRY;
1936
1937         rc = obd_get_info(dt_exp, sizeof(KEY_LOVDESC), KEY_LOVDESC,
1938                           &valsize, &desc, NULL);
1939         if (rc)
1940                 RETURN(rc);
1941
1942         stripes = min(desc.ld_tgt_count, (__u32)LOV_MAX_STRIPE_COUNT);
1943         lsm.lsm_stripe_count = stripes;
1944         easize = obd_size_diskmd(dt_exp, &lsm);
1945
1946         lsm.lsm_stripe_count = desc.ld_default_stripe_count;
1947         def_easize = obd_size_diskmd(dt_exp, &lsm);
1948
1949         cookiesize = stripes * sizeof(struct llog_cookie);
1950
1951         CDEBUG(D_HA, "updating max_mdsize/max_cookiesize: %d/%d\n",
1952                easize, cookiesize);
1953
1954         rc = md_init_ea_size(md_exp, easize, def_easize, cookiesize);
1955         RETURN(rc);
1956 }
1957
1958 static int
1959 llu_fsswop_mount(const char *source,
1960                  unsigned flags,
1961                  const void *data __IS_UNUSED,
1962                  struct pnode *tocover,
1963                  struct mount **mntp)
1964 {
1965         struct filesys *fs;
1966         struct inode *root;
1967         struct pnode_base *rootpb;
1968         struct obd_device *obd;
1969         struct llu_sb_info *sbi;
1970         struct obd_statfs osfs;
1971         static struct qstr noname = { NULL, 0, 0 };
1972         struct ptlrpc_request *request = NULL;
1973         struct lustre_handle md_conn = {0, };
1974         struct lustre_handle dt_conn = {0, };
1975         struct lustre_md md;
1976         class_uuid_t uuid;
1977         struct config_llog_instance cfg = {0, };
1978         char ll_instance[sizeof(sbi) * 2 + 1];
1979         struct lustre_profile *lprof;
1980         char *zconf_mgsnid, *zconf_profile;
1981         char *osc = NULL, *mdc = NULL;
1982         int async = 1, err = -EINVAL;
1983         struct obd_connect_data ocd = {0,};
1984
1985         ENTRY;
1986
1987         if (ll_parse_mount_target(source,
1988                                   &zconf_mgsnid,
1989                                   &zconf_profile)) {
1990                 CERROR("mal-formed target %s\n", source);
1991                 RETURN(err);
1992         }
1993         if (!zconf_mgsnid || !zconf_profile) {
1994                 printf("Liblustre: invalid target %s\n", source);
1995                 RETURN(err);
1996         }
1997         /* allocate & initialize sbi */
1998         OBD_ALLOC(sbi, sizeof(*sbi));
1999         if (!sbi)
2000                 RETURN(-ENOMEM);
2001
2002         CFS_INIT_LIST_HEAD(&sbi->ll_conn_chain);
2003         ll_generate_random_uuid(uuid);
2004         class_uuid_unparse(uuid, &sbi->ll_sb_uuid);
2005
2006         /* generate a string unique to this super, let's try
2007          the address of the super itself.*/
2008         sprintf(ll_instance, "%p", sbi);
2009
2010         /* retrive & parse config log */
2011         cfg.cfg_instance = ll_instance;
2012         cfg.cfg_uuid = sbi->ll_sb_uuid;
2013         err = liblustre_process_log(&cfg, zconf_mgsnid, zconf_profile, 1);
2014         if (err < 0) {
2015                 CERROR("Unable to process log: %s\n", zconf_profile);
2016                 GOTO(out_free, err);
2017         }
2018
2019         lprof = class_get_profile(zconf_profile);
2020         if (lprof == NULL) {
2021                 CERROR("No profile found: %s\n", zconf_profile);
2022                 GOTO(out_free, err = -EINVAL);
2023         }
2024         OBD_ALLOC(osc, strlen(lprof->lp_dt) + strlen(ll_instance) + 2);
2025         sprintf(osc, "%s-%s", lprof->lp_dt, ll_instance);
2026
2027         OBD_ALLOC(mdc, strlen(lprof->lp_md) + strlen(ll_instance) + 2);
2028         sprintf(mdc, "%s-%s", lprof->lp_md, ll_instance);
2029
2030         if (!osc) {
2031                 CERROR("no osc\n");
2032                 GOTO(out_free, err = -EINVAL);
2033         }
2034         if (!mdc) {
2035                 CERROR("no mdc\n");
2036                 GOTO(out_free, err = -EINVAL);
2037         }
2038
2039         fs = _sysio_fs_new(&llu_filesys_ops, flags, sbi);
2040         if (!fs) {
2041                 err = -ENOMEM;
2042                 goto out_free;
2043         }
2044
2045         obd = class_name2obd(mdc);
2046         if (!obd) {
2047                 CERROR("MDC %s: not setup or attached\n", mdc);
2048                 GOTO(out_free, err = -EINVAL);
2049         }
2050         obd_set_info_async(obd->obd_self_export, sizeof(KEY_ASYNC), KEY_ASYNC,
2051                            sizeof(async), &async, NULL);
2052
2053         ocd.ocd_connect_flags = OBD_CONNECT_IBITS | OBD_CONNECT_VERSION |
2054                                 OBD_CONNECT_FID | OBD_CONNECT_AT;
2055 #ifdef LIBLUSTRE_POSIX_ACL
2056         ocd.ocd_connect_flags |= OBD_CONNECT_ACL;
2057 #endif
2058         ocd.ocd_ibits_known = MDS_INODELOCK_FULL;
2059         ocd.ocd_version = LUSTRE_VERSION_CODE;
2060
2061         /* setup mdc */
2062         err = obd_connect(NULL, &md_conn, obd, &sbi->ll_sb_uuid, &ocd, NULL);
2063         if (err) {
2064                 CERROR("cannot connect to %s: rc = %d\n", mdc, err);
2065                 GOTO(out_free, err);
2066         }
2067         sbi->ll_md_exp = class_conn2export(&md_conn);
2068
2069         err = obd_statfs(obd, &osfs, 100000000, 0);
2070         if (err)
2071                 GOTO(out_md, err);
2072
2073         /*
2074          * FIXME fill fs stat data into sbi here!!! FIXME
2075          */
2076
2077         /* setup osc */
2078         obd = class_name2obd(osc);
2079         if (!obd) {
2080                 CERROR("OSC %s: not setup or attached\n", osc);
2081                 GOTO(out_md, err = -EINVAL);
2082         }
2083         obd_set_info_async(obd->obd_self_export, sizeof(KEY_ASYNC), KEY_ASYNC,
2084                            sizeof(async), &async, NULL);
2085
2086         obd->obd_upcall.onu_owner = &sbi->ll_lco;
2087         obd->obd_upcall.onu_upcall = ll_ocd_update;
2088
2089         ocd.ocd_connect_flags = OBD_CONNECT_SRVLOCK | OBD_CONNECT_REQPORTAL |
2090                                 OBD_CONNECT_VERSION | OBD_CONNECT_TRUNCLOCK |
2091                                 OBD_CONNECT_FID | OBD_CONNECT_AT;
2092         ocd.ocd_version = LUSTRE_VERSION_CODE;
2093         err = obd_connect(NULL, &dt_conn, obd, &sbi->ll_sb_uuid, &ocd, NULL);
2094         if (err) {
2095                 CERROR("cannot connect to %s: rc = %d\n", osc, err);
2096                 GOTO(out_md, err);
2097         }
2098         sbi->ll_dt_exp = class_conn2export(&dt_conn);
2099         sbi->ll_lco.lco_flags = ocd.ocd_connect_flags;
2100
2101         if (err) {
2102                 CERROR("cannot register lock cancel callback: rc = %d\n", err);
2103                 GOTO(out_dt, err);
2104         }
2105
2106         llu_init_ea_size(sbi->ll_md_exp, sbi->ll_dt_exp);
2107
2108         fid_zero(&sbi->ll_root_fid);
2109         err = md_getstatus(sbi->ll_md_exp, &sbi->ll_root_fid, NULL);
2110         if (err) {
2111                 CERROR("cannot mds_connect: rc = %d\n", err);
2112                 GOTO(out_lock_cn_cb, err);
2113         }
2114         if (!fid_is_sane(&sbi->ll_root_fid)) {
2115                 CERROR("Invalid root fid during mount\n");
2116                 GOTO(out_lock_cn_cb, err = -EINVAL);
2117         }
2118         CDEBUG(D_SUPER, "rootfid "DFID"\n", PFID(&sbi->ll_root_fid));
2119
2120         /* fetch attr of root inode */
2121         err = md_getattr(sbi->ll_md_exp, &sbi->ll_root_fid, NULL,
2122                          OBD_MD_FLGETATTR | OBD_MD_FLBLOCKS, 0, &request);
2123         if (err) {
2124                 CERROR("md_getattr failed for root: rc = %d\n", err);
2125                 GOTO(out_lock_cn_cb, err);
2126         }
2127
2128         err = md_get_lustre_md(sbi->ll_md_exp, request,
2129                                sbi->ll_dt_exp, sbi->ll_md_exp, &md);
2130         if (err) {
2131                 CERROR("failed to understand root inode md: rc = %d\n",err);
2132                 GOTO(out_request, err);
2133         }
2134
2135         LASSERT(fid_is_sane(&sbi->ll_root_fid));
2136
2137         root = llu_iget(fs, &md);
2138         if (!root || IS_ERR(root)) {
2139                 CERROR("fail to generate root inode\n");
2140                 GOTO(out_request, err = -EBADF);
2141         }
2142
2143         /*
2144          * Generate base path-node for root.
2145          */
2146         rootpb = _sysio_pb_new(&noname, NULL, root);
2147         if (!rootpb) {
2148                 err = -ENOMEM;
2149                 goto out_inode;
2150         }
2151
2152         err = _sysio_do_mount(fs, rootpb, flags, tocover, mntp);
2153         if (err) {
2154                 _sysio_pb_gone(rootpb);
2155                 goto out_inode;
2156         }
2157
2158         cl_sb_init(sbi);
2159
2160         ptlrpc_req_finished(request);
2161
2162         CDEBUG(D_SUPER, "LibLustre: %s mounted successfully!\n", source);
2163         liblustre_wait_idle();
2164
2165         return 0;
2166
2167 out_inode:
2168         _sysio_i_gone(root);
2169 out_request:
2170         ptlrpc_req_finished(request);
2171 out_lock_cn_cb:
2172 out_dt:
2173         obd_disconnect(sbi->ll_dt_exp);
2174 out_md:
2175         obd_disconnect(sbi->ll_md_exp);
2176 out_free:
2177         if (osc)
2178                 OBD_FREE(osc, strlen(osc) + 1);
2179         if (mdc)
2180                 OBD_FREE(mdc, strlen(mdc) + 1);
2181         OBD_FREE(sbi, sizeof(*sbi));
2182         liblustre_wait_idle();
2183         return err;
2184 }
2185
2186 struct fssw_ops llu_fssw_ops = {
2187         llu_fsswop_mount
2188 };
2189
2190 static struct inode_ops llu_inode_ops = {
2191         inop_lookup:    llu_iop_lookup,
2192         inop_getattr:   llu_iop_getattr,
2193         inop_setattr:   llu_iop_setattr,
2194         inop_filldirentries:     llu_iop_filldirentries,
2195         inop_mkdir:     llu_iop_mkdir_raw,
2196         inop_rmdir:     llu_iop_rmdir_raw,
2197         inop_symlink:   llu_iop_symlink_raw,
2198         inop_readlink:  llu_iop_readlink,
2199         inop_open:      llu_iop_open,
2200         inop_close:     llu_iop_close,
2201         inop_link:      llu_iop_link_raw,
2202         inop_unlink:    llu_iop_unlink_raw,
2203         inop_rename:    llu_iop_rename_raw,
2204         inop_pos:       llu_iop_pos,
2205         inop_read:      llu_iop_read,
2206         inop_write:     llu_iop_write,
2207         inop_iodone:    llu_iop_iodone,
2208         inop_fcntl:     llu_iop_fcntl,
2209         inop_sync:      llu_iop_sync,
2210         inop_datasync:  llu_iop_datasync,
2211         inop_ioctl:     llu_iop_ioctl,
2212         inop_mknod:     llu_iop_mknod_raw,
2213 #ifdef _HAVE_STATVFS
2214         inop_statvfs:   llu_iop_statvfs,
2215 #endif
2216         inop_gone:      llu_iop_gone,
2217 };