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 #include "llite_lib.h"
  69
  70 #ifndef MAY_EXEC
  71 #define MAY_EXEC        1
  72 #define MAY_WRITE       2
  73 #define MAY_READ        4
  74 #endif
  75
  76 #define S_IXUGO (S_IXUSR|S_IXGRP|S_IXOTH)
  77
  78 static int ll_permission(struct inode *inode, int mask)
  79 {
  80         struct intnl_stat *st = llu_i2stat(inode);
  81         mode_t mode = st->st_mode;
  82
  83         if (current->fsuid == st->st_uid)
  84                 mode >>= 6;
  85         else if (cfs_curproc_is_in_groups(st->st_gid))
  86                 mode >>= 3;
  87
  88         if ((mode & mask & (MAY_READ|MAY_WRITE|MAY_EXEC)) == mask)
  89                 return 0;
  90
  91         if ((mask & (MAY_READ|MAY_WRITE)) ||
  92             (st->st_mode & S_IXUGO))
  93                 if (cfs_capable(CFS_CAP_DAC_OVERRIDE))
  94                         return 0;
  95
  96         if (mask == MAY_READ ||
  97             (S_ISDIR(st->st_mode) && !(mask & MAY_WRITE))) {
  98                 if (cfs_capable(CFS_CAP_DAC_READ_SEARCH))
  99                         return 0;
 100         }
 101
 102         return -EACCES;
 103 }
 104
 105 static void llu_fsop_gone(struct filesys *fs)
 106 {
 107         struct llu_sb_info *sbi = (struct llu_sb_info *) fs->fs_private;
 108         struct obd_device *obd = class_exp2obd(sbi->ll_md_exp);
 109         int next = 0;
 110         ENTRY;
 111
 112         cfs_list_del(&sbi->ll_conn_chain);
 113         cl_sb_fini(sbi);
 114         obd_disconnect(sbi->ll_dt_exp);
 115         obd_disconnect(sbi->ll_md_exp);
 116
 117         while ((obd = class_devices_in_group(&sbi->ll_sb_uuid, &next)) != NULL)
 118                 class_manual_cleanup(obd);
 119
 120         OBD_FREE(sbi, sizeof(*sbi));
 121
 122         liblustre_wait_idle();
 123         EXIT;
 124 }
 125
 126 static struct inode_ops llu_inode_ops;
 127
 128 static ldlm_mode_t llu_take_md_lock(struct inode *inode, __u64 bits,
 129                                     struct lustre_handle *lockh)
 130 {
 131         ldlm_policy_data_t policy = { .l_inodebits = {bits}};
 132         struct lu_fid *fid;
 133         ldlm_mode_t rc;
 134         int flags;
 135         ENTRY;
 136
 137         fid = &llu_i2info(inode)->lli_fid;
 138         CDEBUG(D_INFO, "trying to match res "DFID"\n", PFID(fid));
 139
 140         flags = LDLM_FL_BLOCK_GRANTED | LDLM_FL_CBPENDING;
 141         rc = md_lock_match(llu_i2mdexp(inode), flags, fid, LDLM_IBITS, &policy,
 142                            LCK_CR|LCK_CW|LCK_PR|LCK_PW, lockh);
 143         RETURN(rc);
 144 }
 145
 146 void llu_update_inode(struct inode *inode, struct lustre_md *md)
 147 {
 148         struct llu_inode_info *lli = llu_i2info(inode);
 149         struct mdt_body *body = md->body;
 150         struct lov_stripe_md *lsm = md->lsm;
 151         struct intnl_stat *st = llu_i2stat(inode);
 152
 153         LASSERT ((lsm != NULL) == ((body->valid & OBD_MD_FLEASIZE) != 0));
 154
 155         if (body->valid & OBD_MD_FLMODE)
 156                 st->st_mode = (st->st_mode & S_IFMT)|(body->mode & ~S_IFMT);
 157         if (body->valid & OBD_MD_FLTYPE)
 158                 st->st_mode = (st->st_mode & ~S_IFMT)|(body->mode & S_IFMT);
 159
 160         if (lsm != NULL) {
 161                 if (lli->lli_smd == NULL) {
 162                         cl_inode_init(inode, md);
 163                         lli->lli_smd = lsm;
 164                         lli->lli_maxbytes = lsm->lsm_maxbytes;
 165                         if (lli->lli_maxbytes > PAGE_CACHE_MAXBYTES)
 166                                 lli->lli_maxbytes = PAGE_CACHE_MAXBYTES;
 167                 } else {
 168                         if (lov_stripe_md_cmp(lli->lli_smd, lsm)) {
 169                                 CERROR("lsm mismatch for inode %lld\n",
 170                                        (long long)st->st_ino);
 171                                 LBUG();
 172                         }
 173                 }
 174         }
 175
 176         if (body->valid & OBD_MD_FLATIME) {
 177                 if (body->atime > LTIME_S(st->st_atime))
 178                         LTIME_S(st->st_atime) = body->atime;
 179                 lli->lli_lvb.lvb_atime = body->atime;
 180         }
 181         if (body->valid & OBD_MD_FLMTIME) {
 182                 if (body->mtime > LTIME_S(st->st_mtime))
 183                         LTIME_S(st->st_mtime) = body->mtime;
 184                 lli->lli_lvb.lvb_mtime = body->mtime;
 185         }
 186         if (body->valid & OBD_MD_FLCTIME) {
 187                 if (body->ctime > LTIME_S(st->st_ctime))
 188                         LTIME_S(st->st_ctime) = body->ctime;
 189                 lli->lli_lvb.lvb_ctime = body->ctime;
 190         }
 191         if (S_ISREG(st->st_mode))
 192                 st->st_blksize = min(2UL * PTLRPC_MAX_BRW_SIZE, LL_MAX_BLKSIZE);
 193         else
 194                 st->st_blksize = 4096;
 195         if (body->valid & OBD_MD_FLUID)
 196                 st->st_uid = body->uid;
 197         if (body->valid & OBD_MD_FLGID)
 198                 st->st_gid = body->gid;
 199         if (body->valid & OBD_MD_FLNLINK)
 200                 st->st_nlink = body->nlink;
 201         if (body->valid & OBD_MD_FLRDEV)
 202                 st->st_rdev = body->rdev;
 203         if (body->valid & OBD_MD_FLFLAGS)
 204                 lli->lli_st_flags = body->flags;
 205         if (body->valid & OBD_MD_FLSIZE) {
 206                 if ((llu_i2sbi(inode)->ll_lco.lco_flags & OBD_CONNECT_SOM) &&
 207                     S_ISREG(st->st_mode) && lli->lli_smd) {
 208                         struct lustre_handle lockh;
 209                         ldlm_mode_t mode;
 210
 211                         /* As it is possible a blocking ast has been processed
 212                          * by this time, we need to check there is an UPDATE
 213                          * lock on the client and set LLIF_MDS_SIZE_LOCK holding
 214                          * it. */
 215                         mode = llu_take_md_lock(inode, MDS_INODELOCK_UPDATE,
 216                                                 &lockh);
 217                         if (mode) {
 218                                 st->st_size = body->size;
 219                                 lli->lli_flags |= LLIF_MDS_SIZE_LOCK;
 220                                 ldlm_lock_decref(&lockh, mode);
 221                         }
 222                 } else {
 223                     st->st_size = body->size;
 224                 }
 225
 226                 if (body->valid & OBD_MD_FLBLOCKS)
 227                         st->st_blocks = body->blocks;
 228         }
 229 }
 230
 231 void obdo_to_inode(struct inode *dst, struct obdo *src, obd_flag valid)
 232 {
 233         struct llu_inode_info *lli = llu_i2info(dst);
 234         struct intnl_stat *st = llu_i2stat(dst);
 235
 236         valid &= src->o_valid;
 237
 238         if (valid & (OBD_MD_FLCTIME | OBD_MD_FLMTIME))
 239                 CDEBUG(D_INODE,"valid "LPX64", cur time "CFS_TIME_T"/"CFS_TIME_T
 240                        ", new %lu/%lu\n",
 241                        src->o_valid,
 242                        LTIME_S(st->st_mtime), LTIME_S(st->st_ctime),
 243                        (long)src->o_mtime, (long)src->o_ctime);
 244
 245         if (valid & OBD_MD_FLATIME)
 246                 LTIME_S(st->st_atime) = src->o_atime;
 247         if (valid & OBD_MD_FLMTIME)
 248                 LTIME_S(st->st_mtime) = src->o_mtime;
 249         if (valid & OBD_MD_FLCTIME && src->o_ctime > LTIME_S(st->st_ctime))
 250                 LTIME_S(st->st_ctime) = src->o_ctime;
 251         if (valid & OBD_MD_FLSIZE)
 252                 st->st_size = src->o_size;
 253         if (valid & OBD_MD_FLBLOCKS) /* allocation of space */
 254                 st->st_blocks = src->o_blocks;
 255         if (valid & OBD_MD_FLBLKSZ)
 256                 st->st_blksize = src->o_blksize;
 257         if (valid & OBD_MD_FLTYPE)
 258                 st->st_mode = (st->st_mode & ~S_IFMT) | (src->o_mode & S_IFMT);
 259         if (valid & OBD_MD_FLMODE)
 260                 st->st_mode = (st->st_mode & S_IFMT) | (src->o_mode & ~S_IFMT);
 261         if (valid & OBD_MD_FLUID)
 262                 st->st_uid = src->o_uid;
 263         if (valid & OBD_MD_FLGID)
 264                 st->st_gid = src->o_gid;
 265         if (valid & OBD_MD_FLFLAGS)
 266                 lli->lli_st_flags = src->o_flags;
 267 }
 268
 269 #define S_IRWXUGO       (S_IRWXU|S_IRWXG|S_IRWXO)
 270 #define S_IALLUGO       (S_ISUID|S_ISGID|S_ISVTX|S_IRWXUGO)
 271
 272 void obdo_from_inode(struct obdo *dst, struct inode *src, obd_flag valid)
 273 {
 274         struct llu_inode_info *lli = llu_i2info(src);
 275         struct intnl_stat *st = llu_i2stat(src);
 276         obd_flag newvalid = 0;
 277
 278         if (valid & (OBD_MD_FLCTIME | OBD_MD_FLMTIME))
 279                 CDEBUG(D_INODE, "valid %x, new time "CFS_TIME_T"/"CFS_TIME_T"\n",
 280                        valid, LTIME_S(st->st_mtime),
 281                        LTIME_S(st->st_ctime));
 282
 283         if (valid & OBD_MD_FLATIME) {
 284                 dst->o_atime = LTIME_S(st->st_atime);
 285                 newvalid |= OBD_MD_FLATIME;
 286         }
 287         if (valid & OBD_MD_FLMTIME) {
 288                 dst->o_mtime = LTIME_S(st->st_mtime);
 289                 newvalid |= OBD_MD_FLMTIME;
 290         }
 291         if (valid & OBD_MD_FLCTIME) {
 292                 dst->o_ctime = LTIME_S(st->st_ctime);
 293                 newvalid |= OBD_MD_FLCTIME;
 294         }
 295         if (valid & OBD_MD_FLSIZE) {
 296                 dst->o_size = st->st_size;
 297                 newvalid |= OBD_MD_FLSIZE;
 298         }
 299         if (valid & OBD_MD_FLBLOCKS) {  /* allocation of space (x512 bytes) */
 300                 dst->o_blocks = st->st_blocks;
 301                 newvalid |= OBD_MD_FLBLOCKS;
 302         }
 303         if (valid & OBD_MD_FLBLKSZ) {   /* optimal block size */
 304                 dst->o_blksize = st->st_blksize;
 305                 newvalid |= OBD_MD_FLBLKSZ;
 306         }
 307         if (valid & OBD_MD_FLTYPE) {
 308                 dst->o_mode = (dst->o_mode & S_IALLUGO)|(st->st_mode & S_IFMT);
 309                 newvalid |= OBD_MD_FLTYPE;
 310         }
 311         if (valid & OBD_MD_FLMODE) {
 312                 dst->o_mode = (dst->o_mode & S_IFMT)|(st->st_mode & S_IALLUGO);
 313                 newvalid |= OBD_MD_FLMODE;
 314         }
 315         if (valid & OBD_MD_FLUID) {
 316                 dst->o_uid = st->st_uid;
 317                 newvalid |= OBD_MD_FLUID;
 318         }
 319         if (valid & OBD_MD_FLGID) {
 320                 dst->o_gid = st->st_gid;
 321                 newvalid |= OBD_MD_FLGID;
 322         }
 323         if (valid & OBD_MD_FLFLAGS) {
 324                 dst->o_flags = lli->lli_st_flags;
 325                 newvalid |= OBD_MD_FLFLAGS;
 326         }
 327         if (valid & OBD_MD_FLGENER) {
 328                 dst->o_generation = lli->lli_st_generation;
 329                 newvalid |= OBD_MD_FLGENER;
 330         }
 331         if (valid & OBD_MD_FLFID) {
 332                 dst->o_fid = st->st_ino;
 333                 newvalid |= OBD_MD_FLFID;
 334         }
 335
 336         dst->o_valid |= newvalid;
 337 }
 338
 339 /**
 340  * Performs the getattr on the inode and updates its fields.
 341  * If @sync != 0, perform the getattr under the server-side lock.
 342  */
 343 int llu_inode_getattr(struct inode *inode, struct obdo *obdo,
 344                       __u64 ioepoch, int sync)
 345 {
 346         struct llu_inode_info *lli = llu_i2info(inode);
 347         struct ptlrpc_request_set *set;
 348         struct lov_stripe_md *lsm = lli->lli_smd;
 349         struct obd_info oinfo = { { { 0 } } };
 350         int rc;
 351         ENTRY;
 352
 353         LASSERT(lsm);
 354
 355         oinfo.oi_md = lsm;
 356         oinfo.oi_oa = obdo;
 357         oinfo.oi_oa->o_id = lsm->lsm_object_id;
 358         oinfo.oi_oa->o_gr = lsm->lsm_object_gr;
 359         oinfo.oi_oa->o_mode = S_IFREG;
 360         oinfo.oi_oa->o_ioepoch = ioepoch;
 361         oinfo.oi_oa->o_valid = OBD_MD_FLID | OBD_MD_FLTYPE |
 362                                OBD_MD_FLSIZE | OBD_MD_FLBLOCKS |
 363                                OBD_MD_FLBLKSZ | OBD_MD_FLMTIME |
 364                                OBD_MD_FLCTIME | OBD_MD_FLGROUP |
 365                                OBD_MD_FLATIME | OBD_MD_FLEPOCH;
 366         if (sync) {
 367                 oinfo.oi_oa->o_valid |= OBD_MD_FLFLAGS;
 368                 oinfo.oi_oa->o_flags |= OBD_FL_SRVLOCK;
 369         }
 370
 371         set = ptlrpc_prep_set();
 372         if (set == NULL) {
 373                 CERROR ("ENOMEM allocing request set\n");
 374                 rc = -ENOMEM;
 375         } else {
 376                 rc = obd_getattr_async(llu_i2obdexp(inode), &oinfo, set);
 377                 if (rc == 0)
 378                         rc = ptlrpc_set_wait(set);
 379                 ptlrpc_set_destroy(set);
 380         }
 381         if (rc)
 382                 RETURN(rc);
 383
 384         oinfo.oi_oa->o_valid = OBD_MD_FLBLOCKS | OBD_MD_FLBLKSZ |
 385                                OBD_MD_FLMTIME | OBD_MD_FLCTIME |
 386                                OBD_MD_FLSIZE;
 387
 388         obdo_refresh_inode(inode, oinfo.oi_oa, oinfo.oi_oa->o_valid);
 389         CDEBUG(D_INODE, "objid "LPX64" size %Lu, blocks %Lu, "
 390                "blksize %Lu\n", lli->lli_smd->lsm_object_id,
 391                (long long unsigned)llu_i2stat(inode)->st_size,
 392                (long long unsigned)llu_i2stat(inode)->st_blocks,
 393                (long long unsigned)llu_i2stat(inode)->st_blksize);
 394         RETURN(0);
 395 }
 396
 397 static struct inode* llu_new_inode(struct filesys *fs,
 398                                    struct lu_fid *fid)
 399 {
 400         struct inode *inode;
 401         struct llu_inode_info *lli;
 402         struct intnl_stat st = {
 403                 .st_dev  = 0,
 404 #if 0
 405 #ifndef AUTOMOUNT_FILE_NAME
 406                 .st_mode = fid->f_type & S_IFMT,
 407 #else
 408                 .st_mode = fid->f_type /* all of the bits! */
 409 #endif
 410 #endif
 411                 /* FIXME: fix this later */
 412                 .st_mode = 0,
 413
 414                 .st_uid  = geteuid(),
 415                 .st_gid  = getegid(),
 416         };
 417
 418         OBD_ALLOC(lli, sizeof(*lli));
 419         if (!lli)
 420                 return NULL;
 421
 422         /* initialize lli here */
 423         lli->lli_sbi = llu_fs2sbi(fs);
 424         lli->lli_smd = NULL;
 425         lli->lli_symlink_name = NULL;
 426         lli->lli_flags = 0;
 427         lli->lli_maxbytes = (__u64)(~0UL);
 428         lli->lli_file_data = NULL;
 429
 430         lli->lli_sysio_fid.fid_data = &lli->lli_fid;
 431         lli->lli_sysio_fid.fid_len = sizeof(lli->lli_fid);
 432         lli->lli_fid = *fid;
 433
 434         /* file identifier is needed by functions like _sysio_i_find() */
 435         inode = _sysio_i_new(fs, &lli->lli_sysio_fid,
 436                              &st, 0, &llu_inode_ops, lli);
 437
 438         if (!inode)
 439                 OBD_FREE(lli, sizeof(*lli));
 440
 441         return inode;
 442 }
 443
 444 static int llu_have_md_lock(struct inode *inode, __u64 lockpart)
 445 {
 446         struct lustre_handle lockh;
 447         ldlm_policy_data_t policy = { .l_inodebits = { lockpart } };
 448         struct lu_fid *fid;
 449         int flags;
 450         ENTRY;
 451
 452         LASSERT(inode);
 453
 454         fid = &llu_i2info(inode)->lli_fid;
 455         CDEBUG(D_INFO, "trying to match res "DFID"\n", PFID(fid));
 456
 457         flags = LDLM_FL_BLOCK_GRANTED | LDLM_FL_CBPENDING | LDLM_FL_TEST_LOCK;
 458         if (md_lock_match(llu_i2mdexp(inode), flags, fid, LDLM_IBITS, &policy,
 459                           LCK_CR|LCK_CW|LCK_PR|LCK_PW, &lockh)) {
 460                 RETURN(1);
 461         }
 462         RETURN(0);
 463 }
 464
 465 static int llu_inode_revalidate(struct inode *inode)
 466 {
 467         struct llu_inode_info *lli = llu_i2info(inode);
 468         struct intnl_stat *st = llu_i2stat(inode);
 469         ENTRY;
 470
 471         if (!inode) {
 472                 CERROR("REPORT THIS LINE TO PETER\n");
 473                 RETURN(0);
 474         }
 475
 476         if (!llu_have_md_lock(inode, MDS_INODELOCK_UPDATE)) {
 477                 struct lustre_md md;
 478                 struct ptlrpc_request *req = NULL;
 479                 struct llu_sb_info *sbi = llu_i2sbi(inode);
 480                 struct md_op_data op_data = { { 0 } };
 481                 unsigned long valid = OBD_MD_FLGETATTR;
 482                 int rc, ealen = 0;
 483
 484                 /* Why don't we update all valid MDS fields here, if we're
 485                  * doing an RPC anyways?  -phil */
 486                 if (S_ISREG(st->st_mode)) {
 487                         ealen = obd_size_diskmd(sbi->ll_dt_exp, NULL);
 488                         valid |= OBD_MD_FLEASIZE;
 489                 }
 490
 491                 llu_prep_md_op_data(&op_data, inode, NULL, NULL, 0, ealen,
 492                                     LUSTRE_OPC_ANY);
 493                 op_data.op_valid = valid;
 494
 495                 rc = md_getattr(sbi->ll_md_exp, &op_data, &req);
 496                 if (rc) {
 497                         CERROR("failure %d inode %llu\n", rc,
 498                                (long long)st->st_ino);
 499                         RETURN(-abs(rc));
 500                 }
 501                 rc = md_get_lustre_md(sbi->ll_md_exp, req,
 502                                       sbi->ll_dt_exp, sbi->ll_md_exp, &md);
 503
 504                 /* XXX Too paranoid? */
 505                 if (((md.body->valid ^ valid) & OBD_MD_FLEASIZE) &&
 506                     !((md.body->valid & OBD_MD_FLNLINK) &&
 507                       (md.body->nlink == 0))) {
 508                         CERROR("Asked for %s eadata but got %s (%d)\n",
 509                                (valid & OBD_MD_FLEASIZE) ? "some" : "no",
 510                                (md.body->valid & OBD_MD_FLEASIZE) ? "some":"none",
 511                                 md.body->eadatasize);
 512                 }
 513                 if (rc) {
 514                         ptlrpc_req_finished(req);
 515                         RETURN(rc);
 516                 }
 517
 518
 519                 llu_update_inode(inode, &md);
 520                 if (md.lsm != NULL && lli->lli_smd != md.lsm)
 521                         obd_free_memmd(sbi->ll_dt_exp, &md.lsm);
 522                 ptlrpc_req_finished(req);
 523         }
 524
 525         if (!lli->lli_smd) {
 526                 /* object not yet allocated, don't validate size */
 527                 st->st_atime = lli->lli_lvb.lvb_atime;
 528                 st->st_mtime = lli->lli_lvb.lvb_mtime;
 529                 st->st_ctime = lli->lli_lvb.lvb_ctime;
 530                 RETURN(0);
 531         }
 532
 533         /* ll_glimpse_size will prefer locally cached writes if they extend
 534          * the file */
 535         RETURN(cl_glimpse_size(inode));
 536 }
 537
 538 static void copy_stat_buf(struct inode *ino, struct intnl_stat *b)
 539 {
 540         *b = *llu_i2stat(ino);
 541 }
 542
 543 static int llu_iop_getattr(struct pnode *pno,
 544                            struct inode *ino,
 545                            struct intnl_stat *b)
 546 {
 547         int rc;
 548         ENTRY;
 549
 550         liblustre_wait_event(0);
 551
 552         if (!ino) {
 553                 LASSERT(pno);
 554                 LASSERT(pno->p_base->pb_ino);
 555                 ino = pno->p_base->pb_ino;
 556         } else {
 557                 LASSERT(!pno || pno->p_base->pb_ino == ino);
 558         }
 559
 560         /* libsysio might call us directly without intent lock,
 561          * we must re-fetch the attrs here
 562          */
 563         rc = llu_inode_revalidate(ino);
 564         if (!rc) {
 565                 copy_stat_buf(ino, b);
 566                 LASSERT(!llu_i2info(ino)->lli_it);
 567         }
 568
 569         liblustre_wait_event(0);
 570         RETURN(rc);
 571 }
 572
 573 static int null_if_equal(struct ldlm_lock *lock, void *data)
 574 {
 575         if (data == lock->l_ast_data) {
 576                 lock->l_ast_data = NULL;
 577
 578                 if (lock->l_req_mode != lock->l_granted_mode)
 579                         LDLM_ERROR(lock,"clearing inode with ungranted lock\n");
 580         }
 581
 582         return LDLM_ITER_CONTINUE;
 583 }
 584
 585 void llu_clear_inode(struct inode *inode)
 586 {
 587         struct llu_inode_info *lli = llu_i2info(inode);
 588         struct llu_sb_info *sbi = llu_i2sbi(inode);
 589         ENTRY;
 590
 591         CDEBUG(D_VFSTRACE, "VFS Op:inode=%llu/%lu(%p)\n",
 592                (long long)llu_i2stat(inode)->st_ino, lli->lli_st_generation,
 593                inode);
 594
 595         lli->lli_flags &= ~LLIF_MDS_SIZE_LOCK;
 596         md_change_cbdata(sbi->ll_md_exp, ll_inode2fid(inode),
 597                          null_if_equal, inode);
 598
 599         if (lli->lli_smd)
 600                 obd_change_cbdata(sbi->ll_dt_exp, lli->lli_smd,
 601                                   null_if_equal, inode);
 602
 603         cl_inode_fini(inode);
 604
 605         if (lli->lli_smd) {
 606                 obd_free_memmd(sbi->ll_dt_exp, &lli->lli_smd);
 607                 lli->lli_smd = NULL;
 608         }
 609
 610         if (lli->lli_symlink_name) {
 611                 OBD_FREE(lli->lli_symlink_name,
 612                          strlen(lli->lli_symlink_name) + 1);
 613                 lli->lli_symlink_name = NULL;
 614         }
 615
 616         EXIT;
 617 }
 618
 619 void llu_iop_gone(struct inode *inode)
 620 {
 621         struct llu_inode_info *lli = llu_i2info(inode);
 622         ENTRY;
 623
 624         liblustre_wait_event(0);
 625         llu_clear_inode(inode);
 626
 627         OBD_FREE(lli, sizeof(*lli));
 628         EXIT;
 629 }
 630
 631 static int inode_setattr(struct inode * inode, struct iattr * attr)
 632 {
 633         unsigned int ia_valid = attr->ia_valid;
 634         struct intnl_stat *st = llu_i2stat(inode);
 635         int error = 0;
 636
 637         /*
 638          * inode_setattr() is only ever invoked with ATTR_SIZE (by
 639          * llu_setattr_raw()) when file has no bodies. Check this.
 640          */
 641         LASSERT(ergo(ia_valid & ATTR_SIZE, llu_i2info(inode)->lli_smd == NULL));
 642
 643         if (ia_valid & ATTR_SIZE)
 644                 st->st_size = attr->ia_size;
 645         if (ia_valid & ATTR_UID)
 646                 st->st_uid = attr->ia_uid;
 647         if (ia_valid & ATTR_GID)
 648                 st->st_gid = attr->ia_gid;
 649         if (ia_valid & ATTR_ATIME)
 650                 st->st_atime = attr->ia_atime;
 651         if (ia_valid & ATTR_MTIME)
 652                 st->st_mtime = attr->ia_mtime;
 653         if (ia_valid & ATTR_CTIME)
 654                 st->st_ctime = attr->ia_ctime;
 655         if (ia_valid & ATTR_MODE) {
 656                 st->st_mode = attr->ia_mode;
 657                 if (!cfs_curproc_is_in_groups(st->st_gid) &&
 658                     !cfs_capable(CFS_CAP_FSETID))
 659                         st->st_mode &= ~S_ISGID;
 660         }
 661         /* mark_inode_dirty(inode); */
 662         return error;
 663 }
 664
 665 int llu_md_setattr(struct inode *inode, struct md_op_data *op_data,
 666                    struct md_open_data **mod)
 667 {
 668         struct lustre_md md;
 669         struct llu_sb_info *sbi = llu_i2sbi(inode);
 670         struct ptlrpc_request *request = NULL;
 671         int rc;
 672         ENTRY;
 673
 674         llu_prep_md_op_data(op_data, inode, NULL, NULL, 0, 0, LUSTRE_OPC_ANY);
 675         rc = md_setattr(sbi->ll_md_exp, op_data, NULL, 0, NULL,
 676                         0, &request, mod);
 677
 678         if (rc) {
 679                 ptlrpc_req_finished(request);
 680                 if (rc != -EPERM && rc != -EACCES)
 681                         CERROR("md_setattr fails: rc = %d\n", rc);
 682                 RETURN(rc);
 683         }
 684
 685         rc = md_get_lustre_md(sbi->ll_md_exp, request,
 686                               sbi->ll_dt_exp, sbi->ll_md_exp, &md);
 687         if (rc) {
 688                 ptlrpc_req_finished(request);
 689                 RETURN(rc);
 690         }
 691
 692         /* We call inode_setattr to adjust timestamps.
 693          * If there is at least some data in file, we cleared ATTR_SIZE
 694          * above to avoid invoking vmtruncate, otherwise it is important
 695          * to call vmtruncate in inode_setattr to update inode->i_size
 696          * (bug 6196) */
 697         inode_setattr(inode, &op_data->op_attr);
 698         llu_update_inode(inode, &md);
 699         ptlrpc_req_finished(request);
 700
 701         RETURN(rc);
 702 }
 703
 704 /* Close IO epoch and send Size-on-MDS attribute update. */
 705 static int llu_setattr_done_writing(struct inode *inode,
 706                                     struct md_op_data *op_data,
 707                                     struct md_open_data *mod)
 708 {
 709         struct llu_inode_info *lli = llu_i2info(inode);
 710         struct intnl_stat *st = llu_i2stat(inode);
 711         int rc = 0;
 712         ENTRY;
 713
 714         LASSERT(op_data != NULL);
 715         if (!S_ISREG(st->st_mode))
 716                 RETURN(0);
 717
 718         /* XXX: pass och here for the recovery purpose. */
 719         CDEBUG(D_INODE, "Epoch "LPU64" closed on "DFID" for truncate\n",
 720                op_data->op_ioepoch, PFID(&lli->lli_fid));
 721
 722         op_data->op_flags = MF_EPOCH_CLOSE;
 723         llu_done_writing_attr(inode, op_data);
 724         llu_pack_inode2opdata(inode, op_data, NULL);
 725
 726         rc = md_done_writing(llu_i2sbi(inode)->ll_md_exp, op_data, mod);
 727         if (rc == -EAGAIN) {
 728                 /* MDS has instructed us to obtain Size-on-MDS attribute
 729                  * from OSTs and send setattr to back to MDS. */
 730                 rc = llu_som_update(inode, op_data);
 731         } else if (rc) {
 732                 CERROR("inode %llu mdc truncate failed: rc = %d\n",
 733                        (unsigned long long)st->st_ino, rc);
 734         }
 735         RETURN(rc);
 736 }
 737
 738 /* If this inode has objects allocated to it (lsm != NULL), then the OST
 739  * object(s) determine the file size and mtime.  Otherwise, the MDS will
 740  * keep these values until such a time that objects are allocated for it.
 741  * We do the MDS operations first, as it is checking permissions for us.
 742  * We don't to the MDS RPC if there is nothing that we want to store there,
 743  * otherwise there is no harm in updating mtime/atime on the MDS if we are
 744  * going to do an RPC anyways.
 745  *
 746  * If we are doing a truncate, we will send the mtime and ctime updates
 747  * to the OST with the punch RPC, otherwise we do an explicit setattr RPC.
 748  * I don't believe it is possible to get e.g. ATTR_MTIME_SET and ATTR_SIZE
 749  * at the same time.
 750  */
 751 int llu_setattr_raw(struct inode *inode, struct iattr *attr)
 752 {
 753         struct lov_stripe_md *lsm = llu_i2info(inode)->lli_smd;
 754         struct intnl_stat *st = llu_i2stat(inode);
 755         int ia_valid = attr->ia_valid;
 756         struct md_op_data op_data = { { 0 } };
 757         struct md_open_data *mod = NULL;
 758         int rc = 0, rc1 = 0;
 759         ENTRY;
 760
 761         CDEBUG(D_VFSTRACE, "VFS Op:inode=%llu\n", (long long)st->st_ino);
 762
 763         if (ia_valid & ATTR_SIZE) {
 764                 if (attr->ia_size > ll_file_maxbytes(inode)) {
 765                         CDEBUG(D_INODE, "file too large %llu > "LPU64"\n",
 766                                (long long)attr->ia_size,
 767                                ll_file_maxbytes(inode));
 768                         RETURN(-EFBIG);
 769                 }
 770
 771                 attr->ia_valid |= ATTR_MTIME | ATTR_CTIME;
 772         }
 773
 774         /* We mark all of the fields "set" so MDS/OST does not re-set them */
 775         if (attr->ia_valid & ATTR_CTIME) {
 776                 attr->ia_ctime = CFS_CURRENT_TIME;
 777                 attr->ia_valid |= ATTR_CTIME_SET;
 778         }
 779         if (!(ia_valid & ATTR_ATIME_SET) && (attr->ia_valid & ATTR_ATIME)) {
 780                 attr->ia_atime = CFS_CURRENT_TIME;
 781                 attr->ia_valid |= ATTR_ATIME_SET;
 782         }
 783         if (!(ia_valid & ATTR_MTIME_SET) && (attr->ia_valid & ATTR_MTIME)) {
 784                 attr->ia_mtime = CFS_CURRENT_TIME;
 785                 attr->ia_valid |= ATTR_MTIME_SET;
 786         }
 787
 788         if (attr->ia_valid & (ATTR_MTIME | ATTR_CTIME))
 789                 CDEBUG(D_INODE, "setting mtime "CFS_TIME_T", ctime "CFS_TIME_T
 790                        ", now = "CFS_TIME_T"\n",
 791                        LTIME_S(attr->ia_mtime), LTIME_S(attr->ia_ctime),
 792                        LTIME_S(CFS_CURRENT_TIME));
 793
 794         /* NB: ATTR_SIZE will only be set after this point if the size
 795          * resides on the MDS, ie, this file has no objects. */
 796         if (lsm)
 797                 attr->ia_valid &= ~ATTR_SIZE;
 798
 799         /* If only OST attributes being set on objects, don't do MDS RPC.
 800          * In that case, we need to check permissions and update the local
 801          * inode ourselves so we can call obdo_from_inode() always. */
 802         if (ia_valid & (lsm ? ~(ATTR_FROM_OPEN | ATTR_RAW) : ~0)) {
 803                 memcpy(&op_data.op_attr, attr, sizeof(*attr));
 804
 805                 /* Open epoch for truncate. */
 806                 if (exp_connect_som(llu_i2mdexp(inode)) &&
 807                     (ia_valid & ATTR_SIZE))
 808                         op_data.op_flags = MF_EPOCH_OPEN;
 809                 rc = llu_md_setattr(inode, &op_data, &mod);
 810                 if (rc)
 811                         RETURN(rc);
 812
 813                 llu_ioepoch_open(llu_i2info(inode), op_data.op_ioepoch);
 814                 if (!lsm || !S_ISREG(st->st_mode)) {
 815                         CDEBUG(D_INODE, "no lsm: not setting attrs on OST\n");
 816                         GOTO(out, rc);
 817                 }
 818         } else {
 819                 /* The OST doesn't check permissions, but the alternative is
 820                  * a gratuitous RPC to the MDS.  We already rely on the client
 821                  * to do read/write/truncate permission checks, so is mtime OK?
 822                  */
 823                 if (ia_valid & (ATTR_MTIME | ATTR_ATIME)) {
 824                         /* from sys_utime() */
 825                         if (!(ia_valid & (ATTR_MTIME_SET | ATTR_ATIME_SET))) {
 826                                 if (current->fsuid != st->st_uid &&
 827                                     (rc = ll_permission(inode, MAY_WRITE)) != 0)
 828                                         RETURN(rc);
 829                         } else {
 830                                 /* from inode_change_ok() */
 831                                 if (current->fsuid != st->st_uid &&
 832                                     !cfs_capable(CFS_CAP_FOWNER))
 833                                         RETURN(-EPERM);
 834                         }
 835                 }
 836
 837
 838                 /* Won't invoke llu_vmtruncate(), as we already cleared
 839                  * ATTR_SIZE */
 840                 inode_setattr(inode, attr);
 841         }
 842
 843         if (ia_valid & (ATTR_SIZE | ATTR_MTIME | ATTR_MTIME_SET)) {
 844                 /* mtime is set to past sending setattr op to osts under PW
 845                  * 0:EOF extent lock (like truncate under PW new_size:EOF), if
 846                  * mtime is not set to past setattr op is not sent to osts */
 847                 if ((ia_valid & ATTR_SIZE) ||
 848                     LTIME_S(attr->ia_mtime) < LTIME_S(attr->ia_ctime))
 849                         rc = cl_setattr_ost(inode, attr, NULL);
 850         }
 851         EXIT;
 852 out:
 853         if (op_data.op_ioepoch)
 854                 rc1 = llu_setattr_done_writing(inode, &op_data, mod);
 855         return rc ? rc : rc1;
 856 }
 857
 858 /* here we simply act as a thin layer to glue it with
 859  * llu_setattr_raw(), which is copy from kernel
 860  */
 861 static int llu_iop_setattr(struct pnode *pno,
 862                            struct inode *ino,
 863                            unsigned mask,
 864                            struct intnl_stat *stbuf)
 865 {
 866         struct iattr iattr;
 867         int rc;
 868         ENTRY;
 869
 870         liblustre_wait_event(0);
 871
 872         LASSERT(!(mask & ~(SETATTR_MTIME | SETATTR_ATIME |
 873                            SETATTR_UID | SETATTR_GID |
 874                            SETATTR_LEN | SETATTR_MODE)));
 875         memset(&iattr, 0, sizeof(iattr));
 876
 877         if (mask & SETATTR_MODE) {
 878                 iattr.ia_mode = stbuf->st_mode;
 879                 iattr.ia_valid |= ATTR_MODE;
 880         }
 881         if (mask & SETATTR_MTIME) {
 882                 iattr.ia_mtime = stbuf->st_mtime;
 883                 iattr.ia_valid |= ATTR_MTIME | ATTR_MTIME_SET;
 884         }
 885         if (mask & SETATTR_ATIME) {
 886                 iattr.ia_atime = stbuf->st_atime;
 887                 iattr.ia_valid |= ATTR_ATIME | ATTR_ATIME_SET;
 888         }
 889         if (mask & SETATTR_UID) {
 890                 iattr.ia_uid = stbuf->st_uid;
 891                 iattr.ia_valid |= ATTR_UID;
 892         }
 893         if (mask & SETATTR_GID) {
 894                 iattr.ia_gid = stbuf->st_gid;
 895                 iattr.ia_valid |= ATTR_GID;
 896         }
 897         if (mask & SETATTR_LEN) {
 898                 iattr.ia_size = stbuf->st_size; /* XXX signed expansion problem */
 899                 iattr.ia_valid |= ATTR_SIZE;
 900         }
 901
 902         iattr.ia_valid |= ATTR_RAW | ATTR_CTIME;
 903         iattr.ia_ctime = CFS_CURRENT_TIME;
 904
 905         rc = llu_setattr_raw(ino, &iattr);
 906         liblustre_wait_idle();
 907         RETURN(rc);
 908 }
 909
 910 #define EXT2_LINK_MAX           32000
 911
 912 static int llu_iop_symlink_raw(struct pnode *pno, const char *tgt)
 913 {
 914         struct inode *dir = pno->p_base->pb_parent->pb_ino;
 915         struct qstr *qstr = &pno->p_base->pb_name;
 916         const char *name = qstr->name;
 917         int len = qstr->len;
 918         struct ptlrpc_request *request = NULL;
 919         struct llu_sb_info *sbi = llu_i2sbi(dir);
 920         struct md_op_data op_data = {{ 0 }};
 921         int err = -EMLINK;
 922         ENTRY;
 923
 924         liblustre_wait_event(0);
 925         if (llu_i2stat(dir)->st_nlink >= EXT2_LINK_MAX)
 926                 RETURN(err);
 927
 928         llu_prep_md_op_data(&op_data, dir, NULL, name, len, 0,
 929                             LUSTRE_OPC_SYMLINK);
 930
 931         err = md_create(sbi->ll_md_exp, &op_data, tgt, strlen(tgt) + 1,
 932                         S_IFLNK | S_IRWXUGO, current->fsuid, current->fsgid,
 933                         cfs_curproc_cap_pack(), 0, &request);
 934         ptlrpc_req_finished(request);
 935         liblustre_wait_event(0);
 936         RETURN(err);
 937 }
 938
 939 static int llu_readlink_internal(struct inode *inode,
 940                                  struct ptlrpc_request **request,
 941                                  char **symname)
 942 {
 943         struct llu_inode_info *lli = llu_i2info(inode);
 944         struct llu_sb_info *sbi = llu_i2sbi(inode);
 945         struct mdt_body *body;
 946         struct intnl_stat *st = llu_i2stat(inode);
 947         struct md_op_data op_data = {{ 0 }};
 948         int rc, symlen = st->st_size + 1;
 949         ENTRY;
 950
 951         *request = NULL;
 952         *symname = NULL;
 953
 954         if (lli->lli_symlink_name) {
 955                 *symname = lli->lli_symlink_name;
 956                 CDEBUG(D_INODE, "using cached symlink %s\n", *symname);
 957                 RETURN(0);
 958         }
 959
 960         llu_prep_md_op_data(&op_data, inode, NULL, NULL, 0, symlen,
 961                             LUSTRE_OPC_ANY);
 962         op_data.op_valid = OBD_MD_LINKNAME;
 963
 964         rc = md_getattr(sbi->ll_md_exp, &op_data, request);
 965         if (rc) {
 966                 CERROR("inode %llu: rc = %d\n", (long long)st->st_ino, rc);
 967                 RETURN(rc);
 968         }
 969
 970         body = req_capsule_server_get(&(*request)->rq_pill, &RMF_MDT_BODY);
 971         LASSERT(body != NULL);
 972
 973         if ((body->valid & OBD_MD_LINKNAME) == 0) {
 974                 CERROR ("OBD_MD_LINKNAME not set on reply\n");
 975                 GOTO (failed, rc = -EPROTO);
 976         }
 977
 978         LASSERT(symlen != 0);
 979         if (body->eadatasize != symlen) {
 980                 CERROR("inode %llu: symlink length %d not expected %d\n",
 981                        (long long)st->st_ino, body->eadatasize - 1, symlen - 1);
 982                 GOTO(failed, rc = -EPROTO);
 983         }
 984
 985         *symname = req_capsule_server_get(&(*request)->rq_pill, &RMF_MDT_MD);
 986         if (*symname == NULL ||
 987             strnlen(*symname, symlen) != symlen - 1) {
 988                 /* not full/NULL terminated */
 989                 CERROR("inode %llu: symlink not NULL terminated string"
 990                        "of length %d\n", (long long)st->st_ino, symlen - 1);
 991                 GOTO(failed, rc = -EPROTO);
 992         }
 993
 994         OBD_ALLOC(lli->lli_symlink_name, symlen);
 995         /* do not return an error if we cannot cache the symlink locally */
 996         if (lli->lli_symlink_name)
 997                 memcpy(lli->lli_symlink_name, *symname, symlen);
 998
 999         RETURN(0);
1000
1001  failed:
1002         ptlrpc_req_finished (*request);
1003         RETURN (-EPROTO);
1004 }
1005
1006 static int llu_iop_readlink(struct pnode *pno, char *data, size_t bufsize)
1007 {
1008         struct inode *inode = pno->p_base->pb_ino;
1009         struct ptlrpc_request *request;
1010         char *symname;
1011         int rc;
1012         ENTRY;
1013
1014         liblustre_wait_event(0);
1015         rc = llu_readlink_internal(inode, &request, &symname);
1016         if (rc)
1017                 GOTO(out, rc);
1018
1019         LASSERT(symname);
1020         strncpy(data, symname, bufsize);
1021         rc = strlen(symname);
1022
1023         ptlrpc_req_finished(request);
1024  out:
1025         liblustre_wait_event(0);
1026         RETURN(rc);
1027 }
1028
1029 static int llu_iop_mknod_raw(struct pnode *pno,
1030                              mode_t mode,
1031                              dev_t dev)
1032 {
1033         struct ptlrpc_request *request = NULL;
1034         struct inode *dir = pno->p_parent->p_base->pb_ino;
1035         struct llu_sb_info *sbi = llu_i2sbi(dir);
1036         struct md_op_data op_data = {{ 0 }};
1037         int err = -EMLINK;
1038         ENTRY;
1039
1040         liblustre_wait_event(0);
1041         CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s,dir=%llu\n",
1042                (int)pno->p_base->pb_name.len, pno->p_base->pb_name.name,
1043                (long long)llu_i2stat(dir)->st_ino);
1044
1045         if (llu_i2stat(dir)->st_nlink >= EXT2_LINK_MAX)
1046                 RETURN(err);
1047
1048         switch (mode & S_IFMT) {
1049         case 0:
1050         case S_IFREG:
1051                 mode |= S_IFREG; /* for mode = 0 case, fallthrough */
1052         case S_IFCHR:
1053         case S_IFBLK:
1054         case S_IFIFO:
1055         case S_IFSOCK:
1056                 llu_prep_md_op_data(&op_data, dir, NULL,
1057                                     pno->p_base->pb_name.name,
1058                                     pno->p_base->pb_name.len, 0,
1059                                     LUSTRE_OPC_MKNOD);
1060
1061                 err = md_create(sbi->ll_md_exp, &op_data, NULL, 0, mode,
1062                                 current->fsuid, current->fsgid,
1063                                 cfs_curproc_cap_pack(), dev, &request);
1064                 ptlrpc_req_finished(request);
1065                 break;
1066         case S_IFDIR:
1067                 err = -EPERM;
1068                 break;
1069         default:
1070                 err = -EINVAL;
1071         }
1072         liblustre_wait_event(0);
1073         RETURN(err);
1074 }
1075
1076 static int llu_iop_link_raw(struct pnode *old, struct pnode *new)
1077 {
1078         struct inode *src = old->p_base->pb_ino;
1079         struct inode *dir = new->p_parent->p_base->pb_ino;
1080         const char *name = new->p_base->pb_name.name;
1081         int namelen = new->p_base->pb_name.len;
1082         struct ptlrpc_request *request = NULL;
1083         struct md_op_data op_data = {{ 0 }};
1084         int rc;
1085         ENTRY;
1086
1087         LASSERT(src);
1088         LASSERT(dir);
1089
1090         liblustre_wait_event(0);
1091         llu_prep_md_op_data(&op_data, src, dir, name, namelen, 0,
1092                             LUSTRE_OPC_ANY);
1093         rc = md_link(llu_i2sbi(src)->ll_md_exp, &op_data, &request);
1094         ptlrpc_req_finished(request);
1095         liblustre_wait_event(0);
1096
1097         RETURN(rc);
1098 }
1099
1100 /*
1101  * libsysio will clear the inode immediately after return
1102  */
1103 static int llu_iop_unlink_raw(struct pnode *pno)
1104 {
1105         struct inode *dir = pno->p_base->pb_parent->pb_ino;
1106         struct qstr *qstr = &pno->p_base->pb_name;
1107         const char *name = qstr->name;
1108         int len = qstr->len;
1109         struct inode *target = pno->p_base->pb_ino;
1110         struct ptlrpc_request *request = NULL;
1111         struct md_op_data op_data = { { 0 } };
1112         int rc;
1113         ENTRY;
1114
1115         LASSERT(target);
1116
1117         liblustre_wait_event(0);
1118         llu_prep_md_op_data(&op_data, dir, NULL, name, len, 0,
1119                             LUSTRE_OPC_ANY);
1120         rc = md_unlink(llu_i2sbi(dir)->ll_md_exp, &op_data, &request);
1121         if (!rc)
1122                 rc = llu_objects_destroy(request, dir);
1123         ptlrpc_req_finished(request);
1124         liblustre_wait_idle();
1125
1126         RETURN(rc);
1127 }
1128
1129 static int llu_iop_rename_raw(struct pnode *old, struct pnode *new)
1130 {
1131         struct inode *src = old->p_parent->p_base->pb_ino;
1132         struct inode *tgt = new->p_parent->p_base->pb_ino;
1133         const char *oldname = old->p_base->pb_name.name;
1134         int oldnamelen = old->p_base->pb_name.len;
1135         const char *newname = new->p_base->pb_name.name;
1136         int newnamelen = new->p_base->pb_name.len;
1137         struct ptlrpc_request *request = NULL;
1138         struct md_op_data op_data = { { 0 } };
1139         int rc;
1140         ENTRY;
1141
1142         LASSERT(src);
1143         LASSERT(tgt);
1144
1145         liblustre_wait_event(0);
1146         llu_prep_md_op_data(&op_data, src, tgt, NULL, 0, 0,
1147                             LUSTRE_OPC_ANY);
1148         rc = md_rename(llu_i2sbi(src)->ll_md_exp, &op_data,
1149                        oldname, oldnamelen, newname, newnamelen,
1150                        &request);
1151         if (!rc) {
1152                 rc = llu_objects_destroy(request, src);
1153         }
1154
1155         ptlrpc_req_finished(request);
1156         liblustre_wait_idle();
1157
1158         RETURN(rc);
1159 }
1160
1161 #ifdef _HAVE_STATVFS
1162 static int llu_statfs_internal(struct llu_sb_info *sbi,
1163                                struct obd_statfs *osfs, __u64 max_age)
1164 {
1165         struct obd_statfs obd_osfs;
1166         int rc;
1167         ENTRY;
1168
1169         rc = obd_statfs(class_exp2obd(sbi->ll_md_exp), osfs, max_age, 0);
1170         if (rc) {
1171                 CERROR("md_statfs fails: rc = %d\n", rc);
1172                 RETURN(rc);
1173         }
1174
1175         CDEBUG(D_SUPER, "MDC blocks "LPU64"/"LPU64" objects "LPU64"/"LPU64"\n",
1176                osfs->os_bavail, osfs->os_blocks, osfs->os_ffree,osfs->os_files);
1177
1178         rc = obd_statfs_rqset(class_exp2obd(sbi->ll_dt_exp),
1179                               &obd_statfs, max_age, 0);
1180         if (rc) {
1181                 CERROR("obd_statfs fails: rc = %d\n", rc);
1182                 RETURN(rc);
1183         }
1184
1185         CDEBUG(D_SUPER, "OSC blocks "LPU64"/"LPU64" objects "LPU64"/"LPU64"\n",
1186                obd_osfs.os_bavail, obd_osfs.os_blocks, obd_osfs.os_ffree,
1187                obd_osfs.os_files);
1188
1189         osfs->os_blocks = obd_osfs.os_blocks;
1190         osfs->os_bfree = obd_osfs.os_bfree;
1191         osfs->os_bavail = obd_osfs.os_bavail;
1192
1193         /* If we don't have as many objects free on the OST as inodes
1194          * on the MDS, we reduce the total number of inodes to
1195          * compensate, so that the "inodes in use" number is correct.
1196          */
1197         if (obd_osfs.os_ffree < osfs->os_ffree) {
1198                 osfs->os_files = (osfs->os_files - osfs->os_ffree) +
1199                         obd_osfs.os_ffree;
1200                 osfs->os_ffree = obd_osfs.os_ffree;
1201         }
1202
1203         RETURN(rc);
1204 }
1205
1206 static int llu_statfs(struct llu_sb_info *sbi, struct statfs *sfs)
1207 {
1208         struct obd_statfs osfs;
1209         int rc;
1210
1211         CDEBUG(D_VFSTRACE, "VFS Op:\n");
1212
1213         /* For now we will always get up-to-date statfs values, but in the
1214          * future we may allow some amount of caching on the client (e.g.
1215          * from QOS or lprocfs updates). */
1216         rc = llu_statfs_internal(sbi, &osfs, cfs_time_current_64() - CFS_HZ);
1217         if (rc)
1218                 return rc;
1219
1220         statfs_unpack(sfs, &osfs);
1221
1222         if (sizeof(sfs->f_blocks) == 4) {
1223                 while (osfs.os_blocks > ~0UL) {
1224                         sfs->f_bsize <<= 1;
1225
1226                         osfs.os_blocks >>= 1;
1227                         osfs.os_bfree >>= 1;
1228                         osfs.os_bavail >>= 1;
1229                 }
1230         }
1231
1232         sfs->f_blocks = osfs.os_blocks;
1233         sfs->f_bfree = osfs.os_bfree;
1234         sfs->f_bavail = osfs.os_bavail;
1235
1236         return 0;
1237 }
1238
1239 static int llu_iop_statvfs(struct pnode *pno,
1240                            struct inode *ino,
1241                            struct intnl_statvfs *buf)
1242 {
1243         struct statfs fs;
1244         int rc;
1245         ENTRY;
1246
1247         liblustre_wait_event(0);
1248
1249 #ifndef __CYGWIN__
1250         LASSERT(pno->p_base->pb_ino);
1251         rc = llu_statfs(llu_i2sbi(pno->p_base->pb_ino), &fs);
1252         if (rc)
1253                 RETURN(rc);
1254
1255         /* from native driver */
1256         buf->f_bsize = fs.f_bsize;  /* file system block size */
1257         buf->f_frsize = fs.f_bsize; /* file system fundamental block size */
1258         buf->f_blocks = fs.f_blocks;
1259         buf->f_bfree = fs.f_bfree;
1260         buf->f_bavail = fs.f_bavail;
1261         buf->f_files = fs.f_files;  /* Total number serial numbers */
1262         buf->f_ffree = fs.f_ffree;  /* Number free serial numbers */
1263         buf->f_favail = fs.f_ffree; /* Number free ser num for non-privileged*/
1264         buf->f_fsid = fs.f_fsid.__val[1];
1265         buf->f_flag = 0;            /* No equiv in statfs; maybe use type? */
1266         buf->f_namemax = fs.f_namelen;
1267 #endif
1268
1269         liblustre_wait_event(0);
1270         RETURN(0);
1271 }
1272 #endif /* _HAVE_STATVFS */
1273
1274 static int llu_iop_mkdir_raw(struct pnode *pno, mode_t mode)
1275 {
1276         struct inode *dir = pno->p_base->pb_parent->pb_ino;
1277         struct qstr *qstr = &pno->p_base->pb_name;
1278         const char *name = qstr->name;
1279         int len = qstr->len;
1280         struct ptlrpc_request *request = NULL;
1281         struct intnl_stat *st = llu_i2stat(dir);
1282         struct md_op_data op_data = {{ 0 }};
1283         int err = -EMLINK;
1284         ENTRY;
1285
1286         liblustre_wait_event(0);
1287         CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s,dir=%llu/%lu(%p)\n", len, name,
1288                (long long)st->st_ino, llu_i2info(dir)->lli_st_generation, dir);
1289
1290         if (st->st_nlink >= EXT2_LINK_MAX)
1291                 RETURN(err);
1292
1293         llu_prep_md_op_data(&op_data, dir, NULL, name, len, 0,
1294                             LUSTRE_OPC_MKDIR);
1295
1296         err = md_create(llu_i2sbi(dir)->ll_md_exp, &op_data, NULL, 0,
1297                         mode | S_IFDIR, current->fsuid, current->fsgid,
1298                         cfs_curproc_cap_pack(), 0, &request);
1299         ptlrpc_req_finished(request);
1300         liblustre_wait_event(0);
1301         RETURN(err);
1302 }
1303
1304 static int llu_iop_rmdir_raw(struct pnode *pno)
1305 {
1306         struct inode *dir = pno->p_base->pb_parent->pb_ino;
1307         struct qstr *qstr = &pno->p_base->pb_name;
1308         const char *name = qstr->name;
1309         int len = qstr->len;
1310         struct ptlrpc_request *request = NULL;
1311         struct md_op_data op_data = {{ 0 }};
1312         int rc;
1313         ENTRY;
1314
1315         liblustre_wait_event(0);
1316         CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s,dir=%llu/%lu(%p)\n", len, name,
1317                (long long)llu_i2stat(dir)->st_ino,
1318                llu_i2info(dir)->lli_st_generation, dir);
1319
1320         llu_prep_md_op_data(&op_data, dir, NULL, name, len, S_IFDIR,
1321                             LUSTRE_OPC_ANY);
1322         rc = md_unlink(llu_i2sbi(dir)->ll_md_exp, &op_data, &request);
1323         ptlrpc_req_finished(request);
1324
1325         liblustre_wait_event(0);
1326         RETURN(rc);
1327 }
1328
1329 #ifdef O_DIRECT
1330 #define FCNTL_FLMASK (O_APPEND|O_NONBLOCK|O_ASYNC|O_DIRECT)
1331 #else
1332 #define FCNTL_FLMASK (O_APPEND|O_NONBLOCK|O_ASYNC)
1333 #endif
1334 #define FCNTL_FLMASK_INVALID (O_NONBLOCK|O_ASYNC)
1335
1336 /* refer to ll_file_flock() for details */
1337 static int llu_file_flock(struct inode *ino,
1338                           int cmd,
1339                           struct file_lock *file_lock)
1340 {
1341         struct llu_inode_info *lli = llu_i2info(ino);
1342         struct intnl_stat *st = llu_i2stat(ino);
1343         struct ldlm_res_id res_id =
1344                 { .name = {fid_seq(&lli->lli_fid),
1345                            fid_oid(&lli->lli_fid),
1346                            fid_ver(&lli->lli_fid),
1347                            LDLM_FLOCK} };
1348         struct ldlm_enqueue_info einfo = { LDLM_FLOCK, 0, NULL,
1349                 ldlm_flock_completion_ast, NULL, NULL, file_lock };
1350
1351         struct lustre_handle lockh = {0};
1352         ldlm_policy_data_t flock;
1353         int flags = 0;
1354         int rc;
1355
1356         CDEBUG(D_VFSTRACE, "VFS Op:inode=%llu file_lock=%p\n",
1357                (unsigned long long)st->st_ino, file_lock);
1358
1359         flock.l_flock.pid = file_lock->fl_pid;
1360         flock.l_flock.start = file_lock->fl_start;
1361         flock.l_flock.end = file_lock->fl_end;
1362
1363         switch (file_lock->fl_type) {
1364         case F_RDLCK:
1365                 einfo.ei_mode = LCK_PR;
1366                 break;
1367         case F_UNLCK:
1368                 einfo.ei_mode = LCK_NL;
1369                 break;
1370         case F_WRLCK:
1371                 einfo.ei_mode = LCK_PW;
1372                 break;
1373         default:
1374                 CERROR("unknown fcntl lock type: %d\n", file_lock->fl_type);
1375                 LBUG();
1376         }
1377
1378         switch (cmd) {
1379         case F_SETLKW:
1380 #ifdef F_SETLKW64
1381 #if F_SETLKW64 != F_SETLKW
1382         case F_SETLKW64:
1383 #endif
1384 #endif
1385                 flags = 0;
1386                 break;
1387         case F_SETLK:
1388 #ifdef F_SETLK64
1389 #if F_SETLK64 != F_SETLK
1390         case F_SETLK64:
1391 #endif
1392 #endif
1393                 flags = LDLM_FL_BLOCK_NOWAIT;
1394                 break;
1395         case F_GETLK:
1396 #ifdef F_GETLK64
1397 #if F_GETLK64 != F_GETLK
1398         case F_GETLK64:
1399 #endif
1400 #endif
1401                 flags = LDLM_FL_TEST_LOCK;
1402                 file_lock->fl_type = einfo.ei_mode;
1403                 break;
1404         default:
1405                 CERROR("unknown fcntl cmd: %d\n", cmd);
1406                 LBUG();
1407         }
1408
1409         CDEBUG(D_DLMTRACE, "inode=%llu, pid=%u, cmd=%d, flags=%#x, mode=%u, "
1410                "start="LPX64", end="LPX64"\n", (unsigned long long)st->st_ino,
1411                flock.l_flock.pid, cmd, flags, einfo.ei_mode, flock.l_flock.start,
1412                flock.l_flock.end);
1413
1414         {
1415                 struct lmv_obd *lmv;
1416                 struct obd_device *lmv_obd;
1417                 lmv_obd = class_exp2obd(llu_i2mdexp(ino));
1418                 lmv = &lmv_obd->u.lmv;
1419
1420                 if (lmv->desc.ld_tgt_count < 1)
1421                         RETURN(rc = -ENODEV);
1422
1423                 if (lmv->tgts[0].ltd_exp != NULL)
1424                         rc = ldlm_cli_enqueue(lmv->tgts[0].ltd_exp, NULL, &einfo, &res_id,
1425                                               &flock, &flags, NULL, 0, &lockh, 0);
1426                 else
1427                         rc = -ENODEV;
1428         }
1429         RETURN(rc);
1430 }
1431
1432 static int assign_type(struct file_lock *fl, int type)
1433 {
1434         switch (type) {
1435         case F_RDLCK:
1436         case F_WRLCK:
1437         case F_UNLCK:
1438                 fl->fl_type = type;
1439                 return 0;
1440         default:
1441                 return -EINVAL;
1442         }
1443 }
1444
1445 static int flock_to_posix_lock(struct inode *ino,
1446                                struct file_lock *fl,
1447                                struct flock *l)
1448 {
1449         switch (l->l_whence) {
1450         /* XXX: only SEEK_SET is supported in lustre */
1451         case SEEK_SET:
1452                 fl->fl_start = 0;
1453                 break;
1454         default:
1455                 return -EINVAL;
1456         }
1457
1458         fl->fl_end = l->l_len - 1;
1459         if (l->l_len < 0)
1460                 return -EINVAL;
1461         if (l->l_len == 0)
1462                 fl->fl_end = OFFSET_MAX;
1463
1464         fl->fl_pid = getpid();
1465         fl->fl_flags = FL_POSIX;
1466         fl->fl_notify = NULL;
1467         fl->fl_insert = NULL;
1468         fl->fl_remove = NULL;
1469         /* XXX: these fields can't be filled with suitable values,
1470                 but I think lustre doesn't use them.
1471          */
1472         fl->fl_owner = NULL;
1473         fl->fl_file = NULL;
1474
1475         return assign_type(fl, l->l_type);
1476 }
1477
1478 static int llu_fcntl_getlk(struct inode *ino, struct flock *flock)
1479 {
1480         struct file_lock fl;
1481         int error;
1482
1483         error = EINVAL;
1484         if ((flock->l_type != F_RDLCK) && (flock->l_type != F_WRLCK))
1485                 goto out;
1486
1487         error = flock_to_posix_lock(ino, &fl, flock);
1488         if (error)
1489                 goto out;
1490
1491         error = llu_file_flock(ino, F_GETLK, &fl);
1492         if (error)
1493                 goto out;
1494
1495         flock->l_type = F_UNLCK;
1496         if (fl.fl_type != F_UNLCK) {
1497                 flock->l_pid = fl.fl_pid;
1498                 flock->l_start = fl.fl_start;
1499                 flock->l_len = fl.fl_end == OFFSET_MAX ? 0:
1500                         fl.fl_end - fl.fl_start + 1;
1501                 flock->l_whence = SEEK_SET;
1502                 flock->l_type = fl.fl_type;
1503         }
1504
1505 out:
1506         return error;
1507 }
1508
1509 static int llu_fcntl_setlk(struct inode *ino, int cmd, struct flock *flock)
1510 {
1511         struct file_lock fl;
1512         int flags = llu_i2info(ino)->lli_open_flags + 1;
1513         int error;
1514
1515         error = flock_to_posix_lock(ino, &fl, flock);
1516         if (error)
1517                 goto out;
1518         if (cmd == F_SETLKW)
1519                 fl.fl_flags |= FL_SLEEP;
1520
1521         error = -EBADF;
1522         switch (flock->l_type) {
1523         case F_RDLCK:
1524                 if (!(flags & FMODE_READ))
1525                         goto out;
1526                 break;
1527         case F_WRLCK:
1528                 if (!(flags & FMODE_WRITE))
1529                         goto out;
1530                 break;
1531         case F_UNLCK:
1532                 break;
1533         default:
1534                 error = -EINVAL;
1535                 goto out;
1536         }
1537
1538         error = llu_file_flock(ino, cmd, &fl);
1539         if (error)
1540                 goto out;
1541
1542 out:
1543         return error;
1544 }
1545
1546 static int llu_iop_fcntl(struct inode *ino, int cmd, va_list ap, int *rtn)
1547 {
1548         struct llu_inode_info *lli = llu_i2info(ino);
1549         long flags;
1550         struct flock *flock;
1551         long err = 0;
1552
1553         liblustre_wait_event(0);
1554         switch (cmd) {
1555         case F_GETFL:
1556                 *rtn = lli->lli_open_flags;
1557                 break;
1558         case F_SETFL:
1559                 flags = va_arg(ap, long);
1560                 flags &= FCNTL_FLMASK;
1561                 if (flags & FCNTL_FLMASK_INVALID) {
1562                         LCONSOLE_ERROR_MSG(0x010, "liblustre does not support "
1563                                            "the O_NONBLOCK or O_ASYNC flags. "
1564                                            "Please fix your application.\n");
1565                         *rtn = -EINVAL;
1566                         err = EINVAL;
1567                         break;
1568                 }
1569                 lli->lli_open_flags = (int)(flags & FCNTL_FLMASK) |
1570                                       (lli->lli_open_flags & ~FCNTL_FLMASK);
1571                 *rtn = 0;
1572                 break;
1573         case F_GETLK:
1574 #ifdef F_GETLK64
1575 #if F_GETLK64 != F_GETLK
1576         case F_GETLK64:
1577 #endif
1578 #endif
1579                 flock = va_arg(ap, struct flock *);
1580                 err = llu_fcntl_getlk(ino, flock);
1581                 *rtn = err? -1: 0;
1582                 break;
1583         case F_SETLK:
1584 #ifdef F_SETLKW64
1585 #if F_SETLKW64 != F_SETLKW
1586         case F_SETLKW64:
1587 #endif
1588 #endif
1589         case F_SETLKW:
1590 #ifdef F_SETLK64
1591 #if F_SETLK64 != F_SETLK
1592         case F_SETLK64:
1593 #endif
1594 #endif
1595                 flock = va_arg(ap, struct flock *);
1596                 err = llu_fcntl_setlk(ino, cmd, flock);
1597                 *rtn = err? -1: 0;
1598                 break;
1599         default:
1600                 CERROR("unsupported fcntl cmd %x\n", cmd);
1601                 *rtn = -ENOSYS;
1602                 err = ENOSYS;
1603                 break;
1604         }
1605
1606         liblustre_wait_event(0);
1607         return err;
1608 }
1609
1610 static int llu_get_grouplock(struct inode *inode, unsigned long arg)
1611 {
1612         struct llu_inode_info *lli = llu_i2info(inode);
1613         struct ll_file_data *fd = lli->lli_file_data;
1614         int rc;
1615         struct ccc_grouplock grouplock;
1616         ENTRY;
1617
1618         if (fd->fd_flags & LL_FILE_IGNORE_LOCK) {
1619                 RETURN(-ENOTSUPP);
1620         }
1621         if (fd->fd_flags & LL_FILE_GROUP_LOCKED) {
1622                 RETURN(-EINVAL);
1623         }
1624         LASSERT(fd->fd_grouplock.cg_lock == NULL);
1625
1626         rc = cl_get_grouplock(cl_i2info(inode)->lli_clob,
1627                               arg, (lli->lli_open_flags & O_NONBLOCK),
1628                               &grouplock);
1629
1630         if (rc)
1631                 RETURN(rc);
1632
1633         fd->fd_flags |= LL_FILE_GROUP_LOCKED;
1634         fd->fd_grouplock = grouplock;
1635
1636         RETURN(0);
1637 }
1638
1639 int llu_put_grouplock(struct inode *inode, unsigned long arg)
1640 {
1641         struct llu_inode_info *lli = llu_i2info(inode);
1642         struct ll_file_data *fd = lli->lli_file_data;
1643         struct ccc_grouplock grouplock;
1644         ENTRY;
1645
1646         if (!(fd->fd_flags & LL_FILE_GROUP_LOCKED))
1647                 RETURN(-EINVAL);
1648
1649         LASSERT(fd->fd_grouplock.cg_lock != NULL);
1650
1651         if (fd->fd_grouplock.cg_gid != arg)
1652                 RETURN(-EINVAL);
1653
1654         grouplock = fd->fd_grouplock;
1655         memset(&fd->fd_grouplock, 0, sizeof(fd->fd_grouplock));
1656         fd->fd_flags &= ~LL_FILE_GROUP_LOCKED;
1657
1658         cl_put_grouplock(&grouplock);
1659
1660         RETURN(0);
1661 }
1662
1663 static int llu_lov_dir_setstripe(struct inode *ino, unsigned long arg)
1664 {
1665         struct llu_sb_info *sbi = llu_i2sbi(ino);
1666         struct ptlrpc_request *request = NULL;
1667         struct md_op_data op_data = {{ 0 }};
1668         struct lov_user_md lum, *lump = (struct lov_user_md *)arg;
1669         int rc = 0;
1670
1671         llu_prep_md_op_data(&op_data, ino, NULL, NULL, 0, 0,
1672                             LUSTRE_OPC_ANY);
1673
1674         LASSERT(sizeof(lum) == sizeof(*lump));
1675         LASSERT(sizeof(lum.lmm_objects[0]) ==
1676                 sizeof(lump->lmm_objects[0]));
1677         if (cfs_copy_from_user(&lum, lump, sizeof(lum)))
1678                 return(-EFAULT);
1679
1680         switch (lum.lmm_magic) {
1681         case LOV_USER_MAGIC_V1: {
1682                 if (lum.lmm_magic != cpu_to_le32(LOV_USER_MAGIC_V1))
1683                         lustre_swab_lov_user_md_v1(&lum);
1684                 break;
1685                 }
1686         case LOV_USER_MAGIC_V3: {
1687                 if (lum.lmm_magic != cpu_to_le32(LOV_USER_MAGIC_V3))
1688                         lustre_swab_lov_user_md_v3((struct lov_user_md_v3 *)&lum);
1689                 break;
1690                 }
1691         default: {
1692                 CDEBUG(D_IOCTL, "bad userland LOV MAGIC:"
1693                                 " %#08x != %#08x nor %#08x\n",
1694                                 lum.lmm_magic, LOV_USER_MAGIC_V1,
1695                                 LOV_USER_MAGIC_V3);
1696                 RETURN(-EINVAL);
1697         }
1698         }
1699
1700         /* swabbing is done in lov_setstripe() on server side */
1701         rc = md_setattr(sbi->ll_md_exp, &op_data, &lum,
1702                         sizeof(lum), NULL, 0, &request, NULL);
1703         if (rc) {
1704                 ptlrpc_req_finished(request);
1705                 if (rc != -EPERM && rc != -EACCES)
1706                         CERROR("md_setattr fails: rc = %d\n", rc);
1707                 return rc;
1708         }
1709         ptlrpc_req_finished(request);
1710
1711         return rc;
1712 }
1713
1714 static int llu_lov_setstripe_ea_info(struct inode *ino, int flags,
1715                                      struct lov_user_md *lum, int lum_size)
1716 {
1717         struct llu_sb_info *sbi = llu_i2sbi(ino);
1718         struct llu_inode_info *lli = llu_i2info(ino);
1719         struct lookup_intent oit = {.it_op = IT_OPEN, .it_flags = flags};
1720         struct ldlm_enqueue_info einfo = { LDLM_IBITS, LCK_CR,
1721                 llu_md_blocking_ast, ldlm_completion_ast, NULL, NULL, NULL };
1722         struct ptlrpc_request *req = NULL;
1723         struct lustre_md md;
1724         struct md_op_data data = {{ 0 }};
1725         struct lustre_handle lockh;
1726         int rc = 0;
1727         ENTRY;
1728
1729         if (lli->lli_smd) {
1730                 CDEBUG(D_IOCTL, "stripe already exists for ino "DFID"\n",
1731                        PFID(&lli->lli_fid));
1732                 return -EEXIST;
1733         }
1734
1735         llu_prep_md_op_data(&data, NULL, ino, NULL, 0, O_RDWR,
1736                             LUSTRE_OPC_ANY);
1737         rc = md_enqueue(sbi->ll_md_exp, &einfo, &oit, &data,
1738                         &lockh, lum, lum_size, NULL, LDLM_FL_INTENT_ONLY);
1739         if (rc)
1740                 GOTO(out, rc);
1741
1742         req = oit.d.lustre.it_data;
1743         rc = it_open_error(DISP_IT_EXECD, &oit);
1744         if (rc) {
1745                 req->rq_replay = 0;
1746                 GOTO(out, rc);
1747         }
1748
1749         rc = it_open_error(DISP_OPEN_OPEN, &oit);
1750         if (rc) {
1751                 req->rq_replay = 0;
1752                 GOTO(out, rc);
1753         }
1754
1755         rc = md_get_lustre_md(sbi->ll_md_exp, req,
1756                               sbi->ll_dt_exp, sbi->ll_md_exp, &md);
1757         if (rc)
1758                 GOTO(out, rc);
1759
1760         llu_update_inode(ino, &md);
1761         llu_local_open(lli, &oit);
1762         /* release intent */
1763         if (lustre_handle_is_used(&lockh))
1764                 ldlm_lock_decref(&lockh, LCK_CR);
1765         ptlrpc_req_finished(req);
1766         req = NULL;
1767         rc = llu_file_release(ino);
1768         EXIT;
1769
1770 out:
1771         if (req != NULL)
1772                 ptlrpc_req_finished(req);
1773         return rc;
1774 }
1775
1776 static int llu_lov_file_setstripe(struct inode *ino, unsigned long arg)
1777 {
1778         struct lov_user_md lum, *lump = (struct lov_user_md *)arg;
1779         int rc;
1780         int flags = FMODE_WRITE;
1781         ENTRY;
1782
1783         LASSERT(sizeof(lum) == sizeof(*lump));
1784         LASSERT(sizeof(lum.lmm_objects[0]) == sizeof(lump->lmm_objects[0]));
1785         if (cfs_copy_from_user(&lum, lump, sizeof(lum)))
1786                 RETURN(-EFAULT);
1787
1788         rc = llu_lov_setstripe_ea_info(ino, flags, &lum, sizeof(lum));
1789         RETURN(rc);
1790 }
1791
1792 static int llu_lov_setstripe(struct inode *ino, unsigned long arg)
1793 {
1794         struct intnl_stat *st = llu_i2stat(ino);
1795         if (S_ISREG(st->st_mode))
1796                 return llu_lov_file_setstripe(ino, arg);
1797         if (S_ISDIR(st->st_mode))
1798                 return llu_lov_dir_setstripe(ino, arg);
1799
1800         return -EINVAL;
1801 }
1802
1803 static int llu_lov_getstripe(struct inode *ino, unsigned long arg)
1804 {
1805         struct lov_stripe_md *lsm = llu_i2info(ino)->lli_smd;
1806
1807         if (!lsm)
1808                 RETURN(-ENODATA);
1809
1810         return obd_iocontrol(LL_IOC_LOV_GETSTRIPE, llu_i2obdexp(ino), 0, lsm,
1811                             (void *)arg);
1812 }
1813
1814 static int llu_iop_ioctl(struct inode *ino, unsigned long int request,
1815                          va_list ap)
1816 {
1817         unsigned long arg;
1818         int rc;
1819
1820         liblustre_wait_event(0);
1821
1822         switch (request) {
1823         case LL_IOC_GROUP_LOCK:
1824                 arg = va_arg(ap, unsigned long);
1825                 rc = llu_get_grouplock(ino, arg);
1826                 break;
1827         case LL_IOC_GROUP_UNLOCK:
1828                 arg = va_arg(ap, unsigned long);
1829                 rc = llu_put_grouplock(ino, arg);
1830                 break;
1831         case LL_IOC_LOV_SETSTRIPE:
1832                 arg = va_arg(ap, unsigned long);
1833                 rc = llu_lov_setstripe(ino, arg);
1834                 break;
1835         case LL_IOC_LOV_GETSTRIPE:
1836                 arg = va_arg(ap, unsigned long);
1837                 rc = llu_lov_getstripe(ino, arg);
1838                 break;
1839         default:
1840                 CERROR("did not support ioctl cmd %lx\n", request);
1841                 rc = -ENOSYS;
1842                 break;
1843         }
1844
1845         liblustre_wait_event(0);
1846         return rc;
1847 }
1848
1849 /*
1850  * we already do syncronous read/write
1851  */
1852 static int llu_iop_sync(struct inode *inode)
1853 {
1854         liblustre_wait_event(0);
1855         return 0;
1856 }
1857
1858 static int llu_iop_datasync(struct inode *inode)
1859 {
1860         liblustre_wait_event(0);
1861         return 0;
1862 }
1863
1864 struct filesys_ops llu_filesys_ops =
1865 {
1866         fsop_gone: llu_fsop_gone,
1867 };
1868
1869 struct inode *llu_iget(struct filesys *fs, struct lustre_md *md)
1870 {
1871         struct inode *inode;
1872         struct lu_fid fid;
1873         struct file_identifier fileid = {&fid, sizeof(fid)};
1874
1875         if ((md->body->valid & (OBD_MD_FLID | OBD_MD_FLTYPE)) !=
1876             (OBD_MD_FLID | OBD_MD_FLTYPE)) {
1877                 CERROR("bad md body valid mask "LPX64"\n", md->body->valid);
1878                 LBUG();
1879                 return ERR_PTR(-EPERM);
1880         }
1881
1882         /* try to find existing inode */
1883         fid = md->body->fid1;
1884
1885         inode = _sysio_i_find(fs, &fileid);
1886         if (inode) {
1887                 if (inode->i_zombie/* ||
1888                     lli->lli_st_generation != md->body->generation*/) {
1889                         I_RELE(inode);
1890                 }
1891                 else {
1892                         llu_update_inode(inode, md);
1893                         return inode;
1894                 }
1895         }
1896
1897         inode = llu_new_inode(fs, &fid);
1898         if (inode)
1899                 llu_update_inode(inode, md);
1900
1901         return inode;
1902 }
1903
1904 static int
1905 llu_fsswop_mount(const char *source,
1906                  unsigned flags,
1907                  const void *data __IS_UNUSED,
1908                  struct pnode *tocover,
1909                  struct mount **mntp)
1910 {
1911         struct filesys *fs;
1912         struct inode *root;
1913         struct pnode_base *rootpb;
1914         struct obd_device *obd;
1915         struct llu_sb_info *sbi;
1916         struct obd_statfs osfs;
1917         static struct qstr noname = { NULL, 0, 0 };
1918         struct ptlrpc_request *request = NULL;
1919         struct lustre_md md;
1920         class_uuid_t uuid;
1921         struct config_llog_instance cfg = {0, };
1922         char ll_instance[sizeof(sbi) * 2 + 1];
1923         struct lustre_profile *lprof;
1924         char *zconf_mgsnid, *zconf_profile;
1925         char *osc = NULL, *mdc = NULL;
1926         int async = 1, err = -EINVAL;
1927         struct obd_connect_data ocd = {0,};
1928         struct md_op_data op_data = {{0}};
1929
1930         ENTRY;
1931
1932         if (ll_parse_mount_target(source,
1933                                   &zconf_mgsnid,
1934                                   &zconf_profile)) {
1935                 CERROR("mal-formed target %s\n", source);
1936                 RETURN(err);
1937         }
1938         if (!zconf_mgsnid || !zconf_profile) {
1939                 printf("Liblustre: invalid target %s\n", source);
1940                 RETURN(err);
1941         }
1942         /* allocate & initialize sbi */
1943         OBD_ALLOC(sbi, sizeof(*sbi));
1944         if (!sbi)
1945                 RETURN(-ENOMEM);
1946
1947         CFS_INIT_LIST_HEAD(&sbi->ll_conn_chain);
1948         ll_generate_random_uuid(uuid);
1949         class_uuid_unparse(uuid, &sbi->ll_sb_uuid);
1950
1951         /* generate a string unique to this super, let's try
1952          the address of the super itself.*/
1953         sprintf(ll_instance, "%p", sbi);
1954
1955         /* retrive & parse config log */
1956         cfg.cfg_instance = ll_instance;
1957         cfg.cfg_uuid = sbi->ll_sb_uuid;
1958         err = liblustre_process_log(&cfg, zconf_mgsnid, zconf_profile, 1);
1959         if (err < 0) {
1960                 CERROR("Unable to process log: %s\n", zconf_profile);
1961                 GOTO(out_free, err);
1962         }
1963
1964         lprof = class_get_profile(zconf_profile);
1965         if (lprof == NULL) {
1966                 CERROR("No profile found: %s\n", zconf_profile);
1967                 GOTO(out_free, err = -EINVAL);
1968         }
1969         OBD_ALLOC(osc, strlen(lprof->lp_dt) + strlen(ll_instance) + 2);
1970         sprintf(osc, "%s-%s", lprof->lp_dt, ll_instance);
1971
1972         OBD_ALLOC(mdc, strlen(lprof->lp_md) + strlen(ll_instance) + 2);
1973         sprintf(mdc, "%s-%s", lprof->lp_md, ll_instance);
1974
1975         if (!osc) {
1976                 CERROR("no osc\n");
1977                 GOTO(out_free, err = -EINVAL);
1978         }
1979         if (!mdc) {
1980                 CERROR("no mdc\n");
1981                 GOTO(out_free, err = -EINVAL);
1982         }
1983
1984         fs = _sysio_fs_new(&llu_filesys_ops, flags, sbi);
1985         if (!fs) {
1986                 err = -ENOMEM;
1987                 goto out_free;
1988         }
1989
1990         obd = class_name2obd(mdc);
1991         if (!obd) {
1992                 CERROR("MDC %s: not setup or attached\n", mdc);
1993                 GOTO(out_free, err = -EINVAL);
1994         }
1995         obd_set_info_async(obd->obd_self_export, sizeof(KEY_ASYNC), KEY_ASYNC,
1996                            sizeof(async), &async, NULL);
1997
1998         ocd.ocd_connect_flags = OBD_CONNECT_IBITS | OBD_CONNECT_VERSION |
1999                                 OBD_CONNECT_FID | OBD_CONNECT_AT |
2000                                 OBD_CONNECT_VBR | OBD_CONNECT_SOM;
2001 #ifdef LIBLUSTRE_POSIX_ACL
2002         ocd.ocd_connect_flags |= OBD_CONNECT_ACL;
2003 #endif
2004         ocd.ocd_ibits_known = MDS_INODELOCK_FULL;
2005         ocd.ocd_version = LUSTRE_VERSION_CODE;
2006
2007         /* setup mdc */
2008         err = obd_connect(NULL, &sbi->ll_md_exp, obd, &sbi->ll_sb_uuid, &ocd, NULL);
2009         if (err) {
2010                 CERROR("cannot connect to %s: rc = %d\n", mdc, err);
2011                 GOTO(out_free, err);
2012         }
2013
2014         err = obd_statfs(obd, &osfs, 100000000, 0);
2015         if (err)
2016                 GOTO(out_md, err);
2017
2018         /*
2019          * FIXME fill fs stat data into sbi here!!! FIXME
2020          */
2021
2022         /* setup osc */
2023         obd = class_name2obd(osc);
2024         if (!obd) {
2025                 CERROR("OSC %s: not setup or attached\n", osc);
2026                 GOTO(out_md, err = -EINVAL);
2027         }
2028         obd_set_info_async(obd->obd_self_export, sizeof(KEY_ASYNC), KEY_ASYNC,
2029                            sizeof(async), &async, NULL);
2030
2031         obd->obd_upcall.onu_owner = &sbi->ll_lco;
2032         obd->obd_upcall.onu_upcall = cl_ocd_update;
2033
2034         ocd.ocd_connect_flags = OBD_CONNECT_SRVLOCK | OBD_CONNECT_REQPORTAL |
2035                                 OBD_CONNECT_VERSION | OBD_CONNECT_TRUNCLOCK |
2036                                 OBD_CONNECT_FID | OBD_CONNECT_AT |
2037                                 OBD_CONNECT_SOM;
2038
2039         ocd.ocd_version = LUSTRE_VERSION_CODE;
2040         err = obd_connect(NULL, &sbi->ll_dt_exp, obd, &sbi->ll_sb_uuid, &ocd, NULL);
2041         if (err) {
2042                 CERROR("cannot connect to %s: rc = %d\n", osc, err);
2043                 GOTO(out_md, err);
2044         }
2045         sbi->ll_lco.lco_flags = ocd.ocd_connect_flags;
2046         sbi->ll_lco.lco_md_exp = sbi->ll_md_exp;
2047         sbi->ll_lco.lco_dt_exp = sbi->ll_dt_exp;
2048
2049         fid_zero(&sbi->ll_root_fid);
2050         err = md_getstatus(sbi->ll_md_exp, &sbi->ll_root_fid, NULL);
2051         if (err) {
2052                 CERROR("cannot mds_connect: rc = %d\n", err);
2053                 GOTO(out_lock_cn_cb, err);
2054         }
2055         if (!fid_is_sane(&sbi->ll_root_fid)) {
2056                 CERROR("Invalid root fid during mount\n");
2057                 GOTO(out_lock_cn_cb, err = -EINVAL);
2058         }
2059         CDEBUG(D_SUPER, "rootfid "DFID"\n", PFID(&sbi->ll_root_fid));
2060
2061         op_data.op_fid1 = sbi->ll_root_fid;
2062         op_data.op_valid = OBD_MD_FLGETATTR | OBD_MD_FLBLOCKS;
2063         /* fetch attr of root inode */
2064         err = md_getattr(sbi->ll_md_exp, &op_data, &request);
2065         if (err) {
2066                 CERROR("md_getattr failed for root: rc = %d\n", err);
2067                 GOTO(out_lock_cn_cb, err);
2068         }
2069
2070         err = md_get_lustre_md(sbi->ll_md_exp, request,
2071                                sbi->ll_dt_exp, sbi->ll_md_exp, &md);
2072         if (err) {
2073                 CERROR("failed to understand root inode md: rc = %d\n",err);
2074                 GOTO(out_request, err);
2075         }
2076
2077         LASSERT(fid_is_sane(&sbi->ll_root_fid));
2078
2079         root = llu_iget(fs, &md);
2080         if (!root || IS_ERR(root)) {
2081                 CERROR("fail to generate root inode\n");
2082                 GOTO(out_request, err = -EBADF);
2083         }
2084
2085         /*
2086          * Generate base path-node for root.
2087          */
2088         rootpb = _sysio_pb_new(&noname, NULL, root);
2089         if (!rootpb) {
2090                 err = -ENOMEM;
2091                 goto out_inode;
2092         }
2093
2094         err = _sysio_do_mount(fs, rootpb, flags, tocover, mntp);
2095         if (err) {
2096                 _sysio_pb_gone(rootpb);
2097                 goto out_inode;
2098         }
2099
2100         cl_sb_init(sbi);
2101
2102         ptlrpc_req_finished(request);
2103
2104         CDEBUG(D_SUPER, "LibLustre: %s mounted successfully!\n", source);
2105         liblustre_wait_idle();
2106
2107         return 0;
2108
2109 out_inode:
2110         _sysio_i_gone(root);
2111 out_request:
2112         ptlrpc_req_finished(request);
2113 out_lock_cn_cb:
2114         obd_disconnect(sbi->ll_dt_exp);
2115 out_md:
2116         obd_disconnect(sbi->ll_md_exp);
2117 out_free:
2118         if (osc)
2119                 OBD_FREE(osc, strlen(osc) + 1);
2120         if (mdc)
2121                 OBD_FREE(mdc, strlen(mdc) + 1);
2122         OBD_FREE(sbi, sizeof(*sbi));
2123         liblustre_wait_idle();
2124         return err;
2125 }
2126
2127 struct fssw_ops llu_fssw_ops = {
2128         llu_fsswop_mount
2129 };
2130
2131 static struct inode_ops llu_inode_ops = {
2132         inop_lookup:    llu_iop_lookup,
2133         inop_getattr:   llu_iop_getattr,
2134         inop_setattr:   llu_iop_setattr,
2135         inop_filldirentries:     llu_iop_filldirentries,
2136         inop_mkdir:     llu_iop_mkdir_raw,
2137         inop_rmdir:     llu_iop_rmdir_raw,
2138         inop_symlink:   llu_iop_symlink_raw,
2139         inop_readlink:  llu_iop_readlink,
2140         inop_open:      llu_iop_open,
2141         inop_close:     llu_iop_close,
2142         inop_link:      llu_iop_link_raw,
2143         inop_unlink:    llu_iop_unlink_raw,
2144         inop_rename:    llu_iop_rename_raw,
2145         inop_pos:       llu_iop_pos,
2146         inop_read:      llu_iop_read,
2147         inop_write:     llu_iop_write,
2148         inop_iodone:    llu_iop_iodone,
2149         inop_fcntl:     llu_iop_fcntl,
2150         inop_sync:      llu_iop_sync,
2151         inop_datasync:  llu_iop_datasync,
2152         inop_ioctl:     llu_iop_ioctl,
2153         inop_mknod:     llu_iop_mknod_raw,
2154 #ifdef _HAVE_STATVFS
2155         inop_statvfs:   llu_iop_statvfs,
2156 #endif
2157         inop_gone:      llu_iop_gone,
2158 };