source: gpfs_3.1_ker2.6.20/lpp/mmfs/src/gpl-linux/file.c @ 16

/***************************************************************************
 *
 * Copyright (C) 2001 International Business Machines
 * All rights reserved.
 *
 * This file is part of the GPFS mmfslinux kernel module.
 *
 * Redistribution and use in source and binary forms, with or without 
 * modification, are permitted provided that the following conditions 
 * are met:
 *
 *  1. Redistributions of source code must retain the above copyright notice, 
 *     this list of conditions and the following disclaimer. 
 *  2. Redistributions in binary form must reproduce the above copyright 
 *     notice, this list of conditions and the following disclaimer in the
 *     documentation and/or other materials provided with the distribution. 
 *  3. The name of the author may not be used to endorse or promote products 
 *     derived from this software without specific prior written
 *     permission. 
 *
 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR 
 * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES 
 * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED. 
 * IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, 
 * SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, 
 * PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; 
 * OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, 
 * WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR
 * OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF 
 * ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 *
 *************************************************************************** */
/* @(#)95       1.95.1.8  src/avs/fs/mmfs/ts/kernext/gpl-linux/file.c, mmfs, avs_rgpfs24, rgpfs24s012a 4/30/07 20:13:26 */
/*
 * File operations
 *
 * Contents:
 *   gpfs_f_llseek
 *   gpfs_f_readdir
 *   gpfs_f_poll
 *   gpfs_f_ioctl
 *   gpfs_filemap_open
 *   gpfs_filemap_close
 *   gpfs_filemap_nopage       (in mmap.c)
 *   gpfs_filemap_nopagedone   (in mmap.c)
 *   gpfs_f_mmap
 *   gpfs_f_open
 *   gpfs_f_release
 *   gpfs_f_fsync
 *   gpfs_f_fasync
 *   fsyncInternal
 *   gpfs_f_lock
 *   gpfs_f_flock
 *   rdwrInternal
 *   gpfs_f_read
 *   gpfs_f_dir_read
 *   gpfs_f_write
 *   gpfs_f_readv
 *   gpfs_f_writev
 *   gpfs_f_cleanup
 *
 */

#include <Shark-gpl.h>

#include <linux/fs.h>
#include <linux/errno.h>
#include <linux/stat.h>
#include <linux/fcntl.h>
#ifdef MODULE
#include <linux/module.h>
#endif
#include <linux/slab.h>
#include <linux/smp_lock.h>
#include <linux/mm.h>
#include <linux/mman.h>
#ifdef NFS4_CLUSTER
#include <linux/nfs4.h>
#endif /* NFS4_CLUSTER */
#if LINUX_KERNEL_VERSION >= 2050000
#include <linux/pagemap.h>
#include <linux/page-flags.h>
#endif

#include <linux2gpfs.h>
#ifdef P_NFS4
#include <linux/sunrpc/svc.h>
#include <linux/nfsd/nfsfh.h>
#include <linux/nfsd/nfsd4_pnfs.h>
#include <linux/nfsd/pnfsd.h>
#endif
#include <cxiTypes.h>
#include <cxiSystem.h>
#include <cxiMode.h>
#include <cxi2gpfs.h>
#include <cxiVFSStats.h>
#include <cxiCred.h>
#include <cxiIOBuffer.h>
#include <cxiMmap.h>
#include <Trace.h>
#include <verdep.h>


/* prototypes */
static int fsyncInternal(struct file *fP);

/* file_operations */

loff_t
gpfs_f_llseek(struct file *fP, loff_t offset, int origin)
{
  struct inode *iP = fP->f_dentry->d_inode;
  loff_t rc = -EINVAL;

  ENTER(0);
  TRACE3(TRACE_VNODE, 1, TRCID_LINUXOPS_LLSEEK_ENTER,
         "gpfs_f_llseek enter: fP 0x%lX offset 0x%llX origin %d\n",
         fP, offset, origin);
  /* BKL is held at entry */

  switch (origin)
  {
    case 2:
      gpfs_i_getattr_internal(iP);
      offset += iP->i_size;
      break;

    case 1:
      offset += fP->f_pos;
  }

  if (offset >= 0)
  {
    rc = offset;
    if (offset != fP->f_pos)
    {
      fP->f_pos = offset;
#if LINUX_KERNEL_VERSION < 2050000
      fP->f_reada = 0;
#endif
    }
  }
  else
    cxiErrorNFS(rc);

  TRACE4(TRACE_VNODE, 1, TRCID_LINUXOPS_LLSEEK_EXIT,
         "gpfs_f_llseek exit: fP 0x%lX offset 0x%llX origin %d rc 0x%llX\n",
         fP, offset, origin, rc);
  EXIT(0);
  return rc;
}

/* Save everything we need to make the OS-specific filldir call. */
typedef struct {
  offset_t  offset;
  ino_t     ino;
  int       namelen;
  char      name[1];
} fillDirEntry;

/* gpfs_f_readdir provides a buffer for NFS_fillDir to place entries,
 * and this structure to keep track of its use over successive calls.
 */
typedef struct {
  fillDirEntry   *firstP;  /* first entry */
  fillDirEntry   *endP;    /* buffer end  */
  fillDirEntry   *bufferP; /* current location */
} NFS_fillDirArgs;

/* Return the location of our next filldir entry.  Allow for the size
 * of the fillDirEntry struct plus the namelen.  Round to dblword.
 */
#define NEXT_FILLDIR_ENTRY(eP, len) \
  (fillDirEntry *)((caddr_t)(eP)+(((sizeof(fillDirEntry)+(len)+3)>>2)<<2))

/* Size of our NFS_fillDir buffer. */
#define FILLDIR_BUFSIZE 700

/* For NFS readdir, we provide our own filldir callback so that we can save
 * the records until after we release our locks.  We can then make the real
 * filldir calls without fear they will deadlock when they loopback to the 
 * filesystem for permission checks, etc.
 */
int
NFS_fillDir(void *myArgP, char *nameP, int namelen,
            offset_t offset, const ino_t ino)
{
  NFS_fillDirArgs *argsP = (NFS_fillDirArgs *)myArgP;
  fillDirEntry *entryP = argsP->bufferP;
  fillDirEntry *nextP = NEXT_FILLDIR_ENTRY(entryP, namelen);

  /* If this entry will not fit, report the full condition. */
  if (nextP > argsP->endP)
    return -EINVAL;

  /* Save filldir information to make the real call later */
  entryP->offset = offset;
  entryP->ino = ino;
  entryP->namelen = namelen;
  cxiMemcpy(entryP->name, nameP, namelen+1);

  /* Bump the entry location in arg structure for the next call */
  argsP->bufferP = nextP;

  return 0;
}

int
gpfs_f_readdir(struct file *fP, void *direntP, filldir_t filldir)
{
  int rc;
  Boolean klock = false;
  struct gpfsVfsData_t *privVfsP;
  cxiNode_t *cnP;
  struct inode *iP;
  cxiFillDirArg_t fillDirArg;
  fillDirEntry *fillDirBuf = NULL;
  NFS_fillDirArgs filldirArgs;

  VFS_STAT_START(readdirCall);
  ENTER(0);
  DBGASSERT(fP != NULL);

  TRACE4(TRACE_VNODE, 1, TRCID_LINUXOPS_READDIR_ENTER,
         "gpfs_f_readdir enter: fP 0x%lX direntP 0x%lX "
         "filldir 0x%lX pos %lld\n", fP, direntP, filldir, fP->f_pos);
  /* BKL is held at entry */

  /* Quick check for EOF */
  if (fP->f_pos == GPFS_DIR_EOF)
  {
    rc = 0;  // end-of-directory
  }
  else
  {

    iP = fP->f_dentry->d_inode;
    DBGASSERT(iP != NULL);
    cnP = VP_TO_CNP(iP);
    privVfsP = VP_TO_PVP(iP);
    DBGASSERT(privVfsP != NULL);

    /* When called by NFS, wait to make the filldar calls until after
     * we return from gpfsReaddir.  The NFS filldir implementation 
     * includes callbacks (e.g., permission checks) into the filesystem 
     * and these calls may result in getting locks out-of-order and 
     * are therefore subject to deadlock.
     */
    if (cxiIsNFSThread())
    {
      /* Specify a special filldir function where we will save entry 
       * information.  Upon our return from gpfsReaddir, we no longer 
       * hold any locks so we will then go through these saved entries
       * and make the real filldir calls.
       */
      fillDirArg.fnP = (void *)NFS_fillDir;
      fillDirArg.argP = &filldirArgs;

      fillDirBuf = (fillDirEntry *)cxiMallocPinned(FILLDIR_BUFSIZE);

      filldirArgs.firstP = fillDirBuf;
      filldirArgs.endP = (fillDirEntry *)((caddr_t)fillDirBuf + FILLDIR_BUFSIZE);
      filldirArgs.bufferP = filldirArgs.firstP;
    }
    else
    {
      /* Unfortunately we can't use the OS version of the filldir 
       * routine directly.  It has different signatures in varying
       * kernel levels, so we use cxiFillDir() in the portability layer
       * to handle the different signatures.
       */
      fillDirArg.fnP = (void *)filldir;
      fillDirArg.argP = direntP;
    }

    if (kernel_locked())
    {
      unlock_kernel();
      klock = true;
    }
    rc = gpfs_ops.gpfsReaddir(privVfsP, cnP, &fillDirArg, cxiFillDir,
                              &fP->f_pos, vnOp, NULL);

    /* Even if gpfsReaddir reports an error, we want to look
     * to see if there were any entries previously returned.
     */
    if (cxiIsNFSThread() && (filldirArgs.bufferP > filldirArgs.firstP))
    {
      int fillrc;
      fillDirEntry *nextP = filldirArgs.firstP;

      /* Set the real filldir fcn/arg pointers */
      fillDirArg.fnP = (void *)filldir;
      fillDirArg.argP = direntP;

      while(nextP < filldirArgs.bufferP)
      {
        /* Do not overlay any gpfsReadir rc. */
        fillrc = CALL_FILLDIR(cxiFillDir,
                              &fillDirArg, 
                              (nextP->name), 
                              (nextP->namelen), 
                              (nextP->offset), 
                              (nextP->ino));
        if (fillrc < 0)
        {
          rc = 0; /* entry doesn't fit is ok (will resume at offset) */

          /* Reset f_pos based on what we've been able to pass back
           * to NFS.  This is where they will start on the next call.
           */
          fP->f_pos = nextP->offset; /* next offset for nfsd_readdir */

          break;
        }
        nextP = NEXT_FILLDIR_ENTRY(nextP, nextP->namelen);
      }
    }

    if (klock)
      lock_kernel();
  }

  TRACE3(TRACE_VNODE, 1, TRCID_LINUXOPS_READDIR_EXIT,
         "gpfs_f_readdir exit: fP 0x%lX pos %lld code 0 rc %d\n",
         fP, fP->f_pos, rc);

  if (fillDirBuf)
    cxiFreePinned(fillDirBuf);

  if (rc)
    cxiErrorNFS(rc);

  VFS_STAT_STOP;
  EXIT(0);
  return (-rc);
}

uint
gpfs_f_poll(struct file *fP, struct poll_table_struct *wait)
{
  TRACE1(TRACE_VNODE, 1, TRCID_LINUXOPS_POLL,
         "gpfs_f_poll: rc bits POLLERR: fP 0x%lX\n", fP);
  return (uint)0; // ?? which POLL* bits
}

int
gpfs_f_ioctl(struct inode *iP, struct file *fP, uint cmd, unsigned long arg)
{
  TRACE3(TRACE_VNODE, 1, TRCID_LINUXOPS_IOCTL,
         "gpfs_f_ioctl: rc -ENOTTY: iP 0x%lX fP 0x%lX cmd %d\n",
         iP, fP, cmd);
  return -ENOTTY; // no one can really explain why this errno, but it is common
}


/* called for every child process forked after mmap */
void gpfs_filemap_open(struct vm_area_struct * vma)
{
  int rc = 0;
  Boolean writeAccess = false;
  cxiNode_t *cnP;
  ext_cred_t eCred;
  struct file *file = vma->vm_file;
  struct inode *inode = file->f_dentry->d_inode;
  struct gpfsVfsData_t *privVfsP;
  struct mm_struct *mm = vma->vm_mm;
  long long offset;
  long long length;

  ENTER(0);
  TRACE4(TRACE_VNODE, 2, TRCID_FM_OPEN,
         "gpfs_filemap_open enter: vma 0x%lX inode %d icount %d name %s\n",
         vma, inode->i_ino, atomic_read((atomic_t *)&inode->i_count),
         file->f_dentry? file->f_dentry->d_name.name: (const unsigned char*)"");

  TRACE2(TRACE_VNODE, 2, TRCID_FM_OPEN_1,
         "gpfs_filemap_open : mm 0x%lX mm_users %d\n",
         mm, atomic_read(&mm->mm_users));

  cnP = VP_TO_CNP(inode);
  privVfsP = VP_TO_PVP(inode);
  DBGASSERT(privVfsP != NULL);

  if ((vma->vm_flags & VM_SHARED) && (vma->vm_flags & VM_MAYWRITE))
    MMAP_WRITE_ACCESS(writeAccess);

  setCred(&eCred);
  
  offset = vma->vm_pgoff<<PAGE_SHIFT;
  length = vma->vm_end - vma->vm_start;
  rc = gpfs_ops.gpfsMmap(privVfsP, cnP, (void *)inode, &eCred, NULL,
                         writeAccess,false, offset,length);

  TRACE2(TRACE_VNODE, 2, TRCID_FM_OPEN_EXIT,
         "gpfs_filemap_open exit: vma 0x%lX icount %d\n",
         vma, atomic_read((atomic_t *)&inode->i_count));

  if (rc)
    cxiErrorNFS(rc);

  EXIT(0);
}

void gpfs_filemap_close(struct vm_area_struct * vma)
{
  struct file *fP = vma->vm_file;
  struct inode *inode = fP->f_dentry->d_inode;
  int flags, rc;
  struct gpfsVfsData_t *privVfsP;
  cxiNode_t *cnP = VP_TO_CNP(inode);
  struct mm_struct *mm = vma->vm_mm;

  VFS_STAT_START(unmapCall);
  ENTER(0);

  if ((vma->vm_flags & VM_SHARED) && (vma->vm_flags & VM_MAYWRITE))
    flags = 0;
  else
    flags = CXI_SHM_RDONLY;

  privVfsP = VP_TO_PVP(inode);

  TRACE3(TRACE_VNODE, 2, TRCID_FM_CLOSE_ENTER,
         "gpfs_filemap_close: vma 0x%lX inode 0x%lX i_count %d\n",
         vma, inode, (Int32)atomic_read((atomic_t *)&inode->i_count));
  TRACE3(TRACE_VNODE, 2, TRCID_FM_CLOSE_ENTER1,
         "gpfs_filemap_close: inode %d, name %s, nrpages %d\n",
         inode->i_ino,
         fP->f_dentry? fP->f_dentry->d_name.name: (const unsigned char*)"",
         inode->i_data.nrpages);
  TRACE2(TRACE_VNODE, 2, TRCID_FM_CLOSE_ENR,
         "gpfs_filemap_close: mm 0x%lX mm_users %d\n",
         mm,atomic_read(&mm->mm_users));
  
  rc = gpfs_ops.gpfsUnmap(privVfsP, cnP, flags);
  cxiPutOSNode((void *)inode);

  TRACE3(TRACE_VNODE, 2, TRCID_FM_CLOSE,
         "gpfs_filemap_close: vma 0x%lX inode 0x%lX i_count %d\n",
         vma, inode, (Int32)atomic_read((atomic_t *)&inode->i_count));

  if (rc)
    cxiErrorNFS(rc);

  VFS_STAT_STOP;
  EXIT(0);
}

int
gpfs_f_mmap(struct file *fP, struct vm_area_struct *vma)
{
  int rc;
  int code = 0;
  Boolean heldVnode = false;
  Boolean writeAccess = false;
  cxiNode_t *cnP;
  struct gpfsVfsData_t *privVfsP;
  struct inode *iP = fP->f_dentry->d_inode;
  ext_cred_t eCred;
  cxiVattr_t vattr;
  struct mm_struct *mm = vma->vm_mm;
  long long offset;
  long long length;

  VFS_STAT_START(map_lloffCall);
  ENTER(0);
  TRACE3(TRACE_VNODE, 1, TRCID_LINUXOPS_MMAP_ENTER,
         "gpfs_f_mmap enter: fP 0x%lX inum %d vma 0x%1X\n",
         fP, iP->i_ino, vma);

  TRACE4(TRACE_VNODE, 1, TRCID_LINUXOPS_MMAP_ENTER_A,
         "gpfs_f_mmap: vm_start 0x%lX vm_end 0x%lX, vmpgoff 0x%lX, "
         "vmflags 0x%lX\n",
         vma->vm_start, vma->vm_end, vma->vm_pgoff, vma->vm_flags);
   TRACE4(TRACE_VNODE, 2, TRCID_LINUXOPS_MMAP_ENTER_A1,
          "gpfs_f_mmap: inode %d icount %d name %s nrpages %d\n",
          iP->i_ino, atomic_read((atomic_t *)&iP->i_count),
          fP->f_dentry ? fP->f_dentry->d_name.name : (const unsigned char*)"",
          iP->i_data.nrpages);
  TRACE2(TRACE_VNODE, 1, TRCID_LINUXOPS_MMAP_ENTER_AB,
         "gpfs_f_mmap: mm 0x%lX mm_users %d\n",
         mm,atomic_read(&mm->mm_users));

  cnP = VP_TO_CNP(iP);
  privVfsP = VP_TO_PVP(iP);
  DBGASSERT(privVfsP != NULL);

  if ((vma->vm_flags & VM_SHARED) && (vma->vm_flags & VM_MAYWRITE))
  {
    MMAP_WRITE_ACCESS(writeAccess);
    if (!writeAccess)
    {
      /* Patch must be applied at this kernel level for mmap write */
      code = 1;
      rc = -EINVAL;
      goto xerror;
    }
  }

  setCred(&eCred);

  offset = vma->vm_pgoff<<PAGE_SHIFT;
  length = vma->vm_end - vma->vm_start;
  rc = gpfs_ops.gpfsMmap(privVfsP, cnP, (void *)iP, &eCred, NULL, 
                         writeAccess, true,offset,length);
  if (rc != 0)
  {
    code = 2;
    rc = -rc;
    goto xerror;
  }
  
  heldVnode = true;

  if ((vma->vm_flags & VM_SHARED) && (vma->vm_flags & VM_MAYWRITE) &&
      !iP->i_mapping->a_ops->writepage)
  {
    code = 3;
    rc = -EINVAL;
    goto xerror;
  }

  if (!iP->i_sb || !S_ISREG(iP->i_mode))
  {
    code = 4;
    rc = -EACCES;
    goto xerror;
  }

  if (!iP->i_mapping->a_ops->readpage)
  {
    code = 5;
    rc = -ENOEXEC;
    goto xerror;
  }

  /* revalidate linux inode */
  /* This has the effect of calling us back under a lock and
   * setting the inode attributes at the OS level (since this
   * operating system caches this info in the vfs layer)
   */
  rc = gpfs_ops.gpfsGetattr(privVfsP, cnP, &vattr, false);
  if (rc != 0)
  {
    code = 6;
    rc = -rc;
    goto xerror;
  }

#ifdef UPDATE_ATIME
  UPDATE_ATIME(iP);
#else
#if LINUX_KERNEL_VERSION >= 2061600
  touch_atime(NULL, fP->f_dentry);
#else
  update_atime(iP);
#endif
#endif
  vma->vm_ops = &gpfs_vmop;

xerror:
  if (rc != 0 && heldVnode)
    cxiPutOSNode((void *)iP); // corresponding hold in gpfsMmap

  TRACE2(TRACE_VNODE, 1, TRCID_LINUXOPS_MMAP_EXIT,
         "gpfs_f_mmap exit: rc %d code %d\n", rc, code);

  if (rc)
    cxiErrorNFS(rc);

  VFS_STAT_STOP;
  EXIT(0);
  return rc;
}

int
gpfs_f_open(struct inode *iP, struct file *fP)
{
  int rc = 0;
  int code = 0;
  Boolean gotBKL = false;
  int flags = cxiOpenFlagsXlate(fP->f_flags);
  int iflags = cxiIsSambaThread()? GPFS_OPEN_NO_SMBLOCK: 0;
  cxiNode_t *cnP;
  struct gpfsVfsData_t *privVfsP;
  ext_cred_t eCred;

  VFS_STAT_START(openCall);
  ENTER(0);
  TRACE7(TRACE_VNODE, 1, TRCID_LINUXOPS_OPEN_ENTER,
         "gpfs_f_open enter: iP 0x%lX fP 0x%lX f_flags 0x%X dP 0x%lX '%s' "
         "flags 0x%X isNFS %d\n", iP, fP, fP->f_flags, fP->f_dentry,
         fP->f_dentry? fP->f_dentry->d_name.name: (const unsigned char*)"",
         flags, cxiIsNFSThread());

  /* BKL is not held at entry, except for NFS calls */
  TraceBKL();
  if (current->lock_depth >= 0)  /* kernel lock is held by me */
  {
    gotBKL = true;
    unlock_kernel();
  }

  cnP = VP_TO_CNP(iP);
  privVfsP = VP_TO_PVP(iP);
  DBGASSERT(privVfsP != NULL);

  /* see comment in gpfs_i_create() on the reason for this code */
  if (cnP->createRaceLoserThreadId &&
      cnP->createRaceLoserThreadId == cxiGetThreadId())
  {
    int fflags = cxiOpenFlagsXlate(fP->f_flags);
    int amode;

    cnP->createRaceLoserThreadId = 0;
    code = EEXIST;

    amode = ((flags & FWRITE ? W_ACC : 0) |
             (flags & FREAD ? R_ACC : 0)  |
             (flags & FTRUNC ? W_ACC : 0));

    TRACE4(TRACE_VNODE, 1, TRCID_LINUXOPS_OPEN_01,
           "gpfs_f_open fileExist iP 0x%lX cnP 0x%lX fflags 0x%X amode 0x%X\n",
           iP, cnP, fflags, amode);

    /* Check if FEXCL and FCREAT are on and the file exists return EEXIST
     * could not do it at create time because the open flgas are not availble
     * on the create call.
     */
    if ((flags & FEXCL) && (flags & FCREAT))
    {
      rc = EEXIST;
      goto xerror;
    }

    setCred(&eCred);
    rc = gpfs_ops.gpfsAccess(privVfsP, cnP, amode, ACC_SELF, &eCred);
    if (rc)
      goto xerror;
  }

  if (cxiIsNFSThread() && GNP_IS_FILE(cnP))
  {
    int NFSflags;
    int code;

    BEGIN_FAR_CODE;
    /* Linux NFS will not do vget so the clone vnode cannot be created then.
       Need to GetNFS here so the NFS structures will be available. */

#ifdef NFS_CLUSTER_LOCKS //??? temp fix for NFSv4
    fP->f_mode |= FMODE_READ;
#endif

    NFSflags = FWRITE|FREAD;
    rc = gpfs_ops.gpfsGetNFS((void *)iP,
                             (struct MMFSVInfo **)&fP->private_data,
                             &NFSflags);
    if (rc != 0)
    {
      code = ENOSYS; //??EGET_NFS;
      goto xerror;
    }
    DBGASSERT((struct MMFSVInfo *)fP->private_data != NULL);

    END_FAR_CODE;
    goto xerror;
  }

  setCred(&eCred);   // rebuild since gpfsAccess may have remapped the ids
  rc = gpfs_ops.gpfsOpen(privVfsP, cnP, flags, iflags, 0,
                         (struct MMFSVInfo **)&fP->private_data, &eCred);

xerror:
  TRACE4(TRACE_VNODE, 1, TRCID_LINUXOPS_OPEN_EXIT,
         "gpfs_f_open exit: iP 0x%lX vinfoP 0x%lX code %d rc %d\n",
         iP, (struct MMFSVInfo *)fP->private_data, code, rc);

  VFS_STAT_STOP;

  if (gotBKL)        /* If held kernel lock on entry then reacquire it */
    lock_kernel();

  if (rc)
    cxiErrorNFS(rc);

  EXIT(0);
  return (-rc);
}

int
gpfs_f_release(struct inode *iP, struct file *fP)
{
  int rc = 0;
  int code = 0;
  int flags = cxiOpenFlagsXlate(fP->f_flags);
  struct MMFSVInfo *vinfoP = (struct MMFSVInfo *)fP->private_data;
  cxiNode_t *cnP;
  struct gpfsVfsData_t *privVfsP;

  VFS_STAT_START(closeCall);
  ENTER(0);
  cnP = VP_TO_CNP(iP);
  privVfsP = VP_TO_PVP(iP);
  DBGASSERT(privVfsP != NULL);

  TRACE4(TRACE_VNODE, 1, TRCID_LINUXOPS_CLOSE_ENTER,
         "gpfs_f_release enter: iP 0x%lX f_flags 0x%X flags 0x%X vinfoP 0x%lX\n",
         iP, fP->f_flags, flags, vinfoP);
  /* BKL is held if the file was open R/W, otherwise not held */

  /* If nfsd is closing one of its files, schedule it for a delayed close. */
  if (cnP && VP_TO_NFSP(iP) && cxiIsNFSThread())
  {
    DBGASSERT(GNP_IS_FILE(cnP));

    /* On the last NFS release, a watchdog will be set to close the file
       after a delay. */

    rc = gpfs_ops.gpfsReleaseNFS(iP);

    goto xerror;
  }

  rc = gpfs_ops.gpfsClose(privVfsP, cnP, flags, vinfoP, true);

  fP->private_data = NULL;  // MMFSVInfo was freed

xerror:
  TRACE2(TRACE_VNODE, 1, TRCID_CLOSE_EXIT,
         "gpfs_f_release exit: code %d rc %d\n", code, rc);

  if (rc)
    cxiErrorNFS(rc);

  VFS_STAT_STOP;
  EXIT(0);
  return (-rc);
}

int
gpfs_f_fsync(struct file *fP, struct dentry *direntP, int datasync)
{
  int rc;

  ENTER(0);
  TRACE3(TRACE_VNODE, 1, TRCID_LINUXOPS_FSYNC_ENTER,
         "gpfs_f_fsync enter: fP 0x%lX dirent 0x%lX datasync %d\n",
         fP, direntP, datasync);
  /* Linux doc says BKL is held, but it does not seem to be */

  rc = fsyncInternal(fP);

  TRACE2(TRACE_VNODE, 1, TRCID_LINUXOPS_FSYNC_EXIT,
         "gpfs_f_fsync exit: file 0x%lX rc %d\n", fP, rc);

  EXIT(0);
  return (-rc);
}

int
gpfs_f_fasync(int fd, struct file *fP, int on)
{
  int rc;

  ENTER(0);
  TRACE3(TRACE_VNODE, 1, TRCID_LINUXOPS_FASYNC_ENTER,
         "gpfs_f_fasync enter: fd %d fP 0x%lX on %d\n",
         fd, fP, on);
  /* Linux doc says BKL is held, but it does not seem to be */

  rc = fsyncInternal(fP);

  TRACE2(TRACE_VNODE, 1, TRCID_LINUXOPS_FASYNC_EXIT,
         "gpfs_f_fasync exit: fP 0x%lX rc %d\n", fP, rc);

  EXIT(0);
  return (-rc);
}

static int
fsyncInternal(struct file *fP)
{
  int rc = 0;
  cxiNode_t *cnP;
  struct inode *iP;
  struct gpfsVfsData_t *privVfsP;
  ext_cred_t eCred;
  struct MMFSVInfo *vinfoP;

  VFS_STAT_START(fsyncCall);
  ENTER(0);
  VFS_INC(fsyncCall);
  /* Creating files via nfs can get us here with a null fP. */
  if (!fP)
    goto xerror;

  vinfoP = (struct MMFSVInfo *)fP->private_data;

  iP = fP->f_dentry->d_inode;
  DBGASSERT(iP != NULL);

  cnP = VP_TO_CNP(iP);
  privVfsP = VP_TO_PVP(iP);
  DBGASSERT(privVfsP != NULL);

  setCred(&eCred);
  rc = gpfs_ops.gpfsFsync(privVfsP, vinfoP, cnP, FFILESYNC, &eCred);

  if (rc)
    cxiErrorNFS(rc);

xerror:
  VFS_STAT_STOP;
  EXIT(0);
  return rc;
}

#ifdef NFS_CLUSTER_LOCKS
void
gpfs_grace(int on_off)
{
  gpfs_ops.gpfsGrace(on_off);
}

/*
 * arg is lock type F_RDLCK:0 F_WRLCK:1 F_UNLCK:2
*/
int
gpfs_f_set_lease(struct file *fP, long arg, struct file_lock **flPP)
{
  int rc = EAGAIN;
  int NFSflags;
  int mode, oplockWant, oplockGot, flags;
  void *cb_token, *cookie;
  struct file_lock *flP;
  ext_cred_t eCred;
  cxiNode_t *cnP;
  struct gpfsVfsData_t *privVfsP;
  struct MMFSVInfo *vinfoP = (struct MMFSVInfo *)fP->private_data;
  struct inode *iP = fP->f_dentry->d_inode;

//VFS_STAT_START(lockctlCall);
  ENTER(0);

  TRACE4(TRACE_VNODE, 1, TRCID_LINUXOPS_LEASE_ENTER,
         "gpfs_f_set_lease: fP 0x%lX iP 0x%lX type %s by %s\n",
          fP, iP, (arg == F_RDLCK) ? "RD" :(arg == F_WRLCK) ? "WR" : "UNLCK",
          (cxiIsNFSThread()) ? "NFS" : "SAMBA");

  privVfsP = VP_TO_PVP(fP->f_dentry->d_inode);
  DBGASSERT(privVfsP != NULL);

  flP = *flPP;
  cookie = NULL;
  NFSflags = FREAD;
  mode = FMODE_WRITE;
  flags = RESERVE_NONE;
  oplockGot = smbOplockNone;
  cb_token = iP;

  if (arg == F_UNLCK) {
    cb_token = NULL;
    oplockWant = smbOplockNone;
    flags = RESERVE_DOWNGRADE;
  }
  else if (arg == F_RDLCK)
    oplockWant = smbOplockShared;
  else if (arg == F_WRLCK) {
    oplockWant = smbOplockExclusive;
    NFSflags |= FWRITE;
    mode = FMODE_READ;
  }
  else goto xerror;

  if (cxiIsNFSThread())
  {
    rc = gpfs_ops.gpfsGetNFS((void *)iP, (struct MMFSVInfo **)&vinfoP,
                                                           &NFSflags);
    if (rc)
      goto xerror;

    cnP = VP_TO_CNP(iP);
    setCred(&eCred);
    rc = gpfs_ops.gpfsOpenNFS(privVfsP, cnP, NFSflags, vinfoP, &eCred);
    if (rc)
      goto xerror2;
  }
  rc = gpfs_ops.gpfsReserveDelegation(fP, vinfoP , privVfsP, oplockWant, flags,
                                                             cb_token, cookie);
  if (rc)
    goto xerror2;

  lock_kernel();
  rc = setlease(fP, arg, flPP);
  unlock_kernel();

  if (rc) {  // if error release the delegation
    gpfs_ops.gpfsReserveDelegation(fP, vinfoP , privVfsP, smbOplockNone,
                                   RESERVE_DOWNGRADE, NULL, NULL);
    if (rc < 0)
      rc = -rc;  /* make it positive */
  }
  else {   // rc=0
    oplockGot = gpfs_ops.SMBGetOplockStateV(vinfoP);
    if (oplockGot == oplockWant)
      goto xerror2;
    else {   // already lost the delegation
      __break_lease(iP, FMODE_WRITE);
    }
  }

xerror2:
  if (cxiIsNFSThread())
    gpfs_ops.gpfsReleaseNFS(iP);

xerror:
  TRACE6(TRACE_VNODE, 1, TRCID_LINUXOPS_LEASE_EXIT,
   "gpfs_f_set_lease: fP 0x%lX flP 0x%lX rc %d oplockWant %d oplockGot %d %s\n",
    fP, flP, rc, oplockWant, oplockGot, (oplockGot == smbOplockShared) ? "RD" :
    (oplockGot == smbOplockExclusive) ? "WR" : "NONE");

//VFS_STAT_STOP;
  EXIT(0);
  return (-rc);
}
#endif

int
gpfs_f_lock(struct file *fP, int cmd, struct file_lock *flP)
{
  int rc = 0;
  int code = 0;
  cxiNode_t *cnP;
  ext_cred_t eCred;
  struct gpfsVfsData_t *privVfsP;
  eflock_t lckdat;
  unsigned long localRetryId = 0;
  int(* vfs_callback)(void *, void *, int) = NULL;

  VFS_STAT_START(lockctlCall);
  ENTER(0);

  /* Linux converts flock64 to flock before calling GPFS lock routine,
     but leaves "cmd" as is. Allow these to go through. */
#if !defined(__64BIT__)
  if (cmd == F_GETLK64) cmd = F_GETLK;
  if (cmd == F_SETLK64) cmd = F_SETLK;
  if (cmd == F_SETLKW64) cmd = F_SETLKW;
#endif

  if ((cmd != F_GETLK) && (cmd != F_SETLK) && (cmd != F_SETLKW))
  {
    code = 2;
    rc = ENOSYS;
    goto xerror;
  }

  setCred(&eCred);
  TRACE6(TRACE_VNODE, 1, TRCID_LINUXOPS_LOCKCTL_ENTER,
         "gpfs_f_lock enter: pid %d fp 0x%lX range 0x%lX:%lX cmd %s type %s\n",
         flP->fl_pid, fP, flP->fl_start, flP->fl_end,
         (cmd == F_GETLK) ? "GETLK" : (cmd == F_SETLK) ? "SETLK" : "SETLKW",
         (flP->fl_type == F_RDLCK) ? "RDLCK" :
         (flP->fl_type == F_WRLCK) ? "WRLCK" : "UNLCK");

  TRACE5(TRACE_VNODE, 3, TRCID_LINUXOPS_LOCKCTL_ENTER2,
         "gpfs_f_lock       : pos 0x%lX iP 0x%lX fl_flags 0x%X uid %d gid %d\n",
         fP->f_pos, fP->f_dentry->d_inode, flP->fl_flags,
         eCred.principal, eCred.group);
  TraceBKL();

  cnP = VP_TO_CNP(fP->f_dentry->d_inode);
  privVfsP = VP_TO_PVP(fP->f_dentry->d_inode);
  DBGASSERT(privVfsP != NULL);

  /* convert file_lock to eflock */
  cxiVFSToFlock((void *)flP, &lckdat);
#ifdef NFS_CLUSTER_LOCKS
#if 0
  if (flP->fl_state == FL_CANCELED && flP->fl_type != F_UNLCK)
  {
    //??? just unblock queued lock
    // use kxCleanupAcquires() which is used only for AIX now
    rc = 0;
    goto xerror;
  }
#endif
#if 0
  if (flP->fl_lmops) // && (flP->fl_flags & (FL_LOCKD|FL_NFSv4)))
  {
    if ((flP->fl_flags & FL_SLEEP) &&
         flP->fl_lmops->fl_vfs_callback &&
         flP->fl_type != F_UNLCK)
    {
      vfs_callback = flP->fl_lmops->fl_vfs_callback;
      cmd = F_SETLKW;
    }
  }
#endif
#endif

  lckdat.l_whence = SEEK_SET;

  rc = gpfs_ops.gpfsFcntl(NULL,    // KernelOperation initialized in gpfsFcntl
                          privVfsP,
                          NULL,     // struct vnode *vP or NULL
                                    // advObjP (advisory lock object) is inode
                          fP->f_dentry->d_inode,
                          flP,      // struct file_lock
                          cnP,
                          0,        // offset
                          &lckdat,  // struct cxiFlock_t
                          cmd,
                          vfs_callback, // lockd callback
                          &localRetryId,
                          &eCred);

xerror:

  TRACE2(TRACE_VNODE, 11, TRCID_LINUXOPS_LOCKCTL_DIAG2,
         "gpfs_f_lock: fP 0x%lX, f_dentry 0x%lX",
         fP, fP->f_dentry);

  VFS_STAT_STOP;

  TRACE1(TRACE_VNODE, 1, TRCID_LINUXOPS_LOCKCTL_EXIT,
         "gpfs_f_lock exit: rc %d",
         rc);

  EXIT(0);
  return (-rc);
}

#ifdef NFS4_CLUSTER
/*
 * cmd: F_SETLKW or F_SETLK
 * fl->fl_flags = FL_FLOCK; if sys_flock(), can use other flags for NFSv4
 * fl->fl_start = 0, fl->fl_end = OFFSET_MAX;
 * fl->fl_type =
 *  LOCK_MAND   allow other processes read
 *  or LOCK_MAND&LOCK_RW  allow other processes read and write
 *  or F_RDLCK    LOCK_SH -- a shared lock.
 *  or F_WRLCK    LOCK_EX -- an exclusive lock.
 *  or F_UNLCK    LOCK_UN -- remove an existing lock.
 *
 *  LOCK_MAND -- a `mandatory' flock. This exists to emulate Windows Share Modes.
 */
int
gpfs_f_flock(struct file *fP, int cmd, struct file_lock *flP)
{
  int rc = 0;
  struct gpfsVfsData_t *privVfsP;
  int shareWant = 0;
  int flags = 0;
  struct MMFSVInfo *vinfoP = (struct MMFSVInfo *)fP->private_data;
  privVfsP = VP_TO_PVP(fP->f_dentry->d_inode);
  DBGASSERT(privVfsP != NULL);

  VFS_STAT_START(flockCall);
  ENTER(0);

  TRACE5(TRACE_VNODE, 1, TRCID_LINUX_FLOCK_ENTER,
        "gpfs_f_flock: enter fP 0x%lX flP 0x%lX cmd %d flags 0x%X type 0x%X\n",
         fP, flP, cmd, flP->fl_flags, flP->fl_type);

  if ((cmd != F_SETLK) && (cmd != F_SETLKW))
  {
    rc = ENOSYS;
    goto xerror;
  }

  shareWant |= ALLOW_SHARE_DELETE;

  if (flP->fl_flags & FL_FLOCK) {
    /* Translate (and validate) the type arguments to our shareWant */
    if (flP->fl_type & LOCK_MAND) {
       if (flP->fl_type & LOCK_RW)
           shareWant |= coRead|coWriteM|coWriteA;
       else
           shareWant |= coRead|coWriteM|coWriteA|coDenyR;
    }
    else {
      switch (flP->fl_type) {
      case F_RDLCK:               /* LOCK_SH */
        shareWant |= coRead|coDenyWM|coDenyWA;
        break;
      case F_WRLCK:               /* LOCK_EX */
        shareWant |= coRead|coWriteM|coWriteA|coDenyWM|coDenyWA|coDenyR;
        break;
      case F_UNLCK:               /* LOCK_UN */
        flags |= RESERVE_DOWNGRADE;
        shareWant |= 0;
        break;
      default:
        rc = EINVAL;
        goto xerror;
      }
    }
  }
  else {

//??? add code for NFSv4 shares and delegations
    shareWant = coNFS4Share;
    if (!(flP->fl_flags & FL_SLEEP))
      flags |= RESERVE_NOWAIT;

  }
  /* Call to make the reservation */
  rc = gpfs_ops.gpfsReserveShare(fP, vinfoP, privVfsP, flags, shareWant,
                                 NULL, NULL);

xerror:

 TRACE1(TRACE_VNODE, 1, TRCID_LINUX_FLOCK_EXIT,
         "gpfs_f_flock: exit rc %d\n", rc);

  if (rc)
    cxiErrorNFS(rc);

  VFS_STAT_STOP;
  EXIT(0);
  return (-rc);
}
#endif /* NFS4_CLUSTER */

static inline ssize_t
rdwrInternal(struct file *fP, cxiRdWr_t op, const struct cxiIovec_t *iovecP,
             unsigned long count, loff_t *offsetP)
{
  int i, rc;
  Boolean gotBKL = false;
  ssize_t total_len = 0;
  struct cxiUio_t tmp_uio;
  int flags = cxiOpenFlagsXlate(fP->f_flags);
  struct gpfsVfsData_t *privVfsP;
  cxiNode_t *cnP;
  struct MMFSVInfo *vinfoP = (struct MMFSVInfo *)fP->private_data;
  struct inode *iP;
  ext_cred_t eCred;
  ssize_t tmp_len;

  VFS_STAT_START((op == CXI_READ)? readCall: writeCall);
  ENTER(0);
  DBGASSERT(fP != NULL);
  iP = fP->f_dentry->d_inode;
  DBGASSERT(iP != NULL);

  TRACE11(TRACE_VNODE, 1, TRCID_LINUXOPS_RDWRINT_ENTER,
          "gpfs_f_rdwr enter: fP 0x%lX f_flags 0x%X flags 0x%X op %d "
          "iovec 0x%lX count %d offset 0x%llX "
          "dentry 0x%lX private 0x%lX iP 0x%lX name '%s'\n",
          fP, fP->f_flags, flags, op, iovecP, count, *offsetP, fP->f_dentry,
          fP->private_data, fP->f_dentry->d_inode, fP->f_dentry->d_name.name);

  /* BKL is not held at entry, except for NFS calls */
  TraceBKL();
  if (current->lock_depth >= 0)  /* kernel lock is held by me */
  {
    gotBKL = true;
    unlock_kernel();
  }

  privVfsP = VP_TO_PVP(iP);
  DBGASSERT(privVfsP != NULL);
  cnP = VP_TO_CNP(iP);

  tmp_uio.uio_iov = (struct cxiIovec_t *)iovecP; /* ptr to iovec struct array */
  tmp_uio.uio_iovcnt = count;         /* #iovec elements left to be processed */
  tmp_uio.uio_iovdcnt = 0;            /* #iovec elements already processed    */
  tmp_uio.uio_offset = *offsetP;      /* byte offset in file/dev to read/write*/
  tmp_uio.uio_segflg = UIO_USERSPACE; /* copy to user space                   */
  tmp_uio.uio_fmode = 0;              /* file modes from open file struct     */

  for (i = 0; i < count; i++)
    total_len += iovecP[i].iov_len;

  tmp_uio.uio_resid = total_len; /* #bytes left in data area */

 /* We should -EINVAL if total length is not >= 0 
  * Be careful here because uio_resid is a unsigned 
  * long not an ssize_t 
  */
  tmp_len = (ssize_t)tmp_uio.uio_resid;
  if ( tmp_len  < 0)
  {
    EXIT(0);
    return (-EINVAL);
  }

  DBGASSERT(vinfoP != NULL);

  setCred(&eCred);
  if (op == CXI_READ)
    rc = gpfs_ops.gpfsRead(privVfsP, NULL, cnP, flags, &tmp_uio,
                           vinfoP, NULL, NULL, &eCred, false, true);
  else
  {
    rc = gpfs_ops.gpfsWrite(privVfsP, NULL, cnP, flags, &tmp_uio,
                            vinfoP, NULL, NULL, &eCred, false, true);
    iP->i_sb->s_dirt = 1;
  }

  TRACE5(TRACE_VNODE, 1, TRCID_LINUXOPS_RDWRINT_EXIT,
         "gpfs_f_rdwr exit: fP 0x%lX total_len %d uio_resid %ld "
         "offset 0x%llX rc %d\n", fP, total_len, tmp_uio.uio_resid,
         tmp_uio.uio_offset, rc);

  VFS_STAT_STOP;

  if (gotBKL)        /* If held kernel lock on entry then reacquire it */
    lock_kernel();

  if (rc)
  {
    cxiErrorNFS(rc);

    EXIT(0);
    return (-rc);
  }

  *offsetP = tmp_uio.uio_offset;
  EXIT(0);
  return (total_len - tmp_uio.uio_resid);
}

ssize_t
gpfs_f_read(struct file *fP, char *bufP, size_t count,
            loff_t *offsetP)
{
  ssize_t rc;
  cxiIovec_t tmp_iovec;

  ENTER(0);
  tmp_iovec.iov_base = bufP;    /* base memory address                  */
  tmp_iovec.iov_len = count;    /* length of transfer for this area     */

  rc = rdwrInternal(fP, CXI_READ, &tmp_iovec, 1, offsetP);

  EXIT(0);
  return rc;
}

ssize_t
gpfs_f_dir_read(struct file *fP, char *bufP, size_t count,
                loff_t *offsetP)
{
  TRACE1(TRACE_VNODE, 1, TRCID_LINUXOPS_READ_DIR,
         "gpfs_f_dir_read: fP 0x%lX EISDIR\n", fP);
  return -EISDIR;
}

ssize_t
gpfs_f_write(struct file *fP, const char *bufP, size_t count,
             loff_t *offsetP)
{
  ssize_t rc;
  cxiIovec_t tmp_iovec;

  ENTER(0);
  tmp_iovec.iov_base = (char *)bufP; /* base memory address              */
  tmp_iovec.iov_len = count;         /* length of transfer for this area */

  rc = rdwrInternal(fP, CXI_WRITE, &tmp_iovec, 1, offsetP);

  EXIT(0);
  return rc;
}

ssize_t
gpfs_f_readv(struct kiocb *iocb, const struct iovec *iovecP,
             unsigned long count, loff_t offsetP)
{
  int rc;
  ENTER(0);
  rc = rdwrInternal(iocb->ki_filp, CXI_READ, (const struct cxiIovec_t *)iovecP,
                    count, &offsetP);
  EXIT(0);
  return rc;
}

ssize_t
gpfs_f_writev(struct kiocb *iocb, const struct iovec *iovecP,
              unsigned long count, loff_t offsetP)
{
  int rc;
  ENTER(0);
  rc = rdwrInternal(iocb->ki_filp, CXI_WRITE, (const struct cxiIovec_t *)iovecP,
                    count, &offsetP);
  EXIT(0);
  return rc;
}

#ifdef NFS4_CLUSTER
int
gpfs_f_share(struct file *fP, unsigned int share_access, unsigned int share_deny)
{
  int err;
  struct inode *iP;
  struct dentry *dentryP;
  int shareHave, shareWant;
  struct gpfsVfsData_t *privVfsP;
  ext_cred_t eCred;
  cxiNode_t *cnP;
  int flags = RESERVE_NONE;
 
  ENTER(0);
  err = 0;
  dentryP = fP? fP->f_dentry: NULL;
  iP = dentryP? dentryP->d_inode: NULL;

  TRACE5(TRACE_VNODE, 1, TRCID_LINUXOPS_SHARE_ENTER,
         "gpfs_f_share enter: fP 0x%lX ino %d (%s) access 0x%X deny 0x%X\n",
         fP, iP? iP->i_ino: -1,
         dentryP? dentryP->d_name.name: (const unsigned char*)"",
         share_access, share_deny);

  if (fP)
    get_file(fP);

  /* Validate the file and obtain privVfsP */
  if (!iP || !(privVfsP = VP_TO_PVP(iP))) {
    err = EBADF;
    goto xerror;
  }

  if ((share_access == 0) && (share_deny == 0))
  {
    /* This type of request can happen after the server recalls a delegation.
     * We reject the request which we recognize since no access/deny flags 
     * are given.  This then causes the client to open the file at the server
     * (no delegation) and continue. */

    TRACE3(TRACE_VNODE, 3, TRCID_LINUXOPS_SHARE_RESET,
           "gpfs_f_share: RESET (fP 0x%lX iP 0x%lX privVfsP 0x%lX)\n",
           fP, iP, iP? VP_TO_PVP(fP->f_dentry->d_inode): NULL);

    goto xerror;
  }

  /* Translate (and validate) the NFS4 share/deny arguments to our shareWant */

  /* setup for the XLATE_NFS4 calls */
  err = EINVAL;
  shareWant = coNFS4Share|ALLOW_SHARE_DELETE;

  XLATE_NFS4_ACCESS(share_access, shareWant);
  XLATE_NFS4_DENY(share_deny, shareWant);

  setCred(&eCred);
  cnP = VP_TO_CNP(iP);
  /* Call to make the reservation */
  err = gpfs_ops.gpfsReserveShare(fP, fP->private_data, privVfsP, flags,
                                  shareWant, cnP, &eCred);

xerror:
  if (fP) fput(fP);

  TRACE1(TRACE_VNODE, 1, TRCID_LINUXOPS_SHARE_EXIT, 
         "gpfs_f_share exit: rc %d\n", err);
  EXIT(0);
  return err;
}
#endif /* NFS4_CLUSTER */

extern int cleanupFD;

/* gpfs_f_cleanup is a routine that runs when the last mmfsd
   process terminates.  It allows us to do some basic cleanup
   so that the daemon can be restarted nicely. */

int gpfs_f_cleanup(struct inode *iP, struct file *fP)
{
  int rc = 0;

  ENTER(0);
  if (cleanupFD)
  {
    rc = gpfs_ops.gpfsCleanup();
    cleanupFD = 0;
  }
  EXIT(0);
  return rc;
}

#if LINUX_KERNEL_VERSION >= 2060000 || defined(SUSE_LINUX)
/* gpfs_f_direct_IO() is never called. Open currently requires a "value" in 
 * gpfs_aops->direct_IO to be successful when O_DIRECT is supplied on the open
 * call. The linux "generic" file routines eventually call this op. We do not use
 * the generic file routines so gpfs_f_direct_IO is never called.
 */
#if LINUX_KERNEL_VERSION >= 2060000
ssize_t gpfs_f_direct_IO(int rw, struct kiocb *iocb, const struct iovec *iovecP,
                         loff_t in_offset, unsigned long count)
#elif defined(SUSE_LINUX)
int gpfs_f_direct_IO(int rw, struct file *file, struct kiobuf *kiobuf,
                        unsigned long in_offset, int count)
#endif
{
  LOGASSERT(!"gpfs_f_direct_IO not supported");
}

#endif

#ifdef P_NFS4
static void printfh2(char *s, int *fh)
{
#ifdef GPFS_PRINTK
  printk("%s: %d: %08x %08x %08x %08x %08x %08x %08x %08x %08x\n",
           s, fh[0],fh[1],fh[2],fh[3],fh[4],fh[5],fh[6],fh[7],fh[8],fh[9]);
#endif
}

int gpfs_get_devicelist(struct super_block *sbP, void *p)
{
  int rc = 0;
  struct gpfsVfsData_t *privVfsP = (struct gpfsVfsData_t *)SBLOCK_PRIVATE(sbP);

//  VFS_STAT_START(gpfs_get_devicelist);
  ENTER(0);
  DBGASSERT(privVfsP != NULL);

  TRACE2(TRACE_VNODE, 1, TRCID_LINUXOPS_GET_DEVICELIST_ENTER,
         "gpfs_get_devicelist: sbP 0x%lX p 0x%lX\n", sbP, p);

  rc = gpfs_ops.gpfsGetDeviceList(privVfsP, p);
#ifdef GPFS_PRINTK
  printk("gpfs_get_devicelist: rc %d\n", rc);
#endif

xerror:
  TRACE1(TRACE_VNODE, 1, TRCID_GET_DEVICELIST_EXIT,
         "gpfs_get_devicelist exit: rc %d\n", rc);

//  VFS_STAT_STOP;
  EXIT(0);
  return (-rc);
}

int
gpfs_layout_get(struct inode *iP, void *p)
{
  int rc = 0;
  int code = 0;
  cxiNode_t *cnP;
  ext_cred_t eCred;
  struct gpfsVfsData_t *privVfsP;
  struct nfsd4_pnfs_layoutget *lgp = (struct nfsd4_pnfs_layoutget *)p;

//  VFS_STAT_START(gpfs_layout_get);
  ENTER(0);
  cnP = VP_TO_CNP(iP);
  privVfsP = VP_TO_PVP(iP);
  DBGASSERT(privVfsP != NULL);

  TRACE2(TRACE_VNODE, 1, TRCID_LINUXOPS_GET_LAYOUT_ENTER,
         "gpfs_layout_get: iP 0x%lX p 0x%lX\n", iP, p);

  setCred(&eCred);

  rc = gpfs_ops.gpfsGetLayout(privVfsP, cnP, p, &eCred);

xerror:
  TRACE2(TRACE_VNODE, 1, TRCID_GET_LAYOUT_EXIT,
         "gpfs_layout_get exit: code %d rc %d\n", code, rc);

//  VFS_STAT_STOP;
  EXIT(0);
  return (-rc);
}

/* pNFS: return layout type */
#define LAYOUT_NFSV4_FILES 1
int
gpfs_layout_type()
{
#ifdef GPFS_PRINTK
  printk("xxx gpfs_layout_type LAYOUT_NFSV4_FILES=%d\n", LAYOUT_NFSV4_FILES);
#endif
  return LAYOUT_NFSV4_FILES;
}

gpfs_get_state(struct inode *iP, void *fh, void *p)
{
  int rc = 0;
  int nodeId, len, code = 0;
  cxiNode_t *cnP;
  ext_cred_t eCred;
  struct gpfsVfsData_t *privVfsP;
  struct pnfs_get_state *osP = (struct pnfs_get_state *)p;
  struct knfsd_fh *fhP = (struct knfsd_fh *)fh;

//  VFS_STAT_START(gpfs_get_state);
  ENTER(0);
  cnP = VP_TO_CNP(iP);
  privVfsP = VP_TO_PVP(iP);
  DBGASSERT(privVfsP != NULL);

#ifdef GPFS_PRINTK
  printk("gpfs_get_state iP %p fh type %d fh size %d\n",
          iP, fhP->fh_fsid_type, fhP->fh_size);
  printfh2("gpfs_get_state:", (int *)fhP);
#endif
  len = sizeof(struct pnfs_get_state);
  if (fhP->fh_fsid_type >= max_fsid_type && fhP->fh_size > 8) {
    nodeId = fhP->fh_base.fh_pad[(fhP->fh_size >> 2) -1];
  }
  else {
    rc = ENOENT;
    goto xerror;
  }
  setCred(&eCred);
  osP->devid = gpfs_ops.gpfsGetMyDevID(privVfsP);

  rc = gpfs_ops.gpfsGetOpenState(privVfsP, cnP, nodeId, p, len, &eCred);

  TRACE7(TRACE_VNODE, 2, TRCID_LINUXOPS_SET_STATEID_ENTER,
      "gpfs_get_state: iP 0x%lX mds %x ds %x p 0x%lX len %d verf 0x%lX:0x%lX\n",
       iP, nodeId, osP->devid, p, len, osP->verifier[0], osP->verifier[1]);
#ifdef GPFS_PRINTK
  printk("gpfs_get_state mds-id %x my-id %x verifier %x:%x\n",
                        nodeId, osP->devid, osP->verifier[0], osP->verifier[1]);
#endif

xerror:
  TRACE2(TRACE_VNODE, 1, TRCID_SET_STETEID_EXIT,
         "gpfs_get_state exit: code %d rc %d\n", code, rc);

//  VFS_STAT_STOP;
  EXIT(0);
  return (-rc);
}

int
gpfs_layout_return(struct inode *iP, void *p)
{
  int rc = 0;
  cxiNode_t *cnP;
  struct gpfsVfsData_t *privVfsP;

//  VFS_STAT_START(gpfs_layout_return);
  ENTER(0);
  cnP = VP_TO_CNP(iP);
  privVfsP = VP_TO_PVP(iP);
  DBGASSERT(privVfsP != NULL);

#ifdef GPFS_PRINTK
  printk("gpfs_layout_return iP %p\n", iP);
#endif
  TRACE1(TRACE_VNODE, 1, TRCID_LINUXOPS_LAYOUT_RET_ENTER,
         "gpfs_layout_return: iP 0x%lX\n", iP);

  rc = gpfs_ops.gpfsLayoutRetrun(privVfsP, cnP, p, sizeof(struct layout_return));

xerror:
  TRACE2(TRACE_VNODE, 1, TRCID_LAYOUT_RET_EXIT,
         "gpfs_layout_return exit: code %d iP 0x%lX\n", rc, iP);

//  VFS_STAT_STOP;
  EXIT(0);
  return (-rc);
}
int
gpfs_get_deviceinfo(struct super_block *sbP, void *p)
{
  int rc;
  struct gpfsVfsData_t *privVfsP = (struct gpfsVfsData_t *)SBLOCK_PRIVATE(sbP);

//  VFS_STAT_START(gpfs_get_deviceinfo);
  ENTER(0);
  DBGASSERT(privVfsP != NULL);

  TRACE2(TRACE_VNODE, 1, TRCID_LINUXOPS_GET_DEVICEINFO_ENTER,
         "gpfs_get_deviceinfo: sbP 0x%lX p 0x%lX\n", sbP, p);

#ifdef GPFS_PRINTK
  printk("gpfs_get_deviceinfo: sbP 0x%lX p 0x%lX\n", sbP, p);
#endif
  rc = gpfs_ops.gpfsGetDeviceInfo(privVfsP, p);
#ifdef GPFS_PRINTK
  printk("gpfs_get_deviceinfo: rc %d\n", rc);
#endif

xerror:
  TRACE1(TRACE_VNODE, 1, TRCID_GET_DEVICEINFO_EXIT,
         "gpfs_get_deviceinfo exit: rc %d\n", rc);

//  VFS_STAT_STOP;
  EXIT(0);
  return (-rc);
}

void gpfs_get_verifier(struct super_block *sbP, u32 *p)
{
  int rc;
  struct gpfsVfsData_t *privVfsP = (struct gpfsVfsData_t *)SBLOCK_PRIVATE(sbP);

//  VFS_STAT_START(gpfs_get_verifier);
  ENTER(0);
  DBGASSERT(privVfsP != NULL);

#ifdef GPFS_PRINTK
  printk("gpfs_get_verifier: sbP 0x%lX p 0x%lX\n", sbP, p);
#endif

  gpfs_ops.gpfsGetVerifier(privVfsP, p);

#ifdef GPFS_PRINTK
  printk("gpfs_get_verifier: sbP 0x%lX v1 0x%lX v2 0x%lX\n", sbP, *p, *(p+1));
#endif

//  VFS_STAT_STOP;
  EXIT(0);
  return;
}

#endif