XrdOfs.cc

Go to the documentation of this file.

/******************************************************************************/
/*                                                                            */
/*                             X r d O f s . c c                              */
/*                                                                            */
/* (c) 2004 by the Board of Trustees of the Leland Stanford, Jr., University  */
/*   Produced by Andrew Hanushevsky for Stanford University under contract    */
/*               DE-AC02-76-SFO0515 with the Deprtment of Energy              */
/*                                                                            */
/* This file is part of the XRootD software suite.                            */
/*                                                                            */
/* XRootD is free software: you can redistribute it and/or modify it under    */
/* the terms of the GNU Lesser General Public License as published by the     */
/* Free Software Foundation, either version 3 of the License, or (at your     */
/* option) any later version.                                                 */
/*                                                                            */
/* XRootD is distributed in the hope that it will be useful, but WITHOUT      */
/* ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or      */
/* FITNESS FOR A PARTICULAR PURPOSE.  See the GNU Lesser General Public       */
/* License for more details.                                                  */
/*                                                                            */
/* You should have received a copy of the GNU Lesser General Public License   */
/* along with XRootD in a file called COPYING.LESSER (LGPL license) and file  */
/* COPYING (GPL license).  If not, see <http://www.gnu.org/licenses/>.        */
/*                                                                            */
/* The copyright holder's institutional names and contributor's names may not */
/* be used to endorse or promote products derived from this software without  */
/* specific prior written permission of the institution or contributor.       */
/******************************************************************************/
 
#include <unistd.h>
#include <dirent.h>
#include <cerrno>
#include <fcntl.h>
#include <memory.h>
#include <cstring>
#include <cstdio>
#include <ctime>
#include <netdb.h>
#include <cstdlib>
#include <memory>
#include <sys/param.h>
#include <sys/stat.h>
#include <sys/time.h>
#include <sys/types.h>
 
#include "XProtocol/XProtocol.hh"
 
#include "XrdCks/XrdCks.hh"
#include "XrdCks/XrdCksConfig.hh"
#include "XrdCks/XrdCksData.hh"
 
#include "XrdNet/XrdNetAddr.hh"
#include "XrdNet/XrdNetIF.hh"
#include "XrdNet/XrdNetUtils.hh"
 
#include "XrdOfs/XrdOfs.hh"
#include "XrdOfs/XrdOfsChkPnt.hh"
#include "XrdOfs/XrdOfsConfigCP.hh"
#include "XrdOfs/XrdOfsEvs.hh"
#include "XrdOfs/XrdOfsHandle.hh"
#include "XrdOfs/XrdOfsPoscq.hh"
#include "XrdOfs/XrdOfsPrepare.hh"
#include "XrdOfs/XrdOfsTrace.hh"
#include "XrdOfs/XrdOfsSecurity.hh"
#include "XrdOfs/XrdOfsStats.hh"
#include "XrdOfs/XrdOfsTPC.hh"
 
#include "XrdCms/XrdCmsClient.hh"
 
#include "XrdOss/XrdOss.hh"
 
#include "XrdSys/XrdSysError.hh"
#include "XrdSys/XrdSysHeaders.hh"
#include "XrdSys/XrdSysLogger.hh"
#include "XrdSys/XrdSysPlatform.hh"
#include "XrdSys/XrdSysPthread.hh"
#include "XrdSys/XrdSysRAtomic.hh"
 
#include "XrdOuc/XrdOuca2x.hh"
#include "XrdOuc/XrdOucEnv.hh"
#include "XrdOuc/XrdOucERoute.hh"
#include "XrdOuc/XrdOucLock.hh"
#include "XrdOuc/XrdOucMsubs.hh"
#include "XrdOuc/XrdOucPgrwUtils.hh"
#include "XrdOuc/XrdOucTList.hh"
#include "XrdOuc/XrdOucTPC.hh"
#include "XrdSec/XrdSecEntity.hh"
#include "XrdSec/XrdSecEntityAttr.hh"
#include "XrdSfs/XrdSfsAio.hh"
#include "XrdSfs/XrdSfsFlags.hh"
#include "XrdSfs/XrdSfsInterface.hh"
 
#ifdef AIX
#include <sys/mode.h>
#endif
 
#ifndef O_DIRECT
#define O_DIRECT 0
#endif
 
/******************************************************************************/
/*                  E r r o r   R o u t i n g   O b j e c t                   */
/******************************************************************************/
 
XrdSysError      OfsEroute(0);
 
XrdSysTrace      OfsTrace("ofs");
 
/******************************************************************************/
/*               S t a t i s t i c a l   D a t a   O b j e c t                */
/******************************************************************************/
  
XrdOfsStats      OfsStats;
 
/******************************************************************************/
/*                       L o c a l   F u n c t i o n s                        */
/******************************************************************************/
 
namespace
{
bool VerPgw(const char *buf, ssize_t off, size_t len, const uint32_t* csv,
            XrdOfsHandle *oh,  XrdOucErrInfo  &error)
{
   EPNAME("VerPgw");
   XrdOucPgrwUtils::dataInfo dInfo(buf, csv, off, len);
   off_t badoff;
   int   badlen;
 
// Verify incoming checksums
//
   if (!XrdOucPgrwUtils::csVer(dInfo, badoff, badlen))
      {char eMsg[512];
       int n;
       n = snprintf(eMsg, sizeof(eMsg), "Checksum error at offset %lld.", (long long) badoff);
       error.setErrInfo(EDOM, eMsg);
       eMsg[n-1] = 0;
       OfsEroute.Emsg(epname, eMsg, "aborted pgwrite to", oh->Name());
       return false;
      }
   return true;
}
}
  
/******************************************************************************/
/*                        S t a t i c   O b j e c t s                         */
/******************************************************************************/
  
XrdOfsHandle     *XrdOfs::dummyHandle;
 
int               XrdOfs::MaxDelay = 60;
int               XrdOfs::OSSDelay = 30;
 
/******************************************************************************/
/*                    F i l e   S y s t e m   O b j e c t                     */
/******************************************************************************/
  
extern XrdOfs* XrdOfsFS;
 
/******************************************************************************/
/*                 S t o r a g e   S y s t e m   O b j e c t                  */
/******************************************************************************/
  
XrdOss *XrdOfsOss;
 
/******************************************************************************/
/*                    X r d O f s   C o n s t r u c t o r                     */
/*****************************************************************************/
 
XrdOfs::XrdOfs() : dMask{0000,0775}, fMask{0000,0775}, // Legacy
                   tpcRdrHost{}, tpcRdrPort{}
{
   const char *bp;
 
// Establish defaults
//
   ofsConfig     = 0;
   FSctl_PC      = 0;
   FSctl_PI      = 0;
   Authorization = 0;
   Finder        = 0;
   Balancer      = 0;
   evsObject     = 0;
   ossRPList     = 0;
   myRole        = strdup("server");
   OssIsProxy    = 0;
   ossRW         =' ';
   ossFeatures   = 0;
 
// Obtain port number we will be using. Note that the constructor must occur
// after the port number is known (i.e., this cannot be a global static).
//
   myPort = (bp = getenv("XRDPORT")) ? strtol(bp, (char **)NULL, 10) : 0;
 
// Defaults for POSC
//
   poscQ   = 0;
   poscLog = 0;
   poscHold= 10*60;
   poscAuto= 0;
   poscSync= 1;
 
// Set the configuration file name and dummy handle
//
   ConfigFN = 0;
   XrdOfsHandle::Alloc(&dummyHandle);
 
// Set checksum pointers
//
   Cks       = 0;
   CksPfn    = true;
   CksRdr    = true;
 
// Prepare handling
//
   prepHandler = 0;
   prepAuth = true;
 
// Eextended attribute limits
//
   usxMaxNsz = kXR_faMaxNlen;
   usxMaxVsz = kXR_faMaxVlen;
 
// Other options
//
   DirRdr    = false;
   reProxy   = false;
   OssHasPGrw= false;
}
 
/******************************************************************************/
/*                                                                            */
/*           D i r e c t o r y   O b j e c t   I n t e r f a c e s            */
/*                                                                            */
/******************************************************************************/
/******************************************************************************/
/*                                  o p e n                                   */
/******************************************************************************/
 
int XrdOfsDirectory::open(const char              *dir_path, // In
                          const XrdSecEntity      *client,   // In
                          const char              *info)      // In
/*
  Function: Open the directory `path' and prepare for reading.
 
  Input:    path      - The fully qualified name of the directory to open.
            client    - Authentication credentials, if any.
            info      - Opaque information to be used as seen fit.
 
  Output:   Returns SFS_OK upon success, otherwise SFS_ERROR.
 
  Notes: 1. The code here assumes that directory file descriptors are never
            shared. Hence, no locks need to be obtained. It works out that
            lock overhead is worse than have a duplicate file descriptor for
            very short durations.
*/
{
   EPNAME("opendir");
   static const int od_mode = SFS_O_RDONLY|SFS_O_META;
   XrdOucEnv Open_Env(info,0,client);
   int retc;
 
// Trace entry
//
   XTRACE(opendir, dir_path, "");
 
// Verify that this object is not already associated with an open directory
//
   if (dp) return
      XrdOfsFS->Emsg(epname, error, EADDRINUSE, "open directory", dir_path);
 
// Apply security, as needed
//
   AUTHORIZE(client,&Open_Env,AOP_Readdir,"open directory",dir_path,error);
 
// Find out where we should open this directory
//
   if (XrdOfsFS->DirRdr && XrdOfsFS->Finder && XrdOfsFS->Finder->isRemote()
   &&  (retc = XrdOfsFS->Finder->Locate(error, dir_path, od_mode, &Open_Env)))
      return XrdOfsFS->fsError(error, retc);
 
// Open the directory and allocate a handle for it
//
   if (!(dp = XrdOfsOss->newDir(tident))) retc = -ENOMEM;
      else if (!(retc = dp->Opendir(dir_path, Open_Env)))
              {fname = strdup(dir_path);
               return SFS_OK;
              }
              else {delete dp; dp = 0;}
 
// Encountered an error
//
   return XrdOfsFS->Emsg(epname, error, retc, "open directory", dir_path);
}
 
/******************************************************************************/
/*                             n e x t E n t r y                              */
/******************************************************************************/
 
const char *XrdOfsDirectory::nextEntry()
/*
  Function: Read the next directory entry.
 
  Input:    n/a
 
  Output:   Upon success, returns the contents of the next directory entry as
            a null terminated string. Returns a null pointer upon EOF or an
            error. To differentiate the two cases, getErrorInfo will return
            0 upon EOF and an actual error code (i.e., not 0) on error.
 
  Notes: 1. The code here assumes that idle directory file descriptors are
            *not* closed. This needs to be the case because we need to return
            non-duplicate directory entries. Anyway, the xrootd readdir protocol
            is handled internally so directories should never be idle.
         2. The code here assumes that directory file descriptors are never
            shared. Hence, no locks need to be obtained. It works out that
            lock overhead is worse than have a duplicate file descriptor for
            very short durations.
*/
{
   EPNAME("readdir");
   int retc;
 
// Check if this directory is actually open
//
   if (!dp) {XrdOfsFS->Emsg(epname, error, EBADF, "read directory");
             return 0;
            }
 
// Check if we are at EOF (once there we stay there)
//
   if (atEOF) return 0;
 
// Read the next directory entry
//
   if ((retc = dp->Readdir(dname, sizeof(dname))) < 0)
      {XrdOfsFS->Emsg(epname, error, retc, "read directory", fname);
       return 0;
      }
 
// Check if we have reached end of file
//
   if (!*dname)
      {atEOF = 1;
       error.clear();
       XTRACE(readdir, fname, "<eof>");
       return 0;
      }
 
// Return the actual entry
//
   XTRACE(readdir, fname, dname);
   return (const char *)(dname);
}
 
/******************************************************************************/
/*                                 c l o s e                                  */
/******************************************************************************/
  
int XrdOfsDirectory::close()
/*
  Function: Close the directory object.
 
  Input:    n/a
 
  Output:   Returns SFS_OK upon success and SFS_ERROR upon failure.
 
  Notes: 1. The code here assumes that directory file descriptors are never
            shared. Hence, no locks need to be obtained. It works out that
            lock overhead is worse than have a duplicate file descriptor for
            very short durations.
*/
{
   EPNAME("closedir");
   int retc;
 
// Check if this directory is actually open
//
   if (!dp) {XrdOfsFS->Emsg(epname, error, EBADF, "close directory");
             return SFS_ERROR;
            }
   XTRACE(closedir, fname, "");
 
// Close this directory
//
    if ((retc = dp->Close()))
       retc = XrdOfsFS->Emsg(epname, error, retc, "close", fname);
       else retc = SFS_OK;
 
// All done
//
   delete dp;
   dp = 0;
   free(fname);
   fname = 0;
   return retc;
}
 
/******************************************************************************/
/*                              a u t o S t a t                               */
/******************************************************************************/
 
int XrdOfsDirectory::autoStat(struct stat *buf)
/*
  Function: Set stat buffer to automaticaly return stat information
 
  Input:    Pointer to stat buffer which will be filled in on each
            nextEntry() and represent stat information for that entry.
 
  Output:   Upon success, returns zero. Upon error returns SFS_ERROR and sets
            the error object to contain the reason.
 
  Notes: 1. If autoStat() is not supported he caller will need to follow up
            with a manual stat() call for the full path, a slow and tedious
            process. The autoStat function significantly reduces overhead by
            automatically providing stat information for the entry read.
*/
{
   EPNAME("autoStat");
   int retc;
 
// Check if this directory is actually open
//
   if (!dp) {XrdOfsFS->Emsg(epname, error, EBADF, "autostat directory");
             return SFS_ERROR;
            }
 
// Set the stat buffer in the storage system directory but don't complain.
//
    if ((retc = dp->StatRet(buf))) return retc;
    return SFS_OK;
}
  
/******************************************************************************/
/*                                                                            */
/*                F i l e   O b j e c t   I n t e r f a c e s                 */
/*                                                                            */
/******************************************************************************/
/******************************************************************************/
/*                X r d O f s F i l e   C o n s t r u c t o r                 */
/******************************************************************************/
 
XrdOfsFile::XrdOfsFile(XrdOucErrInfo &eInfo, const char *user)
                      : XrdSfsFile(eInfo), tident(user ? user : ""),
                        oh(XrdOfs::dummyHandle), myTPC(0), myCKP(0),
                        dorawio(0), viaDel(0), ckpBad(false) {}
 
/******************************************************************************/
/*                                  o p e n                                   */
/******************************************************************************/
 
int XrdOfsFile::open(const char          *path,      // In
                     XrdSfsFileOpenMode   open_mode, // In
                     mode_t               Mode,      // In
               const XrdSecEntity        *client,    // In
               const char                *info)      // In
/*
  Function: Open the file `path' in the mode indicated by `open_mode'.  
 
  Input:    path      - The fully qualified name of the file to open.
            open_mode - One of the following flag values:
                        SFS_O_RDONLY - Open file for reading.
                        SFS_O_WRONLY - Open file for writing.
                        SFS_O_RDWR   - Open file for update
                        SFS_O_NOTPC  - Disallow TPC opens
                        SFS_O_REPLICA- Open file for replication
                        SFS_O_CREAT  - Create the file open in RW mode
                        SFS_O_TRUNC  - Trunc  the file open in RW mode
                        SFS_O_POSC   - Presist    file on successful close
                        SFS_O_SEQIO  - Primarily sequential I/O (e.g. xrdcp)
            Mode      - The Posix access mode bits to be assigned to the file.
                        These bits correspond to the standard Unix permission
                        bits (e.g., 744 == "rwxr--r--"). Additionally, Mode
                        may contain SFS_O_MKPTH to force creation of the full
                        directory path if it does not exist. This parameter is
                        ignored unless open_mode = SFS_O_CREAT.
            client    - Authentication credentials, if any.
            info      - Opaque information to be used as seen fit.
 
  Output:   Returns SFS_OK upon success, otherwise SFS_ERROR is returned.
*/
{
   EPNAME("open");
   static const int crMask = (SFS_O_CREAT  | SFS_O_TRUNC);
   static const int opMask = (SFS_O_RDONLY | SFS_O_WRONLY | SFS_O_RDWR);
 
   struct OpenHelper
         {const char   *Path;
          XrdOfsHandle *hP;
          XrdOssDF     *fP;
          int           poscNum;
 
          int           OK() {hP = 0; fP = 0; poscNum = 0; return SFS_OK;}
 
                        OpenHelper(const char *path)
                       : Path(path), hP(0), fP(0), poscNum(0) {}
 
                       ~OpenHelper()
                       {int retc;
                        if (hP) hP->Retire(retc);
                        if (fP) delete fP;
                        if (poscNum > 0) XrdOfsFS->poscQ->Del(Path, poscNum, 1);
                       }
         } oP(path);
 
   mode_t theMode = (Mode | XrdOfsFS->fMask[0]) & XrdOfsFS->fMask[1];
   const char *tpcKey;
   int retc, isPosc = 0, crOpts = 0, isRW = 0, open_flag = 0;
   int find_flag = open_mode & (SFS_O_NOWAIT | SFS_O_RESET | SFS_O_MULTIW);
   XrdOucEnv Open_Env(info,0,client);
 
// Trace entry
//
   ZTRACE(open, Xrd::hex1 <<open_mode <<"-" <<Xrd::oct1 <<Mode <<" ("
              <<Xrd::oct1 <<theMode <<") fn=" <<path);
 
// Verify that this object is not already associated with an open file
//
   XrdOfsFS->ocMutex.Lock();
   if (oh != XrdOfs::dummyHandle)
      {XrdOfsFS->ocMutex.UnLock();
       return XrdOfsFS->Emsg(epname,error,EADDRINUSE,"open file",path);
      }
   XrdOfsFS->ocMutex.UnLock();
 
// Handle the open mode options
//
   if (open_mode & crMask)
      {crOpts = (Mode & SFS_O_MKPTH ? XRDOSS_mkpath : 0);
       if (XrdOfsFS->poscQ && ((open_mode & SFS_O_POSC) ||
           XrdOfsFS->poscAuto || Open_Env.Get("ofs.posc")))
           {isPosc = 1; isRW = XrdOfsHandle::opPC;}
          else isRW = XrdOfsHandle::opRW;
       if (open_mode & SFS_O_CREAT)
          {open_flag   = O_RDWR     | O_CREAT  | O_EXCL;
           find_flag  |= SFS_O_RDWR | SFS_O_CREAT | (open_mode & SFS_O_REPLICA);
           crOpts     |= XRDOSS_new;
          } else {
           open_flag  |= O_RDWR     | O_CREAT  | O_TRUNC;
           find_flag  |= SFS_O_RDWR | SFS_O_TRUNC;
          }
      }
   else
   switch(open_mode & opMask)
  {case SFS_O_RDONLY: open_flag = O_RDONLY; find_flag |= SFS_O_RDONLY;
                      break;
   case SFS_O_WRONLY: open_flag = O_WRONLY; find_flag |= SFS_O_WRONLY;
                      isRW = XrdOfsHandle::opRW;
                      if (XrdOfsFS->poscQ && ((open_mode & SFS_O_POSC) ||
                          Open_Env.Get("ofs.posc"))) oP.poscNum = -1;
                      break;
   case SFS_O_RDWR:   open_flag = O_RDWR;   find_flag |= SFS_O_RDWR;
                      isRW = XrdOfsHandle::opRW;
                      if (XrdOfsFS->poscQ && ((open_mode & SFS_O_POSC) ||
                          Open_Env.Get("ofs.posc"))) oP.poscNum = -1;
                      break;
   default:           open_flag = O_RDONLY; find_flag |= SFS_O_RDONLY;
                      break;
  }
 
// Preset TPC handling
//
   tpcKey = Open_Env.Get(XrdOucTPC::tpcKey);
 
// Check if we will be redirecting the tpc request
//
   if (tpcKey && isRW && (XrdOfsFS->Options & XrdOfs::RdrTPC))
      {const char *dOn = Open_Env.Get(XrdOucTPC::tpcDlgOn);
       int k = ((dOn && *dOn == '1') || strcmp(tpcKey, "delegate") ? 1 : 0);
       if (XrdOfsFS->tpcRdrHost[k])
          {error.setErrInfo(XrdOfsFS->tpcRdrPort[k], XrdOfsFS->tpcRdrHost[k]);
           return SFS_REDIRECT;
          }
      }
 
// If we have a finder object, use it to direct the client. The final
// destination will apply the security that is needed
//
   if (XrdOfsFS->Finder && (retc = XrdOfsFS->Finder->Locate(error, path,
                                                   find_flag, &Open_Env)))
      return XrdOfsFS->fsError(error, retc);
 
// Preset TPC handling and if not allowed, complain
//
   if (tpcKey && (open_mode & SFS_O_NOTPC))
      return XrdOfsFS->Emsg(epname, error, EPROTOTYPE, "tpc", path);
 
// Create the file if so requested o/w try to attach the file
//
   if (open_flag & O_CREAT)
      {// Apply security, as needed
       //
       // If we aren't requesting O_EXCL, one needs AOP_Create
       bool overwrite_permitted = true;
       if (!(open_flag & O_EXCL))
          {if (client && XrdOfsFS->Authorization &&
               !XrdOfsFS->Authorization->Access(client, path, AOP_Create, &Open_Env))
              { // We don't have the ability to create a file without O_EXCL.  If we have AOP_Excl_Create,
                // then manipulate the open flags and see if we're successful with it.
                AUTHORIZE(client,&Open_Env,AOP_Excl_Create,"create",path,error);
                overwrite_permitted = false;
                open_flag |= O_EXCL;
                open_flag &= ~O_TRUNC;
              }
          }
       // If we are in O_EXCL mode, then we accept either AOP_Excl_Create or AOP_Create
       else if (client && XrdOfsFS->Authorization &&
            !XrdOfsFS->Authorization->Access(client, path, AOP_Create, &Open_Env))
          {AUTHORIZE(client,&Open_Env,AOP_Excl_Create,"create",path,error);
           // In this case, we don't have AOP_Create but we do have AOP_Excl_Create; note that
           // overwrites are not permitted (this is later used to correct an error code).
           overwrite_permitted = false;
          }
 
       OOIDENTENV(client, Open_Env);
 
       // For ephemeral file, we must enter the file into the queue
       //
       if (isPosc)
          {bool isNew = (open_mode & SFS_O_TRUNC) == 0;
           if ((oP.poscNum = XrdOfsFS->poscQ->Add(tident, path, isNew)) < 0)
              return XrdOfsFS->Emsg(epname, error, oP.poscNum, "pcreate", path);
          }
 
       // Create the file. If ENOTSUP is returned, promote the creation to
       // the subsequent open. This is to accomodate proxy support.
       //
       if ((retc = XrdOfsOss->Create(tident, path, theMode, Open_Env,
                                     ((open_flag << 8) | crOpts))))
          {if (retc > 0) return XrdOfsFS->Stall(error, retc, path);
           if (retc == -EINPROGRESS)
              {XrdOfsFS->evrObject.Wait4Event(path,&error);
               return XrdOfsFS->fsError(error, SFS_STARTED);
              }
           if (retc != -ENOTSUP)
              {// If we tried to overwrite an existing file but do not have the AOP_Create
               // privilege, then ensure we generate a 'permission denied' instead of 'exists'
               if ((open_flag & O_EXCL) && retc == -EEXIST && !overwrite_permitted)
                  {retc = -EACCES;}
               if (XrdOfsFS->Balancer) XrdOfsFS->Balancer->Removed(path);
               return XrdOfsFS->Emsg(epname, error, retc, "create", path);
              }
          } else {
            if (XrdOfsFS->Balancer) XrdOfsFS->Balancer->Added(path, isPosc);
            open_flag  = O_RDWR|O_TRUNC;
            if (XrdOfsFS->evsObject 
            &&  XrdOfsFS->evsObject->Enabled(XrdOfsEvs::Create))
               {XrdOfsEvsInfo evInfo(tident,path,info,&Open_Env,Mode);
                XrdOfsFS->evsObject->Notify(XrdOfsEvs::Create, evInfo);
               }
          }
 
      } else {
 
       // Apply security, as needed
       //
          if (tpcKey && !isRW)
             {XrdOfsTPC::Facts Args(client, &error, &Open_Env, tpcKey, path);
              if ((retc = XrdOfsTPC::Authorize(&myTPC, Args))) return retc;
             } else {AUTHORIZE(client, &Open_Env, (isRW?AOP_Update:AOP_Read),
                               "open", path, error);
                    }
       OOIDENTENV(client, Open_Env);
      }
 
// Get a handle for this file.
//
   if ((retc = XrdOfsHandle::Alloc(path, isRW, &oP.hP)))
      {if (retc > 0) return XrdOfsFS->Stall(error, retc, path);
       return XrdOfsFS->Emsg(epname, error, retc, "attach", path);
      }
 
// If this is a third party copy and we are the destination, then validate
// specification at this point and setup to transfer. Note that if the
// call fails and auto removal is enabled, the file we created will be deleted.
//
   if (tpcKey && isRW)
      {char pfnbuff[MAXPATHLEN+8]; const char *pfnP;
       if (!(pfnP = XrdOfsOss->Lfn2Pfn(path, pfnbuff, MAXPATHLEN, retc)))
          return XrdOfsFS->Emsg(epname, error, retc, "open", path);
       XrdOfsTPC::Facts Args(client, &error, &Open_Env, tpcKey, path, pfnP);
       if ((retc = XrdOfsTPC::Validate(&myTPC, Args))) return retc;
      }
 
// Assign/transfer posc ownership. We may need to delay the client if the
// file create ownership does not match and this is not a create request.
//
   if (oP.hP->isRW == XrdOfsHandle::opPC)
      {if (!isRW) return XrdOfsFS->Stall(error, -1, path);
       if ((retc = oP.hP->PoscSet(tident, oP.poscNum, theMode)))
          {if (retc > 0) XrdOfsFS->poscQ->Del(path, retc);
              else return XrdOfsFS->Emsg(epname, error, retc, "access", path);
          }
      }
 
// If this is a previously existing handle, we are almost done. If this is
// the target of a third party copy request, fail it now. We don't support
// multiple writers in tpc mode (this should really never happen).
//
   if (!(oP.hP->Inactive()))
      {dorawio = (oh->isCompressed && open_mode & SFS_O_RAWIO ? 1 : 0);
       if (tpcKey && isRW)
          return XrdOfsFS->Emsg(epname, error, EALREADY, "tpc", path);
       XrdOfsFS->ocMutex.Lock(); oh = oP.hP; XrdOfsFS->ocMutex.UnLock();
       FTRACE(open, "attach use=" <<oh->Usage());
       if (oP.poscNum > 0) XrdOfsFS->poscQ->Commit(path, oP.poscNum);
       oP.hP->UnLock(); 
       OfsStats.sdMutex.Lock();
       isRW ? OfsStats.Data.numOpenW++ : OfsStats.Data.numOpenR++;
       if (oP.poscNum > 0) OfsStats.Data.numOpenP++;
       OfsStats.sdMutex.UnLock();
       return oP.OK();
      }
 
// Get a storage system object
//
   if (!(oP.fP = XrdOfsOss->newFile(tident)))
      return XrdOfsFS->Emsg(epname, error, ENOMEM, "open", path);
 
// We need to make special provisions for proxy servers in the presence of
// the TPC option and possibly cache as it's handled differently in this case.
//
   if (XrdOfsFS->OssIsProxy)
      {if (myTPC) open_flag |= O_NOFOLLOW;
       if (error.getUCap() & XrdOucEI::uUrlOK &&
           error.getUCap() & XrdOucEI::uLclF) open_flag |= O_DIRECT;
      }
 
// Open the file
//
   if ((retc = oP.fP->Open(path, open_flag, theMode, Open_Env)))
      {if (retc > 0) return XrdOfsFS->Stall(error, retc, path);
       if (retc == -EINPROGRESS)
          {XrdOfsFS->evrObject.Wait4Event(path,&error);
           return XrdOfsFS->fsError(error, SFS_STARTED);
          }
       if (retc == -ETXTBSY) return XrdOfsFS->Stall(error, -1, path);
       if (retc == -EDESTADDRREQ)
          {char *url = Open_Env.Get("FileURL");
           if (url) {error.setErrInfo(-1, url); return SFS_REDIRECT;}
          }
       if (XrdOfsFS->Balancer && retc == -ENOENT)
          XrdOfsFS->Balancer->Removed(path);
       return XrdOfsFS->Emsg(epname, error, retc, "open", path);
      }
 
// Verify that we can actually use this file
//
   if (oP.poscNum > 0)
      {if ((retc = oP.fP->Fchmod(static_cast<mode_t>(theMode|XRDSFS_POSCPEND))))
          return XrdOfsFS->Emsg(epname, error, retc, "fchmod", path);
       XrdOfsFS->poscQ->Commit(path, oP.poscNum);
      }
 
// Set compression values and activate the handle
//
   if (oP.fP->isCompressed() > 0)
      {oP.hP->isCompressed = 1;
       dorawio = (open_mode & SFS_O_RAWIO ? 1 : 0);
      }
   oP.hP->Activate(oP.fP);
   oP.hP->UnLock();
 
// If this is being opened for sequential I/O advise the filesystem about it.
//
#if defined(__linux__) || (defined(__FreeBSD_kernel__) && defined(__GLIBC__))
   if (!(XrdOfsFS->OssIsProxy) && open_mode & SFS_O_SEQIO)
      {static RAtomic_int fadFails(0);
       int theFD =  oP.fP->getFD();
       if (theFD >= 0 && fadFails < 4096)
          if (posix_fadvise(theFD, 0, 0, POSIX_FADV_SEQUENTIAL) < 0)
             {OfsEroute.Emsg(epname, errno, "fadsize for sequential I/O.");
              fadFails++;
             }
      }
#endif
 
// Send an open event if we must
//
   if (XrdOfsFS->evsObject)
      {XrdOfsEvs::Event theEvent = (isRW ? XrdOfsEvs::Openw : XrdOfsEvs::Openr);
       if (XrdOfsFS->evsObject->Enabled(theEvent))
          {XrdOfsEvsInfo evInfo(tident, path, info, &Open_Env);
           XrdOfsFS->evsObject->Notify(theEvent, evInfo);
          }
      }
 
// Maintain statistics
//
   OfsStats.sdMutex.Lock();
   isRW ? OfsStats.Data.numOpenW++ : OfsStats.Data.numOpenR++;
   if (oP.poscNum > 0) OfsStats.Data.numOpenP++;
   OfsStats.sdMutex.UnLock();
 
// All done
//
   XrdOfsFS->ocMutex.Lock(); oh = oP.hP; XrdOfsFS->ocMutex.UnLock();
   return oP.OK();
}
 
/******************************************************************************/
/*                                 c l o s e                                  */
/******************************************************************************/
 
int XrdOfsFile::close()  // In
/*
  Function: Close the file object.
 
  Input:    n/a
 
  Output:   Returns SFS_OK upon success and SFS_ERROR upon failure.
*/
{
   EPNAME("close");
 
   class  CloseFH : public XrdOfsHanCB
         {public: void Retired(XrdOfsHandle *hP) {XrdOfsFS->Unpersist(hP);}};
   static XrdOfsHanCB *hCB = static_cast<XrdOfsHanCB *>(new CloseFH);
 
   XrdOfsHandle *hP;
   int   poscNum, retc, cRetc = 0;
   short theMode;
 
// Trace the call
//
    FTRACE(close, "use=" <<oh->Usage()); // Unreliable trace, no origin lock
 
// Verify the handle (we briefly maintain a global lock)
//
    XrdOfsFS->ocMutex.Lock();
    if (oh == XrdOfs::dummyHandle)
       {XrdOfsFS->ocMutex.UnLock(); return SFS_OK;}
    if ((oh->Inactive()))
       {XrdOfsFS->ocMutex.UnLock();
        return XrdOfsFS->Emsg(epname, error, EBADF, "close file");
       }
    hP = oh; oh = XrdOfs::dummyHandle;
    XrdOfsFS->ocMutex.UnLock();
    hP->Lock();
 
// Delete the tpc object, if any
//
   if (myTPC) {myTPC->Del(); myTPC = 0;}
 
// Maintain statistics
//
   OfsStats.sdMutex.Lock();
   if (!(hP->isRW)) OfsStats.Data.numOpenR--;
      else {OfsStats.Data.numOpenW--;
            if (hP->isRW == XrdOfsHandle::opPC) OfsStats.Data.numOpenP--;
           }
   OfsStats.sdMutex.UnLock();
 
// Do an fsync() on this file prior to closing if it was open r/w
//
   if (ocFsync && hP->isRW) hP->Select().Fsync(); // Ignore errors
 
// If this file was tagged as a POSC then we need to make sure it will persist
// Note that we unpersist the file immediately when it's inactive or if no hold
// time is allowed. Also, close events occur only for active handles. If the
// entry was via delete then we ignore the close return code as there is no
// one to handle it on the other side.
//
   if ((poscNum = hP->PoscGet(theMode, !viaDel)))
      {if (viaDel)
          {if (hP->Inactive() || !XrdOfsFS->poscHold)
              {XrdOfsFS->Unpersist(hP, !hP->Inactive()); hP->Retire(cRetc);}
              else hP->Retire(hCB, XrdOfsFS->poscHold);
           return SFS_OK;
          }
       if ((retc = hP->Select().Fchmod(theMode)))
          XrdOfsFS->Emsg(epname, error, retc, "fchmod", hP->Name());
          else {XrdOfsFS->poscQ->Del(hP->Name(), poscNum);
                if (XrdOfsFS->Balancer) XrdOfsFS->Balancer->Added(hP->Name());
               }
      }
 
// Handle any oustanding checkpoint
//
   if (myCKP)
      {retc =  myCKP->Restore();
       if (retc) XrdOfsFS->Emsg(epname,error,retc,"restore chkpnt",hP->Name());
       myCKP->Finished();
       myCKP = 0;
      }
 
// We need to handle the cunudrum that an event may have to be sent upon
// the final close. However, that would cause the path name to be destroyed.
// So, we have two modes of logic where we copy out the pathname if a final
// close actually occurs. The path is not copied if it's not final and we
// don't bother with any of it if we need not generate an event.
//
   if (XrdOfsFS->evsObject && tident
   &&  XrdOfsFS->evsObject->Enabled(hP->isRW ? XrdOfsEvs::Closew
                                             : XrdOfsEvs::Closer))
      {long long FSize, *retsz;
       char pathbuff[MAXPATHLEN+8];
       XrdOfsEvs::Event theEvent;
       if (hP->isRW) {theEvent = XrdOfsEvs::Closew; retsz = &FSize;}
          else {      theEvent = XrdOfsEvs::Closer; retsz = 0; FSize=0;}
       if (!(hP->Retire(cRetc, retsz, pathbuff, sizeof(pathbuff))))
          {XrdOfsEvsInfo evInfo(tident, pathbuff, "" , 0, 0, FSize);
           XrdOfsFS->evsObject->Notify(theEvent, evInfo);
          }
      } else hP->Retire(cRetc);
 
// All done
//
  return (cRetc ? XrdOfsFS->Emsg(epname, error, cRetc, "close file") : SFS_OK);
}
 
/******************************************************************************/
/*                            c h e c k p o i n t                             */
/******************************************************************************/
 
int XrdOfsFile::checkpoint(XrdSfsFile::cpAct act, struct iov *range, int n)
{
   EPNAME("chkpnt");
   const char *ckpName;
   int rc;
   bool readok;
 
// Make sure we are active
//
   if (oh->Inactive()) return XrdOfsFS->Emsg(epname, error, EBADF,
                                        "handle checkpoint", (const char *)0);
 
// If checkpointing is disabled, the don't accept this request.
//
   if (!XrdOfsConfigCP::Enabled) return XrdOfsFS->Emsg(epname, error, ENOTSUP,
                        "handle disabled checkpoint", (const char *)0);
 
// If this checkpoint is bad then only a delete, query or restore is allowed.
//
   if (ckpBad && (act == XrdSfsFile::cpTrunc || act == XrdSfsFile::cpWrite))
      return XrdOfsFS->Emsg(epname, error, EIDRM, "extend checkpoint "
             "(only delete or restore possible) for", oh->Name());
 
// Handle the request
//
   switch(act)
         {case XrdSfsFile::cpCreate:
               ckpName = "create checkpoint for";
               if ((rc = CreateCKP())) return rc;
               if ((rc = myCKP->Create())) {myCKP->Finished(); myCKP = 0;}
               break;
          case XrdSfsFile::cpDelete:
               ckpName = "delete checkpoint for";
               if (!myCKP) rc = ENOENT;
                  else {rc = myCKP->Delete();
                        myCKP->Finished();
                        myCKP = 0;
                        ckpBad = false;
                       }
               break;
          case XrdSfsFile::cpQuery:
               ckpName = "query checkpoint for";
               if (!range || n <= 0)
                  return XrdOfsFS->Emsg(epname, error, EINVAL,
                                   "query checkpoint limits for", oh->Name());
               rc = (myCKP ? myCKP->Query(*range) : ENOENT);
               break;
          case XrdSfsFile::cpRestore:
               ckpName = "restore checkpoint for";
               if (!myCKP) rc = ENOENT;
                  else {if (!(rc = myCKP->Restore(&readok)))
                           {myCKP->Finished();
                            myCKP  = 0;
                            ckpBad = false;
                           } else {
                            if (!(oh->Select().DFType() & XrdOssDF::DF_isProxy))
                               oh->Suppress((readok ? 0 : -EDOM));
                            ckpBad = true;
                           }
                       }
               break;
          case XrdSfsFile::cpTrunc:
               ckpName = "checkpoint truncate";
                    if (!range) rc = EINVAL;
               else if (!myCKP) rc = ENOENT;
               else if ((rc = myCKP->Truncate(range))) ckpBad = true;
               break;
          case XrdSfsFile::cpWrite:
               ckpName = "checkpoint write";
                    if (!range || n <= 0) rc = EINVAL;
               else if (!myCKP)           rc = ENOENT;
               else if ((rc = myCKP->Write(range, n))) ckpBad = true;
               break;
 
          default: return XrdOfsFS->Emsg(epname, error, EINVAL,
                                  "decode checkpoint request for", oh->Name());
         };
 
// Complete as needed
//
   if (rc) return XrdOfsFS->Emsg(epname, error, rc, ckpName, oh->Name());
 
// Trace success and return
//
   FTRACE(chkpnt, ckpName);
   return SFS_OK;
}
  
/******************************************************************************/
/* Private:                    C r e a t e C K P                              */
/******************************************************************************/
 
int            XrdOfsFile::CreateCKP()
{
 
// Verify that a checkpoint does not exist
//
   if (myCKP) return XrdOfsFS->Emsg("CreateCKP", error, EEXIST,
                                    "create checkpoint for", oh->Name());
 
// Verify that this file is open r/w mode
//
   if (!(oh->isRW)) return XrdOfsFS->Emsg("CreateCKP", error, ENOTTY,
                           "create checkpoint for R/O", oh->Name());
 
// POSC and checkpoints are mutally exclusive
//
   if (oh->isRW == XrdOfsHandle::opPC)
      return XrdOfsFS->Emsg("CreateCKP", error, ENOTTY,
                            "create checkpoint for POSC file", oh->Name());
 
// Get a new checkpoint object
//
   if (XrdOfsFS->OssIsProxy)
      {char *resp;
       int rc = oh->Select().Fctl(XrdOssDF::Fctl_ckpObj, 0, 0, &resp);
       if (rc) return XrdOfsFS->Emsg("CreateCKP", error, rc,
                                     "create proxy checkpoint");
       myCKP = (XrdOucChkPnt *)resp;
      } else myCKP = new XrdOfsChkPnt(oh->Select(), oh->Name());
 
// All done
//
   return 0;
}
  
/******************************************************************************/
/*                                  f c t l                                   */
/******************************************************************************/
  
int            XrdOfsFile::fctl(const int               cmd,
                                const char             *args,
                                      XrdOucErrInfo    &out_error)
{
// See if we can do this
//
   if (cmd == SFS_FCTL_GETFD)
      {out_error.setErrCode(oh->Select().getFD());
       return SFS_OK;
      }
 
// We don't support this
//
   out_error.setErrInfo(ENOTSUP, "fctl operation not supported");
 
// Return
//
   return SFS_ERROR;
}
 
/******************************************************************************/
 
int XrdOfsFile::fctl(const int cmd, int alen, const char *args,
                     const XrdSecEntity *client)
{                             // 12345678901234
   static const char *fctlArg = "ofs.tpc cancel";
   static const int   fctlAsz = 15;
 
// See if the is a tpc cancellation (the only thing we support here)
//
   if (cmd != SFS_FCTL_SPEC1 || !args || alen < fctlAsz || strcmp(fctlArg,args))
      return XrdOfsFS->FSctl(*this, cmd, alen, args, client);
 
// Check if we have a tpc operation in progress
//
   if (!myTPC)
      {error.setErrInfo(ESRCH, "tpc operation not found");
       return SFS_ERROR;
      }
 
// Cancel the tpc
//
   myTPC->Del();
   myTPC = 0;
   return SFS_OK;
}
 
/******************************************************************************/
/*                                p g R e a d                                 */
/******************************************************************************/
  
XrdSfsXferSize XrdOfsFile::pgRead(XrdSfsFileOffset   offset,
                                  char              *buffer,
                                  XrdSfsXferSize     rdlen,
                                  uint32_t          *csvec,
                                  uint64_t           opts)
{
   EPNAME("pgRead");
   XrdSfsXferSize nbytes;
   uint64_t pgOpts;
 
// If the oss plugin does not support pgRead and we doing rawio then simulate
// the pgread. As this is relatively common we skip the vtable. This means
// this class cannot be a inherited to override the read() method.
//
   if (!XrdOfsFS->OssHasPGrw || dorawio)
      {if ((nbytes = XrdOfsFile::read(offset, buffer, rdlen)) > 0)
          XrdOucPgrwUtils::csCalc(buffer, offset, nbytes, csvec);
       return nbytes;
      }
 
// Perform required tracing
//
   FTRACE(read, rdlen <<"@" <<offset);
 
// Make sure the offset is not too large
//
#if _FILE_OFFSET_BITS!=64
   if (offset >  0x000000007fffffff)
      return  XrdOfsFS->Emsg(epname, error, EFBIG, "pgRead", oh->Name());
#endif
 
// Pass through any flags of interest
//
   if (opts & XrdSfsFile::Verify) pgOpts = XrdOssDF::Verify;
      else pgOpts = 0;
 
// Now read the actual number of bytes
//
   nbytes = (XrdSfsXferSize)(oh->Select().pgRead((void *)buffer,
                            (off_t)offset, (size_t)rdlen, csvec, pgOpts));
   if (nbytes < 0)
      return XrdOfsFS->Emsg(epname, error, (int)nbytes, "pgRead", oh->Name());
 
// Return number of bytes read
//
   return nbytes;
}
 
/******************************************************************************/
 
XrdSfsXferSize XrdOfsFile::pgRead(XrdSfsAio *aioparm, uint64_t opts)
{
   EPNAME("aiopgread");
   uint64_t pgOpts;
   int rc;
 
// If the oss plugin does not support pgRead or if we are doing rawio or the
// file is compressed then revert to using a standard async read. Note that
// the standard async read will generate checksums if a vector is present.
// Note: we set cksVec in the request to nil to indicate simulation!
//
   if (!XrdOfsFS->OssHasPGrw || dorawio || oh->isCompressed)
      {aioparm->cksVec = 0;
       return XrdOfsFile::read(aioparm);
      }
 
// Perform required tracing
//
   FTRACE(aio, aioparm->sfsAio.aio_nbytes <<"@" <<aioparm->sfsAio.aio_offset);
 
// Make sure the offset is not too large
//
#if _FILE_OFFSET_BITS!=64
   if (aiop->sfsAio.aio_offset >  0x000000007fffffff)
      return  XrdOfsFS->Emsg(epname, error, EFBIG, "pgRead", oh->Name());
#endif
 
// Pass through any flags of interest
//
   if (opts & XrdSfsFile::Verify) pgOpts = XrdOssDF::Verify;
      else pgOpts = 0;
 
// Issue the read. Only true errors are returned here.
//
   if ((rc = oh->Select().pgRead(aioparm, pgOpts)) < 0)
      return XrdOfsFS->Emsg(epname, error, rc, "pgRead", oh->Name());
 
// All done
//
   return SFS_OK;
}
 
/******************************************************************************/
/*                               p g W r i t e                                */
/******************************************************************************/
  
XrdSfsXferSize XrdOfsFile::pgWrite(XrdSfsFileOffset   offset,
                                   char              *buffer,
                                   XrdSfsXferSize     wrlen,
                                   uint32_t          *csvec,
                                   uint64_t           opts)
{
   EPNAME("pgWrite");
   XrdSfsXferSize nbytes;
   uint64_t pgOpts;
 
// If the oss plugin does not support pgWrite revert to using a standard write.
//
   if (!XrdOfsFS->OssHasPGrw)
      {if ((opts & XrdSfsFile::Verify)
       && !VerPgw(buffer, offset, wrlen, csvec, oh, error)) return SFS_ERROR;
       return XrdOfsFile::write(offset, buffer, wrlen);
      }
 
// Perform any required tracing
//
   FTRACE(write, wrlen <<"@" <<offset);
 
// Make sure the offset is not too large
//
#if _FILE_OFFSET_BITS!=64
   if (offset >  0x000000007fffffff)
      return  XrdOfsFS->Emsg(epname, error, EFBIG, "pgwrite", oh);
#endif
 
// Silly Castor stuff
//
   if (XrdOfsFS->evsObject && !(oh->isChanged)
   &&  XrdOfsFS->evsObject->Enabled(XrdOfsEvs::Fwrite)) GenFWEvent();
 
// Pass through any flags of interest
//
   if (opts & XrdSfsFile::Verify) pgOpts = XrdOssDF::Verify;
      else pgOpts = 0;
 
// Write the requested bytes
//
   oh->isPending = 1;
   nbytes = (XrdSfsXferSize)(oh->Select().pgWrite((void *)buffer,
                            (off_t)offset, (size_t)wrlen, csvec, pgOpts));
   if (nbytes < 0)
      return XrdOfsFS->Emsg(epname, error, (int)nbytes, "pgwrite", oh);
 
// Return number of bytes written
//
   return nbytes;
}
 
/******************************************************************************/
 
XrdSfsXferSize XrdOfsFile::pgWrite(XrdSfsAio *aioparm, uint64_t opts)
{
   EPNAME("aiopgWrite");
   uint64_t pgOpts;
   int rc;
 
// If the oss plugin does not support pgWrite revert to using a standard write.
//
   if (!XrdOfsFS->OssHasPGrw)
      {if ((opts & XrdSfsFile::Verify)
       && !VerPgw((char *)aioparm->sfsAio.aio_buf,
                          aioparm->sfsAio.aio_offset,
                          aioparm->sfsAio.aio_nbytes,
                          aioparm->cksVec, oh, error)) return SFS_ERROR;
       return XrdOfsFile::write(aioparm);
      }
 
// If this is a POSC file, we must convert the async call to a sync call as we
// must trap any errors that unpersist the file. We can't do that via aio i/f.
//
   if (oh->isRW == XrdOfsHandle::opPC)
      {aioparm->Result = XrdOfsFile::pgWrite(aioparm->sfsAio.aio_offset,
                                     (char *)aioparm->sfsAio.aio_buf,
                                             aioparm->sfsAio.aio_nbytes,
                                             aioparm->cksVec, opts);
       aioparm->doneWrite();
       return SFS_OK;
      }
 
// Perform any required tracing
//
   FTRACE(aio, aioparm->sfsAio.aio_nbytes <<"@" <<aioparm->sfsAio.aio_offset);
 
// Make sure the offset is not too large
//
#if _FILE_OFFSET_BITS!=64
   if (aiop->sfsAio.aio_offset >  0x000000007fffffff)
      return  XrdOfsFS->Emsg(epname, error, EFBIG, "pgwrite", oh->Name());
#endif
 
// Silly Castor stuff
//
   if (XrdOfsFS->evsObject && !(oh->isChanged)
   &&  XrdOfsFS->evsObject->Enabled(XrdOfsEvs::Fwrite)) GenFWEvent();
 
// Pass through any flags of interest
//
   if (opts & XrdSfsFile::Verify) pgOpts = XrdOssDF::Verify;
      else pgOpts = 0;
 
// Write the requested bytes
//
   oh->isPending = 1;
   if ((rc = oh->Select().pgWrite(aioparm, pgOpts)) < 0)
       return XrdOfsFS->Emsg(epname, error, rc, "pgwrite", oh->Name());
 
// All done
//
   return SFS_OK;
}
 
/******************************************************************************/
/*                                  r e a d                                   */
/******************************************************************************/
 
int            XrdOfsFile::read(XrdSfsFileOffset  offset,    // In
                                XrdSfsXferSize    blen)      // In
/*
  Function: Preread `blen' bytes at `offset'
 
  Input:    offset    - The absolute byte offset at which to start the read.
            blen      - The amount to preread.
 
  Output:   Returns SFS_OK upon success and SFS_ERROR o/w.
*/
{
   EPNAME("read");
   int retc;
 
// Perform required tracing
//
   FTRACE(read, "preread " <<blen <<"@" <<offset);
 
// Make sure the offset is not too large
//
#if _FILE_OFFSET_BITS!=64
   if (offset >  0x000000007fffffff)
      return  XrdOfsFS->Emsg(epname, error, EFBIG, "read", oh->Name());
#endif
 
// Now preread the actual number of bytes
//
   if ((retc = oh->Select().Read((off_t)offset, (size_t)blen)) < 0)
      return XrdOfsFS->Emsg(epname, error, (int)retc, "preread", oh->Name());
 
// Return number of bytes read
//
   return retc;
}
  
/******************************************************************************/
/*                                  r e a d                                   */
/******************************************************************************/
 
XrdSfsXferSize XrdOfsFile::read(XrdSfsFileOffset  offset,    // In
                                char             *buff,      // Out
                                XrdSfsXferSize    blen)      // In
/*
  Function: Read `blen' bytes at `offset' into 'buff' and return the actual
            number of bytes read.
 
  Input:    offset    - The absolute byte offset at which to start the read.
            buff      - Address of the buffer in which to place the data.
            blen      - The size of the buffer. This is the maximum number
                        of bytes that will be read from 'fd'.
 
  Output:   Returns the number of bytes read upon success and SFS_ERROR o/w.
*/
{
   EPNAME("read");
   XrdSfsXferSize nbytes;
 
// Perform required tracing
//
   FTRACE(read, blen <<"@" <<offset);
 
// Make sure the offset is not too large
//
#if _FILE_OFFSET_BITS!=64
   if (offset >  0x000000007fffffff)
      return  XrdOfsFS->Emsg(epname, error, EFBIG, "read", oh->Name());
#endif
 
// Now read the actual number of bytes
//
   nbytes = (dorawio ?
            (XrdSfsXferSize)(oh->Select().ReadRaw((void *)buff,
                            (off_t)offset, (size_t)blen))
          : (XrdSfsXferSize)(oh->Select().Read((void *)buff,
                            (off_t)offset, (size_t)blen)));
   if (nbytes < 0)
      return XrdOfsFS->Emsg(epname, error, (int)nbytes, "read", oh->Name());
 
// Return number of bytes read
//
   return nbytes;
}
 
/******************************************************************************/
/*                                  r e a d v                                 */
/******************************************************************************/
 
XrdSfsXferSize XrdOfsFile::readv(XrdOucIOVec     *readV,     // In
                                 int              readCount) // In
/*
  Function: Perform all the reads specified in the readV vector.
 
  Input:    readV     - A description of the reads to perform; includes the
                        absolute offset, the size of the read, and the buffer
                        to place the data into.
            readCount - The size of the readV vector.
 
  Output:   Returns the number of bytes read upon success and SFS_ERROR o/w.
            If the number of bytes read is less than requested, it is considered
            an error.
*/
{
   EPNAME("readv");
 
   XrdSfsXferSize nbytes = oh->Select().ReadV(readV, readCount);
   if (nbytes < 0)
       return XrdOfsFS->Emsg(epname, error, (int)nbytes, "readv", oh->Name());
 
   return nbytes;
 
}
  
/******************************************************************************/
/*                              r e a d   A I O                               */
/******************************************************************************/
  
/*
  Function: Read `blen' bytes at `offset' into 'buff' and return the actual
            number of bytes read using asynchronous I/O, if possible.
 
  Output:   Returns the 0 if successfullt queued, otherwise returns an error.
            The underlying implementation will convert the request to
            synchronous I/O is async mode is not possible.
*/
 
int XrdOfsFile::read(XrdSfsAio *aiop)
{
   EPNAME("aioread");
   int rc;
 
// Async mode for compressed files is not supported.
//
   if (oh->isCompressed)
      {aiop->Result = this->read((XrdSfsFileOffset)aiop->sfsAio.aio_offset,
                                           (char *)aiop->sfsAio.aio_buf,
                                   (XrdSfsXferSize)aiop->sfsAio.aio_nbytes);
       aiop->doneRead();
       return 0;
      }
 
// Perform required tracing
//
   FTRACE(aio, aiop->sfsAio.aio_nbytes <<"@" <<aiop->sfsAio.aio_offset);
 
// Make sure the offset is not too large
//
#if _FILE_OFFSET_BITS!=64
   if (aiop->sfsAio.aio_offset >  0x000000007fffffff)
      return  XrdOfsFS->Emsg(epname, error, EFBIG, "read", oh->Name());
#endif
 
// Issue the read. Only true errors are returned here.
//
   if ((rc = oh->Select().Read(aiop)) < 0)
      return XrdOfsFS->Emsg(epname, error, rc, "read", oh->Name());
 
// All done
//
   return SFS_OK;
}
 
/******************************************************************************/
/*                                 w r i t e                                  */
/******************************************************************************/
 
XrdSfsXferSize XrdOfsFile::write(XrdSfsFileOffset  offset,    // In
                                 const char       *buff,      // Out
                                 XrdSfsXferSize    blen)      // In
/*
  Function: Write `blen' bytes at `offset' from 'buff' and return the actual
            number of bytes written.
 
  Input:    offset    - The absolute byte offset at which to start the write.
            buff      - Address of the buffer from which to get the data.
            blen      - The size of the buffer. This is the maximum number
                        of bytes that will be written to 'fd'.
 
  Output:   Returns the number of bytes written upon success and SFS_ERROR o/w.
 
  Notes:    An error return may be delayed until the next write(), close(), or
            sync() call.
*/
{
   EPNAME("write");
   XrdSfsXferSize nbytes;
 
// Perform any required tracing
//
   FTRACE(write, blen <<"@" <<offset);
 
// Make sure the offset is not too large
//
#if _FILE_OFFSET_BITS!=64
   if (offset >  0x000000007fffffff)
      return  XrdOfsFS->Emsg(epname, error, EFBIG, "write", oh);
#endif
 
// Silly Castor stuff
//
   if (XrdOfsFS->evsObject && !(oh->isChanged)
   &&  XrdOfsFS->evsObject->Enabled(XrdOfsEvs::Fwrite)) GenFWEvent();
 
// Write the requested bytes
//
   oh->isPending = 1;
   nbytes = (XrdSfsXferSize)(oh->Select().Write((const void *)buff,
                            (off_t)offset, (size_t)blen));
   if (nbytes < 0)
      return XrdOfsFS->Emsg(epname, error, (int)nbytes, "write", oh);
 
// Return number of bytes written
//
   return nbytes;
}
 
/******************************************************************************/
/*                             w r i t e   A I O                              */
/******************************************************************************/
  
// For now, this reverts to synchronous I/O
//
int XrdOfsFile::write(XrdSfsAio *aiop)
{
   EPNAME("aiowrite");
   int rc;
 
// Perform any required tracing
//
   FTRACE(aio, aiop->sfsAio.aio_nbytes <<"@" <<aiop->sfsAio.aio_offset);
 
// If this is a POSC file, we must convert the async call to a sync call as we
// must trap any errors that unpersist the file. We can't do that via aio i/f.
//
   if (oh->isRW == XrdOfsHandle::opPC)
      {aiop->Result = this->write(aiop->sfsAio.aio_offset,
                                  (const char *)aiop->sfsAio.aio_buf,
                                  aiop->sfsAio.aio_nbytes);
       aiop->doneWrite();
       return 0;
      }
 
// Make sure the offset is not too large
//
#if _FILE_OFFSET_BITS!=64
   if (aiop->sfsAio.aio_offset >  0x000000007fffffff)
      return  XrdOfsFS->Emsg(epname, error, EFBIG, "write", oh->Name());
#endif
 
// Silly Castor stuff
//
   if (XrdOfsFS->evsObject && !(oh->isChanged)
   &&  XrdOfsFS->evsObject->Enabled(XrdOfsEvs::Fwrite)) GenFWEvent();
 
// Write the requested bytes
//
   oh->isPending = 1;
   if ((rc = oh->Select().Write(aiop)) < 0)
       return XrdOfsFS->Emsg(epname, error, rc, "write", oh->Name());
 
// All done
//
   return SFS_OK;
}
 
/******************************************************************************/
/*                               g e t M m a p                                */
/******************************************************************************/
 
int XrdOfsFile::getMmap(void **Addr, off_t &Size)         // Out
/*
  Function: Return memory mapping for file, if any.
 
  Output:   Addr        - Address of memory location
            Size        - Size of the file or zero if not memory mapped.
            Returns SFS_OK upon success and SFS_ERROR upon failure.
*/
{
 
// Perform the function
//
   Size = oh->Select().getMmap(Addr);
 
   return SFS_OK;
}
  
/******************************************************************************/
/*                                  s t a t                                   */
/******************************************************************************/
 
int XrdOfsFile::stat(struct stat     *buf)         // Out
/*
  Function: Return file status information
 
  Input:    buf         - The stat structiure to hold the results
 
  Output:   Returns SFS_OK upon success and SFS_ERROR upon failure.
*/
{
   EPNAME("fstat");
   int retc;
 
// Perform any required tracing
//
   FTRACE(stat, "");
 
// Perform the function
//
   if ((retc = oh->Select().Fstat(buf)) < 0)
      return XrdOfsFS->Emsg(epname,error,retc,"get state for",oh->Name());
 
   return SFS_OK;
}
 
/******************************************************************************/
/*                                  s y n c                                   */
/******************************************************************************/
 
int XrdOfsFile::sync()  // In
/*
  Function: Commit all unwritten bytes to physical media.
 
  Input:    n/a
 
  Output:   Returns SFS_OK upon success and SFS_ERROR upon failure.
*/
{
   EPNAME("sync");
   int retc;
 
// Perform any required tracing
//
   FTRACE(sync, "");
 
// If we have a tpc object hanging about, we need to dispatch that first
//
   if (myTPC && (retc = myTPC->Sync(&error))) return retc;
 
// We can test the pendio flag w/o a lock because the person doing this
// sync must have done the previous write. Causality is the synchronizer.
//
   if (!(oh->isPending)) return SFS_OK;
 
// We can also skip the sync if the file is closed. However, we need a file
// object lock in order to test the flag. We can also reset the PENDIO flag.
//
   oh->Lock();
   oh->isPending = 0;
   oh->UnLock();
 
// Perform the function
//
   if ((retc = oh->Select().Fsync()))
      {oh->isPending = 1;
       return XrdOfsFS->Emsg(epname, error, retc, "synchronize", oh);
      }
 
// Indicate all went well
//
   return SFS_OK;
}
 
/******************************************************************************/
/*                              s y n c   A I O                               */
/******************************************************************************/
  
// For now, reverts to synchronous case. This must also be the case for POSC!
//
int XrdOfsFile::sync(XrdSfsAio *aiop)
{
   aiop->Result = this->sync();
   aiop->doneWrite();
   return 0;
}
 
/******************************************************************************/
/*                              t r u n c a t e                               */
/******************************************************************************/
 
int XrdOfsFile::truncate(XrdSfsFileOffset  flen)  // In
/*
  Function: Set the length of the file object to 'flen' bytes.
 
  Input:    flen      - The new size of the file.
 
  Output:   Returns SFS_OK upon success and SFS_ERROR upon failure.
 
  Notes:    If 'flen' is smaller than the current size of the file, the file
            is made smaller and the data past 'flen' is discarded. If 'flen'
            is larger than the current size of the file, a hole is created
            (i.e., the file is logically extended by filling the extra bytes 
            with zeroes).
*/
{
   EPNAME("trunc");
   int retc;
 
// Lock the file handle and perform any tracing
//
   FTRACE(truncate, "len=" <<flen);
 
// Make sure the offset is not too large
//
   if (sizeof(off_t) < sizeof(flen) && flen >  0x000000007fffffff)
      return  XrdOfsFS->Emsg(epname, error, EFBIG, "truncate", oh);
 
// Silly Castor stuff
//
   if (XrdOfsFS->evsObject && !(oh->isChanged)
   &&  XrdOfsFS->evsObject->Enabled(XrdOfsEvs::Fwrite)) GenFWEvent();
 
// Perform the function
//
   oh->isPending = 1;
   if ((retc = oh->Select().Ftruncate(flen)))
      return XrdOfsFS->Emsg(epname, error, retc, "truncate", oh);
 
// Indicate Success
//
   return SFS_OK;
}
 
/******************************************************************************/
/*                             g e t C X i n f o                              */
/******************************************************************************/
  
int XrdOfsFile::getCXinfo(char cxtype[4], int &cxrsz)
/*
  Function: Set the length of the file object to 'flen' bytes.
 
  Input:    n/a
 
  Output:   cxtype - Compression algorithm code
            cxrsz  - Compression region size
 
            Returns SFS_OK upon success and SFS_ERROR upon failure.
*/
{
 
// Copy out the info
//
   cxrsz = oh->Select().isCompressed(cxtype);
   return SFS_OK;
}
 
/******************************************************************************/
/*                  P r i v a t e   F i l e   M e t h o d s                   */
/******************************************************************************/
/******************************************************************************/
/* protected                  G e n F W E v e n t                             */
/******************************************************************************/
  
void XrdOfsFile::GenFWEvent()
{
   int first_write;
 
// This silly code is to generate a 1st write event which slows things down
// but is needed by the one and only Castor. What a big sigh!
//
   oh->Lock();
   if ((first_write = !(oh->isChanged))) oh->isChanged = 1;
   oh->UnLock();
   if (first_write)
      {XrdOfsEvsInfo evInfo(tident, oh->Name());
       XrdOfsFS->evsObject->Notify(XrdOfsEvs::Fwrite, evInfo);
      }
}
 
/******************************************************************************/
/*                                                                            */
/*         F i l e   S y s t e m   O b j e c t   I n t e r f a c e s          */
/*                                                                            */
/******************************************************************************/
/******************************************************************************/
/*                                c h k s u m                                 */
/******************************************************************************/
 
int XrdOfs::chksum(      csFunc            Func,   // In
                         const char       *csName, // In
                         const char       *Path,   // In
                         XrdOucErrInfo    &einfo,  // Out
                   const XrdSecEntity     *client, // In
                   const char             *opaque) // In
/*
  Function: Compute and return file checksum.
 
  Input:    Func      - Function to be performed:
                        csCalc   - Return precomputed or computed checksum.
                        csGet    - Return precomputed checksum.
                        csSize   - Verify csName and get its size.
            Path      - Pathname of file for csCalc and csSize.
            einfo     - Error information object to hold error details.
            client    - Authentication credentials, if any.
            opaque    - Opaque information to be used as seen fit.
 
  Output:   Returns SFS_OK upon success and SFS_ERROR upon failure.
*/
{
   EPNAME("chksum");
   XrdOucEnv  cksEnv(opaque,0,client);
   XrdCksData cksData;
   const char *tident = einfo.getErrUser();
   char buff[MAXPATHLEN+8];
   int rc;
 
// Check if we support checksumming
//
   if (!Cks)
      {einfo.setErrInfo(ENOTSUP, "Checksums are not supported.");
       return SFS_ERROR;
      }
 
// A csSize request is issued usually once to verify everything is working. We
// take this opportunity to also verify the checksum name.
//
   rc = cksData.Set(csName);
   if (!rc || Func == XrdSfsFileSystem::csSize)
      {if (rc && (rc = Cks->Size(csName)))
          {einfo.setErrCode(rc); return SFS_OK;}
       strcpy(buff, csName); strcat(buff, " checksum not supported.");
       einfo.setErrInfo(ENOTSUP, buff);
       return SFS_ERROR;
      }
 
// Everything else requires a path
//
   if (!Path)
      {strcpy(buff, csName);
       strcat(buff, " checksum path not specified.");
       einfo.setErrInfo(EINVAL, buff);
       return SFS_ERROR;
      }
 
// Apply security, as needed
//
   XTRACE(stat, Path, csName);
   AUTHORIZE(client,&cksEnv,AOP_Stat,"checksum",Path,einfo);
 
// If we are a menager then we need to redirect the client to where the file is
//
   if (CksRdr && Finder && Finder->isRemote() 
   &&  (rc = Finder->Locate(einfo, Path, SFS_O_RDONLY, &cksEnv)))
      return fsError(einfo, rc);
 
// At this point we need to convert the lfn to a pfn
//
   if (CksPfn && !(Path = XrdOfsOss->Lfn2Pfn(Path, buff, MAXPATHLEN, rc)))
      return Emsg(epname, einfo, rc, "checksum", Path);
 
// Originally we only passed he env pointer for proxy servers. Due to popular
// demand, we always pass the env as it points to the SecEntity object unless
// we don't have it then we pass the caller's environment.
//
   if (Func == XrdSfsFileSystem::csGet || Func == XrdSfsFileSystem::csCalc)
      {if (client) cksData.envP = &cksEnv;
          else cksData.envP = (einfo.getEnv() ? einfo.getEnv() : &cksEnv);
      }
 
// Now determine what to do
//
        if (Func == XrdSfsFileSystem::csCalc) rc = Cks->Calc(Path, cksData);
   else if (Func == XrdSfsFileSystem::csGet)  rc = Cks->Get( Path, cksData);
   else {einfo.setErrInfo(EINVAL, "Invalid checksum function.");
         return SFS_ERROR;
        }
 
// See if all went well
//
#ifdef ENOATTR
   if (rc >= 0 || rc == -ENOATTR || rc == -ESTALE || rc == -ESRCH)
#else
   if (rc >= 0 || rc == -ENODATA || rc == -ESTALE || rc == -ESRCH)
#endif
      {if (rc >= 0) {cksData.Get(buff, MAXPATHLEN); rc = 0;}
          else {*buff = 0; rc = -rc;}
       einfo.setErrInfo(rc, buff);
       return SFS_OK;
      }
 
// We failed
//
   return Emsg(epname, einfo, rc, "checksum", Path);
}
  
/******************************************************************************/
/*                                 c h m o d                                  */
/******************************************************************************/
 
int XrdOfs::chmod(const char             *path,    // In
                        XrdSfsMode        Mode,    // In
                        XrdOucErrInfo    &einfo,   // Out
                  const XrdSecEntity     *client,  // In
                  const char             *info)    // In
/*
  Function: Change the mode on a file or directory.
 
  Input:    path      - Is the fully qualified name of the file to be removed.
            einfo     - Error information object to hold error details.
            client    - Authentication credentials, if any.
            info      - Opaque information to be used as seen fit.
 
  Output:   Returns SFS_OK upon success and SFS_ERROR upon failure.
*/
{
   EPNAME("chmod");
   static const int locFlags = SFS_O_RDWR|SFS_O_META;
   struct stat Stat;
   mode_t acc_mode = Mode & S_IAMB;
   const char *tident = einfo.getErrUser();
   XrdOucEnv chmod_Env(info,0,client);
   int retc;
   XTRACE(chmod, path, "");
 
// Apply security, as needed
//
   AUTHORIZE(client,&chmod_Env,AOP_Chmod,"chmod",path,einfo);
 
// Find out where we should chmod this file
//
   if (Finder && Finder->isRemote())
      {if (fwdCHMOD.Cmd)
          {char buff[8];
           sprintf(buff, "%o", static_cast<int>(acc_mode));
           if (Forward(retc,einfo,fwdCHMOD,path,buff,&chmod_Env)) return retc;
          }
          else if ((retc = Finder->Locate(einfo, path, locFlags, &chmod_Env)))
                  return fsError(einfo, retc);
      }
 
// We need to adjust the mode based on whether this is a file or directory.
//
   if ((retc = XrdOfsOss->Stat(path, &Stat, 0, &chmod_Env)))
      return XrdOfsFS->Emsg(epname, einfo, retc, "change", path);
   if (S_ISDIR(Stat.st_mode)) acc_mode = (acc_mode | dMask[0]) & dMask[1];
      else acc_mode = (acc_mode | fMask[0]) & fMask[1];
 
// Check if we should generate an event
//
   if (evsObject && evsObject->Enabled(XrdOfsEvs::Chmod))
      {XrdOfsEvsInfo evInfo(tident, path, info, &chmod_Env, acc_mode);
       evsObject->Notify(XrdOfsEvs::Chmod, evInfo);
      }
 
// Now try to find the file or directory
//
   if (!(retc = XrdOfsOss->Chmod(path, acc_mode, &chmod_Env))) return SFS_OK;
 
// An error occurred, return the error info
//
   return XrdOfsFS->Emsg(epname, einfo, retc, "change", path);
}
 
/******************************************************************************/
/*                               C o n n e c t                                */
/******************************************************************************/
 
void XrdOfs::Connect(const XrdSecEntity *client)
{
   XrdOucEnv myEnv(0, 0, client);
 
// Pass this call along
//
   XrdOfsOss->Connect(myEnv);
}
 
/******************************************************************************/
/*                                  D i s c                                   */
/******************************************************************************/
  
void XrdOfs::Disc(const XrdSecEntity *client)
{
   XrdOucEnv myEnv(0, 0, client);
 
// Pass this call along
//
   XrdOfsOss->Disc(myEnv);
}
 
/******************************************************************************/
/*                                e x i s t s                                 */
/******************************************************************************/
 
int XrdOfs::exists(const char                *path,        // In
                         XrdSfsFileExistence &file_exists, // Out
                         XrdOucErrInfo       &einfo,       // Out
                   const XrdSecEntity        *client,      // In
                   const char                *info)        // In
/*
  Function: Determine if file 'path' actually exists.
 
  Input:    path        - Is the fully qualified name of the file to be tested.
            file_exists - Is the address of the variable to hold the status of
                          'path' when success is returned. The values may be:
                          XrdSfsFileExistsIsDirectory - file not found but path is valid.
                          XrdSfsFileExistsIsFile      - file found.
                          XrdSfsFileExistsIsNo        - neither file nor directory.
            einfo       - Error information object holding the details.
            client      - Authentication credentials, if any.
            info        - Opaque information to be used as seen fit.
 
  Output:   Returns SFS_OK upon success and SFS_ERROR upon failure.
 
  Notes:    When failure occurs, 'file_exists' is not modified.
*/
{
   EPNAME("exists");
   struct stat fstat;
   int retc;
   const char *tident = einfo.getErrUser();
   XrdOucEnv stat_Env(info,0,client);
   XTRACE(exists, path, "");
 
// Apply security, as needed
//
   AUTHORIZE(client,&stat_Env,AOP_Stat,"locate",path,einfo);
 
// Find out where we should stat this file
//
   if (Finder && Finder->isRemote() 
   &&  (retc = Finder->Locate(einfo, path, SFS_O_RDONLY, &stat_Env)))
      return fsError(einfo, retc);
 
// Now try to find the file or directory
//
   retc = XrdOfsOss->Stat(path, &fstat, 0, &stat_Env);
   if (!retc)
      {     if (S_ISDIR(fstat.st_mode)) file_exists=XrdSfsFileExistIsDirectory;
       else if (S_ISREG(fstat.st_mode)) file_exists=XrdSfsFileExistIsFile;
       else                             file_exists=XrdSfsFileExistNo;
       return SFS_OK;
      }
   if (retc == -ENOENT)
      {file_exists=XrdSfsFileExistNo;
       return SFS_OK;
      }
 
// An error occurred, return the error info
//
   return XrdOfsFS->Emsg(epname, einfo, retc, "locate", path);
}
 
/******************************************************************************/
/*                              g e t S t a t s                               */
/******************************************************************************/
 
int XrdOfs::getStats(char *buff, int blen)
{
   int n;
 
// See if the size just wanted
//
  if (!buff) return OfsStats.Report(0,0) + XrdOfsOss->Stats(0,0);
 
// Report ofs info followed by the oss info
//
   n = OfsStats.Report(buff, blen);
   buff += n; blen -= n;
   n += XrdOfsOss->Stats(buff, blen);
 
// All done
//
   return n;
}
  
/******************************************************************************/
/*                                 m k d i r                                  */
/******************************************************************************/
 
int XrdOfs::mkdir(const char             *path,    // In
                        XrdSfsMode        Mode,    // In
                        XrdOucErrInfo    &einfo,   // Out
                  const XrdSecEntity     *client,  // In
                  const char             *info)    // In
/*
  Function: Create a directory entry.
 
  Input:    path      - Is the fully qualified name of the file to be removed.
            Mode      - Is the POSIX mode value the directory is to have.
                        Additionally, Mode may contain SFS_O_MKPTH if the
                        full dircectory path should be created.
            einfo     - Error information object to hold error details.
            client    - Authentication credentials, if any.
            info      - Opaque information to be used as seen fit.
 
  Output:   Returns SFS_OK upon success and SFS_ERROR upon failure.
*/
{
   EPNAME("mkdir");
   static const int LocOpts = SFS_O_RDWR | SFS_O_CREAT | SFS_O_META;
   mode_t acc_mode = (Mode | dMask[0]) & dMask[1];
   int retc, mkpath = Mode & SFS_O_MKPTH;
   const char *tident = einfo.getErrUser();
   XrdOucEnv mkdir_Env(info,0,client);
   XTRACE(mkdir, path, "");
 
// Apply security, as needed
//
   AUTHORIZE(client,&mkdir_Env,AOP_Mkdir,"mkdir",path,einfo);
 
// Find out where we should remove this file
//
   if (Finder && Finder->isRemote())
      {if (fwdMKDIR.Cmd)
          {char buff[8];
           sprintf(buff, "%o", static_cast<int>(acc_mode));
           if (Forward(retc, einfo, (mkpath ? fwdMKPATH:fwdMKDIR),
                       path, buff, &mkdir_Env)) return retc;
          }
          else if ((retc = Finder->Locate(einfo,path,LocOpts,&mkdir_Env)))
                  return fsError(einfo, retc);
      }
 
// Perform the actual operation
//
    if ((retc = XrdOfsOss->Mkdir(path, acc_mode, mkpath, &mkdir_Env)))
       return XrdOfsFS->Emsg(epname, einfo, retc, "mkdir", path);
 
// Check if we should generate an event
//
   if (evsObject && evsObject->Enabled(XrdOfsEvs::Mkdir))
      {XrdOfsEvsInfo evInfo(tident, path, info, &mkdir_Env, acc_mode);
       evsObject->Notify(XrdOfsEvs::Mkdir, evInfo);
      }
 
// If we have a redirector, tell it that we now have this path
//
   if (Balancer)
      {if (!mkpath) Balancer->Added(path);
          else     {char *slash, *myPath = strdup(path);
                    do {Balancer->Added(myPath);
                        if ((slash = rindex(myPath, '/'))) *slash = 0;
                       } while(slash && slash != myPath);
                    free(myPath);
                   }
      }
 
    return SFS_OK;
}
  
/******************************************************************************/
/*                               p r e p a r e                                */
/******************************************************************************/
 
int XrdOfs::prepare(      XrdSfsPrep       &pargs,      // In
                          XrdOucErrInfo    &out_error,  // Out
                    const XrdSecEntity     *client)     // In
{
   EPNAME("prepare");
   XrdOucEnv prep_Env(0,0,client);
   XrdOucTList *tp = pargs.paths;
   int retc;
 
// Run through the paths to make sure client can read each one unless we aren't
// supposed to apply authorization.
//
   if (prepAuth)
      while(tp)
           {AUTHORIZE(client,0,AOP_Read,"prepare",tp->text,out_error);
            tp = tp->next;
           }
 
// If there is a prepare plugin, invoke it and return the result.
//
   if (prepHandler)
      {if (pargs.opts & Prep_QUERY)
          return prepHandler->query(pargs, out_error, client);
       if (pargs.opts & Prep_CANCEL)
          return prepHandler->cancel(pargs, out_error, client);
 
       return prepHandler->begin(pargs, out_error, client);
      }
 
// If we have a finder object, use it to prepare the paths. Otherwise,
// ignore this prepare request (we may change this in the future).
//
   if (XrdOfsFS->Finder
   && (retc = XrdOfsFS->Finder->Prepare(out_error, pargs, &prep_Env)))
      return fsError(out_error, retc);
   return 0;
}
  
/******************************************************************************/
/*                                r e m o v e                                 */
/******************************************************************************/
 
int XrdOfs::remove(const char              type,    // In
                   const char             *path,    // In
                         XrdOucErrInfo    &einfo,   // Out
                   const XrdSecEntity     *client,  // In
                   const char             *info)    // In
/*
  Function: Delete a file from the namespace and release it's data storage.
 
  Input:    type      - 'f' for file and 'd' for directory.
            path      - Is the fully qualified name of the file to be removed.
            einfo     - Error information object to hold error details.
            client    - Authentication credentials, if any.
            info      - Opaque information to be used as seen fit.
 
  Output:   Returns SFS_OK upon success and SFS_ERROR upon failure.
*/
{
   EPNAME("remove");
   static const int LocOpts = SFS_O_WRONLY|SFS_O_META;
   int retc, Opt;
   const char *tident = einfo.getErrUser();
   XrdOucEnv rem_Env(info,0,client);
   XTRACE(remove, path, type);
 
// Apply security, as needed
//
   AUTHORIZE(client,&rem_Env,AOP_Delete,"remove",path,einfo);
 
// Find out where we should remove this file
//
   if (Finder && Finder->isRemote())
      {struct fwdOpt *fSpec = (type == 'd' ? &fwdRMDIR : &fwdRM);
       if (fSpec->Cmd)
          {if (Forward(retc, einfo, *fSpec, path, 0, &rem_Env)) return retc;}
          else if ((retc = Finder->Locate(einfo, path, LocOpts, &rem_Env)))
                  return fsError(einfo, retc);
      }
 
// Check if we should generate an event
//
   if (evsObject)
      {XrdOfsEvs::Event theEvent=(type == 'd' ? XrdOfsEvs::Rmdir:XrdOfsEvs::Rm);
       if (evsObject->Enabled(theEvent))
          {XrdOfsEvsInfo evInfo(tident, path, info, &rem_Env);
           evsObject->Notify(theEvent, evInfo);
          }
      }
 
// Check if this is an online deletion only
//
   Opt = (rem_Env.Get("ofs.lcl") ? XRDOSS_Online : 0);
 
// Perform the actual deletion
//
    retc = (type=='d' ? XrdOfsOss->Remdir(path, 0,   &rem_Env)
                      : XrdOfsOss->Unlink(path, Opt, &rem_Env));
    if (retc) return XrdOfsFS->Emsg(epname, einfo, retc, "remove", path);
    if (type == 'f') XrdOfsHandle::Hide(path);
    if (Balancer) Balancer->Removed(path);
    return SFS_OK;
}
 
/******************************************************************************/
/*                                r e n a m e                                 */
/******************************************************************************/
 
int XrdOfs::rename(const char             *old_name,  // In
                   const char             *new_name,  // In
                         XrdOucErrInfo    &einfo,     //Out
                   const XrdSecEntity     *client,    // In
                   const char             *infoO,     // In
                   const char             *infoN)     // In
/*
  Function: Renames a file with name 'old_name' to 'new_name'.
 
  Input:    old_name  - Is the fully qualified name of the file to be renamed.
            new_name  - Is the fully qualified name that the file is to have.
            einfo     - Error information structure, if an error occurs.
            client    - Authentication credentials, if any.
            infoO     - old_name opaque information to be used as seen fit.
            infoN     - new_name opaque information to be used as seen fit.
 
  Output:   Returns SFS_OK upon success and SFS_ERROR upon failure.
*/
{
   EPNAME("rename");
   static const int LocOpts = SFS_O_RDWR|SFS_O_META;
   int retc;
   const char *tident = einfo.getErrUser();
   XrdOucEnv old_Env(infoO,0,client);
   XrdOucEnv new_Env(infoN,0,client);
   XTRACE(rename, new_name, "old fn=" <<old_name <<" new ");
 
// Apply security, as needed
//
   AUTHORIZE(client, &old_Env, AOP_Rename, "renaming", old_name, einfo);
 
// The above authorization may mutate the XrdSecEntity by putting a mapped name
// into the extended attributes.  This mapped name will affect the subsequent
// authorization check below, giving the client access that may not be permitted.
// Hence, we delete this attribute to reset the object back to "pristine" state.
// If there was a way to make a copy of the XrdSecEntity, we could avoid this
// hack-y reach inside the extended attributes.
   if (client) client->eaAPI->Add("request.name", "", true);
 
// If we do not have full-blown insert authorization, we'll need to test for
// destination existence
   bool cannot_overwrite = false;
   if (client && XrdOfsFS->Authorization &&
      !XrdOfsFS->Authorization->Access(client, new_name, AOP_Insert, &new_Env))
      {cannot_overwrite = true;
       AUTHORIZE(client, &new_Env, AOP_Excl_Insert,
             "rename to existing file (overwrite disallowed)", new_name, einfo);
      }
 
// Find out where we should rename this file
//
   if (Finder && Finder->isRemote())
      {if (fwdMV.Cmd)
          {if (Forward(retc,einfo,fwdMV,old_name,new_name,&old_Env,&new_Env))
              return retc;
          }
          else if ((retc = Finder->Locate(einfo, old_name, LocOpts, &old_Env)))
                  return fsError(einfo, retc);
      }
 
// Check if we should generate an event
//
   if (evsObject && evsObject->Enabled(XrdOfsEvs::Mv))
      {XrdOfsEvsInfo evInfo(tident, old_name, infoO, &old_Env, 0, 0,
                                    new_name, infoN, &new_Env);
       evsObject->Notify(XrdOfsEvs::Mv, evInfo);
      }
 
// If we cannot overwrite, we must test for existence first.  This will test whether
// we will destroy data in the rename (without actually destroying data).
// Note there's an obvious race condition here; it was seen as the lesser-of-evils
// compared to creating an exclusive file and potentially leaking it in the event
// of a crash.
//
   if (cannot_overwrite)
      {XrdSfsFileExistence exists_flag = XrdSfsFileExistNo;
       if (SFS_OK != exists(new_name, exists_flag, einfo, client, infoN))
          {// File existence check itself failed; we can't prove that data won't
           // be overwritten so we return an error.
           return fsError(einfo, -einfo.getErrInfo());
          }
       if (exists_flag != XrdSfsFileExistNo)
          {// EPERM mimics the error code set by Linux when you invoke rename()
           // but cannot overwrite the destination file.
           einfo.setErrInfo(EPERM, "Overwrite of existing data not permitted");
           return fsError(einfo, -EPERM);
          }
      }
 
// Perform actual rename operation
//
   if ((retc = XrdOfsOss->Rename(old_name, new_name, &old_Env, &new_Env)))
      {return XrdOfsFS->Emsg(epname, einfo, retc, "rename", old_name);
      }
   XrdOfsHandle::Hide(old_name);
   if (Balancer) {Balancer->Removed(old_name);
                  Balancer->Added(new_name);
                 }
   return SFS_OK;
}
 
/******************************************************************************/
/*                                  s t a t                                   */
/******************************************************************************/
 
int XrdOfs::stat(const char             *path,        // In
                       struct stat      *buf,         // Out
                       XrdOucErrInfo    &einfo,       // Out
                 const XrdSecEntity     *client,      // In
                 const char             *info)        // In
/*
  Function: Return file status information
 
  Input:    path      - The path for which status is wanted
            buf       - The stat structure to hold the results
            einfo     - Error information structure, if an error occurs.
            client    - Authentication credentials, if any.
            info      - opaque information to be used as seen fit.
 
  Output:   Returns SFS_OK upon success and SFS_ERROR upon failure.
*/
{
   EPNAME("stat");
   int retc;
   const char *tident = einfo.getErrUser();
   XrdOucEnv stat_Env(info,0,client);
   XTRACE(stat, path, "");
 
// Apply security, as needed
//
   AUTHORIZE(client,&stat_Env,AOP_Stat,"locate",path,einfo);
 
// Find out where we should stat this file
//
   if (Finder && Finder->isRemote()
   &&  (retc = Finder->Locate(einfo, path, SFS_O_RDONLY|SFS_O_STAT, &stat_Env)))
      return fsError(einfo, retc);
 
// Now try to find the file or directory
//
   if ((retc = XrdOfsOss->Stat(path, buf, 0, &stat_Env)))
      return XrdOfsFS->Emsg(epname, einfo, retc, "locate", path);
   return SFS_OK;
}
 
/******************************************************************************/
 
int XrdOfs::stat(const char             *path,        // In
                       mode_t           &mode,        // Out
                       XrdOucErrInfo    &einfo,       // Out
                 const XrdSecEntity     *client,      // In
                 const char             *info)        // In
/*
  Function: Return file status information (resident files only)
 
  Input:    path      - The path for which status is wanted
            mode      - The stat mode entry (faked -- do not trust it)
            einfo     - Error information structure, if an error occurs.
            client    - Authentication credentials, if any.
            info      - opaque information to be used as seen fit.
 
  Output:   Always returns SFS_ERROR if a delay needs to be imposed. Otherwise,
            SFS_OK is returned and mode is appropriately, if inaccurately, set.
            If file residency cannot be determined, mode is set to -1.
*/
{
   EPNAME("stat");
   struct stat buf;
   int retc;
   const char *tident = einfo.getErrUser();
   XrdOucEnv stat_Env(info,0,client);
   XTRACE(stat, path, "");
 
// Apply security, as needed
//
   AUTHORIZE(client,&stat_Env,AOP_Stat,"locate",path,einfo);
   mode = (mode_t)-1;
 
// Find out where we should stat this file
//
   if (Finder && Finder->isRemote()
   &&  (retc = Finder->Locate(einfo,path,SFS_O_NOWAIT|SFS_O_RDONLY|SFS_O_STAT,
                              &stat_Env)))
      return fsError(einfo, retc);
 
// Now try to find the file or directory
//
   if (!(retc = XrdOfsOss->Stat(path, &buf, XRDOSS_resonly, &stat_Env)))
       mode=buf.st_mode;
      else if ((-ENOMSG) != retc)
              return XrdOfsFS->Emsg(epname, einfo, retc, "locate", path);
   return SFS_OK;
}
 
/******************************************************************************/
/*                              t r u n c a t e                               */
/******************************************************************************/
 
int XrdOfs::truncate(const char             *path,    // In
                           XrdSfsFileOffset  Size,    // In
                           XrdOucErrInfo    &einfo,   // Out
                     const XrdSecEntity     *client,  // In
                     const char             *info)    // In
/*
  Function: Change the mode on a file or directory.
 
  Input:    path      - Is the fully qualified name of the file to be removed.
            Size      - the size the file should have.
            einfo     - Error information object to hold error details.
            client    - Authentication credentials, if any.
            info      - Opaque information to be used as seen fit.
 
  Output:   Returns SFS_OK upon success and SFS_ERROR upon failure.
*/
{
   EPNAME("truncate");
   const char *tident = einfo.getErrUser();
   XrdOucEnv trunc_Env(info,0,client);
   int retc;
   XTRACE(truncate, path, "");
 
// Apply security, as needed
//
   AUTHORIZE(client,&trunc_Env,AOP_Update,"truncate",path,einfo);
 
// Find out where we should chmod this file
//
   if (Finder && Finder->isRemote())
      {if (fwdTRUNC.Cmd)
          {char xSz[32];
           sprintf(xSz, "%lld", static_cast<long long>(Size));
           if (Forward(retc,einfo,fwdTRUNC,path,xSz,&trunc_Env)) return retc;
          }
          else if ((retc = Finder->Locate(einfo,path,SFS_O_RDWR,&trunc_Env)))
                  return fsError(einfo, retc);
      }
 
// Check if we should generate an event
//
   if (evsObject && evsObject->Enabled(XrdOfsEvs::Trunc))
      {XrdOfsEvsInfo evInfo(tident, path, info, &trunc_Env, 0, Size);
       evsObject->Notify(XrdOfsEvs::Trunc, evInfo);
      }
 
// Now try to find the file or directory
//
   if (!(retc = XrdOfsOss->Truncate(path, Size, &trunc_Env))) return SFS_OK;
 
// An error occurred, return the error info
//
   return XrdOfsFS->Emsg(epname, einfo, retc, "trunc", path);
}
 
/******************************************************************************/
/*                                  E m s g                                   */
/******************************************************************************/
 
int XrdOfs::Emsg(const char    *pfx,    // Message prefix value
                 XrdOucErrInfo &einfo,  // Place to put text & error code
                 int            ecode,  // The error code
                 const char    *op,     // Operation being performed
                 XrdOfsHandle  *hP)     // The target handle
{
   int rc;
 
// First issue the error message so if we have to unpersist it makes sense
//
   if ((rc = Emsg(pfx, einfo, ecode, op, hP->Name())) != SFS_ERROR) return rc;
 
// If this is a POSC file then we need to unpersist it. Note that we are always
// called with the handle **unlocked**
//
   if (hP->isRW == XrdOfsHandle::opPC)
      {hP->Lock();
       XrdOfsFS->Unpersist(hP);
       hP->UnLock();
      }
 
// Now return the error
//
   return SFS_ERROR;
}
 
/******************************************************************************/
 
int XrdOfs::Emsg(const char    *pfx,    // Message prefix value
                 XrdOucErrInfo &einfo,  // Place to put text & error code
                 int            ecode,  // The error code
                 const char    *op,     // Operation being performed
                 const char    *target) // The target (e.g., fname)
{
   char buffer[MAXPATHLEN+80];
 
// If the error is EBUSY then we just need to stall the client. This is
// a hack in order to provide for proxy support
//
// The hack unfotunately is now beinng triggered for reads and writes when
// it was never so before (presumably due to client changes). So do not
// apply the hack for these operations. This gets a better fix in R 6.0
//
if (strcmp("read", op) && strcmp("readv", op) && strcmp("pgRead", op) && 
    strcmp("write",op) && strcmp("pgwrite",op)) {
    if (ecode < 0) ecode = -ecode;
    if (ecode == EBUSY) return 5;  // A hack for proxy support
 
// Check for timeout conditions that require a client delay
//
   if (ecode == ETIMEDOUT) return OSSDelay;
   }
 
// Format the error message
//
   XrdOucERoute::Format(buffer, sizeof(buffer), ecode, op, target);
 
// Print it out if debugging is enabled
//
#ifndef NODEBUG
    OfsEroute.Emsg(pfx, einfo.getErrUser(), buffer);
#endif
 
// Place the error message in the error object and return
//
    einfo.setErrInfo(ecode, buffer);
    return SFS_ERROR;
}
 
/******************************************************************************/
/*                     P R I V A T E    S E C T I O N                         */
/******************************************************************************/
 
/******************************************************************************/
/*                                 F n a m e                                  */
/******************************************************************************/
 
const char *XrdOfs::Fname(const char *path)
{
   int i = strlen(path)-1;
   while(i) if (path[i] == '/') return &path[i+1];
               else i--;
   return path;
}
 
/******************************************************************************/
/*                               F o r w a r d                                */
/******************************************************************************/
  
int XrdOfs::Forward(int &Result, XrdOucErrInfo &Resp, struct fwdOpt &Fwd,
                    const char *arg1, const char *arg2,
                    XrdOucEnv  *Env1, XrdOucEnv  *Env2)
{
   int retc;
 
   if ((retc = Finder->Forward(Resp, Fwd.Cmd, arg1, arg2, Env1, Env2)))
      {Result = fsError(Resp, retc);
       return 1;
      }
 
   if (Fwd.Port <= 0)
      {Result = SFS_OK;
       return (Fwd.Port ? 0 : 1);
      }
 
   Resp.setErrInfo(Fwd.Port, Fwd.Host);
   Result = SFS_REDIRECT;
   OfsStats.Data.numRedirect++;
   return 1;
}
 
/******************************************************************************/
/*                               f s E r r o r                                */
/******************************************************************************/
  
int XrdOfs::fsError(XrdOucErrInfo &myError, int rc)
{
 
// Screen the error code (update statistics w/o a lock for speed!)
//
   if (rc == SFS_REDIRECT) {OfsStats.Data.numRedirect++; return SFS_REDIRECT;}
   if (rc == SFS_STARTED)  {OfsStats.Data.numStarted++;  return SFS_STARTED; }
   if (rc > 0)             {OfsStats.Data.numDelays++;   return rc;          }
   if (rc == SFS_DATA)     {OfsStats.Data.numReplies++;  return SFS_DATA;    }
                           {OfsStats.Data.numErrors++;   return SFS_ERROR;   }
}
 
/******************************************************************************/
/*                              R e f o r m a t                               */
/******************************************************************************/
  
int XrdOfs::Reformat(XrdOucErrInfo &myError)
{
   static const char *fmt = "oss.cgroup=all&oss.space=%llu&oss.free=%llu"
                            "&oss.maxf=%llu&oss.used=%llu&oss.quota=-1";
   char qsFmt, *bP;
   unsigned long long totSpace, totFree, maxFree;
   int n, blen;
 
// Get the buffer
//
   bP = myError.getMsgBuff(blen);
 
// Scan out the values
//
   n = sscanf(bP, "%c %llu %llu %llu", &qsFmt, &totSpace, &totFree, &maxFree);
 
// Validate the response. The response will be invalid for older cmsd's
//
   if (n != 4 || qsFmt != 'A')
      {myError.setErrInfo(ENOTSUP,"space fctl operation not supported by cmsd");
       return SFS_ERROR;
      }
 
// Change megabyte values to actual bytes
//
   totSpace = totSpace << 20LL;
   totFree  = totFree  << 20LL;
   maxFree  = maxFree  << 20LL;
 
// Reformat the result
//
   blen = snprintf(bP,blen,fmt,totSpace,totFree,maxFree,(totSpace-totFree));
                                 
   myError.setErrCode(blen);
   return SFS_DATA;
}
 
/******************************************************************************/
/*                                 S p l i t                                  */
/******************************************************************************/
  
const char * XrdOfs::Split(const char *Args, const char **Opq,
                                 char *Path, int Plen)
{
   int xlen;
   *Opq = index(Args, '?');
   if (!(*Opq)) return Args;
   xlen = (*Opq)-Args;
   if (xlen >= Plen) xlen = Plen-1;
   strncpy(Path, Args, xlen);
   Path[xlen] = 0;
   return Path;
}
 
/******************************************************************************/
/*                                 S t a l l                                  */
/******************************************************************************/
  
int XrdOfs::Stall(XrdOucErrInfo   &einfo, // Error text & code
                  int              stime, // Seconds to stall
                  const char      *path)  // The path to stall on
{
    const char *msgfmt = "File %s is being %s; "
                         "estimated time to completion %s";
    EPNAME("Stall")
#ifndef NODEBUG
    const char *tident = "";
#endif
    char Mbuff[2048], Tbuff[32];
    const char *What = "staged";
 
// Check why the stall is occurring
//
   if (stime < 0) {stime = 60; What = "created";}
 
// Format the stall message
//
    snprintf(Mbuff, sizeof(Mbuff)-1, msgfmt,
             Fname(path), What, WaitTime(stime, Tbuff, sizeof(Tbuff)));
    ZTRACE(delay, "Stall " <<stime <<": " <<Mbuff <<" for " <<path);
 
// Place the error message in the error object and return
//
    einfo.setErrInfo(0, Mbuff);
 
// All done
//
   return (stime > MaxDelay ? MaxDelay : stime);
}
 
/******************************************************************************/
/*                             U n p e r s i s t                              */
/******************************************************************************/
  
void XrdOfs::Unpersist(XrdOfsHandle *oh, int xcev)
{
   EPNAME("Unpersist");
   const char *tident = oh->PoscUsr();
   int   poscNum, retc;
   short theMode;
 
// Trace the call
//
    FTRACE(close, "use=0");
 
// Generate a proper close event as one has not yet been generated
//
   if (xcev && XrdOfsFS->evsObject && *tident != '?'
   &&  XrdOfsFS->evsObject->Enabled(XrdOfsEvs::Closew))
       {XrdOfsEvsInfo evInfo(tident, oh->Name());
        XrdOfsFS->evsObject->Notify(XrdOfsEvs::Closew, evInfo);
       }
 
// Now generate a removal event
//
   if (XrdOfsFS->Balancer) XrdOfsFS->Balancer->Removed(oh->Name());
   if (XrdOfsFS->evsObject && XrdOfsFS->evsObject->Enabled(XrdOfsEvs::Rm))
      {XrdOfsEvsInfo evInfo(tident, oh->Name());
       XrdOfsFS->evsObject->Notify(XrdOfsEvs::Rm, evInfo);
      }
 
// Count this
//
   OfsStats.Add(OfsStats.Data.numUnpsist);
 
// Now unpersist the file
//
   OfsEroute.Emsg(epname, "Unpersisting", tident, oh->Name());
   if ((poscNum = oh->PoscGet(theMode))) poscQ->Del(oh->Name(), poscNum, 1);
       else if ((retc = XrdOfsOss->Unlink(oh->Name())))
               OfsEroute.Emsg(epname, retc, "unpersist", oh->Name());
}
  
/******************************************************************************/
/*                              W a i t T i m e                               */
/******************************************************************************/
 
char *XrdOfs::WaitTime(int stime, char *buff, int blen)
{
   int hr, min, sec;
 
// Compute hours, minutes, and seconds
//
   min = stime / 60;
   sec = stime % 60;
   hr  = min   / 60;
   min = min   % 60;
 
// Now format the message based on time duration
//
        if (!hr && !min)
           snprintf(buff,blen,"%d second%s",sec,(sec > 1 ? "s" : ""));
   else if (!hr)
          {if (sec > 10) min++;
           snprintf(buff,blen,"%d minute%s",min,(min > 1 ? "s" : ""));
          }
   else   {if (hr == 1)
              if (min <= 30)
                      snprintf(buff,blen,"%d minutes",min+60);
                 else snprintf(buff,blen,"%d hour and %d minutes",hr,min);
              else {if (min > 30) hr++;
                      snprintf(buff,blen,"%d hours",hr);
                   }
          }
 
// Complete the message
//
   buff[blen-1] = '\0';
   return buff;
}