1 files changed, 545 insertions, 22 deletions

diff --git a/source3/smbd/fileio.c b/source3/smbd/fileio.c
index 1e166275152..8b48a921fde 100644
--- a/source3/smbd/fileio.c
+++ b/source3/smbd/fileio.c
@@ -23,6 +23,7 @@
 
 extern int DEBUGLEVEL;
 
+static BOOL setup_write_cache(files_struct *, SMB_OFF_T);
 
 /****************************************************************************
 seek a file. Try to avoid the seek if possible
@@ -36,7 +37,19 @@ SMB_OFF_T seek_file(files_struct *fsp,SMB_OFF_T pos)
   if (fsp->print_file && lp_postscript(fsp->conn->service))
     offset = 3;
 
-  seek_ret = fsp->conn->vfs_ops.lseek(fsp->fd_ptr->fd,pos+offset,SEEK_SET);
+  seek_ret = sys_lseek(fsp->fd_ptr->fd,pos+offset,SEEK_SET);
+
+  /*
+   * We want to maintain the fiction that we can seek
+   * on a fifo for file system purposes. This allows 
+   * people to set up UNIX fifo's that feed data to Windows
+   * applications. JRA.
+   */
+
+  if((seek_ret == -1) && (errno == ESPIPE)) {
+    seek_ret = pos+offset;
+    errno = 0;
+  }
 
   if((seek_ret == -1) || (seek_ret != pos+offset)) {
     DEBUG(0,("seek_file: sys_lseek failed. Error was %s\n", strerror(errno) ));
@@ -53,6 +66,29 @@ SMB_OFF_T seek_file(files_struct *fsp,SMB_OFF_T pos)
 }
 
 /****************************************************************************
+ Read from write cache if we can.
+****************************************************************************/
+
+static unsigned int cache_read_hits;
+
+BOOL read_from_write_cache(files_struct *fsp,char *data,SMB_OFF_T pos,size_t n)
+{
+  write_cache *wcp = fsp->wcp;
+
+  if(!wcp)
+    return False;
+
+  if(n > wcp->data_size || pos < wcp->offset || pos + n > wcp->offset + wcp->data_size)
+    return False;
+
+  memcpy(data, wcp->data + (pos - wcp->offset), n);
+
+  cache_read_hits++;
+
+  return True;
+}
+
+/****************************************************************************
 read from a file
 ****************************************************************************/
 
@@ -62,7 +98,7 @@ ssize_t read_file(files_struct *fsp,char *data,SMB_OFF_T pos,size_t n)
 
 #if USE_READ_PREDICTION
   if (!fsp->can_write) {
-    ret = read_predict(fsp, fsp->fd_ptr->fd,pos,data,NULL,n);
+    ret = read_predict(fsp->fd_ptr->fd,pos,data,NULL,n);
 
     data += ret;
     n -= ret;
@@ -70,40 +106,64 @@ ssize_t read_file(files_struct *fsp,char *data,SMB_OFF_T pos,size_t n)
   }
 #endif
 
-#if WITH_MMAP
-  if (fsp->mmap_ptr) {
-	  SMB_OFF_T num = (fsp->mmap_size > pos) ? (fsp->mmap_size - pos) : 0;
-	  num = MIN(n,num);
-	  if (num > 0) {
-		  memcpy(data,fsp->mmap_ptr+pos,num);
-		  data += num;
-		  pos += num;
-		  n -= num;
-		  ret += num;
-	  }
-  }
-#endif
+  /*
+   * Serve from write cache if we can.
+   */
+  if(read_from_write_cache(fsp, data, pos, n))
+    return n;
+
+  flush_write_cache(fsp, READ_FLUSH);
 
   if (seek_file(fsp,pos) == -1) {
     DEBUG(3,("read_file: Failed to seek to %.0f\n",(double)pos));
     return(ret);
   }
-
+  
   if (n > 0) {
-    readret = fsp->conn->vfs_ops.read(fsp->fd_ptr->fd,data,n);
+    readret = read(fsp->fd_ptr->fd,data,n);
     if (readret > 0) ret += readret;
   }
 
   return(ret);
 }
 
+/* Write cache static counters. */
+
+static unsigned int abutted_writes;
+static unsigned int total_writes;
+static unsigned int non_oplock_writes;
+static unsigned int direct_writes;
+static unsigned int init_writes;
+static unsigned int flushed_writes;
+static unsigned int num_perfect_writes;
+static unsigned int flush_reasons[NUM_FLUSH_REASONS];
+
+/* how many write cache buffers have been allocated */
+static unsigned int allocated_write_caches;
+static unsigned int num_write_caches;
+
+/****************************************************************************
+ *Really* write to a file
+****************************************************************************/
+
+static ssize_t real_write_file(files_struct *fsp,char *data,SMB_OFF_T pos, size_t n)
+{
+  if ((pos != -1) && (seek_file(fsp,pos) == -1))
+    return -1;
+
+  return write_data(fsp->fd_ptr->fd,data,n);
+}
 
 /****************************************************************************
 write to a file
 ****************************************************************************/
 
-ssize_t write_file(files_struct *fsp,char *data,size_t n)
+ssize_t write_file(files_struct *fsp, char *data, SMB_OFF_T pos, size_t n)
 {
+  write_cache *wcp = fsp->wcp;
+  ssize_t total_written = 0;
+  int write_path = -1; 
+
   if (!fsp->can_write) {
     errno = EPERM;
     return(0);
@@ -112,25 +172,488 @@ ssize_t write_file(files_struct *fsp,char *data,size_t n)
   if (!fsp->modified) {
     SMB_STRUCT_STAT st;
     fsp->modified = True;
-    if (fsp->conn->vfs_ops.fstat(fsp->fd_ptr->fd,&st) == 0) {
+
+    if (sys_fstat(fsp->fd_ptr->fd,&st) == 0) {
       int dosmode = dos_mode(fsp->conn,fsp->fsp_name,&st);
       if (MAP_ARCHIVE(fsp->conn) && !IS_DOS_ARCHIVE(dosmode)) {	
         file_chmod(fsp->conn,fsp->fsp_name,dosmode | aARCH,&st);
       }
+
+      /*
+       * If this is the first write and we have an exclusive oplock then setup
+       * the write cache.
+       */
+
+      if ((fsp->oplock_type == EXCLUSIVE_OPLOCK) && !wcp) {
+        setup_write_cache(fsp, st.st_size);
+        wcp = fsp->wcp;
+      } 
     }  
   }
 
-  return(vfs_write_data(fsp,data,n));
+  total_writes++;
+  if (!fsp->oplock_type) {
+    non_oplock_writes++;
+  }
+
+  /*
+   * If this file is level II oplocked then we need
+   * to grab the shared memory lock and inform all
+   * other files with a level II lock that they need
+   * to flush their read caches. We keep the lock over
+   * the shared memory area whilst doing this.
+   */
+
+  if (LEVEL_II_OPLOCK_TYPE(fsp->oplock_type)) {
+    SMB_DEV_T dev = fsp->fd_ptr->dev;
+    SMB_INO_T inode = fsp->fd_ptr->inode;
+    share_mode_entry *share_list = NULL;
+    pid_t pid = getpid();
+    int token = -1;
+    int num_share_modes = 0;
+    int i;
+
+    if (lock_share_entry(fsp->conn, dev, inode, &token) == False) {
+      DEBUG(0,("write_file: failed to lock share mode entry for file %s.\n", fsp->fsp_name ));
+    }
+
+    num_share_modes = get_share_modes(fsp->conn, token, dev, inode, &share_list);
+
+    for(i = 0; i < num_share_modes; i++) {
+      share_mode_entry *share_entry = &share_list[i];
+
+      /*
+       * As there could have been multiple writes waiting at the lock_share_entry
+       * gate we may not be the first to enter. Hence the state of the op_types
+       * in the share mode entries may be partly NO_OPLOCK and partly LEVEL_II
+       * oplock. It will do no harm to re-send break messages to those smbd's
+       * that are still waiting their turn to remove their LEVEL_II state, and
+       * also no harm to ignore existing NO_OPLOCK states. JRA.
+       */
+
+      if (share_entry->op_type == NO_OPLOCK)
+        continue;
+
+      /* Paranoia .... */
+      if (EXCLUSIVE_OPLOCK_TYPE(share_entry->op_type)) {
+        DEBUG(0,("write_file: PANIC. share mode entry %d is an exlusive oplock !\n", i ));
+        abort();
+      }
+
+      /*
+       * Check if this is a file we have open (including the
+       * file we've been called to do write_file on. If so
+       * then break it directly without releasing the lock.
+       */
+
+      if (pid == share_entry->pid) {
+        files_struct *new_fsp = file_find_dit(dev, inode, &share_entry->time);
+
+        /* Paranoia check... */
+        if(new_fsp == NULL) {
+          DEBUG(0,("write_file: PANIC. share mode entry %d is not a local file !\n", i ));
+          abort();
+        }
+        oplock_break_level2(new_fsp, True, token);
+
+      } else {
+
+        /*
+         * This is a remote file and so we send an asynchronous
+         * message.
+         */
+
+        request_oplock_break(share_entry, dev, inode);
+      }
+    }
+ 
+    free((char *)share_list);
+    unlock_share_entry(fsp->conn, dev, inode, token);
+  }
+
+  /* Paranoia check... */
+  if (LEVEL_II_OPLOCK_TYPE(fsp->oplock_type)) {
+    DEBUG(0,("write_file: PANIC. File %s still has a level II oplock.\n", fsp->fsp_name));
+    abort();
+  }
+
+  if (total_writes % 500 == 0) {
+    DEBUG(3,("WRITECACHE: initwrites=%u abutted=%u flushes=%u total=%u \
+nonop=%u allocated=%u active=%u direct=%u perfect=%u readhits=%u\n",
+         init_writes, abutted_writes, flushed_writes, total_writes, non_oplock_writes,
+         allocated_write_caches,
+         num_write_caches, direct_writes, num_perfect_writes, cache_read_hits ));
+
+    DEBUG(3,("WRITECACHE: SEEK=%d, READ=%d, WRITE=%d, READRAW=%d, OPLOCK=%d, CLOSE=%d, SYNC=%d\n",
+    flush_reasons[SEEK_FLUSH],
+    flush_reasons[READ_FLUSH],
+    flush_reasons[WRITE_FLUSH],
+    flush_reasons[READRAW_FLUSH],
+    flush_reasons[OPLOCK_RELEASE_FLUSH],
+    flush_reasons[CLOSE_FLUSH],
+    flush_reasons[SYNC_FLUSH] ));
+  }
+
+  if(!wcp) {
+    direct_writes++;
+    return real_write_file(fsp, data, pos, n);
+  }
+
+  DEBUG(9,("write_file(fd=%d pos=%d size=%d) wofs=%d wsize=%d\n",
+       fsp->fd_ptr->fd, (int)pos, (int)n, (int)wcp->offset, (int)wcp->data_size));
+
+  /* 
+   * If we have active cache and it isn't contiguous then we flush.
+   * NOTE: There is a small problem with running out of disk ....
+   */
+
+  if (wcp->data_size) {
+
+    BOOL cache_flush_needed = False;
+
+    if ((pos >= wcp->offset) && (pos <= wcp->offset + wcp->data_size)) {
+      
+      /*
+       * Start of write overlaps or abutts the existing data.
+       */
+
+      size_t data_used = MIN((wcp->alloc_size - (pos - wcp->offset)), n);
+
+      memcpy(wcp->data + (pos - wcp->offset), data, data_used);
+
+      /*
+       * Update the current buffer size with the new data.
+       */
+
+      if(pos + data_used > wcp->offset + wcp->data_size)
+        wcp->data_size = pos + data_used - wcp->offset;
+
+      /*
+       * If we used all the data then
+       * return here.
+       */
+
+      if(n == data_used)
+        return n;
+      else
+        cache_flush_needed = True;
+
+      /*
+       * Move the start of data forward by the amount used,
+       * cut down the amount left by the same amount.
+       */
+
+      data += data_used;
+      pos += data_used;
+      n -= data_used;
+
+      abutted_writes++;
+      total_written = data_used;
+
+      write_path = 1;
+
+    } else if ((pos < wcp->offset) && (pos + n > wcp->offset) && 
+               (pos + n <= wcp->offset + wcp->alloc_size)) {
+
+      /*
+       * End of write overlaps the existing data.
+       */
+
+      size_t data_used = pos + n - wcp->offset;
+
+      memcpy(wcp->data, data + n - data_used, data_used);
+
+      /*
+       * Update the current buffer size with the new data.
+       */
+
+      if(pos + n > wcp->offset + wcp->data_size)
+        wcp->data_size = pos + n - wcp->offset;
+
+      /*
+       * We don't need to move the start of data, but we
+       * cut down the amount left by the amount used.
+       */
+
+      n -= data_used;
+
+      /*
+       * We cannot have used all the data here.
+       */
+
+      cache_flush_needed = True;
+
+      abutted_writes++;
+      total_written = data_used;
+
+      write_path = 2;
+
+    } else if ( (pos >= wcp->file_size) && 
+                (pos > wcp->offset + wcp->data_size) && 
+                (pos < wcp->offset + wcp->alloc_size) ) {
+
+      /*
+       * Non-contiguous write part of which fits within
+       * the cache buffer and is extending the file.
+       */
+
+      size_t data_used;
+
+      if(pos + n <= wcp->offset + wcp->alloc_size)
+        data_used = n;
+      else
+        data_used = wcp->offset + wcp->alloc_size - pos;
+
+      /*
+       * Fill in the non-continuous area with zeros.
+       */
+
+      memset(wcp->data + wcp->data_size, '\0',
+             pos - (wcp->offset + wcp->data_size) );
+
+      memcpy(wcp->data + (pos - wcp->offset), data, data_used);
+
+      /*
+       * Update the current buffer size with the new data.
+       */
+
+      if(pos + data_used > wcp->offset + wcp->data_size)
+        wcp->data_size = pos + data_used - wcp->offset;
+
+      /*
+       * Update the known file length.
+       */
+
+      wcp->file_size = wcp->offset + wcp->data_size;
+
+#if 0
+      if (set_filelen(fsp->fd_ptr->fd, wcp->file_size) == -1) {
+        DEBUG(0,("write_file: error %s in setting file to length %.0f\n",
+          strerror(errno), (double)wcp->file_size ));
+        return -1;
+      }
+#endif
+
+      /*
+       * If we used all the data then
+       * return here.
+       */
+
+      if(n == data_used)
+        return n;
+      else
+        cache_flush_needed = True;
+
+      /*
+       * Move the start of data forward by the amount used,
+       * cut down the amount left by the same amount.
+       */
+
+      data += data_used;
+      pos += data_used;
+      n -= data_used;
+
+      abutted_writes++;
+      total_written = data_used;
+
+      write_path = 3;
+
+    } else {
+
+      /*
+       * Write is bigger than buffer, or there is no overlap on the
+       * low or high ends.
+       */
+
+      DEBUG(9,("write_file: non cacheable write : fd = %d, pos = %.0f, len = %u, current cache pos = %.0f \
+len = %u\n",fsp->fd_ptr->fd, (double)pos, (unsigned int)n, (double)wcp->offset, (unsigned int)wcp->data_size ));
+
+      /*
+       * Update the file size if needed.
+       */
+
+      if(pos + n > wcp->file_size)
+        wcp->file_size = pos + n;
+
+      /*
+       * If write would fit in the cache, and is larger than
+       * the data already in the cache, flush the cache and
+       * preferentially copy the data new data into it. Otherwise
+       * just write the data directly.
+       */
+
+      if ( n <= wcp->alloc_size && n > wcp->data_size) {
+        cache_flush_needed = True;
+      } else {
+        direct_writes++;
+        return real_write_file(fsp, data, pos, n);
+      }
+
+      write_path = 4;
+
+    }
+
+    if(wcp->data_size > wcp->file_size)
+      wcp->file_size = wcp->data_size;
+
+    if (cache_flush_needed) {
+      flushed_writes++;
+
+      DEBUG(3,("WRITE_FLUSH:%d: due to noncontinuous write: fd = %d, size = %.0f, pos = %.0f, \
+n = %u, wcp->offset=%.0f, wcp->data_size=%u\n",
+             write_path, fsp->fd_ptr->fd, (double)wcp->file_size, (double)pos, (unsigned int)n,
+             (double)wcp->offset, (unsigned int)wcp->data_size ));
+
+      flush_write_cache(fsp, WRITE_FLUSH);
+    }
+  }
+
+  /*
+   * If the write request is bigger than the cache
+   * size, write it all out.
+   */
+
+  if (n > wcp->alloc_size ) {
+    if(real_write_file(fsp, data, pos, n) == -1)
+      return -1;
+    direct_writes++;
+    return total_written + n;
+  }
+
+  /*
+   * If there's any data left, cache it.
+   */
+
+  if (n) {
+    if (wcp->data_size) {
+      abutted_writes++;
+      DEBUG(9,("abutted write (%u)\n", abutted_writes));
+    } else {
+      init_writes++;
+    }
+    memcpy(wcp->data+wcp->data_size, data, n);
+    if (wcp->data_size == 0) {
+      wcp->offset = pos;
+      num_write_caches++;
+    }
+    wcp->data_size += n;
+    DEBUG(9,("cache return %u\n", (unsigned int)n));
+    total_written += n;
+    return total_written; /* .... that's a write :) */
+  }
+  
+  return total_written;
 }
 
+/****************************************************************************
+ Delete the write cache structure.
+****************************************************************************/
 
+void delete_write_cache(files_struct *fsp)
+{
+  write_cache *wcp;
+
+  if(!fsp)
+    return;
+
+  if(!(wcp = fsp->wcp))
+    return;
+
+  allocated_write_caches--;
+
+  SMB_ASSERT(wcp->data_size == 0);
+
+  free(wcp->data);
+  free(wcp);
+
+  fsp->wcp = NULL;
+}
+
+/****************************************************************************
+ Setup the write cache structure.
+****************************************************************************/
+
+static BOOL setup_write_cache(files_struct *fsp, SMB_OFF_T file_size)
+{
+  ssize_t alloc_size = lp_write_cache_size(SNUM(fsp->conn));
+  write_cache *wcp;
+
+  if (allocated_write_caches >= MAX_WRITE_CACHES) return False;
+
+  if(alloc_size == 0 || fsp->wcp)
+    return False;
+
+  if((wcp = (write_cache *)malloc(sizeof(write_cache))) == NULL) {
+    DEBUG(0,("setup_write_cache: malloc fail.\n"));
+    return False;
+  }
+
+  wcp->file_size = file_size;
+  wcp->offset = 0;
+  wcp->alloc_size = alloc_size;
+  wcp->data_size = 0;
+  if((wcp->data = malloc(wcp->alloc_size)) == NULL) {
+    DEBUG(0,("setup_write_cache: malloc fail for buffer size %u.\n",
+          (unsigned int)wcp->alloc_size ));
+    free(wcp);
+    return False;
+  }
+
+  fsp->wcp = wcp;
+  allocated_write_caches++;
+
+  return True;
+}
+
+/****************************************************************************
+ Cope with a size change.
+****************************************************************************/
+
+void set_filelen_write_cache(files_struct *fsp, SMB_OFF_T file_size)
+{
+  if(fsp->wcp) {
+    flush_write_cache(fsp, SIZECHANGE_FLUSH);
+    fsp->wcp->file_size = file_size;
+  }
+}
+
+/*******************************************************************
+ Flush a write cache struct to disk.
+********************************************************************/
+
+ssize_t flush_write_cache(files_struct *fsp, enum flush_reason_enum reason)
+{
+  write_cache *wcp = fsp->wcp;
+  size_t data_size;
+
+  if(!wcp || !wcp->data_size)
+    return 0;
+
+  data_size = wcp->data_size;
+  wcp->data_size = 0;
+
+  num_write_caches--;
+
+  flush_reasons[(int)reason]++;
+
+  DEBUG(9,("flushing write cache: fd = %d, off=%.0f, size=%u\n",
+       fsp->fd_ptr->fd, (double)wcp->offset, (unsigned int)data_size));
+
+  if(data_size == wcp->alloc_size)
+    num_perfect_writes++;
+
+  return real_write_file(fsp, wcp->data, wcp->offset, data_size);
+}
 /*******************************************************************
 sync a file
 ********************************************************************/
 
-void sys_sync_file(int fd)
+void sync_file(connection_struct *conn, files_struct *fsp)
 {
 #ifdef HAVE_FSYNC
-    fsync(fd);
+    if(lp_strict_sync(SNUM(conn)) && fsp->fd_ptr != NULL) {
+      flush_write_cache(fsp, SYNC_FLUSH);
+      fsync(fsp->fd_ptr->fd);
+    }
 #endif
 }