Add some notes about CF_CORE and new Options{} Plugin

[bacula/docs] / docs / manuals / en / developers / pluginAPI.tex

%%
%%

\chapter{Bacula FD Plugin API}
To write a Bacula plugin, you create a dynamic shared object program (or dll on
Win32) with a particular name and two exported entry points, place it in the
{\bf Plugins Directory}, which is defined in the {\bf bacula-fd.conf} file in
the {\bf Client} resource, and when the FD starts, it will load all the plugins
that end with {\bf -fd.so} (or {\bf -fd.dll} on Win32) found in that directory.

\section{Normal vs Command vs Options Plugins}
In general, there are three ways that plugins are called. The first way, is when
a particular event is detected in Bacula, it will transfer control to each
plugin that is loaded in turn informing the plugin of the event.  This is very
similar to how a {\bf RunScript} works, and the events are very similar.  Once
the plugin gets control, it can interact with Bacula by getting and setting
Bacula variables.  In this way, it behaves much like a RunScript.  Currently
very few Bacula variables are defined, but they will be implemented as the need
arises, and it is very extensible.

We plan to have plugins register to receive events that they normally would
not receive, such as an event for each file examined for backup or restore.
This feature is not yet implemented.

The second type of plugin, which is more useful and fully implemented in the
current version is what we call a command plugin.  As with all plugins, it gets
notified of important events as noted above (details described below), but in
addition, this kind of plugin can accept a command line, which is a:

\begin{verbatim}
   Plugin = <command-string>
\end{verbatim}

directive that is placed in the Include section of a FileSet and is very
similar to the "File = " directive.  When this Plugin directive is encountered
by Bacula during backup, it passes the "command" part of the Plugin directive
only to the plugin that is explicitly named in the first field of that command
string.  This allows that plugin to backup any file or files on the system that
it wants. It can even create "virtual files" in the catalog that contain data
to be restored but do not necessarily correspond to actual files on the
filesystem.

The important features of the command plugin entry points are:
\begin{itemize}
 \item It is triggered by a "Plugin =" directive in the FileSet
 \item Only a single plugin is called that is named on the "Plugin =" directive.
 \item The full command string after the "Plugin =" is passed to the plugin
    so that it can be told what to backup/restore.
\end{itemize}

The third type of plugin is the Options Plugin, this kind of plugin is useful
to implement some custom filter on data. For example, you can implement a
compression algorithm in a very simple way. Bacula will call this plugin for
each file that is selected in a FileSet (according to
Wild/Regex/Exclude/Include rules). As with all plugins, it gets notified of
important events as noted above (details described below), but in addition,
this kind of plugin can be placed in a Options group, which is a:

\begin{verbatim}
 FileSet {
    Name = TestFS
    Include {
       Options {
          Compression = GZIP1
          Signature = MD5
          Wild = "*.txt"
          Plugin = <command-string>
       }
       File = /
    }
}
\end{verbatim}

\section{Loading Plugins}
Once the File daemon loads the plugins, it asks the OS for the
two entry points (loadPlugin and unloadPlugin) then calls the
{\bf loadPlugin} entry point (see below).

Bacula passes information to the plugin through this call and it gets
back information that it needs to use the plugin.  Later, Bacula
 will call particular functions that are defined by the
{\bf loadPlugin} interface.  

When Bacula is finished with the plugin 
(when Bacula is going to exit), it will call the {\bf unloadPlugin}
entry point.

The two entry points are:

\begin{verbatim}
bRC loadPlugin(bInfo *lbinfo, bFuncs *lbfuncs, pInfo **pinfo, pFuncs **pfuncs)

and

bRC unloadPlugin()
\end{verbatim}

both these external entry points to the shared object are defined as C entry
points to avoid name mangling complications with C++.  However, the shared
object can actually be written in any language (preferably C or C++) providing
that it follows C language calling conventions.

The definitions for {\bf bRC} and the arguments are {\bf
src/filed/fd-plugins.h} and so this header file needs to be included in
your plugin.  It along with {\bf src/lib/plugins.h} define basically the whole
plugin interface.  Within this header file, it includes the following
files:

\begin{verbatim}
#include <sys/types.h>
#include "config.h"
#include "bc_types.h"
#include "lib/plugins.h"
#include <sys/stat.h>
\end{verbatim}

Aside from the {\bf bc\_types.h} and {\bf confit.h} headers, the plugin
definition uses the minimum code from Bacula.  The bc\_types.h file is required
to ensure that the data type definitions in arguments correspond to the Bacula
core code.

The return codes are defined as:
\begin{verbatim}
typedef enum {
  bRC_OK    = 0,                         /* OK */
  bRC_Stop  = 1,                         /* Stop calling other plugins */
  bRC_Error = 2,                         /* Some kind of error */
  bRC_More  = 3,                         /* More files to backup */
  bRC_Term  = 4,                         /* Unload me */
  bRC_Seen  = 5,                         /* Return code from checkFiles */
  bRC_Core  = 6,                         /* Let Bacula core handles this file */
  bRC_Skip  = 7,                         /* Skip the proposed file */
} bRC;
\end{verbatim}


At a future point in time, we hope to make the Bacula libbac.a into a
shared object so that the plugin can use much more of Bacula's
infrastructure, but for this first cut, we have tried to minimize the
dependence on Bacula.

\section{loadPlugin}
As previously mentioned, the {\bf loadPlugin} entry point in the plugin
is called immediately after Bacula loads the plugin when the File daemon
itself is first starting.  This entry point is only called once during the
execution of the File daemon.  In calling the
plugin, the first two arguments are information from Bacula that
is passed to the plugin, and the last two arguments are information
about the plugin that the plugin must return to Bacula.  The call is:

\begin{verbatim}
bRC loadPlugin(bInfo *lbinfo, bFuncs *lbfuncs, pInfo **pinfo, pFuncs **pfuncs)
\end{verbatim}

and the arguments are:

\begin{description}
\item [lbinfo]
This is information about Bacula in general. Currently, the only value
defined in the bInfo structure is the version, which is the Bacula plugin 
interface version, currently defined as 1.  The {\bf size} is set to the
byte size of the structure. The exact definition of the bInfo structure
as of this writing is: 

\begin{verbatim}
typedef struct s_baculaInfo {
   uint32_t size;
   uint32_t version;
} bInfo;
\end{verbatim}

\item [lbfuncs]
The bFuncs structure defines the callback entry points within Bacula
that the plugin can use register events, get Bacula values, set
Bacula values, and send messages to the Job output or debug output.

The exact definition as of this writing is:
\begin{verbatim}
typedef struct s_baculaFuncs {
   uint32_t size;
   uint32_t version;
   bRC (*registerBaculaEvents)(bpContext *ctx, ...);
   bRC (*getBaculaValue)(bpContext *ctx, bVariable var, void *value);
   bRC (*setBaculaValue)(bpContext *ctx, bVariable var, void *value);
   bRC (*JobMessage)(bpContext *ctx, const char *file, int line,
       int type, utime_t mtime, const char *fmt, ...);
   bRC (*DebugMessage)(bpContext *ctx, const char *file, int line,
       int level, const char *fmt, ...);
   void *(*baculaMalloc)(bpContext *ctx, const char *file, int line,
       size_t size);
   void (*baculaFree)(bpContext *ctx, const char *file, int line, void *mem);
} bFuncs;
\end{verbatim}

We will discuss these entry points and how to use them a bit later when
describing the plugin code.


\item [pInfo]
When the loadPlugin entry point is called, the plugin must initialize
an information structure about the plugin and return a pointer to
this structure to Bacula.

The exact definition as of this writing is:

\begin{verbatim}
typedef struct s_pluginInfo {
   uint32_t size;
   uint32_t version;
   const char *plugin_magic;
   const char *plugin_license;
   const char *plugin_author;
   const char *plugin_date;
   const char *plugin_version;
   const char *plugin_description;
} pInfo;
\end{verbatim}

Where:
 \begin{description}
 \item [version] is the current Bacula defined plugin interface version, currently
   set to 1. If the interface version differs from the current version of 
   Bacula, the plugin will not be run (not yet implemented).
 \item [plugin\_magic] is a pointer to the text string "*FDPluginData*", a
   sort of sanity check.  If this value is not specified, the plugin
   will not be run (not yet implemented).
 \item [plugin\_license] is a pointer to a text string that describes the
   plugin license. Bacula will only accept compatible licenses (not yet
   implemented).
 \item [plugin\_author] is a pointer to the text name of the author of the program.
   This string can be anything but is generally the author's name.
 \item [plugin\_date] is the pointer text string containing the date of the plugin.
   This string can be anything but is generally some human readable form of 
   the date.
 \item [plugin\_version] is a pointer to a text string containing the version of
   the plugin.  The contents are determined by the plugin writer.
 \item [plugin\_description] is a pointer to a string describing what the
   plugin does. The contents are determined by the plugin writer.
 \end{description}

The pInfo structure must be defined in static memory because Bacula does not
copy it and may refer to the values at any time while the plugin is
loaded. All values must be supplied or the plugin will not run (not yet
implemented).  All text strings must be either ASCII or UTF-8 strings that
are terminated with a zero byte.

\item [pFuncs]
When the loadPlugin entry point is called, the plugin must initialize
an entry point structure about the plugin and return a pointer to
this structure to Bacula. This structure contains pointer to each
of the entry points that the plugin must provide for Bacula. When
Bacula is actually running the plugin, it will call the defined
entry points at particular times.  All entry points must be defined.

The pFuncs structure must be defined in static memory because Bacula does not
copy it and may refer to the values at any time while the plugin is
loaded.

The exact definition as of this writing is:

\begin{verbatim}
typedef struct s_pluginFuncs {
   uint32_t size;
   uint32_t version;
   bRC (*newPlugin)(bpContext *ctx);
   bRC (*freePlugin)(bpContext *ctx);
   bRC (*getPluginValue)(bpContext *ctx, pVariable var, void *value);
   bRC (*setPluginValue)(bpContext *ctx, pVariable var, void *value);
   bRC (*handlePluginEvent)(bpContext *ctx, bEvent *event, void *value);
   bRC (*startBackupFile)(bpContext *ctx, struct save_pkt *sp);
   bRC (*endBackupFile)(bpContext *ctx);
   bRC (*startRestoreFile)(bpContext *ctx, const char *cmd);
   bRC (*endRestoreFile)(bpContext *ctx);
   bRC (*pluginIO)(bpContext *ctx, struct io_pkt *io);
   bRC (*createFile)(bpContext *ctx, struct restore_pkt *rp);
   bRC (*setFileAttributes)(bpContext *ctx, struct restore_pkt *rp);
   bRC (*checkFile)(bpContext *ctx, char *fname);
} pFuncs;
\end{verbatim}

The details of the entry points will be presented in
separate sections below. 

Where:
 \begin{description}
 \item [size] is the byte size of the structure.
 \item [version] is the plugin interface version currently set to 3.
 \end{description}

Sample code for loadPlugin:
\begin{verbatim}
  bfuncs = lbfuncs;                  /* set Bacula funct pointers */
  binfo  = lbinfo;
  *pinfo  = &pluginInfo;             /* return pointer to our info */
  *pfuncs = &pluginFuncs;            /* return pointer to our functions */

   return bRC_OK;
\end{verbatim}

where pluginInfo and pluginFuncs are statically defined structures. 
See bpipe-fd.c for details.



\end{description}

\section{Plugin Entry Points}
This section will describe each of the entry points (subroutines) within
the plugin that the plugin must provide for Bacula, when they are called
and their arguments. As noted above, pointers to these subroutines are
passed back to Bacula in the pFuncs structure when Bacula calls the 
loadPlugin() externally defined entry point.

\subsection{newPlugin(bpContext *ctx)}
  This is the entry point that Bacula will call
  when a new "instance" of the plugin is created. This typically
  happens at the beginning of a Job.  If 10 Jobs are running
  simultaneously, there will be at least 10 instances of the
  plugin.

  The bpContext structure will be passed to the plugin, and
  during this call, if the plugin needs to have its private
  working storage that is associated with the particular
  instance of the plugin, it should create it from the heap
  (malloc the memory) and store a pointer to
  its private working storage in the {\bf pContext} variable.
  Note: since Bacula is a multi-threaded program, you must not
  keep any variable data in your plugin unless it is truly meant
  to apply globally to the whole plugin.  In addition, you must
  be aware that except the first and last call to the plugin
  (loadPlugin and unloadPlugin) all the other calls will be 
  made by threads that correspond to a Bacula job.  The 
  bpContext that will be passed for each thread will remain the
  same throughout the Job thus you can keep your private Job specific
  data in it ({\bf bContext}).

\begin{verbatim}
typedef struct s_bpContext {
  void *pContext;   /* Plugin private context */
  void *bContext;   /* Bacula private context */
} bpContext;

\end{verbatim}
   
  This context pointer will be passed as the first argument to all
  the entry points that Bacula calls within the plugin.  Needless
  to say, the plugin should not change the bContext variable, which
  is Bacula's private context pointer for this instance (Job) of this
  plugin.

\subsection{freePlugin(bpContext *ctx)}
This entry point is called when the
this instance of the plugin is no longer needed (the Job is
ending), and the plugin should release all memory it may
have allocated for this particular instance (Job) i.e. the pContext.  
This is not the final termination
of the plugin signaled by a call to {\bf unloadPlugin}. 
Any other instances (Job) will
continue to run, and the entry point {\bf newPlugin} may be called
again if other jobs start.

\subsection{getPluginValue(bpContext *ctx, pVariable var, void *value)} 
Bacula will call this entry point to get
a value from the plugin.  This entry point is currently not called.

\subsection{setPluginValue(bpContext *ctx, pVariable var, void *value)}
Bacula will call this entry point to set
a value in the plugin.  This entry point is currently not called.
 
\subsection{handlePluginEvent(bpContext *ctx, bEvent *event, void *value)}
This entry point is called when Bacula
encounters certain events (discussed below). This is, in fact, the 
main way that most plugins get control when a Job runs and how
they know what is happening in the job. It can be likened to the
{\bf RunScript} feature that calls external programs and scripts,
and is very similar to the Bacula Python interface.
When the plugin is called, Bacula passes it the pointer to an event
structure (bEvent), which currently has one item, the eventType:

\begin{verbatim}
typedef struct s_bEvent {
   uint32_t eventType;
} bEvent;
\end{verbatim}

  which defines what event has been triggered, and for each event,
  Bacula will pass a pointer to a value associated with that event.
  If no value is associated with a particular event, Bacula will 
  pass a NULL pointer, so the plugin must be careful to always check
  value pointer prior to dereferencing it.
  
  The current list of events are:

\begin{verbatim}
typedef enum {
  bEventJobStart                        = 1,
  bEventJobEnd                          = 2,
  bEventStartBackupJob                  = 3,
  bEventEndBackupJob                    = 4,
  bEventStartRestoreJob                 = 5,
  bEventEndRestoreJob                   = 6,
  bEventStartVerifyJob                  = 7,
  bEventEndVerifyJob                    = 8,
  bEventBackupCommand                   = 9,
  bEventRestoreCommand                  = 10,
  bEventLevel                           = 11,
  bEventSince                           = 12,
  bEventCancelCommand                   = 13,  /* Executed by another thread */
 
  /* Just before bEventVssPrepareSnapshot */
  bEventVssBackupAddComponents          = 14,  

  bEventVssRestoreLoadComponentMetadata = 15,
  bEventVssRestoreSetComponentsSelected = 16,
  bEventRestoreObject                   = 17,
  bEventEndFileSet                      = 18,
  bEventPluginCommand                   = 19,
  bEventVssBeforeCloseRestore           = 21,

  /* Add drives to VSS snapshot 
   *  argument: char[27] drivelist
   * You need to add them without duplicates, 
   * see fd_common.h add_drive() copy_drives() to get help
   */
  bEventVssPrepareSnapshot              = 22,
  bEventOptionPlugin                    = 23,
  bEventHandleBackupFile                = 24 /* Used with Options Plugin */

} bEventType;

\end{verbatim}
 
Most of the above are self-explanatory.

\begin{description}
 \item [bEventJobStart] is called whenever a Job starts. The value
   passed is a pointer to a string that contains: "Jobid=nnn 
   Job=job-name". Where nnn will be replaced by the JobId and job-name
   will be replaced by the Job name. The variable is temporary so if you
   need the values, you must copy them.

 \item [bEventJobEnd] is called whenever a Job ends. No value is passed.

 \item [bEventStartBackupJob] is called when a Backup Job begins. No value
   is passed.

 \item [bEventEndBackupJob] is called when a Backup Job ends. No value is 
   passed.

 \item [bEventStartRestoreJob] is called when a Restore Job starts. No value
   is passed.

 \item [bEventEndRestoreJob] is called when a Restore Job ends. No value is
   passed.

 \item [bEventStartVerifyJob] is called when a Verify Job starts. No value
   is passed.

 \item [bEventEndVerifyJob] is called when a Verify Job ends. No value
   is passed.

 \item [bEventBackupCommand] is called prior to the bEventStartBackupJob and
   the plugin is passed the command string (everything after the equal sign
   in "Plugin =" as the value.

   Note, if you intend to backup a file, this is an important first point to
   write code that copies the command string passed into your pContext area
   so that you will know that a backup is being performed and you will know
   the full contents of the "Plugin =" command (i.e. what to backup and
   what virtual filename the user wants to call it.

 \item [bEventRestoreCommand] is called prior to the bEventStartRestoreJob and
   the plugin is passed the command string (everything after the equal sign
   in "Plugin =" as the value.

   See the notes above concerning backup and the command string. This is the
   point at which Bacula passes you the original command string that was
   specified during the backup, so you will want to save it in your pContext
   area for later use when Bacula calls the plugin again.

 \item [bEventLevel] is called when the level is set for a new Job. The value
   is a 32 bit integer stored in the void*, which represents the Job Level code.

 \item [bEventSince] is called when the since time is set for a new Job. The 
   value is a time\_t time at which the last job was run.

\item [bEventCancelCommand] is called whenever the currently
  running Job is cancelled. Be warned that this event is sent by a different
  thread.

\item [bEventVssBackupAddComponents] 

\item [bEventPluginCommand] is called for each PluginCommand present in the
  current FileSet. The event will be sent only on plugin specifed in the
  command. The argument is the PluginCommand (not valid after the call). 

\item [bEventHandleBackupFile] is called for each file of a FileSet when
  using a Options Plugin. If the plugin returns CF_OK, it will be used for the
  backup, if it returns CF_SKIP, the file will be skipped. Anything else will
  backup the file with Bacula core functions.
\end{description}

During each of the above calls, the plugin receives either no specific value or
only one value, which in some cases may not be sufficient.  However, knowing
the context of the event, the plugin can call back to the Bacula entry points
it was passed during the {\bf loadPlugin} call and get to a number of Bacula
variables.  (at the current time few Bacula variables are implemented, but it
easily extended at a future time and as needs require).

\subsection{startBackupFile(bpContext *ctx, struct save\_pkt *sp)}
This entry point is called only if your plugin is a command plugin, and 
it is called when Bacula encounters the "Plugin = " directive in
the Include section of the FileSet.
Called when beginning the backup of a file. Here Bacula provides you
with a pointer to the {\bf save\_pkt} structure and you must fill in 
this packet with the "attribute" data of the file.

\begin{verbatim}
struct save_pkt {
   int32_t pkt_size;                  /* size of this packet */
   char *fname;                       /* Full path and filename */
   char *link;                        /* Link name if any */
   struct stat statp;                 /* System stat() packet for file */
   int32_t type;                      /* FT_xx for this file */
   uint32_t flags;                    /* Bacula internal flags */
   bool portable;                     /* set if data format is portable */
   char *cmd;                         /* command */
   uint32_t delta_seq;                /* Delta sequence number */
   char *object_name;                 /* Object name to create */
   char *object;                      /* restore object data to save */
   int32_t object_len;                /* restore object length */
   int32_t index;                     /* restore object index */
   int32_t pkt_end;                   /* end packet sentinel */
};
\end{verbatim}

The second argument is a pointer to the {\bf save\_pkt} structure for the file
to be backed up.  The plugin is responsible for filling in all the fields 
of the {\bf save\_pkt}. If you are backing up
a real file, then generally, the statp structure can be filled in by doing
a {\bf stat} system call on the file.  

If you are backing up a database or
something that is more complex, you might want to create a virtual file.
That is a file that does not actually exist on the filesystem, but represents 
say an object that you are backing up.  In that case, you need to ensure
that the {\bf fname} string that you pass back is unique so that it
does not conflict with a real file on the system, and you need to 
artifically create values in the statp packet.

Example programs such as {\bf bpipe-fd.c} show how to set these fields.  You
must take care not to store pointers the stack in the pointer fields such as
fname and link, because when you return from your function, your stack entries
will be destroyed. The solution in that case is to malloc() and return the
pointer to it. In order to not have memory leaks, you should store a pointer to
all memory allocated in your pContext structure so that in subsequent calls or
at termination, you can release it back to the system.

Once the backup has begun, Bacula will call your plugin at the {\bf pluginIO}
entry point to "read" the data to be backed up.  Please see the {\bf bpipe-fd.c}
plugin for how to do I/O.

Example of filling in the save\_pkt as used in bpipe-fd.c:

\begin{verbatim}
   struct plugin_ctx *p_ctx = (struct plugin_ctx *)ctx->pContext;
   time_t now = time(NULL);
   sp->fname = p_ctx->fname;
   sp->statp.st_mode = 0700 | S_IFREG;
   sp->statp.st_ctime = now;
   sp->statp.st_mtime = now;
   sp->statp.st_atime = now;
   sp->statp.st_size = -1;
   sp->statp.st_blksize = 4096;
   sp->statp.st_blocks = 1;
   p_ctx->backup = true;
   return bRC_OK; 
\end{verbatim}

Note: the filename to be created has already been created from the 
command string previously sent to the plugin and is in the plugin 
context (p\_ctx->fname) and is a malloc()ed string.  This example
creates a regular file (S\_IFREG), with various fields being created.

In general, the sequence of commands issued from Bacula to the plugin
to do a backup while processing the "Plugin = " directive are:

\begin{enumerate}
 \item generate a bEventBackupCommand event to the specified plugin
       and pass it the command string.
 \item make a startPluginBackup call to the plugin, which
       fills in the data needed in save\_pkt to save as the file
       attributes and to put on the Volume and in the catalog.
 \item call Bacula's internal save\_file() subroutine to save the specified
       file.  The plugin will then be called at pluginIO() to "open"
       the file, and then to read the file data.
       Note, if you are dealing with a virtual file, the "open" operation
       is something the plugin does internally and it doesn't necessarily
       mean opening a file on the filesystem.  For example in the case of
       the bpipe-fd.c program, it initiates a pipe to the requested program.
       Finally when the plugin signals to Bacula that all the data was read,
       Bacula will call the plugin with the "close" pluginIO() function.
\end{enumerate}


\subsection{endBackupFile(bpContext *ctx)}
Called at the end of backing up a file for a command plugin.  If the plugin's
work is done, it should return bRC\_OK.  If the plugin wishes to create another
file and back it up, then it must return bRC\_More (not yet implemented).  This
is probably a good time to release any malloc()ed memory you used to pass back
filenames.

\subsection{startRestoreFile(bpContext *ctx, const char *cmd)}
Called when the first record is read from the Volume that was 
previously written by the command plugin.

\subsection{createFile(bpContext *ctx, struct restore\_pkt *rp)}
Called for a command plugin to create a file during a Restore job before 
restoring the data. 
This entry point is called before any I/O is done on the file.  After
this call, Bacula will call pluginIO() to open the file for write.

The data in the 
restore\_pkt is passed to the plugin and is based on the data that was
originally given by the plugin during the backup and the current user
restore settings (e.g. where, RegexWhere, replace).  This allows the
plugin to first create a file (if necessary) so that the data can
be transmitted to it.  The next call to the plugin will be a
pluginIO command with a request to open the file write-only.

This call must return one of the following values:

\begin{verbatim}
 enum {
   CF_SKIP = 1,       /* skip file (not newer or something) */
   CF_ERROR,          /* error creating file */
   CF_EXTRACT,        /* file created, data to extract */
   CF_CREATED,        /* file created, no data to extract */
   CF_CORE            /* let bacula core handles the file creation */
};
\end{verbatim}

in the restore\_pkt value {\bf create\_status}.  For a normal file,
unless there is an error, you must return {\bf CF\_EXTRACT}.

\begin{verbatim}
 
struct restore_pkt {
   int32_t pkt_size;                  /* size of this packet */
   int32_t stream;                    /* attribute stream id */
   int32_t data_stream;               /* id of data stream to follow */
   int32_t type;                      /* file type FT */
   int32_t file_index;                /* file index */
   int32_t LinkFI;                    /* file index to data if hard link */
   uid_t uid;                         /* userid */
   struct stat statp;                 /* decoded stat packet */
   const char *attrEx;                /* extended attributes if any */
   const char *ofname;                /* output filename */
   const char *olname;                /* output link name */
   const char *where;                 /* where */
   const char *RegexWhere;            /* regex where */
   int replace;                       /* replace flag */
   int create_status;                 /* status from createFile() */
   int32_t pkt_end;                   /* end packet sentinel */

};
\end{verbatim}

Typical code to create a regular file would be the following:

\begin{verbatim}
   struct plugin_ctx *p_ctx = (struct plugin_ctx *)ctx->pContext;
   time_t now = time(NULL);
   sp->fname = p_ctx->fname;   /* set the full path/filename I want to create */
   sp->type = FT_REG;
   sp->statp.st_mode = 0700 | S_IFREG;
   sp->statp.st_ctime = now;
   sp->statp.st_mtime = now;
   sp->statp.st_atime = now;
   sp->statp.st_size = -1;
   sp->statp.st_blksize = 4096;
   sp->statp.st_blocks = 1;
   return bRC_OK;
\end{verbatim}

This will create a virtual file.  If you are creating a file that actually 
exists, you will most likely want to fill the statp packet using the
stat() system call.

Creating a directory is similar, but requires a few extra steps:

\begin{verbatim}
   struct plugin_ctx *p_ctx = (struct plugin_ctx *)ctx->pContext;
   time_t now = time(NULL);
   sp->fname = p_ctx->fname;   /* set the full path I want to create */
   sp->link = xxx; where xxx is p_ctx->fname with a trailing forward slash
   sp->type = FT_DIREND
   sp->statp.st_mode = 0700 | S_IFDIR;
   sp->statp.st_ctime = now;
   sp->statp.st_mtime = now;
   sp->statp.st_atime = now;
   sp->statp.st_size = -1;
   sp->statp.st_blksize = 4096;
   sp->statp.st_blocks = 1;
   return bRC_OK;
\end{verbatim}

The link field must be set with the full cononical path name, which always 
ends with a forward slash.  If you do not terminate it with a forward slash,
you will surely have problems later.

As with the example that creates a file, if you are backing up a real
directory, you will want to do an stat() on the directory.  

Note, if you want the directory permissions and times to be correctly
restored, you must create the directory {\bf after} all the file directories
have been sent to Bacula. That allows the restore process to restore all the
files in a directory using default directory options, then at the end, restore
the directory permissions.  If you do it the other way around, each time you
restore a file, the OS will modify the time values for the directory entry.

\subsection{setFileAttributes(bpContext *ctx, struct restore\_pkt *rp)}
This is call not yet implemented.  Called for a command plugin.

See the definition of {\bf restre\_pkt} in the above section.

\subsection{endRestoreFile(bpContext *ctx)}
Called when a command plugin is done restoring a file.

\subsection{pluginIO(bpContext *ctx, struct io\_pkt *io)}
Called to do the input (backup) or output (restore) of data from or to a file
for a command plugin. These routines simulate the Unix read(), write(), open(),
close(), and lseek() I/O calls, and the arguments are passed in the packet and
the return values are also placed in the packet.  In addition for Win32 systems
the plugin must return two additional values (described below).

\begin{verbatim}
 enum {
   IO_OPEN = 1,
   IO_READ = 2,
   IO_WRITE = 3,
   IO_CLOSE = 4,
   IO_SEEK = 5
};

struct io_pkt {
   int32_t pkt_size;                  /* Size of this packet */
   int32_t func;                      /* Function code */
   int32_t count;                     /* read/write count */
   mode_t mode;                       /* permissions for created files */
   int32_t flags;                     /* Open flags */
   char *buf;                         /* read/write buffer */
   const char *fname;                 /* open filename */
   int32_t status;                    /* return status */
   int32_t io_errno;                  /* errno code */
   int32_t lerror;                    /* Win32 error code */
   int32_t whence;                    /* lseek argument */
   boffset_t offset;                  /* lseek argument */
   bool win32;                        /* Win32 GetLastError returned */
   int32_t pkt_end;                   /* end packet sentinel */
};
\end{verbatim}

The particular Unix function being simulated is indicated by the {\bf func},
which will have one of the IO\_OPEN, IO\_READ, ... codes listed above.  The
status code that would be returned from a Unix call is returned in {\bf status}
for IO\_OPEN, IO\_CLOSE, IO\_READ, and IO\_WRITE. The return value for IO\_SEEK
is returned in {\bf offset} which in general is a 64 bit value.

When there is an error on Unix systems, you must always set io\_error, and
on a Win32 system, you must always set win32, and the returned value from
the OS call GetLastError() in lerror.

For all except IO\_SEEK, {\bf status} is the return result.  In general it is
a positive integer unless there is an error in which case it is -1.

The following describes each call and what you get and what you
should return:

\begin{description}
 \item [IO\_OPEN]
   You will be passed fname, mode, and flags.
   You must set on return: status, and if there is a Unix error
   io\_errno must be set to the errno value, and if there is a 
   Win32 error win32 and lerror. 

 \item [IO\_READ]
  You will be passed: count, and buf (buffer of size count).
  You must set on return: status to the number of bytes 
  read into the buffer (buf) or -1 on an error, 
  and if there is a Unix error
  io\_errno must be set to the errno value, and if there is a
  Win32 error, win32 and lerror must be set.

 \item [IO\_WRITE]
  You will be passed: count, and buf (buffer of size count).
  You must set on return: status to the number of bytes 
  written from the buffer (buf) or -1 on an error, 
  and if there is a Unix error
  io\_errno must be set to the errno value, and if there is a
  Win32 error, win32 and lerror must be set.

 \item [IO\_CLOSE]
  Nothing will be passed to you.  On return you must set 
  status to 0 on success and -1 on failure.  If there is a Unix error
  io\_errno must be set to the errno value, and if there is a
  Win32 error, win32 and lerror must be set.

 \item [IO\_LSEEK]
  You will be passed: offset, and whence. offset is a 64 bit value
  and is the position to seek to relative to whence.  whence is one
  of the following SEEK\_SET, SEEK\_CUR, or SEEK\_END indicating to
  either to seek to an absolute possition, relative to the current 
  position or relative to the end of the file.
  You must pass back in offset the absolute location to which you 
  seeked. If there is an error, offset should be set to -1.
  If there is a Unix error
  io\_errno must be set to the errno value, and if there is a
  Win32 error, win32 and lerror must be set.

  Note: Bacula will call IO\_SEEK only when writing a sparse file.
  
\end{description}

\subsection{bool checkFile(bpContext *ctx, char *fname)}
If this entry point is set, Bacula will call it after backing up all file
data during an Accurate backup.  It will be passed the full filename for
each file that Bacula is proposing to mark as deleted.  Only files
previously backed up but not backed up in the current session will be
marked to be deleted.  If you return {\bf false}, the file will be be
marked deleted.  If you return {\bf true} the file will not be marked
deleted.  This permits a plugin to ensure that previously saved virtual
files or files controlled by your plugin that have not change (not backed
up in the current job) are not marked to be deleted.  This entry point will
only be called during Accurate Incrmental and Differential backup jobs.


\section{Bacula Plugin Entrypoints}
When Bacula calls one of your plugin entrypoints, you can call back to
the entrypoints in Bacula that were supplied during the xxx plugin call
to get or set information within Bacula.

\subsection{bRC registerBaculaEvents(bpContext *ctx, ...)}
This Bacula entrypoint will allow you to register to receive events
that are not autmatically passed to your plugin by default. This 
entrypoint currently is unimplemented.

\subsection{bRC getBaculaValue(bpContext *ctx, bVariable var, void *value)}
Calling this entrypoint, you can obtain specific values that are available
in Bacula. The following Variables can be referenced:
\begin{itemize}
\item bVarJobId returns an int
\item bVarFDName returns a char *
\item bVarLevel returns an int
\item bVarClient returns a char *
\item bVarJobName returns a char *
\item bVarJobStatus returns an int
\item bVarSinceTime returns an int (time\_t)
\item bVarAccurate returns an int
\end{itemize}

\subsection{bRC setBaculaValue(bpContext *ctx, bVariable var, void *value)}
Calling this entrypoint allows you to set particular values in
Bacula. The only variable that can currently be set is 
{\bf bVarFileSeen} and the value passed is a char * that points
to the full filename for a file that you are indicating has been
seen and hence is not deleted.

\subsection{bRC JobMessage(bpContext *ctx, const char *file, int line,
       int type, utime\_t mtime, const char *fmt, ...)}
This call permits you to put a message in the Job Report.


\subsection{bRC DebugMessage(bpContext *ctx, const char *file, int line,
       int level, const char *fmt, ...)}
This call permits you to print a debug message.


\subsection{void baculaMalloc(bpContext *ctx, const char *file, int line,
       size\_t size)}
This call permits you to obtain memory from Bacula's memory allocator.


\subsection{void baculaFree(bpContext *ctx, const char *file, int line, void *mem)}
This call permits you to free memory obtained from Bacula's memory allocator.

\section{Building Bacula Plugins}
There is currently one sample program {\bf example-plugin-fd.c} and
one working plugin {\bf bpipe-fd.c} that can be found in the Bacula
{\bf src/plugins/fd} directory.  Both are built with the following:

\begin{verbatim}
 cd <bacula-source>
 ./configure <your-options>
 make
 ...
 cd src/plugins/fd
 make
 make test
\end{verbatim}

After building Bacula and changing into the src/plugins/fd directory,
the {\bf make} command will build the {\bf bpipe-fd.so} plugin, which 
is a very useful and working program.

The {\bf make test} command will build the {\bf example-plugin-fd.so}
plugin and a binary named {\bf main}, which is build from the source
code located in {\bf src/filed/fd\_plugins.c}. 

If you execute {\bf ./main}, it will load and run the example-plugin-fd
plugin simulating a small number of the calling sequences that Bacula uses
in calling a real plugin.  This allows you to do initial testing of 
your plugin prior to trying it with Bacula.

You can get a good idea of how to write your own plugin by first 
studying the example-plugin-fd, and actually running it.  Then
it can also be instructive to read the bpipe-fd.c code as it is 
a real plugin, which is still rather simple and small.

When actually writing your own plugin, you may use the example-plugin-fd.c
code as a template for your code.