freebsd-nq/usr.sbin/rpc.statd/file.c
Doug Rabson c675522fc4 Re-implement the client side of rpc.lockd in the kernel. This implementation
provides the correct semantics for flock(2) style locks which are used by the
lockf(1) command line tool and the pidfile(3) library. It also implements
recovery from server restarts and ensures that dirty cache blocks are written
to the server before obtaining locks (allowing multiple clients to use file
locking to safely share data).

Sponsored by:	Isilon Systems
PR:		94256
MFC after:	2 weeks
2008-06-26 10:21:54 +00:00

362 lines
11 KiB
C

/*
* Copyright (c) 1995
* A.R. Gordon (andrew.gordon@net-tel.co.uk). All rights reserved.
*
* Redistribution and use in source and binary forms, with or without
* modification, are permitted provided that the following conditions
* are met:
* 1. Redistributions of source code must retain the above copyright
* notice, this list of conditions and the following disclaimer.
* 2. Redistributions in binary form must reproduce the above copyright
* notice, this list of conditions and the following disclaimer in the
* documentation and/or other materials provided with the distribution.
* 3. All advertising materials mentioning features or use of this software
* must display the following acknowledgement:
* This product includes software developed for the FreeBSD project
* 4. Neither the name of the author nor the names of any co-contributors
* may be used to endorse or promote products derived from this software
* without specific prior written permission.
*
* THIS SOFTWARE IS PROVIDED BY ANDREW GORDON AND CONTRIBUTORS ``AS IS'' AND
* ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
* IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
* ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
* FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
* DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
* OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
* HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
* LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
* OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
* SUCH DAMAGE.
*
* $FreeBSD$
*
*/
#include <err.h>
#include <errno.h>
#include <fcntl.h>
#include <netdb.h>
#include <stdio.h>
#include <string.h>
#include <unistd.h>
#include <sys/types.h>
#include <sys/mman.h> /* For mmap() */
#include <rpc/rpc.h>
#include <syslog.h>
#include <stdlib.h>
#include "statd.h"
FileLayout *status_info; /* Pointer to the mmap()ed status file */
static int status_fd; /* File descriptor for the open file */
static off_t status_file_len; /* Current on-disc length of file */
/* sync_file --------------------------------------------------------------- */
/*
Purpose: Packaged call of msync() to flush changes to mmap()ed file
Returns: Nothing. Errors to syslog.
*/
void sync_file(void)
{
if (msync((void *)status_info, 0, 0) < 0)
{
syslog(LOG_ERR, "msync() failed: %s", strerror(errno));
}
}
/* find_host -------------------------------------------------------------- */
/*
Purpose: Find the entry in the status file for a given host
Returns: Pointer to that entry in the mmap() region, or NULL.
Notes: Also creates entries if requested.
Failure to create also returns NULL.
*/
HostInfo *find_host(char *hostname, int create)
{
HostInfo *hp;
HostInfo *spare_slot = NULL;
HostInfo *result = NULL;
struct addrinfo *ai1, *ai2;
int i;
if (getaddrinfo(hostname, NULL, NULL, &ai1) != 0)
ai1 = NULL;
for (i = 0, hp = status_info->hosts; i < status_info->noOfHosts; i++, hp++)
{
if (!strncasecmp(hostname, hp->hostname, SM_MAXSTRLEN))
{
result = hp;
break;
}
if (hp->hostname[0] &&
getaddrinfo(hp->hostname, NULL, NULL, &ai2) != 0)
ai2 = NULL;
if (ai1 && ai2)
{
struct addrinfo *p1, *p2;
for (p1 = ai1; !result && p1; p1 = p1->ai_next)
{
for (p2 = ai2; !result && p2; p2 = p2->ai_next)
{
if (p1->ai_family == p2->ai_family
&& p1->ai_addrlen == p2->ai_addrlen
&& !memcmp(p1->ai_addr, p2->ai_addr, p1->ai_addrlen))
{
result = hp;
break;
}
}
}
if (result)
break;
}
if (ai2)
freeaddrinfo(ai2);
if (!spare_slot && !hp->monList && !hp->notifyReqd)
spare_slot = hp;
}
if (ai1)
freeaddrinfo(ai1);
/* Return if entry found, or if not asked to create one. */
if (result || !create) return (result);
/* Now create an entry, using the spare slot if one was found or */
/* adding to the end of the list otherwise, extending file if reqd */
if (!spare_slot)
{
off_t desired_size;
spare_slot = &status_info->hosts[status_info->noOfHosts];
desired_size = ((char*)spare_slot - (char*)status_info) + sizeof(HostInfo);
if (desired_size > status_file_len)
{
/* Extend file by writing 1 byte of junk at the desired end pos */
lseek(status_fd, desired_size - 1, SEEK_SET);
i = write(status_fd, &i, 1);
if (i < 1)
{
syslog(LOG_ERR, "Unable to extend status file");
return (NULL);
}
status_file_len = desired_size;
}
status_info->noOfHosts++;
}
/* Initialise the spare slot that has been found/created */
/* Note that we do not msync(), since the caller is presumed to be */
/* about to modify the entry further */
memset(spare_slot, 0, sizeof(HostInfo));
strncpy(spare_slot->hostname, hostname, SM_MAXSTRLEN);
return (spare_slot);
}
/* init_file -------------------------------------------------------------- */
/*
Purpose: Open file, create if necessary, initialise it.
Returns: Nothing - exits on error
Notes: Called before process becomes daemon, hence logs to
stderr rather than syslog.
Opens the file, then mmap()s it for ease of access.
Also performs initial clean-up of the file, zeroing
monitor list pointers, setting the notifyReqd flag in
all hosts that had a monitor list, and incrementing
the state number to the next even value.
*/
void init_file(const char *filename)
{
int new_file = FALSE;
char buf[HEADER_LEN];
int i;
/* try to open existing file - if not present, create one */
status_fd = open(filename, O_RDWR);
if ((status_fd < 0) && (errno == ENOENT))
{
status_fd = open(filename, O_RDWR | O_CREAT, 0644);
new_file = TRUE;
}
if (status_fd < 0)
errx(1, "unable to open status file %s", filename);
/* File now open. mmap() it, with a generous size to allow for */
/* later growth, where we will extend the file but not re-map it. */
status_info = (FileLayout *)
mmap(NULL, 0x10000000, PROT_READ | PROT_WRITE, MAP_SHARED, status_fd, 0);
if (status_info == (FileLayout *) MAP_FAILED)
err(1, "unable to mmap() status file");
status_file_len = lseek(status_fd, 0L, SEEK_END);
/* If the file was not newly created, validate the contents, and if */
/* defective, re-create from scratch. */
if (!new_file)
{
if ((status_file_len < HEADER_LEN) || (status_file_len
< (HEADER_LEN + sizeof(HostInfo) * status_info->noOfHosts)) )
{
warnx("status file is corrupt");
new_file = TRUE;
}
}
/* Initialisation of a new, empty file. */
if (new_file)
{
memset(buf, 0, sizeof(buf));
lseek(status_fd, 0L, SEEK_SET);
write(status_fd, buf, HEADER_LEN);
status_file_len = HEADER_LEN;
}
else
{
/* Clean-up of existing file - monitored hosts will have a pointer */
/* to a list of clients, which refers to memory in the previous */
/* incarnation of the program and so are meaningless now. These */
/* pointers are zeroed and the fact that the host was previously */
/* monitored is recorded by setting the notifyReqd flag, which will */
/* in due course cause a SM_NOTIFY to be sent. */
/* Note that if we crash twice in quick succession, some hosts may */
/* already have notifyReqd set, where we didn't manage to notify */
/* them before the second crash occurred. */
for (i = 0; i < status_info->noOfHosts; i++)
{
HostInfo *this_host = &status_info->hosts[i];
if (this_host->monList)
{
this_host->notifyReqd = TRUE;
this_host->monList = NULL;
}
}
/* Select the next higher even number for the state counter */
status_info->ourState = (status_info->ourState + 2) & 0xfffffffe;
/*???????******/ status_info->ourState++;
}
}
/* notify_one_host --------------------------------------------------------- */
/*
Purpose: Perform SM_NOTIFY procedure at specified host
Returns: TRUE if success, FALSE if failed.
*/
static int notify_one_host(char *hostname)
{
struct timeval timeout = { 20, 0 }; /* 20 secs timeout */
CLIENT *cli;
char dummy;
stat_chge arg;
char our_hostname[SM_MAXSTRLEN+1];
gethostname(our_hostname, sizeof(our_hostname));
our_hostname[SM_MAXSTRLEN] = '\0';
arg.mon_name = our_hostname;
arg.state = status_info->ourState;
if (debug) syslog (LOG_DEBUG, "Sending SM_NOTIFY to host %s from %s", hostname, our_hostname);
cli = clnt_create(hostname, SM_PROG, SM_VERS, "udp");
if (!cli)
{
syslog(LOG_ERR, "Failed to contact host %s%s", hostname,
clnt_spcreateerror(""));
return (FALSE);
}
if (clnt_call(cli, SM_NOTIFY, (xdrproc_t)xdr_stat_chge, &arg,
(xdrproc_t)xdr_void, &dummy, timeout)
!= RPC_SUCCESS)
{
syslog(LOG_ERR, "Failed to contact rpc.statd at host %s", hostname);
clnt_destroy(cli);
return (FALSE);
}
clnt_destroy(cli);
return (TRUE);
}
/* notify_hosts ------------------------------------------------------------ */
/*
Purpose: Send SM_NOTIFY to all hosts marked as requiring it
Returns: Nothing, immediately - forks a process to do the work.
Notes: Does nothing if there are no monitored hosts.
Called after all the initialisation has been done -
logs to syslog.
*/
void notify_hosts(void)
{
int i;
int attempts;
int work_to_do = FALSE;
HostInfo *hp;
pid_t pid;
/* First check if there is in fact any work to do. */
for (i = status_info->noOfHosts, hp = status_info->hosts; i ; i--, hp++)
{
if (hp->notifyReqd)
{
work_to_do = TRUE;
break;
}
}
if (!work_to_do) return; /* No work found */
pid = fork();
if (pid == -1)
{
syslog(LOG_ERR, "Unable to fork notify process - %s", strerror(errno));
return;
}
if (pid) return;
/* Here in the child process. We continue until all the hosts marked */
/* as requiring notification have been duly notified. */
/* If one of the initial attempts fails, we sleep for a while and */
/* have another go. This is necessary because when we have crashed, */
/* (eg. a power outage) it is quite possible that we won't be able to */
/* contact all monitored hosts immediately on restart, either because */
/* they crashed too and take longer to come up (in which case the */
/* notification isn't really required), or more importantly if some */
/* router etc. needed to reach the monitored host has not come back */
/* up yet. In this case, we will be a bit late in re-establishing */
/* locks (after the grace period) but that is the best we can do. */
/* We try 10 times at 5 sec intervals, 10 more times at 1 minute */
/* intervals, then 24 more times at hourly intervals, finally */
/* giving up altogether if the host hasn't come back to life after */
/* 24 hours. */
for (attempts = 0; attempts < 44; attempts++)
{
work_to_do = FALSE; /* Unless anything fails */
for (i = status_info->noOfHosts, hp = status_info->hosts; i ; i--, hp++)
{
if (hp->notifyReqd)
{
if (notify_one_host(hp->hostname))
{
hp->notifyReqd = FALSE;
sync_file();
}
else work_to_do = TRUE;
}
}
if (!work_to_do) break;
if (attempts < 10) sleep(5);
else if (attempts < 20) sleep(60);
else sleep(60*60);
}
exit(0);
}