#ifndef lint
static const char rcsid[] =
- "$Id: tape.c,v 1.80 2004/01/27 10:37:29 stelian Exp $";
+ "$Id: tape.c,v 1.94 2011/06/10 12:41:54 stelian Exp $";
#endif /* not lint */
#include <config.h>
#ifdef __linux__
#include <sys/types.h>
#include <sys/time.h>
-#include <linux/fs.h> /* for definition of BLKFLSBUF */
+#include <sys/ioctl.h>
+#include <sys/mount.h> /* for definition of BLKFLSBUF */
+#ifndef BLKFLSBUF /* last resort... */
+#define BLKFLSBUF _IO(0x12, 97) /* Flush buffer cache. */
+#endif
#include <time.h>
#endif
#include <sys/param.h>
long long tapea_bytes = 0; /* bytes_written at start of current volume */
static int magtapeout; /* output is really a tape */
-static ssize_t dump_atomic_read __P((int, void *, size_t));
-static ssize_t dump_atomic_write __P((int, const void *, size_t));
+static ssize_t dump_atomic_read __P((int, char *, size_t));
+static ssize_t dump_atomic_write __P((int, const char *, size_t));
#ifdef WRITEDEBUG
static void doslave __P((int, int, int));
#else
* The following structure defines the instruction packets sent to slaves.
*/
struct req {
- daddr_t dblk;
+ ext2_loff_t dblk;
int count;
};
int reqsiz;
int master; /* pid of master, for sending error signals */
int tenths; /* length of tape overhead per block written */
-static int caught; /* have we caught the signal to proceed? */
-static int ready; /* have we reached the lock point without having */
+static int caught1; /* have we caught the signal to proceed? */
+static int ready1; /* have we reached the lock point without having */
+ /* received the SIGUSR2 signal from the prev slave? */
+static sigjmp_buf jmpbuf1; /* where to jump to if we are ready when the */
+ /* SIGUSR1 arrives from the previous slave */
+static int caught2; /* have we caught the signal to proceed? */
+static int ready2; /* have we reached the lock point without having */
/* received the SIGUSR2 signal from the prev slave? */
-static sigjmp_buf jmpbuf; /* where to jump to if we are ready when the */
+static sigjmp_buf jmpbuf2; /* where to jump to if we are ready when the */
/* SIGUSR2 arrives from the previous slave */
#ifdef USE_QFA
static int gtperr = 0;
#endif
+/*
+ * Determine if we can use Linux' clone system call. If so, call it
+ * with the CLONE_IO flag so that all processes will share the same I/O
+ * context, allowing the I/O schedulers to make better scheduling decisions.
+ */
+#ifdef __linux__
+/* first, pull in the header files that define sys_clone and CLONE_IO */
+#include <syscall.h>
+#define _GNU_SOURCE
+#include <sched.h>
+#include <unistd.h>
+#undef _GNU_SOURCE
+
+/* If either is not present, fall back on the fork behaviour */
+#if ! defined(SYS_clone) || ! defined (CLONE_IO)
+#define fork_clone_io fork
+#else /* SYS_clone */
+/* CLONE_IO is available, determine which version of sys_clone to use */
+#include <linux/version.h>
+/*
+ * Kernel 2.5.49 introduced two extra parameters to the clone system call.
+ * Neither is useful in our case, so this is easy to handle.
+ */
+#if LINUX_VERSION_CODE >= KERNEL_VERSION(2,5,49)
+/*
+ * Parameters of the sys_clone syscall are
+ * clone_flags, child_stack, parent_tidptr, child_tidptr
+ * on all architectures except s390 and s390x
+ * s390* have child_stack, clone_flags, parent_tidptr, child_tidptr
+ */
+#if defined(__s390__) || defined(__s390x__)
+#define CLONE_ARGS 0, SIGCHLD|CLONE_IO, NULL, NULL
+#else
+#define CLONE_ARGS SIGCHLD|CLONE_IO, 0, NULL, NULL
+#endif
+#else
+#define CLONE_ARGS SIGCHLD|CLONE_IO, 0
+#endif /* LINUX_VERSION_CODE */
+pid_t fork_clone_io(void);
+#endif /* SYS_clone */
+#else /* __linux__ not defined */
+#define fork_clone_io fork
+#endif /* __linux__ */
+
int
alloctape(void)
{
writerec(const void *dp, int isspcl)
{
- slp->req[trecno].dblk = (daddr_t)0;
+ slp->req[trecno].dblk = (ext2_loff_t)0;
slp->req[trecno].count = 1;
/* XXX post increment triggers an egcs-1.1.2-12 bug on alpha/sparc */
*(union u_spcl *)(*(nextblock)) = *(union u_spcl *)dp;
/* Need to write it to the archive file */
if (! AfileActive && isspcl && (spcl.c_type == TS_END))
AfileActive = 1;
- if (AfileActive && Afile >= 0) {
+ if (AfileActive && Afile >= 0 && !(spcl.c_flags & DR_EXTATTRIBUTES)) {
/* When we dump an inode which is not a directory,
* it means we ended the archive contents */
if (isspcl && (spcl.c_type == TS_INODE) &&
}
void
-dumpblock(daddr_t blkno, int size)
+dumpblock(blk_t blkno, int size)
{
- int avail, tpblks, dblkno;
+ int avail, tpblks;
+ ext2_loff_t dblkno;
dblkno = fsbtodb(sblock, blkno);
tpblks = size >> tp_bshift;
int siz = (char *)nextblock - (char *)slp->req;
+ /* make sure returned has sane values in case we don't read
+ * them from the slave in this pass */
+ returned.unclen = returned.clen = writesize;
+
slp->req[trecno].count = 0; /* Sentinel */
if (dump_atomic_write( slp->fd, (char *)slp->req, siz) != siz)
}
blks = 0;
- if (spcl.c_type != TS_END) {
- for (i = 0; i < spcl.c_count; i++)
- if (spcl.c_addr[i] != 0)
- blks++;
+ if (spcl.c_type == TS_CLRI || spcl.c_type == TS_BITS)
+ blks = spcl.c_count;
+ else {
+ if (spcl.c_type != TS_END) {
+ for (i = 0; i < spcl.c_count; i++)
+ if (spcl.c_addr[i] != 0)
+ blks++;
+ }
}
slp->count = lastspclrec + blks + 1 - spcl.c_tapea;
slp->tapea = spcl.c_tapea;
blockswritten += ntrec;
blocksthisvol += ntrec;
if (!pipeout && !unlimited) {
- if (blocksperfiles[blocksperfiles_current]) {
+ if (blocksperfiles && blocksperfiles[blocksperfiles_current]) {
if ( compressed ? (bytes_written - tapea_bytes + SLAVES * (writesize + sizeof(struct tapebuf))) >= (((long long)blocksperfiles[blocksperfiles_current]) * 1024)
: blocksthisvol >= blocksperfiles[blocksperfiles_current] ) {
close_rewind();
{
int eot_code = 1;
(void)trewind();
- if (nexttape || Mflag)
- return;
if (eot_script) {
msg("Launching %s\n", eot_script);
eot_code = system_command(eot_script, tape, tapeno);
}
if (eot_code == 0)
return;
+ if (nexttape || Mflag)
+ return;
if (!nogripe) {
msg("Change Volumes: Mount volume #%d\n", tapeno+1);
broadcast("CHANGE DUMP VOLUMES!\7\7\n");
tslp = &slaves[SLAVES];
ntb = (union u_spcl *)tslp->tblock[1];
+ /* make sure returned has sane values in case we don't read
+ * them from the slave in this pass */
+ returned.unclen = returned.clen = writesize;
+
/*
* Each of the N slaves should have requests that need to
* be replayed on the next tape. Use the extra slave buffers
#endif
}
+#ifdef __linux__
+#if defined(SYS_clone) && defined(CLONE_IO)
+pid_t
+fork_clone_io(void)
+{
+ return syscall(SYS_clone, CLONE_ARGS);
+}
+#endif
+#endif
+
/*
* We implement taking and restoring checkpoints on the tape level.
* When each tape is opened, a new process is created by forking; this
/*
* All signals are inherited...
*/
- childpid = fork();
+ childpid = fork_clone_io();
if (childpid < 0) {
msg("Context save fork fails in parent %d\n", parentpid);
Exit(X_ABORT);
tape[MAXPATHLEN - 1] = '\0';
msg("Dumping volume %d on %s\n", tapeno, tape);
}
- if (blocksperfiles_current < *blocksperfiles)
+ if (blocksperfiles && blocksperfiles_current < *blocksperfiles)
blocksperfiles_current++;
#ifdef RDUMP
while ((tapefd = (host ? rmtopen(tape, O_WRONLY|O_CREAT|O_TRUNC) : pipeout ?
exit(status);
}
+/*
+ * proceed - handler for SIGUSR1, used to synchronize IO between the slaves.
+ */
+static void
+proceed1(UNUSED(int signo))
+{
+ if (ready1)
+ siglongjmp(jmpbuf1, 1);
+ caught1++;
+}
+
/*
* proceed - handler for SIGUSR2, used to synchronize IO between the slaves.
*/
static void
-proceed(UNUSED(int signo))
+proceed2(UNUSED(int signo))
{
- if (ready)
- siglongjmp(jmpbuf, 1);
- caught++;
+ if (ready2)
+ siglongjmp(jmpbuf2, 1);
+ caught2++;
}
void
sigaction(SIGTERM, &sa, NULL); /* Slave sends SIGTERM on dumpabort() */
sa.sa_handler = sigpipe;
sigaction(SIGPIPE, &sa, NULL);
- sa.sa_handler = proceed;
+ sa.sa_handler = proceed1;
+ sa.sa_flags = SA_RESTART;
+ sigaction(SIGUSR1, &sa, NULL); /* Slave sends SIGUSR1 to next slave */
+ sa.sa_handler = proceed2;
sa.sa_flags = SA_RESTART;
sigaction(SIGUSR2, &sa, NULL); /* Slave sends SIGUSR2 to next slave */
}
for (i = 0; i < SLAVES; i++) {
if (i == slp - &slaves[0]) {
- caught = 1;
+ caught1 = 1;
+ caught2 = 1;
} else {
- caught = 0;
+ caught1 = 0;
+ caught2 = 0;
}
if (socketpair(AF_UNIX, SOCK_STREAM, 0, cmd) < 0 ||
- (slaves[i].pid = fork()) < 0)
+ (slaves[i].pid = fork_clone_io()) < 0)
quit("too many slaves, %d (recompile smaller): %s\n",
i, strerror(errno));
* previous process before writing to the tape, and sends SIGUSR2
* to the next process when the tape write completes. On tape errors
* a SIGUSR1 is sent to the master which then terminates all of the
- * slaves.
+ * slaves. Each process sends SIGUSR1 to the next to signal that it
+ * is time to start reading from the disk, after it finishes reading
+ * and moves to the compression phase.
*/
static void
doslave(int cmd,
int compresult;
volatile int do_compress = !first;
unsigned long worklen;
-#ifdef HAVE_BZLIB
- unsigned int worklen2;
-#endif
#ifdef HAVE_LZO
lzo_align_t __LZO_MMODEL *LZO_WorkMem;
#endif
sigset_t set;
sigemptyset(&set);
+ sigaddset(&set, SIGUSR1);
sigaddset(&set, SIGUSR2);
sigprocmask(SIG_BLOCK, &set, NULL);
sigemptyset(&set);
while ((nread = dump_atomic_read( cmd, (char *)slp->req, reqsiz)) == reqsiz) {
struct req *p = slp->req;
+ /* wait for previous slave to finish reading */
+ if (sigsetjmp(jmpbuf1, 1) == 0) {
+ ready1 = 1;
+ if (!caught1)
+ sigsuspend(&set);
+ }
+ ready1 = 0;
+ caught1 = 0;
+
for (trecno = 0; trecno < ntrec;
trecno += p->count, p += p->count) {
if (p->dblk) { /* read a disk block */
quit("master/slave protocol botched.\n");
}
}
+ /* signal next slave to start reading */
+ (void) kill(nextslave, SIGUSR1);
/* Try to write the data... */
wrote = 0;
#endif /* HAVE_ZLIB */
#ifdef HAVE_BZLIB
if (zipflag == COMPRESS_BZLIB) {
- worklen2 = worklen;
+ unsigned int worklen2 = worklen;
compresult = BZ2_bzBuffToBuffCompress(
comp_buf->buf,
&worklen2,
#endif /* HAVE_BZLIB */
#ifdef HAVE_LZO
if (zipflag == COMPRESS_LZO) {
+ lzo_uint worklen2 = worklen;
compresult = lzo1x_1_compress((char *)slp->tblock[0],writesize,
comp_buf->buf,
- (lzo_uintp)&worklen,
+ &worklen2,
LZO_WorkMem);
+ worklen = worklen2;
if (compresult == LZO_E_OK)
compresult = 1;
else
do_compress = compressed;
#endif /* HAVE_ZLIB || HAVE_BZLIB || HAVE_LZO */
- if (sigsetjmp(jmpbuf, 1) == 0) {
- ready = 1;
- if (!caught)
+ if (sigsetjmp(jmpbuf2, 1) == 0) {
+ ready2 = 1;
+ if (!caught2)
sigsuspend(&set);
}
- ready = 0;
- caught = 0;
+ ready2 = 0;
+ caught2 = 0;
#ifdef USE_QFA
if (gTapeposfd >= 0) {
if ((spclptr->c_magic == NFS_MAGIC) &&
(spclptr->c_type == TS_INODE) &&
(spclptr->c_date == gThisDumpDate) &&
- !(spclptr->c_dinode.di_mode & S_IFDIR)
+ !(spclptr->c_dinode.di_mode & S_IFDIR) &&
+ !(spclptr->c_flags & DR_EXTATTRIBUTES)
) {
foundone = 1;
/* if (cntntrecs >= maxntrecs) { only write every maxntrecs amount of data */
* loop until the count is satisfied (or error).
*/
static ssize_t
-dump_atomic_read(int fd, void *buf, size_t count)
+dump_atomic_read(int fd, char *buf, size_t count)
{
int got, need = count;
do {
while ((got = read(fd, buf, need)) > 0 && (need -= got) > 0)
- (char *)buf += got;
+ buf += got;
} while (got == -1 && errno == EINTR);
return (got < 0 ? got : (ssize_t)count - need);
}
* loop until the count is satisfied (or error).
*/
static ssize_t
-dump_atomic_write(int fd, const void *buf, size_t count)
+dump_atomic_write(int fd, const char *buf, size_t count)
{
int got, need = count;
do {
while ((got = write(fd, buf, need)) > 0 && (need -= got) > 0)
- (char *)buf += got;
+ buf += got;
} while (got == -1 && errno == EINTR);
return (got < 0 ? got : (ssize_t)count - need);
}