#include <stdlib.h>
#include <string.h>
#include <zlib.h>
+#include <sys/signal.h>
#include <sys/spa.h>
#include <sys/stat.h>
#include <sys/processor.h>
#include <sys/zfs_context.h>
-#include <sys/zmod.h>
#include <sys/utsname.h>
+#include <sys/time.h>
#include <sys/systeminfo.h>
/*
* threads
* =========================================================================
*/
-/*ARGSUSED*/
+
+pthread_cond_t kthread_cond = PTHREAD_COND_INITIALIZER;
+pthread_mutex_t kthread_lock = PTHREAD_MUTEX_INITIALIZER;
+pthread_key_t kthread_key;
+int kthread_nr = 0;
+
+static void
+thread_init(void)
+{
+ kthread_t *kt;
+
+ VERIFY3S(pthread_key_create(&kthread_key, NULL), ==, 0);
+
+ /* Create entry for primary kthread */
+ kt = umem_zalloc(sizeof(kthread_t), UMEM_NOFAIL);
+ kt->t_tid = pthread_self();
+ kt->t_func = NULL;
+
+ VERIFY3S(pthread_setspecific(kthread_key, kt), ==, 0);
+
+ /* Only the main thread should be running at the moment */
+ ASSERT3S(kthread_nr, ==, 0);
+ kthread_nr = 1;
+}
+
+static void
+thread_fini(void)
+{
+ kthread_t *kt = curthread;
+
+ ASSERT(pthread_equal(kt->t_tid, pthread_self()));
+ ASSERT3P(kt->t_func, ==, NULL);
+
+ umem_free(kt, sizeof(kthread_t));
+
+ /* Wait for all threads to exit via thread_exit() */
+ VERIFY3S(pthread_mutex_lock(&kthread_lock), ==, 0);
+
+ kthread_nr--; /* Main thread is exiting */
+
+ while (kthread_nr > 0)
+ VERIFY3S(pthread_cond_wait(&kthread_cond, &kthread_lock), ==,
+ 0);
+
+ ASSERT3S(kthread_nr, ==, 0);
+ VERIFY3S(pthread_mutex_unlock(&kthread_lock), ==, 0);
+
+ VERIFY3S(pthread_key_delete(kthread_key), ==, 0);
+}
+
kthread_t *
-zk_thread_create(void (*func)(), void *arg)
+zk_thread_current(void)
+{
+ kthread_t *kt = pthread_getspecific(kthread_key);
+
+ ASSERT3P(kt, !=, NULL);
+
+ return kt;
+}
+
+void *
+zk_thread_helper(void *arg)
{
- thread_t tid;
+ kthread_t *kt = (kthread_t *) arg;
+
+ VERIFY3S(pthread_setspecific(kthread_key, kt), ==, 0);
+
+ VERIFY3S(pthread_mutex_lock(&kthread_lock), ==, 0);
+ kthread_nr++;
+ VERIFY3S(pthread_mutex_unlock(&kthread_lock), ==, 0);
+
+ kt->t_tid = pthread_self();
+ ((thread_func_arg_t) kt->t_func)(kt->t_arg);
- VERIFY(thr_create(0, 0, (void *(*)(void *))func, arg, THR_DETACHED,
- &tid) == 0);
+ /* Unreachable, thread must exit with thread_exit() */
+ abort();
- return ((void *)(uintptr_t)tid);
+ return NULL;
+}
+
+kthread_t *
+zk_thread_create(caddr_t stk, size_t stksize, thread_func_t func, void *arg,
+ size_t len, proc_t *pp, int state, pri_t pri, int detachstate)
+{
+ kthread_t *kt;
+ pthread_attr_t attr;
+ size_t stack;
+
+ ASSERT3S(state & ~TS_RUN, ==, 0);
+
+ kt = umem_zalloc(sizeof(kthread_t), UMEM_NOFAIL);
+ kt->t_func = func;
+ kt->t_arg = arg;
+
+ /*
+ * The Solaris kernel stack size is 24k for x86/x86_64.
+ * The Linux kernel stack size is 8k for x86/x86_64.
+ *
+ * We reduce the default stack size in userspace, to ensure
+ * we observe stack overruns in user space as well as in
+ * kernel space. In practice we can't set the userspace stack
+ * size to 8k because differences in stack usage between kernel
+ * space and userspace could lead to spurious stack overflows
+ * (especially when debugging is enabled). Nevertheless, we try
+ * to set it to the lowest value that works (currently 8k*4).
+ * PTHREAD_STACK_MIN is the minimum stack required for a NULL
+ * procedure in user space and is added in to the stack
+ * requirements.
+ *
+ * Some buggy NPTL threading implementations include the
+ * guard area within the stack size allocations. In
+ * this case we allocate an extra page to account for the
+ * guard area since we only have two pages of usable stack
+ * on Linux.
+ */
+
+ stack = PTHREAD_STACK_MIN + MAX(stksize, STACK_SIZE) * 4;
+
+ VERIFY3S(pthread_attr_init(&attr), ==, 0);
+ VERIFY3S(pthread_attr_setstacksize(&attr, stack), ==, 0);
+ VERIFY3S(pthread_attr_setguardsize(&attr, PAGESIZE), ==, 0);
+ VERIFY3S(pthread_attr_setdetachstate(&attr, detachstate), ==, 0);
+
+ VERIFY3S(pthread_create(&kt->t_tid, &attr, &zk_thread_helper, kt),
+ ==, 0);
+
+ VERIFY3S(pthread_attr_destroy(&attr), ==, 0);
+
+ return kt;
+}
+
+void
+zk_thread_exit(void)
+{
+ kthread_t *kt = curthread;
+
+ ASSERT(pthread_equal(kt->t_tid, pthread_self()));
+
+ umem_free(kt, sizeof(kthread_t));
+
+ pthread_mutex_lock(&kthread_lock);
+ kthread_nr--;
+ pthread_mutex_unlock(&kthread_lock);
+
+ pthread_cond_broadcast(&kthread_cond);
+ pthread_exit((void *)TS_MAGIC);
+}
+
+void
+zk_thread_join(kt_did_t tid)
+{
+ void *ret;
+
+ pthread_join((pthread_t)tid, &ret);
+ VERIFY3P(ret, ==, (void *)TS_MAGIC);
}
/*
* mutexes
* =========================================================================
*/
+
void
-zmutex_init(kmutex_t *mp)
+mutex_init(kmutex_t *mp, char *name, int type, void *cookie)
{
- mp->m_owner = NULL;
- mp->initialized = B_TRUE;
- (void) _mutex_init(&mp->m_lock, USYNC_THREAD, NULL);
+ ASSERT3S(type, ==, MUTEX_DEFAULT);
+ ASSERT3P(cookie, ==, NULL);
+ mp->m_owner = MTX_INIT;
+ mp->m_magic = MTX_MAGIC;
+ VERIFY3S(pthread_mutex_init(&mp->m_lock, NULL), ==, 0);
}
void
-zmutex_destroy(kmutex_t *mp)
+mutex_destroy(kmutex_t *mp)
{
- ASSERT(mp->initialized == B_TRUE);
- ASSERT(mp->m_owner == NULL);
- (void) _mutex_destroy(&(mp)->m_lock);
- mp->m_owner = (void *)-1UL;
- mp->initialized = B_FALSE;
+ ASSERT3U(mp->m_magic, ==, MTX_MAGIC);
+ ASSERT3P(mp->m_owner, ==, MTX_INIT);
+ VERIFY3S(pthread_mutex_destroy(&(mp)->m_lock), ==, 0);
+ mp->m_owner = MTX_DEST;
+ mp->m_magic = 0;
}
void
mutex_enter(kmutex_t *mp)
{
- ASSERT(mp->initialized == B_TRUE);
- ASSERT(mp->m_owner != (void *)-1UL);
- ASSERT(mp->m_owner != curthread);
- VERIFY(mutex_lock(&mp->m_lock) == 0);
- ASSERT(mp->m_owner == NULL);
+ ASSERT3U(mp->m_magic, ==, MTX_MAGIC);
+ ASSERT3P(mp->m_owner, !=, MTX_DEST);
+ ASSERT3P(mp->m_owner, !=, curthread);
+ VERIFY3S(pthread_mutex_lock(&mp->m_lock), ==, 0);
+ ASSERT3P(mp->m_owner, ==, MTX_INIT);
mp->m_owner = curthread;
}
int
mutex_tryenter(kmutex_t *mp)
{
- ASSERT(mp->initialized == B_TRUE);
- ASSERT(mp->m_owner != (void *)-1UL);
- if (0 == mutex_trylock(&mp->m_lock)) {
- ASSERT(mp->m_owner == NULL);
+ ASSERT3U(mp->m_magic, ==, MTX_MAGIC);
+ ASSERT3P(mp->m_owner, !=, MTX_DEST);
+ if (0 == pthread_mutex_trylock(&mp->m_lock)) {
+ ASSERT3P(mp->m_owner, ==, MTX_INIT);
mp->m_owner = curthread;
return (1);
} else {
void
mutex_exit(kmutex_t *mp)
{
- ASSERT(mp->initialized == B_TRUE);
- ASSERT(mutex_owner(mp) == curthread);
- mp->m_owner = NULL;
- VERIFY(mutex_unlock(&mp->m_lock) == 0);
+ ASSERT3U(mp->m_magic, ==, MTX_MAGIC);
+ ASSERT3P(mutex_owner(mp), ==, curthread);
+ mp->m_owner = MTX_INIT;
+ VERIFY3S(pthread_mutex_unlock(&mp->m_lock), ==, 0);
}
void *
mutex_owner(kmutex_t *mp)
{
- ASSERT(mp->initialized == B_TRUE);
+ ASSERT3U(mp->m_magic, ==, MTX_MAGIC);
return (mp->m_owner);
}
+int
+mutex_held(kmutex_t *mp)
+{
+ return (mp->m_owner == curthread);
+}
+
/*
* =========================================================================
* rwlocks
* =========================================================================
*/
-/*ARGSUSED*/
+
void
rw_init(krwlock_t *rwlp, char *name, int type, void *arg)
{
- rwlock_init(&rwlp->rw_lock, USYNC_THREAD, NULL);
- rwlp->rw_owner = NULL;
- rwlp->initialized = B_TRUE;
+ ASSERT3S(type, ==, RW_DEFAULT);
+ ASSERT3P(arg, ==, NULL);
+ VERIFY3S(pthread_rwlock_init(&rwlp->rw_lock, NULL), ==, 0);
+ rwlp->rw_owner = RW_INIT;
+ rwlp->rw_wr_owner = RW_INIT;
+ rwlp->rw_readers = 0;
+ rwlp->rw_magic = RW_MAGIC;
}
void
rw_destroy(krwlock_t *rwlp)
{
- rwlock_destroy(&rwlp->rw_lock);
- rwlp->rw_owner = (void *)-1UL;
- rwlp->initialized = B_FALSE;
+ ASSERT3U(rwlp->rw_magic, ==, RW_MAGIC);
+
+ VERIFY3S(pthread_rwlock_destroy(&rwlp->rw_lock), ==, 0);
+ rwlp->rw_magic = 0;
}
void
rw_enter(krwlock_t *rwlp, krw_t rw)
{
- ASSERT(!RW_LOCK_HELD(rwlp));
- ASSERT(rwlp->initialized == B_TRUE);
- ASSERT(rwlp->rw_owner != (void *)-1UL);
- ASSERT(rwlp->rw_owner != curthread);
+ ASSERT3U(rwlp->rw_magic, ==, RW_MAGIC);
+ ASSERT3P(rwlp->rw_owner, !=, curthread);
+ ASSERT3P(rwlp->rw_wr_owner, !=, curthread);
- if (rw == RW_READER)
- VERIFY(rw_rdlock(&rwlp->rw_lock) == 0);
- else
- VERIFY(rw_wrlock(&rwlp->rw_lock) == 0);
+ if (rw == RW_READER) {
+ VERIFY3S(pthread_rwlock_rdlock(&rwlp->rw_lock), ==, 0);
+ ASSERT3P(rwlp->rw_wr_owner, ==, RW_INIT);
+
+ atomic_inc_uint(&rwlp->rw_readers);
+ } else {
+ VERIFY3S(pthread_rwlock_wrlock(&rwlp->rw_lock), ==, 0);
+ ASSERT3P(rwlp->rw_wr_owner, ==, RW_INIT);
+ ASSERT3U(rwlp->rw_readers, ==, 0);
+
+ rwlp->rw_wr_owner = curthread;
+ }
rwlp->rw_owner = curthread;
}
void
rw_exit(krwlock_t *rwlp)
{
- ASSERT(rwlp->initialized == B_TRUE);
- ASSERT(rwlp->rw_owner != (void *)-1UL);
+ ASSERT3U(rwlp->rw_magic, ==, RW_MAGIC);
+ ASSERT(RW_LOCK_HELD(rwlp));
+
+ if (RW_READ_HELD(rwlp))
+ atomic_dec_uint(&rwlp->rw_readers);
+ else
+ rwlp->rw_wr_owner = RW_INIT;
- rwlp->rw_owner = NULL;
- VERIFY(rw_unlock(&rwlp->rw_lock) == 0);
+ rwlp->rw_owner = RW_INIT;
+ VERIFY3S(pthread_rwlock_unlock(&rwlp->rw_lock), ==, 0);
}
int
{
int rv;
- ASSERT(rwlp->initialized == B_TRUE);
- ASSERT(rwlp->rw_owner != (void *)-1UL);
+ ASSERT3U(rwlp->rw_magic, ==, RW_MAGIC);
if (rw == RW_READER)
- rv = rw_tryrdlock(&rwlp->rw_lock);
+ rv = pthread_rwlock_tryrdlock(&rwlp->rw_lock);
else
- rv = rw_trywrlock(&rwlp->rw_lock);
+ rv = pthread_rwlock_trywrlock(&rwlp->rw_lock);
if (rv == 0) {
+ ASSERT3P(rwlp->rw_wr_owner, ==, RW_INIT);
+
+ if (rw == RW_READER)
+ atomic_inc_uint(&rwlp->rw_readers);
+ else {
+ ASSERT3U(rwlp->rw_readers, ==, 0);
+ rwlp->rw_wr_owner = curthread;
+ }
+
rwlp->rw_owner = curthread;
return (1);
}
+ VERIFY3S(rv, ==, EBUSY);
+
return (0);
}
-/*ARGSUSED*/
int
rw_tryupgrade(krwlock_t *rwlp)
{
- ASSERT(rwlp->initialized == B_TRUE);
- ASSERT(rwlp->rw_owner != (void *)-1UL);
+ ASSERT3U(rwlp->rw_magic, ==, RW_MAGIC);
return (0);
}
* condition variables
* =========================================================================
*/
-/*ARGSUSED*/
+
void
cv_init(kcondvar_t *cv, char *name, int type, void *arg)
{
- VERIFY(cond_init(cv, type, NULL) == 0);
+ ASSERT3S(type, ==, CV_DEFAULT);
+ cv->cv_magic = CV_MAGIC;
+ VERIFY3S(pthread_cond_init(&cv->cv, NULL), ==, 0);
}
void
cv_destroy(kcondvar_t *cv)
{
- VERIFY(cond_destroy(cv) == 0);
+ ASSERT3U(cv->cv_magic, ==, CV_MAGIC);
+ VERIFY3S(pthread_cond_destroy(&cv->cv), ==, 0);
+ cv->cv_magic = 0;
}
void
cv_wait(kcondvar_t *cv, kmutex_t *mp)
{
- ASSERT(mutex_owner(mp) == curthread);
- mp->m_owner = NULL;
- int ret = cond_wait(cv, &mp->m_lock);
- VERIFY(ret == 0 || ret == EINTR);
+ ASSERT3U(cv->cv_magic, ==, CV_MAGIC);
+ ASSERT3P(mutex_owner(mp), ==, curthread);
+ mp->m_owner = MTX_INIT;
+ int ret = pthread_cond_wait(&cv->cv, &mp->m_lock);
+ if (ret != 0)
+ VERIFY3S(ret, ==, EINTR);
mp->m_owner = curthread;
}
cv_timedwait(kcondvar_t *cv, kmutex_t *mp, clock_t abstime)
{
int error;
+ struct timeval tv;
timestruc_t ts;
clock_t delta;
+ ASSERT3U(cv->cv_magic, ==, CV_MAGIC);
+
top:
delta = abstime - ddi_get_lbolt();
if (delta <= 0)
return (-1);
- ts.tv_sec = delta / hz;
- ts.tv_nsec = (delta % hz) * (NANOSEC / hz);
+ VERIFY(gettimeofday(&tv, NULL) == 0);
- ASSERT(mutex_owner(mp) == curthread);
- mp->m_owner = NULL;
- error = cond_reltimedwait(cv, &mp->m_lock, &ts);
+ ts.tv_sec = tv.tv_sec + delta / hz;
+ ts.tv_nsec = tv.tv_usec * 1000 + (delta % hz) * (NANOSEC / hz);
+ if (ts.tv_nsec >= NANOSEC) {
+ ts.tv_sec++;
+ ts.tv_nsec -= NANOSEC;
+ }
+
+ ASSERT3P(mutex_owner(mp), ==, curthread);
+ mp->m_owner = MTX_INIT;
+ error = pthread_cond_timedwait(&cv->cv, &mp->m_lock, &ts);
mp->m_owner = curthread;
- if (error == ETIME)
+ if (error == ETIMEDOUT)
return (-1);
if (error == EINTR)
goto top;
- ASSERT(error == 0);
+ VERIFY3S(error, ==, 0);
return (1);
}
void
cv_signal(kcondvar_t *cv)
{
- VERIFY(cond_signal(cv) == 0);
+ ASSERT3U(cv->cv_magic, ==, CV_MAGIC);
+ VERIFY3S(pthread_cond_signal(&cv->cv), ==, 0);
}
void
cv_broadcast(kcondvar_t *cv)
{
- VERIFY(cond_broadcast(cv) == 0);
+ ASSERT3U(cv->cv_magic, ==, CV_MAGIC);
+ VERIFY3S(pthread_cond_broadcast(&cv->cv), ==, 0);
}
/*
{
int fd;
vnode_t *vp;
- int old_umask;
- char realpath[MAXPATHLEN];
+ int old_umask = 0;
+ char *realpath;
struct stat64 st;
int err;
+ realpath = umem_alloc(MAXPATHLEN, UMEM_NOFAIL);
+
/*
* If we're accessing a real disk from userland, we need to use
* the character interface to avoid caching. This is particularly
* for its size. So -- gag -- we open the block device to get
* its size, and remember it for subsequent VOP_GETATTR().
*/
+#if defined(__sun__) || defined(__sun)
if (strncmp(path, "/dev/", 5) == 0) {
+#else
+ if (0) {
+#endif
char *dsk;
fd = open64(path, O_RDONLY);
- if (fd == -1)
- return (errno);
+ if (fd == -1) {
+ err = errno;
+ free(realpath);
+ return (err);
+ }
if (fstat64(fd, &st) == -1) {
+ err = errno;
close(fd);
- return (errno);
+ free(realpath);
+ return (err);
}
close(fd);
(void) sprintf(realpath, "%s", path);
dsk + 1);
} else {
(void) sprintf(realpath, "%s", path);
- if (!(flags & FCREAT) && stat64(realpath, &st) == -1)
- return (errno);
+ if (!(flags & FCREAT) && stat64(realpath, &st) == -1) {
+ err = errno;
+ free(realpath);
+ return (err);
+ }
+ }
+
+ if (!(flags & FCREAT) && S_ISBLK(st.st_mode)) {
+#ifdef __linux__
+ flags |= O_DIRECT;
+#endif
+ /* We shouldn't be writing to block devices in userspace */
+ VERIFY(!(flags & FWRITE));
}
if (flags & FCREAT)
* FREAD and FWRITE to the corresponding O_RDONLY, O_WRONLY, and O_RDWR.
*/
fd = open64(realpath, flags - FREAD, mode);
+ free(realpath);
if (flags & FCREAT)
(void) umask(old_umask);
if (fd == -1)
return (errno);
- if (fstat64(fd, &st) == -1) {
+ if (fstat64_blk(fd, &st) == -1) {
err = errno;
close(fd);
return (err);
* To simulate partial disk writes, we split writes into two
* system calls so that the process can be killed in between.
*/
- split = (len > 0 ? rand() % len : 0);
+ int sectors = len >> SPA_MINBLOCKSHIFT;
+ split = (sectors > 0 ? rand() % sectors : 0) <<
+ SPA_MINBLOCKSHIFT;
rc = pwrite64(vp->v_fd, addr, split, offset);
if (rc != -1) {
done = rc;
}
}
+#ifdef __linux__
+ if (rc == -1 && errno == EINVAL) {
+ /*
+ * Under Linux, this most likely means an alignment issue
+ * (memory or disk) due to O_DIRECT, so we abort() in order to
+ * catch the offender.
+ */
+ abort();
+ }
+#endif
if (rc == -1)
return (errno);
fop_getattr(vnode_t *vp, vattr_t *vap)
{
struct stat64 st;
+ int err;
- if (fstat64(vp->v_fd, &st) == -1) {
+ if (fstat64_blk(vp->v_fd, &st) == -1) {
+ err = errno;
close(vp->v_fd);
- return (errno);
+ return (err);
}
vap->va_size = st.st_size;
return (0);
}
-#ifdef ZFS_DEBUG
-
/*
* =========================================================================
* Figure out which debugging statements to print
if (dprintf_find_string("pid"))
(void) printf("%d ", getpid());
if (dprintf_find_string("tid"))
- (void) printf("%u ", thr_self());
+ (void) printf("%u ", (uint_t) pthread_self());
if (dprintf_find_string("cpu"))
(void) printf("%u ", getcpuid());
if (dprintf_find_string("time"))
}
}
-#endif /* ZFS_DEBUG */
-
/*
* =========================================================================
* cmn_err() and panic()
VERIFY((random_fd = open("/dev/random", O_RDONLY)) != -1);
VERIFY((urandom_fd = open("/dev/urandom", O_RDONLY)) != -1);
+ thread_init();
system_taskq_init();
spa_init(mode);
spa_fini();
system_taskq_fini();
+ thread_fini();
close(random_fd);
close(urandom_fd);
urandom_fd = -1;
}
-int
-z_uncompress(void *dst, size_t *dstlen, const void *src, size_t srclen)
-{
- int ret;
- uLongf len = *dstlen;
-
- if ((ret = uncompress(dst, &len, src, srclen)) == Z_OK)
- *dstlen = (size_t)len;
-
- return (ret);
-}
-
-int
-z_compress_level(void *dst, size_t *dstlen, const void *src, size_t srclen,
- int level)
-{
- int ret;
- uLongf len = *dstlen;
-
- if ((ret = compress2(dst, &len, src, srclen, level)) == Z_OK)
- *dstlen = (size_t)len;
-
- return (ret);
-}
-
uid_t
crgetuid(cred_t *cr)
{
umem_free(ksid, sizeof (ksiddomain_t));
}
-/*
- * Do not change the length of the returned string; it must be freed
- * with strfree().
- */
char *
-kmem_asprintf(const char *fmt, ...)
+kmem_vasprintf(const char *fmt, va_list adx)
{
- int size;
- va_list adx;
- char *buf;
+ char *buf = NULL;
+ va_list adx_copy;
- va_start(adx, fmt);
- size = vsnprintf(NULL, 0, fmt, adx) + 1;
- va_end(adx);
+ va_copy(adx_copy, adx);
+ VERIFY(vasprintf(&buf, fmt, adx_copy) != -1);
+ va_end(adx_copy);
+
+ return (buf);
+}
- buf = kmem_alloc(size, KM_SLEEP);
+char *
+kmem_asprintf(const char *fmt, ...)
+{
+ char *buf = NULL;
+ va_list adx;
va_start(adx, fmt);
- size = vsnprintf(buf, size, fmt, adx);
+ VERIFY(vasprintf(&buf, fmt, adx) != -1);
va_end(adx);
return (buf);
git://git.camperquake.de / zfs.git / blobdiff