#include <sys/resource.h>
#endif
-#include <rx/xdr.h>
+#include <afs/opr.h>
+#ifdef AFS_PTHREAD_ENV
+# include <opr/lock.h>
+#endif
#include <afs/afsint.h>
#include <afs/afssyscalls.h>
#include <afs/afsutil.h>
#include "nfs.h"
#include "ihandle.h"
#include "viceinode.h"
-#include "afs/afs_assert.h"
#ifdef AFS_PTHREAD_ENV
pthread_once_t ih_glock_once = PTHREAD_ONCE_INIT;
IHashBucket_t ihashTable[I_HANDLE_HASH_SIZE];
static int _ih_release_r(IHandle_t * ihP);
-void *ih_sync_thread(void *);
/* start-time configurable I/O limits */
ih_init_params vol_io_params;
/* fd cache size that will be used if/when ih_UseLargeCache()
* is called */
vol_io_params.fd_max_cachesize = FD_MAX_CACHESIZE;
+
+ vol_io_params.sync_behavior = IH_SYNC_ONCLOSE;
+}
+
+int
+ih_SetSyncBehavior(const char *behavior)
+{
+ int val;
+
+ if (strcmp(behavior, "always") == 0) {
+ val = IH_SYNC_ALWAYS;
+
+ } else if (strcmp(behavior, "onclose") == 0) {
+ val = IH_SYNC_ONCLOSE;
+
+ } else if (strcmp(behavior, "never") == 0) {
+ val = IH_SYNC_NEVER;
+
+ } else {
+ /* invalid behavior name */
+ return -1;
+ }
+
+ vol_io_params.sync_behavior = val;
+ return 0;
}
#ifdef AFS_PTHREAD_ENV
void
ih_glock_init(void)
{
- MUTEX_INIT(&ih_glock_mutex, "ih glock", MUTEX_DEFAULT, 0);
+ opr_mutex_init(&ih_glock_mutex);
}
#endif /* AFS_PTHREAD_ENV */
ih_Initialize(void)
{
int i;
- osi_Assert(!ih_Inited);
+ opr_Assert(!ih_Inited);
ih_Inited = 1;
DLL_INIT_LIST(ihAvailHead, ihAvailTail);
DLL_INIT_LIST(fdAvailHead, fdAvailTail);
#elif defined(AFS_SUN5_ENV) || defined(AFS_NBSD_ENV)
{
struct rlimit rlim;
- osi_Assert(getrlimit(RLIMIT_NOFILE, &rlim) == 0);
+ opr_Verify(getrlimit(RLIMIT_NOFILE, &rlim) == 0);
rlim.rlim_cur = rlim.rlim_max;
- osi_Assert(setrlimit(RLIMIT_NOFILE, &rlim) == 0);
+ opr_Verify(setrlimit(RLIMIT_NOFILE, &rlim) == 0);
fdMaxCacheSize = rlim.rlim_cur - vol_io_params.fd_handle_setaside;
#ifdef AFS_NBSD_ENV
/* XXX this is to avoid using up all system fd netbsd is
*/
fdMaxCacheSize /= 4;
#endif
- fdMaxCacheSize = MIN(fdMaxCacheSize, vol_io_params.fd_max_cachesize);
- osi_Assert(fdMaxCacheSize > 0);
+ fdMaxCacheSize = min(fdMaxCacheSize, vol_io_params.fd_max_cachesize);
+ opr_Assert(fdMaxCacheSize > 0);
}
#elif defined(AFS_HPUX_ENV)
/* Avoid problems with "UFSOpen: igetinode failed" panics on HPUX 11.0 */
fdMaxCacheSize = 0;
#else
{
- long fdMax = MAX(sysconf(_SC_OPEN_MAX) - vol_io_params.fd_handle_setaside,
+ long fdMax = max(sysconf(_SC_OPEN_MAX) - vol_io_params.fd_handle_setaside,
0);
- fdMaxCacheSize = (int)MIN(fdMax, vol_io_params.fd_max_cachesize);
+ fdMaxCacheSize = (int)min(fdMax, vol_io_params.fd_max_cachesize);
}
#endif
- fdCacheSize = MIN(fdMaxCacheSize, vol_io_params.fd_initial_cachesize);
-
- {
-#ifdef AFS_PTHREAD_ENV
- pthread_t syncer;
- pthread_attr_t tattr;
-
- pthread_attr_init(&tattr);
- pthread_attr_setdetachstate(&tattr, PTHREAD_CREATE_DETACHED);
-
- pthread_create(&syncer, &tattr, ih_sync_thread, NULL);
-#else /* AFS_PTHREAD_ENV */
- PROCESS syncer;
- LWP_CreateProcess(ih_sync_thread, 16*1024, LWP_MAX_PRIORITY - 2,
- NULL, "ih_syncer", &syncer);
-#endif /* AFS_PTHREAD_ENV */
- }
-
+ fdCacheSize = min(fdMaxCacheSize, vol_io_params.fd_initial_cachesize);
}
/* Make the file descriptor cache as big as possible. Don't this call
int i;
IHandle_t *ihP;
- osi_Assert(ihAvailHead == NULL);
- ihP = (IHandle_t *) malloc(I_HANDLE_MALLOCSIZE * sizeof(IHandle_t));
- osi_Assert(ihP != NULL);
+ opr_Assert(ihAvailHead == NULL);
+ ihP = malloc(I_HANDLE_MALLOCSIZE * sizeof(IHandle_t));
+ opr_Assert(ihP != NULL);
for (i = 0; i < I_HANDLE_MALLOCSIZE; i++) {
ihP[i].ih_refcnt = 0;
DLL_INSERT_TAIL(&ihP[i], ihAvailHead, ihAvailTail, ih_next, ih_prev);
iHandleAllocateChunk();
}
ihP = ihAvailHead;
- osi_Assert(ihP->ih_refcnt == 0);
+ opr_Assert(ihP->ih_refcnt == 0);
DLL_DELETE(ihP, ihAvailHead, ihAvailTail, ih_next, ih_prev);
ihP->ih_dev = dev;
ihP->ih_vid = vid;
ih_copy(IHandle_t * ihP)
{
IH_LOCK;
- osi_Assert(ih_Inited);
- osi_Assert(ihP->ih_refcnt > 0);
+ opr_Assert(ih_Inited);
+ opr_Assert(ihP->ih_refcnt > 0);
ihP->ih_refcnt++;
IH_UNLOCK;
return ihP;
int i;
FdHandle_t *fdP;
- osi_Assert(fdAvailHead == NULL);
- fdP = (FdHandle_t *) malloc(FD_HANDLE_MALLOCSIZE * sizeof(FdHandle_t));
- osi_Assert(fdP != NULL);
+ opr_Assert(fdAvailHead == NULL);
+ fdP = malloc(FD_HANDLE_MALLOCSIZE * sizeof(FdHandle_t));
+ opr_Assert(fdP != NULL);
for (i = 0; i < FD_HANDLE_MALLOCSIZE; i++) {
fdP[i].fd_status = FD_HANDLE_AVAIL;
fdP[i].fd_refcnt = 0;
int i;
StreamHandle_t *streamP;
- osi_Assert(streamAvailHead == NULL);
+ opr_Assert(streamAvailHead == NULL);
streamP = (StreamHandle_t *)
malloc(STREAM_HANDLE_MALLOCSIZE * sizeof(StreamHandle_t));
- osi_Assert(streamP != NULL);
+ opr_Assert(streamP != NULL);
for (i = 0; i < STREAM_HANDLE_MALLOCSIZE; i++) {
streamP[i].str_fd = INVALID_FD;
DLL_INSERT_TAIL(&streamP[i], streamAvailHead, streamAvailTail,
/*
* Get a file descriptor handle given an Inode handle
+ * Takes the given file descriptor, and creates a new FdHandle_t for it,
+ * attached to the given IHandle_t. fd can be INVALID_FD, indicating that the
+ * caller failed to open the relevant file because we had too many FDs open;
+ * ih_attachfd_r will then just evict/close an existing fd in the cache, and
+ * return NULL.
+ */
+static FdHandle_t *
+ih_attachfd_r(IHandle_t *ihP, FD_t fd)
+{
+ FD_t closeFd;
+ FdHandle_t *fdP;
+
+ /* fdCacheSize limits the size of the descriptor cache, but
+ * we permit the number of open files to exceed fdCacheSize.
+ * We only recycle open file descriptors when the number
+ * of open files reaches the size of the cache */
+ if ((fdInUseCount > fdCacheSize || fd == INVALID_FD) && fdLruHead != NULL) {
+ fdP = fdLruHead;
+ opr_Assert(fdP->fd_status == FD_HANDLE_OPEN);
+ DLL_DELETE(fdP, fdLruHead, fdLruTail, fd_next, fd_prev);
+ DLL_DELETE(fdP, fdP->fd_ih->ih_fdhead, fdP->fd_ih->ih_fdtail,
+ fd_ihnext, fd_ihprev);
+ closeFd = fdP->fd_fd;
+ if (fd == INVALID_FD) {
+ fdCacheSize--; /* reduce in order to not run into here too often */
+ DLL_INSERT_TAIL(fdP, fdAvailHead, fdAvailTail, fd_next, fd_prev);
+ fdP->fd_status = FD_HANDLE_AVAIL;
+ fdP->fd_ih = NULL;
+ fdP->fd_fd = INVALID_FD;
+ IH_UNLOCK;
+ OS_CLOSE(closeFd);
+ IH_LOCK;
+ fdInUseCount -= 1;
+ return NULL;
+ }
+ } else {
+ if (fdAvailHead == NULL) {
+ fdHandleAllocateChunk();
+ }
+ fdP = fdAvailHead;
+ opr_Assert(fdP->fd_status == FD_HANDLE_AVAIL);
+ DLL_DELETE(fdP, fdAvailHead, fdAvailTail, fd_next, fd_prev);
+ closeFd = INVALID_FD;
+ }
+
+ fdP->fd_status = FD_HANDLE_INUSE;
+ fdP->fd_fd = fd;
+ fdP->fd_ih = ihP;
+ fdP->fd_refcnt++;
+
+ ihP->ih_refcnt++;
+
+ /* Add this handle to the Inode's list of open descriptors */
+ DLL_INSERT_TAIL(fdP, ihP->ih_fdhead, ihP->ih_fdtail, fd_ihnext,
+ fd_ihprev);
+
+ if (closeFd != INVALID_FD) {
+ IH_UNLOCK;
+ OS_CLOSE(closeFd);
+ IH_LOCK;
+ fdInUseCount -= 1;
+ }
+
+ return fdP;
+}
+
+FdHandle_t *
+ih_attachfd(IHandle_t *ihP, FD_t fd)
+{
+ FdHandle_t *fdP;
+
+ IH_LOCK;
+
+ fdInUseCount += 1;
+
+ fdP = ih_attachfd_r(ihP, fd);
+ if (!fdP) {
+ fdInUseCount -= 1;
+ }
+
+ IH_UNLOCK;
+
+ return fdP;
+}
+
+/*
+ * Get a file descriptor handle given an Inode handle
*/
FdHandle_t *
ih_open(IHandle_t * ihP)
{
FdHandle_t *fdP;
FD_t fd;
- FD_t closeFd;
if (!ihP) /* XXX should log here in the fileserver */
return NULL;
if (fdP->fd_status == FD_HANDLE_INUSE) {
continue;
}
- osi_Assert(fdP->fd_status == FD_HANDLE_OPEN);
+ opr_Assert(fdP->fd_status == FD_HANDLE_OPEN);
#else /* HAVE_PIO */
- osi_Assert(fdP->fd_status != FD_HANDLE_AVAIL);
+ opr_Assert(fdP->fd_status != FD_HANDLE_AVAIL);
#endif /* HAVE_PIO */
fdP->fd_refcnt++;
return NULL;
}
- /* fdCacheSize limits the size of the descriptor cache, but
- * we permit the number of open files to exceed fdCacheSize.
- * We only recycle open file descriptors when the number
- * of open files reaches the size of the cache */
- if ((fdInUseCount > fdCacheSize || fd == INVALID_FD) && fdLruHead != NULL) {
- fdP = fdLruHead;
- osi_Assert(fdP->fd_status == FD_HANDLE_OPEN);
- DLL_DELETE(fdP, fdLruHead, fdLruTail, fd_next, fd_prev);
- DLL_DELETE(fdP, fdP->fd_ih->ih_fdhead, fdP->fd_ih->ih_fdtail,
- fd_ihnext, fd_ihprev);
- closeFd = fdP->fd_fd;
- if (fd == INVALID_FD) {
- fdCacheSize--; /* reduce in order to not run into here too often */
- DLL_INSERT_TAIL(fdP, fdAvailHead, fdAvailTail, fd_next, fd_prev);
- fdP->fd_status = FD_HANDLE_AVAIL;
- fdP->fd_ih = NULL;
- fdP->fd_fd = INVALID_FD;
- IH_UNLOCK;
- OS_CLOSE(closeFd);
- goto ih_open_retry;
- }
- } else {
- if (fdAvailHead == NULL) {
- fdHandleAllocateChunk();
- }
- fdP = fdAvailHead;
- osi_Assert(fdP->fd_status == FD_HANDLE_AVAIL);
- DLL_DELETE(fdP, fdAvailHead, fdAvailTail, fd_next, fd_prev);
- closeFd = INVALID_FD;
- }
-
- fdP->fd_status = FD_HANDLE_INUSE;
- fdP->fd_fd = fd;
- fdP->fd_ih = ihP;
- fdP->fd_refcnt++;
-
- ihP->ih_refcnt++;
-
- /* Add this handle to the Inode's list of open descriptors */
- DLL_INSERT_TAIL(fdP, ihP->ih_fdhead, ihP->ih_fdtail, fd_ihnext,
- fd_ihprev);
-
- if (closeFd != INVALID_FD) {
+ fdP = ih_attachfd_r(ihP, fd);
+ if (!fdP) {
+ opr_Assert(fd == INVALID_FD);
IH_UNLOCK;
- OS_CLOSE(closeFd);
- IH_LOCK;
- fdInUseCount -= 1;
+ goto ih_open_retry;
}
IH_UNLOCK;
+
return fdP;
}
return 0;
IH_LOCK;
- osi_Assert(ih_Inited);
- osi_Assert(fdInUseCount > 0);
- osi_Assert(fdP->fd_status == FD_HANDLE_INUSE ||
+ opr_Assert(ih_Inited);
+ opr_Assert(fdInUseCount > 0);
+ opr_Assert(fdP->fd_status == FD_HANDLE_INUSE ||
fdP->fd_status == FD_HANDLE_CLOSING);
ihP = fdP->fd_ih;
return 0;
IH_LOCK;
- osi_Assert(ih_Inited);
- osi_Assert(fdInUseCount > 0);
- osi_Assert(fdP->fd_status == FD_HANDLE_INUSE ||
+ opr_Assert(ih_Inited);
+ opr_Assert(fdInUseCount > 0);
+ opr_Assert(fdP->fd_status == FD_HANDLE_INUSE ||
fdP->fd_status == FD_HANDLE_CLOSING);
ihP = fdP->fd_ih;
*/
if (!ihP->ih_fdhead) {
ihP->ih_flags &= ~IH_REALLY_CLOSED;
+ } else {
+ FdHandle_t *lfdP, *next;
+ int clear = 1;
+ for (lfdP = ihP->ih_fdhead; lfdP != NULL; lfdP = next) {
+ next = lfdP->fd_ihnext;
+ osi_Assert(lfdP->fd_ih == ihP);
+ if (lfdP->fd_status != FD_HANDLE_CLOSING) {
+ clear = 0;
+ break;
+ }
+ }
+ /* no *future* fd should be subjected to this */
+ if (clear)
+ ihP->ih_flags &= ~IH_REALLY_CLOSED;
}
if (fdP->fd_refcnt == 0) {
} else if (strcmp(mode, "a+") == 0) {
fd = OS_OPEN(filename, O_RDWR | O_APPEND | O_CREAT, 0);
} else {
- osi_Assert(FALSE); /* not implemented */
+ opr_abort(); /* not implemented */
}
if (fd == INVALID_FD) {
streamP->str_bufoff = 0;
streamP->str_buflen = 0;
} else {
- osi_Assert(streamP->str_direction == STREAM_DIRECTION_READ);
+ opr_Assert(streamP->str_direction == STREAM_DIRECTION_READ);
}
bytesRead = 0;
streamP->str_bufoff = 0;
streamP->str_buflen = STREAM_HANDLE_BUFSIZE;
} else {
- osi_Assert(streamP->str_direction == STREAM_DIRECTION_WRITE);
+ opr_Assert(streamP->str_direction == STREAM_DIRECTION_WRITE);
}
nbytes = size * nitems;
ssize_t rc;
int retval = 0;
- osi_Assert(streamP != NULL);
+ opr_Assert(streamP != NULL);
if (streamP->str_direction == STREAM_DIRECTION_WRITE
&& streamP->str_bufoff > 0) {
rc = OS_PWRITE(streamP->str_fd, streamP->str_buffer,
int closeCount, closedAll;
FdHandle_t *fdP, *head, *tail, *next;
- osi_Assert(ihP->ih_refcnt > 0);
+ opr_Assert(ihP->ih_refcnt > 0);
closedAll = 1;
DLL_INIT_LIST(head, tail);
*/
for (fdP = ihP->ih_fdhead; fdP != NULL; fdP = next) {
next = fdP->fd_ihnext;
- osi_Assert(fdP->fd_ih == ihP);
- osi_Assert(fdP->fd_status == FD_HANDLE_OPEN
+ opr_Assert(fdP->fd_ih == ihP);
+ opr_Assert(fdP->fd_status == FD_HANDLE_OPEN
|| fdP->fd_status == FD_HANDLE_INUSE
|| fdP->fd_status == FD_HANDLE_CLOSING);
if (fdP->fd_status == FD_HANDLE_OPEN) {
* closed all file descriptors.
*/
if (ihP->ih_refcnt == 1 || closedAll) {
- osi_Assert(closedAll);
- osi_Assert(!ihP->ih_fdhead);
- osi_Assert(!ihP->ih_fdtail);
+ opr_Assert(closedAll);
+ opr_Assert(!ihP->ih_fdhead);
+ opr_Assert(!ihP->ih_fdtail);
}
if (head == NULL) {
}
IH_LOCK;
- osi_Assert(fdInUseCount >= closeCount);
+ opr_Assert(fdInUseCount >= closeCount);
fdInUseCount -= closeCount;
/*
ihP->ih_refcnt++; /* must not disappear over unlock */
if (ihP->ih_synced) {
FdHandle_t *fdP;
+ opr_Assert(vol_io_params.sync_behavior != IH_SYNC_ALWAYS);
+ opr_Assert(vol_io_params.sync_behavior != IH_SYNC_NEVER);
ihP->ih_synced = 0;
IH_UNLOCK;
IH_LOCK;
}
- osi_Assert(ihP->ih_refcnt > 0);
+ opr_Assert(ihP->ih_refcnt > 0);
ih_fdclose(ihP);
if (!ihP)
return 0;
- osi_Assert(ihP->ih_refcnt > 0);
+ opr_Assert(ihP->ih_refcnt > 0);
if (ihP->ih_refcnt > 1) {
ihP->ih_refcnt--;
return code;
}
-void
-ih_sync_all(void) {
-
- int ihash;
-
- IH_LOCK;
- for (ihash = 0; ihash < I_HANDLE_HASH_SIZE; ihash++) {
- IHandle_t *ihP, *ihPnext;
-
- ihP = ihashTable[ihash].ihash_head;
- if (ihP)
- ihP->ih_refcnt++; /* must not disappear over unlock */
- for (; ihP; ihP = ihPnext) {
-
- if (ihP->ih_synced) {
- FdHandle_t *fdP;
-
- ihP->ih_synced = 0;
- IH_UNLOCK;
-
- fdP = IH_OPEN(ihP);
- if (fdP) {
- OS_SYNC(fdP->fd_fd);
- FDH_CLOSE(fdP);
- }
-
- IH_LOCK;
- }
-
- /* when decrementing the refcount, the ihandle might disappear
- and we might not even be able to proceed to the next one.
- Hence the gymnastics putting a hold on the next one already */
- ihPnext = ihP->ih_next;
- if (ihPnext) ihPnext->ih_refcnt++;
-
- if (ihP->ih_refcnt > 1)
- ihP->ih_refcnt--;
- else
- _ih_release_r(ihP);
- }
- }
- IH_UNLOCK;
-}
-
-void *
-ih_sync_thread(void *dummy) {
- afs_pthread_setname_self("ih_syncer");
- while(1) {
-
-#ifdef AFS_PTHREAD_ENV
- sleep(10);
-#else /* AFS_PTHREAD_ENV */
- IOMGR_Sleep(60);
-#endif /* AFS_PTHREAD_ENV */
-
- ih_sync_all();
- }
- return NULL;
-}
-
-
/*************************************************************************
* OS specific support routines.
*************************************************************************/
}
#endif /* AFS_NAMEI_ENV */
+#if defined(AFS_NT40_ENV) || !defined(AFS_NAMEI_ENV)
+/* Unix namei implements its own more efficient IH_CREATE_INIT; this wrapper
+ * is for everyone else */
+IHandle_t *
+ih_icreate_init(IHandle_t *lh, int dev, char *part, Inode nearInode,
+ afs_uint32 p1, afs_uint32 p2, afs_uint32 p3, afs_uint32 p4)
+{
+ IHandle_t *ihP;
+ Inode ino = IH_CREATE(lh, dev, part, nearInode, p1, p2, p3, p4);
+ if (!VALID_INO(ino)) {
+ return NULL;
+ }
+ IH_INIT(ihP, dev, p1, ino);
+ return ihP;
+}
+#endif
+
afs_sfsize_t
ih_size(FD_t fd)
{
return OS_WRITE(fd, buf, count);
}
#endif /* !HAVE_PIO */
+
+#ifndef AFS_NT40_ENV
+int
+ih_isunlinked(int fd)
+{
+ struct afs_stat_st status;
+ if (afs_fstat(fd, &status) < 0) {
+ return -1;
+ }
+ if (status.st_nlink < 1) {
+ return 1;
+ }
+ return 0;
+}
+#endif /* !AFS_NT40_ENV */
+
+int
+ih_fdsync(FdHandle_t *fdP)
+{
+ switch (vol_io_params.sync_behavior) {
+ case IH_SYNC_ALWAYS:
+ return OS_SYNC(fdP->fd_fd);
+ case IH_SYNC_ONCLOSE:
+ if (fdP->fd_ih) {
+ fdP->fd_ih->ih_synced = 1;
+ return 0;
+ }
+ return 1;
+ case IH_SYNC_NEVER:
+ return 0;
+ default:
+ opr_Assert(0);
+ }
+}