#include "orangefs-kernel.h"
#include "orangefs-bufmap.h"
-static int wait_for_cancellation_downcall(struct orangefs_kernel_op_s *);
-static int wait_for_matching_downcall(struct orangefs_kernel_op_s *);
+static int wait_for_matching_downcall(struct orangefs_kernel_op_s *, long, bool);
+static void orangefs_clean_up_interrupted_operation(struct orangefs_kernel_op_s *);
/*
* What we do in this function is to walk the list of operations that are
"pvfs2-client-core: purging op tag %llu %s\n",
llu(op->tag),
get_opname_string(op));
- spin_lock(&op->lock);
set_op_state_purged(op);
- spin_unlock(&op->lock);
+ gossip_debug(GOSSIP_DEV_DEBUG,
+ "%s: op:%s: op_state:%d: process:%s:\n",
+ __func__,
+ get_opname_string(op),
+ op->op_state,
+ current->comm);
}
spin_unlock(&orangefs_request_list_lock);
}
-static inline void
-add_op_to_request_list(struct orangefs_kernel_op_s *op)
-{
- spin_lock(&orangefs_request_list_lock);
- spin_lock(&op->lock);
- set_op_state_waiting(op);
- list_add_tail(&op->list, &orangefs_request_list);
- spin_unlock(&orangefs_request_list_lock);
- spin_unlock(&op->lock);
- wake_up_interruptible(&orangefs_request_list_waitq);
-}
-
-static inline
-void add_priority_op_to_request_list(struct orangefs_kernel_op_s *op)
-{
- spin_lock(&orangefs_request_list_lock);
- spin_lock(&op->lock);
- set_op_state_waiting(op);
-
- list_add(&op->list, &orangefs_request_list);
- spin_unlock(&orangefs_request_list_lock);
- spin_unlock(&op->lock);
- wake_up_interruptible(&orangefs_request_list_waitq);
-}
-
/*
* submits a ORANGEFS operation and waits for it to complete
*
const char *op_name,
int flags)
{
- /* flags to modify behavior */
- sigset_t orig_sigset;
+ long timeout = MAX_SCHEDULE_TIMEOUT;
int ret = 0;
DEFINE_WAIT(wait_entry);
retry_servicing:
op->downcall.status = 0;
gossip_debug(GOSSIP_WAIT_DEBUG,
- "orangefs: service_operation: %s %p\n",
+ "%s: %s op:%p: process:%s: pid:%d:\n",
+ __func__,
op_name,
- op);
- gossip_debug(GOSSIP_WAIT_DEBUG,
- "orangefs: operation posted by process: %s, pid: %i\n",
+ op,
current->comm,
current->pid);
- /* mask out signals if this operation is not to be interrupted */
- if (!(flags & ORANGEFS_OP_INTERRUPTIBLE))
- orangefs_block_signals(&orig_sigset);
-
- if (!(flags & ORANGEFS_OP_NO_SEMAPHORE)) {
- ret = mutex_lock_interruptible(&request_mutex);
+ /*
+ * If ORANGEFS_OP_NO_MUTEX was set in flags, we need to avoid
+ * acquiring the request_mutex because we're servicing a
+ * high priority remount operation and the request_mutex is
+ * already taken.
+ */
+ if (!(flags & ORANGEFS_OP_NO_MUTEX)) {
+ if (flags & ORANGEFS_OP_INTERRUPTIBLE)
+ ret = mutex_lock_interruptible(&request_mutex);
+ else
+ ret = mutex_lock_killable(&request_mutex);
/*
* check to see if we were interrupted while waiting for
- * semaphore
+ * mutex
*/
if (ret < 0) {
- if (!(flags & ORANGEFS_OP_INTERRUPTIBLE))
- orangefs_set_signals(&orig_sigset);
op->downcall.status = ret;
gossip_debug(GOSSIP_WAIT_DEBUG,
- "orangefs: service_operation interrupted.\n");
+ "%s: service_operation interrupted.\n",
+ __func__);
return ret;
}
}
- gossip_debug(GOSSIP_WAIT_DEBUG,
- "%s:About to call is_daemon_in_service().\n",
- __func__);
-
- if (is_daemon_in_service() < 0) {
- /*
- * By incrementing the per-operation attempt counter, we
- * directly go into the timeout logic while waiting for
- * the matching downcall to be read
- */
- gossip_debug(GOSSIP_WAIT_DEBUG,
- "%s:client core is NOT in service(%d).\n",
- __func__,
- is_daemon_in_service());
- op->attempts++;
- }
-
/* queue up the operation */
- if (flags & ORANGEFS_OP_PRIORITY) {
- add_priority_op_to_request_list(op);
- } else {
+ spin_lock(&orangefs_request_list_lock);
+ spin_lock(&op->lock);
+ set_op_state_waiting(op);
+ gossip_debug(GOSSIP_DEV_DEBUG,
+ "%s: op:%s: op_state:%d: process:%s:\n",
+ __func__,
+ get_opname_string(op),
+ op->op_state,
+ current->comm);
+ /* add high priority remount op to the front of the line. */
+ if (flags & ORANGEFS_OP_PRIORITY)
+ list_add(&op->list, &orangefs_request_list);
+ else
+ list_add_tail(&op->list, &orangefs_request_list);
+ spin_unlock(&op->lock);
+ wake_up_interruptible(&orangefs_request_list_waitq);
+ if (!__is_daemon_in_service()) {
gossip_debug(GOSSIP_WAIT_DEBUG,
- "%s:About to call add_op_to_request_list().\n",
+ "%s:client core is NOT in service.\n",
__func__);
- add_op_to_request_list(op);
+ timeout = op_timeout_secs * HZ;
}
+ spin_unlock(&orangefs_request_list_lock);
- if (!(flags & ORANGEFS_OP_NO_SEMAPHORE))
+ if (!(flags & ORANGEFS_OP_NO_MUTEX))
mutex_unlock(&request_mutex);
- /*
- * If we are asked to service an asynchronous operation from
- * VFS perspective, we are done.
- */
- if (flags & ORANGEFS_OP_ASYNC)
- return 0;
+ ret = wait_for_matching_downcall(op, timeout,
+ flags & ORANGEFS_OP_INTERRUPTIBLE);
- if (flags & ORANGEFS_OP_CANCELLATION) {
- gossip_debug(GOSSIP_WAIT_DEBUG,
- "%s:"
- "About to call wait_for_cancellation_downcall.\n",
- __func__);
- ret = wait_for_cancellation_downcall(op);
- } else {
- ret = wait_for_matching_downcall(op);
- }
+ gossip_debug(GOSSIP_WAIT_DEBUG,
+ "%s: wait_for_matching_downcall returned %d for %p\n",
+ __func__,
+ ret,
+ op);
- if (ret < 0) {
- /* failed to get matching downcall */
- if (ret == -ETIMEDOUT) {
- gossip_err("orangefs: %s -- wait timed out; aborting attempt.\n",
- op_name);
- }
- op->downcall.status = ret;
- } else {
- /* got matching downcall; make sure status is in errno format */
+ /* got matching downcall; make sure status is in errno format */
+ if (!ret) {
+ spin_unlock(&op->lock);
op->downcall.status =
orangefs_normalize_to_errno(op->downcall.status);
ret = op->downcall.status;
+ goto out;
+ }
+
+ /* failed to get matching downcall */
+ if (ret == -ETIMEDOUT) {
+ gossip_err("%s: %s -- wait timed out; aborting attempt.\n",
+ __func__,
+ op_name);
}
- if (!(flags & ORANGEFS_OP_INTERRUPTIBLE))
- orangefs_set_signals(&orig_sigset);
+ /*
+ * remove a waiting op from the request list or
+ * remove an in-progress op from the in-progress list.
+ */
+ orangefs_clean_up_interrupted_operation(op);
- BUG_ON(ret != op->downcall.status);
+ op->downcall.status = ret;
/* retry if operation has not been serviced and if requested */
- if (!op_state_serviced(op) && op->downcall.status == -EAGAIN) {
+ if (ret == -EAGAIN) {
+ op->attempts++;
+ timeout = op_timeout_secs * HZ;
gossip_debug(GOSSIP_WAIT_DEBUG,
"orangefs: tag %llu (%s)"
" -- operation to be retried (%d attempt)\n",
llu(op->tag),
op_name,
- op->attempts + 1);
+ op->attempts);
+ /*
+ * io ops (ops that use the shared memory buffer) have
+ * to be returned to their caller for a retry. Other ops
+ * can just be recycled here.
+ */
if (!op->uses_shared_memory)
- /*
- * this operation doesn't use the shared memory
- * system
- */
goto retry_servicing;
-
- /* op uses shared memory */
- if (orangefs_get_bufmap_init() == 0) {
- /*
- * This operation uses the shared memory system AND
- * the system is not yet ready. This situation occurs
- * when the client-core is restarted AND there were
- * operations waiting to be processed or were already
- * in process.
- */
- gossip_debug(GOSSIP_WAIT_DEBUG,
- "uses_shared_memory is true.\n");
- gossip_debug(GOSSIP_WAIT_DEBUG,
- "Client core in-service status(%d).\n",
- is_daemon_in_service());
- gossip_debug(GOSSIP_WAIT_DEBUG, "bufmap_init:%d.\n",
- orangefs_get_bufmap_init());
- gossip_debug(GOSSIP_WAIT_DEBUG,
- "operation's status is 0x%0x.\n",
- op->op_state);
-
- /*
- * let process sleep for a few seconds so shared
- * memory system can be initialized.
- */
- prepare_to_wait(&orangefs_bufmap_init_waitq,
- &wait_entry,
- TASK_INTERRUPTIBLE);
-
- /*
- * Wait for orangefs_bufmap_initialize() to wake me up
- * within the allotted time.
- */
- ret = schedule_timeout(MSECS_TO_JIFFIES
- (1000 * ORANGEFS_BUFMAP_WAIT_TIMEOUT_SECS));
-
- gossip_debug(GOSSIP_WAIT_DEBUG,
- "Value returned from schedule_timeout:"
- "%d.\n",
- ret);
- gossip_debug(GOSSIP_WAIT_DEBUG,
- "Is shared memory available? (%d).\n",
- orangefs_get_bufmap_init());
-
- finish_wait(&orangefs_bufmap_init_waitq, &wait_entry);
-
- if (orangefs_get_bufmap_init() == 0) {
- gossip_err("%s:The shared memory system has not started in %d seconds after the client core restarted. Aborting user's request(%s).\n",
- __func__,
- ORANGEFS_BUFMAP_WAIT_TIMEOUT_SECS,
- get_opname_string(op));
- return -EIO;
- }
-
- /*
- * Return to the calling function and re-populate a
- * shared memory buffer.
- */
- return -EAGAIN;
- }
}
+out:
gossip_debug(GOSSIP_WAIT_DEBUG,
- "orangefs: service_operation %s returning: %d for %p.\n",
+ "%s: %s returning: %d for %p.\n",
+ __func__,
op_name,
ret,
op);
return ret;
}
-static void orangefs_clean_up_interrupted_operation(struct orangefs_kernel_op_s *op)
+/* This can get called on an I/O op if it had a bad service_operation. */
+bool orangefs_cancel_op_in_progress(struct orangefs_kernel_op_s *op)
+{
+ u64 tag = op->tag;
+ if (!op_state_in_progress(op))
+ return false;
+
+ op->slot_to_free = op->upcall.req.io.buf_index;
+ memset(&op->upcall, 0, sizeof(op->upcall));
+ memset(&op->downcall, 0, sizeof(op->downcall));
+ op->upcall.type = ORANGEFS_VFS_OP_CANCEL;
+ op->upcall.req.cancel.op_tag = tag;
+ op->downcall.type = ORANGEFS_VFS_OP_INVALID;
+ op->downcall.status = -1;
+ orangefs_new_tag(op);
+
+ spin_lock(&orangefs_request_list_lock);
+ /* orangefs_request_list_lock is enough of a barrier here */
+ if (!__is_daemon_in_service()) {
+ spin_unlock(&orangefs_request_list_lock);
+ return false;
+ }
+ spin_lock(&op->lock);
+ set_op_state_waiting(op);
+ gossip_debug(GOSSIP_DEV_DEBUG,
+ "%s: op:%s: op_state:%d: process:%s:\n",
+ __func__,
+ get_opname_string(op),
+ op->op_state,
+ current->comm);
+ list_add(&op->list, &orangefs_request_list);
+ spin_unlock(&op->lock);
+ spin_unlock(&orangefs_request_list_lock);
+
+ gossip_debug(GOSSIP_WAIT_DEBUG,
+ "Attempting ORANGEFS operation cancellation of tag %llu\n",
+ llu(tag));
+ return true;
+}
+
+/*
+ * Change an op to the "given up" state and remove it from its list.
+ */
+static void
+ orangefs_clean_up_interrupted_operation(struct orangefs_kernel_op_s *op)
{
/*
* handle interrupted cases depending on what state we were in when
- * the interruption is detected. there is a coarse grained lock
- * across the operation.
+ * the interruption is detected.
*
- * NOTE: be sure not to reverse lock ordering by locking an op lock
- * while holding the request_list lock. Here, we first lock the op
- * and then lock the appropriate list.
+ * Called with op->lock held.
*/
- if (!op) {
- gossip_debug(GOSSIP_WAIT_DEBUG,
- "%s: op is null, ignoring\n",
- __func__);
- return;
- }
/*
- * one more sanity check, make sure it's in one of the possible states
- * or don't try to cancel it
+ * List manipulation code elsewhere will ignore ops that
+ * have been given up upon.
*/
- if (!(op_state_waiting(op) ||
- op_state_in_progress(op) ||
- op_state_serviced(op) ||
- op_state_purged(op))) {
- gossip_debug(GOSSIP_WAIT_DEBUG,
- "%s: op %p not in a valid state (%0x), "
- "ignoring\n",
- __func__,
- op,
- op->op_state);
- return;
- }
-
- spin_lock(&op->lock);
op->op_state |= OP_VFS_STATE_GIVEN_UP;
- if (op_state_waiting(op)) {
+ if (list_empty(&op->list)) {
+ /* caught copying to/from daemon */
+ BUG_ON(op_state_serviced(op));
+ spin_unlock(&op->lock);
+ wait_for_completion(&op->waitq);
+ } else if (op_state_waiting(op)) {
/*
* upcall hasn't been read; remove op from upcall request
* list.
*/
spin_unlock(&op->lock);
spin_lock(&orangefs_request_list_lock);
- list_del(&op->list);
+ list_del_init(&op->list);
spin_unlock(&orangefs_request_list_lock);
gossip_debug(GOSSIP_WAIT_DEBUG,
"Interrupted: Removed op %p from request_list\n",
/* op must be removed from the in progress htable */
spin_unlock(&op->lock);
spin_lock(&htable_ops_in_progress_lock);
- list_del(&op->list);
+ list_del_init(&op->list);
spin_unlock(&htable_ops_in_progress_lock);
gossip_debug(GOSSIP_WAIT_DEBUG,
"Interrupted: Removed op %p"
" from htable_ops_in_progress\n",
op);
- } else if (!op_state_serviced(op)) {
+ } else {
spin_unlock(&op->lock);
gossip_err("interrupted operation is in a weird state 0x%x\n",
op->op_state);
- } else {
- /*
- * It is not intended for execution to flow here,
- * but having this unlock here makes sparse happy.
- */
- gossip_err("%s: can't get here.\n", __func__);
- spin_unlock(&op->lock);
}
+ reinit_completion(&op->waitq);
}
/*
- * sleeps on waitqueue waiting for matching downcall.
- * if client-core finishes servicing, then we are good to go.
+ * Sleeps on waitqueue waiting for matching downcall.
+ * If client-core finishes servicing, then we are good to go.
* else if client-core exits, we get woken up here, and retry with a timeout
*
- * Post when this call returns to the caller, the specified op will no
- * longer be on any list or htable.
+ * When this call returns to the caller, the specified op will no
+ * longer be in either the in_progress hash table or on the request list.
*
* Returns 0 on success and -errno on failure
* Errors are:
* EINTR/EIO/ETIMEDOUT indicating we are done trying to service this
* operation since client-core seems to be exiting too often
* or if we were interrupted.
+ *
+ * Returns with op->lock taken.
*/
-static int wait_for_matching_downcall(struct orangefs_kernel_op_s *op)
+static int wait_for_matching_downcall(struct orangefs_kernel_op_s *op,
+ long timeout,
+ bool interruptible)
{
- int ret = -EINVAL;
- DEFINE_WAIT(wait_entry);
+ long n;
- while (1) {
- spin_lock(&op->lock);
- prepare_to_wait(&op->waitq, &wait_entry, TASK_INTERRUPTIBLE);
- if (op_state_serviced(op)) {
- spin_unlock(&op->lock);
- ret = 0;
- break;
- }
- spin_unlock(&op->lock);
+ /*
+ * There's a "schedule_timeout" inside of these wait
+ * primitives, during which the op is out of the hands of the
+ * user process that needs something done and is being
+ * manipulated by the client-core process.
+ */
+ if (interruptible)
+ n = wait_for_completion_interruptible_timeout(&op->waitq,
+ timeout);
+ else
+ n = wait_for_completion_killable_timeout(&op->waitq, timeout);
- if (!signal_pending(current)) {
- /*
- * if this was our first attempt and client-core
- * has not purged our operation, we are happy to
- * simply wait
- */
- spin_lock(&op->lock);
- if (op->attempts == 0 && !op_state_purged(op)) {
- spin_unlock(&op->lock);
- schedule();
- } else {
- spin_unlock(&op->lock);
- /*
- * subsequent attempts, we retry exactly once
- * with timeouts
- */
- if (!schedule_timeout(MSECS_TO_JIFFIES
- (1000 * op_timeout_secs))) {
- gossip_debug(GOSSIP_WAIT_DEBUG,
- "*** %s:"
- " operation timed out (tag"
- " %llu, %p, att %d)\n",
- __func__,
- llu(op->tag),
- op,
- op->attempts);
- ret = -ETIMEDOUT;
- orangefs_clean_up_interrupted_operation
- (op);
- break;
- }
- }
- spin_lock(&op->lock);
- op->attempts++;
- /*
- * if the operation was purged in the meantime, it
- * is better to requeue it afresh but ensure that
- * we have not been purged repeatedly. This could
- * happen if client-core crashes when an op
- * is being serviced, so we requeue the op, client
- * core crashes again so we requeue the op, client
- * core starts, and so on...
- */
- if (op_state_purged(op)) {
- ret = (op->attempts < ORANGEFS_PURGE_RETRY_COUNT) ?
- -EAGAIN :
- -EIO;
- spin_unlock(&op->lock);
- gossip_debug(GOSSIP_WAIT_DEBUG,
- "*** %s:"
- " operation purged (tag "
- "%llu, %p, att %d)\n",
- __func__,
- llu(op->tag),
- op,
- op->attempts);
- orangefs_clean_up_interrupted_operation(op);
- break;
- }
- spin_unlock(&op->lock);
- continue;
- }
+ spin_lock(&op->lock);
+ if (op_state_serviced(op))
+ return 0;
+
+ if (unlikely(n < 0)) {
gossip_debug(GOSSIP_WAIT_DEBUG,
- "*** %s:"
- " operation interrupted by a signal (tag "
- "%llu, op %p)\n",
+ "%s: operation interrupted, tag %llu, %p\n",
__func__,
llu(op->tag),
op);
- orangefs_clean_up_interrupted_operation(op);
- ret = -EINTR;
- break;
+ return -EINTR;
}
-
- spin_lock(&op->lock);
- finish_wait(&op->waitq, &wait_entry);
- spin_unlock(&op->lock);
-
- return ret;
-}
-
-/*
- * similar to wait_for_matching_downcall(), but used in the special case
- * of I/O cancellations.
- *
- * Note we need a special wait function because if this is called we already
- * know that a signal is pending in current and need to service the
- * cancellation upcall anyway. the only way to exit this is to either
- * timeout or have the cancellation be serviced properly.
- */
-static int wait_for_cancellation_downcall(struct orangefs_kernel_op_s *op)
-{
- int ret = -EINVAL;
- DEFINE_WAIT(wait_entry);
-
- while (1) {
- spin_lock(&op->lock);
- prepare_to_wait(&op->waitq, &wait_entry, TASK_INTERRUPTIBLE);
- if (op_state_serviced(op)) {
- gossip_debug(GOSSIP_WAIT_DEBUG,
- "%s:op-state is SERVICED.\n",
- __func__);
- spin_unlock(&op->lock);
- ret = 0;
- break;
- }
- spin_unlock(&op->lock);
-
- if (signal_pending(current)) {
- gossip_debug(GOSSIP_WAIT_DEBUG,
- "%s:operation interrupted by a signal (tag"
- " %llu, op %p)\n",
- __func__,
- llu(op->tag),
- op);
- orangefs_clean_up_interrupted_operation(op);
- ret = -EINTR;
- break;
- }
-
- gossip_debug(GOSSIP_WAIT_DEBUG,
- "%s:About to call schedule_timeout.\n",
- __func__);
- ret =
- schedule_timeout(MSECS_TO_JIFFIES(1000 * op_timeout_secs));
-
+ if (op_state_purged(op)) {
gossip_debug(GOSSIP_WAIT_DEBUG,
- "%s:Value returned from schedule_timeout(%d).\n",
+ "%s: operation purged, tag %llu, %p, %d\n",
__func__,
- ret);
- if (!ret) {
- gossip_debug(GOSSIP_WAIT_DEBUG,
- "%s:*** operation timed out: %p\n",
- __func__,
- op);
- orangefs_clean_up_interrupted_operation(op);
- ret = -ETIMEDOUT;
- break;
- }
-
- gossip_debug(GOSSIP_WAIT_DEBUG,
- "%s:Breaking out of loop, regardless of value returned by schedule_timeout.\n",
- __func__);
- ret = -ETIMEDOUT;
- break;
+ llu(op->tag),
+ op,
+ op->attempts);
+ return (op->attempts < ORANGEFS_PURGE_RETRY_COUNT) ?
+ -EAGAIN :
+ -EIO;
}
-
- spin_lock(&op->lock);
- finish_wait(&op->waitq, &wait_entry);
- spin_unlock(&op->lock);
-
+ /* must have timed out, then... */
gossip_debug(GOSSIP_WAIT_DEBUG,
- "%s:returning ret(%d)\n",
+ "%s: operation timed out, tag %llu, %p, %d)\n",
__func__,
- ret);
-
- return ret;
+ llu(op->tag),
+ op,
+ op->attempts);
+ return -ETIMEDOUT;
}