}
}
out:
+ if (unlikely(op_is_cancel(op)))
+ put_cancel(op);
op_release(op);
return ret;
return in_service;
}
+bool __is_daemon_in_service(void)
+{
+ return open_access_count == 1;
+}
+
static inline long check_ioctl_command(unsigned int command)
{
/* Check for valid ioctl codes */
}
if (ret < 0) {
- /*
- * XXX: needs to be optimized - we only need to cancel if it
- * had been seen by daemon and not completed
- */
- if (!op_state_serviced(new_op)) {
- orangefs_cancel_op_in_progress(new_op->tag);
- } else {
- complete(&new_op->done);
- }
- orangefs_bufmap_put(buffer_index);
- buffer_index = -1;
/*
* don't write an error to syslog on signaled operation
* termination unless we've got debugging turned on, as
type == ORANGEFS_IO_READ ?
"read from" : "write to",
handle, ret);
- goto out;
+ if (orangefs_cancel_op_in_progress(new_op))
+ return ret;
+
+ goto done_copying;
}
/*
return "OP_UNKNOWN?";
}
+void orangefs_new_tag(struct orangefs_kernel_op_s *op)
+{
+ spin_lock(&next_tag_value_lock);
+ op->tag = next_tag_value++;
+ if (next_tag_value == 0)
+ next_tag_value = 100;
+ spin_unlock(&next_tag_value_lock);
+}
+
struct orangefs_kernel_op_s *op_alloc(__s32 type)
{
struct orangefs_kernel_op_s *new_op = NULL;
new_op->downcall.status = -1;
new_op->op_state = OP_VFS_STATE_UNKNOWN;
- new_op->tag = 0;
/* initialize the op specific tag and upcall credentials */
- spin_lock(&next_tag_value_lock);
- new_op->tag = next_tag_value++;
- if (next_tag_value == 0)
- next_tag_value = 100;
- spin_unlock(&next_tag_value_lock);
+ orangefs_new_tag(new_op);
new_op->upcall.type = type;
new_op->attempts = 0;
gossip_debug(GOSSIP_CACHE_DEBUG,
/*
* Set uses_shared_memory to 1 if this operation uses shared memory.
* If true, then a retry on the op must also get a new shared memory
- * buffer and re-populate it.
+ * buffer and re-populate it. Cancels don't care - it only matters
+ * for service_operation() retry logics and cancels don't go through
+ * it anymore.
*/
- int uses_shared_memory;
+ union {
+ int uses_shared_memory;
+ int slot_to_free;
+ };
struct orangefs_upcall_s upcall;
struct orangefs_downcall_s downcall;
op->op_state = OP_VFS_STATE_SERVICED;
wake_up_interruptible(&op->waitq);
}
-static inline void set_op_state_purged(struct orangefs_kernel_op_s *op)
-{
- op->op_state |= OP_VFS_STATE_PURGED;
- wake_up_interruptible(&op->waitq);
-}
#define op_state_waiting(op) ((op)->op_state & OP_VFS_STATE_WAITING)
#define op_state_in_progress(op) ((op)->op_state & OP_VFS_STATE_INPROGR)
#define op_state_serviced(op) ((op)->op_state & OP_VFS_STATE_SERVICED)
#define op_state_purged(op) ((op)->op_state & OP_VFS_STATE_PURGED)
#define op_state_given_up(op) ((op)->op_state & OP_VFS_STATE_GIVEN_UP)
+#define op_is_cancel(op) ((op)->upcall.type == ORANGEFS_VFS_OP_CANCEL)
static inline void get_op(struct orangefs_kernel_op_s *op)
{
}
}
+extern void orangefs_bufmap_put(int);
+static inline void put_cancel(struct orangefs_kernel_op_s *op)
+{
+ orangefs_bufmap_put(op->slot_to_free);
+ op_release(op);
+}
+
+static inline void set_op_state_purged(struct orangefs_kernel_op_s *op)
+{
+ spin_lock(&op->lock);
+ if (unlikely(op_is_cancel(op))) {
+ list_del(&op->list);
+ spin_unlock(&op->lock);
+ put_cancel(op);
+ } else {
+ op->op_state |= OP_VFS_STATE_PURGED;
+ wake_up_interruptible(&op->waitq);
+ spin_unlock(&op->lock);
+ }
+}
+
/* per inode private orangefs info */
struct orangefs_inode_s {
struct orangefs_object_kref refn;
int op_cache_initialize(void);
int op_cache_finalize(void);
struct orangefs_kernel_op_s *op_alloc(__s32 type);
+void orangefs_new_tag(struct orangefs_kernel_op_s *op);
char *get_opname_string(struct orangefs_kernel_op_s *new_op);
int orangefs_inode_cache_initialize(void);
int orangefs_dev_init(void);
void orangefs_dev_cleanup(void);
int is_daemon_in_service(void);
+bool __is_daemon_in_service(void);
int fs_mount_pending(__s32 fsid);
/*
int orangefs_unmount_sb(struct super_block *sb);
-int orangefs_cancel_op_in_progress(__u64 tag);
+bool orangefs_cancel_op_in_progress(struct orangefs_kernel_op_s *op);
static inline __u64 orangefs_convert_time_field(const struct timespec *ts)
{
next,
&htable_ops_in_progress[i],
list) {
- spin_lock(&op->lock);
gossip_debug(GOSSIP_INIT_DEBUG,
"pvfs2-client-core: purging in-progress op tag "
"%llu %s\n",
llu(op->tag),
get_opname_string(op));
set_op_state_purged(op);
- spin_unlock(&op->lock);
}
spin_unlock(&htable_ops_in_progress_lock);
}
return ret;
}
-/*
- * NOTE: on successful cancellation, be sure to return -EINTR, as
- * that's the return value the caller expects
- */
-int orangefs_cancel_op_in_progress(__u64 tag)
-{
- int ret = -EINVAL;
- struct orangefs_kernel_op_s *new_op = NULL;
-
- gossip_debug(GOSSIP_UTILS_DEBUG,
- "orangefs_cancel_op_in_progress called on tag %llu\n",
- llu(tag));
-
- new_op = op_alloc(ORANGEFS_VFS_OP_CANCEL);
- if (!new_op)
- return -ENOMEM;
- new_op->upcall.req.cancel.op_tag = tag;
-
- gossip_debug(GOSSIP_UTILS_DEBUG,
- "Attempting ORANGEFS operation cancellation of tag %llu\n",
- llu(new_op->upcall.req.cancel.op_tag));
-
- ret = service_operation(new_op, "orangefs_cancel", ORANGEFS_OP_CANCELLATION);
-
- gossip_debug(GOSSIP_UTILS_DEBUG,
- "orangefs_cancel_op_in_progress: got return value of %d\n",
- ret);
-
- op_release(new_op);
- return ret;
-}
-
void orangefs_make_bad_inode(struct inode *inode)
{
if (is_root_handle(inode)) {
#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 *);
/*
"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);
}
spin_unlock(&orangefs_request_list_lock);
}
static inline void
-add_op_to_request_list(struct orangefs_kernel_op_s *op)
+__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_op_to_request_list(struct orangefs_kernel_op_s *op)
+{
+ spin_lock(&orangefs_request_list_lock);
+ __add_op_to_request_list(op);
+ spin_unlock(&orangefs_request_list_lock);
+}
+
static inline
void add_priority_op_to_request_list(struct orangefs_kernel_op_s *op)
{
if (flags & ORANGEFS_OP_ASYNC)
return 0;
- 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);
- }
+ ret = wait_for_matching_downcall(op);
if (ret < 0) {
/* failed to get matching downcall */
return ret;
}
+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;
+ }
+ __add_op_to_request_list(op);
+ spin_unlock(&orangefs_request_list_lock);
+
+ gossip_debug(GOSSIP_UTILS_DEBUG,
+ "Attempting ORANGEFS operation cancellation of tag %llu\n",
+ llu(tag));
+ return true;
+}
+
static void orangefs_clean_up_interrupted_operation(struct orangefs_kernel_op_s *op)
{
/*
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;
- }
-
- 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__);
- spin_unlock(&op->lock);
- ret = schedule_timeout(op_timeout_secs * HZ);
-
- gossip_debug(GOSSIP_WAIT_DEBUG,
- "%s:Value returned from schedule_timeout(%d).\n",
- __func__,
- ret);
- if (!ret) {
- gossip_debug(GOSSIP_WAIT_DEBUG,
- "%s:*** operation timed out: %p\n",
- __func__,
- op);
- spin_lock(&op->lock);
- 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;
- }
-
- spin_lock(&op->lock);
- finish_wait(&op->waitq, &wait_entry);
- spin_unlock(&op->lock);
-
- gossip_debug(GOSSIP_WAIT_DEBUG,
- "%s:returning ret(%d)\n",
- __func__,
- ret);
-
- return ret;
-}