Re: [PATCH] cleanup of ipc/msg.c

From: Manfred Spraul
Date: Sun Jul 04 2004 - 06:51:18 EST

Hi,

Now I've really attached the lockless receive patch. The previous patch [that I forgot to attach] was buggy, attached is an updated and tested patch:

Description:
- General cleanup of sys_msgrcv and sys_msgsnd: the function were too convoluted.
- Enable lockless receive, update comments.
- Use ipc_getref for sys_msgsnd(), it's better than rechecking that the msqid is still valid.

Signed-Off-By: Manfred Spraul <manfred@xxxxxxxxxxxxxxxx>

--- 2.6/ipc/msg.c 2004-07-04 13:21:52.783832608 +0200
+++ build-2.6/ipc/msg.c 2004-07-04 11:18:53.785610848 +0200
@@ -163,8 +163,10 @@

msr = list_entry(tmp,struct msg_receiver,r_list);
tmp = tmp->next;
- msr->r_msg = ERR_PTR(res);
+ msr->r_msg = NULL;
wake_up_process(msr->r_tsk);
+ smp_mb();
+ msr->r_msg = ERR_PTR(res);
}
}
/*
@@ -525,13 +527,17 @@
!security_msg_queue_msgrcv(msq, msg, msr->r_tsk, msr->r_msgtype, msr->r_mode)) {
list_del(&msr->r_list);
if(msr->r_maxsize < msg->m_ts) {
- msr->r_msg = ERR_PTR(-E2BIG);
+ msr->r_msg = NULL;
wake_up_process(msr->r_tsk);
+ smp_mb();
+ msr->r_msg = ERR_PTR(-E2BIG);
} else {
- msr->r_msg = msg;
+ msr->r_msg = NULL;
msq->q_lrpid = msr->r_tsk->pid;
msq->q_rtime = get_seconds();
wake_up_process(msr->r_tsk);
+ smp_mb();
+ msr->r_msg = msg;
return 1;
}
}
@@ -564,43 +570,49 @@
err=-EINVAL;
if(msq==NULL)
goto out_free;
-retry:
+
err= -EIDRM;
if (msg_checkid(msq,msqid))
goto out_unlock_free;

- err=-EACCES;
- if (ipcperms(&msq->q_perm, S_IWUGO))
- goto out_unlock_free;
+ for (;;) {
+ struct msg_sender s;

- err = security_msg_queue_msgsnd(msq, msg, msgflg);
- if (err)
- goto out_unlock_free;
+ err=-EACCES;
+ if (ipcperms(&msq->q_perm, S_IWUGO))
+ goto out_unlock_free;

- if(msgsz + msq->q_cbytes > msq->q_qbytes ||
- 1 + msq->q_qnum > msq->q_qbytes) {
- struct msg_sender s;
+ err = security_msg_queue_msgsnd(msq, msg, msgflg);
+ if (err)
+ goto out_unlock_free;

+ if(msgsz + msq->q_cbytes <= msq->q_qbytes &&
+ 1 + msq->q_qnum <= msq->q_qbytes) {
+ break;
+ }
+
+ /* queue full, wait: */
if(msgflg&IPC_NOWAIT) {
err=-EAGAIN;
goto out_unlock_free;
}
ss_add(msq, &s);
+ ipc_rcu_getref(msq);
msg_unlock(msq);
schedule();
- current->state= TASK_RUNNING;

- msq = msg_lock(msqid);
- err = -EIDRM;
- if(msq==NULL)
- goto out_free;
+ ipc_lock_by_ptr(&msq->q_perm);
+ ipc_rcu_putref(msq);
+ if (msq->q_perm.deleted) {
+ err = -EIDRM;
+ goto out_unlock_free;
+ }
ss_del(&s);

if (signal_pending(current)) {
- err=-EINTR;
+ err=-ERESTARTNOHAND;
goto out_unlock_free;
}
- goto retry;
}

msq->q_lspid = current->tgid;
@@ -649,10 +661,7 @@
long msgtyp, int msgflg)
{
struct msg_queue *msq;
- struct msg_receiver msr_d;
- struct list_head* tmp;
- struct msg_msg* msg, *found_msg;
- int err;
+ struct msg_msg *msg;
int mode;

if (msqid < 0 || (long) msgsz < 0)
@@ -662,62 +671,57 @@
msq = msg_lock(msqid);
if(msq==NULL)
return -EINVAL;
-retry:
- err = -EIDRM;
+
+ msg = ERR_PTR(-EIDRM);
if (msg_checkid(msq,msqid))
goto out_unlock;

- err=-EACCES;
- if (ipcperms (&msq->q_perm, S_IRUGO))
- goto out_unlock;
+ for (;;) {
+ struct msg_receiver msr_d;
+ struct list_head* tmp;

- tmp = msq->q_messages.next;
- found_msg=NULL;
- while (tmp != &msq->q_messages) {
- msg = list_entry(tmp,struct msg_msg,m_list);
- if(testmsg(msg,msgtyp,mode) &&
- !security_msg_queue_msgrcv(msq, msg, current, msgtyp, mode)) {
- found_msg = msg;
- if(mode == SEARCH_LESSEQUAL && msg->m_type != 1) {
- found_msg=msg;
- msgtyp=msg->m_type-1;
- } else {
- found_msg=msg;
- break;
- }
- }
- tmp = tmp->next;
- }
- if(found_msg) {
- msg=found_msg;
- if ((msgsz < msg->m_ts) && !(msgflg & MSG_NOERROR)) {
- err=-E2BIG;
+ msg = ERR_PTR(-EACCES);
+ if (ipcperms (&msq->q_perm, S_IRUGO))
goto out_unlock;
+
+ msg = ERR_PTR(-EAGAIN);
+ tmp = msq->q_messages.next;
+ while (tmp != &msq->q_messages) {
+ struct msg_msg *walk_msg;
+ walk_msg = list_entry(tmp,struct msg_msg,m_list);
+ if(testmsg(walk_msg,msgtyp,mode) &&
+ !security_msg_queue_msgrcv(msq, walk_msg, current, msgtyp, mode)) {
+ msg = walk_msg;
+ if(mode == SEARCH_LESSEQUAL && walk_msg->m_type != 1) {
+ msg=walk_msg;
+ msgtyp=walk_msg->m_type-1;
+ } else {
+ msg=walk_msg;
+ break;
+ }
+ }
+ tmp = tmp->next;
}
- list_del(&msg->m_list);
- msq->q_qnum--;
- msq->q_rtime = get_seconds();
- msq->q_lrpid = current->tgid;
- msq->q_cbytes -= msg->m_ts;
- atomic_sub(msg->m_ts,&msg_bytes);
- atomic_dec(&msg_hdrs);
- ss_wakeup(&msq->q_senders,0);
- msg_unlock(msq);
-out_success:
- msgsz = (msgsz > msg->m_ts) ? msg->m_ts : msgsz;
- if (put_user (msg->m_type, &msgp->mtype) ||
- store_msg(msgp->mtext, msg, msgsz)) {
- msgsz = -EFAULT;
+ if(!IS_ERR(msg)) {
+ /* Found a suitable message. Unlink it from the queue. */
+ if ((msgsz < msg->m_ts) && !(msgflg & MSG_NOERROR)) {
+ msg = ERR_PTR(-E2BIG);
+ goto out_unlock;
+ }
+ list_del(&msg->m_list);
+ msq->q_qnum--;
+ msq->q_rtime = get_seconds();
+ msq->q_lrpid = current->tgid;
+ msq->q_cbytes -= msg->m_ts;
+ atomic_sub(msg->m_ts,&msg_bytes);
+ atomic_dec(&msg_hdrs);
+ ss_wakeup(&msq->q_senders,0);
+ msg_unlock(msq);
+ break;
}
- free_msg(msg);
- return msgsz;
- } else
- {
- /* no message waiting. Prepare for pipelined
- * receive.
- */
+ /* No message waiting. Wait for a message */
if (msgflg & IPC_NOWAIT) {
- err=-ENOMSG;
+ msg = ERR_PTR(-ENOMSG);
goto out_unlock;
}
list_add_tail(&msr_d.r_list,&msq->q_receivers);
@@ -727,52 +731,76 @@
if(msgflg & MSG_NOERROR)
msr_d.r_maxsize = INT_MAX;
else
- msr_d.r_maxsize = msgsz;
+ msr_d.r_maxsize = msgsz;
msr_d.r_msg = ERR_PTR(-EAGAIN);
current->state = TASK_INTERRUPTIBLE;
msg_unlock(msq);

schedule();

- /*
- * The below optimisation is buggy. A sleeping thread that is
- * woken up checks if it got a message and if so, copies it to
- * userspace and just returns without taking any locks.
- * But this return to user space can be faster than the message
- * send, and if the receiver immediately exits the
- * wake_up_process performed by the sender will oops.
+ /* Lockless receive, part 1:
+ * Disable preemption. We don't hold a reference to the queue
+ * and getting a reference would defeat the idea of a lockless
+ * operation, thus the code relies on rcu to guarantee the
+ * existance of msq:
+ * Prior to destruction, expunge_all(-EIRDM) changes r_msg.
+ * Thus if r_msg is -EAGAIN, then the queue not yet destroyed.
+ * rcu_read_lock() prevents preemption between reading r_msg
+ * and the spin_lock() inside ipc_lock_by_ptr().
+ */
+ rcu_read_lock();
+
+ /* Lockless receive, part 2:
+ * Wait until pipelined_send or expunge_all are outside of
+ * wake_up_process(). There is a race with exit(), see
+ * ipc/mqueue.c for the details.
*/
-#if 0
msg = (struct msg_msg*) msr_d.r_msg;
- if(!IS_ERR(msg))
- goto out_success;
-#endif
+ while (msg == NULL) {
+ cpu_relax();
+ msg = (struct msg_msg*) msr_d.r_msg;
+ }

- msq = msg_lock(msqid);
+ /* Lockless receive, part 3:
+ * If there is a message or an error then accept it without
+ * locking.
+ */
+ if(msg != ERR_PTR(-EAGAIN)) {
+ rcu_read_unlock();
+ break;
+ }
+
+ /* Lockless receive, part 3:
+ * Acquire the queue spinlock.
+ */
+ ipc_lock_by_ptr(&msq->q_perm);
+ rcu_read_unlock();
+
+ /* Lockless receive, part 4:
+ * Repeat test after acquiring the spinlock.
+ */
msg = (struct msg_msg*)msr_d.r_msg;
- if(!IS_ERR(msg)) {
- /* our message arived while we waited for
- * the spinlock. Process it.
- */
- if(msq)
- msg_unlock(msq);
- goto out_success;
- }
- err = PTR_ERR(msg);
- if(err == -EAGAIN) {
- if(!msq)
- BUG();
- list_del(&msr_d.r_list);
- if (signal_pending(current))
- err=-EINTR;
- else
- goto retry;
+ if(msg != ERR_PTR(-EAGAIN))
+ goto out_unlock;
+
+ list_del(&msr_d.r_list);
+ if (signal_pending(current)) {
+ msg = ERR_PTR(-ERESTARTNOHAND);
+out_unlock:
+ msg_unlock(msq);
+ break;
}
}
-out_unlock:
- if(msq)
- msg_unlock(msq);
- return err;
+ if (IS_ERR(msg))
+ return PTR_ERR(msg);
+
+ msgsz = (msgsz > msg->m_ts) ? msg->m_ts : msgsz;
+ if (put_user (msg->m_type, &msgp->mtype) ||
+ store_msg(msgp->mtext, msg, msgsz)) {
+ msgsz = -EFAULT;
+ }
+ free_msg(msg);
+ return msgsz;
}

#ifdef CONFIG_PROC_FS