[PATCH 9/11] LTTng-core 0.5.111 : Relay+DebugFS

From: Mathieu Desnoyers
Date: Sat Sep 16 2006 - 03:57:16 EST


9 - User space tracing trough new system calls
patch-2.6.17-lttng-core-0.5.111-userspace-tracing.diff

OpenPGP public key: http://krystal.dyndns.org:8080/key/compudj.gpg
Key fingerprint: 8CD5 52C3 8E3C 4140 715F BA06 3F25 A8FE 3BAE 9A68
--- a/include/linux/sched.h
+++ b/include/linux/sched.h
@@ -39,6 +39,8 @@ #include <linux/futex.h>

#include <linux/auxvec.h> /* For AT_VECTOR_SIZE */

+#include <linux/ltt-facilities.h>
+
struct exec_domain;

/*
@@ -888,6 +890,9 @@ #endif
* cache last used pipe for splice
*/
struct pipe_inode_info *splice_pipe;
+#ifdef CONFIG_LTT_USERSPACE_GENERIC
+ ltt_facility_t ltt_facilities[LTT_FAC_PER_PROCESS];
+#endif //CONFIG_LTT_USERSPACE_GENERIC
};

static inline pid_t process_group(struct task_struct *tsk)
--- a/kernel/sys_ni.c
+++ b/kernel/sys_ni.c
@@ -110,6 +110,8 @@ cond_syscall(sys_vm86old);
cond_syscall(sys_vm86);
cond_syscall(compat_sys_ipc);
cond_syscall(compat_sys_sysctl);
+cond_syscall(sys_ltt_trace_generic);
+cond_syscall(sys_ltt_register_generic);

/* arch-specific weak syscall entries */
cond_syscall(sys_pciconfig_read);
--- /dev/null
+++ b/kernel/ltt-syscall.c
@@ -0,0 +1,175 @@
+/******************************************************************************
+ * ltt-syscall.c
+ *
+ * Mathieu Desnoyers (mathieu.desnoyers@xxxxxxxxxx)
+ * March 2006
+ *
+ * LTT userspace tracing syscalls
+ */
+
+#include <linux/errno.h>
+#include <linux/syscalls.h>
+#include <linux/sched.h>
+#include <linux/ltt-core.h>
+#include <linux/ltt-facilities.h>
+
+#include <asm/uaccess.h>
+
+/* User event logging function */
+static inline int trace_user_event(unsigned int facility_id,
+ unsigned int event_id,
+ void __user *data, size_t data_size, int blocking,
+ int high_priority)
+{
+ int ret = 0;
+ unsigned int index;
+ struct ltt_channel_struct *channel;
+ struct ltt_trace_struct *trace;
+ void *transport_data;
+ void *buffer = NULL;
+ size_t real_to_base = 0; /* buffer allocated on arch_size alignment */
+ size_t *to_base = &real_to_base;
+ size_t real_to = 0;
+ size_t *to = &real_to;
+ size_t real_len = 0;
+ size_t *len = &real_len;
+ size_t reserve_size;
+ size_t slot_size;
+ u64 tsc;
+ size_t before_hdr_pad, after_hdr_pad, header_size;
+ struct user_dbg_data dbg;
+
+ dbg.avail_size = 0;
+ dbg.write = 0;
+ dbg.read = 0;
+
+ if(ltt_traces.num_active_traces == 0) return 0;
+
+ /* Assume that the padding for alignment starts at a
+ * sizeof(void *) address. */
+
+ reserve_size = data_size;
+
+ if(high_priority) index = GET_CHANNEL_INDEX(processes);
+ else index = GET_CHANNEL_INDEX(cpu);
+
+ preempt_disable();
+
+ if(blocking) {
+ /* User space requested blocking mode :
+ * If one of the active traces has free space below a specific
+ * threshold value, we reenable preemption and block. */
+block_test_begin:
+ list_for_each_entry_rcu(trace, &ltt_traces.head, list) {
+ if(!trace->active)
+ continue;
+
+ if (trace->ops->user_blocking(trace, index, data_size,
+ &dbg))
+ goto block_test_begin;
+ }
+ }
+ ltt_nesting[smp_processor_id()]++;
+ list_for_each_entry_rcu(trace, &ltt_traces.head, list) {
+ if(!trace->active) continue;
+ channel = ltt_get_channel_from_index(trace, index);
+ slot_size = 0;
+ buffer = ltt_reserve_slot(trace, channel, &transport_data,
+ reserve_size, &slot_size, &tsc,
+ &before_hdr_pad, &after_hdr_pad, &header_size);
+ if(!buffer) {
+ if(blocking) trace->ops->user_errors(trace,
+ index, data_size, &dbg);
+ continue; /* buffer full */
+ }
+ *to_base = *to = *len = 0;
+ ltt_write_event_header(trace, channel, buffer,
+ facility_id, event_id,
+ reserve_size, before_hdr_pad, tsc);
+ *to_base += before_hdr_pad + after_hdr_pad + header_size;
+ /* Hope the user pages are not swapped out. In the rare case
+ * where it is, the slot will be zeroed and EFAULT returned. */
+ if(__copy_from_user_inatomic(buffer+*to_base+*to, data,
+ data_size)) {
+ /* Data is garbage in the slot */
+ ret = -EFAULT;
+ }
+ ltt_commit_slot(channel, &transport_data, buffer, slot_size);
+ if(ret != 0) break;
+ }
+ ltt_nesting[smp_processor_id()]--;
+ preempt_enable_no_resched();
+ return ret;
+}
+
+asmlinkage long sys_ltt_trace_generic(unsigned int facility_id,
+ unsigned int event_id,
+ void __user *data,
+ size_t data_size,
+ int blocking,
+ int high_priority)
+{
+ if(!ltt_facility_user_access_ok(facility_id)) return -EPERM;
+ if(!access_ok(VERIFY_READ, data, data_size))
+ return -EFAULT;
+
+ return trace_user_event(facility_id, event_id, data, data_size,
+ blocking, high_priority);
+}
+
+asmlinkage long sys_ltt_register_generic(unsigned int __user *facility_id,
+ const struct user_facility_info __user *info)
+{
+ struct user_facility_info kinfo;
+ int fac_id;
+ unsigned int i;
+
+ /* Check if the process has already registered the maximum number of
+ * allowed facilities */
+ if(current->ltt_facilities[LTT_FAC_PER_PROCESS-1] != 0)
+ return -EPERM;
+
+ if(copy_from_user(&kinfo, info, sizeof(*info))) return -EFAULT;
+
+ /* Verify if facility is already registered */
+ printk(KERN_DEBUG "LTT register generic for %s\n", kinfo.name);
+ fac_id = ltt_facility_verify(LTT_FACILITY_TYPE_USER,
+ kinfo.name,
+ kinfo.num_events,
+ kinfo.checksum,
+ kinfo.int_size,
+ kinfo.long_size,
+ kinfo.pointer_size,
+ kinfo.size_t_size,
+ kinfo.alignment);
+
+ printk(KERN_DEBUG "LTT verify return %d\n", fac_id);
+ if(fac_id > 0) goto found;
+
+ fac_id = ltt_facility_register(LTT_FACILITY_TYPE_USER,
+ kinfo.name,
+ kinfo.num_events,
+ kinfo.checksum,
+ kinfo.int_size,
+ kinfo.long_size,
+ kinfo.pointer_size,
+ kinfo.size_t_size,
+ kinfo.alignment);
+
+ printk(KERN_DEBUG "LTT register return %d\n", fac_id);
+ if(fac_id == 0) return -EPERM;
+ if(fac_id < 0) return fac_id; /* Error */
+found:
+ spin_lock(&current->group_leader->proc_lock);
+ for(i=0; i<LTT_FAC_PER_PROCESS; i++) {
+ if(current->group_leader->ltt_facilities[i] == 0) {
+ current->group_leader->ltt_facilities[i] =
+ (ltt_facility_t)fac_id;
+ break;
+ }
+ }
+ spin_unlock(&current->group_leader->proc_lock);
+ /* Write facility_id */
+ put_user((unsigned int)fac_id, facility_id);
+ return 0;
+}
--- a/kernel/exit.c
+++ b/kernel/exit.c
@@ -4,6 +4,7 @@
* Copyright (C) 1991, 1992 Linus Torvalds
*/

+#include <linux/ltt-facilities.h>
#include <linux/config.h>
#include <linux/mm.h>
#include <linux/slab.h>
@@ -36,6 +37,7 @@ #include <linux/futex.h>
#include <linux/compat.h>
#include <linux/pipe_fs_i.h>
#include <linux/audit.h> /* for audit_free() */
+#include <linux/ltt-core.h>

#include <asm/uaccess.h>
#include <asm/unistd.h>
@@ -59,6 +61,16 @@ static void __unhash_process(struct task

list_del_rcu(&p->tasks);
__get_cpu_var(process_counts)--;
+#ifdef CONFIG_LTT_USERSPACE_GENERIC
+ {
+ int i;
+ for(i=0; i<LTT_FAC_PER_PROCESS; i++) {
+ if(p->ltt_facilities[i] == 0) break;
+ WARN_ON(ltt_facility_unregister(
+ p->ltt_facilities[i]));
+ }
+ }
+#endif //CONFIG_LTT_USERSPACE_GENERIC
}
list_del_rcu(&p->thread_group);
remove_parent(p);
--- a/kernel/fork.c
+++ b/kernel/fork.c
@@ -44,6 +44,8 @@ #include <linux/profile.h>
#include <linux/rmap.h>
#include <linux/acct.h>
#include <linux/cn_proc.h>
+#include <linux/ltt-core.h>
+#include <linux/ltt-facilities.h>

#include <asm/pgtable.h>
#include <asm/pgalloc.h>
@@ -1095,6 +1097,20 @@ #endif

p->parent_exec_id = p->self_exec_id;

+#ifdef CONFIG_LTT_USERSPACE_GENERIC
+ if (clone_flags & CLONE_THREAD)
+ memset(p->ltt_facilities, 0, sizeof(p->ltt_facilities));
+ else {
+ int i;
+ for(i=0; i<LTT_FAC_PER_PROCESS; i++) {
+ p->ltt_facilities[i] = current->ltt_facilities[i];
+ if(p->ltt_facilities[i] != 0)
+ ltt_facility_ref(p->ltt_facilities[i]);
+ }
+ }
+
+#endif //CONFIG_LTT_USERSPACE_GENERIC
+
/* ok, now we should be set up.. */
p->exit_signal = (clone_flags & CLONE_THREAD) ? -1 : (clone_flags & CSIGNAL);
p->pdeath_signal = 0;
@@ -1151,7 +1167,7 @@ #endif
spin_unlock(&current->sighand->siglock);
write_unlock_irq(&tasklist_lock);
retval = -ERESTARTNOINTR;
- goto bad_fork_cleanup_namespace;
+ goto bad_fork_cleanup_ltt_facilities;
}

if (clone_flags & CLONE_THREAD) {
@@ -1164,7 +1180,7 @@ #endif
spin_unlock(&current->sighand->siglock);
write_unlock_irq(&tasklist_lock);
retval = -EAGAIN;
- goto bad_fork_cleanup_namespace;
+ goto bad_fork_cleanup_ltt_facilities;
}

p->group_leader = current->group_leader;
@@ -1216,6 +1232,17 @@ #endif
proc_fork_connector(p);
return p;

+bad_fork_cleanup_ltt_facilities:
+#ifdef CONFIG_LTT_USERSPACE_GENERIC
+ {
+ int i;
+ for(i=0; i<LTT_FAC_PER_PROCESS; i++) {
+ if(p->ltt_facilities[i] == 0) break;
+ WARN_ON(ltt_facility_unregister(
+ p->ltt_facilities[i]));
+ }
+ }
+#endif //CONFIG_LTT_USERSPACE_GENERIC
bad_fork_cleanup_namespace:
exit_namespace(p);
bad_fork_cleanup_keys:
--- a/include/asm-i386/unistd.h
+++ b/include/asm-i386/unistd.h
@@ -322,8 +322,10 @@ #define __NR_splice 313
#define __NR_sync_file_range 314
#define __NR_tee 315
#define __NR_vmsplice 316
+#define __NR_ltt_trace_generic 317
+#define __NR_ltt_register_generic 318

-#define NR_syscalls 317
+#define NR_syscalls 319

/*
* user-visible error numbers are in the range -1 - -128: see
--- a/include/asm-powerpc/unistd.h
+++ b/include/asm-powerpc/unistd.h
@@ -323,8 +323,10 @@ #define __NR_fchmodat 297
#define __NR_faccessat 298
#define __NR_get_robust_list 299
#define __NR_set_robust_list 300
+#define __NR_ltt_trace_generic 301
+#define __NR_ltt_register_generic 302

-#define __NR_syscalls 301
+#define __NR_syscalls 303

#ifdef __KERNEL__
#define __NR__exit __NR_exit
--- a/include/asm-x86_64/ia32_unistd.h
+++ b/include/asm-x86_64/ia32_unistd.h
@@ -317,4 +317,9 @@ #define __NR_ia32_pselect6 308
#define __NR_ia32_ppoll 309
#define __NR_ia32_unshare 310

+/* A few defines seem to have been forgotten by kernel developers.
+ See arch/x86_64/ia32/ia32entry.S and include/asm-i386/unistd.h */
+#define __NR_ia32_ltt_trace_generic 317
+#define __NR_ia32_ltt_register_generic 318
+
#endif /* _ASM_X86_64_IA32_UNISTD_H_ */
--- a/include/asm-x86_64/unistd.h
+++ b/include/asm-x86_64/unistd.h
@@ -617,8 +617,12 @@ #define __NR_sync_file_range 277
__SYSCALL(__NR_sync_file_range, sys_sync_file_range)
#define __NR_vmsplice 278
__SYSCALL(__NR_vmsplice, sys_vmsplice)
+#define __NR_ltt_trace_generic 279
+__SYSCALL(__NR_ltt_trace_generic, sys_ltt_trace_generic)
+#define __NR_ltt_register_generic 280
+__SYSCALL(__NR_ltt_register_generic, sys_ltt_register_generic)

-#define __NR_syscall_max __NR_vmsplice
+#define __NR_syscall_max __NR_ltt_register_generic

#ifndef __NO_STUBS