ln -s `pwd` $LINUX/fs/lustre
fi
+ # -------- linux objects (for 2.6) --
+ AC_MSG_CHECKING([for Linux objects dir])
+ AC_ARG_WITH([linux-obj],
+ AC_HELP_STRING([--with-linux-obj=path],
+ [set path to Linux objects dir (default=\$LINUX)]),
+ [LINUX_OBJ=$with_linux_obj],
+ [LINUX_OBJ=$LINUX])
+ AC_MSG_RESULT([$LINUX_OBJ])
+ AC_SUBST(LINUX_OBJ)
+
# -------- check for .confg --------
AC_ARG_WITH([linux-config],
[AC_HELP_STRING([--with-linux-config=path],
- [set path to Linux .conf (default=\$LINUX/.config)])],
+ [set path to Linux .conf (default=\$LINUX_OBJ/.config)])],
[LINUX_CONFIG=$with_linux_config],
- [LINUX_CONFIG=$LINUX/.config])
+ [LINUX_CONFIG=$LINUX_OBJ/.config])
AC_SUBST(LINUX_CONFIG)
AC_CHECK_FILE([/boot/kernel.h],
AC_DEFUN([LUSTRE_MODULE_COMPILE_IFELSE],
[m4_ifvaln([$1], [LUSTRE_MODULE_CONFTEST([$1])])dnl
rm -f kernel-tests/conftest.o kernel-tests/conftest.mod.c kernel-tests/conftest.ko
-AS_IF([AC_TRY_COMMAND(cp conftest.c kernel-tests && make [$2] -f $PWD/kernel-tests/Makefile LUSTRE_LINUX_CONFIG=$LINUX_CONFIG -o tmp_include_depends -o scripts -o include/config/MARKER -C $LINUX EXTRA_CFLAGS="-Werror-implicit-function-declaration $EXTRA_KCFLAGS" $ARCH_UM SUBDIRS=$PWD/kernel-tests) >/dev/null && AC_TRY_COMMAND([$3])],
+AS_IF([AC_TRY_COMMAND(cp conftest.c kernel-tests && make [$2] CC="$CC" -f $PWD/kernel-tests/Makefile LUSTRE_LINUX_CONFIG=$LINUX_CONFIG -o tmp_include_depends -o scripts -o include/config/MARKER -C $LINUX_OBJ EXTRA_CFLAGS="-Werror-implicit-function-declaration $EXTRA_KCFLAGS" $ARCH_UM $MODULE_TARGET=$PWD/kernel-tests) >/dev/null && AC_TRY_COMMAND([$3])],
[$4],
[_AC_MSG_LOG_CONFTEST
m4_ifvaln([$5],[$5])dnl])dnl
[AC_MSG_ERROR([Kernel config could not be found. If you are building from a kernel-source rpm consult README.kernel-source])])
# ----------- make dep run? ------------------
- AC_CHECK_FILES([$LINUX/include/linux/autoconf.h
- $LINUX/include/linux/version.h
+ AC_CHECK_FILES([$LINUX_OBJ/include/linux/autoconf.h
+ $LINUX_OBJ/include/linux/version.h
$LINUX/include/linux/config.h],[],
[AC_MSG_ERROR([Run make config in $LINUX.])])
# tarred up the tree and ran make dep etc. in it, then
# version.h gets overwritten with a standard linux one.
- if grep rhconfig $LINUX/include/linux/version.h >/dev/null ; then
+ if grep rhconfig $LINUX_OBJ/include/linux/version.h >/dev/null ; then
# This is a clean kernel-source tree, we need to
# enable extensive workarounds to get this to build
# modules
EXTRA_KCFLAGS="-include $KERNEL_SOURCE_HEADER $EXTRA_KCFLAGS"
fi
- # --- check that we can build modules at all
- AC_MSG_CHECKING([that modules can be built])
- LUSTRE_MODULE_TRY_COMPILE([],[],
- [
- AC_MSG_RESULT([yes])
- ],[
- AC_MSG_RESULT([no])
- AC_MSG_WARN([Consult config.log for details.])
- AC_MSG_WARN([If you are trying to build with a kernel-source rpm, consult README.kernel-source])
- AC_MSG_ERROR([Kernel modules could not be built.])
- ])
-
- # ------------ LINUXRELEASE and moduledir ------------------
+ # ------------ external module support ---------------------
MODULE_TARGET="SUBDIRS"
if test $linux25 = 'yes' ; then
- # ------------ external module support ---------------------
makerule="$PWD/kernel-tests"
AC_MSG_CHECKING([for external module build support])
rm -f kernel-tests/conftest.i
makerule="_dir_$PWD/kernel-tests"
fi
AC_SUBST(MODULE_TARGET)
+
+ # --- check that we can build modules at all
+ AC_MSG_CHECKING([that modules can be built])
+ LUSTRE_MODULE_TRY_COMPILE([],[],
+ [
+ AC_MSG_RESULT([yes])
+ ],[
+ AC_MSG_RESULT([no])
+ AC_MSG_WARN([Consult config.log for details.])
+ AC_MSG_WARN([If you are trying to build with a kernel-source rpm, consult README.kernel-source])
+ AC_MSG_ERROR([Kernel modules could not be built.])
+ ])
+
+ # ------------ LINUXRELEASE and moduledir ------------------
LINUXRELEASE=
rm -f kernel-tests/conftest.i
AC_MSG_CHECKING([for Linux release])
AC_SUBST(GMCPPFLAGS)
AC_SUBST(GMNAL)
+ if test $linux25 = 'no' ; then
#### OpenIB
AC_MSG_CHECKING([if OpenIB kernel headers are present])
OPENIBCPPFLAGS="-I$LINUX/drivers/infiniband/include -DIN_TREE_BUILD"
EXTRA_KCFLAGS="$EXTRA_KCFLAGS_save"
AC_SUBST(OPENIBCPPFLAGS)
AC_SUBST(OPENIBNAL)
+ fi
#### Infinicon IB
AC_MSG_CHECKING([if Infinicon IB kernel headers are present])
#include <linux/fs.h>
#include <linux/version.h>
],[
- #if defined(CONFIG_X86_64) && (LINUX_VERSION_CODE < KERNEL_VERSION(2,4,24))
- #error "x86_64 down_read_trylock broken before 2.4.24"
+ #if (LINUX_VERSION_CODE < KERNEL_VERSION(2,4,24))
+ #error "down_read_trylock broken before 2.4.24"
#endif
struct inode i;
return (char *)&i.i_alloc_sem - (char *)&i;
AC_MSG_RESULT([no])
])
+ # ------------ kallsyms (so software watchdogs produce useful stacks)
+ AC_MSG_CHECKING([if kallsyms is enabled])
+ LUSTRE_MODULE_TRY_COMPILE(
+ [
+ #include <linux/config.h>
+ ],[
+ #ifndef CONFIG_KALLSYMS
+ #error CONFIG_KALLSYMS is not #defined
+ #endif
+ ],[
+ AC_MSG_RESULT([yes])
+ ],[
+ AC_MSG_RESULT([no])
+ if test "x$ARCH_UM" = "x" ; then
+ AC_MSG_ERROR([Lustre requires that CONFIG_KALLSYMS is enabled in your kernel.])
+ fi
+ ])
+
+ # ------------ check for our show_task patch
+ AC_MSG_CHECKING([if kernel exports show_task])
+ have_show_task=0
+ for file in ksyms sched ; do
+ if grep -q "EXPORT_SYMBOL(show_task)" \
+ "$LINUX/kernel/$file.c" 2>/dev/null ; then
+ have_show_task=1
+ break
+ fi
+ done
+ if test x$have_show_task = x1 ; then
+ AC_DEFINE(HAVE_SHOW_TASK, 1, [show_task is exported])
+ AC_MSG_RESULT(yes)
+ else
+ AC_MSG_RESULT(no)
+ fi
+
case $BACKINGFS in
ext3)
# --- Check that ext3 and ext3 xattr are enabled in the kernel
echo "CFLAGS: $CFLAGS"
echo "EXTRA_KCFLAGS: $EXTRA_KCFLAGS"
echo "LLCFLAGS: $LLCFLAGS"
+
+ENABLE_INIT_SCRIPTS=0
+if test x$enable_utils = xyes ; then
+ AC_MSG_CHECKING([whether to install init scripts])
+ # our scripts only work on red hat systems
+ if test -f /etc/init.d/functions -a -f /etc/sysconfig/network ; then
+ ENABLE_INIT_SCRIPTS=1
+ AC_MSG_RESULT([yes])
+ else
+ AC_MSG_RESULT([no])
+ fi
+fi
+AM_CONDITIONAL(INIT_SCRIPTS, test x$ENABLE_INIT_SCRIPTS = "x1")
+AC_SUBST(ENABLE_INIT_SCRIPTS)
#define LBUG_WITH_LOC(file, func, line) \
do { \
CEMERG("LBUG\n"); \
- CERROR("STACK: %s\n", portals_debug_dumpstack()); \
+ portals_debug_dumpstack(NULL); \
portals_debug_dumplog(); \
portals_run_lbug_upcall(file, func, line); \
set_task_state(current, TASK_UNINTERRUPTIBLE); \
#endif /* PORTALS_PROFILING */
/* debug.c */
-extern spinlock_t stack_backtrace_lock;
-
-char *portals_debug_dumpstack(void);
+void portals_debug_dumpstack(struct task_struct *tsk);
void portals_run_upcall(char **argv);
void portals_run_lbug_upcall(char * file, const char *fn, const int line);
void portals_debug_dumplog(void);
(unsigned long)(id)->li_stc.u.e3s.l3s_gen
#endif
-
#define S_IBNAL 0x00400000 /* All IB NALs */
#define S_SM 0x00800000
#define S_ASOBD 0x01000000
-#define S_LMV 0x02000000
-#define S_CMOBD 0x04000000
-#define S_LONAL 0x08000000
-
-/* If you change these values, please keep portals/utils/debug.c
- * up to date! */
+#define S_CONFOBD 0x02000000
+#define S_LMV 0x04000000
+#define S_CMOBD 0x08000000
+#define S_LONAL 0x10000000
+/* If you change these values, please keep these files up to date...
+ * portals/utils/debug.c
+ * utils/lconf
+ */
/* Debugging masks (32 bits, non-overlapping) */
#define D_TRACE 0x00000001 /* ENTRY/EXIT markers */
#define D_READA 0x00400000 /* read-ahead */
#define D_MMAP 0x00800000
#define D_CONFIG 0x01000000
+/* If you change these values, please keep these files up to date...
+ * portals/utils/debug.c
+ * utils/lconf
+ */
+
#ifdef __KERNEL__
# include <linux/sched.h> /* THREAD_SIZE */
#else
#define CDEBUG_MAX_LIMIT 600
#define CDEBUG_LIMIT(cdebug_mask, cdebug_format, a...) \
do { \
- static unsigned long cdebug_next; \
- static int cdebug_count, cdebug_delay = 1; \
+ static unsigned long cdebug_next = 0; \
+ static int cdebug_count = 0, cdebug_delay = 1; \
\
CHECK_STACK(CDEBUG_STACK); \
if (time_after(jiffies, cdebug_next)) { \
if (cdebug_count) { \
portals_debug_msg(DEBUG_SUBSYSTEM, cdebug_mask, \
__FILE__, __FUNCTION__, __LINE__, \
- CDEBUG_STACK, cdebug_format, ## a); \
+ 0, "skipped %d similar messages\n", \
+ cdebug_count); \
cdebug_count = 0; \
} \
if (time_after(jiffies, cdebug_next+(CDEBUG_MAX_LIMIT+10)*HZ))\
#endif
-#define _LIBCFS_H
+#ifdef __KERNEL__
+/* libcfs watchdogs */
+struct lc_watchdog;
+
+/* Just use the default handler (dumplog) */
+#define LC_WATCHDOG_DEFAULT_CB NULL
+
+/* Add a watchdog which fires after "time" milliseconds of delay. You have to
+ * touch it once to enable it. */
+struct lc_watchdog *lc_watchdog_add(int time,
+ void (*cb)(struct lc_watchdog *,
+ struct task_struct *,
+ void *),
+ void *data);
+
+/* Enables a watchdog and resets its timer. */
+void lc_watchdog_touch(struct lc_watchdog *lcw);
+
+/* Disable a watchdog; touch it to restart it. */
+void lc_watchdog_disable(struct lc_watchdog *lcw);
+
+/* Clean up the watchdog */
+void lc_watchdog_delete(struct lc_watchdog *lcw);
+
+/* Dump a debug log */
+void lc_watchdog_dumplog(struct lc_watchdog *lcw,
+ struct task_struct *tsk,
+ void *data);
+#endif /* !__KERNEL__ */
#endif /* _LIBCFS_H */
# define USERMODEHELPER(path, argv, envp) \
call_usermodehelper(path, argv, envp, 1)
# define RECALC_SIGPENDING recalc_sigpending()
+# define CLEAR_SIGPENDING clear_tsk_thread_flag(current, \
+ TIF_SIGPENDING)
# define CURRENT_SECONDS get_seconds()
-# define smp_num_cpus NR_CPUS
+# define smp_num_cpus num_online_cpus()
#elif defined(CONFIG_RH_2_4_20) /* RH 2.4.x */
# define USERMODEHELPER(path, argv, envp) \
call_usermodehelper(path, argv, envp)
# define RECALC_SIGPENDING recalc_sigpending()
+# define CLEAR_SIGPENDING (current->sigpending = 0)
# define CURRENT_SECONDS CURRENT_TIME
-# define kernel_text_address(addr) is_kernel_text_address(addr)
-extern int is_kernel_text_address(unsigned long addr);
-
#else /* 2.4.x */
# define SIGNAL_MASK_LOCK(task, flags) \
# define USERMODEHELPER(path, argv, envp) \
call_usermodehelper(path, argv, envp)
# define RECALC_SIGPENDING recalc_sigpending(current)
+# define CLEAR_SIGPENDING (current->sigpending = 0)
# define CURRENT_SECONDS CURRENT_TIME
-# define kernel_text_address(addr) is_kernel_text_address(addr)
-extern int is_kernel_text_address(unsigned long addr);
-
#endif
#if defined(__arch_um__) && (LINUX_VERSION_CODE < KERNEL_VERSION(2,4,20))
-# define THREAD_NAME(comm, len, fmt, a...) \
- snprintf(comm, len, fmt "|%d", ## a, current->thread.extern_pid)
+#define UML_PID(tsk) ((tsk)->thread.extern_pid)
#elif defined(__arch_um__) && (LINUX_VERSION_CODE < KERNEL_VERSION(2,5,0))
+#define UML_PID(tsk) ((tsk)->thread.mode.tt.extern_pid)
+#else
+#define UML_PID(tsk) ((tsk)->pid)
+#endif
+
+#if defined(__arch_um__) && (LINUX_VERSION_CODE < KERNEL_VERSION(2,5,0))
# define THREAD_NAME(comm, len, fmt, a...) \
- snprintf(comm, len,fmt"|%d", ## a,current->thread.mode.tt.extern_pid)
+ snprintf(comm, len,fmt"|%d", ## a, UML_PID(current))
#else
# define THREAD_NAME(comm, len, fmt, a...) \
snprintf(comm, len, fmt, ## a)
#include <linux/stat.h>
#include <linux/errno.h>
#include <linux/uio.h>
+#include <linux/init.h>
#define DEBUG_SUBSYSTEM S_LONAL
struct iovec *iov;
ptl_kiov_t *kiov;
} klod_iov;
-
} klo_desc_t;
typedef struct
#define KQSW_TX_MAXCONTIG (1<<10) /* largest payload that gets made contiguous on transmit */
#define KQSW_NTXMSGS 8 /* # normal transmit messages */
-#define KQSW_NNBLK_TXMSGS 512 /* # reserved transmit messages if can't block */
+#define KQSW_NNBLK_TXMSGS (PAGE_SIZE == 4096 ? 512 : 256) /* # reserved transmit messages if can't block */ /* avoid qsnet crash b=5291 */
#define KQSW_NRXMSGS_LARGE 64 /* # large receive buffers */
#define KQSW_EP_ENVELOPES_LARGE 256 /* # large ep envelopes */
MODULES = libcfs
-libcfs-objs := debug.o lwt.o module.o proc.o tracefile.o
+libcfs-objs := debug.o lwt.o module.o proc.o tracefile.o watchdog.o
@INCLUDE_RULES@
static int handled_panic; /* to avoid recursive calls to notifiers */
char portals_upcall[1024] = "/usr/lib/lustre/portals_upcall";
-int portals_do_debug_dumplog(void *arg)
+void portals_debug_dumplog_internal(void *arg)
{
- void *journal_info;
-
- kportal_daemonize("");
-
- reparent_to_init();
- journal_info = current->journal_info;
+ void *journal_info = current->journal_info;
current->journal_info = NULL;
snprintf(debug_file_name, sizeof(debug_file_path) - 1,
tracefile_dump_all_pages(debug_file_name);
current->journal_info = journal_info;
+}
+
+int portals_debug_dumplog_thread(void *arg)
+{
+ kportal_daemonize("");
+ reparent_to_init();
+ portals_debug_dumplog_internal(arg);
wake_up(&debug_ctlwq);
return 0;
}
set_current_state(TASK_INTERRUPTIBLE);
add_wait_queue(&debug_ctlwq, &wait);
- rc = kernel_thread(portals_do_debug_dumplog, (void *)(long)current->pid,
+ rc = kernel_thread(portals_debug_dumplog_thread,
+ (void *)(long)current->pid,
CLONE_VM | CLONE_FS | CLONE_FILES);
if (rc < 0)
printk(KERN_ERR "LustreError: cannot start log dump thread: "
portals_nid2str(nal, id.nid, str);
len = strlen(str);
- snprintf(str + len, PTL_NALFMT_SIZE - len, ",%u", id.pid);
+ snprintf(str + len, PTL_NALFMT_SIZE - len, "-%u", id.pid);
return str;
}
#ifdef __KERNEL__
-char stack_backtrace[LUSTRE_TRACE_SIZE];
-spinlock_t stack_backtrace_lock = SPIN_LOCK_UNLOCKED;
-
-#if defined(__arch_um__)
-char *portals_debug_dumpstack(void)
+void portals_debug_dumpstack(struct task_struct *tsk)
{
+#if defined(__arch_um__)
+ if (tsk != NULL)
+ CWARN("stack dump for pid %d (%d) requested; wake up gdb.\n",
+ tsk->pid, UML_PID(tsk));
asm("int $3");
- return "dump stack\n";
-}
-
-#elif defined(__i386__)
-
-#if (LINUX_VERSION_CODE < KERNEL_VERSION(2,5,0))
-extern int lookup_symbol(unsigned long address, char *buf, int buflen);
-const char *kallsyms_lookup(unsigned long addr,
- unsigned long *symbolsize,
- unsigned long *offset,
- char **modname, char *namebuf)
-{
- int rc = lookup_symbol(addr, namebuf, 128);
- if (rc == -ENOSYS)
- return NULL;
- return namebuf;
-}
+#elif defined(HAVE_SHOW_TASK)
+ /* this is exported by lustre kernel version 42 */
+ extern void show_task(struct task_struct *);
+
+ if (tsk == NULL)
+ tsk = current;
+ CWARN("showing stack for process %d\n", tsk->pid);
+ show_task(tsk);
+#else
+ CWARN("can't show stack: kernel doesn't export show_task\n");
#endif
-
-char *portals_debug_dumpstack(void)
-{
- unsigned long esp = current->thread.esp, addr;
- unsigned long *stack = (unsigned long *)&esp;
- char *buf = stack_backtrace, *pbuf = buf;
- int size;
-
- /* User space on another CPU? */
- if ((esp ^ (unsigned long)current) & (PAGE_MASK << 1)){
- buf[0] = '\0';
- goto out;
- }
-
- size = sprintf(pbuf, " Call Trace: ");
- pbuf += size;
- while (((long) stack & (THREAD_SIZE - 1)) != 0) {
- addr = *stack++;
- if (kernel_text_address(addr)) {
- const char *sym_name;
- char *modname, buffer[128];
- unsigned long junk, offset;
-
- sym_name = kallsyms_lookup(addr, &junk, &offset,
- &modname, buffer);
- if (sym_name == NULL) {
- if (buf + LUSTRE_TRACE_SIZE <= pbuf + 12)
- break;
- size = sprintf(pbuf, "[<%08lx>] ", addr);
- } else {
- if (buf + LUSTRE_TRACE_SIZE
- /* fix length + sizeof('\0') */
- <= pbuf + strlen(buffer) + 28 + 1)
- break;
- size = sprintf(pbuf, "([<%08lx>] %s (0x%p)) ",
- addr, buffer, stack - 1);
- }
- pbuf += size;
- }
- }
-out:
- return buf;
-}
-
-#else /* !__arch_um__ && !__i386__ */
-
-char *portals_debug_dumpstack(void)
-{
- char *buf = stack_backtrace;
- buf[0] = '\0';
- return buf;
}
-#endif /* __arch_um__ */
struct task_struct *portals_current(void)
{
CWARN("current task struct is %p\n", current);
return current;
}
-EXPORT_SYMBOL(stack_backtrace_lock);
EXPORT_SYMBOL(portals_debug_dumpstack);
EXPORT_SYMBOL(portals_current);
#endif /* __KERNEL__ */
int i;
int rc;
- CDEBUG(D_IOCTL, "Register NAL %d, handler: %p\n", nal, handler);
+ CDEBUG(D_IOCTL, "Register NAL %x, handler: %p\n", nal, handler);
down(&nal_cmd_sem);
{
struct nal_cmd_handler *cmd;
- CDEBUG(D_IOCTL, "Unregister NAL %d\n", nal);
+ CDEBUG(D_IOCTL, "Unregister NAL %x\n", nal);
down(&nal_cmd_sem);
cmd = libcfs_find_nal_cmd_handler(nal);
int
libcfs_nal_cmd(struct portals_cfg *pcfg)
{
+#if CRAY_PORTALS
+ /* pretend success */
+ RETURN(0);
+#else
struct nal_cmd_handler *cmd;
__u32 nal = pcfg->pcfg_nal;
int rc = -EINVAL;
down(&nal_cmd_sem);
cmd = libcfs_find_nal_cmd_handler(nal);
if (cmd != NULL) {
- CDEBUG(D_IOCTL, "calling handler nal: %d, cmd: %d\n", nal,
+ CDEBUG(D_IOCTL, "calling handler nal: %x, cmd: %d\n", nal,
pcfg->pcfg_command);
rc = cmd->nch_handler(pcfg, cmd->nch_private);
} else {
- CERROR("invalid nal: %d, cmd: %d\n", nal, pcfg->pcfg_command);
+ CERROR("invalid nal: %x, cmd: %d\n", nal, pcfg->pcfg_command);
}
up(&nal_cmd_sem);
RETURN(rc);
+#endif
}
EXPORT_SYMBOL(libcfs_nal_cmd);
break;
}
- CDEBUG (D_IOCTL, "nal command nal %d cmd %d\n", pcfg.pcfg_nal,
+ CDEBUG (D_IOCTL, "nal command nal %x cmd %d\n", pcfg.pcfg_nal,
pcfg.pcfg_command);
err = libcfs_nal_cmd(&pcfg);
ent->write_proc = trace_write_daemon_file;
ent->read_proc = trace_read_daemon_file;
- ent = create_proc_entry("sys/portals/debug_size", 0, NULL);
+ ent = create_proc_entry("sys/portals/debug_mb", 0, NULL);
if (ent == NULL) {
- CERROR("couldn't register debug_size\n");
+ CERROR("couldn't register debug_mb\n");
return -1;
}
- ent->write_proc = trace_write_debug_size;
- ent->read_proc = trace_read_debug_size;
+ ent->write_proc = trace_write_debug_mb;
+ ent->read_proc = trace_read_debug_mb;
return 0;
}
remove_proc_entry("sys/portals/dump_kernel", NULL);
remove_proc_entry("sys/portals/daemon_file", NULL);
- remove_proc_entry("sys/portals/debug_size", NULL);
+ remove_proc_entry("sys/portals/debug_mb", NULL);
#ifdef CONFIG_SYSCTL
if (portals_table_header)
#include <linux/portals_compat25.h>
#include <linux/libcfs.h>
-#define TCD_MAX_PAGES 1280
+#define TCD_MAX_PAGES (5 << (20 - PAGE_SHIFT))
/* XXX move things up to the top, comment */
atomic_t tctl_shutdown;
};
+#define TRACEFILE_SIZE (500 << 20)
static DECLARE_RWSEM(tracefile_sem);
static char *tracefile = NULL;
+static long long tracefile_size = TRACEFILE_SIZE;
static struct tracefiled_ctl trace_tctl;
static DECLARE_MUTEX(trace_thread_sem);
static int thread_running = 0;
/* the kernel should print a message for us. fall back
* to using the last page in the ring buffer. */
goto ring_buffer;
- return NULL;
}
page->index = 0;
page->mapping = (void *)(long)smp_processor_id();
struct trace_cpu_data *tcd;
struct ptldebug_header header;
struct page *page;
- char *debug_buf;
- int known_size, needed, max_nob;
+ char *debug_buf = format;
+ int known_size, needed = 85 /* average message length */, max_nob;
va_list ap;
unsigned long flags;
struct timeval tv;
known_size = sizeof(header) + strlen(file) + strlen(fn) + 2; // nulls
- page = trace_get_page(tcd, known_size + 40); /* slop */
retry:
- if (page == NULL)
+ page = trace_get_page(tcd, needed + known_size);
+ if (page == NULL) {
+ debug_buf = format;
+ if (needed + known_size > PAGE_SIZE)
+ mask |= D_ERROR;
+ needed = strlen(format);
goto out;
+ }
debug_buf = page_address(page) + page->index + known_size;
- va_start(ap, format);
max_nob = PAGE_SIZE - page->index - known_size;
LASSERT(max_nob > 0);
+ va_start(ap, format);
needed = vsnprintf(debug_buf, max_nob, format, ap);
va_end(ap);
- if (needed > max_nob) {
- /* overflow. oh poop. */
- page = trace_get_page(tcd, needed + known_size);
+ if (needed > max_nob) /* overflow. oh poop. */
goto retry;
- }
header.ph_len = known_size + needed;
debug_buf = page_address(page) + page->index;
printk(KERN_EMERG "page->index == %lu in portals_debug_msg\n",
page->index);
+ out:
if ((mask & (D_EMERG | D_ERROR | D_WARNING)) || portal_printk)
print_to_console(&header, mask, debug_buf, needed, file, fn);
- out:
trace_put_tcd(tcd, flags);
}
EXPORT_SYMBOL(portals_debug_msg);
down_write(&tracefile_sem);
- filp = filp_open(filename, O_CREAT|O_EXCL|O_WRONLY, 0600);
+ filp = filp_open(filename, O_CREAT|O_EXCL|O_WRONLY|O_LARGEFILE, 0600);
if (IS_ERR(filp)) {
rc = PTR_ERR(filp);
printk(KERN_ERR "LustreError: can't open %s for dump: rc %d\n",
filp = NULL;
down_read(&tracefile_sem);
if (tracefile != NULL) {
- filp = filp_open(tracefile, O_CREAT|O_RDWR|O_APPEND|O_LARGEFILE,
- 0600);
+ filp = filp_open(tracefile, O_CREAT|O_RDWR|O_LARGEFILE,
+ 0600);
if (IS_ERR(filp)) {
printk("couldn't open %s: %ld\n", tracefile,
PTR_ERR(filp));
hdr->ph_flags |= PH_FLAG_FIRST_RECORD;
list_for_each_safe(pos, tmp, &pc.pc_pages) {
+ static loff_t f_pos;
page = list_entry(pos, struct page, PAGE_LIST_ENTRY);
LASSERT(page->index <= PAGE_SIZE);
LASSERT(page_count(page) > 0);
+ if (f_pos >= tracefile_size)
+ f_pos = 0;
+ else if (f_pos > filp->f_dentry->d_inode->i_size)
+ f_pos = filp->f_dentry->d_inode->i_size;
+
rc = filp->f_op->write(filp, page_address(page),
- page->index, &filp->f_pos);
+ page->index, &f_pos);
if (rc != page->index) {
printk(KERN_WARNING "wanted to write %lu but "
"wrote %d\n", page->index, rc);
tracefile = NULL;
trace_stop_thread();
goto out_sem;
+ } else if (strncmp(name, "size=", 5) == 0) {
+ tracefile_size = simple_strtoul(name + 5, NULL, 0);
+ if (tracefile_size < 10 || tracefile_size > 20480)
+ tracefile_size = TRACEFILE_SIZE;
+ else
+ tracefile_size <<= 20;
+ goto out_sem;
}
if (name[0] != '/') {
tracefile = name;
name = NULL;
+
+ printk(KERN_INFO "Lustre: debug daemon will attempt to start writing "
+ "to %s (%lukB max)\n", tracefile, (long)(tracefile_size >> 10));
+
trace_start_thread();
out_sem:
up_write(&tracefile_sem);
out:
- if (name)
- kfree(name);
+ kfree(name);
return count;
}
return rc;
}
-int trace_write_debug_size(struct file *file, const char *buffer,
- unsigned long count, void *data)
+int trace_write_debug_mb(struct file *file, const char *buffer,
+ unsigned long count, void *data)
{
- char *string;
- int rc, i, max;
-
- string = kmalloc(count + 1, GFP_KERNEL);
- if (string == NULL)
- return -ENOMEM;
+ char string[32];
+ int i;
+ unsigned max;
- if (copy_from_user(string, buffer, count)) {
- rc = -EFAULT;
- goto out;
+ if (count >= sizeof(string)) {
+ printk(KERN_ERR "Lustre: value too large (length %lu bytes)\n",
+ count);
+ return -EOVERFLOW;
}
+ if (copy_from_user(string, buffer, count))
+ return -EFAULT;
+
max = simple_strtoul(string, NULL, 0);
- if (max == 0) {
- rc = -EINVAL;
- goto out;
- }
+ if (max == 0)
+ return -EINVAL;
max /= smp_num_cpus;
- if (max > num_physpages / 5 * 4) {
+ if (max * smp_num_cpus > (num_physpages >> (20 - 2 - PAGE_SHIFT)) / 5) {
printk(KERN_ERR "Lustre: Refusing to set debug buffer size to "
- "%d pages, which is more than 80%% of physical pages "
- "(%lu).\n", max * smp_num_cpus, num_physpages / 5 * 4);
- return count;
+ "%d MB, which is more than 80%% of physical RAM "
+ "(%lu).\n", max * smp_num_cpus,
+ (num_physpages >> (20 - 2 - PAGE_SHIFT)) / 5);
+ return -EINVAL;
}
for (i = 0; i < NR_CPUS; i++) {
struct trace_cpu_data *tcd;
tcd = &trace_data[i].tcd;
- tcd->tcd_max_pages = max;
+ tcd->tcd_max_pages = max << (20 - PAGE_SHIFT);
}
- out:
- kfree(string);
return count;
}
-int trace_read_debug_size(char *page, char **start, off_t off, int count,
- int *eof, void *data)
+int trace_read_debug_mb(char *page, char **start, off_t off, int count,
+ int *eof, void *data)
{
struct trace_cpu_data *tcd;
unsigned long flags;
int rc;
tcd = trace_get_tcd(flags);
- rc = snprintf(page, count, "%lu", tcd->tcd_max_pages);
+ rc = snprintf(page, count, "%lu\n",
+ (tcd->tcd_max_pages >> (20 - PAGE_SHIFT)) * smp_num_cpus);
trace_put_tcd(tcd, flags);
return rc;
unsigned long count, void *data);
int trace_read_daemon_file(char *page, char **start, off_t off, int count,
int *eof, void *data);
-int trace_write_debug_size(struct file *file, const char *buffer,
- unsigned long count, void *data);
-int trace_read_debug_size(char *page, char **start, off_t off, int count,
- int *eof, void *data);
+int trace_write_debug_mb(struct file *file, const char *buffer,
+ unsigned long count, void *data);
+int trace_read_debug_mb(char *page, char **start, off_t off, int count,
+ int *eof, void *data);
int trace_dk(struct file *file, const char *buffer, unsigned long count,
void *data);
--- /dev/null
+/* -*- mode: c; c-basic-offset: 8; indent-tabs-mode: nil; -*-
+ * vim:expandtab:shiftwidth=8:tabstop=8:
+ *
+ * Copyright (C) 2004 Cluster File Systems, Inc.
+ * Author: Jacob Berkman <jacob@clusterfs.com>
+ *
+ * This file is part of Lustre, http://www.lustre.org.
+ *
+ * Lustre is free software; you can redistribute it and/or
+ * modify it under the terms of version 2 of the GNU General Public
+ * License as published by the Free Software Foundation.
+ *
+ * Lustre is distributed in the hope that it will be useful,
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
+ * GNU General Public License for more details.
+ *
+ * You should have received a copy of the GNU General Public License
+ * along with Lustre; if not, write to the Free Software
+ * Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.
+ */
+
+#define DEBUG_SUBSYSTEM S_PORTALS
+
+#include <linux/kp30.h>
+#include <linux/libcfs.h>
+#include <linux/portals_compat25.h>
+
+
+
+struct lc_watchdog {
+ struct timer_list lcw_timer; /* kernel timer */
+ struct list_head lcw_list;
+ struct timeval lcw_last_touched;
+ struct task_struct *lcw_task;
+
+ void (*lcw_callback)(struct lc_watchdog *,
+ struct task_struct *,
+ void *data);
+ void *lcw_data;
+
+ int lcw_pid;
+ int lcw_time; /* time until watchdog fires, in ms */
+
+ enum {
+ LC_WATCHDOG_DISABLED,
+ LC_WATCHDOG_ENABLED,
+ LC_WATCHDOG_EXPIRED
+ } lcw_state;
+};
+
+/*
+ * The dispatcher will complete lcw_start_completion when it starts,
+ * and lcw_stop_completion when it exits.
+ * Wake lcw_event_waitq to signal timer callback dispatches.
+ */
+static struct completion lcw_start_completion;
+static struct completion lcw_stop_completion;
+static wait_queue_head_t lcw_event_waitq;
+
+/*
+ * Set this and wake lcw_event_waitq to stop the dispatcher.
+ */
+enum {
+ LCW_FLAG_STOP = 0
+};
+static unsigned long lcw_flags = 0;
+
+/*
+ * Number of outstanding watchdogs.
+ * When it hits 1, we start the dispatcher.
+ * When it hits 0, we stop the distpatcher.
+ */
+static __u32 lcw_refcount = 0;
+static DECLARE_MUTEX(lcw_refcount_sem);
+
+/*
+ * List of timers that have fired that need their callbacks run by the
+ * dispatcher.
+ */
+static spinlock_t lcw_pending_timers_lock = SPIN_LOCK_UNLOCKED;
+static struct list_head lcw_pending_timers = \
+ LIST_HEAD_INIT(lcw_pending_timers);
+
+static struct task_struct *lcw_lookup_task(struct lc_watchdog *lcw)
+{
+ struct task_struct *tsk;
+ unsigned long flags;
+ ENTRY;
+
+ read_lock_irqsave(&tasklist_lock, flags);
+ tsk = find_task_by_pid(lcw->lcw_pid);
+ read_unlock_irqrestore(&tasklist_lock, flags);
+ if (!tsk) {
+ CWARN("Process %d was not found in the task list; "
+ "watchdog callback may be incomplete\n", lcw->lcw_pid);
+ } else if (tsk != lcw->lcw_task) {
+ tsk = NULL;
+ CWARN("The current process %d did not set the watchdog; "
+ "watchdog callback may be incomplete\n", lcw->lcw_pid);
+ }
+
+ RETURN(tsk);
+}
+
+static void lcw_cb(unsigned long data)
+{
+ struct lc_watchdog *lcw = (struct lc_watchdog *)data;
+ struct task_struct *tsk;
+ unsigned long flags;
+
+ ENTRY;
+
+ if (lcw->lcw_state != LC_WATCHDOG_ENABLED) {
+ EXIT;
+ return;
+ }
+
+ lcw->lcw_state = LC_WATCHDOG_EXPIRED;
+
+ CWARN("Watchdog triggered for pid %d: it was inactive for %dus\n",
+ lcw->lcw_pid, (lcw->lcw_time * 1000) / HZ);
+
+ tsk = lcw_lookup_task(lcw);
+ if (tsk != NULL)
+ portals_debug_dumpstack(tsk);
+
+ spin_lock_irqsave(&lcw_pending_timers_lock, flags);
+ if (list_empty(&lcw->lcw_list)) {
+ list_add(&lcw->lcw_list, &lcw_pending_timers);
+ wake_up(&lcw_event_waitq);
+ }
+ spin_unlock_irqrestore(&lcw_pending_timers_lock, flags);
+
+ EXIT;
+}
+
+static int is_watchdog_fired(void)
+{
+ unsigned long flags;
+ int rc;
+
+ if (test_bit(LCW_FLAG_STOP, &lcw_flags))
+ return 1;
+
+ spin_lock_irqsave(&lcw_pending_timers_lock, flags);
+ rc = !list_empty(&lcw_pending_timers);
+ spin_unlock_irqrestore(&lcw_pending_timers_lock, flags);
+ return rc;
+}
+
+static int lcw_dispatch_main(void *data)
+{
+ int rc = 0;
+ unsigned long flags;
+ struct lc_watchdog *lcw;
+ struct task_struct *tsk;
+
+ ENTRY;
+
+ kportal_daemonize("lc_watchdogd");
+
+ SIGNAL_MASK_LOCK(current, flags);
+ sigfillset(¤t->blocked);
+ RECALC_SIGPENDING;
+ SIGNAL_MASK_UNLOCK(current, flags);
+
+ complete(&lcw_start_completion);
+
+ while (1) {
+ wait_event_interruptible(lcw_event_waitq, is_watchdog_fired());
+ CDEBUG(D_INFO, "Watchdog got woken up...\n");
+ if (test_bit(LCW_FLAG_STOP, &lcw_flags)) {
+ CDEBUG(D_INFO, "LCW_FLAG_STOP was set, shutting down...\n");
+
+ spin_lock_irqsave(&lcw_pending_timers_lock, flags);
+ rc = !list_empty(&lcw_pending_timers);
+ spin_unlock_irqrestore(&lcw_pending_timers_lock, flags);
+ if (rc) {
+ CERROR("pending timers list was not empty at "
+ "time of watchdog dispatch shutdown\n");
+ }
+ break;
+ }
+
+ spin_lock_irqsave(&lcw_pending_timers_lock, flags);
+ while (!list_empty(&lcw_pending_timers)) {
+
+ lcw = list_entry(lcw_pending_timers.next,
+ struct lc_watchdog,
+ lcw_list);
+ list_del_init(&lcw->lcw_list);
+ spin_unlock_irqrestore(&lcw_pending_timers_lock, flags);
+
+ CDEBUG(D_INFO, "found lcw for pid %d\n", lcw->lcw_pid);
+
+ if (lcw->lcw_state != LC_WATCHDOG_DISABLED) {
+ /*
+ * sanity check the task against our
+ * watchdog
+ */
+ tsk = lcw_lookup_task(lcw);
+ lcw->lcw_callback(lcw, tsk, lcw->lcw_data);
+ }
+
+ spin_lock_irqsave(&lcw_pending_timers_lock, flags);
+ }
+ spin_unlock_irqrestore(&lcw_pending_timers_lock, flags);
+ }
+
+ complete(&lcw_stop_completion);
+
+ RETURN(rc);
+}
+
+static void lcw_dispatch_start(void)
+{
+ int rc;
+
+ ENTRY;
+ LASSERT(lcw_refcount == 1);
+
+ init_completion(&lcw_stop_completion);
+ init_completion(&lcw_start_completion);
+ init_waitqueue_head(&lcw_event_waitq);
+
+ CDEBUG(D_INFO, "starting dispatch thread\n");
+ rc = kernel_thread(lcw_dispatch_main, NULL, 0);
+ if (rc < 0) {
+ CERROR("error spawning watchdog dispatch thread: %d\n", rc);
+ EXIT;
+ return;
+ }
+ wait_for_completion(&lcw_start_completion);
+ CDEBUG(D_INFO, "watchdog dispatcher initialization complete.\n");
+
+ EXIT;
+}
+
+static void lcw_dispatch_stop(void)
+{
+ ENTRY;
+ LASSERT(lcw_refcount == 0);
+
+ CDEBUG(D_INFO, "trying to stop watchdog dispatcher.\n");
+
+ set_bit(LCW_FLAG_STOP, &lcw_flags);
+ wake_up(&lcw_event_waitq);
+
+ wait_for_completion(&lcw_stop_completion);
+
+ CDEBUG(D_INFO, "watchdog dispatcher has shut down.\n");
+
+ EXIT;
+}
+
+struct lc_watchdog *lc_watchdog_add(int time,
+ void (*callback)(struct lc_watchdog *,
+ struct task_struct *,
+ void *),
+ void *data)
+{
+ struct lc_watchdog *lcw = NULL;
+ ENTRY;
+
+ PORTAL_ALLOC(lcw, sizeof(*lcw));
+ if (!lcw) {
+ CDEBUG(D_INFO, "Could not allocate new lc_watchdog\n");
+ RETURN(ERR_PTR(-ENOMEM));
+ }
+
+ lcw->lcw_task = current;
+ lcw->lcw_pid = current->pid;
+ lcw->lcw_time = (time * HZ) / 1000;
+ lcw->lcw_callback = callback ? callback : lc_watchdog_dumplog;
+ lcw->lcw_data = data;
+ lcw->lcw_state = LC_WATCHDOG_DISABLED;
+
+ INIT_LIST_HEAD(&lcw->lcw_list);
+
+ lcw->lcw_timer.function = lcw_cb;
+ lcw->lcw_timer.data = (unsigned long)lcw;
+ lcw->lcw_timer.expires = jiffies + lcw->lcw_time;
+ init_timer(&lcw->lcw_timer);
+
+ down(&lcw_refcount_sem);
+ if (++lcw_refcount == 1)
+ lcw_dispatch_start();
+ up(&lcw_refcount_sem);
+
+ /* Keep this working in case we enable them by default */
+ if (lcw->lcw_state == LC_WATCHDOG_ENABLED) {
+ do_gettimeofday(&lcw->lcw_last_touched);
+ add_timer(&lcw->lcw_timer);
+ }
+
+ RETURN(lcw);
+}
+EXPORT_SYMBOL(lc_watchdog_add);
+
+static long
+timeval_sub(struct timeval *large, struct timeval *small)
+{
+ return (large->tv_sec - small->tv_sec) * 1000000 +
+ (large->tv_usec - small->tv_usec);
+}
+
+static void lcw_update_time(struct lc_watchdog *lcw, const char *message)
+{
+ struct timeval newtime;
+ unsigned long timediff;
+
+ do_gettimeofday(&newtime);
+ if (lcw->lcw_state == LC_WATCHDOG_EXPIRED) {
+ timediff = timeval_sub(&newtime, &lcw->lcw_last_touched);
+ CWARN("Expired watchdog for pid %d %s after %lu.%.4lus\n",
+ lcw->lcw_pid,
+ message,
+ timediff / 1000000,
+ (timediff % 1000000) / 100);
+ }
+ lcw->lcw_last_touched = newtime;
+}
+
+void lc_watchdog_touch(struct lc_watchdog *lcw)
+{
+ unsigned long flags;
+ ENTRY;
+ LASSERT(lcw != NULL);
+
+ spin_lock_irqsave(&lcw_pending_timers_lock, flags);
+ if (!list_empty(&lcw->lcw_list))
+ list_del_init(&lcw->lcw_list);
+ spin_unlock_irqrestore(&lcw_pending_timers_lock, flags);
+
+ lcw_update_time(lcw, "touched");
+ lcw->lcw_state = LC_WATCHDOG_ENABLED;
+
+ mod_timer(&lcw->lcw_timer, jiffies + lcw->lcw_time);
+
+ EXIT;
+}
+EXPORT_SYMBOL(lc_watchdog_touch);
+
+void lc_watchdog_disable(struct lc_watchdog *lcw)
+{
+ unsigned long flags;
+ ENTRY;
+ LASSERT(lcw != NULL);
+
+ spin_lock_irqsave(&lcw_pending_timers_lock, flags);
+ if (!list_empty(&lcw->lcw_list))
+ list_del_init(&lcw->lcw_list);
+ spin_unlock_irqrestore(&lcw_pending_timers_lock, flags);
+
+ lcw_update_time(lcw, "disabled");
+ lcw->lcw_state = LC_WATCHDOG_DISABLED;
+
+ EXIT;
+}
+EXPORT_SYMBOL(lc_watchdog_disable);
+
+void lc_watchdog_delete(struct lc_watchdog *lcw)
+{
+ unsigned long flags;
+ ENTRY;
+ LASSERT(lcw != NULL);
+
+ del_timer(&lcw->lcw_timer);
+
+ lcw_update_time(lcw, "deleted");
+
+ spin_lock_irqsave(&lcw_pending_timers_lock, flags);
+ if (!list_empty(&lcw->lcw_list))
+ list_del_init(&lcw->lcw_list);
+ spin_unlock_irqrestore(&lcw_pending_timers_lock, flags);
+
+ down(&lcw_refcount_sem);
+ if (--lcw_refcount == 0)
+ lcw_dispatch_stop();
+ up(&lcw_refcount_sem);
+
+ PORTAL_FREE(lcw, sizeof(*lcw));
+
+ EXIT;
+}
+EXPORT_SYMBOL(lc_watchdog_delete);
+
+/*
+ * Provided watchdog handlers
+ */
+
+extern void portals_debug_dumplog_internal(void *arg);
+
+void lc_watchdog_dumplog(struct lc_watchdog *lcw,
+ struct task_struct *tsk,
+ void *data)
+{
+ tsk = tsk ? tsk : current;
+ portals_debug_dumplog_internal((void *)(long)tsk->pid);
+}
+EXPORT_SYMBOL(lc_watchdog_dumplog);
continue;
if (nal->nal_refct != 0) {
- CWARN("NAL %d has outstanding refcount %d\n",
+ CWARN("NAL %x has outstanding refcount %d\n",
i, nal->nal_refct);
nal->nal_ni_fini(nal);
}
nal->nal_handle.nal_idx = (NI_HANDLE_MAGIC & ~NI_HANDLE_MASK) | interface;
nal->nal_handle.cookie = 0;
- CDEBUG(D_OTHER, "Starting up NAL (%d) refs %d\n", interface, nal->nal_refct);
+ CDEBUG(D_OTHER, "Starting up NAL (%x) refs %d\n", interface, nal->nal_refct);
rc = nal->nal_ni_init(nal, requested_pid, desired_limits, actual_limits);
if (rc != PTL_OK) {
- CERROR("Error %d starting up NAL %d, refs %d\n", rc,
+ CERROR("Error %d starting up NAL %x, refs %d\n", rc,
interface, nal->nal_refct);
GOTO(out, rc);
}
ptl_handle_ni_t nih;
ptl_process_id_t pid;
- CDEBUG (D_IOCTL, "Getting nid for nal [%d]\n", data->ioc_nal);
+ CDEBUG (D_IOCTL, "Getting nid for nal [%x]\n", data->ioc_nal);
err = PtlNIInit(data->ioc_nal, LUSTRE_SRV_PTL_PID, NULL,
NULL, &nih);
struct list_head *curr;
unsigned int generation;
off_t skip;
+ rwlock_t proc_route_rwlock;
} kpr_read_routes_data;
/* nal2name support re-used from utils/portals.c */
{ "tcp", SOCKNAL},
{ "gm", GMNAL},
{ "ib", OPENIBNAL},
+ { "iib", IIBNAL},
+ { "lo", LONAL},
{ NULL, -1}
};
static int kpr_proc_routes_read(char *page, char **start, off_t off,
int count, int *eof, void *data)
{
- struct proc_route_data *prd = data;
- kpr_route_entry_t *re;
- kpr_gateway_entry_t *ge;
- int chunk_len = 0;
- int line_len = 0;
- int user_len = 0;
+ struct proc_route_data *prd = data;
+ kpr_route_entry_t *re;
+ kpr_gateway_entry_t *ge;
+ int chunk_len = 0;
+ int line_len = 0;
+ int user_len = 0;
+ int rc = 0;
*eof = 1;
*start = page;
+ write_lock(&(prd->proc_route_rwlock));
+
if (prd->curr == NULL) {
if (off != 0)
- return 0;
+ goto routes_read_exit;
/* First pass, initialize our private data */
prd->curr = kpr_routes.next;
/* Abort route list generation change */
if (prd->generation != kpr_routes_generation) {
prd->curr = NULL;
- return sprintf(page, "\nError: Routes Changed\n");
+ rc = sprintf(page, "\nError: Routes Changed\n");
+ goto routes_read_exit;
}
/* All the routes have been walked */
if (prd->curr == &kpr_routes) {
prd->curr = NULL;
- return 0;
+ goto routes_read_exit;
}
}
if (prd->curr->next == NULL) {
prd->curr = NULL;
read_unlock(&kpr_rwlock);
- return sprintf(page, "\nError: Routes Changed\n");
+ rc = sprintf(page, "\nError: Routes Changed\n");
+ goto routes_read_exit;
}
prd->curr = prd->curr->next;
prd->curr = prd->curr->prev;
prd->skip = line_len - (user_len - count);
read_unlock(&kpr_rwlock);
- return count;
+ rc = count;
+ goto routes_read_exit;
}
/* Not enough data to entirely satify callers request */
prd->skip = 0;
read_unlock(&kpr_rwlock);
- return user_len;
+ rc = user_len;
+
+routes_read_exit:
+ write_unlock(&(prd->proc_route_rwlock));
+ return rc;
}
static int kpr_proc_routes_write(struct file *file, const char *ubuffer,
kpr_read_routes_data.curr = NULL;
kpr_read_routes_data.generation = 0;
kpr_read_routes_data.skip = 0;
+ kpr_read_routes_data.proc_route_rwlock = RW_LOCK_UNLOCKED;
routes_entry->data = &kpr_read_routes_data;
routes_entry->read_proc = kpr_proc_routes_read;
struct list_head *e;
kpr_nal_entry_t *ne;
- CDEBUG (D_NET, "Registering NAL %d\n", nalif->kprni_nalid);
+ CDEBUG (D_NET, "Registering NAL %x\n", nalif->kprni_nalid);
PORTAL_ALLOC (ne, sizeof (*ne));
if (ne == NULL)
{
write_unlock_irqrestore (&kpr_rwlock, flags);
- CERROR ("Attempt to register same NAL %d twice\n", ne->kpne_interface.kprni_nalid);
+ CERROR ("Attempt to register same NAL %x twice\n", ne->kpne_interface.kprni_nalid);
PORTAL_FREE (ne, sizeof (*ne));
return (-EEXIST);
kpr_upcall_t *u = kmalloc (sizeof (kpr_upcall_t), GFP_ATOMIC);
if (u == NULL) {
- CERROR ("Upcall out of memory: nal %d nid "LPX64" (%s) %s\n",
+ CERROR ("Upcall out of memory: nal %x nid "LPX64" (%s) %s\n",
gw_nalid, gw_nid,
portals_nid2str(gw_nalid, gw_nid, str),
alive ? "up" : "down");
struct list_head *n;
char str[PTL_NALFMT_SIZE];
- CDEBUG (D_NET, "%s notifying [%d] "LPX64": %s\n",
+ CDEBUG (D_NET, "%s notifying [%x] "LPX64": %s\n",
byNal ? "NAL" : "userspace",
gateway_nalid, gateway_nid, alive ? "up" : "down");
/* can't do predictions... */
do_gettimeofday (&now);
if (when > now.tv_sec) {
- CWARN ("Ignoring prediction from %s of [%d] "LPX64" %s "
+ CWARN ("Ignoring prediction from %s of [%x] "LPX64" %s "
"%ld seconds in the future\n",
byNal ? "NAL" : "userspace",
gateway_nalid, gateway_nid,
if (byNal) {
/* It wasn't userland that notified me... */
- CWARN ("Upcall: NAL %d NID "LPX64" (%s) is %s\n",
+ CWARN ("Upcall: NAL %x NID "LPX64" (%s) is %s\n",
gateway_nalid, gateway_nid,
portals_nid2str(gateway_nalid, gateway_nid, str),
alive ? "alive" : "dead");
unsigned long flags;
kpr_nal_entry_t *ne = (kpr_nal_entry_t *)arg;
- CDEBUG (D_NET, "Shutting down NAL %d\n", ne->kpne_interface.kprni_nalid);
+ CDEBUG (D_NET, "Shutting down NAL %x\n", ne->kpne_interface.kprni_nalid);
LASSERT (!ne->kpne_shutdown);
LASSERT (!in_interrupt());
unsigned long flags;
kpr_nal_entry_t *ne = (kpr_nal_entry_t *)arg;
- CDEBUG (D_NET, "Deregister NAL %d\n", ne->kpne_interface.kprni_nalid);
+ CDEBUG (D_NET, "Deregister NAL %x\n", ne->kpne_interface.kprni_nalid);
LASSERT (ne->kpne_shutdown); /* caller must have issued shutdown already */
LASSERT (!in_interrupt());
/* Wait until all outstanding messages/notifications have completed */
while (atomic_read (&ne->kpne_refcount) != 0)
{
- CDEBUG (D_NET, "Waiting for refcount on NAL %d to reach zero (%d)\n",
+ CDEBUG (D_NET, "Waiting for refcount on NAL %x to reach zero (%d)\n",
ne->kpne_interface.kprni_nalid, atomic_read (&ne->kpne_refcount));
set_current_state (TASK_UNINTERRUPTIBLE);
/* Caller wants to know if 'target_nid' can be reached via a gateway
* ON HER OWN NETWORK */
- CDEBUG (D_NET, "lookup "LPX64" from NAL %d\n", target_nid,
+ CDEBUG (D_NET, "lookup "LPX64" from NAL %x\n", target_nid,
ne->kpne_interface.kprni_nalid);
LASSERT (!in_interrupt());
/* NB can't deref 're' now; it might have been removed! */
- CDEBUG (D_NET, "lookup "LPX64" from NAL %d: %d ("LPX64")\n",
+ CDEBUG (D_NET, "lookup "LPX64" from NAL %x: %d ("LPX64")\n",
target_nid, ne->kpne_interface.kprni_nalid, rc,
(rc == 0) ? *gateway_nidp : (ptl_nid_t)0);
return (rc);
kpr_nal_entry_t *tmp_ne;
int rc;
- CDEBUG (D_NET, "forward [%p] "LPX64" from NAL %d\n", fwd,
+ CDEBUG (D_NET, "forward [%p] "LPX64" from NAL %x\n", fwd,
target_nid, src_ne->kpne_interface.kprni_nalid);
LASSERT (nob == lib_kiov_nob (fwd->kprfd_niov, fwd->kprfd_kiov));
read_unlock (&kpr_rwlock);
- CDEBUG (D_NET, "forward [%p] "LPX64" from NAL %d: "
- "to "LPX64" on NAL %d\n",
+ CDEBUG (D_NET, "forward [%p] "LPX64" from NAL %x: "
+ "to "LPX64" on NAL %x\n",
fwd, target_nid, src_ne->kpne_interface.kprni_nalid,
fwd->kprfd_gateway_nid, dst_ne->kpne_interface.kprni_nalid);
out:
kpr_fwd_errors++;
- CDEBUG (D_NET, "Failed to forward [%p] "LPX64" from NAL %d: %d\n",
+ CDEBUG (D_NET, "Failed to forward [%p] "LPX64" from NAL %x: %d\n",
fwd, target_nid, src_ne->kpne_interface.kprni_nalid, rc);
(fwd->kprfd_callback)(fwd->kprfd_callback_arg, rc);
kpr_nal_entry_t *dst_ne = (kpr_nal_entry_t *)arg;
kpr_nal_entry_t *src_ne = (kpr_nal_entry_t *)fwd->kprfd_router_arg;
- CDEBUG (D_NET, "complete(1) [%p] from NAL %d to NAL %d: %d\n", fwd,
+ CDEBUG (D_NET, "complete(1) [%p] from NAL %x to NAL %x: %d\n", fwd,
src_ne->kpne_interface.kprni_nalid, dst_ne->kpne_interface.kprni_nalid, error);
atomic_dec (&dst_ne->kpne_refcount); /* CAVEAT EMPTOR dst_ne can disappear now!!! */
(fwd->kprfd_callback)(fwd->kprfd_callback_arg, error);
- CDEBUG (D_NET, "complete(2) [%p] from NAL %d: %d\n", fwd,
+ CDEBUG (D_NET, "complete(2) [%p] from NAL %x: %d\n", fwd,
src_ne->kpne_interface.kprni_nalid, error);
atomic_dec (&kpr_queue_depth);
kpr_gateway_entry_t *ge;
int dup = 0;
- CDEBUG(D_NET, "Add route: %d "LPX64" : "LPX64" - "LPX64"\n",
+ CDEBUG(D_NET, "Add route: %x "LPX64" : "LPX64" - "LPX64"\n",
gateway_nalid, gateway_nid, lo_nid, hi_nid);
if (gateway_nalid == PTL_NID_ANY ||
struct list_head *e;
struct list_head *n;
- CDEBUG(D_NET, "Del route [%d] "LPX64" : "LPX64" - "LPX64"\n",
+ CDEBUG(D_NET, "Del route [%x] "LPX64" : "LPX64" - "LPX64"\n",
gw_nalid, gw_nid, lo, hi);
LASSERT(!in_interrupt());
break;
case NAL_CMD_ADD_ROUTE:
- CDEBUG(D_IOCTL, "Adding route: [%d] "LPU64" : "LPU64" - "LPU64"\n",
+ CDEBUG(D_IOCTL, "Adding route: [%x] "LPU64" : "LPU64" - "LPU64"\n",
pcfg->pcfg_nal, pcfg->pcfg_nid,
pcfg->pcfg_nid2, pcfg->pcfg_nid3);
err = kpr_add_route(pcfg->pcfg_gw_nal, pcfg->pcfg_nid,
break;
case NAL_CMD_DEL_ROUTE:
- CDEBUG (D_IOCTL, "Removing routes via [%d] "LPU64" : "LPU64" - "LPU64"\n",
+ CDEBUG (D_IOCTL, "Removing routes via [%x] "LPU64" : "LPU64" - "LPU64"\n",
pcfg->pcfg_gw_nal, pcfg->pcfg_nid,
pcfg->pcfg_nid2, pcfg->pcfg_nid3);
err = kpr_del_route (pcfg->pcfg_gw_nal, pcfg->pcfg_nid,
break;
case NAL_CMD_NOTIFY_ROUTER: {
- CDEBUG (D_IOCTL, "Notifying peer [%d] "LPU64" %s @ %ld\n",
+ CDEBUG (D_IOCTL, "Notifying peer [%x] "LPU64" %s @ %ld\n",
pcfg->pcfg_gw_nal, pcfg->pcfg_nid,
pcfg->pcfg_flags ? "Enabling" : "Disabling",
(time_t)pcfg->pcfg_nid3);
client->tsk = current;
client->args = args;
CDEBUG (D_OTHER, "pingcli_setup args: nid "LPX64" (%s), \
- nal %d, size %u, count: %u, timeout: %u\n",
+ nal %x, size %u, count: %u, timeout: %u\n",
args->ioc_nid,
portals_nid2str(args->ioc_nal, args->ioc_nid, str),
args->ioc_nal, args->ioc_size,
rc = PtlNIInit(args->ioc_nal, 0, NULL, NULL, &nih);
if (rc != PTL_OK || rc != PTL_IFACE_DUP)
{
- CERROR ("NAL %d not loaded\n", args->ioc_nal);
+ CERROR ("NAL %x not loaded\n", args->ioc_nal);
pingcli_shutdown (nih, 4);
return (NULL);
}
/* Aquire and initialize the proper nal for portals. */
rc = PtlNIInit(nal, 0, NULL, NULL, &server->ni);
if (!(rc == PTL_OK || rc == PTL_IFACE_DUP)) {
- CDEBUG (D_OTHER, "NAL %d not loaded\n", nal);
+ CDEBUG (D_OTHER, "NAL %x not loaded\n", nal);
return pingsrv_shutdown (4);
}
client->args = args;
CDEBUG (D_OTHER, "pingcli_setup args: nid "LPX64" (%s), \
- nal %d, size %u, count: %u, timeout: %u\n",
+ nal %x, size %u, count: %u, timeout: %u\n",
args->ioc_nid,
portals_nid2str(args->ioc_nid, args->ioc_nal, str),
args->ioc_nal, args->ioc_size,
rc = PtlNIInit(args->ioc_nal, 0, NULL, NULL, &nih);
if (rc != PTL_OK && rc != PTL_IFACE_DUP)
{
- CERROR ("NAL %d not loaded.\n", args->ioc_nal);
+ CERROR ("NAL %x not loaded.\n", args->ioc_nal);
pingcli_shutdown (nih, 4);
return (NULL);
}
rc = PtlNIInit(nal, 0, NULL, NULL, &server->ni);
if (rc != PTL_OK && rc != PTL_IFACE_DUP) {
- CDEBUG (D_OTHER, "Nal %d not loaded.\n", nal);
+ CDEBUG (D_OTHER, "Nal %x not loaded.\n", nal);
return pingsrv_shutdown (4);
}
#define MAX_MARK_SIZE 100
static const char *portal_debug_subsystems[] =
- {"undefined", "mdc", "mds", "osc", "ost", "class", "log", "llite",
- "rpc", "mgmt", "portals", "libcfs", "socknal", "qswnal", "pinger",
- "filter", "ptlbd", "echo", "ldlm", "lov", "gmnal", "router", "cobd",
- "ibnal", "lmv", "cmobd", "smfs", NULL};
+ {"undefined", "mdc", "mds", "osc",
+ "ost", "class", "log", "llite",
+ "rpc", "mgmt", "portals", "socknal",
+ "qswnal", "pinger", "filter", "ptlbd",
+ "echo", "ldlm", "lov", "gmnal",
+ "router", "cobd", "ibnal", "sm",
+ "asobd", "confobd", "lmv", "cmobd",
+ "lonal", NULL};
static const char *portal_debug_masks[] =
- {"trace", "inode", "super", "ext2", "malloc", "cache", "info", "ioctl",
- "blocks", "net", "warning", "buffs", "other", "dentry", "portals",
- "page", "dlmtrace", "error", "emerg", "ha", "rpctrace", "vfstrace",
- "reada", "mmap", NULL};
+ {"trace", "inode", "super", "ext2",
+ "malloc", "cache", "info", "ioctl",
+ "blocks", "net", "warning", "buffs",
+ "other", "dentry", "portals", "page",
+ "dlmtrace", "error", "emerg", "ha",
+ "rpctrace", "vfstrace", "reada", "mmap",
+ "config", NULL};
struct debug_daemon_cmd {
char *cmd;
return parse_buffer(in, out);
}
-const char debug_daemon_usage[]="usage: debug_daemon {start file [MB]|stop}\n";
+static int
+dbg_write_cmd(int fd, char *str)
+{
+ int len = strlen(str);
+ int rc = write(fd, str, len);
+
+ return (rc == len ? 0 : 1);
+}
+
+const char debug_daemon_usage[] = "usage: %s {start file [MB]|stop}\n";
+#define DAEMON_FILE "/proc/sys/portals/daemon_file"
int jt_dbg_debug_daemon(int argc, char **argv)
{
- int rc, fd;
+ int rc;
+ int fd;
if (argc <= 1) {
- fprintf(stderr, debug_daemon_usage);
- return 0;
+ fprintf(stderr, debug_daemon_usage, argv[0]);
+ return 1;
}
- fd = open("/proc/sys/portals/daemon_file", O_WRONLY);
+ fd = open(DAEMON_FILE, O_WRONLY);
if (fd < 0) {
- fprintf(stderr, "open(daemon_file) failed: %s\n",
+ fprintf(stderr, "open %s failed: %s\n", DAEMON_FILE,
strerror(errno));
- return 1;
+ return -1;
}
-
+
+ rc = -1;
if (strcasecmp(argv[1], "start") == 0) {
- if (argc != 3) {
- fprintf(stderr, debug_daemon_usage);
- return 1;
+ if (argc < 3 || argc > 4 ||
+ (argc == 4 && strlen(argv[3]) > 5)) {
+ fprintf(stderr, debug_daemon_usage, argv[0]);
+ goto out;
}
- rc = write(fd, argv[2], strlen(argv[2]));
- if (rc != strlen(argv[2])) {
- fprintf(stderr, "write(%s) failed: %s\n", argv[2],
- strerror(errno));
- close(fd);
- return 1;
+ if (argc == 4) {
+ char buf[12];
+ const long min_size = 10;
+ const long max_size = 20480;
+ long size;
+ char *end;
+
+ size = strtoul(argv[3], &end, 0);
+ if (size < min_size ||
+ size > max_size ||
+ *end != 0) {
+ fprintf(stderr, "size %s invalid, must be in "
+ "the range %ld-%ld MB\n", argv[3],
+ min_size, max_size);
+ goto out;
+ }
+
+ snprintf(buf, sizeof(buf), "size=%ld", size);
+ rc = dbg_write_cmd(fd, buf);
+ if (rc != 0) {
+ fprintf(stderr, "set %s failed: %s\n",
+ buf, strerror(errno));
+ goto out;
+ }
}
- } else if (strcasecmp(argv[1], "stop") == 0) {
- rc = write(fd, "stop", 4);
- if (rc != 4) {
- fprintf(stderr, "write(stop) failed: %s\n",
+
+ rc = dbg_write_cmd(fd, "start");
+ if (rc != 0) {
+ fprintf(stderr, "start debug_daemon on %s failed: %s\n",
+ argv[2], strerror(errno));
+ goto out;
+ }
+
+ rc = 0;
+ goto out;
+ }
+
+ if (strcasecmp(argv[1], "stop") == 0) {
+ rc = dbg_write_cmd(fd, "stop");
+ if (rc != 0) {
+ fprintf(stderr, "stopping debug_daemon failed: %s\n",
strerror(errno));
- close(fd);
- return 1;
+ goto out;
}
- } else {
- fprintf(stderr, debug_daemon_usage);
- return 1;
+
+ rc = 0;
+ goto out;
}
+ fprintf(stderr, debug_daemon_usage, argv[0]);
+ rc = -1;
+out:
close(fd);
- return 0;
+ return rc;
}
int jt_dbg_clear_debug_buf(int argc, char **argv)
#include <time.h>
#include <stdarg.h>
#include <endian.h>
+#if CRAY_PORTALS
+#ifdef REDSTORM
+#define __QK__
+#endif
+#include <portals/ipmap.h>
+#endif
#ifdef __CYGWIN__
static name2num_t nalnames[] = {
{"any", 0},
+#if !CRAY_PORTALS
{"tcp", SOCKNAL},
{"elan", QSWNAL},
{"gm", GMNAL},
{"openib", OPENIBNAL},
{"iib", IIBNAL},
{"lo", LONAL},
+#else
+ {"cray_kern_nal", CRAY_KERN_NAL},
+ {"cray_user_nal", CRAY_USER_NAL},
+ {"cray_qk_nal", CRAY_QK_NAL},
+#endif
{NULL, -1}
};
}
if (ptl_parse_ipaddr (&ipaddr, str) == 0) {
+#if !CRAY_PORTALS
*nidp = (ptl_nid_t)ipaddr;
+#else
+ *nidp = (((ptl_nid_t)ipaddr & PNAL_HOSTID_MASK) << PNAL_VNODE_SHIFT);
+#endif
return (0);
}
ln -s `pwd` $LINUX/fs/lustre
fi
+ # -------- linux objects (for 2.6) --
+ AC_MSG_CHECKING([for Linux objects dir])
+ AC_ARG_WITH([linux-obj],
+ AC_HELP_STRING([--with-linux-obj=path],
+ [set path to Linux objects dir (default=\$LINUX)]),
+ [LINUX_OBJ=$with_linux_obj],
+ [LINUX_OBJ=$LINUX])
+ AC_MSG_RESULT([$LINUX_OBJ])
+ AC_SUBST(LINUX_OBJ)
+
# -------- check for .confg --------
AC_ARG_WITH([linux-config],
[AC_HELP_STRING([--with-linux-config=path],
- [set path to Linux .conf (default=\$LINUX/.config)])],
+ [set path to Linux .conf (default=\$LINUX_OBJ/.config)])],
[LINUX_CONFIG=$with_linux_config],
- [LINUX_CONFIG=$LINUX/.config])
+ [LINUX_CONFIG=$LINUX_OBJ/.config])
AC_SUBST(LINUX_CONFIG)
AC_CHECK_FILE([/boot/kernel.h],
AC_DEFUN([LUSTRE_MODULE_COMPILE_IFELSE],
[m4_ifvaln([$1], [LUSTRE_MODULE_CONFTEST([$1])])dnl
rm -f kernel-tests/conftest.o kernel-tests/conftest.mod.c kernel-tests/conftest.ko
-AS_IF([AC_TRY_COMMAND(cp conftest.c kernel-tests && make [$2] -f $PWD/kernel-tests/Makefile LUSTRE_LINUX_CONFIG=$LINUX_CONFIG -o tmp_include_depends -o scripts -o include/config/MARKER -C $LINUX EXTRA_CFLAGS="-Werror-implicit-function-declaration $EXTRA_KCFLAGS" $ARCH_UM SUBDIRS=$PWD/kernel-tests) >/dev/null && AC_TRY_COMMAND([$3])],
+AS_IF([AC_TRY_COMMAND(cp conftest.c kernel-tests && make [$2] CC="$CC" -f $PWD/kernel-tests/Makefile LUSTRE_LINUX_CONFIG=$LINUX_CONFIG -o tmp_include_depends -o scripts -o include/config/MARKER -C $LINUX_OBJ EXTRA_CFLAGS="-Werror-implicit-function-declaration $EXTRA_KCFLAGS" $ARCH_UM $MODULE_TARGET=$PWD/kernel-tests) >/dev/null && AC_TRY_COMMAND([$3])],
[$4],
[_AC_MSG_LOG_CONFTEST
m4_ifvaln([$5],[$5])dnl])dnl
[AC_MSG_ERROR([Kernel config could not be found. If you are building from a kernel-source rpm consult README.kernel-source])])
# ----------- make dep run? ------------------
- AC_CHECK_FILES([$LINUX/include/linux/autoconf.h
- $LINUX/include/linux/version.h
+ AC_CHECK_FILES([$LINUX_OBJ/include/linux/autoconf.h
+ $LINUX_OBJ/include/linux/version.h
$LINUX/include/linux/config.h],[],
[AC_MSG_ERROR([Run make config in $LINUX.])])
# tarred up the tree and ran make dep etc. in it, then
# version.h gets overwritten with a standard linux one.
- if grep rhconfig $LINUX/include/linux/version.h >/dev/null ; then
+ if grep rhconfig $LINUX_OBJ/include/linux/version.h >/dev/null ; then
# This is a clean kernel-source tree, we need to
# enable extensive workarounds to get this to build
# modules
EXTRA_KCFLAGS="-include $KERNEL_SOURCE_HEADER $EXTRA_KCFLAGS"
fi
- # --- check that we can build modules at all
- AC_MSG_CHECKING([that modules can be built])
- LUSTRE_MODULE_TRY_COMPILE([],[],
- [
- AC_MSG_RESULT([yes])
- ],[
- AC_MSG_RESULT([no])
- AC_MSG_WARN([Consult config.log for details.])
- AC_MSG_WARN([If you are trying to build with a kernel-source rpm, consult README.kernel-source])
- AC_MSG_ERROR([Kernel modules could not be built.])
- ])
-
- # ------------ LINUXRELEASE and moduledir ------------------
+ # ------------ external module support ---------------------
MODULE_TARGET="SUBDIRS"
if test $linux25 = 'yes' ; then
- # ------------ external module support ---------------------
makerule="$PWD/kernel-tests"
AC_MSG_CHECKING([for external module build support])
rm -f kernel-tests/conftest.i
makerule="_dir_$PWD/kernel-tests"
fi
AC_SUBST(MODULE_TARGET)
+
+ # --- check that we can build modules at all
+ AC_MSG_CHECKING([that modules can be built])
+ LUSTRE_MODULE_TRY_COMPILE([],[],
+ [
+ AC_MSG_RESULT([yes])
+ ],[
+ AC_MSG_RESULT([no])
+ AC_MSG_WARN([Consult config.log for details.])
+ AC_MSG_WARN([If you are trying to build with a kernel-source rpm, consult README.kernel-source])
+ AC_MSG_ERROR([Kernel modules could not be built.])
+ ])
+
+ # ------------ LINUXRELEASE and moduledir ------------------
LINUXRELEASE=
rm -f kernel-tests/conftest.i
AC_MSG_CHECKING([for Linux release])
AC_SUBST(GMCPPFLAGS)
AC_SUBST(GMNAL)
+ if test $linux25 = 'no' ; then
#### OpenIB
AC_MSG_CHECKING([if OpenIB kernel headers are present])
OPENIBCPPFLAGS="-I$LINUX/drivers/infiniband/include -DIN_TREE_BUILD"
EXTRA_KCFLAGS="$EXTRA_KCFLAGS_save"
AC_SUBST(OPENIBCPPFLAGS)
AC_SUBST(OPENIBNAL)
+ fi
#### Infinicon IB
AC_MSG_CHECKING([if Infinicon IB kernel headers are present])
#include <linux/fs.h>
#include <linux/version.h>
],[
- #if defined(CONFIG_X86_64) && (LINUX_VERSION_CODE < KERNEL_VERSION(2,4,24))
- #error "x86_64 down_read_trylock broken before 2.4.24"
+ #if (LINUX_VERSION_CODE < KERNEL_VERSION(2,4,24))
+ #error "down_read_trylock broken before 2.4.24"
#endif
struct inode i;
return (char *)&i.i_alloc_sem - (char *)&i;
AC_MSG_RESULT([no])
])
+ # ------------ kallsyms (so software watchdogs produce useful stacks)
+ AC_MSG_CHECKING([if kallsyms is enabled])
+ LUSTRE_MODULE_TRY_COMPILE(
+ [
+ #include <linux/config.h>
+ ],[
+ #ifndef CONFIG_KALLSYMS
+ #error CONFIG_KALLSYMS is not #defined
+ #endif
+ ],[
+ AC_MSG_RESULT([yes])
+ ],[
+ AC_MSG_RESULT([no])
+ if test "x$ARCH_UM" = "x" ; then
+ AC_MSG_ERROR([Lustre requires that CONFIG_KALLSYMS is enabled in your kernel.])
+ fi
+ ])
+
+ # ------------ check for our show_task patch
+ AC_MSG_CHECKING([if kernel exports show_task])
+ have_show_task=0
+ for file in ksyms sched ; do
+ if grep -q "EXPORT_SYMBOL(show_task)" \
+ "$LINUX/kernel/$file.c" 2>/dev/null ; then
+ have_show_task=1
+ break
+ fi
+ done
+ if test x$have_show_task = x1 ; then
+ AC_DEFINE(HAVE_SHOW_TASK, 1, [show_task is exported])
+ AC_MSG_RESULT(yes)
+ else
+ AC_MSG_RESULT(no)
+ fi
+
case $BACKINGFS in
ext3)
# --- Check that ext3 and ext3 xattr are enabled in the kernel
echo "CFLAGS: $CFLAGS"
echo "EXTRA_KCFLAGS: $EXTRA_KCFLAGS"
echo "LLCFLAGS: $LLCFLAGS"
+
+ENABLE_INIT_SCRIPTS=0
+if test x$enable_utils = xyes ; then
+ AC_MSG_CHECKING([whether to install init scripts])
+ # our scripts only work on red hat systems
+ if test -f /etc/init.d/functions -a -f /etc/sysconfig/network ; then
+ ENABLE_INIT_SCRIPTS=1
+ AC_MSG_RESULT([yes])
+ else
+ AC_MSG_RESULT([no])
+ fi
+fi
+AM_CONDITIONAL(INIT_SCRIPTS, test x$ENABLE_INIT_SCRIPTS = "x1")
+AC_SUBST(ENABLE_INIT_SCRIPTS)
#define LBUG_WITH_LOC(file, func, line) \
do { \
CEMERG("LBUG\n"); \
- CERROR("STACK: %s\n", portals_debug_dumpstack()); \
+ portals_debug_dumpstack(NULL); \
portals_debug_dumplog(); \
portals_run_lbug_upcall(file, func, line); \
set_task_state(current, TASK_UNINTERRUPTIBLE); \
#endif /* PORTALS_PROFILING */
/* debug.c */
-extern spinlock_t stack_backtrace_lock;
-
-char *portals_debug_dumpstack(void);
+void portals_debug_dumpstack(struct task_struct *tsk);
void portals_run_upcall(char **argv);
void portals_run_lbug_upcall(char * file, const char *fn, const int line);
void portals_debug_dumplog(void);
(unsigned long)(id)->li_stc.u.e3s.l3s_gen
#endif
-
#define S_IBNAL 0x00400000 /* All IB NALs */
#define S_SM 0x00800000
#define S_ASOBD 0x01000000
-#define S_LMV 0x02000000
-#define S_CMOBD 0x04000000
-#define S_LONAL 0x08000000
-
-/* If you change these values, please keep portals/utils/debug.c
- * up to date! */
+#define S_CONFOBD 0x02000000
+#define S_LMV 0x04000000
+#define S_CMOBD 0x08000000
+#define S_LONAL 0x10000000
+/* If you change these values, please keep these files up to date...
+ * portals/utils/debug.c
+ * utils/lconf
+ */
/* Debugging masks (32 bits, non-overlapping) */
#define D_TRACE 0x00000001 /* ENTRY/EXIT markers */
#define D_READA 0x00400000 /* read-ahead */
#define D_MMAP 0x00800000
#define D_CONFIG 0x01000000
+/* If you change these values, please keep these files up to date...
+ * portals/utils/debug.c
+ * utils/lconf
+ */
+
#ifdef __KERNEL__
# include <linux/sched.h> /* THREAD_SIZE */
#else
#define CDEBUG_MAX_LIMIT 600
#define CDEBUG_LIMIT(cdebug_mask, cdebug_format, a...) \
do { \
- static unsigned long cdebug_next; \
- static int cdebug_count, cdebug_delay = 1; \
+ static unsigned long cdebug_next = 0; \
+ static int cdebug_count = 0, cdebug_delay = 1; \
\
CHECK_STACK(CDEBUG_STACK); \
if (time_after(jiffies, cdebug_next)) { \
if (cdebug_count) { \
portals_debug_msg(DEBUG_SUBSYSTEM, cdebug_mask, \
__FILE__, __FUNCTION__, __LINE__, \
- CDEBUG_STACK, cdebug_format, ## a); \
+ 0, "skipped %d similar messages\n", \
+ cdebug_count); \
cdebug_count = 0; \
} \
if (time_after(jiffies, cdebug_next+(CDEBUG_MAX_LIMIT+10)*HZ))\
#endif
-#define _LIBCFS_H
+#ifdef __KERNEL__
+/* libcfs watchdogs */
+struct lc_watchdog;
+
+/* Just use the default handler (dumplog) */
+#define LC_WATCHDOG_DEFAULT_CB NULL
+
+/* Add a watchdog which fires after "time" milliseconds of delay. You have to
+ * touch it once to enable it. */
+struct lc_watchdog *lc_watchdog_add(int time,
+ void (*cb)(struct lc_watchdog *,
+ struct task_struct *,
+ void *),
+ void *data);
+
+/* Enables a watchdog and resets its timer. */
+void lc_watchdog_touch(struct lc_watchdog *lcw);
+
+/* Disable a watchdog; touch it to restart it. */
+void lc_watchdog_disable(struct lc_watchdog *lcw);
+
+/* Clean up the watchdog */
+void lc_watchdog_delete(struct lc_watchdog *lcw);
+
+/* Dump a debug log */
+void lc_watchdog_dumplog(struct lc_watchdog *lcw,
+ struct task_struct *tsk,
+ void *data);
+#endif /* !__KERNEL__ */
#endif /* _LIBCFS_H */
# define USERMODEHELPER(path, argv, envp) \
call_usermodehelper(path, argv, envp, 1)
# define RECALC_SIGPENDING recalc_sigpending()
+# define CLEAR_SIGPENDING clear_tsk_thread_flag(current, \
+ TIF_SIGPENDING)
# define CURRENT_SECONDS get_seconds()
-# define smp_num_cpus NR_CPUS
+# define smp_num_cpus num_online_cpus()
#elif defined(CONFIG_RH_2_4_20) /* RH 2.4.x */
# define USERMODEHELPER(path, argv, envp) \
call_usermodehelper(path, argv, envp)
# define RECALC_SIGPENDING recalc_sigpending()
+# define CLEAR_SIGPENDING (current->sigpending = 0)
# define CURRENT_SECONDS CURRENT_TIME
-# define kernel_text_address(addr) is_kernel_text_address(addr)
-extern int is_kernel_text_address(unsigned long addr);
-
#else /* 2.4.x */
# define SIGNAL_MASK_LOCK(task, flags) \
# define USERMODEHELPER(path, argv, envp) \
call_usermodehelper(path, argv, envp)
# define RECALC_SIGPENDING recalc_sigpending(current)
+# define CLEAR_SIGPENDING (current->sigpending = 0)
# define CURRENT_SECONDS CURRENT_TIME
-# define kernel_text_address(addr) is_kernel_text_address(addr)
-extern int is_kernel_text_address(unsigned long addr);
-
#endif
#if defined(__arch_um__) && (LINUX_VERSION_CODE < KERNEL_VERSION(2,4,20))
-# define THREAD_NAME(comm, len, fmt, a...) \
- snprintf(comm, len, fmt "|%d", ## a, current->thread.extern_pid)
+#define UML_PID(tsk) ((tsk)->thread.extern_pid)
#elif defined(__arch_um__) && (LINUX_VERSION_CODE < KERNEL_VERSION(2,5,0))
+#define UML_PID(tsk) ((tsk)->thread.mode.tt.extern_pid)
+#else
+#define UML_PID(tsk) ((tsk)->pid)
+#endif
+
+#if defined(__arch_um__) && (LINUX_VERSION_CODE < KERNEL_VERSION(2,5,0))
# define THREAD_NAME(comm, len, fmt, a...) \
- snprintf(comm, len,fmt"|%d", ## a,current->thread.mode.tt.extern_pid)
+ snprintf(comm, len,fmt"|%d", ## a, UML_PID(current))
#else
# define THREAD_NAME(comm, len, fmt, a...) \
snprintf(comm, len, fmt, ## a)
#include <linux/stat.h>
#include <linux/errno.h>
#include <linux/uio.h>
+#include <linux/init.h>
#define DEBUG_SUBSYSTEM S_LONAL
struct iovec *iov;
ptl_kiov_t *kiov;
} klod_iov;
-
} klo_desc_t;
typedef struct
#define KQSW_TX_MAXCONTIG (1<<10) /* largest payload that gets made contiguous on transmit */
#define KQSW_NTXMSGS 8 /* # normal transmit messages */
-#define KQSW_NNBLK_TXMSGS 512 /* # reserved transmit messages if can't block */
+#define KQSW_NNBLK_TXMSGS (PAGE_SIZE == 4096 ? 512 : 256) /* # reserved transmit messages if can't block */ /* avoid qsnet crash b=5291 */
#define KQSW_NRXMSGS_LARGE 64 /* # large receive buffers */
#define KQSW_EP_ENVELOPES_LARGE 256 /* # large ep envelopes */
MODULES = libcfs
-libcfs-objs := debug.o lwt.o module.o proc.o tracefile.o
+libcfs-objs := debug.o lwt.o module.o proc.o tracefile.o watchdog.o
@INCLUDE_RULES@
static int handled_panic; /* to avoid recursive calls to notifiers */
char portals_upcall[1024] = "/usr/lib/lustre/portals_upcall";
-int portals_do_debug_dumplog(void *arg)
+void portals_debug_dumplog_internal(void *arg)
{
- void *journal_info;
-
- kportal_daemonize("");
-
- reparent_to_init();
- journal_info = current->journal_info;
+ void *journal_info = current->journal_info;
current->journal_info = NULL;
snprintf(debug_file_name, sizeof(debug_file_path) - 1,
tracefile_dump_all_pages(debug_file_name);
current->journal_info = journal_info;
+}
+
+int portals_debug_dumplog_thread(void *arg)
+{
+ kportal_daemonize("");
+ reparent_to_init();
+ portals_debug_dumplog_internal(arg);
wake_up(&debug_ctlwq);
return 0;
}
set_current_state(TASK_INTERRUPTIBLE);
add_wait_queue(&debug_ctlwq, &wait);
- rc = kernel_thread(portals_do_debug_dumplog, (void *)(long)current->pid,
+ rc = kernel_thread(portals_debug_dumplog_thread,
+ (void *)(long)current->pid,
CLONE_VM | CLONE_FS | CLONE_FILES);
if (rc < 0)
printk(KERN_ERR "LustreError: cannot start log dump thread: "
portals_nid2str(nal, id.nid, str);
len = strlen(str);
- snprintf(str + len, PTL_NALFMT_SIZE - len, ",%u", id.pid);
+ snprintf(str + len, PTL_NALFMT_SIZE - len, "-%u", id.pid);
return str;
}
#ifdef __KERNEL__
-char stack_backtrace[LUSTRE_TRACE_SIZE];
-spinlock_t stack_backtrace_lock = SPIN_LOCK_UNLOCKED;
-
-#if defined(__arch_um__)
-char *portals_debug_dumpstack(void)
+void portals_debug_dumpstack(struct task_struct *tsk)
{
+#if defined(__arch_um__)
+ if (tsk != NULL)
+ CWARN("stack dump for pid %d (%d) requested; wake up gdb.\n",
+ tsk->pid, UML_PID(tsk));
asm("int $3");
- return "dump stack\n";
-}
-
-#elif defined(__i386__)
-
-#if (LINUX_VERSION_CODE < KERNEL_VERSION(2,5,0))
-extern int lookup_symbol(unsigned long address, char *buf, int buflen);
-const char *kallsyms_lookup(unsigned long addr,
- unsigned long *symbolsize,
- unsigned long *offset,
- char **modname, char *namebuf)
-{
- int rc = lookup_symbol(addr, namebuf, 128);
- if (rc == -ENOSYS)
- return NULL;
- return namebuf;
-}
+#elif defined(HAVE_SHOW_TASK)
+ /* this is exported by lustre kernel version 42 */
+ extern void show_task(struct task_struct *);
+
+ if (tsk == NULL)
+ tsk = current;
+ CWARN("showing stack for process %d\n", tsk->pid);
+ show_task(tsk);
+#else
+ CWARN("can't show stack: kernel doesn't export show_task\n");
#endif
-
-char *portals_debug_dumpstack(void)
-{
- unsigned long esp = current->thread.esp, addr;
- unsigned long *stack = (unsigned long *)&esp;
- char *buf = stack_backtrace, *pbuf = buf;
- int size;
-
- /* User space on another CPU? */
- if ((esp ^ (unsigned long)current) & (PAGE_MASK << 1)){
- buf[0] = '\0';
- goto out;
- }
-
- size = sprintf(pbuf, " Call Trace: ");
- pbuf += size;
- while (((long) stack & (THREAD_SIZE - 1)) != 0) {
- addr = *stack++;
- if (kernel_text_address(addr)) {
- const char *sym_name;
- char *modname, buffer[128];
- unsigned long junk, offset;
-
- sym_name = kallsyms_lookup(addr, &junk, &offset,
- &modname, buffer);
- if (sym_name == NULL) {
- if (buf + LUSTRE_TRACE_SIZE <= pbuf + 12)
- break;
- size = sprintf(pbuf, "[<%08lx>] ", addr);
- } else {
- if (buf + LUSTRE_TRACE_SIZE
- /* fix length + sizeof('\0') */
- <= pbuf + strlen(buffer) + 28 + 1)
- break;
- size = sprintf(pbuf, "([<%08lx>] %s (0x%p)) ",
- addr, buffer, stack - 1);
- }
- pbuf += size;
- }
- }
-out:
- return buf;
-}
-
-#else /* !__arch_um__ && !__i386__ */
-
-char *portals_debug_dumpstack(void)
-{
- char *buf = stack_backtrace;
- buf[0] = '\0';
- return buf;
}
-#endif /* __arch_um__ */
struct task_struct *portals_current(void)
{
CWARN("current task struct is %p\n", current);
return current;
}
-EXPORT_SYMBOL(stack_backtrace_lock);
EXPORT_SYMBOL(portals_debug_dumpstack);
EXPORT_SYMBOL(portals_current);
#endif /* __KERNEL__ */
int i;
int rc;
- CDEBUG(D_IOCTL, "Register NAL %d, handler: %p\n", nal, handler);
+ CDEBUG(D_IOCTL, "Register NAL %x, handler: %p\n", nal, handler);
down(&nal_cmd_sem);
{
struct nal_cmd_handler *cmd;
- CDEBUG(D_IOCTL, "Unregister NAL %d\n", nal);
+ CDEBUG(D_IOCTL, "Unregister NAL %x\n", nal);
down(&nal_cmd_sem);
cmd = libcfs_find_nal_cmd_handler(nal);
int
libcfs_nal_cmd(struct portals_cfg *pcfg)
{
+#if CRAY_PORTALS
+ /* pretend success */
+ RETURN(0);
+#else
struct nal_cmd_handler *cmd;
__u32 nal = pcfg->pcfg_nal;
int rc = -EINVAL;
down(&nal_cmd_sem);
cmd = libcfs_find_nal_cmd_handler(nal);
if (cmd != NULL) {
- CDEBUG(D_IOCTL, "calling handler nal: %d, cmd: %d\n", nal,
+ CDEBUG(D_IOCTL, "calling handler nal: %x, cmd: %d\n", nal,
pcfg->pcfg_command);
rc = cmd->nch_handler(pcfg, cmd->nch_private);
} else {
- CERROR("invalid nal: %d, cmd: %d\n", nal, pcfg->pcfg_command);
+ CERROR("invalid nal: %x, cmd: %d\n", nal, pcfg->pcfg_command);
}
up(&nal_cmd_sem);
RETURN(rc);
+#endif
}
EXPORT_SYMBOL(libcfs_nal_cmd);
break;
}
- CDEBUG (D_IOCTL, "nal command nal %d cmd %d\n", pcfg.pcfg_nal,
+ CDEBUG (D_IOCTL, "nal command nal %x cmd %d\n", pcfg.pcfg_nal,
pcfg.pcfg_command);
err = libcfs_nal_cmd(&pcfg);
ent->write_proc = trace_write_daemon_file;
ent->read_proc = trace_read_daemon_file;
- ent = create_proc_entry("sys/portals/debug_size", 0, NULL);
+ ent = create_proc_entry("sys/portals/debug_mb", 0, NULL);
if (ent == NULL) {
- CERROR("couldn't register debug_size\n");
+ CERROR("couldn't register debug_mb\n");
return -1;
}
- ent->write_proc = trace_write_debug_size;
- ent->read_proc = trace_read_debug_size;
+ ent->write_proc = trace_write_debug_mb;
+ ent->read_proc = trace_read_debug_mb;
return 0;
}
remove_proc_entry("sys/portals/dump_kernel", NULL);
remove_proc_entry("sys/portals/daemon_file", NULL);
- remove_proc_entry("sys/portals/debug_size", NULL);
+ remove_proc_entry("sys/portals/debug_mb", NULL);
#ifdef CONFIG_SYSCTL
if (portals_table_header)
#include <linux/portals_compat25.h>
#include <linux/libcfs.h>
-#define TCD_MAX_PAGES 1280
+#define TCD_MAX_PAGES (5 << (20 - PAGE_SHIFT))
/* XXX move things up to the top, comment */
atomic_t tctl_shutdown;
};
+#define TRACEFILE_SIZE (500 << 20)
static DECLARE_RWSEM(tracefile_sem);
static char *tracefile = NULL;
+static long long tracefile_size = TRACEFILE_SIZE;
static struct tracefiled_ctl trace_tctl;
static DECLARE_MUTEX(trace_thread_sem);
static int thread_running = 0;
/* the kernel should print a message for us. fall back
* to using the last page in the ring buffer. */
goto ring_buffer;
- return NULL;
}
page->index = 0;
page->mapping = (void *)(long)smp_processor_id();
struct trace_cpu_data *tcd;
struct ptldebug_header header;
struct page *page;
- char *debug_buf;
- int known_size, needed, max_nob;
+ char *debug_buf = format;
+ int known_size, needed = 85 /* average message length */, max_nob;
va_list ap;
unsigned long flags;
struct timeval tv;
known_size = sizeof(header) + strlen(file) + strlen(fn) + 2; // nulls
- page = trace_get_page(tcd, known_size + 40); /* slop */
retry:
- if (page == NULL)
+ page = trace_get_page(tcd, needed + known_size);
+ if (page == NULL) {
+ debug_buf = format;
+ if (needed + known_size > PAGE_SIZE)
+ mask |= D_ERROR;
+ needed = strlen(format);
goto out;
+ }
debug_buf = page_address(page) + page->index + known_size;
- va_start(ap, format);
max_nob = PAGE_SIZE - page->index - known_size;
LASSERT(max_nob > 0);
+ va_start(ap, format);
needed = vsnprintf(debug_buf, max_nob, format, ap);
va_end(ap);
- if (needed > max_nob) {
- /* overflow. oh poop. */
- page = trace_get_page(tcd, needed + known_size);
+ if (needed > max_nob) /* overflow. oh poop. */
goto retry;
- }
header.ph_len = known_size + needed;
debug_buf = page_address(page) + page->index;
printk(KERN_EMERG "page->index == %lu in portals_debug_msg\n",
page->index);
+ out:
if ((mask & (D_EMERG | D_ERROR | D_WARNING)) || portal_printk)
print_to_console(&header, mask, debug_buf, needed, file, fn);
- out:
trace_put_tcd(tcd, flags);
}
EXPORT_SYMBOL(portals_debug_msg);
down_write(&tracefile_sem);
- filp = filp_open(filename, O_CREAT|O_EXCL|O_WRONLY, 0600);
+ filp = filp_open(filename, O_CREAT|O_EXCL|O_WRONLY|O_LARGEFILE, 0600);
if (IS_ERR(filp)) {
rc = PTR_ERR(filp);
printk(KERN_ERR "LustreError: can't open %s for dump: rc %d\n",
filp = NULL;
down_read(&tracefile_sem);
if (tracefile != NULL) {
- filp = filp_open(tracefile, O_CREAT|O_RDWR|O_APPEND|O_LARGEFILE,
- 0600);
+ filp = filp_open(tracefile, O_CREAT|O_RDWR|O_LARGEFILE,
+ 0600);
if (IS_ERR(filp)) {
printk("couldn't open %s: %ld\n", tracefile,
PTR_ERR(filp));
hdr->ph_flags |= PH_FLAG_FIRST_RECORD;
list_for_each_safe(pos, tmp, &pc.pc_pages) {
+ static loff_t f_pos;
page = list_entry(pos, struct page, PAGE_LIST_ENTRY);
LASSERT(page->index <= PAGE_SIZE);
LASSERT(page_count(page) > 0);
+ if (f_pos >= tracefile_size)
+ f_pos = 0;
+ else if (f_pos > filp->f_dentry->d_inode->i_size)
+ f_pos = filp->f_dentry->d_inode->i_size;
+
rc = filp->f_op->write(filp, page_address(page),
- page->index, &filp->f_pos);
+ page->index, &f_pos);
if (rc != page->index) {
printk(KERN_WARNING "wanted to write %lu but "
"wrote %d\n", page->index, rc);
tracefile = NULL;
trace_stop_thread();
goto out_sem;
+ } else if (strncmp(name, "size=", 5) == 0) {
+ tracefile_size = simple_strtoul(name + 5, NULL, 0);
+ if (tracefile_size < 10 || tracefile_size > 20480)
+ tracefile_size = TRACEFILE_SIZE;
+ else
+ tracefile_size <<= 20;
+ goto out_sem;
}
if (name[0] != '/') {
tracefile = name;
name = NULL;
+
+ printk(KERN_INFO "Lustre: debug daemon will attempt to start writing "
+ "to %s (%lukB max)\n", tracefile, (long)(tracefile_size >> 10));
+
trace_start_thread();
out_sem:
up_write(&tracefile_sem);
out:
- if (name)
- kfree(name);
+ kfree(name);
return count;
}
return rc;
}
-int trace_write_debug_size(struct file *file, const char *buffer,
- unsigned long count, void *data)
+int trace_write_debug_mb(struct file *file, const char *buffer,
+ unsigned long count, void *data)
{
- char *string;
- int rc, i, max;
-
- string = kmalloc(count + 1, GFP_KERNEL);
- if (string == NULL)
- return -ENOMEM;
+ char string[32];
+ int i;
+ unsigned max;
- if (copy_from_user(string, buffer, count)) {
- rc = -EFAULT;
- goto out;
+ if (count >= sizeof(string)) {
+ printk(KERN_ERR "Lustre: value too large (length %lu bytes)\n",
+ count);
+ return -EOVERFLOW;
}
+ if (copy_from_user(string, buffer, count))
+ return -EFAULT;
+
max = simple_strtoul(string, NULL, 0);
- if (max == 0) {
- rc = -EINVAL;
- goto out;
- }
+ if (max == 0)
+ return -EINVAL;
max /= smp_num_cpus;
- if (max > num_physpages / 5 * 4) {
+ if (max * smp_num_cpus > (num_physpages >> (20 - 2 - PAGE_SHIFT)) / 5) {
printk(KERN_ERR "Lustre: Refusing to set debug buffer size to "
- "%d pages, which is more than 80%% of physical pages "
- "(%lu).\n", max * smp_num_cpus, num_physpages / 5 * 4);
- return count;
+ "%d MB, which is more than 80%% of physical RAM "
+ "(%lu).\n", max * smp_num_cpus,
+ (num_physpages >> (20 - 2 - PAGE_SHIFT)) / 5);
+ return -EINVAL;
}
for (i = 0; i < NR_CPUS; i++) {
struct trace_cpu_data *tcd;
tcd = &trace_data[i].tcd;
- tcd->tcd_max_pages = max;
+ tcd->tcd_max_pages = max << (20 - PAGE_SHIFT);
}
- out:
- kfree(string);
return count;
}
-int trace_read_debug_size(char *page, char **start, off_t off, int count,
- int *eof, void *data)
+int trace_read_debug_mb(char *page, char **start, off_t off, int count,
+ int *eof, void *data)
{
struct trace_cpu_data *tcd;
unsigned long flags;
int rc;
tcd = trace_get_tcd(flags);
- rc = snprintf(page, count, "%lu", tcd->tcd_max_pages);
+ rc = snprintf(page, count, "%lu\n",
+ (tcd->tcd_max_pages >> (20 - PAGE_SHIFT)) * smp_num_cpus);
trace_put_tcd(tcd, flags);
return rc;
unsigned long count, void *data);
int trace_read_daemon_file(char *page, char **start, off_t off, int count,
int *eof, void *data);
-int trace_write_debug_size(struct file *file, const char *buffer,
- unsigned long count, void *data);
-int trace_read_debug_size(char *page, char **start, off_t off, int count,
- int *eof, void *data);
+int trace_write_debug_mb(struct file *file, const char *buffer,
+ unsigned long count, void *data);
+int trace_read_debug_mb(char *page, char **start, off_t off, int count,
+ int *eof, void *data);
int trace_dk(struct file *file, const char *buffer, unsigned long count,
void *data);
--- /dev/null
+/* -*- mode: c; c-basic-offset: 8; indent-tabs-mode: nil; -*-
+ * vim:expandtab:shiftwidth=8:tabstop=8:
+ *
+ * Copyright (C) 2004 Cluster File Systems, Inc.
+ * Author: Jacob Berkman <jacob@clusterfs.com>
+ *
+ * This file is part of Lustre, http://www.lustre.org.
+ *
+ * Lustre is free software; you can redistribute it and/or
+ * modify it under the terms of version 2 of the GNU General Public
+ * License as published by the Free Software Foundation.
+ *
+ * Lustre is distributed in the hope that it will be useful,
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
+ * GNU General Public License for more details.
+ *
+ * You should have received a copy of the GNU General Public License
+ * along with Lustre; if not, write to the Free Software
+ * Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.
+ */
+
+#define DEBUG_SUBSYSTEM S_PORTALS
+
+#include <linux/kp30.h>
+#include <linux/libcfs.h>
+#include <linux/portals_compat25.h>
+
+
+
+struct lc_watchdog {
+ struct timer_list lcw_timer; /* kernel timer */
+ struct list_head lcw_list;
+ struct timeval lcw_last_touched;
+ struct task_struct *lcw_task;
+
+ void (*lcw_callback)(struct lc_watchdog *,
+ struct task_struct *,
+ void *data);
+ void *lcw_data;
+
+ int lcw_pid;
+ int lcw_time; /* time until watchdog fires, in ms */
+
+ enum {
+ LC_WATCHDOG_DISABLED,
+ LC_WATCHDOG_ENABLED,
+ LC_WATCHDOG_EXPIRED
+ } lcw_state;
+};
+
+/*
+ * The dispatcher will complete lcw_start_completion when it starts,
+ * and lcw_stop_completion when it exits.
+ * Wake lcw_event_waitq to signal timer callback dispatches.
+ */
+static struct completion lcw_start_completion;
+static struct completion lcw_stop_completion;
+static wait_queue_head_t lcw_event_waitq;
+
+/*
+ * Set this and wake lcw_event_waitq to stop the dispatcher.
+ */
+enum {
+ LCW_FLAG_STOP = 0
+};
+static unsigned long lcw_flags = 0;
+
+/*
+ * Number of outstanding watchdogs.
+ * When it hits 1, we start the dispatcher.
+ * When it hits 0, we stop the distpatcher.
+ */
+static __u32 lcw_refcount = 0;
+static DECLARE_MUTEX(lcw_refcount_sem);
+
+/*
+ * List of timers that have fired that need their callbacks run by the
+ * dispatcher.
+ */
+static spinlock_t lcw_pending_timers_lock = SPIN_LOCK_UNLOCKED;
+static struct list_head lcw_pending_timers = \
+ LIST_HEAD_INIT(lcw_pending_timers);
+
+static struct task_struct *lcw_lookup_task(struct lc_watchdog *lcw)
+{
+ struct task_struct *tsk;
+ unsigned long flags;
+ ENTRY;
+
+ read_lock_irqsave(&tasklist_lock, flags);
+ tsk = find_task_by_pid(lcw->lcw_pid);
+ read_unlock_irqrestore(&tasklist_lock, flags);
+ if (!tsk) {
+ CWARN("Process %d was not found in the task list; "
+ "watchdog callback may be incomplete\n", lcw->lcw_pid);
+ } else if (tsk != lcw->lcw_task) {
+ tsk = NULL;
+ CWARN("The current process %d did not set the watchdog; "
+ "watchdog callback may be incomplete\n", lcw->lcw_pid);
+ }
+
+ RETURN(tsk);
+}
+
+static void lcw_cb(unsigned long data)
+{
+ struct lc_watchdog *lcw = (struct lc_watchdog *)data;
+ struct task_struct *tsk;
+ unsigned long flags;
+
+ ENTRY;
+
+ if (lcw->lcw_state != LC_WATCHDOG_ENABLED) {
+ EXIT;
+ return;
+ }
+
+ lcw->lcw_state = LC_WATCHDOG_EXPIRED;
+
+ CWARN("Watchdog triggered for pid %d: it was inactive for %dus\n",
+ lcw->lcw_pid, (lcw->lcw_time * 1000) / HZ);
+
+ tsk = lcw_lookup_task(lcw);
+ if (tsk != NULL)
+ portals_debug_dumpstack(tsk);
+
+ spin_lock_irqsave(&lcw_pending_timers_lock, flags);
+ if (list_empty(&lcw->lcw_list)) {
+ list_add(&lcw->lcw_list, &lcw_pending_timers);
+ wake_up(&lcw_event_waitq);
+ }
+ spin_unlock_irqrestore(&lcw_pending_timers_lock, flags);
+
+ EXIT;
+}
+
+static int is_watchdog_fired(void)
+{
+ unsigned long flags;
+ int rc;
+
+ if (test_bit(LCW_FLAG_STOP, &lcw_flags))
+ return 1;
+
+ spin_lock_irqsave(&lcw_pending_timers_lock, flags);
+ rc = !list_empty(&lcw_pending_timers);
+ spin_unlock_irqrestore(&lcw_pending_timers_lock, flags);
+ return rc;
+}
+
+static int lcw_dispatch_main(void *data)
+{
+ int rc = 0;
+ unsigned long flags;
+ struct lc_watchdog *lcw;
+ struct task_struct *tsk;
+
+ ENTRY;
+
+ kportal_daemonize("lc_watchdogd");
+
+ SIGNAL_MASK_LOCK(current, flags);
+ sigfillset(¤t->blocked);
+ RECALC_SIGPENDING;
+ SIGNAL_MASK_UNLOCK(current, flags);
+
+ complete(&lcw_start_completion);
+
+ while (1) {
+ wait_event_interruptible(lcw_event_waitq, is_watchdog_fired());
+ CDEBUG(D_INFO, "Watchdog got woken up...\n");
+ if (test_bit(LCW_FLAG_STOP, &lcw_flags)) {
+ CDEBUG(D_INFO, "LCW_FLAG_STOP was set, shutting down...\n");
+
+ spin_lock_irqsave(&lcw_pending_timers_lock, flags);
+ rc = !list_empty(&lcw_pending_timers);
+ spin_unlock_irqrestore(&lcw_pending_timers_lock, flags);
+ if (rc) {
+ CERROR("pending timers list was not empty at "
+ "time of watchdog dispatch shutdown\n");
+ }
+ break;
+ }
+
+ spin_lock_irqsave(&lcw_pending_timers_lock, flags);
+ while (!list_empty(&lcw_pending_timers)) {
+
+ lcw = list_entry(lcw_pending_timers.next,
+ struct lc_watchdog,
+ lcw_list);
+ list_del_init(&lcw->lcw_list);
+ spin_unlock_irqrestore(&lcw_pending_timers_lock, flags);
+
+ CDEBUG(D_INFO, "found lcw for pid %d\n", lcw->lcw_pid);
+
+ if (lcw->lcw_state != LC_WATCHDOG_DISABLED) {
+ /*
+ * sanity check the task against our
+ * watchdog
+ */
+ tsk = lcw_lookup_task(lcw);
+ lcw->lcw_callback(lcw, tsk, lcw->lcw_data);
+ }
+
+ spin_lock_irqsave(&lcw_pending_timers_lock, flags);
+ }
+ spin_unlock_irqrestore(&lcw_pending_timers_lock, flags);
+ }
+
+ complete(&lcw_stop_completion);
+
+ RETURN(rc);
+}
+
+static void lcw_dispatch_start(void)
+{
+ int rc;
+
+ ENTRY;
+ LASSERT(lcw_refcount == 1);
+
+ init_completion(&lcw_stop_completion);
+ init_completion(&lcw_start_completion);
+ init_waitqueue_head(&lcw_event_waitq);
+
+ CDEBUG(D_INFO, "starting dispatch thread\n");
+ rc = kernel_thread(lcw_dispatch_main, NULL, 0);
+ if (rc < 0) {
+ CERROR("error spawning watchdog dispatch thread: %d\n", rc);
+ EXIT;
+ return;
+ }
+ wait_for_completion(&lcw_start_completion);
+ CDEBUG(D_INFO, "watchdog dispatcher initialization complete.\n");
+
+ EXIT;
+}
+
+static void lcw_dispatch_stop(void)
+{
+ ENTRY;
+ LASSERT(lcw_refcount == 0);
+
+ CDEBUG(D_INFO, "trying to stop watchdog dispatcher.\n");
+
+ set_bit(LCW_FLAG_STOP, &lcw_flags);
+ wake_up(&lcw_event_waitq);
+
+ wait_for_completion(&lcw_stop_completion);
+
+ CDEBUG(D_INFO, "watchdog dispatcher has shut down.\n");
+
+ EXIT;
+}
+
+struct lc_watchdog *lc_watchdog_add(int time,
+ void (*callback)(struct lc_watchdog *,
+ struct task_struct *,
+ void *),
+ void *data)
+{
+ struct lc_watchdog *lcw = NULL;
+ ENTRY;
+
+ PORTAL_ALLOC(lcw, sizeof(*lcw));
+ if (!lcw) {
+ CDEBUG(D_INFO, "Could not allocate new lc_watchdog\n");
+ RETURN(ERR_PTR(-ENOMEM));
+ }
+
+ lcw->lcw_task = current;
+ lcw->lcw_pid = current->pid;
+ lcw->lcw_time = (time * HZ) / 1000;
+ lcw->lcw_callback = callback ? callback : lc_watchdog_dumplog;
+ lcw->lcw_data = data;
+ lcw->lcw_state = LC_WATCHDOG_DISABLED;
+
+ INIT_LIST_HEAD(&lcw->lcw_list);
+
+ lcw->lcw_timer.function = lcw_cb;
+ lcw->lcw_timer.data = (unsigned long)lcw;
+ lcw->lcw_timer.expires = jiffies + lcw->lcw_time;
+ init_timer(&lcw->lcw_timer);
+
+ down(&lcw_refcount_sem);
+ if (++lcw_refcount == 1)
+ lcw_dispatch_start();
+ up(&lcw_refcount_sem);
+
+ /* Keep this working in case we enable them by default */
+ if (lcw->lcw_state == LC_WATCHDOG_ENABLED) {
+ do_gettimeofday(&lcw->lcw_last_touched);
+ add_timer(&lcw->lcw_timer);
+ }
+
+ RETURN(lcw);
+}
+EXPORT_SYMBOL(lc_watchdog_add);
+
+static long
+timeval_sub(struct timeval *large, struct timeval *small)
+{
+ return (large->tv_sec - small->tv_sec) * 1000000 +
+ (large->tv_usec - small->tv_usec);
+}
+
+static void lcw_update_time(struct lc_watchdog *lcw, const char *message)
+{
+ struct timeval newtime;
+ unsigned long timediff;
+
+ do_gettimeofday(&newtime);
+ if (lcw->lcw_state == LC_WATCHDOG_EXPIRED) {
+ timediff = timeval_sub(&newtime, &lcw->lcw_last_touched);
+ CWARN("Expired watchdog for pid %d %s after %lu.%.4lus\n",
+ lcw->lcw_pid,
+ message,
+ timediff / 1000000,
+ (timediff % 1000000) / 100);
+ }
+ lcw->lcw_last_touched = newtime;
+}
+
+void lc_watchdog_touch(struct lc_watchdog *lcw)
+{
+ unsigned long flags;
+ ENTRY;
+ LASSERT(lcw != NULL);
+
+ spin_lock_irqsave(&lcw_pending_timers_lock, flags);
+ if (!list_empty(&lcw->lcw_list))
+ list_del_init(&lcw->lcw_list);
+ spin_unlock_irqrestore(&lcw_pending_timers_lock, flags);
+
+ lcw_update_time(lcw, "touched");
+ lcw->lcw_state = LC_WATCHDOG_ENABLED;
+
+ mod_timer(&lcw->lcw_timer, jiffies + lcw->lcw_time);
+
+ EXIT;
+}
+EXPORT_SYMBOL(lc_watchdog_touch);
+
+void lc_watchdog_disable(struct lc_watchdog *lcw)
+{
+ unsigned long flags;
+ ENTRY;
+ LASSERT(lcw != NULL);
+
+ spin_lock_irqsave(&lcw_pending_timers_lock, flags);
+ if (!list_empty(&lcw->lcw_list))
+ list_del_init(&lcw->lcw_list);
+ spin_unlock_irqrestore(&lcw_pending_timers_lock, flags);
+
+ lcw_update_time(lcw, "disabled");
+ lcw->lcw_state = LC_WATCHDOG_DISABLED;
+
+ EXIT;
+}
+EXPORT_SYMBOL(lc_watchdog_disable);
+
+void lc_watchdog_delete(struct lc_watchdog *lcw)
+{
+ unsigned long flags;
+ ENTRY;
+ LASSERT(lcw != NULL);
+
+ del_timer(&lcw->lcw_timer);
+
+ lcw_update_time(lcw, "deleted");
+
+ spin_lock_irqsave(&lcw_pending_timers_lock, flags);
+ if (!list_empty(&lcw->lcw_list))
+ list_del_init(&lcw->lcw_list);
+ spin_unlock_irqrestore(&lcw_pending_timers_lock, flags);
+
+ down(&lcw_refcount_sem);
+ if (--lcw_refcount == 0)
+ lcw_dispatch_stop();
+ up(&lcw_refcount_sem);
+
+ PORTAL_FREE(lcw, sizeof(*lcw));
+
+ EXIT;
+}
+EXPORT_SYMBOL(lc_watchdog_delete);
+
+/*
+ * Provided watchdog handlers
+ */
+
+extern void portals_debug_dumplog_internal(void *arg);
+
+void lc_watchdog_dumplog(struct lc_watchdog *lcw,
+ struct task_struct *tsk,
+ void *data)
+{
+ tsk = tsk ? tsk : current;
+ portals_debug_dumplog_internal((void *)(long)tsk->pid);
+}
+EXPORT_SYMBOL(lc_watchdog_dumplog);
continue;
if (nal->nal_refct != 0) {
- CWARN("NAL %d has outstanding refcount %d\n",
+ CWARN("NAL %x has outstanding refcount %d\n",
i, nal->nal_refct);
nal->nal_ni_fini(nal);
}
nal->nal_handle.nal_idx = (NI_HANDLE_MAGIC & ~NI_HANDLE_MASK) | interface;
nal->nal_handle.cookie = 0;
- CDEBUG(D_OTHER, "Starting up NAL (%d) refs %d\n", interface, nal->nal_refct);
+ CDEBUG(D_OTHER, "Starting up NAL (%x) refs %d\n", interface, nal->nal_refct);
rc = nal->nal_ni_init(nal, requested_pid, desired_limits, actual_limits);
if (rc != PTL_OK) {
- CERROR("Error %d starting up NAL %d, refs %d\n", rc,
+ CERROR("Error %d starting up NAL %x, refs %d\n", rc,
interface, nal->nal_refct);
GOTO(out, rc);
}
ptl_handle_ni_t nih;
ptl_process_id_t pid;
- CDEBUG (D_IOCTL, "Getting nid for nal [%d]\n", data->ioc_nal);
+ CDEBUG (D_IOCTL, "Getting nid for nal [%x]\n", data->ioc_nal);
err = PtlNIInit(data->ioc_nal, LUSTRE_SRV_PTL_PID, NULL,
NULL, &nih);
struct list_head *curr;
unsigned int generation;
off_t skip;
+ rwlock_t proc_route_rwlock;
} kpr_read_routes_data;
/* nal2name support re-used from utils/portals.c */
{ "tcp", SOCKNAL},
{ "gm", GMNAL},
{ "ib", OPENIBNAL},
+ { "iib", IIBNAL},
+ { "lo", LONAL},
{ NULL, -1}
};
static int kpr_proc_routes_read(char *page, char **start, off_t off,
int count, int *eof, void *data)
{
- struct proc_route_data *prd = data;
- kpr_route_entry_t *re;
- kpr_gateway_entry_t *ge;
- int chunk_len = 0;
- int line_len = 0;
- int user_len = 0;
+ struct proc_route_data *prd = data;
+ kpr_route_entry_t *re;
+ kpr_gateway_entry_t *ge;
+ int chunk_len = 0;
+ int line_len = 0;
+ int user_len = 0;
+ int rc = 0;
*eof = 1;
*start = page;
+ write_lock(&(prd->proc_route_rwlock));
+
if (prd->curr == NULL) {
if (off != 0)
- return 0;
+ goto routes_read_exit;
/* First pass, initialize our private data */
prd->curr = kpr_routes.next;
/* Abort route list generation change */
if (prd->generation != kpr_routes_generation) {
prd->curr = NULL;
- return sprintf(page, "\nError: Routes Changed\n");
+ rc = sprintf(page, "\nError: Routes Changed\n");
+ goto routes_read_exit;
}
/* All the routes have been walked */
if (prd->curr == &kpr_routes) {
prd->curr = NULL;
- return 0;
+ goto routes_read_exit;
}
}
if (prd->curr->next == NULL) {
prd->curr = NULL;
read_unlock(&kpr_rwlock);
- return sprintf(page, "\nError: Routes Changed\n");
+ rc = sprintf(page, "\nError: Routes Changed\n");
+ goto routes_read_exit;
}
prd->curr = prd->curr->next;
prd->curr = prd->curr->prev;
prd->skip = line_len - (user_len - count);
read_unlock(&kpr_rwlock);
- return count;
+ rc = count;
+ goto routes_read_exit;
}
/* Not enough data to entirely satify callers request */
prd->skip = 0;
read_unlock(&kpr_rwlock);
- return user_len;
+ rc = user_len;
+
+routes_read_exit:
+ write_unlock(&(prd->proc_route_rwlock));
+ return rc;
}
static int kpr_proc_routes_write(struct file *file, const char *ubuffer,
kpr_read_routes_data.curr = NULL;
kpr_read_routes_data.generation = 0;
kpr_read_routes_data.skip = 0;
+ kpr_read_routes_data.proc_route_rwlock = RW_LOCK_UNLOCKED;
routes_entry->data = &kpr_read_routes_data;
routes_entry->read_proc = kpr_proc_routes_read;
struct list_head *e;
kpr_nal_entry_t *ne;
- CDEBUG (D_NET, "Registering NAL %d\n", nalif->kprni_nalid);
+ CDEBUG (D_NET, "Registering NAL %x\n", nalif->kprni_nalid);
PORTAL_ALLOC (ne, sizeof (*ne));
if (ne == NULL)
{
write_unlock_irqrestore (&kpr_rwlock, flags);
- CERROR ("Attempt to register same NAL %d twice\n", ne->kpne_interface.kprni_nalid);
+ CERROR ("Attempt to register same NAL %x twice\n", ne->kpne_interface.kprni_nalid);
PORTAL_FREE (ne, sizeof (*ne));
return (-EEXIST);
kpr_upcall_t *u = kmalloc (sizeof (kpr_upcall_t), GFP_ATOMIC);
if (u == NULL) {
- CERROR ("Upcall out of memory: nal %d nid "LPX64" (%s) %s\n",
+ CERROR ("Upcall out of memory: nal %x nid "LPX64" (%s) %s\n",
gw_nalid, gw_nid,
portals_nid2str(gw_nalid, gw_nid, str),
alive ? "up" : "down");
struct list_head *n;
char str[PTL_NALFMT_SIZE];
- CDEBUG (D_NET, "%s notifying [%d] "LPX64": %s\n",
+ CDEBUG (D_NET, "%s notifying [%x] "LPX64": %s\n",
byNal ? "NAL" : "userspace",
gateway_nalid, gateway_nid, alive ? "up" : "down");
/* can't do predictions... */
do_gettimeofday (&now);
if (when > now.tv_sec) {
- CWARN ("Ignoring prediction from %s of [%d] "LPX64" %s "
+ CWARN ("Ignoring prediction from %s of [%x] "LPX64" %s "
"%ld seconds in the future\n",
byNal ? "NAL" : "userspace",
gateway_nalid, gateway_nid,
if (byNal) {
/* It wasn't userland that notified me... */
- CWARN ("Upcall: NAL %d NID "LPX64" (%s) is %s\n",
+ CWARN ("Upcall: NAL %x NID "LPX64" (%s) is %s\n",
gateway_nalid, gateway_nid,
portals_nid2str(gateway_nalid, gateway_nid, str),
alive ? "alive" : "dead");
unsigned long flags;
kpr_nal_entry_t *ne = (kpr_nal_entry_t *)arg;
- CDEBUG (D_NET, "Shutting down NAL %d\n", ne->kpne_interface.kprni_nalid);
+ CDEBUG (D_NET, "Shutting down NAL %x\n", ne->kpne_interface.kprni_nalid);
LASSERT (!ne->kpne_shutdown);
LASSERT (!in_interrupt());
unsigned long flags;
kpr_nal_entry_t *ne = (kpr_nal_entry_t *)arg;
- CDEBUG (D_NET, "Deregister NAL %d\n", ne->kpne_interface.kprni_nalid);
+ CDEBUG (D_NET, "Deregister NAL %x\n", ne->kpne_interface.kprni_nalid);
LASSERT (ne->kpne_shutdown); /* caller must have issued shutdown already */
LASSERT (!in_interrupt());
/* Wait until all outstanding messages/notifications have completed */
while (atomic_read (&ne->kpne_refcount) != 0)
{
- CDEBUG (D_NET, "Waiting for refcount on NAL %d to reach zero (%d)\n",
+ CDEBUG (D_NET, "Waiting for refcount on NAL %x to reach zero (%d)\n",
ne->kpne_interface.kprni_nalid, atomic_read (&ne->kpne_refcount));
set_current_state (TASK_UNINTERRUPTIBLE);
/* Caller wants to know if 'target_nid' can be reached via a gateway
* ON HER OWN NETWORK */
- CDEBUG (D_NET, "lookup "LPX64" from NAL %d\n", target_nid,
+ CDEBUG (D_NET, "lookup "LPX64" from NAL %x\n", target_nid,
ne->kpne_interface.kprni_nalid);
LASSERT (!in_interrupt());
/* NB can't deref 're' now; it might have been removed! */
- CDEBUG (D_NET, "lookup "LPX64" from NAL %d: %d ("LPX64")\n",
+ CDEBUG (D_NET, "lookup "LPX64" from NAL %x: %d ("LPX64")\n",
target_nid, ne->kpne_interface.kprni_nalid, rc,
(rc == 0) ? *gateway_nidp : (ptl_nid_t)0);
return (rc);
kpr_nal_entry_t *tmp_ne;
int rc;
- CDEBUG (D_NET, "forward [%p] "LPX64" from NAL %d\n", fwd,
+ CDEBUG (D_NET, "forward [%p] "LPX64" from NAL %x\n", fwd,
target_nid, src_ne->kpne_interface.kprni_nalid);
LASSERT (nob == lib_kiov_nob (fwd->kprfd_niov, fwd->kprfd_kiov));
read_unlock (&kpr_rwlock);
- CDEBUG (D_NET, "forward [%p] "LPX64" from NAL %d: "
- "to "LPX64" on NAL %d\n",
+ CDEBUG (D_NET, "forward [%p] "LPX64" from NAL %x: "
+ "to "LPX64" on NAL %x\n",
fwd, target_nid, src_ne->kpne_interface.kprni_nalid,
fwd->kprfd_gateway_nid, dst_ne->kpne_interface.kprni_nalid);
out:
kpr_fwd_errors++;
- CDEBUG (D_NET, "Failed to forward [%p] "LPX64" from NAL %d: %d\n",
+ CDEBUG (D_NET, "Failed to forward [%p] "LPX64" from NAL %x: %d\n",
fwd, target_nid, src_ne->kpne_interface.kprni_nalid, rc);
(fwd->kprfd_callback)(fwd->kprfd_callback_arg, rc);
kpr_nal_entry_t *dst_ne = (kpr_nal_entry_t *)arg;
kpr_nal_entry_t *src_ne = (kpr_nal_entry_t *)fwd->kprfd_router_arg;
- CDEBUG (D_NET, "complete(1) [%p] from NAL %d to NAL %d: %d\n", fwd,
+ CDEBUG (D_NET, "complete(1) [%p] from NAL %x to NAL %x: %d\n", fwd,
src_ne->kpne_interface.kprni_nalid, dst_ne->kpne_interface.kprni_nalid, error);
atomic_dec (&dst_ne->kpne_refcount); /* CAVEAT EMPTOR dst_ne can disappear now!!! */
(fwd->kprfd_callback)(fwd->kprfd_callback_arg, error);
- CDEBUG (D_NET, "complete(2) [%p] from NAL %d: %d\n", fwd,
+ CDEBUG (D_NET, "complete(2) [%p] from NAL %x: %d\n", fwd,
src_ne->kpne_interface.kprni_nalid, error);
atomic_dec (&kpr_queue_depth);
kpr_gateway_entry_t *ge;
int dup = 0;
- CDEBUG(D_NET, "Add route: %d "LPX64" : "LPX64" - "LPX64"\n",
+ CDEBUG(D_NET, "Add route: %x "LPX64" : "LPX64" - "LPX64"\n",
gateway_nalid, gateway_nid, lo_nid, hi_nid);
if (gateway_nalid == PTL_NID_ANY ||
struct list_head *e;
struct list_head *n;
- CDEBUG(D_NET, "Del route [%d] "LPX64" : "LPX64" - "LPX64"\n",
+ CDEBUG(D_NET, "Del route [%x] "LPX64" : "LPX64" - "LPX64"\n",
gw_nalid, gw_nid, lo, hi);
LASSERT(!in_interrupt());
break;
case NAL_CMD_ADD_ROUTE:
- CDEBUG(D_IOCTL, "Adding route: [%d] "LPU64" : "LPU64" - "LPU64"\n",
+ CDEBUG(D_IOCTL, "Adding route: [%x] "LPU64" : "LPU64" - "LPU64"\n",
pcfg->pcfg_nal, pcfg->pcfg_nid,
pcfg->pcfg_nid2, pcfg->pcfg_nid3);
err = kpr_add_route(pcfg->pcfg_gw_nal, pcfg->pcfg_nid,
break;
case NAL_CMD_DEL_ROUTE:
- CDEBUG (D_IOCTL, "Removing routes via [%d] "LPU64" : "LPU64" - "LPU64"\n",
+ CDEBUG (D_IOCTL, "Removing routes via [%x] "LPU64" : "LPU64" - "LPU64"\n",
pcfg->pcfg_gw_nal, pcfg->pcfg_nid,
pcfg->pcfg_nid2, pcfg->pcfg_nid3);
err = kpr_del_route (pcfg->pcfg_gw_nal, pcfg->pcfg_nid,
break;
case NAL_CMD_NOTIFY_ROUTER: {
- CDEBUG (D_IOCTL, "Notifying peer [%d] "LPU64" %s @ %ld\n",
+ CDEBUG (D_IOCTL, "Notifying peer [%x] "LPU64" %s @ %ld\n",
pcfg->pcfg_gw_nal, pcfg->pcfg_nid,
pcfg->pcfg_flags ? "Enabling" : "Disabling",
(time_t)pcfg->pcfg_nid3);
client->tsk = current;
client->args = args;
CDEBUG (D_OTHER, "pingcli_setup args: nid "LPX64" (%s), \
- nal %d, size %u, count: %u, timeout: %u\n",
+ nal %x, size %u, count: %u, timeout: %u\n",
args->ioc_nid,
portals_nid2str(args->ioc_nal, args->ioc_nid, str),
args->ioc_nal, args->ioc_size,
rc = PtlNIInit(args->ioc_nal, 0, NULL, NULL, &nih);
if (rc != PTL_OK || rc != PTL_IFACE_DUP)
{
- CERROR ("NAL %d not loaded\n", args->ioc_nal);
+ CERROR ("NAL %x not loaded\n", args->ioc_nal);
pingcli_shutdown (nih, 4);
return (NULL);
}
/* Aquire and initialize the proper nal for portals. */
rc = PtlNIInit(nal, 0, NULL, NULL, &server->ni);
if (!(rc == PTL_OK || rc == PTL_IFACE_DUP)) {
- CDEBUG (D_OTHER, "NAL %d not loaded\n", nal);
+ CDEBUG (D_OTHER, "NAL %x not loaded\n", nal);
return pingsrv_shutdown (4);
}
client->args = args;
CDEBUG (D_OTHER, "pingcli_setup args: nid "LPX64" (%s), \
- nal %d, size %u, count: %u, timeout: %u\n",
+ nal %x, size %u, count: %u, timeout: %u\n",
args->ioc_nid,
portals_nid2str(args->ioc_nid, args->ioc_nal, str),
args->ioc_nal, args->ioc_size,
rc = PtlNIInit(args->ioc_nal, 0, NULL, NULL, &nih);
if (rc != PTL_OK && rc != PTL_IFACE_DUP)
{
- CERROR ("NAL %d not loaded.\n", args->ioc_nal);
+ CERROR ("NAL %x not loaded.\n", args->ioc_nal);
pingcli_shutdown (nih, 4);
return (NULL);
}
rc = PtlNIInit(nal, 0, NULL, NULL, &server->ni);
if (rc != PTL_OK && rc != PTL_IFACE_DUP) {
- CDEBUG (D_OTHER, "Nal %d not loaded.\n", nal);
+ CDEBUG (D_OTHER, "Nal %x not loaded.\n", nal);
return pingsrv_shutdown (4);
}
#define MAX_MARK_SIZE 100
static const char *portal_debug_subsystems[] =
- {"undefined", "mdc", "mds", "osc", "ost", "class", "log", "llite",
- "rpc", "mgmt", "portals", "libcfs", "socknal", "qswnal", "pinger",
- "filter", "ptlbd", "echo", "ldlm", "lov", "gmnal", "router", "cobd",
- "ibnal", "lmv", "cmobd", "smfs", NULL};
+ {"undefined", "mdc", "mds", "osc",
+ "ost", "class", "log", "llite",
+ "rpc", "mgmt", "portals", "socknal",
+ "qswnal", "pinger", "filter", "ptlbd",
+ "echo", "ldlm", "lov", "gmnal",
+ "router", "cobd", "ibnal", "sm",
+ "asobd", "confobd", "lmv", "cmobd",
+ "lonal", NULL};
static const char *portal_debug_masks[] =
- {"trace", "inode", "super", "ext2", "malloc", "cache", "info", "ioctl",
- "blocks", "net", "warning", "buffs", "other", "dentry", "portals",
- "page", "dlmtrace", "error", "emerg", "ha", "rpctrace", "vfstrace",
- "reada", "mmap", NULL};
+ {"trace", "inode", "super", "ext2",
+ "malloc", "cache", "info", "ioctl",
+ "blocks", "net", "warning", "buffs",
+ "other", "dentry", "portals", "page",
+ "dlmtrace", "error", "emerg", "ha",
+ "rpctrace", "vfstrace", "reada", "mmap",
+ "config", NULL};
struct debug_daemon_cmd {
char *cmd;
return parse_buffer(in, out);
}
-const char debug_daemon_usage[]="usage: debug_daemon {start file [MB]|stop}\n";
+static int
+dbg_write_cmd(int fd, char *str)
+{
+ int len = strlen(str);
+ int rc = write(fd, str, len);
+
+ return (rc == len ? 0 : 1);
+}
+
+const char debug_daemon_usage[] = "usage: %s {start file [MB]|stop}\n";
+#define DAEMON_FILE "/proc/sys/portals/daemon_file"
int jt_dbg_debug_daemon(int argc, char **argv)
{
- int rc, fd;
+ int rc;
+ int fd;
if (argc <= 1) {
- fprintf(stderr, debug_daemon_usage);
- return 0;
+ fprintf(stderr, debug_daemon_usage, argv[0]);
+ return 1;
}
- fd = open("/proc/sys/portals/daemon_file", O_WRONLY);
+ fd = open(DAEMON_FILE, O_WRONLY);
if (fd < 0) {
- fprintf(stderr, "open(daemon_file) failed: %s\n",
+ fprintf(stderr, "open %s failed: %s\n", DAEMON_FILE,
strerror(errno));
- return 1;
+ return -1;
}
-
+
+ rc = -1;
if (strcasecmp(argv[1], "start") == 0) {
- if (argc != 3) {
- fprintf(stderr, debug_daemon_usage);
- return 1;
+ if (argc < 3 || argc > 4 ||
+ (argc == 4 && strlen(argv[3]) > 5)) {
+ fprintf(stderr, debug_daemon_usage, argv[0]);
+ goto out;
}
- rc = write(fd, argv[2], strlen(argv[2]));
- if (rc != strlen(argv[2])) {
- fprintf(stderr, "write(%s) failed: %s\n", argv[2],
- strerror(errno));
- close(fd);
- return 1;
+ if (argc == 4) {
+ char buf[12];
+ const long min_size = 10;
+ const long max_size = 20480;
+ long size;
+ char *end;
+
+ size = strtoul(argv[3], &end, 0);
+ if (size < min_size ||
+ size > max_size ||
+ *end != 0) {
+ fprintf(stderr, "size %s invalid, must be in "
+ "the range %ld-%ld MB\n", argv[3],
+ min_size, max_size);
+ goto out;
+ }
+
+ snprintf(buf, sizeof(buf), "size=%ld", size);
+ rc = dbg_write_cmd(fd, buf);
+ if (rc != 0) {
+ fprintf(stderr, "set %s failed: %s\n",
+ buf, strerror(errno));
+ goto out;
+ }
}
- } else if (strcasecmp(argv[1], "stop") == 0) {
- rc = write(fd, "stop", 4);
- if (rc != 4) {
- fprintf(stderr, "write(stop) failed: %s\n",
+
+ rc = dbg_write_cmd(fd, "start");
+ if (rc != 0) {
+ fprintf(stderr, "start debug_daemon on %s failed: %s\n",
+ argv[2], strerror(errno));
+ goto out;
+ }
+
+ rc = 0;
+ goto out;
+ }
+
+ if (strcasecmp(argv[1], "stop") == 0) {
+ rc = dbg_write_cmd(fd, "stop");
+ if (rc != 0) {
+ fprintf(stderr, "stopping debug_daemon failed: %s\n",
strerror(errno));
- close(fd);
- return 1;
+ goto out;
}
- } else {
- fprintf(stderr, debug_daemon_usage);
- return 1;
+
+ rc = 0;
+ goto out;
}
+ fprintf(stderr, debug_daemon_usage, argv[0]);
+ rc = -1;
+out:
close(fd);
- return 0;
+ return rc;
}
int jt_dbg_clear_debug_buf(int argc, char **argv)
#include <time.h>
#include <stdarg.h>
#include <endian.h>
+#if CRAY_PORTALS
+#ifdef REDSTORM
+#define __QK__
+#endif
+#include <portals/ipmap.h>
+#endif
#ifdef __CYGWIN__
static name2num_t nalnames[] = {
{"any", 0},
+#if !CRAY_PORTALS
{"tcp", SOCKNAL},
{"elan", QSWNAL},
{"gm", GMNAL},
{"openib", OPENIBNAL},
{"iib", IIBNAL},
{"lo", LONAL},
+#else
+ {"cray_kern_nal", CRAY_KERN_NAL},
+ {"cray_user_nal", CRAY_USER_NAL},
+ {"cray_qk_nal", CRAY_QK_NAL},
+#endif
{NULL, -1}
};
}
if (ptl_parse_ipaddr (&ipaddr, str) == 0) {
+#if !CRAY_PORTALS
*nidp = (ptl_nid_t)ipaddr;
+#else
+ *nidp = (((ptl_nid_t)ipaddr & PNAL_HOSTID_MASK) << PNAL_VNODE_SHIFT);
+#endif
return (0);
}
git://git.whamcloud.com / fs / lustre-release.git / commitdiff