#include "vserver.h"
static int
-create_context(xid_t ctx, uint32_t flags, uint64_t bcaps, const rspec_t *rspec)
+create_context(xid_t ctx, uint32_t flags, uint64_t bcaps)
{
struct vc_ctx_caps vc_caps;
- struct vc_ctx_flags vc_flags;
- struct vc_rlimit vc_rlimit;
/*
* Create context info - this sets the STATE_SETUP and STATE_INIT flags.
* Don't ever clear the STATE_INIT flag, that makes us the init task.
+ *
+ * XXX - the kernel code allows initial flags to be passed as an arg.
*/
if (vc_ctx_create(ctx) == VC_NOCTX)
return -1;
if (vc_set_ccaps(ctx, &vc_caps))
return -1;
- /* set scheduler parameters */
- if (pl_setsched(ctx, rspec->cpu_share, rspec->cpu_sched_flags))
- return -1;
+ /* set default scheduling parameters */
+ pl_setsched(ctx, 32, 0);
- /* set resource limits */
- vc_rlimit.min = VC_LIM_KEEP;
- vc_rlimit.soft = VC_LIM_KEEP;
- vc_rlimit.hard = rspec->mem_limit;
- if (vc_set_rlimit(ctx, RLIMIT_RSS, &vc_rlimit))
- return -1;
+ return 0;
+}
- /* assume min and soft unchanged by set_rlimit */
- vc_rlimit.hard = rspec->task_limit;
- if (vc_set_rlimit(ctx, RLIMIT_NPROC, &vc_rlimit))
- return -1;
+int
+pl_setup_done(xid_t ctx)
+{
+ struct vc_ctx_flags vc_flags;
- /* set flags, unset SETUP flag - this allows other processes to migrate */
- vc_flags.mask = VC_VXF_STATE_SETUP | VC_VXF_SCHED_FLAGS;
- flags = 0; /* XXX - ignore flags parameter */
- vc_flags.flagword = flags | rspec->cpu_sched_flags; /* SETUP cleared */
+ /* unset SETUP flag - this allows other processes to migrate */
+ vc_flags.mask = VC_VXF_STATE_SETUP;
+ vc_flags.flagword = 0;
if (vc_set_cflags(ctx, &vc_flags))
return -1;
return 0;
}
+#define RETRY_LIMIT 10
+
int
-pl_chcontext(xid_t ctx, uint32_t flags, uint64_t bcaps, const rspec_t *rspec)
+pl_chcontext(xid_t ctx, uint32_t flags, uint64_t bcaps)
{
+ int retry_count = 0;
+
for (;;)
{
struct vc_ctx_flags vc_flags;
if (vc_get_cflags(ctx, &vc_flags))
{
+ if (errno != ESRCH)
+ return -1;
+
/* context doesn't exist - create it */
- if (create_context(ctx, flags, bcaps, rspec))
+ if (create_context(ctx, flags, bcaps))
{
if (errno == EEXIST)
/* another process beat us in a race */
}
/* created context and migrated to it i.e., we're done */
- break;
+ return 1;
}
/* check the SETUP flag */
if (vc_flags.flagword & VC_VXF_STATE_SETUP)
{
/* context is still being setup - wait a while then retry */
+ if (retry_count++ >= RETRY_LIMIT)
+ {
+ errno = EBUSY;
+ return -1;
+ }
sleep(1);
continue;
}
} \
while (0)
-
int
pl_setsched(xid_t ctx, uint32_t cpu_share, uint32_t cpu_sched_flags)
{
struct vc_set_sched vc_sched;
struct vc_ctx_flags vc_flags;
-
- if (cpu_sched_flags & ~VC_VXF_SCHED_FLAGS)
- {
- errno = EINVAL;
- return -1;
- }
+ uint32_t new_flags;
vc_sched.set_mask = (VC_VXSM_FILL_RATE | VC_VXSM_INTERVAL | VC_VXSM_TOKENS |
VC_VXSM_TOKENS_MIN | VC_VXSM_TOKENS_MAX);
/* get current flag values */
VC_SYSCALL(vc_get_cflags(ctx, &vc_flags));
- /* the only flag which ever changes is the SCHED_SHARE bit */
- if ((vc_flags.flagword ^ cpu_sched_flags) & VC_VXF_SCHED_SHARE)
+ /* guaranteed CPU corresponds to SCHED_SHARE flag being cleared */
+ new_flags = (cpu_sched_flags & VS_SCHED_CPU_GUARANTEED
+ ? 0
+ : VC_VXF_SCHED_SHARE);
+ if ((vc_flags.flagword & VC_VXF_SCHED_SHARE) != new_flags)
{
- vc_flags.mask = VC_VXF_SCHED_SHARE;
- vc_flags.flagword = cpu_sched_flags & VC_VXF_SCHED_FLAGS;
+ vc_flags.mask = VC_VXF_SCHED_FLAGS;
+ vc_flags.flagword = new_flags | VC_VXF_SCHED_HARD;
VC_SYSCALL(vc_set_cflags(ctx, &vc_flags));
}
#ifndef _LIB_PLANETLAB_H_
#define _LIB_PLANETLAB_H_
-/*
- * context create
- */
-typedef struct {
- uint32_t cpu_share;
- uint32_t cpu_sched_flags;
- uint64_t mem_limit;
- uint64_t task_limit;
-} rspec_t;
-
#define VC_VXF_SCHED_FLAGS (VC_VXF_SCHED_HARD | VC_VXF_SCHED_SHARE)
int
-pl_chcontext(xid_t ctx, uint32_t flags, uint64_t bcaps, const rspec_t *rspec);
+pl_chcontext(xid_t ctx, uint32_t flags, uint64_t bcaps);
+
+int
+pl_setup_done(xid_t ctx);
int
pl_setsched(xid_t ctx, uint32_t cpu_share, uint32_t cpu_sched_flags);
+/* scheduler flags */
+#define VS_SCHED_CPU_GUARANTEED 1
+
#endif
#define NONE ({ Py_INCREF(Py_None); Py_None; })
-static int
-get_rspec(PyObject *resources, rspec_t *rspec)
-{
- int result = -1;
- PyObject *cpu_share;
- PyObject *sched_flags = NULL;
-
- if (!PyMapping_Check(resources))
- {
- PyErr_SetString(PyExc_TypeError, "invalid rspec");
- return -1;
- }
-
- /* get CPU share */
- if (!(cpu_share = PyMapping_GetItemString(resources, "nm_cpu_share")))
- return -1;
- if (!PyInt_Check(cpu_share))
- {
- PyErr_SetString(PyExc_TypeError, "nm_cpu_share not an integer");
- goto out;
- }
- rspec->cpu_share = PyInt_AS_LONG(cpu_share);
-
- /* check whether this share should be guaranteed */
- rspec->cpu_sched_flags = VC_VXF_SCHED_FLAGS;
- result = 0;
- if ((sched_flags = PyMapping_GetItemString(resources, "nm_sched_flags")))
- {
- const char *flagstr;
-
- if (!(flagstr = PyString_AsString(sched_flags)))
- result = -1;
- else if (!strcmp(flagstr, "guaranteed"))
- rspec->cpu_sched_flags &= ~VC_VXF_SCHED_SHARE;
- Py_DECREF(sched_flags);
- }
- else
- /* not an error if nm_sched_flags is missing */
- PyErr_Clear();
-
- out:
- Py_DECREF(cpu_share);
-
- return result;
-}
-
/*
* context create
*/
static PyObject *
vserver_chcontext(PyObject *self, PyObject *args)
{
+ int result;
xid_t ctx;
uint32_t flags = 0;
uint32_t bcaps = ~vc_get_insecurebcaps();
- rspec_t rspec = { 32, VC_VXF_SCHED_FLAGS, -1, -1 };
- PyObject *resources;
- if (!PyArg_ParseTuple(args, "IO|K", &ctx, &resources, &flags) ||
- get_rspec(resources, &rspec))
+ if (!PyArg_ParseTuple(args, "I|K", &ctx, &flags))
+ return NULL;
+
+ if ((result = pl_chcontext(ctx, flags, bcaps)) < 0)
+ return PyErr_SetFromErrno(PyExc_OSError);
+
+ return PyBool_FromLong(result);
+}
+
+static PyObject *
+vserver_setup_done(PyObject *self, PyObject *args)
+{
+ xid_t ctx;
+
+ if (!PyArg_ParseTuple(args, "I", &ctx))
return NULL;
- if (pl_chcontext(ctx, flags, bcaps, &rspec))
+ if (pl_setup_done(ctx) < 0)
return PyErr_SetFromErrno(PyExc_OSError);
return NONE;
vserver_setsched(PyObject *self, PyObject *args)
{
xid_t ctx;
- rspec_t rspec = { 32, VC_VXF_SCHED_FLAGS, -1, -1 };
- PyObject *resources;
+ uint32_t cpu_share;
+ uint32_t cpu_sched_flags = VC_VXF_SCHED_FLAGS;
- if (!PyArg_ParseTuple(args, "IO", &ctx, &resources) ||
- get_rspec(resources, &rspec))
+ if (!PyArg_ParseTuple(args, "II|I", &ctx, &cpu_share, &cpu_sched_flags))
return NULL;
/* ESRCH indicates that there are no processes in the context */
- if (pl_setsched(ctx, rspec.cpu_share, rspec.cpu_sched_flags) &&
+ if (pl_setsched(ctx, cpu_share, cpu_sched_flags) &&
errno != ESRCH)
return PyErr_SetFromErrno(PyExc_OSError);
static PyMethodDef methods[] = {
{ "chcontext", vserver_chcontext, METH_VARARGS,
"chcontext to vserver with provided flags" },
+ { "setup_done", vserver_setup_done, METH_VARARGS,
+ "Release vserver setup lock" },
{ "setsched", vserver_setsched, METH_VARARGS,
"Change vserver scheduling attributes for given vserver context" },
{ "setdlimit", vserver_set_dlimit, METH_VARARGS,
/* export limit-related constants */
PyModule_AddIntConstant(mod, "DLIMIT_KEEP", (int)CDLIM_KEEP);
PyModule_AddIntConstant(mod, "DLIMIT_INF", (int)CDLIM_INFINITY);
+
+ /* scheduler flags */
+ PyModule_AddIntConstant(mod,
+ "VS_SCHED_CPU_GUARANTEED",
+ VS_SCHED_CPU_GUARANTEED);
}
exit(1);
}
#else
- rspec_t rspec;
+ int ctx_is_new;
unsigned long long cpu = VC_LIM_KEEP;
unsigned long long mem = VC_LIM_KEEP;
unsigned long long task = VC_LIM_KEEP;
(void) (sandbox_chroot(ctx));
- rspec.cpu_share = cpu;
- rspec.cpu_sched_flags = (VC_VXF_SCHED_HARD |
- (cpuguaranteed ? 0 : VC_VXF_SCHED_SHARE));
- rspec.mem_limit = mem;
- rspec.task_limit = task;
- if (pl_chcontext(ctx, 0, ~vc_get_insecurebcaps(), &rspec))
+ if ((ctx_is_new = pl_chcontext(ctx, 0, ~vc_get_insecurebcaps())) < 0)
{
PERROR("pl_chcontext(%u)", ctx);
exit(1);
}
+ if (ctx_is_new)
+ {
+ /* set resources */
+ struct vc_rlimit limits;
+
+ limits.min = VC_LIM_KEEP;
+ limits.soft = VC_LIM_KEEP;
+ limits.hard = mem;
+ if (vc_set_rlimit(ctx, RLIMIT_RSS, &limits))
+ {
+ PERROR("pl_setrlimit(%u, RLIMIT_RSS)", ctx);
+ exit(1);
+ }
+ limits.hard = task;
+ if (vc_set_rlimit(ctx, RLIMIT_NPROC, &limits))
+ {
+ PERROR("pl_setrlimit(%u, RLIMIT_NPROC)", ctx);
+ exit(1);
+ }
+ cpuguaranteed &= VS_SCHED_CPU_GUARANTEED;
+ if (pl_setsched(ctx, cpu, cpuguaranteed) < 0)
+ {
+ PERROR("pl_setsched(&u)", ctx);
+ exit(1);
+ }
+ pl_setup_done(ctx);
+ }
#endif
return 0;
}