cr: Support rlimits

Dump the with "new" prlimit syscall that works on arbitrary pid.

Restore is done in restorer _after_ mappings mixup and _before_
caps drop to make it set any max value.

The RLIM_INFINITY is handled explicitly to help future 64<->32
bits migration.

Signed-off-by: Pavel Emelyanov <xemul@parallels.com>

arch/x86/syscall-x86-64.def

@@ -61,6 +61,7 @@ __NR_setpriority	141		sys_setpriority		(int which, int who, int nice)
 __NR_sched_setscheduler	144		sys_sched_setscheduler	(int pid, int policy, struct sched_param *p)
 __NR_prctl		157		sys_prctl		(int option, unsigned long arg2, unsigned long arg3, unsigned long arg4, unsigned long arg5)
 __NR_arch_prctl		158		sys_arch_prctl		(int option, unsigned long addr)
+__NR_setrlimit		160		sys_setrlimit		(int resource, struct rlimit *rlim)
 __NR_mount		165		sys_mount		(char *dev_nmae, char *dir_name, char *type, unsigned long flags, void *data)
 __NR_umount2		166		sys_umount2		(char *name, int flags)
 __NR_gettid		186		sys_gettid		(void)

cr-dump.c

@@ -29,6 +29,7 @@
 #include "protobuf/mm.pb-c.h"
 #include "protobuf/creds.pb-c.h"
 #include "protobuf/core.pb-c.h"
+#include "protobuf/rlimit.pb-c.h"
 
 #include "asm/types.h"
 #include "list.h"
@@ -442,6 +443,36 @@ static int dump_task_fs(pid_t pid, struct parasite_dump_misc *misc, struct cr_fd
 	return pb_write_one(fdset_fd(fdset, CR_FD_FS), &fe, PB_FS);
 }
 
+static inline u_int64_t encode_rlim(unsigned long val)
+{
+	return val == RLIM_INFINITY ? -1 : val;
+}
+
+static int dump_task_rlims(int pid, struct cr_fdset *fds)
+{
+	int res, fd;
+
+	fd = fdset_fd(fds, CR_FD_RLIMIT);
+
+	for (res = 0; res < RLIM_NLIMITS; res++) {
+		struct rlimit lim;
+		RlimitEntry re = RLIMIT_ENTRY__INIT;
+
+		if (prlimit(pid, res, NULL, &lim)) {
+			pr_perror("Can't get rlimit %d", res);
+			return -1;
+		}
+
+		re.cur = encode_rlim(lim.rlim_cur);
+		re.max = encode_rlim(lim.rlim_max);
+
+		if (pb_write_one(fd, &re, PB_RLIMIT))
+			return -1;
+	}
+
+	return 0;
+}
+
 static int dump_filemap(pid_t pid, VmaEntry *vma, int file_fd,
 		const struct cr_fdset *fdset)
 {
@@ -1469,6 +1500,12 @@ static int dump_one_task(struct pstree_item *item)
 		goto err;
 	}
 
+	ret = dump_task_rlims(pid, cr_fdset);
+	if (ret) {
+		pr_err("Dump %d rlimits failed %d\n", pid, ret);
+		goto err;
+	}
+
 	close_cr_fdset(&cr_fdset);
 err:
 	close_pid_proc();

cr-restore.c

@@ -60,6 +60,7 @@
 #include "protobuf/sa.pb-c.h"
 #include "protobuf/itimer.pb-c.h"
 #include "protobuf/vma.pb-c.h"
+#include "protobuf/rlimit.pb-c.h"
 
 #include "asm/restore.h"
 
@@ -1568,6 +1569,58 @@ static int prep_sched_info(struct rst_sched_param *sp, ThreadCoreEntry *tc)
 	return 0;
 }
 
+static unsigned long decode_rlim(u_int64_t ival)
+{
+	return ival == -1 ? RLIM_INFINITY : ival;
+}
+
+static int prepare_rlimits(int pid, struct task_restore_core_args *ta)
+{
+	int fd, ret;
+
+	ta->nr_rlim = 0;
+
+	fd = open_image_ro(CR_FD_RLIMIT, pid);
+	if (fd < 0) {
+		if (errno == ENOENT) {
+			pr_info("Skip rlimits for %d\n", pid);
+			return 0;
+		}
+
+		return -1;
+	}
+
+	while (1) {
+		int l;
+		RlimitEntry *re;
+
+		ret = pb_read_one_eof(fd, &re, PB_RLIMIT);
+		if (ret <= 0)
+			break;
+
+		l = ta->nr_rlim;
+		if (l == RLIM_NLIMITS) {
+			pr_err("Too many rlimits in image for %d\n", pid);
+			ret = -1;
+			break;
+		}
+
+		ta->rlims[l].rlim_cur = decode_rlim(re->cur);
+		ta->rlims[l].rlim_max = decode_rlim(re->max);
+		if (ta->rlims[l].rlim_cur > ta->rlims[l].rlim_max) {
+			pr_err("Can't restore cur > max for %d.%d\n", pid, l);
+			ta->rlims[l].rlim_cur = ta->rlims[l].rlim_max;
+		}
+
+		rlimit_entry__free_unpacked(re, NULL);
+
+		ta->nr_rlim++;
+	}
+
+	close(fd);
+	return ret;
+}
+
 extern void __gcov_flush(void) __attribute__((weak));
 void __gcov_flush(void) {}
 
@@ -1730,6 +1783,9 @@ static int sigreturn_restore(pid_t pid, CoreEntry *core)
 
 	strncpy(task_args->comm, core->tc->comm, sizeof(task_args->comm));
 
+	if (prepare_rlimits(pid, task_args))
+		goto err;
+
 	/*
 	 * Fill up per-thread data.
 	 */

cr-show.c

@@ -129,6 +129,11 @@ void show_vmas(int fd_vma, struct cr_options *o)
 	pb_show_plain(fd_vma, PB_VMAS);
 }
 
+void show_rlimit(int fd, struct cr_options *o)
+{
+	pb_show_plain(fd, PB_RLIMIT);
+}
+
 static int nice_width_for(unsigned long addr)
 {
 	int ret = 3;

image.c

@@ -130,6 +130,7 @@ struct cr_fd_desc_tmpl fdset_template[CR_FD_MAX] = {
 	FD_ENTRY(TMPFS,		"tmpfs-%d.tar.gz", show_raw_image),
 	FD_ENTRY(TTY,		"tty",		 show_tty),
 	FD_ENTRY(TTY_INFO,	"tty-info",	 show_tty_info),
+	FD_ENTRY(RLIMIT,	"rlimit",	 show_rlimit),
 };
 
 static struct cr_fdset *alloc_cr_fdset(int nr)

include/crtools.h

@@ -32,6 +32,7 @@ enum {
 	CR_FD_ITIMERS,
 	CR_FD_CREDS,
 	CR_FD_FS,
+	CR_FD_RLIMIT,
 	_CR_FD_TASK_TO,
 
 	/*
@@ -154,6 +155,7 @@ void show_fown_cont(void *p);
 void show_eventfds(int fd, struct cr_options *o);
 void show_tty(int fd, struct cr_options *o);
 void show_tty_info(int fd, struct cr_options *o);
+void show_rlimit(int fd, struct cr_options *o);
 
 int check_img_inventory(void);
 int write_img_inventory(void);

include/image.h

@@ -61,6 +61,7 @@
 #define NETDEV_MAGIC		0x57373951 /* Yaroslavl */
 #define TTY_MAGIC		0x59433025 /* Pushkin */
 #define TTY_INFO_MAGIC		0x59453036 /* Kolpino */
+#define RLIMIT_MAGIC		0x57113925 /* Rostov */
 
 #define IFADDR_MAGIC		RAW_IMAGE_MAGIC
 #define ROUTE_MAGIC		RAW_IMAGE_MAGIC

include/protobuf.h

@@ -43,6 +43,7 @@ enum {
 	PB_INOTIFY_WD,
 	PB_TTY,
 	PB_TTY_INFO,
+	PB_RLIMIT,
 
 	PB_MAX
 };

include/restorer.h

@@ -3,6 +3,7 @@
 
 #include <signal.h>
 #include <limits.h>
+#include <sys/resource.h>
 
 #include "compiler.h"
 #include "asm/types.h"
@@ -122,6 +123,9 @@ struct task_restore_core_args {
 	u32				mm_saved_auxv_size;
 	char				comm[TASK_COMM_LEN];
 
+	int				nr_rlim;
+	struct rlimit			rlims[RLIM_NLIMITS];
+
 	int				*rst_tcp_socks;
 	int				rst_tcp_socks_size;
 } __aligned(sizeof(long));

include/syscall-types.h

@@ -61,4 +61,6 @@ struct robust_list_head;
 
 #define setns	sys_setns
 
+struct rlimit;
+
 #endif /* __CR_SYSCALL_TYPES_H__ */

pie/restorer.c

@@ -143,6 +143,14 @@ static void restore_sched_info(struct rst_sched_param *p)
 	sys_sched_setscheduler(0, p->policy, &parm);
 }
 
+static void restore_rlims(struct task_restore_core_args *ta)
+{
+	int r;
+
+	for (r = 0; r < ta->nr_rlim; r++)
+		sys_setrlimit(r, &ta->rlims[r]);
+}
+
 static int restore_thread_common(struct rt_sigframe *sigframe,
 		struct thread_restore_args *args)
 {
@@ -656,6 +664,8 @@ long __export_restore_task(struct task_restore_core_args *args)
 		sys_close(fd);
 	}
 
+	restore_rlims(args);
+
 	/* 
 	 * Writing to last-pid is CAP_SYS_ADMIN protected, thus restore
 	 * creds _after_ all threads creation.

protobuf.c

@@ -49,6 +49,7 @@
 #include "protobuf/netdev.pb-c.h"
 #include "protobuf/tcp-stream.pb-c.h"
 #include "protobuf/tty.pb-c.h"
+#include "protobuf/rlimit.pb-c.h"
 
 typedef size_t (*pb_getpksize_t)(void *obj);
 typedef size_t (*pb_pack_t)(void *obj, void *where);
@@ -125,6 +126,7 @@ void cr_pb_init(void)
 	CR_PB_DESC(PACKETSK,		PacketSock,	packet_sock);
 	CR_PB_DESC(TTY,			TtyFile,	tty_file);
 	CR_PB_DESC(TTY_INFO,		TtyInfo,	tty_info);
+	CR_PB_DESC(RLIMIT,		Rlimit,		rlimit);
 }
 
 /*

protobuf/Makefile

@@ -37,6 +37,7 @@ PROTO_FILES	+= vma.proto
 PROTO_FILES	+= core.proto
 PROTO_FILES	+= netdev.proto
 PROTO_FILES	+= tty.proto
+PROTO_FILES	+= rlimit.proto
 
 PROTO_HDRS	:= $(patsubst %.proto,%.pb-c.h,$(PROTO_FILES))
 PROTO_SRCS	:= $(patsubst %.proto,%.pb-c.c,$(PROTO_FILES))

protobuf/rlimit.proto

@@ -0,0 +1,4 @@
+message rlimit_entry {
+	required uint64		cur = 1;
+	required uint64		max = 2;
+}