- A+
前言
在写《[apue] 进程控制那些事儿》/"进程创建"/"更改进程用户 ID 和组 ID"一节时,发现 setreuid 更新实际用户 ID (RUID) 或有效用户 ID (EUID) 时,保存的设置用户 ID (saved set-user-id SUID) 只会随 EUID 变更,并不像 man 上说的会随 RUID 变更 (man setreuid):
If the real user ID is set (i.e., ruid is not -1) or the effective user ID is set to a value not equal to the previous real user ID, the saved set-user-ID will be set to the new effective user ID.
下面是实测结果:
调用参数 (root 身份) | RUID | EUID | SUID |
setreuid (bar, foo) | bar | foo | foo |
setreuid (foo, bar) | foo | bar | bar |
setreuid (-1, foo) | root | foo | foo |
setreuid (bar, -1) | bar | root | root |
setreuid (bar, bar) | bar | bar | bar |
setreuid (foo, foo) | foo | foo | foo |
特别是第 5 行 setreuid(bar, -1)
,RUID 变更为了 bar,SUID 仍保持 root 不变。
为了解答这个问题,找来系统对应版本的 linux 源码查看:
> uname -a Linux goodcitizen.bcc-gzhxy.baidu.com 3.10.0-1160.80.1.el7.x86_64 #1 SMP Tue Nov 8 15:48:59 UTC 2022 x86_64 x86_64 x86_64 GNU/Linux
这里是 3.10。之前推荐过 CodeBrowser,新近又发现一个源码阅读神器 bootlin:
能选代码库版本,功能上也比 CodeBrowser 好用,比如关键字搜索,不仅准确,还能把定义、声明、成员、引用区分清楚,很好的解决了 CodeBrowser 找不到代码的痛点,墙裂推荐~
问题复现
在撸源码之前,先复习一下这个问题的来龙去脉,为节省读者时间,把之前的 demo 贴上来 (setuid):
#include "../apue.h" #include <sys/types.h> #include <sys/file.h> #include <sys/stat.h> #include <unistd.h> void print_ids () { uid_t ruid = 0; uid_t euid = 0; uid_t suid = 0; int ret = getresuid (&ruid, &euid, &suid); if (ret == 0) printf ("%d: ruid %d, euid %d, suid %dn", getpid(), ruid, euid, suid); else err_sys ("getresuid"); } int main (int argc, char *argv[]) { if (argc == 2) { char* uid=argv[1]; int ret = setuid(atol(uid)); if (ret != 0) err_sys ("setuid"); print_ids(); } else if (argc == 3) { char* ruid=argv[1]; char* euid=argv[2]; int ret = setreuid(atol(ruid), atol(euid)); if (ret != 0) err_sys ("setreuid"); print_ids(); } else if (argc > 1) { char* uid=argv[1]; int ret = seteuid(atol(uid)); if (ret != 0) err_sys ("seteuid"); print_ids(); } else { print_ids(); } return 0; }
做个简单说明:
- print_ids 打印当前进程 3 个权限ID:RUID / EUID / SUID,其中用到的 getresuid 仅 Linux 支持,能展示用户当前的 SUID 值
- ./setuid 123:触发 setuid 调用,并打印调用后的结果
- ./setuid 123 456:触发 setreuid 调用,并打印调用后的结果
- ./setuid 123 456 789:触发 seteuid 调用,并打印调用后的结果。注意仅第一个参数有用,后两个占位
通过传递 setuid 程序不同参数,就可以验证不同的接口了,这里只需验证 setreuid,固定传递两个参数即可。把测试脚本也贴上来 (setreuid.sh):
#!/bin/sh groupadd test echo "create group ok" useradd -g test foo useradd -g test bar foo_uid=$(id -u foo) bar_uid=$(id -u bar) echo "create user ok" echo " foo: ${foo_uid}" echo " bar: ${bar_uid}" cd /tmp #chown bar:test ./setuid echo "test foo" ./setuid #chmod u+s ./setuid #echo "test set-uid bar" #su foo -c ./setuid echo "test setreuid(bar, foo)" ./setuid ${bar_uid} ${foo_uid} echo "test setreuid(foo, bar)" ./setuid ${foo_uid} ${bar_uid} echo "test setreuid(-1, foo)" ./setuid -1 ${foo_uid} echo "test setreuid(bar, -1)" ./setuid ${bar_uid} -1 echo "test setreuid(bar, bar)" ./setuid ${bar_uid} ${bar_uid} echo "test setreuid(foo, foo)" ./setuid ${foo_uid} ${foo_uid} userdel bar userdel foo echo "remove user ok" rm -rf /home/bar rm -rf /home/foo echo "remove user home ok" groupdel test echo "delete group ok"
自动创建测试账户并调用 setuid,验证了 6 种用例,需要使用超级用户身份启动:
> sudo sh setreuid.sh create group ok create user ok foo: 1003 bar: 1004 test foo 27253: ruid 0, euid 0, suid 0 test setreuid(bar, foo) 27254: ruid 1004, euid 1003, suid 1003 test setreuid(foo, bar) 27255: ruid 1003, euid 1004, suid 1004 test setreuid(-1, foo) 27256: ruid 0, euid 1003, suid 1003 test setreuid(bar, -1) 27257: ruid 1004, euid 0, suid 0 test setreuid(bar, bar) 27258: ruid 1004, euid 1004, suid 1004 test setreuid(foo, foo) 27259: ruid 1003, euid 1003, suid 1003 remove user ok remove user home ok
现象与表中列出的一致。
源码分析
在 kernel 3.10.0 版本中搜索 setreuid,没搜到,可能是 kernel 在系统函数上加了一堆宏识别不了,搜索 setuid 可以,它俩在同一个文件:
/* * Unprivileged users may change the real uid to the effective uid * or vice versa. (BSD-style) * * If you set the real uid at all, or set the effective uid to a value not * equal to the real uid, then the saved uid is set to the new effective uid. * * This makes it possible for a setuid program to completely drop its * privileges, which is often a useful assertion to make when you are doing * a security audit over a program. * * The general idea is that a program which uses just setreuid() will be * 100% compatible with BSD. A program which uses just setuid() will be * 100% compatible with POSIX with saved IDs. */ SYSCALL_DEFINE2(setreuid, uid_t, ruid, uid_t, euid) { struct user_namespace *ns = current_user_ns(); const struct cred *old; struct cred *new; int retval; kuid_t kruid, keuid; kruid = make_kuid(ns, ruid); keuid = make_kuid(ns, euid); if ((ruid != (uid_t) -1) && !uid_valid(kruid)) return -EINVAL; if ((euid != (uid_t) -1) && !uid_valid(keuid)) return -EINVAL; new = prepare_creds(); if (!new) return -ENOMEM; old = current_cred(); retval = -EPERM; if (ruid != (uid_t) -1) { new->uid = kruid; if (!uid_eq(old->uid, kruid) && !uid_eq(old->euid, kruid) && !nsown_capable(CAP_SETUID)) goto error; } if (euid != (uid_t) -1) { new->euid = keuid; if (!uid_eq(old->uid, keuid) && !uid_eq(old->euid, keuid) && !uid_eq(old->suid, keuid) && !nsown_capable(CAP_SETUID)) goto error; } if (!uid_eq(new->uid, old->uid)) { retval = set_user(new); if (retval < 0) goto error; } if (ruid != (uid_t) -1 || (euid != (uid_t) -1 && !uid_eq(keuid, old->uid))) new->suid = new->euid; new->fsuid = new->euid; retval = security_task_fix_setuid(new, old, LSM_SETID_RE); if (retval < 0) goto error; return commit_creds(new); error: abort_creds(new); return retval; }
代码不长没做删减,主体就是下面的框架:
... new = prepare_creds(); ... old = current_cred(); ... return commit_creds(new); error: abort_creds(new);
prepare_creds 返回的 new 代表新用户权限,会从当前权限复制一份;current_cred 返回的 old 代表原用户权限。经过对 new 的一番操作,如果成功就将它提交 (commit_creds),原权限被替换;否则回滚 (abort_creds),原权限不变。现在关注焦点转移到 new 的变更逻辑:
retval = -EPERM; if (ruid != (uid_t) -1) { new->uid = kruid; if (!uid_eq(old->uid, kruid) && !uid_eq(old->euid, kruid) && !nsown_capable(CAP_SETUID)) goto error; }
先看 ruid 参数,如果参数有效就将它设置到 new 的 uid,但需要同时满足以下条件:
- ruid == old->uid
- ruid == old->euid
- 原用户具有超级用户权限
否则出错。做为对比,再来看 euid 参数:
if (euid != (uid_t) -1) { new->euid = keuid; if (!uid_eq(old->uid, keuid) && !uid_eq(old->euid, keuid) && !uid_eq(old->suid, keuid) && !nsown_capable(CAP_SETUID)) goto error; }
与 ruid 差不多:
- euid == old->uid
- euid == old->euid
- euid == old->suid
- 原用户具有超级用户权限
多了一条规则,可以将 euid 设置为 old->suid。最后来看 SUID 的更新规则:
if (ruid != (uid_t) -1 || (euid != (uid_t) -1 && !uid_eq(keuid, old->uid))) new->suid = new->euid; new->fsuid = new->euid;
第三行准确无误的告诉我们,new ->suid
是固定被设置为 new->euid
的,时机是以下条件之一:
- ruid 有效
- euid 有效且 euid 与原 RUID 不同
看起来 ruid 参数只是影响 SUID 要不要从新 EUID 复制,即便 RUID 没变更、只要 ruid 参数有效就能产生这种作用。再回顾一下 man 的说明:
If the real user ID is set (i.e., ruid is not -1) or the effective user ID is set to a value not equal to the previous real user ID, the saved set-user-ID will be set to the new effective user ID.
简直就是代码的"直译",包括对 ruid 参数有效的说明、对 euid 参数与原 RUID 不同的说明、对 SUID 从新 EUID 复制的说明,一毛一样。之前把这里理解成 SUID 从 RUID 复制了,粗心大意了!
问题验证
了解 SUID 设置规则后,回头来看上面的表,有进一步的理解:
调用参数 (root 身份) | RUID | EUID | SUID | SUID 复制 |
setreuid (bar, foo) | bar * | foo * | foo | 条件 I |
setreuid (foo, bar) | foo * | bar * | bar | 条件 I |
setreuid (-1, foo) | root | foo * | foo | 条件 II |
setreuid (bar, -1) | bar * | root | root | 条件 I |
setreuid (bar, bar) | bar * | bar * | bar | 条件 I |
setreuid (foo, foo) | foo * | foo * | foo | 条件 I |
表中第二列中的星号表示 ruid 参数有效;第三列的星号表示 euid 变更 (!= old.uid);由于 2 个条件之间是短路或的关系,第一个条件满足后就不再检测第二个条件,所以需要最后一列表示 SUID 复制最终是哪个条件触发的。看起来大部分是条件一 ruid 有效,这些用例对条件二的测试不足,需要构造一组新的用例进行验证。
考查这样一个场景,将 demo 设置为 set-user-id 为 root,以普通用户 foo 启动该进程后:RUID = foo、EUID = SUID = root,此时 RUID 与 EUID 不同,满足了上述的条件二;保持 RUID 无效 (-1) 不满足条件一,是不是就能走条件二了?下面来做个测试 (setreuid-setroot.sh):
#!/bin/sh groupadd test echo "create group ok" useradd -g test foo useradd -g test bar foo_uid=$(id -u foo) bar_uid=$(id -u bar) root_uid=0 echo "create user ok" echo " foo: ${foo_uid}" echo " bar: ${bar_uid}" echo " root: ${root_uid}" cd /tmp chown root:test ./setuid echo "test foo" su foo -c ./setuid chmod u+s ./setuid echo "test set-uid root" su foo -c ./setuid echo "test setreuid(-1, foo)" su foo -c "./setuid -1 ${foo_uid}" echo "test setreuid(-1, bar)" su foo -c "./setuid -1 ${bar_uid}" echo "test setreuid(foo, foo)" su foo -c "./setuid ${foo_uid} ${foo_uid}" echo "test setreuid(root, foo)" su foo -c "./setuid ${root_uid} ${foo_uid}" userdel foo userdel bar echo "remove user ok" rm -rf /home/foo rm -rf /home/bar echo "remove user home ok" groupdel test echo "delete group ok"
与之前的最大区别是,这里使用 foo 用户身份启动测试程序 (su foo -c
),且它是 set-user-id 为 root 的。验证以下 4 种 setreuid 用例:
> sudo sh setreuid-setroot.sh create group ok create user ok foo: 1003 bar: 1004 root: 0 test foo 29332: ruid 1003, euid 1003, suid 1003 test set-uid root 29345: ruid 1003, euid 0, suid 0 test setreuid(-1, foo) 29357: ruid 1003, euid 1003, suid 0 test setreuid(-1, bar) 29369: ruid 1003, euid 1004, suid 1004 test setreuid(foo, foo) 29382: ruid 1003, euid 1003, suid 1003 test setreuid(root, foo) 29396: ruid 0, euid 1003, suid 1003 remove user ok remove user home ok delete group ok
清晰起见列表如下:
调用参数 (foo 身份 set-user-id root) | RUID | EUID | SUID | SUID 复制 |
启动后 | foo | root | root | n/a |
setreuid (-1, foo) | foo | foo | root | 未触发 |
setreuid (-1, bar) | foo | bar * | bar | 条件 II |
setreuid (foo, foo) | foo * | foo | foo | 条件 I |
setreuid (root, foo) | root * | foo | foo | 条件 I |
表中星号含义同前,下面分别解释:
- 第 3 行:仅设置 euid 为 foo,此时 ruid == -1 条件 1 不触发;euid == old.uid 条件 2 不触发,所以 SUID 保持 root 不变
- 第 4 行:仅设置 euid 为 bar,此时 ruid == -1 条件 1 不触发;euid != old.uid 条件 2 触发,所以 SUID 被复制为新 EUID:bar
- 第 5 行:同时设置 ruid 和 euid 为 foo,此时 ruid != -1 条件 1 触发;euid == old.uid 条件 2 不触发,所以 SUID 被复制为新 EUID:foo
- 第 6 行:设置 ruid 为 root,euid 为 foo,此时 ruid != -1 条件 1 触发;euid == old.uid 条件 2 不触发,最终 SUID 被复制为新 EUID:foo
与源码逻辑对应上了,正好也解释了原文《[apue] 进程控制那些事儿》中在这种场景下 setreuid(-1, foo)
用例中 SUID 保持为 root 的疑惑。
看看上表中第 3 行在 setreuid(-1,foo)
后的情形:RUID = EUID = foo,SUID = root,如果此时调用 setreuid(foo, -1)
按理说 SUID 会被更新为 foo,一试究竟:
uid_t ruid = 0; uid_t euid = 0; uid_t suid = 0; int ret = getresuid (&ruid, &euid, &suid); if (ret == 0) { printf ("%d: ruid %d, euid %d, suid %dn", getpid(), ruid, euid, suid); #ifdef TEST_UPDATE_RUID if (ruid == euid && euid != suid) { printf ("all uid same except suid %d, try to update ruidn", ruid); ret = setreuid (ruid, -1); if (ret != 0) err_sys ("setreuid"); else { getresuid (&ruid, &euid, &suid); printf ("%d: ruid %d, euid %d, suid %dn", getpid(), ruid, euid, suid); } } #endif } else err_sys ("getresuid");
在 print_ids 中检测到 RUID = EUID != SUID 时,将 RUID 重新设置一下,ruid 参数与原 RUID 一致。重新运行:
> sudo sh setreuid-setroot.sh ... test set-uid root 29511: ruid 1003, euid 0, suid 0 test setreuid(-1, foo) 29523: ruid 1003, euid 1003, suid 0 all uid same except suid 1003, try to update ruid 29523: ruid 1003, euid 1003, suid 1003 ...
SUID 果然随之变更了!这个用例更能说明问题,因为调用 setreuid 前后 RUID 与 EUID 没有发生改变,SUID 却因为 ruid 参数有效而发生了变更,有点意思。下面的表总结了上述过程:
调用参数 (foo 身份 set-user-id root) | RUID | EUID | SUID | SUID 复制 |
启动后 | foo | root | root | n/a |
setreuid (-1, foo) | foo | foo | root | 未触发 |
setreuid (foo, -1) | foo * | foo | foo | 条件 I |
这个表与之前不同的是所有 setreuid 调用均在一个进程中。
意义探寻
知其然,还要知其所以然,上面的探索只是第一步,对于 SUID 复制 EUID 的目的,《[apue] 进程控制那些事儿》已有讨论,这里聚焦 SUID 何时复制 EUID,按照直觉设计成下面的条件看起来更通顺:
ruid != old.uid || euid != old.euid
为了解答这个疑问,就按照设想的条件重跑一下 setreuid 的所有用例:
调用参数 (root 身份) | RUID | EUID | SUID |
setreuid (bar, foo) | bar | foo | foo |
setreuid (foo, bar) | foo | bar | bar |
setreuid (-1, foo) | root | foo | foo |
setreuid (bar, -1) | bar | root | root |
setreuid (bar, bar) | bar | bar | bar |
setreuid (foo, foo) | foo | foo | foo |
调用参数 (foo 身份 set-user-id root) | RUID | EUID | SUID |
启动后 | foo | root | root |
setreuid (-1, foo) | foo | foo | foo |
setreuid (-1, bar) | foo | bar | bar |
setreuid (foo, foo) | foo | foo | foo |
setreuid (root, foo) | root | foo | foo |
发现只有一个用例的结果会发生变化 (表中高亮字体):进程 set-user-id 为 root 且以普通用户身份启动 setreuid(-1, foo)
,从 SUID 不变到现在 SUID 跟随 EUID 改变,这导致整个进程变为普通进程失去重新转变为特权进程的机会。再看这个调用形式特别眼熟,这不就是 seteuid 嘛!它在改变 EUID 时是不希望 SUID 变更的,所以这下全明白了:setreuid 这样的设计是为了给 seteuid 切换特权身份留后门,从而有机会再切换回之前的身份。
结语
关于 seteuid,man 中有一段说明:
Under libc4, libc5 and glibc 2.0 seteuid(euid) is equivalent to setreuid(-1, euid) and hence may change the saved set-user-ID. Under glibc 2.1 and later it is equivalent to setresuid(-1, euid, -1) and hence does not change the saved set-user-ID. Analogous remarks hold for setegid(), with the difference that the change in implementation from setregid(-1, egid) to setresgid(-1, egid, -1) occurred in glibc 2.2 or 2.3 (dependeing on the hardware architecture).
大意是说 seteuid 到底等价于setreuid(-1,euid)
还是setresuid(-1,euid,-1)
要看 glibc 版本,前者在改变 SUID 的逻辑上遵循上面的讨论;后者不遵循,或者说 SUID 不会随 EUID 变更。之前曾经比对过 setuid / setreuid / seteuid,并且推荐使用 seteuid,如果 seteuid 只是 setreuid 的分身,则它们的区别没想象那么大,只是写起来更方便一些。
后记
文章最后再推荐一波 bootlin:
代码库除了 kernel,还可以选择 freebsd、glibc、qemu、dpdk、grub、llvm、busybox 等,点击符号跳转,使用浏览器后退做调用栈回退,非常方便。