summaryrefslogtreecommitdiffstats
path: root/sys/kern
diff options
context:
space:
mode:
Diffstat (limited to 'sys/kern')
-rw-r--r--sys/kern/exec_subr.c4
-rw-r--r--sys/kern/init_main.c4
-rw-r--r--sys/kern/kern_descrip.c8
-rw-r--r--sys/kern/kern_exec.c4
-rw-r--r--sys/kern/kern_exit.c13
-rw-r--r--sys/kern/kern_fork.c11
-rw-r--r--sys/kern/kern_resource.c190
-rw-r--r--sys/kern/kern_sig.c5
-rw-r--r--sys/kern/sys_generic.c4
-rw-r--r--sys/kern/vfs_vnops.c4
10 files changed, 201 insertions, 46 deletions
diff --git a/sys/kern/exec_subr.c b/sys/kern/exec_subr.c
index f2282a4a357..f32be4bddf4 100644
--- a/sys/kern/exec_subr.c
+++ b/sys/kern/exec_subr.c
@@ -1,4 +1,4 @@
-/* $OpenBSD: exec_subr.c,v 1.55 2018/04/12 17:13:44 deraadt Exp $ */
+/* $OpenBSD: exec_subr.c,v 1.56 2019/06/21 09:39:48 visa Exp $ */
/* $NetBSD: exec_subr.c,v 1.9 1994/12/04 03:10:42 mycroft Exp $ */
/*
@@ -351,7 +351,7 @@ exec_setup_stack(struct proc *p, struct exec_package *epp)
epp->ep_maxsaddr = USRSTACK - MAXSSIZ - MAXSSIZ_GUARD;
epp->ep_minsaddr = USRSTACK;
#endif
- epp->ep_ssize = round_page(p->p_rlimit[RLIMIT_STACK].rlim_cur);
+ epp->ep_ssize = round_page(lim_cur(RLIMIT_STACK));
if (stackgap_random != 0) {
sgap = arc4random() & (stackgap_random - 1);
diff --git a/sys/kern/init_main.c b/sys/kern/init_main.c
index b4bacaae1b7..a6a2b6da51e 100644
--- a/sys/kern/init_main.c
+++ b/sys/kern/init_main.c
@@ -1,4 +1,4 @@
-/* $OpenBSD: init_main.c,v 1.289 2019/06/20 14:55:22 anton Exp $ */
+/* $OpenBSD: init_main.c,v 1.290 2019/06/21 09:39:48 visa Exp $ */
/* $NetBSD: init_main.c,v 1.84.4.1 1996/06/02 09:08:06 mrg Exp $ */
/*
@@ -373,7 +373,7 @@ main(void *framep)
cpu_configure();
/* Configure virtual memory system, set vm rlimits. */
- uvm_init_limits(p);
+ uvm_init_limits(&limit0);
/* Per CPU memory allocation */
percpu_init();
diff --git a/sys/kern/kern_descrip.c b/sys/kern/kern_descrip.c
index 7da5dadb4ee..45fae009691 100644
--- a/sys/kern/kern_descrip.c
+++ b/sys/kern/kern_descrip.c
@@ -1,4 +1,4 @@
-/* $OpenBSD: kern_descrip.c,v 1.184 2019/05/13 17:31:51 deraadt Exp $ */
+/* $OpenBSD: kern_descrip.c,v 1.185 2019/06/21 09:39:48 visa Exp $ */
/* $NetBSD: kern_descrip.c,v 1.42 1996/03/30 22:24:38 christos Exp $ */
/*
@@ -353,7 +353,7 @@ restart:
FRELE(fp, p);
return (0);
}
- if ((u_int)new >= p->p_rlimit[RLIMIT_NOFILE].rlim_cur ||
+ if ((u_int)new >= lim_cur(RLIMIT_NOFILE) ||
(u_int)new >= maxfiles) {
FRELE(fp, p);
return (EBADF);
@@ -414,7 +414,7 @@ restart:
case F_DUPFD:
case F_DUPFD_CLOEXEC:
newmin = (long)SCARG(uap, arg);
- if ((u_int)newmin >= p->p_rlimit[RLIMIT_NOFILE].rlim_cur ||
+ if ((u_int)newmin >= lim_cur(RLIMIT_NOFILE) ||
(u_int)newmin >= maxfiles) {
error = EINVAL;
break;
@@ -864,7 +864,7 @@ fdalloc(struct proc *p, int want, int *result)
* expanding the ofile array.
*/
restart:
- lim = min((int)p->p_rlimit[RLIMIT_NOFILE].rlim_cur, maxfiles);
+ lim = min((int)lim_cur(RLIMIT_NOFILE), maxfiles);
last = min(fdp->fd_nfiles, lim);
if ((i = want) < fdp->fd_freefile)
i = fdp->fd_freefile;
diff --git a/sys/kern/kern_exec.c b/sys/kern/kern_exec.c
index 3019366cb1c..301d847f1c8 100644
--- a/sys/kern/kern_exec.c
+++ b/sys/kern/kern_exec.c
@@ -1,4 +1,4 @@
-/* $OpenBSD: kern_exec.c,v 1.205 2019/06/01 14:11:17 mpi Exp $ */
+/* $OpenBSD: kern_exec.c,v 1.206 2019/06/21 09:39:48 visa Exp $ */
/* $NetBSD: kern_exec.c,v 1.75 1996/02/09 18:59:28 christos Exp $ */
/*-
@@ -201,7 +201,7 @@ check_exec(struct proc *p, struct exec_package *epp)
/* check limits */
if ((epp->ep_tsize > MAXTSIZ) ||
- (epp->ep_dsize > p->p_rlimit[RLIMIT_DATA].rlim_cur))
+ (epp->ep_dsize > lim_cur(RLIMIT_DATA)))
error = ENOMEM;
if (!error)
diff --git a/sys/kern/kern_exit.c b/sys/kern/kern_exit.c
index 74ac47d02f8..ab2f0de086d 100644
--- a/sys/kern/kern_exit.c
+++ b/sys/kern/kern_exit.c
@@ -1,4 +1,4 @@
-/* $OpenBSD: kern_exit.c,v 1.177 2019/06/13 21:19:28 mpi Exp $ */
+/* $OpenBSD: kern_exit.c,v 1.178 2019/06/21 09:39:48 visa Exp $ */
/* $NetBSD: kern_exit.c,v 1.39 1996/04/22 01:38:25 christos Exp $ */
/*
@@ -328,6 +328,15 @@ exit1(struct proc *p, int rv, int flags)
KASSERT(pr->ps_refcnt > 0);
}
+ /* Release the thread's read reference of resource limit structure. */
+ if (p->p_limit != NULL) {
+ struct plimit *limit;
+
+ limit = p->p_limit;
+ p->p_limit = NULL;
+ lim_free(limit);
+ }
+
/*
* Other substructures are freed from reaper and wait().
*/
@@ -636,7 +645,7 @@ process_zap(struct process *pr)
free(pr->ps_ptstat, M_SUBPROC, sizeof(*pr->ps_ptstat));
pool_put(&rusage_pool, pr->ps_ru);
KASSERT(TAILQ_EMPTY(&pr->ps_threads));
- limfree(pr->ps_limit);
+ lim_free(pr->ps_limit);
crfree(pr->ps_ucred);
pool_put(&process_pool, pr);
nprocesses--;
diff --git a/sys/kern/kern_fork.c b/sys/kern/kern_fork.c
index 190c749720e..95180d9de31 100644
--- a/sys/kern/kern_fork.c
+++ b/sys/kern/kern_fork.c
@@ -1,4 +1,4 @@
-/* $OpenBSD: kern_fork.c,v 1.212 2019/06/01 14:11:17 mpi Exp $ */
+/* $OpenBSD: kern_fork.c,v 1.213 2019/06/21 09:39:48 visa Exp $ */
/* $NetBSD: kern_fork.c,v 1.29 1996/02/09 18:59:34 christos Exp $ */
/*
@@ -151,6 +151,7 @@ thread_new(struct proc *parent, vaddr_t uaddr)
p = pool_get(&proc_pool, PR_WAITOK);
p->p_stat = SIDL; /* protect against others */
p->p_flag = 0;
+ p->p_limit = NULL;
/*
* Make a proc table entry for the new process.
@@ -210,6 +211,8 @@ process_initialize(struct process *pr, struct proc *p)
LIST_INIT(&pr->ps_kqlist);
LIST_INIT(&pr->ps_sigiolst);
+ mtx_init(&pr->ps_mtx, IPL_MPFLOOR);
+
timeout_set(&pr->ps_realit_to, realitexpire, pr);
timeout_set(&pr->ps_rucheck_to, rucheck, pr);
}
@@ -237,12 +240,10 @@ process_new(struct proc *p, struct process *parent, int flags)
process_initialize(pr, p);
pr->ps_pid = allocpid();
+ lim_fork(parent, pr);
/* post-copy fixups */
pr->ps_pptr = parent;
- pr->ps_limit->pl_refcnt++;
- if (pr->ps_limit->pl_rlimit[RLIMIT_CPU].rlim_cur != RLIM_INFINITY)
- timeout_add_msec(&pr->ps_rucheck_to, RUCHECK_INTERVAL);
/* bump references to the text vnode (for sysctl) */
pr->ps_textvp = parent->ps_textvp;
@@ -373,7 +374,7 @@ fork1(struct proc *curp, int flags, void (*func)(void *), void *arg,
* Don't allow a nonprivileged user to exceed their current limit.
*/
count = chgproccnt(uid, 1);
- if (uid != 0 && count > curp->p_rlimit[RLIMIT_NPROC].rlim_cur) {
+ if (uid != 0 && count > lim_cur(RLIMIT_NPROC)) {
(void)chgproccnt(uid, -1);
nprocesses--;
nthreads--;
diff --git a/sys/kern/kern_resource.c b/sys/kern/kern_resource.c
index 3bc9425020a..050326ab6fd 100644
--- a/sys/kern/kern_resource.c
+++ b/sys/kern/kern_resource.c
@@ -1,4 +1,4 @@
-/* $OpenBSD: kern_resource.c,v 1.64 2019/06/10 03:15:53 visa Exp $ */
+/* $OpenBSD: kern_resource.c,v 1.65 2019/06/21 09:39:48 visa Exp $ */
/* $NetBSD: kern_resource.c,v 1.38 1996/10/23 07:19:38 matthias Exp $ */
/*-
@@ -53,9 +53,16 @@
#include <uvm/uvm_extern.h>
+/* Resource usage check interval in msec */
+#define RUCHECK_INTERVAL 1000
+
/* SIGXCPU interval in seconds of process runtime */
#define SIGXCPU_INTERVAL 5
+struct plimit *lim_copy(struct plimit *);
+struct plimit *lim_write_begin(void);
+void lim_write_commit(struct plimit *);
+
void tuagg_sub(struct tusage *, struct proc *);
/*
@@ -65,6 +72,13 @@ rlim_t maxdmap = MAXDSIZ;
rlim_t maxsmap = MAXSSIZ;
/*
+ * Serializes resource limit updates.
+ * This lock has to be held together with ps_mtx when updating
+ * the process' ps_limit.
+ */
+struct rwlock rlimit_lock = RWLOCK_INITIALIZER("rlimitlk");
+
+/*
* Resource controls and accounting.
*/
@@ -229,25 +243,27 @@ int
dosetrlimit(struct proc *p, u_int which, struct rlimit *limp)
{
struct rlimit *alimp;
+ struct plimit *limit;
rlim_t maxlim;
int error;
if (which >= RLIM_NLIMITS || limp->rlim_cur > limp->rlim_max)
return (EINVAL);
- alimp = &p->p_rlimit[which];
- if (limp->rlim_max > alimp->rlim_max)
- if ((error = suser(p)) != 0)
- return (error);
- if (p->p_p->ps_limit->pl_refcnt > 1) {
- struct plimit *l = p->p_p->ps_limit;
+ rw_enter_write(&rlimit_lock);
- /* limcopy() can sleep, so copy before decrementing refcnt */
- p->p_p->ps_limit = limcopy(l);
- limfree(l);
- alimp = &p->p_rlimit[which];
+ alimp = &p->p_p->ps_limit->pl_rlimit[which];
+ if (limp->rlim_max > alimp->rlim_max) {
+ if ((error = suser(p)) != 0) {
+ rw_exit_write(&rlimit_lock);
+ return (error);
+ }
}
+ /* Get exclusive write access to the limit structure. */
+ limit = lim_write_begin();
+ alimp = &limit->pl_rlimit[which];
+
switch (which) {
case RLIMIT_DATA:
maxlim = maxdmap;
@@ -316,6 +332,10 @@ dosetrlimit(struct proc *p, u_int which, struct rlimit *limp)
}
*alimp = *limp;
+
+ lim_write_commit(limit);
+ rw_exit_write(&rlimit_lock);
+
return (0);
}
@@ -326,16 +346,19 @@ sys_getrlimit(struct proc *p, void *v, register_t *retval)
syscallarg(int) which;
syscallarg(struct rlimit *) rlp;
} */ *uap = v;
- struct rlimit *alimp;
+ struct plimit *limit;
+ struct rlimit alimp;
int error;
if (SCARG(uap, which) < 0 || SCARG(uap, which) >= RLIM_NLIMITS)
return (EINVAL);
- alimp = &p->p_rlimit[SCARG(uap, which)];
- error = copyout(alimp, SCARG(uap, rlp), sizeof(struct rlimit));
+ limit = lim_read_enter();
+ alimp = limit->pl_rlimit[SCARG(uap, which)];
+ lim_read_leave(limit);
+ error = copyout(&alimp, SCARG(uap, rlp), sizeof(struct rlimit));
#ifdef KTRACE
if (error == 0 && KTRPOINT(p, KTR_STRUCT))
- ktrrlimit(p, alimp);
+ ktrrlimit(p, &alimp);
#endif
return (error);
}
@@ -507,8 +530,8 @@ ruadd(struct rusage *ru, struct rusage *ru2)
void
rucheck(void *arg)
{
+ struct rlimit rlim;
struct process *pr = arg;
- struct rlimit *rlim;
time_t runtime;
int s;
@@ -518,9 +541,12 @@ rucheck(void *arg)
runtime = pr->ps_tu.tu_runtime.tv_sec;
SCHED_UNLOCK(s);
- rlim = &pr->ps_limit->pl_rlimit[RLIMIT_CPU];
- if ((rlim_t)runtime >= rlim->rlim_cur) {
- if ((rlim_t)runtime >= rlim->rlim_max) {
+ mtx_enter(&pr->ps_mtx);
+ rlim = pr->ps_limit->pl_rlimit[RLIMIT_CPU];
+ mtx_leave(&pr->ps_mtx);
+
+ if ((rlim_t)runtime >= rlim.rlim_cur) {
+ if ((rlim_t)runtime >= rlim.rlim_max) {
prsignal(pr, SIGKILL);
} else if (runtime >= pr->ps_nextxcpu) {
prsignal(pr, SIGXCPU);
@@ -562,7 +588,7 @@ lim_startup(struct plimit *limit0)
* and copy when a limit is changed.
*/
struct plimit *
-limcopy(struct plimit *lim)
+lim_copy(struct plimit *lim)
{
struct plimit *newlim;
@@ -574,9 +600,129 @@ limcopy(struct plimit *lim)
}
void
-limfree(struct plimit *lim)
+lim_free(struct plimit *lim)
{
- if (--lim->pl_refcnt > 0)
+ if (atomic_dec_int_nv(&lim->pl_refcnt) > 0)
return;
pool_put(&plimit_pool, lim);
}
+
+void
+lim_fork(struct process *parent, struct process *child)
+{
+ struct plimit *limit;
+
+ mtx_enter(&parent->ps_mtx);
+ limit = parent->ps_limit;
+ atomic_inc_int(&limit->pl_refcnt);
+ mtx_leave(&parent->ps_mtx);
+
+ child->ps_limit = limit;
+
+ if (limit->pl_rlimit[RLIMIT_CPU].rlim_cur != RLIM_INFINITY)
+ timeout_add_msec(&child->ps_rucheck_to, RUCHECK_INTERVAL);
+}
+
+/*
+ * Return an exclusive write reference to the process' resource limit structure.
+ * The caller has to release the structure by calling lim_write_commit().
+ *
+ * This invalidates any plimit read reference held by the calling thread.
+ */
+struct plimit *
+lim_write_begin(void)
+{
+ struct plimit *limit;
+ struct proc *p = curproc;
+
+ rw_assert_wrlock(&rlimit_lock);
+
+ if (p->p_limit != NULL)
+ lim_free(p->p_limit);
+ p->p_limit = NULL;
+
+ /*
+ * It is safe to access ps_limit here without holding ps_mtx
+ * because rlimit_lock excludes other writers.
+ */
+
+ limit = p->p_p->ps_limit;
+ if (P_HASSIBLING(p) || limit->pl_refcnt > 1)
+ limit = lim_copy(limit);
+
+ return (limit);
+}
+
+/*
+ * Finish exclusive write access to the plimit structure.
+ * This makes the structure visible to other threads in the process.
+ */
+void
+lim_write_commit(struct plimit *limit)
+{
+ struct plimit *olimit;
+ struct proc *p = curproc;
+
+ rw_assert_wrlock(&rlimit_lock);
+
+ if (limit != p->p_p->ps_limit) {
+ mtx_enter(&p->p_p->ps_mtx);
+ olimit = p->p_p->ps_limit;
+ p->p_p->ps_limit = limit;
+ mtx_leave(&p->p_p->ps_mtx);
+
+ lim_free(olimit);
+ }
+}
+
+/*
+ * Begin read access to the process' resource limit structure.
+ * The access has to be finished by calling lim_read_leave().
+ *
+ * Sections denoted by lim_read_enter() and lim_read_leave() cannot nest.
+ */
+struct plimit *
+lim_read_enter(void)
+{
+ struct plimit *limit;
+ struct proc *p = curproc;
+ struct process *pr = p->p_p;
+
+ /*
+ * This thread might not observe the latest value of ps_limit
+ * if another thread updated the limits very recently on another CPU.
+ * However, the anomaly should disappear quickly, especially if
+ * there is any synchronization activity between the threads (or
+ * the CPUs).
+ */
+
+ limit = p->p_limit;
+ if (limit != pr->ps_limit) {
+ mtx_enter(&pr->ps_mtx);
+ limit = pr->ps_limit;
+ atomic_inc_int(&limit->pl_refcnt);
+ mtx_leave(&pr->ps_mtx);
+ if (p->p_limit != NULL)
+ lim_free(p->p_limit);
+ p->p_limit = limit;
+ }
+ KASSERT(limit != NULL);
+ return (limit);
+}
+
+/*
+ * Get the value of the resource limit in given process.
+ */
+rlim_t
+lim_cur_proc(struct proc *p, int which)
+{
+ struct process *pr = p->p_p;
+ rlim_t val;
+
+ KASSERT(which >= 0 && which < RLIM_NLIMITS);
+
+ mtx_enter(&pr->ps_mtx);
+ val = pr->ps_limit->pl_rlimit[which].rlim_cur;
+ mtx_leave(&pr->ps_mtx);
+ return (val);
+}
diff --git a/sys/kern/kern_sig.c b/sys/kern/kern_sig.c
index 4e3efa9c736..5738cd71af1 100644
--- a/sys/kern/kern_sig.c
+++ b/sys/kern/kern_sig.c
@@ -1,4 +1,4 @@
-/* $OpenBSD: kern_sig.c,v 1.230 2019/05/13 19:21:31 bluhm Exp $ */
+/* $OpenBSD: kern_sig.c,v 1.231 2019/06/21 09:39:48 visa Exp $ */
/* $NetBSD: kern_sig.c,v 1.54 1996/04/22 01:38:32 christos Exp $ */
/*
@@ -1548,8 +1548,7 @@ coredump(struct proc *p)
}
/* Don't dump if will exceed file size limit. */
- if (USPACE + ptoa(vm->vm_dsize + vm->vm_ssize) >=
- p->p_rlimit[RLIMIT_CORE].rlim_cur)
+ if (USPACE + ptoa(vm->vm_dsize + vm->vm_ssize) >= lim_cur(RLIMIT_CORE))
return (EFBIG);
if (incrash && nosuidcoredump == 3) {
diff --git a/sys/kern/sys_generic.c b/sys/kern/sys_generic.c
index ccb8502ed21..5fb4cf04258 100644
--- a/sys/kern/sys_generic.c
+++ b/sys/kern/sys_generic.c
@@ -1,4 +1,4 @@
-/* $OpenBSD: sys_generic.c,v 1.123 2019/01/21 23:41:26 cheloha Exp $ */
+/* $OpenBSD: sys_generic.c,v 1.124 2019/06/21 09:39:48 visa Exp $ */
/* $NetBSD: sys_generic.c,v 1.24 1996/03/29 00:25:32 cgd Exp $ */
/*
@@ -935,7 +935,7 @@ doppoll(struct proc *p, struct pollfd *fds, u_int nfds,
int timo, ncoll, i, s, error;
/* Standards say no more than MAX_OPEN; this is possibly better. */
- if (nfds > min((int)p->p_rlimit[RLIMIT_NOFILE].rlim_cur, maxfiles))
+ if (nfds > min((int)lim_cur(RLIMIT_NOFILE), maxfiles))
return (EINVAL);
/* optimize for the default case, of a small nfds value */
diff --git a/sys/kern/vfs_vnops.c b/sys/kern/vfs_vnops.c
index 2d5d00e5eff..c02f8f0cba1 100644
--- a/sys/kern/vfs_vnops.c
+++ b/sys/kern/vfs_vnops.c
@@ -1,4 +1,4 @@
-/* $OpenBSD: vfs_vnops.c,v 1.97 2018/08/20 16:00:22 mpi Exp $ */
+/* $OpenBSD: vfs_vnops.c,v 1.98 2019/06/21 09:39:48 visa Exp $ */
/* $NetBSD: vfs_vnops.c,v 1.20 1996/02/04 02:18:41 christos Exp $ */
/*
@@ -247,7 +247,7 @@ vn_fsizechk(struct vnode *vp, struct uio *uio, int ioflag, ssize_t *overrun)
*overrun = 0;
if (vp->v_type == VREG && p != NULL && !(ioflag & IO_NOLIMIT)) {
- rlim_t limit = p->p_rlimit[RLIMIT_FSIZE].rlim_cur;
+ rlim_t limit = lim_cur_proc(p, RLIMIT_FSIZE);
/* if already at or over the limit, send the signal and fail */
if (uio->uio_offset >= limit) {