d/freebsd-dev
1
0
Fork 0
Code Issues Pull Requests Packages Projects Releases Wiki Activity

epoll: Store epoll_event udata member in ext member of kevent.

Browse Source 
Current epoll implementation stores udata fields of epoll_event
structure in special dynamically-sized table rather than in udata field
of backing kevent structure because of 2 reasons:
1. Kevent's udata size is smaller than epoll's on 32-bit archs.
2. Kevent's udata can be clobbered on execution EPOLL_CTL_ADD as kqueue
   modifies existing event while epoll returns error in this case.

After r320043 has introduced four new 64bit user data members (ext[]),
we can store epoll udata in one of them and drop aforementioned table.
According to kqueue_register() source code ext members are not updated
when existing kevent is modified that fixes p.2.

As a side effect the patch fixes PR/252582.

Reviewed by:	trasz
MFC after:	1 month
Differential revision:	https://reviews.freebsd.org/D28169
This commit is contained in:
Vladimir Kondratyev 2021-02-08 02:46:14 +03:00
parent d26f2a50ff
commit b3c6fe663b
3 changed files with 11 additions and 99 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

21
sys/compat/linux/linux_emul.c
View File

@ -143,7 +143,6 @@ linux_proc_init(struct thread *td, struct thread *newtd, int flags)
{
struct linux_emuldata *em;
struct linux_pemuldata *pem;
struct epoll_emuldata *emd;
struct proc *p;
if (newtd != NULL) {
@ -185,15 +184,9 @@ linux_proc_init(struct thread *td, struct thread *newtd, int flags)
em->child_clear_tid = NULL;
em->child_set_tid = NULL;
/* epoll should be destroyed in a case of exec. */
pem = pem_find(p);
KASSERT(pem != NULL, ("proc_exit: proc emuldata not found.\n"));
pem->persona = 0;
if (pem->epoll != NULL) {
emd = pem->epoll;
pem->epoll = NULL;
free(emd, M_EPOLL);
}
}
}
@ -202,7 +195,6 @@ void
linux_on_exit(struct proc *p)
{
struct linux_pemuldata *pem;
struct epoll_emuldata *emd;
struct thread *td = curthread;
MPASS(SV_CURPROC_ABI() == SV_ABI_LINUX);
@ -217,12 +209,6 @@ linux_on_exit(struct proc *p)
p->p_emuldata = NULL;
if (pem->epoll != NULL) {
emd = pem->epoll;
pem->epoll = NULL;
free(emd, M_EPOLL);
}
sx_destroy(&pem->pem_sx);
free(pem, M_LINUX);
}
@ -267,7 +253,6 @@ int
linux_common_execve(struct thread *td, struct image_args *eargs)
{
struct linux_pemuldata *pem;
struct epoll_emuldata *emd;
struct vmspace *oldvmspace;
struct linux_emuldata *em;
struct proc *p;
@ -299,12 +284,6 @@ linux_common_execve(struct thread *td, struct image_args *eargs)
p->p_emuldata = NULL;
PROC_UNLOCK(p);
if (pem->epoll != NULL) {
emd = pem->epoll;
pem->epoll = NULL;
free(emd, M_EPOLL);
}
free(em, M_TEMP);
free(pem, M_LINUX);
}

1
sys/compat/linux/linux_emul.h
View File

@ -68,7 +68,6 @@ int linux_common_execve(struct thread *, struct image_args *);
struct linux_pemuldata {
uint32_t flags; /* process emuldata flags */
struct sx pem_sx; /* lock for this struct */
void *epoll; /* epoll data */
uint32_t persona; /* process execution domain */
uint32_t ptrace_flags; /* used by ptrace(2) */
};

88
sys/compat/linux/linux_event.c
View File

@ -71,25 +71,8 @@ __FBSDID("$FreeBSD$");
#include <compat/linux/linux_timer.h>
#include <compat/linux/linux_util.h>
/*
* epoll defines 'struct epoll_event' with the field 'data' as 64 bits
* on all architectures. But on 32 bit architectures BSD 'struct kevent' only
* has 32 bit opaque pointer as 'udata' field. So we can't pass epoll supplied
* data verbatuim. Therefore we allocate 64-bit memory block to pass
* user supplied data for every file descriptor.
*/
typedef uint64_t epoll_udata_t;
struct epoll_emuldata {
uint32_t fdc; /* epoll udata max index */
epoll_udata_t udata[1]; /* epoll user data vector */
};
#define EPOLL_DEF_SZ 16
#define EPOLL_SIZE(fdn) \
(sizeof(struct epoll_emuldata)+(fdn) * sizeof(epoll_udata_t))
struct epoll_event {
uint32_t events;
epoll_udata_t data;
@ -101,7 +84,6 @@ __attribute__((packed))
#define LINUX_MAX_EVENTS (INT_MAX / sizeof(struct epoll_event))
static void epoll_fd_install(struct thread *td, int fd, epoll_udata_t udata);
static int epoll_to_kevent(struct thread *td, int fd,
struct epoll_event *l_event, struct kevent *kevent,
int *nkevents);
@ -175,47 +157,11 @@ struct timerfd {
static void linux_timerfd_expire(void *);
static void linux_timerfd_curval(struct timerfd *, struct itimerspec *);
static void
epoll_fd_install(struct thread *td, int fd, epoll_udata_t udata)
{
struct linux_pemuldata *pem;
struct epoll_emuldata *emd;
struct proc *p;
p = td->td_proc;
pem = pem_find(p);
KASSERT(pem != NULL, ("epoll proc emuldata not found.\n"));
LINUX_PEM_XLOCK(pem);
if (pem->epoll == NULL) {
emd = malloc(EPOLL_SIZE(fd), M_EPOLL, M_WAITOK);
emd->fdc = fd;
pem->epoll = emd;
} else {
emd = pem->epoll;
if (fd > emd->fdc) {
emd = realloc(emd, EPOLL_SIZE(fd), M_EPOLL, M_WAITOK);
emd->fdc = fd;
pem->epoll = emd;
}
}
emd->udata[fd] = udata;
LINUX_PEM_XUNLOCK(pem);
}
static int
epoll_create_common(struct thread *td, int flags)
{
int error;
error = kern_kqueue(td, flags, NULL);
if (error != 0)
return (error);
epoll_fd_install(td, EPOLL_DEF_SZ, 0);
return (0);
return (kern_kqueue(td, flags, NULL));
}
#ifdef LINUX_LEGACY_SYSCALLS
@ -271,11 +217,15 @@ epoll_to_kevent(struct thread *td, int fd, struct epoll_event *l_event,
/* flags related to what event is registered */
if ((levents & LINUX_EPOLL_EVRD) != 0) {
EV_SET(kevent++, fd, EVFILT_READ, kev_flags, 0, 0, 0);
EV_SET(kevent, fd, EVFILT_READ, kev_flags, 0, 0, 0);
kevent->ext[0] = l_event->data;
++kevent;
++(*nkevents);
}
if ((levents & LINUX_EPOLL_EVWR) != 0) {
EV_SET(kevent++, fd, EVFILT_WRITE, kev_flags, 0, 0, 0);
EV_SET(kevent, fd, EVFILT_WRITE, kev_flags, 0, 0, 0);
kevent->ext[0] = l_event->data;
++kevent;
++(*nkevents);
}
/* zero event mask is legal */
@ -289,7 +239,6 @@ epoll_to_kevent(struct thread *td, int fd, struct epoll_event *l_event,
pem = pem_find(p);
KASSERT(pem != NULL, ("epoll proc emuldata not found.\n"));
KASSERT(pem->epoll != NULL, ("epoll proc epolldata not found.\n"));
LINUX_PEM_XLOCK(pem);
if ((pem->flags & LINUX_XUNSUP_EPOLL) == 0) {
@ -314,6 +263,8 @@ static void
kevent_to_epoll(struct kevent *kevent, struct epoll_event *l_event)
{
l_event->data = kevent->ext[0];
if ((kevent->flags & EV_ERROR) != 0) {
l_event->events = LINUX_EPOLLERR;
return;
@ -342,30 +293,15 @@ static int
epoll_kev_copyout(void *arg, struct kevent *kevp, int count)
{
struct epoll_copyout_args *args;
struct linux_pemuldata *pem;
struct epoll_emuldata *emd;
struct epoll_event *eep;
int error, fd, i;
int error, i;
args = (struct epoll_copyout_args*) arg;
eep = malloc(sizeof(*eep) * count, M_EPOLL, M_WAITOK | M_ZERO);
pem = pem_find(args->p);
KASSERT(pem != NULL, ("epoll proc emuldata not found.\n"));
LINUX_PEM_SLOCK(pem);
emd = pem->epoll;
KASSERT(emd != NULL, ("epoll proc epolldata not found.\n"));
for (i = 0; i < count; i++) {
for (i = 0; i < count; i++)
kevent_to_epoll(&kevp[i], &eep[i]);
fd = kevp[i].ident;
KASSERT(fd <= emd->fdc, ("epoll user data vector"
" is too small.\n"));
eep[i].data = emd->udata[fd];
}
LINUX_PEM_SUNLOCK(pem);
error = copyout(eep, args->leventlist, count * sizeof(*eep));
if (error == 0) {
args->leventlist += count;
@ -473,8 +409,6 @@ linux_epoll_ctl(struct thread *td, struct linux_epoll_ctl_args *args)
goto leave0;
}
epoll_fd_install(td, args->fd, le.data);
error = kern_kevent_fp(td, epfp, nchanges, 0, &k_ops, NULL);
leave0: