Hi, Greg KH,

Currently, this patch is under review and no correct solution is available to solve this problem. After a perfect patch is available, I will send the corresponding patch to 5.15y.

On 2022/10/27 0:52, Greg KH wrote:

On Mon, Oct 24, 2022 at 09:49:11AM +0000, Yuanzheng Song wrote:

...

The vma->anon_vma of the child process may be NULL because the entire vma does not contain anonymous pages. In this case, a BUG will occur when the copy_present_page() passes a copy of a non-anonymous page of that vma to the page_add_new_anon_rmap() to set up new anonymous rmap.

------------[ cut here ]------------ kernel BUG at mm/rmap.c:1044! Internal error: Oops - BUG: 0 [#1] SMP Modules linked in: CPU: 2 PID: 3617 Comm: test Not tainted 5.10.149 #1 Hardware name: linux,dummy-virt (DT) pstate: 80000005 (Nzcv daif -PAN -UAO -TCO BTYPE=--) pc : __page_set_anon_rmap+0xbc/0xf8 lr : __page_set_anon_rmap+0xbc/0xf8 sp : ffff800014c1b870 x29: ffff800014c1b870 x28: 0000000000000001 x27: 0000000010100073 x26: ffff1d65c517baa8 x25: ffff1d65cab0f000 x24: ffff1d65c416d800 x23: ffff1d65cab5f248 x22: 0000000020000000 x21: 0000000000000001 x20: 0000000000000000 x19: fffffe75970023c0 x18: 0000000000000000 x17: 0000000000000000 x16: 0000000000000000 x15: 0000000000000000 x14: 0000000000000000 x13: 0000000000000000 x12: 0000000000000000 x11: 0000000000000000 x10: 0000000000000000 x9 : ffffc3096d5fb858 x8 : 0000000000000000 x7 : 0000000000000011 x6 : ffff5a5c9089c000 x5 : 0000000000020000 x4 : ffff5a5c9089c000 x3 : ffffc3096d200000 x2 : ffffc3096e8d0000 x1 : ffff1d65ca3da740 x0 : 0000000000000000 Call trace: __page_set_anon_rmap+0xbc/0xf8 page_add_new_anon_rmap+0x1e0/0x390 copy_pte_range+0xd00/0x1248 copy_page_range+0x39c/0x620 dup_mmap+0x2e0/0x5a8 dup_mm+0x78/0x140 copy_process+0x918/0x1a20 kernel_clone+0xac/0x638 __do_sys_clone+0x78/0xb0 __arm64_sys_clone+0x30/0x40 el0_svc_common.constprop.0+0xb0/0x308 do_el0_svc+0x48/0xb8 el0_svc+0x24/0x38 el0_sync_handler+0x160/0x168 el0_sync+0x180/0x1c0 Code: 97f8ff85 f9400294 17ffffeb 97f8ff82 (d4210000) ---[ end trace a972347688dc9bd4 ]--- Kernel panic - not syncing: Oops - BUG: Fatal exception SMP: stopping secondary CPUs Kernel Offset: 0x43095d200000 from 0xffff800010000000 PHYS_OFFSET: 0xffffe29a80000000 CPU features: 0x08200022,61806082 Memory Limit: none ---[ end Kernel panic - not syncing: Oops - BUG: Fatal exception ]---

This problem has been fixed by the fb3d824d1a46 ("mm/rmap: split page_dup_rmap() into page_dup_file_rmap() and page_try_dup_anon_rmap()"), but still exists in the linux-5.10.y branch.

This patch is not applicable to this version because of the large version differences. Therefore, fix it by adding non-anonymous page check in the copy_present_page().

Fixes: 70e806e4e645 ("mm: Do early cow for pinned pages during fork() for ptes") Signed-off-by: Yuanzheng Song songyuanzheng@huawei.com

---

mm/memory.c | 11 +++++++++++ 1 file changed, 11 insertions(+)

We also need this in 5.15.y, right? Can you provide a working version for there so that no one upgrades from 5.10.y to 5.15.y and has a regression?

I'll wait for that before taking this one.

thanks,

greg k-h .

Thanks,

Yuanzheng .

Hi, Yuanzheng,

On Wed, Oct 26, 2022 at 02:51:16PM -0700, Hugh Dickins wrote:

On Mon, 24 Oct 2022, Yuanzheng Song wrote:

...

The vma->anon_vma of the child process may be NULL because the entire vma does not contain anonymous pages. In this case, a BUG will occur when the copy_present_page() passes a copy of a non-anonymous page of that vma to the page_add_new_anon_rmap() to set up new anonymous rmap.

------------[ cut here ]------------ kernel BUG at mm/rmap.c:1044! Internal error: Oops - BUG: 0 [#1] SMP Modules linked in: CPU: 2 PID: 3617 Comm: test Not tainted 5.10.149 #1 Hardware name: linux,dummy-virt (DT) pstate: 80000005 (Nzcv daif -PAN -UAO -TCO BTYPE=--) pc : __page_set_anon_rmap+0xbc/0xf8 lr : __page_set_anon_rmap+0xbc/0xf8 sp : ffff800014c1b870 x29: ffff800014c1b870 x28: 0000000000000001 x27: 0000000010100073 x26: ffff1d65c517baa8 x25: ffff1d65cab0f000 x24: ffff1d65c416d800 x23: ffff1d65cab5f248 x22: 0000000020000000 x21: 0000000000000001 x20: 0000000000000000 x19: fffffe75970023c0 x18: 0000000000000000 x17: 0000000000000000 x16: 0000000000000000 x15: 0000000000000000 x14: 0000000000000000 x13: 0000000000000000 x12: 0000000000000000 x11: 0000000000000000 x10: 0000000000000000 x9 : ffffc3096d5fb858 x8 : 0000000000000000 x7 : 0000000000000011 x6 : ffff5a5c9089c000 x5 : 0000000000020000 x4 : ffff5a5c9089c000 x3 : ffffc3096d200000 x2 : ffffc3096e8d0000 x1 : ffff1d65ca3da740 x0 : 0000000000000000 Call trace: __page_set_anon_rmap+0xbc/0xf8 page_add_new_anon_rmap+0x1e0/0x390 copy_pte_range+0xd00/0x1248 copy_page_range+0x39c/0x620 dup_mmap+0x2e0/0x5a8 dup_mm+0x78/0x140 copy_process+0x918/0x1a20 kernel_clone+0xac/0x638 __do_sys_clone+0x78/0xb0 __arm64_sys_clone+0x30/0x40 el0_svc_common.constprop.0+0xb0/0x308 do_el0_svc+0x48/0xb8 el0_svc+0x24/0x38 el0_sync_handler+0x160/0x168 el0_sync+0x180/0x1c0 Code: 97f8ff85 f9400294 17ffffeb 97f8ff82 (d4210000) ---[ end trace a972347688dc9bd4 ]--- Kernel panic - not syncing: Oops - BUG: Fatal exception SMP: stopping secondary CPUs Kernel Offset: 0x43095d200000 from 0xffff800010000000 PHYS_OFFSET: 0xffffe29a80000000 CPU features: 0x08200022,61806082 Memory Limit: none ---[ end Kernel panic - not syncing: Oops - BUG: Fatal exception ]---

This problem has been fixed by the fb3d824d1a46 ("mm/rmap: split page_dup_rmap() into page_dup_file_rmap() and page_try_dup_anon_rmap()"), but still exists in the linux-5.10.y branch.

This patch is not applicable to this version because of the large version differences. Therefore, fix it by adding non-anonymous page check in the copy_present_page().

Fixes: 70e806e4e645 ("mm: Do early cow for pinned pages during fork() for ptes") Signed-off-by: Yuanzheng Song songyuanzheng@huawei.com

It's a good point, but this patch should not go into any stable release without an explicit Ack from either Peter Xu or David Hildenbrand.

To my eye, it's simply avoiding the issue, rather than fixing it properly; and even if the issue is so rare, and fixing properly too difficult or inefficent (a cached anon_vma?), that a workaround is good enough, it still looks like the wrong workaround (checking dst_vma->anon_vma instead of PageAnon seems more to the point, and less lenient).

Sorry to have missed this patch. I agree with Hugh that this patch may not really fix the issue.

IIUC it's the case where the vma is privately mapping a file. Some shared pages got pinned, and here we're trying to trigger the CoW assuming it's anonymous page but it's not.

The pin should be RO - if it was a write pin, CoW should have happened on the page cache and there should be an anonymous page, and anon_vma should be there, no issue should happen. Only if with RO pin, we won't trigger CoW, we won't have any anonymous page, we won't have anon_vma, hence the panic.

The thing is if the page is RO pinned, skip copying it (as what was done in this patch) is not correct either, because e.g. a follow up write after fork() from the parent will trigger CoW and the dma RO page that was pinned will be inconsistent to the page in pgtable anymore, I think.

IIUC the correct fix is what David worked on with unshare - when RO pin the page cache we should have triggered CoR already before fork(). But as you mentioned, that's too much change for stable.

So besides this workaround which seems feasible to at least not panic the system (Hugh: I can't quickly tell what'll be the difference here to check dst anon_vma or PageAnon, they all seem to work? I could have missed something for sure..), the other workaround is teach the user app to switch to use RW pin for any DMA pages even if RO, so that it'll always guarantee page coherence even with the old kernel, and afaict that's what RDMA relied on for years. I don't know an easy way to make old kernel work with RO pin solidly if without the unsharing logic.

Thanks,

Hi, Peter Xu:

Thank you for reviewing this patch. This bug was reported by syzkaller.

You can reproduce it using the following test case.

As you said, my patch isn't really fixing the problem, just avoiding the bug.

In my opinion, simply abandoning the creation of new anonymous rmap

by judging that it is not an anonymous page is the minimal change to

circumvent the bug.

This bug can be reproduced from

70e806e4e645 ("mm: Do early cow for pinned pages during fork() for ptes")

to

b51ad4f8679e ("mm/memory: slightly simplify copy_present_pte()")

in the mainline.

(Greg KH: Therefore, this problem involves 5.10y and even 5.15y.)

I hope you can help to look at this problem and find a better solution.

------------------------------------------------------------------------------

#define _GNU_SOURCE

#include <dirent.h> #include <endian.h> #include <errno.h> #include <fcntl.h> #include <pthread.h> #include <signal.h> #include <stdarg.h> #include <stdbool.h> #include <stdint.h> #include <stdio.h> #include <stdlib.h> #include <string.h> #include <sys/prctl.h> #include <sys/stat.h> #include <sys/syscall.h> #include <sys/types.h> #include <sys/wait.h> #include <time.h> #include <unistd.h>

#include <linux/futex.h>

#ifndef __NR_clone #define __NR_clone 220 #endif #ifndef __NR_exit #define __NR_exit 93 #endif #ifndef __NR_gettid #define __NR_gettid 178 #endif #ifndef __NR_mmap #define __NR_mmap 222 #endif #ifndef __NR_process_vm_readv #define __NR_process_vm_readv 270 #endif #ifndef __NR_setsockopt #define __NR_setsockopt 208 #endif #ifndef __NR_socket #define __NR_socket 198 #endif

static unsigned long long procid;

static void sleep_ms(uint64_t ms) {     usleep(ms * 1000); }

static uint64_t current_time_ms(void) {     struct timespec ts;     if (clock_gettime(CLOCK_MONOTONIC, &ts))     exit(1);     return (uint64_t)ts.tv_sec * 1000 + (uint64_t)ts.tv_nsec / 1000000; }

static void thread_start(void* (*fn)(void*), void* arg) {     pthread_t th;     pthread_attr_t attr;     pthread_attr_init(&attr);     pthread_attr_setstacksize(&attr, 128 << 10);     int i = 0;     for (; i < 100; i++) {         if (pthread_create(&th, &attr, fn, arg) == 0) {             pthread_attr_destroy(&attr);             return;         }         if (errno == EAGAIN) {             usleep(50);             continue;         }         break;     }     exit(1); }

typedef struct {     int state; } event_t;

static void event_init(event_t* ev) {     ev->state = 0; }

static void event_reset(event_t* ev) {     ev->state = 0; }

static void event_set(event_t* ev) {     if (ev->state)     exit(1);     __atomic_store_n(&ev->state, 1, __ATOMIC_RELEASE);     syscall(SYS_futex, &ev->state, FUTEX_WAKE | FUTEX_PRIVATE_FLAG, 1000000); }

static void event_wait(event_t* ev) {     while (!__atomic_load_n(&ev->state, __ATOMIC_ACQUIRE))         syscall(SYS_futex, &ev->state, FUTEX_WAIT | FUTEX_PRIVATE_FLAG, 0, 0); }

static int event_isset(event_t* ev) {     return __atomic_load_n(&ev->state, __ATOMIC_ACQUIRE); }

static int event_timedwait(event_t* ev, uint64_t timeout) {     uint64_t start = current_time_ms();     uint64_t now = start;     for (;;) {         uint64_t remain = timeout - (now - start);         struct timespec ts;         ts.tv_sec = remain / 1000;         ts.tv_nsec = (remain % 1000) * 1000 * 1000;         syscall(SYS_futex, &ev->state, FUTEX_WAIT | FUTEX_PRIVATE_FLAG, 0, &ts);         if (__atomic_load_n(&ev->state, __ATOMIC_ACQUIRE))             return 1;         now = current_time_ms();         if (now - start > timeout)             return 0;     } }

static bool write_file(const char* file, const char* what, ...) {     char buf[1024];     va_list args;     va_start(args, what);     vsnprintf(buf, sizeof(buf), what, args);     va_end(args);     buf[sizeof(buf) - 1] = 0;     int len = strlen(buf);     int fd = open(file, O_WRONLY | O_CLOEXEC);     if (fd == -1)         return false;     if (write(fd, buf, len) != len) {         int err = errno;         close(fd);         errno = err;         return false;     }     close(fd);     return true; }

static void kill_and_wait(int pid, int* status) {     kill(-pid, SIGKILL);     kill(pid, SIGKILL);     for (int i = 0; i < 100; i++) {         if (waitpid(-1, status, WNOHANG | __WALL) == pid)             return;         usleep(1000);     }     DIR* dir = opendir("/sys/fs/fuse/connections");     if (dir) {         for (;;) {             struct dirent* ent = readdir(dir);             if (!ent)                 break;             if (strcmp(ent->d_name, ".") == 0 || strcmp(ent->d_name, "..") == 0)                 continue;             char abort[300];             snprintf(abort, sizeof(abort), "/sys/fs/fuse/connections/%s/abort", ent->d_name);             int fd = open(abort, O_WRONLY);             if (fd == -1) {                 continue;             }             if (write(fd, abort, 1) < 0) {             }             close(fd);         }         closedir(dir);     } else {     }     while (waitpid(-1, status, __WALL) != pid) {     } }

static void setup_test() {     prctl(PR_SET_PDEATHSIG, SIGKILL, 0, 0, 0);     setpgrp();     write_file("/proc/self/oom_score_adj", "1000"); }

#define USLEEP_FORKED_CHILD (3 * 50 *1000)

static long handle_clone_ret(long ret) {     if (ret != 0) {         return ret;     }     usleep(USLEEP_FORKED_CHILD);     syscall(__NR_exit, 0);     while (1) {     } }

static long syz_clone(volatile long flags, volatile long stack, volatile long stack_len,               volatile long ptid, volatile long ctid, volatile long tls) {     long sp = (stack + stack_len) & ~15;     long ret = (long)syscall(__NR_clone, flags & ~CLONE_VM, sp, ptid, ctid, tls);     return handle_clone_ret(ret); }

struct thread_t {     int created, call;     event_t ready, done; };

static struct thread_t threads[16]; static void execute_call(int call); static int running;

static void* thr(void* arg) {     struct thread_t* th = (struct thread_t*)arg;     for (;;) {         event_wait(&th->ready);         event_reset(&th->ready);         execute_call(th->call);         __atomic_fetch_sub(&running, 1, __ATOMIC_RELAXED);         event_set(&th->done);     }     return 0; }

static void execute_one(void) {     int i, call, thread;     for (call = 0; call < 6; call++) {         for (thread = 0; thread < (int)(sizeof(threads) / sizeof(threads[0])); thread++) {             struct thread_t* th = &threads[thread];             if (!th->created) {                 th->created = 1;                 event_init(&th->ready);                 event_init(&th->done);                 event_set(&th->done);                 thread_start(thr, th);             }             if (!event_isset(&th->done))                 continue;             event_reset(&th->done);             th->call = call;             __atomic_fetch_add(&running, 1, __ATOMIC_RELAXED);             event_set(&th->ready);             if (call == 3 || call == 4)                 break;             event_timedwait(&th->done, 50);             break;         }     }     for (i = 0; i < 100 && __atomic_load_n(&running, __ATOMIC_RELAXED); i++)         sleep_ms(1); }

static void execute_one(void);

#define WAIT_FLAGS __WALL

static void loop(void) {     int iter = 0;     for (;; iter++) {         int pid = fork();         if (pid < 0)     exit(1);         if (pid == 0) {             setup_test();             execute_one();             exit(0);         }         int status = 0;         uint64_t start = current_time_ms();         for (;;) {             if (waitpid(-1, &status, WNOHANG | WAIT_FLAGS) == pid)                 break;             sleep_ms(1);             if (current_time_ms() - start < 5000)                 continue;             kill_and_wait(pid, &status);             break;         }     } }

uint64_t r[2] = {0xffffffffffffffff, 0x0};

void execute_call(int call) {         intptr_t res = 0;     switch (call) {     case 0:         res = syscall(__NR_socket, 0x11ul, 3ul, 0x300);         if (res != -1)                 r[0] = res;         break;     case 1: *(uint32_t*)0x20000100 = 0x10000; *(uint32_t*)0x20000104 = 7; *(uint32_t*)0x20000108 = 0x1f0; *(uint32_t*)0x2000010c = 0x39c; *(uint32_t*)0x20000110 = 0; *(uint32_t*)0x20000114 = 0; *(uint32_t*)0x20000118 = 0;         syscall(__NR_setsockopt, r[0], 0x107, 0xd, 0x20000100ul, 0x1cul);         break;     case 2:         res = syscall(__NR_gettid);         if (res != -1)                 r[1] = res;         break;     case 3:         syscall(__NR_mmap, 0x20000000ul, 0x70000ul, 0x13ul, 0x12ul, r[0], 0ul);         break;     case 4: syz_clone(0, 0, 0, 0, 0, 0);         break;     case 5: *(uint64_t*)0x20000440 = 0x20000000; *(uint64_t*)0x20000448 = 0xf1; *(uint64_t*)0x20000450 = 0; *(uint64_t*)0x20000458 = 0; *(uint64_t*)0x20000460 = 0; *(uint64_t*)0x20000468 = 0; *(uint64_t*)0x20000470 = 0; *(uint64_t*)0x20000478 = 0; *(uint64_t*)0x20000480 = 0; *(uint64_t*)0x20000488 = 0; *(uint64_t*)0x20000490 = 0; *(uint64_t*)0x20000498 = 0; *(uint64_t*)0x200004a0 = 0; *(uint64_t*)0x200004a8 = 0; *(uint64_t*)0x200004b0 = 0; *(uint64_t*)0x200004b8 = 0; *(uint64_t*)0x200004c0 = 0; *(uint64_t*)0x200004c8 = 0; *(uint64_t*)0x20001700 = 0x20000500; *(uint64_t*)0x20001708 = 0x15; *(uint64_t*)0x20001710 = 0; *(uint64_t*)0x20001718 = 0; *(uint64_t*)0x20001720 = 0; *(uint64_t*)0x20001728 = 0; *(uint64_t*)0x20001730 = 0; *(uint64_t*)0x20001738 = 0; *(uint64_t*)0x20001740 = 0; *(uint64_t*)0x20001748 = 0;         syscall(__NR_process_vm_readv, r[1], 0x20000440ul, 9ul, 0x20001700ul, 5ul, 0ul);         break;     }

} int main(void) {         syscall(__NR_mmap, 0x1ffff000ul, 0x1000ul, 0ul, 0x32ul, -1, 0ul);     syscall(__NR_mmap, 0x20000000ul, 0x1000000ul, 7ul, 0x32ul, -1, 0ul);     syscall(__NR_mmap, 0x21000000ul, 0x1000ul, 0ul, 0x32ul, -1, 0ul);     for (procid = 0; procid < 4; procid++) {         if (fork() == 0) {             loop();         }     }     sleep(1000000);     return 0; }

------------------------------------------------------------------------------

On 2022/10/27 9:48, Hugh Dickins wrote:

On Wed, 26 Oct 2022, Peter Xu wrote:

...

Hi, Yuanzheng,

On Wed, Oct 26, 2022 at 02:51:16PM -0700, Hugh Dickins wrote:

...

On Mon, 24 Oct 2022, Yuanzheng Song wrote:

...

The vma->anon_vma of the child process may be NULL because the entire vma does not contain anonymous pages. In this case, a BUG will occur when the copy_present_page() passes a copy of a non-anonymous page of that vma to the page_add_new_anon_rmap() to set up new anonymous rmap.

------------[ cut here ]------------ kernel BUG at mm/rmap.c:1044! Internal error: Oops - BUG: 0 [#1] SMP Modules linked in: CPU: 2 PID: 3617 Comm: test Not tainted 5.10.149 #1 Hardware name: linux,dummy-virt (DT) pstate: 80000005 (Nzcv daif -PAN -UAO -TCO BTYPE=--) pc : __page_set_anon_rmap+0xbc/0xf8 lr : __page_set_anon_rmap+0xbc/0xf8 sp : ffff800014c1b870 x29: ffff800014c1b870 x28: 0000000000000001 x27: 0000000010100073 x26: ffff1d65c517baa8 x25: ffff1d65cab0f000 x24: ffff1d65c416d800 x23: ffff1d65cab5f248 x22: 0000000020000000 x21: 0000000000000001 x20: 0000000000000000 x19: fffffe75970023c0 x18: 0000000000000000 x17: 0000000000000000 x16: 0000000000000000 x15: 0000000000000000 x14: 0000000000000000 x13: 0000000000000000 x12: 0000000000000000 x11: 0000000000000000 x10: 0000000000000000 x9 : ffffc3096d5fb858 x8 : 0000000000000000 x7 : 0000000000000011 x6 : ffff5a5c9089c000 x5 : 0000000000020000 x4 : ffff5a5c9089c000 x3 : ffffc3096d200000 x2 : ffffc3096e8d0000 x1 : ffff1d65ca3da740 x0 : 0000000000000000 Call trace: __page_set_anon_rmap+0xbc/0xf8 page_add_new_anon_rmap+0x1e0/0x390 copy_pte_range+0xd00/0x1248 copy_page_range+0x39c/0x620 dup_mmap+0x2e0/0x5a8 dup_mm+0x78/0x140 copy_process+0x918/0x1a20 kernel_clone+0xac/0x638 __do_sys_clone+0x78/0xb0 __arm64_sys_clone+0x30/0x40 el0_svc_common.constprop.0+0xb0/0x308 do_el0_svc+0x48/0xb8 el0_svc+0x24/0x38 el0_sync_handler+0x160/0x168 el0_sync+0x180/0x1c0 Code: 97f8ff85 f9400294 17ffffeb 97f8ff82 (d4210000) ---[ end trace a972347688dc9bd4 ]--- Kernel panic - not syncing: Oops - BUG: Fatal exception SMP: stopping secondary CPUs Kernel Offset: 0x43095d200000 from 0xffff800010000000 PHYS_OFFSET: 0xffffe29a80000000 CPU features: 0x08200022,61806082 Memory Limit: none ---[ end Kernel panic - not syncing: Oops - BUG: Fatal exception ]---

This problem has been fixed by the fb3d824d1a46 ("mm/rmap: split page_dup_rmap() into page_dup_file_rmap() and page_try_dup_anon_rmap()"), but still exists in the linux-5.10.y branch.

This patch is not applicable to this version because of the large version differences. Therefore, fix it by adding non-anonymous page check in the copy_present_page().

Fixes: 70e806e4e645 ("mm: Do early cow for pinned pages during fork() for ptes") Signed-off-by: Yuanzheng Song songyuanzheng@huawei.com

It's a good point, but this patch should not go into any stable release without an explicit Ack from either Peter Xu or David Hildenbrand.

To my eye, it's simply avoiding the issue, rather than fixing it properly; and even if the issue is so rare, and fixing properly too difficult or inefficent (a cached anon_vma?), that a workaround is good enough, it still looks like the wrong workaround (checking dst_vma->anon_vma instead of PageAnon seems more to the point, and less lenient).

Sorry to have missed this patch. I agree with Hugh that this patch may not really fix the issue.

IIUC it's the case where the vma is privately mapping a file. Some shared pages got pinned, and here we're trying to trigger the CoW assuming it's anonymous page but it's not.

The pin should be RO - if it was a write pin, CoW should have happened on the page cache and there should be an anonymous page, and anon_vma should be there, no issue should happen. Only if with RO pin, we won't trigger CoW, we won't have any anonymous page, we won't have anon_vma, hence the panic.

The thing is if the page is RO pinned, skip copying it (as what was done in this patch) is not correct either, because e.g. a follow up write after fork() from the parent will trigger CoW and the dma RO page that was pinned will be inconsistent to the page in pgtable anymore, I think.

IIUC the correct fix is what David worked on with unshare - when RO pin the page cache we should have triggered CoR already before fork(). But as you mentioned, that's too much change for stable.

So besides this workaround which seems feasible to at least not panic the system (Hugh: I can't quickly tell what'll be the difference here to check dst anon_vma or PageAnon, they all seem to work? I could have missed

Thanks for all the helpful elucidation above.

My thought here, in favour of checking dst anon_vma rather than PageAnon, was that a common case would be that the private file vma does already have an anon_vma attached (from earlier CoW on some other page), and in that case there is no justification for taking this short cut to avoid the BUG in __page_set_anon_rmap() on every !PageAnon in the vma.

And I imagined that the correct fix (short of going forward with David's full changes) would be to back out to a context where one could add an anon_vma_prepare(), then retry after that - involves dropping pt lock, maybe gets nasty (tedious, anyway).

...

something for sure..), the other workaround is teach the user app to switch to use RW pin for any DMA pages even if RO, so that it'll always guarantee page coherence even with the old kernel, and afaict that's what RDMA relied on for years. I don't know an easy way to make old kernel work with RO pin solidly if without the unsharing logic.

Thanks,

-- Peter Xu

.

Thanks,

On 26.10.22 23:51, Hugh Dickins wrote:

On Mon, 24 Oct 2022, Yuanzheng Song wrote:

...

The vma->anon_vma of the child process may be NULL because the entire vma does not contain anonymous pages. In this case, a BUG will occur when the copy_present_page() passes a copy of a non-anonymous page of that vma to the page_add_new_anon_rmap() to set up new anonymous rmap.

------------[ cut here ]------------ kernel BUG at mm/rmap.c:1044! Internal error: Oops - BUG: 0 [#1] SMP Modules linked in: CPU: 2 PID: 3617 Comm: test Not tainted 5.10.149 #1 Hardware name: linux,dummy-virt (DT) pstate: 80000005 (Nzcv daif -PAN -UAO -TCO BTYPE=--) pc : __page_set_anon_rmap+0xbc/0xf8 lr : __page_set_anon_rmap+0xbc/0xf8 sp : ffff800014c1b870 x29: ffff800014c1b870 x28: 0000000000000001 x27: 0000000010100073 x26: ffff1d65c517baa8 x25: ffff1d65cab0f000 x24: ffff1d65c416d800 x23: ffff1d65cab5f248 x22: 0000000020000000 x21: 0000000000000001 x20: 0000000000000000 x19: fffffe75970023c0 x18: 0000000000000000 x17: 0000000000000000 x16: 0000000000000000 x15: 0000000000000000 x14: 0000000000000000 x13: 0000000000000000 x12: 0000000000000000 x11: 0000000000000000 x10: 0000000000000000 x9 : ffffc3096d5fb858 x8 : 0000000000000000 x7 : 0000000000000011 x6 : ffff5a5c9089c000 x5 : 0000000000020000 x4 : ffff5a5c9089c000 x3 : ffffc3096d200000 x2 : ffffc3096e8d0000 x1 : ffff1d65ca3da740 x0 : 0000000000000000 Call trace: __page_set_anon_rmap+0xbc/0xf8 page_add_new_anon_rmap+0x1e0/0x390 copy_pte_range+0xd00/0x1248 copy_page_range+0x39c/0x620 dup_mmap+0x2e0/0x5a8 dup_mm+0x78/0x140 copy_process+0x918/0x1a20 kernel_clone+0xac/0x638 __do_sys_clone+0x78/0xb0 __arm64_sys_clone+0x30/0x40 el0_svc_common.constprop.0+0xb0/0x308 do_el0_svc+0x48/0xb8 el0_svc+0x24/0x38 el0_sync_handler+0x160/0x168 el0_sync+0x180/0x1c0 Code: 97f8ff85 f9400294 17ffffeb 97f8ff82 (d4210000) ---[ end trace a972347688dc9bd4 ]--- Kernel panic - not syncing: Oops - BUG: Fatal exception SMP: stopping secondary CPUs Kernel Offset: 0x43095d200000 from 0xffff800010000000 PHYS_OFFSET: 0xffffe29a80000000 CPU features: 0x08200022,61806082 Memory Limit: none ---[ end Kernel panic - not syncing: Oops - BUG: Fatal exception ]---

This problem has been fixed by the fb3d824d1a46 ("mm/rmap: split page_dup_rmap() into page_dup_file_rmap() and page_try_dup_anon_rmap()"), but still exists in the linux-5.10.y branch.

This patch is not applicable to this version because of the large version differences. Therefore, fix it by adding non-anonymous page check in the copy_present_page().

Fixes: 70e806e4e645 ("mm: Do early cow for pinned pages during fork() for ptes") Signed-off-by: Yuanzheng Song songyuanzheng@huawei.com

It's a good point, but this patch should not go into any stable release without an explicit Ack from either Peter Xu or David Hildenbrand.

To my eye, it's simply avoiding the issue, rather than fixing it properly; and even if the issue is so rare, and fixing properly too difficult or inefficent (a cached anon_vma?), that a workaround is good enough, it still looks like the wrong workaround (checking dst_vma->anon_vma instead of PageAnon seems more to the point, and less lenient).

But my eye on COW is very poor nowadays, and I may be plain wrong.

I am not aware of any reason for copying a !anon page during fork(). COW regrading fork() is all about sharing private (anon) pages between the parent and the child. The semantics of other pages are untouched.

Yes, I am working on reliable longterm R/O pinning improvements, whereby we never pin such pages in a MAP_PRIVATE mapping but instead break COW before pinning; but this only applies to longterm pinning (FOLL_LONGTERM) and is independent of fork() here.

Let me elaborate: if you have a pagecache page (or the shared zeropage) in a MAP_PRIVATE mapping pinned R/O, the next write fault will replace the page by a copy, *independent* of fork() or not: the page is already mapped write-protected into the page table.

IIUC, the problem here is that we have a writable private mapping (COW mapping) of, say, a file, whereby we never had to COW -- so no anon pages were mapped.

Then, we had the process pin some page (&src_mm->has_pinned) once and detect a pagecache page / shared zeropage as "maybe pinned" during fork(), which can happen easily, for example, due to other process' action, false positives, ... we end up duplicating a !anon page.

Restricting copying during fork() to anon pages is IMHO the right thing to do.