Linux kernel mirror (for testing)
git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git
kernel
os
linux
Merge tag 'io_uring-6.5-2023-07-21' of git://git.kernel.dk/linux
Pull io_uring fixes from Jens Axboe:
- Fix for io-wq not always honoring REQ_F_NOWAIT, if it was set and
punted directly (eg via DRAIN) (me)
- Capability check fix (Ondrej)
- Regression fix for the mmap changes that went into 6.4, which
apparently broke IA64 (Helge)
* tag 'io_uring-6.5-2023-07-21' of git://git.kernel.dk/linux:
ia64: mmap: Consider pgoff when searching for free mapping
io_uring: Fix io_uring mmap() by using architecture-provided get_unmapped_area()
io_uring: treat -EAGAIN for REQ_F_NOWAIT as final for io-wq
io_uring: don't audit the capability check in io_uring_create()
+38
-33
+1
-1
arch/ia64/kernel/sys_ia64.c
+1
-1
arch/ia64/kernel/sys_ia64.c
+10
-5
arch/parisc/kernel/sys_parisc.c
+10
-5
arch/parisc/kernel/sys_parisc.c
···
27
27
#include <linux/elf-randomize.h>
28
28
29
29
/*
30
-
* Construct an artificial page offset for the mapping based on the physical
30
+
* Construct an artificial page offset for the mapping based on the virtual
31
31
* address of the kernel file mapping variable.
32
+
* If filp is zero the calculated pgoff value aliases the memory of the given
33
+
* address. This is useful for io_uring where the mapping shall alias a kernel
34
+
* address and a userspace adress where both the kernel and the userspace
35
+
* access the same memory region.
32
36
*/
33
-
#define GET_FILP_PGOFF(filp) \
34
-
(filp ? (((unsigned long) filp->f_mapping) >> 8) \
35
-
& ((SHM_COLOUR-1) >> PAGE_SHIFT) : 0UL)
37
+
#define GET_FILP_PGOFF(filp, addr) \
38
+
((filp ? (((unsigned long) filp->f_mapping) >> 8) \
39
+
& ((SHM_COLOUR-1) >> PAGE_SHIFT) : 0UL) \
40
+
+ (addr >> PAGE_SHIFT))
36
41
37
42
static unsigned long shared_align_offset(unsigned long filp_pgoff,
38
43
unsigned long pgoff)
···
117
112
do_color_align = 0;
118
113
if (filp || (flags & MAP_SHARED))
119
114
do_color_align = 1;
120
-
filp_pgoff = GET_FILP_PGOFF(filp);
115
+
filp_pgoff = GET_FILP_PGOFF(filp, addr);
121
116
122
117
if (flags & MAP_FIXED) {
123
118
/* Even MAP_FIXED mappings must reside within TASK_SIZE */
+27
-27
io_uring/io_uring.c
+27
-27
io_uring/io_uring.c
···
1948
1948
ret = io_issue_sqe(req, issue_flags);
1949
1949
if (ret != -EAGAIN)
1950
1950
break;
1951
+
1952
+
/*
1953
+
* If REQ_F_NOWAIT is set, then don't wait or retry with
1954
+
* poll. -EAGAIN is final for that case.
1955
+
*/
1956
+
if (req->flags & REQ_F_NOWAIT)
1957
+
break;
1958
+
1951
1959
/*
1952
1960
* We can get EAGAIN for iopolled IO even though we're
1953
1961
* forcing a sync submission from here, since we can't
···
3437
3429
unsigned long addr, unsigned long len,
3438
3430
unsigned long pgoff, unsigned long flags)
3439
3431
{
3440
-
const unsigned long mmap_end = arch_get_mmap_end(addr, len, flags);
3441
-
struct vm_unmapped_area_info info;
3442
3432
void *ptr;
3443
3433
3444
3434
/*
···
3451
3445
if (IS_ERR(ptr))
3452
3446
return -ENOMEM;
3453
3447
3454
-
info.flags = VM_UNMAPPED_AREA_TOPDOWN;
3455
-
info.length = len;
3456
-
info.low_limit = max(PAGE_SIZE, mmap_min_addr);
3457
-
info.high_limit = arch_get_mmap_base(addr, current->mm->mmap_base);
3458
-
#ifdef SHM_COLOUR
3459
-
info.align_mask = PAGE_MASK & (SHM_COLOUR - 1UL);
3460
-
#else
3461
-
info.align_mask = PAGE_MASK & (SHMLBA - 1UL);
3462
-
#endif
3463
-
info.align_offset = (unsigned long) ptr;
3464
-
3465
3448
/*
3466
-
* A failed mmap() very likely causes application failure,
3467
-
* so fall back to the bottom-up function here. This scenario
3468
-
* can happen with large stack limits and large mmap()
3469
-
* allocations.
3449
+
* Some architectures have strong cache aliasing requirements.
3450
+
* For such architectures we need a coherent mapping which aliases
3451
+
* kernel memory *and* userspace memory. To achieve that:
3452
+
* - use a NULL file pointer to reference physical memory, and
3453
+
* - use the kernel virtual address of the shared io_uring context
3454
+
* (instead of the userspace-provided address, which has to be 0UL
3455
+
* anyway).
3456
+
* For architectures without such aliasing requirements, the
3457
+
* architecture will return any suitable mapping because addr is 0.
3470
3458
*/
3471
-
addr = vm_unmapped_area(&info);
3472
-
if (offset_in_page(addr)) {
3473
-
info.flags = 0;
3474
-
info.low_limit = TASK_UNMAPPED_BASE;
3475
-
info.high_limit = mmap_end;
3476
-
addr = vm_unmapped_area(&info);
3477
-
}
3478
-
3479
-
return addr;
3459
+
filp = NULL;
3460
+
flags |= MAP_SHARED;
3461
+
pgoff = 0; /* has been translated to ptr above */
3462
+
#ifdef SHM_COLOUR
3463
+
addr = (uintptr_t) ptr;
3464
+
#else
3465
+
addr = 0UL;
3466
+
#endif
3467
+
return current->mm->get_unmapped_area(filp, addr, len, pgoff, flags);
3480
3468
}
3481
3469
3482
3470
#else /* !CONFIG_MMU */
···
3870
3870
ctx->syscall_iopoll = 1;
3871
3871
3872
3872
ctx->compat = in_compat_syscall();
3873
-
if (!capable(CAP_IPC_LOCK))
3873
+
if (!ns_capable_noaudit(&init_user_ns, CAP_IPC_LOCK))
3874
3874
ctx->user = get_uid(current_user());
3875
3875
3876
3876
/*