brintos

brintos / linux-shallow public Read only

0
0
Text · 4.2 KiB · 7eb7c60 Raw
121 lines · plain
1.. SPDX-License-Identifier: GPL-2.02 3======================4Memory Protection Keys5======================6 7Memory Protection Keys provide a mechanism for enforcing page-based8protections, but without requiring modification of the page tables when an9application changes protection domains.10 11Pkeys Userspace (PKU) is a feature which can be found on:12        * Intel server CPUs, Skylake and later13        * Intel client CPUs, Tiger Lake (11th Gen Core) and later14        * Future AMD CPUs15        * arm64 CPUs implementing the Permission Overlay Extension (FEAT_S1POE)16 17x86_6418======19Pkeys work by dedicating 4 previously Reserved bits in each page table entry to20a "protection key", giving 16 possible keys.21 22Protections for each key are defined with a per-CPU user-accessible register23(PKRU).  Each of these is a 32-bit register storing two bits (Access Disable24and Write Disable) for each of 16 keys.25 26Being a CPU register, PKRU is inherently thread-local, potentially giving each27thread a different set of protections from every other thread.28 29There are two instructions (RDPKRU/WRPKRU) for reading and writing to the30register.  The feature is only available in 64-bit mode, even though there is31theoretically space in the PAE PTEs.  These permissions are enforced on data32access only and have no effect on instruction fetches.33 34arm6435=====36 37Pkeys use 3 bits in each page table entry, to encode a "protection key index",38giving 8 possible keys.39 40Protections for each key are defined with a per-CPU user-writable system41register (POR_EL0).  This is a 64-bit register encoding read, write and execute42overlay permissions for each protection key index.43 44Being a CPU register, POR_EL0 is inherently thread-local, potentially giving45each thread a different set of protections from every other thread.46 47Unlike x86_64, the protection key permissions also apply to instruction48fetches.49 50Syscalls51========52 53There are 3 system calls which directly interact with pkeys::54 55	int pkey_alloc(unsigned long flags, unsigned long init_access_rights)56	int pkey_free(int pkey);57	int pkey_mprotect(unsigned long start, size_t len,58			  unsigned long prot, int pkey);59 60Before a pkey can be used, it must first be allocated with pkey_alloc().  An61application writes to the architecture specific CPU register directly in order62to change access permissions to memory covered with a key.  In this example63this is wrapped by a C function called pkey_set().64::65 66	int real_prot = PROT_READ|PROT_WRITE;67	pkey = pkey_alloc(0, PKEY_DISABLE_WRITE);68	ptr = mmap(NULL, PAGE_SIZE, PROT_NONE, MAP_ANONYMOUS|MAP_PRIVATE, -1, 0);69	ret = pkey_mprotect(ptr, PAGE_SIZE, real_prot, pkey);70	... application runs here71 72Now, if the application needs to update the data at 'ptr', it can73gain access, do the update, then remove its write access::74 75	pkey_set(pkey, 0); // clear PKEY_DISABLE_WRITE76	*ptr = foo; // assign something77	pkey_set(pkey, PKEY_DISABLE_WRITE); // set PKEY_DISABLE_WRITE again78 79Now when it frees the memory, it will also free the pkey since it80is no longer in use::81 82	munmap(ptr, PAGE_SIZE);83	pkey_free(pkey);84 85.. note:: pkey_set() is a wrapper around writing to the CPU register.86          Example implementations can be found in87          tools/testing/selftests/mm/pkey-{arm64,powerpc,x86}.h88 89Behavior90========91 92The kernel attempts to make protection keys consistent with the93behavior of a plain mprotect().  For instance if you do this::94 95	mprotect(ptr, size, PROT_NONE);96	something(ptr);97 98you can expect the same effects with protection keys when doing this::99 100	pkey = pkey_alloc(0, PKEY_DISABLE_WRITE | PKEY_DISABLE_READ);101	pkey_mprotect(ptr, size, PROT_READ|PROT_WRITE, pkey);102	something(ptr);103 104That should be true whether something() is a direct access to 'ptr'105like::106 107	*ptr = foo;108 109or when the kernel does the access on the application's behalf like110with a read()::111 112	read(fd, ptr, 1);113 114The kernel will send a SIGSEGV in both cases, but si_code will be set115to SEGV_PKERR when violating protection keys versus SEGV_ACCERR when116the plain mprotect() permissions are violated.117 118Note that kernel accesses from a kthread (such as io_uring) will use a default119value for the protection key register and so will not be consistent with120userspace's value of the register or mprotect().121