cachepc-linux

Fork of AMDESE/linux with modifications for CachePC side-channel attack
git clone https://git.sinitax.com/sinitax/cachepc-linux
Log | Files | Refs | README | LICENSE | sfeed.txt

trap.c (7986B)


      1// SPDX-License-Identifier: GPL-2.0
      2/*
      3 * Copyright (C) 2000 - 2007 Jeff Dike (jdike@{addtoit,linux.intel}.com)
      4 */
      5
      6#include <linux/mm.h>
      7#include <linux/sched/signal.h>
      8#include <linux/hardirq.h>
      9#include <linux/module.h>
     10#include <linux/uaccess.h>
     11#include <linux/sched/debug.h>
     12#include <asm/current.h>
     13#include <asm/tlbflush.h>
     14#include <arch.h>
     15#include <as-layout.h>
     16#include <kern_util.h>
     17#include <os.h>
     18#include <skas.h>
     19
     20/*
     21 * Note this is constrained to return 0, -EFAULT, -EACCES, -ENOMEM by
     22 * segv().
     23 */
     24int handle_page_fault(unsigned long address, unsigned long ip,
     25		      int is_write, int is_user, int *code_out)
     26{
     27	struct mm_struct *mm = current->mm;
     28	struct vm_area_struct *vma;
     29	pmd_t *pmd;
     30	pte_t *pte;
     31	int err = -EFAULT;
     32	unsigned int flags = FAULT_FLAG_DEFAULT;
     33
     34	*code_out = SEGV_MAPERR;
     35
     36	/*
     37	 * If the fault was with pagefaults disabled, don't take the fault, just
     38	 * fail.
     39	 */
     40	if (faulthandler_disabled())
     41		goto out_nosemaphore;
     42
     43	if (is_user)
     44		flags |= FAULT_FLAG_USER;
     45retry:
     46	mmap_read_lock(mm);
     47	vma = find_vma(mm, address);
     48	if (!vma)
     49		goto out;
     50	else if (vma->vm_start <= address)
     51		goto good_area;
     52	else if (!(vma->vm_flags & VM_GROWSDOWN))
     53		goto out;
     54	else if (is_user && !ARCH_IS_STACKGROW(address))
     55		goto out;
     56	else if (expand_stack(vma, address))
     57		goto out;
     58
     59good_area:
     60	*code_out = SEGV_ACCERR;
     61	if (is_write) {
     62		if (!(vma->vm_flags & VM_WRITE))
     63			goto out;
     64		flags |= FAULT_FLAG_WRITE;
     65	} else {
     66		/* Don't require VM_READ|VM_EXEC for write faults! */
     67		if (!(vma->vm_flags & (VM_READ | VM_EXEC)))
     68			goto out;
     69	}
     70
     71	do {
     72		vm_fault_t fault;
     73
     74		fault = handle_mm_fault(vma, address, flags, NULL);
     75
     76		if ((fault & VM_FAULT_RETRY) && fatal_signal_pending(current))
     77			goto out_nosemaphore;
     78
     79		if (unlikely(fault & VM_FAULT_ERROR)) {
     80			if (fault & VM_FAULT_OOM) {
     81				goto out_of_memory;
     82			} else if (fault & VM_FAULT_SIGSEGV) {
     83				goto out;
     84			} else if (fault & VM_FAULT_SIGBUS) {
     85				err = -EACCES;
     86				goto out;
     87			}
     88			BUG();
     89		}
     90		if (fault & VM_FAULT_RETRY) {
     91			flags |= FAULT_FLAG_TRIED;
     92
     93			goto retry;
     94		}
     95
     96		pmd = pmd_off(mm, address);
     97		pte = pte_offset_kernel(pmd, address);
     98	} while (!pte_present(*pte));
     99	err = 0;
    100	/*
    101	 * The below warning was added in place of
    102	 *	pte_mkyoung(); if (is_write) pte_mkdirty();
    103	 * If it's triggered, we'd see normally a hang here (a clean pte is
    104	 * marked read-only to emulate the dirty bit).
    105	 * However, the generic code can mark a PTE writable but clean on a
    106	 * concurrent read fault, triggering this harmlessly. So comment it out.
    107	 */
    108#if 0
    109	WARN_ON(!pte_young(*pte) || (is_write && !pte_dirty(*pte)));
    110#endif
    111	flush_tlb_page(vma, address);
    112out:
    113	mmap_read_unlock(mm);
    114out_nosemaphore:
    115	return err;
    116
    117out_of_memory:
    118	/*
    119	 * We ran out of memory, call the OOM killer, and return the userspace
    120	 * (which will retry the fault, or kill us if we got oom-killed).
    121	 */
    122	mmap_read_unlock(mm);
    123	if (!is_user)
    124		goto out_nosemaphore;
    125	pagefault_out_of_memory();
    126	return 0;
    127}
    128
    129static void show_segv_info(struct uml_pt_regs *regs)
    130{
    131	struct task_struct *tsk = current;
    132	struct faultinfo *fi = UPT_FAULTINFO(regs);
    133
    134	if (!unhandled_signal(tsk, SIGSEGV))
    135		return;
    136
    137	if (!printk_ratelimit())
    138		return;
    139
    140	printk("%s%s[%d]: segfault at %lx ip %px sp %px error %x",
    141		task_pid_nr(tsk) > 1 ? KERN_INFO : KERN_EMERG,
    142		tsk->comm, task_pid_nr(tsk), FAULT_ADDRESS(*fi),
    143		(void *)UPT_IP(regs), (void *)UPT_SP(regs),
    144		fi->error_code);
    145
    146	print_vma_addr(KERN_CONT " in ", UPT_IP(regs));
    147	printk(KERN_CONT "\n");
    148}
    149
    150static void bad_segv(struct faultinfo fi, unsigned long ip)
    151{
    152	current->thread.arch.faultinfo = fi;
    153	force_sig_fault(SIGSEGV, SEGV_ACCERR, (void __user *) FAULT_ADDRESS(fi));
    154}
    155
    156void fatal_sigsegv(void)
    157{
    158	force_fatal_sig(SIGSEGV);
    159	do_signal(&current->thread.regs);
    160	/*
    161	 * This is to tell gcc that we're not returning - do_signal
    162	 * can, in general, return, but in this case, it's not, since
    163	 * we just got a fatal SIGSEGV queued.
    164	 */
    165	os_dump_core();
    166}
    167
    168/**
    169 * segv_handler() - the SIGSEGV handler
    170 * @sig:	the signal number
    171 * @unused_si:	the signal info struct; unused in this handler
    172 * @regs:	the ptrace register information
    173 *
    174 * The handler first extracts the faultinfo from the UML ptrace regs struct.
    175 * If the userfault did not happen in an UML userspace process, bad_segv is called.
    176 * Otherwise the signal did happen in a cloned userspace process, handle it.
    177 */
    178void segv_handler(int sig, struct siginfo *unused_si, struct uml_pt_regs *regs)
    179{
    180	struct faultinfo * fi = UPT_FAULTINFO(regs);
    181
    182	if (UPT_IS_USER(regs) && !SEGV_IS_FIXABLE(fi)) {
    183		show_segv_info(regs);
    184		bad_segv(*fi, UPT_IP(regs));
    185		return;
    186	}
    187	segv(*fi, UPT_IP(regs), UPT_IS_USER(regs), regs);
    188}
    189
    190/*
    191 * We give a *copy* of the faultinfo in the regs to segv.
    192 * This must be done, since nesting SEGVs could overwrite
    193 * the info in the regs. A pointer to the info then would
    194 * give us bad data!
    195 */
    196unsigned long segv(struct faultinfo fi, unsigned long ip, int is_user,
    197		   struct uml_pt_regs *regs)
    198{
    199	jmp_buf *catcher;
    200	int si_code;
    201	int err;
    202	int is_write = FAULT_WRITE(fi);
    203	unsigned long address = FAULT_ADDRESS(fi);
    204
    205	if (!is_user && regs)
    206		current->thread.segv_regs = container_of(regs, struct pt_regs, regs);
    207
    208	if (!is_user && (address >= start_vm) && (address < end_vm)) {
    209		flush_tlb_kernel_vm();
    210		goto out;
    211	}
    212	else if (current->mm == NULL) {
    213		show_regs(container_of(regs, struct pt_regs, regs));
    214		panic("Segfault with no mm");
    215	}
    216	else if (!is_user && address > PAGE_SIZE && address < TASK_SIZE) {
    217		show_regs(container_of(regs, struct pt_regs, regs));
    218		panic("Kernel tried to access user memory at addr 0x%lx, ip 0x%lx",
    219		       address, ip);
    220	}
    221
    222	if (SEGV_IS_FIXABLE(&fi))
    223		err = handle_page_fault(address, ip, is_write, is_user,
    224					&si_code);
    225	else {
    226		err = -EFAULT;
    227		/*
    228		 * A thread accessed NULL, we get a fault, but CR2 is invalid.
    229		 * This code is used in __do_copy_from_user() of TT mode.
    230		 * XXX tt mode is gone, so maybe this isn't needed any more
    231		 */
    232		address = 0;
    233	}
    234
    235	catcher = current->thread.fault_catcher;
    236	if (!err)
    237		goto out;
    238	else if (catcher != NULL) {
    239		current->thread.fault_addr = (void *) address;
    240		UML_LONGJMP(catcher, 1);
    241	}
    242	else if (current->thread.fault_addr != NULL)
    243		panic("fault_addr set but no fault catcher");
    244	else if (!is_user && arch_fixup(ip, regs))
    245		goto out;
    246
    247	if (!is_user) {
    248		show_regs(container_of(regs, struct pt_regs, regs));
    249		panic("Kernel mode fault at addr 0x%lx, ip 0x%lx",
    250		      address, ip);
    251	}
    252
    253	show_segv_info(regs);
    254
    255	if (err == -EACCES) {
    256		current->thread.arch.faultinfo = fi;
    257		force_sig_fault(SIGBUS, BUS_ADRERR, (void __user *)address);
    258	} else {
    259		BUG_ON(err != -EFAULT);
    260		current->thread.arch.faultinfo = fi;
    261		force_sig_fault(SIGSEGV, si_code, (void __user *) address);
    262	}
    263
    264out:
    265	if (regs)
    266		current->thread.segv_regs = NULL;
    267
    268	return 0;
    269}
    270
    271void relay_signal(int sig, struct siginfo *si, struct uml_pt_regs *regs)
    272{
    273	int code, err;
    274	if (!UPT_IS_USER(regs)) {
    275		if (sig == SIGBUS)
    276			printk(KERN_ERR "Bus error - the host /dev/shm or /tmp "
    277			       "mount likely just ran out of space\n");
    278		panic("Kernel mode signal %d", sig);
    279	}
    280
    281	arch_examine_signal(sig, regs);
    282
    283	/* Is the signal layout for the signal known?
    284	 * Signal data must be scrubbed to prevent information leaks.
    285	 */
    286	code = si->si_code;
    287	err = si->si_errno;
    288	if ((err == 0) && (siginfo_layout(sig, code) == SIL_FAULT)) {
    289		struct faultinfo *fi = UPT_FAULTINFO(regs);
    290		current->thread.arch.faultinfo = *fi;
    291		force_sig_fault(sig, code, (void __user *)FAULT_ADDRESS(*fi));
    292	} else {
    293		printk(KERN_ERR "Attempted to relay unknown signal %d (si_code = %d) with errno %d\n",
    294		       sig, code, err);
    295		force_sig(sig);
    296	}
    297}
    298
    299void bus_handler(int sig, struct siginfo *si, struct uml_pt_regs *regs)
    300{
    301	if (current->thread.fault_catcher != NULL)
    302		UML_LONGJMP(current->thread.fault_catcher, 1);
    303	else
    304		relay_signal(sig, si, regs);
    305}
    306
    307void winch(int sig, struct siginfo *unused_si, struct uml_pt_regs *regs)
    308{
    309	do_IRQ(WINCH_IRQ, regs);
    310}