1 .. SPDX-License-Identifier: GPL-2.0
3 ===============================
4 Kernel level exception handling
5 ===============================
7 Commentary by Joerg Pommnitz <joerg@raleigh.ibm.com>
9 When a process runs in kernel mode, it often has to access user
10 mode memory whose address has been passed by an untrusted program.
11 To protect itself the kernel has to verify this address.
13 In older versions of Linux this was done with the
14 int verify_area(int type, const void * addr, unsigned long size)
15 function (which has since been replaced by access_ok()).
17 This function verified that the memory area starting at address
18 'addr' and of size 'size' was accessible for the operation specified
19 in type (read or write). To do this, verify_read had to look up the
20 virtual memory area (vma) that contained the address addr. In the
21 normal case (correctly working program), this test was successful.
22 It only failed for a few buggy programs. In some kernel profiling
23 tests, this normally unneeded verification used up a considerable
26 To overcome this situation, Linus decided to let the virtual memory
27 hardware present in every Linux-capable CPU handle this test.
31 Whenever the kernel tries to access an address that is currently not
32 accessible, the CPU generates a page fault exception and calls the
35 void exc_page_fault(struct pt_regs *regs, unsigned long error_code)
37 in arch/x86/mm/fault.c. The parameters on the stack are set up by
38 the low level assembly glue in arch/x86/entry/entry_32.S. The parameter
39 regs is a pointer to the saved registers on the stack, error_code
40 contains a reason code for the exception.
42 exc_page_fault() first obtains the inaccessible address from the CPU
43 control register CR2. If the address is within the virtual address
44 space of the process, the fault probably occurred, because the page
45 was not swapped in, write protected or something similar. However,
46 we are interested in the other case: the address is not valid, there
47 is no vma that contains this address. In this case, the kernel jumps
48 to the bad_area label.
50 There it uses the address of the instruction that caused the exception
51 (i.e. regs->eip) to find an address where the execution can continue
52 (fixup). If this search is successful, the fault handler modifies the
53 return address (again regs->eip) and returns. The execution will
54 continue at the address in fixup.
56 Where does fixup point to?
58 Since we jump to the contents of fixup, fixup obviously points
59 to executable code. This code is hidden inside the user access macros.
60 I have picked the get_user() macro defined in arch/x86/include/asm/uaccess.h
61 as an example. The definition is somewhat hard to follow, so let's peek at
62 the code generated by the preprocessor and the compiler. I selected
63 the get_user() call in drivers/char/sysrq.c for a detailed examination.
65 The original code in sysrq.c line 587::
69 The preprocessor output (edited to become somewhat readable)::
73 long __gu_err = - 14 , __gu_val = 0;
74 const __typeof__(*( ( buf ) )) *__gu_addr = ((buf));
75 if (((((0 + current_set[0])->tss.segment) == 0x18 ) ||
76 (((sizeof(*(buf))) <= 0xC0000000UL) &&
77 ((unsigned long)(__gu_addr ) <= 0xC0000000UL - (sizeof(*(buf)))))))
80 switch ((sizeof(*(buf)))) {
83 "1: mov" "b" " %2,%" "b" "1\n"
85 ".section .fixup,\"ax\"\n"
87 " xor" "b" " %" "b" "1,%" "b" "1\n"
89 ".section __ex_table,\"a\"\n"
92 ".text" : "=r"(__gu_err), "=q" (__gu_val): "m"((*(struct __large_struct *)
93 ( __gu_addr )) ), "i"(- 14 ), "0"( __gu_err )) ;
97 "1: mov" "w" " %2,%" "w" "1\n"
99 ".section .fixup,\"ax\"\n"
101 " xor" "w" " %" "w" "1,%" "w" "1\n"
103 ".section __ex_table,\"a\"\n"
106 ".text" : "=r"(__gu_err), "=r" (__gu_val) : "m"((*(struct __large_struct *)
107 ( __gu_addr )) ), "i"(- 14 ), "0"( __gu_err ));
110 __asm__ __volatile__(
111 "1: mov" "l" " %2,%" "" "1\n"
113 ".section .fixup,\"ax\"\n"
115 " xor" "l" " %" "" "1,%" "" "1\n"
117 ".section __ex_table,\"a\"\n"
118 " .align 4\n" " .long 1b,3b\n"
119 ".text" : "=r"(__gu_err), "=r" (__gu_val) : "m"((*(struct __large_struct *)
120 ( __gu_addr )) ), "i"(- 14 ), "0"(__gu_err));
123 (__gu_val) = __get_user_bad();
126 ((c)) = (__typeof__(*((buf))))__gu_val;
131 WOW! Black GCC/assembly magic. This is impossible to follow, so let's
132 see what code gcc generates::
135 > movl current_set,%eax
138 > cmpl $-1073741825,64(%esp)
144 > 1: movb (%ebx),%dl /* this is the actual user access */
146 > .section .fixup,"ax"
150 > .section __ex_table,"a"
158 The optimizer does a good job and gives us something we can actually
159 understand. Can we? The actual user access is quite obvious. Thanks
160 to the unified address space we can just access the address in user
161 memory. But what does the .section stuff do?????
163 To understand this we have to look at the final kernel::
165 > objdump --section-headers vmlinux
167 > vmlinux: file format elf32-i386
170 > Idx Name Size VMA LMA File off Algn
171 > 0 .text 00098f40 c0100000 c0100000 00001000 2**4
172 > CONTENTS, ALLOC, LOAD, READONLY, CODE
173 > 1 .fixup 000016bc c0198f40 c0198f40 00099f40 2**0
174 > CONTENTS, ALLOC, LOAD, READONLY, CODE
175 > 2 .rodata 0000f127 c019a5fc c019a5fc 0009b5fc 2**2
176 > CONTENTS, ALLOC, LOAD, READONLY, DATA
177 > 3 __ex_table 000015c0 c01a9724 c01a9724 000aa724 2**2
178 > CONTENTS, ALLOC, LOAD, READONLY, DATA
179 > 4 .data 0000ea58 c01abcf0 c01abcf0 000abcf0 2**4
180 > CONTENTS, ALLOC, LOAD, DATA
181 > 5 .bss 00018e21 c01ba748 c01ba748 000ba748 2**2
183 > 6 .comment 00000ec4 00000000 00000000 000ba748 2**0
185 > 7 .note 00001068 00000ec4 00000ec4 000bb60c 2**0
188 There are obviously 2 non standard ELF sections in the generated object
189 file. But first we want to find out what happened to our code in the
190 final kernel executable::
192 > objdump --disassemble --section=.text vmlinux
194 > c017e785 <do_con_write+c1> xorl %edx,%edx
195 > c017e787 <do_con_write+c3> movl 0xc01c7bec,%eax
196 > c017e78c <do_con_write+c8> cmpl $0x18,0x314(%eax)
197 > c017e793 <do_con_write+cf> je c017e79f <do_con_write+db>
198 > c017e795 <do_con_write+d1> cmpl $0xbfffffff,0x40(%esp,1)
199 > c017e79d <do_con_write+d9> ja c017e7a7 <do_con_write+e3>
200 > c017e79f <do_con_write+db> movl %edx,%eax
201 > c017e7a1 <do_con_write+dd> movl 0x40(%esp,1),%ebx
202 > c017e7a5 <do_con_write+e1> movb (%ebx),%dl
203 > c017e7a7 <do_con_write+e3> movzbl %dl,%esi
205 The whole user memory access is reduced to 10 x86 machine instructions.
206 The instructions bracketed in the .section directives are no longer
207 in the normal execution path. They are located in a different section
208 of the executable file::
210 > objdump --disassemble --section=.fixup vmlinux
212 > c0199ff5 <.fixup+10b5> movl $0xfffffff2,%eax
213 > c0199ffa <.fixup+10ba> xorb %dl,%dl
214 > c0199ffc <.fixup+10bc> jmp c017e7a7 <do_con_write+e3>
218 > objdump --full-contents --section=__ex_table vmlinux
220 > c01aa7c4 93c017c0 e09f19c0 97c017c0 99c017c0 ................
221 > c01aa7d4 f6c217c0 e99f19c0 a5e717c0 f59f19c0 ................
222 > c01aa7e4 080a18c0 01a019c0 0a0a18c0 04a019c0 ................
224 or in human readable byte order::
226 > c01aa7c4 c017c093 c0199fe0 c017c097 c017c099 ................
227 > c01aa7d4 c017c2f6 c0199fe9 c017e7a5 c0199ff5 ................
229 this is the interesting part!
230 > c01aa7e4 c0180a08 c019a001 c0180a0a c019a004 ................
232 What happened? The assembly directives::
235 .section __ex_table,"a"
237 told the assembler to move the following code to the specified
238 sections in the ELF object file. So the instructions::
244 ended up in the .fixup section of the object file and the addresses::
248 ended up in the __ex_table section of the object file. 1b and 3b
249 are local labels. The local label 1b (1b stands for next label 1
250 backward) is the address of the instruction that might fault, i.e.
251 in our case the address of the label 1 is c017e7a5:
252 the original assembly code: > 1: movb (%ebx),%dl
253 and linked in vmlinux : > c017e7a5 <do_con_write+e1> movb (%ebx),%dl
255 The local label 3 (backwards again) is the address of the code to handle
256 the fault, in our case the actual value is c0199ff5:
257 the original assembly code: > 3: movl $-14,%eax
258 and linked in vmlinux : > c0199ff5 <.fixup+10b5> movl $0xfffffff2,%eax
260 If the fixup was able to handle the exception, control flow may be returned
261 to the instruction after the one that triggered the fault, ie. local label 2b.
265 > .section __ex_table,"a"
269 becomes the value pair::
271 > c01aa7d4 c017c2f6 c0199fe9 c017e7a5 c0199ff5 ................
275 c017e7a5,c0199ff5 in the exception table of the kernel.
277 So, what actually happens if a fault from kernel mode with no suitable
280 #. access to invalid address::
282 > c017e7a5 <do_con_write+e1> movb (%ebx),%dl
283 #. MMU generates exception
284 #. CPU calls exc_page_fault()
285 #. exc_page_fault() calls do_user_addr_fault()
286 #. do_user_addr_fault() calls kernelmode_fixup_or_oops()
287 #. kernelmode_fixup_or_oops() calls fixup_exception() (regs->eip == c017e7a5);
288 #. fixup_exception() calls search_exception_tables()
289 #. search_exception_tables() looks up the address c017e7a5 in the
290 exception table (i.e. the contents of the ELF section __ex_table)
291 and returns the address of the associated fault handle code c0199ff5.
292 #. fixup_exception() modifies its own return address to point to the fault
293 handle code and returns.
294 #. execution continues in the fault handling code.
295 #. a) EAX becomes -EFAULT (== -14)
296 b) DL becomes zero (the value we "read" from user space)
297 c) execution continues at local label 2 (address of the
298 instruction immediately after the faulting user access).
300 The steps 8a to 8c in a certain way emulate the faulting instruction.
302 That's it, mostly. If you look at our example, you might ask why
303 we set EAX to -EFAULT in the exception handler code. Well, the
304 get_user() macro actually returns a value: 0, if the user access was
305 successful, -EFAULT on failure. Our original code did not test this
306 return value, however the inline assembly code in get_user() tries to
307 return -EFAULT. GCC selected EAX to return this value.
310 Due to the way that the exception table is built and needs to be ordered,
311 only use exceptions for code in the .text section. Any other section
312 will cause the exception table to not be sorted correctly, and the
313 exceptions will fail.
315 Things changed when 64-bit support was added to x86 Linux. Rather than
316 double the size of the exception table by expanding the two entries
317 from 32-bits to 64 bits, a clever trick was used to store addresses
318 as relative offsets from the table itself. The assembly code changed
326 and the C-code that uses these values converts back to absolute addresses
329 ex_insn_addr(const struct exception_table_entry *x)
331 return (unsigned long)&x->insn + x->insn;
334 In v4.6 the exception table entry was expanded with a new field "handler".
335 This is also 32-bits wide and contains a third relative function
336 pointer which points to one of:
338 1) ``int ex_handler_default(const struct exception_table_entry *fixup)``
339 This is legacy case that just jumps to the fixup code
341 2) ``int ex_handler_fault(const struct exception_table_entry *fixup)``
342 This case provides the fault number of the trap that occurred at
343 entry->insn. It is used to distinguish page faults from machine
346 More functions can easily be added.
348 CONFIG_BUILDTIME_TABLE_SORT allows the __ex_table section to be sorted post
349 link of the kernel image, via a host utility scripts/sorttable. It will set the
350 symbol main_extable_sort_needed to 0, avoiding sorting the __ex_table section
351 at boot time. With the exception table sorted, at runtime when an exception
352 occurs we can quickly lookup the __ex_table entry via binary search.
354 This is not just a boot time optimization, some architectures require this
355 table to be sorted in order to handle exceptions relatively early in the boot
356 process. For example, i386 makes use of this form of exception handling before
357 paging support is even enabled!