nfsd: don't fail unchecked creates of non-special files
[zen-stable.git] / arch / microblaze / mm / fault.c
blobae97d2ccdc22c88da115c8fd8400ebc697d04804
1 /*
2 * arch/microblaze/mm/fault.c
4 * Copyright (C) 2007 Xilinx, Inc. All rights reserved.
6 * Derived from "arch/ppc/mm/fault.c"
7 * Copyright (C) 1995-1996 Gary Thomas (gdt@linuxppc.org)
9 * Derived from "arch/i386/mm/fault.c"
10 * Copyright (C) 1991, 1992, 1993, 1994 Linus Torvalds
12 * Modified by Cort Dougan and Paul Mackerras.
14 * This file is subject to the terms and conditions of the GNU General
15 * Public License. See the file COPYING in the main directory of this
16 * archive for more details.
20 #include <linux/module.h>
21 #include <linux/signal.h>
22 #include <linux/sched.h>
23 #include <linux/kernel.h>
24 #include <linux/errno.h>
25 #include <linux/string.h>
26 #include <linux/types.h>
27 #include <linux/ptrace.h>
28 #include <linux/mman.h>
29 #include <linux/mm.h>
30 #include <linux/interrupt.h>
32 #include <asm/page.h>
33 #include <asm/pgtable.h>
34 #include <asm/mmu.h>
35 #include <asm/mmu_context.h>
36 #include <asm/system.h>
37 #include <linux/uaccess.h>
38 #include <asm/exceptions.h>
40 static unsigned long pte_misses; /* updated by do_page_fault() */
41 static unsigned long pte_errors; /* updated by do_page_fault() */
44 * Check whether the instruction at regs->pc is a store using
45 * an update addressing form which will update r1.
47 static int store_updates_sp(struct pt_regs *regs)
49 unsigned int inst;
51 if (get_user(inst, (unsigned int __user *)regs->pc))
52 return 0;
53 /* check for 1 in the rD field */
54 if (((inst >> 21) & 0x1f) != 1)
55 return 0;
56 /* check for store opcodes */
57 if ((inst & 0xd0000000) == 0xd0000000)
58 return 1;
59 return 0;
64 * bad_page_fault is called when we have a bad access from the kernel.
65 * It is called from do_page_fault above and from some of the procedures
66 * in traps.c.
68 void bad_page_fault(struct pt_regs *regs, unsigned long address, int sig)
70 const struct exception_table_entry *fixup;
71 /* MS: no context */
72 /* Are we prepared to handle this fault? */
73 fixup = search_exception_tables(regs->pc);
74 if (fixup) {
75 regs->pc = fixup->fixup;
76 return;
79 /* kernel has accessed a bad area */
80 die("kernel access of bad area", regs, sig);
84 * The error_code parameter is ESR for a data fault,
85 * 0 for an instruction fault.
87 void do_page_fault(struct pt_regs *regs, unsigned long address,
88 unsigned long error_code)
90 struct vm_area_struct *vma;
91 struct mm_struct *mm = current->mm;
92 siginfo_t info;
93 int code = SEGV_MAPERR;
94 int is_write = error_code & ESR_S;
95 int fault;
97 regs->ear = address;
98 regs->esr = error_code;
100 /* On a kernel SLB miss we can only check for a valid exception entry */
101 if (unlikely(kernel_mode(regs) && (address >= TASK_SIZE))) {
102 printk(KERN_WARNING "kernel task_size exceed");
103 _exception(SIGSEGV, regs, code, address);
106 /* for instr TLB miss and instr storage exception ESR_S is undefined */
107 if ((error_code & 0x13) == 0x13 || (error_code & 0x11) == 0x11)
108 is_write = 0;
110 if (unlikely(in_atomic() || !mm)) {
111 if (kernel_mode(regs))
112 goto bad_area_nosemaphore;
114 /* in_atomic() in user mode is really bad,
115 as is current->mm == NULL. */
116 printk(KERN_EMERG "Page fault in user mode with "
117 "in_atomic(), mm = %p\n", mm);
118 printk(KERN_EMERG "r15 = %lx MSR = %lx\n",
119 regs->r15, regs->msr);
120 die("Weird page fault", regs, SIGSEGV);
123 /* When running in the kernel we expect faults to occur only to
124 * addresses in user space. All other faults represent errors in the
125 * kernel and should generate an OOPS. Unfortunately, in the case of an
126 * erroneous fault occurring in a code path which already holds mmap_sem
127 * we will deadlock attempting to validate the fault against the
128 * address space. Luckily the kernel only validly references user
129 * space from well defined areas of code, which are listed in the
130 * exceptions table.
132 * As the vast majority of faults will be valid we will only perform
133 * the source reference check when there is a possibility of a deadlock.
134 * Attempt to lock the address space, if we cannot we then validate the
135 * source. If this is invalid we can skip the address space check,
136 * thus avoiding the deadlock.
138 if (unlikely(!down_read_trylock(&mm->mmap_sem))) {
139 if (kernel_mode(regs) && !search_exception_tables(regs->pc))
140 goto bad_area_nosemaphore;
142 down_read(&mm->mmap_sem);
145 vma = find_vma(mm, address);
146 if (unlikely(!vma))
147 goto bad_area;
149 if (vma->vm_start <= address)
150 goto good_area;
152 if (unlikely(!(vma->vm_flags & VM_GROWSDOWN)))
153 goto bad_area;
155 if (unlikely(!is_write))
156 goto bad_area;
159 * N.B. The ABI allows programs to access up to
160 * a few hundred bytes below the stack pointer (TBD).
161 * The kernel signal delivery code writes up to about 1.5kB
162 * below the stack pointer (r1) before decrementing it.
163 * The exec code can write slightly over 640kB to the stack
164 * before setting the user r1. Thus we allow the stack to
165 * expand to 1MB without further checks.
167 if (unlikely(address + 0x100000 < vma->vm_end)) {
169 /* get user regs even if this fault is in kernel mode */
170 struct pt_regs *uregs = current->thread.regs;
171 if (uregs == NULL)
172 goto bad_area;
175 * A user-mode access to an address a long way below
176 * the stack pointer is only valid if the instruction
177 * is one which would update the stack pointer to the
178 * address accessed if the instruction completed,
179 * i.e. either stwu rs,n(r1) or stwux rs,r1,rb
180 * (or the byte, halfword, float or double forms).
182 * If we don't check this then any write to the area
183 * between the last mapped region and the stack will
184 * expand the stack rather than segfaulting.
186 if (address + 2048 < uregs->r1
187 && (kernel_mode(regs) || !store_updates_sp(regs)))
188 goto bad_area;
190 if (expand_stack(vma, address))
191 goto bad_area;
193 good_area:
194 code = SEGV_ACCERR;
196 /* a write */
197 if (unlikely(is_write)) {
198 if (unlikely(!(vma->vm_flags & VM_WRITE)))
199 goto bad_area;
200 /* a read */
201 } else {
202 /* protection fault */
203 if (unlikely(error_code & 0x08000000))
204 goto bad_area;
205 if (unlikely(!(vma->vm_flags & (VM_READ | VM_EXEC))))
206 goto bad_area;
210 * If for any reason at all we couldn't handle the fault,
211 * make sure we exit gracefully rather than endlessly redo
212 * the fault.
214 fault = handle_mm_fault(mm, vma, address, is_write ? FAULT_FLAG_WRITE : 0);
215 if (unlikely(fault & VM_FAULT_ERROR)) {
216 if (fault & VM_FAULT_OOM)
217 goto out_of_memory;
218 else if (fault & VM_FAULT_SIGBUS)
219 goto do_sigbus;
220 BUG();
222 if (unlikely(fault & VM_FAULT_MAJOR))
223 current->maj_flt++;
224 else
225 current->min_flt++;
226 up_read(&mm->mmap_sem);
228 * keep track of tlb+htab misses that are good addrs but
229 * just need pte's created via handle_mm_fault()
230 * -- Cort
232 pte_misses++;
233 return;
235 bad_area:
236 up_read(&mm->mmap_sem);
238 bad_area_nosemaphore:
239 pte_errors++;
241 /* User mode accesses cause a SIGSEGV */
242 if (user_mode(regs)) {
243 _exception(SIGSEGV, regs, code, address);
244 /* info.si_signo = SIGSEGV;
245 info.si_errno = 0;
246 info.si_code = code;
247 info.si_addr = (void *) address;
248 force_sig_info(SIGSEGV, &info, current);*/
249 return;
252 bad_page_fault(regs, address, SIGSEGV);
253 return;
256 * We ran out of memory, or some other thing happened to us that made
257 * us unable to handle the page fault gracefully.
259 out_of_memory:
260 up_read(&mm->mmap_sem);
261 if (!user_mode(regs))
262 bad_page_fault(regs, address, SIGKILL);
263 else
264 pagefault_out_of_memory();
265 return;
267 do_sigbus:
268 up_read(&mm->mmap_sem);
269 if (user_mode(regs)) {
270 info.si_signo = SIGBUS;
271 info.si_errno = 0;
272 info.si_code = BUS_ADRERR;
273 info.si_addr = (void __user *)address;
274 force_sig_info(SIGBUS, &info, current);
275 return;
277 bad_page_fault(regs, address, SIGBUS);