Back to home page

OSCL-LXR
 
 

     

0001 // SPDX-License-Identifier: GPL-2.0
0002 /*
0003  *  Copyright (C) 1995  Linus Torvalds
0004  *  Copyright (C) 2001, 2002 Andi Kleen, SuSE Labs.
0005  *  Copyright (C) 2008-2009, Red Hat Inc., Ingo Molnar
0006  */
0007 #include <linux/sched.h>        /* test_thread_flag(), ...  */
0008 #include <linux/sched/task_stack.h> /* task_stack_*(), ...      */
0009 #include <linux/kdebug.h>       /* oops_begin/end, ...      */
0010 #include <linux/extable.h>      /* search_exception_tables  */
0011 #include <linux/memblock.h>     /* max_low_pfn          */
0012 #include <linux/kfence.h>       /* kfence_handle_page_fault */
0013 #include <linux/kprobes.h>      /* NOKPROBE_SYMBOL, ...     */
0014 #include <linux/mmiotrace.h>        /* kmmio_handler, ...       */
0015 #include <linux/perf_event.h>       /* perf_sw_event        */
0016 #include <linux/hugetlb.h>      /* hstate_index_to_shift    */
0017 #include <linux/prefetch.h>     /* prefetchw            */
0018 #include <linux/context_tracking.h> /* exception_enter(), ...   */
0019 #include <linux/uaccess.h>      /* faulthandler_disabled()  */
0020 #include <linux/efi.h>          /* efi_crash_gracefully_on_page_fault()*/
0021 #include <linux/mm_types.h>
0022 
0023 #include <asm/cpufeature.h>     /* boot_cpu_has, ...        */
0024 #include <asm/traps.h>          /* dotraplinkage, ...       */
0025 #include <asm/fixmap.h>         /* VSYSCALL_ADDR        */
0026 #include <asm/vsyscall.h>       /* emulate_vsyscall     */
0027 #include <asm/vm86.h>           /* struct vm86          */
0028 #include <asm/mmu_context.h>        /* vma_pkey()           */
0029 #include <asm/efi.h>            /* efi_crash_gracefully_on_page_fault()*/
0030 #include <asm/desc.h>           /* store_idt(), ...     */
0031 #include <asm/cpu_entry_area.h>     /* exception stack      */
0032 #include <asm/pgtable_areas.h>      /* VMALLOC_START, ...       */
0033 #include <asm/kvm_para.h>       /* kvm_handle_async_pf      */
0034 #include <asm/vdso.h>           /* fixup_vdso_exception()   */
0035 #include <asm/irq_stack.h>
0036 
0037 #define CREATE_TRACE_POINTS
0038 #include <asm/trace/exceptions.h>
0039 
0040 /*
0041  * Returns 0 if mmiotrace is disabled, or if the fault is not
0042  * handled by mmiotrace:
0043  */
0044 static nokprobe_inline int
0045 kmmio_fault(struct pt_regs *regs, unsigned long addr)
0046 {
0047     if (unlikely(is_kmmio_active()))
0048         if (kmmio_handler(regs, addr) == 1)
0049             return -1;
0050     return 0;
0051 }
0052 
0053 /*
0054  * Prefetch quirks:
0055  *
0056  * 32-bit mode:
0057  *
0058  *   Sometimes AMD Athlon/Opteron CPUs report invalid exceptions on prefetch.
0059  *   Check that here and ignore it.  This is AMD erratum #91.
0060  *
0061  * 64-bit mode:
0062  *
0063  *   Sometimes the CPU reports invalid exceptions on prefetch.
0064  *   Check that here and ignore it.
0065  *
0066  * Opcode checker based on code by Richard Brunner.
0067  */
0068 static inline int
0069 check_prefetch_opcode(struct pt_regs *regs, unsigned char *instr,
0070               unsigned char opcode, int *prefetch)
0071 {
0072     unsigned char instr_hi = opcode & 0xf0;
0073     unsigned char instr_lo = opcode & 0x0f;
0074 
0075     switch (instr_hi) {
0076     case 0x20:
0077     case 0x30:
0078         /*
0079          * Values 0x26,0x2E,0x36,0x3E are valid x86 prefixes.
0080          * In X86_64 long mode, the CPU will signal invalid
0081          * opcode if some of these prefixes are present so
0082          * X86_64 will never get here anyway
0083          */
0084         return ((instr_lo & 7) == 0x6);
0085 #ifdef CONFIG_X86_64
0086     case 0x40:
0087         /*
0088          * In 64-bit mode 0x40..0x4F are valid REX prefixes
0089          */
0090         return (!user_mode(regs) || user_64bit_mode(regs));
0091 #endif
0092     case 0x60:
0093         /* 0x64 thru 0x67 are valid prefixes in all modes. */
0094         return (instr_lo & 0xC) == 0x4;
0095     case 0xF0:
0096         /* 0xF0, 0xF2, 0xF3 are valid prefixes in all modes. */
0097         return !instr_lo || (instr_lo>>1) == 1;
0098     case 0x00:
0099         /* Prefetch instruction is 0x0F0D or 0x0F18 */
0100         if (get_kernel_nofault(opcode, instr))
0101             return 0;
0102 
0103         *prefetch = (instr_lo == 0xF) &&
0104             (opcode == 0x0D || opcode == 0x18);
0105         return 0;
0106     default:
0107         return 0;
0108     }
0109 }
0110 
0111 static bool is_amd_k8_pre_npt(void)
0112 {
0113     struct cpuinfo_x86 *c = &boot_cpu_data;
0114 
0115     return unlikely(IS_ENABLED(CONFIG_CPU_SUP_AMD) &&
0116             c->x86_vendor == X86_VENDOR_AMD &&
0117             c->x86 == 0xf && c->x86_model < 0x40);
0118 }
0119 
0120 static int
0121 is_prefetch(struct pt_regs *regs, unsigned long error_code, unsigned long addr)
0122 {
0123     unsigned char *max_instr;
0124     unsigned char *instr;
0125     int prefetch = 0;
0126 
0127     /* Erratum #91 affects AMD K8, pre-NPT CPUs */
0128     if (!is_amd_k8_pre_npt())
0129         return 0;
0130 
0131     /*
0132      * If it was a exec (instruction fetch) fault on NX page, then
0133      * do not ignore the fault:
0134      */
0135     if (error_code & X86_PF_INSTR)
0136         return 0;
0137 
0138     instr = (void *)convert_ip_to_linear(current, regs);
0139     max_instr = instr + 15;
0140 
0141     /*
0142      * This code has historically always bailed out if IP points to a
0143      * not-present page (e.g. due to a race).  No one has ever
0144      * complained about this.
0145      */
0146     pagefault_disable();
0147 
0148     while (instr < max_instr) {
0149         unsigned char opcode;
0150 
0151         if (user_mode(regs)) {
0152             if (get_user(opcode, (unsigned char __user *) instr))
0153                 break;
0154         } else {
0155             if (get_kernel_nofault(opcode, instr))
0156                 break;
0157         }
0158 
0159         instr++;
0160 
0161         if (!check_prefetch_opcode(regs, instr, opcode, &prefetch))
0162             break;
0163     }
0164 
0165     pagefault_enable();
0166     return prefetch;
0167 }
0168 
0169 DEFINE_SPINLOCK(pgd_lock);
0170 LIST_HEAD(pgd_list);
0171 
0172 #ifdef CONFIG_X86_32
0173 static inline pmd_t *vmalloc_sync_one(pgd_t *pgd, unsigned long address)
0174 {
0175     unsigned index = pgd_index(address);
0176     pgd_t *pgd_k;
0177     p4d_t *p4d, *p4d_k;
0178     pud_t *pud, *pud_k;
0179     pmd_t *pmd, *pmd_k;
0180 
0181     pgd += index;
0182     pgd_k = init_mm.pgd + index;
0183 
0184     if (!pgd_present(*pgd_k))
0185         return NULL;
0186 
0187     /*
0188      * set_pgd(pgd, *pgd_k); here would be useless on PAE
0189      * and redundant with the set_pmd() on non-PAE. As would
0190      * set_p4d/set_pud.
0191      */
0192     p4d = p4d_offset(pgd, address);
0193     p4d_k = p4d_offset(pgd_k, address);
0194     if (!p4d_present(*p4d_k))
0195         return NULL;
0196 
0197     pud = pud_offset(p4d, address);
0198     pud_k = pud_offset(p4d_k, address);
0199     if (!pud_present(*pud_k))
0200         return NULL;
0201 
0202     pmd = pmd_offset(pud, address);
0203     pmd_k = pmd_offset(pud_k, address);
0204 
0205     if (pmd_present(*pmd) != pmd_present(*pmd_k))
0206         set_pmd(pmd, *pmd_k);
0207 
0208     if (!pmd_present(*pmd_k))
0209         return NULL;
0210     else
0211         BUG_ON(pmd_pfn(*pmd) != pmd_pfn(*pmd_k));
0212 
0213     return pmd_k;
0214 }
0215 
0216 /*
0217  *   Handle a fault on the vmalloc or module mapping area
0218  *
0219  *   This is needed because there is a race condition between the time
0220  *   when the vmalloc mapping code updates the PMD to the point in time
0221  *   where it synchronizes this update with the other page-tables in the
0222  *   system.
0223  *
0224  *   In this race window another thread/CPU can map an area on the same
0225  *   PMD, finds it already present and does not synchronize it with the
0226  *   rest of the system yet. As a result v[mz]alloc might return areas
0227  *   which are not mapped in every page-table in the system, causing an
0228  *   unhandled page-fault when they are accessed.
0229  */
0230 static noinline int vmalloc_fault(unsigned long address)
0231 {
0232     unsigned long pgd_paddr;
0233     pmd_t *pmd_k;
0234     pte_t *pte_k;
0235 
0236     /* Make sure we are in vmalloc area: */
0237     if (!(address >= VMALLOC_START && address < VMALLOC_END))
0238         return -1;
0239 
0240     /*
0241      * Synchronize this task's top level page-table
0242      * with the 'reference' page table.
0243      *
0244      * Do _not_ use "current" here. We might be inside
0245      * an interrupt in the middle of a task switch..
0246      */
0247     pgd_paddr = read_cr3_pa();
0248     pmd_k = vmalloc_sync_one(__va(pgd_paddr), address);
0249     if (!pmd_k)
0250         return -1;
0251 
0252     if (pmd_large(*pmd_k))
0253         return 0;
0254 
0255     pte_k = pte_offset_kernel(pmd_k, address);
0256     if (!pte_present(*pte_k))
0257         return -1;
0258 
0259     return 0;
0260 }
0261 NOKPROBE_SYMBOL(vmalloc_fault);
0262 
0263 void arch_sync_kernel_mappings(unsigned long start, unsigned long end)
0264 {
0265     unsigned long addr;
0266 
0267     for (addr = start & PMD_MASK;
0268          addr >= TASK_SIZE_MAX && addr < VMALLOC_END;
0269          addr += PMD_SIZE) {
0270         struct page *page;
0271 
0272         spin_lock(&pgd_lock);
0273         list_for_each_entry(page, &pgd_list, lru) {
0274             spinlock_t *pgt_lock;
0275 
0276             /* the pgt_lock only for Xen */
0277             pgt_lock = &pgd_page_get_mm(page)->page_table_lock;
0278 
0279             spin_lock(pgt_lock);
0280             vmalloc_sync_one(page_address(page), addr);
0281             spin_unlock(pgt_lock);
0282         }
0283         spin_unlock(&pgd_lock);
0284     }
0285 }
0286 
0287 static bool low_pfn(unsigned long pfn)
0288 {
0289     return pfn < max_low_pfn;
0290 }
0291 
0292 static void dump_pagetable(unsigned long address)
0293 {
0294     pgd_t *base = __va(read_cr3_pa());
0295     pgd_t *pgd = &base[pgd_index(address)];
0296     p4d_t *p4d;
0297     pud_t *pud;
0298     pmd_t *pmd;
0299     pte_t *pte;
0300 
0301 #ifdef CONFIG_X86_PAE
0302     pr_info("*pdpt = %016Lx ", pgd_val(*pgd));
0303     if (!low_pfn(pgd_val(*pgd) >> PAGE_SHIFT) || !pgd_present(*pgd))
0304         goto out;
0305 #define pr_pde pr_cont
0306 #else
0307 #define pr_pde pr_info
0308 #endif
0309     p4d = p4d_offset(pgd, address);
0310     pud = pud_offset(p4d, address);
0311     pmd = pmd_offset(pud, address);
0312     pr_pde("*pde = %0*Lx ", sizeof(*pmd) * 2, (u64)pmd_val(*pmd));
0313 #undef pr_pde
0314 
0315     /*
0316      * We must not directly access the pte in the highpte
0317      * case if the page table is located in highmem.
0318      * And let's rather not kmap-atomic the pte, just in case
0319      * it's allocated already:
0320      */
0321     if (!low_pfn(pmd_pfn(*pmd)) || !pmd_present(*pmd) || pmd_large(*pmd))
0322         goto out;
0323 
0324     pte = pte_offset_kernel(pmd, address);
0325     pr_cont("*pte = %0*Lx ", sizeof(*pte) * 2, (u64)pte_val(*pte));
0326 out:
0327     pr_cont("\n");
0328 }
0329 
0330 #else /* CONFIG_X86_64: */
0331 
0332 #ifdef CONFIG_CPU_SUP_AMD
0333 static const char errata93_warning[] =
0334 KERN_ERR 
0335 "******* Your BIOS seems to not contain a fix for K8 errata #93\n"
0336 "******* Working around it, but it may cause SEGVs or burn power.\n"
0337 "******* Please consider a BIOS update.\n"
0338 "******* Disabling USB legacy in the BIOS may also help.\n";
0339 #endif
0340 
0341 static int bad_address(void *p)
0342 {
0343     unsigned long dummy;
0344 
0345     return get_kernel_nofault(dummy, (unsigned long *)p);
0346 }
0347 
0348 static void dump_pagetable(unsigned long address)
0349 {
0350     pgd_t *base = __va(read_cr3_pa());
0351     pgd_t *pgd = base + pgd_index(address);
0352     p4d_t *p4d;
0353     pud_t *pud;
0354     pmd_t *pmd;
0355     pte_t *pte;
0356 
0357     if (bad_address(pgd))
0358         goto bad;
0359 
0360     pr_info("PGD %lx ", pgd_val(*pgd));
0361 
0362     if (!pgd_present(*pgd))
0363         goto out;
0364 
0365     p4d = p4d_offset(pgd, address);
0366     if (bad_address(p4d))
0367         goto bad;
0368 
0369     pr_cont("P4D %lx ", p4d_val(*p4d));
0370     if (!p4d_present(*p4d) || p4d_large(*p4d))
0371         goto out;
0372 
0373     pud = pud_offset(p4d, address);
0374     if (bad_address(pud))
0375         goto bad;
0376 
0377     pr_cont("PUD %lx ", pud_val(*pud));
0378     if (!pud_present(*pud) || pud_large(*pud))
0379         goto out;
0380 
0381     pmd = pmd_offset(pud, address);
0382     if (bad_address(pmd))
0383         goto bad;
0384 
0385     pr_cont("PMD %lx ", pmd_val(*pmd));
0386     if (!pmd_present(*pmd) || pmd_large(*pmd))
0387         goto out;
0388 
0389     pte = pte_offset_kernel(pmd, address);
0390     if (bad_address(pte))
0391         goto bad;
0392 
0393     pr_cont("PTE %lx", pte_val(*pte));
0394 out:
0395     pr_cont("\n");
0396     return;
0397 bad:
0398     pr_info("BAD\n");
0399 }
0400 
0401 #endif /* CONFIG_X86_64 */
0402 
0403 /*
0404  * Workaround for K8 erratum #93 & buggy BIOS.
0405  *
0406  * BIOS SMM functions are required to use a specific workaround
0407  * to avoid corruption of the 64bit RIP register on C stepping K8.
0408  *
0409  * A lot of BIOS that didn't get tested properly miss this.
0410  *
0411  * The OS sees this as a page fault with the upper 32bits of RIP cleared.
0412  * Try to work around it here.
0413  *
0414  * Note we only handle faults in kernel here.
0415  * Does nothing on 32-bit.
0416  */
0417 static int is_errata93(struct pt_regs *regs, unsigned long address)
0418 {
0419 #if defined(CONFIG_X86_64) && defined(CONFIG_CPU_SUP_AMD)
0420     if (boot_cpu_data.x86_vendor != X86_VENDOR_AMD
0421         || boot_cpu_data.x86 != 0xf)
0422         return 0;
0423 
0424     if (user_mode(regs))
0425         return 0;
0426 
0427     if (address != regs->ip)
0428         return 0;
0429 
0430     if ((address >> 32) != 0)
0431         return 0;
0432 
0433     address |= 0xffffffffUL << 32;
0434     if ((address >= (u64)_stext && address <= (u64)_etext) ||
0435         (address >= MODULES_VADDR && address <= MODULES_END)) {
0436         printk_once(errata93_warning);
0437         regs->ip = address;
0438         return 1;
0439     }
0440 #endif
0441     return 0;
0442 }
0443 
0444 /*
0445  * Work around K8 erratum #100 K8 in compat mode occasionally jumps
0446  * to illegal addresses >4GB.
0447  *
0448  * We catch this in the page fault handler because these addresses
0449  * are not reachable. Just detect this case and return.  Any code
0450  * segment in LDT is compatibility mode.
0451  */
0452 static int is_errata100(struct pt_regs *regs, unsigned long address)
0453 {
0454 #ifdef CONFIG_X86_64
0455     if ((regs->cs == __USER32_CS || (regs->cs & (1<<2))) && (address >> 32))
0456         return 1;
0457 #endif
0458     return 0;
0459 }
0460 
0461 /* Pentium F0 0F C7 C8 bug workaround: */
0462 static int is_f00f_bug(struct pt_regs *regs, unsigned long error_code,
0463                unsigned long address)
0464 {
0465 #ifdef CONFIG_X86_F00F_BUG
0466     if (boot_cpu_has_bug(X86_BUG_F00F) && !(error_code & X86_PF_USER) &&
0467         idt_is_f00f_address(address)) {
0468         handle_invalid_op(regs);
0469         return 1;
0470     }
0471 #endif
0472     return 0;
0473 }
0474 
0475 static void show_ldttss(const struct desc_ptr *gdt, const char *name, u16 index)
0476 {
0477     u32 offset = (index >> 3) * sizeof(struct desc_struct);
0478     unsigned long addr;
0479     struct ldttss_desc desc;
0480 
0481     if (index == 0) {
0482         pr_alert("%s: NULL\n", name);
0483         return;
0484     }
0485 
0486     if (offset + sizeof(struct ldttss_desc) >= gdt->size) {
0487         pr_alert("%s: 0x%hx -- out of bounds\n", name, index);
0488         return;
0489     }
0490 
0491     if (copy_from_kernel_nofault(&desc, (void *)(gdt->address + offset),
0492                   sizeof(struct ldttss_desc))) {
0493         pr_alert("%s: 0x%hx -- GDT entry is not readable\n",
0494              name, index);
0495         return;
0496     }
0497 
0498     addr = desc.base0 | (desc.base1 << 16) | ((unsigned long)desc.base2 << 24);
0499 #ifdef CONFIG_X86_64
0500     addr |= ((u64)desc.base3 << 32);
0501 #endif
0502     pr_alert("%s: 0x%hx -- base=0x%lx limit=0x%x\n",
0503          name, index, addr, (desc.limit0 | (desc.limit1 << 16)));
0504 }
0505 
0506 static void
0507 show_fault_oops(struct pt_regs *regs, unsigned long error_code, unsigned long address)
0508 {
0509     if (!oops_may_print())
0510         return;
0511 
0512     if (error_code & X86_PF_INSTR) {
0513         unsigned int level;
0514         pgd_t *pgd;
0515         pte_t *pte;
0516 
0517         pgd = __va(read_cr3_pa());
0518         pgd += pgd_index(address);
0519 
0520         pte = lookup_address_in_pgd(pgd, address, &level);
0521 
0522         if (pte && pte_present(*pte) && !pte_exec(*pte))
0523             pr_crit("kernel tried to execute NX-protected page - exploit attempt? (uid: %d)\n",
0524                 from_kuid(&init_user_ns, current_uid()));
0525         if (pte && pte_present(*pte) && pte_exec(*pte) &&
0526                 (pgd_flags(*pgd) & _PAGE_USER) &&
0527                 (__read_cr4() & X86_CR4_SMEP))
0528             pr_crit("unable to execute userspace code (SMEP?) (uid: %d)\n",
0529                 from_kuid(&init_user_ns, current_uid()));
0530     }
0531 
0532     if (address < PAGE_SIZE && !user_mode(regs))
0533         pr_alert("BUG: kernel NULL pointer dereference, address: %px\n",
0534             (void *)address);
0535     else
0536         pr_alert("BUG: unable to handle page fault for address: %px\n",
0537             (void *)address);
0538 
0539     pr_alert("#PF: %s %s in %s mode\n",
0540          (error_code & X86_PF_USER)  ? "user" : "supervisor",
0541          (error_code & X86_PF_INSTR) ? "instruction fetch" :
0542          (error_code & X86_PF_WRITE) ? "write access" :
0543                            "read access",
0544                  user_mode(regs) ? "user" : "kernel");
0545     pr_alert("#PF: error_code(0x%04lx) - %s\n", error_code,
0546          !(error_code & X86_PF_PROT) ? "not-present page" :
0547          (error_code & X86_PF_RSVD)  ? "reserved bit violation" :
0548          (error_code & X86_PF_PK)    ? "protection keys violation" :
0549                            "permissions violation");
0550 
0551     if (!(error_code & X86_PF_USER) && user_mode(regs)) {
0552         struct desc_ptr idt, gdt;
0553         u16 ldtr, tr;
0554 
0555         /*
0556          * This can happen for quite a few reasons.  The more obvious
0557          * ones are faults accessing the GDT, or LDT.  Perhaps
0558          * surprisingly, if the CPU tries to deliver a benign or
0559          * contributory exception from user code and gets a page fault
0560          * during delivery, the page fault can be delivered as though
0561          * it originated directly from user code.  This could happen
0562          * due to wrong permissions on the IDT, GDT, LDT, TSS, or
0563          * kernel or IST stack.
0564          */
0565         store_idt(&idt);
0566 
0567         /* Usable even on Xen PV -- it's just slow. */
0568         native_store_gdt(&gdt);
0569 
0570         pr_alert("IDT: 0x%lx (limit=0x%hx) GDT: 0x%lx (limit=0x%hx)\n",
0571              idt.address, idt.size, gdt.address, gdt.size);
0572 
0573         store_ldt(ldtr);
0574         show_ldttss(&gdt, "LDTR", ldtr);
0575 
0576         store_tr(tr);
0577         show_ldttss(&gdt, "TR", tr);
0578     }
0579 
0580     dump_pagetable(address);
0581 }
0582 
0583 static noinline void
0584 pgtable_bad(struct pt_regs *regs, unsigned long error_code,
0585         unsigned long address)
0586 {
0587     struct task_struct *tsk;
0588     unsigned long flags;
0589     int sig;
0590 
0591     flags = oops_begin();
0592     tsk = current;
0593     sig = SIGKILL;
0594 
0595     printk(KERN_ALERT "%s: Corrupted page table at address %lx\n",
0596            tsk->comm, address);
0597     dump_pagetable(address);
0598 
0599     if (__die("Bad pagetable", regs, error_code))
0600         sig = 0;
0601 
0602     oops_end(flags, regs, sig);
0603 }
0604 
0605 static void sanitize_error_code(unsigned long address,
0606                 unsigned long *error_code)
0607 {
0608     /*
0609      * To avoid leaking information about the kernel page
0610      * table layout, pretend that user-mode accesses to
0611      * kernel addresses are always protection faults.
0612      *
0613      * NB: This means that failed vsyscalls with vsyscall=none
0614      * will have the PROT bit.  This doesn't leak any
0615      * information and does not appear to cause any problems.
0616      */
0617     if (address >= TASK_SIZE_MAX)
0618         *error_code |= X86_PF_PROT;
0619 }
0620 
0621 static void set_signal_archinfo(unsigned long address,
0622                 unsigned long error_code)
0623 {
0624     struct task_struct *tsk = current;
0625 
0626     tsk->thread.trap_nr = X86_TRAP_PF;
0627     tsk->thread.error_code = error_code | X86_PF_USER;
0628     tsk->thread.cr2 = address;
0629 }
0630 
0631 static noinline void
0632 page_fault_oops(struct pt_regs *regs, unsigned long error_code,
0633         unsigned long address)
0634 {
0635 #ifdef CONFIG_VMAP_STACK
0636     struct stack_info info;
0637 #endif
0638     unsigned long flags;
0639     int sig;
0640 
0641     if (user_mode(regs)) {
0642         /*
0643          * Implicit kernel access from user mode?  Skip the stack
0644          * overflow and EFI special cases.
0645          */
0646         goto oops;
0647     }
0648 
0649 #ifdef CONFIG_VMAP_STACK
0650     /*
0651      * Stack overflow?  During boot, we can fault near the initial
0652      * stack in the direct map, but that's not an overflow -- check
0653      * that we're in vmalloc space to avoid this.
0654      */
0655     if (is_vmalloc_addr((void *)address) &&
0656         get_stack_guard_info((void *)address, &info)) {
0657         /*
0658          * We're likely to be running with very little stack space
0659          * left.  It's plausible that we'd hit this condition but
0660          * double-fault even before we get this far, in which case
0661          * we're fine: the double-fault handler will deal with it.
0662          *
0663          * We don't want to make it all the way into the oops code
0664          * and then double-fault, though, because we're likely to
0665          * break the console driver and lose most of the stack dump.
0666          */
0667         call_on_stack(__this_cpu_ist_top_va(DF) - sizeof(void*),
0668                   handle_stack_overflow,
0669                   ASM_CALL_ARG3,
0670                   , [arg1] "r" (regs), [arg2] "r" (address), [arg3] "r" (&info));
0671 
0672         unreachable();
0673     }
0674 #endif
0675 
0676     /*
0677      * Buggy firmware could access regions which might page fault.  If
0678      * this happens, EFI has a special OOPS path that will try to
0679      * avoid hanging the system.
0680      */
0681     if (IS_ENABLED(CONFIG_EFI))
0682         efi_crash_gracefully_on_page_fault(address);
0683 
0684     /* Only not-present faults should be handled by KFENCE. */
0685     if (!(error_code & X86_PF_PROT) &&
0686         kfence_handle_page_fault(address, error_code & X86_PF_WRITE, regs))
0687         return;
0688 
0689 oops:
0690     /*
0691      * Oops. The kernel tried to access some bad page. We'll have to
0692      * terminate things with extreme prejudice:
0693      */
0694     flags = oops_begin();
0695 
0696     show_fault_oops(regs, error_code, address);
0697 
0698     if (task_stack_end_corrupted(current))
0699         printk(KERN_EMERG "Thread overran stack, or stack corrupted\n");
0700 
0701     sig = SIGKILL;
0702     if (__die("Oops", regs, error_code))
0703         sig = 0;
0704 
0705     /* Executive summary in case the body of the oops scrolled away */
0706     printk(KERN_DEFAULT "CR2: %016lx\n", address);
0707 
0708     oops_end(flags, regs, sig);
0709 }
0710 
0711 static noinline void
0712 kernelmode_fixup_or_oops(struct pt_regs *regs, unsigned long error_code,
0713              unsigned long address, int signal, int si_code,
0714              u32 pkey)
0715 {
0716     WARN_ON_ONCE(user_mode(regs));
0717 
0718     /* Are we prepared to handle this kernel fault? */
0719     if (fixup_exception(regs, X86_TRAP_PF, error_code, address)) {
0720         /*
0721          * Any interrupt that takes a fault gets the fixup. This makes
0722          * the below recursive fault logic only apply to a faults from
0723          * task context.
0724          */
0725         if (in_interrupt())
0726             return;
0727 
0728         /*
0729          * Per the above we're !in_interrupt(), aka. task context.
0730          *
0731          * In this case we need to make sure we're not recursively
0732          * faulting through the emulate_vsyscall() logic.
0733          */
0734         if (current->thread.sig_on_uaccess_err && signal) {
0735             sanitize_error_code(address, &error_code);
0736 
0737             set_signal_archinfo(address, error_code);
0738 
0739             if (si_code == SEGV_PKUERR) {
0740                 force_sig_pkuerr((void __user *)address, pkey);
0741             } else {
0742                 /* XXX: hwpoison faults will set the wrong code. */
0743                 force_sig_fault(signal, si_code, (void __user *)address);
0744             }
0745         }
0746 
0747         /*
0748          * Barring that, we can do the fixup and be happy.
0749          */
0750         return;
0751     }
0752 
0753     /*
0754      * AMD erratum #91 manifests as a spurious page fault on a PREFETCH
0755      * instruction.
0756      */
0757     if (is_prefetch(regs, error_code, address))
0758         return;
0759 
0760     page_fault_oops(regs, error_code, address);
0761 }
0762 
0763 /*
0764  * Print out info about fatal segfaults, if the show_unhandled_signals
0765  * sysctl is set:
0766  */
0767 static inline void
0768 show_signal_msg(struct pt_regs *regs, unsigned long error_code,
0769         unsigned long address, struct task_struct *tsk)
0770 {
0771     const char *loglvl = task_pid_nr(tsk) > 1 ? KERN_INFO : KERN_EMERG;
0772 
0773     if (!unhandled_signal(tsk, SIGSEGV))
0774         return;
0775 
0776     if (!printk_ratelimit())
0777         return;
0778 
0779     printk("%s%s[%d]: segfault at %lx ip %px sp %px error %lx",
0780         loglvl, tsk->comm, task_pid_nr(tsk), address,
0781         (void *)regs->ip, (void *)regs->sp, error_code);
0782 
0783     print_vma_addr(KERN_CONT " in ", regs->ip);
0784 
0785     printk(KERN_CONT "\n");
0786 
0787     show_opcodes(regs, loglvl);
0788 }
0789 
0790 /*
0791  * The (legacy) vsyscall page is the long page in the kernel portion
0792  * of the address space that has user-accessible permissions.
0793  */
0794 static bool is_vsyscall_vaddr(unsigned long vaddr)
0795 {
0796     return unlikely((vaddr & PAGE_MASK) == VSYSCALL_ADDR);
0797 }
0798 
0799 static void
0800 __bad_area_nosemaphore(struct pt_regs *regs, unsigned long error_code,
0801                unsigned long address, u32 pkey, int si_code)
0802 {
0803     struct task_struct *tsk = current;
0804 
0805     if (!user_mode(regs)) {
0806         kernelmode_fixup_or_oops(regs, error_code, address,
0807                      SIGSEGV, si_code, pkey);
0808         return;
0809     }
0810 
0811     if (!(error_code & X86_PF_USER)) {
0812         /* Implicit user access to kernel memory -- just oops */
0813         page_fault_oops(regs, error_code, address);
0814         return;
0815     }
0816 
0817     /*
0818      * User mode accesses just cause a SIGSEGV.
0819      * It's possible to have interrupts off here:
0820      */
0821     local_irq_enable();
0822 
0823     /*
0824      * Valid to do another page fault here because this one came
0825      * from user space:
0826      */
0827     if (is_prefetch(regs, error_code, address))
0828         return;
0829 
0830     if (is_errata100(regs, address))
0831         return;
0832 
0833     sanitize_error_code(address, &error_code);
0834 
0835     if (fixup_vdso_exception(regs, X86_TRAP_PF, error_code, address))
0836         return;
0837 
0838     if (likely(show_unhandled_signals))
0839         show_signal_msg(regs, error_code, address, tsk);
0840 
0841     set_signal_archinfo(address, error_code);
0842 
0843     if (si_code == SEGV_PKUERR)
0844         force_sig_pkuerr((void __user *)address, pkey);
0845     else
0846         force_sig_fault(SIGSEGV, si_code, (void __user *)address);
0847 
0848     local_irq_disable();
0849 }
0850 
0851 static noinline void
0852 bad_area_nosemaphore(struct pt_regs *regs, unsigned long error_code,
0853              unsigned long address)
0854 {
0855     __bad_area_nosemaphore(regs, error_code, address, 0, SEGV_MAPERR);
0856 }
0857 
0858 static void
0859 __bad_area(struct pt_regs *regs, unsigned long error_code,
0860        unsigned long address, u32 pkey, int si_code)
0861 {
0862     struct mm_struct *mm = current->mm;
0863     /*
0864      * Something tried to access memory that isn't in our memory map..
0865      * Fix it, but check if it's kernel or user first..
0866      */
0867     mmap_read_unlock(mm);
0868 
0869     __bad_area_nosemaphore(regs, error_code, address, pkey, si_code);
0870 }
0871 
0872 static noinline void
0873 bad_area(struct pt_regs *regs, unsigned long error_code, unsigned long address)
0874 {
0875     __bad_area(regs, error_code, address, 0, SEGV_MAPERR);
0876 }
0877 
0878 static inline bool bad_area_access_from_pkeys(unsigned long error_code,
0879         struct vm_area_struct *vma)
0880 {
0881     /* This code is always called on the current mm */
0882     bool foreign = false;
0883 
0884     if (!cpu_feature_enabled(X86_FEATURE_OSPKE))
0885         return false;
0886     if (error_code & X86_PF_PK)
0887         return true;
0888     /* this checks permission keys on the VMA: */
0889     if (!arch_vma_access_permitted(vma, (error_code & X86_PF_WRITE),
0890                        (error_code & X86_PF_INSTR), foreign))
0891         return true;
0892     return false;
0893 }
0894 
0895 static noinline void
0896 bad_area_access_error(struct pt_regs *regs, unsigned long error_code,
0897               unsigned long address, struct vm_area_struct *vma)
0898 {
0899     /*
0900      * This OSPKE check is not strictly necessary at runtime.
0901      * But, doing it this way allows compiler optimizations
0902      * if pkeys are compiled out.
0903      */
0904     if (bad_area_access_from_pkeys(error_code, vma)) {
0905         /*
0906          * A protection key fault means that the PKRU value did not allow
0907          * access to some PTE.  Userspace can figure out what PKRU was
0908          * from the XSAVE state.  This function captures the pkey from
0909          * the vma and passes it to userspace so userspace can discover
0910          * which protection key was set on the PTE.
0911          *
0912          * If we get here, we know that the hardware signaled a X86_PF_PK
0913          * fault and that there was a VMA once we got in the fault
0914          * handler.  It does *not* guarantee that the VMA we find here
0915          * was the one that we faulted on.
0916          *
0917          * 1. T1   : mprotect_key(foo, PAGE_SIZE, pkey=4);
0918          * 2. T1   : set PKRU to deny access to pkey=4, touches page
0919          * 3. T1   : faults...
0920          * 4.    T2: mprotect_key(foo, PAGE_SIZE, pkey=5);
0921          * 5. T1   : enters fault handler, takes mmap_lock, etc...
0922          * 6. T1   : reaches here, sees vma_pkey(vma)=5, when we really
0923          *       faulted on a pte with its pkey=4.
0924          */
0925         u32 pkey = vma_pkey(vma);
0926 
0927         __bad_area(regs, error_code, address, pkey, SEGV_PKUERR);
0928     } else {
0929         __bad_area(regs, error_code, address, 0, SEGV_ACCERR);
0930     }
0931 }
0932 
0933 static void
0934 do_sigbus(struct pt_regs *regs, unsigned long error_code, unsigned long address,
0935       vm_fault_t fault)
0936 {
0937     /* Kernel mode? Handle exceptions or die: */
0938     if (!user_mode(regs)) {
0939         kernelmode_fixup_or_oops(regs, error_code, address,
0940                      SIGBUS, BUS_ADRERR, ARCH_DEFAULT_PKEY);
0941         return;
0942     }
0943 
0944     /* User-space => ok to do another page fault: */
0945     if (is_prefetch(regs, error_code, address))
0946         return;
0947 
0948     sanitize_error_code(address, &error_code);
0949 
0950     if (fixup_vdso_exception(regs, X86_TRAP_PF, error_code, address))
0951         return;
0952 
0953     set_signal_archinfo(address, error_code);
0954 
0955 #ifdef CONFIG_MEMORY_FAILURE
0956     if (fault & (VM_FAULT_HWPOISON|VM_FAULT_HWPOISON_LARGE)) {
0957         struct task_struct *tsk = current;
0958         unsigned lsb = 0;
0959 
0960         pr_err(
0961     "MCE: Killing %s:%d due to hardware memory corruption fault at %lx\n",
0962             tsk->comm, tsk->pid, address);
0963         if (fault & VM_FAULT_HWPOISON_LARGE)
0964             lsb = hstate_index_to_shift(VM_FAULT_GET_HINDEX(fault));
0965         if (fault & VM_FAULT_HWPOISON)
0966             lsb = PAGE_SHIFT;
0967         force_sig_mceerr(BUS_MCEERR_AR, (void __user *)address, lsb);
0968         return;
0969     }
0970 #endif
0971     force_sig_fault(SIGBUS, BUS_ADRERR, (void __user *)address);
0972 }
0973 
0974 static int spurious_kernel_fault_check(unsigned long error_code, pte_t *pte)
0975 {
0976     if ((error_code & X86_PF_WRITE) && !pte_write(*pte))
0977         return 0;
0978 
0979     if ((error_code & X86_PF_INSTR) && !pte_exec(*pte))
0980         return 0;
0981 
0982     return 1;
0983 }
0984 
0985 /*
0986  * Handle a spurious fault caused by a stale TLB entry.
0987  *
0988  * This allows us to lazily refresh the TLB when increasing the
0989  * permissions of a kernel page (RO -> RW or NX -> X).  Doing it
0990  * eagerly is very expensive since that implies doing a full
0991  * cross-processor TLB flush, even if no stale TLB entries exist
0992  * on other processors.
0993  *
0994  * Spurious faults may only occur if the TLB contains an entry with
0995  * fewer permission than the page table entry.  Non-present (P = 0)
0996  * and reserved bit (R = 1) faults are never spurious.
0997  *
0998  * There are no security implications to leaving a stale TLB when
0999  * increasing the permissions on a page.
1000  *
1001  * Returns non-zero if a spurious fault was handled, zero otherwise.
1002  *
1003  * See Intel Developer's Manual Vol 3 Section 4.10.4.3, bullet 3
1004  * (Optional Invalidation).
1005  */
1006 static noinline int
1007 spurious_kernel_fault(unsigned long error_code, unsigned long address)
1008 {
1009     pgd_t *pgd;
1010     p4d_t *p4d;
1011     pud_t *pud;
1012     pmd_t *pmd;
1013     pte_t *pte;
1014     int ret;
1015 
1016     /*
1017      * Only writes to RO or instruction fetches from NX may cause
1018      * spurious faults.
1019      *
1020      * These could be from user or supervisor accesses but the TLB
1021      * is only lazily flushed after a kernel mapping protection
1022      * change, so user accesses are not expected to cause spurious
1023      * faults.
1024      */
1025     if (error_code != (X86_PF_WRITE | X86_PF_PROT) &&
1026         error_code != (X86_PF_INSTR | X86_PF_PROT))
1027         return 0;
1028 
1029     pgd = init_mm.pgd + pgd_index(address);
1030     if (!pgd_present(*pgd))
1031         return 0;
1032 
1033     p4d = p4d_offset(pgd, address);
1034     if (!p4d_present(*p4d))
1035         return 0;
1036 
1037     if (p4d_large(*p4d))
1038         return spurious_kernel_fault_check(error_code, (pte_t *) p4d);
1039 
1040     pud = pud_offset(p4d, address);
1041     if (!pud_present(*pud))
1042         return 0;
1043 
1044     if (pud_large(*pud))
1045         return spurious_kernel_fault_check(error_code, (pte_t *) pud);
1046 
1047     pmd = pmd_offset(pud, address);
1048     if (!pmd_present(*pmd))
1049         return 0;
1050 
1051     if (pmd_large(*pmd))
1052         return spurious_kernel_fault_check(error_code, (pte_t *) pmd);
1053 
1054     pte = pte_offset_kernel(pmd, address);
1055     if (!pte_present(*pte))
1056         return 0;
1057 
1058     ret = spurious_kernel_fault_check(error_code, pte);
1059     if (!ret)
1060         return 0;
1061 
1062     /*
1063      * Make sure we have permissions in PMD.
1064      * If not, then there's a bug in the page tables:
1065      */
1066     ret = spurious_kernel_fault_check(error_code, (pte_t *) pmd);
1067     WARN_ONCE(!ret, "PMD has incorrect permission bits\n");
1068 
1069     return ret;
1070 }
1071 NOKPROBE_SYMBOL(spurious_kernel_fault);
1072 
1073 int show_unhandled_signals = 1;
1074 
1075 static inline int
1076 access_error(unsigned long error_code, struct vm_area_struct *vma)
1077 {
1078     /* This is only called for the current mm, so: */
1079     bool foreign = false;
1080 
1081     /*
1082      * Read or write was blocked by protection keys.  This is
1083      * always an unconditional error and can never result in
1084      * a follow-up action to resolve the fault, like a COW.
1085      */
1086     if (error_code & X86_PF_PK)
1087         return 1;
1088 
1089     /*
1090      * SGX hardware blocked the access.  This usually happens
1091      * when the enclave memory contents have been destroyed, like
1092      * after a suspend/resume cycle. In any case, the kernel can't
1093      * fix the cause of the fault.  Handle the fault as an access
1094      * error even in cases where no actual access violation
1095      * occurred.  This allows userspace to rebuild the enclave in
1096      * response to the signal.
1097      */
1098     if (unlikely(error_code & X86_PF_SGX))
1099         return 1;
1100 
1101     /*
1102      * Make sure to check the VMA so that we do not perform
1103      * faults just to hit a X86_PF_PK as soon as we fill in a
1104      * page.
1105      */
1106     if (!arch_vma_access_permitted(vma, (error_code & X86_PF_WRITE),
1107                        (error_code & X86_PF_INSTR), foreign))
1108         return 1;
1109 
1110     if (error_code & X86_PF_WRITE) {
1111         /* write, present and write, not present: */
1112         if (unlikely(!(vma->vm_flags & VM_WRITE)))
1113             return 1;
1114         return 0;
1115     }
1116 
1117     /* read, present: */
1118     if (unlikely(error_code & X86_PF_PROT))
1119         return 1;
1120 
1121     /* read, not present: */
1122     if (unlikely(!vma_is_accessible(vma)))
1123         return 1;
1124 
1125     return 0;
1126 }
1127 
1128 bool fault_in_kernel_space(unsigned long address)
1129 {
1130     /*
1131      * On 64-bit systems, the vsyscall page is at an address above
1132      * TASK_SIZE_MAX, but is not considered part of the kernel
1133      * address space.
1134      */
1135     if (IS_ENABLED(CONFIG_X86_64) && is_vsyscall_vaddr(address))
1136         return false;
1137 
1138     return address >= TASK_SIZE_MAX;
1139 }
1140 
1141 /*
1142  * Called for all faults where 'address' is part of the kernel address
1143  * space.  Might get called for faults that originate from *code* that
1144  * ran in userspace or the kernel.
1145  */
1146 static void
1147 do_kern_addr_fault(struct pt_regs *regs, unsigned long hw_error_code,
1148            unsigned long address)
1149 {
1150     /*
1151      * Protection keys exceptions only happen on user pages.  We
1152      * have no user pages in the kernel portion of the address
1153      * space, so do not expect them here.
1154      */
1155     WARN_ON_ONCE(hw_error_code & X86_PF_PK);
1156 
1157 #ifdef CONFIG_X86_32
1158     /*
1159      * We can fault-in kernel-space virtual memory on-demand. The
1160      * 'reference' page table is init_mm.pgd.
1161      *
1162      * NOTE! We MUST NOT take any locks for this case. We may
1163      * be in an interrupt or a critical region, and should
1164      * only copy the information from the master page table,
1165      * nothing more.
1166      *
1167      * Before doing this on-demand faulting, ensure that the
1168      * fault is not any of the following:
1169      * 1. A fault on a PTE with a reserved bit set.
1170      * 2. A fault caused by a user-mode access.  (Do not demand-
1171      *    fault kernel memory due to user-mode accesses).
1172      * 3. A fault caused by a page-level protection violation.
1173      *    (A demand fault would be on a non-present page which
1174      *     would have X86_PF_PROT==0).
1175      *
1176      * This is only needed to close a race condition on x86-32 in
1177      * the vmalloc mapping/unmapping code. See the comment above
1178      * vmalloc_fault() for details. On x86-64 the race does not
1179      * exist as the vmalloc mappings don't need to be synchronized
1180      * there.
1181      */
1182     if (!(hw_error_code & (X86_PF_RSVD | X86_PF_USER | X86_PF_PROT))) {
1183         if (vmalloc_fault(address) >= 0)
1184             return;
1185     }
1186 #endif
1187 
1188     if (is_f00f_bug(regs, hw_error_code, address))
1189         return;
1190 
1191     /* Was the fault spurious, caused by lazy TLB invalidation? */
1192     if (spurious_kernel_fault(hw_error_code, address))
1193         return;
1194 
1195     /* kprobes don't want to hook the spurious faults: */
1196     if (WARN_ON_ONCE(kprobe_page_fault(regs, X86_TRAP_PF)))
1197         return;
1198 
1199     /*
1200      * Note, despite being a "bad area", there are quite a few
1201      * acceptable reasons to get here, such as erratum fixups
1202      * and handling kernel code that can fault, like get_user().
1203      *
1204      * Don't take the mm semaphore here. If we fixup a prefetch
1205      * fault we could otherwise deadlock:
1206      */
1207     bad_area_nosemaphore(regs, hw_error_code, address);
1208 }
1209 NOKPROBE_SYMBOL(do_kern_addr_fault);
1210 
1211 /*
1212  * Handle faults in the user portion of the address space.  Nothing in here
1213  * should check X86_PF_USER without a specific justification: for almost
1214  * all purposes, we should treat a normal kernel access to user memory
1215  * (e.g. get_user(), put_user(), etc.) the same as the WRUSS instruction.
1216  * The one exception is AC flag handling, which is, per the x86
1217  * architecture, special for WRUSS.
1218  */
1219 static inline
1220 void do_user_addr_fault(struct pt_regs *regs,
1221             unsigned long error_code,
1222             unsigned long address)
1223 {
1224     struct vm_area_struct *vma;
1225     struct task_struct *tsk;
1226     struct mm_struct *mm;
1227     vm_fault_t fault;
1228     unsigned int flags = FAULT_FLAG_DEFAULT;
1229 
1230     tsk = current;
1231     mm = tsk->mm;
1232 
1233     if (unlikely((error_code & (X86_PF_USER | X86_PF_INSTR)) == X86_PF_INSTR)) {
1234         /*
1235          * Whoops, this is kernel mode code trying to execute from
1236          * user memory.  Unless this is AMD erratum #93, which
1237          * corrupts RIP such that it looks like a user address,
1238          * this is unrecoverable.  Don't even try to look up the
1239          * VMA or look for extable entries.
1240          */
1241         if (is_errata93(regs, address))
1242             return;
1243 
1244         page_fault_oops(regs, error_code, address);
1245         return;
1246     }
1247 
1248     /* kprobes don't want to hook the spurious faults: */
1249     if (WARN_ON_ONCE(kprobe_page_fault(regs, X86_TRAP_PF)))
1250         return;
1251 
1252     /*
1253      * Reserved bits are never expected to be set on
1254      * entries in the user portion of the page tables.
1255      */
1256     if (unlikely(error_code & X86_PF_RSVD))
1257         pgtable_bad(regs, error_code, address);
1258 
1259     /*
1260      * If SMAP is on, check for invalid kernel (supervisor) access to user
1261      * pages in the user address space.  The odd case here is WRUSS,
1262      * which, according to the preliminary documentation, does not respect
1263      * SMAP and will have the USER bit set so, in all cases, SMAP
1264      * enforcement appears to be consistent with the USER bit.
1265      */
1266     if (unlikely(cpu_feature_enabled(X86_FEATURE_SMAP) &&
1267              !(error_code & X86_PF_USER) &&
1268              !(regs->flags & X86_EFLAGS_AC))) {
1269         /*
1270          * No extable entry here.  This was a kernel access to an
1271          * invalid pointer.  get_kernel_nofault() will not get here.
1272          */
1273         page_fault_oops(regs, error_code, address);
1274         return;
1275     }
1276 
1277     /*
1278      * If we're in an interrupt, have no user context or are running
1279      * in a region with pagefaults disabled then we must not take the fault
1280      */
1281     if (unlikely(faulthandler_disabled() || !mm)) {
1282         bad_area_nosemaphore(regs, error_code, address);
1283         return;
1284     }
1285 
1286     /*
1287      * It's safe to allow irq's after cr2 has been saved and the
1288      * vmalloc fault has been handled.
1289      *
1290      * User-mode registers count as a user access even for any
1291      * potential system fault or CPU buglet:
1292      */
1293     if (user_mode(regs)) {
1294         local_irq_enable();
1295         flags |= FAULT_FLAG_USER;
1296     } else {
1297         if (regs->flags & X86_EFLAGS_IF)
1298             local_irq_enable();
1299     }
1300 
1301     perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS, 1, regs, address);
1302 
1303     if (error_code & X86_PF_WRITE)
1304         flags |= FAULT_FLAG_WRITE;
1305     if (error_code & X86_PF_INSTR)
1306         flags |= FAULT_FLAG_INSTRUCTION;
1307 
1308 #ifdef CONFIG_X86_64
1309     /*
1310      * Faults in the vsyscall page might need emulation.  The
1311      * vsyscall page is at a high address (>PAGE_OFFSET), but is
1312      * considered to be part of the user address space.
1313      *
1314      * The vsyscall page does not have a "real" VMA, so do this
1315      * emulation before we go searching for VMAs.
1316      *
1317      * PKRU never rejects instruction fetches, so we don't need
1318      * to consider the PF_PK bit.
1319      */
1320     if (is_vsyscall_vaddr(address)) {
1321         if (emulate_vsyscall(error_code, regs, address))
1322             return;
1323     }
1324 #endif
1325 
1326     /*
1327      * Kernel-mode access to the user address space should only occur
1328      * on well-defined single instructions listed in the exception
1329      * tables.  But, an erroneous kernel fault occurring outside one of
1330      * those areas which also holds mmap_lock might deadlock attempting
1331      * to validate the fault against the address space.
1332      *
1333      * Only do the expensive exception table search when we might be at
1334      * risk of a deadlock.  This happens if we
1335      * 1. Failed to acquire mmap_lock, and
1336      * 2. The access did not originate in userspace.
1337      */
1338     if (unlikely(!mmap_read_trylock(mm))) {
1339         if (!user_mode(regs) && !search_exception_tables(regs->ip)) {
1340             /*
1341              * Fault from code in kernel from
1342              * which we do not expect faults.
1343              */
1344             bad_area_nosemaphore(regs, error_code, address);
1345             return;
1346         }
1347 retry:
1348         mmap_read_lock(mm);
1349     } else {
1350         /*
1351          * The above down_read_trylock() might have succeeded in
1352          * which case we'll have missed the might_sleep() from
1353          * down_read():
1354          */
1355         might_sleep();
1356     }
1357 
1358     vma = find_vma(mm, address);
1359     if (unlikely(!vma)) {
1360         bad_area(regs, error_code, address);
1361         return;
1362     }
1363     if (likely(vma->vm_start <= address))
1364         goto good_area;
1365     if (unlikely(!(vma->vm_flags & VM_GROWSDOWN))) {
1366         bad_area(regs, error_code, address);
1367         return;
1368     }
1369     if (unlikely(expand_stack(vma, address))) {
1370         bad_area(regs, error_code, address);
1371         return;
1372     }
1373 
1374     /*
1375      * Ok, we have a good vm_area for this memory access, so
1376      * we can handle it..
1377      */
1378 good_area:
1379     if (unlikely(access_error(error_code, vma))) {
1380         bad_area_access_error(regs, error_code, address, vma);
1381         return;
1382     }
1383 
1384     /*
1385      * If for any reason at all we couldn't handle the fault,
1386      * make sure we exit gracefully rather than endlessly redo
1387      * the fault.  Since we never set FAULT_FLAG_RETRY_NOWAIT, if
1388      * we get VM_FAULT_RETRY back, the mmap_lock has been unlocked.
1389      *
1390      * Note that handle_userfault() may also release and reacquire mmap_lock
1391      * (and not return with VM_FAULT_RETRY), when returning to userland to
1392      * repeat the page fault later with a VM_FAULT_NOPAGE retval
1393      * (potentially after handling any pending signal during the return to
1394      * userland). The return to userland is identified whenever
1395      * FAULT_FLAG_USER|FAULT_FLAG_KILLABLE are both set in flags.
1396      */
1397     fault = handle_mm_fault(vma, address, flags, regs);
1398 
1399     if (fault_signal_pending(fault, regs)) {
1400         /*
1401          * Quick path to respond to signals.  The core mm code
1402          * has unlocked the mm for us if we get here.
1403          */
1404         if (!user_mode(regs))
1405             kernelmode_fixup_or_oops(regs, error_code, address,
1406                          SIGBUS, BUS_ADRERR,
1407                          ARCH_DEFAULT_PKEY);
1408         return;
1409     }
1410 
1411     /* The fault is fully completed (including releasing mmap lock) */
1412     if (fault & VM_FAULT_COMPLETED)
1413         return;
1414 
1415     /*
1416      * If we need to retry the mmap_lock has already been released,
1417      * and if there is a fatal signal pending there is no guarantee
1418      * that we made any progress. Handle this case first.
1419      */
1420     if (unlikely(fault & VM_FAULT_RETRY)) {
1421         flags |= FAULT_FLAG_TRIED;
1422         goto retry;
1423     }
1424 
1425     mmap_read_unlock(mm);
1426     if (likely(!(fault & VM_FAULT_ERROR)))
1427         return;
1428 
1429     if (fatal_signal_pending(current) && !user_mode(regs)) {
1430         kernelmode_fixup_or_oops(regs, error_code, address,
1431                      0, 0, ARCH_DEFAULT_PKEY);
1432         return;
1433     }
1434 
1435     if (fault & VM_FAULT_OOM) {
1436         /* Kernel mode? Handle exceptions or die: */
1437         if (!user_mode(regs)) {
1438             kernelmode_fixup_or_oops(regs, error_code, address,
1439                          SIGSEGV, SEGV_MAPERR,
1440                          ARCH_DEFAULT_PKEY);
1441             return;
1442         }
1443 
1444         /*
1445          * We ran out of memory, call the OOM killer, and return the
1446          * userspace (which will retry the fault, or kill us if we got
1447          * oom-killed):
1448          */
1449         pagefault_out_of_memory();
1450     } else {
1451         if (fault & (VM_FAULT_SIGBUS|VM_FAULT_HWPOISON|
1452                  VM_FAULT_HWPOISON_LARGE))
1453             do_sigbus(regs, error_code, address, fault);
1454         else if (fault & VM_FAULT_SIGSEGV)
1455             bad_area_nosemaphore(regs, error_code, address);
1456         else
1457             BUG();
1458     }
1459 }
1460 NOKPROBE_SYMBOL(do_user_addr_fault);
1461 
1462 static __always_inline void
1463 trace_page_fault_entries(struct pt_regs *regs, unsigned long error_code,
1464              unsigned long address)
1465 {
1466     if (!trace_pagefault_enabled())
1467         return;
1468 
1469     if (user_mode(regs))
1470         trace_page_fault_user(address, regs, error_code);
1471     else
1472         trace_page_fault_kernel(address, regs, error_code);
1473 }
1474 
1475 static __always_inline void
1476 handle_page_fault(struct pt_regs *regs, unsigned long error_code,
1477                   unsigned long address)
1478 {
1479     trace_page_fault_entries(regs, error_code, address);
1480 
1481     if (unlikely(kmmio_fault(regs, address)))
1482         return;
1483 
1484     /* Was the fault on kernel-controlled part of the address space? */
1485     if (unlikely(fault_in_kernel_space(address))) {
1486         do_kern_addr_fault(regs, error_code, address);
1487     } else {
1488         do_user_addr_fault(regs, error_code, address);
1489         /*
1490          * User address page fault handling might have reenabled
1491          * interrupts. Fixing up all potential exit points of
1492          * do_user_addr_fault() and its leaf functions is just not
1493          * doable w/o creating an unholy mess or turning the code
1494          * upside down.
1495          */
1496         local_irq_disable();
1497     }
1498 }
1499 
1500 DEFINE_IDTENTRY_RAW_ERRORCODE(exc_page_fault)
1501 {
1502     unsigned long address = read_cr2();
1503     irqentry_state_t state;
1504 
1505     prefetchw(&current->mm->mmap_lock);
1506 
1507     /*
1508      * KVM uses #PF vector to deliver 'page not present' events to guests
1509      * (asynchronous page fault mechanism). The event happens when a
1510      * userspace task is trying to access some valid (from guest's point of
1511      * view) memory which is not currently mapped by the host (e.g. the
1512      * memory is swapped out). Note, the corresponding "page ready" event
1513      * which is injected when the memory becomes available, is delivered via
1514      * an interrupt mechanism and not a #PF exception
1515      * (see arch/x86/kernel/kvm.c: sysvec_kvm_asyncpf_interrupt()).
1516      *
1517      * We are relying on the interrupted context being sane (valid RSP,
1518      * relevant locks not held, etc.), which is fine as long as the
1519      * interrupted context had IF=1.  We are also relying on the KVM
1520      * async pf type field and CR2 being read consistently instead of
1521      * getting values from real and async page faults mixed up.
1522      *
1523      * Fingers crossed.
1524      *
1525      * The async #PF handling code takes care of idtentry handling
1526      * itself.
1527      */
1528     if (kvm_handle_async_pf(regs, (u32)address))
1529         return;
1530 
1531     /*
1532      * Entry handling for valid #PF from kernel mode is slightly
1533      * different: RCU is already watching and ct_irq_enter() must not
1534      * be invoked because a kernel fault on a user space address might
1535      * sleep.
1536      *
1537      * In case the fault hit a RCU idle region the conditional entry
1538      * code reenabled RCU to avoid subsequent wreckage which helps
1539      * debuggability.
1540      */
1541     state = irqentry_enter(regs);
1542 
1543     instrumentation_begin();
1544     handle_page_fault(regs, error_code, address);
1545     instrumentation_end();
1546 
1547     irqentry_exit(regs, state);
1548 }
[ Source navigation ] [ Diff markup ] [ Identifier search ] [ general search ]

This page was automatically generated by the 2.1.0 LXR engine.

The LXR team
Valid CSS 2.1! Valid HTML 4.01!