11da177e4SLinus Torvalds /* 21da177e4SLinus Torvalds * linux/mm/memory.c 31da177e4SLinus Torvalds * 41da177e4SLinus Torvalds * Copyright (C) 1991, 1992, 1993, 1994 Linus Torvalds 51da177e4SLinus Torvalds */ 61da177e4SLinus Torvalds 71da177e4SLinus Torvalds /* 81da177e4SLinus Torvalds * demand-loading started 01.12.91 - seems it is high on the list of 91da177e4SLinus Torvalds * things wanted, and it should be easy to implement. - Linus 101da177e4SLinus Torvalds */ 111da177e4SLinus Torvalds 121da177e4SLinus Torvalds /* 131da177e4SLinus Torvalds * Ok, demand-loading was easy, shared pages a little bit tricker. Shared 141da177e4SLinus Torvalds * pages started 02.12.91, seems to work. - Linus. 151da177e4SLinus Torvalds * 161da177e4SLinus Torvalds * Tested sharing by executing about 30 /bin/sh: under the old kernel it 171da177e4SLinus Torvalds * would have taken more than the 6M I have free, but it worked well as 181da177e4SLinus Torvalds * far as I could see. 191da177e4SLinus Torvalds * 201da177e4SLinus Torvalds * Also corrected some "invalidate()"s - I wasn't doing enough of them. 211da177e4SLinus Torvalds */ 221da177e4SLinus Torvalds 231da177e4SLinus Torvalds /* 241da177e4SLinus Torvalds * Real VM (paging to/from disk) started 18.12.91. Much more work and 251da177e4SLinus Torvalds * thought has to go into this. Oh, well.. 261da177e4SLinus Torvalds * 19.12.91 - works, somewhat. Sometimes I get faults, don't know why. 271da177e4SLinus Torvalds * Found it. Everything seems to work now. 281da177e4SLinus Torvalds * 20.12.91 - Ok, making the swap-device changeable like the root. 291da177e4SLinus Torvalds */ 301da177e4SLinus Torvalds 311da177e4SLinus Torvalds /* 321da177e4SLinus Torvalds * 05.04.94 - Multi-page memory management added for v1.1. 331da177e4SLinus Torvalds * Idea by Alex Bligh (alex@cconcepts.co.uk) 341da177e4SLinus Torvalds * 351da177e4SLinus Torvalds * 16.07.99 - Support of BIGMEM added by Gerhard Wichert, Siemens AG 361da177e4SLinus Torvalds * (Gerhard.Wichert@pdb.siemens.de) 371da177e4SLinus Torvalds * 381da177e4SLinus Torvalds * Aug/Sep 2004 Changed to four level page tables (Andi Kleen) 391da177e4SLinus Torvalds */ 401da177e4SLinus Torvalds 411da177e4SLinus Torvalds #include <linux/kernel_stat.h> 421da177e4SLinus Torvalds #include <linux/mm.h> 431da177e4SLinus Torvalds #include <linux/hugetlb.h> 441da177e4SLinus Torvalds #include <linux/mman.h> 451da177e4SLinus Torvalds #include <linux/swap.h> 461da177e4SLinus Torvalds #include <linux/highmem.h> 471da177e4SLinus Torvalds #include <linux/pagemap.h> 481da177e4SLinus Torvalds #include <linux/rmap.h> 491da177e4SLinus Torvalds #include <linux/module.h> 500ff92245SShailabh Nagar #include <linux/delayacct.h> 511da177e4SLinus Torvalds #include <linux/init.h> 52edc79b2aSPeter Zijlstra #include <linux/writeback.h> 531da177e4SLinus Torvalds 541da177e4SLinus Torvalds #include <asm/pgalloc.h> 551da177e4SLinus Torvalds #include <asm/uaccess.h> 561da177e4SLinus Torvalds #include <asm/tlb.h> 571da177e4SLinus Torvalds #include <asm/tlbflush.h> 581da177e4SLinus Torvalds #include <asm/pgtable.h> 591da177e4SLinus Torvalds 601da177e4SLinus Torvalds #include <linux/swapops.h> 611da177e4SLinus Torvalds #include <linux/elf.h> 621da177e4SLinus Torvalds 63d41dee36SAndy Whitcroft #ifndef CONFIG_NEED_MULTIPLE_NODES 641da177e4SLinus Torvalds /* use the per-pgdat data instead for discontigmem - mbligh */ 651da177e4SLinus Torvalds unsigned long max_mapnr; 661da177e4SLinus Torvalds struct page *mem_map; 671da177e4SLinus Torvalds 681da177e4SLinus Torvalds EXPORT_SYMBOL(max_mapnr); 691da177e4SLinus Torvalds EXPORT_SYMBOL(mem_map); 701da177e4SLinus Torvalds #endif 711da177e4SLinus Torvalds 721da177e4SLinus Torvalds unsigned long num_physpages; 731da177e4SLinus Torvalds /* 741da177e4SLinus Torvalds * A number of key systems in x86 including ioremap() rely on the assumption 751da177e4SLinus Torvalds * that high_memory defines the upper bound on direct map memory, then end 761da177e4SLinus Torvalds * of ZONE_NORMAL. Under CONFIG_DISCONTIG this means that max_low_pfn and 771da177e4SLinus Torvalds * highstart_pfn must be the same; there must be no gap between ZONE_NORMAL 781da177e4SLinus Torvalds * and ZONE_HIGHMEM. 791da177e4SLinus Torvalds */ 801da177e4SLinus Torvalds void * high_memory; 811da177e4SLinus Torvalds 821da177e4SLinus Torvalds EXPORT_SYMBOL(num_physpages); 831da177e4SLinus Torvalds EXPORT_SYMBOL(high_memory); 841da177e4SLinus Torvalds 85a62eaf15SAndi Kleen int randomize_va_space __read_mostly = 1; 86a62eaf15SAndi Kleen 87a62eaf15SAndi Kleen static int __init disable_randmaps(char *s) 88a62eaf15SAndi Kleen { 89a62eaf15SAndi Kleen randomize_va_space = 0; 909b41046cSOGAWA Hirofumi return 1; 91a62eaf15SAndi Kleen } 92a62eaf15SAndi Kleen __setup("norandmaps", disable_randmaps); 93a62eaf15SAndi Kleen 94a62eaf15SAndi Kleen 951da177e4SLinus Torvalds /* 961da177e4SLinus Torvalds * If a p?d_bad entry is found while walking page tables, report 971da177e4SLinus Torvalds * the error, before resetting entry to p?d_none. Usually (but 981da177e4SLinus Torvalds * very seldom) called out from the p?d_none_or_clear_bad macros. 991da177e4SLinus Torvalds */ 1001da177e4SLinus Torvalds 1011da177e4SLinus Torvalds void pgd_clear_bad(pgd_t *pgd) 1021da177e4SLinus Torvalds { 1031da177e4SLinus Torvalds pgd_ERROR(*pgd); 1041da177e4SLinus Torvalds pgd_clear(pgd); 1051da177e4SLinus Torvalds } 1061da177e4SLinus Torvalds 1071da177e4SLinus Torvalds void pud_clear_bad(pud_t *pud) 1081da177e4SLinus Torvalds { 1091da177e4SLinus Torvalds pud_ERROR(*pud); 1101da177e4SLinus Torvalds pud_clear(pud); 1111da177e4SLinus Torvalds } 1121da177e4SLinus Torvalds 1131da177e4SLinus Torvalds void pmd_clear_bad(pmd_t *pmd) 1141da177e4SLinus Torvalds { 1151da177e4SLinus Torvalds pmd_ERROR(*pmd); 1161da177e4SLinus Torvalds pmd_clear(pmd); 1171da177e4SLinus Torvalds } 1181da177e4SLinus Torvalds 1191da177e4SLinus Torvalds /* 1201da177e4SLinus Torvalds * Note: this doesn't free the actual pages themselves. That 1211da177e4SLinus Torvalds * has been handled earlier when unmapping all the memory regions. 1221da177e4SLinus Torvalds */ 123e0da382cSHugh Dickins static void free_pte_range(struct mmu_gather *tlb, pmd_t *pmd) 1241da177e4SLinus Torvalds { 1251da177e4SLinus Torvalds struct page *page = pmd_page(*pmd); 1261da177e4SLinus Torvalds pmd_clear(pmd); 1274c21e2f2SHugh Dickins pte_lock_deinit(page); 128e0da382cSHugh Dickins pte_free_tlb(tlb, page); 129df849a15SChristoph Lameter dec_zone_page_state(page, NR_PAGETABLE); 1301da177e4SLinus Torvalds tlb->mm->nr_ptes--; 1311da177e4SLinus Torvalds } 1321da177e4SLinus Torvalds 133e0da382cSHugh Dickins static inline void free_pmd_range(struct mmu_gather *tlb, pud_t *pud, 134e0da382cSHugh Dickins unsigned long addr, unsigned long end, 135e0da382cSHugh Dickins unsigned long floor, unsigned long ceiling) 1361da177e4SLinus Torvalds { 1371da177e4SLinus Torvalds pmd_t *pmd; 1381da177e4SLinus Torvalds unsigned long next; 139e0da382cSHugh Dickins unsigned long start; 1401da177e4SLinus Torvalds 141e0da382cSHugh Dickins start = addr; 1421da177e4SLinus Torvalds pmd = pmd_offset(pud, addr); 1431da177e4SLinus Torvalds do { 1441da177e4SLinus Torvalds next = pmd_addr_end(addr, end); 1451da177e4SLinus Torvalds if (pmd_none_or_clear_bad(pmd)) 1461da177e4SLinus Torvalds continue; 147e0da382cSHugh Dickins free_pte_range(tlb, pmd); 1481da177e4SLinus Torvalds } while (pmd++, addr = next, addr != end); 1491da177e4SLinus Torvalds 150e0da382cSHugh Dickins start &= PUD_MASK; 151e0da382cSHugh Dickins if (start < floor) 152e0da382cSHugh Dickins return; 153e0da382cSHugh Dickins if (ceiling) { 154e0da382cSHugh Dickins ceiling &= PUD_MASK; 155e0da382cSHugh Dickins if (!ceiling) 156e0da382cSHugh Dickins return; 1571da177e4SLinus Torvalds } 158e0da382cSHugh Dickins if (end - 1 > ceiling - 1) 159e0da382cSHugh Dickins return; 160e0da382cSHugh Dickins 161e0da382cSHugh Dickins pmd = pmd_offset(pud, start); 162e0da382cSHugh Dickins pud_clear(pud); 163e0da382cSHugh Dickins pmd_free_tlb(tlb, pmd); 1641da177e4SLinus Torvalds } 1651da177e4SLinus Torvalds 166e0da382cSHugh Dickins static inline void free_pud_range(struct mmu_gather *tlb, pgd_t *pgd, 167e0da382cSHugh Dickins unsigned long addr, unsigned long end, 168e0da382cSHugh Dickins unsigned long floor, unsigned long ceiling) 1691da177e4SLinus Torvalds { 1701da177e4SLinus Torvalds pud_t *pud; 1711da177e4SLinus Torvalds unsigned long next; 172e0da382cSHugh Dickins unsigned long start; 1731da177e4SLinus Torvalds 174e0da382cSHugh Dickins start = addr; 1751da177e4SLinus Torvalds pud = pud_offset(pgd, addr); 1761da177e4SLinus Torvalds do { 1771da177e4SLinus Torvalds next = pud_addr_end(addr, end); 1781da177e4SLinus Torvalds if (pud_none_or_clear_bad(pud)) 1791da177e4SLinus Torvalds continue; 180e0da382cSHugh Dickins free_pmd_range(tlb, pud, addr, next, floor, ceiling); 1811da177e4SLinus Torvalds } while (pud++, addr = next, addr != end); 1821da177e4SLinus Torvalds 183e0da382cSHugh Dickins start &= PGDIR_MASK; 184e0da382cSHugh Dickins if (start < floor) 185e0da382cSHugh Dickins return; 186e0da382cSHugh Dickins if (ceiling) { 187e0da382cSHugh Dickins ceiling &= PGDIR_MASK; 188e0da382cSHugh Dickins if (!ceiling) 189e0da382cSHugh Dickins return; 1901da177e4SLinus Torvalds } 191e0da382cSHugh Dickins if (end - 1 > ceiling - 1) 192e0da382cSHugh Dickins return; 193e0da382cSHugh Dickins 194e0da382cSHugh Dickins pud = pud_offset(pgd, start); 195e0da382cSHugh Dickins pgd_clear(pgd); 196e0da382cSHugh Dickins pud_free_tlb(tlb, pud); 1971da177e4SLinus Torvalds } 1981da177e4SLinus Torvalds 1991da177e4SLinus Torvalds /* 200e0da382cSHugh Dickins * This function frees user-level page tables of a process. 201e0da382cSHugh Dickins * 2021da177e4SLinus Torvalds * Must be called with pagetable lock held. 2031da177e4SLinus Torvalds */ 2043bf5ee95SHugh Dickins void free_pgd_range(struct mmu_gather **tlb, 205e0da382cSHugh Dickins unsigned long addr, unsigned long end, 206e0da382cSHugh Dickins unsigned long floor, unsigned long ceiling) 2071da177e4SLinus Torvalds { 2081da177e4SLinus Torvalds pgd_t *pgd; 2091da177e4SLinus Torvalds unsigned long next; 210e0da382cSHugh Dickins unsigned long start; 2111da177e4SLinus Torvalds 212e0da382cSHugh Dickins /* 213e0da382cSHugh Dickins * The next few lines have given us lots of grief... 214e0da382cSHugh Dickins * 215e0da382cSHugh Dickins * Why are we testing PMD* at this top level? Because often 216e0da382cSHugh Dickins * there will be no work to do at all, and we'd prefer not to 217e0da382cSHugh Dickins * go all the way down to the bottom just to discover that. 218e0da382cSHugh Dickins * 219e0da382cSHugh Dickins * Why all these "- 1"s? Because 0 represents both the bottom 220e0da382cSHugh Dickins * of the address space and the top of it (using -1 for the 221e0da382cSHugh Dickins * top wouldn't help much: the masks would do the wrong thing). 222e0da382cSHugh Dickins * The rule is that addr 0 and floor 0 refer to the bottom of 223e0da382cSHugh Dickins * the address space, but end 0 and ceiling 0 refer to the top 224e0da382cSHugh Dickins * Comparisons need to use "end - 1" and "ceiling - 1" (though 225e0da382cSHugh Dickins * that end 0 case should be mythical). 226e0da382cSHugh Dickins * 227e0da382cSHugh Dickins * Wherever addr is brought up or ceiling brought down, we must 228e0da382cSHugh Dickins * be careful to reject "the opposite 0" before it confuses the 229e0da382cSHugh Dickins * subsequent tests. But what about where end is brought down 230e0da382cSHugh Dickins * by PMD_SIZE below? no, end can't go down to 0 there. 231e0da382cSHugh Dickins * 232e0da382cSHugh Dickins * Whereas we round start (addr) and ceiling down, by different 233e0da382cSHugh Dickins * masks at different levels, in order to test whether a table 234e0da382cSHugh Dickins * now has no other vmas using it, so can be freed, we don't 235e0da382cSHugh Dickins * bother to round floor or end up - the tests don't need that. 236e0da382cSHugh Dickins */ 237e0da382cSHugh Dickins 238e0da382cSHugh Dickins addr &= PMD_MASK; 239e0da382cSHugh Dickins if (addr < floor) { 240e0da382cSHugh Dickins addr += PMD_SIZE; 241e0da382cSHugh Dickins if (!addr) 242e0da382cSHugh Dickins return; 243e0da382cSHugh Dickins } 244e0da382cSHugh Dickins if (ceiling) { 245e0da382cSHugh Dickins ceiling &= PMD_MASK; 246e0da382cSHugh Dickins if (!ceiling) 247e0da382cSHugh Dickins return; 248e0da382cSHugh Dickins } 249e0da382cSHugh Dickins if (end - 1 > ceiling - 1) 250e0da382cSHugh Dickins end -= PMD_SIZE; 251e0da382cSHugh Dickins if (addr > end - 1) 252e0da382cSHugh Dickins return; 253e0da382cSHugh Dickins 254e0da382cSHugh Dickins start = addr; 2553bf5ee95SHugh Dickins pgd = pgd_offset((*tlb)->mm, addr); 2561da177e4SLinus Torvalds do { 2571da177e4SLinus Torvalds next = pgd_addr_end(addr, end); 2581da177e4SLinus Torvalds if (pgd_none_or_clear_bad(pgd)) 2591da177e4SLinus Torvalds continue; 2603bf5ee95SHugh Dickins free_pud_range(*tlb, pgd, addr, next, floor, ceiling); 2611da177e4SLinus Torvalds } while (pgd++, addr = next, addr != end); 262e0da382cSHugh Dickins 2634d6ddfa9SHugh Dickins if (!(*tlb)->fullmm) 2643bf5ee95SHugh Dickins flush_tlb_pgtables((*tlb)->mm, start, end); 265e0da382cSHugh Dickins } 266e0da382cSHugh Dickins 267e0da382cSHugh Dickins void free_pgtables(struct mmu_gather **tlb, struct vm_area_struct *vma, 268e0da382cSHugh Dickins unsigned long floor, unsigned long ceiling) 269e0da382cSHugh Dickins { 270e0da382cSHugh Dickins while (vma) { 271e0da382cSHugh Dickins struct vm_area_struct *next = vma->vm_next; 272e0da382cSHugh Dickins unsigned long addr = vma->vm_start; 273e0da382cSHugh Dickins 2748f4f8c16SHugh Dickins /* 2758f4f8c16SHugh Dickins * Hide vma from rmap and vmtruncate before freeing pgtables 2768f4f8c16SHugh Dickins */ 2778f4f8c16SHugh Dickins anon_vma_unlink(vma); 2788f4f8c16SHugh Dickins unlink_file_vma(vma); 2798f4f8c16SHugh Dickins 2809da61aefSDavid Gibson if (is_vm_hugetlb_page(vma)) { 2813bf5ee95SHugh Dickins hugetlb_free_pgd_range(tlb, addr, vma->vm_end, 2823bf5ee95SHugh Dickins floor, next? next->vm_start: ceiling); 2833bf5ee95SHugh Dickins } else { 2843bf5ee95SHugh Dickins /* 2853bf5ee95SHugh Dickins * Optimization: gather nearby vmas into one call down 2863bf5ee95SHugh Dickins */ 2873bf5ee95SHugh Dickins while (next && next->vm_start <= vma->vm_end + PMD_SIZE 2884866920bSDavid Gibson && !is_vm_hugetlb_page(next)) { 289e0da382cSHugh Dickins vma = next; 290e0da382cSHugh Dickins next = vma->vm_next; 2918f4f8c16SHugh Dickins anon_vma_unlink(vma); 2928f4f8c16SHugh Dickins unlink_file_vma(vma); 293e0da382cSHugh Dickins } 2943bf5ee95SHugh Dickins free_pgd_range(tlb, addr, vma->vm_end, 295e0da382cSHugh Dickins floor, next? next->vm_start: ceiling); 2963bf5ee95SHugh Dickins } 297e0da382cSHugh Dickins vma = next; 298e0da382cSHugh Dickins } 2991da177e4SLinus Torvalds } 3001da177e4SLinus Torvalds 3011bb3630eSHugh Dickins int __pte_alloc(struct mm_struct *mm, pmd_t *pmd, unsigned long address) 3021da177e4SLinus Torvalds { 303c74df32cSHugh Dickins struct page *new = pte_alloc_one(mm, address); 3041da177e4SLinus Torvalds if (!new) 3051bb3630eSHugh Dickins return -ENOMEM; 3061bb3630eSHugh Dickins 3074c21e2f2SHugh Dickins pte_lock_init(new); 308c74df32cSHugh Dickins spin_lock(&mm->page_table_lock); 3094c21e2f2SHugh Dickins if (pmd_present(*pmd)) { /* Another has populated it */ 3104c21e2f2SHugh Dickins pte_lock_deinit(new); 3111da177e4SLinus Torvalds pte_free(new); 3124c21e2f2SHugh Dickins } else { 3131da177e4SLinus Torvalds mm->nr_ptes++; 314df849a15SChristoph Lameter inc_zone_page_state(new, NR_PAGETABLE); 3151da177e4SLinus Torvalds pmd_populate(mm, pmd, new); 3161da177e4SLinus Torvalds } 317c74df32cSHugh Dickins spin_unlock(&mm->page_table_lock); 3181bb3630eSHugh Dickins return 0; 3191da177e4SLinus Torvalds } 3201da177e4SLinus Torvalds 3211bb3630eSHugh Dickins int __pte_alloc_kernel(pmd_t *pmd, unsigned long address) 3221da177e4SLinus Torvalds { 3231bb3630eSHugh Dickins pte_t *new = pte_alloc_one_kernel(&init_mm, address); 3241da177e4SLinus Torvalds if (!new) 3251bb3630eSHugh Dickins return -ENOMEM; 3261da177e4SLinus Torvalds 327872fec16SHugh Dickins spin_lock(&init_mm.page_table_lock); 3281bb3630eSHugh Dickins if (pmd_present(*pmd)) /* Another has populated it */ 3291da177e4SLinus Torvalds pte_free_kernel(new); 330872fec16SHugh Dickins else 331872fec16SHugh Dickins pmd_populate_kernel(&init_mm, pmd, new); 332872fec16SHugh Dickins spin_unlock(&init_mm.page_table_lock); 3331bb3630eSHugh Dickins return 0; 3341da177e4SLinus Torvalds } 3351da177e4SLinus Torvalds 336ae859762SHugh Dickins static inline void add_mm_rss(struct mm_struct *mm, int file_rss, int anon_rss) 337ae859762SHugh Dickins { 338ae859762SHugh Dickins if (file_rss) 339ae859762SHugh Dickins add_mm_counter(mm, file_rss, file_rss); 340ae859762SHugh Dickins if (anon_rss) 341ae859762SHugh Dickins add_mm_counter(mm, anon_rss, anon_rss); 342ae859762SHugh Dickins } 343ae859762SHugh Dickins 3441da177e4SLinus Torvalds /* 3456aab341eSLinus Torvalds * This function is called to print an error when a bad pte 3466aab341eSLinus Torvalds * is found. For example, we might have a PFN-mapped pte in 3476aab341eSLinus Torvalds * a region that doesn't allow it. 348b5810039SNick Piggin * 349b5810039SNick Piggin * The calling function must still handle the error. 350b5810039SNick Piggin */ 351b5810039SNick Piggin void print_bad_pte(struct vm_area_struct *vma, pte_t pte, unsigned long vaddr) 352b5810039SNick Piggin { 353b5810039SNick Piggin printk(KERN_ERR "Bad pte = %08llx, process = %s, " 354b5810039SNick Piggin "vm_flags = %lx, vaddr = %lx\n", 355b5810039SNick Piggin (long long)pte_val(pte), 356b5810039SNick Piggin (vma->vm_mm == current->mm ? current->comm : "???"), 357b5810039SNick Piggin vma->vm_flags, vaddr); 358b5810039SNick Piggin dump_stack(); 359b5810039SNick Piggin } 360b5810039SNick Piggin 36167121172SLinus Torvalds static inline int is_cow_mapping(unsigned int flags) 36267121172SLinus Torvalds { 36367121172SLinus Torvalds return (flags & (VM_SHARED | VM_MAYWRITE)) == VM_MAYWRITE; 36467121172SLinus Torvalds } 36567121172SLinus Torvalds 366b5810039SNick Piggin /* 3676aab341eSLinus Torvalds * This function gets the "struct page" associated with a pte. 3686aab341eSLinus Torvalds * 3696aab341eSLinus Torvalds * NOTE! Some mappings do not have "struct pages". A raw PFN mapping 3706aab341eSLinus Torvalds * will have each page table entry just pointing to a raw page frame 3716aab341eSLinus Torvalds * number, and as far as the VM layer is concerned, those do not have 3726aab341eSLinus Torvalds * pages associated with them - even if the PFN might point to memory 3736aab341eSLinus Torvalds * that otherwise is perfectly fine and has a "struct page". 3746aab341eSLinus Torvalds * 3756aab341eSLinus Torvalds * The way we recognize those mappings is through the rules set up 3766aab341eSLinus Torvalds * by "remap_pfn_range()": the vma will have the VM_PFNMAP bit set, 3776aab341eSLinus Torvalds * and the vm_pgoff will point to the first PFN mapped: thus every 3786aab341eSLinus Torvalds * page that is a raw mapping will always honor the rule 3796aab341eSLinus Torvalds * 3806aab341eSLinus Torvalds * pfn_of_page == vma->vm_pgoff + ((addr - vma->vm_start) >> PAGE_SHIFT) 3816aab341eSLinus Torvalds * 3826aab341eSLinus Torvalds * and if that isn't true, the page has been COW'ed (in which case it 3836aab341eSLinus Torvalds * _does_ have a "struct page" associated with it even if it is in a 3846aab341eSLinus Torvalds * VM_PFNMAP range). 385ee498ed7SHugh Dickins */ 3866aab341eSLinus Torvalds struct page *vm_normal_page(struct vm_area_struct *vma, unsigned long addr, pte_t pte) 387ee498ed7SHugh Dickins { 3886aab341eSLinus Torvalds unsigned long pfn = pte_pfn(pte); 3896aab341eSLinus Torvalds 390b7ab795bSNick Piggin if (unlikely(vma->vm_flags & VM_PFNMAP)) { 3916aab341eSLinus Torvalds unsigned long off = (addr - vma->vm_start) >> PAGE_SHIFT; 3926aab341eSLinus Torvalds if (pfn == vma->vm_pgoff + off) 3936aab341eSLinus Torvalds return NULL; 39467121172SLinus Torvalds if (!is_cow_mapping(vma->vm_flags)) 395fb155c16SLinus Torvalds return NULL; 3966aab341eSLinus Torvalds } 3976aab341eSLinus Torvalds 398315ab19aSNick Piggin /* 399315ab19aSNick Piggin * Add some anal sanity checks for now. Eventually, 400315ab19aSNick Piggin * we should just do "return pfn_to_page(pfn)", but 401315ab19aSNick Piggin * in the meantime we check that we get a valid pfn, 402315ab19aSNick Piggin * and that the resulting page looks ok. 403315ab19aSNick Piggin */ 4046aab341eSLinus Torvalds if (unlikely(!pfn_valid(pfn))) { 4056aab341eSLinus Torvalds print_bad_pte(vma, pte, addr); 4066aab341eSLinus Torvalds return NULL; 4076aab341eSLinus Torvalds } 4086aab341eSLinus Torvalds 4096aab341eSLinus Torvalds /* 4106aab341eSLinus Torvalds * NOTE! We still have PageReserved() pages in the page 4116aab341eSLinus Torvalds * tables. 4126aab341eSLinus Torvalds * 4136aab341eSLinus Torvalds * The PAGE_ZERO() pages and various VDSO mappings can 4146aab341eSLinus Torvalds * cause them to exist. 4156aab341eSLinus Torvalds */ 4166aab341eSLinus Torvalds return pfn_to_page(pfn); 417ee498ed7SHugh Dickins } 418ee498ed7SHugh Dickins 419ee498ed7SHugh Dickins /* 4201da177e4SLinus Torvalds * copy one vm_area from one task to the other. Assumes the page tables 4211da177e4SLinus Torvalds * already present in the new task to be cleared in the whole range 4221da177e4SLinus Torvalds * covered by this vma. 4231da177e4SLinus Torvalds */ 4241da177e4SLinus Torvalds 4258c103762SHugh Dickins static inline void 4261da177e4SLinus Torvalds copy_one_pte(struct mm_struct *dst_mm, struct mm_struct *src_mm, 427b5810039SNick Piggin pte_t *dst_pte, pte_t *src_pte, struct vm_area_struct *vma, 4288c103762SHugh Dickins unsigned long addr, int *rss) 4291da177e4SLinus Torvalds { 430b5810039SNick Piggin unsigned long vm_flags = vma->vm_flags; 4311da177e4SLinus Torvalds pte_t pte = *src_pte; 4321da177e4SLinus Torvalds struct page *page; 4331da177e4SLinus Torvalds 4341da177e4SLinus Torvalds /* pte contains position in swap or file, so copy. */ 4351da177e4SLinus Torvalds if (unlikely(!pte_present(pte))) { 4361da177e4SLinus Torvalds if (!pte_file(pte)) { 4370697212aSChristoph Lameter swp_entry_t entry = pte_to_swp_entry(pte); 4380697212aSChristoph Lameter 4390697212aSChristoph Lameter swap_duplicate(entry); 4401da177e4SLinus Torvalds /* make sure dst_mm is on swapoff's mmlist. */ 4411da177e4SLinus Torvalds if (unlikely(list_empty(&dst_mm->mmlist))) { 4421da177e4SLinus Torvalds spin_lock(&mmlist_lock); 443f412ac08SHugh Dickins if (list_empty(&dst_mm->mmlist)) 444f412ac08SHugh Dickins list_add(&dst_mm->mmlist, 445f412ac08SHugh Dickins &src_mm->mmlist); 4461da177e4SLinus Torvalds spin_unlock(&mmlist_lock); 4471da177e4SLinus Torvalds } 4480697212aSChristoph Lameter if (is_write_migration_entry(entry) && 4490697212aSChristoph Lameter is_cow_mapping(vm_flags)) { 4500697212aSChristoph Lameter /* 4510697212aSChristoph Lameter * COW mappings require pages in both parent 4520697212aSChristoph Lameter * and child to be set to read. 4530697212aSChristoph Lameter */ 4540697212aSChristoph Lameter make_migration_entry_read(&entry); 4550697212aSChristoph Lameter pte = swp_entry_to_pte(entry); 4560697212aSChristoph Lameter set_pte_at(src_mm, addr, src_pte, pte); 4570697212aSChristoph Lameter } 4581da177e4SLinus Torvalds } 459ae859762SHugh Dickins goto out_set_pte; 4601da177e4SLinus Torvalds } 4611da177e4SLinus Torvalds 4621da177e4SLinus Torvalds /* 4631da177e4SLinus Torvalds * If it's a COW mapping, write protect it both 4641da177e4SLinus Torvalds * in the parent and the child 4651da177e4SLinus Torvalds */ 46667121172SLinus Torvalds if (is_cow_mapping(vm_flags)) { 4671da177e4SLinus Torvalds ptep_set_wrprotect(src_mm, addr, src_pte); 4683dc90795SZachary Amsden pte = pte_wrprotect(pte); 4691da177e4SLinus Torvalds } 4701da177e4SLinus Torvalds 4711da177e4SLinus Torvalds /* 4721da177e4SLinus Torvalds * If it's a shared mapping, mark it clean in 4731da177e4SLinus Torvalds * the child 4741da177e4SLinus Torvalds */ 4751da177e4SLinus Torvalds if (vm_flags & VM_SHARED) 4761da177e4SLinus Torvalds pte = pte_mkclean(pte); 4771da177e4SLinus Torvalds pte = pte_mkold(pte); 4786aab341eSLinus Torvalds 4796aab341eSLinus Torvalds page = vm_normal_page(vma, addr, pte); 4806aab341eSLinus Torvalds if (page) { 4811da177e4SLinus Torvalds get_page(page); 482c97a9e10SNick Piggin page_dup_rmap(page, vma, addr); 4838c103762SHugh Dickins rss[!!PageAnon(page)]++; 4846aab341eSLinus Torvalds } 485ae859762SHugh Dickins 486ae859762SHugh Dickins out_set_pte: 487ae859762SHugh Dickins set_pte_at(dst_mm, addr, dst_pte, pte); 4881da177e4SLinus Torvalds } 4891da177e4SLinus Torvalds 4901da177e4SLinus Torvalds static int copy_pte_range(struct mm_struct *dst_mm, struct mm_struct *src_mm, 4911da177e4SLinus Torvalds pmd_t *dst_pmd, pmd_t *src_pmd, struct vm_area_struct *vma, 4921da177e4SLinus Torvalds unsigned long addr, unsigned long end) 4931da177e4SLinus Torvalds { 4941da177e4SLinus Torvalds pte_t *src_pte, *dst_pte; 495c74df32cSHugh Dickins spinlock_t *src_ptl, *dst_ptl; 496e040f218SHugh Dickins int progress = 0; 4978c103762SHugh Dickins int rss[2]; 4981da177e4SLinus Torvalds 4991da177e4SLinus Torvalds again: 500ae859762SHugh Dickins rss[1] = rss[0] = 0; 501c74df32cSHugh Dickins dst_pte = pte_alloc_map_lock(dst_mm, dst_pmd, addr, &dst_ptl); 5021da177e4SLinus Torvalds if (!dst_pte) 5031da177e4SLinus Torvalds return -ENOMEM; 5041da177e4SLinus Torvalds src_pte = pte_offset_map_nested(src_pmd, addr); 5054c21e2f2SHugh Dickins src_ptl = pte_lockptr(src_mm, src_pmd); 506f20dc5f7SIngo Molnar spin_lock_nested(src_ptl, SINGLE_DEPTH_NESTING); 5076606c3e0SZachary Amsden arch_enter_lazy_mmu_mode(); 5081da177e4SLinus Torvalds 5091da177e4SLinus Torvalds do { 5101da177e4SLinus Torvalds /* 5111da177e4SLinus Torvalds * We are holding two locks at this point - either of them 5121da177e4SLinus Torvalds * could generate latencies in another task on another CPU. 5131da177e4SLinus Torvalds */ 514e040f218SHugh Dickins if (progress >= 32) { 515e040f218SHugh Dickins progress = 0; 516e040f218SHugh Dickins if (need_resched() || 517c74df32cSHugh Dickins need_lockbreak(src_ptl) || 518c74df32cSHugh Dickins need_lockbreak(dst_ptl)) 5191da177e4SLinus Torvalds break; 520e040f218SHugh Dickins } 5211da177e4SLinus Torvalds if (pte_none(*src_pte)) { 5221da177e4SLinus Torvalds progress++; 5231da177e4SLinus Torvalds continue; 5241da177e4SLinus Torvalds } 5258c103762SHugh Dickins copy_one_pte(dst_mm, src_mm, dst_pte, src_pte, vma, addr, rss); 5261da177e4SLinus Torvalds progress += 8; 5271da177e4SLinus Torvalds } while (dst_pte++, src_pte++, addr += PAGE_SIZE, addr != end); 5281da177e4SLinus Torvalds 5296606c3e0SZachary Amsden arch_leave_lazy_mmu_mode(); 530c74df32cSHugh Dickins spin_unlock(src_ptl); 5311da177e4SLinus Torvalds pte_unmap_nested(src_pte - 1); 532ae859762SHugh Dickins add_mm_rss(dst_mm, rss[0], rss[1]); 533c74df32cSHugh Dickins pte_unmap_unlock(dst_pte - 1, dst_ptl); 534c74df32cSHugh Dickins cond_resched(); 5351da177e4SLinus Torvalds if (addr != end) 5361da177e4SLinus Torvalds goto again; 5371da177e4SLinus Torvalds return 0; 5381da177e4SLinus Torvalds } 5391da177e4SLinus Torvalds 5401da177e4SLinus Torvalds static inline int copy_pmd_range(struct mm_struct *dst_mm, struct mm_struct *src_mm, 5411da177e4SLinus Torvalds pud_t *dst_pud, pud_t *src_pud, struct vm_area_struct *vma, 5421da177e4SLinus Torvalds unsigned long addr, unsigned long end) 5431da177e4SLinus Torvalds { 5441da177e4SLinus Torvalds pmd_t *src_pmd, *dst_pmd; 5451da177e4SLinus Torvalds unsigned long next; 5461da177e4SLinus Torvalds 5471da177e4SLinus Torvalds dst_pmd = pmd_alloc(dst_mm, dst_pud, addr); 5481da177e4SLinus Torvalds if (!dst_pmd) 5491da177e4SLinus Torvalds return -ENOMEM; 5501da177e4SLinus Torvalds src_pmd = pmd_offset(src_pud, addr); 5511da177e4SLinus Torvalds do { 5521da177e4SLinus Torvalds next = pmd_addr_end(addr, end); 5531da177e4SLinus Torvalds if (pmd_none_or_clear_bad(src_pmd)) 5541da177e4SLinus Torvalds continue; 5551da177e4SLinus Torvalds if (copy_pte_range(dst_mm, src_mm, dst_pmd, src_pmd, 5561da177e4SLinus Torvalds vma, addr, next)) 5571da177e4SLinus Torvalds return -ENOMEM; 5581da177e4SLinus Torvalds } while (dst_pmd++, src_pmd++, addr = next, addr != end); 5591da177e4SLinus Torvalds return 0; 5601da177e4SLinus Torvalds } 5611da177e4SLinus Torvalds 5621da177e4SLinus Torvalds static inline int copy_pud_range(struct mm_struct *dst_mm, struct mm_struct *src_mm, 5631da177e4SLinus Torvalds pgd_t *dst_pgd, pgd_t *src_pgd, struct vm_area_struct *vma, 5641da177e4SLinus Torvalds unsigned long addr, unsigned long end) 5651da177e4SLinus Torvalds { 5661da177e4SLinus Torvalds pud_t *src_pud, *dst_pud; 5671da177e4SLinus Torvalds unsigned long next; 5681da177e4SLinus Torvalds 5691da177e4SLinus Torvalds dst_pud = pud_alloc(dst_mm, dst_pgd, addr); 5701da177e4SLinus Torvalds if (!dst_pud) 5711da177e4SLinus Torvalds return -ENOMEM; 5721da177e4SLinus Torvalds src_pud = pud_offset(src_pgd, addr); 5731da177e4SLinus Torvalds do { 5741da177e4SLinus Torvalds next = pud_addr_end(addr, end); 5751da177e4SLinus Torvalds if (pud_none_or_clear_bad(src_pud)) 5761da177e4SLinus Torvalds continue; 5771da177e4SLinus Torvalds if (copy_pmd_range(dst_mm, src_mm, dst_pud, src_pud, 5781da177e4SLinus Torvalds vma, addr, next)) 5791da177e4SLinus Torvalds return -ENOMEM; 5801da177e4SLinus Torvalds } while (dst_pud++, src_pud++, addr = next, addr != end); 5811da177e4SLinus Torvalds return 0; 5821da177e4SLinus Torvalds } 5831da177e4SLinus Torvalds 5841da177e4SLinus Torvalds int copy_page_range(struct mm_struct *dst_mm, struct mm_struct *src_mm, 5851da177e4SLinus Torvalds struct vm_area_struct *vma) 5861da177e4SLinus Torvalds { 5871da177e4SLinus Torvalds pgd_t *src_pgd, *dst_pgd; 5881da177e4SLinus Torvalds unsigned long next; 5891da177e4SLinus Torvalds unsigned long addr = vma->vm_start; 5901da177e4SLinus Torvalds unsigned long end = vma->vm_end; 5911da177e4SLinus Torvalds 592d992895bSNick Piggin /* 593d992895bSNick Piggin * Don't copy ptes where a page fault will fill them correctly. 594d992895bSNick Piggin * Fork becomes much lighter when there are big shared or private 595d992895bSNick Piggin * readonly mappings. The tradeoff is that copy_page_range is more 596d992895bSNick Piggin * efficient than faulting. 597d992895bSNick Piggin */ 5984d7672b4SLinus Torvalds if (!(vma->vm_flags & (VM_HUGETLB|VM_NONLINEAR|VM_PFNMAP|VM_INSERTPAGE))) { 599d992895bSNick Piggin if (!vma->anon_vma) 600d992895bSNick Piggin return 0; 601d992895bSNick Piggin } 602d992895bSNick Piggin 6031da177e4SLinus Torvalds if (is_vm_hugetlb_page(vma)) 6041da177e4SLinus Torvalds return copy_hugetlb_page_range(dst_mm, src_mm, vma); 6051da177e4SLinus Torvalds 6061da177e4SLinus Torvalds dst_pgd = pgd_offset(dst_mm, addr); 6071da177e4SLinus Torvalds src_pgd = pgd_offset(src_mm, addr); 6081da177e4SLinus Torvalds do { 6091da177e4SLinus Torvalds next = pgd_addr_end(addr, end); 6101da177e4SLinus Torvalds if (pgd_none_or_clear_bad(src_pgd)) 6111da177e4SLinus Torvalds continue; 6121da177e4SLinus Torvalds if (copy_pud_range(dst_mm, src_mm, dst_pgd, src_pgd, 6131da177e4SLinus Torvalds vma, addr, next)) 6141da177e4SLinus Torvalds return -ENOMEM; 6151da177e4SLinus Torvalds } while (dst_pgd++, src_pgd++, addr = next, addr != end); 6161da177e4SLinus Torvalds return 0; 6171da177e4SLinus Torvalds } 6181da177e4SLinus Torvalds 61951c6f666SRobin Holt static unsigned long zap_pte_range(struct mmu_gather *tlb, 620b5810039SNick Piggin struct vm_area_struct *vma, pmd_t *pmd, 6211da177e4SLinus Torvalds unsigned long addr, unsigned long end, 62251c6f666SRobin Holt long *zap_work, struct zap_details *details) 6231da177e4SLinus Torvalds { 624b5810039SNick Piggin struct mm_struct *mm = tlb->mm; 6251da177e4SLinus Torvalds pte_t *pte; 626508034a3SHugh Dickins spinlock_t *ptl; 627ae859762SHugh Dickins int file_rss = 0; 628ae859762SHugh Dickins int anon_rss = 0; 6291da177e4SLinus Torvalds 630508034a3SHugh Dickins pte = pte_offset_map_lock(mm, pmd, addr, &ptl); 6316606c3e0SZachary Amsden arch_enter_lazy_mmu_mode(); 6321da177e4SLinus Torvalds do { 6331da177e4SLinus Torvalds pte_t ptent = *pte; 63451c6f666SRobin Holt if (pte_none(ptent)) { 63551c6f666SRobin Holt (*zap_work)--; 6361da177e4SLinus Torvalds continue; 63751c6f666SRobin Holt } 63851c6f666SRobin Holt 63951c6f666SRobin Holt (*zap_work) -= PAGE_SIZE; 64051c6f666SRobin Holt 6416f5e6b9eSHugh Dickins if (pte_present(ptent)) { 6426f5e6b9eSHugh Dickins struct page *page; 6436f5e6b9eSHugh Dickins 6446aab341eSLinus Torvalds page = vm_normal_page(vma, addr, ptent); 6451da177e4SLinus Torvalds if (unlikely(details) && page) { 6461da177e4SLinus Torvalds /* 6471da177e4SLinus Torvalds * unmap_shared_mapping_pages() wants to 6481da177e4SLinus Torvalds * invalidate cache without truncating: 6491da177e4SLinus Torvalds * unmap shared but keep private pages. 6501da177e4SLinus Torvalds */ 6511da177e4SLinus Torvalds if (details->check_mapping && 6521da177e4SLinus Torvalds details->check_mapping != page->mapping) 6531da177e4SLinus Torvalds continue; 6541da177e4SLinus Torvalds /* 6551da177e4SLinus Torvalds * Each page->index must be checked when 6561da177e4SLinus Torvalds * invalidating or truncating nonlinear. 6571da177e4SLinus Torvalds */ 6581da177e4SLinus Torvalds if (details->nonlinear_vma && 6591da177e4SLinus Torvalds (page->index < details->first_index || 6601da177e4SLinus Torvalds page->index > details->last_index)) 6611da177e4SLinus Torvalds continue; 6621da177e4SLinus Torvalds } 663b5810039SNick Piggin ptent = ptep_get_and_clear_full(mm, addr, pte, 664a600388dSZachary Amsden tlb->fullmm); 6651da177e4SLinus Torvalds tlb_remove_tlb_entry(tlb, pte, addr); 6661da177e4SLinus Torvalds if (unlikely(!page)) 6671da177e4SLinus Torvalds continue; 6681da177e4SLinus Torvalds if (unlikely(details) && details->nonlinear_vma 6691da177e4SLinus Torvalds && linear_page_index(details->nonlinear_vma, 6701da177e4SLinus Torvalds addr) != page->index) 671b5810039SNick Piggin set_pte_at(mm, addr, pte, 6721da177e4SLinus Torvalds pgoff_to_pte(page->index)); 6731da177e4SLinus Torvalds if (PageAnon(page)) 67486d912f4SHugh Dickins anon_rss--; 6756237bcd9SHugh Dickins else { 6766237bcd9SHugh Dickins if (pte_dirty(ptent)) 6776237bcd9SHugh Dickins set_page_dirty(page); 6786237bcd9SHugh Dickins if (pte_young(ptent)) 679daa88c8dSKen Chen SetPageReferenced(page); 68086d912f4SHugh Dickins file_rss--; 6816237bcd9SHugh Dickins } 6827de6b805SNick Piggin page_remove_rmap(page, vma); 6831da177e4SLinus Torvalds tlb_remove_page(tlb, page); 6841da177e4SLinus Torvalds continue; 6851da177e4SLinus Torvalds } 6861da177e4SLinus Torvalds /* 6871da177e4SLinus Torvalds * If details->check_mapping, we leave swap entries; 6881da177e4SLinus Torvalds * if details->nonlinear_vma, we leave file entries. 6891da177e4SLinus Torvalds */ 6901da177e4SLinus Torvalds if (unlikely(details)) 6911da177e4SLinus Torvalds continue; 6921da177e4SLinus Torvalds if (!pte_file(ptent)) 6931da177e4SLinus Torvalds free_swap_and_cache(pte_to_swp_entry(ptent)); 6949888a1caSZachary Amsden pte_clear_not_present_full(mm, addr, pte, tlb->fullmm); 69551c6f666SRobin Holt } while (pte++, addr += PAGE_SIZE, (addr != end && *zap_work > 0)); 696ae859762SHugh Dickins 69786d912f4SHugh Dickins add_mm_rss(mm, file_rss, anon_rss); 6986606c3e0SZachary Amsden arch_leave_lazy_mmu_mode(); 699508034a3SHugh Dickins pte_unmap_unlock(pte - 1, ptl); 70051c6f666SRobin Holt 70151c6f666SRobin Holt return addr; 7021da177e4SLinus Torvalds } 7031da177e4SLinus Torvalds 70451c6f666SRobin Holt static inline unsigned long zap_pmd_range(struct mmu_gather *tlb, 705b5810039SNick Piggin struct vm_area_struct *vma, pud_t *pud, 7061da177e4SLinus Torvalds unsigned long addr, unsigned long end, 70751c6f666SRobin Holt long *zap_work, struct zap_details *details) 7081da177e4SLinus Torvalds { 7091da177e4SLinus Torvalds pmd_t *pmd; 7101da177e4SLinus Torvalds unsigned long next; 7111da177e4SLinus Torvalds 7121da177e4SLinus Torvalds pmd = pmd_offset(pud, addr); 7131da177e4SLinus Torvalds do { 7141da177e4SLinus Torvalds next = pmd_addr_end(addr, end); 71551c6f666SRobin Holt if (pmd_none_or_clear_bad(pmd)) { 71651c6f666SRobin Holt (*zap_work)--; 7171da177e4SLinus Torvalds continue; 71851c6f666SRobin Holt } 71951c6f666SRobin Holt next = zap_pte_range(tlb, vma, pmd, addr, next, 72051c6f666SRobin Holt zap_work, details); 72151c6f666SRobin Holt } while (pmd++, addr = next, (addr != end && *zap_work > 0)); 72251c6f666SRobin Holt 72351c6f666SRobin Holt return addr; 7241da177e4SLinus Torvalds } 7251da177e4SLinus Torvalds 72651c6f666SRobin Holt static inline unsigned long zap_pud_range(struct mmu_gather *tlb, 727b5810039SNick Piggin struct vm_area_struct *vma, pgd_t *pgd, 7281da177e4SLinus Torvalds unsigned long addr, unsigned long end, 72951c6f666SRobin Holt long *zap_work, struct zap_details *details) 7301da177e4SLinus Torvalds { 7311da177e4SLinus Torvalds pud_t *pud; 7321da177e4SLinus Torvalds unsigned long next; 7331da177e4SLinus Torvalds 7341da177e4SLinus Torvalds pud = pud_offset(pgd, addr); 7351da177e4SLinus Torvalds do { 7361da177e4SLinus Torvalds next = pud_addr_end(addr, end); 73751c6f666SRobin Holt if (pud_none_or_clear_bad(pud)) { 73851c6f666SRobin Holt (*zap_work)--; 7391da177e4SLinus Torvalds continue; 74051c6f666SRobin Holt } 74151c6f666SRobin Holt next = zap_pmd_range(tlb, vma, pud, addr, next, 74251c6f666SRobin Holt zap_work, details); 74351c6f666SRobin Holt } while (pud++, addr = next, (addr != end && *zap_work > 0)); 74451c6f666SRobin Holt 74551c6f666SRobin Holt return addr; 7461da177e4SLinus Torvalds } 7471da177e4SLinus Torvalds 74851c6f666SRobin Holt static unsigned long unmap_page_range(struct mmu_gather *tlb, 74951c6f666SRobin Holt struct vm_area_struct *vma, 7501da177e4SLinus Torvalds unsigned long addr, unsigned long end, 75151c6f666SRobin Holt long *zap_work, struct zap_details *details) 7521da177e4SLinus Torvalds { 7531da177e4SLinus Torvalds pgd_t *pgd; 7541da177e4SLinus Torvalds unsigned long next; 7551da177e4SLinus Torvalds 7561da177e4SLinus Torvalds if (details && !details->check_mapping && !details->nonlinear_vma) 7571da177e4SLinus Torvalds details = NULL; 7581da177e4SLinus Torvalds 7591da177e4SLinus Torvalds BUG_ON(addr >= end); 7601da177e4SLinus Torvalds tlb_start_vma(tlb, vma); 7611da177e4SLinus Torvalds pgd = pgd_offset(vma->vm_mm, addr); 7621da177e4SLinus Torvalds do { 7631da177e4SLinus Torvalds next = pgd_addr_end(addr, end); 76451c6f666SRobin Holt if (pgd_none_or_clear_bad(pgd)) { 76551c6f666SRobin Holt (*zap_work)--; 7661da177e4SLinus Torvalds continue; 76751c6f666SRobin Holt } 76851c6f666SRobin Holt next = zap_pud_range(tlb, vma, pgd, addr, next, 76951c6f666SRobin Holt zap_work, details); 77051c6f666SRobin Holt } while (pgd++, addr = next, (addr != end && *zap_work > 0)); 7711da177e4SLinus Torvalds tlb_end_vma(tlb, vma); 77251c6f666SRobin Holt 77351c6f666SRobin Holt return addr; 7741da177e4SLinus Torvalds } 7751da177e4SLinus Torvalds 7761da177e4SLinus Torvalds #ifdef CONFIG_PREEMPT 7771da177e4SLinus Torvalds # define ZAP_BLOCK_SIZE (8 * PAGE_SIZE) 7781da177e4SLinus Torvalds #else 7791da177e4SLinus Torvalds /* No preempt: go for improved straight-line efficiency */ 7801da177e4SLinus Torvalds # define ZAP_BLOCK_SIZE (1024 * PAGE_SIZE) 7811da177e4SLinus Torvalds #endif 7821da177e4SLinus Torvalds 7831da177e4SLinus Torvalds /** 7841da177e4SLinus Torvalds * unmap_vmas - unmap a range of memory covered by a list of vma's 7851da177e4SLinus Torvalds * @tlbp: address of the caller's struct mmu_gather 7861da177e4SLinus Torvalds * @vma: the starting vma 7871da177e4SLinus Torvalds * @start_addr: virtual address at which to start unmapping 7881da177e4SLinus Torvalds * @end_addr: virtual address at which to end unmapping 7891da177e4SLinus Torvalds * @nr_accounted: Place number of unmapped pages in vm-accountable vma's here 7901da177e4SLinus Torvalds * @details: details of nonlinear truncation or shared cache invalidation 7911da177e4SLinus Torvalds * 792ee39b37bSHugh Dickins * Returns the end address of the unmapping (restart addr if interrupted). 7931da177e4SLinus Torvalds * 794508034a3SHugh Dickins * Unmap all pages in the vma list. 7951da177e4SLinus Torvalds * 796508034a3SHugh Dickins * We aim to not hold locks for too long (for scheduling latency reasons). 797508034a3SHugh Dickins * So zap pages in ZAP_BLOCK_SIZE bytecounts. This means we need to 7981da177e4SLinus Torvalds * return the ending mmu_gather to the caller. 7991da177e4SLinus Torvalds * 8001da177e4SLinus Torvalds * Only addresses between `start' and `end' will be unmapped. 8011da177e4SLinus Torvalds * 8021da177e4SLinus Torvalds * The VMA list must be sorted in ascending virtual address order. 8031da177e4SLinus Torvalds * 8041da177e4SLinus Torvalds * unmap_vmas() assumes that the caller will flush the whole unmapped address 8051da177e4SLinus Torvalds * range after unmap_vmas() returns. So the only responsibility here is to 8061da177e4SLinus Torvalds * ensure that any thus-far unmapped pages are flushed before unmap_vmas() 8071da177e4SLinus Torvalds * drops the lock and schedules. 8081da177e4SLinus Torvalds */ 809508034a3SHugh Dickins unsigned long unmap_vmas(struct mmu_gather **tlbp, 8101da177e4SLinus Torvalds struct vm_area_struct *vma, unsigned long start_addr, 8111da177e4SLinus Torvalds unsigned long end_addr, unsigned long *nr_accounted, 8121da177e4SLinus Torvalds struct zap_details *details) 8131da177e4SLinus Torvalds { 81451c6f666SRobin Holt long zap_work = ZAP_BLOCK_SIZE; 8151da177e4SLinus Torvalds unsigned long tlb_start = 0; /* For tlb_finish_mmu */ 8161da177e4SLinus Torvalds int tlb_start_valid = 0; 817ee39b37bSHugh Dickins unsigned long start = start_addr; 8181da177e4SLinus Torvalds spinlock_t *i_mmap_lock = details? details->i_mmap_lock: NULL; 8194d6ddfa9SHugh Dickins int fullmm = (*tlbp)->fullmm; 8201da177e4SLinus Torvalds 8211da177e4SLinus Torvalds for ( ; vma && vma->vm_start < end_addr; vma = vma->vm_next) { 8221da177e4SLinus Torvalds unsigned long end; 8231da177e4SLinus Torvalds 8241da177e4SLinus Torvalds start = max(vma->vm_start, start_addr); 8251da177e4SLinus Torvalds if (start >= vma->vm_end) 8261da177e4SLinus Torvalds continue; 8271da177e4SLinus Torvalds end = min(vma->vm_end, end_addr); 8281da177e4SLinus Torvalds if (end <= vma->vm_start) 8291da177e4SLinus Torvalds continue; 8301da177e4SLinus Torvalds 8311da177e4SLinus Torvalds if (vma->vm_flags & VM_ACCOUNT) 8321da177e4SLinus Torvalds *nr_accounted += (end - start) >> PAGE_SHIFT; 8331da177e4SLinus Torvalds 8341da177e4SLinus Torvalds while (start != end) { 8351da177e4SLinus Torvalds if (!tlb_start_valid) { 8361da177e4SLinus Torvalds tlb_start = start; 8371da177e4SLinus Torvalds tlb_start_valid = 1; 8381da177e4SLinus Torvalds } 8391da177e4SLinus Torvalds 84051c6f666SRobin Holt if (unlikely(is_vm_hugetlb_page(vma))) { 8411da177e4SLinus Torvalds unmap_hugepage_range(vma, start, end); 84251c6f666SRobin Holt zap_work -= (end - start) / 84351c6f666SRobin Holt (HPAGE_SIZE / PAGE_SIZE); 84451c6f666SRobin Holt start = end; 84551c6f666SRobin Holt } else 84651c6f666SRobin Holt start = unmap_page_range(*tlbp, vma, 84751c6f666SRobin Holt start, end, &zap_work, details); 8481da177e4SLinus Torvalds 84951c6f666SRobin Holt if (zap_work > 0) { 85051c6f666SRobin Holt BUG_ON(start != end); 85151c6f666SRobin Holt break; 85251c6f666SRobin Holt } 8531da177e4SLinus Torvalds 8541da177e4SLinus Torvalds tlb_finish_mmu(*tlbp, tlb_start, start); 8551da177e4SLinus Torvalds 8561da177e4SLinus Torvalds if (need_resched() || 8571da177e4SLinus Torvalds (i_mmap_lock && need_lockbreak(i_mmap_lock))) { 8581da177e4SLinus Torvalds if (i_mmap_lock) { 859508034a3SHugh Dickins *tlbp = NULL; 8601da177e4SLinus Torvalds goto out; 8611da177e4SLinus Torvalds } 8621da177e4SLinus Torvalds cond_resched(); 8631da177e4SLinus Torvalds } 8641da177e4SLinus Torvalds 865508034a3SHugh Dickins *tlbp = tlb_gather_mmu(vma->vm_mm, fullmm); 8661da177e4SLinus Torvalds tlb_start_valid = 0; 86751c6f666SRobin Holt zap_work = ZAP_BLOCK_SIZE; 8681da177e4SLinus Torvalds } 8691da177e4SLinus Torvalds } 8701da177e4SLinus Torvalds out: 871ee39b37bSHugh Dickins return start; /* which is now the end (or restart) address */ 8721da177e4SLinus Torvalds } 8731da177e4SLinus Torvalds 8741da177e4SLinus Torvalds /** 8751da177e4SLinus Torvalds * zap_page_range - remove user pages in a given range 8761da177e4SLinus Torvalds * @vma: vm_area_struct holding the applicable pages 8771da177e4SLinus Torvalds * @address: starting address of pages to zap 8781da177e4SLinus Torvalds * @size: number of bytes to zap 8791da177e4SLinus Torvalds * @details: details of nonlinear truncation or shared cache invalidation 8801da177e4SLinus Torvalds */ 881ee39b37bSHugh Dickins unsigned long zap_page_range(struct vm_area_struct *vma, unsigned long address, 8821da177e4SLinus Torvalds unsigned long size, struct zap_details *details) 8831da177e4SLinus Torvalds { 8841da177e4SLinus Torvalds struct mm_struct *mm = vma->vm_mm; 8851da177e4SLinus Torvalds struct mmu_gather *tlb; 8861da177e4SLinus Torvalds unsigned long end = address + size; 8871da177e4SLinus Torvalds unsigned long nr_accounted = 0; 8881da177e4SLinus Torvalds 8891da177e4SLinus Torvalds lru_add_drain(); 8901da177e4SLinus Torvalds tlb = tlb_gather_mmu(mm, 0); 891365e9c87SHugh Dickins update_hiwater_rss(mm); 892508034a3SHugh Dickins end = unmap_vmas(&tlb, vma, address, end, &nr_accounted, details); 893508034a3SHugh Dickins if (tlb) 8948f4f8c16SHugh Dickins tlb_finish_mmu(tlb, address, end); 895ee39b37bSHugh Dickins return end; 8961da177e4SLinus Torvalds } 8971da177e4SLinus Torvalds 8981da177e4SLinus Torvalds /* 8991da177e4SLinus Torvalds * Do a quick page-table lookup for a single page. 9001da177e4SLinus Torvalds */ 9016aab341eSLinus Torvalds struct page *follow_page(struct vm_area_struct *vma, unsigned long address, 902deceb6cdSHugh Dickins unsigned int flags) 9031da177e4SLinus Torvalds { 9041da177e4SLinus Torvalds pgd_t *pgd; 9051da177e4SLinus Torvalds pud_t *pud; 9061da177e4SLinus Torvalds pmd_t *pmd; 9071da177e4SLinus Torvalds pte_t *ptep, pte; 908deceb6cdSHugh Dickins spinlock_t *ptl; 9091da177e4SLinus Torvalds struct page *page; 9106aab341eSLinus Torvalds struct mm_struct *mm = vma->vm_mm; 9111da177e4SLinus Torvalds 912deceb6cdSHugh Dickins page = follow_huge_addr(mm, address, flags & FOLL_WRITE); 913deceb6cdSHugh Dickins if (!IS_ERR(page)) { 914deceb6cdSHugh Dickins BUG_ON(flags & FOLL_GET); 915deceb6cdSHugh Dickins goto out; 916deceb6cdSHugh Dickins } 9171da177e4SLinus Torvalds 918deceb6cdSHugh Dickins page = NULL; 9191da177e4SLinus Torvalds pgd = pgd_offset(mm, address); 9201da177e4SLinus Torvalds if (pgd_none(*pgd) || unlikely(pgd_bad(*pgd))) 921deceb6cdSHugh Dickins goto no_page_table; 9221da177e4SLinus Torvalds 9231da177e4SLinus Torvalds pud = pud_offset(pgd, address); 9241da177e4SLinus Torvalds if (pud_none(*pud) || unlikely(pud_bad(*pud))) 925deceb6cdSHugh Dickins goto no_page_table; 9261da177e4SLinus Torvalds 9271da177e4SLinus Torvalds pmd = pmd_offset(pud, address); 9281da177e4SLinus Torvalds if (pmd_none(*pmd) || unlikely(pmd_bad(*pmd))) 929deceb6cdSHugh Dickins goto no_page_table; 9301da177e4SLinus Torvalds 931deceb6cdSHugh Dickins if (pmd_huge(*pmd)) { 932deceb6cdSHugh Dickins BUG_ON(flags & FOLL_GET); 933deceb6cdSHugh Dickins page = follow_huge_pmd(mm, address, pmd, flags & FOLL_WRITE); 934deceb6cdSHugh Dickins goto out; 935deceb6cdSHugh Dickins } 936deceb6cdSHugh Dickins 937deceb6cdSHugh Dickins ptep = pte_offset_map_lock(mm, pmd, address, &ptl); 9381da177e4SLinus Torvalds if (!ptep) 9391da177e4SLinus Torvalds goto out; 9401da177e4SLinus Torvalds 9411da177e4SLinus Torvalds pte = *ptep; 942deceb6cdSHugh Dickins if (!pte_present(pte)) 943deceb6cdSHugh Dickins goto unlock; 944deceb6cdSHugh Dickins if ((flags & FOLL_WRITE) && !pte_write(pte)) 945deceb6cdSHugh Dickins goto unlock; 9466aab341eSLinus Torvalds page = vm_normal_page(vma, address, pte); 9476aab341eSLinus Torvalds if (unlikely(!page)) 948deceb6cdSHugh Dickins goto unlock; 949deceb6cdSHugh Dickins 950deceb6cdSHugh Dickins if (flags & FOLL_GET) 951deceb6cdSHugh Dickins get_page(page); 952deceb6cdSHugh Dickins if (flags & FOLL_TOUCH) { 953deceb6cdSHugh Dickins if ((flags & FOLL_WRITE) && 954deceb6cdSHugh Dickins !pte_dirty(pte) && !PageDirty(page)) 955f33ea7f4SNick Piggin set_page_dirty(page); 9561da177e4SLinus Torvalds mark_page_accessed(page); 9571da177e4SLinus Torvalds } 958deceb6cdSHugh Dickins unlock: 959deceb6cdSHugh Dickins pte_unmap_unlock(ptep, ptl); 9601da177e4SLinus Torvalds out: 961deceb6cdSHugh Dickins return page; 962deceb6cdSHugh Dickins 963deceb6cdSHugh Dickins no_page_table: 964deceb6cdSHugh Dickins /* 965deceb6cdSHugh Dickins * When core dumping an enormous anonymous area that nobody 966deceb6cdSHugh Dickins * has touched so far, we don't want to allocate page tables. 967deceb6cdSHugh Dickins */ 968deceb6cdSHugh Dickins if (flags & FOLL_ANON) { 969deceb6cdSHugh Dickins page = ZERO_PAGE(address); 970deceb6cdSHugh Dickins if (flags & FOLL_GET) 971deceb6cdSHugh Dickins get_page(page); 972deceb6cdSHugh Dickins BUG_ON(flags & FOLL_WRITE); 9731da177e4SLinus Torvalds } 974deceb6cdSHugh Dickins return page; 9751da177e4SLinus Torvalds } 9761da177e4SLinus Torvalds 9771da177e4SLinus Torvalds int get_user_pages(struct task_struct *tsk, struct mm_struct *mm, 9781da177e4SLinus Torvalds unsigned long start, int len, int write, int force, 9791da177e4SLinus Torvalds struct page **pages, struct vm_area_struct **vmas) 9801da177e4SLinus Torvalds { 9811da177e4SLinus Torvalds int i; 982deceb6cdSHugh Dickins unsigned int vm_flags; 9831da177e4SLinus Torvalds 9841da177e4SLinus Torvalds /* 9851da177e4SLinus Torvalds * Require read or write permissions. 9861da177e4SLinus Torvalds * If 'force' is set, we only require the "MAY" flags. 9871da177e4SLinus Torvalds */ 988deceb6cdSHugh Dickins vm_flags = write ? (VM_WRITE | VM_MAYWRITE) : (VM_READ | VM_MAYREAD); 989deceb6cdSHugh Dickins vm_flags &= force ? (VM_MAYREAD | VM_MAYWRITE) : (VM_READ | VM_WRITE); 9901da177e4SLinus Torvalds i = 0; 9911da177e4SLinus Torvalds 9921da177e4SLinus Torvalds do { 9931da177e4SLinus Torvalds struct vm_area_struct *vma; 994deceb6cdSHugh Dickins unsigned int foll_flags; 9951da177e4SLinus Torvalds 9961da177e4SLinus Torvalds vma = find_extend_vma(mm, start); 9971da177e4SLinus Torvalds if (!vma && in_gate_area(tsk, start)) { 9981da177e4SLinus Torvalds unsigned long pg = start & PAGE_MASK; 9991da177e4SLinus Torvalds struct vm_area_struct *gate_vma = get_gate_vma(tsk); 10001da177e4SLinus Torvalds pgd_t *pgd; 10011da177e4SLinus Torvalds pud_t *pud; 10021da177e4SLinus Torvalds pmd_t *pmd; 10031da177e4SLinus Torvalds pte_t *pte; 10041da177e4SLinus Torvalds if (write) /* user gate pages are read-only */ 10051da177e4SLinus Torvalds return i ? : -EFAULT; 10061da177e4SLinus Torvalds if (pg > TASK_SIZE) 10071da177e4SLinus Torvalds pgd = pgd_offset_k(pg); 10081da177e4SLinus Torvalds else 10091da177e4SLinus Torvalds pgd = pgd_offset_gate(mm, pg); 10101da177e4SLinus Torvalds BUG_ON(pgd_none(*pgd)); 10111da177e4SLinus Torvalds pud = pud_offset(pgd, pg); 10121da177e4SLinus Torvalds BUG_ON(pud_none(*pud)); 10131da177e4SLinus Torvalds pmd = pmd_offset(pud, pg); 1014690dbe1cSHugh Dickins if (pmd_none(*pmd)) 1015690dbe1cSHugh Dickins return i ? : -EFAULT; 10161da177e4SLinus Torvalds pte = pte_offset_map(pmd, pg); 1017690dbe1cSHugh Dickins if (pte_none(*pte)) { 1018690dbe1cSHugh Dickins pte_unmap(pte); 1019690dbe1cSHugh Dickins return i ? : -EFAULT; 1020690dbe1cSHugh Dickins } 10211da177e4SLinus Torvalds if (pages) { 1022fa2a455bSNick Piggin struct page *page = vm_normal_page(gate_vma, start, *pte); 10236aab341eSLinus Torvalds pages[i] = page; 10246aab341eSLinus Torvalds if (page) 10256aab341eSLinus Torvalds get_page(page); 10261da177e4SLinus Torvalds } 10271da177e4SLinus Torvalds pte_unmap(pte); 10281da177e4SLinus Torvalds if (vmas) 10291da177e4SLinus Torvalds vmas[i] = gate_vma; 10301da177e4SLinus Torvalds i++; 10311da177e4SLinus Torvalds start += PAGE_SIZE; 10321da177e4SLinus Torvalds len--; 10331da177e4SLinus Torvalds continue; 10341da177e4SLinus Torvalds } 10351da177e4SLinus Torvalds 10361ff80389SLinus Torvalds if (!vma || (vma->vm_flags & (VM_IO | VM_PFNMAP)) 1037deceb6cdSHugh Dickins || !(vm_flags & vma->vm_flags)) 10381da177e4SLinus Torvalds return i ? : -EFAULT; 10391da177e4SLinus Torvalds 10401da177e4SLinus Torvalds if (is_vm_hugetlb_page(vma)) { 10411da177e4SLinus Torvalds i = follow_hugetlb_page(mm, vma, pages, vmas, 10421da177e4SLinus Torvalds &start, &len, i); 10431da177e4SLinus Torvalds continue; 10441da177e4SLinus Torvalds } 1045deceb6cdSHugh Dickins 1046deceb6cdSHugh Dickins foll_flags = FOLL_TOUCH; 1047deceb6cdSHugh Dickins if (pages) 1048deceb6cdSHugh Dickins foll_flags |= FOLL_GET; 1049deceb6cdSHugh Dickins if (!write && !(vma->vm_flags & VM_LOCKED) && 105054cb8821SNick Piggin (!vma->vm_ops || (!vma->vm_ops->nopage && 105154cb8821SNick Piggin !vma->vm_ops->fault))) 1052deceb6cdSHugh Dickins foll_flags |= FOLL_ANON; 1053deceb6cdSHugh Dickins 10541da177e4SLinus Torvalds do { 105508ef4729SHugh Dickins struct page *page; 10561da177e4SLinus Torvalds 1057462e00ccSEthan Solomita /* 1058462e00ccSEthan Solomita * If tsk is ooming, cut off its access to large memory 1059462e00ccSEthan Solomita * allocations. It has a pending SIGKILL, but it can't 1060462e00ccSEthan Solomita * be processed until returning to user space. 1061462e00ccSEthan Solomita */ 1062462e00ccSEthan Solomita if (unlikely(test_tsk_thread_flag(tsk, TIF_MEMDIE))) 1063462e00ccSEthan Solomita return -ENOMEM; 1064462e00ccSEthan Solomita 1065deceb6cdSHugh Dickins if (write) 1066deceb6cdSHugh Dickins foll_flags |= FOLL_WRITE; 1067deceb6cdSHugh Dickins 1068deceb6cdSHugh Dickins cond_resched(); 10696aab341eSLinus Torvalds while (!(page = follow_page(vma, start, foll_flags))) { 1070a68d2ebcSLinus Torvalds int ret; 1071*83c54070SNick Piggin ret = handle_mm_fault(mm, vma, start, 1072deceb6cdSHugh Dickins foll_flags & FOLL_WRITE); 1073*83c54070SNick Piggin if (ret & VM_FAULT_ERROR) { 1074*83c54070SNick Piggin if (ret & VM_FAULT_OOM) 1075*83c54070SNick Piggin return i ? i : -ENOMEM; 1076*83c54070SNick Piggin else if (ret & VM_FAULT_SIGBUS) 1077*83c54070SNick Piggin return i ? i : -EFAULT; 1078*83c54070SNick Piggin BUG(); 1079*83c54070SNick Piggin } 1080*83c54070SNick Piggin if (ret & VM_FAULT_MAJOR) 1081*83c54070SNick Piggin tsk->maj_flt++; 1082*83c54070SNick Piggin else 1083*83c54070SNick Piggin tsk->min_flt++; 1084*83c54070SNick Piggin 1085f33ea7f4SNick Piggin /* 1086*83c54070SNick Piggin * The VM_FAULT_WRITE bit tells us that 1087*83c54070SNick Piggin * do_wp_page has broken COW when necessary, 1088*83c54070SNick Piggin * even if maybe_mkwrite decided not to set 1089*83c54070SNick Piggin * pte_write. We can thus safely do subsequent 1090*83c54070SNick Piggin * page lookups as if they were reads. 1091f33ea7f4SNick Piggin */ 1092a68d2ebcSLinus Torvalds if (ret & VM_FAULT_WRITE) 1093deceb6cdSHugh Dickins foll_flags &= ~FOLL_WRITE; 1094a68d2ebcSLinus Torvalds 10957f7bbbe5SBenjamin Herrenschmidt cond_resched(); 10961da177e4SLinus Torvalds } 10971da177e4SLinus Torvalds if (pages) { 109808ef4729SHugh Dickins pages[i] = page; 109903beb076SJames Bottomley 1100a6f36be3SRussell King flush_anon_page(vma, page, start); 110108ef4729SHugh Dickins flush_dcache_page(page); 11021da177e4SLinus Torvalds } 11031da177e4SLinus Torvalds if (vmas) 11041da177e4SLinus Torvalds vmas[i] = vma; 11051da177e4SLinus Torvalds i++; 11061da177e4SLinus Torvalds start += PAGE_SIZE; 11071da177e4SLinus Torvalds len--; 11081da177e4SLinus Torvalds } while (len && start < vma->vm_end); 11091da177e4SLinus Torvalds } while (len); 11101da177e4SLinus Torvalds return i; 11111da177e4SLinus Torvalds } 11121da177e4SLinus Torvalds EXPORT_SYMBOL(get_user_pages); 11131da177e4SLinus Torvalds 11141da177e4SLinus Torvalds static int zeromap_pte_range(struct mm_struct *mm, pmd_t *pmd, 11151da177e4SLinus Torvalds unsigned long addr, unsigned long end, pgprot_t prot) 11161da177e4SLinus Torvalds { 11171da177e4SLinus Torvalds pte_t *pte; 1118c74df32cSHugh Dickins spinlock_t *ptl; 11195fcf7bb7SHugh Dickins int err = 0; 11201da177e4SLinus Torvalds 1121c74df32cSHugh Dickins pte = pte_alloc_map_lock(mm, pmd, addr, &ptl); 11221da177e4SLinus Torvalds if (!pte) 11235fcf7bb7SHugh Dickins return -EAGAIN; 11246606c3e0SZachary Amsden arch_enter_lazy_mmu_mode(); 11251da177e4SLinus Torvalds do { 1126b5810039SNick Piggin struct page *page = ZERO_PAGE(addr); 1127b5810039SNick Piggin pte_t zero_pte = pte_wrprotect(mk_pte(page, prot)); 11285fcf7bb7SHugh Dickins 11295fcf7bb7SHugh Dickins if (unlikely(!pte_none(*pte))) { 11305fcf7bb7SHugh Dickins err = -EEXIST; 11315fcf7bb7SHugh Dickins pte++; 11325fcf7bb7SHugh Dickins break; 11335fcf7bb7SHugh Dickins } 1134b5810039SNick Piggin page_cache_get(page); 1135b5810039SNick Piggin page_add_file_rmap(page); 1136b5810039SNick Piggin inc_mm_counter(mm, file_rss); 11371da177e4SLinus Torvalds set_pte_at(mm, addr, pte, zero_pte); 11381da177e4SLinus Torvalds } while (pte++, addr += PAGE_SIZE, addr != end); 11396606c3e0SZachary Amsden arch_leave_lazy_mmu_mode(); 1140c74df32cSHugh Dickins pte_unmap_unlock(pte - 1, ptl); 11415fcf7bb7SHugh Dickins return err; 11421da177e4SLinus Torvalds } 11431da177e4SLinus Torvalds 11441da177e4SLinus Torvalds static inline int zeromap_pmd_range(struct mm_struct *mm, pud_t *pud, 11451da177e4SLinus Torvalds unsigned long addr, unsigned long end, pgprot_t prot) 11461da177e4SLinus Torvalds { 11471da177e4SLinus Torvalds pmd_t *pmd; 11481da177e4SLinus Torvalds unsigned long next; 11495fcf7bb7SHugh Dickins int err; 11501da177e4SLinus Torvalds 11511da177e4SLinus Torvalds pmd = pmd_alloc(mm, pud, addr); 11521da177e4SLinus Torvalds if (!pmd) 11535fcf7bb7SHugh Dickins return -EAGAIN; 11541da177e4SLinus Torvalds do { 11551da177e4SLinus Torvalds next = pmd_addr_end(addr, end); 11565fcf7bb7SHugh Dickins err = zeromap_pte_range(mm, pmd, addr, next, prot); 11575fcf7bb7SHugh Dickins if (err) 11585fcf7bb7SHugh Dickins break; 11591da177e4SLinus Torvalds } while (pmd++, addr = next, addr != end); 11605fcf7bb7SHugh Dickins return err; 11611da177e4SLinus Torvalds } 11621da177e4SLinus Torvalds 11631da177e4SLinus Torvalds static inline int zeromap_pud_range(struct mm_struct *mm, pgd_t *pgd, 11641da177e4SLinus Torvalds unsigned long addr, unsigned long end, pgprot_t prot) 11651da177e4SLinus Torvalds { 11661da177e4SLinus Torvalds pud_t *pud; 11671da177e4SLinus Torvalds unsigned long next; 11685fcf7bb7SHugh Dickins int err; 11691da177e4SLinus Torvalds 11701da177e4SLinus Torvalds pud = pud_alloc(mm, pgd, addr); 11711da177e4SLinus Torvalds if (!pud) 11725fcf7bb7SHugh Dickins return -EAGAIN; 11731da177e4SLinus Torvalds do { 11741da177e4SLinus Torvalds next = pud_addr_end(addr, end); 11755fcf7bb7SHugh Dickins err = zeromap_pmd_range(mm, pud, addr, next, prot); 11765fcf7bb7SHugh Dickins if (err) 11775fcf7bb7SHugh Dickins break; 11781da177e4SLinus Torvalds } while (pud++, addr = next, addr != end); 11795fcf7bb7SHugh Dickins return err; 11801da177e4SLinus Torvalds } 11811da177e4SLinus Torvalds 11821da177e4SLinus Torvalds int zeromap_page_range(struct vm_area_struct *vma, 11831da177e4SLinus Torvalds unsigned long addr, unsigned long size, pgprot_t prot) 11841da177e4SLinus Torvalds { 11851da177e4SLinus Torvalds pgd_t *pgd; 11861da177e4SLinus Torvalds unsigned long next; 11871da177e4SLinus Torvalds unsigned long end = addr + size; 11881da177e4SLinus Torvalds struct mm_struct *mm = vma->vm_mm; 11891da177e4SLinus Torvalds int err; 11901da177e4SLinus Torvalds 11911da177e4SLinus Torvalds BUG_ON(addr >= end); 11921da177e4SLinus Torvalds pgd = pgd_offset(mm, addr); 11931da177e4SLinus Torvalds flush_cache_range(vma, addr, end); 11941da177e4SLinus Torvalds do { 11951da177e4SLinus Torvalds next = pgd_addr_end(addr, end); 11961da177e4SLinus Torvalds err = zeromap_pud_range(mm, pgd, addr, next, prot); 11971da177e4SLinus Torvalds if (err) 11981da177e4SLinus Torvalds break; 11991da177e4SLinus Torvalds } while (pgd++, addr = next, addr != end); 12001da177e4SLinus Torvalds return err; 12011da177e4SLinus Torvalds } 12021da177e4SLinus Torvalds 120349c91fb0STrond Myklebust pte_t * fastcall get_locked_pte(struct mm_struct *mm, unsigned long addr, spinlock_t **ptl) 1204c9cfcddfSLinus Torvalds { 1205c9cfcddfSLinus Torvalds pgd_t * pgd = pgd_offset(mm, addr); 1206c9cfcddfSLinus Torvalds pud_t * pud = pud_alloc(mm, pgd, addr); 1207c9cfcddfSLinus Torvalds if (pud) { 120849c91fb0STrond Myklebust pmd_t * pmd = pmd_alloc(mm, pud, addr); 1209c9cfcddfSLinus Torvalds if (pmd) 1210c9cfcddfSLinus Torvalds return pte_alloc_map_lock(mm, pmd, addr, ptl); 1211c9cfcddfSLinus Torvalds } 1212c9cfcddfSLinus Torvalds return NULL; 1213c9cfcddfSLinus Torvalds } 1214c9cfcddfSLinus Torvalds 12151da177e4SLinus Torvalds /* 1216238f58d8SLinus Torvalds * This is the old fallback for page remapping. 1217238f58d8SLinus Torvalds * 1218238f58d8SLinus Torvalds * For historical reasons, it only allows reserved pages. Only 1219238f58d8SLinus Torvalds * old drivers should use this, and they needed to mark their 1220238f58d8SLinus Torvalds * pages reserved for the old functions anyway. 1221238f58d8SLinus Torvalds */ 1222238f58d8SLinus Torvalds static int insert_page(struct mm_struct *mm, unsigned long addr, struct page *page, pgprot_t prot) 1223238f58d8SLinus Torvalds { 1224238f58d8SLinus Torvalds int retval; 1225238f58d8SLinus Torvalds pte_t *pte; 1226238f58d8SLinus Torvalds spinlock_t *ptl; 1227238f58d8SLinus Torvalds 1228238f58d8SLinus Torvalds retval = -EINVAL; 1229a145dd41SLinus Torvalds if (PageAnon(page)) 1230238f58d8SLinus Torvalds goto out; 1231238f58d8SLinus Torvalds retval = -ENOMEM; 1232238f58d8SLinus Torvalds flush_dcache_page(page); 1233c9cfcddfSLinus Torvalds pte = get_locked_pte(mm, addr, &ptl); 1234238f58d8SLinus Torvalds if (!pte) 1235238f58d8SLinus Torvalds goto out; 1236238f58d8SLinus Torvalds retval = -EBUSY; 1237238f58d8SLinus Torvalds if (!pte_none(*pte)) 1238238f58d8SLinus Torvalds goto out_unlock; 1239238f58d8SLinus Torvalds 1240238f58d8SLinus Torvalds /* Ok, finally just insert the thing.. */ 1241238f58d8SLinus Torvalds get_page(page); 1242238f58d8SLinus Torvalds inc_mm_counter(mm, file_rss); 1243238f58d8SLinus Torvalds page_add_file_rmap(page); 1244238f58d8SLinus Torvalds set_pte_at(mm, addr, pte, mk_pte(page, prot)); 1245238f58d8SLinus Torvalds 1246238f58d8SLinus Torvalds retval = 0; 1247238f58d8SLinus Torvalds out_unlock: 1248238f58d8SLinus Torvalds pte_unmap_unlock(pte, ptl); 1249238f58d8SLinus Torvalds out: 1250238f58d8SLinus Torvalds return retval; 1251238f58d8SLinus Torvalds } 1252238f58d8SLinus Torvalds 1253bfa5bf6dSRolf Eike Beer /** 1254bfa5bf6dSRolf Eike Beer * vm_insert_page - insert single page into user vma 1255bfa5bf6dSRolf Eike Beer * @vma: user vma to map to 1256bfa5bf6dSRolf Eike Beer * @addr: target user address of this page 1257bfa5bf6dSRolf Eike Beer * @page: source kernel page 1258bfa5bf6dSRolf Eike Beer * 1259a145dd41SLinus Torvalds * This allows drivers to insert individual pages they've allocated 1260a145dd41SLinus Torvalds * into a user vma. 1261a145dd41SLinus Torvalds * 1262a145dd41SLinus Torvalds * The page has to be a nice clean _individual_ kernel allocation. 1263a145dd41SLinus Torvalds * If you allocate a compound page, you need to have marked it as 1264a145dd41SLinus Torvalds * such (__GFP_COMP), or manually just split the page up yourself 12658dfcc9baSNick Piggin * (see split_page()). 1266a145dd41SLinus Torvalds * 1267a145dd41SLinus Torvalds * NOTE! Traditionally this was done with "remap_pfn_range()" which 1268a145dd41SLinus Torvalds * took an arbitrary page protection parameter. This doesn't allow 1269a145dd41SLinus Torvalds * that. Your vma protection will have to be set up correctly, which 1270a145dd41SLinus Torvalds * means that if you want a shared writable mapping, you'd better 1271a145dd41SLinus Torvalds * ask for a shared writable mapping! 1272a145dd41SLinus Torvalds * 1273a145dd41SLinus Torvalds * The page does not need to be reserved. 1274a145dd41SLinus Torvalds */ 1275a145dd41SLinus Torvalds int vm_insert_page(struct vm_area_struct *vma, unsigned long addr, struct page *page) 1276a145dd41SLinus Torvalds { 1277a145dd41SLinus Torvalds if (addr < vma->vm_start || addr >= vma->vm_end) 1278a145dd41SLinus Torvalds return -EFAULT; 1279a145dd41SLinus Torvalds if (!page_count(page)) 1280a145dd41SLinus Torvalds return -EINVAL; 12814d7672b4SLinus Torvalds vma->vm_flags |= VM_INSERTPAGE; 1282a145dd41SLinus Torvalds return insert_page(vma->vm_mm, addr, page, vma->vm_page_prot); 1283a145dd41SLinus Torvalds } 1284e3c3374fSLinus Torvalds EXPORT_SYMBOL(vm_insert_page); 1285a145dd41SLinus Torvalds 1286e0dc0d8fSNick Piggin /** 1287e0dc0d8fSNick Piggin * vm_insert_pfn - insert single pfn into user vma 1288e0dc0d8fSNick Piggin * @vma: user vma to map to 1289e0dc0d8fSNick Piggin * @addr: target user address of this page 1290e0dc0d8fSNick Piggin * @pfn: source kernel pfn 1291e0dc0d8fSNick Piggin * 1292e0dc0d8fSNick Piggin * Similar to vm_inert_page, this allows drivers to insert individual pages 1293e0dc0d8fSNick Piggin * they've allocated into a user vma. Same comments apply. 1294e0dc0d8fSNick Piggin * 1295e0dc0d8fSNick Piggin * This function should only be called from a vm_ops->fault handler, and 1296e0dc0d8fSNick Piggin * in that case the handler should return NULL. 1297e0dc0d8fSNick Piggin */ 1298e0dc0d8fSNick Piggin int vm_insert_pfn(struct vm_area_struct *vma, unsigned long addr, 1299e0dc0d8fSNick Piggin unsigned long pfn) 1300e0dc0d8fSNick Piggin { 1301e0dc0d8fSNick Piggin struct mm_struct *mm = vma->vm_mm; 1302e0dc0d8fSNick Piggin int retval; 1303e0dc0d8fSNick Piggin pte_t *pte, entry; 1304e0dc0d8fSNick Piggin spinlock_t *ptl; 1305e0dc0d8fSNick Piggin 1306e0dc0d8fSNick Piggin BUG_ON(!(vma->vm_flags & VM_PFNMAP)); 1307e0dc0d8fSNick Piggin BUG_ON(is_cow_mapping(vma->vm_flags)); 1308e0dc0d8fSNick Piggin 1309e0dc0d8fSNick Piggin retval = -ENOMEM; 1310e0dc0d8fSNick Piggin pte = get_locked_pte(mm, addr, &ptl); 1311e0dc0d8fSNick Piggin if (!pte) 1312e0dc0d8fSNick Piggin goto out; 1313e0dc0d8fSNick Piggin retval = -EBUSY; 1314e0dc0d8fSNick Piggin if (!pte_none(*pte)) 1315e0dc0d8fSNick Piggin goto out_unlock; 1316e0dc0d8fSNick Piggin 1317e0dc0d8fSNick Piggin /* Ok, finally just insert the thing.. */ 1318e0dc0d8fSNick Piggin entry = pfn_pte(pfn, vma->vm_page_prot); 1319e0dc0d8fSNick Piggin set_pte_at(mm, addr, pte, entry); 1320e0dc0d8fSNick Piggin update_mmu_cache(vma, addr, entry); 1321e0dc0d8fSNick Piggin 1322e0dc0d8fSNick Piggin retval = 0; 1323e0dc0d8fSNick Piggin out_unlock: 1324e0dc0d8fSNick Piggin pte_unmap_unlock(pte, ptl); 1325e0dc0d8fSNick Piggin 1326e0dc0d8fSNick Piggin out: 1327e0dc0d8fSNick Piggin return retval; 1328e0dc0d8fSNick Piggin } 1329e0dc0d8fSNick Piggin EXPORT_SYMBOL(vm_insert_pfn); 1330e0dc0d8fSNick Piggin 1331a145dd41SLinus Torvalds /* 13321da177e4SLinus Torvalds * maps a range of physical memory into the requested pages. the old 13331da177e4SLinus Torvalds * mappings are removed. any references to nonexistent pages results 13341da177e4SLinus Torvalds * in null mappings (currently treated as "copy-on-access") 13351da177e4SLinus Torvalds */ 13361da177e4SLinus Torvalds static int remap_pte_range(struct mm_struct *mm, pmd_t *pmd, 13371da177e4SLinus Torvalds unsigned long addr, unsigned long end, 13381da177e4SLinus Torvalds unsigned long pfn, pgprot_t prot) 13391da177e4SLinus Torvalds { 13401da177e4SLinus Torvalds pte_t *pte; 1341c74df32cSHugh Dickins spinlock_t *ptl; 13421da177e4SLinus Torvalds 1343c74df32cSHugh Dickins pte = pte_alloc_map_lock(mm, pmd, addr, &ptl); 13441da177e4SLinus Torvalds if (!pte) 13451da177e4SLinus Torvalds return -ENOMEM; 13466606c3e0SZachary Amsden arch_enter_lazy_mmu_mode(); 13471da177e4SLinus Torvalds do { 13481da177e4SLinus Torvalds BUG_ON(!pte_none(*pte)); 13491da177e4SLinus Torvalds set_pte_at(mm, addr, pte, pfn_pte(pfn, prot)); 13501da177e4SLinus Torvalds pfn++; 13511da177e4SLinus Torvalds } while (pte++, addr += PAGE_SIZE, addr != end); 13526606c3e0SZachary Amsden arch_leave_lazy_mmu_mode(); 1353c74df32cSHugh Dickins pte_unmap_unlock(pte - 1, ptl); 13541da177e4SLinus Torvalds return 0; 13551da177e4SLinus Torvalds } 13561da177e4SLinus Torvalds 13571da177e4SLinus Torvalds static inline int remap_pmd_range(struct mm_struct *mm, pud_t *pud, 13581da177e4SLinus Torvalds unsigned long addr, unsigned long end, 13591da177e4SLinus Torvalds unsigned long pfn, pgprot_t prot) 13601da177e4SLinus Torvalds { 13611da177e4SLinus Torvalds pmd_t *pmd; 13621da177e4SLinus Torvalds unsigned long next; 13631da177e4SLinus Torvalds 13641da177e4SLinus Torvalds pfn -= addr >> PAGE_SHIFT; 13651da177e4SLinus Torvalds pmd = pmd_alloc(mm, pud, addr); 13661da177e4SLinus Torvalds if (!pmd) 13671da177e4SLinus Torvalds return -ENOMEM; 13681da177e4SLinus Torvalds do { 13691da177e4SLinus Torvalds next = pmd_addr_end(addr, end); 13701da177e4SLinus Torvalds if (remap_pte_range(mm, pmd, addr, next, 13711da177e4SLinus Torvalds pfn + (addr >> PAGE_SHIFT), prot)) 13721da177e4SLinus Torvalds return -ENOMEM; 13731da177e4SLinus Torvalds } while (pmd++, addr = next, addr != end); 13741da177e4SLinus Torvalds return 0; 13751da177e4SLinus Torvalds } 13761da177e4SLinus Torvalds 13771da177e4SLinus Torvalds static inline int remap_pud_range(struct mm_struct *mm, pgd_t *pgd, 13781da177e4SLinus Torvalds unsigned long addr, unsigned long end, 13791da177e4SLinus Torvalds unsigned long pfn, pgprot_t prot) 13801da177e4SLinus Torvalds { 13811da177e4SLinus Torvalds pud_t *pud; 13821da177e4SLinus Torvalds unsigned long next; 13831da177e4SLinus Torvalds 13841da177e4SLinus Torvalds pfn -= addr >> PAGE_SHIFT; 13851da177e4SLinus Torvalds pud = pud_alloc(mm, pgd, addr); 13861da177e4SLinus Torvalds if (!pud) 13871da177e4SLinus Torvalds return -ENOMEM; 13881da177e4SLinus Torvalds do { 13891da177e4SLinus Torvalds next = pud_addr_end(addr, end); 13901da177e4SLinus Torvalds if (remap_pmd_range(mm, pud, addr, next, 13911da177e4SLinus Torvalds pfn + (addr >> PAGE_SHIFT), prot)) 13921da177e4SLinus Torvalds return -ENOMEM; 13931da177e4SLinus Torvalds } while (pud++, addr = next, addr != end); 13941da177e4SLinus Torvalds return 0; 13951da177e4SLinus Torvalds } 13961da177e4SLinus Torvalds 1397bfa5bf6dSRolf Eike Beer /** 1398bfa5bf6dSRolf Eike Beer * remap_pfn_range - remap kernel memory to userspace 1399bfa5bf6dSRolf Eike Beer * @vma: user vma to map to 1400bfa5bf6dSRolf Eike Beer * @addr: target user address to start at 1401bfa5bf6dSRolf Eike Beer * @pfn: physical address of kernel memory 1402bfa5bf6dSRolf Eike Beer * @size: size of map area 1403bfa5bf6dSRolf Eike Beer * @prot: page protection flags for this mapping 1404bfa5bf6dSRolf Eike Beer * 1405bfa5bf6dSRolf Eike Beer * Note: this is only safe if the mm semaphore is held when called. 1406bfa5bf6dSRolf Eike Beer */ 14071da177e4SLinus Torvalds int remap_pfn_range(struct vm_area_struct *vma, unsigned long addr, 14081da177e4SLinus Torvalds unsigned long pfn, unsigned long size, pgprot_t prot) 14091da177e4SLinus Torvalds { 14101da177e4SLinus Torvalds pgd_t *pgd; 14111da177e4SLinus Torvalds unsigned long next; 14122d15cab8SHugh Dickins unsigned long end = addr + PAGE_ALIGN(size); 14131da177e4SLinus Torvalds struct mm_struct *mm = vma->vm_mm; 14141da177e4SLinus Torvalds int err; 14151da177e4SLinus Torvalds 14161da177e4SLinus Torvalds /* 14171da177e4SLinus Torvalds * Physically remapped pages are special. Tell the 14181da177e4SLinus Torvalds * rest of the world about it: 14191da177e4SLinus Torvalds * VM_IO tells people not to look at these pages 14201da177e4SLinus Torvalds * (accesses can have side effects). 14210b14c179SHugh Dickins * VM_RESERVED is specified all over the place, because 14220b14c179SHugh Dickins * in 2.4 it kept swapout's vma scan off this vma; but 14230b14c179SHugh Dickins * in 2.6 the LRU scan won't even find its pages, so this 14240b14c179SHugh Dickins * flag means no more than count its pages in reserved_vm, 14250b14c179SHugh Dickins * and omit it from core dump, even when VM_IO turned off. 14266aab341eSLinus Torvalds * VM_PFNMAP tells the core MM that the base pages are just 14276aab341eSLinus Torvalds * raw PFN mappings, and do not have a "struct page" associated 14286aab341eSLinus Torvalds * with them. 1429fb155c16SLinus Torvalds * 1430fb155c16SLinus Torvalds * There's a horrible special case to handle copy-on-write 1431fb155c16SLinus Torvalds * behaviour that some programs depend on. We mark the "original" 1432fb155c16SLinus Torvalds * un-COW'ed pages by matching them up with "vma->vm_pgoff". 14331da177e4SLinus Torvalds */ 143467121172SLinus Torvalds if (is_cow_mapping(vma->vm_flags)) { 1435fb155c16SLinus Torvalds if (addr != vma->vm_start || end != vma->vm_end) 14367fc7e2eeSLinus Torvalds return -EINVAL; 14376aab341eSLinus Torvalds vma->vm_pgoff = pfn; 1438fb155c16SLinus Torvalds } 1439fb155c16SLinus Torvalds 1440fb155c16SLinus Torvalds vma->vm_flags |= VM_IO | VM_RESERVED | VM_PFNMAP; 14411da177e4SLinus Torvalds 14421da177e4SLinus Torvalds BUG_ON(addr >= end); 14431da177e4SLinus Torvalds pfn -= addr >> PAGE_SHIFT; 14441da177e4SLinus Torvalds pgd = pgd_offset(mm, addr); 14451da177e4SLinus Torvalds flush_cache_range(vma, addr, end); 14461da177e4SLinus Torvalds do { 14471da177e4SLinus Torvalds next = pgd_addr_end(addr, end); 14481da177e4SLinus Torvalds err = remap_pud_range(mm, pgd, addr, next, 14491da177e4SLinus Torvalds pfn + (addr >> PAGE_SHIFT), prot); 14501da177e4SLinus Torvalds if (err) 14511da177e4SLinus Torvalds break; 14521da177e4SLinus Torvalds } while (pgd++, addr = next, addr != end); 14531da177e4SLinus Torvalds return err; 14541da177e4SLinus Torvalds } 14551da177e4SLinus Torvalds EXPORT_SYMBOL(remap_pfn_range); 14561da177e4SLinus Torvalds 1457aee16b3cSJeremy Fitzhardinge static int apply_to_pte_range(struct mm_struct *mm, pmd_t *pmd, 1458aee16b3cSJeremy Fitzhardinge unsigned long addr, unsigned long end, 1459aee16b3cSJeremy Fitzhardinge pte_fn_t fn, void *data) 1460aee16b3cSJeremy Fitzhardinge { 1461aee16b3cSJeremy Fitzhardinge pte_t *pte; 1462aee16b3cSJeremy Fitzhardinge int err; 1463aee16b3cSJeremy Fitzhardinge struct page *pmd_page; 146494909914SBorislav Petkov spinlock_t *uninitialized_var(ptl); 1465aee16b3cSJeremy Fitzhardinge 1466aee16b3cSJeremy Fitzhardinge pte = (mm == &init_mm) ? 1467aee16b3cSJeremy Fitzhardinge pte_alloc_kernel(pmd, addr) : 1468aee16b3cSJeremy Fitzhardinge pte_alloc_map_lock(mm, pmd, addr, &ptl); 1469aee16b3cSJeremy Fitzhardinge if (!pte) 1470aee16b3cSJeremy Fitzhardinge return -ENOMEM; 1471aee16b3cSJeremy Fitzhardinge 1472aee16b3cSJeremy Fitzhardinge BUG_ON(pmd_huge(*pmd)); 1473aee16b3cSJeremy Fitzhardinge 1474aee16b3cSJeremy Fitzhardinge pmd_page = pmd_page(*pmd); 1475aee16b3cSJeremy Fitzhardinge 1476aee16b3cSJeremy Fitzhardinge do { 1477aee16b3cSJeremy Fitzhardinge err = fn(pte, pmd_page, addr, data); 1478aee16b3cSJeremy Fitzhardinge if (err) 1479aee16b3cSJeremy Fitzhardinge break; 1480aee16b3cSJeremy Fitzhardinge } while (pte++, addr += PAGE_SIZE, addr != end); 1481aee16b3cSJeremy Fitzhardinge 1482aee16b3cSJeremy Fitzhardinge if (mm != &init_mm) 1483aee16b3cSJeremy Fitzhardinge pte_unmap_unlock(pte-1, ptl); 1484aee16b3cSJeremy Fitzhardinge return err; 1485aee16b3cSJeremy Fitzhardinge } 1486aee16b3cSJeremy Fitzhardinge 1487aee16b3cSJeremy Fitzhardinge static int apply_to_pmd_range(struct mm_struct *mm, pud_t *pud, 1488aee16b3cSJeremy Fitzhardinge unsigned long addr, unsigned long end, 1489aee16b3cSJeremy Fitzhardinge pte_fn_t fn, void *data) 1490aee16b3cSJeremy Fitzhardinge { 1491aee16b3cSJeremy Fitzhardinge pmd_t *pmd; 1492aee16b3cSJeremy Fitzhardinge unsigned long next; 1493aee16b3cSJeremy Fitzhardinge int err; 1494aee16b3cSJeremy Fitzhardinge 1495aee16b3cSJeremy Fitzhardinge pmd = pmd_alloc(mm, pud, addr); 1496aee16b3cSJeremy Fitzhardinge if (!pmd) 1497aee16b3cSJeremy Fitzhardinge return -ENOMEM; 1498aee16b3cSJeremy Fitzhardinge do { 1499aee16b3cSJeremy Fitzhardinge next = pmd_addr_end(addr, end); 1500aee16b3cSJeremy Fitzhardinge err = apply_to_pte_range(mm, pmd, addr, next, fn, data); 1501aee16b3cSJeremy Fitzhardinge if (err) 1502aee16b3cSJeremy Fitzhardinge break; 1503aee16b3cSJeremy Fitzhardinge } while (pmd++, addr = next, addr != end); 1504aee16b3cSJeremy Fitzhardinge return err; 1505aee16b3cSJeremy Fitzhardinge } 1506aee16b3cSJeremy Fitzhardinge 1507aee16b3cSJeremy Fitzhardinge static int apply_to_pud_range(struct mm_struct *mm, pgd_t *pgd, 1508aee16b3cSJeremy Fitzhardinge unsigned long addr, unsigned long end, 1509aee16b3cSJeremy Fitzhardinge pte_fn_t fn, void *data) 1510aee16b3cSJeremy Fitzhardinge { 1511aee16b3cSJeremy Fitzhardinge pud_t *pud; 1512aee16b3cSJeremy Fitzhardinge unsigned long next; 1513aee16b3cSJeremy Fitzhardinge int err; 1514aee16b3cSJeremy Fitzhardinge 1515aee16b3cSJeremy Fitzhardinge pud = pud_alloc(mm, pgd, addr); 1516aee16b3cSJeremy Fitzhardinge if (!pud) 1517aee16b3cSJeremy Fitzhardinge return -ENOMEM; 1518aee16b3cSJeremy Fitzhardinge do { 1519aee16b3cSJeremy Fitzhardinge next = pud_addr_end(addr, end); 1520aee16b3cSJeremy Fitzhardinge err = apply_to_pmd_range(mm, pud, addr, next, fn, data); 1521aee16b3cSJeremy Fitzhardinge if (err) 1522aee16b3cSJeremy Fitzhardinge break; 1523aee16b3cSJeremy Fitzhardinge } while (pud++, addr = next, addr != end); 1524aee16b3cSJeremy Fitzhardinge return err; 1525aee16b3cSJeremy Fitzhardinge } 1526aee16b3cSJeremy Fitzhardinge 1527aee16b3cSJeremy Fitzhardinge /* 1528aee16b3cSJeremy Fitzhardinge * Scan a region of virtual memory, filling in page tables as necessary 1529aee16b3cSJeremy Fitzhardinge * and calling a provided function on each leaf page table. 1530aee16b3cSJeremy Fitzhardinge */ 1531aee16b3cSJeremy Fitzhardinge int apply_to_page_range(struct mm_struct *mm, unsigned long addr, 1532aee16b3cSJeremy Fitzhardinge unsigned long size, pte_fn_t fn, void *data) 1533aee16b3cSJeremy Fitzhardinge { 1534aee16b3cSJeremy Fitzhardinge pgd_t *pgd; 1535aee16b3cSJeremy Fitzhardinge unsigned long next; 1536aee16b3cSJeremy Fitzhardinge unsigned long end = addr + size; 1537aee16b3cSJeremy Fitzhardinge int err; 1538aee16b3cSJeremy Fitzhardinge 1539aee16b3cSJeremy Fitzhardinge BUG_ON(addr >= end); 1540aee16b3cSJeremy Fitzhardinge pgd = pgd_offset(mm, addr); 1541aee16b3cSJeremy Fitzhardinge do { 1542aee16b3cSJeremy Fitzhardinge next = pgd_addr_end(addr, end); 1543aee16b3cSJeremy Fitzhardinge err = apply_to_pud_range(mm, pgd, addr, next, fn, data); 1544aee16b3cSJeremy Fitzhardinge if (err) 1545aee16b3cSJeremy Fitzhardinge break; 1546aee16b3cSJeremy Fitzhardinge } while (pgd++, addr = next, addr != end); 1547aee16b3cSJeremy Fitzhardinge return err; 1548aee16b3cSJeremy Fitzhardinge } 1549aee16b3cSJeremy Fitzhardinge EXPORT_SYMBOL_GPL(apply_to_page_range); 1550aee16b3cSJeremy Fitzhardinge 15511da177e4SLinus Torvalds /* 15528f4e2101SHugh Dickins * handle_pte_fault chooses page fault handler according to an entry 15538f4e2101SHugh Dickins * which was read non-atomically. Before making any commitment, on 15548f4e2101SHugh Dickins * those architectures or configurations (e.g. i386 with PAE) which 15558f4e2101SHugh Dickins * might give a mix of unmatched parts, do_swap_page and do_file_page 15568f4e2101SHugh Dickins * must check under lock before unmapping the pte and proceeding 15578f4e2101SHugh Dickins * (but do_wp_page is only called after already making such a check; 15588f4e2101SHugh Dickins * and do_anonymous_page and do_no_page can safely check later on). 15598f4e2101SHugh Dickins */ 15604c21e2f2SHugh Dickins static inline int pte_unmap_same(struct mm_struct *mm, pmd_t *pmd, 15618f4e2101SHugh Dickins pte_t *page_table, pte_t orig_pte) 15628f4e2101SHugh Dickins { 15638f4e2101SHugh Dickins int same = 1; 15648f4e2101SHugh Dickins #if defined(CONFIG_SMP) || defined(CONFIG_PREEMPT) 15658f4e2101SHugh Dickins if (sizeof(pte_t) > sizeof(unsigned long)) { 15664c21e2f2SHugh Dickins spinlock_t *ptl = pte_lockptr(mm, pmd); 15674c21e2f2SHugh Dickins spin_lock(ptl); 15688f4e2101SHugh Dickins same = pte_same(*page_table, orig_pte); 15694c21e2f2SHugh Dickins spin_unlock(ptl); 15708f4e2101SHugh Dickins } 15718f4e2101SHugh Dickins #endif 15728f4e2101SHugh Dickins pte_unmap(page_table); 15738f4e2101SHugh Dickins return same; 15748f4e2101SHugh Dickins } 15758f4e2101SHugh Dickins 15768f4e2101SHugh Dickins /* 15771da177e4SLinus Torvalds * Do pte_mkwrite, but only if the vma says VM_WRITE. We do this when 15781da177e4SLinus Torvalds * servicing faults for write access. In the normal case, do always want 15791da177e4SLinus Torvalds * pte_mkwrite. But get_user_pages can cause write faults for mappings 15801da177e4SLinus Torvalds * that do not have writing enabled, when used by access_process_vm. 15811da177e4SLinus Torvalds */ 15821da177e4SLinus Torvalds static inline pte_t maybe_mkwrite(pte_t pte, struct vm_area_struct *vma) 15831da177e4SLinus Torvalds { 15841da177e4SLinus Torvalds if (likely(vma->vm_flags & VM_WRITE)) 15851da177e4SLinus Torvalds pte = pte_mkwrite(pte); 15861da177e4SLinus Torvalds return pte; 15871da177e4SLinus Torvalds } 15881da177e4SLinus Torvalds 15899de455b2SAtsushi Nemoto static inline void cow_user_page(struct page *dst, struct page *src, unsigned long va, struct vm_area_struct *vma) 15906aab341eSLinus Torvalds { 15916aab341eSLinus Torvalds /* 15926aab341eSLinus Torvalds * If the source page was a PFN mapping, we don't have 15936aab341eSLinus Torvalds * a "struct page" for it. We do a best-effort copy by 15946aab341eSLinus Torvalds * just copying from the original user address. If that 15956aab341eSLinus Torvalds * fails, we just zero-fill it. Live with it. 15966aab341eSLinus Torvalds */ 15976aab341eSLinus Torvalds if (unlikely(!src)) { 15986aab341eSLinus Torvalds void *kaddr = kmap_atomic(dst, KM_USER0); 15995d2a2dbbSLinus Torvalds void __user *uaddr = (void __user *)(va & PAGE_MASK); 16005d2a2dbbSLinus Torvalds 16015d2a2dbbSLinus Torvalds /* 16025d2a2dbbSLinus Torvalds * This really shouldn't fail, because the page is there 16035d2a2dbbSLinus Torvalds * in the page tables. But it might just be unreadable, 16045d2a2dbbSLinus Torvalds * in which case we just give up and fill the result with 16055d2a2dbbSLinus Torvalds * zeroes. 16065d2a2dbbSLinus Torvalds */ 16075d2a2dbbSLinus Torvalds if (__copy_from_user_inatomic(kaddr, uaddr, PAGE_SIZE)) 16086aab341eSLinus Torvalds memset(kaddr, 0, PAGE_SIZE); 16096aab341eSLinus Torvalds kunmap_atomic(kaddr, KM_USER0); 1610c4ec7b0dSDmitriy Monakhov flush_dcache_page(dst); 16116aab341eSLinus Torvalds return; 16126aab341eSLinus Torvalds 16136aab341eSLinus Torvalds } 16149de455b2SAtsushi Nemoto copy_user_highpage(dst, src, va, vma); 16156aab341eSLinus Torvalds } 16166aab341eSLinus Torvalds 16171da177e4SLinus Torvalds /* 16181da177e4SLinus Torvalds * This routine handles present pages, when users try to write 16191da177e4SLinus Torvalds * to a shared page. It is done by copying the page to a new address 16201da177e4SLinus Torvalds * and decrementing the shared-page counter for the old page. 16211da177e4SLinus Torvalds * 16221da177e4SLinus Torvalds * Note that this routine assumes that the protection checks have been 16231da177e4SLinus Torvalds * done by the caller (the low-level page fault routine in most cases). 16241da177e4SLinus Torvalds * Thus we can safely just mark it writable once we've done any necessary 16251da177e4SLinus Torvalds * COW. 16261da177e4SLinus Torvalds * 16271da177e4SLinus Torvalds * We also mark the page dirty at this point even though the page will 16281da177e4SLinus Torvalds * change only once the write actually happens. This avoids a few races, 16291da177e4SLinus Torvalds * and potentially makes it more efficient. 16301da177e4SLinus Torvalds * 16318f4e2101SHugh Dickins * We enter with non-exclusive mmap_sem (to exclude vma changes, 16328f4e2101SHugh Dickins * but allow concurrent faults), with pte both mapped and locked. 16338f4e2101SHugh Dickins * We return with mmap_sem still held, but pte unmapped and unlocked. 16341da177e4SLinus Torvalds */ 16351da177e4SLinus Torvalds static int do_wp_page(struct mm_struct *mm, struct vm_area_struct *vma, 163665500d23SHugh Dickins unsigned long address, pte_t *page_table, pmd_t *pmd, 16378f4e2101SHugh Dickins spinlock_t *ptl, pte_t orig_pte) 16381da177e4SLinus Torvalds { 1639e5bbe4dfSHugh Dickins struct page *old_page, *new_page; 16401da177e4SLinus Torvalds pte_t entry; 1641*83c54070SNick Piggin int reuse = 0, ret = 0; 1642d08b3851SPeter Zijlstra struct page *dirty_page = NULL; 16431da177e4SLinus Torvalds 16446aab341eSLinus Torvalds old_page = vm_normal_page(vma, address, orig_pte); 16456aab341eSLinus Torvalds if (!old_page) 1646920fc356SHugh Dickins goto gotten; 16471da177e4SLinus Torvalds 1648d08b3851SPeter Zijlstra /* 1649ee6a6457SPeter Zijlstra * Take out anonymous pages first, anonymous shared vmas are 1650ee6a6457SPeter Zijlstra * not dirty accountable. 1651d08b3851SPeter Zijlstra */ 1652ee6a6457SPeter Zijlstra if (PageAnon(old_page)) { 1653ee6a6457SPeter Zijlstra if (!TestSetPageLocked(old_page)) { 1654ee6a6457SPeter Zijlstra reuse = can_share_swap_page(old_page); 1655ee6a6457SPeter Zijlstra unlock_page(old_page); 1656ee6a6457SPeter Zijlstra } 1657ee6a6457SPeter Zijlstra } else if (unlikely((vma->vm_flags & (VM_WRITE|VM_SHARED)) == 1658d08b3851SPeter Zijlstra (VM_WRITE|VM_SHARED))) { 1659ee6a6457SPeter Zijlstra /* 1660ee6a6457SPeter Zijlstra * Only catch write-faults on shared writable pages, 1661ee6a6457SPeter Zijlstra * read-only shared pages can get COWed by 1662ee6a6457SPeter Zijlstra * get_user_pages(.write=1, .force=1). 1663ee6a6457SPeter Zijlstra */ 16649637a5efSDavid Howells if (vma->vm_ops && vma->vm_ops->page_mkwrite) { 16659637a5efSDavid Howells /* 16669637a5efSDavid Howells * Notify the address space that the page is about to 16679637a5efSDavid Howells * become writable so that it can prohibit this or wait 16689637a5efSDavid Howells * for the page to get into an appropriate state. 16699637a5efSDavid Howells * 16709637a5efSDavid Howells * We do this without the lock held, so that it can 16719637a5efSDavid Howells * sleep if it needs to. 16729637a5efSDavid Howells */ 16739637a5efSDavid Howells page_cache_get(old_page); 16749637a5efSDavid Howells pte_unmap_unlock(page_table, ptl); 16759637a5efSDavid Howells 16769637a5efSDavid Howells if (vma->vm_ops->page_mkwrite(vma, old_page) < 0) 16779637a5efSDavid Howells goto unwritable_page; 16789637a5efSDavid Howells 16799637a5efSDavid Howells /* 16809637a5efSDavid Howells * Since we dropped the lock we need to revalidate 16819637a5efSDavid Howells * the PTE as someone else may have changed it. If 16829637a5efSDavid Howells * they did, we just return, as we can count on the 16839637a5efSDavid Howells * MMU to tell us if they didn't also make it writable. 16849637a5efSDavid Howells */ 16859637a5efSDavid Howells page_table = pte_offset_map_lock(mm, pmd, address, 16869637a5efSDavid Howells &ptl); 1687c3704cebSHugh Dickins page_cache_release(old_page); 16889637a5efSDavid Howells if (!pte_same(*page_table, orig_pte)) 16899637a5efSDavid Howells goto unlock; 16909637a5efSDavid Howells } 1691d08b3851SPeter Zijlstra dirty_page = old_page; 1692d08b3851SPeter Zijlstra get_page(dirty_page); 16939637a5efSDavid Howells reuse = 1; 16949637a5efSDavid Howells } 16959637a5efSDavid Howells 16961da177e4SLinus Torvalds if (reuse) { 1697eca35133SBen Collins flush_cache_page(vma, address, pte_pfn(orig_pte)); 169865500d23SHugh Dickins entry = pte_mkyoung(orig_pte); 169965500d23SHugh Dickins entry = maybe_mkwrite(pte_mkdirty(entry), vma); 17008dab5241SBenjamin Herrenschmidt if (ptep_set_access_flags(vma, address, page_table, entry,1)) { 17011da177e4SLinus Torvalds update_mmu_cache(vma, address, entry); 17021da177e4SLinus Torvalds lazy_mmu_prot_update(entry); 17038dab5241SBenjamin Herrenschmidt } 170465500d23SHugh Dickins ret |= VM_FAULT_WRITE; 170565500d23SHugh Dickins goto unlock; 17061da177e4SLinus Torvalds } 17071da177e4SLinus Torvalds 17081da177e4SLinus Torvalds /* 17091da177e4SLinus Torvalds * Ok, we need to copy. Oh, well.. 17101da177e4SLinus Torvalds */ 17111da177e4SLinus Torvalds page_cache_get(old_page); 1712920fc356SHugh Dickins gotten: 17138f4e2101SHugh Dickins pte_unmap_unlock(page_table, ptl); 17141da177e4SLinus Torvalds 17151da177e4SLinus Torvalds if (unlikely(anon_vma_prepare(vma))) 171665500d23SHugh Dickins goto oom; 1717e5bbe4dfSHugh Dickins if (old_page == ZERO_PAGE(address)) { 1718769848c0SMel Gorman new_page = alloc_zeroed_user_highpage_movable(vma, address); 17191da177e4SLinus Torvalds if (!new_page) 172065500d23SHugh Dickins goto oom; 17211da177e4SLinus Torvalds } else { 1722769848c0SMel Gorman new_page = alloc_page_vma(GFP_HIGHUSER_MOVABLE, vma, address); 17231da177e4SLinus Torvalds if (!new_page) 172465500d23SHugh Dickins goto oom; 17259de455b2SAtsushi Nemoto cow_user_page(new_page, old_page, address, vma); 17261da177e4SLinus Torvalds } 172765500d23SHugh Dickins 17281da177e4SLinus Torvalds /* 17291da177e4SLinus Torvalds * Re-check the pte - we dropped the lock 17301da177e4SLinus Torvalds */ 17318f4e2101SHugh Dickins page_table = pte_offset_map_lock(mm, pmd, address, &ptl); 173265500d23SHugh Dickins if (likely(pte_same(*page_table, orig_pte))) { 1733920fc356SHugh Dickins if (old_page) { 17347de6b805SNick Piggin page_remove_rmap(old_page, vma); 17354294621fSHugh Dickins if (!PageAnon(old_page)) { 17364294621fSHugh Dickins dec_mm_counter(mm, file_rss); 1737920fc356SHugh Dickins inc_mm_counter(mm, anon_rss); 17384294621fSHugh Dickins } 1739920fc356SHugh Dickins } else 1740920fc356SHugh Dickins inc_mm_counter(mm, anon_rss); 1741eca35133SBen Collins flush_cache_page(vma, address, pte_pfn(orig_pte)); 174265500d23SHugh Dickins entry = mk_pte(new_page, vma->vm_page_prot); 174365500d23SHugh Dickins entry = maybe_mkwrite(pte_mkdirty(entry), vma); 1744c38c8db7SAnil Keshavamurthy lazy_mmu_prot_update(entry); 17454ce072f1SSiddha, Suresh B /* 17464ce072f1SSiddha, Suresh B * Clear the pte entry and flush it first, before updating the 17474ce072f1SSiddha, Suresh B * pte with the new entry. This will avoid a race condition 17484ce072f1SSiddha, Suresh B * seen in the presence of one thread doing SMC and another 17494ce072f1SSiddha, Suresh B * thread doing COW. 17504ce072f1SSiddha, Suresh B */ 17514ce072f1SSiddha, Suresh B ptep_clear_flush(vma, address, page_table); 17524ce072f1SSiddha, Suresh B set_pte_at(mm, address, page_table, entry); 175365500d23SHugh Dickins update_mmu_cache(vma, address, entry); 17541da177e4SLinus Torvalds lru_cache_add_active(new_page); 17559617d95eSNick Piggin page_add_new_anon_rmap(new_page, vma, address); 17561da177e4SLinus Torvalds 17571da177e4SLinus Torvalds /* Free the old page.. */ 17581da177e4SLinus Torvalds new_page = old_page; 1759f33ea7f4SNick Piggin ret |= VM_FAULT_WRITE; 17601da177e4SLinus Torvalds } 1761920fc356SHugh Dickins if (new_page) 17621da177e4SLinus Torvalds page_cache_release(new_page); 1763920fc356SHugh Dickins if (old_page) 17641da177e4SLinus Torvalds page_cache_release(old_page); 176565500d23SHugh Dickins unlock: 17668f4e2101SHugh Dickins pte_unmap_unlock(page_table, ptl); 1767d08b3851SPeter Zijlstra if (dirty_page) { 1768edc79b2aSPeter Zijlstra set_page_dirty_balance(dirty_page); 1769d08b3851SPeter Zijlstra put_page(dirty_page); 1770d08b3851SPeter Zijlstra } 1771f33ea7f4SNick Piggin return ret; 177265500d23SHugh Dickins oom: 1773920fc356SHugh Dickins if (old_page) 17741da177e4SLinus Torvalds page_cache_release(old_page); 17751da177e4SLinus Torvalds return VM_FAULT_OOM; 17769637a5efSDavid Howells 17779637a5efSDavid Howells unwritable_page: 17789637a5efSDavid Howells page_cache_release(old_page); 17799637a5efSDavid Howells return VM_FAULT_SIGBUS; 17801da177e4SLinus Torvalds } 17811da177e4SLinus Torvalds 17821da177e4SLinus Torvalds /* 17831da177e4SLinus Torvalds * Helper functions for unmap_mapping_range(). 17841da177e4SLinus Torvalds * 17851da177e4SLinus Torvalds * __ Notes on dropping i_mmap_lock to reduce latency while unmapping __ 17861da177e4SLinus Torvalds * 17871da177e4SLinus Torvalds * We have to restart searching the prio_tree whenever we drop the lock, 17881da177e4SLinus Torvalds * since the iterator is only valid while the lock is held, and anyway 17891da177e4SLinus Torvalds * a later vma might be split and reinserted earlier while lock dropped. 17901da177e4SLinus Torvalds * 17911da177e4SLinus Torvalds * The list of nonlinear vmas could be handled more efficiently, using 17921da177e4SLinus Torvalds * a placeholder, but handle it in the same way until a need is shown. 17931da177e4SLinus Torvalds * It is important to search the prio_tree before nonlinear list: a vma 17941da177e4SLinus Torvalds * may become nonlinear and be shifted from prio_tree to nonlinear list 17951da177e4SLinus Torvalds * while the lock is dropped; but never shifted from list to prio_tree. 17961da177e4SLinus Torvalds * 17971da177e4SLinus Torvalds * In order to make forward progress despite restarting the search, 17981da177e4SLinus Torvalds * vm_truncate_count is used to mark a vma as now dealt with, so we can 17991da177e4SLinus Torvalds * quickly skip it next time around. Since the prio_tree search only 18001da177e4SLinus Torvalds * shows us those vmas affected by unmapping the range in question, we 18011da177e4SLinus Torvalds * can't efficiently keep all vmas in step with mapping->truncate_count: 18021da177e4SLinus Torvalds * so instead reset them all whenever it wraps back to 0 (then go to 1). 18031da177e4SLinus Torvalds * mapping->truncate_count and vma->vm_truncate_count are protected by 18041da177e4SLinus Torvalds * i_mmap_lock. 18051da177e4SLinus Torvalds * 18061da177e4SLinus Torvalds * In order to make forward progress despite repeatedly restarting some 1807ee39b37bSHugh Dickins * large vma, note the restart_addr from unmap_vmas when it breaks out: 18081da177e4SLinus Torvalds * and restart from that address when we reach that vma again. It might 18091da177e4SLinus Torvalds * have been split or merged, shrunk or extended, but never shifted: so 18101da177e4SLinus Torvalds * restart_addr remains valid so long as it remains in the vma's range. 18111da177e4SLinus Torvalds * unmap_mapping_range forces truncate_count to leap over page-aligned 18121da177e4SLinus Torvalds * values so we can save vma's restart_addr in its truncate_count field. 18131da177e4SLinus Torvalds */ 18141da177e4SLinus Torvalds #define is_restart_addr(truncate_count) (!((truncate_count) & ~PAGE_MASK)) 18151da177e4SLinus Torvalds 18161da177e4SLinus Torvalds static void reset_vma_truncate_counts(struct address_space *mapping) 18171da177e4SLinus Torvalds { 18181da177e4SLinus Torvalds struct vm_area_struct *vma; 18191da177e4SLinus Torvalds struct prio_tree_iter iter; 18201da177e4SLinus Torvalds 18211da177e4SLinus Torvalds vma_prio_tree_foreach(vma, &iter, &mapping->i_mmap, 0, ULONG_MAX) 18221da177e4SLinus Torvalds vma->vm_truncate_count = 0; 18231da177e4SLinus Torvalds list_for_each_entry(vma, &mapping->i_mmap_nonlinear, shared.vm_set.list) 18241da177e4SLinus Torvalds vma->vm_truncate_count = 0; 18251da177e4SLinus Torvalds } 18261da177e4SLinus Torvalds 18271da177e4SLinus Torvalds static int unmap_mapping_range_vma(struct vm_area_struct *vma, 18281da177e4SLinus Torvalds unsigned long start_addr, unsigned long end_addr, 18291da177e4SLinus Torvalds struct zap_details *details) 18301da177e4SLinus Torvalds { 18311da177e4SLinus Torvalds unsigned long restart_addr; 18321da177e4SLinus Torvalds int need_break; 18331da177e4SLinus Torvalds 1834d00806b1SNick Piggin /* 1835d00806b1SNick Piggin * files that support invalidating or truncating portions of the 1836d0217ac0SNick Piggin * file from under mmaped areas must have their ->fault function 1837*83c54070SNick Piggin * return a locked page (and set VM_FAULT_LOCKED in the return). 1838*83c54070SNick Piggin * This provides synchronisation against concurrent unmapping here. 1839d00806b1SNick Piggin */ 1840d00806b1SNick Piggin 18411da177e4SLinus Torvalds again: 18421da177e4SLinus Torvalds restart_addr = vma->vm_truncate_count; 18431da177e4SLinus Torvalds if (is_restart_addr(restart_addr) && start_addr < restart_addr) { 18441da177e4SLinus Torvalds start_addr = restart_addr; 18451da177e4SLinus Torvalds if (start_addr >= end_addr) { 18461da177e4SLinus Torvalds /* Top of vma has been split off since last time */ 18471da177e4SLinus Torvalds vma->vm_truncate_count = details->truncate_count; 18481da177e4SLinus Torvalds return 0; 18491da177e4SLinus Torvalds } 18501da177e4SLinus Torvalds } 18511da177e4SLinus Torvalds 1852ee39b37bSHugh Dickins restart_addr = zap_page_range(vma, start_addr, 1853ee39b37bSHugh Dickins end_addr - start_addr, details); 18541da177e4SLinus Torvalds need_break = need_resched() || 18551da177e4SLinus Torvalds need_lockbreak(details->i_mmap_lock); 18561da177e4SLinus Torvalds 1857ee39b37bSHugh Dickins if (restart_addr >= end_addr) { 18581da177e4SLinus Torvalds /* We have now completed this vma: mark it so */ 18591da177e4SLinus Torvalds vma->vm_truncate_count = details->truncate_count; 18601da177e4SLinus Torvalds if (!need_break) 18611da177e4SLinus Torvalds return 0; 18621da177e4SLinus Torvalds } else { 18631da177e4SLinus Torvalds /* Note restart_addr in vma's truncate_count field */ 1864ee39b37bSHugh Dickins vma->vm_truncate_count = restart_addr; 18651da177e4SLinus Torvalds if (!need_break) 18661da177e4SLinus Torvalds goto again; 18671da177e4SLinus Torvalds } 18681da177e4SLinus Torvalds 18691da177e4SLinus Torvalds spin_unlock(details->i_mmap_lock); 18701da177e4SLinus Torvalds cond_resched(); 18711da177e4SLinus Torvalds spin_lock(details->i_mmap_lock); 18721da177e4SLinus Torvalds return -EINTR; 18731da177e4SLinus Torvalds } 18741da177e4SLinus Torvalds 18751da177e4SLinus Torvalds static inline void unmap_mapping_range_tree(struct prio_tree_root *root, 18761da177e4SLinus Torvalds struct zap_details *details) 18771da177e4SLinus Torvalds { 18781da177e4SLinus Torvalds struct vm_area_struct *vma; 18791da177e4SLinus Torvalds struct prio_tree_iter iter; 18801da177e4SLinus Torvalds pgoff_t vba, vea, zba, zea; 18811da177e4SLinus Torvalds 18821da177e4SLinus Torvalds restart: 18831da177e4SLinus Torvalds vma_prio_tree_foreach(vma, &iter, root, 18841da177e4SLinus Torvalds details->first_index, details->last_index) { 18851da177e4SLinus Torvalds /* Skip quickly over those we have already dealt with */ 18861da177e4SLinus Torvalds if (vma->vm_truncate_count == details->truncate_count) 18871da177e4SLinus Torvalds continue; 18881da177e4SLinus Torvalds 18891da177e4SLinus Torvalds vba = vma->vm_pgoff; 18901da177e4SLinus Torvalds vea = vba + ((vma->vm_end - vma->vm_start) >> PAGE_SHIFT) - 1; 18911da177e4SLinus Torvalds /* Assume for now that PAGE_CACHE_SHIFT == PAGE_SHIFT */ 18921da177e4SLinus Torvalds zba = details->first_index; 18931da177e4SLinus Torvalds if (zba < vba) 18941da177e4SLinus Torvalds zba = vba; 18951da177e4SLinus Torvalds zea = details->last_index; 18961da177e4SLinus Torvalds if (zea > vea) 18971da177e4SLinus Torvalds zea = vea; 18981da177e4SLinus Torvalds 18991da177e4SLinus Torvalds if (unmap_mapping_range_vma(vma, 19001da177e4SLinus Torvalds ((zba - vba) << PAGE_SHIFT) + vma->vm_start, 19011da177e4SLinus Torvalds ((zea - vba + 1) << PAGE_SHIFT) + vma->vm_start, 19021da177e4SLinus Torvalds details) < 0) 19031da177e4SLinus Torvalds goto restart; 19041da177e4SLinus Torvalds } 19051da177e4SLinus Torvalds } 19061da177e4SLinus Torvalds 19071da177e4SLinus Torvalds static inline void unmap_mapping_range_list(struct list_head *head, 19081da177e4SLinus Torvalds struct zap_details *details) 19091da177e4SLinus Torvalds { 19101da177e4SLinus Torvalds struct vm_area_struct *vma; 19111da177e4SLinus Torvalds 19121da177e4SLinus Torvalds /* 19131da177e4SLinus Torvalds * In nonlinear VMAs there is no correspondence between virtual address 19141da177e4SLinus Torvalds * offset and file offset. So we must perform an exhaustive search 19151da177e4SLinus Torvalds * across *all* the pages in each nonlinear VMA, not just the pages 19161da177e4SLinus Torvalds * whose virtual address lies outside the file truncation point. 19171da177e4SLinus Torvalds */ 19181da177e4SLinus Torvalds restart: 19191da177e4SLinus Torvalds list_for_each_entry(vma, head, shared.vm_set.list) { 19201da177e4SLinus Torvalds /* Skip quickly over those we have already dealt with */ 19211da177e4SLinus Torvalds if (vma->vm_truncate_count == details->truncate_count) 19221da177e4SLinus Torvalds continue; 19231da177e4SLinus Torvalds details->nonlinear_vma = vma; 19241da177e4SLinus Torvalds if (unmap_mapping_range_vma(vma, vma->vm_start, 19251da177e4SLinus Torvalds vma->vm_end, details) < 0) 19261da177e4SLinus Torvalds goto restart; 19271da177e4SLinus Torvalds } 19281da177e4SLinus Torvalds } 19291da177e4SLinus Torvalds 19301da177e4SLinus Torvalds /** 193172fd4a35SRobert P. J. Day * unmap_mapping_range - unmap the portion of all mmaps in the specified address_space corresponding to the specified page range in the underlying file. 19323d41088fSMartin Waitz * @mapping: the address space containing mmaps to be unmapped. 19331da177e4SLinus Torvalds * @holebegin: byte in first page to unmap, relative to the start of 19341da177e4SLinus Torvalds * the underlying file. This will be rounded down to a PAGE_SIZE 19351da177e4SLinus Torvalds * boundary. Note that this is different from vmtruncate(), which 19361da177e4SLinus Torvalds * must keep the partial page. In contrast, we must get rid of 19371da177e4SLinus Torvalds * partial pages. 19381da177e4SLinus Torvalds * @holelen: size of prospective hole in bytes. This will be rounded 19391da177e4SLinus Torvalds * up to a PAGE_SIZE boundary. A holelen of zero truncates to the 19401da177e4SLinus Torvalds * end of the file. 19411da177e4SLinus Torvalds * @even_cows: 1 when truncating a file, unmap even private COWed pages; 19421da177e4SLinus Torvalds * but 0 when invalidating pagecache, don't throw away private data. 19431da177e4SLinus Torvalds */ 19441da177e4SLinus Torvalds void unmap_mapping_range(struct address_space *mapping, 19451da177e4SLinus Torvalds loff_t const holebegin, loff_t const holelen, int even_cows) 19461da177e4SLinus Torvalds { 19471da177e4SLinus Torvalds struct zap_details details; 19481da177e4SLinus Torvalds pgoff_t hba = holebegin >> PAGE_SHIFT; 19491da177e4SLinus Torvalds pgoff_t hlen = (holelen + PAGE_SIZE - 1) >> PAGE_SHIFT; 19501da177e4SLinus Torvalds 19511da177e4SLinus Torvalds /* Check for overflow. */ 19521da177e4SLinus Torvalds if (sizeof(holelen) > sizeof(hlen)) { 19531da177e4SLinus Torvalds long long holeend = 19541da177e4SLinus Torvalds (holebegin + holelen + PAGE_SIZE - 1) >> PAGE_SHIFT; 19551da177e4SLinus Torvalds if (holeend & ~(long long)ULONG_MAX) 19561da177e4SLinus Torvalds hlen = ULONG_MAX - hba + 1; 19571da177e4SLinus Torvalds } 19581da177e4SLinus Torvalds 19591da177e4SLinus Torvalds details.check_mapping = even_cows? NULL: mapping; 19601da177e4SLinus Torvalds details.nonlinear_vma = NULL; 19611da177e4SLinus Torvalds details.first_index = hba; 19621da177e4SLinus Torvalds details.last_index = hba + hlen - 1; 19631da177e4SLinus Torvalds if (details.last_index < details.first_index) 19641da177e4SLinus Torvalds details.last_index = ULONG_MAX; 19651da177e4SLinus Torvalds details.i_mmap_lock = &mapping->i_mmap_lock; 19661da177e4SLinus Torvalds 19671da177e4SLinus Torvalds spin_lock(&mapping->i_mmap_lock); 19681da177e4SLinus Torvalds 1969d00806b1SNick Piggin /* Protect against endless unmapping loops */ 19701da177e4SLinus Torvalds mapping->truncate_count++; 19711da177e4SLinus Torvalds if (unlikely(is_restart_addr(mapping->truncate_count))) { 19721da177e4SLinus Torvalds if (mapping->truncate_count == 0) 19731da177e4SLinus Torvalds reset_vma_truncate_counts(mapping); 19741da177e4SLinus Torvalds mapping->truncate_count++; 19751da177e4SLinus Torvalds } 19761da177e4SLinus Torvalds details.truncate_count = mapping->truncate_count; 19771da177e4SLinus Torvalds 19781da177e4SLinus Torvalds if (unlikely(!prio_tree_empty(&mapping->i_mmap))) 19791da177e4SLinus Torvalds unmap_mapping_range_tree(&mapping->i_mmap, &details); 19801da177e4SLinus Torvalds if (unlikely(!list_empty(&mapping->i_mmap_nonlinear))) 19811da177e4SLinus Torvalds unmap_mapping_range_list(&mapping->i_mmap_nonlinear, &details); 19821da177e4SLinus Torvalds spin_unlock(&mapping->i_mmap_lock); 19831da177e4SLinus Torvalds } 19841da177e4SLinus Torvalds EXPORT_SYMBOL(unmap_mapping_range); 19851da177e4SLinus Torvalds 1986bfa5bf6dSRolf Eike Beer /** 1987bfa5bf6dSRolf Eike Beer * vmtruncate - unmap mappings "freed" by truncate() syscall 1988bfa5bf6dSRolf Eike Beer * @inode: inode of the file used 1989bfa5bf6dSRolf Eike Beer * @offset: file offset to start truncating 19901da177e4SLinus Torvalds * 19911da177e4SLinus Torvalds * NOTE! We have to be ready to update the memory sharing 19921da177e4SLinus Torvalds * between the file and the memory map for a potential last 19931da177e4SLinus Torvalds * incomplete page. Ugly, but necessary. 19941da177e4SLinus Torvalds */ 19951da177e4SLinus Torvalds int vmtruncate(struct inode * inode, loff_t offset) 19961da177e4SLinus Torvalds { 19971da177e4SLinus Torvalds struct address_space *mapping = inode->i_mapping; 19981da177e4SLinus Torvalds unsigned long limit; 19991da177e4SLinus Torvalds 20001da177e4SLinus Torvalds if (inode->i_size < offset) 20011da177e4SLinus Torvalds goto do_expand; 20021da177e4SLinus Torvalds /* 20031da177e4SLinus Torvalds * truncation of in-use swapfiles is disallowed - it would cause 20041da177e4SLinus Torvalds * subsequent swapout to scribble on the now-freed blocks. 20051da177e4SLinus Torvalds */ 20061da177e4SLinus Torvalds if (IS_SWAPFILE(inode)) 20071da177e4SLinus Torvalds goto out_busy; 20081da177e4SLinus Torvalds i_size_write(inode, offset); 2009d00806b1SNick Piggin 2010d00806b1SNick Piggin /* 2011d00806b1SNick Piggin * unmap_mapping_range is called twice, first simply for efficiency 2012d00806b1SNick Piggin * so that truncate_inode_pages does fewer single-page unmaps. However 2013d00806b1SNick Piggin * after this first call, and before truncate_inode_pages finishes, 2014d00806b1SNick Piggin * it is possible for private pages to be COWed, which remain after 2015d00806b1SNick Piggin * truncate_inode_pages finishes, hence the second unmap_mapping_range 2016d00806b1SNick Piggin * call must be made for correctness. 2017d00806b1SNick Piggin */ 20181da177e4SLinus Torvalds unmap_mapping_range(mapping, offset + PAGE_SIZE - 1, 0, 1); 20191da177e4SLinus Torvalds truncate_inode_pages(mapping, offset); 2020d00806b1SNick Piggin unmap_mapping_range(mapping, offset + PAGE_SIZE - 1, 0, 1); 20211da177e4SLinus Torvalds goto out_truncate; 20221da177e4SLinus Torvalds 20231da177e4SLinus Torvalds do_expand: 20241da177e4SLinus Torvalds limit = current->signal->rlim[RLIMIT_FSIZE].rlim_cur; 20251da177e4SLinus Torvalds if (limit != RLIM_INFINITY && offset > limit) 20261da177e4SLinus Torvalds goto out_sig; 20271da177e4SLinus Torvalds if (offset > inode->i_sb->s_maxbytes) 20281da177e4SLinus Torvalds goto out_big; 20291da177e4SLinus Torvalds i_size_write(inode, offset); 20301da177e4SLinus Torvalds 20311da177e4SLinus Torvalds out_truncate: 20321da177e4SLinus Torvalds if (inode->i_op && inode->i_op->truncate) 20331da177e4SLinus Torvalds inode->i_op->truncate(inode); 20341da177e4SLinus Torvalds return 0; 20351da177e4SLinus Torvalds out_sig: 20361da177e4SLinus Torvalds send_sig(SIGXFSZ, current, 0); 20371da177e4SLinus Torvalds out_big: 20381da177e4SLinus Torvalds return -EFBIG; 20391da177e4SLinus Torvalds out_busy: 20401da177e4SLinus Torvalds return -ETXTBSY; 20411da177e4SLinus Torvalds } 20421da177e4SLinus Torvalds EXPORT_SYMBOL(vmtruncate); 20431da177e4SLinus Torvalds 2044f6b3ec23SBadari Pulavarty int vmtruncate_range(struct inode *inode, loff_t offset, loff_t end) 2045f6b3ec23SBadari Pulavarty { 2046f6b3ec23SBadari Pulavarty struct address_space *mapping = inode->i_mapping; 2047f6b3ec23SBadari Pulavarty 2048f6b3ec23SBadari Pulavarty /* 2049f6b3ec23SBadari Pulavarty * If the underlying filesystem is not going to provide 2050f6b3ec23SBadari Pulavarty * a way to truncate a range of blocks (punch a hole) - 2051f6b3ec23SBadari Pulavarty * we should return failure right now. 2052f6b3ec23SBadari Pulavarty */ 2053f6b3ec23SBadari Pulavarty if (!inode->i_op || !inode->i_op->truncate_range) 2054f6b3ec23SBadari Pulavarty return -ENOSYS; 2055f6b3ec23SBadari Pulavarty 20561b1dcc1bSJes Sorensen mutex_lock(&inode->i_mutex); 2057f6b3ec23SBadari Pulavarty down_write(&inode->i_alloc_sem); 2058f6b3ec23SBadari Pulavarty unmap_mapping_range(mapping, offset, (end - offset), 1); 2059f6b3ec23SBadari Pulavarty truncate_inode_pages_range(mapping, offset, end); 2060d00806b1SNick Piggin unmap_mapping_range(mapping, offset, (end - offset), 1); 2061f6b3ec23SBadari Pulavarty inode->i_op->truncate_range(inode, offset, end); 2062f6b3ec23SBadari Pulavarty up_write(&inode->i_alloc_sem); 20631b1dcc1bSJes Sorensen mutex_unlock(&inode->i_mutex); 2064f6b3ec23SBadari Pulavarty 2065f6b3ec23SBadari Pulavarty return 0; 2066f6b3ec23SBadari Pulavarty } 2067f6b3ec23SBadari Pulavarty 2068bfa5bf6dSRolf Eike Beer /** 2069bfa5bf6dSRolf Eike Beer * swapin_readahead - swap in pages in hope we need them soon 2070bfa5bf6dSRolf Eike Beer * @entry: swap entry of this memory 2071bfa5bf6dSRolf Eike Beer * @addr: address to start 2072bfa5bf6dSRolf Eike Beer * @vma: user vma this addresses belong to 2073bfa5bf6dSRolf Eike Beer * 20741da177e4SLinus Torvalds * Primitive swap readahead code. We simply read an aligned block of 20751da177e4SLinus Torvalds * (1 << page_cluster) entries in the swap area. This method is chosen 20761da177e4SLinus Torvalds * because it doesn't cost us any seek time. We also make sure to queue 20771da177e4SLinus Torvalds * the 'original' request together with the readahead ones... 20781da177e4SLinus Torvalds * 20791da177e4SLinus Torvalds * This has been extended to use the NUMA policies from the mm triggering 20801da177e4SLinus Torvalds * the readahead. 20811da177e4SLinus Torvalds * 20821da177e4SLinus Torvalds * Caller must hold down_read on the vma->vm_mm if vma is not NULL. 20831da177e4SLinus Torvalds */ 20841da177e4SLinus Torvalds void swapin_readahead(swp_entry_t entry, unsigned long addr,struct vm_area_struct *vma) 20851da177e4SLinus Torvalds { 20861da177e4SLinus Torvalds #ifdef CONFIG_NUMA 20871da177e4SLinus Torvalds struct vm_area_struct *next_vma = vma ? vma->vm_next : NULL; 20881da177e4SLinus Torvalds #endif 20891da177e4SLinus Torvalds int i, num; 20901da177e4SLinus Torvalds struct page *new_page; 20911da177e4SLinus Torvalds unsigned long offset; 20921da177e4SLinus Torvalds 20931da177e4SLinus Torvalds /* 20941da177e4SLinus Torvalds * Get the number of handles we should do readahead io to. 20951da177e4SLinus Torvalds */ 20961da177e4SLinus Torvalds num = valid_swaphandles(entry, &offset); 20971da177e4SLinus Torvalds for (i = 0; i < num; offset++, i++) { 20981da177e4SLinus Torvalds /* Ok, do the async read-ahead now */ 20991da177e4SLinus Torvalds new_page = read_swap_cache_async(swp_entry(swp_type(entry), 21001da177e4SLinus Torvalds offset), vma, addr); 21011da177e4SLinus Torvalds if (!new_page) 21021da177e4SLinus Torvalds break; 21031da177e4SLinus Torvalds page_cache_release(new_page); 21041da177e4SLinus Torvalds #ifdef CONFIG_NUMA 21051da177e4SLinus Torvalds /* 21061da177e4SLinus Torvalds * Find the next applicable VMA for the NUMA policy. 21071da177e4SLinus Torvalds */ 21081da177e4SLinus Torvalds addr += PAGE_SIZE; 21091da177e4SLinus Torvalds if (addr == 0) 21101da177e4SLinus Torvalds vma = NULL; 21111da177e4SLinus Torvalds if (vma) { 21121da177e4SLinus Torvalds if (addr >= vma->vm_end) { 21131da177e4SLinus Torvalds vma = next_vma; 21141da177e4SLinus Torvalds next_vma = vma ? vma->vm_next : NULL; 21151da177e4SLinus Torvalds } 21161da177e4SLinus Torvalds if (vma && addr < vma->vm_start) 21171da177e4SLinus Torvalds vma = NULL; 21181da177e4SLinus Torvalds } else { 21191da177e4SLinus Torvalds if (next_vma && addr >= next_vma->vm_start) { 21201da177e4SLinus Torvalds vma = next_vma; 21211da177e4SLinus Torvalds next_vma = vma->vm_next; 21221da177e4SLinus Torvalds } 21231da177e4SLinus Torvalds } 21241da177e4SLinus Torvalds #endif 21251da177e4SLinus Torvalds } 21261da177e4SLinus Torvalds lru_add_drain(); /* Push any new pages onto the LRU now */ 21271da177e4SLinus Torvalds } 21281da177e4SLinus Torvalds 21291da177e4SLinus Torvalds /* 21308f4e2101SHugh Dickins * We enter with non-exclusive mmap_sem (to exclude vma changes, 21318f4e2101SHugh Dickins * but allow concurrent faults), and pte mapped but not yet locked. 21328f4e2101SHugh Dickins * We return with mmap_sem still held, but pte unmapped and unlocked. 21331da177e4SLinus Torvalds */ 213465500d23SHugh Dickins static int do_swap_page(struct mm_struct *mm, struct vm_area_struct *vma, 213565500d23SHugh Dickins unsigned long address, pte_t *page_table, pmd_t *pmd, 213665500d23SHugh Dickins int write_access, pte_t orig_pte) 21371da177e4SLinus Torvalds { 21388f4e2101SHugh Dickins spinlock_t *ptl; 21391da177e4SLinus Torvalds struct page *page; 214065500d23SHugh Dickins swp_entry_t entry; 21411da177e4SLinus Torvalds pte_t pte; 2142*83c54070SNick Piggin int ret = 0; 21431da177e4SLinus Torvalds 21444c21e2f2SHugh Dickins if (!pte_unmap_same(mm, pmd, page_table, orig_pte)) 21458f4e2101SHugh Dickins goto out; 214665500d23SHugh Dickins 214765500d23SHugh Dickins entry = pte_to_swp_entry(orig_pte); 21480697212aSChristoph Lameter if (is_migration_entry(entry)) { 21490697212aSChristoph Lameter migration_entry_wait(mm, pmd, address); 21500697212aSChristoph Lameter goto out; 21510697212aSChristoph Lameter } 21520ff92245SShailabh Nagar delayacct_set_flag(DELAYACCT_PF_SWAPIN); 21531da177e4SLinus Torvalds page = lookup_swap_cache(entry); 21541da177e4SLinus Torvalds if (!page) { 2155098fe651SAshwin Chaugule grab_swap_token(); /* Contend for token _before_ read-in */ 21561da177e4SLinus Torvalds swapin_readahead(entry, address, vma); 21571da177e4SLinus Torvalds page = read_swap_cache_async(entry, vma, address); 21581da177e4SLinus Torvalds if (!page) { 21591da177e4SLinus Torvalds /* 21608f4e2101SHugh Dickins * Back out if somebody else faulted in this pte 21618f4e2101SHugh Dickins * while we released the pte lock. 21621da177e4SLinus Torvalds */ 21638f4e2101SHugh Dickins page_table = pte_offset_map_lock(mm, pmd, address, &ptl); 21641da177e4SLinus Torvalds if (likely(pte_same(*page_table, orig_pte))) 21651da177e4SLinus Torvalds ret = VM_FAULT_OOM; 21660ff92245SShailabh Nagar delayacct_clear_flag(DELAYACCT_PF_SWAPIN); 216765500d23SHugh Dickins goto unlock; 21681da177e4SLinus Torvalds } 21691da177e4SLinus Torvalds 21701da177e4SLinus Torvalds /* Had to read the page from swap area: Major fault */ 21711da177e4SLinus Torvalds ret = VM_FAULT_MAJOR; 2172f8891e5eSChristoph Lameter count_vm_event(PGMAJFAULT); 21731da177e4SLinus Torvalds } 21741da177e4SLinus Torvalds 21750ff92245SShailabh Nagar delayacct_clear_flag(DELAYACCT_PF_SWAPIN); 21761da177e4SLinus Torvalds mark_page_accessed(page); 21771da177e4SLinus Torvalds lock_page(page); 21781da177e4SLinus Torvalds 21791da177e4SLinus Torvalds /* 21808f4e2101SHugh Dickins * Back out if somebody else already faulted in this pte. 21811da177e4SLinus Torvalds */ 21828f4e2101SHugh Dickins page_table = pte_offset_map_lock(mm, pmd, address, &ptl); 21839e9bef07SHugh Dickins if (unlikely(!pte_same(*page_table, orig_pte))) 2184b8107480SKirill Korotaev goto out_nomap; 2185b8107480SKirill Korotaev 2186b8107480SKirill Korotaev if (unlikely(!PageUptodate(page))) { 2187b8107480SKirill Korotaev ret = VM_FAULT_SIGBUS; 2188b8107480SKirill Korotaev goto out_nomap; 21891da177e4SLinus Torvalds } 21901da177e4SLinus Torvalds 21911da177e4SLinus Torvalds /* The page isn't present yet, go ahead with the fault. */ 21921da177e4SLinus Torvalds 21934294621fSHugh Dickins inc_mm_counter(mm, anon_rss); 21941da177e4SLinus Torvalds pte = mk_pte(page, vma->vm_page_prot); 21951da177e4SLinus Torvalds if (write_access && can_share_swap_page(page)) { 21961da177e4SLinus Torvalds pte = maybe_mkwrite(pte_mkdirty(pte), vma); 21971da177e4SLinus Torvalds write_access = 0; 21981da177e4SLinus Torvalds } 21991da177e4SLinus Torvalds 22001da177e4SLinus Torvalds flush_icache_page(vma, page); 22011da177e4SLinus Torvalds set_pte_at(mm, address, page_table, pte); 22021da177e4SLinus Torvalds page_add_anon_rmap(page, vma, address); 22031da177e4SLinus Torvalds 2204c475a8abSHugh Dickins swap_free(entry); 2205c475a8abSHugh Dickins if (vm_swap_full()) 2206c475a8abSHugh Dickins remove_exclusive_swap_page(page); 2207c475a8abSHugh Dickins unlock_page(page); 2208c475a8abSHugh Dickins 22091da177e4SLinus Torvalds if (write_access) { 2210*83c54070SNick Piggin /* XXX: We could OR the do_wp_page code with this one? */ 22111da177e4SLinus Torvalds if (do_wp_page(mm, vma, address, 2212*83c54070SNick Piggin page_table, pmd, ptl, pte) & VM_FAULT_OOM) 22131da177e4SLinus Torvalds ret = VM_FAULT_OOM; 22141da177e4SLinus Torvalds goto out; 22151da177e4SLinus Torvalds } 22161da177e4SLinus Torvalds 22171da177e4SLinus Torvalds /* No need to invalidate - it was non-present before */ 22181da177e4SLinus Torvalds update_mmu_cache(vma, address, pte); 221965500d23SHugh Dickins unlock: 22208f4e2101SHugh Dickins pte_unmap_unlock(page_table, ptl); 22211da177e4SLinus Torvalds out: 22221da177e4SLinus Torvalds return ret; 2223b8107480SKirill Korotaev out_nomap: 22248f4e2101SHugh Dickins pte_unmap_unlock(page_table, ptl); 2225b8107480SKirill Korotaev unlock_page(page); 2226b8107480SKirill Korotaev page_cache_release(page); 222765500d23SHugh Dickins return ret; 22281da177e4SLinus Torvalds } 22291da177e4SLinus Torvalds 22301da177e4SLinus Torvalds /* 22318f4e2101SHugh Dickins * We enter with non-exclusive mmap_sem (to exclude vma changes, 22328f4e2101SHugh Dickins * but allow concurrent faults), and pte mapped but not yet locked. 22338f4e2101SHugh Dickins * We return with mmap_sem still held, but pte unmapped and unlocked. 22341da177e4SLinus Torvalds */ 223565500d23SHugh Dickins static int do_anonymous_page(struct mm_struct *mm, struct vm_area_struct *vma, 223665500d23SHugh Dickins unsigned long address, pte_t *page_table, pmd_t *pmd, 223765500d23SHugh Dickins int write_access) 22381da177e4SLinus Torvalds { 22398f4e2101SHugh Dickins struct page *page; 22408f4e2101SHugh Dickins spinlock_t *ptl; 22411da177e4SLinus Torvalds pte_t entry; 22421da177e4SLinus Torvalds 22436aab341eSLinus Torvalds if (write_access) { 22441da177e4SLinus Torvalds /* Allocate our own private page. */ 22451da177e4SLinus Torvalds pte_unmap(page_table); 22461da177e4SLinus Torvalds 22471da177e4SLinus Torvalds if (unlikely(anon_vma_prepare(vma))) 224865500d23SHugh Dickins goto oom; 2249769848c0SMel Gorman page = alloc_zeroed_user_highpage_movable(vma, address); 22501da177e4SLinus Torvalds if (!page) 225165500d23SHugh Dickins goto oom; 22521da177e4SLinus Torvalds 225365500d23SHugh Dickins entry = mk_pte(page, vma->vm_page_prot); 225465500d23SHugh Dickins entry = maybe_mkwrite(pte_mkdirty(entry), vma); 22558f4e2101SHugh Dickins 22568f4e2101SHugh Dickins page_table = pte_offset_map_lock(mm, pmd, address, &ptl); 22578f4e2101SHugh Dickins if (!pte_none(*page_table)) 22588f4e2101SHugh Dickins goto release; 22598f4e2101SHugh Dickins inc_mm_counter(mm, anon_rss); 22601da177e4SLinus Torvalds lru_cache_add_active(page); 22619617d95eSNick Piggin page_add_new_anon_rmap(page, vma, address); 2262b5810039SNick Piggin } else { 22638f4e2101SHugh Dickins /* Map the ZERO_PAGE - vm_page_prot is readonly */ 22648f4e2101SHugh Dickins page = ZERO_PAGE(address); 22658f4e2101SHugh Dickins page_cache_get(page); 22668f4e2101SHugh Dickins entry = mk_pte(page, vma->vm_page_prot); 22678f4e2101SHugh Dickins 22684c21e2f2SHugh Dickins ptl = pte_lockptr(mm, pmd); 22698f4e2101SHugh Dickins spin_lock(ptl); 22708f4e2101SHugh Dickins if (!pte_none(*page_table)) 22718f4e2101SHugh Dickins goto release; 2272b5810039SNick Piggin inc_mm_counter(mm, file_rss); 2273b5810039SNick Piggin page_add_file_rmap(page); 22741da177e4SLinus Torvalds } 22751da177e4SLinus Torvalds 227665500d23SHugh Dickins set_pte_at(mm, address, page_table, entry); 22771da177e4SLinus Torvalds 22781da177e4SLinus Torvalds /* No need to invalidate - it was non-present before */ 227965500d23SHugh Dickins update_mmu_cache(vma, address, entry); 22801da177e4SLinus Torvalds lazy_mmu_prot_update(entry); 228165500d23SHugh Dickins unlock: 22828f4e2101SHugh Dickins pte_unmap_unlock(page_table, ptl); 2283*83c54070SNick Piggin return 0; 22848f4e2101SHugh Dickins release: 22858f4e2101SHugh Dickins page_cache_release(page); 22868f4e2101SHugh Dickins goto unlock; 228765500d23SHugh Dickins oom: 22881da177e4SLinus Torvalds return VM_FAULT_OOM; 22891da177e4SLinus Torvalds } 22901da177e4SLinus Torvalds 22911da177e4SLinus Torvalds /* 229254cb8821SNick Piggin * __do_fault() tries to create a new page mapping. It aggressively 22931da177e4SLinus Torvalds * tries to share with existing pages, but makes a separate copy if 229454cb8821SNick Piggin * the FAULT_FLAG_WRITE is set in the flags parameter in order to avoid 229554cb8821SNick Piggin * the next page fault. 22961da177e4SLinus Torvalds * 22971da177e4SLinus Torvalds * As this is called only for pages that do not currently exist, we 22981da177e4SLinus Torvalds * do not need to flush old virtual caches or the TLB. 22991da177e4SLinus Torvalds * 23008f4e2101SHugh Dickins * We enter with non-exclusive mmap_sem (to exclude vma changes, 23018f4e2101SHugh Dickins * but allow concurrent faults), and pte mapped but not yet locked. 23028f4e2101SHugh Dickins * We return with mmap_sem still held, but pte unmapped and unlocked. 23031da177e4SLinus Torvalds */ 230454cb8821SNick Piggin static int __do_fault(struct mm_struct *mm, struct vm_area_struct *vma, 230565500d23SHugh Dickins unsigned long address, pte_t *page_table, pmd_t *pmd, 230654cb8821SNick Piggin pgoff_t pgoff, unsigned int flags, pte_t orig_pte) 23071da177e4SLinus Torvalds { 23088f4e2101SHugh Dickins spinlock_t *ptl; 2309d0217ac0SNick Piggin struct page *page; 23101da177e4SLinus Torvalds pte_t entry; 23111da177e4SLinus Torvalds int anon = 0; 2312d08b3851SPeter Zijlstra struct page *dirty_page = NULL; 2313d0217ac0SNick Piggin struct vm_fault vmf; 2314d0217ac0SNick Piggin int ret; 231554cb8821SNick Piggin 2316d0217ac0SNick Piggin vmf.virtual_address = (void __user *)(address & PAGE_MASK); 2317d0217ac0SNick Piggin vmf.pgoff = pgoff; 2318d0217ac0SNick Piggin vmf.flags = flags; 2319d0217ac0SNick Piggin vmf.page = NULL; 23201da177e4SLinus Torvalds 23211da177e4SLinus Torvalds pte_unmap(page_table); 2322325f04dbSHugh Dickins BUG_ON(vma->vm_flags & VM_PFNMAP); 2323325f04dbSHugh Dickins 232454cb8821SNick Piggin if (likely(vma->vm_ops->fault)) { 2325d0217ac0SNick Piggin ret = vma->vm_ops->fault(vma, &vmf); 2326*83c54070SNick Piggin if (unlikely(ret & (VM_FAULT_ERROR | VM_FAULT_NOPAGE))) 2327*83c54070SNick Piggin return ret; 232854cb8821SNick Piggin } else { 232954cb8821SNick Piggin /* Legacy ->nopage path */ 2330*83c54070SNick Piggin ret = 0; 2331d0217ac0SNick Piggin vmf.page = vma->vm_ops->nopage(vma, address & PAGE_MASK, &ret); 233254cb8821SNick Piggin /* no page was available -- either SIGBUS or OOM */ 2333d0217ac0SNick Piggin if (unlikely(vmf.page == NOPAGE_SIGBUS)) 23341da177e4SLinus Torvalds return VM_FAULT_SIGBUS; 2335d0217ac0SNick Piggin else if (unlikely(vmf.page == NOPAGE_OOM)) 23361da177e4SLinus Torvalds return VM_FAULT_OOM; 233754cb8821SNick Piggin } 23381da177e4SLinus Torvalds 2339d00806b1SNick Piggin /* 2340d0217ac0SNick Piggin * For consistency in subsequent calls, make the faulted page always 2341d00806b1SNick Piggin * locked. 2342d00806b1SNick Piggin */ 2343*83c54070SNick Piggin if (unlikely(!(ret & VM_FAULT_LOCKED))) 2344d0217ac0SNick Piggin lock_page(vmf.page); 234554cb8821SNick Piggin else 2346d0217ac0SNick Piggin VM_BUG_ON(!PageLocked(vmf.page)); 2347d00806b1SNick Piggin 23481da177e4SLinus Torvalds /* 23491da177e4SLinus Torvalds * Should we do an early C-O-W break? 23501da177e4SLinus Torvalds */ 2351d0217ac0SNick Piggin page = vmf.page; 235254cb8821SNick Piggin if (flags & FAULT_FLAG_WRITE) { 23539637a5efSDavid Howells if (!(vma->vm_flags & VM_SHARED)) { 235454cb8821SNick Piggin anon = 1; 2355d00806b1SNick Piggin if (unlikely(anon_vma_prepare(vma))) { 2356d0217ac0SNick Piggin ret = VM_FAULT_OOM; 235754cb8821SNick Piggin goto out; 2358d00806b1SNick Piggin } 2359*83c54070SNick Piggin page = alloc_page_vma(GFP_HIGHUSER_MOVABLE, 2360*83c54070SNick Piggin vma, address); 2361d00806b1SNick Piggin if (!page) { 2362d0217ac0SNick Piggin ret = VM_FAULT_OOM; 236354cb8821SNick Piggin goto out; 2364d00806b1SNick Piggin } 2365d0217ac0SNick Piggin copy_user_highpage(page, vmf.page, address, vma); 23669637a5efSDavid Howells } else { 236754cb8821SNick Piggin /* 236854cb8821SNick Piggin * If the page will be shareable, see if the backing 23699637a5efSDavid Howells * address space wants to know that the page is about 237054cb8821SNick Piggin * to become writable 237154cb8821SNick Piggin */ 237269676147SMark Fasheh if (vma->vm_ops->page_mkwrite) { 237369676147SMark Fasheh unlock_page(page); 237469676147SMark Fasheh if (vma->vm_ops->page_mkwrite(vma, page) < 0) { 2375d0217ac0SNick Piggin ret = VM_FAULT_SIGBUS; 2376d0217ac0SNick Piggin anon = 1; /* no anon but release vmf.page */ 237769676147SMark Fasheh goto out_unlocked; 237869676147SMark Fasheh } 237969676147SMark Fasheh lock_page(page); 2380d0217ac0SNick Piggin /* 2381d0217ac0SNick Piggin * XXX: this is not quite right (racy vs 2382d0217ac0SNick Piggin * invalidate) to unlock and relock the page 2383d0217ac0SNick Piggin * like this, however a better fix requires 2384d0217ac0SNick Piggin * reworking page_mkwrite locking API, which 2385d0217ac0SNick Piggin * is better done later. 2386d0217ac0SNick Piggin */ 2387d0217ac0SNick Piggin if (!page->mapping) { 2388*83c54070SNick Piggin ret = 0; 2389d0217ac0SNick Piggin anon = 1; /* no anon but release vmf.page */ 2390d0217ac0SNick Piggin goto out; 2391d0217ac0SNick Piggin } 23929637a5efSDavid Howells } 23939637a5efSDavid Howells } 239454cb8821SNick Piggin 23951da177e4SLinus Torvalds } 23961da177e4SLinus Torvalds 23978f4e2101SHugh Dickins page_table = pte_offset_map_lock(mm, pmd, address, &ptl); 23981da177e4SLinus Torvalds 23991da177e4SLinus Torvalds /* 24001da177e4SLinus Torvalds * This silly early PAGE_DIRTY setting removes a race 24011da177e4SLinus Torvalds * due to the bad i386 page protection. But it's valid 24021da177e4SLinus Torvalds * for other architectures too. 24031da177e4SLinus Torvalds * 24041da177e4SLinus Torvalds * Note that if write_access is true, we either now have 24051da177e4SLinus Torvalds * an exclusive copy of the page, or this is a shared mapping, 24061da177e4SLinus Torvalds * so we can make it writable and dirty to avoid having to 24071da177e4SLinus Torvalds * handle that later. 24081da177e4SLinus Torvalds */ 24091da177e4SLinus Torvalds /* Only go through if we didn't race with anybody else... */ 241054cb8821SNick Piggin if (likely(pte_same(*page_table, orig_pte))) { 2411d00806b1SNick Piggin flush_icache_page(vma, page); 2412d00806b1SNick Piggin entry = mk_pte(page, vma->vm_page_prot); 241354cb8821SNick Piggin if (flags & FAULT_FLAG_WRITE) 24141da177e4SLinus Torvalds entry = maybe_mkwrite(pte_mkdirty(entry), vma); 24151da177e4SLinus Torvalds set_pte_at(mm, address, page_table, entry); 24161da177e4SLinus Torvalds if (anon) { 24174294621fSHugh Dickins inc_mm_counter(mm, anon_rss); 2418d00806b1SNick Piggin lru_cache_add_active(page); 2419d00806b1SNick Piggin page_add_new_anon_rmap(page, vma, address); 2420f57e88a8SHugh Dickins } else { 24214294621fSHugh Dickins inc_mm_counter(mm, file_rss); 2422d00806b1SNick Piggin page_add_file_rmap(page); 242354cb8821SNick Piggin if (flags & FAULT_FLAG_WRITE) { 2424d00806b1SNick Piggin dirty_page = page; 2425d08b3851SPeter Zijlstra get_page(dirty_page); 2426d08b3851SPeter Zijlstra } 24274294621fSHugh Dickins } 24281da177e4SLinus Torvalds 2429d00806b1SNick Piggin /* no need to invalidate: a not-present page won't be cached */ 24301da177e4SLinus Torvalds update_mmu_cache(vma, address, entry); 24311da177e4SLinus Torvalds lazy_mmu_prot_update(entry); 2432d00806b1SNick Piggin } else { 2433d00806b1SNick Piggin if (anon) 2434d00806b1SNick Piggin page_cache_release(page); 2435d00806b1SNick Piggin else 243654cb8821SNick Piggin anon = 1; /* no anon but release faulted_page */ 2437d00806b1SNick Piggin } 2438d00806b1SNick Piggin 24398f4e2101SHugh Dickins pte_unmap_unlock(page_table, ptl); 2440d00806b1SNick Piggin 2441d00806b1SNick Piggin out: 2442d0217ac0SNick Piggin unlock_page(vmf.page); 244369676147SMark Fasheh out_unlocked: 2444d00806b1SNick Piggin if (anon) 2445d0217ac0SNick Piggin page_cache_release(vmf.page); 2446d00806b1SNick Piggin else if (dirty_page) { 2447edc79b2aSPeter Zijlstra set_page_dirty_balance(dirty_page); 2448d08b3851SPeter Zijlstra put_page(dirty_page); 2449d08b3851SPeter Zijlstra } 2450d00806b1SNick Piggin 2451*83c54070SNick Piggin return ret; 245254cb8821SNick Piggin } 2453d00806b1SNick Piggin 245454cb8821SNick Piggin static int do_linear_fault(struct mm_struct *mm, struct vm_area_struct *vma, 245554cb8821SNick Piggin unsigned long address, pte_t *page_table, pmd_t *pmd, 245654cb8821SNick Piggin int write_access, pte_t orig_pte) 245754cb8821SNick Piggin { 245854cb8821SNick Piggin pgoff_t pgoff = (((address & PAGE_MASK) 245954cb8821SNick Piggin - vma->vm_start) >> PAGE_CACHE_SHIFT) + vma->vm_pgoff; 246054cb8821SNick Piggin unsigned int flags = (write_access ? FAULT_FLAG_WRITE : 0); 246154cb8821SNick Piggin 2462d0217ac0SNick Piggin return __do_fault(mm, vma, address, page_table, pmd, pgoff, 2463d0217ac0SNick Piggin flags, orig_pte); 246454cb8821SNick Piggin } 246554cb8821SNick Piggin 24661da177e4SLinus Torvalds 24671da177e4SLinus Torvalds /* 2468f4b81804SJes Sorensen * do_no_pfn() tries to create a new page mapping for a page without 2469f4b81804SJes Sorensen * a struct_page backing it 2470f4b81804SJes Sorensen * 2471f4b81804SJes Sorensen * As this is called only for pages that do not currently exist, we 2472f4b81804SJes Sorensen * do not need to flush old virtual caches or the TLB. 2473f4b81804SJes Sorensen * 2474f4b81804SJes Sorensen * We enter with non-exclusive mmap_sem (to exclude vma changes, 2475f4b81804SJes Sorensen * but allow concurrent faults), and pte mapped but not yet locked. 2476f4b81804SJes Sorensen * We return with mmap_sem still held, but pte unmapped and unlocked. 2477f4b81804SJes Sorensen * 2478f4b81804SJes Sorensen * It is expected that the ->nopfn handler always returns the same pfn 2479f4b81804SJes Sorensen * for a given virtual mapping. 2480f4b81804SJes Sorensen * 2481f4b81804SJes Sorensen * Mark this `noinline' to prevent it from bloating the main pagefault code. 2482f4b81804SJes Sorensen */ 2483f4b81804SJes Sorensen static noinline int do_no_pfn(struct mm_struct *mm, struct vm_area_struct *vma, 2484f4b81804SJes Sorensen unsigned long address, pte_t *page_table, pmd_t *pmd, 2485f4b81804SJes Sorensen int write_access) 2486f4b81804SJes Sorensen { 2487f4b81804SJes Sorensen spinlock_t *ptl; 2488f4b81804SJes Sorensen pte_t entry; 2489f4b81804SJes Sorensen unsigned long pfn; 2490f4b81804SJes Sorensen 2491f4b81804SJes Sorensen pte_unmap(page_table); 2492f4b81804SJes Sorensen BUG_ON(!(vma->vm_flags & VM_PFNMAP)); 2493f4b81804SJes Sorensen BUG_ON(is_cow_mapping(vma->vm_flags)); 2494f4b81804SJes Sorensen 2495f4b81804SJes Sorensen pfn = vma->vm_ops->nopfn(vma, address & PAGE_MASK); 249622cd25edSBenjamin Herrenschmidt if (unlikely(pfn == NOPFN_OOM)) 2497f4b81804SJes Sorensen return VM_FAULT_OOM; 249822cd25edSBenjamin Herrenschmidt else if (unlikely(pfn == NOPFN_SIGBUS)) 2499f4b81804SJes Sorensen return VM_FAULT_SIGBUS; 250022cd25edSBenjamin Herrenschmidt else if (unlikely(pfn == NOPFN_REFAULT)) 2501*83c54070SNick Piggin return 0; 2502f4b81804SJes Sorensen 2503f4b81804SJes Sorensen page_table = pte_offset_map_lock(mm, pmd, address, &ptl); 2504f4b81804SJes Sorensen 2505f4b81804SJes Sorensen /* Only go through if we didn't race with anybody else... */ 2506f4b81804SJes Sorensen if (pte_none(*page_table)) { 2507f4b81804SJes Sorensen entry = pfn_pte(pfn, vma->vm_page_prot); 2508f4b81804SJes Sorensen if (write_access) 2509f4b81804SJes Sorensen entry = maybe_mkwrite(pte_mkdirty(entry), vma); 2510f4b81804SJes Sorensen set_pte_at(mm, address, page_table, entry); 2511f4b81804SJes Sorensen } 2512f4b81804SJes Sorensen pte_unmap_unlock(page_table, ptl); 2513*83c54070SNick Piggin return 0; 2514f4b81804SJes Sorensen } 2515f4b81804SJes Sorensen 2516f4b81804SJes Sorensen /* 25171da177e4SLinus Torvalds * Fault of a previously existing named mapping. Repopulate the pte 25181da177e4SLinus Torvalds * from the encoded file_pte if possible. This enables swappable 25191da177e4SLinus Torvalds * nonlinear vmas. 25208f4e2101SHugh Dickins * 25218f4e2101SHugh Dickins * We enter with non-exclusive mmap_sem (to exclude vma changes, 25228f4e2101SHugh Dickins * but allow concurrent faults), and pte mapped but not yet locked. 25238f4e2101SHugh Dickins * We return with mmap_sem still held, but pte unmapped and unlocked. 25241da177e4SLinus Torvalds */ 2525d0217ac0SNick Piggin static int do_nonlinear_fault(struct mm_struct *mm, struct vm_area_struct *vma, 252665500d23SHugh Dickins unsigned long address, pte_t *page_table, pmd_t *pmd, 252765500d23SHugh Dickins int write_access, pte_t orig_pte) 25281da177e4SLinus Torvalds { 2529d0217ac0SNick Piggin unsigned int flags = FAULT_FLAG_NONLINEAR | 2530d0217ac0SNick Piggin (write_access ? FAULT_FLAG_WRITE : 0); 253165500d23SHugh Dickins pgoff_t pgoff; 25321da177e4SLinus Torvalds 25334c21e2f2SHugh Dickins if (!pte_unmap_same(mm, pmd, page_table, orig_pte)) 2534*83c54070SNick Piggin return 0; 25351da177e4SLinus Torvalds 2536d0217ac0SNick Piggin if (unlikely(!(vma->vm_flags & VM_NONLINEAR) || 2537d0217ac0SNick Piggin !(vma->vm_flags & VM_CAN_NONLINEAR))) { 253865500d23SHugh Dickins /* 253965500d23SHugh Dickins * Page table corrupted: show pte and kill process. 254065500d23SHugh Dickins */ 2541b5810039SNick Piggin print_bad_pte(vma, orig_pte, address); 254265500d23SHugh Dickins return VM_FAULT_OOM; 254365500d23SHugh Dickins } 254465500d23SHugh Dickins 254565500d23SHugh Dickins pgoff = pte_to_pgoff(orig_pte); 254654cb8821SNick Piggin 2547d0217ac0SNick Piggin return __do_fault(mm, vma, address, page_table, pmd, pgoff, 2548d0217ac0SNick Piggin flags, orig_pte); 25491da177e4SLinus Torvalds } 25501da177e4SLinus Torvalds 25511da177e4SLinus Torvalds /* 25521da177e4SLinus Torvalds * These routines also need to handle stuff like marking pages dirty 25531da177e4SLinus Torvalds * and/or accessed for architectures that don't do it in hardware (most 25541da177e4SLinus Torvalds * RISC architectures). The early dirtying is also good on the i386. 25551da177e4SLinus Torvalds * 25561da177e4SLinus Torvalds * There is also a hook called "update_mmu_cache()" that architectures 25571da177e4SLinus Torvalds * with external mmu caches can use to update those (ie the Sparc or 25581da177e4SLinus Torvalds * PowerPC hashed page tables that act as extended TLBs). 25591da177e4SLinus Torvalds * 2560c74df32cSHugh Dickins * We enter with non-exclusive mmap_sem (to exclude vma changes, 2561c74df32cSHugh Dickins * but allow concurrent faults), and pte mapped but not yet locked. 2562c74df32cSHugh Dickins * We return with mmap_sem still held, but pte unmapped and unlocked. 25631da177e4SLinus Torvalds */ 25641da177e4SLinus Torvalds static inline int handle_pte_fault(struct mm_struct *mm, 25651da177e4SLinus Torvalds struct vm_area_struct *vma, unsigned long address, 256665500d23SHugh Dickins pte_t *pte, pmd_t *pmd, int write_access) 25671da177e4SLinus Torvalds { 25681da177e4SLinus Torvalds pte_t entry; 25698f4e2101SHugh Dickins spinlock_t *ptl; 25701da177e4SLinus Torvalds 25718dab5241SBenjamin Herrenschmidt entry = *pte; 25721da177e4SLinus Torvalds if (!pte_present(entry)) { 257365500d23SHugh Dickins if (pte_none(entry)) { 2574f4b81804SJes Sorensen if (vma->vm_ops) { 257554cb8821SNick Piggin if (vma->vm_ops->fault || vma->vm_ops->nopage) 257654cb8821SNick Piggin return do_linear_fault(mm, vma, address, 257754cb8821SNick Piggin pte, pmd, write_access, entry); 2578f4b81804SJes Sorensen if (unlikely(vma->vm_ops->nopfn)) 2579f4b81804SJes Sorensen return do_no_pfn(mm, vma, address, pte, 2580f4b81804SJes Sorensen pmd, write_access); 2581f4b81804SJes Sorensen } 2582f4b81804SJes Sorensen return do_anonymous_page(mm, vma, address, 258365500d23SHugh Dickins pte, pmd, write_access); 258465500d23SHugh Dickins } 25851da177e4SLinus Torvalds if (pte_file(entry)) 2586d0217ac0SNick Piggin return do_nonlinear_fault(mm, vma, address, 258765500d23SHugh Dickins pte, pmd, write_access, entry); 258865500d23SHugh Dickins return do_swap_page(mm, vma, address, 258965500d23SHugh Dickins pte, pmd, write_access, entry); 25901da177e4SLinus Torvalds } 25911da177e4SLinus Torvalds 25924c21e2f2SHugh Dickins ptl = pte_lockptr(mm, pmd); 25938f4e2101SHugh Dickins spin_lock(ptl); 25948f4e2101SHugh Dickins if (unlikely(!pte_same(*pte, entry))) 25958f4e2101SHugh Dickins goto unlock; 25961da177e4SLinus Torvalds if (write_access) { 25971da177e4SLinus Torvalds if (!pte_write(entry)) 25988f4e2101SHugh Dickins return do_wp_page(mm, vma, address, 25998f4e2101SHugh Dickins pte, pmd, ptl, entry); 26001da177e4SLinus Torvalds entry = pte_mkdirty(entry); 26011da177e4SLinus Torvalds } 26021da177e4SLinus Torvalds entry = pte_mkyoung(entry); 26038dab5241SBenjamin Herrenschmidt if (ptep_set_access_flags(vma, address, pte, entry, write_access)) { 26041da177e4SLinus Torvalds update_mmu_cache(vma, address, entry); 26051da177e4SLinus Torvalds lazy_mmu_prot_update(entry); 26061a44e149SAndrea Arcangeli } else { 26071a44e149SAndrea Arcangeli /* 26081a44e149SAndrea Arcangeli * This is needed only for protection faults but the arch code 26091a44e149SAndrea Arcangeli * is not yet telling us if this is a protection fault or not. 26101a44e149SAndrea Arcangeli * This still avoids useless tlb flushes for .text page faults 26111a44e149SAndrea Arcangeli * with threads. 26121a44e149SAndrea Arcangeli */ 26131a44e149SAndrea Arcangeli if (write_access) 26141a44e149SAndrea Arcangeli flush_tlb_page(vma, address); 26151a44e149SAndrea Arcangeli } 26168f4e2101SHugh Dickins unlock: 26178f4e2101SHugh Dickins pte_unmap_unlock(pte, ptl); 2618*83c54070SNick Piggin return 0; 26191da177e4SLinus Torvalds } 26201da177e4SLinus Torvalds 26211da177e4SLinus Torvalds /* 26221da177e4SLinus Torvalds * By the time we get here, we already hold the mm semaphore 26231da177e4SLinus Torvalds */ 2624*83c54070SNick Piggin int handle_mm_fault(struct mm_struct *mm, struct vm_area_struct *vma, 26251da177e4SLinus Torvalds unsigned long address, int write_access) 26261da177e4SLinus Torvalds { 26271da177e4SLinus Torvalds pgd_t *pgd; 26281da177e4SLinus Torvalds pud_t *pud; 26291da177e4SLinus Torvalds pmd_t *pmd; 26301da177e4SLinus Torvalds pte_t *pte; 26311da177e4SLinus Torvalds 26321da177e4SLinus Torvalds __set_current_state(TASK_RUNNING); 26331da177e4SLinus Torvalds 2634f8891e5eSChristoph Lameter count_vm_event(PGFAULT); 26351da177e4SLinus Torvalds 2636ac9b9c66SHugh Dickins if (unlikely(is_vm_hugetlb_page(vma))) 2637ac9b9c66SHugh Dickins return hugetlb_fault(mm, vma, address, write_access); 26381da177e4SLinus Torvalds 26391da177e4SLinus Torvalds pgd = pgd_offset(mm, address); 26401da177e4SLinus Torvalds pud = pud_alloc(mm, pgd, address); 26411da177e4SLinus Torvalds if (!pud) 2642c74df32cSHugh Dickins return VM_FAULT_OOM; 26431da177e4SLinus Torvalds pmd = pmd_alloc(mm, pud, address); 26441da177e4SLinus Torvalds if (!pmd) 2645c74df32cSHugh Dickins return VM_FAULT_OOM; 26461da177e4SLinus Torvalds pte = pte_alloc_map(mm, pmd, address); 26471da177e4SLinus Torvalds if (!pte) 2648c74df32cSHugh Dickins return VM_FAULT_OOM; 26491da177e4SLinus Torvalds 265065500d23SHugh Dickins return handle_pte_fault(mm, vma, address, pte, pmd, write_access); 26511da177e4SLinus Torvalds } 26521da177e4SLinus Torvalds 2653*83c54070SNick Piggin EXPORT_SYMBOL_GPL(handle_mm_fault); 265467207b96SArnd Bergmann 26551da177e4SLinus Torvalds #ifndef __PAGETABLE_PUD_FOLDED 26561da177e4SLinus Torvalds /* 26571da177e4SLinus Torvalds * Allocate page upper directory. 2658872fec16SHugh Dickins * We've already handled the fast-path in-line. 26591da177e4SLinus Torvalds */ 26601bb3630eSHugh Dickins int __pud_alloc(struct mm_struct *mm, pgd_t *pgd, unsigned long address) 26611da177e4SLinus Torvalds { 2662c74df32cSHugh Dickins pud_t *new = pud_alloc_one(mm, address); 2663c74df32cSHugh Dickins if (!new) 26641bb3630eSHugh Dickins return -ENOMEM; 26651da177e4SLinus Torvalds 2666872fec16SHugh Dickins spin_lock(&mm->page_table_lock); 26671bb3630eSHugh Dickins if (pgd_present(*pgd)) /* Another has populated it */ 26681da177e4SLinus Torvalds pud_free(new); 26691bb3630eSHugh Dickins else 26701da177e4SLinus Torvalds pgd_populate(mm, pgd, new); 2671872fec16SHugh Dickins spin_unlock(&mm->page_table_lock); 26721bb3630eSHugh Dickins return 0; 26731da177e4SLinus Torvalds } 26741da177e4SLinus Torvalds #endif /* __PAGETABLE_PUD_FOLDED */ 26751da177e4SLinus Torvalds 26761da177e4SLinus Torvalds #ifndef __PAGETABLE_PMD_FOLDED 26771da177e4SLinus Torvalds /* 26781da177e4SLinus Torvalds * Allocate page middle directory. 2679872fec16SHugh Dickins * We've already handled the fast-path in-line. 26801da177e4SLinus Torvalds */ 26811bb3630eSHugh Dickins int __pmd_alloc(struct mm_struct *mm, pud_t *pud, unsigned long address) 26821da177e4SLinus Torvalds { 2683c74df32cSHugh Dickins pmd_t *new = pmd_alloc_one(mm, address); 2684c74df32cSHugh Dickins if (!new) 26851bb3630eSHugh Dickins return -ENOMEM; 26861da177e4SLinus Torvalds 2687872fec16SHugh Dickins spin_lock(&mm->page_table_lock); 26881da177e4SLinus Torvalds #ifndef __ARCH_HAS_4LEVEL_HACK 26891bb3630eSHugh Dickins if (pud_present(*pud)) /* Another has populated it */ 26901da177e4SLinus Torvalds pmd_free(new); 26911bb3630eSHugh Dickins else 26921da177e4SLinus Torvalds pud_populate(mm, pud, new); 26931da177e4SLinus Torvalds #else 26941bb3630eSHugh Dickins if (pgd_present(*pud)) /* Another has populated it */ 26951da177e4SLinus Torvalds pmd_free(new); 26961bb3630eSHugh Dickins else 26971da177e4SLinus Torvalds pgd_populate(mm, pud, new); 26981da177e4SLinus Torvalds #endif /* __ARCH_HAS_4LEVEL_HACK */ 2699872fec16SHugh Dickins spin_unlock(&mm->page_table_lock); 27001bb3630eSHugh Dickins return 0; 27011da177e4SLinus Torvalds } 27021da177e4SLinus Torvalds #endif /* __PAGETABLE_PMD_FOLDED */ 27031da177e4SLinus Torvalds 27041da177e4SLinus Torvalds int make_pages_present(unsigned long addr, unsigned long end) 27051da177e4SLinus Torvalds { 27061da177e4SLinus Torvalds int ret, len, write; 27071da177e4SLinus Torvalds struct vm_area_struct * vma; 27081da177e4SLinus Torvalds 27091da177e4SLinus Torvalds vma = find_vma(current->mm, addr); 27101da177e4SLinus Torvalds if (!vma) 27111da177e4SLinus Torvalds return -1; 27121da177e4SLinus Torvalds write = (vma->vm_flags & VM_WRITE) != 0; 27135bcb28b1SEric Sesterhenn BUG_ON(addr >= end); 27145bcb28b1SEric Sesterhenn BUG_ON(end > vma->vm_end); 271568e116a3SRolf Eike Beer len = DIV_ROUND_UP(end, PAGE_SIZE) - addr/PAGE_SIZE; 27161da177e4SLinus Torvalds ret = get_user_pages(current, current->mm, addr, 27171da177e4SLinus Torvalds len, write, 0, NULL, NULL); 27181da177e4SLinus Torvalds if (ret < 0) 27191da177e4SLinus Torvalds return ret; 27201da177e4SLinus Torvalds return ret == len ? 0 : -1; 27211da177e4SLinus Torvalds } 27221da177e4SLinus Torvalds 27231da177e4SLinus Torvalds /* 27241da177e4SLinus Torvalds * Map a vmalloc()-space virtual address to the physical page. 27251da177e4SLinus Torvalds */ 27261da177e4SLinus Torvalds struct page * vmalloc_to_page(void * vmalloc_addr) 27271da177e4SLinus Torvalds { 27281da177e4SLinus Torvalds unsigned long addr = (unsigned long) vmalloc_addr; 27291da177e4SLinus Torvalds struct page *page = NULL; 27301da177e4SLinus Torvalds pgd_t *pgd = pgd_offset_k(addr); 27311da177e4SLinus Torvalds pud_t *pud; 27321da177e4SLinus Torvalds pmd_t *pmd; 27331da177e4SLinus Torvalds pte_t *ptep, pte; 27341da177e4SLinus Torvalds 27351da177e4SLinus Torvalds if (!pgd_none(*pgd)) { 27361da177e4SLinus Torvalds pud = pud_offset(pgd, addr); 27371da177e4SLinus Torvalds if (!pud_none(*pud)) { 27381da177e4SLinus Torvalds pmd = pmd_offset(pud, addr); 27391da177e4SLinus Torvalds if (!pmd_none(*pmd)) { 27401da177e4SLinus Torvalds ptep = pte_offset_map(pmd, addr); 27411da177e4SLinus Torvalds pte = *ptep; 27421da177e4SLinus Torvalds if (pte_present(pte)) 27431da177e4SLinus Torvalds page = pte_page(pte); 27441da177e4SLinus Torvalds pte_unmap(ptep); 27451da177e4SLinus Torvalds } 27461da177e4SLinus Torvalds } 27471da177e4SLinus Torvalds } 27481da177e4SLinus Torvalds return page; 27491da177e4SLinus Torvalds } 27501da177e4SLinus Torvalds 27511da177e4SLinus Torvalds EXPORT_SYMBOL(vmalloc_to_page); 27521da177e4SLinus Torvalds 27531da177e4SLinus Torvalds /* 27541da177e4SLinus Torvalds * Map a vmalloc()-space virtual address to the physical page frame number. 27551da177e4SLinus Torvalds */ 27561da177e4SLinus Torvalds unsigned long vmalloc_to_pfn(void * vmalloc_addr) 27571da177e4SLinus Torvalds { 27581da177e4SLinus Torvalds return page_to_pfn(vmalloc_to_page(vmalloc_addr)); 27591da177e4SLinus Torvalds } 27601da177e4SLinus Torvalds 27611da177e4SLinus Torvalds EXPORT_SYMBOL(vmalloc_to_pfn); 27621da177e4SLinus Torvalds 27631da177e4SLinus Torvalds #if !defined(__HAVE_ARCH_GATE_AREA) 27641da177e4SLinus Torvalds 27651da177e4SLinus Torvalds #if defined(AT_SYSINFO_EHDR) 27665ce7852cSAdrian Bunk static struct vm_area_struct gate_vma; 27671da177e4SLinus Torvalds 27681da177e4SLinus Torvalds static int __init gate_vma_init(void) 27691da177e4SLinus Torvalds { 27701da177e4SLinus Torvalds gate_vma.vm_mm = NULL; 27711da177e4SLinus Torvalds gate_vma.vm_start = FIXADDR_USER_START; 27721da177e4SLinus Torvalds gate_vma.vm_end = FIXADDR_USER_END; 2773b6558c4aSRoland McGrath gate_vma.vm_flags = VM_READ | VM_MAYREAD | VM_EXEC | VM_MAYEXEC; 2774b6558c4aSRoland McGrath gate_vma.vm_page_prot = __P101; 2775f47aef55SRoland McGrath /* 2776f47aef55SRoland McGrath * Make sure the vDSO gets into every core dump. 2777f47aef55SRoland McGrath * Dumping its contents makes post-mortem fully interpretable later 2778f47aef55SRoland McGrath * without matching up the same kernel and hardware config to see 2779f47aef55SRoland McGrath * what PC values meant. 2780f47aef55SRoland McGrath */ 2781f47aef55SRoland McGrath gate_vma.vm_flags |= VM_ALWAYSDUMP; 27821da177e4SLinus Torvalds return 0; 27831da177e4SLinus Torvalds } 27841da177e4SLinus Torvalds __initcall(gate_vma_init); 27851da177e4SLinus Torvalds #endif 27861da177e4SLinus Torvalds 27871da177e4SLinus Torvalds struct vm_area_struct *get_gate_vma(struct task_struct *tsk) 27881da177e4SLinus Torvalds { 27891da177e4SLinus Torvalds #ifdef AT_SYSINFO_EHDR 27901da177e4SLinus Torvalds return &gate_vma; 27911da177e4SLinus Torvalds #else 27921da177e4SLinus Torvalds return NULL; 27931da177e4SLinus Torvalds #endif 27941da177e4SLinus Torvalds } 27951da177e4SLinus Torvalds 27961da177e4SLinus Torvalds int in_gate_area_no_task(unsigned long addr) 27971da177e4SLinus Torvalds { 27981da177e4SLinus Torvalds #ifdef AT_SYSINFO_EHDR 27991da177e4SLinus Torvalds if ((addr >= FIXADDR_USER_START) && (addr < FIXADDR_USER_END)) 28001da177e4SLinus Torvalds return 1; 28011da177e4SLinus Torvalds #endif 28021da177e4SLinus Torvalds return 0; 28031da177e4SLinus Torvalds } 28041da177e4SLinus Torvalds 28051da177e4SLinus Torvalds #endif /* __HAVE_ARCH_GATE_AREA */ 28060ec76a11SDavid Howells 28070ec76a11SDavid Howells /* 28080ec76a11SDavid Howells * Access another process' address space. 28090ec76a11SDavid Howells * Source/target buffer must be kernel space, 28100ec76a11SDavid Howells * Do not walk the page table directly, use get_user_pages 28110ec76a11SDavid Howells */ 28120ec76a11SDavid Howells int access_process_vm(struct task_struct *tsk, unsigned long addr, void *buf, int len, int write) 28130ec76a11SDavid Howells { 28140ec76a11SDavid Howells struct mm_struct *mm; 28150ec76a11SDavid Howells struct vm_area_struct *vma; 28160ec76a11SDavid Howells struct page *page; 28170ec76a11SDavid Howells void *old_buf = buf; 28180ec76a11SDavid Howells 28190ec76a11SDavid Howells mm = get_task_mm(tsk); 28200ec76a11SDavid Howells if (!mm) 28210ec76a11SDavid Howells return 0; 28220ec76a11SDavid Howells 28230ec76a11SDavid Howells down_read(&mm->mmap_sem); 28240ec76a11SDavid Howells /* ignore errors, just check how much was sucessfully transfered */ 28250ec76a11SDavid Howells while (len) { 28260ec76a11SDavid Howells int bytes, ret, offset; 28270ec76a11SDavid Howells void *maddr; 28280ec76a11SDavid Howells 28290ec76a11SDavid Howells ret = get_user_pages(tsk, mm, addr, 1, 28300ec76a11SDavid Howells write, 1, &page, &vma); 28310ec76a11SDavid Howells if (ret <= 0) 28320ec76a11SDavid Howells break; 28330ec76a11SDavid Howells 28340ec76a11SDavid Howells bytes = len; 28350ec76a11SDavid Howells offset = addr & (PAGE_SIZE-1); 28360ec76a11SDavid Howells if (bytes > PAGE_SIZE-offset) 28370ec76a11SDavid Howells bytes = PAGE_SIZE-offset; 28380ec76a11SDavid Howells 28390ec76a11SDavid Howells maddr = kmap(page); 28400ec76a11SDavid Howells if (write) { 28410ec76a11SDavid Howells copy_to_user_page(vma, page, addr, 28420ec76a11SDavid Howells maddr + offset, buf, bytes); 28430ec76a11SDavid Howells set_page_dirty_lock(page); 28440ec76a11SDavid Howells } else { 28450ec76a11SDavid Howells copy_from_user_page(vma, page, addr, 28460ec76a11SDavid Howells buf, maddr + offset, bytes); 28470ec76a11SDavid Howells } 28480ec76a11SDavid Howells kunmap(page); 28490ec76a11SDavid Howells page_cache_release(page); 28500ec76a11SDavid Howells len -= bytes; 28510ec76a11SDavid Howells buf += bytes; 28520ec76a11SDavid Howells addr += bytes; 28530ec76a11SDavid Howells } 28540ec76a11SDavid Howells up_read(&mm->mmap_sem); 28550ec76a11SDavid Howells mmput(mm); 28560ec76a11SDavid Howells 28570ec76a11SDavid Howells return buf - old_buf; 28580ec76a11SDavid Howells } 2859