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> 501da177e4SLinus Torvalds #include <linux/init.h> 511da177e4SLinus Torvalds 521da177e4SLinus Torvalds #include <asm/pgalloc.h> 531da177e4SLinus Torvalds #include <asm/uaccess.h> 541da177e4SLinus Torvalds #include <asm/tlb.h> 551da177e4SLinus Torvalds #include <asm/tlbflush.h> 561da177e4SLinus Torvalds #include <asm/pgtable.h> 571da177e4SLinus Torvalds 581da177e4SLinus Torvalds #include <linux/swapops.h> 591da177e4SLinus Torvalds #include <linux/elf.h> 601da177e4SLinus Torvalds 61d41dee36SAndy Whitcroft #ifndef CONFIG_NEED_MULTIPLE_NODES 621da177e4SLinus Torvalds /* use the per-pgdat data instead for discontigmem - mbligh */ 631da177e4SLinus Torvalds unsigned long max_mapnr; 641da177e4SLinus Torvalds struct page *mem_map; 651da177e4SLinus Torvalds 661da177e4SLinus Torvalds EXPORT_SYMBOL(max_mapnr); 671da177e4SLinus Torvalds EXPORT_SYMBOL(mem_map); 681da177e4SLinus Torvalds #endif 691da177e4SLinus Torvalds 701da177e4SLinus Torvalds unsigned long num_physpages; 711da177e4SLinus Torvalds /* 721da177e4SLinus Torvalds * A number of key systems in x86 including ioremap() rely on the assumption 731da177e4SLinus Torvalds * that high_memory defines the upper bound on direct map memory, then end 741da177e4SLinus Torvalds * of ZONE_NORMAL. Under CONFIG_DISCONTIG this means that max_low_pfn and 751da177e4SLinus Torvalds * highstart_pfn must be the same; there must be no gap between ZONE_NORMAL 761da177e4SLinus Torvalds * and ZONE_HIGHMEM. 771da177e4SLinus Torvalds */ 781da177e4SLinus Torvalds void * high_memory; 791da177e4SLinus Torvalds unsigned long vmalloc_earlyreserve; 801da177e4SLinus Torvalds 811da177e4SLinus Torvalds EXPORT_SYMBOL(num_physpages); 821da177e4SLinus Torvalds EXPORT_SYMBOL(high_memory); 831da177e4SLinus Torvalds EXPORT_SYMBOL(vmalloc_earlyreserve); 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); 4681da177e4SLinus Torvalds pte = *src_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); 4821da177e4SLinus Torvalds page_dup_rmap(page); 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); 506c74df32cSHugh Dickins spin_lock(src_ptl); 5071da177e4SLinus Torvalds 5081da177e4SLinus Torvalds do { 5091da177e4SLinus Torvalds /* 5101da177e4SLinus Torvalds * We are holding two locks at this point - either of them 5111da177e4SLinus Torvalds * could generate latencies in another task on another CPU. 5121da177e4SLinus Torvalds */ 513e040f218SHugh Dickins if (progress >= 32) { 514e040f218SHugh Dickins progress = 0; 515e040f218SHugh Dickins if (need_resched() || 516c74df32cSHugh Dickins need_lockbreak(src_ptl) || 517c74df32cSHugh Dickins need_lockbreak(dst_ptl)) 5181da177e4SLinus Torvalds break; 519e040f218SHugh Dickins } 5201da177e4SLinus Torvalds if (pte_none(*src_pte)) { 5211da177e4SLinus Torvalds progress++; 5221da177e4SLinus Torvalds continue; 5231da177e4SLinus Torvalds } 5248c103762SHugh Dickins copy_one_pte(dst_mm, src_mm, dst_pte, src_pte, vma, addr, rss); 5251da177e4SLinus Torvalds progress += 8; 5261da177e4SLinus Torvalds } while (dst_pte++, src_pte++, addr += PAGE_SIZE, addr != end); 5271da177e4SLinus Torvalds 528c74df32cSHugh Dickins spin_unlock(src_ptl); 5291da177e4SLinus Torvalds pte_unmap_nested(src_pte - 1); 530ae859762SHugh Dickins add_mm_rss(dst_mm, rss[0], rss[1]); 531c74df32cSHugh Dickins pte_unmap_unlock(dst_pte - 1, dst_ptl); 532c74df32cSHugh Dickins cond_resched(); 5331da177e4SLinus Torvalds if (addr != end) 5341da177e4SLinus Torvalds goto again; 5351da177e4SLinus Torvalds return 0; 5361da177e4SLinus Torvalds } 5371da177e4SLinus Torvalds 5381da177e4SLinus Torvalds static inline int copy_pmd_range(struct mm_struct *dst_mm, struct mm_struct *src_mm, 5391da177e4SLinus Torvalds pud_t *dst_pud, pud_t *src_pud, struct vm_area_struct *vma, 5401da177e4SLinus Torvalds unsigned long addr, unsigned long end) 5411da177e4SLinus Torvalds { 5421da177e4SLinus Torvalds pmd_t *src_pmd, *dst_pmd; 5431da177e4SLinus Torvalds unsigned long next; 5441da177e4SLinus Torvalds 5451da177e4SLinus Torvalds dst_pmd = pmd_alloc(dst_mm, dst_pud, addr); 5461da177e4SLinus Torvalds if (!dst_pmd) 5471da177e4SLinus Torvalds return -ENOMEM; 5481da177e4SLinus Torvalds src_pmd = pmd_offset(src_pud, addr); 5491da177e4SLinus Torvalds do { 5501da177e4SLinus Torvalds next = pmd_addr_end(addr, end); 5511da177e4SLinus Torvalds if (pmd_none_or_clear_bad(src_pmd)) 5521da177e4SLinus Torvalds continue; 5531da177e4SLinus Torvalds if (copy_pte_range(dst_mm, src_mm, dst_pmd, src_pmd, 5541da177e4SLinus Torvalds vma, addr, next)) 5551da177e4SLinus Torvalds return -ENOMEM; 5561da177e4SLinus Torvalds } while (dst_pmd++, src_pmd++, addr = next, addr != end); 5571da177e4SLinus Torvalds return 0; 5581da177e4SLinus Torvalds } 5591da177e4SLinus Torvalds 5601da177e4SLinus Torvalds static inline int copy_pud_range(struct mm_struct *dst_mm, struct mm_struct *src_mm, 5611da177e4SLinus Torvalds pgd_t *dst_pgd, pgd_t *src_pgd, struct vm_area_struct *vma, 5621da177e4SLinus Torvalds unsigned long addr, unsigned long end) 5631da177e4SLinus Torvalds { 5641da177e4SLinus Torvalds pud_t *src_pud, *dst_pud; 5651da177e4SLinus Torvalds unsigned long next; 5661da177e4SLinus Torvalds 5671da177e4SLinus Torvalds dst_pud = pud_alloc(dst_mm, dst_pgd, addr); 5681da177e4SLinus Torvalds if (!dst_pud) 5691da177e4SLinus Torvalds return -ENOMEM; 5701da177e4SLinus Torvalds src_pud = pud_offset(src_pgd, addr); 5711da177e4SLinus Torvalds do { 5721da177e4SLinus Torvalds next = pud_addr_end(addr, end); 5731da177e4SLinus Torvalds if (pud_none_or_clear_bad(src_pud)) 5741da177e4SLinus Torvalds continue; 5751da177e4SLinus Torvalds if (copy_pmd_range(dst_mm, src_mm, dst_pud, src_pud, 5761da177e4SLinus Torvalds vma, addr, next)) 5771da177e4SLinus Torvalds return -ENOMEM; 5781da177e4SLinus Torvalds } while (dst_pud++, src_pud++, addr = next, addr != end); 5791da177e4SLinus Torvalds return 0; 5801da177e4SLinus Torvalds } 5811da177e4SLinus Torvalds 5821da177e4SLinus Torvalds int copy_page_range(struct mm_struct *dst_mm, struct mm_struct *src_mm, 5831da177e4SLinus Torvalds struct vm_area_struct *vma) 5841da177e4SLinus Torvalds { 5851da177e4SLinus Torvalds pgd_t *src_pgd, *dst_pgd; 5861da177e4SLinus Torvalds unsigned long next; 5871da177e4SLinus Torvalds unsigned long addr = vma->vm_start; 5881da177e4SLinus Torvalds unsigned long end = vma->vm_end; 5891da177e4SLinus Torvalds 590d992895bSNick Piggin /* 591d992895bSNick Piggin * Don't copy ptes where a page fault will fill them correctly. 592d992895bSNick Piggin * Fork becomes much lighter when there are big shared or private 593d992895bSNick Piggin * readonly mappings. The tradeoff is that copy_page_range is more 594d992895bSNick Piggin * efficient than faulting. 595d992895bSNick Piggin */ 5964d7672b4SLinus Torvalds if (!(vma->vm_flags & (VM_HUGETLB|VM_NONLINEAR|VM_PFNMAP|VM_INSERTPAGE))) { 597d992895bSNick Piggin if (!vma->anon_vma) 598d992895bSNick Piggin return 0; 599d992895bSNick Piggin } 600d992895bSNick Piggin 6011da177e4SLinus Torvalds if (is_vm_hugetlb_page(vma)) 6021da177e4SLinus Torvalds return copy_hugetlb_page_range(dst_mm, src_mm, vma); 6031da177e4SLinus Torvalds 6041da177e4SLinus Torvalds dst_pgd = pgd_offset(dst_mm, addr); 6051da177e4SLinus Torvalds src_pgd = pgd_offset(src_mm, addr); 6061da177e4SLinus Torvalds do { 6071da177e4SLinus Torvalds next = pgd_addr_end(addr, end); 6081da177e4SLinus Torvalds if (pgd_none_or_clear_bad(src_pgd)) 6091da177e4SLinus Torvalds continue; 6101da177e4SLinus Torvalds if (copy_pud_range(dst_mm, src_mm, dst_pgd, src_pgd, 6111da177e4SLinus Torvalds vma, addr, next)) 6121da177e4SLinus Torvalds return -ENOMEM; 6131da177e4SLinus Torvalds } while (dst_pgd++, src_pgd++, addr = next, addr != end); 6141da177e4SLinus Torvalds return 0; 6151da177e4SLinus Torvalds } 6161da177e4SLinus Torvalds 61751c6f666SRobin Holt static unsigned long zap_pte_range(struct mmu_gather *tlb, 618b5810039SNick Piggin struct vm_area_struct *vma, pmd_t *pmd, 6191da177e4SLinus Torvalds unsigned long addr, unsigned long end, 62051c6f666SRobin Holt long *zap_work, struct zap_details *details) 6211da177e4SLinus Torvalds { 622b5810039SNick Piggin struct mm_struct *mm = tlb->mm; 6231da177e4SLinus Torvalds pte_t *pte; 624508034a3SHugh Dickins spinlock_t *ptl; 625ae859762SHugh Dickins int file_rss = 0; 626ae859762SHugh Dickins int anon_rss = 0; 6271da177e4SLinus Torvalds 628508034a3SHugh Dickins pte = pte_offset_map_lock(mm, pmd, addr, &ptl); 6291da177e4SLinus Torvalds do { 6301da177e4SLinus Torvalds pte_t ptent = *pte; 63151c6f666SRobin Holt if (pte_none(ptent)) { 63251c6f666SRobin Holt (*zap_work)--; 6331da177e4SLinus Torvalds continue; 63451c6f666SRobin Holt } 63551c6f666SRobin Holt 63651c6f666SRobin Holt (*zap_work) -= PAGE_SIZE; 63751c6f666SRobin Holt 6386f5e6b9eSHugh Dickins if (pte_present(ptent)) { 6396f5e6b9eSHugh Dickins struct page *page; 6406f5e6b9eSHugh Dickins 6416aab341eSLinus Torvalds page = vm_normal_page(vma, addr, ptent); 6421da177e4SLinus Torvalds if (unlikely(details) && page) { 6431da177e4SLinus Torvalds /* 6441da177e4SLinus Torvalds * unmap_shared_mapping_pages() wants to 6451da177e4SLinus Torvalds * invalidate cache without truncating: 6461da177e4SLinus Torvalds * unmap shared but keep private pages. 6471da177e4SLinus Torvalds */ 6481da177e4SLinus Torvalds if (details->check_mapping && 6491da177e4SLinus Torvalds details->check_mapping != page->mapping) 6501da177e4SLinus Torvalds continue; 6511da177e4SLinus Torvalds /* 6521da177e4SLinus Torvalds * Each page->index must be checked when 6531da177e4SLinus Torvalds * invalidating or truncating nonlinear. 6541da177e4SLinus Torvalds */ 6551da177e4SLinus Torvalds if (details->nonlinear_vma && 6561da177e4SLinus Torvalds (page->index < details->first_index || 6571da177e4SLinus Torvalds page->index > details->last_index)) 6581da177e4SLinus Torvalds continue; 6591da177e4SLinus Torvalds } 660b5810039SNick Piggin ptent = ptep_get_and_clear_full(mm, addr, pte, 661a600388dSZachary Amsden tlb->fullmm); 6621da177e4SLinus Torvalds tlb_remove_tlb_entry(tlb, pte, addr); 6631da177e4SLinus Torvalds if (unlikely(!page)) 6641da177e4SLinus Torvalds continue; 6651da177e4SLinus Torvalds if (unlikely(details) && details->nonlinear_vma 6661da177e4SLinus Torvalds && linear_page_index(details->nonlinear_vma, 6671da177e4SLinus Torvalds addr) != page->index) 668b5810039SNick Piggin set_pte_at(mm, addr, pte, 6691da177e4SLinus Torvalds pgoff_to_pte(page->index)); 6701da177e4SLinus Torvalds if (PageAnon(page)) 67186d912f4SHugh Dickins anon_rss--; 6726237bcd9SHugh Dickins else { 6736237bcd9SHugh Dickins if (pte_dirty(ptent)) 6746237bcd9SHugh Dickins set_page_dirty(page); 6756237bcd9SHugh Dickins if (pte_young(ptent)) 6761da177e4SLinus Torvalds mark_page_accessed(page); 67786d912f4SHugh Dickins file_rss--; 6786237bcd9SHugh Dickins } 6791da177e4SLinus Torvalds page_remove_rmap(page); 6801da177e4SLinus Torvalds tlb_remove_page(tlb, page); 6811da177e4SLinus Torvalds continue; 6821da177e4SLinus Torvalds } 6831da177e4SLinus Torvalds /* 6841da177e4SLinus Torvalds * If details->check_mapping, we leave swap entries; 6851da177e4SLinus Torvalds * if details->nonlinear_vma, we leave file entries. 6861da177e4SLinus Torvalds */ 6871da177e4SLinus Torvalds if (unlikely(details)) 6881da177e4SLinus Torvalds continue; 6891da177e4SLinus Torvalds if (!pte_file(ptent)) 6901da177e4SLinus Torvalds free_swap_and_cache(pte_to_swp_entry(ptent)); 691b5810039SNick Piggin pte_clear_full(mm, addr, pte, tlb->fullmm); 69251c6f666SRobin Holt } while (pte++, addr += PAGE_SIZE, (addr != end && *zap_work > 0)); 693ae859762SHugh Dickins 69486d912f4SHugh Dickins add_mm_rss(mm, file_rss, anon_rss); 695508034a3SHugh Dickins pte_unmap_unlock(pte - 1, ptl); 69651c6f666SRobin Holt 69751c6f666SRobin Holt return addr; 6981da177e4SLinus Torvalds } 6991da177e4SLinus Torvalds 70051c6f666SRobin Holt static inline unsigned long zap_pmd_range(struct mmu_gather *tlb, 701b5810039SNick Piggin struct vm_area_struct *vma, pud_t *pud, 7021da177e4SLinus Torvalds unsigned long addr, unsigned long end, 70351c6f666SRobin Holt long *zap_work, struct zap_details *details) 7041da177e4SLinus Torvalds { 7051da177e4SLinus Torvalds pmd_t *pmd; 7061da177e4SLinus Torvalds unsigned long next; 7071da177e4SLinus Torvalds 7081da177e4SLinus Torvalds pmd = pmd_offset(pud, addr); 7091da177e4SLinus Torvalds do { 7101da177e4SLinus Torvalds next = pmd_addr_end(addr, end); 71151c6f666SRobin Holt if (pmd_none_or_clear_bad(pmd)) { 71251c6f666SRobin Holt (*zap_work)--; 7131da177e4SLinus Torvalds continue; 71451c6f666SRobin Holt } 71551c6f666SRobin Holt next = zap_pte_range(tlb, vma, pmd, addr, next, 71651c6f666SRobin Holt zap_work, details); 71751c6f666SRobin Holt } while (pmd++, addr = next, (addr != end && *zap_work > 0)); 71851c6f666SRobin Holt 71951c6f666SRobin Holt return addr; 7201da177e4SLinus Torvalds } 7211da177e4SLinus Torvalds 72251c6f666SRobin Holt static inline unsigned long zap_pud_range(struct mmu_gather *tlb, 723b5810039SNick Piggin struct vm_area_struct *vma, pgd_t *pgd, 7241da177e4SLinus Torvalds unsigned long addr, unsigned long end, 72551c6f666SRobin Holt long *zap_work, struct zap_details *details) 7261da177e4SLinus Torvalds { 7271da177e4SLinus Torvalds pud_t *pud; 7281da177e4SLinus Torvalds unsigned long next; 7291da177e4SLinus Torvalds 7301da177e4SLinus Torvalds pud = pud_offset(pgd, addr); 7311da177e4SLinus Torvalds do { 7321da177e4SLinus Torvalds next = pud_addr_end(addr, end); 73351c6f666SRobin Holt if (pud_none_or_clear_bad(pud)) { 73451c6f666SRobin Holt (*zap_work)--; 7351da177e4SLinus Torvalds continue; 73651c6f666SRobin Holt } 73751c6f666SRobin Holt next = zap_pmd_range(tlb, vma, pud, addr, next, 73851c6f666SRobin Holt zap_work, details); 73951c6f666SRobin Holt } while (pud++, addr = next, (addr != end && *zap_work > 0)); 74051c6f666SRobin Holt 74151c6f666SRobin Holt return addr; 7421da177e4SLinus Torvalds } 7431da177e4SLinus Torvalds 74451c6f666SRobin Holt static unsigned long unmap_page_range(struct mmu_gather *tlb, 74551c6f666SRobin Holt struct vm_area_struct *vma, 7461da177e4SLinus Torvalds unsigned long addr, unsigned long end, 74751c6f666SRobin Holt long *zap_work, struct zap_details *details) 7481da177e4SLinus Torvalds { 7491da177e4SLinus Torvalds pgd_t *pgd; 7501da177e4SLinus Torvalds unsigned long next; 7511da177e4SLinus Torvalds 7521da177e4SLinus Torvalds if (details && !details->check_mapping && !details->nonlinear_vma) 7531da177e4SLinus Torvalds details = NULL; 7541da177e4SLinus Torvalds 7551da177e4SLinus Torvalds BUG_ON(addr >= end); 7561da177e4SLinus Torvalds tlb_start_vma(tlb, vma); 7571da177e4SLinus Torvalds pgd = pgd_offset(vma->vm_mm, addr); 7581da177e4SLinus Torvalds do { 7591da177e4SLinus Torvalds next = pgd_addr_end(addr, end); 76051c6f666SRobin Holt if (pgd_none_or_clear_bad(pgd)) { 76151c6f666SRobin Holt (*zap_work)--; 7621da177e4SLinus Torvalds continue; 76351c6f666SRobin Holt } 76451c6f666SRobin Holt next = zap_pud_range(tlb, vma, pgd, addr, next, 76551c6f666SRobin Holt zap_work, details); 76651c6f666SRobin Holt } while (pgd++, addr = next, (addr != end && *zap_work > 0)); 7671da177e4SLinus Torvalds tlb_end_vma(tlb, vma); 76851c6f666SRobin Holt 76951c6f666SRobin Holt return addr; 7701da177e4SLinus Torvalds } 7711da177e4SLinus Torvalds 7721da177e4SLinus Torvalds #ifdef CONFIG_PREEMPT 7731da177e4SLinus Torvalds # define ZAP_BLOCK_SIZE (8 * PAGE_SIZE) 7741da177e4SLinus Torvalds #else 7751da177e4SLinus Torvalds /* No preempt: go for improved straight-line efficiency */ 7761da177e4SLinus Torvalds # define ZAP_BLOCK_SIZE (1024 * PAGE_SIZE) 7771da177e4SLinus Torvalds #endif 7781da177e4SLinus Torvalds 7791da177e4SLinus Torvalds /** 7801da177e4SLinus Torvalds * unmap_vmas - unmap a range of memory covered by a list of vma's 7811da177e4SLinus Torvalds * @tlbp: address of the caller's struct mmu_gather 7821da177e4SLinus Torvalds * @vma: the starting vma 7831da177e4SLinus Torvalds * @start_addr: virtual address at which to start unmapping 7841da177e4SLinus Torvalds * @end_addr: virtual address at which to end unmapping 7851da177e4SLinus Torvalds * @nr_accounted: Place number of unmapped pages in vm-accountable vma's here 7861da177e4SLinus Torvalds * @details: details of nonlinear truncation or shared cache invalidation 7871da177e4SLinus Torvalds * 788ee39b37bSHugh Dickins * Returns the end address of the unmapping (restart addr if interrupted). 7891da177e4SLinus Torvalds * 790508034a3SHugh Dickins * Unmap all pages in the vma list. 7911da177e4SLinus Torvalds * 792508034a3SHugh Dickins * We aim to not hold locks for too long (for scheduling latency reasons). 793508034a3SHugh Dickins * So zap pages in ZAP_BLOCK_SIZE bytecounts. This means we need to 7941da177e4SLinus Torvalds * return the ending mmu_gather to the caller. 7951da177e4SLinus Torvalds * 7961da177e4SLinus Torvalds * Only addresses between `start' and `end' will be unmapped. 7971da177e4SLinus Torvalds * 7981da177e4SLinus Torvalds * The VMA list must be sorted in ascending virtual address order. 7991da177e4SLinus Torvalds * 8001da177e4SLinus Torvalds * unmap_vmas() assumes that the caller will flush the whole unmapped address 8011da177e4SLinus Torvalds * range after unmap_vmas() returns. So the only responsibility here is to 8021da177e4SLinus Torvalds * ensure that any thus-far unmapped pages are flushed before unmap_vmas() 8031da177e4SLinus Torvalds * drops the lock and schedules. 8041da177e4SLinus Torvalds */ 805508034a3SHugh Dickins unsigned long unmap_vmas(struct mmu_gather **tlbp, 8061da177e4SLinus Torvalds struct vm_area_struct *vma, unsigned long start_addr, 8071da177e4SLinus Torvalds unsigned long end_addr, unsigned long *nr_accounted, 8081da177e4SLinus Torvalds struct zap_details *details) 8091da177e4SLinus Torvalds { 81051c6f666SRobin Holt long zap_work = ZAP_BLOCK_SIZE; 8111da177e4SLinus Torvalds unsigned long tlb_start = 0; /* For tlb_finish_mmu */ 8121da177e4SLinus Torvalds int tlb_start_valid = 0; 813ee39b37bSHugh Dickins unsigned long start = start_addr; 8141da177e4SLinus Torvalds spinlock_t *i_mmap_lock = details? details->i_mmap_lock: NULL; 8154d6ddfa9SHugh Dickins int fullmm = (*tlbp)->fullmm; 8161da177e4SLinus Torvalds 8171da177e4SLinus Torvalds for ( ; vma && vma->vm_start < end_addr; vma = vma->vm_next) { 8181da177e4SLinus Torvalds unsigned long end; 8191da177e4SLinus Torvalds 8201da177e4SLinus Torvalds start = max(vma->vm_start, start_addr); 8211da177e4SLinus Torvalds if (start >= vma->vm_end) 8221da177e4SLinus Torvalds continue; 8231da177e4SLinus Torvalds end = min(vma->vm_end, end_addr); 8241da177e4SLinus Torvalds if (end <= vma->vm_start) 8251da177e4SLinus Torvalds continue; 8261da177e4SLinus Torvalds 8271da177e4SLinus Torvalds if (vma->vm_flags & VM_ACCOUNT) 8281da177e4SLinus Torvalds *nr_accounted += (end - start) >> PAGE_SHIFT; 8291da177e4SLinus Torvalds 8301da177e4SLinus Torvalds while (start != end) { 8311da177e4SLinus Torvalds if (!tlb_start_valid) { 8321da177e4SLinus Torvalds tlb_start = start; 8331da177e4SLinus Torvalds tlb_start_valid = 1; 8341da177e4SLinus Torvalds } 8351da177e4SLinus Torvalds 83651c6f666SRobin Holt if (unlikely(is_vm_hugetlb_page(vma))) { 8371da177e4SLinus Torvalds unmap_hugepage_range(vma, start, end); 83851c6f666SRobin Holt zap_work -= (end - start) / 83951c6f666SRobin Holt (HPAGE_SIZE / PAGE_SIZE); 84051c6f666SRobin Holt start = end; 84151c6f666SRobin Holt } else 84251c6f666SRobin Holt start = unmap_page_range(*tlbp, vma, 84351c6f666SRobin Holt start, end, &zap_work, details); 8441da177e4SLinus Torvalds 84551c6f666SRobin Holt if (zap_work > 0) { 84651c6f666SRobin Holt BUG_ON(start != end); 84751c6f666SRobin Holt break; 84851c6f666SRobin Holt } 8491da177e4SLinus Torvalds 8501da177e4SLinus Torvalds tlb_finish_mmu(*tlbp, tlb_start, start); 8511da177e4SLinus Torvalds 8521da177e4SLinus Torvalds if (need_resched() || 8531da177e4SLinus Torvalds (i_mmap_lock && need_lockbreak(i_mmap_lock))) { 8541da177e4SLinus Torvalds if (i_mmap_lock) { 855508034a3SHugh Dickins *tlbp = NULL; 8561da177e4SLinus Torvalds goto out; 8571da177e4SLinus Torvalds } 8581da177e4SLinus Torvalds cond_resched(); 8591da177e4SLinus Torvalds } 8601da177e4SLinus Torvalds 861508034a3SHugh Dickins *tlbp = tlb_gather_mmu(vma->vm_mm, fullmm); 8621da177e4SLinus Torvalds tlb_start_valid = 0; 86351c6f666SRobin Holt zap_work = ZAP_BLOCK_SIZE; 8641da177e4SLinus Torvalds } 8651da177e4SLinus Torvalds } 8661da177e4SLinus Torvalds out: 867ee39b37bSHugh Dickins return start; /* which is now the end (or restart) address */ 8681da177e4SLinus Torvalds } 8691da177e4SLinus Torvalds 8701da177e4SLinus Torvalds /** 8711da177e4SLinus Torvalds * zap_page_range - remove user pages in a given range 8721da177e4SLinus Torvalds * @vma: vm_area_struct holding the applicable pages 8731da177e4SLinus Torvalds * @address: starting address of pages to zap 8741da177e4SLinus Torvalds * @size: number of bytes to zap 8751da177e4SLinus Torvalds * @details: details of nonlinear truncation or shared cache invalidation 8761da177e4SLinus Torvalds */ 877ee39b37bSHugh Dickins unsigned long zap_page_range(struct vm_area_struct *vma, unsigned long address, 8781da177e4SLinus Torvalds unsigned long size, struct zap_details *details) 8791da177e4SLinus Torvalds { 8801da177e4SLinus Torvalds struct mm_struct *mm = vma->vm_mm; 8811da177e4SLinus Torvalds struct mmu_gather *tlb; 8821da177e4SLinus Torvalds unsigned long end = address + size; 8831da177e4SLinus Torvalds unsigned long nr_accounted = 0; 8841da177e4SLinus Torvalds 8851da177e4SLinus Torvalds lru_add_drain(); 8861da177e4SLinus Torvalds tlb = tlb_gather_mmu(mm, 0); 887365e9c87SHugh Dickins update_hiwater_rss(mm); 888508034a3SHugh Dickins end = unmap_vmas(&tlb, vma, address, end, &nr_accounted, details); 889508034a3SHugh Dickins if (tlb) 8908f4f8c16SHugh Dickins tlb_finish_mmu(tlb, address, end); 891ee39b37bSHugh Dickins return end; 8921da177e4SLinus Torvalds } 8931da177e4SLinus Torvalds 8941da177e4SLinus Torvalds /* 8951da177e4SLinus Torvalds * Do a quick page-table lookup for a single page. 8961da177e4SLinus Torvalds */ 8976aab341eSLinus Torvalds struct page *follow_page(struct vm_area_struct *vma, unsigned long address, 898deceb6cdSHugh Dickins unsigned int flags) 8991da177e4SLinus Torvalds { 9001da177e4SLinus Torvalds pgd_t *pgd; 9011da177e4SLinus Torvalds pud_t *pud; 9021da177e4SLinus Torvalds pmd_t *pmd; 9031da177e4SLinus Torvalds pte_t *ptep, pte; 904deceb6cdSHugh Dickins spinlock_t *ptl; 9051da177e4SLinus Torvalds struct page *page; 9066aab341eSLinus Torvalds struct mm_struct *mm = vma->vm_mm; 9071da177e4SLinus Torvalds 908deceb6cdSHugh Dickins page = follow_huge_addr(mm, address, flags & FOLL_WRITE); 909deceb6cdSHugh Dickins if (!IS_ERR(page)) { 910deceb6cdSHugh Dickins BUG_ON(flags & FOLL_GET); 911deceb6cdSHugh Dickins goto out; 912deceb6cdSHugh Dickins } 9131da177e4SLinus Torvalds 914deceb6cdSHugh Dickins page = NULL; 9151da177e4SLinus Torvalds pgd = pgd_offset(mm, address); 9161da177e4SLinus Torvalds if (pgd_none(*pgd) || unlikely(pgd_bad(*pgd))) 917deceb6cdSHugh Dickins goto no_page_table; 9181da177e4SLinus Torvalds 9191da177e4SLinus Torvalds pud = pud_offset(pgd, address); 9201da177e4SLinus Torvalds if (pud_none(*pud) || unlikely(pud_bad(*pud))) 921deceb6cdSHugh Dickins goto no_page_table; 9221da177e4SLinus Torvalds 9231da177e4SLinus Torvalds pmd = pmd_offset(pud, address); 9241da177e4SLinus Torvalds if (pmd_none(*pmd) || unlikely(pmd_bad(*pmd))) 925deceb6cdSHugh Dickins goto no_page_table; 9261da177e4SLinus Torvalds 927deceb6cdSHugh Dickins if (pmd_huge(*pmd)) { 928deceb6cdSHugh Dickins BUG_ON(flags & FOLL_GET); 929deceb6cdSHugh Dickins page = follow_huge_pmd(mm, address, pmd, flags & FOLL_WRITE); 930deceb6cdSHugh Dickins goto out; 931deceb6cdSHugh Dickins } 932deceb6cdSHugh Dickins 933deceb6cdSHugh Dickins ptep = pte_offset_map_lock(mm, pmd, address, &ptl); 9341da177e4SLinus Torvalds if (!ptep) 9351da177e4SLinus Torvalds goto out; 9361da177e4SLinus Torvalds 9371da177e4SLinus Torvalds pte = *ptep; 938deceb6cdSHugh Dickins if (!pte_present(pte)) 939deceb6cdSHugh Dickins goto unlock; 940deceb6cdSHugh Dickins if ((flags & FOLL_WRITE) && !pte_write(pte)) 941deceb6cdSHugh Dickins goto unlock; 9426aab341eSLinus Torvalds page = vm_normal_page(vma, address, pte); 9436aab341eSLinus Torvalds if (unlikely(!page)) 944deceb6cdSHugh Dickins goto unlock; 945deceb6cdSHugh Dickins 946deceb6cdSHugh Dickins if (flags & FOLL_GET) 947deceb6cdSHugh Dickins get_page(page); 948deceb6cdSHugh Dickins if (flags & FOLL_TOUCH) { 949deceb6cdSHugh Dickins if ((flags & FOLL_WRITE) && 950deceb6cdSHugh Dickins !pte_dirty(pte) && !PageDirty(page)) 951f33ea7f4SNick Piggin set_page_dirty(page); 9521da177e4SLinus Torvalds mark_page_accessed(page); 9531da177e4SLinus Torvalds } 954deceb6cdSHugh Dickins unlock: 955deceb6cdSHugh Dickins pte_unmap_unlock(ptep, ptl); 9561da177e4SLinus Torvalds out: 957deceb6cdSHugh Dickins return page; 958deceb6cdSHugh Dickins 959deceb6cdSHugh Dickins no_page_table: 960deceb6cdSHugh Dickins /* 961deceb6cdSHugh Dickins * When core dumping an enormous anonymous area that nobody 962deceb6cdSHugh Dickins * has touched so far, we don't want to allocate page tables. 963deceb6cdSHugh Dickins */ 964deceb6cdSHugh Dickins if (flags & FOLL_ANON) { 965deceb6cdSHugh Dickins page = ZERO_PAGE(address); 966deceb6cdSHugh Dickins if (flags & FOLL_GET) 967deceb6cdSHugh Dickins get_page(page); 968deceb6cdSHugh Dickins BUG_ON(flags & FOLL_WRITE); 9691da177e4SLinus Torvalds } 970deceb6cdSHugh Dickins return page; 9711da177e4SLinus Torvalds } 9721da177e4SLinus Torvalds 9731da177e4SLinus Torvalds int get_user_pages(struct task_struct *tsk, struct mm_struct *mm, 9741da177e4SLinus Torvalds unsigned long start, int len, int write, int force, 9751da177e4SLinus Torvalds struct page **pages, struct vm_area_struct **vmas) 9761da177e4SLinus Torvalds { 9771da177e4SLinus Torvalds int i; 978deceb6cdSHugh Dickins unsigned int vm_flags; 9791da177e4SLinus Torvalds 9801da177e4SLinus Torvalds /* 9811da177e4SLinus Torvalds * Require read or write permissions. 9821da177e4SLinus Torvalds * If 'force' is set, we only require the "MAY" flags. 9831da177e4SLinus Torvalds */ 984deceb6cdSHugh Dickins vm_flags = write ? (VM_WRITE | VM_MAYWRITE) : (VM_READ | VM_MAYREAD); 985deceb6cdSHugh Dickins vm_flags &= force ? (VM_MAYREAD | VM_MAYWRITE) : (VM_READ | VM_WRITE); 9861da177e4SLinus Torvalds i = 0; 9871da177e4SLinus Torvalds 9881da177e4SLinus Torvalds do { 9891da177e4SLinus Torvalds struct vm_area_struct *vma; 990deceb6cdSHugh Dickins unsigned int foll_flags; 9911da177e4SLinus Torvalds 9921da177e4SLinus Torvalds vma = find_extend_vma(mm, start); 9931da177e4SLinus Torvalds if (!vma && in_gate_area(tsk, start)) { 9941da177e4SLinus Torvalds unsigned long pg = start & PAGE_MASK; 9951da177e4SLinus Torvalds struct vm_area_struct *gate_vma = get_gate_vma(tsk); 9961da177e4SLinus Torvalds pgd_t *pgd; 9971da177e4SLinus Torvalds pud_t *pud; 9981da177e4SLinus Torvalds pmd_t *pmd; 9991da177e4SLinus Torvalds pte_t *pte; 10001da177e4SLinus Torvalds if (write) /* user gate pages are read-only */ 10011da177e4SLinus Torvalds return i ? : -EFAULT; 10021da177e4SLinus Torvalds if (pg > TASK_SIZE) 10031da177e4SLinus Torvalds pgd = pgd_offset_k(pg); 10041da177e4SLinus Torvalds else 10051da177e4SLinus Torvalds pgd = pgd_offset_gate(mm, pg); 10061da177e4SLinus Torvalds BUG_ON(pgd_none(*pgd)); 10071da177e4SLinus Torvalds pud = pud_offset(pgd, pg); 10081da177e4SLinus Torvalds BUG_ON(pud_none(*pud)); 10091da177e4SLinus Torvalds pmd = pmd_offset(pud, pg); 1010690dbe1cSHugh Dickins if (pmd_none(*pmd)) 1011690dbe1cSHugh Dickins return i ? : -EFAULT; 10121da177e4SLinus Torvalds pte = pte_offset_map(pmd, pg); 1013690dbe1cSHugh Dickins if (pte_none(*pte)) { 1014690dbe1cSHugh Dickins pte_unmap(pte); 1015690dbe1cSHugh Dickins return i ? : -EFAULT; 1016690dbe1cSHugh Dickins } 10171da177e4SLinus Torvalds if (pages) { 1018fa2a455bSNick Piggin struct page *page = vm_normal_page(gate_vma, start, *pte); 10196aab341eSLinus Torvalds pages[i] = page; 10206aab341eSLinus Torvalds if (page) 10216aab341eSLinus Torvalds get_page(page); 10221da177e4SLinus Torvalds } 10231da177e4SLinus Torvalds pte_unmap(pte); 10241da177e4SLinus Torvalds if (vmas) 10251da177e4SLinus Torvalds vmas[i] = gate_vma; 10261da177e4SLinus Torvalds i++; 10271da177e4SLinus Torvalds start += PAGE_SIZE; 10281da177e4SLinus Torvalds len--; 10291da177e4SLinus Torvalds continue; 10301da177e4SLinus Torvalds } 10311da177e4SLinus Torvalds 10321ff80389SLinus Torvalds if (!vma || (vma->vm_flags & (VM_IO | VM_PFNMAP)) 1033deceb6cdSHugh Dickins || !(vm_flags & vma->vm_flags)) 10341da177e4SLinus Torvalds return i ? : -EFAULT; 10351da177e4SLinus Torvalds 10361da177e4SLinus Torvalds if (is_vm_hugetlb_page(vma)) { 10371da177e4SLinus Torvalds i = follow_hugetlb_page(mm, vma, pages, vmas, 10381da177e4SLinus Torvalds &start, &len, i); 10391da177e4SLinus Torvalds continue; 10401da177e4SLinus Torvalds } 1041deceb6cdSHugh Dickins 1042deceb6cdSHugh Dickins foll_flags = FOLL_TOUCH; 1043deceb6cdSHugh Dickins if (pages) 1044deceb6cdSHugh Dickins foll_flags |= FOLL_GET; 1045deceb6cdSHugh Dickins if (!write && !(vma->vm_flags & VM_LOCKED) && 1046deceb6cdSHugh Dickins (!vma->vm_ops || !vma->vm_ops->nopage)) 1047deceb6cdSHugh Dickins foll_flags |= FOLL_ANON; 1048deceb6cdSHugh Dickins 10491da177e4SLinus Torvalds do { 105008ef4729SHugh Dickins struct page *page; 10511da177e4SLinus Torvalds 1052deceb6cdSHugh Dickins if (write) 1053deceb6cdSHugh Dickins foll_flags |= FOLL_WRITE; 1054deceb6cdSHugh Dickins 1055deceb6cdSHugh Dickins cond_resched(); 10566aab341eSLinus Torvalds while (!(page = follow_page(vma, start, foll_flags))) { 1057a68d2ebcSLinus Torvalds int ret; 1058deceb6cdSHugh Dickins ret = __handle_mm_fault(mm, vma, start, 1059deceb6cdSHugh Dickins foll_flags & FOLL_WRITE); 1060f33ea7f4SNick Piggin /* 1061a68d2ebcSLinus Torvalds * The VM_FAULT_WRITE bit tells us that do_wp_page has 1062a68d2ebcSLinus Torvalds * broken COW when necessary, even if maybe_mkwrite 1063a68d2ebcSLinus Torvalds * decided not to set pte_write. We can thus safely do 1064a68d2ebcSLinus Torvalds * subsequent page lookups as if they were reads. 1065f33ea7f4SNick Piggin */ 1066a68d2ebcSLinus Torvalds if (ret & VM_FAULT_WRITE) 1067deceb6cdSHugh Dickins foll_flags &= ~FOLL_WRITE; 1068a68d2ebcSLinus Torvalds 1069a68d2ebcSLinus Torvalds switch (ret & ~VM_FAULT_WRITE) { 10701da177e4SLinus Torvalds case VM_FAULT_MINOR: 10711da177e4SLinus Torvalds tsk->min_flt++; 10721da177e4SLinus Torvalds break; 10731da177e4SLinus Torvalds case VM_FAULT_MAJOR: 10741da177e4SLinus Torvalds tsk->maj_flt++; 10751da177e4SLinus Torvalds break; 10761da177e4SLinus Torvalds case VM_FAULT_SIGBUS: 10771da177e4SLinus Torvalds return i ? i : -EFAULT; 10781da177e4SLinus Torvalds case VM_FAULT_OOM: 10791da177e4SLinus Torvalds return i ? i : -ENOMEM; 10801da177e4SLinus Torvalds default: 10811da177e4SLinus Torvalds BUG(); 10821da177e4SLinus Torvalds } 10831da177e4SLinus Torvalds } 10841da177e4SLinus Torvalds if (pages) { 108508ef4729SHugh Dickins pages[i] = page; 108603beb076SJames Bottomley 108703beb076SJames Bottomley flush_anon_page(page, start); 108808ef4729SHugh Dickins flush_dcache_page(page); 10891da177e4SLinus Torvalds } 10901da177e4SLinus Torvalds if (vmas) 10911da177e4SLinus Torvalds vmas[i] = vma; 10921da177e4SLinus Torvalds i++; 10931da177e4SLinus Torvalds start += PAGE_SIZE; 10941da177e4SLinus Torvalds len--; 10951da177e4SLinus Torvalds } while (len && start < vma->vm_end); 10961da177e4SLinus Torvalds } while (len); 10971da177e4SLinus Torvalds return i; 10981da177e4SLinus Torvalds } 10991da177e4SLinus Torvalds EXPORT_SYMBOL(get_user_pages); 11001da177e4SLinus Torvalds 11011da177e4SLinus Torvalds static int zeromap_pte_range(struct mm_struct *mm, pmd_t *pmd, 11021da177e4SLinus Torvalds unsigned long addr, unsigned long end, pgprot_t prot) 11031da177e4SLinus Torvalds { 11041da177e4SLinus Torvalds pte_t *pte; 1105c74df32cSHugh Dickins spinlock_t *ptl; 11061da177e4SLinus Torvalds 1107c74df32cSHugh Dickins pte = pte_alloc_map_lock(mm, pmd, addr, &ptl); 11081da177e4SLinus Torvalds if (!pte) 11091da177e4SLinus Torvalds return -ENOMEM; 11101da177e4SLinus Torvalds do { 1111b5810039SNick Piggin struct page *page = ZERO_PAGE(addr); 1112b5810039SNick Piggin pte_t zero_pte = pte_wrprotect(mk_pte(page, prot)); 1113b5810039SNick Piggin page_cache_get(page); 1114b5810039SNick Piggin page_add_file_rmap(page); 1115b5810039SNick Piggin inc_mm_counter(mm, file_rss); 11161da177e4SLinus Torvalds BUG_ON(!pte_none(*pte)); 11171da177e4SLinus Torvalds set_pte_at(mm, addr, pte, zero_pte); 11181da177e4SLinus Torvalds } while (pte++, addr += PAGE_SIZE, addr != end); 1119c74df32cSHugh Dickins pte_unmap_unlock(pte - 1, ptl); 11201da177e4SLinus Torvalds return 0; 11211da177e4SLinus Torvalds } 11221da177e4SLinus Torvalds 11231da177e4SLinus Torvalds static inline int zeromap_pmd_range(struct mm_struct *mm, pud_t *pud, 11241da177e4SLinus Torvalds unsigned long addr, unsigned long end, pgprot_t prot) 11251da177e4SLinus Torvalds { 11261da177e4SLinus Torvalds pmd_t *pmd; 11271da177e4SLinus Torvalds unsigned long next; 11281da177e4SLinus Torvalds 11291da177e4SLinus Torvalds pmd = pmd_alloc(mm, pud, addr); 11301da177e4SLinus Torvalds if (!pmd) 11311da177e4SLinus Torvalds return -ENOMEM; 11321da177e4SLinus Torvalds do { 11331da177e4SLinus Torvalds next = pmd_addr_end(addr, end); 11341da177e4SLinus Torvalds if (zeromap_pte_range(mm, pmd, addr, next, prot)) 11351da177e4SLinus Torvalds return -ENOMEM; 11361da177e4SLinus Torvalds } while (pmd++, addr = next, addr != end); 11371da177e4SLinus Torvalds return 0; 11381da177e4SLinus Torvalds } 11391da177e4SLinus Torvalds 11401da177e4SLinus Torvalds static inline int zeromap_pud_range(struct mm_struct *mm, pgd_t *pgd, 11411da177e4SLinus Torvalds unsigned long addr, unsigned long end, pgprot_t prot) 11421da177e4SLinus Torvalds { 11431da177e4SLinus Torvalds pud_t *pud; 11441da177e4SLinus Torvalds unsigned long next; 11451da177e4SLinus Torvalds 11461da177e4SLinus Torvalds pud = pud_alloc(mm, pgd, addr); 11471da177e4SLinus Torvalds if (!pud) 11481da177e4SLinus Torvalds return -ENOMEM; 11491da177e4SLinus Torvalds do { 11501da177e4SLinus Torvalds next = pud_addr_end(addr, end); 11511da177e4SLinus Torvalds if (zeromap_pmd_range(mm, pud, addr, next, prot)) 11521da177e4SLinus Torvalds return -ENOMEM; 11531da177e4SLinus Torvalds } while (pud++, addr = next, addr != end); 11541da177e4SLinus Torvalds return 0; 11551da177e4SLinus Torvalds } 11561da177e4SLinus Torvalds 11571da177e4SLinus Torvalds int zeromap_page_range(struct vm_area_struct *vma, 11581da177e4SLinus Torvalds unsigned long addr, unsigned long size, pgprot_t prot) 11591da177e4SLinus Torvalds { 11601da177e4SLinus Torvalds pgd_t *pgd; 11611da177e4SLinus Torvalds unsigned long next; 11621da177e4SLinus Torvalds unsigned long end = addr + size; 11631da177e4SLinus Torvalds struct mm_struct *mm = vma->vm_mm; 11641da177e4SLinus Torvalds int err; 11651da177e4SLinus Torvalds 11661da177e4SLinus Torvalds BUG_ON(addr >= end); 11671da177e4SLinus Torvalds pgd = pgd_offset(mm, addr); 11681da177e4SLinus Torvalds flush_cache_range(vma, addr, end); 11691da177e4SLinus Torvalds do { 11701da177e4SLinus Torvalds next = pgd_addr_end(addr, end); 11711da177e4SLinus Torvalds err = zeromap_pud_range(mm, pgd, addr, next, prot); 11721da177e4SLinus Torvalds if (err) 11731da177e4SLinus Torvalds break; 11741da177e4SLinus Torvalds } while (pgd++, addr = next, addr != end); 11751da177e4SLinus Torvalds return err; 11761da177e4SLinus Torvalds } 11771da177e4SLinus Torvalds 117849c91fb0STrond Myklebust pte_t * fastcall get_locked_pte(struct mm_struct *mm, unsigned long addr, spinlock_t **ptl) 1179c9cfcddfSLinus Torvalds { 1180c9cfcddfSLinus Torvalds pgd_t * pgd = pgd_offset(mm, addr); 1181c9cfcddfSLinus Torvalds pud_t * pud = pud_alloc(mm, pgd, addr); 1182c9cfcddfSLinus Torvalds if (pud) { 118349c91fb0STrond Myklebust pmd_t * pmd = pmd_alloc(mm, pud, addr); 1184c9cfcddfSLinus Torvalds if (pmd) 1185c9cfcddfSLinus Torvalds return pte_alloc_map_lock(mm, pmd, addr, ptl); 1186c9cfcddfSLinus Torvalds } 1187c9cfcddfSLinus Torvalds return NULL; 1188c9cfcddfSLinus Torvalds } 1189c9cfcddfSLinus Torvalds 11901da177e4SLinus Torvalds /* 1191238f58d8SLinus Torvalds * This is the old fallback for page remapping. 1192238f58d8SLinus Torvalds * 1193238f58d8SLinus Torvalds * For historical reasons, it only allows reserved pages. Only 1194238f58d8SLinus Torvalds * old drivers should use this, and they needed to mark their 1195238f58d8SLinus Torvalds * pages reserved for the old functions anyway. 1196238f58d8SLinus Torvalds */ 1197238f58d8SLinus Torvalds static int insert_page(struct mm_struct *mm, unsigned long addr, struct page *page, pgprot_t prot) 1198238f58d8SLinus Torvalds { 1199238f58d8SLinus Torvalds int retval; 1200238f58d8SLinus Torvalds pte_t *pte; 1201238f58d8SLinus Torvalds spinlock_t *ptl; 1202238f58d8SLinus Torvalds 1203238f58d8SLinus Torvalds retval = -EINVAL; 1204a145dd41SLinus Torvalds if (PageAnon(page)) 1205238f58d8SLinus Torvalds goto out; 1206238f58d8SLinus Torvalds retval = -ENOMEM; 1207238f58d8SLinus Torvalds flush_dcache_page(page); 1208c9cfcddfSLinus Torvalds pte = get_locked_pte(mm, addr, &ptl); 1209238f58d8SLinus Torvalds if (!pte) 1210238f58d8SLinus Torvalds goto out; 1211238f58d8SLinus Torvalds retval = -EBUSY; 1212238f58d8SLinus Torvalds if (!pte_none(*pte)) 1213238f58d8SLinus Torvalds goto out_unlock; 1214238f58d8SLinus Torvalds 1215238f58d8SLinus Torvalds /* Ok, finally just insert the thing.. */ 1216238f58d8SLinus Torvalds get_page(page); 1217238f58d8SLinus Torvalds inc_mm_counter(mm, file_rss); 1218238f58d8SLinus Torvalds page_add_file_rmap(page); 1219238f58d8SLinus Torvalds set_pte_at(mm, addr, pte, mk_pte(page, prot)); 1220238f58d8SLinus Torvalds 1221238f58d8SLinus Torvalds retval = 0; 1222238f58d8SLinus Torvalds out_unlock: 1223238f58d8SLinus Torvalds pte_unmap_unlock(pte, ptl); 1224238f58d8SLinus Torvalds out: 1225238f58d8SLinus Torvalds return retval; 1226238f58d8SLinus Torvalds } 1227238f58d8SLinus Torvalds 1228238f58d8SLinus Torvalds /* 1229a145dd41SLinus Torvalds * This allows drivers to insert individual pages they've allocated 1230a145dd41SLinus Torvalds * into a user vma. 1231a145dd41SLinus Torvalds * 1232a145dd41SLinus Torvalds * The page has to be a nice clean _individual_ kernel allocation. 1233a145dd41SLinus Torvalds * If you allocate a compound page, you need to have marked it as 1234a145dd41SLinus Torvalds * such (__GFP_COMP), or manually just split the page up yourself 12358dfcc9baSNick Piggin * (see split_page()). 1236a145dd41SLinus Torvalds * 1237a145dd41SLinus Torvalds * NOTE! Traditionally this was done with "remap_pfn_range()" which 1238a145dd41SLinus Torvalds * took an arbitrary page protection parameter. This doesn't allow 1239a145dd41SLinus Torvalds * that. Your vma protection will have to be set up correctly, which 1240a145dd41SLinus Torvalds * means that if you want a shared writable mapping, you'd better 1241a145dd41SLinus Torvalds * ask for a shared writable mapping! 1242a145dd41SLinus Torvalds * 1243a145dd41SLinus Torvalds * The page does not need to be reserved. 1244a145dd41SLinus Torvalds */ 1245a145dd41SLinus Torvalds int vm_insert_page(struct vm_area_struct *vma, unsigned long addr, struct page *page) 1246a145dd41SLinus Torvalds { 1247a145dd41SLinus Torvalds if (addr < vma->vm_start || addr >= vma->vm_end) 1248a145dd41SLinus Torvalds return -EFAULT; 1249a145dd41SLinus Torvalds if (!page_count(page)) 1250a145dd41SLinus Torvalds return -EINVAL; 12514d7672b4SLinus Torvalds vma->vm_flags |= VM_INSERTPAGE; 1252a145dd41SLinus Torvalds return insert_page(vma->vm_mm, addr, page, vma->vm_page_prot); 1253a145dd41SLinus Torvalds } 1254e3c3374fSLinus Torvalds EXPORT_SYMBOL(vm_insert_page); 1255a145dd41SLinus Torvalds 1256a145dd41SLinus Torvalds /* 12571da177e4SLinus Torvalds * maps a range of physical memory into the requested pages. the old 12581da177e4SLinus Torvalds * mappings are removed. any references to nonexistent pages results 12591da177e4SLinus Torvalds * in null mappings (currently treated as "copy-on-access") 12601da177e4SLinus Torvalds */ 12611da177e4SLinus Torvalds static int remap_pte_range(struct mm_struct *mm, pmd_t *pmd, 12621da177e4SLinus Torvalds unsigned long addr, unsigned long end, 12631da177e4SLinus Torvalds unsigned long pfn, pgprot_t prot) 12641da177e4SLinus Torvalds { 12651da177e4SLinus Torvalds pte_t *pte; 1266c74df32cSHugh Dickins spinlock_t *ptl; 12671da177e4SLinus Torvalds 1268c74df32cSHugh Dickins pte = pte_alloc_map_lock(mm, pmd, addr, &ptl); 12691da177e4SLinus Torvalds if (!pte) 12701da177e4SLinus Torvalds return -ENOMEM; 12711da177e4SLinus Torvalds do { 12721da177e4SLinus Torvalds BUG_ON(!pte_none(*pte)); 12731da177e4SLinus Torvalds set_pte_at(mm, addr, pte, pfn_pte(pfn, prot)); 12741da177e4SLinus Torvalds pfn++; 12751da177e4SLinus Torvalds } while (pte++, addr += PAGE_SIZE, addr != end); 1276c74df32cSHugh Dickins pte_unmap_unlock(pte - 1, ptl); 12771da177e4SLinus Torvalds return 0; 12781da177e4SLinus Torvalds } 12791da177e4SLinus Torvalds 12801da177e4SLinus Torvalds static inline int remap_pmd_range(struct mm_struct *mm, pud_t *pud, 12811da177e4SLinus Torvalds unsigned long addr, unsigned long end, 12821da177e4SLinus Torvalds unsigned long pfn, pgprot_t prot) 12831da177e4SLinus Torvalds { 12841da177e4SLinus Torvalds pmd_t *pmd; 12851da177e4SLinus Torvalds unsigned long next; 12861da177e4SLinus Torvalds 12871da177e4SLinus Torvalds pfn -= addr >> PAGE_SHIFT; 12881da177e4SLinus Torvalds pmd = pmd_alloc(mm, pud, addr); 12891da177e4SLinus Torvalds if (!pmd) 12901da177e4SLinus Torvalds return -ENOMEM; 12911da177e4SLinus Torvalds do { 12921da177e4SLinus Torvalds next = pmd_addr_end(addr, end); 12931da177e4SLinus Torvalds if (remap_pte_range(mm, pmd, addr, next, 12941da177e4SLinus Torvalds pfn + (addr >> PAGE_SHIFT), prot)) 12951da177e4SLinus Torvalds return -ENOMEM; 12961da177e4SLinus Torvalds } while (pmd++, addr = next, addr != end); 12971da177e4SLinus Torvalds return 0; 12981da177e4SLinus Torvalds } 12991da177e4SLinus Torvalds 13001da177e4SLinus Torvalds static inline int remap_pud_range(struct mm_struct *mm, pgd_t *pgd, 13011da177e4SLinus Torvalds unsigned long addr, unsigned long end, 13021da177e4SLinus Torvalds unsigned long pfn, pgprot_t prot) 13031da177e4SLinus Torvalds { 13041da177e4SLinus Torvalds pud_t *pud; 13051da177e4SLinus Torvalds unsigned long next; 13061da177e4SLinus Torvalds 13071da177e4SLinus Torvalds pfn -= addr >> PAGE_SHIFT; 13081da177e4SLinus Torvalds pud = pud_alloc(mm, pgd, addr); 13091da177e4SLinus Torvalds if (!pud) 13101da177e4SLinus Torvalds return -ENOMEM; 13111da177e4SLinus Torvalds do { 13121da177e4SLinus Torvalds next = pud_addr_end(addr, end); 13131da177e4SLinus Torvalds if (remap_pmd_range(mm, pud, addr, next, 13141da177e4SLinus Torvalds pfn + (addr >> PAGE_SHIFT), prot)) 13151da177e4SLinus Torvalds return -ENOMEM; 13161da177e4SLinus Torvalds } while (pud++, addr = next, addr != end); 13171da177e4SLinus Torvalds return 0; 13181da177e4SLinus Torvalds } 13191da177e4SLinus Torvalds 13201da177e4SLinus Torvalds /* Note: this is only safe if the mm semaphore is held when called. */ 13211da177e4SLinus Torvalds int remap_pfn_range(struct vm_area_struct *vma, unsigned long addr, 13221da177e4SLinus Torvalds unsigned long pfn, unsigned long size, pgprot_t prot) 13231da177e4SLinus Torvalds { 13241da177e4SLinus Torvalds pgd_t *pgd; 13251da177e4SLinus Torvalds unsigned long next; 13262d15cab8SHugh Dickins unsigned long end = addr + PAGE_ALIGN(size); 13271da177e4SLinus Torvalds struct mm_struct *mm = vma->vm_mm; 13281da177e4SLinus Torvalds int err; 13291da177e4SLinus Torvalds 13301da177e4SLinus Torvalds /* 13311da177e4SLinus Torvalds * Physically remapped pages are special. Tell the 13321da177e4SLinus Torvalds * rest of the world about it: 13331da177e4SLinus Torvalds * VM_IO tells people not to look at these pages 13341da177e4SLinus Torvalds * (accesses can have side effects). 13350b14c179SHugh Dickins * VM_RESERVED is specified all over the place, because 13360b14c179SHugh Dickins * in 2.4 it kept swapout's vma scan off this vma; but 13370b14c179SHugh Dickins * in 2.6 the LRU scan won't even find its pages, so this 13380b14c179SHugh Dickins * flag means no more than count its pages in reserved_vm, 13390b14c179SHugh Dickins * and omit it from core dump, even when VM_IO turned off. 13406aab341eSLinus Torvalds * VM_PFNMAP tells the core MM that the base pages are just 13416aab341eSLinus Torvalds * raw PFN mappings, and do not have a "struct page" associated 13426aab341eSLinus Torvalds * with them. 1343fb155c16SLinus Torvalds * 1344fb155c16SLinus Torvalds * There's a horrible special case to handle copy-on-write 1345fb155c16SLinus Torvalds * behaviour that some programs depend on. We mark the "original" 1346fb155c16SLinus Torvalds * un-COW'ed pages by matching them up with "vma->vm_pgoff". 13471da177e4SLinus Torvalds */ 134867121172SLinus Torvalds if (is_cow_mapping(vma->vm_flags)) { 1349fb155c16SLinus Torvalds if (addr != vma->vm_start || end != vma->vm_end) 13507fc7e2eeSLinus Torvalds return -EINVAL; 13516aab341eSLinus Torvalds vma->vm_pgoff = pfn; 1352fb155c16SLinus Torvalds } 1353fb155c16SLinus Torvalds 1354fb155c16SLinus Torvalds vma->vm_flags |= VM_IO | VM_RESERVED | VM_PFNMAP; 13551da177e4SLinus Torvalds 13561da177e4SLinus Torvalds BUG_ON(addr >= end); 13571da177e4SLinus Torvalds pfn -= addr >> PAGE_SHIFT; 13581da177e4SLinus Torvalds pgd = pgd_offset(mm, addr); 13591da177e4SLinus Torvalds flush_cache_range(vma, addr, end); 13601da177e4SLinus Torvalds do { 13611da177e4SLinus Torvalds next = pgd_addr_end(addr, end); 13621da177e4SLinus Torvalds err = remap_pud_range(mm, pgd, addr, next, 13631da177e4SLinus Torvalds pfn + (addr >> PAGE_SHIFT), prot); 13641da177e4SLinus Torvalds if (err) 13651da177e4SLinus Torvalds break; 13661da177e4SLinus Torvalds } while (pgd++, addr = next, addr != end); 13671da177e4SLinus Torvalds return err; 13681da177e4SLinus Torvalds } 13691da177e4SLinus Torvalds EXPORT_SYMBOL(remap_pfn_range); 13701da177e4SLinus Torvalds 13711da177e4SLinus Torvalds /* 13728f4e2101SHugh Dickins * handle_pte_fault chooses page fault handler according to an entry 13738f4e2101SHugh Dickins * which was read non-atomically. Before making any commitment, on 13748f4e2101SHugh Dickins * those architectures or configurations (e.g. i386 with PAE) which 13758f4e2101SHugh Dickins * might give a mix of unmatched parts, do_swap_page and do_file_page 13768f4e2101SHugh Dickins * must check under lock before unmapping the pte and proceeding 13778f4e2101SHugh Dickins * (but do_wp_page is only called after already making such a check; 13788f4e2101SHugh Dickins * and do_anonymous_page and do_no_page can safely check later on). 13798f4e2101SHugh Dickins */ 13804c21e2f2SHugh Dickins static inline int pte_unmap_same(struct mm_struct *mm, pmd_t *pmd, 13818f4e2101SHugh Dickins pte_t *page_table, pte_t orig_pte) 13828f4e2101SHugh Dickins { 13838f4e2101SHugh Dickins int same = 1; 13848f4e2101SHugh Dickins #if defined(CONFIG_SMP) || defined(CONFIG_PREEMPT) 13858f4e2101SHugh Dickins if (sizeof(pte_t) > sizeof(unsigned long)) { 13864c21e2f2SHugh Dickins spinlock_t *ptl = pte_lockptr(mm, pmd); 13874c21e2f2SHugh Dickins spin_lock(ptl); 13888f4e2101SHugh Dickins same = pte_same(*page_table, orig_pte); 13894c21e2f2SHugh Dickins spin_unlock(ptl); 13908f4e2101SHugh Dickins } 13918f4e2101SHugh Dickins #endif 13928f4e2101SHugh Dickins pte_unmap(page_table); 13938f4e2101SHugh Dickins return same; 13948f4e2101SHugh Dickins } 13958f4e2101SHugh Dickins 13968f4e2101SHugh Dickins /* 13971da177e4SLinus Torvalds * Do pte_mkwrite, but only if the vma says VM_WRITE. We do this when 13981da177e4SLinus Torvalds * servicing faults for write access. In the normal case, do always want 13991da177e4SLinus Torvalds * pte_mkwrite. But get_user_pages can cause write faults for mappings 14001da177e4SLinus Torvalds * that do not have writing enabled, when used by access_process_vm. 14011da177e4SLinus Torvalds */ 14021da177e4SLinus Torvalds static inline pte_t maybe_mkwrite(pte_t pte, struct vm_area_struct *vma) 14031da177e4SLinus Torvalds { 14041da177e4SLinus Torvalds if (likely(vma->vm_flags & VM_WRITE)) 14051da177e4SLinus Torvalds pte = pte_mkwrite(pte); 14061da177e4SLinus Torvalds return pte; 14071da177e4SLinus Torvalds } 14081da177e4SLinus Torvalds 14096aab341eSLinus Torvalds static inline void cow_user_page(struct page *dst, struct page *src, unsigned long va) 14106aab341eSLinus Torvalds { 14116aab341eSLinus Torvalds /* 14126aab341eSLinus Torvalds * If the source page was a PFN mapping, we don't have 14136aab341eSLinus Torvalds * a "struct page" for it. We do a best-effort copy by 14146aab341eSLinus Torvalds * just copying from the original user address. If that 14156aab341eSLinus Torvalds * fails, we just zero-fill it. Live with it. 14166aab341eSLinus Torvalds */ 14176aab341eSLinus Torvalds if (unlikely(!src)) { 14186aab341eSLinus Torvalds void *kaddr = kmap_atomic(dst, KM_USER0); 14195d2a2dbbSLinus Torvalds void __user *uaddr = (void __user *)(va & PAGE_MASK); 14205d2a2dbbSLinus Torvalds 14215d2a2dbbSLinus Torvalds /* 14225d2a2dbbSLinus Torvalds * This really shouldn't fail, because the page is there 14235d2a2dbbSLinus Torvalds * in the page tables. But it might just be unreadable, 14245d2a2dbbSLinus Torvalds * in which case we just give up and fill the result with 14255d2a2dbbSLinus Torvalds * zeroes. 14265d2a2dbbSLinus Torvalds */ 14275d2a2dbbSLinus Torvalds if (__copy_from_user_inatomic(kaddr, uaddr, PAGE_SIZE)) 14286aab341eSLinus Torvalds memset(kaddr, 0, PAGE_SIZE); 14296aab341eSLinus Torvalds kunmap_atomic(kaddr, KM_USER0); 14306aab341eSLinus Torvalds return; 14316aab341eSLinus Torvalds 14326aab341eSLinus Torvalds } 14336aab341eSLinus Torvalds copy_user_highpage(dst, src, va); 14346aab341eSLinus Torvalds } 14356aab341eSLinus Torvalds 14361da177e4SLinus Torvalds /* 14371da177e4SLinus Torvalds * This routine handles present pages, when users try to write 14381da177e4SLinus Torvalds * to a shared page. It is done by copying the page to a new address 14391da177e4SLinus Torvalds * and decrementing the shared-page counter for the old page. 14401da177e4SLinus Torvalds * 14411da177e4SLinus Torvalds * Note that this routine assumes that the protection checks have been 14421da177e4SLinus Torvalds * done by the caller (the low-level page fault routine in most cases). 14431da177e4SLinus Torvalds * Thus we can safely just mark it writable once we've done any necessary 14441da177e4SLinus Torvalds * COW. 14451da177e4SLinus Torvalds * 14461da177e4SLinus Torvalds * We also mark the page dirty at this point even though the page will 14471da177e4SLinus Torvalds * change only once the write actually happens. This avoids a few races, 14481da177e4SLinus Torvalds * and potentially makes it more efficient. 14491da177e4SLinus Torvalds * 14508f4e2101SHugh Dickins * We enter with non-exclusive mmap_sem (to exclude vma changes, 14518f4e2101SHugh Dickins * but allow concurrent faults), with pte both mapped and locked. 14528f4e2101SHugh Dickins * We return with mmap_sem still held, but pte unmapped and unlocked. 14531da177e4SLinus Torvalds */ 14541da177e4SLinus Torvalds static int do_wp_page(struct mm_struct *mm, struct vm_area_struct *vma, 145565500d23SHugh Dickins unsigned long address, pte_t *page_table, pmd_t *pmd, 14568f4e2101SHugh Dickins spinlock_t *ptl, pte_t orig_pte) 14571da177e4SLinus Torvalds { 1458e5bbe4dfSHugh Dickins struct page *old_page, *new_page; 14591da177e4SLinus Torvalds pte_t entry; 14609637a5efSDavid Howells int reuse, ret = VM_FAULT_MINOR; 14611da177e4SLinus Torvalds 14626aab341eSLinus Torvalds old_page = vm_normal_page(vma, address, orig_pte); 14636aab341eSLinus Torvalds if (!old_page) 1464920fc356SHugh Dickins goto gotten; 14651da177e4SLinus Torvalds 14669637a5efSDavid Howells if (unlikely((vma->vm_flags & (VM_SHARED|VM_WRITE)) == 14679637a5efSDavid Howells (VM_SHARED|VM_WRITE))) { 14689637a5efSDavid Howells if (vma->vm_ops && vma->vm_ops->page_mkwrite) { 14699637a5efSDavid Howells /* 14709637a5efSDavid Howells * Notify the address space that the page is about to 14719637a5efSDavid Howells * become writable so that it can prohibit this or wait 14729637a5efSDavid Howells * for the page to get into an appropriate state. 14739637a5efSDavid Howells * 14749637a5efSDavid Howells * We do this without the lock held, so that it can 14759637a5efSDavid Howells * sleep if it needs to. 14769637a5efSDavid Howells */ 14779637a5efSDavid Howells page_cache_get(old_page); 14789637a5efSDavid Howells pte_unmap_unlock(page_table, ptl); 14799637a5efSDavid Howells 14809637a5efSDavid Howells if (vma->vm_ops->page_mkwrite(vma, old_page) < 0) 14819637a5efSDavid Howells goto unwritable_page; 14829637a5efSDavid Howells 14839637a5efSDavid Howells page_cache_release(old_page); 14849637a5efSDavid Howells 14859637a5efSDavid Howells /* 14869637a5efSDavid Howells * Since we dropped the lock we need to revalidate 14879637a5efSDavid Howells * the PTE as someone else may have changed it. If 14889637a5efSDavid Howells * they did, we just return, as we can count on the 14899637a5efSDavid Howells * MMU to tell us if they didn't also make it writable. 14909637a5efSDavid Howells */ 14919637a5efSDavid Howells page_table = pte_offset_map_lock(mm, pmd, address, 14929637a5efSDavid Howells &ptl); 14939637a5efSDavid Howells if (!pte_same(*page_table, orig_pte)) 14949637a5efSDavid Howells goto unlock; 14959637a5efSDavid Howells } 14969637a5efSDavid Howells 14979637a5efSDavid Howells reuse = 1; 14989637a5efSDavid Howells } else if (PageAnon(old_page) && !TestSetPageLocked(old_page)) { 14999637a5efSDavid Howells reuse = can_share_swap_page(old_page); 15001da177e4SLinus Torvalds unlock_page(old_page); 15019637a5efSDavid Howells } else { 15029637a5efSDavid Howells reuse = 0; 15039637a5efSDavid Howells } 15049637a5efSDavid Howells 15051da177e4SLinus Torvalds if (reuse) { 1506eca35133SBen Collins flush_cache_page(vma, address, pte_pfn(orig_pte)); 150765500d23SHugh Dickins entry = pte_mkyoung(orig_pte); 150865500d23SHugh Dickins entry = maybe_mkwrite(pte_mkdirty(entry), vma); 15091da177e4SLinus Torvalds ptep_set_access_flags(vma, address, page_table, entry, 1); 15101da177e4SLinus Torvalds update_mmu_cache(vma, address, entry); 15111da177e4SLinus Torvalds lazy_mmu_prot_update(entry); 151265500d23SHugh Dickins ret |= VM_FAULT_WRITE; 151365500d23SHugh Dickins goto unlock; 15141da177e4SLinus Torvalds } 15151da177e4SLinus Torvalds 15161da177e4SLinus Torvalds /* 15171da177e4SLinus Torvalds * Ok, we need to copy. Oh, well.. 15181da177e4SLinus Torvalds */ 15191da177e4SLinus Torvalds page_cache_get(old_page); 1520920fc356SHugh Dickins gotten: 15218f4e2101SHugh Dickins pte_unmap_unlock(page_table, ptl); 15221da177e4SLinus Torvalds 15231da177e4SLinus Torvalds if (unlikely(anon_vma_prepare(vma))) 152465500d23SHugh Dickins goto oom; 1525e5bbe4dfSHugh Dickins if (old_page == ZERO_PAGE(address)) { 15261da177e4SLinus Torvalds new_page = alloc_zeroed_user_highpage(vma, address); 15271da177e4SLinus Torvalds if (!new_page) 152865500d23SHugh Dickins goto oom; 15291da177e4SLinus Torvalds } else { 15301da177e4SLinus Torvalds new_page = alloc_page_vma(GFP_HIGHUSER, vma, address); 15311da177e4SLinus Torvalds if (!new_page) 153265500d23SHugh Dickins goto oom; 1533e5bbe4dfSHugh Dickins cow_user_page(new_page, old_page, address); 15341da177e4SLinus Torvalds } 153565500d23SHugh Dickins 15361da177e4SLinus Torvalds /* 15371da177e4SLinus Torvalds * Re-check the pte - we dropped the lock 15381da177e4SLinus Torvalds */ 15398f4e2101SHugh Dickins page_table = pte_offset_map_lock(mm, pmd, address, &ptl); 154065500d23SHugh Dickins if (likely(pte_same(*page_table, orig_pte))) { 1541920fc356SHugh Dickins if (old_page) { 15421da177e4SLinus Torvalds page_remove_rmap(old_page); 15434294621fSHugh Dickins if (!PageAnon(old_page)) { 15444294621fSHugh Dickins dec_mm_counter(mm, file_rss); 1545920fc356SHugh Dickins inc_mm_counter(mm, anon_rss); 15464294621fSHugh Dickins } 1547920fc356SHugh Dickins } else 1548920fc356SHugh Dickins inc_mm_counter(mm, anon_rss); 1549eca35133SBen Collins flush_cache_page(vma, address, pte_pfn(orig_pte)); 155065500d23SHugh Dickins entry = mk_pte(new_page, vma->vm_page_prot); 155165500d23SHugh Dickins entry = maybe_mkwrite(pte_mkdirty(entry), vma); 155265500d23SHugh Dickins ptep_establish(vma, address, page_table, entry); 155365500d23SHugh Dickins update_mmu_cache(vma, address, entry); 155465500d23SHugh Dickins lazy_mmu_prot_update(entry); 15551da177e4SLinus Torvalds lru_cache_add_active(new_page); 15569617d95eSNick Piggin page_add_new_anon_rmap(new_page, vma, address); 15571da177e4SLinus Torvalds 15581da177e4SLinus Torvalds /* Free the old page.. */ 15591da177e4SLinus Torvalds new_page = old_page; 1560f33ea7f4SNick Piggin ret |= VM_FAULT_WRITE; 15611da177e4SLinus Torvalds } 1562920fc356SHugh Dickins if (new_page) 15631da177e4SLinus Torvalds page_cache_release(new_page); 1564920fc356SHugh Dickins if (old_page) 15651da177e4SLinus Torvalds page_cache_release(old_page); 156665500d23SHugh Dickins unlock: 15678f4e2101SHugh Dickins pte_unmap_unlock(page_table, ptl); 1568f33ea7f4SNick Piggin return ret; 156965500d23SHugh Dickins oom: 1570920fc356SHugh Dickins if (old_page) 15711da177e4SLinus Torvalds page_cache_release(old_page); 15721da177e4SLinus Torvalds return VM_FAULT_OOM; 15739637a5efSDavid Howells 15749637a5efSDavid Howells unwritable_page: 15759637a5efSDavid Howells page_cache_release(old_page); 15769637a5efSDavid Howells return VM_FAULT_SIGBUS; 15771da177e4SLinus Torvalds } 15781da177e4SLinus Torvalds 15791da177e4SLinus Torvalds /* 15801da177e4SLinus Torvalds * Helper functions for unmap_mapping_range(). 15811da177e4SLinus Torvalds * 15821da177e4SLinus Torvalds * __ Notes on dropping i_mmap_lock to reduce latency while unmapping __ 15831da177e4SLinus Torvalds * 15841da177e4SLinus Torvalds * We have to restart searching the prio_tree whenever we drop the lock, 15851da177e4SLinus Torvalds * since the iterator is only valid while the lock is held, and anyway 15861da177e4SLinus Torvalds * a later vma might be split and reinserted earlier while lock dropped. 15871da177e4SLinus Torvalds * 15881da177e4SLinus Torvalds * The list of nonlinear vmas could be handled more efficiently, using 15891da177e4SLinus Torvalds * a placeholder, but handle it in the same way until a need is shown. 15901da177e4SLinus Torvalds * It is important to search the prio_tree before nonlinear list: a vma 15911da177e4SLinus Torvalds * may become nonlinear and be shifted from prio_tree to nonlinear list 15921da177e4SLinus Torvalds * while the lock is dropped; but never shifted from list to prio_tree. 15931da177e4SLinus Torvalds * 15941da177e4SLinus Torvalds * In order to make forward progress despite restarting the search, 15951da177e4SLinus Torvalds * vm_truncate_count is used to mark a vma as now dealt with, so we can 15961da177e4SLinus Torvalds * quickly skip it next time around. Since the prio_tree search only 15971da177e4SLinus Torvalds * shows us those vmas affected by unmapping the range in question, we 15981da177e4SLinus Torvalds * can't efficiently keep all vmas in step with mapping->truncate_count: 15991da177e4SLinus Torvalds * so instead reset them all whenever it wraps back to 0 (then go to 1). 16001da177e4SLinus Torvalds * mapping->truncate_count and vma->vm_truncate_count are protected by 16011da177e4SLinus Torvalds * i_mmap_lock. 16021da177e4SLinus Torvalds * 16031da177e4SLinus Torvalds * In order to make forward progress despite repeatedly restarting some 1604ee39b37bSHugh Dickins * large vma, note the restart_addr from unmap_vmas when it breaks out: 16051da177e4SLinus Torvalds * and restart from that address when we reach that vma again. It might 16061da177e4SLinus Torvalds * have been split or merged, shrunk or extended, but never shifted: so 16071da177e4SLinus Torvalds * restart_addr remains valid so long as it remains in the vma's range. 16081da177e4SLinus Torvalds * unmap_mapping_range forces truncate_count to leap over page-aligned 16091da177e4SLinus Torvalds * values so we can save vma's restart_addr in its truncate_count field. 16101da177e4SLinus Torvalds */ 16111da177e4SLinus Torvalds #define is_restart_addr(truncate_count) (!((truncate_count) & ~PAGE_MASK)) 16121da177e4SLinus Torvalds 16131da177e4SLinus Torvalds static void reset_vma_truncate_counts(struct address_space *mapping) 16141da177e4SLinus Torvalds { 16151da177e4SLinus Torvalds struct vm_area_struct *vma; 16161da177e4SLinus Torvalds struct prio_tree_iter iter; 16171da177e4SLinus Torvalds 16181da177e4SLinus Torvalds vma_prio_tree_foreach(vma, &iter, &mapping->i_mmap, 0, ULONG_MAX) 16191da177e4SLinus Torvalds vma->vm_truncate_count = 0; 16201da177e4SLinus Torvalds list_for_each_entry(vma, &mapping->i_mmap_nonlinear, shared.vm_set.list) 16211da177e4SLinus Torvalds vma->vm_truncate_count = 0; 16221da177e4SLinus Torvalds } 16231da177e4SLinus Torvalds 16241da177e4SLinus Torvalds static int unmap_mapping_range_vma(struct vm_area_struct *vma, 16251da177e4SLinus Torvalds unsigned long start_addr, unsigned long end_addr, 16261da177e4SLinus Torvalds struct zap_details *details) 16271da177e4SLinus Torvalds { 16281da177e4SLinus Torvalds unsigned long restart_addr; 16291da177e4SLinus Torvalds int need_break; 16301da177e4SLinus Torvalds 16311da177e4SLinus Torvalds again: 16321da177e4SLinus Torvalds restart_addr = vma->vm_truncate_count; 16331da177e4SLinus Torvalds if (is_restart_addr(restart_addr) && start_addr < restart_addr) { 16341da177e4SLinus Torvalds start_addr = restart_addr; 16351da177e4SLinus Torvalds if (start_addr >= end_addr) { 16361da177e4SLinus Torvalds /* Top of vma has been split off since last time */ 16371da177e4SLinus Torvalds vma->vm_truncate_count = details->truncate_count; 16381da177e4SLinus Torvalds return 0; 16391da177e4SLinus Torvalds } 16401da177e4SLinus Torvalds } 16411da177e4SLinus Torvalds 1642ee39b37bSHugh Dickins restart_addr = zap_page_range(vma, start_addr, 1643ee39b37bSHugh Dickins end_addr - start_addr, details); 16441da177e4SLinus Torvalds need_break = need_resched() || 16451da177e4SLinus Torvalds need_lockbreak(details->i_mmap_lock); 16461da177e4SLinus Torvalds 1647ee39b37bSHugh Dickins if (restart_addr >= end_addr) { 16481da177e4SLinus Torvalds /* We have now completed this vma: mark it so */ 16491da177e4SLinus Torvalds vma->vm_truncate_count = details->truncate_count; 16501da177e4SLinus Torvalds if (!need_break) 16511da177e4SLinus Torvalds return 0; 16521da177e4SLinus Torvalds } else { 16531da177e4SLinus Torvalds /* Note restart_addr in vma's truncate_count field */ 1654ee39b37bSHugh Dickins vma->vm_truncate_count = restart_addr; 16551da177e4SLinus Torvalds if (!need_break) 16561da177e4SLinus Torvalds goto again; 16571da177e4SLinus Torvalds } 16581da177e4SLinus Torvalds 16591da177e4SLinus Torvalds spin_unlock(details->i_mmap_lock); 16601da177e4SLinus Torvalds cond_resched(); 16611da177e4SLinus Torvalds spin_lock(details->i_mmap_lock); 16621da177e4SLinus Torvalds return -EINTR; 16631da177e4SLinus Torvalds } 16641da177e4SLinus Torvalds 16651da177e4SLinus Torvalds static inline void unmap_mapping_range_tree(struct prio_tree_root *root, 16661da177e4SLinus Torvalds struct zap_details *details) 16671da177e4SLinus Torvalds { 16681da177e4SLinus Torvalds struct vm_area_struct *vma; 16691da177e4SLinus Torvalds struct prio_tree_iter iter; 16701da177e4SLinus Torvalds pgoff_t vba, vea, zba, zea; 16711da177e4SLinus Torvalds 16721da177e4SLinus Torvalds restart: 16731da177e4SLinus Torvalds vma_prio_tree_foreach(vma, &iter, root, 16741da177e4SLinus Torvalds details->first_index, details->last_index) { 16751da177e4SLinus Torvalds /* Skip quickly over those we have already dealt with */ 16761da177e4SLinus Torvalds if (vma->vm_truncate_count == details->truncate_count) 16771da177e4SLinus Torvalds continue; 16781da177e4SLinus Torvalds 16791da177e4SLinus Torvalds vba = vma->vm_pgoff; 16801da177e4SLinus Torvalds vea = vba + ((vma->vm_end - vma->vm_start) >> PAGE_SHIFT) - 1; 16811da177e4SLinus Torvalds /* Assume for now that PAGE_CACHE_SHIFT == PAGE_SHIFT */ 16821da177e4SLinus Torvalds zba = details->first_index; 16831da177e4SLinus Torvalds if (zba < vba) 16841da177e4SLinus Torvalds zba = vba; 16851da177e4SLinus Torvalds zea = details->last_index; 16861da177e4SLinus Torvalds if (zea > vea) 16871da177e4SLinus Torvalds zea = vea; 16881da177e4SLinus Torvalds 16891da177e4SLinus Torvalds if (unmap_mapping_range_vma(vma, 16901da177e4SLinus Torvalds ((zba - vba) << PAGE_SHIFT) + vma->vm_start, 16911da177e4SLinus Torvalds ((zea - vba + 1) << PAGE_SHIFT) + vma->vm_start, 16921da177e4SLinus Torvalds details) < 0) 16931da177e4SLinus Torvalds goto restart; 16941da177e4SLinus Torvalds } 16951da177e4SLinus Torvalds } 16961da177e4SLinus Torvalds 16971da177e4SLinus Torvalds static inline void unmap_mapping_range_list(struct list_head *head, 16981da177e4SLinus Torvalds struct zap_details *details) 16991da177e4SLinus Torvalds { 17001da177e4SLinus Torvalds struct vm_area_struct *vma; 17011da177e4SLinus Torvalds 17021da177e4SLinus Torvalds /* 17031da177e4SLinus Torvalds * In nonlinear VMAs there is no correspondence between virtual address 17041da177e4SLinus Torvalds * offset and file offset. So we must perform an exhaustive search 17051da177e4SLinus Torvalds * across *all* the pages in each nonlinear VMA, not just the pages 17061da177e4SLinus Torvalds * whose virtual address lies outside the file truncation point. 17071da177e4SLinus Torvalds */ 17081da177e4SLinus Torvalds restart: 17091da177e4SLinus Torvalds list_for_each_entry(vma, head, shared.vm_set.list) { 17101da177e4SLinus Torvalds /* Skip quickly over those we have already dealt with */ 17111da177e4SLinus Torvalds if (vma->vm_truncate_count == details->truncate_count) 17121da177e4SLinus Torvalds continue; 17131da177e4SLinus Torvalds details->nonlinear_vma = vma; 17141da177e4SLinus Torvalds if (unmap_mapping_range_vma(vma, vma->vm_start, 17151da177e4SLinus Torvalds vma->vm_end, details) < 0) 17161da177e4SLinus Torvalds goto restart; 17171da177e4SLinus Torvalds } 17181da177e4SLinus Torvalds } 17191da177e4SLinus Torvalds 17201da177e4SLinus Torvalds /** 17211da177e4SLinus Torvalds * unmap_mapping_range - unmap the portion of all mmaps 17221da177e4SLinus Torvalds * in the specified address_space corresponding to the specified 17231da177e4SLinus Torvalds * page range in the underlying file. 17243d41088fSMartin Waitz * @mapping: the address space containing mmaps to be unmapped. 17251da177e4SLinus Torvalds * @holebegin: byte in first page to unmap, relative to the start of 17261da177e4SLinus Torvalds * the underlying file. This will be rounded down to a PAGE_SIZE 17271da177e4SLinus Torvalds * boundary. Note that this is different from vmtruncate(), which 17281da177e4SLinus Torvalds * must keep the partial page. In contrast, we must get rid of 17291da177e4SLinus Torvalds * partial pages. 17301da177e4SLinus Torvalds * @holelen: size of prospective hole in bytes. This will be rounded 17311da177e4SLinus Torvalds * up to a PAGE_SIZE boundary. A holelen of zero truncates to the 17321da177e4SLinus Torvalds * end of the file. 17331da177e4SLinus Torvalds * @even_cows: 1 when truncating a file, unmap even private COWed pages; 17341da177e4SLinus Torvalds * but 0 when invalidating pagecache, don't throw away private data. 17351da177e4SLinus Torvalds */ 17361da177e4SLinus Torvalds void unmap_mapping_range(struct address_space *mapping, 17371da177e4SLinus Torvalds loff_t const holebegin, loff_t const holelen, int even_cows) 17381da177e4SLinus Torvalds { 17391da177e4SLinus Torvalds struct zap_details details; 17401da177e4SLinus Torvalds pgoff_t hba = holebegin >> PAGE_SHIFT; 17411da177e4SLinus Torvalds pgoff_t hlen = (holelen + PAGE_SIZE - 1) >> PAGE_SHIFT; 17421da177e4SLinus Torvalds 17431da177e4SLinus Torvalds /* Check for overflow. */ 17441da177e4SLinus Torvalds if (sizeof(holelen) > sizeof(hlen)) { 17451da177e4SLinus Torvalds long long holeend = 17461da177e4SLinus Torvalds (holebegin + holelen + PAGE_SIZE - 1) >> PAGE_SHIFT; 17471da177e4SLinus Torvalds if (holeend & ~(long long)ULONG_MAX) 17481da177e4SLinus Torvalds hlen = ULONG_MAX - hba + 1; 17491da177e4SLinus Torvalds } 17501da177e4SLinus Torvalds 17511da177e4SLinus Torvalds details.check_mapping = even_cows? NULL: mapping; 17521da177e4SLinus Torvalds details.nonlinear_vma = NULL; 17531da177e4SLinus Torvalds details.first_index = hba; 17541da177e4SLinus Torvalds details.last_index = hba + hlen - 1; 17551da177e4SLinus Torvalds if (details.last_index < details.first_index) 17561da177e4SLinus Torvalds details.last_index = ULONG_MAX; 17571da177e4SLinus Torvalds details.i_mmap_lock = &mapping->i_mmap_lock; 17581da177e4SLinus Torvalds 17591da177e4SLinus Torvalds spin_lock(&mapping->i_mmap_lock); 17601da177e4SLinus Torvalds 17611da177e4SLinus Torvalds /* serialize i_size write against truncate_count write */ 17621da177e4SLinus Torvalds smp_wmb(); 17631da177e4SLinus Torvalds /* Protect against page faults, and endless unmapping loops */ 17641da177e4SLinus Torvalds mapping->truncate_count++; 17651da177e4SLinus Torvalds /* 17661da177e4SLinus Torvalds * For archs where spin_lock has inclusive semantics like ia64 17671da177e4SLinus Torvalds * this smp_mb() will prevent to read pagetable contents 17681da177e4SLinus Torvalds * before the truncate_count increment is visible to 17691da177e4SLinus Torvalds * other cpus. 17701da177e4SLinus Torvalds */ 17711da177e4SLinus Torvalds smp_mb(); 17721da177e4SLinus Torvalds if (unlikely(is_restart_addr(mapping->truncate_count))) { 17731da177e4SLinus Torvalds if (mapping->truncate_count == 0) 17741da177e4SLinus Torvalds reset_vma_truncate_counts(mapping); 17751da177e4SLinus Torvalds mapping->truncate_count++; 17761da177e4SLinus Torvalds } 17771da177e4SLinus Torvalds details.truncate_count = mapping->truncate_count; 17781da177e4SLinus Torvalds 17791da177e4SLinus Torvalds if (unlikely(!prio_tree_empty(&mapping->i_mmap))) 17801da177e4SLinus Torvalds unmap_mapping_range_tree(&mapping->i_mmap, &details); 17811da177e4SLinus Torvalds if (unlikely(!list_empty(&mapping->i_mmap_nonlinear))) 17821da177e4SLinus Torvalds unmap_mapping_range_list(&mapping->i_mmap_nonlinear, &details); 17831da177e4SLinus Torvalds spin_unlock(&mapping->i_mmap_lock); 17841da177e4SLinus Torvalds } 17851da177e4SLinus Torvalds EXPORT_SYMBOL(unmap_mapping_range); 17861da177e4SLinus Torvalds 17871da177e4SLinus Torvalds /* 17881da177e4SLinus Torvalds * Handle all mappings that got truncated by a "truncate()" 17891da177e4SLinus Torvalds * system call. 17901da177e4SLinus Torvalds * 17911da177e4SLinus Torvalds * NOTE! We have to be ready to update the memory sharing 17921da177e4SLinus Torvalds * between the file and the memory map for a potential last 17931da177e4SLinus Torvalds * incomplete page. Ugly, but necessary. 17941da177e4SLinus Torvalds */ 17951da177e4SLinus Torvalds int vmtruncate(struct inode * inode, loff_t offset) 17961da177e4SLinus Torvalds { 17971da177e4SLinus Torvalds struct address_space *mapping = inode->i_mapping; 17981da177e4SLinus Torvalds unsigned long limit; 17991da177e4SLinus Torvalds 18001da177e4SLinus Torvalds if (inode->i_size < offset) 18011da177e4SLinus Torvalds goto do_expand; 18021da177e4SLinus Torvalds /* 18031da177e4SLinus Torvalds * truncation of in-use swapfiles is disallowed - it would cause 18041da177e4SLinus Torvalds * subsequent swapout to scribble on the now-freed blocks. 18051da177e4SLinus Torvalds */ 18061da177e4SLinus Torvalds if (IS_SWAPFILE(inode)) 18071da177e4SLinus Torvalds goto out_busy; 18081da177e4SLinus Torvalds i_size_write(inode, offset); 18091da177e4SLinus Torvalds unmap_mapping_range(mapping, offset + PAGE_SIZE - 1, 0, 1); 18101da177e4SLinus Torvalds truncate_inode_pages(mapping, offset); 18111da177e4SLinus Torvalds goto out_truncate; 18121da177e4SLinus Torvalds 18131da177e4SLinus Torvalds do_expand: 18141da177e4SLinus Torvalds limit = current->signal->rlim[RLIMIT_FSIZE].rlim_cur; 18151da177e4SLinus Torvalds if (limit != RLIM_INFINITY && offset > limit) 18161da177e4SLinus Torvalds goto out_sig; 18171da177e4SLinus Torvalds if (offset > inode->i_sb->s_maxbytes) 18181da177e4SLinus Torvalds goto out_big; 18191da177e4SLinus Torvalds i_size_write(inode, offset); 18201da177e4SLinus Torvalds 18211da177e4SLinus Torvalds out_truncate: 18221da177e4SLinus Torvalds if (inode->i_op && inode->i_op->truncate) 18231da177e4SLinus Torvalds inode->i_op->truncate(inode); 18241da177e4SLinus Torvalds return 0; 18251da177e4SLinus Torvalds out_sig: 18261da177e4SLinus Torvalds send_sig(SIGXFSZ, current, 0); 18271da177e4SLinus Torvalds out_big: 18281da177e4SLinus Torvalds return -EFBIG; 18291da177e4SLinus Torvalds out_busy: 18301da177e4SLinus Torvalds return -ETXTBSY; 18311da177e4SLinus Torvalds } 18321da177e4SLinus Torvalds EXPORT_SYMBOL(vmtruncate); 18331da177e4SLinus Torvalds 1834f6b3ec23SBadari Pulavarty int vmtruncate_range(struct inode *inode, loff_t offset, loff_t end) 1835f6b3ec23SBadari Pulavarty { 1836f6b3ec23SBadari Pulavarty struct address_space *mapping = inode->i_mapping; 1837f6b3ec23SBadari Pulavarty 1838f6b3ec23SBadari Pulavarty /* 1839f6b3ec23SBadari Pulavarty * If the underlying filesystem is not going to provide 1840f6b3ec23SBadari Pulavarty * a way to truncate a range of blocks (punch a hole) - 1841f6b3ec23SBadari Pulavarty * we should return failure right now. 1842f6b3ec23SBadari Pulavarty */ 1843f6b3ec23SBadari Pulavarty if (!inode->i_op || !inode->i_op->truncate_range) 1844f6b3ec23SBadari Pulavarty return -ENOSYS; 1845f6b3ec23SBadari Pulavarty 18461b1dcc1bSJes Sorensen mutex_lock(&inode->i_mutex); 1847f6b3ec23SBadari Pulavarty down_write(&inode->i_alloc_sem); 1848f6b3ec23SBadari Pulavarty unmap_mapping_range(mapping, offset, (end - offset), 1); 1849f6b3ec23SBadari Pulavarty truncate_inode_pages_range(mapping, offset, end); 1850f6b3ec23SBadari Pulavarty inode->i_op->truncate_range(inode, offset, end); 1851f6b3ec23SBadari Pulavarty up_write(&inode->i_alloc_sem); 18521b1dcc1bSJes Sorensen mutex_unlock(&inode->i_mutex); 1853f6b3ec23SBadari Pulavarty 1854f6b3ec23SBadari Pulavarty return 0; 1855f6b3ec23SBadari Pulavarty } 1856f6b3ec23SBadari Pulavarty EXPORT_SYMBOL(vmtruncate_range); 1857f6b3ec23SBadari Pulavarty 18581da177e4SLinus Torvalds /* 18591da177e4SLinus Torvalds * Primitive swap readahead code. We simply read an aligned block of 18601da177e4SLinus Torvalds * (1 << page_cluster) entries in the swap area. This method is chosen 18611da177e4SLinus Torvalds * because it doesn't cost us any seek time. We also make sure to queue 18621da177e4SLinus Torvalds * the 'original' request together with the readahead ones... 18631da177e4SLinus Torvalds * 18641da177e4SLinus Torvalds * This has been extended to use the NUMA policies from the mm triggering 18651da177e4SLinus Torvalds * the readahead. 18661da177e4SLinus Torvalds * 18671da177e4SLinus Torvalds * Caller must hold down_read on the vma->vm_mm if vma is not NULL. 18681da177e4SLinus Torvalds */ 18691da177e4SLinus Torvalds void swapin_readahead(swp_entry_t entry, unsigned long addr,struct vm_area_struct *vma) 18701da177e4SLinus Torvalds { 18711da177e4SLinus Torvalds #ifdef CONFIG_NUMA 18721da177e4SLinus Torvalds struct vm_area_struct *next_vma = vma ? vma->vm_next : NULL; 18731da177e4SLinus Torvalds #endif 18741da177e4SLinus Torvalds int i, num; 18751da177e4SLinus Torvalds struct page *new_page; 18761da177e4SLinus Torvalds unsigned long offset; 18771da177e4SLinus Torvalds 18781da177e4SLinus Torvalds /* 18791da177e4SLinus Torvalds * Get the number of handles we should do readahead io to. 18801da177e4SLinus Torvalds */ 18811da177e4SLinus Torvalds num = valid_swaphandles(entry, &offset); 18821da177e4SLinus Torvalds for (i = 0; i < num; offset++, i++) { 18831da177e4SLinus Torvalds /* Ok, do the async read-ahead now */ 18841da177e4SLinus Torvalds new_page = read_swap_cache_async(swp_entry(swp_type(entry), 18851da177e4SLinus Torvalds offset), vma, addr); 18861da177e4SLinus Torvalds if (!new_page) 18871da177e4SLinus Torvalds break; 18881da177e4SLinus Torvalds page_cache_release(new_page); 18891da177e4SLinus Torvalds #ifdef CONFIG_NUMA 18901da177e4SLinus Torvalds /* 18911da177e4SLinus Torvalds * Find the next applicable VMA for the NUMA policy. 18921da177e4SLinus Torvalds */ 18931da177e4SLinus Torvalds addr += PAGE_SIZE; 18941da177e4SLinus Torvalds if (addr == 0) 18951da177e4SLinus Torvalds vma = NULL; 18961da177e4SLinus Torvalds if (vma) { 18971da177e4SLinus Torvalds if (addr >= vma->vm_end) { 18981da177e4SLinus Torvalds vma = next_vma; 18991da177e4SLinus Torvalds next_vma = vma ? vma->vm_next : NULL; 19001da177e4SLinus Torvalds } 19011da177e4SLinus Torvalds if (vma && addr < vma->vm_start) 19021da177e4SLinus Torvalds vma = NULL; 19031da177e4SLinus Torvalds } else { 19041da177e4SLinus Torvalds if (next_vma && addr >= next_vma->vm_start) { 19051da177e4SLinus Torvalds vma = next_vma; 19061da177e4SLinus Torvalds next_vma = vma->vm_next; 19071da177e4SLinus Torvalds } 19081da177e4SLinus Torvalds } 19091da177e4SLinus Torvalds #endif 19101da177e4SLinus Torvalds } 19111da177e4SLinus Torvalds lru_add_drain(); /* Push any new pages onto the LRU now */ 19121da177e4SLinus Torvalds } 19131da177e4SLinus Torvalds 19141da177e4SLinus Torvalds /* 19158f4e2101SHugh Dickins * We enter with non-exclusive mmap_sem (to exclude vma changes, 19168f4e2101SHugh Dickins * but allow concurrent faults), and pte mapped but not yet locked. 19178f4e2101SHugh Dickins * We return with mmap_sem still held, but pte unmapped and unlocked. 19181da177e4SLinus Torvalds */ 191965500d23SHugh Dickins static int do_swap_page(struct mm_struct *mm, struct vm_area_struct *vma, 192065500d23SHugh Dickins unsigned long address, pte_t *page_table, pmd_t *pmd, 192165500d23SHugh Dickins int write_access, pte_t orig_pte) 19221da177e4SLinus Torvalds { 19238f4e2101SHugh Dickins spinlock_t *ptl; 19241da177e4SLinus Torvalds struct page *page; 192565500d23SHugh Dickins swp_entry_t entry; 19261da177e4SLinus Torvalds pte_t pte; 19271da177e4SLinus Torvalds int ret = VM_FAULT_MINOR; 19281da177e4SLinus Torvalds 19294c21e2f2SHugh Dickins if (!pte_unmap_same(mm, pmd, page_table, orig_pte)) 19308f4e2101SHugh Dickins goto out; 193165500d23SHugh Dickins 193265500d23SHugh Dickins entry = pte_to_swp_entry(orig_pte); 19330697212aSChristoph Lameter if (is_migration_entry(entry)) { 19340697212aSChristoph Lameter migration_entry_wait(mm, pmd, address); 19350697212aSChristoph Lameter goto out; 19360697212aSChristoph Lameter } 19371da177e4SLinus Torvalds page = lookup_swap_cache(entry); 19381da177e4SLinus Torvalds if (!page) { 19391da177e4SLinus Torvalds swapin_readahead(entry, address, vma); 19401da177e4SLinus Torvalds page = read_swap_cache_async(entry, vma, address); 19411da177e4SLinus Torvalds if (!page) { 19421da177e4SLinus Torvalds /* 19438f4e2101SHugh Dickins * Back out if somebody else faulted in this pte 19448f4e2101SHugh Dickins * while we released the pte lock. 19451da177e4SLinus Torvalds */ 19468f4e2101SHugh Dickins page_table = pte_offset_map_lock(mm, pmd, address, &ptl); 19471da177e4SLinus Torvalds if (likely(pte_same(*page_table, orig_pte))) 19481da177e4SLinus Torvalds ret = VM_FAULT_OOM; 194965500d23SHugh Dickins goto unlock; 19501da177e4SLinus Torvalds } 19511da177e4SLinus Torvalds 19521da177e4SLinus Torvalds /* Had to read the page from swap area: Major fault */ 19531da177e4SLinus Torvalds ret = VM_FAULT_MAJOR; 1954*f8891e5eSChristoph Lameter count_vm_event(PGMAJFAULT); 19551da177e4SLinus Torvalds grab_swap_token(); 19561da177e4SLinus Torvalds } 19571da177e4SLinus Torvalds 19581da177e4SLinus Torvalds mark_page_accessed(page); 19591da177e4SLinus Torvalds lock_page(page); 19601da177e4SLinus Torvalds 19611da177e4SLinus Torvalds /* 19628f4e2101SHugh Dickins * Back out if somebody else already faulted in this pte. 19631da177e4SLinus Torvalds */ 19648f4e2101SHugh Dickins page_table = pte_offset_map_lock(mm, pmd, address, &ptl); 19659e9bef07SHugh Dickins if (unlikely(!pte_same(*page_table, orig_pte))) 1966b8107480SKirill Korotaev goto out_nomap; 1967b8107480SKirill Korotaev 1968b8107480SKirill Korotaev if (unlikely(!PageUptodate(page))) { 1969b8107480SKirill Korotaev ret = VM_FAULT_SIGBUS; 1970b8107480SKirill Korotaev goto out_nomap; 19711da177e4SLinus Torvalds } 19721da177e4SLinus Torvalds 19731da177e4SLinus Torvalds /* The page isn't present yet, go ahead with the fault. */ 19741da177e4SLinus Torvalds 19754294621fSHugh Dickins inc_mm_counter(mm, anon_rss); 19761da177e4SLinus Torvalds pte = mk_pte(page, vma->vm_page_prot); 19771da177e4SLinus Torvalds if (write_access && can_share_swap_page(page)) { 19781da177e4SLinus Torvalds pte = maybe_mkwrite(pte_mkdirty(pte), vma); 19791da177e4SLinus Torvalds write_access = 0; 19801da177e4SLinus Torvalds } 19811da177e4SLinus Torvalds 19821da177e4SLinus Torvalds flush_icache_page(vma, page); 19831da177e4SLinus Torvalds set_pte_at(mm, address, page_table, pte); 19841da177e4SLinus Torvalds page_add_anon_rmap(page, vma, address); 19851da177e4SLinus Torvalds 1986c475a8abSHugh Dickins swap_free(entry); 1987c475a8abSHugh Dickins if (vm_swap_full()) 1988c475a8abSHugh Dickins remove_exclusive_swap_page(page); 1989c475a8abSHugh Dickins unlock_page(page); 1990c475a8abSHugh Dickins 19911da177e4SLinus Torvalds if (write_access) { 19921da177e4SLinus Torvalds if (do_wp_page(mm, vma, address, 19938f4e2101SHugh Dickins page_table, pmd, ptl, pte) == VM_FAULT_OOM) 19941da177e4SLinus Torvalds ret = VM_FAULT_OOM; 19951da177e4SLinus Torvalds goto out; 19961da177e4SLinus Torvalds } 19971da177e4SLinus Torvalds 19981da177e4SLinus Torvalds /* No need to invalidate - it was non-present before */ 19991da177e4SLinus Torvalds update_mmu_cache(vma, address, pte); 20001da177e4SLinus Torvalds lazy_mmu_prot_update(pte); 200165500d23SHugh Dickins unlock: 20028f4e2101SHugh Dickins pte_unmap_unlock(page_table, ptl); 20031da177e4SLinus Torvalds out: 20041da177e4SLinus Torvalds return ret; 2005b8107480SKirill Korotaev out_nomap: 20068f4e2101SHugh Dickins pte_unmap_unlock(page_table, ptl); 2007b8107480SKirill Korotaev unlock_page(page); 2008b8107480SKirill Korotaev page_cache_release(page); 200965500d23SHugh Dickins return ret; 20101da177e4SLinus Torvalds } 20111da177e4SLinus Torvalds 20121da177e4SLinus Torvalds /* 20138f4e2101SHugh Dickins * We enter with non-exclusive mmap_sem (to exclude vma changes, 20148f4e2101SHugh Dickins * but allow concurrent faults), and pte mapped but not yet locked. 20158f4e2101SHugh Dickins * We return with mmap_sem still held, but pte unmapped and unlocked. 20161da177e4SLinus Torvalds */ 201765500d23SHugh Dickins static int do_anonymous_page(struct mm_struct *mm, struct vm_area_struct *vma, 201865500d23SHugh Dickins unsigned long address, pte_t *page_table, pmd_t *pmd, 201965500d23SHugh Dickins int write_access) 20201da177e4SLinus Torvalds { 20218f4e2101SHugh Dickins struct page *page; 20228f4e2101SHugh Dickins spinlock_t *ptl; 20231da177e4SLinus Torvalds pte_t entry; 20241da177e4SLinus Torvalds 20256aab341eSLinus Torvalds if (write_access) { 20261da177e4SLinus Torvalds /* Allocate our own private page. */ 20271da177e4SLinus Torvalds pte_unmap(page_table); 20281da177e4SLinus Torvalds 20291da177e4SLinus Torvalds if (unlikely(anon_vma_prepare(vma))) 203065500d23SHugh Dickins goto oom; 203165500d23SHugh Dickins page = alloc_zeroed_user_highpage(vma, address); 20321da177e4SLinus Torvalds if (!page) 203365500d23SHugh Dickins goto oom; 20341da177e4SLinus Torvalds 203565500d23SHugh Dickins entry = mk_pte(page, vma->vm_page_prot); 203665500d23SHugh Dickins entry = maybe_mkwrite(pte_mkdirty(entry), vma); 20378f4e2101SHugh Dickins 20388f4e2101SHugh Dickins page_table = pte_offset_map_lock(mm, pmd, address, &ptl); 20398f4e2101SHugh Dickins if (!pte_none(*page_table)) 20408f4e2101SHugh Dickins goto release; 20418f4e2101SHugh Dickins inc_mm_counter(mm, anon_rss); 20421da177e4SLinus Torvalds lru_cache_add_active(page); 20439617d95eSNick Piggin page_add_new_anon_rmap(page, vma, address); 2044b5810039SNick Piggin } else { 20458f4e2101SHugh Dickins /* Map the ZERO_PAGE - vm_page_prot is readonly */ 20468f4e2101SHugh Dickins page = ZERO_PAGE(address); 20478f4e2101SHugh Dickins page_cache_get(page); 20488f4e2101SHugh Dickins entry = mk_pte(page, vma->vm_page_prot); 20498f4e2101SHugh Dickins 20504c21e2f2SHugh Dickins ptl = pte_lockptr(mm, pmd); 20518f4e2101SHugh Dickins spin_lock(ptl); 20528f4e2101SHugh Dickins if (!pte_none(*page_table)) 20538f4e2101SHugh Dickins goto release; 2054b5810039SNick Piggin inc_mm_counter(mm, file_rss); 2055b5810039SNick Piggin page_add_file_rmap(page); 20561da177e4SLinus Torvalds } 20571da177e4SLinus Torvalds 205865500d23SHugh Dickins set_pte_at(mm, address, page_table, entry); 20591da177e4SLinus Torvalds 20601da177e4SLinus Torvalds /* No need to invalidate - it was non-present before */ 206165500d23SHugh Dickins update_mmu_cache(vma, address, entry); 20621da177e4SLinus Torvalds lazy_mmu_prot_update(entry); 206365500d23SHugh Dickins unlock: 20648f4e2101SHugh Dickins pte_unmap_unlock(page_table, ptl); 20651da177e4SLinus Torvalds return VM_FAULT_MINOR; 20668f4e2101SHugh Dickins release: 20678f4e2101SHugh Dickins page_cache_release(page); 20688f4e2101SHugh Dickins goto unlock; 206965500d23SHugh Dickins oom: 20701da177e4SLinus Torvalds return VM_FAULT_OOM; 20711da177e4SLinus Torvalds } 20721da177e4SLinus Torvalds 20731da177e4SLinus Torvalds /* 20741da177e4SLinus Torvalds * do_no_page() tries to create a new page mapping. It aggressively 20751da177e4SLinus Torvalds * tries to share with existing pages, but makes a separate copy if 20761da177e4SLinus Torvalds * the "write_access" parameter is true in order to avoid the next 20771da177e4SLinus Torvalds * page fault. 20781da177e4SLinus Torvalds * 20791da177e4SLinus Torvalds * As this is called only for pages that do not currently exist, we 20801da177e4SLinus Torvalds * do not need to flush old virtual caches or the TLB. 20811da177e4SLinus Torvalds * 20828f4e2101SHugh Dickins * We enter with non-exclusive mmap_sem (to exclude vma changes, 20838f4e2101SHugh Dickins * but allow concurrent faults), and pte mapped but not yet locked. 20848f4e2101SHugh Dickins * We return with mmap_sem still held, but pte unmapped and unlocked. 20851da177e4SLinus Torvalds */ 208665500d23SHugh Dickins static int do_no_page(struct mm_struct *mm, struct vm_area_struct *vma, 208765500d23SHugh Dickins unsigned long address, pte_t *page_table, pmd_t *pmd, 208865500d23SHugh Dickins int write_access) 20891da177e4SLinus Torvalds { 20908f4e2101SHugh Dickins spinlock_t *ptl; 20911da177e4SLinus Torvalds struct page *new_page; 20921da177e4SLinus Torvalds struct address_space *mapping = NULL; 20931da177e4SLinus Torvalds pte_t entry; 20941da177e4SLinus Torvalds unsigned int sequence = 0; 20951da177e4SLinus Torvalds int ret = VM_FAULT_MINOR; 20961da177e4SLinus Torvalds int anon = 0; 20971da177e4SLinus Torvalds 20981da177e4SLinus Torvalds pte_unmap(page_table); 2099325f04dbSHugh Dickins BUG_ON(vma->vm_flags & VM_PFNMAP); 2100325f04dbSHugh Dickins 21011da177e4SLinus Torvalds if (vma->vm_file) { 21021da177e4SLinus Torvalds mapping = vma->vm_file->f_mapping; 21031da177e4SLinus Torvalds sequence = mapping->truncate_count; 21041da177e4SLinus Torvalds smp_rmb(); /* serializes i_size against truncate_count */ 21051da177e4SLinus Torvalds } 21061da177e4SLinus Torvalds retry: 21071da177e4SLinus Torvalds new_page = vma->vm_ops->nopage(vma, address & PAGE_MASK, &ret); 21081da177e4SLinus Torvalds /* 21091da177e4SLinus Torvalds * No smp_rmb is needed here as long as there's a full 21101da177e4SLinus Torvalds * spin_lock/unlock sequence inside the ->nopage callback 21111da177e4SLinus Torvalds * (for the pagecache lookup) that acts as an implicit 21121da177e4SLinus Torvalds * smp_mb() and prevents the i_size read to happen 21131da177e4SLinus Torvalds * after the next truncate_count read. 21141da177e4SLinus Torvalds */ 21151da177e4SLinus Torvalds 21161da177e4SLinus Torvalds /* no page was available -- either SIGBUS or OOM */ 21171da177e4SLinus Torvalds if (new_page == NOPAGE_SIGBUS) 21181da177e4SLinus Torvalds return VM_FAULT_SIGBUS; 21191da177e4SLinus Torvalds if (new_page == NOPAGE_OOM) 21201da177e4SLinus Torvalds return VM_FAULT_OOM; 21211da177e4SLinus Torvalds 21221da177e4SLinus Torvalds /* 21231da177e4SLinus Torvalds * Should we do an early C-O-W break? 21241da177e4SLinus Torvalds */ 21259637a5efSDavid Howells if (write_access) { 21269637a5efSDavid Howells if (!(vma->vm_flags & VM_SHARED)) { 21271da177e4SLinus Torvalds struct page *page; 21281da177e4SLinus Torvalds 21291da177e4SLinus Torvalds if (unlikely(anon_vma_prepare(vma))) 21301da177e4SLinus Torvalds goto oom; 21311da177e4SLinus Torvalds page = alloc_page_vma(GFP_HIGHUSER, vma, address); 21321da177e4SLinus Torvalds if (!page) 21331da177e4SLinus Torvalds goto oom; 2134325f04dbSHugh Dickins copy_user_highpage(page, new_page, address); 21351da177e4SLinus Torvalds page_cache_release(new_page); 21361da177e4SLinus Torvalds new_page = page; 21371da177e4SLinus Torvalds anon = 1; 21389637a5efSDavid Howells 21399637a5efSDavid Howells } else { 21409637a5efSDavid Howells /* if the page will be shareable, see if the backing 21419637a5efSDavid Howells * address space wants to know that the page is about 21429637a5efSDavid Howells * to become writable */ 21439637a5efSDavid Howells if (vma->vm_ops->page_mkwrite && 21449637a5efSDavid Howells vma->vm_ops->page_mkwrite(vma, new_page) < 0 21459637a5efSDavid Howells ) { 21469637a5efSDavid Howells page_cache_release(new_page); 21479637a5efSDavid Howells return VM_FAULT_SIGBUS; 21489637a5efSDavid Howells } 21499637a5efSDavid Howells } 21501da177e4SLinus Torvalds } 21511da177e4SLinus Torvalds 21528f4e2101SHugh Dickins page_table = pte_offset_map_lock(mm, pmd, address, &ptl); 21531da177e4SLinus Torvalds /* 21541da177e4SLinus Torvalds * For a file-backed vma, someone could have truncated or otherwise 21551da177e4SLinus Torvalds * invalidated this page. If unmap_mapping_range got called, 21561da177e4SLinus Torvalds * retry getting the page. 21571da177e4SLinus Torvalds */ 21581da177e4SLinus Torvalds if (mapping && unlikely(sequence != mapping->truncate_count)) { 21598f4e2101SHugh Dickins pte_unmap_unlock(page_table, ptl); 21601da177e4SLinus Torvalds page_cache_release(new_page); 216165500d23SHugh Dickins cond_resched(); 216265500d23SHugh Dickins sequence = mapping->truncate_count; 216365500d23SHugh Dickins smp_rmb(); 21641da177e4SLinus Torvalds goto retry; 21651da177e4SLinus Torvalds } 21661da177e4SLinus Torvalds 21671da177e4SLinus Torvalds /* 21681da177e4SLinus Torvalds * This silly early PAGE_DIRTY setting removes a race 21691da177e4SLinus Torvalds * due to the bad i386 page protection. But it's valid 21701da177e4SLinus Torvalds * for other architectures too. 21711da177e4SLinus Torvalds * 21721da177e4SLinus Torvalds * Note that if write_access is true, we either now have 21731da177e4SLinus Torvalds * an exclusive copy of the page, or this is a shared mapping, 21741da177e4SLinus Torvalds * so we can make it writable and dirty to avoid having to 21751da177e4SLinus Torvalds * handle that later. 21761da177e4SLinus Torvalds */ 21771da177e4SLinus Torvalds /* Only go through if we didn't race with anybody else... */ 21781da177e4SLinus Torvalds if (pte_none(*page_table)) { 21791da177e4SLinus Torvalds flush_icache_page(vma, new_page); 21801da177e4SLinus Torvalds entry = mk_pte(new_page, vma->vm_page_prot); 21811da177e4SLinus Torvalds if (write_access) 21821da177e4SLinus Torvalds entry = maybe_mkwrite(pte_mkdirty(entry), vma); 21831da177e4SLinus Torvalds set_pte_at(mm, address, page_table, entry); 21841da177e4SLinus Torvalds if (anon) { 21854294621fSHugh Dickins inc_mm_counter(mm, anon_rss); 21861da177e4SLinus Torvalds lru_cache_add_active(new_page); 21879617d95eSNick Piggin page_add_new_anon_rmap(new_page, vma, address); 2188f57e88a8SHugh Dickins } else { 21894294621fSHugh Dickins inc_mm_counter(mm, file_rss); 21901da177e4SLinus Torvalds page_add_file_rmap(new_page); 21914294621fSHugh Dickins } 21921da177e4SLinus Torvalds } else { 21931da177e4SLinus Torvalds /* One of our sibling threads was faster, back out. */ 21941da177e4SLinus Torvalds page_cache_release(new_page); 219565500d23SHugh Dickins goto unlock; 21961da177e4SLinus Torvalds } 21971da177e4SLinus Torvalds 21981da177e4SLinus Torvalds /* no need to invalidate: a not-present page shouldn't be cached */ 21991da177e4SLinus Torvalds update_mmu_cache(vma, address, entry); 22001da177e4SLinus Torvalds lazy_mmu_prot_update(entry); 220165500d23SHugh Dickins unlock: 22028f4e2101SHugh Dickins pte_unmap_unlock(page_table, ptl); 22031da177e4SLinus Torvalds return ret; 22041da177e4SLinus Torvalds oom: 22051da177e4SLinus Torvalds page_cache_release(new_page); 220665500d23SHugh Dickins return VM_FAULT_OOM; 22071da177e4SLinus Torvalds } 22081da177e4SLinus Torvalds 22091da177e4SLinus Torvalds /* 22101da177e4SLinus Torvalds * Fault of a previously existing named mapping. Repopulate the pte 22111da177e4SLinus Torvalds * from the encoded file_pte if possible. This enables swappable 22121da177e4SLinus Torvalds * nonlinear vmas. 22138f4e2101SHugh Dickins * 22148f4e2101SHugh Dickins * We enter with non-exclusive mmap_sem (to exclude vma changes, 22158f4e2101SHugh Dickins * but allow concurrent faults), and pte mapped but not yet locked. 22168f4e2101SHugh Dickins * We return with mmap_sem still held, but pte unmapped and unlocked. 22171da177e4SLinus Torvalds */ 22181da177e4SLinus Torvalds static int do_file_page(struct mm_struct *mm, struct vm_area_struct *vma, 221965500d23SHugh Dickins unsigned long address, pte_t *page_table, pmd_t *pmd, 222065500d23SHugh Dickins int write_access, pte_t orig_pte) 22211da177e4SLinus Torvalds { 222265500d23SHugh Dickins pgoff_t pgoff; 22231da177e4SLinus Torvalds int err; 22241da177e4SLinus Torvalds 22254c21e2f2SHugh Dickins if (!pte_unmap_same(mm, pmd, page_table, orig_pte)) 22268f4e2101SHugh Dickins return VM_FAULT_MINOR; 22271da177e4SLinus Torvalds 222865500d23SHugh Dickins if (unlikely(!(vma->vm_flags & VM_NONLINEAR))) { 222965500d23SHugh Dickins /* 223065500d23SHugh Dickins * Page table corrupted: show pte and kill process. 223165500d23SHugh Dickins */ 2232b5810039SNick Piggin print_bad_pte(vma, orig_pte, address); 223365500d23SHugh Dickins return VM_FAULT_OOM; 223465500d23SHugh Dickins } 223565500d23SHugh Dickins /* We can then assume vm->vm_ops && vma->vm_ops->populate */ 223665500d23SHugh Dickins 223765500d23SHugh Dickins pgoff = pte_to_pgoff(orig_pte); 223865500d23SHugh Dickins err = vma->vm_ops->populate(vma, address & PAGE_MASK, PAGE_SIZE, 223965500d23SHugh Dickins vma->vm_page_prot, pgoff, 0); 22401da177e4SLinus Torvalds if (err == -ENOMEM) 22411da177e4SLinus Torvalds return VM_FAULT_OOM; 22421da177e4SLinus Torvalds if (err) 22431da177e4SLinus Torvalds return VM_FAULT_SIGBUS; 22441da177e4SLinus Torvalds return VM_FAULT_MAJOR; 22451da177e4SLinus Torvalds } 22461da177e4SLinus Torvalds 22471da177e4SLinus Torvalds /* 22481da177e4SLinus Torvalds * These routines also need to handle stuff like marking pages dirty 22491da177e4SLinus Torvalds * and/or accessed for architectures that don't do it in hardware (most 22501da177e4SLinus Torvalds * RISC architectures). The early dirtying is also good on the i386. 22511da177e4SLinus Torvalds * 22521da177e4SLinus Torvalds * There is also a hook called "update_mmu_cache()" that architectures 22531da177e4SLinus Torvalds * with external mmu caches can use to update those (ie the Sparc or 22541da177e4SLinus Torvalds * PowerPC hashed page tables that act as extended TLBs). 22551da177e4SLinus Torvalds * 2256c74df32cSHugh Dickins * We enter with non-exclusive mmap_sem (to exclude vma changes, 2257c74df32cSHugh Dickins * but allow concurrent faults), and pte mapped but not yet locked. 2258c74df32cSHugh Dickins * We return with mmap_sem still held, but pte unmapped and unlocked. 22591da177e4SLinus Torvalds */ 22601da177e4SLinus Torvalds static inline int handle_pte_fault(struct mm_struct *mm, 22611da177e4SLinus Torvalds struct vm_area_struct *vma, unsigned long address, 226265500d23SHugh Dickins pte_t *pte, pmd_t *pmd, int write_access) 22631da177e4SLinus Torvalds { 22641da177e4SLinus Torvalds pte_t entry; 22651a44e149SAndrea Arcangeli pte_t old_entry; 22668f4e2101SHugh Dickins spinlock_t *ptl; 22671da177e4SLinus Torvalds 22681a44e149SAndrea Arcangeli old_entry = entry = *pte; 22691da177e4SLinus Torvalds if (!pte_present(entry)) { 227065500d23SHugh Dickins if (pte_none(entry)) { 227165500d23SHugh Dickins if (!vma->vm_ops || !vma->vm_ops->nopage) 227265500d23SHugh Dickins return do_anonymous_page(mm, vma, address, 227365500d23SHugh Dickins pte, pmd, write_access); 227465500d23SHugh Dickins return do_no_page(mm, vma, address, 227565500d23SHugh Dickins pte, pmd, write_access); 227665500d23SHugh Dickins } 22771da177e4SLinus Torvalds if (pte_file(entry)) 227865500d23SHugh Dickins return do_file_page(mm, vma, address, 227965500d23SHugh Dickins pte, pmd, write_access, entry); 228065500d23SHugh Dickins return do_swap_page(mm, vma, address, 228165500d23SHugh Dickins pte, pmd, write_access, entry); 22821da177e4SLinus Torvalds } 22831da177e4SLinus Torvalds 22844c21e2f2SHugh Dickins ptl = pte_lockptr(mm, pmd); 22858f4e2101SHugh Dickins spin_lock(ptl); 22868f4e2101SHugh Dickins if (unlikely(!pte_same(*pte, entry))) 22878f4e2101SHugh Dickins goto unlock; 22881da177e4SLinus Torvalds if (write_access) { 22891da177e4SLinus Torvalds if (!pte_write(entry)) 22908f4e2101SHugh Dickins return do_wp_page(mm, vma, address, 22918f4e2101SHugh Dickins pte, pmd, ptl, entry); 22921da177e4SLinus Torvalds entry = pte_mkdirty(entry); 22931da177e4SLinus Torvalds } 22941da177e4SLinus Torvalds entry = pte_mkyoung(entry); 22951a44e149SAndrea Arcangeli if (!pte_same(old_entry, entry)) { 22961da177e4SLinus Torvalds ptep_set_access_flags(vma, address, pte, entry, write_access); 22971da177e4SLinus Torvalds update_mmu_cache(vma, address, entry); 22981da177e4SLinus Torvalds lazy_mmu_prot_update(entry); 22991a44e149SAndrea Arcangeli } else { 23001a44e149SAndrea Arcangeli /* 23011a44e149SAndrea Arcangeli * This is needed only for protection faults but the arch code 23021a44e149SAndrea Arcangeli * is not yet telling us if this is a protection fault or not. 23031a44e149SAndrea Arcangeli * This still avoids useless tlb flushes for .text page faults 23041a44e149SAndrea Arcangeli * with threads. 23051a44e149SAndrea Arcangeli */ 23061a44e149SAndrea Arcangeli if (write_access) 23071a44e149SAndrea Arcangeli flush_tlb_page(vma, address); 23081a44e149SAndrea Arcangeli } 23098f4e2101SHugh Dickins unlock: 23108f4e2101SHugh Dickins pte_unmap_unlock(pte, ptl); 23111da177e4SLinus Torvalds return VM_FAULT_MINOR; 23121da177e4SLinus Torvalds } 23131da177e4SLinus Torvalds 23141da177e4SLinus Torvalds /* 23151da177e4SLinus Torvalds * By the time we get here, we already hold the mm semaphore 23161da177e4SLinus Torvalds */ 2317f33ea7f4SNick Piggin int __handle_mm_fault(struct mm_struct *mm, struct vm_area_struct *vma, 23181da177e4SLinus Torvalds unsigned long address, int write_access) 23191da177e4SLinus Torvalds { 23201da177e4SLinus Torvalds pgd_t *pgd; 23211da177e4SLinus Torvalds pud_t *pud; 23221da177e4SLinus Torvalds pmd_t *pmd; 23231da177e4SLinus Torvalds pte_t *pte; 23241da177e4SLinus Torvalds 23251da177e4SLinus Torvalds __set_current_state(TASK_RUNNING); 23261da177e4SLinus Torvalds 2327*f8891e5eSChristoph Lameter count_vm_event(PGFAULT); 23281da177e4SLinus Torvalds 2329ac9b9c66SHugh Dickins if (unlikely(is_vm_hugetlb_page(vma))) 2330ac9b9c66SHugh Dickins return hugetlb_fault(mm, vma, address, write_access); 23311da177e4SLinus Torvalds 23321da177e4SLinus Torvalds pgd = pgd_offset(mm, address); 23331da177e4SLinus Torvalds pud = pud_alloc(mm, pgd, address); 23341da177e4SLinus Torvalds if (!pud) 2335c74df32cSHugh Dickins return VM_FAULT_OOM; 23361da177e4SLinus Torvalds pmd = pmd_alloc(mm, pud, address); 23371da177e4SLinus Torvalds if (!pmd) 2338c74df32cSHugh Dickins return VM_FAULT_OOM; 23391da177e4SLinus Torvalds pte = pte_alloc_map(mm, pmd, address); 23401da177e4SLinus Torvalds if (!pte) 2341c74df32cSHugh Dickins return VM_FAULT_OOM; 23421da177e4SLinus Torvalds 234365500d23SHugh Dickins return handle_pte_fault(mm, vma, address, pte, pmd, write_access); 23441da177e4SLinus Torvalds } 23451da177e4SLinus Torvalds 234667207b96SArnd Bergmann EXPORT_SYMBOL_GPL(__handle_mm_fault); 234767207b96SArnd Bergmann 23481da177e4SLinus Torvalds #ifndef __PAGETABLE_PUD_FOLDED 23491da177e4SLinus Torvalds /* 23501da177e4SLinus Torvalds * Allocate page upper directory. 2351872fec16SHugh Dickins * We've already handled the fast-path in-line. 23521da177e4SLinus Torvalds */ 23531bb3630eSHugh Dickins int __pud_alloc(struct mm_struct *mm, pgd_t *pgd, unsigned long address) 23541da177e4SLinus Torvalds { 2355c74df32cSHugh Dickins pud_t *new = pud_alloc_one(mm, address); 2356c74df32cSHugh Dickins if (!new) 23571bb3630eSHugh Dickins return -ENOMEM; 23581da177e4SLinus Torvalds 2359872fec16SHugh Dickins spin_lock(&mm->page_table_lock); 23601bb3630eSHugh Dickins if (pgd_present(*pgd)) /* Another has populated it */ 23611da177e4SLinus Torvalds pud_free(new); 23621bb3630eSHugh Dickins else 23631da177e4SLinus Torvalds pgd_populate(mm, pgd, new); 2364872fec16SHugh Dickins spin_unlock(&mm->page_table_lock); 23651bb3630eSHugh Dickins return 0; 23661da177e4SLinus Torvalds } 2367e0f39591SAlan Stern #else 2368e0f39591SAlan Stern /* Workaround for gcc 2.96 */ 2369e0f39591SAlan Stern int __pud_alloc(struct mm_struct *mm, pgd_t *pgd, unsigned long address) 2370e0f39591SAlan Stern { 2371e0f39591SAlan Stern return 0; 2372e0f39591SAlan Stern } 23731da177e4SLinus Torvalds #endif /* __PAGETABLE_PUD_FOLDED */ 23741da177e4SLinus Torvalds 23751da177e4SLinus Torvalds #ifndef __PAGETABLE_PMD_FOLDED 23761da177e4SLinus Torvalds /* 23771da177e4SLinus Torvalds * Allocate page middle directory. 2378872fec16SHugh Dickins * We've already handled the fast-path in-line. 23791da177e4SLinus Torvalds */ 23801bb3630eSHugh Dickins int __pmd_alloc(struct mm_struct *mm, pud_t *pud, unsigned long address) 23811da177e4SLinus Torvalds { 2382c74df32cSHugh Dickins pmd_t *new = pmd_alloc_one(mm, address); 2383c74df32cSHugh Dickins if (!new) 23841bb3630eSHugh Dickins return -ENOMEM; 23851da177e4SLinus Torvalds 2386872fec16SHugh Dickins spin_lock(&mm->page_table_lock); 23871da177e4SLinus Torvalds #ifndef __ARCH_HAS_4LEVEL_HACK 23881bb3630eSHugh Dickins if (pud_present(*pud)) /* Another has populated it */ 23891da177e4SLinus Torvalds pmd_free(new); 23901bb3630eSHugh Dickins else 23911da177e4SLinus Torvalds pud_populate(mm, pud, new); 23921da177e4SLinus Torvalds #else 23931bb3630eSHugh Dickins if (pgd_present(*pud)) /* Another has populated it */ 23941da177e4SLinus Torvalds pmd_free(new); 23951bb3630eSHugh Dickins else 23961da177e4SLinus Torvalds pgd_populate(mm, pud, new); 23971da177e4SLinus Torvalds #endif /* __ARCH_HAS_4LEVEL_HACK */ 2398872fec16SHugh Dickins spin_unlock(&mm->page_table_lock); 23991bb3630eSHugh Dickins return 0; 24001da177e4SLinus Torvalds } 2401e0f39591SAlan Stern #else 2402e0f39591SAlan Stern /* Workaround for gcc 2.96 */ 2403e0f39591SAlan Stern int __pmd_alloc(struct mm_struct *mm, pud_t *pud, unsigned long address) 2404e0f39591SAlan Stern { 2405e0f39591SAlan Stern return 0; 2406e0f39591SAlan Stern } 24071da177e4SLinus Torvalds #endif /* __PAGETABLE_PMD_FOLDED */ 24081da177e4SLinus Torvalds 24091da177e4SLinus Torvalds int make_pages_present(unsigned long addr, unsigned long end) 24101da177e4SLinus Torvalds { 24111da177e4SLinus Torvalds int ret, len, write; 24121da177e4SLinus Torvalds struct vm_area_struct * vma; 24131da177e4SLinus Torvalds 24141da177e4SLinus Torvalds vma = find_vma(current->mm, addr); 24151da177e4SLinus Torvalds if (!vma) 24161da177e4SLinus Torvalds return -1; 24171da177e4SLinus Torvalds write = (vma->vm_flags & VM_WRITE) != 0; 24185bcb28b1SEric Sesterhenn BUG_ON(addr >= end); 24195bcb28b1SEric Sesterhenn BUG_ON(end > vma->vm_end); 24201da177e4SLinus Torvalds len = (end+PAGE_SIZE-1)/PAGE_SIZE-addr/PAGE_SIZE; 24211da177e4SLinus Torvalds ret = get_user_pages(current, current->mm, addr, 24221da177e4SLinus Torvalds len, write, 0, NULL, NULL); 24231da177e4SLinus Torvalds if (ret < 0) 24241da177e4SLinus Torvalds return ret; 24251da177e4SLinus Torvalds return ret == len ? 0 : -1; 24261da177e4SLinus Torvalds } 24271da177e4SLinus Torvalds 24281da177e4SLinus Torvalds /* 24291da177e4SLinus Torvalds * Map a vmalloc()-space virtual address to the physical page. 24301da177e4SLinus Torvalds */ 24311da177e4SLinus Torvalds struct page * vmalloc_to_page(void * vmalloc_addr) 24321da177e4SLinus Torvalds { 24331da177e4SLinus Torvalds unsigned long addr = (unsigned long) vmalloc_addr; 24341da177e4SLinus Torvalds struct page *page = NULL; 24351da177e4SLinus Torvalds pgd_t *pgd = pgd_offset_k(addr); 24361da177e4SLinus Torvalds pud_t *pud; 24371da177e4SLinus Torvalds pmd_t *pmd; 24381da177e4SLinus Torvalds pte_t *ptep, pte; 24391da177e4SLinus Torvalds 24401da177e4SLinus Torvalds if (!pgd_none(*pgd)) { 24411da177e4SLinus Torvalds pud = pud_offset(pgd, addr); 24421da177e4SLinus Torvalds if (!pud_none(*pud)) { 24431da177e4SLinus Torvalds pmd = pmd_offset(pud, addr); 24441da177e4SLinus Torvalds if (!pmd_none(*pmd)) { 24451da177e4SLinus Torvalds ptep = pte_offset_map(pmd, addr); 24461da177e4SLinus Torvalds pte = *ptep; 24471da177e4SLinus Torvalds if (pte_present(pte)) 24481da177e4SLinus Torvalds page = pte_page(pte); 24491da177e4SLinus Torvalds pte_unmap(ptep); 24501da177e4SLinus Torvalds } 24511da177e4SLinus Torvalds } 24521da177e4SLinus Torvalds } 24531da177e4SLinus Torvalds return page; 24541da177e4SLinus Torvalds } 24551da177e4SLinus Torvalds 24561da177e4SLinus Torvalds EXPORT_SYMBOL(vmalloc_to_page); 24571da177e4SLinus Torvalds 24581da177e4SLinus Torvalds /* 24591da177e4SLinus Torvalds * Map a vmalloc()-space virtual address to the physical page frame number. 24601da177e4SLinus Torvalds */ 24611da177e4SLinus Torvalds unsigned long vmalloc_to_pfn(void * vmalloc_addr) 24621da177e4SLinus Torvalds { 24631da177e4SLinus Torvalds return page_to_pfn(vmalloc_to_page(vmalloc_addr)); 24641da177e4SLinus Torvalds } 24651da177e4SLinus Torvalds 24661da177e4SLinus Torvalds EXPORT_SYMBOL(vmalloc_to_pfn); 24671da177e4SLinus Torvalds 24681da177e4SLinus Torvalds #if !defined(__HAVE_ARCH_GATE_AREA) 24691da177e4SLinus Torvalds 24701da177e4SLinus Torvalds #if defined(AT_SYSINFO_EHDR) 24715ce7852cSAdrian Bunk static struct vm_area_struct gate_vma; 24721da177e4SLinus Torvalds 24731da177e4SLinus Torvalds static int __init gate_vma_init(void) 24741da177e4SLinus Torvalds { 24751da177e4SLinus Torvalds gate_vma.vm_mm = NULL; 24761da177e4SLinus Torvalds gate_vma.vm_start = FIXADDR_USER_START; 24771da177e4SLinus Torvalds gate_vma.vm_end = FIXADDR_USER_END; 24781da177e4SLinus Torvalds gate_vma.vm_page_prot = PAGE_READONLY; 24790b14c179SHugh Dickins gate_vma.vm_flags = 0; 24801da177e4SLinus Torvalds return 0; 24811da177e4SLinus Torvalds } 24821da177e4SLinus Torvalds __initcall(gate_vma_init); 24831da177e4SLinus Torvalds #endif 24841da177e4SLinus Torvalds 24851da177e4SLinus Torvalds struct vm_area_struct *get_gate_vma(struct task_struct *tsk) 24861da177e4SLinus Torvalds { 24871da177e4SLinus Torvalds #ifdef AT_SYSINFO_EHDR 24881da177e4SLinus Torvalds return &gate_vma; 24891da177e4SLinus Torvalds #else 24901da177e4SLinus Torvalds return NULL; 24911da177e4SLinus Torvalds #endif 24921da177e4SLinus Torvalds } 24931da177e4SLinus Torvalds 24941da177e4SLinus Torvalds int in_gate_area_no_task(unsigned long addr) 24951da177e4SLinus Torvalds { 24961da177e4SLinus Torvalds #ifdef AT_SYSINFO_EHDR 24971da177e4SLinus Torvalds if ((addr >= FIXADDR_USER_START) && (addr < FIXADDR_USER_END)) 24981da177e4SLinus Torvalds return 1; 24991da177e4SLinus Torvalds #endif 25001da177e4SLinus Torvalds return 0; 25011da177e4SLinus Torvalds } 25021da177e4SLinus Torvalds 25031da177e4SLinus Torvalds #endif /* __HAVE_ARCH_GATE_AREA */ 2504