11da177e4SLinus Torvalds /* 21da177e4SLinus Torvalds * hugetlbpage-backed filesystem. Based on ramfs. 31da177e4SLinus Torvalds * 46d49e352SNadia Yvette Chambers * Nadia Yvette Chambers, 2002 51da177e4SLinus Torvalds * 61da177e4SLinus Torvalds * Copyright (C) 2002 Linus Torvalds. 73e89e1c5SPaul Gortmaker * License: GPL 81da177e4SLinus Torvalds */ 91da177e4SLinus Torvalds 109b857d26SAndrew Morton #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt 119b857d26SAndrew Morton 121da177e4SLinus Torvalds #include <linux/thread_info.h> 131da177e4SLinus Torvalds #include <asm/current.h> 1470c3547eSMike Kravetz #include <linux/falloc.h> 151da177e4SLinus Torvalds #include <linux/fs.h> 161da177e4SLinus Torvalds #include <linux/mount.h> 171da177e4SLinus Torvalds #include <linux/file.h> 18e73a75faSRandy Dunlap #include <linux/kernel.h> 191da177e4SLinus Torvalds #include <linux/writeback.h> 201da177e4SLinus Torvalds #include <linux/pagemap.h> 211da177e4SLinus Torvalds #include <linux/highmem.h> 221da177e4SLinus Torvalds #include <linux/init.h> 231da177e4SLinus Torvalds #include <linux/string.h> 2416f7e0feSRandy Dunlap #include <linux/capability.h> 25e73a75faSRandy Dunlap #include <linux/ctype.h> 261da177e4SLinus Torvalds #include <linux/backing-dev.h> 271da177e4SLinus Torvalds #include <linux/hugetlb.h> 281da177e4SLinus Torvalds #include <linux/pagevec.h> 2932021982SDavid Howells #include <linux/fs_parser.h> 30036e0856SBenjamin Herrenschmidt #include <linux/mman.h> 311da177e4SLinus Torvalds #include <linux/slab.h> 321da177e4SLinus Torvalds #include <linux/dnotify.h> 331da177e4SLinus Torvalds #include <linux/statfs.h> 341da177e4SLinus Torvalds #include <linux/security.h> 351fd7317dSNick Black #include <linux/magic.h> 36290408d4SNaoya Horiguchi #include <linux/migrate.h> 3734d0640eSAl Viro #include <linux/uio.h> 381da177e4SLinus Torvalds 397c0f6ba6SLinus Torvalds #include <linux/uaccess.h> 4088590253SShijie Hu #include <linux/sched/mm.h> 411da177e4SLinus Torvalds 42f5e54d6eSChristoph Hellwig static const struct address_space_operations hugetlbfs_aops; 434b6f5d20SArjan van de Ven const struct file_operations hugetlbfs_file_operations; 4492e1d5beSArjan van de Ven static const struct inode_operations hugetlbfs_dir_inode_operations; 4592e1d5beSArjan van de Ven static const struct inode_operations hugetlbfs_inode_operations; 461da177e4SLinus Torvalds 4732021982SDavid Howells enum hugetlbfs_size_type { NO_SIZE, SIZE_STD, SIZE_PERCENT }; 4832021982SDavid Howells 4932021982SDavid Howells struct hugetlbfs_fs_context { 504a25220dSDavid Howells struct hstate *hstate; 5132021982SDavid Howells unsigned long long max_size_opt; 5232021982SDavid Howells unsigned long long min_size_opt; 534a25220dSDavid Howells long max_hpages; 544a25220dSDavid Howells long nr_inodes; 554a25220dSDavid Howells long min_hpages; 5632021982SDavid Howells enum hugetlbfs_size_type max_val_type; 5732021982SDavid Howells enum hugetlbfs_size_type min_val_type; 58a0eb3a05SEric W. Biederman kuid_t uid; 59a0eb3a05SEric W. Biederman kgid_t gid; 60a1d776eeSDavid Gibson umode_t mode; 61a1d776eeSDavid Gibson }; 62a1d776eeSDavid Gibson 631da177e4SLinus Torvalds int sysctl_hugetlb_shm_group; 641da177e4SLinus Torvalds 6532021982SDavid Howells enum hugetlb_param { 6632021982SDavid Howells Opt_gid, 6732021982SDavid Howells Opt_min_size, 6832021982SDavid Howells Opt_mode, 6932021982SDavid Howells Opt_nr_inodes, 7032021982SDavid Howells Opt_pagesize, 7132021982SDavid Howells Opt_size, 7232021982SDavid Howells Opt_uid, 73e73a75faSRandy Dunlap }; 74e73a75faSRandy Dunlap 75d7167b14SAl Viro static const struct fs_parameter_spec hugetlb_fs_parameters[] = { 7632021982SDavid Howells fsparam_u32 ("gid", Opt_gid), 7732021982SDavid Howells fsparam_string("min_size", Opt_min_size), 78e0f7e2b2SMike Kravetz fsparam_u32oct("mode", Opt_mode), 7932021982SDavid Howells fsparam_string("nr_inodes", Opt_nr_inodes), 8032021982SDavid Howells fsparam_string("pagesize", Opt_pagesize), 8132021982SDavid Howells fsparam_string("size", Opt_size), 8232021982SDavid Howells fsparam_u32 ("uid", Opt_uid), 8332021982SDavid Howells {} 8432021982SDavid Howells }; 8532021982SDavid Howells 8670c3547eSMike Kravetz #ifdef CONFIG_NUMA 8770c3547eSMike Kravetz static inline void hugetlb_set_vma_policy(struct vm_area_struct *vma, 8870c3547eSMike Kravetz struct inode *inode, pgoff_t index) 8970c3547eSMike Kravetz { 9070c3547eSMike Kravetz vma->vm_policy = mpol_shared_policy_lookup(&HUGETLBFS_I(inode)->policy, 9170c3547eSMike Kravetz index); 9270c3547eSMike Kravetz } 9370c3547eSMike Kravetz 9470c3547eSMike Kravetz static inline void hugetlb_drop_vma_policy(struct vm_area_struct *vma) 9570c3547eSMike Kravetz { 9670c3547eSMike Kravetz mpol_cond_put(vma->vm_policy); 9770c3547eSMike Kravetz } 9870c3547eSMike Kravetz #else 9970c3547eSMike Kravetz static inline void hugetlb_set_vma_policy(struct vm_area_struct *vma, 10070c3547eSMike Kravetz struct inode *inode, pgoff_t index) 10170c3547eSMike Kravetz { 10270c3547eSMike Kravetz } 10370c3547eSMike Kravetz 10470c3547eSMike Kravetz static inline void hugetlb_drop_vma_policy(struct vm_area_struct *vma) 10570c3547eSMike Kravetz { 10670c3547eSMike Kravetz } 10770c3547eSMike Kravetz #endif 10870c3547eSMike Kravetz 10963489f8eSMike Kravetz /* 11063489f8eSMike Kravetz * Mask used when checking the page offset value passed in via system 11163489f8eSMike Kravetz * calls. This value will be converted to a loff_t which is signed. 11263489f8eSMike Kravetz * Therefore, we want to check the upper PAGE_SHIFT + 1 bits of the 11363489f8eSMike Kravetz * value. The extra bit (- 1 in the shift value) is to take the sign 11463489f8eSMike Kravetz * bit into account. 11563489f8eSMike Kravetz */ 11663489f8eSMike Kravetz #define PGOFF_LOFFT_MAX \ 11763489f8eSMike Kravetz (((1UL << (PAGE_SHIFT + 1)) - 1) << (BITS_PER_LONG - (PAGE_SHIFT + 1))) 11863489f8eSMike Kravetz 1191da177e4SLinus Torvalds static int hugetlbfs_file_mmap(struct file *file, struct vm_area_struct *vma) 1201da177e4SLinus Torvalds { 121496ad9aaSAl Viro struct inode *inode = file_inode(file); 12222247efdSPeter Xu struct hugetlbfs_inode_info *info = HUGETLBFS_I(inode); 1231da177e4SLinus Torvalds loff_t len, vma_len; 1241da177e4SLinus Torvalds int ret; 125a5516438SAndi Kleen struct hstate *h = hstate_file(file); 1261da177e4SLinus Torvalds 12768589bc3SHugh Dickins /* 128dec4ad86SDavid Gibson * vma address alignment (but not the pgoff alignment) has 129dec4ad86SDavid Gibson * already been checked by prepare_hugepage_range. If you add 130dec4ad86SDavid Gibson * any error returns here, do so after setting VM_HUGETLB, so 131dec4ad86SDavid Gibson * is_vm_hugetlb_page tests below unmap_region go the right 13245e55300SPeter Collingbourne * way when do_mmap unwinds (may be important on powerpc 133dec4ad86SDavid Gibson * and ia64). 13468589bc3SHugh Dickins */ 1351c71222eSSuren Baghdasaryan vm_flags_set(vma, VM_HUGETLB | VM_DONTEXPAND); 13668589bc3SHugh Dickins vma->vm_ops = &hugetlb_vm_ops; 1371da177e4SLinus Torvalds 13822247efdSPeter Xu ret = seal_check_future_write(info->seals, vma); 13922247efdSPeter Xu if (ret) 14022247efdSPeter Xu return ret; 14122247efdSPeter Xu 142045c7a3fSMike Kravetz /* 14363489f8eSMike Kravetz * page based offset in vm_pgoff could be sufficiently large to 1445df63c2aSMike Kravetz * overflow a loff_t when converted to byte offset. This can 1455df63c2aSMike Kravetz * only happen on architectures where sizeof(loff_t) == 1465df63c2aSMike Kravetz * sizeof(unsigned long). So, only check in those instances. 147045c7a3fSMike Kravetz */ 1485df63c2aSMike Kravetz if (sizeof(unsigned long) == sizeof(loff_t)) { 14963489f8eSMike Kravetz if (vma->vm_pgoff & PGOFF_LOFFT_MAX) 150045c7a3fSMike Kravetz return -EINVAL; 1515df63c2aSMike Kravetz } 152045c7a3fSMike Kravetz 15363489f8eSMike Kravetz /* must be huge page aligned */ 1542b37c35eSBecky Bruce if (vma->vm_pgoff & (~huge_page_mask(h) >> PAGE_SHIFT)) 155dec4ad86SDavid Gibson return -EINVAL; 156dec4ad86SDavid Gibson 1571da177e4SLinus Torvalds vma_len = (loff_t)(vma->vm_end - vma->vm_start); 158045c7a3fSMike Kravetz len = vma_len + ((loff_t)vma->vm_pgoff << PAGE_SHIFT); 159045c7a3fSMike Kravetz /* check for overflow */ 160045c7a3fSMike Kravetz if (len < vma_len) 161045c7a3fSMike Kravetz return -EINVAL; 1621da177e4SLinus Torvalds 1635955102cSAl Viro inode_lock(inode); 1641da177e4SLinus Torvalds file_accessed(file); 1651da177e4SLinus Torvalds 1661da177e4SLinus Torvalds ret = -ENOMEM; 16733b8f84aSMike Kravetz if (!hugetlb_reserve_pages(inode, 168a5516438SAndi Kleen vma->vm_pgoff >> huge_page_order(h), 1695a6fe125SMel Gorman len >> huge_page_shift(h), vma, 1705a6fe125SMel Gorman vma->vm_flags)) 171b45b5bd6SDavid Gibson goto out; 172b45b5bd6SDavid Gibson 1734c887265SAdam Litke ret = 0; 174b6174df5SZhang, Yanmin if (vma->vm_flags & VM_WRITE && inode->i_size < len) 175045c7a3fSMike Kravetz i_size_write(inode, len); 1761da177e4SLinus Torvalds out: 1775955102cSAl Viro inode_unlock(inode); 1781da177e4SLinus Torvalds 1791da177e4SLinus Torvalds return ret; 1801da177e4SLinus Torvalds } 1811da177e4SLinus Torvalds 1821da177e4SLinus Torvalds /* 1833e4e28c5SMichel Lespinasse * Called under mmap_write_lock(mm). 1841da177e4SLinus Torvalds */ 1851da177e4SLinus Torvalds 1861da177e4SLinus Torvalds static unsigned long 18788590253SShijie Hu hugetlb_get_unmapped_area_bottomup(struct file *file, unsigned long addr, 18888590253SShijie Hu unsigned long len, unsigned long pgoff, unsigned long flags) 18988590253SShijie Hu { 19088590253SShijie Hu struct hstate *h = hstate_file(file); 19188590253SShijie Hu struct vm_unmapped_area_info info; 19288590253SShijie Hu 19388590253SShijie Hu info.flags = 0; 19488590253SShijie Hu info.length = len; 19588590253SShijie Hu info.low_limit = current->mm->mmap_base; 1962cb4de08SChristophe Leroy info.high_limit = arch_get_mmap_end(addr, len, flags); 19788590253SShijie Hu info.align_mask = PAGE_MASK & ~huge_page_mask(h); 19888590253SShijie Hu info.align_offset = 0; 19988590253SShijie Hu return vm_unmapped_area(&info); 20088590253SShijie Hu } 20188590253SShijie Hu 20288590253SShijie Hu static unsigned long 20388590253SShijie Hu hugetlb_get_unmapped_area_topdown(struct file *file, unsigned long addr, 20488590253SShijie Hu unsigned long len, unsigned long pgoff, unsigned long flags) 20588590253SShijie Hu { 20688590253SShijie Hu struct hstate *h = hstate_file(file); 20788590253SShijie Hu struct vm_unmapped_area_info info; 20888590253SShijie Hu 20988590253SShijie Hu info.flags = VM_UNMAPPED_AREA_TOPDOWN; 21088590253SShijie Hu info.length = len; 2116b008640SLinus Torvalds info.low_limit = PAGE_SIZE; 2125f24d5a5SChristophe Leroy info.high_limit = arch_get_mmap_base(addr, current->mm->mmap_base); 21388590253SShijie Hu info.align_mask = PAGE_MASK & ~huge_page_mask(h); 21488590253SShijie Hu info.align_offset = 0; 21588590253SShijie Hu addr = vm_unmapped_area(&info); 21688590253SShijie Hu 21788590253SShijie Hu /* 21888590253SShijie Hu * A failed mmap() very likely causes application failure, 21988590253SShijie Hu * so fall back to the bottom-up function here. This scenario 22088590253SShijie Hu * can happen with large stack limits and large mmap() 22188590253SShijie Hu * allocations. 22288590253SShijie Hu */ 22388590253SShijie Hu if (unlikely(offset_in_page(addr))) { 22488590253SShijie Hu VM_BUG_ON(addr != -ENOMEM); 22588590253SShijie Hu info.flags = 0; 22688590253SShijie Hu info.low_limit = current->mm->mmap_base; 2272cb4de08SChristophe Leroy info.high_limit = arch_get_mmap_end(addr, len, flags); 22888590253SShijie Hu addr = vm_unmapped_area(&info); 22988590253SShijie Hu } 23088590253SShijie Hu 23188590253SShijie Hu return addr; 23288590253SShijie Hu } 23388590253SShijie Hu 2344b439e25SChristophe Leroy unsigned long 2354b439e25SChristophe Leroy generic_hugetlb_get_unmapped_area(struct file *file, unsigned long addr, 2364b439e25SChristophe Leroy unsigned long len, unsigned long pgoff, 2374b439e25SChristophe Leroy unsigned long flags) 2381da177e4SLinus Torvalds { 2391da177e4SLinus Torvalds struct mm_struct *mm = current->mm; 2401da177e4SLinus Torvalds struct vm_area_struct *vma; 241a5516438SAndi Kleen struct hstate *h = hstate_file(file); 2422cb4de08SChristophe Leroy const unsigned long mmap_end = arch_get_mmap_end(addr, len, flags); 2431da177e4SLinus Torvalds 244a5516438SAndi Kleen if (len & ~huge_page_mask(h)) 2451da177e4SLinus Torvalds return -EINVAL; 2461da177e4SLinus Torvalds if (len > TASK_SIZE) 2471da177e4SLinus Torvalds return -ENOMEM; 2481da177e4SLinus Torvalds 249036e0856SBenjamin Herrenschmidt if (flags & MAP_FIXED) { 250a5516438SAndi Kleen if (prepare_hugepage_range(file, addr, len)) 251036e0856SBenjamin Herrenschmidt return -EINVAL; 252036e0856SBenjamin Herrenschmidt return addr; 253036e0856SBenjamin Herrenschmidt } 254036e0856SBenjamin Herrenschmidt 2551da177e4SLinus Torvalds if (addr) { 256a5516438SAndi Kleen addr = ALIGN(addr, huge_page_size(h)); 2571da177e4SLinus Torvalds vma = find_vma(mm, addr); 2585f24d5a5SChristophe Leroy if (mmap_end - len >= addr && 2591be7107fSHugh Dickins (!vma || addr + len <= vm_start_gap(vma))) 2601da177e4SLinus Torvalds return addr; 2611da177e4SLinus Torvalds } 2621da177e4SLinus Torvalds 26388590253SShijie Hu /* 26488590253SShijie Hu * Use mm->get_unmapped_area value as a hint to use topdown routine. 26588590253SShijie Hu * If architectures have special needs, they should define their own 26688590253SShijie Hu * version of hugetlb_get_unmapped_area. 26788590253SShijie Hu */ 26888590253SShijie Hu if (mm->get_unmapped_area == arch_get_unmapped_area_topdown) 26988590253SShijie Hu return hugetlb_get_unmapped_area_topdown(file, addr, len, 27088590253SShijie Hu pgoff, flags); 27188590253SShijie Hu return hugetlb_get_unmapped_area_bottomup(file, addr, len, 27288590253SShijie Hu pgoff, flags); 2731da177e4SLinus Torvalds } 2744b439e25SChristophe Leroy 2754b439e25SChristophe Leroy #ifndef HAVE_ARCH_HUGETLB_UNMAPPED_AREA 2764b439e25SChristophe Leroy static unsigned long 2774b439e25SChristophe Leroy hugetlb_get_unmapped_area(struct file *file, unsigned long addr, 2784b439e25SChristophe Leroy unsigned long len, unsigned long pgoff, 2794b439e25SChristophe Leroy unsigned long flags) 2804b439e25SChristophe Leroy { 2814b439e25SChristophe Leroy return generic_hugetlb_get_unmapped_area(file, addr, len, pgoff, flags); 2824b439e25SChristophe Leroy } 2831da177e4SLinus Torvalds #endif 2841da177e4SLinus Torvalds 285e63e1e5aSBadari Pulavarty /* 286*38c1ddbdSJiaqi Yan * Someone wants to read @bytes from a HWPOISON hugetlb @page from @offset. 287*38c1ddbdSJiaqi Yan * Returns the maximum number of bytes one can read without touching the 1st raw 288*38c1ddbdSJiaqi Yan * HWPOISON subpage. 289*38c1ddbdSJiaqi Yan * 290*38c1ddbdSJiaqi Yan * The implementation borrows the iteration logic from copy_page_to_iter*. 291*38c1ddbdSJiaqi Yan */ 292*38c1ddbdSJiaqi Yan static size_t adjust_range_hwpoison(struct page *page, size_t offset, size_t bytes) 293*38c1ddbdSJiaqi Yan { 294*38c1ddbdSJiaqi Yan size_t n = 0; 295*38c1ddbdSJiaqi Yan size_t res = 0; 296*38c1ddbdSJiaqi Yan 297*38c1ddbdSJiaqi Yan /* First subpage to start the loop. */ 298*38c1ddbdSJiaqi Yan page += offset / PAGE_SIZE; 299*38c1ddbdSJiaqi Yan offset %= PAGE_SIZE; 300*38c1ddbdSJiaqi Yan while (1) { 301*38c1ddbdSJiaqi Yan if (is_raw_hwpoison_page_in_hugepage(page)) 302*38c1ddbdSJiaqi Yan break; 303*38c1ddbdSJiaqi Yan 304*38c1ddbdSJiaqi Yan /* Safe to read n bytes without touching HWPOISON subpage. */ 305*38c1ddbdSJiaqi Yan n = min(bytes, (size_t)PAGE_SIZE - offset); 306*38c1ddbdSJiaqi Yan res += n; 307*38c1ddbdSJiaqi Yan bytes -= n; 308*38c1ddbdSJiaqi Yan if (!bytes || !n) 309*38c1ddbdSJiaqi Yan break; 310*38c1ddbdSJiaqi Yan offset += n; 311*38c1ddbdSJiaqi Yan if (offset == PAGE_SIZE) { 312*38c1ddbdSJiaqi Yan page++; 313*38c1ddbdSJiaqi Yan offset = 0; 314*38c1ddbdSJiaqi Yan } 315*38c1ddbdSJiaqi Yan } 316*38c1ddbdSJiaqi Yan 317*38c1ddbdSJiaqi Yan return res; 318*38c1ddbdSJiaqi Yan } 319*38c1ddbdSJiaqi Yan 320*38c1ddbdSJiaqi Yan /* 321e63e1e5aSBadari Pulavarty * Support for read() - Find the page attached to f_mapping and copy out the 322445c8098SMiaohe Lin * data. This provides functionality similar to filemap_read(). 323e63e1e5aSBadari Pulavarty */ 32434d0640eSAl Viro static ssize_t hugetlbfs_read_iter(struct kiocb *iocb, struct iov_iter *to) 325e63e1e5aSBadari Pulavarty { 32634d0640eSAl Viro struct file *file = iocb->ki_filp; 32734d0640eSAl Viro struct hstate *h = hstate_file(file); 32834d0640eSAl Viro struct address_space *mapping = file->f_mapping; 329e63e1e5aSBadari Pulavarty struct inode *inode = mapping->host; 33034d0640eSAl Viro unsigned long index = iocb->ki_pos >> huge_page_shift(h); 33134d0640eSAl Viro unsigned long offset = iocb->ki_pos & ~huge_page_mask(h); 332e63e1e5aSBadari Pulavarty unsigned long end_index; 333e63e1e5aSBadari Pulavarty loff_t isize; 334e63e1e5aSBadari Pulavarty ssize_t retval = 0; 335e63e1e5aSBadari Pulavarty 33634d0640eSAl Viro while (iov_iter_count(to)) { 337e63e1e5aSBadari Pulavarty struct page *page; 338*38c1ddbdSJiaqi Yan size_t nr, copied, want; 339e63e1e5aSBadari Pulavarty 340e63e1e5aSBadari Pulavarty /* nr is the maximum number of bytes to copy from this page */ 341a5516438SAndi Kleen nr = huge_page_size(h); 342a05b0855SAneesh Kumar K.V isize = i_size_read(inode); 343a05b0855SAneesh Kumar K.V if (!isize) 34434d0640eSAl Viro break; 345a05b0855SAneesh Kumar K.V end_index = (isize - 1) >> huge_page_shift(h); 346e63e1e5aSBadari Pulavarty if (index > end_index) 34734d0640eSAl Viro break; 34834d0640eSAl Viro if (index == end_index) { 349a5516438SAndi Kleen nr = ((isize - 1) & ~huge_page_mask(h)) + 1; 350a05b0855SAneesh Kumar K.V if (nr <= offset) 35134d0640eSAl Viro break; 352e63e1e5aSBadari Pulavarty } 353e63e1e5aSBadari Pulavarty nr = nr - offset; 354e63e1e5aSBadari Pulavarty 355e63e1e5aSBadari Pulavarty /* Find the page */ 356a05b0855SAneesh Kumar K.V page = find_lock_page(mapping, index); 357e63e1e5aSBadari Pulavarty if (unlikely(page == NULL)) { 358e63e1e5aSBadari Pulavarty /* 359e63e1e5aSBadari Pulavarty * We have a HOLE, zero out the user-buffer for the 360e63e1e5aSBadari Pulavarty * length of the hole or request. 361e63e1e5aSBadari Pulavarty */ 36234d0640eSAl Viro copied = iov_iter_zero(nr, to); 363e63e1e5aSBadari Pulavarty } else { 364a05b0855SAneesh Kumar K.V unlock_page(page); 365a05b0855SAneesh Kumar K.V 366*38c1ddbdSJiaqi Yan if (!PageHWPoison(page)) 367*38c1ddbdSJiaqi Yan want = nr; 368*38c1ddbdSJiaqi Yan else { 369*38c1ddbdSJiaqi Yan /* 370*38c1ddbdSJiaqi Yan * Adjust how many bytes safe to read without 371*38c1ddbdSJiaqi Yan * touching the 1st raw HWPOISON subpage after 372*38c1ddbdSJiaqi Yan * offset. 373*38c1ddbdSJiaqi Yan */ 374*38c1ddbdSJiaqi Yan want = adjust_range_hwpoison(page, offset, nr); 375*38c1ddbdSJiaqi Yan if (want == 0) { 3768625147cSJames Houghton put_page(page); 3778625147cSJames Houghton retval = -EIO; 3788625147cSJames Houghton break; 3798625147cSJames Houghton } 380*38c1ddbdSJiaqi Yan } 3818625147cSJames Houghton 382e63e1e5aSBadari Pulavarty /* 383e63e1e5aSBadari Pulavarty * We have the page, copy it to user space buffer. 384e63e1e5aSBadari Pulavarty */ 385*38c1ddbdSJiaqi Yan copied = copy_page_to_iter(page, offset, want, to); 38609cbfeafSKirill A. Shutemov put_page(page); 387e63e1e5aSBadari Pulavarty } 38834d0640eSAl Viro offset += copied; 38934d0640eSAl Viro retval += copied; 39034d0640eSAl Viro if (copied != nr && iov_iter_count(to)) { 39134d0640eSAl Viro if (!retval) 39234d0640eSAl Viro retval = -EFAULT; 393e63e1e5aSBadari Pulavarty break; 394e63e1e5aSBadari Pulavarty } 39534d0640eSAl Viro index += offset >> huge_page_shift(h); 39634d0640eSAl Viro offset &= ~huge_page_mask(h); 39734d0640eSAl Viro } 39834d0640eSAl Viro iocb->ki_pos = ((loff_t)index << huge_page_shift(h)) + offset; 399e63e1e5aSBadari Pulavarty return retval; 400e63e1e5aSBadari Pulavarty } 401e63e1e5aSBadari Pulavarty 402800d15a5SNick Piggin static int hugetlbfs_write_begin(struct file *file, 403800d15a5SNick Piggin struct address_space *mapping, 4049d6b0cd7SMatthew Wilcox (Oracle) loff_t pos, unsigned len, 405800d15a5SNick Piggin struct page **pagep, void **fsdata) 4061da177e4SLinus Torvalds { 4071da177e4SLinus Torvalds return -EINVAL; 4081da177e4SLinus Torvalds } 4091da177e4SLinus Torvalds 410800d15a5SNick Piggin static int hugetlbfs_write_end(struct file *file, struct address_space *mapping, 411800d15a5SNick Piggin loff_t pos, unsigned len, unsigned copied, 412800d15a5SNick Piggin struct page *page, void *fsdata) 4131da177e4SLinus Torvalds { 414800d15a5SNick Piggin BUG(); 4151da177e4SLinus Torvalds return -EINVAL; 4161da177e4SLinus Torvalds } 4171da177e4SLinus Torvalds 418ece62684SSidhartha Kumar static void hugetlb_delete_from_page_cache(struct folio *folio) 4191da177e4SLinus Torvalds { 420ece62684SSidhartha Kumar folio_clear_dirty(folio); 421ece62684SSidhartha Kumar folio_clear_uptodate(folio); 422ece62684SSidhartha Kumar filemap_remove_folio(folio); 4231da177e4SLinus Torvalds } 4241da177e4SLinus Torvalds 425378397ccSMike Kravetz /* 426378397ccSMike Kravetz * Called with i_mmap_rwsem held for inode based vma maps. This makes 427378397ccSMike Kravetz * sure vma (and vm_mm) will not go away. We also hold the hugetlb fault 428378397ccSMike Kravetz * mutex for the page in the mapping. So, we can not race with page being 429378397ccSMike Kravetz * faulted into the vma. 430378397ccSMike Kravetz */ 431378397ccSMike Kravetz static bool hugetlb_vma_maps_page(struct vm_area_struct *vma, 432378397ccSMike Kravetz unsigned long addr, struct page *page) 433378397ccSMike Kravetz { 434378397ccSMike Kravetz pte_t *ptep, pte; 435378397ccSMike Kravetz 4369c67a207SPeter Xu ptep = hugetlb_walk(vma, addr, huge_page_size(hstate_vma(vma))); 437378397ccSMike Kravetz if (!ptep) 438378397ccSMike Kravetz return false; 439378397ccSMike Kravetz 440378397ccSMike Kravetz pte = huge_ptep_get(ptep); 441378397ccSMike Kravetz if (huge_pte_none(pte) || !pte_present(pte)) 442378397ccSMike Kravetz return false; 443378397ccSMike Kravetz 444378397ccSMike Kravetz if (pte_page(pte) == page) 445378397ccSMike Kravetz return true; 446378397ccSMike Kravetz 447378397ccSMike Kravetz return false; 448378397ccSMike Kravetz } 449378397ccSMike Kravetz 450378397ccSMike Kravetz /* 451378397ccSMike Kravetz * Can vma_offset_start/vma_offset_end overflow on 32-bit arches? 452378397ccSMike Kravetz * No, because the interval tree returns us only those vmas 453378397ccSMike Kravetz * which overlap the truncated area starting at pgoff, 454378397ccSMike Kravetz * and no vma on a 32-bit arch can span beyond the 4GB. 455378397ccSMike Kravetz */ 456378397ccSMike Kravetz static unsigned long vma_offset_start(struct vm_area_struct *vma, pgoff_t start) 457378397ccSMike Kravetz { 458243b1f2dSPeter Xu unsigned long offset = 0; 459243b1f2dSPeter Xu 460378397ccSMike Kravetz if (vma->vm_pgoff < start) 461243b1f2dSPeter Xu offset = (start - vma->vm_pgoff) << PAGE_SHIFT; 462243b1f2dSPeter Xu 463243b1f2dSPeter Xu return vma->vm_start + offset; 464378397ccSMike Kravetz } 465378397ccSMike Kravetz 466378397ccSMike Kravetz static unsigned long vma_offset_end(struct vm_area_struct *vma, pgoff_t end) 467378397ccSMike Kravetz { 468378397ccSMike Kravetz unsigned long t_end; 469378397ccSMike Kravetz 470378397ccSMike Kravetz if (!end) 471378397ccSMike Kravetz return vma->vm_end; 472378397ccSMike Kravetz 473378397ccSMike Kravetz t_end = ((end - vma->vm_pgoff) << PAGE_SHIFT) + vma->vm_start; 474378397ccSMike Kravetz if (t_end > vma->vm_end) 475378397ccSMike Kravetz t_end = vma->vm_end; 476378397ccSMike Kravetz return t_end; 477378397ccSMike Kravetz } 478378397ccSMike Kravetz 479378397ccSMike Kravetz /* 480378397ccSMike Kravetz * Called with hugetlb fault mutex held. Therefore, no more mappings to 481378397ccSMike Kravetz * this folio can be created while executing the routine. 482378397ccSMike Kravetz */ 483378397ccSMike Kravetz static void hugetlb_unmap_file_folio(struct hstate *h, 484378397ccSMike Kravetz struct address_space *mapping, 485378397ccSMike Kravetz struct folio *folio, pgoff_t index) 486378397ccSMike Kravetz { 487378397ccSMike Kravetz struct rb_root_cached *root = &mapping->i_mmap; 48840549ba8SMike Kravetz struct hugetlb_vma_lock *vma_lock; 489378397ccSMike Kravetz struct page *page = &folio->page; 490378397ccSMike Kravetz struct vm_area_struct *vma; 491378397ccSMike Kravetz unsigned long v_start; 492378397ccSMike Kravetz unsigned long v_end; 493378397ccSMike Kravetz pgoff_t start, end; 494378397ccSMike Kravetz 495378397ccSMike Kravetz start = index * pages_per_huge_page(h); 496378397ccSMike Kravetz end = (index + 1) * pages_per_huge_page(h); 497378397ccSMike Kravetz 498378397ccSMike Kravetz i_mmap_lock_write(mapping); 49940549ba8SMike Kravetz retry: 50040549ba8SMike Kravetz vma_lock = NULL; 501378397ccSMike Kravetz vma_interval_tree_foreach(vma, root, start, end - 1) { 502378397ccSMike Kravetz v_start = vma_offset_start(vma, start); 503378397ccSMike Kravetz v_end = vma_offset_end(vma, end); 504378397ccSMike Kravetz 505243b1f2dSPeter Xu if (!hugetlb_vma_maps_page(vma, v_start, page)) 506378397ccSMike Kravetz continue; 507378397ccSMike Kravetz 50840549ba8SMike Kravetz if (!hugetlb_vma_trylock_write(vma)) { 50940549ba8SMike Kravetz vma_lock = vma->vm_private_data; 51040549ba8SMike Kravetz /* 51140549ba8SMike Kravetz * If we can not get vma lock, we need to drop 51240549ba8SMike Kravetz * immap_sema and take locks in order. First, 51340549ba8SMike Kravetz * take a ref on the vma_lock structure so that 51440549ba8SMike Kravetz * we can be guaranteed it will not go away when 51540549ba8SMike Kravetz * dropping immap_sema. 51640549ba8SMike Kravetz */ 51740549ba8SMike Kravetz kref_get(&vma_lock->refs); 51840549ba8SMike Kravetz break; 51940549ba8SMike Kravetz } 52040549ba8SMike Kravetz 521243b1f2dSPeter Xu unmap_hugepage_range(vma, v_start, v_end, NULL, 522243b1f2dSPeter Xu ZAP_FLAG_DROP_MARKER); 52340549ba8SMike Kravetz hugetlb_vma_unlock_write(vma); 524378397ccSMike Kravetz } 525378397ccSMike Kravetz 526378397ccSMike Kravetz i_mmap_unlock_write(mapping); 52740549ba8SMike Kravetz 52840549ba8SMike Kravetz if (vma_lock) { 52940549ba8SMike Kravetz /* 53040549ba8SMike Kravetz * Wait on vma_lock. We know it is still valid as we have 53140549ba8SMike Kravetz * a reference. We must 'open code' vma locking as we do 53240549ba8SMike Kravetz * not know if vma_lock is still attached to vma. 53340549ba8SMike Kravetz */ 53440549ba8SMike Kravetz down_write(&vma_lock->rw_sema); 53540549ba8SMike Kravetz i_mmap_lock_write(mapping); 53640549ba8SMike Kravetz 53740549ba8SMike Kravetz vma = vma_lock->vma; 53840549ba8SMike Kravetz if (!vma) { 53940549ba8SMike Kravetz /* 54040549ba8SMike Kravetz * If lock is no longer attached to vma, then just 54140549ba8SMike Kravetz * unlock, drop our reference and retry looking for 54240549ba8SMike Kravetz * other vmas. 54340549ba8SMike Kravetz */ 54440549ba8SMike Kravetz up_write(&vma_lock->rw_sema); 54540549ba8SMike Kravetz kref_put(&vma_lock->refs, hugetlb_vma_lock_release); 54640549ba8SMike Kravetz goto retry; 54740549ba8SMike Kravetz } 54840549ba8SMike Kravetz 54940549ba8SMike Kravetz /* 55040549ba8SMike Kravetz * vma_lock is still attached to vma. Check to see if vma 55140549ba8SMike Kravetz * still maps page and if so, unmap. 55240549ba8SMike Kravetz */ 55340549ba8SMike Kravetz v_start = vma_offset_start(vma, start); 55440549ba8SMike Kravetz v_end = vma_offset_end(vma, end); 555243b1f2dSPeter Xu if (hugetlb_vma_maps_page(vma, v_start, page)) 556243b1f2dSPeter Xu unmap_hugepage_range(vma, v_start, v_end, NULL, 55740549ba8SMike Kravetz ZAP_FLAG_DROP_MARKER); 55840549ba8SMike Kravetz 55940549ba8SMike Kravetz kref_put(&vma_lock->refs, hugetlb_vma_lock_release); 56040549ba8SMike Kravetz hugetlb_vma_unlock_write(vma); 56140549ba8SMike Kravetz 56240549ba8SMike Kravetz goto retry; 56340549ba8SMike Kravetz } 564378397ccSMike Kravetz } 565378397ccSMike Kravetz 5664aae8d1cSMike Kravetz static void 56705e90bd0SPeter Xu hugetlb_vmdelete_list(struct rb_root_cached *root, pgoff_t start, pgoff_t end, 56805e90bd0SPeter Xu zap_flags_t zap_flags) 5694aae8d1cSMike Kravetz { 5704aae8d1cSMike Kravetz struct vm_area_struct *vma; 5714aae8d1cSMike Kravetz 5724aae8d1cSMike Kravetz /* 573d6aba4c8SSean Christopherson * end == 0 indicates that the entire range after start should be 574d6aba4c8SSean Christopherson * unmapped. Note, end is exclusive, whereas the interval tree takes 575d6aba4c8SSean Christopherson * an inclusive "last". 5764aae8d1cSMike Kravetz */ 577d6aba4c8SSean Christopherson vma_interval_tree_foreach(vma, root, start, end ? end - 1 : ULONG_MAX) { 578378397ccSMike Kravetz unsigned long v_start; 5794aae8d1cSMike Kravetz unsigned long v_end; 5804aae8d1cSMike Kravetz 58140549ba8SMike Kravetz if (!hugetlb_vma_trylock_write(vma)) 58240549ba8SMike Kravetz continue; 58340549ba8SMike Kravetz 584378397ccSMike Kravetz v_start = vma_offset_start(vma, start); 585378397ccSMike Kravetz v_end = vma_offset_end(vma, end); 5864aae8d1cSMike Kravetz 587243b1f2dSPeter Xu unmap_hugepage_range(vma, v_start, v_end, NULL, zap_flags); 58840549ba8SMike Kravetz 58940549ba8SMike Kravetz /* 59040549ba8SMike Kravetz * Note that vma lock only exists for shared/non-private 59140549ba8SMike Kravetz * vmas. Therefore, lock is not held when calling 59240549ba8SMike Kravetz * unmap_hugepage_range for private vmas. 59340549ba8SMike Kravetz */ 59440549ba8SMike Kravetz hugetlb_vma_unlock_write(vma); 5954aae8d1cSMike Kravetz } 5964aae8d1cSMike Kravetz } 597b5cec28dSMike Kravetz 598b5cec28dSMike Kravetz /* 599c8627228SMike Kravetz * Called with hugetlb fault mutex held. 600c8627228SMike Kravetz * Returns true if page was actually removed, false otherwise. 601c8627228SMike Kravetz */ 602c8627228SMike Kravetz static bool remove_inode_single_folio(struct hstate *h, struct inode *inode, 603c8627228SMike Kravetz struct address_space *mapping, 604c8627228SMike Kravetz struct folio *folio, pgoff_t index, 605c8627228SMike Kravetz bool truncate_op) 606c8627228SMike Kravetz { 607c8627228SMike Kravetz bool ret = false; 608c8627228SMike Kravetz 609c8627228SMike Kravetz /* 610c8627228SMike Kravetz * If folio is mapped, it was faulted in after being 611c8627228SMike Kravetz * unmapped in caller. Unmap (again) while holding 612c8627228SMike Kravetz * the fault mutex. The mutex will prevent faults 613c8627228SMike Kravetz * until we finish removing the folio. 614c8627228SMike Kravetz */ 615378397ccSMike Kravetz if (unlikely(folio_mapped(folio))) 616378397ccSMike Kravetz hugetlb_unmap_file_folio(h, mapping, folio, index); 617c8627228SMike Kravetz 618c8627228SMike Kravetz folio_lock(folio); 619c8627228SMike Kravetz /* 620c8627228SMike Kravetz * We must remove the folio from page cache before removing 621c8627228SMike Kravetz * the region/ reserve map (hugetlb_unreserve_pages). In 622c8627228SMike Kravetz * rare out of memory conditions, removal of the region/reserve 623c8627228SMike Kravetz * map could fail. Correspondingly, the subpool and global 624c8627228SMike Kravetz * reserve usage count can need to be adjusted. 625c8627228SMike Kravetz */ 626ece62684SSidhartha Kumar VM_BUG_ON_FOLIO(folio_test_hugetlb_restore_reserve(folio), folio); 627ece62684SSidhartha Kumar hugetlb_delete_from_page_cache(folio); 628c8627228SMike Kravetz ret = true; 629c8627228SMike Kravetz if (!truncate_op) { 630c8627228SMike Kravetz if (unlikely(hugetlb_unreserve_pages(inode, index, 631c8627228SMike Kravetz index + 1, 1))) 632c8627228SMike Kravetz hugetlb_fix_reserve_counts(inode); 633c8627228SMike Kravetz } 634c8627228SMike Kravetz 635c8627228SMike Kravetz folio_unlock(folio); 636c8627228SMike Kravetz return ret; 637c8627228SMike Kravetz } 638c8627228SMike Kravetz 639c8627228SMike Kravetz /* 640b5cec28dSMike Kravetz * remove_inode_hugepages handles two distinct cases: truncation and hole 641b5cec28dSMike Kravetz * punch. There are subtle differences in operation for each case. 6424aae8d1cSMike Kravetz * 643b5cec28dSMike Kravetz * truncation is indicated by end of range being LLONG_MAX 644b5cec28dSMike Kravetz * In this case, we first scan the range and release found pages. 6451935ebd3SMiaohe Lin * After releasing pages, hugetlb_unreserve_pages cleans up region/reserve 646c8627228SMike Kravetz * maps and global counts. Page faults can race with truncation. 647c8627228SMike Kravetz * During faults, hugetlb_no_page() checks i_size before page allocation, 648c8627228SMike Kravetz * and again after obtaining page table lock. It will 'back out' 649c8627228SMike Kravetz * allocations in the truncated range. 650b5cec28dSMike Kravetz * hole punch is indicated if end is not LLONG_MAX 651b5cec28dSMike Kravetz * In the hole punch case we scan the range and release found pages. 6521935ebd3SMiaohe Lin * Only when releasing a page is the associated region/reserve map 6531935ebd3SMiaohe Lin * deleted. The region/reserve map for ranges without associated 654e7c58097SMike Kravetz * pages are not modified. Page faults can race with hole punch. 655e7c58097SMike Kravetz * This is indicated if we find a mapped page. 656b5cec28dSMike Kravetz * Note: If the passed end of range value is beyond the end of file, but 657b5cec28dSMike Kravetz * not LLONG_MAX this routine still performs a hole punch operation. 658b5cec28dSMike Kravetz */ 659b5cec28dSMike Kravetz static void remove_inode_hugepages(struct inode *inode, loff_t lstart, 660b5cec28dSMike Kravetz loff_t lend) 6611da177e4SLinus Torvalds { 662a5516438SAndi Kleen struct hstate *h = hstate_inode(inode); 663b45b5bd6SDavid Gibson struct address_space *mapping = &inode->i_data; 664a5516438SAndi Kleen const pgoff_t start = lstart >> huge_page_shift(h); 665b5cec28dSMike Kravetz const pgoff_t end = lend >> huge_page_shift(h); 6661508062eSMatthew Wilcox (Oracle) struct folio_batch fbatch; 667d72dc8a2SJan Kara pgoff_t next, index; 668a43a8c39SChen, Kenneth W int i, freed = 0; 669b5cec28dSMike Kravetz bool truncate_op = (lend == LLONG_MAX); 6701da177e4SLinus Torvalds 6711508062eSMatthew Wilcox (Oracle) folio_batch_init(&fbatch); 6721da177e4SLinus Torvalds next = start; 6731508062eSMatthew Wilcox (Oracle) while (filemap_get_folios(mapping, &next, end - 1, &fbatch)) { 6741508062eSMatthew Wilcox (Oracle) for (i = 0; i < folio_batch_count(&fbatch); ++i) { 6751508062eSMatthew Wilcox (Oracle) struct folio *folio = fbatch.folios[i]; 676d4241a04SMiaohe Lin u32 hash = 0; 677b5cec28dSMike Kravetz 6781508062eSMatthew Wilcox (Oracle) index = folio->index; 679d4241a04SMiaohe Lin hash = hugetlb_fault_mutex_hash(mapping, index); 680e7c58097SMike Kravetz mutex_lock(&hugetlb_fault_mutex_table[hash]); 681e7c58097SMike Kravetz 682b5cec28dSMike Kravetz /* 683c8627228SMike Kravetz * Remove folio that was part of folio_batch. 684b5cec28dSMike Kravetz */ 685c8627228SMike Kravetz if (remove_inode_single_folio(h, inode, mapping, folio, 686c8627228SMike Kravetz index, truncate_op)) 687b5cec28dSMike Kravetz freed++; 688b5cec28dSMike Kravetz 689e7c58097SMike Kravetz mutex_unlock(&hugetlb_fault_mutex_table[hash]); 6901da177e4SLinus Torvalds } 6911508062eSMatthew Wilcox (Oracle) folio_batch_release(&fbatch); 6921817889eSMike Kravetz cond_resched(); 6931da177e4SLinus Torvalds } 694b5cec28dSMike Kravetz 695b5cec28dSMike Kravetz if (truncate_op) 696b5cec28dSMike Kravetz (void)hugetlb_unreserve_pages(inode, start, LONG_MAX, freed); 6971da177e4SLinus Torvalds } 6981da177e4SLinus Torvalds 6992bbbda30SAl Viro static void hugetlbfs_evict_inode(struct inode *inode) 7001da177e4SLinus Torvalds { 7019119a41eSJoonsoo Kim struct resv_map *resv_map; 7029119a41eSJoonsoo Kim 703b5cec28dSMike Kravetz remove_inode_hugepages(inode, 0, LLONG_MAX); 704f27a5136SMike Kravetz 705f27a5136SMike Kravetz /* 706f27a5136SMike Kravetz * Get the resv_map from the address space embedded in the inode. 707f27a5136SMike Kravetz * This is the address space which points to any resv_map allocated 708f27a5136SMike Kravetz * at inode creation time. If this is a device special inode, 709f27a5136SMike Kravetz * i_mapping may not point to the original address space. 710f27a5136SMike Kravetz */ 711f27a5136SMike Kravetz resv_map = (struct resv_map *)(&inode->i_data)->private_data; 712f27a5136SMike Kravetz /* Only regular and link inodes have associated reserve maps */ 7139119a41eSJoonsoo Kim if (resv_map) 7149119a41eSJoonsoo Kim resv_map_release(&resv_map->refs); 715dbd5768fSJan Kara clear_inode(inode); 716149f4211SChristoph Hellwig } 717149f4211SChristoph Hellwig 718e5d319deSMiaohe Lin static void hugetlb_vmtruncate(struct inode *inode, loff_t offset) 7191da177e4SLinus Torvalds { 720856fc295SHugh Dickins pgoff_t pgoff; 7211da177e4SLinus Torvalds struct address_space *mapping = inode->i_mapping; 722a5516438SAndi Kleen struct hstate *h = hstate_inode(inode); 7231da177e4SLinus Torvalds 724a5516438SAndi Kleen BUG_ON(offset & ~huge_page_mask(h)); 725856fc295SHugh Dickins pgoff = offset >> PAGE_SHIFT; 7261da177e4SLinus Torvalds 72787bf91d3SMike Kravetz i_size_write(inode, offset); 728188a3972SMike Kravetz i_mmap_lock_write(mapping); 729f808c13fSDavidlohr Bueso if (!RB_EMPTY_ROOT(&mapping->i_mmap.rb_root)) 73005e90bd0SPeter Xu hugetlb_vmdelete_list(&mapping->i_mmap, pgoff, 0, 73105e90bd0SPeter Xu ZAP_FLAG_DROP_MARKER); 732c86aa7bbSMike Kravetz i_mmap_unlock_write(mapping); 733e7c58097SMike Kravetz remove_inode_hugepages(inode, offset, LLONG_MAX); 7341da177e4SLinus Torvalds } 7351da177e4SLinus Torvalds 73668d32527SMike Kravetz static void hugetlbfs_zero_partial_page(struct hstate *h, 73768d32527SMike Kravetz struct address_space *mapping, 73868d32527SMike Kravetz loff_t start, 73968d32527SMike Kravetz loff_t end) 74068d32527SMike Kravetz { 74168d32527SMike Kravetz pgoff_t idx = start >> huge_page_shift(h); 74268d32527SMike Kravetz struct folio *folio; 74368d32527SMike Kravetz 74468d32527SMike Kravetz folio = filemap_lock_folio(mapping, idx); 74566dabbb6SChristoph Hellwig if (IS_ERR(folio)) 74668d32527SMike Kravetz return; 74768d32527SMike Kravetz 74868d32527SMike Kravetz start = start & ~huge_page_mask(h); 74968d32527SMike Kravetz end = end & ~huge_page_mask(h); 75068d32527SMike Kravetz if (!end) 75168d32527SMike Kravetz end = huge_page_size(h); 75268d32527SMike Kravetz 75368d32527SMike Kravetz folio_zero_segment(folio, (size_t)start, (size_t)end); 75468d32527SMike Kravetz 75568d32527SMike Kravetz folio_unlock(folio); 75668d32527SMike Kravetz folio_put(folio); 75768d32527SMike Kravetz } 75868d32527SMike Kravetz 75970c3547eSMike Kravetz static long hugetlbfs_punch_hole(struct inode *inode, loff_t offset, loff_t len) 76070c3547eSMike Kravetz { 76168d32527SMike Kravetz struct hugetlbfs_inode_info *info = HUGETLBFS_I(inode); 76268d32527SMike Kravetz struct address_space *mapping = inode->i_mapping; 76370c3547eSMike Kravetz struct hstate *h = hstate_inode(inode); 76470c3547eSMike Kravetz loff_t hpage_size = huge_page_size(h); 76570c3547eSMike Kravetz loff_t hole_start, hole_end; 76670c3547eSMike Kravetz 76770c3547eSMike Kravetz /* 76868d32527SMike Kravetz * hole_start and hole_end indicate the full pages within the hole. 76970c3547eSMike Kravetz */ 77070c3547eSMike Kravetz hole_start = round_up(offset, hpage_size); 77170c3547eSMike Kravetz hole_end = round_down(offset + len, hpage_size); 77270c3547eSMike Kravetz 7735955102cSAl Viro inode_lock(inode); 774ff62a342SMarc-André Lureau 775398c0da7SMiaohe Lin /* protected by i_rwsem */ 776ab3948f5SJoel Fernandes (Google) if (info->seals & (F_SEAL_WRITE | F_SEAL_FUTURE_WRITE)) { 777ff62a342SMarc-André Lureau inode_unlock(inode); 778ff62a342SMarc-André Lureau return -EPERM; 779ff62a342SMarc-André Lureau } 780ff62a342SMarc-André Lureau 78170c3547eSMike Kravetz i_mmap_lock_write(mapping); 78268d32527SMike Kravetz 78368d32527SMike Kravetz /* If range starts before first full page, zero partial page. */ 78468d32527SMike Kravetz if (offset < hole_start) 78568d32527SMike Kravetz hugetlbfs_zero_partial_page(h, mapping, 78668d32527SMike Kravetz offset, min(offset + len, hole_start)); 78768d32527SMike Kravetz 78868d32527SMike Kravetz /* Unmap users of full pages in the hole. */ 78968d32527SMike Kravetz if (hole_end > hole_start) { 790f808c13fSDavidlohr Bueso if (!RB_EMPTY_ROOT(&mapping->i_mmap.rb_root)) 79170c3547eSMike Kravetz hugetlb_vmdelete_list(&mapping->i_mmap, 79270c3547eSMike Kravetz hole_start >> PAGE_SHIFT, 79305e90bd0SPeter Xu hole_end >> PAGE_SHIFT, 0); 79470c3547eSMike Kravetz } 79570c3547eSMike Kravetz 79668d32527SMike Kravetz /* If range extends beyond last full page, zero partial page. */ 79768d32527SMike Kravetz if ((offset + len) > hole_end && (offset + len) > hole_start) 79868d32527SMike Kravetz hugetlbfs_zero_partial_page(h, mapping, 79968d32527SMike Kravetz hole_end, offset + len); 80068d32527SMike Kravetz 80168d32527SMike Kravetz i_mmap_unlock_write(mapping); 80268d32527SMike Kravetz 80368d32527SMike Kravetz /* Remove full pages from the file. */ 80468d32527SMike Kravetz if (hole_end > hole_start) 80568d32527SMike Kravetz remove_inode_hugepages(inode, hole_start, hole_end); 80668d32527SMike Kravetz 80768d32527SMike Kravetz inode_unlock(inode); 80868d32527SMike Kravetz 80970c3547eSMike Kravetz return 0; 81070c3547eSMike Kravetz } 81170c3547eSMike Kravetz 81270c3547eSMike Kravetz static long hugetlbfs_fallocate(struct file *file, int mode, loff_t offset, 81370c3547eSMike Kravetz loff_t len) 81470c3547eSMike Kravetz { 81570c3547eSMike Kravetz struct inode *inode = file_inode(file); 816ff62a342SMarc-André Lureau struct hugetlbfs_inode_info *info = HUGETLBFS_I(inode); 81770c3547eSMike Kravetz struct address_space *mapping = inode->i_mapping; 81870c3547eSMike Kravetz struct hstate *h = hstate_inode(inode); 81970c3547eSMike Kravetz struct vm_area_struct pseudo_vma; 82070c3547eSMike Kravetz struct mm_struct *mm = current->mm; 82170c3547eSMike Kravetz loff_t hpage_size = huge_page_size(h); 82270c3547eSMike Kravetz unsigned long hpage_shift = huge_page_shift(h); 82370c3547eSMike Kravetz pgoff_t start, index, end; 82470c3547eSMike Kravetz int error; 82570c3547eSMike Kravetz u32 hash; 82670c3547eSMike Kravetz 82770c3547eSMike Kravetz if (mode & ~(FALLOC_FL_KEEP_SIZE | FALLOC_FL_PUNCH_HOLE)) 82870c3547eSMike Kravetz return -EOPNOTSUPP; 82970c3547eSMike Kravetz 83070c3547eSMike Kravetz if (mode & FALLOC_FL_PUNCH_HOLE) 83170c3547eSMike Kravetz return hugetlbfs_punch_hole(inode, offset, len); 83270c3547eSMike Kravetz 83370c3547eSMike Kravetz /* 83470c3547eSMike Kravetz * Default preallocate case. 83570c3547eSMike Kravetz * For this range, start is rounded down and end is rounded up 83670c3547eSMike Kravetz * as well as being converted to page offsets. 83770c3547eSMike Kravetz */ 83870c3547eSMike Kravetz start = offset >> hpage_shift; 83970c3547eSMike Kravetz end = (offset + len + hpage_size - 1) >> hpage_shift; 84070c3547eSMike Kravetz 8415955102cSAl Viro inode_lock(inode); 84270c3547eSMike Kravetz 84370c3547eSMike Kravetz /* We need to check rlimit even when FALLOC_FL_KEEP_SIZE */ 84470c3547eSMike Kravetz error = inode_newsize_ok(inode, offset + len); 84570c3547eSMike Kravetz if (error) 84670c3547eSMike Kravetz goto out; 84770c3547eSMike Kravetz 848ff62a342SMarc-André Lureau if ((info->seals & F_SEAL_GROW) && offset + len > inode->i_size) { 849ff62a342SMarc-André Lureau error = -EPERM; 850ff62a342SMarc-André Lureau goto out; 851ff62a342SMarc-André Lureau } 852ff62a342SMarc-André Lureau 85370c3547eSMike Kravetz /* 85470c3547eSMike Kravetz * Initialize a pseudo vma as this is required by the huge page 85570c3547eSMike Kravetz * allocation routines. If NUMA is configured, use page index 85670c3547eSMike Kravetz * as input to create an allocation policy. 85770c3547eSMike Kravetz */ 8582c4541e2SKirill A. Shutemov vma_init(&pseudo_vma, mm); 8591c71222eSSuren Baghdasaryan vm_flags_init(&pseudo_vma, VM_HUGETLB | VM_MAYSHARE | VM_SHARED); 86070c3547eSMike Kravetz pseudo_vma.vm_file = file; 86170c3547eSMike Kravetz 86270c3547eSMike Kravetz for (index = start; index < end; index++) { 86370c3547eSMike Kravetz /* 86470c3547eSMike Kravetz * This is supposed to be the vaddr where the page is being 86570c3547eSMike Kravetz * faulted in, but we have no vaddr here. 86670c3547eSMike Kravetz */ 867d0ce0e47SSidhartha Kumar struct folio *folio; 86870c3547eSMike Kravetz unsigned long addr; 86970c3547eSMike Kravetz 87070c3547eSMike Kravetz cond_resched(); 87170c3547eSMike Kravetz 87270c3547eSMike Kravetz /* 87370c3547eSMike Kravetz * fallocate(2) manpage permits EINTR; we may have been 87470c3547eSMike Kravetz * interrupted because we are using up too much memory. 87570c3547eSMike Kravetz */ 87670c3547eSMike Kravetz if (signal_pending(current)) { 87770c3547eSMike Kravetz error = -EINTR; 87870c3547eSMike Kravetz break; 87970c3547eSMike Kravetz } 88070c3547eSMike Kravetz 88170c3547eSMike Kravetz /* addr is the offset within the file (zero based) */ 88270c3547eSMike Kravetz addr = index * hpage_size; 88370c3547eSMike Kravetz 884188a3972SMike Kravetz /* mutex taken here, fault path and hole punch */ 885188b04a7SWei Yang hash = hugetlb_fault_mutex_hash(mapping, index); 88670c3547eSMike Kravetz mutex_lock(&hugetlb_fault_mutex_table[hash]); 88770c3547eSMike Kravetz 88870c3547eSMike Kravetz /* See if already present in mapping to avoid alloc/free */ 889fd4aed8dSMike Kravetz folio = filemap_get_folio(mapping, index); 890fd4aed8dSMike Kravetz if (!IS_ERR(folio)) { 891fd4aed8dSMike Kravetz folio_put(folio); 89270c3547eSMike Kravetz mutex_unlock(&hugetlb_fault_mutex_table[hash]); 89370c3547eSMike Kravetz continue; 89470c3547eSMike Kravetz } 89570c3547eSMike Kravetz 89688ce3fefSMiaohe Lin /* 897d0ce0e47SSidhartha Kumar * Allocate folio without setting the avoid_reserve argument. 89888ce3fefSMiaohe Lin * There certainly are no reserves associated with the 89988ce3fefSMiaohe Lin * pseudo_vma. However, there could be shared mappings with 90088ce3fefSMiaohe Lin * reserves for the file at the inode level. If we fallocate 901d0ce0e47SSidhartha Kumar * folios in these areas, we need to consume the reserves 90288ce3fefSMiaohe Lin * to keep reservation accounting consistent. 90388ce3fefSMiaohe Lin */ 904adef0803SAckerley Tng hugetlb_set_vma_policy(&pseudo_vma, inode, index); 905d0ce0e47SSidhartha Kumar folio = alloc_hugetlb_folio(&pseudo_vma, addr, 0); 90670c3547eSMike Kravetz hugetlb_drop_vma_policy(&pseudo_vma); 907d0ce0e47SSidhartha Kumar if (IS_ERR(folio)) { 90870c3547eSMike Kravetz mutex_unlock(&hugetlb_fault_mutex_table[hash]); 909d0ce0e47SSidhartha Kumar error = PTR_ERR(folio); 91070c3547eSMike Kravetz goto out; 91170c3547eSMike Kravetz } 912d0ce0e47SSidhartha Kumar clear_huge_page(&folio->page, addr, pages_per_huge_page(h)); 913d0ce0e47SSidhartha Kumar __folio_mark_uptodate(folio); 9149b91c0e2SSidhartha Kumar error = hugetlb_add_to_page_cache(folio, mapping, index); 91570c3547eSMike Kravetz if (unlikely(error)) { 916d2d7bb44SSidhartha Kumar restore_reserve_on_error(h, &pseudo_vma, addr, folio); 917d0ce0e47SSidhartha Kumar folio_put(folio); 91870c3547eSMike Kravetz mutex_unlock(&hugetlb_fault_mutex_table[hash]); 91970c3547eSMike Kravetz goto out; 92070c3547eSMike Kravetz } 92170c3547eSMike Kravetz 92270c3547eSMike Kravetz mutex_unlock(&hugetlb_fault_mutex_table[hash]); 92370c3547eSMike Kravetz 924d0ce0e47SSidhartha Kumar folio_set_hugetlb_migratable(folio); 92570c3547eSMike Kravetz /* 926d0ce0e47SSidhartha Kumar * folio_unlock because locked by hugetlb_add_to_page_cache() 927d0ce0e47SSidhartha Kumar * folio_put() due to reference from alloc_hugetlb_folio() 92870c3547eSMike Kravetz */ 929d0ce0e47SSidhartha Kumar folio_unlock(folio); 930d0ce0e47SSidhartha Kumar folio_put(folio); 93170c3547eSMike Kravetz } 93270c3547eSMike Kravetz 93370c3547eSMike Kravetz if (!(mode & FALLOC_FL_KEEP_SIZE) && offset + len > inode->i_size) 93470c3547eSMike Kravetz i_size_write(inode, offset + len); 935078cd827SDeepa Dinamani inode->i_ctime = current_time(inode); 93670c3547eSMike Kravetz out: 9375955102cSAl Viro inode_unlock(inode); 93870c3547eSMike Kravetz return error; 93970c3547eSMike Kravetz } 94070c3547eSMike Kravetz 941c1632a0fSChristian Brauner static int hugetlbfs_setattr(struct mnt_idmap *idmap, 942549c7297SChristian Brauner struct dentry *dentry, struct iattr *attr) 9431da177e4SLinus Torvalds { 9442b0143b5SDavid Howells struct inode *inode = d_inode(dentry); 945a5516438SAndi Kleen struct hstate *h = hstate_inode(inode); 9461da177e4SLinus Torvalds int error; 9471da177e4SLinus Torvalds unsigned int ia_valid = attr->ia_valid; 948ff62a342SMarc-André Lureau struct hugetlbfs_inode_info *info = HUGETLBFS_I(inode); 9491da177e4SLinus Torvalds 950c1632a0fSChristian Brauner error = setattr_prepare(&nop_mnt_idmap, dentry, attr); 9511da177e4SLinus Torvalds if (error) 9521025774cSChristoph Hellwig return error; 9531da177e4SLinus Torvalds 9541da177e4SLinus Torvalds if (ia_valid & ATTR_SIZE) { 955ff62a342SMarc-André Lureau loff_t oldsize = inode->i_size; 956ff62a342SMarc-André Lureau loff_t newsize = attr->ia_size; 957ff62a342SMarc-André Lureau 958ff62a342SMarc-André Lureau if (newsize & ~huge_page_mask(h)) 9591025774cSChristoph Hellwig return -EINVAL; 960398c0da7SMiaohe Lin /* protected by i_rwsem */ 961ff62a342SMarc-André Lureau if ((newsize < oldsize && (info->seals & F_SEAL_SHRINK)) || 962ff62a342SMarc-André Lureau (newsize > oldsize && (info->seals & F_SEAL_GROW))) 963ff62a342SMarc-André Lureau return -EPERM; 964e5d319deSMiaohe Lin hugetlb_vmtruncate(inode, newsize); 9651da177e4SLinus Torvalds } 9661da177e4SLinus Torvalds 967c1632a0fSChristian Brauner setattr_copy(&nop_mnt_idmap, inode, attr); 9681025774cSChristoph Hellwig mark_inode_dirty(inode); 9691025774cSChristoph Hellwig return 0; 9701025774cSChristoph Hellwig } 9711025774cSChristoph Hellwig 9727d54fa64SAl Viro static struct inode *hugetlbfs_get_root(struct super_block *sb, 97332021982SDavid Howells struct hugetlbfs_fs_context *ctx) 9741da177e4SLinus Torvalds { 9751da177e4SLinus Torvalds struct inode *inode; 9761da177e4SLinus Torvalds 9771da177e4SLinus Torvalds inode = new_inode(sb); 9781da177e4SLinus Torvalds if (inode) { 97985fe4025SChristoph Hellwig inode->i_ino = get_next_ino(); 98032021982SDavid Howells inode->i_mode = S_IFDIR | ctx->mode; 98132021982SDavid Howells inode->i_uid = ctx->uid; 98232021982SDavid Howells inode->i_gid = ctx->gid; 983078cd827SDeepa Dinamani inode->i_atime = inode->i_mtime = inode->i_ctime = current_time(inode); 9847d54fa64SAl Viro inode->i_op = &hugetlbfs_dir_inode_operations; 9857d54fa64SAl Viro inode->i_fop = &simple_dir_operations; 9867d54fa64SAl Viro /* directory inodes start off with i_nlink == 2 (for "." entry) */ 9877d54fa64SAl Viro inc_nlink(inode); 98865ed7601SAneesh Kumar K.V lockdep_annotate_inode_mutex_key(inode); 9897d54fa64SAl Viro } 9907d54fa64SAl Viro return inode; 9917d54fa64SAl Viro } 9927d54fa64SAl Viro 993b610ded7SMichal Hocko /* 994c8c06efaSDavidlohr Bueso * Hugetlbfs is not reclaimable; therefore its i_mmap_rwsem will never 995b610ded7SMichal Hocko * be taken from reclaim -- unlike regular filesystems. This needs an 99688f306b6SKirill A. Shutemov * annotation because huge_pmd_share() does an allocation under hugetlb's 997c8c06efaSDavidlohr Bueso * i_mmap_rwsem. 998b610ded7SMichal Hocko */ 999c8c06efaSDavidlohr Bueso static struct lock_class_key hugetlbfs_i_mmap_rwsem_key; 1000b610ded7SMichal Hocko 10017d54fa64SAl Viro static struct inode *hugetlbfs_get_inode(struct super_block *sb, 10027d54fa64SAl Viro struct inode *dir, 100318df2252SAl Viro umode_t mode, dev_t dev) 10047d54fa64SAl Viro { 10057d54fa64SAl Viro struct inode *inode; 100658b6e5e8SMike Kravetz struct resv_map *resv_map = NULL; 10079119a41eSJoonsoo Kim 100858b6e5e8SMike Kravetz /* 100958b6e5e8SMike Kravetz * Reserve maps are only needed for inodes that can have associated 101058b6e5e8SMike Kravetz * page allocations. 101158b6e5e8SMike Kravetz */ 101258b6e5e8SMike Kravetz if (S_ISREG(mode) || S_ISLNK(mode)) { 10139119a41eSJoonsoo Kim resv_map = resv_map_alloc(); 10149119a41eSJoonsoo Kim if (!resv_map) 10159119a41eSJoonsoo Kim return NULL; 101658b6e5e8SMike Kravetz } 10177d54fa64SAl Viro 10187d54fa64SAl Viro inode = new_inode(sb); 10197d54fa64SAl Viro if (inode) { 1020ff62a342SMarc-André Lureau struct hugetlbfs_inode_info *info = HUGETLBFS_I(inode); 1021ff62a342SMarc-André Lureau 10227d54fa64SAl Viro inode->i_ino = get_next_ino(); 1023f2d40141SChristian Brauner inode_init_owner(&nop_mnt_idmap, inode, dir, mode); 1024c8c06efaSDavidlohr Bueso lockdep_set_class(&inode->i_mapping->i_mmap_rwsem, 1025c8c06efaSDavidlohr Bueso &hugetlbfs_i_mmap_rwsem_key); 10261da177e4SLinus Torvalds inode->i_mapping->a_ops = &hugetlbfs_aops; 1027078cd827SDeepa Dinamani inode->i_atime = inode->i_mtime = inode->i_ctime = current_time(inode); 10289119a41eSJoonsoo Kim inode->i_mapping->private_data = resv_map; 1029ff62a342SMarc-André Lureau info->seals = F_SEAL_SEAL; 10301da177e4SLinus Torvalds switch (mode & S_IFMT) { 10311da177e4SLinus Torvalds default: 10321da177e4SLinus Torvalds init_special_inode(inode, mode, dev); 10331da177e4SLinus Torvalds break; 10341da177e4SLinus Torvalds case S_IFREG: 10351da177e4SLinus Torvalds inode->i_op = &hugetlbfs_inode_operations; 10361da177e4SLinus Torvalds inode->i_fop = &hugetlbfs_file_operations; 10371da177e4SLinus Torvalds break; 10381da177e4SLinus Torvalds case S_IFDIR: 10391da177e4SLinus Torvalds inode->i_op = &hugetlbfs_dir_inode_operations; 10401da177e4SLinus Torvalds inode->i_fop = &simple_dir_operations; 10411da177e4SLinus Torvalds 10421da177e4SLinus Torvalds /* directory inodes start off with i_nlink == 2 (for "." entry) */ 1043d8c76e6fSDave Hansen inc_nlink(inode); 10441da177e4SLinus Torvalds break; 10451da177e4SLinus Torvalds case S_IFLNK: 10461da177e4SLinus Torvalds inode->i_op = &page_symlink_inode_operations; 104721fc61c7SAl Viro inode_nohighmem(inode); 10481da177e4SLinus Torvalds break; 10491da177e4SLinus Torvalds } 1050e096d0c7SJosh Boyer lockdep_annotate_inode_mutex_key(inode); 105158b6e5e8SMike Kravetz } else { 105258b6e5e8SMike Kravetz if (resv_map) 10539119a41eSJoonsoo Kim kref_put(&resv_map->refs, resv_map_release); 105458b6e5e8SMike Kravetz } 10559119a41eSJoonsoo Kim 10561da177e4SLinus Torvalds return inode; 10571da177e4SLinus Torvalds } 10581da177e4SLinus Torvalds 10591da177e4SLinus Torvalds /* 10601da177e4SLinus Torvalds * File creation. Allocate an inode, and we're done.. 10611da177e4SLinus Torvalds */ 10625ebb29beSChristian Brauner static int hugetlbfs_mknod(struct mnt_idmap *idmap, struct inode *dir, 10631ab5b82fSPiotr Sarna struct dentry *dentry, umode_t mode, dev_t dev) 10641ab5b82fSPiotr Sarna { 106519ee5345SAl Viro struct inode *inode; 106619ee5345SAl Viro 106719ee5345SAl Viro inode = hugetlbfs_get_inode(dir->i_sb, dir, mode, dev); 106819ee5345SAl Viro if (!inode) 106919ee5345SAl Viro return -ENOSPC; 107019ee5345SAl Viro dir->i_ctime = dir->i_mtime = current_time(dir); 107119ee5345SAl Viro d_instantiate(dentry, inode); 107219ee5345SAl Viro dget(dentry);/* Extra count - pin the dentry in core */ 107319ee5345SAl Viro return 0; 10741ab5b82fSPiotr Sarna } 10751ab5b82fSPiotr Sarna 1076c54bd91eSChristian Brauner static int hugetlbfs_mkdir(struct mnt_idmap *idmap, struct inode *dir, 1077549c7297SChristian Brauner struct dentry *dentry, umode_t mode) 10781da177e4SLinus Torvalds { 10795ebb29beSChristian Brauner int retval = hugetlbfs_mknod(&nop_mnt_idmap, dir, dentry, 1080549c7297SChristian Brauner mode | S_IFDIR, 0); 10811da177e4SLinus Torvalds if (!retval) 1082d8c76e6fSDave Hansen inc_nlink(dir); 10831da177e4SLinus Torvalds return retval; 10841da177e4SLinus Torvalds } 10851da177e4SLinus Torvalds 10866c960e68SChristian Brauner static int hugetlbfs_create(struct mnt_idmap *idmap, 1087549c7297SChristian Brauner struct inode *dir, struct dentry *dentry, 1088549c7297SChristian Brauner umode_t mode, bool excl) 10891da177e4SLinus Torvalds { 10905ebb29beSChristian Brauner return hugetlbfs_mknod(&nop_mnt_idmap, dir, dentry, mode | S_IFREG, 0); 10911da177e4SLinus Torvalds } 10921da177e4SLinus Torvalds 1093011e2b71SChristian Brauner static int hugetlbfs_tmpfile(struct mnt_idmap *idmap, 1094863f144fSMiklos Szeredi struct inode *dir, struct file *file, 1095549c7297SChristian Brauner umode_t mode) 10961ab5b82fSPiotr Sarna { 109719ee5345SAl Viro struct inode *inode; 109819ee5345SAl Viro 109919ee5345SAl Viro inode = hugetlbfs_get_inode(dir->i_sb, dir, mode | S_IFREG, 0); 110019ee5345SAl Viro if (!inode) 110119ee5345SAl Viro return -ENOSPC; 110219ee5345SAl Viro dir->i_ctime = dir->i_mtime = current_time(dir); 1103863f144fSMiklos Szeredi d_tmpfile(file, inode); 1104863f144fSMiklos Szeredi return finish_open_simple(file, 0); 11051ab5b82fSPiotr Sarna } 11061ab5b82fSPiotr Sarna 11077a77db95SChristian Brauner static int hugetlbfs_symlink(struct mnt_idmap *idmap, 1108549c7297SChristian Brauner struct inode *dir, struct dentry *dentry, 1109549c7297SChristian Brauner const char *symname) 11101da177e4SLinus Torvalds { 11111da177e4SLinus Torvalds struct inode *inode; 11121da177e4SLinus Torvalds int error = -ENOSPC; 11131da177e4SLinus Torvalds 11147d54fa64SAl Viro inode = hugetlbfs_get_inode(dir->i_sb, dir, S_IFLNK|S_IRWXUGO, 0); 11151da177e4SLinus Torvalds if (inode) { 11161da177e4SLinus Torvalds int l = strlen(symname)+1; 11171da177e4SLinus Torvalds error = page_symlink(inode, symname, l); 11181da177e4SLinus Torvalds if (!error) { 11191da177e4SLinus Torvalds d_instantiate(dentry, inode); 11201da177e4SLinus Torvalds dget(dentry); 11211da177e4SLinus Torvalds } else 11221da177e4SLinus Torvalds iput(inode); 11231da177e4SLinus Torvalds } 1124078cd827SDeepa Dinamani dir->i_ctime = dir->i_mtime = current_time(dir); 11251da177e4SLinus Torvalds 11261da177e4SLinus Torvalds return error; 11271da177e4SLinus Torvalds } 11281da177e4SLinus Torvalds 1129b890ec2aSMatthew Wilcox (Oracle) #ifdef CONFIG_MIGRATION 1130b890ec2aSMatthew Wilcox (Oracle) static int hugetlbfs_migrate_folio(struct address_space *mapping, 1131b890ec2aSMatthew Wilcox (Oracle) struct folio *dst, struct folio *src, 1132a6bc32b8SMel Gorman enum migrate_mode mode) 1133290408d4SNaoya Horiguchi { 1134290408d4SNaoya Horiguchi int rc; 1135290408d4SNaoya Horiguchi 1136b890ec2aSMatthew Wilcox (Oracle) rc = migrate_huge_page_move_mapping(mapping, dst, src); 113778bd5209SRafael Aquini if (rc != MIGRATEPAGE_SUCCESS) 1138290408d4SNaoya Horiguchi return rc; 1139cb6acd01SMike Kravetz 1140149562f7SSidhartha Kumar if (hugetlb_folio_subpool(src)) { 1141149562f7SSidhartha Kumar hugetlb_set_folio_subpool(dst, 1142149562f7SSidhartha Kumar hugetlb_folio_subpool(src)); 1143149562f7SSidhartha Kumar hugetlb_set_folio_subpool(src, NULL); 1144cb6acd01SMike Kravetz } 1145cb6acd01SMike Kravetz 11462916ecc0SJérôme Glisse if (mode != MIGRATE_SYNC_NO_COPY) 1147b890ec2aSMatthew Wilcox (Oracle) folio_migrate_copy(dst, src); 11482916ecc0SJérôme Glisse else 1149b890ec2aSMatthew Wilcox (Oracle) folio_migrate_flags(dst, src); 1150290408d4SNaoya Horiguchi 115178bd5209SRafael Aquini return MIGRATEPAGE_SUCCESS; 1152290408d4SNaoya Horiguchi } 1153b890ec2aSMatthew Wilcox (Oracle) #else 1154b890ec2aSMatthew Wilcox (Oracle) #define hugetlbfs_migrate_folio NULL 1155b890ec2aSMatthew Wilcox (Oracle) #endif 1156290408d4SNaoya Horiguchi 115778bb9203SNaoya Horiguchi static int hugetlbfs_error_remove_page(struct address_space *mapping, 115878bb9203SNaoya Horiguchi struct page *page) 115978bb9203SNaoya Horiguchi { 116078bb9203SNaoya Horiguchi return 0; 116178bb9203SNaoya Horiguchi } 116278bb9203SNaoya Horiguchi 11634a25220dSDavid Howells /* 11644a25220dSDavid Howells * Display the mount options in /proc/mounts. 11654a25220dSDavid Howells */ 11664a25220dSDavid Howells static int hugetlbfs_show_options(struct seq_file *m, struct dentry *root) 11674a25220dSDavid Howells { 11684a25220dSDavid Howells struct hugetlbfs_sb_info *sbinfo = HUGETLBFS_SB(root->d_sb); 11694a25220dSDavid Howells struct hugepage_subpool *spool = sbinfo->spool; 11704a25220dSDavid Howells unsigned long hpage_size = huge_page_size(sbinfo->hstate); 11714a25220dSDavid Howells unsigned hpage_shift = huge_page_shift(sbinfo->hstate); 11724a25220dSDavid Howells char mod; 11734a25220dSDavid Howells 11744a25220dSDavid Howells if (!uid_eq(sbinfo->uid, GLOBAL_ROOT_UID)) 11754a25220dSDavid Howells seq_printf(m, ",uid=%u", 11764a25220dSDavid Howells from_kuid_munged(&init_user_ns, sbinfo->uid)); 11774a25220dSDavid Howells if (!gid_eq(sbinfo->gid, GLOBAL_ROOT_GID)) 11784a25220dSDavid Howells seq_printf(m, ",gid=%u", 11794a25220dSDavid Howells from_kgid_munged(&init_user_ns, sbinfo->gid)); 11804a25220dSDavid Howells if (sbinfo->mode != 0755) 11814a25220dSDavid Howells seq_printf(m, ",mode=%o", sbinfo->mode); 11824a25220dSDavid Howells if (sbinfo->max_inodes != -1) 11834a25220dSDavid Howells seq_printf(m, ",nr_inodes=%lu", sbinfo->max_inodes); 11844a25220dSDavid Howells 11854a25220dSDavid Howells hpage_size /= 1024; 11864a25220dSDavid Howells mod = 'K'; 11874a25220dSDavid Howells if (hpage_size >= 1024) { 11884a25220dSDavid Howells hpage_size /= 1024; 11894a25220dSDavid Howells mod = 'M'; 11904a25220dSDavid Howells } 11914a25220dSDavid Howells seq_printf(m, ",pagesize=%lu%c", hpage_size, mod); 11924a25220dSDavid Howells if (spool) { 11934a25220dSDavid Howells if (spool->max_hpages != -1) 11944a25220dSDavid Howells seq_printf(m, ",size=%llu", 11954a25220dSDavid Howells (unsigned long long)spool->max_hpages << hpage_shift); 11964a25220dSDavid Howells if (spool->min_hpages != -1) 11974a25220dSDavid Howells seq_printf(m, ",min_size=%llu", 11984a25220dSDavid Howells (unsigned long long)spool->min_hpages << hpage_shift); 11994a25220dSDavid Howells } 12004a25220dSDavid Howells return 0; 12014a25220dSDavid Howells } 12024a25220dSDavid Howells 1203726c3342SDavid Howells static int hugetlbfs_statfs(struct dentry *dentry, struct kstatfs *buf) 12041da177e4SLinus Torvalds { 1205726c3342SDavid Howells struct hugetlbfs_sb_info *sbinfo = HUGETLBFS_SB(dentry->d_sb); 12062b0143b5SDavid Howells struct hstate *h = hstate_inode(d_inode(dentry)); 12071da177e4SLinus Torvalds 12081da177e4SLinus Torvalds buf->f_type = HUGETLBFS_MAGIC; 1209a5516438SAndi Kleen buf->f_bsize = huge_page_size(h); 12101da177e4SLinus Torvalds if (sbinfo) { 12111da177e4SLinus Torvalds spin_lock(&sbinfo->stat_lock); 121211680763SMiaohe Lin /* If no limits set, just report 0 or -1 for max/free/used 121374a8a65cSDavid Gibson * blocks, like simple_statfs() */ 121490481622SDavid Gibson if (sbinfo->spool) { 121590481622SDavid Gibson long free_pages; 121690481622SDavid Gibson 12174b25f030SMina Almasry spin_lock_irq(&sbinfo->spool->lock); 121890481622SDavid Gibson buf->f_blocks = sbinfo->spool->max_hpages; 121990481622SDavid Gibson free_pages = sbinfo->spool->max_hpages 122090481622SDavid Gibson - sbinfo->spool->used_hpages; 122190481622SDavid Gibson buf->f_bavail = buf->f_bfree = free_pages; 12224b25f030SMina Almasry spin_unlock_irq(&sbinfo->spool->lock); 12231da177e4SLinus Torvalds buf->f_files = sbinfo->max_inodes; 12241da177e4SLinus Torvalds buf->f_ffree = sbinfo->free_inodes; 122574a8a65cSDavid Gibson } 12261da177e4SLinus Torvalds spin_unlock(&sbinfo->stat_lock); 12271da177e4SLinus Torvalds } 12281da177e4SLinus Torvalds buf->f_namelen = NAME_MAX; 12291da177e4SLinus Torvalds return 0; 12301da177e4SLinus Torvalds } 12311da177e4SLinus Torvalds 12321da177e4SLinus Torvalds static void hugetlbfs_put_super(struct super_block *sb) 12331da177e4SLinus Torvalds { 12341da177e4SLinus Torvalds struct hugetlbfs_sb_info *sbi = HUGETLBFS_SB(sb); 12351da177e4SLinus Torvalds 12361da177e4SLinus Torvalds if (sbi) { 12371da177e4SLinus Torvalds sb->s_fs_info = NULL; 123890481622SDavid Gibson 123990481622SDavid Gibson if (sbi->spool) 124090481622SDavid Gibson hugepage_put_subpool(sbi->spool); 124190481622SDavid Gibson 12421da177e4SLinus Torvalds kfree(sbi); 12431da177e4SLinus Torvalds } 12441da177e4SLinus Torvalds } 12451da177e4SLinus Torvalds 124696527980SChristoph Hellwig static inline int hugetlbfs_dec_free_inodes(struct hugetlbfs_sb_info *sbinfo) 124796527980SChristoph Hellwig { 124896527980SChristoph Hellwig if (sbinfo->free_inodes >= 0) { 124996527980SChristoph Hellwig spin_lock(&sbinfo->stat_lock); 125096527980SChristoph Hellwig if (unlikely(!sbinfo->free_inodes)) { 125196527980SChristoph Hellwig spin_unlock(&sbinfo->stat_lock); 125296527980SChristoph Hellwig return 0; 125396527980SChristoph Hellwig } 125496527980SChristoph Hellwig sbinfo->free_inodes--; 125596527980SChristoph Hellwig spin_unlock(&sbinfo->stat_lock); 125696527980SChristoph Hellwig } 125796527980SChristoph Hellwig 125896527980SChristoph Hellwig return 1; 125996527980SChristoph Hellwig } 126096527980SChristoph Hellwig 126196527980SChristoph Hellwig static void hugetlbfs_inc_free_inodes(struct hugetlbfs_sb_info *sbinfo) 126296527980SChristoph Hellwig { 126396527980SChristoph Hellwig if (sbinfo->free_inodes >= 0) { 126496527980SChristoph Hellwig spin_lock(&sbinfo->stat_lock); 126596527980SChristoph Hellwig sbinfo->free_inodes++; 126696527980SChristoph Hellwig spin_unlock(&sbinfo->stat_lock); 126796527980SChristoph Hellwig } 126896527980SChristoph Hellwig } 126996527980SChristoph Hellwig 127096527980SChristoph Hellwig 1271e18b890bSChristoph Lameter static struct kmem_cache *hugetlbfs_inode_cachep; 12721da177e4SLinus Torvalds 12731da177e4SLinus Torvalds static struct inode *hugetlbfs_alloc_inode(struct super_block *sb) 12741da177e4SLinus Torvalds { 127596527980SChristoph Hellwig struct hugetlbfs_sb_info *sbinfo = HUGETLBFS_SB(sb); 12761da177e4SLinus Torvalds struct hugetlbfs_inode_info *p; 12771da177e4SLinus Torvalds 127896527980SChristoph Hellwig if (unlikely(!hugetlbfs_dec_free_inodes(sbinfo))) 12791da177e4SLinus Torvalds return NULL; 1280fd60b288SMuchun Song p = alloc_inode_sb(sb, hugetlbfs_inode_cachep, GFP_KERNEL); 128196527980SChristoph Hellwig if (unlikely(!p)) { 128296527980SChristoph Hellwig hugetlbfs_inc_free_inodes(sbinfo); 128396527980SChristoph Hellwig return NULL; 12841da177e4SLinus Torvalds } 12854742a35dSMike Kravetz 12864742a35dSMike Kravetz /* 12874742a35dSMike Kravetz * Any time after allocation, hugetlbfs_destroy_inode can be called 12884742a35dSMike Kravetz * for the inode. mpol_free_shared_policy is unconditionally called 12894742a35dSMike Kravetz * as part of hugetlbfs_destroy_inode. So, initialize policy here 12904742a35dSMike Kravetz * in case of a quick call to destroy. 12914742a35dSMike Kravetz * 12924742a35dSMike Kravetz * Note that the policy is initialized even if we are creating a 12934742a35dSMike Kravetz * private inode. This simplifies hugetlbfs_destroy_inode. 12944742a35dSMike Kravetz */ 12954742a35dSMike Kravetz mpol_shared_policy_init(&p->policy, NULL); 12964742a35dSMike Kravetz 129796527980SChristoph Hellwig return &p->vfs_inode; 12981da177e4SLinus Torvalds } 12991da177e4SLinus Torvalds 1300b62de322SAl Viro static void hugetlbfs_free_inode(struct inode *inode) 1301fa0d7e3dSNick Piggin { 1302fa0d7e3dSNick Piggin kmem_cache_free(hugetlbfs_inode_cachep, HUGETLBFS_I(inode)); 1303fa0d7e3dSNick Piggin } 1304fa0d7e3dSNick Piggin 13051da177e4SLinus Torvalds static void hugetlbfs_destroy_inode(struct inode *inode) 13061da177e4SLinus Torvalds { 130796527980SChristoph Hellwig hugetlbfs_inc_free_inodes(HUGETLBFS_SB(inode->i_sb)); 13081da177e4SLinus Torvalds mpol_free_shared_policy(&HUGETLBFS_I(inode)->policy); 13091da177e4SLinus Torvalds } 13101da177e4SLinus Torvalds 1311f5e54d6eSChristoph Hellwig static const struct address_space_operations hugetlbfs_aops = { 1312800d15a5SNick Piggin .write_begin = hugetlbfs_write_begin, 1313800d15a5SNick Piggin .write_end = hugetlbfs_write_end, 131446de8b97SMatthew Wilcox (Oracle) .dirty_folio = noop_dirty_folio, 1315b890ec2aSMatthew Wilcox (Oracle) .migrate_folio = hugetlbfs_migrate_folio, 131678bb9203SNaoya Horiguchi .error_remove_page = hugetlbfs_error_remove_page, 13171da177e4SLinus Torvalds }; 13181da177e4SLinus Torvalds 131996527980SChristoph Hellwig 132051cc5068SAlexey Dobriyan static void init_once(void *foo) 132196527980SChristoph Hellwig { 1322dbaf7dc9SLi zeming struct hugetlbfs_inode_info *ei = foo; 132396527980SChristoph Hellwig 132496527980SChristoph Hellwig inode_init_once(&ei->vfs_inode); 132596527980SChristoph Hellwig } 132696527980SChristoph Hellwig 13274b6f5d20SArjan van de Ven const struct file_operations hugetlbfs_file_operations = { 132834d0640eSAl Viro .read_iter = hugetlbfs_read_iter, 13291da177e4SLinus Torvalds .mmap = hugetlbfs_file_mmap, 13301b061d92SChristoph Hellwig .fsync = noop_fsync, 13311da177e4SLinus Torvalds .get_unmapped_area = hugetlb_get_unmapped_area, 13326038f373SArnd Bergmann .llseek = default_llseek, 133370c3547eSMike Kravetz .fallocate = hugetlbfs_fallocate, 13341da177e4SLinus Torvalds }; 13351da177e4SLinus Torvalds 133692e1d5beSArjan van de Ven static const struct inode_operations hugetlbfs_dir_inode_operations = { 13371da177e4SLinus Torvalds .create = hugetlbfs_create, 13381da177e4SLinus Torvalds .lookup = simple_lookup, 13391da177e4SLinus Torvalds .link = simple_link, 13401da177e4SLinus Torvalds .unlink = simple_unlink, 13411da177e4SLinus Torvalds .symlink = hugetlbfs_symlink, 13421da177e4SLinus Torvalds .mkdir = hugetlbfs_mkdir, 13431da177e4SLinus Torvalds .rmdir = simple_rmdir, 13441da177e4SLinus Torvalds .mknod = hugetlbfs_mknod, 13451da177e4SLinus Torvalds .rename = simple_rename, 13461da177e4SLinus Torvalds .setattr = hugetlbfs_setattr, 13471ab5b82fSPiotr Sarna .tmpfile = hugetlbfs_tmpfile, 13481da177e4SLinus Torvalds }; 13491da177e4SLinus Torvalds 135092e1d5beSArjan van de Ven static const struct inode_operations hugetlbfs_inode_operations = { 13511da177e4SLinus Torvalds .setattr = hugetlbfs_setattr, 13521da177e4SLinus Torvalds }; 13531da177e4SLinus Torvalds 1354ee9b6d61SJosef 'Jeff' Sipek static const struct super_operations hugetlbfs_ops = { 13551da177e4SLinus Torvalds .alloc_inode = hugetlbfs_alloc_inode, 1356b62de322SAl Viro .free_inode = hugetlbfs_free_inode, 13571da177e4SLinus Torvalds .destroy_inode = hugetlbfs_destroy_inode, 13582bbbda30SAl Viro .evict_inode = hugetlbfs_evict_inode, 13591da177e4SLinus Torvalds .statfs = hugetlbfs_statfs, 13601da177e4SLinus Torvalds .put_super = hugetlbfs_put_super, 13614a25220dSDavid Howells .show_options = hugetlbfs_show_options, 13621da177e4SLinus Torvalds }; 13631da177e4SLinus Torvalds 13647ca02d0aSMike Kravetz /* 13657ca02d0aSMike Kravetz * Convert size option passed from command line to number of huge pages 13667ca02d0aSMike Kravetz * in the pool specified by hstate. Size option could be in bytes 13677ca02d0aSMike Kravetz * (val_type == SIZE_STD) or percentage of the pool (val_type == SIZE_PERCENT). 13687ca02d0aSMike Kravetz */ 13694a25220dSDavid Howells static long 13707ca02d0aSMike Kravetz hugetlbfs_size_to_hpages(struct hstate *h, unsigned long long size_opt, 13714a25220dSDavid Howells enum hugetlbfs_size_type val_type) 13727ca02d0aSMike Kravetz { 13737ca02d0aSMike Kravetz if (val_type == NO_SIZE) 13747ca02d0aSMike Kravetz return -1; 13757ca02d0aSMike Kravetz 13767ca02d0aSMike Kravetz if (val_type == SIZE_PERCENT) { 13777ca02d0aSMike Kravetz size_opt <<= huge_page_shift(h); 13787ca02d0aSMike Kravetz size_opt *= h->max_huge_pages; 13797ca02d0aSMike Kravetz do_div(size_opt, 100); 13807ca02d0aSMike Kravetz } 13817ca02d0aSMike Kravetz 13827ca02d0aSMike Kravetz size_opt >>= huge_page_shift(h); 13837ca02d0aSMike Kravetz return size_opt; 13847ca02d0aSMike Kravetz } 13857ca02d0aSMike Kravetz 138632021982SDavid Howells /* 138732021982SDavid Howells * Parse one mount parameter. 138832021982SDavid Howells */ 138932021982SDavid Howells static int hugetlbfs_parse_param(struct fs_context *fc, struct fs_parameter *param) 13901da177e4SLinus Torvalds { 139132021982SDavid Howells struct hugetlbfs_fs_context *ctx = fc->fs_private; 139232021982SDavid Howells struct fs_parse_result result; 139332021982SDavid Howells char *rest; 139432021982SDavid Howells unsigned long ps; 139532021982SDavid Howells int opt; 13961da177e4SLinus Torvalds 1397d7167b14SAl Viro opt = fs_parse(fc, hugetlb_fs_parameters, param, &result); 139832021982SDavid Howells if (opt < 0) 139932021982SDavid Howells return opt; 140032021982SDavid Howells 140132021982SDavid Howells switch (opt) { 140232021982SDavid Howells case Opt_uid: 140332021982SDavid Howells ctx->uid = make_kuid(current_user_ns(), result.uint_32); 140432021982SDavid Howells if (!uid_valid(ctx->uid)) 140532021982SDavid Howells goto bad_val; 14061da177e4SLinus Torvalds return 0; 14071da177e4SLinus Torvalds 1408e73a75faSRandy Dunlap case Opt_gid: 140932021982SDavid Howells ctx->gid = make_kgid(current_user_ns(), result.uint_32); 141032021982SDavid Howells if (!gid_valid(ctx->gid)) 1411e73a75faSRandy Dunlap goto bad_val; 141232021982SDavid Howells return 0; 1413e73a75faSRandy Dunlap 1414e73a75faSRandy Dunlap case Opt_mode: 141532021982SDavid Howells ctx->mode = result.uint_32 & 01777U; 141632021982SDavid Howells return 0; 1417e73a75faSRandy Dunlap 141832021982SDavid Howells case Opt_size: 1419e73a75faSRandy Dunlap /* memparse() will accept a K/M/G without a digit */ 142026215b7eSHawkins Jiawei if (!param->string || !isdigit(param->string[0])) 1421e73a75faSRandy Dunlap goto bad_val; 142232021982SDavid Howells ctx->max_size_opt = memparse(param->string, &rest); 142332021982SDavid Howells ctx->max_val_type = SIZE_STD; 1424a137e1ccSAndi Kleen if (*rest == '%') 142532021982SDavid Howells ctx->max_val_type = SIZE_PERCENT; 142632021982SDavid Howells return 0; 14271da177e4SLinus Torvalds 1428e73a75faSRandy Dunlap case Opt_nr_inodes: 1429e73a75faSRandy Dunlap /* memparse() will accept a K/M/G without a digit */ 143026215b7eSHawkins Jiawei if (!param->string || !isdigit(param->string[0])) 1431e73a75faSRandy Dunlap goto bad_val; 143232021982SDavid Howells ctx->nr_inodes = memparse(param->string, &rest); 143332021982SDavid Howells return 0; 1434e73a75faSRandy Dunlap 143532021982SDavid Howells case Opt_pagesize: 143632021982SDavid Howells ps = memparse(param->string, &rest); 143732021982SDavid Howells ctx->hstate = size_to_hstate(ps); 143832021982SDavid Howells if (!ctx->hstate) { 1439d0036517SMiaohe Lin pr_err("Unsupported page size %lu MB\n", ps / SZ_1M); 1440a137e1ccSAndi Kleen return -EINVAL; 1441a137e1ccSAndi Kleen } 144232021982SDavid Howells return 0; 1443a137e1ccSAndi Kleen 144432021982SDavid Howells case Opt_min_size: 14457ca02d0aSMike Kravetz /* memparse() will accept a K/M/G without a digit */ 144626215b7eSHawkins Jiawei if (!param->string || !isdigit(param->string[0])) 14477ca02d0aSMike Kravetz goto bad_val; 144832021982SDavid Howells ctx->min_size_opt = memparse(param->string, &rest); 144932021982SDavid Howells ctx->min_val_type = SIZE_STD; 14507ca02d0aSMike Kravetz if (*rest == '%') 145132021982SDavid Howells ctx->min_val_type = SIZE_PERCENT; 145232021982SDavid Howells return 0; 14537ca02d0aSMike Kravetz 1454e73a75faSRandy Dunlap default: 1455b4c07bceSLee Schermerhorn return -EINVAL; 1456e73a75faSRandy Dunlap } 145732021982SDavid Howells 145832021982SDavid Howells bad_val: 1459b5db30cfSAl Viro return invalfc(fc, "Bad value '%s' for mount option '%s'\n", 146032021982SDavid Howells param->string, param->key); 14611da177e4SLinus Torvalds } 1462a137e1ccSAndi Kleen 14637ca02d0aSMike Kravetz /* 146432021982SDavid Howells * Validate the parsed options. 146532021982SDavid Howells */ 146632021982SDavid Howells static int hugetlbfs_validate(struct fs_context *fc) 146732021982SDavid Howells { 146832021982SDavid Howells struct hugetlbfs_fs_context *ctx = fc->fs_private; 146932021982SDavid Howells 147032021982SDavid Howells /* 14717ca02d0aSMike Kravetz * Use huge page pool size (in hstate) to convert the size 14727ca02d0aSMike Kravetz * options to number of huge pages. If NO_SIZE, -1 is returned. 14737ca02d0aSMike Kravetz */ 147432021982SDavid Howells ctx->max_hpages = hugetlbfs_size_to_hpages(ctx->hstate, 147532021982SDavid Howells ctx->max_size_opt, 147632021982SDavid Howells ctx->max_val_type); 147732021982SDavid Howells ctx->min_hpages = hugetlbfs_size_to_hpages(ctx->hstate, 147832021982SDavid Howells ctx->min_size_opt, 147932021982SDavid Howells ctx->min_val_type); 14807ca02d0aSMike Kravetz 14817ca02d0aSMike Kravetz /* 14827ca02d0aSMike Kravetz * If max_size was specified, then min_size must be smaller 14837ca02d0aSMike Kravetz */ 148432021982SDavid Howells if (ctx->max_val_type > NO_SIZE && 148532021982SDavid Howells ctx->min_hpages > ctx->max_hpages) { 148632021982SDavid Howells pr_err("Minimum size can not be greater than maximum size\n"); 14877ca02d0aSMike Kravetz return -EINVAL; 1488a137e1ccSAndi Kleen } 1489a137e1ccSAndi Kleen 14901da177e4SLinus Torvalds return 0; 14911da177e4SLinus Torvalds } 14921da177e4SLinus Torvalds 14931da177e4SLinus Torvalds static int 149432021982SDavid Howells hugetlbfs_fill_super(struct super_block *sb, struct fs_context *fc) 14951da177e4SLinus Torvalds { 149632021982SDavid Howells struct hugetlbfs_fs_context *ctx = fc->fs_private; 14971da177e4SLinus Torvalds struct hugetlbfs_sb_info *sbinfo; 14981da177e4SLinus Torvalds 14991da177e4SLinus Torvalds sbinfo = kmalloc(sizeof(struct hugetlbfs_sb_info), GFP_KERNEL); 15001da177e4SLinus Torvalds if (!sbinfo) 15011da177e4SLinus Torvalds return -ENOMEM; 15021da177e4SLinus Torvalds sb->s_fs_info = sbinfo; 15031da177e4SLinus Torvalds spin_lock_init(&sbinfo->stat_lock); 150432021982SDavid Howells sbinfo->hstate = ctx->hstate; 150532021982SDavid Howells sbinfo->max_inodes = ctx->nr_inodes; 150632021982SDavid Howells sbinfo->free_inodes = ctx->nr_inodes; 150790481622SDavid Gibson sbinfo->spool = NULL; 150832021982SDavid Howells sbinfo->uid = ctx->uid; 150932021982SDavid Howells sbinfo->gid = ctx->gid; 151032021982SDavid Howells sbinfo->mode = ctx->mode; 15114a25220dSDavid Howells 15127ca02d0aSMike Kravetz /* 15137ca02d0aSMike Kravetz * Allocate and initialize subpool if maximum or minimum size is 15141935ebd3SMiaohe Lin * specified. Any needed reservations (for minimum size) are taken 1515445c8098SMiaohe Lin * when the subpool is created. 15167ca02d0aSMike Kravetz */ 151732021982SDavid Howells if (ctx->max_hpages != -1 || ctx->min_hpages != -1) { 151832021982SDavid Howells sbinfo->spool = hugepage_new_subpool(ctx->hstate, 151932021982SDavid Howells ctx->max_hpages, 152032021982SDavid Howells ctx->min_hpages); 152190481622SDavid Gibson if (!sbinfo->spool) 152290481622SDavid Gibson goto out_free; 152390481622SDavid Gibson } 15241da177e4SLinus Torvalds sb->s_maxbytes = MAX_LFS_FILESIZE; 152532021982SDavid Howells sb->s_blocksize = huge_page_size(ctx->hstate); 152632021982SDavid Howells sb->s_blocksize_bits = huge_page_shift(ctx->hstate); 15271da177e4SLinus Torvalds sb->s_magic = HUGETLBFS_MAGIC; 15281da177e4SLinus Torvalds sb->s_op = &hugetlbfs_ops; 15291da177e4SLinus Torvalds sb->s_time_gran = 1; 153015568299SMike Kravetz 153115568299SMike Kravetz /* 153215568299SMike Kravetz * Due to the special and limited functionality of hugetlbfs, it does 153315568299SMike Kravetz * not work well as a stacking filesystem. 153415568299SMike Kravetz */ 153515568299SMike Kravetz sb->s_stack_depth = FILESYSTEM_MAX_STACK_DEPTH; 153632021982SDavid Howells sb->s_root = d_make_root(hugetlbfs_get_root(sb, ctx)); 153748fde701SAl Viro if (!sb->s_root) 15381da177e4SLinus Torvalds goto out_free; 15391da177e4SLinus Torvalds return 0; 15401da177e4SLinus Torvalds out_free: 154190481622SDavid Gibson kfree(sbinfo->spool); 15421da177e4SLinus Torvalds kfree(sbinfo); 15431da177e4SLinus Torvalds return -ENOMEM; 15441da177e4SLinus Torvalds } 15451da177e4SLinus Torvalds 154632021982SDavid Howells static int hugetlbfs_get_tree(struct fs_context *fc) 15471da177e4SLinus Torvalds { 154832021982SDavid Howells int err = hugetlbfs_validate(fc); 154932021982SDavid Howells if (err) 155032021982SDavid Howells return err; 15512ac295d4SAl Viro return get_tree_nodev(fc, hugetlbfs_fill_super); 155232021982SDavid Howells } 155332021982SDavid Howells 155432021982SDavid Howells static void hugetlbfs_fs_context_free(struct fs_context *fc) 155532021982SDavid Howells { 155632021982SDavid Howells kfree(fc->fs_private); 155732021982SDavid Howells } 155832021982SDavid Howells 155932021982SDavid Howells static const struct fs_context_operations hugetlbfs_fs_context_ops = { 156032021982SDavid Howells .free = hugetlbfs_fs_context_free, 156132021982SDavid Howells .parse_param = hugetlbfs_parse_param, 156232021982SDavid Howells .get_tree = hugetlbfs_get_tree, 156332021982SDavid Howells }; 156432021982SDavid Howells 156532021982SDavid Howells static int hugetlbfs_init_fs_context(struct fs_context *fc) 156632021982SDavid Howells { 156732021982SDavid Howells struct hugetlbfs_fs_context *ctx; 156832021982SDavid Howells 156932021982SDavid Howells ctx = kzalloc(sizeof(struct hugetlbfs_fs_context), GFP_KERNEL); 157032021982SDavid Howells if (!ctx) 157132021982SDavid Howells return -ENOMEM; 157232021982SDavid Howells 157332021982SDavid Howells ctx->max_hpages = -1; /* No limit on size by default */ 157432021982SDavid Howells ctx->nr_inodes = -1; /* No limit on number of inodes by default */ 157532021982SDavid Howells ctx->uid = current_fsuid(); 157632021982SDavid Howells ctx->gid = current_fsgid(); 157732021982SDavid Howells ctx->mode = 0755; 157832021982SDavid Howells ctx->hstate = &default_hstate; 157932021982SDavid Howells ctx->min_hpages = -1; /* No default minimum size */ 158032021982SDavid Howells ctx->max_val_type = NO_SIZE; 158132021982SDavid Howells ctx->min_val_type = NO_SIZE; 158232021982SDavid Howells fc->fs_private = ctx; 158332021982SDavid Howells fc->ops = &hugetlbfs_fs_context_ops; 158432021982SDavid Howells return 0; 15851da177e4SLinus Torvalds } 15861da177e4SLinus Torvalds 15871da177e4SLinus Torvalds static struct file_system_type hugetlbfs_fs_type = { 15881da177e4SLinus Torvalds .name = "hugetlbfs", 158932021982SDavid Howells .init_fs_context = hugetlbfs_init_fs_context, 1590d7167b14SAl Viro .parameters = hugetlb_fs_parameters, 15911da177e4SLinus Torvalds .kill_sb = kill_litter_super, 15921da177e4SLinus Torvalds }; 15931da177e4SLinus Torvalds 159442d7395fSAndi Kleen static struct vfsmount *hugetlbfs_vfsmount[HUGE_MAX_HSTATE]; 15951da177e4SLinus Torvalds 1596ef1ff6b8SFrom: Mel Gorman static int can_do_hugetlb_shm(void) 15971da177e4SLinus Torvalds { 1598a0eb3a05SEric W. Biederman kgid_t shm_group; 1599a0eb3a05SEric W. Biederman shm_group = make_kgid(&init_user_ns, sysctl_hugetlb_shm_group); 1600a0eb3a05SEric W. Biederman return capable(CAP_IPC_LOCK) || in_group_p(shm_group); 16011da177e4SLinus Torvalds } 16021da177e4SLinus Torvalds 160342d7395fSAndi Kleen static int get_hstate_idx(int page_size_log) 160442d7395fSAndi Kleen { 1605af73e4d9SNaoya Horiguchi struct hstate *h = hstate_sizelog(page_size_log); 160642d7395fSAndi Kleen 160742d7395fSAndi Kleen if (!h) 160842d7395fSAndi Kleen return -1; 160904adbc3fSMiaohe Lin return hstate_index(h); 161042d7395fSAndi Kleen } 161142d7395fSAndi Kleen 1612af73e4d9SNaoya Horiguchi /* 1613af73e4d9SNaoya Horiguchi * Note that size should be aligned to proper hugepage size in caller side, 1614af73e4d9SNaoya Horiguchi * otherwise hugetlb_reserve_pages reserves one less hugepages than intended. 1615af73e4d9SNaoya Horiguchi */ 1616af73e4d9SNaoya Horiguchi struct file *hugetlb_file_setup(const char *name, size_t size, 161783c1fd76Szhangyiru vm_flags_t acctflag, int creat_flags, 161883c1fd76Szhangyiru int page_size_log) 16191da177e4SLinus Torvalds { 16201da177e4SLinus Torvalds struct inode *inode; 1621e68375c8SAl Viro struct vfsmount *mnt; 162242d7395fSAndi Kleen int hstate_idx; 1623e68375c8SAl Viro struct file *file; 162442d7395fSAndi Kleen 162542d7395fSAndi Kleen hstate_idx = get_hstate_idx(page_size_log); 162642d7395fSAndi Kleen if (hstate_idx < 0) 162742d7395fSAndi Kleen return ERR_PTR(-ENODEV); 16281da177e4SLinus Torvalds 1629e68375c8SAl Viro mnt = hugetlbfs_vfsmount[hstate_idx]; 1630e68375c8SAl Viro if (!mnt) 16315bc98594SAkinobu Mita return ERR_PTR(-ENOENT); 16325bc98594SAkinobu Mita 1633ef1ff6b8SFrom: Mel Gorman if (creat_flags == HUGETLB_SHMFS_INODE && !can_do_hugetlb_shm()) { 163483c1fd76Szhangyiru struct ucounts *ucounts = current_ucounts(); 163583c1fd76Szhangyiru 163683c1fd76Szhangyiru if (user_shm_lock(size, ucounts)) { 163783c1fd76Szhangyiru pr_warn_once("%s (%d): Using mlock ulimits for SHM_HUGETLB is obsolete\n", 163821a3c273SDavid Rientjes current->comm, current->pid); 163983c1fd76Szhangyiru user_shm_unlock(size, ucounts); 16402584e517SRavikiran G Thirumalai } 164183c1fd76Szhangyiru return ERR_PTR(-EPERM); 1642353d5c30SHugh Dickins } 16431da177e4SLinus Torvalds 164439b65252SAnatol Pomozov file = ERR_PTR(-ENOSPC); 1645e68375c8SAl Viro inode = hugetlbfs_get_inode(mnt->mnt_sb, NULL, S_IFREG | S_IRWXUGO, 0); 16461da177e4SLinus Torvalds if (!inode) 1647e68375c8SAl Viro goto out; 1648e1832f29SStephen Smalley if (creat_flags == HUGETLB_SHMFS_INODE) 1649e1832f29SStephen Smalley inode->i_flags |= S_PRIVATE; 16501da177e4SLinus Torvalds 16511da177e4SLinus Torvalds inode->i_size = size; 16526d6b77f1SMiklos Szeredi clear_nlink(inode); 1653ce8d2cdfSDave Hansen 165433b8f84aSMike Kravetz if (!hugetlb_reserve_pages(inode, 0, 1655e68375c8SAl Viro size >> huge_page_shift(hstate_inode(inode)), NULL, 1656e68375c8SAl Viro acctflag)) 1657e68375c8SAl Viro file = ERR_PTR(-ENOMEM); 1658e68375c8SAl Viro else 1659e68375c8SAl Viro file = alloc_file_pseudo(inode, mnt, name, O_RDWR, 1660ce8d2cdfSDave Hansen &hugetlbfs_file_operations); 1661e68375c8SAl Viro if (!IS_ERR(file)) 16621da177e4SLinus Torvalds return file; 16631da177e4SLinus Torvalds 1664b45b5bd6SDavid Gibson iput(inode); 1665e68375c8SAl Viro out: 166639b65252SAnatol Pomozov return file; 16671da177e4SLinus Torvalds } 16681da177e4SLinus Torvalds 166932021982SDavid Howells static struct vfsmount *__init mount_one_hugetlbfs(struct hstate *h) 167032021982SDavid Howells { 167132021982SDavid Howells struct fs_context *fc; 167232021982SDavid Howells struct vfsmount *mnt; 167332021982SDavid Howells 167432021982SDavid Howells fc = fs_context_for_mount(&hugetlbfs_fs_type, SB_KERNMOUNT); 167532021982SDavid Howells if (IS_ERR(fc)) { 167632021982SDavid Howells mnt = ERR_CAST(fc); 167732021982SDavid Howells } else { 167832021982SDavid Howells struct hugetlbfs_fs_context *ctx = fc->fs_private; 167932021982SDavid Howells ctx->hstate = h; 168032021982SDavid Howells mnt = fc_mount(fc); 168132021982SDavid Howells put_fs_context(fc); 168232021982SDavid Howells } 168332021982SDavid Howells if (IS_ERR(mnt)) 1684a25fddceSMiaohe Lin pr_err("Cannot mount internal hugetlbfs for page size %luK", 1685d0036517SMiaohe Lin huge_page_size(h) / SZ_1K); 168632021982SDavid Howells return mnt; 168732021982SDavid Howells } 168832021982SDavid Howells 16891da177e4SLinus Torvalds static int __init init_hugetlbfs_fs(void) 16901da177e4SLinus Torvalds { 169132021982SDavid Howells struct vfsmount *mnt; 169242d7395fSAndi Kleen struct hstate *h; 16931da177e4SLinus Torvalds int error; 169442d7395fSAndi Kleen int i; 16951da177e4SLinus Torvalds 1696457c1b27SNishanth Aravamudan if (!hugepages_supported()) { 16979b857d26SAndrew Morton pr_info("disabling because there are no supported hugepage sizes\n"); 1698457c1b27SNishanth Aravamudan return -ENOTSUPP; 1699457c1b27SNishanth Aravamudan } 1700457c1b27SNishanth Aravamudan 1701d1d5e05fSHillf Danton error = -ENOMEM; 17021da177e4SLinus Torvalds hugetlbfs_inode_cachep = kmem_cache_create("hugetlbfs_inode_cache", 17031da177e4SLinus Torvalds sizeof(struct hugetlbfs_inode_info), 17045d097056SVladimir Davydov 0, SLAB_ACCOUNT, init_once); 17051da177e4SLinus Torvalds if (hugetlbfs_inode_cachep == NULL) 17068fc312b3SMike Kravetz goto out; 17071da177e4SLinus Torvalds 17081da177e4SLinus Torvalds error = register_filesystem(&hugetlbfs_fs_type); 17091da177e4SLinus Torvalds if (error) 17108fc312b3SMike Kravetz goto out_free; 17111da177e4SLinus Torvalds 17128fc312b3SMike Kravetz /* default hstate mount is required */ 17133b2275a8SMiaohe Lin mnt = mount_one_hugetlbfs(&default_hstate); 17148fc312b3SMike Kravetz if (IS_ERR(mnt)) { 17158fc312b3SMike Kravetz error = PTR_ERR(mnt); 17168fc312b3SMike Kravetz goto out_unreg; 17178fc312b3SMike Kravetz } 17188fc312b3SMike Kravetz hugetlbfs_vfsmount[default_hstate_idx] = mnt; 17198fc312b3SMike Kravetz 17208fc312b3SMike Kravetz /* other hstates are optional */ 172142d7395fSAndi Kleen i = 0; 172242d7395fSAndi Kleen for_each_hstate(h) { 172315f0ec94SJan Stancek if (i == default_hstate_idx) { 172415f0ec94SJan Stancek i++; 17258fc312b3SMike Kravetz continue; 172615f0ec94SJan Stancek } 17278fc312b3SMike Kravetz 172832021982SDavid Howells mnt = mount_one_hugetlbfs(h); 17298fc312b3SMike Kravetz if (IS_ERR(mnt)) 17308fc312b3SMike Kravetz hugetlbfs_vfsmount[i] = NULL; 17318fc312b3SMike Kravetz else 173232021982SDavid Howells hugetlbfs_vfsmount[i] = mnt; 173342d7395fSAndi Kleen i++; 173442d7395fSAndi Kleen } 173532021982SDavid Howells 173642d7395fSAndi Kleen return 0; 17371da177e4SLinus Torvalds 17388fc312b3SMike Kravetz out_unreg: 17398fc312b3SMike Kravetz (void)unregister_filesystem(&hugetlbfs_fs_type); 17408fc312b3SMike Kravetz out_free: 17411da177e4SLinus Torvalds kmem_cache_destroy(hugetlbfs_inode_cachep); 17428fc312b3SMike Kravetz out: 17431da177e4SLinus Torvalds return error; 17441da177e4SLinus Torvalds } 17453e89e1c5SPaul Gortmaker fs_initcall(init_hugetlbfs_fs) 1746