Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1 | /* |
| 2 | * hugetlbpage-backed filesystem. Based on ramfs. |
| 3 | * |
Nadia Yvette Chambers | 6d49e35 | 2012-12-06 10:39:54 +0100 | [diff] [blame] | 4 | * Nadia Yvette Chambers, 2002 |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 5 | * |
| 6 | * Copyright (C) 2002 Linus Torvalds. |
Paul Gortmaker | 3e89e1c | 2016-01-14 15:21:52 -0800 | [diff] [blame] | 7 | * License: GPL |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 8 | */ |
| 9 | |
Andrew Morton | 9b857d2 | 2014-06-04 16:07:21 -0700 | [diff] [blame] | 10 | #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt |
| 11 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 12 | #include <linux/thread_info.h> |
| 13 | #include <asm/current.h> |
Ingo Molnar | 174cd4b | 2017-02-02 19:15:33 +0100 | [diff] [blame] | 14 | #include <linux/sched/signal.h> /* remove ASAP */ |
Mike Kravetz | 70c3547 | 2015-09-08 15:01:54 -0700 | [diff] [blame] | 15 | #include <linux/falloc.h> |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 16 | #include <linux/fs.h> |
| 17 | #include <linux/mount.h> |
| 18 | #include <linux/file.h> |
Randy Dunlap | e73a75f | 2007-07-15 23:40:52 -0700 | [diff] [blame] | 19 | #include <linux/kernel.h> |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 20 | #include <linux/writeback.h> |
| 21 | #include <linux/pagemap.h> |
| 22 | #include <linux/highmem.h> |
| 23 | #include <linux/init.h> |
| 24 | #include <linux/string.h> |
Randy Dunlap | 16f7e0f | 2006-01-11 12:17:46 -0800 | [diff] [blame] | 25 | #include <linux/capability.h> |
Randy Dunlap | e73a75f | 2007-07-15 23:40:52 -0700 | [diff] [blame] | 26 | #include <linux/ctype.h> |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 27 | #include <linux/backing-dev.h> |
| 28 | #include <linux/hugetlb.h> |
| 29 | #include <linux/pagevec.h> |
Randy Dunlap | e73a75f | 2007-07-15 23:40:52 -0700 | [diff] [blame] | 30 | #include <linux/parser.h> |
Benjamin Herrenschmidt | 036e085 | 2007-05-06 14:50:12 -0700 | [diff] [blame] | 31 | #include <linux/mman.h> |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 32 | #include <linux/slab.h> |
| 33 | #include <linux/dnotify.h> |
| 34 | #include <linux/statfs.h> |
| 35 | #include <linux/security.h> |
Nick Black | 1fd7317d | 2009-09-22 16:43:33 -0700 | [diff] [blame] | 36 | #include <linux/magic.h> |
Naoya Horiguchi | 290408d | 2010-09-08 10:19:35 +0900 | [diff] [blame] | 37 | #include <linux/migrate.h> |
Al Viro | 34d0640 | 2015-04-03 11:31:35 -0400 | [diff] [blame] | 38 | #include <linux/uio.h> |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 39 | |
Linus Torvalds | 7c0f6ba | 2016-12-24 11:46:01 -0800 | [diff] [blame] | 40 | #include <linux/uaccess.h> |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 41 | |
Josef 'Jeff' Sipek | ee9b6d6 | 2007-02-12 00:55:41 -0800 | [diff] [blame] | 42 | static const struct super_operations hugetlbfs_ops; |
Christoph Hellwig | f5e54d6 | 2006-06-28 04:26:44 -0700 | [diff] [blame] | 43 | static const struct address_space_operations hugetlbfs_aops; |
Arjan van de Ven | 4b6f5d2 | 2006-03-28 01:56:42 -0800 | [diff] [blame] | 44 | const struct file_operations hugetlbfs_file_operations; |
Arjan van de Ven | 92e1d5b | 2007-02-12 00:55:39 -0800 | [diff] [blame] | 45 | static const struct inode_operations hugetlbfs_dir_inode_operations; |
| 46 | static const struct inode_operations hugetlbfs_inode_operations; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 47 | |
David Gibson | a1d776e | 2012-03-21 16:34:12 -0700 | [diff] [blame] | 48 | struct hugetlbfs_config { |
David Howells | 4a25220 | 2017-07-05 16:24:18 +0100 | [diff] [blame] | 49 | struct hstate *hstate; |
| 50 | long max_hpages; |
| 51 | long nr_inodes; |
| 52 | long min_hpages; |
| 53 | kuid_t uid; |
| 54 | kgid_t gid; |
| 55 | umode_t mode; |
David Gibson | a1d776e | 2012-03-21 16:34:12 -0700 | [diff] [blame] | 56 | }; |
| 57 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 58 | int sysctl_hugetlb_shm_group; |
| 59 | |
Randy Dunlap | e73a75f | 2007-07-15 23:40:52 -0700 | [diff] [blame] | 60 | enum { |
| 61 | Opt_size, Opt_nr_inodes, |
| 62 | Opt_mode, Opt_uid, Opt_gid, |
Mike Kravetz | 7ca02d0a | 2015-04-15 16:13:42 -0700 | [diff] [blame] | 63 | Opt_pagesize, Opt_min_size, |
Randy Dunlap | e73a75f | 2007-07-15 23:40:52 -0700 | [diff] [blame] | 64 | Opt_err, |
| 65 | }; |
| 66 | |
Steven Whitehouse | a447c09 | 2008-10-13 10:46:57 +0100 | [diff] [blame] | 67 | static const match_table_t tokens = { |
Randy Dunlap | e73a75f | 2007-07-15 23:40:52 -0700 | [diff] [blame] | 68 | {Opt_size, "size=%s"}, |
| 69 | {Opt_nr_inodes, "nr_inodes=%s"}, |
| 70 | {Opt_mode, "mode=%o"}, |
| 71 | {Opt_uid, "uid=%u"}, |
| 72 | {Opt_gid, "gid=%u"}, |
Andi Kleen | a137e1c | 2008-07-23 21:27:43 -0700 | [diff] [blame] | 73 | {Opt_pagesize, "pagesize=%s"}, |
Mike Kravetz | 7ca02d0a | 2015-04-15 16:13:42 -0700 | [diff] [blame] | 74 | {Opt_min_size, "min_size=%s"}, |
Randy Dunlap | e73a75f | 2007-07-15 23:40:52 -0700 | [diff] [blame] | 75 | {Opt_err, NULL}, |
| 76 | }; |
| 77 | |
Mike Kravetz | 70c3547 | 2015-09-08 15:01:54 -0700 | [diff] [blame] | 78 | #ifdef CONFIG_NUMA |
| 79 | static inline void hugetlb_set_vma_policy(struct vm_area_struct *vma, |
| 80 | struct inode *inode, pgoff_t index) |
| 81 | { |
| 82 | vma->vm_policy = mpol_shared_policy_lookup(&HUGETLBFS_I(inode)->policy, |
| 83 | index); |
| 84 | } |
| 85 | |
| 86 | static inline void hugetlb_drop_vma_policy(struct vm_area_struct *vma) |
| 87 | { |
| 88 | mpol_cond_put(vma->vm_policy); |
| 89 | } |
| 90 | #else |
| 91 | static inline void hugetlb_set_vma_policy(struct vm_area_struct *vma, |
| 92 | struct inode *inode, pgoff_t index) |
| 93 | { |
| 94 | } |
| 95 | |
| 96 | static inline void hugetlb_drop_vma_policy(struct vm_area_struct *vma) |
| 97 | { |
| 98 | } |
| 99 | #endif |
| 100 | |
Adam Litke | 2e9b367c | 2005-10-29 18:16:47 -0700 | [diff] [blame] | 101 | static void huge_pagevec_release(struct pagevec *pvec) |
| 102 | { |
| 103 | int i; |
| 104 | |
| 105 | for (i = 0; i < pagevec_count(pvec); ++i) |
| 106 | put_page(pvec->pages[i]); |
| 107 | |
| 108 | pagevec_reinit(pvec); |
| 109 | } |
| 110 | |
Mike Kravetz | 63489f8 | 2018-03-22 16:17:13 -0700 | [diff] [blame] | 111 | /* |
| 112 | * Mask used when checking the page offset value passed in via system |
| 113 | * calls. This value will be converted to a loff_t which is signed. |
| 114 | * Therefore, we want to check the upper PAGE_SHIFT + 1 bits of the |
| 115 | * value. The extra bit (- 1 in the shift value) is to take the sign |
| 116 | * bit into account. |
| 117 | */ |
| 118 | #define PGOFF_LOFFT_MAX \ |
| 119 | (((1UL << (PAGE_SHIFT + 1)) - 1) << (BITS_PER_LONG - (PAGE_SHIFT + 1))) |
| 120 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 121 | static int hugetlbfs_file_mmap(struct file *file, struct vm_area_struct *vma) |
| 122 | { |
Al Viro | 496ad9a | 2013-01-23 17:07:38 -0500 | [diff] [blame] | 123 | struct inode *inode = file_inode(file); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 124 | loff_t len, vma_len; |
| 125 | int ret; |
Andi Kleen | a551643 | 2008-07-23 21:27:41 -0700 | [diff] [blame] | 126 | struct hstate *h = hstate_file(file); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 127 | |
Hugh Dickins | 68589bc | 2006-11-14 02:03:32 -0800 | [diff] [blame] | 128 | /* |
David Gibson | dec4ad8 | 2007-08-30 23:56:40 -0700 | [diff] [blame] | 129 | * vma address alignment (but not the pgoff alignment) has |
| 130 | * already been checked by prepare_hugepage_range. If you add |
| 131 | * any error returns here, do so after setting VM_HUGETLB, so |
| 132 | * is_vm_hugetlb_page tests below unmap_region go the right |
| 133 | * way when do_mmap_pgoff unwinds (may be important on powerpc |
| 134 | * and ia64). |
Hugh Dickins | 68589bc | 2006-11-14 02:03:32 -0800 | [diff] [blame] | 135 | */ |
Naoya Horiguchi | a2fce91 | 2013-04-17 15:58:27 -0700 | [diff] [blame] | 136 | vma->vm_flags |= VM_HUGETLB | VM_DONTEXPAND; |
Hugh Dickins | 68589bc | 2006-11-14 02:03:32 -0800 | [diff] [blame] | 137 | vma->vm_ops = &hugetlb_vm_ops; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 138 | |
Mike Kravetz | 045c7a3 | 2017-04-13 14:56:32 -0700 | [diff] [blame] | 139 | /* |
Mike Kravetz | 63489f8 | 2018-03-22 16:17:13 -0700 | [diff] [blame] | 140 | * page based offset in vm_pgoff could be sufficiently large to |
Mike Kravetz | 5df63c2 | 2018-04-05 16:18:21 -0700 | [diff] [blame] | 141 | * overflow a loff_t when converted to byte offset. This can |
| 142 | * only happen on architectures where sizeof(loff_t) == |
| 143 | * sizeof(unsigned long). So, only check in those instances. |
Mike Kravetz | 045c7a3 | 2017-04-13 14:56:32 -0700 | [diff] [blame] | 144 | */ |
Mike Kravetz | 5df63c2 | 2018-04-05 16:18:21 -0700 | [diff] [blame] | 145 | if (sizeof(unsigned long) == sizeof(loff_t)) { |
| 146 | if (vma->vm_pgoff & PGOFF_LOFFT_MAX) |
| 147 | return -EINVAL; |
| 148 | } |
Mike Kravetz | 045c7a3 | 2017-04-13 14:56:32 -0700 | [diff] [blame] | 149 | |
Mike Kravetz | 63489f8 | 2018-03-22 16:17:13 -0700 | [diff] [blame] | 150 | /* must be huge page aligned */ |
Becky Bruce | 2b37c35 | 2011-07-25 17:11:49 -0700 | [diff] [blame] | 151 | if (vma->vm_pgoff & (~huge_page_mask(h) >> PAGE_SHIFT)) |
David Gibson | dec4ad8 | 2007-08-30 23:56:40 -0700 | [diff] [blame] | 152 | return -EINVAL; |
| 153 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 154 | vma_len = (loff_t)(vma->vm_end - vma->vm_start); |
Mike Kravetz | 045c7a3 | 2017-04-13 14:56:32 -0700 | [diff] [blame] | 155 | len = vma_len + ((loff_t)vma->vm_pgoff << PAGE_SHIFT); |
| 156 | /* check for overflow */ |
| 157 | if (len < vma_len) |
| 158 | return -EINVAL; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 159 | |
Al Viro | 5955102 | 2016-01-22 15:40:57 -0500 | [diff] [blame] | 160 | inode_lock(inode); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 161 | file_accessed(file); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 162 | |
| 163 | ret = -ENOMEM; |
Mel Gorman | a1e7877 | 2008-07-23 21:27:23 -0700 | [diff] [blame] | 164 | if (hugetlb_reserve_pages(inode, |
Andi Kleen | a551643 | 2008-07-23 21:27:41 -0700 | [diff] [blame] | 165 | vma->vm_pgoff >> huge_page_order(h), |
Mel Gorman | 5a6fe12 | 2009-02-10 14:02:27 +0000 | [diff] [blame] | 166 | len >> huge_page_shift(h), vma, |
| 167 | vma->vm_flags)) |
Chen, Kenneth W | a43a8c3 | 2006-06-23 02:03:15 -0700 | [diff] [blame] | 168 | goto out; |
David Gibson | b45b5bd | 2006-03-22 00:08:55 -0800 | [diff] [blame] | 169 | |
Adam Litke | 4c88726 | 2005-10-29 18:16:46 -0700 | [diff] [blame] | 170 | ret = 0; |
Zhang, Yanmin | b6174df | 2006-07-10 04:44:49 -0700 | [diff] [blame] | 171 | if (vma->vm_flags & VM_WRITE && inode->i_size < len) |
Mike Kravetz | 045c7a3 | 2017-04-13 14:56:32 -0700 | [diff] [blame] | 172 | i_size_write(inode, len); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 173 | out: |
Al Viro | 5955102 | 2016-01-22 15:40:57 -0500 | [diff] [blame] | 174 | inode_unlock(inode); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 175 | |
| 176 | return ret; |
| 177 | } |
| 178 | |
| 179 | /* |
Hugh Dickins | 508034a | 2005-10-29 18:16:30 -0700 | [diff] [blame] | 180 | * Called under down_write(mmap_sem). |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 181 | */ |
| 182 | |
Adrian Bunk | d2ba27e8 | 2007-05-06 14:49:00 -0700 | [diff] [blame] | 183 | #ifndef HAVE_ARCH_HUGETLB_UNMAPPED_AREA |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 184 | static unsigned long |
| 185 | hugetlb_get_unmapped_area(struct file *file, unsigned long addr, |
| 186 | unsigned long len, unsigned long pgoff, unsigned long flags) |
| 187 | { |
| 188 | struct mm_struct *mm = current->mm; |
| 189 | struct vm_area_struct *vma; |
Andi Kleen | a551643 | 2008-07-23 21:27:41 -0700 | [diff] [blame] | 190 | struct hstate *h = hstate_file(file); |
Michel Lespinasse | 0865935 | 2012-12-11 16:02:00 -0800 | [diff] [blame] | 191 | struct vm_unmapped_area_info info; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 192 | |
Andi Kleen | a551643 | 2008-07-23 21:27:41 -0700 | [diff] [blame] | 193 | if (len & ~huge_page_mask(h)) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 194 | return -EINVAL; |
| 195 | if (len > TASK_SIZE) |
| 196 | return -ENOMEM; |
| 197 | |
Benjamin Herrenschmidt | 036e085 | 2007-05-06 14:50:12 -0700 | [diff] [blame] | 198 | if (flags & MAP_FIXED) { |
Andi Kleen | a551643 | 2008-07-23 21:27:41 -0700 | [diff] [blame] | 199 | if (prepare_hugepage_range(file, addr, len)) |
Benjamin Herrenschmidt | 036e085 | 2007-05-06 14:50:12 -0700 | [diff] [blame] | 200 | return -EINVAL; |
| 201 | return addr; |
| 202 | } |
| 203 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 204 | if (addr) { |
Andi Kleen | a551643 | 2008-07-23 21:27:41 -0700 | [diff] [blame] | 205 | addr = ALIGN(addr, huge_page_size(h)); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 206 | vma = find_vma(mm, addr); |
| 207 | if (TASK_SIZE - len >= addr && |
Hugh Dickins | 1be7107 | 2017-06-19 04:03:24 -0700 | [diff] [blame] | 208 | (!vma || addr + len <= vm_start_gap(vma))) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 209 | return addr; |
| 210 | } |
| 211 | |
Michel Lespinasse | 0865935 | 2012-12-11 16:02:00 -0800 | [diff] [blame] | 212 | info.flags = 0; |
| 213 | info.length = len; |
| 214 | info.low_limit = TASK_UNMAPPED_BASE; |
| 215 | info.high_limit = TASK_SIZE; |
| 216 | info.align_mask = PAGE_MASK & ~huge_page_mask(h); |
| 217 | info.align_offset = 0; |
| 218 | return vm_unmapped_area(&info); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 219 | } |
| 220 | #endif |
| 221 | |
Al Viro | 34d0640 | 2015-04-03 11:31:35 -0400 | [diff] [blame] | 222 | static size_t |
Badari Pulavarty | e63e1e5 | 2007-10-16 01:26:22 -0700 | [diff] [blame] | 223 | hugetlbfs_read_actor(struct page *page, unsigned long offset, |
Al Viro | 34d0640 | 2015-04-03 11:31:35 -0400 | [diff] [blame] | 224 | struct iov_iter *to, unsigned long size) |
Badari Pulavarty | e63e1e5 | 2007-10-16 01:26:22 -0700 | [diff] [blame] | 225 | { |
Al Viro | 34d0640 | 2015-04-03 11:31:35 -0400 | [diff] [blame] | 226 | size_t copied = 0; |
Badari Pulavarty | e63e1e5 | 2007-10-16 01:26:22 -0700 | [diff] [blame] | 227 | int i, chunksize; |
| 228 | |
Badari Pulavarty | e63e1e5 | 2007-10-16 01:26:22 -0700 | [diff] [blame] | 229 | /* Find which 4k chunk and offset with in that chunk */ |
Kirill A. Shutemov | 09cbfea | 2016-04-01 15:29:47 +0300 | [diff] [blame] | 230 | i = offset >> PAGE_SHIFT; |
| 231 | offset = offset & ~PAGE_MASK; |
Badari Pulavarty | e63e1e5 | 2007-10-16 01:26:22 -0700 | [diff] [blame] | 232 | |
| 233 | while (size) { |
Al Viro | 34d0640 | 2015-04-03 11:31:35 -0400 | [diff] [blame] | 234 | size_t n; |
Kirill A. Shutemov | 09cbfea | 2016-04-01 15:29:47 +0300 | [diff] [blame] | 235 | chunksize = PAGE_SIZE; |
Badari Pulavarty | e63e1e5 | 2007-10-16 01:26:22 -0700 | [diff] [blame] | 236 | if (offset) |
| 237 | chunksize -= offset; |
| 238 | if (chunksize > size) |
| 239 | chunksize = size; |
Al Viro | 34d0640 | 2015-04-03 11:31:35 -0400 | [diff] [blame] | 240 | n = copy_page_to_iter(&page[i], offset, chunksize, to); |
| 241 | copied += n; |
| 242 | if (n != chunksize) |
| 243 | return copied; |
Badari Pulavarty | e63e1e5 | 2007-10-16 01:26:22 -0700 | [diff] [blame] | 244 | offset = 0; |
| 245 | size -= chunksize; |
Badari Pulavarty | e63e1e5 | 2007-10-16 01:26:22 -0700 | [diff] [blame] | 246 | i++; |
| 247 | } |
Al Viro | 34d0640 | 2015-04-03 11:31:35 -0400 | [diff] [blame] | 248 | return copied; |
Badari Pulavarty | e63e1e5 | 2007-10-16 01:26:22 -0700 | [diff] [blame] | 249 | } |
| 250 | |
| 251 | /* |
| 252 | * Support for read() - Find the page attached to f_mapping and copy out the |
| 253 | * data. Its *very* similar to do_generic_mapping_read(), we can't use that |
Kirill A. Shutemov | ea1754a | 2016-04-01 15:29:48 +0300 | [diff] [blame] | 254 | * since it has PAGE_SIZE assumptions. |
Badari Pulavarty | e63e1e5 | 2007-10-16 01:26:22 -0700 | [diff] [blame] | 255 | */ |
Al Viro | 34d0640 | 2015-04-03 11:31:35 -0400 | [diff] [blame] | 256 | static ssize_t hugetlbfs_read_iter(struct kiocb *iocb, struct iov_iter *to) |
Badari Pulavarty | e63e1e5 | 2007-10-16 01:26:22 -0700 | [diff] [blame] | 257 | { |
Al Viro | 34d0640 | 2015-04-03 11:31:35 -0400 | [diff] [blame] | 258 | struct file *file = iocb->ki_filp; |
| 259 | struct hstate *h = hstate_file(file); |
| 260 | struct address_space *mapping = file->f_mapping; |
Badari Pulavarty | e63e1e5 | 2007-10-16 01:26:22 -0700 | [diff] [blame] | 261 | struct inode *inode = mapping->host; |
Al Viro | 34d0640 | 2015-04-03 11:31:35 -0400 | [diff] [blame] | 262 | unsigned long index = iocb->ki_pos >> huge_page_shift(h); |
| 263 | unsigned long offset = iocb->ki_pos & ~huge_page_mask(h); |
Badari Pulavarty | e63e1e5 | 2007-10-16 01:26:22 -0700 | [diff] [blame] | 264 | unsigned long end_index; |
| 265 | loff_t isize; |
| 266 | ssize_t retval = 0; |
| 267 | |
Al Viro | 34d0640 | 2015-04-03 11:31:35 -0400 | [diff] [blame] | 268 | while (iov_iter_count(to)) { |
Badari Pulavarty | e63e1e5 | 2007-10-16 01:26:22 -0700 | [diff] [blame] | 269 | struct page *page; |
Al Viro | 34d0640 | 2015-04-03 11:31:35 -0400 | [diff] [blame] | 270 | size_t nr, copied; |
Badari Pulavarty | e63e1e5 | 2007-10-16 01:26:22 -0700 | [diff] [blame] | 271 | |
| 272 | /* nr is the maximum number of bytes to copy from this page */ |
Andi Kleen | a551643 | 2008-07-23 21:27:41 -0700 | [diff] [blame] | 273 | nr = huge_page_size(h); |
Aneesh Kumar K.V | a05b085 | 2012-03-21 16:34:08 -0700 | [diff] [blame] | 274 | isize = i_size_read(inode); |
| 275 | if (!isize) |
Al Viro | 34d0640 | 2015-04-03 11:31:35 -0400 | [diff] [blame] | 276 | break; |
Aneesh Kumar K.V | a05b085 | 2012-03-21 16:34:08 -0700 | [diff] [blame] | 277 | end_index = (isize - 1) >> huge_page_shift(h); |
Al Viro | 34d0640 | 2015-04-03 11:31:35 -0400 | [diff] [blame] | 278 | if (index > end_index) |
| 279 | break; |
| 280 | if (index == end_index) { |
Andi Kleen | a551643 | 2008-07-23 21:27:41 -0700 | [diff] [blame] | 281 | nr = ((isize - 1) & ~huge_page_mask(h)) + 1; |
Aneesh Kumar K.V | a05b085 | 2012-03-21 16:34:08 -0700 | [diff] [blame] | 282 | if (nr <= offset) |
Al Viro | 34d0640 | 2015-04-03 11:31:35 -0400 | [diff] [blame] | 283 | break; |
Badari Pulavarty | e63e1e5 | 2007-10-16 01:26:22 -0700 | [diff] [blame] | 284 | } |
| 285 | nr = nr - offset; |
| 286 | |
| 287 | /* Find the page */ |
Aneesh Kumar K.V | a05b085 | 2012-03-21 16:34:08 -0700 | [diff] [blame] | 288 | page = find_lock_page(mapping, index); |
Badari Pulavarty | e63e1e5 | 2007-10-16 01:26:22 -0700 | [diff] [blame] | 289 | if (unlikely(page == NULL)) { |
| 290 | /* |
| 291 | * We have a HOLE, zero out the user-buffer for the |
| 292 | * length of the hole or request. |
| 293 | */ |
Al Viro | 34d0640 | 2015-04-03 11:31:35 -0400 | [diff] [blame] | 294 | copied = iov_iter_zero(nr, to); |
Badari Pulavarty | e63e1e5 | 2007-10-16 01:26:22 -0700 | [diff] [blame] | 295 | } else { |
Aneesh Kumar K.V | a05b085 | 2012-03-21 16:34:08 -0700 | [diff] [blame] | 296 | unlock_page(page); |
| 297 | |
Badari Pulavarty | e63e1e5 | 2007-10-16 01:26:22 -0700 | [diff] [blame] | 298 | /* |
| 299 | * We have the page, copy it to user space buffer. |
| 300 | */ |
Al Viro | 34d0640 | 2015-04-03 11:31:35 -0400 | [diff] [blame] | 301 | copied = hugetlbfs_read_actor(page, offset, to, nr); |
Kirill A. Shutemov | 09cbfea | 2016-04-01 15:29:47 +0300 | [diff] [blame] | 302 | put_page(page); |
Badari Pulavarty | e63e1e5 | 2007-10-16 01:26:22 -0700 | [diff] [blame] | 303 | } |
Al Viro | 34d0640 | 2015-04-03 11:31:35 -0400 | [diff] [blame] | 304 | offset += copied; |
| 305 | retval += copied; |
| 306 | if (copied != nr && iov_iter_count(to)) { |
| 307 | if (!retval) |
| 308 | retval = -EFAULT; |
| 309 | break; |
Badari Pulavarty | e63e1e5 | 2007-10-16 01:26:22 -0700 | [diff] [blame] | 310 | } |
Andi Kleen | a551643 | 2008-07-23 21:27:41 -0700 | [diff] [blame] | 311 | index += offset >> huge_page_shift(h); |
| 312 | offset &= ~huge_page_mask(h); |
Badari Pulavarty | e63e1e5 | 2007-10-16 01:26:22 -0700 | [diff] [blame] | 313 | } |
Al Viro | 34d0640 | 2015-04-03 11:31:35 -0400 | [diff] [blame] | 314 | iocb->ki_pos = ((loff_t)index << huge_page_shift(h)) + offset; |
Badari Pulavarty | e63e1e5 | 2007-10-16 01:26:22 -0700 | [diff] [blame] | 315 | return retval; |
| 316 | } |
| 317 | |
Nick Piggin | 800d15a | 2007-10-16 01:25:03 -0700 | [diff] [blame] | 318 | static int hugetlbfs_write_begin(struct file *file, |
| 319 | struct address_space *mapping, |
| 320 | loff_t pos, unsigned len, unsigned flags, |
| 321 | struct page **pagep, void **fsdata) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 322 | { |
| 323 | return -EINVAL; |
| 324 | } |
| 325 | |
Nick Piggin | 800d15a | 2007-10-16 01:25:03 -0700 | [diff] [blame] | 326 | static int hugetlbfs_write_end(struct file *file, struct address_space *mapping, |
| 327 | loff_t pos, unsigned len, unsigned copied, |
| 328 | struct page *page, void *fsdata) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 329 | { |
Nick Piggin | 800d15a | 2007-10-16 01:25:03 -0700 | [diff] [blame] | 330 | BUG(); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 331 | return -EINVAL; |
| 332 | } |
| 333 | |
Mike Kravetz | b5cec28 | 2015-09-08 15:01:41 -0700 | [diff] [blame] | 334 | static void remove_huge_page(struct page *page) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 335 | { |
Konstantin Khlebnikov | b9ea251 | 2015-04-14 15:45:27 -0700 | [diff] [blame] | 336 | ClearPageDirty(page); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 337 | ClearPageUptodate(page); |
Minchan Kim | bd65cb8 | 2011-03-22 16:30:54 -0700 | [diff] [blame] | 338 | delete_from_page_cache(page); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 339 | } |
| 340 | |
Mike Kravetz | 4aae8d1 | 2016-01-15 16:57:40 -0800 | [diff] [blame] | 341 | static void |
Davidlohr Bueso | f808c13 | 2017-09-08 16:15:08 -0700 | [diff] [blame] | 342 | hugetlb_vmdelete_list(struct rb_root_cached *root, pgoff_t start, pgoff_t end) |
Mike Kravetz | 4aae8d1 | 2016-01-15 16:57:40 -0800 | [diff] [blame] | 343 | { |
| 344 | struct vm_area_struct *vma; |
| 345 | |
| 346 | /* |
| 347 | * end == 0 indicates that the entire range after |
| 348 | * start should be unmapped. |
| 349 | */ |
| 350 | vma_interval_tree_foreach(vma, root, start, end ? end : ULONG_MAX) { |
| 351 | unsigned long v_offset; |
| 352 | unsigned long v_end; |
| 353 | |
| 354 | /* |
| 355 | * Can the expression below overflow on 32-bit arches? |
| 356 | * No, because the interval tree returns us only those vmas |
| 357 | * which overlap the truncated area starting at pgoff, |
| 358 | * and no vma on a 32-bit arch can span beyond the 4GB. |
| 359 | */ |
| 360 | if (vma->vm_pgoff < start) |
| 361 | v_offset = (start - vma->vm_pgoff) << PAGE_SHIFT; |
| 362 | else |
| 363 | v_offset = 0; |
| 364 | |
| 365 | if (!end) |
| 366 | v_end = vma->vm_end; |
| 367 | else { |
| 368 | v_end = ((end - vma->vm_pgoff) << PAGE_SHIFT) |
| 369 | + vma->vm_start; |
| 370 | if (v_end > vma->vm_end) |
| 371 | v_end = vma->vm_end; |
| 372 | } |
| 373 | |
| 374 | unmap_hugepage_range(vma, vma->vm_start + v_offset, v_end, |
| 375 | NULL); |
| 376 | } |
| 377 | } |
Mike Kravetz | b5cec28 | 2015-09-08 15:01:41 -0700 | [diff] [blame] | 378 | |
| 379 | /* |
| 380 | * remove_inode_hugepages handles two distinct cases: truncation and hole |
| 381 | * punch. There are subtle differences in operation for each case. |
Mike Kravetz | 4aae8d1 | 2016-01-15 16:57:40 -0800 | [diff] [blame] | 382 | * |
Mike Kravetz | b5cec28 | 2015-09-08 15:01:41 -0700 | [diff] [blame] | 383 | * truncation is indicated by end of range being LLONG_MAX |
| 384 | * In this case, we first scan the range and release found pages. |
| 385 | * After releasing pages, hugetlb_unreserve_pages cleans up region/reserv |
Mike Kravetz | 1817889 | 2015-11-20 15:57:13 -0800 | [diff] [blame] | 386 | * maps and global counts. Page faults can not race with truncation |
| 387 | * in this routine. hugetlb_no_page() prevents page faults in the |
| 388 | * truncated range. It checks i_size before allocation, and again after |
| 389 | * with the page table lock for the page held. The same lock must be |
| 390 | * acquired to unmap a page. |
Mike Kravetz | b5cec28 | 2015-09-08 15:01:41 -0700 | [diff] [blame] | 391 | * hole punch is indicated if end is not LLONG_MAX |
| 392 | * In the hole punch case we scan the range and release found pages. |
| 393 | * Only when releasing a page is the associated region/reserv map |
| 394 | * deleted. The region/reserv map for ranges without associated |
Mike Kravetz | 1817889 | 2015-11-20 15:57:13 -0800 | [diff] [blame] | 395 | * pages are not modified. Page faults can race with hole punch. |
| 396 | * This is indicated if we find a mapped page. |
Mike Kravetz | b5cec28 | 2015-09-08 15:01:41 -0700 | [diff] [blame] | 397 | * Note: If the passed end of range value is beyond the end of file, but |
| 398 | * not LLONG_MAX this routine still performs a hole punch operation. |
| 399 | */ |
| 400 | static void remove_inode_hugepages(struct inode *inode, loff_t lstart, |
| 401 | loff_t lend) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 402 | { |
Andi Kleen | a551643 | 2008-07-23 21:27:41 -0700 | [diff] [blame] | 403 | struct hstate *h = hstate_inode(inode); |
David Gibson | b45b5bd | 2006-03-22 00:08:55 -0800 | [diff] [blame] | 404 | struct address_space *mapping = &inode->i_data; |
Andi Kleen | a551643 | 2008-07-23 21:27:41 -0700 | [diff] [blame] | 405 | const pgoff_t start = lstart >> huge_page_shift(h); |
Mike Kravetz | b5cec28 | 2015-09-08 15:01:41 -0700 | [diff] [blame] | 406 | const pgoff_t end = lend >> huge_page_shift(h); |
| 407 | struct vm_area_struct pseudo_vma; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 408 | struct pagevec pvec; |
Jan Kara | d72dc8a | 2017-09-06 16:21:18 -0700 | [diff] [blame] | 409 | pgoff_t next, index; |
Chen, Kenneth W | a43a8c3 | 2006-06-23 02:03:15 -0700 | [diff] [blame] | 410 | int i, freed = 0; |
Mike Kravetz | b5cec28 | 2015-09-08 15:01:41 -0700 | [diff] [blame] | 411 | bool truncate_op = (lend == LLONG_MAX); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 412 | |
Kirill A. Shutemov | 2c4541e | 2018-07-26 16:37:30 -0700 | [diff] [blame] | 413 | vma_init(&pseudo_vma, current->mm); |
Mike Kravetz | b5cec28 | 2015-09-08 15:01:41 -0700 | [diff] [blame] | 414 | pseudo_vma.vm_flags = (VM_HUGETLB | VM_MAYSHARE | VM_SHARED); |
Mel Gorman | 8667982 | 2017-11-15 17:37:52 -0800 | [diff] [blame] | 415 | pagevec_init(&pvec); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 416 | next = start; |
Mike Kravetz | b5cec28 | 2015-09-08 15:01:41 -0700 | [diff] [blame] | 417 | while (next < end) { |
| 418 | /* |
Mike Kravetz | 1817889 | 2015-11-20 15:57:13 -0800 | [diff] [blame] | 419 | * When no more pages are found, we are done. |
Mike Kravetz | b5cec28 | 2015-09-08 15:01:41 -0700 | [diff] [blame] | 420 | */ |
Jan Kara | 397162f | 2017-09-06 16:21:43 -0700 | [diff] [blame] | 421 | if (!pagevec_lookup_range(&pvec, mapping, &next, end - 1)) |
Mike Kravetz | 1817889 | 2015-11-20 15:57:13 -0800 | [diff] [blame] | 422 | break; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 423 | |
| 424 | for (i = 0; i < pagevec_count(&pvec); ++i) { |
| 425 | struct page *page = pvec.pages[i]; |
Mike Kravetz | b5cec28 | 2015-09-08 15:01:41 -0700 | [diff] [blame] | 426 | u32 hash; |
| 427 | |
Jan Kara | d72dc8a | 2017-09-06 16:21:18 -0700 | [diff] [blame] | 428 | index = page->index; |
Mike Kravetz | b5cec28 | 2015-09-08 15:01:41 -0700 | [diff] [blame] | 429 | hash = hugetlb_fault_mutex_hash(h, current->mm, |
| 430 | &pseudo_vma, |
Jan Kara | d72dc8a | 2017-09-06 16:21:18 -0700 | [diff] [blame] | 431 | mapping, index, 0); |
Mike Kravetz | b5cec28 | 2015-09-08 15:01:41 -0700 | [diff] [blame] | 432 | mutex_lock(&hugetlb_fault_mutex_table[hash]); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 433 | |
Mike Kravetz | 4aae8d1 | 2016-01-15 16:57:40 -0800 | [diff] [blame] | 434 | /* |
| 435 | * If page is mapped, it was faulted in after being |
| 436 | * unmapped in caller. Unmap (again) now after taking |
| 437 | * the fault mutex. The mutex will prevent faults |
| 438 | * until we finish removing the page. |
| 439 | * |
| 440 | * This race can only happen in the hole punch case. |
| 441 | * Getting here in a truncate operation is a bug. |
| 442 | */ |
| 443 | if (unlikely(page_mapped(page))) { |
Mike Kravetz | 1817889 | 2015-11-20 15:57:13 -0800 | [diff] [blame] | 444 | BUG_ON(truncate_op); |
Mike Kravetz | 4aae8d1 | 2016-01-15 16:57:40 -0800 | [diff] [blame] | 445 | |
| 446 | i_mmap_lock_write(mapping); |
| 447 | hugetlb_vmdelete_list(&mapping->i_mmap, |
Jan Kara | d72dc8a | 2017-09-06 16:21:18 -0700 | [diff] [blame] | 448 | index * pages_per_huge_page(h), |
| 449 | (index + 1) * pages_per_huge_page(h)); |
Mike Kravetz | 4aae8d1 | 2016-01-15 16:57:40 -0800 | [diff] [blame] | 450 | i_mmap_unlock_write(mapping); |
| 451 | } |
| 452 | |
| 453 | lock_page(page); |
| 454 | /* |
| 455 | * We must free the huge page and remove from page |
| 456 | * cache (remove_huge_page) BEFORE removing the |
| 457 | * region/reserve map (hugetlb_unreserve_pages). In |
| 458 | * rare out of memory conditions, removal of the |
zhong jiang | 72e2936 | 2016-10-07 17:02:01 -0700 | [diff] [blame] | 459 | * region/reserve map could fail. Correspondingly, |
| 460 | * the subpool and global reserve usage count can need |
| 461 | * to be adjusted. |
Mike Kravetz | 4aae8d1 | 2016-01-15 16:57:40 -0800 | [diff] [blame] | 462 | */ |
zhong jiang | 72e2936 | 2016-10-07 17:02:01 -0700 | [diff] [blame] | 463 | VM_BUG_ON(PagePrivate(page)); |
Mike Kravetz | 4aae8d1 | 2016-01-15 16:57:40 -0800 | [diff] [blame] | 464 | remove_huge_page(page); |
| 465 | freed++; |
| 466 | if (!truncate_op) { |
| 467 | if (unlikely(hugetlb_unreserve_pages(inode, |
Jan Kara | d72dc8a | 2017-09-06 16:21:18 -0700 | [diff] [blame] | 468 | index, index + 1, 1))) |
zhong jiang | 72e2936 | 2016-10-07 17:02:01 -0700 | [diff] [blame] | 469 | hugetlb_fix_reserve_counts(inode); |
Mike Kravetz | b5cec28 | 2015-09-08 15:01:41 -0700 | [diff] [blame] | 470 | } |
| 471 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 472 | unlock_page(page); |
Mike Kravetz | b5cec28 | 2015-09-08 15:01:41 -0700 | [diff] [blame] | 473 | mutex_unlock(&hugetlb_fault_mutex_table[hash]); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 474 | } |
| 475 | huge_pagevec_release(&pvec); |
Mike Kravetz | 1817889 | 2015-11-20 15:57:13 -0800 | [diff] [blame] | 476 | cond_resched(); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 477 | } |
Mike Kravetz | b5cec28 | 2015-09-08 15:01:41 -0700 | [diff] [blame] | 478 | |
| 479 | if (truncate_op) |
| 480 | (void)hugetlb_unreserve_pages(inode, start, LONG_MAX, freed); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 481 | } |
| 482 | |
Al Viro | 2bbbda3 | 2010-06-04 19:52:12 -0400 | [diff] [blame] | 483 | static void hugetlbfs_evict_inode(struct inode *inode) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 484 | { |
Joonsoo Kim | 9119a41 | 2014-04-03 14:47:25 -0700 | [diff] [blame] | 485 | struct resv_map *resv_map; |
| 486 | |
Mike Kravetz | b5cec28 | 2015-09-08 15:01:41 -0700 | [diff] [blame] | 487 | remove_inode_hugepages(inode, 0, LLONG_MAX); |
Joonsoo Kim | 9119a41 | 2014-04-03 14:47:25 -0700 | [diff] [blame] | 488 | resv_map = (struct resv_map *)inode->i_mapping->private_data; |
| 489 | /* root inode doesn't have the resv_map, so we should check it */ |
| 490 | if (resv_map) |
| 491 | resv_map_release(&resv_map->refs); |
Jan Kara | dbd5768 | 2012-05-03 14:48:02 +0200 | [diff] [blame] | 492 | clear_inode(inode); |
Christoph Hellwig | 149f421 | 2005-10-29 18:16:43 -0700 | [diff] [blame] | 493 | } |
| 494 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 495 | static int hugetlb_vmtruncate(struct inode *inode, loff_t offset) |
| 496 | { |
Hugh Dickins | 856fc29 | 2006-10-28 10:38:43 -0700 | [diff] [blame] | 497 | pgoff_t pgoff; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 498 | struct address_space *mapping = inode->i_mapping; |
Andi Kleen | a551643 | 2008-07-23 21:27:41 -0700 | [diff] [blame] | 499 | struct hstate *h = hstate_inode(inode); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 500 | |
Andi Kleen | a551643 | 2008-07-23 21:27:41 -0700 | [diff] [blame] | 501 | BUG_ON(offset & ~huge_page_mask(h)); |
Hugh Dickins | 856fc29 | 2006-10-28 10:38:43 -0700 | [diff] [blame] | 502 | pgoff = offset >> PAGE_SHIFT; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 503 | |
Ken Chen | 7aa91e1 | 2007-10-16 01:26:21 -0700 | [diff] [blame] | 504 | i_size_write(inode, offset); |
Davidlohr Bueso | 83cde9e | 2014-12-12 16:54:21 -0800 | [diff] [blame] | 505 | i_mmap_lock_write(mapping); |
Davidlohr Bueso | f808c13 | 2017-09-08 16:15:08 -0700 | [diff] [blame] | 506 | if (!RB_EMPTY_ROOT(&mapping->i_mmap.rb_root)) |
Mike Kravetz | 1bfad99 | 2015-09-08 15:01:38 -0700 | [diff] [blame] | 507 | hugetlb_vmdelete_list(&mapping->i_mmap, pgoff, 0); |
Davidlohr Bueso | 83cde9e | 2014-12-12 16:54:21 -0800 | [diff] [blame] | 508 | i_mmap_unlock_write(mapping); |
Mike Kravetz | b5cec28 | 2015-09-08 15:01:41 -0700 | [diff] [blame] | 509 | remove_inode_hugepages(inode, offset, LLONG_MAX); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 510 | return 0; |
| 511 | } |
| 512 | |
Mike Kravetz | 70c3547 | 2015-09-08 15:01:54 -0700 | [diff] [blame] | 513 | static long hugetlbfs_punch_hole(struct inode *inode, loff_t offset, loff_t len) |
| 514 | { |
| 515 | struct hstate *h = hstate_inode(inode); |
| 516 | loff_t hpage_size = huge_page_size(h); |
| 517 | loff_t hole_start, hole_end; |
| 518 | |
| 519 | /* |
| 520 | * For hole punch round up the beginning offset of the hole and |
| 521 | * round down the end. |
| 522 | */ |
| 523 | hole_start = round_up(offset, hpage_size); |
| 524 | hole_end = round_down(offset + len, hpage_size); |
| 525 | |
| 526 | if (hole_end > hole_start) { |
| 527 | struct address_space *mapping = inode->i_mapping; |
Marc-André Lureau | ff62a34 | 2018-01-31 16:19:25 -0800 | [diff] [blame] | 528 | struct hugetlbfs_inode_info *info = HUGETLBFS_I(inode); |
Mike Kravetz | 70c3547 | 2015-09-08 15:01:54 -0700 | [diff] [blame] | 529 | |
Al Viro | 5955102 | 2016-01-22 15:40:57 -0500 | [diff] [blame] | 530 | inode_lock(inode); |
Marc-André Lureau | ff62a34 | 2018-01-31 16:19:25 -0800 | [diff] [blame] | 531 | |
| 532 | /* protected by i_mutex */ |
| 533 | if (info->seals & F_SEAL_WRITE) { |
| 534 | inode_unlock(inode); |
| 535 | return -EPERM; |
| 536 | } |
| 537 | |
Mike Kravetz | 70c3547 | 2015-09-08 15:01:54 -0700 | [diff] [blame] | 538 | i_mmap_lock_write(mapping); |
Davidlohr Bueso | f808c13 | 2017-09-08 16:15:08 -0700 | [diff] [blame] | 539 | if (!RB_EMPTY_ROOT(&mapping->i_mmap.rb_root)) |
Mike Kravetz | 70c3547 | 2015-09-08 15:01:54 -0700 | [diff] [blame] | 540 | hugetlb_vmdelete_list(&mapping->i_mmap, |
| 541 | hole_start >> PAGE_SHIFT, |
| 542 | hole_end >> PAGE_SHIFT); |
| 543 | i_mmap_unlock_write(mapping); |
| 544 | remove_inode_hugepages(inode, hole_start, hole_end); |
Al Viro | 5955102 | 2016-01-22 15:40:57 -0500 | [diff] [blame] | 545 | inode_unlock(inode); |
Mike Kravetz | 70c3547 | 2015-09-08 15:01:54 -0700 | [diff] [blame] | 546 | } |
| 547 | |
| 548 | return 0; |
| 549 | } |
| 550 | |
| 551 | static long hugetlbfs_fallocate(struct file *file, int mode, loff_t offset, |
| 552 | loff_t len) |
| 553 | { |
| 554 | struct inode *inode = file_inode(file); |
Marc-André Lureau | ff62a34 | 2018-01-31 16:19:25 -0800 | [diff] [blame] | 555 | struct hugetlbfs_inode_info *info = HUGETLBFS_I(inode); |
Mike Kravetz | 70c3547 | 2015-09-08 15:01:54 -0700 | [diff] [blame] | 556 | struct address_space *mapping = inode->i_mapping; |
| 557 | struct hstate *h = hstate_inode(inode); |
| 558 | struct vm_area_struct pseudo_vma; |
| 559 | struct mm_struct *mm = current->mm; |
| 560 | loff_t hpage_size = huge_page_size(h); |
| 561 | unsigned long hpage_shift = huge_page_shift(h); |
| 562 | pgoff_t start, index, end; |
| 563 | int error; |
| 564 | u32 hash; |
| 565 | |
| 566 | if (mode & ~(FALLOC_FL_KEEP_SIZE | FALLOC_FL_PUNCH_HOLE)) |
| 567 | return -EOPNOTSUPP; |
| 568 | |
| 569 | if (mode & FALLOC_FL_PUNCH_HOLE) |
| 570 | return hugetlbfs_punch_hole(inode, offset, len); |
| 571 | |
| 572 | /* |
| 573 | * Default preallocate case. |
| 574 | * For this range, start is rounded down and end is rounded up |
| 575 | * as well as being converted to page offsets. |
| 576 | */ |
| 577 | start = offset >> hpage_shift; |
| 578 | end = (offset + len + hpage_size - 1) >> hpage_shift; |
| 579 | |
Al Viro | 5955102 | 2016-01-22 15:40:57 -0500 | [diff] [blame] | 580 | inode_lock(inode); |
Mike Kravetz | 70c3547 | 2015-09-08 15:01:54 -0700 | [diff] [blame] | 581 | |
| 582 | /* We need to check rlimit even when FALLOC_FL_KEEP_SIZE */ |
| 583 | error = inode_newsize_ok(inode, offset + len); |
| 584 | if (error) |
| 585 | goto out; |
| 586 | |
Marc-André Lureau | ff62a34 | 2018-01-31 16:19:25 -0800 | [diff] [blame] | 587 | if ((info->seals & F_SEAL_GROW) && offset + len > inode->i_size) { |
| 588 | error = -EPERM; |
| 589 | goto out; |
| 590 | } |
| 591 | |
Mike Kravetz | 70c3547 | 2015-09-08 15:01:54 -0700 | [diff] [blame] | 592 | /* |
| 593 | * Initialize a pseudo vma as this is required by the huge page |
| 594 | * allocation routines. If NUMA is configured, use page index |
| 595 | * as input to create an allocation policy. |
| 596 | */ |
Kirill A. Shutemov | 2c4541e | 2018-07-26 16:37:30 -0700 | [diff] [blame] | 597 | vma_init(&pseudo_vma, mm); |
Mike Kravetz | 70c3547 | 2015-09-08 15:01:54 -0700 | [diff] [blame] | 598 | pseudo_vma.vm_flags = (VM_HUGETLB | VM_MAYSHARE | VM_SHARED); |
| 599 | pseudo_vma.vm_file = file; |
| 600 | |
| 601 | for (index = start; index < end; index++) { |
| 602 | /* |
| 603 | * This is supposed to be the vaddr where the page is being |
| 604 | * faulted in, but we have no vaddr here. |
| 605 | */ |
| 606 | struct page *page; |
| 607 | unsigned long addr; |
| 608 | int avoid_reserve = 0; |
| 609 | |
| 610 | cond_resched(); |
| 611 | |
| 612 | /* |
| 613 | * fallocate(2) manpage permits EINTR; we may have been |
| 614 | * interrupted because we are using up too much memory. |
| 615 | */ |
| 616 | if (signal_pending(current)) { |
| 617 | error = -EINTR; |
| 618 | break; |
| 619 | } |
| 620 | |
| 621 | /* Set numa allocation policy based on index */ |
| 622 | hugetlb_set_vma_policy(&pseudo_vma, inode, index); |
| 623 | |
| 624 | /* addr is the offset within the file (zero based) */ |
| 625 | addr = index * hpage_size; |
| 626 | |
| 627 | /* mutex taken here, fault path and hole punch */ |
| 628 | hash = hugetlb_fault_mutex_hash(h, mm, &pseudo_vma, mapping, |
| 629 | index, addr); |
| 630 | mutex_lock(&hugetlb_fault_mutex_table[hash]); |
| 631 | |
| 632 | /* See if already present in mapping to avoid alloc/free */ |
| 633 | page = find_get_page(mapping, index); |
| 634 | if (page) { |
| 635 | put_page(page); |
| 636 | mutex_unlock(&hugetlb_fault_mutex_table[hash]); |
| 637 | hugetlb_drop_vma_policy(&pseudo_vma); |
| 638 | continue; |
| 639 | } |
| 640 | |
| 641 | /* Allocate page and add to page cache */ |
| 642 | page = alloc_huge_page(&pseudo_vma, addr, avoid_reserve); |
| 643 | hugetlb_drop_vma_policy(&pseudo_vma); |
| 644 | if (IS_ERR(page)) { |
| 645 | mutex_unlock(&hugetlb_fault_mutex_table[hash]); |
| 646 | error = PTR_ERR(page); |
| 647 | goto out; |
| 648 | } |
| 649 | clear_huge_page(page, addr, pages_per_huge_page(h)); |
| 650 | __SetPageUptodate(page); |
| 651 | error = huge_add_to_page_cache(page, mapping, index); |
| 652 | if (unlikely(error)) { |
| 653 | put_page(page); |
| 654 | mutex_unlock(&hugetlb_fault_mutex_table[hash]); |
| 655 | goto out; |
| 656 | } |
| 657 | |
| 658 | mutex_unlock(&hugetlb_fault_mutex_table[hash]); |
| 659 | |
| 660 | /* |
Mike Kravetz | 70c3547 | 2015-09-08 15:01:54 -0700 | [diff] [blame] | 661 | * unlock_page because locked by add_to_page_cache() |
Nadav Amit | 72639e6 | 2017-11-29 16:11:33 -0800 | [diff] [blame] | 662 | * page_put due to reference from alloc_huge_page() |
Mike Kravetz | 70c3547 | 2015-09-08 15:01:54 -0700 | [diff] [blame] | 663 | */ |
Mike Kravetz | 70c3547 | 2015-09-08 15:01:54 -0700 | [diff] [blame] | 664 | unlock_page(page); |
Nadav Amit | 72639e6 | 2017-11-29 16:11:33 -0800 | [diff] [blame] | 665 | put_page(page); |
Mike Kravetz | 70c3547 | 2015-09-08 15:01:54 -0700 | [diff] [blame] | 666 | } |
| 667 | |
| 668 | if (!(mode & FALLOC_FL_KEEP_SIZE) && offset + len > inode->i_size) |
| 669 | i_size_write(inode, offset + len); |
Deepa Dinamani | 078cd82 | 2016-09-14 07:48:04 -0700 | [diff] [blame] | 670 | inode->i_ctime = current_time(inode); |
Mike Kravetz | 70c3547 | 2015-09-08 15:01:54 -0700 | [diff] [blame] | 671 | out: |
Al Viro | 5955102 | 2016-01-22 15:40:57 -0500 | [diff] [blame] | 672 | inode_unlock(inode); |
Mike Kravetz | 70c3547 | 2015-09-08 15:01:54 -0700 | [diff] [blame] | 673 | return error; |
| 674 | } |
| 675 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 676 | static int hugetlbfs_setattr(struct dentry *dentry, struct iattr *attr) |
| 677 | { |
David Howells | 2b0143b | 2015-03-17 22:25:59 +0000 | [diff] [blame] | 678 | struct inode *inode = d_inode(dentry); |
Andi Kleen | a551643 | 2008-07-23 21:27:41 -0700 | [diff] [blame] | 679 | struct hstate *h = hstate_inode(inode); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 680 | int error; |
| 681 | unsigned int ia_valid = attr->ia_valid; |
Marc-André Lureau | ff62a34 | 2018-01-31 16:19:25 -0800 | [diff] [blame] | 682 | struct hugetlbfs_inode_info *info = HUGETLBFS_I(inode); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 683 | |
| 684 | BUG_ON(!inode); |
| 685 | |
Jan Kara | 31051c8 | 2016-05-26 16:55:18 +0200 | [diff] [blame] | 686 | error = setattr_prepare(dentry, attr); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 687 | if (error) |
Christoph Hellwig | 1025774 | 2010-06-04 11:30:02 +0200 | [diff] [blame] | 688 | return error; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 689 | |
| 690 | if (ia_valid & ATTR_SIZE) { |
Marc-André Lureau | ff62a34 | 2018-01-31 16:19:25 -0800 | [diff] [blame] | 691 | loff_t oldsize = inode->i_size; |
| 692 | loff_t newsize = attr->ia_size; |
| 693 | |
| 694 | if (newsize & ~huge_page_mask(h)) |
Christoph Hellwig | 1025774 | 2010-06-04 11:30:02 +0200 | [diff] [blame] | 695 | return -EINVAL; |
Marc-André Lureau | ff62a34 | 2018-01-31 16:19:25 -0800 | [diff] [blame] | 696 | /* protected by i_mutex */ |
| 697 | if ((newsize < oldsize && (info->seals & F_SEAL_SHRINK)) || |
| 698 | (newsize > oldsize && (info->seals & F_SEAL_GROW))) |
| 699 | return -EPERM; |
| 700 | error = hugetlb_vmtruncate(inode, newsize); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 701 | if (error) |
Christoph Hellwig | 1025774 | 2010-06-04 11:30:02 +0200 | [diff] [blame] | 702 | return error; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 703 | } |
Christoph Hellwig | 1025774 | 2010-06-04 11:30:02 +0200 | [diff] [blame] | 704 | |
| 705 | setattr_copy(inode, attr); |
| 706 | mark_inode_dirty(inode); |
| 707 | return 0; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 708 | } |
| 709 | |
Al Viro | 7d54fa6 | 2011-07-24 20:20:48 -0400 | [diff] [blame] | 710 | static struct inode *hugetlbfs_get_root(struct super_block *sb, |
| 711 | struct hugetlbfs_config *config) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 712 | { |
| 713 | struct inode *inode; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 714 | |
| 715 | inode = new_inode(sb); |
| 716 | if (inode) { |
Christoph Hellwig | 85fe402 | 2010-10-23 11:19:54 -0400 | [diff] [blame] | 717 | inode->i_ino = get_next_ino(); |
Al Viro | 7d54fa6 | 2011-07-24 20:20:48 -0400 | [diff] [blame] | 718 | inode->i_mode = S_IFDIR | config->mode; |
| 719 | inode->i_uid = config->uid; |
| 720 | inode->i_gid = config->gid; |
Deepa Dinamani | 078cd82 | 2016-09-14 07:48:04 -0700 | [diff] [blame] | 721 | inode->i_atime = inode->i_mtime = inode->i_ctime = current_time(inode); |
Al Viro | 7d54fa6 | 2011-07-24 20:20:48 -0400 | [diff] [blame] | 722 | inode->i_op = &hugetlbfs_dir_inode_operations; |
| 723 | inode->i_fop = &simple_dir_operations; |
| 724 | /* directory inodes start off with i_nlink == 2 (for "." entry) */ |
| 725 | inc_nlink(inode); |
Aneesh Kumar K.V | 65ed760 | 2012-04-25 16:01:50 -0700 | [diff] [blame] | 726 | lockdep_annotate_inode_mutex_key(inode); |
Al Viro | 7d54fa6 | 2011-07-24 20:20:48 -0400 | [diff] [blame] | 727 | } |
| 728 | return inode; |
| 729 | } |
| 730 | |
Michal Hocko | b610ded | 2013-08-13 16:00:55 -0700 | [diff] [blame] | 731 | /* |
Davidlohr Bueso | c8c06ef | 2014-12-12 16:54:24 -0800 | [diff] [blame] | 732 | * Hugetlbfs is not reclaimable; therefore its i_mmap_rwsem will never |
Michal Hocko | b610ded | 2013-08-13 16:00:55 -0700 | [diff] [blame] | 733 | * be taken from reclaim -- unlike regular filesystems. This needs an |
Kirill A. Shutemov | 88f306b | 2016-01-15 16:57:31 -0800 | [diff] [blame] | 734 | * annotation because huge_pmd_share() does an allocation under hugetlb's |
Davidlohr Bueso | c8c06ef | 2014-12-12 16:54:24 -0800 | [diff] [blame] | 735 | * i_mmap_rwsem. |
Michal Hocko | b610ded | 2013-08-13 16:00:55 -0700 | [diff] [blame] | 736 | */ |
Davidlohr Bueso | c8c06ef | 2014-12-12 16:54:24 -0800 | [diff] [blame] | 737 | static struct lock_class_key hugetlbfs_i_mmap_rwsem_key; |
Michal Hocko | b610ded | 2013-08-13 16:00:55 -0700 | [diff] [blame] | 738 | |
Al Viro | 7d54fa6 | 2011-07-24 20:20:48 -0400 | [diff] [blame] | 739 | static struct inode *hugetlbfs_get_inode(struct super_block *sb, |
| 740 | struct inode *dir, |
Al Viro | 18df225 | 2011-07-24 23:17:40 -0400 | [diff] [blame] | 741 | umode_t mode, dev_t dev) |
Al Viro | 7d54fa6 | 2011-07-24 20:20:48 -0400 | [diff] [blame] | 742 | { |
| 743 | struct inode *inode; |
Joonsoo Kim | 9119a41 | 2014-04-03 14:47:25 -0700 | [diff] [blame] | 744 | struct resv_map *resv_map; |
| 745 | |
| 746 | resv_map = resv_map_alloc(); |
| 747 | if (!resv_map) |
| 748 | return NULL; |
Al Viro | 7d54fa6 | 2011-07-24 20:20:48 -0400 | [diff] [blame] | 749 | |
| 750 | inode = new_inode(sb); |
| 751 | if (inode) { |
Marc-André Lureau | ff62a34 | 2018-01-31 16:19:25 -0800 | [diff] [blame] | 752 | struct hugetlbfs_inode_info *info = HUGETLBFS_I(inode); |
| 753 | |
Al Viro | 7d54fa6 | 2011-07-24 20:20:48 -0400 | [diff] [blame] | 754 | inode->i_ino = get_next_ino(); |
| 755 | inode_init_owner(inode, dir, mode); |
Davidlohr Bueso | c8c06ef | 2014-12-12 16:54:24 -0800 | [diff] [blame] | 756 | lockdep_set_class(&inode->i_mapping->i_mmap_rwsem, |
| 757 | &hugetlbfs_i_mmap_rwsem_key); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 758 | inode->i_mapping->a_ops = &hugetlbfs_aops; |
Deepa Dinamani | 078cd82 | 2016-09-14 07:48:04 -0700 | [diff] [blame] | 759 | inode->i_atime = inode->i_mtime = inode->i_ctime = current_time(inode); |
Joonsoo Kim | 9119a41 | 2014-04-03 14:47:25 -0700 | [diff] [blame] | 760 | inode->i_mapping->private_data = resv_map; |
Marc-André Lureau | ff62a34 | 2018-01-31 16:19:25 -0800 | [diff] [blame] | 761 | info->seals = F_SEAL_SEAL; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 762 | switch (mode & S_IFMT) { |
| 763 | default: |
| 764 | init_special_inode(inode, mode, dev); |
| 765 | break; |
| 766 | case S_IFREG: |
| 767 | inode->i_op = &hugetlbfs_inode_operations; |
| 768 | inode->i_fop = &hugetlbfs_file_operations; |
| 769 | break; |
| 770 | case S_IFDIR: |
| 771 | inode->i_op = &hugetlbfs_dir_inode_operations; |
| 772 | inode->i_fop = &simple_dir_operations; |
| 773 | |
| 774 | /* directory inodes start off with i_nlink == 2 (for "." entry) */ |
Dave Hansen | d8c76e6 | 2006-09-30 23:29:04 -0700 | [diff] [blame] | 775 | inc_nlink(inode); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 776 | break; |
| 777 | case S_IFLNK: |
| 778 | inode->i_op = &page_symlink_inode_operations; |
Al Viro | 21fc61c | 2015-11-17 01:07:57 -0500 | [diff] [blame] | 779 | inode_nohighmem(inode); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 780 | break; |
| 781 | } |
Josh Boyer | e096d0c | 2011-08-25 07:48:12 -0400 | [diff] [blame] | 782 | lockdep_annotate_inode_mutex_key(inode); |
Joonsoo Kim | 9119a41 | 2014-04-03 14:47:25 -0700 | [diff] [blame] | 783 | } else |
| 784 | kref_put(&resv_map->refs, resv_map_release); |
| 785 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 786 | return inode; |
| 787 | } |
| 788 | |
| 789 | /* |
| 790 | * File creation. Allocate an inode, and we're done.. |
| 791 | */ |
| 792 | static int hugetlbfs_mknod(struct inode *dir, |
Al Viro | 1a67aaf | 2011-07-26 01:52:52 -0400 | [diff] [blame] | 793 | struct dentry *dentry, umode_t mode, dev_t dev) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 794 | { |
| 795 | struct inode *inode; |
| 796 | int error = -ENOSPC; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 797 | |
Al Viro | 7d54fa6 | 2011-07-24 20:20:48 -0400 | [diff] [blame] | 798 | inode = hugetlbfs_get_inode(dir->i_sb, dir, mode, dev); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 799 | if (inode) { |
Deepa Dinamani | 078cd82 | 2016-09-14 07:48:04 -0700 | [diff] [blame] | 800 | dir->i_ctime = dir->i_mtime = current_time(dir); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 801 | d_instantiate(dentry, inode); |
| 802 | dget(dentry); /* Extra count - pin the dentry in core */ |
| 803 | error = 0; |
| 804 | } |
| 805 | return error; |
| 806 | } |
| 807 | |
Al Viro | 18bb1db | 2011-07-26 01:41:39 -0400 | [diff] [blame] | 808 | static int hugetlbfs_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 809 | { |
| 810 | int retval = hugetlbfs_mknod(dir, dentry, mode | S_IFDIR, 0); |
| 811 | if (!retval) |
Dave Hansen | d8c76e6 | 2006-09-30 23:29:04 -0700 | [diff] [blame] | 812 | inc_nlink(dir); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 813 | return retval; |
| 814 | } |
| 815 | |
Al Viro | ebfc3b4 | 2012-06-10 18:05:36 -0400 | [diff] [blame] | 816 | static int hugetlbfs_create(struct inode *dir, struct dentry *dentry, umode_t mode, bool excl) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 817 | { |
| 818 | return hugetlbfs_mknod(dir, dentry, mode | S_IFREG, 0); |
| 819 | } |
| 820 | |
| 821 | static int hugetlbfs_symlink(struct inode *dir, |
| 822 | struct dentry *dentry, const char *symname) |
| 823 | { |
| 824 | struct inode *inode; |
| 825 | int error = -ENOSPC; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 826 | |
Al Viro | 7d54fa6 | 2011-07-24 20:20:48 -0400 | [diff] [blame] | 827 | inode = hugetlbfs_get_inode(dir->i_sb, dir, S_IFLNK|S_IRWXUGO, 0); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 828 | if (inode) { |
| 829 | int l = strlen(symname)+1; |
| 830 | error = page_symlink(inode, symname, l); |
| 831 | if (!error) { |
| 832 | d_instantiate(dentry, inode); |
| 833 | dget(dentry); |
| 834 | } else |
| 835 | iput(inode); |
| 836 | } |
Deepa Dinamani | 078cd82 | 2016-09-14 07:48:04 -0700 | [diff] [blame] | 837 | dir->i_ctime = dir->i_mtime = current_time(dir); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 838 | |
| 839 | return error; |
| 840 | } |
| 841 | |
| 842 | /* |
Ken Chen | 6649a38 | 2007-02-08 14:20:27 -0800 | [diff] [blame] | 843 | * mark the head page dirty |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 844 | */ |
| 845 | static int hugetlbfs_set_page_dirty(struct page *page) |
| 846 | { |
Christoph Lameter | d85f338 | 2007-05-06 14:49:39 -0700 | [diff] [blame] | 847 | struct page *head = compound_head(page); |
Ken Chen | 6649a38 | 2007-02-08 14:20:27 -0800 | [diff] [blame] | 848 | |
| 849 | SetPageDirty(head); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 850 | return 0; |
| 851 | } |
| 852 | |
Naoya Horiguchi | 290408d | 2010-09-08 10:19:35 +0900 | [diff] [blame] | 853 | static int hugetlbfs_migrate_page(struct address_space *mapping, |
Mel Gorman | b969c4a | 2012-01-12 17:19:34 -0800 | [diff] [blame] | 854 | struct page *newpage, struct page *page, |
Mel Gorman | a6bc32b | 2012-01-12 17:19:43 -0800 | [diff] [blame] | 855 | enum migrate_mode mode) |
Naoya Horiguchi | 290408d | 2010-09-08 10:19:35 +0900 | [diff] [blame] | 856 | { |
| 857 | int rc; |
| 858 | |
| 859 | rc = migrate_huge_page_move_mapping(mapping, newpage, page); |
Rafael Aquini | 78bd520 | 2012-12-11 16:02:31 -0800 | [diff] [blame] | 860 | if (rc != MIGRATEPAGE_SUCCESS) |
Naoya Horiguchi | 290408d | 2010-09-08 10:19:35 +0900 | [diff] [blame] | 861 | return rc; |
Jérôme Glisse | 2916ecc | 2017-09-08 16:12:06 -0700 | [diff] [blame] | 862 | if (mode != MIGRATE_SYNC_NO_COPY) |
| 863 | migrate_page_copy(newpage, page); |
| 864 | else |
| 865 | migrate_page_states(newpage, page); |
Naoya Horiguchi | 290408d | 2010-09-08 10:19:35 +0900 | [diff] [blame] | 866 | |
Rafael Aquini | 78bd520 | 2012-12-11 16:02:31 -0800 | [diff] [blame] | 867 | return MIGRATEPAGE_SUCCESS; |
Naoya Horiguchi | 290408d | 2010-09-08 10:19:35 +0900 | [diff] [blame] | 868 | } |
| 869 | |
Naoya Horiguchi | 78bb920 | 2017-07-10 15:47:50 -0700 | [diff] [blame] | 870 | static int hugetlbfs_error_remove_page(struct address_space *mapping, |
| 871 | struct page *page) |
| 872 | { |
| 873 | struct inode *inode = mapping->host; |
Mike Kravetz | ab615a5 | 2017-11-02 15:59:41 -0700 | [diff] [blame] | 874 | pgoff_t index = page->index; |
Naoya Horiguchi | 78bb920 | 2017-07-10 15:47:50 -0700 | [diff] [blame] | 875 | |
| 876 | remove_huge_page(page); |
Mike Kravetz | ab615a5 | 2017-11-02 15:59:41 -0700 | [diff] [blame] | 877 | if (unlikely(hugetlb_unreserve_pages(inode, index, index + 1, 1))) |
| 878 | hugetlb_fix_reserve_counts(inode); |
| 879 | |
Naoya Horiguchi | 78bb920 | 2017-07-10 15:47:50 -0700 | [diff] [blame] | 880 | return 0; |
| 881 | } |
| 882 | |
David Howells | 4a25220 | 2017-07-05 16:24:18 +0100 | [diff] [blame] | 883 | /* |
| 884 | * Display the mount options in /proc/mounts. |
| 885 | */ |
| 886 | static int hugetlbfs_show_options(struct seq_file *m, struct dentry *root) |
| 887 | { |
| 888 | struct hugetlbfs_sb_info *sbinfo = HUGETLBFS_SB(root->d_sb); |
| 889 | struct hugepage_subpool *spool = sbinfo->spool; |
| 890 | unsigned long hpage_size = huge_page_size(sbinfo->hstate); |
| 891 | unsigned hpage_shift = huge_page_shift(sbinfo->hstate); |
| 892 | char mod; |
| 893 | |
| 894 | if (!uid_eq(sbinfo->uid, GLOBAL_ROOT_UID)) |
| 895 | seq_printf(m, ",uid=%u", |
| 896 | from_kuid_munged(&init_user_ns, sbinfo->uid)); |
| 897 | if (!gid_eq(sbinfo->gid, GLOBAL_ROOT_GID)) |
| 898 | seq_printf(m, ",gid=%u", |
| 899 | from_kgid_munged(&init_user_ns, sbinfo->gid)); |
| 900 | if (sbinfo->mode != 0755) |
| 901 | seq_printf(m, ",mode=%o", sbinfo->mode); |
| 902 | if (sbinfo->max_inodes != -1) |
| 903 | seq_printf(m, ",nr_inodes=%lu", sbinfo->max_inodes); |
| 904 | |
| 905 | hpage_size /= 1024; |
| 906 | mod = 'K'; |
| 907 | if (hpage_size >= 1024) { |
| 908 | hpage_size /= 1024; |
| 909 | mod = 'M'; |
| 910 | } |
| 911 | seq_printf(m, ",pagesize=%lu%c", hpage_size, mod); |
| 912 | if (spool) { |
| 913 | if (spool->max_hpages != -1) |
| 914 | seq_printf(m, ",size=%llu", |
| 915 | (unsigned long long)spool->max_hpages << hpage_shift); |
| 916 | if (spool->min_hpages != -1) |
| 917 | seq_printf(m, ",min_size=%llu", |
| 918 | (unsigned long long)spool->min_hpages << hpage_shift); |
| 919 | } |
| 920 | return 0; |
| 921 | } |
| 922 | |
David Howells | 726c334 | 2006-06-23 02:02:58 -0700 | [diff] [blame] | 923 | static int hugetlbfs_statfs(struct dentry *dentry, struct kstatfs *buf) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 924 | { |
David Howells | 726c334 | 2006-06-23 02:02:58 -0700 | [diff] [blame] | 925 | struct hugetlbfs_sb_info *sbinfo = HUGETLBFS_SB(dentry->d_sb); |
David Howells | 2b0143b | 2015-03-17 22:25:59 +0000 | [diff] [blame] | 926 | struct hstate *h = hstate_inode(d_inode(dentry)); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 927 | |
| 928 | buf->f_type = HUGETLBFS_MAGIC; |
Andi Kleen | a551643 | 2008-07-23 21:27:41 -0700 | [diff] [blame] | 929 | buf->f_bsize = huge_page_size(h); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 930 | if (sbinfo) { |
| 931 | spin_lock(&sbinfo->stat_lock); |
David Gibson | 74a8a65 | 2005-11-21 21:32:24 -0800 | [diff] [blame] | 932 | /* If no limits set, just report 0 for max/free/used |
| 933 | * blocks, like simple_statfs() */ |
David Gibson | 9048162 | 2012-03-21 16:34:12 -0700 | [diff] [blame] | 934 | if (sbinfo->spool) { |
| 935 | long free_pages; |
| 936 | |
| 937 | spin_lock(&sbinfo->spool->lock); |
| 938 | buf->f_blocks = sbinfo->spool->max_hpages; |
| 939 | free_pages = sbinfo->spool->max_hpages |
| 940 | - sbinfo->spool->used_hpages; |
| 941 | buf->f_bavail = buf->f_bfree = free_pages; |
| 942 | spin_unlock(&sbinfo->spool->lock); |
David Gibson | 74a8a65 | 2005-11-21 21:32:24 -0800 | [diff] [blame] | 943 | buf->f_files = sbinfo->max_inodes; |
| 944 | buf->f_ffree = sbinfo->free_inodes; |
| 945 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 946 | spin_unlock(&sbinfo->stat_lock); |
| 947 | } |
| 948 | buf->f_namelen = NAME_MAX; |
| 949 | return 0; |
| 950 | } |
| 951 | |
| 952 | static void hugetlbfs_put_super(struct super_block *sb) |
| 953 | { |
| 954 | struct hugetlbfs_sb_info *sbi = HUGETLBFS_SB(sb); |
| 955 | |
| 956 | if (sbi) { |
| 957 | sb->s_fs_info = NULL; |
David Gibson | 9048162 | 2012-03-21 16:34:12 -0700 | [diff] [blame] | 958 | |
| 959 | if (sbi->spool) |
| 960 | hugepage_put_subpool(sbi->spool); |
| 961 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 962 | kfree(sbi); |
| 963 | } |
| 964 | } |
| 965 | |
Christoph Hellwig | 9652798 | 2005-10-29 18:16:42 -0700 | [diff] [blame] | 966 | static inline int hugetlbfs_dec_free_inodes(struct hugetlbfs_sb_info *sbinfo) |
| 967 | { |
| 968 | if (sbinfo->free_inodes >= 0) { |
| 969 | spin_lock(&sbinfo->stat_lock); |
| 970 | if (unlikely(!sbinfo->free_inodes)) { |
| 971 | spin_unlock(&sbinfo->stat_lock); |
| 972 | return 0; |
| 973 | } |
| 974 | sbinfo->free_inodes--; |
| 975 | spin_unlock(&sbinfo->stat_lock); |
| 976 | } |
| 977 | |
| 978 | return 1; |
| 979 | } |
| 980 | |
| 981 | static void hugetlbfs_inc_free_inodes(struct hugetlbfs_sb_info *sbinfo) |
| 982 | { |
| 983 | if (sbinfo->free_inodes >= 0) { |
| 984 | spin_lock(&sbinfo->stat_lock); |
| 985 | sbinfo->free_inodes++; |
| 986 | spin_unlock(&sbinfo->stat_lock); |
| 987 | } |
| 988 | } |
| 989 | |
| 990 | |
Christoph Lameter | e18b890 | 2006-12-06 20:33:20 -0800 | [diff] [blame] | 991 | static struct kmem_cache *hugetlbfs_inode_cachep; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 992 | |
| 993 | static struct inode *hugetlbfs_alloc_inode(struct super_block *sb) |
| 994 | { |
Christoph Hellwig | 9652798 | 2005-10-29 18:16:42 -0700 | [diff] [blame] | 995 | struct hugetlbfs_sb_info *sbinfo = HUGETLBFS_SB(sb); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 996 | struct hugetlbfs_inode_info *p; |
| 997 | |
Christoph Hellwig | 9652798 | 2005-10-29 18:16:42 -0700 | [diff] [blame] | 998 | if (unlikely(!hugetlbfs_dec_free_inodes(sbinfo))) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 999 | return NULL; |
Christoph Lameter | e94b176 | 2006-12-06 20:33:17 -0800 | [diff] [blame] | 1000 | p = kmem_cache_alloc(hugetlbfs_inode_cachep, GFP_KERNEL); |
Christoph Hellwig | 9652798 | 2005-10-29 18:16:42 -0700 | [diff] [blame] | 1001 | if (unlikely(!p)) { |
| 1002 | hugetlbfs_inc_free_inodes(sbinfo); |
| 1003 | return NULL; |
| 1004 | } |
Mike Kravetz | 4742a35 | 2017-03-31 15:12:01 -0700 | [diff] [blame] | 1005 | |
| 1006 | /* |
| 1007 | * Any time after allocation, hugetlbfs_destroy_inode can be called |
| 1008 | * for the inode. mpol_free_shared_policy is unconditionally called |
| 1009 | * as part of hugetlbfs_destroy_inode. So, initialize policy here |
| 1010 | * in case of a quick call to destroy. |
| 1011 | * |
| 1012 | * Note that the policy is initialized even if we are creating a |
| 1013 | * private inode. This simplifies hugetlbfs_destroy_inode. |
| 1014 | */ |
| 1015 | mpol_shared_policy_init(&p->policy, NULL); |
| 1016 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1017 | return &p->vfs_inode; |
| 1018 | } |
| 1019 | |
Nick Piggin | fa0d7e3d | 2011-01-07 17:49:49 +1100 | [diff] [blame] | 1020 | static void hugetlbfs_i_callback(struct rcu_head *head) |
| 1021 | { |
| 1022 | struct inode *inode = container_of(head, struct inode, i_rcu); |
Nick Piggin | fa0d7e3d | 2011-01-07 17:49:49 +1100 | [diff] [blame] | 1023 | kmem_cache_free(hugetlbfs_inode_cachep, HUGETLBFS_I(inode)); |
| 1024 | } |
| 1025 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1026 | static void hugetlbfs_destroy_inode(struct inode *inode) |
| 1027 | { |
Christoph Hellwig | 9652798 | 2005-10-29 18:16:42 -0700 | [diff] [blame] | 1028 | hugetlbfs_inc_free_inodes(HUGETLBFS_SB(inode->i_sb)); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1029 | mpol_free_shared_policy(&HUGETLBFS_I(inode)->policy); |
Nick Piggin | fa0d7e3d | 2011-01-07 17:49:49 +1100 | [diff] [blame] | 1030 | call_rcu(&inode->i_rcu, hugetlbfs_i_callback); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1031 | } |
| 1032 | |
Christoph Hellwig | f5e54d6 | 2006-06-28 04:26:44 -0700 | [diff] [blame] | 1033 | static const struct address_space_operations hugetlbfs_aops = { |
Nick Piggin | 800d15a | 2007-10-16 01:25:03 -0700 | [diff] [blame] | 1034 | .write_begin = hugetlbfs_write_begin, |
| 1035 | .write_end = hugetlbfs_write_end, |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1036 | .set_page_dirty = hugetlbfs_set_page_dirty, |
Naoya Horiguchi | 290408d | 2010-09-08 10:19:35 +0900 | [diff] [blame] | 1037 | .migratepage = hugetlbfs_migrate_page, |
Naoya Horiguchi | 78bb920 | 2017-07-10 15:47:50 -0700 | [diff] [blame] | 1038 | .error_remove_page = hugetlbfs_error_remove_page, |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1039 | }; |
| 1040 | |
Christoph Hellwig | 9652798 | 2005-10-29 18:16:42 -0700 | [diff] [blame] | 1041 | |
Alexey Dobriyan | 51cc506 | 2008-07-25 19:45:34 -0700 | [diff] [blame] | 1042 | static void init_once(void *foo) |
Christoph Hellwig | 9652798 | 2005-10-29 18:16:42 -0700 | [diff] [blame] | 1043 | { |
| 1044 | struct hugetlbfs_inode_info *ei = (struct hugetlbfs_inode_info *)foo; |
| 1045 | |
Christoph Lameter | a35afb8 | 2007-05-16 22:10:57 -0700 | [diff] [blame] | 1046 | inode_init_once(&ei->vfs_inode); |
Christoph Hellwig | 9652798 | 2005-10-29 18:16:42 -0700 | [diff] [blame] | 1047 | } |
| 1048 | |
Arjan van de Ven | 4b6f5d2 | 2006-03-28 01:56:42 -0800 | [diff] [blame] | 1049 | const struct file_operations hugetlbfs_file_operations = { |
Al Viro | 34d0640 | 2015-04-03 11:31:35 -0400 | [diff] [blame] | 1050 | .read_iter = hugetlbfs_read_iter, |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1051 | .mmap = hugetlbfs_file_mmap, |
Christoph Hellwig | 1b061d9 | 2010-05-26 17:53:41 +0200 | [diff] [blame] | 1052 | .fsync = noop_fsync, |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1053 | .get_unmapped_area = hugetlb_get_unmapped_area, |
Mike Kravetz | 70c3547 | 2015-09-08 15:01:54 -0700 | [diff] [blame] | 1054 | .llseek = default_llseek, |
| 1055 | .fallocate = hugetlbfs_fallocate, |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1056 | }; |
| 1057 | |
Arjan van de Ven | 92e1d5b | 2007-02-12 00:55:39 -0800 | [diff] [blame] | 1058 | static const struct inode_operations hugetlbfs_dir_inode_operations = { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1059 | .create = hugetlbfs_create, |
| 1060 | .lookup = simple_lookup, |
| 1061 | .link = simple_link, |
| 1062 | .unlink = simple_unlink, |
| 1063 | .symlink = hugetlbfs_symlink, |
| 1064 | .mkdir = hugetlbfs_mkdir, |
| 1065 | .rmdir = simple_rmdir, |
| 1066 | .mknod = hugetlbfs_mknod, |
| 1067 | .rename = simple_rename, |
| 1068 | .setattr = hugetlbfs_setattr, |
| 1069 | }; |
| 1070 | |
Arjan van de Ven | 92e1d5b | 2007-02-12 00:55:39 -0800 | [diff] [blame] | 1071 | static const struct inode_operations hugetlbfs_inode_operations = { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1072 | .setattr = hugetlbfs_setattr, |
| 1073 | }; |
| 1074 | |
Josef 'Jeff' Sipek | ee9b6d6 | 2007-02-12 00:55:41 -0800 | [diff] [blame] | 1075 | static const struct super_operations hugetlbfs_ops = { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1076 | .alloc_inode = hugetlbfs_alloc_inode, |
| 1077 | .destroy_inode = hugetlbfs_destroy_inode, |
Al Viro | 2bbbda3 | 2010-06-04 19:52:12 -0400 | [diff] [blame] | 1078 | .evict_inode = hugetlbfs_evict_inode, |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1079 | .statfs = hugetlbfs_statfs, |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1080 | .put_super = hugetlbfs_put_super, |
David Howells | 4a25220 | 2017-07-05 16:24:18 +0100 | [diff] [blame] | 1081 | .show_options = hugetlbfs_show_options, |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1082 | }; |
| 1083 | |
David Howells | 4a25220 | 2017-07-05 16:24:18 +0100 | [diff] [blame] | 1084 | enum hugetlbfs_size_type { NO_SIZE, SIZE_STD, SIZE_PERCENT }; |
Mike Kravetz | 7ca02d0a | 2015-04-15 16:13:42 -0700 | [diff] [blame] | 1085 | |
| 1086 | /* |
| 1087 | * Convert size option passed from command line to number of huge pages |
| 1088 | * in the pool specified by hstate. Size option could be in bytes |
| 1089 | * (val_type == SIZE_STD) or percentage of the pool (val_type == SIZE_PERCENT). |
| 1090 | */ |
David Howells | 4a25220 | 2017-07-05 16:24:18 +0100 | [diff] [blame] | 1091 | static long |
Mike Kravetz | 7ca02d0a | 2015-04-15 16:13:42 -0700 | [diff] [blame] | 1092 | hugetlbfs_size_to_hpages(struct hstate *h, unsigned long long size_opt, |
David Howells | 4a25220 | 2017-07-05 16:24:18 +0100 | [diff] [blame] | 1093 | enum hugetlbfs_size_type val_type) |
Mike Kravetz | 7ca02d0a | 2015-04-15 16:13:42 -0700 | [diff] [blame] | 1094 | { |
| 1095 | if (val_type == NO_SIZE) |
| 1096 | return -1; |
| 1097 | |
| 1098 | if (val_type == SIZE_PERCENT) { |
| 1099 | size_opt <<= huge_page_shift(h); |
| 1100 | size_opt *= h->max_huge_pages; |
| 1101 | do_div(size_opt, 100); |
| 1102 | } |
| 1103 | |
| 1104 | size_opt >>= huge_page_shift(h); |
| 1105 | return size_opt; |
| 1106 | } |
| 1107 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1108 | static int |
| 1109 | hugetlbfs_parse_options(char *options, struct hugetlbfs_config *pconfig) |
| 1110 | { |
Randy Dunlap | e73a75f | 2007-07-15 23:40:52 -0700 | [diff] [blame] | 1111 | char *p, *rest; |
| 1112 | substring_t args[MAX_OPT_ARGS]; |
| 1113 | int option; |
Mike Kravetz | 7ca02d0a | 2015-04-15 16:13:42 -0700 | [diff] [blame] | 1114 | unsigned long long max_size_opt = 0, min_size_opt = 0; |
David Howells | 4a25220 | 2017-07-05 16:24:18 +0100 | [diff] [blame] | 1115 | enum hugetlbfs_size_type max_val_type = NO_SIZE, min_val_type = NO_SIZE; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1116 | |
| 1117 | if (!options) |
| 1118 | return 0; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1119 | |
Randy Dunlap | e73a75f | 2007-07-15 23:40:52 -0700 | [diff] [blame] | 1120 | while ((p = strsep(&options, ",")) != NULL) { |
| 1121 | int token; |
Lee Schermerhorn | b4c07bc | 2007-07-15 23:40:54 -0700 | [diff] [blame] | 1122 | if (!*p) |
| 1123 | continue; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1124 | |
Randy Dunlap | e73a75f | 2007-07-15 23:40:52 -0700 | [diff] [blame] | 1125 | token = match_token(p, tokens, args); |
| 1126 | switch (token) { |
| 1127 | case Opt_uid: |
| 1128 | if (match_int(&args[0], &option)) |
| 1129 | goto bad_val; |
Eric W. Biederman | a0eb3a0 | 2012-02-07 16:19:25 -0800 | [diff] [blame] | 1130 | pconfig->uid = make_kuid(current_user_ns(), option); |
| 1131 | if (!uid_valid(pconfig->uid)) |
| 1132 | goto bad_val; |
Randy Dunlap | e73a75f | 2007-07-15 23:40:52 -0700 | [diff] [blame] | 1133 | break; |
| 1134 | |
| 1135 | case Opt_gid: |
| 1136 | if (match_int(&args[0], &option)) |
| 1137 | goto bad_val; |
Eric W. Biederman | a0eb3a0 | 2012-02-07 16:19:25 -0800 | [diff] [blame] | 1138 | pconfig->gid = make_kgid(current_user_ns(), option); |
| 1139 | if (!gid_valid(pconfig->gid)) |
| 1140 | goto bad_val; |
Randy Dunlap | e73a75f | 2007-07-15 23:40:52 -0700 | [diff] [blame] | 1141 | break; |
| 1142 | |
| 1143 | case Opt_mode: |
| 1144 | if (match_octal(&args[0], &option)) |
| 1145 | goto bad_val; |
Ken Chen | 75897d6 | 2008-02-04 22:28:36 -0800 | [diff] [blame] | 1146 | pconfig->mode = option & 01777U; |
Randy Dunlap | e73a75f | 2007-07-15 23:40:52 -0700 | [diff] [blame] | 1147 | break; |
| 1148 | |
| 1149 | case Opt_size: { |
Randy Dunlap | e73a75f | 2007-07-15 23:40:52 -0700 | [diff] [blame] | 1150 | /* memparse() will accept a K/M/G without a digit */ |
| 1151 | if (!isdigit(*args[0].from)) |
| 1152 | goto bad_val; |
Mike Kravetz | 7ca02d0a | 2015-04-15 16:13:42 -0700 | [diff] [blame] | 1153 | max_size_opt = memparse(args[0].from, &rest); |
| 1154 | max_val_type = SIZE_STD; |
Andi Kleen | a137e1c | 2008-07-23 21:27:43 -0700 | [diff] [blame] | 1155 | if (*rest == '%') |
Mike Kravetz | 7ca02d0a | 2015-04-15 16:13:42 -0700 | [diff] [blame] | 1156 | max_val_type = SIZE_PERCENT; |
Randy Dunlap | e73a75f | 2007-07-15 23:40:52 -0700 | [diff] [blame] | 1157 | break; |
| 1158 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1159 | |
Randy Dunlap | e73a75f | 2007-07-15 23:40:52 -0700 | [diff] [blame] | 1160 | case Opt_nr_inodes: |
| 1161 | /* memparse() will accept a K/M/G without a digit */ |
| 1162 | if (!isdigit(*args[0].from)) |
| 1163 | goto bad_val; |
| 1164 | pconfig->nr_inodes = memparse(args[0].from, &rest); |
| 1165 | break; |
| 1166 | |
Andi Kleen | a137e1c | 2008-07-23 21:27:43 -0700 | [diff] [blame] | 1167 | case Opt_pagesize: { |
| 1168 | unsigned long ps; |
| 1169 | ps = memparse(args[0].from, &rest); |
| 1170 | pconfig->hstate = size_to_hstate(ps); |
| 1171 | if (!pconfig->hstate) { |
Andrew Morton | 9b857d2 | 2014-06-04 16:07:21 -0700 | [diff] [blame] | 1172 | pr_err("Unsupported page size %lu MB\n", |
Andi Kleen | a137e1c | 2008-07-23 21:27:43 -0700 | [diff] [blame] | 1173 | ps >> 20); |
| 1174 | return -EINVAL; |
| 1175 | } |
| 1176 | break; |
| 1177 | } |
| 1178 | |
Mike Kravetz | 7ca02d0a | 2015-04-15 16:13:42 -0700 | [diff] [blame] | 1179 | case Opt_min_size: { |
| 1180 | /* memparse() will accept a K/M/G without a digit */ |
| 1181 | if (!isdigit(*args[0].from)) |
| 1182 | goto bad_val; |
| 1183 | min_size_opt = memparse(args[0].from, &rest); |
| 1184 | min_val_type = SIZE_STD; |
| 1185 | if (*rest == '%') |
| 1186 | min_val_type = SIZE_PERCENT; |
| 1187 | break; |
| 1188 | } |
| 1189 | |
Randy Dunlap | e73a75f | 2007-07-15 23:40:52 -0700 | [diff] [blame] | 1190 | default: |
Andrew Morton | 9b857d2 | 2014-06-04 16:07:21 -0700 | [diff] [blame] | 1191 | pr_err("Bad mount option: \"%s\"\n", p); |
Lee Schermerhorn | b4c07bc | 2007-07-15 23:40:54 -0700 | [diff] [blame] | 1192 | return -EINVAL; |
Randy Dunlap | e73a75f | 2007-07-15 23:40:52 -0700 | [diff] [blame] | 1193 | break; |
| 1194 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1195 | } |
Andi Kleen | a137e1c | 2008-07-23 21:27:43 -0700 | [diff] [blame] | 1196 | |
Mike Kravetz | 7ca02d0a | 2015-04-15 16:13:42 -0700 | [diff] [blame] | 1197 | /* |
| 1198 | * Use huge page pool size (in hstate) to convert the size |
| 1199 | * options to number of huge pages. If NO_SIZE, -1 is returned. |
| 1200 | */ |
| 1201 | pconfig->max_hpages = hugetlbfs_size_to_hpages(pconfig->hstate, |
| 1202 | max_size_opt, max_val_type); |
| 1203 | pconfig->min_hpages = hugetlbfs_size_to_hpages(pconfig->hstate, |
| 1204 | min_size_opt, min_val_type); |
| 1205 | |
| 1206 | /* |
| 1207 | * If max_size was specified, then min_size must be smaller |
| 1208 | */ |
| 1209 | if (max_val_type > NO_SIZE && |
| 1210 | pconfig->min_hpages > pconfig->max_hpages) { |
| 1211 | pr_err("minimum size can not be greater than maximum size\n"); |
| 1212 | return -EINVAL; |
Andi Kleen | a137e1c | 2008-07-23 21:27:43 -0700 | [diff] [blame] | 1213 | } |
| 1214 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1215 | return 0; |
Randy Dunlap | e73a75f | 2007-07-15 23:40:52 -0700 | [diff] [blame] | 1216 | |
| 1217 | bad_val: |
Andrew Morton | 9b857d2 | 2014-06-04 16:07:21 -0700 | [diff] [blame] | 1218 | pr_err("Bad value '%s' for mount option '%s'\n", args[0].from, p); |
Akinobu Mita | c12ddba | 2009-04-21 12:24:05 -0700 | [diff] [blame] | 1219 | return -EINVAL; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1220 | } |
| 1221 | |
| 1222 | static int |
| 1223 | hugetlbfs_fill_super(struct super_block *sb, void *data, int silent) |
| 1224 | { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1225 | int ret; |
| 1226 | struct hugetlbfs_config config; |
| 1227 | struct hugetlbfs_sb_info *sbinfo; |
| 1228 | |
Mike Kravetz | 7ca02d0a | 2015-04-15 16:13:42 -0700 | [diff] [blame] | 1229 | config.max_hpages = -1; /* No limit on size by default */ |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1230 | config.nr_inodes = -1; /* No limit on number of inodes by default */ |
David Howells | 77c70de | 2008-11-14 10:38:56 +1100 | [diff] [blame] | 1231 | config.uid = current_fsuid(); |
| 1232 | config.gid = current_fsgid(); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1233 | config.mode = 0755; |
Andi Kleen | a137e1c | 2008-07-23 21:27:43 -0700 | [diff] [blame] | 1234 | config.hstate = &default_hstate; |
Mike Kravetz | 7ca02d0a | 2015-04-15 16:13:42 -0700 | [diff] [blame] | 1235 | config.min_hpages = -1; /* No default minimum size */ |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1236 | ret = hugetlbfs_parse_options(data, &config); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1237 | if (ret) |
| 1238 | return ret; |
| 1239 | |
| 1240 | sbinfo = kmalloc(sizeof(struct hugetlbfs_sb_info), GFP_KERNEL); |
| 1241 | if (!sbinfo) |
| 1242 | return -ENOMEM; |
| 1243 | sb->s_fs_info = sbinfo; |
Andi Kleen | a137e1c | 2008-07-23 21:27:43 -0700 | [diff] [blame] | 1244 | sbinfo->hstate = config.hstate; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1245 | spin_lock_init(&sbinfo->stat_lock); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1246 | sbinfo->max_inodes = config.nr_inodes; |
| 1247 | sbinfo->free_inodes = config.nr_inodes; |
David Gibson | 9048162 | 2012-03-21 16:34:12 -0700 | [diff] [blame] | 1248 | sbinfo->spool = NULL; |
David Howells | 4a25220 | 2017-07-05 16:24:18 +0100 | [diff] [blame] | 1249 | sbinfo->uid = config.uid; |
| 1250 | sbinfo->gid = config.gid; |
| 1251 | sbinfo->mode = config.mode; |
| 1252 | |
Mike Kravetz | 7ca02d0a | 2015-04-15 16:13:42 -0700 | [diff] [blame] | 1253 | /* |
| 1254 | * Allocate and initialize subpool if maximum or minimum size is |
| 1255 | * specified. Any needed reservations (for minimim size) are taken |
| 1256 | * taken when the subpool is created. |
| 1257 | */ |
| 1258 | if (config.max_hpages != -1 || config.min_hpages != -1) { |
| 1259 | sbinfo->spool = hugepage_new_subpool(config.hstate, |
| 1260 | config.max_hpages, |
| 1261 | config.min_hpages); |
David Gibson | 9048162 | 2012-03-21 16:34:12 -0700 | [diff] [blame] | 1262 | if (!sbinfo->spool) |
| 1263 | goto out_free; |
| 1264 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1265 | sb->s_maxbytes = MAX_LFS_FILESIZE; |
Andi Kleen | a137e1c | 2008-07-23 21:27:43 -0700 | [diff] [blame] | 1266 | sb->s_blocksize = huge_page_size(config.hstate); |
| 1267 | sb->s_blocksize_bits = huge_page_shift(config.hstate); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1268 | sb->s_magic = HUGETLBFS_MAGIC; |
| 1269 | sb->s_op = &hugetlbfs_ops; |
| 1270 | sb->s_time_gran = 1; |
Al Viro | 48fde70 | 2012-01-08 22:15:13 -0500 | [diff] [blame] | 1271 | sb->s_root = d_make_root(hugetlbfs_get_root(sb, &config)); |
| 1272 | if (!sb->s_root) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1273 | goto out_free; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1274 | return 0; |
| 1275 | out_free: |
Fabian Frederick | 6e6870d | 2014-06-04 16:10:40 -0700 | [diff] [blame] | 1276 | kfree(sbinfo->spool); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1277 | kfree(sbinfo); |
| 1278 | return -ENOMEM; |
| 1279 | } |
| 1280 | |
Al Viro | 3c26ff6 | 2010-07-25 11:46:36 +0400 | [diff] [blame] | 1281 | static struct dentry *hugetlbfs_mount(struct file_system_type *fs_type, |
| 1282 | int flags, const char *dev_name, void *data) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1283 | { |
Al Viro | 3c26ff6 | 2010-07-25 11:46:36 +0400 | [diff] [blame] | 1284 | return mount_nodev(fs_type, flags, data, hugetlbfs_fill_super); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1285 | } |
| 1286 | |
| 1287 | static struct file_system_type hugetlbfs_fs_type = { |
| 1288 | .name = "hugetlbfs", |
Al Viro | 3c26ff6 | 2010-07-25 11:46:36 +0400 | [diff] [blame] | 1289 | .mount = hugetlbfs_mount, |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1290 | .kill_sb = kill_litter_super, |
| 1291 | }; |
| 1292 | |
Andi Kleen | 42d7395 | 2012-12-11 16:01:34 -0800 | [diff] [blame] | 1293 | static struct vfsmount *hugetlbfs_vfsmount[HUGE_MAX_HSTATE]; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1294 | |
From: Mel Gorman | ef1ff6b | 2009-09-23 15:56:05 -0700 | [diff] [blame] | 1295 | static int can_do_hugetlb_shm(void) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1296 | { |
Eric W. Biederman | a0eb3a0 | 2012-02-07 16:19:25 -0800 | [diff] [blame] | 1297 | kgid_t shm_group; |
| 1298 | shm_group = make_kgid(&init_user_ns, sysctl_hugetlb_shm_group); |
| 1299 | return capable(CAP_IPC_LOCK) || in_group_p(shm_group); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1300 | } |
| 1301 | |
Andi Kleen | 42d7395 | 2012-12-11 16:01:34 -0800 | [diff] [blame] | 1302 | static int get_hstate_idx(int page_size_log) |
| 1303 | { |
Naoya Horiguchi | af73e4d | 2013-05-07 16:18:13 -0700 | [diff] [blame] | 1304 | struct hstate *h = hstate_sizelog(page_size_log); |
Andi Kleen | 42d7395 | 2012-12-11 16:01:34 -0800 | [diff] [blame] | 1305 | |
Andi Kleen | 42d7395 | 2012-12-11 16:01:34 -0800 | [diff] [blame] | 1306 | if (!h) |
| 1307 | return -1; |
| 1308 | return h - hstates; |
| 1309 | } |
| 1310 | |
Naoya Horiguchi | af73e4d | 2013-05-07 16:18:13 -0700 | [diff] [blame] | 1311 | /* |
| 1312 | * Note that size should be aligned to proper hugepage size in caller side, |
| 1313 | * otherwise hugetlb_reserve_pages reserves one less hugepages than intended. |
| 1314 | */ |
| 1315 | struct file *hugetlb_file_setup(const char *name, size_t size, |
| 1316 | vm_flags_t acctflag, struct user_struct **user, |
Andi Kleen | 42d7395 | 2012-12-11 16:01:34 -0800 | [diff] [blame] | 1317 | int creat_flags, int page_size_log) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1318 | { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1319 | struct inode *inode; |
Al Viro | e68375c | 2018-06-09 09:50:46 -0400 | [diff] [blame] | 1320 | struct vfsmount *mnt; |
Andi Kleen | 42d7395 | 2012-12-11 16:01:34 -0800 | [diff] [blame] | 1321 | int hstate_idx; |
Al Viro | e68375c | 2018-06-09 09:50:46 -0400 | [diff] [blame] | 1322 | struct file *file; |
Andi Kleen | 42d7395 | 2012-12-11 16:01:34 -0800 | [diff] [blame] | 1323 | |
| 1324 | hstate_idx = get_hstate_idx(page_size_log); |
| 1325 | if (hstate_idx < 0) |
| 1326 | return ERR_PTR(-ENODEV); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1327 | |
Hugh Dickins | 353d5c3 | 2009-08-24 16:30:28 +0100 | [diff] [blame] | 1328 | *user = NULL; |
Al Viro | e68375c | 2018-06-09 09:50:46 -0400 | [diff] [blame] | 1329 | mnt = hugetlbfs_vfsmount[hstate_idx]; |
| 1330 | if (!mnt) |
Akinobu Mita | 5bc9859 | 2007-05-06 14:50:18 -0700 | [diff] [blame] | 1331 | return ERR_PTR(-ENOENT); |
| 1332 | |
From: Mel Gorman | ef1ff6b | 2009-09-23 15:56:05 -0700 | [diff] [blame] | 1333 | if (creat_flags == HUGETLB_SHMFS_INODE && !can_do_hugetlb_shm()) { |
Hugh Dickins | 353d5c3 | 2009-08-24 16:30:28 +0100 | [diff] [blame] | 1334 | *user = current_user(); |
| 1335 | if (user_shm_lock(size, *user)) { |
David Rientjes | 21a3c27 | 2012-03-21 16:34:13 -0700 | [diff] [blame] | 1336 | task_lock(current); |
Andrew Morton | 9b857d2 | 2014-06-04 16:07:21 -0700 | [diff] [blame] | 1337 | pr_warn_once("%s (%d): Using mlock ulimits for SHM_HUGETLB is deprecated\n", |
David Rientjes | 21a3c27 | 2012-03-21 16:34:13 -0700 | [diff] [blame] | 1338 | current->comm, current->pid); |
| 1339 | task_unlock(current); |
Hugh Dickins | 353d5c3 | 2009-08-24 16:30:28 +0100 | [diff] [blame] | 1340 | } else { |
| 1341 | *user = NULL; |
Ravikiran G Thirumalai | 2584e51 | 2009-03-31 15:21:26 -0700 | [diff] [blame] | 1342 | return ERR_PTR(-EPERM); |
Hugh Dickins | 353d5c3 | 2009-08-24 16:30:28 +0100 | [diff] [blame] | 1343 | } |
Ravikiran G Thirumalai | 2584e51 | 2009-03-31 15:21:26 -0700 | [diff] [blame] | 1344 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1345 | |
Anatol Pomozov | 39b6525 | 2012-09-12 20:11:55 -0700 | [diff] [blame] | 1346 | file = ERR_PTR(-ENOSPC); |
Al Viro | e68375c | 2018-06-09 09:50:46 -0400 | [diff] [blame] | 1347 | inode = hugetlbfs_get_inode(mnt->mnt_sb, NULL, S_IFREG | S_IRWXUGO, 0); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1348 | if (!inode) |
Al Viro | e68375c | 2018-06-09 09:50:46 -0400 | [diff] [blame] | 1349 | goto out; |
Stephen Smalley | e1832f2 | 2015-08-06 15:46:55 -0700 | [diff] [blame] | 1350 | if (creat_flags == HUGETLB_SHMFS_INODE) |
| 1351 | inode->i_flags |= S_PRIVATE; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1352 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1353 | inode->i_size = size; |
Miklos Szeredi | 6d6b77f | 2011-10-28 14:13:28 +0200 | [diff] [blame] | 1354 | clear_nlink(inode); |
Dave Hansen | ce8d2cd | 2007-10-16 23:31:13 -0700 | [diff] [blame] | 1355 | |
Al Viro | e68375c | 2018-06-09 09:50:46 -0400 | [diff] [blame] | 1356 | if (hugetlb_reserve_pages(inode, 0, |
| 1357 | size >> huge_page_shift(hstate_inode(inode)), NULL, |
| 1358 | acctflag)) |
| 1359 | file = ERR_PTR(-ENOMEM); |
| 1360 | else |
| 1361 | file = alloc_file_pseudo(inode, mnt, name, O_RDWR, |
| 1362 | &hugetlbfs_file_operations); |
| 1363 | if (!IS_ERR(file)) |
| 1364 | return file; |
Dave Hansen | ce8d2cd | 2007-10-16 23:31:13 -0700 | [diff] [blame] | 1365 | |
David Gibson | b45b5bd | 2006-03-22 00:08:55 -0800 | [diff] [blame] | 1366 | iput(inode); |
Al Viro | e68375c | 2018-06-09 09:50:46 -0400 | [diff] [blame] | 1367 | out: |
Hugh Dickins | 353d5c3 | 2009-08-24 16:30:28 +0100 | [diff] [blame] | 1368 | if (*user) { |
| 1369 | user_shm_unlock(size, *user); |
| 1370 | *user = NULL; |
| 1371 | } |
Anatol Pomozov | 39b6525 | 2012-09-12 20:11:55 -0700 | [diff] [blame] | 1372 | return file; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1373 | } |
| 1374 | |
| 1375 | static int __init init_hugetlbfs_fs(void) |
| 1376 | { |
Andi Kleen | 42d7395 | 2012-12-11 16:01:34 -0800 | [diff] [blame] | 1377 | struct hstate *h; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1378 | int error; |
Andi Kleen | 42d7395 | 2012-12-11 16:01:34 -0800 | [diff] [blame] | 1379 | int i; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1380 | |
Nishanth Aravamudan | 457c1b2 | 2014-05-06 12:50:00 -0700 | [diff] [blame] | 1381 | if (!hugepages_supported()) { |
Andrew Morton | 9b857d2 | 2014-06-04 16:07:21 -0700 | [diff] [blame] | 1382 | pr_info("disabling because there are no supported hugepage sizes\n"); |
Nishanth Aravamudan | 457c1b2 | 2014-05-06 12:50:00 -0700 | [diff] [blame] | 1383 | return -ENOTSUPP; |
| 1384 | } |
| 1385 | |
Hillf Danton | d1d5e05ff | 2012-03-21 16:34:15 -0700 | [diff] [blame] | 1386 | error = -ENOMEM; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1387 | hugetlbfs_inode_cachep = kmem_cache_create("hugetlbfs_inode_cache", |
| 1388 | sizeof(struct hugetlbfs_inode_info), |
Vladimir Davydov | 5d09705 | 2016-01-14 15:18:21 -0800 | [diff] [blame] | 1389 | 0, SLAB_ACCOUNT, init_once); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1390 | if (hugetlbfs_inode_cachep == NULL) |
Peter Zijlstra | e0bf68d | 2007-10-16 23:25:46 -0700 | [diff] [blame] | 1391 | goto out2; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1392 | |
| 1393 | error = register_filesystem(&hugetlbfs_fs_type); |
| 1394 | if (error) |
| 1395 | goto out; |
| 1396 | |
Andi Kleen | 42d7395 | 2012-12-11 16:01:34 -0800 | [diff] [blame] | 1397 | i = 0; |
| 1398 | for_each_hstate(h) { |
| 1399 | char buf[50]; |
| 1400 | unsigned ps_kb = 1U << (h->order + PAGE_SHIFT - 10); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1401 | |
Andi Kleen | 42d7395 | 2012-12-11 16:01:34 -0800 | [diff] [blame] | 1402 | snprintf(buf, sizeof(buf), "pagesize=%uK", ps_kb); |
| 1403 | hugetlbfs_vfsmount[i] = kern_mount_data(&hugetlbfs_fs_type, |
| 1404 | buf); |
| 1405 | |
| 1406 | if (IS_ERR(hugetlbfs_vfsmount[i])) { |
Andrew Morton | 9b857d2 | 2014-06-04 16:07:21 -0700 | [diff] [blame] | 1407 | pr_err("Cannot mount internal hugetlbfs for " |
Andi Kleen | 42d7395 | 2012-12-11 16:01:34 -0800 | [diff] [blame] | 1408 | "page size %uK", ps_kb); |
| 1409 | error = PTR_ERR(hugetlbfs_vfsmount[i]); |
| 1410 | hugetlbfs_vfsmount[i] = NULL; |
| 1411 | } |
| 1412 | i++; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1413 | } |
Andi Kleen | 42d7395 | 2012-12-11 16:01:34 -0800 | [diff] [blame] | 1414 | /* Non default hstates are optional */ |
| 1415 | if (!IS_ERR_OR_NULL(hugetlbfs_vfsmount[default_hstate_idx])) |
| 1416 | return 0; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1417 | |
| 1418 | out: |
Hillf Danton | d1d5e05ff | 2012-03-21 16:34:15 -0700 | [diff] [blame] | 1419 | kmem_cache_destroy(hugetlbfs_inode_cachep); |
Peter Zijlstra | e0bf68d | 2007-10-16 23:25:46 -0700 | [diff] [blame] | 1420 | out2: |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1421 | return error; |
| 1422 | } |
Paul Gortmaker | 3e89e1c | 2016-01-14 15:21:52 -0800 | [diff] [blame] | 1423 | fs_initcall(init_hugetlbfs_fs) |