blob: 4ea71eba40a57a3dd1c2e32460021878e12f56d3 [file] [log] [blame]
Linus Torvalds1da177e2005-04-16 15:20:36 -07001/*
2 * hugetlbpage-backed filesystem. Based on ramfs.
3 *
Nadia Yvette Chambers6d49e352012-12-06 10:39:54 +01004 * Nadia Yvette Chambers, 2002
Linus Torvalds1da177e2005-04-16 15:20:36 -07005 *
6 * Copyright (C) 2002 Linus Torvalds.
Paul Gortmaker3e89e1c2016-01-14 15:21:52 -08007 * License: GPL
Linus Torvalds1da177e2005-04-16 15:20:36 -07008 */
9
Andrew Morton9b857d22014-06-04 16:07:21 -070010#define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
11
Linus Torvalds1da177e2005-04-16 15:20:36 -070012#include <linux/thread_info.h>
13#include <asm/current.h>
14#include <linux/sched.h> /* remove ASAP */
Mike Kravetz70c35472015-09-08 15:01:54 -070015#include <linux/falloc.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070016#include <linux/fs.h>
17#include <linux/mount.h>
18#include <linux/file.h>
Randy Dunlape73a75f2007-07-15 23:40:52 -070019#include <linux/kernel.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070020#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 Dunlap16f7e0f2006-01-11 12:17:46 -080025#include <linux/capability.h>
Randy Dunlape73a75f2007-07-15 23:40:52 -070026#include <linux/ctype.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070027#include <linux/backing-dev.h>
28#include <linux/hugetlb.h>
29#include <linux/pagevec.h>
Randy Dunlape73a75f2007-07-15 23:40:52 -070030#include <linux/parser.h>
Benjamin Herrenschmidt036e0852007-05-06 14:50:12 -070031#include <linux/mman.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070032#include <linux/slab.h>
33#include <linux/dnotify.h>
34#include <linux/statfs.h>
35#include <linux/security.h>
Nick Black1fd7317d2009-09-22 16:43:33 -070036#include <linux/magic.h>
Naoya Horiguchi290408d2010-09-08 10:19:35 +090037#include <linux/migrate.h>
Al Viro34d06402015-04-03 11:31:35 -040038#include <linux/uio.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070039
40#include <asm/uaccess.h>
41
Josef 'Jeff' Sipekee9b6d62007-02-12 00:55:41 -080042static const struct super_operations hugetlbfs_ops;
Christoph Hellwigf5e54d62006-06-28 04:26:44 -070043static const struct address_space_operations hugetlbfs_aops;
Arjan van de Ven4b6f5d22006-03-28 01:56:42 -080044const struct file_operations hugetlbfs_file_operations;
Arjan van de Ven92e1d5b2007-02-12 00:55:39 -080045static const struct inode_operations hugetlbfs_dir_inode_operations;
46static const struct inode_operations hugetlbfs_inode_operations;
Linus Torvalds1da177e2005-04-16 15:20:36 -070047
David Gibsona1d776e2012-03-21 16:34:12 -070048struct hugetlbfs_config {
Eric W. Biedermana0eb3a02012-02-07 16:19:25 -080049 kuid_t uid;
50 kgid_t gid;
David Gibsona1d776e2012-03-21 16:34:12 -070051 umode_t mode;
Mike Kravetz7ca02d02015-04-15 16:13:42 -070052 long max_hpages;
David Gibsona1d776e2012-03-21 16:34:12 -070053 long nr_inodes;
54 struct hstate *hstate;
Mike Kravetz7ca02d02015-04-15 16:13:42 -070055 long min_hpages;
David Gibsona1d776e2012-03-21 16:34:12 -070056};
57
58struct hugetlbfs_inode_info {
59 struct shared_policy policy;
60 struct inode vfs_inode;
61};
62
63static inline struct hugetlbfs_inode_info *HUGETLBFS_I(struct inode *inode)
64{
65 return container_of(inode, struct hugetlbfs_inode_info, vfs_inode);
66}
67
Linus Torvalds1da177e2005-04-16 15:20:36 -070068int sysctl_hugetlb_shm_group;
69
Randy Dunlape73a75f2007-07-15 23:40:52 -070070enum {
71 Opt_size, Opt_nr_inodes,
72 Opt_mode, Opt_uid, Opt_gid,
Mike Kravetz7ca02d02015-04-15 16:13:42 -070073 Opt_pagesize, Opt_min_size,
Randy Dunlape73a75f2007-07-15 23:40:52 -070074 Opt_err,
75};
76
Steven Whitehousea447c092008-10-13 10:46:57 +010077static const match_table_t tokens = {
Randy Dunlape73a75f2007-07-15 23:40:52 -070078 {Opt_size, "size=%s"},
79 {Opt_nr_inodes, "nr_inodes=%s"},
80 {Opt_mode, "mode=%o"},
81 {Opt_uid, "uid=%u"},
82 {Opt_gid, "gid=%u"},
Andi Kleena137e1c2008-07-23 21:27:43 -070083 {Opt_pagesize, "pagesize=%s"},
Mike Kravetz7ca02d02015-04-15 16:13:42 -070084 {Opt_min_size, "min_size=%s"},
Randy Dunlape73a75f2007-07-15 23:40:52 -070085 {Opt_err, NULL},
86};
87
Mike Kravetz70c35472015-09-08 15:01:54 -070088#ifdef CONFIG_NUMA
89static inline void hugetlb_set_vma_policy(struct vm_area_struct *vma,
90 struct inode *inode, pgoff_t index)
91{
92 vma->vm_policy = mpol_shared_policy_lookup(&HUGETLBFS_I(inode)->policy,
93 index);
94}
95
96static inline void hugetlb_drop_vma_policy(struct vm_area_struct *vma)
97{
98 mpol_cond_put(vma->vm_policy);
99}
100#else
101static inline void hugetlb_set_vma_policy(struct vm_area_struct *vma,
102 struct inode *inode, pgoff_t index)
103{
104}
105
106static inline void hugetlb_drop_vma_policy(struct vm_area_struct *vma)
107{
108}
109#endif
110
Adam Litke2e9b3672005-10-29 18:16:47 -0700111static void huge_pagevec_release(struct pagevec *pvec)
112{
113 int i;
114
115 for (i = 0; i < pagevec_count(pvec); ++i)
116 put_page(pvec->pages[i]);
117
118 pagevec_reinit(pvec);
119}
120
Linus Torvalds1da177e2005-04-16 15:20:36 -0700121static int hugetlbfs_file_mmap(struct file *file, struct vm_area_struct *vma)
122{
Al Viro496ad9a2013-01-23 17:07:38 -0500123 struct inode *inode = file_inode(file);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700124 loff_t len, vma_len;
125 int ret;
Andi Kleena5516432008-07-23 21:27:41 -0700126 struct hstate *h = hstate_file(file);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700127
Hugh Dickins68589bc2006-11-14 02:03:32 -0800128 /*
David Gibsondec4ad82007-08-30 23:56:40 -0700129 * 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 Dickins68589bc2006-11-14 02:03:32 -0800135 */
Naoya Horiguchia2fce912013-04-17 15:58:27 -0700136 vma->vm_flags |= VM_HUGETLB | VM_DONTEXPAND;
Hugh Dickins68589bc2006-11-14 02:03:32 -0800137 vma->vm_ops = &hugetlb_vm_ops;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700138
Becky Bruce2b37c352011-07-25 17:11:49 -0700139 if (vma->vm_pgoff & (~huge_page_mask(h) >> PAGE_SHIFT))
David Gibsondec4ad82007-08-30 23:56:40 -0700140 return -EINVAL;
141
Linus Torvalds1da177e2005-04-16 15:20:36 -0700142 vma_len = (loff_t)(vma->vm_end - vma->vm_start);
143
Al Viro59551022016-01-22 15:40:57 -0500144 inode_lock(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700145 file_accessed(file);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700146
147 ret = -ENOMEM;
148 len = vma_len + ((loff_t)vma->vm_pgoff << PAGE_SHIFT);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700149
Mel Gormana1e78772008-07-23 21:27:23 -0700150 if (hugetlb_reserve_pages(inode,
Andi Kleena5516432008-07-23 21:27:41 -0700151 vma->vm_pgoff >> huge_page_order(h),
Mel Gorman5a6fe122009-02-10 14:02:27 +0000152 len >> huge_page_shift(h), vma,
153 vma->vm_flags))
Chen, Kenneth Wa43a8c32006-06-23 02:03:15 -0700154 goto out;
David Gibsonb45b5bd2006-03-22 00:08:55 -0800155
Adam Litke4c887262005-10-29 18:16:46 -0700156 ret = 0;
Zhang, Yanminb6174df2006-07-10 04:44:49 -0700157 if (vma->vm_flags & VM_WRITE && inode->i_size < len)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700158 inode->i_size = len;
159out:
Al Viro59551022016-01-22 15:40:57 -0500160 inode_unlock(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700161
162 return ret;
163}
164
165/*
Hugh Dickins508034a2005-10-29 18:16:30 -0700166 * Called under down_write(mmap_sem).
Linus Torvalds1da177e2005-04-16 15:20:36 -0700167 */
168
Adrian Bunkd2ba27e82007-05-06 14:49:00 -0700169#ifndef HAVE_ARCH_HUGETLB_UNMAPPED_AREA
Linus Torvalds1da177e2005-04-16 15:20:36 -0700170static unsigned long
171hugetlb_get_unmapped_area(struct file *file, unsigned long addr,
172 unsigned long len, unsigned long pgoff, unsigned long flags)
173{
174 struct mm_struct *mm = current->mm;
175 struct vm_area_struct *vma;
Andi Kleena5516432008-07-23 21:27:41 -0700176 struct hstate *h = hstate_file(file);
Michel Lespinasse08659352012-12-11 16:02:00 -0800177 struct vm_unmapped_area_info info;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700178
Andi Kleena5516432008-07-23 21:27:41 -0700179 if (len & ~huge_page_mask(h))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700180 return -EINVAL;
181 if (len > TASK_SIZE)
182 return -ENOMEM;
183
Benjamin Herrenschmidt036e0852007-05-06 14:50:12 -0700184 if (flags & MAP_FIXED) {
Andi Kleena5516432008-07-23 21:27:41 -0700185 if (prepare_hugepage_range(file, addr, len))
Benjamin Herrenschmidt036e0852007-05-06 14:50:12 -0700186 return -EINVAL;
187 return addr;
188 }
189
Linus Torvalds1da177e2005-04-16 15:20:36 -0700190 if (addr) {
Andi Kleena5516432008-07-23 21:27:41 -0700191 addr = ALIGN(addr, huge_page_size(h));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700192 vma = find_vma(mm, addr);
193 if (TASK_SIZE - len >= addr &&
194 (!vma || addr + len <= vma->vm_start))
195 return addr;
196 }
197
Michel Lespinasse08659352012-12-11 16:02:00 -0800198 info.flags = 0;
199 info.length = len;
200 info.low_limit = TASK_UNMAPPED_BASE;
201 info.high_limit = TASK_SIZE;
202 info.align_mask = PAGE_MASK & ~huge_page_mask(h);
203 info.align_offset = 0;
204 return vm_unmapped_area(&info);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700205}
206#endif
207
Al Viro34d06402015-04-03 11:31:35 -0400208static size_t
Badari Pulavartye63e1e52007-10-16 01:26:22 -0700209hugetlbfs_read_actor(struct page *page, unsigned long offset,
Al Viro34d06402015-04-03 11:31:35 -0400210 struct iov_iter *to, unsigned long size)
Badari Pulavartye63e1e52007-10-16 01:26:22 -0700211{
Al Viro34d06402015-04-03 11:31:35 -0400212 size_t copied = 0;
Badari Pulavartye63e1e52007-10-16 01:26:22 -0700213 int i, chunksize;
214
Badari Pulavartye63e1e52007-10-16 01:26:22 -0700215 /* Find which 4k chunk and offset with in that chunk */
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +0300216 i = offset >> PAGE_SHIFT;
217 offset = offset & ~PAGE_MASK;
Badari Pulavartye63e1e52007-10-16 01:26:22 -0700218
219 while (size) {
Al Viro34d06402015-04-03 11:31:35 -0400220 size_t n;
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +0300221 chunksize = PAGE_SIZE;
Badari Pulavartye63e1e52007-10-16 01:26:22 -0700222 if (offset)
223 chunksize -= offset;
224 if (chunksize > size)
225 chunksize = size;
Al Viro34d06402015-04-03 11:31:35 -0400226 n = copy_page_to_iter(&page[i], offset, chunksize, to);
227 copied += n;
228 if (n != chunksize)
229 return copied;
Badari Pulavartye63e1e52007-10-16 01:26:22 -0700230 offset = 0;
231 size -= chunksize;
Badari Pulavartye63e1e52007-10-16 01:26:22 -0700232 i++;
233 }
Al Viro34d06402015-04-03 11:31:35 -0400234 return copied;
Badari Pulavartye63e1e52007-10-16 01:26:22 -0700235}
236
237/*
238 * Support for read() - Find the page attached to f_mapping and copy out the
239 * data. Its *very* similar to do_generic_mapping_read(), we can't use that
Kirill A. Shutemovea1754a2016-04-01 15:29:48 +0300240 * since it has PAGE_SIZE assumptions.
Badari Pulavartye63e1e52007-10-16 01:26:22 -0700241 */
Al Viro34d06402015-04-03 11:31:35 -0400242static ssize_t hugetlbfs_read_iter(struct kiocb *iocb, struct iov_iter *to)
Badari Pulavartye63e1e52007-10-16 01:26:22 -0700243{
Al Viro34d06402015-04-03 11:31:35 -0400244 struct file *file = iocb->ki_filp;
245 struct hstate *h = hstate_file(file);
246 struct address_space *mapping = file->f_mapping;
Badari Pulavartye63e1e52007-10-16 01:26:22 -0700247 struct inode *inode = mapping->host;
Al Viro34d06402015-04-03 11:31:35 -0400248 unsigned long index = iocb->ki_pos >> huge_page_shift(h);
249 unsigned long offset = iocb->ki_pos & ~huge_page_mask(h);
Badari Pulavartye63e1e52007-10-16 01:26:22 -0700250 unsigned long end_index;
251 loff_t isize;
252 ssize_t retval = 0;
253
Al Viro34d06402015-04-03 11:31:35 -0400254 while (iov_iter_count(to)) {
Badari Pulavartye63e1e52007-10-16 01:26:22 -0700255 struct page *page;
Al Viro34d06402015-04-03 11:31:35 -0400256 size_t nr, copied;
Badari Pulavartye63e1e52007-10-16 01:26:22 -0700257
258 /* nr is the maximum number of bytes to copy from this page */
Andi Kleena5516432008-07-23 21:27:41 -0700259 nr = huge_page_size(h);
Aneesh Kumar K.Va05b0852012-03-21 16:34:08 -0700260 isize = i_size_read(inode);
261 if (!isize)
Al Viro34d06402015-04-03 11:31:35 -0400262 break;
Aneesh Kumar K.Va05b0852012-03-21 16:34:08 -0700263 end_index = (isize - 1) >> huge_page_shift(h);
Al Viro34d06402015-04-03 11:31:35 -0400264 if (index > end_index)
265 break;
266 if (index == end_index) {
Andi Kleena5516432008-07-23 21:27:41 -0700267 nr = ((isize - 1) & ~huge_page_mask(h)) + 1;
Aneesh Kumar K.Va05b0852012-03-21 16:34:08 -0700268 if (nr <= offset)
Al Viro34d06402015-04-03 11:31:35 -0400269 break;
Badari Pulavartye63e1e52007-10-16 01:26:22 -0700270 }
271 nr = nr - offset;
272
273 /* Find the page */
Aneesh Kumar K.Va05b0852012-03-21 16:34:08 -0700274 page = find_lock_page(mapping, index);
Badari Pulavartye63e1e52007-10-16 01:26:22 -0700275 if (unlikely(page == NULL)) {
276 /*
277 * We have a HOLE, zero out the user-buffer for the
278 * length of the hole or request.
279 */
Al Viro34d06402015-04-03 11:31:35 -0400280 copied = iov_iter_zero(nr, to);
Badari Pulavartye63e1e52007-10-16 01:26:22 -0700281 } else {
Aneesh Kumar K.Va05b0852012-03-21 16:34:08 -0700282 unlock_page(page);
283
Badari Pulavartye63e1e52007-10-16 01:26:22 -0700284 /*
285 * We have the page, copy it to user space buffer.
286 */
Al Viro34d06402015-04-03 11:31:35 -0400287 copied = hugetlbfs_read_actor(page, offset, to, nr);
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +0300288 put_page(page);
Badari Pulavartye63e1e52007-10-16 01:26:22 -0700289 }
Al Viro34d06402015-04-03 11:31:35 -0400290 offset += copied;
291 retval += copied;
292 if (copied != nr && iov_iter_count(to)) {
293 if (!retval)
294 retval = -EFAULT;
295 break;
Badari Pulavartye63e1e52007-10-16 01:26:22 -0700296 }
Andi Kleena5516432008-07-23 21:27:41 -0700297 index += offset >> huge_page_shift(h);
298 offset &= ~huge_page_mask(h);
Badari Pulavartye63e1e52007-10-16 01:26:22 -0700299 }
Al Viro34d06402015-04-03 11:31:35 -0400300 iocb->ki_pos = ((loff_t)index << huge_page_shift(h)) + offset;
Badari Pulavartye63e1e52007-10-16 01:26:22 -0700301 return retval;
302}
303
Nick Piggin800d15a2007-10-16 01:25:03 -0700304static int hugetlbfs_write_begin(struct file *file,
305 struct address_space *mapping,
306 loff_t pos, unsigned len, unsigned flags,
307 struct page **pagep, void **fsdata)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700308{
309 return -EINVAL;
310}
311
Nick Piggin800d15a2007-10-16 01:25:03 -0700312static int hugetlbfs_write_end(struct file *file, struct address_space *mapping,
313 loff_t pos, unsigned len, unsigned copied,
314 struct page *page, void *fsdata)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700315{
Nick Piggin800d15a2007-10-16 01:25:03 -0700316 BUG();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700317 return -EINVAL;
318}
319
Mike Kravetzb5cec282015-09-08 15:01:41 -0700320static void remove_huge_page(struct page *page)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700321{
Konstantin Khlebnikovb9ea2512015-04-14 15:45:27 -0700322 ClearPageDirty(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700323 ClearPageUptodate(page);
Minchan Kimbd65cb82011-03-22 16:30:54 -0700324 delete_from_page_cache(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700325}
326
Mike Kravetz4aae8d12016-01-15 16:57:40 -0800327static void
328hugetlb_vmdelete_list(struct rb_root *root, pgoff_t start, pgoff_t end)
329{
330 struct vm_area_struct *vma;
331
332 /*
333 * end == 0 indicates that the entire range after
334 * start should be unmapped.
335 */
336 vma_interval_tree_foreach(vma, root, start, end ? end : ULONG_MAX) {
337 unsigned long v_offset;
338 unsigned long v_end;
339
340 /*
341 * Can the expression below overflow on 32-bit arches?
342 * No, because the interval tree returns us only those vmas
343 * which overlap the truncated area starting at pgoff,
344 * and no vma on a 32-bit arch can span beyond the 4GB.
345 */
346 if (vma->vm_pgoff < start)
347 v_offset = (start - vma->vm_pgoff) << PAGE_SHIFT;
348 else
349 v_offset = 0;
350
351 if (!end)
352 v_end = vma->vm_end;
353 else {
354 v_end = ((end - vma->vm_pgoff) << PAGE_SHIFT)
355 + vma->vm_start;
356 if (v_end > vma->vm_end)
357 v_end = vma->vm_end;
358 }
359
360 unmap_hugepage_range(vma, vma->vm_start + v_offset, v_end,
361 NULL);
362 }
363}
Mike Kravetzb5cec282015-09-08 15:01:41 -0700364
365/*
366 * remove_inode_hugepages handles two distinct cases: truncation and hole
367 * punch. There are subtle differences in operation for each case.
Mike Kravetz4aae8d12016-01-15 16:57:40 -0800368 *
Mike Kravetzb5cec282015-09-08 15:01:41 -0700369 * truncation is indicated by end of range being LLONG_MAX
370 * In this case, we first scan the range and release found pages.
371 * After releasing pages, hugetlb_unreserve_pages cleans up region/reserv
Mike Kravetz18178892015-11-20 15:57:13 -0800372 * maps and global counts. Page faults can not race with truncation
373 * in this routine. hugetlb_no_page() prevents page faults in the
374 * truncated range. It checks i_size before allocation, and again after
375 * with the page table lock for the page held. The same lock must be
376 * acquired to unmap a page.
Mike Kravetzb5cec282015-09-08 15:01:41 -0700377 * hole punch is indicated if end is not LLONG_MAX
378 * In the hole punch case we scan the range and release found pages.
379 * Only when releasing a page is the associated region/reserv map
380 * deleted. The region/reserv map for ranges without associated
Mike Kravetz18178892015-11-20 15:57:13 -0800381 * pages are not modified. Page faults can race with hole punch.
382 * This is indicated if we find a mapped page.
Mike Kravetzb5cec282015-09-08 15:01:41 -0700383 * Note: If the passed end of range value is beyond the end of file, but
384 * not LLONG_MAX this routine still performs a hole punch operation.
385 */
386static void remove_inode_hugepages(struct inode *inode, loff_t lstart,
387 loff_t lend)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700388{
Andi Kleena5516432008-07-23 21:27:41 -0700389 struct hstate *h = hstate_inode(inode);
David Gibsonb45b5bd2006-03-22 00:08:55 -0800390 struct address_space *mapping = &inode->i_data;
Andi Kleena5516432008-07-23 21:27:41 -0700391 const pgoff_t start = lstart >> huge_page_shift(h);
Mike Kravetzb5cec282015-09-08 15:01:41 -0700392 const pgoff_t end = lend >> huge_page_shift(h);
393 struct vm_area_struct pseudo_vma;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700394 struct pagevec pvec;
395 pgoff_t next;
Chen, Kenneth Wa43a8c32006-06-23 02:03:15 -0700396 int i, freed = 0;
Mike Kravetzb5cec282015-09-08 15:01:41 -0700397 long lookup_nr = PAGEVEC_SIZE;
398 bool truncate_op = (lend == LLONG_MAX);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700399
Mike Kravetzb5cec282015-09-08 15:01:41 -0700400 memset(&pseudo_vma, 0, sizeof(struct vm_area_struct));
401 pseudo_vma.vm_flags = (VM_HUGETLB | VM_MAYSHARE | VM_SHARED);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700402 pagevec_init(&pvec, 0);
403 next = start;
Mike Kravetzb5cec282015-09-08 15:01:41 -0700404 while (next < end) {
405 /*
Mike Kravetz18178892015-11-20 15:57:13 -0800406 * Don't grab more pages than the number left in the range.
Mike Kravetzb5cec282015-09-08 15:01:41 -0700407 */
408 if (end - next < lookup_nr)
409 lookup_nr = end - next;
410
411 /*
Mike Kravetz18178892015-11-20 15:57:13 -0800412 * When no more pages are found, we are done.
Mike Kravetzb5cec282015-09-08 15:01:41 -0700413 */
Mike Kravetz18178892015-11-20 15:57:13 -0800414 if (!pagevec_lookup(&pvec, mapping, next, lookup_nr))
415 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700416
417 for (i = 0; i < pagevec_count(&pvec); ++i) {
418 struct page *page = pvec.pages[i];
Mike Kravetz4aae8d12016-01-15 16:57:40 -0800419 bool rsv_on_error;
Mike Kravetzb5cec282015-09-08 15:01:41 -0700420 u32 hash;
421
Mike Kravetz18178892015-11-20 15:57:13 -0800422 /*
423 * The page (index) could be beyond end. This is
424 * only possible in the punch hole case as end is
425 * max page offset in the truncate case.
426 */
427 next = page->index;
428 if (next >= end)
429 break;
430
Mike Kravetzb5cec282015-09-08 15:01:41 -0700431 hash = hugetlb_fault_mutex_hash(h, current->mm,
432 &pseudo_vma,
433 mapping, next, 0);
434 mutex_lock(&hugetlb_fault_mutex_table[hash]);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700435
Mike Kravetz4aae8d12016-01-15 16:57:40 -0800436 /*
437 * If page is mapped, it was faulted in after being
438 * unmapped in caller. Unmap (again) now after taking
439 * the fault mutex. The mutex will prevent faults
440 * until we finish removing the page.
441 *
442 * This race can only happen in the hole punch case.
443 * Getting here in a truncate operation is a bug.
444 */
445 if (unlikely(page_mapped(page))) {
Mike Kravetz18178892015-11-20 15:57:13 -0800446 BUG_ON(truncate_op);
Mike Kravetz4aae8d12016-01-15 16:57:40 -0800447
448 i_mmap_lock_write(mapping);
449 hugetlb_vmdelete_list(&mapping->i_mmap,
450 next * pages_per_huge_page(h),
451 (next + 1) * pages_per_huge_page(h));
452 i_mmap_unlock_write(mapping);
453 }
454
455 lock_page(page);
456 /*
457 * We must free the huge page and remove from page
458 * cache (remove_huge_page) BEFORE removing the
459 * region/reserve map (hugetlb_unreserve_pages). In
460 * rare out of memory conditions, removal of the
461 * region/reserve map could fail. Before free'ing
462 * the page, note PagePrivate which is used in case
463 * of error.
464 */
465 rsv_on_error = !PagePrivate(page);
466 remove_huge_page(page);
467 freed++;
468 if (!truncate_op) {
469 if (unlikely(hugetlb_unreserve_pages(inode,
470 next, next + 1, 1)))
471 hugetlb_fix_reserve_counts(inode,
472 rsv_on_error);
Mike Kravetzb5cec282015-09-08 15:01:41 -0700473 }
474
Linus Torvalds1da177e2005-04-16 15:20:36 -0700475 unlock_page(page);
Mike Kravetzb5cec282015-09-08 15:01:41 -0700476 mutex_unlock(&hugetlb_fault_mutex_table[hash]);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700477 }
Mike Kravetz18178892015-11-20 15:57:13 -0800478 ++next;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700479 huge_pagevec_release(&pvec);
Mike Kravetz18178892015-11-20 15:57:13 -0800480 cond_resched();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700481 }
Mike Kravetzb5cec282015-09-08 15:01:41 -0700482
483 if (truncate_op)
484 (void)hugetlb_unreserve_pages(inode, start, LONG_MAX, freed);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700485}
486
Al Viro2bbbda32010-06-04 19:52:12 -0400487static void hugetlbfs_evict_inode(struct inode *inode)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700488{
Joonsoo Kim9119a412014-04-03 14:47:25 -0700489 struct resv_map *resv_map;
490
Mike Kravetzb5cec282015-09-08 15:01:41 -0700491 remove_inode_hugepages(inode, 0, LLONG_MAX);
Joonsoo Kim9119a412014-04-03 14:47:25 -0700492 resv_map = (struct resv_map *)inode->i_mapping->private_data;
493 /* root inode doesn't have the resv_map, so we should check it */
494 if (resv_map)
495 resv_map_release(&resv_map->refs);
Jan Karadbd57682012-05-03 14:48:02 +0200496 clear_inode(inode);
Christoph Hellwig149f4212005-10-29 18:16:43 -0700497}
498
Linus Torvalds1da177e2005-04-16 15:20:36 -0700499static int hugetlb_vmtruncate(struct inode *inode, loff_t offset)
500{
Hugh Dickins856fc292006-10-28 10:38:43 -0700501 pgoff_t pgoff;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700502 struct address_space *mapping = inode->i_mapping;
Andi Kleena5516432008-07-23 21:27:41 -0700503 struct hstate *h = hstate_inode(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700504
Andi Kleena5516432008-07-23 21:27:41 -0700505 BUG_ON(offset & ~huge_page_mask(h));
Hugh Dickins856fc292006-10-28 10:38:43 -0700506 pgoff = offset >> PAGE_SHIFT;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700507
Ken Chen7aa91e12007-10-16 01:26:21 -0700508 i_size_write(inode, offset);
Davidlohr Bueso83cde9e2014-12-12 16:54:21 -0800509 i_mmap_lock_write(mapping);
Michel Lespinasse6b2dbba2012-10-08 16:31:25 -0700510 if (!RB_EMPTY_ROOT(&mapping->i_mmap))
Mike Kravetz1bfad992015-09-08 15:01:38 -0700511 hugetlb_vmdelete_list(&mapping->i_mmap, pgoff, 0);
Davidlohr Bueso83cde9e2014-12-12 16:54:21 -0800512 i_mmap_unlock_write(mapping);
Mike Kravetzb5cec282015-09-08 15:01:41 -0700513 remove_inode_hugepages(inode, offset, LLONG_MAX);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700514 return 0;
515}
516
Mike Kravetz70c35472015-09-08 15:01:54 -0700517static long hugetlbfs_punch_hole(struct inode *inode, loff_t offset, loff_t len)
518{
519 struct hstate *h = hstate_inode(inode);
520 loff_t hpage_size = huge_page_size(h);
521 loff_t hole_start, hole_end;
522
523 /*
524 * For hole punch round up the beginning offset of the hole and
525 * round down the end.
526 */
527 hole_start = round_up(offset, hpage_size);
528 hole_end = round_down(offset + len, hpage_size);
529
530 if (hole_end > hole_start) {
531 struct address_space *mapping = inode->i_mapping;
532
Al Viro59551022016-01-22 15:40:57 -0500533 inode_lock(inode);
Mike Kravetz70c35472015-09-08 15:01:54 -0700534 i_mmap_lock_write(mapping);
535 if (!RB_EMPTY_ROOT(&mapping->i_mmap))
536 hugetlb_vmdelete_list(&mapping->i_mmap,
537 hole_start >> PAGE_SHIFT,
538 hole_end >> PAGE_SHIFT);
539 i_mmap_unlock_write(mapping);
540 remove_inode_hugepages(inode, hole_start, hole_end);
Al Viro59551022016-01-22 15:40:57 -0500541 inode_unlock(inode);
Mike Kravetz70c35472015-09-08 15:01:54 -0700542 }
543
544 return 0;
545}
546
547static long hugetlbfs_fallocate(struct file *file, int mode, loff_t offset,
548 loff_t len)
549{
550 struct inode *inode = file_inode(file);
551 struct address_space *mapping = inode->i_mapping;
552 struct hstate *h = hstate_inode(inode);
553 struct vm_area_struct pseudo_vma;
554 struct mm_struct *mm = current->mm;
555 loff_t hpage_size = huge_page_size(h);
556 unsigned long hpage_shift = huge_page_shift(h);
557 pgoff_t start, index, end;
558 int error;
559 u32 hash;
560
561 if (mode & ~(FALLOC_FL_KEEP_SIZE | FALLOC_FL_PUNCH_HOLE))
562 return -EOPNOTSUPP;
563
564 if (mode & FALLOC_FL_PUNCH_HOLE)
565 return hugetlbfs_punch_hole(inode, offset, len);
566
567 /*
568 * Default preallocate case.
569 * For this range, start is rounded down and end is rounded up
570 * as well as being converted to page offsets.
571 */
572 start = offset >> hpage_shift;
573 end = (offset + len + hpage_size - 1) >> hpage_shift;
574
Al Viro59551022016-01-22 15:40:57 -0500575 inode_lock(inode);
Mike Kravetz70c35472015-09-08 15:01:54 -0700576
577 /* We need to check rlimit even when FALLOC_FL_KEEP_SIZE */
578 error = inode_newsize_ok(inode, offset + len);
579 if (error)
580 goto out;
581
582 /*
583 * Initialize a pseudo vma as this is required by the huge page
584 * allocation routines. If NUMA is configured, use page index
585 * as input to create an allocation policy.
586 */
587 memset(&pseudo_vma, 0, sizeof(struct vm_area_struct));
588 pseudo_vma.vm_flags = (VM_HUGETLB | VM_MAYSHARE | VM_SHARED);
589 pseudo_vma.vm_file = file;
590
591 for (index = start; index < end; index++) {
592 /*
593 * This is supposed to be the vaddr where the page is being
594 * faulted in, but we have no vaddr here.
595 */
596 struct page *page;
597 unsigned long addr;
598 int avoid_reserve = 0;
599
600 cond_resched();
601
602 /*
603 * fallocate(2) manpage permits EINTR; we may have been
604 * interrupted because we are using up too much memory.
605 */
606 if (signal_pending(current)) {
607 error = -EINTR;
608 break;
609 }
610
611 /* Set numa allocation policy based on index */
612 hugetlb_set_vma_policy(&pseudo_vma, inode, index);
613
614 /* addr is the offset within the file (zero based) */
615 addr = index * hpage_size;
616
617 /* mutex taken here, fault path and hole punch */
618 hash = hugetlb_fault_mutex_hash(h, mm, &pseudo_vma, mapping,
619 index, addr);
620 mutex_lock(&hugetlb_fault_mutex_table[hash]);
621
622 /* See if already present in mapping to avoid alloc/free */
623 page = find_get_page(mapping, index);
624 if (page) {
625 put_page(page);
626 mutex_unlock(&hugetlb_fault_mutex_table[hash]);
627 hugetlb_drop_vma_policy(&pseudo_vma);
628 continue;
629 }
630
631 /* Allocate page and add to page cache */
632 page = alloc_huge_page(&pseudo_vma, addr, avoid_reserve);
633 hugetlb_drop_vma_policy(&pseudo_vma);
634 if (IS_ERR(page)) {
635 mutex_unlock(&hugetlb_fault_mutex_table[hash]);
636 error = PTR_ERR(page);
637 goto out;
638 }
639 clear_huge_page(page, addr, pages_per_huge_page(h));
640 __SetPageUptodate(page);
641 error = huge_add_to_page_cache(page, mapping, index);
642 if (unlikely(error)) {
643 put_page(page);
644 mutex_unlock(&hugetlb_fault_mutex_table[hash]);
645 goto out;
646 }
647
648 mutex_unlock(&hugetlb_fault_mutex_table[hash]);
649
650 /*
651 * page_put due to reference from alloc_huge_page()
652 * unlock_page because locked by add_to_page_cache()
653 */
654 put_page(page);
655 unlock_page(page);
656 }
657
658 if (!(mode & FALLOC_FL_KEEP_SIZE) && offset + len > inode->i_size)
659 i_size_write(inode, offset + len);
660 inode->i_ctime = CURRENT_TIME;
Mike Kravetz70c35472015-09-08 15:01:54 -0700661out:
Al Viro59551022016-01-22 15:40:57 -0500662 inode_unlock(inode);
Mike Kravetz70c35472015-09-08 15:01:54 -0700663 return error;
664}
665
Linus Torvalds1da177e2005-04-16 15:20:36 -0700666static int hugetlbfs_setattr(struct dentry *dentry, struct iattr *attr)
667{
David Howells2b0143b2015-03-17 22:25:59 +0000668 struct inode *inode = d_inode(dentry);
Andi Kleena5516432008-07-23 21:27:41 -0700669 struct hstate *h = hstate_inode(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700670 int error;
671 unsigned int ia_valid = attr->ia_valid;
672
673 BUG_ON(!inode);
674
675 error = inode_change_ok(inode, attr);
676 if (error)
Christoph Hellwig10257742010-06-04 11:30:02 +0200677 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700678
679 if (ia_valid & ATTR_SIZE) {
680 error = -EINVAL;
Christoph Hellwig10257742010-06-04 11:30:02 +0200681 if (attr->ia_size & ~huge_page_mask(h))
682 return -EINVAL;
683 error = hugetlb_vmtruncate(inode, attr->ia_size);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700684 if (error)
Christoph Hellwig10257742010-06-04 11:30:02 +0200685 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700686 }
Christoph Hellwig10257742010-06-04 11:30:02 +0200687
688 setattr_copy(inode, attr);
689 mark_inode_dirty(inode);
690 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700691}
692
Al Viro7d54fa62011-07-24 20:20:48 -0400693static struct inode *hugetlbfs_get_root(struct super_block *sb,
694 struct hugetlbfs_config *config)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700695{
696 struct inode *inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700697
698 inode = new_inode(sb);
699 if (inode) {
700 struct hugetlbfs_inode_info *info;
Christoph Hellwig85fe4022010-10-23 11:19:54 -0400701 inode->i_ino = get_next_ino();
Al Viro7d54fa62011-07-24 20:20:48 -0400702 inode->i_mode = S_IFDIR | config->mode;
703 inode->i_uid = config->uid;
704 inode->i_gid = config->gid;
705 inode->i_atime = inode->i_mtime = inode->i_ctime = CURRENT_TIME;
706 info = HUGETLBFS_I(inode);
707 mpol_shared_policy_init(&info->policy, NULL);
708 inode->i_op = &hugetlbfs_dir_inode_operations;
709 inode->i_fop = &simple_dir_operations;
710 /* directory inodes start off with i_nlink == 2 (for "." entry) */
711 inc_nlink(inode);
Aneesh Kumar K.V65ed7602012-04-25 16:01:50 -0700712 lockdep_annotate_inode_mutex_key(inode);
Al Viro7d54fa62011-07-24 20:20:48 -0400713 }
714 return inode;
715}
716
Michal Hockob610ded2013-08-13 16:00:55 -0700717/*
Davidlohr Buesoc8c06ef2014-12-12 16:54:24 -0800718 * Hugetlbfs is not reclaimable; therefore its i_mmap_rwsem will never
Michal Hockob610ded2013-08-13 16:00:55 -0700719 * be taken from reclaim -- unlike regular filesystems. This needs an
Kirill A. Shutemov88f306b2016-01-15 16:57:31 -0800720 * annotation because huge_pmd_share() does an allocation under hugetlb's
Davidlohr Buesoc8c06ef2014-12-12 16:54:24 -0800721 * i_mmap_rwsem.
Michal Hockob610ded2013-08-13 16:00:55 -0700722 */
Davidlohr Buesoc8c06ef2014-12-12 16:54:24 -0800723static struct lock_class_key hugetlbfs_i_mmap_rwsem_key;
Michal Hockob610ded2013-08-13 16:00:55 -0700724
Al Viro7d54fa62011-07-24 20:20:48 -0400725static struct inode *hugetlbfs_get_inode(struct super_block *sb,
726 struct inode *dir,
Al Viro18df2252011-07-24 23:17:40 -0400727 umode_t mode, dev_t dev)
Al Viro7d54fa62011-07-24 20:20:48 -0400728{
729 struct inode *inode;
Joonsoo Kim9119a412014-04-03 14:47:25 -0700730 struct resv_map *resv_map;
731
732 resv_map = resv_map_alloc();
733 if (!resv_map)
734 return NULL;
Al Viro7d54fa62011-07-24 20:20:48 -0400735
736 inode = new_inode(sb);
737 if (inode) {
738 struct hugetlbfs_inode_info *info;
739 inode->i_ino = get_next_ino();
740 inode_init_owner(inode, dir, mode);
Davidlohr Buesoc8c06ef2014-12-12 16:54:24 -0800741 lockdep_set_class(&inode->i_mapping->i_mmap_rwsem,
742 &hugetlbfs_i_mmap_rwsem_key);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700743 inode->i_mapping->a_ops = &hugetlbfs_aops;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700744 inode->i_atime = inode->i_mtime = inode->i_ctime = CURRENT_TIME;
Joonsoo Kim9119a412014-04-03 14:47:25 -0700745 inode->i_mapping->private_data = resv_map;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700746 info = HUGETLBFS_I(inode);
Eric B Munson6bfde052009-09-21 17:03:43 -0700747 /*
748 * The policy is initialized here even if we are creating a
749 * private inode because initialization simply creates an
Nathan Zimmer4a8c7bb2016-01-14 15:18:36 -0800750 * an empty rb tree and calls rwlock_init(), later when we
Eric B Munson6bfde052009-09-21 17:03:43 -0700751 * call mpol_free_shared_policy() it will just return because
752 * the rb tree will still be empty.
753 */
Lee Schermerhorn71fe8042008-04-28 02:13:26 -0700754 mpol_shared_policy_init(&info->policy, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700755 switch (mode & S_IFMT) {
756 default:
757 init_special_inode(inode, mode, dev);
758 break;
759 case S_IFREG:
760 inode->i_op = &hugetlbfs_inode_operations;
761 inode->i_fop = &hugetlbfs_file_operations;
762 break;
763 case S_IFDIR:
764 inode->i_op = &hugetlbfs_dir_inode_operations;
765 inode->i_fop = &simple_dir_operations;
766
767 /* directory inodes start off with i_nlink == 2 (for "." entry) */
Dave Hansend8c76e62006-09-30 23:29:04 -0700768 inc_nlink(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700769 break;
770 case S_IFLNK:
771 inode->i_op = &page_symlink_inode_operations;
Al Viro21fc61c2015-11-17 01:07:57 -0500772 inode_nohighmem(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700773 break;
774 }
Josh Boyere096d0c2011-08-25 07:48:12 -0400775 lockdep_annotate_inode_mutex_key(inode);
Joonsoo Kim9119a412014-04-03 14:47:25 -0700776 } else
777 kref_put(&resv_map->refs, resv_map_release);
778
Linus Torvalds1da177e2005-04-16 15:20:36 -0700779 return inode;
780}
781
782/*
783 * File creation. Allocate an inode, and we're done..
784 */
785static int hugetlbfs_mknod(struct inode *dir,
Al Viro1a67aaf2011-07-26 01:52:52 -0400786 struct dentry *dentry, umode_t mode, dev_t dev)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700787{
788 struct inode *inode;
789 int error = -ENOSPC;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700790
Al Viro7d54fa62011-07-24 20:20:48 -0400791 inode = hugetlbfs_get_inode(dir->i_sb, dir, mode, dev);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700792 if (inode) {
793 dir->i_ctime = dir->i_mtime = CURRENT_TIME;
794 d_instantiate(dentry, inode);
795 dget(dentry); /* Extra count - pin the dentry in core */
796 error = 0;
797 }
798 return error;
799}
800
Al Viro18bb1db2011-07-26 01:41:39 -0400801static int hugetlbfs_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700802{
803 int retval = hugetlbfs_mknod(dir, dentry, mode | S_IFDIR, 0);
804 if (!retval)
Dave Hansend8c76e62006-09-30 23:29:04 -0700805 inc_nlink(dir);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700806 return retval;
807}
808
Al Viroebfc3b42012-06-10 18:05:36 -0400809static int hugetlbfs_create(struct inode *dir, struct dentry *dentry, umode_t mode, bool excl)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700810{
811 return hugetlbfs_mknod(dir, dentry, mode | S_IFREG, 0);
812}
813
814static int hugetlbfs_symlink(struct inode *dir,
815 struct dentry *dentry, const char *symname)
816{
817 struct inode *inode;
818 int error = -ENOSPC;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700819
Al Viro7d54fa62011-07-24 20:20:48 -0400820 inode = hugetlbfs_get_inode(dir->i_sb, dir, S_IFLNK|S_IRWXUGO, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700821 if (inode) {
822 int l = strlen(symname)+1;
823 error = page_symlink(inode, symname, l);
824 if (!error) {
825 d_instantiate(dentry, inode);
826 dget(dentry);
827 } else
828 iput(inode);
829 }
830 dir->i_ctime = dir->i_mtime = CURRENT_TIME;
831
832 return error;
833}
834
835/*
Ken Chen6649a382007-02-08 14:20:27 -0800836 * mark the head page dirty
Linus Torvalds1da177e2005-04-16 15:20:36 -0700837 */
838static int hugetlbfs_set_page_dirty(struct page *page)
839{
Christoph Lameterd85f3382007-05-06 14:49:39 -0700840 struct page *head = compound_head(page);
Ken Chen6649a382007-02-08 14:20:27 -0800841
842 SetPageDirty(head);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700843 return 0;
844}
845
Naoya Horiguchi290408d2010-09-08 10:19:35 +0900846static int hugetlbfs_migrate_page(struct address_space *mapping,
Mel Gormanb969c4a2012-01-12 17:19:34 -0800847 struct page *newpage, struct page *page,
Mel Gormana6bc32b2012-01-12 17:19:43 -0800848 enum migrate_mode mode)
Naoya Horiguchi290408d2010-09-08 10:19:35 +0900849{
850 int rc;
851
852 rc = migrate_huge_page_move_mapping(mapping, newpage, page);
Rafael Aquini78bd5202012-12-11 16:02:31 -0800853 if (rc != MIGRATEPAGE_SUCCESS)
Naoya Horiguchi290408d2010-09-08 10:19:35 +0900854 return rc;
855 migrate_page_copy(newpage, page);
856
Rafael Aquini78bd5202012-12-11 16:02:31 -0800857 return MIGRATEPAGE_SUCCESS;
Naoya Horiguchi290408d2010-09-08 10:19:35 +0900858}
859
David Howells726c3342006-06-23 02:02:58 -0700860static int hugetlbfs_statfs(struct dentry *dentry, struct kstatfs *buf)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700861{
David Howells726c3342006-06-23 02:02:58 -0700862 struct hugetlbfs_sb_info *sbinfo = HUGETLBFS_SB(dentry->d_sb);
David Howells2b0143b2015-03-17 22:25:59 +0000863 struct hstate *h = hstate_inode(d_inode(dentry));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700864
865 buf->f_type = HUGETLBFS_MAGIC;
Andi Kleena5516432008-07-23 21:27:41 -0700866 buf->f_bsize = huge_page_size(h);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700867 if (sbinfo) {
868 spin_lock(&sbinfo->stat_lock);
David Gibson74a8a652005-11-21 21:32:24 -0800869 /* If no limits set, just report 0 for max/free/used
870 * blocks, like simple_statfs() */
David Gibson90481622012-03-21 16:34:12 -0700871 if (sbinfo->spool) {
872 long free_pages;
873
874 spin_lock(&sbinfo->spool->lock);
875 buf->f_blocks = sbinfo->spool->max_hpages;
876 free_pages = sbinfo->spool->max_hpages
877 - sbinfo->spool->used_hpages;
878 buf->f_bavail = buf->f_bfree = free_pages;
879 spin_unlock(&sbinfo->spool->lock);
David Gibson74a8a652005-11-21 21:32:24 -0800880 buf->f_files = sbinfo->max_inodes;
881 buf->f_ffree = sbinfo->free_inodes;
882 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700883 spin_unlock(&sbinfo->stat_lock);
884 }
885 buf->f_namelen = NAME_MAX;
886 return 0;
887}
888
889static void hugetlbfs_put_super(struct super_block *sb)
890{
891 struct hugetlbfs_sb_info *sbi = HUGETLBFS_SB(sb);
892
893 if (sbi) {
894 sb->s_fs_info = NULL;
David Gibson90481622012-03-21 16:34:12 -0700895
896 if (sbi->spool)
897 hugepage_put_subpool(sbi->spool);
898
Linus Torvalds1da177e2005-04-16 15:20:36 -0700899 kfree(sbi);
900 }
901}
902
Christoph Hellwig96527982005-10-29 18:16:42 -0700903static inline int hugetlbfs_dec_free_inodes(struct hugetlbfs_sb_info *sbinfo)
904{
905 if (sbinfo->free_inodes >= 0) {
906 spin_lock(&sbinfo->stat_lock);
907 if (unlikely(!sbinfo->free_inodes)) {
908 spin_unlock(&sbinfo->stat_lock);
909 return 0;
910 }
911 sbinfo->free_inodes--;
912 spin_unlock(&sbinfo->stat_lock);
913 }
914
915 return 1;
916}
917
918static void hugetlbfs_inc_free_inodes(struct hugetlbfs_sb_info *sbinfo)
919{
920 if (sbinfo->free_inodes >= 0) {
921 spin_lock(&sbinfo->stat_lock);
922 sbinfo->free_inodes++;
923 spin_unlock(&sbinfo->stat_lock);
924 }
925}
926
927
Christoph Lametere18b8902006-12-06 20:33:20 -0800928static struct kmem_cache *hugetlbfs_inode_cachep;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700929
930static struct inode *hugetlbfs_alloc_inode(struct super_block *sb)
931{
Christoph Hellwig96527982005-10-29 18:16:42 -0700932 struct hugetlbfs_sb_info *sbinfo = HUGETLBFS_SB(sb);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700933 struct hugetlbfs_inode_info *p;
934
Christoph Hellwig96527982005-10-29 18:16:42 -0700935 if (unlikely(!hugetlbfs_dec_free_inodes(sbinfo)))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700936 return NULL;
Christoph Lametere94b1762006-12-06 20:33:17 -0800937 p = kmem_cache_alloc(hugetlbfs_inode_cachep, GFP_KERNEL);
Christoph Hellwig96527982005-10-29 18:16:42 -0700938 if (unlikely(!p)) {
939 hugetlbfs_inc_free_inodes(sbinfo);
940 return NULL;
941 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700942 return &p->vfs_inode;
943}
944
Nick Pigginfa0d7e3d2011-01-07 17:49:49 +1100945static void hugetlbfs_i_callback(struct rcu_head *head)
946{
947 struct inode *inode = container_of(head, struct inode, i_rcu);
Nick Pigginfa0d7e3d2011-01-07 17:49:49 +1100948 kmem_cache_free(hugetlbfs_inode_cachep, HUGETLBFS_I(inode));
949}
950
Linus Torvalds1da177e2005-04-16 15:20:36 -0700951static void hugetlbfs_destroy_inode(struct inode *inode)
952{
Christoph Hellwig96527982005-10-29 18:16:42 -0700953 hugetlbfs_inc_free_inodes(HUGETLBFS_SB(inode->i_sb));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700954 mpol_free_shared_policy(&HUGETLBFS_I(inode)->policy);
Nick Pigginfa0d7e3d2011-01-07 17:49:49 +1100955 call_rcu(&inode->i_rcu, hugetlbfs_i_callback);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700956}
957
Christoph Hellwigf5e54d62006-06-28 04:26:44 -0700958static const struct address_space_operations hugetlbfs_aops = {
Nick Piggin800d15a2007-10-16 01:25:03 -0700959 .write_begin = hugetlbfs_write_begin,
960 .write_end = hugetlbfs_write_end,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700961 .set_page_dirty = hugetlbfs_set_page_dirty,
Naoya Horiguchi290408d2010-09-08 10:19:35 +0900962 .migratepage = hugetlbfs_migrate_page,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700963};
964
Christoph Hellwig96527982005-10-29 18:16:42 -0700965
Alexey Dobriyan51cc5062008-07-25 19:45:34 -0700966static void init_once(void *foo)
Christoph Hellwig96527982005-10-29 18:16:42 -0700967{
968 struct hugetlbfs_inode_info *ei = (struct hugetlbfs_inode_info *)foo;
969
Christoph Lametera35afb82007-05-16 22:10:57 -0700970 inode_init_once(&ei->vfs_inode);
Christoph Hellwig96527982005-10-29 18:16:42 -0700971}
972
Arjan van de Ven4b6f5d22006-03-28 01:56:42 -0800973const struct file_operations hugetlbfs_file_operations = {
Al Viro34d06402015-04-03 11:31:35 -0400974 .read_iter = hugetlbfs_read_iter,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700975 .mmap = hugetlbfs_file_mmap,
Christoph Hellwig1b061d92010-05-26 17:53:41 +0200976 .fsync = noop_fsync,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700977 .get_unmapped_area = hugetlb_get_unmapped_area,
Mike Kravetz70c35472015-09-08 15:01:54 -0700978 .llseek = default_llseek,
979 .fallocate = hugetlbfs_fallocate,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700980};
981
Arjan van de Ven92e1d5b2007-02-12 00:55:39 -0800982static const struct inode_operations hugetlbfs_dir_inode_operations = {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700983 .create = hugetlbfs_create,
984 .lookup = simple_lookup,
985 .link = simple_link,
986 .unlink = simple_unlink,
987 .symlink = hugetlbfs_symlink,
988 .mkdir = hugetlbfs_mkdir,
989 .rmdir = simple_rmdir,
990 .mknod = hugetlbfs_mknod,
991 .rename = simple_rename,
992 .setattr = hugetlbfs_setattr,
993};
994
Arjan van de Ven92e1d5b2007-02-12 00:55:39 -0800995static const struct inode_operations hugetlbfs_inode_operations = {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700996 .setattr = hugetlbfs_setattr,
997};
998
Josef 'Jeff' Sipekee9b6d62007-02-12 00:55:41 -0800999static const struct super_operations hugetlbfs_ops = {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001000 .alloc_inode = hugetlbfs_alloc_inode,
1001 .destroy_inode = hugetlbfs_destroy_inode,
Al Viro2bbbda32010-06-04 19:52:12 -04001002 .evict_inode = hugetlbfs_evict_inode,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001003 .statfs = hugetlbfs_statfs,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001004 .put_super = hugetlbfs_put_super,
Miklos Szeredi10f19a82008-02-08 04:21:45 -08001005 .show_options = generic_show_options,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001006};
1007
Mike Kravetz7ca02d02015-04-15 16:13:42 -07001008enum { NO_SIZE, SIZE_STD, SIZE_PERCENT };
1009
1010/*
1011 * Convert size option passed from command line to number of huge pages
1012 * in the pool specified by hstate. Size option could be in bytes
1013 * (val_type == SIZE_STD) or percentage of the pool (val_type == SIZE_PERCENT).
1014 */
1015static long long
1016hugetlbfs_size_to_hpages(struct hstate *h, unsigned long long size_opt,
1017 int val_type)
1018{
1019 if (val_type == NO_SIZE)
1020 return -1;
1021
1022 if (val_type == SIZE_PERCENT) {
1023 size_opt <<= huge_page_shift(h);
1024 size_opt *= h->max_huge_pages;
1025 do_div(size_opt, 100);
1026 }
1027
1028 size_opt >>= huge_page_shift(h);
1029 return size_opt;
1030}
1031
Linus Torvalds1da177e2005-04-16 15:20:36 -07001032static int
1033hugetlbfs_parse_options(char *options, struct hugetlbfs_config *pconfig)
1034{
Randy Dunlape73a75f2007-07-15 23:40:52 -07001035 char *p, *rest;
1036 substring_t args[MAX_OPT_ARGS];
1037 int option;
Mike Kravetz7ca02d02015-04-15 16:13:42 -07001038 unsigned long long max_size_opt = 0, min_size_opt = 0;
1039 int max_val_type = NO_SIZE, min_val_type = NO_SIZE;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001040
1041 if (!options)
1042 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001043
Randy Dunlape73a75f2007-07-15 23:40:52 -07001044 while ((p = strsep(&options, ",")) != NULL) {
1045 int token;
Lee Schermerhornb4c07bc2007-07-15 23:40:54 -07001046 if (!*p)
1047 continue;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001048
Randy Dunlape73a75f2007-07-15 23:40:52 -07001049 token = match_token(p, tokens, args);
1050 switch (token) {
1051 case Opt_uid:
1052 if (match_int(&args[0], &option))
1053 goto bad_val;
Eric W. Biedermana0eb3a02012-02-07 16:19:25 -08001054 pconfig->uid = make_kuid(current_user_ns(), option);
1055 if (!uid_valid(pconfig->uid))
1056 goto bad_val;
Randy Dunlape73a75f2007-07-15 23:40:52 -07001057 break;
1058
1059 case Opt_gid:
1060 if (match_int(&args[0], &option))
1061 goto bad_val;
Eric W. Biedermana0eb3a02012-02-07 16:19:25 -08001062 pconfig->gid = make_kgid(current_user_ns(), option);
1063 if (!gid_valid(pconfig->gid))
1064 goto bad_val;
Randy Dunlape73a75f2007-07-15 23:40:52 -07001065 break;
1066
1067 case Opt_mode:
1068 if (match_octal(&args[0], &option))
1069 goto bad_val;
Ken Chen75897d62008-02-04 22:28:36 -08001070 pconfig->mode = option & 01777U;
Randy Dunlape73a75f2007-07-15 23:40:52 -07001071 break;
1072
1073 case Opt_size: {
Randy Dunlape73a75f2007-07-15 23:40:52 -07001074 /* memparse() will accept a K/M/G without a digit */
1075 if (!isdigit(*args[0].from))
1076 goto bad_val;
Mike Kravetz7ca02d02015-04-15 16:13:42 -07001077 max_size_opt = memparse(args[0].from, &rest);
1078 max_val_type = SIZE_STD;
Andi Kleena137e1c2008-07-23 21:27:43 -07001079 if (*rest == '%')
Mike Kravetz7ca02d02015-04-15 16:13:42 -07001080 max_val_type = SIZE_PERCENT;
Randy Dunlape73a75f2007-07-15 23:40:52 -07001081 break;
1082 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001083
Randy Dunlape73a75f2007-07-15 23:40:52 -07001084 case Opt_nr_inodes:
1085 /* memparse() will accept a K/M/G without a digit */
1086 if (!isdigit(*args[0].from))
1087 goto bad_val;
1088 pconfig->nr_inodes = memparse(args[0].from, &rest);
1089 break;
1090
Andi Kleena137e1c2008-07-23 21:27:43 -07001091 case Opt_pagesize: {
1092 unsigned long ps;
1093 ps = memparse(args[0].from, &rest);
1094 pconfig->hstate = size_to_hstate(ps);
1095 if (!pconfig->hstate) {
Andrew Morton9b857d22014-06-04 16:07:21 -07001096 pr_err("Unsupported page size %lu MB\n",
Andi Kleena137e1c2008-07-23 21:27:43 -07001097 ps >> 20);
1098 return -EINVAL;
1099 }
1100 break;
1101 }
1102
Mike Kravetz7ca02d02015-04-15 16:13:42 -07001103 case Opt_min_size: {
1104 /* memparse() will accept a K/M/G without a digit */
1105 if (!isdigit(*args[0].from))
1106 goto bad_val;
1107 min_size_opt = memparse(args[0].from, &rest);
1108 min_val_type = SIZE_STD;
1109 if (*rest == '%')
1110 min_val_type = SIZE_PERCENT;
1111 break;
1112 }
1113
Randy Dunlape73a75f2007-07-15 23:40:52 -07001114 default:
Andrew Morton9b857d22014-06-04 16:07:21 -07001115 pr_err("Bad mount option: \"%s\"\n", p);
Lee Schermerhornb4c07bc2007-07-15 23:40:54 -07001116 return -EINVAL;
Randy Dunlape73a75f2007-07-15 23:40:52 -07001117 break;
1118 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001119 }
Andi Kleena137e1c2008-07-23 21:27:43 -07001120
Mike Kravetz7ca02d02015-04-15 16:13:42 -07001121 /*
1122 * Use huge page pool size (in hstate) to convert the size
1123 * options to number of huge pages. If NO_SIZE, -1 is returned.
1124 */
1125 pconfig->max_hpages = hugetlbfs_size_to_hpages(pconfig->hstate,
1126 max_size_opt, max_val_type);
1127 pconfig->min_hpages = hugetlbfs_size_to_hpages(pconfig->hstate,
1128 min_size_opt, min_val_type);
1129
1130 /*
1131 * If max_size was specified, then min_size must be smaller
1132 */
1133 if (max_val_type > NO_SIZE &&
1134 pconfig->min_hpages > pconfig->max_hpages) {
1135 pr_err("minimum size can not be greater than maximum size\n");
1136 return -EINVAL;
Andi Kleena137e1c2008-07-23 21:27:43 -07001137 }
1138
Linus Torvalds1da177e2005-04-16 15:20:36 -07001139 return 0;
Randy Dunlape73a75f2007-07-15 23:40:52 -07001140
1141bad_val:
Andrew Morton9b857d22014-06-04 16:07:21 -07001142 pr_err("Bad value '%s' for mount option '%s'\n", args[0].from, p);
Akinobu Mitac12ddba2009-04-21 12:24:05 -07001143 return -EINVAL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001144}
1145
1146static int
1147hugetlbfs_fill_super(struct super_block *sb, void *data, int silent)
1148{
Linus Torvalds1da177e2005-04-16 15:20:36 -07001149 int ret;
1150 struct hugetlbfs_config config;
1151 struct hugetlbfs_sb_info *sbinfo;
1152
Miklos Szeredi10f19a82008-02-08 04:21:45 -08001153 save_mount_options(sb, data);
1154
Mike Kravetz7ca02d02015-04-15 16:13:42 -07001155 config.max_hpages = -1; /* No limit on size by default */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001156 config.nr_inodes = -1; /* No limit on number of inodes by default */
David Howells77c70de2008-11-14 10:38:56 +11001157 config.uid = current_fsuid();
1158 config.gid = current_fsgid();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001159 config.mode = 0755;
Andi Kleena137e1c2008-07-23 21:27:43 -07001160 config.hstate = &default_hstate;
Mike Kravetz7ca02d02015-04-15 16:13:42 -07001161 config.min_hpages = -1; /* No default minimum size */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001162 ret = hugetlbfs_parse_options(data, &config);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001163 if (ret)
1164 return ret;
1165
1166 sbinfo = kmalloc(sizeof(struct hugetlbfs_sb_info), GFP_KERNEL);
1167 if (!sbinfo)
1168 return -ENOMEM;
1169 sb->s_fs_info = sbinfo;
Andi Kleena137e1c2008-07-23 21:27:43 -07001170 sbinfo->hstate = config.hstate;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001171 spin_lock_init(&sbinfo->stat_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001172 sbinfo->max_inodes = config.nr_inodes;
1173 sbinfo->free_inodes = config.nr_inodes;
David Gibson90481622012-03-21 16:34:12 -07001174 sbinfo->spool = NULL;
Mike Kravetz7ca02d02015-04-15 16:13:42 -07001175 /*
1176 * Allocate and initialize subpool if maximum or minimum size is
1177 * specified. Any needed reservations (for minimim size) are taken
1178 * taken when the subpool is created.
1179 */
1180 if (config.max_hpages != -1 || config.min_hpages != -1) {
1181 sbinfo->spool = hugepage_new_subpool(config.hstate,
1182 config.max_hpages,
1183 config.min_hpages);
David Gibson90481622012-03-21 16:34:12 -07001184 if (!sbinfo->spool)
1185 goto out_free;
1186 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001187 sb->s_maxbytes = MAX_LFS_FILESIZE;
Andi Kleena137e1c2008-07-23 21:27:43 -07001188 sb->s_blocksize = huge_page_size(config.hstate);
1189 sb->s_blocksize_bits = huge_page_shift(config.hstate);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001190 sb->s_magic = HUGETLBFS_MAGIC;
1191 sb->s_op = &hugetlbfs_ops;
1192 sb->s_time_gran = 1;
Al Viro48fde702012-01-08 22:15:13 -05001193 sb->s_root = d_make_root(hugetlbfs_get_root(sb, &config));
1194 if (!sb->s_root)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001195 goto out_free;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001196 return 0;
1197out_free:
Fabian Frederick6e6870d2014-06-04 16:10:40 -07001198 kfree(sbinfo->spool);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001199 kfree(sbinfo);
1200 return -ENOMEM;
1201}
1202
Al Viro3c26ff62010-07-25 11:46:36 +04001203static struct dentry *hugetlbfs_mount(struct file_system_type *fs_type,
1204 int flags, const char *dev_name, void *data)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001205{
Al Viro3c26ff62010-07-25 11:46:36 +04001206 return mount_nodev(fs_type, flags, data, hugetlbfs_fill_super);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001207}
1208
1209static struct file_system_type hugetlbfs_fs_type = {
1210 .name = "hugetlbfs",
Al Viro3c26ff62010-07-25 11:46:36 +04001211 .mount = hugetlbfs_mount,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001212 .kill_sb = kill_litter_super,
1213};
1214
Andi Kleen42d73952012-12-11 16:01:34 -08001215static struct vfsmount *hugetlbfs_vfsmount[HUGE_MAX_HSTATE];
Linus Torvalds1da177e2005-04-16 15:20:36 -07001216
From: Mel Gormanef1ff6b2009-09-23 15:56:05 -07001217static int can_do_hugetlb_shm(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001218{
Eric W. Biedermana0eb3a02012-02-07 16:19:25 -08001219 kgid_t shm_group;
1220 shm_group = make_kgid(&init_user_ns, sysctl_hugetlb_shm_group);
1221 return capable(CAP_IPC_LOCK) || in_group_p(shm_group);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001222}
1223
Andi Kleen42d73952012-12-11 16:01:34 -08001224static int get_hstate_idx(int page_size_log)
1225{
Naoya Horiguchiaf73e4d2013-05-07 16:18:13 -07001226 struct hstate *h = hstate_sizelog(page_size_log);
Andi Kleen42d73952012-12-11 16:01:34 -08001227
Andi Kleen42d73952012-12-11 16:01:34 -08001228 if (!h)
1229 return -1;
1230 return h - hstates;
1231}
1232
Fabian Frederickbe1d2cf2014-06-04 16:10:39 -07001233static const struct dentry_operations anon_ops = {
Al Viro118b2302013-08-24 12:08:17 -04001234 .d_dname = simple_dname
Al Viro0df4d6e2013-02-14 22:39:53 -05001235};
1236
Naoya Horiguchiaf73e4d2013-05-07 16:18:13 -07001237/*
1238 * Note that size should be aligned to proper hugepage size in caller side,
1239 * otherwise hugetlb_reserve_pages reserves one less hugepages than intended.
1240 */
1241struct file *hugetlb_file_setup(const char *name, size_t size,
1242 vm_flags_t acctflag, struct user_struct **user,
Andi Kleen42d73952012-12-11 16:01:34 -08001243 int creat_flags, int page_size_log)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001244{
Anatol Pomozov39b65252012-09-12 20:11:55 -07001245 struct file *file = ERR_PTR(-ENOMEM);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001246 struct inode *inode;
Al Viro2c48b9c2009-08-09 00:52:35 +04001247 struct path path;
Al Viro0df4d6e2013-02-14 22:39:53 -05001248 struct super_block *sb;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001249 struct qstr quick_string;
Andi Kleen42d73952012-12-11 16:01:34 -08001250 int hstate_idx;
1251
1252 hstate_idx = get_hstate_idx(page_size_log);
1253 if (hstate_idx < 0)
1254 return ERR_PTR(-ENODEV);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001255
Hugh Dickins353d5c32009-08-24 16:30:28 +01001256 *user = NULL;
Andi Kleen42d73952012-12-11 16:01:34 -08001257 if (!hugetlbfs_vfsmount[hstate_idx])
Akinobu Mita5bc98592007-05-06 14:50:18 -07001258 return ERR_PTR(-ENOENT);
1259
From: Mel Gormanef1ff6b2009-09-23 15:56:05 -07001260 if (creat_flags == HUGETLB_SHMFS_INODE && !can_do_hugetlb_shm()) {
Hugh Dickins353d5c32009-08-24 16:30:28 +01001261 *user = current_user();
1262 if (user_shm_lock(size, *user)) {
David Rientjes21a3c272012-03-21 16:34:13 -07001263 task_lock(current);
Andrew Morton9b857d22014-06-04 16:07:21 -07001264 pr_warn_once("%s (%d): Using mlock ulimits for SHM_HUGETLB is deprecated\n",
David Rientjes21a3c272012-03-21 16:34:13 -07001265 current->comm, current->pid);
1266 task_unlock(current);
Hugh Dickins353d5c32009-08-24 16:30:28 +01001267 } else {
1268 *user = NULL;
Ravikiran G Thirumalai2584e512009-03-31 15:21:26 -07001269 return ERR_PTR(-EPERM);
Hugh Dickins353d5c32009-08-24 16:30:28 +01001270 }
Ravikiran G Thirumalai2584e512009-03-31 15:21:26 -07001271 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001272
Al Viro0df4d6e2013-02-14 22:39:53 -05001273 sb = hugetlbfs_vfsmount[hstate_idx]->mnt_sb;
Eric W. Biederman9d665862007-06-16 10:16:16 -07001274 quick_string.name = name;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001275 quick_string.len = strlen(quick_string.name);
1276 quick_string.hash = 0;
Al Viro0df4d6e2013-02-14 22:39:53 -05001277 path.dentry = d_alloc_pseudo(sb, &quick_string);
Al Viro2c48b9c2009-08-09 00:52:35 +04001278 if (!path.dentry)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001279 goto out_shm_unlock;
1280
Al Viro0df4d6e2013-02-14 22:39:53 -05001281 d_set_d_op(path.dentry, &anon_ops);
Andi Kleen42d73952012-12-11 16:01:34 -08001282 path.mnt = mntget(hugetlbfs_vfsmount[hstate_idx]);
Anatol Pomozov39b65252012-09-12 20:11:55 -07001283 file = ERR_PTR(-ENOSPC);
Al Viro0df4d6e2013-02-14 22:39:53 -05001284 inode = hugetlbfs_get_inode(sb, NULL, S_IFREG | S_IRWXUGO, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001285 if (!inode)
Dave Hansence8d2cd2007-10-16 23:31:13 -07001286 goto out_dentry;
Stephen Smalleye1832f22015-08-06 15:46:55 -07001287 if (creat_flags == HUGETLB_SHMFS_INODE)
1288 inode->i_flags |= S_PRIVATE;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001289
Anatol Pomozov39b65252012-09-12 20:11:55 -07001290 file = ERR_PTR(-ENOMEM);
Naoya Horiguchiaf73e4d2013-05-07 16:18:13 -07001291 if (hugetlb_reserve_pages(inode, 0,
1292 size >> huge_page_shift(hstate_inode(inode)), NULL,
1293 acctflag))
David Gibsonb45b5bd2006-03-22 00:08:55 -08001294 goto out_inode;
1295
Al Viro2c48b9c2009-08-09 00:52:35 +04001296 d_instantiate(path.dentry, inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001297 inode->i_size = size;
Miklos Szeredi6d6b77f2011-10-28 14:13:28 +02001298 clear_nlink(inode);
Dave Hansence8d2cd2007-10-16 23:31:13 -07001299
Al Viro2c48b9c2009-08-09 00:52:35 +04001300 file = alloc_file(&path, FMODE_WRITE | FMODE_READ,
Dave Hansence8d2cd2007-10-16 23:31:13 -07001301 &hugetlbfs_file_operations);
Anatol Pomozov39b65252012-09-12 20:11:55 -07001302 if (IS_ERR(file))
Al Virob4d232e2008-02-23 05:59:19 -05001303 goto out_dentry; /* inode is already attached */
Dave Hansence8d2cd2007-10-16 23:31:13 -07001304
Linus Torvalds1da177e2005-04-16 15:20:36 -07001305 return file;
1306
David Gibsonb45b5bd2006-03-22 00:08:55 -08001307out_inode:
1308 iput(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001309out_dentry:
Al Viro2c48b9c2009-08-09 00:52:35 +04001310 path_put(&path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001311out_shm_unlock:
Hugh Dickins353d5c32009-08-24 16:30:28 +01001312 if (*user) {
1313 user_shm_unlock(size, *user);
1314 *user = NULL;
1315 }
Anatol Pomozov39b65252012-09-12 20:11:55 -07001316 return file;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001317}
1318
1319static int __init init_hugetlbfs_fs(void)
1320{
Andi Kleen42d73952012-12-11 16:01:34 -08001321 struct hstate *h;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001322 int error;
Andi Kleen42d73952012-12-11 16:01:34 -08001323 int i;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001324
Nishanth Aravamudan457c1b22014-05-06 12:50:00 -07001325 if (!hugepages_supported()) {
Andrew Morton9b857d22014-06-04 16:07:21 -07001326 pr_info("disabling because there are no supported hugepage sizes\n");
Nishanth Aravamudan457c1b22014-05-06 12:50:00 -07001327 return -ENOTSUPP;
1328 }
1329
Hillf Dantond1d5e052012-03-21 16:34:15 -07001330 error = -ENOMEM;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001331 hugetlbfs_inode_cachep = kmem_cache_create("hugetlbfs_inode_cache",
1332 sizeof(struct hugetlbfs_inode_info),
Vladimir Davydov5d097052016-01-14 15:18:21 -08001333 0, SLAB_ACCOUNT, init_once);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001334 if (hugetlbfs_inode_cachep == NULL)
Peter Zijlstrae0bf68d2007-10-16 23:25:46 -07001335 goto out2;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001336
1337 error = register_filesystem(&hugetlbfs_fs_type);
1338 if (error)
1339 goto out;
1340
Andi Kleen42d73952012-12-11 16:01:34 -08001341 i = 0;
1342 for_each_hstate(h) {
1343 char buf[50];
1344 unsigned ps_kb = 1U << (h->order + PAGE_SHIFT - 10);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001345
Andi Kleen42d73952012-12-11 16:01:34 -08001346 snprintf(buf, sizeof(buf), "pagesize=%uK", ps_kb);
1347 hugetlbfs_vfsmount[i] = kern_mount_data(&hugetlbfs_fs_type,
1348 buf);
1349
1350 if (IS_ERR(hugetlbfs_vfsmount[i])) {
Andrew Morton9b857d22014-06-04 16:07:21 -07001351 pr_err("Cannot mount internal hugetlbfs for "
Andi Kleen42d73952012-12-11 16:01:34 -08001352 "page size %uK", ps_kb);
1353 error = PTR_ERR(hugetlbfs_vfsmount[i]);
1354 hugetlbfs_vfsmount[i] = NULL;
1355 }
1356 i++;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001357 }
Andi Kleen42d73952012-12-11 16:01:34 -08001358 /* Non default hstates are optional */
1359 if (!IS_ERR_OR_NULL(hugetlbfs_vfsmount[default_hstate_idx]))
1360 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001361
1362 out:
Hillf Dantond1d5e052012-03-21 16:34:15 -07001363 kmem_cache_destroy(hugetlbfs_inode_cachep);
Peter Zijlstrae0bf68d2007-10-16 23:25:46 -07001364 out2:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001365 return error;
1366}
Paul Gortmaker3e89e1c2016-01-14 15:21:52 -08001367fs_initcall(init_hugetlbfs_fs)