blob: 98959b87cdf8f1f674c668673b10bf110a7bf653 [file] [log] [blame]
Linus Torvalds1da177e2005-04-16 15:20:36 -07001/*
2 * hugetlbpage-backed filesystem. Based on ramfs.
3 *
4 * William Irwin, 2002
5 *
6 * Copyright (C) 2002 Linus Torvalds.
7 */
8
9#include <linux/module.h>
10#include <linux/thread_info.h>
11#include <asm/current.h>
12#include <linux/sched.h> /* remove ASAP */
13#include <linux/fs.h>
14#include <linux/mount.h>
15#include <linux/file.h>
16#include <linux/writeback.h>
17#include <linux/pagemap.h>
18#include <linux/highmem.h>
19#include <linux/init.h>
20#include <linux/string.h>
Randy Dunlap16f7e0f2006-01-11 12:17:46 -080021#include <linux/capability.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070022#include <linux/backing-dev.h>
23#include <linux/hugetlb.h>
24#include <linux/pagevec.h>
Benjamin Herrenschmidt036e0852007-05-06 14:50:12 -070025#include <linux/mman.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070026#include <linux/quotaops.h>
27#include <linux/slab.h>
28#include <linux/dnotify.h>
29#include <linux/statfs.h>
30#include <linux/security.h>
31
32#include <asm/uaccess.h>
33
34/* some random number */
35#define HUGETLBFS_MAGIC 0x958458f6
36
Josef 'Jeff' Sipekee9b6d62007-02-12 00:55:41 -080037static const struct super_operations hugetlbfs_ops;
Christoph Hellwigf5e54d62006-06-28 04:26:44 -070038static const struct address_space_operations hugetlbfs_aops;
Arjan van de Ven4b6f5d22006-03-28 01:56:42 -080039const struct file_operations hugetlbfs_file_operations;
Arjan van de Ven92e1d5b2007-02-12 00:55:39 -080040static const struct inode_operations hugetlbfs_dir_inode_operations;
41static const struct inode_operations hugetlbfs_inode_operations;
Linus Torvalds1da177e2005-04-16 15:20:36 -070042
43static struct backing_dev_info hugetlbfs_backing_dev_info = {
44 .ra_pages = 0, /* No readahead */
45 .capabilities = BDI_CAP_NO_ACCT_DIRTY | BDI_CAP_NO_WRITEBACK,
46};
47
48int sysctl_hugetlb_shm_group;
49
Adam Litke2e9b3672005-10-29 18:16:47 -070050static void huge_pagevec_release(struct pagevec *pvec)
51{
52 int i;
53
54 for (i = 0; i < pagevec_count(pvec); ++i)
55 put_page(pvec->pages[i]);
56
57 pagevec_reinit(pvec);
58}
59
Linus Torvalds1da177e2005-04-16 15:20:36 -070060static int hugetlbfs_file_mmap(struct file *file, struct vm_area_struct *vma)
61{
Josef Sipekb39424e2006-12-08 02:37:07 -080062 struct inode *inode = file->f_path.dentry->d_inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -070063 loff_t len, vma_len;
64 int ret;
65
Hugh Dickins68589bc2006-11-14 02:03:32 -080066 /*
67 * vma alignment has already been checked by prepare_hugepage_range.
68 * If you add any error returns here, do so after setting VM_HUGETLB,
69 * so is_vm_hugetlb_page tests below unmap_region go the right way
70 * when do_mmap_pgoff unwinds (may be important on powerpc and ia64).
71 */
72 vma->vm_flags |= VM_HUGETLB | VM_RESERVED;
73 vma->vm_ops = &hugetlb_vm_ops;
Linus Torvalds1da177e2005-04-16 15:20:36 -070074
75 vma_len = (loff_t)(vma->vm_end - vma->vm_start);
76
Jes Sorensen1b1dcc12006-01-09 15:59:24 -080077 mutex_lock(&inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -070078 file_accessed(file);
Linus Torvalds1da177e2005-04-16 15:20:36 -070079
80 ret = -ENOMEM;
81 len = vma_len + ((loff_t)vma->vm_pgoff << PAGE_SHIFT);
Linus Torvalds1da177e2005-04-16 15:20:36 -070082
Chen, Kenneth Wa43a8c32006-06-23 02:03:15 -070083 if (vma->vm_flags & VM_MAYSHARE &&
84 hugetlb_reserve_pages(inode, vma->vm_pgoff >> (HPAGE_SHIFT-PAGE_SHIFT),
85 len >> HPAGE_SHIFT))
86 goto out;
David Gibsonb45b5bd2006-03-22 00:08:55 -080087
Adam Litke4c887262005-10-29 18:16:46 -070088 ret = 0;
89 hugetlb_prefault_arch_hook(vma->vm_mm);
Zhang, Yanminb6174df2006-07-10 04:44:49 -070090 if (vma->vm_flags & VM_WRITE && inode->i_size < len)
Linus Torvalds1da177e2005-04-16 15:20:36 -070091 inode->i_size = len;
92out:
Jes Sorensen1b1dcc12006-01-09 15:59:24 -080093 mutex_unlock(&inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -070094
95 return ret;
96}
97
98/*
Hugh Dickins508034a2005-10-29 18:16:30 -070099 * Called under down_write(mmap_sem).
Linus Torvalds1da177e2005-04-16 15:20:36 -0700100 */
101
Adrian Bunkd2ba27e82007-05-06 14:49:00 -0700102#ifndef HAVE_ARCH_HUGETLB_UNMAPPED_AREA
Linus Torvalds1da177e2005-04-16 15:20:36 -0700103static unsigned long
104hugetlb_get_unmapped_area(struct file *file, unsigned long addr,
105 unsigned long len, unsigned long pgoff, unsigned long flags)
106{
107 struct mm_struct *mm = current->mm;
108 struct vm_area_struct *vma;
109 unsigned long start_addr;
110
111 if (len & ~HPAGE_MASK)
112 return -EINVAL;
113 if (len > TASK_SIZE)
114 return -ENOMEM;
115
Benjamin Herrenschmidt036e0852007-05-06 14:50:12 -0700116 if (flags & MAP_FIXED) {
117 if (prepare_hugepage_range(addr, len, pgoff))
118 return -EINVAL;
119 return addr;
120 }
121
Linus Torvalds1da177e2005-04-16 15:20:36 -0700122 if (addr) {
123 addr = ALIGN(addr, HPAGE_SIZE);
124 vma = find_vma(mm, addr);
125 if (TASK_SIZE - len >= addr &&
126 (!vma || addr + len <= vma->vm_start))
127 return addr;
128 }
129
130 start_addr = mm->free_area_cache;
131
Wolfgang Wander1363c3c2005-06-21 17:14:49 -0700132 if (len <= mm->cached_hole_size)
133 start_addr = TASK_UNMAPPED_BASE;
134
Linus Torvalds1da177e2005-04-16 15:20:36 -0700135full_search:
136 addr = ALIGN(start_addr, HPAGE_SIZE);
137
138 for (vma = find_vma(mm, addr); ; vma = vma->vm_next) {
139 /* At this point: (!vma || addr < vma->vm_end). */
140 if (TASK_SIZE - len < addr) {
141 /*
142 * Start a new search - just in case we missed
143 * some holes.
144 */
145 if (start_addr != TASK_UNMAPPED_BASE) {
146 start_addr = TASK_UNMAPPED_BASE;
147 goto full_search;
148 }
149 return -ENOMEM;
150 }
151
152 if (!vma || addr + len <= vma->vm_start)
153 return addr;
154 addr = ALIGN(vma->vm_end, HPAGE_SIZE);
155 }
156}
157#endif
158
159/*
160 * Read a page. Again trivial. If it didn't already exist
161 * in the page cache, it is zero-filled.
162 */
163static int hugetlbfs_readpage(struct file *file, struct page * page)
164{
165 unlock_page(page);
166 return -EINVAL;
167}
168
169static int hugetlbfs_prepare_write(struct file *file,
170 struct page *page, unsigned offset, unsigned to)
171{
172 return -EINVAL;
173}
174
175static int hugetlbfs_commit_write(struct file *file,
176 struct page *page, unsigned offset, unsigned to)
177{
178 return -EINVAL;
179}
180
Linus Torvalds1da177e2005-04-16 15:20:36 -0700181static void truncate_huge_page(struct page *page)
182{
Linus Torvaldsfba25912006-12-20 13:46:42 -0800183 cancel_dirty_page(page, /* No IO accounting for huge pages? */0);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700184 ClearPageUptodate(page);
185 remove_from_page_cache(page);
186 put_page(page);
187}
188
David Gibsonb45b5bd2006-03-22 00:08:55 -0800189static void truncate_hugepages(struct inode *inode, loff_t lstart)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700190{
David Gibsonb45b5bd2006-03-22 00:08:55 -0800191 struct address_space *mapping = &inode->i_data;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700192 const pgoff_t start = lstart >> HPAGE_SHIFT;
193 struct pagevec pvec;
194 pgoff_t next;
Chen, Kenneth Wa43a8c32006-06-23 02:03:15 -0700195 int i, freed = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700196
197 pagevec_init(&pvec, 0);
198 next = start;
199 while (1) {
200 if (!pagevec_lookup(&pvec, mapping, next, PAGEVEC_SIZE)) {
201 if (next == start)
202 break;
203 next = start;
204 continue;
205 }
206
207 for (i = 0; i < pagevec_count(&pvec); ++i) {
208 struct page *page = pvec.pages[i];
209
210 lock_page(page);
211 if (page->index > next)
212 next = page->index;
213 ++next;
214 truncate_huge_page(page);
215 unlock_page(page);
216 hugetlb_put_quota(mapping);
Chen, Kenneth Wa43a8c32006-06-23 02:03:15 -0700217 freed++;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700218 }
219 huge_pagevec_release(&pvec);
220 }
221 BUG_ON(!lstart && mapping->nrpages);
Chen, Kenneth Wa43a8c32006-06-23 02:03:15 -0700222 hugetlb_unreserve_pages(inode, start, freed);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700223}
224
225static void hugetlbfs_delete_inode(struct inode *inode)
226{
David Gibsonb45b5bd2006-03-22 00:08:55 -0800227 truncate_hugepages(inode, 0);
Christoph Hellwig149f4212005-10-29 18:16:43 -0700228 clear_inode(inode);
229}
230
Josh Triplettddc0a512006-09-29 01:59:27 -0700231static void hugetlbfs_forget_inode(struct inode *inode) __releases(inode_lock)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700232{
Christoph Hellwig0b1533f2005-10-29 18:16:45 -0700233 struct super_block *sb = inode->i_sb;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700234
Christoph Hellwig0b1533f2005-10-29 18:16:45 -0700235 if (!hlist_unhashed(&inode->i_hash)) {
236 if (!(inode->i_state & (I_DIRTY|I_LOCK)))
237 list_move(&inode->i_list, &inode_unused);
238 inodes_stat.nr_unused++;
239 if (!sb || (sb->s_flags & MS_ACTIVE)) {
240 spin_unlock(&inode_lock);
241 return;
242 }
243 inode->i_state |= I_WILL_FREE;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700244 spin_unlock(&inode_lock);
Christoph Hellwig0b1533f2005-10-29 18:16:45 -0700245 /*
246 * write_inode_now is a noop as we set BDI_CAP_NO_WRITEBACK
247 * in our backing_dev_info.
248 */
249 write_inode_now(inode, 1);
250 spin_lock(&inode_lock);
251 inode->i_state &= ~I_WILL_FREE;
252 inodes_stat.nr_unused--;
253 hlist_del_init(&inode->i_hash);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700254 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700255 list_del_init(&inode->i_list);
256 list_del_init(&inode->i_sb_list);
257 inode->i_state |= I_FREEING;
258 inodes_stat.nr_inodes--;
259 spin_unlock(&inode_lock);
David Gibsonb45b5bd2006-03-22 00:08:55 -0800260 truncate_hugepages(inode, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700261 clear_inode(inode);
262 destroy_inode(inode);
263}
264
265static void hugetlbfs_drop_inode(struct inode *inode)
266{
267 if (!inode->i_nlink)
Christoph Hellwig6b09b9d2005-10-29 18:16:44 -0700268 generic_delete_inode(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700269 else
270 hugetlbfs_forget_inode(inode);
271}
272
Linus Torvalds1da177e2005-04-16 15:20:36 -0700273static inline void
Hugh Dickins856fc292006-10-28 10:38:43 -0700274hugetlb_vmtruncate_list(struct prio_tree_root *root, pgoff_t pgoff)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700275{
276 struct vm_area_struct *vma;
277 struct prio_tree_iter iter;
278
Hugh Dickins856fc292006-10-28 10:38:43 -0700279 vma_prio_tree_foreach(vma, &iter, root, pgoff, ULONG_MAX) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700280 unsigned long v_offset;
281
Linus Torvalds1da177e2005-04-16 15:20:36 -0700282 /*
Hugh Dickins856fc292006-10-28 10:38:43 -0700283 * Can the expression below overflow on 32-bit arches?
284 * No, because the prio_tree returns us only those vmas
285 * which overlap the truncated area starting at pgoff,
286 * and no vma on a 32-bit arch can span beyond the 4GB.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700287 */
Hugh Dickins856fc292006-10-28 10:38:43 -0700288 if (vma->vm_pgoff < pgoff)
289 v_offset = (pgoff - vma->vm_pgoff) << PAGE_SHIFT;
290 else
Linus Torvalds1da177e2005-04-16 15:20:36 -0700291 v_offset = 0;
292
Chen, Kenneth W502717f2006-10-11 01:20:46 -0700293 __unmap_hugepage_range(vma,
Hugh Dickins508034a2005-10-29 18:16:30 -0700294 vma->vm_start + v_offset, vma->vm_end);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700295 }
296}
297
298/*
299 * Expanding truncates are not allowed.
300 */
301static int hugetlb_vmtruncate(struct inode *inode, loff_t offset)
302{
Hugh Dickins856fc292006-10-28 10:38:43 -0700303 pgoff_t pgoff;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700304 struct address_space *mapping = inode->i_mapping;
305
306 if (offset > inode->i_size)
307 return -EINVAL;
308
309 BUG_ON(offset & ~HPAGE_MASK);
Hugh Dickins856fc292006-10-28 10:38:43 -0700310 pgoff = offset >> PAGE_SHIFT;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700311
312 inode->i_size = offset;
313 spin_lock(&mapping->i_mmap_lock);
314 if (!prio_tree_empty(&mapping->i_mmap))
315 hugetlb_vmtruncate_list(&mapping->i_mmap, pgoff);
316 spin_unlock(&mapping->i_mmap_lock);
David Gibsonb45b5bd2006-03-22 00:08:55 -0800317 truncate_hugepages(inode, offset);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700318 return 0;
319}
320
321static int hugetlbfs_setattr(struct dentry *dentry, struct iattr *attr)
322{
323 struct inode *inode = dentry->d_inode;
324 int error;
325 unsigned int ia_valid = attr->ia_valid;
326
327 BUG_ON(!inode);
328
329 error = inode_change_ok(inode, attr);
330 if (error)
331 goto out;
332
333 if (ia_valid & ATTR_SIZE) {
334 error = -EINVAL;
335 if (!(attr->ia_size & ~HPAGE_MASK))
336 error = hugetlb_vmtruncate(inode, attr->ia_size);
337 if (error)
338 goto out;
339 attr->ia_valid &= ~ATTR_SIZE;
340 }
341 error = inode_setattr(inode, attr);
342out:
343 return error;
344}
345
346static struct inode *hugetlbfs_get_inode(struct super_block *sb, uid_t uid,
347 gid_t gid, int mode, dev_t dev)
348{
349 struct inode *inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700350
351 inode = new_inode(sb);
352 if (inode) {
353 struct hugetlbfs_inode_info *info;
354 inode->i_mode = mode;
355 inode->i_uid = uid;
356 inode->i_gid = gid;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700357 inode->i_blocks = 0;
358 inode->i_mapping->a_ops = &hugetlbfs_aops;
359 inode->i_mapping->backing_dev_info =&hugetlbfs_backing_dev_info;
360 inode->i_atime = inode->i_mtime = inode->i_ctime = CURRENT_TIME;
Chen, Kenneth Wa43a8c32006-06-23 02:03:15 -0700361 INIT_LIST_HEAD(&inode->i_mapping->private_list);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700362 info = HUGETLBFS_I(inode);
Robin Holt7339ff82006-01-14 13:20:48 -0800363 mpol_shared_policy_init(&info->policy, MPOL_DEFAULT, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700364 switch (mode & S_IFMT) {
365 default:
366 init_special_inode(inode, mode, dev);
367 break;
368 case S_IFREG:
369 inode->i_op = &hugetlbfs_inode_operations;
370 inode->i_fop = &hugetlbfs_file_operations;
371 break;
372 case S_IFDIR:
373 inode->i_op = &hugetlbfs_dir_inode_operations;
374 inode->i_fop = &simple_dir_operations;
375
376 /* directory inodes start off with i_nlink == 2 (for "." entry) */
Dave Hansend8c76e62006-09-30 23:29:04 -0700377 inc_nlink(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700378 break;
379 case S_IFLNK:
380 inode->i_op = &page_symlink_inode_operations;
381 break;
382 }
383 }
384 return inode;
385}
386
387/*
388 * File creation. Allocate an inode, and we're done..
389 */
390static int hugetlbfs_mknod(struct inode *dir,
391 struct dentry *dentry, int mode, dev_t dev)
392{
393 struct inode *inode;
394 int error = -ENOSPC;
395 gid_t gid;
396
397 if (dir->i_mode & S_ISGID) {
398 gid = dir->i_gid;
399 if (S_ISDIR(mode))
400 mode |= S_ISGID;
401 } else {
402 gid = current->fsgid;
403 }
404 inode = hugetlbfs_get_inode(dir->i_sb, current->fsuid, gid, mode, dev);
405 if (inode) {
406 dir->i_ctime = dir->i_mtime = CURRENT_TIME;
407 d_instantiate(dentry, inode);
408 dget(dentry); /* Extra count - pin the dentry in core */
409 error = 0;
410 }
411 return error;
412}
413
414static int hugetlbfs_mkdir(struct inode *dir, struct dentry *dentry, int mode)
415{
416 int retval = hugetlbfs_mknod(dir, dentry, mode | S_IFDIR, 0);
417 if (!retval)
Dave Hansend8c76e62006-09-30 23:29:04 -0700418 inc_nlink(dir);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700419 return retval;
420}
421
422static int hugetlbfs_create(struct inode *dir, struct dentry *dentry, int mode, struct nameidata *nd)
423{
424 return hugetlbfs_mknod(dir, dentry, mode | S_IFREG, 0);
425}
426
427static int hugetlbfs_symlink(struct inode *dir,
428 struct dentry *dentry, const char *symname)
429{
430 struct inode *inode;
431 int error = -ENOSPC;
432 gid_t gid;
433
434 if (dir->i_mode & S_ISGID)
435 gid = dir->i_gid;
436 else
437 gid = current->fsgid;
438
439 inode = hugetlbfs_get_inode(dir->i_sb, current->fsuid,
440 gid, S_IFLNK|S_IRWXUGO, 0);
441 if (inode) {
442 int l = strlen(symname)+1;
443 error = page_symlink(inode, symname, l);
444 if (!error) {
445 d_instantiate(dentry, inode);
446 dget(dentry);
447 } else
448 iput(inode);
449 }
450 dir->i_ctime = dir->i_mtime = CURRENT_TIME;
451
452 return error;
453}
454
455/*
Ken Chen6649a382007-02-08 14:20:27 -0800456 * mark the head page dirty
Linus Torvalds1da177e2005-04-16 15:20:36 -0700457 */
458static int hugetlbfs_set_page_dirty(struct page *page)
459{
Christoph Lameterd85f3382007-05-06 14:49:39 -0700460 struct page *head = compound_head(page);
Ken Chen6649a382007-02-08 14:20:27 -0800461
462 SetPageDirty(head);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700463 return 0;
464}
465
David Howells726c3342006-06-23 02:02:58 -0700466static int hugetlbfs_statfs(struct dentry *dentry, struct kstatfs *buf)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700467{
David Howells726c3342006-06-23 02:02:58 -0700468 struct hugetlbfs_sb_info *sbinfo = HUGETLBFS_SB(dentry->d_sb);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700469
470 buf->f_type = HUGETLBFS_MAGIC;
471 buf->f_bsize = HPAGE_SIZE;
472 if (sbinfo) {
473 spin_lock(&sbinfo->stat_lock);
David Gibson74a8a652005-11-21 21:32:24 -0800474 /* If no limits set, just report 0 for max/free/used
475 * blocks, like simple_statfs() */
476 if (sbinfo->max_blocks >= 0) {
477 buf->f_blocks = sbinfo->max_blocks;
478 buf->f_bavail = buf->f_bfree = sbinfo->free_blocks;
479 buf->f_files = sbinfo->max_inodes;
480 buf->f_ffree = sbinfo->free_inodes;
481 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700482 spin_unlock(&sbinfo->stat_lock);
483 }
484 buf->f_namelen = NAME_MAX;
485 return 0;
486}
487
488static void hugetlbfs_put_super(struct super_block *sb)
489{
490 struct hugetlbfs_sb_info *sbi = HUGETLBFS_SB(sb);
491
492 if (sbi) {
493 sb->s_fs_info = NULL;
494 kfree(sbi);
495 }
496}
497
Christoph Hellwig96527982005-10-29 18:16:42 -0700498static inline int hugetlbfs_dec_free_inodes(struct hugetlbfs_sb_info *sbinfo)
499{
500 if (sbinfo->free_inodes >= 0) {
501 spin_lock(&sbinfo->stat_lock);
502 if (unlikely(!sbinfo->free_inodes)) {
503 spin_unlock(&sbinfo->stat_lock);
504 return 0;
505 }
506 sbinfo->free_inodes--;
507 spin_unlock(&sbinfo->stat_lock);
508 }
509
510 return 1;
511}
512
513static void hugetlbfs_inc_free_inodes(struct hugetlbfs_sb_info *sbinfo)
514{
515 if (sbinfo->free_inodes >= 0) {
516 spin_lock(&sbinfo->stat_lock);
517 sbinfo->free_inodes++;
518 spin_unlock(&sbinfo->stat_lock);
519 }
520}
521
522
Christoph Lametere18b8902006-12-06 20:33:20 -0800523static struct kmem_cache *hugetlbfs_inode_cachep;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700524
525static struct inode *hugetlbfs_alloc_inode(struct super_block *sb)
526{
Christoph Hellwig96527982005-10-29 18:16:42 -0700527 struct hugetlbfs_sb_info *sbinfo = HUGETLBFS_SB(sb);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700528 struct hugetlbfs_inode_info *p;
529
Christoph Hellwig96527982005-10-29 18:16:42 -0700530 if (unlikely(!hugetlbfs_dec_free_inodes(sbinfo)))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700531 return NULL;
Christoph Lametere94b1762006-12-06 20:33:17 -0800532 p = kmem_cache_alloc(hugetlbfs_inode_cachep, GFP_KERNEL);
Christoph Hellwig96527982005-10-29 18:16:42 -0700533 if (unlikely(!p)) {
534 hugetlbfs_inc_free_inodes(sbinfo);
535 return NULL;
536 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700537 return &p->vfs_inode;
538}
539
Linus Torvalds1da177e2005-04-16 15:20:36 -0700540static void hugetlbfs_destroy_inode(struct inode *inode)
541{
Christoph Hellwig96527982005-10-29 18:16:42 -0700542 hugetlbfs_inc_free_inodes(HUGETLBFS_SB(inode->i_sb));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700543 mpol_free_shared_policy(&HUGETLBFS_I(inode)->policy);
544 kmem_cache_free(hugetlbfs_inode_cachep, HUGETLBFS_I(inode));
545}
546
Christoph Hellwigf5e54d62006-06-28 04:26:44 -0700547static const struct address_space_operations hugetlbfs_aops = {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700548 .readpage = hugetlbfs_readpage,
549 .prepare_write = hugetlbfs_prepare_write,
550 .commit_write = hugetlbfs_commit_write,
551 .set_page_dirty = hugetlbfs_set_page_dirty,
552};
553
Christoph Hellwig96527982005-10-29 18:16:42 -0700554
Christoph Lametere18b8902006-12-06 20:33:20 -0800555static void init_once(void *foo, struct kmem_cache *cachep, unsigned long flags)
Christoph Hellwig96527982005-10-29 18:16:42 -0700556{
557 struct hugetlbfs_inode_info *ei = (struct hugetlbfs_inode_info *)foo;
558
Christoph Lameter50953fe2007-05-06 14:50:16 -0700559 if (flags & SLAB_CTOR_CONSTRUCTOR)
Christoph Hellwig96527982005-10-29 18:16:42 -0700560 inode_init_once(&ei->vfs_inode);
561}
562
Arjan van de Ven4b6f5d22006-03-28 01:56:42 -0800563const struct file_operations hugetlbfs_file_operations = {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700564 .mmap = hugetlbfs_file_mmap,
565 .fsync = simple_sync_file,
566 .get_unmapped_area = hugetlb_get_unmapped_area,
567};
568
Arjan van de Ven92e1d5b2007-02-12 00:55:39 -0800569static const struct inode_operations hugetlbfs_dir_inode_operations = {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700570 .create = hugetlbfs_create,
571 .lookup = simple_lookup,
572 .link = simple_link,
573 .unlink = simple_unlink,
574 .symlink = hugetlbfs_symlink,
575 .mkdir = hugetlbfs_mkdir,
576 .rmdir = simple_rmdir,
577 .mknod = hugetlbfs_mknod,
578 .rename = simple_rename,
579 .setattr = hugetlbfs_setattr,
580};
581
Arjan van de Ven92e1d5b2007-02-12 00:55:39 -0800582static const struct inode_operations hugetlbfs_inode_operations = {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700583 .setattr = hugetlbfs_setattr,
584};
585
Josef 'Jeff' Sipekee9b6d62007-02-12 00:55:41 -0800586static const struct super_operations hugetlbfs_ops = {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700587 .alloc_inode = hugetlbfs_alloc_inode,
588 .destroy_inode = hugetlbfs_destroy_inode,
589 .statfs = hugetlbfs_statfs,
Christoph Hellwig149f4212005-10-29 18:16:43 -0700590 .delete_inode = hugetlbfs_delete_inode,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700591 .drop_inode = hugetlbfs_drop_inode,
592 .put_super = hugetlbfs_put_super,
593};
594
595static int
596hugetlbfs_parse_options(char *options, struct hugetlbfs_config *pconfig)
597{
598 char *opt, *value, *rest;
599
600 if (!options)
601 return 0;
602 while ((opt = strsep(&options, ",")) != NULL) {
603 if (!*opt)
604 continue;
605
606 value = strchr(opt, '=');
607 if (!value || !*value)
608 return -EINVAL;
609 else
610 *value++ = '\0';
611
612 if (!strcmp(opt, "uid"))
613 pconfig->uid = simple_strtoul(value, &value, 0);
614 else if (!strcmp(opt, "gid"))
615 pconfig->gid = simple_strtoul(value, &value, 0);
616 else if (!strcmp(opt, "mode"))
617 pconfig->mode = simple_strtoul(value,&value,0) & 0777U;
618 else if (!strcmp(opt, "size")) {
619 unsigned long long size = memparse(value, &rest);
620 if (*rest == '%') {
621 size <<= HPAGE_SHIFT;
622 size *= max_huge_pages;
623 do_div(size, 100);
624 rest++;
625 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700626 pconfig->nr_blocks = (size >> HPAGE_SHIFT);
627 value = rest;
628 } else if (!strcmp(opt,"nr_inodes")) {
629 pconfig->nr_inodes = memparse(value, &rest);
630 value = rest;
631 } else
632 return -EINVAL;
633
634 if (*value)
635 return -EINVAL;
636 }
637 return 0;
638}
639
640static int
641hugetlbfs_fill_super(struct super_block *sb, void *data, int silent)
642{
643 struct inode * inode;
644 struct dentry * root;
645 int ret;
646 struct hugetlbfs_config config;
647 struct hugetlbfs_sb_info *sbinfo;
648
649 config.nr_blocks = -1; /* No limit on size by default */
650 config.nr_inodes = -1; /* No limit on number of inodes by default */
651 config.uid = current->fsuid;
652 config.gid = current->fsgid;
653 config.mode = 0755;
654 ret = hugetlbfs_parse_options(data, &config);
655
656 if (ret)
657 return ret;
658
659 sbinfo = kmalloc(sizeof(struct hugetlbfs_sb_info), GFP_KERNEL);
660 if (!sbinfo)
661 return -ENOMEM;
662 sb->s_fs_info = sbinfo;
663 spin_lock_init(&sbinfo->stat_lock);
664 sbinfo->max_blocks = config.nr_blocks;
665 sbinfo->free_blocks = config.nr_blocks;
666 sbinfo->max_inodes = config.nr_inodes;
667 sbinfo->free_inodes = config.nr_inodes;
668 sb->s_maxbytes = MAX_LFS_FILESIZE;
669 sb->s_blocksize = HPAGE_SIZE;
670 sb->s_blocksize_bits = HPAGE_SHIFT;
671 sb->s_magic = HUGETLBFS_MAGIC;
672 sb->s_op = &hugetlbfs_ops;
673 sb->s_time_gran = 1;
674 inode = hugetlbfs_get_inode(sb, config.uid, config.gid,
675 S_IFDIR | config.mode, 0);
676 if (!inode)
677 goto out_free;
678
679 root = d_alloc_root(inode);
680 if (!root) {
681 iput(inode);
682 goto out_free;
683 }
684 sb->s_root = root;
685 return 0;
686out_free:
687 kfree(sbinfo);
688 return -ENOMEM;
689}
690
691int hugetlb_get_quota(struct address_space *mapping)
692{
693 int ret = 0;
694 struct hugetlbfs_sb_info *sbinfo = HUGETLBFS_SB(mapping->host->i_sb);
695
696 if (sbinfo->free_blocks > -1) {
697 spin_lock(&sbinfo->stat_lock);
698 if (sbinfo->free_blocks > 0)
699 sbinfo->free_blocks--;
700 else
701 ret = -ENOMEM;
702 spin_unlock(&sbinfo->stat_lock);
703 }
704
705 return ret;
706}
707
708void hugetlb_put_quota(struct address_space *mapping)
709{
710 struct hugetlbfs_sb_info *sbinfo = HUGETLBFS_SB(mapping->host->i_sb);
711
712 if (sbinfo->free_blocks > -1) {
713 spin_lock(&sbinfo->stat_lock);
714 sbinfo->free_blocks++;
715 spin_unlock(&sbinfo->stat_lock);
716 }
717}
718
David Howells454e2392006-06-23 02:02:57 -0700719static int hugetlbfs_get_sb(struct file_system_type *fs_type,
720 int flags, const char *dev_name, void *data, struct vfsmount *mnt)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700721{
David Howells454e2392006-06-23 02:02:57 -0700722 return get_sb_nodev(fs_type, flags, data, hugetlbfs_fill_super, mnt);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700723}
724
725static struct file_system_type hugetlbfs_fs_type = {
726 .name = "hugetlbfs",
727 .get_sb = hugetlbfs_get_sb,
728 .kill_sb = kill_litter_super,
729};
730
731static struct vfsmount *hugetlbfs_vfsmount;
732
Linus Torvalds1da177e2005-04-16 15:20:36 -0700733static int can_do_hugetlb_shm(void)
734{
735 return likely(capable(CAP_IPC_LOCK) ||
736 in_group_p(sysctl_hugetlb_shm_group) ||
737 can_do_mlock());
738}
739
740struct file *hugetlb_zero_setup(size_t size)
741{
742 int error = -ENOMEM;
743 struct file *file;
744 struct inode *inode;
745 struct dentry *dentry, *root;
746 struct qstr quick_string;
747 char buf[16];
Chen, Kenneth Wbba1e9b2006-03-22 00:09:02 -0800748 static atomic_t counter;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700749
Akinobu Mita5bc98592007-05-06 14:50:18 -0700750 if (!hugetlbfs_vfsmount)
751 return ERR_PTR(-ENOENT);
752
Linus Torvalds1da177e2005-04-16 15:20:36 -0700753 if (!can_do_hugetlb_shm())
754 return ERR_PTR(-EPERM);
755
Linus Torvalds1da177e2005-04-16 15:20:36 -0700756 if (!user_shm_lock(size, current->user))
757 return ERR_PTR(-ENOMEM);
758
759 root = hugetlbfs_vfsmount->mnt_root;
Chen, Kenneth Wbba1e9b2006-03-22 00:09:02 -0800760 snprintf(buf, 16, "%u", atomic_inc_return(&counter));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700761 quick_string.name = buf;
762 quick_string.len = strlen(quick_string.name);
763 quick_string.hash = 0;
764 dentry = d_alloc(root, &quick_string);
765 if (!dentry)
766 goto out_shm_unlock;
767
768 error = -ENFILE;
769 file = get_empty_filp();
770 if (!file)
771 goto out_dentry;
772
773 error = -ENOSPC;
774 inode = hugetlbfs_get_inode(root->d_sb, current->fsuid,
775 current->fsgid, S_IFREG | S_IRWXUGO, 0);
776 if (!inode)
777 goto out_file;
778
David Gibsonb45b5bd2006-03-22 00:08:55 -0800779 error = -ENOMEM;
Chen, Kenneth Wa43a8c32006-06-23 02:03:15 -0700780 if (hugetlb_reserve_pages(inode, 0, size >> HPAGE_SHIFT))
David Gibsonb45b5bd2006-03-22 00:08:55 -0800781 goto out_inode;
782
Linus Torvalds1da177e2005-04-16 15:20:36 -0700783 d_instantiate(dentry, inode);
784 inode->i_size = size;
785 inode->i_nlink = 0;
Josef Sipekb39424e2006-12-08 02:37:07 -0800786 file->f_path.mnt = mntget(hugetlbfs_vfsmount);
787 file->f_path.dentry = dentry;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700788 file->f_mapping = inode->i_mapping;
789 file->f_op = &hugetlbfs_file_operations;
790 file->f_mode = FMODE_WRITE | FMODE_READ;
791 return file;
792
David Gibsonb45b5bd2006-03-22 00:08:55 -0800793out_inode:
794 iput(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700795out_file:
796 put_filp(file);
797out_dentry:
798 dput(dentry);
799out_shm_unlock:
800 user_shm_unlock(size, current->user);
801 return ERR_PTR(error);
802}
803
804static int __init init_hugetlbfs_fs(void)
805{
806 int error;
807 struct vfsmount *vfsmount;
808
809 hugetlbfs_inode_cachep = kmem_cache_create("hugetlbfs_inode_cache",
810 sizeof(struct hugetlbfs_inode_info),
811 0, 0, init_once, NULL);
812 if (hugetlbfs_inode_cachep == NULL)
813 return -ENOMEM;
814
815 error = register_filesystem(&hugetlbfs_fs_type);
816 if (error)
817 goto out;
818
819 vfsmount = kern_mount(&hugetlbfs_fs_type);
820
821 if (!IS_ERR(vfsmount)) {
822 hugetlbfs_vfsmount = vfsmount;
823 return 0;
824 }
825
826 error = PTR_ERR(vfsmount);
827
828 out:
829 if (error)
830 kmem_cache_destroy(hugetlbfs_inode_cachep);
831 return error;
832}
833
834static void __exit exit_hugetlbfs_fs(void)
835{
836 kmem_cache_destroy(hugetlbfs_inode_cachep);
837 unregister_filesystem(&hugetlbfs_fs_type);
838}
839
840module_init(init_hugetlbfs_fs)
841module_exit(exit_hugetlbfs_fs)
842
843MODULE_LICENSE("GPL");