~ [ source navigation ] ~ [ diff markup ] ~ [ identifier search ] ~

TOMOYO Linux Cross Reference
Linux/fs/hugetlbfs/inode.c

Version: ~ [ linux-5.16-rc1 ] ~ [ linux-5.15.2 ] ~ [ linux-5.14.18 ] ~ [ linux-5.13.19 ] ~ [ linux-5.12.19 ] ~ [ linux-5.11.22 ] ~ [ linux-5.10.79 ] ~ [ linux-5.9.16 ] ~ [ linux-5.8.18 ] ~ [ linux-5.7.19 ] ~ [ linux-5.6.19 ] ~ [ linux-5.5.19 ] ~ [ linux-5.4.159 ] ~ [ linux-5.3.18 ] ~ [ linux-5.2.21 ] ~ [ linux-5.1.21 ] ~ [ linux-5.0.21 ] ~ [ linux-4.20.17 ] ~ [ linux-4.19.217 ] ~ [ linux-4.18.20 ] ~ [ linux-4.17.19 ] ~ [ linux-4.16.18 ] ~ [ linux-4.15.18 ] ~ [ linux-4.14.255 ] ~ [ linux-4.13.16 ] ~ [ linux-4.12.14 ] ~ [ linux-4.11.12 ] ~ [ linux-4.10.17 ] ~ [ linux-4.9.290 ] ~ [ linux-4.8.17 ] ~ [ linux-4.7.10 ] ~ [ linux-4.6.7 ] ~ [ linux-4.5.7 ] ~ [ linux-4.4.292 ] ~ [ linux-4.3.6 ] ~ [ linux-4.2.8 ] ~ [ linux-4.1.52 ] ~ [ linux-4.0.9 ] ~ [ linux-3.18.140 ] ~ [ linux-3.16.85 ] ~ [ linux-3.14.79 ] ~ [ linux-3.12.74 ] ~ [ linux-3.10.108 ] ~ [ linux-2.6.32.71 ] ~ [ linux-2.6.0 ] ~ [ linux-2.4.37.11 ] ~ [ unix-v6-master ] ~ [ ccs-tools-1.8.5 ] ~ [ policy-sample ] ~
Architecture: ~ [ i386 ] ~ [ alpha ] ~ [ m68k ] ~ [ mips ] ~ [ ppc ] ~ [ sparc ] ~ [ sparc64 ] ~

  1 /*
  2  * hugetlbpage-backed filesystem.  Based on ramfs.
  3  *
  4  * Nadia Yvette Chambers, 2002
  5  *
  6  * Copyright (C) 2002 Linus Torvalds.
  7  * License: GPL
  8  */
  9 
 10 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
 11 
 12 #include <linux/thread_info.h>
 13 #include <asm/current.h>
 14 #include <linux/sched/signal.h>         /* remove ASAP */
 15 #include <linux/falloc.h>
 16 #include <linux/fs.h>
 17 #include <linux/mount.h>
 18 #include <linux/file.h>
 19 #include <linux/kernel.h>
 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>
 25 #include <linux/capability.h>
 26 #include <linux/ctype.h>
 27 #include <linux/backing-dev.h>
 28 #include <linux/hugetlb.h>
 29 #include <linux/pagevec.h>
 30 #include <linux/parser.h>
 31 #include <linux/mman.h>
 32 #include <linux/slab.h>
 33 #include <linux/dnotify.h>
 34 #include <linux/statfs.h>
 35 #include <linux/security.h>
 36 #include <linux/magic.h>
 37 #include <linux/migrate.h>
 38 #include <linux/uio.h>
 39 
 40 #include <linux/uaccess.h>
 41 
 42 static const struct super_operations hugetlbfs_ops;
 43 static const struct address_space_operations hugetlbfs_aops;
 44 const struct file_operations hugetlbfs_file_operations;
 45 static const struct inode_operations hugetlbfs_dir_inode_operations;
 46 static const struct inode_operations hugetlbfs_inode_operations;
 47 
 48 struct hugetlbfs_config {
 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;
 56 };
 57 
 58 int sysctl_hugetlb_shm_group;
 59 
 60 enum {
 61         Opt_size, Opt_nr_inodes,
 62         Opt_mode, Opt_uid, Opt_gid,
 63         Opt_pagesize, Opt_min_size,
 64         Opt_err,
 65 };
 66 
 67 static const match_table_t tokens = {
 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"},
 73         {Opt_pagesize,  "pagesize=%s"},
 74         {Opt_min_size,  "min_size=%s"},
 75         {Opt_err,       NULL},
 76 };
 77 
 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 
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 
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 
121 static int hugetlbfs_file_mmap(struct file *file, struct vm_area_struct *vma)
122 {
123         struct inode *inode = file_inode(file);
124         loff_t len, vma_len;
125         int ret;
126         struct hstate *h = hstate_file(file);
127 
128         /*
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).
135          */
136         vma->vm_flags |= VM_HUGETLB | VM_DONTEXPAND;
137         vma->vm_ops = &hugetlb_vm_ops;
138 
139         /*
140          * page based offset in vm_pgoff could be sufficiently large to
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.
144          */
145         if (sizeof(unsigned long) == sizeof(loff_t)) {
146                 if (vma->vm_pgoff & PGOFF_LOFFT_MAX)
147                         return -EINVAL;
148         }
149 
150         /* must be huge page aligned */
151         if (vma->vm_pgoff & (~huge_page_mask(h) >> PAGE_SHIFT))
152                 return -EINVAL;
153 
154         vma_len = (loff_t)(vma->vm_end - vma->vm_start);
155         len = vma_len + ((loff_t)vma->vm_pgoff << PAGE_SHIFT);
156         /* check for overflow */
157         if (len < vma_len)
158                 return -EINVAL;
159 
160         inode_lock(inode);
161         file_accessed(file);
162 
163         ret = -ENOMEM;
164         if (hugetlb_reserve_pages(inode,
165                                 vma->vm_pgoff >> huge_page_order(h),
166                                 len >> huge_page_shift(h), vma,
167                                 vma->vm_flags))
168                 goto out;
169 
170         ret = 0;
171         if (vma->vm_flags & VM_WRITE && inode->i_size < len)
172                 i_size_write(inode, len);
173 out:
174         inode_unlock(inode);
175 
176         return ret;
177 }
178 
179 /*
180  * Called under down_write(mmap_sem).
181  */
182 
183 #ifndef HAVE_ARCH_HUGETLB_UNMAPPED_AREA
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;
190         struct hstate *h = hstate_file(file);
191         struct vm_unmapped_area_info info;
192 
193         if (len & ~huge_page_mask(h))
194                 return -EINVAL;
195         if (len > TASK_SIZE)
196                 return -ENOMEM;
197 
198         if (flags & MAP_FIXED) {
199                 if (prepare_hugepage_range(file, addr, len))
200                         return -EINVAL;
201                 return addr;
202         }
203 
204         if (addr) {
205                 addr = ALIGN(addr, huge_page_size(h));
206                 vma = find_vma(mm, addr);
207                 if (TASK_SIZE - len >= addr &&
208                     (!vma || addr + len <= vm_start_gap(vma)))
209                         return addr;
210         }
211 
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);
219 }
220 #endif
221 
222 static size_t
223 hugetlbfs_read_actor(struct page *page, unsigned long offset,
224                         struct iov_iter *to, unsigned long size)
225 {
226         size_t copied = 0;
227         int i, chunksize;
228 
229         /* Find which 4k chunk and offset with in that chunk */
230         i = offset >> PAGE_SHIFT;
231         offset = offset & ~PAGE_MASK;
232 
233         while (size) {
234                 size_t n;
235                 chunksize = PAGE_SIZE;
236                 if (offset)
237                         chunksize -= offset;
238                 if (chunksize > size)
239                         chunksize = size;
240                 n = copy_page_to_iter(&page[i], offset, chunksize, to);
241                 copied += n;
242                 if (n != chunksize)
243                         return copied;
244                 offset = 0;
245                 size -= chunksize;
246                 i++;
247         }
248         return copied;
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
254  * since it has PAGE_SIZE assumptions.
255  */
256 static ssize_t hugetlbfs_read_iter(struct kiocb *iocb, struct iov_iter *to)
257 {
258         struct file *file = iocb->ki_filp;
259         struct hstate *h = hstate_file(file);
260         struct address_space *mapping = file->f_mapping;
261         struct inode *inode = mapping->host;
262         unsigned long index = iocb->ki_pos >> huge_page_shift(h);
263         unsigned long offset = iocb->ki_pos & ~huge_page_mask(h);
264         unsigned long end_index;
265         loff_t isize;
266         ssize_t retval = 0;
267 
268         while (iov_iter_count(to)) {
269                 struct page *page;
270                 size_t nr, copied;
271 
272                 /* nr is the maximum number of bytes to copy from this page */
273                 nr = huge_page_size(h);
274                 isize = i_size_read(inode);
275                 if (!isize)
276                         break;
277                 end_index = (isize - 1) >> huge_page_shift(h);
278                 if (index > end_index)
279                         break;
280                 if (index == end_index) {
281                         nr = ((isize - 1) & ~huge_page_mask(h)) + 1;
282                         if (nr <= offset)
283                                 break;
284                 }
285                 nr = nr - offset;
286 
287                 /* Find the page */
288                 page = find_lock_page(mapping, index);
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                          */
294                         copied = iov_iter_zero(nr, to);
295                 } else {
296                         unlock_page(page);
297 
298                         /*
299                          * We have the page, copy it to user space buffer.
300                          */
301                         copied = hugetlbfs_read_actor(page, offset, to, nr);
302                         put_page(page);
303                 }
304                 offset += copied;
305                 retval += copied;
306                 if (copied != nr && iov_iter_count(to)) {
307                         if (!retval)
308                                 retval = -EFAULT;
309                         break;
310                 }
311                 index += offset >> huge_page_shift(h);
312                 offset &= ~huge_page_mask(h);
313         }
314         iocb->ki_pos = ((loff_t)index << huge_page_shift(h)) + offset;
315         return retval;
316 }
317 
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)
322 {
323         return -EINVAL;
324 }
325 
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)
329 {
330         BUG();
331         return -EINVAL;
332 }
333 
334 static void remove_huge_page(struct page *page)
335 {
336         ClearPageDirty(page);
337         ClearPageUptodate(page);
338         delete_from_page_cache(page);
339 }
340 
341 static void
342 hugetlb_vmdelete_list(struct rb_root_cached *root, pgoff_t start, pgoff_t end)
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 }
378 
379 /*
380  * remove_inode_hugepages handles two distinct cases: truncation and hole
381  * punch.  There are subtle differences in operation for each case.
382  *
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
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.
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
395  *      pages are not modified.  Page faults can race with hole punch.
396  *      This is indicated if we find a mapped page.
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)
402 {
403         struct hstate *h = hstate_inode(inode);
404         struct address_space *mapping = &inode->i_data;
405         const pgoff_t start = lstart >> huge_page_shift(h);
406         const pgoff_t end = lend >> huge_page_shift(h);
407         struct vm_area_struct pseudo_vma;
408         struct pagevec pvec;
409         pgoff_t next, index;
410         int i, freed = 0;
411         bool truncate_op = (lend == LLONG_MAX);
412 
413         vma_init(&pseudo_vma, current->mm);
414         pseudo_vma.vm_flags = (VM_HUGETLB | VM_MAYSHARE | VM_SHARED);
415         pagevec_init(&pvec);
416         next = start;
417         while (next < end) {
418                 /*
419                  * When no more pages are found, we are done.
420                  */
421                 if (!pagevec_lookup_range(&pvec, mapping, &next, end - 1))
422                         break;
423 
424                 for (i = 0; i < pagevec_count(&pvec); ++i) {
425                         struct page *page = pvec.pages[i];
426                         u32 hash;
427 
428                         index = page->index;
429                         hash = hugetlb_fault_mutex_hash(h, mapping, index, 0);
430                         mutex_lock(&hugetlb_fault_mutex_table[hash]);
431 
432                         /*
433                          * If page is mapped, it was faulted in after being
434                          * unmapped in caller.  Unmap (again) now after taking
435                          * the fault mutex.  The mutex will prevent faults
436                          * until we finish removing the page.
437                          *
438                          * This race can only happen in the hole punch case.
439                          * Getting here in a truncate operation is a bug.
440                          */
441                         if (unlikely(page_mapped(page))) {
442                                 BUG_ON(truncate_op);
443 
444                                 i_mmap_lock_write(mapping);
445                                 hugetlb_vmdelete_list(&mapping->i_mmap,
446                                         index * pages_per_huge_page(h),
447                                         (index + 1) * pages_per_huge_page(h));
448                                 i_mmap_unlock_write(mapping);
449                         }
450 
451                         lock_page(page);
452                         /*
453                          * We must free the huge page and remove from page
454                          * cache (remove_huge_page) BEFORE removing the
455                          * region/reserve map (hugetlb_unreserve_pages).  In
456                          * rare out of memory conditions, removal of the
457                          * region/reserve map could fail. Correspondingly,
458                          * the subpool and global reserve usage count can need
459                          * to be adjusted.
460                          */
461                         VM_BUG_ON(PagePrivate(page));
462                         remove_huge_page(page);
463                         freed++;
464                         if (!truncate_op) {
465                                 if (unlikely(hugetlb_unreserve_pages(inode,
466                                                         index, index + 1, 1)))
467                                         hugetlb_fix_reserve_counts(inode);
468                         }
469 
470                         unlock_page(page);
471                         mutex_unlock(&hugetlb_fault_mutex_table[hash]);
472                 }
473                 huge_pagevec_release(&pvec);
474                 cond_resched();
475         }
476 
477         if (truncate_op)
478                 (void)hugetlb_unreserve_pages(inode, start, LONG_MAX, freed);
479 }
480 
481 static void hugetlbfs_evict_inode(struct inode *inode)
482 {
483         struct resv_map *resv_map;
484 
485         remove_inode_hugepages(inode, 0, LLONG_MAX);
486         resv_map = (struct resv_map *)inode->i_mapping->private_data;
487         /* root inode doesn't have the resv_map, so we should check it */
488         if (resv_map)
489                 resv_map_release(&resv_map->refs);
490         clear_inode(inode);
491 }
492 
493 static int hugetlb_vmtruncate(struct inode *inode, loff_t offset)
494 {
495         pgoff_t pgoff;
496         struct address_space *mapping = inode->i_mapping;
497         struct hstate *h = hstate_inode(inode);
498 
499         BUG_ON(offset & ~huge_page_mask(h));
500         pgoff = offset >> PAGE_SHIFT;
501 
502         i_size_write(inode, offset);
503         i_mmap_lock_write(mapping);
504         if (!RB_EMPTY_ROOT(&mapping->i_mmap.rb_root))
505                 hugetlb_vmdelete_list(&mapping->i_mmap, pgoff, 0);
506         i_mmap_unlock_write(mapping);
507         remove_inode_hugepages(inode, offset, LLONG_MAX);
508         return 0;
509 }
510 
511 static long hugetlbfs_punch_hole(struct inode *inode, loff_t offset, loff_t len)
512 {
513         struct hstate *h = hstate_inode(inode);
514         loff_t hpage_size = huge_page_size(h);
515         loff_t hole_start, hole_end;
516 
517         /*
518          * For hole punch round up the beginning offset of the hole and
519          * round down the end.
520          */
521         hole_start = round_up(offset, hpage_size);
522         hole_end = round_down(offset + len, hpage_size);
523 
524         if (hole_end > hole_start) {
525                 struct address_space *mapping = inode->i_mapping;
526                 struct hugetlbfs_inode_info *info = HUGETLBFS_I(inode);
527 
528                 inode_lock(inode);
529 
530                 /* protected by i_mutex */
531                 if (info->seals & F_SEAL_WRITE) {
532                         inode_unlock(inode);
533                         return -EPERM;
534                 }
535 
536                 i_mmap_lock_write(mapping);
537                 if (!RB_EMPTY_ROOT(&mapping->i_mmap.rb_root))
538                         hugetlb_vmdelete_list(&mapping->i_mmap,
539                                                 hole_start >> PAGE_SHIFT,
540                                                 hole_end  >> PAGE_SHIFT);
541                 i_mmap_unlock_write(mapping);
542                 remove_inode_hugepages(inode, hole_start, hole_end);
543                 inode_unlock(inode);
544         }
545 
546         return 0;
547 }
548 
549 static long hugetlbfs_fallocate(struct file *file, int mode, loff_t offset,
550                                 loff_t len)
551 {
552         struct inode *inode = file_inode(file);
553         struct hugetlbfs_inode_info *info = HUGETLBFS_I(inode);
554         struct address_space *mapping = inode->i_mapping;
555         struct hstate *h = hstate_inode(inode);
556         struct vm_area_struct pseudo_vma;
557         struct mm_struct *mm = current->mm;
558         loff_t hpage_size = huge_page_size(h);
559         unsigned long hpage_shift = huge_page_shift(h);
560         pgoff_t start, index, end;
561         int error;
562         u32 hash;
563 
564         if (mode & ~(FALLOC_FL_KEEP_SIZE | FALLOC_FL_PUNCH_HOLE))
565                 return -EOPNOTSUPP;
566 
567         if (mode & FALLOC_FL_PUNCH_HOLE)
568                 return hugetlbfs_punch_hole(inode, offset, len);
569 
570         /*
571          * Default preallocate case.
572          * For this range, start is rounded down and end is rounded up
573          * as well as being converted to page offsets.
574          */
575         start = offset >> hpage_shift;
576         end = (offset + len + hpage_size - 1) >> hpage_shift;
577 
578         inode_lock(inode);
579 
580         /* We need to check rlimit even when FALLOC_FL_KEEP_SIZE */
581         error = inode_newsize_ok(inode, offset + len);
582         if (error)
583                 goto out;
584 
585         if ((info->seals & F_SEAL_GROW) && offset + len > inode->i_size) {
586                 error = -EPERM;
587                 goto out;
588         }
589 
590         /*
591          * Initialize a pseudo vma as this is required by the huge page
592          * allocation routines.  If NUMA is configured, use page index
593          * as input to create an allocation policy.
594          */
595         vma_init(&pseudo_vma, mm);
596         pseudo_vma.vm_flags = (VM_HUGETLB | VM_MAYSHARE | VM_SHARED);
597         pseudo_vma.vm_file = file;
598 
599         for (index = start; index < end; index++) {
600                 /*
601                  * This is supposed to be the vaddr where the page is being
602                  * faulted in, but we have no vaddr here.
603                  */
604                 struct page *page;
605                 unsigned long addr;
606                 int avoid_reserve = 0;
607 
608                 cond_resched();
609 
610                 /*
611                  * fallocate(2) manpage permits EINTR; we may have been
612                  * interrupted because we are using up too much memory.
613                  */
614                 if (signal_pending(current)) {
615                         error = -EINTR;
616                         break;
617                 }
618 
619                 /* Set numa allocation policy based on index */
620                 hugetlb_set_vma_policy(&pseudo_vma, inode, index);
621 
622                 /* addr is the offset within the file (zero based) */
623                 addr = index * hpage_size;
624 
625                 /* mutex taken here, fault path and hole punch */
626                 hash = hugetlb_fault_mutex_hash(h, mapping, index, addr);
627                 mutex_lock(&hugetlb_fault_mutex_table[hash]);
628 
629                 /* See if already present in mapping to avoid alloc/free */
630                 page = find_get_page(mapping, index);
631                 if (page) {
632                         put_page(page);
633                         mutex_unlock(&hugetlb_fault_mutex_table[hash]);
634                         hugetlb_drop_vma_policy(&pseudo_vma);
635                         continue;
636                 }
637 
638                 /* Allocate page and add to page cache */
639                 page = alloc_huge_page(&pseudo_vma, addr, avoid_reserve);
640                 hugetlb_drop_vma_policy(&pseudo_vma);
641                 if (IS_ERR(page)) {
642                         mutex_unlock(&hugetlb_fault_mutex_table[hash]);
643                         error = PTR_ERR(page);
644                         goto out;
645                 }
646                 clear_huge_page(page, addr, pages_per_huge_page(h));
647                 __SetPageUptodate(page);
648                 error = huge_add_to_page_cache(page, mapping, index);
649                 if (unlikely(error)) {
650                         put_page(page);
651                         mutex_unlock(&hugetlb_fault_mutex_table[hash]);
652                         goto out;
653                 }
654 
655                 mutex_unlock(&hugetlb_fault_mutex_table[hash]);
656 
657                 /*
658                  * unlock_page because locked by add_to_page_cache()
659                  * page_put due to reference from alloc_huge_page()
660                  */
661                 unlock_page(page);
662                 put_page(page);
663         }
664 
665         if (!(mode & FALLOC_FL_KEEP_SIZE) && offset + len > inode->i_size)
666                 i_size_write(inode, offset + len);
667         inode->i_ctime = current_time(inode);
668 out:
669         inode_unlock(inode);
670         return error;
671 }
672 
673 static int hugetlbfs_setattr(struct dentry *dentry, struct iattr *attr)
674 {
675         struct inode *inode = d_inode(dentry);
676         struct hstate *h = hstate_inode(inode);
677         int error;
678         unsigned int ia_valid = attr->ia_valid;
679         struct hugetlbfs_inode_info *info = HUGETLBFS_I(inode);
680 
681         BUG_ON(!inode);
682 
683         error = setattr_prepare(dentry, attr);
684         if (error)
685                 return error;
686 
687         if (ia_valid & ATTR_SIZE) {
688                 loff_t oldsize = inode->i_size;
689                 loff_t newsize = attr->ia_size;
690 
691                 if (newsize & ~huge_page_mask(h))
692                         return -EINVAL;
693                 /* protected by i_mutex */
694                 if ((newsize < oldsize && (info->seals & F_SEAL_SHRINK)) ||
695                     (newsize > oldsize && (info->seals & F_SEAL_GROW)))
696                         return -EPERM;
697                 error = hugetlb_vmtruncate(inode, newsize);
698                 if (error)
699                         return error;
700         }
701 
702         setattr_copy(inode, attr);
703         mark_inode_dirty(inode);
704         return 0;
705 }
706 
707 static struct inode *hugetlbfs_get_root(struct super_block *sb,
708                                         struct hugetlbfs_config *config)
709 {
710         struct inode *inode;
711 
712         inode = new_inode(sb);
713         if (inode) {
714                 inode->i_ino = get_next_ino();
715                 inode->i_mode = S_IFDIR | config->mode;
716                 inode->i_uid = config->uid;
717                 inode->i_gid = config->gid;
718                 inode->i_atime = inode->i_mtime = inode->i_ctime = current_time(inode);
719                 inode->i_op = &hugetlbfs_dir_inode_operations;
720                 inode->i_fop = &simple_dir_operations;
721                 /* directory inodes start off with i_nlink == 2 (for "." entry) */
722                 inc_nlink(inode);
723                 lockdep_annotate_inode_mutex_key(inode);
724         }
725         return inode;
726 }
727 
728 /*
729  * Hugetlbfs is not reclaimable; therefore its i_mmap_rwsem will never
730  * be taken from reclaim -- unlike regular filesystems. This needs an
731  * annotation because huge_pmd_share() does an allocation under hugetlb's
732  * i_mmap_rwsem.
733  */
734 static struct lock_class_key hugetlbfs_i_mmap_rwsem_key;
735 
736 static struct inode *hugetlbfs_get_inode(struct super_block *sb,
737                                         struct inode *dir,
738                                         umode_t mode, dev_t dev)
739 {
740         struct inode *inode;
741         struct resv_map *resv_map = NULL;
742 
743         /*
744          * Reserve maps are only needed for inodes that can have associated
745          * page allocations.
746          */
747         if (S_ISREG(mode) || S_ISLNK(mode)) {
748                 resv_map = resv_map_alloc();
749                 if (!resv_map)
750                         return NULL;
751         }
752 
753         inode = new_inode(sb);
754         if (inode) {
755                 struct hugetlbfs_inode_info *info = HUGETLBFS_I(inode);
756 
757                 inode->i_ino = get_next_ino();
758                 inode_init_owner(inode, dir, mode);
759                 lockdep_set_class(&inode->i_mapping->i_mmap_rwsem,
760                                 &hugetlbfs_i_mmap_rwsem_key);
761                 inode->i_mapping->a_ops = &hugetlbfs_aops;
762                 inode->i_atime = inode->i_mtime = inode->i_ctime = current_time(inode);
763                 inode->i_mapping->private_data = resv_map;
764                 info->seals = F_SEAL_SEAL;
765                 switch (mode & S_IFMT) {
766                 default:
767                         init_special_inode(inode, mode, dev);
768                         break;
769                 case S_IFREG:
770                         inode->i_op = &hugetlbfs_inode_operations;
771                         inode->i_fop = &hugetlbfs_file_operations;
772                         break;
773                 case S_IFDIR:
774                         inode->i_op = &hugetlbfs_dir_inode_operations;
775                         inode->i_fop = &simple_dir_operations;
776 
777                         /* directory inodes start off with i_nlink == 2 (for "." entry) */
778                         inc_nlink(inode);
779                         break;
780                 case S_IFLNK:
781                         inode->i_op = &page_symlink_inode_operations;
782                         inode_nohighmem(inode);
783                         break;
784                 }
785                 lockdep_annotate_inode_mutex_key(inode);
786         } else {
787                 if (resv_map)
788                         kref_put(&resv_map->refs, resv_map_release);
789         }
790 
791         return inode;
792 }
793 
794 /*
795  * File creation. Allocate an inode, and we're done..
796  */
797 static int hugetlbfs_mknod(struct inode *dir,
798                         struct dentry *dentry, umode_t mode, dev_t dev)
799 {
800         struct inode *inode;
801         int error = -ENOSPC;
802 
803         inode = hugetlbfs_get_inode(dir->i_sb, dir, mode, dev);
804         if (inode) {
805                 dir->i_ctime = dir->i_mtime = current_time(dir);
806                 d_instantiate(dentry, inode);
807                 dget(dentry);   /* Extra count - pin the dentry in core */
808                 error = 0;
809         }
810         return error;
811 }
812 
813 static int hugetlbfs_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode)
814 {
815         int retval = hugetlbfs_mknod(dir, dentry, mode | S_IFDIR, 0);
816         if (!retval)
817                 inc_nlink(dir);
818         return retval;
819 }
820 
821 static int hugetlbfs_create(struct inode *dir, struct dentry *dentry, umode_t mode, bool excl)
822 {
823         return hugetlbfs_mknod(dir, dentry, mode | S_IFREG, 0);
824 }
825 
826 static int hugetlbfs_symlink(struct inode *dir,
827                         struct dentry *dentry, const char *symname)
828 {
829         struct inode *inode;
830         int error = -ENOSPC;
831 
832         inode = hugetlbfs_get_inode(dir->i_sb, dir, S_IFLNK|S_IRWXUGO, 0);
833         if (inode) {
834                 int l = strlen(symname)+1;
835                 error = page_symlink(inode, symname, l);
836                 if (!error) {
837                         d_instantiate(dentry, inode);
838                         dget(dentry);
839                 } else
840                         iput(inode);
841         }
842         dir->i_ctime = dir->i_mtime = current_time(dir);
843 
844         return error;
845 }
846 
847 /*
848  * mark the head page dirty
849  */
850 static int hugetlbfs_set_page_dirty(struct page *page)
851 {
852         struct page *head = compound_head(page);
853 
854         SetPageDirty(head);
855         return 0;
856 }
857 
858 static int hugetlbfs_migrate_page(struct address_space *mapping,
859                                 struct page *newpage, struct page *page,
860                                 enum migrate_mode mode)
861 {
862         int rc;
863 
864         rc = migrate_huge_page_move_mapping(mapping, newpage, page);
865         if (rc != MIGRATEPAGE_SUCCESS)
866                 return rc;
867 
868         /*
869          * page_private is subpool pointer in hugetlb pages.  Transfer to
870          * new page.  PagePrivate is not associated with page_private for
871          * hugetlb pages and can not be set here as only page_huge_active
872          * pages can be migrated.
873          */
874         if (page_private(page)) {
875                 set_page_private(newpage, page_private(page));
876                 set_page_private(page, 0);
877         }
878 
879         if (mode != MIGRATE_SYNC_NO_COPY)
880                 migrate_page_copy(newpage, page);
881         else
882                 migrate_page_states(newpage, page);
883 
884         return MIGRATEPAGE_SUCCESS;
885 }
886 
887 static int hugetlbfs_error_remove_page(struct address_space *mapping,
888                                 struct page *page)
889 {
890         struct inode *inode = mapping->host;
891         pgoff_t index = page->index;
892 
893         remove_huge_page(page);
894         if (unlikely(hugetlb_unreserve_pages(inode, index, index + 1, 1)))
895                 hugetlb_fix_reserve_counts(inode);
896 
897         return 0;
898 }
899 
900 /*
901  * Display the mount options in /proc/mounts.
902  */
903 static int hugetlbfs_show_options(struct seq_file *m, struct dentry *root)
904 {
905         struct hugetlbfs_sb_info *sbinfo = HUGETLBFS_SB(root->d_sb);
906         struct hugepage_subpool *spool = sbinfo->spool;
907         unsigned long hpage_size = huge_page_size(sbinfo->hstate);
908         unsigned hpage_shift = huge_page_shift(sbinfo->hstate);
909         char mod;
910 
911         if (!uid_eq(sbinfo->uid, GLOBAL_ROOT_UID))
912                 seq_printf(m, ",uid=%u",
913                            from_kuid_munged(&init_user_ns, sbinfo->uid));
914         if (!gid_eq(sbinfo->gid, GLOBAL_ROOT_GID))
915                 seq_printf(m, ",gid=%u",
916                            from_kgid_munged(&init_user_ns, sbinfo->gid));
917         if (sbinfo->mode != 0755)
918                 seq_printf(m, ",mode=%o", sbinfo->mode);
919         if (sbinfo->max_inodes != -1)
920                 seq_printf(m, ",nr_inodes=%lu", sbinfo->max_inodes);
921 
922         hpage_size /= 1024;
923         mod = 'K';
924         if (hpage_size >= 1024) {
925                 hpage_size /= 1024;
926                 mod = 'M';
927         }
928         seq_printf(m, ",pagesize=%lu%c", hpage_size, mod);
929         if (spool) {
930                 if (spool->max_hpages != -1)
931                         seq_printf(m, ",size=%llu",
932                                    (unsigned long long)spool->max_hpages << hpage_shift);
933                 if (spool->min_hpages != -1)
934                         seq_printf(m, ",min_size=%llu",
935                                    (unsigned long long)spool->min_hpages << hpage_shift);
936         }
937         return 0;
938 }
939 
940 static int hugetlbfs_statfs(struct dentry *dentry, struct kstatfs *buf)
941 {
942         struct hugetlbfs_sb_info *sbinfo = HUGETLBFS_SB(dentry->d_sb);
943         struct hstate *h = hstate_inode(d_inode(dentry));
944 
945         buf->f_type = HUGETLBFS_MAGIC;
946         buf->f_bsize = huge_page_size(h);
947         if (sbinfo) {
948                 spin_lock(&sbinfo->stat_lock);
949                 /* If no limits set, just report 0 for max/free/used
950                  * blocks, like simple_statfs() */
951                 if (sbinfo->spool) {
952                         long free_pages;
953 
954                         spin_lock(&sbinfo->spool->lock);
955                         buf->f_blocks = sbinfo->spool->max_hpages;
956                         free_pages = sbinfo->spool->max_hpages
957                                 - sbinfo->spool->used_hpages;
958                         buf->f_bavail = buf->f_bfree = free_pages;
959                         spin_unlock(&sbinfo->spool->lock);
960                         buf->f_files = sbinfo->max_inodes;
961                         buf->f_ffree = sbinfo->free_inodes;
962                 }
963                 spin_unlock(&sbinfo->stat_lock);
964         }
965         buf->f_namelen = NAME_MAX;
966         return 0;
967 }
968 
969 static void hugetlbfs_put_super(struct super_block *sb)
970 {
971         struct hugetlbfs_sb_info *sbi = HUGETLBFS_SB(sb);
972 
973         if (sbi) {
974                 sb->s_fs_info = NULL;
975 
976                 if (sbi->spool)
977                         hugepage_put_subpool(sbi->spool);
978 
979                 kfree(sbi);
980         }
981 }
982 
983 static inline int hugetlbfs_dec_free_inodes(struct hugetlbfs_sb_info *sbinfo)
984 {
985         if (sbinfo->free_inodes >= 0) {
986                 spin_lock(&sbinfo->stat_lock);
987                 if (unlikely(!sbinfo->free_inodes)) {
988                         spin_unlock(&sbinfo->stat_lock);
989                         return 0;
990                 }
991                 sbinfo->free_inodes--;
992                 spin_unlock(&sbinfo->stat_lock);
993         }
994 
995         return 1;
996 }
997 
998 static void hugetlbfs_inc_free_inodes(struct hugetlbfs_sb_info *sbinfo)
999 {
1000         if (sbinfo->free_inodes >= 0) {
1001                 spin_lock(&sbinfo->stat_lock);
1002                 sbinfo->free_inodes++;
1003                 spin_unlock(&sbinfo->stat_lock);
1004         }
1005 }
1006 
1007 
1008 static struct kmem_cache *hugetlbfs_inode_cachep;
1009 
1010 static struct inode *hugetlbfs_alloc_inode(struct super_block *sb)
1011 {
1012         struct hugetlbfs_sb_info *sbinfo = HUGETLBFS_SB(sb);
1013         struct hugetlbfs_inode_info *p;
1014 
1015         if (unlikely(!hugetlbfs_dec_free_inodes(sbinfo)))
1016                 return NULL;
1017         p = kmem_cache_alloc(hugetlbfs_inode_cachep, GFP_KERNEL);
1018         if (unlikely(!p)) {
1019                 hugetlbfs_inc_free_inodes(sbinfo);
1020                 return NULL;
1021         }
1022 
1023         /*
1024          * Any time after allocation, hugetlbfs_destroy_inode can be called
1025          * for the inode.  mpol_free_shared_policy is unconditionally called
1026          * as part of hugetlbfs_destroy_inode.  So, initialize policy here
1027          * in case of a quick call to destroy.
1028          *
1029          * Note that the policy is initialized even if we are creating a
1030          * private inode.  This simplifies hugetlbfs_destroy_inode.
1031          */
1032         mpol_shared_policy_init(&p->policy, NULL);
1033 
1034         return &p->vfs_inode;
1035 }
1036 
1037 static void hugetlbfs_i_callback(struct rcu_head *head)
1038 {
1039         struct inode *inode = container_of(head, struct inode, i_rcu);
1040         kmem_cache_free(hugetlbfs_inode_cachep, HUGETLBFS_I(inode));
1041 }
1042 
1043 static void hugetlbfs_destroy_inode(struct inode *inode)
1044 {
1045         hugetlbfs_inc_free_inodes(HUGETLBFS_SB(inode->i_sb));
1046         mpol_free_shared_policy(&HUGETLBFS_I(inode)->policy);
1047         call_rcu(&inode->i_rcu, hugetlbfs_i_callback);
1048 }
1049 
1050 static const struct address_space_operations hugetlbfs_aops = {
1051         .write_begin    = hugetlbfs_write_begin,
1052         .write_end      = hugetlbfs_write_end,
1053         .set_page_dirty = hugetlbfs_set_page_dirty,
1054         .migratepage    = hugetlbfs_migrate_page,
1055         .error_remove_page      = hugetlbfs_error_remove_page,
1056 };
1057 
1058 
1059 static void init_once(void *foo)
1060 {
1061         struct hugetlbfs_inode_info *ei = (struct hugetlbfs_inode_info *)foo;
1062 
1063         inode_init_once(&ei->vfs_inode);
1064 }
1065 
1066 const struct file_operations hugetlbfs_file_operations = {
1067         .read_iter              = hugetlbfs_read_iter,
1068         .mmap                   = hugetlbfs_file_mmap,
1069         .fsync                  = noop_fsync,
1070         .get_unmapped_area      = hugetlb_get_unmapped_area,
1071         .llseek                 = default_llseek,
1072         .fallocate              = hugetlbfs_fallocate,
1073 };
1074 
1075 static const struct inode_operations hugetlbfs_dir_inode_operations = {
1076         .create         = hugetlbfs_create,
1077         .lookup         = simple_lookup,
1078         .link           = simple_link,
1079         .unlink         = simple_unlink,
1080         .symlink        = hugetlbfs_symlink,
1081         .mkdir          = hugetlbfs_mkdir,
1082         .rmdir          = simple_rmdir,
1083         .mknod          = hugetlbfs_mknod,
1084         .rename         = simple_rename,
1085         .setattr        = hugetlbfs_setattr,
1086 };
1087 
1088 static const struct inode_operations hugetlbfs_inode_operations = {
1089         .setattr        = hugetlbfs_setattr,
1090 };
1091 
1092 static const struct super_operations hugetlbfs_ops = {
1093         .alloc_inode    = hugetlbfs_alloc_inode,
1094         .destroy_inode  = hugetlbfs_destroy_inode,
1095         .evict_inode    = hugetlbfs_evict_inode,
1096         .statfs         = hugetlbfs_statfs,
1097         .put_super      = hugetlbfs_put_super,
1098         .show_options   = hugetlbfs_show_options,
1099 };
1100 
1101 enum hugetlbfs_size_type { NO_SIZE, SIZE_STD, SIZE_PERCENT };
1102 
1103 /*
1104  * Convert size option passed from command line to number of huge pages
1105  * in the pool specified by hstate.  Size option could be in bytes
1106  * (val_type == SIZE_STD) or percentage of the pool (val_type == SIZE_PERCENT).
1107  */
1108 static long
1109 hugetlbfs_size_to_hpages(struct hstate *h, unsigned long long size_opt,
1110                          enum hugetlbfs_size_type val_type)
1111 {
1112         if (val_type == NO_SIZE)
1113                 return -1;
1114 
1115         if (val_type == SIZE_PERCENT) {
1116                 size_opt <<= huge_page_shift(h);
1117                 size_opt *= h->max_huge_pages;
1118                 do_div(size_opt, 100);
1119         }
1120 
1121         size_opt >>= huge_page_shift(h);
1122         return size_opt;
1123 }
1124 
1125 static int
1126 hugetlbfs_parse_options(char *options, struct hugetlbfs_config *pconfig)
1127 {
1128         char *p, *rest;
1129         substring_t args[MAX_OPT_ARGS];
1130         int option;
1131         unsigned long long max_size_opt = 0, min_size_opt = 0;
1132         enum hugetlbfs_size_type max_val_type = NO_SIZE, min_val_type = NO_SIZE;
1133 
1134         if (!options)
1135                 return 0;
1136 
1137         while ((p = strsep(&options, ",")) != NULL) {
1138                 int token;
1139                 if (!*p)
1140                         continue;
1141 
1142                 token = match_token(p, tokens, args);
1143                 switch (token) {
1144                 case Opt_uid:
1145                         if (match_int(&args[0], &option))
1146                                 goto bad_val;
1147                         pconfig->uid = make_kuid(current_user_ns(), option);
1148                         if (!uid_valid(pconfig->uid))
1149                                 goto bad_val;
1150                         break;
1151 
1152                 case Opt_gid:
1153                         if (match_int(&args[0], &option))
1154                                 goto bad_val;
1155                         pconfig->gid = make_kgid(current_user_ns(), option);
1156                         if (!gid_valid(pconfig->gid))
1157                                 goto bad_val;
1158                         break;
1159 
1160                 case Opt_mode:
1161                         if (match_octal(&args[0], &option))
1162                                 goto bad_val;
1163                         pconfig->mode = option & 01777U;
1164                         break;
1165 
1166                 case Opt_size: {
1167                         /* memparse() will accept a K/M/G without a digit */
1168                         if (!isdigit(*args[0].from))
1169                                 goto bad_val;
1170                         max_size_opt = memparse(args[0].from, &rest);
1171                         max_val_type = SIZE_STD;
1172                         if (*rest == '%')
1173                                 max_val_type = SIZE_PERCENT;
1174                         break;
1175                 }
1176 
1177                 case Opt_nr_inodes:
1178                         /* memparse() will accept a K/M/G without a digit */
1179                         if (!isdigit(*args[0].from))
1180                                 goto bad_val;
1181                         pconfig->nr_inodes = memparse(args[0].from, &rest);
1182                         break;
1183 
1184                 case Opt_pagesize: {
1185                         unsigned long ps;
1186                         ps = memparse(args[0].from, &rest);
1187                         pconfig->hstate = size_to_hstate(ps);
1188                         if (!pconfig->hstate) {
1189                                 pr_err("Unsupported page size %lu MB\n",
1190                                         ps >> 20);
1191                                 return -EINVAL;
1192                         }
1193                         break;
1194                 }
1195 
1196                 case Opt_min_size: {
1197                         /* memparse() will accept a K/M/G without a digit */
1198                         if (!isdigit(*args[0].from))
1199                                 goto bad_val;
1200                         min_size_opt = memparse(args[0].from, &rest);
1201                         min_val_type = SIZE_STD;
1202                         if (*rest == '%')
1203                                 min_val_type = SIZE_PERCENT;
1204                         break;
1205                 }
1206 
1207                 default:
1208                         pr_err("Bad mount option: \"%s\"\n", p);
1209                         return -EINVAL;
1210                         break;
1211                 }
1212         }
1213 
1214         /*
1215          * Use huge page pool size (in hstate) to convert the size
1216          * options to number of huge pages.  If NO_SIZE, -1 is returned.
1217          */
1218         pconfig->max_hpages = hugetlbfs_size_to_hpages(pconfig->hstate,
1219                                                 max_size_opt, max_val_type);
1220         pconfig->min_hpages = hugetlbfs_size_to_hpages(pconfig->hstate,
1221                                                 min_size_opt, min_val_type);
1222 
1223         /*
1224          * If max_size was specified, then min_size must be smaller
1225          */
1226         if (max_val_type > NO_SIZE &&
1227             pconfig->min_hpages > pconfig->max_hpages) {
1228                 pr_err("minimum size can not be greater than maximum size\n");
1229                 return -EINVAL;
1230         }
1231 
1232         return 0;
1233 
1234 bad_val:
1235         pr_err("Bad value '%s' for mount option '%s'\n", args[0].from, p);
1236         return -EINVAL;
1237 }
1238 
1239 static int
1240 hugetlbfs_fill_super(struct super_block *sb, void *data, int silent)
1241 {
1242         int ret;
1243         struct hugetlbfs_config config;
1244         struct hugetlbfs_sb_info *sbinfo;
1245 
1246         config.max_hpages = -1; /* No limit on size by default */
1247         config.nr_inodes = -1; /* No limit on number of inodes by default */
1248         config.uid = current_fsuid();
1249         config.gid = current_fsgid();
1250         config.mode = 0755;
1251         config.hstate = &default_hstate;
1252         config.min_hpages = -1; /* No default minimum size */
1253         ret = hugetlbfs_parse_options(data, &config);
1254         if (ret)
1255                 return ret;
1256 
1257         sbinfo = kmalloc(sizeof(struct hugetlbfs_sb_info), GFP_KERNEL);
1258         if (!sbinfo)
1259                 return -ENOMEM;
1260         sb->s_fs_info = sbinfo;
1261         sbinfo->hstate = config.hstate;
1262         spin_lock_init(&sbinfo->stat_lock);
1263         sbinfo->max_inodes = config.nr_inodes;
1264         sbinfo->free_inodes = config.nr_inodes;
1265         sbinfo->spool = NULL;
1266         sbinfo->uid = config.uid;
1267         sbinfo->gid = config.gid;
1268         sbinfo->mode = config.mode;
1269 
1270         /*
1271          * Allocate and initialize subpool if maximum or minimum size is
1272          * specified.  Any needed reservations (for minimim size) are taken
1273          * taken when the subpool is created.
1274          */
1275         if (config.max_hpages != -1 || config.min_hpages != -1) {
1276                 sbinfo->spool = hugepage_new_subpool(config.hstate,
1277                                                         config.max_hpages,
1278                                                         config.min_hpages);
1279                 if (!sbinfo->spool)
1280                         goto out_free;
1281         }
1282         sb->s_maxbytes = MAX_LFS_FILESIZE;
1283         sb->s_blocksize = huge_page_size(config.hstate);
1284         sb->s_blocksize_bits = huge_page_shift(config.hstate);
1285         sb->s_magic = HUGETLBFS_MAGIC;
1286         sb->s_op = &hugetlbfs_ops;
1287         sb->s_time_gran = 1;
1288         sb->s_root = d_make_root(hugetlbfs_get_root(sb, &config));
1289         if (!sb->s_root)
1290                 goto out_free;
1291         return 0;
1292 out_free:
1293         kfree(sbinfo->spool);
1294         kfree(sbinfo);
1295         return -ENOMEM;
1296 }
1297 
1298 static struct dentry *hugetlbfs_mount(struct file_system_type *fs_type,
1299         int flags, const char *dev_name, void *data)
1300 {
1301         return mount_nodev(fs_type, flags, data, hugetlbfs_fill_super);
1302 }
1303 
1304 static struct file_system_type hugetlbfs_fs_type = {
1305         .name           = "hugetlbfs",
1306         .mount          = hugetlbfs_mount,
1307         .kill_sb        = kill_litter_super,
1308 };
1309 
1310 static struct vfsmount *hugetlbfs_vfsmount[HUGE_MAX_HSTATE];
1311 
1312 static int can_do_hugetlb_shm(void)
1313 {
1314         kgid_t shm_group;
1315         shm_group = make_kgid(&init_user_ns, sysctl_hugetlb_shm_group);
1316         return capable(CAP_IPC_LOCK) || in_group_p(shm_group);
1317 }
1318 
1319 static int get_hstate_idx(int page_size_log)
1320 {
1321         struct hstate *h = hstate_sizelog(page_size_log);
1322 
1323         if (!h)
1324                 return -1;
1325         return h - hstates;
1326 }
1327 
1328 /*
1329  * Note that size should be aligned to proper hugepage size in caller side,
1330  * otherwise hugetlb_reserve_pages reserves one less hugepages than intended.
1331  */
1332 struct file *hugetlb_file_setup(const char *name, size_t size,
1333                                 vm_flags_t acctflag, struct user_struct **user,
1334                                 int creat_flags, int page_size_log)
1335 {
1336         struct inode *inode;
1337         struct vfsmount *mnt;
1338         int hstate_idx;
1339         struct file *file;
1340 
1341         hstate_idx = get_hstate_idx(page_size_log);
1342         if (hstate_idx < 0)
1343                 return ERR_PTR(-ENODEV);
1344 
1345         *user = NULL;
1346         mnt = hugetlbfs_vfsmount[hstate_idx];
1347         if (!mnt)
1348                 return ERR_PTR(-ENOENT);
1349 
1350         if (creat_flags == HUGETLB_SHMFS_INODE && !can_do_hugetlb_shm()) {
1351                 *user = current_user();
1352                 if (user_shm_lock(size, *user)) {
1353                         task_lock(current);
1354                         pr_warn_once("%s (%d): Using mlock ulimits for SHM_HUGETLB is deprecated\n",
1355                                 current->comm, current->pid);
1356                         task_unlock(current);
1357                 } else {
1358                         *user = NULL;
1359                         return ERR_PTR(-EPERM);
1360                 }
1361         }
1362 
1363         file = ERR_PTR(-ENOSPC);
1364         inode = hugetlbfs_get_inode(mnt->mnt_sb, NULL, S_IFREG | S_IRWXUGO, 0);
1365         if (!inode)
1366                 goto out;
1367         if (creat_flags == HUGETLB_SHMFS_INODE)
1368                 inode->i_flags |= S_PRIVATE;
1369 
1370         inode->i_size = size;
1371         clear_nlink(inode);
1372 
1373         if (hugetlb_reserve_pages(inode, 0,
1374                         size >> huge_page_shift(hstate_inode(inode)), NULL,
1375                         acctflag))
1376                 file = ERR_PTR(-ENOMEM);
1377         else
1378                 file = alloc_file_pseudo(inode, mnt, name, O_RDWR,
1379                                         &hugetlbfs_file_operations);
1380         if (!IS_ERR(file))
1381                 return file;
1382 
1383         iput(inode);
1384 out:
1385         if (*user) {
1386                 user_shm_unlock(size, *user);
1387                 *user = NULL;
1388         }
1389         return file;
1390 }
1391 
1392 static int __init init_hugetlbfs_fs(void)
1393 {
1394         struct hstate *h;
1395         int error;
1396         int i;
1397 
1398         if (!hugepages_supported()) {
1399                 pr_info("disabling because there are no supported hugepage sizes\n");
1400                 return -ENOTSUPP;
1401         }
1402 
1403         error = -ENOMEM;
1404         hugetlbfs_inode_cachep = kmem_cache_create("hugetlbfs_inode_cache",
1405                                         sizeof(struct hugetlbfs_inode_info),
1406                                         0, SLAB_ACCOUNT, init_once);
1407         if (hugetlbfs_inode_cachep == NULL)
1408                 goto out2;
1409 
1410         error = register_filesystem(&hugetlbfs_fs_type);
1411         if (error)
1412                 goto out;
1413 
1414         i = 0;
1415         for_each_hstate(h) {
1416                 char buf[50];
1417                 unsigned ps_kb = 1U << (h->order + PAGE_SHIFT - 10);
1418 
1419                 snprintf(buf, sizeof(buf), "pagesize=%uK", ps_kb);
1420                 hugetlbfs_vfsmount[i] = kern_mount_data(&hugetlbfs_fs_type,
1421                                                         buf);
1422 
1423                 if (IS_ERR(hugetlbfs_vfsmount[i])) {
1424                         pr_err("Cannot mount internal hugetlbfs for "
1425                                 "page size %uK", ps_kb);
1426                         error = PTR_ERR(hugetlbfs_vfsmount[i]);
1427                         hugetlbfs_vfsmount[i] = NULL;
1428                 }
1429                 i++;
1430         }
1431         /* Non default hstates are optional */
1432         if (!IS_ERR_OR_NULL(hugetlbfs_vfsmount[default_hstate_idx]))
1433                 return 0;
1434 
1435  out:
1436         kmem_cache_destroy(hugetlbfs_inode_cachep);
1437  out2:
1438         return error;
1439 }
1440 fs_initcall(init_hugetlbfs_fs)
1441 

~ [ source navigation ] ~ [ diff markup ] ~ [ identifier search ] ~

kernel.org | git.kernel.org | LWN.net | Project Home | Wiki (Japanese) | Wiki (English) | SVN repository | Mail admin

Linux® is a registered trademark of Linus Torvalds in the United States and other countries.
TOMOYO® is a registered trademark of NTT DATA CORPORATION.

osdn.jp