Loading...
1/*
2 * Copyright (C) Sistina Software, Inc. 1997-2003 All rights reserved.
3 * Copyright (C) 2004-2006 Red Hat, Inc. All rights reserved.
4 *
5 * This copyrighted material is made available to anyone wishing to use,
6 * modify, copy, or redistribute it subject to the terms and conditions
7 * of the GNU General Public License version 2.
8 */
9
10#include <linux/slab.h>
11#include <linux/spinlock.h>
12#include <linux/completion.h>
13#include <linux/buffer_head.h>
14#include <linux/pagemap.h>
15#include <linux/uio.h>
16#include <linux/blkdev.h>
17#include <linux/mm.h>
18#include <linux/mount.h>
19#include <linux/fs.h>
20#include <linux/gfs2_ondisk.h>
21#include <linux/ext2_fs.h>
22#include <linux/falloc.h>
23#include <linux/swap.h>
24#include <linux/crc32.h>
25#include <linux/writeback.h>
26#include <asm/uaccess.h>
27#include <linux/dlm.h>
28#include <linux/dlm_plock.h>
29
30#include "gfs2.h"
31#include "incore.h"
32#include "bmap.h"
33#include "dir.h"
34#include "glock.h"
35#include "glops.h"
36#include "inode.h"
37#include "log.h"
38#include "meta_io.h"
39#include "quota.h"
40#include "rgrp.h"
41#include "trans.h"
42#include "util.h"
43
44/**
45 * gfs2_llseek - seek to a location in a file
46 * @file: the file
47 * @offset: the offset
48 * @origin: Where to seek from (SEEK_SET, SEEK_CUR, or SEEK_END)
49 *
50 * SEEK_END requires the glock for the file because it references the
51 * file's size.
52 *
53 * Returns: The new offset, or errno
54 */
55
56static loff_t gfs2_llseek(struct file *file, loff_t offset, int origin)
57{
58 struct gfs2_inode *ip = GFS2_I(file->f_mapping->host);
59 struct gfs2_holder i_gh;
60 loff_t error;
61
62 if (origin == 2) {
63 error = gfs2_glock_nq_init(ip->i_gl, LM_ST_SHARED, LM_FLAG_ANY,
64 &i_gh);
65 if (!error) {
66 error = generic_file_llseek_unlocked(file, offset, origin);
67 gfs2_glock_dq_uninit(&i_gh);
68 }
69 } else
70 error = generic_file_llseek_unlocked(file, offset, origin);
71
72 return error;
73}
74
75/**
76 * gfs2_readdir - Read directory entries from a directory
77 * @file: The directory to read from
78 * @dirent: Buffer for dirents
79 * @filldir: Function used to do the copying
80 *
81 * Returns: errno
82 */
83
84static int gfs2_readdir(struct file *file, void *dirent, filldir_t filldir)
85{
86 struct inode *dir = file->f_mapping->host;
87 struct gfs2_inode *dip = GFS2_I(dir);
88 struct gfs2_holder d_gh;
89 u64 offset = file->f_pos;
90 int error;
91
92 gfs2_holder_init(dip->i_gl, LM_ST_SHARED, 0, &d_gh);
93 error = gfs2_glock_nq(&d_gh);
94 if (error) {
95 gfs2_holder_uninit(&d_gh);
96 return error;
97 }
98
99 error = gfs2_dir_read(dir, &offset, dirent, filldir);
100
101 gfs2_glock_dq_uninit(&d_gh);
102
103 file->f_pos = offset;
104
105 return error;
106}
107
108/**
109 * fsflags_cvt
110 * @table: A table of 32 u32 flags
111 * @val: a 32 bit value to convert
112 *
113 * This function can be used to convert between fsflags values and
114 * GFS2's own flags values.
115 *
116 * Returns: the converted flags
117 */
118static u32 fsflags_cvt(const u32 *table, u32 val)
119{
120 u32 res = 0;
121 while(val) {
122 if (val & 1)
123 res |= *table;
124 table++;
125 val >>= 1;
126 }
127 return res;
128}
129
130static const u32 fsflags_to_gfs2[32] = {
131 [3] = GFS2_DIF_SYNC,
132 [4] = GFS2_DIF_IMMUTABLE,
133 [5] = GFS2_DIF_APPENDONLY,
134 [7] = GFS2_DIF_NOATIME,
135 [12] = GFS2_DIF_EXHASH,
136 [14] = GFS2_DIF_INHERIT_JDATA,
137};
138
139static const u32 gfs2_to_fsflags[32] = {
140 [gfs2fl_Sync] = FS_SYNC_FL,
141 [gfs2fl_Immutable] = FS_IMMUTABLE_FL,
142 [gfs2fl_AppendOnly] = FS_APPEND_FL,
143 [gfs2fl_NoAtime] = FS_NOATIME_FL,
144 [gfs2fl_ExHash] = FS_INDEX_FL,
145 [gfs2fl_InheritJdata] = FS_JOURNAL_DATA_FL,
146};
147
148static int gfs2_get_flags(struct file *filp, u32 __user *ptr)
149{
150 struct inode *inode = filp->f_path.dentry->d_inode;
151 struct gfs2_inode *ip = GFS2_I(inode);
152 struct gfs2_holder gh;
153 int error;
154 u32 fsflags;
155
156 gfs2_holder_init(ip->i_gl, LM_ST_SHARED, 0, &gh);
157 error = gfs2_glock_nq(&gh);
158 if (error)
159 return error;
160
161 fsflags = fsflags_cvt(gfs2_to_fsflags, ip->i_diskflags);
162 if (!S_ISDIR(inode->i_mode) && ip->i_diskflags & GFS2_DIF_JDATA)
163 fsflags |= FS_JOURNAL_DATA_FL;
164 if (put_user(fsflags, ptr))
165 error = -EFAULT;
166
167 gfs2_glock_dq(&gh);
168 gfs2_holder_uninit(&gh);
169 return error;
170}
171
172void gfs2_set_inode_flags(struct inode *inode)
173{
174 struct gfs2_inode *ip = GFS2_I(inode);
175 unsigned int flags = inode->i_flags;
176
177 flags &= ~(S_SYNC|S_APPEND|S_IMMUTABLE|S_NOATIME|S_DIRSYNC|S_NOSEC);
178 if ((ip->i_eattr == 0) && !is_sxid(inode->i_mode))
179 inode->i_flags |= S_NOSEC;
180 if (ip->i_diskflags & GFS2_DIF_IMMUTABLE)
181 flags |= S_IMMUTABLE;
182 if (ip->i_diskflags & GFS2_DIF_APPENDONLY)
183 flags |= S_APPEND;
184 if (ip->i_diskflags & GFS2_DIF_NOATIME)
185 flags |= S_NOATIME;
186 if (ip->i_diskflags & GFS2_DIF_SYNC)
187 flags |= S_SYNC;
188 inode->i_flags = flags;
189}
190
191/* Flags that can be set by user space */
192#define GFS2_FLAGS_USER_SET (GFS2_DIF_JDATA| \
193 GFS2_DIF_IMMUTABLE| \
194 GFS2_DIF_APPENDONLY| \
195 GFS2_DIF_NOATIME| \
196 GFS2_DIF_SYNC| \
197 GFS2_DIF_SYSTEM| \
198 GFS2_DIF_INHERIT_JDATA)
199
200/**
201 * gfs2_set_flags - set flags on an inode
202 * @inode: The inode
203 * @flags: The flags to set
204 * @mask: Indicates which flags are valid
205 *
206 */
207static int do_gfs2_set_flags(struct file *filp, u32 reqflags, u32 mask)
208{
209 struct inode *inode = filp->f_path.dentry->d_inode;
210 struct gfs2_inode *ip = GFS2_I(inode);
211 struct gfs2_sbd *sdp = GFS2_SB(inode);
212 struct buffer_head *bh;
213 struct gfs2_holder gh;
214 int error;
215 u32 new_flags, flags;
216
217 error = mnt_want_write(filp->f_path.mnt);
218 if (error)
219 return error;
220
221 error = gfs2_glock_nq_init(ip->i_gl, LM_ST_EXCLUSIVE, 0, &gh);
222 if (error)
223 goto out_drop_write;
224
225 error = -EACCES;
226 if (!inode_owner_or_capable(inode))
227 goto out;
228
229 error = 0;
230 flags = ip->i_diskflags;
231 new_flags = (flags & ~mask) | (reqflags & mask);
232 if ((new_flags ^ flags) == 0)
233 goto out;
234
235 error = -EINVAL;
236 if ((new_flags ^ flags) & ~GFS2_FLAGS_USER_SET)
237 goto out;
238
239 error = -EPERM;
240 if (IS_IMMUTABLE(inode) && (new_flags & GFS2_DIF_IMMUTABLE))
241 goto out;
242 if (IS_APPEND(inode) && (new_flags & GFS2_DIF_APPENDONLY))
243 goto out;
244 if (((new_flags ^ flags) & GFS2_DIF_IMMUTABLE) &&
245 !capable(CAP_LINUX_IMMUTABLE))
246 goto out;
247 if (!IS_IMMUTABLE(inode)) {
248 error = gfs2_permission(inode, MAY_WRITE);
249 if (error)
250 goto out;
251 }
252 if ((flags ^ new_flags) & GFS2_DIF_JDATA) {
253 if (flags & GFS2_DIF_JDATA)
254 gfs2_log_flush(sdp, ip->i_gl);
255 error = filemap_fdatawrite(inode->i_mapping);
256 if (error)
257 goto out;
258 error = filemap_fdatawait(inode->i_mapping);
259 if (error)
260 goto out;
261 }
262 error = gfs2_trans_begin(sdp, RES_DINODE, 0);
263 if (error)
264 goto out;
265 error = gfs2_meta_inode_buffer(ip, &bh);
266 if (error)
267 goto out_trans_end;
268 gfs2_trans_add_bh(ip->i_gl, bh, 1);
269 ip->i_diskflags = new_flags;
270 gfs2_dinode_out(ip, bh->b_data);
271 brelse(bh);
272 gfs2_set_inode_flags(inode);
273 gfs2_set_aops(inode);
274out_trans_end:
275 gfs2_trans_end(sdp);
276out:
277 gfs2_glock_dq_uninit(&gh);
278out_drop_write:
279 mnt_drop_write(filp->f_path.mnt);
280 return error;
281}
282
283static int gfs2_set_flags(struct file *filp, u32 __user *ptr)
284{
285 struct inode *inode = filp->f_path.dentry->d_inode;
286 u32 fsflags, gfsflags;
287
288 if (get_user(fsflags, ptr))
289 return -EFAULT;
290
291 gfsflags = fsflags_cvt(fsflags_to_gfs2, fsflags);
292 if (!S_ISDIR(inode->i_mode)) {
293 if (gfsflags & GFS2_DIF_INHERIT_JDATA)
294 gfsflags ^= (GFS2_DIF_JDATA | GFS2_DIF_INHERIT_JDATA);
295 return do_gfs2_set_flags(filp, gfsflags, ~0);
296 }
297 return do_gfs2_set_flags(filp, gfsflags, ~GFS2_DIF_JDATA);
298}
299
300static long gfs2_ioctl(struct file *filp, unsigned int cmd, unsigned long arg)
301{
302 switch(cmd) {
303 case FS_IOC_GETFLAGS:
304 return gfs2_get_flags(filp, (u32 __user *)arg);
305 case FS_IOC_SETFLAGS:
306 return gfs2_set_flags(filp, (u32 __user *)arg);
307 }
308 return -ENOTTY;
309}
310
311/**
312 * gfs2_allocate_page_backing - Use bmap to allocate blocks
313 * @page: The (locked) page to allocate backing for
314 *
315 * We try to allocate all the blocks required for the page in
316 * one go. This might fail for various reasons, so we keep
317 * trying until all the blocks to back this page are allocated.
318 * If some of the blocks are already allocated, thats ok too.
319 */
320
321static int gfs2_allocate_page_backing(struct page *page)
322{
323 struct inode *inode = page->mapping->host;
324 struct buffer_head bh;
325 unsigned long size = PAGE_CACHE_SIZE;
326 u64 lblock = page->index << (PAGE_CACHE_SHIFT - inode->i_blkbits);
327
328 do {
329 bh.b_state = 0;
330 bh.b_size = size;
331 gfs2_block_map(inode, lblock, &bh, 1);
332 if (!buffer_mapped(&bh))
333 return -EIO;
334 size -= bh.b_size;
335 lblock += (bh.b_size >> inode->i_blkbits);
336 } while(size > 0);
337 return 0;
338}
339
340/**
341 * gfs2_page_mkwrite - Make a shared, mmap()ed, page writable
342 * @vma: The virtual memory area
343 * @page: The page which is about to become writable
344 *
345 * When the page becomes writable, we need to ensure that we have
346 * blocks allocated on disk to back that page.
347 */
348
349static int gfs2_page_mkwrite(struct vm_area_struct *vma, struct vm_fault *vmf)
350{
351 struct page *page = vmf->page;
352 struct inode *inode = vma->vm_file->f_path.dentry->d_inode;
353 struct gfs2_inode *ip = GFS2_I(inode);
354 struct gfs2_sbd *sdp = GFS2_SB(inode);
355 unsigned long last_index;
356 u64 pos = page->index << PAGE_CACHE_SHIFT;
357 unsigned int data_blocks, ind_blocks, rblocks;
358 struct gfs2_holder gh;
359 struct gfs2_alloc *al;
360 int ret;
361
362 gfs2_holder_init(ip->i_gl, LM_ST_EXCLUSIVE, 0, &gh);
363 ret = gfs2_glock_nq(&gh);
364 if (ret)
365 goto out;
366
367 set_bit(GLF_DIRTY, &ip->i_gl->gl_flags);
368 set_bit(GIF_SW_PAGED, &ip->i_flags);
369
370 if (!gfs2_write_alloc_required(ip, pos, PAGE_CACHE_SIZE))
371 goto out_unlock;
372 ret = -ENOMEM;
373 al = gfs2_alloc_get(ip);
374 if (al == NULL)
375 goto out_unlock;
376
377 ret = gfs2_quota_lock_check(ip);
378 if (ret)
379 goto out_alloc_put;
380 gfs2_write_calc_reserv(ip, PAGE_CACHE_SIZE, &data_blocks, &ind_blocks);
381 al->al_requested = data_blocks + ind_blocks;
382 ret = gfs2_inplace_reserve(ip);
383 if (ret)
384 goto out_quota_unlock;
385
386 rblocks = RES_DINODE + ind_blocks;
387 if (gfs2_is_jdata(ip))
388 rblocks += data_blocks ? data_blocks : 1;
389 if (ind_blocks || data_blocks) {
390 rblocks += RES_STATFS + RES_QUOTA;
391 rblocks += gfs2_rg_blocks(al);
392 }
393 ret = gfs2_trans_begin(sdp, rblocks, 0);
394 if (ret)
395 goto out_trans_fail;
396
397 lock_page(page);
398 ret = -EINVAL;
399 last_index = ip->i_inode.i_size >> PAGE_CACHE_SHIFT;
400 if (page->index > last_index)
401 goto out_unlock_page;
402 ret = 0;
403 if (!PageUptodate(page) || page->mapping != ip->i_inode.i_mapping)
404 goto out_unlock_page;
405 if (gfs2_is_stuffed(ip)) {
406 ret = gfs2_unstuff_dinode(ip, page);
407 if (ret)
408 goto out_unlock_page;
409 }
410 ret = gfs2_allocate_page_backing(page);
411
412out_unlock_page:
413 unlock_page(page);
414 gfs2_trans_end(sdp);
415out_trans_fail:
416 gfs2_inplace_release(ip);
417out_quota_unlock:
418 gfs2_quota_unlock(ip);
419out_alloc_put:
420 gfs2_alloc_put(ip);
421out_unlock:
422 gfs2_glock_dq(&gh);
423out:
424 gfs2_holder_uninit(&gh);
425 if (ret == -ENOMEM)
426 ret = VM_FAULT_OOM;
427 else if (ret)
428 ret = VM_FAULT_SIGBUS;
429 return ret;
430}
431
432static const struct vm_operations_struct gfs2_vm_ops = {
433 .fault = filemap_fault,
434 .page_mkwrite = gfs2_page_mkwrite,
435};
436
437/**
438 * gfs2_mmap -
439 * @file: The file to map
440 * @vma: The VMA which described the mapping
441 *
442 * There is no need to get a lock here unless we should be updating
443 * atime. We ignore any locking errors since the only consequence is
444 * a missed atime update (which will just be deferred until later).
445 *
446 * Returns: 0
447 */
448
449static int gfs2_mmap(struct file *file, struct vm_area_struct *vma)
450{
451 struct gfs2_inode *ip = GFS2_I(file->f_mapping->host);
452
453 if (!(file->f_flags & O_NOATIME) &&
454 !IS_NOATIME(&ip->i_inode)) {
455 struct gfs2_holder i_gh;
456 int error;
457
458 gfs2_holder_init(ip->i_gl, LM_ST_SHARED, LM_FLAG_ANY, &i_gh);
459 error = gfs2_glock_nq(&i_gh);
460 if (error == 0) {
461 file_accessed(file);
462 gfs2_glock_dq(&i_gh);
463 }
464 gfs2_holder_uninit(&i_gh);
465 if (error)
466 return error;
467 }
468 vma->vm_ops = &gfs2_vm_ops;
469 vma->vm_flags |= VM_CAN_NONLINEAR;
470
471 return 0;
472}
473
474/**
475 * gfs2_open - open a file
476 * @inode: the inode to open
477 * @file: the struct file for this opening
478 *
479 * Returns: errno
480 */
481
482static int gfs2_open(struct inode *inode, struct file *file)
483{
484 struct gfs2_inode *ip = GFS2_I(inode);
485 struct gfs2_holder i_gh;
486 struct gfs2_file *fp;
487 int error;
488
489 fp = kzalloc(sizeof(struct gfs2_file), GFP_KERNEL);
490 if (!fp)
491 return -ENOMEM;
492
493 mutex_init(&fp->f_fl_mutex);
494
495 gfs2_assert_warn(GFS2_SB(inode), !file->private_data);
496 file->private_data = fp;
497
498 if (S_ISREG(ip->i_inode.i_mode)) {
499 error = gfs2_glock_nq_init(ip->i_gl, LM_ST_SHARED, LM_FLAG_ANY,
500 &i_gh);
501 if (error)
502 goto fail;
503
504 if (!(file->f_flags & O_LARGEFILE) &&
505 i_size_read(inode) > MAX_NON_LFS) {
506 error = -EOVERFLOW;
507 goto fail_gunlock;
508 }
509
510 gfs2_glock_dq_uninit(&i_gh);
511 }
512
513 return 0;
514
515fail_gunlock:
516 gfs2_glock_dq_uninit(&i_gh);
517fail:
518 file->private_data = NULL;
519 kfree(fp);
520 return error;
521}
522
523/**
524 * gfs2_close - called to close a struct file
525 * @inode: the inode the struct file belongs to
526 * @file: the struct file being closed
527 *
528 * Returns: errno
529 */
530
531static int gfs2_close(struct inode *inode, struct file *file)
532{
533 struct gfs2_sbd *sdp = inode->i_sb->s_fs_info;
534 struct gfs2_file *fp;
535
536 fp = file->private_data;
537 file->private_data = NULL;
538
539 if (gfs2_assert_warn(sdp, fp))
540 return -EIO;
541
542 kfree(fp);
543
544 return 0;
545}
546
547/**
548 * gfs2_fsync - sync the dirty data for a file (across the cluster)
549 * @file: the file that points to the dentry
550 * @start: the start position in the file to sync
551 * @end: the end position in the file to sync
552 * @datasync: set if we can ignore timestamp changes
553 *
554 * The VFS will flush data for us. We only need to worry
555 * about metadata here.
556 *
557 * Returns: errno
558 */
559
560static int gfs2_fsync(struct file *file, loff_t start, loff_t end,
561 int datasync)
562{
563 struct inode *inode = file->f_mapping->host;
564 int sync_state = inode->i_state & (I_DIRTY_SYNC|I_DIRTY_DATASYNC);
565 struct gfs2_inode *ip = GFS2_I(inode);
566 int ret;
567
568 ret = filemap_write_and_wait_range(inode->i_mapping, start, end);
569 if (ret)
570 return ret;
571 mutex_lock(&inode->i_mutex);
572
573 if (datasync)
574 sync_state &= ~I_DIRTY_SYNC;
575
576 if (sync_state) {
577 ret = sync_inode_metadata(inode, 1);
578 if (ret) {
579 mutex_unlock(&inode->i_mutex);
580 return ret;
581 }
582 gfs2_ail_flush(ip->i_gl);
583 }
584
585 mutex_unlock(&inode->i_mutex);
586 return 0;
587}
588
589/**
590 * gfs2_file_aio_write - Perform a write to a file
591 * @iocb: The io context
592 * @iov: The data to write
593 * @nr_segs: Number of @iov segments
594 * @pos: The file position
595 *
596 * We have to do a lock/unlock here to refresh the inode size for
597 * O_APPEND writes, otherwise we can land up writing at the wrong
598 * offset. There is still a race, but provided the app is using its
599 * own file locking, this will make O_APPEND work as expected.
600 *
601 */
602
603static ssize_t gfs2_file_aio_write(struct kiocb *iocb, const struct iovec *iov,
604 unsigned long nr_segs, loff_t pos)
605{
606 struct file *file = iocb->ki_filp;
607
608 if (file->f_flags & O_APPEND) {
609 struct dentry *dentry = file->f_dentry;
610 struct gfs2_inode *ip = GFS2_I(dentry->d_inode);
611 struct gfs2_holder gh;
612 int ret;
613
614 ret = gfs2_glock_nq_init(ip->i_gl, LM_ST_SHARED, 0, &gh);
615 if (ret)
616 return ret;
617 gfs2_glock_dq_uninit(&gh);
618 }
619
620 return generic_file_aio_write(iocb, iov, nr_segs, pos);
621}
622
623static int empty_write_end(struct page *page, unsigned from,
624 unsigned to, int mode)
625{
626 struct inode *inode = page->mapping->host;
627 struct gfs2_inode *ip = GFS2_I(inode);
628 struct buffer_head *bh;
629 unsigned offset, blksize = 1 << inode->i_blkbits;
630 pgoff_t end_index = i_size_read(inode) >> PAGE_CACHE_SHIFT;
631
632 zero_user(page, from, to-from);
633 mark_page_accessed(page);
634
635 if (page->index < end_index || !(mode & FALLOC_FL_KEEP_SIZE)) {
636 if (!gfs2_is_writeback(ip))
637 gfs2_page_add_databufs(ip, page, from, to);
638
639 block_commit_write(page, from, to);
640 return 0;
641 }
642
643 offset = 0;
644 bh = page_buffers(page);
645 while (offset < to) {
646 if (offset >= from) {
647 set_buffer_uptodate(bh);
648 mark_buffer_dirty(bh);
649 clear_buffer_new(bh);
650 write_dirty_buffer(bh, WRITE);
651 }
652 offset += blksize;
653 bh = bh->b_this_page;
654 }
655
656 offset = 0;
657 bh = page_buffers(page);
658 while (offset < to) {
659 if (offset >= from) {
660 wait_on_buffer(bh);
661 if (!buffer_uptodate(bh))
662 return -EIO;
663 }
664 offset += blksize;
665 bh = bh->b_this_page;
666 }
667 return 0;
668}
669
670static int needs_empty_write(sector_t block, struct inode *inode)
671{
672 int error;
673 struct buffer_head bh_map = { .b_state = 0, .b_blocknr = 0 };
674
675 bh_map.b_size = 1 << inode->i_blkbits;
676 error = gfs2_block_map(inode, block, &bh_map, 0);
677 if (unlikely(error))
678 return error;
679 return !buffer_mapped(&bh_map);
680}
681
682static int write_empty_blocks(struct page *page, unsigned from, unsigned to,
683 int mode)
684{
685 struct inode *inode = page->mapping->host;
686 unsigned start, end, next, blksize;
687 sector_t block = page->index << (PAGE_CACHE_SHIFT - inode->i_blkbits);
688 int ret;
689
690 blksize = 1 << inode->i_blkbits;
691 next = end = 0;
692 while (next < from) {
693 next += blksize;
694 block++;
695 }
696 start = next;
697 do {
698 next += blksize;
699 ret = needs_empty_write(block, inode);
700 if (unlikely(ret < 0))
701 return ret;
702 if (ret == 0) {
703 if (end) {
704 ret = __block_write_begin(page, start, end - start,
705 gfs2_block_map);
706 if (unlikely(ret))
707 return ret;
708 ret = empty_write_end(page, start, end, mode);
709 if (unlikely(ret))
710 return ret;
711 end = 0;
712 }
713 start = next;
714 }
715 else
716 end = next;
717 block++;
718 } while (next < to);
719
720 if (end) {
721 ret = __block_write_begin(page, start, end - start, gfs2_block_map);
722 if (unlikely(ret))
723 return ret;
724 ret = empty_write_end(page, start, end, mode);
725 if (unlikely(ret))
726 return ret;
727 }
728
729 return 0;
730}
731
732static int fallocate_chunk(struct inode *inode, loff_t offset, loff_t len,
733 int mode)
734{
735 struct gfs2_inode *ip = GFS2_I(inode);
736 struct buffer_head *dibh;
737 int error;
738 u64 start = offset >> PAGE_CACHE_SHIFT;
739 unsigned int start_offset = offset & ~PAGE_CACHE_MASK;
740 u64 end = (offset + len - 1) >> PAGE_CACHE_SHIFT;
741 pgoff_t curr;
742 struct page *page;
743 unsigned int end_offset = (offset + len) & ~PAGE_CACHE_MASK;
744 unsigned int from, to;
745
746 if (!end_offset)
747 end_offset = PAGE_CACHE_SIZE;
748
749 error = gfs2_meta_inode_buffer(ip, &dibh);
750 if (unlikely(error))
751 goto out;
752
753 gfs2_trans_add_bh(ip->i_gl, dibh, 1);
754
755 if (gfs2_is_stuffed(ip)) {
756 error = gfs2_unstuff_dinode(ip, NULL);
757 if (unlikely(error))
758 goto out;
759 }
760
761 curr = start;
762 offset = start << PAGE_CACHE_SHIFT;
763 from = start_offset;
764 to = PAGE_CACHE_SIZE;
765 while (curr <= end) {
766 page = grab_cache_page_write_begin(inode->i_mapping, curr,
767 AOP_FLAG_NOFS);
768 if (unlikely(!page)) {
769 error = -ENOMEM;
770 goto out;
771 }
772
773 if (curr == end)
774 to = end_offset;
775 error = write_empty_blocks(page, from, to, mode);
776 if (!error && offset + to > inode->i_size &&
777 !(mode & FALLOC_FL_KEEP_SIZE)) {
778 i_size_write(inode, offset + to);
779 }
780 unlock_page(page);
781 page_cache_release(page);
782 if (error)
783 goto out;
784 curr++;
785 offset += PAGE_CACHE_SIZE;
786 from = 0;
787 }
788
789 gfs2_dinode_out(ip, dibh->b_data);
790 mark_inode_dirty(inode);
791
792 brelse(dibh);
793
794out:
795 return error;
796}
797
798static void calc_max_reserv(struct gfs2_inode *ip, loff_t max, loff_t *len,
799 unsigned int *data_blocks, unsigned int *ind_blocks)
800{
801 const struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode);
802 unsigned int max_blocks = ip->i_alloc->al_rgd->rd_free_clone;
803 unsigned int tmp, max_data = max_blocks - 3 * (sdp->sd_max_height - 1);
804
805 for (tmp = max_data; tmp > sdp->sd_diptrs;) {
806 tmp = DIV_ROUND_UP(tmp, sdp->sd_inptrs);
807 max_data -= tmp;
808 }
809 /* This calculation isn't the exact reverse of gfs2_write_calc_reserve,
810 so it might end up with fewer data blocks */
811 if (max_data <= *data_blocks)
812 return;
813 *data_blocks = max_data;
814 *ind_blocks = max_blocks - max_data;
815 *len = ((loff_t)max_data - 3) << sdp->sd_sb.sb_bsize_shift;
816 if (*len > max) {
817 *len = max;
818 gfs2_write_calc_reserv(ip, max, data_blocks, ind_blocks);
819 }
820}
821
822static long gfs2_fallocate(struct file *file, int mode, loff_t offset,
823 loff_t len)
824{
825 struct inode *inode = file->f_path.dentry->d_inode;
826 struct gfs2_sbd *sdp = GFS2_SB(inode);
827 struct gfs2_inode *ip = GFS2_I(inode);
828 unsigned int data_blocks = 0, ind_blocks = 0, rblocks;
829 loff_t bytes, max_bytes;
830 struct gfs2_alloc *al;
831 int error;
832 loff_t bsize_mask = ~((loff_t)sdp->sd_sb.sb_bsize - 1);
833 loff_t next = (offset + len - 1) >> sdp->sd_sb.sb_bsize_shift;
834 next = (next + 1) << sdp->sd_sb.sb_bsize_shift;
835
836 /* We only support the FALLOC_FL_KEEP_SIZE mode */
837 if (mode & ~FALLOC_FL_KEEP_SIZE)
838 return -EOPNOTSUPP;
839
840 offset &= bsize_mask;
841
842 len = next - offset;
843 bytes = sdp->sd_max_rg_data * sdp->sd_sb.sb_bsize / 2;
844 if (!bytes)
845 bytes = UINT_MAX;
846 bytes &= bsize_mask;
847 if (bytes == 0)
848 bytes = sdp->sd_sb.sb_bsize;
849
850 gfs2_holder_init(ip->i_gl, LM_ST_EXCLUSIVE, 0, &ip->i_gh);
851 error = gfs2_glock_nq(&ip->i_gh);
852 if (unlikely(error))
853 goto out_uninit;
854
855 if (!gfs2_write_alloc_required(ip, offset, len))
856 goto out_unlock;
857
858 while (len > 0) {
859 if (len < bytes)
860 bytes = len;
861 al = gfs2_alloc_get(ip);
862 if (!al) {
863 error = -ENOMEM;
864 goto out_unlock;
865 }
866
867 error = gfs2_quota_lock_check(ip);
868 if (error)
869 goto out_alloc_put;
870
871retry:
872 gfs2_write_calc_reserv(ip, bytes, &data_blocks, &ind_blocks);
873
874 al->al_requested = data_blocks + ind_blocks;
875 error = gfs2_inplace_reserve(ip);
876 if (error) {
877 if (error == -ENOSPC && bytes > sdp->sd_sb.sb_bsize) {
878 bytes >>= 1;
879 bytes &= bsize_mask;
880 if (bytes == 0)
881 bytes = sdp->sd_sb.sb_bsize;
882 goto retry;
883 }
884 goto out_qunlock;
885 }
886 max_bytes = bytes;
887 calc_max_reserv(ip, len, &max_bytes, &data_blocks, &ind_blocks);
888 al->al_requested = data_blocks + ind_blocks;
889
890 rblocks = RES_DINODE + ind_blocks + RES_STATFS + RES_QUOTA +
891 RES_RG_HDR + gfs2_rg_blocks(al);
892 if (gfs2_is_jdata(ip))
893 rblocks += data_blocks ? data_blocks : 1;
894
895 error = gfs2_trans_begin(sdp, rblocks,
896 PAGE_CACHE_SIZE/sdp->sd_sb.sb_bsize);
897 if (error)
898 goto out_trans_fail;
899
900 error = fallocate_chunk(inode, offset, max_bytes, mode);
901 gfs2_trans_end(sdp);
902
903 if (error)
904 goto out_trans_fail;
905
906 len -= max_bytes;
907 offset += max_bytes;
908 gfs2_inplace_release(ip);
909 gfs2_quota_unlock(ip);
910 gfs2_alloc_put(ip);
911 }
912 goto out_unlock;
913
914out_trans_fail:
915 gfs2_inplace_release(ip);
916out_qunlock:
917 gfs2_quota_unlock(ip);
918out_alloc_put:
919 gfs2_alloc_put(ip);
920out_unlock:
921 gfs2_glock_dq(&ip->i_gh);
922out_uninit:
923 gfs2_holder_uninit(&ip->i_gh);
924 return error;
925}
926
927#ifdef CONFIG_GFS2_FS_LOCKING_DLM
928
929/**
930 * gfs2_setlease - acquire/release a file lease
931 * @file: the file pointer
932 * @arg: lease type
933 * @fl: file lock
934 *
935 * We don't currently have a way to enforce a lease across the whole
936 * cluster; until we do, disable leases (by just returning -EINVAL),
937 * unless the administrator has requested purely local locking.
938 *
939 * Locking: called under lock_flocks
940 *
941 * Returns: errno
942 */
943
944static int gfs2_setlease(struct file *file, long arg, struct file_lock **fl)
945{
946 return -EINVAL;
947}
948
949/**
950 * gfs2_lock - acquire/release a posix lock on a file
951 * @file: the file pointer
952 * @cmd: either modify or retrieve lock state, possibly wait
953 * @fl: type and range of lock
954 *
955 * Returns: errno
956 */
957
958static int gfs2_lock(struct file *file, int cmd, struct file_lock *fl)
959{
960 struct gfs2_inode *ip = GFS2_I(file->f_mapping->host);
961 struct gfs2_sbd *sdp = GFS2_SB(file->f_mapping->host);
962 struct lm_lockstruct *ls = &sdp->sd_lockstruct;
963
964 if (!(fl->fl_flags & FL_POSIX))
965 return -ENOLCK;
966 if (__mandatory_lock(&ip->i_inode) && fl->fl_type != F_UNLCK)
967 return -ENOLCK;
968
969 if (cmd == F_CANCELLK) {
970 /* Hack: */
971 cmd = F_SETLK;
972 fl->fl_type = F_UNLCK;
973 }
974 if (unlikely(test_bit(SDF_SHUTDOWN, &sdp->sd_flags)))
975 return -EIO;
976 if (IS_GETLK(cmd))
977 return dlm_posix_get(ls->ls_dlm, ip->i_no_addr, file, fl);
978 else if (fl->fl_type == F_UNLCK)
979 return dlm_posix_unlock(ls->ls_dlm, ip->i_no_addr, file, fl);
980 else
981 return dlm_posix_lock(ls->ls_dlm, ip->i_no_addr, file, cmd, fl);
982}
983
984static int do_flock(struct file *file, int cmd, struct file_lock *fl)
985{
986 struct gfs2_file *fp = file->private_data;
987 struct gfs2_holder *fl_gh = &fp->f_fl_gh;
988 struct gfs2_inode *ip = GFS2_I(file->f_path.dentry->d_inode);
989 struct gfs2_glock *gl;
990 unsigned int state;
991 int flags;
992 int error = 0;
993
994 state = (fl->fl_type == F_WRLCK) ? LM_ST_EXCLUSIVE : LM_ST_SHARED;
995 flags = (IS_SETLKW(cmd) ? 0 : LM_FLAG_TRY) | GL_EXACT | GL_NOCACHE;
996
997 mutex_lock(&fp->f_fl_mutex);
998
999 gl = fl_gh->gh_gl;
1000 if (gl) {
1001 if (fl_gh->gh_state == state)
1002 goto out;
1003 flock_lock_file_wait(file,
1004 &(struct file_lock){.fl_type = F_UNLCK});
1005 gfs2_glock_dq_wait(fl_gh);
1006 gfs2_holder_reinit(state, flags, fl_gh);
1007 } else {
1008 error = gfs2_glock_get(GFS2_SB(&ip->i_inode), ip->i_no_addr,
1009 &gfs2_flock_glops, CREATE, &gl);
1010 if (error)
1011 goto out;
1012 gfs2_holder_init(gl, state, flags, fl_gh);
1013 gfs2_glock_put(gl);
1014 }
1015 error = gfs2_glock_nq(fl_gh);
1016 if (error) {
1017 gfs2_holder_uninit(fl_gh);
1018 if (error == GLR_TRYFAILED)
1019 error = -EAGAIN;
1020 } else {
1021 error = flock_lock_file_wait(file, fl);
1022 gfs2_assert_warn(GFS2_SB(&ip->i_inode), !error);
1023 }
1024
1025out:
1026 mutex_unlock(&fp->f_fl_mutex);
1027 return error;
1028}
1029
1030static void do_unflock(struct file *file, struct file_lock *fl)
1031{
1032 struct gfs2_file *fp = file->private_data;
1033 struct gfs2_holder *fl_gh = &fp->f_fl_gh;
1034
1035 mutex_lock(&fp->f_fl_mutex);
1036 flock_lock_file_wait(file, fl);
1037 if (fl_gh->gh_gl) {
1038 gfs2_glock_dq_wait(fl_gh);
1039 gfs2_holder_uninit(fl_gh);
1040 }
1041 mutex_unlock(&fp->f_fl_mutex);
1042}
1043
1044/**
1045 * gfs2_flock - acquire/release a flock lock on a file
1046 * @file: the file pointer
1047 * @cmd: either modify or retrieve lock state, possibly wait
1048 * @fl: type and range of lock
1049 *
1050 * Returns: errno
1051 */
1052
1053static int gfs2_flock(struct file *file, int cmd, struct file_lock *fl)
1054{
1055 if (!(fl->fl_flags & FL_FLOCK))
1056 return -ENOLCK;
1057 if (fl->fl_type & LOCK_MAND)
1058 return -EOPNOTSUPP;
1059
1060 if (fl->fl_type == F_UNLCK) {
1061 do_unflock(file, fl);
1062 return 0;
1063 } else {
1064 return do_flock(file, cmd, fl);
1065 }
1066}
1067
1068const struct file_operations gfs2_file_fops = {
1069 .llseek = gfs2_llseek,
1070 .read = do_sync_read,
1071 .aio_read = generic_file_aio_read,
1072 .write = do_sync_write,
1073 .aio_write = gfs2_file_aio_write,
1074 .unlocked_ioctl = gfs2_ioctl,
1075 .mmap = gfs2_mmap,
1076 .open = gfs2_open,
1077 .release = gfs2_close,
1078 .fsync = gfs2_fsync,
1079 .lock = gfs2_lock,
1080 .flock = gfs2_flock,
1081 .splice_read = generic_file_splice_read,
1082 .splice_write = generic_file_splice_write,
1083 .setlease = gfs2_setlease,
1084 .fallocate = gfs2_fallocate,
1085};
1086
1087const struct file_operations gfs2_dir_fops = {
1088 .readdir = gfs2_readdir,
1089 .unlocked_ioctl = gfs2_ioctl,
1090 .open = gfs2_open,
1091 .release = gfs2_close,
1092 .fsync = gfs2_fsync,
1093 .lock = gfs2_lock,
1094 .flock = gfs2_flock,
1095 .llseek = default_llseek,
1096};
1097
1098#endif /* CONFIG_GFS2_FS_LOCKING_DLM */
1099
1100const struct file_operations gfs2_file_fops_nolock = {
1101 .llseek = gfs2_llseek,
1102 .read = do_sync_read,
1103 .aio_read = generic_file_aio_read,
1104 .write = do_sync_write,
1105 .aio_write = gfs2_file_aio_write,
1106 .unlocked_ioctl = gfs2_ioctl,
1107 .mmap = gfs2_mmap,
1108 .open = gfs2_open,
1109 .release = gfs2_close,
1110 .fsync = gfs2_fsync,
1111 .splice_read = generic_file_splice_read,
1112 .splice_write = generic_file_splice_write,
1113 .setlease = generic_setlease,
1114 .fallocate = gfs2_fallocate,
1115};
1116
1117const struct file_operations gfs2_dir_fops_nolock = {
1118 .readdir = gfs2_readdir,
1119 .unlocked_ioctl = gfs2_ioctl,
1120 .open = gfs2_open,
1121 .release = gfs2_close,
1122 .fsync = gfs2_fsync,
1123 .llseek = default_llseek,
1124};
1125
1// SPDX-License-Identifier: GPL-2.0-only
2/*
3 * Copyright (C) Sistina Software, Inc. 1997-2003 All rights reserved.
4 * Copyright (C) 2004-2006 Red Hat, Inc. All rights reserved.
5 */
6
7#include <linux/slab.h>
8#include <linux/spinlock.h>
9#include <linux/compat.h>
10#include <linux/completion.h>
11#include <linux/buffer_head.h>
12#include <linux/pagemap.h>
13#include <linux/uio.h>
14#include <linux/blkdev.h>
15#include <linux/mm.h>
16#include <linux/mount.h>
17#include <linux/fs.h>
18#include <linux/gfs2_ondisk.h>
19#include <linux/falloc.h>
20#include <linux/swap.h>
21#include <linux/crc32.h>
22#include <linux/writeback.h>
23#include <linux/uaccess.h>
24#include <linux/dlm.h>
25#include <linux/dlm_plock.h>
26#include <linux/delay.h>
27#include <linux/backing-dev.h>
28
29#include "gfs2.h"
30#include "incore.h"
31#include "bmap.h"
32#include "aops.h"
33#include "dir.h"
34#include "glock.h"
35#include "glops.h"
36#include "inode.h"
37#include "log.h"
38#include "meta_io.h"
39#include "quota.h"
40#include "rgrp.h"
41#include "trans.h"
42#include "util.h"
43
44/**
45 * gfs2_llseek - seek to a location in a file
46 * @file: the file
47 * @offset: the offset
48 * @whence: Where to seek from (SEEK_SET, SEEK_CUR, or SEEK_END)
49 *
50 * SEEK_END requires the glock for the file because it references the
51 * file's size.
52 *
53 * Returns: The new offset, or errno
54 */
55
56static loff_t gfs2_llseek(struct file *file, loff_t offset, int whence)
57{
58 struct gfs2_inode *ip = GFS2_I(file->f_mapping->host);
59 struct gfs2_holder i_gh;
60 loff_t error;
61
62 switch (whence) {
63 case SEEK_END:
64 error = gfs2_glock_nq_init(ip->i_gl, LM_ST_SHARED, LM_FLAG_ANY,
65 &i_gh);
66 if (!error) {
67 error = generic_file_llseek(file, offset, whence);
68 gfs2_glock_dq_uninit(&i_gh);
69 }
70 break;
71
72 case SEEK_DATA:
73 error = gfs2_seek_data(file, offset);
74 break;
75
76 case SEEK_HOLE:
77 error = gfs2_seek_hole(file, offset);
78 break;
79
80 case SEEK_CUR:
81 case SEEK_SET:
82 /*
83 * These don't reference inode->i_size and don't depend on the
84 * block mapping, so we don't need the glock.
85 */
86 error = generic_file_llseek(file, offset, whence);
87 break;
88 default:
89 error = -EINVAL;
90 }
91
92 return error;
93}
94
95/**
96 * gfs2_readdir - Iterator for a directory
97 * @file: The directory to read from
98 * @ctx: What to feed directory entries to
99 *
100 * Returns: errno
101 */
102
103static int gfs2_readdir(struct file *file, struct dir_context *ctx)
104{
105 struct inode *dir = file->f_mapping->host;
106 struct gfs2_inode *dip = GFS2_I(dir);
107 struct gfs2_holder d_gh;
108 int error;
109
110 error = gfs2_glock_nq_init(dip->i_gl, LM_ST_SHARED, 0, &d_gh);
111 if (error)
112 return error;
113
114 error = gfs2_dir_read(dir, ctx, &file->f_ra);
115
116 gfs2_glock_dq_uninit(&d_gh);
117
118 return error;
119}
120
121/**
122 * fsflag_gfs2flag
123 *
124 * The FS_JOURNAL_DATA_FL flag maps to GFS2_DIF_INHERIT_JDATA for directories,
125 * and to GFS2_DIF_JDATA for non-directories.
126 */
127static struct {
128 u32 fsflag;
129 u32 gfsflag;
130} fsflag_gfs2flag[] = {
131 {FS_SYNC_FL, GFS2_DIF_SYNC},
132 {FS_IMMUTABLE_FL, GFS2_DIF_IMMUTABLE},
133 {FS_APPEND_FL, GFS2_DIF_APPENDONLY},
134 {FS_NOATIME_FL, GFS2_DIF_NOATIME},
135 {FS_INDEX_FL, GFS2_DIF_EXHASH},
136 {FS_TOPDIR_FL, GFS2_DIF_TOPDIR},
137 {FS_JOURNAL_DATA_FL, GFS2_DIF_JDATA | GFS2_DIF_INHERIT_JDATA},
138};
139
140static inline u32 gfs2_gfsflags_to_fsflags(struct inode *inode, u32 gfsflags)
141{
142 int i;
143 u32 fsflags = 0;
144
145 if (S_ISDIR(inode->i_mode))
146 gfsflags &= ~GFS2_DIF_JDATA;
147 else
148 gfsflags &= ~GFS2_DIF_INHERIT_JDATA;
149
150 for (i = 0; i < ARRAY_SIZE(fsflag_gfs2flag); i++)
151 if (gfsflags & fsflag_gfs2flag[i].gfsflag)
152 fsflags |= fsflag_gfs2flag[i].fsflag;
153 return fsflags;
154}
155
156static int gfs2_get_flags(struct file *filp, u32 __user *ptr)
157{
158 struct inode *inode = file_inode(filp);
159 struct gfs2_inode *ip = GFS2_I(inode);
160 struct gfs2_holder gh;
161 int error;
162 u32 fsflags;
163
164 gfs2_holder_init(ip->i_gl, LM_ST_SHARED, 0, &gh);
165 error = gfs2_glock_nq(&gh);
166 if (error)
167 goto out_uninit;
168
169 fsflags = gfs2_gfsflags_to_fsflags(inode, ip->i_diskflags);
170
171 if (put_user(fsflags, ptr))
172 error = -EFAULT;
173
174 gfs2_glock_dq(&gh);
175out_uninit:
176 gfs2_holder_uninit(&gh);
177 return error;
178}
179
180void gfs2_set_inode_flags(struct inode *inode)
181{
182 struct gfs2_inode *ip = GFS2_I(inode);
183 unsigned int flags = inode->i_flags;
184
185 flags &= ~(S_SYNC|S_APPEND|S_IMMUTABLE|S_NOATIME|S_DIRSYNC|S_NOSEC);
186 if ((ip->i_eattr == 0) && !is_sxid(inode->i_mode))
187 flags |= S_NOSEC;
188 if (ip->i_diskflags & GFS2_DIF_IMMUTABLE)
189 flags |= S_IMMUTABLE;
190 if (ip->i_diskflags & GFS2_DIF_APPENDONLY)
191 flags |= S_APPEND;
192 if (ip->i_diskflags & GFS2_DIF_NOATIME)
193 flags |= S_NOATIME;
194 if (ip->i_diskflags & GFS2_DIF_SYNC)
195 flags |= S_SYNC;
196 inode->i_flags = flags;
197}
198
199/* Flags that can be set by user space */
200#define GFS2_FLAGS_USER_SET (GFS2_DIF_JDATA| \
201 GFS2_DIF_IMMUTABLE| \
202 GFS2_DIF_APPENDONLY| \
203 GFS2_DIF_NOATIME| \
204 GFS2_DIF_SYNC| \
205 GFS2_DIF_TOPDIR| \
206 GFS2_DIF_INHERIT_JDATA)
207
208/**
209 * do_gfs2_set_flags - set flags on an inode
210 * @filp: file pointer
211 * @reqflags: The flags to set
212 * @mask: Indicates which flags are valid
213 * @fsflags: The FS_* inode flags passed in
214 *
215 */
216static int do_gfs2_set_flags(struct file *filp, u32 reqflags, u32 mask,
217 const u32 fsflags)
218{
219 struct inode *inode = file_inode(filp);
220 struct gfs2_inode *ip = GFS2_I(inode);
221 struct gfs2_sbd *sdp = GFS2_SB(inode);
222 struct buffer_head *bh;
223 struct gfs2_holder gh;
224 int error;
225 u32 new_flags, flags, oldflags;
226
227 error = mnt_want_write_file(filp);
228 if (error)
229 return error;
230
231 error = gfs2_glock_nq_init(ip->i_gl, LM_ST_EXCLUSIVE, 0, &gh);
232 if (error)
233 goto out_drop_write;
234
235 oldflags = gfs2_gfsflags_to_fsflags(inode, ip->i_diskflags);
236 error = vfs_ioc_setflags_prepare(inode, oldflags, fsflags);
237 if (error)
238 goto out;
239
240 error = -EACCES;
241 if (!inode_owner_or_capable(inode))
242 goto out;
243
244 error = 0;
245 flags = ip->i_diskflags;
246 new_flags = (flags & ~mask) | (reqflags & mask);
247 if ((new_flags ^ flags) == 0)
248 goto out;
249
250 error = -EPERM;
251 if (IS_IMMUTABLE(inode) && (new_flags & GFS2_DIF_IMMUTABLE))
252 goto out;
253 if (IS_APPEND(inode) && (new_flags & GFS2_DIF_APPENDONLY))
254 goto out;
255 if (((new_flags ^ flags) & GFS2_DIF_IMMUTABLE) &&
256 !capable(CAP_LINUX_IMMUTABLE))
257 goto out;
258 if (!IS_IMMUTABLE(inode)) {
259 error = gfs2_permission(inode, MAY_WRITE);
260 if (error)
261 goto out;
262 }
263 if ((flags ^ new_flags) & GFS2_DIF_JDATA) {
264 if (new_flags & GFS2_DIF_JDATA)
265 gfs2_log_flush(sdp, ip->i_gl,
266 GFS2_LOG_HEAD_FLUSH_NORMAL |
267 GFS2_LFC_SET_FLAGS);
268 error = filemap_fdatawrite(inode->i_mapping);
269 if (error)
270 goto out;
271 error = filemap_fdatawait(inode->i_mapping);
272 if (error)
273 goto out;
274 if (new_flags & GFS2_DIF_JDATA)
275 gfs2_ordered_del_inode(ip);
276 }
277 error = gfs2_trans_begin(sdp, RES_DINODE, 0);
278 if (error)
279 goto out;
280 error = gfs2_meta_inode_buffer(ip, &bh);
281 if (error)
282 goto out_trans_end;
283 inode->i_ctime = current_time(inode);
284 gfs2_trans_add_meta(ip->i_gl, bh);
285 ip->i_diskflags = new_flags;
286 gfs2_dinode_out(ip, bh->b_data);
287 brelse(bh);
288 gfs2_set_inode_flags(inode);
289 gfs2_set_aops(inode);
290out_trans_end:
291 gfs2_trans_end(sdp);
292out:
293 gfs2_glock_dq_uninit(&gh);
294out_drop_write:
295 mnt_drop_write_file(filp);
296 return error;
297}
298
299static int gfs2_set_flags(struct file *filp, u32 __user *ptr)
300{
301 struct inode *inode = file_inode(filp);
302 u32 fsflags, gfsflags = 0;
303 u32 mask;
304 int i;
305
306 if (get_user(fsflags, ptr))
307 return -EFAULT;
308
309 for (i = 0; i < ARRAY_SIZE(fsflag_gfs2flag); i++) {
310 if (fsflags & fsflag_gfs2flag[i].fsflag) {
311 fsflags &= ~fsflag_gfs2flag[i].fsflag;
312 gfsflags |= fsflag_gfs2flag[i].gfsflag;
313 }
314 }
315 if (fsflags || gfsflags & ~GFS2_FLAGS_USER_SET)
316 return -EINVAL;
317
318 mask = GFS2_FLAGS_USER_SET;
319 if (S_ISDIR(inode->i_mode)) {
320 mask &= ~GFS2_DIF_JDATA;
321 } else {
322 /* The GFS2_DIF_TOPDIR flag is only valid for directories. */
323 if (gfsflags & GFS2_DIF_TOPDIR)
324 return -EINVAL;
325 mask &= ~(GFS2_DIF_TOPDIR | GFS2_DIF_INHERIT_JDATA);
326 }
327
328 return do_gfs2_set_flags(filp, gfsflags, mask, fsflags);
329}
330
331static int gfs2_getlabel(struct file *filp, char __user *label)
332{
333 struct inode *inode = file_inode(filp);
334 struct gfs2_sbd *sdp = GFS2_SB(inode);
335
336 if (copy_to_user(label, sdp->sd_sb.sb_locktable, GFS2_LOCKNAME_LEN))
337 return -EFAULT;
338
339 return 0;
340}
341
342static long gfs2_ioctl(struct file *filp, unsigned int cmd, unsigned long arg)
343{
344 switch(cmd) {
345 case FS_IOC_GETFLAGS:
346 return gfs2_get_flags(filp, (u32 __user *)arg);
347 case FS_IOC_SETFLAGS:
348 return gfs2_set_flags(filp, (u32 __user *)arg);
349 case FITRIM:
350 return gfs2_fitrim(filp, (void __user *)arg);
351 case FS_IOC_GETFSLABEL:
352 return gfs2_getlabel(filp, (char __user *)arg);
353 }
354
355 return -ENOTTY;
356}
357
358#ifdef CONFIG_COMPAT
359static long gfs2_compat_ioctl(struct file *filp, unsigned int cmd, unsigned long arg)
360{
361 switch(cmd) {
362 /* These are just misnamed, they actually get/put from/to user an int */
363 case FS_IOC32_GETFLAGS:
364 cmd = FS_IOC_GETFLAGS;
365 break;
366 case FS_IOC32_SETFLAGS:
367 cmd = FS_IOC_SETFLAGS;
368 break;
369 /* Keep this list in sync with gfs2_ioctl */
370 case FITRIM:
371 case FS_IOC_GETFSLABEL:
372 break;
373 default:
374 return -ENOIOCTLCMD;
375 }
376
377 return gfs2_ioctl(filp, cmd, (unsigned long)compat_ptr(arg));
378}
379#else
380#define gfs2_compat_ioctl NULL
381#endif
382
383/**
384 * gfs2_size_hint - Give a hint to the size of a write request
385 * @filep: The struct file
386 * @offset: The file offset of the write
387 * @size: The length of the write
388 *
389 * When we are about to do a write, this function records the total
390 * write size in order to provide a suitable hint to the lower layers
391 * about how many blocks will be required.
392 *
393 */
394
395static void gfs2_size_hint(struct file *filep, loff_t offset, size_t size)
396{
397 struct inode *inode = file_inode(filep);
398 struct gfs2_sbd *sdp = GFS2_SB(inode);
399 struct gfs2_inode *ip = GFS2_I(inode);
400 size_t blks = (size + sdp->sd_sb.sb_bsize - 1) >> sdp->sd_sb.sb_bsize_shift;
401 int hint = min_t(size_t, INT_MAX, blks);
402
403 if (hint > atomic_read(&ip->i_sizehint))
404 atomic_set(&ip->i_sizehint, hint);
405}
406
407/**
408 * gfs2_allocate_page_backing - Allocate blocks for a write fault
409 * @page: The (locked) page to allocate backing for
410 * @length: Size of the allocation
411 *
412 * We try to allocate all the blocks required for the page in one go. This
413 * might fail for various reasons, so we keep trying until all the blocks to
414 * back this page are allocated. If some of the blocks are already allocated,
415 * that is ok too.
416 */
417static int gfs2_allocate_page_backing(struct page *page, unsigned int length)
418{
419 u64 pos = page_offset(page);
420
421 do {
422 struct iomap iomap = { };
423
424 if (gfs2_iomap_get_alloc(page->mapping->host, pos, length, &iomap))
425 return -EIO;
426
427 if (length < iomap.length)
428 iomap.length = length;
429 length -= iomap.length;
430 pos += iomap.length;
431 } while (length > 0);
432
433 return 0;
434}
435
436/**
437 * gfs2_page_mkwrite - Make a shared, mmap()ed, page writable
438 * @vma: The virtual memory area
439 * @vmf: The virtual memory fault containing the page to become writable
440 *
441 * When the page becomes writable, we need to ensure that we have
442 * blocks allocated on disk to back that page.
443 */
444
445static vm_fault_t gfs2_page_mkwrite(struct vm_fault *vmf)
446{
447 struct page *page = vmf->page;
448 struct inode *inode = file_inode(vmf->vma->vm_file);
449 struct gfs2_inode *ip = GFS2_I(inode);
450 struct gfs2_sbd *sdp = GFS2_SB(inode);
451 struct gfs2_alloc_parms ap = { .aflags = 0, };
452 u64 offset = page_offset(page);
453 unsigned int data_blocks, ind_blocks, rblocks;
454 struct gfs2_holder gh;
455 unsigned int length;
456 loff_t size;
457 int ret;
458
459 sb_start_pagefault(inode->i_sb);
460
461 gfs2_holder_init(ip->i_gl, LM_ST_EXCLUSIVE, 0, &gh);
462 ret = gfs2_glock_nq(&gh);
463 if (ret)
464 goto out_uninit;
465
466 /* Check page index against inode size */
467 size = i_size_read(inode);
468 if (offset >= size) {
469 ret = -EINVAL;
470 goto out_unlock;
471 }
472
473 /* Update file times before taking page lock */
474 file_update_time(vmf->vma->vm_file);
475
476 /* page is wholly or partially inside EOF */
477 if (offset > size - PAGE_SIZE)
478 length = offset_in_page(size);
479 else
480 length = PAGE_SIZE;
481
482 gfs2_size_hint(vmf->vma->vm_file, offset, length);
483
484 set_bit(GLF_DIRTY, &ip->i_gl->gl_flags);
485 set_bit(GIF_SW_PAGED, &ip->i_flags);
486
487 /*
488 * iomap_writepage / iomap_writepages currently don't support inline
489 * files, so always unstuff here.
490 */
491
492 if (!gfs2_is_stuffed(ip) &&
493 !gfs2_write_alloc_required(ip, offset, length)) {
494 lock_page(page);
495 if (!PageUptodate(page) || page->mapping != inode->i_mapping) {
496 ret = -EAGAIN;
497 unlock_page(page);
498 }
499 goto out_unlock;
500 }
501
502 ret = gfs2_rindex_update(sdp);
503 if (ret)
504 goto out_unlock;
505
506 gfs2_write_calc_reserv(ip, length, &data_blocks, &ind_blocks);
507 ap.target = data_blocks + ind_blocks;
508 ret = gfs2_quota_lock_check(ip, &ap);
509 if (ret)
510 goto out_unlock;
511 ret = gfs2_inplace_reserve(ip, &ap);
512 if (ret)
513 goto out_quota_unlock;
514
515 rblocks = RES_DINODE + ind_blocks;
516 if (gfs2_is_jdata(ip))
517 rblocks += data_blocks ? data_blocks : 1;
518 if (ind_blocks || data_blocks) {
519 rblocks += RES_STATFS + RES_QUOTA;
520 rblocks += gfs2_rg_blocks(ip, data_blocks + ind_blocks);
521 }
522 ret = gfs2_trans_begin(sdp, rblocks, 0);
523 if (ret)
524 goto out_trans_fail;
525
526 lock_page(page);
527 ret = -EAGAIN;
528 /* If truncated, we must retry the operation, we may have raced
529 * with the glock demotion code.
530 */
531 if (!PageUptodate(page) || page->mapping != inode->i_mapping)
532 goto out_trans_end;
533
534 /* Unstuff, if required, and allocate backing blocks for page */
535 ret = 0;
536 if (gfs2_is_stuffed(ip))
537 ret = gfs2_unstuff_dinode(ip, page);
538 if (ret == 0)
539 ret = gfs2_allocate_page_backing(page, length);
540
541out_trans_end:
542 if (ret)
543 unlock_page(page);
544 gfs2_trans_end(sdp);
545out_trans_fail:
546 gfs2_inplace_release(ip);
547out_quota_unlock:
548 gfs2_quota_unlock(ip);
549out_unlock:
550 gfs2_glock_dq(&gh);
551out_uninit:
552 gfs2_holder_uninit(&gh);
553 if (ret == 0) {
554 set_page_dirty(page);
555 wait_for_stable_page(page);
556 }
557 sb_end_pagefault(inode->i_sb);
558 return block_page_mkwrite_return(ret);
559}
560
561static vm_fault_t gfs2_fault(struct vm_fault *vmf)
562{
563 struct inode *inode = file_inode(vmf->vma->vm_file);
564 struct gfs2_inode *ip = GFS2_I(inode);
565 struct gfs2_holder gh;
566 vm_fault_t ret;
567 int err;
568
569 gfs2_holder_init(ip->i_gl, LM_ST_SHARED, 0, &gh);
570 err = gfs2_glock_nq(&gh);
571 if (err) {
572 ret = block_page_mkwrite_return(err);
573 goto out_uninit;
574 }
575 ret = filemap_fault(vmf);
576 gfs2_glock_dq(&gh);
577out_uninit:
578 gfs2_holder_uninit(&gh);
579 return ret;
580}
581
582static const struct vm_operations_struct gfs2_vm_ops = {
583 .fault = gfs2_fault,
584 .map_pages = filemap_map_pages,
585 .page_mkwrite = gfs2_page_mkwrite,
586};
587
588/**
589 * gfs2_mmap -
590 * @file: The file to map
591 * @vma: The VMA which described the mapping
592 *
593 * There is no need to get a lock here unless we should be updating
594 * atime. We ignore any locking errors since the only consequence is
595 * a missed atime update (which will just be deferred until later).
596 *
597 * Returns: 0
598 */
599
600static int gfs2_mmap(struct file *file, struct vm_area_struct *vma)
601{
602 struct gfs2_inode *ip = GFS2_I(file->f_mapping->host);
603
604 if (!(file->f_flags & O_NOATIME) &&
605 !IS_NOATIME(&ip->i_inode)) {
606 struct gfs2_holder i_gh;
607 int error;
608
609 error = gfs2_glock_nq_init(ip->i_gl, LM_ST_SHARED, LM_FLAG_ANY,
610 &i_gh);
611 if (error)
612 return error;
613 /* grab lock to update inode */
614 gfs2_glock_dq_uninit(&i_gh);
615 file_accessed(file);
616 }
617 vma->vm_ops = &gfs2_vm_ops;
618
619 return 0;
620}
621
622/**
623 * gfs2_open_common - This is common to open and atomic_open
624 * @inode: The inode being opened
625 * @file: The file being opened
626 *
627 * This maybe called under a glock or not depending upon how it has
628 * been called. We must always be called under a glock for regular
629 * files, however. For other file types, it does not matter whether
630 * we hold the glock or not.
631 *
632 * Returns: Error code or 0 for success
633 */
634
635int gfs2_open_common(struct inode *inode, struct file *file)
636{
637 struct gfs2_file *fp;
638 int ret;
639
640 if (S_ISREG(inode->i_mode)) {
641 ret = generic_file_open(inode, file);
642 if (ret)
643 return ret;
644 }
645
646 fp = kzalloc(sizeof(struct gfs2_file), GFP_NOFS);
647 if (!fp)
648 return -ENOMEM;
649
650 mutex_init(&fp->f_fl_mutex);
651
652 gfs2_assert_warn(GFS2_SB(inode), !file->private_data);
653 file->private_data = fp;
654 if (file->f_mode & FMODE_WRITE) {
655 ret = gfs2_qa_get(GFS2_I(inode));
656 if (ret)
657 goto fail;
658 }
659 return 0;
660
661fail:
662 kfree(file->private_data);
663 file->private_data = NULL;
664 return ret;
665}
666
667/**
668 * gfs2_open - open a file
669 * @inode: the inode to open
670 * @file: the struct file for this opening
671 *
672 * After atomic_open, this function is only used for opening files
673 * which are already cached. We must still get the glock for regular
674 * files to ensure that we have the file size uptodate for the large
675 * file check which is in the common code. That is only an issue for
676 * regular files though.
677 *
678 * Returns: errno
679 */
680
681static int gfs2_open(struct inode *inode, struct file *file)
682{
683 struct gfs2_inode *ip = GFS2_I(inode);
684 struct gfs2_holder i_gh;
685 int error;
686 bool need_unlock = false;
687
688 if (S_ISREG(ip->i_inode.i_mode)) {
689 error = gfs2_glock_nq_init(ip->i_gl, LM_ST_SHARED, LM_FLAG_ANY,
690 &i_gh);
691 if (error)
692 return error;
693 need_unlock = true;
694 }
695
696 error = gfs2_open_common(inode, file);
697
698 if (need_unlock)
699 gfs2_glock_dq_uninit(&i_gh);
700
701 return error;
702}
703
704/**
705 * gfs2_release - called to close a struct file
706 * @inode: the inode the struct file belongs to
707 * @file: the struct file being closed
708 *
709 * Returns: errno
710 */
711
712static int gfs2_release(struct inode *inode, struct file *file)
713{
714 struct gfs2_inode *ip = GFS2_I(inode);
715
716 kfree(file->private_data);
717 file->private_data = NULL;
718
719 if (file->f_mode & FMODE_WRITE) {
720 gfs2_rs_delete(ip, &inode->i_writecount);
721 gfs2_qa_put(ip);
722 }
723 return 0;
724}
725
726/**
727 * gfs2_fsync - sync the dirty data for a file (across the cluster)
728 * @file: the file that points to the dentry
729 * @start: the start position in the file to sync
730 * @end: the end position in the file to sync
731 * @datasync: set if we can ignore timestamp changes
732 *
733 * We split the data flushing here so that we don't wait for the data
734 * until after we've also sent the metadata to disk. Note that for
735 * data=ordered, we will write & wait for the data at the log flush
736 * stage anyway, so this is unlikely to make much of a difference
737 * except in the data=writeback case.
738 *
739 * If the fdatawrite fails due to any reason except -EIO, we will
740 * continue the remainder of the fsync, although we'll still report
741 * the error at the end. This is to match filemap_write_and_wait_range()
742 * behaviour.
743 *
744 * Returns: errno
745 */
746
747static int gfs2_fsync(struct file *file, loff_t start, loff_t end,
748 int datasync)
749{
750 struct address_space *mapping = file->f_mapping;
751 struct inode *inode = mapping->host;
752 int sync_state = inode->i_state & I_DIRTY_ALL;
753 struct gfs2_inode *ip = GFS2_I(inode);
754 int ret = 0, ret1 = 0;
755
756 if (mapping->nrpages) {
757 ret1 = filemap_fdatawrite_range(mapping, start, end);
758 if (ret1 == -EIO)
759 return ret1;
760 }
761
762 if (!gfs2_is_jdata(ip))
763 sync_state &= ~I_DIRTY_PAGES;
764 if (datasync)
765 sync_state &= ~(I_DIRTY_SYNC | I_DIRTY_TIME);
766
767 if (sync_state) {
768 ret = sync_inode_metadata(inode, 1);
769 if (ret)
770 return ret;
771 if (gfs2_is_jdata(ip))
772 ret = file_write_and_wait(file);
773 if (ret)
774 return ret;
775 gfs2_ail_flush(ip->i_gl, 1);
776 }
777
778 if (mapping->nrpages)
779 ret = file_fdatawait_range(file, start, end);
780
781 return ret ? ret : ret1;
782}
783
784static ssize_t gfs2_file_direct_read(struct kiocb *iocb, struct iov_iter *to,
785 struct gfs2_holder *gh)
786{
787 struct file *file = iocb->ki_filp;
788 struct gfs2_inode *ip = GFS2_I(file->f_mapping->host);
789 size_t count = iov_iter_count(to);
790 ssize_t ret;
791
792 if (!count)
793 return 0; /* skip atime */
794
795 gfs2_holder_init(ip->i_gl, LM_ST_DEFERRED, 0, gh);
796 ret = gfs2_glock_nq(gh);
797 if (ret)
798 goto out_uninit;
799
800 ret = iomap_dio_rw(iocb, to, &gfs2_iomap_ops, NULL,
801 is_sync_kiocb(iocb));
802
803 gfs2_glock_dq(gh);
804out_uninit:
805 gfs2_holder_uninit(gh);
806 return ret;
807}
808
809static ssize_t gfs2_file_direct_write(struct kiocb *iocb, struct iov_iter *from,
810 struct gfs2_holder *gh)
811{
812 struct file *file = iocb->ki_filp;
813 struct inode *inode = file->f_mapping->host;
814 struct gfs2_inode *ip = GFS2_I(inode);
815 size_t len = iov_iter_count(from);
816 loff_t offset = iocb->ki_pos;
817 ssize_t ret;
818
819 /*
820 * Deferred lock, even if its a write, since we do no allocation on
821 * this path. All we need to change is the atime, and this lock mode
822 * ensures that other nodes have flushed their buffered read caches
823 * (i.e. their page cache entries for this inode). We do not,
824 * unfortunately, have the option of only flushing a range like the
825 * VFS does.
826 */
827 gfs2_holder_init(ip->i_gl, LM_ST_DEFERRED, 0, gh);
828 ret = gfs2_glock_nq(gh);
829 if (ret)
830 goto out_uninit;
831
832 /* Silently fall back to buffered I/O when writing beyond EOF */
833 if (offset + len > i_size_read(&ip->i_inode))
834 goto out;
835
836 ret = iomap_dio_rw(iocb, from, &gfs2_iomap_ops, NULL,
837 is_sync_kiocb(iocb));
838 if (ret == -ENOTBLK)
839 ret = 0;
840out:
841 gfs2_glock_dq(gh);
842out_uninit:
843 gfs2_holder_uninit(gh);
844 return ret;
845}
846
847static ssize_t gfs2_file_read_iter(struct kiocb *iocb, struct iov_iter *to)
848{
849 struct gfs2_inode *ip;
850 struct gfs2_holder gh;
851 size_t written = 0;
852 ssize_t ret;
853
854 if (iocb->ki_flags & IOCB_DIRECT) {
855 ret = gfs2_file_direct_read(iocb, to, &gh);
856 if (likely(ret != -ENOTBLK))
857 return ret;
858 iocb->ki_flags &= ~IOCB_DIRECT;
859 }
860 iocb->ki_flags |= IOCB_NOIO;
861 ret = generic_file_read_iter(iocb, to);
862 iocb->ki_flags &= ~IOCB_NOIO;
863 if (ret >= 0) {
864 if (!iov_iter_count(to))
865 return ret;
866 written = ret;
867 } else {
868 if (ret != -EAGAIN)
869 return ret;
870 if (iocb->ki_flags & IOCB_NOWAIT)
871 return ret;
872 }
873 ip = GFS2_I(iocb->ki_filp->f_mapping->host);
874 gfs2_holder_init(ip->i_gl, LM_ST_SHARED, 0, &gh);
875 ret = gfs2_glock_nq(&gh);
876 if (ret)
877 goto out_uninit;
878 ret = generic_file_read_iter(iocb, to);
879 if (ret > 0)
880 written += ret;
881 gfs2_glock_dq(&gh);
882out_uninit:
883 gfs2_holder_uninit(&gh);
884 return written ? written : ret;
885}
886
887/**
888 * gfs2_file_write_iter - Perform a write to a file
889 * @iocb: The io context
890 * @from: The data to write
891 *
892 * We have to do a lock/unlock here to refresh the inode size for
893 * O_APPEND writes, otherwise we can land up writing at the wrong
894 * offset. There is still a race, but provided the app is using its
895 * own file locking, this will make O_APPEND work as expected.
896 *
897 */
898
899static ssize_t gfs2_file_write_iter(struct kiocb *iocb, struct iov_iter *from)
900{
901 struct file *file = iocb->ki_filp;
902 struct inode *inode = file_inode(file);
903 struct gfs2_inode *ip = GFS2_I(inode);
904 struct gfs2_holder gh;
905 ssize_t ret;
906
907 gfs2_size_hint(file, iocb->ki_pos, iov_iter_count(from));
908
909 if (iocb->ki_flags & IOCB_APPEND) {
910 ret = gfs2_glock_nq_init(ip->i_gl, LM_ST_SHARED, 0, &gh);
911 if (ret)
912 return ret;
913 gfs2_glock_dq_uninit(&gh);
914 }
915
916 inode_lock(inode);
917 ret = generic_write_checks(iocb, from);
918 if (ret <= 0)
919 goto out_unlock;
920
921 ret = file_remove_privs(file);
922 if (ret)
923 goto out_unlock;
924
925 ret = file_update_time(file);
926 if (ret)
927 goto out_unlock;
928
929 if (iocb->ki_flags & IOCB_DIRECT) {
930 struct address_space *mapping = file->f_mapping;
931 ssize_t buffered, ret2;
932
933 ret = gfs2_file_direct_write(iocb, from, &gh);
934 if (ret < 0 || !iov_iter_count(from))
935 goto out_unlock;
936
937 iocb->ki_flags |= IOCB_DSYNC;
938 current->backing_dev_info = inode_to_bdi(inode);
939 buffered = iomap_file_buffered_write(iocb, from, &gfs2_iomap_ops);
940 current->backing_dev_info = NULL;
941 if (unlikely(buffered <= 0))
942 goto out_unlock;
943
944 /*
945 * We need to ensure that the page cache pages are written to
946 * disk and invalidated to preserve the expected O_DIRECT
947 * semantics. If the writeback or invalidate fails, only report
948 * the direct I/O range as we don't know if the buffered pages
949 * made it to disk.
950 */
951 iocb->ki_pos += buffered;
952 ret2 = generic_write_sync(iocb, buffered);
953 invalidate_mapping_pages(mapping,
954 (iocb->ki_pos - buffered) >> PAGE_SHIFT,
955 (iocb->ki_pos - 1) >> PAGE_SHIFT);
956 if (!ret || ret2 > 0)
957 ret += ret2;
958 } else {
959 current->backing_dev_info = inode_to_bdi(inode);
960 ret = iomap_file_buffered_write(iocb, from, &gfs2_iomap_ops);
961 current->backing_dev_info = NULL;
962 if (likely(ret > 0)) {
963 iocb->ki_pos += ret;
964 ret = generic_write_sync(iocb, ret);
965 }
966 }
967
968out_unlock:
969 inode_unlock(inode);
970 return ret;
971}
972
973static int fallocate_chunk(struct inode *inode, loff_t offset, loff_t len,
974 int mode)
975{
976 struct super_block *sb = inode->i_sb;
977 struct gfs2_inode *ip = GFS2_I(inode);
978 loff_t end = offset + len;
979 struct buffer_head *dibh;
980 int error;
981
982 error = gfs2_meta_inode_buffer(ip, &dibh);
983 if (unlikely(error))
984 return error;
985
986 gfs2_trans_add_meta(ip->i_gl, dibh);
987
988 if (gfs2_is_stuffed(ip)) {
989 error = gfs2_unstuff_dinode(ip, NULL);
990 if (unlikely(error))
991 goto out;
992 }
993
994 while (offset < end) {
995 struct iomap iomap = { };
996
997 error = gfs2_iomap_get_alloc(inode, offset, end - offset,
998 &iomap);
999 if (error)
1000 goto out;
1001 offset = iomap.offset + iomap.length;
1002 if (!(iomap.flags & IOMAP_F_NEW))
1003 continue;
1004 error = sb_issue_zeroout(sb, iomap.addr >> inode->i_blkbits,
1005 iomap.length >> inode->i_blkbits,
1006 GFP_NOFS);
1007 if (error) {
1008 fs_err(GFS2_SB(inode), "Failed to zero data buffers\n");
1009 goto out;
1010 }
1011 }
1012out:
1013 brelse(dibh);
1014 return error;
1015}
1016
1017/**
1018 * calc_max_reserv() - Reverse of write_calc_reserv. Given a number of
1019 * blocks, determine how many bytes can be written.
1020 * @ip: The inode in question.
1021 * @len: Max cap of bytes. What we return in *len must be <= this.
1022 * @data_blocks: Compute and return the number of data blocks needed
1023 * @ind_blocks: Compute and return the number of indirect blocks needed
1024 * @max_blocks: The total blocks available to work with.
1025 *
1026 * Returns: void, but @len, @data_blocks and @ind_blocks are filled in.
1027 */
1028static void calc_max_reserv(struct gfs2_inode *ip, loff_t *len,
1029 unsigned int *data_blocks, unsigned int *ind_blocks,
1030 unsigned int max_blocks)
1031{
1032 loff_t max = *len;
1033 const struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode);
1034 unsigned int tmp, max_data = max_blocks - 3 * (sdp->sd_max_height - 1);
1035
1036 for (tmp = max_data; tmp > sdp->sd_diptrs;) {
1037 tmp = DIV_ROUND_UP(tmp, sdp->sd_inptrs);
1038 max_data -= tmp;
1039 }
1040
1041 *data_blocks = max_data;
1042 *ind_blocks = max_blocks - max_data;
1043 *len = ((loff_t)max_data - 3) << sdp->sd_sb.sb_bsize_shift;
1044 if (*len > max) {
1045 *len = max;
1046 gfs2_write_calc_reserv(ip, max, data_blocks, ind_blocks);
1047 }
1048}
1049
1050static long __gfs2_fallocate(struct file *file, int mode, loff_t offset, loff_t len)
1051{
1052 struct inode *inode = file_inode(file);
1053 struct gfs2_sbd *sdp = GFS2_SB(inode);
1054 struct gfs2_inode *ip = GFS2_I(inode);
1055 struct gfs2_alloc_parms ap = { .aflags = 0, };
1056 unsigned int data_blocks = 0, ind_blocks = 0, rblocks;
1057 loff_t bytes, max_bytes, max_blks;
1058 int error;
1059 const loff_t pos = offset;
1060 const loff_t count = len;
1061 loff_t bsize_mask = ~((loff_t)sdp->sd_sb.sb_bsize - 1);
1062 loff_t next = (offset + len - 1) >> sdp->sd_sb.sb_bsize_shift;
1063 loff_t max_chunk_size = UINT_MAX & bsize_mask;
1064
1065 next = (next + 1) << sdp->sd_sb.sb_bsize_shift;
1066
1067 offset &= bsize_mask;
1068
1069 len = next - offset;
1070 bytes = sdp->sd_max_rg_data * sdp->sd_sb.sb_bsize / 2;
1071 if (!bytes)
1072 bytes = UINT_MAX;
1073 bytes &= bsize_mask;
1074 if (bytes == 0)
1075 bytes = sdp->sd_sb.sb_bsize;
1076
1077 gfs2_size_hint(file, offset, len);
1078
1079 gfs2_write_calc_reserv(ip, PAGE_SIZE, &data_blocks, &ind_blocks);
1080 ap.min_target = data_blocks + ind_blocks;
1081
1082 while (len > 0) {
1083 if (len < bytes)
1084 bytes = len;
1085 if (!gfs2_write_alloc_required(ip, offset, bytes)) {
1086 len -= bytes;
1087 offset += bytes;
1088 continue;
1089 }
1090
1091 /* We need to determine how many bytes we can actually
1092 * fallocate without exceeding quota or going over the
1093 * end of the fs. We start off optimistically by assuming
1094 * we can write max_bytes */
1095 max_bytes = (len > max_chunk_size) ? max_chunk_size : len;
1096
1097 /* Since max_bytes is most likely a theoretical max, we
1098 * calculate a more realistic 'bytes' to serve as a good
1099 * starting point for the number of bytes we may be able
1100 * to write */
1101 gfs2_write_calc_reserv(ip, bytes, &data_blocks, &ind_blocks);
1102 ap.target = data_blocks + ind_blocks;
1103
1104 error = gfs2_quota_lock_check(ip, &ap);
1105 if (error)
1106 return error;
1107 /* ap.allowed tells us how many blocks quota will allow
1108 * us to write. Check if this reduces max_blks */
1109 max_blks = UINT_MAX;
1110 if (ap.allowed)
1111 max_blks = ap.allowed;
1112
1113 error = gfs2_inplace_reserve(ip, &ap);
1114 if (error)
1115 goto out_qunlock;
1116
1117 /* check if the selected rgrp limits our max_blks further */
1118 if (ap.allowed && ap.allowed < max_blks)
1119 max_blks = ap.allowed;
1120
1121 /* Almost done. Calculate bytes that can be written using
1122 * max_blks. We also recompute max_bytes, data_blocks and
1123 * ind_blocks */
1124 calc_max_reserv(ip, &max_bytes, &data_blocks,
1125 &ind_blocks, max_blks);
1126
1127 rblocks = RES_DINODE + ind_blocks + RES_STATFS + RES_QUOTA +
1128 RES_RG_HDR + gfs2_rg_blocks(ip, data_blocks + ind_blocks);
1129 if (gfs2_is_jdata(ip))
1130 rblocks += data_blocks ? data_blocks : 1;
1131
1132 error = gfs2_trans_begin(sdp, rblocks,
1133 PAGE_SIZE >> inode->i_blkbits);
1134 if (error)
1135 goto out_trans_fail;
1136
1137 error = fallocate_chunk(inode, offset, max_bytes, mode);
1138 gfs2_trans_end(sdp);
1139
1140 if (error)
1141 goto out_trans_fail;
1142
1143 len -= max_bytes;
1144 offset += max_bytes;
1145 gfs2_inplace_release(ip);
1146 gfs2_quota_unlock(ip);
1147 }
1148
1149 if (!(mode & FALLOC_FL_KEEP_SIZE) && (pos + count) > inode->i_size)
1150 i_size_write(inode, pos + count);
1151 file_update_time(file);
1152 mark_inode_dirty(inode);
1153
1154 if ((file->f_flags & O_DSYNC) || IS_SYNC(file->f_mapping->host))
1155 return vfs_fsync_range(file, pos, pos + count - 1,
1156 (file->f_flags & __O_SYNC) ? 0 : 1);
1157 return 0;
1158
1159out_trans_fail:
1160 gfs2_inplace_release(ip);
1161out_qunlock:
1162 gfs2_quota_unlock(ip);
1163 return error;
1164}
1165
1166static long gfs2_fallocate(struct file *file, int mode, loff_t offset, loff_t len)
1167{
1168 struct inode *inode = file_inode(file);
1169 struct gfs2_sbd *sdp = GFS2_SB(inode);
1170 struct gfs2_inode *ip = GFS2_I(inode);
1171 struct gfs2_holder gh;
1172 int ret;
1173
1174 if (mode & ~(FALLOC_FL_PUNCH_HOLE | FALLOC_FL_KEEP_SIZE))
1175 return -EOPNOTSUPP;
1176 /* fallocate is needed by gfs2_grow to reserve space in the rindex */
1177 if (gfs2_is_jdata(ip) && inode != sdp->sd_rindex)
1178 return -EOPNOTSUPP;
1179
1180 inode_lock(inode);
1181
1182 gfs2_holder_init(ip->i_gl, LM_ST_EXCLUSIVE, 0, &gh);
1183 ret = gfs2_glock_nq(&gh);
1184 if (ret)
1185 goto out_uninit;
1186
1187 if (!(mode & FALLOC_FL_KEEP_SIZE) &&
1188 (offset + len) > inode->i_size) {
1189 ret = inode_newsize_ok(inode, offset + len);
1190 if (ret)
1191 goto out_unlock;
1192 }
1193
1194 ret = get_write_access(inode);
1195 if (ret)
1196 goto out_unlock;
1197
1198 if (mode & FALLOC_FL_PUNCH_HOLE) {
1199 ret = __gfs2_punch_hole(file, offset, len);
1200 } else {
1201 ret = __gfs2_fallocate(file, mode, offset, len);
1202 if (ret)
1203 gfs2_rs_deltree(&ip->i_res);
1204 }
1205
1206 put_write_access(inode);
1207out_unlock:
1208 gfs2_glock_dq(&gh);
1209out_uninit:
1210 gfs2_holder_uninit(&gh);
1211 inode_unlock(inode);
1212 return ret;
1213}
1214
1215static ssize_t gfs2_file_splice_write(struct pipe_inode_info *pipe,
1216 struct file *out, loff_t *ppos,
1217 size_t len, unsigned int flags)
1218{
1219 ssize_t ret;
1220
1221 gfs2_size_hint(out, *ppos, len);
1222
1223 ret = iter_file_splice_write(pipe, out, ppos, len, flags);
1224 return ret;
1225}
1226
1227#ifdef CONFIG_GFS2_FS_LOCKING_DLM
1228
1229/**
1230 * gfs2_lock - acquire/release a posix lock on a file
1231 * @file: the file pointer
1232 * @cmd: either modify or retrieve lock state, possibly wait
1233 * @fl: type and range of lock
1234 *
1235 * Returns: errno
1236 */
1237
1238static int gfs2_lock(struct file *file, int cmd, struct file_lock *fl)
1239{
1240 struct gfs2_inode *ip = GFS2_I(file->f_mapping->host);
1241 struct gfs2_sbd *sdp = GFS2_SB(file->f_mapping->host);
1242 struct lm_lockstruct *ls = &sdp->sd_lockstruct;
1243
1244 if (!(fl->fl_flags & FL_POSIX))
1245 return -ENOLCK;
1246 if (__mandatory_lock(&ip->i_inode) && fl->fl_type != F_UNLCK)
1247 return -ENOLCK;
1248
1249 if (cmd == F_CANCELLK) {
1250 /* Hack: */
1251 cmd = F_SETLK;
1252 fl->fl_type = F_UNLCK;
1253 }
1254 if (unlikely(gfs2_withdrawn(sdp))) {
1255 if (fl->fl_type == F_UNLCK)
1256 locks_lock_file_wait(file, fl);
1257 return -EIO;
1258 }
1259 if (IS_GETLK(cmd))
1260 return dlm_posix_get(ls->ls_dlm, ip->i_no_addr, file, fl);
1261 else if (fl->fl_type == F_UNLCK)
1262 return dlm_posix_unlock(ls->ls_dlm, ip->i_no_addr, file, fl);
1263 else
1264 return dlm_posix_lock(ls->ls_dlm, ip->i_no_addr, file, cmd, fl);
1265}
1266
1267static int do_flock(struct file *file, int cmd, struct file_lock *fl)
1268{
1269 struct gfs2_file *fp = file->private_data;
1270 struct gfs2_holder *fl_gh = &fp->f_fl_gh;
1271 struct gfs2_inode *ip = GFS2_I(file_inode(file));
1272 struct gfs2_glock *gl;
1273 unsigned int state;
1274 u16 flags;
1275 int error = 0;
1276 int sleeptime;
1277
1278 state = (fl->fl_type == F_WRLCK) ? LM_ST_EXCLUSIVE : LM_ST_SHARED;
1279 flags = (IS_SETLKW(cmd) ? 0 : LM_FLAG_TRY_1CB) | GL_EXACT;
1280
1281 mutex_lock(&fp->f_fl_mutex);
1282
1283 if (gfs2_holder_initialized(fl_gh)) {
1284 struct file_lock request;
1285 if (fl_gh->gh_state == state)
1286 goto out;
1287 locks_init_lock(&request);
1288 request.fl_type = F_UNLCK;
1289 request.fl_flags = FL_FLOCK;
1290 locks_lock_file_wait(file, &request);
1291 gfs2_glock_dq(fl_gh);
1292 gfs2_holder_reinit(state, flags, fl_gh);
1293 } else {
1294 error = gfs2_glock_get(GFS2_SB(&ip->i_inode), ip->i_no_addr,
1295 &gfs2_flock_glops, CREATE, &gl);
1296 if (error)
1297 goto out;
1298 gfs2_holder_init(gl, state, flags, fl_gh);
1299 gfs2_glock_put(gl);
1300 }
1301 for (sleeptime = 1; sleeptime <= 4; sleeptime <<= 1) {
1302 error = gfs2_glock_nq(fl_gh);
1303 if (error != GLR_TRYFAILED)
1304 break;
1305 fl_gh->gh_flags = LM_FLAG_TRY | GL_EXACT;
1306 fl_gh->gh_error = 0;
1307 msleep(sleeptime);
1308 }
1309 if (error) {
1310 gfs2_holder_uninit(fl_gh);
1311 if (error == GLR_TRYFAILED)
1312 error = -EAGAIN;
1313 } else {
1314 error = locks_lock_file_wait(file, fl);
1315 gfs2_assert_warn(GFS2_SB(&ip->i_inode), !error);
1316 }
1317
1318out:
1319 mutex_unlock(&fp->f_fl_mutex);
1320 return error;
1321}
1322
1323static void do_unflock(struct file *file, struct file_lock *fl)
1324{
1325 struct gfs2_file *fp = file->private_data;
1326 struct gfs2_holder *fl_gh = &fp->f_fl_gh;
1327
1328 mutex_lock(&fp->f_fl_mutex);
1329 locks_lock_file_wait(file, fl);
1330 if (gfs2_holder_initialized(fl_gh)) {
1331 gfs2_glock_dq(fl_gh);
1332 gfs2_holder_uninit(fl_gh);
1333 }
1334 mutex_unlock(&fp->f_fl_mutex);
1335}
1336
1337/**
1338 * gfs2_flock - acquire/release a flock lock on a file
1339 * @file: the file pointer
1340 * @cmd: either modify or retrieve lock state, possibly wait
1341 * @fl: type and range of lock
1342 *
1343 * Returns: errno
1344 */
1345
1346static int gfs2_flock(struct file *file, int cmd, struct file_lock *fl)
1347{
1348 if (!(fl->fl_flags & FL_FLOCK))
1349 return -ENOLCK;
1350 if (fl->fl_type & LOCK_MAND)
1351 return -EOPNOTSUPP;
1352
1353 if (fl->fl_type == F_UNLCK) {
1354 do_unflock(file, fl);
1355 return 0;
1356 } else {
1357 return do_flock(file, cmd, fl);
1358 }
1359}
1360
1361const struct file_operations gfs2_file_fops = {
1362 .llseek = gfs2_llseek,
1363 .read_iter = gfs2_file_read_iter,
1364 .write_iter = gfs2_file_write_iter,
1365 .iopoll = iomap_dio_iopoll,
1366 .unlocked_ioctl = gfs2_ioctl,
1367 .compat_ioctl = gfs2_compat_ioctl,
1368 .mmap = gfs2_mmap,
1369 .open = gfs2_open,
1370 .release = gfs2_release,
1371 .fsync = gfs2_fsync,
1372 .lock = gfs2_lock,
1373 .flock = gfs2_flock,
1374 .splice_read = generic_file_splice_read,
1375 .splice_write = gfs2_file_splice_write,
1376 .setlease = simple_nosetlease,
1377 .fallocate = gfs2_fallocate,
1378};
1379
1380const struct file_operations gfs2_dir_fops = {
1381 .iterate_shared = gfs2_readdir,
1382 .unlocked_ioctl = gfs2_ioctl,
1383 .compat_ioctl = gfs2_compat_ioctl,
1384 .open = gfs2_open,
1385 .release = gfs2_release,
1386 .fsync = gfs2_fsync,
1387 .lock = gfs2_lock,
1388 .flock = gfs2_flock,
1389 .llseek = default_llseek,
1390};
1391
1392#endif /* CONFIG_GFS2_FS_LOCKING_DLM */
1393
1394const struct file_operations gfs2_file_fops_nolock = {
1395 .llseek = gfs2_llseek,
1396 .read_iter = gfs2_file_read_iter,
1397 .write_iter = gfs2_file_write_iter,
1398 .iopoll = iomap_dio_iopoll,
1399 .unlocked_ioctl = gfs2_ioctl,
1400 .compat_ioctl = gfs2_compat_ioctl,
1401 .mmap = gfs2_mmap,
1402 .open = gfs2_open,
1403 .release = gfs2_release,
1404 .fsync = gfs2_fsync,
1405 .splice_read = generic_file_splice_read,
1406 .splice_write = gfs2_file_splice_write,
1407 .setlease = generic_setlease,
1408 .fallocate = gfs2_fallocate,
1409};
1410
1411const struct file_operations gfs2_dir_fops_nolock = {
1412 .iterate_shared = gfs2_readdir,
1413 .unlocked_ioctl = gfs2_ioctl,
1414 .compat_ioctl = gfs2_compat_ioctl,
1415 .open = gfs2_open,
1416 .release = gfs2_release,
1417 .fsync = gfs2_fsync,
1418 .llseek = default_llseek,
1419};
1420