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#include <linux/fileattr.h>
29
30#include "gfs2.h"
31#include "incore.h"
32#include "bmap.h"
33#include "aops.h"
34#include "dir.h"
35#include "glock.h"
36#include "glops.h"
37#include "inode.h"
38#include "log.h"
39#include "meta_io.h"
40#include "quota.h"
41#include "rgrp.h"
42#include "trans.h"
43#include "util.h"
44
45/**
46 * gfs2_llseek - seek to a location in a file
47 * @file: the file
48 * @offset: the offset
49 * @whence: Where to seek from (SEEK_SET, SEEK_CUR, or SEEK_END)
50 *
51 * SEEK_END requires the glock for the file because it references the
52 * file's size.
53 *
54 * Returns: The new offset, or errno
55 */
56
57static loff_t gfs2_llseek(struct file *file, loff_t offset, int whence)
58{
59 struct gfs2_inode *ip = GFS2_I(file->f_mapping->host);
60 struct gfs2_holder i_gh;
61 loff_t error;
62
63 switch (whence) {
64 case SEEK_END:
65 error = gfs2_glock_nq_init(ip->i_gl, LM_ST_SHARED, LM_FLAG_ANY,
66 &i_gh);
67 if (!error) {
68 error = generic_file_llseek(file, offset, whence);
69 gfs2_glock_dq_uninit(&i_gh);
70 }
71 break;
72
73 case SEEK_DATA:
74 error = gfs2_seek_data(file, offset);
75 break;
76
77 case SEEK_HOLE:
78 error = gfs2_seek_hole(file, offset);
79 break;
80
81 case SEEK_CUR:
82 case SEEK_SET:
83 /*
84 * These don't reference inode->i_size and don't depend on the
85 * block mapping, so we don't need the glock.
86 */
87 error = generic_file_llseek(file, offset, whence);
88 break;
89 default:
90 error = -EINVAL;
91 }
92
93 return error;
94}
95
96/**
97 * gfs2_readdir - Iterator for a directory
98 * @file: The directory to read from
99 * @ctx: What to feed directory entries to
100 *
101 * Returns: errno
102 */
103
104static int gfs2_readdir(struct file *file, struct dir_context *ctx)
105{
106 struct inode *dir = file->f_mapping->host;
107 struct gfs2_inode *dip = GFS2_I(dir);
108 struct gfs2_holder d_gh;
109 int error;
110
111 error = gfs2_glock_nq_init(dip->i_gl, LM_ST_SHARED, 0, &d_gh);
112 if (error)
113 return error;
114
115 error = gfs2_dir_read(dir, ctx, &file->f_ra);
116
117 gfs2_glock_dq_uninit(&d_gh);
118
119 return error;
120}
121
122/*
123 * struct fsflag_gfs2flag
124 *
125 * The FS_JOURNAL_DATA_FL flag maps to GFS2_DIF_INHERIT_JDATA for directories,
126 * and to GFS2_DIF_JDATA for non-directories.
127 */
128static struct {
129 u32 fsflag;
130 u32 gfsflag;
131} fsflag_gfs2flag[] = {
132 {FS_SYNC_FL, GFS2_DIF_SYNC},
133 {FS_IMMUTABLE_FL, GFS2_DIF_IMMUTABLE},
134 {FS_APPEND_FL, GFS2_DIF_APPENDONLY},
135 {FS_NOATIME_FL, GFS2_DIF_NOATIME},
136 {FS_INDEX_FL, GFS2_DIF_EXHASH},
137 {FS_TOPDIR_FL, GFS2_DIF_TOPDIR},
138 {FS_JOURNAL_DATA_FL, GFS2_DIF_JDATA | GFS2_DIF_INHERIT_JDATA},
139};
140
141static inline u32 gfs2_gfsflags_to_fsflags(struct inode *inode, u32 gfsflags)
142{
143 int i;
144 u32 fsflags = 0;
145
146 if (S_ISDIR(inode->i_mode))
147 gfsflags &= ~GFS2_DIF_JDATA;
148 else
149 gfsflags &= ~GFS2_DIF_INHERIT_JDATA;
150
151 for (i = 0; i < ARRAY_SIZE(fsflag_gfs2flag); i++)
152 if (gfsflags & fsflag_gfs2flag[i].gfsflag)
153 fsflags |= fsflag_gfs2flag[i].fsflag;
154 return fsflags;
155}
156
157int gfs2_fileattr_get(struct dentry *dentry, struct fileattr *fa)
158{
159 struct inode *inode = d_inode(dentry);
160 struct gfs2_inode *ip = GFS2_I(inode);
161 struct gfs2_holder gh;
162 int error;
163 u32 fsflags;
164
165 if (d_is_special(dentry))
166 return -ENOTTY;
167
168 gfs2_holder_init(ip->i_gl, LM_ST_SHARED, 0, &gh);
169 error = gfs2_glock_nq(&gh);
170 if (error)
171 goto out_uninit;
172
173 fsflags = gfs2_gfsflags_to_fsflags(inode, ip->i_diskflags);
174
175 fileattr_fill_flags(fa, fsflags);
176
177 gfs2_glock_dq(&gh);
178out_uninit:
179 gfs2_holder_uninit(&gh);
180 return error;
181}
182
183void gfs2_set_inode_flags(struct inode *inode)
184{
185 struct gfs2_inode *ip = GFS2_I(inode);
186 unsigned int flags = inode->i_flags;
187
188 flags &= ~(S_SYNC|S_APPEND|S_IMMUTABLE|S_NOATIME|S_DIRSYNC|S_NOSEC);
189 if ((ip->i_eattr == 0) && !is_sxid(inode->i_mode))
190 flags |= S_NOSEC;
191 if (ip->i_diskflags & GFS2_DIF_IMMUTABLE)
192 flags |= S_IMMUTABLE;
193 if (ip->i_diskflags & GFS2_DIF_APPENDONLY)
194 flags |= S_APPEND;
195 if (ip->i_diskflags & GFS2_DIF_NOATIME)
196 flags |= S_NOATIME;
197 if (ip->i_diskflags & GFS2_DIF_SYNC)
198 flags |= S_SYNC;
199 inode->i_flags = flags;
200}
201
202/* Flags that can be set by user space */
203#define GFS2_FLAGS_USER_SET (GFS2_DIF_JDATA| \
204 GFS2_DIF_IMMUTABLE| \
205 GFS2_DIF_APPENDONLY| \
206 GFS2_DIF_NOATIME| \
207 GFS2_DIF_SYNC| \
208 GFS2_DIF_TOPDIR| \
209 GFS2_DIF_INHERIT_JDATA)
210
211/**
212 * do_gfs2_set_flags - set flags on an inode
213 * @inode: The inode
214 * @reqflags: The flags to set
215 * @mask: Indicates which flags are valid
216 * @fsflags: The FS_* inode flags passed in
217 *
218 */
219static int do_gfs2_set_flags(struct inode *inode, u32 reqflags, u32 mask,
220 const u32 fsflags)
221{
222 struct gfs2_inode *ip = GFS2_I(inode);
223 struct gfs2_sbd *sdp = GFS2_SB(inode);
224 struct buffer_head *bh;
225 struct gfs2_holder gh;
226 int error;
227 u32 new_flags, flags;
228
229 error = gfs2_glock_nq_init(ip->i_gl, LM_ST_EXCLUSIVE, 0, &gh);
230 if (error)
231 return error;
232
233 error = 0;
234 flags = ip->i_diskflags;
235 new_flags = (flags & ~mask) | (reqflags & mask);
236 if ((new_flags ^ flags) == 0)
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 (!IS_IMMUTABLE(inode)) {
245 error = gfs2_permission(&init_user_ns, inode, MAY_WRITE);
246 if (error)
247 goto out;
248 }
249 if ((flags ^ new_flags) & GFS2_DIF_JDATA) {
250 if (new_flags & GFS2_DIF_JDATA)
251 gfs2_log_flush(sdp, ip->i_gl,
252 GFS2_LOG_HEAD_FLUSH_NORMAL |
253 GFS2_LFC_SET_FLAGS);
254 error = filemap_fdatawrite(inode->i_mapping);
255 if (error)
256 goto out;
257 error = filemap_fdatawait(inode->i_mapping);
258 if (error)
259 goto out;
260 if (new_flags & GFS2_DIF_JDATA)
261 gfs2_ordered_del_inode(ip);
262 }
263 error = gfs2_trans_begin(sdp, RES_DINODE, 0);
264 if (error)
265 goto out;
266 error = gfs2_meta_inode_buffer(ip, &bh);
267 if (error)
268 goto out_trans_end;
269 inode->i_ctime = current_time(inode);
270 gfs2_trans_add_meta(ip->i_gl, bh);
271 ip->i_diskflags = new_flags;
272 gfs2_dinode_out(ip, bh->b_data);
273 brelse(bh);
274 gfs2_set_inode_flags(inode);
275 gfs2_set_aops(inode);
276out_trans_end:
277 gfs2_trans_end(sdp);
278out:
279 gfs2_glock_dq_uninit(&gh);
280 return error;
281}
282
283int gfs2_fileattr_set(struct user_namespace *mnt_userns,
284 struct dentry *dentry, struct fileattr *fa)
285{
286 struct inode *inode = d_inode(dentry);
287 u32 fsflags = fa->flags, gfsflags = 0;
288 u32 mask;
289 int i;
290
291 if (d_is_special(dentry))
292 return -ENOTTY;
293
294 if (fileattr_has_fsx(fa))
295 return -EOPNOTSUPP;
296
297 for (i = 0; i < ARRAY_SIZE(fsflag_gfs2flag); i++) {
298 if (fsflags & fsflag_gfs2flag[i].fsflag) {
299 fsflags &= ~fsflag_gfs2flag[i].fsflag;
300 gfsflags |= fsflag_gfs2flag[i].gfsflag;
301 }
302 }
303 if (fsflags || gfsflags & ~GFS2_FLAGS_USER_SET)
304 return -EINVAL;
305
306 mask = GFS2_FLAGS_USER_SET;
307 if (S_ISDIR(inode->i_mode)) {
308 mask &= ~GFS2_DIF_JDATA;
309 } else {
310 /* The GFS2_DIF_TOPDIR flag is only valid for directories. */
311 if (gfsflags & GFS2_DIF_TOPDIR)
312 return -EINVAL;
313 mask &= ~(GFS2_DIF_TOPDIR | GFS2_DIF_INHERIT_JDATA);
314 }
315
316 return do_gfs2_set_flags(inode, gfsflags, mask, fsflags);
317}
318
319static int gfs2_getlabel(struct file *filp, char __user *label)
320{
321 struct inode *inode = file_inode(filp);
322 struct gfs2_sbd *sdp = GFS2_SB(inode);
323
324 if (copy_to_user(label, sdp->sd_sb.sb_locktable, GFS2_LOCKNAME_LEN))
325 return -EFAULT;
326
327 return 0;
328}
329
330static long gfs2_ioctl(struct file *filp, unsigned int cmd, unsigned long arg)
331{
332 switch(cmd) {
333 case FITRIM:
334 return gfs2_fitrim(filp, (void __user *)arg);
335 case FS_IOC_GETFSLABEL:
336 return gfs2_getlabel(filp, (char __user *)arg);
337 }
338
339 return -ENOTTY;
340}
341
342#ifdef CONFIG_COMPAT
343static long gfs2_compat_ioctl(struct file *filp, unsigned int cmd, unsigned long arg)
344{
345 switch(cmd) {
346 /* Keep this list in sync with gfs2_ioctl */
347 case FITRIM:
348 case FS_IOC_GETFSLABEL:
349 break;
350 default:
351 return -ENOIOCTLCMD;
352 }
353
354 return gfs2_ioctl(filp, cmd, (unsigned long)compat_ptr(arg));
355}
356#else
357#define gfs2_compat_ioctl NULL
358#endif
359
360/**
361 * gfs2_size_hint - Give a hint to the size of a write request
362 * @filep: The struct file
363 * @offset: The file offset of the write
364 * @size: The length of the write
365 *
366 * When we are about to do a write, this function records the total
367 * write size in order to provide a suitable hint to the lower layers
368 * about how many blocks will be required.
369 *
370 */
371
372static void gfs2_size_hint(struct file *filep, loff_t offset, size_t size)
373{
374 struct inode *inode = file_inode(filep);
375 struct gfs2_sbd *sdp = GFS2_SB(inode);
376 struct gfs2_inode *ip = GFS2_I(inode);
377 size_t blks = (size + sdp->sd_sb.sb_bsize - 1) >> sdp->sd_sb.sb_bsize_shift;
378 int hint = min_t(size_t, INT_MAX, blks);
379
380 if (hint > atomic_read(&ip->i_sizehint))
381 atomic_set(&ip->i_sizehint, hint);
382}
383
384/**
385 * gfs2_allocate_page_backing - Allocate blocks for a write fault
386 * @page: The (locked) page to allocate backing for
387 * @length: Size of the allocation
388 *
389 * We try to allocate all the blocks required for the page in one go. This
390 * might fail for various reasons, so we keep trying until all the blocks to
391 * back this page are allocated. If some of the blocks are already allocated,
392 * that is ok too.
393 */
394static int gfs2_allocate_page_backing(struct page *page, unsigned int length)
395{
396 u64 pos = page_offset(page);
397
398 do {
399 struct iomap iomap = { };
400
401 if (gfs2_iomap_alloc(page->mapping->host, pos, length, &iomap))
402 return -EIO;
403
404 if (length < iomap.length)
405 iomap.length = length;
406 length -= iomap.length;
407 pos += iomap.length;
408 } while (length > 0);
409
410 return 0;
411}
412
413/**
414 * gfs2_page_mkwrite - Make a shared, mmap()ed, page writable
415 * @vmf: The virtual memory fault containing the page to become writable
416 *
417 * When the page becomes writable, we need to ensure that we have
418 * blocks allocated on disk to back that page.
419 */
420
421static vm_fault_t gfs2_page_mkwrite(struct vm_fault *vmf)
422{
423 struct page *page = vmf->page;
424 struct inode *inode = file_inode(vmf->vma->vm_file);
425 struct gfs2_inode *ip = GFS2_I(inode);
426 struct gfs2_sbd *sdp = GFS2_SB(inode);
427 struct gfs2_alloc_parms ap = { .aflags = 0, };
428 u64 offset = page_offset(page);
429 unsigned int data_blocks, ind_blocks, rblocks;
430 vm_fault_t ret = VM_FAULT_LOCKED;
431 struct gfs2_holder gh;
432 unsigned int length;
433 loff_t size;
434 int err;
435
436 sb_start_pagefault(inode->i_sb);
437
438 gfs2_holder_init(ip->i_gl, LM_ST_EXCLUSIVE, 0, &gh);
439 err = gfs2_glock_nq(&gh);
440 if (err) {
441 ret = block_page_mkwrite_return(err);
442 goto out_uninit;
443 }
444
445 /* Check page index against inode size */
446 size = i_size_read(inode);
447 if (offset >= size) {
448 ret = VM_FAULT_SIGBUS;
449 goto out_unlock;
450 }
451
452 /* Update file times before taking page lock */
453 file_update_time(vmf->vma->vm_file);
454
455 /* page is wholly or partially inside EOF */
456 if (size - offset < PAGE_SIZE)
457 length = size - offset;
458 else
459 length = PAGE_SIZE;
460
461 gfs2_size_hint(vmf->vma->vm_file, offset, length);
462
463 set_bit(GLF_DIRTY, &ip->i_gl->gl_flags);
464 set_bit(GIF_SW_PAGED, &ip->i_flags);
465
466 /*
467 * iomap_writepage / iomap_writepages currently don't support inline
468 * files, so always unstuff here.
469 */
470
471 if (!gfs2_is_stuffed(ip) &&
472 !gfs2_write_alloc_required(ip, offset, length)) {
473 lock_page(page);
474 if (!PageUptodate(page) || page->mapping != inode->i_mapping) {
475 ret = VM_FAULT_NOPAGE;
476 unlock_page(page);
477 }
478 goto out_unlock;
479 }
480
481 err = gfs2_rindex_update(sdp);
482 if (err) {
483 ret = block_page_mkwrite_return(err);
484 goto out_unlock;
485 }
486
487 gfs2_write_calc_reserv(ip, length, &data_blocks, &ind_blocks);
488 ap.target = data_blocks + ind_blocks;
489 err = gfs2_quota_lock_check(ip, &ap);
490 if (err) {
491 ret = block_page_mkwrite_return(err);
492 goto out_unlock;
493 }
494 err = gfs2_inplace_reserve(ip, &ap);
495 if (err) {
496 ret = block_page_mkwrite_return(err);
497 goto out_quota_unlock;
498 }
499
500 rblocks = RES_DINODE + ind_blocks;
501 if (gfs2_is_jdata(ip))
502 rblocks += data_blocks ? data_blocks : 1;
503 if (ind_blocks || data_blocks) {
504 rblocks += RES_STATFS + RES_QUOTA;
505 rblocks += gfs2_rg_blocks(ip, data_blocks + ind_blocks);
506 }
507 err = gfs2_trans_begin(sdp, rblocks, 0);
508 if (err) {
509 ret = block_page_mkwrite_return(err);
510 goto out_trans_fail;
511 }
512
513 /* Unstuff, if required, and allocate backing blocks for page */
514 if (gfs2_is_stuffed(ip)) {
515 err = gfs2_unstuff_dinode(ip);
516 if (err) {
517 ret = block_page_mkwrite_return(err);
518 goto out_trans_end;
519 }
520 }
521
522 lock_page(page);
523 /* If truncated, we must retry the operation, we may have raced
524 * with the glock demotion code.
525 */
526 if (!PageUptodate(page) || page->mapping != inode->i_mapping) {
527 ret = VM_FAULT_NOPAGE;
528 goto out_page_locked;
529 }
530
531 err = gfs2_allocate_page_backing(page, length);
532 if (err)
533 ret = block_page_mkwrite_return(err);
534
535out_page_locked:
536 if (ret != VM_FAULT_LOCKED)
537 unlock_page(page);
538out_trans_end:
539 gfs2_trans_end(sdp);
540out_trans_fail:
541 gfs2_inplace_release(ip);
542out_quota_unlock:
543 gfs2_quota_unlock(ip);
544out_unlock:
545 gfs2_glock_dq(&gh);
546out_uninit:
547 gfs2_holder_uninit(&gh);
548 if (ret == VM_FAULT_LOCKED) {
549 set_page_dirty(page);
550 wait_for_stable_page(page);
551 }
552 sb_end_pagefault(inode->i_sb);
553 return ret;
554}
555
556static vm_fault_t gfs2_fault(struct vm_fault *vmf)
557{
558 struct inode *inode = file_inode(vmf->vma->vm_file);
559 struct gfs2_inode *ip = GFS2_I(inode);
560 struct gfs2_holder gh;
561 vm_fault_t ret;
562 int err;
563
564 gfs2_holder_init(ip->i_gl, LM_ST_SHARED, 0, &gh);
565 err = gfs2_glock_nq(&gh);
566 if (err) {
567 ret = block_page_mkwrite_return(err);
568 goto out_uninit;
569 }
570 ret = filemap_fault(vmf);
571 gfs2_glock_dq(&gh);
572out_uninit:
573 gfs2_holder_uninit(&gh);
574 return ret;
575}
576
577static const struct vm_operations_struct gfs2_vm_ops = {
578 .fault = gfs2_fault,
579 .map_pages = filemap_map_pages,
580 .page_mkwrite = gfs2_page_mkwrite,
581};
582
583/**
584 * gfs2_mmap
585 * @file: The file to map
586 * @vma: The VMA which described the mapping
587 *
588 * There is no need to get a lock here unless we should be updating
589 * atime. We ignore any locking errors since the only consequence is
590 * a missed atime update (which will just be deferred until later).
591 *
592 * Returns: 0
593 */
594
595static int gfs2_mmap(struct file *file, struct vm_area_struct *vma)
596{
597 struct gfs2_inode *ip = GFS2_I(file->f_mapping->host);
598
599 if (!(file->f_flags & O_NOATIME) &&
600 !IS_NOATIME(&ip->i_inode)) {
601 struct gfs2_holder i_gh;
602 int error;
603
604 error = gfs2_glock_nq_init(ip->i_gl, LM_ST_SHARED, LM_FLAG_ANY,
605 &i_gh);
606 if (error)
607 return error;
608 /* grab lock to update inode */
609 gfs2_glock_dq_uninit(&i_gh);
610 file_accessed(file);
611 }
612 vma->vm_ops = &gfs2_vm_ops;
613
614 return 0;
615}
616
617/**
618 * gfs2_open_common - This is common to open and atomic_open
619 * @inode: The inode being opened
620 * @file: The file being opened
621 *
622 * This maybe called under a glock or not depending upon how it has
623 * been called. We must always be called under a glock for regular
624 * files, however. For other file types, it does not matter whether
625 * we hold the glock or not.
626 *
627 * Returns: Error code or 0 for success
628 */
629
630int gfs2_open_common(struct inode *inode, struct file *file)
631{
632 struct gfs2_file *fp;
633 int ret;
634
635 if (S_ISREG(inode->i_mode)) {
636 ret = generic_file_open(inode, file);
637 if (ret)
638 return ret;
639 }
640
641 fp = kzalloc(sizeof(struct gfs2_file), GFP_NOFS);
642 if (!fp)
643 return -ENOMEM;
644
645 mutex_init(&fp->f_fl_mutex);
646
647 gfs2_assert_warn(GFS2_SB(inode), !file->private_data);
648 file->private_data = fp;
649 if (file->f_mode & FMODE_WRITE) {
650 ret = gfs2_qa_get(GFS2_I(inode));
651 if (ret)
652 goto fail;
653 }
654 return 0;
655
656fail:
657 kfree(file->private_data);
658 file->private_data = NULL;
659 return ret;
660}
661
662/**
663 * gfs2_open - open a file
664 * @inode: the inode to open
665 * @file: the struct file for this opening
666 *
667 * After atomic_open, this function is only used for opening files
668 * which are already cached. We must still get the glock for regular
669 * files to ensure that we have the file size uptodate for the large
670 * file check which is in the common code. That is only an issue for
671 * regular files though.
672 *
673 * Returns: errno
674 */
675
676static int gfs2_open(struct inode *inode, struct file *file)
677{
678 struct gfs2_inode *ip = GFS2_I(inode);
679 struct gfs2_holder i_gh;
680 int error;
681 bool need_unlock = false;
682
683 if (S_ISREG(ip->i_inode.i_mode)) {
684 error = gfs2_glock_nq_init(ip->i_gl, LM_ST_SHARED, LM_FLAG_ANY,
685 &i_gh);
686 if (error)
687 return error;
688 need_unlock = true;
689 }
690
691 error = gfs2_open_common(inode, file);
692
693 if (need_unlock)
694 gfs2_glock_dq_uninit(&i_gh);
695
696 return error;
697}
698
699/**
700 * gfs2_release - called to close a struct file
701 * @inode: the inode the struct file belongs to
702 * @file: the struct file being closed
703 *
704 * Returns: errno
705 */
706
707static int gfs2_release(struct inode *inode, struct file *file)
708{
709 struct gfs2_inode *ip = GFS2_I(inode);
710
711 kfree(file->private_data);
712 file->private_data = NULL;
713
714 if (gfs2_rs_active(&ip->i_res))
715 gfs2_rs_delete(ip, &inode->i_writecount);
716 if (file->f_mode & FMODE_WRITE)
717 gfs2_qa_put(ip);
718 return 0;
719}
720
721/**
722 * gfs2_fsync - sync the dirty data for a file (across the cluster)
723 * @file: the file that points to the dentry
724 * @start: the start position in the file to sync
725 * @end: the end position in the file to sync
726 * @datasync: set if we can ignore timestamp changes
727 *
728 * We split the data flushing here so that we don't wait for the data
729 * until after we've also sent the metadata to disk. Note that for
730 * data=ordered, we will write & wait for the data at the log flush
731 * stage anyway, so this is unlikely to make much of a difference
732 * except in the data=writeback case.
733 *
734 * If the fdatawrite fails due to any reason except -EIO, we will
735 * continue the remainder of the fsync, although we'll still report
736 * the error at the end. This is to match filemap_write_and_wait_range()
737 * behaviour.
738 *
739 * Returns: errno
740 */
741
742static int gfs2_fsync(struct file *file, loff_t start, loff_t end,
743 int datasync)
744{
745 struct address_space *mapping = file->f_mapping;
746 struct inode *inode = mapping->host;
747 int sync_state = inode->i_state & I_DIRTY;
748 struct gfs2_inode *ip = GFS2_I(inode);
749 int ret = 0, ret1 = 0;
750
751 if (mapping->nrpages) {
752 ret1 = filemap_fdatawrite_range(mapping, start, end);
753 if (ret1 == -EIO)
754 return ret1;
755 }
756
757 if (!gfs2_is_jdata(ip))
758 sync_state &= ~I_DIRTY_PAGES;
759 if (datasync)
760 sync_state &= ~I_DIRTY_SYNC;
761
762 if (sync_state) {
763 ret = sync_inode_metadata(inode, 1);
764 if (ret)
765 return ret;
766 if (gfs2_is_jdata(ip))
767 ret = file_write_and_wait(file);
768 if (ret)
769 return ret;
770 gfs2_ail_flush(ip->i_gl, 1);
771 }
772
773 if (mapping->nrpages)
774 ret = file_fdatawait_range(file, start, end);
775
776 return ret ? ret : ret1;
777}
778
779static ssize_t gfs2_file_direct_read(struct kiocb *iocb, struct iov_iter *to,
780 struct gfs2_holder *gh)
781{
782 struct file *file = iocb->ki_filp;
783 struct gfs2_inode *ip = GFS2_I(file->f_mapping->host);
784 size_t count = iov_iter_count(to);
785 ssize_t ret;
786
787 if (!count)
788 return 0; /* skip atime */
789
790 gfs2_holder_init(ip->i_gl, LM_ST_DEFERRED, 0, gh);
791 ret = gfs2_glock_nq(gh);
792 if (ret)
793 goto out_uninit;
794
795 ret = iomap_dio_rw(iocb, to, &gfs2_iomap_ops, NULL, 0);
796 gfs2_glock_dq(gh);
797out_uninit:
798 gfs2_holder_uninit(gh);
799 return ret;
800}
801
802static ssize_t gfs2_file_direct_write(struct kiocb *iocb, struct iov_iter *from,
803 struct gfs2_holder *gh)
804{
805 struct file *file = iocb->ki_filp;
806 struct inode *inode = file->f_mapping->host;
807 struct gfs2_inode *ip = GFS2_I(inode);
808 size_t len = iov_iter_count(from);
809 loff_t offset = iocb->ki_pos;
810 ssize_t ret;
811
812 /*
813 * Deferred lock, even if its a write, since we do no allocation on
814 * this path. All we need to change is the atime, and this lock mode
815 * ensures that other nodes have flushed their buffered read caches
816 * (i.e. their page cache entries for this inode). We do not,
817 * unfortunately, have the option of only flushing a range like the
818 * VFS does.
819 */
820 gfs2_holder_init(ip->i_gl, LM_ST_DEFERRED, 0, gh);
821 ret = gfs2_glock_nq(gh);
822 if (ret)
823 goto out_uninit;
824
825 /* Silently fall back to buffered I/O when writing beyond EOF */
826 if (offset + len > i_size_read(&ip->i_inode))
827 goto out;
828
829 ret = iomap_dio_rw(iocb, from, &gfs2_iomap_ops, NULL, 0);
830 if (ret == -ENOTBLK)
831 ret = 0;
832out:
833 gfs2_glock_dq(gh);
834out_uninit:
835 gfs2_holder_uninit(gh);
836 return ret;
837}
838
839static ssize_t gfs2_file_read_iter(struct kiocb *iocb, struct iov_iter *to)
840{
841 struct gfs2_inode *ip;
842 struct gfs2_holder gh;
843 size_t written = 0;
844 ssize_t ret;
845
846 if (iocb->ki_flags & IOCB_DIRECT) {
847 ret = gfs2_file_direct_read(iocb, to, &gh);
848 if (likely(ret != -ENOTBLK))
849 return ret;
850 iocb->ki_flags &= ~IOCB_DIRECT;
851 }
852 iocb->ki_flags |= IOCB_NOIO;
853 ret = generic_file_read_iter(iocb, to);
854 iocb->ki_flags &= ~IOCB_NOIO;
855 if (ret >= 0) {
856 if (!iov_iter_count(to))
857 return ret;
858 written = ret;
859 } else {
860 if (ret != -EAGAIN)
861 return ret;
862 if (iocb->ki_flags & IOCB_NOWAIT)
863 return ret;
864 }
865 ip = GFS2_I(iocb->ki_filp->f_mapping->host);
866 gfs2_holder_init(ip->i_gl, LM_ST_SHARED, 0, &gh);
867 ret = gfs2_glock_nq(&gh);
868 if (ret)
869 goto out_uninit;
870 ret = generic_file_read_iter(iocb, to);
871 if (ret > 0)
872 written += ret;
873 gfs2_glock_dq(&gh);
874out_uninit:
875 gfs2_holder_uninit(&gh);
876 return written ? written : ret;
877}
878
879/**
880 * gfs2_file_write_iter - Perform a write to a file
881 * @iocb: The io context
882 * @from: The data to write
883 *
884 * We have to do a lock/unlock here to refresh the inode size for
885 * O_APPEND writes, otherwise we can land up writing at the wrong
886 * offset. There is still a race, but provided the app is using its
887 * own file locking, this will make O_APPEND work as expected.
888 *
889 */
890
891static ssize_t gfs2_file_write_iter(struct kiocb *iocb, struct iov_iter *from)
892{
893 struct file *file = iocb->ki_filp;
894 struct inode *inode = file_inode(file);
895 struct gfs2_inode *ip = GFS2_I(inode);
896 struct gfs2_holder gh;
897 ssize_t ret;
898
899 gfs2_size_hint(file, iocb->ki_pos, iov_iter_count(from));
900
901 if (iocb->ki_flags & IOCB_APPEND) {
902 ret = gfs2_glock_nq_init(ip->i_gl, LM_ST_SHARED, 0, &gh);
903 if (ret)
904 return ret;
905 gfs2_glock_dq_uninit(&gh);
906 }
907
908 inode_lock(inode);
909 ret = generic_write_checks(iocb, from);
910 if (ret <= 0)
911 goto out_unlock;
912
913 ret = file_remove_privs(file);
914 if (ret)
915 goto out_unlock;
916
917 ret = file_update_time(file);
918 if (ret)
919 goto out_unlock;
920
921 if (iocb->ki_flags & IOCB_DIRECT) {
922 struct address_space *mapping = file->f_mapping;
923 ssize_t buffered, ret2;
924
925 ret = gfs2_file_direct_write(iocb, from, &gh);
926 if (ret < 0 || !iov_iter_count(from))
927 goto out_unlock;
928
929 iocb->ki_flags |= IOCB_DSYNC;
930 current->backing_dev_info = inode_to_bdi(inode);
931 buffered = iomap_file_buffered_write(iocb, from, &gfs2_iomap_ops);
932 current->backing_dev_info = NULL;
933 if (unlikely(buffered <= 0)) {
934 if (!ret)
935 ret = buffered;
936 goto out_unlock;
937 }
938
939 /*
940 * We need to ensure that the page cache pages are written to
941 * disk and invalidated to preserve the expected O_DIRECT
942 * semantics. If the writeback or invalidate fails, only report
943 * the direct I/O range as we don't know if the buffered pages
944 * made it to disk.
945 */
946 iocb->ki_pos += buffered;
947 ret2 = generic_write_sync(iocb, buffered);
948 invalidate_mapping_pages(mapping,
949 (iocb->ki_pos - buffered) >> PAGE_SHIFT,
950 (iocb->ki_pos - 1) >> PAGE_SHIFT);
951 if (!ret || ret2 > 0)
952 ret += ret2;
953 } else {
954 current->backing_dev_info = inode_to_bdi(inode);
955 ret = iomap_file_buffered_write(iocb, from, &gfs2_iomap_ops);
956 current->backing_dev_info = NULL;
957 if (likely(ret > 0)) {
958 iocb->ki_pos += ret;
959 ret = generic_write_sync(iocb, ret);
960 }
961 }
962
963out_unlock:
964 inode_unlock(inode);
965 return ret;
966}
967
968static int fallocate_chunk(struct inode *inode, loff_t offset, loff_t len,
969 int mode)
970{
971 struct super_block *sb = inode->i_sb;
972 struct gfs2_inode *ip = GFS2_I(inode);
973 loff_t end = offset + len;
974 struct buffer_head *dibh;
975 int error;
976
977 error = gfs2_meta_inode_buffer(ip, &dibh);
978 if (unlikely(error))
979 return error;
980
981 gfs2_trans_add_meta(ip->i_gl, dibh);
982
983 if (gfs2_is_stuffed(ip)) {
984 error = gfs2_unstuff_dinode(ip);
985 if (unlikely(error))
986 goto out;
987 }
988
989 while (offset < end) {
990 struct iomap iomap = { };
991
992 error = gfs2_iomap_alloc(inode, offset, end - offset, &iomap);
993 if (error)
994 goto out;
995 offset = iomap.offset + iomap.length;
996 if (!(iomap.flags & IOMAP_F_NEW))
997 continue;
998 error = sb_issue_zeroout(sb, iomap.addr >> inode->i_blkbits,
999 iomap.length >> inode->i_blkbits,
1000 GFP_NOFS);
1001 if (error) {
1002 fs_err(GFS2_SB(inode), "Failed to zero data buffers\n");
1003 goto out;
1004 }
1005 }
1006out:
1007 brelse(dibh);
1008 return error;
1009}
1010
1011/**
1012 * calc_max_reserv() - Reverse of write_calc_reserv. Given a number of
1013 * blocks, determine how many bytes can be written.
1014 * @ip: The inode in question.
1015 * @len: Max cap of bytes. What we return in *len must be <= this.
1016 * @data_blocks: Compute and return the number of data blocks needed
1017 * @ind_blocks: Compute and return the number of indirect blocks needed
1018 * @max_blocks: The total blocks available to work with.
1019 *
1020 * Returns: void, but @len, @data_blocks and @ind_blocks are filled in.
1021 */
1022static void calc_max_reserv(struct gfs2_inode *ip, loff_t *len,
1023 unsigned int *data_blocks, unsigned int *ind_blocks,
1024 unsigned int max_blocks)
1025{
1026 loff_t max = *len;
1027 const struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode);
1028 unsigned int tmp, max_data = max_blocks - 3 * (sdp->sd_max_height - 1);
1029
1030 for (tmp = max_data; tmp > sdp->sd_diptrs;) {
1031 tmp = DIV_ROUND_UP(tmp, sdp->sd_inptrs);
1032 max_data -= tmp;
1033 }
1034
1035 *data_blocks = max_data;
1036 *ind_blocks = max_blocks - max_data;
1037 *len = ((loff_t)max_data - 3) << sdp->sd_sb.sb_bsize_shift;
1038 if (*len > max) {
1039 *len = max;
1040 gfs2_write_calc_reserv(ip, max, data_blocks, ind_blocks);
1041 }
1042}
1043
1044static long __gfs2_fallocate(struct file *file, int mode, loff_t offset, loff_t len)
1045{
1046 struct inode *inode = file_inode(file);
1047 struct gfs2_sbd *sdp = GFS2_SB(inode);
1048 struct gfs2_inode *ip = GFS2_I(inode);
1049 struct gfs2_alloc_parms ap = { .aflags = 0, };
1050 unsigned int data_blocks = 0, ind_blocks = 0, rblocks;
1051 loff_t bytes, max_bytes, max_blks;
1052 int error;
1053 const loff_t pos = offset;
1054 const loff_t count = len;
1055 loff_t bsize_mask = ~((loff_t)sdp->sd_sb.sb_bsize - 1);
1056 loff_t next = (offset + len - 1) >> sdp->sd_sb.sb_bsize_shift;
1057 loff_t max_chunk_size = UINT_MAX & bsize_mask;
1058
1059 next = (next + 1) << sdp->sd_sb.sb_bsize_shift;
1060
1061 offset &= bsize_mask;
1062
1063 len = next - offset;
1064 bytes = sdp->sd_max_rg_data * sdp->sd_sb.sb_bsize / 2;
1065 if (!bytes)
1066 bytes = UINT_MAX;
1067 bytes &= bsize_mask;
1068 if (bytes == 0)
1069 bytes = sdp->sd_sb.sb_bsize;
1070
1071 gfs2_size_hint(file, offset, len);
1072
1073 gfs2_write_calc_reserv(ip, PAGE_SIZE, &data_blocks, &ind_blocks);
1074 ap.min_target = data_blocks + ind_blocks;
1075
1076 while (len > 0) {
1077 if (len < bytes)
1078 bytes = len;
1079 if (!gfs2_write_alloc_required(ip, offset, bytes)) {
1080 len -= bytes;
1081 offset += bytes;
1082 continue;
1083 }
1084
1085 /* We need to determine how many bytes we can actually
1086 * fallocate without exceeding quota or going over the
1087 * end of the fs. We start off optimistically by assuming
1088 * we can write max_bytes */
1089 max_bytes = (len > max_chunk_size) ? max_chunk_size : len;
1090
1091 /* Since max_bytes is most likely a theoretical max, we
1092 * calculate a more realistic 'bytes' to serve as a good
1093 * starting point for the number of bytes we may be able
1094 * to write */
1095 gfs2_write_calc_reserv(ip, bytes, &data_blocks, &ind_blocks);
1096 ap.target = data_blocks + ind_blocks;
1097
1098 error = gfs2_quota_lock_check(ip, &ap);
1099 if (error)
1100 return error;
1101 /* ap.allowed tells us how many blocks quota will allow
1102 * us to write. Check if this reduces max_blks */
1103 max_blks = UINT_MAX;
1104 if (ap.allowed)
1105 max_blks = ap.allowed;
1106
1107 error = gfs2_inplace_reserve(ip, &ap);
1108 if (error)
1109 goto out_qunlock;
1110
1111 /* check if the selected rgrp limits our max_blks further */
1112 if (ip->i_res.rs_reserved < max_blks)
1113 max_blks = ip->i_res.rs_reserved;
1114
1115 /* Almost done. Calculate bytes that can be written using
1116 * max_blks. We also recompute max_bytes, data_blocks and
1117 * ind_blocks */
1118 calc_max_reserv(ip, &max_bytes, &data_blocks,
1119 &ind_blocks, max_blks);
1120
1121 rblocks = RES_DINODE + ind_blocks + RES_STATFS + RES_QUOTA +
1122 RES_RG_HDR + gfs2_rg_blocks(ip, data_blocks + ind_blocks);
1123 if (gfs2_is_jdata(ip))
1124 rblocks += data_blocks ? data_blocks : 1;
1125
1126 error = gfs2_trans_begin(sdp, rblocks,
1127 PAGE_SIZE >> inode->i_blkbits);
1128 if (error)
1129 goto out_trans_fail;
1130
1131 error = fallocate_chunk(inode, offset, max_bytes, mode);
1132 gfs2_trans_end(sdp);
1133
1134 if (error)
1135 goto out_trans_fail;
1136
1137 len -= max_bytes;
1138 offset += max_bytes;
1139 gfs2_inplace_release(ip);
1140 gfs2_quota_unlock(ip);
1141 }
1142
1143 if (!(mode & FALLOC_FL_KEEP_SIZE) && (pos + count) > inode->i_size)
1144 i_size_write(inode, pos + count);
1145 file_update_time(file);
1146 mark_inode_dirty(inode);
1147
1148 if ((file->f_flags & O_DSYNC) || IS_SYNC(file->f_mapping->host))
1149 return vfs_fsync_range(file, pos, pos + count - 1,
1150 (file->f_flags & __O_SYNC) ? 0 : 1);
1151 return 0;
1152
1153out_trans_fail:
1154 gfs2_inplace_release(ip);
1155out_qunlock:
1156 gfs2_quota_unlock(ip);
1157 return error;
1158}
1159
1160static long gfs2_fallocate(struct file *file, int mode, loff_t offset, loff_t len)
1161{
1162 struct inode *inode = file_inode(file);
1163 struct gfs2_sbd *sdp = GFS2_SB(inode);
1164 struct gfs2_inode *ip = GFS2_I(inode);
1165 struct gfs2_holder gh;
1166 int ret;
1167
1168 if (mode & ~(FALLOC_FL_PUNCH_HOLE | FALLOC_FL_KEEP_SIZE))
1169 return -EOPNOTSUPP;
1170 /* fallocate is needed by gfs2_grow to reserve space in the rindex */
1171 if (gfs2_is_jdata(ip) && inode != sdp->sd_rindex)
1172 return -EOPNOTSUPP;
1173
1174 inode_lock(inode);
1175
1176 gfs2_holder_init(ip->i_gl, LM_ST_EXCLUSIVE, 0, &gh);
1177 ret = gfs2_glock_nq(&gh);
1178 if (ret)
1179 goto out_uninit;
1180
1181 if (!(mode & FALLOC_FL_KEEP_SIZE) &&
1182 (offset + len) > inode->i_size) {
1183 ret = inode_newsize_ok(inode, offset + len);
1184 if (ret)
1185 goto out_unlock;
1186 }
1187
1188 ret = get_write_access(inode);
1189 if (ret)
1190 goto out_unlock;
1191
1192 if (mode & FALLOC_FL_PUNCH_HOLE) {
1193 ret = __gfs2_punch_hole(file, offset, len);
1194 } else {
1195 ret = __gfs2_fallocate(file, mode, offset, len);
1196 if (ret)
1197 gfs2_rs_deltree(&ip->i_res);
1198 }
1199
1200 put_write_access(inode);
1201out_unlock:
1202 gfs2_glock_dq(&gh);
1203out_uninit:
1204 gfs2_holder_uninit(&gh);
1205 inode_unlock(inode);
1206 return ret;
1207}
1208
1209static ssize_t gfs2_file_splice_write(struct pipe_inode_info *pipe,
1210 struct file *out, loff_t *ppos,
1211 size_t len, unsigned int flags)
1212{
1213 ssize_t ret;
1214
1215 gfs2_size_hint(out, *ppos, len);
1216
1217 ret = iter_file_splice_write(pipe, out, ppos, len, flags);
1218 return ret;
1219}
1220
1221#ifdef CONFIG_GFS2_FS_LOCKING_DLM
1222
1223/**
1224 * gfs2_lock - acquire/release a posix lock on a file
1225 * @file: the file pointer
1226 * @cmd: either modify or retrieve lock state, possibly wait
1227 * @fl: type and range of lock
1228 *
1229 * Returns: errno
1230 */
1231
1232static int gfs2_lock(struct file *file, int cmd, struct file_lock *fl)
1233{
1234 struct gfs2_inode *ip = GFS2_I(file->f_mapping->host);
1235 struct gfs2_sbd *sdp = GFS2_SB(file->f_mapping->host);
1236 struct lm_lockstruct *ls = &sdp->sd_lockstruct;
1237
1238 if (!(fl->fl_flags & FL_POSIX))
1239 return -ENOLCK;
1240 if (__mandatory_lock(&ip->i_inode) && fl->fl_type != F_UNLCK)
1241 return -ENOLCK;
1242
1243 if (cmd == F_CANCELLK) {
1244 /* Hack: */
1245 cmd = F_SETLK;
1246 fl->fl_type = F_UNLCK;
1247 }
1248 if (unlikely(gfs2_withdrawn(sdp))) {
1249 if (fl->fl_type == F_UNLCK)
1250 locks_lock_file_wait(file, fl);
1251 return -EIO;
1252 }
1253 if (IS_GETLK(cmd))
1254 return dlm_posix_get(ls->ls_dlm, ip->i_no_addr, file, fl);
1255 else if (fl->fl_type == F_UNLCK)
1256 return dlm_posix_unlock(ls->ls_dlm, ip->i_no_addr, file, fl);
1257 else
1258 return dlm_posix_lock(ls->ls_dlm, ip->i_no_addr, file, cmd, fl);
1259}
1260
1261static int do_flock(struct file *file, int cmd, struct file_lock *fl)
1262{
1263 struct gfs2_file *fp = file->private_data;
1264 struct gfs2_holder *fl_gh = &fp->f_fl_gh;
1265 struct gfs2_inode *ip = GFS2_I(file_inode(file));
1266 struct gfs2_glock *gl;
1267 unsigned int state;
1268 u16 flags;
1269 int error = 0;
1270 int sleeptime;
1271
1272 state = (fl->fl_type == F_WRLCK) ? LM_ST_EXCLUSIVE : LM_ST_SHARED;
1273 flags = (IS_SETLKW(cmd) ? 0 : LM_FLAG_TRY_1CB) | GL_EXACT;
1274
1275 mutex_lock(&fp->f_fl_mutex);
1276
1277 if (gfs2_holder_initialized(fl_gh)) {
1278 struct file_lock request;
1279 if (fl_gh->gh_state == state)
1280 goto out;
1281 locks_init_lock(&request);
1282 request.fl_type = F_UNLCK;
1283 request.fl_flags = FL_FLOCK;
1284 locks_lock_file_wait(file, &request);
1285 gfs2_glock_dq(fl_gh);
1286 gfs2_holder_reinit(state, flags, fl_gh);
1287 } else {
1288 error = gfs2_glock_get(GFS2_SB(&ip->i_inode), ip->i_no_addr,
1289 &gfs2_flock_glops, CREATE, &gl);
1290 if (error)
1291 goto out;
1292 gfs2_holder_init(gl, state, flags, fl_gh);
1293 gfs2_glock_put(gl);
1294 }
1295 for (sleeptime = 1; sleeptime <= 4; sleeptime <<= 1) {
1296 error = gfs2_glock_nq(fl_gh);
1297 if (error != GLR_TRYFAILED)
1298 break;
1299 fl_gh->gh_flags = LM_FLAG_TRY | GL_EXACT;
1300 fl_gh->gh_error = 0;
1301 msleep(sleeptime);
1302 }
1303 if (error) {
1304 gfs2_holder_uninit(fl_gh);
1305 if (error == GLR_TRYFAILED)
1306 error = -EAGAIN;
1307 } else {
1308 error = locks_lock_file_wait(file, fl);
1309 gfs2_assert_warn(GFS2_SB(&ip->i_inode), !error);
1310 }
1311
1312out:
1313 mutex_unlock(&fp->f_fl_mutex);
1314 return error;
1315}
1316
1317static void do_unflock(struct file *file, struct file_lock *fl)
1318{
1319 struct gfs2_file *fp = file->private_data;
1320 struct gfs2_holder *fl_gh = &fp->f_fl_gh;
1321
1322 mutex_lock(&fp->f_fl_mutex);
1323 locks_lock_file_wait(file, fl);
1324 if (gfs2_holder_initialized(fl_gh)) {
1325 gfs2_glock_dq(fl_gh);
1326 gfs2_holder_uninit(fl_gh);
1327 }
1328 mutex_unlock(&fp->f_fl_mutex);
1329}
1330
1331/**
1332 * gfs2_flock - acquire/release a flock lock on a file
1333 * @file: the file pointer
1334 * @cmd: either modify or retrieve lock state, possibly wait
1335 * @fl: type and range of lock
1336 *
1337 * Returns: errno
1338 */
1339
1340static int gfs2_flock(struct file *file, int cmd, struct file_lock *fl)
1341{
1342 if (!(fl->fl_flags & FL_FLOCK))
1343 return -ENOLCK;
1344 if (fl->fl_type & LOCK_MAND)
1345 return -EOPNOTSUPP;
1346
1347 if (fl->fl_type == F_UNLCK) {
1348 do_unflock(file, fl);
1349 return 0;
1350 } else {
1351 return do_flock(file, cmd, fl);
1352 }
1353}
1354
1355const struct file_operations gfs2_file_fops = {
1356 .llseek = gfs2_llseek,
1357 .read_iter = gfs2_file_read_iter,
1358 .write_iter = gfs2_file_write_iter,
1359 .iopoll = iomap_dio_iopoll,
1360 .unlocked_ioctl = gfs2_ioctl,
1361 .compat_ioctl = gfs2_compat_ioctl,
1362 .mmap = gfs2_mmap,
1363 .open = gfs2_open,
1364 .release = gfs2_release,
1365 .fsync = gfs2_fsync,
1366 .lock = gfs2_lock,
1367 .flock = gfs2_flock,
1368 .splice_read = generic_file_splice_read,
1369 .splice_write = gfs2_file_splice_write,
1370 .setlease = simple_nosetlease,
1371 .fallocate = gfs2_fallocate,
1372};
1373
1374const struct file_operations gfs2_dir_fops = {
1375 .iterate_shared = gfs2_readdir,
1376 .unlocked_ioctl = gfs2_ioctl,
1377 .compat_ioctl = gfs2_compat_ioctl,
1378 .open = gfs2_open,
1379 .release = gfs2_release,
1380 .fsync = gfs2_fsync,
1381 .lock = gfs2_lock,
1382 .flock = gfs2_flock,
1383 .llseek = default_llseek,
1384};
1385
1386#endif /* CONFIG_GFS2_FS_LOCKING_DLM */
1387
1388const struct file_operations gfs2_file_fops_nolock = {
1389 .llseek = gfs2_llseek,
1390 .read_iter = gfs2_file_read_iter,
1391 .write_iter = gfs2_file_write_iter,
1392 .iopoll = iomap_dio_iopoll,
1393 .unlocked_ioctl = gfs2_ioctl,
1394 .compat_ioctl = gfs2_compat_ioctl,
1395 .mmap = gfs2_mmap,
1396 .open = gfs2_open,
1397 .release = gfs2_release,
1398 .fsync = gfs2_fsync,
1399 .splice_read = generic_file_splice_read,
1400 .splice_write = gfs2_file_splice_write,
1401 .setlease = generic_setlease,
1402 .fallocate = gfs2_fallocate,
1403};
1404
1405const struct file_operations gfs2_dir_fops_nolock = {
1406 .iterate_shared = gfs2_readdir,
1407 .unlocked_ioctl = gfs2_ioctl,
1408 .compat_ioctl = gfs2_compat_ioctl,
1409 .open = gfs2_open,
1410 .release = gfs2_release,
1411 .fsync = gfs2_fsync,
1412 .llseek = default_llseek,
1413};
1414