Loading...
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 *
217 */
218static int do_gfs2_set_flags(struct inode *inode, u32 reqflags, u32 mask)
219{
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;
226
227 error = gfs2_glock_nq_init(ip->i_gl, LM_ST_EXCLUSIVE, 0, &gh);
228 if (error)
229 return error;
230
231 error = 0;
232 flags = ip->i_diskflags;
233 new_flags = (flags & ~mask) | (reqflags & mask);
234 if ((new_flags ^ flags) == 0)
235 goto out;
236
237 if (!IS_IMMUTABLE(inode)) {
238 error = gfs2_permission(&init_user_ns, inode, MAY_WRITE);
239 if (error)
240 goto out;
241 }
242 if ((flags ^ new_flags) & GFS2_DIF_JDATA) {
243 if (new_flags & GFS2_DIF_JDATA)
244 gfs2_log_flush(sdp, ip->i_gl,
245 GFS2_LOG_HEAD_FLUSH_NORMAL |
246 GFS2_LFC_SET_FLAGS);
247 error = filemap_fdatawrite(inode->i_mapping);
248 if (error)
249 goto out;
250 error = filemap_fdatawait(inode->i_mapping);
251 if (error)
252 goto out;
253 if (new_flags & GFS2_DIF_JDATA)
254 gfs2_ordered_del_inode(ip);
255 }
256 error = gfs2_trans_begin(sdp, RES_DINODE, 0);
257 if (error)
258 goto out;
259 error = gfs2_meta_inode_buffer(ip, &bh);
260 if (error)
261 goto out_trans_end;
262 inode->i_ctime = current_time(inode);
263 gfs2_trans_add_meta(ip->i_gl, bh);
264 ip->i_diskflags = new_flags;
265 gfs2_dinode_out(ip, bh->b_data);
266 brelse(bh);
267 gfs2_set_inode_flags(inode);
268 gfs2_set_aops(inode);
269out_trans_end:
270 gfs2_trans_end(sdp);
271out:
272 gfs2_glock_dq_uninit(&gh);
273 return error;
274}
275
276int gfs2_fileattr_set(struct user_namespace *mnt_userns,
277 struct dentry *dentry, struct fileattr *fa)
278{
279 struct inode *inode = d_inode(dentry);
280 u32 fsflags = fa->flags, gfsflags = 0;
281 u32 mask;
282 int i;
283
284 if (d_is_special(dentry))
285 return -ENOTTY;
286
287 if (fileattr_has_fsx(fa))
288 return -EOPNOTSUPP;
289
290 for (i = 0; i < ARRAY_SIZE(fsflag_gfs2flag); i++) {
291 if (fsflags & fsflag_gfs2flag[i].fsflag) {
292 fsflags &= ~fsflag_gfs2flag[i].fsflag;
293 gfsflags |= fsflag_gfs2flag[i].gfsflag;
294 }
295 }
296 if (fsflags || gfsflags & ~GFS2_FLAGS_USER_SET)
297 return -EINVAL;
298
299 mask = GFS2_FLAGS_USER_SET;
300 if (S_ISDIR(inode->i_mode)) {
301 mask &= ~GFS2_DIF_JDATA;
302 } else {
303 /* The GFS2_DIF_TOPDIR flag is only valid for directories. */
304 if (gfsflags & GFS2_DIF_TOPDIR)
305 return -EINVAL;
306 mask &= ~(GFS2_DIF_TOPDIR | GFS2_DIF_INHERIT_JDATA);
307 }
308
309 return do_gfs2_set_flags(inode, gfsflags, mask);
310}
311
312static int gfs2_getlabel(struct file *filp, char __user *label)
313{
314 struct inode *inode = file_inode(filp);
315 struct gfs2_sbd *sdp = GFS2_SB(inode);
316
317 if (copy_to_user(label, sdp->sd_sb.sb_locktable, GFS2_LOCKNAME_LEN))
318 return -EFAULT;
319
320 return 0;
321}
322
323static long gfs2_ioctl(struct file *filp, unsigned int cmd, unsigned long arg)
324{
325 switch(cmd) {
326 case FITRIM:
327 return gfs2_fitrim(filp, (void __user *)arg);
328 case FS_IOC_GETFSLABEL:
329 return gfs2_getlabel(filp, (char __user *)arg);
330 }
331
332 return -ENOTTY;
333}
334
335#ifdef CONFIG_COMPAT
336static long gfs2_compat_ioctl(struct file *filp, unsigned int cmd, unsigned long arg)
337{
338 switch(cmd) {
339 /* Keep this list in sync with gfs2_ioctl */
340 case FITRIM:
341 case FS_IOC_GETFSLABEL:
342 break;
343 default:
344 return -ENOIOCTLCMD;
345 }
346
347 return gfs2_ioctl(filp, cmd, (unsigned long)compat_ptr(arg));
348}
349#else
350#define gfs2_compat_ioctl NULL
351#endif
352
353/**
354 * gfs2_size_hint - Give a hint to the size of a write request
355 * @filep: The struct file
356 * @offset: The file offset of the write
357 * @size: The length of the write
358 *
359 * When we are about to do a write, this function records the total
360 * write size in order to provide a suitable hint to the lower layers
361 * about how many blocks will be required.
362 *
363 */
364
365static void gfs2_size_hint(struct file *filep, loff_t offset, size_t size)
366{
367 struct inode *inode = file_inode(filep);
368 struct gfs2_sbd *sdp = GFS2_SB(inode);
369 struct gfs2_inode *ip = GFS2_I(inode);
370 size_t blks = (size + sdp->sd_sb.sb_bsize - 1) >> sdp->sd_sb.sb_bsize_shift;
371 int hint = min_t(size_t, INT_MAX, blks);
372
373 if (hint > atomic_read(&ip->i_sizehint))
374 atomic_set(&ip->i_sizehint, hint);
375}
376
377/**
378 * gfs2_allocate_page_backing - Allocate blocks for a write fault
379 * @page: The (locked) page to allocate backing for
380 * @length: Size of the allocation
381 *
382 * We try to allocate all the blocks required for the page in one go. This
383 * might fail for various reasons, so we keep trying until all the blocks to
384 * back this page are allocated. If some of the blocks are already allocated,
385 * that is ok too.
386 */
387static int gfs2_allocate_page_backing(struct page *page, unsigned int length)
388{
389 u64 pos = page_offset(page);
390
391 do {
392 struct iomap iomap = { };
393
394 if (gfs2_iomap_alloc(page->mapping->host, pos, length, &iomap))
395 return -EIO;
396
397 if (length < iomap.length)
398 iomap.length = length;
399 length -= iomap.length;
400 pos += iomap.length;
401 } while (length > 0);
402
403 return 0;
404}
405
406/**
407 * gfs2_page_mkwrite - Make a shared, mmap()ed, page writable
408 * @vmf: The virtual memory fault containing the page to become writable
409 *
410 * When the page becomes writable, we need to ensure that we have
411 * blocks allocated on disk to back that page.
412 */
413
414static vm_fault_t gfs2_page_mkwrite(struct vm_fault *vmf)
415{
416 struct page *page = vmf->page;
417 struct inode *inode = file_inode(vmf->vma->vm_file);
418 struct gfs2_inode *ip = GFS2_I(inode);
419 struct gfs2_sbd *sdp = GFS2_SB(inode);
420 struct gfs2_alloc_parms ap = { .aflags = 0, };
421 u64 offset = page_offset(page);
422 unsigned int data_blocks, ind_blocks, rblocks;
423 vm_fault_t ret = VM_FAULT_LOCKED;
424 struct gfs2_holder gh;
425 unsigned int length;
426 loff_t size;
427 int err;
428
429 sb_start_pagefault(inode->i_sb);
430
431 gfs2_holder_init(ip->i_gl, LM_ST_EXCLUSIVE, 0, &gh);
432 err = gfs2_glock_nq(&gh);
433 if (err) {
434 ret = block_page_mkwrite_return(err);
435 goto out_uninit;
436 }
437
438 /* Check page index against inode size */
439 size = i_size_read(inode);
440 if (offset >= size) {
441 ret = VM_FAULT_SIGBUS;
442 goto out_unlock;
443 }
444
445 /* Update file times before taking page lock */
446 file_update_time(vmf->vma->vm_file);
447
448 /* page is wholly or partially inside EOF */
449 if (size - offset < PAGE_SIZE)
450 length = size - offset;
451 else
452 length = PAGE_SIZE;
453
454 gfs2_size_hint(vmf->vma->vm_file, offset, length);
455
456 set_bit(GLF_DIRTY, &ip->i_gl->gl_flags);
457 set_bit(GIF_SW_PAGED, &ip->i_flags);
458
459 /*
460 * iomap_writepage / iomap_writepages currently don't support inline
461 * files, so always unstuff here.
462 */
463
464 if (!gfs2_is_stuffed(ip) &&
465 !gfs2_write_alloc_required(ip, offset, length)) {
466 lock_page(page);
467 if (!PageUptodate(page) || page->mapping != inode->i_mapping) {
468 ret = VM_FAULT_NOPAGE;
469 unlock_page(page);
470 }
471 goto out_unlock;
472 }
473
474 err = gfs2_rindex_update(sdp);
475 if (err) {
476 ret = block_page_mkwrite_return(err);
477 goto out_unlock;
478 }
479
480 gfs2_write_calc_reserv(ip, length, &data_blocks, &ind_blocks);
481 ap.target = data_blocks + ind_blocks;
482 err = gfs2_quota_lock_check(ip, &ap);
483 if (err) {
484 ret = block_page_mkwrite_return(err);
485 goto out_unlock;
486 }
487 err = gfs2_inplace_reserve(ip, &ap);
488 if (err) {
489 ret = block_page_mkwrite_return(err);
490 goto out_quota_unlock;
491 }
492
493 rblocks = RES_DINODE + ind_blocks;
494 if (gfs2_is_jdata(ip))
495 rblocks += data_blocks ? data_blocks : 1;
496 if (ind_blocks || data_blocks) {
497 rblocks += RES_STATFS + RES_QUOTA;
498 rblocks += gfs2_rg_blocks(ip, data_blocks + ind_blocks);
499 }
500 err = gfs2_trans_begin(sdp, rblocks, 0);
501 if (err) {
502 ret = block_page_mkwrite_return(err);
503 goto out_trans_fail;
504 }
505
506 /* Unstuff, if required, and allocate backing blocks for page */
507 if (gfs2_is_stuffed(ip)) {
508 err = gfs2_unstuff_dinode(ip);
509 if (err) {
510 ret = block_page_mkwrite_return(err);
511 goto out_trans_end;
512 }
513 }
514
515 lock_page(page);
516 /* If truncated, we must retry the operation, we may have raced
517 * with the glock demotion code.
518 */
519 if (!PageUptodate(page) || page->mapping != inode->i_mapping) {
520 ret = VM_FAULT_NOPAGE;
521 goto out_page_locked;
522 }
523
524 err = gfs2_allocate_page_backing(page, length);
525 if (err)
526 ret = block_page_mkwrite_return(err);
527
528out_page_locked:
529 if (ret != VM_FAULT_LOCKED)
530 unlock_page(page);
531out_trans_end:
532 gfs2_trans_end(sdp);
533out_trans_fail:
534 gfs2_inplace_release(ip);
535out_quota_unlock:
536 gfs2_quota_unlock(ip);
537out_unlock:
538 gfs2_glock_dq(&gh);
539out_uninit:
540 gfs2_holder_uninit(&gh);
541 if (ret == VM_FAULT_LOCKED) {
542 set_page_dirty(page);
543 wait_for_stable_page(page);
544 }
545 sb_end_pagefault(inode->i_sb);
546 return ret;
547}
548
549static vm_fault_t gfs2_fault(struct vm_fault *vmf)
550{
551 struct inode *inode = file_inode(vmf->vma->vm_file);
552 struct gfs2_inode *ip = GFS2_I(inode);
553 struct gfs2_holder gh;
554 vm_fault_t ret;
555 int err;
556
557 gfs2_holder_init(ip->i_gl, LM_ST_SHARED, 0, &gh);
558 err = gfs2_glock_nq(&gh);
559 if (err) {
560 ret = block_page_mkwrite_return(err);
561 goto out_uninit;
562 }
563 ret = filemap_fault(vmf);
564 gfs2_glock_dq(&gh);
565out_uninit:
566 gfs2_holder_uninit(&gh);
567 return ret;
568}
569
570static const struct vm_operations_struct gfs2_vm_ops = {
571 .fault = gfs2_fault,
572 .map_pages = filemap_map_pages,
573 .page_mkwrite = gfs2_page_mkwrite,
574};
575
576/**
577 * gfs2_mmap
578 * @file: The file to map
579 * @vma: The VMA which described the mapping
580 *
581 * There is no need to get a lock here unless we should be updating
582 * atime. We ignore any locking errors since the only consequence is
583 * a missed atime update (which will just be deferred until later).
584 *
585 * Returns: 0
586 */
587
588static int gfs2_mmap(struct file *file, struct vm_area_struct *vma)
589{
590 struct gfs2_inode *ip = GFS2_I(file->f_mapping->host);
591
592 if (!(file->f_flags & O_NOATIME) &&
593 !IS_NOATIME(&ip->i_inode)) {
594 struct gfs2_holder i_gh;
595 int error;
596
597 error = gfs2_glock_nq_init(ip->i_gl, LM_ST_SHARED, LM_FLAG_ANY,
598 &i_gh);
599 if (error)
600 return error;
601 /* grab lock to update inode */
602 gfs2_glock_dq_uninit(&i_gh);
603 file_accessed(file);
604 }
605 vma->vm_ops = &gfs2_vm_ops;
606
607 return 0;
608}
609
610/**
611 * gfs2_open_common - This is common to open and atomic_open
612 * @inode: The inode being opened
613 * @file: The file being opened
614 *
615 * This maybe called under a glock or not depending upon how it has
616 * been called. We must always be called under a glock for regular
617 * files, however. For other file types, it does not matter whether
618 * we hold the glock or not.
619 *
620 * Returns: Error code or 0 for success
621 */
622
623int gfs2_open_common(struct inode *inode, struct file *file)
624{
625 struct gfs2_file *fp;
626 int ret;
627
628 if (S_ISREG(inode->i_mode)) {
629 ret = generic_file_open(inode, file);
630 if (ret)
631 return ret;
632 }
633
634 fp = kzalloc(sizeof(struct gfs2_file), GFP_NOFS);
635 if (!fp)
636 return -ENOMEM;
637
638 mutex_init(&fp->f_fl_mutex);
639
640 gfs2_assert_warn(GFS2_SB(inode), !file->private_data);
641 file->private_data = fp;
642 if (file->f_mode & FMODE_WRITE) {
643 ret = gfs2_qa_get(GFS2_I(inode));
644 if (ret)
645 goto fail;
646 }
647 return 0;
648
649fail:
650 kfree(file->private_data);
651 file->private_data = NULL;
652 return ret;
653}
654
655/**
656 * gfs2_open - open a file
657 * @inode: the inode to open
658 * @file: the struct file for this opening
659 *
660 * After atomic_open, this function is only used for opening files
661 * which are already cached. We must still get the glock for regular
662 * files to ensure that we have the file size uptodate for the large
663 * file check which is in the common code. That is only an issue for
664 * regular files though.
665 *
666 * Returns: errno
667 */
668
669static int gfs2_open(struct inode *inode, struct file *file)
670{
671 struct gfs2_inode *ip = GFS2_I(inode);
672 struct gfs2_holder i_gh;
673 int error;
674 bool need_unlock = false;
675
676 if (S_ISREG(ip->i_inode.i_mode)) {
677 error = gfs2_glock_nq_init(ip->i_gl, LM_ST_SHARED, LM_FLAG_ANY,
678 &i_gh);
679 if (error)
680 return error;
681 need_unlock = true;
682 }
683
684 error = gfs2_open_common(inode, file);
685
686 if (need_unlock)
687 gfs2_glock_dq_uninit(&i_gh);
688
689 return error;
690}
691
692/**
693 * gfs2_release - called to close a struct file
694 * @inode: the inode the struct file belongs to
695 * @file: the struct file being closed
696 *
697 * Returns: errno
698 */
699
700static int gfs2_release(struct inode *inode, struct file *file)
701{
702 struct gfs2_inode *ip = GFS2_I(inode);
703
704 kfree(file->private_data);
705 file->private_data = NULL;
706
707 if (file->f_mode & FMODE_WRITE) {
708 if (gfs2_rs_active(&ip->i_res))
709 gfs2_rs_delete(ip);
710 gfs2_qa_put(ip);
711 }
712 return 0;
713}
714
715/**
716 * gfs2_fsync - sync the dirty data for a file (across the cluster)
717 * @file: the file that points to the dentry
718 * @start: the start position in the file to sync
719 * @end: the end position in the file to sync
720 * @datasync: set if we can ignore timestamp changes
721 *
722 * We split the data flushing here so that we don't wait for the data
723 * until after we've also sent the metadata to disk. Note that for
724 * data=ordered, we will write & wait for the data at the log flush
725 * stage anyway, so this is unlikely to make much of a difference
726 * except in the data=writeback case.
727 *
728 * If the fdatawrite fails due to any reason except -EIO, we will
729 * continue the remainder of the fsync, although we'll still report
730 * the error at the end. This is to match filemap_write_and_wait_range()
731 * behaviour.
732 *
733 * Returns: errno
734 */
735
736static int gfs2_fsync(struct file *file, loff_t start, loff_t end,
737 int datasync)
738{
739 struct address_space *mapping = file->f_mapping;
740 struct inode *inode = mapping->host;
741 int sync_state = inode->i_state & I_DIRTY;
742 struct gfs2_inode *ip = GFS2_I(inode);
743 int ret = 0, ret1 = 0;
744
745 if (mapping->nrpages) {
746 ret1 = filemap_fdatawrite_range(mapping, start, end);
747 if (ret1 == -EIO)
748 return ret1;
749 }
750
751 if (!gfs2_is_jdata(ip))
752 sync_state &= ~I_DIRTY_PAGES;
753 if (datasync)
754 sync_state &= ~I_DIRTY_SYNC;
755
756 if (sync_state) {
757 ret = sync_inode_metadata(inode, 1);
758 if (ret)
759 return ret;
760 if (gfs2_is_jdata(ip))
761 ret = file_write_and_wait(file);
762 if (ret)
763 return ret;
764 gfs2_ail_flush(ip->i_gl, 1);
765 }
766
767 if (mapping->nrpages)
768 ret = file_fdatawait_range(file, start, end);
769
770 return ret ? ret : ret1;
771}
772
773static inline bool should_fault_in_pages(struct iov_iter *i,
774 struct kiocb *iocb,
775 size_t *prev_count,
776 size_t *window_size)
777{
778 size_t count = iov_iter_count(i);
779 size_t size, offs;
780
781 if (!count)
782 return false;
783 if (!user_backed_iter(i))
784 return false;
785
786 size = PAGE_SIZE;
787 offs = offset_in_page(iocb->ki_pos);
788 if (*prev_count != count || !*window_size) {
789 size_t nr_dirtied;
790
791 nr_dirtied = max(current->nr_dirtied_pause -
792 current->nr_dirtied, 8);
793 size = min_t(size_t, SZ_1M, nr_dirtied << PAGE_SHIFT);
794 }
795
796 *prev_count = count;
797 *window_size = size - offs;
798 return true;
799}
800
801static ssize_t gfs2_file_direct_read(struct kiocb *iocb, struct iov_iter *to,
802 struct gfs2_holder *gh)
803{
804 struct file *file = iocb->ki_filp;
805 struct gfs2_inode *ip = GFS2_I(file->f_mapping->host);
806 size_t prev_count = 0, window_size = 0;
807 size_t read = 0;
808 ssize_t ret;
809
810 /*
811 * In this function, we disable page faults when we're holding the
812 * inode glock while doing I/O. If a page fault occurs, we indicate
813 * that the inode glock may be dropped, fault in the pages manually,
814 * and retry.
815 *
816 * Unlike generic_file_read_iter, for reads, iomap_dio_rw can trigger
817 * physical as well as manual page faults, and we need to disable both
818 * kinds.
819 *
820 * For direct I/O, gfs2 takes the inode glock in deferred mode. This
821 * locking mode is compatible with other deferred holders, so multiple
822 * processes and nodes can do direct I/O to a file at the same time.
823 * There's no guarantee that reads or writes will be atomic. Any
824 * coordination among readers and writers needs to happen externally.
825 */
826
827 if (!iov_iter_count(to))
828 return 0; /* skip atime */
829
830 gfs2_holder_init(ip->i_gl, LM_ST_DEFERRED, 0, gh);
831retry:
832 ret = gfs2_glock_nq(gh);
833 if (ret)
834 goto out_uninit;
835 pagefault_disable();
836 to->nofault = true;
837 ret = iomap_dio_rw(iocb, to, &gfs2_iomap_ops, NULL,
838 IOMAP_DIO_PARTIAL, NULL, read);
839 to->nofault = false;
840 pagefault_enable();
841 if (ret <= 0 && ret != -EFAULT)
842 goto out_unlock;
843 /* No increment (+=) because iomap_dio_rw returns a cumulative value. */
844 if (ret > 0)
845 read = ret;
846
847 if (should_fault_in_pages(to, iocb, &prev_count, &window_size)) {
848 gfs2_glock_dq(gh);
849 window_size -= fault_in_iov_iter_writeable(to, window_size);
850 if (window_size)
851 goto retry;
852 }
853out_unlock:
854 if (gfs2_holder_queued(gh))
855 gfs2_glock_dq(gh);
856out_uninit:
857 gfs2_holder_uninit(gh);
858 /* User space doesn't expect partial success. */
859 if (ret < 0)
860 return ret;
861 return read;
862}
863
864static ssize_t gfs2_file_direct_write(struct kiocb *iocb, struct iov_iter *from,
865 struct gfs2_holder *gh)
866{
867 struct file *file = iocb->ki_filp;
868 struct inode *inode = file->f_mapping->host;
869 struct gfs2_inode *ip = GFS2_I(inode);
870 size_t prev_count = 0, window_size = 0;
871 size_t written = 0;
872 ssize_t ret;
873
874 /*
875 * In this function, we disable page faults when we're holding the
876 * inode glock while doing I/O. If a page fault occurs, we indicate
877 * that the inode glock may be dropped, fault in the pages manually,
878 * and retry.
879 *
880 * For writes, iomap_dio_rw only triggers manual page faults, so we
881 * don't need to disable physical ones.
882 */
883
884 /*
885 * Deferred lock, even if its a write, since we do no allocation on
886 * this path. All we need to change is the atime, and this lock mode
887 * ensures that other nodes have flushed their buffered read caches
888 * (i.e. their page cache entries for this inode). We do not,
889 * unfortunately, have the option of only flushing a range like the
890 * VFS does.
891 */
892 gfs2_holder_init(ip->i_gl, LM_ST_DEFERRED, 0, gh);
893retry:
894 ret = gfs2_glock_nq(gh);
895 if (ret)
896 goto out_uninit;
897 /* Silently fall back to buffered I/O when writing beyond EOF */
898 if (iocb->ki_pos + iov_iter_count(from) > i_size_read(&ip->i_inode))
899 goto out_unlock;
900
901 from->nofault = true;
902 ret = iomap_dio_rw(iocb, from, &gfs2_iomap_ops, NULL,
903 IOMAP_DIO_PARTIAL, NULL, written);
904 from->nofault = false;
905 if (ret <= 0) {
906 if (ret == -ENOTBLK)
907 ret = 0;
908 if (ret != -EFAULT)
909 goto out_unlock;
910 }
911 /* No increment (+=) because iomap_dio_rw returns a cumulative value. */
912 if (ret > 0)
913 written = ret;
914
915 if (should_fault_in_pages(from, iocb, &prev_count, &window_size)) {
916 gfs2_glock_dq(gh);
917 window_size -= fault_in_iov_iter_readable(from, window_size);
918 if (window_size)
919 goto retry;
920 }
921out_unlock:
922 if (gfs2_holder_queued(gh))
923 gfs2_glock_dq(gh);
924out_uninit:
925 gfs2_holder_uninit(gh);
926 /* User space doesn't expect partial success. */
927 if (ret < 0)
928 return ret;
929 return written;
930}
931
932static ssize_t gfs2_file_read_iter(struct kiocb *iocb, struct iov_iter *to)
933{
934 struct gfs2_inode *ip;
935 struct gfs2_holder gh;
936 size_t prev_count = 0, window_size = 0;
937 size_t read = 0;
938 ssize_t ret;
939
940 /*
941 * In this function, we disable page faults when we're holding the
942 * inode glock while doing I/O. If a page fault occurs, we indicate
943 * that the inode glock may be dropped, fault in the pages manually,
944 * and retry.
945 */
946
947 if (iocb->ki_flags & IOCB_DIRECT)
948 return gfs2_file_direct_read(iocb, to, &gh);
949
950 pagefault_disable();
951 iocb->ki_flags |= IOCB_NOIO;
952 ret = generic_file_read_iter(iocb, to);
953 iocb->ki_flags &= ~IOCB_NOIO;
954 pagefault_enable();
955 if (ret >= 0) {
956 if (!iov_iter_count(to))
957 return ret;
958 read = ret;
959 } else if (ret != -EFAULT) {
960 if (ret != -EAGAIN)
961 return ret;
962 if (iocb->ki_flags & IOCB_NOWAIT)
963 return ret;
964 }
965 ip = GFS2_I(iocb->ki_filp->f_mapping->host);
966 gfs2_holder_init(ip->i_gl, LM_ST_SHARED, 0, &gh);
967retry:
968 ret = gfs2_glock_nq(&gh);
969 if (ret)
970 goto out_uninit;
971 pagefault_disable();
972 ret = generic_file_read_iter(iocb, to);
973 pagefault_enable();
974 if (ret <= 0 && ret != -EFAULT)
975 goto out_unlock;
976 if (ret > 0)
977 read += ret;
978
979 if (should_fault_in_pages(to, iocb, &prev_count, &window_size)) {
980 gfs2_glock_dq(&gh);
981 window_size -= fault_in_iov_iter_writeable(to, window_size);
982 if (window_size)
983 goto retry;
984 }
985out_unlock:
986 if (gfs2_holder_queued(&gh))
987 gfs2_glock_dq(&gh);
988out_uninit:
989 gfs2_holder_uninit(&gh);
990 return read ? read : ret;
991}
992
993static ssize_t gfs2_file_buffered_write(struct kiocb *iocb,
994 struct iov_iter *from,
995 struct gfs2_holder *gh)
996{
997 struct file *file = iocb->ki_filp;
998 struct inode *inode = file_inode(file);
999 struct gfs2_inode *ip = GFS2_I(inode);
1000 struct gfs2_sbd *sdp = GFS2_SB(inode);
1001 struct gfs2_holder *statfs_gh = NULL;
1002 size_t prev_count = 0, window_size = 0;
1003 size_t orig_count = iov_iter_count(from);
1004 size_t written = 0;
1005 ssize_t ret;
1006
1007 /*
1008 * In this function, we disable page faults when we're holding the
1009 * inode glock while doing I/O. If a page fault occurs, we indicate
1010 * that the inode glock may be dropped, fault in the pages manually,
1011 * and retry.
1012 */
1013
1014 if (inode == sdp->sd_rindex) {
1015 statfs_gh = kmalloc(sizeof(*statfs_gh), GFP_NOFS);
1016 if (!statfs_gh)
1017 return -ENOMEM;
1018 }
1019
1020 gfs2_holder_init(ip->i_gl, LM_ST_EXCLUSIVE, 0, gh);
1021retry:
1022 if (should_fault_in_pages(from, iocb, &prev_count, &window_size)) {
1023 window_size -= fault_in_iov_iter_readable(from, window_size);
1024 if (!window_size) {
1025 ret = -EFAULT;
1026 goto out_uninit;
1027 }
1028 from->count = min(from->count, window_size);
1029 }
1030 ret = gfs2_glock_nq(gh);
1031 if (ret)
1032 goto out_uninit;
1033
1034 if (inode == sdp->sd_rindex) {
1035 struct gfs2_inode *m_ip = GFS2_I(sdp->sd_statfs_inode);
1036
1037 ret = gfs2_glock_nq_init(m_ip->i_gl, LM_ST_EXCLUSIVE,
1038 GL_NOCACHE, statfs_gh);
1039 if (ret)
1040 goto out_unlock;
1041 }
1042
1043 current->backing_dev_info = inode_to_bdi(inode);
1044 pagefault_disable();
1045 ret = iomap_file_buffered_write(iocb, from, &gfs2_iomap_ops);
1046 pagefault_enable();
1047 current->backing_dev_info = NULL;
1048 if (ret > 0) {
1049 iocb->ki_pos += ret;
1050 written += ret;
1051 }
1052
1053 if (inode == sdp->sd_rindex)
1054 gfs2_glock_dq_uninit(statfs_gh);
1055
1056 if (ret <= 0 && ret != -EFAULT)
1057 goto out_unlock;
1058
1059 from->count = orig_count - written;
1060 if (should_fault_in_pages(from, iocb, &prev_count, &window_size)) {
1061 gfs2_glock_dq(gh);
1062 goto retry;
1063 }
1064out_unlock:
1065 if (gfs2_holder_queued(gh))
1066 gfs2_glock_dq(gh);
1067out_uninit:
1068 gfs2_holder_uninit(gh);
1069 kfree(statfs_gh);
1070 from->count = orig_count - written;
1071 return written ? written : ret;
1072}
1073
1074/**
1075 * gfs2_file_write_iter - Perform a write to a file
1076 * @iocb: The io context
1077 * @from: The data to write
1078 *
1079 * We have to do a lock/unlock here to refresh the inode size for
1080 * O_APPEND writes, otherwise we can land up writing at the wrong
1081 * offset. There is still a race, but provided the app is using its
1082 * own file locking, this will make O_APPEND work as expected.
1083 *
1084 */
1085
1086static ssize_t gfs2_file_write_iter(struct kiocb *iocb, struct iov_iter *from)
1087{
1088 struct file *file = iocb->ki_filp;
1089 struct inode *inode = file_inode(file);
1090 struct gfs2_inode *ip = GFS2_I(inode);
1091 struct gfs2_holder gh;
1092 ssize_t ret;
1093
1094 gfs2_size_hint(file, iocb->ki_pos, iov_iter_count(from));
1095
1096 if (iocb->ki_flags & IOCB_APPEND) {
1097 ret = gfs2_glock_nq_init(ip->i_gl, LM_ST_SHARED, 0, &gh);
1098 if (ret)
1099 return ret;
1100 gfs2_glock_dq_uninit(&gh);
1101 }
1102
1103 inode_lock(inode);
1104 ret = generic_write_checks(iocb, from);
1105 if (ret <= 0)
1106 goto out_unlock;
1107
1108 ret = file_remove_privs(file);
1109 if (ret)
1110 goto out_unlock;
1111
1112 ret = file_update_time(file);
1113 if (ret)
1114 goto out_unlock;
1115
1116 if (iocb->ki_flags & IOCB_DIRECT) {
1117 struct address_space *mapping = file->f_mapping;
1118 ssize_t buffered, ret2;
1119
1120 ret = gfs2_file_direct_write(iocb, from, &gh);
1121 if (ret < 0 || !iov_iter_count(from))
1122 goto out_unlock;
1123
1124 iocb->ki_flags |= IOCB_DSYNC;
1125 buffered = gfs2_file_buffered_write(iocb, from, &gh);
1126 if (unlikely(buffered <= 0)) {
1127 if (!ret)
1128 ret = buffered;
1129 goto out_unlock;
1130 }
1131
1132 /*
1133 * We need to ensure that the page cache pages are written to
1134 * disk and invalidated to preserve the expected O_DIRECT
1135 * semantics. If the writeback or invalidate fails, only report
1136 * the direct I/O range as we don't know if the buffered pages
1137 * made it to disk.
1138 */
1139 ret2 = generic_write_sync(iocb, buffered);
1140 invalidate_mapping_pages(mapping,
1141 (iocb->ki_pos - buffered) >> PAGE_SHIFT,
1142 (iocb->ki_pos - 1) >> PAGE_SHIFT);
1143 if (!ret || ret2 > 0)
1144 ret += ret2;
1145 } else {
1146 ret = gfs2_file_buffered_write(iocb, from, &gh);
1147 if (likely(ret > 0))
1148 ret = generic_write_sync(iocb, ret);
1149 }
1150
1151out_unlock:
1152 inode_unlock(inode);
1153 return ret;
1154}
1155
1156static int fallocate_chunk(struct inode *inode, loff_t offset, loff_t len,
1157 int mode)
1158{
1159 struct super_block *sb = inode->i_sb;
1160 struct gfs2_inode *ip = GFS2_I(inode);
1161 loff_t end = offset + len;
1162 struct buffer_head *dibh;
1163 int error;
1164
1165 error = gfs2_meta_inode_buffer(ip, &dibh);
1166 if (unlikely(error))
1167 return error;
1168
1169 gfs2_trans_add_meta(ip->i_gl, dibh);
1170
1171 if (gfs2_is_stuffed(ip)) {
1172 error = gfs2_unstuff_dinode(ip);
1173 if (unlikely(error))
1174 goto out;
1175 }
1176
1177 while (offset < end) {
1178 struct iomap iomap = { };
1179
1180 error = gfs2_iomap_alloc(inode, offset, end - offset, &iomap);
1181 if (error)
1182 goto out;
1183 offset = iomap.offset + iomap.length;
1184 if (!(iomap.flags & IOMAP_F_NEW))
1185 continue;
1186 error = sb_issue_zeroout(sb, iomap.addr >> inode->i_blkbits,
1187 iomap.length >> inode->i_blkbits,
1188 GFP_NOFS);
1189 if (error) {
1190 fs_err(GFS2_SB(inode), "Failed to zero data buffers\n");
1191 goto out;
1192 }
1193 }
1194out:
1195 brelse(dibh);
1196 return error;
1197}
1198
1199/**
1200 * calc_max_reserv() - Reverse of write_calc_reserv. Given a number of
1201 * blocks, determine how many bytes can be written.
1202 * @ip: The inode in question.
1203 * @len: Max cap of bytes. What we return in *len must be <= this.
1204 * @data_blocks: Compute and return the number of data blocks needed
1205 * @ind_blocks: Compute and return the number of indirect blocks needed
1206 * @max_blocks: The total blocks available to work with.
1207 *
1208 * Returns: void, but @len, @data_blocks and @ind_blocks are filled in.
1209 */
1210static void calc_max_reserv(struct gfs2_inode *ip, loff_t *len,
1211 unsigned int *data_blocks, unsigned int *ind_blocks,
1212 unsigned int max_blocks)
1213{
1214 loff_t max = *len;
1215 const struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode);
1216 unsigned int tmp, max_data = max_blocks - 3 * (sdp->sd_max_height - 1);
1217
1218 for (tmp = max_data; tmp > sdp->sd_diptrs;) {
1219 tmp = DIV_ROUND_UP(tmp, sdp->sd_inptrs);
1220 max_data -= tmp;
1221 }
1222
1223 *data_blocks = max_data;
1224 *ind_blocks = max_blocks - max_data;
1225 *len = ((loff_t)max_data - 3) << sdp->sd_sb.sb_bsize_shift;
1226 if (*len > max) {
1227 *len = max;
1228 gfs2_write_calc_reserv(ip, max, data_blocks, ind_blocks);
1229 }
1230}
1231
1232static long __gfs2_fallocate(struct file *file, int mode, loff_t offset, loff_t len)
1233{
1234 struct inode *inode = file_inode(file);
1235 struct gfs2_sbd *sdp = GFS2_SB(inode);
1236 struct gfs2_inode *ip = GFS2_I(inode);
1237 struct gfs2_alloc_parms ap = { .aflags = 0, };
1238 unsigned int data_blocks = 0, ind_blocks = 0, rblocks;
1239 loff_t bytes, max_bytes, max_blks;
1240 int error;
1241 const loff_t pos = offset;
1242 const loff_t count = len;
1243 loff_t bsize_mask = ~((loff_t)sdp->sd_sb.sb_bsize - 1);
1244 loff_t next = (offset + len - 1) >> sdp->sd_sb.sb_bsize_shift;
1245 loff_t max_chunk_size = UINT_MAX & bsize_mask;
1246
1247 next = (next + 1) << sdp->sd_sb.sb_bsize_shift;
1248
1249 offset &= bsize_mask;
1250
1251 len = next - offset;
1252 bytes = sdp->sd_max_rg_data * sdp->sd_sb.sb_bsize / 2;
1253 if (!bytes)
1254 bytes = UINT_MAX;
1255 bytes &= bsize_mask;
1256 if (bytes == 0)
1257 bytes = sdp->sd_sb.sb_bsize;
1258
1259 gfs2_size_hint(file, offset, len);
1260
1261 gfs2_write_calc_reserv(ip, PAGE_SIZE, &data_blocks, &ind_blocks);
1262 ap.min_target = data_blocks + ind_blocks;
1263
1264 while (len > 0) {
1265 if (len < bytes)
1266 bytes = len;
1267 if (!gfs2_write_alloc_required(ip, offset, bytes)) {
1268 len -= bytes;
1269 offset += bytes;
1270 continue;
1271 }
1272
1273 /* We need to determine how many bytes we can actually
1274 * fallocate without exceeding quota or going over the
1275 * end of the fs. We start off optimistically by assuming
1276 * we can write max_bytes */
1277 max_bytes = (len > max_chunk_size) ? max_chunk_size : len;
1278
1279 /* Since max_bytes is most likely a theoretical max, we
1280 * calculate a more realistic 'bytes' to serve as a good
1281 * starting point for the number of bytes we may be able
1282 * to write */
1283 gfs2_write_calc_reserv(ip, bytes, &data_blocks, &ind_blocks);
1284 ap.target = data_blocks + ind_blocks;
1285
1286 error = gfs2_quota_lock_check(ip, &ap);
1287 if (error)
1288 return error;
1289 /* ap.allowed tells us how many blocks quota will allow
1290 * us to write. Check if this reduces max_blks */
1291 max_blks = UINT_MAX;
1292 if (ap.allowed)
1293 max_blks = ap.allowed;
1294
1295 error = gfs2_inplace_reserve(ip, &ap);
1296 if (error)
1297 goto out_qunlock;
1298
1299 /* check if the selected rgrp limits our max_blks further */
1300 if (ip->i_res.rs_reserved < max_blks)
1301 max_blks = ip->i_res.rs_reserved;
1302
1303 /* Almost done. Calculate bytes that can be written using
1304 * max_blks. We also recompute max_bytes, data_blocks and
1305 * ind_blocks */
1306 calc_max_reserv(ip, &max_bytes, &data_blocks,
1307 &ind_blocks, max_blks);
1308
1309 rblocks = RES_DINODE + ind_blocks + RES_STATFS + RES_QUOTA +
1310 RES_RG_HDR + gfs2_rg_blocks(ip, data_blocks + ind_blocks);
1311 if (gfs2_is_jdata(ip))
1312 rblocks += data_blocks ? data_blocks : 1;
1313
1314 error = gfs2_trans_begin(sdp, rblocks,
1315 PAGE_SIZE >> inode->i_blkbits);
1316 if (error)
1317 goto out_trans_fail;
1318
1319 error = fallocate_chunk(inode, offset, max_bytes, mode);
1320 gfs2_trans_end(sdp);
1321
1322 if (error)
1323 goto out_trans_fail;
1324
1325 len -= max_bytes;
1326 offset += max_bytes;
1327 gfs2_inplace_release(ip);
1328 gfs2_quota_unlock(ip);
1329 }
1330
1331 if (!(mode & FALLOC_FL_KEEP_SIZE) && (pos + count) > inode->i_size)
1332 i_size_write(inode, pos + count);
1333 file_update_time(file);
1334 mark_inode_dirty(inode);
1335
1336 if ((file->f_flags & O_DSYNC) || IS_SYNC(file->f_mapping->host))
1337 return vfs_fsync_range(file, pos, pos + count - 1,
1338 (file->f_flags & __O_SYNC) ? 0 : 1);
1339 return 0;
1340
1341out_trans_fail:
1342 gfs2_inplace_release(ip);
1343out_qunlock:
1344 gfs2_quota_unlock(ip);
1345 return error;
1346}
1347
1348static long gfs2_fallocate(struct file *file, int mode, loff_t offset, loff_t len)
1349{
1350 struct inode *inode = file_inode(file);
1351 struct gfs2_sbd *sdp = GFS2_SB(inode);
1352 struct gfs2_inode *ip = GFS2_I(inode);
1353 struct gfs2_holder gh;
1354 int ret;
1355
1356 if (mode & ~(FALLOC_FL_PUNCH_HOLE | FALLOC_FL_KEEP_SIZE))
1357 return -EOPNOTSUPP;
1358 /* fallocate is needed by gfs2_grow to reserve space in the rindex */
1359 if (gfs2_is_jdata(ip) && inode != sdp->sd_rindex)
1360 return -EOPNOTSUPP;
1361
1362 inode_lock(inode);
1363
1364 gfs2_holder_init(ip->i_gl, LM_ST_EXCLUSIVE, 0, &gh);
1365 ret = gfs2_glock_nq(&gh);
1366 if (ret)
1367 goto out_uninit;
1368
1369 if (!(mode & FALLOC_FL_KEEP_SIZE) &&
1370 (offset + len) > inode->i_size) {
1371 ret = inode_newsize_ok(inode, offset + len);
1372 if (ret)
1373 goto out_unlock;
1374 }
1375
1376 ret = get_write_access(inode);
1377 if (ret)
1378 goto out_unlock;
1379
1380 if (mode & FALLOC_FL_PUNCH_HOLE) {
1381 ret = __gfs2_punch_hole(file, offset, len);
1382 } else {
1383 ret = __gfs2_fallocate(file, mode, offset, len);
1384 if (ret)
1385 gfs2_rs_deltree(&ip->i_res);
1386 }
1387
1388 put_write_access(inode);
1389out_unlock:
1390 gfs2_glock_dq(&gh);
1391out_uninit:
1392 gfs2_holder_uninit(&gh);
1393 inode_unlock(inode);
1394 return ret;
1395}
1396
1397static ssize_t gfs2_file_splice_write(struct pipe_inode_info *pipe,
1398 struct file *out, loff_t *ppos,
1399 size_t len, unsigned int flags)
1400{
1401 ssize_t ret;
1402
1403 gfs2_size_hint(out, *ppos, len);
1404
1405 ret = iter_file_splice_write(pipe, out, ppos, len, flags);
1406 return ret;
1407}
1408
1409#ifdef CONFIG_GFS2_FS_LOCKING_DLM
1410
1411/**
1412 * gfs2_lock - acquire/release a posix lock on a file
1413 * @file: the file pointer
1414 * @cmd: either modify or retrieve lock state, possibly wait
1415 * @fl: type and range of lock
1416 *
1417 * Returns: errno
1418 */
1419
1420static int gfs2_lock(struct file *file, int cmd, struct file_lock *fl)
1421{
1422 struct gfs2_inode *ip = GFS2_I(file->f_mapping->host);
1423 struct gfs2_sbd *sdp = GFS2_SB(file->f_mapping->host);
1424 struct lm_lockstruct *ls = &sdp->sd_lockstruct;
1425
1426 if (!(fl->fl_flags & FL_POSIX))
1427 return -ENOLCK;
1428 if (cmd == F_CANCELLK) {
1429 /* Hack: */
1430 cmd = F_SETLK;
1431 fl->fl_type = F_UNLCK;
1432 }
1433 if (unlikely(gfs2_withdrawn(sdp))) {
1434 if (fl->fl_type == F_UNLCK)
1435 locks_lock_file_wait(file, fl);
1436 return -EIO;
1437 }
1438 if (IS_GETLK(cmd))
1439 return dlm_posix_get(ls->ls_dlm, ip->i_no_addr, file, fl);
1440 else if (fl->fl_type == F_UNLCK)
1441 return dlm_posix_unlock(ls->ls_dlm, ip->i_no_addr, file, fl);
1442 else
1443 return dlm_posix_lock(ls->ls_dlm, ip->i_no_addr, file, cmd, fl);
1444}
1445
1446static void __flock_holder_uninit(struct file *file, struct gfs2_holder *fl_gh)
1447{
1448 struct gfs2_glock *gl = gfs2_glock_hold(fl_gh->gh_gl);
1449
1450 /*
1451 * Make sure gfs2_glock_put() won't sleep under the file->f_lock
1452 * spinlock.
1453 */
1454
1455 spin_lock(&file->f_lock);
1456 gfs2_holder_uninit(fl_gh);
1457 spin_unlock(&file->f_lock);
1458 gfs2_glock_put(gl);
1459}
1460
1461static int do_flock(struct file *file, int cmd, struct file_lock *fl)
1462{
1463 struct gfs2_file *fp = file->private_data;
1464 struct gfs2_holder *fl_gh = &fp->f_fl_gh;
1465 struct gfs2_inode *ip = GFS2_I(file_inode(file));
1466 struct gfs2_glock *gl;
1467 unsigned int state;
1468 u16 flags;
1469 int error = 0;
1470 int sleeptime;
1471
1472 state = (fl->fl_type == F_WRLCK) ? LM_ST_EXCLUSIVE : LM_ST_SHARED;
1473 flags = GL_EXACT | GL_NOPID;
1474 if (!IS_SETLKW(cmd))
1475 flags |= LM_FLAG_TRY_1CB;
1476
1477 mutex_lock(&fp->f_fl_mutex);
1478
1479 if (gfs2_holder_initialized(fl_gh)) {
1480 struct file_lock request;
1481 if (fl_gh->gh_state == state)
1482 goto out;
1483 locks_init_lock(&request);
1484 request.fl_type = F_UNLCK;
1485 request.fl_flags = FL_FLOCK;
1486 locks_lock_file_wait(file, &request);
1487 gfs2_glock_dq(fl_gh);
1488 gfs2_holder_reinit(state, flags, fl_gh);
1489 } else {
1490 error = gfs2_glock_get(GFS2_SB(&ip->i_inode), ip->i_no_addr,
1491 &gfs2_flock_glops, CREATE, &gl);
1492 if (error)
1493 goto out;
1494 spin_lock(&file->f_lock);
1495 gfs2_holder_init(gl, state, flags, fl_gh);
1496 spin_unlock(&file->f_lock);
1497 gfs2_glock_put(gl);
1498 }
1499 for (sleeptime = 1; sleeptime <= 4; sleeptime <<= 1) {
1500 error = gfs2_glock_nq(fl_gh);
1501 if (error != GLR_TRYFAILED)
1502 break;
1503 fl_gh->gh_flags &= ~LM_FLAG_TRY_1CB;
1504 fl_gh->gh_flags |= LM_FLAG_TRY;
1505 msleep(sleeptime);
1506 }
1507 if (error) {
1508 __flock_holder_uninit(file, fl_gh);
1509 if (error == GLR_TRYFAILED)
1510 error = -EAGAIN;
1511 } else {
1512 error = locks_lock_file_wait(file, fl);
1513 gfs2_assert_warn(GFS2_SB(&ip->i_inode), !error);
1514 }
1515
1516out:
1517 mutex_unlock(&fp->f_fl_mutex);
1518 return error;
1519}
1520
1521static void do_unflock(struct file *file, struct file_lock *fl)
1522{
1523 struct gfs2_file *fp = file->private_data;
1524 struct gfs2_holder *fl_gh = &fp->f_fl_gh;
1525
1526 mutex_lock(&fp->f_fl_mutex);
1527 locks_lock_file_wait(file, fl);
1528 if (gfs2_holder_initialized(fl_gh)) {
1529 gfs2_glock_dq(fl_gh);
1530 __flock_holder_uninit(file, fl_gh);
1531 }
1532 mutex_unlock(&fp->f_fl_mutex);
1533}
1534
1535/**
1536 * gfs2_flock - acquire/release a flock lock on a file
1537 * @file: the file pointer
1538 * @cmd: either modify or retrieve lock state, possibly wait
1539 * @fl: type and range of lock
1540 *
1541 * Returns: errno
1542 */
1543
1544static int gfs2_flock(struct file *file, int cmd, struct file_lock *fl)
1545{
1546 if (!(fl->fl_flags & FL_FLOCK))
1547 return -ENOLCK;
1548
1549 if (fl->fl_type == F_UNLCK) {
1550 do_unflock(file, fl);
1551 return 0;
1552 } else {
1553 return do_flock(file, cmd, fl);
1554 }
1555}
1556
1557const struct file_operations gfs2_file_fops = {
1558 .llseek = gfs2_llseek,
1559 .read_iter = gfs2_file_read_iter,
1560 .write_iter = gfs2_file_write_iter,
1561 .iopoll = iocb_bio_iopoll,
1562 .unlocked_ioctl = gfs2_ioctl,
1563 .compat_ioctl = gfs2_compat_ioctl,
1564 .mmap = gfs2_mmap,
1565 .open = gfs2_open,
1566 .release = gfs2_release,
1567 .fsync = gfs2_fsync,
1568 .lock = gfs2_lock,
1569 .flock = gfs2_flock,
1570 .splice_read = generic_file_splice_read,
1571 .splice_write = gfs2_file_splice_write,
1572 .setlease = simple_nosetlease,
1573 .fallocate = gfs2_fallocate,
1574};
1575
1576const struct file_operations gfs2_dir_fops = {
1577 .iterate_shared = gfs2_readdir,
1578 .unlocked_ioctl = gfs2_ioctl,
1579 .compat_ioctl = gfs2_compat_ioctl,
1580 .open = gfs2_open,
1581 .release = gfs2_release,
1582 .fsync = gfs2_fsync,
1583 .lock = gfs2_lock,
1584 .flock = gfs2_flock,
1585 .llseek = default_llseek,
1586};
1587
1588#endif /* CONFIG_GFS2_FS_LOCKING_DLM */
1589
1590const struct file_operations gfs2_file_fops_nolock = {
1591 .llseek = gfs2_llseek,
1592 .read_iter = gfs2_file_read_iter,
1593 .write_iter = gfs2_file_write_iter,
1594 .iopoll = iocb_bio_iopoll,
1595 .unlocked_ioctl = gfs2_ioctl,
1596 .compat_ioctl = gfs2_compat_ioctl,
1597 .mmap = gfs2_mmap,
1598 .open = gfs2_open,
1599 .release = gfs2_release,
1600 .fsync = gfs2_fsync,
1601 .splice_read = generic_file_splice_read,
1602 .splice_write = gfs2_file_splice_write,
1603 .setlease = generic_setlease,
1604 .fallocate = gfs2_fallocate,
1605};
1606
1607const struct file_operations gfs2_dir_fops_nolock = {
1608 .iterate_shared = gfs2_readdir,
1609 .unlocked_ioctl = gfs2_ioctl,
1610 .compat_ioctl = gfs2_compat_ioctl,
1611 .open = gfs2_open,
1612 .release = gfs2_release,
1613 .fsync = gfs2_fsync,
1614 .llseek = default_llseek,
1615};
1616
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/filelock.h>
19#include <linux/gfs2_ondisk.h>
20#include <linux/falloc.h>
21#include <linux/swap.h>
22#include <linux/crc32.h>
23#include <linux/writeback.h>
24#include <linux/uaccess.h>
25#include <linux/dlm.h>
26#include <linux/dlm_plock.h>
27#include <linux/delay.h>
28#include <linux/backing-dev.h>
29#include <linux/fileattr.h>
30
31#include "gfs2.h"
32#include "incore.h"
33#include "bmap.h"
34#include "aops.h"
35#include "dir.h"
36#include "glock.h"
37#include "glops.h"
38#include "inode.h"
39#include "log.h"
40#include "meta_io.h"
41#include "quota.h"
42#include "rgrp.h"
43#include "trans.h"
44#include "util.h"
45
46/**
47 * gfs2_llseek - seek to a location in a file
48 * @file: the file
49 * @offset: the offset
50 * @whence: Where to seek from (SEEK_SET, SEEK_CUR, or SEEK_END)
51 *
52 * SEEK_END requires the glock for the file because it references the
53 * file's size.
54 *
55 * Returns: The new offset, or errno
56 */
57
58static loff_t gfs2_llseek(struct file *file, loff_t offset, int whence)
59{
60 struct gfs2_inode *ip = GFS2_I(file->f_mapping->host);
61 struct gfs2_holder i_gh;
62 loff_t error;
63
64 switch (whence) {
65 case SEEK_END:
66 error = gfs2_glock_nq_init(ip->i_gl, LM_ST_SHARED, LM_FLAG_ANY,
67 &i_gh);
68 if (!error) {
69 error = generic_file_llseek(file, offset, whence);
70 gfs2_glock_dq_uninit(&i_gh);
71 }
72 break;
73
74 case SEEK_DATA:
75 error = gfs2_seek_data(file, offset);
76 break;
77
78 case SEEK_HOLE:
79 error = gfs2_seek_hole(file, offset);
80 break;
81
82 case SEEK_CUR:
83 case SEEK_SET:
84 /*
85 * These don't reference inode->i_size and don't depend on the
86 * block mapping, so we don't need the glock.
87 */
88 error = generic_file_llseek(file, offset, whence);
89 break;
90 default:
91 error = -EINVAL;
92 }
93
94 return error;
95}
96
97/**
98 * gfs2_readdir - Iterator for a directory
99 * @file: The directory to read from
100 * @ctx: What to feed directory entries to
101 *
102 * Returns: errno
103 */
104
105static int gfs2_readdir(struct file *file, struct dir_context *ctx)
106{
107 struct inode *dir = file->f_mapping->host;
108 struct gfs2_inode *dip = GFS2_I(dir);
109 struct gfs2_holder d_gh;
110 int error;
111
112 error = gfs2_glock_nq_init(dip->i_gl, LM_ST_SHARED, 0, &d_gh);
113 if (error)
114 return error;
115
116 error = gfs2_dir_read(dir, ctx, &file->f_ra);
117
118 gfs2_glock_dq_uninit(&d_gh);
119
120 return error;
121}
122
123/*
124 * struct fsflag_gfs2flag
125 *
126 * The FS_JOURNAL_DATA_FL flag maps to GFS2_DIF_INHERIT_JDATA for directories,
127 * and to GFS2_DIF_JDATA for non-directories.
128 */
129static struct {
130 u32 fsflag;
131 u32 gfsflag;
132} fsflag_gfs2flag[] = {
133 {FS_SYNC_FL, GFS2_DIF_SYNC},
134 {FS_IMMUTABLE_FL, GFS2_DIF_IMMUTABLE},
135 {FS_APPEND_FL, GFS2_DIF_APPENDONLY},
136 {FS_NOATIME_FL, GFS2_DIF_NOATIME},
137 {FS_INDEX_FL, GFS2_DIF_EXHASH},
138 {FS_TOPDIR_FL, GFS2_DIF_TOPDIR},
139 {FS_JOURNAL_DATA_FL, GFS2_DIF_JDATA | GFS2_DIF_INHERIT_JDATA},
140};
141
142static inline u32 gfs2_gfsflags_to_fsflags(struct inode *inode, u32 gfsflags)
143{
144 int i;
145 u32 fsflags = 0;
146
147 if (S_ISDIR(inode->i_mode))
148 gfsflags &= ~GFS2_DIF_JDATA;
149 else
150 gfsflags &= ~GFS2_DIF_INHERIT_JDATA;
151
152 for (i = 0; i < ARRAY_SIZE(fsflag_gfs2flag); i++)
153 if (gfsflags & fsflag_gfs2flag[i].gfsflag)
154 fsflags |= fsflag_gfs2flag[i].fsflag;
155 return fsflags;
156}
157
158int gfs2_fileattr_get(struct dentry *dentry, struct fileattr *fa)
159{
160 struct inode *inode = d_inode(dentry);
161 struct gfs2_inode *ip = GFS2_I(inode);
162 struct gfs2_holder gh;
163 int error;
164 u32 fsflags;
165
166 if (d_is_special(dentry))
167 return -ENOTTY;
168
169 gfs2_holder_init(ip->i_gl, LM_ST_SHARED, 0, &gh);
170 error = gfs2_glock_nq(&gh);
171 if (error)
172 goto out_uninit;
173
174 fsflags = gfs2_gfsflags_to_fsflags(inode, ip->i_diskflags);
175
176 fileattr_fill_flags(fa, fsflags);
177
178 gfs2_glock_dq(&gh);
179out_uninit:
180 gfs2_holder_uninit(&gh);
181 return error;
182}
183
184void gfs2_set_inode_flags(struct inode *inode)
185{
186 struct gfs2_inode *ip = GFS2_I(inode);
187 unsigned int flags = inode->i_flags;
188
189 flags &= ~(S_SYNC|S_APPEND|S_IMMUTABLE|S_NOATIME|S_DIRSYNC|S_NOSEC);
190 if ((ip->i_eattr == 0) && !is_sxid(inode->i_mode))
191 flags |= S_NOSEC;
192 if (ip->i_diskflags & GFS2_DIF_IMMUTABLE)
193 flags |= S_IMMUTABLE;
194 if (ip->i_diskflags & GFS2_DIF_APPENDONLY)
195 flags |= S_APPEND;
196 if (ip->i_diskflags & GFS2_DIF_NOATIME)
197 flags |= S_NOATIME;
198 if (ip->i_diskflags & GFS2_DIF_SYNC)
199 flags |= S_SYNC;
200 inode->i_flags = flags;
201}
202
203/* Flags that can be set by user space */
204#define GFS2_FLAGS_USER_SET (GFS2_DIF_JDATA| \
205 GFS2_DIF_IMMUTABLE| \
206 GFS2_DIF_APPENDONLY| \
207 GFS2_DIF_NOATIME| \
208 GFS2_DIF_SYNC| \
209 GFS2_DIF_TOPDIR| \
210 GFS2_DIF_INHERIT_JDATA)
211
212/**
213 * do_gfs2_set_flags - set flags on an inode
214 * @inode: The inode
215 * @reqflags: The flags to set
216 * @mask: Indicates which flags are valid
217 *
218 */
219static int do_gfs2_set_flags(struct inode *inode, u32 reqflags, u32 mask)
220{
221 struct gfs2_inode *ip = GFS2_I(inode);
222 struct gfs2_sbd *sdp = GFS2_SB(inode);
223 struct buffer_head *bh;
224 struct gfs2_holder gh;
225 int error;
226 u32 new_flags, flags;
227
228 error = gfs2_glock_nq_init(ip->i_gl, LM_ST_EXCLUSIVE, 0, &gh);
229 if (error)
230 return error;
231
232 error = 0;
233 flags = ip->i_diskflags;
234 new_flags = (flags & ~mask) | (reqflags & mask);
235 if ((new_flags ^ flags) == 0)
236 goto out;
237
238 if (!IS_IMMUTABLE(inode)) {
239 error = gfs2_permission(&nop_mnt_idmap, inode, MAY_WRITE);
240 if (error)
241 goto out;
242 }
243 if ((flags ^ new_flags) & GFS2_DIF_JDATA) {
244 if (new_flags & GFS2_DIF_JDATA)
245 gfs2_log_flush(sdp, ip->i_gl,
246 GFS2_LOG_HEAD_FLUSH_NORMAL |
247 GFS2_LFC_SET_FLAGS);
248 error = filemap_fdatawrite(inode->i_mapping);
249 if (error)
250 goto out;
251 error = filemap_fdatawait(inode->i_mapping);
252 if (error)
253 goto out;
254 truncate_inode_pages(inode->i_mapping, 0);
255 if (new_flags & GFS2_DIF_JDATA)
256 gfs2_ordered_del_inode(ip);
257 }
258 error = gfs2_trans_begin(sdp, RES_DINODE, 0);
259 if (error)
260 goto out;
261 error = gfs2_meta_inode_buffer(ip, &bh);
262 if (error)
263 goto out_trans_end;
264 inode_set_ctime_current(inode);
265 gfs2_trans_add_meta(ip->i_gl, bh);
266 ip->i_diskflags = new_flags;
267 gfs2_dinode_out(ip, bh->b_data);
268 brelse(bh);
269 gfs2_set_inode_flags(inode);
270 gfs2_set_aops(inode);
271out_trans_end:
272 gfs2_trans_end(sdp);
273out:
274 gfs2_glock_dq_uninit(&gh);
275 return error;
276}
277
278int gfs2_fileattr_set(struct mnt_idmap *idmap,
279 struct dentry *dentry, struct fileattr *fa)
280{
281 struct inode *inode = d_inode(dentry);
282 u32 fsflags = fa->flags, gfsflags = 0;
283 u32 mask;
284 int i;
285
286 if (d_is_special(dentry))
287 return -ENOTTY;
288
289 if (fileattr_has_fsx(fa))
290 return -EOPNOTSUPP;
291
292 for (i = 0; i < ARRAY_SIZE(fsflag_gfs2flag); i++) {
293 if (fsflags & fsflag_gfs2flag[i].fsflag) {
294 fsflags &= ~fsflag_gfs2flag[i].fsflag;
295 gfsflags |= fsflag_gfs2flag[i].gfsflag;
296 }
297 }
298 if (fsflags || gfsflags & ~GFS2_FLAGS_USER_SET)
299 return -EINVAL;
300
301 mask = GFS2_FLAGS_USER_SET;
302 if (S_ISDIR(inode->i_mode)) {
303 mask &= ~GFS2_DIF_JDATA;
304 } else {
305 /* The GFS2_DIF_TOPDIR flag is only valid for directories. */
306 if (gfsflags & GFS2_DIF_TOPDIR)
307 return -EINVAL;
308 mask &= ~(GFS2_DIF_TOPDIR | GFS2_DIF_INHERIT_JDATA);
309 }
310
311 return do_gfs2_set_flags(inode, gfsflags, mask);
312}
313
314static int gfs2_getlabel(struct file *filp, char __user *label)
315{
316 struct inode *inode = file_inode(filp);
317 struct gfs2_sbd *sdp = GFS2_SB(inode);
318
319 if (copy_to_user(label, sdp->sd_sb.sb_locktable, GFS2_LOCKNAME_LEN))
320 return -EFAULT;
321
322 return 0;
323}
324
325static long gfs2_ioctl(struct file *filp, unsigned int cmd, unsigned long arg)
326{
327 switch(cmd) {
328 case FITRIM:
329 return gfs2_fitrim(filp, (void __user *)arg);
330 case FS_IOC_GETFSLABEL:
331 return gfs2_getlabel(filp, (char __user *)arg);
332 }
333
334 return -ENOTTY;
335}
336
337#ifdef CONFIG_COMPAT
338static long gfs2_compat_ioctl(struct file *filp, unsigned int cmd, unsigned long arg)
339{
340 switch(cmd) {
341 /* Keep this list in sync with gfs2_ioctl */
342 case FITRIM:
343 case FS_IOC_GETFSLABEL:
344 break;
345 default:
346 return -ENOIOCTLCMD;
347 }
348
349 return gfs2_ioctl(filp, cmd, (unsigned long)compat_ptr(arg));
350}
351#else
352#define gfs2_compat_ioctl NULL
353#endif
354
355/**
356 * gfs2_size_hint - Give a hint to the size of a write request
357 * @filep: The struct file
358 * @offset: The file offset of the write
359 * @size: The length of the write
360 *
361 * When we are about to do a write, this function records the total
362 * write size in order to provide a suitable hint to the lower layers
363 * about how many blocks will be required.
364 *
365 */
366
367static void gfs2_size_hint(struct file *filep, loff_t offset, size_t size)
368{
369 struct inode *inode = file_inode(filep);
370 struct gfs2_sbd *sdp = GFS2_SB(inode);
371 struct gfs2_inode *ip = GFS2_I(inode);
372 size_t blks = (size + sdp->sd_sb.sb_bsize - 1) >> sdp->sd_sb.sb_bsize_shift;
373 int hint = min_t(size_t, INT_MAX, blks);
374
375 if (hint > atomic_read(&ip->i_sizehint))
376 atomic_set(&ip->i_sizehint, hint);
377}
378
379/**
380 * gfs2_allocate_folio_backing - Allocate blocks for a write fault
381 * @folio: The (locked) folio to allocate backing for
382 * @length: Size of the allocation
383 *
384 * We try to allocate all the blocks required for the folio in one go. This
385 * might fail for various reasons, so we keep trying until all the blocks to
386 * back this folio are allocated. If some of the blocks are already allocated,
387 * that is ok too.
388 */
389static int gfs2_allocate_folio_backing(struct folio *folio, size_t length)
390{
391 u64 pos = folio_pos(folio);
392
393 do {
394 struct iomap iomap = { };
395
396 if (gfs2_iomap_alloc(folio->mapping->host, pos, length, &iomap))
397 return -EIO;
398
399 if (length < iomap.length)
400 iomap.length = length;
401 length -= iomap.length;
402 pos += iomap.length;
403 } while (length > 0);
404
405 return 0;
406}
407
408/**
409 * gfs2_page_mkwrite - Make a shared, mmap()ed, page writable
410 * @vmf: The virtual memory fault containing the page to become writable
411 *
412 * When the page becomes writable, we need to ensure that we have
413 * blocks allocated on disk to back that page.
414 */
415
416static vm_fault_t gfs2_page_mkwrite(struct vm_fault *vmf)
417{
418 struct folio *folio = page_folio(vmf->page);
419 struct inode *inode = file_inode(vmf->vma->vm_file);
420 struct gfs2_inode *ip = GFS2_I(inode);
421 struct gfs2_sbd *sdp = GFS2_SB(inode);
422 struct gfs2_alloc_parms ap = {};
423 u64 pos = folio_pos(folio);
424 unsigned int data_blocks, ind_blocks, rblocks;
425 vm_fault_t ret = VM_FAULT_LOCKED;
426 struct gfs2_holder gh;
427 size_t length;
428 loff_t size;
429 int err;
430
431 sb_start_pagefault(inode->i_sb);
432
433 gfs2_holder_init(ip->i_gl, LM_ST_EXCLUSIVE, 0, &gh);
434 err = gfs2_glock_nq(&gh);
435 if (err) {
436 ret = vmf_fs_error(err);
437 goto out_uninit;
438 }
439
440 /* Check folio index against inode size */
441 size = i_size_read(inode);
442 if (pos >= size) {
443 ret = VM_FAULT_SIGBUS;
444 goto out_unlock;
445 }
446
447 /* Update file times before taking folio lock */
448 file_update_time(vmf->vma->vm_file);
449
450 /* folio is wholly or partially inside EOF */
451 if (size - pos < folio_size(folio))
452 length = size - pos;
453 else
454 length = folio_size(folio);
455
456 gfs2_size_hint(vmf->vma->vm_file, pos, length);
457
458 set_bit(GLF_DIRTY, &ip->i_gl->gl_flags);
459 set_bit(GIF_SW_PAGED, &ip->i_flags);
460
461 /*
462 * iomap_writepage / iomap_writepages currently don't support inline
463 * files, so always unstuff here.
464 */
465
466 if (!gfs2_is_stuffed(ip) &&
467 !gfs2_write_alloc_required(ip, pos, length)) {
468 folio_lock(folio);
469 if (!folio_test_uptodate(folio) ||
470 folio->mapping != inode->i_mapping) {
471 ret = VM_FAULT_NOPAGE;
472 folio_unlock(folio);
473 }
474 goto out_unlock;
475 }
476
477 err = gfs2_rindex_update(sdp);
478 if (err) {
479 ret = vmf_fs_error(err);
480 goto out_unlock;
481 }
482
483 gfs2_write_calc_reserv(ip, length, &data_blocks, &ind_blocks);
484 ap.target = data_blocks + ind_blocks;
485 err = gfs2_quota_lock_check(ip, &ap);
486 if (err) {
487 ret = vmf_fs_error(err);
488 goto out_unlock;
489 }
490 err = gfs2_inplace_reserve(ip, &ap);
491 if (err) {
492 ret = vmf_fs_error(err);
493 goto out_quota_unlock;
494 }
495
496 rblocks = RES_DINODE + ind_blocks;
497 if (gfs2_is_jdata(ip))
498 rblocks += data_blocks ? data_blocks : 1;
499 if (ind_blocks || data_blocks) {
500 rblocks += RES_STATFS + RES_QUOTA;
501 rblocks += gfs2_rg_blocks(ip, data_blocks + ind_blocks);
502 }
503 err = gfs2_trans_begin(sdp, rblocks, 0);
504 if (err) {
505 ret = vmf_fs_error(err);
506 goto out_trans_fail;
507 }
508
509 /* Unstuff, if required, and allocate backing blocks for folio */
510 if (gfs2_is_stuffed(ip)) {
511 err = gfs2_unstuff_dinode(ip);
512 if (err) {
513 ret = vmf_fs_error(err);
514 goto out_trans_end;
515 }
516 }
517
518 folio_lock(folio);
519 /* If truncated, we must retry the operation, we may have raced
520 * with the glock demotion code.
521 */
522 if (!folio_test_uptodate(folio) || folio->mapping != inode->i_mapping) {
523 ret = VM_FAULT_NOPAGE;
524 goto out_page_locked;
525 }
526
527 err = gfs2_allocate_folio_backing(folio, length);
528 if (err)
529 ret = vmf_fs_error(err);
530
531out_page_locked:
532 if (ret != VM_FAULT_LOCKED)
533 folio_unlock(folio);
534out_trans_end:
535 gfs2_trans_end(sdp);
536out_trans_fail:
537 gfs2_inplace_release(ip);
538out_quota_unlock:
539 gfs2_quota_unlock(ip);
540out_unlock:
541 gfs2_glock_dq(&gh);
542out_uninit:
543 gfs2_holder_uninit(&gh);
544 if (ret == VM_FAULT_LOCKED) {
545 folio_mark_dirty(folio);
546 folio_wait_stable(folio);
547 }
548 sb_end_pagefault(inode->i_sb);
549 return ret;
550}
551
552static vm_fault_t gfs2_fault(struct vm_fault *vmf)
553{
554 struct inode *inode = file_inode(vmf->vma->vm_file);
555 struct gfs2_inode *ip = GFS2_I(inode);
556 struct gfs2_holder gh;
557 vm_fault_t ret;
558 int err;
559
560 gfs2_holder_init(ip->i_gl, LM_ST_SHARED, 0, &gh);
561 err = gfs2_glock_nq(&gh);
562 if (err) {
563 ret = vmf_fs_error(err);
564 goto out_uninit;
565 }
566 ret = filemap_fault(vmf);
567 gfs2_glock_dq(&gh);
568out_uninit:
569 gfs2_holder_uninit(&gh);
570 return ret;
571}
572
573static const struct vm_operations_struct gfs2_vm_ops = {
574 .fault = gfs2_fault,
575 .map_pages = filemap_map_pages,
576 .page_mkwrite = gfs2_page_mkwrite,
577};
578
579/**
580 * gfs2_mmap
581 * @file: The file to map
582 * @vma: The VMA which described the mapping
583 *
584 * There is no need to get a lock here unless we should be updating
585 * atime. We ignore any locking errors since the only consequence is
586 * a missed atime update (which will just be deferred until later).
587 *
588 * Returns: 0
589 */
590
591static int gfs2_mmap(struct file *file, struct vm_area_struct *vma)
592{
593 struct gfs2_inode *ip = GFS2_I(file->f_mapping->host);
594
595 if (!(file->f_flags & O_NOATIME) &&
596 !IS_NOATIME(&ip->i_inode)) {
597 struct gfs2_holder i_gh;
598 int error;
599
600 error = gfs2_glock_nq_init(ip->i_gl, LM_ST_SHARED, LM_FLAG_ANY,
601 &i_gh);
602 if (error)
603 return error;
604 /* grab lock to update inode */
605 gfs2_glock_dq_uninit(&i_gh);
606 file_accessed(file);
607 }
608 vma->vm_ops = &gfs2_vm_ops;
609
610 return 0;
611}
612
613/**
614 * gfs2_open_common - This is common to open and atomic_open
615 * @inode: The inode being opened
616 * @file: The file being opened
617 *
618 * This maybe called under a glock or not depending upon how it has
619 * been called. We must always be called under a glock for regular
620 * files, however. For other file types, it does not matter whether
621 * we hold the glock or not.
622 *
623 * Returns: Error code or 0 for success
624 */
625
626int gfs2_open_common(struct inode *inode, struct file *file)
627{
628 struct gfs2_file *fp;
629 int ret;
630
631 if (S_ISREG(inode->i_mode)) {
632 ret = generic_file_open(inode, file);
633 if (ret)
634 return ret;
635
636 if (!gfs2_is_jdata(GFS2_I(inode)))
637 file->f_mode |= FMODE_CAN_ODIRECT;
638 }
639
640 fp = kzalloc(sizeof(struct gfs2_file), GFP_NOFS);
641 if (!fp)
642 return -ENOMEM;
643
644 mutex_init(&fp->f_fl_mutex);
645
646 gfs2_assert_warn(GFS2_SB(inode), !file->private_data);
647 file->private_data = fp;
648 if (file->f_mode & FMODE_WRITE) {
649 ret = gfs2_qa_get(GFS2_I(inode));
650 if (ret)
651 goto fail;
652 }
653 return 0;
654
655fail:
656 kfree(file->private_data);
657 file->private_data = NULL;
658 return ret;
659}
660
661/**
662 * gfs2_open - open a file
663 * @inode: the inode to open
664 * @file: the struct file for this opening
665 *
666 * After atomic_open, this function is only used for opening files
667 * which are already cached. We must still get the glock for regular
668 * files to ensure that we have the file size uptodate for the large
669 * file check which is in the common code. That is only an issue for
670 * regular files though.
671 *
672 * Returns: errno
673 */
674
675static int gfs2_open(struct inode *inode, struct file *file)
676{
677 struct gfs2_inode *ip = GFS2_I(inode);
678 struct gfs2_holder i_gh;
679 int error;
680 bool need_unlock = false;
681
682 if (S_ISREG(ip->i_inode.i_mode)) {
683 error = gfs2_glock_nq_init(ip->i_gl, LM_ST_SHARED, LM_FLAG_ANY,
684 &i_gh);
685 if (error)
686 return error;
687 need_unlock = true;
688 }
689
690 error = gfs2_open_common(inode, file);
691
692 if (need_unlock)
693 gfs2_glock_dq_uninit(&i_gh);
694
695 return error;
696}
697
698/**
699 * gfs2_release - called to close a struct file
700 * @inode: the inode the struct file belongs to
701 * @file: the struct file being closed
702 *
703 * Returns: errno
704 */
705
706static int gfs2_release(struct inode *inode, struct file *file)
707{
708 struct gfs2_inode *ip = GFS2_I(inode);
709
710 kfree(file->private_data);
711 file->private_data = NULL;
712
713 if (file->f_mode & FMODE_WRITE) {
714 if (gfs2_rs_active(&ip->i_res))
715 gfs2_rs_delete(ip);
716 gfs2_qa_put(ip);
717 }
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 inline bool should_fault_in_pages(struct iov_iter *i,
780 struct kiocb *iocb,
781 size_t *prev_count,
782 size_t *window_size)
783{
784 size_t count = iov_iter_count(i);
785 size_t size, offs;
786
787 if (!count)
788 return false;
789 if (!user_backed_iter(i))
790 return false;
791
792 /*
793 * Try to fault in multiple pages initially. When that doesn't result
794 * in any progress, fall back to a single page.
795 */
796 size = PAGE_SIZE;
797 offs = offset_in_page(iocb->ki_pos);
798 if (*prev_count != count) {
799 size_t nr_dirtied;
800
801 nr_dirtied = max(current->nr_dirtied_pause -
802 current->nr_dirtied, 8);
803 size = min_t(size_t, SZ_1M, nr_dirtied << PAGE_SHIFT);
804 }
805
806 *prev_count = count;
807 *window_size = size - offs;
808 return true;
809}
810
811static ssize_t gfs2_file_direct_read(struct kiocb *iocb, struct iov_iter *to,
812 struct gfs2_holder *gh)
813{
814 struct file *file = iocb->ki_filp;
815 struct gfs2_inode *ip = GFS2_I(file->f_mapping->host);
816 size_t prev_count = 0, window_size = 0;
817 size_t read = 0;
818 ssize_t ret;
819
820 /*
821 * In this function, we disable page faults when we're holding the
822 * inode glock while doing I/O. If a page fault occurs, we indicate
823 * that the inode glock may be dropped, fault in the pages manually,
824 * and retry.
825 *
826 * Unlike generic_file_read_iter, for reads, iomap_dio_rw can trigger
827 * physical as well as manual page faults, and we need to disable both
828 * kinds.
829 *
830 * For direct I/O, gfs2 takes the inode glock in deferred mode. This
831 * locking mode is compatible with other deferred holders, so multiple
832 * processes and nodes can do direct I/O to a file at the same time.
833 * There's no guarantee that reads or writes will be atomic. Any
834 * coordination among readers and writers needs to happen externally.
835 */
836
837 if (!iov_iter_count(to))
838 return 0; /* skip atime */
839
840 gfs2_holder_init(ip->i_gl, LM_ST_DEFERRED, 0, gh);
841retry:
842 ret = gfs2_glock_nq(gh);
843 if (ret)
844 goto out_uninit;
845 pagefault_disable();
846 to->nofault = true;
847 ret = iomap_dio_rw(iocb, to, &gfs2_iomap_ops, NULL,
848 IOMAP_DIO_PARTIAL, NULL, read);
849 to->nofault = false;
850 pagefault_enable();
851 if (ret <= 0 && ret != -EFAULT)
852 goto out_unlock;
853 /* No increment (+=) because iomap_dio_rw returns a cumulative value. */
854 if (ret > 0)
855 read = ret;
856
857 if (should_fault_in_pages(to, iocb, &prev_count, &window_size)) {
858 gfs2_glock_dq(gh);
859 window_size -= fault_in_iov_iter_writeable(to, window_size);
860 if (window_size)
861 goto retry;
862 }
863out_unlock:
864 if (gfs2_holder_queued(gh))
865 gfs2_glock_dq(gh);
866out_uninit:
867 gfs2_holder_uninit(gh);
868 /* User space doesn't expect partial success. */
869 if (ret < 0)
870 return ret;
871 return read;
872}
873
874static ssize_t gfs2_file_direct_write(struct kiocb *iocb, struct iov_iter *from,
875 struct gfs2_holder *gh)
876{
877 struct file *file = iocb->ki_filp;
878 struct inode *inode = file->f_mapping->host;
879 struct gfs2_inode *ip = GFS2_I(inode);
880 size_t prev_count = 0, window_size = 0;
881 size_t written = 0;
882 bool enough_retries;
883 ssize_t ret;
884
885 /*
886 * In this function, we disable page faults when we're holding the
887 * inode glock while doing I/O. If a page fault occurs, we indicate
888 * that the inode glock may be dropped, fault in the pages manually,
889 * and retry.
890 *
891 * For writes, iomap_dio_rw only triggers manual page faults, so we
892 * don't need to disable physical ones.
893 */
894
895 /*
896 * Deferred lock, even if its a write, since we do no allocation on
897 * this path. All we need to change is the atime, and this lock mode
898 * ensures that other nodes have flushed their buffered read caches
899 * (i.e. their page cache entries for this inode). We do not,
900 * unfortunately, have the option of only flushing a range like the
901 * VFS does.
902 */
903 gfs2_holder_init(ip->i_gl, LM_ST_DEFERRED, 0, gh);
904retry:
905 ret = gfs2_glock_nq(gh);
906 if (ret)
907 goto out_uninit;
908 /* Silently fall back to buffered I/O when writing beyond EOF */
909 if (iocb->ki_pos + iov_iter_count(from) > i_size_read(&ip->i_inode))
910 goto out_unlock;
911
912 from->nofault = true;
913 ret = iomap_dio_rw(iocb, from, &gfs2_iomap_ops, NULL,
914 IOMAP_DIO_PARTIAL, NULL, written);
915 from->nofault = false;
916 if (ret <= 0) {
917 if (ret == -ENOTBLK)
918 ret = 0;
919 if (ret != -EFAULT)
920 goto out_unlock;
921 }
922 /* No increment (+=) because iomap_dio_rw returns a cumulative value. */
923 if (ret > 0)
924 written = ret;
925
926 enough_retries = prev_count == iov_iter_count(from) &&
927 window_size <= PAGE_SIZE;
928 if (should_fault_in_pages(from, iocb, &prev_count, &window_size)) {
929 gfs2_glock_dq(gh);
930 window_size -= fault_in_iov_iter_readable(from, window_size);
931 if (window_size) {
932 if (!enough_retries)
933 goto retry;
934 /* fall back to buffered I/O */
935 ret = 0;
936 }
937 }
938out_unlock:
939 if (gfs2_holder_queued(gh))
940 gfs2_glock_dq(gh);
941out_uninit:
942 gfs2_holder_uninit(gh);
943 /* User space doesn't expect partial success. */
944 if (ret < 0)
945 return ret;
946 return written;
947}
948
949static ssize_t gfs2_file_read_iter(struct kiocb *iocb, struct iov_iter *to)
950{
951 struct gfs2_inode *ip;
952 struct gfs2_holder gh;
953 size_t prev_count = 0, window_size = 0;
954 size_t read = 0;
955 ssize_t ret;
956
957 /*
958 * In this function, we disable page faults when we're holding the
959 * inode glock while doing I/O. If a page fault occurs, we indicate
960 * that the inode glock may be dropped, fault in the pages manually,
961 * and retry.
962 */
963
964 if (iocb->ki_flags & IOCB_DIRECT)
965 return gfs2_file_direct_read(iocb, to, &gh);
966
967 pagefault_disable();
968 iocb->ki_flags |= IOCB_NOIO;
969 ret = generic_file_read_iter(iocb, to);
970 iocb->ki_flags &= ~IOCB_NOIO;
971 pagefault_enable();
972 if (ret >= 0) {
973 if (!iov_iter_count(to))
974 return ret;
975 read = ret;
976 } else if (ret != -EFAULT) {
977 if (ret != -EAGAIN)
978 return ret;
979 if (iocb->ki_flags & IOCB_NOWAIT)
980 return ret;
981 }
982 ip = GFS2_I(iocb->ki_filp->f_mapping->host);
983 gfs2_holder_init(ip->i_gl, LM_ST_SHARED, 0, &gh);
984retry:
985 ret = gfs2_glock_nq(&gh);
986 if (ret)
987 goto out_uninit;
988 pagefault_disable();
989 ret = generic_file_read_iter(iocb, to);
990 pagefault_enable();
991 if (ret <= 0 && ret != -EFAULT)
992 goto out_unlock;
993 if (ret > 0)
994 read += ret;
995
996 if (should_fault_in_pages(to, iocb, &prev_count, &window_size)) {
997 gfs2_glock_dq(&gh);
998 window_size -= fault_in_iov_iter_writeable(to, window_size);
999 if (window_size)
1000 goto retry;
1001 }
1002out_unlock:
1003 if (gfs2_holder_queued(&gh))
1004 gfs2_glock_dq(&gh);
1005out_uninit:
1006 gfs2_holder_uninit(&gh);
1007 return read ? read : ret;
1008}
1009
1010static ssize_t gfs2_file_buffered_write(struct kiocb *iocb,
1011 struct iov_iter *from,
1012 struct gfs2_holder *gh)
1013{
1014 struct file *file = iocb->ki_filp;
1015 struct inode *inode = file_inode(file);
1016 struct gfs2_inode *ip = GFS2_I(inode);
1017 struct gfs2_sbd *sdp = GFS2_SB(inode);
1018 struct gfs2_holder *statfs_gh = NULL;
1019 size_t prev_count = 0, window_size = 0;
1020 size_t orig_count = iov_iter_count(from);
1021 size_t written = 0;
1022 ssize_t ret;
1023
1024 /*
1025 * In this function, we disable page faults when we're holding the
1026 * inode glock while doing I/O. If a page fault occurs, we indicate
1027 * that the inode glock may be dropped, fault in the pages manually,
1028 * and retry.
1029 */
1030
1031 if (inode == sdp->sd_rindex) {
1032 statfs_gh = kmalloc(sizeof(*statfs_gh), GFP_NOFS);
1033 if (!statfs_gh)
1034 return -ENOMEM;
1035 }
1036
1037 gfs2_holder_init(ip->i_gl, LM_ST_EXCLUSIVE, 0, gh);
1038 if (should_fault_in_pages(from, iocb, &prev_count, &window_size)) {
1039retry:
1040 window_size -= fault_in_iov_iter_readable(from, window_size);
1041 if (!window_size) {
1042 ret = -EFAULT;
1043 goto out_uninit;
1044 }
1045 from->count = min(from->count, window_size);
1046 }
1047 ret = gfs2_glock_nq(gh);
1048 if (ret)
1049 goto out_uninit;
1050
1051 if (inode == sdp->sd_rindex) {
1052 struct gfs2_inode *m_ip = GFS2_I(sdp->sd_statfs_inode);
1053
1054 ret = gfs2_glock_nq_init(m_ip->i_gl, LM_ST_EXCLUSIVE,
1055 GL_NOCACHE, statfs_gh);
1056 if (ret)
1057 goto out_unlock;
1058 }
1059
1060 pagefault_disable();
1061 ret = iomap_file_buffered_write(iocb, from, &gfs2_iomap_ops, NULL);
1062 pagefault_enable();
1063 if (ret > 0)
1064 written += ret;
1065
1066 if (inode == sdp->sd_rindex)
1067 gfs2_glock_dq_uninit(statfs_gh);
1068
1069 if (ret <= 0 && ret != -EFAULT)
1070 goto out_unlock;
1071
1072 from->count = orig_count - written;
1073 if (should_fault_in_pages(from, iocb, &prev_count, &window_size)) {
1074 gfs2_glock_dq(gh);
1075 goto retry;
1076 }
1077out_unlock:
1078 if (gfs2_holder_queued(gh))
1079 gfs2_glock_dq(gh);
1080out_uninit:
1081 gfs2_holder_uninit(gh);
1082 kfree(statfs_gh);
1083 from->count = orig_count - written;
1084 return written ? written : ret;
1085}
1086
1087/**
1088 * gfs2_file_write_iter - Perform a write to a file
1089 * @iocb: The io context
1090 * @from: The data to write
1091 *
1092 * We have to do a lock/unlock here to refresh the inode size for
1093 * O_APPEND writes, otherwise we can land up writing at the wrong
1094 * offset. There is still a race, but provided the app is using its
1095 * own file locking, this will make O_APPEND work as expected.
1096 *
1097 */
1098
1099static ssize_t gfs2_file_write_iter(struct kiocb *iocb, struct iov_iter *from)
1100{
1101 struct file *file = iocb->ki_filp;
1102 struct inode *inode = file_inode(file);
1103 struct gfs2_inode *ip = GFS2_I(inode);
1104 struct gfs2_holder gh;
1105 ssize_t ret;
1106
1107 gfs2_size_hint(file, iocb->ki_pos, iov_iter_count(from));
1108
1109 if (iocb->ki_flags & IOCB_APPEND) {
1110 ret = gfs2_glock_nq_init(ip->i_gl, LM_ST_SHARED, 0, &gh);
1111 if (ret)
1112 return ret;
1113 gfs2_glock_dq_uninit(&gh);
1114 }
1115
1116 inode_lock(inode);
1117 ret = generic_write_checks(iocb, from);
1118 if (ret <= 0)
1119 goto out_unlock;
1120
1121 ret = file_remove_privs(file);
1122 if (ret)
1123 goto out_unlock;
1124
1125 if (iocb->ki_flags & IOCB_DIRECT) {
1126 struct address_space *mapping = file->f_mapping;
1127 ssize_t buffered, ret2;
1128
1129 /*
1130 * Note that under direct I/O, we don't allow and inode
1131 * timestamp updates, so we're not calling file_update_time()
1132 * here.
1133 */
1134
1135 ret = gfs2_file_direct_write(iocb, from, &gh);
1136 if (ret < 0 || !iov_iter_count(from))
1137 goto out_unlock;
1138
1139 iocb->ki_flags |= IOCB_DSYNC;
1140 buffered = gfs2_file_buffered_write(iocb, from, &gh);
1141 if (unlikely(buffered <= 0)) {
1142 if (!ret)
1143 ret = buffered;
1144 goto out_unlock;
1145 }
1146
1147 /*
1148 * We need to ensure that the page cache pages are written to
1149 * disk and invalidated to preserve the expected O_DIRECT
1150 * semantics. If the writeback or invalidate fails, only report
1151 * the direct I/O range as we don't know if the buffered pages
1152 * made it to disk.
1153 */
1154 ret2 = generic_write_sync(iocb, buffered);
1155 invalidate_mapping_pages(mapping,
1156 (iocb->ki_pos - buffered) >> PAGE_SHIFT,
1157 (iocb->ki_pos - 1) >> PAGE_SHIFT);
1158 if (!ret || ret2 > 0)
1159 ret += ret2;
1160 } else {
1161 ret = file_update_time(file);
1162 if (ret)
1163 goto out_unlock;
1164
1165 ret = gfs2_file_buffered_write(iocb, from, &gh);
1166 if (likely(ret > 0))
1167 ret = generic_write_sync(iocb, ret);
1168 }
1169
1170out_unlock:
1171 inode_unlock(inode);
1172 return ret;
1173}
1174
1175static int fallocate_chunk(struct inode *inode, loff_t offset, loff_t len,
1176 int mode)
1177{
1178 struct super_block *sb = inode->i_sb;
1179 struct gfs2_inode *ip = GFS2_I(inode);
1180 loff_t end = offset + len;
1181 struct buffer_head *dibh;
1182 int error;
1183
1184 error = gfs2_meta_inode_buffer(ip, &dibh);
1185 if (unlikely(error))
1186 return error;
1187
1188 gfs2_trans_add_meta(ip->i_gl, dibh);
1189
1190 if (gfs2_is_stuffed(ip)) {
1191 error = gfs2_unstuff_dinode(ip);
1192 if (unlikely(error))
1193 goto out;
1194 }
1195
1196 while (offset < end) {
1197 struct iomap iomap = { };
1198
1199 error = gfs2_iomap_alloc(inode, offset, end - offset, &iomap);
1200 if (error)
1201 goto out;
1202 offset = iomap.offset + iomap.length;
1203 if (!(iomap.flags & IOMAP_F_NEW))
1204 continue;
1205 error = sb_issue_zeroout(sb, iomap.addr >> inode->i_blkbits,
1206 iomap.length >> inode->i_blkbits,
1207 GFP_NOFS);
1208 if (error) {
1209 fs_err(GFS2_SB(inode), "Failed to zero data buffers\n");
1210 goto out;
1211 }
1212 }
1213out:
1214 brelse(dibh);
1215 return error;
1216}
1217
1218/**
1219 * calc_max_reserv() - Reverse of write_calc_reserv. Given a number of
1220 * blocks, determine how many bytes can be written.
1221 * @ip: The inode in question.
1222 * @len: Max cap of bytes. What we return in *len must be <= this.
1223 * @data_blocks: Compute and return the number of data blocks needed
1224 * @ind_blocks: Compute and return the number of indirect blocks needed
1225 * @max_blocks: The total blocks available to work with.
1226 *
1227 * Returns: void, but @len, @data_blocks and @ind_blocks are filled in.
1228 */
1229static void calc_max_reserv(struct gfs2_inode *ip, loff_t *len,
1230 unsigned int *data_blocks, unsigned int *ind_blocks,
1231 unsigned int max_blocks)
1232{
1233 loff_t max = *len;
1234 const struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode);
1235 unsigned int tmp, max_data = max_blocks - 3 * (sdp->sd_max_height - 1);
1236
1237 for (tmp = max_data; tmp > sdp->sd_diptrs;) {
1238 tmp = DIV_ROUND_UP(tmp, sdp->sd_inptrs);
1239 max_data -= tmp;
1240 }
1241
1242 *data_blocks = max_data;
1243 *ind_blocks = max_blocks - max_data;
1244 *len = ((loff_t)max_data - 3) << sdp->sd_sb.sb_bsize_shift;
1245 if (*len > max) {
1246 *len = max;
1247 gfs2_write_calc_reserv(ip, max, data_blocks, ind_blocks);
1248 }
1249}
1250
1251static long __gfs2_fallocate(struct file *file, int mode, loff_t offset, loff_t len)
1252{
1253 struct inode *inode = file_inode(file);
1254 struct gfs2_sbd *sdp = GFS2_SB(inode);
1255 struct gfs2_inode *ip = GFS2_I(inode);
1256 struct gfs2_alloc_parms ap = {};
1257 unsigned int data_blocks = 0, ind_blocks = 0, rblocks;
1258 loff_t bytes, max_bytes, max_blks;
1259 int error;
1260 const loff_t pos = offset;
1261 const loff_t count = len;
1262 loff_t bsize_mask = ~((loff_t)sdp->sd_sb.sb_bsize - 1);
1263 loff_t next = (offset + len - 1) >> sdp->sd_sb.sb_bsize_shift;
1264 loff_t max_chunk_size = UINT_MAX & bsize_mask;
1265
1266 next = (next + 1) << sdp->sd_sb.sb_bsize_shift;
1267
1268 offset &= bsize_mask;
1269
1270 len = next - offset;
1271 bytes = sdp->sd_max_rg_data * sdp->sd_sb.sb_bsize / 2;
1272 if (!bytes)
1273 bytes = UINT_MAX;
1274 bytes &= bsize_mask;
1275 if (bytes == 0)
1276 bytes = sdp->sd_sb.sb_bsize;
1277
1278 gfs2_size_hint(file, offset, len);
1279
1280 gfs2_write_calc_reserv(ip, PAGE_SIZE, &data_blocks, &ind_blocks);
1281 ap.min_target = data_blocks + ind_blocks;
1282
1283 while (len > 0) {
1284 if (len < bytes)
1285 bytes = len;
1286 if (!gfs2_write_alloc_required(ip, offset, bytes)) {
1287 len -= bytes;
1288 offset += bytes;
1289 continue;
1290 }
1291
1292 /* We need to determine how many bytes we can actually
1293 * fallocate without exceeding quota or going over the
1294 * end of the fs. We start off optimistically by assuming
1295 * we can write max_bytes */
1296 max_bytes = (len > max_chunk_size) ? max_chunk_size : len;
1297
1298 /* Since max_bytes is most likely a theoretical max, we
1299 * calculate a more realistic 'bytes' to serve as a good
1300 * starting point for the number of bytes we may be able
1301 * to write */
1302 gfs2_write_calc_reserv(ip, bytes, &data_blocks, &ind_blocks);
1303 ap.target = data_blocks + ind_blocks;
1304
1305 error = gfs2_quota_lock_check(ip, &ap);
1306 if (error)
1307 return error;
1308 /* ap.allowed tells us how many blocks quota will allow
1309 * us to write. Check if this reduces max_blks */
1310 max_blks = UINT_MAX;
1311 if (ap.allowed)
1312 max_blks = ap.allowed;
1313
1314 error = gfs2_inplace_reserve(ip, &ap);
1315 if (error)
1316 goto out_qunlock;
1317
1318 /* check if the selected rgrp limits our max_blks further */
1319 if (ip->i_res.rs_reserved < max_blks)
1320 max_blks = ip->i_res.rs_reserved;
1321
1322 /* Almost done. Calculate bytes that can be written using
1323 * max_blks. We also recompute max_bytes, data_blocks and
1324 * ind_blocks */
1325 calc_max_reserv(ip, &max_bytes, &data_blocks,
1326 &ind_blocks, max_blks);
1327
1328 rblocks = RES_DINODE + ind_blocks + RES_STATFS + RES_QUOTA +
1329 RES_RG_HDR + gfs2_rg_blocks(ip, data_blocks + ind_blocks);
1330 if (gfs2_is_jdata(ip))
1331 rblocks += data_blocks ? data_blocks : 1;
1332
1333 error = gfs2_trans_begin(sdp, rblocks,
1334 PAGE_SIZE >> inode->i_blkbits);
1335 if (error)
1336 goto out_trans_fail;
1337
1338 error = fallocate_chunk(inode, offset, max_bytes, mode);
1339 gfs2_trans_end(sdp);
1340
1341 if (error)
1342 goto out_trans_fail;
1343
1344 len -= max_bytes;
1345 offset += max_bytes;
1346 gfs2_inplace_release(ip);
1347 gfs2_quota_unlock(ip);
1348 }
1349
1350 if (!(mode & FALLOC_FL_KEEP_SIZE) && (pos + count) > inode->i_size)
1351 i_size_write(inode, pos + count);
1352 file_update_time(file);
1353 mark_inode_dirty(inode);
1354
1355 if ((file->f_flags & O_DSYNC) || IS_SYNC(file->f_mapping->host))
1356 return vfs_fsync_range(file, pos, pos + count - 1,
1357 (file->f_flags & __O_SYNC) ? 0 : 1);
1358 return 0;
1359
1360out_trans_fail:
1361 gfs2_inplace_release(ip);
1362out_qunlock:
1363 gfs2_quota_unlock(ip);
1364 return error;
1365}
1366
1367static long gfs2_fallocate(struct file *file, int mode, loff_t offset, loff_t len)
1368{
1369 struct inode *inode = file_inode(file);
1370 struct gfs2_sbd *sdp = GFS2_SB(inode);
1371 struct gfs2_inode *ip = GFS2_I(inode);
1372 struct gfs2_holder gh;
1373 int ret;
1374
1375 if (mode & ~(FALLOC_FL_PUNCH_HOLE | FALLOC_FL_KEEP_SIZE))
1376 return -EOPNOTSUPP;
1377 /* fallocate is needed by gfs2_grow to reserve space in the rindex */
1378 if (gfs2_is_jdata(ip) && inode != sdp->sd_rindex)
1379 return -EOPNOTSUPP;
1380
1381 inode_lock(inode);
1382
1383 gfs2_holder_init(ip->i_gl, LM_ST_EXCLUSIVE, 0, &gh);
1384 ret = gfs2_glock_nq(&gh);
1385 if (ret)
1386 goto out_uninit;
1387
1388 if (!(mode & FALLOC_FL_KEEP_SIZE) &&
1389 (offset + len) > inode->i_size) {
1390 ret = inode_newsize_ok(inode, offset + len);
1391 if (ret)
1392 goto out_unlock;
1393 }
1394
1395 ret = get_write_access(inode);
1396 if (ret)
1397 goto out_unlock;
1398
1399 if (mode & FALLOC_FL_PUNCH_HOLE) {
1400 ret = __gfs2_punch_hole(file, offset, len);
1401 } else {
1402 ret = __gfs2_fallocate(file, mode, offset, len);
1403 if (ret)
1404 gfs2_rs_deltree(&ip->i_res);
1405 }
1406
1407 put_write_access(inode);
1408out_unlock:
1409 gfs2_glock_dq(&gh);
1410out_uninit:
1411 gfs2_holder_uninit(&gh);
1412 inode_unlock(inode);
1413 return ret;
1414}
1415
1416static ssize_t gfs2_file_splice_write(struct pipe_inode_info *pipe,
1417 struct file *out, loff_t *ppos,
1418 size_t len, unsigned int flags)
1419{
1420 ssize_t ret;
1421
1422 gfs2_size_hint(out, *ppos, len);
1423
1424 ret = iter_file_splice_write(pipe, out, ppos, len, flags);
1425 return ret;
1426}
1427
1428#ifdef CONFIG_GFS2_FS_LOCKING_DLM
1429
1430/**
1431 * gfs2_lock - acquire/release a posix lock on a file
1432 * @file: the file pointer
1433 * @cmd: either modify or retrieve lock state, possibly wait
1434 * @fl: type and range of lock
1435 *
1436 * Returns: errno
1437 */
1438
1439static int gfs2_lock(struct file *file, int cmd, struct file_lock *fl)
1440{
1441 struct gfs2_inode *ip = GFS2_I(file->f_mapping->host);
1442 struct gfs2_sbd *sdp = GFS2_SB(file->f_mapping->host);
1443 struct lm_lockstruct *ls = &sdp->sd_lockstruct;
1444
1445 if (!(fl->c.flc_flags & FL_POSIX))
1446 return -ENOLCK;
1447 if (gfs2_withdrawing_or_withdrawn(sdp)) {
1448 if (lock_is_unlock(fl))
1449 locks_lock_file_wait(file, fl);
1450 return -EIO;
1451 }
1452 if (cmd == F_CANCELLK)
1453 return dlm_posix_cancel(ls->ls_dlm, ip->i_no_addr, file, fl);
1454 else if (IS_GETLK(cmd))
1455 return dlm_posix_get(ls->ls_dlm, ip->i_no_addr, file, fl);
1456 else if (lock_is_unlock(fl))
1457 return dlm_posix_unlock(ls->ls_dlm, ip->i_no_addr, file, fl);
1458 else
1459 return dlm_posix_lock(ls->ls_dlm, ip->i_no_addr, file, cmd, fl);
1460}
1461
1462static void __flock_holder_uninit(struct file *file, struct gfs2_holder *fl_gh)
1463{
1464 struct gfs2_glock *gl = gfs2_glock_hold(fl_gh->gh_gl);
1465
1466 /*
1467 * Make sure gfs2_glock_put() won't sleep under the file->f_lock
1468 * spinlock.
1469 */
1470
1471 spin_lock(&file->f_lock);
1472 gfs2_holder_uninit(fl_gh);
1473 spin_unlock(&file->f_lock);
1474 gfs2_glock_put(gl);
1475}
1476
1477static int do_flock(struct file *file, int cmd, struct file_lock *fl)
1478{
1479 struct gfs2_file *fp = file->private_data;
1480 struct gfs2_holder *fl_gh = &fp->f_fl_gh;
1481 struct gfs2_inode *ip = GFS2_I(file_inode(file));
1482 struct gfs2_glock *gl;
1483 unsigned int state;
1484 u16 flags;
1485 int error = 0;
1486 int sleeptime;
1487
1488 state = lock_is_write(fl) ? LM_ST_EXCLUSIVE : LM_ST_SHARED;
1489 flags = GL_EXACT | GL_NOPID;
1490 if (!IS_SETLKW(cmd))
1491 flags |= LM_FLAG_TRY_1CB;
1492
1493 mutex_lock(&fp->f_fl_mutex);
1494
1495 if (gfs2_holder_initialized(fl_gh)) {
1496 struct file_lock request;
1497 if (fl_gh->gh_state == state)
1498 goto out;
1499 locks_init_lock(&request);
1500 request.c.flc_type = F_UNLCK;
1501 request.c.flc_flags = FL_FLOCK;
1502 locks_lock_file_wait(file, &request);
1503 gfs2_glock_dq(fl_gh);
1504 gfs2_holder_reinit(state, flags, fl_gh);
1505 } else {
1506 error = gfs2_glock_get(GFS2_SB(&ip->i_inode), ip->i_no_addr,
1507 &gfs2_flock_glops, CREATE, &gl);
1508 if (error)
1509 goto out;
1510 spin_lock(&file->f_lock);
1511 gfs2_holder_init(gl, state, flags, fl_gh);
1512 spin_unlock(&file->f_lock);
1513 gfs2_glock_put(gl);
1514 }
1515 for (sleeptime = 1; sleeptime <= 4; sleeptime <<= 1) {
1516 error = gfs2_glock_nq(fl_gh);
1517 if (error != GLR_TRYFAILED)
1518 break;
1519 fl_gh->gh_flags &= ~LM_FLAG_TRY_1CB;
1520 fl_gh->gh_flags |= LM_FLAG_TRY;
1521 msleep(sleeptime);
1522 }
1523 if (error) {
1524 __flock_holder_uninit(file, fl_gh);
1525 if (error == GLR_TRYFAILED)
1526 error = -EAGAIN;
1527 } else {
1528 error = locks_lock_file_wait(file, fl);
1529 gfs2_assert_warn(GFS2_SB(&ip->i_inode), !error);
1530 }
1531
1532out:
1533 mutex_unlock(&fp->f_fl_mutex);
1534 return error;
1535}
1536
1537static void do_unflock(struct file *file, struct file_lock *fl)
1538{
1539 struct gfs2_file *fp = file->private_data;
1540 struct gfs2_holder *fl_gh = &fp->f_fl_gh;
1541
1542 mutex_lock(&fp->f_fl_mutex);
1543 locks_lock_file_wait(file, fl);
1544 if (gfs2_holder_initialized(fl_gh)) {
1545 gfs2_glock_dq(fl_gh);
1546 __flock_holder_uninit(file, fl_gh);
1547 }
1548 mutex_unlock(&fp->f_fl_mutex);
1549}
1550
1551/**
1552 * gfs2_flock - acquire/release a flock lock on a file
1553 * @file: the file pointer
1554 * @cmd: either modify or retrieve lock state, possibly wait
1555 * @fl: type and range of lock
1556 *
1557 * Returns: errno
1558 */
1559
1560static int gfs2_flock(struct file *file, int cmd, struct file_lock *fl)
1561{
1562 if (!(fl->c.flc_flags & FL_FLOCK))
1563 return -ENOLCK;
1564
1565 if (lock_is_unlock(fl)) {
1566 do_unflock(file, fl);
1567 return 0;
1568 } else {
1569 return do_flock(file, cmd, fl);
1570 }
1571}
1572
1573const struct file_operations gfs2_file_fops = {
1574 .llseek = gfs2_llseek,
1575 .read_iter = gfs2_file_read_iter,
1576 .write_iter = gfs2_file_write_iter,
1577 .iopoll = iocb_bio_iopoll,
1578 .unlocked_ioctl = gfs2_ioctl,
1579 .compat_ioctl = gfs2_compat_ioctl,
1580 .mmap = gfs2_mmap,
1581 .open = gfs2_open,
1582 .release = gfs2_release,
1583 .fsync = gfs2_fsync,
1584 .lock = gfs2_lock,
1585 .flock = gfs2_flock,
1586 .splice_read = copy_splice_read,
1587 .splice_write = gfs2_file_splice_write,
1588 .setlease = simple_nosetlease,
1589 .fallocate = gfs2_fallocate,
1590 .fop_flags = FOP_ASYNC_LOCK,
1591};
1592
1593const struct file_operations gfs2_dir_fops = {
1594 .iterate_shared = gfs2_readdir,
1595 .unlocked_ioctl = gfs2_ioctl,
1596 .compat_ioctl = gfs2_compat_ioctl,
1597 .open = gfs2_open,
1598 .release = gfs2_release,
1599 .fsync = gfs2_fsync,
1600 .lock = gfs2_lock,
1601 .flock = gfs2_flock,
1602 .llseek = default_llseek,
1603 .fop_flags = FOP_ASYNC_LOCK,
1604};
1605
1606#endif /* CONFIG_GFS2_FS_LOCKING_DLM */
1607
1608const struct file_operations gfs2_file_fops_nolock = {
1609 .llseek = gfs2_llseek,
1610 .read_iter = gfs2_file_read_iter,
1611 .write_iter = gfs2_file_write_iter,
1612 .iopoll = iocb_bio_iopoll,
1613 .unlocked_ioctl = gfs2_ioctl,
1614 .compat_ioctl = gfs2_compat_ioctl,
1615 .mmap = gfs2_mmap,
1616 .open = gfs2_open,
1617 .release = gfs2_release,
1618 .fsync = gfs2_fsync,
1619 .splice_read = copy_splice_read,
1620 .splice_write = gfs2_file_splice_write,
1621 .setlease = generic_setlease,
1622 .fallocate = gfs2_fallocate,
1623};
1624
1625const struct file_operations gfs2_dir_fops_nolock = {
1626 .iterate_shared = gfs2_readdir,
1627 .unlocked_ioctl = gfs2_ioctl,
1628 .compat_ioctl = gfs2_compat_ioctl,
1629 .open = gfs2_open,
1630 .release = gfs2_release,
1631 .fsync = gfs2_fsync,
1632 .llseek = default_llseek,
1633};
1634