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/*
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