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
gfs2_llseek(struct file * file,loff_t offset,int whence)57 static 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
gfs2_readdir(struct file * file,struct dir_context * ctx)104 static 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 */
128 static 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
gfs2_gfsflags_to_fsflags(struct inode * inode,u32 gfsflags)141 static 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
gfs2_fileattr_get(struct dentry * dentry,struct fileattr * fa)157 int 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);
178 out_uninit:
179 gfs2_holder_uninit(&gh);
180 return error;
181 }
182
gfs2_set_inode_flags(struct inode * inode)183 void 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 */
do_gfs2_set_flags(struct inode * inode,u32 reqflags,u32 mask)218 static 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);
269 out_trans_end:
270 gfs2_trans_end(sdp);
271 out:
272 gfs2_glock_dq_uninit(&gh);
273 return error;
274 }
275
gfs2_fileattr_set(struct user_namespace * mnt_userns,struct dentry * dentry,struct fileattr * fa)276 int 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
gfs2_getlabel(struct file * filp,char __user * label)312 static 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
gfs2_ioctl(struct file * filp,unsigned int cmd,unsigned long arg)323 static 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
gfs2_compat_ioctl(struct file * filp,unsigned int cmd,unsigned long arg)336 static 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
gfs2_size_hint(struct file * filep,loff_t offset,size_t size)365 static 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 */
gfs2_allocate_page_backing(struct page * page,unsigned int length)387 static 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
gfs2_page_mkwrite(struct vm_fault * vmf)414 static 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
528 out_page_locked:
529 if (ret != VM_FAULT_LOCKED)
530 unlock_page(page);
531 out_trans_end:
532 gfs2_trans_end(sdp);
533 out_trans_fail:
534 gfs2_inplace_release(ip);
535 out_quota_unlock:
536 gfs2_quota_unlock(ip);
537 out_unlock:
538 gfs2_glock_dq(&gh);
539 out_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
gfs2_fault(struct vm_fault * vmf)549 static 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);
565 out_uninit:
566 gfs2_holder_uninit(&gh);
567 return ret;
568 }
569
570 static 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
gfs2_mmap(struct file * file,struct vm_area_struct * vma)588 static 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
gfs2_open_common(struct inode * inode,struct file * file)623 int 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
649 fail:
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
gfs2_open(struct inode * inode,struct file * file)669 static 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
gfs2_release(struct inode * inode,struct file * file)700 static 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 (gfs2_rs_active(&ip->i_res))
708 gfs2_rs_delete(ip, &inode->i_writecount);
709 if (file->f_mode & FMODE_WRITE)
710 gfs2_qa_put(ip);
711 return 0;
712 }
713
714 /**
715 * gfs2_fsync - sync the dirty data for a file (across the cluster)
716 * @file: the file that points to the dentry
717 * @start: the start position in the file to sync
718 * @end: the end position in the file to sync
719 * @datasync: set if we can ignore timestamp changes
720 *
721 * We split the data flushing here so that we don't wait for the data
722 * until after we've also sent the metadata to disk. Note that for
723 * data=ordered, we will write & wait for the data at the log flush
724 * stage anyway, so this is unlikely to make much of a difference
725 * except in the data=writeback case.
726 *
727 * If the fdatawrite fails due to any reason except -EIO, we will
728 * continue the remainder of the fsync, although we'll still report
729 * the error at the end. This is to match filemap_write_and_wait_range()
730 * behaviour.
731 *
732 * Returns: errno
733 */
734
gfs2_fsync(struct file * file,loff_t start,loff_t end,int datasync)735 static int gfs2_fsync(struct file *file, loff_t start, loff_t end,
736 int datasync)
737 {
738 struct address_space *mapping = file->f_mapping;
739 struct inode *inode = mapping->host;
740 int sync_state = inode->i_state & I_DIRTY;
741 struct gfs2_inode *ip = GFS2_I(inode);
742 int ret = 0, ret1 = 0;
743
744 if (mapping->nrpages) {
745 ret1 = filemap_fdatawrite_range(mapping, start, end);
746 if (ret1 == -EIO)
747 return ret1;
748 }
749
750 if (!gfs2_is_jdata(ip))
751 sync_state &= ~I_DIRTY_PAGES;
752 if (datasync)
753 sync_state &= ~I_DIRTY_SYNC;
754
755 if (sync_state) {
756 ret = sync_inode_metadata(inode, 1);
757 if (ret)
758 return ret;
759 if (gfs2_is_jdata(ip))
760 ret = file_write_and_wait(file);
761 if (ret)
762 return ret;
763 gfs2_ail_flush(ip->i_gl, 1);
764 }
765
766 if (mapping->nrpages)
767 ret = file_fdatawait_range(file, start, end);
768
769 return ret ? ret : ret1;
770 }
771
should_fault_in_pages(ssize_t ret,struct iov_iter * i,size_t * prev_count,size_t * window_size)772 static inline bool should_fault_in_pages(ssize_t ret, struct iov_iter *i,
773 size_t *prev_count,
774 size_t *window_size)
775 {
776 size_t count = iov_iter_count(i);
777 char __user *p;
778 int pages = 1;
779
780 if (likely(!count))
781 return false;
782 if (ret <= 0 && ret != -EFAULT)
783 return false;
784 if (!iter_is_iovec(i))
785 return false;
786
787 if (*prev_count != count || !*window_size) {
788 int pages, nr_dirtied;
789
790 pages = min_t(int, BIO_MAX_VECS, DIV_ROUND_UP(count, PAGE_SIZE));
791 nr_dirtied = max(current->nr_dirtied_pause -
792 current->nr_dirtied, 1);
793 pages = min(pages, nr_dirtied);
794 }
795
796 *prev_count = count;
797 p = i->iov[0].iov_base + i->iov_offset;
798 *window_size = (size_t)PAGE_SIZE * pages - offset_in_page(p);
799 return true;
800 }
801
gfs2_file_direct_read(struct kiocb * iocb,struct iov_iter * to,struct gfs2_holder * gh)802 static ssize_t gfs2_file_direct_read(struct kiocb *iocb, struct iov_iter *to,
803 struct gfs2_holder *gh)
804 {
805 struct file *file = iocb->ki_filp;
806 struct gfs2_inode *ip = GFS2_I(file->f_mapping->host);
807 size_t prev_count = 0, window_size = 0;
808 size_t written = 0;
809 ssize_t ret;
810
811 /*
812 * In this function, we disable page faults when we're holding the
813 * inode glock while doing I/O. If a page fault occurs, we indicate
814 * that the inode glock may be dropped, fault in the pages manually,
815 * and retry.
816 *
817 * Unlike generic_file_read_iter, for reads, iomap_dio_rw can trigger
818 * physical as well as manual page faults, and we need to disable both
819 * kinds.
820 *
821 * For direct I/O, gfs2 takes the inode glock in deferred mode. This
822 * locking mode is compatible with other deferred holders, so multiple
823 * processes and nodes can do direct I/O to a file at the same time.
824 * There's no guarantee that reads or writes will be atomic. Any
825 * coordination among readers and writers needs to happen externally.
826 */
827
828 if (!iov_iter_count(to))
829 return 0; /* skip atime */
830
831 gfs2_holder_init(ip->i_gl, LM_ST_DEFERRED, 0, gh);
832 retry:
833 ret = gfs2_glock_nq(gh);
834 if (ret)
835 goto out_uninit;
836 retry_under_glock:
837 pagefault_disable();
838 to->nofault = true;
839 ret = iomap_dio_rw(iocb, to, &gfs2_iomap_ops, NULL,
840 IOMAP_DIO_PARTIAL, written);
841 to->nofault = false;
842 pagefault_enable();
843 if (ret > 0)
844 written = ret;
845
846 if (should_fault_in_pages(ret, to, &prev_count, &window_size)) {
847 size_t leftover;
848
849 gfs2_holder_allow_demote(gh);
850 leftover = fault_in_iov_iter_writeable(to, window_size);
851 gfs2_holder_disallow_demote(gh);
852 if (leftover != window_size) {
853 if (!gfs2_holder_queued(gh))
854 goto retry;
855 goto retry_under_glock;
856 }
857 }
858 if (gfs2_holder_queued(gh))
859 gfs2_glock_dq(gh);
860 out_uninit:
861 gfs2_holder_uninit(gh);
862 if (ret < 0)
863 return ret;
864 return written;
865 }
866
gfs2_file_direct_write(struct kiocb * iocb,struct iov_iter * from,struct gfs2_holder * gh)867 static ssize_t gfs2_file_direct_write(struct kiocb *iocb, struct iov_iter *from,
868 struct gfs2_holder *gh)
869 {
870 struct file *file = iocb->ki_filp;
871 struct inode *inode = file->f_mapping->host;
872 struct gfs2_inode *ip = GFS2_I(inode);
873 size_t prev_count = 0, window_size = 0;
874 size_t read = 0;
875 ssize_t ret;
876
877 /*
878 * In this function, we disable page faults when we're holding the
879 * inode glock while doing I/O. If a page fault occurs, we indicate
880 * that the inode glock may be dropped, fault in the pages manually,
881 * and retry.
882 *
883 * For writes, iomap_dio_rw only triggers manual page faults, so we
884 * don't need to disable physical ones.
885 */
886
887 /*
888 * Deferred lock, even if its a write, since we do no allocation on
889 * this path. All we need to change is the atime, and this lock mode
890 * ensures that other nodes have flushed their buffered read caches
891 * (i.e. their page cache entries for this inode). We do not,
892 * unfortunately, have the option of only flushing a range like the
893 * VFS does.
894 */
895 gfs2_holder_init(ip->i_gl, LM_ST_DEFERRED, 0, gh);
896 retry:
897 ret = gfs2_glock_nq(gh);
898 if (ret)
899 goto out_uninit;
900 retry_under_glock:
901 /* Silently fall back to buffered I/O when writing beyond EOF */
902 if (iocb->ki_pos + iov_iter_count(from) > i_size_read(&ip->i_inode))
903 goto out;
904
905 from->nofault = true;
906 ret = iomap_dio_rw(iocb, from, &gfs2_iomap_ops, NULL,
907 IOMAP_DIO_PARTIAL, read);
908 from->nofault = false;
909
910 if (ret == -ENOTBLK)
911 ret = 0;
912 if (ret > 0)
913 read = ret;
914
915 if (should_fault_in_pages(ret, from, &prev_count, &window_size)) {
916 size_t leftover;
917
918 gfs2_holder_allow_demote(gh);
919 leftover = fault_in_iov_iter_readable(from, window_size);
920 gfs2_holder_disallow_demote(gh);
921 if (leftover != window_size) {
922 if (!gfs2_holder_queued(gh))
923 goto retry;
924 goto retry_under_glock;
925 }
926 }
927 out:
928 if (gfs2_holder_queued(gh))
929 gfs2_glock_dq(gh);
930 out_uninit:
931 gfs2_holder_uninit(gh);
932 if (ret < 0)
933 return ret;
934 return read;
935 }
936
gfs2_file_read_iter(struct kiocb * iocb,struct iov_iter * to)937 static ssize_t gfs2_file_read_iter(struct kiocb *iocb, struct iov_iter *to)
938 {
939 struct gfs2_inode *ip;
940 struct gfs2_holder gh;
941 size_t prev_count = 0, window_size = 0;
942 size_t written = 0;
943 ssize_t ret;
944
945 /*
946 * In this function, we disable page faults when we're holding the
947 * inode glock while doing I/O. If a page fault occurs, we indicate
948 * that the inode glock may be dropped, fault in the pages manually,
949 * and retry.
950 */
951
952 if (iocb->ki_flags & IOCB_DIRECT) {
953 ret = gfs2_file_direct_read(iocb, to, &gh);
954 if (likely(ret != -ENOTBLK))
955 return ret;
956 iocb->ki_flags &= ~IOCB_DIRECT;
957 }
958 iocb->ki_flags |= IOCB_NOIO;
959 ret = generic_file_read_iter(iocb, to);
960 iocb->ki_flags &= ~IOCB_NOIO;
961 if (ret >= 0) {
962 if (!iov_iter_count(to))
963 return ret;
964 written = ret;
965 } else {
966 if (ret != -EAGAIN)
967 return ret;
968 if (iocb->ki_flags & IOCB_NOWAIT)
969 return ret;
970 }
971 ip = GFS2_I(iocb->ki_filp->f_mapping->host);
972 gfs2_holder_init(ip->i_gl, LM_ST_SHARED, 0, &gh);
973 retry:
974 ret = gfs2_glock_nq(&gh);
975 if (ret)
976 goto out_uninit;
977 retry_under_glock:
978 pagefault_disable();
979 ret = generic_file_read_iter(iocb, to);
980 pagefault_enable();
981 if (ret > 0)
982 written += ret;
983
984 if (should_fault_in_pages(ret, to, &prev_count, &window_size)) {
985 size_t leftover;
986
987 gfs2_holder_allow_demote(&gh);
988 leftover = fault_in_iov_iter_writeable(to, window_size);
989 gfs2_holder_disallow_demote(&gh);
990 if (leftover != window_size) {
991 if (!gfs2_holder_queued(&gh)) {
992 if (written)
993 goto out_uninit;
994 goto retry;
995 }
996 goto retry_under_glock;
997 }
998 }
999 if (gfs2_holder_queued(&gh))
1000 gfs2_glock_dq(&gh);
1001 out_uninit:
1002 gfs2_holder_uninit(&gh);
1003 return written ? written : ret;
1004 }
1005
gfs2_file_buffered_write(struct kiocb * iocb,struct iov_iter * from,struct gfs2_holder * gh)1006 static ssize_t gfs2_file_buffered_write(struct kiocb *iocb,
1007 struct iov_iter *from,
1008 struct gfs2_holder *gh)
1009 {
1010 struct file *file = iocb->ki_filp;
1011 struct inode *inode = file_inode(file);
1012 struct gfs2_inode *ip = GFS2_I(inode);
1013 struct gfs2_sbd *sdp = GFS2_SB(inode);
1014 struct gfs2_holder *statfs_gh = NULL;
1015 size_t prev_count = 0, window_size = 0;
1016 size_t orig_count = iov_iter_count(from);
1017 size_t read = 0;
1018 ssize_t ret;
1019
1020 /*
1021 * In this function, we disable page faults when we're holding the
1022 * inode glock while doing I/O. If a page fault occurs, we indicate
1023 * that the inode glock may be dropped, fault in the pages manually,
1024 * and retry.
1025 */
1026
1027 if (inode == sdp->sd_rindex) {
1028 statfs_gh = kmalloc(sizeof(*statfs_gh), GFP_NOFS);
1029 if (!statfs_gh)
1030 return -ENOMEM;
1031 }
1032
1033 gfs2_holder_init(ip->i_gl, LM_ST_EXCLUSIVE, 0, gh);
1034 retry:
1035 ret = gfs2_glock_nq(gh);
1036 if (ret)
1037 goto out_uninit;
1038 retry_under_glock:
1039 if (inode == sdp->sd_rindex) {
1040 struct gfs2_inode *m_ip = GFS2_I(sdp->sd_statfs_inode);
1041
1042 ret = gfs2_glock_nq_init(m_ip->i_gl, LM_ST_EXCLUSIVE,
1043 GL_NOCACHE, statfs_gh);
1044 if (ret)
1045 goto out_unlock;
1046 }
1047
1048 current->backing_dev_info = inode_to_bdi(inode);
1049 pagefault_disable();
1050 ret = iomap_file_buffered_write(iocb, from, &gfs2_iomap_ops);
1051 pagefault_enable();
1052 current->backing_dev_info = NULL;
1053 if (ret > 0) {
1054 iocb->ki_pos += ret;
1055 read += ret;
1056 }
1057
1058 if (inode == sdp->sd_rindex)
1059 gfs2_glock_dq_uninit(statfs_gh);
1060
1061 from->count = orig_count - read;
1062 if (should_fault_in_pages(ret, from, &prev_count, &window_size)) {
1063 size_t leftover;
1064
1065 gfs2_holder_allow_demote(gh);
1066 leftover = fault_in_iov_iter_readable(from, window_size);
1067 gfs2_holder_disallow_demote(gh);
1068 if (leftover != window_size) {
1069 from->count = min(from->count, window_size - leftover);
1070 if (!gfs2_holder_queued(gh)) {
1071 if (read)
1072 goto out_uninit;
1073 goto retry;
1074 }
1075 goto retry_under_glock;
1076 }
1077 }
1078 out_unlock:
1079 if (gfs2_holder_queued(gh))
1080 gfs2_glock_dq(gh);
1081 out_uninit:
1082 gfs2_holder_uninit(gh);
1083 if (statfs_gh)
1084 kfree(statfs_gh);
1085 return read ? read : ret;
1086 }
1087
1088 /**
1089 * gfs2_file_write_iter - Perform a write to a file
1090 * @iocb: The io context
1091 * @from: The data to write
1092 *
1093 * We have to do a lock/unlock here to refresh the inode size for
1094 * O_APPEND writes, otherwise we can land up writing at the wrong
1095 * offset. There is still a race, but provided the app is using its
1096 * own file locking, this will make O_APPEND work as expected.
1097 *
1098 */
1099
gfs2_file_write_iter(struct kiocb * iocb,struct iov_iter * from)1100 static ssize_t gfs2_file_write_iter(struct kiocb *iocb, struct iov_iter *from)
1101 {
1102 struct file *file = iocb->ki_filp;
1103 struct inode *inode = file_inode(file);
1104 struct gfs2_inode *ip = GFS2_I(inode);
1105 struct gfs2_holder gh;
1106 ssize_t ret;
1107
1108 gfs2_size_hint(file, iocb->ki_pos, iov_iter_count(from));
1109
1110 if (iocb->ki_flags & IOCB_APPEND) {
1111 ret = gfs2_glock_nq_init(ip->i_gl, LM_ST_SHARED, 0, &gh);
1112 if (ret)
1113 return ret;
1114 gfs2_glock_dq_uninit(&gh);
1115 }
1116
1117 inode_lock(inode);
1118 ret = generic_write_checks(iocb, from);
1119 if (ret <= 0)
1120 goto out_unlock;
1121
1122 ret = file_remove_privs(file);
1123 if (ret)
1124 goto out_unlock;
1125
1126 ret = file_update_time(file);
1127 if (ret)
1128 goto out_unlock;
1129
1130 if (iocb->ki_flags & IOCB_DIRECT) {
1131 struct address_space *mapping = file->f_mapping;
1132 ssize_t buffered, ret2;
1133
1134 ret = gfs2_file_direct_write(iocb, from, &gh);
1135 if (ret < 0 || !iov_iter_count(from))
1136 goto out_unlock;
1137
1138 iocb->ki_flags |= IOCB_DSYNC;
1139 buffered = gfs2_file_buffered_write(iocb, from, &gh);
1140 if (unlikely(buffered <= 0)) {
1141 if (!ret)
1142 ret = buffered;
1143 goto out_unlock;
1144 }
1145
1146 /*
1147 * We need to ensure that the page cache pages are written to
1148 * disk and invalidated to preserve the expected O_DIRECT
1149 * semantics. If the writeback or invalidate fails, only report
1150 * the direct I/O range as we don't know if the buffered pages
1151 * made it to disk.
1152 */
1153 ret2 = generic_write_sync(iocb, buffered);
1154 invalidate_mapping_pages(mapping,
1155 (iocb->ki_pos - buffered) >> PAGE_SHIFT,
1156 (iocb->ki_pos - 1) >> PAGE_SHIFT);
1157 if (!ret || ret2 > 0)
1158 ret += ret2;
1159 } else {
1160 ret = gfs2_file_buffered_write(iocb, from, &gh);
1161 if (likely(ret > 0))
1162 ret = generic_write_sync(iocb, ret);
1163 }
1164
1165 out_unlock:
1166 inode_unlock(inode);
1167 return ret;
1168 }
1169
fallocate_chunk(struct inode * inode,loff_t offset,loff_t len,int mode)1170 static int fallocate_chunk(struct inode *inode, loff_t offset, loff_t len,
1171 int mode)
1172 {
1173 struct super_block *sb = inode->i_sb;
1174 struct gfs2_inode *ip = GFS2_I(inode);
1175 loff_t end = offset + len;
1176 struct buffer_head *dibh;
1177 int error;
1178
1179 error = gfs2_meta_inode_buffer(ip, &dibh);
1180 if (unlikely(error))
1181 return error;
1182
1183 gfs2_trans_add_meta(ip->i_gl, dibh);
1184
1185 if (gfs2_is_stuffed(ip)) {
1186 error = gfs2_unstuff_dinode(ip);
1187 if (unlikely(error))
1188 goto out;
1189 }
1190
1191 while (offset < end) {
1192 struct iomap iomap = { };
1193
1194 error = gfs2_iomap_alloc(inode, offset, end - offset, &iomap);
1195 if (error)
1196 goto out;
1197 offset = iomap.offset + iomap.length;
1198 if (!(iomap.flags & IOMAP_F_NEW))
1199 continue;
1200 error = sb_issue_zeroout(sb, iomap.addr >> inode->i_blkbits,
1201 iomap.length >> inode->i_blkbits,
1202 GFP_NOFS);
1203 if (error) {
1204 fs_err(GFS2_SB(inode), "Failed to zero data buffers\n");
1205 goto out;
1206 }
1207 }
1208 out:
1209 brelse(dibh);
1210 return error;
1211 }
1212
1213 /**
1214 * calc_max_reserv() - Reverse of write_calc_reserv. Given a number of
1215 * blocks, determine how many bytes can be written.
1216 * @ip: The inode in question.
1217 * @len: Max cap of bytes. What we return in *len must be <= this.
1218 * @data_blocks: Compute and return the number of data blocks needed
1219 * @ind_blocks: Compute and return the number of indirect blocks needed
1220 * @max_blocks: The total blocks available to work with.
1221 *
1222 * Returns: void, but @len, @data_blocks and @ind_blocks are filled in.
1223 */
calc_max_reserv(struct gfs2_inode * ip,loff_t * len,unsigned int * data_blocks,unsigned int * ind_blocks,unsigned int max_blocks)1224 static void calc_max_reserv(struct gfs2_inode *ip, loff_t *len,
1225 unsigned int *data_blocks, unsigned int *ind_blocks,
1226 unsigned int max_blocks)
1227 {
1228 loff_t max = *len;
1229 const struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode);
1230 unsigned int tmp, max_data = max_blocks - 3 * (sdp->sd_max_height - 1);
1231
1232 for (tmp = max_data; tmp > sdp->sd_diptrs;) {
1233 tmp = DIV_ROUND_UP(tmp, sdp->sd_inptrs);
1234 max_data -= tmp;
1235 }
1236
1237 *data_blocks = max_data;
1238 *ind_blocks = max_blocks - max_data;
1239 *len = ((loff_t)max_data - 3) << sdp->sd_sb.sb_bsize_shift;
1240 if (*len > max) {
1241 *len = max;
1242 gfs2_write_calc_reserv(ip, max, data_blocks, ind_blocks);
1243 }
1244 }
1245
__gfs2_fallocate(struct file * file,int mode,loff_t offset,loff_t len)1246 static long __gfs2_fallocate(struct file *file, int mode, loff_t offset, loff_t len)
1247 {
1248 struct inode *inode = file_inode(file);
1249 struct gfs2_sbd *sdp = GFS2_SB(inode);
1250 struct gfs2_inode *ip = GFS2_I(inode);
1251 struct gfs2_alloc_parms ap = { .aflags = 0, };
1252 unsigned int data_blocks = 0, ind_blocks = 0, rblocks;
1253 loff_t bytes, max_bytes, max_blks;
1254 int error;
1255 const loff_t pos = offset;
1256 const loff_t count = len;
1257 loff_t bsize_mask = ~((loff_t)sdp->sd_sb.sb_bsize - 1);
1258 loff_t next = (offset + len - 1) >> sdp->sd_sb.sb_bsize_shift;
1259 loff_t max_chunk_size = UINT_MAX & bsize_mask;
1260
1261 next = (next + 1) << sdp->sd_sb.sb_bsize_shift;
1262
1263 offset &= bsize_mask;
1264
1265 len = next - offset;
1266 bytes = sdp->sd_max_rg_data * sdp->sd_sb.sb_bsize / 2;
1267 if (!bytes)
1268 bytes = UINT_MAX;
1269 bytes &= bsize_mask;
1270 if (bytes == 0)
1271 bytes = sdp->sd_sb.sb_bsize;
1272
1273 gfs2_size_hint(file, offset, len);
1274
1275 gfs2_write_calc_reserv(ip, PAGE_SIZE, &data_blocks, &ind_blocks);
1276 ap.min_target = data_blocks + ind_blocks;
1277
1278 while (len > 0) {
1279 if (len < bytes)
1280 bytes = len;
1281 if (!gfs2_write_alloc_required(ip, offset, bytes)) {
1282 len -= bytes;
1283 offset += bytes;
1284 continue;
1285 }
1286
1287 /* We need to determine how many bytes we can actually
1288 * fallocate without exceeding quota or going over the
1289 * end of the fs. We start off optimistically by assuming
1290 * we can write max_bytes */
1291 max_bytes = (len > max_chunk_size) ? max_chunk_size : len;
1292
1293 /* Since max_bytes is most likely a theoretical max, we
1294 * calculate a more realistic 'bytes' to serve as a good
1295 * starting point for the number of bytes we may be able
1296 * to write */
1297 gfs2_write_calc_reserv(ip, bytes, &data_blocks, &ind_blocks);
1298 ap.target = data_blocks + ind_blocks;
1299
1300 error = gfs2_quota_lock_check(ip, &ap);
1301 if (error)
1302 return error;
1303 /* ap.allowed tells us how many blocks quota will allow
1304 * us to write. Check if this reduces max_blks */
1305 max_blks = UINT_MAX;
1306 if (ap.allowed)
1307 max_blks = ap.allowed;
1308
1309 error = gfs2_inplace_reserve(ip, &ap);
1310 if (error)
1311 goto out_qunlock;
1312
1313 /* check if the selected rgrp limits our max_blks further */
1314 if (ip->i_res.rs_reserved < max_blks)
1315 max_blks = ip->i_res.rs_reserved;
1316
1317 /* Almost done. Calculate bytes that can be written using
1318 * max_blks. We also recompute max_bytes, data_blocks and
1319 * ind_blocks */
1320 calc_max_reserv(ip, &max_bytes, &data_blocks,
1321 &ind_blocks, max_blks);
1322
1323 rblocks = RES_DINODE + ind_blocks + RES_STATFS + RES_QUOTA +
1324 RES_RG_HDR + gfs2_rg_blocks(ip, data_blocks + ind_blocks);
1325 if (gfs2_is_jdata(ip))
1326 rblocks += data_blocks ? data_blocks : 1;
1327
1328 error = gfs2_trans_begin(sdp, rblocks,
1329 PAGE_SIZE >> inode->i_blkbits);
1330 if (error)
1331 goto out_trans_fail;
1332
1333 error = fallocate_chunk(inode, offset, max_bytes, mode);
1334 gfs2_trans_end(sdp);
1335
1336 if (error)
1337 goto out_trans_fail;
1338
1339 len -= max_bytes;
1340 offset += max_bytes;
1341 gfs2_inplace_release(ip);
1342 gfs2_quota_unlock(ip);
1343 }
1344
1345 if (!(mode & FALLOC_FL_KEEP_SIZE) && (pos + count) > inode->i_size)
1346 i_size_write(inode, pos + count);
1347 file_update_time(file);
1348 mark_inode_dirty(inode);
1349
1350 if ((file->f_flags & O_DSYNC) || IS_SYNC(file->f_mapping->host))
1351 return vfs_fsync_range(file, pos, pos + count - 1,
1352 (file->f_flags & __O_SYNC) ? 0 : 1);
1353 return 0;
1354
1355 out_trans_fail:
1356 gfs2_inplace_release(ip);
1357 out_qunlock:
1358 gfs2_quota_unlock(ip);
1359 return error;
1360 }
1361
gfs2_fallocate(struct file * file,int mode,loff_t offset,loff_t len)1362 static long gfs2_fallocate(struct file *file, int mode, loff_t offset, loff_t len)
1363 {
1364 struct inode *inode = file_inode(file);
1365 struct gfs2_sbd *sdp = GFS2_SB(inode);
1366 struct gfs2_inode *ip = GFS2_I(inode);
1367 struct gfs2_holder gh;
1368 int ret;
1369
1370 if (mode & ~(FALLOC_FL_PUNCH_HOLE | FALLOC_FL_KEEP_SIZE))
1371 return -EOPNOTSUPP;
1372 /* fallocate is needed by gfs2_grow to reserve space in the rindex */
1373 if (gfs2_is_jdata(ip) && inode != sdp->sd_rindex)
1374 return -EOPNOTSUPP;
1375
1376 inode_lock(inode);
1377
1378 gfs2_holder_init(ip->i_gl, LM_ST_EXCLUSIVE, 0, &gh);
1379 ret = gfs2_glock_nq(&gh);
1380 if (ret)
1381 goto out_uninit;
1382
1383 if (!(mode & FALLOC_FL_KEEP_SIZE) &&
1384 (offset + len) > inode->i_size) {
1385 ret = inode_newsize_ok(inode, offset + len);
1386 if (ret)
1387 goto out_unlock;
1388 }
1389
1390 ret = get_write_access(inode);
1391 if (ret)
1392 goto out_unlock;
1393
1394 if (mode & FALLOC_FL_PUNCH_HOLE) {
1395 ret = __gfs2_punch_hole(file, offset, len);
1396 } else {
1397 ret = __gfs2_fallocate(file, mode, offset, len);
1398 if (ret)
1399 gfs2_rs_deltree(&ip->i_res);
1400 }
1401
1402 put_write_access(inode);
1403 out_unlock:
1404 gfs2_glock_dq(&gh);
1405 out_uninit:
1406 gfs2_holder_uninit(&gh);
1407 inode_unlock(inode);
1408 return ret;
1409 }
1410
gfs2_file_splice_write(struct pipe_inode_info * pipe,struct file * out,loff_t * ppos,size_t len,unsigned int flags)1411 static ssize_t gfs2_file_splice_write(struct pipe_inode_info *pipe,
1412 struct file *out, loff_t *ppos,
1413 size_t len, unsigned int flags)
1414 {
1415 ssize_t ret;
1416
1417 gfs2_size_hint(out, *ppos, len);
1418
1419 ret = iter_file_splice_write(pipe, out, ppos, len, flags);
1420 return ret;
1421 }
1422
1423 #ifdef CONFIG_GFS2_FS_LOCKING_DLM
1424
1425 /**
1426 * gfs2_lock - acquire/release a posix lock on a file
1427 * @file: the file pointer
1428 * @cmd: either modify or retrieve lock state, possibly wait
1429 * @fl: type and range of lock
1430 *
1431 * Returns: errno
1432 */
1433
gfs2_lock(struct file * file,int cmd,struct file_lock * fl)1434 static int gfs2_lock(struct file *file, int cmd, struct file_lock *fl)
1435 {
1436 struct gfs2_inode *ip = GFS2_I(file->f_mapping->host);
1437 struct gfs2_sbd *sdp = GFS2_SB(file->f_mapping->host);
1438 struct lm_lockstruct *ls = &sdp->sd_lockstruct;
1439
1440 if (!(fl->fl_flags & FL_POSIX))
1441 return -ENOLCK;
1442 if (cmd == F_CANCELLK) {
1443 /* Hack: */
1444 cmd = F_SETLK;
1445 fl->fl_type = F_UNLCK;
1446 }
1447 if (unlikely(gfs2_withdrawn(sdp))) {
1448 if (fl->fl_type == F_UNLCK)
1449 locks_lock_file_wait(file, fl);
1450 return -EIO;
1451 }
1452 if (IS_GETLK(cmd))
1453 return dlm_posix_get(ls->ls_dlm, ip->i_no_addr, file, fl);
1454 else if (fl->fl_type == F_UNLCK)
1455 return dlm_posix_unlock(ls->ls_dlm, ip->i_no_addr, file, fl);
1456 else
1457 return dlm_posix_lock(ls->ls_dlm, ip->i_no_addr, file, cmd, fl);
1458 }
1459
do_flock(struct file * file,int cmd,struct file_lock * fl)1460 static int do_flock(struct file *file, int cmd, struct file_lock *fl)
1461 {
1462 struct gfs2_file *fp = file->private_data;
1463 struct gfs2_holder *fl_gh = &fp->f_fl_gh;
1464 struct gfs2_inode *ip = GFS2_I(file_inode(file));
1465 struct gfs2_glock *gl;
1466 unsigned int state;
1467 u16 flags;
1468 int error = 0;
1469 int sleeptime;
1470
1471 state = (fl->fl_type == F_WRLCK) ? LM_ST_EXCLUSIVE : LM_ST_SHARED;
1472 flags = (IS_SETLKW(cmd) ? 0 : LM_FLAG_TRY_1CB) | GL_EXACT;
1473
1474 mutex_lock(&fp->f_fl_mutex);
1475
1476 if (gfs2_holder_initialized(fl_gh)) {
1477 struct file_lock request;
1478 if (fl_gh->gh_state == state)
1479 goto out;
1480 locks_init_lock(&request);
1481 request.fl_type = F_UNLCK;
1482 request.fl_flags = FL_FLOCK;
1483 locks_lock_file_wait(file, &request);
1484 gfs2_glock_dq(fl_gh);
1485 gfs2_holder_reinit(state, flags, fl_gh);
1486 } else {
1487 error = gfs2_glock_get(GFS2_SB(&ip->i_inode), ip->i_no_addr,
1488 &gfs2_flock_glops, CREATE, &gl);
1489 if (error)
1490 goto out;
1491 gfs2_holder_init(gl, state, flags, fl_gh);
1492 gfs2_glock_put(gl);
1493 }
1494 for (sleeptime = 1; sleeptime <= 4; sleeptime <<= 1) {
1495 error = gfs2_glock_nq(fl_gh);
1496 if (error != GLR_TRYFAILED)
1497 break;
1498 fl_gh->gh_flags = LM_FLAG_TRY | GL_EXACT;
1499 fl_gh->gh_error = 0;
1500 msleep(sleeptime);
1501 }
1502 if (error) {
1503 gfs2_holder_uninit(fl_gh);
1504 if (error == GLR_TRYFAILED)
1505 error = -EAGAIN;
1506 } else {
1507 error = locks_lock_file_wait(file, fl);
1508 gfs2_assert_warn(GFS2_SB(&ip->i_inode), !error);
1509 }
1510
1511 out:
1512 mutex_unlock(&fp->f_fl_mutex);
1513 return error;
1514 }
1515
do_unflock(struct file * file,struct file_lock * fl)1516 static void do_unflock(struct file *file, struct file_lock *fl)
1517 {
1518 struct gfs2_file *fp = file->private_data;
1519 struct gfs2_holder *fl_gh = &fp->f_fl_gh;
1520
1521 mutex_lock(&fp->f_fl_mutex);
1522 locks_lock_file_wait(file, fl);
1523 if (gfs2_holder_initialized(fl_gh)) {
1524 gfs2_glock_dq(fl_gh);
1525 gfs2_holder_uninit(fl_gh);
1526 }
1527 mutex_unlock(&fp->f_fl_mutex);
1528 }
1529
1530 /**
1531 * gfs2_flock - acquire/release a flock lock on a file
1532 * @file: the file pointer
1533 * @cmd: either modify or retrieve lock state, possibly wait
1534 * @fl: type and range of lock
1535 *
1536 * Returns: errno
1537 */
1538
gfs2_flock(struct file * file,int cmd,struct file_lock * fl)1539 static int gfs2_flock(struct file *file, int cmd, struct file_lock *fl)
1540 {
1541 if (!(fl->fl_flags & FL_FLOCK))
1542 return -ENOLCK;
1543
1544 if (fl->fl_type == F_UNLCK) {
1545 do_unflock(file, fl);
1546 return 0;
1547 } else {
1548 return do_flock(file, cmd, fl);
1549 }
1550 }
1551
1552 const struct file_operations gfs2_file_fops = {
1553 .llseek = gfs2_llseek,
1554 .read_iter = gfs2_file_read_iter,
1555 .write_iter = gfs2_file_write_iter,
1556 .iopoll = iocb_bio_iopoll,
1557 .unlocked_ioctl = gfs2_ioctl,
1558 .compat_ioctl = gfs2_compat_ioctl,
1559 .mmap = gfs2_mmap,
1560 .open = gfs2_open,
1561 .release = gfs2_release,
1562 .fsync = gfs2_fsync,
1563 .lock = gfs2_lock,
1564 .flock = gfs2_flock,
1565 .splice_read = generic_file_splice_read,
1566 .splice_write = gfs2_file_splice_write,
1567 .setlease = simple_nosetlease,
1568 .fallocate = gfs2_fallocate,
1569 };
1570
1571 const struct file_operations gfs2_dir_fops = {
1572 .iterate_shared = gfs2_readdir,
1573 .unlocked_ioctl = gfs2_ioctl,
1574 .compat_ioctl = gfs2_compat_ioctl,
1575 .open = gfs2_open,
1576 .release = gfs2_release,
1577 .fsync = gfs2_fsync,
1578 .lock = gfs2_lock,
1579 .flock = gfs2_flock,
1580 .llseek = default_llseek,
1581 };
1582
1583 #endif /* CONFIG_GFS2_FS_LOCKING_DLM */
1584
1585 const struct file_operations gfs2_file_fops_nolock = {
1586 .llseek = gfs2_llseek,
1587 .read_iter = gfs2_file_read_iter,
1588 .write_iter = gfs2_file_write_iter,
1589 .iopoll = iocb_bio_iopoll,
1590 .unlocked_ioctl = gfs2_ioctl,
1591 .compat_ioctl = gfs2_compat_ioctl,
1592 .mmap = gfs2_mmap,
1593 .open = gfs2_open,
1594 .release = gfs2_release,
1595 .fsync = gfs2_fsync,
1596 .splice_read = generic_file_splice_read,
1597 .splice_write = gfs2_file_splice_write,
1598 .setlease = generic_setlease,
1599 .fallocate = gfs2_fallocate,
1600 };
1601
1602 const struct file_operations gfs2_dir_fops_nolock = {
1603 .iterate_shared = gfs2_readdir,
1604 .unlocked_ioctl = gfs2_ioctl,
1605 .compat_ioctl = gfs2_compat_ioctl,
1606 .open = gfs2_open,
1607 .release = gfs2_release,
1608 .fsync = gfs2_fsync,
1609 .llseek = default_llseek,
1610 };
1611
1612