1 // SPDX-License-Identifier: GPL-2.0-only
2 /*
3 * Copyright (C) Sistina Software, Inc. 1997-2003 All rights reserved.
4 * Copyright (C) 2004-2006 Red Hat, Inc. All rights reserved.
5 */
6
7 #include <linux/slab.h>
8 #include <linux/spinlock.h>
9 #include <linux/compat.h>
10 #include <linux/completion.h>
11 #include <linux/buffer_head.h>
12 #include <linux/pagemap.h>
13 #include <linux/uio.h>
14 #include <linux/blkdev.h>
15 #include <linux/mm.h>
16 #include <linux/mount.h>
17 #include <linux/fs.h>
18 #include <linux/gfs2_ondisk.h>
19 #include <linux/falloc.h>
20 #include <linux/swap.h>
21 #include <linux/crc32.h>
22 #include <linux/writeback.h>
23 #include <linux/uaccess.h>
24 #include <linux/dlm.h>
25 #include <linux/dlm_plock.h>
26 #include <linux/delay.h>
27 #include <linux/backing-dev.h>
28
29 #include "gfs2.h"
30 #include "incore.h"
31 #include "bmap.h"
32 #include "aops.h"
33 #include "dir.h"
34 #include "glock.h"
35 #include "glops.h"
36 #include "inode.h"
37 #include "log.h"
38 #include "meta_io.h"
39 #include "quota.h"
40 #include "rgrp.h"
41 #include "trans.h"
42 #include "util.h"
43
44 /**
45 * gfs2_llseek - seek to a location in a file
46 * @file: the file
47 * @offset: the offset
48 * @whence: Where to seek from (SEEK_SET, SEEK_CUR, or SEEK_END)
49 *
50 * SEEK_END requires the glock for the file because it references the
51 * file's size.
52 *
53 * Returns: The new offset, or errno
54 */
55
gfs2_llseek(struct file * file,loff_t offset,int whence)56 static loff_t gfs2_llseek(struct file *file, loff_t offset, int whence)
57 {
58 struct gfs2_inode *ip = GFS2_I(file->f_mapping->host);
59 struct gfs2_holder i_gh;
60 loff_t error;
61
62 switch (whence) {
63 case SEEK_END:
64 error = gfs2_glock_nq_init(ip->i_gl, LM_ST_SHARED, LM_FLAG_ANY,
65 &i_gh);
66 if (!error) {
67 error = generic_file_llseek(file, offset, whence);
68 gfs2_glock_dq_uninit(&i_gh);
69 }
70 break;
71
72 case SEEK_DATA:
73 error = gfs2_seek_data(file, offset);
74 break;
75
76 case SEEK_HOLE:
77 error = gfs2_seek_hole(file, offset);
78 break;
79
80 case SEEK_CUR:
81 case SEEK_SET:
82 /*
83 * These don't reference inode->i_size and don't depend on the
84 * block mapping, so we don't need the glock.
85 */
86 error = generic_file_llseek(file, offset, whence);
87 break;
88 default:
89 error = -EINVAL;
90 }
91
92 return error;
93 }
94
95 /**
96 * gfs2_readdir - Iterator for a directory
97 * @file: The directory to read from
98 * @ctx: What to feed directory entries to
99 *
100 * Returns: errno
101 */
102
gfs2_readdir(struct file * file,struct dir_context * ctx)103 static int gfs2_readdir(struct file *file, struct dir_context *ctx)
104 {
105 struct inode *dir = file->f_mapping->host;
106 struct gfs2_inode *dip = GFS2_I(dir);
107 struct gfs2_holder d_gh;
108 int error;
109
110 error = gfs2_glock_nq_init(dip->i_gl, LM_ST_SHARED, 0, &d_gh);
111 if (error)
112 return error;
113
114 error = gfs2_dir_read(dir, ctx, &file->f_ra);
115
116 gfs2_glock_dq_uninit(&d_gh);
117
118 return error;
119 }
120
121 /**
122 * fsflag_gfs2flag
123 *
124 * The FS_JOURNAL_DATA_FL flag maps to GFS2_DIF_INHERIT_JDATA for directories,
125 * and to GFS2_DIF_JDATA for non-directories.
126 */
127 static struct {
128 u32 fsflag;
129 u32 gfsflag;
130 } fsflag_gfs2flag[] = {
131 {FS_SYNC_FL, GFS2_DIF_SYNC},
132 {FS_IMMUTABLE_FL, GFS2_DIF_IMMUTABLE},
133 {FS_APPEND_FL, GFS2_DIF_APPENDONLY},
134 {FS_NOATIME_FL, GFS2_DIF_NOATIME},
135 {FS_INDEX_FL, GFS2_DIF_EXHASH},
136 {FS_TOPDIR_FL, GFS2_DIF_TOPDIR},
137 {FS_JOURNAL_DATA_FL, GFS2_DIF_JDATA | GFS2_DIF_INHERIT_JDATA},
138 };
139
gfs2_gfsflags_to_fsflags(struct inode * inode,u32 gfsflags)140 static inline u32 gfs2_gfsflags_to_fsflags(struct inode *inode, u32 gfsflags)
141 {
142 int i;
143 u32 fsflags = 0;
144
145 if (S_ISDIR(inode->i_mode))
146 gfsflags &= ~GFS2_DIF_JDATA;
147 else
148 gfsflags &= ~GFS2_DIF_INHERIT_JDATA;
149
150 for (i = 0; i < ARRAY_SIZE(fsflag_gfs2flag); i++)
151 if (gfsflags & fsflag_gfs2flag[i].gfsflag)
152 fsflags |= fsflag_gfs2flag[i].fsflag;
153 return fsflags;
154 }
155
gfs2_get_flags(struct file * filp,u32 __user * ptr)156 static int gfs2_get_flags(struct file *filp, u32 __user *ptr)
157 {
158 struct inode *inode = file_inode(filp);
159 struct gfs2_inode *ip = GFS2_I(inode);
160 struct gfs2_holder gh;
161 int error;
162 u32 fsflags;
163
164 gfs2_holder_init(ip->i_gl, LM_ST_SHARED, 0, &gh);
165 error = gfs2_glock_nq(&gh);
166 if (error)
167 goto out_uninit;
168
169 fsflags = gfs2_gfsflags_to_fsflags(inode, ip->i_diskflags);
170
171 if (put_user(fsflags, ptr))
172 error = -EFAULT;
173
174 gfs2_glock_dq(&gh);
175 out_uninit:
176 gfs2_holder_uninit(&gh);
177 return error;
178 }
179
gfs2_set_inode_flags(struct inode * inode)180 void gfs2_set_inode_flags(struct inode *inode)
181 {
182 struct gfs2_inode *ip = GFS2_I(inode);
183 unsigned int flags = inode->i_flags;
184
185 flags &= ~(S_SYNC|S_APPEND|S_IMMUTABLE|S_NOATIME|S_DIRSYNC|S_NOSEC);
186 if ((ip->i_eattr == 0) && !is_sxid(inode->i_mode))
187 flags |= S_NOSEC;
188 if (ip->i_diskflags & GFS2_DIF_IMMUTABLE)
189 flags |= S_IMMUTABLE;
190 if (ip->i_diskflags & GFS2_DIF_APPENDONLY)
191 flags |= S_APPEND;
192 if (ip->i_diskflags & GFS2_DIF_NOATIME)
193 flags |= S_NOATIME;
194 if (ip->i_diskflags & GFS2_DIF_SYNC)
195 flags |= S_SYNC;
196 inode->i_flags = flags;
197 }
198
199 /* Flags that can be set by user space */
200 #define GFS2_FLAGS_USER_SET (GFS2_DIF_JDATA| \
201 GFS2_DIF_IMMUTABLE| \
202 GFS2_DIF_APPENDONLY| \
203 GFS2_DIF_NOATIME| \
204 GFS2_DIF_SYNC| \
205 GFS2_DIF_TOPDIR| \
206 GFS2_DIF_INHERIT_JDATA)
207
208 /**
209 * do_gfs2_set_flags - set flags on an inode
210 * @filp: file pointer
211 * @reqflags: The flags to set
212 * @mask: Indicates which flags are valid
213 * @fsflags: The FS_* inode flags passed in
214 *
215 */
do_gfs2_set_flags(struct file * filp,u32 reqflags,u32 mask,const u32 fsflags)216 static int do_gfs2_set_flags(struct file *filp, u32 reqflags, u32 mask,
217 const u32 fsflags)
218 {
219 struct inode *inode = file_inode(filp);
220 struct gfs2_inode *ip = GFS2_I(inode);
221 struct gfs2_sbd *sdp = GFS2_SB(inode);
222 struct buffer_head *bh;
223 struct gfs2_holder gh;
224 int error;
225 u32 new_flags, flags, oldflags;
226
227 error = mnt_want_write_file(filp);
228 if (error)
229 return error;
230
231 error = gfs2_glock_nq_init(ip->i_gl, LM_ST_EXCLUSIVE, 0, &gh);
232 if (error)
233 goto out_drop_write;
234
235 oldflags = gfs2_gfsflags_to_fsflags(inode, ip->i_diskflags);
236 error = vfs_ioc_setflags_prepare(inode, oldflags, fsflags);
237 if (error)
238 goto out;
239
240 error = -EACCES;
241 if (!inode_owner_or_capable(inode))
242 goto out;
243
244 error = 0;
245 flags = ip->i_diskflags;
246 new_flags = (flags & ~mask) | (reqflags & mask);
247 if ((new_flags ^ flags) == 0)
248 goto out;
249
250 error = -EPERM;
251 if (IS_IMMUTABLE(inode) && (new_flags & GFS2_DIF_IMMUTABLE))
252 goto out;
253 if (IS_APPEND(inode) && (new_flags & GFS2_DIF_APPENDONLY))
254 goto out;
255 if (((new_flags ^ flags) & GFS2_DIF_IMMUTABLE) &&
256 !capable(CAP_LINUX_IMMUTABLE))
257 goto out;
258 if (!IS_IMMUTABLE(inode)) {
259 error = gfs2_permission(inode, MAY_WRITE);
260 if (error)
261 goto out;
262 }
263 if ((flags ^ new_flags) & GFS2_DIF_JDATA) {
264 if (new_flags & GFS2_DIF_JDATA)
265 gfs2_log_flush(sdp, ip->i_gl,
266 GFS2_LOG_HEAD_FLUSH_NORMAL |
267 GFS2_LFC_SET_FLAGS);
268 error = filemap_fdatawrite(inode->i_mapping);
269 if (error)
270 goto out;
271 error = filemap_fdatawait(inode->i_mapping);
272 if (error)
273 goto out;
274 if (new_flags & GFS2_DIF_JDATA)
275 gfs2_ordered_del_inode(ip);
276 }
277 error = gfs2_trans_begin(sdp, RES_DINODE, 0);
278 if (error)
279 goto out;
280 error = gfs2_meta_inode_buffer(ip, &bh);
281 if (error)
282 goto out_trans_end;
283 inode->i_ctime = current_time(inode);
284 gfs2_trans_add_meta(ip->i_gl, bh);
285 ip->i_diskflags = new_flags;
286 gfs2_dinode_out(ip, bh->b_data);
287 brelse(bh);
288 gfs2_set_inode_flags(inode);
289 gfs2_set_aops(inode);
290 out_trans_end:
291 gfs2_trans_end(sdp);
292 out:
293 gfs2_glock_dq_uninit(&gh);
294 out_drop_write:
295 mnt_drop_write_file(filp);
296 return error;
297 }
298
gfs2_set_flags(struct file * filp,u32 __user * ptr)299 static int gfs2_set_flags(struct file *filp, u32 __user *ptr)
300 {
301 struct inode *inode = file_inode(filp);
302 u32 fsflags, gfsflags = 0;
303 u32 mask;
304 int i;
305
306 if (get_user(fsflags, ptr))
307 return -EFAULT;
308
309 for (i = 0; i < ARRAY_SIZE(fsflag_gfs2flag); i++) {
310 if (fsflags & fsflag_gfs2flag[i].fsflag) {
311 fsflags &= ~fsflag_gfs2flag[i].fsflag;
312 gfsflags |= fsflag_gfs2flag[i].gfsflag;
313 }
314 }
315 if (fsflags || gfsflags & ~GFS2_FLAGS_USER_SET)
316 return -EINVAL;
317
318 mask = GFS2_FLAGS_USER_SET;
319 if (S_ISDIR(inode->i_mode)) {
320 mask &= ~GFS2_DIF_JDATA;
321 } else {
322 /* The GFS2_DIF_TOPDIR flag is only valid for directories. */
323 if (gfsflags & GFS2_DIF_TOPDIR)
324 return -EINVAL;
325 mask &= ~(GFS2_DIF_TOPDIR | GFS2_DIF_INHERIT_JDATA);
326 }
327
328 return do_gfs2_set_flags(filp, gfsflags, mask, fsflags);
329 }
330
gfs2_getlabel(struct file * filp,char __user * label)331 static int gfs2_getlabel(struct file *filp, char __user *label)
332 {
333 struct inode *inode = file_inode(filp);
334 struct gfs2_sbd *sdp = GFS2_SB(inode);
335
336 if (copy_to_user(label, sdp->sd_sb.sb_locktable, GFS2_LOCKNAME_LEN))
337 return -EFAULT;
338
339 return 0;
340 }
341
gfs2_ioctl(struct file * filp,unsigned int cmd,unsigned long arg)342 static long gfs2_ioctl(struct file *filp, unsigned int cmd, unsigned long arg)
343 {
344 switch(cmd) {
345 case FS_IOC_GETFLAGS:
346 return gfs2_get_flags(filp, (u32 __user *)arg);
347 case FS_IOC_SETFLAGS:
348 return gfs2_set_flags(filp, (u32 __user *)arg);
349 case FITRIM:
350 return gfs2_fitrim(filp, (void __user *)arg);
351 case FS_IOC_GETFSLABEL:
352 return gfs2_getlabel(filp, (char __user *)arg);
353 }
354
355 return -ENOTTY;
356 }
357
358 #ifdef CONFIG_COMPAT
gfs2_compat_ioctl(struct file * filp,unsigned int cmd,unsigned long arg)359 static long gfs2_compat_ioctl(struct file *filp, unsigned int cmd, unsigned long arg)
360 {
361 switch(cmd) {
362 /* These are just misnamed, they actually get/put from/to user an int */
363 case FS_IOC32_GETFLAGS:
364 cmd = FS_IOC_GETFLAGS;
365 break;
366 case FS_IOC32_SETFLAGS:
367 cmd = FS_IOC_SETFLAGS;
368 break;
369 /* Keep this list in sync with gfs2_ioctl */
370 case FITRIM:
371 case FS_IOC_GETFSLABEL:
372 break;
373 default:
374 return -ENOIOCTLCMD;
375 }
376
377 return gfs2_ioctl(filp, cmd, (unsigned long)compat_ptr(arg));
378 }
379 #else
380 #define gfs2_compat_ioctl NULL
381 #endif
382
383 /**
384 * gfs2_size_hint - Give a hint to the size of a write request
385 * @filep: The struct file
386 * @offset: The file offset of the write
387 * @size: The length of the write
388 *
389 * When we are about to do a write, this function records the total
390 * write size in order to provide a suitable hint to the lower layers
391 * about how many blocks will be required.
392 *
393 */
394
gfs2_size_hint(struct file * filep,loff_t offset,size_t size)395 static void gfs2_size_hint(struct file *filep, loff_t offset, size_t size)
396 {
397 struct inode *inode = file_inode(filep);
398 struct gfs2_sbd *sdp = GFS2_SB(inode);
399 struct gfs2_inode *ip = GFS2_I(inode);
400 size_t blks = (size + sdp->sd_sb.sb_bsize - 1) >> sdp->sd_sb.sb_bsize_shift;
401 int hint = min_t(size_t, INT_MAX, blks);
402
403 if (hint > atomic_read(&ip->i_sizehint))
404 atomic_set(&ip->i_sizehint, hint);
405 }
406
407 /**
408 * gfs2_allocate_page_backing - Allocate blocks for a write fault
409 * @page: The (locked) page to allocate backing for
410 * @length: Size of the allocation
411 *
412 * We try to allocate all the blocks required for the page in one go. This
413 * might fail for various reasons, so we keep trying until all the blocks to
414 * back this page are allocated. If some of the blocks are already allocated,
415 * that is ok too.
416 */
gfs2_allocate_page_backing(struct page * page,unsigned int length)417 static int gfs2_allocate_page_backing(struct page *page, unsigned int length)
418 {
419 u64 pos = page_offset(page);
420
421 do {
422 struct iomap iomap = { };
423
424 if (gfs2_iomap_get_alloc(page->mapping->host, pos, length, &iomap))
425 return -EIO;
426
427 if (length < iomap.length)
428 iomap.length = length;
429 length -= iomap.length;
430 pos += iomap.length;
431 } while (length > 0);
432
433 return 0;
434 }
435
436 /**
437 * gfs2_page_mkwrite - Make a shared, mmap()ed, page writable
438 * @vma: The virtual memory area
439 * @vmf: The virtual memory fault containing the page to become writable
440 *
441 * When the page becomes writable, we need to ensure that we have
442 * blocks allocated on disk to back that page.
443 */
444
gfs2_page_mkwrite(struct vm_fault * vmf)445 static vm_fault_t gfs2_page_mkwrite(struct vm_fault *vmf)
446 {
447 struct page *page = vmf->page;
448 struct inode *inode = file_inode(vmf->vma->vm_file);
449 struct gfs2_inode *ip = GFS2_I(inode);
450 struct gfs2_sbd *sdp = GFS2_SB(inode);
451 struct gfs2_alloc_parms ap = { .aflags = 0, };
452 u64 offset = page_offset(page);
453 unsigned int data_blocks, ind_blocks, rblocks;
454 struct gfs2_holder gh;
455 unsigned int length;
456 loff_t size;
457 int ret;
458
459 sb_start_pagefault(inode->i_sb);
460
461 gfs2_holder_init(ip->i_gl, LM_ST_EXCLUSIVE, 0, &gh);
462 ret = gfs2_glock_nq(&gh);
463 if (ret)
464 goto out_uninit;
465
466 /* Check page index against inode size */
467 size = i_size_read(inode);
468 if (offset >= size) {
469 ret = -EINVAL;
470 goto out_unlock;
471 }
472
473 /* Update file times before taking page lock */
474 file_update_time(vmf->vma->vm_file);
475
476 /* page is wholly or partially inside EOF */
477 if (size - offset < PAGE_SIZE)
478 length = size - offset;
479 else
480 length = PAGE_SIZE;
481
482 gfs2_size_hint(vmf->vma->vm_file, offset, length);
483
484 set_bit(GLF_DIRTY, &ip->i_gl->gl_flags);
485 set_bit(GIF_SW_PAGED, &ip->i_flags);
486
487 /*
488 * iomap_writepage / iomap_writepages currently don't support inline
489 * files, so always unstuff here.
490 */
491
492 if (!gfs2_is_stuffed(ip) &&
493 !gfs2_write_alloc_required(ip, offset, length)) {
494 lock_page(page);
495 if (!PageUptodate(page) || page->mapping != inode->i_mapping) {
496 ret = -EAGAIN;
497 unlock_page(page);
498 }
499 goto out_unlock;
500 }
501
502 ret = gfs2_rindex_update(sdp);
503 if (ret)
504 goto out_unlock;
505
506 gfs2_write_calc_reserv(ip, length, &data_blocks, &ind_blocks);
507 ap.target = data_blocks + ind_blocks;
508 ret = gfs2_quota_lock_check(ip, &ap);
509 if (ret)
510 goto out_unlock;
511 ret = gfs2_inplace_reserve(ip, &ap);
512 if (ret)
513 goto out_quota_unlock;
514
515 rblocks = RES_DINODE + ind_blocks;
516 if (gfs2_is_jdata(ip))
517 rblocks += data_blocks ? data_blocks : 1;
518 if (ind_blocks || data_blocks) {
519 rblocks += RES_STATFS + RES_QUOTA;
520 rblocks += gfs2_rg_blocks(ip, data_blocks + ind_blocks);
521 }
522 ret = gfs2_trans_begin(sdp, rblocks, 0);
523 if (ret)
524 goto out_trans_fail;
525
526 lock_page(page);
527 ret = -EAGAIN;
528 /* If truncated, we must retry the operation, we may have raced
529 * with the glock demotion code.
530 */
531 if (!PageUptodate(page) || page->mapping != inode->i_mapping)
532 goto out_trans_end;
533
534 /* Unstuff, if required, and allocate backing blocks for page */
535 ret = 0;
536 if (gfs2_is_stuffed(ip))
537 ret = gfs2_unstuff_dinode(ip, page);
538 if (ret == 0)
539 ret = gfs2_allocate_page_backing(page, length);
540
541 out_trans_end:
542 if (ret)
543 unlock_page(page);
544 gfs2_trans_end(sdp);
545 out_trans_fail:
546 gfs2_inplace_release(ip);
547 out_quota_unlock:
548 gfs2_quota_unlock(ip);
549 out_unlock:
550 gfs2_glock_dq(&gh);
551 out_uninit:
552 gfs2_holder_uninit(&gh);
553 if (ret == 0) {
554 set_page_dirty(page);
555 wait_for_stable_page(page);
556 }
557 sb_end_pagefault(inode->i_sb);
558 return block_page_mkwrite_return(ret);
559 }
560
gfs2_fault(struct vm_fault * vmf)561 static vm_fault_t gfs2_fault(struct vm_fault *vmf)
562 {
563 struct inode *inode = file_inode(vmf->vma->vm_file);
564 struct gfs2_inode *ip = GFS2_I(inode);
565 struct gfs2_holder gh;
566 vm_fault_t ret;
567 int err;
568
569 gfs2_holder_init(ip->i_gl, LM_ST_SHARED, 0, &gh);
570 err = gfs2_glock_nq(&gh);
571 if (err) {
572 ret = block_page_mkwrite_return(err);
573 goto out_uninit;
574 }
575 ret = filemap_fault(vmf);
576 gfs2_glock_dq(&gh);
577 out_uninit:
578 gfs2_holder_uninit(&gh);
579 return ret;
580 }
581
582 static const struct vm_operations_struct gfs2_vm_ops = {
583 .fault = gfs2_fault,
584 .map_pages = filemap_map_pages,
585 .page_mkwrite = gfs2_page_mkwrite,
586 };
587
588 /**
589 * gfs2_mmap -
590 * @file: The file to map
591 * @vma: The VMA which described the mapping
592 *
593 * There is no need to get a lock here unless we should be updating
594 * atime. We ignore any locking errors since the only consequence is
595 * a missed atime update (which will just be deferred until later).
596 *
597 * Returns: 0
598 */
599
gfs2_mmap(struct file * file,struct vm_area_struct * vma)600 static int gfs2_mmap(struct file *file, struct vm_area_struct *vma)
601 {
602 struct gfs2_inode *ip = GFS2_I(file->f_mapping->host);
603
604 if (!(file->f_flags & O_NOATIME) &&
605 !IS_NOATIME(&ip->i_inode)) {
606 struct gfs2_holder i_gh;
607 int error;
608
609 error = gfs2_glock_nq_init(ip->i_gl, LM_ST_SHARED, LM_FLAG_ANY,
610 &i_gh);
611 if (error)
612 return error;
613 /* grab lock to update inode */
614 gfs2_glock_dq_uninit(&i_gh);
615 file_accessed(file);
616 }
617 vma->vm_ops = &gfs2_vm_ops;
618
619 return 0;
620 }
621
622 /**
623 * gfs2_open_common - This is common to open and atomic_open
624 * @inode: The inode being opened
625 * @file: The file being opened
626 *
627 * This maybe called under a glock or not depending upon how it has
628 * been called. We must always be called under a glock for regular
629 * files, however. For other file types, it does not matter whether
630 * we hold the glock or not.
631 *
632 * Returns: Error code or 0 for success
633 */
634
gfs2_open_common(struct inode * inode,struct file * file)635 int gfs2_open_common(struct inode *inode, struct file *file)
636 {
637 struct gfs2_file *fp;
638 int ret;
639
640 if (S_ISREG(inode->i_mode)) {
641 ret = generic_file_open(inode, file);
642 if (ret)
643 return ret;
644 }
645
646 fp = kzalloc(sizeof(struct gfs2_file), GFP_NOFS);
647 if (!fp)
648 return -ENOMEM;
649
650 mutex_init(&fp->f_fl_mutex);
651
652 gfs2_assert_warn(GFS2_SB(inode), !file->private_data);
653 file->private_data = fp;
654 if (file->f_mode & FMODE_WRITE) {
655 ret = gfs2_qa_get(GFS2_I(inode));
656 if (ret)
657 goto fail;
658 }
659 return 0;
660
661 fail:
662 kfree(file->private_data);
663 file->private_data = NULL;
664 return ret;
665 }
666
667 /**
668 * gfs2_open - open a file
669 * @inode: the inode to open
670 * @file: the struct file for this opening
671 *
672 * After atomic_open, this function is only used for opening files
673 * which are already cached. We must still get the glock for regular
674 * files to ensure that we have the file size uptodate for the large
675 * file check which is in the common code. That is only an issue for
676 * regular files though.
677 *
678 * Returns: errno
679 */
680
gfs2_open(struct inode * inode,struct file * file)681 static int gfs2_open(struct inode *inode, struct file *file)
682 {
683 struct gfs2_inode *ip = GFS2_I(inode);
684 struct gfs2_holder i_gh;
685 int error;
686 bool need_unlock = false;
687
688 if (S_ISREG(ip->i_inode.i_mode)) {
689 error = gfs2_glock_nq_init(ip->i_gl, LM_ST_SHARED, LM_FLAG_ANY,
690 &i_gh);
691 if (error)
692 return error;
693 need_unlock = true;
694 }
695
696 error = gfs2_open_common(inode, file);
697
698 if (need_unlock)
699 gfs2_glock_dq_uninit(&i_gh);
700
701 return error;
702 }
703
704 /**
705 * gfs2_release - called to close a struct file
706 * @inode: the inode the struct file belongs to
707 * @file: the struct file being closed
708 *
709 * Returns: errno
710 */
711
gfs2_release(struct inode * inode,struct file * file)712 static int gfs2_release(struct inode *inode, struct file *file)
713 {
714 struct gfs2_inode *ip = GFS2_I(inode);
715
716 kfree(file->private_data);
717 file->private_data = NULL;
718
719 if (file->f_mode & FMODE_WRITE) {
720 if (gfs2_rs_active(&ip->i_res))
721 gfs2_rs_delete(ip);
722 gfs2_qa_put(ip);
723 }
724 return 0;
725 }
726
727 /**
728 * gfs2_fsync - sync the dirty data for a file (across the cluster)
729 * @file: the file that points to the dentry
730 * @start: the start position in the file to sync
731 * @end: the end position in the file to sync
732 * @datasync: set if we can ignore timestamp changes
733 *
734 * We split the data flushing here so that we don't wait for the data
735 * until after we've also sent the metadata to disk. Note that for
736 * data=ordered, we will write & wait for the data at the log flush
737 * stage anyway, so this is unlikely to make much of a difference
738 * except in the data=writeback case.
739 *
740 * If the fdatawrite fails due to any reason except -EIO, we will
741 * continue the remainder of the fsync, although we'll still report
742 * the error at the end. This is to match filemap_write_and_wait_range()
743 * behaviour.
744 *
745 * Returns: errno
746 */
747
gfs2_fsync(struct file * file,loff_t start,loff_t end,int datasync)748 static int gfs2_fsync(struct file *file, loff_t start, loff_t end,
749 int datasync)
750 {
751 struct address_space *mapping = file->f_mapping;
752 struct inode *inode = mapping->host;
753 int sync_state = inode->i_state & I_DIRTY_ALL;
754 struct gfs2_inode *ip = GFS2_I(inode);
755 int ret = 0, ret1 = 0;
756
757 if (mapping->nrpages) {
758 ret1 = filemap_fdatawrite_range(mapping, start, end);
759 if (ret1 == -EIO)
760 return ret1;
761 }
762
763 if (!gfs2_is_jdata(ip))
764 sync_state &= ~I_DIRTY_PAGES;
765 if (datasync)
766 sync_state &= ~(I_DIRTY_SYNC | I_DIRTY_TIME);
767
768 if (sync_state) {
769 ret = sync_inode_metadata(inode, 1);
770 if (ret)
771 return ret;
772 if (gfs2_is_jdata(ip))
773 ret = file_write_and_wait(file);
774 if (ret)
775 return ret;
776 gfs2_ail_flush(ip->i_gl, 1);
777 }
778
779 if (mapping->nrpages)
780 ret = file_fdatawait_range(file, start, end);
781
782 return ret ? ret : ret1;
783 }
784
gfs2_file_direct_read(struct kiocb * iocb,struct iov_iter * to,struct gfs2_holder * gh)785 static ssize_t gfs2_file_direct_read(struct kiocb *iocb, struct iov_iter *to,
786 struct gfs2_holder *gh)
787 {
788 struct file *file = iocb->ki_filp;
789 struct gfs2_inode *ip = GFS2_I(file->f_mapping->host);
790 size_t count = iov_iter_count(to);
791 ssize_t ret;
792
793 if (!count)
794 return 0; /* skip atime */
795
796 gfs2_holder_init(ip->i_gl, LM_ST_DEFERRED, 0, gh);
797 ret = gfs2_glock_nq(gh);
798 if (ret)
799 goto out_uninit;
800
801 ret = iomap_dio_rw(iocb, to, &gfs2_iomap_ops, NULL,
802 is_sync_kiocb(iocb));
803
804 gfs2_glock_dq(gh);
805 out_uninit:
806 gfs2_holder_uninit(gh);
807 return ret;
808 }
809
gfs2_file_direct_write(struct kiocb * iocb,struct iov_iter * from,struct gfs2_holder * gh)810 static ssize_t gfs2_file_direct_write(struct kiocb *iocb, struct iov_iter *from,
811 struct gfs2_holder *gh)
812 {
813 struct file *file = iocb->ki_filp;
814 struct inode *inode = file->f_mapping->host;
815 struct gfs2_inode *ip = GFS2_I(inode);
816 size_t len = iov_iter_count(from);
817 loff_t offset = iocb->ki_pos;
818 ssize_t ret;
819
820 /*
821 * Deferred lock, even if its a write, since we do no allocation on
822 * this path. All we need to change is the atime, and this lock mode
823 * ensures that other nodes have flushed their buffered read caches
824 * (i.e. their page cache entries for this inode). We do not,
825 * unfortunately, have the option of only flushing a range like the
826 * VFS does.
827 */
828 gfs2_holder_init(ip->i_gl, LM_ST_DEFERRED, 0, gh);
829 ret = gfs2_glock_nq(gh);
830 if (ret)
831 goto out_uninit;
832
833 /* Silently fall back to buffered I/O when writing beyond EOF */
834 if (offset + len > i_size_read(&ip->i_inode))
835 goto out;
836
837 ret = iomap_dio_rw(iocb, from, &gfs2_iomap_ops, NULL,
838 is_sync_kiocb(iocb));
839 if (ret == -ENOTBLK)
840 ret = 0;
841 out:
842 gfs2_glock_dq(gh);
843 out_uninit:
844 gfs2_holder_uninit(gh);
845 return ret;
846 }
847
gfs2_file_read_iter(struct kiocb * iocb,struct iov_iter * to)848 static ssize_t gfs2_file_read_iter(struct kiocb *iocb, struct iov_iter *to)
849 {
850 struct gfs2_inode *ip;
851 struct gfs2_holder gh;
852 size_t written = 0;
853 ssize_t ret;
854
855 if (iocb->ki_flags & IOCB_DIRECT) {
856 ret = gfs2_file_direct_read(iocb, to, &gh);
857 if (likely(ret != -ENOTBLK))
858 return ret;
859 iocb->ki_flags &= ~IOCB_DIRECT;
860 }
861 pagefault_disable();
862 iocb->ki_flags |= IOCB_NOIO;
863 ret = generic_file_read_iter(iocb, to);
864 iocb->ki_flags &= ~IOCB_NOIO;
865 pagefault_enable();
866 if (ret >= 0) {
867 if (!iov_iter_count(to))
868 return ret;
869 written = ret;
870 } else if (ret != -EFAULT) {
871 if (ret != -EAGAIN)
872 return ret;
873 if (iocb->ki_flags & IOCB_NOWAIT)
874 return ret;
875 }
876 ip = GFS2_I(iocb->ki_filp->f_mapping->host);
877 gfs2_holder_init(ip->i_gl, LM_ST_SHARED, 0, &gh);
878 ret = gfs2_glock_nq(&gh);
879 if (ret)
880 goto out_uninit;
881 ret = generic_file_read_iter(iocb, to);
882 if (ret > 0)
883 written += ret;
884 gfs2_glock_dq(&gh);
885 out_uninit:
886 gfs2_holder_uninit(&gh);
887 return written ? written : ret;
888 }
889
890 /**
891 * gfs2_file_write_iter - Perform a write to a file
892 * @iocb: The io context
893 * @from: The data to write
894 *
895 * We have to do a lock/unlock here to refresh the inode size for
896 * O_APPEND writes, otherwise we can land up writing at the wrong
897 * offset. There is still a race, but provided the app is using its
898 * own file locking, this will make O_APPEND work as expected.
899 *
900 */
901
gfs2_file_write_iter(struct kiocb * iocb,struct iov_iter * from)902 static ssize_t gfs2_file_write_iter(struct kiocb *iocb, struct iov_iter *from)
903 {
904 struct file *file = iocb->ki_filp;
905 struct inode *inode = file_inode(file);
906 struct gfs2_inode *ip = GFS2_I(inode);
907 struct gfs2_holder gh;
908 ssize_t ret;
909
910 gfs2_size_hint(file, iocb->ki_pos, iov_iter_count(from));
911
912 if (iocb->ki_flags & IOCB_APPEND) {
913 ret = gfs2_glock_nq_init(ip->i_gl, LM_ST_SHARED, 0, &gh);
914 if (ret)
915 return ret;
916 gfs2_glock_dq_uninit(&gh);
917 }
918
919 inode_lock(inode);
920 ret = generic_write_checks(iocb, from);
921 if (ret <= 0)
922 goto out_unlock;
923
924 ret = file_remove_privs(file);
925 if (ret)
926 goto out_unlock;
927
928 ret = file_update_time(file);
929 if (ret)
930 goto out_unlock;
931
932 if (iocb->ki_flags & IOCB_DIRECT) {
933 struct address_space *mapping = file->f_mapping;
934 ssize_t buffered, ret2;
935
936 ret = gfs2_file_direct_write(iocb, from, &gh);
937 if (ret < 0 || !iov_iter_count(from))
938 goto out_unlock;
939
940 iocb->ki_flags |= IOCB_DSYNC;
941 current->backing_dev_info = inode_to_bdi(inode);
942 buffered = iomap_file_buffered_write(iocb, from, &gfs2_iomap_ops);
943 current->backing_dev_info = NULL;
944 if (unlikely(buffered <= 0)) {
945 if (!ret)
946 ret = buffered;
947 goto out_unlock;
948 }
949
950 /*
951 * We need to ensure that the page cache pages are written to
952 * disk and invalidated to preserve the expected O_DIRECT
953 * semantics. If the writeback or invalidate fails, only report
954 * the direct I/O range as we don't know if the buffered pages
955 * made it to disk.
956 */
957 iocb->ki_pos += buffered;
958 ret2 = generic_write_sync(iocb, buffered);
959 invalidate_mapping_pages(mapping,
960 (iocb->ki_pos - buffered) >> PAGE_SHIFT,
961 (iocb->ki_pos - 1) >> PAGE_SHIFT);
962 if (!ret || ret2 > 0)
963 ret += ret2;
964 } else {
965 current->backing_dev_info = inode_to_bdi(inode);
966 ret = iomap_file_buffered_write(iocb, from, &gfs2_iomap_ops);
967 current->backing_dev_info = NULL;
968 if (likely(ret > 0)) {
969 iocb->ki_pos += ret;
970 ret = generic_write_sync(iocb, ret);
971 }
972 }
973
974 out_unlock:
975 inode_unlock(inode);
976 return ret;
977 }
978
fallocate_chunk(struct inode * inode,loff_t offset,loff_t len,int mode)979 static int fallocate_chunk(struct inode *inode, loff_t offset, loff_t len,
980 int mode)
981 {
982 struct super_block *sb = inode->i_sb;
983 struct gfs2_inode *ip = GFS2_I(inode);
984 loff_t end = offset + len;
985 struct buffer_head *dibh;
986 int error;
987
988 error = gfs2_meta_inode_buffer(ip, &dibh);
989 if (unlikely(error))
990 return error;
991
992 gfs2_trans_add_meta(ip->i_gl, dibh);
993
994 if (gfs2_is_stuffed(ip)) {
995 error = gfs2_unstuff_dinode(ip, NULL);
996 if (unlikely(error))
997 goto out;
998 }
999
1000 while (offset < end) {
1001 struct iomap iomap = { };
1002
1003 error = gfs2_iomap_get_alloc(inode, offset, end - offset,
1004 &iomap);
1005 if (error)
1006 goto out;
1007 offset = iomap.offset + iomap.length;
1008 if (!(iomap.flags & IOMAP_F_NEW))
1009 continue;
1010 error = sb_issue_zeroout(sb, iomap.addr >> inode->i_blkbits,
1011 iomap.length >> inode->i_blkbits,
1012 GFP_NOFS);
1013 if (error) {
1014 fs_err(GFS2_SB(inode), "Failed to zero data buffers\n");
1015 goto out;
1016 }
1017 }
1018 out:
1019 brelse(dibh);
1020 return error;
1021 }
1022
1023 /**
1024 * calc_max_reserv() - Reverse of write_calc_reserv. Given a number of
1025 * blocks, determine how many bytes can be written.
1026 * @ip: The inode in question.
1027 * @len: Max cap of bytes. What we return in *len must be <= this.
1028 * @data_blocks: Compute and return the number of data blocks needed
1029 * @ind_blocks: Compute and return the number of indirect blocks needed
1030 * @max_blocks: The total blocks available to work with.
1031 *
1032 * Returns: void, but @len, @data_blocks and @ind_blocks are filled in.
1033 */
calc_max_reserv(struct gfs2_inode * ip,loff_t * len,unsigned int * data_blocks,unsigned int * ind_blocks,unsigned int max_blocks)1034 static void calc_max_reserv(struct gfs2_inode *ip, loff_t *len,
1035 unsigned int *data_blocks, unsigned int *ind_blocks,
1036 unsigned int max_blocks)
1037 {
1038 loff_t max = *len;
1039 const struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode);
1040 unsigned int tmp, max_data = max_blocks - 3 * (sdp->sd_max_height - 1);
1041
1042 for (tmp = max_data; tmp > sdp->sd_diptrs;) {
1043 tmp = DIV_ROUND_UP(tmp, sdp->sd_inptrs);
1044 max_data -= tmp;
1045 }
1046
1047 *data_blocks = max_data;
1048 *ind_blocks = max_blocks - max_data;
1049 *len = ((loff_t)max_data - 3) << sdp->sd_sb.sb_bsize_shift;
1050 if (*len > max) {
1051 *len = max;
1052 gfs2_write_calc_reserv(ip, max, data_blocks, ind_blocks);
1053 }
1054 }
1055
__gfs2_fallocate(struct file * file,int mode,loff_t offset,loff_t len)1056 static long __gfs2_fallocate(struct file *file, int mode, loff_t offset, loff_t len)
1057 {
1058 struct inode *inode = file_inode(file);
1059 struct gfs2_sbd *sdp = GFS2_SB(inode);
1060 struct gfs2_inode *ip = GFS2_I(inode);
1061 struct gfs2_alloc_parms ap = { .aflags = 0, };
1062 unsigned int data_blocks = 0, ind_blocks = 0, rblocks;
1063 loff_t bytes, max_bytes, max_blks;
1064 int error;
1065 const loff_t pos = offset;
1066 const loff_t count = len;
1067 loff_t bsize_mask = ~((loff_t)sdp->sd_sb.sb_bsize - 1);
1068 loff_t next = (offset + len - 1) >> sdp->sd_sb.sb_bsize_shift;
1069 loff_t max_chunk_size = UINT_MAX & bsize_mask;
1070
1071 next = (next + 1) << sdp->sd_sb.sb_bsize_shift;
1072
1073 offset &= bsize_mask;
1074
1075 len = next - offset;
1076 bytes = sdp->sd_max_rg_data * sdp->sd_sb.sb_bsize / 2;
1077 if (!bytes)
1078 bytes = UINT_MAX;
1079 bytes &= bsize_mask;
1080 if (bytes == 0)
1081 bytes = sdp->sd_sb.sb_bsize;
1082
1083 gfs2_size_hint(file, offset, len);
1084
1085 gfs2_write_calc_reserv(ip, PAGE_SIZE, &data_blocks, &ind_blocks);
1086 ap.min_target = data_blocks + ind_blocks;
1087
1088 while (len > 0) {
1089 if (len < bytes)
1090 bytes = len;
1091 if (!gfs2_write_alloc_required(ip, offset, bytes)) {
1092 len -= bytes;
1093 offset += bytes;
1094 continue;
1095 }
1096
1097 /* We need to determine how many bytes we can actually
1098 * fallocate without exceeding quota or going over the
1099 * end of the fs. We start off optimistically by assuming
1100 * we can write max_bytes */
1101 max_bytes = (len > max_chunk_size) ? max_chunk_size : len;
1102
1103 /* Since max_bytes is most likely a theoretical max, we
1104 * calculate a more realistic 'bytes' to serve as a good
1105 * starting point for the number of bytes we may be able
1106 * to write */
1107 gfs2_write_calc_reserv(ip, bytes, &data_blocks, &ind_blocks);
1108 ap.target = data_blocks + ind_blocks;
1109
1110 error = gfs2_quota_lock_check(ip, &ap);
1111 if (error)
1112 return error;
1113 /* ap.allowed tells us how many blocks quota will allow
1114 * us to write. Check if this reduces max_blks */
1115 max_blks = UINT_MAX;
1116 if (ap.allowed)
1117 max_blks = ap.allowed;
1118
1119 error = gfs2_inplace_reserve(ip, &ap);
1120 if (error)
1121 goto out_qunlock;
1122
1123 /* check if the selected rgrp limits our max_blks further */
1124 if (ap.allowed && ap.allowed < max_blks)
1125 max_blks = ap.allowed;
1126
1127 /* Almost done. Calculate bytes that can be written using
1128 * max_blks. We also recompute max_bytes, data_blocks and
1129 * ind_blocks */
1130 calc_max_reserv(ip, &max_bytes, &data_blocks,
1131 &ind_blocks, max_blks);
1132
1133 rblocks = RES_DINODE + ind_blocks + RES_STATFS + RES_QUOTA +
1134 RES_RG_HDR + gfs2_rg_blocks(ip, data_blocks + ind_blocks);
1135 if (gfs2_is_jdata(ip))
1136 rblocks += data_blocks ? data_blocks : 1;
1137
1138 error = gfs2_trans_begin(sdp, rblocks,
1139 PAGE_SIZE >> inode->i_blkbits);
1140 if (error)
1141 goto out_trans_fail;
1142
1143 error = fallocate_chunk(inode, offset, max_bytes, mode);
1144 gfs2_trans_end(sdp);
1145
1146 if (error)
1147 goto out_trans_fail;
1148
1149 len -= max_bytes;
1150 offset += max_bytes;
1151 gfs2_inplace_release(ip);
1152 gfs2_quota_unlock(ip);
1153 }
1154
1155 if (!(mode & FALLOC_FL_KEEP_SIZE) && (pos + count) > inode->i_size)
1156 i_size_write(inode, pos + count);
1157 file_update_time(file);
1158 mark_inode_dirty(inode);
1159
1160 if ((file->f_flags & O_DSYNC) || IS_SYNC(file->f_mapping->host))
1161 return vfs_fsync_range(file, pos, pos + count - 1,
1162 (file->f_flags & __O_SYNC) ? 0 : 1);
1163 return 0;
1164
1165 out_trans_fail:
1166 gfs2_inplace_release(ip);
1167 out_qunlock:
1168 gfs2_quota_unlock(ip);
1169 return error;
1170 }
1171
gfs2_fallocate(struct file * file,int mode,loff_t offset,loff_t len)1172 static long gfs2_fallocate(struct file *file, int mode, loff_t offset, loff_t len)
1173 {
1174 struct inode *inode = file_inode(file);
1175 struct gfs2_sbd *sdp = GFS2_SB(inode);
1176 struct gfs2_inode *ip = GFS2_I(inode);
1177 struct gfs2_holder gh;
1178 int ret;
1179
1180 if (mode & ~(FALLOC_FL_PUNCH_HOLE | FALLOC_FL_KEEP_SIZE))
1181 return -EOPNOTSUPP;
1182 /* fallocate is needed by gfs2_grow to reserve space in the rindex */
1183 if (gfs2_is_jdata(ip) && inode != sdp->sd_rindex)
1184 return -EOPNOTSUPP;
1185
1186 inode_lock(inode);
1187
1188 gfs2_holder_init(ip->i_gl, LM_ST_EXCLUSIVE, 0, &gh);
1189 ret = gfs2_glock_nq(&gh);
1190 if (ret)
1191 goto out_uninit;
1192
1193 if (!(mode & FALLOC_FL_KEEP_SIZE) &&
1194 (offset + len) > inode->i_size) {
1195 ret = inode_newsize_ok(inode, offset + len);
1196 if (ret)
1197 goto out_unlock;
1198 }
1199
1200 ret = get_write_access(inode);
1201 if (ret)
1202 goto out_unlock;
1203
1204 if (mode & FALLOC_FL_PUNCH_HOLE) {
1205 ret = __gfs2_punch_hole(file, offset, len);
1206 } else {
1207 ret = __gfs2_fallocate(file, mode, offset, len);
1208 if (ret)
1209 gfs2_rs_deltree(&ip->i_res);
1210 }
1211
1212 put_write_access(inode);
1213 out_unlock:
1214 gfs2_glock_dq(&gh);
1215 out_uninit:
1216 gfs2_holder_uninit(&gh);
1217 inode_unlock(inode);
1218 return ret;
1219 }
1220
gfs2_file_splice_write(struct pipe_inode_info * pipe,struct file * out,loff_t * ppos,size_t len,unsigned int flags)1221 static ssize_t gfs2_file_splice_write(struct pipe_inode_info *pipe,
1222 struct file *out, loff_t *ppos,
1223 size_t len, unsigned int flags)
1224 {
1225 ssize_t ret;
1226
1227 gfs2_size_hint(out, *ppos, len);
1228
1229 ret = iter_file_splice_write(pipe, out, ppos, len, flags);
1230 return ret;
1231 }
1232
1233 #ifdef CONFIG_GFS2_FS_LOCKING_DLM
1234
1235 /**
1236 * gfs2_lock - acquire/release a posix lock on a file
1237 * @file: the file pointer
1238 * @cmd: either modify or retrieve lock state, possibly wait
1239 * @fl: type and range of lock
1240 *
1241 * Returns: errno
1242 */
1243
gfs2_lock(struct file * file,int cmd,struct file_lock * fl)1244 static int gfs2_lock(struct file *file, int cmd, struct file_lock *fl)
1245 {
1246 struct gfs2_inode *ip = GFS2_I(file->f_mapping->host);
1247 struct gfs2_sbd *sdp = GFS2_SB(file->f_mapping->host);
1248 struct lm_lockstruct *ls = &sdp->sd_lockstruct;
1249
1250 if (!(fl->fl_flags & FL_POSIX))
1251 return -ENOLCK;
1252 if (__mandatory_lock(&ip->i_inode) && fl->fl_type != F_UNLCK)
1253 return -ENOLCK;
1254
1255 if (cmd == F_CANCELLK) {
1256 /* Hack: */
1257 cmd = F_SETLK;
1258 fl->fl_type = F_UNLCK;
1259 }
1260 if (unlikely(gfs2_withdrawn(sdp))) {
1261 if (fl->fl_type == F_UNLCK)
1262 locks_lock_file_wait(file, fl);
1263 return -EIO;
1264 }
1265 if (IS_GETLK(cmd))
1266 return dlm_posix_get(ls->ls_dlm, ip->i_no_addr, file, fl);
1267 else if (fl->fl_type == F_UNLCK)
1268 return dlm_posix_unlock(ls->ls_dlm, ip->i_no_addr, file, fl);
1269 else
1270 return dlm_posix_lock(ls->ls_dlm, ip->i_no_addr, file, cmd, fl);
1271 }
1272
do_flock(struct file * file,int cmd,struct file_lock * fl)1273 static int do_flock(struct file *file, int cmd, struct file_lock *fl)
1274 {
1275 struct gfs2_file *fp = file->private_data;
1276 struct gfs2_holder *fl_gh = &fp->f_fl_gh;
1277 struct gfs2_inode *ip = GFS2_I(file_inode(file));
1278 struct gfs2_glock *gl;
1279 unsigned int state;
1280 u16 flags;
1281 int error = 0;
1282 int sleeptime;
1283
1284 state = (fl->fl_type == F_WRLCK) ? LM_ST_EXCLUSIVE : LM_ST_SHARED;
1285 flags = (IS_SETLKW(cmd) ? 0 : LM_FLAG_TRY_1CB) | GL_EXACT;
1286
1287 mutex_lock(&fp->f_fl_mutex);
1288
1289 if (gfs2_holder_initialized(fl_gh)) {
1290 struct file_lock request;
1291 if (fl_gh->gh_state == state)
1292 goto out;
1293 locks_init_lock(&request);
1294 request.fl_type = F_UNLCK;
1295 request.fl_flags = FL_FLOCK;
1296 locks_lock_file_wait(file, &request);
1297 gfs2_glock_dq(fl_gh);
1298 gfs2_holder_reinit(state, flags, fl_gh);
1299 } else {
1300 error = gfs2_glock_get(GFS2_SB(&ip->i_inode), ip->i_no_addr,
1301 &gfs2_flock_glops, CREATE, &gl);
1302 if (error)
1303 goto out;
1304 gfs2_holder_init(gl, state, flags, fl_gh);
1305 gfs2_glock_put(gl);
1306 }
1307 for (sleeptime = 1; sleeptime <= 4; sleeptime <<= 1) {
1308 error = gfs2_glock_nq(fl_gh);
1309 if (error != GLR_TRYFAILED)
1310 break;
1311 fl_gh->gh_flags = LM_FLAG_TRY | GL_EXACT;
1312 fl_gh->gh_error = 0;
1313 msleep(sleeptime);
1314 }
1315 if (error) {
1316 gfs2_holder_uninit(fl_gh);
1317 if (error == GLR_TRYFAILED)
1318 error = -EAGAIN;
1319 } else {
1320 error = locks_lock_file_wait(file, fl);
1321 gfs2_assert_warn(GFS2_SB(&ip->i_inode), !error);
1322 }
1323
1324 out:
1325 mutex_unlock(&fp->f_fl_mutex);
1326 return error;
1327 }
1328
do_unflock(struct file * file,struct file_lock * fl)1329 static void do_unflock(struct file *file, struct file_lock *fl)
1330 {
1331 struct gfs2_file *fp = file->private_data;
1332 struct gfs2_holder *fl_gh = &fp->f_fl_gh;
1333
1334 mutex_lock(&fp->f_fl_mutex);
1335 locks_lock_file_wait(file, fl);
1336 if (gfs2_holder_initialized(fl_gh)) {
1337 gfs2_glock_dq(fl_gh);
1338 gfs2_holder_uninit(fl_gh);
1339 }
1340 mutex_unlock(&fp->f_fl_mutex);
1341 }
1342
1343 /**
1344 * gfs2_flock - acquire/release a flock lock on a file
1345 * @file: the file pointer
1346 * @cmd: either modify or retrieve lock state, possibly wait
1347 * @fl: type and range of lock
1348 *
1349 * Returns: errno
1350 */
1351
gfs2_flock(struct file * file,int cmd,struct file_lock * fl)1352 static int gfs2_flock(struct file *file, int cmd, struct file_lock *fl)
1353 {
1354 if (!(fl->fl_flags & FL_FLOCK))
1355 return -ENOLCK;
1356 if (fl->fl_type & LOCK_MAND)
1357 return -EOPNOTSUPP;
1358
1359 if (fl->fl_type == F_UNLCK) {
1360 do_unflock(file, fl);
1361 return 0;
1362 } else {
1363 return do_flock(file, cmd, fl);
1364 }
1365 }
1366
1367 const struct file_operations gfs2_file_fops = {
1368 .llseek = gfs2_llseek,
1369 .read_iter = gfs2_file_read_iter,
1370 .write_iter = gfs2_file_write_iter,
1371 .iopoll = iomap_dio_iopoll,
1372 .unlocked_ioctl = gfs2_ioctl,
1373 .compat_ioctl = gfs2_compat_ioctl,
1374 .mmap = gfs2_mmap,
1375 .open = gfs2_open,
1376 .release = gfs2_release,
1377 .fsync = gfs2_fsync,
1378 .lock = gfs2_lock,
1379 .flock = gfs2_flock,
1380 .splice_read = generic_file_splice_read,
1381 .splice_write = gfs2_file_splice_write,
1382 .setlease = simple_nosetlease,
1383 .fallocate = gfs2_fallocate,
1384 };
1385
1386 const struct file_operations gfs2_dir_fops = {
1387 .iterate_shared = gfs2_readdir,
1388 .unlocked_ioctl = gfs2_ioctl,
1389 .compat_ioctl = gfs2_compat_ioctl,
1390 .open = gfs2_open,
1391 .release = gfs2_release,
1392 .fsync = gfs2_fsync,
1393 .lock = gfs2_lock,
1394 .flock = gfs2_flock,
1395 .llseek = default_llseek,
1396 };
1397
1398 #endif /* CONFIG_GFS2_FS_LOCKING_DLM */
1399
1400 const struct file_operations gfs2_file_fops_nolock = {
1401 .llseek = gfs2_llseek,
1402 .read_iter = gfs2_file_read_iter,
1403 .write_iter = gfs2_file_write_iter,
1404 .iopoll = iomap_dio_iopoll,
1405 .unlocked_ioctl = gfs2_ioctl,
1406 .compat_ioctl = gfs2_compat_ioctl,
1407 .mmap = gfs2_mmap,
1408 .open = gfs2_open,
1409 .release = gfs2_release,
1410 .fsync = gfs2_fsync,
1411 .splice_read = generic_file_splice_read,
1412 .splice_write = gfs2_file_splice_write,
1413 .setlease = generic_setlease,
1414 .fallocate = gfs2_fallocate,
1415 };
1416
1417 const struct file_operations gfs2_dir_fops_nolock = {
1418 .iterate_shared = gfs2_readdir,
1419 .unlocked_ioctl = gfs2_ioctl,
1420 .compat_ioctl = gfs2_compat_ioctl,
1421 .open = gfs2_open,
1422 .release = gfs2_release,
1423 .fsync = gfs2_fsync,
1424 .llseek = default_llseek,
1425 };
1426
1427