]> bbs.cooldavid.org Git - net-next-2.6.git/blame - fs/gfs2/file.c
GFS2: New truncate sequence
[net-next-2.6.git] / fs / gfs2 / file.c
CommitLineData
b3b94faa
DT
1/*
2 * Copyright (C) Sistina Software, Inc. 1997-2003 All rights reserved.
3a8a9a10 3 * Copyright (C) 2004-2006 Red Hat, Inc. All rights reserved.
b3b94faa
DT
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
e9fc2aa0 7 * of the GNU General Public License version 2.
b3b94faa
DT
8 */
9
b3b94faa
DT
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>
f58ba889 18#include <linux/mount.h>
18ec7d5c 19#include <linux/fs.h>
5c676f6d 20#include <linux/gfs2_ondisk.h>
71b86f56
SW
21#include <linux/ext2_fs.h>
22#include <linux/crc32.h>
33c3de32 23#include <linux/writeback.h>
b3b94faa 24#include <asm/uaccess.h>
f057f6cd
SW
25#include <linux/dlm.h>
26#include <linux/dlm_plock.h>
b3b94faa
DT
27
28#include "gfs2.h"
5c676f6d 29#include "incore.h"
b3b94faa
DT
30#include "bmap.h"
31#include "dir.h"
32#include "glock.h"
33#include "glops.h"
34#include "inode.h"
b3b94faa
DT
35#include "log.h"
36#include "meta_io.h"
b3b94faa
DT
37#include "quota.h"
38#include "rgrp.h"
39#include "trans.h"
5c676f6d 40#include "util.h"
b3b94faa 41
b3b94faa
DT
42/**
43 * gfs2_llseek - seek to a location in a file
44 * @file: the file
45 * @offset: the offset
46 * @origin: Where to seek from (SEEK_SET, SEEK_CUR, or SEEK_END)
47 *
48 * SEEK_END requires the glock for the file because it references the
49 * file's size.
50 *
51 * Returns: The new offset, or errno
52 */
53
54static loff_t gfs2_llseek(struct file *file, loff_t offset, int origin)
55{
feaa7bba 56 struct gfs2_inode *ip = GFS2_I(file->f_mapping->host);
b3b94faa
DT
57 struct gfs2_holder i_gh;
58 loff_t error;
59
b3b94faa
DT
60 if (origin == 2) {
61 error = gfs2_glock_nq_init(ip->i_gl, LM_ST_SHARED, LM_FLAG_ANY,
62 &i_gh);
63 if (!error) {
9465efc9 64 error = generic_file_llseek_unlocked(file, offset, origin);
b3b94faa
DT
65 gfs2_glock_dq_uninit(&i_gh);
66 }
67 } else
9465efc9 68 error = generic_file_llseek_unlocked(file, offset, origin);
b3b94faa
DT
69
70 return error;
71}
72
b3b94faa 73/**
f0e522a9 74 * gfs2_readdir - Read directory entries from a directory
b3b94faa
DT
75 * @file: The directory to read from
76 * @dirent: Buffer for dirents
77 * @filldir: Function used to do the copying
78 *
79 * Returns: errno
80 */
81
f0e522a9 82static int gfs2_readdir(struct file *file, void *dirent, filldir_t filldir)
b3b94faa 83{
71b86f56 84 struct inode *dir = file->f_mapping->host;
feaa7bba 85 struct gfs2_inode *dip = GFS2_I(dir);
b3b94faa 86 struct gfs2_holder d_gh;
cd915493 87 u64 offset = file->f_pos;
b3b94faa
DT
88 int error;
89
719ee344
SW
90 gfs2_holder_init(dip->i_gl, LM_ST_SHARED, 0, &d_gh);
91 error = gfs2_glock_nq(&d_gh);
b3b94faa
DT
92 if (error) {
93 gfs2_holder_uninit(&d_gh);
94 return error;
95 }
96
3699e3a4 97 error = gfs2_dir_read(dir, &offset, dirent, filldir);
b3b94faa
DT
98
99 gfs2_glock_dq_uninit(&d_gh);
100
b3b94faa
DT
101 file->f_pos = offset;
102
b3b94faa
DT
103 return error;
104}
105
128e5eba
SW
106/**
107 * fsflags_cvt
108 * @table: A table of 32 u32 flags
109 * @val: a 32 bit value to convert
110 *
111 * This function can be used to convert between fsflags values and
112 * GFS2's own flags values.
113 *
114 * Returns: the converted flags
115 */
116static u32 fsflags_cvt(const u32 *table, u32 val)
117{
118 u32 res = 0;
119 while(val) {
120 if (val & 1)
121 res |= *table;
122 table++;
123 val >>= 1;
124 }
125 return res;
126}
b3b94faa 127
128e5eba
SW
128static const u32 fsflags_to_gfs2[32] = {
129 [3] = GFS2_DIF_SYNC,
130 [4] = GFS2_DIF_IMMUTABLE,
131 [5] = GFS2_DIF_APPENDONLY,
132 [7] = GFS2_DIF_NOATIME,
133 [12] = GFS2_DIF_EXHASH,
b9af7ca6 134 [14] = GFS2_DIF_INHERIT_JDATA,
71b86f56
SW
135};
136
128e5eba
SW
137static const u32 gfs2_to_fsflags[32] = {
138 [gfs2fl_Sync] = FS_SYNC_FL,
139 [gfs2fl_Immutable] = FS_IMMUTABLE_FL,
140 [gfs2fl_AppendOnly] = FS_APPEND_FL,
141 [gfs2fl_NoAtime] = FS_NOATIME_FL,
142 [gfs2fl_ExHash] = FS_INDEX_FL,
128e5eba 143 [gfs2fl_InheritJdata] = FS_JOURNAL_DATA_FL,
7ea9ea83 144};
71b86f56 145
b09e593d 146static int gfs2_get_flags(struct file *filp, u32 __user *ptr)
71b86f56 147{
81454098 148 struct inode *inode = filp->f_path.dentry->d_inode;
feaa7bba 149 struct gfs2_inode *ip = GFS2_I(inode);
71b86f56
SW
150 struct gfs2_holder gh;
151 int error;
128e5eba 152 u32 fsflags;
71b86f56 153
719ee344
SW
154 gfs2_holder_init(ip->i_gl, LM_ST_SHARED, 0, &gh);
155 error = gfs2_glock_nq(&gh);
71b86f56
SW
156 if (error)
157 return error;
907b9bce 158
383f01fb
SW
159 fsflags = fsflags_cvt(gfs2_to_fsflags, ip->i_diskflags);
160 if (!S_ISDIR(inode->i_mode) && ip->i_diskflags & GFS2_DIF_JDATA)
c9f6a6bb 161 fsflags |= FS_JOURNAL_DATA_FL;
128e5eba 162 if (put_user(fsflags, ptr))
71b86f56
SW
163 error = -EFAULT;
164
3cc3f710 165 gfs2_glock_dq(&gh);
71b86f56
SW
166 gfs2_holder_uninit(&gh);
167 return error;
168}
169
6b124d8d
SW
170void gfs2_set_inode_flags(struct inode *inode)
171{
172 struct gfs2_inode *ip = GFS2_I(inode);
6b124d8d
SW
173 unsigned int flags = inode->i_flags;
174
175 flags &= ~(S_SYNC|S_APPEND|S_IMMUTABLE|S_NOATIME|S_DIRSYNC);
383f01fb 176 if (ip->i_diskflags & GFS2_DIF_IMMUTABLE)
6b124d8d 177 flags |= S_IMMUTABLE;
383f01fb 178 if (ip->i_diskflags & GFS2_DIF_APPENDONLY)
6b124d8d 179 flags |= S_APPEND;
383f01fb 180 if (ip->i_diskflags & GFS2_DIF_NOATIME)
6b124d8d 181 flags |= S_NOATIME;
383f01fb 182 if (ip->i_diskflags & GFS2_DIF_SYNC)
6b124d8d
SW
183 flags |= S_SYNC;
184 inode->i_flags = flags;
185}
186
71b86f56
SW
187/* Flags that can be set by user space */
188#define GFS2_FLAGS_USER_SET (GFS2_DIF_JDATA| \
71b86f56
SW
189 GFS2_DIF_IMMUTABLE| \
190 GFS2_DIF_APPENDONLY| \
191 GFS2_DIF_NOATIME| \
192 GFS2_DIF_SYNC| \
193 GFS2_DIF_SYSTEM| \
71b86f56
SW
194 GFS2_DIF_INHERIT_JDATA)
195
196/**
197 * gfs2_set_flags - set flags on an inode
198 * @inode: The inode
199 * @flags: The flags to set
200 * @mask: Indicates which flags are valid
201 *
202 */
b09e593d 203static int do_gfs2_set_flags(struct file *filp, u32 reqflags, u32 mask)
71b86f56 204{
81454098 205 struct inode *inode = filp->f_path.dentry->d_inode;
feaa7bba
SW
206 struct gfs2_inode *ip = GFS2_I(inode);
207 struct gfs2_sbd *sdp = GFS2_SB(inode);
71b86f56
SW
208 struct buffer_head *bh;
209 struct gfs2_holder gh;
210 int error;
55eccc6d 211 u32 new_flags, flags;
71b86f56 212
f58ba889 213 error = mnt_want_write(filp->f_path.mnt);
52f341cf 214 if (error)
71b86f56
SW
215 return error;
216
f58ba889
MS
217 error = gfs2_glock_nq_init(ip->i_gl, LM_ST_EXCLUSIVE, 0, &gh);
218 if (error)
219 goto out_drop_write;
220
7df0e039
SW
221 error = -EACCES;
222 if (!is_owner_or_cap(inode))
223 goto out;
224
225 error = 0;
383f01fb 226 flags = ip->i_diskflags;
55eccc6d 227 new_flags = (flags & ~mask) | (reqflags & mask);
71b86f56
SW
228 if ((new_flags ^ flags) == 0)
229 goto out;
230
231 error = -EINVAL;
232 if ((new_flags ^ flags) & ~GFS2_FLAGS_USER_SET)
233 goto out;
234
71b86f56
SW
235 error = -EPERM;
236 if (IS_IMMUTABLE(inode) && (new_flags & GFS2_DIF_IMMUTABLE))
237 goto out;
238 if (IS_APPEND(inode) && (new_flags & GFS2_DIF_APPENDONLY))
239 goto out;
907b9bce 240 if (((new_flags ^ flags) & GFS2_DIF_IMMUTABLE) &&
b9cb9813 241 !capable(CAP_LINUX_IMMUTABLE))
71b86f56 242 goto out;
b9cb9813 243 if (!IS_IMMUTABLE(inode)) {
f58ba889 244 error = gfs2_permission(inode, MAY_WRITE);
b9cb9813
SW
245 if (error)
246 goto out;
247 }
5561093e
SW
248 if ((flags ^ new_flags) & GFS2_DIF_JDATA) {
249 if (flags & GFS2_DIF_JDATA)
250 gfs2_log_flush(sdp, ip->i_gl);
251 error = filemap_fdatawrite(inode->i_mapping);
252 if (error)
253 goto out;
254 error = filemap_fdatawait(inode->i_mapping);
255 if (error)
256 goto out;
257 }
55eccc6d 258 error = gfs2_trans_begin(sdp, RES_DINODE, 0);
71b86f56
SW
259 if (error)
260 goto out;
55eccc6d
SW
261 error = gfs2_meta_inode_buffer(ip, &bh);
262 if (error)
263 goto out_trans_end;
71b86f56 264 gfs2_trans_add_bh(ip->i_gl, bh, 1);
383f01fb 265 ip->i_diskflags = new_flags;
539e5d6b 266 gfs2_dinode_out(ip, bh->b_data);
71b86f56 267 brelse(bh);
6b124d8d 268 gfs2_set_inode_flags(inode);
5561093e 269 gfs2_set_aops(inode);
55eccc6d
SW
270out_trans_end:
271 gfs2_trans_end(sdp);
71b86f56
SW
272out:
273 gfs2_glock_dq_uninit(&gh);
f58ba889
MS
274out_drop_write:
275 mnt_drop_write(filp->f_path.mnt);
71b86f56
SW
276 return error;
277}
278
b09e593d 279static int gfs2_set_flags(struct file *filp, u32 __user *ptr)
71b86f56 280{
b9af7ca6 281 struct inode *inode = filp->f_path.dentry->d_inode;
128e5eba 282 u32 fsflags, gfsflags;
7df0e039 283
128e5eba 284 if (get_user(fsflags, ptr))
71b86f56 285 return -EFAULT;
7df0e039 286
128e5eba 287 gfsflags = fsflags_cvt(fsflags_to_gfs2, fsflags);
b9af7ca6
SW
288 if (!S_ISDIR(inode->i_mode)) {
289 if (gfsflags & GFS2_DIF_INHERIT_JDATA)
290 gfsflags ^= (GFS2_DIF_JDATA | GFS2_DIF_INHERIT_JDATA);
b9af7ca6
SW
291 return do_gfs2_set_flags(filp, gfsflags, ~0);
292 }
293 return do_gfs2_set_flags(filp, gfsflags, ~GFS2_DIF_JDATA);
71b86f56
SW
294}
295
b09e593d 296static long gfs2_ioctl(struct file *filp, unsigned int cmd, unsigned long arg)
71b86f56
SW
297{
298 switch(cmd) {
128e5eba 299 case FS_IOC_GETFLAGS:
b09e593d 300 return gfs2_get_flags(filp, (u32 __user *)arg);
128e5eba 301 case FS_IOC_SETFLAGS:
b09e593d 302 return gfs2_set_flags(filp, (u32 __user *)arg);
71b86f56
SW
303 }
304 return -ENOTTY;
305}
306
3cc3f710
SW
307/**
308 * gfs2_allocate_page_backing - Use bmap to allocate blocks
309 * @page: The (locked) page to allocate backing for
310 *
311 * We try to allocate all the blocks required for the page in
312 * one go. This might fail for various reasons, so we keep
313 * trying until all the blocks to back this page are allocated.
314 * If some of the blocks are already allocated, thats ok too.
315 */
316
317static int gfs2_allocate_page_backing(struct page *page)
318{
319 struct inode *inode = page->mapping->host;
320 struct buffer_head bh;
321 unsigned long size = PAGE_CACHE_SIZE;
322 u64 lblock = page->index << (PAGE_CACHE_SHIFT - inode->i_blkbits);
323
324 do {
325 bh.b_state = 0;
326 bh.b_size = size;
e9e1ef2b 327 gfs2_block_map(inode, lblock, &bh, 1);
3cc3f710
SW
328 if (!buffer_mapped(&bh))
329 return -EIO;
330 size -= bh.b_size;
331 lblock += (bh.b_size >> inode->i_blkbits);
332 } while(size > 0);
333 return 0;
334}
335
336/**
337 * gfs2_page_mkwrite - Make a shared, mmap()ed, page writable
338 * @vma: The virtual memory area
339 * @page: The page which is about to become writable
340 *
341 * When the page becomes writable, we need to ensure that we have
342 * blocks allocated on disk to back that page.
343 */
344
c2ec175c 345static int gfs2_page_mkwrite(struct vm_area_struct *vma, struct vm_fault *vmf)
3cc3f710 346{
c2ec175c 347 struct page *page = vmf->page;
3cc3f710
SW
348 struct inode *inode = vma->vm_file->f_path.dentry->d_inode;
349 struct gfs2_inode *ip = GFS2_I(inode);
350 struct gfs2_sbd *sdp = GFS2_SB(inode);
351 unsigned long last_index;
c8f554b9 352 u64 pos = page->index << PAGE_CACHE_SHIFT;
3cc3f710 353 unsigned int data_blocks, ind_blocks, rblocks;
3cc3f710
SW
354 struct gfs2_holder gh;
355 struct gfs2_alloc *al;
356 int ret;
357
719ee344
SW
358 gfs2_holder_init(ip->i_gl, LM_ST_EXCLUSIVE, 0, &gh);
359 ret = gfs2_glock_nq(&gh);
3cc3f710
SW
360 if (ret)
361 goto out;
362
9c538837
SW
363 set_bit(GLF_DIRTY, &ip->i_gl->gl_flags);
364 set_bit(GIF_SW_PAGED, &ip->i_flags);
365
461cb419 366 if (!gfs2_write_alloc_required(ip, pos, PAGE_CACHE_SIZE))
3cc3f710 367 goto out_unlock;
6dbd8224 368 ret = -ENOMEM;
3cc3f710 369 al = gfs2_alloc_get(ip);
6dbd8224
SW
370 if (al == NULL)
371 goto out_unlock;
372
d82661d9 373 ret = gfs2_quota_lock_check(ip);
3cc3f710
SW
374 if (ret)
375 goto out_alloc_put;
7ed122e4 376 gfs2_write_calc_reserv(ip, PAGE_CACHE_SIZE, &data_blocks, &ind_blocks);
3cc3f710
SW
377 al->al_requested = data_blocks + ind_blocks;
378 ret = gfs2_inplace_reserve(ip);
379 if (ret)
380 goto out_quota_unlock;
381
382 rblocks = RES_DINODE + ind_blocks;
383 if (gfs2_is_jdata(ip))
384 rblocks += data_blocks ? data_blocks : 1;
385 if (ind_blocks || data_blocks)
386 rblocks += RES_STATFS + RES_QUOTA;
387 ret = gfs2_trans_begin(sdp, rblocks, 0);
388 if (ret)
389 goto out_trans_fail;
390
391 lock_page(page);
392 ret = -EINVAL;
393 last_index = ip->i_inode.i_size >> PAGE_CACHE_SHIFT;
394 if (page->index > last_index)
395 goto out_unlock_page;
b7fe2e39 396 ret = 0;
3cc3f710
SW
397 if (!PageUptodate(page) || page->mapping != ip->i_inode.i_mapping)
398 goto out_unlock_page;
399 if (gfs2_is_stuffed(ip)) {
400 ret = gfs2_unstuff_dinode(ip, page);
401 if (ret)
402 goto out_unlock_page;
403 }
404 ret = gfs2_allocate_page_backing(page);
405
406out_unlock_page:
407 unlock_page(page);
408 gfs2_trans_end(sdp);
409out_trans_fail:
410 gfs2_inplace_release(ip);
411out_quota_unlock:
412 gfs2_quota_unlock(ip);
413out_alloc_put:
414 gfs2_alloc_put(ip);
415out_unlock:
416 gfs2_glock_dq(&gh);
417out:
418 gfs2_holder_uninit(&gh);
e56985da
SW
419 if (ret == -ENOMEM)
420 ret = VM_FAULT_OOM;
421 else if (ret)
c2ec175c 422 ret = VM_FAULT_SIGBUS;
3cc3f710
SW
423 return ret;
424}
425
f0f37e2f 426static const struct vm_operations_struct gfs2_vm_ops = {
3cc3f710
SW
427 .fault = filemap_fault,
428 .page_mkwrite = gfs2_page_mkwrite,
429};
430
b3b94faa
DT
431/**
432 * gfs2_mmap -
433 * @file: The file to map
434 * @vma: The VMA which described the mapping
435 *
48bf2b17
SW
436 * There is no need to get a lock here unless we should be updating
437 * atime. We ignore any locking errors since the only consequence is
438 * a missed atime update (which will just be deferred until later).
439 *
440 * Returns: 0
b3b94faa
DT
441 */
442
443static int gfs2_mmap(struct file *file, struct vm_area_struct *vma)
444{
feaa7bba 445 struct gfs2_inode *ip = GFS2_I(file->f_mapping->host);
b3b94faa 446
48bf2b17
SW
447 if (!(file->f_flags & O_NOATIME)) {
448 struct gfs2_holder i_gh;
449 int error;
b3b94faa 450
48bf2b17
SW
451 gfs2_holder_init(ip->i_gl, LM_ST_EXCLUSIVE, 0, &i_gh);
452 error = gfs2_glock_nq(&i_gh);
453 file_accessed(file);
454 if (error == 0)
455 gfs2_glock_dq_uninit(&i_gh);
456 }
3cc3f710 457 vma->vm_ops = &gfs2_vm_ops;
48bf2b17 458 vma->vm_flags |= VM_CAN_NONLINEAR;
b3b94faa 459
48bf2b17 460 return 0;
b3b94faa
DT
461}
462
463/**
464 * gfs2_open - open a file
465 * @inode: the inode to open
466 * @file: the struct file for this opening
467 *
468 * Returns: errno
469 */
470
471static int gfs2_open(struct inode *inode, struct file *file)
472{
feaa7bba 473 struct gfs2_inode *ip = GFS2_I(inode);
b3b94faa
DT
474 struct gfs2_holder i_gh;
475 struct gfs2_file *fp;
476 int error;
477
b3b94faa
DT
478 fp = kzalloc(sizeof(struct gfs2_file), GFP_KERNEL);
479 if (!fp)
480 return -ENOMEM;
481
f55ab26a 482 mutex_init(&fp->f_fl_mutex);
b3b94faa 483
feaa7bba 484 gfs2_assert_warn(GFS2_SB(inode), !file->private_data);
5c676f6d 485 file->private_data = fp;
b3b94faa 486
b60623c2 487 if (S_ISREG(ip->i_inode.i_mode)) {
b3b94faa
DT
488 error = gfs2_glock_nq_init(ip->i_gl, LM_ST_SHARED, LM_FLAG_ANY,
489 &i_gh);
490 if (error)
491 goto fail;
492
493 if (!(file->f_flags & O_LARGEFILE) &&
c9e98886 494 ip->i_disksize > MAX_NON_LFS) {
a9c62a18 495 error = -EOVERFLOW;
b3b94faa
DT
496 goto fail_gunlock;
497 }
498
b3b94faa
DT
499 gfs2_glock_dq_uninit(&i_gh);
500 }
501
502 return 0;
503
420b9e5e 504fail_gunlock:
b3b94faa 505 gfs2_glock_dq_uninit(&i_gh);
420b9e5e 506fail:
5c676f6d 507 file->private_data = NULL;
b3b94faa 508 kfree(fp);
b3b94faa
DT
509 return error;
510}
511
512/**
513 * gfs2_close - called to close a struct file
514 * @inode: the inode the struct file belongs to
515 * @file: the struct file being closed
516 *
517 * Returns: errno
518 */
519
520static int gfs2_close(struct inode *inode, struct file *file)
521{
5c676f6d 522 struct gfs2_sbd *sdp = inode->i_sb->s_fs_info;
b3b94faa
DT
523 struct gfs2_file *fp;
524
5c676f6d
SW
525 fp = file->private_data;
526 file->private_data = NULL;
b3b94faa
DT
527
528 if (gfs2_assert_warn(sdp, fp))
529 return -EIO;
530
531 kfree(fp);
532
533 return 0;
534}
535
536/**
537 * gfs2_fsync - sync the dirty data for a file (across the cluster)
538 * @file: the file that points to the dentry (we ignore this)
539 * @dentry: the dentry that points to the inode to sync
540 *
33c3de32
SW
541 * The VFS will flush "normal" data for us. We only need to worry
542 * about metadata here. For journaled data, we just do a log flush
543 * as we can't avoid it. Otherwise we can just bale out if datasync
544 * is set. For stuffed inodes we must flush the log in order to
545 * ensure that all data is on disk.
546 *
34126f9f
SW
547 * The call to write_inode_now() is there to write back metadata and
548 * the inode itself. It does also try and write the data, but thats
549 * (hopefully) a no-op due to the VFS having already called filemap_fdatawrite()
550 * for us.
551 *
b3b94faa
DT
552 * Returns: errno
553 */
554
7ea80859 555static int gfs2_fsync(struct file *file, int datasync)
b3b94faa 556{
7ea80859 557 struct inode *inode = file->f_mapping->host;
33c3de32
SW
558 int sync_state = inode->i_state & (I_DIRTY_SYNC|I_DIRTY_DATASYNC);
559 int ret = 0;
33c3de32
SW
560
561 if (gfs2_is_jdata(GFS2_I(inode))) {
562 gfs2_log_flush(GFS2_SB(inode), GFS2_I(inode)->i_gl);
563 return 0;
564 }
b3b94faa 565
33c3de32
SW
566 if (sync_state != 0) {
567 if (!datasync)
34126f9f 568 ret = write_inode_now(inode, 0);
b3b94faa 569
33c3de32
SW
570 if (gfs2_is_stuffed(GFS2_I(inode)))
571 gfs2_log_flush(GFS2_SB(inode), GFS2_I(inode)->i_gl);
572 }
573
574 return ret;
b3b94faa
DT
575}
576
56aa616a
SW
577/**
578 * gfs2_file_aio_write - Perform a write to a file
579 * @iocb: The io context
580 * @iov: The data to write
581 * @nr_segs: Number of @iov segments
582 * @pos: The file position
583 *
584 * We have to do a lock/unlock here to refresh the inode size for
585 * O_APPEND writes, otherwise we can land up writing at the wrong
586 * offset. There is still a race, but provided the app is using its
587 * own file locking, this will make O_APPEND work as expected.
588 *
589 */
590
591static ssize_t gfs2_file_aio_write(struct kiocb *iocb, const struct iovec *iov,
592 unsigned long nr_segs, loff_t pos)
593{
594 struct file *file = iocb->ki_filp;
595
596 if (file->f_flags & O_APPEND) {
597 struct dentry *dentry = file->f_dentry;
598 struct gfs2_inode *ip = GFS2_I(dentry->d_inode);
599 struct gfs2_holder gh;
600 int ret;
601
602 ret = gfs2_glock_nq_init(ip->i_gl, LM_ST_SHARED, 0, &gh);
603 if (ret)
604 return ret;
605 gfs2_glock_dq_uninit(&gh);
606 }
607
608 return generic_file_aio_write(iocb, iov, nr_segs, pos);
609}
610
f057f6cd
SW
611#ifdef CONFIG_GFS2_FS_LOCKING_DLM
612
60446067
ME
613/**
614 * gfs2_setlease - acquire/release a file lease
615 * @file: the file pointer
616 * @arg: lease type
617 * @fl: file lock
618 *
f057f6cd
SW
619 * We don't currently have a way to enforce a lease across the whole
620 * cluster; until we do, disable leases (by just returning -EINVAL),
621 * unless the administrator has requested purely local locking.
622 *
60446067
ME
623 * Returns: errno
624 */
625
626static int gfs2_setlease(struct file *file, long arg, struct file_lock **fl)
627{
f057f6cd 628 return -EINVAL;
da755fdb
SW
629}
630
b3b94faa
DT
631/**
632 * gfs2_lock - acquire/release a posix lock on a file
633 * @file: the file pointer
634 * @cmd: either modify or retrieve lock state, possibly wait
635 * @fl: type and range of lock
636 *
637 * Returns: errno
638 */
639
640static int gfs2_lock(struct file *file, int cmd, struct file_lock *fl)
641{
feaa7bba
SW
642 struct gfs2_inode *ip = GFS2_I(file->f_mapping->host);
643 struct gfs2_sbd *sdp = GFS2_SB(file->f_mapping->host);
f057f6cd 644 struct lm_lockstruct *ls = &sdp->sd_lockstruct;
b3b94faa 645
b3b94faa
DT
646 if (!(fl->fl_flags & FL_POSIX))
647 return -ENOLCK;
720e7749 648 if (__mandatory_lock(&ip->i_inode) && fl->fl_type != F_UNLCK)
b3b94faa
DT
649 return -ENOLCK;
650
586759f0
ME
651 if (cmd == F_CANCELLK) {
652 /* Hack: */
653 cmd = F_SETLK;
654 fl->fl_type = F_UNLCK;
655 }
f057f6cd
SW
656 if (unlikely(test_bit(SDF_SHUTDOWN, &sdp->sd_flags)))
657 return -EIO;
b3b94faa 658 if (IS_GETLK(cmd))
f057f6cd 659 return dlm_posix_get(ls->ls_dlm, ip->i_no_addr, file, fl);
b3b94faa 660 else if (fl->fl_type == F_UNLCK)
f057f6cd 661 return dlm_posix_unlock(ls->ls_dlm, ip->i_no_addr, file, fl);
b3b94faa 662 else
f057f6cd 663 return dlm_posix_lock(ls->ls_dlm, ip->i_no_addr, file, cmd, fl);
b3b94faa
DT
664}
665
b3b94faa
DT
666static int do_flock(struct file *file, int cmd, struct file_lock *fl)
667{
5c676f6d 668 struct gfs2_file *fp = file->private_data;
b3b94faa 669 struct gfs2_holder *fl_gh = &fp->f_fl_gh;
81454098 670 struct gfs2_inode *ip = GFS2_I(file->f_path.dentry->d_inode);
b3b94faa
DT
671 struct gfs2_glock *gl;
672 unsigned int state;
673 int flags;
674 int error = 0;
675
676 state = (fl->fl_type == F_WRLCK) ? LM_ST_EXCLUSIVE : LM_ST_SHARED;
6802e340 677 flags = (IS_SETLKW(cmd) ? 0 : LM_FLAG_TRY) | GL_EXACT | GL_NOCACHE;
b3b94faa 678
f55ab26a 679 mutex_lock(&fp->f_fl_mutex);
b3b94faa
DT
680
681 gl = fl_gh->gh_gl;
682 if (gl) {
683 if (fl_gh->gh_state == state)
684 goto out;
b3b94faa 685 flock_lock_file_wait(file,
907b9bce 686 &(struct file_lock){.fl_type = F_UNLCK});
b4c20166
AD
687 gfs2_glock_dq_wait(fl_gh);
688 gfs2_holder_reinit(state, flags, fl_gh);
b3b94faa 689 } else {
6802e340
SW
690 error = gfs2_glock_get(GFS2_SB(&ip->i_inode), ip->i_no_addr,
691 &gfs2_flock_glops, CREATE, &gl);
b3b94faa
DT
692 if (error)
693 goto out;
b4c20166
AD
694 gfs2_holder_init(gl, state, flags, fl_gh);
695 gfs2_glock_put(gl);
b3b94faa 696 }
b3b94faa
DT
697 error = gfs2_glock_nq(fl_gh);
698 if (error) {
699 gfs2_holder_uninit(fl_gh);
700 if (error == GLR_TRYFAILED)
701 error = -EAGAIN;
702 } else {
703 error = flock_lock_file_wait(file, fl);
feaa7bba 704 gfs2_assert_warn(GFS2_SB(&ip->i_inode), !error);
b3b94faa
DT
705 }
706
420b9e5e 707out:
f55ab26a 708 mutex_unlock(&fp->f_fl_mutex);
b3b94faa
DT
709 return error;
710}
711
712static void do_unflock(struct file *file, struct file_lock *fl)
713{
5c676f6d 714 struct gfs2_file *fp = file->private_data;
b3b94faa
DT
715 struct gfs2_holder *fl_gh = &fp->f_fl_gh;
716
f55ab26a 717 mutex_lock(&fp->f_fl_mutex);
b3b94faa
DT
718 flock_lock_file_wait(file, fl);
719 if (fl_gh->gh_gl)
720 gfs2_glock_dq_uninit(fl_gh);
f55ab26a 721 mutex_unlock(&fp->f_fl_mutex);
b3b94faa
DT
722}
723
724/**
725 * gfs2_flock - acquire/release a flock lock on a file
726 * @file: the file pointer
727 * @cmd: either modify or retrieve lock state, possibly wait
728 * @fl: type and range of lock
729 *
730 * Returns: errno
731 */
732
733static int gfs2_flock(struct file *file, int cmd, struct file_lock *fl)
734{
b3b94faa
DT
735 if (!(fl->fl_flags & FL_FLOCK))
736 return -ENOLCK;
a12af1eb
AD
737 if (fl->fl_type & LOCK_MAND)
738 return -EOPNOTSUPP;
b3b94faa 739
b3b94faa
DT
740 if (fl->fl_type == F_UNLCK) {
741 do_unflock(file, fl);
742 return 0;
d00223f1 743 } else {
b3b94faa 744 return do_flock(file, cmd, fl);
d00223f1 745 }
b3b94faa
DT
746}
747
10d21988 748const struct file_operations gfs2_file_fops = {
26c1a574 749 .llseek = gfs2_llseek,
d00223f1 750 .read = do_sync_read,
26c1a574 751 .aio_read = generic_file_aio_read,
d00223f1 752 .write = do_sync_write,
56aa616a 753 .aio_write = gfs2_file_aio_write,
26c1a574
SW
754 .unlocked_ioctl = gfs2_ioctl,
755 .mmap = gfs2_mmap,
756 .open = gfs2_open,
757 .release = gfs2_close,
758 .fsync = gfs2_fsync,
759 .lock = gfs2_lock,
26c1a574
SW
760 .flock = gfs2_flock,
761 .splice_read = generic_file_splice_read,
762 .splice_write = generic_file_splice_write,
60446067 763 .setlease = gfs2_setlease,
b3b94faa
DT
764};
765
10d21988 766const struct file_operations gfs2_dir_fops = {
26c1a574
SW
767 .readdir = gfs2_readdir,
768 .unlocked_ioctl = gfs2_ioctl,
769 .open = gfs2_open,
770 .release = gfs2_close,
771 .fsync = gfs2_fsync,
772 .lock = gfs2_lock,
773 .flock = gfs2_flock,
b3b94faa
DT
774};
775
f057f6cd
SW
776#endif /* CONFIG_GFS2_FS_LOCKING_DLM */
777
10d21988 778const struct file_operations gfs2_file_fops_nolock = {
c97bfe43
WC
779 .llseek = gfs2_llseek,
780 .read = do_sync_read,
781 .aio_read = generic_file_aio_read,
782 .write = do_sync_write,
56aa616a 783 .aio_write = gfs2_file_aio_write,
c97bfe43
WC
784 .unlocked_ioctl = gfs2_ioctl,
785 .mmap = gfs2_mmap,
786 .open = gfs2_open,
787 .release = gfs2_close,
788 .fsync = gfs2_fsync,
789 .splice_read = generic_file_splice_read,
790 .splice_write = generic_file_splice_write,
f057f6cd 791 .setlease = generic_setlease,
c97bfe43
WC
792};
793
10d21988 794const struct file_operations gfs2_dir_fops_nolock = {
c97bfe43
WC
795 .readdir = gfs2_readdir,
796 .unlocked_ioctl = gfs2_ioctl,
797 .open = gfs2_open,
798 .release = gfs2_close,
799 .fsync = gfs2_fsync,
800};
801