?  search:
summary | log | graphiclog1 | graphiclog2 | commit | commitdiff | tree | refs | edit | fork
blame | history | raw | HEAD
Linux 3.8-rc7
[cris-mirror.git] / fs / gfs2 / file.c
blob991ab2d484dd888e9432dd66ecee92247e16fac9
1 /*
2 * Copyright (C) Sistina Software, Inc. 1997-2003 All rights reserved.
3 * Copyright (C) 2004-2006 Red Hat, Inc. All rights reserved.
4 *
5 * This copyrighted material is made available to anyone wishing to use,
6 * modify, copy, or redistribute it subject to the terms and conditions
7 * of the GNU General Public License version 2.
8 */
9
10 #include <linux/slab.h>
11 #include <linux/spinlock.h>
12 #include <linux/completion.h>
13 #include <linux/buffer_head.h>
14 #include <linux/pagemap.h>
15 #include <linux/uio.h>
16 #include <linux/blkdev.h>
17 #include <linux/mm.h>
18 #include <linux/mount.h>
19 #include <linux/fs.h>
20 #include <linux/gfs2_ondisk.h>
21 #include <linux/falloc.h>
22 #include <linux/swap.h>
23 #include <linux/crc32.h>
24 #include <linux/writeback.h>
25 #include <asm/uaccess.h>
26 #include <linux/dlm.h>
27 #include <linux/dlm_plock.h>
28
29 #include "gfs2.h"
30 #include "incore.h"
31 #include "bmap.h"
32 #include "dir.h"
33 #include "glock.h"
34 #include "glops.h"
35 #include "inode.h"
36 #include "log.h"
37 #include "meta_io.h"
38 #include "quota.h"
39 #include "rgrp.h"
40 #include "trans.h"
41 #include "util.h"
42
43 /**
44 * gfs2_llseek - seek to a location in a file
45 * @file: the file
46 * @offset: the offset
47 * @whence: Where to seek from (SEEK_SET, SEEK_CUR, or SEEK_END)
48 *
49 * SEEK_END requires the glock for the file because it references the
50 * file's size.
51 *
52 * Returns: The new offset, or errno
53 */
54
55 static loff_t gfs2_llseek(struct file *file, loff_t offset, int whence)
56 {
57 struct gfs2_inode *ip = GFS2_I(file->f_mapping->host);
58 struct gfs2_holder i_gh;
59 loff_t error;
60
61 switch (whence) {
62 case SEEK_END: /* These reference inode->i_size */
63 case SEEK_DATA:
64 case SEEK_HOLE:
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 case SEEK_CUR:
73 case SEEK_SET:
74 error = generic_file_llseek(file, offset, whence);
75 break;
76 default:
77 error = -EINVAL;
78 }
79
80 return error;
81 }
82
83 /**
84 * gfs2_readdir - Read directory entries from a directory
85 * @file: The directory to read from
86 * @dirent: Buffer for dirents
87 * @filldir: Function used to do the copying
88 *
89 * Returns: errno
90 */
91
92 static int gfs2_readdir(struct file *file, void *dirent, filldir_t filldir)
93 {
94 struct inode *dir = file->f_mapping->host;
95 struct gfs2_inode *dip = GFS2_I(dir);
96 struct gfs2_holder d_gh;
97 u64 offset = file->f_pos;
98 int error;
99
100 gfs2_holder_init(dip->i_gl, LM_ST_SHARED, 0, &d_gh);
101 error = gfs2_glock_nq(&d_gh);
102 if (error) {
103 gfs2_holder_uninit(&d_gh);
104 return error;
105 }
106
107 error = gfs2_dir_read(dir, &offset, dirent, filldir, &file->f_ra);
108
109 gfs2_glock_dq_uninit(&d_gh);
110
111 file->f_pos = offset;
112
113 return error;
114 }
115
116 /**
117 * fsflags_cvt
118 * @table: A table of 32 u32 flags
119 * @val: a 32 bit value to convert
120 *
121 * This function can be used to convert between fsflags values and
122 * GFS2's own flags values.
123 *
124 * Returns: the converted flags
125 */
126 static u32 fsflags_cvt(const u32 *table, u32 val)
127 {
128 u32 res = 0;
129 while(val) {
130 if (val & 1)
131 res |= *table;
132 table++;
133 val >>= 1;
134 }
135 return res;
136 }
137
138 static const u32 fsflags_to_gfs2[32] = {
139 [3] = GFS2_DIF_SYNC,
140 [4] = GFS2_DIF_IMMUTABLE,
141 [5] = GFS2_DIF_APPENDONLY,
142 [7] = GFS2_DIF_NOATIME,
143 [12] = GFS2_DIF_EXHASH,
144 [14] = GFS2_DIF_INHERIT_JDATA,
145 [17] = GFS2_DIF_TOPDIR,
146 };
147
148 static const u32 gfs2_to_fsflags[32] = {
149 [gfs2fl_Sync] = FS_SYNC_FL,
150 [gfs2fl_Immutable] = FS_IMMUTABLE_FL,
151 [gfs2fl_AppendOnly] = FS_APPEND_FL,
152 [gfs2fl_NoAtime] = FS_NOATIME_FL,
153 [gfs2fl_ExHash] = FS_INDEX_FL,
154 [gfs2fl_TopLevel] = FS_TOPDIR_FL,
155 [gfs2fl_InheritJdata] = FS_JOURNAL_DATA_FL,
156 };
157
158 static int gfs2_get_flags(struct file *filp, u32 __user *ptr)
159 {
160 struct inode *inode = filp->f_path.dentry->d_inode;
161 struct gfs2_inode *ip = GFS2_I(inode);
162 struct gfs2_holder gh;
163 int error;
164 u32 fsflags;
165
166 gfs2_holder_init(ip->i_gl, LM_ST_SHARED, 0, &gh);
167 error = gfs2_glock_nq(&gh);
168 if (error)
169 return error;
170
171 fsflags = fsflags_cvt(gfs2_to_fsflags, ip->i_diskflags);
172 if (!S_ISDIR(inode->i_mode) && ip->i_diskflags & GFS2_DIF_JDATA)
173 fsflags |= FS_JOURNAL_DATA_FL;
174 if (put_user(fsflags, ptr))
175 error = -EFAULT;
176
177 gfs2_glock_dq(&gh);
178 gfs2_holder_uninit(&gh);
179 return error;
180 }
181
182 void gfs2_set_inode_flags(struct inode *inode)
183 {
184 struct gfs2_inode *ip = GFS2_I(inode);
185 unsigned int flags = inode->i_flags;
186
187 flags &= ~(S_SYNC|S_APPEND|S_IMMUTABLE|S_NOATIME|S_DIRSYNC|S_NOSEC);
188 if ((ip->i_eattr == 0) && !is_sxid(inode->i_mode))
189 inode->i_flags |= S_NOSEC;
190 if (ip->i_diskflags & GFS2_DIF_IMMUTABLE)
191 flags |= S_IMMUTABLE;
192 if (ip->i_diskflags & GFS2_DIF_APPENDONLY)
193 flags |= S_APPEND;
194 if (ip->i_diskflags & GFS2_DIF_NOATIME)
195 flags |= S_NOATIME;
196 if (ip->i_diskflags & GFS2_DIF_SYNC)
197 flags |= S_SYNC;
198 inode->i_flags = flags;
199 }
200
201 /* Flags that can be set by user space */
202 #define GFS2_FLAGS_USER_SET (GFS2_DIF_JDATA| \
203 GFS2_DIF_IMMUTABLE| \
204 GFS2_DIF_APPENDONLY| \
205 GFS2_DIF_NOATIME| \
206 GFS2_DIF_SYNC| \
207 GFS2_DIF_SYSTEM| \
208 GFS2_DIF_TOPDIR| \
209 GFS2_DIF_INHERIT_JDATA)
210
211 /**
212 * gfs2_set_flags - set flags on an inode
213 * @inode: The inode
214 * @flags: The flags to set
215 * @mask: Indicates which flags are valid
216 *
217 */
218 static int do_gfs2_set_flags(struct file *filp, u32 reqflags, u32 mask)
219 {
220 struct inode *inode = filp->f_path.dentry->d_inode;
221 struct gfs2_inode *ip = GFS2_I(inode);
222 struct gfs2_sbd *sdp = GFS2_SB(inode);
223 struct buffer_head *bh;
224 struct gfs2_holder gh;
225 int error;
226 u32 new_flags, flags;
227
228 error = mnt_want_write_file(filp);
229 if (error)
230 return error;
231
232 error = gfs2_glock_nq_init(ip->i_gl, LM_ST_EXCLUSIVE, 0, &gh);
233 if (error)
234 goto out_drop_write;
235
236 error = -EACCES;
237 if (!inode_owner_or_capable(inode))
238 goto out;
239
240 error = 0;
241 flags = ip->i_diskflags;
242 new_flags = (flags & ~mask) | (reqflags & mask);
243 if ((new_flags ^ flags) == 0)
244 goto out;
245
246 error = -EINVAL;
247 if ((new_flags ^ flags) & ~GFS2_FLAGS_USER_SET)
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 (flags & GFS2_DIF_JDATA)
265 gfs2_log_flush(sdp, ip->i_gl);
266 error = filemap_fdatawrite(inode->i_mapping);
267 if (error)
268 goto out;
269 error = filemap_fdatawait(inode->i_mapping);
270 if (error)
271 goto out;
272 }
273 error = gfs2_trans_begin(sdp, RES_DINODE, 0);
274 if (error)
275 goto out;
276 error = gfs2_meta_inode_buffer(ip, &bh);
277 if (error)
278 goto out_trans_end;
279 gfs2_trans_add_bh(ip->i_gl, bh, 1);
280 ip->i_diskflags = new_flags;
281 gfs2_dinode_out(ip, bh->b_data);
282 brelse(bh);
283 gfs2_set_inode_flags(inode);
284 gfs2_set_aops(inode);
285 out_trans_end:
286 gfs2_trans_end(sdp);
287 out:
288 gfs2_glock_dq_uninit(&gh);
289 out_drop_write:
290 mnt_drop_write_file(filp);
291 return error;
292 }
293
294 static int gfs2_set_flags(struct file *filp, u32 __user *ptr)
295 {
296 struct inode *inode = filp->f_path.dentry->d_inode;
297 u32 fsflags, gfsflags;
298
299 if (get_user(fsflags, ptr))
300 return -EFAULT;
301
302 gfsflags = fsflags_cvt(fsflags_to_gfs2, fsflags);
303 if (!S_ISDIR(inode->i_mode)) {
304 gfsflags &= ~GFS2_DIF_TOPDIR;
305 if (gfsflags & GFS2_DIF_INHERIT_JDATA)
306 gfsflags ^= (GFS2_DIF_JDATA | GFS2_DIF_INHERIT_JDATA);
307 return do_gfs2_set_flags(filp, gfsflags, ~0);
308 }
309 return do_gfs2_set_flags(filp, gfsflags, ~GFS2_DIF_JDATA);
310 }
311
312 static long gfs2_ioctl(struct file *filp, unsigned int cmd, unsigned long arg)
313 {
314 switch(cmd) {
315 case FS_IOC_GETFLAGS:
316 return gfs2_get_flags(filp, (u32 __user *)arg);
317 case FS_IOC_SETFLAGS:
318 return gfs2_set_flags(filp, (u32 __user *)arg);
319 case FITRIM:
320 return gfs2_fitrim(filp, (void __user *)arg);
321 }
322 return -ENOTTY;
323 }
324
325 /**
326 * gfs2_size_hint - Give a hint to the size of a write request
327 * @file: The struct file
328 * @offset: The file offset of the write
329 * @size: The length of the write
330 *
331 * When we are about to do a write, this function records the total
332 * write size in order to provide a suitable hint to the lower layers
333 * about how many blocks will be required.
334 *
335 */
336
337 static void gfs2_size_hint(struct file *filep, loff_t offset, size_t size)
338 {
339 struct inode *inode = filep->f_dentry->d_inode;
340 struct gfs2_sbd *sdp = GFS2_SB(inode);
341 struct gfs2_inode *ip = GFS2_I(inode);
342 size_t blks = (size + sdp->sd_sb.sb_bsize - 1) >> sdp->sd_sb.sb_bsize_shift;
343 int hint = min_t(size_t, INT_MAX, blks);
344
345 atomic_set(&ip->i_res->rs_sizehint, hint);
346 }
347
348 /**
349 * gfs2_allocate_page_backing - Use bmap to allocate blocks
350 * @page: The (locked) page to allocate backing for
351 *
352 * We try to allocate all the blocks required for the page in
353 * one go. This might fail for various reasons, so we keep
354 * trying until all the blocks to back this page are allocated.
355 * If some of the blocks are already allocated, thats ok too.
356 */
357
358 static int gfs2_allocate_page_backing(struct page *page)
359 {
360 struct inode *inode = page->mapping->host;
361 struct buffer_head bh;
362 unsigned long size = PAGE_CACHE_SIZE;
363 u64 lblock = page->index << (PAGE_CACHE_SHIFT - inode->i_blkbits);
364
365 do {
366 bh.b_state = 0;
367 bh.b_size = size;
368 gfs2_block_map(inode, lblock, &bh, 1);
369 if (!buffer_mapped(&bh))
370 return -EIO;
371 size -= bh.b_size;
372 lblock += (bh.b_size >> inode->i_blkbits);
373 } while(size > 0);
374 return 0;
375 }
376
377 /**
378 * gfs2_page_mkwrite - Make a shared, mmap()ed, page writable
379 * @vma: The virtual memory area
380 * @page: The page which is about to become writable
381 *
382 * When the page becomes writable, we need to ensure that we have
383 * blocks allocated on disk to back that page.
384 */
385
386 static int gfs2_page_mkwrite(struct vm_area_struct *vma, struct vm_fault *vmf)
387 {
388 struct page *page = vmf->page;
389 struct inode *inode = vma->vm_file->f_path.dentry->d_inode;
390 struct gfs2_inode *ip = GFS2_I(inode);
391 struct gfs2_sbd *sdp = GFS2_SB(inode);
392 unsigned long last_index;
393 u64 pos = page->index << PAGE_CACHE_SHIFT;
394 unsigned int data_blocks, ind_blocks, rblocks;
395 struct gfs2_holder gh;
396 loff_t size;
397 int ret;
398
399 sb_start_pagefault(inode->i_sb);
400
401 /* Update file times before taking page lock */
402 file_update_time(vma->vm_file);
403
404 ret = gfs2_rs_alloc(ip);
405 if (ret)
406 return ret;
407
408 gfs2_size_hint(vma->vm_file, pos, PAGE_CACHE_SIZE);
409
410 gfs2_holder_init(ip->i_gl, LM_ST_EXCLUSIVE, 0, &gh);
411 ret = gfs2_glock_nq(&gh);
412 if (ret)
413 goto out;
414
415 set_bit(GLF_DIRTY, &ip->i_gl->gl_flags);
416 set_bit(GIF_SW_PAGED, &ip->i_flags);
417
418 if (!gfs2_write_alloc_required(ip, pos, PAGE_CACHE_SIZE)) {
419 lock_page(page);
420 if (!PageUptodate(page) || page->mapping != inode->i_mapping) {
421 ret = -EAGAIN;
422 unlock_page(page);
423 }
424 goto out_unlock;
425 }
426
427 ret = gfs2_rindex_update(sdp);
428 if (ret)
429 goto out_unlock;
430
431 ret = gfs2_quota_lock_check(ip);
432 if (ret)
433 goto out_unlock;
434 gfs2_write_calc_reserv(ip, PAGE_CACHE_SIZE, &data_blocks, &ind_blocks);
435 ret = gfs2_inplace_reserve(ip, data_blocks + ind_blocks, 0);
436 if (ret)
437 goto out_quota_unlock;
438
439 rblocks = RES_DINODE + ind_blocks;
440 if (gfs2_is_jdata(ip))
441 rblocks += data_blocks ? data_blocks : 1;
442 if (ind_blocks || data_blocks) {
443 rblocks += RES_STATFS + RES_QUOTA;
444 rblocks += gfs2_rg_blocks(ip, data_blocks + ind_blocks);
445 }
446 ret = gfs2_trans_begin(sdp, rblocks, 0);
447 if (ret)
448 goto out_trans_fail;
449
450 lock_page(page);
451 ret = -EINVAL;
452 size = i_size_read(inode);
453 last_index = (size - 1) >> PAGE_CACHE_SHIFT;
454 /* Check page index against inode size */
455 if (size == 0 || (page->index > last_index))
456 goto out_trans_end;
457
458 ret = -EAGAIN;
459 /* If truncated, we must retry the operation, we may have raced
460 * with the glock demotion code.
461 */
462 if (!PageUptodate(page) || page->mapping != inode->i_mapping)
463 goto out_trans_end;
464
465 /* Unstuff, if required, and allocate backing blocks for page */
466 ret = 0;
467 if (gfs2_is_stuffed(ip))
468 ret = gfs2_unstuff_dinode(ip, page);
469 if (ret == 0)
470 ret = gfs2_allocate_page_backing(page);
471
472 out_trans_end:
473 if (ret)
474 unlock_page(page);
475 gfs2_trans_end(sdp);
476 out_trans_fail:
477 gfs2_inplace_release(ip);
478 out_quota_unlock:
479 gfs2_quota_unlock(ip);
480 out_unlock:
481 gfs2_glock_dq(&gh);
482 out:
483 gfs2_holder_uninit(&gh);
484 if (ret == 0) {
485 set_page_dirty(page);
486 wait_on_page_writeback(page);
487 }
488 sb_end_pagefault(inode->i_sb);
489 return block_page_mkwrite_return(ret);
490 }
491
492 static const struct vm_operations_struct gfs2_vm_ops = {
493 .fault = filemap_fault,
494 .page_mkwrite = gfs2_page_mkwrite,
495 .remap_pages = generic_file_remap_pages,
496 };
497
498 /**
499 * gfs2_mmap -
500 * @file: The file to map
501 * @vma: The VMA which described the mapping
502 *
503 * There is no need to get a lock here unless we should be updating
504 * atime. We ignore any locking errors since the only consequence is
505 * a missed atime update (which will just be deferred until later).
506 *
507 * Returns: 0
508 */
509
510 static int gfs2_mmap(struct file *file, struct vm_area_struct *vma)
511 {
512 struct gfs2_inode *ip = GFS2_I(file->f_mapping->host);
513
514 if (!(file->f_flags & O_NOATIME) &&
515 !IS_NOATIME(&ip->i_inode)) {
516 struct gfs2_holder i_gh;
517 int error;
518
519 error = gfs2_glock_nq_init(ip->i_gl, LM_ST_SHARED, LM_FLAG_ANY,
520 &i_gh);
521 if (error)
522 return error;
523 /* grab lock to update inode */
524 gfs2_glock_dq_uninit(&i_gh);
525 file_accessed(file);
526 }
527 vma->vm_ops = &gfs2_vm_ops;
528
529 return 0;
530 }
531
532 /**
533 * gfs2_open - open a file
534 * @inode: the inode to open
535 * @file: the struct file for this opening
536 *
537 * Returns: errno
538 */
539
540 static int gfs2_open(struct inode *inode, struct file *file)
541 {
542 struct gfs2_inode *ip = GFS2_I(inode);
543 struct gfs2_holder i_gh;
544 struct gfs2_file *fp;
545 int error;
546
547 fp = kzalloc(sizeof(struct gfs2_file), GFP_KERNEL);
548 if (!fp)
549 return -ENOMEM;
550
551 mutex_init(&fp->f_fl_mutex);
552
553 gfs2_assert_warn(GFS2_SB(inode), !file->private_data);
554 file->private_data = fp;
555
556 if (S_ISREG(ip->i_inode.i_mode)) {
557 error = gfs2_glock_nq_init(ip->i_gl, LM_ST_SHARED, LM_FLAG_ANY,
558 &i_gh);
559 if (error)
560 goto fail;
561
562 if (!(file->f_flags & O_LARGEFILE) &&
563 i_size_read(inode) > MAX_NON_LFS) {
564 error = -EOVERFLOW;
565 goto fail_gunlock;
566 }
567
568 gfs2_glock_dq_uninit(&i_gh);
569 }
570
571 return 0;
572
573 fail_gunlock:
574 gfs2_glock_dq_uninit(&i_gh);
575 fail:
576 file->private_data = NULL;
577 kfree(fp);
578 return error;
579 }
580
581 /**
582 * gfs2_release - called to close a struct file
583 * @inode: the inode the struct file belongs to
584 * @file: the struct file being closed
585 *
586 * Returns: errno
587 */
588
589 static int gfs2_release(struct inode *inode, struct file *file)
590 {
591 struct gfs2_inode *ip = GFS2_I(inode);
592
593 kfree(file->private_data);
594 file->private_data = NULL;
595
596 if ((file->f_mode & FMODE_WRITE) &&
597 (atomic_read(&inode->i_writecount) == 1))
598 gfs2_rs_delete(ip);
599
600 return 0;
601 }
602
603 /**
604 * gfs2_fsync - sync the dirty data for a file (across the cluster)
605 * @file: the file that points to the dentry
606 * @start: the start position in the file to sync
607 * @end: the end position in the file to sync
608 * @datasync: set if we can ignore timestamp changes
609 *
610 * We split the data flushing here so that we don't wait for the data
611 * until after we've also sent the metadata to disk. Note that for
612 * data=ordered, we will write & wait for the data at the log flush
613 * stage anyway, so this is unlikely to make much of a difference
614 * except in the data=writeback case.
615 *
616 * If the fdatawrite fails due to any reason except -EIO, we will
617 * continue the remainder of the fsync, although we'll still report
618 * the error at the end. This is to match filemap_write_and_wait_range()
619 * behaviour.
620 *
621 * Returns: errno
622 */
623
624 static int gfs2_fsync(struct file *file, loff_t start, loff_t end,
625 int datasync)
626 {
627 struct address_space *mapping = file->f_mapping;
628 struct inode *inode = mapping->host;
629 int sync_state = inode->i_state & (I_DIRTY_SYNC|I_DIRTY_DATASYNC);
630 struct gfs2_inode *ip = GFS2_I(inode);
631 int ret = 0, ret1 = 0;
632
633 if (mapping->nrpages) {
634 ret1 = filemap_fdatawrite_range(mapping, start, end);
635 if (ret1 == -EIO)
636 return ret1;
637 }
638
639 if (datasync)
640 sync_state &= ~I_DIRTY_SYNC;
641
642 if (sync_state) {
643 ret = sync_inode_metadata(inode, 1);
644 if (ret)
645 return ret;
646 if (gfs2_is_jdata(ip))
647 filemap_write_and_wait(mapping);
648 gfs2_ail_flush(ip->i_gl, 1);
649 }
650
651 if (mapping->nrpages)
652 ret = filemap_fdatawait_range(mapping, start, end);
653
654 return ret ? ret : ret1;
655 }
656
657 /**
658 * gfs2_file_aio_write - Perform a write to a file
659 * @iocb: The io context
660 * @iov: The data to write
661 * @nr_segs: Number of @iov segments
662 * @pos: The file position
663 *
664 * We have to do a lock/unlock here to refresh the inode size for
665 * O_APPEND writes, otherwise we can land up writing at the wrong
666 * offset. There is still a race, but provided the app is using its
667 * own file locking, this will make O_APPEND work as expected.
668 *
669 */
670
671 static ssize_t gfs2_file_aio_write(struct kiocb *iocb, const struct iovec *iov,
672 unsigned long nr_segs, loff_t pos)
673 {
674 struct file *file = iocb->ki_filp;
675 size_t writesize = iov_length(iov, nr_segs);
676 struct dentry *dentry = file->f_dentry;
677 struct gfs2_inode *ip = GFS2_I(dentry->d_inode);
678 int ret;
679
680 ret = gfs2_rs_alloc(ip);
681 if (ret)
682 return ret;
683
684 gfs2_size_hint(file, pos, writesize);
685
686 if (file->f_flags & O_APPEND) {
687 struct gfs2_holder gh;
688
689 ret = gfs2_glock_nq_init(ip->i_gl, LM_ST_SHARED, 0, &gh);
690 if (ret)
691 return ret;
692 gfs2_glock_dq_uninit(&gh);
693 }
694
695 return generic_file_aio_write(iocb, iov, nr_segs, pos);
696 }
697
698 static int fallocate_chunk(struct inode *inode, loff_t offset, loff_t len,
699 int mode)
700 {
701 struct gfs2_inode *ip = GFS2_I(inode);
702 struct buffer_head *dibh;
703 int error;
704 loff_t size = len;
705 unsigned int nr_blks;
706 sector_t lblock = offset >> inode->i_blkbits;
707
708 error = gfs2_meta_inode_buffer(ip, &dibh);
709 if (unlikely(error))
710 return error;
711
712 gfs2_trans_add_bh(ip->i_gl, dibh, 1);
713
714 if (gfs2_is_stuffed(ip)) {
715 error = gfs2_unstuff_dinode(ip, NULL);
716 if (unlikely(error))
717 goto out;
718 }
719
720 while (len) {
721 struct buffer_head bh_map = { .b_state = 0, .b_blocknr = 0 };
722 bh_map.b_size = len;
723 set_buffer_zeronew(&bh_map);
724
725 error = gfs2_block_map(inode, lblock, &bh_map, 1);
726 if (unlikely(error))
727 goto out;
728 len -= bh_map.b_size;
729 nr_blks = bh_map.b_size >> inode->i_blkbits;
730 lblock += nr_blks;
731 if (!buffer_new(&bh_map))
732 continue;
733 if (unlikely(!buffer_zeronew(&bh_map))) {
734 error = -EIO;
735 goto out;
736 }
737 }
738 if (offset + size > inode->i_size && !(mode & FALLOC_FL_KEEP_SIZE))
739 i_size_write(inode, offset + size);
740
741 mark_inode_dirty(inode);
742
743 out:
744 brelse(dibh);
745 return error;
746 }
747
748 static void calc_max_reserv(struct gfs2_inode *ip, loff_t max, loff_t *len,
749 unsigned int *data_blocks, unsigned int *ind_blocks)
750 {
751 const struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode);
752 unsigned int max_blocks = ip->i_rgd->rd_free_clone;
753 unsigned int tmp, max_data = max_blocks - 3 * (sdp->sd_max_height - 1);
754
755 for (tmp = max_data; tmp > sdp->sd_diptrs;) {
756 tmp = DIV_ROUND_UP(tmp, sdp->sd_inptrs);
757 max_data -= tmp;
758 }
759 /* This calculation isn't the exact reverse of gfs2_write_calc_reserve,
760 so it might end up with fewer data blocks */
761 if (max_data <= *data_blocks)
762 return;
763 *data_blocks = max_data;
764 *ind_blocks = max_blocks - max_data;
765 *len = ((loff_t)max_data - 3) << sdp->sd_sb.sb_bsize_shift;
766 if (*len > max) {
767 *len = max;
768 gfs2_write_calc_reserv(ip, max, data_blocks, ind_blocks);
769 }
770 }
771
772 static long gfs2_fallocate(struct file *file, int mode, loff_t offset,
773 loff_t len)
774 {
775 struct inode *inode = file->f_path.dentry->d_inode;
776 struct gfs2_sbd *sdp = GFS2_SB(inode);
777 struct gfs2_inode *ip = GFS2_I(inode);
778 unsigned int data_blocks = 0, ind_blocks = 0, rblocks;
779 loff_t bytes, max_bytes;
780 int error;
781 const loff_t pos = offset;
782 const loff_t count = len;
783 loff_t bsize_mask = ~((loff_t)sdp->sd_sb.sb_bsize - 1);
784 loff_t next = (offset + len - 1) >> sdp->sd_sb.sb_bsize_shift;
785 loff_t max_chunk_size = UINT_MAX & bsize_mask;
786 next = (next + 1) << sdp->sd_sb.sb_bsize_shift;
787
788 /* We only support the FALLOC_FL_KEEP_SIZE mode */
789 if (mode & ~FALLOC_FL_KEEP_SIZE)
790 return -EOPNOTSUPP;
791
792 offset &= bsize_mask;
793
794 len = next - offset;
795 bytes = sdp->sd_max_rg_data * sdp->sd_sb.sb_bsize / 2;
796 if (!bytes)
797 bytes = UINT_MAX;
798 bytes &= bsize_mask;
799 if (bytes == 0)
800 bytes = sdp->sd_sb.sb_bsize;
801
802 error = gfs2_rs_alloc(ip);
803 if (error)
804 return error;
805
806 gfs2_holder_init(ip->i_gl, LM_ST_EXCLUSIVE, 0, &ip->i_gh);
807 error = gfs2_glock_nq(&ip->i_gh);
808 if (unlikely(error))
809 goto out_uninit;
810
811 gfs2_size_hint(file, offset, len);
812
813 while (len > 0) {
814 if (len < bytes)
815 bytes = len;
816 if (!gfs2_write_alloc_required(ip, offset, bytes)) {
817 len -= bytes;
818 offset += bytes;
819 continue;
820 }
821 error = gfs2_quota_lock_check(ip);
822 if (error)
823 goto out_unlock;
824
825 retry:
826 gfs2_write_calc_reserv(ip, bytes, &data_blocks, &ind_blocks);
827
828 error = gfs2_inplace_reserve(ip, data_blocks + ind_blocks, 0);
829 if (error) {
830 if (error == -ENOSPC && bytes > sdp->sd_sb.sb_bsize) {
831 bytes >>= 1;
832 bytes &= bsize_mask;
833 if (bytes == 0)
834 bytes = sdp->sd_sb.sb_bsize;
835 goto retry;
836 }
837 goto out_qunlock;
838 }
839 max_bytes = bytes;
840 calc_max_reserv(ip, (len > max_chunk_size)? max_chunk_size: len,
841 &max_bytes, &data_blocks, &ind_blocks);
842
843 rblocks = RES_DINODE + ind_blocks + RES_STATFS + RES_QUOTA +
844 RES_RG_HDR + gfs2_rg_blocks(ip, data_blocks + ind_blocks);
845 if (gfs2_is_jdata(ip))
846 rblocks += data_blocks ? data_blocks : 1;
847
848 error = gfs2_trans_begin(sdp, rblocks,
849 PAGE_CACHE_SIZE/sdp->sd_sb.sb_bsize);
850 if (error)
851 goto out_trans_fail;
852
853 error = fallocate_chunk(inode, offset, max_bytes, mode);
854 gfs2_trans_end(sdp);
855
856 if (error)
857 goto out_trans_fail;
858
859 len -= max_bytes;
860 offset += max_bytes;
861 gfs2_inplace_release(ip);
862 gfs2_quota_unlock(ip);
863 }
864
865 if (error == 0)
866 error = generic_write_sync(file, pos, count);
867 goto out_unlock;
868
869 out_trans_fail:
870 gfs2_inplace_release(ip);
871 out_qunlock:
872 gfs2_quota_unlock(ip);
873 out_unlock:
874 gfs2_glock_dq(&ip->i_gh);
875 out_uninit:
876 gfs2_holder_uninit(&ip->i_gh);
877 return error;
878 }
879
880 #ifdef CONFIG_GFS2_FS_LOCKING_DLM
881
882 /**
883 * gfs2_setlease - acquire/release a file lease
884 * @file: the file pointer
885 * @arg: lease type
886 * @fl: file lock
887 *
888 * We don't currently have a way to enforce a lease across the whole
889 * cluster; until we do, disable leases (by just returning -EINVAL),
890 * unless the administrator has requested purely local locking.
891 *
892 * Locking: called under lock_flocks
893 *
894 * Returns: errno
895 */
896
897 static int gfs2_setlease(struct file *file, long arg, struct file_lock **fl)
898 {
899 return -EINVAL;
900 }
901
902 /**
903 * gfs2_lock - acquire/release a posix lock on a file
904 * @file: the file pointer
905 * @cmd: either modify or retrieve lock state, possibly wait
906 * @fl: type and range of lock
907 *
908 * Returns: errno
909 */
910
911 static int gfs2_lock(struct file *file, int cmd, struct file_lock *fl)
912 {
913 struct gfs2_inode *ip = GFS2_I(file->f_mapping->host);
914 struct gfs2_sbd *sdp = GFS2_SB(file->f_mapping->host);
915 struct lm_lockstruct *ls = &sdp->sd_lockstruct;
916
917 if (!(fl->fl_flags & FL_POSIX))
918 return -ENOLCK;
919 if (__mandatory_lock(&ip->i_inode) && fl->fl_type != F_UNLCK)
920 return -ENOLCK;
921
922 if (cmd == F_CANCELLK) {
923 /* Hack: */
924 cmd = F_SETLK;
925 fl->fl_type = F_UNLCK;
926 }
927 if (unlikely(test_bit(SDF_SHUTDOWN, &sdp->sd_flags)))
928 return -EIO;
929 if (IS_GETLK(cmd))
930 return dlm_posix_get(ls->ls_dlm, ip->i_no_addr, file, fl);
931 else if (fl->fl_type == F_UNLCK)
932 return dlm_posix_unlock(ls->ls_dlm, ip->i_no_addr, file, fl);
933 else
934 return dlm_posix_lock(ls->ls_dlm, ip->i_no_addr, file, cmd, fl);
935 }
936
937 static int do_flock(struct file *file, int cmd, struct file_lock *fl)
938 {
939 struct gfs2_file *fp = file->private_data;
940 struct gfs2_holder *fl_gh = &fp->f_fl_gh;
941 struct gfs2_inode *ip = GFS2_I(file->f_path.dentry->d_inode);
942 struct gfs2_glock *gl;
943 unsigned int state;
944 int flags;
945 int error = 0;
946
947 state = (fl->fl_type == F_WRLCK) ? LM_ST_EXCLUSIVE : LM_ST_SHARED;
948 flags = (IS_SETLKW(cmd) ? 0 : LM_FLAG_TRY) | GL_EXACT | GL_NOCACHE;
949
950 mutex_lock(&fp->f_fl_mutex);
951
952 gl = fl_gh->gh_gl;
953 if (gl) {
954 if (fl_gh->gh_state == state)
955 goto out;
956 flock_lock_file_wait(file,
957 &(struct file_lock){.fl_type = F_UNLCK});
958 gfs2_glock_dq_wait(fl_gh);
959 gfs2_holder_reinit(state, flags, fl_gh);
960 } else {
961 error = gfs2_glock_get(GFS2_SB(&ip->i_inode), ip->i_no_addr,
962 &gfs2_flock_glops, CREATE, &gl);
963 if (error)
964 goto out;
965 gfs2_holder_init(gl, state, flags, fl_gh);
966 gfs2_glock_put(gl);
967 }
968 error = gfs2_glock_nq(fl_gh);
969 if (error) {
970 gfs2_holder_uninit(fl_gh);
971 if (error == GLR_TRYFAILED)
972 error = -EAGAIN;
973 } else {
974 error = flock_lock_file_wait(file, fl);
975 gfs2_assert_warn(GFS2_SB(&ip->i_inode), !error);
976 }
977
978 out:
979 mutex_unlock(&fp->f_fl_mutex);
980 return error;
981 }
982
983 static void do_unflock(struct file *file, struct file_lock *fl)
984 {
985 struct gfs2_file *fp = file->private_data;
986 struct gfs2_holder *fl_gh = &fp->f_fl_gh;
987
988 mutex_lock(&fp->f_fl_mutex);
989 flock_lock_file_wait(file, fl);
990 if (fl_gh->gh_gl) {
991 gfs2_glock_dq_wait(fl_gh);
992 gfs2_holder_uninit(fl_gh);
993 }
994 mutex_unlock(&fp->f_fl_mutex);
995 }
996
997 /**
998 * gfs2_flock - acquire/release a flock lock on a file
999 * @file: the file pointer
1000 * @cmd: either modify or retrieve lock state, possibly wait
1001 * @fl: type and range of lock
1002 *
1003 * Returns: errno
1004 */
1005
1006 static int gfs2_flock(struct file *file, int cmd, struct file_lock *fl)
1007 {
1008 if (!(fl->fl_flags & FL_FLOCK))
1009 return -ENOLCK;
1010 if (fl->fl_type & LOCK_MAND)
1011 return -EOPNOTSUPP;
1012
1013 if (fl->fl_type == F_UNLCK) {
1014 do_unflock(file, fl);
1015 return 0;
1016 } else {
1017 return do_flock(file, cmd, fl);
1018 }
1019 }
1020
1021 const struct file_operations gfs2_file_fops = {
1022 .llseek = gfs2_llseek,
1023 .read = do_sync_read,
1024 .aio_read = generic_file_aio_read,
1025 .write = do_sync_write,
1026 .aio_write = gfs2_file_aio_write,
1027 .unlocked_ioctl = gfs2_ioctl,
1028 .mmap = gfs2_mmap,
1029 .open = gfs2_open,
1030 .release = gfs2_release,
1031 .fsync = gfs2_fsync,
1032 .lock = gfs2_lock,
1033 .flock = gfs2_flock,
1034 .splice_read = generic_file_splice_read,
1035 .splice_write = generic_file_splice_write,
1036 .setlease = gfs2_setlease,
1037 .fallocate = gfs2_fallocate,
1038 };
1039
1040 const struct file_operations gfs2_dir_fops = {
1041 .readdir = gfs2_readdir,
1042 .unlocked_ioctl = gfs2_ioctl,
1043 .open = gfs2_open,
1044 .release = gfs2_release,
1045 .fsync = gfs2_fsync,
1046 .lock = gfs2_lock,
1047 .flock = gfs2_flock,
1048 .llseek = default_llseek,
1049 };
1050
1051 #endif /* CONFIG_GFS2_FS_LOCKING_DLM */
1052
1053 const struct file_operations gfs2_file_fops_nolock = {
1054 .llseek = gfs2_llseek,
1055 .read = do_sync_read,
1056 .aio_read = generic_file_aio_read,
1057 .write = do_sync_write,
1058 .aio_write = gfs2_file_aio_write,
1059 .unlocked_ioctl = gfs2_ioctl,
1060 .mmap = gfs2_mmap,
1061 .open = gfs2_open,
1062 .release = gfs2_release,
1063 .fsync = gfs2_fsync,
1064 .splice_read = generic_file_splice_read,
1065 .splice_write = generic_file_splice_write,
1066 .setlease = generic_setlease,
1067 .fallocate = gfs2_fallocate,
1068 };
1069
1070 const struct file_operations gfs2_dir_fops_nolock = {
1071 .readdir = gfs2_readdir,
1072 .unlocked_ioctl = gfs2_ioctl,
1073 .open = gfs2_open,
1074 .release = gfs2_release,
1075 .fsync = gfs2_fsync,
1076 .llseek = default_llseek,
1077 };
1078
CRIS linux kernel tree