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  * @origin: 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 
gfs2_llseek(struct file * file,loff_t offset,int origin)55 static loff_t gfs2_llseek(struct file *file, loff_t offset, int origin)
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 (origin) {
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, origin);
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, origin);
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 
gfs2_readdir(struct file * file,void * dirent,filldir_t filldir)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  */
fsflags_cvt(const u32 * table,u32 val)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 };
146 
147 static const u32 gfs2_to_fsflags[32] = {
148 	[gfs2fl_Sync] = FS_SYNC_FL,
149 	[gfs2fl_Immutable] = FS_IMMUTABLE_FL,
150 	[gfs2fl_AppendOnly] = FS_APPEND_FL,
151 	[gfs2fl_NoAtime] = FS_NOATIME_FL,
152 	[gfs2fl_ExHash] = FS_INDEX_FL,
153 	[gfs2fl_InheritJdata] = FS_JOURNAL_DATA_FL,
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 = filp->f_path.dentry->d_inode;
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 		return error;
168 
169 	fsflags = fsflags_cvt(gfs2_to_fsflags, ip->i_diskflags);
170 	if (!S_ISDIR(inode->i_mode) && ip->i_diskflags & GFS2_DIF_JDATA)
171 		fsflags |= FS_JOURNAL_DATA_FL;
172 	if (put_user(fsflags, ptr))
173 		error = -EFAULT;
174 
175 	gfs2_glock_dq(&gh);
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 		inode->i_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_SYSTEM|			\
206 			     GFS2_DIF_INHERIT_JDATA)
207 
208 /**
209  * gfs2_set_flags - set flags on an inode
210  * @inode: The inode
211  * @flags: The flags to set
212  * @mask: Indicates which flags are valid
213  *
214  */
do_gfs2_set_flags(struct file * filp,u32 reqflags,u32 mask)215 static int do_gfs2_set_flags(struct file *filp, u32 reqflags, u32 mask)
216 {
217 	struct inode *inode = filp->f_path.dentry->d_inode;
218 	struct gfs2_inode *ip = GFS2_I(inode);
219 	struct gfs2_sbd *sdp = GFS2_SB(inode);
220 	struct buffer_head *bh;
221 	struct gfs2_holder gh;
222 	int error;
223 	u32 new_flags, flags;
224 
225 	error = mnt_want_write_file(filp);
226 	if (error)
227 		return error;
228 
229 	error = gfs2_glock_nq_init(ip->i_gl, LM_ST_EXCLUSIVE, 0, &gh);
230 	if (error)
231 		goto out_drop_write;
232 
233 	error = -EACCES;
234 	if (!inode_owner_or_capable(inode))
235 		goto out;
236 
237 	error = 0;
238 	flags = ip->i_diskflags;
239 	new_flags = (flags & ~mask) | (reqflags & mask);
240 	if ((new_flags ^ flags) == 0)
241 		goto out;
242 
243 	error = -EINVAL;
244 	if ((new_flags ^ flags) & ~GFS2_FLAGS_USER_SET)
245 		goto out;
246 
247 	error = -EPERM;
248 	if (IS_IMMUTABLE(inode) && (new_flags & GFS2_DIF_IMMUTABLE))
249 		goto out;
250 	if (IS_APPEND(inode) && (new_flags & GFS2_DIF_APPENDONLY))
251 		goto out;
252 	if (((new_flags ^ flags) & GFS2_DIF_IMMUTABLE) &&
253 	    !capable(CAP_LINUX_IMMUTABLE))
254 		goto out;
255 	if (!IS_IMMUTABLE(inode)) {
256 		error = gfs2_permission(inode, MAY_WRITE);
257 		if (error)
258 			goto out;
259 	}
260 	if ((flags ^ new_flags) & GFS2_DIF_JDATA) {
261 		if (flags & GFS2_DIF_JDATA)
262 			gfs2_log_flush(sdp, ip->i_gl);
263 		error = filemap_fdatawrite(inode->i_mapping);
264 		if (error)
265 			goto out;
266 		error = filemap_fdatawait(inode->i_mapping);
267 		if (error)
268 			goto out;
269 	}
270 	error = gfs2_trans_begin(sdp, RES_DINODE, 0);
271 	if (error)
272 		goto out;
273 	error = gfs2_meta_inode_buffer(ip, &bh);
274 	if (error)
275 		goto out_trans_end;
276 	gfs2_trans_add_bh(ip->i_gl, bh, 1);
277 	ip->i_diskflags = new_flags;
278 	gfs2_dinode_out(ip, bh->b_data);
279 	brelse(bh);
280 	gfs2_set_inode_flags(inode);
281 	gfs2_set_aops(inode);
282 out_trans_end:
283 	gfs2_trans_end(sdp);
284 out:
285 	gfs2_glock_dq_uninit(&gh);
286 out_drop_write:
287 	mnt_drop_write_file(filp);
288 	return error;
289 }
290 
gfs2_set_flags(struct file * filp,u32 __user * ptr)291 static int gfs2_set_flags(struct file *filp, u32 __user *ptr)
292 {
293 	struct inode *inode = filp->f_path.dentry->d_inode;
294 	u32 fsflags, gfsflags;
295 
296 	if (get_user(fsflags, ptr))
297 		return -EFAULT;
298 
299 	gfsflags = fsflags_cvt(fsflags_to_gfs2, fsflags);
300 	if (!S_ISDIR(inode->i_mode)) {
301 		if (gfsflags & GFS2_DIF_INHERIT_JDATA)
302 			gfsflags ^= (GFS2_DIF_JDATA | GFS2_DIF_INHERIT_JDATA);
303 		return do_gfs2_set_flags(filp, gfsflags, ~0);
304 	}
305 	return do_gfs2_set_flags(filp, gfsflags, ~GFS2_DIF_JDATA);
306 }
307 
gfs2_ioctl(struct file * filp,unsigned int cmd,unsigned long arg)308 static long gfs2_ioctl(struct file *filp, unsigned int cmd, unsigned long arg)
309 {
310 	switch(cmd) {
311 	case FS_IOC_GETFLAGS:
312 		return gfs2_get_flags(filp, (u32 __user *)arg);
313 	case FS_IOC_SETFLAGS:
314 		return gfs2_set_flags(filp, (u32 __user *)arg);
315 	case FITRIM:
316 		return gfs2_fitrim(filp, (void __user *)arg);
317 	}
318 	return -ENOTTY;
319 }
320 
321 /**
322  * gfs2_allocate_page_backing - Use bmap to allocate blocks
323  * @page: The (locked) page to allocate backing for
324  *
325  * We try to allocate all the blocks required for the page in
326  * one go. This might fail for various reasons, so we keep
327  * trying until all the blocks to back this page are allocated.
328  * If some of the blocks are already allocated, thats ok too.
329  */
330 
gfs2_allocate_page_backing(struct page * page)331 static int gfs2_allocate_page_backing(struct page *page)
332 {
333 	struct inode *inode = page->mapping->host;
334 	struct buffer_head bh;
335 	unsigned long size = PAGE_CACHE_SIZE;
336 	u64 lblock = page->index << (PAGE_CACHE_SHIFT - inode->i_blkbits);
337 
338 	do {
339 		bh.b_state = 0;
340 		bh.b_size = size;
341 		gfs2_block_map(inode, lblock, &bh, 1);
342 		if (!buffer_mapped(&bh))
343 			return -EIO;
344 		size -= bh.b_size;
345 		lblock += (bh.b_size >> inode->i_blkbits);
346 	} while(size > 0);
347 	return 0;
348 }
349 
350 /**
351  * gfs2_page_mkwrite - Make a shared, mmap()ed, page writable
352  * @vma: The virtual memory area
353  * @page: The page which is about to become writable
354  *
355  * When the page becomes writable, we need to ensure that we have
356  * blocks allocated on disk to back that page.
357  */
358 
gfs2_page_mkwrite(struct vm_area_struct * vma,struct vm_fault * vmf)359 static int gfs2_page_mkwrite(struct vm_area_struct *vma, struct vm_fault *vmf)
360 {
361 	struct page *page = vmf->page;
362 	struct inode *inode = vma->vm_file->f_path.dentry->d_inode;
363 	struct gfs2_inode *ip = GFS2_I(inode);
364 	struct gfs2_sbd *sdp = GFS2_SB(inode);
365 	unsigned long last_index;
366 	u64 pos = page->index << PAGE_CACHE_SHIFT;
367 	unsigned int data_blocks, ind_blocks, rblocks;
368 	struct gfs2_holder gh;
369 	struct gfs2_qadata *qa;
370 	loff_t size;
371 	int ret;
372 
373 	/* Wait if fs is frozen. This is racy so we check again later on
374 	 * and retry if the fs has been frozen after the page lock has
375 	 * been acquired
376 	 */
377 	vfs_check_frozen(inode->i_sb, SB_FREEZE_WRITE);
378 
379 	gfs2_holder_init(ip->i_gl, LM_ST_EXCLUSIVE, 0, &gh);
380 	ret = gfs2_glock_nq(&gh);
381 	if (ret)
382 		goto out;
383 
384 	set_bit(GLF_DIRTY, &ip->i_gl->gl_flags);
385 	set_bit(GIF_SW_PAGED, &ip->i_flags);
386 
387 	if (!gfs2_write_alloc_required(ip, pos, PAGE_CACHE_SIZE)) {
388 		lock_page(page);
389 		if (!PageUptodate(page) || page->mapping != inode->i_mapping) {
390 			ret = -EAGAIN;
391 			unlock_page(page);
392 		}
393 		goto out_unlock;
394 	}
395 
396 	ret = -ENOMEM;
397 	qa = gfs2_qadata_get(ip);
398 	if (qa == NULL)
399 		goto out_unlock;
400 
401 	ret = gfs2_quota_lock_check(ip);
402 	if (ret)
403 		goto out_alloc_put;
404 	gfs2_write_calc_reserv(ip, PAGE_CACHE_SIZE, &data_blocks, &ind_blocks);
405 	ret = gfs2_inplace_reserve(ip, data_blocks + ind_blocks);
406 	if (ret)
407 		goto out_quota_unlock;
408 
409 	rblocks = RES_DINODE + ind_blocks;
410 	if (gfs2_is_jdata(ip))
411 		rblocks += data_blocks ? data_blocks : 1;
412 	if (ind_blocks || data_blocks) {
413 		rblocks += RES_STATFS + RES_QUOTA;
414 		rblocks += gfs2_rg_blocks(ip);
415 	}
416 	ret = gfs2_trans_begin(sdp, rblocks, 0);
417 	if (ret)
418 		goto out_trans_fail;
419 
420 	lock_page(page);
421 	ret = -EINVAL;
422 	size = i_size_read(inode);
423 	last_index = (size - 1) >> PAGE_CACHE_SHIFT;
424 	/* Check page index against inode size */
425 	if (size == 0 || (page->index > last_index))
426 		goto out_trans_end;
427 
428 	ret = -EAGAIN;
429 	/* If truncated, we must retry the operation, we may have raced
430 	 * with the glock demotion code.
431 	 */
432 	if (!PageUptodate(page) || page->mapping != inode->i_mapping)
433 		goto out_trans_end;
434 
435 	/* Unstuff, if required, and allocate backing blocks for page */
436 	ret = 0;
437 	if (gfs2_is_stuffed(ip))
438 		ret = gfs2_unstuff_dinode(ip, page);
439 	if (ret == 0)
440 		ret = gfs2_allocate_page_backing(page);
441 
442 out_trans_end:
443 	if (ret)
444 		unlock_page(page);
445 	gfs2_trans_end(sdp);
446 out_trans_fail:
447 	gfs2_inplace_release(ip);
448 out_quota_unlock:
449 	gfs2_quota_unlock(ip);
450 out_alloc_put:
451 	gfs2_qadata_put(ip);
452 out_unlock:
453 	gfs2_glock_dq(&gh);
454 out:
455 	gfs2_holder_uninit(&gh);
456 	if (ret == 0) {
457 		set_page_dirty(page);
458 		/* This check must be post dropping of transaction lock */
459 		if (inode->i_sb->s_frozen == SB_UNFROZEN) {
460 			wait_on_page_writeback(page);
461 		} else {
462 			ret = -EAGAIN;
463 			unlock_page(page);
464 		}
465 	}
466 	return block_page_mkwrite_return(ret);
467 }
468 
469 static const struct vm_operations_struct gfs2_vm_ops = {
470 	.fault = filemap_fault,
471 	.page_mkwrite = gfs2_page_mkwrite,
472 };
473 
474 /**
475  * gfs2_mmap -
476  * @file: The file to map
477  * @vma: The VMA which described the mapping
478  *
479  * There is no need to get a lock here unless we should be updating
480  * atime. We ignore any locking errors since the only consequence is
481  * a missed atime update (which will just be deferred until later).
482  *
483  * Returns: 0
484  */
485 
gfs2_mmap(struct file * file,struct vm_area_struct * vma)486 static int gfs2_mmap(struct file *file, struct vm_area_struct *vma)
487 {
488 	struct gfs2_inode *ip = GFS2_I(file->f_mapping->host);
489 
490 	if (!(file->f_flags & O_NOATIME) &&
491 	    !IS_NOATIME(&ip->i_inode)) {
492 		struct gfs2_holder i_gh;
493 		int error;
494 
495 		gfs2_holder_init(ip->i_gl, LM_ST_SHARED, LM_FLAG_ANY, &i_gh);
496 		error = gfs2_glock_nq(&i_gh);
497 		if (error == 0) {
498 			file_accessed(file);
499 			gfs2_glock_dq(&i_gh);
500 		}
501 		gfs2_holder_uninit(&i_gh);
502 		if (error)
503 			return error;
504 	}
505 	vma->vm_ops = &gfs2_vm_ops;
506 	vma->vm_flags |= VM_CAN_NONLINEAR;
507 
508 	return 0;
509 }
510 
511 /**
512  * gfs2_open - open a file
513  * @inode: the inode to open
514  * @file: the struct file for this opening
515  *
516  * Returns: errno
517  */
518 
gfs2_open(struct inode * inode,struct file * file)519 static int gfs2_open(struct inode *inode, struct file *file)
520 {
521 	struct gfs2_inode *ip = GFS2_I(inode);
522 	struct gfs2_holder i_gh;
523 	struct gfs2_file *fp;
524 	int error;
525 
526 	fp = kzalloc(sizeof(struct gfs2_file), GFP_KERNEL);
527 	if (!fp)
528 		return -ENOMEM;
529 
530 	mutex_init(&fp->f_fl_mutex);
531 
532 	gfs2_assert_warn(GFS2_SB(inode), !file->private_data);
533 	file->private_data = fp;
534 
535 	if (S_ISREG(ip->i_inode.i_mode)) {
536 		error = gfs2_glock_nq_init(ip->i_gl, LM_ST_SHARED, LM_FLAG_ANY,
537 					   &i_gh);
538 		if (error)
539 			goto fail;
540 
541 		if (!(file->f_flags & O_LARGEFILE) &&
542 		    i_size_read(inode) > MAX_NON_LFS) {
543 			error = -EOVERFLOW;
544 			goto fail_gunlock;
545 		}
546 
547 		gfs2_glock_dq_uninit(&i_gh);
548 	}
549 
550 	return 0;
551 
552 fail_gunlock:
553 	gfs2_glock_dq_uninit(&i_gh);
554 fail:
555 	file->private_data = NULL;
556 	kfree(fp);
557 	return error;
558 }
559 
560 /**
561  * gfs2_close - called to close a struct file
562  * @inode: the inode the struct file belongs to
563  * @file: the struct file being closed
564  *
565  * Returns: errno
566  */
567 
gfs2_close(struct inode * inode,struct file * file)568 static int gfs2_close(struct inode *inode, struct file *file)
569 {
570 	struct gfs2_sbd *sdp = inode->i_sb->s_fs_info;
571 	struct gfs2_file *fp;
572 
573 	fp = file->private_data;
574 	file->private_data = NULL;
575 
576 	if (gfs2_assert_warn(sdp, fp))
577 		return -EIO;
578 
579 	kfree(fp);
580 
581 	return 0;
582 }
583 
584 /**
585  * gfs2_fsync - sync the dirty data for a file (across the cluster)
586  * @file: the file that points to the dentry
587  * @start: the start position in the file to sync
588  * @end: the end position in the file to sync
589  * @datasync: set if we can ignore timestamp changes
590  *
591  * We split the data flushing here so that we don't wait for the data
592  * until after we've also sent the metadata to disk. Note that for
593  * data=ordered, we will write & wait for the data at the log flush
594  * stage anyway, so this is unlikely to make much of a difference
595  * except in the data=writeback case.
596  *
597  * If the fdatawrite fails due to any reason except -EIO, we will
598  * continue the remainder of the fsync, although we'll still report
599  * the error at the end. This is to match filemap_write_and_wait_range()
600  * behaviour.
601  *
602  * Returns: errno
603  */
604 
gfs2_fsync(struct file * file,loff_t start,loff_t end,int datasync)605 static int gfs2_fsync(struct file *file, loff_t start, loff_t end,
606 		      int datasync)
607 {
608 	struct address_space *mapping = file->f_mapping;
609 	struct inode *inode = mapping->host;
610 	int sync_state = inode->i_state & (I_DIRTY_SYNC|I_DIRTY_DATASYNC);
611 	struct gfs2_inode *ip = GFS2_I(inode);
612 	int ret = 0, ret1 = 0;
613 
614 	if (mapping->nrpages) {
615 		ret1 = filemap_fdatawrite_range(mapping, start, end);
616 		if (ret1 == -EIO)
617 			return ret1;
618 	}
619 
620 	if (datasync)
621 		sync_state &= ~I_DIRTY_SYNC;
622 
623 	if (sync_state) {
624 		ret = sync_inode_metadata(inode, 1);
625 		if (ret)
626 			return ret;
627 		if (gfs2_is_jdata(ip))
628 			filemap_write_and_wait(mapping);
629 		gfs2_ail_flush(ip->i_gl, 1);
630 	}
631 
632 	if (mapping->nrpages)
633 		ret = filemap_fdatawait_range(mapping, start, end);
634 
635 	return ret ? ret : ret1;
636 }
637 
638 /**
639  * gfs2_file_aio_write - Perform a write to a file
640  * @iocb: The io context
641  * @iov: The data to write
642  * @nr_segs: Number of @iov segments
643  * @pos: The file position
644  *
645  * We have to do a lock/unlock here to refresh the inode size for
646  * O_APPEND writes, otherwise we can land up writing at the wrong
647  * offset. There is still a race, but provided the app is using its
648  * own file locking, this will make O_APPEND work as expected.
649  *
650  */
651 
gfs2_file_aio_write(struct kiocb * iocb,const struct iovec * iov,unsigned long nr_segs,loff_t pos)652 static ssize_t gfs2_file_aio_write(struct kiocb *iocb, const struct iovec *iov,
653 				   unsigned long nr_segs, loff_t pos)
654 {
655 	struct file *file = iocb->ki_filp;
656 
657 	if (file->f_flags & O_APPEND) {
658 		struct dentry *dentry = file->f_dentry;
659 		struct gfs2_inode *ip = GFS2_I(dentry->d_inode);
660 		struct gfs2_holder gh;
661 		int ret;
662 
663 		ret = gfs2_glock_nq_init(ip->i_gl, LM_ST_SHARED, 0, &gh);
664 		if (ret)
665 			return ret;
666 		gfs2_glock_dq_uninit(&gh);
667 	}
668 
669 	return generic_file_aio_write(iocb, iov, nr_segs, pos);
670 }
671 
fallocate_chunk(struct inode * inode,loff_t offset,loff_t len,int mode)672 static int fallocate_chunk(struct inode *inode, loff_t offset, loff_t len,
673 			   int mode)
674 {
675 	struct gfs2_inode *ip = GFS2_I(inode);
676 	struct buffer_head *dibh;
677 	int error;
678 	loff_t size = len;
679 	unsigned int nr_blks;
680 	sector_t lblock = offset >> inode->i_blkbits;
681 
682 	error = gfs2_meta_inode_buffer(ip, &dibh);
683 	if (unlikely(error))
684 		return error;
685 
686 	gfs2_trans_add_bh(ip->i_gl, dibh, 1);
687 
688 	if (gfs2_is_stuffed(ip)) {
689 		error = gfs2_unstuff_dinode(ip, NULL);
690 		if (unlikely(error))
691 			goto out;
692 	}
693 
694 	while (len) {
695 		struct buffer_head bh_map = { .b_state = 0, .b_blocknr = 0 };
696 		bh_map.b_size = len;
697 		set_buffer_zeronew(&bh_map);
698 
699 		error = gfs2_block_map(inode, lblock, &bh_map, 1);
700 		if (unlikely(error))
701 			goto out;
702 		len -= bh_map.b_size;
703 		nr_blks = bh_map.b_size >> inode->i_blkbits;
704 		lblock += nr_blks;
705 		if (!buffer_new(&bh_map))
706 			continue;
707 		if (unlikely(!buffer_zeronew(&bh_map))) {
708 			error = -EIO;
709 			goto out;
710 		}
711 	}
712 	if (offset + size > inode->i_size && !(mode & FALLOC_FL_KEEP_SIZE))
713 		i_size_write(inode, offset + size);
714 
715 	mark_inode_dirty(inode);
716 
717 out:
718 	brelse(dibh);
719 	return error;
720 }
721 
calc_max_reserv(struct gfs2_inode * ip,loff_t max,loff_t * len,unsigned int * data_blocks,unsigned int * ind_blocks)722 static void calc_max_reserv(struct gfs2_inode *ip, loff_t max, loff_t *len,
723 			    unsigned int *data_blocks, unsigned int *ind_blocks)
724 {
725 	const struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode);
726 	unsigned int max_blocks = ip->i_rgd->rd_free_clone;
727 	unsigned int tmp, max_data = max_blocks - 3 * (sdp->sd_max_height - 1);
728 
729 	for (tmp = max_data; tmp > sdp->sd_diptrs;) {
730 		tmp = DIV_ROUND_UP(tmp, sdp->sd_inptrs);
731 		max_data -= tmp;
732 	}
733 	/* This calculation isn't the exact reverse of gfs2_write_calc_reserve,
734 	   so it might end up with fewer data blocks */
735 	if (max_data <= *data_blocks)
736 		return;
737 	*data_blocks = max_data;
738 	*ind_blocks = max_blocks - max_data;
739 	*len = ((loff_t)max_data - 3) << sdp->sd_sb.sb_bsize_shift;
740 	if (*len > max) {
741 		*len = max;
742 		gfs2_write_calc_reserv(ip, max, data_blocks, ind_blocks);
743 	}
744 }
745 
gfs2_fallocate(struct file * file,int mode,loff_t offset,loff_t len)746 static long gfs2_fallocate(struct file *file, int mode, loff_t offset,
747 			   loff_t len)
748 {
749 	struct inode *inode = file->f_path.dentry->d_inode;
750 	struct gfs2_sbd *sdp = GFS2_SB(inode);
751 	struct gfs2_inode *ip = GFS2_I(inode);
752 	unsigned int data_blocks = 0, ind_blocks = 0, rblocks;
753 	loff_t bytes, max_bytes;
754 	struct gfs2_qadata *qa;
755 	int error;
756 	const loff_t pos = offset;
757 	const loff_t count = len;
758 	loff_t bsize_mask = ~((loff_t)sdp->sd_sb.sb_bsize - 1);
759 	loff_t next = (offset + len - 1) >> sdp->sd_sb.sb_bsize_shift;
760 	loff_t max_chunk_size = UINT_MAX & bsize_mask;
761 	next = (next + 1) << sdp->sd_sb.sb_bsize_shift;
762 
763 	/* We only support the FALLOC_FL_KEEP_SIZE mode */
764 	if (mode & ~FALLOC_FL_KEEP_SIZE)
765 		return -EOPNOTSUPP;
766 
767 	offset &= bsize_mask;
768 
769 	len = next - offset;
770 	bytes = sdp->sd_max_rg_data * sdp->sd_sb.sb_bsize / 2;
771 	if (!bytes)
772 		bytes = UINT_MAX;
773 	bytes &= bsize_mask;
774 	if (bytes == 0)
775 		bytes = sdp->sd_sb.sb_bsize;
776 
777 	gfs2_holder_init(ip->i_gl, LM_ST_EXCLUSIVE, 0, &ip->i_gh);
778 	error = gfs2_glock_nq(&ip->i_gh);
779 	if (unlikely(error))
780 		goto out_uninit;
781 
782 	while (len > 0) {
783 		if (len < bytes)
784 			bytes = len;
785 		if (!gfs2_write_alloc_required(ip, offset, bytes)) {
786 			len -= bytes;
787 			offset += bytes;
788 			continue;
789 		}
790 		qa = gfs2_qadata_get(ip);
791 		if (!qa) {
792 			error = -ENOMEM;
793 			goto out_unlock;
794 		}
795 
796 		error = gfs2_quota_lock_check(ip);
797 		if (error)
798 			goto out_alloc_put;
799 
800 retry:
801 		gfs2_write_calc_reserv(ip, bytes, &data_blocks, &ind_blocks);
802 
803 		error = gfs2_inplace_reserve(ip, data_blocks + ind_blocks);
804 		if (error) {
805 			if (error == -ENOSPC && bytes > sdp->sd_sb.sb_bsize) {
806 				bytes >>= 1;
807 				bytes &= bsize_mask;
808 				if (bytes == 0)
809 					bytes = sdp->sd_sb.sb_bsize;
810 				goto retry;
811 			}
812 			goto out_qunlock;
813 		}
814 		max_bytes = bytes;
815 		calc_max_reserv(ip, (len > max_chunk_size)? max_chunk_size: len,
816 				&max_bytes, &data_blocks, &ind_blocks);
817 
818 		rblocks = RES_DINODE + ind_blocks + RES_STATFS + RES_QUOTA +
819 			  RES_RG_HDR + gfs2_rg_blocks(ip);
820 		if (gfs2_is_jdata(ip))
821 			rblocks += data_blocks ? data_blocks : 1;
822 
823 		error = gfs2_trans_begin(sdp, rblocks,
824 					 PAGE_CACHE_SIZE/sdp->sd_sb.sb_bsize);
825 		if (error)
826 			goto out_trans_fail;
827 
828 		error = fallocate_chunk(inode, offset, max_bytes, mode);
829 		gfs2_trans_end(sdp);
830 
831 		if (error)
832 			goto out_trans_fail;
833 
834 		len -= max_bytes;
835 		offset += max_bytes;
836 		gfs2_inplace_release(ip);
837 		gfs2_quota_unlock(ip);
838 		gfs2_qadata_put(ip);
839 	}
840 
841 	if (error == 0)
842 		error = generic_write_sync(file, pos, count);
843 	goto out_unlock;
844 
845 out_trans_fail:
846 	gfs2_inplace_release(ip);
847 out_qunlock:
848 	gfs2_quota_unlock(ip);
849 out_alloc_put:
850 	gfs2_qadata_put(ip);
851 out_unlock:
852 	gfs2_glock_dq(&ip->i_gh);
853 out_uninit:
854 	gfs2_holder_uninit(&ip->i_gh);
855 	return error;
856 }
857 
858 #ifdef CONFIG_GFS2_FS_LOCKING_DLM
859 
860 /**
861  * gfs2_setlease - acquire/release a file lease
862  * @file: the file pointer
863  * @arg: lease type
864  * @fl: file lock
865  *
866  * We don't currently have a way to enforce a lease across the whole
867  * cluster; until we do, disable leases (by just returning -EINVAL),
868  * unless the administrator has requested purely local locking.
869  *
870  * Locking: called under lock_flocks
871  *
872  * Returns: errno
873  */
874 
gfs2_setlease(struct file * file,long arg,struct file_lock ** fl)875 static int gfs2_setlease(struct file *file, long arg, struct file_lock **fl)
876 {
877 	return -EINVAL;
878 }
879 
880 /**
881  * gfs2_lock - acquire/release a posix lock on a file
882  * @file: the file pointer
883  * @cmd: either modify or retrieve lock state, possibly wait
884  * @fl: type and range of lock
885  *
886  * Returns: errno
887  */
888 
gfs2_lock(struct file * file,int cmd,struct file_lock * fl)889 static int gfs2_lock(struct file *file, int cmd, struct file_lock *fl)
890 {
891 	struct gfs2_inode *ip = GFS2_I(file->f_mapping->host);
892 	struct gfs2_sbd *sdp = GFS2_SB(file->f_mapping->host);
893 	struct lm_lockstruct *ls = &sdp->sd_lockstruct;
894 
895 	if (!(fl->fl_flags & FL_POSIX))
896 		return -ENOLCK;
897 	if (__mandatory_lock(&ip->i_inode) && fl->fl_type != F_UNLCK)
898 		return -ENOLCK;
899 
900 	if (cmd == F_CANCELLK) {
901 		/* Hack: */
902 		cmd = F_SETLK;
903 		fl->fl_type = F_UNLCK;
904 	}
905 	if (unlikely(test_bit(SDF_SHUTDOWN, &sdp->sd_flags)))
906 		return -EIO;
907 	if (IS_GETLK(cmd))
908 		return dlm_posix_get(ls->ls_dlm, ip->i_no_addr, file, fl);
909 	else if (fl->fl_type == F_UNLCK)
910 		return dlm_posix_unlock(ls->ls_dlm, ip->i_no_addr, file, fl);
911 	else
912 		return dlm_posix_lock(ls->ls_dlm, ip->i_no_addr, file, cmd, fl);
913 }
914 
do_flock(struct file * file,int cmd,struct file_lock * fl)915 static int do_flock(struct file *file, int cmd, struct file_lock *fl)
916 {
917 	struct gfs2_file *fp = file->private_data;
918 	struct gfs2_holder *fl_gh = &fp->f_fl_gh;
919 	struct gfs2_inode *ip = GFS2_I(file->f_path.dentry->d_inode);
920 	struct gfs2_glock *gl;
921 	unsigned int state;
922 	int flags;
923 	int error = 0;
924 
925 	state = (fl->fl_type == F_WRLCK) ? LM_ST_EXCLUSIVE : LM_ST_SHARED;
926 	flags = (IS_SETLKW(cmd) ? 0 : LM_FLAG_TRY) | GL_EXACT | GL_NOCACHE;
927 
928 	mutex_lock(&fp->f_fl_mutex);
929 
930 	gl = fl_gh->gh_gl;
931 	if (gl) {
932 		if (fl_gh->gh_state == state)
933 			goto out;
934 		flock_lock_file_wait(file,
935 				     &(struct file_lock){.fl_type = F_UNLCK});
936 		gfs2_glock_dq_wait(fl_gh);
937 		gfs2_holder_reinit(state, flags, fl_gh);
938 	} else {
939 		error = gfs2_glock_get(GFS2_SB(&ip->i_inode), ip->i_no_addr,
940 				       &gfs2_flock_glops, CREATE, &gl);
941 		if (error)
942 			goto out;
943 		gfs2_holder_init(gl, state, flags, fl_gh);
944 		gfs2_glock_put(gl);
945 	}
946 	error = gfs2_glock_nq(fl_gh);
947 	if (error) {
948 		gfs2_holder_uninit(fl_gh);
949 		if (error == GLR_TRYFAILED)
950 			error = -EAGAIN;
951 	} else {
952 		error = flock_lock_file_wait(file, fl);
953 		gfs2_assert_warn(GFS2_SB(&ip->i_inode), !error);
954 	}
955 
956 out:
957 	mutex_unlock(&fp->f_fl_mutex);
958 	return error;
959 }
960 
do_unflock(struct file * file,struct file_lock * fl)961 static void do_unflock(struct file *file, struct file_lock *fl)
962 {
963 	struct gfs2_file *fp = file->private_data;
964 	struct gfs2_holder *fl_gh = &fp->f_fl_gh;
965 
966 	mutex_lock(&fp->f_fl_mutex);
967 	flock_lock_file_wait(file, fl);
968 	if (fl_gh->gh_gl) {
969 		gfs2_glock_dq_wait(fl_gh);
970 		gfs2_holder_uninit(fl_gh);
971 	}
972 	mutex_unlock(&fp->f_fl_mutex);
973 }
974 
975 /**
976  * gfs2_flock - acquire/release a flock lock on a file
977  * @file: the file pointer
978  * @cmd: either modify or retrieve lock state, possibly wait
979  * @fl: type and range of lock
980  *
981  * Returns: errno
982  */
983 
gfs2_flock(struct file * file,int cmd,struct file_lock * fl)984 static int gfs2_flock(struct file *file, int cmd, struct file_lock *fl)
985 {
986 	if (!(fl->fl_flags & FL_FLOCK))
987 		return -ENOLCK;
988 	if (fl->fl_type & LOCK_MAND)
989 		return -EOPNOTSUPP;
990 
991 	if (fl->fl_type == F_UNLCK) {
992 		do_unflock(file, fl);
993 		return 0;
994 	} else {
995 		return do_flock(file, cmd, fl);
996 	}
997 }
998 
999 const struct file_operations gfs2_file_fops = {
1000 	.llseek		= gfs2_llseek,
1001 	.read		= do_sync_read,
1002 	.aio_read	= generic_file_aio_read,
1003 	.write		= do_sync_write,
1004 	.aio_write	= gfs2_file_aio_write,
1005 	.unlocked_ioctl	= gfs2_ioctl,
1006 	.mmap		= gfs2_mmap,
1007 	.open		= gfs2_open,
1008 	.release	= gfs2_close,
1009 	.fsync		= gfs2_fsync,
1010 	.lock		= gfs2_lock,
1011 	.flock		= gfs2_flock,
1012 	.splice_read	= generic_file_splice_read,
1013 	.splice_write	= generic_file_splice_write,
1014 	.setlease	= gfs2_setlease,
1015 	.fallocate	= gfs2_fallocate,
1016 };
1017 
1018 const struct file_operations gfs2_dir_fops = {
1019 	.readdir	= gfs2_readdir,
1020 	.unlocked_ioctl	= gfs2_ioctl,
1021 	.open		= gfs2_open,
1022 	.release	= gfs2_close,
1023 	.fsync		= gfs2_fsync,
1024 	.lock		= gfs2_lock,
1025 	.flock		= gfs2_flock,
1026 	.llseek		= default_llseek,
1027 };
1028 
1029 #endif /* CONFIG_GFS2_FS_LOCKING_DLM */
1030 
1031 const struct file_operations gfs2_file_fops_nolock = {
1032 	.llseek		= gfs2_llseek,
1033 	.read		= do_sync_read,
1034 	.aio_read	= generic_file_aio_read,
1035 	.write		= do_sync_write,
1036 	.aio_write	= gfs2_file_aio_write,
1037 	.unlocked_ioctl	= gfs2_ioctl,
1038 	.mmap		= gfs2_mmap,
1039 	.open		= gfs2_open,
1040 	.release	= gfs2_close,
1041 	.fsync		= gfs2_fsync,
1042 	.splice_read	= generic_file_splice_read,
1043 	.splice_write	= generic_file_splice_write,
1044 	.setlease	= generic_setlease,
1045 	.fallocate	= gfs2_fallocate,
1046 };
1047 
1048 const struct file_operations gfs2_dir_fops_nolock = {
1049 	.readdir	= gfs2_readdir,
1050 	.unlocked_ioctl	= gfs2_ioctl,
1051 	.open		= gfs2_open,
1052 	.release	= gfs2_close,
1053 	.fsync		= gfs2_fsync,
1054 	.llseek		= default_llseek,
1055 };
1056 
1057