1/*
  2 * fs/logfs/file.c	- prepare_write, commit_write and friends
  3 *
  4 * As should be obvious for Linux kernel code, license is GPLv2
  5 *
  6 * Copyright (c) 2005-2008 Joern Engel <joern@logfs.org>
  7 */
  8#include "logfs.h"
  9#include <linux/sched.h>
 10#include <linux/writeback.h>
 11
 12static int logfs_write_begin(struct file *file, struct address_space *mapping,
 13		loff_t pos, unsigned len, unsigned flags,
 14		struct page **pagep, void **fsdata)
 15{
 16	struct inode *inode = mapping->host;
 17	struct page *page;
 18	pgoff_t index = pos >> PAGE_SHIFT;
 19
 20	page = grab_cache_page_write_begin(mapping, index, flags);
 21	if (!page)
 22		return -ENOMEM;
 23	*pagep = page;
 24
 25	if ((len == PAGE_SIZE) || PageUptodate(page))
 26		return 0;
 27	if ((pos & PAGE_MASK) >= i_size_read(inode)) {
 28		unsigned start = pos & (PAGE_SIZE - 1);
 29		unsigned end = start + len;
 30
 31		/* Reading beyond i_size is simple: memset to zero */
 32		zero_user_segments(page, 0, start, end, PAGE_SIZE);
 33		return 0;
 34	}
 35	return logfs_readpage_nolock(page);
 36}
 37
 38static int logfs_write_end(struct file *file, struct address_space *mapping,
 39		loff_t pos, unsigned len, unsigned copied, struct page *page,
 40		void *fsdata)
 41{
 42	struct inode *inode = mapping->host;
 43	pgoff_t index = page->index;
 44	unsigned start = pos & (PAGE_SIZE - 1);
 45	unsigned end = start + copied;
 46	int ret = 0;
 47
 48	BUG_ON(PAGE_SIZE != inode->i_sb->s_blocksize);
 49	BUG_ON(page->index > I3_BLOCKS);
 50
 51	if (copied < len) {
 52		/*
 53		 * Short write of a non-initialized paged.  Just tell userspace
 54		 * to retry the entire page.
 55		 */
 56		if (!PageUptodate(page)) {
 57			copied = 0;
 58			goto out;
 59		}
 60	}
 61	if (copied == 0)
 62		goto out; /* FIXME: do we need to update inode? */
 63
 64	if (i_size_read(inode) < (index << PAGE_SHIFT) + end) {
 65		i_size_write(inode, (index << PAGE_SHIFT) + end);
 66		mark_inode_dirty_sync(inode);
 67	}
 68
 69	SetPageUptodate(page);
 70	if (!PageDirty(page)) {
 71		if (!get_page_reserve(inode, page))
 72			__set_page_dirty_nobuffers(page);
 73		else
 74			ret = logfs_write_buf(inode, page, WF_LOCK);
 75	}
 76out:
 77	unlock_page(page);
 78	put_page(page);
 79	return ret ? ret : copied;
 80}
 81
 82int logfs_readpage(struct file *file, struct page *page)
 83{
 84	int ret;
 85
 86	ret = logfs_readpage_nolock(page);
 87	unlock_page(page);
 88	return ret;
 89}
 90
 91/* Clear the page's dirty flag in the radix tree. */
 92/* TODO: mucking with PageWriteback is silly.  Add a generic function to clear
 93 * the dirty bit from the radix tree for filesystems that don't have to wait
 94 * for page writeback to finish (i.e. any compressing filesystem).
 95 */
 96static void clear_radix_tree_dirty(struct page *page)
 97{
 98	BUG_ON(PagePrivate(page) || page->private);
 99	set_page_writeback(page);
100	end_page_writeback(page);
101}
102
103static int __logfs_writepage(struct page *page)
104{
105	struct inode *inode = page->mapping->host;
106	int err;
107
108	err = logfs_write_buf(inode, page, WF_LOCK);
109	if (err)
110		set_page_dirty(page);
111	else
112		clear_radix_tree_dirty(page);
113	unlock_page(page);
114	return err;
115}
116
117static int logfs_writepage(struct page *page, struct writeback_control *wbc)
118{
119	struct inode *inode = page->mapping->host;
120	loff_t i_size = i_size_read(inode);
121	pgoff_t end_index = i_size >> PAGE_SHIFT;
122	unsigned offset;
123	u64 bix;
124	level_t level;
125
126	log_file("logfs_writepage(%lx, %lx, %p)\n", inode->i_ino, page->index,
127			page);
128
129	logfs_unpack_index(page->index, &bix, &level);
130
131	/* Indirect blocks are never truncated */
132	if (level != 0)
133		return __logfs_writepage(page);
134
135	/*
136	 * TODO: everything below is a near-verbatim copy of nobh_writepage().
137	 * The relevant bits should be factored out after logfs is merged.
138	 */
139
140	/* Is the page fully inside i_size? */
141	if (bix < end_index)
142		return __logfs_writepage(page);
143
144	 /* Is the page fully outside i_size? (truncate in progress) */
145	offset = i_size & (PAGE_SIZE-1);
146	if (bix > end_index || offset == 0) {
147		unlock_page(page);
148		return 0; /* don't care */
149	}
150
151	/*
152	 * The page straddles i_size.  It must be zeroed out on each and every
153	 * writepage invokation because it may be mmapped.  "A file is mapped
154	 * in multiples of the page size.  For a file that is not a multiple of
155	 * the  page size, the remaining memory is zeroed when mapped, and
156	 * writes to that region are not written out to the file."
157	 */
158	zero_user_segment(page, offset, PAGE_SIZE);
159	return __logfs_writepage(page);
160}
161
162static void logfs_invalidatepage(struct page *page, unsigned int offset,
163				 unsigned int length)
164{
165	struct logfs_block *block = logfs_block(page);
166
167	if (block->reserved_bytes) {
168		struct super_block *sb = page->mapping->host->i_sb;
169		struct logfs_super *super = logfs_super(sb);
170
171		super->s_dirty_pages -= block->reserved_bytes;
172		block->ops->free_block(sb, block);
173		BUG_ON(bitmap_weight(block->alias_map, LOGFS_BLOCK_FACTOR));
174	} else
175		move_page_to_btree(page);
176	BUG_ON(PagePrivate(page) || page->private);
177}
178
179static int logfs_releasepage(struct page *page, gfp_t only_xfs_uses_this)
180{
181	return 0; /* None of these are easy to release */
182}
183
184
185long logfs_ioctl(struct file *file, unsigned int cmd, unsigned long arg)
186{
187	struct inode *inode = file_inode(file);
188	struct logfs_inode *li = logfs_inode(inode);
189	unsigned int oldflags, flags;
190	int err;
191
192	switch (cmd) {
193	case FS_IOC_GETFLAGS:
194		flags = li->li_flags & LOGFS_FL_USER_VISIBLE;
195		return put_user(flags, (int __user *)arg);
196	case FS_IOC_SETFLAGS:
197		if (IS_RDONLY(inode))
198			return -EROFS;
199
200		if (!inode_owner_or_capable(inode))
201			return -EACCES;
202
203		err = get_user(flags, (int __user *)arg);
204		if (err)
205			return err;
206
207		inode_lock(inode);
208		oldflags = li->li_flags;
209		flags &= LOGFS_FL_USER_MODIFIABLE;
210		flags |= oldflags & ~LOGFS_FL_USER_MODIFIABLE;
211		li->li_flags = flags;
212		inode_unlock(inode);
213
214		inode->i_ctime = CURRENT_TIME;
215		mark_inode_dirty_sync(inode);
216		return 0;
217
218	default:
219		return -ENOTTY;
220	}
221}
222
223int logfs_fsync(struct file *file, loff_t start, loff_t end, int datasync)
224{
225	struct super_block *sb = file->f_mapping->host->i_sb;
226	struct inode *inode = file->f_mapping->host;
227	int ret;
228
229	ret = filemap_write_and_wait_range(inode->i_mapping, start, end);
230	if (ret)
231		return ret;
232
233	inode_lock(inode);
234	logfs_get_wblocks(sb, NULL, WF_LOCK);
235	logfs_write_anchor(sb);
236	logfs_put_wblocks(sb, NULL, WF_LOCK);
237	inode_unlock(inode);
238
239	return 0;
240}
241
242static int logfs_setattr(struct dentry *dentry, struct iattr *attr)
243{
244	struct inode *inode = d_inode(dentry);
245	int err = 0;
246
247	err = inode_change_ok(inode, attr);
248	if (err)
249		return err;
250
251	if (attr->ia_valid & ATTR_SIZE) {
252		err = logfs_truncate(inode, attr->ia_size);
253		if (err)
254			return err;
255	}
256
257	setattr_copy(inode, attr);
258	mark_inode_dirty(inode);
259	return 0;
260}
261
262const struct inode_operations logfs_reg_iops = {
263	.setattr	= logfs_setattr,
264};
265
266const struct file_operations logfs_reg_fops = {
267	.read_iter	= generic_file_read_iter,
268	.write_iter	= generic_file_write_iter,
269	.fsync		= logfs_fsync,
270	.unlocked_ioctl	= logfs_ioctl,
271	.llseek		= generic_file_llseek,
272	.mmap		= generic_file_readonly_mmap,
273	.open		= generic_file_open,
274};
275
276const struct address_space_operations logfs_reg_aops = {
277	.invalidatepage	= logfs_invalidatepage,
278	.readpage	= logfs_readpage,
279	.releasepage	= logfs_releasepage,
280	.set_page_dirty	= __set_page_dirty_nobuffers,
281	.writepage	= logfs_writepage,
282	.writepages	= generic_writepages,
283	.write_begin	= logfs_write_begin,
284	.write_end	= logfs_write_end,
285};