xref: /linux/fs/reiserfs/file.c (revision 861e10be08c69808065d755d3e3cab5d520a2d32)
1 /*
2  * Copyright 2000 by Hans Reiser, licensing governed by reiserfs/README
3  */
4 
5 #include <linux/time.h>
6 #include "reiserfs.h"
7 #include "acl.h"
8 #include "xattr.h"
9 #include <asm/uaccess.h>
10 #include <linux/pagemap.h>
11 #include <linux/swap.h>
12 #include <linux/writeback.h>
13 #include <linux/blkdev.h>
14 #include <linux/buffer_head.h>
15 #include <linux/quotaops.h>
16 
17 /*
18 ** We pack the tails of files on file close, not at the time they are written.
19 ** This implies an unnecessary copy of the tail and an unnecessary indirect item
20 ** insertion/balancing, for files that are written in one write.
21 ** It avoids unnecessary tail packings (balances) for files that are written in
22 ** multiple writes and are small enough to have tails.
23 **
24 ** file_release is called by the VFS layer when the file is closed.  If
25 ** this is the last open file descriptor, and the file
26 ** small enough to have a tail, and the tail is currently in an
27 ** unformatted node, the tail is converted back into a direct item.
28 **
29 ** We use reiserfs_truncate_file to pack the tail, since it already has
30 ** all the conditions coded.
31 */
32 static int reiserfs_file_release(struct inode *inode, struct file *filp)
33 {
34 
35 	struct reiserfs_transaction_handle th;
36 	int err;
37 	int jbegin_failure = 0;
38 
39 	BUG_ON(!S_ISREG(inode->i_mode));
40 
41         if (atomic_add_unless(&REISERFS_I(inode)->openers, -1, 1))
42 		return 0;
43 
44 	mutex_lock(&(REISERFS_I(inode)->tailpack));
45 
46         if (!atomic_dec_and_test(&REISERFS_I(inode)->openers)) {
47 		mutex_unlock(&(REISERFS_I(inode)->tailpack));
48 		return 0;
49 	}
50 
51 	/* fast out for when nothing needs to be done */
52 	if ((!(REISERFS_I(inode)->i_flags & i_pack_on_close_mask) ||
53 	     !tail_has_to_be_packed(inode)) &&
54 	    REISERFS_I(inode)->i_prealloc_count <= 0) {
55 		mutex_unlock(&(REISERFS_I(inode)->tailpack));
56 		return 0;
57 	}
58 
59 	reiserfs_write_lock(inode->i_sb);
60 	/* freeing preallocation only involves relogging blocks that
61 	 * are already in the current transaction.  preallocation gets
62 	 * freed at the end of each transaction, so it is impossible for
63 	 * us to log any additional blocks (including quota blocks)
64 	 */
65 	err = journal_begin(&th, inode->i_sb, 1);
66 	if (err) {
67 		/* uh oh, we can't allow the inode to go away while there
68 		 * is still preallocation blocks pending.  Try to join the
69 		 * aborted transaction
70 		 */
71 		jbegin_failure = err;
72 		err = journal_join_abort(&th, inode->i_sb, 1);
73 
74 		if (err) {
75 			/* hmpf, our choices here aren't good.  We can pin the inode
76 			 * which will disallow unmount from every happening, we can
77 			 * do nothing, which will corrupt random memory on unmount,
78 			 * or we can forcibly remove the file from the preallocation
79 			 * list, which will leak blocks on disk.  Lets pin the inode
80 			 * and let the admin know what is going on.
81 			 */
82 			igrab(inode);
83 			reiserfs_warning(inode->i_sb, "clm-9001",
84 					 "pinning inode %lu because the "
85 					 "preallocation can't be freed",
86 					 inode->i_ino);
87 			goto out;
88 		}
89 	}
90 	reiserfs_update_inode_transaction(inode);
91 
92 #ifdef REISERFS_PREALLOCATE
93 	reiserfs_discard_prealloc(&th, inode);
94 #endif
95 	err = journal_end(&th, inode->i_sb, 1);
96 
97 	/* copy back the error code from journal_begin */
98 	if (!err)
99 		err = jbegin_failure;
100 
101 	if (!err &&
102 	    (REISERFS_I(inode)->i_flags & i_pack_on_close_mask) &&
103 	    tail_has_to_be_packed(inode)) {
104 
105 		/* if regular file is released by last holder and it has been
106 		   appended (we append by unformatted node only) or its direct
107 		   item(s) had to be converted, then it may have to be
108 		   indirect2direct converted */
109 		err = reiserfs_truncate_file(inode, 0);
110 	}
111       out:
112 	reiserfs_write_unlock(inode->i_sb);
113 	mutex_unlock(&(REISERFS_I(inode)->tailpack));
114 	return err;
115 }
116 
117 static int reiserfs_file_open(struct inode *inode, struct file *file)
118 {
119 	int err = dquot_file_open(inode, file);
120         if (!atomic_inc_not_zero(&REISERFS_I(inode)->openers)) {
121 		/* somebody might be tailpacking on final close; wait for it */
122 		mutex_lock(&(REISERFS_I(inode)->tailpack));
123 		atomic_inc(&REISERFS_I(inode)->openers);
124 		mutex_unlock(&(REISERFS_I(inode)->tailpack));
125 	}
126 	return err;
127 }
128 
129 void reiserfs_vfs_truncate_file(struct inode *inode)
130 {
131 	mutex_lock(&(REISERFS_I(inode)->tailpack));
132 	reiserfs_truncate_file(inode, 1);
133 	mutex_unlock(&(REISERFS_I(inode)->tailpack));
134 }
135 
136 /* Sync a reiserfs file. */
137 
138 /*
139  * FIXME: sync_mapping_buffers() never has anything to sync.  Can
140  * be removed...
141  */
142 
143 static int reiserfs_sync_file(struct file *filp, loff_t start, loff_t end,
144 			      int datasync)
145 {
146 	struct inode *inode = filp->f_mapping->host;
147 	int err;
148 	int barrier_done;
149 
150 	err = filemap_write_and_wait_range(inode->i_mapping, start, end);
151 	if (err)
152 		return err;
153 
154 	mutex_lock(&inode->i_mutex);
155 	BUG_ON(!S_ISREG(inode->i_mode));
156 	err = sync_mapping_buffers(inode->i_mapping);
157 	reiserfs_write_lock(inode->i_sb);
158 	barrier_done = reiserfs_commit_for_inode(inode);
159 	reiserfs_write_unlock(inode->i_sb);
160 	if (barrier_done != 1 && reiserfs_barrier_flush(inode->i_sb))
161 		blkdev_issue_flush(inode->i_sb->s_bdev, GFP_KERNEL, NULL);
162 	mutex_unlock(&inode->i_mutex);
163 	if (barrier_done < 0)
164 		return barrier_done;
165 	return (err < 0) ? -EIO : 0;
166 }
167 
168 /* taken fs/buffer.c:__block_commit_write */
169 int reiserfs_commit_page(struct inode *inode, struct page *page,
170 			 unsigned from, unsigned to)
171 {
172 	unsigned block_start, block_end;
173 	int partial = 0;
174 	unsigned blocksize;
175 	struct buffer_head *bh, *head;
176 	unsigned long i_size_index = inode->i_size >> PAGE_CACHE_SHIFT;
177 	int new;
178 	int logit = reiserfs_file_data_log(inode);
179 	struct super_block *s = inode->i_sb;
180 	int bh_per_page = PAGE_CACHE_SIZE / s->s_blocksize;
181 	struct reiserfs_transaction_handle th;
182 	int ret = 0;
183 
184 	th.t_trans_id = 0;
185 	blocksize = 1 << inode->i_blkbits;
186 
187 	if (logit) {
188 		reiserfs_write_lock(s);
189 		ret = journal_begin(&th, s, bh_per_page + 1);
190 		if (ret)
191 			goto drop_write_lock;
192 		reiserfs_update_inode_transaction(inode);
193 	}
194 	for (bh = head = page_buffers(page), block_start = 0;
195 	     bh != head || !block_start;
196 	     block_start = block_end, bh = bh->b_this_page) {
197 
198 		new = buffer_new(bh);
199 		clear_buffer_new(bh);
200 		block_end = block_start + blocksize;
201 		if (block_end <= from || block_start >= to) {
202 			if (!buffer_uptodate(bh))
203 				partial = 1;
204 		} else {
205 			set_buffer_uptodate(bh);
206 			if (logit) {
207 				reiserfs_prepare_for_journal(s, bh, 1);
208 				journal_mark_dirty(&th, s, bh);
209 			} else if (!buffer_dirty(bh)) {
210 				mark_buffer_dirty(bh);
211 				/* do data=ordered on any page past the end
212 				 * of file and any buffer marked BH_New.
213 				 */
214 				if (reiserfs_data_ordered(inode->i_sb) &&
215 				    (new || page->index >= i_size_index)) {
216 					reiserfs_add_ordered_list(inode, bh);
217 				}
218 			}
219 		}
220 	}
221 	if (logit) {
222 		ret = journal_end(&th, s, bh_per_page + 1);
223 	      drop_write_lock:
224 		reiserfs_write_unlock(s);
225 	}
226 	/*
227 	 * If this is a partial write which happened to make all buffers
228 	 * uptodate then we can optimize away a bogus readpage() for
229 	 * the next read(). Here we 'discover' whether the page went
230 	 * uptodate as a result of this (potentially partial) write.
231 	 */
232 	if (!partial)
233 		SetPageUptodate(page);
234 	return ret;
235 }
236 
237 /* Write @count bytes at position @ppos in a file indicated by @file
238    from the buffer @buf.
239 
240    generic_file_write() is only appropriate for filesystems that are not seeking to optimize performance and want
241    something simple that works.  It is not for serious use by general purpose filesystems, excepting the one that it was
242    written for (ext2/3).  This is for several reasons:
243 
244    * It has no understanding of any filesystem specific optimizations.
245 
246    * It enters the filesystem repeatedly for each page that is written.
247 
248    * It depends on reiserfs_get_block() function which if implemented by reiserfs performs costly search_by_key
249    * operation for each page it is supplied with. By contrast reiserfs_file_write() feeds as much as possible at a time
250    * to reiserfs which allows for fewer tree traversals.
251 
252    * Each indirect pointer insertion takes a lot of cpu, because it involves memory moves inside of blocks.
253 
254    * Asking the block allocation code for blocks one at a time is slightly less efficient.
255 
256    All of these reasons for not using only generic file write were understood back when reiserfs was first miscoded to
257    use it, but we were in a hurry to make code freeze, and so it couldn't be revised then.  This new code should make
258    things right finally.
259 
260    Future Features: providing search_by_key with hints.
261 
262 */
263 static ssize_t reiserfs_file_write(struct file *file,	/* the file we are going to write into */
264 				   const char __user * buf,	/*  pointer to user supplied data
265 								   (in userspace) */
266 				   size_t count,	/* amount of bytes to write */
267 				   loff_t * ppos	/* pointer to position in file that we start writing at. Should be updated to
268 							 * new current position before returning. */
269 				   )
270 {
271 	struct inode *inode = file->f_path.dentry->d_inode;	// Inode of the file that we are writing to.
272 	/* To simplify coding at this time, we store
273 	   locked pages in array for now */
274 	struct reiserfs_transaction_handle th;
275 	th.t_trans_id = 0;
276 
277 	/* If a filesystem is converted from 3.5 to 3.6, we'll have v3.5 items
278 	* lying around (most of the disk, in fact). Despite the filesystem
279 	* now being a v3.6 format, the old items still can't support large
280 	* file sizes. Catch this case here, as the rest of the VFS layer is
281 	* oblivious to the different limitations between old and new items.
282 	* reiserfs_setattr catches this for truncates. This chunk is lifted
283 	* from generic_write_checks. */
284 	if (get_inode_item_key_version (inode) == KEY_FORMAT_3_5 &&
285 	    *ppos + count > MAX_NON_LFS) {
286 		if (*ppos >= MAX_NON_LFS) {
287 			return -EFBIG;
288 		}
289 		if (count > MAX_NON_LFS - (unsigned long)*ppos)
290 			count = MAX_NON_LFS - (unsigned long)*ppos;
291 	}
292 
293 	return do_sync_write(file, buf, count, ppos);
294 }
295 
296 const struct file_operations reiserfs_file_operations = {
297 	.read = do_sync_read,
298 	.write = reiserfs_file_write,
299 	.unlocked_ioctl = reiserfs_ioctl,
300 #ifdef CONFIG_COMPAT
301 	.compat_ioctl = reiserfs_compat_ioctl,
302 #endif
303 	.mmap = generic_file_mmap,
304 	.open = reiserfs_file_open,
305 	.release = reiserfs_file_release,
306 	.fsync = reiserfs_sync_file,
307 	.aio_read = generic_file_aio_read,
308 	.aio_write = generic_file_aio_write,
309 	.splice_read = generic_file_splice_read,
310 	.splice_write = generic_file_splice_write,
311 	.llseek = generic_file_llseek,
312 };
313 
314 const struct inode_operations reiserfs_file_inode_operations = {
315 	.setattr = reiserfs_setattr,
316 	.setxattr = reiserfs_setxattr,
317 	.getxattr = reiserfs_getxattr,
318 	.listxattr = reiserfs_listxattr,
319 	.removexattr = reiserfs_removexattr,
320 	.permission = reiserfs_permission,
321 	.get_acl = reiserfs_get_acl,
322 };
323