aboutsummaryrefslogtreecommitdiffstats
path: root/fs/reiserfs/file.c
diff options
context:
space:
mode:
Diffstat (limited to 'fs/reiserfs/file.c')
-rw-r--r--fs/reiserfs/file.c2564
1 files changed, 1362 insertions, 1202 deletions
diff --git a/fs/reiserfs/file.c b/fs/reiserfs/file.c
index 12e91209544e..c9f178fb494f 100644
--- a/fs/reiserfs/file.c
+++ b/fs/reiserfs/file.c
@@ -2,7 +2,6 @@
2 * Copyright 2000 by Hans Reiser, licensing governed by reiserfs/README 2 * Copyright 2000 by Hans Reiser, licensing governed by reiserfs/README
3 */ 3 */
4 4
5
6#include <linux/time.h> 5#include <linux/time.h>
7#include <linux/reiserfs_fs.h> 6#include <linux/reiserfs_fs.h>
8#include <linux/reiserfs_acl.h> 7#include <linux/reiserfs_acl.h>
@@ -31,82 +30,84 @@
31** We use reiserfs_truncate_file to pack the tail, since it already has 30** We use reiserfs_truncate_file to pack the tail, since it already has
32** all the conditions coded. 31** all the conditions coded.
33*/ 32*/
34static int reiserfs_file_release (struct inode * inode, struct file * filp) 33static int reiserfs_file_release(struct inode *inode, struct file *filp)
35{ 34{
36 35
37 struct reiserfs_transaction_handle th ; 36 struct reiserfs_transaction_handle th;
38 int err; 37 int err;
39 int jbegin_failure = 0; 38 int jbegin_failure = 0;
40 39
41 if (!S_ISREG (inode->i_mode)) 40 if (!S_ISREG(inode->i_mode))
42 BUG (); 41 BUG();
43 42
44 /* fast out for when nothing needs to be done */ 43 /* fast out for when nothing needs to be done */
45 if ((atomic_read(&inode->i_count) > 1 || 44 if ((atomic_read(&inode->i_count) > 1 ||
46 !(REISERFS_I(inode)->i_flags & i_pack_on_close_mask) || 45 !(REISERFS_I(inode)->i_flags & i_pack_on_close_mask) ||
47 !tail_has_to_be_packed(inode)) && 46 !tail_has_to_be_packed(inode)) &&
48 REISERFS_I(inode)->i_prealloc_count <= 0) { 47 REISERFS_I(inode)->i_prealloc_count <= 0) {
49 return 0; 48 return 0;
50 } 49 }
51
52 reiserfs_write_lock(inode->i_sb);
53 down (&inode->i_sem);
54 /* freeing preallocation only involves relogging blocks that
55 * are already in the current transaction. preallocation gets
56 * freed at the end of each transaction, so it is impossible for
57 * us to log any additional blocks (including quota blocks)
58 */
59 err = journal_begin(&th, inode->i_sb, 1);
60 if (err) {
61 /* uh oh, we can't allow the inode to go away while there
62 * is still preallocation blocks pending. Try to join the
63 * aborted transaction
64 */
65 jbegin_failure = err;
66 err = journal_join_abort(&th, inode->i_sb, 1);
67 50
51 reiserfs_write_lock(inode->i_sb);
52 down(&inode->i_sem);
53 /* freeing preallocation only involves relogging blocks that
54 * are already in the current transaction. preallocation gets
55 * freed at the end of each transaction, so it is impossible for
56 * us to log any additional blocks (including quota blocks)
57 */
58 err = journal_begin(&th, inode->i_sb, 1);
68 if (err) { 59 if (err) {
69 /* hmpf, our choices here aren't good. We can pin the inode 60 /* uh oh, we can't allow the inode to go away while there
70 * which will disallow unmount from every happening, we can 61 * is still preallocation blocks pending. Try to join the
71 * do nothing, which will corrupt random memory on unmount, 62 * aborted transaction
72 * or we can forcibly remove the file from the preallocation 63 */
73 * list, which will leak blocks on disk. Lets pin the inode 64 jbegin_failure = err;
74 * and let the admin know what is going on. 65 err = journal_join_abort(&th, inode->i_sb, 1);
75 */ 66
76 igrab(inode); 67 if (err) {
77 reiserfs_warning(inode->i_sb, "pinning inode %lu because the " 68 /* hmpf, our choices here aren't good. We can pin the inode
78 "preallocation can't be freed"); 69 * which will disallow unmount from every happening, we can
79 goto out; 70 * do nothing, which will corrupt random memory on unmount,
71 * or we can forcibly remove the file from the preallocation
72 * list, which will leak blocks on disk. Lets pin the inode
73 * and let the admin know what is going on.
74 */
75 igrab(inode);
76 reiserfs_warning(inode->i_sb,
77 "pinning inode %lu because the "
78 "preallocation can't be freed");
79 goto out;
80 }
80 } 81 }
81 } 82 reiserfs_update_inode_transaction(inode);
82 reiserfs_update_inode_transaction(inode) ;
83 83
84#ifdef REISERFS_PREALLOCATE 84#ifdef REISERFS_PREALLOCATE
85 reiserfs_discard_prealloc (&th, inode); 85 reiserfs_discard_prealloc(&th, inode);
86#endif 86#endif
87 err = journal_end(&th, inode->i_sb, 1); 87 err = journal_end(&th, inode->i_sb, 1);
88 88
89 /* copy back the error code from journal_begin */ 89 /* copy back the error code from journal_begin */
90 if (!err) 90 if (!err)
91 err = jbegin_failure; 91 err = jbegin_failure;
92 92
93 if (!err && atomic_read(&inode->i_count) <= 1 && 93 if (!err && atomic_read(&inode->i_count) <= 1 &&
94 (REISERFS_I(inode)->i_flags & i_pack_on_close_mask) && 94 (REISERFS_I(inode)->i_flags & i_pack_on_close_mask) &&
95 tail_has_to_be_packed (inode)) { 95 tail_has_to_be_packed(inode)) {
96 /* if regular file is released by last holder and it has been 96 /* if regular file is released by last holder and it has been
97 appended (we append by unformatted node only) or its direct 97 appended (we append by unformatted node only) or its direct
98 item(s) had to be converted, then it may have to be 98 item(s) had to be converted, then it may have to be
99 indirect2direct converted */ 99 indirect2direct converted */
100 err = reiserfs_truncate_file(inode, 0) ; 100 err = reiserfs_truncate_file(inode, 0);
101 } 101 }
102out: 102 out:
103 up (&inode->i_sem); 103 up(&inode->i_sem);
104 reiserfs_write_unlock(inode->i_sb); 104 reiserfs_write_unlock(inode->i_sb);
105 return err; 105 return err;
106} 106}
107 107
108static void reiserfs_vfs_truncate_file(struct inode *inode) { 108static void reiserfs_vfs_truncate_file(struct inode *inode)
109 reiserfs_truncate_file(inode, 1) ; 109{
110 reiserfs_truncate_file(inode, 1);
110} 111}
111 112
112/* Sync a reiserfs file. */ 113/* Sync a reiserfs file. */
@@ -116,26 +117,24 @@ static void reiserfs_vfs_truncate_file(struct inode *inode) {
116 * be removed... 117 * be removed...
117 */ 118 */
118 119
119static int reiserfs_sync_file( 120static int reiserfs_sync_file(struct file *p_s_filp,
120 struct file * p_s_filp, 121 struct dentry *p_s_dentry, int datasync)
121 struct dentry * p_s_dentry, 122{
122 int datasync 123 struct inode *p_s_inode = p_s_dentry->d_inode;
123 ) { 124 int n_err;
124 struct inode * p_s_inode = p_s_dentry->d_inode; 125 int barrier_done;
125 int n_err; 126
126 int barrier_done; 127 if (!S_ISREG(p_s_inode->i_mode))
127 128 BUG();
128 if (!S_ISREG(p_s_inode->i_mode)) 129 n_err = sync_mapping_buffers(p_s_inode->i_mapping);
129 BUG (); 130 reiserfs_write_lock(p_s_inode->i_sb);
130 n_err = sync_mapping_buffers(p_s_inode->i_mapping) ; 131 barrier_done = reiserfs_commit_for_inode(p_s_inode);
131 reiserfs_write_lock(p_s_inode->i_sb); 132 reiserfs_write_unlock(p_s_inode->i_sb);
132 barrier_done = reiserfs_commit_for_inode(p_s_inode); 133 if (barrier_done != 1)
133 reiserfs_write_unlock(p_s_inode->i_sb); 134 blkdev_issue_flush(p_s_inode->i_sb->s_bdev, NULL);
134 if (barrier_done != 1) 135 if (barrier_done < 0)
135 blkdev_issue_flush(p_s_inode->i_sb->s_bdev, NULL); 136 return barrier_done;
136 if (barrier_done < 0) 137 return (n_err < 0) ? -EIO : 0;
137 return barrier_done;
138 return ( n_err < 0 ) ? -EIO : 0;
139} 138}
140 139
141/* I really do not want to play with memory shortage right now, so 140/* I really do not want to play with memory shortage right now, so
@@ -147,700 +146,797 @@ static int reiserfs_sync_file(
147/* Allocates blocks for a file to fulfil write request. 146/* Allocates blocks for a file to fulfil write request.
148 Maps all unmapped but prepared pages from the list. 147 Maps all unmapped but prepared pages from the list.
149 Updates metadata with newly allocated blocknumbers as needed */ 148 Updates metadata with newly allocated blocknumbers as needed */
150static int reiserfs_allocate_blocks_for_region( 149static int reiserfs_allocate_blocks_for_region(struct reiserfs_transaction_handle *th, struct inode *inode, /* Inode we work with */
151 struct reiserfs_transaction_handle *th, 150 loff_t pos, /* Writing position */
152 struct inode *inode, /* Inode we work with */ 151 int num_pages, /* number of pages write going
153 loff_t pos, /* Writing position */ 152 to touch */
154 int num_pages, /* number of pages write going 153 int write_bytes, /* amount of bytes to write */
155 to touch */ 154 struct page **prepared_pages, /* array of
156 int write_bytes, /* amount of bytes to write */ 155 prepared pages
157 struct page **prepared_pages, /* array of 156 */
158 prepared pages 157 int blocks_to_allocate /* Amount of blocks we
159 */ 158 need to allocate to
160 int blocks_to_allocate /* Amount of blocks we 159 fit the data into file
161 need to allocate to 160 */
162 fit the data into file 161 )
163 */
164 )
165{ 162{
166 struct cpu_key key; // cpu key of item that we are going to deal with 163 struct cpu_key key; // cpu key of item that we are going to deal with
167 struct item_head *ih; // pointer to item head that we are going to deal with 164 struct item_head *ih; // pointer to item head that we are going to deal with
168 struct buffer_head *bh; // Buffer head that contains items that we are going to deal with 165 struct buffer_head *bh; // Buffer head that contains items that we are going to deal with
169 __le32 * item; // pointer to item we are going to deal with 166 __le32 *item; // pointer to item we are going to deal with
170 INITIALIZE_PATH(path); // path to item, that we are going to deal with. 167 INITIALIZE_PATH(path); // path to item, that we are going to deal with.
171 b_blocknr_t *allocated_blocks; // Pointer to a place where allocated blocknumbers would be stored. 168 b_blocknr_t *allocated_blocks; // Pointer to a place where allocated blocknumbers would be stored.
172 reiserfs_blocknr_hint_t hint; // hint structure for block allocator. 169 reiserfs_blocknr_hint_t hint; // hint structure for block allocator.
173 size_t res; // return value of various functions that we call. 170 size_t res; // return value of various functions that we call.
174 int curr_block; // current block used to keep track of unmapped blocks. 171 int curr_block; // current block used to keep track of unmapped blocks.
175 int i; // loop counter 172 int i; // loop counter
176 int itempos; // position in item 173 int itempos; // position in item
177 unsigned int from = (pos & (PAGE_CACHE_SIZE - 1)); // writing position in 174 unsigned int from = (pos & (PAGE_CACHE_SIZE - 1)); // writing position in
178 // first page 175 // first page
179 unsigned int to = ((pos + write_bytes - 1) & (PAGE_CACHE_SIZE - 1)) + 1; /* last modified byte offset in last page */ 176 unsigned int to = ((pos + write_bytes - 1) & (PAGE_CACHE_SIZE - 1)) + 1; /* last modified byte offset in last page */
180 __u64 hole_size ; // amount of blocks for a file hole, if it needed to be created. 177 __u64 hole_size; // amount of blocks for a file hole, if it needed to be created.
181 int modifying_this_item = 0; // Flag for items traversal code to keep track 178 int modifying_this_item = 0; // Flag for items traversal code to keep track
182 // of the fact that we already prepared 179 // of the fact that we already prepared
183 // current block for journal 180 // current block for journal
184 int will_prealloc = 0; 181 int will_prealloc = 0;
185 RFALSE(!blocks_to_allocate, "green-9004: tried to allocate zero blocks?"); 182 RFALSE(!blocks_to_allocate,
186 183 "green-9004: tried to allocate zero blocks?");
187 /* only preallocate if this is a small write */ 184
188 if (REISERFS_I(inode)->i_prealloc_count || 185 /* only preallocate if this is a small write */
189 (!(write_bytes & (inode->i_sb->s_blocksize -1)) && 186 if (REISERFS_I(inode)->i_prealloc_count ||
190 blocks_to_allocate < 187 (!(write_bytes & (inode->i_sb->s_blocksize - 1)) &&
191 REISERFS_SB(inode->i_sb)->s_alloc_options.preallocsize)) 188 blocks_to_allocate <
192 will_prealloc = REISERFS_SB(inode->i_sb)->s_alloc_options.preallocsize; 189 REISERFS_SB(inode->i_sb)->s_alloc_options.preallocsize))
193 190 will_prealloc =
194 allocated_blocks = kmalloc((blocks_to_allocate + will_prealloc) * 191 REISERFS_SB(inode->i_sb)->s_alloc_options.preallocsize;
195 sizeof(b_blocknr_t), GFP_NOFS); 192
196 193 allocated_blocks = kmalloc((blocks_to_allocate + will_prealloc) *
197 /* First we compose a key to point at the writing position, we want to do 194 sizeof(b_blocknr_t), GFP_NOFS);
198 that outside of any locking region. */ 195
199 make_cpu_key (&key, inode, pos+1, TYPE_ANY, 3/*key length*/); 196 /* First we compose a key to point at the writing position, we want to do
200 197 that outside of any locking region. */
201 /* If we came here, it means we absolutely need to open a transaction, 198 make_cpu_key(&key, inode, pos + 1, TYPE_ANY, 3 /*key length */ );
202 since we need to allocate some blocks */ 199
203 reiserfs_write_lock(inode->i_sb); // Journaling stuff and we need that. 200 /* If we came here, it means we absolutely need to open a transaction,
204 res = journal_begin(th, inode->i_sb, JOURNAL_PER_BALANCE_CNT * 3 + 1 + 2 * REISERFS_QUOTA_TRANS_BLOCKS(inode->i_sb)); // Wish I know if this number enough 201 since we need to allocate some blocks */
205 if (res) 202 reiserfs_write_lock(inode->i_sb); // Journaling stuff and we need that.
206 goto error_exit; 203 res = journal_begin(th, inode->i_sb, JOURNAL_PER_BALANCE_CNT * 3 + 1 + 2 * REISERFS_QUOTA_TRANS_BLOCKS(inode->i_sb)); // Wish I know if this number enough
207 reiserfs_update_inode_transaction(inode) ; 204 if (res)
208
209 /* Look for the in-tree position of our write, need path for block allocator */
210 res = search_for_position_by_key(inode->i_sb, &key, &path);
211 if ( res == IO_ERROR ) {
212 res = -EIO;
213 goto error_exit;
214 }
215
216 /* Allocate blocks */
217 /* First fill in "hint" structure for block allocator */
218 hint.th = th; // transaction handle.
219 hint.path = &path; // Path, so that block allocator can determine packing locality or whatever it needs to determine.
220 hint.inode = inode; // Inode is needed by block allocator too.
221 hint.search_start = 0; // We have no hint on where to search free blocks for block allocator.
222 hint.key = key.on_disk_key; // on disk key of file.
223 hint.block = inode->i_blocks>>(inode->i_sb->s_blocksize_bits-9); // Number of disk blocks this file occupies already.
224 hint.formatted_node = 0; // We are allocating blocks for unformatted node.
225 hint.preallocate = will_prealloc;
226
227 /* Call block allocator to allocate blocks */
228 res = reiserfs_allocate_blocknrs(&hint, allocated_blocks, blocks_to_allocate, blocks_to_allocate);
229 if ( res != CARRY_ON ) {
230 if ( res == NO_DISK_SPACE ) {
231 /* We flush the transaction in case of no space. This way some
232 blocks might become free */
233 SB_JOURNAL(inode->i_sb)->j_must_wait = 1;
234 res = restart_transaction(th, inode, &path);
235 if (res)
236 goto error_exit;
237
238 /* We might have scheduled, so search again */
239 res = search_for_position_by_key(inode->i_sb, &key, &path);
240 if ( res == IO_ERROR ) {
241 res = -EIO;
242 goto error_exit; 205 goto error_exit;
243 } 206 reiserfs_update_inode_transaction(inode);
244 207
245 /* update changed info for hint structure. */ 208 /* Look for the in-tree position of our write, need path for block allocator */
246 res = reiserfs_allocate_blocknrs(&hint, allocated_blocks, blocks_to_allocate, blocks_to_allocate); 209 res = search_for_position_by_key(inode->i_sb, &key, &path);
247 if ( res != CARRY_ON ) { 210 if (res == IO_ERROR) {
248 res = -ENOSPC; 211 res = -EIO;
249 pathrelse(&path);
250 goto error_exit; 212 goto error_exit;
251 }
252 } else {
253 res = -ENOSPC;
254 pathrelse(&path);
255 goto error_exit;
256 } 213 }
257 }
258 214
259#ifdef __BIG_ENDIAN 215 /* Allocate blocks */
260 // Too bad, I have not found any way to convert a given region from 216 /* First fill in "hint" structure for block allocator */
261 // cpu format to little endian format 217 hint.th = th; // transaction handle.
262 { 218 hint.path = &path; // Path, so that block allocator can determine packing locality or whatever it needs to determine.
263 int i; 219 hint.inode = inode; // Inode is needed by block allocator too.
264 for ( i = 0; i < blocks_to_allocate ; i++) 220 hint.search_start = 0; // We have no hint on where to search free blocks for block allocator.
265 allocated_blocks[i]=cpu_to_le32(allocated_blocks[i]); 221 hint.key = key.on_disk_key; // on disk key of file.
266 } 222 hint.block = inode->i_blocks >> (inode->i_sb->s_blocksize_bits - 9); // Number of disk blocks this file occupies already.
267#endif 223 hint.formatted_node = 0; // We are allocating blocks for unformatted node.
268 224 hint.preallocate = will_prealloc;
269 /* Blocks allocating well might have scheduled and tree might have changed, 225
270 let's search the tree again */ 226 /* Call block allocator to allocate blocks */
271 /* find where in the tree our write should go */ 227 res =
272 res = search_for_position_by_key(inode->i_sb, &key, &path); 228 reiserfs_allocate_blocknrs(&hint, allocated_blocks,
273 if ( res == IO_ERROR ) { 229 blocks_to_allocate, blocks_to_allocate);
274 res = -EIO; 230 if (res != CARRY_ON) {
275 goto error_exit_free_blocks; 231 if (res == NO_DISK_SPACE) {
276 } 232 /* We flush the transaction in case of no space. This way some
277 233 blocks might become free */
278 bh = get_last_bh( &path ); // Get a bufferhead for last element in path. 234 SB_JOURNAL(inode->i_sb)->j_must_wait = 1;
279 ih = get_ih( &path ); // Get a pointer to last item head in path. 235 res = restart_transaction(th, inode, &path);
280 item = get_item( &path ); // Get a pointer to last item in path 236 if (res)
281 237 goto error_exit;
282 /* Let's see what we have found */ 238
283 if ( res != POSITION_FOUND ) { /* position not found, this means that we 239 /* We might have scheduled, so search again */
284 might need to append file with holes 240 res =
285 first */ 241 search_for_position_by_key(inode->i_sb, &key,
286 // Since we are writing past the file's end, we need to find out if 242 &path);
287 // there is a hole that needs to be inserted before our writing 243 if (res == IO_ERROR) {
288 // position, and how many blocks it is going to cover (we need to 244 res = -EIO;
289 // populate pointers to file blocks representing the hole with zeros) 245 goto error_exit;
246 }
290 247
248 /* update changed info for hint structure. */
249 res =
250 reiserfs_allocate_blocknrs(&hint, allocated_blocks,
251 blocks_to_allocate,
252 blocks_to_allocate);
253 if (res != CARRY_ON) {
254 res = -ENOSPC;
255 pathrelse(&path);
256 goto error_exit;
257 }
258 } else {
259 res = -ENOSPC;
260 pathrelse(&path);
261 goto error_exit;
262 }
263 }
264#ifdef __BIG_ENDIAN
265 // Too bad, I have not found any way to convert a given region from
266 // cpu format to little endian format
291 { 267 {
292 int item_offset = 1; 268 int i;
293 /* 269 for (i = 0; i < blocks_to_allocate; i++)
294 * if ih is stat data, its offset is 0 and we don't want to 270 allocated_blocks[i] = cpu_to_le32(allocated_blocks[i]);
295 * add 1 to pos in the hole_size calculation
296 */
297 if (is_statdata_le_ih(ih))
298 item_offset = 0;
299 hole_size = (pos + item_offset -
300 (le_key_k_offset( get_inode_item_key_version(inode),
301 &(ih->ih_key)) +
302 op_bytes_number(ih, inode->i_sb->s_blocksize))) >>
303 inode->i_sb->s_blocksize_bits;
304 } 271 }
272#endif
305 273
306 if ( hole_size > 0 ) { 274 /* Blocks allocating well might have scheduled and tree might have changed,
307 int to_paste = min_t(__u64, hole_size, MAX_ITEM_LEN(inode->i_sb->s_blocksize)/UNFM_P_SIZE ); // How much data to insert first time. 275 let's search the tree again */
308 /* area filled with zeroes, to supply as list of zero blocknumbers 276 /* find where in the tree our write should go */
309 We allocate it outside of loop just in case loop would spin for 277 res = search_for_position_by_key(inode->i_sb, &key, &path);
310 several iterations. */ 278 if (res == IO_ERROR) {
311 char *zeros = kmalloc(to_paste*UNFM_P_SIZE, GFP_ATOMIC); // We cannot insert more than MAX_ITEM_LEN bytes anyway. 279 res = -EIO;
312 if ( !zeros ) {
313 res = -ENOMEM;
314 goto error_exit_free_blocks; 280 goto error_exit_free_blocks;
315 } 281 }
316 memset ( zeros, 0, to_paste*UNFM_P_SIZE); 282
317 do { 283 bh = get_last_bh(&path); // Get a bufferhead for last element in path.
318 to_paste = min_t(__u64, hole_size, MAX_ITEM_LEN(inode->i_sb->s_blocksize)/UNFM_P_SIZE ); 284 ih = get_ih(&path); // Get a pointer to last item head in path.
319 if ( is_indirect_le_ih(ih) ) { 285 item = get_item(&path); // Get a pointer to last item in path
320 /* Ok, there is existing indirect item already. Need to append it */ 286
321 /* Calculate position past inserted item */ 287 /* Let's see what we have found */
322 make_cpu_key( &key, inode, le_key_k_offset( get_inode_item_key_version(inode), &(ih->ih_key)) + op_bytes_number(ih, inode->i_sb->s_blocksize), TYPE_INDIRECT, 3); 288 if (res != POSITION_FOUND) { /* position not found, this means that we
323 res = reiserfs_paste_into_item( th, &path, &key, inode, (char *)zeros, UNFM_P_SIZE*to_paste); 289 might need to append file with holes
324 if ( res ) { 290 first */
325 kfree(zeros); 291 // Since we are writing past the file's end, we need to find out if
326 goto error_exit_free_blocks; 292 // there is a hole that needs to be inserted before our writing
327 } 293 // position, and how many blocks it is going to cover (we need to
328 } else if ( is_statdata_le_ih(ih) ) { 294 // populate pointers to file blocks representing the hole with zeros)
329 /* No existing item, create it */ 295
330 /* item head for new item */ 296 {
331 struct item_head ins_ih; 297 int item_offset = 1;
332 298 /*
333 /* create a key for our new item */ 299 * if ih is stat data, its offset is 0 and we don't want to
334 make_cpu_key( &key, inode, 1, TYPE_INDIRECT, 3); 300 * add 1 to pos in the hole_size calculation
335 301 */
336 /* Create new item head for our new item */ 302 if (is_statdata_le_ih(ih))
337 make_le_item_head (&ins_ih, &key, key.version, 1, 303 item_offset = 0;
338 TYPE_INDIRECT, to_paste*UNFM_P_SIZE, 304 hole_size = (pos + item_offset -
339 0 /* free space */); 305 (le_key_k_offset
340 306 (get_inode_item_key_version(inode),
341 /* Find where such item should live in the tree */ 307 &(ih->ih_key)) + op_bytes_number(ih,
342 res = search_item (inode->i_sb, &key, &path); 308 inode->
343 if ( res != ITEM_NOT_FOUND ) { 309 i_sb->
344 /* item should not exist, otherwise we have error */ 310 s_blocksize)))
345 if ( res != -ENOSPC ) { 311 >> inode->i_sb->s_blocksize_bits;
346 reiserfs_warning (inode->i_sb, 312 }
347 "green-9008: search_by_key (%K) returned %d", 313
348 &key, res); 314 if (hole_size > 0) {
315 int to_paste = min_t(__u64, hole_size, MAX_ITEM_LEN(inode->i_sb->s_blocksize) / UNFM_P_SIZE); // How much data to insert first time.
316 /* area filled with zeroes, to supply as list of zero blocknumbers
317 We allocate it outside of loop just in case loop would spin for
318 several iterations. */
319 char *zeros = kmalloc(to_paste * UNFM_P_SIZE, GFP_ATOMIC); // We cannot insert more than MAX_ITEM_LEN bytes anyway.
320 if (!zeros) {
321 res = -ENOMEM;
322 goto error_exit_free_blocks;
349 } 323 }
350 res = -EIO; 324 memset(zeros, 0, to_paste * UNFM_P_SIZE);
351 kfree(zeros); 325 do {
352 goto error_exit_free_blocks; 326 to_paste =
353 } 327 min_t(__u64, hole_size,
354 res = reiserfs_insert_item( th, &path, &key, &ins_ih, inode, (char *)zeros); 328 MAX_ITEM_LEN(inode->i_sb->
355 } else { 329 s_blocksize) /
356 reiserfs_panic(inode->i_sb, "green-9011: Unexpected key type %K\n", &key); 330 UNFM_P_SIZE);
331 if (is_indirect_le_ih(ih)) {
332 /* Ok, there is existing indirect item already. Need to append it */
333 /* Calculate position past inserted item */
334 make_cpu_key(&key, inode,
335 le_key_k_offset
336 (get_inode_item_key_version
337 (inode),
338 &(ih->ih_key)) +
339 op_bytes_number(ih,
340 inode->
341 i_sb->
342 s_blocksize),
343 TYPE_INDIRECT, 3);
344 res =
345 reiserfs_paste_into_item(th, &path,
346 &key,
347 inode,
348 (char *)
349 zeros,
350 UNFM_P_SIZE
351 *
352 to_paste);
353 if (res) {
354 kfree(zeros);
355 goto error_exit_free_blocks;
356 }
357 } else if (is_statdata_le_ih(ih)) {
358 /* No existing item, create it */
359 /* item head for new item */
360 struct item_head ins_ih;
361
362 /* create a key for our new item */
363 make_cpu_key(&key, inode, 1,
364 TYPE_INDIRECT, 3);
365
366 /* Create new item head for our new item */
367 make_le_item_head(&ins_ih, &key,
368 key.version, 1,
369 TYPE_INDIRECT,
370 to_paste *
371 UNFM_P_SIZE,
372 0 /* free space */ );
373
374 /* Find where such item should live in the tree */
375 res =
376 search_item(inode->i_sb, &key,
377 &path);
378 if (res != ITEM_NOT_FOUND) {
379 /* item should not exist, otherwise we have error */
380 if (res != -ENOSPC) {
381 reiserfs_warning(inode->
382 i_sb,
383 "green-9008: search_by_key (%K) returned %d",
384 &key,
385 res);
386 }
387 res = -EIO;
388 kfree(zeros);
389 goto error_exit_free_blocks;
390 }
391 res =
392 reiserfs_insert_item(th, &path,
393 &key, &ins_ih,
394 inode,
395 (char *)zeros);
396 } else {
397 reiserfs_panic(inode->i_sb,
398 "green-9011: Unexpected key type %K\n",
399 &key);
400 }
401 if (res) {
402 kfree(zeros);
403 goto error_exit_free_blocks;
404 }
405 /* Now we want to check if transaction is too full, and if it is
406 we restart it. This will also free the path. */
407 if (journal_transaction_should_end
408 (th, th->t_blocks_allocated)) {
409 res =
410 restart_transaction(th, inode,
411 &path);
412 if (res) {
413 pathrelse(&path);
414 kfree(zeros);
415 goto error_exit;
416 }
417 }
418
419 /* Well, need to recalculate path and stuff */
420 set_cpu_key_k_offset(&key,
421 cpu_key_k_offset(&key) +
422 (to_paste << inode->
423 i_blkbits));
424 res =
425 search_for_position_by_key(inode->i_sb,
426 &key, &path);
427 if (res == IO_ERROR) {
428 res = -EIO;
429 kfree(zeros);
430 goto error_exit_free_blocks;
431 }
432 bh = get_last_bh(&path);
433 ih = get_ih(&path);
434 item = get_item(&path);
435 hole_size -= to_paste;
436 } while (hole_size);
437 kfree(zeros);
357 } 438 }
358 if ( res ) { 439 }
359 kfree(zeros); 440 // Go through existing indirect items first
360 goto error_exit_free_blocks; 441 // replace all zeroes with blocknumbers from list
442 // Note that if no corresponding item was found, by previous search,
443 // it means there are no existing in-tree representation for file area
444 // we are going to overwrite, so there is nothing to scan through for holes.
445 for (curr_block = 0, itempos = path.pos_in_item;
446 curr_block < blocks_to_allocate && res == POSITION_FOUND;) {
447 retry:
448
449 if (itempos >= ih_item_len(ih) / UNFM_P_SIZE) {
450 /* We run out of data in this indirect item, let's look for another
451 one. */
452 /* First if we are already modifying current item, log it */
453 if (modifying_this_item) {
454 journal_mark_dirty(th, inode->i_sb, bh);
455 modifying_this_item = 0;
456 }
457 /* Then set the key to look for a new indirect item (offset of old
458 item is added to old item length */
459 set_cpu_key_k_offset(&key,
460 le_key_k_offset
461 (get_inode_item_key_version(inode),
462 &(ih->ih_key)) +
463 op_bytes_number(ih,
464 inode->i_sb->
465 s_blocksize));
466 /* Search ofor position of new key in the tree. */
467 res =
468 search_for_position_by_key(inode->i_sb, &key,
469 &path);
470 if (res == IO_ERROR) {
471 res = -EIO;
472 goto error_exit_free_blocks;
473 }
474 bh = get_last_bh(&path);
475 ih = get_ih(&path);
476 item = get_item(&path);
477 itempos = path.pos_in_item;
478 continue; // loop to check all kinds of conditions and so on.
361 } 479 }
362 /* Now we want to check if transaction is too full, and if it is 480 /* Ok, we have correct position in item now, so let's see if it is
363 we restart it. This will also free the path. */ 481 representing file hole (blocknumber is zero) and fill it if needed */
364 if (journal_transaction_should_end(th, th->t_blocks_allocated)) { 482 if (!item[itempos]) {
365 res = restart_transaction(th, inode, &path); 483 /* Ok, a hole. Now we need to check if we already prepared this
366 if (res) { 484 block to be journaled */
367 pathrelse (&path); 485 while (!modifying_this_item) { // loop until succeed
368 kfree(zeros); 486 /* Well, this item is not journaled yet, so we must prepare
369 goto error_exit; 487 it for journal first, before we can change it */
370 } 488 struct item_head tmp_ih; // We copy item head of found item,
371 } 489 // here to detect if fs changed under
372 490 // us while we were preparing for
373 /* Well, need to recalculate path and stuff */ 491 // journal.
374 set_cpu_key_k_offset( &key, cpu_key_k_offset(&key) + (to_paste << inode->i_blkbits)); 492 int fs_gen; // We store fs generation here to find if someone
375 res = search_for_position_by_key(inode->i_sb, &key, &path); 493 // changes fs under our feet
376 if ( res == IO_ERROR ) { 494
377 res = -EIO; 495 copy_item_head(&tmp_ih, ih); // Remember itemhead
378 kfree(zeros); 496 fs_gen = get_generation(inode->i_sb); // remember fs generation
379 goto error_exit_free_blocks; 497 reiserfs_prepare_for_journal(inode->i_sb, bh, 1); // Prepare a buffer within which indirect item is stored for changing.
498 if (fs_changed(fs_gen, inode->i_sb)
499 && item_moved(&tmp_ih, &path)) {
500 // Sigh, fs was changed under us, we need to look for new
501 // location of item we are working with
502
503 /* unmark prepaerd area as journaled and search for it's
504 new position */
505 reiserfs_restore_prepared_buffer(inode->
506 i_sb,
507 bh);
508 res =
509 search_for_position_by_key(inode->
510 i_sb,
511 &key,
512 &path);
513 if (res == IO_ERROR) {
514 res = -EIO;
515 goto error_exit_free_blocks;
516 }
517 bh = get_last_bh(&path);
518 ih = get_ih(&path);
519 item = get_item(&path);
520 itempos = path.pos_in_item;
521 goto retry;
522 }
523 modifying_this_item = 1;
524 }
525 item[itempos] = allocated_blocks[curr_block]; // Assign new block
526 curr_block++;
380 } 527 }
381 bh=get_last_bh(&path); 528 itempos++;
382 ih=get_ih(&path);
383 item = get_item(&path);
384 hole_size -= to_paste;
385 } while ( hole_size );
386 kfree(zeros);
387 } 529 }
388 } 530
389 531 if (modifying_this_item) { // We need to log last-accessed block, if it
390 // Go through existing indirect items first 532 // was modified, but not logged yet.
391 // replace all zeroes with blocknumbers from list 533 journal_mark_dirty(th, inode->i_sb, bh);
392 // Note that if no corresponding item was found, by previous search,
393 // it means there are no existing in-tree representation for file area
394 // we are going to overwrite, so there is nothing to scan through for holes.
395 for ( curr_block = 0, itempos = path.pos_in_item ; curr_block < blocks_to_allocate && res == POSITION_FOUND ; ) {
396retry:
397
398 if ( itempos >= ih_item_len(ih)/UNFM_P_SIZE ) {
399 /* We run out of data in this indirect item, let's look for another
400 one. */
401 /* First if we are already modifying current item, log it */
402 if ( modifying_this_item ) {
403 journal_mark_dirty (th, inode->i_sb, bh);
404 modifying_this_item = 0;
405 }
406 /* Then set the key to look for a new indirect item (offset of old
407 item is added to old item length */
408 set_cpu_key_k_offset( &key, le_key_k_offset( get_inode_item_key_version(inode), &(ih->ih_key)) + op_bytes_number(ih, inode->i_sb->s_blocksize));
409 /* Search ofor position of new key in the tree. */
410 res = search_for_position_by_key(inode->i_sb, &key, &path);
411 if ( res == IO_ERROR) {
412 res = -EIO;
413 goto error_exit_free_blocks;
414 }
415 bh=get_last_bh(&path);
416 ih=get_ih(&path);
417 item = get_item(&path);
418 itempos = path.pos_in_item;
419 continue; // loop to check all kinds of conditions and so on.
420 } 534 }
421 /* Ok, we have correct position in item now, so let's see if it is 535
422 representing file hole (blocknumber is zero) and fill it if needed */ 536 if (curr_block < blocks_to_allocate) {
423 if ( !item[itempos] ) { 537 // Oh, well need to append to indirect item, or to create indirect item
424 /* Ok, a hole. Now we need to check if we already prepared this 538 // if there weren't any
425 block to be journaled */ 539 if (is_indirect_le_ih(ih)) {
426 while ( !modifying_this_item ) { // loop until succeed 540 // Existing indirect item - append. First calculate key for append
427 /* Well, this item is not journaled yet, so we must prepare 541 // position. We do not need to recalculate path as it should
428 it for journal first, before we can change it */ 542 // already point to correct place.
429 struct item_head tmp_ih; // We copy item head of found item, 543 make_cpu_key(&key, inode,
430 // here to detect if fs changed under 544 le_key_k_offset(get_inode_item_key_version
431 // us while we were preparing for 545 (inode),
432 // journal. 546 &(ih->ih_key)) +
433 int fs_gen; // We store fs generation here to find if someone 547 op_bytes_number(ih,
434 // changes fs under our feet 548 inode->i_sb->s_blocksize),
435 549 TYPE_INDIRECT, 3);
436 copy_item_head (&tmp_ih, ih); // Remember itemhead 550 res =
437 fs_gen = get_generation (inode->i_sb); // remember fs generation 551 reiserfs_paste_into_item(th, &path, &key, inode,
438 reiserfs_prepare_for_journal(inode->i_sb, bh, 1); // Prepare a buffer within which indirect item is stored for changing. 552 (char *)(allocated_blocks +
439 if (fs_changed (fs_gen, inode->i_sb) && item_moved (&tmp_ih, &path)) { 553 curr_block),
440 // Sigh, fs was changed under us, we need to look for new 554 UNFM_P_SIZE *
441 // location of item we are working with 555 (blocks_to_allocate -
442 556 curr_block));
443 /* unmark prepaerd area as journaled and search for it's 557 if (res) {
444 new position */ 558 goto error_exit_free_blocks;
445 reiserfs_restore_prepared_buffer(inode->i_sb, bh); 559 }
446 res = search_for_position_by_key(inode->i_sb, &key, &path); 560 } else if (is_statdata_le_ih(ih)) {
447 if ( res == IO_ERROR) { 561 // Last found item was statdata. That means we need to create indirect item.
448 res = -EIO; 562 struct item_head ins_ih; /* itemhead for new item */
449 goto error_exit_free_blocks; 563
450 } 564 /* create a key for our new item */
451 bh=get_last_bh(&path); 565 make_cpu_key(&key, inode, 1, TYPE_INDIRECT, 3); // Position one,
452 ih=get_ih(&path); 566 // because that's
453 item = get_item(&path); 567 // where first
454 itempos = path.pos_in_item; 568 // indirect item
455 goto retry; 569 // begins
570 /* Create new item head for our new item */
571 make_le_item_head(&ins_ih, &key, key.version, 1,
572 TYPE_INDIRECT,
573 (blocks_to_allocate -
574 curr_block) * UNFM_P_SIZE,
575 0 /* free space */ );
576 /* Find where such item should live in the tree */
577 res = search_item(inode->i_sb, &key, &path);
578 if (res != ITEM_NOT_FOUND) {
579 /* Well, if we have found such item already, or some error
580 occured, we need to warn user and return error */
581 if (res != -ENOSPC) {
582 reiserfs_warning(inode->i_sb,
583 "green-9009: search_by_key (%K) "
584 "returned %d", &key,
585 res);
586 }
587 res = -EIO;
588 goto error_exit_free_blocks;
589 }
590 /* Insert item into the tree with the data as its body */
591 res =
592 reiserfs_insert_item(th, &path, &key, &ins_ih,
593 inode,
594 (char *)(allocated_blocks +
595 curr_block));
596 } else {
597 reiserfs_panic(inode->i_sb,
598 "green-9010: unexpected item type for key %K\n",
599 &key);
456 } 600 }
457 modifying_this_item = 1;
458 }
459 item[itempos] = allocated_blocks[curr_block]; // Assign new block
460 curr_block++;
461 } 601 }
462 itempos++; 602 // the caller is responsible for closing the transaction
463 } 603 // unless we return an error, they are also responsible for logging
464 604 // the inode.
465 if ( modifying_this_item ) { // We need to log last-accessed block, if it 605 //
466 // was modified, but not logged yet. 606 pathrelse(&path);
467 journal_mark_dirty (th, inode->i_sb, bh); 607 /*
468 } 608 * cleanup prellocation from previous writes
469 609 * if this is a partial block write
470 if ( curr_block < blocks_to_allocate ) { 610 */
471 // Oh, well need to append to indirect item, or to create indirect item 611 if (write_bytes & (inode->i_sb->s_blocksize - 1))
472 // if there weren't any 612 reiserfs_discard_prealloc(th, inode);
473 if ( is_indirect_le_ih(ih) ) { 613 reiserfs_write_unlock(inode->i_sb);
474 // Existing indirect item - append. First calculate key for append 614
475 // position. We do not need to recalculate path as it should 615 // go through all the pages/buffers and map the buffers to newly allocated
476 // already point to correct place. 616 // blocks (so that system knows where to write these pages later).
477 make_cpu_key( &key, inode, le_key_k_offset( get_inode_item_key_version(inode), &(ih->ih_key)) + op_bytes_number(ih, inode->i_sb->s_blocksize), TYPE_INDIRECT, 3); 617 curr_block = 0;
478 res = reiserfs_paste_into_item( th, &path, &key, inode, (char *)(allocated_blocks+curr_block), UNFM_P_SIZE*(blocks_to_allocate-curr_block)); 618 for (i = 0; i < num_pages; i++) {
479 if ( res ) { 619 struct page *page = prepared_pages[i]; //current page
480 goto error_exit_free_blocks; 620 struct buffer_head *head = page_buffers(page); // first buffer for a page
481 } 621 int block_start, block_end; // in-page offsets for buffers.
482 } else if (is_statdata_le_ih(ih) ) { 622
483 // Last found item was statdata. That means we need to create indirect item. 623 if (!page_buffers(page))
484 struct item_head ins_ih; /* itemhead for new item */ 624 reiserfs_panic(inode->i_sb,
485 625 "green-9005: No buffers for prepared page???");
486 /* create a key for our new item */ 626
487 make_cpu_key( &key, inode, 1, TYPE_INDIRECT, 3); // Position one, 627 /* For each buffer in page */
488 // because that's 628 for (bh = head, block_start = 0; bh != head || !block_start;
489 // where first 629 block_start = block_end, bh = bh->b_this_page) {
490 // indirect item 630 if (!bh)
491 // begins 631 reiserfs_panic(inode->i_sb,
492 /* Create new item head for our new item */ 632 "green-9006: Allocated but absent buffer for a page?");
493 make_le_item_head (&ins_ih, &key, key.version, 1, TYPE_INDIRECT, 633 block_end = block_start + inode->i_sb->s_blocksize;
494 (blocks_to_allocate-curr_block)*UNFM_P_SIZE, 634 if (i == 0 && block_end <= from)
495 0 /* free space */); 635 /* if this buffer is before requested data to map, skip it */
496 /* Find where such item should live in the tree */ 636 continue;
497 res = search_item (inode->i_sb, &key, &path); 637 if (i == num_pages - 1 && block_start >= to)
498 if ( res != ITEM_NOT_FOUND ) { 638 /* If this buffer is after requested data to map, abort
499 /* Well, if we have found such item already, or some error 639 processing of current page */
500 occured, we need to warn user and return error */ 640 break;
501 if ( res != -ENOSPC ) { 641
502 reiserfs_warning (inode->i_sb, 642 if (!buffer_mapped(bh)) { // Ok, unmapped buffer, need to map it
503 "green-9009: search_by_key (%K) " 643 map_bh(bh, inode->i_sb,
504 "returned %d", &key, res); 644 le32_to_cpu(allocated_blocks
645 [curr_block]));
646 curr_block++;
647 set_buffer_new(bh);
648 }
505 } 649 }
506 res = -EIO;
507 goto error_exit_free_blocks;
508 }
509 /* Insert item into the tree with the data as its body */
510 res = reiserfs_insert_item( th, &path, &key, &ins_ih, inode, (char *)(allocated_blocks+curr_block));
511 } else {
512 reiserfs_panic(inode->i_sb, "green-9010: unexpected item type for key %K\n",&key);
513 }
514 }
515
516 // the caller is responsible for closing the transaction
517 // unless we return an error, they are also responsible for logging
518 // the inode.
519 //
520 pathrelse(&path);
521 /*
522 * cleanup prellocation from previous writes
523 * if this is a partial block write
524 */
525 if (write_bytes & (inode->i_sb->s_blocksize -1))
526 reiserfs_discard_prealloc(th, inode);
527 reiserfs_write_unlock(inode->i_sb);
528
529 // go through all the pages/buffers and map the buffers to newly allocated
530 // blocks (so that system knows where to write these pages later).
531 curr_block = 0;
532 for ( i = 0; i < num_pages ; i++ ) {
533 struct page *page=prepared_pages[i]; //current page
534 struct buffer_head *head = page_buffers(page);// first buffer for a page
535 int block_start, block_end; // in-page offsets for buffers.
536
537 if (!page_buffers(page))
538 reiserfs_panic(inode->i_sb, "green-9005: No buffers for prepared page???");
539
540 /* For each buffer in page */
541 for(bh = head, block_start = 0; bh != head || !block_start;
542 block_start=block_end, bh = bh->b_this_page) {
543 if (!bh)
544 reiserfs_panic(inode->i_sb, "green-9006: Allocated but absent buffer for a page?");
545 block_end = block_start+inode->i_sb->s_blocksize;
546 if (i == 0 && block_end <= from )
547 /* if this buffer is before requested data to map, skip it */
548 continue;
549 if (i == num_pages - 1 && block_start >= to)
550 /* If this buffer is after requested data to map, abort
551 processing of current page */
552 break;
553
554 if ( !buffer_mapped(bh) ) { // Ok, unmapped buffer, need to map it
555 map_bh( bh, inode->i_sb, le32_to_cpu(allocated_blocks[curr_block]));
556 curr_block++;
557 set_buffer_new(bh);
558 }
559 } 650 }
560 }
561 651
562 RFALSE( curr_block > blocks_to_allocate, "green-9007: Used too many blocks? weird"); 652 RFALSE(curr_block > blocks_to_allocate,
653 "green-9007: Used too many blocks? weird");
563 654
564 kfree(allocated_blocks); 655 kfree(allocated_blocks);
565 return 0; 656 return 0;
566 657
567// Need to deal with transaction here. 658// Need to deal with transaction here.
568error_exit_free_blocks: 659 error_exit_free_blocks:
569 pathrelse(&path); 660 pathrelse(&path);
570 // free blocks 661 // free blocks
571 for( i = 0; i < blocks_to_allocate; i++ ) 662 for (i = 0; i < blocks_to_allocate; i++)
572 reiserfs_free_block(th, inode, le32_to_cpu(allocated_blocks[i]), 1); 663 reiserfs_free_block(th, inode, le32_to_cpu(allocated_blocks[i]),
573 664 1);
574error_exit: 665
575 if (th->t_trans_id) { 666 error_exit:
576 int err; 667 if (th->t_trans_id) {
577 // update any changes we made to blk count 668 int err;
578 reiserfs_update_sd(th, inode); 669 // update any changes we made to blk count
579 err = journal_end(th, inode->i_sb, JOURNAL_PER_BALANCE_CNT * 3 + 1 + 2 * REISERFS_QUOTA_TRANS_BLOCKS(inode->i_sb)); 670 reiserfs_update_sd(th, inode);
580 if (err) 671 err =
581 res = err; 672 journal_end(th, inode->i_sb,
582 } 673 JOURNAL_PER_BALANCE_CNT * 3 + 1 +
583 reiserfs_write_unlock(inode->i_sb); 674 2 * REISERFS_QUOTA_TRANS_BLOCKS(inode->i_sb));
584 kfree(allocated_blocks); 675 if (err)
585 676 res = err;
586 return res; 677 }
678 reiserfs_write_unlock(inode->i_sb);
679 kfree(allocated_blocks);
680
681 return res;
587} 682}
588 683
589/* Unlock pages prepared by reiserfs_prepare_file_region_for_write */ 684/* Unlock pages prepared by reiserfs_prepare_file_region_for_write */
590static void reiserfs_unprepare_pages(struct page **prepared_pages, /* list of locked pages */ 685static void reiserfs_unprepare_pages(struct page **prepared_pages, /* list of locked pages */
591 size_t num_pages /* amount of pages */) { 686 size_t num_pages /* amount of pages */ )
592 int i; // loop counter 687{
688 int i; // loop counter
593 689
594 for (i=0; i < num_pages ; i++) { 690 for (i = 0; i < num_pages; i++) {
595 struct page *page = prepared_pages[i]; 691 struct page *page = prepared_pages[i];
596 692
597 try_to_free_buffers(page); 693 try_to_free_buffers(page);
598 unlock_page(page); 694 unlock_page(page);
599 page_cache_release(page); 695 page_cache_release(page);
600 } 696 }
601} 697}
602 698
603/* This function will copy data from userspace to specified pages within 699/* This function will copy data from userspace to specified pages within
604 supplied byte range */ 700 supplied byte range */
605static int reiserfs_copy_from_user_to_file_region( 701static int reiserfs_copy_from_user_to_file_region(loff_t pos, /* In-file position */
606 loff_t pos, /* In-file position */ 702 int num_pages, /* Number of pages affected */
607 int num_pages, /* Number of pages affected */ 703 int write_bytes, /* Amount of bytes to write */
608 int write_bytes, /* Amount of bytes to write */ 704 struct page **prepared_pages, /* pointer to
609 struct page **prepared_pages, /* pointer to 705 array to
610 array to 706 prepared pages
611 prepared pages 707 */
612 */ 708 const char __user * buf /* Pointer to user-supplied
613 const char __user *buf /* Pointer to user-supplied 709 data */
614 data*/ 710 )
615 )
616{ 711{
617 long page_fault=0; // status of copy_from_user. 712 long page_fault = 0; // status of copy_from_user.
618 int i; // loop counter. 713 int i; // loop counter.
619 int offset; // offset in page 714 int offset; // offset in page
620 715
621 for ( i = 0, offset = (pos & (PAGE_CACHE_SIZE-1)); i < num_pages ; i++,offset=0) { 716 for (i = 0, offset = (pos & (PAGE_CACHE_SIZE - 1)); i < num_pages;
622 size_t count = min_t(size_t,PAGE_CACHE_SIZE-offset,write_bytes); // How much of bytes to write to this page 717 i++, offset = 0) {
623 struct page *page=prepared_pages[i]; // Current page we process. 718 size_t count = min_t(size_t, PAGE_CACHE_SIZE - offset, write_bytes); // How much of bytes to write to this page
624 719 struct page *page = prepared_pages[i]; // Current page we process.
625 fault_in_pages_readable( buf, count); 720
626 721 fault_in_pages_readable(buf, count);
627 /* Copy data from userspace to the current page */ 722
628 kmap(page); 723 /* Copy data from userspace to the current page */
629 page_fault = __copy_from_user(page_address(page)+offset, buf, count); // Copy the data. 724 kmap(page);
630 /* Flush processor's dcache for this page */ 725 page_fault = __copy_from_user(page_address(page) + offset, buf, count); // Copy the data.
631 flush_dcache_page(page); 726 /* Flush processor's dcache for this page */
632 kunmap(page); 727 flush_dcache_page(page);
633 buf+=count; 728 kunmap(page);
634 write_bytes-=count; 729 buf += count;
635 730 write_bytes -= count;
636 if (page_fault) 731
637 break; // Was there a fault? abort. 732 if (page_fault)
638 } 733 break; // Was there a fault? abort.
639 734 }
640 return page_fault?-EFAULT:0; 735
736 return page_fault ? -EFAULT : 0;
641} 737}
642 738
643/* taken fs/buffer.c:__block_commit_write */ 739/* taken fs/buffer.c:__block_commit_write */
644int reiserfs_commit_page(struct inode *inode, struct page *page, 740int reiserfs_commit_page(struct inode *inode, struct page *page,
645 unsigned from, unsigned to) 741 unsigned from, unsigned to)
646{ 742{
647 unsigned block_start, block_end; 743 unsigned block_start, block_end;
648 int partial = 0; 744 int partial = 0;
649 unsigned blocksize; 745 unsigned blocksize;
650 struct buffer_head *bh, *head; 746 struct buffer_head *bh, *head;
651 unsigned long i_size_index = inode->i_size >> PAGE_CACHE_SHIFT; 747 unsigned long i_size_index = inode->i_size >> PAGE_CACHE_SHIFT;
652 int new; 748 int new;
653 int logit = reiserfs_file_data_log(inode); 749 int logit = reiserfs_file_data_log(inode);
654 struct super_block *s = inode->i_sb; 750 struct super_block *s = inode->i_sb;
655 int bh_per_page = PAGE_CACHE_SIZE / s->s_blocksize; 751 int bh_per_page = PAGE_CACHE_SIZE / s->s_blocksize;
656 struct reiserfs_transaction_handle th; 752 struct reiserfs_transaction_handle th;
657 int ret = 0; 753 int ret = 0;
658 754
659 th.t_trans_id = 0; 755 th.t_trans_id = 0;
660 blocksize = 1 << inode->i_blkbits; 756 blocksize = 1 << inode->i_blkbits;
661 757
662 if (logit) { 758 if (logit) {
663 reiserfs_write_lock(s); 759 reiserfs_write_lock(s);
664 ret = journal_begin(&th, s, bh_per_page + 1); 760 ret = journal_begin(&th, s, bh_per_page + 1);
665 if (ret) 761 if (ret)
666 goto drop_write_lock; 762 goto drop_write_lock;
667 reiserfs_update_inode_transaction(inode); 763 reiserfs_update_inode_transaction(inode);
668 } 764 }
669 for(bh = head = page_buffers(page), block_start = 0; 765 for (bh = head = page_buffers(page), block_start = 0;
670 bh != head || !block_start; 766 bh != head || !block_start;
671 block_start=block_end, bh = bh->b_this_page) 767 block_start = block_end, bh = bh->b_this_page) {
672 { 768
673 769 new = buffer_new(bh);
674 new = buffer_new(bh); 770 clear_buffer_new(bh);
675 clear_buffer_new(bh); 771 block_end = block_start + blocksize;
676 block_end = block_start + blocksize; 772 if (block_end <= from || block_start >= to) {
677 if (block_end <= from || block_start >= to) { 773 if (!buffer_uptodate(bh))
678 if (!buffer_uptodate(bh)) 774 partial = 1;
679 partial = 1; 775 } else {
680 } else { 776 set_buffer_uptodate(bh);
681 set_buffer_uptodate(bh); 777 if (logit) {
682 if (logit) { 778 reiserfs_prepare_for_journal(s, bh, 1);
683 reiserfs_prepare_for_journal(s, bh, 1); 779 journal_mark_dirty(&th, s, bh);
684 journal_mark_dirty(&th, s, bh); 780 } else if (!buffer_dirty(bh)) {
685 } else if (!buffer_dirty(bh)) { 781 mark_buffer_dirty(bh);
686 mark_buffer_dirty(bh); 782 /* do data=ordered on any page past the end
687 /* do data=ordered on any page past the end 783 * of file and any buffer marked BH_New.
688 * of file and any buffer marked BH_New. 784 */
689 */ 785 if (reiserfs_data_ordered(inode->i_sb) &&
690 if (reiserfs_data_ordered(inode->i_sb) && 786 (new || page->index >= i_size_index)) {
691 (new || page->index >= i_size_index)) { 787 reiserfs_add_ordered_list(inode, bh);
692 reiserfs_add_ordered_list(inode, bh); 788 }
693 } 789 }
694 } 790 }
695 } 791 }
696 } 792 if (logit) {
697 if (logit) { 793 ret = journal_end(&th, s, bh_per_page + 1);
698 ret = journal_end(&th, s, bh_per_page + 1); 794 drop_write_lock:
699drop_write_lock: 795 reiserfs_write_unlock(s);
700 reiserfs_write_unlock(s); 796 }
701 } 797 /*
702 /* 798 * If this is a partial write which happened to make all buffers
703 * If this is a partial write which happened to make all buffers 799 * uptodate then we can optimize away a bogus readpage() for
704 * uptodate then we can optimize away a bogus readpage() for 800 * the next read(). Here we 'discover' whether the page went
705 * the next read(). Here we 'discover' whether the page went 801 * uptodate as a result of this (potentially partial) write.
706 * uptodate as a result of this (potentially partial) write. 802 */
707 */ 803 if (!partial)
708 if (!partial) 804 SetPageUptodate(page);
709 SetPageUptodate(page); 805 return ret;
710 return ret;
711} 806}
712 807
713
714/* Submit pages for write. This was separated from actual file copying 808/* Submit pages for write. This was separated from actual file copying
715 because we might want to allocate block numbers in-between. 809 because we might want to allocate block numbers in-between.
716 This function assumes that caller will adjust file size to correct value. */ 810 This function assumes that caller will adjust file size to correct value. */
717static int reiserfs_submit_file_region_for_write( 811static int reiserfs_submit_file_region_for_write(struct reiserfs_transaction_handle *th, struct inode *inode, loff_t pos, /* Writing position offset */
718 struct reiserfs_transaction_handle *th, 812 size_t num_pages, /* Number of pages to write */
719 struct inode *inode, 813 size_t write_bytes, /* number of bytes to write */
720 loff_t pos, /* Writing position offset */ 814 struct page **prepared_pages /* list of pages */
721 size_t num_pages, /* Number of pages to write */ 815 )
722 size_t write_bytes, /* number of bytes to write */
723 struct page **prepared_pages /* list of pages */
724 )
725{ 816{
726 int status; // return status of block_commit_write. 817 int status; // return status of block_commit_write.
727 int retval = 0; // Return value we are going to return. 818 int retval = 0; // Return value we are going to return.
728 int i; // loop counter 819 int i; // loop counter
729 int offset; // Writing offset in page. 820 int offset; // Writing offset in page.
730 int orig_write_bytes = write_bytes; 821 int orig_write_bytes = write_bytes;
731 int sd_update = 0; 822 int sd_update = 0;
732 823
733 for ( i = 0, offset = (pos & (PAGE_CACHE_SIZE-1)); i < num_pages ; i++,offset=0) { 824 for (i = 0, offset = (pos & (PAGE_CACHE_SIZE - 1)); i < num_pages;
734 int count = min_t(int,PAGE_CACHE_SIZE-offset,write_bytes); // How much of bytes to write to this page 825 i++, offset = 0) {
735 struct page *page=prepared_pages[i]; // Current page we process. 826 int count = min_t(int, PAGE_CACHE_SIZE - offset, write_bytes); // How much of bytes to write to this page
736 827 struct page *page = prepared_pages[i]; // Current page we process.
737 status = reiserfs_commit_page(inode, page, offset, offset+count); 828
738 if ( status ) 829 status =
739 retval = status; // To not overcomplicate matters We are going to 830 reiserfs_commit_page(inode, page, offset, offset + count);
740 // submit all the pages even if there was error. 831 if (status)
741 // we only remember error status to report it on 832 retval = status; // To not overcomplicate matters We are going to
742 // exit. 833 // submit all the pages even if there was error.
743 write_bytes-=count; 834 // we only remember error status to report it on
744 } 835 // exit.
745 /* now that we've gotten all the ordered buffers marked dirty, 836 write_bytes -= count;
746 * we can safely update i_size and close any running transaction 837 }
747 */ 838 /* now that we've gotten all the ordered buffers marked dirty,
748 if ( pos + orig_write_bytes > inode->i_size) { 839 * we can safely update i_size and close any running transaction
749 inode->i_size = pos + orig_write_bytes; // Set new size 840 */
750 /* If the file have grown so much that tail packing is no 841 if (pos + orig_write_bytes > inode->i_size) {
751 * longer possible, reset "need to pack" flag */ 842 inode->i_size = pos + orig_write_bytes; // Set new size
752 if ( (have_large_tails (inode->i_sb) && 843 /* If the file have grown so much that tail packing is no
753 inode->i_size > i_block_size (inode)*4) || 844 * longer possible, reset "need to pack" flag */
754 (have_small_tails (inode->i_sb) && 845 if ((have_large_tails(inode->i_sb) &&
755 inode->i_size > i_block_size(inode)) ) 846 inode->i_size > i_block_size(inode) * 4) ||
756 REISERFS_I(inode)->i_flags &= ~i_pack_on_close_mask ; 847 (have_small_tails(inode->i_sb) &&
757 else if ( (have_large_tails (inode->i_sb) && 848 inode->i_size > i_block_size(inode)))
758 inode->i_size < i_block_size (inode)*4) || 849 REISERFS_I(inode)->i_flags &= ~i_pack_on_close_mask;
759 (have_small_tails (inode->i_sb) && 850 else if ((have_large_tails(inode->i_sb) &&
760 inode->i_size < i_block_size(inode)) ) 851 inode->i_size < i_block_size(inode) * 4) ||
761 REISERFS_I(inode)->i_flags |= i_pack_on_close_mask ; 852 (have_small_tails(inode->i_sb) &&
762 853 inode->i_size < i_block_size(inode)))
854 REISERFS_I(inode)->i_flags |= i_pack_on_close_mask;
855
856 if (th->t_trans_id) {
857 reiserfs_write_lock(inode->i_sb);
858 reiserfs_update_sd(th, inode); // And update on-disk metadata
859 reiserfs_write_unlock(inode->i_sb);
860 } else
861 inode->i_sb->s_op->dirty_inode(inode);
862
863 sd_update = 1;
864 }
763 if (th->t_trans_id) { 865 if (th->t_trans_id) {
764 reiserfs_write_lock(inode->i_sb); 866 reiserfs_write_lock(inode->i_sb);
765 reiserfs_update_sd(th, inode); // And update on-disk metadata 867 if (!sd_update)
766 reiserfs_write_unlock(inode->i_sb); 868 reiserfs_update_sd(th, inode);
767 } else 869 status = journal_end(th, th->t_super, th->t_blocks_allocated);
768 inode->i_sb->s_op->dirty_inode(inode); 870 if (status)
871 retval = status;
872 reiserfs_write_unlock(inode->i_sb);
873 }
874 th->t_trans_id = 0;
769 875
770 sd_update = 1; 876 /*
771 } 877 * we have to unlock the pages after updating i_size, otherwise
772 if (th->t_trans_id) { 878 * we race with writepage
773 reiserfs_write_lock(inode->i_sb); 879 */
774 if (!sd_update) 880 for (i = 0; i < num_pages; i++) {
775 reiserfs_update_sd(th, inode); 881 struct page *page = prepared_pages[i];
776 status = journal_end(th, th->t_super, th->t_blocks_allocated); 882 unlock_page(page);
777 if (status) 883 mark_page_accessed(page);
778 retval = status; 884 page_cache_release(page);
779 reiserfs_write_unlock(inode->i_sb); 885 }
780 } 886 return retval;
781 th->t_trans_id = 0;
782
783 /*
784 * we have to unlock the pages after updating i_size, otherwise
785 * we race with writepage
786 */
787 for ( i = 0; i < num_pages ; i++) {
788 struct page *page=prepared_pages[i];
789 unlock_page(page);
790 mark_page_accessed(page);
791 page_cache_release(page);
792 }
793 return retval;
794} 887}
795 888
796/* Look if passed writing region is going to touch file's tail 889/* Look if passed writing region is going to touch file's tail
797 (if it is present). And if it is, convert the tail to unformatted node */ 890 (if it is present). And if it is, convert the tail to unformatted node */
798static int reiserfs_check_for_tail_and_convert( struct inode *inode, /* inode to deal with */ 891static int reiserfs_check_for_tail_and_convert(struct inode *inode, /* inode to deal with */
799 loff_t pos, /* Writing position */ 892 loff_t pos, /* Writing position */
800 int write_bytes /* amount of bytes to write */ 893 int write_bytes /* amount of bytes to write */
801 ) 894 )
802{ 895{
803 INITIALIZE_PATH(path); // needed for search_for_position 896 INITIALIZE_PATH(path); // needed for search_for_position
804 struct cpu_key key; // Key that would represent last touched writing byte. 897 struct cpu_key key; // Key that would represent last touched writing byte.
805 struct item_head *ih; // item header of found block; 898 struct item_head *ih; // item header of found block;
806 int res; // Return value of various functions we call. 899 int res; // Return value of various functions we call.
807 int cont_expand_offset; // We will put offset for generic_cont_expand here 900 int cont_expand_offset; // We will put offset for generic_cont_expand here
808 // This can be int just because tails are created 901 // This can be int just because tails are created
809 // only for small files. 902 // only for small files.
810 903
811/* this embodies a dependency on a particular tail policy */ 904/* this embodies a dependency on a particular tail policy */
812 if ( inode->i_size >= inode->i_sb->s_blocksize*4 ) { 905 if (inode->i_size >= inode->i_sb->s_blocksize * 4) {
813 /* such a big files do not have tails, so we won't bother ourselves 906 /* such a big files do not have tails, so we won't bother ourselves
814 to look for tails, simply return */ 907 to look for tails, simply return */
815 return 0; 908 return 0;
816 } 909 }
817
818 reiserfs_write_lock(inode->i_sb);
819 /* find the item containing the last byte to be written, or if
820 * writing past the end of the file then the last item of the
821 * file (and then we check its type). */
822 make_cpu_key (&key, inode, pos+write_bytes+1, TYPE_ANY, 3/*key length*/);
823 res = search_for_position_by_key(inode->i_sb, &key, &path);
824 if ( res == IO_ERROR ) {
825 reiserfs_write_unlock(inode->i_sb);
826 return -EIO;
827 }
828 ih = get_ih(&path);
829 res = 0;
830 if ( is_direct_le_ih(ih) ) {
831 /* Ok, closest item is file tail (tails are stored in "direct"
832 * items), so we need to unpack it. */
833 /* To not overcomplicate matters, we just call generic_cont_expand
834 which will in turn call other stuff and finally will boil down to
835 reiserfs_get_block() that would do necessary conversion. */
836 cont_expand_offset = le_key_k_offset(get_inode_item_key_version(inode), &(ih->ih_key));
837 pathrelse(&path);
838 res = generic_cont_expand( inode, cont_expand_offset);
839 } else
840 pathrelse(&path);
841 910
842 reiserfs_write_unlock(inode->i_sb); 911 reiserfs_write_lock(inode->i_sb);
843 return res; 912 /* find the item containing the last byte to be written, or if
913 * writing past the end of the file then the last item of the
914 * file (and then we check its type). */
915 make_cpu_key(&key, inode, pos + write_bytes + 1, TYPE_ANY,
916 3 /*key length */ );
917 res = search_for_position_by_key(inode->i_sb, &key, &path);
918 if (res == IO_ERROR) {
919 reiserfs_write_unlock(inode->i_sb);
920 return -EIO;
921 }
922 ih = get_ih(&path);
923 res = 0;
924 if (is_direct_le_ih(ih)) {
925 /* Ok, closest item is file tail (tails are stored in "direct"
926 * items), so we need to unpack it. */
927 /* To not overcomplicate matters, we just call generic_cont_expand
928 which will in turn call other stuff and finally will boil down to
929 reiserfs_get_block() that would do necessary conversion. */
930 cont_expand_offset =
931 le_key_k_offset(get_inode_item_key_version(inode),
932 &(ih->ih_key));
933 pathrelse(&path);
934 res = generic_cont_expand(inode, cont_expand_offset);
935 } else
936 pathrelse(&path);
937
938 reiserfs_write_unlock(inode->i_sb);
939 return res;
844} 940}
845 941
846/* This function locks pages starting from @pos for @inode. 942/* This function locks pages starting from @pos for @inode.
@@ -851,275 +947,296 @@ static int reiserfs_check_for_tail_and_convert( struct inode *inode, /* inode to
851 append), it is zeroed, then. 947 append), it is zeroed, then.
852 Returns number of unallocated blocks that should be allocated to cover 948 Returns number of unallocated blocks that should be allocated to cover
853 new file data.*/ 949 new file data.*/
854static int reiserfs_prepare_file_region_for_write( 950static int reiserfs_prepare_file_region_for_write(struct inode *inode
855 struct inode *inode /* Inode of the file */, 951 /* Inode of the file */ ,
856 loff_t pos, /* position in the file */ 952 loff_t pos, /* position in the file */
857 size_t num_pages, /* number of pages to 953 size_t num_pages, /* number of pages to
858 prepare */ 954 prepare */
859 size_t write_bytes, /* Amount of bytes to be 955 size_t write_bytes, /* Amount of bytes to be
860 overwritten from 956 overwritten from
861 @pos */ 957 @pos */
862 struct page **prepared_pages /* pointer to array 958 struct page **prepared_pages /* pointer to array
863 where to store 959 where to store
864 prepared pages */ 960 prepared pages */
865 ) 961 )
866{ 962{
867 int res=0; // Return values of different functions we call. 963 int res = 0; // Return values of different functions we call.
868 unsigned long index = pos >> PAGE_CACHE_SHIFT; // Offset in file in pages. 964 unsigned long index = pos >> PAGE_CACHE_SHIFT; // Offset in file in pages.
869 int from = (pos & (PAGE_CACHE_SIZE - 1)); // Writing offset in first page 965 int from = (pos & (PAGE_CACHE_SIZE - 1)); // Writing offset in first page
870 int to = ((pos + write_bytes - 1) & (PAGE_CACHE_SIZE - 1)) + 1; 966 int to = ((pos + write_bytes - 1) & (PAGE_CACHE_SIZE - 1)) + 1;
871 /* offset of last modified byte in last 967 /* offset of last modified byte in last
872 page */ 968 page */
873 struct address_space *mapping = inode->i_mapping; // Pages are mapped here. 969 struct address_space *mapping = inode->i_mapping; // Pages are mapped here.
874 int i; // Simple counter 970 int i; // Simple counter
875 int blocks = 0; /* Return value (blocks that should be allocated) */ 971 int blocks = 0; /* Return value (blocks that should be allocated) */
876 struct buffer_head *bh, *head; // Current bufferhead and first bufferhead 972 struct buffer_head *bh, *head; // Current bufferhead and first bufferhead
877 // of a page. 973 // of a page.
878 unsigned block_start, block_end; // Starting and ending offsets of current 974 unsigned block_start, block_end; // Starting and ending offsets of current
879 // buffer in the page. 975 // buffer in the page.
880 struct buffer_head *wait[2], **wait_bh=wait; // Buffers for page, if 976 struct buffer_head *wait[2], **wait_bh = wait; // Buffers for page, if
881 // Page appeared to be not up 977 // Page appeared to be not up
882 // to date. Note how we have 978 // to date. Note how we have
883 // at most 2 buffers, this is 979 // at most 2 buffers, this is
884 // because we at most may 980 // because we at most may
885 // partially overwrite two 981 // partially overwrite two
886 // buffers for one page. One at // the beginning of write area 982 // buffers for one page. One at // the beginning of write area
887 // and one at the end. 983 // and one at the end.
888 // Everything inthe middle gets // overwritten totally. 984 // Everything inthe middle gets // overwritten totally.
889 985
890 struct cpu_key key; // cpu key of item that we are going to deal with 986 struct cpu_key key; // cpu key of item that we are going to deal with
891 struct item_head *ih = NULL; // pointer to item head that we are going to deal with 987 struct item_head *ih = NULL; // pointer to item head that we are going to deal with
892 struct buffer_head *itembuf=NULL; // Buffer head that contains items that we are going to deal with 988 struct buffer_head *itembuf = NULL; // Buffer head that contains items that we are going to deal with
893 INITIALIZE_PATH(path); // path to item, that we are going to deal with. 989 INITIALIZE_PATH(path); // path to item, that we are going to deal with.
894 __le32 * item=NULL; // pointer to item we are going to deal with 990 __le32 *item = NULL; // pointer to item we are going to deal with
895 int item_pos=-1; /* Position in indirect item */ 991 int item_pos = -1; /* Position in indirect item */
896 992
897 993 if (num_pages < 1) {
898 if ( num_pages < 1 ) { 994 reiserfs_warning(inode->i_sb,
899 reiserfs_warning (inode->i_sb, 995 "green-9001: reiserfs_prepare_file_region_for_write "
900 "green-9001: reiserfs_prepare_file_region_for_write " 996 "called with zero number of pages to process");
901 "called with zero number of pages to process"); 997 return -EFAULT;
902 return -EFAULT;
903 }
904
905 /* We have 2 loops for pages. In first loop we grab and lock the pages, so
906 that nobody would touch these until we release the pages. Then
907 we'd start to deal with mapping buffers to blocks. */
908 for ( i = 0; i < num_pages; i++) {
909 prepared_pages[i] = grab_cache_page(mapping, index + i); // locks the page
910 if ( !prepared_pages[i]) {
911 res = -ENOMEM;
912 goto failed_page_grabbing;
913 }
914 if (!page_has_buffers(prepared_pages[i]))
915 create_empty_buffers(prepared_pages[i], inode->i_sb->s_blocksize, 0);
916 }
917
918 /* Let's count amount of blocks for a case where all the blocks
919 overwritten are new (we will substract already allocated blocks later)*/
920 if ( num_pages > 2 )
921 /* These are full-overwritten pages so we count all the blocks in
922 these pages are counted as needed to be allocated */
923 blocks = (num_pages - 2) << (PAGE_CACHE_SHIFT - inode->i_blkbits);
924
925 /* count blocks needed for first page (possibly partially written) */
926 blocks += ((PAGE_CACHE_SIZE - from) >> inode->i_blkbits) +
927 !!(from & (inode->i_sb->s_blocksize-1)); /* roundup */
928
929 /* Now we account for last page. If last page == first page (we
930 overwrite only one page), we substract all the blocks past the
931 last writing position in a page out of already calculated number
932 of blocks */
933 blocks += ((num_pages > 1) << (PAGE_CACHE_SHIFT-inode->i_blkbits)) -
934 ((PAGE_CACHE_SIZE - to) >> inode->i_blkbits);
935 /* Note how we do not roundup here since partial blocks still
936 should be allocated */
937
938 /* Now if all the write area lies past the file end, no point in
939 maping blocks, since there is none, so we just zero out remaining
940 parts of first and last pages in write area (if needed) */
941 if ( (pos & ~((loff_t)PAGE_CACHE_SIZE - 1)) > inode->i_size ) {
942 if ( from != 0 ) {/* First page needs to be partially zeroed */
943 char *kaddr = kmap_atomic(prepared_pages[0], KM_USER0);
944 memset(kaddr, 0, from);
945 kunmap_atomic( kaddr, KM_USER0);
946 }
947 if ( to != PAGE_CACHE_SIZE ) { /* Last page needs to be partially zeroed */
948 char *kaddr = kmap_atomic(prepared_pages[num_pages-1], KM_USER0);
949 memset(kaddr+to, 0, PAGE_CACHE_SIZE - to);
950 kunmap_atomic( kaddr, KM_USER0);
951 } 998 }
952 999
953 /* Since all blocks are new - use already calculated value */ 1000 /* We have 2 loops for pages. In first loop we grab and lock the pages, so
954 return blocks; 1001 that nobody would touch these until we release the pages. Then
955 } 1002 we'd start to deal with mapping buffers to blocks. */
956 1003 for (i = 0; i < num_pages; i++) {
957 /* Well, since we write somewhere into the middle of a file, there is 1004 prepared_pages[i] = grab_cache_page(mapping, index + i); // locks the page
958 possibility we are writing over some already allocated blocks, so 1005 if (!prepared_pages[i]) {
959 let's map these blocks and substract number of such blocks out of blocks 1006 res = -ENOMEM;
960 we need to allocate (calculated above) */ 1007 goto failed_page_grabbing;
961 /* Mask write position to start on blocksize, we do it out of the
962 loop for performance reasons */
963 pos &= ~((loff_t) inode->i_sb->s_blocksize - 1);
964 /* Set cpu key to the starting position in a file (on left block boundary)*/
965 make_cpu_key (&key, inode, 1 + ((pos) & ~((loff_t) inode->i_sb->s_blocksize - 1)), TYPE_ANY, 3/*key length*/);
966
967 reiserfs_write_lock(inode->i_sb); // We need that for at least search_by_key()
968 for ( i = 0; i < num_pages ; i++ ) {
969
970 head = page_buffers(prepared_pages[i]);
971 /* For each buffer in the page */
972 for(bh = head, block_start = 0; bh != head || !block_start;
973 block_start=block_end, bh = bh->b_this_page) {
974 if (!bh)
975 reiserfs_panic(inode->i_sb, "green-9002: Allocated but absent buffer for a page?");
976 /* Find where this buffer ends */
977 block_end = block_start+inode->i_sb->s_blocksize;
978 if (i == 0 && block_end <= from )
979 /* if this buffer is before requested data to map, skip it*/
980 continue;
981
982 if (i == num_pages - 1 && block_start >= to) {
983 /* If this buffer is after requested data to map, abort
984 processing of current page */
985 break;
986 } 1008 }
1009 if (!page_has_buffers(prepared_pages[i]))
1010 create_empty_buffers(prepared_pages[i],
1011 inode->i_sb->s_blocksize, 0);
1012 }
987 1013
988 if ( buffer_mapped(bh) && bh->b_blocknr !=0 ) { 1014 /* Let's count amount of blocks for a case where all the blocks
989 /* This is optimisation for a case where buffer is mapped 1015 overwritten are new (we will substract already allocated blocks later) */
990 and have blocknumber assigned. In case significant amount 1016 if (num_pages > 2)
991 of such buffers are present, we may avoid some amount 1017 /* These are full-overwritten pages so we count all the blocks in
992 of search_by_key calls. 1018 these pages are counted as needed to be allocated */
993 Probably it would be possible to move parts of this code 1019 blocks =
994 out of BKL, but I afraid that would overcomplicate code 1020 (num_pages - 2) << (PAGE_CACHE_SHIFT - inode->i_blkbits);
995 without any noticeable benefit. 1021
996 */ 1022 /* count blocks needed for first page (possibly partially written) */
997 item_pos++; 1023 blocks += ((PAGE_CACHE_SIZE - from) >> inode->i_blkbits) + !!(from & (inode->i_sb->s_blocksize - 1)); /* roundup */
998 /* Update the key */ 1024
999 set_cpu_key_k_offset( &key, cpu_key_k_offset(&key) + inode->i_sb->s_blocksize); 1025 /* Now we account for last page. If last page == first page (we
1000 blocks--; // Decrease the amount of blocks that need to be 1026 overwrite only one page), we substract all the blocks past the
1001 // allocated 1027 last writing position in a page out of already calculated number
1002 continue; // Go to the next buffer 1028 of blocks */
1029 blocks += ((num_pages > 1) << (PAGE_CACHE_SHIFT - inode->i_blkbits)) -
1030 ((PAGE_CACHE_SIZE - to) >> inode->i_blkbits);
1031 /* Note how we do not roundup here since partial blocks still
1032 should be allocated */
1033
1034 /* Now if all the write area lies past the file end, no point in
1035 maping blocks, since there is none, so we just zero out remaining
1036 parts of first and last pages in write area (if needed) */
1037 if ((pos & ~((loff_t) PAGE_CACHE_SIZE - 1)) > inode->i_size) {
1038 if (from != 0) { /* First page needs to be partially zeroed */
1039 char *kaddr = kmap_atomic(prepared_pages[0], KM_USER0);
1040 memset(kaddr, 0, from);
1041 kunmap_atomic(kaddr, KM_USER0);
1042 }
1043 if (to != PAGE_CACHE_SIZE) { /* Last page needs to be partially zeroed */
1044 char *kaddr =
1045 kmap_atomic(prepared_pages[num_pages - 1],
1046 KM_USER0);
1047 memset(kaddr + to, 0, PAGE_CACHE_SIZE - to);
1048 kunmap_atomic(kaddr, KM_USER0);
1003 } 1049 }
1004 1050
1005 if ( !itembuf || /* if first iteration */ 1051 /* Since all blocks are new - use already calculated value */
1006 item_pos >= ih_item_len(ih)/UNFM_P_SIZE) 1052 return blocks;
1007 { /* or if we progressed past the 1053 }
1008 current unformatted_item */ 1054
1009 /* Try to find next item */ 1055 /* Well, since we write somewhere into the middle of a file, there is
1010 res = search_for_position_by_key(inode->i_sb, &key, &path); 1056 possibility we are writing over some already allocated blocks, so
1011 /* Abort if no more items */ 1057 let's map these blocks and substract number of such blocks out of blocks
1012 if ( res != POSITION_FOUND ) { 1058 we need to allocate (calculated above) */
1013 /* make sure later loops don't use this item */ 1059 /* Mask write position to start on blocksize, we do it out of the
1014 itembuf = NULL; 1060 loop for performance reasons */
1015 item = NULL; 1061 pos &= ~((loff_t) inode->i_sb->s_blocksize - 1);
1016 break; 1062 /* Set cpu key to the starting position in a file (on left block boundary) */
1063 make_cpu_key(&key, inode,
1064 1 + ((pos) & ~((loff_t) inode->i_sb->s_blocksize - 1)),
1065 TYPE_ANY, 3 /*key length */ );
1066
1067 reiserfs_write_lock(inode->i_sb); // We need that for at least search_by_key()
1068 for (i = 0; i < num_pages; i++) {
1069
1070 head = page_buffers(prepared_pages[i]);
1071 /* For each buffer in the page */
1072 for (bh = head, block_start = 0; bh != head || !block_start;
1073 block_start = block_end, bh = bh->b_this_page) {
1074 if (!bh)
1075 reiserfs_panic(inode->i_sb,
1076 "green-9002: Allocated but absent buffer for a page?");
1077 /* Find where this buffer ends */
1078 block_end = block_start + inode->i_sb->s_blocksize;
1079 if (i == 0 && block_end <= from)
1080 /* if this buffer is before requested data to map, skip it */
1081 continue;
1082
1083 if (i == num_pages - 1 && block_start >= to) {
1084 /* If this buffer is after requested data to map, abort
1085 processing of current page */
1086 break;
1017 } 1087 }
1018 1088
1019 /* Update information about current indirect item */ 1089 if (buffer_mapped(bh) && bh->b_blocknr != 0) {
1020 itembuf = get_last_bh( &path ); 1090 /* This is optimisation for a case where buffer is mapped
1021 ih = get_ih( &path ); 1091 and have blocknumber assigned. In case significant amount
1022 item = get_item( &path ); 1092 of such buffers are present, we may avoid some amount
1023 item_pos = path.pos_in_item; 1093 of search_by_key calls.
1094 Probably it would be possible to move parts of this code
1095 out of BKL, but I afraid that would overcomplicate code
1096 without any noticeable benefit.
1097 */
1098 item_pos++;
1099 /* Update the key */
1100 set_cpu_key_k_offset(&key,
1101 cpu_key_k_offset(&key) +
1102 inode->i_sb->s_blocksize);
1103 blocks--; // Decrease the amount of blocks that need to be
1104 // allocated
1105 continue; // Go to the next buffer
1106 }
1024 1107
1025 RFALSE( !is_indirect_le_ih (ih), "green-9003: indirect item expected"); 1108 if (!itembuf || /* if first iteration */
1026 } 1109 item_pos >= ih_item_len(ih) / UNFM_P_SIZE) { /* or if we progressed past the
1110 current unformatted_item */
1111 /* Try to find next item */
1112 res =
1113 search_for_position_by_key(inode->i_sb,
1114 &key, &path);
1115 /* Abort if no more items */
1116 if (res != POSITION_FOUND) {
1117 /* make sure later loops don't use this item */
1118 itembuf = NULL;
1119 item = NULL;
1120 break;
1121 }
1122
1123 /* Update information about current indirect item */
1124 itembuf = get_last_bh(&path);
1125 ih = get_ih(&path);
1126 item = get_item(&path);
1127 item_pos = path.pos_in_item;
1128
1129 RFALSE(!is_indirect_le_ih(ih),
1130 "green-9003: indirect item expected");
1131 }
1027 1132
1028 /* See if there is some block associated with the file 1133 /* See if there is some block associated with the file
1029 at that position, map the buffer to this block */ 1134 at that position, map the buffer to this block */
1030 if ( get_block_num(item,item_pos) ) { 1135 if (get_block_num(item, item_pos)) {
1031 map_bh(bh, inode->i_sb, get_block_num(item,item_pos)); 1136 map_bh(bh, inode->i_sb,
1032 blocks--; // Decrease the amount of blocks that need to be 1137 get_block_num(item, item_pos));
1033 // allocated 1138 blocks--; // Decrease the amount of blocks that need to be
1139 // allocated
1140 }
1141 item_pos++;
1142 /* Update the key */
1143 set_cpu_key_k_offset(&key,
1144 cpu_key_k_offset(&key) +
1145 inode->i_sb->s_blocksize);
1034 } 1146 }
1035 item_pos++;
1036 /* Update the key */
1037 set_cpu_key_k_offset( &key, cpu_key_k_offset(&key) + inode->i_sb->s_blocksize);
1038 } 1147 }
1039 } 1148 pathrelse(&path); // Free the path
1040 pathrelse(&path); // Free the path 1149 reiserfs_write_unlock(inode->i_sb);
1041 reiserfs_write_unlock(inode->i_sb);
1042 1150
1043 /* Now zero out unmappend buffers for the first and last pages of 1151 /* Now zero out unmappend buffers for the first and last pages of
1044 write area or issue read requests if page is mapped. */ 1152 write area or issue read requests if page is mapped. */
1045 /* First page, see if it is not uptodate */ 1153 /* First page, see if it is not uptodate */
1046 if ( !PageUptodate(prepared_pages[0]) ) { 1154 if (!PageUptodate(prepared_pages[0])) {
1047 head = page_buffers(prepared_pages[0]); 1155 head = page_buffers(prepared_pages[0]);
1048 1156
1049 /* For each buffer in page */ 1157 /* For each buffer in page */
1050 for(bh = head, block_start = 0; bh != head || !block_start; 1158 for (bh = head, block_start = 0; bh != head || !block_start;
1051 block_start=block_end, bh = bh->b_this_page) { 1159 block_start = block_end, bh = bh->b_this_page) {
1052 1160
1053 if (!bh) 1161 if (!bh)
1054 reiserfs_panic(inode->i_sb, "green-9002: Allocated but absent buffer for a page?"); 1162 reiserfs_panic(inode->i_sb,
1055 /* Find where this buffer ends */ 1163 "green-9002: Allocated but absent buffer for a page?");
1056 block_end = block_start+inode->i_sb->s_blocksize; 1164 /* Find where this buffer ends */
1057 if ( block_end <= from ) 1165 block_end = block_start + inode->i_sb->s_blocksize;
1058 /* if this buffer is before requested data to map, skip it*/ 1166 if (block_end <= from)
1059 continue; 1167 /* if this buffer is before requested data to map, skip it */
1060 if ( block_start < from ) { /* Aha, our partial buffer */ 1168 continue;
1061 if ( buffer_mapped(bh) ) { /* If it is mapped, we need to 1169 if (block_start < from) { /* Aha, our partial buffer */
1062 issue READ request for it to 1170 if (buffer_mapped(bh)) { /* If it is mapped, we need to
1063 not loose data */ 1171 issue READ request for it to
1064 ll_rw_block(READ, 1, &bh); 1172 not loose data */
1065 *wait_bh++=bh; 1173 ll_rw_block(READ, 1, &bh);
1066 } else { /* Not mapped, zero it */ 1174 *wait_bh++ = bh;
1067 char *kaddr = kmap_atomic(prepared_pages[0], KM_USER0); 1175 } else { /* Not mapped, zero it */
1068 memset(kaddr+block_start, 0, from-block_start); 1176 char *kaddr =
1069 kunmap_atomic( kaddr, KM_USER0); 1177 kmap_atomic(prepared_pages[0],
1070 set_buffer_uptodate(bh); 1178 KM_USER0);
1071 } 1179 memset(kaddr + block_start, 0,
1180 from - block_start);
1181 kunmap_atomic(kaddr, KM_USER0);
1182 set_buffer_uptodate(bh);
1183 }
1184 }
1072 } 1185 }
1073 }
1074 } 1186 }
1075 1187
1076 /* Last page, see if it is not uptodate, or if the last page is past the end of the file. */ 1188 /* Last page, see if it is not uptodate, or if the last page is past the end of the file. */
1077 if ( !PageUptodate(prepared_pages[num_pages-1]) || 1189 if (!PageUptodate(prepared_pages[num_pages - 1]) ||
1078 ((pos+write_bytes)>>PAGE_CACHE_SHIFT) > (inode->i_size>>PAGE_CACHE_SHIFT) ) { 1190 ((pos + write_bytes) >> PAGE_CACHE_SHIFT) >
1079 head = page_buffers(prepared_pages[num_pages-1]); 1191 (inode->i_size >> PAGE_CACHE_SHIFT)) {
1080 1192 head = page_buffers(prepared_pages[num_pages - 1]);
1081 /* for each buffer in page */ 1193
1082 for(bh = head, block_start = 0; bh != head || !block_start; 1194 /* for each buffer in page */
1083 block_start=block_end, bh = bh->b_this_page) { 1195 for (bh = head, block_start = 0; bh != head || !block_start;
1084 1196 block_start = block_end, bh = bh->b_this_page) {
1085 if (!bh) 1197
1086 reiserfs_panic(inode->i_sb, "green-9002: Allocated but absent buffer for a page?"); 1198 if (!bh)
1087 /* Find where this buffer ends */ 1199 reiserfs_panic(inode->i_sb,
1088 block_end = block_start+inode->i_sb->s_blocksize; 1200 "green-9002: Allocated but absent buffer for a page?");
1089 if ( block_start >= to ) 1201 /* Find where this buffer ends */
1090 /* if this buffer is after requested data to map, skip it*/ 1202 block_end = block_start + inode->i_sb->s_blocksize;
1091 break; 1203 if (block_start >= to)
1092 if ( block_end > to ) { /* Aha, our partial buffer */ 1204 /* if this buffer is after requested data to map, skip it */
1093 if ( buffer_mapped(bh) ) { /* If it is mapped, we need to 1205 break;
1094 issue READ request for it to 1206 if (block_end > to) { /* Aha, our partial buffer */
1095 not loose data */ 1207 if (buffer_mapped(bh)) { /* If it is mapped, we need to
1096 ll_rw_block(READ, 1, &bh); 1208 issue READ request for it to
1097 *wait_bh++=bh; 1209 not loose data */
1098 } else { /* Not mapped, zero it */ 1210 ll_rw_block(READ, 1, &bh);
1099 char *kaddr = kmap_atomic(prepared_pages[num_pages-1], KM_USER0); 1211 *wait_bh++ = bh;
1100 memset(kaddr+to, 0, block_end-to); 1212 } else { /* Not mapped, zero it */
1101 kunmap_atomic( kaddr, KM_USER0); 1213 char *kaddr =
1102 set_buffer_uptodate(bh); 1214 kmap_atomic(prepared_pages
1103 } 1215 [num_pages - 1],
1216 KM_USER0);
1217 memset(kaddr + to, 0, block_end - to);
1218 kunmap_atomic(kaddr, KM_USER0);
1219 set_buffer_uptodate(bh);
1220 }
1221 }
1104 } 1222 }
1105 }
1106 } 1223 }
1107 1224
1108 /* Wait for read requests we made to happen, if necessary */ 1225 /* Wait for read requests we made to happen, if necessary */
1109 while(wait_bh > wait) { 1226 while (wait_bh > wait) {
1110 wait_on_buffer(*--wait_bh); 1227 wait_on_buffer(*--wait_bh);
1111 if (!buffer_uptodate(*wait_bh)) { 1228 if (!buffer_uptodate(*wait_bh)) {
1112 res = -EIO; 1229 res = -EIO;
1113 goto failed_read; 1230 goto failed_read;
1231 }
1114 } 1232 }
1115 } 1233
1116 1234 return blocks;
1117 return blocks; 1235 failed_page_grabbing:
1118failed_page_grabbing: 1236 num_pages = i;
1119 num_pages = i; 1237 failed_read:
1120failed_read: 1238 reiserfs_unprepare_pages(prepared_pages, num_pages);
1121 reiserfs_unprepare_pages(prepared_pages, num_pages); 1239 return res;
1122 return res;
1123} 1240}
1124 1241
1125/* Write @count bytes at position @ppos in a file indicated by @file 1242/* Write @count bytes at position @ppos in a file indicated by @file
@@ -1148,262 +1265,305 @@ failed_read:
1148 Future Features: providing search_by_key with hints. 1265 Future Features: providing search_by_key with hints.
1149 1266
1150*/ 1267*/
1151static ssize_t reiserfs_file_write( struct file *file, /* the file we are going to write into */ 1268static ssize_t reiserfs_file_write(struct file *file, /* the file we are going to write into */
1152 const char __user *buf, /* pointer to user supplied data 1269 const char __user * buf, /* pointer to user supplied data
1153(in userspace) */ 1270 (in userspace) */
1154 size_t count, /* amount of bytes to write */ 1271 size_t count, /* amount of bytes to write */
1155 loff_t *ppos /* pointer to position in file that we start writing at. Should be updated to 1272 loff_t * ppos /* pointer to position in file that we start writing at. Should be updated to
1156 * new current position before returning. */ ) 1273 * new current position before returning. */
1274 )
1157{ 1275{
1158 size_t already_written = 0; // Number of bytes already written to the file. 1276 size_t already_written = 0; // Number of bytes already written to the file.
1159 loff_t pos; // Current position in the file. 1277 loff_t pos; // Current position in the file.
1160 ssize_t res; // return value of various functions that we call. 1278 ssize_t res; // return value of various functions that we call.
1161 int err = 0; 1279 int err = 0;
1162 struct inode *inode = file->f_dentry->d_inode; // Inode of the file that we are writing to. 1280 struct inode *inode = file->f_dentry->d_inode; // Inode of the file that we are writing to.
1163 /* To simplify coding at this time, we store 1281 /* To simplify coding at this time, we store
1164 locked pages in array for now */ 1282 locked pages in array for now */
1165 struct page * prepared_pages[REISERFS_WRITE_PAGES_AT_A_TIME]; 1283 struct page *prepared_pages[REISERFS_WRITE_PAGES_AT_A_TIME];
1166 struct reiserfs_transaction_handle th; 1284 struct reiserfs_transaction_handle th;
1167 th.t_trans_id = 0; 1285 th.t_trans_id = 0;
1168 1286
1169 if ( file->f_flags & O_DIRECT) { // Direct IO needs treatment 1287 if (file->f_flags & O_DIRECT) { // Direct IO needs treatment
1170 ssize_t result, after_file_end = 0; 1288 ssize_t result, after_file_end = 0;
1171 if ( (*ppos + count >= inode->i_size) || (file->f_flags & O_APPEND) ) { 1289 if ((*ppos + count >= inode->i_size)
1172 /* If we are appending a file, we need to put this savelink in here. 1290 || (file->f_flags & O_APPEND)) {
1173 If we will crash while doing direct io, finish_unfinished will 1291 /* If we are appending a file, we need to put this savelink in here.
1174 cut the garbage from the file end. */ 1292 If we will crash while doing direct io, finish_unfinished will
1175 reiserfs_write_lock(inode->i_sb); 1293 cut the garbage from the file end. */
1176 err = journal_begin(&th, inode->i_sb, JOURNAL_PER_BALANCE_CNT ); 1294 reiserfs_write_lock(inode->i_sb);
1177 if (err) { 1295 err =
1178 reiserfs_write_unlock (inode->i_sb); 1296 journal_begin(&th, inode->i_sb,
1179 return err; 1297 JOURNAL_PER_BALANCE_CNT);
1180 } 1298 if (err) {
1181 reiserfs_update_inode_transaction(inode); 1299 reiserfs_write_unlock(inode->i_sb);
1182 add_save_link (&th, inode, 1 /* Truncate */); 1300 return err;
1183 after_file_end = 1; 1301 }
1184 err = journal_end(&th, inode->i_sb, JOURNAL_PER_BALANCE_CNT ); 1302 reiserfs_update_inode_transaction(inode);
1185 reiserfs_write_unlock(inode->i_sb); 1303 add_save_link(&th, inode, 1 /* Truncate */ );
1186 if (err) 1304 after_file_end = 1;
1187 return err; 1305 err =
1188 } 1306 journal_end(&th, inode->i_sb,
1189 result = generic_file_write(file, buf, count, ppos); 1307 JOURNAL_PER_BALANCE_CNT);
1190 1308 reiserfs_write_unlock(inode->i_sb);
1191 if ( after_file_end ) { /* Now update i_size and remove the savelink */ 1309 if (err)
1192 struct reiserfs_transaction_handle th; 1310 return err;
1193 reiserfs_write_lock(inode->i_sb); 1311 }
1194 err = journal_begin(&th, inode->i_sb, 1); 1312 result = generic_file_write(file, buf, count, ppos);
1195 if (err) { 1313
1196 reiserfs_write_unlock (inode->i_sb); 1314 if (after_file_end) { /* Now update i_size and remove the savelink */
1197 return err; 1315 struct reiserfs_transaction_handle th;
1198 } 1316 reiserfs_write_lock(inode->i_sb);
1199 reiserfs_update_inode_transaction(inode); 1317 err = journal_begin(&th, inode->i_sb, 1);
1200 reiserfs_update_sd(&th, inode); 1318 if (err) {
1201 err = journal_end(&th, inode->i_sb, 1); 1319 reiserfs_write_unlock(inode->i_sb);
1202 if (err) { 1320 return err;
1203 reiserfs_write_unlock (inode->i_sb); 1321 }
1204 return err; 1322 reiserfs_update_inode_transaction(inode);
1205 } 1323 reiserfs_update_sd(&th, inode);
1206 err = remove_save_link (inode, 1/* truncate */); 1324 err = journal_end(&th, inode->i_sb, 1);
1207 reiserfs_write_unlock(inode->i_sb); 1325 if (err) {
1208 if (err) 1326 reiserfs_write_unlock(inode->i_sb);
1209 return err; 1327 return err;
1210 } 1328 }
1211 1329 err = remove_save_link(inode, 1 /* truncate */ );
1212 return result; 1330 reiserfs_write_unlock(inode->i_sb);
1213 } 1331 if (err)
1214 1332 return err;
1215 if ( unlikely((ssize_t) count < 0 )) 1333 }
1216 return -EINVAL;
1217
1218 if (unlikely(!access_ok(VERIFY_READ, buf, count)))
1219 return -EFAULT;
1220
1221 down(&inode->i_sem); // locks the entire file for just us
1222
1223 pos = *ppos;
1224
1225 /* Check if we can write to specified region of file, file
1226 is not overly big and this kind of stuff. Adjust pos and
1227 count, if needed */
1228 res = generic_write_checks(file, &pos, &count, 0);
1229 if (res)
1230 goto out;
1231
1232 if ( count == 0 )
1233 goto out;
1234
1235 res = remove_suid(file->f_dentry);
1236 if (res)
1237 goto out;
1238
1239 inode_update_time(inode, 1); /* Both mtime and ctime */
1240
1241 // Ok, we are done with all the checks.
1242 1334
1243 // Now we should start real work 1335 return result;
1336 }
1244 1337
1245 /* If we are going to write past the file's packed tail or if we are going 1338 if (unlikely((ssize_t) count < 0))
1246 to overwrite part of the tail, we need that tail to be converted into 1339 return -EINVAL;
1247 unformatted node */ 1340
1248 res = reiserfs_check_for_tail_and_convert( inode, pos, count); 1341 if (unlikely(!access_ok(VERIFY_READ, buf, count)))
1249 if (res) 1342 return -EFAULT;
1250 goto out; 1343
1344 down(&inode->i_sem); // locks the entire file for just us
1345
1346 pos = *ppos;
1347
1348 /* Check if we can write to specified region of file, file
1349 is not overly big and this kind of stuff. Adjust pos and
1350 count, if needed */
1351 res = generic_write_checks(file, &pos, &count, 0);
1352 if (res)
1353 goto out;
1354
1355 if (count == 0)
1356 goto out;
1357
1358 res = remove_suid(file->f_dentry);
1359 if (res)
1360 goto out;
1361
1362 inode_update_time(inode, 1); /* Both mtime and ctime */
1363
1364 // Ok, we are done with all the checks.
1365
1366 // Now we should start real work
1367
1368 /* If we are going to write past the file's packed tail or if we are going
1369 to overwrite part of the tail, we need that tail to be converted into
1370 unformatted node */
1371 res = reiserfs_check_for_tail_and_convert(inode, pos, count);
1372 if (res)
1373 goto out;
1374
1375 while (count > 0) {
1376 /* This is the main loop in which we running until some error occures
1377 or until we write all of the data. */
1378 size_t num_pages; /* amount of pages we are going to write this iteration */
1379 size_t write_bytes; /* amount of bytes to write during this iteration */
1380 size_t blocks_to_allocate; /* how much blocks we need to allocate for this iteration */
1381
1382 /* (pos & (PAGE_CACHE_SIZE-1)) is an idiom for offset into a page of pos */
1383 num_pages = !!((pos + count) & (PAGE_CACHE_SIZE - 1)) + /* round up partial
1384 pages */
1385 ((count +
1386 (pos & (PAGE_CACHE_SIZE - 1))) >> PAGE_CACHE_SHIFT);
1387 /* convert size to amount of
1388 pages */
1389 reiserfs_write_lock(inode->i_sb);
1390 if (num_pages > REISERFS_WRITE_PAGES_AT_A_TIME
1391 || num_pages > reiserfs_can_fit_pages(inode->i_sb)) {
1392 /* If we were asked to write more data than we want to or if there
1393 is not that much space, then we shorten amount of data to write
1394 for this iteration. */
1395 num_pages =
1396 min_t(size_t, REISERFS_WRITE_PAGES_AT_A_TIME,
1397 reiserfs_can_fit_pages(inode->i_sb));
1398 /* Also we should not forget to set size in bytes accordingly */
1399 write_bytes = (num_pages << PAGE_CACHE_SHIFT) -
1400 (pos & (PAGE_CACHE_SIZE - 1));
1401 /* If position is not on the
1402 start of the page, we need
1403 to substract the offset
1404 within page */
1405 } else
1406 write_bytes = count;
1407
1408 /* reserve the blocks to be allocated later, so that later on
1409 we still have the space to write the blocks to */
1410 reiserfs_claim_blocks_to_be_allocated(inode->i_sb,
1411 num_pages <<
1412 (PAGE_CACHE_SHIFT -
1413 inode->i_blkbits));
1414 reiserfs_write_unlock(inode->i_sb);
1415
1416 if (!num_pages) { /* If we do not have enough space even for a single page... */
1417 if (pos >
1418 inode->i_size + inode->i_sb->s_blocksize -
1419 (pos & (inode->i_sb->s_blocksize - 1))) {
1420 res = -ENOSPC;
1421 break; // In case we are writing past the end of the last file block, break.
1422 }
1423 // Otherwise we are possibly overwriting the file, so
1424 // let's set write size to be equal or less than blocksize.
1425 // This way we get it correctly for file holes.
1426 // But overwriting files on absolutelly full volumes would not
1427 // be very efficient. Well, people are not supposed to fill
1428 // 100% of disk space anyway.
1429 write_bytes =
1430 min_t(size_t, count,
1431 inode->i_sb->s_blocksize -
1432 (pos & (inode->i_sb->s_blocksize - 1)));
1433 num_pages = 1;
1434 // No blocks were claimed before, so do it now.
1435 reiserfs_claim_blocks_to_be_allocated(inode->i_sb,
1436 1 <<
1437 (PAGE_CACHE_SHIFT
1438 -
1439 inode->
1440 i_blkbits));
1441 }
1251 1442
1252 while ( count > 0) { 1443 /* Prepare for writing into the region, read in all the
1253 /* This is the main loop in which we running until some error occures 1444 partially overwritten pages, if needed. And lock the pages,
1254 or until we write all of the data. */ 1445 so that nobody else can access these until we are done.
1255 size_t num_pages;/* amount of pages we are going to write this iteration */ 1446 We get number of actual blocks needed as a result. */
1256 size_t write_bytes; /* amount of bytes to write during this iteration */ 1447 blocks_to_allocate =
1257 size_t blocks_to_allocate; /* how much blocks we need to allocate for this iteration */ 1448 reiserfs_prepare_file_region_for_write(inode, pos,
1258 1449 num_pages,
1259 /* (pos & (PAGE_CACHE_SIZE-1)) is an idiom for offset into a page of pos*/ 1450 write_bytes,
1260 num_pages = !!((pos+count) & (PAGE_CACHE_SIZE - 1)) + /* round up partial 1451 prepared_pages);
1261 pages */ 1452 if (blocks_to_allocate < 0) {
1262 ((count + (pos & (PAGE_CACHE_SIZE-1))) >> PAGE_CACHE_SHIFT); 1453 res = blocks_to_allocate;
1263 /* convert size to amount of 1454 reiserfs_release_claimed_blocks(inode->i_sb,
1264 pages */ 1455 num_pages <<
1265 reiserfs_write_lock(inode->i_sb); 1456 (PAGE_CACHE_SHIFT -
1266 if ( num_pages > REISERFS_WRITE_PAGES_AT_A_TIME 1457 inode->i_blkbits));
1267 || num_pages > reiserfs_can_fit_pages(inode->i_sb) ) { 1458 break;
1268 /* If we were asked to write more data than we want to or if there 1459 }
1269 is not that much space, then we shorten amount of data to write
1270 for this iteration. */
1271 num_pages = min_t(size_t, REISERFS_WRITE_PAGES_AT_A_TIME, reiserfs_can_fit_pages(inode->i_sb));
1272 /* Also we should not forget to set size in bytes accordingly */
1273 write_bytes = (num_pages << PAGE_CACHE_SHIFT) -
1274 (pos & (PAGE_CACHE_SIZE-1));
1275 /* If position is not on the
1276 start of the page, we need
1277 to substract the offset
1278 within page */
1279 } else
1280 write_bytes = count;
1281 1460
1282 /* reserve the blocks to be allocated later, so that later on 1461 /* First we correct our estimate of how many blocks we need */
1283 we still have the space to write the blocks to */ 1462 reiserfs_release_claimed_blocks(inode->i_sb,
1284 reiserfs_claim_blocks_to_be_allocated(inode->i_sb, num_pages << (PAGE_CACHE_SHIFT - inode->i_blkbits)); 1463 (num_pages <<
1285 reiserfs_write_unlock(inode->i_sb); 1464 (PAGE_CACHE_SHIFT -
1465 inode->i_sb->
1466 s_blocksize_bits)) -
1467 blocks_to_allocate);
1468
1469 if (blocks_to_allocate > 0) { /*We only allocate blocks if we need to */
1470 /* Fill in all the possible holes and append the file if needed */
1471 res =
1472 reiserfs_allocate_blocks_for_region(&th, inode, pos,
1473 num_pages,
1474 write_bytes,
1475 prepared_pages,
1476 blocks_to_allocate);
1477 }
1286 1478
1287 if ( !num_pages ) { /* If we do not have enough space even for a single page... */ 1479 /* well, we have allocated the blocks, so it is time to free
1288 if ( pos > inode->i_size+inode->i_sb->s_blocksize-(pos & (inode->i_sb->s_blocksize-1))) { 1480 the reservation we made earlier. */
1289 res = -ENOSPC; 1481 reiserfs_release_claimed_blocks(inode->i_sb,
1290 break; // In case we are writing past the end of the last file block, break. 1482 blocks_to_allocate);
1291 } 1483 if (res) {
1292 // Otherwise we are possibly overwriting the file, so 1484 reiserfs_unprepare_pages(prepared_pages, num_pages);
1293 // let's set write size to be equal or less than blocksize. 1485 break;
1294 // This way we get it correctly for file holes. 1486 }
1295 // But overwriting files on absolutelly full volumes would not
1296 // be very efficient. Well, people are not supposed to fill
1297 // 100% of disk space anyway.
1298 write_bytes = min_t(size_t, count, inode->i_sb->s_blocksize - (pos & (inode->i_sb->s_blocksize - 1)));
1299 num_pages = 1;
1300 // No blocks were claimed before, so do it now.
1301 reiserfs_claim_blocks_to_be_allocated(inode->i_sb, 1 << (PAGE_CACHE_SHIFT - inode->i_blkbits));
1302 }
1303 1487
1304 /* Prepare for writing into the region, read in all the 1488/* NOTE that allocating blocks and filling blocks can be done in reverse order
1305 partially overwritten pages, if needed. And lock the pages, 1489 and probably we would do that just to get rid of garbage in files after a
1306 so that nobody else can access these until we are done. 1490 crash */
1307 We get number of actual blocks needed as a result.*/
1308 blocks_to_allocate = reiserfs_prepare_file_region_for_write(inode, pos, num_pages, write_bytes, prepared_pages);
1309 if ( blocks_to_allocate < 0 ) {
1310 res = blocks_to_allocate;
1311 reiserfs_release_claimed_blocks(inode->i_sb, num_pages << (PAGE_CACHE_SHIFT - inode->i_blkbits));
1312 break;
1313 }
1314 1491
1315 /* First we correct our estimate of how many blocks we need */ 1492 /* Copy data from user-supplied buffer to file's pages */
1316 reiserfs_release_claimed_blocks(inode->i_sb, (num_pages << (PAGE_CACHE_SHIFT - inode->i_sb->s_blocksize_bits)) - blocks_to_allocate ); 1493 res =
1494 reiserfs_copy_from_user_to_file_region(pos, num_pages,
1495 write_bytes,
1496 prepared_pages, buf);
1497 if (res) {
1498 reiserfs_unprepare_pages(prepared_pages, num_pages);
1499 break;
1500 }
1317 1501
1318 if ( blocks_to_allocate > 0) {/*We only allocate blocks if we need to*/ 1502 /* Send the pages to disk and unlock them. */
1319 /* Fill in all the possible holes and append the file if needed */ 1503 res =
1320 res = reiserfs_allocate_blocks_for_region(&th, inode, pos, num_pages, write_bytes, prepared_pages, blocks_to_allocate); 1504 reiserfs_submit_file_region_for_write(&th, inode, pos,
1505 num_pages,
1506 write_bytes,
1507 prepared_pages);
1508 if (res)
1509 break;
1510
1511 already_written += write_bytes;
1512 buf += write_bytes;
1513 *ppos = pos += write_bytes;
1514 count -= write_bytes;
1515 balance_dirty_pages_ratelimited(inode->i_mapping);
1321 } 1516 }
1322 1517
1323 /* well, we have allocated the blocks, so it is time to free 1518 /* this is only true on error */
1324 the reservation we made earlier. */ 1519 if (th.t_trans_id) {
1325 reiserfs_release_claimed_blocks(inode->i_sb, blocks_to_allocate); 1520 reiserfs_write_lock(inode->i_sb);
1326 if ( res ) { 1521 err = journal_end(&th, th.t_super, th.t_blocks_allocated);
1327 reiserfs_unprepare_pages(prepared_pages, num_pages); 1522 reiserfs_write_unlock(inode->i_sb);
1328 break; 1523 if (err) {
1524 res = err;
1525 goto out;
1526 }
1329 } 1527 }
1330 1528
1331/* NOTE that allocating blocks and filling blocks can be done in reverse order 1529 if ((file->f_flags & O_SYNC) || IS_SYNC(inode))
1332 and probably we would do that just to get rid of garbage in files after a 1530 res =
1333 crash */ 1531 generic_osync_inode(inode, file->f_mapping,
1532 OSYNC_METADATA | OSYNC_DATA);
1334 1533
1335 /* Copy data from user-supplied buffer to file's pages */ 1534 up(&inode->i_sem);
1336 res = reiserfs_copy_from_user_to_file_region(pos, num_pages, write_bytes, prepared_pages, buf); 1535 reiserfs_async_progress_wait(inode->i_sb);
1337 if ( res ) { 1536 return (already_written != 0) ? already_written : res;
1338 reiserfs_unprepare_pages(prepared_pages, num_pages);
1339 break;
1340 }
1341 1537
1342 /* Send the pages to disk and unlock them. */ 1538 out:
1343 res = reiserfs_submit_file_region_for_write(&th, inode, pos, num_pages, 1539 up(&inode->i_sem); // unlock the file on exit.
1344 write_bytes,prepared_pages); 1540 return res;
1345 if ( res )
1346 break;
1347
1348 already_written += write_bytes;
1349 buf += write_bytes;
1350 *ppos = pos += write_bytes;
1351 count -= write_bytes;
1352 balance_dirty_pages_ratelimited(inode->i_mapping);
1353 }
1354
1355 /* this is only true on error */
1356 if (th.t_trans_id) {
1357 reiserfs_write_lock(inode->i_sb);
1358 err = journal_end(&th, th.t_super, th.t_blocks_allocated);
1359 reiserfs_write_unlock(inode->i_sb);
1360 if (err) {
1361 res = err;
1362 goto out;
1363 }
1364 }
1365
1366 if ((file->f_flags & O_SYNC) || IS_SYNC(inode))
1367 res = generic_osync_inode(inode, file->f_mapping, OSYNC_METADATA|OSYNC_DATA);
1368
1369 up(&inode->i_sem);
1370 reiserfs_async_progress_wait(inode->i_sb);
1371 return (already_written != 0)?already_written:res;
1372
1373out:
1374 up(&inode->i_sem); // unlock the file on exit.
1375 return res;
1376} 1541}
1377 1542
1378static ssize_t reiserfs_aio_write(struct kiocb *iocb, const char __user *buf, 1543static ssize_t reiserfs_aio_write(struct kiocb *iocb, const char __user * buf,
1379 size_t count, loff_t pos) 1544 size_t count, loff_t pos)
1380{ 1545{
1381 return generic_file_aio_write(iocb, buf, count, pos); 1546 return generic_file_aio_write(iocb, buf, count, pos);
1382} 1547}
1383 1548
1384
1385
1386struct file_operations reiserfs_file_operations = { 1549struct file_operations reiserfs_file_operations = {
1387 .read = generic_file_read, 1550 .read = generic_file_read,
1388 .write = reiserfs_file_write, 1551 .write = reiserfs_file_write,
1389 .ioctl = reiserfs_ioctl, 1552 .ioctl = reiserfs_ioctl,
1390 .mmap = generic_file_mmap, 1553 .mmap = generic_file_mmap,
1391 .release = reiserfs_file_release, 1554 .release = reiserfs_file_release,
1392 .fsync = reiserfs_sync_file, 1555 .fsync = reiserfs_sync_file,
1393 .sendfile = generic_file_sendfile, 1556 .sendfile = generic_file_sendfile,
1394 .aio_read = generic_file_aio_read, 1557 .aio_read = generic_file_aio_read,
1395 .aio_write = reiserfs_aio_write, 1558 .aio_write = reiserfs_aio_write,
1396}; 1559};
1397 1560
1398 1561struct inode_operations reiserfs_file_inode_operations = {
1399struct inode_operations reiserfs_file_inode_operations = { 1562 .truncate = reiserfs_vfs_truncate_file,
1400 .truncate = reiserfs_vfs_truncate_file, 1563 .setattr = reiserfs_setattr,
1401 .setattr = reiserfs_setattr, 1564 .setxattr = reiserfs_setxattr,
1402 .setxattr = reiserfs_setxattr, 1565 .getxattr = reiserfs_getxattr,
1403 .getxattr = reiserfs_getxattr, 1566 .listxattr = reiserfs_listxattr,
1404 .listxattr = reiserfs_listxattr, 1567 .removexattr = reiserfs_removexattr,
1405 .removexattr = reiserfs_removexattr, 1568 .permission = reiserfs_permission,
1406 .permission = reiserfs_permission,
1407}; 1569};
1408
1409
generated by cgit v1.2.2 (git 2.25.0) at 2024-11-12 09:54:14 -0500