[PATCH] Introduce sys_splice() system call
[linux-2.6.git] / fs / reiserfs / file.c
1 /*
2  * Copyright 2000 by Hans Reiser, licensing governed by reiserfs/README
3  */
4
5 #include <linux/time.h>
6 #include <linux/reiserfs_fs.h>
7 #include <linux/reiserfs_acl.h>
8 #include <linux/reiserfs_xattr.h>
9 #include <linux/smp_lock.h>
10 #include <asm/uaccess.h>
11 #include <linux/pagemap.h>
12 #include <linux/swap.h>
13 #include <linux/writeback.h>
14 #include <linux/blkdev.h>
15 #include <linux/buffer_head.h>
16 #include <linux/quotaops.h>
17
18 /*
19 ** We pack the tails of files on file close, not at the time they are written.
20 ** This implies an unnecessary copy of the tail and an unnecessary indirect item
21 ** insertion/balancing, for files that are written in one write.
22 ** It avoids unnecessary tail packings (balances) for files that are written in
23 ** multiple writes and are small enough to have tails.
24 ** 
25 ** file_release is called by the VFS layer when the file is closed.  If
26 ** this is the last open file descriptor, and the file
27 ** small enough to have a tail, and the tail is currently in an
28 ** unformatted node, the tail is converted back into a direct item.
29 ** 
30 ** We use reiserfs_truncate_file to pack the tail, since it already has
31 ** all the conditions coded.  
32 */
33 static int reiserfs_file_release(struct inode *inode, struct file *filp)
34 {
35
36         struct reiserfs_transaction_handle th;
37         int err;
38         int jbegin_failure = 0;
39
40         if (!S_ISREG(inode->i_mode))
41                 BUG();
42
43         /* fast out for when nothing needs to be done */
44         if ((atomic_read(&inode->i_count) > 1 ||
45              !(REISERFS_I(inode)->i_flags & i_pack_on_close_mask) ||
46              !tail_has_to_be_packed(inode)) &&
47             REISERFS_I(inode)->i_prealloc_count <= 0) {
48                 return 0;
49         }
50
51         reiserfs_write_lock(inode->i_sb);
52         mutex_lock(&inode->i_mutex);
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);
59         if (err) {
60                 /* uh oh, we can't allow the inode to go away while there
61                  * is still preallocation blocks pending.  Try to join the
62                  * aborted transaction
63                  */
64                 jbegin_failure = err;
65                 err = journal_join_abort(&th, inode->i_sb, 1);
66
67                 if (err) {
68                         /* hmpf, our choices here aren't good.  We can pin the inode
69                          * which will disallow unmount from every happening, we can
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                 }
81         }
82         reiserfs_update_inode_transaction(inode);
83
84 #ifdef REISERFS_PREALLOCATE
85         reiserfs_discard_prealloc(&th, inode);
86 #endif
87         err = journal_end(&th, inode->i_sb, 1);
88
89         /* copy back the error code from journal_begin */
90         if (!err)
91                 err = jbegin_failure;
92
93         if (!err && atomic_read(&inode->i_count) <= 1 &&
94             (REISERFS_I(inode)->i_flags & i_pack_on_close_mask) &&
95             tail_has_to_be_packed(inode)) {
96                 /* if regular file is released by last holder and it has been
97                    appended (we append by unformatted node only) or its direct
98                    item(s) had to be converted, then it may have to be
99                    indirect2direct converted */
100                 err = reiserfs_truncate_file(inode, 0);
101         }
102       out:
103         mutex_unlock(&inode->i_mutex);
104         reiserfs_write_unlock(inode->i_sb);
105         return err;
106 }
107
108 static void reiserfs_vfs_truncate_file(struct inode *inode)
109 {
110         reiserfs_truncate_file(inode, 1);
111 }
112
113 /* Sync a reiserfs file. */
114
115 /*
116  * FIXME: sync_mapping_buffers() never has anything to sync.  Can
117  * be removed...
118  */
119
120 static int reiserfs_sync_file(struct file *p_s_filp,
121                               struct dentry *p_s_dentry, int datasync)
122 {
123         struct inode *p_s_inode = p_s_dentry->d_inode;
124         int n_err;
125         int barrier_done;
126
127         if (!S_ISREG(p_s_inode->i_mode))
128                 BUG();
129         n_err = sync_mapping_buffers(p_s_inode->i_mapping);
130         reiserfs_write_lock(p_s_inode->i_sb);
131         barrier_done = reiserfs_commit_for_inode(p_s_inode);
132         reiserfs_write_unlock(p_s_inode->i_sb);
133         if (barrier_done != 1)
134                 blkdev_issue_flush(p_s_inode->i_sb->s_bdev, NULL);
135         if (barrier_done < 0)
136                 return barrier_done;
137         return (n_err < 0) ? -EIO : 0;
138 }
139
140 /* I really do not want to play with memory shortage right now, so
141    to simplify the code, we are not going to write more than this much pages at
142    a time. This still should considerably improve performance compared to 4k
143    at a time case. This is 32 pages of 4k size. */
144 #define REISERFS_WRITE_PAGES_AT_A_TIME (128 * 1024) / PAGE_CACHE_SIZE
145
146 /* Allocates blocks for a file to fulfil write request.
147    Maps all unmapped but prepared pages from the list.
148    Updates metadata with newly allocated blocknumbers as needed */
149 static int reiserfs_allocate_blocks_for_region(struct reiserfs_transaction_handle *th, struct inode *inode,     /* Inode we work with */
150                                                loff_t pos,      /* Writing position */
151                                                int num_pages,   /* number of pages write going
152                                                                    to touch */
153                                                int write_bytes, /* amount of bytes to write */
154                                                struct page **prepared_pages,    /* array of
155                                                                                    prepared pages
156                                                                                  */
157                                                int blocks_to_allocate   /* Amount of blocks we
158                                                                            need to allocate to
159                                                                            fit the data into file
160                                                                          */
161     )
162 {
163         struct cpu_key key;     // cpu key of item that we are going to deal with
164         struct item_head *ih;   // pointer to item head that we are going to deal with
165         struct buffer_head *bh; // Buffer head that contains items that we are going to deal with
166         __le32 *item;           // pointer to item we are going to deal with
167         INITIALIZE_PATH(path);  // path to item, that we are going to deal with.
168         b_blocknr_t *allocated_blocks;  // Pointer to a place where allocated blocknumbers would be stored.
169         reiserfs_blocknr_hint_t hint;   // hint structure for block allocator.
170         size_t res;             // return value of various functions that we call.
171         int curr_block;         // current block used to keep track of unmapped blocks.
172         int i;                  // loop counter
173         int itempos;            // position in item
174         unsigned int from = (pos & (PAGE_CACHE_SIZE - 1));      // writing position in
175         // first page
176         unsigned int to = ((pos + write_bytes - 1) & (PAGE_CACHE_SIZE - 1)) + 1;        /* last modified byte offset in last page */
177         __u64 hole_size;        // amount of blocks for a file hole, if it needed to be created.
178         int modifying_this_item = 0;    // Flag for items traversal code to keep track
179         // of the fact that we already prepared
180         // current block for journal
181         int will_prealloc = 0;
182         RFALSE(!blocks_to_allocate,
183                "green-9004: tried to allocate zero blocks?");
184
185         /* only preallocate if this is a small write */
186         if (REISERFS_I(inode)->i_prealloc_count ||
187             (!(write_bytes & (inode->i_sb->s_blocksize - 1)) &&
188              blocks_to_allocate <
189              REISERFS_SB(inode->i_sb)->s_alloc_options.preallocsize))
190                 will_prealloc =
191                     REISERFS_SB(inode->i_sb)->s_alloc_options.preallocsize;
192
193         allocated_blocks = kmalloc((blocks_to_allocate + will_prealloc) *
194                                    sizeof(b_blocknr_t), GFP_NOFS);
195         if (!allocated_blocks)
196                 return -ENOMEM;
197
198         /* First we compose a key to point at the writing position, we want to do
199            that outside of any locking region. */
200         make_cpu_key(&key, inode, pos + 1, TYPE_ANY, 3 /*key length */ );
201
202         /* If we came here, it means we absolutely need to open a transaction,
203            since we need to allocate some blocks */
204         reiserfs_write_lock(inode->i_sb);       // Journaling stuff and we need that.
205         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
206         if (res)
207                 goto error_exit;
208         reiserfs_update_inode_transaction(inode);
209
210         /* Look for the in-tree position of our write, need path for block allocator */
211         res = search_for_position_by_key(inode->i_sb, &key, &path);
212         if (res == IO_ERROR) {
213                 res = -EIO;
214                 goto error_exit;
215         }
216
217         /* Allocate blocks */
218         /* First fill in "hint" structure for block allocator */
219         hint.th = th;           // transaction handle.
220         hint.path = &path;      // Path, so that block allocator can determine packing locality or whatever it needs to determine.
221         hint.inode = inode;     // Inode is needed by block allocator too.
222         hint.search_start = 0;  // We have no hint on where to search free blocks for block allocator.
223         hint.key = key.on_disk_key;     // on disk key of file.
224         hint.block = inode->i_blocks >> (inode->i_sb->s_blocksize_bits - 9);    // Number of disk blocks this file occupies already.
225         hint.formatted_node = 0;        // We are allocating blocks for unformatted node.
226         hint.preallocate = will_prealloc;
227
228         /* Call block allocator to allocate blocks */
229         res =
230             reiserfs_allocate_blocknrs(&hint, allocated_blocks,
231                                        blocks_to_allocate, blocks_to_allocate);
232         if (res != CARRY_ON) {
233                 if (res == NO_DISK_SPACE) {
234                         /* We flush the transaction in case of no space. This way some
235                            blocks might become free */
236                         SB_JOURNAL(inode->i_sb)->j_must_wait = 1;
237                         res = restart_transaction(th, inode, &path);
238                         if (res)
239                                 goto error_exit;
240
241                         /* We might have scheduled, so search again */
242                         res =
243                             search_for_position_by_key(inode->i_sb, &key,
244                                                        &path);
245                         if (res == IO_ERROR) {
246                                 res = -EIO;
247                                 goto error_exit;
248                         }
249
250                         /* update changed info for hint structure. */
251                         res =
252                             reiserfs_allocate_blocknrs(&hint, allocated_blocks,
253                                                        blocks_to_allocate,
254                                                        blocks_to_allocate);
255                         if (res != CARRY_ON) {
256                                 res = res == QUOTA_EXCEEDED ? -EDQUOT : -ENOSPC;
257                                 pathrelse(&path);
258                                 goto error_exit;
259                         }
260                 } else {
261                         res = res == QUOTA_EXCEEDED ? -EDQUOT : -ENOSPC;
262                         pathrelse(&path);
263                         goto error_exit;
264                 }
265         }
266 #ifdef __BIG_ENDIAN
267         // Too bad, I have not found any way to convert a given region from
268         // cpu format to little endian format
269         {
270                 int i;
271                 for (i = 0; i < blocks_to_allocate; i++)
272                         allocated_blocks[i] = cpu_to_le32(allocated_blocks[i]);
273         }
274 #endif
275
276         /* Blocks allocating well might have scheduled and tree might have changed,
277            let's search the tree again */
278         /* find where in the tree our write should go */
279         res = search_for_position_by_key(inode->i_sb, &key, &path);
280         if (res == IO_ERROR) {
281                 res = -EIO;
282                 goto error_exit_free_blocks;
283         }
284
285         bh = get_last_bh(&path);        // Get a bufferhead for last element in path.
286         ih = get_ih(&path);     // Get a pointer to last item head in path.
287         item = get_item(&path); // Get a pointer to last item in path
288
289         /* Let's see what we have found */
290         if (res != POSITION_FOUND) {    /* position not found, this means that we
291                                            might need to append file with holes
292                                            first */
293                 // Since we are writing past the file's end, we need to find out if
294                 // there is a hole that needs to be inserted before our writing
295                 // position, and how many blocks it is going to cover (we need to
296                 //  populate pointers to file blocks representing the hole with zeros)
297
298                 {
299                         int item_offset = 1;
300                         /*
301                          * if ih is stat data, its offset is 0 and we don't want to
302                          * add 1 to pos in the hole_size calculation
303                          */
304                         if (is_statdata_le_ih(ih))
305                                 item_offset = 0;
306                         hole_size = (pos + item_offset -
307                                      (le_key_k_offset
308                                       (get_inode_item_key_version(inode),
309                                        &(ih->ih_key)) + op_bytes_number(ih,
310                                                                         inode->
311                                                                         i_sb->
312                                                                         s_blocksize)))
313                             >> inode->i_sb->s_blocksize_bits;
314                 }
315
316                 if (hole_size > 0) {
317                         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.
318                         /* area filled with zeroes, to supply as list of zero blocknumbers
319                            We allocate it outside of loop just in case loop would spin for
320                            several iterations. */
321                         char *zeros = kmalloc(to_paste * UNFM_P_SIZE, GFP_ATOMIC);      // We cannot insert more than MAX_ITEM_LEN bytes anyway.
322                         if (!zeros) {
323                                 res = -ENOMEM;
324                                 goto error_exit_free_blocks;
325                         }
326                         memset(zeros, 0, to_paste * UNFM_P_SIZE);
327                         do {
328                                 to_paste =
329                                     min_t(__u64, hole_size,
330                                           MAX_ITEM_LEN(inode->i_sb->
331                                                        s_blocksize) /
332                                           UNFM_P_SIZE);
333                                 if (is_indirect_le_ih(ih)) {
334                                         /* Ok, there is existing indirect item already. Need to append it */
335                                         /* Calculate position past inserted item */
336                                         make_cpu_key(&key, inode,
337                                                      le_key_k_offset
338                                                      (get_inode_item_key_version
339                                                       (inode),
340                                                       &(ih->ih_key)) +
341                                                      op_bytes_number(ih,
342                                                                      inode->
343                                                                      i_sb->
344                                                                      s_blocksize),
345                                                      TYPE_INDIRECT, 3);
346                                         res =
347                                             reiserfs_paste_into_item(th, &path,
348                                                                      &key,
349                                                                      inode,
350                                                                      (char *)
351                                                                      zeros,
352                                                                      UNFM_P_SIZE
353                                                                      *
354                                                                      to_paste);
355                                         if (res) {
356                                                 kfree(zeros);
357                                                 goto error_exit_free_blocks;
358                                         }
359                                 } else if (is_statdata_le_ih(ih)) {
360                                         /* No existing item, create it */
361                                         /* item head for new item */
362                                         struct item_head ins_ih;
363
364                                         /* create a key for our new item */
365                                         make_cpu_key(&key, inode, 1,
366                                                      TYPE_INDIRECT, 3);
367
368                                         /* Create new item head for our new item */
369                                         make_le_item_head(&ins_ih, &key,
370                                                           key.version, 1,
371                                                           TYPE_INDIRECT,
372                                                           to_paste *
373                                                           UNFM_P_SIZE,
374                                                           0 /* free space */ );
375
376                                         /* Find where such item should live in the tree */
377                                         res =
378                                             search_item(inode->i_sb, &key,
379                                                         &path);
380                                         if (res != ITEM_NOT_FOUND) {
381                                                 /* item should not exist, otherwise we have error */
382                                                 if (res != -ENOSPC) {
383                                                         reiserfs_warning(inode->
384                                                                          i_sb,
385                                                                          "green-9008: search_by_key (%K) returned %d",
386                                                                          &key,
387                                                                          res);
388                                                 }
389                                                 res = -EIO;
390                                                 kfree(zeros);
391                                                 goto error_exit_free_blocks;
392                                         }
393                                         res =
394                                             reiserfs_insert_item(th, &path,
395                                                                  &key, &ins_ih,
396                                                                  inode,
397                                                                  (char *)zeros);
398                                 } else {
399                                         reiserfs_panic(inode->i_sb,
400                                                        "green-9011: Unexpected key type %K\n",
401                                                        &key);
402                                 }
403                                 if (res) {
404                                         kfree(zeros);
405                                         goto error_exit_free_blocks;
406                                 }
407                                 /* Now we want to check if transaction is too full, and if it is
408                                    we restart it. This will also free the path. */
409                                 if (journal_transaction_should_end
410                                     (th, th->t_blocks_allocated)) {
411                                         res =
412                                             restart_transaction(th, inode,
413                                                                 &path);
414                                         if (res) {
415                                                 pathrelse(&path);
416                                                 kfree(zeros);
417                                                 goto error_exit;
418                                         }
419                                 }
420
421                                 /* Well, need to recalculate path and stuff */
422                                 set_cpu_key_k_offset(&key,
423                                                      cpu_key_k_offset(&key) +
424                                                      (to_paste << inode->
425                                                       i_blkbits));
426                                 res =
427                                     search_for_position_by_key(inode->i_sb,
428                                                                &key, &path);
429                                 if (res == IO_ERROR) {
430                                         res = -EIO;
431                                         kfree(zeros);
432                                         goto error_exit_free_blocks;
433                                 }
434                                 bh = get_last_bh(&path);
435                                 ih = get_ih(&path);
436                                 item = get_item(&path);
437                                 hole_size -= to_paste;
438                         } while (hole_size);
439                         kfree(zeros);
440                 }
441         }
442         // Go through existing indirect items first
443         // replace all zeroes with blocknumbers from list
444         // Note that if no corresponding item was found, by previous search,
445         // it means there are no existing in-tree representation for file area
446         // we are going to overwrite, so there is nothing to scan through for holes.
447         for (curr_block = 0, itempos = path.pos_in_item;
448              curr_block < blocks_to_allocate && res == POSITION_FOUND;) {
449               retry:
450
451                 if (itempos >= ih_item_len(ih) / UNFM_P_SIZE) {
452                         /* We run out of data in this indirect item, let's look for another
453                            one. */
454                         /* First if we are already modifying current item, log it */
455                         if (modifying_this_item) {
456                                 journal_mark_dirty(th, inode->i_sb, bh);
457                                 modifying_this_item = 0;
458                         }
459                         /* Then set the key to look for a new indirect item (offset of old
460                            item is added to old item length */
461                         set_cpu_key_k_offset(&key,
462                                              le_key_k_offset
463                                              (get_inode_item_key_version(inode),
464                                               &(ih->ih_key)) +
465                                              op_bytes_number(ih,
466                                                              inode->i_sb->
467                                                              s_blocksize));
468                         /* Search ofor position of new key in the tree. */
469                         res =
470                             search_for_position_by_key(inode->i_sb, &key,
471                                                        &path);
472                         if (res == IO_ERROR) {
473                                 res = -EIO;
474                                 goto error_exit_free_blocks;
475                         }
476                         bh = get_last_bh(&path);
477                         ih = get_ih(&path);
478                         item = get_item(&path);
479                         itempos = path.pos_in_item;
480                         continue;       // loop to check all kinds of conditions and so on.
481                 }
482                 /* Ok, we have correct position in item now, so let's see if it is
483                    representing file hole (blocknumber is zero) and fill it if needed */
484                 if (!item[itempos]) {
485                         /* Ok, a hole. Now we need to check if we already prepared this
486                            block to be journaled */
487                         while (!modifying_this_item) {  // loop until succeed
488                                 /* Well, this item is not journaled yet, so we must prepare
489                                    it for journal first, before we can change it */
490                                 struct item_head tmp_ih;        // We copy item head of found item,
491                                 // here to detect if fs changed under
492                                 // us while we were preparing for
493                                 // journal.
494                                 int fs_gen;     // We store fs generation here to find if someone
495                                 // changes fs under our feet
496
497                                 copy_item_head(&tmp_ih, ih);    // Remember itemhead
498                                 fs_gen = get_generation(inode->i_sb);   // remember fs generation
499                                 reiserfs_prepare_for_journal(inode->i_sb, bh, 1);       // Prepare a buffer within which indirect item is stored for changing.
500                                 if (fs_changed(fs_gen, inode->i_sb)
501                                     && item_moved(&tmp_ih, &path)) {
502                                         // Sigh, fs was changed under us, we need to look for new
503                                         // location of item we are working with
504
505                                         /* unmark prepaerd area as journaled and search for it's
506                                            new position */
507                                         reiserfs_restore_prepared_buffer(inode->
508                                                                          i_sb,
509                                                                          bh);
510                                         res =
511                                             search_for_position_by_key(inode->
512                                                                        i_sb,
513                                                                        &key,
514                                                                        &path);
515                                         if (res == IO_ERROR) {
516                                                 res = -EIO;
517                                                 goto error_exit_free_blocks;
518                                         }
519                                         bh = get_last_bh(&path);
520                                         ih = get_ih(&path);
521                                         item = get_item(&path);
522                                         itempos = path.pos_in_item;
523                                         goto retry;
524                                 }
525                                 modifying_this_item = 1;
526                         }
527                         item[itempos] = allocated_blocks[curr_block];   // Assign new block
528                         curr_block++;
529                 }
530                 itempos++;
531         }
532
533         if (modifying_this_item) {      // We need to log last-accessed block, if it
534                 // was modified, but not logged yet.
535                 journal_mark_dirty(th, inode->i_sb, bh);
536         }
537
538         if (curr_block < blocks_to_allocate) {
539                 // Oh, well need to append to indirect item, or to create indirect item
540                 // if there weren't any
541                 if (is_indirect_le_ih(ih)) {
542                         // Existing indirect item - append. First calculate key for append
543                         // position. We do not need to recalculate path as it should
544                         // already point to correct place.
545                         make_cpu_key(&key, inode,
546                                      le_key_k_offset(get_inode_item_key_version
547                                                      (inode),
548                                                      &(ih->ih_key)) +
549                                      op_bytes_number(ih,
550                                                      inode->i_sb->s_blocksize),
551                                      TYPE_INDIRECT, 3);
552                         res =
553                             reiserfs_paste_into_item(th, &path, &key, inode,
554                                                      (char *)(allocated_blocks +
555                                                               curr_block),
556                                                      UNFM_P_SIZE *
557                                                      (blocks_to_allocate -
558                                                       curr_block));
559                         if (res) {
560                                 goto error_exit_free_blocks;
561                         }
562                 } else if (is_statdata_le_ih(ih)) {
563                         // Last found item was statdata. That means we need to create indirect item.
564                         struct item_head ins_ih;        /* itemhead for new item */
565
566                         /* create a key for our new item */
567                         make_cpu_key(&key, inode, 1, TYPE_INDIRECT, 3); // Position one,
568                         // because that's
569                         // where first
570                         // indirect item
571                         // begins
572                         /* Create new item head for our new item */
573                         make_le_item_head(&ins_ih, &key, key.version, 1,
574                                           TYPE_INDIRECT,
575                                           (blocks_to_allocate -
576                                            curr_block) * UNFM_P_SIZE,
577                                           0 /* free space */ );
578                         /* Find where such item should live in the tree */
579                         res = search_item(inode->i_sb, &key, &path);
580                         if (res != ITEM_NOT_FOUND) {
581                                 /* Well, if we have found such item already, or some error
582                                    occured, we need to warn user and return error */
583                                 if (res != -ENOSPC) {
584                                         reiserfs_warning(inode->i_sb,
585                                                          "green-9009: search_by_key (%K) "
586                                                          "returned %d", &key,
587                                                          res);
588                                 }
589                                 res = -EIO;
590                                 goto error_exit_free_blocks;
591                         }
592                         /* Insert item into the tree with the data as its body */
593                         res =
594                             reiserfs_insert_item(th, &path, &key, &ins_ih,
595                                                  inode,
596                                                  (char *)(allocated_blocks +
597                                                           curr_block));
598                 } else {
599                         reiserfs_panic(inode->i_sb,
600                                        "green-9010: unexpected item type for key %K\n",
601                                        &key);
602                 }
603         }
604         // the caller is responsible for closing the transaction
605         // unless we return an error, they are also responsible for logging
606         // the inode.
607         //
608         pathrelse(&path);
609         /*
610          * cleanup prellocation from previous writes
611          * if this is a partial block write
612          */
613         if (write_bytes & (inode->i_sb->s_blocksize - 1))
614                 reiserfs_discard_prealloc(th, inode);
615         reiserfs_write_unlock(inode->i_sb);
616
617         // go through all the pages/buffers and map the buffers to newly allocated
618         // blocks (so that system knows where to write these pages later).
619         curr_block = 0;
620         for (i = 0; i < num_pages; i++) {
621                 struct page *page = prepared_pages[i];  //current page
622                 struct buffer_head *head = page_buffers(page);  // first buffer for a page
623                 int block_start, block_end;     // in-page offsets for buffers.
624
625                 if (!page_buffers(page))
626                         reiserfs_panic(inode->i_sb,
627                                        "green-9005: No buffers for prepared page???");
628
629                 /* For each buffer in page */
630                 for (bh = head, block_start = 0; bh != head || !block_start;
631                      block_start = block_end, bh = bh->b_this_page) {
632                         if (!bh)
633                                 reiserfs_panic(inode->i_sb,
634                                                "green-9006: Allocated but absent buffer for a page?");
635                         block_end = block_start + inode->i_sb->s_blocksize;
636                         if (i == 0 && block_end <= from)
637                                 /* if this buffer is before requested data to map, skip it */
638                                 continue;
639                         if (i == num_pages - 1 && block_start >= to)
640                                 /* If this buffer is after requested data to map, abort
641                                    processing of current page */
642                                 break;
643
644                         if (!buffer_mapped(bh)) {       // Ok, unmapped buffer, need to map it
645                                 map_bh(bh, inode->i_sb,
646                                        le32_to_cpu(allocated_blocks
647                                                    [curr_block]));
648                                 curr_block++;
649                                 set_buffer_new(bh);
650                         }
651                 }
652         }
653
654         RFALSE(curr_block > blocks_to_allocate,
655                "green-9007: Used too many blocks? weird");
656
657         kfree(allocated_blocks);
658         return 0;
659
660 // Need to deal with transaction here.
661       error_exit_free_blocks:
662         pathrelse(&path);
663         // free blocks
664         for (i = 0; i < blocks_to_allocate; i++)
665                 reiserfs_free_block(th, inode, le32_to_cpu(allocated_blocks[i]),
666                                     1);
667
668       error_exit:
669         if (th->t_trans_id) {
670                 int err;
671                 // update any changes we made to blk count
672                 mark_inode_dirty(inode);
673                 err =
674                     journal_end(th, inode->i_sb,
675                                 JOURNAL_PER_BALANCE_CNT * 3 + 1 +
676                                 2 * REISERFS_QUOTA_TRANS_BLOCKS(inode->i_sb));
677                 if (err)
678                         res = err;
679         }
680         reiserfs_write_unlock(inode->i_sb);
681         kfree(allocated_blocks);
682
683         return res;
684 }
685
686 /* Unlock pages prepared by reiserfs_prepare_file_region_for_write */
687 static void reiserfs_unprepare_pages(struct page **prepared_pages,      /* list of locked pages */
688                                      size_t num_pages /* amount of pages */ )
689 {
690         int i;                  // loop counter
691
692         for (i = 0; i < num_pages; i++) {
693                 struct page *page = prepared_pages[i];
694
695                 try_to_free_buffers(page);
696                 unlock_page(page);
697                 page_cache_release(page);
698         }
699 }
700
701 /* This function will copy data from userspace to specified pages within
702    supplied byte range */
703 static int reiserfs_copy_from_user_to_file_region(loff_t pos,   /* In-file position */
704                                                   int num_pages,        /* Number of pages affected */
705                                                   int write_bytes,      /* Amount of bytes to write */
706                                                   struct page **prepared_pages, /* pointer to 
707                                                                                    array to
708                                                                                    prepared pages
709                                                                                  */
710                                                   const char __user * buf       /* Pointer to user-supplied
711                                                                                    data */
712     )
713 {
714         long page_fault = 0;    // status of copy_from_user.
715         int i;                  // loop counter.
716         int offset;             // offset in page
717
718         for (i = 0, offset = (pos & (PAGE_CACHE_SIZE - 1)); i < num_pages;
719              i++, offset = 0) {
720                 size_t count = min_t(size_t, PAGE_CACHE_SIZE - offset, write_bytes);    // How much of bytes to write to this page
721                 struct page *page = prepared_pages[i];  // Current page we process.
722
723                 fault_in_pages_readable(buf, count);
724
725                 /* Copy data from userspace to the current page */
726                 kmap(page);
727                 page_fault = __copy_from_user(page_address(page) + offset, buf, count); // Copy the data.
728                 /* Flush processor's dcache for this page */
729                 flush_dcache_page(page);
730                 kunmap(page);
731                 buf += count;
732                 write_bytes -= count;
733
734                 if (page_fault)
735                         break;  // Was there a fault? abort.
736         }
737
738         return page_fault ? -EFAULT : 0;
739 }
740
741 /* taken fs/buffer.c:__block_commit_write */
742 int reiserfs_commit_page(struct inode *inode, struct page *page,
743                          unsigned from, unsigned to)
744 {
745         unsigned block_start, block_end;
746         int partial = 0;
747         unsigned blocksize;
748         struct buffer_head *bh, *head;
749         unsigned long i_size_index = inode->i_size >> PAGE_CACHE_SHIFT;
750         int new;
751         int logit = reiserfs_file_data_log(inode);
752         struct super_block *s = inode->i_sb;
753         int bh_per_page = PAGE_CACHE_SIZE / s->s_blocksize;
754         struct reiserfs_transaction_handle th;
755         int ret = 0;
756
757         th.t_trans_id = 0;
758         blocksize = 1 << inode->i_blkbits;
759
760         if (logit) {
761                 reiserfs_write_lock(s);
762                 ret = journal_begin(&th, s, bh_per_page + 1);
763                 if (ret)
764                         goto drop_write_lock;
765                 reiserfs_update_inode_transaction(inode);
766         }
767         for (bh = head = page_buffers(page), block_start = 0;
768              bh != head || !block_start;
769              block_start = block_end, bh = bh->b_this_page) {
770
771                 new = buffer_new(bh);
772                 clear_buffer_new(bh);
773                 block_end = block_start + blocksize;
774                 if (block_end <= from || block_start >= to) {
775                         if (!buffer_uptodate(bh))
776                                 partial = 1;
777                 } else {
778                         set_buffer_uptodate(bh);
779                         if (logit) {
780                                 reiserfs_prepare_for_journal(s, bh, 1);
781                                 journal_mark_dirty(&th, s, bh);
782                         } else if (!buffer_dirty(bh)) {
783                                 mark_buffer_dirty(bh);
784                                 /* do data=ordered on any page past the end
785                                  * of file and any buffer marked BH_New.
786                                  */
787                                 if (reiserfs_data_ordered(inode->i_sb) &&
788                                     (new || page->index >= i_size_index)) {
789                                         reiserfs_add_ordered_list(inode, bh);
790                                 }
791                         }
792                 }
793         }
794         if (logit) {
795                 ret = journal_end(&th, s, bh_per_page + 1);
796               drop_write_lock:
797                 reiserfs_write_unlock(s);
798         }
799         /*
800          * If this is a partial write which happened to make all buffers
801          * uptodate then we can optimize away a bogus readpage() for
802          * the next read(). Here we 'discover' whether the page went
803          * uptodate as a result of this (potentially partial) write.
804          */
805         if (!partial)
806                 SetPageUptodate(page);
807         return ret;
808 }
809
810 /* Submit pages for write. This was separated from actual file copying
811    because we might want to allocate block numbers in-between.
812    This function assumes that caller will adjust file size to correct value. */
813 static int reiserfs_submit_file_region_for_write(struct reiserfs_transaction_handle *th, struct inode *inode, loff_t pos,       /* Writing position offset */
814                                                  size_t num_pages,      /* Number of pages to write */
815                                                  size_t write_bytes,    /* number of bytes to write */
816                                                  struct page **prepared_pages   /* list of pages */
817     )
818 {
819         int status;             // return status of block_commit_write.
820         int retval = 0;         // Return value we are going to return.
821         int i;                  // loop counter
822         int offset;             // Writing offset in page.
823         int orig_write_bytes = write_bytes;
824         int sd_update = 0;
825
826         for (i = 0, offset = (pos & (PAGE_CACHE_SIZE - 1)); i < num_pages;
827              i++, offset = 0) {
828                 int count = min_t(int, PAGE_CACHE_SIZE - offset, write_bytes);  // How much of bytes to write to this page
829                 struct page *page = prepared_pages[i];  // Current page we process.
830
831                 status =
832                     reiserfs_commit_page(inode, page, offset, offset + count);
833                 if (status)
834                         retval = status;        // To not overcomplicate matters We are going to
835                 // submit all the pages even if there was error.
836                 // we only remember error status to report it on
837                 // exit.
838                 write_bytes -= count;
839         }
840         /* now that we've gotten all the ordered buffers marked dirty,
841          * we can safely update i_size and close any running transaction
842          */
843         if (pos + orig_write_bytes > inode->i_size) {
844                 inode->i_size = pos + orig_write_bytes; // Set new size
845                 /* If the file have grown so much that tail packing is no
846                  * longer possible, reset "need to pack" flag */
847                 if ((have_large_tails(inode->i_sb) &&
848                      inode->i_size > i_block_size(inode) * 4) ||
849                     (have_small_tails(inode->i_sb) &&
850                      inode->i_size > i_block_size(inode)))
851                         REISERFS_I(inode)->i_flags &= ~i_pack_on_close_mask;
852                 else if ((have_large_tails(inode->i_sb) &&
853                           inode->i_size < i_block_size(inode) * 4) ||
854                          (have_small_tails(inode->i_sb) &&
855                           inode->i_size < i_block_size(inode)))
856                         REISERFS_I(inode)->i_flags |= i_pack_on_close_mask;
857
858                 if (th->t_trans_id) {
859                         reiserfs_write_lock(inode->i_sb);
860                         // this sets the proper flags for O_SYNC to trigger a commit
861                         mark_inode_dirty(inode);
862                         reiserfs_write_unlock(inode->i_sb);
863                 } else
864                         mark_inode_dirty(inode);
865
866                 sd_update = 1;
867         }
868         if (th->t_trans_id) {
869                 reiserfs_write_lock(inode->i_sb);
870                 if (!sd_update)
871                         mark_inode_dirty(inode);
872                 status = journal_end(th, th->t_super, th->t_blocks_allocated);
873                 if (status)
874                         retval = status;
875                 reiserfs_write_unlock(inode->i_sb);
876         }
877         th->t_trans_id = 0;
878
879         /* 
880          * we have to unlock the pages after updating i_size, otherwise
881          * we race with writepage
882          */
883         for (i = 0; i < num_pages; i++) {
884                 struct page *page = prepared_pages[i];
885                 unlock_page(page);
886                 mark_page_accessed(page);
887                 page_cache_release(page);
888         }
889         return retval;
890 }
891
892 /* Look if passed writing region is going to touch file's tail
893    (if it is present). And if it is, convert the tail to unformatted node */
894 static int reiserfs_check_for_tail_and_convert(struct inode *inode,     /* inode to deal with */
895                                                loff_t pos,      /* Writing position */
896                                                int write_bytes  /* amount of bytes to write */
897     )
898 {
899         INITIALIZE_PATH(path);  // needed for search_for_position
900         struct cpu_key key;     // Key that would represent last touched writing byte.
901         struct item_head *ih;   // item header of found block;
902         int res;                // Return value of various functions we call.
903         int cont_expand_offset; // We will put offset for generic_cont_expand here
904         // This can be int just because tails are created
905         // only for small files.
906
907 /* this embodies a dependency on a particular tail policy */
908         if (inode->i_size >= inode->i_sb->s_blocksize * 4) {
909                 /* such a big files do not have tails, so we won't bother ourselves
910                    to look for tails, simply return */
911                 return 0;
912         }
913
914         reiserfs_write_lock(inode->i_sb);
915         /* find the item containing the last byte to be written, or if
916          * writing past the end of the file then the last item of the
917          * file (and then we check its type). */
918         make_cpu_key(&key, inode, pos + write_bytes + 1, TYPE_ANY,
919                      3 /*key length */ );
920         res = search_for_position_by_key(inode->i_sb, &key, &path);
921         if (res == IO_ERROR) {
922                 reiserfs_write_unlock(inode->i_sb);
923                 return -EIO;
924         }
925         ih = get_ih(&path);
926         res = 0;
927         if (is_direct_le_ih(ih)) {
928                 /* Ok, closest item is file tail (tails are stored in "direct"
929                  * items), so we need to unpack it. */
930                 /* To not overcomplicate matters, we just call generic_cont_expand
931                    which will in turn call other stuff and finally will boil down to
932                    reiserfs_get_block() that would do necessary conversion. */
933                 cont_expand_offset =
934                     le_key_k_offset(get_inode_item_key_version(inode),
935                                     &(ih->ih_key));
936                 pathrelse(&path);
937                 res = generic_cont_expand(inode, cont_expand_offset);
938         } else
939                 pathrelse(&path);
940
941         reiserfs_write_unlock(inode->i_sb);
942         return res;
943 }
944
945 /* This function locks pages starting from @pos for @inode.
946    @num_pages pages are locked and stored in
947    @prepared_pages array. Also buffers are allocated for these pages.
948    First and last page of the region is read if it is overwritten only
949    partially. If last page did not exist before write (file hole or file
950    append), it is zeroed, then. 
951    Returns number of unallocated blocks that should be allocated to cover
952    new file data.*/
953 static int reiserfs_prepare_file_region_for_write(struct inode *inode
954                                                   /* Inode of the file */ ,
955                                                   loff_t pos,   /* position in the file */
956                                                   size_t num_pages,     /* number of pages to
957                                                                            prepare */
958                                                   size_t write_bytes,   /* Amount of bytes to be
959                                                                            overwritten from
960                                                                            @pos */
961                                                   struct page **prepared_pages  /* pointer to array
962                                                                                    where to store
963                                                                                    prepared pages */
964     )
965 {
966         int res = 0;            // Return values of different functions we call.
967         unsigned long index = pos >> PAGE_CACHE_SHIFT;  // Offset in file in pages.
968         int from = (pos & (PAGE_CACHE_SIZE - 1));       // Writing offset in first page
969         int to = ((pos + write_bytes - 1) & (PAGE_CACHE_SIZE - 1)) + 1;
970         /* offset of last modified byte in last
971            page */
972         struct address_space *mapping = inode->i_mapping;       // Pages are mapped here.
973         int i;                  // Simple counter
974         int blocks = 0;         /* Return value (blocks that should be allocated) */
975         struct buffer_head *bh, *head;  // Current bufferhead and first bufferhead
976         // of a page.
977         unsigned block_start, block_end;        // Starting and ending offsets of current
978         // buffer in the page.
979         struct buffer_head *wait[2], **wait_bh = wait;  // Buffers for page, if
980         // Page appeared to be not up
981         // to date. Note how we have
982         // at most 2 buffers, this is
983         // because we at most may
984         // partially overwrite two
985         // buffers for one page. One at                                                 // the beginning of write area
986         // and one at the end.
987         // Everything inthe middle gets                                                 // overwritten totally.
988
989         struct cpu_key key;     // cpu key of item that we are going to deal with
990         struct item_head *ih = NULL;    // pointer to item head that we are going to deal with
991         struct buffer_head *itembuf = NULL;     // Buffer head that contains items that we are going to deal with
992         INITIALIZE_PATH(path);  // path to item, that we are going to deal with.
993         __le32 *item = NULL;    // pointer to item we are going to deal with
994         int item_pos = -1;      /* Position in indirect item */
995
996         if (num_pages < 1) {
997                 reiserfs_warning(inode->i_sb,
998                                  "green-9001: reiserfs_prepare_file_region_for_write "
999                                  "called with zero number of pages to process");
1000                 return -EFAULT;
1001         }
1002
1003         /* We have 2 loops for pages. In first loop we grab and lock the pages, so
1004            that nobody would touch these until we release the pages. Then
1005            we'd start to deal with mapping buffers to blocks. */
1006         for (i = 0; i < num_pages; i++) {
1007                 prepared_pages[i] = grab_cache_page(mapping, index + i);        // locks the page
1008                 if (!prepared_pages[i]) {
1009                         res = -ENOMEM;
1010                         goto failed_page_grabbing;
1011                 }
1012                 if (!page_has_buffers(prepared_pages[i]))
1013                         create_empty_buffers(prepared_pages[i],
1014                                              inode->i_sb->s_blocksize, 0);
1015         }
1016
1017         /* Let's count amount of blocks for a case where all the blocks
1018            overwritten are new (we will substract already allocated blocks later) */
1019         if (num_pages > 2)
1020                 /* These are full-overwritten pages so we count all the blocks in
1021                    these pages are counted as needed to be allocated */
1022                 blocks =
1023                     (num_pages - 2) << (PAGE_CACHE_SHIFT - inode->i_blkbits);
1024
1025         /* count blocks needed for first page (possibly partially written) */
1026         blocks += ((PAGE_CACHE_SIZE - from) >> inode->i_blkbits) + !!(from & (inode->i_sb->s_blocksize - 1));   /* roundup */
1027
1028         /* Now we account for last page. If last page == first page (we
1029            overwrite only one page), we substract all the blocks past the
1030            last writing position in a page out of already calculated number
1031            of blocks */
1032         blocks += ((num_pages > 1) << (PAGE_CACHE_SHIFT - inode->i_blkbits)) -
1033             ((PAGE_CACHE_SIZE - to) >> inode->i_blkbits);
1034         /* Note how we do not roundup here since partial blocks still
1035            should be allocated */
1036
1037         /* Now if all the write area lies past the file end, no point in
1038            maping blocks, since there is none, so we just zero out remaining
1039            parts of first and last pages in write area (if needed) */
1040         if ((pos & ~((loff_t) PAGE_CACHE_SIZE - 1)) > inode->i_size) {
1041                 if (from != 0) {        /* First page needs to be partially zeroed */
1042                         char *kaddr = kmap_atomic(prepared_pages[0], KM_USER0);
1043                         memset(kaddr, 0, from);
1044                         kunmap_atomic(kaddr, KM_USER0);
1045                 }
1046                 if (to != PAGE_CACHE_SIZE) {    /* Last page needs to be partially zeroed */
1047                         char *kaddr =
1048                             kmap_atomic(prepared_pages[num_pages - 1],
1049                                         KM_USER0);
1050                         memset(kaddr + to, 0, PAGE_CACHE_SIZE - to);
1051                         kunmap_atomic(kaddr, KM_USER0);
1052                 }
1053
1054                 /* Since all blocks are new - use already calculated value */
1055                 return blocks;
1056         }
1057
1058         /* Well, since we write somewhere into the middle of a file, there is
1059            possibility we are writing over some already allocated blocks, so
1060            let's map these blocks and substract number of such blocks out of blocks
1061            we need to allocate (calculated above) */
1062         /* Mask write position to start on blocksize, we do it out of the
1063            loop for performance reasons */
1064         pos &= ~((loff_t) inode->i_sb->s_blocksize - 1);
1065         /* Set cpu key to the starting position in a file (on left block boundary) */
1066         make_cpu_key(&key, inode,
1067                      1 + ((pos) & ~((loff_t) inode->i_sb->s_blocksize - 1)),
1068                      TYPE_ANY, 3 /*key length */ );
1069
1070         reiserfs_write_lock(inode->i_sb);       // We need that for at least search_by_key()
1071         for (i = 0; i < num_pages; i++) {
1072
1073                 head = page_buffers(prepared_pages[i]);
1074                 /* For each buffer in the page */
1075                 for (bh = head, block_start = 0; bh != head || !block_start;
1076                      block_start = block_end, bh = bh->b_this_page) {
1077                         if (!bh)
1078                                 reiserfs_panic(inode->i_sb,
1079                                                "green-9002: Allocated but absent buffer for a page?");
1080                         /* Find where this buffer ends */
1081                         block_end = block_start + inode->i_sb->s_blocksize;
1082                         if (i == 0 && block_end <= from)
1083                                 /* if this buffer is before requested data to map, skip it */
1084                                 continue;
1085
1086                         if (i == num_pages - 1 && block_start >= to) {
1087                                 /* If this buffer is after requested data to map, abort
1088                                    processing of current page */
1089                                 break;
1090                         }
1091
1092                         if (buffer_mapped(bh) && bh->b_blocknr != 0) {
1093                                 /* This is optimisation for a case where buffer is mapped
1094                                    and have blocknumber assigned. In case significant amount
1095                                    of such buffers are present, we may avoid some amount
1096                                    of search_by_key calls.
1097                                    Probably it would be possible to move parts of this code
1098                                    out of BKL, but I afraid that would overcomplicate code
1099                                    without any noticeable benefit.
1100                                  */
1101                                 item_pos++;
1102                                 /* Update the key */
1103                                 set_cpu_key_k_offset(&key,
1104                                                      cpu_key_k_offset(&key) +
1105                                                      inode->i_sb->s_blocksize);
1106                                 blocks--;       // Decrease the amount of blocks that need to be
1107                                 // allocated
1108                                 continue;       // Go to the next buffer
1109                         }
1110
1111                         if (!itembuf || /* if first iteration */
1112                             item_pos >= ih_item_len(ih) / UNFM_P_SIZE) {        /* or if we progressed past the
1113                                                                                    current unformatted_item */
1114                                 /* Try to find next item */
1115                                 res =
1116                                     search_for_position_by_key(inode->i_sb,
1117                                                                &key, &path);
1118                                 /* Abort if no more items */
1119                                 if (res != POSITION_FOUND) {
1120                                         /* make sure later loops don't use this item */
1121                                         itembuf = NULL;
1122                                         item = NULL;
1123                                         break;
1124                                 }
1125
1126                                 /* Update information about current indirect item */
1127                                 itembuf = get_last_bh(&path);
1128                                 ih = get_ih(&path);
1129                                 item = get_item(&path);
1130                                 item_pos = path.pos_in_item;
1131
1132                                 RFALSE(!is_indirect_le_ih(ih),
1133                                        "green-9003: indirect item expected");
1134                         }
1135
1136                         /* See if there is some block associated with the file
1137                            at that position, map the buffer to this block */
1138                         if (get_block_num(item, item_pos)) {
1139                                 map_bh(bh, inode->i_sb,
1140                                        get_block_num(item, item_pos));
1141                                 blocks--;       // Decrease the amount of blocks that need to be
1142                                 // allocated
1143                         }
1144                         item_pos++;
1145                         /* Update the key */
1146                         set_cpu_key_k_offset(&key,
1147                                              cpu_key_k_offset(&key) +
1148                                              inode->i_sb->s_blocksize);
1149                 }
1150         }
1151         pathrelse(&path);       // Free the path
1152         reiserfs_write_unlock(inode->i_sb);
1153
1154         /* Now zero out unmappend buffers for the first and last pages of
1155            write area or issue read requests if page is mapped. */
1156         /* First page, see if it is not uptodate */
1157         if (!PageUptodate(prepared_pages[0])) {
1158                 head = page_buffers(prepared_pages[0]);
1159
1160                 /* For each buffer in page */
1161                 for (bh = head, block_start = 0; bh != head || !block_start;
1162                      block_start = block_end, bh = bh->b_this_page) {
1163
1164                         if (!bh)
1165                                 reiserfs_panic(inode->i_sb,
1166                                                "green-9002: Allocated but absent buffer for a page?");
1167                         /* Find where this buffer ends */
1168                         block_end = block_start + inode->i_sb->s_blocksize;
1169                         if (block_end <= from)
1170                                 /* if this buffer is before requested data to map, skip it */
1171                                 continue;
1172                         if (block_start < from) {       /* Aha, our partial buffer */
1173                                 if (buffer_mapped(bh)) {        /* If it is mapped, we need to
1174                                                                    issue READ request for it to
1175                                                                    not loose data */
1176                                         ll_rw_block(READ, 1, &bh);
1177                                         *wait_bh++ = bh;
1178                                 } else {        /* Not mapped, zero it */
1179                                         char *kaddr =
1180                                             kmap_atomic(prepared_pages[0],
1181                                                         KM_USER0);
1182                                         memset(kaddr + block_start, 0,
1183                                                from - block_start);
1184                                         kunmap_atomic(kaddr, KM_USER0);
1185                                         set_buffer_uptodate(bh);
1186                                 }
1187                         }
1188                 }
1189         }
1190
1191         /* Last page, see if it is not uptodate, or if the last page is past the end of the file. */
1192         if (!PageUptodate(prepared_pages[num_pages - 1]) ||
1193             ((pos + write_bytes) >> PAGE_CACHE_SHIFT) >
1194             (inode->i_size >> PAGE_CACHE_SHIFT)) {
1195                 head = page_buffers(prepared_pages[num_pages - 1]);
1196
1197                 /* for each buffer in page */
1198                 for (bh = head, block_start = 0; bh != head || !block_start;
1199                      block_start = block_end, bh = bh->b_this_page) {
1200
1201                         if (!bh)
1202                                 reiserfs_panic(inode->i_sb,
1203                                                "green-9002: Allocated but absent buffer for a page?");
1204                         /* Find where this buffer ends */
1205                         block_end = block_start + inode->i_sb->s_blocksize;
1206                         if (block_start >= to)
1207                                 /* if this buffer is after requested data to map, skip it */
1208                                 break;
1209                         if (block_end > to) {   /* Aha, our partial buffer */
1210                                 if (buffer_mapped(bh)) {        /* If it is mapped, we need to
1211                                                                    issue READ request for it to
1212                                                                    not loose data */
1213                                         ll_rw_block(READ, 1, &bh);
1214                                         *wait_bh++ = bh;
1215                                 } else {        /* Not mapped, zero it */
1216                                         char *kaddr =
1217                                             kmap_atomic(prepared_pages
1218                                                         [num_pages - 1],
1219                                                         KM_USER0);
1220                                         memset(kaddr + to, 0, block_end - to);
1221                                         kunmap_atomic(kaddr, KM_USER0);
1222                                         set_buffer_uptodate(bh);
1223                                 }
1224                         }
1225                 }
1226         }
1227
1228         /* Wait for read requests we made to happen, if necessary */
1229         while (wait_bh > wait) {
1230                 wait_on_buffer(*--wait_bh);
1231                 if (!buffer_uptodate(*wait_bh)) {
1232                         res = -EIO;
1233                         goto failed_read;
1234                 }
1235         }
1236
1237         return blocks;
1238       failed_page_grabbing:
1239         num_pages = i;
1240       failed_read:
1241         reiserfs_unprepare_pages(prepared_pages, num_pages);
1242         return res;
1243 }
1244
1245 /* Write @count bytes at position @ppos in a file indicated by @file
1246    from the buffer @buf.  
1247
1248    generic_file_write() is only appropriate for filesystems that are not seeking to optimize performance and want
1249    something simple that works.  It is not for serious use by general purpose filesystems, excepting the one that it was
1250    written for (ext2/3).  This is for several reasons:
1251
1252    * It has no understanding of any filesystem specific optimizations.
1253
1254    * It enters the filesystem repeatedly for each page that is written.
1255
1256    * It depends on reiserfs_get_block() function which if implemented by reiserfs performs costly search_by_key
1257    * operation for each page it is supplied with. By contrast reiserfs_file_write() feeds as much as possible at a time
1258    * to reiserfs which allows for fewer tree traversals.
1259
1260    * Each indirect pointer insertion takes a lot of cpu, because it involves memory moves inside of blocks.
1261
1262    * Asking the block allocation code for blocks one at a time is slightly less efficient.
1263
1264    All of these reasons for not using only generic file write were understood back when reiserfs was first miscoded to
1265    use it, but we were in a hurry to make code freeze, and so it couldn't be revised then.  This new code should make
1266    things right finally.
1267
1268    Future Features: providing search_by_key with hints.
1269
1270 */
1271 static ssize_t reiserfs_file_write(struct file *file,   /* the file we are going to write into */
1272                                    const char __user * buf,     /*  pointer to user supplied data
1273                                                                    (in userspace) */
1274                                    size_t count,        /* amount of bytes to write */
1275                                    loff_t * ppos        /* pointer to position in file that we start writing at. Should be updated to
1276                                                          * new current position before returning. */
1277                                    )
1278 {
1279         size_t already_written = 0;     // Number of bytes already written to the file.
1280         loff_t pos;             // Current position in the file.
1281         ssize_t res;            // return value of various functions that we call.
1282         int err = 0;
1283         struct inode *inode = file->f_dentry->d_inode;  // Inode of the file that we are writing to.
1284         /* To simplify coding at this time, we store
1285            locked pages in array for now */
1286         struct page *prepared_pages[REISERFS_WRITE_PAGES_AT_A_TIME];
1287         struct reiserfs_transaction_handle th;
1288         th.t_trans_id = 0;
1289
1290         /* If a filesystem is converted from 3.5 to 3.6, we'll have v3.5 items
1291         * lying around (most of the disk, in fact). Despite the filesystem
1292         * now being a v3.6 format, the old items still can't support large
1293         * file sizes. Catch this case here, as the rest of the VFS layer is
1294         * oblivious to the different limitations between old and new items.
1295         * reiserfs_setattr catches this for truncates. This chunk is lifted
1296         * from generic_write_checks. */
1297         if (get_inode_item_key_version (inode) == KEY_FORMAT_3_5 &&
1298             *ppos + count > MAX_NON_LFS) {
1299                 if (*ppos >= MAX_NON_LFS) {
1300                         send_sig(SIGXFSZ, current, 0);
1301                         return -EFBIG;
1302                 }
1303                 if (count > MAX_NON_LFS - (unsigned long)*ppos)
1304                         count = MAX_NON_LFS - (unsigned long)*ppos;
1305         }
1306
1307         if (file->f_flags & O_DIRECT) { // Direct IO needs treatment
1308                 ssize_t result, after_file_end = 0;
1309                 if ((*ppos + count >= inode->i_size)
1310                     || (file->f_flags & O_APPEND)) {
1311                         /* If we are appending a file, we need to put this savelink in here.
1312                            If we will crash while doing direct io, finish_unfinished will
1313                            cut the garbage from the file end. */
1314                         reiserfs_write_lock(inode->i_sb);
1315                         err =
1316                             journal_begin(&th, inode->i_sb,
1317                                           JOURNAL_PER_BALANCE_CNT);
1318                         if (err) {
1319                                 reiserfs_write_unlock(inode->i_sb);
1320                                 return err;
1321                         }
1322                         reiserfs_update_inode_transaction(inode);
1323                         add_save_link(&th, inode, 1 /* Truncate */ );
1324                         after_file_end = 1;
1325                         err =
1326                             journal_end(&th, inode->i_sb,
1327                                         JOURNAL_PER_BALANCE_CNT);
1328                         reiserfs_write_unlock(inode->i_sb);
1329                         if (err)
1330                                 return err;
1331                 }
1332                 result = generic_file_write(file, buf, count, ppos);
1333
1334                 if (after_file_end) {   /* Now update i_size and remove the savelink */
1335                         struct reiserfs_transaction_handle th;
1336                         reiserfs_write_lock(inode->i_sb);
1337                         err = journal_begin(&th, inode->i_sb, 1);
1338                         if (err) {
1339                                 reiserfs_write_unlock(inode->i_sb);
1340                                 return err;
1341                         }
1342                         reiserfs_update_inode_transaction(inode);
1343                         mark_inode_dirty(inode);
1344                         err = journal_end(&th, inode->i_sb, 1);
1345                         if (err) {
1346                                 reiserfs_write_unlock(inode->i_sb);
1347                                 return err;
1348                         }
1349                         err = remove_save_link(inode, 1 /* truncate */ );
1350                         reiserfs_write_unlock(inode->i_sb);
1351                         if (err)
1352                                 return err;
1353                 }
1354
1355                 return result;
1356         }
1357
1358         if (unlikely((ssize_t) count < 0))
1359                 return -EINVAL;
1360
1361         if (unlikely(!access_ok(VERIFY_READ, buf, count)))
1362                 return -EFAULT;
1363
1364         mutex_lock(&inode->i_mutex);    // locks the entire file for just us
1365
1366         pos = *ppos;
1367
1368         /* Check if we can write to specified region of file, file
1369            is not overly big and this kind of stuff. Adjust pos and
1370            count, if needed */
1371         res = generic_write_checks(file, &pos, &count, 0);
1372         if (res)
1373                 goto out;
1374
1375         if (count == 0)
1376                 goto out;
1377
1378         res = remove_suid(file->f_dentry);
1379         if (res)
1380                 goto out;
1381
1382         file_update_time(file);
1383
1384         // Ok, we are done with all the checks.
1385
1386         // Now we should start real work
1387
1388         /* If we are going to write past the file's packed tail or if we are going
1389            to overwrite part of the tail, we need that tail to be converted into
1390            unformatted node */
1391         res = reiserfs_check_for_tail_and_convert(inode, pos, count);
1392         if (res)
1393                 goto out;
1394
1395         while (count > 0) {
1396                 /* This is the main loop in which we running until some error occures
1397                    or until we write all of the data. */
1398                 size_t num_pages;       /* amount of pages we are going to write this iteration */
1399                 size_t write_bytes;     /* amount of bytes to write during this iteration */
1400                 size_t blocks_to_allocate;      /* how much blocks we need to allocate for this iteration */
1401
1402                 /*  (pos & (PAGE_CACHE_SIZE-1)) is an idiom for offset into a page of pos */
1403                 num_pages = !!((pos + count) & (PAGE_CACHE_SIZE - 1)) + /* round up partial
1404                                                                            pages */
1405                     ((count +
1406                       (pos & (PAGE_CACHE_SIZE - 1))) >> PAGE_CACHE_SHIFT);
1407                 /* convert size to amount of
1408                    pages */
1409                 reiserfs_write_lock(inode->i_sb);
1410                 if (num_pages > REISERFS_WRITE_PAGES_AT_A_TIME
1411                     || num_pages > reiserfs_can_fit_pages(inode->i_sb)) {
1412                         /* If we were asked to write more data than we want to or if there
1413                            is not that much space, then we shorten amount of data to write
1414                            for this iteration. */
1415                         num_pages =
1416                             min_t(size_t, REISERFS_WRITE_PAGES_AT_A_TIME,
1417                                   reiserfs_can_fit_pages(inode->i_sb));
1418                         /* Also we should not forget to set size in bytes accordingly */
1419                         write_bytes = (num_pages << PAGE_CACHE_SHIFT) -
1420                             (pos & (PAGE_CACHE_SIZE - 1));
1421                         /* If position is not on the
1422                            start of the page, we need
1423                            to substract the offset
1424                            within page */
1425                 } else
1426                         write_bytes = count;
1427
1428                 /* reserve the blocks to be allocated later, so that later on
1429                    we still have the space to write the blocks to */
1430                 reiserfs_claim_blocks_to_be_allocated(inode->i_sb,
1431                                                       num_pages <<
1432                                                       (PAGE_CACHE_SHIFT -
1433                                                        inode->i_blkbits));
1434                 reiserfs_write_unlock(inode->i_sb);
1435
1436                 if (!num_pages) {       /* If we do not have enough space even for a single page... */
1437                         if (pos >
1438                             inode->i_size + inode->i_sb->s_blocksize -
1439                             (pos & (inode->i_sb->s_blocksize - 1))) {
1440                                 res = -ENOSPC;
1441                                 break;  // In case we are writing past the end of the last file block, break.
1442                         }
1443                         // Otherwise we are possibly overwriting the file, so
1444                         // let's set write size to be equal or less than blocksize.
1445                         // This way we get it correctly for file holes.
1446                         // But overwriting files on absolutelly full volumes would not
1447                         // be very efficient. Well, people are not supposed to fill
1448                         // 100% of disk space anyway.
1449                         write_bytes =
1450                             min_t(size_t, count,
1451                                   inode->i_sb->s_blocksize -
1452                                   (pos & (inode->i_sb->s_blocksize - 1)));
1453                         num_pages = 1;
1454                         // No blocks were claimed before, so do it now.
1455                         reiserfs_claim_blocks_to_be_allocated(inode->i_sb,
1456                                                               1 <<
1457                                                               (PAGE_CACHE_SHIFT
1458                                                                -
1459                                                                inode->
1460                                                                i_blkbits));
1461                 }
1462
1463                 /* Prepare for writing into the region, read in all the
1464                    partially overwritten pages, if needed. And lock the pages,
1465                    so that nobody else can access these until we are done.
1466                    We get number of actual blocks needed as a result. */
1467                 res = reiserfs_prepare_file_region_for_write(inode, pos,
1468                                                              num_pages,
1469                                                              write_bytes,
1470                                                              prepared_pages);
1471                 if (res < 0) {
1472                         reiserfs_release_claimed_blocks(inode->i_sb,
1473                                                         num_pages <<
1474                                                         (PAGE_CACHE_SHIFT -
1475                                                          inode->i_blkbits));
1476                         break;
1477                 }
1478
1479                 blocks_to_allocate = res;
1480
1481                 /* First we correct our estimate of how many blocks we need */
1482                 reiserfs_release_claimed_blocks(inode->i_sb,
1483                                                 (num_pages <<
1484                                                  (PAGE_CACHE_SHIFT -
1485                                                   inode->i_sb->
1486                                                   s_blocksize_bits)) -
1487                                                 blocks_to_allocate);
1488
1489                 if (blocks_to_allocate > 0) {   /*We only allocate blocks if we need to */
1490                         /* Fill in all the possible holes and append the file if needed */
1491                         res =
1492                             reiserfs_allocate_blocks_for_region(&th, inode, pos,
1493                                                                 num_pages,
1494                                                                 write_bytes,
1495                                                                 prepared_pages,
1496                                                                 blocks_to_allocate);
1497                 }
1498
1499                 /* well, we have allocated the blocks, so it is time to free
1500                    the reservation we made earlier. */
1501                 reiserfs_release_claimed_blocks(inode->i_sb,
1502                                                 blocks_to_allocate);
1503                 if (res) {
1504                         reiserfs_unprepare_pages(prepared_pages, num_pages);
1505                         break;
1506                 }
1507
1508 /* NOTE that allocating blocks and filling blocks can be done in reverse order
1509    and probably we would do that just to get rid of garbage in files after a
1510    crash */
1511
1512                 /* Copy data from user-supplied buffer to file's pages */
1513                 res =
1514                     reiserfs_copy_from_user_to_file_region(pos, num_pages,
1515                                                            write_bytes,
1516                                                            prepared_pages, buf);
1517                 if (res) {
1518                         reiserfs_unprepare_pages(prepared_pages, num_pages);
1519                         break;
1520                 }
1521
1522                 /* Send the pages to disk and unlock them. */
1523                 res =
1524                     reiserfs_submit_file_region_for_write(&th, inode, pos,
1525                                                           num_pages,
1526                                                           write_bytes,
1527                                                           prepared_pages);
1528                 if (res)
1529                         break;
1530
1531                 already_written += write_bytes;
1532                 buf += write_bytes;
1533                 *ppos = pos += write_bytes;
1534                 count -= write_bytes;
1535                 balance_dirty_pages_ratelimited_nr(inode->i_mapping, num_pages);
1536         }
1537
1538         /* this is only true on error */
1539         if (th.t_trans_id) {
1540                 reiserfs_write_lock(inode->i_sb);
1541                 err = journal_end(&th, th.t_super, th.t_blocks_allocated);
1542                 reiserfs_write_unlock(inode->i_sb);
1543                 if (err) {
1544                         res = err;
1545                         goto out;
1546                 }
1547         }
1548
1549         if (likely(res >= 0) &&
1550             (unlikely((file->f_flags & O_SYNC) || IS_SYNC(inode))))
1551                 res = generic_osync_inode(inode, file->f_mapping,
1552                                           OSYNC_METADATA | OSYNC_DATA);
1553
1554         mutex_unlock(&inode->i_mutex);
1555         reiserfs_async_progress_wait(inode->i_sb);
1556         return (already_written != 0) ? already_written : res;
1557
1558       out:
1559         mutex_unlock(&inode->i_mutex);  // unlock the file on exit.
1560         return res;
1561 }
1562
1563 static ssize_t reiserfs_aio_write(struct kiocb *iocb, const char __user * buf,
1564                                   size_t count, loff_t pos)
1565 {
1566         return generic_file_aio_write(iocb, buf, count, pos);
1567 }
1568
1569 const struct file_operations reiserfs_file_operations = {
1570         .read = generic_file_read,
1571         .write = reiserfs_file_write,
1572         .ioctl = reiserfs_ioctl,
1573         .mmap = generic_file_mmap,
1574         .release = reiserfs_file_release,
1575         .fsync = reiserfs_sync_file,
1576         .sendfile = generic_file_sendfile,
1577         .aio_read = generic_file_aio_read,
1578         .aio_write = reiserfs_aio_write,
1579         .splice_read = generic_file_splice_read,
1580         .splice_write = generic_file_splice_write,
1581 };
1582
1583 struct inode_operations reiserfs_file_inode_operations = {
1584         .truncate = reiserfs_vfs_truncate_file,
1585         .setattr = reiserfs_setattr,
1586         .setxattr = reiserfs_setxattr,
1587         .getxattr = reiserfs_getxattr,
1588         .listxattr = reiserfs_listxattr,
1589         .removexattr = reiserfs_removexattr,
1590         .permission = reiserfs_permission,
1591 };