Pull netlink into release branch
[linux-2.6] / 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 <asm/uaccess.h>
10 #include <linux/pagemap.h>
11 #include <linux/swap.h>
12 #include <linux/writeback.h>
13 #include <linux/blkdev.h>
14 #include <linux/buffer_head.h>
15 #include <linux/quotaops.h>
16
17 /*
18 ** We pack the tails of files on file close, not at the time they are written.
19 ** This implies an unnecessary copy of the tail and an unnecessary indirect item
20 ** insertion/balancing, for files that are written in one write.
21 ** It avoids unnecessary tail packings (balances) for files that are written in
22 ** multiple writes and are small enough to have tails.
23 ** 
24 ** file_release is called by the VFS layer when the file is closed.  If
25 ** this is the last open file descriptor, and the file
26 ** small enough to have a tail, and the tail is currently in an
27 ** unformatted node, the tail is converted back into a direct item.
28 ** 
29 ** We use reiserfs_truncate_file to pack the tail, since it already has
30 ** all the conditions coded.  
31 */
32 static int reiserfs_file_release(struct inode *inode, struct file *filp)
33 {
34
35         struct reiserfs_transaction_handle th;
36         int err;
37         int jbegin_failure = 0;
38
39         BUG_ON(!S_ISREG(inode->i_mode));
40
41         /* fast out for when nothing needs to be done */
42         if ((atomic_read(&inode->i_count) > 1 ||
43              !(REISERFS_I(inode)->i_flags & i_pack_on_close_mask) ||
44              !tail_has_to_be_packed(inode)) &&
45             REISERFS_I(inode)->i_prealloc_count <= 0) {
46                 return 0;
47         }
48
49         mutex_lock(&inode->i_mutex);
50
51         mutex_lock(&(REISERFS_I(inode)->i_mmap));
52         if (REISERFS_I(inode)->i_flags & i_ever_mapped)
53                 REISERFS_I(inode)->i_flags &= ~i_pack_on_close_mask;
54
55         reiserfs_write_lock(inode->i_sb);
56         /* freeing preallocation only involves relogging blocks that
57          * are already in the current transaction.  preallocation gets
58          * freed at the end of each transaction, so it is impossible for
59          * us to log any additional blocks (including quota blocks)
60          */
61         err = journal_begin(&th, inode->i_sb, 1);
62         if (err) {
63                 /* uh oh, we can't allow the inode to go away while there
64                  * is still preallocation blocks pending.  Try to join the
65                  * aborted transaction
66                  */
67                 jbegin_failure = err;
68                 err = journal_join_abort(&th, inode->i_sb, 1);
69
70                 if (err) {
71                         /* hmpf, our choices here aren't good.  We can pin the inode
72                          * which will disallow unmount from every happening, we can
73                          * do nothing, which will corrupt random memory on unmount,
74                          * or we can forcibly remove the file from the preallocation
75                          * list, which will leak blocks on disk.  Lets pin the inode
76                          * and let the admin know what is going on.
77                          */
78                         igrab(inode);
79                         reiserfs_warning(inode->i_sb,
80                                          "pinning inode %lu because the "
81                                          "preallocation can't be freed",
82                                          inode->i_ino);
83                         goto out;
84                 }
85         }
86         reiserfs_update_inode_transaction(inode);
87
88 #ifdef REISERFS_PREALLOCATE
89         reiserfs_discard_prealloc(&th, inode);
90 #endif
91         err = journal_end(&th, inode->i_sb, 1);
92
93         /* copy back the error code from journal_begin */
94         if (!err)
95                 err = jbegin_failure;
96
97         if (!err && atomic_read(&inode->i_count) <= 1 &&
98             (REISERFS_I(inode)->i_flags & i_pack_on_close_mask) &&
99             tail_has_to_be_packed(inode)) {
100                 /* if regular file is released by last holder and it has been
101                    appended (we append by unformatted node only) or its direct
102                    item(s) had to be converted, then it may have to be
103                    indirect2direct converted */
104                 err = reiserfs_truncate_file(inode, 0);
105         }
106       out:
107         mutex_unlock(&(REISERFS_I(inode)->i_mmap));
108         mutex_unlock(&inode->i_mutex);
109         reiserfs_write_unlock(inode->i_sb);
110         return err;
111 }
112
113 static int reiserfs_file_mmap(struct file *file, struct vm_area_struct *vma)
114 {
115         struct inode *inode;
116
117         inode = file->f_path.dentry->d_inode;
118         mutex_lock(&(REISERFS_I(inode)->i_mmap));
119         REISERFS_I(inode)->i_flags |= i_ever_mapped;
120         mutex_unlock(&(REISERFS_I(inode)->i_mmap));
121
122         return generic_file_mmap(file, vma);
123 }
124
125 static void reiserfs_vfs_truncate_file(struct inode *inode)
126 {
127         reiserfs_truncate_file(inode, 1);
128 }
129
130 /* Sync a reiserfs file. */
131
132 /*
133  * FIXME: sync_mapping_buffers() never has anything to sync.  Can
134  * be removed...
135  */
136
137 static int reiserfs_sync_file(struct file *p_s_filp,
138                               struct dentry *p_s_dentry, int datasync)
139 {
140         struct inode *p_s_inode = p_s_dentry->d_inode;
141         int n_err;
142         int barrier_done;
143
144         BUG_ON(!S_ISREG(p_s_inode->i_mode));
145         n_err = sync_mapping_buffers(p_s_inode->i_mapping);
146         reiserfs_write_lock(p_s_inode->i_sb);
147         barrier_done = reiserfs_commit_for_inode(p_s_inode);
148         reiserfs_write_unlock(p_s_inode->i_sb);
149         if (barrier_done != 1 && reiserfs_barrier_flush(p_s_inode->i_sb))
150                 blkdev_issue_flush(p_s_inode->i_sb->s_bdev, NULL);
151         if (barrier_done < 0)
152                 return barrier_done;
153         return (n_err < 0) ? -EIO : 0;
154 }
155
156 /* I really do not want to play with memory shortage right now, so
157    to simplify the code, we are not going to write more than this much pages at
158    a time. This still should considerably improve performance compared to 4k
159    at a time case. This is 32 pages of 4k size. */
160 #define REISERFS_WRITE_PAGES_AT_A_TIME (128 * 1024) / PAGE_CACHE_SIZE
161
162 /* Allocates blocks for a file to fulfil write request.
163    Maps all unmapped but prepared pages from the list.
164    Updates metadata with newly allocated blocknumbers as needed */
165 static int reiserfs_allocate_blocks_for_region(struct reiserfs_transaction_handle *th, struct inode *inode,     /* Inode we work with */
166                                                loff_t pos,      /* Writing position */
167                                                int num_pages,   /* number of pages write going
168                                                                    to touch */
169                                                int write_bytes, /* amount of bytes to write */
170                                                struct page **prepared_pages,    /* array of
171                                                                                    prepared pages
172                                                                                  */
173                                                int blocks_to_allocate   /* Amount of blocks we
174                                                                            need to allocate to
175                                                                            fit the data into file
176                                                                          */
177     )
178 {
179         struct cpu_key key;     // cpu key of item that we are going to deal with
180         struct item_head *ih;   // pointer to item head that we are going to deal with
181         struct buffer_head *bh; // Buffer head that contains items that we are going to deal with
182         __le32 *item;           // pointer to item we are going to deal with
183         INITIALIZE_PATH(path);  // path to item, that we are going to deal with.
184         b_blocknr_t *allocated_blocks;  // Pointer to a place where allocated blocknumbers would be stored.
185         reiserfs_blocknr_hint_t hint;   // hint structure for block allocator.
186         size_t res;             // return value of various functions that we call.
187         int curr_block;         // current block used to keep track of unmapped blocks.
188         int i;                  // loop counter
189         int itempos;            // position in item
190         unsigned int from = (pos & (PAGE_CACHE_SIZE - 1));      // writing position in
191         // first page
192         unsigned int to = ((pos + write_bytes - 1) & (PAGE_CACHE_SIZE - 1)) + 1;        /* last modified byte offset in last page */
193         __u64 hole_size;        // amount of blocks for a file hole, if it needed to be created.
194         int modifying_this_item = 0;    // Flag for items traversal code to keep track
195         // of the fact that we already prepared
196         // current block for journal
197         int will_prealloc = 0;
198         RFALSE(!blocks_to_allocate,
199                "green-9004: tried to allocate zero blocks?");
200
201         /* only preallocate if this is a small write */
202         if (REISERFS_I(inode)->i_prealloc_count ||
203             (!(write_bytes & (inode->i_sb->s_blocksize - 1)) &&
204              blocks_to_allocate <
205              REISERFS_SB(inode->i_sb)->s_alloc_options.preallocsize))
206                 will_prealloc =
207                     REISERFS_SB(inode->i_sb)->s_alloc_options.preallocsize;
208
209         allocated_blocks = kmalloc((blocks_to_allocate + will_prealloc) *
210                                    sizeof(b_blocknr_t), GFP_NOFS);
211         if (!allocated_blocks)
212                 return -ENOMEM;
213
214         /* First we compose a key to point at the writing position, we want to do
215            that outside of any locking region. */
216         make_cpu_key(&key, inode, pos + 1, TYPE_ANY, 3 /*key length */ );
217
218         /* If we came here, it means we absolutely need to open a transaction,
219            since we need to allocate some blocks */
220         reiserfs_write_lock(inode->i_sb);       // Journaling stuff and we need that.
221         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
222         if (res)
223                 goto error_exit;
224         reiserfs_update_inode_transaction(inode);
225
226         /* Look for the in-tree position of our write, need path for block allocator */
227         res = search_for_position_by_key(inode->i_sb, &key, &path);
228         if (res == IO_ERROR) {
229                 res = -EIO;
230                 goto error_exit;
231         }
232
233         /* Allocate blocks */
234         /* First fill in "hint" structure for block allocator */
235         hint.th = th;           // transaction handle.
236         hint.path = &path;      // Path, so that block allocator can determine packing locality or whatever it needs to determine.
237         hint.inode = inode;     // Inode is needed by block allocator too.
238         hint.search_start = 0;  // We have no hint on where to search free blocks for block allocator.
239         hint.key = key.on_disk_key;     // on disk key of file.
240         hint.block = inode->i_blocks >> (inode->i_sb->s_blocksize_bits - 9);    // Number of disk blocks this file occupies already.
241         hint.formatted_node = 0;        // We are allocating blocks for unformatted node.
242         hint.preallocate = will_prealloc;
243
244         /* Call block allocator to allocate blocks */
245         res =
246             reiserfs_allocate_blocknrs(&hint, allocated_blocks,
247                                        blocks_to_allocate, blocks_to_allocate);
248         if (res != CARRY_ON) {
249                 if (res == NO_DISK_SPACE) {
250                         /* We flush the transaction in case of no space. This way some
251                            blocks might become free */
252                         SB_JOURNAL(inode->i_sb)->j_must_wait = 1;
253                         res = restart_transaction(th, inode, &path);
254                         if (res)
255                                 goto error_exit;
256
257                         /* We might have scheduled, so search again */
258                         res =
259                             search_for_position_by_key(inode->i_sb, &key,
260                                                        &path);
261                         if (res == IO_ERROR) {
262                                 res = -EIO;
263                                 goto error_exit;
264                         }
265
266                         /* update changed info for hint structure. */
267                         res =
268                             reiserfs_allocate_blocknrs(&hint, allocated_blocks,
269                                                        blocks_to_allocate,
270                                                        blocks_to_allocate);
271                         if (res != CARRY_ON) {
272                                 res = res == QUOTA_EXCEEDED ? -EDQUOT : -ENOSPC;
273                                 pathrelse(&path);
274                                 goto error_exit;
275                         }
276                 } else {
277                         res = res == QUOTA_EXCEEDED ? -EDQUOT : -ENOSPC;
278                         pathrelse(&path);
279                         goto error_exit;
280                 }
281         }
282 #ifdef __BIG_ENDIAN
283         // Too bad, I have not found any way to convert a given region from
284         // cpu format to little endian format
285         {
286                 int i;
287                 for (i = 0; i < blocks_to_allocate; i++)
288                         allocated_blocks[i] = cpu_to_le32(allocated_blocks[i]);
289         }
290 #endif
291
292         /* Blocks allocating well might have scheduled and tree might have changed,
293            let's search the tree again */
294         /* find where in the tree our write should go */
295         res = search_for_position_by_key(inode->i_sb, &key, &path);
296         if (res == IO_ERROR) {
297                 res = -EIO;
298                 goto error_exit_free_blocks;
299         }
300
301         bh = get_last_bh(&path);        // Get a bufferhead for last element in path.
302         ih = get_ih(&path);     // Get a pointer to last item head in path.
303         item = get_item(&path); // Get a pointer to last item in path
304
305         /* Let's see what we have found */
306         if (res != POSITION_FOUND) {    /* position not found, this means that we
307                                            might need to append file with holes
308                                            first */
309                 // Since we are writing past the file's end, we need to find out if
310                 // there is a hole that needs to be inserted before our writing
311                 // position, and how many blocks it is going to cover (we need to
312                 //  populate pointers to file blocks representing the hole with zeros)
313
314                 {
315                         int item_offset = 1;
316                         /*
317                          * if ih is stat data, its offset is 0 and we don't want to
318                          * add 1 to pos in the hole_size calculation
319                          */
320                         if (is_statdata_le_ih(ih))
321                                 item_offset = 0;
322                         hole_size = (pos + item_offset -
323                                      (le_key_k_offset
324                                       (get_inode_item_key_version(inode),
325                                        &(ih->ih_key)) + op_bytes_number(ih,
326                                                                         inode->
327                                                                         i_sb->
328                                                                         s_blocksize)))
329                             >> inode->i_sb->s_blocksize_bits;
330                 }
331
332                 if (hole_size > 0) {
333                         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.
334                         /* area filled with zeroes, to supply as list of zero blocknumbers
335                            We allocate it outside of loop just in case loop would spin for
336                            several iterations. */
337                         char *zeros = kzalloc(to_paste * UNFM_P_SIZE, GFP_ATOMIC);      // We cannot insert more than MAX_ITEM_LEN bytes anyway.
338                         if (!zeros) {
339                                 res = -ENOMEM;
340                                 goto error_exit_free_blocks;
341                         }
342                         do {
343                                 to_paste =
344                                     min_t(__u64, hole_size,
345                                           MAX_ITEM_LEN(inode->i_sb->
346                                                        s_blocksize) /
347                                           UNFM_P_SIZE);
348                                 if (is_indirect_le_ih(ih)) {
349                                         /* Ok, there is existing indirect item already. Need to append it */
350                                         /* Calculate position past inserted item */
351                                         make_cpu_key(&key, inode,
352                                                      le_key_k_offset
353                                                      (get_inode_item_key_version
354                                                       (inode),
355                                                       &(ih->ih_key)) +
356                                                      op_bytes_number(ih,
357                                                                      inode->
358                                                                      i_sb->
359                                                                      s_blocksize),
360                                                      TYPE_INDIRECT, 3);
361                                         res =
362                                             reiserfs_paste_into_item(th, &path,
363                                                                      &key,
364                                                                      inode,
365                                                                      (char *)
366                                                                      zeros,
367                                                                      UNFM_P_SIZE
368                                                                      *
369                                                                      to_paste);
370                                         if (res) {
371                                                 kfree(zeros);
372                                                 goto error_exit_free_blocks;
373                                         }
374                                 } else if (is_statdata_le_ih(ih)) {
375                                         /* No existing item, create it */
376                                         /* item head for new item */
377                                         struct item_head ins_ih;
378
379                                         /* create a key for our new item */
380                                         make_cpu_key(&key, inode, 1,
381                                                      TYPE_INDIRECT, 3);
382
383                                         /* Create new item head for our new item */
384                                         make_le_item_head(&ins_ih, &key,
385                                                           key.version, 1,
386                                                           TYPE_INDIRECT,
387                                                           to_paste *
388                                                           UNFM_P_SIZE,
389                                                           0 /* free space */ );
390
391                                         /* Find where such item should live in the tree */
392                                         res =
393                                             search_item(inode->i_sb, &key,
394                                                         &path);
395                                         if (res != ITEM_NOT_FOUND) {
396                                                 /* item should not exist, otherwise we have error */
397                                                 if (res != -ENOSPC) {
398                                                         reiserfs_warning(inode->
399                                                                          i_sb,
400                                                                          "green-9008: search_by_key (%K) returned %d",
401                                                                          &key,
402                                                                          res);
403                                                 }
404                                                 res = -EIO;
405                                                 kfree(zeros);
406                                                 goto error_exit_free_blocks;
407                                         }
408                                         res =
409                                             reiserfs_insert_item(th, &path,
410                                                                  &key, &ins_ih,
411                                                                  inode,
412                                                                  (char *)zeros);
413                                 } else {
414                                         reiserfs_panic(inode->i_sb,
415                                                        "green-9011: Unexpected key type %K\n",
416                                                        &key);
417                                 }
418                                 if (res) {
419                                         kfree(zeros);
420                                         goto error_exit_free_blocks;
421                                 }
422                                 /* Now we want to check if transaction is too full, and if it is
423                                    we restart it. This will also free the path. */
424                                 if (journal_transaction_should_end
425                                     (th, th->t_blocks_allocated)) {
426                                         inode->i_size = cpu_key_k_offset(&key) +
427                                                 (to_paste << inode->i_blkbits);
428                                         res =
429                                             restart_transaction(th, inode,
430                                                                 &path);
431                                         if (res) {
432                                                 pathrelse(&path);
433                                                 kfree(zeros);
434                                                 goto error_exit;
435                                         }
436                                 }
437
438                                 /* Well, need to recalculate path and stuff */
439                                 set_cpu_key_k_offset(&key,
440                                                      cpu_key_k_offset(&key) +
441                                                      (to_paste << inode->
442                                                       i_blkbits));
443                                 res =
444                                     search_for_position_by_key(inode->i_sb,
445                                                                &key, &path);
446                                 if (res == IO_ERROR) {
447                                         res = -EIO;
448                                         kfree(zeros);
449                                         goto error_exit_free_blocks;
450                                 }
451                                 bh = get_last_bh(&path);
452                                 ih = get_ih(&path);
453                                 item = get_item(&path);
454                                 hole_size -= to_paste;
455                         } while (hole_size);
456                         kfree(zeros);
457                 }
458         }
459         // Go through existing indirect items first
460         // replace all zeroes with blocknumbers from list
461         // Note that if no corresponding item was found, by previous search,
462         // it means there are no existing in-tree representation for file area
463         // we are going to overwrite, so there is nothing to scan through for holes.
464         for (curr_block = 0, itempos = path.pos_in_item;
465              curr_block < blocks_to_allocate && res == POSITION_FOUND;) {
466               retry:
467
468                 if (itempos >= ih_item_len(ih) / UNFM_P_SIZE) {
469                         /* We run out of data in this indirect item, let's look for another
470                            one. */
471                         /* First if we are already modifying current item, log it */
472                         if (modifying_this_item) {
473                                 journal_mark_dirty(th, inode->i_sb, bh);
474                                 modifying_this_item = 0;
475                         }
476                         /* Then set the key to look for a new indirect item (offset of old
477                            item is added to old item length */
478                         set_cpu_key_k_offset(&key,
479                                              le_key_k_offset
480                                              (get_inode_item_key_version(inode),
481                                               &(ih->ih_key)) +
482                                              op_bytes_number(ih,
483                                                              inode->i_sb->
484                                                              s_blocksize));
485                         /* Search ofor position of new key in the tree. */
486                         res =
487                             search_for_position_by_key(inode->i_sb, &key,
488                                                        &path);
489                         if (res == IO_ERROR) {
490                                 res = -EIO;
491                                 goto error_exit_free_blocks;
492                         }
493                         bh = get_last_bh(&path);
494                         ih = get_ih(&path);
495                         item = get_item(&path);
496                         itempos = path.pos_in_item;
497                         continue;       // loop to check all kinds of conditions and so on.
498                 }
499                 /* Ok, we have correct position in item now, so let's see if it is
500                    representing file hole (blocknumber is zero) and fill it if needed */
501                 if (!item[itempos]) {
502                         /* Ok, a hole. Now we need to check if we already prepared this
503                            block to be journaled */
504                         while (!modifying_this_item) {  // loop until succeed
505                                 /* Well, this item is not journaled yet, so we must prepare
506                                    it for journal first, before we can change it */
507                                 struct item_head tmp_ih;        // We copy item head of found item,
508                                 // here to detect if fs changed under
509                                 // us while we were preparing for
510                                 // journal.
511                                 int fs_gen;     // We store fs generation here to find if someone
512                                 // changes fs under our feet
513
514                                 copy_item_head(&tmp_ih, ih);    // Remember itemhead
515                                 fs_gen = get_generation(inode->i_sb);   // remember fs generation
516                                 reiserfs_prepare_for_journal(inode->i_sb, bh, 1);       // Prepare a buffer within which indirect item is stored for changing.
517                                 if (fs_changed(fs_gen, inode->i_sb)
518                                     && item_moved(&tmp_ih, &path)) {
519                                         // Sigh, fs was changed under us, we need to look for new
520                                         // location of item we are working with
521
522                                         /* unmark prepaerd area as journaled and search for it's
523                                            new position */
524                                         reiserfs_restore_prepared_buffer(inode->
525                                                                          i_sb,
526                                                                          bh);
527                                         res =
528                                             search_for_position_by_key(inode->
529                                                                        i_sb,
530                                                                        &key,
531                                                                        &path);
532                                         if (res == IO_ERROR) {
533                                                 res = -EIO;
534                                                 goto error_exit_free_blocks;
535                                         }
536                                         bh = get_last_bh(&path);
537                                         ih = get_ih(&path);
538                                         item = get_item(&path);
539                                         itempos = path.pos_in_item;
540                                         goto retry;
541                                 }
542                                 modifying_this_item = 1;
543                         }
544                         item[itempos] = allocated_blocks[curr_block];   // Assign new block
545                         curr_block++;
546                 }
547                 itempos++;
548         }
549
550         if (modifying_this_item) {      // We need to log last-accessed block, if it
551                 // was modified, but not logged yet.
552                 journal_mark_dirty(th, inode->i_sb, bh);
553         }
554
555         if (curr_block < blocks_to_allocate) {
556                 // Oh, well need to append to indirect item, or to create indirect item
557                 // if there weren't any
558                 if (is_indirect_le_ih(ih)) {
559                         // Existing indirect item - append. First calculate key for append
560                         // position. We do not need to recalculate path as it should
561                         // already point to correct place.
562                         make_cpu_key(&key, inode,
563                                      le_key_k_offset(get_inode_item_key_version
564                                                      (inode),
565                                                      &(ih->ih_key)) +
566                                      op_bytes_number(ih,
567                                                      inode->i_sb->s_blocksize),
568                                      TYPE_INDIRECT, 3);
569                         res =
570                             reiserfs_paste_into_item(th, &path, &key, inode,
571                                                      (char *)(allocated_blocks +
572                                                               curr_block),
573                                                      UNFM_P_SIZE *
574                                                      (blocks_to_allocate -
575                                                       curr_block));
576                         if (res) {
577                                 goto error_exit_free_blocks;
578                         }
579                 } else if (is_statdata_le_ih(ih)) {
580                         // Last found item was statdata. That means we need to create indirect item.
581                         struct item_head ins_ih;        /* itemhead for new item */
582
583                         /* create a key for our new item */
584                         make_cpu_key(&key, inode, 1, TYPE_INDIRECT, 3); // Position one,
585                         // because that's
586                         // where first
587                         // indirect item
588                         // begins
589                         /* Create new item head for our new item */
590                         make_le_item_head(&ins_ih, &key, key.version, 1,
591                                           TYPE_INDIRECT,
592                                           (blocks_to_allocate -
593                                            curr_block) * UNFM_P_SIZE,
594                                           0 /* free space */ );
595                         /* Find where such item should live in the tree */
596                         res = search_item(inode->i_sb, &key, &path);
597                         if (res != ITEM_NOT_FOUND) {
598                                 /* Well, if we have found such item already, or some error
599                                    occured, we need to warn user and return error */
600                                 if (res != -ENOSPC) {
601                                         reiserfs_warning(inode->i_sb,
602                                                          "green-9009: search_by_key (%K) "
603                                                          "returned %d", &key,
604                                                          res);
605                                 }
606                                 res = -EIO;
607                                 goto error_exit_free_blocks;
608                         }
609                         /* Insert item into the tree with the data as its body */
610                         res =
611                             reiserfs_insert_item(th, &path, &key, &ins_ih,
612                                                  inode,
613                                                  (char *)(allocated_blocks +
614                                                           curr_block));
615                 } else {
616                         reiserfs_panic(inode->i_sb,
617                                        "green-9010: unexpected item type for key %K\n",
618                                        &key);
619                 }
620         }
621         // the caller is responsible for closing the transaction
622         // unless we return an error, they are also responsible for logging
623         // the inode.
624         //
625         pathrelse(&path);
626         /*
627          * cleanup prellocation from previous writes
628          * if this is a partial block write
629          */
630         if (write_bytes & (inode->i_sb->s_blocksize - 1))
631                 reiserfs_discard_prealloc(th, inode);
632         reiserfs_write_unlock(inode->i_sb);
633
634         // go through all the pages/buffers and map the buffers to newly allocated
635         // blocks (so that system knows where to write these pages later).
636         curr_block = 0;
637         for (i = 0; i < num_pages; i++) {
638                 struct page *page = prepared_pages[i];  //current page
639                 struct buffer_head *head = page_buffers(page);  // first buffer for a page
640                 int block_start, block_end;     // in-page offsets for buffers.
641
642                 if (!page_buffers(page))
643                         reiserfs_panic(inode->i_sb,
644                                        "green-9005: No buffers for prepared page???");
645
646                 /* For each buffer in page */
647                 for (bh = head, block_start = 0; bh != head || !block_start;
648                      block_start = block_end, bh = bh->b_this_page) {
649                         if (!bh)
650                                 reiserfs_panic(inode->i_sb,
651                                                "green-9006: Allocated but absent buffer for a page?");
652                         block_end = block_start + inode->i_sb->s_blocksize;
653                         if (i == 0 && block_end <= from)
654                                 /* if this buffer is before requested data to map, skip it */
655                                 continue;
656                         if (i == num_pages - 1 && block_start >= to)
657                                 /* If this buffer is after requested data to map, abort
658                                    processing of current page */
659                                 break;
660
661                         if (!buffer_mapped(bh)) {       // Ok, unmapped buffer, need to map it
662                                 map_bh(bh, inode->i_sb,
663                                        le32_to_cpu(allocated_blocks
664                                                    [curr_block]));
665                                 curr_block++;
666                                 set_buffer_new(bh);
667                         }
668                 }
669         }
670
671         RFALSE(curr_block > blocks_to_allocate,
672                "green-9007: Used too many blocks? weird");
673
674         kfree(allocated_blocks);
675         return 0;
676
677 // Need to deal with transaction here.
678       error_exit_free_blocks:
679         pathrelse(&path);
680         // free blocks
681         for (i = 0; i < blocks_to_allocate; i++)
682                 reiserfs_free_block(th, inode, le32_to_cpu(allocated_blocks[i]),
683                                     1);
684
685       error_exit:
686         if (th->t_trans_id) {
687                 int err;
688                 // update any changes we made to blk count
689                 mark_inode_dirty(inode);
690                 err =
691                     journal_end(th, inode->i_sb,
692                                 JOURNAL_PER_BALANCE_CNT * 3 + 1 +
693                                 2 * REISERFS_QUOTA_TRANS_BLOCKS(inode->i_sb));
694                 if (err)
695                         res = err;
696         }
697         reiserfs_write_unlock(inode->i_sb);
698         kfree(allocated_blocks);
699
700         return res;
701 }
702
703 /* Unlock pages prepared by reiserfs_prepare_file_region_for_write */
704 static void reiserfs_unprepare_pages(struct page **prepared_pages,      /* list of locked pages */
705                                      size_t num_pages /* amount of pages */ )
706 {
707         int i;                  // loop counter
708
709         for (i = 0; i < num_pages; i++) {
710                 struct page *page = prepared_pages[i];
711
712                 try_to_free_buffers(page);
713                 unlock_page(page);
714                 page_cache_release(page);
715         }
716 }
717
718 /* This function will copy data from userspace to specified pages within
719    supplied byte range */
720 static int reiserfs_copy_from_user_to_file_region(loff_t pos,   /* In-file position */
721                                                   int num_pages,        /* Number of pages affected */
722                                                   int write_bytes,      /* Amount of bytes to write */
723                                                   struct page **prepared_pages, /* pointer to 
724                                                                                    array to
725                                                                                    prepared pages
726                                                                                  */
727                                                   const char __user * buf       /* Pointer to user-supplied
728                                                                                    data */
729     )
730 {
731         long page_fault = 0;    // status of copy_from_user.
732         int i;                  // loop counter.
733         int offset;             // offset in page
734
735         for (i = 0, offset = (pos & (PAGE_CACHE_SIZE - 1)); i < num_pages;
736              i++, offset = 0) {
737                 size_t count = min_t(size_t, PAGE_CACHE_SIZE - offset, write_bytes);    // How much of bytes to write to this page
738                 struct page *page = prepared_pages[i];  // Current page we process.
739
740                 fault_in_pages_readable(buf, count);
741
742                 /* Copy data from userspace to the current page */
743                 kmap(page);
744                 page_fault = __copy_from_user(page_address(page) + offset, buf, count); // Copy the data.
745                 /* Flush processor's dcache for this page */
746                 flush_dcache_page(page);
747                 kunmap(page);
748                 buf += count;
749                 write_bytes -= count;
750
751                 if (page_fault)
752                         break;  // Was there a fault? abort.
753         }
754
755         return page_fault ? -EFAULT : 0;
756 }
757
758 /* taken fs/buffer.c:__block_commit_write */
759 int reiserfs_commit_page(struct inode *inode, struct page *page,
760                          unsigned from, unsigned to)
761 {
762         unsigned block_start, block_end;
763         int partial = 0;
764         unsigned blocksize;
765         struct buffer_head *bh, *head;
766         unsigned long i_size_index = inode->i_size >> PAGE_CACHE_SHIFT;
767         int new;
768         int logit = reiserfs_file_data_log(inode);
769         struct super_block *s = inode->i_sb;
770         int bh_per_page = PAGE_CACHE_SIZE / s->s_blocksize;
771         struct reiserfs_transaction_handle th;
772         int ret = 0;
773
774         th.t_trans_id = 0;
775         blocksize = 1 << inode->i_blkbits;
776
777         if (logit) {
778                 reiserfs_write_lock(s);
779                 ret = journal_begin(&th, s, bh_per_page + 1);
780                 if (ret)
781                         goto drop_write_lock;
782                 reiserfs_update_inode_transaction(inode);
783         }
784         for (bh = head = page_buffers(page), block_start = 0;
785              bh != head || !block_start;
786              block_start = block_end, bh = bh->b_this_page) {
787
788                 new = buffer_new(bh);
789                 clear_buffer_new(bh);
790                 block_end = block_start + blocksize;
791                 if (block_end <= from || block_start >= to) {
792                         if (!buffer_uptodate(bh))
793                                 partial = 1;
794                 } else {
795                         set_buffer_uptodate(bh);
796                         if (logit) {
797                                 reiserfs_prepare_for_journal(s, bh, 1);
798                                 journal_mark_dirty(&th, s, bh);
799                         } else if (!buffer_dirty(bh)) {
800                                 mark_buffer_dirty(bh);
801                                 /* do data=ordered on any page past the end
802                                  * of file and any buffer marked BH_New.
803                                  */
804                                 if (reiserfs_data_ordered(inode->i_sb) &&
805                                     (new || page->index >= i_size_index)) {
806                                         reiserfs_add_ordered_list(inode, bh);
807                                 }
808                         }
809                 }
810         }
811         if (logit) {
812                 ret = journal_end(&th, s, bh_per_page + 1);
813               drop_write_lock:
814                 reiserfs_write_unlock(s);
815         }
816         /*
817          * If this is a partial write which happened to make all buffers
818          * uptodate then we can optimize away a bogus readpage() for
819          * the next read(). Here we 'discover' whether the page went
820          * uptodate as a result of this (potentially partial) write.
821          */
822         if (!partial)
823                 SetPageUptodate(page);
824         return ret;
825 }
826
827 /* Submit pages for write. This was separated from actual file copying
828    because we might want to allocate block numbers in-between.
829    This function assumes that caller will adjust file size to correct value. */
830 static int reiserfs_submit_file_region_for_write(struct reiserfs_transaction_handle *th, struct inode *inode, loff_t pos,       /* Writing position offset */
831                                                  size_t num_pages,      /* Number of pages to write */
832                                                  size_t write_bytes,    /* number of bytes to write */
833                                                  struct page **prepared_pages   /* list of pages */
834     )
835 {
836         int status;             // return status of block_commit_write.
837         int retval = 0;         // Return value we are going to return.
838         int i;                  // loop counter
839         int offset;             // Writing offset in page.
840         int orig_write_bytes = write_bytes;
841         int sd_update = 0;
842
843         for (i = 0, offset = (pos & (PAGE_CACHE_SIZE - 1)); i < num_pages;
844              i++, offset = 0) {
845                 int count = min_t(int, PAGE_CACHE_SIZE - offset, write_bytes);  // How much of bytes to write to this page
846                 struct page *page = prepared_pages[i];  // Current page we process.
847
848                 status =
849                     reiserfs_commit_page(inode, page, offset, offset + count);
850                 if (status)
851                         retval = status;        // To not overcomplicate matters We are going to
852                 // submit all the pages even if there was error.
853                 // we only remember error status to report it on
854                 // exit.
855                 write_bytes -= count;
856         }
857         /* now that we've gotten all the ordered buffers marked dirty,
858          * we can safely update i_size and close any running transaction
859          */
860         if (pos + orig_write_bytes > inode->i_size) {
861                 inode->i_size = pos + orig_write_bytes; // Set new size
862                 /* If the file have grown so much that tail packing is no
863                  * longer possible, reset "need to pack" flag */
864                 if ((have_large_tails(inode->i_sb) &&
865                      inode->i_size > i_block_size(inode) * 4) ||
866                     (have_small_tails(inode->i_sb) &&
867                      inode->i_size > i_block_size(inode)))
868                         REISERFS_I(inode)->i_flags &= ~i_pack_on_close_mask;
869                 else if ((have_large_tails(inode->i_sb) &&
870                           inode->i_size < i_block_size(inode) * 4) ||
871                          (have_small_tails(inode->i_sb) &&
872                           inode->i_size < i_block_size(inode)))
873                         REISERFS_I(inode)->i_flags |= i_pack_on_close_mask;
874
875                 if (th->t_trans_id) {
876                         reiserfs_write_lock(inode->i_sb);
877                         // this sets the proper flags for O_SYNC to trigger a commit
878                         mark_inode_dirty(inode);
879                         reiserfs_write_unlock(inode->i_sb);
880                 } else {
881                         reiserfs_write_lock(inode->i_sb);
882                         reiserfs_update_inode_transaction(inode);
883                         mark_inode_dirty(inode);
884                         reiserfs_write_unlock(inode->i_sb);
885                 }
886
887                 sd_update = 1;
888         }
889         if (th->t_trans_id) {
890                 reiserfs_write_lock(inode->i_sb);
891                 if (!sd_update)
892                         mark_inode_dirty(inode);
893                 status = journal_end(th, th->t_super, th->t_blocks_allocated);
894                 if (status)
895                         retval = status;
896                 reiserfs_write_unlock(inode->i_sb);
897         }
898         th->t_trans_id = 0;
899
900         /* 
901          * we have to unlock the pages after updating i_size, otherwise
902          * we race with writepage
903          */
904         for (i = 0; i < num_pages; i++) {
905                 struct page *page = prepared_pages[i];
906                 unlock_page(page);
907                 mark_page_accessed(page);
908                 page_cache_release(page);
909         }
910         return retval;
911 }
912
913 /* Look if passed writing region is going to touch file's tail
914    (if it is present). And if it is, convert the tail to unformatted node */
915 static int reiserfs_check_for_tail_and_convert(struct inode *inode,     /* inode to deal with */
916                                                loff_t pos,      /* Writing position */
917                                                int write_bytes  /* amount of bytes to write */
918     )
919 {
920         INITIALIZE_PATH(path);  // needed for search_for_position
921         struct cpu_key key;     // Key that would represent last touched writing byte.
922         struct item_head *ih;   // item header of found block;
923         int res;                // Return value of various functions we call.
924         int cont_expand_offset; // We will put offset for generic_cont_expand here
925         // This can be int just because tails are created
926         // only for small files.
927
928 /* this embodies a dependency on a particular tail policy */
929         if (inode->i_size >= inode->i_sb->s_blocksize * 4) {
930                 /* such a big files do not have tails, so we won't bother ourselves
931                    to look for tails, simply return */
932                 return 0;
933         }
934
935         reiserfs_write_lock(inode->i_sb);
936         /* find the item containing the last byte to be written, or if
937          * writing past the end of the file then the last item of the
938          * file (and then we check its type). */
939         make_cpu_key(&key, inode, pos + write_bytes + 1, TYPE_ANY,
940                      3 /*key length */ );
941         res = search_for_position_by_key(inode->i_sb, &key, &path);
942         if (res == IO_ERROR) {
943                 reiserfs_write_unlock(inode->i_sb);
944                 return -EIO;
945         }
946         ih = get_ih(&path);
947         res = 0;
948         if (is_direct_le_ih(ih)) {
949                 /* Ok, closest item is file tail (tails are stored in "direct"
950                  * items), so we need to unpack it. */
951                 /* To not overcomplicate matters, we just call generic_cont_expand
952                    which will in turn call other stuff and finally will boil down to
953                    reiserfs_get_block() that would do necessary conversion. */
954                 cont_expand_offset =
955                     le_key_k_offset(get_inode_item_key_version(inode),
956                                     &(ih->ih_key));
957                 pathrelse(&path);
958                 res = generic_cont_expand(inode, cont_expand_offset);
959         } else
960                 pathrelse(&path);
961
962         reiserfs_write_unlock(inode->i_sb);
963         return res;
964 }
965
966 /* This function locks pages starting from @pos for @inode.
967    @num_pages pages are locked and stored in
968    @prepared_pages array. Also buffers are allocated for these pages.
969    First and last page of the region is read if it is overwritten only
970    partially. If last page did not exist before write (file hole or file
971    append), it is zeroed, then. 
972    Returns number of unallocated blocks that should be allocated to cover
973    new file data.*/
974 static int reiserfs_prepare_file_region_for_write(struct inode *inode
975                                                   /* Inode of the file */ ,
976                                                   loff_t pos,   /* position in the file */
977                                                   size_t num_pages,     /* number of pages to
978                                                                            prepare */
979                                                   size_t write_bytes,   /* Amount of bytes to be
980                                                                            overwritten from
981                                                                            @pos */
982                                                   struct page **prepared_pages  /* pointer to array
983                                                                                    where to store
984                                                                                    prepared pages */
985     )
986 {
987         int res = 0;            // Return values of different functions we call.
988         unsigned long index = pos >> PAGE_CACHE_SHIFT;  // Offset in file in pages.
989         int from = (pos & (PAGE_CACHE_SIZE - 1));       // Writing offset in first page
990         int to = ((pos + write_bytes - 1) & (PAGE_CACHE_SIZE - 1)) + 1;
991         /* offset of last modified byte in last
992            page */
993         struct address_space *mapping = inode->i_mapping;       // Pages are mapped here.
994         int i;                  // Simple counter
995         int blocks = 0;         /* Return value (blocks that should be allocated) */
996         struct buffer_head *bh, *head;  // Current bufferhead and first bufferhead
997         // of a page.
998         unsigned block_start, block_end;        // Starting and ending offsets of current
999         // buffer in the page.
1000         struct buffer_head *wait[2], **wait_bh = wait;  // Buffers for page, if
1001         // Page appeared to be not up
1002         // to date. Note how we have
1003         // at most 2 buffers, this is
1004         // because we at most may
1005         // partially overwrite two
1006         // buffers for one page. One at                                                 // the beginning of write area
1007         // and one at the end.
1008         // Everything inthe middle gets                                                 // overwritten totally.
1009
1010         struct cpu_key key;     // cpu key of item that we are going to deal with
1011         struct item_head *ih = NULL;    // pointer to item head that we are going to deal with
1012         struct buffer_head *itembuf = NULL;     // Buffer head that contains items that we are going to deal with
1013         INITIALIZE_PATH(path);  // path to item, that we are going to deal with.
1014         __le32 *item = NULL;    // pointer to item we are going to deal with
1015         int item_pos = -1;      /* Position in indirect item */
1016
1017         if (num_pages < 1) {
1018                 reiserfs_warning(inode->i_sb,
1019                                  "green-9001: reiserfs_prepare_file_region_for_write "
1020                                  "called with zero number of pages to process");
1021                 return -EFAULT;
1022         }
1023
1024         /* We have 2 loops for pages. In first loop we grab and lock the pages, so
1025            that nobody would touch these until we release the pages. Then
1026            we'd start to deal with mapping buffers to blocks. */
1027         for (i = 0; i < num_pages; i++) {
1028                 prepared_pages[i] = grab_cache_page(mapping, index + i);        // locks the page
1029                 if (!prepared_pages[i]) {
1030                         res = -ENOMEM;
1031                         goto failed_page_grabbing;
1032                 }
1033                 if (!page_has_buffers(prepared_pages[i]))
1034                         create_empty_buffers(prepared_pages[i],
1035                                              inode->i_sb->s_blocksize, 0);
1036         }
1037
1038         /* Let's count amount of blocks for a case where all the blocks
1039            overwritten are new (we will substract already allocated blocks later) */
1040         if (num_pages > 2)
1041                 /* These are full-overwritten pages so we count all the blocks in
1042                    these pages are counted as needed to be allocated */
1043                 blocks =
1044                     (num_pages - 2) << (PAGE_CACHE_SHIFT - inode->i_blkbits);
1045
1046         /* count blocks needed for first page (possibly partially written) */
1047         blocks += ((PAGE_CACHE_SIZE - from) >> inode->i_blkbits) + !!(from & (inode->i_sb->s_blocksize - 1));   /* roundup */
1048
1049         /* Now we account for last page. If last page == first page (we
1050            overwrite only one page), we substract all the blocks past the
1051            last writing position in a page out of already calculated number
1052            of blocks */
1053         blocks += ((num_pages > 1) << (PAGE_CACHE_SHIFT - inode->i_blkbits)) -
1054             ((PAGE_CACHE_SIZE - to) >> inode->i_blkbits);
1055         /* Note how we do not roundup here since partial blocks still
1056            should be allocated */
1057
1058         /* Now if all the write area lies past the file end, no point in
1059            maping blocks, since there is none, so we just zero out remaining
1060            parts of first and last pages in write area (if needed) */
1061         if ((pos & ~((loff_t) PAGE_CACHE_SIZE - 1)) > inode->i_size) {
1062                 if (from != 0)          /* First page needs to be partially zeroed */
1063                         zero_user_page(prepared_pages[0], 0, from, KM_USER0);
1064
1065                 if (to != PAGE_CACHE_SIZE)      /* Last page needs to be partially zeroed */
1066                         zero_user_page(prepared_pages[num_pages-1], to,
1067                                         PAGE_CACHE_SIZE - to, KM_USER0);
1068
1069                 /* Since all blocks are new - use already calculated value */
1070                 return blocks;
1071         }
1072
1073         /* Well, since we write somewhere into the middle of a file, there is
1074            possibility we are writing over some already allocated blocks, so
1075            let's map these blocks and substract number of such blocks out of blocks
1076            we need to allocate (calculated above) */
1077         /* Mask write position to start on blocksize, we do it out of the
1078            loop for performance reasons */
1079         pos &= ~((loff_t) inode->i_sb->s_blocksize - 1);
1080         /* Set cpu key to the starting position in a file (on left block boundary) */
1081         make_cpu_key(&key, inode,
1082                      1 + ((pos) & ~((loff_t) inode->i_sb->s_blocksize - 1)),
1083                      TYPE_ANY, 3 /*key length */ );
1084
1085         reiserfs_write_lock(inode->i_sb);       // We need that for at least search_by_key()
1086         for (i = 0; i < num_pages; i++) {
1087
1088                 head = page_buffers(prepared_pages[i]);
1089                 /* For each buffer in the page */
1090                 for (bh = head, block_start = 0; bh != head || !block_start;
1091                      block_start = block_end, bh = bh->b_this_page) {
1092                         if (!bh)
1093                                 reiserfs_panic(inode->i_sb,
1094                                                "green-9002: Allocated but absent buffer for a page?");
1095                         /* Find where this buffer ends */
1096                         block_end = block_start + inode->i_sb->s_blocksize;
1097                         if (i == 0 && block_end <= from)
1098                                 /* if this buffer is before requested data to map, skip it */
1099                                 continue;
1100
1101                         if (i == num_pages - 1 && block_start >= to) {
1102                                 /* If this buffer is after requested data to map, abort
1103                                    processing of current page */
1104                                 break;
1105                         }
1106
1107                         if (buffer_mapped(bh) && bh->b_blocknr != 0) {
1108                                 /* This is optimisation for a case where buffer is mapped
1109                                    and have blocknumber assigned. In case significant amount
1110                                    of such buffers are present, we may avoid some amount
1111                                    of search_by_key calls.
1112                                    Probably it would be possible to move parts of this code
1113                                    out of BKL, but I afraid that would overcomplicate code
1114                                    without any noticeable benefit.
1115                                  */
1116                                 item_pos++;
1117                                 /* Update the key */
1118                                 set_cpu_key_k_offset(&key,
1119                                                      cpu_key_k_offset(&key) +
1120                                                      inode->i_sb->s_blocksize);
1121                                 blocks--;       // Decrease the amount of blocks that need to be
1122                                 // allocated
1123                                 continue;       // Go to the next buffer
1124                         }
1125
1126                         if (!itembuf || /* if first iteration */
1127                             item_pos >= ih_item_len(ih) / UNFM_P_SIZE) {        /* or if we progressed past the
1128                                                                                    current unformatted_item */
1129                                 /* Try to find next item */
1130                                 res =
1131                                     search_for_position_by_key(inode->i_sb,
1132                                                                &key, &path);
1133                                 /* Abort if no more items */
1134                                 if (res != POSITION_FOUND) {
1135                                         /* make sure later loops don't use this item */
1136                                         itembuf = NULL;
1137                                         item = NULL;
1138                                         break;
1139                                 }
1140
1141                                 /* Update information about current indirect item */
1142                                 itembuf = get_last_bh(&path);
1143                                 ih = get_ih(&path);
1144                                 item = get_item(&path);
1145                                 item_pos = path.pos_in_item;
1146
1147                                 RFALSE(!is_indirect_le_ih(ih),
1148                                        "green-9003: indirect item expected");
1149                         }
1150
1151                         /* See if there is some block associated with the file
1152                            at that position, map the buffer to this block */
1153                         if (get_block_num(item, item_pos)) {
1154                                 map_bh(bh, inode->i_sb,
1155                                        get_block_num(item, item_pos));
1156                                 blocks--;       // Decrease the amount of blocks that need to be
1157                                 // allocated
1158                         }
1159                         item_pos++;
1160                         /* Update the key */
1161                         set_cpu_key_k_offset(&key,
1162                                              cpu_key_k_offset(&key) +
1163                                              inode->i_sb->s_blocksize);
1164                 }
1165         }
1166         pathrelse(&path);       // Free the path
1167         reiserfs_write_unlock(inode->i_sb);
1168
1169         /* Now zero out unmappend buffers for the first and last pages of
1170            write area or issue read requests if page is mapped. */
1171         /* First page, see if it is not uptodate */
1172         if (!PageUptodate(prepared_pages[0])) {
1173                 head = page_buffers(prepared_pages[0]);
1174
1175                 /* For each buffer in page */
1176                 for (bh = head, block_start = 0; bh != head || !block_start;
1177                      block_start = block_end, bh = bh->b_this_page) {
1178
1179                         if (!bh)
1180                                 reiserfs_panic(inode->i_sb,
1181                                                "green-9002: Allocated but absent buffer for a page?");
1182                         /* Find where this buffer ends */
1183                         block_end = block_start + inode->i_sb->s_blocksize;
1184                         if (block_end <= from)
1185                                 /* if this buffer is before requested data to map, skip it */
1186                                 continue;
1187                         if (block_start < from) {       /* Aha, our partial buffer */
1188                                 if (buffer_mapped(bh)) {        /* If it is mapped, we need to
1189                                                                    issue READ request for it to
1190                                                                    not loose data */
1191                                         ll_rw_block(READ, 1, &bh);
1192                                         *wait_bh++ = bh;
1193                                 } else {        /* Not mapped, zero it */
1194                                         zero_user_page(prepared_pages[0],
1195                                                        block_start,
1196                                                        from - block_start, KM_USER0);
1197                                         set_buffer_uptodate(bh);
1198                                 }
1199                         }
1200                 }
1201         }
1202
1203         /* Last page, see if it is not uptodate, or if the last page is past the end of the file. */
1204         if (!PageUptodate(prepared_pages[num_pages - 1]) ||
1205             ((pos + write_bytes) >> PAGE_CACHE_SHIFT) >
1206             (inode->i_size >> PAGE_CACHE_SHIFT)) {
1207                 head = page_buffers(prepared_pages[num_pages - 1]);
1208
1209                 /* for each buffer in page */
1210                 for (bh = head, block_start = 0; bh != head || !block_start;
1211                      block_start = block_end, bh = bh->b_this_page) {
1212
1213                         if (!bh)
1214                                 reiserfs_panic(inode->i_sb,
1215                                                "green-9002: Allocated but absent buffer for a page?");
1216                         /* Find where this buffer ends */
1217                         block_end = block_start + inode->i_sb->s_blocksize;
1218                         if (block_start >= to)
1219                                 /* if this buffer is after requested data to map, skip it */
1220                                 break;
1221                         if (block_end > to) {   /* Aha, our partial buffer */
1222                                 if (buffer_mapped(bh)) {        /* If it is mapped, we need to
1223                                                                    issue READ request for it to
1224                                                                    not loose data */
1225                                         ll_rw_block(READ, 1, &bh);
1226                                         *wait_bh++ = bh;
1227                                 } else {        /* Not mapped, zero it */
1228                                         zero_user_page(prepared_pages[num_pages-1],
1229                                                         to, block_end - to, KM_USER0);
1230                                         set_buffer_uptodate(bh);
1231                                 }
1232                         }
1233                 }
1234         }
1235
1236         /* Wait for read requests we made to happen, if necessary */
1237         while (wait_bh > wait) {
1238                 wait_on_buffer(*--wait_bh);
1239                 if (!buffer_uptodate(*wait_bh)) {
1240                         res = -EIO;
1241                         goto failed_read;
1242                 }
1243         }
1244
1245         return blocks;
1246       failed_page_grabbing:
1247         num_pages = i;
1248       failed_read:
1249         reiserfs_unprepare_pages(prepared_pages, num_pages);
1250         return res;
1251 }
1252
1253 /* Write @count bytes at position @ppos in a file indicated by @file
1254    from the buffer @buf.  
1255
1256    generic_file_write() is only appropriate for filesystems that are not seeking to optimize performance and want
1257    something simple that works.  It is not for serious use by general purpose filesystems, excepting the one that it was
1258    written for (ext2/3).  This is for several reasons:
1259
1260    * It has no understanding of any filesystem specific optimizations.
1261
1262    * It enters the filesystem repeatedly for each page that is written.
1263
1264    * It depends on reiserfs_get_block() function which if implemented by reiserfs performs costly search_by_key
1265    * operation for each page it is supplied with. By contrast reiserfs_file_write() feeds as much as possible at a time
1266    * to reiserfs which allows for fewer tree traversals.
1267
1268    * Each indirect pointer insertion takes a lot of cpu, because it involves memory moves inside of blocks.
1269
1270    * Asking the block allocation code for blocks one at a time is slightly less efficient.
1271
1272    All of these reasons for not using only generic file write were understood back when reiserfs was first miscoded to
1273    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
1274    things right finally.
1275
1276    Future Features: providing search_by_key with hints.
1277
1278 */
1279 static ssize_t reiserfs_file_write(struct file *file,   /* the file we are going to write into */
1280                                    const char __user * buf,     /*  pointer to user supplied data
1281                                                                    (in userspace) */
1282                                    size_t count,        /* amount of bytes to write */
1283                                    loff_t * ppos        /* pointer to position in file that we start writing at. Should be updated to
1284                                                          * new current position before returning. */
1285                                    )
1286 {
1287         size_t already_written = 0;     // Number of bytes already written to the file.
1288         loff_t pos;             // Current position in the file.
1289         ssize_t res;            // return value of various functions that we call.
1290         int err = 0;
1291         struct inode *inode = file->f_path.dentry->d_inode;     // Inode of the file that we are writing to.
1292         /* To simplify coding at this time, we store
1293            locked pages in array for now */
1294         struct page *prepared_pages[REISERFS_WRITE_PAGES_AT_A_TIME];
1295         struct reiserfs_transaction_handle th;
1296         th.t_trans_id = 0;
1297
1298         /* If a filesystem is converted from 3.5 to 3.6, we'll have v3.5 items
1299         * lying around (most of the disk, in fact). Despite the filesystem
1300         * now being a v3.6 format, the old items still can't support large
1301         * file sizes. Catch this case here, as the rest of the VFS layer is
1302         * oblivious to the different limitations between old and new items.
1303         * reiserfs_setattr catches this for truncates. This chunk is lifted
1304         * from generic_write_checks. */
1305         if (get_inode_item_key_version (inode) == KEY_FORMAT_3_5 &&
1306             *ppos + count > MAX_NON_LFS) {
1307                 if (*ppos >= MAX_NON_LFS) {
1308                         return -EFBIG;
1309                 }
1310                 if (count > MAX_NON_LFS - (unsigned long)*ppos)
1311                         count = MAX_NON_LFS - (unsigned long)*ppos;
1312         }
1313
1314         if (file->f_flags & O_DIRECT)
1315                 return do_sync_write(file, buf, count, ppos);
1316
1317         if (unlikely((ssize_t) count < 0))
1318                 return -EINVAL;
1319
1320         if (unlikely(!access_ok(VERIFY_READ, buf, count)))
1321                 return -EFAULT;
1322
1323         mutex_lock(&inode->i_mutex);    // locks the entire file for just us
1324
1325         pos = *ppos;
1326
1327         /* Check if we can write to specified region of file, file
1328            is not overly big and this kind of stuff. Adjust pos and
1329            count, if needed */
1330         res = generic_write_checks(file, &pos, &count, 0);
1331         if (res)
1332                 goto out;
1333
1334         if (count == 0)
1335                 goto out;
1336
1337         res = remove_suid(file->f_path.dentry);
1338         if (res)
1339                 goto out;
1340
1341         file_update_time(file);
1342
1343         // Ok, we are done with all the checks.
1344
1345         // Now we should start real work
1346
1347         /* If we are going to write past the file's packed tail or if we are going
1348            to overwrite part of the tail, we need that tail to be converted into
1349            unformatted node */
1350         res = reiserfs_check_for_tail_and_convert(inode, pos, count);
1351         if (res)
1352                 goto out;
1353
1354         while (count > 0) {
1355                 /* This is the main loop in which we running until some error occures
1356                    or until we write all of the data. */
1357                 size_t num_pages;       /* amount of pages we are going to write this iteration */
1358                 size_t write_bytes;     /* amount of bytes to write during this iteration */
1359                 size_t blocks_to_allocate;      /* how much blocks we need to allocate for this iteration */
1360
1361                 /*  (pos & (PAGE_CACHE_SIZE-1)) is an idiom for offset into a page of pos */
1362                 num_pages = !!((pos + count) & (PAGE_CACHE_SIZE - 1)) + /* round up partial
1363                                                                            pages */
1364                     ((count +
1365                       (pos & (PAGE_CACHE_SIZE - 1))) >> PAGE_CACHE_SHIFT);
1366                 /* convert size to amount of
1367                    pages */
1368                 reiserfs_write_lock(inode->i_sb);
1369                 if (num_pages > REISERFS_WRITE_PAGES_AT_A_TIME
1370                     || num_pages > reiserfs_can_fit_pages(inode->i_sb)) {
1371                         /* If we were asked to write more data than we want to or if there
1372                            is not that much space, then we shorten amount of data to write
1373                            for this iteration. */
1374                         num_pages =
1375                             min_t(size_t, REISERFS_WRITE_PAGES_AT_A_TIME,
1376                                   reiserfs_can_fit_pages(inode->i_sb));
1377                         /* Also we should not forget to set size in bytes accordingly */
1378                         write_bytes = (num_pages << PAGE_CACHE_SHIFT) -
1379                             (pos & (PAGE_CACHE_SIZE - 1));
1380                         /* If position is not on the
1381                            start of the page, we need
1382                            to substract the offset
1383                            within page */
1384                 } else
1385                         write_bytes = count;
1386
1387                 /* reserve the blocks to be allocated later, so that later on
1388                    we still have the space to write the blocks to */
1389                 reiserfs_claim_blocks_to_be_allocated(inode->i_sb,
1390                                                       num_pages <<
1391                                                       (PAGE_CACHE_SHIFT -
1392                                                        inode->i_blkbits));
1393                 reiserfs_write_unlock(inode->i_sb);
1394
1395                 if (!num_pages) {       /* If we do not have enough space even for a single page... */
1396                         if (pos >
1397                             inode->i_size + inode->i_sb->s_blocksize -
1398                             (pos & (inode->i_sb->s_blocksize - 1))) {
1399                                 res = -ENOSPC;
1400                                 break;  // In case we are writing past the end of the last file block, break.
1401                         }
1402                         // Otherwise we are possibly overwriting the file, so
1403                         // let's set write size to be equal or less than blocksize.
1404                         // This way we get it correctly for file holes.
1405                         // But overwriting files on absolutelly full volumes would not
1406                         // be very efficient. Well, people are not supposed to fill
1407                         // 100% of disk space anyway.
1408                         write_bytes =
1409                             min_t(size_t, count,
1410                                   inode->i_sb->s_blocksize -
1411                                   (pos & (inode->i_sb->s_blocksize - 1)));
1412                         num_pages = 1;
1413                         // No blocks were claimed before, so do it now.
1414                         reiserfs_claim_blocks_to_be_allocated(inode->i_sb,
1415                                                               1 <<
1416                                                               (PAGE_CACHE_SHIFT
1417                                                                -
1418                                                                inode->
1419                                                                i_blkbits));
1420                 }
1421
1422                 /* Prepare for writing into the region, read in all the
1423                    partially overwritten pages, if needed. And lock the pages,
1424                    so that nobody else can access these until we are done.
1425                    We get number of actual blocks needed as a result. */
1426                 res = reiserfs_prepare_file_region_for_write(inode, pos,
1427                                                              num_pages,
1428                                                              write_bytes,
1429                                                              prepared_pages);
1430                 if (res < 0) {
1431                         reiserfs_release_claimed_blocks(inode->i_sb,
1432                                                         num_pages <<
1433                                                         (PAGE_CACHE_SHIFT -
1434                                                          inode->i_blkbits));
1435                         break;
1436                 }
1437
1438                 blocks_to_allocate = res;
1439
1440                 /* First we correct our estimate of how many blocks we need */
1441                 reiserfs_release_claimed_blocks(inode->i_sb,
1442                                                 (num_pages <<
1443                                                  (PAGE_CACHE_SHIFT -
1444                                                   inode->i_sb->
1445                                                   s_blocksize_bits)) -
1446                                                 blocks_to_allocate);
1447
1448                 if (blocks_to_allocate > 0) {   /*We only allocate blocks if we need to */
1449                         /* Fill in all the possible holes and append the file if needed */
1450                         res =
1451                             reiserfs_allocate_blocks_for_region(&th, inode, pos,
1452                                                                 num_pages,
1453                                                                 write_bytes,
1454                                                                 prepared_pages,
1455                                                                 blocks_to_allocate);
1456                 }
1457
1458                 /* well, we have allocated the blocks, so it is time to free
1459                    the reservation we made earlier. */
1460                 reiserfs_release_claimed_blocks(inode->i_sb,
1461                                                 blocks_to_allocate);
1462                 if (res) {
1463                         reiserfs_unprepare_pages(prepared_pages, num_pages);
1464                         break;
1465                 }
1466
1467 /* NOTE that allocating blocks and filling blocks can be done in reverse order
1468    and probably we would do that just to get rid of garbage in files after a
1469    crash */
1470
1471                 /* Copy data from user-supplied buffer to file's pages */
1472                 res =
1473                     reiserfs_copy_from_user_to_file_region(pos, num_pages,
1474                                                            write_bytes,
1475                                                            prepared_pages, buf);
1476                 if (res) {
1477                         reiserfs_unprepare_pages(prepared_pages, num_pages);
1478                         break;
1479                 }
1480
1481                 /* Send the pages to disk and unlock them. */
1482                 res =
1483                     reiserfs_submit_file_region_for_write(&th, inode, pos,
1484                                                           num_pages,
1485                                                           write_bytes,
1486                                                           prepared_pages);
1487                 if (res)
1488                         break;
1489
1490                 already_written += write_bytes;
1491                 buf += write_bytes;
1492                 *ppos = pos += write_bytes;
1493                 count -= write_bytes;
1494                 balance_dirty_pages_ratelimited_nr(inode->i_mapping, num_pages);
1495         }
1496
1497         /* this is only true on error */
1498         if (th.t_trans_id) {
1499                 reiserfs_write_lock(inode->i_sb);
1500                 err = journal_end(&th, th.t_super, th.t_blocks_allocated);
1501                 reiserfs_write_unlock(inode->i_sb);
1502                 if (err) {
1503                         res = err;
1504                         goto out;
1505                 }
1506         }
1507
1508         if (likely(res >= 0) &&
1509             (unlikely((file->f_flags & O_SYNC) || IS_SYNC(inode))))
1510                 res = generic_osync_inode(inode, file->f_mapping,
1511                                           OSYNC_METADATA | OSYNC_DATA);
1512
1513         mutex_unlock(&inode->i_mutex);
1514         reiserfs_async_progress_wait(inode->i_sb);
1515         return (already_written != 0) ? already_written : res;
1516
1517       out:
1518         mutex_unlock(&inode->i_mutex);  // unlock the file on exit.
1519         return res;
1520 }
1521
1522 const struct file_operations reiserfs_file_operations = {
1523         .read = do_sync_read,
1524         .write = reiserfs_file_write,
1525         .ioctl = reiserfs_ioctl,
1526 #ifdef CONFIG_COMPAT
1527         .compat_ioctl = reiserfs_compat_ioctl,
1528 #endif
1529         .mmap = reiserfs_file_mmap,
1530         .open = generic_file_open,
1531         .release = reiserfs_file_release,
1532         .fsync = reiserfs_sync_file,
1533         .aio_read = generic_file_aio_read,
1534         .aio_write = generic_file_aio_write,
1535         .splice_read = generic_file_splice_read,
1536         .splice_write = generic_file_splice_write,
1537 };
1538
1539 const struct inode_operations reiserfs_file_inode_operations = {
1540         .truncate = reiserfs_vfs_truncate_file,
1541         .setattr = reiserfs_setattr,
1542         .setxattr = reiserfs_setxattr,
1543         .getxattr = reiserfs_getxattr,
1544         .listxattr = reiserfs_listxattr,
1545         .removexattr = reiserfs_removexattr,
1546         .permission = reiserfs_permission,
1547 };