vserver 1.9.5.x5
[linux-2.6.git] / fs / isofs / compress.c
1 /* -*- linux-c -*- ------------------------------------------------------- *
2  *   
3  *   Copyright 2001 H. Peter Anvin - All Rights Reserved
4  *
5  *   This program is free software; you can redistribute it and/or modify
6  *   it under the terms of the GNU General Public License as published by
7  *   the Free Software Foundation, Inc., 675 Mass Ave, Cambridge MA 02139,
8  *   USA; either version 2 of the License, or (at your option) any later
9  *   version; incorporated herein by reference.
10  *
11  * ----------------------------------------------------------------------- */
12
13 /*
14  * linux/fs/isofs/compress.c
15  *
16  * Transparent decompression of files on an iso9660 filesystem
17  */
18
19 #include <linux/config.h>
20 #include <linux/module.h>
21
22 #include <linux/stat.h>
23 #include <linux/time.h>
24 #include <linux/iso_fs.h>
25 #include <linux/kernel.h>
26 #include <linux/major.h>
27 #include <linux/mm.h>
28 #include <linux/string.h>
29 #include <linux/slab.h>
30 #include <linux/errno.h>
31 #include <linux/init.h>
32 #include <linux/nls.h>
33 #include <linux/ctype.h>
34 #include <linux/smp_lock.h>
35 #include <linux/blkdev.h>
36 #include <linux/vmalloc.h>
37 #include <linux/zlib.h>
38 #include <linux/buffer_head.h>
39
40 #include <asm/system.h>
41 #include <asm/uaccess.h>
42 #include <asm/semaphore.h>
43
44 #include "zisofs.h"
45
46 /* This should probably be global. */
47 static char zisofs_sink_page[PAGE_CACHE_SIZE];
48
49 /*
50  * This contains the zlib memory allocation and the mutex for the
51  * allocation; this avoids failures at block-decompression time.
52  */
53 static void *zisofs_zlib_workspace;
54 static struct semaphore zisofs_zlib_semaphore;
55
56 /*
57  * When decompressing, we typically obtain more than one page
58  * per reference.  We inject the additional pages into the page
59  * cache as a form of readahead.
60  */
61 static int zisofs_readpage(struct file *file, struct page *page)
62 {
63         struct inode *inode = file->f_dentry->d_inode;
64         struct address_space *mapping = inode->i_mapping;
65         unsigned int maxpage, xpage, fpage, blockindex;
66         unsigned long offset;
67         unsigned long blockptr, blockendptr, cstart, cend, csize;
68         struct buffer_head *bh, *ptrbh[2];
69         unsigned long bufsize = ISOFS_BUFFER_SIZE(inode);
70         unsigned int bufshift = ISOFS_BUFFER_BITS(inode);
71         unsigned long bufmask  = bufsize - 1;
72         int err = -EIO;
73         int i;
74         unsigned int header_size = ISOFS_I(inode)->i_format_parm[0];
75         unsigned int zisofs_block_shift = ISOFS_I(inode)->i_format_parm[1];
76         /* unsigned long zisofs_block_size = 1UL << zisofs_block_shift; */
77         unsigned int zisofs_block_page_shift = zisofs_block_shift-PAGE_CACHE_SHIFT;
78         unsigned long zisofs_block_pages = 1UL << zisofs_block_page_shift;
79         unsigned long zisofs_block_page_mask = zisofs_block_pages-1;
80         struct page *pages[zisofs_block_pages];
81         unsigned long index = page->index;
82         int indexblocks;
83
84         /* We have already been given one page, this is the one
85            we must do. */
86         xpage = index & zisofs_block_page_mask;
87         pages[xpage] = page;
88  
89         /* The remaining pages need to be allocated and inserted */
90         offset = index & ~zisofs_block_page_mask;
91         blockindex = offset >> zisofs_block_page_shift;
92         maxpage = (inode->i_size + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT;
93         maxpage = min(zisofs_block_pages, maxpage-offset);
94
95         for ( i = 0 ; i < maxpage ; i++, offset++ ) {
96                 if ( i != xpage ) {
97                         pages[i] = grab_cache_page_nowait(mapping, offset);
98                 }
99                 page = pages[i];
100                 if ( page ) {
101                         ClearPageError(page);
102                         kmap(page);
103                 }
104         }
105
106         /* This is the last page filled, plus one; used in case of abort. */
107         fpage = 0;
108
109         /* Find the pointer to this specific chunk */
110         /* Note: we're not using isonum_731() here because the data is known aligned */
111         /* Note: header_size is in 32-bit words (4 bytes) */
112         blockptr = (header_size + blockindex) << 2;
113         blockendptr = blockptr + 4;
114
115         indexblocks = ((blockptr^blockendptr) >> bufshift) ? 2 : 1;
116         ptrbh[0] = ptrbh[1] = NULL;
117
118         if ( isofs_get_blocks(inode, blockptr >> bufshift, ptrbh, indexblocks) != indexblocks ) {
119                 if ( ptrbh[0] ) brelse(ptrbh[0]);
120                 printk(KERN_DEBUG "zisofs: Null buffer on reading block table, inode = %lu, block = %lu\n",
121                        inode->i_ino, blockptr >> bufshift);
122                 goto eio;
123         }
124         ll_rw_block(READ, indexblocks, ptrbh);
125
126         bh = ptrbh[0];
127         if ( !bh || (wait_on_buffer(bh), !buffer_uptodate(bh)) ) {
128                 printk(KERN_DEBUG "zisofs: Failed to read block table, inode = %lu, block = %lu\n",
129                        inode->i_ino, blockptr >> bufshift);
130                 if ( ptrbh[1] )
131                         brelse(ptrbh[1]);
132                 goto eio;
133         }
134         cstart = le32_to_cpu(*(__le32 *)(bh->b_data + (blockptr & bufmask)));
135
136         if ( indexblocks == 2 ) {
137                 /* We just crossed a block boundary.  Switch to the next block */
138                 brelse(bh);
139                 bh = ptrbh[1];
140                 if ( !bh || (wait_on_buffer(bh), !buffer_uptodate(bh)) ) {
141                         printk(KERN_DEBUG "zisofs: Failed to read block table, inode = %lu, block = %lu\n",
142                                inode->i_ino, blockendptr >> bufshift);
143                         goto eio;
144                 }
145         }
146         cend = le32_to_cpu(*(__le32 *)(bh->b_data + (blockendptr & bufmask)));
147         brelse(bh);
148
149         csize = cend-cstart;
150
151         /* Now page[] contains an array of pages, any of which can be NULL,
152            and the locks on which we hold.  We should now read the data and
153            release the pages.  If the pages are NULL the decompressed data
154            for that particular page should be discarded. */
155         
156         if ( csize == 0 ) {
157                 /* This data block is empty. */
158
159                 for ( fpage = 0 ; fpage < maxpage ; fpage++ ) {
160                         if ( (page = pages[fpage]) != NULL ) {
161                                 memset(page_address(page), 0, PAGE_CACHE_SIZE);
162                                 
163                                 flush_dcache_page(page);
164                                 SetPageUptodate(page);
165                                 kunmap(page);
166                                 unlock_page(page);
167                                 if ( fpage == xpage )
168                                         err = 0; /* The critical page */
169                                 else
170                                         page_cache_release(page);
171                         }
172                 }
173         } else {
174                 /* This data block is compressed. */
175                 z_stream stream;
176                 int bail = 0, left_out = -1;
177                 int zerr;
178                 int needblocks = (csize + (cstart & bufmask) + bufmask) >> bufshift;
179                 int haveblocks;
180                 struct buffer_head *bhs[needblocks+1];
181                 struct buffer_head **bhptr;
182
183                 /* Because zlib is not thread-safe, do all the I/O at the top. */
184
185                 blockptr = cstart >> bufshift;
186                 memset(bhs, 0, (needblocks+1)*sizeof(struct buffer_head *));
187                 haveblocks = isofs_get_blocks(inode, blockptr, bhs, needblocks);
188                 ll_rw_block(READ, haveblocks, bhs);
189
190                 bhptr = &bhs[0];
191                 bh = *bhptr++;
192
193                 /* First block is special since it may be fractional.
194                    We also wait for it before grabbing the zlib
195                    semaphore; odds are that the subsequent blocks are
196                    going to come in in short order so we don't hold
197                    the zlib semaphore longer than necessary. */
198
199                 if ( !bh || (wait_on_buffer(bh), !buffer_uptodate(bh)) ) {
200                         printk(KERN_DEBUG "zisofs: Hit null buffer, fpage = %d, xpage = %d, csize = %ld\n",
201                                fpage, xpage, csize);
202                         goto b_eio;
203                 }
204                 stream.next_in  = bh->b_data + (cstart & bufmask);
205                 stream.avail_in = min(bufsize-(cstart & bufmask), csize);
206                 csize -= stream.avail_in;
207
208                 stream.workspace = zisofs_zlib_workspace;
209                 down(&zisofs_zlib_semaphore);
210                 
211                 zerr = zlib_inflateInit(&stream);
212                 if ( zerr != Z_OK ) {
213                         if ( err && zerr == Z_MEM_ERROR )
214                                 err = -ENOMEM;
215                         printk(KERN_DEBUG "zisofs: zisofs_inflateInit returned %d\n",
216                                zerr);
217                         goto z_eio;
218                 }
219
220                 while ( !bail && fpage < maxpage ) {
221                         page = pages[fpage];
222                         if ( page )
223                                 stream.next_out = page_address(page);
224                         else
225                                 stream.next_out = (void *)&zisofs_sink_page;
226                         stream.avail_out = PAGE_CACHE_SIZE;
227
228                         while ( stream.avail_out ) {
229                                 int ao, ai;
230                                 if ( stream.avail_in == 0 && left_out ) {
231                                         if ( !csize ) {
232                                                 printk(KERN_WARNING "zisofs: ZF read beyond end of input\n");
233                                                 bail = 1;
234                                                 break;
235                                         } else {
236                                                 bh = *bhptr++;
237                                                 if ( !bh ||
238                                                      (wait_on_buffer(bh), !buffer_uptodate(bh)) ) {
239                                                         /* Reached an EIO */
240                                                         printk(KERN_DEBUG "zisofs: Hit null buffer, fpage = %d, xpage = %d, csize = %ld\n",
241                                                                fpage, xpage, csize);
242                                                                
243                                                         bail = 1;
244                                                         break;
245                                                 }
246                                                 stream.next_in = bh->b_data;
247                                                 stream.avail_in = min(csize,bufsize);
248                                                 csize -= stream.avail_in;
249                                         }
250                                 }
251                                 ao = stream.avail_out;  ai = stream.avail_in;
252                                 zerr = zlib_inflate(&stream, Z_SYNC_FLUSH);
253                                 left_out = stream.avail_out;
254                                 if ( zerr == Z_BUF_ERROR && stream.avail_in == 0 )
255                                         continue;
256                                 if ( zerr != Z_OK ) {
257                                         /* EOF, error, or trying to read beyond end of input */
258                                         if ( err && zerr == Z_MEM_ERROR )
259                                                 err = -ENOMEM;
260                                         if ( zerr != Z_STREAM_END )
261                                                 printk(KERN_DEBUG "zisofs: zisofs_inflate returned %d, inode = %lu, index = %lu, fpage = %d, xpage = %d, avail_in = %d, avail_out = %d, ai = %d, ao = %d\n",
262                                                        zerr, inode->i_ino, index,
263                                                        fpage, xpage,
264                                                        stream.avail_in, stream.avail_out,
265                                                        ai, ao);
266                                         bail = 1;
267                                         break;
268                                 }
269                         }
270
271                         if ( stream.avail_out && zerr == Z_STREAM_END ) {
272                                 /* Fractional page written before EOF.  This may
273                                    be the last page in the file. */
274                                 memset(stream.next_out, 0, stream.avail_out);
275                                 stream.avail_out = 0;
276                         }
277
278                         if ( !stream.avail_out ) {
279                                 /* This page completed */
280                                 if ( page ) {
281                                         flush_dcache_page(page);
282                                         SetPageUptodate(page);
283                                         kunmap(page);
284                                         unlock_page(page);
285                                         if ( fpage == xpage )
286                                                 err = 0; /* The critical page */
287                                         else
288                                                 page_cache_release(page);
289                                 }
290                                 fpage++;
291                         }
292                 }
293                 zlib_inflateEnd(&stream);
294
295         z_eio:
296                 up(&zisofs_zlib_semaphore);
297
298         b_eio:
299                 for ( i = 0 ; i < haveblocks ; i++ ) {
300                         if ( bhs[i] )
301                                 brelse(bhs[i]);
302                 }
303         }
304
305 eio:
306
307         /* Release any residual pages, do not SetPageUptodate */
308         while ( fpage < maxpage ) {
309                 page = pages[fpage];
310                 if ( page ) {
311                         flush_dcache_page(page);
312                         if ( fpage == xpage )
313                                 SetPageError(page);
314                         kunmap(page);
315                         unlock_page(page);
316                         if ( fpage != xpage )
317                                 page_cache_release(page);
318                 }
319                 fpage++;
320         }                       
321
322         /* At this point, err contains 0 or -EIO depending on the "critical" page */
323         return err;
324 }
325
326 struct address_space_operations zisofs_aops = {
327         .readpage = zisofs_readpage,
328         /* No sync_page operation supported? */
329         /* No bmap operation supported */
330 };
331
332 static int initialized;
333
334 int __init zisofs_init(void)
335 {
336         if ( initialized ) {
337                 printk("zisofs_init: called more than once\n");
338                 return 0;
339         }
340
341         zisofs_zlib_workspace = vmalloc(zlib_inflate_workspacesize());
342         if ( !zisofs_zlib_workspace )
343                 return -ENOMEM;
344         init_MUTEX(&zisofs_zlib_semaphore);
345
346         initialized = 1;
347         return 0;
348 }
349
350 void zisofs_cleanup(void)
351 {
352         if ( !initialized ) {
353                 printk("zisofs_cleanup: called without initialization\n");
354                 return;
355         }
356
357         vfree(zisofs_zlib_workspace);
358         initialized = 0;
359 }