mm, hugetlb: get rid of surplus page accounting tricks
[linux/fpc-iii.git] / fs / isofs / compress.c
blob9bb2fe35799d3d306fde5b06ff699af753d66aff
1 /* -*- linux-c -*- ------------------------------------------------------- *
2 *
3 * Copyright 2001 H. Peter Anvin - All Rights Reserved
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.
11 * ----------------------------------------------------------------------- */
14 * linux/fs/isofs/compress.c
16 * Transparent decompression of files on an iso9660 filesystem
19 #include <linux/module.h>
20 #include <linux/init.h>
21 #include <linux/bio.h>
23 #include <linux/vmalloc.h>
24 #include <linux/zlib.h>
26 #include "isofs.h"
27 #include "zisofs.h"
29 /* This should probably be global. */
30 static char zisofs_sink_page[PAGE_SIZE];
33 * This contains the zlib memory allocation and the mutex for the
34 * allocation; this avoids failures at block-decompression time.
36 static void *zisofs_zlib_workspace;
37 static DEFINE_MUTEX(zisofs_zlib_lock);
40 * Read data of @inode from @block_start to @block_end and uncompress
41 * to one zisofs block. Store the data in the @pages array with @pcount
42 * entries. Start storing at offset @poffset of the first page.
44 static loff_t zisofs_uncompress_block(struct inode *inode, loff_t block_start,
45 loff_t block_end, int pcount,
46 struct page **pages, unsigned poffset,
47 int *errp)
49 unsigned int zisofs_block_shift = ISOFS_I(inode)->i_format_parm[1];
50 unsigned int bufsize = ISOFS_BUFFER_SIZE(inode);
51 unsigned int bufshift = ISOFS_BUFFER_BITS(inode);
52 unsigned int bufmask = bufsize - 1;
53 int i, block_size = block_end - block_start;
54 z_stream stream = { .total_out = 0,
55 .avail_in = 0,
56 .avail_out = 0, };
57 int zerr;
58 int needblocks = (block_size + (block_start & bufmask) + bufmask)
59 >> bufshift;
60 int haveblocks;
61 blkcnt_t blocknum;
62 struct buffer_head *bhs[needblocks + 1];
63 int curbh, curpage;
65 if (block_size > deflateBound(1UL << zisofs_block_shift)) {
66 *errp = -EIO;
67 return 0;
69 /* Empty block? */
70 if (block_size == 0) {
71 for ( i = 0 ; i < pcount ; i++ ) {
72 if (!pages[i])
73 continue;
74 memset(page_address(pages[i]), 0, PAGE_SIZE);
75 flush_dcache_page(pages[i]);
76 SetPageUptodate(pages[i]);
78 return ((loff_t)pcount) << PAGE_SHIFT;
81 /* Because zlib is not thread-safe, do all the I/O at the top. */
82 blocknum = block_start >> bufshift;
83 memset(bhs, 0, (needblocks + 1) * sizeof(struct buffer_head *));
84 haveblocks = isofs_get_blocks(inode, blocknum, bhs, needblocks);
85 ll_rw_block(REQ_OP_READ, 0, haveblocks, bhs);
87 curbh = 0;
88 curpage = 0;
90 * First block is special since it may be fractional. We also wait for
91 * it before grabbing the zlib mutex; odds are that the subsequent
92 * blocks are going to come in in short order so we don't hold the zlib
93 * mutex longer than necessary.
96 if (!bhs[0])
97 goto b_eio;
99 wait_on_buffer(bhs[0]);
100 if (!buffer_uptodate(bhs[0])) {
101 *errp = -EIO;
102 goto b_eio;
105 stream.workspace = zisofs_zlib_workspace;
106 mutex_lock(&zisofs_zlib_lock);
108 zerr = zlib_inflateInit(&stream);
109 if (zerr != Z_OK) {
110 if (zerr == Z_MEM_ERROR)
111 *errp = -ENOMEM;
112 else
113 *errp = -EIO;
114 printk(KERN_DEBUG "zisofs: zisofs_inflateInit returned %d\n",
115 zerr);
116 goto z_eio;
119 while (curpage < pcount && curbh < haveblocks &&
120 zerr != Z_STREAM_END) {
121 if (!stream.avail_out) {
122 if (pages[curpage]) {
123 stream.next_out = page_address(pages[curpage])
124 + poffset;
125 stream.avail_out = PAGE_SIZE - poffset;
126 poffset = 0;
127 } else {
128 stream.next_out = (void *)&zisofs_sink_page;
129 stream.avail_out = PAGE_SIZE;
132 if (!stream.avail_in) {
133 wait_on_buffer(bhs[curbh]);
134 if (!buffer_uptodate(bhs[curbh])) {
135 *errp = -EIO;
136 break;
138 stream.next_in = bhs[curbh]->b_data +
139 (block_start & bufmask);
140 stream.avail_in = min_t(unsigned, bufsize -
141 (block_start & bufmask),
142 block_size);
143 block_size -= stream.avail_in;
144 block_start = 0;
147 while (stream.avail_out && stream.avail_in) {
148 zerr = zlib_inflate(&stream, Z_SYNC_FLUSH);
149 if (zerr == Z_BUF_ERROR && stream.avail_in == 0)
150 break;
151 if (zerr == Z_STREAM_END)
152 break;
153 if (zerr != Z_OK) {
154 /* EOF, error, or trying to read beyond end of input */
155 if (zerr == Z_MEM_ERROR)
156 *errp = -ENOMEM;
157 else {
158 printk(KERN_DEBUG
159 "zisofs: zisofs_inflate returned"
160 " %d, inode = %lu,"
161 " page idx = %d, bh idx = %d,"
162 " avail_in = %ld,"
163 " avail_out = %ld\n",
164 zerr, inode->i_ino, curpage,
165 curbh, stream.avail_in,
166 stream.avail_out);
167 *errp = -EIO;
169 goto inflate_out;
173 if (!stream.avail_out) {
174 /* This page completed */
175 if (pages[curpage]) {
176 flush_dcache_page(pages[curpage]);
177 SetPageUptodate(pages[curpage]);
179 curpage++;
181 if (!stream.avail_in)
182 curbh++;
184 inflate_out:
185 zlib_inflateEnd(&stream);
187 z_eio:
188 mutex_unlock(&zisofs_zlib_lock);
190 b_eio:
191 for (i = 0; i < haveblocks; i++)
192 brelse(bhs[i]);
193 return stream.total_out;
197 * Uncompress data so that pages[full_page] is fully uptodate and possibly
198 * fills in other pages if we have data for them.
200 static int zisofs_fill_pages(struct inode *inode, int full_page, int pcount,
201 struct page **pages)
203 loff_t start_off, end_off;
204 loff_t block_start, block_end;
205 unsigned int header_size = ISOFS_I(inode)->i_format_parm[0];
206 unsigned int zisofs_block_shift = ISOFS_I(inode)->i_format_parm[1];
207 unsigned int blockptr;
208 loff_t poffset = 0;
209 blkcnt_t cstart_block, cend_block;
210 struct buffer_head *bh;
211 unsigned int blkbits = ISOFS_BUFFER_BITS(inode);
212 unsigned int blksize = 1 << blkbits;
213 int err;
214 loff_t ret;
216 BUG_ON(!pages[full_page]);
219 * We want to read at least 'full_page' page. Because we have to
220 * uncompress the whole compression block anyway, fill the surrounding
221 * pages with the data we have anyway...
223 start_off = page_offset(pages[full_page]);
224 end_off = min_t(loff_t, start_off + PAGE_SIZE, inode->i_size);
226 cstart_block = start_off >> zisofs_block_shift;
227 cend_block = (end_off + (1 << zisofs_block_shift) - 1)
228 >> zisofs_block_shift;
230 WARN_ON(start_off - (full_page << PAGE_SHIFT) !=
231 ((cstart_block << zisofs_block_shift) & PAGE_MASK));
233 /* Find the pointer to this specific chunk */
234 /* Note: we're not using isonum_731() here because the data is known aligned */
235 /* Note: header_size is in 32-bit words (4 bytes) */
236 blockptr = (header_size + cstart_block) << 2;
237 bh = isofs_bread(inode, blockptr >> blkbits);
238 if (!bh)
239 return -EIO;
240 block_start = le32_to_cpu(*(__le32 *)
241 (bh->b_data + (blockptr & (blksize - 1))));
243 while (cstart_block < cend_block && pcount > 0) {
244 /* Load end of the compressed block in the file */
245 blockptr += 4;
246 /* Traversed to next block? */
247 if (!(blockptr & (blksize - 1))) {
248 brelse(bh);
250 bh = isofs_bread(inode, blockptr >> blkbits);
251 if (!bh)
252 return -EIO;
254 block_end = le32_to_cpu(*(__le32 *)
255 (bh->b_data + (blockptr & (blksize - 1))));
256 if (block_start > block_end) {
257 brelse(bh);
258 return -EIO;
260 err = 0;
261 ret = zisofs_uncompress_block(inode, block_start, block_end,
262 pcount, pages, poffset, &err);
263 poffset += ret;
264 pages += poffset >> PAGE_SHIFT;
265 pcount -= poffset >> PAGE_SHIFT;
266 full_page -= poffset >> PAGE_SHIFT;
267 poffset &= ~PAGE_MASK;
269 if (err) {
270 brelse(bh);
272 * Did we finish reading the page we really wanted
273 * to read?
275 if (full_page < 0)
276 return 0;
277 return err;
280 block_start = block_end;
281 cstart_block++;
284 if (poffset && *pages) {
285 memset(page_address(*pages) + poffset, 0,
286 PAGE_SIZE - poffset);
287 flush_dcache_page(*pages);
288 SetPageUptodate(*pages);
290 return 0;
294 * When decompressing, we typically obtain more than one page
295 * per reference. We inject the additional pages into the page
296 * cache as a form of readahead.
298 static int zisofs_readpage(struct file *file, struct page *page)
300 struct inode *inode = file_inode(file);
301 struct address_space *mapping = inode->i_mapping;
302 int err;
303 int i, pcount, full_page;
304 unsigned int zisofs_block_shift = ISOFS_I(inode)->i_format_parm[1];
305 unsigned int zisofs_pages_per_cblock =
306 PAGE_SHIFT <= zisofs_block_shift ?
307 (1 << (zisofs_block_shift - PAGE_SHIFT)) : 0;
308 struct page *pages[max_t(unsigned, zisofs_pages_per_cblock, 1)];
309 pgoff_t index = page->index, end_index;
311 end_index = (inode->i_size + PAGE_SIZE - 1) >> PAGE_SHIFT;
313 * If this page is wholly outside i_size we just return zero;
314 * do_generic_file_read() will handle this for us
316 if (index >= end_index) {
317 SetPageUptodate(page);
318 unlock_page(page);
319 return 0;
322 if (PAGE_SHIFT <= zisofs_block_shift) {
323 /* We have already been given one page, this is the one
324 we must do. */
325 full_page = index & (zisofs_pages_per_cblock - 1);
326 pcount = min_t(int, zisofs_pages_per_cblock,
327 end_index - (index & ~(zisofs_pages_per_cblock - 1)));
328 index -= full_page;
329 } else {
330 full_page = 0;
331 pcount = 1;
333 pages[full_page] = page;
335 for (i = 0; i < pcount; i++, index++) {
336 if (i != full_page)
337 pages[i] = grab_cache_page_nowait(mapping, index);
338 if (pages[i]) {
339 ClearPageError(pages[i]);
340 kmap(pages[i]);
344 err = zisofs_fill_pages(inode, full_page, pcount, pages);
346 /* Release any residual pages, do not SetPageUptodate */
347 for (i = 0; i < pcount; i++) {
348 if (pages[i]) {
349 flush_dcache_page(pages[i]);
350 if (i == full_page && err)
351 SetPageError(pages[i]);
352 kunmap(pages[i]);
353 unlock_page(pages[i]);
354 if (i != full_page)
355 put_page(pages[i]);
359 /* At this point, err contains 0 or -EIO depending on the "critical" page */
360 return err;
363 const struct address_space_operations zisofs_aops = {
364 .readpage = zisofs_readpage,
365 /* No bmap operation supported */
368 int __init zisofs_init(void)
370 zisofs_zlib_workspace = vmalloc(zlib_inflate_workspacesize());
371 if ( !zisofs_zlib_workspace )
372 return -ENOMEM;
374 return 0;
377 void zisofs_cleanup(void)
379 vfree(zisofs_zlib_workspace);