2 * Copyright (C) 2008 Oracle. All rights reserved.
4 * This program is free software; you can redistribute it and/or
5 * modify it under the terms of the GNU General Public
6 * License v2 as published by the Free Software Foundation.
8 * This program is distributed in the hope that it will be useful,
9 * but WITHOUT ANY WARRANTY; without even the implied warranty of
10 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
11 * General Public License for more details.
13 * You should have received a copy of the GNU General Public
14 * License along with this program; if not, write to the
15 * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
16 * Boston, MA 021110-1307, USA.
18 * Based on jffs2 zlib code:
19 * Copyright © 2001-2007 Red Hat, Inc.
20 * Created by David Woodhouse <dwmw2@infradead.org>
23 #include <linux/kernel.h>
24 #include <linux/slab.h>
25 #include <linux/zlib.h>
26 #include <linux/zutil.h>
27 #include <linux/vmalloc.h>
28 #include <linux/init.h>
29 #include <linux/err.h>
30 #include <linux/sched.h>
31 #include <linux/pagemap.h>
32 #include <linux/bio.h>
33 #include "compression.h"
35 /* Plan: call deflate() with avail_in == *sourcelen,
36 avail_out = *dstlen - 12 and flush == Z_FINISH.
37 If it doesn't manage to finish, call it again with
38 avail_in == 0 and avail_out set to the remaining 12
39 bytes for it to clean up.
40 Q: Is 12 bytes sufficient?
42 #define STREAM_END_SPACE 12
48 struct list_head list;
51 static LIST_HEAD(idle_workspace);
52 static DEFINE_SPINLOCK(workspace_lock);
53 static unsigned long num_workspace;
54 static atomic_t alloc_workspace = ATOMIC_INIT(0);
55 static DECLARE_WAIT_QUEUE_HEAD(workspace_wait);
58 * this finds an available zlib workspace or allocates a new one
59 * NULL or an ERR_PTR is returned if things go bad.
61 static struct workspace *find_zlib_workspace(void)
63 struct workspace *workspace;
65 int cpus = num_online_cpus();
68 spin_lock(&workspace_lock);
69 if (!list_empty(&idle_workspace)) {
70 workspace = list_entry(idle_workspace.next, struct workspace,
72 list_del(&workspace->list);
74 spin_unlock(&workspace_lock);
78 spin_unlock(&workspace_lock);
79 if (atomic_read(&alloc_workspace) > cpus) {
81 prepare_to_wait(&workspace_wait, &wait, TASK_UNINTERRUPTIBLE);
82 if (atomic_read(&alloc_workspace) > cpus)
84 finish_wait(&workspace_wait, &wait);
87 atomic_inc(&alloc_workspace);
88 workspace = kzalloc(sizeof(*workspace), GFP_NOFS);
94 workspace->def_strm.workspace = vmalloc(zlib_deflate_workspacesize());
95 if (!workspace->def_strm.workspace) {
99 workspace->inf_strm.workspace = vmalloc(zlib_inflate_workspacesize());
100 if (!workspace->inf_strm.workspace) {
104 workspace->buf = kmalloc(PAGE_CACHE_SIZE, GFP_NOFS);
105 if (!workspace->buf) {
112 vfree(workspace->inf_strm.workspace);
114 vfree(workspace->def_strm.workspace);
117 atomic_dec(&alloc_workspace);
118 wake_up(&workspace_wait);
123 * put a workspace struct back on the list or free it if we have enough
124 * idle ones sitting around
126 static int free_workspace(struct workspace *workspace)
128 spin_lock(&workspace_lock);
129 if (num_workspace < num_online_cpus()) {
130 list_add_tail(&workspace->list, &idle_workspace);
132 spin_unlock(&workspace_lock);
133 if (waitqueue_active(&workspace_wait))
134 wake_up(&workspace_wait);
137 spin_unlock(&workspace_lock);
138 vfree(workspace->def_strm.workspace);
139 vfree(workspace->inf_strm.workspace);
140 kfree(workspace->buf);
143 atomic_dec(&alloc_workspace);
144 if (waitqueue_active(&workspace_wait))
145 wake_up(&workspace_wait);
150 * cleanup function for module exit
152 static void free_workspaces(void)
154 struct workspace *workspace;
155 while(!list_empty(&idle_workspace)) {
156 workspace = list_entry(idle_workspace.next, struct workspace,
158 list_del(&workspace->list);
159 vfree(workspace->def_strm.workspace);
160 vfree(workspace->inf_strm.workspace);
161 kfree(workspace->buf);
163 atomic_dec(&alloc_workspace);
168 * given an address space and start/len, compress the bytes.
170 * pages are allocated to hold the compressed result and stored
173 * out_pages is used to return the number of pages allocated. There
174 * may be pages allocated even if we return an error
176 * total_in is used to return the number of bytes actually read. It
177 * may be smaller then len if we had to exit early because we
178 * ran out of room in the pages array or because we cross the
181 * total_out is used to return the total number of compressed bytes
183 * max_out tells us the max number of bytes that we're allowed to
186 int btrfs_zlib_compress_pages(struct address_space *mapping,
187 u64 start, unsigned long len,
189 unsigned long nr_dest_pages,
190 unsigned long *out_pages,
191 unsigned long *total_in,
192 unsigned long *total_out,
193 unsigned long max_out)
196 struct workspace *workspace;
200 struct page *in_page = NULL;
201 struct page *out_page = NULL;
204 unsigned long bytes_left;
210 workspace = find_zlib_workspace();
214 if (Z_OK != zlib_deflateInit(&workspace->def_strm, 3)) {
215 printk(KERN_WARNING "deflateInit failed\n");
220 workspace->def_strm.total_in = 0;
221 workspace->def_strm.total_out = 0;
223 in_page = find_get_page(mapping, start >> PAGE_CACHE_SHIFT);
224 data_in = kmap(in_page);
226 out_page = alloc_page(GFP_NOFS | __GFP_HIGHMEM);
227 cpage_out = kmap(out_page);
231 workspace->def_strm.next_in = data_in;
232 workspace->def_strm.next_out = cpage_out;
233 workspace->def_strm.avail_out = PAGE_CACHE_SIZE;
234 workspace->def_strm.avail_in = min(len, PAGE_CACHE_SIZE);
239 while (workspace->def_strm.total_in < len) {
240 ret = zlib_deflate(&workspace->def_strm, Z_SYNC_FLUSH);
242 printk(KERN_DEBUG "btrfs deflate in loop returned %d\n",
244 zlib_deflateEnd(&workspace->def_strm);
249 /* we're making it bigger, give up */
250 if (workspace->def_strm.total_in > 8192 &&
251 workspace->def_strm.total_in <
252 workspace->def_strm.total_out) {
256 /* we need another page for writing out. Test this
257 * before the total_in so we will pull in a new page for
258 * the stream end if required
260 if (workspace->def_strm.avail_out == 0) {
262 if (nr_pages == nr_dest_pages) {
267 out_page = alloc_page(GFP_NOFS | __GFP_HIGHMEM);
268 cpage_out = kmap(out_page);
269 pages[nr_pages] = out_page;
271 workspace->def_strm.avail_out = PAGE_CACHE_SIZE;
272 workspace->def_strm.next_out = cpage_out;
275 if (workspace->def_strm.total_in >= len)
278 /* we've read in a full page, get a new one */
279 if (workspace->def_strm.avail_in == 0) {
280 if (workspace->def_strm.total_out > max_out)
283 bytes_left = len - workspace->def_strm.total_in;
285 page_cache_release(in_page);
287 start += PAGE_CACHE_SIZE;
288 in_page = find_get_page(mapping,
289 start >> PAGE_CACHE_SHIFT);
290 data_in = kmap(in_page);
291 workspace->def_strm.avail_in = min(bytes_left,
293 workspace->def_strm.next_in = data_in;
296 workspace->def_strm.avail_in = 0;
297 ret = zlib_deflate(&workspace->def_strm, Z_FINISH);
298 zlib_deflateEnd(&workspace->def_strm);
300 if (ret != Z_STREAM_END) {
305 if (workspace->def_strm.total_out >= workspace->def_strm.total_in) {
311 *total_out = workspace->def_strm.total_out;
312 *total_in = workspace->def_strm.total_in;
314 *out_pages = nr_pages;
320 page_cache_release(in_page);
322 free_workspace(workspace);
327 * pages_in is an array of pages with compressed data.
329 * disk_start is the starting logical offset of this array in the file
331 * bvec is a bio_vec of pages from the file that we want to decompress into
333 * vcnt is the count of pages in the biovec
335 * srclen is the number of bytes in pages_in
337 * The basic idea is that we have a bio that was created by readpages.
338 * The pages in the bio are for the uncompressed data, and they may not
339 * be contiguous. They all correspond to the range of bytes covered by
340 * the compressed extent.
342 int btrfs_zlib_decompress_biovec(struct page **pages_in,
344 struct bio_vec *bvec,
349 int wbits = MAX_WBITS;
350 struct workspace *workspace;
352 size_t total_out = 0;
353 unsigned long page_bytes_left;
354 unsigned long page_in_index = 0;
355 unsigned long page_out_index = 0;
356 struct page *page_out;
357 unsigned long total_pages_in = (srclen + PAGE_CACHE_SIZE - 1) /
359 unsigned long buf_start;
360 unsigned long buf_offset;
362 unsigned long working_bytes;
363 unsigned long pg_offset;
364 unsigned long start_byte;
365 unsigned long current_buf_start;
368 workspace = find_zlib_workspace();
372 data_in = kmap(pages_in[page_in_index]);
373 workspace->inf_strm.next_in = data_in;
374 workspace->inf_strm.avail_in = min_t(size_t, srclen, PAGE_CACHE_SIZE);
375 workspace->inf_strm.total_in = 0;
377 workspace->inf_strm.total_out = 0;
378 workspace->inf_strm.next_out = workspace->buf;
379 workspace->inf_strm.avail_out = PAGE_CACHE_SIZE;
380 page_out = bvec[page_out_index].bv_page;
381 page_bytes_left = PAGE_CACHE_SIZE;
384 /* If it's deflate, and it's got no preset dictionary, then
385 we can tell zlib to skip the adler32 check. */
386 if (srclen > 2 && !(data_in[1] & PRESET_DICT) &&
387 ((data_in[0] & 0x0f) == Z_DEFLATED) &&
388 !(((data_in[0]<<8) + data_in[1]) % 31)) {
390 wbits = -((data_in[0] >> 4) + 8);
391 workspace->inf_strm.next_in += 2;
392 workspace->inf_strm.avail_in -= 2;
395 if (Z_OK != zlib_inflateInit2(&workspace->inf_strm, wbits)) {
396 printk(KERN_WARNING "inflateInit failed\n");
400 while(workspace->inf_strm.total_in < srclen) {
401 ret = zlib_inflate(&workspace->inf_strm, Z_NO_FLUSH);
402 if (ret != Z_OK && ret != Z_STREAM_END) {
407 * buf start is the byte offset we're of the start of
408 * our workspace buffer
410 buf_start = total_out;
412 /* total_out is the last byte of the workspace buffer */
413 total_out = workspace->inf_strm.total_out;
415 working_bytes = total_out - buf_start;
418 * start byte is the first byte of the page we're currently
419 * copying into relative to the start of the compressed data.
421 start_byte = page_offset(page_out) - disk_start;
423 if (working_bytes == 0) {
424 /* we didn't make progress in this inflate
427 if (ret != Z_STREAM_END) {
433 /* we haven't yet hit data corresponding to this page */
434 if (total_out <= start_byte) {
439 * the start of the data we care about is offset into
440 * the middle of our working buffer
442 if (total_out > start_byte && buf_start < start_byte) {
443 buf_offset = start_byte - buf_start;
444 working_bytes -= buf_offset;
448 current_buf_start = buf_start;
450 /* copy bytes from the working buffer into the pages */
451 while(working_bytes > 0) {
452 bytes = min(PAGE_CACHE_SIZE - pg_offset,
453 PAGE_CACHE_SIZE - buf_offset);
454 bytes = min(bytes, working_bytes);
455 kaddr = kmap_atomic(page_out, KM_USER0);
456 memcpy(kaddr + pg_offset, workspace->buf + buf_offset,
458 kunmap_atomic(kaddr, KM_USER0);
459 flush_dcache_page(page_out);
462 page_bytes_left -= bytes;
464 working_bytes -= bytes;
465 current_buf_start += bytes;
467 /* check if we need to pick another page */
468 if (page_bytes_left == 0) {
470 if (page_out_index >= vcnt) {
474 page_out = bvec[page_out_index].bv_page;
476 page_bytes_left = PAGE_CACHE_SIZE;
477 start_byte = page_offset(page_out) - disk_start;
480 * make sure our new page is covered by this
483 if (total_out <= start_byte) {
487 /* the next page in the biovec might not
488 * be adjacent to the last page, but it
489 * might still be found inside this working
490 * buffer. bump our offset pointer
492 if (total_out > start_byte &&
493 current_buf_start < start_byte) {
494 buf_offset = start_byte - buf_start;
495 working_bytes = total_out - start_byte;
496 current_buf_start = buf_start +
502 workspace->inf_strm.next_out = workspace->buf;
503 workspace->inf_strm.avail_out = PAGE_CACHE_SIZE;
505 if (workspace->inf_strm.avail_in == 0) {
507 kunmap(pages_in[page_in_index]);
509 if (page_in_index >= total_pages_in) {
513 data_in = kmap(pages_in[page_in_index]);
514 workspace->inf_strm.next_in = data_in;
515 tmp = srclen - workspace->inf_strm.total_in;
516 workspace->inf_strm.avail_in = min(tmp,
520 if (ret != Z_STREAM_END) {
526 zlib_inflateEnd(&workspace->inf_strm);
528 kunmap(pages_in[page_in_index]);
530 free_workspace(workspace);
535 * a less complex decompression routine. Our compressed data fits in a
536 * single page, and we want to read a single page out of it.
537 * start_byte tells us the offset into the compressed data we're interested in
539 int btrfs_zlib_decompress(unsigned char *data_in,
540 struct page *dest_page,
541 unsigned long start_byte,
542 size_t srclen, size_t destlen)
545 int wbits = MAX_WBITS;
546 struct workspace *workspace;
547 unsigned long bytes_left = destlen;
548 unsigned long total_out = 0;
551 if (destlen > PAGE_CACHE_SIZE)
554 workspace = find_zlib_workspace();
558 workspace->inf_strm.next_in = data_in;
559 workspace->inf_strm.avail_in = srclen;
560 workspace->inf_strm.total_in = 0;
562 workspace->inf_strm.next_out = workspace->buf;
563 workspace->inf_strm.avail_out = PAGE_CACHE_SIZE;
564 workspace->inf_strm.total_out = 0;
565 /* If it's deflate, and it's got no preset dictionary, then
566 we can tell zlib to skip the adler32 check. */
567 if (srclen > 2 && !(data_in[1] & PRESET_DICT) &&
568 ((data_in[0] & 0x0f) == Z_DEFLATED) &&
569 !(((data_in[0]<<8) + data_in[1]) % 31)) {
571 wbits = -((data_in[0] >> 4) + 8);
572 workspace->inf_strm.next_in += 2;
573 workspace->inf_strm.avail_in -= 2;
576 if (Z_OK != zlib_inflateInit2(&workspace->inf_strm, wbits)) {
577 printk(KERN_WARNING "inflateInit failed\n");
582 while(bytes_left > 0) {
583 unsigned long buf_start;
584 unsigned long buf_offset;
586 unsigned long pg_offset = 0;
588 ret = zlib_inflate(&workspace->inf_strm, Z_NO_FLUSH);
589 if (ret != Z_OK && ret != Z_STREAM_END) {
593 buf_start = total_out;
594 total_out = workspace->inf_strm.total_out;
596 if (total_out == buf_start) {
601 if (total_out <= start_byte) {
605 if (total_out > start_byte && buf_start < start_byte) {
606 buf_offset = start_byte - buf_start;
611 bytes = min(PAGE_CACHE_SIZE - pg_offset,
612 PAGE_CACHE_SIZE - buf_offset);
613 bytes = min(bytes, bytes_left);
615 kaddr = kmap_atomic(dest_page, KM_USER0);
616 memcpy(kaddr + pg_offset, workspace->buf + buf_offset, bytes);
617 kunmap_atomic(kaddr, KM_USER0);
622 workspace->inf_strm.next_out = workspace->buf;
623 workspace->inf_strm.avail_out = PAGE_CACHE_SIZE;
625 if (ret != Z_STREAM_END && bytes_left != 0) {
630 zlib_inflateEnd(&workspace->inf_strm);
632 free_workspace(workspace);
636 void btrfs_zlib_exit(void)