aboutsummaryrefslogtreecommitdiffstats
path: root/drivers/block/rd.c
diff options
context:
space:
mode:
authorNick Piggin <npiggin@suse.de>2008-02-08 07:19:49 -0500
committerLinus Torvalds <torvalds@woody.linux-foundation.org>2008-02-08 12:22:30 -0500
commit9db5579be4bb5320c3248f6acf807aedf05ae143 (patch)
treefde09bbeb427946b30d2e0fb6b00494a42488052 /drivers/block/rd.c
parentdaeb51e62cacde31c8245866e1096ff79a0c83fe (diff)
rewrite rd
This is a rewrite of the ramdisk block device driver. The old one is really difficult because it effectively implements a block device which serves data out of its own buffer cache. It relies on the dirty bit being set, to pin its backing store in cache, however there are non trivial paths which can clear the dirty bit (eg. try_to_free_buffers()), which had recently lead to data corruption. And in general it is completely wrong for a block device driver to do this. The new one is more like a regular block device driver. It has no idea about vm/vfs stuff. It's backing store is similar to the buffer cache (a simple radix-tree of pages), but it doesn't know anything about page cache (the pages in the radix tree are not pagecache pages). There is one slight downside -- direct block device access and filesystem metadata access goes through an extra copy and gets stored in RAM twice. However, this downside is only slight, because the real buffercache of the device is now reclaimable (because we're not playing crazy games with it), so under memory intensive situations, footprint should effectively be the same -- maybe even a slight advantage to the new driver because it can also reclaim buffer heads. The fact that it now goes through all the regular vm/fs paths makes it much more useful for testing, too. text data bss dec hex filename 2837 849 384 4070 fe6 drivers/block/rd.o 3528 371 12 3911 f47 drivers/block/brd.o Text is larger, but data and bss are smaller, making total size smaller. A few other nice things about it: - Similar structure and layout to the new loop device handlinag. - Dynamic ramdisk creation. - Runtime flexible buffer head size (because it is no longer part of the ramdisk code). - Boot / load time flexible ramdisk size, which could easily be extended to a per-ramdisk runtime changeable size (eg. with an ioctl). - Can use highmem for the backing store. [akpm@linux-foundation.org: fix build] [byron.bbradley@gmail.com: make rd_size non-static] Signed-off-by: Nick Piggin <npiggin@suse.de> Signed-off-by: Byron Bradley <byron.bbradley@gmail.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
Diffstat (limited to 'drivers/block/rd.c')
-rw-r--r--drivers/block/rd.c537
1 files changed, 0 insertions, 537 deletions
diff --git a/drivers/block/rd.c b/drivers/block/rd.c
deleted file mode 100644
index 06e23be70904..000000000000
--- a/drivers/block/rd.c
+++ /dev/null
@@ -1,537 +0,0 @@
1/*
2 * ramdisk.c - Multiple RAM disk driver - gzip-loading version - v. 0.8 beta.
3 *
4 * (C) Chad Page, Theodore Ts'o, et. al, 1995.
5 *
6 * This RAM disk is designed to have filesystems created on it and mounted
7 * just like a regular floppy disk.
8 *
9 * It also does something suggested by Linus: use the buffer cache as the
10 * RAM disk data. This makes it possible to dynamically allocate the RAM disk
11 * buffer - with some consequences I have to deal with as I write this.
12 *
13 * This code is based on the original ramdisk.c, written mostly by
14 * Theodore Ts'o (TYT) in 1991. The code was largely rewritten by
15 * Chad Page to use the buffer cache to store the RAM disk data in
16 * 1995; Theodore then took over the driver again, and cleaned it up
17 * for inclusion in the mainline kernel.
18 *
19 * The original CRAMDISK code was written by Richard Lyons, and
20 * adapted by Chad Page to use the new RAM disk interface. Theodore
21 * Ts'o rewrote it so that both the compressed RAM disk loader and the
22 * kernel decompressor uses the same inflate.c codebase. The RAM disk
23 * loader now also loads into a dynamic (buffer cache based) RAM disk,
24 * not the old static RAM disk. Support for the old static RAM disk has
25 * been completely removed.
26 *
27 * Loadable module support added by Tom Dyas.
28 *
29 * Further cleanups by Chad Page (page0588@sundance.sjsu.edu):
30 * Cosmetic changes in #ifdef MODULE, code movement, etc.
31 * When the RAM disk module is removed, free the protected buffers
32 * Default RAM disk size changed to 2.88 MB
33 *
34 * Added initrd: Werner Almesberger & Hans Lermen, Feb '96
35 *
36 * 4/25/96 : Made RAM disk size a parameter (default is now 4 MB)
37 * - Chad Page
38 *
39 * Add support for fs images split across >1 disk, Paul Gortmaker, Mar '98
40 *
41 * Make block size and block size shift for RAM disks a global macro
42 * and set blk_size for -ENOSPC, Werner Fink <werner@suse.de>, Apr '99
43 */
44
45#include <linux/string.h>
46#include <linux/slab.h>
47#include <asm/atomic.h>
48#include <linux/bio.h>
49#include <linux/module.h>
50#include <linux/moduleparam.h>
51#include <linux/init.h>
52#include <linux/pagemap.h>
53#include <linux/blkdev.h>
54#include <linux/genhd.h>
55#include <linux/buffer_head.h> /* for invalidate_bdev() */
56#include <linux/backing-dev.h>
57#include <linux/blkpg.h>
58#include <linux/writeback.h>
59#include <linux/log2.h>
60
61#include <asm/uaccess.h>
62
63/* Various static variables go here. Most are used only in the RAM disk code.
64 */
65
66static struct gendisk *rd_disks[CONFIG_BLK_DEV_RAM_COUNT];
67static struct block_device *rd_bdev[CONFIG_BLK_DEV_RAM_COUNT];/* Protected device data */
68static struct request_queue *rd_queue[CONFIG_BLK_DEV_RAM_COUNT];
69
70/*
71 * Parameters for the boot-loading of the RAM disk. These are set by
72 * init/main.c (from arguments to the kernel command line) or from the
73 * architecture-specific setup routine (from the stored boot sector
74 * information).
75 */
76int rd_size = CONFIG_BLK_DEV_RAM_SIZE; /* Size of the RAM disks */
77/*
78 * It would be very desirable to have a soft-blocksize (that in the case
79 * of the ramdisk driver is also the hardblocksize ;) of PAGE_SIZE because
80 * doing that we'll achieve a far better MM footprint. Using a rd_blocksize of
81 * BLOCK_SIZE in the worst case we'll make PAGE_SIZE/BLOCK_SIZE buffer-pages
82 * unfreeable. With a rd_blocksize of PAGE_SIZE instead we are sure that only
83 * 1 page will be protected. Depending on the size of the ramdisk you
84 * may want to change the ramdisk blocksize to achieve a better or worse MM
85 * behaviour. The default is still BLOCK_SIZE (needed by rd_load_image that
86 * supposes the filesystem in the image uses a BLOCK_SIZE blocksize).
87 */
88static int rd_blocksize = CONFIG_BLK_DEV_RAM_BLOCKSIZE;
89
90/*
91 * Copyright (C) 2000 Linus Torvalds.
92 * 2000 Transmeta Corp.
93 * aops copied from ramfs.
94 */
95
96/*
97 * If a ramdisk page has buffers, some may be uptodate and some may be not.
98 * To bring the page uptodate we zero out the non-uptodate buffers. The
99 * page must be locked.
100 */
101static void make_page_uptodate(struct page *page)
102{
103 if (page_has_buffers(page)) {
104 struct buffer_head *bh = page_buffers(page);
105 struct buffer_head *head = bh;
106
107 do {
108 if (!buffer_uptodate(bh)) {
109 memset(bh->b_data, 0, bh->b_size);
110 /*
111 * akpm: I'm totally undecided about this. The
112 * buffer has just been magically brought "up to
113 * date", but nobody should want to be reading
114 * it anyway, because it hasn't been used for
115 * anything yet. It is still in a "not read
116 * from disk yet" state.
117 *
118 * But non-uptodate buffers against an uptodate
119 * page are against the rules. So do it anyway.
120 */
121 set_buffer_uptodate(bh);
122 }
123 } while ((bh = bh->b_this_page) != head);
124 } else {
125 memset(page_address(page), 0, PAGE_CACHE_SIZE);
126 }
127 flush_dcache_page(page);
128 SetPageUptodate(page);
129}
130
131static int ramdisk_readpage(struct file *file, struct page *page)
132{
133 if (!PageUptodate(page))
134 make_page_uptodate(page);
135 unlock_page(page);
136 return 0;
137}
138
139static int ramdisk_prepare_write(struct file *file, struct page *page,
140 unsigned offset, unsigned to)
141{
142 if (!PageUptodate(page))
143 make_page_uptodate(page);
144 return 0;
145}
146
147static int ramdisk_commit_write(struct file *file, struct page *page,
148 unsigned offset, unsigned to)
149{
150 set_page_dirty(page);
151 return 0;
152}
153
154/*
155 * ->writepage to the blockdev's mapping has to redirty the page so that the
156 * VM doesn't go and steal it. We return AOP_WRITEPAGE_ACTIVATE so that the VM
157 * won't try to (pointlessly) write the page again for a while.
158 *
159 * Really, these pages should not be on the LRU at all.
160 */
161static int ramdisk_writepage(struct page *page, struct writeback_control *wbc)
162{
163 if (!PageUptodate(page))
164 make_page_uptodate(page);
165 SetPageDirty(page);
166 if (wbc->for_reclaim)
167 return AOP_WRITEPAGE_ACTIVATE;
168 unlock_page(page);
169 return 0;
170}
171
172/*
173 * This is a little speedup thing: short-circuit attempts to write back the
174 * ramdisk blockdev inode to its non-existent backing store.
175 */
176static int ramdisk_writepages(struct address_space *mapping,
177 struct writeback_control *wbc)
178{
179 return 0;
180}
181
182/*
183 * ramdisk blockdev pages have their own ->set_page_dirty() because we don't
184 * want them to contribute to dirty memory accounting.
185 */
186static int ramdisk_set_page_dirty(struct page *page)
187{
188 if (!TestSetPageDirty(page))
189 return 1;
190 return 0;
191}
192
193/*
194 * releasepage is called by pagevec_strip/try_to_release_page if
195 * buffers_heads_over_limit is true. Without a releasepage function
196 * try_to_free_buffers is called instead. That can unset the dirty
197 * bit of our ram disk pages, which will be eventually freed, even
198 * if the page is still in use.
199 */
200static int ramdisk_releasepage(struct page *page, gfp_t dummy)
201{
202 return 0;
203}
204
205static const struct address_space_operations ramdisk_aops = {
206 .readpage = ramdisk_readpage,
207 .prepare_write = ramdisk_prepare_write,
208 .commit_write = ramdisk_commit_write,
209 .writepage = ramdisk_writepage,
210 .set_page_dirty = ramdisk_set_page_dirty,
211 .writepages = ramdisk_writepages,
212 .releasepage = ramdisk_releasepage,
213};
214
215static int rd_blkdev_pagecache_IO(int rw, struct bio_vec *vec, sector_t sector,
216 struct address_space *mapping)
217{
218 pgoff_t index = sector >> (PAGE_CACHE_SHIFT - 9);
219 unsigned int vec_offset = vec->bv_offset;
220 int offset = (sector << 9) & ~PAGE_CACHE_MASK;
221 int size = vec->bv_len;
222 int err = 0;
223
224 do {
225 int count;
226 struct page *page;
227 char *src;
228 char *dst;
229
230 count = PAGE_CACHE_SIZE - offset;
231 if (count > size)
232 count = size;
233 size -= count;
234
235 page = grab_cache_page(mapping, index);
236 if (!page) {
237 err = -ENOMEM;
238 goto out;
239 }
240
241 if (!PageUptodate(page))
242 make_page_uptodate(page);
243
244 index++;
245
246 if (rw == READ) {
247 src = kmap_atomic(page, KM_USER0) + offset;
248 dst = kmap_atomic(vec->bv_page, KM_USER1) + vec_offset;
249 } else {
250 src = kmap_atomic(vec->bv_page, KM_USER0) + vec_offset;
251 dst = kmap_atomic(page, KM_USER1) + offset;
252 }
253 offset = 0;
254 vec_offset += count;
255
256 memcpy(dst, src, count);
257
258 kunmap_atomic(src, KM_USER0);
259 kunmap_atomic(dst, KM_USER1);
260
261 if (rw == READ)
262 flush_dcache_page(vec->bv_page);
263 else
264 set_page_dirty(page);
265 unlock_page(page);
266 put_page(page);
267 } while (size);
268
269 out:
270 return err;
271}
272
273/*
274 * Basically, my strategy here is to set up a buffer-head which can't be
275 * deleted, and make that my Ramdisk. If the request is outside of the
276 * allocated size, we must get rid of it...
277 *
278 * 19-JAN-1998 Richard Gooch <rgooch@atnf.csiro.au> Added devfs support
279 *
280 */
281static int rd_make_request(struct request_queue *q, struct bio *bio)
282{
283 struct block_device *bdev = bio->bi_bdev;
284 struct address_space * mapping = bdev->bd_inode->i_mapping;
285 sector_t sector = bio->bi_sector;
286 unsigned long len = bio->bi_size >> 9;
287 int rw = bio_data_dir(bio);
288 struct bio_vec *bvec;
289 int ret = 0, i;
290
291 if (sector + len > get_capacity(bdev->bd_disk))
292 goto fail;
293
294 if (rw==READA)
295 rw=READ;
296
297 bio_for_each_segment(bvec, bio, i) {
298 ret |= rd_blkdev_pagecache_IO(rw, bvec, sector, mapping);
299 sector += bvec->bv_len >> 9;
300 }
301 if (ret)
302 goto fail;
303
304 bio_endio(bio, 0);
305 return 0;
306fail:
307 bio_io_error(bio);
308 return 0;
309}
310
311static int rd_ioctl(struct inode *inode, struct file *file,
312 unsigned int cmd, unsigned long arg)
313{
314 int error;
315 struct block_device *bdev = inode->i_bdev;
316
317 if (cmd != BLKFLSBUF)
318 return -ENOTTY;
319
320 /*
321 * special: we want to release the ramdisk memory, it's not like with
322 * the other blockdevices where this ioctl only flushes away the buffer
323 * cache
324 */
325 error = -EBUSY;
326 mutex_lock(&bdev->bd_mutex);
327 if (bdev->bd_openers <= 2) {
328 truncate_inode_pages(bdev->bd_inode->i_mapping, 0);
329 error = 0;
330 }
331 mutex_unlock(&bdev->bd_mutex);
332 return error;
333}
334
335/*
336 * This is the backing_dev_info for the blockdev inode itself. It doesn't need
337 * writeback and it does not contribute to dirty memory accounting.
338 */
339static struct backing_dev_info rd_backing_dev_info = {
340 .ra_pages = 0, /* No readahead */
341 .capabilities = BDI_CAP_NO_ACCT_DIRTY | BDI_CAP_NO_WRITEBACK | BDI_CAP_MAP_COPY,
342 .unplug_io_fn = default_unplug_io_fn,
343};
344
345/*
346 * This is the backing_dev_info for the files which live atop the ramdisk
347 * "device". These files do need writeback and they do contribute to dirty
348 * memory accounting.
349 */
350static struct backing_dev_info rd_file_backing_dev_info = {
351 .ra_pages = 0, /* No readahead */
352 .capabilities = BDI_CAP_MAP_COPY, /* Does contribute to dirty memory */
353 .unplug_io_fn = default_unplug_io_fn,
354};
355
356static int rd_open(struct inode *inode, struct file *filp)
357{
358 unsigned unit = iminor(inode);
359
360 if (rd_bdev[unit] == NULL) {
361 struct block_device *bdev = inode->i_bdev;
362 struct address_space *mapping;
363 unsigned bsize;
364 gfp_t gfp_mask;
365
366 inode = igrab(bdev->bd_inode);
367 rd_bdev[unit] = bdev;
368 bdev->bd_openers++;
369 bsize = bdev_hardsect_size(bdev);
370 bdev->bd_block_size = bsize;
371 inode->i_blkbits = blksize_bits(bsize);
372 inode->i_size = get_capacity(bdev->bd_disk)<<9;
373
374 mapping = inode->i_mapping;
375 mapping->a_ops = &ramdisk_aops;
376 mapping->backing_dev_info = &rd_backing_dev_info;
377 bdev->bd_inode_backing_dev_info = &rd_file_backing_dev_info;
378
379 /*
380 * Deep badness. rd_blkdev_pagecache_IO() needs to allocate
381 * pagecache pages within a request_fn. We cannot recur back
382 * into the filesystem which is mounted atop the ramdisk, because
383 * that would deadlock on fs locks. And we really don't want
384 * to reenter rd_blkdev_pagecache_IO when we're already within
385 * that function.
386 *
387 * So we turn off __GFP_FS and __GFP_IO.
388 *
389 * And to give this thing a hope of working, turn on __GFP_HIGH.
390 * Hopefully, there's enough regular memory allocation going on
391 * for the page allocator emergency pools to keep the ramdisk
392 * driver happy.
393 */
394 gfp_mask = mapping_gfp_mask(mapping);
395 gfp_mask &= ~(__GFP_FS|__GFP_IO);
396 gfp_mask |= __GFP_HIGH;
397 mapping_set_gfp_mask(mapping, gfp_mask);
398 }
399
400 return 0;
401}
402
403static struct block_device_operations rd_bd_op = {
404 .owner = THIS_MODULE,
405 .open = rd_open,
406 .ioctl = rd_ioctl,
407};
408
409/*
410 * Before freeing the module, invalidate all of the protected buffers!
411 */
412static void __exit rd_cleanup(void)
413{
414 int i;
415
416 for (i = 0; i < CONFIG_BLK_DEV_RAM_COUNT; i++) {
417 struct block_device *bdev = rd_bdev[i];
418 rd_bdev[i] = NULL;
419 if (bdev) {
420 invalidate_bdev(bdev);
421 blkdev_put(bdev);
422 }
423 del_gendisk(rd_disks[i]);
424 put_disk(rd_disks[i]);
425 blk_cleanup_queue(rd_queue[i]);
426 }
427 unregister_blkdev(RAMDISK_MAJOR, "ramdisk");
428
429 bdi_destroy(&rd_file_backing_dev_info);
430 bdi_destroy(&rd_backing_dev_info);
431}
432
433/*
434 * This is the registration and initialization section of the RAM disk driver
435 */
436static int __init rd_init(void)
437{
438 int i;
439 int err;
440
441 err = bdi_init(&rd_backing_dev_info);
442 if (err)
443 goto out2;
444
445 err = bdi_init(&rd_file_backing_dev_info);
446 if (err) {
447 bdi_destroy(&rd_backing_dev_info);
448 goto out2;
449 }
450
451 err = -ENOMEM;
452
453 if (rd_blocksize > PAGE_SIZE || rd_blocksize < 512 ||
454 !is_power_of_2(rd_blocksize)) {
455 printk("RAMDISK: wrong blocksize %d, reverting to defaults\n",
456 rd_blocksize);
457 rd_blocksize = BLOCK_SIZE;
458 }
459
460 for (i = 0; i < CONFIG_BLK_DEV_RAM_COUNT; i++) {
461 rd_disks[i] = alloc_disk(1);
462 if (!rd_disks[i])
463 goto out;
464
465 rd_queue[i] = blk_alloc_queue(GFP_KERNEL);
466 if (!rd_queue[i]) {
467 put_disk(rd_disks[i]);
468 goto out;
469 }
470 }
471
472 if (register_blkdev(RAMDISK_MAJOR, "ramdisk")) {
473 err = -EIO;
474 goto out;
475 }
476
477 for (i = 0; i < CONFIG_BLK_DEV_RAM_COUNT; i++) {
478 struct gendisk *disk = rd_disks[i];
479
480 blk_queue_make_request(rd_queue[i], &rd_make_request);
481 blk_queue_hardsect_size(rd_queue[i], rd_blocksize);
482
483 /* rd_size is given in kB */
484 disk->major = RAMDISK_MAJOR;
485 disk->first_minor = i;
486 disk->fops = &rd_bd_op;
487 disk->queue = rd_queue[i];
488 disk->flags |= GENHD_FL_SUPPRESS_PARTITION_INFO;
489 sprintf(disk->disk_name, "ram%d", i);
490 set_capacity(disk, rd_size * 2);
491 add_disk(rd_disks[i]);
492 }
493
494 /* rd_size is given in kB */
495 printk("RAMDISK driver initialized: "
496 "%d RAM disks of %dK size %d blocksize\n",
497 CONFIG_BLK_DEV_RAM_COUNT, rd_size, rd_blocksize);
498
499 return 0;
500out:
501 while (i--) {
502 put_disk(rd_disks[i]);
503 blk_cleanup_queue(rd_queue[i]);
504 }
505 bdi_destroy(&rd_backing_dev_info);
506 bdi_destroy(&rd_file_backing_dev_info);
507out2:
508 return err;
509}
510
511module_init(rd_init);
512module_exit(rd_cleanup);
513
514/* options - nonmodular */
515#ifndef MODULE
516static int __init ramdisk_size(char *str)
517{
518 rd_size = simple_strtol(str,NULL,0);
519 return 1;
520}
521static int __init ramdisk_blocksize(char *str)
522{
523 rd_blocksize = simple_strtol(str,NULL,0);
524 return 1;
525}
526__setup("ramdisk_size=", ramdisk_size);
527__setup("ramdisk_blocksize=", ramdisk_blocksize);
528#endif
529
530/* options - modular */
531module_param(rd_size, int, 0);
532MODULE_PARM_DESC(rd_size, "Size of each RAM disk in kbytes.");
533module_param(rd_blocksize, int, 0);
534MODULE_PARM_DESC(rd_blocksize, "Blocksize of each RAM disk in bytes.");
535MODULE_ALIAS_BLOCKDEV_MAJOR(RAMDISK_MAJOR);
536
537MODULE_LICENSE("GPL");
generated by cgit v1.2.2 (git 2.25.0) at 2024-12-14 03:19:49 -0500