2 * ramdisk.c - Multiple RAM disk driver - gzip-loading version - v. 0.8 beta.
4 * (C) Chad Page, Theodore Ts'o, et. al, 1995.
6 * This RAM disk is designed to have filesystems created on it and mounted
7 * just like a regular floppy disk.
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.
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.
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.
27 * Loadable module support added by Tom Dyas.
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
34 * Added initrd: Werner Almesberger & Hans Lermen, Feb '96
36 * 4/25/96 : Made RAM disk size a parameter (default is now 4 MB)
39 * Add support for fs images split across >1 disk, Paul Gortmaker, Mar '98
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
45 #include <linux/config.h>
46 #include <linux/string.h>
47 #include <linux/slab.h>
48 #include <asm/atomic.h>
49 #include <linux/bio.h>
50 #include <linux/module.h>
51 #include <linux/init.h>
52 #include <linux/devfs_fs_kernel.h>
53 #include <linux/pagemap.h>
54 #include <linux/blkdev.h>
55 #include <linux/genhd.h>
56 #include <linux/buffer_head.h> /* for invalidate_bdev() */
57 #include <linux/backing-dev.h>
58 #include <linux/blkpg.h>
59 #include <linux/writeback.h>
61 #include <asm/uaccess.h>
63 /* The RAM disk size is now a parameter */
64 #define NUM_RAMDISKS 16 /* This cannot be overridden (yet) */
66 /* Various static variables go here. Most are used only in the RAM disk code.
69 static struct gendisk *rd_disks[NUM_RAMDISKS];
70 static struct block_device *rd_bdev[NUM_RAMDISKS];/* Protected device data */
71 static struct request_queue *rd_queue[NUM_RAMDISKS];
74 * Parameters for the boot-loading of the RAM disk. These are set by
75 * init/main.c (from arguments to the kernel command line) or from the
76 * architecture-specific setup routine (from the stored boot sector
79 int rd_size = CONFIG_BLK_DEV_RAM_SIZE; /* Size of the RAM disks */
81 * It would be very desirable to have a soft-blocksize (that in the case
82 * of the ramdisk driver is also the hardblocksize ;) of PAGE_SIZE because
83 * doing that we'll achieve a far better MM footprint. Using a rd_blocksize of
84 * BLOCK_SIZE in the worst case we'll make PAGE_SIZE/BLOCK_SIZE buffer-pages
85 * unfreeable. With a rd_blocksize of PAGE_SIZE instead we are sure that only
86 * 1 page will be protected. Depending on the size of the ramdisk you
87 * may want to change the ramdisk blocksize to achieve a better or worse MM
88 * behaviour. The default is still BLOCK_SIZE (needed by rd_load_image that
89 * supposes the filesystem in the image uses a BLOCK_SIZE blocksize).
91 int rd_blocksize = BLOCK_SIZE; /* blocksize of the RAM disks */
94 * Copyright (C) 2000 Linus Torvalds.
95 * 2000 Transmeta Corp.
96 * aops copied from ramfs.
100 * If a ramdisk page has buffers, some may be uptodate and some may be not.
101 * To bring the page uptodate we zero out the non-uptodate buffers. The
102 * page must be locked.
104 static void make_page_uptodate(struct page *page)
106 if (page_has_buffers(page)) {
107 struct buffer_head *bh = page_buffers(page);
108 struct buffer_head *head = bh;
111 if (!buffer_uptodate(bh))
112 memset(bh->b_data, 0, bh->b_size);
113 } while ((bh = bh->b_this_page) != head);
115 memset(page_address(page), 0, PAGE_CACHE_SIZE);
117 flush_dcache_page(page);
118 SetPageUptodate(page);
121 static int ramdisk_readpage(struct file *file, struct page *page)
123 if (!PageUptodate(page))
124 make_page_uptodate(page);
129 static int ramdisk_prepare_write(struct file *file, struct page *page,
130 unsigned offset, unsigned to)
132 if (!PageUptodate(page))
133 make_page_uptodate(page);
137 static int ramdisk_commit_write(struct file *file, struct page *page,
138 unsigned offset, unsigned to)
140 set_page_dirty(page);
145 * ->writepage to the the blockdev's mapping has to redirty the page so that the
146 * VM doesn't go and steal it. We return WRITEPAGE_ACTIVATE so that the VM
147 * won't try to (pointlessly) write the page again for a while.
149 * Really, these pages should not be on the LRU at all.
151 static int ramdisk_writepage(struct page *page, struct writeback_control *wbc)
153 if (!PageUptodate(page))
154 make_page_uptodate(page);
156 if (wbc->for_reclaim)
157 return WRITEPAGE_ACTIVATE;
163 * This is a little speedup thing: short-circuit attempts to write back the
164 * ramdisk blockdev inode to its non-existent backing store.
166 static int ramdisk_writepages(struct address_space *mapping,
167 struct writeback_control *wbc)
173 * ramdisk blockdev pages have their own ->set_page_dirty() because we don't
174 * want them to contribute to dirty memory accounting.
176 static int ramdisk_set_page_dirty(struct page *page)
182 static struct address_space_operations ramdisk_aops = {
183 .readpage = ramdisk_readpage,
184 .prepare_write = ramdisk_prepare_write,
185 .commit_write = ramdisk_commit_write,
186 .writepage = ramdisk_writepage,
187 .set_page_dirty = ramdisk_set_page_dirty,
188 .writepages = ramdisk_writepages,
191 static int rd_blkdev_pagecache_IO(int rw, struct bio_vec *vec, sector_t sector,
192 struct address_space *mapping)
194 pgoff_t index = sector >> (PAGE_CACHE_SHIFT - 9);
195 unsigned int vec_offset = vec->bv_offset;
196 int offset = (sector << 9) & ~PAGE_CACHE_MASK;
197 int size = vec->bv_len;
206 count = PAGE_CACHE_SIZE - offset;
211 page = grab_cache_page(mapping, index);
217 if (!PageUptodate(page))
218 make_page_uptodate(page);
223 src = kmap_atomic(page, KM_USER0) + offset;
224 dst = kmap_atomic(vec->bv_page, KM_USER1) + vec_offset;
226 src = kmap_atomic(vec->bv_page, KM_USER0) + vec_offset;
227 dst = kmap_atomic(page, KM_USER1) + offset;
232 memcpy(dst, src, count);
234 kunmap_atomic(src, KM_USER0);
235 kunmap_atomic(dst, KM_USER1);
238 flush_dcache_page(vec->bv_page);
240 set_page_dirty(page);
250 * Basically, my strategy here is to set up a buffer-head which can't be
251 * deleted, and make that my Ramdisk. If the request is outside of the
252 * allocated size, we must get rid of it...
254 * 19-JAN-1998 Richard Gooch <rgooch@atnf.csiro.au> Added devfs support
257 static int rd_make_request(request_queue_t *q, struct bio *bio)
259 struct block_device *bdev = bio->bi_bdev;
260 struct address_space * mapping = bdev->bd_inode->i_mapping;
261 sector_t sector = bio->bi_sector;
262 unsigned long len = bio->bi_size >> 9;
263 int rw = bio_data_dir(bio);
264 struct bio_vec *bvec;
267 if (sector + len > get_capacity(bdev->bd_disk))
273 bio_for_each_segment(bvec, bio, i) {
274 ret |= rd_blkdev_pagecache_IO(rw, bvec, sector, mapping);
275 sector += bvec->bv_len >> 9;
280 bio_endio(bio, bio->bi_size, 0);
283 bio_io_error(bio, bio->bi_size);
287 static int rd_ioctl(struct inode *inode, struct file *file,
288 unsigned int cmd, unsigned long arg)
291 struct block_device *bdev = inode->i_bdev;
293 if (cmd != BLKFLSBUF)
297 * special: we want to release the ramdisk memory, it's not like with
298 * the other blockdevices where this ioctl only flushes away the buffer
303 if (bdev->bd_openers <= 2) {
304 truncate_inode_pages(bdev->bd_inode->i_mapping, 0);
312 * This is the backing_dev_info for the blockdev inode itself. It doesn't need
313 * writeback and it does not contribute to dirty memory accounting.
315 static struct backing_dev_info rd_backing_dev_info = {
316 .ra_pages = 0, /* No readahead */
317 .memory_backed = 1, /* Does not contribute to dirty memory */
318 .unplug_io_fn = default_unplug_io_fn,
322 * This is the backing_dev_info for the files which live atop the ramdisk
323 * "device". These files do need writeback and they do contribute to dirty
326 static struct backing_dev_info rd_file_backing_dev_info = {
327 .ra_pages = 0, /* No readahead */
328 .memory_backed = 0, /* Does contribute to dirty memory */
329 .unplug_io_fn = default_unplug_io_fn,
332 static int rd_open(struct inode *inode, struct file *filp)
334 unsigned unit = iminor(inode);
336 if (rd_bdev[unit] == NULL) {
337 struct block_device *bdev = inode->i_bdev;
338 struct address_space *mapping;
341 inode = igrab(bdev->bd_inode);
342 rd_bdev[unit] = bdev;
344 bdev->bd_block_size = rd_blocksize;
345 inode->i_size = get_capacity(rd_disks[unit])<<9;
346 mapping = inode->i_mapping;
347 mapping->a_ops = &ramdisk_aops;
348 mapping->backing_dev_info = &rd_backing_dev_info;
349 bdev->bd_inode_backing_dev_info = &rd_file_backing_dev_info;
352 * Deep badness. rd_blkdev_pagecache_IO() needs to allocate
353 * pagecache pages within a request_fn. We cannot recur back
354 * into the filesytem which is mounted atop the ramdisk, because
355 * that would deadlock on fs locks. And we really don't want
356 * to reenter rd_blkdev_pagecache_IO when we're already within
359 * So we turn off __GFP_FS and __GFP_IO.
361 * And to give this thing a hope of working, turn on __GFP_HIGH.
362 * Hopefully, there's enough regular memory allocation going on
363 * for the page allocator emergency pools to keep the ramdisk
366 gfp_mask = mapping_gfp_mask(mapping);
367 gfp_mask &= ~(__GFP_FS|__GFP_IO);
368 gfp_mask |= __GFP_HIGH;
369 mapping_set_gfp_mask(mapping, gfp_mask);
375 static struct block_device_operations rd_bd_op = {
376 .owner = THIS_MODULE,
382 * Before freeing the module, invalidate all of the protected buffers!
384 static void __exit rd_cleanup(void)
388 for (i = 0; i < NUM_RAMDISKS; i++) {
389 struct block_device *bdev = rd_bdev[i];
392 invalidate_bdev(bdev, 1);
395 del_gendisk(rd_disks[i]);
396 put_disk(rd_disks[i]);
397 blk_cleanup_queue(rd_queue[i]);
400 unregister_blkdev(RAMDISK_MAJOR, "ramdisk");
404 * This is the registration and initialization section of the RAM disk driver
406 static int __init rd_init(void)
411 if (rd_blocksize > PAGE_SIZE || rd_blocksize < 512 ||
412 (rd_blocksize & (rd_blocksize-1))) {
413 printk("RAMDISK: wrong blocksize %d, reverting to defaults\n",
415 rd_blocksize = BLOCK_SIZE;
418 for (i = 0; i < NUM_RAMDISKS; i++) {
419 rd_disks[i] = alloc_disk(1);
424 if (register_blkdev(RAMDISK_MAJOR, "ramdisk")) {
431 for (i = 0; i < NUM_RAMDISKS; i++) {
432 struct gendisk *disk = rd_disks[i];
434 rd_queue[i] = blk_alloc_queue(GFP_KERNEL);
438 blk_queue_make_request(rd_queue[i], &rd_make_request);
440 /* rd_size is given in kB */
441 disk->major = RAMDISK_MAJOR;
442 disk->first_minor = i;
443 disk->fops = &rd_bd_op;
444 disk->queue = rd_queue[i];
445 disk->flags |= GENHD_FL_SUPPRESS_PARTITION_INFO;
446 sprintf(disk->disk_name, "ram%d", i);
447 sprintf(disk->devfs_name, "rd/%d", i);
448 set_capacity(disk, rd_size * 2);
449 add_disk(rd_disks[i]);
452 /* rd_size is given in kB */
453 printk("RAMDISK driver initialized: "
454 "%d RAM disks of %dK size %d blocksize\n",
455 NUM_RAMDISKS, rd_size, rd_blocksize);
459 unregister_blkdev(RAMDISK_MAJOR, "ramdisk");
462 put_disk(rd_disks[i]);
463 blk_cleanup_queue(rd_queue[i]);
468 module_init(rd_init);
469 module_exit(rd_cleanup);
471 /* options - nonmodular */
473 static int __init ramdisk_size(char *str)
475 rd_size = simple_strtol(str,NULL,0);
478 static int __init ramdisk_size2(char *str) /* kludge */
480 return ramdisk_size(str);
482 static int __init ramdisk_blocksize(char *str)
484 rd_blocksize = simple_strtol(str,NULL,0);
487 __setup("ramdisk=", ramdisk_size);
488 __setup("ramdisk_size=", ramdisk_size2);
489 __setup("ramdisk_blocksize=", ramdisk_blocksize);
492 /* options - modular */
493 MODULE_PARM (rd_size, "1i");
494 MODULE_PARM_DESC(rd_size, "Size of each RAM disk in kbytes.");
495 MODULE_PARM (rd_blocksize, "i");
496 MODULE_PARM_DESC(rd_blocksize, "Blocksize of each RAM disk in bytes.");
498 MODULE_LICENSE("GPL");