Thomas Gleixner | 09c434b | 2019-05-19 13:08:20 +0100 | [diff] [blame] | 1 | // SPDX-License-Identifier: GPL-2.0-only |
Nick Piggin | 9db5579 | 2008-02-08 04:19:49 -0800 | [diff] [blame] | 2 | /* |
| 3 | * Ram backed block device driver. |
| 4 | * |
| 5 | * Copyright (C) 2007 Nick Piggin |
| 6 | * Copyright (C) 2007 Novell Inc. |
| 7 | * |
| 8 | * Parts derived from drivers/block/rd.c, and drivers/block/loop.c, copyright |
| 9 | * of their respective owners. |
| 10 | */ |
| 11 | |
| 12 | #include <linux/init.h> |
Bart Van Assche | 287f3ca | 2017-07-10 15:51:10 -0700 | [diff] [blame] | 13 | #include <linux/initrd.h> |
Nick Piggin | 9db5579 | 2008-02-08 04:19:49 -0800 | [diff] [blame] | 14 | #include <linux/module.h> |
| 15 | #include <linux/moduleparam.h> |
| 16 | #include <linux/major.h> |
| 17 | #include <linux/blkdev.h> |
| 18 | #include <linux/bio.h> |
| 19 | #include <linux/highmem.h> |
Arnd Bergmann | 2a48fc0 | 2010-06-02 14:28:52 +0200 | [diff] [blame] | 20 | #include <linux/mutex.h> |
Nick Piggin | 9db5579 | 2008-02-08 04:19:49 -0800 | [diff] [blame] | 21 | #include <linux/radix-tree.h> |
Al Viro | ff01bb4 | 2011-09-16 02:31:11 -0400 | [diff] [blame] | 22 | #include <linux/fs.h> |
Tejun Heo | 5a0e3ad | 2010-03-24 17:04:11 +0900 | [diff] [blame] | 23 | #include <linux/slab.h> |
Minchan Kim | 23c47d2 | 2017-11-15 17:33:00 -0800 | [diff] [blame] | 24 | #include <linux/backing-dev.h> |
Nick Piggin | 9db5579 | 2008-02-08 04:19:49 -0800 | [diff] [blame] | 25 | |
Linus Torvalds | 7c0f6ba | 2016-12-24 11:46:01 -0800 | [diff] [blame] | 26 | #include <linux/uaccess.h> |
Nick Piggin | 9db5579 | 2008-02-08 04:19:49 -0800 | [diff] [blame] | 27 | |
Nick Piggin | 9db5579 | 2008-02-08 04:19:49 -0800 | [diff] [blame] | 28 | #define PAGE_SECTORS_SHIFT (PAGE_SHIFT - SECTOR_SHIFT) |
| 29 | #define PAGE_SECTORS (1 << PAGE_SECTORS_SHIFT) |
| 30 | |
| 31 | /* |
| 32 | * Each block ramdisk device has a radix_tree brd_pages of pages that stores |
| 33 | * the pages containing the block device's contents. A brd page's ->index is |
| 34 | * its offset in PAGE_SIZE units. This is similar to, but in no way connected |
| 35 | * with, the kernel's pagecache or buffer cache (which sit above our block |
| 36 | * device). |
| 37 | */ |
| 38 | struct brd_device { |
| 39 | int brd_number; |
Nick Piggin | 9db5579 | 2008-02-08 04:19:49 -0800 | [diff] [blame] | 40 | |
| 41 | struct request_queue *brd_queue; |
| 42 | struct gendisk *brd_disk; |
| 43 | struct list_head brd_list; |
| 44 | |
| 45 | /* |
| 46 | * Backing store of pages and lock to protect it. This is the contents |
| 47 | * of the block device. |
| 48 | */ |
| 49 | spinlock_t brd_lock; |
| 50 | struct radix_tree_root brd_pages; |
| 51 | }; |
| 52 | |
| 53 | /* |
| 54 | * Look up and return a brd's page for a given sector. |
| 55 | */ |
| 56 | static struct page *brd_lookup_page(struct brd_device *brd, sector_t sector) |
| 57 | { |
| 58 | pgoff_t idx; |
| 59 | struct page *page; |
| 60 | |
| 61 | /* |
| 62 | * The page lifetime is protected by the fact that we have opened the |
| 63 | * device node -- brd pages will never be deleted under us, so we |
| 64 | * don't need any further locking or refcounting. |
| 65 | * |
| 66 | * This is strictly true for the radix-tree nodes as well (ie. we |
| 67 | * don't actually need the rcu_read_lock()), however that is not a |
| 68 | * documented feature of the radix-tree API so it is better to be |
| 69 | * safe here (we don't have total exclusion from radix tree updates |
| 70 | * here, only deletes). |
| 71 | */ |
| 72 | rcu_read_lock(); |
| 73 | idx = sector >> PAGE_SECTORS_SHIFT; /* sector to page index */ |
| 74 | page = radix_tree_lookup(&brd->brd_pages, idx); |
| 75 | rcu_read_unlock(); |
| 76 | |
| 77 | BUG_ON(page && page->index != idx); |
| 78 | |
| 79 | return page; |
| 80 | } |
| 81 | |
| 82 | /* |
| 83 | * Look up and return a brd's page for a given sector. |
| 84 | * If one does not exist, allocate an empty page, and insert that. Then |
| 85 | * return it. |
| 86 | */ |
| 87 | static struct page *brd_insert_page(struct brd_device *brd, sector_t sector) |
| 88 | { |
| 89 | pgoff_t idx; |
| 90 | struct page *page; |
Nick Piggin | 75acb9c | 2008-02-08 04:19:50 -0800 | [diff] [blame] | 91 | gfp_t gfp_flags; |
Nick Piggin | 9db5579 | 2008-02-08 04:19:49 -0800 | [diff] [blame] | 92 | |
| 93 | page = brd_lookup_page(brd, sector); |
| 94 | if (page) |
| 95 | return page; |
| 96 | |
| 97 | /* |
| 98 | * Must use NOIO because we don't want to recurse back into the |
| 99 | * block or filesystem layers from page reclaim. |
| 100 | */ |
Hou Tao | f6b5016 | 2019-04-22 21:23:21 +0800 | [diff] [blame] | 101 | gfp_flags = GFP_NOIO | __GFP_ZERO | __GFP_HIGHMEM; |
Petr Tesarik | 26defe3 | 2008-04-22 05:36:52 +0200 | [diff] [blame] | 102 | page = alloc_page(gfp_flags); |
Nick Piggin | 9db5579 | 2008-02-08 04:19:49 -0800 | [diff] [blame] | 103 | if (!page) |
| 104 | return NULL; |
| 105 | |
| 106 | if (radix_tree_preload(GFP_NOIO)) { |
| 107 | __free_page(page); |
| 108 | return NULL; |
| 109 | } |
| 110 | |
| 111 | spin_lock(&brd->brd_lock); |
| 112 | idx = sector >> PAGE_SECTORS_SHIFT; |
Brian Behlendorf | dfd20b2 | 2013-05-24 15:55:28 -0700 | [diff] [blame] | 113 | page->index = idx; |
Nick Piggin | 9db5579 | 2008-02-08 04:19:49 -0800 | [diff] [blame] | 114 | if (radix_tree_insert(&brd->brd_pages, idx, page)) { |
| 115 | __free_page(page); |
| 116 | page = radix_tree_lookup(&brd->brd_pages, idx); |
| 117 | BUG_ON(!page); |
| 118 | BUG_ON(page->index != idx); |
Brian Behlendorf | dfd20b2 | 2013-05-24 15:55:28 -0700 | [diff] [blame] | 119 | } |
Nick Piggin | 9db5579 | 2008-02-08 04:19:49 -0800 | [diff] [blame] | 120 | spin_unlock(&brd->brd_lock); |
| 121 | |
| 122 | radix_tree_preload_end(); |
| 123 | |
| 124 | return page; |
| 125 | } |
| 126 | |
| 127 | /* |
| 128 | * Free all backing store pages and radix tree. This must only be called when |
| 129 | * there are no other users of the device. |
| 130 | */ |
| 131 | #define FREE_BATCH 16 |
| 132 | static void brd_free_pages(struct brd_device *brd) |
| 133 | { |
| 134 | unsigned long pos = 0; |
| 135 | struct page *pages[FREE_BATCH]; |
| 136 | int nr_pages; |
| 137 | |
| 138 | do { |
| 139 | int i; |
| 140 | |
| 141 | nr_pages = radix_tree_gang_lookup(&brd->brd_pages, |
| 142 | (void **)pages, pos, FREE_BATCH); |
| 143 | |
| 144 | for (i = 0; i < nr_pages; i++) { |
| 145 | void *ret; |
| 146 | |
| 147 | BUG_ON(pages[i]->index < pos); |
| 148 | pos = pages[i]->index; |
| 149 | ret = radix_tree_delete(&brd->brd_pages, pos); |
| 150 | BUG_ON(!ret || ret != pages[i]); |
| 151 | __free_page(pages[i]); |
| 152 | } |
| 153 | |
| 154 | pos++; |
| 155 | |
| 156 | /* |
Mikulas Patocka | 936b33f | 2019-05-09 12:51:27 -0600 | [diff] [blame] | 157 | * It takes 3.4 seconds to remove 80GiB ramdisk. |
| 158 | * So, we need cond_resched to avoid stalling the CPU. |
| 159 | */ |
| 160 | cond_resched(); |
| 161 | |
| 162 | /* |
Nick Piggin | 9db5579 | 2008-02-08 04:19:49 -0800 | [diff] [blame] | 163 | * This assumes radix_tree_gang_lookup always returns as |
| 164 | * many pages as possible. If the radix-tree code changes, |
| 165 | * so will this have to. |
| 166 | */ |
| 167 | } while (nr_pages == FREE_BATCH); |
| 168 | } |
| 169 | |
| 170 | /* |
| 171 | * copy_to_brd_setup must be called before copy_to_brd. It may sleep. |
| 172 | */ |
| 173 | static int copy_to_brd_setup(struct brd_device *brd, sector_t sector, size_t n) |
| 174 | { |
| 175 | unsigned int offset = (sector & (PAGE_SECTORS-1)) << SECTOR_SHIFT; |
| 176 | size_t copy; |
| 177 | |
| 178 | copy = min_t(size_t, n, PAGE_SIZE - offset); |
| 179 | if (!brd_insert_page(brd, sector)) |
Matthew Wilcox | 96f8d8e | 2014-06-04 16:07:50 -0700 | [diff] [blame] | 180 | return -ENOSPC; |
Nick Piggin | 9db5579 | 2008-02-08 04:19:49 -0800 | [diff] [blame] | 181 | if (copy < n) { |
| 182 | sector += copy >> SECTOR_SHIFT; |
| 183 | if (!brd_insert_page(brd, sector)) |
Matthew Wilcox | 96f8d8e | 2014-06-04 16:07:50 -0700 | [diff] [blame] | 184 | return -ENOSPC; |
Nick Piggin | 9db5579 | 2008-02-08 04:19:49 -0800 | [diff] [blame] | 185 | } |
| 186 | return 0; |
| 187 | } |
| 188 | |
| 189 | /* |
| 190 | * Copy n bytes from src to the brd starting at sector. Does not sleep. |
| 191 | */ |
| 192 | static void copy_to_brd(struct brd_device *brd, const void *src, |
| 193 | sector_t sector, size_t n) |
| 194 | { |
| 195 | struct page *page; |
| 196 | void *dst; |
| 197 | unsigned int offset = (sector & (PAGE_SECTORS-1)) << SECTOR_SHIFT; |
| 198 | size_t copy; |
| 199 | |
| 200 | copy = min_t(size_t, n, PAGE_SIZE - offset); |
| 201 | page = brd_lookup_page(brd, sector); |
| 202 | BUG_ON(!page); |
| 203 | |
Cong Wang | cfd8005 | 2011-11-25 23:14:18 +0800 | [diff] [blame] | 204 | dst = kmap_atomic(page); |
Nick Piggin | 9db5579 | 2008-02-08 04:19:49 -0800 | [diff] [blame] | 205 | memcpy(dst + offset, src, copy); |
Cong Wang | cfd8005 | 2011-11-25 23:14:18 +0800 | [diff] [blame] | 206 | kunmap_atomic(dst); |
Nick Piggin | 9db5579 | 2008-02-08 04:19:49 -0800 | [diff] [blame] | 207 | |
| 208 | if (copy < n) { |
| 209 | src += copy; |
| 210 | sector += copy >> SECTOR_SHIFT; |
| 211 | copy = n - copy; |
| 212 | page = brd_lookup_page(brd, sector); |
| 213 | BUG_ON(!page); |
| 214 | |
Cong Wang | cfd8005 | 2011-11-25 23:14:18 +0800 | [diff] [blame] | 215 | dst = kmap_atomic(page); |
Nick Piggin | 9db5579 | 2008-02-08 04:19:49 -0800 | [diff] [blame] | 216 | memcpy(dst, src, copy); |
Cong Wang | cfd8005 | 2011-11-25 23:14:18 +0800 | [diff] [blame] | 217 | kunmap_atomic(dst); |
Nick Piggin | 9db5579 | 2008-02-08 04:19:49 -0800 | [diff] [blame] | 218 | } |
| 219 | } |
| 220 | |
| 221 | /* |
| 222 | * Copy n bytes to dst from the brd starting at sector. Does not sleep. |
| 223 | */ |
| 224 | static void copy_from_brd(void *dst, struct brd_device *brd, |
| 225 | sector_t sector, size_t n) |
| 226 | { |
| 227 | struct page *page; |
| 228 | void *src; |
| 229 | unsigned int offset = (sector & (PAGE_SECTORS-1)) << SECTOR_SHIFT; |
| 230 | size_t copy; |
| 231 | |
| 232 | copy = min_t(size_t, n, PAGE_SIZE - offset); |
| 233 | page = brd_lookup_page(brd, sector); |
| 234 | if (page) { |
Cong Wang | cfd8005 | 2011-11-25 23:14:18 +0800 | [diff] [blame] | 235 | src = kmap_atomic(page); |
Nick Piggin | 9db5579 | 2008-02-08 04:19:49 -0800 | [diff] [blame] | 236 | memcpy(dst, src + offset, copy); |
Cong Wang | cfd8005 | 2011-11-25 23:14:18 +0800 | [diff] [blame] | 237 | kunmap_atomic(src); |
Nick Piggin | 9db5579 | 2008-02-08 04:19:49 -0800 | [diff] [blame] | 238 | } else |
| 239 | memset(dst, 0, copy); |
| 240 | |
| 241 | if (copy < n) { |
| 242 | dst += copy; |
| 243 | sector += copy >> SECTOR_SHIFT; |
| 244 | copy = n - copy; |
| 245 | page = brd_lookup_page(brd, sector); |
| 246 | if (page) { |
Cong Wang | cfd8005 | 2011-11-25 23:14:18 +0800 | [diff] [blame] | 247 | src = kmap_atomic(page); |
Nick Piggin | 9db5579 | 2008-02-08 04:19:49 -0800 | [diff] [blame] | 248 | memcpy(dst, src, copy); |
Cong Wang | cfd8005 | 2011-11-25 23:14:18 +0800 | [diff] [blame] | 249 | kunmap_atomic(src); |
Nick Piggin | 9db5579 | 2008-02-08 04:19:49 -0800 | [diff] [blame] | 250 | } else |
| 251 | memset(dst, 0, copy); |
| 252 | } |
| 253 | } |
| 254 | |
| 255 | /* |
| 256 | * Process a single bvec of a bio. |
| 257 | */ |
| 258 | static int brd_do_bvec(struct brd_device *brd, struct page *page, |
Tejun Heo | 3f289dc | 2018-07-18 04:47:36 -0700 | [diff] [blame] | 259 | unsigned int len, unsigned int off, unsigned int op, |
Nick Piggin | 9db5579 | 2008-02-08 04:19:49 -0800 | [diff] [blame] | 260 | sector_t sector) |
| 261 | { |
| 262 | void *mem; |
| 263 | int err = 0; |
| 264 | |
Tejun Heo | 3f289dc | 2018-07-18 04:47:36 -0700 | [diff] [blame] | 265 | if (op_is_write(op)) { |
Nick Piggin | 9db5579 | 2008-02-08 04:19:49 -0800 | [diff] [blame] | 266 | err = copy_to_brd_setup(brd, sector, len); |
| 267 | if (err) |
| 268 | goto out; |
| 269 | } |
| 270 | |
Cong Wang | cfd8005 | 2011-11-25 23:14:18 +0800 | [diff] [blame] | 271 | mem = kmap_atomic(page); |
Tejun Heo | 3f289dc | 2018-07-18 04:47:36 -0700 | [diff] [blame] | 272 | if (!op_is_write(op)) { |
Nick Piggin | 9db5579 | 2008-02-08 04:19:49 -0800 | [diff] [blame] | 273 | copy_from_brd(mem + off, brd, sector, len); |
| 274 | flush_dcache_page(page); |
Nick Piggin | c2572f2 | 2009-04-15 10:32:07 +0200 | [diff] [blame] | 275 | } else { |
| 276 | flush_dcache_page(page); |
Nick Piggin | 9db5579 | 2008-02-08 04:19:49 -0800 | [diff] [blame] | 277 | copy_to_brd(brd, mem + off, sector, len); |
Nick Piggin | c2572f2 | 2009-04-15 10:32:07 +0200 | [diff] [blame] | 278 | } |
Cong Wang | cfd8005 | 2011-11-25 23:14:18 +0800 | [diff] [blame] | 279 | kunmap_atomic(mem); |
Nick Piggin | 9db5579 | 2008-02-08 04:19:49 -0800 | [diff] [blame] | 280 | |
| 281 | out: |
| 282 | return err; |
| 283 | } |
| 284 | |
Jens Axboe | dece163 | 2015-11-05 10:41:16 -0700 | [diff] [blame] | 285 | static blk_qc_t brd_make_request(struct request_queue *q, struct bio *bio) |
Nick Piggin | 9db5579 | 2008-02-08 04:19:49 -0800 | [diff] [blame] | 286 | { |
Christoph Hellwig | 74d4699 | 2017-08-23 19:10:32 +0200 | [diff] [blame] | 287 | struct brd_device *brd = bio->bi_disk->private_data; |
Kent Overstreet | 7988613 | 2013-11-23 17:19:00 -0800 | [diff] [blame] | 288 | struct bio_vec bvec; |
Nick Piggin | 9db5579 | 2008-02-08 04:19:49 -0800 | [diff] [blame] | 289 | sector_t sector; |
Kent Overstreet | 7988613 | 2013-11-23 17:19:00 -0800 | [diff] [blame] | 290 | struct bvec_iter iter; |
Nick Piggin | 9db5579 | 2008-02-08 04:19:49 -0800 | [diff] [blame] | 291 | |
Kent Overstreet | 4f024f3 | 2013-10-11 15:44:27 -0700 | [diff] [blame] | 292 | sector = bio->bi_iter.bi_sector; |
Christoph Hellwig | 74d4699 | 2017-08-23 19:10:32 +0200 | [diff] [blame] | 293 | if (bio_end_sector(bio) > get_capacity(bio->bi_disk)) |
Christoph Hellwig | 4246a0b | 2015-07-20 15:29:37 +0200 | [diff] [blame] | 294 | goto io_error; |
Nick Piggin | 9db5579 | 2008-02-08 04:19:49 -0800 | [diff] [blame] | 295 | |
Kent Overstreet | 7988613 | 2013-11-23 17:19:00 -0800 | [diff] [blame] | 296 | bio_for_each_segment(bvec, bio, iter) { |
| 297 | unsigned int len = bvec.bv_len; |
Christoph Hellwig | 4246a0b | 2015-07-20 15:29:37 +0200 | [diff] [blame] | 298 | int err; |
| 299 | |
Ming Lei | f1acbf2 | 2019-12-04 19:31:15 +0800 | [diff] [blame] | 300 | /* Don't support un-aligned buffer */ |
| 301 | WARN_ON_ONCE((bvec.bv_offset & (SECTOR_SIZE - 1)) || |
| 302 | (len & (SECTOR_SIZE - 1))); |
| 303 | |
Jens Axboe | c11f0c0 | 2016-08-05 08:11:04 -0600 | [diff] [blame] | 304 | err = brd_do_bvec(brd, bvec.bv_page, len, bvec.bv_offset, |
Tejun Heo | 3f289dc | 2018-07-18 04:47:36 -0700 | [diff] [blame] | 305 | bio_op(bio), sector); |
Nick Piggin | 9db5579 | 2008-02-08 04:19:49 -0800 | [diff] [blame] | 306 | if (err) |
Christoph Hellwig | 4246a0b | 2015-07-20 15:29:37 +0200 | [diff] [blame] | 307 | goto io_error; |
Nick Piggin | 9db5579 | 2008-02-08 04:19:49 -0800 | [diff] [blame] | 308 | sector += len >> SECTOR_SHIFT; |
| 309 | } |
| 310 | |
Christoph Hellwig | 4246a0b | 2015-07-20 15:29:37 +0200 | [diff] [blame] | 311 | bio_endio(bio); |
Jens Axboe | dece163 | 2015-11-05 10:41:16 -0700 | [diff] [blame] | 312 | return BLK_QC_T_NONE; |
Christoph Hellwig | 4246a0b | 2015-07-20 15:29:37 +0200 | [diff] [blame] | 313 | io_error: |
| 314 | bio_io_error(bio); |
Jens Axboe | dece163 | 2015-11-05 10:41:16 -0700 | [diff] [blame] | 315 | return BLK_QC_T_NONE; |
Nick Piggin | 9db5579 | 2008-02-08 04:19:49 -0800 | [diff] [blame] | 316 | } |
| 317 | |
Matthew Wilcox | a72132c | 2014-06-04 16:07:49 -0700 | [diff] [blame] | 318 | static int brd_rw_page(struct block_device *bdev, sector_t sector, |
Tejun Heo | 3f289dc | 2018-07-18 04:47:36 -0700 | [diff] [blame] | 319 | struct page *page, unsigned int op) |
Matthew Wilcox | a72132c | 2014-06-04 16:07:49 -0700 | [diff] [blame] | 320 | { |
| 321 | struct brd_device *brd = bdev->bd_disk->private_data; |
Huang Ying | 98cc093 | 2017-09-06 16:22:27 -0700 | [diff] [blame] | 322 | int err; |
| 323 | |
| 324 | if (PageTransHuge(page)) |
| 325 | return -ENOTSUPP; |
Tejun Heo | 3f289dc | 2018-07-18 04:47:36 -0700 | [diff] [blame] | 326 | err = brd_do_bvec(brd, page, PAGE_SIZE, 0, op, sector); |
| 327 | page_endio(page, op_is_write(op), err); |
Matthew Wilcox | a72132c | 2014-06-04 16:07:49 -0700 | [diff] [blame] | 328 | return err; |
| 329 | } |
| 330 | |
Alexey Dobriyan | 83d5cde | 2009-09-21 17:01:13 -0700 | [diff] [blame] | 331 | static const struct block_device_operations brd_fops = { |
Nick Piggin | 75acb9c | 2008-02-08 04:19:50 -0800 | [diff] [blame] | 332 | .owner = THIS_MODULE, |
Matthew Wilcox | a72132c | 2014-06-04 16:07:49 -0700 | [diff] [blame] | 333 | .rw_page = brd_rw_page, |
Nick Piggin | 9db5579 | 2008-02-08 04:19:49 -0800 | [diff] [blame] | 334 | }; |
| 335 | |
| 336 | /* |
| 337 | * And now the modules code and kernel interface. |
| 338 | */ |
Boaz Harrosh | 937af5e | 2015-01-07 18:07:56 +0200 | [diff] [blame] | 339 | static int rd_nr = CONFIG_BLK_DEV_RAM_COUNT; |
Joe Perches | 5657a81 | 2018-05-24 13:38:59 -0600 | [diff] [blame] | 340 | module_param(rd_nr, int, 0444); |
Nick Piggin | 9db5579 | 2008-02-08 04:19:49 -0800 | [diff] [blame] | 341 | MODULE_PARM_DESC(rd_nr, "Maximum number of brd devices"); |
Boaz Harrosh | 937af5e | 2015-01-07 18:07:56 +0200 | [diff] [blame] | 342 | |
Jan Kara | 366f4ae | 2016-10-25 13:53:27 +0200 | [diff] [blame] | 343 | unsigned long rd_size = CONFIG_BLK_DEV_RAM_SIZE; |
Joe Perches | 5657a81 | 2018-05-24 13:38:59 -0600 | [diff] [blame] | 344 | module_param(rd_size, ulong, 0444); |
Nick Piggin | 9db5579 | 2008-02-08 04:19:49 -0800 | [diff] [blame] | 345 | MODULE_PARM_DESC(rd_size, "Size of each RAM disk in kbytes."); |
Boaz Harrosh | 937af5e | 2015-01-07 18:07:56 +0200 | [diff] [blame] | 346 | |
| 347 | static int max_part = 1; |
Joe Perches | 5657a81 | 2018-05-24 13:38:59 -0600 | [diff] [blame] | 348 | module_param(max_part, int, 0444); |
Boaz Harrosh | 937af5e | 2015-01-07 18:07:56 +0200 | [diff] [blame] | 349 | MODULE_PARM_DESC(max_part, "Num Minors to reserve between devices"); |
| 350 | |
Nick Piggin | 9db5579 | 2008-02-08 04:19:49 -0800 | [diff] [blame] | 351 | MODULE_LICENSE("GPL"); |
| 352 | MODULE_ALIAS_BLOCKDEV_MAJOR(RAMDISK_MAJOR); |
Nick Piggin | efedf51 | 2008-06-04 17:18:42 +0200 | [diff] [blame] | 353 | MODULE_ALIAS("rd"); |
Nick Piggin | 9db5579 | 2008-02-08 04:19:49 -0800 | [diff] [blame] | 354 | |
| 355 | #ifndef MODULE |
| 356 | /* Legacy boot options - nonmodular */ |
| 357 | static int __init ramdisk_size(char *str) |
| 358 | { |
| 359 | rd_size = simple_strtol(str, NULL, 0); |
| 360 | return 1; |
| 361 | } |
Robert P. J. Day | 1adbee5 | 2009-06-10 12:57:08 -0700 | [diff] [blame] | 362 | __setup("ramdisk_size=", ramdisk_size); |
Nick Piggin | 9db5579 | 2008-02-08 04:19:49 -0800 | [diff] [blame] | 363 | #endif |
| 364 | |
| 365 | /* |
| 366 | * The device scheme is derived from loop.c. Keep them in synch where possible |
| 367 | * (should share code eventually). |
| 368 | */ |
| 369 | static LIST_HEAD(brd_devices); |
| 370 | static DEFINE_MUTEX(brd_devices_mutex); |
| 371 | |
| 372 | static struct brd_device *brd_alloc(int i) |
| 373 | { |
| 374 | struct brd_device *brd; |
| 375 | struct gendisk *disk; |
| 376 | |
| 377 | brd = kzalloc(sizeof(*brd), GFP_KERNEL); |
| 378 | if (!brd) |
| 379 | goto out; |
| 380 | brd->brd_number = i; |
| 381 | spin_lock_init(&brd->brd_lock); |
| 382 | INIT_RADIX_TREE(&brd->brd_pages, GFP_ATOMIC); |
| 383 | |
Christoph Hellwig | 3d745ea | 2020-03-27 09:30:11 +0100 | [diff] [blame] | 384 | brd->brd_queue = blk_alloc_queue(brd_make_request, NUMA_NO_NODE); |
Nick Piggin | 9db5579 | 2008-02-08 04:19:49 -0800 | [diff] [blame] | 385 | if (!brd->brd_queue) |
| 386 | goto out_free_dev; |
Boaz Harrosh | c8fa317 | 2015-01-07 18:09:38 +0200 | [diff] [blame] | 387 | |
Boaz Harrosh | c8fa317 | 2015-01-07 18:09:38 +0200 | [diff] [blame] | 388 | /* This is so fdisk will align partitions on 4k, because of |
| 389 | * direct_access API needing 4k alignment, returning a PFN |
| 390 | * (This is only a problem on very small devices <= 4M, |
| 391 | * otherwise fdisk will align on 1M. Regardless this call |
| 392 | * is harmless) |
| 393 | */ |
| 394 | blk_queue_physical_block_size(brd->brd_queue, PAGE_SIZE); |
Boaz Harrosh | 937af5e | 2015-01-07 18:07:56 +0200 | [diff] [blame] | 395 | disk = brd->brd_disk = alloc_disk(max_part); |
Nick Piggin | 9db5579 | 2008-02-08 04:19:49 -0800 | [diff] [blame] | 396 | if (!disk) |
| 397 | goto out_free_queue; |
| 398 | disk->major = RAMDISK_MAJOR; |
Boaz Harrosh | 937af5e | 2015-01-07 18:07:56 +0200 | [diff] [blame] | 399 | disk->first_minor = i * max_part; |
Nick Piggin | 9db5579 | 2008-02-08 04:19:49 -0800 | [diff] [blame] | 400 | disk->fops = &brd_fops; |
| 401 | disk->private_data = brd; |
Boaz Harrosh | 937af5e | 2015-01-07 18:07:56 +0200 | [diff] [blame] | 402 | disk->flags = GENHD_FL_EXT_DEVT; |
Nick Piggin | 9db5579 | 2008-02-08 04:19:49 -0800 | [diff] [blame] | 403 | sprintf(disk->disk_name, "ram%d", i); |
| 404 | set_capacity(disk, rd_size * 2); |
Ming Lei | 153fcd5 | 2018-11-02 08:50:51 +0800 | [diff] [blame] | 405 | brd->brd_queue->backing_dev_info->capabilities |= BDI_CAP_SYNCHRONOUS_IO; |
Nick Piggin | 9db5579 | 2008-02-08 04:19:49 -0800 | [diff] [blame] | 406 | |
SeongJae Park | 316ba57 | 2018-05-03 18:53:26 +0900 | [diff] [blame] | 407 | /* Tell the block layer that this is not a rotational device */ |
Ming Lei | 153fcd5 | 2018-11-02 08:50:51 +0800 | [diff] [blame] | 408 | blk_queue_flag_set(QUEUE_FLAG_NONROT, brd->brd_queue); |
| 409 | blk_queue_flag_clear(QUEUE_FLAG_ADD_RANDOM, brd->brd_queue); |
SeongJae Park | 316ba57 | 2018-05-03 18:53:26 +0900 | [diff] [blame] | 410 | |
Nick Piggin | 9db5579 | 2008-02-08 04:19:49 -0800 | [diff] [blame] | 411 | return brd; |
| 412 | |
| 413 | out_free_queue: |
| 414 | blk_cleanup_queue(brd->brd_queue); |
| 415 | out_free_dev: |
| 416 | kfree(brd); |
| 417 | out: |
| 418 | return NULL; |
| 419 | } |
| 420 | |
| 421 | static void brd_free(struct brd_device *brd) |
| 422 | { |
| 423 | put_disk(brd->brd_disk); |
| 424 | blk_cleanup_queue(brd->brd_queue); |
| 425 | brd_free_pages(brd); |
| 426 | kfree(brd); |
| 427 | } |
| 428 | |
Boaz Harrosh | 937af5e | 2015-01-07 18:07:56 +0200 | [diff] [blame] | 429 | static struct brd_device *brd_init_one(int i, bool *new) |
Nick Piggin | 9db5579 | 2008-02-08 04:19:49 -0800 | [diff] [blame] | 430 | { |
| 431 | struct brd_device *brd; |
| 432 | |
Boaz Harrosh | 937af5e | 2015-01-07 18:07:56 +0200 | [diff] [blame] | 433 | *new = false; |
Nick Piggin | 9db5579 | 2008-02-08 04:19:49 -0800 | [diff] [blame] | 434 | list_for_each_entry(brd, &brd_devices, brd_list) { |
| 435 | if (brd->brd_number == i) |
| 436 | goto out; |
| 437 | } |
| 438 | |
| 439 | brd = brd_alloc(i); |
| 440 | if (brd) { |
Ming Lei | 153fcd5 | 2018-11-02 08:50:51 +0800 | [diff] [blame] | 441 | brd->brd_disk->queue = brd->brd_queue; |
Nick Piggin | 9db5579 | 2008-02-08 04:19:49 -0800 | [diff] [blame] | 442 | add_disk(brd->brd_disk); |
| 443 | list_add_tail(&brd->brd_list, &brd_devices); |
| 444 | } |
Boaz Harrosh | 937af5e | 2015-01-07 18:07:56 +0200 | [diff] [blame] | 445 | *new = true; |
Nick Piggin | 9db5579 | 2008-02-08 04:19:49 -0800 | [diff] [blame] | 446 | out: |
| 447 | return brd; |
| 448 | } |
| 449 | |
| 450 | static void brd_del_one(struct brd_device *brd) |
| 451 | { |
| 452 | list_del(&brd->brd_list); |
| 453 | del_gendisk(brd->brd_disk); |
| 454 | brd_free(brd); |
| 455 | } |
| 456 | |
| 457 | static struct kobject *brd_probe(dev_t dev, int *part, void *data) |
| 458 | { |
| 459 | struct brd_device *brd; |
| 460 | struct kobject *kobj; |
Boaz Harrosh | 937af5e | 2015-01-07 18:07:56 +0200 | [diff] [blame] | 461 | bool new; |
Nick Piggin | 9db5579 | 2008-02-08 04:19:49 -0800 | [diff] [blame] | 462 | |
| 463 | mutex_lock(&brd_devices_mutex); |
Boaz Harrosh | 937af5e | 2015-01-07 18:07:56 +0200 | [diff] [blame] | 464 | brd = brd_init_one(MINOR(dev) / max_part, &new); |
Jan Kara | 3079c22 | 2018-02-26 13:01:38 +0100 | [diff] [blame] | 465 | kobj = brd ? get_disk_and_module(brd->brd_disk) : NULL; |
Nick Piggin | 9db5579 | 2008-02-08 04:19:49 -0800 | [diff] [blame] | 466 | mutex_unlock(&brd_devices_mutex); |
| 467 | |
Boaz Harrosh | 937af5e | 2015-01-07 18:07:56 +0200 | [diff] [blame] | 468 | if (new) |
| 469 | *part = 0; |
| 470 | |
Nick Piggin | 9db5579 | 2008-02-08 04:19:49 -0800 | [diff] [blame] | 471 | return kobj; |
| 472 | } |
| 473 | |
Zhiqiang Liu | c8ab422 | 2020-02-04 19:30:20 +0800 | [diff] [blame] | 474 | static inline void brd_check_and_reset_par(void) |
| 475 | { |
| 476 | if (unlikely(!max_part)) |
| 477 | max_part = 1; |
| 478 | |
| 479 | /* |
| 480 | * make sure 'max_part' can be divided exactly by (1U << MINORBITS), |
| 481 | * otherwise, it is possiable to get same dev_t when adding partitions. |
| 482 | */ |
| 483 | if ((1U << MINORBITS) % max_part != 0) |
| 484 | max_part = 1UL << fls(max_part); |
| 485 | |
| 486 | if (max_part > DISK_MAX_PARTS) { |
| 487 | pr_info("brd: max_part can't be larger than %d, reset max_part = %d.\n", |
| 488 | DISK_MAX_PARTS, DISK_MAX_PARTS); |
| 489 | max_part = DISK_MAX_PARTS; |
| 490 | } |
| 491 | } |
| 492 | |
Nick Piggin | 9db5579 | 2008-02-08 04:19:49 -0800 | [diff] [blame] | 493 | static int __init brd_init(void) |
| 494 | { |
Nick Piggin | 9db5579 | 2008-02-08 04:19:49 -0800 | [diff] [blame] | 495 | struct brd_device *brd, *next; |
Boaz Harrosh | 937af5e | 2015-01-07 18:07:56 +0200 | [diff] [blame] | 496 | int i; |
Nick Piggin | 9db5579 | 2008-02-08 04:19:49 -0800 | [diff] [blame] | 497 | |
| 498 | /* |
| 499 | * brd module now has a feature to instantiate underlying device |
| 500 | * structure on-demand, provided that there is an access dev node. |
Nick Piggin | 9db5579 | 2008-02-08 04:19:49 -0800 | [diff] [blame] | 501 | * |
Boaz Harrosh | 937af5e | 2015-01-07 18:07:56 +0200 | [diff] [blame] | 502 | * (1) if rd_nr is specified, create that many upfront. else |
| 503 | * it defaults to CONFIG_BLK_DEV_RAM_COUNT |
| 504 | * (2) User can further extend brd devices by create dev node themselves |
| 505 | * and have kernel automatically instantiate actual device |
| 506 | * on-demand. Example: |
| 507 | * mknod /path/devnod_name b 1 X # 1 is the rd major |
| 508 | * fdisk -l /path/devnod_name |
| 509 | * If (X / max_part) was not already created it will be created |
| 510 | * dynamically. |
Nick Piggin | 9db5579 | 2008-02-08 04:19:49 -0800 | [diff] [blame] | 511 | */ |
Laurent Vivier | d7853d1 | 2008-04-30 00:55:06 -0700 | [diff] [blame] | 512 | |
Nick Piggin | 9db5579 | 2008-02-08 04:19:49 -0800 | [diff] [blame] | 513 | if (register_blkdev(RAMDISK_MAJOR, "ramdisk")) |
| 514 | return -EIO; |
| 515 | |
Zhiqiang Liu | c8ab422 | 2020-02-04 19:30:20 +0800 | [diff] [blame] | 516 | brd_check_and_reset_par(); |
Boaz Harrosh | 937af5e | 2015-01-07 18:07:56 +0200 | [diff] [blame] | 517 | |
| 518 | for (i = 0; i < rd_nr; i++) { |
Nick Piggin | 9db5579 | 2008-02-08 04:19:49 -0800 | [diff] [blame] | 519 | brd = brd_alloc(i); |
| 520 | if (!brd) |
| 521 | goto out_free; |
| 522 | list_add_tail(&brd->brd_list, &brd_devices); |
| 523 | } |
| 524 | |
| 525 | /* point of no return */ |
| 526 | |
Ming Lei | 153fcd5 | 2018-11-02 08:50:51 +0800 | [diff] [blame] | 527 | list_for_each_entry(brd, &brd_devices, brd_list) { |
| 528 | /* |
| 529 | * associate with queue just before adding disk for |
| 530 | * avoiding to mess up failure path |
| 531 | */ |
| 532 | brd->brd_disk->queue = brd->brd_queue; |
Nick Piggin | 9db5579 | 2008-02-08 04:19:49 -0800 | [diff] [blame] | 533 | add_disk(brd->brd_disk); |
Ming Lei | 153fcd5 | 2018-11-02 08:50:51 +0800 | [diff] [blame] | 534 | } |
Nick Piggin | 9db5579 | 2008-02-08 04:19:49 -0800 | [diff] [blame] | 535 | |
Boaz Harrosh | 937af5e | 2015-01-07 18:07:56 +0200 | [diff] [blame] | 536 | blk_register_region(MKDEV(RAMDISK_MAJOR, 0), 1UL << MINORBITS, |
Nick Piggin | 9db5579 | 2008-02-08 04:19:49 -0800 | [diff] [blame] | 537 | THIS_MODULE, brd_probe, NULL, NULL); |
| 538 | |
Boaz Harrosh | 937af5e | 2015-01-07 18:07:56 +0200 | [diff] [blame] | 539 | pr_info("brd: module loaded\n"); |
Nick Piggin | 9db5579 | 2008-02-08 04:19:49 -0800 | [diff] [blame] | 540 | return 0; |
| 541 | |
| 542 | out_free: |
| 543 | list_for_each_entry_safe(brd, next, &brd_devices, brd_list) { |
| 544 | list_del(&brd->brd_list); |
| 545 | brd_free(brd); |
| 546 | } |
Akinobu Mita | c82f296 | 2008-08-20 14:09:09 -0700 | [diff] [blame] | 547 | unregister_blkdev(RAMDISK_MAJOR, "ramdisk"); |
Nick Piggin | 9db5579 | 2008-02-08 04:19:49 -0800 | [diff] [blame] | 548 | |
Boaz Harrosh | 937af5e | 2015-01-07 18:07:56 +0200 | [diff] [blame] | 549 | pr_info("brd: module NOT loaded !!!\n"); |
Nick Piggin | 9db5579 | 2008-02-08 04:19:49 -0800 | [diff] [blame] | 550 | return -ENOMEM; |
| 551 | } |
| 552 | |
| 553 | static void __exit brd_exit(void) |
| 554 | { |
Nick Piggin | 9db5579 | 2008-02-08 04:19:49 -0800 | [diff] [blame] | 555 | struct brd_device *brd, *next; |
| 556 | |
Nick Piggin | 9db5579 | 2008-02-08 04:19:49 -0800 | [diff] [blame] | 557 | list_for_each_entry_safe(brd, next, &brd_devices, brd_list) |
| 558 | brd_del_one(brd); |
| 559 | |
Boaz Harrosh | 937af5e | 2015-01-07 18:07:56 +0200 | [diff] [blame] | 560 | blk_unregister_region(MKDEV(RAMDISK_MAJOR, 0), 1UL << MINORBITS); |
Nick Piggin | 9db5579 | 2008-02-08 04:19:49 -0800 | [diff] [blame] | 561 | unregister_blkdev(RAMDISK_MAJOR, "ramdisk"); |
Boaz Harrosh | 937af5e | 2015-01-07 18:07:56 +0200 | [diff] [blame] | 562 | |
| 563 | pr_info("brd: module unloaded\n"); |
Nick Piggin | 9db5579 | 2008-02-08 04:19:49 -0800 | [diff] [blame] | 564 | } |
| 565 | |
| 566 | module_init(brd_init); |
| 567 | module_exit(brd_exit); |
| 568 | |