| 1 | |
| 2 | * linux/drivers/block/loop.c
|
| 3 | *
|
| 4 | * Written by Theodore Ts'o, 3/29/93
|
| 5 | *
|
| 6 | * Copyright 1993 by Theodore Ts'o. Redistribution of this file is
|
| 7 | * permitted under the GNU General Public License.
|
| 8 | *
|
| 9 | * DES encryption plus some minor changes by Werner Almesberger, 30-MAY-1993
|
| 10 | * more DES encryption plus IDEA encryption by Nicholas J. Leon, June 20, 1996
|
| 11 | *
|
| 12 | * Modularized and updated for 1.1.16 kernel - Mitch Dsouza 28th May 1994
|
| 13 | * Adapted for 1.3.59 kernel - Andries Brouwer, 1 Feb 1996
|
| 14 | *
|
| 15 | * Fixed do_loop_request() re-entrancy - [email protected] Mar 20, 1997
|
| 16 | *
|
| 17 | * Added devfs support - Richard Gooch <[email protected]> 16-Jan-1998
|
| 18 | *
|
| 19 | * Handle sparse backing files correctly - Kenn Humborg, Jun 28, 1998
|
| 20 | *
|
| 21 | * Loadable modules and other fixes by AK, 1998
|
| 22 | *
|
| 23 | * Make real block number available to downstream transfer functions, enables
|
| 24 | * CBC (and relatives) mode encryption requiring unique IVs per data block.
|
| 25 | * Reed H. Petty, [email protected]
|
| 26 | *
|
| 27 | * Maximum number of loop devices now dynamic via max_loop module parameter.
|
| 28 | * Russell Kroll <[email protected]> 19990701
|
| 29 | *
|
| 30 | * Maximum number of loop devices when compiled-in now selectable by passing
|
| 31 | * max_loop=<1-255> to the kernel on boot.
|
| 32 | * Erik I. Bolsø, <[email protected]>, Oct 31, 1999
|
| 33 | *
|
| 34 | * Completely rewrite request handling to be make_request_fn style and
|
| 35 | * non blocking, pushing work to a helper thread. Lots of fixes from
|
| 36 | * Al Viro too.
|
| 37 | * Jens Axboe <[email protected]>, Nov 2000
|
| 38 | *
|
| 39 | * Support up to 256 loop devices
|
| 40 | * Heinz Mauelshagen <[email protected]>, Feb 2002
|
| 41 | *
|
| 42 | * Support for falling back on the write file operation when the address space
|
| 43 | * operations write_begin is not available on the backing filesystem.
|
| 44 | * Anton Altaparmakov, 16 Feb 2005
|
| 45 | *
|
| 46 | * Still To Fix:
|
| 47 | * - Advisory locking is ignored here.
|
| 48 | * - Should use an own CAP_* category instead of CAP_SYS_ADMIN
|
| 49 | *
|
| 50 |
|
| 51 |
|
| 52 | #include <linux/module.h>
|
| 53 | #include <linux/moduleparam.h>
|
| 54 | #include <linux/sched.h>
|
| 55 | #include <linux/fs.h>
|
| 56 | #include <linux/file.h>
|
| 57 | #include <linux/stat.h>
|
| 58 | #include <linux/errno.h>
|
| 59 | #include <linux/major.h>
|
| 60 | #include <linux/wait.h>
|
| 61 | #include <linux/blkdev.h>
|
| 62 | #include <linux/blkpg.h>
|
| 63 | #include <linux/init.h>
|
| 64 | #include <linux/swap.h>
|
| 65 | #include <linux/slab.h>
|
| 66 | #include <linux/loop.h>
|
| 67 | #include <linux/compat.h>
|
| 68 | #include <linux/suspend.h>
|
| 69 | #include <linux/freezer.h>
|
| 70 | #include <linux/smp_lock.h>
|
| 71 | #include <linux/writeback.h>
|
| 72 | #include <linux/buffer_head.h>
|
| 73 | #include <linux/completion.h>
|
| 74 | #include <linux/highmem.h>
|
| 75 | #include <linux/kthread.h>
|
| 76 | #include <linux/splice.h>
|
| 77 |
|
| 78 | #include <asm/uaccess.h>
|
| 79 |
|
| 80 | static LIST_HEAD(loop_devices);
|
| 81 | static DEFINE_MUTEX(loop_devices_mutex);
|
| 82 |
|
| 83 | static int max_part;
|
| 84 | static int part_shift;
|
| 85 |
|
| 86 | |
| 87 | * Transfer functions
|
| 88 |
|
| 89 | static int transfer_none(struct loop_device *lo, int cmd,
|
| 90 | struct page *raw_page, unsigned raw_off,
|
| 91 | struct page *loop_page, unsigned loop_off,
|
| 92 | int size, sector_t real_block)
|
| 93 | {
|
| 94 | char *raw_buf = kmap_atomic(raw_page, KM_USER0) + raw_off;
|
| 95 | char *loop_buf = kmap_atomic(loop_page, KM_USER1) + loop_off;
|
| 96 |
|
| 97 | if (cmd == READ)
|
| 98 | memcpy(loop_buf, raw_buf, size);
|
| 99 | else
|
| 100 | memcpy(raw_buf, loop_buf, size);
|
| 101 |
|
| 102 | kunmap_atomic(raw_buf, KM_USER0);
|
| 103 | kunmap_atomic(loop_buf, KM_USER1);
|
| 104 | cond_resched();
|
| 105 | return 0;
|
| 106 | }
|
| 107 |
|
| 108 | static int transfer_xor(struct loop_device *lo, int cmd,
|
| 109 | struct page *raw_page, unsigned raw_off,
|
| 110 | struct page *loop_page, unsigned loop_off,
|
| 111 | int size, sector_t real_block)
|
| 112 | {
|
| 113 | char *raw_buf = kmap_atomic(raw_page, KM_USER0) + raw_off;
|
| 114 | char *loop_buf = kmap_atomic(loop_page, KM_USER1) + loop_off;
|
| 115 | char *in, *out, *key;
|
| 116 | int i, keysize;
|
| 117 |
|
| 118 | if (cmd == READ) {
|
| 119 | in = raw_buf;
|
| 120 | out = loop_buf;
|
| 121 | } else {
|
| 122 | in = loop_buf;
|
| 123 | out = raw_buf;
|
| 124 | }
|
| 125 |
|
| 126 | key = lo->lo_encrypt_key;
|
| 127 | keysize = lo->lo_encrypt_key_size;
|
| 128 | for (i = 0; i < size; i++)
|
| 129 | *out++ = *in++ ^ key[(i & 511) % keysize];
|
| 130 |
|
| 131 | kunmap_atomic(raw_buf, KM_USER0);
|
| 132 | kunmap_atomic(loop_buf, KM_USER1);
|
| 133 | cond_resched();
|
| 134 | return 0;
|
| 135 | }
|
| 136 |
|
| 137 | static int xor_init(struct loop_device *lo, const struct loop_info64 *info)
|
| 138 | {
|
| 139 | if (unlikely(info->lo_encrypt_key_size <= 0))
|
| 140 | return -EINVAL;
|
| 141 | return 0;
|
| 142 | }
|
| 143 |
|
| 144 | static struct loop_func_table none_funcs = {
|
| 145 | .number = LO_CRYPT_NONE,
|
| 146 | .transfer = transfer_none,
|
| 147 | };
|
| 148 |
|
| 149 | static struct loop_func_table xor_funcs = {
|
| 150 | .number = LO_CRYPT_XOR,
|
| 151 | .transfer = transfer_xor,
|
| 152 | .init = xor_init
|
| 153 | };
|
| 154 |
|
| 155 |
|
| 156 | static struct loop_func_table *xfer_funcs[MAX_LO_CRYPT] = {
|
| 157 | &none_funcs,
|
| 158 | &xor_funcs
|
| 159 | };
|
| 160 |
|
| 161 | static loff_t get_loop_size(struct loop_device *lo, struct file *file)
|
| 162 | {
|
| 163 | loff_t size, offset, loopsize;
|
| 164 |
|
| 165 |
|
| 166 | size = i_size_read(file->f_mapping->host);
|
| 167 | offset = lo->lo_offset;
|
| 168 | loopsize = size - offset;
|
| 169 | if (lo->lo_sizelimit > 0 && lo->lo_sizelimit < loopsize)
|
| 170 | loopsize = lo->lo_sizelimit;
|
| 171 |
|
| 172 | |
| 173 | * Unfortunately, if we want to do I/O on the device,
|
| 174 | * the number of 512-byte sectors has to fit into a sector_t.
|
| 175 |
|
| 176 | return loopsize >> 9;
|
| 177 | }
|
| 178 |
|
| 179 | static int
|
| 180 | figure_loop_size(struct loop_device *lo)
|
| 181 | {
|
| 182 | loff_t size = get_loop_size(lo, lo->lo_backing_file);
|
| 183 | sector_t x = (sector_t)size;
|
| 184 |
|
| 185 | if (unlikely((loff_t)x != size))
|
| 186 | return -EFBIG;
|
| 187 |
|
| 188 | set_capacity(lo->lo_disk, x);
|
| 189 | return 0;
|
| 190 | }
|
| 191 |
|
| 192 | static inline int
|
| 193 | lo_do_transfer(struct loop_device *lo, int cmd,
|
| 194 | struct page *rpage, unsigned roffs,
|
| 195 | struct page *lpage, unsigned loffs,
|
| 196 | int size, sector_t rblock)
|
| 197 | {
|
| 198 | if (unlikely(!lo->transfer))
|
| 199 | return 0;
|
| 200 |
|
| 201 | return lo->transfer(lo, cmd, rpage, roffs, lpage, loffs, size, rblock);
|
| 202 | }
|
| 203 |
|
| 204 | |
| 205 | * do_lo_send_aops - helper for writing data to a loop device
|
| 206 | *
|
| 207 | * This is the fast version for backing filesystems which implement the address
|
| 208 | * space operations write_begin and write_end.
|
| 209 |
|
| 210 | static int do_lo_send_aops(struct loop_device *lo, struct bio_vec *bvec,
|
| 211 | loff_t pos, struct page *unused)
|
| 212 | {
|
| 213 | struct file *file = lo->lo_backing_file;
|
| 214 | struct address_space *mapping = file->f_mapping;
|
| 215 | pgoff_t index;
|
| 216 | unsigned offset, bv_offs;
|
| 217 | int len, ret;
|
| 218 |
|
| 219 | mutex_lock(&mapping->host->i_mutex);
|
| 220 | index = pos >> PAGE_CACHE_SHIFT;
|
| 221 | offset = pos & ((pgoff_t)PAGE_CACHE_SIZE - 1);
|
| 222 | bv_offs = bvec->bv_offset;
|
| 223 | len = bvec->bv_len;
|
| 224 | while (len > 0) {
|
| 225 | sector_t IV;
|
| 226 | unsigned size, copied;
|
| 227 | int transfer_result;
|
| 228 | struct page *page;
|
| 229 | void *fsdata;
|
| 230 |
|
| 231 | IV = ((sector_t)index << (PAGE_CACHE_SHIFT - 9))+(offset >> 9);
|
| 232 | size = PAGE_CACHE_SIZE - offset;
|
| 233 | if (size > len)
|
| 234 | size = len;
|
| 235 |
|
| 236 | ret = pagecache_write_begin(file, mapping, pos, size, 0,
|
| 237 | &page, &fsdata);
|
| 238 | if (ret)
|
| 239 | goto fail;
|
| 240 |
|
| 241 | file_update_time(file);
|
| 242 |
|
| 243 | transfer_result = lo_do_transfer(lo, WRITE, page, offset,
|
| 244 | bvec->bv_page, bv_offs, size, IV);
|
| 245 | copied = size;
|
| 246 | if (unlikely(transfer_result))
|
| 247 | copied = 0;
|
| 248 |
|
| 249 | ret = pagecache_write_end(file, mapping, pos, size, copied,
|
| 250 | page, fsdata);
|
| 251 | if (ret < 0 || ret != copied)
|
| 252 | goto fail;
|
| 253 |
|
| 254 | if (unlikely(transfer_result))
|
| 255 | goto fail;
|
| 256 |
|
| 257 | bv_offs += copied;
|
| 258 | len -= copied;
|
| 259 | offset = 0;
|
| 260 | index++;
|
| 261 | pos += copied;
|
| 262 | }
|
| 263 | ret = 0;
|
| 264 | out:
|
| 265 | mutex_unlock(&mapping->host->i_mutex);
|
| 266 | return ret;
|
| 267 | fail:
|
| 268 | ret = -1;
|
| 269 | goto out;
|
| 270 | }
|
| 271 |
|
| 272 | |
| 273 | * __do_lo_send_write - helper for writing data to a loop device
|
| 274 | *
|
| 275 | * This helper just factors out common code between do_lo_send_direct_write()
|
| 276 | * and do_lo_send_write().
|
| 277 |
|
| 278 | static int __do_lo_send_write(struct file *file,
|
| 279 | u8 *buf, const int len, loff_t pos)
|
| 280 | {
|
| 281 | ssize_t bw;
|
| 282 | mm_segment_t old_fs = get_fs();
|
| 283 |
|
| 284 | set_fs(get_ds());
|
| 285 | bw = file->f_op->write(file, buf, len, &pos);
|
| 286 | set_fs(old_fs);
|
| 287 | if (likely(bw == len))
|
| 288 | return 0;
|
| 289 | printk(KERN_ERR "loop: Write error at byte offset %llu, length %i.\n",
|
| 290 | (unsigned long long)pos, len);
|
| 291 | if (bw >= 0)
|
| 292 | bw = -EIO;
|
| 293 | return bw;
|
| 294 | }
|
| 295 |
|
| 296 | |
| 297 | * do_lo_send_direct_write - helper for writing data to a loop device
|
| 298 | *
|
| 299 | * This is the fast, non-transforming version for backing filesystems which do
|
| 300 | * not implement the address space operations write_begin and write_end.
|
| 301 | * It uses the write file operation which should be present on all writeable
|
| 302 | * filesystems.
|
| 303 |
|
| 304 | static int do_lo_send_direct_write(struct loop_device *lo,
|
| 305 | struct bio_vec *bvec, loff_t pos, struct page *page)
|
| 306 | {
|
| 307 | ssize_t bw = __do_lo_send_write(lo->lo_backing_file,
|
| 308 | kmap(bvec->bv_page) + bvec->bv_offset,
|
| 309 | bvec->bv_len, pos);
|
| 310 | kunmap(bvec->bv_page);
|
| 311 | cond_resched();
|
| 312 | return bw;
|
| 313 | }
|
| 314 |
|
| 315 | |
| 316 | * do_lo_send_write - helper for writing data to a loop device
|
| 317 | *
|
| 318 | * This is the slow, transforming version for filesystems which do not
|
| 319 | * implement the address space operations write_begin and write_end. It
|
| 320 | * uses the write file operation which should be present on all writeable
|
| 321 | * filesystems.
|
| 322 | *
|
| 323 | * Using fops->write is slower than using aops->{prepare,commit}_write in the
|
| 324 | * transforming case because we need to double buffer the data as we cannot do
|
| 325 | * the transformations in place as we do not have direct access to the
|
| 326 | * destination pages of the backing file.
|
| 327 |
|
| 328 | static int do_lo_send_write(struct loop_device *lo, struct bio_vec *bvec,
|
| 329 | loff_t pos, struct page *page)
|
| 330 | {
|
| 331 | int ret = lo_do_transfer(lo, WRITE, page, 0, bvec->bv_page,
|
| 332 | bvec->bv_offset, bvec->bv_len, pos >> 9);
|
| 333 | if (likely(!ret))
|
| 334 | return __do_lo_send_write(lo->lo_backing_file,
|
| 335 | page_address(page), bvec->bv_len,
|
| 336 | pos);
|
| 337 | printk(KERN_ERR "loop: Transfer error at byte offset %llu, "
|
| 338 | "length %i.\n", (unsigned long long)pos, bvec->bv_len);
|
| 339 | if (ret > 0)
|
| 340 | ret = -EIO;
|
| 341 | return ret;
|
| 342 | }
|
| 343 |
|
| 344 | static int lo_send(struct loop_device *lo, struct bio *bio, loff_t pos)
|
| 345 | {
|
| 346 | int (*do_lo_send)(struct loop_device *, struct bio_vec *, loff_t,
|
| 347 | struct page *page);
|
| 348 | struct bio_vec *bvec;
|
| 349 | struct page *page = NULL;
|
| 350 | int i, ret = 0;
|
| 351 |
|
| 352 | do_lo_send = do_lo_send_aops;
|
| 353 | if (!(lo->lo_flags & LO_FLAGS_USE_AOPS)) {
|
| 354 | do_lo_send = do_lo_send_direct_write;
|
| 355 | if (lo->transfer != transfer_none) {
|
| 356 | page = alloc_page(GFP_NOIO | __GFP_HIGHMEM);
|
| 357 | if (unlikely(!page))
|
| 358 | goto fail;
|
| 359 | kmap(page);
|
| 360 | do_lo_send = do_lo_send_write;
|
| 361 | }
|
| 362 | }
|
| 363 | bio_for_each_segment(bvec, bio, i) {
|
| 364 | ret = do_lo_send(lo, bvec, pos, page);
|
| 365 | if (ret < 0)
|
| 366 | break;
|
| 367 | pos += bvec->bv_len;
|
| 368 | }
|
| 369 | if (page) {
|
| 370 | kunmap(page);
|
| 371 | __free_page(page);
|
| 372 | }
|
| 373 | out:
|
| 374 | return ret;
|
| 375 | fail:
|
| 376 | printk(KERN_ERR "loop: Failed to allocate temporary page for write.\n");
|
| 377 | ret = -ENOMEM;
|
| 378 | goto out;
|
| 379 | }
|
| 380 |
|
| 381 | struct lo_read_data {
|
| 382 | struct loop_device *lo;
|
| 383 | struct page *page;
|
| 384 | unsigned offset;
|
| 385 | int bsize;
|
| 386 | };
|
| 387 |
|
| 388 | static int
|
| 389 | lo_splice_actor(struct pipe_inode_info *pipe, struct pipe_buffer *buf,
|
| 390 | struct splice_desc *sd)
|
| 391 | {
|
| 392 | struct lo_read_data *p = sd->u.data;
|
| 393 | struct loop_device *lo = p->lo;
|
| 394 | struct page *page = buf->page;
|
| 395 | sector_t IV;
|
| 396 | int size, ret;
|
| 397 |
|
| 398 | ret = buf->ops->confirm(pipe, buf);
|
| 399 | if (unlikely(ret))
|
| 400 | return ret;
|
| 401 |
|
| 402 | IV = ((sector_t) page->index << (PAGE_CACHE_SHIFT - 9)) +
|
| 403 | (buf->offset >> 9);
|
| 404 | size = sd->len;
|
| 405 | if (size > p->bsize)
|
| 406 | size = p->bsize;
|
| 407 |
|
| 408 | if (lo_do_transfer(lo, READ, page, buf->offset, p->page, p->offset, size, IV)) {
|
| 409 | printk(KERN_ERR "loop: transfer error block %ld\n",
|
| 410 | page->index);
|
| 411 | size = -EINVAL;
|
| 412 | }
|
| 413 |
|
| 414 | flush_dcache_page(p->page);
|
| 415 |
|
| 416 | if (size > 0)
|
| 417 | p->offset += size;
|
| 418 |
|
| 419 | return size;
|
| 420 | }
|
| 421 |
|
| 422 | static int
|
| 423 | lo_direct_splice_actor(struct pipe_inode_info *pipe, struct splice_desc *sd)
|
| 424 | {
|
| 425 | return __splice_from_pipe(pipe, sd, lo_splice_actor);
|
| 426 | }
|
| 427 |
|
| 428 | static int
|
| 429 | do_lo_receive(struct loop_device *lo,
|
| 430 | struct bio_vec *bvec, int bsize, loff_t pos)
|
| 431 | {
|
| 432 | struct lo_read_data cookie;
|
| 433 | struct splice_desc sd;
|
| 434 | struct file *file;
|
| 435 | long retval;
|
| 436 |
|
| 437 | cookie.lo = lo;
|
| 438 | cookie.page = bvec->bv_page;
|
| 439 | cookie.offset = bvec->bv_offset;
|
| 440 | cookie.bsize = bsize;
|
| 441 |
|
| 442 | sd.len = 0;
|
| 443 | sd.total_len = bvec->bv_len;
|
| 444 | sd.flags = 0;
|
| 445 | sd.pos = pos;
|
| 446 | sd.u.data = &cookie;
|
| 447 |
|
| 448 | file = lo->lo_backing_file;
|
| 449 | retval = splice_direct_to_actor(file, &sd, lo_direct_splice_actor);
|
| 450 |
|
| 451 | if (retval < 0)
|
| 452 | return retval;
|
| 453 |
|
| 454 | return 0;
|
| 455 | }
|
| 456 |
|
| 457 | static int
|
| 458 | lo_receive(struct loop_device *lo, struct bio *bio, int bsize, loff_t pos)
|
| 459 | {
|
| 460 | struct bio_vec *bvec;
|
| 461 | int i, ret = 0;
|
| 462 |
|
| 463 | bio_for_each_segment(bvec, bio, i) {
|
| 464 | ret = do_lo_receive(lo, bvec, bsize, pos);
|
| 465 | if (ret < 0)
|
| 466 | break;
|
| 467 | pos += bvec->bv_len;
|
| 468 | }
|
| 469 | return ret;
|
| 470 | }
|
| 471 |
|
| 472 | static int do_bio_filebacked(struct loop_device *lo, struct bio *bio)
|
| 473 | {
|
| 474 | loff_t pos;
|
| 475 | int ret;
|
| 476 |
|
| 477 | pos = ((loff_t) bio->bi_sector << 9) + lo->lo_offset;
|
| 478 |
|
| 479 | if (bio_rw(bio) == WRITE) {
|
| 480 | bool barrier = !!(bio->bi_rw & REQ_HARDBARRIER);
|
| 481 | struct file *file = lo->lo_backing_file;
|
| 482 |
|
| 483 | if (barrier) {
|
| 484 | if (unlikely(!file->f_op->fsync)) {
|
| 485 | ret = -EOPNOTSUPP;
|
| 486 | goto out;
|
| 487 | }
|
| 488 |
|
| 489 | ret = vfs_fsync(file, 0);
|
| 490 | if (unlikely(ret)) {
|
| 491 | ret = -EIO;
|
| 492 | goto out;
|
| 493 | }
|
| 494 | }
|
| 495 |
|
| 496 | ret = lo_send(lo, bio, pos);
|
| 497 |
|
| 498 | if (barrier && !ret) {
|
| 499 | ret = vfs_fsync(file, 0);
|
| 500 | if (unlikely(ret))
|
| 501 | ret = -EIO;
|
| 502 | }
|
| 503 | } else
|
| 504 | ret = lo_receive(lo, bio, lo->lo_blocksize, pos);
|
| 505 |
|
| 506 | out:
|
| 507 | return ret;
|
| 508 | }
|
| 509 |
|
| 510 | |
| 511 | * Add bio to back of pending list
|
| 512 |
|
| 513 | static void loop_add_bio(struct loop_device *lo, struct bio *bio)
|
| 514 | {
|
| 515 | bio_list_add(&lo->lo_bio_list, bio);
|
| 516 | }
|
| 517 |
|
| 518 | |
| 519 | * Grab first pending buffer
|
| 520 |
|
| 521 | static struct bio *loop_get_bio(struct loop_device *lo)
|
| 522 | {
|
| 523 | return bio_list_pop(&lo->lo_bio_list);
|
| 524 | }
|
| 525 |
|
| 526 | static int loop_make_request(struct request_queue *q, struct bio *old_bio)
|
| 527 | {
|
| 528 | struct loop_device *lo = q->queuedata;
|
| 529 | int rw = bio_rw(old_bio);
|
| 530 |
|
| 531 | if (rw == READA)
|
| 532 | rw = READ;
|
| 533 |
|
| 534 | BUG_ON(!lo || (rw != READ && rw != WRITE));
|
| 535 |
|
| 536 | spin_lock_irq(&lo->lo_lock);
|
| 537 | if (lo->lo_state != Lo_bound)
|
| 538 | goto out;
|
| 539 | if (unlikely(rw == WRITE && (lo->lo_flags & LO_FLAGS_READ_ONLY)))
|
| 540 | goto out;
|
| 541 | loop_add_bio(lo, old_bio);
|
| 542 | wake_up(&lo->lo_event);
|
| 543 | spin_unlock_irq(&lo->lo_lock);
|
| 544 | return 0;
|
| 545 |
|
| 546 | out:
|
| 547 | spin_unlock_irq(&lo->lo_lock);
|
| 548 | bio_io_error(old_bio);
|
| 549 | return 0;
|
| 550 | }
|
| 551 |
|
| 552 | |
| 553 | * kick off io on the underlying address space
|
| 554 |
|
| 555 | static void loop_unplug(struct request_queue *q)
|
| 556 | {
|
| 557 | struct loop_device *lo = q->queuedata;
|
| 558 |
|
| 559 | queue_flag_clear_unlocked(QUEUE_FLAG_PLUGGED, q);
|
| 560 | blk_run_address_space(lo->lo_backing_file->f_mapping);
|
| 561 | }
|
| 562 |
|
| 563 | struct switch_request {
|
| 564 | struct file *file;
|
| 565 | struct completion wait;
|
| 566 | };
|
| 567 |
|
| 568 | static void do_loop_switch(struct loop_device *, struct switch_request *);
|
| 569 |
|
| 570 | static inline void loop_handle_bio(struct loop_device *lo, struct bio *bio)
|
| 571 | {
|
| 572 | if (unlikely(!bio->bi_bdev)) {
|
| 573 | do_loop_switch(lo, bio->bi_private);
|
| 574 | bio_put(bio);
|
| 575 | } else {
|
| 576 | int ret = do_bio_filebacked(lo, bio);
|
| 577 | bio_endio(bio, ret);
|
| 578 | }
|
| 579 | }
|
| 580 |
|
| 581 | |
| 582 | * worker thread that handles reads/writes to file backed loop devices,
|
| 583 | * to avoid blocking in our make_request_fn. it also does loop decrypting
|
| 584 | * on reads for block backed loop, as that is too heavy to do from
|
| 585 | * b_end_io context where irqs may be disabled.
|
| 586 | *
|
| 587 | * Loop explanation: loop_clr_fd() sets lo_state to Lo_rundown before
|
| 588 | * calling kthread_stop(). Therefore once kthread_should_stop() is
|
| 589 | * true, make_request will not place any more requests. Therefore
|
| 590 | * once kthread_should_stop() is true and lo_bio is NULL, we are
|
| 591 | * done with the loop.
|
| 592 |
|
| 593 | static int loop_thread(void *data)
|
| 594 | {
|
| 595 | struct loop_device *lo = data;
|
| 596 | struct bio *bio;
|
| 597 |
|
| 598 | set_user_nice(current, -20);
|
| 599 |
|
| 600 | while (!kthread_should_stop() || !bio_list_empty(&lo->lo_bio_list)) {
|
| 601 |
|
| 602 | wait_event_interruptible(lo->lo_event,
|
| 603 | !bio_list_empty(&lo->lo_bio_list) ||
|
| 604 | kthread_should_stop());
|
| 605 |
|
| 606 | if (bio_list_empty(&lo->lo_bio_list))
|
| 607 | continue;
|
| 608 | spin_lock_irq(&lo->lo_lock);
|
| 609 | bio = loop_get_bio(lo);
|
| 610 | spin_unlock_irq(&lo->lo_lock);
|
| 611 |
|
| 612 | BUG_ON(!bio);
|
| 613 | loop_handle_bio(lo, bio);
|
| 614 | }
|
| 615 |
|
| 616 | return 0;
|
| 617 | }
|
| 618 |
|
| 619 | |
| 620 | * loop_switch performs the hard work of switching a backing store.
|
| 621 | * First it needs to flush existing IO, it does this by sending a magic
|
| 622 | * BIO down the pipe. The completion of this BIO does the actual switch.
|
| 623 |
|
| 624 | static int loop_switch(struct loop_device *lo, struct file *file)
|
| 625 | {
|
| 626 | struct switch_request w;
|
| 627 | struct bio *bio = bio_alloc(GFP_KERNEL, 0);
|
| 628 | if (!bio)
|
| 629 | return -ENOMEM;
|
| 630 | init_completion(&w.wait);
|
| 631 | w.file = file;
|
| 632 | bio->bi_private = &w;
|
| 633 | bio->bi_bdev = NULL;
|
| 634 | loop_make_request(lo->lo_queue, bio);
|
| 635 | wait_for_completion(&w.wait);
|
| 636 | return 0;
|
| 637 | }
|
| 638 |
|
| 639 | |
| 640 | * Helper to flush the IOs in loop, but keeping loop thread running
|
| 641 |
|
| 642 | static int loop_flush(struct loop_device *lo)
|
| 643 | {
|
| 644 |
|
| 645 | if (!lo->lo_thread)
|
| 646 | return 0;
|
| 647 |
|
| 648 | return loop_switch(lo, NULL);
|
| 649 | }
|
| 650 |
|
| 651 | |
| 652 | * Do the actual switch; called from the BIO completion routine
|
| 653 |
|
| 654 | static void do_loop_switch(struct loop_device *lo, struct switch_request *p)
|
| 655 | {
|
| 656 | struct file *file = p->file;
|
| 657 | struct file *old_file = lo->lo_backing_file;
|
| 658 | struct address_space *mapping;
|
| 659 |
|
| 660 |
|
| 661 | if (!file)
|
| 662 | goto out;
|
| 663 |
|
| 664 | mapping = file->f_mapping;
|
| 665 | mapping_set_gfp_mask(old_file->f_mapping, lo->old_gfp_mask);
|
| 666 | lo->lo_backing_file = file;
|
| 667 | lo->lo_blocksize = S_ISBLK(mapping->host->i_mode) ?
|
| 668 | mapping->host->i_bdev->bd_block_size : PAGE_SIZE;
|
| 669 | lo->old_gfp_mask = mapping_gfp_mask(mapping);
|
| 670 | mapping_set_gfp_mask(mapping, lo->old_gfp_mask & ~(__GFP_IO|__GFP_FS));
|
| 671 | out:
|
| 672 | complete(&p->wait);
|
| 673 | }
|
| 674 |
|
| 675 |
|
| 676 | |
| 677 | * loop_change_fd switched the backing store of a loopback device to
|
| 678 | * a new file. This is useful for operating system installers to free up
|
| 679 | * the original file and in High Availability environments to switch to
|
| 680 | * an alternative location for the content in case of server meltdown.
|
| 681 | * This can only work if the loop device is used read-only, and if the
|
| 682 | * new backing store is the same size and type as the old backing store.
|
| 683 |
|
| 684 | static int loop_change_fd(struct loop_device *lo, struct block_device *bdev,
|
| 685 | unsigned int arg)
|
| 686 | {
|
| 687 | struct file *file, *old_file;
|
| 688 | struct inode *inode;
|
| 689 | int error;
|
| 690 |
|
| 691 | error = -ENXIO;
|
| 692 | if (lo->lo_state != Lo_bound)
|
| 693 | goto out;
|
| 694 |
|
| 695 |
|
| 696 | error = -EINVAL;
|
| 697 | if (!(lo->lo_flags & LO_FLAGS_READ_ONLY))
|
| 698 | goto out;
|
| 699 |
|
| 700 | error = -EBADF;
|
| 701 | file = fget(arg);
|
| 702 | if (!file)
|
| 703 | goto out;
|
| 704 |
|
| 705 | inode = file->f_mapping->host;
|
| 706 | old_file = lo->lo_backing_file;
|
| 707 |
|
| 708 | error = -EINVAL;
|
| 709 |
|
| 710 | if (!S_ISREG(inode->i_mode) && !S_ISBLK(inode->i_mode))
|
| 711 | goto out_putf;
|
| 712 |
|
| 713 |
|
| 714 | if (get_loop_size(lo, file) != get_loop_size(lo, old_file))
|
| 715 | goto out_putf;
|
| 716 |
|
| 717 |
|
| 718 | error = loop_switch(lo, file);
|
| 719 | if (error)
|
| 720 | goto out_putf;
|
| 721 |
|
| 722 | fput(old_file);
|
| 723 | if (max_part > 0)
|
| 724 | ioctl_by_bdev(bdev, BLKRRPART, 0);
|
| 725 | return 0;
|
| 726 |
|
| 727 | out_putf:
|
| 728 | fput(file);
|
| 729 | out:
|
| 730 | return error;
|
| 731 | }
|
| 732 |
|
| 733 | static inline int is_loop_device(struct file *file)
|
| 734 | {
|
| 735 | struct inode *i = file->f_mapping->host;
|
| 736 |
|
| 737 | return i && S_ISBLK(i->i_mode) && MAJOR(i->i_rdev) == LOOP_MAJOR;
|
| 738 | }
|
| 739 |
|
| 740 | static int loop_set_fd(struct loop_device *lo, fmode_t mode,
|
| 741 | struct block_device *bdev, unsigned int arg)
|
| 742 | {
|
| 743 | struct file *file, *f;
|
| 744 | struct inode *inode;
|
| 745 | struct address_space *mapping;
|
| 746 | unsigned lo_blocksize;
|
| 747 | int lo_flags = 0;
|
| 748 | int error;
|
| 749 | loff_t size;
|
| 750 |
|
| 751 |
|
| 752 | __module_get(THIS_MODULE);
|
| 753 |
|
| 754 | error = -EBADF;
|
| 755 | file = fget(arg);
|
| 756 | if (!file)
|
| 757 | goto out;
|
| 758 |
|
| 759 | error = -EBUSY;
|
| 760 | if (lo->lo_state != Lo_unbound)
|
| 761 | goto out_putf;
|
| 762 |
|
| 763 |
|
| 764 | f = file;
|
| 765 | while (is_loop_device(f)) {
|
| 766 | struct loop_device *l;
|
| 767 |
|
| 768 | if (f->f_mapping->host->i_bdev == bdev)
|
| 769 | goto out_putf;
|
| 770 |
|
| 771 | l = f->f_mapping->host->i_bdev->bd_disk->private_data;
|
| 772 | if (l->lo_state == Lo_unbound) {
|
| 773 | error = -EINVAL;
|
| 774 | goto out_putf;
|
| 775 | }
|
| 776 | f = l->lo_backing_file;
|
| 777 | }
|
| 778 |
|
| 779 | mapping = file->f_mapping;
|
| 780 | inode = mapping->host;
|
| 781 |
|
| 782 | if (!(file->f_mode & FMODE_WRITE))
|
| 783 | lo_flags |= LO_FLAGS_READ_ONLY;
|
| 784 |
|
| 785 | error = -EINVAL;
|
| 786 | if (S_ISREG(inode->i_mode) || S_ISBLK(inode->i_mode)) {
|
| 787 | const struct address_space_operations *aops = mapping->a_ops;
|
| 788 |
|
| 789 | if (aops->write_begin)
|
| 790 | lo_flags |= LO_FLAGS_USE_AOPS;
|
| 791 | if (!(lo_flags & LO_FLAGS_USE_AOPS) && !file->f_op->write)
|
| 792 | lo_flags |= LO_FLAGS_READ_ONLY;
|
| 793 |
|
| 794 | lo_blocksize = S_ISBLK(inode->i_mode) ?
|
| 795 | inode->i_bdev->bd_block_size : PAGE_SIZE;
|
| 796 |
|
| 797 | error = 0;
|
| 798 | } else {
|
| 799 | goto out_putf;
|
| 800 | }
|
| 801 |
|
| 802 | size = get_loop_size(lo, file);
|
| 803 |
|
| 804 | if ((loff_t)(sector_t)size != size) {
|
| 805 | error = -EFBIG;
|
| 806 | goto out_putf;
|
| 807 | }
|
| 808 |
|
| 809 | if (!(mode & FMODE_WRITE))
|
| 810 | lo_flags |= LO_FLAGS_READ_ONLY;
|
| 811 |
|
| 812 | set_device_ro(bdev, (lo_flags & LO_FLAGS_READ_ONLY) != 0);
|
| 813 |
|
| 814 | lo->lo_blocksize = lo_blocksize;
|
| 815 | lo->lo_device = bdev;
|
| 816 | lo->lo_flags = lo_flags;
|
| 817 | lo->lo_backing_file = file;
|
| 818 | lo->transfer = transfer_none;
|
| 819 | lo->ioctl = NULL;
|
| 820 | lo->lo_sizelimit = 0;
|
| 821 | lo->old_gfp_mask = mapping_gfp_mask(mapping);
|
| 822 | mapping_set_gfp_mask(mapping, lo->old_gfp_mask & ~(__GFP_IO|__GFP_FS));
|
| 823 |
|
| 824 | bio_list_init(&lo->lo_bio_list);
|
| 825 |
|
| 826 | |
| 827 | * set queue make_request_fn, and add limits based on lower level
|
| 828 | * device
|
| 829 |
|
| 830 | blk_queue_make_request(lo->lo_queue, loop_make_request);
|
| 831 | lo->lo_queue->queuedata = lo;
|
| 832 | lo->lo_queue->unplug_fn = loop_unplug;
|
| 833 |
|
| 834 | if (!(lo_flags & LO_FLAGS_READ_ONLY) && file->f_op->fsync)
|
| 835 | blk_queue_ordered(lo->lo_queue, QUEUE_ORDERED_DRAIN);
|
| 836 |
|
| 837 | set_capacity(lo->lo_disk, size);
|
| 838 | bd_set_size(bdev, size << 9);
|
| 839 |
|
| 840 | kobject_uevent(&disk_to_dev(bdev->bd_disk)->kobj, KOBJ_CHANGE);
|
| 841 |
|
| 842 | set_blocksize(bdev, lo_blocksize);
|
| 843 |
|
| 844 | lo->lo_thread = kthread_create(loop_thread, lo, "loop%d",
|
| 845 | lo->lo_number);
|
| 846 | if (IS_ERR(lo->lo_thread)) {
|
| 847 | error = PTR_ERR(lo->lo_thread);
|
| 848 | goto out_clr;
|
| 849 | }
|
| 850 | lo->lo_state = Lo_bound;
|
| 851 | wake_up_process(lo->lo_thread);
|
| 852 | if (max_part > 0)
|
| 853 | ioctl_by_bdev(bdev, BLKRRPART, 0);
|
| 854 | return 0;
|
| 855 |
|
| 856 | out_clr:
|
| 857 | lo->lo_thread = NULL;
|
| 858 | lo->lo_device = NULL;
|
| 859 | lo->lo_backing_file = NULL;
|
| 860 | lo->lo_flags = 0;
|
| 861 | set_capacity(lo->lo_disk, 0);
|
| 862 | invalidate_bdev(bdev);
|
| 863 | bd_set_size(bdev, 0);
|
| 864 | kobject_uevent(&disk_to_dev(bdev->bd_disk)->kobj, KOBJ_CHANGE);
|
| 865 | mapping_set_gfp_mask(mapping, lo->old_gfp_mask);
|
| 866 | lo->lo_state = Lo_unbound;
|
| 867 | out_putf:
|
| 868 | fput(file);
|
| 869 | out:
|
| 870 |
|
| 871 | module_put(THIS_MODULE);
|
| 872 | return error;
|
| 873 | }
|
| 874 |
|
| 875 | static int
|
| 876 | loop_release_xfer(struct loop_device *lo)
|
| 877 | {
|
| 878 | int err = 0;
|
| 879 | struct loop_func_table *xfer = lo->lo_encryption;
|
| 880 |
|
| 881 | if (xfer) {
|
| 882 | if (xfer->release)
|
| 883 | err = xfer->release(lo);
|
| 884 | lo->transfer = NULL;
|
| 885 | lo->lo_encryption = NULL;
|
| 886 | module_put(xfer->owner);
|
| 887 | }
|
| 888 | return err;
|
| 889 | }
|
| 890 |
|
| 891 | static int
|
| 892 | loop_init_xfer(struct loop_device *lo, struct loop_func_table *xfer,
|
| 893 | const struct loop_info64 *i)
|
| 894 | {
|
| 895 | int err = 0;
|
| 896 |
|
| 897 | if (xfer) {
|
| 898 | struct module *owner = xfer->owner;
|
| 899 |
|
| 900 | if (!try_module_get(owner))
|
| 901 | return -EINVAL;
|
| 902 | if (xfer->init)
|
| 903 | err = xfer->init(lo, i);
|
| 904 | if (err)
|
| 905 | module_put(owner);
|
| 906 | else
|
| 907 | lo->lo_encryption = xfer;
|
| 908 | }
|
| 909 | return err;
|
| 910 | }
|
| 911 |
|
| 912 | static int loop_clr_fd(struct loop_device *lo, struct block_device *bdev)
|
| 913 | {
|
| 914 | struct file *filp = lo->lo_backing_file;
|
| 915 | gfp_t gfp = lo->old_gfp_mask;
|
| 916 |
|
| 917 | if (lo->lo_state != Lo_bound)
|
| 918 | return -ENXIO;
|
| 919 |
|
| 920 | if (lo->lo_refcnt > 1)
|
| 921 | return -EBUSY;
|
| 922 |
|
| 923 | if (filp == NULL)
|
| 924 | return -EINVAL;
|
| 925 |
|
| 926 | spin_lock_irq(&lo->lo_lock);
|
| 927 | lo->lo_state = Lo_rundown;
|
| 928 | spin_unlock_irq(&lo->lo_lock);
|
| 929 |
|
| 930 | kthread_stop(lo->lo_thread);
|
| 931 |
|
| 932 | lo->lo_queue->unplug_fn = NULL;
|
| 933 | lo->lo_backing_file = NULL;
|
| 934 |
|
| 935 | loop_release_xfer(lo);
|
| 936 | lo->transfer = NULL;
|
| 937 | lo->ioctl = NULL;
|
| 938 | lo->lo_device = NULL;
|
| 939 | lo->lo_encryption = NULL;
|
| 940 | lo->lo_offset = 0;
|
| 941 | lo->lo_sizelimit = 0;
|
| 942 | lo->lo_encrypt_key_size = 0;
|
| 943 | lo->lo_flags = 0;
|
| 944 | lo->lo_thread = NULL;
|
| 945 | memset(lo->lo_encrypt_key, 0, LO_KEY_SIZE);
|
| 946 | memset(lo->lo_crypt_name, 0, LO_NAME_SIZE);
|
| 947 | memset(lo->lo_file_name, 0, LO_NAME_SIZE);
|
| 948 | if (bdev)
|
| 949 | invalidate_bdev(bdev);
|
| 950 | set_capacity(lo->lo_disk, 0);
|
| 951 | if (bdev) {
|
| 952 | bd_set_size(bdev, 0);
|
| 953 |
|
| 954 | kobject_uevent(&disk_to_dev(bdev->bd_disk)->kobj, KOBJ_CHANGE);
|
| 955 | }
|
| 956 | mapping_set_gfp_mask(filp->f_mapping, gfp);
|
| 957 | lo->lo_state = Lo_unbound;
|
| 958 |
|
| 959 | module_put(THIS_MODULE);
|
| 960 | if (max_part > 0 && bdev)
|
| 961 | ioctl_by_bdev(bdev, BLKRRPART, 0);
|
| 962 | mutex_unlock(&lo->lo_ctl_mutex);
|
| 963 | |
| 964 | * Need not hold lo_ctl_mutex to fput backing file.
|
| 965 | * Calling fput holding lo_ctl_mutex triggers a circular
|
| 966 | * lock dependency possibility warning as fput can take
|
| 967 | * bd_mutex which is usually taken before lo_ctl_mutex.
|
| 968 |
|
| 969 | fput(filp);
|
| 970 | return 0;
|
| 971 | }
|
| 972 |
|
| 973 | static int
|
| 974 | loop_set_status(struct loop_device *lo, const struct loop_info64 *info)
|
| 975 | {
|
| 976 | int err;
|
| 977 | struct loop_func_table *xfer;
|
| 978 | uid_t uid = current_uid();
|
| 979 |
|
| 980 | if (lo->lo_encrypt_key_size &&
|
| 981 | lo->lo_key_owner != uid &&
|
| 982 | !capable(CAP_SYS_ADMIN))
|
| 983 | return -EPERM;
|
| 984 | if (lo->lo_state != Lo_bound)
|
| 985 | return -ENXIO;
|
| 986 | if ((unsigned int) info->lo_encrypt_key_size > LO_KEY_SIZE)
|
| 987 | return -EINVAL;
|
| 988 |
|
| 989 | err = loop_release_xfer(lo);
|
| 990 | if (err)
|
| 991 | return err;
|
| 992 |
|
| 993 | if (info->lo_encrypt_type) {
|
| 994 | unsigned int type = info->lo_encrypt_type;
|
| 995 |
|
| 996 | if (type >= MAX_LO_CRYPT)
|
| 997 | return -EINVAL;
|
| 998 | xfer = xfer_funcs[type];
|
| 999 | if (xfer == NULL)
|
| 1000 | return -EINVAL;
|
| 1001 | } else
|
| 1002 | xfer = NULL;
|
| 1003 |
|
| 1004 | err = loop_init_xfer(lo, xfer, info);
|
| 1005 | if (err)
|
| 1006 | return err;
|
| 1007 |
|
| 1008 | if (lo->lo_offset != info->lo_offset ||
|
| 1009 | lo->lo_sizelimit != info->lo_sizelimit) {
|
| 1010 | lo->lo_offset = info->lo_offset;
|
| 1011 | lo->lo_sizelimit = info->lo_sizelimit;
|
| 1012 | if (figure_loop_size(lo))
|
| 1013 | return -EFBIG;
|
| 1014 | }
|
| 1015 |
|
| 1016 | memcpy(lo->lo_file_name, info->lo_file_name, LO_NAME_SIZE);
|
| 1017 | memcpy(lo->lo_crypt_name, info->lo_crypt_name, LO_NAME_SIZE);
|
| 1018 | lo->lo_file_name[LO_NAME_SIZE-1] = 0;
|
| 1019 | lo->lo_crypt_name[LO_NAME_SIZE-1] = 0;
|
| 1020 |
|
| 1021 | if (!xfer)
|
| 1022 | xfer = &none_funcs;
|
| 1023 | lo->transfer = xfer->transfer;
|
| 1024 | lo->ioctl = xfer->ioctl;
|
| 1025 |
|
| 1026 | if ((lo->lo_flags & LO_FLAGS_AUTOCLEAR) !=
|
| 1027 | (info->lo_flags & LO_FLAGS_AUTOCLEAR))
|
| 1028 | lo->lo_flags ^= LO_FLAGS_AUTOCLEAR;
|
| 1029 |
|
| 1030 | lo->lo_encrypt_key_size = info->lo_encrypt_key_size;
|
| 1031 | lo->lo_init[0] = info->lo_init[0];
|
| 1032 | lo->lo_init[1] = info->lo_init[1];
|
| 1033 | if (info->lo_encrypt_key_size) {
|
| 1034 | memcpy(lo->lo_encrypt_key, info->lo_encrypt_key,
|
| 1035 | info->lo_encrypt_key_size);
|
| 1036 | lo->lo_key_owner = uid;
|
| 1037 | }
|
| 1038 |
|
| 1039 | return 0;
|
| 1040 | }
|
| 1041 |
|
| 1042 | static int
|
| 1043 | loop_get_status(struct loop_device *lo, struct loop_info64 *info)
|
| 1044 | {
|
| 1045 | struct file *file = lo->lo_backing_file;
|
| 1046 | struct kstat stat;
|
| 1047 | int error;
|
| 1048 |
|
| 1049 | if (lo->lo_state != Lo_bound)
|
| 1050 | return -ENXIO;
|
| 1051 | error = vfs_getattr(file->f_path.mnt, file->f_path.dentry, &stat);
|
| 1052 | if (error)
|
| 1053 | return error;
|
| 1054 | memset(info, 0, sizeof(*info));
|
| 1055 | info->lo_number = lo->lo_number;
|
| 1056 | info->lo_device = huge_encode_dev(stat.dev);
|
| 1057 | info->lo_inode = stat.ino;
|
| 1058 | info->lo_rdevice = huge_encode_dev(lo->lo_device ? stat.rdev : stat.dev);
|
| 1059 | info->lo_offset = lo->lo_offset;
|
| 1060 | info->lo_sizelimit = lo->lo_sizelimit;
|
| 1061 | info->lo_flags = lo->lo_flags;
|
| 1062 | memcpy(info->lo_file_name, lo->lo_file_name, LO_NAME_SIZE);
|
| 1063 | memcpy(info->lo_crypt_name, lo->lo_crypt_name, LO_NAME_SIZE);
|
| 1064 | info->lo_encrypt_type =
|
| 1065 | lo->lo_encryption ? lo->lo_encryption->number : 0;
|
| 1066 | if (lo->lo_encrypt_key_size && capable(CAP_SYS_ADMIN)) {
|
| 1067 | info->lo_encrypt_key_size = lo->lo_encrypt_key_size;
|
| 1068 | memcpy(info->lo_encrypt_key, lo->lo_encrypt_key,
|
| 1069 | lo->lo_encrypt_key_size);
|
| 1070 | }
|
| 1071 | return 0;
|
| 1072 | }
|
| 1073 |
|
| 1074 | static void
|
| 1075 | loop_info64_from_old(const struct loop_info *info, struct loop_info64 *info64)
|
| 1076 | {
|
| 1077 | memset(info64, 0, sizeof(*info64));
|
| 1078 | info64->lo_number = info->lo_number;
|
| 1079 | info64->lo_device = info->lo_device;
|
| 1080 | info64->lo_inode = info->lo_inode;
|
| 1081 | info64->lo_rdevice = info->lo_rdevice;
|
| 1082 | info64->lo_offset = info->lo_offset;
|
| 1083 | info64->lo_sizelimit = 0;
|
| 1084 | info64->lo_encrypt_type = info->lo_encrypt_type;
|
| 1085 | info64->lo_encrypt_key_size = info->lo_encrypt_key_size;
|
| 1086 | info64->lo_flags = info->lo_flags;
|
| 1087 | info64->lo_init[0] = info->lo_init[0];
|
| 1088 | info64->lo_init[1] = info->lo_init[1];
|
| 1089 | if (info->lo_encrypt_type == LO_CRYPT_CRYPTOAPI)
|
| 1090 | memcpy(info64->lo_crypt_name, info->lo_name, LO_NAME_SIZE);
|
| 1091 | else
|
| 1092 | memcpy(info64->lo_file_name, info->lo_name, LO_NAME_SIZE);
|
| 1093 | memcpy(info64->lo_encrypt_key, info->lo_encrypt_key, LO_KEY_SIZE);
|
| 1094 | }
|
| 1095 |
|
| 1096 | static int
|
| 1097 | loop_info64_to_old(const struct loop_info64 *info64, struct loop_info *info)
|
| 1098 | {
|
| 1099 | memset(info, 0, sizeof(*info));
|
| 1100 | info->lo_number = info64->lo_number;
|
| 1101 | info->lo_device = info64->lo_device;
|
| 1102 | info->lo_inode = info64->lo_inode;
|
| 1103 | info->lo_rdevice = info64->lo_rdevice;
|
| 1104 | info->lo_offset = info64->lo_offset;
|
| 1105 | info->lo_encrypt_type = info64->lo_encrypt_type;
|
| 1106 | info->lo_encrypt_key_size = info64->lo_encrypt_key_size;
|
| 1107 | info->lo_flags = info64->lo_flags;
|
| 1108 | info->lo_init[0] = info64->lo_init[0];
|
| 1109 | info->lo_init[1] = info64->lo_init[1];
|
| 1110 | if (info->lo_encrypt_type == LO_CRYPT_CRYPTOAPI)
|
| 1111 | memcpy(info->lo_name, info64->lo_crypt_name, LO_NAME_SIZE);
|
| 1112 | else
|
| 1113 | memcpy(info->lo_name, info64->lo_file_name, LO_NAME_SIZE);
|
| 1114 | memcpy(info->lo_encrypt_key, info64->lo_encrypt_key, LO_KEY_SIZE);
|
| 1115 |
|
| 1116 |
|
| 1117 | if (info->lo_device != info64->lo_device ||
|
| 1118 | info->lo_rdevice != info64->lo_rdevice ||
|
| 1119 | info->lo_inode != info64->lo_inode ||
|
| 1120 | info->lo_offset != info64->lo_offset)
|
| 1121 | return -EOVERFLOW;
|
| 1122 |
|
| 1123 | return 0;
|
| 1124 | }
|
| 1125 |
|
| 1126 | static int
|
| 1127 | loop_set_status_old(struct loop_device *lo, const struct loop_info __user *arg)
|
| 1128 | {
|
| 1129 | struct loop_info info;
|
| 1130 | struct loop_info64 info64;
|
| 1131 |
|
| 1132 | if (copy_from_user(&info, arg, sizeof (struct loop_info)))
|
| 1133 | return -EFAULT;
|
| 1134 | loop_info64_from_old(&info, &info64);
|
| 1135 | return loop_set_status(lo, &info64);
|
| 1136 | }
|
| 1137 |
|
| 1138 | static int
|
| 1139 | loop_set_status64(struct loop_device *lo, const struct loop_info64 __user *arg)
|
| 1140 | {
|
| 1141 | struct loop_info64 info64;
|
| 1142 |
|
| 1143 | if (copy_from_user(&info64, arg, sizeof (struct loop_info64)))
|
| 1144 | return -EFAULT;
|
| 1145 | return loop_set_status(lo, &info64);
|
| 1146 | }
|
| 1147 |
|
| 1148 | static int
|
| 1149 | loop_get_status_old(struct loop_device *lo, struct loop_info __user *arg) {
|
| 1150 | struct loop_info info;
|
| 1151 | struct loop_info64 info64;
|
| 1152 | int err = 0;
|
| 1153 |
|
| 1154 | if (!arg)
|
| 1155 | err = -EINVAL;
|
| 1156 | if (!err)
|
| 1157 | err = loop_get_status(lo, &info64);
|
| 1158 | if (!err)
|
| 1159 | err = loop_info64_to_old(&info64, &info);
|
| 1160 | if (!err && copy_to_user(arg, &info, sizeof(info)))
|
| 1161 | err = -EFAULT;
|
| 1162 |
|
| 1163 | return err;
|
| 1164 | }
|
| 1165 |
|
| 1166 | static int
|
| 1167 | loop_get_status64(struct loop_device *lo, struct loop_info64 __user *arg) {
|
| 1168 | struct loop_info64 info64;
|
| 1169 | int err = 0;
|
| 1170 |
|
| 1171 | if (!arg)
|
| 1172 | err = -EINVAL;
|
| 1173 | if (!err)
|
| 1174 | err = loop_get_status(lo, &info64);
|
| 1175 | if (!err && copy_to_user(arg, &info64, sizeof(info64)))
|
| 1176 | err = -EFAULT;
|
| 1177 |
|
| 1178 | return err;
|
| 1179 | }
|
| 1180 |
|
| 1181 | static int loop_set_capacity(struct loop_device *lo, struct block_device *bdev)
|
| 1182 | {
|
| 1183 | int err;
|
| 1184 | sector_t sec;
|
| 1185 | loff_t sz;
|
| 1186 |
|
| 1187 | err = -ENXIO;
|
| 1188 | if (unlikely(lo->lo_state != Lo_bound))
|
| 1189 | goto out;
|
| 1190 | err = figure_loop_size(lo);
|
| 1191 | if (unlikely(err))
|
| 1192 | goto out;
|
| 1193 | sec = get_capacity(lo->lo_disk);
|
| 1194 |
|
| 1195 | sz = sec;
|
| 1196 | sz <<= 9;
|
| 1197 | mutex_lock(&bdev->bd_mutex);
|
| 1198 | bd_set_size(bdev, sz);
|
| 1199 |
|
| 1200 | kobject_uevent(&disk_to_dev(bdev->bd_disk)->kobj, KOBJ_CHANGE);
|
| 1201 | mutex_unlock(&bdev->bd_mutex);
|
| 1202 |
|
| 1203 | out:
|
| 1204 | return err;
|
| 1205 | }
|
| 1206 |
|
| 1207 | static int lo_ioctl(struct block_device *bdev, fmode_t mode,
|
| 1208 | unsigned int cmd, unsigned long arg)
|
| 1209 | {
|
| 1210 | struct loop_device *lo = bdev->bd_disk->private_data;
|
| 1211 | int err;
|
| 1212 |
|
| 1213 | mutex_lock_nested(&lo->lo_ctl_mutex, 1);
|
| 1214 | switch (cmd) {
|
| 1215 | case LOOP_SET_FD:
|
| 1216 | err = loop_set_fd(lo, mode, bdev, arg);
|
| 1217 | break;
|
| 1218 | case LOOP_CHANGE_FD:
|
| 1219 | err = loop_change_fd(lo, bdev, arg);
|
| 1220 | break;
|
| 1221 | case LOOP_CLR_FD:
|
| 1222 |
|
| 1223 | err = loop_clr_fd(lo, bdev);
|
| 1224 | if (!err)
|
| 1225 | goto out_unlocked;
|
| 1226 | break;
|
| 1227 | case LOOP_SET_STATUS:
|
| 1228 | err = loop_set_status_old(lo, (struct loop_info __user *) arg);
|
| 1229 | break;
|
| 1230 | case LOOP_GET_STATUS:
|
| 1231 | err = loop_get_status_old(lo, (struct loop_info __user *) arg);
|
| 1232 | break;
|
| 1233 | case LOOP_SET_STATUS64:
|
| 1234 | err = loop_set_status64(lo, (struct loop_info64 __user *) arg);
|
| 1235 | break;
|
| 1236 | case LOOP_GET_STATUS64:
|
| 1237 | err = loop_get_status64(lo, (struct loop_info64 __user *) arg);
|
| 1238 | break;
|
| 1239 | case LOOP_SET_CAPACITY:
|
| 1240 | err = -EPERM;
|
| 1241 | if ((mode & FMODE_WRITE) || capable(CAP_SYS_ADMIN))
|
| 1242 | err = loop_set_capacity(lo, bdev);
|
| 1243 | break;
|
| 1244 | default:
|
| 1245 | err = lo->ioctl ? lo->ioctl(lo, cmd, arg) : -EINVAL;
|
| 1246 | }
|
| 1247 | mutex_unlock(&lo->lo_ctl_mutex);
|
| 1248 |
|
| 1249 | out_unlocked:
|
| 1250 | return err;
|
| 1251 | }
|
| 1252 |
|
| 1253 | #ifdef CONFIG_COMPAT
|
| 1254 | struct compat_loop_info {
|
| 1255 | compat_int_t lo_number;
|
| 1256 | compat_dev_t lo_device;
|
| 1257 | compat_ulong_t lo_inode;
|
| 1258 | compat_dev_t lo_rdevice;
|
| 1259 | compat_int_t lo_offset;
|
| 1260 | compat_int_t lo_encrypt_type;
|
| 1261 | compat_int_t lo_encrypt_key_size;
|
| 1262 | compat_int_t lo_flags;
|
| 1263 | char lo_name[LO_NAME_SIZE];
|
| 1264 | unsigned char lo_encrypt_key[LO_KEY_SIZE];
|
| 1265 | compat_ulong_t lo_init[2];
|
| 1266 | char reserved[4];
|
| 1267 | };
|
| 1268 |
|
| 1269 | |
| 1270 | * Transfer 32-bit compatibility structure in userspace to 64-bit loop info
|
| 1271 | * - noinlined to reduce stack space usage in main part of driver
|
| 1272 |
|
| 1273 | static noinline int
|
| 1274 | loop_info64_from_compat(const struct compat_loop_info __user *arg,
|
| 1275 | struct loop_info64 *info64)
|
| 1276 | {
|
| 1277 | struct compat_loop_info info;
|
| 1278 |
|
| 1279 | if (copy_from_user(&info, arg, sizeof(info)))
|
| 1280 | return -EFAULT;
|
| 1281 |
|
| 1282 | memset(info64, 0, sizeof(*info64));
|
| 1283 | info64->lo_number = info.lo_number;
|
| 1284 | info64->lo_device = info.lo_device;
|
| 1285 | info64->lo_inode = info.lo_inode;
|
| 1286 | info64->lo_rdevice = info.lo_rdevice;
|
| 1287 | info64->lo_offset = info.lo_offset;
|
| 1288 | info64->lo_sizelimit = 0;
|
| 1289 | info64->lo_encrypt_type = info.lo_encrypt_type;
|
| 1290 | info64->lo_encrypt_key_size = info.lo_encrypt_key_size;
|
| 1291 | info64->lo_flags = info.lo_flags;
|
| 1292 | info64->lo_init[0] = info.lo_init[0];
|
| 1293 | info64->lo_init[1] = info.lo_init[1];
|
| 1294 | if (info.lo_encrypt_type == LO_CRYPT_CRYPTOAPI)
|
| 1295 | memcpy(info64->lo_crypt_name, info.lo_name, LO_NAME_SIZE);
|
| 1296 | else
|
| 1297 | memcpy(info64->lo_file_name, info.lo_name, LO_NAME_SIZE);
|
| 1298 | memcpy(info64->lo_encrypt_key, info.lo_encrypt_key, LO_KEY_SIZE);
|
| 1299 | return 0;
|
| 1300 | }
|
| 1301 |
|
| 1302 | |
| 1303 | * Transfer 64-bit loop info to 32-bit compatibility structure in userspace
|
| 1304 | * - noinlined to reduce stack space usage in main part of driver
|
| 1305 |
|
| 1306 | static noinline int
|
| 1307 | loop_info64_to_compat(const struct loop_info64 *info64,
|
| 1308 | struct compat_loop_info __user *arg)
|
| 1309 | {
|
| 1310 | struct compat_loop_info info;
|
| 1311 |
|
| 1312 | memset(&info, 0, sizeof(info));
|
| 1313 | info.lo_number = info64->lo_number;
|
| 1314 | info.lo_device = info64->lo_device;
|
| 1315 | info.lo_inode = info64->lo_inode;
|
| 1316 | info.lo_rdevice = info64->lo_rdevice;
|
| 1317 | info.lo_offset = info64->lo_offset;
|
| 1318 | info.lo_encrypt_type = info64->lo_encrypt_type;
|
| 1319 | info.lo_encrypt_key_size = info64->lo_encrypt_key_size;
|
| 1320 | info.lo_flags = info64->lo_flags;
|
| 1321 | info.lo_init[0] = info64->lo_init[0];
|
| 1322 | info.lo_init[1] = info64->lo_init[1];
|
| 1323 | if (info.lo_encrypt_type == LO_CRYPT_CRYPTOAPI)
|
| 1324 | memcpy(info.lo_name, info64->lo_crypt_name, LO_NAME_SIZE);
|
| 1325 | else
|
| 1326 | memcpy(info.lo_name, info64->lo_file_name, LO_NAME_SIZE);
|
| 1327 | memcpy(info.lo_encrypt_key, info64->lo_encrypt_key, LO_KEY_SIZE);
|
| 1328 |
|
| 1329 |
|
| 1330 | if (info.lo_device != info64->lo_device ||
|
| 1331 | info.lo_rdevice != info64->lo_rdevice ||
|
| 1332 | info.lo_inode != info64->lo_inode ||
|
| 1333 | info.lo_offset != info64->lo_offset ||
|
| 1334 | info.lo_init[0] != info64->lo_init[0] ||
|
| 1335 | info.lo_init[1] != info64->lo_init[1])
|
| 1336 | return -EOVERFLOW;
|
| 1337 |
|
| 1338 | if (copy_to_user(arg, &info, sizeof(info)))
|
| 1339 | return -EFAULT;
|
| 1340 | return 0;
|
| 1341 | }
|
| 1342 |
|
| 1343 | static int
|
| 1344 | loop_set_status_compat(struct loop_device *lo,
|
| 1345 | const struct compat_loop_info __user *arg)
|
| 1346 | {
|
| 1347 | struct loop_info64 info64;
|
| 1348 | int ret;
|
| 1349 |
|
| 1350 | ret = loop_info64_from_compat(arg, &info64);
|
| 1351 | if (ret < 0)
|
| 1352 | return ret;
|
| 1353 | return loop_set_status(lo, &info64);
|
| 1354 | }
|
| 1355 |
|
| 1356 | static int
|
| 1357 | loop_get_status_compat(struct loop_device *lo,
|
| 1358 | struct compat_loop_info __user *arg)
|
| 1359 | {
|
| 1360 | struct loop_info64 info64;
|
| 1361 | int err = 0;
|
| 1362 |
|
| 1363 | if (!arg)
|
| 1364 | err = -EINVAL;
|
| 1365 | if (!err)
|
| 1366 | err = loop_get_status(lo, &info64);
|
| 1367 | if (!err)
|
| 1368 | err = loop_info64_to_compat(&info64, arg);
|
| 1369 | return err;
|
| 1370 | }
|
| 1371 |
|
| 1372 | static int lo_compat_ioctl(struct block_device *bdev, fmode_t mode,
|
| 1373 | unsigned int cmd, unsigned long arg)
|
| 1374 | {
|
| 1375 | struct loop_device *lo = bdev->bd_disk->private_data;
|
| 1376 | int err;
|
| 1377 |
|
| 1378 | switch(cmd) {
|
| 1379 | case LOOP_SET_STATUS:
|
| 1380 | mutex_lock(&lo->lo_ctl_mutex);
|
| 1381 | err = loop_set_status_compat(
|
| 1382 | lo, (const struct compat_loop_info __user *) arg);
|
| 1383 | mutex_unlock(&lo->lo_ctl_mutex);
|
| 1384 | break;
|
| 1385 | case LOOP_GET_STATUS:
|
| 1386 | mutex_lock(&lo->lo_ctl_mutex);
|
| 1387 | err = loop_get_status_compat(
|
| 1388 | lo, (struct compat_loop_info __user *) arg);
|
| 1389 | mutex_unlock(&lo->lo_ctl_mutex);
|
| 1390 | break;
|
| 1391 | case LOOP_SET_CAPACITY:
|
| 1392 | case LOOP_CLR_FD:
|
| 1393 | case LOOP_GET_STATUS64:
|
| 1394 | case LOOP_SET_STATUS64:
|
| 1395 | arg = (unsigned long) compat_ptr(arg);
|
| 1396 | case LOOP_SET_FD:
|
| 1397 | case LOOP_CHANGE_FD:
|
| 1398 | err = lo_ioctl(bdev, mode, cmd, arg);
|
| 1399 | break;
|
| 1400 | default:
|
| 1401 | err = -ENOIOCTLCMD;
|
| 1402 | break;
|
| 1403 | }
|
| 1404 | return err;
|
| 1405 | }
|
| 1406 | #endif
|
| 1407 |
|
| 1408 | static int lo_open(struct block_device *bdev, fmode_t mode)
|
| 1409 | {
|
| 1410 | struct loop_device *lo = bdev->bd_disk->private_data;
|
| 1411 |
|
| 1412 | lock_kernel();
|
| 1413 | mutex_lock(&lo->lo_ctl_mutex);
|
| 1414 | lo->lo_refcnt++;
|
| 1415 | mutex_unlock(&lo->lo_ctl_mutex);
|
| 1416 | unlock_kernel();
|
| 1417 |
|
| 1418 | return 0;
|
| 1419 | }
|
| 1420 |
|
| 1421 | static int lo_release(struct gendisk *disk, fmode_t mode)
|
| 1422 | {
|
| 1423 | struct loop_device *lo = disk->private_data;
|
| 1424 | int err;
|
| 1425 |
|
| 1426 | lock_kernel();
|
| 1427 | mutex_lock(&lo->lo_ctl_mutex);
|
| 1428 |
|
| 1429 | if (--lo->lo_refcnt)
|
| 1430 | goto out;
|
| 1431 |
|
| 1432 | if (lo->lo_flags & LO_FLAGS_AUTOCLEAR) {
|
| 1433 | |
| 1434 | * In autoclear mode, stop the loop thread
|
| 1435 | * and remove configuration after last close.
|
| 1436 |
|
| 1437 | err = loop_clr_fd(lo, NULL);
|
| 1438 | if (!err)
|
| 1439 | goto out_unlocked;
|
| 1440 | } else {
|
| 1441 | |
| 1442 | * Otherwise keep thread (if running) and config,
|
| 1443 | * but flush possible ongoing bios in thread.
|
| 1444 |
|
| 1445 | loop_flush(lo);
|
| 1446 | }
|
| 1447 |
|
| 1448 | out:
|
| 1449 | mutex_unlock(&lo->lo_ctl_mutex);
|
| 1450 | out_unlocked:
|
| 1451 | lock_kernel();
|
| 1452 | return 0;
|
| 1453 | }
|
| 1454 |
|
| 1455 | static const struct block_device_operations lo_fops = {
|
| 1456 | .owner = THIS_MODULE,
|
| 1457 | .open = lo_open,
|
| 1458 | .release = lo_release,
|
| 1459 | .ioctl = lo_ioctl,
|
| 1460 | #ifdef CONFIG_COMPAT
|
| 1461 | .compat_ioctl = lo_compat_ioctl,
|
| 1462 | #endif
|
| 1463 | };
|
| 1464 |
|
| 1465 | |
| 1466 | * And now the modules code and kernel interface.
|
| 1467 |
|
| 1468 | static int max_loop;
|
| 1469 | module_param(max_loop, int, 0);
|
| 1470 | MODULE_PARM_DESC(max_loop, "Maximum number of loop devices");
|
| 1471 | module_param(max_part, int, 0);
|
| 1472 | MODULE_PARM_DESC(max_part, "Maximum number of partitions per loop device");
|
| 1473 | MODULE_LICENSE("GPL");
|
| 1474 | MODULE_ALIAS_BLOCKDEV_MAJOR(LOOP_MAJOR);
|
| 1475 |
|
| 1476 | int loop_register_transfer(struct loop_func_table *funcs)
|
| 1477 | {
|
| 1478 | unsigned int n = funcs->number;
|
| 1479 |
|
| 1480 | if (n >= MAX_LO_CRYPT || xfer_funcs[n])
|
| 1481 | return -EINVAL;
|
| 1482 | xfer_funcs[n] = funcs;
|
| 1483 | return 0;
|
| 1484 | }
|
| 1485 |
|
| 1486 | int loop_unregister_transfer(int number)
|
| 1487 | {
|
| 1488 | unsigned int n = number;
|
| 1489 | struct loop_device *lo;
|
| 1490 | struct loop_func_table *xfer;
|
| 1491 |
|
| 1492 | if (n == 0 || n >= MAX_LO_CRYPT || (xfer = xfer_funcs[n]) == NULL)
|
| 1493 | return -EINVAL;
|
| 1494 |
|
| 1495 | xfer_funcs[n] = NULL;
|
| 1496 |
|
| 1497 | list_for_each_entry(lo, &loop_devices, lo_list) {
|
| 1498 | mutex_lock(&lo->lo_ctl_mutex);
|
| 1499 |
|
| 1500 | if (lo->lo_encryption == xfer)
|
| 1501 | loop_release_xfer(lo);
|
| 1502 |
|
| 1503 | mutex_unlock(&lo->lo_ctl_mutex);
|
| 1504 | }
|
| 1505 |
|
| 1506 | return 0;
|
| 1507 | }
|
| 1508 |
|
| 1509 | EXPORT_SYMBOL(loop_register_transfer);
|
| 1510 | EXPORT_SYMBOL(loop_unregister_transfer);
|
| 1511 |
|
| 1512 | static struct loop_device *loop_alloc(int i)
|
| 1513 | {
|
| 1514 | struct loop_device *lo;
|
| 1515 | struct gendisk *disk;
|
| 1516 |
|
| 1517 | lo = kzalloc(sizeof(*lo), GFP_KERNEL);
|
| 1518 | if (!lo)
|
| 1519 | goto out;
|
| 1520 |
|
| 1521 | lo->lo_queue = blk_alloc_queue(GFP_KERNEL);
|
| 1522 | if (!lo->lo_queue)
|
| 1523 | goto out_free_dev;
|
| 1524 |
|
| 1525 | disk = lo->lo_disk = alloc_disk(1 << part_shift);
|
| 1526 | if (!disk)
|
| 1527 | goto out_free_queue;
|
| 1528 |
|
| 1529 | mutex_init(&lo->lo_ctl_mutex);
|
| 1530 | lo->lo_number = i;
|
| 1531 | lo->lo_thread = NULL;
|
| 1532 | init_waitqueue_head(&lo->lo_event);
|
| 1533 | spin_lock_init(&lo->lo_lock);
|
| 1534 | disk->major = LOOP_MAJOR;
|
| 1535 | disk->first_minor = i << part_shift;
|
| 1536 | disk->fops = &lo_fops;
|
| 1537 | disk->private_data = lo;
|
| 1538 | disk->queue = lo->lo_queue;
|
| 1539 | sprintf(disk->disk_name, "loop%d", i);
|
| 1540 | return lo;
|
| 1541 |
|
| 1542 | out_free_queue:
|
| 1543 | blk_cleanup_queue(lo->lo_queue);
|
| 1544 | out_free_dev:
|
| 1545 | kfree(lo);
|
| 1546 | out:
|
| 1547 | return NULL;
|
| 1548 | }
|
| 1549 |
|
| 1550 | static void loop_free(struct loop_device *lo)
|
| 1551 | {
|
| 1552 | blk_cleanup_queue(lo->lo_queue);
|
| 1553 | put_disk(lo->lo_disk);
|
| 1554 | list_del(&lo->lo_list);
|
| 1555 | kfree(lo);
|
| 1556 | }
|
| 1557 |
|
| 1558 | static struct loop_device *loop_init_one(int i)
|
| 1559 | {
|
| 1560 | struct loop_device *lo;
|
| 1561 |
|
| 1562 | list_for_each_entry(lo, &loop_devices, lo_list) {
|
| 1563 | if (lo->lo_number == i)
|
| 1564 | return lo;
|
| 1565 | }
|
| 1566 |
|
| 1567 | lo = loop_alloc(i);
|
| 1568 | if (lo) {
|
| 1569 | add_disk(lo->lo_disk);
|
| 1570 | list_add_tail(&lo->lo_list, &loop_devices);
|
| 1571 | }
|
| 1572 | return lo;
|
| 1573 | }
|
| 1574 |
|
| 1575 | static void loop_del_one(struct loop_device *lo)
|
| 1576 | {
|
| 1577 | del_gendisk(lo->lo_disk);
|
| 1578 | loop_free(lo);
|
| 1579 | }
|
| 1580 |
|
| 1581 | static struct kobject *loop_probe(dev_t dev, int *part, void *data)
|
| 1582 | {
|
| 1583 | struct loop_device *lo;
|
| 1584 | struct kobject *kobj;
|
| 1585 |
|
| 1586 | mutex_lock(&loop_devices_mutex);
|
| 1587 | lo = loop_init_one(dev & MINORMASK);
|
| 1588 | kobj = lo ? get_disk(lo->lo_disk) : ERR_PTR(-ENOMEM);
|
| 1589 | mutex_unlock(&loop_devices_mutex);
|
| 1590 |
|
| 1591 | *part = 0;
|
| 1592 | return kobj;
|
| 1593 | }
|
| 1594 |
|
| 1595 | static int __init loop_init(void)
|
| 1596 | {
|
| 1597 | int i, nr;
|
| 1598 | unsigned long range;
|
| 1599 | struct loop_device *lo, *next;
|
| 1600 |
|
| 1601 | |
| 1602 | * loop module now has a feature to instantiate underlying device
|
| 1603 | * structure on-demand, provided that there is an access dev node.
|
| 1604 | * However, this will not work well with user space tool that doesn't
|
| 1605 | * know about such "feature". In order to not break any existing
|
| 1606 | * tool, we do the following:
|
| 1607 | *
|
| 1608 | * (1) if max_loop is specified, create that many upfront, and this
|
| 1609 | * also becomes a hard limit.
|
| 1610 | * (2) if max_loop is not specified, create 8 loop device on module
|
| 1611 | * load, user can further extend loop device by create dev node
|
| 1612 | * themselves and have kernel automatically instantiate actual
|
| 1613 | * device on-demand.
|
| 1614 |
|
| 1615 |
|
| 1616 | part_shift = 0;
|
| 1617 | if (max_part > 0)
|
| 1618 | part_shift = fls(max_part);
|
| 1619 |
|
| 1620 | if (max_loop > 1UL << (MINORBITS - part_shift))
|
| 1621 | return -EINVAL;
|
| 1622 |
|
| 1623 | if (max_loop) {
|
| 1624 | nr = max_loop;
|
| 1625 | range = max_loop;
|
| 1626 | } else {
|
| 1627 | nr = 8;
|
| 1628 | range = 1UL << (MINORBITS - part_shift);
|
| 1629 | }
|
| 1630 |
|
| 1631 | if (register_blkdev(LOOP_MAJOR, "loop"))
|
| 1632 | return -EIO;
|
| 1633 |
|
| 1634 | for (i = 0; i < nr; i++) {
|
| 1635 | lo = loop_alloc(i);
|
| 1636 | if (!lo)
|
| 1637 | goto Enomem;
|
| 1638 | list_add_tail(&lo->lo_list, &loop_devices);
|
| 1639 | }
|
| 1640 |
|
| 1641 |
|
| 1642 |
|
| 1643 | list_for_each_entry(lo, &loop_devices, lo_list)
|
| 1644 | add_disk(lo->lo_disk);
|
| 1645 |
|
| 1646 | blk_register_region(MKDEV(LOOP_MAJOR, 0), range,
|
| 1647 | THIS_MODULE, loop_probe, NULL, NULL);
|
| 1648 |
|
| 1649 | printk(KERN_INFO "loop: module loaded\n");
|
| 1650 | return 0;
|
| 1651 |
|
| 1652 | Enomem:
|
| 1653 | printk(KERN_INFO "loop: out of memory\n");
|
| 1654 |
|
| 1655 | list_for_each_entry_safe(lo, next, &loop_devices, lo_list)
|
| 1656 | loop_free(lo);
|
| 1657 |
|
| 1658 | unregister_blkdev(LOOP_MAJOR, "loop");
|
| 1659 | return -ENOMEM;
|
| 1660 | }
|
| 1661 |
|
| 1662 | static void __exit loop_exit(void)
|
| 1663 | {
|
| 1664 | unsigned long range;
|
| 1665 | struct loop_device *lo, *next;
|
| 1666 |
|
| 1667 | range = max_loop ? max_loop : 1UL << (MINORBITS - part_shift);
|
| 1668 |
|
| 1669 | list_for_each_entry_safe(lo, next, &loop_devices, lo_list)
|
| 1670 | loop_del_one(lo);
|
| 1671 |
|
| 1672 | blk_unregister_region(MKDEV(LOOP_MAJOR, 0), range);
|
| 1673 | unregister_blkdev(LOOP_MAJOR, "loop");
|
| 1674 | }
|
| 1675 |
|
| 1676 | module_init(loop_init);
|
| 1677 | module_exit(loop_exit);
|
| 1678 |
|
| 1679 | #ifndef MODULE
|
| 1680 | static int __init max_loop_setup(char *str)
|
| 1681 | {
|
| 1682 | max_loop = simple_strtol(str, NULL, 0);
|
| 1683 | return 1;
|
| 1684 | }
|
| 1685 |
|
| 1686 | __setup("max_loop=", max_loop_setup);
|
| 1687 | #endif
|