| 1 | /* |
| 2 | * GPL HEADER START |
| 3 | * |
| 4 | * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER. |
| 5 | * |
| 6 | * This program is free software; you can redistribute it and/or modify |
| 7 | * it under the terms of the GNU General Public License version 2 only, |
| 8 | * as published by the Free Software Foundation. |
| 9 | * |
| 10 | * This program is distributed in the hope that it will be useful, but |
| 11 | * WITHOUT ANY WARRANTY; without even the implied warranty of |
| 12 | * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU |
| 13 | * General Public License version 2 for more details (a copy is included |
| 14 | * in the LICENSE file that accompanied this code). |
| 15 | * |
| 16 | * You should have received a copy of the GNU General Public License |
| 17 | * version 2 along with this program; If not, see |
| 18 | * http://www.sun.com/software/products/lustre/docs/GPLv2.pdf |
| 19 | * |
| 20 | * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara, |
| 21 | * CA 95054 USA or visit www.sun.com if you need additional information or |
| 22 | * have any questions. |
| 23 | * |
| 24 | * GPL HEADER END |
| 25 | */ |
| 26 | /* |
| 27 | * Copyright (c) 2007, 2010, Oracle and/or its affiliates. All rights reserved. |
| 28 | * Use is subject to license terms. |
| 29 | * |
| 30 | * Copyright (c) 2011, 2012, Intel Corporation. |
| 31 | */ |
| 32 | /* |
| 33 | * This file is part of Lustre, http://www.lustre.org/ |
| 34 | * Lustre is a trademark of Sun Microsystems, Inc. |
| 35 | */ |
| 36 | |
| 37 | /* |
| 38 | * linux/drivers/block/loop.c |
| 39 | * |
| 40 | * Written by Theodore Ts'o, 3/29/93 |
| 41 | * |
| 42 | * Copyright 1993 by Theodore Ts'o. Redistribution of this file is |
| 43 | * permitted under the GNU General Public License. |
| 44 | * |
| 45 | * Modularized and updated for 1.1.16 kernel - Mitch Dsouza 28th May 1994 |
| 46 | * Adapted for 1.3.59 kernel - Andries Brouwer, 1 Feb 1996 |
| 47 | * |
| 48 | * Fixed do_loop_request() re-entrancy - Vincent.Renardias@waw.com Mar 20, 1997 |
| 49 | * |
| 50 | * Added devfs support - Richard Gooch <rgooch@atnf.csiro.au> 16-Jan-1998 |
| 51 | * |
| 52 | * Handle sparse backing files correctly - Kenn Humborg, Jun 28, 1998 |
| 53 | * |
| 54 | * Loadable modules and other fixes by AK, 1998 |
| 55 | * |
| 56 | * Maximum number of loop devices now dynamic via max_loop module parameter. |
| 57 | * Russell Kroll <rkroll@exploits.org> 19990701 |
| 58 | * |
| 59 | * Maximum number of loop devices when compiled-in now selectable by passing |
| 60 | * max_loop=<1-255> to the kernel on boot. |
| 61 | * Erik I. Bols?, <eriki@himolde.no>, Oct 31, 1999 |
| 62 | * |
| 63 | * Completely rewrite request handling to be make_request_fn style and |
| 64 | * non blocking, pushing work to a helper thread. Lots of fixes from |
| 65 | * Al Viro too. |
| 66 | * Jens Axboe <axboe@suse.de>, Nov 2000 |
| 67 | * |
| 68 | * Support up to 256 loop devices |
| 69 | * Heinz Mauelshagen <mge@sistina.com>, Feb 2002 |
| 70 | * |
| 71 | * Support for falling back on the write file operation when the address space |
| 72 | * operations prepare_write and/or commit_write are not available on the |
| 73 | * backing filesystem. |
| 74 | * Anton Altaparmakov, 16 Feb 2005 |
| 75 | * |
| 76 | * Still To Fix: |
| 77 | * - Advisory locking is ignored here. |
| 78 | * - Should use an own CAP_* category instead of CAP_SYS_ADMIN |
| 79 | * |
| 80 | */ |
| 81 | |
| 82 | #include <linux/module.h> |
| 83 | |
| 84 | #include <linux/sched.h> |
| 85 | #include <linux/fs.h> |
| 86 | #include <linux/file.h> |
| 87 | #include <linux/stat.h> |
| 88 | #include <linux/errno.h> |
| 89 | #include <linux/major.h> |
| 90 | #include <linux/wait.h> |
| 91 | #include <linux/blkdev.h> |
| 92 | #include <linux/blkpg.h> |
| 93 | #include <linux/init.h> |
| 94 | #include <linux/swap.h> |
| 95 | #include <linux/slab.h> |
| 96 | #include <linux/suspend.h> |
| 97 | #include <linux/writeback.h> |
| 98 | #include <linux/buffer_head.h> /* for invalidate_bdev() */ |
| 99 | #include <linux/completion.h> |
| 100 | #include <linux/highmem.h> |
| 101 | #include <linux/gfp.h> |
| 102 | #include <linux/pagevec.h> |
| 103 | |
| 104 | #include <asm/uaccess.h> |
| 105 | |
| 106 | #include <lustre_lib.h> |
| 107 | #include <lustre_lite.h> |
| 108 | #include "llite_internal.h" |
| 109 | |
| 110 | #define LLOOP_MAX_SEGMENTS LNET_MAX_IOV |
| 111 | |
| 112 | /* Possible states of device */ |
| 113 | enum { |
| 114 | LLOOP_UNBOUND, |
| 115 | LLOOP_BOUND, |
| 116 | LLOOP_RUNDOWN, |
| 117 | }; |
| 118 | |
| 119 | struct lloop_device { |
| 120 | int lo_number; |
| 121 | int lo_refcnt; |
| 122 | loff_t lo_offset; |
| 123 | loff_t lo_sizelimit; |
| 124 | int lo_flags; |
| 125 | int (*ioctl)(struct lloop_device *, int cmd, |
| 126 | unsigned long arg); |
| 127 | |
| 128 | struct file *lo_backing_file; |
| 129 | struct block_device *lo_device; |
| 130 | unsigned lo_blocksize; |
| 131 | |
| 132 | int old_gfp_mask; |
| 133 | |
| 134 | spinlock_t lo_lock; |
| 135 | struct bio *lo_bio; |
| 136 | struct bio *lo_biotail; |
| 137 | int lo_state; |
| 138 | struct semaphore lo_sem; |
| 139 | struct mutex lo_ctl_mutex; |
| 140 | atomic_t lo_pending; |
| 141 | wait_queue_head_t lo_bh_wait; |
| 142 | |
| 143 | struct request_queue *lo_queue; |
| 144 | |
| 145 | const struct lu_env *lo_env; |
| 146 | struct cl_io lo_io; |
| 147 | struct ll_dio_pages lo_pvec; |
| 148 | |
| 149 | /* data to handle bio for lustre. */ |
| 150 | struct lo_request_data { |
| 151 | struct page *lrd_pages[LLOOP_MAX_SEGMENTS]; |
| 152 | loff_t lrd_offsets[LLOOP_MAX_SEGMENTS]; |
| 153 | } lo_requests[1]; |
| 154 | }; |
| 155 | |
| 156 | /* |
| 157 | * Loop flags |
| 158 | */ |
| 159 | enum { |
| 160 | LO_FLAGS_READ_ONLY = 1, |
| 161 | }; |
| 162 | |
| 163 | static int lloop_major; |
| 164 | #define MAX_LOOP_DEFAULT 16 |
| 165 | static int max_loop = MAX_LOOP_DEFAULT; |
| 166 | static struct lloop_device *loop_dev; |
| 167 | static struct gendisk **disks; |
| 168 | static struct mutex lloop_mutex; |
| 169 | static void *ll_iocontrol_magic = NULL; |
| 170 | |
| 171 | static loff_t get_loop_size(struct lloop_device *lo, struct file *file) |
| 172 | { |
| 173 | loff_t size, offset, loopsize; |
| 174 | |
| 175 | /* Compute loopsize in bytes */ |
| 176 | size = i_size_read(file->f_mapping->host); |
| 177 | offset = lo->lo_offset; |
| 178 | loopsize = size - offset; |
| 179 | if (lo->lo_sizelimit > 0 && lo->lo_sizelimit < loopsize) |
| 180 | loopsize = lo->lo_sizelimit; |
| 181 | |
| 182 | /* |
| 183 | * Unfortunately, if we want to do I/O on the device, |
| 184 | * the number of 512-byte sectors has to fit into a sector_t. |
| 185 | */ |
| 186 | return loopsize >> 9; |
| 187 | } |
| 188 | |
| 189 | static int do_bio_lustrebacked(struct lloop_device *lo, struct bio *head) |
| 190 | { |
| 191 | const struct lu_env *env = lo->lo_env; |
| 192 | struct cl_io *io = &lo->lo_io; |
| 193 | struct inode *inode = lo->lo_backing_file->f_dentry->d_inode; |
| 194 | struct cl_object *obj = ll_i2info(inode)->lli_clob; |
| 195 | pgoff_t offset; |
| 196 | int ret; |
| 197 | int rw; |
| 198 | obd_count page_count = 0; |
| 199 | struct bio_vec bvec; |
| 200 | struct bvec_iter iter; |
| 201 | struct bio *bio; |
| 202 | ssize_t bytes; |
| 203 | |
| 204 | struct ll_dio_pages *pvec = &lo->lo_pvec; |
| 205 | struct page **pages = pvec->ldp_pages; |
| 206 | loff_t *offsets = pvec->ldp_offsets; |
| 207 | |
| 208 | truncate_inode_pages(inode->i_mapping, 0); |
| 209 | |
| 210 | /* initialize the IO */ |
| 211 | memset(io, 0, sizeof(*io)); |
| 212 | io->ci_obj = obj; |
| 213 | ret = cl_io_init(env, io, CIT_MISC, obj); |
| 214 | if (ret) |
| 215 | return io->ci_result; |
| 216 | io->ci_lockreq = CILR_NEVER; |
| 217 | |
| 218 | LASSERT(head != NULL); |
| 219 | rw = head->bi_rw; |
| 220 | for (bio = head; bio != NULL; bio = bio->bi_next) { |
| 221 | LASSERT(rw == bio->bi_rw); |
| 222 | |
| 223 | offset = (pgoff_t)(bio->bi_iter.bi_sector << 9) + lo->lo_offset; |
| 224 | bio_for_each_segment(bvec, bio, iter) { |
| 225 | BUG_ON(bvec.bv_offset != 0); |
| 226 | BUG_ON(bvec.bv_len != PAGE_CACHE_SIZE); |
| 227 | |
| 228 | pages[page_count] = bvec.bv_page; |
| 229 | offsets[page_count] = offset; |
| 230 | page_count++; |
| 231 | offset += bvec.bv_len; |
| 232 | } |
| 233 | LASSERT(page_count <= LLOOP_MAX_SEGMENTS); |
| 234 | } |
| 235 | |
| 236 | ll_stats_ops_tally(ll_i2sbi(inode), |
| 237 | (rw == WRITE) ? LPROC_LL_BRW_WRITE : LPROC_LL_BRW_READ, |
| 238 | page_count); |
| 239 | |
| 240 | pvec->ldp_size = page_count << PAGE_CACHE_SHIFT; |
| 241 | pvec->ldp_nr = page_count; |
| 242 | |
| 243 | /* FIXME: in ll_direct_rw_pages, it has to allocate many cl_page{}s to |
| 244 | * write those pages into OST. Even worse case is that more pages |
| 245 | * would be asked to write out to swap space, and then finally get here |
| 246 | * again. |
| 247 | * Unfortunately this is NOT easy to fix. |
| 248 | * Thoughts on solution: |
| 249 | * 0. Define a reserved pool for cl_pages, which could be a list of |
| 250 | * pre-allocated cl_pages; |
| 251 | * 1. Define a new operation in cl_object_operations{}, says clo_depth, |
| 252 | * which measures how many layers for this lustre object. Generally |
| 253 | * speaking, the depth would be 2, one for llite, and one for lovsub. |
| 254 | * However, for SNS, there will be more since we need additional page |
| 255 | * to store parity; |
| 256 | * 2. Reserve the # of (page_count * depth) cl_pages from the reserved |
| 257 | * pool. Afterwards, the clio would allocate the pages from reserved |
| 258 | * pool, this guarantees we needn't allocate the cl_pages from |
| 259 | * generic cl_page slab cache. |
| 260 | * Of course, if there is NOT enough pages in the pool, we might |
| 261 | * be asked to write less pages once, this purely depends on |
| 262 | * implementation. Anyway, we should be careful to avoid deadlocking. |
| 263 | */ |
| 264 | mutex_lock(&inode->i_mutex); |
| 265 | bytes = ll_direct_rw_pages(env, io, rw, inode, pvec); |
| 266 | mutex_unlock(&inode->i_mutex); |
| 267 | cl_io_fini(env, io); |
| 268 | return (bytes == pvec->ldp_size) ? 0 : (int)bytes; |
| 269 | } |
| 270 | |
| 271 | /* |
| 272 | * Add bio to back of pending list |
| 273 | */ |
| 274 | static void loop_add_bio(struct lloop_device *lo, struct bio *bio) |
| 275 | { |
| 276 | unsigned long flags; |
| 277 | |
| 278 | spin_lock_irqsave(&lo->lo_lock, flags); |
| 279 | if (lo->lo_biotail) { |
| 280 | lo->lo_biotail->bi_next = bio; |
| 281 | lo->lo_biotail = bio; |
| 282 | } else |
| 283 | lo->lo_bio = lo->lo_biotail = bio; |
| 284 | spin_unlock_irqrestore(&lo->lo_lock, flags); |
| 285 | |
| 286 | atomic_inc(&lo->lo_pending); |
| 287 | if (waitqueue_active(&lo->lo_bh_wait)) |
| 288 | wake_up(&lo->lo_bh_wait); |
| 289 | } |
| 290 | |
| 291 | /* |
| 292 | * Grab first pending buffer |
| 293 | */ |
| 294 | static unsigned int loop_get_bio(struct lloop_device *lo, struct bio **req) |
| 295 | { |
| 296 | struct bio *first; |
| 297 | struct bio **bio; |
| 298 | unsigned int count = 0; |
| 299 | unsigned int page_count = 0; |
| 300 | int rw; |
| 301 | |
| 302 | spin_lock_irq(&lo->lo_lock); |
| 303 | first = lo->lo_bio; |
| 304 | if (unlikely(first == NULL)) { |
| 305 | spin_unlock_irq(&lo->lo_lock); |
| 306 | return 0; |
| 307 | } |
| 308 | |
| 309 | /* TODO: need to split the bio, too bad. */ |
| 310 | LASSERT(first->bi_vcnt <= LLOOP_MAX_SEGMENTS); |
| 311 | |
| 312 | rw = first->bi_rw; |
| 313 | bio = &lo->lo_bio; |
| 314 | while (*bio && (*bio)->bi_rw == rw) { |
| 315 | CDEBUG(D_INFO, "bio sector %llu size %u count %u vcnt%u \n", |
| 316 | (unsigned long long)(*bio)->bi_iter.bi_sector, |
| 317 | (*bio)->bi_iter.bi_size, |
| 318 | page_count, (*bio)->bi_vcnt); |
| 319 | if (page_count + (*bio)->bi_vcnt > LLOOP_MAX_SEGMENTS) |
| 320 | break; |
| 321 | |
| 322 | |
| 323 | page_count += (*bio)->bi_vcnt; |
| 324 | count++; |
| 325 | bio = &(*bio)->bi_next; |
| 326 | } |
| 327 | if (*bio) { |
| 328 | /* Some of bios can't be mergeable. */ |
| 329 | lo->lo_bio = *bio; |
| 330 | *bio = NULL; |
| 331 | } else { |
| 332 | /* Hit the end of queue */ |
| 333 | lo->lo_biotail = NULL; |
| 334 | lo->lo_bio = NULL; |
| 335 | } |
| 336 | *req = first; |
| 337 | spin_unlock_irq(&lo->lo_lock); |
| 338 | return count; |
| 339 | } |
| 340 | |
| 341 | static void loop_make_request(struct request_queue *q, struct bio *old_bio) |
| 342 | { |
| 343 | struct lloop_device *lo = q->queuedata; |
| 344 | int rw = bio_rw(old_bio); |
| 345 | int inactive; |
| 346 | |
| 347 | if (!lo) |
| 348 | goto err; |
| 349 | |
| 350 | CDEBUG(D_INFO, "submit bio sector %llu size %u\n", |
| 351 | (unsigned long long)old_bio->bi_iter.bi_sector, |
| 352 | old_bio->bi_iter.bi_size); |
| 353 | |
| 354 | spin_lock_irq(&lo->lo_lock); |
| 355 | inactive = (lo->lo_state != LLOOP_BOUND); |
| 356 | spin_unlock_irq(&lo->lo_lock); |
| 357 | if (inactive) |
| 358 | goto err; |
| 359 | |
| 360 | if (rw == WRITE) { |
| 361 | if (lo->lo_flags & LO_FLAGS_READ_ONLY) |
| 362 | goto err; |
| 363 | } else if (rw == READA) { |
| 364 | rw = READ; |
| 365 | } else if (rw != READ) { |
| 366 | CERROR("lloop: unknown command (%x)\n", rw); |
| 367 | goto err; |
| 368 | } |
| 369 | loop_add_bio(lo, old_bio); |
| 370 | return; |
| 371 | err: |
| 372 | cfs_bio_io_error(old_bio, old_bio->bi_iter.bi_size); |
| 373 | } |
| 374 | |
| 375 | |
| 376 | static inline void loop_handle_bio(struct lloop_device *lo, struct bio *bio) |
| 377 | { |
| 378 | int ret; |
| 379 | ret = do_bio_lustrebacked(lo, bio); |
| 380 | while (bio) { |
| 381 | struct bio *tmp = bio->bi_next; |
| 382 | bio->bi_next = NULL; |
| 383 | cfs_bio_endio(bio, bio->bi_iter.bi_size, ret); |
| 384 | bio = tmp; |
| 385 | } |
| 386 | } |
| 387 | |
| 388 | static inline int loop_active(struct lloop_device *lo) |
| 389 | { |
| 390 | return atomic_read(&lo->lo_pending) || |
| 391 | (lo->lo_state == LLOOP_RUNDOWN); |
| 392 | } |
| 393 | |
| 394 | /* |
| 395 | * worker thread that handles reads/writes to file backed loop devices, |
| 396 | * to avoid blocking in our make_request_fn. |
| 397 | */ |
| 398 | static int loop_thread(void *data) |
| 399 | { |
| 400 | struct lloop_device *lo = data; |
| 401 | struct bio *bio; |
| 402 | unsigned int count; |
| 403 | unsigned long times = 0; |
| 404 | unsigned long total_count = 0; |
| 405 | |
| 406 | struct lu_env *env; |
| 407 | int refcheck; |
| 408 | int ret = 0; |
| 409 | |
| 410 | set_user_nice(current, -20); |
| 411 | |
| 412 | lo->lo_state = LLOOP_BOUND; |
| 413 | |
| 414 | env = cl_env_get(&refcheck); |
| 415 | if (IS_ERR(env)) |
| 416 | GOTO(out, ret = PTR_ERR(env)); |
| 417 | |
| 418 | lo->lo_env = env; |
| 419 | memset(&lo->lo_pvec, 0, sizeof(lo->lo_pvec)); |
| 420 | lo->lo_pvec.ldp_pages = lo->lo_requests[0].lrd_pages; |
| 421 | lo->lo_pvec.ldp_offsets = lo->lo_requests[0].lrd_offsets; |
| 422 | |
| 423 | /* |
| 424 | * up sem, we are running |
| 425 | */ |
| 426 | up(&lo->lo_sem); |
| 427 | |
| 428 | for (;;) { |
| 429 | wait_event(lo->lo_bh_wait, loop_active(lo)); |
| 430 | if (!atomic_read(&lo->lo_pending)) { |
| 431 | int exiting = 0; |
| 432 | spin_lock_irq(&lo->lo_lock); |
| 433 | exiting = (lo->lo_state == LLOOP_RUNDOWN); |
| 434 | spin_unlock_irq(&lo->lo_lock); |
| 435 | if (exiting) |
| 436 | break; |
| 437 | } |
| 438 | |
| 439 | bio = NULL; |
| 440 | count = loop_get_bio(lo, &bio); |
| 441 | if (!count) { |
| 442 | CWARN("lloop(minor: %d): missing bio\n", lo->lo_number); |
| 443 | continue; |
| 444 | } |
| 445 | |
| 446 | total_count += count; |
| 447 | if (total_count < count) { /* overflow */ |
| 448 | total_count = count; |
| 449 | times = 1; |
| 450 | } else { |
| 451 | times++; |
| 452 | } |
| 453 | if ((times & 127) == 0) { |
| 454 | CDEBUG(D_INFO, "total: %lu, count: %lu, avg: %lu\n", |
| 455 | total_count, times, total_count / times); |
| 456 | } |
| 457 | |
| 458 | LASSERT(bio != NULL); |
| 459 | LASSERT(count <= atomic_read(&lo->lo_pending)); |
| 460 | loop_handle_bio(lo, bio); |
| 461 | atomic_sub(count, &lo->lo_pending); |
| 462 | } |
| 463 | cl_env_put(env, &refcheck); |
| 464 | |
| 465 | out: |
| 466 | up(&lo->lo_sem); |
| 467 | return ret; |
| 468 | } |
| 469 | |
| 470 | static int loop_set_fd(struct lloop_device *lo, struct file *unused, |
| 471 | struct block_device *bdev, struct file *file) |
| 472 | { |
| 473 | struct inode *inode; |
| 474 | struct address_space *mapping; |
| 475 | int lo_flags = 0; |
| 476 | int error; |
| 477 | loff_t size; |
| 478 | |
| 479 | if (!try_module_get(THIS_MODULE)) |
| 480 | return -ENODEV; |
| 481 | |
| 482 | error = -EBUSY; |
| 483 | if (lo->lo_state != LLOOP_UNBOUND) |
| 484 | goto out; |
| 485 | |
| 486 | mapping = file->f_mapping; |
| 487 | inode = mapping->host; |
| 488 | |
| 489 | error = -EINVAL; |
| 490 | if (!S_ISREG(inode->i_mode) || inode->i_sb->s_magic != LL_SUPER_MAGIC) |
| 491 | goto out; |
| 492 | |
| 493 | if (!(file->f_mode & FMODE_WRITE)) |
| 494 | lo_flags |= LO_FLAGS_READ_ONLY; |
| 495 | |
| 496 | size = get_loop_size(lo, file); |
| 497 | |
| 498 | if ((loff_t)(sector_t)size != size) { |
| 499 | error = -EFBIG; |
| 500 | goto out; |
| 501 | } |
| 502 | |
| 503 | /* remove all pages in cache so as dirty pages not to be existent. */ |
| 504 | truncate_inode_pages(mapping, 0); |
| 505 | |
| 506 | set_device_ro(bdev, (lo_flags & LO_FLAGS_READ_ONLY) != 0); |
| 507 | |
| 508 | lo->lo_blocksize = PAGE_CACHE_SIZE; |
| 509 | lo->lo_device = bdev; |
| 510 | lo->lo_flags = lo_flags; |
| 511 | lo->lo_backing_file = file; |
| 512 | lo->ioctl = NULL; |
| 513 | lo->lo_sizelimit = 0; |
| 514 | lo->old_gfp_mask = mapping_gfp_mask(mapping); |
| 515 | mapping_set_gfp_mask(mapping, lo->old_gfp_mask & ~(__GFP_IO|__GFP_FS)); |
| 516 | |
| 517 | lo->lo_bio = lo->lo_biotail = NULL; |
| 518 | |
| 519 | /* |
| 520 | * set queue make_request_fn, and add limits based on lower level |
| 521 | * device |
| 522 | */ |
| 523 | blk_queue_make_request(lo->lo_queue, loop_make_request); |
| 524 | lo->lo_queue->queuedata = lo; |
| 525 | |
| 526 | /* queue parameters */ |
| 527 | CLASSERT(PAGE_CACHE_SIZE < (1 << (sizeof(unsigned short) * 8))); |
| 528 | blk_queue_logical_block_size(lo->lo_queue, |
| 529 | (unsigned short)PAGE_CACHE_SIZE); |
| 530 | blk_queue_max_hw_sectors(lo->lo_queue, |
| 531 | LLOOP_MAX_SEGMENTS << (PAGE_CACHE_SHIFT - 9)); |
| 532 | blk_queue_max_segments(lo->lo_queue, LLOOP_MAX_SEGMENTS); |
| 533 | |
| 534 | set_capacity(disks[lo->lo_number], size); |
| 535 | bd_set_size(bdev, size << 9); |
| 536 | |
| 537 | set_blocksize(bdev, lo->lo_blocksize); |
| 538 | |
| 539 | kthread_run(loop_thread, lo, "lloop%d", lo->lo_number); |
| 540 | down(&lo->lo_sem); |
| 541 | return 0; |
| 542 | |
| 543 | out: |
| 544 | /* This is safe: open() is still holding a reference. */ |
| 545 | module_put(THIS_MODULE); |
| 546 | return error; |
| 547 | } |
| 548 | |
| 549 | static int loop_clr_fd(struct lloop_device *lo, struct block_device *bdev, |
| 550 | int count) |
| 551 | { |
| 552 | struct file *filp = lo->lo_backing_file; |
| 553 | int gfp = lo->old_gfp_mask; |
| 554 | |
| 555 | if (lo->lo_state != LLOOP_BOUND) |
| 556 | return -ENXIO; |
| 557 | |
| 558 | if (lo->lo_refcnt > count) /* we needed one fd for the ioctl */ |
| 559 | return -EBUSY; |
| 560 | |
| 561 | if (filp == NULL) |
| 562 | return -EINVAL; |
| 563 | |
| 564 | spin_lock_irq(&lo->lo_lock); |
| 565 | lo->lo_state = LLOOP_RUNDOWN; |
| 566 | spin_unlock_irq(&lo->lo_lock); |
| 567 | wake_up(&lo->lo_bh_wait); |
| 568 | |
| 569 | down(&lo->lo_sem); |
| 570 | lo->lo_backing_file = NULL; |
| 571 | lo->ioctl = NULL; |
| 572 | lo->lo_device = NULL; |
| 573 | lo->lo_offset = 0; |
| 574 | lo->lo_sizelimit = 0; |
| 575 | lo->lo_flags = 0; |
| 576 | invalidate_bdev(bdev); |
| 577 | set_capacity(disks[lo->lo_number], 0); |
| 578 | bd_set_size(bdev, 0); |
| 579 | mapping_set_gfp_mask(filp->f_mapping, gfp); |
| 580 | lo->lo_state = LLOOP_UNBOUND; |
| 581 | fput(filp); |
| 582 | /* This is safe: open() is still holding a reference. */ |
| 583 | module_put(THIS_MODULE); |
| 584 | return 0; |
| 585 | } |
| 586 | |
| 587 | static int lo_open(struct block_device *bdev, fmode_t mode) |
| 588 | { |
| 589 | struct lloop_device *lo = bdev->bd_disk->private_data; |
| 590 | |
| 591 | mutex_lock(&lo->lo_ctl_mutex); |
| 592 | lo->lo_refcnt++; |
| 593 | mutex_unlock(&lo->lo_ctl_mutex); |
| 594 | |
| 595 | return 0; |
| 596 | } |
| 597 | |
| 598 | static void lo_release(struct gendisk *disk, fmode_t mode) |
| 599 | { |
| 600 | struct lloop_device *lo = disk->private_data; |
| 601 | |
| 602 | mutex_lock(&lo->lo_ctl_mutex); |
| 603 | --lo->lo_refcnt; |
| 604 | mutex_unlock(&lo->lo_ctl_mutex); |
| 605 | } |
| 606 | |
| 607 | /* lloop device node's ioctl function. */ |
| 608 | static int lo_ioctl(struct block_device *bdev, fmode_t mode, |
| 609 | unsigned int cmd, unsigned long arg) |
| 610 | { |
| 611 | struct lloop_device *lo = bdev->bd_disk->private_data; |
| 612 | struct inode *inode = NULL; |
| 613 | int err = 0; |
| 614 | |
| 615 | mutex_lock(&lloop_mutex); |
| 616 | switch (cmd) { |
| 617 | case LL_IOC_LLOOP_DETACH: { |
| 618 | err = loop_clr_fd(lo, bdev, 2); |
| 619 | if (err == 0) |
| 620 | blkdev_put(bdev, 0); /* grabbed in LLOOP_ATTACH */ |
| 621 | break; |
| 622 | } |
| 623 | |
| 624 | case LL_IOC_LLOOP_INFO: { |
| 625 | struct lu_fid fid; |
| 626 | |
| 627 | LASSERT(lo->lo_backing_file != NULL); |
| 628 | if (inode == NULL) |
| 629 | inode = lo->lo_backing_file->f_dentry->d_inode; |
| 630 | if (lo->lo_state == LLOOP_BOUND) |
| 631 | fid = ll_i2info(inode)->lli_fid; |
| 632 | else |
| 633 | fid_zero(&fid); |
| 634 | |
| 635 | if (copy_to_user((struct lu_fid *)arg, &fid, sizeof(fid))) |
| 636 | err = -EFAULT; |
| 637 | break; |
| 638 | } |
| 639 | |
| 640 | default: |
| 641 | err = -EINVAL; |
| 642 | break; |
| 643 | } |
| 644 | mutex_unlock(&lloop_mutex); |
| 645 | |
| 646 | return err; |
| 647 | } |
| 648 | |
| 649 | static struct block_device_operations lo_fops = { |
| 650 | .owner = THIS_MODULE, |
| 651 | .open = lo_open, |
| 652 | .release = lo_release, |
| 653 | .ioctl = lo_ioctl, |
| 654 | }; |
| 655 | |
| 656 | /* dynamic iocontrol callback. |
| 657 | * This callback is registered in lloop_init and will be called by |
| 658 | * ll_iocontrol_call. |
| 659 | * |
| 660 | * This is a llite regular file ioctl function. It takes the responsibility |
| 661 | * of attaching or detaching a file by a lloop's device number. |
| 662 | */ |
| 663 | static enum llioc_iter lloop_ioctl(struct inode *unused, struct file *file, |
| 664 | unsigned int cmd, unsigned long arg, |
| 665 | void *magic, int *rcp) |
| 666 | { |
| 667 | struct lloop_device *lo = NULL; |
| 668 | struct block_device *bdev = NULL; |
| 669 | int err = 0; |
| 670 | dev_t dev; |
| 671 | |
| 672 | if (magic != ll_iocontrol_magic) |
| 673 | return LLIOC_CONT; |
| 674 | |
| 675 | if (disks == NULL) |
| 676 | GOTO(out1, err = -ENODEV); |
| 677 | |
| 678 | CWARN("Enter llop_ioctl\n"); |
| 679 | |
| 680 | mutex_lock(&lloop_mutex); |
| 681 | switch (cmd) { |
| 682 | case LL_IOC_LLOOP_ATTACH: { |
| 683 | struct lloop_device *lo_free = NULL; |
| 684 | int i; |
| 685 | |
| 686 | for (i = 0; i < max_loop; i++, lo = NULL) { |
| 687 | lo = &loop_dev[i]; |
| 688 | if (lo->lo_state == LLOOP_UNBOUND) { |
| 689 | if (!lo_free) |
| 690 | lo_free = lo; |
| 691 | continue; |
| 692 | } |
| 693 | if (lo->lo_backing_file->f_dentry->d_inode == |
| 694 | file->f_dentry->d_inode) |
| 695 | break; |
| 696 | } |
| 697 | if (lo || !lo_free) |
| 698 | GOTO(out, err = -EBUSY); |
| 699 | |
| 700 | lo = lo_free; |
| 701 | dev = MKDEV(lloop_major, lo->lo_number); |
| 702 | |
| 703 | /* quit if the used pointer is writable */ |
| 704 | if (put_user((long)old_encode_dev(dev), (long*)arg)) |
| 705 | GOTO(out, err = -EFAULT); |
| 706 | |
| 707 | bdev = blkdev_get_by_dev(dev, file->f_mode, NULL); |
| 708 | if (IS_ERR(bdev)) |
| 709 | GOTO(out, err = PTR_ERR(bdev)); |
| 710 | |
| 711 | get_file(file); |
| 712 | err = loop_set_fd(lo, NULL, bdev, file); |
| 713 | if (err) { |
| 714 | fput(file); |
| 715 | blkdev_put(bdev, 0); |
| 716 | } |
| 717 | |
| 718 | break; |
| 719 | } |
| 720 | |
| 721 | case LL_IOC_LLOOP_DETACH_BYDEV: { |
| 722 | int minor; |
| 723 | |
| 724 | dev = old_decode_dev(arg); |
| 725 | if (MAJOR(dev) != lloop_major) |
| 726 | GOTO(out, err = -EINVAL); |
| 727 | |
| 728 | minor = MINOR(dev); |
| 729 | if (minor > max_loop - 1) |
| 730 | GOTO(out, err = -EINVAL); |
| 731 | |
| 732 | lo = &loop_dev[minor]; |
| 733 | if (lo->lo_state != LLOOP_BOUND) |
| 734 | GOTO(out, err = -EINVAL); |
| 735 | |
| 736 | bdev = lo->lo_device; |
| 737 | err = loop_clr_fd(lo, bdev, 1); |
| 738 | if (err == 0) |
| 739 | blkdev_put(bdev, 0); /* grabbed in LLOOP_ATTACH */ |
| 740 | |
| 741 | break; |
| 742 | } |
| 743 | |
| 744 | default: |
| 745 | err = -EINVAL; |
| 746 | break; |
| 747 | } |
| 748 | |
| 749 | out: |
| 750 | mutex_unlock(&lloop_mutex); |
| 751 | out1: |
| 752 | if (rcp) |
| 753 | *rcp = err; |
| 754 | return LLIOC_STOP; |
| 755 | } |
| 756 | |
| 757 | static int __init lloop_init(void) |
| 758 | { |
| 759 | int i; |
| 760 | unsigned int cmdlist[] = { |
| 761 | LL_IOC_LLOOP_ATTACH, |
| 762 | LL_IOC_LLOOP_DETACH_BYDEV, |
| 763 | }; |
| 764 | |
| 765 | if (max_loop < 1 || max_loop > 256) { |
| 766 | max_loop = MAX_LOOP_DEFAULT; |
| 767 | CWARN("lloop: invalid max_loop (must be between" |
| 768 | " 1 and 256), using default (%u)\n", max_loop); |
| 769 | } |
| 770 | |
| 771 | lloop_major = register_blkdev(0, "lloop"); |
| 772 | if (lloop_major < 0) |
| 773 | return -EIO; |
| 774 | |
| 775 | CDEBUG(D_CONFIG, "registered lloop major %d with %u minors\n", |
| 776 | lloop_major, max_loop); |
| 777 | |
| 778 | ll_iocontrol_magic = ll_iocontrol_register(lloop_ioctl, 2, cmdlist); |
| 779 | if (ll_iocontrol_magic == NULL) |
| 780 | goto out_mem1; |
| 781 | |
| 782 | OBD_ALLOC_WAIT(loop_dev, max_loop * sizeof(*loop_dev)); |
| 783 | if (!loop_dev) |
| 784 | goto out_mem1; |
| 785 | |
| 786 | OBD_ALLOC_WAIT(disks, max_loop * sizeof(*disks)); |
| 787 | if (!disks) |
| 788 | goto out_mem2; |
| 789 | |
| 790 | for (i = 0; i < max_loop; i++) { |
| 791 | disks[i] = alloc_disk(1); |
| 792 | if (!disks[i]) |
| 793 | goto out_mem3; |
| 794 | } |
| 795 | |
| 796 | mutex_init(&lloop_mutex); |
| 797 | |
| 798 | for (i = 0; i < max_loop; i++) { |
| 799 | struct lloop_device *lo = &loop_dev[i]; |
| 800 | struct gendisk *disk = disks[i]; |
| 801 | |
| 802 | lo->lo_queue = blk_alloc_queue(GFP_KERNEL); |
| 803 | if (!lo->lo_queue) |
| 804 | goto out_mem4; |
| 805 | |
| 806 | mutex_init(&lo->lo_ctl_mutex); |
| 807 | sema_init(&lo->lo_sem, 0); |
| 808 | init_waitqueue_head(&lo->lo_bh_wait); |
| 809 | lo->lo_number = i; |
| 810 | spin_lock_init(&lo->lo_lock); |
| 811 | disk->major = lloop_major; |
| 812 | disk->first_minor = i; |
| 813 | disk->fops = &lo_fops; |
| 814 | sprintf(disk->disk_name, "lloop%d", i); |
| 815 | disk->private_data = lo; |
| 816 | disk->queue = lo->lo_queue; |
| 817 | } |
| 818 | |
| 819 | /* We cannot fail after we call this, so another loop!*/ |
| 820 | for (i = 0; i < max_loop; i++) |
| 821 | add_disk(disks[i]); |
| 822 | return 0; |
| 823 | |
| 824 | out_mem4: |
| 825 | while (i--) |
| 826 | blk_cleanup_queue(loop_dev[i].lo_queue); |
| 827 | i = max_loop; |
| 828 | out_mem3: |
| 829 | while (i--) |
| 830 | put_disk(disks[i]); |
| 831 | OBD_FREE(disks, max_loop * sizeof(*disks)); |
| 832 | out_mem2: |
| 833 | OBD_FREE(loop_dev, max_loop * sizeof(*loop_dev)); |
| 834 | out_mem1: |
| 835 | unregister_blkdev(lloop_major, "lloop"); |
| 836 | ll_iocontrol_unregister(ll_iocontrol_magic); |
| 837 | CERROR("lloop: ran out of memory\n"); |
| 838 | return -ENOMEM; |
| 839 | } |
| 840 | |
| 841 | static void lloop_exit(void) |
| 842 | { |
| 843 | int i; |
| 844 | |
| 845 | ll_iocontrol_unregister(ll_iocontrol_magic); |
| 846 | for (i = 0; i < max_loop; i++) { |
| 847 | del_gendisk(disks[i]); |
| 848 | blk_cleanup_queue(loop_dev[i].lo_queue); |
| 849 | put_disk(disks[i]); |
| 850 | } |
| 851 | |
| 852 | unregister_blkdev(lloop_major, "lloop"); |
| 853 | |
| 854 | OBD_FREE(disks, max_loop * sizeof(*disks)); |
| 855 | OBD_FREE(loop_dev, max_loop * sizeof(*loop_dev)); |
| 856 | } |
| 857 | |
| 858 | module_init(lloop_init); |
| 859 | module_exit(lloop_exit); |
| 860 | |
| 861 | module_param(max_loop, int, 0444); |
| 862 | MODULE_PARM_DESC(max_loop, "maximum of lloop_device"); |
| 863 | MODULE_AUTHOR("Sun Microsystems, Inc. <http://www.lustre.org/>"); |
| 864 | MODULE_DESCRIPTION("Lustre virtual block device"); |
| 865 | MODULE_LICENSE("GPL"); |