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 #include <linux/uaccess.h>
104
105 #include "../include/lustre_lib.h"
106 #include "../include/lustre_lite.h"
107 #include "llite_internal.h"
108
109 #define LLOOP_MAX_SEGMENTS LNET_MAX_IOV
110
111 /* Possible states of device */
112 enum {
113 LLOOP_UNBOUND,
114 LLOOP_BOUND,
115 LLOOP_RUNDOWN,
116 };
117
118 struct lloop_device {
119 int lo_number;
120 int lo_refcnt;
121 loff_t lo_offset;
122 loff_t lo_sizelimit;
123 int lo_flags;
124 struct file *lo_backing_file;
125 struct block_device *lo_device;
126 unsigned lo_blocksize;
127
128 gfp_t old_gfp_mask;
129
130 spinlock_t lo_lock;
131 struct bio *lo_bio;
132 struct bio *lo_biotail;
133 int lo_state;
134 struct semaphore lo_sem;
135 struct mutex lo_ctl_mutex;
136 atomic_t lo_pending;
137 wait_queue_head_t lo_bh_wait;
138
139 struct request_queue *lo_queue;
140
141 const struct lu_env *lo_env;
142 struct cl_io lo_io;
143 struct ll_dio_pages lo_pvec;
144
145 /* data to handle bio for lustre. */
146 struct lo_request_data {
147 struct page *lrd_pages[LLOOP_MAX_SEGMENTS];
148 loff_t lrd_offsets[LLOOP_MAX_SEGMENTS];
149 } lo_requests[1];
150 };
151
152 /*
153 * Loop flags
154 */
155 enum {
156 LO_FLAGS_READ_ONLY = 1,
157 };
158
159 static int lloop_major;
160 #define MAX_LOOP_DEFAULT 16
161 static int max_loop = MAX_LOOP_DEFAULT;
162 static struct lloop_device *loop_dev;
163 static struct gendisk **disks;
164 static struct mutex lloop_mutex;
165 static void *ll_iocontrol_magic;
166
get_loop_size(struct lloop_device * lo,struct file * file)167 static loff_t get_loop_size(struct lloop_device *lo, struct file *file)
168 {
169 loff_t size, offset, loopsize;
170
171 /* Compute loopsize in bytes */
172 size = i_size_read(file->f_mapping->host);
173 offset = lo->lo_offset;
174 loopsize = size - offset;
175 if (lo->lo_sizelimit > 0 && lo->lo_sizelimit < loopsize)
176 loopsize = lo->lo_sizelimit;
177
178 /*
179 * Unfortunately, if we want to do I/O on the device,
180 * the number of 512-byte sectors has to fit into a sector_t.
181 */
182 return loopsize >> 9;
183 }
184
do_bio_lustrebacked(struct lloop_device * lo,struct bio * head)185 static int do_bio_lustrebacked(struct lloop_device *lo, struct bio *head)
186 {
187 const struct lu_env *env = lo->lo_env;
188 struct cl_io *io = &lo->lo_io;
189 struct inode *inode = file_inode(lo->lo_backing_file);
190 struct cl_object *obj = ll_i2info(inode)->lli_clob;
191 pgoff_t offset;
192 int ret;
193 int rw;
194 u32 page_count = 0;
195 struct bio_vec bvec;
196 struct bvec_iter iter;
197 struct bio *bio;
198 ssize_t bytes;
199
200 struct ll_dio_pages *pvec = &lo->lo_pvec;
201 struct page **pages = pvec->ldp_pages;
202 loff_t *offsets = pvec->ldp_offsets;
203
204 truncate_inode_pages(inode->i_mapping, 0);
205
206 /* initialize the IO */
207 memset(io, 0, sizeof(*io));
208 io->ci_obj = obj;
209 ret = cl_io_init(env, io, CIT_MISC, obj);
210 if (ret)
211 return io->ci_result;
212 io->ci_lockreq = CILR_NEVER;
213
214 LASSERT(head != NULL);
215 rw = head->bi_rw;
216 for (bio = head; bio != NULL; bio = bio->bi_next) {
217 LASSERT(rw == bio->bi_rw);
218
219 offset = (pgoff_t)(bio->bi_iter.bi_sector << 9) + lo->lo_offset;
220 bio_for_each_segment(bvec, bio, iter) {
221 BUG_ON(bvec.bv_offset != 0);
222 BUG_ON(bvec.bv_len != PAGE_CACHE_SIZE);
223
224 pages[page_count] = bvec.bv_page;
225 offsets[page_count] = offset;
226 page_count++;
227 offset += bvec.bv_len;
228 }
229 LASSERT(page_count <= LLOOP_MAX_SEGMENTS);
230 }
231
232 ll_stats_ops_tally(ll_i2sbi(inode),
233 (rw == WRITE) ? LPROC_LL_BRW_WRITE : LPROC_LL_BRW_READ,
234 page_count);
235
236 pvec->ldp_size = page_count << PAGE_CACHE_SHIFT;
237 pvec->ldp_nr = page_count;
238
239 /* FIXME: in ll_direct_rw_pages, it has to allocate many cl_page{}s to
240 * write those pages into OST. Even worse case is that more pages
241 * would be asked to write out to swap space, and then finally get here
242 * again.
243 * Unfortunately this is NOT easy to fix.
244 * Thoughts on solution:
245 * 0. Define a reserved pool for cl_pages, which could be a list of
246 * pre-allocated cl_pages;
247 * 1. Define a new operation in cl_object_operations{}, says clo_depth,
248 * which measures how many layers for this lustre object. Generally
249 * speaking, the depth would be 2, one for llite, and one for lovsub.
250 * However, for SNS, there will be more since we need additional page
251 * to store parity;
252 * 2. Reserve the # of (page_count * depth) cl_pages from the reserved
253 * pool. Afterwards, the clio would allocate the pages from reserved
254 * pool, this guarantees we needn't allocate the cl_pages from
255 * generic cl_page slab cache.
256 * Of course, if there is NOT enough pages in the pool, we might
257 * be asked to write less pages once, this purely depends on
258 * implementation. Anyway, we should be careful to avoid deadlocking.
259 */
260 mutex_lock(&inode->i_mutex);
261 bytes = ll_direct_rw_pages(env, io, rw, inode, pvec);
262 mutex_unlock(&inode->i_mutex);
263 cl_io_fini(env, io);
264 return (bytes == pvec->ldp_size) ? 0 : (int)bytes;
265 }
266
267 /*
268 * Add bio to back of pending list
269 */
loop_add_bio(struct lloop_device * lo,struct bio * bio)270 static void loop_add_bio(struct lloop_device *lo, struct bio *bio)
271 {
272 unsigned long flags;
273
274 spin_lock_irqsave(&lo->lo_lock, flags);
275 if (lo->lo_biotail) {
276 lo->lo_biotail->bi_next = bio;
277 lo->lo_biotail = bio;
278 } else
279 lo->lo_bio = lo->lo_biotail = bio;
280 spin_unlock_irqrestore(&lo->lo_lock, flags);
281
282 atomic_inc(&lo->lo_pending);
283 if (waitqueue_active(&lo->lo_bh_wait))
284 wake_up(&lo->lo_bh_wait);
285 }
286
287 /*
288 * Grab first pending buffer
289 */
loop_get_bio(struct lloop_device * lo,struct bio ** req)290 static unsigned int loop_get_bio(struct lloop_device *lo, struct bio **req)
291 {
292 struct bio *first;
293 struct bio **bio;
294 unsigned int count = 0;
295 unsigned int page_count = 0;
296 int rw;
297
298 spin_lock_irq(&lo->lo_lock);
299 first = lo->lo_bio;
300 if (unlikely(first == NULL)) {
301 spin_unlock_irq(&lo->lo_lock);
302 return 0;
303 }
304
305 /* TODO: need to split the bio, too bad. */
306 LASSERT(first->bi_vcnt <= LLOOP_MAX_SEGMENTS);
307
308 rw = first->bi_rw;
309 bio = &lo->lo_bio;
310 while (*bio && (*bio)->bi_rw == rw) {
311 CDEBUG(D_INFO, "bio sector %llu size %u count %u vcnt%u \n",
312 (unsigned long long)(*bio)->bi_iter.bi_sector,
313 (*bio)->bi_iter.bi_size,
314 page_count, (*bio)->bi_vcnt);
315 if (page_count + (*bio)->bi_vcnt > LLOOP_MAX_SEGMENTS)
316 break;
317
318 page_count += (*bio)->bi_vcnt;
319 count++;
320 bio = &(*bio)->bi_next;
321 }
322 if (*bio) {
323 /* Some of bios can't be mergeable. */
324 lo->lo_bio = *bio;
325 *bio = NULL;
326 } else {
327 /* Hit the end of queue */
328 lo->lo_biotail = NULL;
329 lo->lo_bio = NULL;
330 }
331 *req = first;
332 spin_unlock_irq(&lo->lo_lock);
333 return count;
334 }
335
loop_make_request(struct request_queue * q,struct bio * old_bio)336 static blk_qc_t loop_make_request(struct request_queue *q, struct bio *old_bio)
337 {
338 struct lloop_device *lo = q->queuedata;
339 int rw = bio_rw(old_bio);
340 int inactive;
341
342 blk_queue_split(q, &old_bio, q->bio_split);
343
344 if (!lo)
345 goto err;
346
347 CDEBUG(D_INFO, "submit bio sector %llu size %u\n",
348 (unsigned long long)old_bio->bi_iter.bi_sector,
349 old_bio->bi_iter.bi_size);
350
351 spin_lock_irq(&lo->lo_lock);
352 inactive = lo->lo_state != LLOOP_BOUND;
353 spin_unlock_irq(&lo->lo_lock);
354 if (inactive)
355 goto err;
356
357 if (rw == WRITE) {
358 if (lo->lo_flags & LO_FLAGS_READ_ONLY)
359 goto err;
360 } else if (rw == READA) {
361 rw = READ;
362 } else if (rw != READ) {
363 CERROR("lloop: unknown command (%x)\n", rw);
364 goto err;
365 }
366 loop_add_bio(lo, old_bio);
367 return BLK_QC_T_NONE;
368 err:
369 bio_io_error(old_bio);
370 return BLK_QC_T_NONE;
371 }
372
loop_handle_bio(struct lloop_device * lo,struct bio * bio)373 static inline void loop_handle_bio(struct lloop_device *lo, struct bio *bio)
374 {
375 int ret;
376
377 ret = do_bio_lustrebacked(lo, bio);
378 while (bio) {
379 struct bio *tmp = bio->bi_next;
380
381 bio->bi_next = NULL;
382 bio->bi_error = ret;
383 bio_endio(bio);
384 bio = tmp;
385 }
386 }
387
loop_active(struct lloop_device * lo)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 */
loop_thread(void * data)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, MIN_NICE);
411
412 lo->lo_state = LLOOP_BOUND;
413
414 env = cl_env_get(&refcheck);
415 if (IS_ERR(env)) {
416 ret = PTR_ERR(env);
417 goto out;
418 }
419
420 lo->lo_env = env;
421 memset(&lo->lo_pvec, 0, sizeof(lo->lo_pvec));
422 lo->lo_pvec.ldp_pages = lo->lo_requests[0].lrd_pages;
423 lo->lo_pvec.ldp_offsets = lo->lo_requests[0].lrd_offsets;
424
425 /*
426 * up sem, we are running
427 */
428 up(&lo->lo_sem);
429
430 for (;;) {
431 wait_event(lo->lo_bh_wait, loop_active(lo));
432 if (!atomic_read(&lo->lo_pending)) {
433 int exiting = 0;
434
435 spin_lock_irq(&lo->lo_lock);
436 exiting = (lo->lo_state == LLOOP_RUNDOWN);
437 spin_unlock_irq(&lo->lo_lock);
438 if (exiting)
439 break;
440 }
441
442 bio = NULL;
443 count = loop_get_bio(lo, &bio);
444 if (!count) {
445 CWARN("lloop(minor: %d): missing bio\n", lo->lo_number);
446 continue;
447 }
448
449 total_count += count;
450 if (total_count < count) { /* overflow */
451 total_count = count;
452 times = 1;
453 } else {
454 times++;
455 }
456 if ((times & 127) == 0) {
457 CDEBUG(D_INFO, "total: %lu, count: %lu, avg: %lu\n",
458 total_count, times, total_count / times);
459 }
460
461 LASSERT(bio != NULL);
462 LASSERT(count <= atomic_read(&lo->lo_pending));
463 loop_handle_bio(lo, bio);
464 atomic_sub(count, &lo->lo_pending);
465 }
466 cl_env_put(env, &refcheck);
467
468 out:
469 up(&lo->lo_sem);
470 return ret;
471 }
472
loop_set_fd(struct lloop_device * lo,struct file * unused,struct block_device * bdev,struct file * file)473 static int loop_set_fd(struct lloop_device *lo, struct file *unused,
474 struct block_device *bdev, struct file *file)
475 {
476 struct inode *inode;
477 struct address_space *mapping;
478 int lo_flags = 0;
479 int error;
480 loff_t size;
481
482 if (!try_module_get(THIS_MODULE))
483 return -ENODEV;
484
485 error = -EBUSY;
486 if (lo->lo_state != LLOOP_UNBOUND)
487 goto out;
488
489 mapping = file->f_mapping;
490 inode = mapping->host;
491
492 error = -EINVAL;
493 if (!S_ISREG(inode->i_mode) || inode->i_sb->s_magic != LL_SUPER_MAGIC)
494 goto out;
495
496 if (!(file->f_mode & FMODE_WRITE))
497 lo_flags |= LO_FLAGS_READ_ONLY;
498
499 size = get_loop_size(lo, file);
500
501 if ((loff_t)(sector_t)size != size) {
502 error = -EFBIG;
503 goto out;
504 }
505
506 /* remove all pages in cache so as dirty pages not to be existent. */
507 truncate_inode_pages(mapping, 0);
508
509 set_device_ro(bdev, (lo_flags & LO_FLAGS_READ_ONLY) != 0);
510
511 lo->lo_blocksize = PAGE_CACHE_SIZE;
512 lo->lo_device = bdev;
513 lo->lo_flags = lo_flags;
514 lo->lo_backing_file = file;
515 lo->lo_sizelimit = 0;
516 lo->old_gfp_mask = mapping_gfp_mask(mapping);
517 mapping_set_gfp_mask(mapping, lo->old_gfp_mask & ~(__GFP_IO|__GFP_FS));
518
519 lo->lo_bio = lo->lo_biotail = NULL;
520
521 /*
522 * set queue make_request_fn, and add limits based on lower level
523 * device
524 */
525 blk_queue_make_request(lo->lo_queue, loop_make_request);
526 lo->lo_queue->queuedata = lo;
527
528 /* queue parameters */
529 CLASSERT(PAGE_CACHE_SIZE < (1 << (sizeof(unsigned short) * 8)));
530 blk_queue_logical_block_size(lo->lo_queue,
531 (unsigned short)PAGE_CACHE_SIZE);
532 blk_queue_max_hw_sectors(lo->lo_queue,
533 LLOOP_MAX_SEGMENTS << (PAGE_CACHE_SHIFT - 9));
534 blk_queue_max_segments(lo->lo_queue, LLOOP_MAX_SEGMENTS);
535
536 set_capacity(disks[lo->lo_number], size);
537 bd_set_size(bdev, size << 9);
538
539 set_blocksize(bdev, lo->lo_blocksize);
540
541 kthread_run(loop_thread, lo, "lloop%d", lo->lo_number);
542 down(&lo->lo_sem);
543 return 0;
544
545 out:
546 /* This is safe: open() is still holding a reference. */
547 module_put(THIS_MODULE);
548 return error;
549 }
550
loop_clr_fd(struct lloop_device * lo,struct block_device * bdev,int count)551 static int loop_clr_fd(struct lloop_device *lo, struct block_device *bdev,
552 int count)
553 {
554 struct file *filp = lo->lo_backing_file;
555 gfp_t gfp = lo->old_gfp_mask;
556
557 if (lo->lo_state != LLOOP_BOUND)
558 return -ENXIO;
559
560 if (lo->lo_refcnt > count) /* we needed one fd for the ioctl */
561 return -EBUSY;
562
563 if (filp == NULL)
564 return -EINVAL;
565
566 spin_lock_irq(&lo->lo_lock);
567 lo->lo_state = LLOOP_RUNDOWN;
568 spin_unlock_irq(&lo->lo_lock);
569 wake_up(&lo->lo_bh_wait);
570
571 down(&lo->lo_sem);
572 lo->lo_backing_file = NULL;
573 lo->lo_device = NULL;
574 lo->lo_offset = 0;
575 lo->lo_sizelimit = 0;
576 lo->lo_flags = 0;
577 invalidate_bdev(bdev);
578 set_capacity(disks[lo->lo_number], 0);
579 bd_set_size(bdev, 0);
580 mapping_set_gfp_mask(filp->f_mapping, gfp);
581 lo->lo_state = LLOOP_UNBOUND;
582 fput(filp);
583 /* This is safe: open() is still holding a reference. */
584 module_put(THIS_MODULE);
585 return 0;
586 }
587
lo_open(struct block_device * bdev,fmode_t mode)588 static int lo_open(struct block_device *bdev, fmode_t mode)
589 {
590 struct lloop_device *lo = bdev->bd_disk->private_data;
591
592 mutex_lock(&lo->lo_ctl_mutex);
593 lo->lo_refcnt++;
594 mutex_unlock(&lo->lo_ctl_mutex);
595
596 return 0;
597 }
598
lo_release(struct gendisk * disk,fmode_t mode)599 static void lo_release(struct gendisk *disk, fmode_t mode)
600 {
601 struct lloop_device *lo = disk->private_data;
602
603 mutex_lock(&lo->lo_ctl_mutex);
604 --lo->lo_refcnt;
605 mutex_unlock(&lo->lo_ctl_mutex);
606 }
607
608 /* lloop device node's ioctl function. */
lo_ioctl(struct block_device * bdev,fmode_t mode,unsigned int cmd,unsigned long arg)609 static int lo_ioctl(struct block_device *bdev, fmode_t mode,
610 unsigned int cmd, unsigned long arg)
611 {
612 struct lloop_device *lo = bdev->bd_disk->private_data;
613 struct inode *inode = NULL;
614 int err = 0;
615
616 mutex_lock(&lloop_mutex);
617 switch (cmd) {
618 case LL_IOC_LLOOP_DETACH: {
619 err = loop_clr_fd(lo, bdev, 2);
620 if (err == 0)
621 blkdev_put(bdev, 0); /* grabbed in LLOOP_ATTACH */
622 break;
623 }
624
625 case LL_IOC_LLOOP_INFO: {
626 struct lu_fid fid;
627
628 if (lo->lo_backing_file == NULL) {
629 err = -ENOENT;
630 break;
631 }
632 if (inode == NULL)
633 inode = file_inode(lo->lo_backing_file);
634 if (lo->lo_state == LLOOP_BOUND)
635 fid = ll_i2info(inode)->lli_fid;
636 else
637 fid_zero(&fid);
638
639 if (copy_to_user((struct lu_fid *)arg, &fid, sizeof(fid)))
640 err = -EFAULT;
641 break;
642 }
643
644 default:
645 err = -EINVAL;
646 break;
647 }
648 mutex_unlock(&lloop_mutex);
649
650 return err;
651 }
652
653 static struct block_device_operations lo_fops = {
654 .owner = THIS_MODULE,
655 .open = lo_open,
656 .release = lo_release,
657 .ioctl = lo_ioctl,
658 };
659
660 /* dynamic iocontrol callback.
661 * This callback is registered in lloop_init and will be called by
662 * ll_iocontrol_call.
663 *
664 * This is a llite regular file ioctl function. It takes the responsibility
665 * of attaching or detaching a file by a lloop's device number.
666 */
lloop_ioctl(struct inode * unused,struct file * file,unsigned int cmd,unsigned long arg,void * magic,int * rcp)667 static enum llioc_iter lloop_ioctl(struct inode *unused, struct file *file,
668 unsigned int cmd, unsigned long arg,
669 void *magic, int *rcp)
670 {
671 struct lloop_device *lo = NULL;
672 struct block_device *bdev = NULL;
673 int err = 0;
674 dev_t dev;
675
676 if (magic != ll_iocontrol_magic)
677 return LLIOC_CONT;
678
679 if (disks == NULL) {
680 err = -ENODEV;
681 goto out1;
682 }
683
684 CWARN("Enter llop_ioctl\n");
685
686 mutex_lock(&lloop_mutex);
687 switch (cmd) {
688 case LL_IOC_LLOOP_ATTACH: {
689 struct lloop_device *lo_free = NULL;
690 int i;
691
692 for (i = 0; i < max_loop; i++, lo = NULL) {
693 lo = &loop_dev[i];
694 if (lo->lo_state == LLOOP_UNBOUND) {
695 if (!lo_free)
696 lo_free = lo;
697 continue;
698 }
699 if (file_inode(lo->lo_backing_file) == file_inode(file))
700 break;
701 }
702 if (lo || !lo_free) {
703 err = -EBUSY;
704 goto out;
705 }
706
707 lo = lo_free;
708 dev = MKDEV(lloop_major, lo->lo_number);
709
710 /* quit if the used pointer is writable */
711 if (put_user((long)old_encode_dev(dev), (long *)arg)) {
712 err = -EFAULT;
713 goto out;
714 }
715
716 bdev = blkdev_get_by_dev(dev, file->f_mode, NULL);
717 if (IS_ERR(bdev)) {
718 err = PTR_ERR(bdev);
719 goto out;
720 }
721
722 get_file(file);
723 err = loop_set_fd(lo, NULL, bdev, file);
724 if (err) {
725 fput(file);
726 blkdev_put(bdev, 0);
727 }
728
729 break;
730 }
731
732 case LL_IOC_LLOOP_DETACH_BYDEV: {
733 int minor;
734
735 dev = old_decode_dev(arg);
736 if (MAJOR(dev) != lloop_major) {
737 err = -EINVAL;
738 goto out;
739 }
740
741 minor = MINOR(dev);
742 if (minor > max_loop - 1) {
743 err = -EINVAL;
744 goto out;
745 }
746
747 lo = &loop_dev[minor];
748 if (lo->lo_state != LLOOP_BOUND) {
749 err = -EINVAL;
750 goto out;
751 }
752
753 bdev = lo->lo_device;
754 err = loop_clr_fd(lo, bdev, 1);
755 if (err == 0)
756 blkdev_put(bdev, 0); /* grabbed in LLOOP_ATTACH */
757
758 break;
759 }
760
761 default:
762 err = -EINVAL;
763 break;
764 }
765
766 out:
767 mutex_unlock(&lloop_mutex);
768 out1:
769 if (rcp)
770 *rcp = err;
771 return LLIOC_STOP;
772 }
773
lloop_init(void)774 static int __init lloop_init(void)
775 {
776 int i;
777 unsigned int cmdlist[] = {
778 LL_IOC_LLOOP_ATTACH,
779 LL_IOC_LLOOP_DETACH_BYDEV,
780 };
781
782 if (max_loop < 1 || max_loop > 256) {
783 max_loop = MAX_LOOP_DEFAULT;
784 CWARN("lloop: invalid max_loop (must be between 1 and 256), using default (%u)\n",
785 max_loop);
786 }
787
788 lloop_major = register_blkdev(0, "lloop");
789 if (lloop_major < 0)
790 return -EIO;
791
792 CDEBUG(D_CONFIG, "registered lloop major %d with %u minors\n",
793 lloop_major, max_loop);
794
795 ll_iocontrol_magic = ll_iocontrol_register(lloop_ioctl, 2, cmdlist);
796 if (ll_iocontrol_magic == NULL)
797 goto out_mem1;
798
799 loop_dev = kcalloc(max_loop, sizeof(*loop_dev), GFP_KERNEL);
800 if (!loop_dev)
801 goto out_mem1;
802
803 disks = kcalloc(max_loop, sizeof(*disks), GFP_KERNEL);
804 if (!disks)
805 goto out_mem2;
806
807 for (i = 0; i < max_loop; i++) {
808 disks[i] = alloc_disk(1);
809 if (!disks[i])
810 goto out_mem3;
811 }
812
813 mutex_init(&lloop_mutex);
814
815 for (i = 0; i < max_loop; i++) {
816 struct lloop_device *lo = &loop_dev[i];
817 struct gendisk *disk = disks[i];
818
819 lo->lo_queue = blk_alloc_queue(GFP_KERNEL);
820 if (!lo->lo_queue)
821 goto out_mem4;
822
823 mutex_init(&lo->lo_ctl_mutex);
824 sema_init(&lo->lo_sem, 0);
825 init_waitqueue_head(&lo->lo_bh_wait);
826 lo->lo_number = i;
827 spin_lock_init(&lo->lo_lock);
828 disk->major = lloop_major;
829 disk->first_minor = i;
830 disk->fops = &lo_fops;
831 sprintf(disk->disk_name, "lloop%d", i);
832 disk->private_data = lo;
833 disk->queue = lo->lo_queue;
834 }
835
836 /* We cannot fail after we call this, so another loop!*/
837 for (i = 0; i < max_loop; i++)
838 add_disk(disks[i]);
839 return 0;
840
841 out_mem4:
842 while (i--)
843 blk_cleanup_queue(loop_dev[i].lo_queue);
844 i = max_loop;
845 out_mem3:
846 while (i--)
847 put_disk(disks[i]);
848 kfree(disks);
849 out_mem2:
850 kfree(loop_dev);
851 out_mem1:
852 unregister_blkdev(lloop_major, "lloop");
853 ll_iocontrol_unregister(ll_iocontrol_magic);
854 CERROR("lloop: ran out of memory\n");
855 return -ENOMEM;
856 }
857
lloop_exit(void)858 static void lloop_exit(void)
859 {
860 int i;
861
862 ll_iocontrol_unregister(ll_iocontrol_magic);
863 for (i = 0; i < max_loop; i++) {
864 del_gendisk(disks[i]);
865 blk_cleanup_queue(loop_dev[i].lo_queue);
866 put_disk(disks[i]);
867 }
868
869 unregister_blkdev(lloop_major, "lloop");
870
871 kfree(disks);
872 kfree(loop_dev);
873 }
874
875 module_init(lloop_init);
876 module_exit(lloop_exit);
877
878 module_param(max_loop, int, 0444);
879 MODULE_PARM_DESC(max_loop, "maximum of lloop_device");
880 MODULE_AUTHOR("Sun Microsystems, Inc. <http://www.lustre.org/>");
881 MODULE_DESCRIPTION("Lustre virtual block device");
882 MODULE_LICENSE("GPL");
883