projects / linux-block.git / blame
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (incremental) | history | HEAD
staging: lustre: fld: remove ccflags from Makefile
[linux-block.git] / drivers / staging / lustre / lustre / llite / lloop.c
CommitLineData
d7e09d03
PT		 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>
d7e09d03
PT		 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 */
113enum {
114 LLOOP_UNBOUND,
115 LLOOP_BOUND,
116 LLOOP_RUNDOWN,
117};
118
119struct lloop_device {
120 int lo_number;
121 int lo_refcnt;
122 loff_t lo_offset;
123 loff_t lo_sizelimit;
124 int lo_flags;
d7e09d03
PT		 125 struct file *lo_backing_file;
126 struct block_device *lo_device;
127 unsigned lo_blocksize;
128
129 int old_gfp_mask;
130
131 spinlock_t lo_lock;
132 struct bio *lo_bio;
133 struct bio *lo_biotail;
134 int lo_state;
135 struct semaphore lo_sem;
136 struct mutex lo_ctl_mutex;
137 atomic_t lo_pending;
138 wait_queue_head_t lo_bh_wait;
139
140 struct request_queue *lo_queue;
141
142 const struct lu_env *lo_env;
143 struct cl_io lo_io;
144 struct ll_dio_pages lo_pvec;
145
146 /* data to handle bio for lustre. */
147 struct lo_request_data {
148 struct page *lrd_pages[LLOOP_MAX_SEGMENTS];
149 loff_t lrd_offsets[LLOOP_MAX_SEGMENTS];
150 } lo_requests[1];
151};
152
153/*
154 * Loop flags
155 */
156enum {
157 LO_FLAGS_READ_ONLY = 1,
158};
159
160static int lloop_major;
161#define MAX_LOOP_DEFAULT 16
162static int max_loop = MAX_LOOP_DEFAULT;
163static struct lloop_device *loop_dev;
164static struct gendisk **disks;
165static struct mutex lloop_mutex;
166static void *ll_iocontrol_magic = NULL;
167
168static loff_t get_loop_size(struct lloop_device *lo, struct file *file)
169{
170 loff_t size, offset, loopsize;
171
172 /* Compute loopsize in bytes */
173 size = i_size_read(file->f_mapping->host);
174 offset = lo->lo_offset;
175 loopsize = size - offset;
176 if (lo->lo_sizelimit > 0 && lo->lo_sizelimit < loopsize)
177 loopsize = lo->lo_sizelimit;
178
179 /*
180 * Unfortunately, if we want to do I/O on the device,
181 * the number of 512-byte sectors has to fit into a sector_t.
182 */
183 return loopsize >> 9;
184}
185
186static int do_bio_lustrebacked(struct lloop_device *lo, struct bio *head)
187{
188 const struct lu_env *env = lo->lo_env;
189 struct cl_io *io = &lo->lo_io;
190 struct inode *inode = lo->lo_backing_file->f_dentry->d_inode;
191 struct cl_object *obj = ll_i2info(inode)->lli_clob;
192 pgoff_t offset;
193 int ret;
d7e09d03
PT		 194 int rw;
195 obd_count page_count = 0;
7988613b
KO		 196 struct bio_vec bvec;
197 struct bvec_iter iter;
d7e09d03
PT		 198 struct bio *bio;
199 ssize_t bytes;
200
201 struct ll_dio_pages *pvec = &lo->lo_pvec;
202 struct page **pages = pvec->ldp_pages;
203 loff_t *offsets = pvec->ldp_offsets;
204
205 truncate_inode_pages(inode->i_mapping, 0);
206
207 /* initialize the IO */
208 memset(io, 0, sizeof(*io));
209 io->ci_obj = obj;
210 ret = cl_io_init(env, io, CIT_MISC, obj);
211 if (ret)
212 return io->ci_result;
213 io->ci_lockreq = CILR_NEVER;
214
215 LASSERT(head != NULL);
216 rw = head->bi_rw;
217 for (bio = head; bio != NULL; bio = bio->bi_next) {
218 LASSERT(rw == bio->bi_rw);
219
4f024f37 220 offset = (pgoff_t)(bio->bi_iter.bi_sector << 9) + lo->lo_offset;
7988613b
KO		 221 bio_for_each_segment(bvec, bio, iter) {
222 BUG_ON(bvec.bv_offset != 0);
223 BUG_ON(bvec.bv_len != PAGE_CACHE_SIZE);
d7e09d03 224
7988613b 225 pages[page_count] = bvec.bv_page;
d7e09d03
PT		 226 offsets[page_count] = offset;
227 page_count++;
7988613b 228 offset += bvec.bv_len;
d7e09d03
PT		 229 }
230 LASSERT(page_count <= LLOOP_MAX_SEGMENTS);
231 }
232
233 ll_stats_ops_tally(ll_i2sbi(inode),
234 (rw == WRITE) ? LPROC_LL_BRW_WRITE : LPROC_LL_BRW_READ,
235 page_count);
236
237 pvec->ldp_size = page_count << PAGE_CACHE_SHIFT;
238 pvec->ldp_nr = page_count;
239
240 /* FIXME: in ll_direct_rw_pages, it has to allocate many cl_page{}s to
241 * write those pages into OST. Even worse case is that more pages
242 * would be asked to write out to swap space, and then finally get here
243 * again.
244 * Unfortunately this is NOT easy to fix.
245 * Thoughts on solution:
246 * 0. Define a reserved pool for cl_pages, which could be a list of
247 * pre-allocated cl_pages;
248 * 1. Define a new operation in cl_object_operations{}, says clo_depth,
249 * which measures how many layers for this lustre object. Generally
250 * speaking, the depth would be 2, one for llite, and one for lovsub.
251 * However, for SNS, there will be more since we need additional page
252 * to store parity;
253 * 2. Reserve the # of (page_count * depth) cl_pages from the reserved
254 * pool. Afterwards, the clio would allocate the pages from reserved
d0a0acc3 255 * pool, this guarantees we needn't allocate the cl_pages from
d7e09d03
PT		 256 * generic cl_page slab cache.
257 * Of course, if there is NOT enough pages in the pool, we might
258 * be asked to write less pages once, this purely depends on
259 * implementation. Anyway, we should be careful to avoid deadlocking.
260 */
261 mutex_lock(&inode->i_mutex);
262 bytes = ll_direct_rw_pages(env, io, rw, inode, pvec);
263 mutex_unlock(&inode->i_mutex);
264 cl_io_fini(env, io);
265 return (bytes == pvec->ldp_size) ? 0 : (int)bytes;
266}
267
268/*
269 * Add bio to back of pending list
270 */
271static void loop_add_bio(struct lloop_device *lo, struct bio *bio)
272{
273 unsigned long flags;
274
275 spin_lock_irqsave(&lo->lo_lock, flags);
276 if (lo->lo_biotail) {
277 lo->lo_biotail->bi_next = bio;
278 lo->lo_biotail = bio;
279 } else
280 lo->lo_bio = lo->lo_biotail = bio;
281 spin_unlock_irqrestore(&lo->lo_lock, flags);
282
283 atomic_inc(&lo->lo_pending);
284 if (waitqueue_active(&lo->lo_bh_wait))
285 wake_up(&lo->lo_bh_wait);
286}
287
288/*
289 * Grab first pending buffer
290 */
291static unsigned int loop_get_bio(struct lloop_device *lo, struct bio **req)
292{
293 struct bio *first;
294 struct bio **bio;
295 unsigned int count = 0;
296 unsigned int page_count = 0;
297 int rw;
298
299 spin_lock_irq(&lo->lo_lock);
300 first = lo->lo_bio;
301 if (unlikely(first == NULL)) {
302 spin_unlock_irq(&lo->lo_lock);
303 return 0;
304 }
305
306 /* TODO: need to split the bio, too bad. */
307 LASSERT(first->bi_vcnt <= LLOOP_MAX_SEGMENTS);
308
309 rw = first->bi_rw;
310 bio = &lo->lo_bio;
311 while (*bio && (*bio)->bi_rw == rw) {
312 CDEBUG(D_INFO, "bio sector %llu size %u count %u vcnt%u \n",
4f024f37
KO		 313 (unsigned long long)(*bio)->bi_iter.bi_sector,
314 (*bio)->bi_iter.bi_size,
d7e09d03
PT		 315 page_count, (*bio)->bi_vcnt);
316 if (page_count + (*bio)->bi_vcnt > LLOOP_MAX_SEGMENTS)
317 break;
318
319
320 page_count += (*bio)->bi_vcnt;
321 count++;
322 bio = &(*bio)->bi_next;
323 }
324 if (*bio) {
d0a0acc3 325 /* Some of bios can't be mergeable. */
d7e09d03
PT		 326 lo->lo_bio = *bio;
327 *bio = NULL;
328 } else {
329 /* Hit the end of queue */
330 lo->lo_biotail = NULL;
331 lo->lo_bio = NULL;
332 }
333 *req = first;
334 spin_unlock_irq(&lo->lo_lock);
335 return count;
336}
337
3ec65cb3 338static void loop_make_request(struct request_queue *q, struct bio *old_bio)
d7e09d03
PT		 339{
340 struct lloop_device *lo = q->queuedata;
341 int rw = bio_rw(old_bio);
342 int inactive;
343
344 if (!lo)
345 goto err;
346
347 CDEBUG(D_INFO, "submit bio sector %llu size %u\n",
4f024f37
KO		 348 (unsigned long long)old_bio->bi_iter.bi_sector,
349 old_bio->bi_iter.bi_size);
d7e09d03
PT		 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);
de40d120 367 return;
d7e09d03 368err:
4f024f37 369 cfs_bio_io_error(old_bio, old_bio->bi_iter.bi_size);
d7e09d03
PT		 370}
371
372
373static inline void loop_handle_bio(struct lloop_device *lo, struct bio *bio)
374{
375 int ret;
376 ret = do_bio_lustrebacked(lo, bio);
377 while (bio) {
378 struct bio *tmp = bio->bi_next;
379 bio->bi_next = NULL;
4f024f37 380 cfs_bio_endio(bio, bio->bi_iter.bi_size, ret);
d7e09d03
PT		 381 bio = tmp;
382 }
383}
384
385static inline int loop_active(struct lloop_device *lo)
386{
387 return atomic_read(&lo->lo_pending) ||
388 (lo->lo_state == LLOOP_RUNDOWN);
389}
390
391/*
392 * worker thread that handles reads/writes to file backed loop devices,
393 * to avoid blocking in our make_request_fn.
394 */
395static int loop_thread(void *data)
396{
397 struct lloop_device *lo = data;
398 struct bio *bio;
399 unsigned int count;
400 unsigned long times = 0;
401 unsigned long total_count = 0;
402
403 struct lu_env *env;
404 int refcheck;
405 int ret = 0;
406
8698a745 407 set_user_nice(current, MIN_NICE);
d7e09d03
PT		 408
409 lo->lo_state = LLOOP_BOUND;
410
411 env = cl_env_get(&refcheck);
412 if (IS_ERR(env))
413 GOTO(out, ret = PTR_ERR(env));
414
415 lo->lo_env = env;
416 memset(&lo->lo_pvec, 0, sizeof(lo->lo_pvec));
417 lo->lo_pvec.ldp_pages = lo->lo_requests[0].lrd_pages;
418 lo->lo_pvec.ldp_offsets = lo->lo_requests[0].lrd_offsets;
419
420 /*
421 * up sem, we are running
422 */
423 up(&lo->lo_sem);
424
425 for (;;) {
426 wait_event(lo->lo_bh_wait, loop_active(lo));
427 if (!atomic_read(&lo->lo_pending)) {
428 int exiting = 0;
429 spin_lock_irq(&lo->lo_lock);
430 exiting = (lo->lo_state == LLOOP_RUNDOWN);
431 spin_unlock_irq(&lo->lo_lock);
432 if (exiting)
433 break;
434 }
435
436 bio = NULL;
437 count = loop_get_bio(lo, &bio);
438 if (!count) {
439 CWARN("lloop(minor: %d): missing bio\n", lo->lo_number);
440 continue;
441 }
442
443 total_count += count;
444 if (total_count < count) { /* overflow */
445 total_count = count;
446 times = 1;
447 } else {
448 times++;
449 }
450 if ((times & 127) == 0) {
451 CDEBUG(D_INFO, "total: %lu, count: %lu, avg: %lu\n",
452 total_count, times, total_count / times);
453 }
454
455 LASSERT(bio != NULL);
456 LASSERT(count <= atomic_read(&lo->lo_pending));
457 loop_handle_bio(lo, bio);
458 atomic_sub(count, &lo->lo_pending);
459 }
460 cl_env_put(env, &refcheck);
461
462out:
463 up(&lo->lo_sem);
464 return ret;
465}
466
467static int loop_set_fd(struct lloop_device *lo, struct file *unused,
468 struct block_device *bdev, struct file *file)
469{
470 struct inode *inode;
471 struct address_space *mapping;
472 int lo_flags = 0;
473 int error;
474 loff_t size;
475
476 if (!try_module_get(THIS_MODULE))
477 return -ENODEV;
478
479 error = -EBUSY;
480 if (lo->lo_state != LLOOP_UNBOUND)
481 goto out;
482
483 mapping = file->f_mapping;
484 inode = mapping->host;
485
486 error = -EINVAL;
487 if (!S_ISREG(inode->i_mode) || inode->i_sb->s_magic != LL_SUPER_MAGIC)
488 goto out;
489
490 if (!(file->f_mode & FMODE_WRITE))
491 lo_flags |= LO_FLAGS_READ_ONLY;
492
493 size = get_loop_size(lo, file);
494
495 if ((loff_t)(sector_t)size != size) {
496 error = -EFBIG;
497 goto out;
498 }
499
500 /* remove all pages in cache so as dirty pages not to be existent. */
501 truncate_inode_pages(mapping, 0);
502
503 set_device_ro(bdev, (lo_flags & LO_FLAGS_READ_ONLY) != 0);
504
505 lo->lo_blocksize = PAGE_CACHE_SIZE;
506 lo->lo_device = bdev;
507 lo->lo_flags = lo_flags;
508 lo->lo_backing_file = file;
d7e09d03
PT		 509 lo->lo_sizelimit = 0;
510 lo->old_gfp_mask = mapping_gfp_mask(mapping);
511 mapping_set_gfp_mask(mapping, lo->old_gfp_mask & ~(__GFP_IO|__GFP_FS));
512
513 lo->lo_bio = lo->lo_biotail = NULL;
514
515 /*
516 * set queue make_request_fn, and add limits based on lower level
517 * device
518 */
519 blk_queue_make_request(lo->lo_queue, loop_make_request);
520 lo->lo_queue->queuedata = lo;
521
522 /* queue parameters */
523 CLASSERT(PAGE_CACHE_SIZE < (1 << (sizeof(unsigned short) * 8)));
524 blk_queue_logical_block_size(lo->lo_queue,
525 (unsigned short)PAGE_CACHE_SIZE);
526 blk_queue_max_hw_sectors(lo->lo_queue,
527 LLOOP_MAX_SEGMENTS << (PAGE_CACHE_SHIFT - 9));
528 blk_queue_max_segments(lo->lo_queue, LLOOP_MAX_SEGMENTS);
529
530 set_capacity(disks[lo->lo_number], size);
531 bd_set_size(bdev, size << 9);
532
533 set_blocksize(bdev, lo->lo_blocksize);
534
535 kthread_run(loop_thread, lo, "lloop%d", lo->lo_number);
536 down(&lo->lo_sem);
537 return 0;
538
539out:
540 /* This is safe: open() is still holding a reference. */
541 module_put(THIS_MODULE);
542 return error;
543}
544
545static int loop_clr_fd(struct lloop_device *lo, struct block_device *bdev,
546 int count)
547{
548 struct file *filp = lo->lo_backing_file;
549 int gfp = lo->old_gfp_mask;
550
551 if (lo->lo_state != LLOOP_BOUND)
552 return -ENXIO;
553
554 if (lo->lo_refcnt > count) /* we needed one fd for the ioctl */
555 return -EBUSY;
556
557 if (filp == NULL)
558 return -EINVAL;
559
560 spin_lock_irq(&lo->lo_lock);
561 lo->lo_state = LLOOP_RUNDOWN;
562 spin_unlock_irq(&lo->lo_lock);
563 wake_up(&lo->lo_bh_wait);
564
565 down(&lo->lo_sem);
566 lo->lo_backing_file = NULL;
d7e09d03
PT		 567 lo->lo_device = NULL;
568 lo->lo_offset = 0;
569 lo->lo_sizelimit = 0;
570 lo->lo_flags = 0;
db7392c2 571 invalidate_bdev(bdev);
d7e09d03
PT		 572 set_capacity(disks[lo->lo_number], 0);
573 bd_set_size(bdev, 0);
574 mapping_set_gfp_mask(filp->f_mapping, gfp);
575 lo->lo_state = LLOOP_UNBOUND;
576 fput(filp);
577 /* This is safe: open() is still holding a reference. */
578 module_put(THIS_MODULE);
579 return 0;
580}
581
582static int lo_open(struct block_device *bdev, fmode_t mode)
583{
584 struct lloop_device *lo = bdev->bd_disk->private_data;
585
586 mutex_lock(&lo->lo_ctl_mutex);
587 lo->lo_refcnt++;
588 mutex_unlock(&lo->lo_ctl_mutex);
589
590 return 0;
591}
592
f1e2f53e 593static void lo_release(struct gendisk *disk, fmode_t mode)
d7e09d03
PT		 594{
595 struct lloop_device *lo = disk->private_data;
596
597 mutex_lock(&lo->lo_ctl_mutex);
598 --lo->lo_refcnt;
599 mutex_unlock(&lo->lo_ctl_mutex);
d7e09d03
PT		 600}
601
602/* lloop device node's ioctl function. */
603static int lo_ioctl(struct block_device *bdev, fmode_t mode,
604 unsigned int cmd, unsigned long arg)
605{
606 struct lloop_device *lo = bdev->bd_disk->private_data;
607 struct inode *inode = NULL;
608 int err = 0;
609
610 mutex_lock(&lloop_mutex);
611 switch (cmd) {
612 case LL_IOC_LLOOP_DETACH: {
613 err = loop_clr_fd(lo, bdev, 2);
614 if (err == 0)
418690b6 615 blkdev_put(bdev, 0); /* grabbed in LLOOP_ATTACH */
d7e09d03
PT		 616 break;
617 }
618
619 case LL_IOC_LLOOP_INFO: {
620 struct lu_fid fid;
621
4de665c1
BG		 622 if (lo->lo_backing_file == NULL) {
623 err = -ENOENT;
624 break;
625 }
d7e09d03
PT		 626 if (inode == NULL)
627 inode = lo->lo_backing_file->f_dentry->d_inode;
628 if (lo->lo_state == LLOOP_BOUND)
629 fid = ll_i2info(inode)->lli_fid;
630 else
631 fid_zero(&fid);
632
633 if (copy_to_user((struct lu_fid *)arg, &fid, sizeof(fid)))
634 err = -EFAULT;
635 break;
636 }
637
638 default:
639 err = -EINVAL;
640 break;
641 }
642 mutex_unlock(&lloop_mutex);
643
644 return err;
645}
646
647static struct block_device_operations lo_fops = {
648 .owner = THIS_MODULE,
649 .open = lo_open,
650 .release = lo_release,
651 .ioctl = lo_ioctl,
652};
653
654/* dynamic iocontrol callback.
655 * This callback is registered in lloop_init and will be called by
656 * ll_iocontrol_call.
657 *
658 * This is a llite regular file ioctl function. It takes the responsibility
d0a0acc3 659 * of attaching or detaching a file by a lloop's device number.
d7e09d03
PT		 660 */
661static enum llioc_iter lloop_ioctl(struct inode *unused, struct file *file,
662 unsigned int cmd, unsigned long arg,
663 void *magic, int *rcp)
664{
665 struct lloop_device *lo = NULL;
666 struct block_device *bdev = NULL;
667 int err = 0;
668 dev_t dev;
669
670 if (magic != ll_iocontrol_magic)
671 return LLIOC_CONT;
672
673 if (disks == NULL)
674 GOTO(out1, err = -ENODEV);
675
676 CWARN("Enter llop_ioctl\n");
677
678 mutex_lock(&lloop_mutex);
679 switch (cmd) {
680 case LL_IOC_LLOOP_ATTACH: {
681 struct lloop_device *lo_free = NULL;
682 int i;
683
684 for (i = 0; i < max_loop; i++, lo = NULL) {
685 lo = &loop_dev[i];
686 if (lo->lo_state == LLOOP_UNBOUND) {
687 if (!lo_free)
688 lo_free = lo;
689 continue;
690 }
691 if (lo->lo_backing_file->f_dentry->d_inode ==
692 file->f_dentry->d_inode)
693 break;
694 }
695 if (lo || !lo_free)
696 GOTO(out, err = -EBUSY);
697
698 lo = lo_free;
699 dev = MKDEV(lloop_major, lo->lo_number);
700
701 /* quit if the used pointer is writable */
702 if (put_user((long)old_encode_dev(dev), (long*)arg))
703 GOTO(out, err = -EFAULT);
704
705 bdev = blkdev_get_by_dev(dev, file->f_mode, NULL);
706 if (IS_ERR(bdev))
707 GOTO(out, err = PTR_ERR(bdev));
708
709 get_file(file);
710 err = loop_set_fd(lo, NULL, bdev, file);
711 if (err) {
712 fput(file);
418690b6 713 blkdev_put(bdev, 0);
d7e09d03
PT		 714 }
715
716 break;
717 }
718
719 case LL_IOC_LLOOP_DETACH_BYDEV: {
720 int minor;
721
722 dev = old_decode_dev(arg);
723 if (MAJOR(dev) != lloop_major)
724 GOTO(out, err = -EINVAL);
725
726 minor = MINOR(dev);
727 if (minor > max_loop - 1)
728 GOTO(out, err = -EINVAL);
729
730 lo = &loop_dev[minor];
731 if (lo->lo_state != LLOOP_BOUND)
732 GOTO(out, err = -EINVAL);
733
734 bdev = lo->lo_device;
735 err = loop_clr_fd(lo, bdev, 1);
736 if (err == 0)
418690b6 737 blkdev_put(bdev, 0); /* grabbed in LLOOP_ATTACH */
d7e09d03
PT		 738
739 break;
740 }
741
742 default:
743 err = -EINVAL;
744 break;
745 }
746
747out:
748 mutex_unlock(&lloop_mutex);
749out1:
750 if (rcp)
751 *rcp = err;
752 return LLIOC_STOP;
753}
754
755static int __init lloop_init(void)
756{
757 int i;
758 unsigned int cmdlist[] = {
759 LL_IOC_LLOOP_ATTACH,
760 LL_IOC_LLOOP_DETACH_BYDEV,
761 };
762
763 if (max_loop < 1 || max_loop > 256) {
764 max_loop = MAX_LOOP_DEFAULT;
765 CWARN("lloop: invalid max_loop (must be between"
766 " 1 and 256), using default (%u)\n", max_loop);
767 }
768
769 lloop_major = register_blkdev(0, "lloop");
770 if (lloop_major < 0)
771 return -EIO;
772
773 CDEBUG(D_CONFIG, "registered lloop major %d with %u minors\n",
774 lloop_major, max_loop);
775
776 ll_iocontrol_magic = ll_iocontrol_register(lloop_ioctl, 2, cmdlist);
777 if (ll_iocontrol_magic == NULL)
778 goto out_mem1;
779
780 OBD_ALLOC_WAIT(loop_dev, max_loop * sizeof(*loop_dev));
781 if (!loop_dev)
782 goto out_mem1;
783
784 OBD_ALLOC_WAIT(disks, max_loop * sizeof(*disks));
785 if (!disks)
786 goto out_mem2;
787
788 for (i = 0; i < max_loop; i++) {
789 disks[i] = alloc_disk(1);
790 if (!disks[i])
791 goto out_mem3;
792 }
793
794 mutex_init(&lloop_mutex);
795
796 for (i = 0; i < max_loop; i++) {
797 struct lloop_device *lo = &loop_dev[i];
798 struct gendisk *disk = disks[i];
799
800 lo->lo_queue = blk_alloc_queue(GFP_KERNEL);
801 if (!lo->lo_queue)
802 goto out_mem4;
803
804 mutex_init(&lo->lo_ctl_mutex);
805 sema_init(&lo->lo_sem, 0);
806 init_waitqueue_head(&lo->lo_bh_wait);
807 lo->lo_number = i;
808 spin_lock_init(&lo->lo_lock);
809 disk->major = lloop_major;
810 disk->first_minor = i;
811 disk->fops = &lo_fops;
812 sprintf(disk->disk_name, "lloop%d", i);
813 disk->private_data = lo;
814 disk->queue = lo->lo_queue;
815 }
816
817 /* We cannot fail after we call this, so another loop!*/
818 for (i = 0; i < max_loop; i++)
819 add_disk(disks[i]);
820 return 0;
821
822out_mem4:
823 while (i--)
824 blk_cleanup_queue(loop_dev[i].lo_queue);
825 i = max_loop;
826out_mem3:
827 while (i--)
828 put_disk(disks[i]);
829 OBD_FREE(disks, max_loop * sizeof(*disks));
830out_mem2:
831 OBD_FREE(loop_dev, max_loop * sizeof(*loop_dev));
832out_mem1:
833 unregister_blkdev(lloop_major, "lloop");
834 ll_iocontrol_unregister(ll_iocontrol_magic);
835 CERROR("lloop: ran out of memory\n");
836 return -ENOMEM;
837}
838
839static void lloop_exit(void)
840{
841 int i;
842
843 ll_iocontrol_unregister(ll_iocontrol_magic);
844 for (i = 0; i < max_loop; i++) {
845 del_gendisk(disks[i]);
846 blk_cleanup_queue(loop_dev[i].lo_queue);
847 put_disk(disks[i]);
848 }
49126e47
XZ		 849
850 unregister_blkdev(lloop_major, "lloop");
d7e09d03
PT		 851
852 OBD_FREE(disks, max_loop * sizeof(*disks));
853 OBD_FREE(loop_dev, max_loop * sizeof(*loop_dev));
854}
855
856module_init(lloop_init);
857module_exit(lloop_exit);
858
8cc7b4b9
PT		 859module_param(max_loop, int, 0444);
860MODULE_PARM_DESC(max_loop, "maximum of lloop_device");
d7e09d03
PT		 861MODULE_AUTHOR("Sun Microsystems, Inc. <http://www.lustre.org/>");
862MODULE_DESCRIPTION("Lustre virtual block device");
863MODULE_LICENSE("GPL");
Linux 6.x block layer and io_uring tree(s)