projects / deliverable / linux.git / blame
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (incremental) | history | HEAD
drivers/staging/dwc2: don't check resource with devm_ioremap_resource
[deliverable/linux.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>
102#include <linux/swap.h>
103#include <linux/pagevec.h>
104
105#include <asm/uaccess.h>
106
107#include <lustre_lib.h>
108#include <lustre_lite.h>
109#include "llite_internal.h"
110
111#define LLOOP_MAX_SEGMENTS LNET_MAX_IOV
112
113/* Possible states of device */
114enum {
115 LLOOP_UNBOUND,
116 LLOOP_BOUND,
117 LLOOP_RUNDOWN,
118};
119
120struct lloop_device {
121 int lo_number;
122 int lo_refcnt;
123 loff_t lo_offset;
124 loff_t lo_sizelimit;
125 int lo_flags;
126 int (*ioctl)(struct lloop_device *, int cmd,
127 unsigned long arg);
128
129 struct file *lo_backing_file;
130 struct block_device *lo_device;
131 unsigned lo_blocksize;
132
133 int old_gfp_mask;
134
135 spinlock_t lo_lock;
136 struct bio *lo_bio;
137 struct bio *lo_biotail;
138 int lo_state;
139 struct semaphore lo_sem;
140 struct mutex lo_ctl_mutex;
141 atomic_t lo_pending;
142 wait_queue_head_t lo_bh_wait;
143
144 struct request_queue *lo_queue;
145
146 const struct lu_env *lo_env;
147 struct cl_io lo_io;
148 struct ll_dio_pages lo_pvec;
149
150 /* data to handle bio for lustre. */
151 struct lo_request_data {
152 struct page *lrd_pages[LLOOP_MAX_SEGMENTS];
153 loff_t lrd_offsets[LLOOP_MAX_SEGMENTS];
154 } lo_requests[1];
155};
156
157/*
158 * Loop flags
159 */
160enum {
161 LO_FLAGS_READ_ONLY = 1,
162};
163
164static int lloop_major;
165#define MAX_LOOP_DEFAULT 16
166static int max_loop = MAX_LOOP_DEFAULT;
167static struct lloop_device *loop_dev;
168static struct gendisk **disks;
169static struct mutex lloop_mutex;
170static void *ll_iocontrol_magic = NULL;
171
172static loff_t get_loop_size(struct lloop_device *lo, struct file *file)
173{
174 loff_t size, offset, loopsize;
175
176 /* Compute loopsize in bytes */
177 size = i_size_read(file->f_mapping->host);
178 offset = lo->lo_offset;
179 loopsize = size - offset;
180 if (lo->lo_sizelimit > 0 && lo->lo_sizelimit < loopsize)
181 loopsize = lo->lo_sizelimit;
182
183 /*
184 * Unfortunately, if we want to do I/O on the device,
185 * the number of 512-byte sectors has to fit into a sector_t.
186 */
187 return loopsize >> 9;
188}
189
190static int do_bio_lustrebacked(struct lloop_device *lo, struct bio *head)
191{
192 const struct lu_env *env = lo->lo_env;
193 struct cl_io *io = &lo->lo_io;
194 struct inode *inode = lo->lo_backing_file->f_dentry->d_inode;
195 struct cl_object *obj = ll_i2info(inode)->lli_clob;
196 pgoff_t offset;
197 int ret;
198 int i;
199 int rw;
200 obd_count page_count = 0;
201 struct bio_vec *bvec;
202 struct bio *bio;
203 ssize_t bytes;
204
205 struct ll_dio_pages *pvec = &lo->lo_pvec;
206 struct page **pages = pvec->ldp_pages;
207 loff_t *offsets = pvec->ldp_offsets;
208
209 truncate_inode_pages(inode->i_mapping, 0);
210
211 /* initialize the IO */
212 memset(io, 0, sizeof(*io));
213 io->ci_obj = obj;
214 ret = cl_io_init(env, io, CIT_MISC, obj);
215 if (ret)
216 return io->ci_result;
217 io->ci_lockreq = CILR_NEVER;
218
219 LASSERT(head != NULL);
220 rw = head->bi_rw;
221 for (bio = head; bio != NULL; bio = bio->bi_next) {
222 LASSERT(rw == bio->bi_rw);
223
224 offset = (pgoff_t)(bio->bi_sector << 9) + lo->lo_offset;
225 bio_for_each_segment(bvec, bio, i) {
226 BUG_ON(bvec->bv_offset != 0);
227 BUG_ON(bvec->bv_len != PAGE_CACHE_SIZE);
228
229 pages[page_count] = bvec->bv_page;
230 offsets[page_count] = offset;
231 page_count++;
232 offset += bvec->bv_len;
233 }
234 LASSERT(page_count <= LLOOP_MAX_SEGMENTS);
235 }
236
237 ll_stats_ops_tally(ll_i2sbi(inode),
238 (rw == WRITE) ? LPROC_LL_BRW_WRITE : LPROC_LL_BRW_READ,
239 page_count);
240
241 pvec->ldp_size = page_count << PAGE_CACHE_SHIFT;
242 pvec->ldp_nr = page_count;
243
244 /* FIXME: in ll_direct_rw_pages, it has to allocate many cl_page{}s to
245 * write those pages into OST. Even worse case is that more pages
246 * would be asked to write out to swap space, and then finally get here
247 * again.
248 * Unfortunately this is NOT easy to fix.
249 * Thoughts on solution:
250 * 0. Define a reserved pool for cl_pages, which could be a list of
251 * pre-allocated cl_pages;
252 * 1. Define a new operation in cl_object_operations{}, says clo_depth,
253 * which measures how many layers for this lustre object. Generally
254 * speaking, the depth would be 2, one for llite, and one for lovsub.
255 * However, for SNS, there will be more since we need additional page
256 * to store parity;
257 * 2. Reserve the # of (page_count * depth) cl_pages from the reserved
258 * pool. Afterwards, the clio would allocate the pages from reserved
259 * pool, this guarantees we neeedn't allocate the cl_pages from
260 * generic cl_page slab cache.
261 * Of course, if there is NOT enough pages in the pool, we might
262 * be asked to write less pages once, this purely depends on
263 * implementation. Anyway, we should be careful to avoid deadlocking.
264 */
265 mutex_lock(&inode->i_mutex);
266 bytes = ll_direct_rw_pages(env, io, rw, inode, pvec);
267 mutex_unlock(&inode->i_mutex);
268 cl_io_fini(env, io);
269 return (bytes == pvec->ldp_size) ? 0 : (int)bytes;
270}
271
272/*
273 * Add bio to back of pending list
274 */
275static void loop_add_bio(struct lloop_device *lo, struct bio *bio)
276{
277 unsigned long flags;
278
279 spin_lock_irqsave(&lo->lo_lock, flags);
280 if (lo->lo_biotail) {
281 lo->lo_biotail->bi_next = bio;
282 lo->lo_biotail = bio;
283 } else
284 lo->lo_bio = lo->lo_biotail = bio;
285 spin_unlock_irqrestore(&lo->lo_lock, flags);
286
287 atomic_inc(&lo->lo_pending);
288 if (waitqueue_active(&lo->lo_bh_wait))
289 wake_up(&lo->lo_bh_wait);
290}
291
292/*
293 * Grab first pending buffer
294 */
295static unsigned int loop_get_bio(struct lloop_device *lo, struct bio **req)
296{
297 struct bio *first;
298 struct bio **bio;
299 unsigned int count = 0;
300 unsigned int page_count = 0;
301 int rw;
302
303 spin_lock_irq(&lo->lo_lock);
304 first = lo->lo_bio;
305 if (unlikely(first == NULL)) {
306 spin_unlock_irq(&lo->lo_lock);
307 return 0;
308 }
309
310 /* TODO: need to split the bio, too bad. */
311 LASSERT(first->bi_vcnt <= LLOOP_MAX_SEGMENTS);
312
313 rw = first->bi_rw;
314 bio = &lo->lo_bio;
315 while (*bio && (*bio)->bi_rw == rw) {
316 CDEBUG(D_INFO, "bio sector %llu size %u count %u vcnt%u \n",
317 (unsigned long long)(*bio)->bi_sector, (*bio)->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 mergable. */
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
341static ll_mrf_ret
342loop_make_request(struct request_queue *q, struct bio *old_bio)
343{
344 struct lloop_device *lo = q->queuedata;
345 int rw = bio_rw(old_bio);
346 int inactive;
347
348 if (!lo)
349 goto err;
350
351 CDEBUG(D_INFO, "submit bio sector %llu size %u\n",
352 (unsigned long long)old_bio->bi_sector, old_bio->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 LL_MRF_RETURN(0);
371err:
372 cfs_bio_io_error(old_bio, old_bio->bi_size);
373 LL_MRF_RETURN(0);
374}
375
376
377static inline void loop_handle_bio(struct lloop_device *lo, struct bio *bio)
378{
379 int ret;
380 ret = do_bio_lustrebacked(lo, bio);
381 while (bio) {
382 struct bio *tmp = bio->bi_next;
383 bio->bi_next = NULL;
384 cfs_bio_endio(bio, bio->bi_size, ret);
385 bio = tmp;
386 }
387}
388
389static inline int loop_active(struct lloop_device *lo)
390{
391 return atomic_read(&lo->lo_pending) ||
392 (lo->lo_state == LLOOP_RUNDOWN);
393}
394
395/*
396 * worker thread that handles reads/writes to file backed loop devices,
397 * to avoid blocking in our make_request_fn.
398 */
399static int loop_thread(void *data)
400{
401 struct lloop_device *lo = data;
402 struct bio *bio;
403 unsigned int count;
404 unsigned long times = 0;
405 unsigned long total_count = 0;
406
407 struct lu_env *env;
408 int refcheck;
409 int ret = 0;
410
411 set_user_nice(current, -20);
412
413 lo->lo_state = LLOOP_BOUND;
414
415 env = cl_env_get(&refcheck);
416 if (IS_ERR(env))
417 GOTO(out, ret = PTR_ERR(env));
418
419 lo->lo_env = env;
420 memset(&lo->lo_pvec, 0, sizeof(lo->lo_pvec));
421 lo->lo_pvec.ldp_pages = lo->lo_requests[0].lrd_pages;
422 lo->lo_pvec.ldp_offsets = lo->lo_requests[0].lrd_offsets;
423
424 /*
425 * up sem, we are running
426 */
427 up(&lo->lo_sem);
428
429 for (;;) {
430 wait_event(lo->lo_bh_wait, loop_active(lo));
431 if (!atomic_read(&lo->lo_pending)) {
432 int exiting = 0;
433 spin_lock_irq(&lo->lo_lock);
434 exiting = (lo->lo_state == LLOOP_RUNDOWN);
435 spin_unlock_irq(&lo->lo_lock);
436 if (exiting)
437 break;
438 }
439
440 bio = NULL;
441 count = loop_get_bio(lo, &bio);
442 if (!count) {
443 CWARN("lloop(minor: %d): missing bio\n", lo->lo_number);
444 continue;
445 }
446
447 total_count += count;
448 if (total_count < count) { /* overflow */
449 total_count = count;
450 times = 1;
451 } else {
452 times++;
453 }
454 if ((times & 127) == 0) {
455 CDEBUG(D_INFO, "total: %lu, count: %lu, avg: %lu\n",
456 total_count, times, total_count / times);
457 }
458
459 LASSERT(bio != NULL);
460 LASSERT(count <= atomic_read(&lo->lo_pending));
461 loop_handle_bio(lo, bio);
462 atomic_sub(count, &lo->lo_pending);
463 }
464 cl_env_put(env, &refcheck);
465
466out:
467 up(&lo->lo_sem);
468 return ret;
469}
470
471static int loop_set_fd(struct lloop_device *lo, struct file *unused,
472 struct block_device *bdev, struct file *file)
473{
474 struct inode *inode;
475 struct address_space *mapping;
476 int lo_flags = 0;
477 int error;
478 loff_t size;
479
480 if (!try_module_get(THIS_MODULE))
481 return -ENODEV;
482
483 error = -EBUSY;
484 if (lo->lo_state != LLOOP_UNBOUND)
485 goto out;
486
487 mapping = file->f_mapping;
488 inode = mapping->host;
489
490 error = -EINVAL;
491 if (!S_ISREG(inode->i_mode) || inode->i_sb->s_magic != LL_SUPER_MAGIC)
492 goto out;
493
494 if (!(file->f_mode & FMODE_WRITE))
495 lo_flags |= LO_FLAGS_READ_ONLY;
496
497 size = get_loop_size(lo, file);
498
499 if ((loff_t)(sector_t)size != size) {
500 error = -EFBIG;
501 goto out;
502 }
503
504 /* remove all pages in cache so as dirty pages not to be existent. */
505 truncate_inode_pages(mapping, 0);
506
507 set_device_ro(bdev, (lo_flags & LO_FLAGS_READ_ONLY) != 0);
508
509 lo->lo_blocksize = PAGE_CACHE_SIZE;
510 lo->lo_device = bdev;
511 lo->lo_flags = lo_flags;
512 lo->lo_backing_file = file;
513 lo->ioctl = NULL;
514 lo->lo_sizelimit = 0;
515 lo->old_gfp_mask = mapping_gfp_mask(mapping);
516 mapping_set_gfp_mask(mapping, lo->old_gfp_mask & ~(__GFP_IO|__GFP_FS));
517
518 lo->lo_bio = lo->lo_biotail = NULL;
519
520 /*
521 * set queue make_request_fn, and add limits based on lower level
522 * device
523 */
524 blk_queue_make_request(lo->lo_queue, loop_make_request);
525 lo->lo_queue->queuedata = lo;
526
527 /* queue parameters */
528 CLASSERT(PAGE_CACHE_SIZE < (1 << (sizeof(unsigned short) * 8)));
529 blk_queue_logical_block_size(lo->lo_queue,
530 (unsigned short)PAGE_CACHE_SIZE);
531 blk_queue_max_hw_sectors(lo->lo_queue,
532 LLOOP_MAX_SEGMENTS << (PAGE_CACHE_SHIFT - 9));
533 blk_queue_max_segments(lo->lo_queue, LLOOP_MAX_SEGMENTS);
534
535 set_capacity(disks[lo->lo_number], size);
536 bd_set_size(bdev, size << 9);
537
538 set_blocksize(bdev, lo->lo_blocksize);
539
540 kthread_run(loop_thread, lo, "lloop%d", lo->lo_number);
541 down(&lo->lo_sem);
542 return 0;
543
544out:
545 /* This is safe: open() is still holding a reference. */
546 module_put(THIS_MODULE);
547 return error;
548}
549
550static int loop_clr_fd(struct lloop_device *lo, struct block_device *bdev,
551 int count)
552{
553 struct file *filp = lo->lo_backing_file;
554 int gfp = lo->old_gfp_mask;
555
556 if (lo->lo_state != LLOOP_BOUND)
557 return -ENXIO;
558
559 if (lo->lo_refcnt > count) /* we needed one fd for the ioctl */
560 return -EBUSY;
561
562 if (filp == NULL)
563 return -EINVAL;
564
565 spin_lock_irq(&lo->lo_lock);
566 lo->lo_state = LLOOP_RUNDOWN;
567 spin_unlock_irq(&lo->lo_lock);
568 wake_up(&lo->lo_bh_wait);
569
570 down(&lo->lo_sem);
571 lo->lo_backing_file = NULL;
572 lo->ioctl = NULL;
573 lo->lo_device = NULL;
574 lo->lo_offset = 0;
575 lo->lo_sizelimit = 0;
576 lo->lo_flags = 0;
577 ll_invalidate_bdev(bdev, 0);
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
588static 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
599static int 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 return 0;
608}
609
610/* lloop device node's ioctl function. */
611static int lo_ioctl(struct block_device *bdev, fmode_t mode,
612 unsigned int cmd, unsigned long arg)
613{
614 struct lloop_device *lo = bdev->bd_disk->private_data;
615 struct inode *inode = NULL;
616 int err = 0;
617
618 mutex_lock(&lloop_mutex);
619 switch (cmd) {
620 case LL_IOC_LLOOP_DETACH: {
621 err = loop_clr_fd(lo, bdev, 2);
622 if (err == 0)
623 ll_blkdev_put(bdev, 0); /* grabbed in LLOOP_ATTACH */
624 break;
625 }
626
627 case LL_IOC_LLOOP_INFO: {
628 struct lu_fid fid;
629
630 LASSERT(lo->lo_backing_file != NULL);
631 if (inode == NULL)
632 inode = lo->lo_backing_file->f_dentry->d_inode;
633 if (lo->lo_state == LLOOP_BOUND)
634 fid = ll_i2info(inode)->lli_fid;
635 else
636 fid_zero(&fid);
637
638 if (copy_to_user((struct lu_fid *)arg, &fid, sizeof(fid)))
639 err = -EFAULT;
640 break;
641 }
642
643 default:
644 err = -EINVAL;
645 break;
646 }
647 mutex_unlock(&lloop_mutex);
648
649 return err;
650}
651
652static struct block_device_operations lo_fops = {
653 .owner = THIS_MODULE,
654 .open = lo_open,
655 .release = lo_release,
656 .ioctl = lo_ioctl,
657};
658
659/* dynamic iocontrol callback.
660 * This callback is registered in lloop_init and will be called by
661 * ll_iocontrol_call.
662 *
663 * This is a llite regular file ioctl function. It takes the responsibility
664 * of attaching or detaching a file by a lloop's device numner.
665 */
666static enum llioc_iter lloop_ioctl(struct inode *unused, struct file *file,
667 unsigned int cmd, unsigned long arg,
668 void *magic, int *rcp)
669{
670 struct lloop_device *lo = NULL;
671 struct block_device *bdev = NULL;
672 int err = 0;
673 dev_t dev;
674
675 if (magic != ll_iocontrol_magic)
676 return LLIOC_CONT;
677
678 if (disks == NULL)
679 GOTO(out1, err = -ENODEV);
680
681 CWARN("Enter llop_ioctl\n");
682
683 mutex_lock(&lloop_mutex);
684 switch (cmd) {
685 case LL_IOC_LLOOP_ATTACH: {
686 struct lloop_device *lo_free = NULL;
687 int i;
688
689 for (i = 0; i < max_loop; i++, lo = NULL) {
690 lo = &loop_dev[i];
691 if (lo->lo_state == LLOOP_UNBOUND) {
692 if (!lo_free)
693 lo_free = lo;
694 continue;
695 }
696 if (lo->lo_backing_file->f_dentry->d_inode ==
697 file->f_dentry->d_inode)
698 break;
699 }
700 if (lo || !lo_free)
701 GOTO(out, err = -EBUSY);
702
703 lo = lo_free;
704 dev = MKDEV(lloop_major, lo->lo_number);
705
706 /* quit if the used pointer is writable */
707 if (put_user((long)old_encode_dev(dev), (long*)arg))
708 GOTO(out, err = -EFAULT);
709
710 bdev = blkdev_get_by_dev(dev, file->f_mode, NULL);
711 if (IS_ERR(bdev))
712 GOTO(out, err = PTR_ERR(bdev));
713
714 get_file(file);
715 err = loop_set_fd(lo, NULL, bdev, file);
716 if (err) {
717 fput(file);
718 ll_blkdev_put(bdev, 0);
719 }
720
721 break;
722 }
723
724 case LL_IOC_LLOOP_DETACH_BYDEV: {
725 int minor;
726
727 dev = old_decode_dev(arg);
728 if (MAJOR(dev) != lloop_major)
729 GOTO(out, err = -EINVAL);
730
731 minor = MINOR(dev);
732 if (minor > max_loop - 1)
733 GOTO(out, err = -EINVAL);
734
735 lo = &loop_dev[minor];
736 if (lo->lo_state != LLOOP_BOUND)
737 GOTO(out, err = -EINVAL);
738
739 bdev = lo->lo_device;
740 err = loop_clr_fd(lo, bdev, 1);
741 if (err == 0)
742 ll_blkdev_put(bdev, 0); /* grabbed in LLOOP_ATTACH */
743
744 break;
745 }
746
747 default:
748 err = -EINVAL;
749 break;
750 }
751
752out:
753 mutex_unlock(&lloop_mutex);
754out1:
755 if (rcp)
756 *rcp = err;
757 return LLIOC_STOP;
758}
759
760static int __init lloop_init(void)
761{
762 int i;
763 unsigned int cmdlist[] = {
764 LL_IOC_LLOOP_ATTACH,
765 LL_IOC_LLOOP_DETACH_BYDEV,
766 };
767
768 if (max_loop < 1 || max_loop > 256) {
769 max_loop = MAX_LOOP_DEFAULT;
770 CWARN("lloop: invalid max_loop (must be between"
771 " 1 and 256), using default (%u)\n", max_loop);
772 }
773
774 lloop_major = register_blkdev(0, "lloop");
775 if (lloop_major < 0)
776 return -EIO;
777
778 CDEBUG(D_CONFIG, "registered lloop major %d with %u minors\n",
779 lloop_major, max_loop);
780
781 ll_iocontrol_magic = ll_iocontrol_register(lloop_ioctl, 2, cmdlist);
782 if (ll_iocontrol_magic == NULL)
783 goto out_mem1;
784
785 OBD_ALLOC_WAIT(loop_dev, max_loop * sizeof(*loop_dev));
786 if (!loop_dev)
787 goto out_mem1;
788
789 OBD_ALLOC_WAIT(disks, max_loop * sizeof(*disks));
790 if (!disks)
791 goto out_mem2;
792
793 for (i = 0; i < max_loop; i++) {
794 disks[i] = alloc_disk(1);
795 if (!disks[i])
796 goto out_mem3;
797 }
798
799 mutex_init(&lloop_mutex);
800
801 for (i = 0; i < max_loop; i++) {
802 struct lloop_device *lo = &loop_dev[i];
803 struct gendisk *disk = disks[i];
804
805 lo->lo_queue = blk_alloc_queue(GFP_KERNEL);
806 if (!lo->lo_queue)
807 goto out_mem4;
808
809 mutex_init(&lo->lo_ctl_mutex);
810 sema_init(&lo->lo_sem, 0);
811 init_waitqueue_head(&lo->lo_bh_wait);
812 lo->lo_number = i;
813 spin_lock_init(&lo->lo_lock);
814 disk->major = lloop_major;
815 disk->first_minor = i;
816 disk->fops = &lo_fops;
817 sprintf(disk->disk_name, "lloop%d", i);
818 disk->private_data = lo;
819 disk->queue = lo->lo_queue;
820 }
821
822 /* We cannot fail after we call this, so another loop!*/
823 for (i = 0; i < max_loop; i++)
824 add_disk(disks[i]);
825 return 0;
826
827out_mem4:
828 while (i--)
829 blk_cleanup_queue(loop_dev[i].lo_queue);
830 i = max_loop;
831out_mem3:
832 while (i--)
833 put_disk(disks[i]);
834 OBD_FREE(disks, max_loop * sizeof(*disks));
835out_mem2:
836 OBD_FREE(loop_dev, max_loop * sizeof(*loop_dev));
837out_mem1:
838 unregister_blkdev(lloop_major, "lloop");
839 ll_iocontrol_unregister(ll_iocontrol_magic);
840 CERROR("lloop: ran out of memory\n");
841 return -ENOMEM;
842}
843
844static void lloop_exit(void)
845{
846 int i;
847
848 ll_iocontrol_unregister(ll_iocontrol_magic);
849 for (i = 0; i < max_loop; i++) {
850 del_gendisk(disks[i]);
851 blk_cleanup_queue(loop_dev[i].lo_queue);
852 put_disk(disks[i]);
853 }
854 if (ll_unregister_blkdev(lloop_major, "lloop"))
855 CWARN("lloop: cannot unregister blkdev\n");
856 else
857 CDEBUG(D_CONFIG, "unregistered lloop major %d\n", lloop_major);
858
859 OBD_FREE(disks, max_loop * sizeof(*disks));
860 OBD_FREE(loop_dev, max_loop * sizeof(*loop_dev));
861}
862
863module_init(lloop_init);
864module_exit(lloop_exit);
865
866CFS_MODULE_PARM(max_loop, "i", int, 0444, "maximum of lloop_device");
867MODULE_AUTHOR("Sun Microsystems, Inc. <http://www.lustre.org/>");
868MODULE_DESCRIPTION("Lustre virtual block device");
869MODULE_LICENSE("GPL");
Linux kernel - Deliverables
This page took 0.092359 seconds and 5 git commands to generate.