projects / deliverable / linux.git / blame
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (incremental) | history | HEAD
staging: lustre: remove ll_mrf_ret
[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>
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;
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 */
159enum {
160 LO_FLAGS_READ_ONLY = 1,
161};
162
163static int lloop_major;
164#define MAX_LOOP_DEFAULT 16
165static int max_loop = MAX_LOOP_DEFAULT;
166static struct lloop_device *loop_dev;
167static struct gendisk **disks;
168static struct mutex lloop_mutex;
169static void *ll_iocontrol_magic = NULL;
170
171static 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
189static 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 i;
198 int rw;
199 obd_count page_count = 0;
200 struct bio_vec *bvec;
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_sector << 9) + lo->lo_offset;
224 bio_for_each_segment(bvec, bio, i) {
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 neeedn'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 */
274static 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 */
294static 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_sector, (*bio)->bi_size,
317 page_count, (*bio)->bi_vcnt);
318 if (page_count + (*bio)->bi_vcnt > LLOOP_MAX_SEGMENTS)
319 break;
320
321
322 page_count += (*bio)->bi_vcnt;
323 count++;
324 bio = &(*bio)->bi_next;
325 }
326 if (*bio) {
327 /* Some of bios can't be mergable. */
328 lo->lo_bio = *bio;
329 *bio = NULL;
330 } else {
331 /* Hit the end of queue */
332 lo->lo_biotail = NULL;
333 lo->lo_bio = NULL;
334 }
335 *req = first;
336 spin_unlock_irq(&lo->lo_lock);
337 return count;
338}
339
3ec65cb3 340static void loop_make_request(struct request_queue *q, struct bio *old_bio)
d7e09d03
PT		 341{
342 struct lloop_device *lo = q->queuedata;
343 int rw = bio_rw(old_bio);
344 int inactive;
345
346 if (!lo)
347 goto err;
348
349 CDEBUG(D_INFO, "submit bio sector %llu size %u\n",
350 (unsigned long long)old_bio->bi_sector, old_bio->bi_size);
351
352 spin_lock_irq(&lo->lo_lock);
353 inactive = (lo->lo_state != LLOOP_BOUND);
354 spin_unlock_irq(&lo->lo_lock);
355 if (inactive)
356 goto err;
357
358 if (rw == WRITE) {
359 if (lo->lo_flags & LO_FLAGS_READ_ONLY)
360 goto err;
361 } else if (rw == READA) {
362 rw = READ;
363 } else if (rw != READ) {
364 CERROR("lloop: unknown command (%x)\n", rw);
365 goto err;
366 }
367 loop_add_bio(lo, old_bio);
368 LL_MRF_RETURN(0);
369err:
370 cfs_bio_io_error(old_bio, old_bio->bi_size);
371 LL_MRF_RETURN(0);
372}
373
374
375static inline void loop_handle_bio(struct lloop_device *lo, struct bio *bio)
376{
377 int ret;
378 ret = do_bio_lustrebacked(lo, bio);
379 while (bio) {
380 struct bio *tmp = bio->bi_next;
381 bio->bi_next = NULL;
382 cfs_bio_endio(bio, bio->bi_size, ret);
383 bio = tmp;
384 }
385}
386
387static inline int loop_active(struct lloop_device *lo)
388{
389 return atomic_read(&lo->lo_pending) ||
390 (lo->lo_state == LLOOP_RUNDOWN);
391}
392
393/*
394 * worker thread that handles reads/writes to file backed loop devices,
395 * to avoid blocking in our make_request_fn.
396 */
397static int loop_thread(void *data)
398{
399 struct lloop_device *lo = data;
400 struct bio *bio;
401 unsigned int count;
402 unsigned long times = 0;
403 unsigned long total_count = 0;
404
405 struct lu_env *env;
406 int refcheck;
407 int ret = 0;
408
409 set_user_nice(current, -20);
410
411 lo->lo_state = LLOOP_BOUND;
412
413 env = cl_env_get(&refcheck);
414 if (IS_ERR(env))
415 GOTO(out, ret = PTR_ERR(env));
416
417 lo->lo_env = env;
418 memset(&lo->lo_pvec, 0, sizeof(lo->lo_pvec));
419 lo->lo_pvec.ldp_pages = lo->lo_requests[0].lrd_pages;
420 lo->lo_pvec.ldp_offsets = lo->lo_requests[0].lrd_offsets;
421
422 /*
423 * up sem, we are running
424 */
425 up(&lo->lo_sem);
426
427 for (;;) {
428 wait_event(lo->lo_bh_wait, loop_active(lo));
429 if (!atomic_read(&lo->lo_pending)) {
430 int exiting = 0;
431 spin_lock_irq(&lo->lo_lock);
432 exiting = (lo->lo_state == LLOOP_RUNDOWN);
433 spin_unlock_irq(&lo->lo_lock);
434 if (exiting)
435 break;
436 }
437
438 bio = NULL;
439 count = loop_get_bio(lo, &bio);
440 if (!count) {
441 CWARN("lloop(minor: %d): missing bio\n", lo->lo_number);
442 continue;
443 }
444
445 total_count += count;
446 if (total_count < count) { /* overflow */
447 total_count = count;
448 times = 1;
449 } else {
450 times++;
451 }
452 if ((times & 127) == 0) {
453 CDEBUG(D_INFO, "total: %lu, count: %lu, avg: %lu\n",
454 total_count, times, total_count / times);
455 }
456
457 LASSERT(bio != NULL);
458 LASSERT(count <= atomic_read(&lo->lo_pending));
459 loop_handle_bio(lo, bio);
460 atomic_sub(count, &lo->lo_pending);
461 }
462 cl_env_put(env, &refcheck);
463
464out:
465 up(&lo->lo_sem);
466 return ret;
467}
468
469static int loop_set_fd(struct lloop_device *lo, struct file *unused,
470 struct block_device *bdev, struct file *file)
471{
472 struct inode *inode;
473 struct address_space *mapping;
474 int lo_flags = 0;
475 int error;
476 loff_t size;
477
478 if (!try_module_get(THIS_MODULE))
479 return -ENODEV;
480
481 error = -EBUSY;
482 if (lo->lo_state != LLOOP_UNBOUND)
483 goto out;
484
485 mapping = file->f_mapping;
486 inode = mapping->host;
487
488 error = -EINVAL;
489 if (!S_ISREG(inode->i_mode) || inode->i_sb->s_magic != LL_SUPER_MAGIC)
490 goto out;
491
492 if (!(file->f_mode & FMODE_WRITE))
493 lo_flags |= LO_FLAGS_READ_ONLY;
494
495 size = get_loop_size(lo, file);
496
497 if ((loff_t)(sector_t)size != size) {
498 error = -EFBIG;
499 goto out;
500 }
501
502 /* remove all pages in cache so as dirty pages not to be existent. */
503 truncate_inode_pages(mapping, 0);
504
505 set_device_ro(bdev, (lo_flags & LO_FLAGS_READ_ONLY) != 0);
506
507 lo->lo_blocksize = PAGE_CACHE_SIZE;
508 lo->lo_device = bdev;
509 lo->lo_flags = lo_flags;
510 lo->lo_backing_file = file;
511 lo->ioctl = NULL;
512 lo->lo_sizelimit = 0;
513 lo->old_gfp_mask = mapping_gfp_mask(mapping);
514 mapping_set_gfp_mask(mapping, lo->old_gfp_mask & ~(__GFP_IO|__GFP_FS));
515
516 lo->lo_bio = lo->lo_biotail = NULL;
517
518 /*
519 * set queue make_request_fn, and add limits based on lower level
520 * device
521 */
522 blk_queue_make_request(lo->lo_queue, loop_make_request);
523 lo->lo_queue->queuedata = lo;
524
525 /* queue parameters */
526 CLASSERT(PAGE_CACHE_SIZE < (1 << (sizeof(unsigned short) * 8)));
527 blk_queue_logical_block_size(lo->lo_queue,
528 (unsigned short)PAGE_CACHE_SIZE);
529 blk_queue_max_hw_sectors(lo->lo_queue,
530 LLOOP_MAX_SEGMENTS << (PAGE_CACHE_SHIFT - 9));
531 blk_queue_max_segments(lo->lo_queue, LLOOP_MAX_SEGMENTS);
532
533 set_capacity(disks[lo->lo_number], size);
534 bd_set_size(bdev, size << 9);
535
536 set_blocksize(bdev, lo->lo_blocksize);
537
538 kthread_run(loop_thread, lo, "lloop%d", lo->lo_number);
539 down(&lo->lo_sem);
540 return 0;
541
542out:
543 /* This is safe: open() is still holding a reference. */
544 module_put(THIS_MODULE);
545 return error;
546}
547
548static int loop_clr_fd(struct lloop_device *lo, struct block_device *bdev,
549 int count)
550{
551 struct file *filp = lo->lo_backing_file;
552 int gfp = lo->old_gfp_mask;
553
554 if (lo->lo_state != LLOOP_BOUND)
555 return -ENXIO;
556
557 if (lo->lo_refcnt > count) /* we needed one fd for the ioctl */
558 return -EBUSY;
559
560 if (filp == NULL)
561 return -EINVAL;
562
563 spin_lock_irq(&lo->lo_lock);
564 lo->lo_state = LLOOP_RUNDOWN;
565 spin_unlock_irq(&lo->lo_lock);
566 wake_up(&lo->lo_bh_wait);
567
568 down(&lo->lo_sem);
569 lo->lo_backing_file = NULL;
570 lo->ioctl = NULL;
571 lo->lo_device = NULL;
572 lo->lo_offset = 0;
573 lo->lo_sizelimit = 0;
574 lo->lo_flags = 0;
db7392c2 575 invalidate_bdev(bdev);
d7e09d03
PT		 576 set_capacity(disks[lo->lo_number], 0);
577 bd_set_size(bdev, 0);
578 mapping_set_gfp_mask(filp->f_mapping, gfp);
579 lo->lo_state = LLOOP_UNBOUND;
580 fput(filp);
581 /* This is safe: open() is still holding a reference. */
582 module_put(THIS_MODULE);
583 return 0;
584}
585
586static int lo_open(struct block_device *bdev, fmode_t mode)
587{
588 struct lloop_device *lo = bdev->bd_disk->private_data;
589
590 mutex_lock(&lo->lo_ctl_mutex);
591 lo->lo_refcnt++;
592 mutex_unlock(&lo->lo_ctl_mutex);
593
594 return 0;
595}
596
f1e2f53e 597static void lo_release(struct gendisk *disk, fmode_t mode)
d7e09d03
PT		 598{
599 struct lloop_device *lo = disk->private_data;
600
601 mutex_lock(&lo->lo_ctl_mutex);
602 --lo->lo_refcnt;
603 mutex_unlock(&lo->lo_ctl_mutex);
d7e09d03
PT		 604}
605
606/* lloop device node's ioctl function. */
607static int lo_ioctl(struct block_device *bdev, fmode_t mode,
608 unsigned int cmd, unsigned long arg)
609{
610 struct lloop_device *lo = bdev->bd_disk->private_data;
611 struct inode *inode = NULL;
612 int err = 0;
613
614 mutex_lock(&lloop_mutex);
615 switch (cmd) {
616 case LL_IOC_LLOOP_DETACH: {
617 err = loop_clr_fd(lo, bdev, 2);
618 if (err == 0)
418690b6 619 blkdev_put(bdev, 0); /* grabbed in LLOOP_ATTACH */
d7e09d03
PT		 620 break;
621 }
622
623 case LL_IOC_LLOOP_INFO: {
624 struct lu_fid fid;
625
626 LASSERT(lo->lo_backing_file != NULL);
627 if (inode == NULL)
628 inode = lo->lo_backing_file->f_dentry->d_inode;
629 if (lo->lo_state == LLOOP_BOUND)
630 fid = ll_i2info(inode)->lli_fid;
631 else
632 fid_zero(&fid);
633
634 if (copy_to_user((struct lu_fid *)arg, &fid, sizeof(fid)))
635 err = -EFAULT;
636 break;
637 }
638
639 default:
640 err = -EINVAL;
641 break;
642 }
643 mutex_unlock(&lloop_mutex);
644
645 return err;
646}
647
648static struct block_device_operations lo_fops = {
649 .owner = THIS_MODULE,
650 .open = lo_open,
651 .release = lo_release,
652 .ioctl = lo_ioctl,
653};
654
655/* dynamic iocontrol callback.
656 * This callback is registered in lloop_init and will be called by
657 * ll_iocontrol_call.
658 *
659 * This is a llite regular file ioctl function. It takes the responsibility
660 * of attaching or detaching a file by a lloop's device numner.
661 */
662static enum llioc_iter lloop_ioctl(struct inode *unused, struct file *file,
663 unsigned int cmd, unsigned long arg,
664 void *magic, int *rcp)
665{
666 struct lloop_device *lo = NULL;
667 struct block_device *bdev = NULL;
668 int err = 0;
669 dev_t dev;
670
671 if (magic != ll_iocontrol_magic)
672 return LLIOC_CONT;
673
674 if (disks == NULL)
675 GOTO(out1, err = -ENODEV);
676
677 CWARN("Enter llop_ioctl\n");
678
679 mutex_lock(&lloop_mutex);
680 switch (cmd) {
681 case LL_IOC_LLOOP_ATTACH: {
682 struct lloop_device *lo_free = NULL;
683 int i;
684
685 for (i = 0; i < max_loop; i++, lo = NULL) {
686 lo = &loop_dev[i];
687 if (lo->lo_state == LLOOP_UNBOUND) {
688 if (!lo_free)
689 lo_free = lo;
690 continue;
691 }
692 if (lo->lo_backing_file->f_dentry->d_inode ==
693 file->f_dentry->d_inode)
694 break;
695 }
696 if (lo || !lo_free)
697 GOTO(out, err = -EBUSY);
698
699 lo = lo_free;
700 dev = MKDEV(lloop_major, lo->lo_number);
701
702 /* quit if the used pointer is writable */
703 if (put_user((long)old_encode_dev(dev), (long*)arg))
704 GOTO(out, err = -EFAULT);
705
706 bdev = blkdev_get_by_dev(dev, file->f_mode, NULL);
707 if (IS_ERR(bdev))
708 GOTO(out, err = PTR_ERR(bdev));
709
710 get_file(file);
711 err = loop_set_fd(lo, NULL, bdev, file);
712 if (err) {
713 fput(file);
418690b6 714 blkdev_put(bdev, 0);
d7e09d03
PT		 715 }
716
717 break;
718 }
719
720 case LL_IOC_LLOOP_DETACH_BYDEV: {
721 int minor;
722
723 dev = old_decode_dev(arg);
724 if (MAJOR(dev) != lloop_major)
725 GOTO(out, err = -EINVAL);
726
727 minor = MINOR(dev);
728 if (minor > max_loop - 1)
729 GOTO(out, err = -EINVAL);
730
731 lo = &loop_dev[minor];
732 if (lo->lo_state != LLOOP_BOUND)
733 GOTO(out, err = -EINVAL);
734
735 bdev = lo->lo_device;
736 err = loop_clr_fd(lo, bdev, 1);
737 if (err == 0)
418690b6 738 blkdev_put(bdev, 0); /* grabbed in LLOOP_ATTACH */
d7e09d03
PT		 739
740 break;
741 }
742
743 default:
744 err = -EINVAL;
745 break;
746 }
747
748out:
749 mutex_unlock(&lloop_mutex);
750out1:
751 if (rcp)
752 *rcp = err;
753 return LLIOC_STOP;
754}
755
756static int __init lloop_init(void)
757{
758 int i;
759 unsigned int cmdlist[] = {
760 LL_IOC_LLOOP_ATTACH,
761 LL_IOC_LLOOP_DETACH_BYDEV,
762 };
763
764 if (max_loop < 1 || max_loop > 256) {
765 max_loop = MAX_LOOP_DEFAULT;
766 CWARN("lloop: invalid max_loop (must be between"
767 " 1 and 256), using default (%u)\n", max_loop);
768 }
769
770 lloop_major = register_blkdev(0, "lloop");
771 if (lloop_major < 0)
772 return -EIO;
773
774 CDEBUG(D_CONFIG, "registered lloop major %d with %u minors\n",
775 lloop_major, max_loop);
776
777 ll_iocontrol_magic = ll_iocontrol_register(lloop_ioctl, 2, cmdlist);
778 if (ll_iocontrol_magic == NULL)
779 goto out_mem1;
780
781 OBD_ALLOC_WAIT(loop_dev, max_loop * sizeof(*loop_dev));
782 if (!loop_dev)
783 goto out_mem1;
784
785 OBD_ALLOC_WAIT(disks, max_loop * sizeof(*disks));
786 if (!disks)
787 goto out_mem2;
788
789 for (i = 0; i < max_loop; i++) {
790 disks[i] = alloc_disk(1);
791 if (!disks[i])
792 goto out_mem3;
793 }
794
795 mutex_init(&lloop_mutex);
796
797 for (i = 0; i < max_loop; i++) {
798 struct lloop_device *lo = &loop_dev[i];
799 struct gendisk *disk = disks[i];
800
801 lo->lo_queue = blk_alloc_queue(GFP_KERNEL);
802 if (!lo->lo_queue)
803 goto out_mem4;
804
805 mutex_init(&lo->lo_ctl_mutex);
806 sema_init(&lo->lo_sem, 0);
807 init_waitqueue_head(&lo->lo_bh_wait);
808 lo->lo_number = i;
809 spin_lock_init(&lo->lo_lock);
810 disk->major = lloop_major;
811 disk->first_minor = i;
812 disk->fops = &lo_fops;
813 sprintf(disk->disk_name, "lloop%d", i);
814 disk->private_data = lo;
815 disk->queue = lo->lo_queue;
816 }
817
818 /* We cannot fail after we call this, so another loop!*/
819 for (i = 0; i < max_loop; i++)
820 add_disk(disks[i]);
821 return 0;
822
823out_mem4:
824 while (i--)
825 blk_cleanup_queue(loop_dev[i].lo_queue);
826 i = max_loop;
827out_mem3:
828 while (i--)
829 put_disk(disks[i]);
830 OBD_FREE(disks, max_loop * sizeof(*disks));
831out_mem2:
832 OBD_FREE(loop_dev, max_loop * sizeof(*loop_dev));
833out_mem1:
834 unregister_blkdev(lloop_major, "lloop");
835 ll_iocontrol_unregister(ll_iocontrol_magic);
836 CERROR("lloop: ran out of memory\n");
837 return -ENOMEM;
838}
839
840static void lloop_exit(void)
841{
842 int i;
843
844 ll_iocontrol_unregister(ll_iocontrol_magic);
845 for (i = 0; i < max_loop; i++) {
846 del_gendisk(disks[i]);
847 blk_cleanup_queue(loop_dev[i].lo_queue);
848 put_disk(disks[i]);
849 }
49126e47
XZ		 850
851 unregister_blkdev(lloop_major, "lloop");
d7e09d03
PT		 852
853 OBD_FREE(disks, max_loop * sizeof(*disks));
854 OBD_FREE(loop_dev, max_loop * sizeof(*loop_dev));
855}
856
857module_init(lloop_init);
858module_exit(lloop_exit);
859
860CFS_MODULE_PARM(max_loop, "i", int, 0444, "maximum of lloop_device");
861MODULE_AUTHOR("Sun Microsystems, Inc. <http://www.lustre.org/>");
862MODULE_DESCRIPTION("Lustre virtual block device");
863MODULE_LICENSE("GPL");
Linux kernel - Deliverables
This page took 0.138436 seconds and 5 git commands to generate.