Line data Source code
1 : // SPDX-License-Identifier: GPL-2.0
2 : /*
3 : * gendisk handling
4 : *
5 : * Portions Copyright (C) 2020 Christoph Hellwig
6 : */
7 :
8 : #include <linux/module.h>
9 : #include <linux/ctype.h>
10 : #include <linux/fs.h>
11 : #include <linux/kdev_t.h>
12 : #include <linux/kernel.h>
13 : #include <linux/blkdev.h>
14 : #include <linux/backing-dev.h>
15 : #include <linux/init.h>
16 : #include <linux/spinlock.h>
17 : #include <linux/proc_fs.h>
18 : #include <linux/seq_file.h>
19 : #include <linux/slab.h>
20 : #include <linux/kmod.h>
21 : #include <linux/major.h>
22 : #include <linux/mutex.h>
23 : #include <linux/idr.h>
24 : #include <linux/log2.h>
25 : #include <linux/pm_runtime.h>
26 : #include <linux/badblocks.h>
27 : #include <linux/part_stat.h>
28 : #include "blk-throttle.h"
29 :
30 : #include "blk.h"
31 : #include "blk-mq-sched.h"
32 : #include "blk-rq-qos.h"
33 : #include "blk-cgroup.h"
34 :
35 : static struct kobject *block_depr;
36 :
37 : /*
38 : * Unique, monotonically increasing sequential number associated with block
39 : * devices instances (i.e. incremented each time a device is attached).
40 : * Associating uevents with block devices in userspace is difficult and racy:
41 : * the uevent netlink socket is lossy, and on slow and overloaded systems has
42 : * a very high latency.
43 : * Block devices do not have exclusive owners in userspace, any process can set
44 : * one up (e.g. loop devices). Moreover, device names can be reused (e.g. loop0
45 : * can be reused again and again).
46 : * A userspace process setting up a block device and watching for its events
47 : * cannot thus reliably tell whether an event relates to the device it just set
48 : * up or another earlier instance with the same name.
49 : * This sequential number allows userspace processes to solve this problem, and
50 : * uniquely associate an uevent to the lifetime to a device.
51 : */
52 : static atomic64_t diskseq;
53 :
54 : /* for extended dynamic devt allocation, currently only one major is used */
55 : #define NR_EXT_DEVT (1 << MINORBITS)
56 : static DEFINE_IDA(ext_devt_ida);
57 :
58 0 : void set_capacity(struct gendisk *disk, sector_t sectors)
59 : {
60 0 : bdev_set_nr_sectors(disk->part0, sectors);
61 0 : }
62 : EXPORT_SYMBOL(set_capacity);
63 :
64 : /*
65 : * Set disk capacity and notify if the size is not currently zero and will not
66 : * be set to zero. Returns true if a uevent was sent, otherwise false.
67 : */
68 0 : bool set_capacity_and_notify(struct gendisk *disk, sector_t size)
69 : {
70 0 : sector_t capacity = get_capacity(disk);
71 0 : char *envp[] = { "RESIZE=1", NULL };
72 :
73 0 : set_capacity(disk, size);
74 :
75 : /*
76 : * Only print a message and send a uevent if the gendisk is user visible
77 : * and alive. This avoids spamming the log and udev when setting the
78 : * initial capacity during probing.
79 : */
80 0 : if (size == capacity ||
81 0 : !disk_live(disk) ||
82 0 : (disk->flags & GENHD_FL_HIDDEN))
83 : return false;
84 :
85 0 : pr_info("%s: detected capacity change from %lld to %lld\n",
86 : disk->disk_name, capacity, size);
87 :
88 : /*
89 : * Historically we did not send a uevent for changes to/from an empty
90 : * device.
91 : */
92 0 : if (!capacity || !size)
93 : return false;
94 0 : kobject_uevent_env(&disk_to_dev(disk)->kobj, KOBJ_CHANGE, envp);
95 0 : return true;
96 : }
97 : EXPORT_SYMBOL_GPL(set_capacity_and_notify);
98 :
99 0 : static void part_stat_read_all(struct block_device *part,
100 : struct disk_stats *stat)
101 : {
102 : int cpu;
103 :
104 0 : memset(stat, 0, sizeof(struct disk_stats));
105 0 : for_each_possible_cpu(cpu) {
106 0 : struct disk_stats *ptr = per_cpu_ptr(part->bd_stats, cpu);
107 : int group;
108 :
109 0 : for (group = 0; group < NR_STAT_GROUPS; group++) {
110 0 : stat->nsecs[group] += ptr->nsecs[group];
111 0 : stat->sectors[group] += ptr->sectors[group];
112 0 : stat->ios[group] += ptr->ios[group];
113 0 : stat->merges[group] += ptr->merges[group];
114 : }
115 :
116 0 : stat->io_ticks += ptr->io_ticks;
117 : }
118 0 : }
119 :
120 : static unsigned int part_in_flight(struct block_device *part)
121 : {
122 : unsigned int inflight = 0;
123 : int cpu;
124 :
125 0 : for_each_possible_cpu(cpu) {
126 0 : inflight += part_stat_local_read_cpu(part, in_flight[0], cpu) +
127 0 : part_stat_local_read_cpu(part, in_flight[1], cpu);
128 : }
129 0 : if ((int)inflight < 0)
130 0 : inflight = 0;
131 :
132 : return inflight;
133 : }
134 :
135 : static void part_in_flight_rw(struct block_device *part,
136 : unsigned int inflight[2])
137 : {
138 : int cpu;
139 :
140 0 : inflight[0] = 0;
141 0 : inflight[1] = 0;
142 0 : for_each_possible_cpu(cpu) {
143 0 : inflight[0] += part_stat_local_read_cpu(part, in_flight[0], cpu);
144 0 : inflight[1] += part_stat_local_read_cpu(part, in_flight[1], cpu);
145 : }
146 0 : if ((int)inflight[0] < 0)
147 0 : inflight[0] = 0;
148 0 : if ((int)inflight[1] < 0)
149 0 : inflight[1] = 0;
150 : }
151 :
152 : /*
153 : * Can be deleted altogether. Later.
154 : *
155 : */
156 : #define BLKDEV_MAJOR_HASH_SIZE 255
157 : static struct blk_major_name {
158 : struct blk_major_name *next;
159 : int major;
160 : char name[16];
161 : #ifdef CONFIG_BLOCK_LEGACY_AUTOLOAD
162 : void (*probe)(dev_t devt);
163 : #endif
164 : } *major_names[BLKDEV_MAJOR_HASH_SIZE];
165 : static DEFINE_MUTEX(major_names_lock);
166 : static DEFINE_SPINLOCK(major_names_spinlock);
167 :
168 : /* index in the above - for now: assume no multimajor ranges */
169 : static inline int major_to_index(unsigned major)
170 : {
171 1 : return major % BLKDEV_MAJOR_HASH_SIZE;
172 : }
173 :
174 : #ifdef CONFIG_PROC_FS
175 0 : void blkdev_show(struct seq_file *seqf, off_t offset)
176 : {
177 : struct blk_major_name *dp;
178 :
179 0 : spin_lock(&major_names_spinlock);
180 0 : for (dp = major_names[major_to_index(offset)]; dp; dp = dp->next)
181 0 : if (dp->major == offset)
182 0 : seq_printf(seqf, "%3d %s\n", dp->major, dp->name);
183 0 : spin_unlock(&major_names_spinlock);
184 0 : }
185 : #endif /* CONFIG_PROC_FS */
186 :
187 : /**
188 : * __register_blkdev - register a new block device
189 : *
190 : * @major: the requested major device number [1..BLKDEV_MAJOR_MAX-1]. If
191 : * @major = 0, try to allocate any unused major number.
192 : * @name: the name of the new block device as a zero terminated string
193 : * @probe: pre-devtmpfs / pre-udev callback used to create disks when their
194 : * pre-created device node is accessed. When a probe call uses
195 : * add_disk() and it fails the driver must cleanup resources. This
196 : * interface may soon be removed.
197 : *
198 : * The @name must be unique within the system.
199 : *
200 : * The return value depends on the @major input parameter:
201 : *
202 : * - if a major device number was requested in range [1..BLKDEV_MAJOR_MAX-1]
203 : * then the function returns zero on success, or a negative error code
204 : * - if any unused major number was requested with @major = 0 parameter
205 : * then the return value is the allocated major number in range
206 : * [1..BLKDEV_MAJOR_MAX-1] or a negative error code otherwise
207 : *
208 : * See Documentation/admin-guide/devices.txt for the list of allocated
209 : * major numbers.
210 : *
211 : * Use register_blkdev instead for any new code.
212 : */
213 1 : int __register_blkdev(unsigned int major, const char *name,
214 : void (*probe)(dev_t devt))
215 : {
216 : struct blk_major_name **n, *p;
217 1 : int index, ret = 0;
218 :
219 1 : mutex_lock(&major_names_lock);
220 :
221 : /* temporary */
222 1 : if (major == 0) {
223 0 : for (index = ARRAY_SIZE(major_names)-1; index > 0; index--) {
224 0 : if (major_names[index] == NULL)
225 : break;
226 : }
227 :
228 0 : if (index == 0) {
229 0 : printk("%s: failed to get major for %s\n",
230 : __func__, name);
231 0 : ret = -EBUSY;
232 0 : goto out;
233 : }
234 0 : major = index;
235 0 : ret = major;
236 : }
237 :
238 1 : if (major >= BLKDEV_MAJOR_MAX) {
239 0 : pr_err("%s: major requested (%u) is greater than the maximum (%u) for %s\n",
240 : __func__, major, BLKDEV_MAJOR_MAX-1, name);
241 :
242 0 : ret = -EINVAL;
243 0 : goto out;
244 : }
245 :
246 1 : p = kmalloc(sizeof(struct blk_major_name), GFP_KERNEL);
247 1 : if (p == NULL) {
248 : ret = -ENOMEM;
249 : goto out;
250 : }
251 :
252 1 : p->major = major;
253 : #ifdef CONFIG_BLOCK_LEGACY_AUTOLOAD
254 1 : p->probe = probe;
255 : #endif
256 1 : strlcpy(p->name, name, sizeof(p->name));
257 1 : p->next = NULL;
258 1 : index = major_to_index(major);
259 :
260 1 : spin_lock(&major_names_spinlock);
261 1 : for (n = &major_names[index]; *n; n = &(*n)->next) {
262 0 : if ((*n)->major == major)
263 : break;
264 : }
265 1 : if (!*n)
266 1 : *n = p;
267 : else
268 : ret = -EBUSY;
269 1 : spin_unlock(&major_names_spinlock);
270 :
271 1 : if (ret < 0) {
272 0 : printk("register_blkdev: cannot get major %u for %s\n",
273 : major, name);
274 0 : kfree(p);
275 : }
276 : out:
277 1 : mutex_unlock(&major_names_lock);
278 1 : return ret;
279 : }
280 : EXPORT_SYMBOL(__register_blkdev);
281 :
282 0 : void unregister_blkdev(unsigned int major, const char *name)
283 : {
284 : struct blk_major_name **n;
285 0 : struct blk_major_name *p = NULL;
286 0 : int index = major_to_index(major);
287 :
288 0 : mutex_lock(&major_names_lock);
289 0 : spin_lock(&major_names_spinlock);
290 0 : for (n = &major_names[index]; *n; n = &(*n)->next)
291 0 : if ((*n)->major == major)
292 : break;
293 0 : if (!*n || strcmp((*n)->name, name)) {
294 0 : WARN_ON(1);
295 : } else {
296 0 : p = *n;
297 0 : *n = p->next;
298 : }
299 0 : spin_unlock(&major_names_spinlock);
300 0 : mutex_unlock(&major_names_lock);
301 0 : kfree(p);
302 0 : }
303 :
304 : EXPORT_SYMBOL(unregister_blkdev);
305 :
306 0 : int blk_alloc_ext_minor(void)
307 : {
308 : int idx;
309 :
310 0 : idx = ida_alloc_range(&ext_devt_ida, 0, NR_EXT_DEVT - 1, GFP_KERNEL);
311 0 : if (idx == -ENOSPC)
312 : return -EBUSY;
313 0 : return idx;
314 : }
315 :
316 0 : void blk_free_ext_minor(unsigned int minor)
317 : {
318 0 : ida_free(&ext_devt_ida, minor);
319 0 : }
320 :
321 0 : static char *bdevt_str(dev_t devt, char *buf)
322 : {
323 0 : if (MAJOR(devt) <= 0xff && MINOR(devt) <= 0xff) {
324 : char tbuf[BDEVT_SIZE];
325 0 : snprintf(tbuf, BDEVT_SIZE, "%02x%02x", MAJOR(devt), MINOR(devt));
326 0 : snprintf(buf, BDEVT_SIZE, "%-9s", tbuf);
327 : } else
328 0 : snprintf(buf, BDEVT_SIZE, "%03x:%05x", MAJOR(devt), MINOR(devt));
329 :
330 0 : return buf;
331 : }
332 :
333 0 : void disk_uevent(struct gendisk *disk, enum kobject_action action)
334 : {
335 : struct block_device *part;
336 : unsigned long idx;
337 :
338 : rcu_read_lock();
339 0 : xa_for_each(&disk->part_tbl, idx, part) {
340 0 : if (bdev_is_partition(part) && !bdev_nr_sectors(part))
341 0 : continue;
342 0 : if (!kobject_get_unless_zero(&part->bd_device.kobj))
343 0 : continue;
344 :
345 : rcu_read_unlock();
346 0 : kobject_uevent(bdev_kobj(part), action);
347 0 : put_device(&part->bd_device);
348 : rcu_read_lock();
349 : }
350 : rcu_read_unlock();
351 0 : }
352 : EXPORT_SYMBOL_GPL(disk_uevent);
353 :
354 0 : int disk_scan_partitions(struct gendisk *disk, fmode_t mode)
355 : {
356 : struct block_device *bdev;
357 0 : int ret = 0;
358 :
359 0 : if (disk->flags & (GENHD_FL_NO_PART | GENHD_FL_HIDDEN))
360 : return -EINVAL;
361 0 : if (test_bit(GD_SUPPRESS_PART_SCAN, &disk->state))
362 : return -EINVAL;
363 0 : if (disk->open_partitions)
364 : return -EBUSY;
365 :
366 : /*
367 : * If the device is opened exclusively by current thread already, it's
368 : * safe to scan partitons, otherwise, use bd_prepare_to_claim() to
369 : * synchronize with other exclusive openers and other partition
370 : * scanners.
371 : */
372 0 : if (!(mode & FMODE_EXCL)) {
373 0 : ret = bd_prepare_to_claim(disk->part0, disk_scan_partitions);
374 0 : if (ret)
375 : return ret;
376 : }
377 :
378 0 : set_bit(GD_NEED_PART_SCAN, &disk->state);
379 0 : bdev = blkdev_get_by_dev(disk_devt(disk), mode & ~FMODE_EXCL, NULL);
380 0 : if (IS_ERR(bdev))
381 0 : ret = PTR_ERR(bdev);
382 : else
383 0 : blkdev_put(bdev, mode & ~FMODE_EXCL);
384 :
385 : /*
386 : * If blkdev_get_by_dev() failed early, GD_NEED_PART_SCAN is still set,
387 : * and this will cause that re-assemble partitioned raid device will
388 : * creat partition for underlying disk.
389 : */
390 0 : clear_bit(GD_NEED_PART_SCAN, &disk->state);
391 0 : if (!(mode & FMODE_EXCL))
392 0 : bd_abort_claiming(disk->part0, disk_scan_partitions);
393 : return ret;
394 : }
395 :
396 : /**
397 : * device_add_disk - add disk information to kernel list
398 : * @parent: parent device for the disk
399 : * @disk: per-device partitioning information
400 : * @groups: Additional per-device sysfs groups
401 : *
402 : * This function registers the partitioning information in @disk
403 : * with the kernel.
404 : */
405 0 : int __must_check device_add_disk(struct device *parent, struct gendisk *disk,
406 : const struct attribute_group **groups)
407 :
408 : {
409 0 : struct device *ddev = disk_to_dev(disk);
410 : int ret;
411 :
412 : /* Only makes sense for bio-based to set ->poll_bio */
413 0 : if (queue_is_mq(disk->queue) && disk->fops->poll_bio)
414 : return -EINVAL;
415 :
416 : /*
417 : * The disk queue should now be all set with enough information about
418 : * the device for the elevator code to pick an adequate default
419 : * elevator if one is needed, that is, for devices requesting queue
420 : * registration.
421 : */
422 0 : elevator_init_mq(disk->queue);
423 :
424 : /* Mark bdev as having a submit_bio, if needed */
425 0 : disk->part0->bd_has_submit_bio = disk->fops->submit_bio != NULL;
426 :
427 : /*
428 : * If the driver provides an explicit major number it also must provide
429 : * the number of minors numbers supported, and those will be used to
430 : * setup the gendisk.
431 : * Otherwise just allocate the device numbers for both the whole device
432 : * and all partitions from the extended dev_t space.
433 : */
434 0 : ret = -EINVAL;
435 0 : if (disk->major) {
436 0 : if (WARN_ON(!disk->minors))
437 : goto out_exit_elevator;
438 :
439 0 : if (disk->minors > DISK_MAX_PARTS) {
440 0 : pr_err("block: can't allocate more than %d partitions\n",
441 : DISK_MAX_PARTS);
442 0 : disk->minors = DISK_MAX_PARTS;
443 : }
444 0 : if (disk->first_minor + disk->minors > MINORMASK + 1)
445 : goto out_exit_elevator;
446 : } else {
447 0 : if (WARN_ON(disk->minors))
448 : goto out_exit_elevator;
449 :
450 0 : ret = blk_alloc_ext_minor();
451 0 : if (ret < 0)
452 : goto out_exit_elevator;
453 0 : disk->major = BLOCK_EXT_MAJOR;
454 0 : disk->first_minor = ret;
455 : }
456 :
457 : /* delay uevents, until we scanned partition table */
458 0 : dev_set_uevent_suppress(ddev, 1);
459 :
460 0 : ddev->parent = parent;
461 0 : ddev->groups = groups;
462 0 : dev_set_name(ddev, "%s", disk->disk_name);
463 0 : if (!(disk->flags & GENHD_FL_HIDDEN))
464 0 : ddev->devt = MKDEV(disk->major, disk->first_minor);
465 0 : ret = device_add(ddev);
466 0 : if (ret)
467 : goto out_free_ext_minor;
468 :
469 0 : ret = disk_alloc_events(disk);
470 0 : if (ret)
471 : goto out_device_del;
472 :
473 0 : ret = sysfs_create_link(block_depr, &ddev->kobj,
474 0 : kobject_name(&ddev->kobj));
475 0 : if (ret)
476 : goto out_device_del;
477 :
478 : /*
479 : * avoid probable deadlock caused by allocating memory with
480 : * GFP_KERNEL in runtime_resume callback of its all ancestor
481 : * devices
482 : */
483 0 : pm_runtime_set_memalloc_noio(ddev, true);
484 :
485 0 : disk->part0->bd_holder_dir =
486 0 : kobject_create_and_add("holders", &ddev->kobj);
487 0 : if (!disk->part0->bd_holder_dir) {
488 : ret = -ENOMEM;
489 : goto out_del_block_link;
490 : }
491 0 : disk->slave_dir = kobject_create_and_add("slaves", &ddev->kobj);
492 0 : if (!disk->slave_dir) {
493 : ret = -ENOMEM;
494 : goto out_put_holder_dir;
495 : }
496 :
497 0 : ret = blk_register_queue(disk);
498 0 : if (ret)
499 : goto out_put_slave_dir;
500 :
501 0 : if (!(disk->flags & GENHD_FL_HIDDEN)) {
502 0 : ret = bdi_register(disk->bdi, "%u:%u",
503 : disk->major, disk->first_minor);
504 0 : if (ret)
505 : goto out_unregister_queue;
506 0 : bdi_set_owner(disk->bdi, ddev);
507 0 : ret = sysfs_create_link(&ddev->kobj,
508 0 : &disk->bdi->dev->kobj, "bdi");
509 0 : if (ret)
510 : goto out_unregister_bdi;
511 :
512 : /* Make sure the first partition scan will be proceed */
513 0 : if (get_capacity(disk) && !(disk->flags & GENHD_FL_NO_PART) &&
514 0 : !test_bit(GD_SUPPRESS_PART_SCAN, &disk->state))
515 0 : set_bit(GD_NEED_PART_SCAN, &disk->state);
516 :
517 0 : bdev_add(disk->part0, ddev->devt);
518 0 : if (get_capacity(disk))
519 0 : disk_scan_partitions(disk, FMODE_READ);
520 :
521 : /*
522 : * Announce the disk and partitions after all partitions are
523 : * created. (for hidden disks uevents remain suppressed forever)
524 : */
525 0 : dev_set_uevent_suppress(ddev, 0);
526 0 : disk_uevent(disk, KOBJ_ADD);
527 : } else {
528 : /*
529 : * Even if the block_device for a hidden gendisk is not
530 : * registered, it needs to have a valid bd_dev so that the
531 : * freeing of the dynamic major works.
532 : */
533 0 : disk->part0->bd_dev = MKDEV(disk->major, disk->first_minor);
534 : }
535 :
536 0 : disk_update_readahead(disk);
537 0 : disk_add_events(disk);
538 0 : set_bit(GD_ADDED, &disk->state);
539 0 : return 0;
540 :
541 : out_unregister_bdi:
542 0 : if (!(disk->flags & GENHD_FL_HIDDEN))
543 0 : bdi_unregister(disk->bdi);
544 : out_unregister_queue:
545 0 : blk_unregister_queue(disk);
546 0 : rq_qos_exit(disk->queue);
547 : out_put_slave_dir:
548 0 : kobject_put(disk->slave_dir);
549 0 : disk->slave_dir = NULL;
550 : out_put_holder_dir:
551 0 : kobject_put(disk->part0->bd_holder_dir);
552 : out_del_block_link:
553 0 : sysfs_remove_link(block_depr, dev_name(ddev));
554 : out_device_del:
555 0 : device_del(ddev);
556 : out_free_ext_minor:
557 0 : if (disk->major == BLOCK_EXT_MAJOR)
558 0 : blk_free_ext_minor(disk->first_minor);
559 : out_exit_elevator:
560 0 : if (disk->queue->elevator)
561 0 : elevator_exit(disk->queue);
562 : return ret;
563 : }
564 : EXPORT_SYMBOL(device_add_disk);
565 :
566 : /**
567 : * blk_mark_disk_dead - mark a disk as dead
568 : * @disk: disk to mark as dead
569 : *
570 : * Mark as disk as dead (e.g. surprise removed) and don't accept any new I/O
571 : * to this disk.
572 : */
573 0 : void blk_mark_disk_dead(struct gendisk *disk)
574 : {
575 0 : set_bit(GD_DEAD, &disk->state);
576 0 : blk_queue_start_drain(disk->queue);
577 :
578 : /*
579 : * Stop buffered writers from dirtying pages that can't be written out.
580 : */
581 0 : set_capacity_and_notify(disk, 0);
582 0 : }
583 : EXPORT_SYMBOL_GPL(blk_mark_disk_dead);
584 :
585 : /**
586 : * del_gendisk - remove the gendisk
587 : * @disk: the struct gendisk to remove
588 : *
589 : * Removes the gendisk and all its associated resources. This deletes the
590 : * partitions associated with the gendisk, and unregisters the associated
591 : * request_queue.
592 : *
593 : * This is the counter to the respective __device_add_disk() call.
594 : *
595 : * The final removal of the struct gendisk happens when its refcount reaches 0
596 : * with put_disk(), which should be called after del_gendisk(), if
597 : * __device_add_disk() was used.
598 : *
599 : * Drivers exist which depend on the release of the gendisk to be synchronous,
600 : * it should not be deferred.
601 : *
602 : * Context: can sleep
603 : */
604 0 : void del_gendisk(struct gendisk *disk)
605 : {
606 0 : struct request_queue *q = disk->queue;
607 :
608 : might_sleep();
609 :
610 0 : if (WARN_ON_ONCE(!disk_live(disk) && !(disk->flags & GENHD_FL_HIDDEN)))
611 : return;
612 :
613 0 : disk_del_events(disk);
614 :
615 0 : mutex_lock(&disk->open_mutex);
616 0 : remove_inode_hash(disk->part0->bd_inode);
617 0 : blk_drop_partitions(disk);
618 0 : mutex_unlock(&disk->open_mutex);
619 :
620 0 : fsync_bdev(disk->part0);
621 0 : __invalidate_device(disk->part0, true);
622 :
623 : /*
624 : * Fail any new I/O.
625 : */
626 0 : set_bit(GD_DEAD, &disk->state);
627 0 : if (test_bit(GD_OWNS_QUEUE, &disk->state))
628 0 : blk_queue_flag_set(QUEUE_FLAG_DYING, q);
629 0 : set_capacity(disk, 0);
630 :
631 : /*
632 : * Prevent new I/O from crossing bio_queue_enter().
633 : */
634 0 : blk_queue_start_drain(q);
635 :
636 0 : if (!(disk->flags & GENHD_FL_HIDDEN)) {
637 0 : sysfs_remove_link(&disk_to_dev(disk)->kobj, "bdi");
638 :
639 : /*
640 : * Unregister bdi before releasing device numbers (as they can
641 : * get reused and we'd get clashes in sysfs).
642 : */
643 0 : bdi_unregister(disk->bdi);
644 : }
645 :
646 0 : blk_unregister_queue(disk);
647 :
648 0 : kobject_put(disk->part0->bd_holder_dir);
649 0 : kobject_put(disk->slave_dir);
650 0 : disk->slave_dir = NULL;
651 :
652 0 : part_stat_set_all(disk->part0, 0);
653 0 : disk->part0->bd_stamp = 0;
654 0 : sysfs_remove_link(block_depr, dev_name(disk_to_dev(disk)));
655 0 : pm_runtime_set_memalloc_noio(disk_to_dev(disk), false);
656 0 : device_del(disk_to_dev(disk));
657 :
658 0 : blk_mq_freeze_queue_wait(q);
659 :
660 0 : blk_throtl_cancel_bios(disk);
661 :
662 0 : blk_sync_queue(q);
663 : blk_flush_integrity();
664 :
665 0 : if (queue_is_mq(q))
666 0 : blk_mq_cancel_work_sync(q);
667 :
668 0 : blk_mq_quiesce_queue(q);
669 0 : if (q->elevator) {
670 0 : mutex_lock(&q->sysfs_lock);
671 0 : elevator_exit(q);
672 0 : mutex_unlock(&q->sysfs_lock);
673 : }
674 0 : rq_qos_exit(q);
675 0 : blk_mq_unquiesce_queue(q);
676 :
677 : /*
678 : * If the disk does not own the queue, allow using passthrough requests
679 : * again. Else leave the queue frozen to fail all I/O.
680 : */
681 0 : if (!test_bit(GD_OWNS_QUEUE, &disk->state)) {
682 0 : blk_queue_flag_clear(QUEUE_FLAG_INIT_DONE, q);
683 0 : __blk_mq_unfreeze_queue(q, true);
684 : } else {
685 0 : if (queue_is_mq(q))
686 0 : blk_mq_exit_queue(q);
687 : }
688 : }
689 : EXPORT_SYMBOL(del_gendisk);
690 :
691 : /**
692 : * invalidate_disk - invalidate the disk
693 : * @disk: the struct gendisk to invalidate
694 : *
695 : * A helper to invalidates the disk. It will clean the disk's associated
696 : * buffer/page caches and reset its internal states so that the disk
697 : * can be reused by the drivers.
698 : *
699 : * Context: can sleep
700 : */
701 0 : void invalidate_disk(struct gendisk *disk)
702 : {
703 0 : struct block_device *bdev = disk->part0;
704 :
705 0 : invalidate_bdev(bdev);
706 0 : bdev->bd_inode->i_mapping->wb_err = 0;
707 0 : set_capacity(disk, 0);
708 0 : }
709 : EXPORT_SYMBOL(invalidate_disk);
710 :
711 : /* sysfs access to bad-blocks list. */
712 0 : static ssize_t disk_badblocks_show(struct device *dev,
713 : struct device_attribute *attr,
714 : char *page)
715 : {
716 0 : struct gendisk *disk = dev_to_disk(dev);
717 :
718 0 : if (!disk->bb)
719 0 : return sprintf(page, "\n");
720 :
721 0 : return badblocks_show(disk->bb, page, 0);
722 : }
723 :
724 0 : static ssize_t disk_badblocks_store(struct device *dev,
725 : struct device_attribute *attr,
726 : const char *page, size_t len)
727 : {
728 0 : struct gendisk *disk = dev_to_disk(dev);
729 :
730 0 : if (!disk->bb)
731 : return -ENXIO;
732 :
733 0 : return badblocks_store(disk->bb, page, len, 0);
734 : }
735 :
736 : #ifdef CONFIG_BLOCK_LEGACY_AUTOLOAD
737 0 : void blk_request_module(dev_t devt)
738 : {
739 0 : unsigned int major = MAJOR(devt);
740 : struct blk_major_name **n;
741 :
742 0 : mutex_lock(&major_names_lock);
743 0 : for (n = &major_names[major_to_index(major)]; *n; n = &(*n)->next) {
744 0 : if ((*n)->major == major && (*n)->probe) {
745 0 : (*n)->probe(devt);
746 0 : mutex_unlock(&major_names_lock);
747 0 : return;
748 : }
749 : }
750 0 : mutex_unlock(&major_names_lock);
751 :
752 0 : if (request_module("block-major-%d-%d", MAJOR(devt), MINOR(devt)) > 0)
753 : /* Make old-style 2.4 aliases work */
754 : request_module("block-major-%d", MAJOR(devt));
755 : }
756 : #endif /* CONFIG_BLOCK_LEGACY_AUTOLOAD */
757 :
758 : /*
759 : * print a full list of all partitions - intended for places where the root
760 : * filesystem can't be mounted and thus to give the victim some idea of what
761 : * went wrong
762 : */
763 0 : void __init printk_all_partitions(void)
764 : {
765 : struct class_dev_iter iter;
766 : struct device *dev;
767 :
768 0 : class_dev_iter_init(&iter, &block_class, NULL, &disk_type);
769 0 : while ((dev = class_dev_iter_next(&iter))) {
770 0 : struct gendisk *disk = dev_to_disk(dev);
771 : struct block_device *part;
772 : char devt_buf[BDEVT_SIZE];
773 : unsigned long idx;
774 :
775 : /*
776 : * Don't show empty devices or things that have been
777 : * suppressed
778 : */
779 0 : if (get_capacity(disk) == 0 || (disk->flags & GENHD_FL_HIDDEN))
780 0 : continue;
781 :
782 : /*
783 : * Note, unlike /proc/partitions, I am showing the numbers in
784 : * hex - the same format as the root= option takes.
785 : */
786 : rcu_read_lock();
787 0 : xa_for_each(&disk->part_tbl, idx, part) {
788 0 : if (!bdev_nr_sectors(part))
789 0 : continue;
790 0 : printk("%s%s %10llu %pg %s",
791 : bdev_is_partition(part) ? " " : "",
792 : bdevt_str(part->bd_dev, devt_buf),
793 : bdev_nr_sectors(part) >> 1, part,
794 : part->bd_meta_info ?
795 : part->bd_meta_info->uuid : "");
796 0 : if (bdev_is_partition(part))
797 0 : printk("\n");
798 0 : else if (dev->parent && dev->parent->driver)
799 0 : printk(" driver: %s\n",
800 : dev->parent->driver->name);
801 : else
802 0 : printk(" (driver?)\n");
803 : }
804 : rcu_read_unlock();
805 : }
806 0 : class_dev_iter_exit(&iter);
807 0 : }
808 :
809 : #ifdef CONFIG_PROC_FS
810 : /* iterator */
811 0 : static void *disk_seqf_start(struct seq_file *seqf, loff_t *pos)
812 : {
813 0 : loff_t skip = *pos;
814 : struct class_dev_iter *iter;
815 : struct device *dev;
816 :
817 0 : iter = kmalloc(sizeof(*iter), GFP_KERNEL);
818 0 : if (!iter)
819 : return ERR_PTR(-ENOMEM);
820 :
821 0 : seqf->private = iter;
822 0 : class_dev_iter_init(iter, &block_class, NULL, &disk_type);
823 : do {
824 0 : dev = class_dev_iter_next(iter);
825 0 : if (!dev)
826 : return NULL;
827 0 : } while (skip--);
828 :
829 0 : return dev_to_disk(dev);
830 : }
831 :
832 0 : static void *disk_seqf_next(struct seq_file *seqf, void *v, loff_t *pos)
833 : {
834 : struct device *dev;
835 :
836 0 : (*pos)++;
837 0 : dev = class_dev_iter_next(seqf->private);
838 0 : if (dev)
839 0 : return dev_to_disk(dev);
840 :
841 : return NULL;
842 : }
843 :
844 0 : static void disk_seqf_stop(struct seq_file *seqf, void *v)
845 : {
846 0 : struct class_dev_iter *iter = seqf->private;
847 :
848 : /* stop is called even after start failed :-( */
849 0 : if (iter) {
850 0 : class_dev_iter_exit(iter);
851 0 : kfree(iter);
852 0 : seqf->private = NULL;
853 : }
854 0 : }
855 :
856 0 : static void *show_partition_start(struct seq_file *seqf, loff_t *pos)
857 : {
858 : void *p;
859 :
860 0 : p = disk_seqf_start(seqf, pos);
861 0 : if (!IS_ERR_OR_NULL(p) && !*pos)
862 0 : seq_puts(seqf, "major minor #blocks name\n\n");
863 0 : return p;
864 : }
865 :
866 0 : static int show_partition(struct seq_file *seqf, void *v)
867 : {
868 0 : struct gendisk *sgp = v;
869 : struct block_device *part;
870 : unsigned long idx;
871 :
872 0 : if (!get_capacity(sgp) || (sgp->flags & GENHD_FL_HIDDEN))
873 : return 0;
874 :
875 : rcu_read_lock();
876 0 : xa_for_each(&sgp->part_tbl, idx, part) {
877 0 : if (!bdev_nr_sectors(part))
878 0 : continue;
879 0 : seq_printf(seqf, "%4d %7d %10llu %pg\n",
880 0 : MAJOR(part->bd_dev), MINOR(part->bd_dev),
881 0 : bdev_nr_sectors(part) >> 1, part);
882 : }
883 : rcu_read_unlock();
884 0 : return 0;
885 : }
886 :
887 : static const struct seq_operations partitions_op = {
888 : .start = show_partition_start,
889 : .next = disk_seqf_next,
890 : .stop = disk_seqf_stop,
891 : .show = show_partition
892 : };
893 : #endif
894 :
895 1 : static int __init genhd_device_init(void)
896 : {
897 : int error;
898 :
899 1 : error = class_register(&block_class);
900 1 : if (unlikely(error))
901 : return error;
902 1 : blk_dev_init();
903 :
904 1 : register_blkdev(BLOCK_EXT_MAJOR, "blkext");
905 :
906 : /* create top-level block dir */
907 1 : block_depr = kobject_create_and_add("block", NULL);
908 1 : return 0;
909 : }
910 :
911 : subsys_initcall(genhd_device_init);
912 :
913 0 : static ssize_t disk_range_show(struct device *dev,
914 : struct device_attribute *attr, char *buf)
915 : {
916 0 : struct gendisk *disk = dev_to_disk(dev);
917 :
918 0 : return sprintf(buf, "%d\n", disk->minors);
919 : }
920 :
921 0 : static ssize_t disk_ext_range_show(struct device *dev,
922 : struct device_attribute *attr, char *buf)
923 : {
924 0 : struct gendisk *disk = dev_to_disk(dev);
925 :
926 0 : return sprintf(buf, "%d\n",
927 0 : (disk->flags & GENHD_FL_NO_PART) ? 1 : DISK_MAX_PARTS);
928 : }
929 :
930 0 : static ssize_t disk_removable_show(struct device *dev,
931 : struct device_attribute *attr, char *buf)
932 : {
933 0 : struct gendisk *disk = dev_to_disk(dev);
934 :
935 0 : return sprintf(buf, "%d\n",
936 0 : (disk->flags & GENHD_FL_REMOVABLE ? 1 : 0));
937 : }
938 :
939 0 : static ssize_t disk_hidden_show(struct device *dev,
940 : struct device_attribute *attr, char *buf)
941 : {
942 0 : struct gendisk *disk = dev_to_disk(dev);
943 :
944 0 : return sprintf(buf, "%d\n",
945 0 : (disk->flags & GENHD_FL_HIDDEN ? 1 : 0));
946 : }
947 :
948 0 : static ssize_t disk_ro_show(struct device *dev,
949 : struct device_attribute *attr, char *buf)
950 : {
951 0 : struct gendisk *disk = dev_to_disk(dev);
952 :
953 0 : return sprintf(buf, "%d\n", get_disk_ro(disk) ? 1 : 0);
954 : }
955 :
956 0 : ssize_t part_size_show(struct device *dev,
957 : struct device_attribute *attr, char *buf)
958 : {
959 0 : return sprintf(buf, "%llu\n", bdev_nr_sectors(dev_to_bdev(dev)));
960 : }
961 :
962 0 : ssize_t part_stat_show(struct device *dev,
963 : struct device_attribute *attr, char *buf)
964 : {
965 0 : struct block_device *bdev = dev_to_bdev(dev);
966 0 : struct request_queue *q = bdev_get_queue(bdev);
967 : struct disk_stats stat;
968 : unsigned int inflight;
969 :
970 0 : if (queue_is_mq(q))
971 0 : inflight = blk_mq_in_flight(q, bdev);
972 : else
973 : inflight = part_in_flight(bdev);
974 :
975 0 : if (inflight) {
976 0 : part_stat_lock();
977 0 : update_io_ticks(bdev, jiffies, true);
978 0 : part_stat_unlock();
979 : }
980 0 : part_stat_read_all(bdev, &stat);
981 0 : return sprintf(buf,
982 : "%8lu %8lu %8llu %8u "
983 : "%8lu %8lu %8llu %8u "
984 : "%8u %8u %8u "
985 : "%8lu %8lu %8llu %8u "
986 : "%8lu %8u"
987 : "\n",
988 : stat.ios[STAT_READ],
989 : stat.merges[STAT_READ],
990 0 : (unsigned long long)stat.sectors[STAT_READ],
991 0 : (unsigned int)div_u64(stat.nsecs[STAT_READ], NSEC_PER_MSEC),
992 : stat.ios[STAT_WRITE],
993 : stat.merges[STAT_WRITE],
994 0 : (unsigned long long)stat.sectors[STAT_WRITE],
995 0 : (unsigned int)div_u64(stat.nsecs[STAT_WRITE], NSEC_PER_MSEC),
996 : inflight,
997 : jiffies_to_msecs(stat.io_ticks),
998 0 : (unsigned int)div_u64(stat.nsecs[STAT_READ] +
999 0 : stat.nsecs[STAT_WRITE] +
1000 : stat.nsecs[STAT_DISCARD] +
1001 : stat.nsecs[STAT_FLUSH],
1002 : NSEC_PER_MSEC),
1003 : stat.ios[STAT_DISCARD],
1004 : stat.merges[STAT_DISCARD],
1005 0 : (unsigned long long)stat.sectors[STAT_DISCARD],
1006 0 : (unsigned int)div_u64(stat.nsecs[STAT_DISCARD], NSEC_PER_MSEC),
1007 : stat.ios[STAT_FLUSH],
1008 0 : (unsigned int)div_u64(stat.nsecs[STAT_FLUSH], NSEC_PER_MSEC));
1009 : }
1010 :
1011 0 : ssize_t part_inflight_show(struct device *dev, struct device_attribute *attr,
1012 : char *buf)
1013 : {
1014 0 : struct block_device *bdev = dev_to_bdev(dev);
1015 0 : struct request_queue *q = bdev_get_queue(bdev);
1016 : unsigned int inflight[2];
1017 :
1018 0 : if (queue_is_mq(q))
1019 0 : blk_mq_in_flight_rw(q, bdev, inflight);
1020 : else
1021 0 : part_in_flight_rw(bdev, inflight);
1022 :
1023 0 : return sprintf(buf, "%8u %8u\n", inflight[0], inflight[1]);
1024 : }
1025 :
1026 0 : static ssize_t disk_capability_show(struct device *dev,
1027 : struct device_attribute *attr, char *buf)
1028 : {
1029 0 : dev_warn_once(dev, "the capability attribute has been deprecated.\n");
1030 0 : return sprintf(buf, "0\n");
1031 : }
1032 :
1033 0 : static ssize_t disk_alignment_offset_show(struct device *dev,
1034 : struct device_attribute *attr,
1035 : char *buf)
1036 : {
1037 0 : struct gendisk *disk = dev_to_disk(dev);
1038 :
1039 0 : return sprintf(buf, "%d\n", bdev_alignment_offset(disk->part0));
1040 : }
1041 :
1042 0 : static ssize_t disk_discard_alignment_show(struct device *dev,
1043 : struct device_attribute *attr,
1044 : char *buf)
1045 : {
1046 0 : struct gendisk *disk = dev_to_disk(dev);
1047 :
1048 0 : return sprintf(buf, "%d\n", bdev_alignment_offset(disk->part0));
1049 : }
1050 :
1051 0 : static ssize_t diskseq_show(struct device *dev,
1052 : struct device_attribute *attr, char *buf)
1053 : {
1054 0 : struct gendisk *disk = dev_to_disk(dev);
1055 :
1056 0 : return sprintf(buf, "%llu\n", disk->diskseq);
1057 : }
1058 :
1059 : static DEVICE_ATTR(range, 0444, disk_range_show, NULL);
1060 : static DEVICE_ATTR(ext_range, 0444, disk_ext_range_show, NULL);
1061 : static DEVICE_ATTR(removable, 0444, disk_removable_show, NULL);
1062 : static DEVICE_ATTR(hidden, 0444, disk_hidden_show, NULL);
1063 : static DEVICE_ATTR(ro, 0444, disk_ro_show, NULL);
1064 : static DEVICE_ATTR(size, 0444, part_size_show, NULL);
1065 : static DEVICE_ATTR(alignment_offset, 0444, disk_alignment_offset_show, NULL);
1066 : static DEVICE_ATTR(discard_alignment, 0444, disk_discard_alignment_show, NULL);
1067 : static DEVICE_ATTR(capability, 0444, disk_capability_show, NULL);
1068 : static DEVICE_ATTR(stat, 0444, part_stat_show, NULL);
1069 : static DEVICE_ATTR(inflight, 0444, part_inflight_show, NULL);
1070 : static DEVICE_ATTR(badblocks, 0644, disk_badblocks_show, disk_badblocks_store);
1071 : static DEVICE_ATTR(diskseq, 0444, diskseq_show, NULL);
1072 :
1073 : #ifdef CONFIG_FAIL_MAKE_REQUEST
1074 : ssize_t part_fail_show(struct device *dev,
1075 : struct device_attribute *attr, char *buf)
1076 : {
1077 : return sprintf(buf, "%d\n", dev_to_bdev(dev)->bd_make_it_fail);
1078 : }
1079 :
1080 : ssize_t part_fail_store(struct device *dev,
1081 : struct device_attribute *attr,
1082 : const char *buf, size_t count)
1083 : {
1084 : int i;
1085 :
1086 : if (count > 0 && sscanf(buf, "%d", &i) > 0)
1087 : dev_to_bdev(dev)->bd_make_it_fail = i;
1088 :
1089 : return count;
1090 : }
1091 :
1092 : static struct device_attribute dev_attr_fail =
1093 : __ATTR(make-it-fail, 0644, part_fail_show, part_fail_store);
1094 : #endif /* CONFIG_FAIL_MAKE_REQUEST */
1095 :
1096 : #ifdef CONFIG_FAIL_IO_TIMEOUT
1097 : static struct device_attribute dev_attr_fail_timeout =
1098 : __ATTR(io-timeout-fail, 0644, part_timeout_show, part_timeout_store);
1099 : #endif
1100 :
1101 : static struct attribute *disk_attrs[] = {
1102 : &dev_attr_range.attr,
1103 : &dev_attr_ext_range.attr,
1104 : &dev_attr_removable.attr,
1105 : &dev_attr_hidden.attr,
1106 : &dev_attr_ro.attr,
1107 : &dev_attr_size.attr,
1108 : &dev_attr_alignment_offset.attr,
1109 : &dev_attr_discard_alignment.attr,
1110 : &dev_attr_capability.attr,
1111 : &dev_attr_stat.attr,
1112 : &dev_attr_inflight.attr,
1113 : &dev_attr_badblocks.attr,
1114 : &dev_attr_events.attr,
1115 : &dev_attr_events_async.attr,
1116 : &dev_attr_events_poll_msecs.attr,
1117 : &dev_attr_diskseq.attr,
1118 : #ifdef CONFIG_FAIL_MAKE_REQUEST
1119 : &dev_attr_fail.attr,
1120 : #endif
1121 : #ifdef CONFIG_FAIL_IO_TIMEOUT
1122 : &dev_attr_fail_timeout.attr,
1123 : #endif
1124 : NULL
1125 : };
1126 :
1127 0 : static umode_t disk_visible(struct kobject *kobj, struct attribute *a, int n)
1128 : {
1129 0 : struct device *dev = container_of(kobj, typeof(*dev), kobj);
1130 0 : struct gendisk *disk = dev_to_disk(dev);
1131 :
1132 0 : if (a == &dev_attr_badblocks.attr && !disk->bb)
1133 : return 0;
1134 0 : return a->mode;
1135 : }
1136 :
1137 : static struct attribute_group disk_attr_group = {
1138 : .attrs = disk_attrs,
1139 : .is_visible = disk_visible,
1140 : };
1141 :
1142 : static const struct attribute_group *disk_attr_groups[] = {
1143 : &disk_attr_group,
1144 : #ifdef CONFIG_BLK_DEV_IO_TRACE
1145 : &blk_trace_attr_group,
1146 : #endif
1147 : #ifdef CONFIG_BLK_DEV_INTEGRITY
1148 : &blk_integrity_attr_group,
1149 : #endif
1150 : NULL
1151 : };
1152 :
1153 : /**
1154 : * disk_release - releases all allocated resources of the gendisk
1155 : * @dev: the device representing this disk
1156 : *
1157 : * This function releases all allocated resources of the gendisk.
1158 : *
1159 : * Drivers which used __device_add_disk() have a gendisk with a request_queue
1160 : * assigned. Since the request_queue sits on top of the gendisk for these
1161 : * drivers we also call blk_put_queue() for them, and we expect the
1162 : * request_queue refcount to reach 0 at this point, and so the request_queue
1163 : * will also be freed prior to the disk.
1164 : *
1165 : * Context: can sleep
1166 : */
1167 0 : static void disk_release(struct device *dev)
1168 : {
1169 0 : struct gendisk *disk = dev_to_disk(dev);
1170 :
1171 : might_sleep();
1172 0 : WARN_ON_ONCE(disk_live(disk));
1173 :
1174 : /*
1175 : * To undo the all initialization from blk_mq_init_allocated_queue in
1176 : * case of a probe failure where add_disk is never called we have to
1177 : * call blk_mq_exit_queue here. We can't do this for the more common
1178 : * teardown case (yet) as the tagset can be gone by the time the disk
1179 : * is released once it was added.
1180 : */
1181 0 : if (queue_is_mq(disk->queue) &&
1182 0 : test_bit(GD_OWNS_QUEUE, &disk->state) &&
1183 0 : !test_bit(GD_ADDED, &disk->state))
1184 0 : blk_mq_exit_queue(disk->queue);
1185 :
1186 0 : blkcg_exit_disk(disk);
1187 :
1188 0 : bioset_exit(&disk->bio_split);
1189 :
1190 0 : disk_release_events(disk);
1191 0 : kfree(disk->random);
1192 0 : disk_free_zone_bitmaps(disk);
1193 0 : xa_destroy(&disk->part_tbl);
1194 :
1195 0 : disk->queue->disk = NULL;
1196 0 : blk_put_queue(disk->queue);
1197 :
1198 0 : if (test_bit(GD_ADDED, &disk->state) && disk->fops->free_disk)
1199 0 : disk->fops->free_disk(disk);
1200 :
1201 0 : iput(disk->part0->bd_inode); /* frees the disk */
1202 0 : }
1203 :
1204 0 : static int block_uevent(const struct device *dev, struct kobj_uevent_env *env)
1205 : {
1206 0 : const struct gendisk *disk = dev_to_disk(dev);
1207 :
1208 0 : return add_uevent_var(env, "DISKSEQ=%llu", disk->diskseq);
1209 : }
1210 :
1211 : struct class block_class = {
1212 : .name = "block",
1213 : .dev_uevent = block_uevent,
1214 : };
1215 :
1216 0 : static char *block_devnode(const struct device *dev, umode_t *mode,
1217 : kuid_t *uid, kgid_t *gid)
1218 : {
1219 0 : struct gendisk *disk = dev_to_disk(dev);
1220 :
1221 0 : if (disk->fops->devnode)
1222 0 : return disk->fops->devnode(disk, mode);
1223 : return NULL;
1224 : }
1225 :
1226 : const struct device_type disk_type = {
1227 : .name = "disk",
1228 : .groups = disk_attr_groups,
1229 : .release = disk_release,
1230 : .devnode = block_devnode,
1231 : };
1232 :
1233 : #ifdef CONFIG_PROC_FS
1234 : /*
1235 : * aggregate disk stat collector. Uses the same stats that the sysfs
1236 : * entries do, above, but makes them available through one seq_file.
1237 : *
1238 : * The output looks suspiciously like /proc/partitions with a bunch of
1239 : * extra fields.
1240 : */
1241 0 : static int diskstats_show(struct seq_file *seqf, void *v)
1242 : {
1243 0 : struct gendisk *gp = v;
1244 : struct block_device *hd;
1245 : unsigned int inflight;
1246 : struct disk_stats stat;
1247 : unsigned long idx;
1248 :
1249 : /*
1250 : if (&disk_to_dev(gp)->kobj.entry == block_class.devices.next)
1251 : seq_puts(seqf, "major minor name"
1252 : " rio rmerge rsect ruse wio wmerge "
1253 : "wsect wuse running use aveq"
1254 : "\n\n");
1255 : */
1256 :
1257 : rcu_read_lock();
1258 0 : xa_for_each(&gp->part_tbl, idx, hd) {
1259 0 : if (bdev_is_partition(hd) && !bdev_nr_sectors(hd))
1260 0 : continue;
1261 0 : if (queue_is_mq(gp->queue))
1262 0 : inflight = blk_mq_in_flight(gp->queue, hd);
1263 : else
1264 : inflight = part_in_flight(hd);
1265 :
1266 0 : if (inflight) {
1267 0 : part_stat_lock();
1268 0 : update_io_ticks(hd, jiffies, true);
1269 0 : part_stat_unlock();
1270 : }
1271 0 : part_stat_read_all(hd, &stat);
1272 0 : seq_printf(seqf, "%4d %7d %pg "
1273 : "%lu %lu %lu %u "
1274 : "%lu %lu %lu %u "
1275 : "%u %u %u "
1276 : "%lu %lu %lu %u "
1277 : "%lu %u"
1278 : "\n",
1279 0 : MAJOR(hd->bd_dev), MINOR(hd->bd_dev), hd,
1280 : stat.ios[STAT_READ],
1281 : stat.merges[STAT_READ],
1282 : stat.sectors[STAT_READ],
1283 0 : (unsigned int)div_u64(stat.nsecs[STAT_READ],
1284 : NSEC_PER_MSEC),
1285 : stat.ios[STAT_WRITE],
1286 : stat.merges[STAT_WRITE],
1287 : stat.sectors[STAT_WRITE],
1288 0 : (unsigned int)div_u64(stat.nsecs[STAT_WRITE],
1289 : NSEC_PER_MSEC),
1290 : inflight,
1291 : jiffies_to_msecs(stat.io_ticks),
1292 0 : (unsigned int)div_u64(stat.nsecs[STAT_READ] +
1293 0 : stat.nsecs[STAT_WRITE] +
1294 : stat.nsecs[STAT_DISCARD] +
1295 : stat.nsecs[STAT_FLUSH],
1296 : NSEC_PER_MSEC),
1297 : stat.ios[STAT_DISCARD],
1298 : stat.merges[STAT_DISCARD],
1299 : stat.sectors[STAT_DISCARD],
1300 0 : (unsigned int)div_u64(stat.nsecs[STAT_DISCARD],
1301 : NSEC_PER_MSEC),
1302 : stat.ios[STAT_FLUSH],
1303 0 : (unsigned int)div_u64(stat.nsecs[STAT_FLUSH],
1304 : NSEC_PER_MSEC)
1305 : );
1306 : }
1307 : rcu_read_unlock();
1308 :
1309 0 : return 0;
1310 : }
1311 :
1312 : static const struct seq_operations diskstats_op = {
1313 : .start = disk_seqf_start,
1314 : .next = disk_seqf_next,
1315 : .stop = disk_seqf_stop,
1316 : .show = diskstats_show
1317 : };
1318 :
1319 1 : static int __init proc_genhd_init(void)
1320 : {
1321 1 : proc_create_seq("diskstats", 0, NULL, &diskstats_op);
1322 1 : proc_create_seq("partitions", 0, NULL, &partitions_op);
1323 1 : return 0;
1324 : }
1325 : module_init(proc_genhd_init);
1326 : #endif /* CONFIG_PROC_FS */
1327 :
1328 0 : dev_t part_devt(struct gendisk *disk, u8 partno)
1329 : {
1330 : struct block_device *part;
1331 0 : dev_t devt = 0;
1332 :
1333 : rcu_read_lock();
1334 0 : part = xa_load(&disk->part_tbl, partno);
1335 0 : if (part)
1336 0 : devt = part->bd_dev;
1337 : rcu_read_unlock();
1338 :
1339 0 : return devt;
1340 : }
1341 :
1342 0 : dev_t blk_lookup_devt(const char *name, int partno)
1343 : {
1344 0 : dev_t devt = MKDEV(0, 0);
1345 : struct class_dev_iter iter;
1346 : struct device *dev;
1347 :
1348 0 : class_dev_iter_init(&iter, &block_class, NULL, &disk_type);
1349 0 : while ((dev = class_dev_iter_next(&iter))) {
1350 0 : struct gendisk *disk = dev_to_disk(dev);
1351 :
1352 0 : if (strcmp(dev_name(dev), name))
1353 0 : continue;
1354 :
1355 0 : if (partno < disk->minors) {
1356 : /* We need to return the right devno, even
1357 : * if the partition doesn't exist yet.
1358 : */
1359 0 : devt = MKDEV(MAJOR(dev->devt),
1360 : MINOR(dev->devt) + partno);
1361 : } else {
1362 0 : devt = part_devt(disk, partno);
1363 0 : if (devt)
1364 : break;
1365 : }
1366 : }
1367 0 : class_dev_iter_exit(&iter);
1368 0 : return devt;
1369 : }
1370 :
1371 0 : struct gendisk *__alloc_disk_node(struct request_queue *q, int node_id,
1372 : struct lock_class_key *lkclass)
1373 : {
1374 : struct gendisk *disk;
1375 :
1376 0 : disk = kzalloc_node(sizeof(struct gendisk), GFP_KERNEL, node_id);
1377 0 : if (!disk)
1378 : return NULL;
1379 :
1380 0 : if (bioset_init(&disk->bio_split, BIO_POOL_SIZE, 0, 0))
1381 : goto out_free_disk;
1382 :
1383 0 : disk->bdi = bdi_alloc(node_id);
1384 0 : if (!disk->bdi)
1385 : goto out_free_bioset;
1386 :
1387 : /* bdev_alloc() might need the queue, set before the first call */
1388 0 : disk->queue = q;
1389 :
1390 0 : disk->part0 = bdev_alloc(disk, 0);
1391 0 : if (!disk->part0)
1392 : goto out_free_bdi;
1393 :
1394 0 : disk->node_id = node_id;
1395 0 : mutex_init(&disk->open_mutex);
1396 0 : xa_init(&disk->part_tbl);
1397 0 : if (xa_insert(&disk->part_tbl, 0, disk->part0, GFP_KERNEL))
1398 : goto out_destroy_part_tbl;
1399 :
1400 0 : if (blkcg_init_disk(disk))
1401 : goto out_erase_part0;
1402 :
1403 0 : rand_initialize_disk(disk);
1404 0 : disk_to_dev(disk)->class = &block_class;
1405 0 : disk_to_dev(disk)->type = &disk_type;
1406 0 : device_initialize(disk_to_dev(disk));
1407 0 : inc_diskseq(disk);
1408 0 : q->disk = disk;
1409 : lockdep_init_map(&disk->lockdep_map, "(bio completion)", lkclass, 0);
1410 : #ifdef CONFIG_BLOCK_HOLDER_DEPRECATED
1411 : INIT_LIST_HEAD(&disk->slave_bdevs);
1412 : #endif
1413 0 : return disk;
1414 :
1415 : out_erase_part0:
1416 : xa_erase(&disk->part_tbl, 0);
1417 : out_destroy_part_tbl:
1418 0 : xa_destroy(&disk->part_tbl);
1419 0 : disk->part0->bd_disk = NULL;
1420 0 : iput(disk->part0->bd_inode);
1421 : out_free_bdi:
1422 0 : bdi_put(disk->bdi);
1423 : out_free_bioset:
1424 0 : bioset_exit(&disk->bio_split);
1425 : out_free_disk:
1426 0 : kfree(disk);
1427 0 : return NULL;
1428 : }
1429 :
1430 0 : struct gendisk *__blk_alloc_disk(int node, struct lock_class_key *lkclass)
1431 : {
1432 : struct request_queue *q;
1433 : struct gendisk *disk;
1434 :
1435 0 : q = blk_alloc_queue(node);
1436 0 : if (!q)
1437 : return NULL;
1438 :
1439 0 : disk = __alloc_disk_node(q, node, lkclass);
1440 0 : if (!disk) {
1441 0 : blk_put_queue(q);
1442 0 : return NULL;
1443 : }
1444 0 : set_bit(GD_OWNS_QUEUE, &disk->state);
1445 0 : return disk;
1446 : }
1447 : EXPORT_SYMBOL(__blk_alloc_disk);
1448 :
1449 : /**
1450 : * put_disk - decrements the gendisk refcount
1451 : * @disk: the struct gendisk to decrement the refcount for
1452 : *
1453 : * This decrements the refcount for the struct gendisk. When this reaches 0
1454 : * we'll have disk_release() called.
1455 : *
1456 : * Note: for blk-mq disk put_disk must be called before freeing the tag_set
1457 : * when handling probe errors (that is before add_disk() is called).
1458 : *
1459 : * Context: Any context, but the last reference must not be dropped from
1460 : * atomic context.
1461 : */
1462 0 : void put_disk(struct gendisk *disk)
1463 : {
1464 0 : if (disk)
1465 0 : put_device(disk_to_dev(disk));
1466 0 : }
1467 : EXPORT_SYMBOL(put_disk);
1468 :
1469 0 : static void set_disk_ro_uevent(struct gendisk *gd, int ro)
1470 : {
1471 0 : char event[] = "DISK_RO=1";
1472 0 : char *envp[] = { event, NULL };
1473 :
1474 0 : if (!ro)
1475 0 : event[8] = '0';
1476 0 : kobject_uevent_env(&disk_to_dev(gd)->kobj, KOBJ_CHANGE, envp);
1477 0 : }
1478 :
1479 : /**
1480 : * set_disk_ro - set a gendisk read-only
1481 : * @disk: gendisk to operate on
1482 : * @read_only: %true to set the disk read-only, %false set the disk read/write
1483 : *
1484 : * This function is used to indicate whether a given disk device should have its
1485 : * read-only flag set. set_disk_ro() is typically used by device drivers to
1486 : * indicate whether the underlying physical device is write-protected.
1487 : */
1488 0 : void set_disk_ro(struct gendisk *disk, bool read_only)
1489 : {
1490 0 : if (read_only) {
1491 0 : if (test_and_set_bit(GD_READ_ONLY, &disk->state))
1492 : return;
1493 : } else {
1494 0 : if (!test_and_clear_bit(GD_READ_ONLY, &disk->state))
1495 : return;
1496 : }
1497 0 : set_disk_ro_uevent(disk, read_only);
1498 : }
1499 : EXPORT_SYMBOL(set_disk_ro);
1500 :
1501 0 : void inc_diskseq(struct gendisk *disk)
1502 : {
1503 0 : disk->diskseq = atomic64_inc_return(&diskseq);
1504 0 : }
|
