1 // SPDX-License-Identifier: GPL-2.0
2 #ifndef NO_BCACHEFS_CHARDEV
5 #include "bcachefs_ioctl.h"
14 #include <linux/anon_inodes.h>
15 #include <linux/cdev.h>
16 #include <linux/device.h>
17 #include <linux/file.h>
19 #include <linux/ioctl.h>
20 #include <linux/kthread.h>
21 #include <linux/major.h>
22 #include <linux/sched/task.h>
23 #include <linux/slab.h>
24 #include <linux/uaccess.h>
27 static int copy_to_user_errcode(void __user *to, const void *from, unsigned long n)
29 return copy_to_user(to, from, n) ? -EFAULT : 0;
32 /* returns with ref on ca->ref */
33 static struct bch_dev *bch2_device_lookup(struct bch_fs *c, u64 dev,
38 if (flags & BCH_BY_INDEX) {
39 if (dev >= c->sb.nr_devices)
40 return ERR_PTR(-EINVAL);
43 ca = rcu_dereference(c->devs[dev]);
45 percpu_ref_get(&ca->ref);
49 return ERR_PTR(-EINVAL);
53 path = strndup_user((const char __user *)
54 (unsigned long) dev, PATH_MAX);
56 return ERR_CAST(path);
58 ca = bch2_dev_lookup(c, path);
66 static long bch2_ioctl_assemble(struct bch_ioctl_assemble __user *user_arg)
68 struct bch_ioctl_assemble arg;
70 u64 *user_devs = NULL;
75 if (copy_from_user(&arg, user_arg, sizeof(arg)))
78 if (arg.flags || arg.pad)
81 user_devs = kmalloc_array(arg.nr_devs, sizeof(u64), GFP_KERNEL);
85 devs = kcalloc(arg.nr_devs, sizeof(char *), GFP_KERNEL);
87 if (copy_from_user(user_devs, user_arg->devs,
88 sizeof(u64) * arg.nr_devs))
91 for (i = 0; i < arg.nr_devs; i++) {
92 devs[i] = strndup_user((const char __user *)(unsigned long)
95 ret= PTR_ERR_OR_ZERO(devs[i]);
100 c = bch2_fs_open(devs, arg.nr_devs, bch2_opts_empty());
101 ret = PTR_ERR_OR_ZERO(c);
106 for (i = 0; i < arg.nr_devs; i++)
112 static long bch2_ioctl_incremental(struct bch_ioctl_incremental __user *user_arg)
114 struct bch_ioctl_incremental arg;
118 if (copy_from_user(&arg, user_arg, sizeof(arg)))
121 if (arg.flags || arg.pad)
124 path = strndup_user((const char __user *)(unsigned long) arg.dev, PATH_MAX);
125 ret = PTR_ERR_OR_ZERO(path);
129 err = bch2_fs_open_incremental(path);
133 pr_err("Could not register bcachefs devices: %s", err);
141 static long bch2_global_ioctl(unsigned cmd, void __user *arg)
145 case BCH_IOCTL_ASSEMBLE:
146 return bch2_ioctl_assemble(arg);
147 case BCH_IOCTL_INCREMENTAL:
148 return bch2_ioctl_incremental(arg);
155 static long bch2_ioctl_query_uuid(struct bch_fs *c,
156 struct bch_ioctl_query_uuid __user *user_arg)
158 return copy_to_user_errcode(&user_arg->uuid, &c->sb.user_uuid,
159 sizeof(c->sb.user_uuid));
163 static long bch2_ioctl_start(struct bch_fs *c, struct bch_ioctl_start arg)
165 if (!capable(CAP_SYS_ADMIN))
168 if (arg.flags || arg.pad)
171 return bch2_fs_start(c);
174 static long bch2_ioctl_stop(struct bch_fs *c)
176 if (!capable(CAP_SYS_ADMIN))
184 static long bch2_ioctl_disk_add(struct bch_fs *c, struct bch_ioctl_disk arg)
189 if (!capable(CAP_SYS_ADMIN))
192 if (arg.flags || arg.pad)
195 path = strndup_user((const char __user *)(unsigned long) arg.dev, PATH_MAX);
196 ret = PTR_ERR_OR_ZERO(path);
200 ret = bch2_dev_add(c, path);
206 static long bch2_ioctl_disk_remove(struct bch_fs *c, struct bch_ioctl_disk arg)
210 if (!capable(CAP_SYS_ADMIN))
213 if ((arg.flags & ~(BCH_FORCE_IF_DATA_LOST|
214 BCH_FORCE_IF_METADATA_LOST|
215 BCH_FORCE_IF_DEGRADED|
220 ca = bch2_device_lookup(c, arg.dev, arg.flags);
224 return bch2_dev_remove(c, ca, arg.flags);
227 static long bch2_ioctl_disk_online(struct bch_fs *c, struct bch_ioctl_disk arg)
232 if (!capable(CAP_SYS_ADMIN))
235 if (arg.flags || arg.pad)
238 path = strndup_user((const char __user *)(unsigned long) arg.dev, PATH_MAX);
239 ret = PTR_ERR_OR_ZERO(path);
243 ret = bch2_dev_online(c, path);
248 static long bch2_ioctl_disk_offline(struct bch_fs *c, struct bch_ioctl_disk arg)
253 if (!capable(CAP_SYS_ADMIN))
256 if ((arg.flags & ~(BCH_FORCE_IF_DATA_LOST|
257 BCH_FORCE_IF_METADATA_LOST|
258 BCH_FORCE_IF_DEGRADED|
263 ca = bch2_device_lookup(c, arg.dev, arg.flags);
267 ret = bch2_dev_offline(c, ca, arg.flags);
268 percpu_ref_put(&ca->ref);
272 static long bch2_ioctl_disk_set_state(struct bch_fs *c,
273 struct bch_ioctl_disk_set_state arg)
278 if (!capable(CAP_SYS_ADMIN))
281 if ((arg.flags & ~(BCH_FORCE_IF_DATA_LOST|
282 BCH_FORCE_IF_METADATA_LOST|
283 BCH_FORCE_IF_DEGRADED|
285 arg.pad[0] || arg.pad[1] || arg.pad[2] ||
286 arg.new_state >= BCH_MEMBER_STATE_NR)
289 ca = bch2_device_lookup(c, arg.dev, arg.flags);
293 ret = bch2_dev_set_state(c, ca, arg.new_state, arg.flags);
295 bch_err(c, "Error setting device state: %s", bch2_err_str(ret));
297 percpu_ref_put(&ca->ref);
301 struct bch_data_ctx {
303 struct bch_ioctl_data arg;
304 struct bch_move_stats stats;
308 struct task_struct *thread;
311 static int bch2_data_thread(void *arg)
313 struct bch_data_ctx *ctx = arg;
315 ctx->ret = bch2_data_job(ctx->c, &ctx->stats, ctx->arg);
317 ctx->stats.data_type = U8_MAX;
321 static int bch2_data_job_release(struct inode *inode, struct file *file)
323 struct bch_data_ctx *ctx = file->private_data;
325 kthread_stop(ctx->thread);
326 put_task_struct(ctx->thread);
331 static ssize_t bch2_data_job_read(struct file *file, char __user *buf,
332 size_t len, loff_t *ppos)
334 struct bch_data_ctx *ctx = file->private_data;
335 struct bch_fs *c = ctx->c;
336 struct bch_ioctl_data_event e = {
337 .type = BCH_DATA_EVENT_PROGRESS,
338 .p.data_type = ctx->stats.data_type,
339 .p.btree_id = ctx->stats.pos.btree,
340 .p.pos = ctx->stats.pos.pos,
341 .p.sectors_done = atomic64_read(&ctx->stats.sectors_seen),
342 .p.sectors_total = bch2_fs_usage_read_short(c).used,
348 return copy_to_user_errcode(buf, &e, sizeof(e)) ?: sizeof(e);
351 static const struct file_operations bcachefs_data_ops = {
352 .release = bch2_data_job_release,
353 .read = bch2_data_job_read,
357 static long bch2_ioctl_data(struct bch_fs *c,
358 struct bch_ioctl_data arg)
360 struct bch_data_ctx *ctx = NULL;
361 struct file *file = NULL;
362 unsigned flags = O_RDONLY|O_CLOEXEC|O_NONBLOCK;
365 if (!capable(CAP_SYS_ADMIN))
368 if (arg.op >= BCH_DATA_OP_NR || arg.flags)
371 ctx = kzalloc(sizeof(*ctx), GFP_KERNEL);
378 ctx->thread = kthread_create(bch2_data_thread, ctx,
379 "bch-data/%s", c->name);
380 if (IS_ERR(ctx->thread)) {
381 ret = PTR_ERR(ctx->thread);
385 ret = get_unused_fd_flags(flags);
390 file = anon_inode_getfile("[bcachefs]", &bcachefs_data_ops, ctx, flags);
396 fd_install(fd, file);
398 get_task_struct(ctx->thread);
399 wake_up_process(ctx->thread);
405 if (!IS_ERR_OR_NULL(ctx->thread))
406 kthread_stop(ctx->thread);
411 static long bch2_ioctl_fs_usage(struct bch_fs *c,
412 struct bch_ioctl_fs_usage __user *user_arg)
414 struct bch_ioctl_fs_usage *arg = NULL;
415 struct bch_replicas_usage *dst_e, *dst_end;
416 struct bch_fs_usage_online *src;
417 u32 replica_entries_bytes;
421 if (!test_bit(BCH_FS_STARTED, &c->flags))
424 if (get_user(replica_entries_bytes, &user_arg->replica_entries_bytes))
427 arg = kzalloc(size_add(sizeof(*arg), replica_entries_bytes), GFP_KERNEL);
431 src = bch2_fs_usage_read(c);
437 arg->capacity = c->capacity;
438 arg->used = bch2_fs_sectors_used(c, src);
439 arg->online_reserved = src->online_reserved;
441 for (i = 0; i < BCH_REPLICAS_MAX; i++)
442 arg->persistent_reserved[i] = src->u.persistent_reserved[i];
444 dst_e = arg->replicas;
445 dst_end = (void *) arg->replicas + replica_entries_bytes;
447 for (i = 0; i < c->replicas.nr; i++) {
448 struct bch_replicas_entry_v1 *src_e =
449 cpu_replicas_entry(&c->replicas, i);
451 /* check that we have enough space for one replicas entry */
452 if (dst_e + 1 > dst_end) {
457 dst_e->sectors = src->u.replicas[i];
460 /* recheck after setting nr_devs: */
461 if (replicas_usage_next(dst_e) > dst_end) {
466 memcpy(dst_e->r.devs, src_e->devs, src_e->nr_devs);
468 dst_e = replicas_usage_next(dst_e);
471 arg->replica_entries_bytes = (void *) dst_e - (void *) arg->replicas;
473 percpu_up_read(&c->mark_lock);
479 ret = copy_to_user_errcode(user_arg, arg,
480 sizeof(*arg) + arg->replica_entries_bytes);
486 /* obsolete, didn't allow for new data types: */
487 static long bch2_ioctl_dev_usage(struct bch_fs *c,
488 struct bch_ioctl_dev_usage __user *user_arg)
490 struct bch_ioctl_dev_usage arg;
491 struct bch_dev_usage src;
495 if (!test_bit(BCH_FS_STARTED, &c->flags))
498 if (copy_from_user(&arg, user_arg, sizeof(arg)))
501 if ((arg.flags & ~BCH_BY_INDEX) ||
507 ca = bch2_device_lookup(c, arg.dev, arg.flags);
511 src = bch2_dev_usage_read(ca);
513 arg.state = ca->mi.state;
514 arg.bucket_size = ca->mi.bucket_size;
515 arg.nr_buckets = ca->mi.nbuckets - ca->mi.first_bucket;
517 for (i = 0; i < BCH_DATA_NR; i++) {
518 arg.d[i].buckets = src.d[i].buckets;
519 arg.d[i].sectors = src.d[i].sectors;
520 arg.d[i].fragmented = src.d[i].fragmented;
523 percpu_ref_put(&ca->ref);
525 return copy_to_user_errcode(user_arg, &arg, sizeof(arg));
528 static long bch2_ioctl_dev_usage_v2(struct bch_fs *c,
529 struct bch_ioctl_dev_usage_v2 __user *user_arg)
531 struct bch_ioctl_dev_usage_v2 arg;
532 struct bch_dev_usage src;
536 if (!test_bit(BCH_FS_STARTED, &c->flags))
539 if (copy_from_user(&arg, user_arg, sizeof(arg)))
542 if ((arg.flags & ~BCH_BY_INDEX) ||
548 ca = bch2_device_lookup(c, arg.dev, arg.flags);
552 src = bch2_dev_usage_read(ca);
554 arg.state = ca->mi.state;
555 arg.bucket_size = ca->mi.bucket_size;
556 arg.nr_data_types = min(arg.nr_data_types, BCH_DATA_NR);
557 arg.nr_buckets = ca->mi.nbuckets - ca->mi.first_bucket;
559 ret = copy_to_user_errcode(user_arg, &arg, sizeof(arg));
563 for (unsigned i = 0; i < arg.nr_data_types; i++) {
564 struct bch_ioctl_dev_usage_type t = {
565 .buckets = src.d[i].buckets,
566 .sectors = src.d[i].sectors,
567 .fragmented = src.d[i].fragmented,
570 ret = copy_to_user_errcode(&user_arg->d[i], &t, sizeof(t));
575 percpu_ref_put(&ca->ref);
579 static long bch2_ioctl_read_super(struct bch_fs *c,
580 struct bch_ioctl_read_super arg)
582 struct bch_dev *ca = NULL;
586 if (!capable(CAP_SYS_ADMIN))
589 if ((arg.flags & ~(BCH_BY_INDEX|BCH_READ_DEV)) ||
593 mutex_lock(&c->sb_lock);
595 if (arg.flags & BCH_READ_DEV) {
596 ca = bch2_device_lookup(c, arg.dev, arg.flags);
608 if (vstruct_bytes(sb) > arg.size) {
613 ret = copy_to_user_errcode((void __user *)(unsigned long)arg.sb, sb,
616 if (!IS_ERR_OR_NULL(ca))
617 percpu_ref_put(&ca->ref);
618 mutex_unlock(&c->sb_lock);
622 static long bch2_ioctl_disk_get_idx(struct bch_fs *c,
623 struct bch_ioctl_disk_get_idx arg)
625 dev_t dev = huge_decode_dev(arg.dev);
629 if (!capable(CAP_SYS_ADMIN))
635 for_each_online_member(ca, c, i)
636 if (ca->dev == dev) {
637 percpu_ref_put(&ca->io_ref);
641 return -BCH_ERR_ENOENT_dev_idx_not_found;
644 static long bch2_ioctl_disk_resize(struct bch_fs *c,
645 struct bch_ioctl_disk_resize arg)
650 if (!capable(CAP_SYS_ADMIN))
653 if ((arg.flags & ~BCH_BY_INDEX) ||
657 ca = bch2_device_lookup(c, arg.dev, arg.flags);
661 ret = bch2_dev_resize(c, ca, arg.nbuckets);
663 percpu_ref_put(&ca->ref);
667 static long bch2_ioctl_disk_resize_journal(struct bch_fs *c,
668 struct bch_ioctl_disk_resize_journal arg)
673 if (!capable(CAP_SYS_ADMIN))
676 if ((arg.flags & ~BCH_BY_INDEX) ||
680 if (arg.nbuckets > U32_MAX)
683 ca = bch2_device_lookup(c, arg.dev, arg.flags);
687 ret = bch2_set_nr_journal_buckets(c, ca, arg.nbuckets);
689 percpu_ref_put(&ca->ref);
693 #define BCH_IOCTL(_name, _argtype) \
697 if (copy_from_user(&i, arg, sizeof(i))) \
699 ret = bch2_ioctl_##_name(c, i); \
703 long bch2_fs_ioctl(struct bch_fs *c, unsigned cmd, void __user *arg)
708 case BCH_IOCTL_QUERY_UUID:
709 return bch2_ioctl_query_uuid(c, arg);
710 case BCH_IOCTL_FS_USAGE:
711 return bch2_ioctl_fs_usage(c, arg);
712 case BCH_IOCTL_DEV_USAGE:
713 return bch2_ioctl_dev_usage(c, arg);
714 case BCH_IOCTL_DEV_USAGE_V2:
715 return bch2_ioctl_dev_usage_v2(c, arg);
717 case BCH_IOCTL_START:
718 BCH_IOCTL(start, struct bch_ioctl_start);
720 return bch2_ioctl_stop(c);
722 case BCH_IOCTL_READ_SUPER:
723 BCH_IOCTL(read_super, struct bch_ioctl_read_super);
724 case BCH_IOCTL_DISK_GET_IDX:
725 BCH_IOCTL(disk_get_idx, struct bch_ioctl_disk_get_idx);
728 if (!test_bit(BCH_FS_STARTED, &c->flags))
732 case BCH_IOCTL_DISK_ADD:
733 BCH_IOCTL(disk_add, struct bch_ioctl_disk);
734 case BCH_IOCTL_DISK_REMOVE:
735 BCH_IOCTL(disk_remove, struct bch_ioctl_disk);
736 case BCH_IOCTL_DISK_ONLINE:
737 BCH_IOCTL(disk_online, struct bch_ioctl_disk);
738 case BCH_IOCTL_DISK_OFFLINE:
739 BCH_IOCTL(disk_offline, struct bch_ioctl_disk);
740 case BCH_IOCTL_DISK_SET_STATE:
741 BCH_IOCTL(disk_set_state, struct bch_ioctl_disk_set_state);
743 BCH_IOCTL(data, struct bch_ioctl_data);
744 case BCH_IOCTL_DISK_RESIZE:
745 BCH_IOCTL(disk_resize, struct bch_ioctl_disk_resize);
746 case BCH_IOCTL_DISK_RESIZE_JOURNAL:
747 BCH_IOCTL(disk_resize_journal, struct bch_ioctl_disk_resize_journal);
754 ret = bch2_err_class(ret);
758 static DEFINE_IDR(bch_chardev_minor);
760 static long bch2_chardev_ioctl(struct file *filp, unsigned cmd, unsigned long v)
762 unsigned minor = iminor(file_inode(filp));
763 struct bch_fs *c = minor < U8_MAX ? idr_find(&bch_chardev_minor, minor) : NULL;
764 void __user *arg = (void __user *) v;
767 ? bch2_fs_ioctl(c, cmd, arg)
768 : bch2_global_ioctl(cmd, arg);
771 static const struct file_operations bch_chardev_fops = {
772 .owner = THIS_MODULE,
773 .unlocked_ioctl = bch2_chardev_ioctl,
774 .open = nonseekable_open,
777 static int bch_chardev_major;
778 static struct class *bch_chardev_class;
779 static struct device *bch_chardev;
781 void bch2_fs_chardev_exit(struct bch_fs *c)
783 if (!IS_ERR_OR_NULL(c->chardev))
784 device_unregister(c->chardev);
786 idr_remove(&bch_chardev_minor, c->minor);
789 int bch2_fs_chardev_init(struct bch_fs *c)
791 c->minor = idr_alloc(&bch_chardev_minor, c, 0, 0, GFP_KERNEL);
795 c->chardev = device_create(bch_chardev_class, NULL,
796 MKDEV(bch_chardev_major, c->minor), c,
797 "bcachefs%u-ctl", c->minor);
798 if (IS_ERR(c->chardev))
799 return PTR_ERR(c->chardev);
804 void bch2_chardev_exit(void)
806 if (!IS_ERR_OR_NULL(bch_chardev_class))
807 device_destroy(bch_chardev_class,
808 MKDEV(bch_chardev_major, U8_MAX));
809 if (!IS_ERR_OR_NULL(bch_chardev_class))
810 class_destroy(bch_chardev_class);
811 if (bch_chardev_major > 0)
812 unregister_chrdev(bch_chardev_major, "bcachefs");
815 int __init bch2_chardev_init(void)
817 bch_chardev_major = register_chrdev(0, "bcachefs-ctl", &bch_chardev_fops);
818 if (bch_chardev_major < 0)
819 return bch_chardev_major;
821 bch_chardev_class = class_create("bcachefs");
822 if (IS_ERR(bch_chardev_class))
823 return PTR_ERR(bch_chardev_class);
825 bch_chardev = device_create(bch_chardev_class, NULL,
826 MKDEV(bch_chardev_major, U8_MAX),
827 NULL, "bcachefs-ctl");
828 if (IS_ERR(bch_chardev))
829 return PTR_ERR(bch_chardev);
834 #endif /* NO_BCACHEFS_CHARDEV */