1 // SPDX-License-Identifier: GPL-2.0
3 * Copyright (C) 2018 Cambridge Greys Ltd
4 * Copyright (C) 2015-2016 Anton Ivanov (aivanov@brocade.com)
5 * Copyright (C) 2000 Jeff Dike (jdike@karaya.com)
8 /* 2001-09-28...2002-04-17
9 * Partition stuff by James_McMechan@hotmail.com
10 * old style ubd by setting UBD_SHIFT to 0
11 * 2002-09-27...2002-10-18 massive tinkering for 2.5
12 * partitions have changed in 2.5
13 * 2003-01-29 more tinkering for 2.5.59-1
14 * This should now address the sysfs problems and has
15 * the symlink for devfs to allow for booting with
16 * the common /dev/ubd/discX/... names rather than
17 * only /dev/ubdN/discN this version also has lots of
18 * clean ups preparing for ubd-many.
24 #include <linux/module.h>
25 #include <linux/init.h>
26 #include <linux/blkdev.h>
27 #include <linux/blk-mq.h>
28 #include <linux/ata.h>
29 #include <linux/hdreg.h>
30 #include <linux/major.h>
31 #include <linux/cdrom.h>
32 #include <linux/proc_fs.h>
33 #include <linux/seq_file.h>
34 #include <linux/ctype.h>
35 #include <linux/slab.h>
36 #include <linux/vmalloc.h>
37 #include <linux/platform_device.h>
38 #include <linux/scatterlist.h>
39 #include <asm/tlbflush.h>
40 #include <kern_util.h>
41 #include "mconsole_kern.h"
48 /* Max request size is determined by sector mask - 32K */
49 #define UBD_MAX_REQUEST (8 * sizeof(long))
54 unsigned long sector_mask;
55 unsigned long long cow_offset;
56 unsigned long bitmap_words[2];
59 struct io_thread_req {
62 unsigned long offsets[2];
63 unsigned long long offset;
68 /* io_desc has to be the last element of the struct */
69 struct io_desc io_desc[];
73 static struct io_thread_req * (*irq_req_buffer)[];
74 static struct io_thread_req *irq_remainder;
75 static int irq_remainder_size;
77 static struct io_thread_req * (*io_req_buffer)[];
78 static struct io_thread_req *io_remainder;
79 static int io_remainder_size;
83 static inline int ubd_test_bit(__u64 bit, unsigned char *data)
88 bits = sizeof(data[0]) * 8;
91 return (data[n] & (1 << off)) != 0;
94 static inline void ubd_set_bit(__u64 bit, unsigned char *data)
99 bits = sizeof(data[0]) * 8;
102 data[n] |= (1 << off);
104 /*End stuff from ubd_user.h*/
106 #define DRIVER_NAME "uml-blkdev"
108 static DEFINE_MUTEX(ubd_lock);
109 static DEFINE_MUTEX(ubd_mutex); /* replaces BKL, might not be needed */
111 static int ubd_open(struct block_device *bdev, fmode_t mode);
112 static void ubd_release(struct gendisk *disk, fmode_t mode);
113 static int ubd_ioctl(struct block_device *bdev, fmode_t mode,
114 unsigned int cmd, unsigned long arg);
115 static int ubd_getgeo(struct block_device *bdev, struct hd_geometry *geo);
119 static const struct block_device_operations ubd_blops = {
120 .owner = THIS_MODULE,
122 .release = ubd_release,
124 .compat_ioctl = blkdev_compat_ptr_ioctl,
125 .getgeo = ubd_getgeo,
128 /* Protected by ubd_lock */
129 static struct gendisk *ubd_gendisk[MAX_DEV];
131 #ifdef CONFIG_BLK_DEV_UBD_SYNC
132 #define OPEN_FLAGS ((struct openflags) { .r = 1, .w = 1, .s = 1, .c = 0, \
135 #define OPEN_FLAGS ((struct openflags) { .r = 1, .w = 1, .s = 0, .c = 0, \
138 static struct openflags global_openflags = OPEN_FLAGS;
141 /* backing file name */
143 /* backing file fd */
145 unsigned long *bitmap;
146 unsigned long bitmap_len;
154 /* name (and fd, below) of the file opened for writing, either the
155 * backing or the cow file. */
161 struct openflags boot_openflags;
162 struct openflags openflags;
167 struct platform_device pdev;
168 struct request_queue *queue;
169 struct blk_mq_tag_set tag_set;
173 #define DEFAULT_COW { \
177 .bitmap_offset = 0, \
181 #define DEFAULT_UBD { \
187 .boot_openflags = OPEN_FLAGS, \
188 .openflags = OPEN_FLAGS, \
192 .cow = DEFAULT_COW, \
193 .lock = __SPIN_LOCK_UNLOCKED(ubd_devs.lock), \
196 /* Protected by ubd_lock */
197 static struct ubd ubd_devs[MAX_DEV] = { [0 ... MAX_DEV - 1] = DEFAULT_UBD };
199 static blk_status_t ubd_queue_rq(struct blk_mq_hw_ctx *hctx,
200 const struct blk_mq_queue_data *bd);
202 static int fake_ide_setup(char *str)
204 pr_warn("The fake_ide option has been removed\n");
207 __setup("fake_ide", fake_ide_setup);
209 __uml_help(fake_ide_setup,
211 " Obsolete stub.\n\n"
214 static int parse_unit(char **ptr)
216 char *str = *ptr, *end;
220 n = simple_strtoul(str, &end, 0);
225 else if (('a' <= *str) && (*str <= 'z')) {
233 /* If *index_out == -1 at exit, the passed option was a general one;
234 * otherwise, the str pointer is used (and owned) inside ubd_devs array, so it
235 * should not be freed on exit.
237 static int ubd_setup_common(char *str, int *index_out, char **error_out)
240 struct openflags flags = global_openflags;
241 char *file, *backing_file, *serial;
244 if(index_out) *index_out = -1;
248 if(!strcmp(str, "sync")){
249 global_openflags = of_sync(global_openflags);
253 pr_warn("fake major not supported any more\n");
257 n = parse_unit(&str);
259 *error_out = "Couldn't parse device number";
263 *error_out = "Device number out of range";
268 mutex_lock(&ubd_lock);
270 ubd_dev = &ubd_devs[n];
271 if(ubd_dev->file != NULL){
272 *error_out = "Device is already configured";
280 for (i = 0; i < sizeof("rscdt="); i++) {
295 ubd_dev->no_trim = 1;
301 *error_out = "Expected '=' or flag letter "
309 *error_out = "Too many flags specified";
311 *error_out = "Missing '='";
315 file = strsep(&str, ",:");
319 backing_file = strsep(&str, ",:");
320 if (backing_file && *backing_file == '\0')
323 serial = strsep(&str, ",:");
324 if (serial && *serial == '\0')
327 if (backing_file && ubd_dev->no_cow) {
328 *error_out = "Can't specify both 'd' and a cow file";
333 ubd_dev->file = file;
334 ubd_dev->cow.file = backing_file;
335 ubd_dev->serial = serial;
336 ubd_dev->boot_openflags = flags;
338 mutex_unlock(&ubd_lock);
342 static int ubd_setup(char *str)
347 err = ubd_setup_common(str, NULL, &error);
349 printk(KERN_ERR "Failed to initialize device with \"%s\" : "
354 __setup("ubd", ubd_setup);
355 __uml_help(ubd_setup,
356 "ubd<n><flags>=<filename>[(:|,)<filename2>][(:|,)<serial>]\n"
357 " This is used to associate a device with a file in the underlying\n"
358 " filesystem. When specifying two filenames, the first one is the\n"
359 " COW name and the second is the backing file name. As separator you can\n"
360 " use either a ':' or a ',': the first one allows writing things like;\n"
361 " ubd0=~/Uml/root_cow:~/Uml/root_backing_file\n"
362 " while with a ',' the shell would not expand the 2nd '~'.\n"
363 " When using only one filename, UML will detect whether to treat it like\n"
364 " a COW file or a backing file. To override this detection, add the 'd'\n"
366 " ubd0d=BackingFile\n"
367 " Usually, there is a filesystem in the file, but \n"
368 " that's not required. Swap devices containing swap files can be\n"
369 " specified like this. Also, a file which doesn't contain a\n"
370 " filesystem can have its contents read in the virtual \n"
371 " machine by running 'dd' on the device. <n> must be in the range\n"
372 " 0 to 7. Appending an 'r' to the number will cause that device\n"
373 " to be mounted read-only. For example ubd1r=./ext_fs. Appending\n"
374 " an 's' will cause data to be written to disk on the host immediately.\n"
375 " 'c' will cause the device to be treated as being shared between multiple\n"
376 " UMLs and file locking will be turned off - this is appropriate for a\n"
377 " cluster filesystem and inappropriate at almost all other times.\n\n"
378 " 't' will disable trim/discard support on the device (enabled by default).\n\n"
379 " An optional device serial number can be exposed using the serial parameter\n"
380 " on the cmdline which is exposed as a sysfs entry. This is particularly\n"
381 " useful when a unique number should be given to the device. Note when\n"
382 " specifying a label, the filename2 must be also presented. It can be\n"
383 " an empty string, in which case the backing file is not used:\n"
384 " ubd0=File,,Serial\n"
387 static int udb_setup(char *str)
389 printk("udb%s specified on command line is almost certainly a ubd -> "
394 __setup("udb", udb_setup);
395 __uml_help(udb_setup,
397 " This option is here solely to catch ubd -> udb typos, which can be\n"
398 " to impossible to catch visually unless you specifically look for\n"
399 " them. The only result of any option starting with 'udb' is an error\n"
400 " in the boot output.\n\n"
403 /* Only changed by ubd_init, which is an initcall. */
404 static int thread_fd = -1;
406 /* Function to read several request pointers at a time
407 * handling fractional reads if (and as) needed
410 static int bulk_req_safe_read(
412 struct io_thread_req * (*request_buffer)[],
413 struct io_thread_req **remainder,
421 if (*remainder_size > 0) {
423 (char *) request_buffer,
424 (char *) remainder, *remainder_size
431 ((char *) request_buffer) + *remainder_size,
432 sizeof(struct io_thread_req *)*max_recs
437 if ((n % sizeof(struct io_thread_req *)) > 0) {
439 * Read somehow returned not a multiple of dword
440 * theoretically possible, but never observed in the
441 * wild, so read routine must be able to handle it
443 *remainder_size = n % sizeof(struct io_thread_req *);
444 WARN(*remainder_size > 0, "UBD IPC read returned a partial result");
447 ((char *) request_buffer) +
448 (n/sizeof(struct io_thread_req *))*sizeof(struct io_thread_req *),
451 n = n - *remainder_size;
459 /* Called without dev->lock held, and only in interrupt context. */
460 static void ubd_handler(void)
466 n = bulk_req_safe_read(
476 printk(KERN_ERR "spurious interrupt in ubd_handler, "
480 for (count = 0; count < n/sizeof(struct io_thread_req *); count++) {
481 struct io_thread_req *io_req = (*irq_req_buffer)[count];
483 if ((io_req->error == BLK_STS_NOTSUPP) && (req_op(io_req->req) == REQ_OP_DISCARD)) {
484 blk_queue_max_discard_sectors(io_req->req->q, 0);
485 blk_queue_max_write_zeroes_sectors(io_req->req->q, 0);
486 blk_queue_flag_clear(QUEUE_FLAG_DISCARD, io_req->req->q);
488 blk_mq_end_request(io_req->req, io_req->error);
494 static irqreturn_t ubd_intr(int irq, void *dev)
500 /* Only changed by ubd_init, which is an initcall. */
501 static int io_pid = -1;
503 static void kill_io_thread(void)
506 os_kill_process(io_pid, 1);
509 __uml_exitcall(kill_io_thread);
511 static inline int ubd_file_size(struct ubd *ubd_dev, __u64 *size_out)
521 unsigned long long size;
525 if (ubd_dev->file && ubd_dev->cow.file) {
526 file = ubd_dev->cow.file;
531 fd = os_open_file(ubd_dev->file, of_read(OPENFLAGS()), 0);
535 err = read_cow_header(file_reader, &fd, &version, &backing_file, \
536 &mtime, &size, §or_size, &align, &bitmap_offset);
540 file = ubd_dev->file;
545 return os_file_size(file, size_out);
548 static int read_cow_bitmap(int fd, void *buf, int offset, int len)
552 err = os_pread_file(fd, buf, len, offset);
559 static int backing_file_mismatch(char *file, __u64 size, time64_t mtime)
562 unsigned long long actual;
565 err = os_file_modtime(file, &modtime);
567 printk(KERN_ERR "Failed to get modification time of backing "
568 "file \"%s\", err = %d\n", file, -err);
572 err = os_file_size(file, &actual);
574 printk(KERN_ERR "Failed to get size of backing file \"%s\", "
575 "err = %d\n", file, -err);
579 if (actual != size) {
580 /*__u64 can be a long on AMD64 and with %lu GCC complains; so
582 printk(KERN_ERR "Size mismatch (%llu vs %llu) of COW header "
583 "vs backing file\n", (unsigned long long) size, actual);
586 if (modtime != mtime) {
587 printk(KERN_ERR "mtime mismatch (%lld vs %lld) of COW header vs "
588 "backing file\n", mtime, modtime);
594 static int path_requires_switch(char *from_cmdline, char *from_cow, char *cow)
596 struct uml_stat buf1, buf2;
599 if (from_cmdline == NULL)
601 if (!strcmp(from_cmdline, from_cow))
604 err = os_stat_file(from_cmdline, &buf1);
606 printk(KERN_ERR "Couldn't stat '%s', err = %d\n", from_cmdline,
610 err = os_stat_file(from_cow, &buf2);
612 printk(KERN_ERR "Couldn't stat '%s', err = %d\n", from_cow,
616 if ((buf1.ust_dev == buf2.ust_dev) && (buf1.ust_ino == buf2.ust_ino))
619 printk(KERN_ERR "Backing file mismatch - \"%s\" requested, "
620 "\"%s\" specified in COW header of \"%s\"\n",
621 from_cmdline, from_cow, cow);
625 static int open_ubd_file(char *file, struct openflags *openflags, int shared,
626 char **backing_file_out, int *bitmap_offset_out,
627 unsigned long *bitmap_len_out, int *data_offset_out,
631 unsigned long long size;
632 __u32 version, align;
634 int fd, err, sectorsize, asked_switch, mode = 0644;
636 fd = os_open_file(file, *openflags, mode);
638 if ((fd == -ENOENT) && (create_cow_out != NULL))
641 ((fd != -EROFS) && (fd != -EACCES)))
644 fd = os_open_file(file, *openflags, mode);
650 printk(KERN_INFO "Not locking \"%s\" on the host\n", file);
652 err = os_lock_file(fd, openflags->w);
654 printk(KERN_ERR "Failed to lock '%s', err = %d\n",
660 /* Successful return case! */
661 if (backing_file_out == NULL)
664 err = read_cow_header(file_reader, &fd, &version, &backing_file, &mtime,
665 &size, §orsize, &align, bitmap_offset_out);
666 if (err && (*backing_file_out != NULL)) {
667 printk(KERN_ERR "Failed to read COW header from COW file "
668 "\"%s\", errno = %d\n", file, -err);
674 asked_switch = path_requires_switch(*backing_file_out, backing_file,
677 /* Allow switching only if no mismatch. */
678 if (asked_switch && !backing_file_mismatch(*backing_file_out, size,
680 printk(KERN_ERR "Switching backing file to '%s'\n",
682 err = write_cow_header(file, fd, *backing_file_out,
683 sectorsize, align, &size);
685 printk(KERN_ERR "Switch failed, errno = %d\n", -err);
689 *backing_file_out = backing_file;
690 err = backing_file_mismatch(*backing_file_out, size, mtime);
695 cow_sizes(version, size, sectorsize, align, *bitmap_offset_out,
696 bitmap_len_out, data_offset_out);
704 static int create_cow_file(char *cow_file, char *backing_file,
705 struct openflags flags,
706 int sectorsize, int alignment, int *bitmap_offset_out,
707 unsigned long *bitmap_len_out, int *data_offset_out)
712 fd = open_ubd_file(cow_file, &flags, 0, NULL, NULL, NULL, NULL, NULL);
715 printk(KERN_ERR "Open of COW file '%s' failed, errno = %d\n",
720 err = init_cow_file(fd, cow_file, backing_file, sectorsize, alignment,
721 bitmap_offset_out, bitmap_len_out,
730 static void ubd_close_dev(struct ubd *ubd_dev)
732 os_close_file(ubd_dev->fd);
733 if(ubd_dev->cow.file == NULL)
736 os_close_file(ubd_dev->cow.fd);
737 vfree(ubd_dev->cow.bitmap);
738 ubd_dev->cow.bitmap = NULL;
741 static int ubd_open_dev(struct ubd *ubd_dev)
743 struct openflags flags;
745 int err, create_cow, *create_ptr;
748 ubd_dev->openflags = ubd_dev->boot_openflags;
750 create_ptr = (ubd_dev->cow.file != NULL) ? &create_cow : NULL;
751 back_ptr = ubd_dev->no_cow ? NULL : &ubd_dev->cow.file;
753 fd = open_ubd_file(ubd_dev->file, &ubd_dev->openflags, ubd_dev->shared,
754 back_ptr, &ubd_dev->cow.bitmap_offset,
755 &ubd_dev->cow.bitmap_len, &ubd_dev->cow.data_offset,
758 if((fd == -ENOENT) && create_cow){
759 fd = create_cow_file(ubd_dev->file, ubd_dev->cow.file,
760 ubd_dev->openflags, SECTOR_SIZE, PAGE_SIZE,
761 &ubd_dev->cow.bitmap_offset,
762 &ubd_dev->cow.bitmap_len,
763 &ubd_dev->cow.data_offset);
765 printk(KERN_INFO "Creating \"%s\" as COW file for "
766 "\"%s\"\n", ubd_dev->file, ubd_dev->cow.file);
771 printk("Failed to open '%s', errno = %d\n", ubd_dev->file,
777 if(ubd_dev->cow.file != NULL){
778 blk_queue_max_hw_sectors(ubd_dev->queue, 8 * sizeof(long));
781 ubd_dev->cow.bitmap = vmalloc(ubd_dev->cow.bitmap_len);
782 if(ubd_dev->cow.bitmap == NULL){
783 printk(KERN_ERR "Failed to vmalloc COW bitmap\n");
786 flush_tlb_kernel_vm();
788 err = read_cow_bitmap(ubd_dev->fd, ubd_dev->cow.bitmap,
789 ubd_dev->cow.bitmap_offset,
790 ubd_dev->cow.bitmap_len);
794 flags = ubd_dev->openflags;
796 err = open_ubd_file(ubd_dev->cow.file, &flags, ubd_dev->shared, NULL,
797 NULL, NULL, NULL, NULL);
798 if(err < 0) goto error;
799 ubd_dev->cow.fd = err;
801 if (ubd_dev->no_trim == 0) {
802 ubd_dev->queue->limits.discard_granularity = SECTOR_SIZE;
803 ubd_dev->queue->limits.discard_alignment = SECTOR_SIZE;
804 blk_queue_max_discard_sectors(ubd_dev->queue, UBD_MAX_REQUEST);
805 blk_queue_max_write_zeroes_sectors(ubd_dev->queue, UBD_MAX_REQUEST);
806 blk_queue_flag_set(QUEUE_FLAG_DISCARD, ubd_dev->queue);
808 blk_queue_flag_set(QUEUE_FLAG_NONROT, ubd_dev->queue);
811 os_close_file(ubd_dev->fd);
815 static void ubd_device_release(struct device *dev)
817 struct ubd *ubd_dev = dev_get_drvdata(dev);
819 blk_mq_free_tag_set(&ubd_dev->tag_set);
820 *ubd_dev = ((struct ubd) DEFAULT_UBD);
823 static ssize_t serial_show(struct device *dev,
824 struct device_attribute *attr, char *buf)
826 struct gendisk *disk = dev_to_disk(dev);
827 struct ubd *ubd_dev = disk->private_data;
832 return sprintf(buf, "%s", ubd_dev->serial);
835 static DEVICE_ATTR_RO(serial);
837 static struct attribute *ubd_attrs[] = {
838 &dev_attr_serial.attr,
842 static umode_t ubd_attrs_are_visible(struct kobject *kobj,
843 struct attribute *a, int n)
848 static const struct attribute_group ubd_attr_group = {
850 .is_visible = ubd_attrs_are_visible,
853 static const struct attribute_group *ubd_attr_groups[] = {
858 static int ubd_disk_register(int major, u64 size, int unit,
859 struct gendisk *disk)
862 disk->first_minor = unit << UBD_SHIFT;
863 disk->minors = 1 << UBD_SHIFT;
864 disk->fops = &ubd_blops;
865 set_capacity(disk, size / 512);
866 sprintf(disk->disk_name, "ubd%c", 'a' + unit);
868 ubd_devs[unit].pdev.id = unit;
869 ubd_devs[unit].pdev.name = DRIVER_NAME;
870 ubd_devs[unit].pdev.dev.release = ubd_device_release;
871 dev_set_drvdata(&ubd_devs[unit].pdev.dev, &ubd_devs[unit]);
872 platform_device_register(&ubd_devs[unit].pdev);
874 disk->private_data = &ubd_devs[unit];
875 disk->queue = ubd_devs[unit].queue;
876 return device_add_disk(&ubd_devs[unit].pdev.dev, disk, ubd_attr_groups);
879 #define ROUND_BLOCK(n) ((n + (SECTOR_SIZE - 1)) & (-SECTOR_SIZE))
881 static const struct blk_mq_ops ubd_mq_ops = {
882 .queue_rq = ubd_queue_rq,
885 static int ubd_add(int n, char **error_out)
887 struct ubd *ubd_dev = &ubd_devs[n];
888 struct gendisk *disk;
891 if(ubd_dev->file == NULL)
894 err = ubd_file_size(ubd_dev, &ubd_dev->size);
896 *error_out = "Couldn't determine size of device's file";
900 ubd_dev->size = ROUND_BLOCK(ubd_dev->size);
902 ubd_dev->tag_set.ops = &ubd_mq_ops;
903 ubd_dev->tag_set.queue_depth = 64;
904 ubd_dev->tag_set.numa_node = NUMA_NO_NODE;
905 ubd_dev->tag_set.flags = BLK_MQ_F_SHOULD_MERGE;
906 ubd_dev->tag_set.driver_data = ubd_dev;
907 ubd_dev->tag_set.nr_hw_queues = 1;
909 err = blk_mq_alloc_tag_set(&ubd_dev->tag_set);
913 disk = blk_mq_alloc_disk(&ubd_dev->tag_set, ubd_dev);
916 goto out_cleanup_tags;
918 ubd_dev->queue = disk->queue;
920 blk_queue_write_cache(ubd_dev->queue, true, false);
921 blk_queue_max_segments(ubd_dev->queue, MAX_SG);
922 blk_queue_segment_boundary(ubd_dev->queue, PAGE_SIZE - 1);
923 err = ubd_disk_register(UBD_MAJOR, ubd_dev->size, n, disk);
925 goto out_cleanup_disk;
927 ubd_gendisk[n] = disk;
931 blk_cleanup_disk(disk);
933 blk_mq_free_tag_set(&ubd_dev->tag_set);
938 static int ubd_config(char *str, char **error_out)
942 /* This string is possibly broken up and stored, so it's only
943 * freed if ubd_setup_common fails, or if only general options
946 str = kstrdup(str, GFP_KERNEL);
948 *error_out = "Failed to allocate memory";
952 ret = ubd_setup_common(str, &n, error_out);
961 mutex_lock(&ubd_lock);
962 ret = ubd_add(n, error_out);
964 ubd_devs[n].file = NULL;
965 mutex_unlock(&ubd_lock);
975 static int ubd_get_config(char *name, char *str, int size, char **error_out)
980 n = parse_unit(&name);
981 if((n >= MAX_DEV) || (n < 0)){
982 *error_out = "ubd_get_config : device number out of range";
986 ubd_dev = &ubd_devs[n];
987 mutex_lock(&ubd_lock);
989 if(ubd_dev->file == NULL){
990 CONFIG_CHUNK(str, size, len, "", 1);
994 CONFIG_CHUNK(str, size, len, ubd_dev->file, 0);
996 if(ubd_dev->cow.file != NULL){
997 CONFIG_CHUNK(str, size, len, ",", 0);
998 CONFIG_CHUNK(str, size, len, ubd_dev->cow.file, 1);
1000 else CONFIG_CHUNK(str, size, len, "", 1);
1003 mutex_unlock(&ubd_lock);
1007 static int ubd_id(char **str, int *start_out, int *end_out)
1011 n = parse_unit(str);
1013 *end_out = MAX_DEV - 1;
1017 static int ubd_remove(int n, char **error_out)
1019 struct gendisk *disk = ubd_gendisk[n];
1020 struct ubd *ubd_dev;
1023 mutex_lock(&ubd_lock);
1025 ubd_dev = &ubd_devs[n];
1027 if(ubd_dev->file == NULL)
1030 /* you cannot remove a open disk */
1032 if(ubd_dev->count > 0)
1035 ubd_gendisk[n] = NULL;
1038 blk_cleanup_disk(disk);
1042 platform_device_unregister(&ubd_dev->pdev);
1044 mutex_unlock(&ubd_lock);
1048 /* All these are called by mconsole in process context and without
1049 * ubd-specific locks. The structure itself is const except for .list.
1051 static struct mc_device ubd_mc = {
1052 .list = LIST_HEAD_INIT(ubd_mc.list),
1054 .config = ubd_config,
1055 .get_config = ubd_get_config,
1057 .remove = ubd_remove,
1060 static int __init ubd_mc_init(void)
1062 mconsole_register_dev(&ubd_mc);
1066 __initcall(ubd_mc_init);
1068 static int __init ubd0_init(void)
1070 struct ubd *ubd_dev = &ubd_devs[0];
1072 mutex_lock(&ubd_lock);
1073 if(ubd_dev->file == NULL)
1074 ubd_dev->file = "root_fs";
1075 mutex_unlock(&ubd_lock);
1080 __initcall(ubd0_init);
1082 /* Used in ubd_init, which is an initcall */
1083 static struct platform_driver ubd_driver = {
1085 .name = DRIVER_NAME,
1089 static int __init ubd_init(void)
1094 if (register_blkdev(UBD_MAJOR, "ubd"))
1097 irq_req_buffer = kmalloc_array(UBD_REQ_BUFFER_SIZE,
1098 sizeof(struct io_thread_req *),
1103 if (irq_req_buffer == NULL) {
1104 printk(KERN_ERR "Failed to initialize ubd buffering\n");
1107 io_req_buffer = kmalloc_array(UBD_REQ_BUFFER_SIZE,
1108 sizeof(struct io_thread_req *),
1114 if (io_req_buffer == NULL) {
1115 printk(KERN_ERR "Failed to initialize ubd buffering\n");
1118 platform_driver_register(&ubd_driver);
1119 mutex_lock(&ubd_lock);
1120 for (i = 0; i < MAX_DEV; i++){
1121 err = ubd_add(i, &error);
1123 printk(KERN_ERR "Failed to initialize ubd device %d :"
1126 mutex_unlock(&ubd_lock);
1130 late_initcall(ubd_init);
1132 static int __init ubd_driver_init(void){
1133 unsigned long stack;
1136 /* Set by CONFIG_BLK_DEV_UBD_SYNC or ubd=sync.*/
1137 if(global_openflags.s){
1138 printk(KERN_INFO "ubd: Synchronous mode\n");
1139 /* Letting ubd=sync be like using ubd#s= instead of ubd#= is
1140 * enough. So use anyway the io thread. */
1142 stack = alloc_stack(0, 0);
1143 io_pid = start_io_thread(stack + PAGE_SIZE, &thread_fd);
1146 "ubd : Failed to start I/O thread (errno = %d) - "
1147 "falling back to synchronous I/O\n", -io_pid);
1151 err = um_request_irq(UBD_IRQ, thread_fd, IRQ_READ, ubd_intr,
1152 0, "ubd", ubd_devs);
1154 printk(KERN_ERR "um_request_irq failed - errno = %d\n", -err);
1158 device_initcall(ubd_driver_init);
1160 static int ubd_open(struct block_device *bdev, fmode_t mode)
1162 struct gendisk *disk = bdev->bd_disk;
1163 struct ubd *ubd_dev = disk->private_data;
1166 mutex_lock(&ubd_mutex);
1167 if(ubd_dev->count == 0){
1168 err = ubd_open_dev(ubd_dev);
1170 printk(KERN_ERR "%s: Can't open \"%s\": errno = %d\n",
1171 disk->disk_name, ubd_dev->file, -err);
1176 set_disk_ro(disk, !ubd_dev->openflags.w);
1178 /* This should no more be needed. And it didn't work anyway to exclude
1179 * read-write remounting of filesystems.*/
1180 /*if((mode & FMODE_WRITE) && !ubd_dev->openflags.w){
1181 if(--ubd_dev->count == 0) ubd_close_dev(ubd_dev);
1185 mutex_unlock(&ubd_mutex);
1189 static void ubd_release(struct gendisk *disk, fmode_t mode)
1191 struct ubd *ubd_dev = disk->private_data;
1193 mutex_lock(&ubd_mutex);
1194 if(--ubd_dev->count == 0)
1195 ubd_close_dev(ubd_dev);
1196 mutex_unlock(&ubd_mutex);
1199 static void cowify_bitmap(__u64 io_offset, int length, unsigned long *cow_mask,
1200 __u64 *cow_offset, unsigned long *bitmap,
1201 __u64 bitmap_offset, unsigned long *bitmap_words,
1204 __u64 sector = io_offset >> SECTOR_SHIFT;
1205 int i, update_bitmap = 0;
1207 for (i = 0; i < length >> SECTOR_SHIFT; i++) {
1208 if(cow_mask != NULL)
1209 ubd_set_bit(i, (unsigned char *) cow_mask);
1210 if(ubd_test_bit(sector + i, (unsigned char *) bitmap))
1214 ubd_set_bit(sector + i, (unsigned char *) bitmap);
1220 *cow_offset = sector / (sizeof(unsigned long) * 8);
1222 /* This takes care of the case where we're exactly at the end of the
1223 * device, and *cow_offset + 1 is off the end. So, just back it up
1224 * by one word. Thanks to Lynn Kerby for the fix and James McMechan
1225 * for the original diagnosis.
1227 if (*cow_offset == (DIV_ROUND_UP(bitmap_len,
1228 sizeof(unsigned long)) - 1))
1231 bitmap_words[0] = bitmap[*cow_offset];
1232 bitmap_words[1] = bitmap[*cow_offset + 1];
1234 *cow_offset *= sizeof(unsigned long);
1235 *cow_offset += bitmap_offset;
1238 static void cowify_req(struct io_thread_req *req, struct io_desc *segment,
1239 unsigned long offset, unsigned long *bitmap,
1240 __u64 bitmap_offset, __u64 bitmap_len)
1242 __u64 sector = offset >> SECTOR_SHIFT;
1245 if (segment->length > (sizeof(segment->sector_mask) * 8) << SECTOR_SHIFT)
1246 panic("Operation too long");
1248 if (req_op(req->req) == REQ_OP_READ) {
1249 for (i = 0; i < segment->length >> SECTOR_SHIFT; i++) {
1250 if(ubd_test_bit(sector + i, (unsigned char *) bitmap))
1251 ubd_set_bit(i, (unsigned char *)
1252 &segment->sector_mask);
1255 cowify_bitmap(offset, segment->length, &segment->sector_mask,
1256 &segment->cow_offset, bitmap, bitmap_offset,
1257 segment->bitmap_words, bitmap_len);
1261 static void ubd_map_req(struct ubd *dev, struct io_thread_req *io_req,
1262 struct request *req)
1264 struct bio_vec bvec;
1265 struct req_iterator iter;
1267 unsigned long byte_offset = io_req->offset;
1268 int op = req_op(req);
1270 if (op == REQ_OP_WRITE_ZEROES || op == REQ_OP_DISCARD) {
1271 io_req->io_desc[0].buffer = NULL;
1272 io_req->io_desc[0].length = blk_rq_bytes(req);
1274 rq_for_each_segment(bvec, req, iter) {
1275 BUG_ON(i >= io_req->desc_cnt);
1277 io_req->io_desc[i].buffer = bvec_virt(&bvec);
1278 io_req->io_desc[i].length = bvec.bv_len;
1283 if (dev->cow.file) {
1284 for (i = 0; i < io_req->desc_cnt; i++) {
1285 cowify_req(io_req, &io_req->io_desc[i], byte_offset,
1286 dev->cow.bitmap, dev->cow.bitmap_offset,
1287 dev->cow.bitmap_len);
1288 byte_offset += io_req->io_desc[i].length;
1294 static struct io_thread_req *ubd_alloc_req(struct ubd *dev, struct request *req,
1297 struct io_thread_req *io_req;
1300 io_req = kmalloc(sizeof(*io_req) +
1301 (desc_cnt * sizeof(struct io_desc)),
1308 io_req->fds[0] = dev->cow.fd;
1310 io_req->fds[0] = dev->fd;
1312 io_req->sectorsize = SECTOR_SIZE;
1313 io_req->fds[1] = dev->fd;
1314 io_req->offset = (u64) blk_rq_pos(req) << SECTOR_SHIFT;
1315 io_req->offsets[0] = 0;
1316 io_req->offsets[1] = dev->cow.data_offset;
1318 for (i = 0 ; i < desc_cnt; i++) {
1319 io_req->io_desc[i].sector_mask = 0;
1320 io_req->io_desc[i].cow_offset = -1;
1326 static int ubd_submit_request(struct ubd *dev, struct request *req)
1329 struct io_thread_req *io_req;
1331 int op = req_op(req);
1333 if (op == REQ_OP_FLUSH)
1335 else if (op == REQ_OP_WRITE_ZEROES || op == REQ_OP_DISCARD)
1338 segs = blk_rq_nr_phys_segments(req);
1340 io_req = ubd_alloc_req(dev, req, segs);
1344 io_req->desc_cnt = segs;
1346 ubd_map_req(dev, io_req, req);
1348 ret = os_write_file(thread_fd, &io_req, sizeof(io_req));
1349 if (ret != sizeof(io_req)) {
1351 pr_err("write to io thread failed: %d\n", -ret);
1357 static blk_status_t ubd_queue_rq(struct blk_mq_hw_ctx *hctx,
1358 const struct blk_mq_queue_data *bd)
1360 struct ubd *ubd_dev = hctx->queue->queuedata;
1361 struct request *req = bd->rq;
1362 int ret = 0, res = BLK_STS_OK;
1364 blk_mq_start_request(req);
1366 spin_lock_irq(&ubd_dev->lock);
1368 switch (req_op(req)) {
1372 case REQ_OP_DISCARD:
1373 case REQ_OP_WRITE_ZEROES:
1374 ret = ubd_submit_request(ubd_dev, req);
1378 res = BLK_STS_NOTSUPP;
1381 spin_unlock_irq(&ubd_dev->lock);
1385 res = BLK_STS_RESOURCE;
1387 res = BLK_STS_DEV_RESOURCE;
1393 static int ubd_getgeo(struct block_device *bdev, struct hd_geometry *geo)
1395 struct ubd *ubd_dev = bdev->bd_disk->private_data;
1399 geo->cylinders = ubd_dev->size / (128 * 32 * 512);
1403 static int ubd_ioctl(struct block_device *bdev, fmode_t mode,
1404 unsigned int cmd, unsigned long arg)
1406 struct ubd *ubd_dev = bdev->bd_disk->private_data;
1407 u16 ubd_id[ATA_ID_WORDS];
1410 struct cdrom_volctrl volume;
1411 case HDIO_GET_IDENTITY:
1412 memset(&ubd_id, 0, ATA_ID_WORDS * 2);
1413 ubd_id[ATA_ID_CYLS] = ubd_dev->size / (128 * 32 * 512);
1414 ubd_id[ATA_ID_HEADS] = 128;
1415 ubd_id[ATA_ID_SECTORS] = 32;
1416 if(copy_to_user((char __user *) arg, (char *) &ubd_id,
1422 if(copy_from_user(&volume, (char __user *) arg, sizeof(volume)))
1424 volume.channel0 = 255;
1425 volume.channel1 = 255;
1426 volume.channel2 = 255;
1427 volume.channel3 = 255;
1428 if(copy_to_user((char __user *) arg, &volume, sizeof(volume)))
1435 static int map_error(int error_code)
1437 switch (error_code) {
1442 return BLK_STS_NOTSUPP;
1444 return BLK_STS_NOSPC;
1446 return BLK_STS_IOERR;
1450 * Everything from here onwards *IS NOT PART OF THE KERNEL*
1452 * The following functions are part of UML hypervisor code.
1453 * All functions from here onwards are executed as a helper
1454 * thread and are not allowed to execute any kernel functions.
1456 * Any communication must occur strictly via shared memory and IPC.
1458 * Do not add printks, locks, kernel memory operations, etc - it
1459 * will result in unpredictable behaviour and/or crashes.
1462 static int update_bitmap(struct io_thread_req *req, struct io_desc *segment)
1466 if (segment->cow_offset == -1)
1467 return map_error(0);
1469 n = os_pwrite_file(req->fds[1], &segment->bitmap_words,
1470 sizeof(segment->bitmap_words), segment->cow_offset);
1471 if (n != sizeof(segment->bitmap_words))
1472 return map_error(-n);
1474 return map_error(0);
1477 static void do_io(struct io_thread_req *req, struct io_desc *desc)
1481 int n, nsectors, start, end, bit;
1484 /* FLUSH is really a special case, we cannot "case" it with others */
1486 if (req_op(req->req) == REQ_OP_FLUSH) {
1487 /* fds[0] is always either the rw image or our cow file */
1488 req->error = map_error(-os_sync_file(req->fds[0]));
1492 nsectors = desc->length / req->sectorsize;
1495 bit = ubd_test_bit(start, (unsigned char *) &desc->sector_mask);
1497 while((end < nsectors) &&
1498 (ubd_test_bit(end, (unsigned char *) &desc->sector_mask) == bit))
1501 off = req->offset + req->offsets[bit] +
1502 start * req->sectorsize;
1503 len = (end - start) * req->sectorsize;
1504 if (desc->buffer != NULL)
1505 buf = &desc->buffer[start * req->sectorsize];
1507 switch (req_op(req->req)) {
1513 n = os_pread_file(req->fds[bit], buf, len, off);
1515 req->error = map_error(-n);
1518 } while((n < len) && (n != 0));
1519 if (n < len) memset(&buf[n], 0, len - n);
1522 n = os_pwrite_file(req->fds[bit], buf, len, off);
1524 req->error = map_error(-n);
1528 case REQ_OP_DISCARD:
1529 case REQ_OP_WRITE_ZEROES:
1530 n = os_falloc_punch(req->fds[bit], off, len);
1532 req->error = map_error(-n);
1538 req->error = BLK_STS_NOTSUPP;
1543 } while(start < nsectors);
1546 req->error = update_bitmap(req, desc);
1549 /* Changed in start_io_thread, which is serialized by being called only
1550 * from ubd_init, which is an initcall.
1554 /* Only changed by the io thread. XXX: currently unused. */
1555 static int io_count = 0;
1557 int io_thread(void *arg)
1559 int n, count, written, res;
1561 os_fix_helper_signals();
1564 n = bulk_req_safe_read(
1578 for (count = 0; count < n/sizeof(struct io_thread_req *); count++) {
1579 struct io_thread_req *req = (*io_req_buffer)[count];
1583 for (i = 0; !req->error && i < req->desc_cnt; i++)
1584 do_io(req, &(req->io_desc[i]));
1591 res = os_write_file(kernel_fd,
1592 ((char *) io_req_buffer) + written,
1600 } while (written < n);