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/cdrom.h>
31 #include <linux/proc_fs.h>
32 #include <linux/seq_file.h>
33 #include <linux/ctype.h>
34 #include <linux/slab.h>
35 #include <linux/vmalloc.h>
36 #include <linux/platform_device.h>
37 #include <linux/scatterlist.h>
38 #include <asm/tlbflush.h>
39 #include <kern_util.h>
40 #include "mconsole_kern.h"
47 /* Max request size is determined by sector mask - 32K */
48 #define UBD_MAX_REQUEST (8 * sizeof(long))
53 unsigned long sector_mask;
54 unsigned long long cow_offset;
55 unsigned long bitmap_words[2];
58 struct io_thread_req {
61 unsigned long offsets[2];
62 unsigned long long offset;
67 /* io_desc has to be the last element of the struct */
68 struct io_desc io_desc[];
72 static struct io_thread_req * (*irq_req_buffer)[];
73 static struct io_thread_req *irq_remainder;
74 static int irq_remainder_size;
76 static struct io_thread_req * (*io_req_buffer)[];
77 static struct io_thread_req *io_remainder;
78 static int io_remainder_size;
82 static inline int ubd_test_bit(__u64 bit, unsigned char *data)
87 bits = sizeof(data[0]) * 8;
90 return (data[n] & (1 << off)) != 0;
93 static inline void ubd_set_bit(__u64 bit, unsigned char *data)
98 bits = sizeof(data[0]) * 8;
101 data[n] |= (1 << off);
103 /*End stuff from ubd_user.h*/
105 #define DRIVER_NAME "uml-blkdev"
107 static DEFINE_MUTEX(ubd_lock);
108 static DEFINE_MUTEX(ubd_mutex); /* replaces BKL, might not be needed */
110 static int ubd_open(struct block_device *bdev, fmode_t mode);
111 static void ubd_release(struct gendisk *disk, fmode_t mode);
112 static int ubd_ioctl(struct block_device *bdev, fmode_t mode,
113 unsigned int cmd, unsigned long arg);
114 static int ubd_getgeo(struct block_device *bdev, struct hd_geometry *geo);
118 static const struct block_device_operations ubd_blops = {
119 .owner = THIS_MODULE,
121 .release = ubd_release,
123 .compat_ioctl = blkdev_compat_ptr_ioctl,
124 .getgeo = ubd_getgeo,
127 /* Protected by ubd_lock */
128 static struct gendisk *ubd_gendisk[MAX_DEV];
130 #ifdef CONFIG_BLK_DEV_UBD_SYNC
131 #define OPEN_FLAGS ((struct openflags) { .r = 1, .w = 1, .s = 1, .c = 0, \
134 #define OPEN_FLAGS ((struct openflags) { .r = 1, .w = 1, .s = 0, .c = 0, \
137 static struct openflags global_openflags = OPEN_FLAGS;
140 /* backing file name */
142 /* backing file fd */
144 unsigned long *bitmap;
145 unsigned long bitmap_len;
153 /* name (and fd, below) of the file opened for writing, either the
154 * backing or the cow file. */
160 struct openflags boot_openflags;
161 struct openflags openflags;
166 struct platform_device pdev;
167 struct request_queue *queue;
168 struct blk_mq_tag_set tag_set;
172 #define DEFAULT_COW { \
176 .bitmap_offset = 0, \
180 #define DEFAULT_UBD { \
186 .boot_openflags = OPEN_FLAGS, \
187 .openflags = OPEN_FLAGS, \
191 .cow = DEFAULT_COW, \
192 .lock = __SPIN_LOCK_UNLOCKED(ubd_devs.lock), \
195 /* Protected by ubd_lock */
196 static struct ubd ubd_devs[MAX_DEV] = { [0 ... MAX_DEV - 1] = DEFAULT_UBD };
198 static blk_status_t ubd_queue_rq(struct blk_mq_hw_ctx *hctx,
199 const struct blk_mq_queue_data *bd);
201 static int fake_ide_setup(char *str)
203 pr_warn("The fake_ide option has been removed\n");
206 __setup("fake_ide", fake_ide_setup);
208 __uml_help(fake_ide_setup,
210 " Obsolete stub.\n\n"
213 static int parse_unit(char **ptr)
215 char *str = *ptr, *end;
219 n = simple_strtoul(str, &end, 0);
224 else if (('a' <= *str) && (*str <= 'z')) {
232 /* If *index_out == -1 at exit, the passed option was a general one;
233 * otherwise, the str pointer is used (and owned) inside ubd_devs array, so it
234 * should not be freed on exit.
236 static int ubd_setup_common(char *str, int *index_out, char **error_out)
239 struct openflags flags = global_openflags;
240 char *file, *backing_file, *serial;
243 if(index_out) *index_out = -1;
247 if(!strcmp(str, "sync")){
248 global_openflags = of_sync(global_openflags);
252 pr_warn("fake major not supported any more\n");
256 n = parse_unit(&str);
258 *error_out = "Couldn't parse device number";
262 *error_out = "Device number out of range";
267 mutex_lock(&ubd_lock);
269 ubd_dev = &ubd_devs[n];
270 if(ubd_dev->file != NULL){
271 *error_out = "Device is already configured";
279 for (i = 0; i < sizeof("rscdt="); i++) {
294 ubd_dev->no_trim = 1;
300 *error_out = "Expected '=' or flag letter "
308 *error_out = "Too many flags specified";
310 *error_out = "Missing '='";
314 file = strsep(&str, ",:");
318 backing_file = strsep(&str, ",:");
319 if (backing_file && *backing_file == '\0')
322 serial = strsep(&str, ",:");
323 if (serial && *serial == '\0')
326 if (backing_file && ubd_dev->no_cow) {
327 *error_out = "Can't specify both 'd' and a cow file";
332 ubd_dev->file = file;
333 ubd_dev->cow.file = backing_file;
334 ubd_dev->serial = serial;
335 ubd_dev->boot_openflags = flags;
337 mutex_unlock(&ubd_lock);
341 static int ubd_setup(char *str)
346 err = ubd_setup_common(str, NULL, &error);
348 printk(KERN_ERR "Failed to initialize device with \"%s\" : "
353 __setup("ubd", ubd_setup);
354 __uml_help(ubd_setup,
355 "ubd<n><flags>=<filename>[(:|,)<filename2>][(:|,)<serial>]\n"
356 " This is used to associate a device with a file in the underlying\n"
357 " filesystem. When specifying two filenames, the first one is the\n"
358 " COW name and the second is the backing file name. As separator you can\n"
359 " use either a ':' or a ',': the first one allows writing things like;\n"
360 " ubd0=~/Uml/root_cow:~/Uml/root_backing_file\n"
361 " while with a ',' the shell would not expand the 2nd '~'.\n"
362 " When using only one filename, UML will detect whether to treat it like\n"
363 " a COW file or a backing file. To override this detection, add the 'd'\n"
365 " ubd0d=BackingFile\n"
366 " Usually, there is a filesystem in the file, but \n"
367 " that's not required. Swap devices containing swap files can be\n"
368 " specified like this. Also, a file which doesn't contain a\n"
369 " filesystem can have its contents read in the virtual \n"
370 " machine by running 'dd' on the device. <n> must be in the range\n"
371 " 0 to 7. Appending an 'r' to the number will cause that device\n"
372 " to be mounted read-only. For example ubd1r=./ext_fs. Appending\n"
373 " an 's' will cause data to be written to disk on the host immediately.\n"
374 " 'c' will cause the device to be treated as being shared between multiple\n"
375 " UMLs and file locking will be turned off - this is appropriate for a\n"
376 " cluster filesystem and inappropriate at almost all other times.\n\n"
377 " 't' will disable trim/discard support on the device (enabled by default).\n\n"
378 " An optional device serial number can be exposed using the serial parameter\n"
379 " on the cmdline which is exposed as a sysfs entry. This is particularly\n"
380 " useful when a unique number should be given to the device. Note when\n"
381 " specifying a label, the filename2 must be also presented. It can be\n"
382 " an empty string, in which case the backing file is not used:\n"
383 " ubd0=File,,Serial\n"
386 static int udb_setup(char *str)
388 printk("udb%s specified on command line is almost certainly a ubd -> "
393 __setup("udb", udb_setup);
394 __uml_help(udb_setup,
396 " This option is here solely to catch ubd -> udb typos, which can be\n"
397 " to impossible to catch visually unless you specifically look for\n"
398 " them. The only result of any option starting with 'udb' is an error\n"
399 " in the boot output.\n\n"
402 /* Only changed by ubd_init, which is an initcall. */
403 static int thread_fd = -1;
405 /* Function to read several request pointers at a time
406 * handling fractional reads if (and as) needed
409 static int bulk_req_safe_read(
411 struct io_thread_req * (*request_buffer)[],
412 struct io_thread_req **remainder,
420 if (*remainder_size > 0) {
422 (char *) request_buffer,
423 (char *) remainder, *remainder_size
430 ((char *) request_buffer) + *remainder_size,
431 sizeof(struct io_thread_req *)*max_recs
436 if ((n % sizeof(struct io_thread_req *)) > 0) {
438 * Read somehow returned not a multiple of dword
439 * theoretically possible, but never observed in the
440 * wild, so read routine must be able to handle it
442 *remainder_size = n % sizeof(struct io_thread_req *);
443 WARN(*remainder_size > 0, "UBD IPC read returned a partial result");
446 ((char *) request_buffer) +
447 (n/sizeof(struct io_thread_req *))*sizeof(struct io_thread_req *),
450 n = n - *remainder_size;
458 /* Called without dev->lock held, and only in interrupt context. */
459 static void ubd_handler(void)
465 n = bulk_req_safe_read(
475 printk(KERN_ERR "spurious interrupt in ubd_handler, "
479 for (count = 0; count < n/sizeof(struct io_thread_req *); count++) {
480 struct io_thread_req *io_req = (*irq_req_buffer)[count];
482 if ((io_req->error == BLK_STS_NOTSUPP) && (req_op(io_req->req) == REQ_OP_DISCARD)) {
483 blk_queue_max_discard_sectors(io_req->req->q, 0);
484 blk_queue_max_write_zeroes_sectors(io_req->req->q, 0);
485 blk_queue_flag_clear(QUEUE_FLAG_DISCARD, io_req->req->q);
487 blk_mq_end_request(io_req->req, io_req->error);
493 static irqreturn_t ubd_intr(int irq, void *dev)
499 /* Only changed by ubd_init, which is an initcall. */
500 static int io_pid = -1;
502 static void kill_io_thread(void)
505 os_kill_process(io_pid, 1);
508 __uml_exitcall(kill_io_thread);
510 static inline int ubd_file_size(struct ubd *ubd_dev, __u64 *size_out)
520 unsigned long long size;
524 if (ubd_dev->file && ubd_dev->cow.file) {
525 file = ubd_dev->cow.file;
530 fd = os_open_file(ubd_dev->file, of_read(OPENFLAGS()), 0);
534 err = read_cow_header(file_reader, &fd, &version, &backing_file, \
535 &mtime, &size, §or_size, &align, &bitmap_offset);
539 file = ubd_dev->file;
544 return os_file_size(file, size_out);
547 static int read_cow_bitmap(int fd, void *buf, int offset, int len)
551 err = os_pread_file(fd, buf, len, offset);
558 static int backing_file_mismatch(char *file, __u64 size, time64_t mtime)
561 unsigned long long actual;
564 err = os_file_modtime(file, &modtime);
566 printk(KERN_ERR "Failed to get modification time of backing "
567 "file \"%s\", err = %d\n", file, -err);
571 err = os_file_size(file, &actual);
573 printk(KERN_ERR "Failed to get size of backing file \"%s\", "
574 "err = %d\n", file, -err);
578 if (actual != size) {
579 /*__u64 can be a long on AMD64 and with %lu GCC complains; so
581 printk(KERN_ERR "Size mismatch (%llu vs %llu) of COW header "
582 "vs backing file\n", (unsigned long long) size, actual);
585 if (modtime != mtime) {
586 printk(KERN_ERR "mtime mismatch (%lld vs %lld) of COW header vs "
587 "backing file\n", mtime, modtime);
593 static int path_requires_switch(char *from_cmdline, char *from_cow, char *cow)
595 struct uml_stat buf1, buf2;
598 if (from_cmdline == NULL)
600 if (!strcmp(from_cmdline, from_cow))
603 err = os_stat_file(from_cmdline, &buf1);
605 printk(KERN_ERR "Couldn't stat '%s', err = %d\n", from_cmdline,
609 err = os_stat_file(from_cow, &buf2);
611 printk(KERN_ERR "Couldn't stat '%s', err = %d\n", from_cow,
615 if ((buf1.ust_dev == buf2.ust_dev) && (buf1.ust_ino == buf2.ust_ino))
618 printk(KERN_ERR "Backing file mismatch - \"%s\" requested, "
619 "\"%s\" specified in COW header of \"%s\"\n",
620 from_cmdline, from_cow, cow);
624 static int open_ubd_file(char *file, struct openflags *openflags, int shared,
625 char **backing_file_out, int *bitmap_offset_out,
626 unsigned long *bitmap_len_out, int *data_offset_out,
630 unsigned long long size;
631 __u32 version, align;
633 int fd, err, sectorsize, asked_switch, mode = 0644;
635 fd = os_open_file(file, *openflags, mode);
637 if ((fd == -ENOENT) && (create_cow_out != NULL))
640 ((fd != -EROFS) && (fd != -EACCES)))
643 fd = os_open_file(file, *openflags, mode);
649 printk(KERN_INFO "Not locking \"%s\" on the host\n", file);
651 err = os_lock_file(fd, openflags->w);
653 printk(KERN_ERR "Failed to lock '%s', err = %d\n",
659 /* Successful return case! */
660 if (backing_file_out == NULL)
663 err = read_cow_header(file_reader, &fd, &version, &backing_file, &mtime,
664 &size, §orsize, &align, bitmap_offset_out);
665 if (err && (*backing_file_out != NULL)) {
666 printk(KERN_ERR "Failed to read COW header from COW file "
667 "\"%s\", errno = %d\n", file, -err);
673 asked_switch = path_requires_switch(*backing_file_out, backing_file,
676 /* Allow switching only if no mismatch. */
677 if (asked_switch && !backing_file_mismatch(*backing_file_out, size,
679 printk(KERN_ERR "Switching backing file to '%s'\n",
681 err = write_cow_header(file, fd, *backing_file_out,
682 sectorsize, align, &size);
684 printk(KERN_ERR "Switch failed, errno = %d\n", -err);
688 *backing_file_out = backing_file;
689 err = backing_file_mismatch(*backing_file_out, size, mtime);
694 cow_sizes(version, size, sectorsize, align, *bitmap_offset_out,
695 bitmap_len_out, data_offset_out);
703 static int create_cow_file(char *cow_file, char *backing_file,
704 struct openflags flags,
705 int sectorsize, int alignment, int *bitmap_offset_out,
706 unsigned long *bitmap_len_out, int *data_offset_out)
711 fd = open_ubd_file(cow_file, &flags, 0, NULL, NULL, NULL, NULL, NULL);
714 printk(KERN_ERR "Open of COW file '%s' failed, errno = %d\n",
719 err = init_cow_file(fd, cow_file, backing_file, sectorsize, alignment,
720 bitmap_offset_out, bitmap_len_out,
729 static void ubd_close_dev(struct ubd *ubd_dev)
731 os_close_file(ubd_dev->fd);
732 if(ubd_dev->cow.file == NULL)
735 os_close_file(ubd_dev->cow.fd);
736 vfree(ubd_dev->cow.bitmap);
737 ubd_dev->cow.bitmap = NULL;
740 static int ubd_open_dev(struct ubd *ubd_dev)
742 struct openflags flags;
744 int err, create_cow, *create_ptr;
747 ubd_dev->openflags = ubd_dev->boot_openflags;
749 create_ptr = (ubd_dev->cow.file != NULL) ? &create_cow : NULL;
750 back_ptr = ubd_dev->no_cow ? NULL : &ubd_dev->cow.file;
752 fd = open_ubd_file(ubd_dev->file, &ubd_dev->openflags, ubd_dev->shared,
753 back_ptr, &ubd_dev->cow.bitmap_offset,
754 &ubd_dev->cow.bitmap_len, &ubd_dev->cow.data_offset,
757 if((fd == -ENOENT) && create_cow){
758 fd = create_cow_file(ubd_dev->file, ubd_dev->cow.file,
759 ubd_dev->openflags, SECTOR_SIZE, PAGE_SIZE,
760 &ubd_dev->cow.bitmap_offset,
761 &ubd_dev->cow.bitmap_len,
762 &ubd_dev->cow.data_offset);
764 printk(KERN_INFO "Creating \"%s\" as COW file for "
765 "\"%s\"\n", ubd_dev->file, ubd_dev->cow.file);
770 printk("Failed to open '%s', errno = %d\n", ubd_dev->file,
776 if(ubd_dev->cow.file != NULL){
777 blk_queue_max_hw_sectors(ubd_dev->queue, 8 * sizeof(long));
780 ubd_dev->cow.bitmap = vmalloc(ubd_dev->cow.bitmap_len);
781 if(ubd_dev->cow.bitmap == NULL){
782 printk(KERN_ERR "Failed to vmalloc COW bitmap\n");
785 flush_tlb_kernel_vm();
787 err = read_cow_bitmap(ubd_dev->fd, ubd_dev->cow.bitmap,
788 ubd_dev->cow.bitmap_offset,
789 ubd_dev->cow.bitmap_len);
793 flags = ubd_dev->openflags;
795 err = open_ubd_file(ubd_dev->cow.file, &flags, ubd_dev->shared, NULL,
796 NULL, NULL, NULL, NULL);
797 if(err < 0) goto error;
798 ubd_dev->cow.fd = err;
800 if (ubd_dev->no_trim == 0) {
801 ubd_dev->queue->limits.discard_granularity = SECTOR_SIZE;
802 ubd_dev->queue->limits.discard_alignment = SECTOR_SIZE;
803 blk_queue_max_discard_sectors(ubd_dev->queue, UBD_MAX_REQUEST);
804 blk_queue_max_write_zeroes_sectors(ubd_dev->queue, UBD_MAX_REQUEST);
805 blk_queue_flag_set(QUEUE_FLAG_DISCARD, ubd_dev->queue);
807 blk_queue_flag_set(QUEUE_FLAG_NONROT, ubd_dev->queue);
810 os_close_file(ubd_dev->fd);
814 static void ubd_device_release(struct device *dev)
816 struct ubd *ubd_dev = dev_get_drvdata(dev);
818 blk_mq_free_tag_set(&ubd_dev->tag_set);
819 *ubd_dev = ((struct ubd) DEFAULT_UBD);
822 static ssize_t serial_show(struct device *dev,
823 struct device_attribute *attr, char *buf)
825 struct gendisk *disk = dev_to_disk(dev);
826 struct ubd *ubd_dev = disk->private_data;
831 return sprintf(buf, "%s", ubd_dev->serial);
834 static DEVICE_ATTR_RO(serial);
836 static struct attribute *ubd_attrs[] = {
837 &dev_attr_serial.attr,
841 static umode_t ubd_attrs_are_visible(struct kobject *kobj,
842 struct attribute *a, int n)
847 static const struct attribute_group ubd_attr_group = {
849 .is_visible = ubd_attrs_are_visible,
852 static const struct attribute_group *ubd_attr_groups[] = {
857 static void ubd_disk_register(int major, u64 size, int unit,
858 struct gendisk *disk)
861 disk->first_minor = unit << UBD_SHIFT;
862 disk->minors = 1 << UBD_SHIFT;
863 disk->fops = &ubd_blops;
864 set_capacity(disk, size / 512);
865 sprintf(disk->disk_name, "ubd%c", 'a' + unit);
867 ubd_devs[unit].pdev.id = unit;
868 ubd_devs[unit].pdev.name = DRIVER_NAME;
869 ubd_devs[unit].pdev.dev.release = ubd_device_release;
870 dev_set_drvdata(&ubd_devs[unit].pdev.dev, &ubd_devs[unit]);
871 platform_device_register(&ubd_devs[unit].pdev);
873 disk->private_data = &ubd_devs[unit];
874 disk->queue = ubd_devs[unit].queue;
875 device_add_disk(&ubd_devs[unit].pdev.dev, disk, ubd_attr_groups);
878 #define ROUND_BLOCK(n) ((n + (SECTOR_SIZE - 1)) & (-SECTOR_SIZE))
880 static const struct blk_mq_ops ubd_mq_ops = {
881 .queue_rq = ubd_queue_rq,
884 static int ubd_add(int n, char **error_out)
886 struct ubd *ubd_dev = &ubd_devs[n];
887 struct gendisk *disk;
890 if(ubd_dev->file == NULL)
893 err = ubd_file_size(ubd_dev, &ubd_dev->size);
895 *error_out = "Couldn't determine size of device's file";
899 ubd_dev->size = ROUND_BLOCK(ubd_dev->size);
901 ubd_dev->tag_set.ops = &ubd_mq_ops;
902 ubd_dev->tag_set.queue_depth = 64;
903 ubd_dev->tag_set.numa_node = NUMA_NO_NODE;
904 ubd_dev->tag_set.flags = BLK_MQ_F_SHOULD_MERGE;
905 ubd_dev->tag_set.driver_data = ubd_dev;
906 ubd_dev->tag_set.nr_hw_queues = 1;
908 err = blk_mq_alloc_tag_set(&ubd_dev->tag_set);
912 disk = blk_mq_alloc_disk(&ubd_dev->tag_set, ubd_dev);
915 goto out_cleanup_tags;
917 ubd_dev->queue = disk->queue;
919 blk_queue_write_cache(ubd_dev->queue, true, false);
920 blk_queue_max_segments(ubd_dev->queue, MAX_SG);
921 blk_queue_segment_boundary(ubd_dev->queue, PAGE_SIZE - 1);
922 ubd_disk_register(UBD_MAJOR, ubd_dev->size, n, disk);
923 ubd_gendisk[n] = disk;
927 blk_mq_free_tag_set(&ubd_dev->tag_set);
932 static int ubd_config(char *str, char **error_out)
936 /* This string is possibly broken up and stored, so it's only
937 * freed if ubd_setup_common fails, or if only general options
940 str = kstrdup(str, GFP_KERNEL);
942 *error_out = "Failed to allocate memory";
946 ret = ubd_setup_common(str, &n, error_out);
955 mutex_lock(&ubd_lock);
956 ret = ubd_add(n, error_out);
958 ubd_devs[n].file = NULL;
959 mutex_unlock(&ubd_lock);
969 static int ubd_get_config(char *name, char *str, int size, char **error_out)
974 n = parse_unit(&name);
975 if((n >= MAX_DEV) || (n < 0)){
976 *error_out = "ubd_get_config : device number out of range";
980 ubd_dev = &ubd_devs[n];
981 mutex_lock(&ubd_lock);
983 if(ubd_dev->file == NULL){
984 CONFIG_CHUNK(str, size, len, "", 1);
988 CONFIG_CHUNK(str, size, len, ubd_dev->file, 0);
990 if(ubd_dev->cow.file != NULL){
991 CONFIG_CHUNK(str, size, len, ",", 0);
992 CONFIG_CHUNK(str, size, len, ubd_dev->cow.file, 1);
994 else CONFIG_CHUNK(str, size, len, "", 1);
997 mutex_unlock(&ubd_lock);
1001 static int ubd_id(char **str, int *start_out, int *end_out)
1005 n = parse_unit(str);
1007 *end_out = MAX_DEV - 1;
1011 static int ubd_remove(int n, char **error_out)
1013 struct gendisk *disk = ubd_gendisk[n];
1014 struct ubd *ubd_dev;
1017 mutex_lock(&ubd_lock);
1019 ubd_dev = &ubd_devs[n];
1021 if(ubd_dev->file == NULL)
1024 /* you cannot remove a open disk */
1026 if(ubd_dev->count > 0)
1029 ubd_gendisk[n] = NULL;
1032 blk_cleanup_disk(disk);
1036 platform_device_unregister(&ubd_dev->pdev);
1038 mutex_unlock(&ubd_lock);
1042 /* All these are called by mconsole in process context and without
1043 * ubd-specific locks. The structure itself is const except for .list.
1045 static struct mc_device ubd_mc = {
1046 .list = LIST_HEAD_INIT(ubd_mc.list),
1048 .config = ubd_config,
1049 .get_config = ubd_get_config,
1051 .remove = ubd_remove,
1054 static int __init ubd_mc_init(void)
1056 mconsole_register_dev(&ubd_mc);
1060 __initcall(ubd_mc_init);
1062 static int __init ubd0_init(void)
1064 struct ubd *ubd_dev = &ubd_devs[0];
1066 mutex_lock(&ubd_lock);
1067 if(ubd_dev->file == NULL)
1068 ubd_dev->file = "root_fs";
1069 mutex_unlock(&ubd_lock);
1074 __initcall(ubd0_init);
1076 /* Used in ubd_init, which is an initcall */
1077 static struct platform_driver ubd_driver = {
1079 .name = DRIVER_NAME,
1083 static int __init ubd_init(void)
1088 if (register_blkdev(UBD_MAJOR, "ubd"))
1091 irq_req_buffer = kmalloc_array(UBD_REQ_BUFFER_SIZE,
1092 sizeof(struct io_thread_req *),
1097 if (irq_req_buffer == NULL) {
1098 printk(KERN_ERR "Failed to initialize ubd buffering\n");
1101 io_req_buffer = kmalloc_array(UBD_REQ_BUFFER_SIZE,
1102 sizeof(struct io_thread_req *),
1108 if (io_req_buffer == NULL) {
1109 printk(KERN_ERR "Failed to initialize ubd buffering\n");
1112 platform_driver_register(&ubd_driver);
1113 mutex_lock(&ubd_lock);
1114 for (i = 0; i < MAX_DEV; i++){
1115 err = ubd_add(i, &error);
1117 printk(KERN_ERR "Failed to initialize ubd device %d :"
1120 mutex_unlock(&ubd_lock);
1124 late_initcall(ubd_init);
1126 static int __init ubd_driver_init(void){
1127 unsigned long stack;
1130 /* Set by CONFIG_BLK_DEV_UBD_SYNC or ubd=sync.*/
1131 if(global_openflags.s){
1132 printk(KERN_INFO "ubd: Synchronous mode\n");
1133 /* Letting ubd=sync be like using ubd#s= instead of ubd#= is
1134 * enough. So use anyway the io thread. */
1136 stack = alloc_stack(0, 0);
1137 io_pid = start_io_thread(stack + PAGE_SIZE, &thread_fd);
1140 "ubd : Failed to start I/O thread (errno = %d) - "
1141 "falling back to synchronous I/O\n", -io_pid);
1145 err = um_request_irq(UBD_IRQ, thread_fd, IRQ_READ, ubd_intr,
1146 0, "ubd", ubd_devs);
1148 printk(KERN_ERR "um_request_irq failed - errno = %d\n", -err);
1152 device_initcall(ubd_driver_init);
1154 static int ubd_open(struct block_device *bdev, fmode_t mode)
1156 struct gendisk *disk = bdev->bd_disk;
1157 struct ubd *ubd_dev = disk->private_data;
1160 mutex_lock(&ubd_mutex);
1161 if(ubd_dev->count == 0){
1162 err = ubd_open_dev(ubd_dev);
1164 printk(KERN_ERR "%s: Can't open \"%s\": errno = %d\n",
1165 disk->disk_name, ubd_dev->file, -err);
1170 set_disk_ro(disk, !ubd_dev->openflags.w);
1172 /* This should no more be needed. And it didn't work anyway to exclude
1173 * read-write remounting of filesystems.*/
1174 /*if((mode & FMODE_WRITE) && !ubd_dev->openflags.w){
1175 if(--ubd_dev->count == 0) ubd_close_dev(ubd_dev);
1179 mutex_unlock(&ubd_mutex);
1183 static void ubd_release(struct gendisk *disk, fmode_t mode)
1185 struct ubd *ubd_dev = disk->private_data;
1187 mutex_lock(&ubd_mutex);
1188 if(--ubd_dev->count == 0)
1189 ubd_close_dev(ubd_dev);
1190 mutex_unlock(&ubd_mutex);
1193 static void cowify_bitmap(__u64 io_offset, int length, unsigned long *cow_mask,
1194 __u64 *cow_offset, unsigned long *bitmap,
1195 __u64 bitmap_offset, unsigned long *bitmap_words,
1198 __u64 sector = io_offset >> SECTOR_SHIFT;
1199 int i, update_bitmap = 0;
1201 for (i = 0; i < length >> SECTOR_SHIFT; i++) {
1202 if(cow_mask != NULL)
1203 ubd_set_bit(i, (unsigned char *) cow_mask);
1204 if(ubd_test_bit(sector + i, (unsigned char *) bitmap))
1208 ubd_set_bit(sector + i, (unsigned char *) bitmap);
1214 *cow_offset = sector / (sizeof(unsigned long) * 8);
1216 /* This takes care of the case where we're exactly at the end of the
1217 * device, and *cow_offset + 1 is off the end. So, just back it up
1218 * by one word. Thanks to Lynn Kerby for the fix and James McMechan
1219 * for the original diagnosis.
1221 if (*cow_offset == (DIV_ROUND_UP(bitmap_len,
1222 sizeof(unsigned long)) - 1))
1225 bitmap_words[0] = bitmap[*cow_offset];
1226 bitmap_words[1] = bitmap[*cow_offset + 1];
1228 *cow_offset *= sizeof(unsigned long);
1229 *cow_offset += bitmap_offset;
1232 static void cowify_req(struct io_thread_req *req, struct io_desc *segment,
1233 unsigned long offset, unsigned long *bitmap,
1234 __u64 bitmap_offset, __u64 bitmap_len)
1236 __u64 sector = offset >> SECTOR_SHIFT;
1239 if (segment->length > (sizeof(segment->sector_mask) * 8) << SECTOR_SHIFT)
1240 panic("Operation too long");
1242 if (req_op(req->req) == REQ_OP_READ) {
1243 for (i = 0; i < segment->length >> SECTOR_SHIFT; i++) {
1244 if(ubd_test_bit(sector + i, (unsigned char *) bitmap))
1245 ubd_set_bit(i, (unsigned char *)
1246 &segment->sector_mask);
1249 cowify_bitmap(offset, segment->length, &segment->sector_mask,
1250 &segment->cow_offset, bitmap, bitmap_offset,
1251 segment->bitmap_words, bitmap_len);
1255 static void ubd_map_req(struct ubd *dev, struct io_thread_req *io_req,
1256 struct request *req)
1258 struct bio_vec bvec;
1259 struct req_iterator iter;
1261 unsigned long byte_offset = io_req->offset;
1262 int op = req_op(req);
1264 if (op == REQ_OP_WRITE_ZEROES || op == REQ_OP_DISCARD) {
1265 io_req->io_desc[0].buffer = NULL;
1266 io_req->io_desc[0].length = blk_rq_bytes(req);
1268 rq_for_each_segment(bvec, req, iter) {
1269 BUG_ON(i >= io_req->desc_cnt);
1271 io_req->io_desc[i].buffer = bvec_virt(&bvec);
1272 io_req->io_desc[i].length = bvec.bv_len;
1277 if (dev->cow.file) {
1278 for (i = 0; i < io_req->desc_cnt; i++) {
1279 cowify_req(io_req, &io_req->io_desc[i], byte_offset,
1280 dev->cow.bitmap, dev->cow.bitmap_offset,
1281 dev->cow.bitmap_len);
1282 byte_offset += io_req->io_desc[i].length;
1288 static struct io_thread_req *ubd_alloc_req(struct ubd *dev, struct request *req,
1291 struct io_thread_req *io_req;
1294 io_req = kmalloc(sizeof(*io_req) +
1295 (desc_cnt * sizeof(struct io_desc)),
1302 io_req->fds[0] = dev->cow.fd;
1304 io_req->fds[0] = dev->fd;
1306 io_req->sectorsize = SECTOR_SIZE;
1307 io_req->fds[1] = dev->fd;
1308 io_req->offset = (u64) blk_rq_pos(req) << SECTOR_SHIFT;
1309 io_req->offsets[0] = 0;
1310 io_req->offsets[1] = dev->cow.data_offset;
1312 for (i = 0 ; i < desc_cnt; i++) {
1313 io_req->io_desc[i].sector_mask = 0;
1314 io_req->io_desc[i].cow_offset = -1;
1320 static int ubd_submit_request(struct ubd *dev, struct request *req)
1323 struct io_thread_req *io_req;
1325 int op = req_op(req);
1327 if (op == REQ_OP_FLUSH)
1329 else if (op == REQ_OP_WRITE_ZEROES || op == REQ_OP_DISCARD)
1332 segs = blk_rq_nr_phys_segments(req);
1334 io_req = ubd_alloc_req(dev, req, segs);
1338 io_req->desc_cnt = segs;
1340 ubd_map_req(dev, io_req, req);
1342 ret = os_write_file(thread_fd, &io_req, sizeof(io_req));
1343 if (ret != sizeof(io_req)) {
1345 pr_err("write to io thread failed: %d\n", -ret);
1351 static blk_status_t ubd_queue_rq(struct blk_mq_hw_ctx *hctx,
1352 const struct blk_mq_queue_data *bd)
1354 struct ubd *ubd_dev = hctx->queue->queuedata;
1355 struct request *req = bd->rq;
1356 int ret = 0, res = BLK_STS_OK;
1358 blk_mq_start_request(req);
1360 spin_lock_irq(&ubd_dev->lock);
1362 switch (req_op(req)) {
1366 case REQ_OP_DISCARD:
1367 case REQ_OP_WRITE_ZEROES:
1368 ret = ubd_submit_request(ubd_dev, req);
1372 res = BLK_STS_NOTSUPP;
1375 spin_unlock_irq(&ubd_dev->lock);
1379 res = BLK_STS_RESOURCE;
1381 res = BLK_STS_DEV_RESOURCE;
1387 static int ubd_getgeo(struct block_device *bdev, struct hd_geometry *geo)
1389 struct ubd *ubd_dev = bdev->bd_disk->private_data;
1393 geo->cylinders = ubd_dev->size / (128 * 32 * 512);
1397 static int ubd_ioctl(struct block_device *bdev, fmode_t mode,
1398 unsigned int cmd, unsigned long arg)
1400 struct ubd *ubd_dev = bdev->bd_disk->private_data;
1401 u16 ubd_id[ATA_ID_WORDS];
1404 struct cdrom_volctrl volume;
1405 case HDIO_GET_IDENTITY:
1406 memset(&ubd_id, 0, ATA_ID_WORDS * 2);
1407 ubd_id[ATA_ID_CYLS] = ubd_dev->size / (128 * 32 * 512);
1408 ubd_id[ATA_ID_HEADS] = 128;
1409 ubd_id[ATA_ID_SECTORS] = 32;
1410 if(copy_to_user((char __user *) arg, (char *) &ubd_id,
1416 if(copy_from_user(&volume, (char __user *) arg, sizeof(volume)))
1418 volume.channel0 = 255;
1419 volume.channel1 = 255;
1420 volume.channel2 = 255;
1421 volume.channel3 = 255;
1422 if(copy_to_user((char __user *) arg, &volume, sizeof(volume)))
1429 static int map_error(int error_code)
1431 switch (error_code) {
1436 return BLK_STS_NOTSUPP;
1438 return BLK_STS_NOSPC;
1440 return BLK_STS_IOERR;
1444 * Everything from here onwards *IS NOT PART OF THE KERNEL*
1446 * The following functions are part of UML hypervisor code.
1447 * All functions from here onwards are executed as a helper
1448 * thread and are not allowed to execute any kernel functions.
1450 * Any communication must occur strictly via shared memory and IPC.
1452 * Do not add printks, locks, kernel memory operations, etc - it
1453 * will result in unpredictable behaviour and/or crashes.
1456 static int update_bitmap(struct io_thread_req *req, struct io_desc *segment)
1460 if (segment->cow_offset == -1)
1461 return map_error(0);
1463 n = os_pwrite_file(req->fds[1], &segment->bitmap_words,
1464 sizeof(segment->bitmap_words), segment->cow_offset);
1465 if (n != sizeof(segment->bitmap_words))
1466 return map_error(-n);
1468 return map_error(0);
1471 static void do_io(struct io_thread_req *req, struct io_desc *desc)
1475 int n, nsectors, start, end, bit;
1478 /* FLUSH is really a special case, we cannot "case" it with others */
1480 if (req_op(req->req) == REQ_OP_FLUSH) {
1481 /* fds[0] is always either the rw image or our cow file */
1482 req->error = map_error(-os_sync_file(req->fds[0]));
1486 nsectors = desc->length / req->sectorsize;
1489 bit = ubd_test_bit(start, (unsigned char *) &desc->sector_mask);
1491 while((end < nsectors) &&
1492 (ubd_test_bit(end, (unsigned char *) &desc->sector_mask) == bit))
1495 off = req->offset + req->offsets[bit] +
1496 start * req->sectorsize;
1497 len = (end - start) * req->sectorsize;
1498 if (desc->buffer != NULL)
1499 buf = &desc->buffer[start * req->sectorsize];
1501 switch (req_op(req->req)) {
1507 n = os_pread_file(req->fds[bit], buf, len, off);
1509 req->error = map_error(-n);
1512 } while((n < len) && (n != 0));
1513 if (n < len) memset(&buf[n], 0, len - n);
1516 n = os_pwrite_file(req->fds[bit], buf, len, off);
1518 req->error = map_error(-n);
1522 case REQ_OP_DISCARD:
1523 case REQ_OP_WRITE_ZEROES:
1524 n = os_falloc_punch(req->fds[bit], off, len);
1526 req->error = map_error(-n);
1532 req->error = BLK_STS_NOTSUPP;
1537 } while(start < nsectors);
1540 req->error = update_bitmap(req, desc);
1543 /* Changed in start_io_thread, which is serialized by being called only
1544 * from ubd_init, which is an initcall.
1548 /* Only changed by the io thread. XXX: currently unused. */
1549 static int io_count = 0;
1551 int io_thread(void *arg)
1553 int n, count, written, res;
1555 os_fix_helper_signals();
1558 n = bulk_req_safe_read(
1572 for (count = 0; count < n/sizeof(struct io_thread_req *); count++) {
1573 struct io_thread_req *req = (*io_req_buffer)[count];
1577 for (i = 0; !req->error && i < req->desc_cnt; i++)
1578 do_io(req, &(req->io_desc[i]));
1585 res = os_write_file(kernel_fd,
1586 ((char *) io_req_buffer) + written,
1594 } while (written < n);