1 // SPDX-License-Identifier: GPL-2.0
3 * Copyright (C) 2018 Cambridge Greys Ltd
8 /* 2001-09-28...2002-04-17
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 <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);
109 static int ubd_ioctl(struct block_device *bdev, blk_mode_t mode,
110 unsigned int cmd, unsigned long arg);
111 static int ubd_getgeo(struct block_device *bdev, struct hd_geometry *geo);
115 static const struct block_device_operations ubd_blops = {
116 .owner = THIS_MODULE,
118 .compat_ioctl = blkdev_compat_ptr_ioctl,
119 .getgeo = ubd_getgeo,
122 #ifdef CONFIG_BLK_DEV_UBD_SYNC
123 #define OPEN_FLAGS ((struct openflags) { .r = 1, .w = 1, .s = 1, .c = 0, \
126 #define OPEN_FLAGS ((struct openflags) { .r = 1, .w = 1, .s = 0, .c = 0, \
129 static struct openflags global_openflags = OPEN_FLAGS;
132 /* backing file name */
134 /* backing file fd */
136 unsigned long *bitmap;
137 unsigned long bitmap_len;
145 /* name (and fd, below) of the file opened for writing, either the
146 * backing or the cow file. */
151 struct openflags boot_openflags;
152 struct openflags openflags;
157 struct platform_device pdev;
158 struct gendisk *disk;
159 struct blk_mq_tag_set tag_set;
163 #define DEFAULT_COW { \
167 .bitmap_offset = 0, \
171 #define DEFAULT_UBD { \
176 .boot_openflags = OPEN_FLAGS, \
177 .openflags = OPEN_FLAGS, \
181 .cow = DEFAULT_COW, \
182 .lock = __SPIN_LOCK_UNLOCKED(ubd_devs.lock), \
185 /* Protected by ubd_lock */
186 static struct ubd ubd_devs[MAX_DEV] = { [0 ... MAX_DEV - 1] = DEFAULT_UBD };
188 static blk_status_t ubd_queue_rq(struct blk_mq_hw_ctx *hctx,
189 const struct blk_mq_queue_data *bd);
191 static int fake_ide_setup(char *str)
193 pr_warn("The fake_ide option has been removed\n");
196 __setup("fake_ide", fake_ide_setup);
198 __uml_help(fake_ide_setup,
200 " Obsolete stub.\n\n"
203 static int parse_unit(char **ptr)
205 char *str = *ptr, *end;
209 n = simple_strtoul(str, &end, 0);
214 else if (('a' <= *str) && (*str <= 'z')) {
222 /* If *index_out == -1 at exit, the passed option was a general one;
223 * otherwise, the str pointer is used (and owned) inside ubd_devs array, so it
224 * should not be freed on exit.
226 static int ubd_setup_common(char *str, int *index_out, char **error_out)
229 struct openflags flags = global_openflags;
230 char *file, *backing_file, *serial;
233 if(index_out) *index_out = -1;
237 if(!strcmp(str, "sync")){
238 global_openflags = of_sync(global_openflags);
242 pr_warn("fake major not supported any more\n");
246 n = parse_unit(&str);
248 *error_out = "Couldn't parse device number";
252 *error_out = "Device number out of range";
257 mutex_lock(&ubd_lock);
259 ubd_dev = &ubd_devs[n];
260 if(ubd_dev->file != NULL){
261 *error_out = "Device is already configured";
269 for (i = 0; i < sizeof("rscdt="); i++) {
284 ubd_dev->no_trim = 1;
290 *error_out = "Expected '=' or flag letter "
298 *error_out = "Too many flags specified";
300 *error_out = "Missing '='";
304 file = strsep(&str, ",:");
308 backing_file = strsep(&str, ",:");
309 if (backing_file && *backing_file == '\0')
312 serial = strsep(&str, ",:");
313 if (serial && *serial == '\0')
316 if (backing_file && ubd_dev->no_cow) {
317 *error_out = "Can't specify both 'd' and a cow file";
322 ubd_dev->file = file;
323 ubd_dev->cow.file = backing_file;
324 ubd_dev->serial = serial;
325 ubd_dev->boot_openflags = flags;
327 mutex_unlock(&ubd_lock);
331 static int ubd_setup(char *str)
336 err = ubd_setup_common(str, NULL, &error);
338 printk(KERN_ERR "Failed to initialize device with \"%s\" : "
343 __setup("ubd", ubd_setup);
344 __uml_help(ubd_setup,
345 "ubd<n><flags>=<filename>[(:|,)<filename2>][(:|,)<serial>]\n"
346 " This is used to associate a device with a file in the underlying\n"
347 " filesystem. When specifying two filenames, the first one is the\n"
348 " COW name and the second is the backing file name. As separator you can\n"
349 " use either a ':' or a ',': the first one allows writing things like;\n"
350 " ubd0=~/Uml/root_cow:~/Uml/root_backing_file\n"
351 " while with a ',' the shell would not expand the 2nd '~'.\n"
352 " When using only one filename, UML will detect whether to treat it like\n"
353 " a COW file or a backing file. To override this detection, add the 'd'\n"
355 " ubd0d=BackingFile\n"
356 " Usually, there is a filesystem in the file, but \n"
357 " that's not required. Swap devices containing swap files can be\n"
358 " specified like this. Also, a file which doesn't contain a\n"
359 " filesystem can have its contents read in the virtual \n"
360 " machine by running 'dd' on the device. <n> must be in the range\n"
361 " 0 to 7. Appending an 'r' to the number will cause that device\n"
362 " to be mounted read-only. For example ubd1r=./ext_fs. Appending\n"
363 " an 's' will cause data to be written to disk on the host immediately.\n"
364 " 'c' will cause the device to be treated as being shared between multiple\n"
365 " UMLs and file locking will be turned off - this is appropriate for a\n"
366 " cluster filesystem and inappropriate at almost all other times.\n\n"
367 " 't' will disable trim/discard support on the device (enabled by default).\n\n"
368 " An optional device serial number can be exposed using the serial parameter\n"
369 " on the cmdline which is exposed as a sysfs entry. This is particularly\n"
370 " useful when a unique number should be given to the device. Note when\n"
371 " specifying a label, the filename2 must be also presented. It can be\n"
372 " an empty string, in which case the backing file is not used:\n"
373 " ubd0=File,,Serial\n"
376 static int udb_setup(char *str)
378 printk("udb%s specified on command line is almost certainly a ubd -> "
383 __setup("udb", udb_setup);
384 __uml_help(udb_setup,
386 " This option is here solely to catch ubd -> udb typos, which can be\n"
387 " to impossible to catch visually unless you specifically look for\n"
388 " them. The only result of any option starting with 'udb' is an error\n"
389 " in the boot output.\n\n"
392 /* Only changed by ubd_init, which is an initcall. */
393 static int thread_fd = -1;
395 /* Function to read several request pointers at a time
396 * handling fractional reads if (and as) needed
399 static int bulk_req_safe_read(
401 struct io_thread_req * (*request_buffer)[],
402 struct io_thread_req **remainder,
410 if (*remainder_size > 0) {
412 (char *) request_buffer,
413 (char *) remainder, *remainder_size
420 ((char *) request_buffer) + *remainder_size,
421 sizeof(struct io_thread_req *)*max_recs
426 if ((n % sizeof(struct io_thread_req *)) > 0) {
428 * Read somehow returned not a multiple of dword
429 * theoretically possible, but never observed in the
430 * wild, so read routine must be able to handle it
432 *remainder_size = n % sizeof(struct io_thread_req *);
433 WARN(*remainder_size > 0, "UBD IPC read returned a partial result");
436 ((char *) request_buffer) +
437 (n/sizeof(struct io_thread_req *))*sizeof(struct io_thread_req *),
440 n = n - *remainder_size;
448 static void ubd_end_request(struct io_thread_req *io_req)
450 if (io_req->error == BLK_STS_NOTSUPP) {
451 if (req_op(io_req->req) == REQ_OP_DISCARD)
452 blk_queue_disable_discard(io_req->req->q);
453 else if (req_op(io_req->req) == REQ_OP_WRITE_ZEROES)
454 blk_queue_disable_write_zeroes(io_req->req->q);
456 blk_mq_end_request(io_req->req, io_req->error);
460 static irqreturn_t ubd_intr(int irq, void *dev)
464 while ((len = bulk_req_safe_read(thread_fd, irq_req_buffer,
465 &irq_remainder, &irq_remainder_size,
466 UBD_REQ_BUFFER_SIZE)) >= 0) {
467 for (i = 0; i < len / sizeof(struct io_thread_req *); i++)
468 ubd_end_request((*irq_req_buffer)[i]);
471 if (len < 0 && len != -EAGAIN)
472 pr_err("spurious interrupt in %s, err = %d\n", __func__, len);
476 /* Only changed by ubd_init, which is an initcall. */
477 static int io_pid = -1;
479 static void kill_io_thread(void)
482 os_kill_process(io_pid, 1);
485 __uml_exitcall(kill_io_thread);
487 static inline int ubd_file_size(struct ubd *ubd_dev, __u64 *size_out)
497 unsigned long long size;
501 if (ubd_dev->file && ubd_dev->cow.file) {
502 file = ubd_dev->cow.file;
507 fd = os_open_file(ubd_dev->file, of_read(OPENFLAGS()), 0);
511 err = read_cow_header(file_reader, &fd, &version, &backing_file, \
512 &mtime, &size, §or_size, &align, &bitmap_offset);
516 file = ubd_dev->file;
521 return os_file_size(file, size_out);
524 static int read_cow_bitmap(int fd, void *buf, int offset, int len)
528 err = os_pread_file(fd, buf, len, offset);
535 static int backing_file_mismatch(char *file, __u64 size, time64_t mtime)
538 unsigned long long actual;
541 err = os_file_modtime(file, &modtime);
543 printk(KERN_ERR "Failed to get modification time of backing "
544 "file \"%s\", err = %d\n", file, -err);
548 err = os_file_size(file, &actual);
550 printk(KERN_ERR "Failed to get size of backing file \"%s\", "
551 "err = %d\n", file, -err);
555 if (actual != size) {
556 /*__u64 can be a long on AMD64 and with %lu GCC complains; so
558 printk(KERN_ERR "Size mismatch (%llu vs %llu) of COW header "
559 "vs backing file\n", (unsigned long long) size, actual);
562 if (modtime != mtime) {
563 printk(KERN_ERR "mtime mismatch (%lld vs %lld) of COW header vs "
564 "backing file\n", mtime, modtime);
570 static int path_requires_switch(char *from_cmdline, char *from_cow, char *cow)
572 struct uml_stat buf1, buf2;
575 if (from_cmdline == NULL)
577 if (!strcmp(from_cmdline, from_cow))
580 err = os_stat_file(from_cmdline, &buf1);
582 printk(KERN_ERR "Couldn't stat '%s', err = %d\n", from_cmdline,
586 err = os_stat_file(from_cow, &buf2);
588 printk(KERN_ERR "Couldn't stat '%s', err = %d\n", from_cow,
592 if ((buf1.ust_dev == buf2.ust_dev) && (buf1.ust_ino == buf2.ust_ino))
595 printk(KERN_ERR "Backing file mismatch - \"%s\" requested, "
596 "\"%s\" specified in COW header of \"%s\"\n",
597 from_cmdline, from_cow, cow);
601 static int open_ubd_file(char *file, struct openflags *openflags, int shared,
602 char **backing_file_out, int *bitmap_offset_out,
603 unsigned long *bitmap_len_out, int *data_offset_out,
607 unsigned long long size;
608 __u32 version, align;
610 int fd, err, sectorsize, asked_switch, mode = 0644;
612 fd = os_open_file(file, *openflags, mode);
614 if ((fd == -ENOENT) && (create_cow_out != NULL))
617 ((fd != -EROFS) && (fd != -EACCES)))
620 fd = os_open_file(file, *openflags, mode);
626 printk(KERN_INFO "Not locking \"%s\" on the host\n", file);
628 err = os_lock_file(fd, openflags->w);
630 printk(KERN_ERR "Failed to lock '%s', err = %d\n",
636 /* Successful return case! */
637 if (backing_file_out == NULL)
640 err = read_cow_header(file_reader, &fd, &version, &backing_file, &mtime,
641 &size, §orsize, &align, bitmap_offset_out);
642 if (err && (*backing_file_out != NULL)) {
643 printk(KERN_ERR "Failed to read COW header from COW file "
644 "\"%s\", errno = %d\n", file, -err);
650 asked_switch = path_requires_switch(*backing_file_out, backing_file,
653 /* Allow switching only if no mismatch. */
654 if (asked_switch && !backing_file_mismatch(*backing_file_out, size,
656 printk(KERN_ERR "Switching backing file to '%s'\n",
658 err = write_cow_header(file, fd, *backing_file_out,
659 sectorsize, align, &size);
661 printk(KERN_ERR "Switch failed, errno = %d\n", -err);
665 *backing_file_out = backing_file;
666 err = backing_file_mismatch(*backing_file_out, size, mtime);
671 cow_sizes(version, size, sectorsize, align, *bitmap_offset_out,
672 bitmap_len_out, data_offset_out);
680 static int create_cow_file(char *cow_file, char *backing_file,
681 struct openflags flags,
682 int sectorsize, int alignment, int *bitmap_offset_out,
683 unsigned long *bitmap_len_out, int *data_offset_out)
688 fd = open_ubd_file(cow_file, &flags, 0, NULL, NULL, NULL, NULL, NULL);
691 printk(KERN_ERR "Open of COW file '%s' failed, errno = %d\n",
696 err = init_cow_file(fd, cow_file, backing_file, sectorsize, alignment,
697 bitmap_offset_out, bitmap_len_out,
706 static void ubd_close_dev(struct ubd *ubd_dev)
708 os_close_file(ubd_dev->fd);
709 if(ubd_dev->cow.file == NULL)
712 os_close_file(ubd_dev->cow.fd);
713 vfree(ubd_dev->cow.bitmap);
714 ubd_dev->cow.bitmap = NULL;
717 static int ubd_open_dev(struct ubd *ubd_dev)
719 struct openflags flags;
721 int err, create_cow, *create_ptr;
724 ubd_dev->openflags = ubd_dev->boot_openflags;
726 create_ptr = (ubd_dev->cow.file != NULL) ? &create_cow : NULL;
727 back_ptr = ubd_dev->no_cow ? NULL : &ubd_dev->cow.file;
729 fd = open_ubd_file(ubd_dev->file, &ubd_dev->openflags, ubd_dev->shared,
730 back_ptr, &ubd_dev->cow.bitmap_offset,
731 &ubd_dev->cow.bitmap_len, &ubd_dev->cow.data_offset,
734 if((fd == -ENOENT) && create_cow){
735 fd = create_cow_file(ubd_dev->file, ubd_dev->cow.file,
736 ubd_dev->openflags, SECTOR_SIZE, PAGE_SIZE,
737 &ubd_dev->cow.bitmap_offset,
738 &ubd_dev->cow.bitmap_len,
739 &ubd_dev->cow.data_offset);
741 printk(KERN_INFO "Creating \"%s\" as COW file for "
742 "\"%s\"\n", ubd_dev->file, ubd_dev->cow.file);
747 printk("Failed to open '%s', errno = %d\n", ubd_dev->file,
753 if(ubd_dev->cow.file != NULL){
755 ubd_dev->cow.bitmap = vmalloc(ubd_dev->cow.bitmap_len);
756 if(ubd_dev->cow.bitmap == NULL){
757 printk(KERN_ERR "Failed to vmalloc COW bitmap\n");
761 err = read_cow_bitmap(ubd_dev->fd, ubd_dev->cow.bitmap,
762 ubd_dev->cow.bitmap_offset,
763 ubd_dev->cow.bitmap_len);
767 flags = ubd_dev->openflags;
769 err = open_ubd_file(ubd_dev->cow.file, &flags, ubd_dev->shared, NULL,
770 NULL, NULL, NULL, NULL);
771 if(err < 0) goto error;
772 ubd_dev->cow.fd = err;
776 os_close_file(ubd_dev->fd);
780 static void ubd_device_release(struct device *dev)
782 struct ubd *ubd_dev = container_of(dev, struct ubd, pdev.dev);
784 blk_mq_free_tag_set(&ubd_dev->tag_set);
785 *ubd_dev = ((struct ubd) DEFAULT_UBD);
788 static ssize_t serial_show(struct device *dev,
789 struct device_attribute *attr, char *buf)
791 struct gendisk *disk = dev_to_disk(dev);
792 struct ubd *ubd_dev = disk->private_data;
797 return sprintf(buf, "%s", ubd_dev->serial);
800 static DEVICE_ATTR_RO(serial);
802 static struct attribute *ubd_attrs[] = {
803 &dev_attr_serial.attr,
807 static umode_t ubd_attrs_are_visible(struct kobject *kobj,
808 struct attribute *a, int n)
813 static const struct attribute_group ubd_attr_group = {
815 .is_visible = ubd_attrs_are_visible,
818 static const struct attribute_group *ubd_attr_groups[] = {
823 #define ROUND_BLOCK(n) ((n + (SECTOR_SIZE - 1)) & (-SECTOR_SIZE))
825 static const struct blk_mq_ops ubd_mq_ops = {
826 .queue_rq = ubd_queue_rq,
829 static int ubd_add(int n, char **error_out)
831 struct ubd *ubd_dev = &ubd_devs[n];
832 struct queue_limits lim = {
833 .max_segments = MAX_SG,
834 .seg_boundary_mask = PAGE_SIZE - 1,
835 .features = BLK_FEAT_WRITE_CACHE,
837 struct gendisk *disk;
840 if(ubd_dev->file == NULL)
843 if (ubd_dev->cow.file)
844 lim.max_hw_sectors = 8 * sizeof(long);
845 if (!ubd_dev->no_trim) {
846 lim.max_hw_discard_sectors = UBD_MAX_REQUEST;
847 lim.max_write_zeroes_sectors = UBD_MAX_REQUEST;
850 err = ubd_file_size(ubd_dev, &ubd_dev->size);
852 *error_out = "Couldn't determine size of device's file";
856 err = ubd_open_dev(ubd_dev);
858 pr_err("ubd%c: Can't open \"%s\": errno = %d\n",
859 'a' + n, ubd_dev->file, -err);
863 ubd_dev->size = ROUND_BLOCK(ubd_dev->size);
865 ubd_dev->tag_set.ops = &ubd_mq_ops;
866 ubd_dev->tag_set.queue_depth = 64;
867 ubd_dev->tag_set.numa_node = NUMA_NO_NODE;
868 ubd_dev->tag_set.driver_data = ubd_dev;
869 ubd_dev->tag_set.nr_hw_queues = 1;
871 err = blk_mq_alloc_tag_set(&ubd_dev->tag_set);
875 disk = blk_mq_alloc_disk(&ubd_dev->tag_set, &lim, ubd_dev);
878 goto out_cleanup_tags;
881 disk->major = UBD_MAJOR;
882 disk->first_minor = n << UBD_SHIFT;
883 disk->minors = 1 << UBD_SHIFT;
884 disk->fops = &ubd_blops;
885 set_capacity(disk, ubd_dev->size / 512);
886 sprintf(disk->disk_name, "ubd%c", 'a' + n);
887 disk->private_data = ubd_dev;
888 set_disk_ro(disk, !ubd_dev->openflags.w);
890 ubd_dev->pdev.id = n;
891 ubd_dev->pdev.name = DRIVER_NAME;
892 ubd_dev->pdev.dev.release = ubd_device_release;
893 dev_set_drvdata(&ubd_dev->pdev.dev, ubd_dev);
894 platform_device_register(&ubd_dev->pdev);
896 err = device_add_disk(&ubd_dev->pdev.dev, disk, ubd_attr_groups);
898 goto out_cleanup_disk;
900 ubd_dev->disk = disk;
907 blk_mq_free_tag_set(&ubd_dev->tag_set);
909 ubd_close_dev(ubd_dev);
914 static int ubd_config(char *str, char **error_out)
918 /* This string is possibly broken up and stored, so it's only
919 * freed if ubd_setup_common fails, or if only general options
922 str = kstrdup(str, GFP_KERNEL);
924 *error_out = "Failed to allocate memory";
928 ret = ubd_setup_common(str, &n, error_out);
937 mutex_lock(&ubd_lock);
938 ret = ubd_add(n, error_out);
940 ubd_devs[n].file = NULL;
941 mutex_unlock(&ubd_lock);
951 static int ubd_get_config(char *name, char *str, int size, char **error_out)
956 n = parse_unit(&name);
957 if((n >= MAX_DEV) || (n < 0)){
958 *error_out = "ubd_get_config : device number out of range";
962 ubd_dev = &ubd_devs[n];
963 mutex_lock(&ubd_lock);
965 if(ubd_dev->file == NULL){
966 CONFIG_CHUNK(str, size, len, "", 1);
970 CONFIG_CHUNK(str, size, len, ubd_dev->file, 0);
972 if(ubd_dev->cow.file != NULL){
973 CONFIG_CHUNK(str, size, len, ",", 0);
974 CONFIG_CHUNK(str, size, len, ubd_dev->cow.file, 1);
976 else CONFIG_CHUNK(str, size, len, "", 1);
979 mutex_unlock(&ubd_lock);
983 static int ubd_id(char **str, int *start_out, int *end_out)
989 *end_out = MAX_DEV - 1;
993 static int ubd_remove(int n, char **error_out)
998 mutex_lock(&ubd_lock);
1000 ubd_dev = &ubd_devs[n];
1002 if(ubd_dev->file == NULL)
1005 if (ubd_dev->disk) {
1006 /* you cannot remove a open disk */
1008 if (disk_openers(ubd_dev->disk))
1011 del_gendisk(ubd_dev->disk);
1012 ubd_close_dev(ubd_dev);
1013 put_disk(ubd_dev->disk);
1017 platform_device_unregister(&ubd_dev->pdev);
1019 mutex_unlock(&ubd_lock);
1023 /* All these are called by mconsole in process context and without
1024 * ubd-specific locks. The structure itself is const except for .list.
1026 static struct mc_device ubd_mc = {
1027 .list = LIST_HEAD_INIT(ubd_mc.list),
1029 .config = ubd_config,
1030 .get_config = ubd_get_config,
1032 .remove = ubd_remove,
1035 static int __init ubd_mc_init(void)
1037 mconsole_register_dev(&ubd_mc);
1041 __initcall(ubd_mc_init);
1043 static int __init ubd0_init(void)
1045 struct ubd *ubd_dev = &ubd_devs[0];
1047 mutex_lock(&ubd_lock);
1048 if(ubd_dev->file == NULL)
1049 ubd_dev->file = "root_fs";
1050 mutex_unlock(&ubd_lock);
1055 __initcall(ubd0_init);
1057 /* Used in ubd_init, which is an initcall */
1058 static struct platform_driver ubd_driver = {
1060 .name = DRIVER_NAME,
1064 static int __init ubd_init(void)
1069 if (register_blkdev(UBD_MAJOR, "ubd"))
1072 irq_req_buffer = kmalloc_array(UBD_REQ_BUFFER_SIZE,
1073 sizeof(struct io_thread_req *),
1078 if (irq_req_buffer == NULL) {
1079 printk(KERN_ERR "Failed to initialize ubd buffering\n");
1082 io_req_buffer = kmalloc_array(UBD_REQ_BUFFER_SIZE,
1083 sizeof(struct io_thread_req *),
1089 if (io_req_buffer == NULL) {
1090 printk(KERN_ERR "Failed to initialize ubd buffering\n");
1093 platform_driver_register(&ubd_driver);
1094 mutex_lock(&ubd_lock);
1095 for (i = 0; i < MAX_DEV; i++){
1096 err = ubd_add(i, &error);
1098 printk(KERN_ERR "Failed to initialize ubd device %d :"
1101 mutex_unlock(&ubd_lock);
1105 late_initcall(ubd_init);
1107 static int __init ubd_driver_init(void){
1108 unsigned long stack;
1111 /* Set by CONFIG_BLK_DEV_UBD_SYNC or ubd=sync.*/
1112 if(global_openflags.s){
1113 printk(KERN_INFO "ubd: Synchronous mode\n");
1114 /* Letting ubd=sync be like using ubd#s= instead of ubd#= is
1115 * enough. So use anyway the io thread. */
1117 stack = alloc_stack(0, 0);
1118 io_pid = start_io_thread(stack + PAGE_SIZE, &thread_fd);
1121 "ubd : Failed to start I/O thread (errno = %d) - "
1122 "falling back to synchronous I/O\n", -io_pid);
1126 err = um_request_irq(UBD_IRQ, thread_fd, IRQ_READ, ubd_intr,
1127 0, "ubd", ubd_devs);
1129 printk(KERN_ERR "um_request_irq failed - errno = %d\n", -err);
1133 device_initcall(ubd_driver_init);
1135 static void cowify_bitmap(__u64 io_offset, int length, unsigned long *cow_mask,
1136 __u64 *cow_offset, unsigned long *bitmap,
1137 __u64 bitmap_offset, unsigned long *bitmap_words,
1140 __u64 sector = io_offset >> SECTOR_SHIFT;
1141 int i, update_bitmap = 0;
1143 for (i = 0; i < length >> SECTOR_SHIFT; i++) {
1144 if(cow_mask != NULL)
1145 ubd_set_bit(i, (unsigned char *) cow_mask);
1146 if(ubd_test_bit(sector + i, (unsigned char *) bitmap))
1150 ubd_set_bit(sector + i, (unsigned char *) bitmap);
1156 *cow_offset = sector / (sizeof(unsigned long) * 8);
1158 /* This takes care of the case where we're exactly at the end of the
1159 * device, and *cow_offset + 1 is off the end. So, just back it up
1160 * by one word. Thanks to Lynn Kerby for the fix and James McMechan
1161 * for the original diagnosis.
1163 if (*cow_offset == (DIV_ROUND_UP(bitmap_len,
1164 sizeof(unsigned long)) - 1))
1167 bitmap_words[0] = bitmap[*cow_offset];
1168 bitmap_words[1] = bitmap[*cow_offset + 1];
1170 *cow_offset *= sizeof(unsigned long);
1171 *cow_offset += bitmap_offset;
1174 static void cowify_req(struct io_thread_req *req, struct io_desc *segment,
1175 unsigned long offset, unsigned long *bitmap,
1176 __u64 bitmap_offset, __u64 bitmap_len)
1178 __u64 sector = offset >> SECTOR_SHIFT;
1181 if (segment->length > (sizeof(segment->sector_mask) * 8) << SECTOR_SHIFT)
1182 panic("Operation too long");
1184 if (req_op(req->req) == REQ_OP_READ) {
1185 for (i = 0; i < segment->length >> SECTOR_SHIFT; i++) {
1186 if(ubd_test_bit(sector + i, (unsigned char *) bitmap))
1187 ubd_set_bit(i, (unsigned char *)
1188 &segment->sector_mask);
1191 cowify_bitmap(offset, segment->length, &segment->sector_mask,
1192 &segment->cow_offset, bitmap, bitmap_offset,
1193 segment->bitmap_words, bitmap_len);
1197 static void ubd_map_req(struct ubd *dev, struct io_thread_req *io_req,
1198 struct request *req)
1200 struct bio_vec bvec;
1201 struct req_iterator iter;
1203 unsigned long byte_offset = io_req->offset;
1204 enum req_op op = req_op(req);
1206 if (op == REQ_OP_WRITE_ZEROES || op == REQ_OP_DISCARD) {
1207 io_req->io_desc[0].buffer = NULL;
1208 io_req->io_desc[0].length = blk_rq_bytes(req);
1210 rq_for_each_segment(bvec, req, iter) {
1211 BUG_ON(i >= io_req->desc_cnt);
1213 io_req->io_desc[i].buffer = bvec_virt(&bvec);
1214 io_req->io_desc[i].length = bvec.bv_len;
1219 if (dev->cow.file) {
1220 for (i = 0; i < io_req->desc_cnt; i++) {
1221 cowify_req(io_req, &io_req->io_desc[i], byte_offset,
1222 dev->cow.bitmap, dev->cow.bitmap_offset,
1223 dev->cow.bitmap_len);
1224 byte_offset += io_req->io_desc[i].length;
1230 static struct io_thread_req *ubd_alloc_req(struct ubd *dev, struct request *req,
1233 struct io_thread_req *io_req;
1236 io_req = kmalloc(sizeof(*io_req) +
1237 (desc_cnt * sizeof(struct io_desc)),
1244 io_req->fds[0] = dev->cow.fd;
1246 io_req->fds[0] = dev->fd;
1248 io_req->sectorsize = SECTOR_SIZE;
1249 io_req->fds[1] = dev->fd;
1250 io_req->offset = (u64) blk_rq_pos(req) << SECTOR_SHIFT;
1251 io_req->offsets[0] = 0;
1252 io_req->offsets[1] = dev->cow.data_offset;
1254 for (i = 0 ; i < desc_cnt; i++) {
1255 io_req->io_desc[i].sector_mask = 0;
1256 io_req->io_desc[i].cow_offset = -1;
1262 static int ubd_submit_request(struct ubd *dev, struct request *req)
1265 struct io_thread_req *io_req;
1267 enum req_op op = req_op(req);
1269 if (op == REQ_OP_FLUSH)
1271 else if (op == REQ_OP_WRITE_ZEROES || op == REQ_OP_DISCARD)
1274 segs = blk_rq_nr_phys_segments(req);
1276 io_req = ubd_alloc_req(dev, req, segs);
1280 io_req->desc_cnt = segs;
1282 ubd_map_req(dev, io_req, req);
1284 ret = os_write_file(thread_fd, &io_req, sizeof(io_req));
1285 if (ret != sizeof(io_req)) {
1287 pr_err("write to io thread failed: %d\n", -ret);
1293 static blk_status_t ubd_queue_rq(struct blk_mq_hw_ctx *hctx,
1294 const struct blk_mq_queue_data *bd)
1296 struct ubd *ubd_dev = hctx->queue->queuedata;
1297 struct request *req = bd->rq;
1298 int ret = 0, res = BLK_STS_OK;
1300 blk_mq_start_request(req);
1302 spin_lock_irq(&ubd_dev->lock);
1304 switch (req_op(req)) {
1308 case REQ_OP_DISCARD:
1309 case REQ_OP_WRITE_ZEROES:
1310 ret = ubd_submit_request(ubd_dev, req);
1314 res = BLK_STS_NOTSUPP;
1317 spin_unlock_irq(&ubd_dev->lock);
1321 res = BLK_STS_RESOURCE;
1323 res = BLK_STS_DEV_RESOURCE;
1329 static int ubd_getgeo(struct block_device *bdev, struct hd_geometry *geo)
1331 struct ubd *ubd_dev = bdev->bd_disk->private_data;
1335 geo->cylinders = ubd_dev->size / (128 * 32 * 512);
1339 static int ubd_ioctl(struct block_device *bdev, blk_mode_t mode,
1340 unsigned int cmd, unsigned long arg)
1342 struct ubd *ubd_dev = bdev->bd_disk->private_data;
1343 u16 ubd_id[ATA_ID_WORDS];
1346 struct cdrom_volctrl volume;
1347 case HDIO_GET_IDENTITY:
1348 memset(&ubd_id, 0, ATA_ID_WORDS * 2);
1349 ubd_id[ATA_ID_CYLS] = ubd_dev->size / (128 * 32 * 512);
1350 ubd_id[ATA_ID_HEADS] = 128;
1351 ubd_id[ATA_ID_SECTORS] = 32;
1352 if(copy_to_user((char __user *) arg, (char *) &ubd_id,
1358 if(copy_from_user(&volume, (char __user *) arg, sizeof(volume)))
1360 volume.channel0 = 255;
1361 volume.channel1 = 255;
1362 volume.channel2 = 255;
1363 volume.channel3 = 255;
1364 if(copy_to_user((char __user *) arg, &volume, sizeof(volume)))
1371 static int map_error(int error_code)
1373 switch (error_code) {
1378 return BLK_STS_NOTSUPP;
1380 return BLK_STS_NOSPC;
1382 return BLK_STS_IOERR;
1386 * Everything from here onwards *IS NOT PART OF THE KERNEL*
1388 * The following functions are part of UML hypervisor code.
1389 * All functions from here onwards are executed as a helper
1390 * thread and are not allowed to execute any kernel functions.
1392 * Any communication must occur strictly via shared memory and IPC.
1394 * Do not add printks, locks, kernel memory operations, etc - it
1395 * will result in unpredictable behaviour and/or crashes.
1398 static int update_bitmap(struct io_thread_req *req, struct io_desc *segment)
1402 if (segment->cow_offset == -1)
1403 return map_error(0);
1405 n = os_pwrite_file(req->fds[1], &segment->bitmap_words,
1406 sizeof(segment->bitmap_words), segment->cow_offset);
1407 if (n != sizeof(segment->bitmap_words))
1408 return map_error(-n);
1410 return map_error(0);
1413 static void do_io(struct io_thread_req *req, struct io_desc *desc)
1417 int n, nsectors, start, end, bit;
1420 /* FLUSH is really a special case, we cannot "case" it with others */
1422 if (req_op(req->req) == REQ_OP_FLUSH) {
1423 /* fds[0] is always either the rw image or our cow file */
1424 req->error = map_error(-os_sync_file(req->fds[0]));
1428 nsectors = desc->length / req->sectorsize;
1431 bit = ubd_test_bit(start, (unsigned char *) &desc->sector_mask);
1433 while((end < nsectors) &&
1434 (ubd_test_bit(end, (unsigned char *) &desc->sector_mask) == bit))
1437 off = req->offset + req->offsets[bit] +
1438 start * req->sectorsize;
1439 len = (end - start) * req->sectorsize;
1440 if (desc->buffer != NULL)
1441 buf = &desc->buffer[start * req->sectorsize];
1443 switch (req_op(req->req)) {
1449 n = os_pread_file(req->fds[bit], buf, len, off);
1451 req->error = map_error(-n);
1454 } while((n < len) && (n != 0));
1455 if (n < len) memset(&buf[n], 0, len - n);
1458 n = os_pwrite_file(req->fds[bit], buf, len, off);
1460 req->error = map_error(-n);
1464 case REQ_OP_DISCARD:
1465 n = os_falloc_punch(req->fds[bit], off, len);
1467 req->error = map_error(-n);
1471 case REQ_OP_WRITE_ZEROES:
1472 n = os_falloc_zeroes(req->fds[bit], off, len);
1474 req->error = map_error(-n);
1480 req->error = BLK_STS_NOTSUPP;
1485 } while(start < nsectors);
1488 req->error = update_bitmap(req, desc);
1491 /* Changed in start_io_thread, which is serialized by being called only
1492 * from ubd_init, which is an initcall.
1496 /* Only changed by the io thread. XXX: currently unused. */
1497 static int io_count;
1499 int io_thread(void *arg)
1501 int n, count, written, res;
1504 os_fix_helper_signals();
1507 n = bulk_req_safe_read(
1521 for (count = 0; count < n/sizeof(struct io_thread_req *); count++) {
1522 struct io_thread_req *req = (*io_req_buffer)[count];
1526 for (i = 0; !req->error && i < req->desc_cnt; i++)
1527 do_io(req, &(req->io_desc[i]));
1534 res = os_write_file(kernel_fd,
1535 ((char *) io_req_buffer) + written,
1543 } while (written < n);