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_ioctl(struct block_device *bdev, blk_mode_t mode,
112 unsigned int cmd, unsigned long arg);
113 static int ubd_getgeo(struct block_device *bdev, struct hd_geometry *geo);
117 static const struct block_device_operations ubd_blops = {
118 .owner = THIS_MODULE,
120 .compat_ioctl = blkdev_compat_ptr_ioctl,
121 .getgeo = ubd_getgeo,
124 #ifdef CONFIG_BLK_DEV_UBD_SYNC
125 #define OPEN_FLAGS ((struct openflags) { .r = 1, .w = 1, .s = 1, .c = 0, \
128 #define OPEN_FLAGS ((struct openflags) { .r = 1, .w = 1, .s = 0, .c = 0, \
131 static struct openflags global_openflags = OPEN_FLAGS;
134 /* backing file name */
136 /* backing file fd */
138 unsigned long *bitmap;
139 unsigned long bitmap_len;
147 /* name (and fd, below) of the file opened for writing, either the
148 * backing or the cow file. */
153 struct openflags boot_openflags;
154 struct openflags openflags;
159 struct platform_device pdev;
160 struct gendisk *disk;
161 struct blk_mq_tag_set tag_set;
165 #define DEFAULT_COW { \
169 .bitmap_offset = 0, \
173 #define DEFAULT_UBD { \
178 .boot_openflags = OPEN_FLAGS, \
179 .openflags = OPEN_FLAGS, \
183 .cow = DEFAULT_COW, \
184 .lock = __SPIN_LOCK_UNLOCKED(ubd_devs.lock), \
187 /* Protected by ubd_lock */
188 static struct ubd ubd_devs[MAX_DEV] = { [0 ... MAX_DEV - 1] = DEFAULT_UBD };
190 static blk_status_t ubd_queue_rq(struct blk_mq_hw_ctx *hctx,
191 const struct blk_mq_queue_data *bd);
193 static int fake_ide_setup(char *str)
195 pr_warn("The fake_ide option has been removed\n");
198 __setup("fake_ide", fake_ide_setup);
200 __uml_help(fake_ide_setup,
202 " Obsolete stub.\n\n"
205 static int parse_unit(char **ptr)
207 char *str = *ptr, *end;
211 n = simple_strtoul(str, &end, 0);
216 else if (('a' <= *str) && (*str <= 'z')) {
224 /* If *index_out == -1 at exit, the passed option was a general one;
225 * otherwise, the str pointer is used (and owned) inside ubd_devs array, so it
226 * should not be freed on exit.
228 static int ubd_setup_common(char *str, int *index_out, char **error_out)
231 struct openflags flags = global_openflags;
232 char *file, *backing_file, *serial;
235 if(index_out) *index_out = -1;
239 if(!strcmp(str, "sync")){
240 global_openflags = of_sync(global_openflags);
244 pr_warn("fake major not supported any more\n");
248 n = parse_unit(&str);
250 *error_out = "Couldn't parse device number";
254 *error_out = "Device number out of range";
259 mutex_lock(&ubd_lock);
261 ubd_dev = &ubd_devs[n];
262 if(ubd_dev->file != NULL){
263 *error_out = "Device is already configured";
271 for (i = 0; i < sizeof("rscdt="); i++) {
286 ubd_dev->no_trim = 1;
292 *error_out = "Expected '=' or flag letter "
300 *error_out = "Too many flags specified";
302 *error_out = "Missing '='";
306 file = strsep(&str, ",:");
310 backing_file = strsep(&str, ",:");
311 if (backing_file && *backing_file == '\0')
314 serial = strsep(&str, ",:");
315 if (serial && *serial == '\0')
318 if (backing_file && ubd_dev->no_cow) {
319 *error_out = "Can't specify both 'd' and a cow file";
324 ubd_dev->file = file;
325 ubd_dev->cow.file = backing_file;
326 ubd_dev->serial = serial;
327 ubd_dev->boot_openflags = flags;
329 mutex_unlock(&ubd_lock);
333 static int ubd_setup(char *str)
338 err = ubd_setup_common(str, NULL, &error);
340 printk(KERN_ERR "Failed to initialize device with \"%s\" : "
345 __setup("ubd", ubd_setup);
346 __uml_help(ubd_setup,
347 "ubd<n><flags>=<filename>[(:|,)<filename2>][(:|,)<serial>]\n"
348 " This is used to associate a device with a file in the underlying\n"
349 " filesystem. When specifying two filenames, the first one is the\n"
350 " COW name and the second is the backing file name. As separator you can\n"
351 " use either a ':' or a ',': the first one allows writing things like;\n"
352 " ubd0=~/Uml/root_cow:~/Uml/root_backing_file\n"
353 " while with a ',' the shell would not expand the 2nd '~'.\n"
354 " When using only one filename, UML will detect whether to treat it like\n"
355 " a COW file or a backing file. To override this detection, add the 'd'\n"
357 " ubd0d=BackingFile\n"
358 " Usually, there is a filesystem in the file, but \n"
359 " that's not required. Swap devices containing swap files can be\n"
360 " specified like this. Also, a file which doesn't contain a\n"
361 " filesystem can have its contents read in the virtual \n"
362 " machine by running 'dd' on the device. <n> must be in the range\n"
363 " 0 to 7. Appending an 'r' to the number will cause that device\n"
364 " to be mounted read-only. For example ubd1r=./ext_fs. Appending\n"
365 " an 's' will cause data to be written to disk on the host immediately.\n"
366 " 'c' will cause the device to be treated as being shared between multiple\n"
367 " UMLs and file locking will be turned off - this is appropriate for a\n"
368 " cluster filesystem and inappropriate at almost all other times.\n\n"
369 " 't' will disable trim/discard support on the device (enabled by default).\n\n"
370 " An optional device serial number can be exposed using the serial parameter\n"
371 " on the cmdline which is exposed as a sysfs entry. This is particularly\n"
372 " useful when a unique number should be given to the device. Note when\n"
373 " specifying a label, the filename2 must be also presented. It can be\n"
374 " an empty string, in which case the backing file is not used:\n"
375 " ubd0=File,,Serial\n"
378 static int udb_setup(char *str)
380 printk("udb%s specified on command line is almost certainly a ubd -> "
385 __setup("udb", udb_setup);
386 __uml_help(udb_setup,
388 " This option is here solely to catch ubd -> udb typos, which can be\n"
389 " to impossible to catch visually unless you specifically look for\n"
390 " them. The only result of any option starting with 'udb' is an error\n"
391 " in the boot output.\n\n"
394 /* Only changed by ubd_init, which is an initcall. */
395 static int thread_fd = -1;
397 /* Function to read several request pointers at a time
398 * handling fractional reads if (and as) needed
401 static int bulk_req_safe_read(
403 struct io_thread_req * (*request_buffer)[],
404 struct io_thread_req **remainder,
412 if (*remainder_size > 0) {
414 (char *) request_buffer,
415 (char *) remainder, *remainder_size
422 ((char *) request_buffer) + *remainder_size,
423 sizeof(struct io_thread_req *)*max_recs
428 if ((n % sizeof(struct io_thread_req *)) > 0) {
430 * Read somehow returned not a multiple of dword
431 * theoretically possible, but never observed in the
432 * wild, so read routine must be able to handle it
434 *remainder_size = n % sizeof(struct io_thread_req *);
435 WARN(*remainder_size > 0, "UBD IPC read returned a partial result");
438 ((char *) request_buffer) +
439 (n/sizeof(struct io_thread_req *))*sizeof(struct io_thread_req *),
442 n = n - *remainder_size;
450 /* Called without dev->lock held, and only in interrupt context. */
451 static void ubd_handler(void)
457 n = bulk_req_safe_read(
467 printk(KERN_ERR "spurious interrupt in ubd_handler, "
471 for (count = 0; count < n/sizeof(struct io_thread_req *); count++) {
472 struct io_thread_req *io_req = (*irq_req_buffer)[count];
474 if ((io_req->error == BLK_STS_NOTSUPP) && (req_op(io_req->req) == REQ_OP_DISCARD)) {
475 blk_queue_max_discard_sectors(io_req->req->q, 0);
476 blk_queue_max_write_zeroes_sectors(io_req->req->q, 0);
478 blk_mq_end_request(io_req->req, io_req->error);
484 static irqreturn_t ubd_intr(int irq, void *dev)
490 /* Only changed by ubd_init, which is an initcall. */
491 static int io_pid = -1;
493 static void kill_io_thread(void)
496 os_kill_process(io_pid, 1);
499 __uml_exitcall(kill_io_thread);
501 static inline int ubd_file_size(struct ubd *ubd_dev, __u64 *size_out)
511 unsigned long long size;
515 if (ubd_dev->file && ubd_dev->cow.file) {
516 file = ubd_dev->cow.file;
521 fd = os_open_file(ubd_dev->file, of_read(OPENFLAGS()), 0);
525 err = read_cow_header(file_reader, &fd, &version, &backing_file, \
526 &mtime, &size, §or_size, &align, &bitmap_offset);
530 file = ubd_dev->file;
535 return os_file_size(file, size_out);
538 static int read_cow_bitmap(int fd, void *buf, int offset, int len)
542 err = os_pread_file(fd, buf, len, offset);
549 static int backing_file_mismatch(char *file, __u64 size, time64_t mtime)
552 unsigned long long actual;
555 err = os_file_modtime(file, &modtime);
557 printk(KERN_ERR "Failed to get modification time of backing "
558 "file \"%s\", err = %d\n", file, -err);
562 err = os_file_size(file, &actual);
564 printk(KERN_ERR "Failed to get size of backing file \"%s\", "
565 "err = %d\n", file, -err);
569 if (actual != size) {
570 /*__u64 can be a long on AMD64 and with %lu GCC complains; so
572 printk(KERN_ERR "Size mismatch (%llu vs %llu) of COW header "
573 "vs backing file\n", (unsigned long long) size, actual);
576 if (modtime != mtime) {
577 printk(KERN_ERR "mtime mismatch (%lld vs %lld) of COW header vs "
578 "backing file\n", mtime, modtime);
584 static int path_requires_switch(char *from_cmdline, char *from_cow, char *cow)
586 struct uml_stat buf1, buf2;
589 if (from_cmdline == NULL)
591 if (!strcmp(from_cmdline, from_cow))
594 err = os_stat_file(from_cmdline, &buf1);
596 printk(KERN_ERR "Couldn't stat '%s', err = %d\n", from_cmdline,
600 err = os_stat_file(from_cow, &buf2);
602 printk(KERN_ERR "Couldn't stat '%s', err = %d\n", from_cow,
606 if ((buf1.ust_dev == buf2.ust_dev) && (buf1.ust_ino == buf2.ust_ino))
609 printk(KERN_ERR "Backing file mismatch - \"%s\" requested, "
610 "\"%s\" specified in COW header of \"%s\"\n",
611 from_cmdline, from_cow, cow);
615 static int open_ubd_file(char *file, struct openflags *openflags, int shared,
616 char **backing_file_out, int *bitmap_offset_out,
617 unsigned long *bitmap_len_out, int *data_offset_out,
621 unsigned long long size;
622 __u32 version, align;
624 int fd, err, sectorsize, asked_switch, mode = 0644;
626 fd = os_open_file(file, *openflags, mode);
628 if ((fd == -ENOENT) && (create_cow_out != NULL))
631 ((fd != -EROFS) && (fd != -EACCES)))
634 fd = os_open_file(file, *openflags, mode);
640 printk(KERN_INFO "Not locking \"%s\" on the host\n", file);
642 err = os_lock_file(fd, openflags->w);
644 printk(KERN_ERR "Failed to lock '%s', err = %d\n",
650 /* Successful return case! */
651 if (backing_file_out == NULL)
654 err = read_cow_header(file_reader, &fd, &version, &backing_file, &mtime,
655 &size, §orsize, &align, bitmap_offset_out);
656 if (err && (*backing_file_out != NULL)) {
657 printk(KERN_ERR "Failed to read COW header from COW file "
658 "\"%s\", errno = %d\n", file, -err);
664 asked_switch = path_requires_switch(*backing_file_out, backing_file,
667 /* Allow switching only if no mismatch. */
668 if (asked_switch && !backing_file_mismatch(*backing_file_out, size,
670 printk(KERN_ERR "Switching backing file to '%s'\n",
672 err = write_cow_header(file, fd, *backing_file_out,
673 sectorsize, align, &size);
675 printk(KERN_ERR "Switch failed, errno = %d\n", -err);
679 *backing_file_out = backing_file;
680 err = backing_file_mismatch(*backing_file_out, size, mtime);
685 cow_sizes(version, size, sectorsize, align, *bitmap_offset_out,
686 bitmap_len_out, data_offset_out);
694 static int create_cow_file(char *cow_file, char *backing_file,
695 struct openflags flags,
696 int sectorsize, int alignment, int *bitmap_offset_out,
697 unsigned long *bitmap_len_out, int *data_offset_out)
702 fd = open_ubd_file(cow_file, &flags, 0, NULL, NULL, NULL, NULL, NULL);
705 printk(KERN_ERR "Open of COW file '%s' failed, errno = %d\n",
710 err = init_cow_file(fd, cow_file, backing_file, sectorsize, alignment,
711 bitmap_offset_out, bitmap_len_out,
720 static void ubd_close_dev(struct ubd *ubd_dev)
722 os_close_file(ubd_dev->fd);
723 if(ubd_dev->cow.file == NULL)
726 os_close_file(ubd_dev->cow.fd);
727 vfree(ubd_dev->cow.bitmap);
728 ubd_dev->cow.bitmap = NULL;
731 static int ubd_open_dev(struct ubd *ubd_dev)
733 struct openflags flags;
735 int err, create_cow, *create_ptr;
738 ubd_dev->openflags = ubd_dev->boot_openflags;
740 create_ptr = (ubd_dev->cow.file != NULL) ? &create_cow : NULL;
741 back_ptr = ubd_dev->no_cow ? NULL : &ubd_dev->cow.file;
743 fd = open_ubd_file(ubd_dev->file, &ubd_dev->openflags, ubd_dev->shared,
744 back_ptr, &ubd_dev->cow.bitmap_offset,
745 &ubd_dev->cow.bitmap_len, &ubd_dev->cow.data_offset,
748 if((fd == -ENOENT) && create_cow){
749 fd = create_cow_file(ubd_dev->file, ubd_dev->cow.file,
750 ubd_dev->openflags, SECTOR_SIZE, PAGE_SIZE,
751 &ubd_dev->cow.bitmap_offset,
752 &ubd_dev->cow.bitmap_len,
753 &ubd_dev->cow.data_offset);
755 printk(KERN_INFO "Creating \"%s\" as COW file for "
756 "\"%s\"\n", ubd_dev->file, ubd_dev->cow.file);
761 printk("Failed to open '%s', errno = %d\n", ubd_dev->file,
767 if(ubd_dev->cow.file != NULL){
769 ubd_dev->cow.bitmap = vmalloc(ubd_dev->cow.bitmap_len);
770 if(ubd_dev->cow.bitmap == NULL){
771 printk(KERN_ERR "Failed to vmalloc COW bitmap\n");
774 flush_tlb_kernel_vm();
776 err = read_cow_bitmap(ubd_dev->fd, ubd_dev->cow.bitmap,
777 ubd_dev->cow.bitmap_offset,
778 ubd_dev->cow.bitmap_len);
782 flags = ubd_dev->openflags;
784 err = open_ubd_file(ubd_dev->cow.file, &flags, ubd_dev->shared, NULL,
785 NULL, NULL, NULL, NULL);
786 if(err < 0) goto error;
787 ubd_dev->cow.fd = err;
791 os_close_file(ubd_dev->fd);
795 static void ubd_device_release(struct device *dev)
797 struct ubd *ubd_dev = dev_get_drvdata(dev);
799 blk_mq_free_tag_set(&ubd_dev->tag_set);
800 *ubd_dev = ((struct ubd) DEFAULT_UBD);
803 static ssize_t serial_show(struct device *dev,
804 struct device_attribute *attr, char *buf)
806 struct gendisk *disk = dev_to_disk(dev);
807 struct ubd *ubd_dev = disk->private_data;
812 return sprintf(buf, "%s", ubd_dev->serial);
815 static DEVICE_ATTR_RO(serial);
817 static struct attribute *ubd_attrs[] = {
818 &dev_attr_serial.attr,
822 static umode_t ubd_attrs_are_visible(struct kobject *kobj,
823 struct attribute *a, int n)
828 static const struct attribute_group ubd_attr_group = {
830 .is_visible = ubd_attrs_are_visible,
833 static const struct attribute_group *ubd_attr_groups[] = {
838 #define ROUND_BLOCK(n) ((n + (SECTOR_SIZE - 1)) & (-SECTOR_SIZE))
840 static const struct blk_mq_ops ubd_mq_ops = {
841 .queue_rq = ubd_queue_rq,
844 static int ubd_add(int n, char **error_out)
846 struct ubd *ubd_dev = &ubd_devs[n];
847 struct queue_limits lim = {
848 .max_segments = MAX_SG,
849 .seg_boundary_mask = PAGE_SIZE - 1,
851 struct gendisk *disk;
854 if(ubd_dev->file == NULL)
857 if (ubd_dev->cow.file)
858 lim.max_hw_sectors = 8 * sizeof(long);
859 if (!ubd_dev->no_trim) {
860 lim.max_hw_discard_sectors = UBD_MAX_REQUEST;
861 lim.max_write_zeroes_sectors = UBD_MAX_REQUEST;
864 err = ubd_file_size(ubd_dev, &ubd_dev->size);
866 *error_out = "Couldn't determine size of device's file";
870 err = ubd_open_dev(ubd_dev);
872 pr_err("ubd%c: Can't open \"%s\": errno = %d\n",
873 'a' + n, ubd_dev->file, -err);
877 ubd_dev->size = ROUND_BLOCK(ubd_dev->size);
879 ubd_dev->tag_set.ops = &ubd_mq_ops;
880 ubd_dev->tag_set.queue_depth = 64;
881 ubd_dev->tag_set.numa_node = NUMA_NO_NODE;
882 ubd_dev->tag_set.flags = BLK_MQ_F_SHOULD_MERGE;
883 ubd_dev->tag_set.driver_data = ubd_dev;
884 ubd_dev->tag_set.nr_hw_queues = 1;
886 err = blk_mq_alloc_tag_set(&ubd_dev->tag_set);
890 disk = blk_mq_alloc_disk(&ubd_dev->tag_set, &lim, ubd_dev);
893 goto out_cleanup_tags;
896 blk_queue_flag_set(QUEUE_FLAG_NONROT, disk->queue);
897 blk_queue_write_cache(disk->queue, true, false);
898 disk->major = UBD_MAJOR;
899 disk->first_minor = n << UBD_SHIFT;
900 disk->minors = 1 << UBD_SHIFT;
901 disk->fops = &ubd_blops;
902 set_capacity(disk, ubd_dev->size / 512);
903 sprintf(disk->disk_name, "ubd%c", 'a' + n);
904 disk->private_data = ubd_dev;
905 set_disk_ro(disk, !ubd_dev->openflags.w);
907 ubd_dev->pdev.id = n;
908 ubd_dev->pdev.name = DRIVER_NAME;
909 ubd_dev->pdev.dev.release = ubd_device_release;
910 dev_set_drvdata(&ubd_dev->pdev.dev, ubd_dev);
911 platform_device_register(&ubd_dev->pdev);
913 err = device_add_disk(&ubd_dev->pdev.dev, disk, ubd_attr_groups);
915 goto out_cleanup_disk;
922 blk_mq_free_tag_set(&ubd_dev->tag_set);
924 ubd_close_dev(ubd_dev);
929 static int ubd_config(char *str, char **error_out)
933 /* This string is possibly broken up and stored, so it's only
934 * freed if ubd_setup_common fails, or if only general options
937 str = kstrdup(str, GFP_KERNEL);
939 *error_out = "Failed to allocate memory";
943 ret = ubd_setup_common(str, &n, error_out);
952 mutex_lock(&ubd_lock);
953 ret = ubd_add(n, error_out);
955 ubd_devs[n].file = NULL;
956 mutex_unlock(&ubd_lock);
966 static int ubd_get_config(char *name, char *str, int size, char **error_out)
971 n = parse_unit(&name);
972 if((n >= MAX_DEV) || (n < 0)){
973 *error_out = "ubd_get_config : device number out of range";
977 ubd_dev = &ubd_devs[n];
978 mutex_lock(&ubd_lock);
980 if(ubd_dev->file == NULL){
981 CONFIG_CHUNK(str, size, len, "", 1);
985 CONFIG_CHUNK(str, size, len, ubd_dev->file, 0);
987 if(ubd_dev->cow.file != NULL){
988 CONFIG_CHUNK(str, size, len, ",", 0);
989 CONFIG_CHUNK(str, size, len, ubd_dev->cow.file, 1);
991 else CONFIG_CHUNK(str, size, len, "", 1);
994 mutex_unlock(&ubd_lock);
998 static int ubd_id(char **str, int *start_out, int *end_out)
1002 n = parse_unit(str);
1004 *end_out = MAX_DEV - 1;
1008 static int ubd_remove(int n, char **error_out)
1010 struct ubd *ubd_dev;
1013 mutex_lock(&ubd_lock);
1015 ubd_dev = &ubd_devs[n];
1017 if(ubd_dev->file == NULL)
1020 if (ubd_dev->disk) {
1021 /* you cannot remove a open disk */
1023 if (disk_openers(ubd_dev->disk))
1026 del_gendisk(ubd_dev->disk);
1027 ubd_close_dev(ubd_dev);
1028 put_disk(ubd_dev->disk);
1032 platform_device_unregister(&ubd_dev->pdev);
1034 mutex_unlock(&ubd_lock);
1038 /* All these are called by mconsole in process context and without
1039 * ubd-specific locks. The structure itself is const except for .list.
1041 static struct mc_device ubd_mc = {
1042 .list = LIST_HEAD_INIT(ubd_mc.list),
1044 .config = ubd_config,
1045 .get_config = ubd_get_config,
1047 .remove = ubd_remove,
1050 static int __init ubd_mc_init(void)
1052 mconsole_register_dev(&ubd_mc);
1056 __initcall(ubd_mc_init);
1058 static int __init ubd0_init(void)
1060 struct ubd *ubd_dev = &ubd_devs[0];
1062 mutex_lock(&ubd_lock);
1063 if(ubd_dev->file == NULL)
1064 ubd_dev->file = "root_fs";
1065 mutex_unlock(&ubd_lock);
1070 __initcall(ubd0_init);
1072 /* Used in ubd_init, which is an initcall */
1073 static struct platform_driver ubd_driver = {
1075 .name = DRIVER_NAME,
1079 static int __init ubd_init(void)
1084 if (register_blkdev(UBD_MAJOR, "ubd"))
1087 irq_req_buffer = kmalloc_array(UBD_REQ_BUFFER_SIZE,
1088 sizeof(struct io_thread_req *),
1093 if (irq_req_buffer == NULL) {
1094 printk(KERN_ERR "Failed to initialize ubd buffering\n");
1097 io_req_buffer = kmalloc_array(UBD_REQ_BUFFER_SIZE,
1098 sizeof(struct io_thread_req *),
1104 if (io_req_buffer == NULL) {
1105 printk(KERN_ERR "Failed to initialize ubd buffering\n");
1108 platform_driver_register(&ubd_driver);
1109 mutex_lock(&ubd_lock);
1110 for (i = 0; i < MAX_DEV; i++){
1111 err = ubd_add(i, &error);
1113 printk(KERN_ERR "Failed to initialize ubd device %d :"
1116 mutex_unlock(&ubd_lock);
1120 late_initcall(ubd_init);
1122 static int __init ubd_driver_init(void){
1123 unsigned long stack;
1126 /* Set by CONFIG_BLK_DEV_UBD_SYNC or ubd=sync.*/
1127 if(global_openflags.s){
1128 printk(KERN_INFO "ubd: Synchronous mode\n");
1129 /* Letting ubd=sync be like using ubd#s= instead of ubd#= is
1130 * enough. So use anyway the io thread. */
1132 stack = alloc_stack(0, 0);
1133 io_pid = start_io_thread(stack + PAGE_SIZE, &thread_fd);
1136 "ubd : Failed to start I/O thread (errno = %d) - "
1137 "falling back to synchronous I/O\n", -io_pid);
1141 err = um_request_irq(UBD_IRQ, thread_fd, IRQ_READ, ubd_intr,
1142 0, "ubd", ubd_devs);
1144 printk(KERN_ERR "um_request_irq failed - errno = %d\n", -err);
1148 device_initcall(ubd_driver_init);
1150 static void cowify_bitmap(__u64 io_offset, int length, unsigned long *cow_mask,
1151 __u64 *cow_offset, unsigned long *bitmap,
1152 __u64 bitmap_offset, unsigned long *bitmap_words,
1155 __u64 sector = io_offset >> SECTOR_SHIFT;
1156 int i, update_bitmap = 0;
1158 for (i = 0; i < length >> SECTOR_SHIFT; i++) {
1159 if(cow_mask != NULL)
1160 ubd_set_bit(i, (unsigned char *) cow_mask);
1161 if(ubd_test_bit(sector + i, (unsigned char *) bitmap))
1165 ubd_set_bit(sector + i, (unsigned char *) bitmap);
1171 *cow_offset = sector / (sizeof(unsigned long) * 8);
1173 /* This takes care of the case where we're exactly at the end of the
1174 * device, and *cow_offset + 1 is off the end. So, just back it up
1175 * by one word. Thanks to Lynn Kerby for the fix and James McMechan
1176 * for the original diagnosis.
1178 if (*cow_offset == (DIV_ROUND_UP(bitmap_len,
1179 sizeof(unsigned long)) - 1))
1182 bitmap_words[0] = bitmap[*cow_offset];
1183 bitmap_words[1] = bitmap[*cow_offset + 1];
1185 *cow_offset *= sizeof(unsigned long);
1186 *cow_offset += bitmap_offset;
1189 static void cowify_req(struct io_thread_req *req, struct io_desc *segment,
1190 unsigned long offset, unsigned long *bitmap,
1191 __u64 bitmap_offset, __u64 bitmap_len)
1193 __u64 sector = offset >> SECTOR_SHIFT;
1196 if (segment->length > (sizeof(segment->sector_mask) * 8) << SECTOR_SHIFT)
1197 panic("Operation too long");
1199 if (req_op(req->req) == REQ_OP_READ) {
1200 for (i = 0; i < segment->length >> SECTOR_SHIFT; i++) {
1201 if(ubd_test_bit(sector + i, (unsigned char *) bitmap))
1202 ubd_set_bit(i, (unsigned char *)
1203 &segment->sector_mask);
1206 cowify_bitmap(offset, segment->length, &segment->sector_mask,
1207 &segment->cow_offset, bitmap, bitmap_offset,
1208 segment->bitmap_words, bitmap_len);
1212 static void ubd_map_req(struct ubd *dev, struct io_thread_req *io_req,
1213 struct request *req)
1215 struct bio_vec bvec;
1216 struct req_iterator iter;
1218 unsigned long byte_offset = io_req->offset;
1219 enum req_op op = req_op(req);
1221 if (op == REQ_OP_WRITE_ZEROES || op == REQ_OP_DISCARD) {
1222 io_req->io_desc[0].buffer = NULL;
1223 io_req->io_desc[0].length = blk_rq_bytes(req);
1225 rq_for_each_segment(bvec, req, iter) {
1226 BUG_ON(i >= io_req->desc_cnt);
1228 io_req->io_desc[i].buffer = bvec_virt(&bvec);
1229 io_req->io_desc[i].length = bvec.bv_len;
1234 if (dev->cow.file) {
1235 for (i = 0; i < io_req->desc_cnt; i++) {
1236 cowify_req(io_req, &io_req->io_desc[i], byte_offset,
1237 dev->cow.bitmap, dev->cow.bitmap_offset,
1238 dev->cow.bitmap_len);
1239 byte_offset += io_req->io_desc[i].length;
1245 static struct io_thread_req *ubd_alloc_req(struct ubd *dev, struct request *req,
1248 struct io_thread_req *io_req;
1251 io_req = kmalloc(sizeof(*io_req) +
1252 (desc_cnt * sizeof(struct io_desc)),
1259 io_req->fds[0] = dev->cow.fd;
1261 io_req->fds[0] = dev->fd;
1263 io_req->sectorsize = SECTOR_SIZE;
1264 io_req->fds[1] = dev->fd;
1265 io_req->offset = (u64) blk_rq_pos(req) << SECTOR_SHIFT;
1266 io_req->offsets[0] = 0;
1267 io_req->offsets[1] = dev->cow.data_offset;
1269 for (i = 0 ; i < desc_cnt; i++) {
1270 io_req->io_desc[i].sector_mask = 0;
1271 io_req->io_desc[i].cow_offset = -1;
1277 static int ubd_submit_request(struct ubd *dev, struct request *req)
1280 struct io_thread_req *io_req;
1282 enum req_op op = req_op(req);
1284 if (op == REQ_OP_FLUSH)
1286 else if (op == REQ_OP_WRITE_ZEROES || op == REQ_OP_DISCARD)
1289 segs = blk_rq_nr_phys_segments(req);
1291 io_req = ubd_alloc_req(dev, req, segs);
1295 io_req->desc_cnt = segs;
1297 ubd_map_req(dev, io_req, req);
1299 ret = os_write_file(thread_fd, &io_req, sizeof(io_req));
1300 if (ret != sizeof(io_req)) {
1302 pr_err("write to io thread failed: %d\n", -ret);
1308 static blk_status_t ubd_queue_rq(struct blk_mq_hw_ctx *hctx,
1309 const struct blk_mq_queue_data *bd)
1311 struct ubd *ubd_dev = hctx->queue->queuedata;
1312 struct request *req = bd->rq;
1313 int ret = 0, res = BLK_STS_OK;
1315 blk_mq_start_request(req);
1317 spin_lock_irq(&ubd_dev->lock);
1319 switch (req_op(req)) {
1323 case REQ_OP_DISCARD:
1324 case REQ_OP_WRITE_ZEROES:
1325 ret = ubd_submit_request(ubd_dev, req);
1329 res = BLK_STS_NOTSUPP;
1332 spin_unlock_irq(&ubd_dev->lock);
1336 res = BLK_STS_RESOURCE;
1338 res = BLK_STS_DEV_RESOURCE;
1344 static int ubd_getgeo(struct block_device *bdev, struct hd_geometry *geo)
1346 struct ubd *ubd_dev = bdev->bd_disk->private_data;
1350 geo->cylinders = ubd_dev->size / (128 * 32 * 512);
1354 static int ubd_ioctl(struct block_device *bdev, blk_mode_t mode,
1355 unsigned int cmd, unsigned long arg)
1357 struct ubd *ubd_dev = bdev->bd_disk->private_data;
1358 u16 ubd_id[ATA_ID_WORDS];
1361 struct cdrom_volctrl volume;
1362 case HDIO_GET_IDENTITY:
1363 memset(&ubd_id, 0, ATA_ID_WORDS * 2);
1364 ubd_id[ATA_ID_CYLS] = ubd_dev->size / (128 * 32 * 512);
1365 ubd_id[ATA_ID_HEADS] = 128;
1366 ubd_id[ATA_ID_SECTORS] = 32;
1367 if(copy_to_user((char __user *) arg, (char *) &ubd_id,
1373 if(copy_from_user(&volume, (char __user *) arg, sizeof(volume)))
1375 volume.channel0 = 255;
1376 volume.channel1 = 255;
1377 volume.channel2 = 255;
1378 volume.channel3 = 255;
1379 if(copy_to_user((char __user *) arg, &volume, sizeof(volume)))
1386 static int map_error(int error_code)
1388 switch (error_code) {
1393 return BLK_STS_NOTSUPP;
1395 return BLK_STS_NOSPC;
1397 return BLK_STS_IOERR;
1401 * Everything from here onwards *IS NOT PART OF THE KERNEL*
1403 * The following functions are part of UML hypervisor code.
1404 * All functions from here onwards are executed as a helper
1405 * thread and are not allowed to execute any kernel functions.
1407 * Any communication must occur strictly via shared memory and IPC.
1409 * Do not add printks, locks, kernel memory operations, etc - it
1410 * will result in unpredictable behaviour and/or crashes.
1413 static int update_bitmap(struct io_thread_req *req, struct io_desc *segment)
1417 if (segment->cow_offset == -1)
1418 return map_error(0);
1420 n = os_pwrite_file(req->fds[1], &segment->bitmap_words,
1421 sizeof(segment->bitmap_words), segment->cow_offset);
1422 if (n != sizeof(segment->bitmap_words))
1423 return map_error(-n);
1425 return map_error(0);
1428 static void do_io(struct io_thread_req *req, struct io_desc *desc)
1432 int n, nsectors, start, end, bit;
1435 /* FLUSH is really a special case, we cannot "case" it with others */
1437 if (req_op(req->req) == REQ_OP_FLUSH) {
1438 /* fds[0] is always either the rw image or our cow file */
1439 req->error = map_error(-os_sync_file(req->fds[0]));
1443 nsectors = desc->length / req->sectorsize;
1446 bit = ubd_test_bit(start, (unsigned char *) &desc->sector_mask);
1448 while((end < nsectors) &&
1449 (ubd_test_bit(end, (unsigned char *) &desc->sector_mask) == bit))
1452 off = req->offset + req->offsets[bit] +
1453 start * req->sectorsize;
1454 len = (end - start) * req->sectorsize;
1455 if (desc->buffer != NULL)
1456 buf = &desc->buffer[start * req->sectorsize];
1458 switch (req_op(req->req)) {
1464 n = os_pread_file(req->fds[bit], buf, len, off);
1466 req->error = map_error(-n);
1469 } while((n < len) && (n != 0));
1470 if (n < len) memset(&buf[n], 0, len - n);
1473 n = os_pwrite_file(req->fds[bit], buf, len, off);
1475 req->error = map_error(-n);
1479 case REQ_OP_DISCARD:
1480 n = os_falloc_punch(req->fds[bit], off, len);
1482 req->error = map_error(-n);
1486 case REQ_OP_WRITE_ZEROES:
1487 n = os_falloc_zeroes(req->fds[bit], off, len);
1489 req->error = map_error(-n);
1495 req->error = BLK_STS_NOTSUPP;
1500 } while(start < nsectors);
1503 req->error = update_bitmap(req, desc);
1506 /* Changed in start_io_thread, which is serialized by being called only
1507 * from ubd_init, which is an initcall.
1511 /* Only changed by the io thread. XXX: currently unused. */
1512 static int io_count;
1514 int io_thread(void *arg)
1516 int n, count, written, res;
1518 os_fix_helper_signals();
1521 n = bulk_req_safe_read(
1535 for (count = 0; count < n/sizeof(struct io_thread_req *); count++) {
1536 struct io_thread_req *req = (*io_req_buffer)[count];
1540 for (i = 0; !req->error && i < req->desc_cnt; i++)
1541 do_io(req, &(req->io_desc[i]));
1548 res = os_write_file(kernel_fd,
1549 ((char *) io_req_buffer) + written,
1557 } while (written < n);