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))
50 struct io_thread_req {
53 unsigned long offsets[2];
54 unsigned long long offset;
58 unsigned long sector_mask;
59 unsigned long long cow_offset;
60 unsigned long bitmap_words[2];
65 static struct io_thread_req * (*irq_req_buffer)[];
66 static struct io_thread_req *irq_remainder;
67 static int irq_remainder_size;
69 static struct io_thread_req * (*io_req_buffer)[];
70 static struct io_thread_req *io_remainder;
71 static int io_remainder_size;
75 static inline int ubd_test_bit(__u64 bit, unsigned char *data)
80 bits = sizeof(data[0]) * 8;
83 return (data[n] & (1 << off)) != 0;
86 static inline void ubd_set_bit(__u64 bit, unsigned char *data)
91 bits = sizeof(data[0]) * 8;
94 data[n] |= (1 << off);
96 /*End stuff from ubd_user.h*/
98 #define DRIVER_NAME "uml-blkdev"
100 static DEFINE_MUTEX(ubd_lock);
101 static DEFINE_MUTEX(ubd_mutex); /* replaces BKL, might not be needed */
103 static int ubd_open(struct block_device *bdev, fmode_t mode);
104 static void ubd_release(struct gendisk *disk, fmode_t mode);
105 static int ubd_ioctl(struct block_device *bdev, fmode_t mode,
106 unsigned int cmd, unsigned long arg);
107 static int ubd_getgeo(struct block_device *bdev, struct hd_geometry *geo);
111 static const struct block_device_operations ubd_blops = {
112 .owner = THIS_MODULE,
114 .release = ubd_release,
116 .compat_ioctl = blkdev_compat_ptr_ioctl,
117 .getgeo = ubd_getgeo,
120 /* Protected by ubd_lock */
121 static int fake_major = UBD_MAJOR;
122 static struct gendisk *ubd_gendisk[MAX_DEV];
123 static struct gendisk *fake_gendisk[MAX_DEV];
125 #ifdef CONFIG_BLK_DEV_UBD_SYNC
126 #define OPEN_FLAGS ((struct openflags) { .r = 1, .w = 1, .s = 1, .c = 0, \
129 #define OPEN_FLAGS ((struct openflags) { .r = 1, .w = 1, .s = 0, .c = 0, \
132 static struct openflags global_openflags = OPEN_FLAGS;
135 /* backing file name */
137 /* backing file fd */
139 unsigned long *bitmap;
140 unsigned long bitmap_len;
148 /* name (and fd, below) of the file opened for writing, either the
149 * backing or the cow file. */
154 struct openflags boot_openflags;
155 struct openflags openflags;
160 struct platform_device pdev;
161 struct request_queue *queue;
162 struct blk_mq_tag_set tag_set;
166 #define DEFAULT_COW { \
170 .bitmap_offset = 0, \
174 #define DEFAULT_UBD { \
179 .boot_openflags = OPEN_FLAGS, \
180 .openflags = OPEN_FLAGS, \
184 .cow = DEFAULT_COW, \
185 .lock = __SPIN_LOCK_UNLOCKED(ubd_devs.lock), \
188 /* Protected by ubd_lock */
189 static struct ubd ubd_devs[MAX_DEV] = { [0 ... MAX_DEV - 1] = DEFAULT_UBD };
191 /* Only changed by fake_ide_setup which is a setup */
192 static int fake_ide = 0;
193 static struct proc_dir_entry *proc_ide_root = NULL;
194 static struct proc_dir_entry *proc_ide = NULL;
196 static blk_status_t ubd_queue_rq(struct blk_mq_hw_ctx *hctx,
197 const struct blk_mq_queue_data *bd);
199 static void make_proc_ide(void)
201 proc_ide_root = proc_mkdir("ide", NULL);
202 proc_ide = proc_mkdir("ide0", proc_ide_root);
205 static int fake_ide_media_proc_show(struct seq_file *m, void *v)
207 seq_puts(m, "disk\n");
211 static void make_ide_entries(const char *dev_name)
213 struct proc_dir_entry *dir, *ent;
216 if(proc_ide_root == NULL) make_proc_ide();
218 dir = proc_mkdir(dev_name, proc_ide);
221 ent = proc_create_single("media", S_IRUGO, dir,
222 fake_ide_media_proc_show);
224 snprintf(name, sizeof(name), "ide0/%s", dev_name);
225 proc_symlink(dev_name, proc_ide_root, name);
228 static int fake_ide_setup(char *str)
234 __setup("fake_ide", fake_ide_setup);
236 __uml_help(fake_ide_setup,
238 " Create ide0 entries that map onto ubd devices.\n\n"
241 static int parse_unit(char **ptr)
243 char *str = *ptr, *end;
247 n = simple_strtoul(str, &end, 0);
252 else if (('a' <= *str) && (*str <= 'z')) {
260 /* If *index_out == -1 at exit, the passed option was a general one;
261 * otherwise, the str pointer is used (and owned) inside ubd_devs array, so it
262 * should not be freed on exit.
264 static int ubd_setup_common(char *str, int *index_out, char **error_out)
267 struct openflags flags = global_openflags;
271 if(index_out) *index_out = -1;
278 if(!strcmp(str, "sync")){
279 global_openflags = of_sync(global_openflags);
284 major = simple_strtoul(str, &end, 0);
285 if((*end != '\0') || (end == str)){
286 *error_out = "Didn't parse major number";
290 mutex_lock(&ubd_lock);
291 if (fake_major != UBD_MAJOR) {
292 *error_out = "Can't assign a fake major twice";
298 printk(KERN_INFO "Setting extra ubd major number to %d\n",
302 mutex_unlock(&ubd_lock);
306 n = parse_unit(&str);
308 *error_out = "Couldn't parse device number";
312 *error_out = "Device number out of range";
317 mutex_lock(&ubd_lock);
319 ubd_dev = &ubd_devs[n];
320 if(ubd_dev->file != NULL){
321 *error_out = "Device is already configured";
329 for (i = 0; i < sizeof("rscdt="); i++) {
344 ubd_dev->no_trim = 1;
350 *error_out = "Expected '=' or flag letter "
358 *error_out = "Too many flags specified";
360 *error_out = "Missing '='";
364 backing_file = strchr(str, ',');
366 if (backing_file == NULL)
367 backing_file = strchr(str, ':');
369 if(backing_file != NULL){
371 *error_out = "Can't specify both 'd' and a cow file";
375 *backing_file = '\0';
381 ubd_dev->cow.file = backing_file;
382 ubd_dev->boot_openflags = flags;
384 mutex_unlock(&ubd_lock);
388 static int ubd_setup(char *str)
393 err = ubd_setup_common(str, NULL, &error);
395 printk(KERN_ERR "Failed to initialize device with \"%s\" : "
400 __setup("ubd", ubd_setup);
401 __uml_help(ubd_setup,
402 "ubd<n><flags>=<filename>[(:|,)<filename2>]\n"
403 " This is used to associate a device with a file in the underlying\n"
404 " filesystem. When specifying two filenames, the first one is the\n"
405 " COW name and the second is the backing file name. As separator you can\n"
406 " use either a ':' or a ',': the first one allows writing things like;\n"
407 " ubd0=~/Uml/root_cow:~/Uml/root_backing_file\n"
408 " while with a ',' the shell would not expand the 2nd '~'.\n"
409 " When using only one filename, UML will detect whether to treat it like\n"
410 " a COW file or a backing file. To override this detection, add the 'd'\n"
412 " ubd0d=BackingFile\n"
413 " Usually, there is a filesystem in the file, but \n"
414 " that's not required. Swap devices containing swap files can be\n"
415 " specified like this. Also, a file which doesn't contain a\n"
416 " filesystem can have its contents read in the virtual \n"
417 " machine by running 'dd' on the device. <n> must be in the range\n"
418 " 0 to 7. Appending an 'r' to the number will cause that device\n"
419 " to be mounted read-only. For example ubd1r=./ext_fs. Appending\n"
420 " an 's' will cause data to be written to disk on the host immediately.\n"
421 " 'c' will cause the device to be treated as being shared between multiple\n"
422 " UMLs and file locking will be turned off - this is appropriate for a\n"
423 " cluster filesystem and inappropriate at almost all other times.\n\n"
424 " 't' will disable trim/discard support on the device (enabled by default).\n\n"
427 static int udb_setup(char *str)
429 printk("udb%s specified on command line is almost certainly a ubd -> "
434 __setup("udb", udb_setup);
435 __uml_help(udb_setup,
437 " This option is here solely to catch ubd -> udb typos, which can be\n"
438 " to impossible to catch visually unless you specifically look for\n"
439 " them. The only result of any option starting with 'udb' is an error\n"
440 " in the boot output.\n\n"
443 /* Only changed by ubd_init, which is an initcall. */
444 static int thread_fd = -1;
446 /* Function to read several request pointers at a time
447 * handling fractional reads if (and as) needed
450 static int bulk_req_safe_read(
452 struct io_thread_req * (*request_buffer)[],
453 struct io_thread_req **remainder,
461 if (*remainder_size > 0) {
463 (char *) request_buffer,
464 (char *) remainder, *remainder_size
471 ((char *) request_buffer) + *remainder_size,
472 sizeof(struct io_thread_req *)*max_recs
477 if ((n % sizeof(struct io_thread_req *)) > 0) {
479 * Read somehow returned not a multiple of dword
480 * theoretically possible, but never observed in the
481 * wild, so read routine must be able to handle it
483 *remainder_size = n % sizeof(struct io_thread_req *);
484 WARN(*remainder_size > 0, "UBD IPC read returned a partial result");
487 ((char *) request_buffer) +
488 (n/sizeof(struct io_thread_req *))*sizeof(struct io_thread_req *),
491 n = n - *remainder_size;
499 /* Called without dev->lock held, and only in interrupt context. */
500 static void ubd_handler(void)
506 n = bulk_req_safe_read(
516 printk(KERN_ERR "spurious interrupt in ubd_handler, "
520 for (count = 0; count < n/sizeof(struct io_thread_req *); count++) {
521 struct io_thread_req *io_req = (*irq_req_buffer)[count];
523 if ((io_req->error == BLK_STS_NOTSUPP) && (req_op(io_req->req) == REQ_OP_DISCARD)) {
524 blk_queue_max_discard_sectors(io_req->req->q, 0);
525 blk_queue_max_write_zeroes_sectors(io_req->req->q, 0);
526 blk_queue_flag_clear(QUEUE_FLAG_DISCARD, io_req->req->q);
528 if ((io_req->error) || (io_req->buffer == NULL))
529 blk_mq_end_request(io_req->req, io_req->error);
531 if (!blk_update_request(io_req->req, io_req->error, io_req->length))
532 __blk_mq_end_request(io_req->req, io_req->error);
539 static irqreturn_t ubd_intr(int irq, void *dev)
545 /* Only changed by ubd_init, which is an initcall. */
546 static int io_pid = -1;
548 static void kill_io_thread(void)
551 os_kill_process(io_pid, 1);
554 __uml_exitcall(kill_io_thread);
556 static inline int ubd_file_size(struct ubd *ubd_dev, __u64 *size_out)
566 unsigned long long size;
570 if (ubd_dev->file && ubd_dev->cow.file) {
571 file = ubd_dev->cow.file;
576 fd = os_open_file(ubd_dev->file, of_read(OPENFLAGS()), 0);
580 err = read_cow_header(file_reader, &fd, &version, &backing_file, \
581 &mtime, &size, §or_size, &align, &bitmap_offset);
585 file = ubd_dev->file;
590 return os_file_size(file, size_out);
593 static int read_cow_bitmap(int fd, void *buf, int offset, int len)
597 err = os_pread_file(fd, buf, len, offset);
604 static int backing_file_mismatch(char *file, __u64 size, time64_t mtime)
607 unsigned long long actual;
610 err = os_file_modtime(file, &modtime);
612 printk(KERN_ERR "Failed to get modification time of backing "
613 "file \"%s\", err = %d\n", file, -err);
617 err = os_file_size(file, &actual);
619 printk(KERN_ERR "Failed to get size of backing file \"%s\", "
620 "err = %d\n", file, -err);
624 if (actual != size) {
625 /*__u64 can be a long on AMD64 and with %lu GCC complains; so
627 printk(KERN_ERR "Size mismatch (%llu vs %llu) of COW header "
628 "vs backing file\n", (unsigned long long) size, actual);
631 if (modtime != mtime) {
632 printk(KERN_ERR "mtime mismatch (%lld vs %lld) of COW header vs "
633 "backing file\n", mtime, modtime);
639 static int path_requires_switch(char *from_cmdline, char *from_cow, char *cow)
641 struct uml_stat buf1, buf2;
644 if (from_cmdline == NULL)
646 if (!strcmp(from_cmdline, from_cow))
649 err = os_stat_file(from_cmdline, &buf1);
651 printk(KERN_ERR "Couldn't stat '%s', err = %d\n", from_cmdline,
655 err = os_stat_file(from_cow, &buf2);
657 printk(KERN_ERR "Couldn't stat '%s', err = %d\n", from_cow,
661 if ((buf1.ust_dev == buf2.ust_dev) && (buf1.ust_ino == buf2.ust_ino))
664 printk(KERN_ERR "Backing file mismatch - \"%s\" requested, "
665 "\"%s\" specified in COW header of \"%s\"\n",
666 from_cmdline, from_cow, cow);
670 static int open_ubd_file(char *file, struct openflags *openflags, int shared,
671 char **backing_file_out, int *bitmap_offset_out,
672 unsigned long *bitmap_len_out, int *data_offset_out,
676 unsigned long long size;
677 __u32 version, align;
679 int fd, err, sectorsize, asked_switch, mode = 0644;
681 fd = os_open_file(file, *openflags, mode);
683 if ((fd == -ENOENT) && (create_cow_out != NULL))
686 ((fd != -EROFS) && (fd != -EACCES)))
689 fd = os_open_file(file, *openflags, mode);
695 printk(KERN_INFO "Not locking \"%s\" on the host\n", file);
697 err = os_lock_file(fd, openflags->w);
699 printk(KERN_ERR "Failed to lock '%s', err = %d\n",
705 /* Successful return case! */
706 if (backing_file_out == NULL)
709 err = read_cow_header(file_reader, &fd, &version, &backing_file, &mtime,
710 &size, §orsize, &align, bitmap_offset_out);
711 if (err && (*backing_file_out != NULL)) {
712 printk(KERN_ERR "Failed to read COW header from COW file "
713 "\"%s\", errno = %d\n", file, -err);
719 asked_switch = path_requires_switch(*backing_file_out, backing_file,
722 /* Allow switching only if no mismatch. */
723 if (asked_switch && !backing_file_mismatch(*backing_file_out, size,
725 printk(KERN_ERR "Switching backing file to '%s'\n",
727 err = write_cow_header(file, fd, *backing_file_out,
728 sectorsize, align, &size);
730 printk(KERN_ERR "Switch failed, errno = %d\n", -err);
734 *backing_file_out = backing_file;
735 err = backing_file_mismatch(*backing_file_out, size, mtime);
740 cow_sizes(version, size, sectorsize, align, *bitmap_offset_out,
741 bitmap_len_out, data_offset_out);
749 static int create_cow_file(char *cow_file, char *backing_file,
750 struct openflags flags,
751 int sectorsize, int alignment, int *bitmap_offset_out,
752 unsigned long *bitmap_len_out, int *data_offset_out)
757 fd = open_ubd_file(cow_file, &flags, 0, NULL, NULL, NULL, NULL, NULL);
760 printk(KERN_ERR "Open of COW file '%s' failed, errno = %d\n",
765 err = init_cow_file(fd, cow_file, backing_file, sectorsize, alignment,
766 bitmap_offset_out, bitmap_len_out,
775 static void ubd_close_dev(struct ubd *ubd_dev)
777 os_close_file(ubd_dev->fd);
778 if(ubd_dev->cow.file == NULL)
781 os_close_file(ubd_dev->cow.fd);
782 vfree(ubd_dev->cow.bitmap);
783 ubd_dev->cow.bitmap = NULL;
786 static int ubd_open_dev(struct ubd *ubd_dev)
788 struct openflags flags;
790 int err, create_cow, *create_ptr;
793 ubd_dev->openflags = ubd_dev->boot_openflags;
795 create_ptr = (ubd_dev->cow.file != NULL) ? &create_cow : NULL;
796 back_ptr = ubd_dev->no_cow ? NULL : &ubd_dev->cow.file;
798 fd = open_ubd_file(ubd_dev->file, &ubd_dev->openflags, ubd_dev->shared,
799 back_ptr, &ubd_dev->cow.bitmap_offset,
800 &ubd_dev->cow.bitmap_len, &ubd_dev->cow.data_offset,
803 if((fd == -ENOENT) && create_cow){
804 fd = create_cow_file(ubd_dev->file, ubd_dev->cow.file,
805 ubd_dev->openflags, SECTOR_SIZE, PAGE_SIZE,
806 &ubd_dev->cow.bitmap_offset,
807 &ubd_dev->cow.bitmap_len,
808 &ubd_dev->cow.data_offset);
810 printk(KERN_INFO "Creating \"%s\" as COW file for "
811 "\"%s\"\n", ubd_dev->file, ubd_dev->cow.file);
816 printk("Failed to open '%s', errno = %d\n", ubd_dev->file,
822 if(ubd_dev->cow.file != NULL){
823 blk_queue_max_hw_sectors(ubd_dev->queue, 8 * sizeof(long));
826 ubd_dev->cow.bitmap = vmalloc(ubd_dev->cow.bitmap_len);
827 if(ubd_dev->cow.bitmap == NULL){
828 printk(KERN_ERR "Failed to vmalloc COW bitmap\n");
831 flush_tlb_kernel_vm();
833 err = read_cow_bitmap(ubd_dev->fd, ubd_dev->cow.bitmap,
834 ubd_dev->cow.bitmap_offset,
835 ubd_dev->cow.bitmap_len);
839 flags = ubd_dev->openflags;
841 err = open_ubd_file(ubd_dev->cow.file, &flags, ubd_dev->shared, NULL,
842 NULL, NULL, NULL, NULL);
843 if(err < 0) goto error;
844 ubd_dev->cow.fd = err;
846 if (ubd_dev->no_trim == 0) {
847 ubd_dev->queue->limits.discard_granularity = SECTOR_SIZE;
848 ubd_dev->queue->limits.discard_alignment = SECTOR_SIZE;
849 blk_queue_max_discard_sectors(ubd_dev->queue, UBD_MAX_REQUEST);
850 blk_queue_max_write_zeroes_sectors(ubd_dev->queue, UBD_MAX_REQUEST);
851 blk_queue_flag_set(QUEUE_FLAG_DISCARD, ubd_dev->queue);
853 blk_queue_flag_set(QUEUE_FLAG_NONROT, ubd_dev->queue);
856 os_close_file(ubd_dev->fd);
860 static void ubd_device_release(struct device *dev)
862 struct ubd *ubd_dev = dev_get_drvdata(dev);
864 blk_cleanup_queue(ubd_dev->queue);
865 blk_mq_free_tag_set(&ubd_dev->tag_set);
866 *ubd_dev = ((struct ubd) DEFAULT_UBD);
869 static int ubd_disk_register(int major, u64 size, int unit,
870 struct gendisk **disk_out)
872 struct device *parent = NULL;
873 struct gendisk *disk;
875 disk = alloc_disk(1 << UBD_SHIFT);
880 disk->first_minor = unit << UBD_SHIFT;
881 disk->fops = &ubd_blops;
882 set_capacity(disk, size / 512);
883 if (major == UBD_MAJOR)
884 sprintf(disk->disk_name, "ubd%c", 'a' + unit);
886 sprintf(disk->disk_name, "ubd_fake%d", unit);
888 /* sysfs register (not for ide fake devices) */
889 if (major == UBD_MAJOR) {
890 ubd_devs[unit].pdev.id = unit;
891 ubd_devs[unit].pdev.name = DRIVER_NAME;
892 ubd_devs[unit].pdev.dev.release = ubd_device_release;
893 dev_set_drvdata(&ubd_devs[unit].pdev.dev, &ubd_devs[unit]);
894 platform_device_register(&ubd_devs[unit].pdev);
895 parent = &ubd_devs[unit].pdev.dev;
898 disk->private_data = &ubd_devs[unit];
899 disk->queue = ubd_devs[unit].queue;
900 device_add_disk(parent, disk, NULL);
906 #define ROUND_BLOCK(n) ((n + (SECTOR_SIZE - 1)) & (-SECTOR_SIZE))
908 static const struct blk_mq_ops ubd_mq_ops = {
909 .queue_rq = ubd_queue_rq,
912 static int ubd_add(int n, char **error_out)
914 struct ubd *ubd_dev = &ubd_devs[n];
917 if(ubd_dev->file == NULL)
920 err = ubd_file_size(ubd_dev, &ubd_dev->size);
922 *error_out = "Couldn't determine size of device's file";
926 ubd_dev->size = ROUND_BLOCK(ubd_dev->size);
928 ubd_dev->tag_set.ops = &ubd_mq_ops;
929 ubd_dev->tag_set.queue_depth = 64;
930 ubd_dev->tag_set.numa_node = NUMA_NO_NODE;
931 ubd_dev->tag_set.flags = BLK_MQ_F_SHOULD_MERGE;
932 ubd_dev->tag_set.driver_data = ubd_dev;
933 ubd_dev->tag_set.nr_hw_queues = 1;
935 err = blk_mq_alloc_tag_set(&ubd_dev->tag_set);
939 ubd_dev->queue = blk_mq_init_queue(&ubd_dev->tag_set);
940 if (IS_ERR(ubd_dev->queue)) {
941 err = PTR_ERR(ubd_dev->queue);
942 goto out_cleanup_tags;
945 ubd_dev->queue->queuedata = ubd_dev;
946 blk_queue_write_cache(ubd_dev->queue, true, false);
948 blk_queue_max_segments(ubd_dev->queue, MAX_SG);
949 err = ubd_disk_register(UBD_MAJOR, ubd_dev->size, n, &ubd_gendisk[n]);
951 *error_out = "Failed to register device";
952 goto out_cleanup_tags;
955 if (fake_major != UBD_MAJOR)
956 ubd_disk_register(fake_major, ubd_dev->size, n,
960 * Perhaps this should also be under the "if (fake_major)" above
961 * using the fake_disk->disk_name
964 make_ide_entries(ubd_gendisk[n]->disk_name);
971 blk_mq_free_tag_set(&ubd_dev->tag_set);
972 if (!(IS_ERR(ubd_dev->queue)))
973 blk_cleanup_queue(ubd_dev->queue);
977 static int ubd_config(char *str, char **error_out)
981 /* This string is possibly broken up and stored, so it's only
982 * freed if ubd_setup_common fails, or if only general options
985 str = kstrdup(str, GFP_KERNEL);
987 *error_out = "Failed to allocate memory";
991 ret = ubd_setup_common(str, &n, error_out);
1000 mutex_lock(&ubd_lock);
1001 ret = ubd_add(n, error_out);
1003 ubd_devs[n].file = NULL;
1004 mutex_unlock(&ubd_lock);
1014 static int ubd_get_config(char *name, char *str, int size, char **error_out)
1016 struct ubd *ubd_dev;
1019 n = parse_unit(&name);
1020 if((n >= MAX_DEV) || (n < 0)){
1021 *error_out = "ubd_get_config : device number out of range";
1025 ubd_dev = &ubd_devs[n];
1026 mutex_lock(&ubd_lock);
1028 if(ubd_dev->file == NULL){
1029 CONFIG_CHUNK(str, size, len, "", 1);
1033 CONFIG_CHUNK(str, size, len, ubd_dev->file, 0);
1035 if(ubd_dev->cow.file != NULL){
1036 CONFIG_CHUNK(str, size, len, ",", 0);
1037 CONFIG_CHUNK(str, size, len, ubd_dev->cow.file, 1);
1039 else CONFIG_CHUNK(str, size, len, "", 1);
1042 mutex_unlock(&ubd_lock);
1046 static int ubd_id(char **str, int *start_out, int *end_out)
1050 n = parse_unit(str);
1052 *end_out = MAX_DEV - 1;
1056 static int ubd_remove(int n, char **error_out)
1058 struct gendisk *disk = ubd_gendisk[n];
1059 struct ubd *ubd_dev;
1062 mutex_lock(&ubd_lock);
1064 ubd_dev = &ubd_devs[n];
1066 if(ubd_dev->file == NULL)
1069 /* you cannot remove a open disk */
1071 if(ubd_dev->count > 0)
1074 ubd_gendisk[n] = NULL;
1080 if(fake_gendisk[n] != NULL){
1081 del_gendisk(fake_gendisk[n]);
1082 put_disk(fake_gendisk[n]);
1083 fake_gendisk[n] = NULL;
1087 platform_device_unregister(&ubd_dev->pdev);
1089 mutex_unlock(&ubd_lock);
1093 /* All these are called by mconsole in process context and without
1094 * ubd-specific locks. The structure itself is const except for .list.
1096 static struct mc_device ubd_mc = {
1097 .list = LIST_HEAD_INIT(ubd_mc.list),
1099 .config = ubd_config,
1100 .get_config = ubd_get_config,
1102 .remove = ubd_remove,
1105 static int __init ubd_mc_init(void)
1107 mconsole_register_dev(&ubd_mc);
1111 __initcall(ubd_mc_init);
1113 static int __init ubd0_init(void)
1115 struct ubd *ubd_dev = &ubd_devs[0];
1117 mutex_lock(&ubd_lock);
1118 if(ubd_dev->file == NULL)
1119 ubd_dev->file = "root_fs";
1120 mutex_unlock(&ubd_lock);
1125 __initcall(ubd0_init);
1127 /* Used in ubd_init, which is an initcall */
1128 static struct platform_driver ubd_driver = {
1130 .name = DRIVER_NAME,
1134 static int __init ubd_init(void)
1139 if (register_blkdev(UBD_MAJOR, "ubd"))
1142 if (fake_major != UBD_MAJOR) {
1143 char name[sizeof("ubd_nnn\0")];
1145 snprintf(name, sizeof(name), "ubd_%d", fake_major);
1146 if (register_blkdev(fake_major, "ubd"))
1150 irq_req_buffer = kmalloc_array(UBD_REQ_BUFFER_SIZE,
1151 sizeof(struct io_thread_req *),
1156 if (irq_req_buffer == NULL) {
1157 printk(KERN_ERR "Failed to initialize ubd buffering\n");
1160 io_req_buffer = kmalloc_array(UBD_REQ_BUFFER_SIZE,
1161 sizeof(struct io_thread_req *),
1167 if (io_req_buffer == NULL) {
1168 printk(KERN_ERR "Failed to initialize ubd buffering\n");
1171 platform_driver_register(&ubd_driver);
1172 mutex_lock(&ubd_lock);
1173 for (i = 0; i < MAX_DEV; i++){
1174 err = ubd_add(i, &error);
1176 printk(KERN_ERR "Failed to initialize ubd device %d :"
1179 mutex_unlock(&ubd_lock);
1183 late_initcall(ubd_init);
1185 static int __init ubd_driver_init(void){
1186 unsigned long stack;
1189 /* Set by CONFIG_BLK_DEV_UBD_SYNC or ubd=sync.*/
1190 if(global_openflags.s){
1191 printk(KERN_INFO "ubd: Synchronous mode\n");
1192 /* Letting ubd=sync be like using ubd#s= instead of ubd#= is
1193 * enough. So use anyway the io thread. */
1195 stack = alloc_stack(0, 0);
1196 io_pid = start_io_thread(stack + PAGE_SIZE - sizeof(void *),
1200 "ubd : Failed to start I/O thread (errno = %d) - "
1201 "falling back to synchronous I/O\n", -io_pid);
1205 err = um_request_irq(UBD_IRQ, thread_fd, IRQ_READ, ubd_intr,
1206 0, "ubd", ubd_devs);
1208 printk(KERN_ERR "um_request_irq failed - errno = %d\n", -err);
1212 device_initcall(ubd_driver_init);
1214 static int ubd_open(struct block_device *bdev, fmode_t mode)
1216 struct gendisk *disk = bdev->bd_disk;
1217 struct ubd *ubd_dev = disk->private_data;
1220 mutex_lock(&ubd_mutex);
1221 if(ubd_dev->count == 0){
1222 err = ubd_open_dev(ubd_dev);
1224 printk(KERN_ERR "%s: Can't open \"%s\": errno = %d\n",
1225 disk->disk_name, ubd_dev->file, -err);
1230 set_disk_ro(disk, !ubd_dev->openflags.w);
1232 /* This should no more be needed. And it didn't work anyway to exclude
1233 * read-write remounting of filesystems.*/
1234 /*if((mode & FMODE_WRITE) && !ubd_dev->openflags.w){
1235 if(--ubd_dev->count == 0) ubd_close_dev(ubd_dev);
1239 mutex_unlock(&ubd_mutex);
1243 static void ubd_release(struct gendisk *disk, fmode_t mode)
1245 struct ubd *ubd_dev = disk->private_data;
1247 mutex_lock(&ubd_mutex);
1248 if(--ubd_dev->count == 0)
1249 ubd_close_dev(ubd_dev);
1250 mutex_unlock(&ubd_mutex);
1253 static void cowify_bitmap(__u64 io_offset, int length, unsigned long *cow_mask,
1254 __u64 *cow_offset, unsigned long *bitmap,
1255 __u64 bitmap_offset, unsigned long *bitmap_words,
1258 __u64 sector = io_offset >> SECTOR_SHIFT;
1259 int i, update_bitmap = 0;
1261 for (i = 0; i < length >> SECTOR_SHIFT; i++) {
1262 if(cow_mask != NULL)
1263 ubd_set_bit(i, (unsigned char *) cow_mask);
1264 if(ubd_test_bit(sector + i, (unsigned char *) bitmap))
1268 ubd_set_bit(sector + i, (unsigned char *) bitmap);
1274 *cow_offset = sector / (sizeof(unsigned long) * 8);
1276 /* This takes care of the case where we're exactly at the end of the
1277 * device, and *cow_offset + 1 is off the end. So, just back it up
1278 * by one word. Thanks to Lynn Kerby for the fix and James McMechan
1279 * for the original diagnosis.
1281 if (*cow_offset == (DIV_ROUND_UP(bitmap_len,
1282 sizeof(unsigned long)) - 1))
1285 bitmap_words[0] = bitmap[*cow_offset];
1286 bitmap_words[1] = bitmap[*cow_offset + 1];
1288 *cow_offset *= sizeof(unsigned long);
1289 *cow_offset += bitmap_offset;
1292 static void cowify_req(struct io_thread_req *req, unsigned long *bitmap,
1293 __u64 bitmap_offset, __u64 bitmap_len)
1295 __u64 sector = req->offset >> SECTOR_SHIFT;
1298 if (req->length > (sizeof(req->sector_mask) * 8) << SECTOR_SHIFT)
1299 panic("Operation too long");
1301 if (req_op(req->req) == REQ_OP_READ) {
1302 for (i = 0; i < req->length >> SECTOR_SHIFT; i++) {
1303 if(ubd_test_bit(sector + i, (unsigned char *) bitmap))
1304 ubd_set_bit(i, (unsigned char *)
1308 else cowify_bitmap(req->offset, req->length, &req->sector_mask,
1309 &req->cow_offset, bitmap, bitmap_offset,
1310 req->bitmap_words, bitmap_len);
1313 static int ubd_queue_one_vec(struct blk_mq_hw_ctx *hctx, struct request *req,
1314 u64 off, struct bio_vec *bvec)
1316 struct ubd *dev = hctx->queue->queuedata;
1317 struct io_thread_req *io_req;
1320 io_req = kmalloc(sizeof(struct io_thread_req), GFP_ATOMIC);
1326 io_req->fds[0] = dev->cow.fd;
1328 io_req->fds[0] = dev->fd;
1332 io_req->buffer = page_address(bvec->bv_page) + bvec->bv_offset;
1333 io_req->length = bvec->bv_len;
1335 io_req->buffer = NULL;
1336 io_req->length = blk_rq_bytes(req);
1339 io_req->sectorsize = SECTOR_SIZE;
1340 io_req->fds[1] = dev->fd;
1341 io_req->cow_offset = -1;
1342 io_req->offset = off;
1343 io_req->sector_mask = 0;
1344 io_req->offsets[0] = 0;
1345 io_req->offsets[1] = dev->cow.data_offset;
1348 cowify_req(io_req, dev->cow.bitmap,
1349 dev->cow.bitmap_offset, dev->cow.bitmap_len);
1351 ret = os_write_file(thread_fd, &io_req, sizeof(io_req));
1352 if (ret != sizeof(io_req)) {
1354 pr_err("write to io thread failed: %d\n", -ret);
1360 static int queue_rw_req(struct blk_mq_hw_ctx *hctx, struct request *req)
1362 struct req_iterator iter;
1363 struct bio_vec bvec;
1365 u64 off = (u64)blk_rq_pos(req) << SECTOR_SHIFT;
1367 rq_for_each_segment(bvec, req, iter) {
1368 ret = ubd_queue_one_vec(hctx, req, off, &bvec);
1376 static blk_status_t ubd_queue_rq(struct blk_mq_hw_ctx *hctx,
1377 const struct blk_mq_queue_data *bd)
1379 struct ubd *ubd_dev = hctx->queue->queuedata;
1380 struct request *req = bd->rq;
1381 int ret = 0, res = BLK_STS_OK;
1383 blk_mq_start_request(req);
1385 spin_lock_irq(&ubd_dev->lock);
1387 switch (req_op(req)) {
1388 /* operations with no lentgth/offset arguments */
1390 ret = ubd_queue_one_vec(hctx, req, 0, NULL);
1394 ret = queue_rw_req(hctx, req);
1396 case REQ_OP_DISCARD:
1397 case REQ_OP_WRITE_ZEROES:
1398 ret = ubd_queue_one_vec(hctx, req, (u64)blk_rq_pos(req) << 9, NULL);
1402 res = BLK_STS_NOTSUPP;
1405 spin_unlock_irq(&ubd_dev->lock);
1409 res = BLK_STS_RESOURCE;
1411 res = BLK_STS_DEV_RESOURCE;
1417 static int ubd_getgeo(struct block_device *bdev, struct hd_geometry *geo)
1419 struct ubd *ubd_dev = bdev->bd_disk->private_data;
1423 geo->cylinders = ubd_dev->size / (128 * 32 * 512);
1427 static int ubd_ioctl(struct block_device *bdev, fmode_t mode,
1428 unsigned int cmd, unsigned long arg)
1430 struct ubd *ubd_dev = bdev->bd_disk->private_data;
1431 u16 ubd_id[ATA_ID_WORDS];
1434 struct cdrom_volctrl volume;
1435 case HDIO_GET_IDENTITY:
1436 memset(&ubd_id, 0, ATA_ID_WORDS * 2);
1437 ubd_id[ATA_ID_CYLS] = ubd_dev->size / (128 * 32 * 512);
1438 ubd_id[ATA_ID_HEADS] = 128;
1439 ubd_id[ATA_ID_SECTORS] = 32;
1440 if(copy_to_user((char __user *) arg, (char *) &ubd_id,
1446 if(copy_from_user(&volume, (char __user *) arg, sizeof(volume)))
1448 volume.channel0 = 255;
1449 volume.channel1 = 255;
1450 volume.channel2 = 255;
1451 volume.channel3 = 255;
1452 if(copy_to_user((char __user *) arg, &volume, sizeof(volume)))
1459 static int map_error(int error_code)
1461 switch (error_code) {
1466 return BLK_STS_NOTSUPP;
1468 return BLK_STS_NOSPC;
1470 return BLK_STS_IOERR;
1474 * Everything from here onwards *IS NOT PART OF THE KERNEL*
1476 * The following functions are part of UML hypervisor code.
1477 * All functions from here onwards are executed as a helper
1478 * thread and are not allowed to execute any kernel functions.
1480 * Any communication must occur strictly via shared memory and IPC.
1482 * Do not add printks, locks, kernel memory operations, etc - it
1483 * will result in unpredictable behaviour and/or crashes.
1486 static int update_bitmap(struct io_thread_req *req)
1490 if(req->cow_offset == -1)
1491 return map_error(0);
1493 n = os_pwrite_file(req->fds[1], &req->bitmap_words,
1494 sizeof(req->bitmap_words), req->cow_offset);
1495 if (n != sizeof(req->bitmap_words))
1496 return map_error(-n);
1498 return map_error(0);
1501 static void do_io(struct io_thread_req *req)
1505 int n, nsectors, start, end, bit;
1508 /* FLUSH is really a special case, we cannot "case" it with others */
1510 if (req_op(req->req) == REQ_OP_FLUSH) {
1511 /* fds[0] is always either the rw image or our cow file */
1512 req->error = map_error(-os_sync_file(req->fds[0]));
1516 nsectors = req->length / req->sectorsize;
1519 bit = ubd_test_bit(start, (unsigned char *) &req->sector_mask);
1521 while((end < nsectors) &&
1522 (ubd_test_bit(end, (unsigned char *)
1523 &req->sector_mask) == bit))
1526 off = req->offset + req->offsets[bit] +
1527 start * req->sectorsize;
1528 len = (end - start) * req->sectorsize;
1529 if (req->buffer != NULL)
1530 buf = &req->buffer[start * req->sectorsize];
1532 switch (req_op(req->req)) {
1538 n = os_pread_file(req->fds[bit], buf, len, off);
1540 req->error = map_error(-n);
1543 } while((n < len) && (n != 0));
1544 if (n < len) memset(&buf[n], 0, len - n);
1547 n = os_pwrite_file(req->fds[bit], buf, len, off);
1549 req->error = map_error(-n);
1553 case REQ_OP_DISCARD:
1554 case REQ_OP_WRITE_ZEROES:
1555 n = os_falloc_punch(req->fds[bit], off, len);
1557 req->error = map_error(-n);
1563 req->error = BLK_STS_NOTSUPP;
1568 } while(start < nsectors);
1570 req->error = update_bitmap(req);
1573 /* Changed in start_io_thread, which is serialized by being called only
1574 * from ubd_init, which is an initcall.
1578 /* Only changed by the io thread. XXX: currently unused. */
1579 static int io_count = 0;
1581 int io_thread(void *arg)
1583 int n, count, written, res;
1585 os_fix_helper_signals();
1588 n = bulk_req_safe_read(
1602 for (count = 0; count < n/sizeof(struct io_thread_req *); count++) {
1604 do_io((*io_req_buffer)[count]);
1610 res = os_write_file(kernel_fd,
1611 ((char *) io_req_buffer) + written,
1619 } while (written < n);