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 .getgeo = ubd_getgeo,
119 /* Protected by ubd_lock */
120 static int fake_major = UBD_MAJOR;
121 static struct gendisk *ubd_gendisk[MAX_DEV];
122 static struct gendisk *fake_gendisk[MAX_DEV];
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 request_queue *queue;
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 /* Only changed by fake_ide_setup which is a setup */
191 static int fake_ide = 0;
192 static struct proc_dir_entry *proc_ide_root = NULL;
193 static struct proc_dir_entry *proc_ide = NULL;
195 static blk_status_t ubd_queue_rq(struct blk_mq_hw_ctx *hctx,
196 const struct blk_mq_queue_data *bd);
198 static void make_proc_ide(void)
200 proc_ide_root = proc_mkdir("ide", NULL);
201 proc_ide = proc_mkdir("ide0", proc_ide_root);
204 static int fake_ide_media_proc_show(struct seq_file *m, void *v)
206 seq_puts(m, "disk\n");
210 static void make_ide_entries(const char *dev_name)
212 struct proc_dir_entry *dir, *ent;
215 if(proc_ide_root == NULL) make_proc_ide();
217 dir = proc_mkdir(dev_name, proc_ide);
220 ent = proc_create_single("media", S_IRUGO, dir,
221 fake_ide_media_proc_show);
223 snprintf(name, sizeof(name), "ide0/%s", dev_name);
224 proc_symlink(dev_name, proc_ide_root, name);
227 static int fake_ide_setup(char *str)
233 __setup("fake_ide", fake_ide_setup);
235 __uml_help(fake_ide_setup,
237 " Create ide0 entries that map onto ubd devices.\n\n"
240 static int parse_unit(char **ptr)
242 char *str = *ptr, *end;
246 n = simple_strtoul(str, &end, 0);
251 else if (('a' <= *str) && (*str <= 'z')) {
259 /* If *index_out == -1 at exit, the passed option was a general one;
260 * otherwise, the str pointer is used (and owned) inside ubd_devs array, so it
261 * should not be freed on exit.
263 static int ubd_setup_common(char *str, int *index_out, char **error_out)
266 struct openflags flags = global_openflags;
270 if(index_out) *index_out = -1;
277 if(!strcmp(str, "sync")){
278 global_openflags = of_sync(global_openflags);
283 major = simple_strtoul(str, &end, 0);
284 if((*end != '\0') || (end == str)){
285 *error_out = "Didn't parse major number";
289 mutex_lock(&ubd_lock);
290 if (fake_major != UBD_MAJOR) {
291 *error_out = "Can't assign a fake major twice";
297 printk(KERN_INFO "Setting extra ubd major number to %d\n",
301 mutex_unlock(&ubd_lock);
305 n = parse_unit(&str);
307 *error_out = "Couldn't parse device number";
311 *error_out = "Device number out of range";
316 mutex_lock(&ubd_lock);
318 ubd_dev = &ubd_devs[n];
319 if(ubd_dev->file != NULL){
320 *error_out = "Device is already configured";
328 for (i = 0; i < sizeof("rscdt="); i++) {
343 ubd_dev->no_trim = 1;
349 *error_out = "Expected '=' or flag letter "
357 *error_out = "Too many flags specified";
359 *error_out = "Missing '='";
363 backing_file = strchr(str, ',');
365 if (backing_file == NULL)
366 backing_file = strchr(str, ':');
368 if(backing_file != NULL){
370 *error_out = "Can't specify both 'd' and a cow file";
374 *backing_file = '\0';
380 ubd_dev->cow.file = backing_file;
381 ubd_dev->boot_openflags = flags;
383 mutex_unlock(&ubd_lock);
387 static int ubd_setup(char *str)
392 err = ubd_setup_common(str, NULL, &error);
394 printk(KERN_ERR "Failed to initialize device with \"%s\" : "
399 __setup("ubd", ubd_setup);
400 __uml_help(ubd_setup,
401 "ubd<n><flags>=<filename>[(:|,)<filename2>]\n"
402 " This is used to associate a device with a file in the underlying\n"
403 " filesystem. When specifying two filenames, the first one is the\n"
404 " COW name and the second is the backing file name. As separator you can\n"
405 " use either a ':' or a ',': the first one allows writing things like;\n"
406 " ubd0=~/Uml/root_cow:~/Uml/root_backing_file\n"
407 " while with a ',' the shell would not expand the 2nd '~'.\n"
408 " When using only one filename, UML will detect whether to treat it like\n"
409 " a COW file or a backing file. To override this detection, add the 'd'\n"
411 " ubd0d=BackingFile\n"
412 " Usually, there is a filesystem in the file, but \n"
413 " that's not required. Swap devices containing swap files can be\n"
414 " specified like this. Also, a file which doesn't contain a\n"
415 " filesystem can have its contents read in the virtual \n"
416 " machine by running 'dd' on the device. <n> must be in the range\n"
417 " 0 to 7. Appending an 'r' to the number will cause that device\n"
418 " to be mounted read-only. For example ubd1r=./ext_fs. Appending\n"
419 " an 's' will cause data to be written to disk on the host immediately.\n"
420 " 'c' will cause the device to be treated as being shared between multiple\n"
421 " UMLs and file locking will be turned off - this is appropriate for a\n"
422 " cluster filesystem and inappropriate at almost all other times.\n\n"
423 " 't' will disable trim/discard support on the device (enabled by default).\n\n"
426 static int udb_setup(char *str)
428 printk("udb%s specified on command line is almost certainly a ubd -> "
433 __setup("udb", udb_setup);
434 __uml_help(udb_setup,
436 " This option is here solely to catch ubd -> udb typos, which can be\n"
437 " to impossible to catch visually unless you specifically look for\n"
438 " them. The only result of any option starting with 'udb' is an error\n"
439 " in the boot output.\n\n"
442 /* Only changed by ubd_init, which is an initcall. */
443 static int thread_fd = -1;
445 /* Function to read several request pointers at a time
446 * handling fractional reads if (and as) needed
449 static int bulk_req_safe_read(
451 struct io_thread_req * (*request_buffer)[],
452 struct io_thread_req **remainder,
460 if (*remainder_size > 0) {
462 (char *) request_buffer,
463 (char *) remainder, *remainder_size
470 ((char *) request_buffer) + *remainder_size,
471 sizeof(struct io_thread_req *)*max_recs
476 if ((n % sizeof(struct io_thread_req *)) > 0) {
478 * Read somehow returned not a multiple of dword
479 * theoretically possible, but never observed in the
480 * wild, so read routine must be able to handle it
482 *remainder_size = n % sizeof(struct io_thread_req *);
483 WARN(*remainder_size > 0, "UBD IPC read returned a partial result");
486 ((char *) request_buffer) +
487 (n/sizeof(struct io_thread_req *))*sizeof(struct io_thread_req *),
490 n = n - *remainder_size;
498 /* Called without dev->lock held, and only in interrupt context. */
499 static void ubd_handler(void)
505 n = bulk_req_safe_read(
515 printk(KERN_ERR "spurious interrupt in ubd_handler, "
519 for (count = 0; count < n/sizeof(struct io_thread_req *); count++) {
520 struct io_thread_req *io_req = (*irq_req_buffer)[count];
522 if ((io_req->error == BLK_STS_NOTSUPP) && (req_op(io_req->req) == REQ_OP_DISCARD)) {
523 blk_queue_max_discard_sectors(io_req->req->q, 0);
524 blk_queue_max_write_zeroes_sectors(io_req->req->q, 0);
525 blk_queue_flag_clear(QUEUE_FLAG_DISCARD, io_req->req->q);
527 if ((io_req->error) || (io_req->buffer == NULL))
528 blk_mq_end_request(io_req->req, io_req->error);
530 if (!blk_update_request(io_req->req, io_req->error, io_req->length))
531 __blk_mq_end_request(io_req->req, io_req->error);
538 static irqreturn_t ubd_intr(int irq, void *dev)
544 /* Only changed by ubd_init, which is an initcall. */
545 static int io_pid = -1;
547 static void kill_io_thread(void)
550 os_kill_process(io_pid, 1);
553 __uml_exitcall(kill_io_thread);
555 static inline int ubd_file_size(struct ubd *ubd_dev, __u64 *size_out)
565 unsigned long long size;
569 if (ubd_dev->file && ubd_dev->cow.file) {
570 file = ubd_dev->cow.file;
575 fd = os_open_file(ubd_dev->file, of_read(OPENFLAGS()), 0);
579 err = read_cow_header(file_reader, &fd, &version, &backing_file, \
580 &mtime, &size, §or_size, &align, &bitmap_offset);
584 file = ubd_dev->file;
589 return os_file_size(file, size_out);
592 static int read_cow_bitmap(int fd, void *buf, int offset, int len)
596 err = os_pread_file(fd, buf, len, offset);
603 static int backing_file_mismatch(char *file, __u64 size, time64_t mtime)
606 unsigned long long actual;
609 err = os_file_modtime(file, &modtime);
611 printk(KERN_ERR "Failed to get modification time of backing "
612 "file \"%s\", err = %d\n", file, -err);
616 err = os_file_size(file, &actual);
618 printk(KERN_ERR "Failed to get size of backing file \"%s\", "
619 "err = %d\n", file, -err);
623 if (actual != size) {
624 /*__u64 can be a long on AMD64 and with %lu GCC complains; so
626 printk(KERN_ERR "Size mismatch (%llu vs %llu) of COW header "
627 "vs backing file\n", (unsigned long long) size, actual);
630 if (modtime != mtime) {
631 printk(KERN_ERR "mtime mismatch (%lld vs %lld) of COW header vs "
632 "backing file\n", mtime, modtime);
638 static int path_requires_switch(char *from_cmdline, char *from_cow, char *cow)
640 struct uml_stat buf1, buf2;
643 if (from_cmdline == NULL)
645 if (!strcmp(from_cmdline, from_cow))
648 err = os_stat_file(from_cmdline, &buf1);
650 printk(KERN_ERR "Couldn't stat '%s', err = %d\n", from_cmdline,
654 err = os_stat_file(from_cow, &buf2);
656 printk(KERN_ERR "Couldn't stat '%s', err = %d\n", from_cow,
660 if ((buf1.ust_dev == buf2.ust_dev) && (buf1.ust_ino == buf2.ust_ino))
663 printk(KERN_ERR "Backing file mismatch - \"%s\" requested, "
664 "\"%s\" specified in COW header of \"%s\"\n",
665 from_cmdline, from_cow, cow);
669 static int open_ubd_file(char *file, struct openflags *openflags, int shared,
670 char **backing_file_out, int *bitmap_offset_out,
671 unsigned long *bitmap_len_out, int *data_offset_out,
675 unsigned long long size;
676 __u32 version, align;
678 int fd, err, sectorsize, asked_switch, mode = 0644;
680 fd = os_open_file(file, *openflags, mode);
682 if ((fd == -ENOENT) && (create_cow_out != NULL))
685 ((fd != -EROFS) && (fd != -EACCES)))
688 fd = os_open_file(file, *openflags, mode);
694 printk(KERN_INFO "Not locking \"%s\" on the host\n", file);
696 err = os_lock_file(fd, openflags->w);
698 printk(KERN_ERR "Failed to lock '%s', err = %d\n",
704 /* Successful return case! */
705 if (backing_file_out == NULL)
708 err = read_cow_header(file_reader, &fd, &version, &backing_file, &mtime,
709 &size, §orsize, &align, bitmap_offset_out);
710 if (err && (*backing_file_out != NULL)) {
711 printk(KERN_ERR "Failed to read COW header from COW file "
712 "\"%s\", errno = %d\n", file, -err);
718 asked_switch = path_requires_switch(*backing_file_out, backing_file,
721 /* Allow switching only if no mismatch. */
722 if (asked_switch && !backing_file_mismatch(*backing_file_out, size,
724 printk(KERN_ERR "Switching backing file to '%s'\n",
726 err = write_cow_header(file, fd, *backing_file_out,
727 sectorsize, align, &size);
729 printk(KERN_ERR "Switch failed, errno = %d\n", -err);
733 *backing_file_out = backing_file;
734 err = backing_file_mismatch(*backing_file_out, size, mtime);
739 cow_sizes(version, size, sectorsize, align, *bitmap_offset_out,
740 bitmap_len_out, data_offset_out);
748 static int create_cow_file(char *cow_file, char *backing_file,
749 struct openflags flags,
750 int sectorsize, int alignment, int *bitmap_offset_out,
751 unsigned long *bitmap_len_out, int *data_offset_out)
756 fd = open_ubd_file(cow_file, &flags, 0, NULL, NULL, NULL, NULL, NULL);
759 printk(KERN_ERR "Open of COW file '%s' failed, errno = %d\n",
764 err = init_cow_file(fd, cow_file, backing_file, sectorsize, alignment,
765 bitmap_offset_out, bitmap_len_out,
774 static void ubd_close_dev(struct ubd *ubd_dev)
776 os_close_file(ubd_dev->fd);
777 if(ubd_dev->cow.file == NULL)
780 os_close_file(ubd_dev->cow.fd);
781 vfree(ubd_dev->cow.bitmap);
782 ubd_dev->cow.bitmap = NULL;
785 static int ubd_open_dev(struct ubd *ubd_dev)
787 struct openflags flags;
789 int err, create_cow, *create_ptr;
792 ubd_dev->openflags = ubd_dev->boot_openflags;
794 create_ptr = (ubd_dev->cow.file != NULL) ? &create_cow : NULL;
795 back_ptr = ubd_dev->no_cow ? NULL : &ubd_dev->cow.file;
797 fd = open_ubd_file(ubd_dev->file, &ubd_dev->openflags, ubd_dev->shared,
798 back_ptr, &ubd_dev->cow.bitmap_offset,
799 &ubd_dev->cow.bitmap_len, &ubd_dev->cow.data_offset,
802 if((fd == -ENOENT) && create_cow){
803 fd = create_cow_file(ubd_dev->file, ubd_dev->cow.file,
804 ubd_dev->openflags, SECTOR_SIZE, PAGE_SIZE,
805 &ubd_dev->cow.bitmap_offset,
806 &ubd_dev->cow.bitmap_len,
807 &ubd_dev->cow.data_offset);
809 printk(KERN_INFO "Creating \"%s\" as COW file for "
810 "\"%s\"\n", ubd_dev->file, ubd_dev->cow.file);
815 printk("Failed to open '%s', errno = %d\n", ubd_dev->file,
821 if(ubd_dev->cow.file != NULL){
822 blk_queue_max_hw_sectors(ubd_dev->queue, 8 * sizeof(long));
825 ubd_dev->cow.bitmap = vmalloc(ubd_dev->cow.bitmap_len);
826 if(ubd_dev->cow.bitmap == NULL){
827 printk(KERN_ERR "Failed to vmalloc COW bitmap\n");
830 flush_tlb_kernel_vm();
832 err = read_cow_bitmap(ubd_dev->fd, ubd_dev->cow.bitmap,
833 ubd_dev->cow.bitmap_offset,
834 ubd_dev->cow.bitmap_len);
838 flags = ubd_dev->openflags;
840 err = open_ubd_file(ubd_dev->cow.file, &flags, ubd_dev->shared, NULL,
841 NULL, NULL, NULL, NULL);
842 if(err < 0) goto error;
843 ubd_dev->cow.fd = err;
845 if (ubd_dev->no_trim == 0) {
846 ubd_dev->queue->limits.discard_granularity = SECTOR_SIZE;
847 ubd_dev->queue->limits.discard_alignment = SECTOR_SIZE;
848 blk_queue_max_discard_sectors(ubd_dev->queue, UBD_MAX_REQUEST);
849 blk_queue_max_write_zeroes_sectors(ubd_dev->queue, UBD_MAX_REQUEST);
850 blk_queue_flag_set(QUEUE_FLAG_DISCARD, ubd_dev->queue);
852 blk_queue_flag_set(QUEUE_FLAG_NONROT, ubd_dev->queue);
855 os_close_file(ubd_dev->fd);
859 static void ubd_device_release(struct device *dev)
861 struct ubd *ubd_dev = dev_get_drvdata(dev);
863 blk_cleanup_queue(ubd_dev->queue);
864 blk_mq_free_tag_set(&ubd_dev->tag_set);
865 *ubd_dev = ((struct ubd) DEFAULT_UBD);
868 static int ubd_disk_register(int major, u64 size, int unit,
869 struct gendisk **disk_out)
871 struct device *parent = NULL;
872 struct gendisk *disk;
874 disk = alloc_disk(1 << UBD_SHIFT);
879 disk->first_minor = unit << UBD_SHIFT;
880 disk->fops = &ubd_blops;
881 set_capacity(disk, size / 512);
882 if (major == UBD_MAJOR)
883 sprintf(disk->disk_name, "ubd%c", 'a' + unit);
885 sprintf(disk->disk_name, "ubd_fake%d", unit);
887 /* sysfs register (not for ide fake devices) */
888 if (major == UBD_MAJOR) {
889 ubd_devs[unit].pdev.id = unit;
890 ubd_devs[unit].pdev.name = DRIVER_NAME;
891 ubd_devs[unit].pdev.dev.release = ubd_device_release;
892 dev_set_drvdata(&ubd_devs[unit].pdev.dev, &ubd_devs[unit]);
893 platform_device_register(&ubd_devs[unit].pdev);
894 parent = &ubd_devs[unit].pdev.dev;
897 disk->private_data = &ubd_devs[unit];
898 disk->queue = ubd_devs[unit].queue;
899 device_add_disk(parent, disk, NULL);
905 #define ROUND_BLOCK(n) ((n + (SECTOR_SIZE - 1)) & (-SECTOR_SIZE))
907 static const struct blk_mq_ops ubd_mq_ops = {
908 .queue_rq = ubd_queue_rq,
911 static int ubd_add(int n, char **error_out)
913 struct ubd *ubd_dev = &ubd_devs[n];
916 if(ubd_dev->file == NULL)
919 err = ubd_file_size(ubd_dev, &ubd_dev->size);
921 *error_out = "Couldn't determine size of device's file";
925 ubd_dev->size = ROUND_BLOCK(ubd_dev->size);
927 ubd_dev->tag_set.ops = &ubd_mq_ops;
928 ubd_dev->tag_set.queue_depth = 64;
929 ubd_dev->tag_set.numa_node = NUMA_NO_NODE;
930 ubd_dev->tag_set.flags = BLK_MQ_F_SHOULD_MERGE;
931 ubd_dev->tag_set.driver_data = ubd_dev;
932 ubd_dev->tag_set.nr_hw_queues = 1;
934 err = blk_mq_alloc_tag_set(&ubd_dev->tag_set);
938 ubd_dev->queue = blk_mq_init_queue(&ubd_dev->tag_set);
939 if (IS_ERR(ubd_dev->queue)) {
940 err = PTR_ERR(ubd_dev->queue);
941 goto out_cleanup_tags;
944 ubd_dev->queue->queuedata = ubd_dev;
945 blk_queue_write_cache(ubd_dev->queue, true, false);
947 blk_queue_max_segments(ubd_dev->queue, MAX_SG);
948 err = ubd_disk_register(UBD_MAJOR, ubd_dev->size, n, &ubd_gendisk[n]);
950 *error_out = "Failed to register device";
951 goto out_cleanup_tags;
954 if (fake_major != UBD_MAJOR)
955 ubd_disk_register(fake_major, ubd_dev->size, n,
959 * Perhaps this should also be under the "if (fake_major)" above
960 * using the fake_disk->disk_name
963 make_ide_entries(ubd_gendisk[n]->disk_name);
970 blk_mq_free_tag_set(&ubd_dev->tag_set);
971 if (!(IS_ERR(ubd_dev->queue)))
972 blk_cleanup_queue(ubd_dev->queue);
976 static int ubd_config(char *str, char **error_out)
980 /* This string is possibly broken up and stored, so it's only
981 * freed if ubd_setup_common fails, or if only general options
984 str = kstrdup(str, GFP_KERNEL);
986 *error_out = "Failed to allocate memory";
990 ret = ubd_setup_common(str, &n, error_out);
999 mutex_lock(&ubd_lock);
1000 ret = ubd_add(n, error_out);
1002 ubd_devs[n].file = NULL;
1003 mutex_unlock(&ubd_lock);
1013 static int ubd_get_config(char *name, char *str, int size, char **error_out)
1015 struct ubd *ubd_dev;
1018 n = parse_unit(&name);
1019 if((n >= MAX_DEV) || (n < 0)){
1020 *error_out = "ubd_get_config : device number out of range";
1024 ubd_dev = &ubd_devs[n];
1025 mutex_lock(&ubd_lock);
1027 if(ubd_dev->file == NULL){
1028 CONFIG_CHUNK(str, size, len, "", 1);
1032 CONFIG_CHUNK(str, size, len, ubd_dev->file, 0);
1034 if(ubd_dev->cow.file != NULL){
1035 CONFIG_CHUNK(str, size, len, ",", 0);
1036 CONFIG_CHUNK(str, size, len, ubd_dev->cow.file, 1);
1038 else CONFIG_CHUNK(str, size, len, "", 1);
1041 mutex_unlock(&ubd_lock);
1045 static int ubd_id(char **str, int *start_out, int *end_out)
1049 n = parse_unit(str);
1051 *end_out = MAX_DEV - 1;
1055 static int ubd_remove(int n, char **error_out)
1057 struct gendisk *disk = ubd_gendisk[n];
1058 struct ubd *ubd_dev;
1061 mutex_lock(&ubd_lock);
1063 ubd_dev = &ubd_devs[n];
1065 if(ubd_dev->file == NULL)
1068 /* you cannot remove a open disk */
1070 if(ubd_dev->count > 0)
1073 ubd_gendisk[n] = NULL;
1079 if(fake_gendisk[n] != NULL){
1080 del_gendisk(fake_gendisk[n]);
1081 put_disk(fake_gendisk[n]);
1082 fake_gendisk[n] = NULL;
1086 platform_device_unregister(&ubd_dev->pdev);
1088 mutex_unlock(&ubd_lock);
1092 /* All these are called by mconsole in process context and without
1093 * ubd-specific locks. The structure itself is const except for .list.
1095 static struct mc_device ubd_mc = {
1096 .list = LIST_HEAD_INIT(ubd_mc.list),
1098 .config = ubd_config,
1099 .get_config = ubd_get_config,
1101 .remove = ubd_remove,
1104 static int __init ubd_mc_init(void)
1106 mconsole_register_dev(&ubd_mc);
1110 __initcall(ubd_mc_init);
1112 static int __init ubd0_init(void)
1114 struct ubd *ubd_dev = &ubd_devs[0];
1116 mutex_lock(&ubd_lock);
1117 if(ubd_dev->file == NULL)
1118 ubd_dev->file = "root_fs";
1119 mutex_unlock(&ubd_lock);
1124 __initcall(ubd0_init);
1126 /* Used in ubd_init, which is an initcall */
1127 static struct platform_driver ubd_driver = {
1129 .name = DRIVER_NAME,
1133 static int __init ubd_init(void)
1138 if (register_blkdev(UBD_MAJOR, "ubd"))
1141 if (fake_major != UBD_MAJOR) {
1142 char name[sizeof("ubd_nnn\0")];
1144 snprintf(name, sizeof(name), "ubd_%d", fake_major);
1145 if (register_blkdev(fake_major, "ubd"))
1149 irq_req_buffer = kmalloc_array(UBD_REQ_BUFFER_SIZE,
1150 sizeof(struct io_thread_req *),
1155 if (irq_req_buffer == NULL) {
1156 printk(KERN_ERR "Failed to initialize ubd buffering\n");
1159 io_req_buffer = kmalloc_array(UBD_REQ_BUFFER_SIZE,
1160 sizeof(struct io_thread_req *),
1166 if (io_req_buffer == NULL) {
1167 printk(KERN_ERR "Failed to initialize ubd buffering\n");
1170 platform_driver_register(&ubd_driver);
1171 mutex_lock(&ubd_lock);
1172 for (i = 0; i < MAX_DEV; i++){
1173 err = ubd_add(i, &error);
1175 printk(KERN_ERR "Failed to initialize ubd device %d :"
1178 mutex_unlock(&ubd_lock);
1182 late_initcall(ubd_init);
1184 static int __init ubd_driver_init(void){
1185 unsigned long stack;
1188 /* Set by CONFIG_BLK_DEV_UBD_SYNC or ubd=sync.*/
1189 if(global_openflags.s){
1190 printk(KERN_INFO "ubd: Synchronous mode\n");
1191 /* Letting ubd=sync be like using ubd#s= instead of ubd#= is
1192 * enough. So use anyway the io thread. */
1194 stack = alloc_stack(0, 0);
1195 io_pid = start_io_thread(stack + PAGE_SIZE - sizeof(void *),
1199 "ubd : Failed to start I/O thread (errno = %d) - "
1200 "falling back to synchronous I/O\n", -io_pid);
1204 err = um_request_irq(UBD_IRQ, thread_fd, IRQ_READ, ubd_intr,
1205 0, "ubd", ubd_devs);
1207 printk(KERN_ERR "um_request_irq failed - errno = %d\n", -err);
1211 device_initcall(ubd_driver_init);
1213 static int ubd_open(struct block_device *bdev, fmode_t mode)
1215 struct gendisk *disk = bdev->bd_disk;
1216 struct ubd *ubd_dev = disk->private_data;
1219 mutex_lock(&ubd_mutex);
1220 if(ubd_dev->count == 0){
1221 err = ubd_open_dev(ubd_dev);
1223 printk(KERN_ERR "%s: Can't open \"%s\": errno = %d\n",
1224 disk->disk_name, ubd_dev->file, -err);
1229 set_disk_ro(disk, !ubd_dev->openflags.w);
1231 /* This should no more be needed. And it didn't work anyway to exclude
1232 * read-write remounting of filesystems.*/
1233 /*if((mode & FMODE_WRITE) && !ubd_dev->openflags.w){
1234 if(--ubd_dev->count == 0) ubd_close_dev(ubd_dev);
1238 mutex_unlock(&ubd_mutex);
1242 static void ubd_release(struct gendisk *disk, fmode_t mode)
1244 struct ubd *ubd_dev = disk->private_data;
1246 mutex_lock(&ubd_mutex);
1247 if(--ubd_dev->count == 0)
1248 ubd_close_dev(ubd_dev);
1249 mutex_unlock(&ubd_mutex);
1252 static void cowify_bitmap(__u64 io_offset, int length, unsigned long *cow_mask,
1253 __u64 *cow_offset, unsigned long *bitmap,
1254 __u64 bitmap_offset, unsigned long *bitmap_words,
1257 __u64 sector = io_offset >> SECTOR_SHIFT;
1258 int i, update_bitmap = 0;
1260 for (i = 0; i < length >> SECTOR_SHIFT; i++) {
1261 if(cow_mask != NULL)
1262 ubd_set_bit(i, (unsigned char *) cow_mask);
1263 if(ubd_test_bit(sector + i, (unsigned char *) bitmap))
1267 ubd_set_bit(sector + i, (unsigned char *) bitmap);
1273 *cow_offset = sector / (sizeof(unsigned long) * 8);
1275 /* This takes care of the case where we're exactly at the end of the
1276 * device, and *cow_offset + 1 is off the end. So, just back it up
1277 * by one word. Thanks to Lynn Kerby for the fix and James McMechan
1278 * for the original diagnosis.
1280 if (*cow_offset == (DIV_ROUND_UP(bitmap_len,
1281 sizeof(unsigned long)) - 1))
1284 bitmap_words[0] = bitmap[*cow_offset];
1285 bitmap_words[1] = bitmap[*cow_offset + 1];
1287 *cow_offset *= sizeof(unsigned long);
1288 *cow_offset += bitmap_offset;
1291 static void cowify_req(struct io_thread_req *req, unsigned long *bitmap,
1292 __u64 bitmap_offset, __u64 bitmap_len)
1294 __u64 sector = req->offset >> SECTOR_SHIFT;
1297 if (req->length > (sizeof(req->sector_mask) * 8) << SECTOR_SHIFT)
1298 panic("Operation too long");
1300 if (req_op(req->req) == REQ_OP_READ) {
1301 for (i = 0; i < req->length >> SECTOR_SHIFT; i++) {
1302 if(ubd_test_bit(sector + i, (unsigned char *) bitmap))
1303 ubd_set_bit(i, (unsigned char *)
1307 else cowify_bitmap(req->offset, req->length, &req->sector_mask,
1308 &req->cow_offset, bitmap, bitmap_offset,
1309 req->bitmap_words, bitmap_len);
1312 static int ubd_queue_one_vec(struct blk_mq_hw_ctx *hctx, struct request *req,
1313 u64 off, struct bio_vec *bvec)
1315 struct ubd *dev = hctx->queue->queuedata;
1316 struct io_thread_req *io_req;
1319 io_req = kmalloc(sizeof(struct io_thread_req), GFP_ATOMIC);
1325 io_req->fds[0] = dev->cow.fd;
1327 io_req->fds[0] = dev->fd;
1331 io_req->buffer = page_address(bvec->bv_page) + bvec->bv_offset;
1332 io_req->length = bvec->bv_len;
1334 io_req->buffer = NULL;
1335 io_req->length = blk_rq_bytes(req);
1338 io_req->sectorsize = SECTOR_SIZE;
1339 io_req->fds[1] = dev->fd;
1340 io_req->cow_offset = -1;
1341 io_req->offset = off;
1342 io_req->sector_mask = 0;
1343 io_req->offsets[0] = 0;
1344 io_req->offsets[1] = dev->cow.data_offset;
1347 cowify_req(io_req, dev->cow.bitmap,
1348 dev->cow.bitmap_offset, dev->cow.bitmap_len);
1350 ret = os_write_file(thread_fd, &io_req, sizeof(io_req));
1351 if (ret != sizeof(io_req)) {
1353 pr_err("write to io thread failed: %d\n", -ret);
1359 static int queue_rw_req(struct blk_mq_hw_ctx *hctx, struct request *req)
1361 struct req_iterator iter;
1362 struct bio_vec bvec;
1364 u64 off = (u64)blk_rq_pos(req) << SECTOR_SHIFT;
1366 rq_for_each_segment(bvec, req, iter) {
1367 ret = ubd_queue_one_vec(hctx, req, off, &bvec);
1375 static blk_status_t ubd_queue_rq(struct blk_mq_hw_ctx *hctx,
1376 const struct blk_mq_queue_data *bd)
1378 struct ubd *ubd_dev = hctx->queue->queuedata;
1379 struct request *req = bd->rq;
1380 int ret = 0, res = BLK_STS_OK;
1382 blk_mq_start_request(req);
1384 spin_lock_irq(&ubd_dev->lock);
1386 switch (req_op(req)) {
1387 /* operations with no lentgth/offset arguments */
1389 ret = ubd_queue_one_vec(hctx, req, 0, NULL);
1393 ret = queue_rw_req(hctx, req);
1395 case REQ_OP_DISCARD:
1396 case REQ_OP_WRITE_ZEROES:
1397 ret = ubd_queue_one_vec(hctx, req, (u64)blk_rq_pos(req) << 9, NULL);
1401 res = BLK_STS_NOTSUPP;
1404 spin_unlock_irq(&ubd_dev->lock);
1408 res = BLK_STS_RESOURCE;
1410 res = BLK_STS_DEV_RESOURCE;
1416 static int ubd_getgeo(struct block_device *bdev, struct hd_geometry *geo)
1418 struct ubd *ubd_dev = bdev->bd_disk->private_data;
1422 geo->cylinders = ubd_dev->size / (128 * 32 * 512);
1426 static int ubd_ioctl(struct block_device *bdev, fmode_t mode,
1427 unsigned int cmd, unsigned long arg)
1429 struct ubd *ubd_dev = bdev->bd_disk->private_data;
1430 u16 ubd_id[ATA_ID_WORDS];
1433 struct cdrom_volctrl volume;
1434 case HDIO_GET_IDENTITY:
1435 memset(&ubd_id, 0, ATA_ID_WORDS * 2);
1436 ubd_id[ATA_ID_CYLS] = ubd_dev->size / (128 * 32 * 512);
1437 ubd_id[ATA_ID_HEADS] = 128;
1438 ubd_id[ATA_ID_SECTORS] = 32;
1439 if(copy_to_user((char __user *) arg, (char *) &ubd_id,
1445 if(copy_from_user(&volume, (char __user *) arg, sizeof(volume)))
1447 volume.channel0 = 255;
1448 volume.channel1 = 255;
1449 volume.channel2 = 255;
1450 volume.channel3 = 255;
1451 if(copy_to_user((char __user *) arg, &volume, sizeof(volume)))
1458 static int map_error(int error_code)
1460 switch (error_code) {
1465 return BLK_STS_NOTSUPP;
1467 return BLK_STS_NOSPC;
1469 return BLK_STS_IOERR;
1473 * Everything from here onwards *IS NOT PART OF THE KERNEL*
1475 * The following functions are part of UML hypervisor code.
1476 * All functions from here onwards are executed as a helper
1477 * thread and are not allowed to execute any kernel functions.
1479 * Any communication must occur strictly via shared memory and IPC.
1481 * Do not add printks, locks, kernel memory operations, etc - it
1482 * will result in unpredictable behaviour and/or crashes.
1485 static int update_bitmap(struct io_thread_req *req)
1489 if(req->cow_offset == -1)
1490 return map_error(0);
1492 n = os_pwrite_file(req->fds[1], &req->bitmap_words,
1493 sizeof(req->bitmap_words), req->cow_offset);
1494 if (n != sizeof(req->bitmap_words))
1495 return map_error(-n);
1497 return map_error(0);
1500 static void do_io(struct io_thread_req *req)
1504 int n, nsectors, start, end, bit;
1507 /* FLUSH is really a special case, we cannot "case" it with others */
1509 if (req_op(req->req) == REQ_OP_FLUSH) {
1510 /* fds[0] is always either the rw image or our cow file */
1511 req->error = map_error(-os_sync_file(req->fds[0]));
1515 nsectors = req->length / req->sectorsize;
1518 bit = ubd_test_bit(start, (unsigned char *) &req->sector_mask);
1520 while((end < nsectors) &&
1521 (ubd_test_bit(end, (unsigned char *)
1522 &req->sector_mask) == bit))
1525 off = req->offset + req->offsets[bit] +
1526 start * req->sectorsize;
1527 len = (end - start) * req->sectorsize;
1528 if (req->buffer != NULL)
1529 buf = &req->buffer[start * req->sectorsize];
1531 switch (req_op(req->req)) {
1537 n = os_pread_file(req->fds[bit], buf, len, off);
1539 req->error = map_error(-n);
1542 } while((n < len) && (n != 0));
1543 if (n < len) memset(&buf[n], 0, len - n);
1546 n = os_pwrite_file(req->fds[bit], buf, len, off);
1548 req->error = map_error(-n);
1552 case REQ_OP_DISCARD:
1553 case REQ_OP_WRITE_ZEROES:
1554 n = os_falloc_punch(req->fds[bit], off, len);
1556 req->error = map_error(-n);
1562 req->error = BLK_STS_NOTSUPP;
1567 } while(start < nsectors);
1569 req->error = update_bitmap(req);
1572 /* Changed in start_io_thread, which is serialized by being called only
1573 * from ubd_init, which is an initcall.
1577 /* Only changed by the io thread. XXX: currently unused. */
1578 static int io_count = 0;
1580 int io_thread(void *arg)
1582 int n, count, written, res;
1584 os_fix_helper_signals();
1587 n = bulk_req_safe_read(
1601 for (count = 0; count < n/sizeof(struct io_thread_req *); count++) {
1603 do_io((*io_req_buffer)[count]);
1609 res = os_write_file(kernel_fd, ((char *) io_req_buffer) + written, n);
1616 } while (written < n);