11da177e4SLinus Torvalds /* 21da177e4SLinus Torvalds * Copyright (C) 2000 Jeff Dike (jdike@karaya.com) 31da177e4SLinus Torvalds * Licensed under the GPL 41da177e4SLinus Torvalds */ 51da177e4SLinus Torvalds 61da177e4SLinus Torvalds /* 2001-09-28...2002-04-17 71da177e4SLinus Torvalds * Partition stuff by James_McMechan@hotmail.com 81da177e4SLinus Torvalds * old style ubd by setting UBD_SHIFT to 0 91da177e4SLinus Torvalds * 2002-09-27...2002-10-18 massive tinkering for 2.5 101da177e4SLinus Torvalds * partitions have changed in 2.5 111da177e4SLinus Torvalds * 2003-01-29 more tinkering for 2.5.59-1 121da177e4SLinus Torvalds * This should now address the sysfs problems and has 131da177e4SLinus Torvalds * the symlink for devfs to allow for booting with 141da177e4SLinus Torvalds * the common /dev/ubd/discX/... names rather than 151da177e4SLinus Torvalds * only /dev/ubdN/discN this version also has lots of 161da177e4SLinus Torvalds * clean ups preparing for ubd-many. 171da177e4SLinus Torvalds * James McMechan 181da177e4SLinus Torvalds */ 191da177e4SLinus Torvalds 201da177e4SLinus Torvalds #define MAJOR_NR UBD_MAJOR 211da177e4SLinus Torvalds #define UBD_SHIFT 4 221da177e4SLinus Torvalds 231da177e4SLinus Torvalds #include "linux/config.h" 241da177e4SLinus Torvalds #include "linux/module.h" 251da177e4SLinus Torvalds #include "linux/blkdev.h" 261da177e4SLinus Torvalds #include "linux/hdreg.h" 271da177e4SLinus Torvalds #include "linux/init.h" 281da177e4SLinus Torvalds #include "linux/cdrom.h" 291da177e4SLinus Torvalds #include "linux/proc_fs.h" 301da177e4SLinus Torvalds #include "linux/ctype.h" 311da177e4SLinus Torvalds #include "linux/capability.h" 321da177e4SLinus Torvalds #include "linux/mm.h" 331da177e4SLinus Torvalds #include "linux/vmalloc.h" 341da177e4SLinus Torvalds #include "linux/blkpg.h" 351da177e4SLinus Torvalds #include "linux/genhd.h" 361da177e4SLinus Torvalds #include "linux/spinlock.h" 37d052d1beSRussell King #include "linux/platform_device.h" 381da177e4SLinus Torvalds #include "asm/segment.h" 391da177e4SLinus Torvalds #include "asm/uaccess.h" 401da177e4SLinus Torvalds #include "asm/irq.h" 411da177e4SLinus Torvalds #include "asm/types.h" 421da177e4SLinus Torvalds #include "asm/tlbflush.h" 431da177e4SLinus Torvalds #include "user_util.h" 441da177e4SLinus Torvalds #include "mem_user.h" 451da177e4SLinus Torvalds #include "kern_util.h" 461da177e4SLinus Torvalds #include "kern.h" 471da177e4SLinus Torvalds #include "mconsole_kern.h" 481da177e4SLinus Torvalds #include "init.h" 491da177e4SLinus Torvalds #include "irq_user.h" 501da177e4SLinus Torvalds #include "irq_kern.h" 511da177e4SLinus Torvalds #include "ubd_user.h" 521da177e4SLinus Torvalds #include "os.h" 531da177e4SLinus Torvalds #include "mem.h" 541da177e4SLinus Torvalds #include "mem_kern.h" 551da177e4SLinus Torvalds #include "cow.h" 561da177e4SLinus Torvalds 577b9014c1SJeff Dike enum ubd_req { UBD_READ, UBD_WRITE }; 581da177e4SLinus Torvalds 591da177e4SLinus Torvalds struct io_thread_req { 6091acb21fSJeff Dike enum ubd_req op; 611da177e4SLinus Torvalds int fds[2]; 621da177e4SLinus Torvalds unsigned long offsets[2]; 631da177e4SLinus Torvalds unsigned long long offset; 641da177e4SLinus Torvalds unsigned long length; 651da177e4SLinus Torvalds char *buffer; 661da177e4SLinus Torvalds int sectorsize; 6791acb21fSJeff Dike unsigned long sector_mask; 6891acb21fSJeff Dike unsigned long long cow_offset; 6991acb21fSJeff Dike unsigned long bitmap_words[2]; 701da177e4SLinus Torvalds int error; 711da177e4SLinus Torvalds }; 721da177e4SLinus Torvalds 736c29256cSJeff Dike extern int open_ubd_file(char *file, struct openflags *openflags, int shared, 741da177e4SLinus Torvalds char **backing_file_out, int *bitmap_offset_out, 751da177e4SLinus Torvalds unsigned long *bitmap_len_out, int *data_offset_out, 761da177e4SLinus Torvalds int *create_cow_out); 771da177e4SLinus Torvalds extern int create_cow_file(char *cow_file, char *backing_file, 781da177e4SLinus Torvalds struct openflags flags, int sectorsize, 791da177e4SLinus Torvalds int alignment, int *bitmap_offset_out, 801da177e4SLinus Torvalds unsigned long *bitmap_len_out, 811da177e4SLinus Torvalds int *data_offset_out); 821da177e4SLinus Torvalds extern int read_cow_bitmap(int fd, void *buf, int offset, int len); 8391acb21fSJeff Dike extern void do_io(struct io_thread_req *req); 841da177e4SLinus Torvalds 8591acb21fSJeff Dike static inline int ubd_test_bit(__u64 bit, unsigned char *data) 861da177e4SLinus Torvalds { 871da177e4SLinus Torvalds __u64 n; 881da177e4SLinus Torvalds int bits, off; 891da177e4SLinus Torvalds 9091acb21fSJeff Dike bits = sizeof(data[0]) * 8; 911da177e4SLinus Torvalds n = bit / bits; 921da177e4SLinus Torvalds off = bit % bits; 9391acb21fSJeff Dike return((data[n] & (1 << off)) != 0); 941da177e4SLinus Torvalds } 951da177e4SLinus Torvalds 9691acb21fSJeff Dike static inline void ubd_set_bit(__u64 bit, unsigned char *data) 971da177e4SLinus Torvalds { 981da177e4SLinus Torvalds __u64 n; 991da177e4SLinus Torvalds int bits, off; 1001da177e4SLinus Torvalds 10191acb21fSJeff Dike bits = sizeof(data[0]) * 8; 1021da177e4SLinus Torvalds n = bit / bits; 1031da177e4SLinus Torvalds off = bit % bits; 10491acb21fSJeff Dike data[n] |= (1 << off); 1051da177e4SLinus Torvalds } 1061da177e4SLinus Torvalds /*End stuff from ubd_user.h*/ 1071da177e4SLinus Torvalds 1081da177e4SLinus Torvalds #define DRIVER_NAME "uml-blkdev" 1091da177e4SLinus Torvalds 1101da177e4SLinus Torvalds static DEFINE_SPINLOCK(ubd_io_lock); 1111da177e4SLinus Torvalds static DEFINE_SPINLOCK(ubd_lock); 1121da177e4SLinus Torvalds 11391acb21fSJeff Dike static void (*do_ubd)(void); 11491acb21fSJeff Dike 1151da177e4SLinus Torvalds static int ubd_open(struct inode * inode, struct file * filp); 1161da177e4SLinus Torvalds static int ubd_release(struct inode * inode, struct file * file); 1171da177e4SLinus Torvalds static int ubd_ioctl(struct inode * inode, struct file * file, 1181da177e4SLinus Torvalds unsigned int cmd, unsigned long arg); 119a885c8c4SChristoph Hellwig static int ubd_getgeo(struct block_device *bdev, struct hd_geometry *geo); 1201da177e4SLinus Torvalds 1211da177e4SLinus Torvalds #define MAX_DEV (8) 1221da177e4SLinus Torvalds 1231da177e4SLinus Torvalds static struct block_device_operations ubd_blops = { 1241da177e4SLinus Torvalds .owner = THIS_MODULE, 1251da177e4SLinus Torvalds .open = ubd_open, 1261da177e4SLinus Torvalds .release = ubd_release, 1271da177e4SLinus Torvalds .ioctl = ubd_ioctl, 128a885c8c4SChristoph Hellwig .getgeo = ubd_getgeo, 1291da177e4SLinus Torvalds }; 1301da177e4SLinus Torvalds 1311da177e4SLinus Torvalds /* Protected by the queue_lock */ 1321da177e4SLinus Torvalds static request_queue_t *ubd_queue; 1331da177e4SLinus Torvalds 1341da177e4SLinus Torvalds /* Protected by ubd_lock */ 1351da177e4SLinus Torvalds static int fake_major = MAJOR_NR; 1361da177e4SLinus Torvalds 1371da177e4SLinus Torvalds static struct gendisk *ubd_gendisk[MAX_DEV]; 1381da177e4SLinus Torvalds static struct gendisk *fake_gendisk[MAX_DEV]; 1391da177e4SLinus Torvalds 1401da177e4SLinus Torvalds #ifdef CONFIG_BLK_DEV_UBD_SYNC 1411da177e4SLinus Torvalds #define OPEN_FLAGS ((struct openflags) { .r = 1, .w = 1, .s = 1, .c = 0, \ 1421da177e4SLinus Torvalds .cl = 1 }) 1431da177e4SLinus Torvalds #else 1441da177e4SLinus Torvalds #define OPEN_FLAGS ((struct openflags) { .r = 1, .w = 1, .s = 0, .c = 0, \ 1451da177e4SLinus Torvalds .cl = 1 }) 1461da177e4SLinus Torvalds #endif 1471da177e4SLinus Torvalds 1481da177e4SLinus Torvalds /* Not protected - changed only in ubd_setup_common and then only to 1491da177e4SLinus Torvalds * to enable O_SYNC. 1501da177e4SLinus Torvalds */ 1511da177e4SLinus Torvalds static struct openflags global_openflags = OPEN_FLAGS; 1521da177e4SLinus Torvalds 1531da177e4SLinus Torvalds struct cow { 1542c49be99SPaolo 'Blaisorblade' Giarrusso /* This is the backing file, actually */ 1551da177e4SLinus Torvalds char *file; 1561da177e4SLinus Torvalds int fd; 1571da177e4SLinus Torvalds unsigned long *bitmap; 1581da177e4SLinus Torvalds unsigned long bitmap_len; 1591da177e4SLinus Torvalds int bitmap_offset; 1601da177e4SLinus Torvalds int data_offset; 1611da177e4SLinus Torvalds }; 1621da177e4SLinus Torvalds 1631da177e4SLinus Torvalds struct ubd { 1641da177e4SLinus Torvalds char *file; 1651da177e4SLinus Torvalds int count; 1661da177e4SLinus Torvalds int fd; 1671da177e4SLinus Torvalds __u64 size; 1681da177e4SLinus Torvalds struct openflags boot_openflags; 1691da177e4SLinus Torvalds struct openflags openflags; 1706c29256cSJeff Dike int shared; 1711da177e4SLinus Torvalds int no_cow; 1721da177e4SLinus Torvalds struct cow cow; 1731da177e4SLinus Torvalds struct platform_device pdev; 1741da177e4SLinus Torvalds }; 1751da177e4SLinus Torvalds 1761da177e4SLinus Torvalds #define DEFAULT_COW { \ 1771da177e4SLinus Torvalds .file = NULL, \ 1781da177e4SLinus Torvalds .fd = -1, \ 1791da177e4SLinus Torvalds .bitmap = NULL, \ 1801da177e4SLinus Torvalds .bitmap_offset = 0, \ 1811da177e4SLinus Torvalds .data_offset = 0, \ 1821da177e4SLinus Torvalds } 1831da177e4SLinus Torvalds 1841da177e4SLinus Torvalds #define DEFAULT_UBD { \ 1851da177e4SLinus Torvalds .file = NULL, \ 1861da177e4SLinus Torvalds .count = 0, \ 1871da177e4SLinus Torvalds .fd = -1, \ 1881da177e4SLinus Torvalds .size = -1, \ 1891da177e4SLinus Torvalds .boot_openflags = OPEN_FLAGS, \ 1901da177e4SLinus Torvalds .openflags = OPEN_FLAGS, \ 1911da177e4SLinus Torvalds .no_cow = 0, \ 1926c29256cSJeff Dike .shared = 0, \ 1931da177e4SLinus Torvalds .cow = DEFAULT_COW, \ 1941da177e4SLinus Torvalds } 1951da177e4SLinus Torvalds 1961da177e4SLinus Torvalds struct ubd ubd_dev[MAX_DEV] = { [ 0 ... MAX_DEV - 1 ] = DEFAULT_UBD }; 1971da177e4SLinus Torvalds 1981da177e4SLinus Torvalds static int ubd0_init(void) 1991da177e4SLinus Torvalds { 2001da177e4SLinus Torvalds struct ubd *dev = &ubd_dev[0]; 2011da177e4SLinus Torvalds 2021da177e4SLinus Torvalds if(dev->file == NULL) 2031da177e4SLinus Torvalds dev->file = "root_fs"; 2041da177e4SLinus Torvalds return(0); 2051da177e4SLinus Torvalds } 2061da177e4SLinus Torvalds 2071da177e4SLinus Torvalds __initcall(ubd0_init); 2081da177e4SLinus Torvalds 2091da177e4SLinus Torvalds /* Only changed by fake_ide_setup which is a setup */ 2101da177e4SLinus Torvalds static int fake_ide = 0; 2111da177e4SLinus Torvalds static struct proc_dir_entry *proc_ide_root = NULL; 2121da177e4SLinus Torvalds static struct proc_dir_entry *proc_ide = NULL; 2131da177e4SLinus Torvalds 2141da177e4SLinus Torvalds static void make_proc_ide(void) 2151da177e4SLinus Torvalds { 2161da177e4SLinus Torvalds proc_ide_root = proc_mkdir("ide", NULL); 2171da177e4SLinus Torvalds proc_ide = proc_mkdir("ide0", proc_ide_root); 2181da177e4SLinus Torvalds } 2191da177e4SLinus Torvalds 2201da177e4SLinus Torvalds static int proc_ide_read_media(char *page, char **start, off_t off, int count, 2211da177e4SLinus Torvalds int *eof, void *data) 2221da177e4SLinus Torvalds { 2231da177e4SLinus Torvalds int len; 2241da177e4SLinus Torvalds 2251da177e4SLinus Torvalds strcpy(page, "disk\n"); 2261da177e4SLinus Torvalds len = strlen("disk\n"); 2271da177e4SLinus Torvalds len -= off; 2281da177e4SLinus Torvalds if (len < count){ 2291da177e4SLinus Torvalds *eof = 1; 2301da177e4SLinus Torvalds if (len <= 0) return 0; 2311da177e4SLinus Torvalds } 2321da177e4SLinus Torvalds else len = count; 2331da177e4SLinus Torvalds *start = page + off; 2341da177e4SLinus Torvalds return len; 2351da177e4SLinus Torvalds } 2361da177e4SLinus Torvalds 2371da177e4SLinus Torvalds static void make_ide_entries(char *dev_name) 2381da177e4SLinus Torvalds { 2391da177e4SLinus Torvalds struct proc_dir_entry *dir, *ent; 2401da177e4SLinus Torvalds char name[64]; 2411da177e4SLinus Torvalds 2421da177e4SLinus Torvalds if(proc_ide_root == NULL) make_proc_ide(); 2431da177e4SLinus Torvalds 2441da177e4SLinus Torvalds dir = proc_mkdir(dev_name, proc_ide); 2451da177e4SLinus Torvalds if(!dir) return; 2461da177e4SLinus Torvalds 2471da177e4SLinus Torvalds ent = create_proc_entry("media", S_IFREG|S_IRUGO, dir); 2481da177e4SLinus Torvalds if(!ent) return; 2491da177e4SLinus Torvalds ent->nlink = 1; 2501da177e4SLinus Torvalds ent->data = NULL; 2511da177e4SLinus Torvalds ent->read_proc = proc_ide_read_media; 2521da177e4SLinus Torvalds ent->write_proc = NULL; 2531da177e4SLinus Torvalds sprintf(name,"ide0/%s", dev_name); 2541da177e4SLinus Torvalds proc_symlink(dev_name, proc_ide_root, name); 2551da177e4SLinus Torvalds } 2561da177e4SLinus Torvalds 2571da177e4SLinus Torvalds static int fake_ide_setup(char *str) 2581da177e4SLinus Torvalds { 2591da177e4SLinus Torvalds fake_ide = 1; 2601da177e4SLinus Torvalds return(1); 2611da177e4SLinus Torvalds } 2621da177e4SLinus Torvalds 2631da177e4SLinus Torvalds __setup("fake_ide", fake_ide_setup); 2641da177e4SLinus Torvalds 2651da177e4SLinus Torvalds __uml_help(fake_ide_setup, 2661da177e4SLinus Torvalds "fake_ide\n" 2671da177e4SLinus Torvalds " Create ide0 entries that map onto ubd devices.\n\n" 2681da177e4SLinus Torvalds ); 2691da177e4SLinus Torvalds 2701da177e4SLinus Torvalds static int parse_unit(char **ptr) 2711da177e4SLinus Torvalds { 2721da177e4SLinus Torvalds char *str = *ptr, *end; 2731da177e4SLinus Torvalds int n = -1; 2741da177e4SLinus Torvalds 2751da177e4SLinus Torvalds if(isdigit(*str)) { 2761da177e4SLinus Torvalds n = simple_strtoul(str, &end, 0); 2771da177e4SLinus Torvalds if(end == str) 2781da177e4SLinus Torvalds return(-1); 2791da177e4SLinus Torvalds *ptr = end; 2801da177e4SLinus Torvalds } 2811da177e4SLinus Torvalds else if (('a' <= *str) && (*str <= 'h')) { 2821da177e4SLinus Torvalds n = *str - 'a'; 2831da177e4SLinus Torvalds str++; 2841da177e4SLinus Torvalds *ptr = str; 2851da177e4SLinus Torvalds } 2861da177e4SLinus Torvalds return(n); 2871da177e4SLinus Torvalds } 2881da177e4SLinus Torvalds 2891da177e4SLinus Torvalds static int ubd_setup_common(char *str, int *index_out) 2901da177e4SLinus Torvalds { 2911da177e4SLinus Torvalds struct ubd *dev; 2921da177e4SLinus Torvalds struct openflags flags = global_openflags; 2931da177e4SLinus Torvalds char *backing_file; 2941da177e4SLinus Torvalds int n, err, i; 2951da177e4SLinus Torvalds 2961da177e4SLinus Torvalds if(index_out) *index_out = -1; 2971da177e4SLinus Torvalds n = *str; 2981da177e4SLinus Torvalds if(n == '='){ 2991da177e4SLinus Torvalds char *end; 3001da177e4SLinus Torvalds int major; 3011da177e4SLinus Torvalds 3021da177e4SLinus Torvalds str++; 3031da177e4SLinus Torvalds if(!strcmp(str, "sync")){ 3041da177e4SLinus Torvalds global_openflags = of_sync(global_openflags); 3051da177e4SLinus Torvalds return(0); 3061da177e4SLinus Torvalds } 3071da177e4SLinus Torvalds major = simple_strtoul(str, &end, 0); 3081da177e4SLinus Torvalds if((*end != '\0') || (end == str)){ 3091da177e4SLinus Torvalds printk(KERN_ERR 3101da177e4SLinus Torvalds "ubd_setup : didn't parse major number\n"); 3111da177e4SLinus Torvalds return(1); 3121da177e4SLinus Torvalds } 3131da177e4SLinus Torvalds 3141da177e4SLinus Torvalds err = 1; 3151da177e4SLinus Torvalds spin_lock(&ubd_lock); 3161da177e4SLinus Torvalds if(fake_major != MAJOR_NR){ 3171da177e4SLinus Torvalds printk(KERN_ERR "Can't assign a fake major twice\n"); 3181da177e4SLinus Torvalds goto out1; 3191da177e4SLinus Torvalds } 3201da177e4SLinus Torvalds 3211da177e4SLinus Torvalds fake_major = major; 3221da177e4SLinus Torvalds 3231da177e4SLinus Torvalds printk(KERN_INFO "Setting extra ubd major number to %d\n", 3241da177e4SLinus Torvalds major); 3251da177e4SLinus Torvalds err = 0; 3261da177e4SLinus Torvalds out1: 3271da177e4SLinus Torvalds spin_unlock(&ubd_lock); 3281da177e4SLinus Torvalds return(err); 3291da177e4SLinus Torvalds } 3301da177e4SLinus Torvalds 3311da177e4SLinus Torvalds n = parse_unit(&str); 3321da177e4SLinus Torvalds if(n < 0){ 3331da177e4SLinus Torvalds printk(KERN_ERR "ubd_setup : couldn't parse unit number " 3341da177e4SLinus Torvalds "'%s'\n", str); 3351da177e4SLinus Torvalds return(1); 3361da177e4SLinus Torvalds } 3371da177e4SLinus Torvalds if(n >= MAX_DEV){ 3381da177e4SLinus Torvalds printk(KERN_ERR "ubd_setup : index %d out of range " 3391da177e4SLinus Torvalds "(%d devices, from 0 to %d)\n", n, MAX_DEV, MAX_DEV - 1); 3401da177e4SLinus Torvalds return(1); 3411da177e4SLinus Torvalds } 3421da177e4SLinus Torvalds 3431da177e4SLinus Torvalds err = 1; 3441da177e4SLinus Torvalds spin_lock(&ubd_lock); 3451da177e4SLinus Torvalds 3461da177e4SLinus Torvalds dev = &ubd_dev[n]; 3471da177e4SLinus Torvalds if(dev->file != NULL){ 3481da177e4SLinus Torvalds printk(KERN_ERR "ubd_setup : device already configured\n"); 3491da177e4SLinus Torvalds goto out; 3501da177e4SLinus Torvalds } 3511da177e4SLinus Torvalds 3521da177e4SLinus Torvalds if (index_out) 3531da177e4SLinus Torvalds *index_out = n; 3541da177e4SLinus Torvalds 3556c29256cSJeff Dike for (i = 0; i < sizeof("rscd="); i++) { 3561da177e4SLinus Torvalds switch (*str) { 3571da177e4SLinus Torvalds case 'r': 3581da177e4SLinus Torvalds flags.w = 0; 3591da177e4SLinus Torvalds break; 3601da177e4SLinus Torvalds case 's': 3611da177e4SLinus Torvalds flags.s = 1; 3621da177e4SLinus Torvalds break; 3631da177e4SLinus Torvalds case 'd': 3641da177e4SLinus Torvalds dev->no_cow = 1; 3651da177e4SLinus Torvalds break; 3666c29256cSJeff Dike case 'c': 3676c29256cSJeff Dike dev->shared = 1; 3686c29256cSJeff Dike break; 3691da177e4SLinus Torvalds case '=': 3701da177e4SLinus Torvalds str++; 3711da177e4SLinus Torvalds goto break_loop; 3721da177e4SLinus Torvalds default: 3736c29256cSJeff Dike printk(KERN_ERR "ubd_setup : Expected '=' or flag letter (r, s, c, or d)\n"); 3741da177e4SLinus Torvalds goto out; 3751da177e4SLinus Torvalds } 3761da177e4SLinus Torvalds str++; 3771da177e4SLinus Torvalds } 3781da177e4SLinus Torvalds 3791da177e4SLinus Torvalds if (*str == '=') 3801da177e4SLinus Torvalds printk(KERN_ERR "ubd_setup : Too many flags specified\n"); 3811da177e4SLinus Torvalds else 3821da177e4SLinus Torvalds printk(KERN_ERR "ubd_setup : Expected '='\n"); 3831da177e4SLinus Torvalds goto out; 3841da177e4SLinus Torvalds 3851da177e4SLinus Torvalds break_loop: 3861da177e4SLinus Torvalds err = 0; 3871da177e4SLinus Torvalds backing_file = strchr(str, ','); 3881da177e4SLinus Torvalds 3891da177e4SLinus Torvalds if (!backing_file) { 3901da177e4SLinus Torvalds backing_file = strchr(str, ':'); 3911da177e4SLinus Torvalds } 3921da177e4SLinus Torvalds 3931da177e4SLinus Torvalds if(backing_file){ 3941da177e4SLinus Torvalds if(dev->no_cow) 3951da177e4SLinus Torvalds printk(KERN_ERR "Can't specify both 'd' and a " 3961da177e4SLinus Torvalds "cow file\n"); 3971da177e4SLinus Torvalds else { 3981da177e4SLinus Torvalds *backing_file = '\0'; 3991da177e4SLinus Torvalds backing_file++; 4001da177e4SLinus Torvalds } 4011da177e4SLinus Torvalds } 4021da177e4SLinus Torvalds dev->file = str; 4031da177e4SLinus Torvalds dev->cow.file = backing_file; 4041da177e4SLinus Torvalds dev->boot_openflags = flags; 4051da177e4SLinus Torvalds out: 4061da177e4SLinus Torvalds spin_unlock(&ubd_lock); 4071da177e4SLinus Torvalds return(err); 4081da177e4SLinus Torvalds } 4091da177e4SLinus Torvalds 4101da177e4SLinus Torvalds static int ubd_setup(char *str) 4111da177e4SLinus Torvalds { 4121da177e4SLinus Torvalds ubd_setup_common(str, NULL); 4131da177e4SLinus Torvalds return(1); 4141da177e4SLinus Torvalds } 4151da177e4SLinus Torvalds 4161da177e4SLinus Torvalds __setup("ubd", ubd_setup); 4171da177e4SLinus Torvalds __uml_help(ubd_setup, 4181da177e4SLinus Torvalds "ubd<n><flags>=<filename>[(:|,)<filename2>]\n" 4191da177e4SLinus Torvalds " This is used to associate a device with a file in the underlying\n" 4201da177e4SLinus Torvalds " filesystem. When specifying two filenames, the first one is the\n" 4211da177e4SLinus Torvalds " COW name and the second is the backing file name. As separator you can\n" 4221da177e4SLinus Torvalds " use either a ':' or a ',': the first one allows writing things like;\n" 4231da177e4SLinus Torvalds " ubd0=~/Uml/root_cow:~/Uml/root_backing_file\n" 4241da177e4SLinus Torvalds " while with a ',' the shell would not expand the 2nd '~'.\n" 4251da177e4SLinus Torvalds " When using only one filename, UML will detect whether to thread it like\n" 4261da177e4SLinus Torvalds " a COW file or a backing file. To override this detection, add the 'd'\n" 4271da177e4SLinus Torvalds " flag:\n" 4281da177e4SLinus Torvalds " ubd0d=BackingFile\n" 4291da177e4SLinus Torvalds " Usually, there is a filesystem in the file, but \n" 4301da177e4SLinus Torvalds " that's not required. Swap devices containing swap files can be\n" 4311da177e4SLinus Torvalds " specified like this. Also, a file which doesn't contain a\n" 4321da177e4SLinus Torvalds " filesystem can have its contents read in the virtual \n" 4331da177e4SLinus Torvalds " machine by running 'dd' on the device. <n> must be in the range\n" 4341da177e4SLinus Torvalds " 0 to 7. Appending an 'r' to the number will cause that device\n" 4351da177e4SLinus Torvalds " to be mounted read-only. For example ubd1r=./ext_fs. Appending\n" 4361da177e4SLinus Torvalds " an 's' will cause data to be written to disk on the host immediately.\n\n" 4371da177e4SLinus Torvalds ); 4381da177e4SLinus Torvalds 4391da177e4SLinus Torvalds static int udb_setup(char *str) 4401da177e4SLinus Torvalds { 4411da177e4SLinus Torvalds printk("udb%s specified on command line is almost certainly a ubd -> " 4421da177e4SLinus Torvalds "udb TYPO\n", str); 4431da177e4SLinus Torvalds return(1); 4441da177e4SLinus Torvalds } 4451da177e4SLinus Torvalds 4461da177e4SLinus Torvalds __setup("udb", udb_setup); 4471da177e4SLinus Torvalds __uml_help(udb_setup, 4481da177e4SLinus Torvalds "udb\n" 4490894e27eSJeff Dike " This option is here solely to catch ubd -> udb typos, which can be\n" 4500894e27eSJeff Dike " to impossible to catch visually unless you specifically look for\n" 4510894e27eSJeff Dike " them. The only result of any option starting with 'udb' is an error\n" 4521da177e4SLinus Torvalds " in the boot output.\n\n" 4531da177e4SLinus Torvalds ); 4541da177e4SLinus Torvalds 4551da177e4SLinus Torvalds static int fakehd_set = 0; 4561da177e4SLinus Torvalds static int fakehd(char *str) 4571da177e4SLinus Torvalds { 4581da177e4SLinus Torvalds printk(KERN_INFO "fakehd : Changing ubd name to \"hd\".\n"); 4591da177e4SLinus Torvalds fakehd_set = 1; 4601da177e4SLinus Torvalds return 1; 4611da177e4SLinus Torvalds } 4621da177e4SLinus Torvalds 4631da177e4SLinus Torvalds __setup("fakehd", fakehd); 4641da177e4SLinus Torvalds __uml_help(fakehd, 4651da177e4SLinus Torvalds "fakehd\n" 4661da177e4SLinus Torvalds " Change the ubd device name to \"hd\".\n\n" 4671da177e4SLinus Torvalds ); 4681da177e4SLinus Torvalds 4691da177e4SLinus Torvalds static void do_ubd_request(request_queue_t * q); 47091acb21fSJeff Dike 47191acb21fSJeff Dike /* Only changed by ubd_init, which is an initcall. */ 47291acb21fSJeff Dike int thread_fd = -1; 4731da177e4SLinus Torvalds 4741da177e4SLinus Torvalds /* Changed by ubd_handler, which is serialized because interrupts only 4751da177e4SLinus Torvalds * happen on CPU 0. 4761da177e4SLinus Torvalds */ 4771da177e4SLinus Torvalds int intr_count = 0; 4781da177e4SLinus Torvalds 4791da177e4SLinus Torvalds /* call ubd_finish if you need to serialize */ 48091acb21fSJeff Dike static void __ubd_finish(struct request *req, int error) 4811da177e4SLinus Torvalds { 48291acb21fSJeff Dike int nsect; 48391acb21fSJeff Dike 48491acb21fSJeff Dike if(error){ 48591acb21fSJeff Dike end_request(req, 0); 4861da177e4SLinus Torvalds return; 4871da177e4SLinus Torvalds } 48891acb21fSJeff Dike nsect = req->current_nr_sectors; 48991acb21fSJeff Dike req->sector += nsect; 49091acb21fSJeff Dike req->buffer += nsect << 9; 49191acb21fSJeff Dike req->errors = 0; 49291acb21fSJeff Dike req->nr_sectors -= nsect; 49391acb21fSJeff Dike req->current_nr_sectors = 0; 49491acb21fSJeff Dike end_request(req, 1); 4951da177e4SLinus Torvalds } 4961da177e4SLinus Torvalds 49791acb21fSJeff Dike static inline void ubd_finish(struct request *req, int error) 49809ace81cSJeff Dike { 49909ace81cSJeff Dike spin_lock(&ubd_io_lock); 50091acb21fSJeff Dike __ubd_finish(req, error); 50109ace81cSJeff Dike spin_unlock(&ubd_io_lock); 50209ace81cSJeff Dike } 50309ace81cSJeff Dike 50491acb21fSJeff Dike /* Called without ubd_io_lock held */ 50591acb21fSJeff Dike static void ubd_handler(void) 50691acb21fSJeff Dike { 50791acb21fSJeff Dike struct io_thread_req req; 50891acb21fSJeff Dike struct request *rq = elv_next_request(ubd_queue); 50991acb21fSJeff Dike int n; 51009ace81cSJeff Dike 51191acb21fSJeff Dike do_ubd = NULL; 51291acb21fSJeff Dike intr_count++; 51391acb21fSJeff Dike n = os_read_file(thread_fd, &req, sizeof(req)); 51491acb21fSJeff Dike if(n != sizeof(req)){ 51591acb21fSJeff Dike printk(KERN_ERR "Pid %d - spurious interrupt in ubd_handler, " 51691acb21fSJeff Dike "err = %d\n", os_getpid(), -n); 51791acb21fSJeff Dike spin_lock(&ubd_io_lock); 51891acb21fSJeff Dike end_request(rq, 0); 51991acb21fSJeff Dike spin_unlock(&ubd_io_lock); 52091acb21fSJeff Dike return; 52191acb21fSJeff Dike } 52209ace81cSJeff Dike 52391acb21fSJeff Dike ubd_finish(rq, req.error); 52491acb21fSJeff Dike reactivate_fd(thread_fd, UBD_IRQ); 52591acb21fSJeff Dike do_ubd_request(ubd_queue); 52691acb21fSJeff Dike } 52709ace81cSJeff Dike 5281da177e4SLinus Torvalds static irqreturn_t ubd_intr(int irq, void *dev, struct pt_regs *unused) 5291da177e4SLinus Torvalds { 53091acb21fSJeff Dike ubd_handler(); 5311da177e4SLinus Torvalds return(IRQ_HANDLED); 5321da177e4SLinus Torvalds } 5331da177e4SLinus Torvalds 53491acb21fSJeff Dike /* Only changed by ubd_init, which is an initcall. */ 53591acb21fSJeff Dike static int io_pid = -1; 53691acb21fSJeff Dike 53791acb21fSJeff Dike void kill_io_thread(void) 53891acb21fSJeff Dike { 53991acb21fSJeff Dike if(io_pid != -1) 54091acb21fSJeff Dike os_kill_process(io_pid, 1); 54191acb21fSJeff Dike } 54291acb21fSJeff Dike 54391acb21fSJeff Dike __uml_exitcall(kill_io_thread); 54491acb21fSJeff Dike 5451da177e4SLinus Torvalds static int ubd_file_size(struct ubd *dev, __u64 *size_out) 5461da177e4SLinus Torvalds { 5471da177e4SLinus Torvalds char *file; 5481da177e4SLinus Torvalds 5491da177e4SLinus Torvalds file = dev->cow.file ? dev->cow.file : dev->file; 5501da177e4SLinus Torvalds return(os_file_size(file, size_out)); 5511da177e4SLinus Torvalds } 5521da177e4SLinus Torvalds 5531da177e4SLinus Torvalds static void ubd_close(struct ubd *dev) 5541da177e4SLinus Torvalds { 5551da177e4SLinus Torvalds os_close_file(dev->fd); 5561da177e4SLinus Torvalds if(dev->cow.file == NULL) 5571da177e4SLinus Torvalds return; 5581da177e4SLinus Torvalds 5591da177e4SLinus Torvalds os_close_file(dev->cow.fd); 5601da177e4SLinus Torvalds vfree(dev->cow.bitmap); 5611da177e4SLinus Torvalds dev->cow.bitmap = NULL; 5621da177e4SLinus Torvalds } 5631da177e4SLinus Torvalds 5641da177e4SLinus Torvalds static int ubd_open_dev(struct ubd *dev) 5651da177e4SLinus Torvalds { 5661da177e4SLinus Torvalds struct openflags flags; 5671da177e4SLinus Torvalds char **back_ptr; 5681da177e4SLinus Torvalds int err, create_cow, *create_ptr; 5691da177e4SLinus Torvalds 5701da177e4SLinus Torvalds dev->openflags = dev->boot_openflags; 5711da177e4SLinus Torvalds create_cow = 0; 5721da177e4SLinus Torvalds create_ptr = (dev->cow.file != NULL) ? &create_cow : NULL; 5731da177e4SLinus Torvalds back_ptr = dev->no_cow ? NULL : &dev->cow.file; 5746c29256cSJeff Dike dev->fd = open_ubd_file(dev->file, &dev->openflags, dev->shared, 5756c29256cSJeff Dike back_ptr, &dev->cow.bitmap_offset, 5766c29256cSJeff Dike &dev->cow.bitmap_len, &dev->cow.data_offset, 5776c29256cSJeff Dike create_ptr); 5781da177e4SLinus Torvalds 5791da177e4SLinus Torvalds if((dev->fd == -ENOENT) && create_cow){ 5801da177e4SLinus Torvalds dev->fd = create_cow_file(dev->file, dev->cow.file, 5811da177e4SLinus Torvalds dev->openflags, 1 << 9, PAGE_SIZE, 5821da177e4SLinus Torvalds &dev->cow.bitmap_offset, 5831da177e4SLinus Torvalds &dev->cow.bitmap_len, 5841da177e4SLinus Torvalds &dev->cow.data_offset); 5851da177e4SLinus Torvalds if(dev->fd >= 0){ 5861da177e4SLinus Torvalds printk(KERN_INFO "Creating \"%s\" as COW file for " 5871da177e4SLinus Torvalds "\"%s\"\n", dev->file, dev->cow.file); 5881da177e4SLinus Torvalds } 5891da177e4SLinus Torvalds } 5901da177e4SLinus Torvalds 5911da177e4SLinus Torvalds if(dev->fd < 0){ 5921da177e4SLinus Torvalds printk("Failed to open '%s', errno = %d\n", dev->file, 5931da177e4SLinus Torvalds -dev->fd); 5941da177e4SLinus Torvalds return(dev->fd); 5951da177e4SLinus Torvalds } 5961da177e4SLinus Torvalds 5971da177e4SLinus Torvalds if(dev->cow.file != NULL){ 5981da177e4SLinus Torvalds err = -ENOMEM; 5991da177e4SLinus Torvalds dev->cow.bitmap = (void *) vmalloc(dev->cow.bitmap_len); 6001da177e4SLinus Torvalds if(dev->cow.bitmap == NULL){ 6011da177e4SLinus Torvalds printk(KERN_ERR "Failed to vmalloc COW bitmap\n"); 6021da177e4SLinus Torvalds goto error; 6031da177e4SLinus Torvalds } 6041da177e4SLinus Torvalds flush_tlb_kernel_vm(); 6051da177e4SLinus Torvalds 6061da177e4SLinus Torvalds err = read_cow_bitmap(dev->fd, dev->cow.bitmap, 6071da177e4SLinus Torvalds dev->cow.bitmap_offset, 6081da177e4SLinus Torvalds dev->cow.bitmap_len); 6091da177e4SLinus Torvalds if(err < 0) 6101da177e4SLinus Torvalds goto error; 6111da177e4SLinus Torvalds 6121da177e4SLinus Torvalds flags = dev->openflags; 6131da177e4SLinus Torvalds flags.w = 0; 6146c29256cSJeff Dike err = open_ubd_file(dev->cow.file, &flags, dev->shared, NULL, 6156c29256cSJeff Dike NULL, NULL, NULL, NULL); 6161da177e4SLinus Torvalds if(err < 0) goto error; 6171da177e4SLinus Torvalds dev->cow.fd = err; 6181da177e4SLinus Torvalds } 6191da177e4SLinus Torvalds return(0); 6201da177e4SLinus Torvalds error: 6211da177e4SLinus Torvalds os_close_file(dev->fd); 6221da177e4SLinus Torvalds return(err); 6231da177e4SLinus Torvalds } 6241da177e4SLinus Torvalds 6251da177e4SLinus Torvalds static int ubd_new_disk(int major, u64 size, int unit, 6261da177e4SLinus Torvalds struct gendisk **disk_out) 6271da177e4SLinus Torvalds 6281da177e4SLinus Torvalds { 6291da177e4SLinus Torvalds struct gendisk *disk; 6301da177e4SLinus Torvalds 6311da177e4SLinus Torvalds disk = alloc_disk(1 << UBD_SHIFT); 6321da177e4SLinus Torvalds if(disk == NULL) 6331da177e4SLinus Torvalds return(-ENOMEM); 6341da177e4SLinus Torvalds 6351da177e4SLinus Torvalds disk->major = major; 6361da177e4SLinus Torvalds disk->first_minor = unit << UBD_SHIFT; 6371da177e4SLinus Torvalds disk->fops = &ubd_blops; 6381da177e4SLinus Torvalds set_capacity(disk, size / 512); 639ce7b0f46SGreg Kroah-Hartman if(major == MAJOR_NR) 6401da177e4SLinus Torvalds sprintf(disk->disk_name, "ubd%c", 'a' + unit); 641ce7b0f46SGreg Kroah-Hartman else 6421da177e4SLinus Torvalds sprintf(disk->disk_name, "ubd_fake%d", unit); 6431da177e4SLinus Torvalds 6441da177e4SLinus Torvalds /* sysfs register (not for ide fake devices) */ 6451da177e4SLinus Torvalds if (major == MAJOR_NR) { 6461da177e4SLinus Torvalds ubd_dev[unit].pdev.id = unit; 6471da177e4SLinus Torvalds ubd_dev[unit].pdev.name = DRIVER_NAME; 6481da177e4SLinus Torvalds platform_device_register(&ubd_dev[unit].pdev); 6491da177e4SLinus Torvalds disk->driverfs_dev = &ubd_dev[unit].pdev.dev; 6501da177e4SLinus Torvalds } 6511da177e4SLinus Torvalds 6521da177e4SLinus Torvalds disk->private_data = &ubd_dev[unit]; 6531da177e4SLinus Torvalds disk->queue = ubd_queue; 6541da177e4SLinus Torvalds add_disk(disk); 6551da177e4SLinus Torvalds 6561da177e4SLinus Torvalds *disk_out = disk; 6571da177e4SLinus Torvalds return 0; 6581da177e4SLinus Torvalds } 6591da177e4SLinus Torvalds 6601da177e4SLinus Torvalds #define ROUND_BLOCK(n) ((n + ((1 << 9) - 1)) & (-1 << 9)) 6611da177e4SLinus Torvalds 6621da177e4SLinus Torvalds static int ubd_add(int n) 6631da177e4SLinus Torvalds { 6641da177e4SLinus Torvalds struct ubd *dev = &ubd_dev[n]; 6651da177e4SLinus Torvalds int err; 6661da177e4SLinus Torvalds 667ec7cf783SJeff Dike err = -ENODEV; 6681da177e4SLinus Torvalds if(dev->file == NULL) 669ec7cf783SJeff Dike goto out; 6701da177e4SLinus Torvalds 6711da177e4SLinus Torvalds err = ubd_file_size(dev, &dev->size); 6721da177e4SLinus Torvalds if(err < 0) 67380c13749SJeff Dike goto out; 6741da177e4SLinus Torvalds 6751da177e4SLinus Torvalds dev->size = ROUND_BLOCK(dev->size); 6761da177e4SLinus Torvalds 6771da177e4SLinus Torvalds err = ubd_new_disk(MAJOR_NR, dev->size, n, &ubd_gendisk[n]); 6781da177e4SLinus Torvalds if(err) 67980c13749SJeff Dike goto out; 6801da177e4SLinus Torvalds 6811da177e4SLinus Torvalds if(fake_major != MAJOR_NR) 6821da177e4SLinus Torvalds ubd_new_disk(fake_major, dev->size, n, 6831da177e4SLinus Torvalds &fake_gendisk[n]); 6841da177e4SLinus Torvalds 6851da177e4SLinus Torvalds /* perhaps this should also be under the "if (fake_major)" above */ 6861da177e4SLinus Torvalds /* using the fake_disk->disk_name and also the fakehd_set name */ 6871da177e4SLinus Torvalds if (fake_ide) 6881da177e4SLinus Torvalds make_ide_entries(ubd_gendisk[n]->disk_name); 6891da177e4SLinus Torvalds 690ec7cf783SJeff Dike err = 0; 691ec7cf783SJeff Dike out: 692ec7cf783SJeff Dike return err; 6931da177e4SLinus Torvalds } 6941da177e4SLinus Torvalds 6951da177e4SLinus Torvalds static int ubd_config(char *str) 6961da177e4SLinus Torvalds { 6971da177e4SLinus Torvalds int n, err; 6981da177e4SLinus Torvalds 699970d6e3aSJeff Dike str = kstrdup(str, GFP_KERNEL); 7001da177e4SLinus Torvalds if(str == NULL){ 7011da177e4SLinus Torvalds printk(KERN_ERR "ubd_config failed to strdup string\n"); 7021da177e4SLinus Torvalds return(1); 7031da177e4SLinus Torvalds } 7041da177e4SLinus Torvalds err = ubd_setup_common(str, &n); 7051da177e4SLinus Torvalds if(err){ 7061da177e4SLinus Torvalds kfree(str); 7071da177e4SLinus Torvalds return(-1); 7081da177e4SLinus Torvalds } 7091da177e4SLinus Torvalds if(n == -1) return(0); 7101da177e4SLinus Torvalds 7111da177e4SLinus Torvalds spin_lock(&ubd_lock); 7121da177e4SLinus Torvalds err = ubd_add(n); 7131da177e4SLinus Torvalds if(err) 7141da177e4SLinus Torvalds ubd_dev[n].file = NULL; 7151da177e4SLinus Torvalds spin_unlock(&ubd_lock); 7161da177e4SLinus Torvalds 7171da177e4SLinus Torvalds return(err); 7181da177e4SLinus Torvalds } 7191da177e4SLinus Torvalds 7201da177e4SLinus Torvalds static int ubd_get_config(char *name, char *str, int size, char **error_out) 7211da177e4SLinus Torvalds { 7221da177e4SLinus Torvalds struct ubd *dev; 7231da177e4SLinus Torvalds int n, len = 0; 7241da177e4SLinus Torvalds 7251da177e4SLinus Torvalds n = parse_unit(&name); 7261da177e4SLinus Torvalds if((n >= MAX_DEV) || (n < 0)){ 7271da177e4SLinus Torvalds *error_out = "ubd_get_config : device number out of range"; 7281da177e4SLinus Torvalds return(-1); 7291da177e4SLinus Torvalds } 7301da177e4SLinus Torvalds 7311da177e4SLinus Torvalds dev = &ubd_dev[n]; 7321da177e4SLinus Torvalds spin_lock(&ubd_lock); 7331da177e4SLinus Torvalds 7341da177e4SLinus Torvalds if(dev->file == NULL){ 7351da177e4SLinus Torvalds CONFIG_CHUNK(str, size, len, "", 1); 7361da177e4SLinus Torvalds goto out; 7371da177e4SLinus Torvalds } 7381da177e4SLinus Torvalds 7391da177e4SLinus Torvalds CONFIG_CHUNK(str, size, len, dev->file, 0); 7401da177e4SLinus Torvalds 7411da177e4SLinus Torvalds if(dev->cow.file != NULL){ 7421da177e4SLinus Torvalds CONFIG_CHUNK(str, size, len, ",", 0); 7431da177e4SLinus Torvalds CONFIG_CHUNK(str, size, len, dev->cow.file, 1); 7441da177e4SLinus Torvalds } 7451da177e4SLinus Torvalds else CONFIG_CHUNK(str, size, len, "", 1); 7461da177e4SLinus Torvalds 7471da177e4SLinus Torvalds out: 7481da177e4SLinus Torvalds spin_unlock(&ubd_lock); 7491da177e4SLinus Torvalds return(len); 7501da177e4SLinus Torvalds } 7511da177e4SLinus Torvalds 75229d56cfeSJeff Dike static int ubd_id(char **str, int *start_out, int *end_out) 75329d56cfeSJeff Dike { 75429d56cfeSJeff Dike int n; 75529d56cfeSJeff Dike 75629d56cfeSJeff Dike n = parse_unit(str); 75729d56cfeSJeff Dike *start_out = 0; 75829d56cfeSJeff Dike *end_out = MAX_DEV - 1; 75929d56cfeSJeff Dike return n; 76029d56cfeSJeff Dike } 76129d56cfeSJeff Dike 76229d56cfeSJeff Dike static int ubd_remove(int n) 7631da177e4SLinus Torvalds { 7641da177e4SLinus Torvalds struct ubd *dev; 76529d56cfeSJeff Dike int err = -ENODEV; 7661da177e4SLinus Torvalds 7671da177e4SLinus Torvalds spin_lock(&ubd_lock); 7681da177e4SLinus Torvalds 7691da177e4SLinus Torvalds if(ubd_gendisk[n] == NULL) 7701da177e4SLinus Torvalds goto out; 7711da177e4SLinus Torvalds 77229d56cfeSJeff Dike dev = &ubd_dev[n]; 77329d56cfeSJeff Dike 77429d56cfeSJeff Dike if(dev->file == NULL) 77529d56cfeSJeff Dike goto out; 77629d56cfeSJeff Dike 77729d56cfeSJeff Dike /* you cannot remove a open disk */ 77829d56cfeSJeff Dike err = -EBUSY; 77929d56cfeSJeff Dike if(dev->count > 0) 78029d56cfeSJeff Dike goto out; 78129d56cfeSJeff Dike 7821da177e4SLinus Torvalds del_gendisk(ubd_gendisk[n]); 7831da177e4SLinus Torvalds put_disk(ubd_gendisk[n]); 7841da177e4SLinus Torvalds ubd_gendisk[n] = NULL; 7851da177e4SLinus Torvalds 7861da177e4SLinus Torvalds if(fake_gendisk[n] != NULL){ 7871da177e4SLinus Torvalds del_gendisk(fake_gendisk[n]); 7881da177e4SLinus Torvalds put_disk(fake_gendisk[n]); 7891da177e4SLinus Torvalds fake_gendisk[n] = NULL; 7901da177e4SLinus Torvalds } 7911da177e4SLinus Torvalds 7921da177e4SLinus Torvalds platform_device_unregister(&dev->pdev); 7931da177e4SLinus Torvalds *dev = ((struct ubd) DEFAULT_UBD); 7941da177e4SLinus Torvalds err = 0; 7951da177e4SLinus Torvalds out: 7961da177e4SLinus Torvalds spin_unlock(&ubd_lock); 79729d56cfeSJeff Dike return err; 7981da177e4SLinus Torvalds } 7991da177e4SLinus Torvalds 8001da177e4SLinus Torvalds static struct mc_device ubd_mc = { 8011da177e4SLinus Torvalds .name = "ubd", 8021da177e4SLinus Torvalds .config = ubd_config, 8031da177e4SLinus Torvalds .get_config = ubd_get_config, 80429d56cfeSJeff Dike .id = ubd_id, 8051da177e4SLinus Torvalds .remove = ubd_remove, 8061da177e4SLinus Torvalds }; 8071da177e4SLinus Torvalds 8081da177e4SLinus Torvalds static int ubd_mc_init(void) 8091da177e4SLinus Torvalds { 8101da177e4SLinus Torvalds mconsole_register_dev(&ubd_mc); 8111da177e4SLinus Torvalds return 0; 8121da177e4SLinus Torvalds } 8131da177e4SLinus Torvalds 8141da177e4SLinus Torvalds __initcall(ubd_mc_init); 8151da177e4SLinus Torvalds 8163ae5eaecSRussell King static struct platform_driver ubd_driver = { 8173ae5eaecSRussell King .driver = { 8181da177e4SLinus Torvalds .name = DRIVER_NAME, 8193ae5eaecSRussell King }, 8201da177e4SLinus Torvalds }; 8211da177e4SLinus Torvalds 8221da177e4SLinus Torvalds int ubd_init(void) 8231da177e4SLinus Torvalds { 8241da177e4SLinus Torvalds int i; 8251da177e4SLinus Torvalds 8261da177e4SLinus Torvalds if (register_blkdev(MAJOR_NR, "ubd")) 8271da177e4SLinus Torvalds return -1; 8281da177e4SLinus Torvalds 8291da177e4SLinus Torvalds ubd_queue = blk_init_queue(do_ubd_request, &ubd_io_lock); 8301da177e4SLinus Torvalds if (!ubd_queue) { 8311da177e4SLinus Torvalds unregister_blkdev(MAJOR_NR, "ubd"); 8321da177e4SLinus Torvalds return -1; 8331da177e4SLinus Torvalds } 8341da177e4SLinus Torvalds 8351da177e4SLinus Torvalds if (fake_major != MAJOR_NR) { 8361da177e4SLinus Torvalds char name[sizeof("ubd_nnn\0")]; 8371da177e4SLinus Torvalds 8381da177e4SLinus Torvalds snprintf(name, sizeof(name), "ubd_%d", fake_major); 8391da177e4SLinus Torvalds if (register_blkdev(fake_major, "ubd")) 8401da177e4SLinus Torvalds return -1; 8411da177e4SLinus Torvalds } 8423ae5eaecSRussell King platform_driver_register(&ubd_driver); 8431da177e4SLinus Torvalds for (i = 0; i < MAX_DEV; i++) 8441da177e4SLinus Torvalds ubd_add(i); 8451da177e4SLinus Torvalds return 0; 8461da177e4SLinus Torvalds } 8471da177e4SLinus Torvalds 8481da177e4SLinus Torvalds late_initcall(ubd_init); 8491da177e4SLinus Torvalds 85091acb21fSJeff Dike int ubd_driver_init(void){ 85191acb21fSJeff Dike unsigned long stack; 85291acb21fSJeff Dike int err; 85391acb21fSJeff Dike 85491acb21fSJeff Dike /* Set by CONFIG_BLK_DEV_UBD_SYNC or ubd=sync.*/ 85591acb21fSJeff Dike if(global_openflags.s){ 85691acb21fSJeff Dike printk(KERN_INFO "ubd: Synchronous mode\n"); 85791acb21fSJeff Dike /* Letting ubd=sync be like using ubd#s= instead of ubd#= is 85891acb21fSJeff Dike * enough. So use anyway the io thread. */ 85991acb21fSJeff Dike } 86091acb21fSJeff Dike stack = alloc_stack(0, 0); 86191acb21fSJeff Dike io_pid = start_io_thread(stack + PAGE_SIZE - sizeof(void *), 86291acb21fSJeff Dike &thread_fd); 86391acb21fSJeff Dike if(io_pid < 0){ 86491acb21fSJeff Dike printk(KERN_ERR 86591acb21fSJeff Dike "ubd : Failed to start I/O thread (errno = %d) - " 86691acb21fSJeff Dike "falling back to synchronous I/O\n", -io_pid); 86791acb21fSJeff Dike io_pid = -1; 86891acb21fSJeff Dike return(0); 86991acb21fSJeff Dike } 87091acb21fSJeff Dike err = um_request_irq(UBD_IRQ, thread_fd, IRQ_READ, ubd_intr, 871bd6aa650SThomas Gleixner IRQF_DISABLED, "ubd", ubd_dev); 87291acb21fSJeff Dike if(err != 0) 87391acb21fSJeff Dike printk(KERN_ERR "um_request_irq failed - errno = %d\n", -err); 874f4c57a78SJeff Dike return 0; 87591acb21fSJeff Dike } 87691acb21fSJeff Dike 87791acb21fSJeff Dike device_initcall(ubd_driver_init); 87891acb21fSJeff Dike 8791da177e4SLinus Torvalds static int ubd_open(struct inode *inode, struct file *filp) 8801da177e4SLinus Torvalds { 8811da177e4SLinus Torvalds struct gendisk *disk = inode->i_bdev->bd_disk; 8821da177e4SLinus Torvalds struct ubd *dev = disk->private_data; 8831da177e4SLinus Torvalds int err = 0; 8841da177e4SLinus Torvalds 8851da177e4SLinus Torvalds if(dev->count == 0){ 8861da177e4SLinus Torvalds err = ubd_open_dev(dev); 8871da177e4SLinus Torvalds if(err){ 8881da177e4SLinus Torvalds printk(KERN_ERR "%s: Can't open \"%s\": errno = %d\n", 8891da177e4SLinus Torvalds disk->disk_name, dev->file, -err); 8901da177e4SLinus Torvalds goto out; 8911da177e4SLinus Torvalds } 8921da177e4SLinus Torvalds } 8931da177e4SLinus Torvalds dev->count++; 8942c49be99SPaolo 'Blaisorblade' Giarrusso set_disk_ro(disk, !dev->openflags.w); 8952c49be99SPaolo 'Blaisorblade' Giarrusso 8962c49be99SPaolo 'Blaisorblade' Giarrusso /* This should no more be needed. And it didn't work anyway to exclude 8972c49be99SPaolo 'Blaisorblade' Giarrusso * read-write remounting of filesystems.*/ 8982c49be99SPaolo 'Blaisorblade' Giarrusso /*if((filp->f_mode & FMODE_WRITE) && !dev->openflags.w){ 8991da177e4SLinus Torvalds if(--dev->count == 0) ubd_close(dev); 9001da177e4SLinus Torvalds err = -EROFS; 9012c49be99SPaolo 'Blaisorblade' Giarrusso }*/ 9021da177e4SLinus Torvalds out: 9031da177e4SLinus Torvalds return(err); 9041da177e4SLinus Torvalds } 9051da177e4SLinus Torvalds 9061da177e4SLinus Torvalds static int ubd_release(struct inode * inode, struct file * file) 9071da177e4SLinus Torvalds { 9081da177e4SLinus Torvalds struct gendisk *disk = inode->i_bdev->bd_disk; 9091da177e4SLinus Torvalds struct ubd *dev = disk->private_data; 9101da177e4SLinus Torvalds 9111da177e4SLinus Torvalds if(--dev->count == 0) 9121da177e4SLinus Torvalds ubd_close(dev); 9131da177e4SLinus Torvalds return(0); 9141da177e4SLinus Torvalds } 9151da177e4SLinus Torvalds 91691acb21fSJeff Dike static void cowify_bitmap(__u64 io_offset, int length, unsigned long *cow_mask, 91791acb21fSJeff Dike __u64 *cow_offset, unsigned long *bitmap, 91891acb21fSJeff Dike __u64 bitmap_offset, unsigned long *bitmap_words, 91991acb21fSJeff Dike __u64 bitmap_len) 9201da177e4SLinus Torvalds { 92191acb21fSJeff Dike __u64 sector = io_offset >> 9; 92291acb21fSJeff Dike int i, update_bitmap = 0; 9231da177e4SLinus Torvalds 92491acb21fSJeff Dike for(i = 0; i < length >> 9; i++){ 92591acb21fSJeff Dike if(cow_mask != NULL) 92691acb21fSJeff Dike ubd_set_bit(i, (unsigned char *) cow_mask); 92791acb21fSJeff Dike if(ubd_test_bit(sector + i, (unsigned char *) bitmap)) 92809ace81cSJeff Dike continue; 9291da177e4SLinus Torvalds 93091acb21fSJeff Dike update_bitmap = 1; 93191acb21fSJeff Dike ubd_set_bit(sector + i, (unsigned char *) bitmap); 9321da177e4SLinus Torvalds } 93391acb21fSJeff Dike 93491acb21fSJeff Dike if(!update_bitmap) 93591acb21fSJeff Dike return; 93691acb21fSJeff Dike 93791acb21fSJeff Dike *cow_offset = sector / (sizeof(unsigned long) * 8); 93891acb21fSJeff Dike 93991acb21fSJeff Dike /* This takes care of the case where we're exactly at the end of the 94091acb21fSJeff Dike * device, and *cow_offset + 1 is off the end. So, just back it up 94191acb21fSJeff Dike * by one word. Thanks to Lynn Kerby for the fix and James McMechan 94291acb21fSJeff Dike * for the original diagnosis. 94391acb21fSJeff Dike */ 94491acb21fSJeff Dike if(*cow_offset == ((bitmap_len + sizeof(unsigned long) - 1) / 94591acb21fSJeff Dike sizeof(unsigned long) - 1)) 94691acb21fSJeff Dike (*cow_offset)--; 94791acb21fSJeff Dike 94891acb21fSJeff Dike bitmap_words[0] = bitmap[*cow_offset]; 94991acb21fSJeff Dike bitmap_words[1] = bitmap[*cow_offset + 1]; 95091acb21fSJeff Dike 95191acb21fSJeff Dike *cow_offset *= sizeof(unsigned long); 95291acb21fSJeff Dike *cow_offset += bitmap_offset; 95391acb21fSJeff Dike } 95491acb21fSJeff Dike 95591acb21fSJeff Dike static void cowify_req(struct io_thread_req *req, unsigned long *bitmap, 95691acb21fSJeff Dike __u64 bitmap_offset, __u64 bitmap_len) 95791acb21fSJeff Dike { 95891acb21fSJeff Dike __u64 sector = req->offset >> 9; 95991acb21fSJeff Dike int i; 96091acb21fSJeff Dike 96191acb21fSJeff Dike if(req->length > (sizeof(req->sector_mask) * 8) << 9) 96291acb21fSJeff Dike panic("Operation too long"); 96391acb21fSJeff Dike 96491acb21fSJeff Dike if(req->op == UBD_READ) { 96591acb21fSJeff Dike for(i = 0; i < req->length >> 9; i++){ 96691acb21fSJeff Dike if(ubd_test_bit(sector + i, (unsigned char *) bitmap)) 96791acb21fSJeff Dike ubd_set_bit(i, (unsigned char *) 96891acb21fSJeff Dike &req->sector_mask); 96991acb21fSJeff Dike } 97091acb21fSJeff Dike } 97191acb21fSJeff Dike else cowify_bitmap(req->offset, req->length, &req->sector_mask, 97291acb21fSJeff Dike &req->cow_offset, bitmap, bitmap_offset, 97391acb21fSJeff Dike req->bitmap_words, bitmap_len); 9741da177e4SLinus Torvalds } 9751da177e4SLinus Torvalds 9761da177e4SLinus Torvalds /* Called with ubd_io_lock held */ 97791acb21fSJeff Dike static int prepare_request(struct request *req, struct io_thread_req *io_req) 9781da177e4SLinus Torvalds { 9791da177e4SLinus Torvalds struct gendisk *disk = req->rq_disk; 9801da177e4SLinus Torvalds struct ubd *dev = disk->private_data; 98191acb21fSJeff Dike __u64 offset; 98291acb21fSJeff Dike int len; 98391acb21fSJeff Dike 98491acb21fSJeff Dike if(req->rq_status == RQ_INACTIVE) return(1); 9851da177e4SLinus Torvalds 9862c49be99SPaolo 'Blaisorblade' Giarrusso /* This should be impossible now */ 9871da177e4SLinus Torvalds if((rq_data_dir(req) == WRITE) && !dev->openflags.w){ 9881da177e4SLinus Torvalds printk("Write attempted on readonly ubd device %s\n", 9891da177e4SLinus Torvalds disk->disk_name); 99091acb21fSJeff Dike end_request(req, 0); 9911da177e4SLinus Torvalds return(1); 9921da177e4SLinus Torvalds } 9931da177e4SLinus Torvalds 99491acb21fSJeff Dike offset = ((__u64) req->sector) << 9; 99591acb21fSJeff Dike len = req->current_nr_sectors << 9; 99691acb21fSJeff Dike 9971da177e4SLinus Torvalds io_req->fds[0] = (dev->cow.file != NULL) ? dev->cow.fd : dev->fd; 9981da177e4SLinus Torvalds io_req->fds[1] = dev->fd; 99991acb21fSJeff Dike io_req->cow_offset = -1; 10001da177e4SLinus Torvalds io_req->offset = offset; 10011da177e4SLinus Torvalds io_req->length = len; 10021da177e4SLinus Torvalds io_req->error = 0; 100391acb21fSJeff Dike io_req->sector_mask = 0; 100491acb21fSJeff Dike 100591acb21fSJeff Dike io_req->op = (rq_data_dir(req) == READ) ? UBD_READ : UBD_WRITE; 10061da177e4SLinus Torvalds io_req->offsets[0] = 0; 10071da177e4SLinus Torvalds io_req->offsets[1] = dev->cow.data_offset; 100891acb21fSJeff Dike io_req->buffer = req->buffer; 10091da177e4SLinus Torvalds io_req->sectorsize = 1 << 9; 10101da177e4SLinus Torvalds 101191acb21fSJeff Dike if(dev->cow.file != NULL) 101291acb21fSJeff Dike cowify_req(io_req, dev->cow.bitmap, dev->cow.bitmap_offset, 101391acb21fSJeff Dike dev->cow.bitmap_len); 101491acb21fSJeff Dike 10151da177e4SLinus Torvalds return(0); 10161da177e4SLinus Torvalds } 10171da177e4SLinus Torvalds 10181da177e4SLinus Torvalds /* Called with ubd_io_lock held */ 10191da177e4SLinus Torvalds static void do_ubd_request(request_queue_t *q) 10201da177e4SLinus Torvalds { 10211da177e4SLinus Torvalds struct io_thread_req io_req; 10221da177e4SLinus Torvalds struct request *req; 102391acb21fSJeff Dike int err, n; 10241da177e4SLinus Torvalds 102591acb21fSJeff Dike if(thread_fd == -1){ 102609ace81cSJeff Dike while((req = elv_next_request(q)) != NULL){ 102791acb21fSJeff Dike err = prepare_request(req, &io_req); 102891acb21fSJeff Dike if(!err){ 102991acb21fSJeff Dike do_io(&io_req); 103091acb21fSJeff Dike __ubd_finish(req, io_req.error); 10311da177e4SLinus Torvalds } 10321da177e4SLinus Torvalds } 103391acb21fSJeff Dike } 103491acb21fSJeff Dike else { 103591acb21fSJeff Dike if(do_ubd || (req = elv_next_request(q)) == NULL) 103691acb21fSJeff Dike return; 103791acb21fSJeff Dike err = prepare_request(req, &io_req); 103891acb21fSJeff Dike if(!err){ 103991acb21fSJeff Dike do_ubd = ubd_handler; 104091acb21fSJeff Dike n = os_write_file(thread_fd, (char *) &io_req, 104191acb21fSJeff Dike sizeof(io_req)); 104291acb21fSJeff Dike if(n != sizeof(io_req)) 104391acb21fSJeff Dike printk("write to io thread failed, " 104491acb21fSJeff Dike "errno = %d\n", -n); 104591acb21fSJeff Dike } 104691acb21fSJeff Dike } 10471da177e4SLinus Torvalds } 10481da177e4SLinus Torvalds 1049a885c8c4SChristoph Hellwig static int ubd_getgeo(struct block_device *bdev, struct hd_geometry *geo) 1050a885c8c4SChristoph Hellwig { 1051a885c8c4SChristoph Hellwig struct ubd *dev = bdev->bd_disk->private_data; 1052a885c8c4SChristoph Hellwig 1053a885c8c4SChristoph Hellwig geo->heads = 128; 1054a885c8c4SChristoph Hellwig geo->sectors = 32; 1055a885c8c4SChristoph Hellwig geo->cylinders = dev->size / (128 * 32 * 512); 1056a885c8c4SChristoph Hellwig return 0; 1057a885c8c4SChristoph Hellwig } 1058a885c8c4SChristoph Hellwig 10591da177e4SLinus Torvalds static int ubd_ioctl(struct inode * inode, struct file * file, 10601da177e4SLinus Torvalds unsigned int cmd, unsigned long arg) 10611da177e4SLinus Torvalds { 10621da177e4SLinus Torvalds struct ubd *dev = inode->i_bdev->bd_disk->private_data; 10631da177e4SLinus Torvalds struct hd_driveid ubd_id = { 10641da177e4SLinus Torvalds .cyls = 0, 10651da177e4SLinus Torvalds .heads = 128, 10661da177e4SLinus Torvalds .sectors = 32, 10671da177e4SLinus Torvalds }; 10681da177e4SLinus Torvalds 10691da177e4SLinus Torvalds switch (cmd) { 10701da177e4SLinus Torvalds struct cdrom_volctrl volume; 10711da177e4SLinus Torvalds case HDIO_GET_IDENTITY: 10721da177e4SLinus Torvalds ubd_id.cyls = dev->size / (128 * 32 * 512); 10731da177e4SLinus Torvalds if(copy_to_user((char __user *) arg, (char *) &ubd_id, 10741da177e4SLinus Torvalds sizeof(ubd_id))) 10751da177e4SLinus Torvalds return(-EFAULT); 10761da177e4SLinus Torvalds return(0); 10771da177e4SLinus Torvalds 10781da177e4SLinus Torvalds case CDROMVOLREAD: 10791da177e4SLinus Torvalds if(copy_from_user(&volume, (char __user *) arg, sizeof(volume))) 10801da177e4SLinus Torvalds return(-EFAULT); 10811da177e4SLinus Torvalds volume.channel0 = 255; 10821da177e4SLinus Torvalds volume.channel1 = 255; 10831da177e4SLinus Torvalds volume.channel2 = 255; 10841da177e4SLinus Torvalds volume.channel3 = 255; 10851da177e4SLinus Torvalds if(copy_to_user((char __user *) arg, &volume, sizeof(volume))) 10861da177e4SLinus Torvalds return(-EFAULT); 10871da177e4SLinus Torvalds return(0); 10881da177e4SLinus Torvalds } 10891da177e4SLinus Torvalds return(-EINVAL); 10901da177e4SLinus Torvalds } 10911da177e4SLinus Torvalds 10924833aff7SPaolo 'Blaisorblade' Giarrusso static int path_requires_switch(char *from_cmdline, char *from_cow, char *cow) 10931da177e4SLinus Torvalds { 10941da177e4SLinus Torvalds struct uml_stat buf1, buf2; 10951da177e4SLinus Torvalds int err; 10961da177e4SLinus Torvalds 10974833aff7SPaolo 'Blaisorblade' Giarrusso if(from_cmdline == NULL) 10984833aff7SPaolo 'Blaisorblade' Giarrusso return 0; 10994833aff7SPaolo 'Blaisorblade' Giarrusso if(!strcmp(from_cmdline, from_cow)) 11004833aff7SPaolo 'Blaisorblade' Giarrusso return 0; 11011da177e4SLinus Torvalds 11021da177e4SLinus Torvalds err = os_stat_file(from_cmdline, &buf1); 11031da177e4SLinus Torvalds if(err < 0){ 11041da177e4SLinus Torvalds printk("Couldn't stat '%s', err = %d\n", from_cmdline, -err); 11054833aff7SPaolo 'Blaisorblade' Giarrusso return 0; 11061da177e4SLinus Torvalds } 11071da177e4SLinus Torvalds err = os_stat_file(from_cow, &buf2); 11081da177e4SLinus Torvalds if(err < 0){ 11091da177e4SLinus Torvalds printk("Couldn't stat '%s', err = %d\n", from_cow, -err); 11104833aff7SPaolo 'Blaisorblade' Giarrusso return 1; 11111da177e4SLinus Torvalds } 11121da177e4SLinus Torvalds if((buf1.ust_dev == buf2.ust_dev) && (buf1.ust_ino == buf2.ust_ino)) 11134833aff7SPaolo 'Blaisorblade' Giarrusso return 0; 11141da177e4SLinus Torvalds 11151da177e4SLinus Torvalds printk("Backing file mismatch - \"%s\" requested,\n" 11161da177e4SLinus Torvalds "\"%s\" specified in COW header of \"%s\"\n", 11171da177e4SLinus Torvalds from_cmdline, from_cow, cow); 11184833aff7SPaolo 'Blaisorblade' Giarrusso return 1; 11191da177e4SLinus Torvalds } 11201da177e4SLinus Torvalds 11211da177e4SLinus Torvalds static int backing_file_mismatch(char *file, __u64 size, time_t mtime) 11221da177e4SLinus Torvalds { 11231da177e4SLinus Torvalds unsigned long modtime; 1124fe1db50cSPaolo 'Blaisorblade' Giarrusso unsigned long long actual; 11251da177e4SLinus Torvalds int err; 11261da177e4SLinus Torvalds 11271da177e4SLinus Torvalds err = os_file_modtime(file, &modtime); 11281da177e4SLinus Torvalds if(err < 0){ 11291da177e4SLinus Torvalds printk("Failed to get modification time of backing file " 11301da177e4SLinus Torvalds "\"%s\", err = %d\n", file, -err); 11311da177e4SLinus Torvalds return(err); 11321da177e4SLinus Torvalds } 11331da177e4SLinus Torvalds 11341da177e4SLinus Torvalds err = os_file_size(file, &actual); 11351da177e4SLinus Torvalds if(err < 0){ 11361da177e4SLinus Torvalds printk("Failed to get size of backing file \"%s\", " 11371da177e4SLinus Torvalds "err = %d\n", file, -err); 11381da177e4SLinus Torvalds return(err); 11391da177e4SLinus Torvalds } 11401da177e4SLinus Torvalds 11411da177e4SLinus Torvalds if(actual != size){ 11421da177e4SLinus Torvalds /*__u64 can be a long on AMD64 and with %lu GCC complains; so 11431da177e4SLinus Torvalds * the typecast.*/ 11441da177e4SLinus Torvalds printk("Size mismatch (%llu vs %llu) of COW header vs backing " 11451da177e4SLinus Torvalds "file\n", (unsigned long long) size, actual); 11461da177e4SLinus Torvalds return(-EINVAL); 11471da177e4SLinus Torvalds } 11481da177e4SLinus Torvalds if(modtime != mtime){ 11491da177e4SLinus Torvalds printk("mtime mismatch (%ld vs %ld) of COW header vs backing " 11501da177e4SLinus Torvalds "file\n", mtime, modtime); 11511da177e4SLinus Torvalds return(-EINVAL); 11521da177e4SLinus Torvalds } 11531da177e4SLinus Torvalds return(0); 11541da177e4SLinus Torvalds } 11551da177e4SLinus Torvalds 11561da177e4SLinus Torvalds int read_cow_bitmap(int fd, void *buf, int offset, int len) 11571da177e4SLinus Torvalds { 11581da177e4SLinus Torvalds int err; 11591da177e4SLinus Torvalds 11601da177e4SLinus Torvalds err = os_seek_file(fd, offset); 11611da177e4SLinus Torvalds if(err < 0) 11621da177e4SLinus Torvalds return(err); 11631da177e4SLinus Torvalds 11641da177e4SLinus Torvalds err = os_read_file(fd, buf, len); 11651da177e4SLinus Torvalds if(err < 0) 11661da177e4SLinus Torvalds return(err); 11671da177e4SLinus Torvalds 11681da177e4SLinus Torvalds return(0); 11691da177e4SLinus Torvalds } 11701da177e4SLinus Torvalds 11716c29256cSJeff Dike int open_ubd_file(char *file, struct openflags *openflags, int shared, 11721da177e4SLinus Torvalds char **backing_file_out, int *bitmap_offset_out, 11731da177e4SLinus Torvalds unsigned long *bitmap_len_out, int *data_offset_out, 11741da177e4SLinus Torvalds int *create_cow_out) 11751da177e4SLinus Torvalds { 11761da177e4SLinus Torvalds time_t mtime; 11771da177e4SLinus Torvalds unsigned long long size; 11781da177e4SLinus Torvalds __u32 version, align; 11791da177e4SLinus Torvalds char *backing_file; 11804833aff7SPaolo 'Blaisorblade' Giarrusso int fd, err, sectorsize, asked_switch, mode = 0644; 11811da177e4SLinus Torvalds 11821da177e4SLinus Torvalds fd = os_open_file(file, *openflags, mode); 11831da177e4SLinus Torvalds if (fd < 0) { 11841da177e4SLinus Torvalds if ((fd == -ENOENT) && (create_cow_out != NULL)) 11851da177e4SLinus Torvalds *create_cow_out = 1; 11861da177e4SLinus Torvalds if (!openflags->w || 1187a374a48fSPaolo 'Blaisorblade' Giarrusso ((fd != -EROFS) && (fd != -EACCES))) 1188a374a48fSPaolo 'Blaisorblade' Giarrusso return fd; 11891da177e4SLinus Torvalds openflags->w = 0; 11901da177e4SLinus Torvalds fd = os_open_file(file, *openflags, mode); 11911da177e4SLinus Torvalds if (fd < 0) 1192a374a48fSPaolo 'Blaisorblade' Giarrusso return fd; 11931da177e4SLinus Torvalds } 11941da177e4SLinus Torvalds 11956c29256cSJeff Dike if(shared) 11966c29256cSJeff Dike printk("Not locking \"%s\" on the host\n", file); 11976c29256cSJeff Dike else { 11981da177e4SLinus Torvalds err = os_lock_file(fd, openflags->w); 11991da177e4SLinus Torvalds if(err < 0){ 12001da177e4SLinus Torvalds printk("Failed to lock '%s', err = %d\n", file, -err); 12011da177e4SLinus Torvalds goto out_close; 12021da177e4SLinus Torvalds } 12036c29256cSJeff Dike } 12041da177e4SLinus Torvalds 1205d6e05edcSAndreas Mohr /* Successful return case! */ 1206a374a48fSPaolo 'Blaisorblade' Giarrusso if(backing_file_out == NULL) 1207a374a48fSPaolo 'Blaisorblade' Giarrusso return(fd); 12081da177e4SLinus Torvalds 12091da177e4SLinus Torvalds err = read_cow_header(file_reader, &fd, &version, &backing_file, &mtime, 12101da177e4SLinus Torvalds &size, §orsize, &align, bitmap_offset_out); 12111da177e4SLinus Torvalds if(err && (*backing_file_out != NULL)){ 12121da177e4SLinus Torvalds printk("Failed to read COW header from COW file \"%s\", " 12131da177e4SLinus Torvalds "errno = %d\n", file, -err); 12141da177e4SLinus Torvalds goto out_close; 12151da177e4SLinus Torvalds } 1216a374a48fSPaolo 'Blaisorblade' Giarrusso if(err) 1217a374a48fSPaolo 'Blaisorblade' Giarrusso return(fd); 12181da177e4SLinus Torvalds 12194833aff7SPaolo 'Blaisorblade' Giarrusso asked_switch = path_requires_switch(*backing_file_out, backing_file, file); 12201da177e4SLinus Torvalds 12214833aff7SPaolo 'Blaisorblade' Giarrusso /* Allow switching only if no mismatch. */ 12224833aff7SPaolo 'Blaisorblade' Giarrusso if (asked_switch && !backing_file_mismatch(*backing_file_out, size, mtime)) { 12231da177e4SLinus Torvalds printk("Switching backing file to '%s'\n", *backing_file_out); 12241da177e4SLinus Torvalds err = write_cow_header(file, fd, *backing_file_out, 12251da177e4SLinus Torvalds sectorsize, align, &size); 12261da177e4SLinus Torvalds if (err) { 12271da177e4SLinus Torvalds printk("Switch failed, errno = %d\n", -err); 12284833aff7SPaolo 'Blaisorblade' Giarrusso goto out_close; 12291da177e4SLinus Torvalds } 1230a374a48fSPaolo 'Blaisorblade' Giarrusso } else { 12311da177e4SLinus Torvalds *backing_file_out = backing_file; 12321da177e4SLinus Torvalds err = backing_file_mismatch(*backing_file_out, size, mtime); 1233a374a48fSPaolo 'Blaisorblade' Giarrusso if (err) 1234a374a48fSPaolo 'Blaisorblade' Giarrusso goto out_close; 12351da177e4SLinus Torvalds } 12361da177e4SLinus Torvalds 12371da177e4SLinus Torvalds cow_sizes(version, size, sectorsize, align, *bitmap_offset_out, 12381da177e4SLinus Torvalds bitmap_len_out, data_offset_out); 12391da177e4SLinus Torvalds 1240a374a48fSPaolo 'Blaisorblade' Giarrusso return fd; 12411da177e4SLinus Torvalds out_close: 12421da177e4SLinus Torvalds os_close_file(fd); 1243a374a48fSPaolo 'Blaisorblade' Giarrusso return err; 12441da177e4SLinus Torvalds } 12451da177e4SLinus Torvalds 12461da177e4SLinus Torvalds int create_cow_file(char *cow_file, char *backing_file, struct openflags flags, 12471da177e4SLinus Torvalds int sectorsize, int alignment, int *bitmap_offset_out, 12481da177e4SLinus Torvalds unsigned long *bitmap_len_out, int *data_offset_out) 12491da177e4SLinus Torvalds { 12501da177e4SLinus Torvalds int err, fd; 12511da177e4SLinus Torvalds 12521da177e4SLinus Torvalds flags.c = 1; 12536c29256cSJeff Dike fd = open_ubd_file(cow_file, &flags, 0, NULL, NULL, NULL, NULL, NULL); 12541da177e4SLinus Torvalds if(fd < 0){ 12551da177e4SLinus Torvalds err = fd; 12561da177e4SLinus Torvalds printk("Open of COW file '%s' failed, errno = %d\n", cow_file, 12571da177e4SLinus Torvalds -err); 12581da177e4SLinus Torvalds goto out; 12591da177e4SLinus Torvalds } 12601da177e4SLinus Torvalds 12611da177e4SLinus Torvalds err = init_cow_file(fd, cow_file, backing_file, sectorsize, alignment, 12621da177e4SLinus Torvalds bitmap_offset_out, bitmap_len_out, 12631da177e4SLinus Torvalds data_offset_out); 12641da177e4SLinus Torvalds if(!err) 12651da177e4SLinus Torvalds return(fd); 12661da177e4SLinus Torvalds os_close_file(fd); 12671da177e4SLinus Torvalds out: 12681da177e4SLinus Torvalds return(err); 12691da177e4SLinus Torvalds } 12701da177e4SLinus Torvalds 127191acb21fSJeff Dike static int update_bitmap(struct io_thread_req *req) 12721da177e4SLinus Torvalds { 127391acb21fSJeff Dike int n; 127491acb21fSJeff Dike 127591acb21fSJeff Dike if(req->cow_offset == -1) 127691acb21fSJeff Dike return(0); 127791acb21fSJeff Dike 127891acb21fSJeff Dike n = os_seek_file(req->fds[1], req->cow_offset); 127991acb21fSJeff Dike if(n < 0){ 128091acb21fSJeff Dike printk("do_io - bitmap lseek failed : err = %d\n", -n); 128191acb21fSJeff Dike return(1); 128291acb21fSJeff Dike } 128391acb21fSJeff Dike 128491acb21fSJeff Dike n = os_write_file(req->fds[1], &req->bitmap_words, 128591acb21fSJeff Dike sizeof(req->bitmap_words)); 128691acb21fSJeff Dike if(n != sizeof(req->bitmap_words)){ 128791acb21fSJeff Dike printk("do_io - bitmap update failed, err = %d fd = %d\n", -n, 128891acb21fSJeff Dike req->fds[1]); 128991acb21fSJeff Dike return(1); 129091acb21fSJeff Dike } 129191acb21fSJeff Dike 129291acb21fSJeff Dike return(0); 129391acb21fSJeff Dike } 129491acb21fSJeff Dike 129591acb21fSJeff Dike void do_io(struct io_thread_req *req) 129691acb21fSJeff Dike { 12971da177e4SLinus Torvalds char *buf; 129891acb21fSJeff Dike unsigned long len; 129991acb21fSJeff Dike int n, nsectors, start, end, bit; 130091acb21fSJeff Dike int err; 13011da177e4SLinus Torvalds __u64 off; 13021da177e4SLinus Torvalds 13031da177e4SLinus Torvalds nsectors = req->length / req->sectorsize; 13041da177e4SLinus Torvalds start = 0; 13051da177e4SLinus Torvalds do { 130691acb21fSJeff Dike bit = ubd_test_bit(start, (unsigned char *) &req->sector_mask); 13071da177e4SLinus Torvalds end = start; 13081da177e4SLinus Torvalds while((end < nsectors) && 130991acb21fSJeff Dike (ubd_test_bit(end, (unsigned char *) 131091acb21fSJeff Dike &req->sector_mask) == bit)) 13111da177e4SLinus Torvalds end++; 13121da177e4SLinus Torvalds 131391acb21fSJeff Dike off = req->offset + req->offsets[bit] + 13141da177e4SLinus Torvalds start * req->sectorsize; 13151da177e4SLinus Torvalds len = (end - start) * req->sectorsize; 13161da177e4SLinus Torvalds buf = &req->buffer[start * req->sectorsize]; 13171da177e4SLinus Torvalds 131891acb21fSJeff Dike err = os_seek_file(req->fds[bit], off); 131991acb21fSJeff Dike if(err < 0){ 132091acb21fSJeff Dike printk("do_io - lseek failed : err = %d\n", -err); 13211da177e4SLinus Torvalds req->error = 1; 13221da177e4SLinus Torvalds return; 13231da177e4SLinus Torvalds } 132491acb21fSJeff Dike if(req->op == UBD_READ){ 132591acb21fSJeff Dike n = 0; 132691acb21fSJeff Dike do { 132791acb21fSJeff Dike buf = &buf[n]; 132891acb21fSJeff Dike len -= n; 132991acb21fSJeff Dike n = os_read_file(req->fds[bit], buf, len); 133091acb21fSJeff Dike if (n < 0) { 133191acb21fSJeff Dike printk("do_io - read failed, err = %d " 133291acb21fSJeff Dike "fd = %d\n", -n, req->fds[bit]); 13331da177e4SLinus Torvalds req->error = 1; 13341da177e4SLinus Torvalds return; 13351da177e4SLinus Torvalds } 133691acb21fSJeff Dike } while((n < len) && (n != 0)); 133791acb21fSJeff Dike if (n < len) memset(&buf[n], 0, len - n); 133891acb21fSJeff Dike } else { 133991acb21fSJeff Dike n = os_write_file(req->fds[bit], buf, len); 134091acb21fSJeff Dike if(n != len){ 134191acb21fSJeff Dike printk("do_io - write failed err = %d " 134291acb21fSJeff Dike "fd = %d\n", -n, req->fds[bit]); 134391acb21fSJeff Dike req->error = 1; 134491acb21fSJeff Dike return; 134591acb21fSJeff Dike } 134691acb21fSJeff Dike } 13471da177e4SLinus Torvalds 13481da177e4SLinus Torvalds start = end; 13491da177e4SLinus Torvalds } while(start < nsectors); 135091acb21fSJeff Dike 135191acb21fSJeff Dike req->error = update_bitmap(req); 13521da177e4SLinus Torvalds } 135391acb21fSJeff Dike 135491acb21fSJeff Dike /* Changed in start_io_thread, which is serialized by being called only 135591acb21fSJeff Dike * from ubd_init, which is an initcall. 135691acb21fSJeff Dike */ 135791acb21fSJeff Dike int kernel_fd = -1; 135891acb21fSJeff Dike 135991acb21fSJeff Dike /* Only changed by the io thread */ 136091acb21fSJeff Dike int io_count = 0; 136191acb21fSJeff Dike 136291acb21fSJeff Dike int io_thread(void *arg) 136391acb21fSJeff Dike { 136491acb21fSJeff Dike struct io_thread_req req; 136591acb21fSJeff Dike int n; 136691acb21fSJeff Dike 136791acb21fSJeff Dike ignore_sigwinch_sig(); 136891acb21fSJeff Dike while(1){ 136991acb21fSJeff Dike n = os_read_file(kernel_fd, &req, sizeof(req)); 137091acb21fSJeff Dike if(n != sizeof(req)){ 137191acb21fSJeff Dike if(n < 0) 137291acb21fSJeff Dike printk("io_thread - read failed, fd = %d, " 137391acb21fSJeff Dike "err = %d\n", kernel_fd, -n); 137491acb21fSJeff Dike else { 137591acb21fSJeff Dike printk("io_thread - short read, fd = %d, " 137691acb21fSJeff Dike "length = %d\n", kernel_fd, n); 137791acb21fSJeff Dike } 137891acb21fSJeff Dike continue; 137991acb21fSJeff Dike } 138091acb21fSJeff Dike io_count++; 138191acb21fSJeff Dike do_io(&req); 138291acb21fSJeff Dike n = os_write_file(kernel_fd, &req, sizeof(req)); 138391acb21fSJeff Dike if(n != sizeof(req)) 138491acb21fSJeff Dike printk("io_thread - write failed, fd = %d, err = %d\n", 138591acb21fSJeff Dike kernel_fd, -n); 138691acb21fSJeff Dike } 138791acb21fSJeff Dike 13881b57e9c2SJeff Dike return 0; 13891b57e9c2SJeff Dike } 1390