1dbddf429SAlex Dewar // SPDX-License-Identifier: GPL-2.0 21da177e4SLinus Torvalds /* 3a41421edSAnton Ivanov * Copyright (C) 2018 Cambridge Greys Ltd 4f88f0bdfSAnton Ivanov * Copyright (C) 2015-2016 Anton Ivanov (aivanov@brocade.com) 51da177e4SLinus Torvalds * Copyright (C) 2000 Jeff Dike (jdike@karaya.com) 61da177e4SLinus Torvalds */ 71da177e4SLinus Torvalds 81da177e4SLinus Torvalds /* 2001-09-28...2002-04-17 91da177e4SLinus Torvalds * Partition stuff by James_McMechan@hotmail.com 101da177e4SLinus Torvalds * old style ubd by setting UBD_SHIFT to 0 111da177e4SLinus Torvalds * 2002-09-27...2002-10-18 massive tinkering for 2.5 121da177e4SLinus Torvalds * partitions have changed in 2.5 131da177e4SLinus Torvalds * 2003-01-29 more tinkering for 2.5.59-1 141da177e4SLinus Torvalds * This should now address the sysfs problems and has 151da177e4SLinus Torvalds * the symlink for devfs to allow for booting with 161da177e4SLinus Torvalds * the common /dev/ubd/discX/... names rather than 171da177e4SLinus Torvalds * only /dev/ubdN/discN this version also has lots of 181da177e4SLinus Torvalds * clean ups preparing for ubd-many. 191da177e4SLinus Torvalds * James McMechan 201da177e4SLinus Torvalds */ 211da177e4SLinus Torvalds 221da177e4SLinus Torvalds #define UBD_SHIFT 4 231da177e4SLinus Torvalds 248ea3c06aSAl Viro #include <linux/module.h> 258ea3c06aSAl Viro #include <linux/init.h> 268ea3c06aSAl Viro #include <linux/blkdev.h> 274e6da0feSRichard Weinberger #include <linux/blk-mq.h> 288ea3c06aSAl Viro #include <linux/ata.h> 298ea3c06aSAl Viro #include <linux/hdreg.h> 30*b81e0c23SChristoph Hellwig #include <linux/major.h> 318ea3c06aSAl Viro #include <linux/cdrom.h> 328ea3c06aSAl Viro #include <linux/proc_fs.h> 338ea3c06aSAl Viro #include <linux/seq_file.h> 348ea3c06aSAl Viro #include <linux/ctype.h> 358ea3c06aSAl Viro #include <linux/slab.h> 368ea3c06aSAl Viro #include <linux/vmalloc.h> 378ea3c06aSAl Viro #include <linux/platform_device.h> 388ea3c06aSAl Viro #include <linux/scatterlist.h> 398ea3c06aSAl Viro #include <asm/tlbflush.h> 4037185b33SAl Viro #include <kern_util.h> 411da177e4SLinus Torvalds #include "mconsole_kern.h" 4237185b33SAl Viro #include <init.h> 4337185b33SAl Viro #include <irq_kern.h> 448ea3c06aSAl Viro #include "ubd.h" 4537185b33SAl Viro #include <os.h> 461da177e4SLinus Torvalds #include "cow.h" 471da177e4SLinus Torvalds 48a43c8316SAnton Ivanov /* Max request size is determined by sector mask - 32K */ 49a43c8316SAnton Ivanov #define UBD_MAX_REQUEST (8 * sizeof(long)) 501da177e4SLinus Torvalds 51fc6b6a87SGabriel Krisman Bertazi struct io_desc { 52fc6b6a87SGabriel Krisman Bertazi char *buffer; 53fc6b6a87SGabriel Krisman Bertazi unsigned long length; 54fc6b6a87SGabriel Krisman Bertazi unsigned long sector_mask; 55fc6b6a87SGabriel Krisman Bertazi unsigned long long cow_offset; 56fc6b6a87SGabriel Krisman Bertazi unsigned long bitmap_words[2]; 57fc6b6a87SGabriel Krisman Bertazi }; 58fc6b6a87SGabriel Krisman Bertazi 591da177e4SLinus Torvalds struct io_thread_req { 6062f96cb0SJeff Dike struct request *req; 611da177e4SLinus Torvalds int fds[2]; 621da177e4SLinus Torvalds unsigned long offsets[2]; 631da177e4SLinus Torvalds unsigned long long offset; 641da177e4SLinus Torvalds int sectorsize; 651da177e4SLinus Torvalds int error; 66fc6b6a87SGabriel Krisman Bertazi 67fc6b6a87SGabriel Krisman Bertazi int desc_cnt; 68fc6b6a87SGabriel Krisman Bertazi /* io_desc has to be the last element of the struct */ 69fc6b6a87SGabriel Krisman Bertazi struct io_desc io_desc[]; 701da177e4SLinus Torvalds }; 711da177e4SLinus Torvalds 72f88f0bdfSAnton Ivanov 73f88f0bdfSAnton Ivanov static struct io_thread_req * (*irq_req_buffer)[]; 74f88f0bdfSAnton Ivanov static struct io_thread_req *irq_remainder; 75f88f0bdfSAnton Ivanov static int irq_remainder_size; 76f88f0bdfSAnton Ivanov 77f88f0bdfSAnton Ivanov static struct io_thread_req * (*io_req_buffer)[]; 78f88f0bdfSAnton Ivanov static struct io_thread_req *io_remainder; 79f88f0bdfSAnton Ivanov static int io_remainder_size; 80f88f0bdfSAnton Ivanov 81f88f0bdfSAnton Ivanov 82f88f0bdfSAnton Ivanov 8391acb21fSJeff Dike static inline int ubd_test_bit(__u64 bit, unsigned char *data) 841da177e4SLinus Torvalds { 851da177e4SLinus Torvalds __u64 n; 861da177e4SLinus Torvalds int bits, off; 871da177e4SLinus Torvalds 8891acb21fSJeff Dike bits = sizeof(data[0]) * 8; 891da177e4SLinus Torvalds n = bit / bits; 901da177e4SLinus Torvalds off = bit % bits; 91dc764e50SJeff Dike return (data[n] & (1 << off)) != 0; 921da177e4SLinus Torvalds } 931da177e4SLinus Torvalds 9491acb21fSJeff Dike static inline void ubd_set_bit(__u64 bit, unsigned char *data) 951da177e4SLinus Torvalds { 961da177e4SLinus Torvalds __u64 n; 971da177e4SLinus Torvalds int bits, off; 981da177e4SLinus Torvalds 9991acb21fSJeff Dike bits = sizeof(data[0]) * 8; 1001da177e4SLinus Torvalds n = bit / bits; 1011da177e4SLinus Torvalds off = bit % bits; 10291acb21fSJeff Dike data[n] |= (1 << off); 1031da177e4SLinus Torvalds } 1041da177e4SLinus Torvalds /*End stuff from ubd_user.h*/ 1051da177e4SLinus Torvalds 1061da177e4SLinus Torvalds #define DRIVER_NAME "uml-blkdev" 1071da177e4SLinus Torvalds 108d7fb2c38SPaolo 'Blaisorblade' Giarrusso static DEFINE_MUTEX(ubd_lock); 1099a181c58SArnd Bergmann static DEFINE_MUTEX(ubd_mutex); /* replaces BKL, might not be needed */ 1101da177e4SLinus Torvalds 111a625c998SAl Viro static int ubd_open(struct block_device *bdev, fmode_t mode); 112db2a144bSAl Viro static void ubd_release(struct gendisk *disk, fmode_t mode); 113a625c998SAl Viro static int ubd_ioctl(struct block_device *bdev, fmode_t mode, 1141da177e4SLinus Torvalds unsigned int cmd, unsigned long arg); 115a885c8c4SChristoph Hellwig static int ubd_getgeo(struct block_device *bdev, struct hd_geometry *geo); 1161da177e4SLinus Torvalds 11797d88ac8SPaolo 'Blaisorblade' Giarrusso #define MAX_DEV (16) 1181da177e4SLinus Torvalds 11983d5cde4SAlexey Dobriyan static const struct block_device_operations ubd_blops = { 1201da177e4SLinus Torvalds .owner = THIS_MODULE, 121a625c998SAl Viro .open = ubd_open, 122a625c998SAl Viro .release = ubd_release, 123a625c998SAl Viro .ioctl = ubd_ioctl, 124ab0cf1e4SArnd Bergmann .compat_ioctl = blkdev_compat_ptr_ioctl, 125a885c8c4SChristoph Hellwig .getgeo = ubd_getgeo, 1261da177e4SLinus Torvalds }; 1271da177e4SLinus Torvalds 1281da177e4SLinus Torvalds /* Protected by ubd_lock */ 1291da177e4SLinus Torvalds static struct gendisk *ubd_gendisk[MAX_DEV]; 1301da177e4SLinus Torvalds 1311da177e4SLinus Torvalds #ifdef CONFIG_BLK_DEV_UBD_SYNC 1321da177e4SLinus Torvalds #define OPEN_FLAGS ((struct openflags) { .r = 1, .w = 1, .s = 1, .c = 0, \ 1331da177e4SLinus Torvalds .cl = 1 }) 1341da177e4SLinus Torvalds #else 1351da177e4SLinus Torvalds #define OPEN_FLAGS ((struct openflags) { .r = 1, .w = 1, .s = 0, .c = 0, \ 1361da177e4SLinus Torvalds .cl = 1 }) 1371da177e4SLinus Torvalds #endif 1381da177e4SLinus Torvalds static struct openflags global_openflags = OPEN_FLAGS; 1391da177e4SLinus Torvalds 1401da177e4SLinus Torvalds struct cow { 1412a9d32f6SPaolo 'Blaisorblade' Giarrusso /* backing file name */ 1421da177e4SLinus Torvalds char *file; 1432a9d32f6SPaolo 'Blaisorblade' Giarrusso /* backing file fd */ 1441da177e4SLinus Torvalds int fd; 1451da177e4SLinus Torvalds unsigned long *bitmap; 1461da177e4SLinus Torvalds unsigned long bitmap_len; 1471da177e4SLinus Torvalds int bitmap_offset; 1481da177e4SLinus Torvalds int data_offset; 1491da177e4SLinus Torvalds }; 1501da177e4SLinus Torvalds 151a0044bdfSJeff Dike #define MAX_SG 64 152a0044bdfSJeff Dike 1531da177e4SLinus Torvalds struct ubd { 1542a9d32f6SPaolo 'Blaisorblade' Giarrusso /* name (and fd, below) of the file opened for writing, either the 1552a9d32f6SPaolo 'Blaisorblade' Giarrusso * backing or the cow file. */ 1561da177e4SLinus Torvalds char *file; 157ef3ba87cSChristopher Obbard char *serial; 1581da177e4SLinus Torvalds int count; 1591da177e4SLinus Torvalds int fd; 1601da177e4SLinus Torvalds __u64 size; 1611da177e4SLinus Torvalds struct openflags boot_openflags; 1621da177e4SLinus Torvalds struct openflags openflags; 16384e945e3SPaolo 'Blaisorblade' Giarrusso unsigned shared:1; 16484e945e3SPaolo 'Blaisorblade' Giarrusso unsigned no_cow:1; 16550109b5aSAnton Ivanov unsigned no_trim:1; 1661da177e4SLinus Torvalds struct cow cow; 1671da177e4SLinus Torvalds struct platform_device pdev; 16862f96cb0SJeff Dike struct request_queue *queue; 1694e6da0feSRichard Weinberger struct blk_mq_tag_set tag_set; 17062f96cb0SJeff Dike spinlock_t lock; 1714e6da0feSRichard Weinberger }; 1724e6da0feSRichard Weinberger 1731da177e4SLinus Torvalds #define DEFAULT_COW { \ 1741da177e4SLinus Torvalds .file = NULL, \ 1751da177e4SLinus Torvalds .fd = -1, \ 1761da177e4SLinus Torvalds .bitmap = NULL, \ 1771da177e4SLinus Torvalds .bitmap_offset = 0, \ 1781da177e4SLinus Torvalds .data_offset = 0, \ 1791da177e4SLinus Torvalds } 1801da177e4SLinus Torvalds 1811da177e4SLinus Torvalds #define DEFAULT_UBD { \ 1821da177e4SLinus Torvalds .file = NULL, \ 183ef3ba87cSChristopher Obbard .serial = NULL, \ 1841da177e4SLinus Torvalds .count = 0, \ 1851da177e4SLinus Torvalds .fd = -1, \ 1861da177e4SLinus Torvalds .size = -1, \ 1871da177e4SLinus Torvalds .boot_openflags = OPEN_FLAGS, \ 1881da177e4SLinus Torvalds .openflags = OPEN_FLAGS, \ 1891da177e4SLinus Torvalds .no_cow = 0, \ 19050109b5aSAnton Ivanov .no_trim = 0, \ 1916c29256cSJeff Dike .shared = 0, \ 1921da177e4SLinus Torvalds .cow = DEFAULT_COW, \ 19322e65004SThomas Gleixner .lock = __SPIN_LOCK_UNLOCKED(ubd_devs.lock), \ 1941da177e4SLinus Torvalds } 1951da177e4SLinus Torvalds 196b8831a1dSJeff Dike /* Protected by ubd_lock */ 1975dc62b1bSWANG Cong static struct ubd ubd_devs[MAX_DEV] = { [0 ... MAX_DEV - 1] = DEFAULT_UBD }; 1981da177e4SLinus Torvalds 1994e6da0feSRichard Weinberger static blk_status_t ubd_queue_rq(struct blk_mq_hw_ctx *hctx, 2004e6da0feSRichard Weinberger const struct blk_mq_queue_data *bd); 2014e6da0feSRichard Weinberger 2021da177e4SLinus Torvalds static int fake_ide_setup(char *str) 2031da177e4SLinus Torvalds { 2047eb90f7eSChristoph Hellwig pr_warn("The fake_ide option has been removed\n"); 205dc764e50SJeff Dike return 1; 2061da177e4SLinus Torvalds } 2071da177e4SLinus Torvalds __setup("fake_ide", fake_ide_setup); 2081da177e4SLinus Torvalds 2091da177e4SLinus Torvalds __uml_help(fake_ide_setup, 2101da177e4SLinus Torvalds "fake_ide\n" 2117eb90f7eSChristoph Hellwig " Obsolete stub.\n\n" 2121da177e4SLinus Torvalds ); 2131da177e4SLinus Torvalds 2141da177e4SLinus Torvalds static int parse_unit(char **ptr) 2151da177e4SLinus Torvalds { 2161da177e4SLinus Torvalds char *str = *ptr, *end; 2171da177e4SLinus Torvalds int n = -1; 2181da177e4SLinus Torvalds 2191da177e4SLinus Torvalds if(isdigit(*str)) { 2201da177e4SLinus Torvalds n = simple_strtoul(str, &end, 0); 2211da177e4SLinus Torvalds if(end == str) 222dc764e50SJeff Dike return -1; 2231da177e4SLinus Torvalds *ptr = end; 2241da177e4SLinus Torvalds } 22597d88ac8SPaolo 'Blaisorblade' Giarrusso else if (('a' <= *str) && (*str <= 'z')) { 2261da177e4SLinus Torvalds n = *str - 'a'; 2271da177e4SLinus Torvalds str++; 2281da177e4SLinus Torvalds *ptr = str; 2291da177e4SLinus Torvalds } 230dc764e50SJeff Dike return n; 2311da177e4SLinus Torvalds } 2321da177e4SLinus Torvalds 233d8d7c28eSPaolo 'Blaisorblade' Giarrusso /* If *index_out == -1 at exit, the passed option was a general one; 234d8d7c28eSPaolo 'Blaisorblade' Giarrusso * otherwise, the str pointer is used (and owned) inside ubd_devs array, so it 235d8d7c28eSPaolo 'Blaisorblade' Giarrusso * should not be freed on exit. 236d8d7c28eSPaolo 'Blaisorblade' Giarrusso */ 237f28169d2SJeff Dike static int ubd_setup_common(char *str, int *index_out, char **error_out) 2381da177e4SLinus Torvalds { 2397d314e34SPaolo 'Blaisorblade' Giarrusso struct ubd *ubd_dev; 2401da177e4SLinus Torvalds struct openflags flags = global_openflags; 241ef3ba87cSChristopher Obbard char *file, *backing_file, *serial; 242b8831a1dSJeff Dike int n, err = 0, i; 2431da177e4SLinus Torvalds 2441da177e4SLinus Torvalds if(index_out) *index_out = -1; 2451da177e4SLinus Torvalds n = *str; 2461da177e4SLinus Torvalds if(n == '='){ 2471da177e4SLinus Torvalds str++; 2481da177e4SLinus Torvalds if(!strcmp(str, "sync")){ 2491da177e4SLinus Torvalds global_openflags = of_sync(global_openflags); 2509ca55299SDaniel Walter return err; 2511da177e4SLinus Torvalds } 2521da177e4SLinus Torvalds 2537eb90f7eSChristoph Hellwig pr_warn("fake major not supported any more\n"); 2547eb90f7eSChristoph Hellwig return 0; 2551da177e4SLinus Torvalds } 2561da177e4SLinus Torvalds 2571da177e4SLinus Torvalds n = parse_unit(&str); 2581da177e4SLinus Torvalds if(n < 0){ 259f28169d2SJeff Dike *error_out = "Couldn't parse device number"; 260f28169d2SJeff Dike return -EINVAL; 2611da177e4SLinus Torvalds } 2621da177e4SLinus Torvalds if(n >= MAX_DEV){ 263f28169d2SJeff Dike *error_out = "Device number out of range"; 264f28169d2SJeff Dike return 1; 2651da177e4SLinus Torvalds } 2661da177e4SLinus Torvalds 267f28169d2SJeff Dike err = -EBUSY; 268d7fb2c38SPaolo 'Blaisorblade' Giarrusso mutex_lock(&ubd_lock); 2691da177e4SLinus Torvalds 2707d314e34SPaolo 'Blaisorblade' Giarrusso ubd_dev = &ubd_devs[n]; 2717d314e34SPaolo 'Blaisorblade' Giarrusso if(ubd_dev->file != NULL){ 272f28169d2SJeff Dike *error_out = "Device is already configured"; 2731da177e4SLinus Torvalds goto out; 2741da177e4SLinus Torvalds } 2751da177e4SLinus Torvalds 2761da177e4SLinus Torvalds if (index_out) 2771da177e4SLinus Torvalds *index_out = n; 2781da177e4SLinus Torvalds 279f28169d2SJeff Dike err = -EINVAL; 28050109b5aSAnton Ivanov for (i = 0; i < sizeof("rscdt="); i++) { 2811da177e4SLinus Torvalds switch (*str) { 2821da177e4SLinus Torvalds case 'r': 2831da177e4SLinus Torvalds flags.w = 0; 2841da177e4SLinus Torvalds break; 2851da177e4SLinus Torvalds case 's': 2861da177e4SLinus Torvalds flags.s = 1; 2871da177e4SLinus Torvalds break; 2881da177e4SLinus Torvalds case 'd': 2897d314e34SPaolo 'Blaisorblade' Giarrusso ubd_dev->no_cow = 1; 2901da177e4SLinus Torvalds break; 2916c29256cSJeff Dike case 'c': 2927d314e34SPaolo 'Blaisorblade' Giarrusso ubd_dev->shared = 1; 2936c29256cSJeff Dike break; 29450109b5aSAnton Ivanov case 't': 29550109b5aSAnton Ivanov ubd_dev->no_trim = 1; 29650109b5aSAnton Ivanov break; 2971da177e4SLinus Torvalds case '=': 2981da177e4SLinus Torvalds str++; 2991da177e4SLinus Torvalds goto break_loop; 3001da177e4SLinus Torvalds default: 301f28169d2SJeff Dike *error_out = "Expected '=' or flag letter " 30250109b5aSAnton Ivanov "(r, s, c, t or d)"; 3031da177e4SLinus Torvalds goto out; 3041da177e4SLinus Torvalds } 3051da177e4SLinus Torvalds str++; 3061da177e4SLinus Torvalds } 3071da177e4SLinus Torvalds 3081da177e4SLinus Torvalds if (*str == '=') 309f28169d2SJeff Dike *error_out = "Too many flags specified"; 3101da177e4SLinus Torvalds else 311f28169d2SJeff Dike *error_out = "Missing '='"; 3121da177e4SLinus Torvalds goto out; 3131da177e4SLinus Torvalds 3141da177e4SLinus Torvalds break_loop: 315ef3ba87cSChristopher Obbard file = strsep(&str, ",:"); 316ef3ba87cSChristopher Obbard if (*file == '\0') 317ef3ba87cSChristopher Obbard file = NULL; 3181da177e4SLinus Torvalds 319ef3ba87cSChristopher Obbard backing_file = strsep(&str, ",:"); 32094c41b3aSHajime Tazaki if (backing_file && *backing_file == '\0') 321ef3ba87cSChristopher Obbard backing_file = NULL; 3221da177e4SLinus Torvalds 323ef3ba87cSChristopher Obbard serial = strsep(&str, ",:"); 32494c41b3aSHajime Tazaki if (serial && *serial == '\0') 325ef3ba87cSChristopher Obbard serial = NULL; 326ef3ba87cSChristopher Obbard 327ef3ba87cSChristopher Obbard if (backing_file && ubd_dev->no_cow) { 328f28169d2SJeff Dike *error_out = "Can't specify both 'd' and a cow file"; 329f28169d2SJeff Dike goto out; 330f28169d2SJeff Dike } 331ef3ba87cSChristopher Obbard 332f28169d2SJeff Dike err = 0; 333ef3ba87cSChristopher Obbard ubd_dev->file = file; 3347d314e34SPaolo 'Blaisorblade' Giarrusso ubd_dev->cow.file = backing_file; 335ef3ba87cSChristopher Obbard ubd_dev->serial = serial; 3367d314e34SPaolo 'Blaisorblade' Giarrusso ubd_dev->boot_openflags = flags; 3371da177e4SLinus Torvalds out: 338d7fb2c38SPaolo 'Blaisorblade' Giarrusso mutex_unlock(&ubd_lock); 339f28169d2SJeff Dike return err; 3401da177e4SLinus Torvalds } 3411da177e4SLinus Torvalds 3421da177e4SLinus Torvalds static int ubd_setup(char *str) 3431da177e4SLinus Torvalds { 344f28169d2SJeff Dike char *error; 345f28169d2SJeff Dike int err; 346f28169d2SJeff Dike 347f28169d2SJeff Dike err = ubd_setup_common(str, NULL, &error); 348f28169d2SJeff Dike if(err) 349f28169d2SJeff Dike printk(KERN_ERR "Failed to initialize device with \"%s\" : " 350f28169d2SJeff Dike "%s\n", str, error); 351f28169d2SJeff Dike return 1; 3521da177e4SLinus Torvalds } 3531da177e4SLinus Torvalds 3541da177e4SLinus Torvalds __setup("ubd", ubd_setup); 3551da177e4SLinus Torvalds __uml_help(ubd_setup, 356ef3ba87cSChristopher Obbard "ubd<n><flags>=<filename>[(:|,)<filename2>][(:|,)<serial>]\n" 3571da177e4SLinus Torvalds " This is used to associate a device with a file in the underlying\n" 3581da177e4SLinus Torvalds " filesystem. When specifying two filenames, the first one is the\n" 3591da177e4SLinus Torvalds " COW name and the second is the backing file name. As separator you can\n" 3601da177e4SLinus Torvalds " use either a ':' or a ',': the first one allows writing things like;\n" 3611da177e4SLinus Torvalds " ubd0=~/Uml/root_cow:~/Uml/root_backing_file\n" 3621da177e4SLinus Torvalds " while with a ',' the shell would not expand the 2nd '~'.\n" 363f28169d2SJeff Dike " When using only one filename, UML will detect whether to treat it like\n" 3641da177e4SLinus Torvalds " a COW file or a backing file. To override this detection, add the 'd'\n" 3651da177e4SLinus Torvalds " flag:\n" 3661da177e4SLinus Torvalds " ubd0d=BackingFile\n" 3671da177e4SLinus Torvalds " Usually, there is a filesystem in the file, but \n" 3681da177e4SLinus Torvalds " that's not required. Swap devices containing swap files can be\n" 3691da177e4SLinus Torvalds " specified like this. Also, a file which doesn't contain a\n" 3701da177e4SLinus Torvalds " filesystem can have its contents read in the virtual \n" 3711da177e4SLinus Torvalds " machine by running 'dd' on the device. <n> must be in the range\n" 3721da177e4SLinus Torvalds " 0 to 7. Appending an 'r' to the number will cause that device\n" 3731da177e4SLinus Torvalds " to be mounted read-only. For example ubd1r=./ext_fs. Appending\n" 37420ede453SJeff Dike " an 's' will cause data to be written to disk on the host immediately.\n" 37520ede453SJeff Dike " 'c' will cause the device to be treated as being shared between multiple\n" 37620ede453SJeff Dike " UMLs and file locking will be turned off - this is appropriate for a\n" 37720ede453SJeff Dike " cluster filesystem and inappropriate at almost all other times.\n\n" 37850109b5aSAnton Ivanov " 't' will disable trim/discard support on the device (enabled by default).\n\n" 379ef3ba87cSChristopher Obbard " An optional device serial number can be exposed using the serial parameter\n" 380ef3ba87cSChristopher Obbard " on the cmdline which is exposed as a sysfs entry. This is particularly\n" 381ef3ba87cSChristopher Obbard " useful when a unique number should be given to the device. Note when\n" 382ef3ba87cSChristopher Obbard " specifying a label, the filename2 must be also presented. It can be\n" 383ef3ba87cSChristopher Obbard " an empty string, in which case the backing file is not used:\n" 384ef3ba87cSChristopher Obbard " ubd0=File,,Serial\n" 3851da177e4SLinus Torvalds ); 3861da177e4SLinus Torvalds 3878299ca5cSJeff Dike static int udb_setup(char *str) 3881da177e4SLinus Torvalds { 3891da177e4SLinus Torvalds printk("udb%s specified on command line is almost certainly a ubd -> " 3901da177e4SLinus Torvalds "udb TYPO\n", str); 391dc764e50SJeff Dike return 1; 3921da177e4SLinus Torvalds } 3931da177e4SLinus Torvalds 3941da177e4SLinus Torvalds __setup("udb", udb_setup); 3951da177e4SLinus Torvalds __uml_help(udb_setup, 3961da177e4SLinus Torvalds "udb\n" 3970894e27eSJeff Dike " This option is here solely to catch ubd -> udb typos, which can be\n" 3980894e27eSJeff Dike " to impossible to catch visually unless you specifically look for\n" 3990894e27eSJeff Dike " them. The only result of any option starting with 'udb' is an error\n" 4001da177e4SLinus Torvalds " in the boot output.\n\n" 4011da177e4SLinus Torvalds ); 4021da177e4SLinus Torvalds 40391acb21fSJeff Dike /* Only changed by ubd_init, which is an initcall. */ 4045dc62b1bSWANG Cong static int thread_fd = -1; 40509ace81cSJeff Dike 406f88f0bdfSAnton Ivanov /* Function to read several request pointers at a time 407f88f0bdfSAnton Ivanov * handling fractional reads if (and as) needed 408f88f0bdfSAnton Ivanov */ 409f88f0bdfSAnton Ivanov 410f88f0bdfSAnton Ivanov static int bulk_req_safe_read( 411f88f0bdfSAnton Ivanov int fd, 412f88f0bdfSAnton Ivanov struct io_thread_req * (*request_buffer)[], 413f88f0bdfSAnton Ivanov struct io_thread_req **remainder, 414f88f0bdfSAnton Ivanov int *remainder_size, 415f88f0bdfSAnton Ivanov int max_recs 416f88f0bdfSAnton Ivanov ) 417f88f0bdfSAnton Ivanov { 418f88f0bdfSAnton Ivanov int n = 0; 419f88f0bdfSAnton Ivanov int res = 0; 420f88f0bdfSAnton Ivanov 421f88f0bdfSAnton Ivanov if (*remainder_size > 0) { 422f88f0bdfSAnton Ivanov memmove( 423f88f0bdfSAnton Ivanov (char *) request_buffer, 424f88f0bdfSAnton Ivanov (char *) remainder, *remainder_size 425f88f0bdfSAnton Ivanov ); 426f88f0bdfSAnton Ivanov n = *remainder_size; 427f88f0bdfSAnton Ivanov } 428f88f0bdfSAnton Ivanov 429f88f0bdfSAnton Ivanov res = os_read_file( 430f88f0bdfSAnton Ivanov fd, 431f88f0bdfSAnton Ivanov ((char *) request_buffer) + *remainder_size, 432f88f0bdfSAnton Ivanov sizeof(struct io_thread_req *)*max_recs 433f88f0bdfSAnton Ivanov - *remainder_size 434f88f0bdfSAnton Ivanov ); 435f88f0bdfSAnton Ivanov if (res > 0) { 436f88f0bdfSAnton Ivanov n += res; 437f88f0bdfSAnton Ivanov if ((n % sizeof(struct io_thread_req *)) > 0) { 438f88f0bdfSAnton Ivanov /* 439f88f0bdfSAnton Ivanov * Read somehow returned not a multiple of dword 440f88f0bdfSAnton Ivanov * theoretically possible, but never observed in the 441f88f0bdfSAnton Ivanov * wild, so read routine must be able to handle it 442f88f0bdfSAnton Ivanov */ 443f88f0bdfSAnton Ivanov *remainder_size = n % sizeof(struct io_thread_req *); 444f88f0bdfSAnton Ivanov WARN(*remainder_size > 0, "UBD IPC read returned a partial result"); 445f88f0bdfSAnton Ivanov memmove( 446f88f0bdfSAnton Ivanov remainder, 447f88f0bdfSAnton Ivanov ((char *) request_buffer) + 448f88f0bdfSAnton Ivanov (n/sizeof(struct io_thread_req *))*sizeof(struct io_thread_req *), 449f88f0bdfSAnton Ivanov *remainder_size 450f88f0bdfSAnton Ivanov ); 451f88f0bdfSAnton Ivanov n = n - *remainder_size; 452f88f0bdfSAnton Ivanov } 453f88f0bdfSAnton Ivanov } else { 454f88f0bdfSAnton Ivanov n = res; 455f88f0bdfSAnton Ivanov } 456f88f0bdfSAnton Ivanov return n; 457f88f0bdfSAnton Ivanov } 458f88f0bdfSAnton Ivanov 45962f96cb0SJeff Dike /* Called without dev->lock held, and only in interrupt context. */ 46091acb21fSJeff Dike static void ubd_handler(void) 46191acb21fSJeff Dike { 46291acb21fSJeff Dike int n; 463f88f0bdfSAnton Ivanov int count; 46409ace81cSJeff Dike 465a0044bdfSJeff Dike while(1){ 466f88f0bdfSAnton Ivanov n = bulk_req_safe_read( 467f88f0bdfSAnton Ivanov thread_fd, 468f88f0bdfSAnton Ivanov irq_req_buffer, 469f88f0bdfSAnton Ivanov &irq_remainder, 470f88f0bdfSAnton Ivanov &irq_remainder_size, 471f88f0bdfSAnton Ivanov UBD_REQ_BUFFER_SIZE 472f88f0bdfSAnton Ivanov ); 473f88f0bdfSAnton Ivanov if (n < 0) { 474a0044bdfSJeff Dike if(n == -EAGAIN) 475a0044bdfSJeff Dike break; 476a0044bdfSJeff Dike printk(KERN_ERR "spurious interrupt in ubd_handler, " 477a0044bdfSJeff Dike "err = %d\n", -n); 47891acb21fSJeff Dike return; 47991acb21fSJeff Dike } 480f88f0bdfSAnton Ivanov for (count = 0; count < n/sizeof(struct io_thread_req *); count++) { 4814e6da0feSRichard Weinberger struct io_thread_req *io_req = (*irq_req_buffer)[count]; 482a0044bdfSJeff Dike 48350109b5aSAnton Ivanov if ((io_req->error == BLK_STS_NOTSUPP) && (req_op(io_req->req) == REQ_OP_DISCARD)) { 48450109b5aSAnton Ivanov blk_queue_max_discard_sectors(io_req->req->q, 0); 48550109b5aSAnton Ivanov blk_queue_max_write_zeroes_sectors(io_req->req->q, 0); 48650109b5aSAnton Ivanov blk_queue_flag_clear(QUEUE_FLAG_DISCARD, io_req->req->q); 48750109b5aSAnton Ivanov } 48850109b5aSAnton Ivanov blk_mq_end_request(io_req->req, io_req->error); 4894e6da0feSRichard Weinberger kfree(io_req); 490a0044bdfSJeff Dike } 49191acb21fSJeff Dike } 4924e6da0feSRichard Weinberger } 4934e6da0feSRichard Weinberger 4947bea96fdSAl Viro static irqreturn_t ubd_intr(int irq, void *dev) 4951da177e4SLinus Torvalds { 49691acb21fSJeff Dike ubd_handler(); 497dc764e50SJeff Dike return IRQ_HANDLED; 4981da177e4SLinus Torvalds } 4991da177e4SLinus Torvalds 50091acb21fSJeff Dike /* Only changed by ubd_init, which is an initcall. */ 50191acb21fSJeff Dike static int io_pid = -1; 50291acb21fSJeff Dike 5035dc62b1bSWANG Cong static void kill_io_thread(void) 50491acb21fSJeff Dike { 50591acb21fSJeff Dike if(io_pid != -1) 50691acb21fSJeff Dike os_kill_process(io_pid, 1); 50791acb21fSJeff Dike } 50891acb21fSJeff Dike 50991acb21fSJeff Dike __uml_exitcall(kill_io_thread); 51091acb21fSJeff Dike 511d8d7c28eSPaolo 'Blaisorblade' Giarrusso static inline int ubd_file_size(struct ubd *ubd_dev, __u64 *size_out) 5121da177e4SLinus Torvalds { 5131da177e4SLinus Torvalds char *file; 51485356398SRichard Weinberger int fd; 51585356398SRichard Weinberger int err; 5161da177e4SLinus Torvalds 51785356398SRichard Weinberger __u32 version; 51885356398SRichard Weinberger __u32 align; 51985356398SRichard Weinberger char *backing_file; 520853bc0abSArnd Bergmann time64_t mtime; 52185356398SRichard Weinberger unsigned long long size; 52285356398SRichard Weinberger int sector_size; 52385356398SRichard Weinberger int bitmap_offset; 52485356398SRichard Weinberger 52585356398SRichard Weinberger if (ubd_dev->file && ubd_dev->cow.file) { 52685356398SRichard Weinberger file = ubd_dev->cow.file; 52785356398SRichard Weinberger 52885356398SRichard Weinberger goto out; 52985356398SRichard Weinberger } 53085356398SRichard Weinberger 531d4afcba9SMartin Pärtel fd = os_open_file(ubd_dev->file, of_read(OPENFLAGS()), 0); 53285356398SRichard Weinberger if (fd < 0) 53385356398SRichard Weinberger return fd; 53485356398SRichard Weinberger 53585356398SRichard Weinberger err = read_cow_header(file_reader, &fd, &version, &backing_file, \ 53685356398SRichard Weinberger &mtime, &size, §or_size, &align, &bitmap_offset); 53785356398SRichard Weinberger os_close_file(fd); 53885356398SRichard Weinberger 53985356398SRichard Weinberger if(err == -EINVAL) 54085356398SRichard Weinberger file = ubd_dev->file; 54185356398SRichard Weinberger else 54285356398SRichard Weinberger file = backing_file; 54385356398SRichard Weinberger 54485356398SRichard Weinberger out: 545dc764e50SJeff Dike return os_file_size(file, size_out); 5461da177e4SLinus Torvalds } 5471da177e4SLinus Torvalds 5485dc62b1bSWANG Cong static int read_cow_bitmap(int fd, void *buf, int offset, int len) 5495dc62b1bSWANG Cong { 5505dc62b1bSWANG Cong int err; 5515dc62b1bSWANG Cong 5528c6157b6SAnton Ivanov err = os_pread_file(fd, buf, len, offset); 5535dc62b1bSWANG Cong if (err < 0) 5545dc62b1bSWANG Cong return err; 5555dc62b1bSWANG Cong 5565dc62b1bSWANG Cong return 0; 5575dc62b1bSWANG Cong } 5585dc62b1bSWANG Cong 559853bc0abSArnd Bergmann static int backing_file_mismatch(char *file, __u64 size, time64_t mtime) 5605dc62b1bSWANG Cong { 561853bc0abSArnd Bergmann time64_t modtime; 5625dc62b1bSWANG Cong unsigned long long actual; 5635dc62b1bSWANG Cong int err; 5645dc62b1bSWANG Cong 5655dc62b1bSWANG Cong err = os_file_modtime(file, &modtime); 5665dc62b1bSWANG Cong if (err < 0) { 5675dc62b1bSWANG Cong printk(KERN_ERR "Failed to get modification time of backing " 5685dc62b1bSWANG Cong "file \"%s\", err = %d\n", file, -err); 5695dc62b1bSWANG Cong return err; 5705dc62b1bSWANG Cong } 5715dc62b1bSWANG Cong 5725dc62b1bSWANG Cong err = os_file_size(file, &actual); 5735dc62b1bSWANG Cong if (err < 0) { 5745dc62b1bSWANG Cong printk(KERN_ERR "Failed to get size of backing file \"%s\", " 5755dc62b1bSWANG Cong "err = %d\n", file, -err); 5765dc62b1bSWANG Cong return err; 5775dc62b1bSWANG Cong } 5785dc62b1bSWANG Cong 5795dc62b1bSWANG Cong if (actual != size) { 5805dc62b1bSWANG Cong /*__u64 can be a long on AMD64 and with %lu GCC complains; so 5815dc62b1bSWANG Cong * the typecast.*/ 5825dc62b1bSWANG Cong printk(KERN_ERR "Size mismatch (%llu vs %llu) of COW header " 5835dc62b1bSWANG Cong "vs backing file\n", (unsigned long long) size, actual); 5845dc62b1bSWANG Cong return -EINVAL; 5855dc62b1bSWANG Cong } 5865dc62b1bSWANG Cong if (modtime != mtime) { 587853bc0abSArnd Bergmann printk(KERN_ERR "mtime mismatch (%lld vs %lld) of COW header vs " 5885dc62b1bSWANG Cong "backing file\n", mtime, modtime); 5895dc62b1bSWANG Cong return -EINVAL; 5905dc62b1bSWANG Cong } 5915dc62b1bSWANG Cong return 0; 5925dc62b1bSWANG Cong } 5935dc62b1bSWANG Cong 5945dc62b1bSWANG Cong static int path_requires_switch(char *from_cmdline, char *from_cow, char *cow) 5955dc62b1bSWANG Cong { 5965dc62b1bSWANG Cong struct uml_stat buf1, buf2; 5975dc62b1bSWANG Cong int err; 5985dc62b1bSWANG Cong 5995dc62b1bSWANG Cong if (from_cmdline == NULL) 6005dc62b1bSWANG Cong return 0; 6015dc62b1bSWANG Cong if (!strcmp(from_cmdline, from_cow)) 6025dc62b1bSWANG Cong return 0; 6035dc62b1bSWANG Cong 6045dc62b1bSWANG Cong err = os_stat_file(from_cmdline, &buf1); 6055dc62b1bSWANG Cong if (err < 0) { 6065dc62b1bSWANG Cong printk(KERN_ERR "Couldn't stat '%s', err = %d\n", from_cmdline, 6075dc62b1bSWANG Cong -err); 6085dc62b1bSWANG Cong return 0; 6095dc62b1bSWANG Cong } 6105dc62b1bSWANG Cong err = os_stat_file(from_cow, &buf2); 6115dc62b1bSWANG Cong if (err < 0) { 6125dc62b1bSWANG Cong printk(KERN_ERR "Couldn't stat '%s', err = %d\n", from_cow, 6135dc62b1bSWANG Cong -err); 6145dc62b1bSWANG Cong return 1; 6155dc62b1bSWANG Cong } 6165dc62b1bSWANG Cong if ((buf1.ust_dev == buf2.ust_dev) && (buf1.ust_ino == buf2.ust_ino)) 6175dc62b1bSWANG Cong return 0; 6185dc62b1bSWANG Cong 6195dc62b1bSWANG Cong printk(KERN_ERR "Backing file mismatch - \"%s\" requested, " 6205dc62b1bSWANG Cong "\"%s\" specified in COW header of \"%s\"\n", 6215dc62b1bSWANG Cong from_cmdline, from_cow, cow); 6225dc62b1bSWANG Cong return 1; 6235dc62b1bSWANG Cong } 6245dc62b1bSWANG Cong 6255dc62b1bSWANG Cong static int open_ubd_file(char *file, struct openflags *openflags, int shared, 6265dc62b1bSWANG Cong char **backing_file_out, int *bitmap_offset_out, 6275dc62b1bSWANG Cong unsigned long *bitmap_len_out, int *data_offset_out, 6285dc62b1bSWANG Cong int *create_cow_out) 6295dc62b1bSWANG Cong { 630853bc0abSArnd Bergmann time64_t mtime; 6315dc62b1bSWANG Cong unsigned long long size; 6325dc62b1bSWANG Cong __u32 version, align; 6335dc62b1bSWANG Cong char *backing_file; 6345dc62b1bSWANG Cong int fd, err, sectorsize, asked_switch, mode = 0644; 6355dc62b1bSWANG Cong 6365dc62b1bSWANG Cong fd = os_open_file(file, *openflags, mode); 6375dc62b1bSWANG Cong if (fd < 0) { 6385dc62b1bSWANG Cong if ((fd == -ENOENT) && (create_cow_out != NULL)) 6395dc62b1bSWANG Cong *create_cow_out = 1; 6405dc62b1bSWANG Cong if (!openflags->w || 6415dc62b1bSWANG Cong ((fd != -EROFS) && (fd != -EACCES))) 6425dc62b1bSWANG Cong return fd; 6435dc62b1bSWANG Cong openflags->w = 0; 6445dc62b1bSWANG Cong fd = os_open_file(file, *openflags, mode); 6455dc62b1bSWANG Cong if (fd < 0) 6465dc62b1bSWANG Cong return fd; 6475dc62b1bSWANG Cong } 6485dc62b1bSWANG Cong 6495dc62b1bSWANG Cong if (shared) 6505dc62b1bSWANG Cong printk(KERN_INFO "Not locking \"%s\" on the host\n", file); 6515dc62b1bSWANG Cong else { 6525dc62b1bSWANG Cong err = os_lock_file(fd, openflags->w); 6535dc62b1bSWANG Cong if (err < 0) { 6545dc62b1bSWANG Cong printk(KERN_ERR "Failed to lock '%s', err = %d\n", 6555dc62b1bSWANG Cong file, -err); 6565dc62b1bSWANG Cong goto out_close; 6575dc62b1bSWANG Cong } 6585dc62b1bSWANG Cong } 6595dc62b1bSWANG Cong 6605dc62b1bSWANG Cong /* Successful return case! */ 6615dc62b1bSWANG Cong if (backing_file_out == NULL) 6625dc62b1bSWANG Cong return fd; 6635dc62b1bSWANG Cong 6645dc62b1bSWANG Cong err = read_cow_header(file_reader, &fd, &version, &backing_file, &mtime, 6655dc62b1bSWANG Cong &size, §orsize, &align, bitmap_offset_out); 6665dc62b1bSWANG Cong if (err && (*backing_file_out != NULL)) { 6675dc62b1bSWANG Cong printk(KERN_ERR "Failed to read COW header from COW file " 6685dc62b1bSWANG Cong "\"%s\", errno = %d\n", file, -err); 6695dc62b1bSWANG Cong goto out_close; 6705dc62b1bSWANG Cong } 6715dc62b1bSWANG Cong if (err) 6725dc62b1bSWANG Cong return fd; 6735dc62b1bSWANG Cong 6745dc62b1bSWANG Cong asked_switch = path_requires_switch(*backing_file_out, backing_file, 6755dc62b1bSWANG Cong file); 6765dc62b1bSWANG Cong 6775dc62b1bSWANG Cong /* Allow switching only if no mismatch. */ 6785dc62b1bSWANG Cong if (asked_switch && !backing_file_mismatch(*backing_file_out, size, 6795dc62b1bSWANG Cong mtime)) { 6805dc62b1bSWANG Cong printk(KERN_ERR "Switching backing file to '%s'\n", 6815dc62b1bSWANG Cong *backing_file_out); 6825dc62b1bSWANG Cong err = write_cow_header(file, fd, *backing_file_out, 6835dc62b1bSWANG Cong sectorsize, align, &size); 6845dc62b1bSWANG Cong if (err) { 6855dc62b1bSWANG Cong printk(KERN_ERR "Switch failed, errno = %d\n", -err); 6865dc62b1bSWANG Cong goto out_close; 6875dc62b1bSWANG Cong } 6885dc62b1bSWANG Cong } else { 6895dc62b1bSWANG Cong *backing_file_out = backing_file; 6905dc62b1bSWANG Cong err = backing_file_mismatch(*backing_file_out, size, mtime); 6915dc62b1bSWANG Cong if (err) 6925dc62b1bSWANG Cong goto out_close; 6935dc62b1bSWANG Cong } 6945dc62b1bSWANG Cong 6955dc62b1bSWANG Cong cow_sizes(version, size, sectorsize, align, *bitmap_offset_out, 6965dc62b1bSWANG Cong bitmap_len_out, data_offset_out); 6975dc62b1bSWANG Cong 6985dc62b1bSWANG Cong return fd; 6995dc62b1bSWANG Cong out_close: 7005dc62b1bSWANG Cong os_close_file(fd); 7015dc62b1bSWANG Cong return err; 7025dc62b1bSWANG Cong } 7035dc62b1bSWANG Cong 7045dc62b1bSWANG Cong static int create_cow_file(char *cow_file, char *backing_file, 7055dc62b1bSWANG Cong struct openflags flags, 7065dc62b1bSWANG Cong int sectorsize, int alignment, int *bitmap_offset_out, 7075dc62b1bSWANG Cong unsigned long *bitmap_len_out, int *data_offset_out) 7085dc62b1bSWANG Cong { 7095dc62b1bSWANG Cong int err, fd; 7105dc62b1bSWANG Cong 7115dc62b1bSWANG Cong flags.c = 1; 7125dc62b1bSWANG Cong fd = open_ubd_file(cow_file, &flags, 0, NULL, NULL, NULL, NULL, NULL); 7135dc62b1bSWANG Cong if (fd < 0) { 7145dc62b1bSWANG Cong err = fd; 7155dc62b1bSWANG Cong printk(KERN_ERR "Open of COW file '%s' failed, errno = %d\n", 7165dc62b1bSWANG Cong cow_file, -err); 7175dc62b1bSWANG Cong goto out; 7185dc62b1bSWANG Cong } 7195dc62b1bSWANG Cong 7205dc62b1bSWANG Cong err = init_cow_file(fd, cow_file, backing_file, sectorsize, alignment, 7215dc62b1bSWANG Cong bitmap_offset_out, bitmap_len_out, 7225dc62b1bSWANG Cong data_offset_out); 7235dc62b1bSWANG Cong if (!err) 7245dc62b1bSWANG Cong return fd; 7255dc62b1bSWANG Cong os_close_file(fd); 7265dc62b1bSWANG Cong out: 7275dc62b1bSWANG Cong return err; 7285dc62b1bSWANG Cong } 7295dc62b1bSWANG Cong 7305f75a4f8SPaolo 'Blaisorblade' Giarrusso static void ubd_close_dev(struct ubd *ubd_dev) 7311da177e4SLinus Torvalds { 7327d314e34SPaolo 'Blaisorblade' Giarrusso os_close_file(ubd_dev->fd); 7337d314e34SPaolo 'Blaisorblade' Giarrusso if(ubd_dev->cow.file == NULL) 7341da177e4SLinus Torvalds return; 7351da177e4SLinus Torvalds 7367d314e34SPaolo 'Blaisorblade' Giarrusso os_close_file(ubd_dev->cow.fd); 7377d314e34SPaolo 'Blaisorblade' Giarrusso vfree(ubd_dev->cow.bitmap); 7387d314e34SPaolo 'Blaisorblade' Giarrusso ubd_dev->cow.bitmap = NULL; 7391da177e4SLinus Torvalds } 7401da177e4SLinus Torvalds 7417d314e34SPaolo 'Blaisorblade' Giarrusso static int ubd_open_dev(struct ubd *ubd_dev) 7421da177e4SLinus Torvalds { 7431da177e4SLinus Torvalds struct openflags flags; 7441da177e4SLinus Torvalds char **back_ptr; 7451da177e4SLinus Torvalds int err, create_cow, *create_ptr; 7460bf16bffSPaolo 'Blaisorblade' Giarrusso int fd; 7471da177e4SLinus Torvalds 7487d314e34SPaolo 'Blaisorblade' Giarrusso ubd_dev->openflags = ubd_dev->boot_openflags; 7491da177e4SLinus Torvalds create_cow = 0; 7507d314e34SPaolo 'Blaisorblade' Giarrusso create_ptr = (ubd_dev->cow.file != NULL) ? &create_cow : NULL; 7517d314e34SPaolo 'Blaisorblade' Giarrusso back_ptr = ubd_dev->no_cow ? NULL : &ubd_dev->cow.file; 7520bf16bffSPaolo 'Blaisorblade' Giarrusso 7530bf16bffSPaolo 'Blaisorblade' Giarrusso fd = open_ubd_file(ubd_dev->file, &ubd_dev->openflags, ubd_dev->shared, 7547d314e34SPaolo 'Blaisorblade' Giarrusso back_ptr, &ubd_dev->cow.bitmap_offset, 7557d314e34SPaolo 'Blaisorblade' Giarrusso &ubd_dev->cow.bitmap_len, &ubd_dev->cow.data_offset, 7566c29256cSJeff Dike create_ptr); 7571da177e4SLinus Torvalds 7580bf16bffSPaolo 'Blaisorblade' Giarrusso if((fd == -ENOENT) && create_cow){ 7590bf16bffSPaolo 'Blaisorblade' Giarrusso fd = create_cow_file(ubd_dev->file, ubd_dev->cow.file, 760a43c8316SAnton Ivanov ubd_dev->openflags, SECTOR_SIZE, PAGE_SIZE, 7617d314e34SPaolo 'Blaisorblade' Giarrusso &ubd_dev->cow.bitmap_offset, 7627d314e34SPaolo 'Blaisorblade' Giarrusso &ubd_dev->cow.bitmap_len, 7637d314e34SPaolo 'Blaisorblade' Giarrusso &ubd_dev->cow.data_offset); 7640bf16bffSPaolo 'Blaisorblade' Giarrusso if(fd >= 0){ 7651da177e4SLinus Torvalds printk(KERN_INFO "Creating \"%s\" as COW file for " 7667d314e34SPaolo 'Blaisorblade' Giarrusso "\"%s\"\n", ubd_dev->file, ubd_dev->cow.file); 7671da177e4SLinus Torvalds } 7681da177e4SLinus Torvalds } 7691da177e4SLinus Torvalds 7700bf16bffSPaolo 'Blaisorblade' Giarrusso if(fd < 0){ 7717d314e34SPaolo 'Blaisorblade' Giarrusso printk("Failed to open '%s', errno = %d\n", ubd_dev->file, 7720bf16bffSPaolo 'Blaisorblade' Giarrusso -fd); 7730bf16bffSPaolo 'Blaisorblade' Giarrusso return fd; 7741da177e4SLinus Torvalds } 7750bf16bffSPaolo 'Blaisorblade' Giarrusso ubd_dev->fd = fd; 7761da177e4SLinus Torvalds 7777d314e34SPaolo 'Blaisorblade' Giarrusso if(ubd_dev->cow.file != NULL){ 778086fa5ffSMartin K. Petersen blk_queue_max_hw_sectors(ubd_dev->queue, 8 * sizeof(long)); 779f4768ffdSJeff Dike 7801da177e4SLinus Torvalds err = -ENOMEM; 781da2486baSJesper Juhl ubd_dev->cow.bitmap = vmalloc(ubd_dev->cow.bitmap_len); 7827d314e34SPaolo 'Blaisorblade' Giarrusso if(ubd_dev->cow.bitmap == NULL){ 7831da177e4SLinus Torvalds printk(KERN_ERR "Failed to vmalloc COW bitmap\n"); 7841da177e4SLinus Torvalds goto error; 7851da177e4SLinus Torvalds } 7861da177e4SLinus Torvalds flush_tlb_kernel_vm(); 7871da177e4SLinus Torvalds 7887d314e34SPaolo 'Blaisorblade' Giarrusso err = read_cow_bitmap(ubd_dev->fd, ubd_dev->cow.bitmap, 7897d314e34SPaolo 'Blaisorblade' Giarrusso ubd_dev->cow.bitmap_offset, 7907d314e34SPaolo 'Blaisorblade' Giarrusso ubd_dev->cow.bitmap_len); 7911da177e4SLinus Torvalds if(err < 0) 7921da177e4SLinus Torvalds goto error; 7931da177e4SLinus Torvalds 7947d314e34SPaolo 'Blaisorblade' Giarrusso flags = ubd_dev->openflags; 7951da177e4SLinus Torvalds flags.w = 0; 7967d314e34SPaolo 'Blaisorblade' Giarrusso err = open_ubd_file(ubd_dev->cow.file, &flags, ubd_dev->shared, NULL, 7976c29256cSJeff Dike NULL, NULL, NULL, NULL); 7981da177e4SLinus Torvalds if(err < 0) goto error; 7997d314e34SPaolo 'Blaisorblade' Giarrusso ubd_dev->cow.fd = err; 8001da177e4SLinus Torvalds } 80150109b5aSAnton Ivanov if (ubd_dev->no_trim == 0) { 80250109b5aSAnton Ivanov ubd_dev->queue->limits.discard_granularity = SECTOR_SIZE; 80350109b5aSAnton Ivanov ubd_dev->queue->limits.discard_alignment = SECTOR_SIZE; 80450109b5aSAnton Ivanov blk_queue_max_discard_sectors(ubd_dev->queue, UBD_MAX_REQUEST); 80550109b5aSAnton Ivanov blk_queue_max_write_zeroes_sectors(ubd_dev->queue, UBD_MAX_REQUEST); 80650109b5aSAnton Ivanov blk_queue_flag_set(QUEUE_FLAG_DISCARD, ubd_dev->queue); 80750109b5aSAnton Ivanov } 808a43c8316SAnton Ivanov blk_queue_flag_set(QUEUE_FLAG_NONROT, ubd_dev->queue); 809dc764e50SJeff Dike return 0; 8101da177e4SLinus Torvalds error: 8117d314e34SPaolo 'Blaisorblade' Giarrusso os_close_file(ubd_dev->fd); 812dc764e50SJeff Dike return err; 8131da177e4SLinus Torvalds } 8141da177e4SLinus Torvalds 8152e3f5251SJeff Dike static void ubd_device_release(struct device *dev) 8162e3f5251SJeff Dike { 8178691b97bSGreg Kroah-Hartman struct ubd *ubd_dev = dev_get_drvdata(dev); 8182e3f5251SJeff Dike 8194e6da0feSRichard Weinberger blk_mq_free_tag_set(&ubd_dev->tag_set); 8202e3f5251SJeff Dike *ubd_dev = ((struct ubd) DEFAULT_UBD); 8212e3f5251SJeff Dike } 8222e3f5251SJeff Dike 823ef3ba87cSChristopher Obbard static ssize_t serial_show(struct device *dev, 824ef3ba87cSChristopher Obbard struct device_attribute *attr, char *buf) 825ef3ba87cSChristopher Obbard { 826ef3ba87cSChristopher Obbard struct gendisk *disk = dev_to_disk(dev); 827ef3ba87cSChristopher Obbard struct ubd *ubd_dev = disk->private_data; 828ef3ba87cSChristopher Obbard 829ef3ba87cSChristopher Obbard if (!ubd_dev) 830ef3ba87cSChristopher Obbard return 0; 831ef3ba87cSChristopher Obbard 832ef3ba87cSChristopher Obbard return sprintf(buf, "%s", ubd_dev->serial); 833ef3ba87cSChristopher Obbard } 834ef3ba87cSChristopher Obbard 835ef3ba87cSChristopher Obbard static DEVICE_ATTR_RO(serial); 836ef3ba87cSChristopher Obbard 837ef3ba87cSChristopher Obbard static struct attribute *ubd_attrs[] = { 838ef3ba87cSChristopher Obbard &dev_attr_serial.attr, 839ef3ba87cSChristopher Obbard NULL, 840ef3ba87cSChristopher Obbard }; 841ef3ba87cSChristopher Obbard 842ef3ba87cSChristopher Obbard static umode_t ubd_attrs_are_visible(struct kobject *kobj, 843ef3ba87cSChristopher Obbard struct attribute *a, int n) 844ef3ba87cSChristopher Obbard { 845ef3ba87cSChristopher Obbard return a->mode; 846ef3ba87cSChristopher Obbard } 847ef3ba87cSChristopher Obbard 848ef3ba87cSChristopher Obbard static const struct attribute_group ubd_attr_group = { 849ef3ba87cSChristopher Obbard .attrs = ubd_attrs, 850ef3ba87cSChristopher Obbard .is_visible = ubd_attrs_are_visible, 851ef3ba87cSChristopher Obbard }; 852ef3ba87cSChristopher Obbard 853ef3ba87cSChristopher Obbard static const struct attribute_group *ubd_attr_groups[] = { 854ef3ba87cSChristopher Obbard &ubd_attr_group, 855ef3ba87cSChristopher Obbard NULL, 856ef3ba87cSChristopher Obbard }; 857ef3ba87cSChristopher Obbard 85835efb594SChristoph Hellwig static void ubd_disk_register(int major, u64 size, int unit, 85935efb594SChristoph Hellwig struct gendisk *disk) 8601da177e4SLinus Torvalds { 8611da177e4SLinus Torvalds disk->major = major; 8621da177e4SLinus Torvalds disk->first_minor = unit << UBD_SHIFT; 86335efb594SChristoph Hellwig disk->minors = 1 << UBD_SHIFT; 8641da177e4SLinus Torvalds disk->fops = &ubd_blops; 8651da177e4SLinus Torvalds set_capacity(disk, size / 512); 8661da177e4SLinus Torvalds sprintf(disk->disk_name, "ubd%c", 'a' + unit); 8671da177e4SLinus Torvalds 8687d314e34SPaolo 'Blaisorblade' Giarrusso ubd_devs[unit].pdev.id = unit; 8697d314e34SPaolo 'Blaisorblade' Giarrusso ubd_devs[unit].pdev.name = DRIVER_NAME; 8702e3f5251SJeff Dike ubd_devs[unit].pdev.dev.release = ubd_device_release; 8718691b97bSGreg Kroah-Hartman dev_set_drvdata(&ubd_devs[unit].pdev.dev, &ubd_devs[unit]); 8727d314e34SPaolo 'Blaisorblade' Giarrusso platform_device_register(&ubd_devs[unit].pdev); 8731da177e4SLinus Torvalds 8747d314e34SPaolo 'Blaisorblade' Giarrusso disk->private_data = &ubd_devs[unit]; 87562f96cb0SJeff Dike disk->queue = ubd_devs[unit].queue; 8767eb90f7eSChristoph Hellwig device_add_disk(&ubd_devs[unit].pdev.dev, disk, ubd_attr_groups); 8771da177e4SLinus Torvalds } 8781da177e4SLinus Torvalds 879a43c8316SAnton Ivanov #define ROUND_BLOCK(n) ((n + (SECTOR_SIZE - 1)) & (-SECTOR_SIZE)) 8801da177e4SLinus Torvalds 8814e6da0feSRichard Weinberger static const struct blk_mq_ops ubd_mq_ops = { 8824e6da0feSRichard Weinberger .queue_rq = ubd_queue_rq, 8834e6da0feSRichard Weinberger }; 8844e6da0feSRichard Weinberger 885f28169d2SJeff Dike static int ubd_add(int n, char **error_out) 8861da177e4SLinus Torvalds { 8877d314e34SPaolo 'Blaisorblade' Giarrusso struct ubd *ubd_dev = &ubd_devs[n]; 88835efb594SChristoph Hellwig struct gendisk *disk; 889f28169d2SJeff Dike int err = 0; 8901da177e4SLinus Torvalds 8917d314e34SPaolo 'Blaisorblade' Giarrusso if(ubd_dev->file == NULL) 892ec7cf783SJeff Dike goto out; 8931da177e4SLinus Torvalds 8947d314e34SPaolo 'Blaisorblade' Giarrusso err = ubd_file_size(ubd_dev, &ubd_dev->size); 895f28169d2SJeff Dike if(err < 0){ 896f28169d2SJeff Dike *error_out = "Couldn't determine size of device's file"; 89780c13749SJeff Dike goto out; 898f28169d2SJeff Dike } 8991da177e4SLinus Torvalds 9007d314e34SPaolo 'Blaisorblade' Giarrusso ubd_dev->size = ROUND_BLOCK(ubd_dev->size); 9011da177e4SLinus Torvalds 9024e6da0feSRichard Weinberger ubd_dev->tag_set.ops = &ubd_mq_ops; 9034e6da0feSRichard Weinberger ubd_dev->tag_set.queue_depth = 64; 9044e6da0feSRichard Weinberger ubd_dev->tag_set.numa_node = NUMA_NO_NODE; 9054e6da0feSRichard Weinberger ubd_dev->tag_set.flags = BLK_MQ_F_SHOULD_MERGE; 9064e6da0feSRichard Weinberger ubd_dev->tag_set.driver_data = ubd_dev; 9074e6da0feSRichard Weinberger ubd_dev->tag_set.nr_hw_queues = 1; 908a0044bdfSJeff Dike 9094e6da0feSRichard Weinberger err = blk_mq_alloc_tag_set(&ubd_dev->tag_set); 9104e6da0feSRichard Weinberger if (err) 91180c13749SJeff Dike goto out; 9124e6da0feSRichard Weinberger 91335efb594SChristoph Hellwig disk = blk_mq_alloc_disk(&ubd_dev->tag_set, ubd_dev); 91435efb594SChristoph Hellwig if (IS_ERR(disk)) { 91535efb594SChristoph Hellwig err = PTR_ERR(disk); 916aea05eb5SAnton Ivanov goto out_cleanup_tags; 91762f96cb0SJeff Dike } 91835efb594SChristoph Hellwig ubd_dev->queue = disk->queue; 9194e6da0feSRichard Weinberger 920f935a8ceSJens Axboe blk_queue_write_cache(ubd_dev->queue, true, false); 9218a78362cSMartin K. Petersen blk_queue_max_segments(ubd_dev->queue, MAX_SG); 922fc6b6a87SGabriel Krisman Bertazi blk_queue_segment_boundary(ubd_dev->queue, PAGE_SIZE - 1); 92335efb594SChristoph Hellwig ubd_disk_register(UBD_MAJOR, ubd_dev->size, n, disk); 92435efb594SChristoph Hellwig ubd_gendisk[n] = disk; 92535efb594SChristoph Hellwig return 0; 92662f96cb0SJeff Dike 9274e6da0feSRichard Weinberger out_cleanup_tags: 9284e6da0feSRichard Weinberger blk_mq_free_tag_set(&ubd_dev->tag_set); 92935efb594SChristoph Hellwig out: 93035efb594SChristoph Hellwig return err; 9311da177e4SLinus Torvalds } 9321da177e4SLinus Torvalds 933f28169d2SJeff Dike static int ubd_config(char *str, char **error_out) 9341da177e4SLinus Torvalds { 935e7f6552fSPaolo 'Blaisorblade' Giarrusso int n, ret; 9361da177e4SLinus Torvalds 937f28169d2SJeff Dike /* This string is possibly broken up and stored, so it's only 938f28169d2SJeff Dike * freed if ubd_setup_common fails, or if only general options 939f28169d2SJeff Dike * were set. 940f28169d2SJeff Dike */ 941970d6e3aSJeff Dike str = kstrdup(str, GFP_KERNEL); 9421da177e4SLinus Torvalds if (str == NULL) { 943f28169d2SJeff Dike *error_out = "Failed to allocate memory"; 944f28169d2SJeff Dike return -ENOMEM; 9451da177e4SLinus Torvalds } 946f28169d2SJeff Dike 947f28169d2SJeff Dike ret = ubd_setup_common(str, &n, error_out); 948f28169d2SJeff Dike if (ret) 949e7f6552fSPaolo 'Blaisorblade' Giarrusso goto err_free; 950f28169d2SJeff Dike 951e7f6552fSPaolo 'Blaisorblade' Giarrusso if (n == -1) { 952e7f6552fSPaolo 'Blaisorblade' Giarrusso ret = 0; 953d8d7c28eSPaolo 'Blaisorblade' Giarrusso goto err_free; 954e7f6552fSPaolo 'Blaisorblade' Giarrusso } 9551da177e4SLinus Torvalds 956d7fb2c38SPaolo 'Blaisorblade' Giarrusso mutex_lock(&ubd_lock); 957f28169d2SJeff Dike ret = ubd_add(n, error_out); 958e7f6552fSPaolo 'Blaisorblade' Giarrusso if (ret) 9597d314e34SPaolo 'Blaisorblade' Giarrusso ubd_devs[n].file = NULL; 960d7fb2c38SPaolo 'Blaisorblade' Giarrusso mutex_unlock(&ubd_lock); 9611da177e4SLinus Torvalds 962e7f6552fSPaolo 'Blaisorblade' Giarrusso out: 963e7f6552fSPaolo 'Blaisorblade' Giarrusso return ret; 964e7f6552fSPaolo 'Blaisorblade' Giarrusso 965e7f6552fSPaolo 'Blaisorblade' Giarrusso err_free: 966e7f6552fSPaolo 'Blaisorblade' Giarrusso kfree(str); 967e7f6552fSPaolo 'Blaisorblade' Giarrusso goto out; 9681da177e4SLinus Torvalds } 9691da177e4SLinus Torvalds 9701da177e4SLinus Torvalds static int ubd_get_config(char *name, char *str, int size, char **error_out) 9711da177e4SLinus Torvalds { 9727d314e34SPaolo 'Blaisorblade' Giarrusso struct ubd *ubd_dev; 9731da177e4SLinus Torvalds int n, len = 0; 9741da177e4SLinus Torvalds 9751da177e4SLinus Torvalds n = parse_unit(&name); 9761da177e4SLinus Torvalds if((n >= MAX_DEV) || (n < 0)){ 9771da177e4SLinus Torvalds *error_out = "ubd_get_config : device number out of range"; 978dc764e50SJeff Dike return -1; 9791da177e4SLinus Torvalds } 9801da177e4SLinus Torvalds 9817d314e34SPaolo 'Blaisorblade' Giarrusso ubd_dev = &ubd_devs[n]; 982d7fb2c38SPaolo 'Blaisorblade' Giarrusso mutex_lock(&ubd_lock); 9831da177e4SLinus Torvalds 9847d314e34SPaolo 'Blaisorblade' Giarrusso if(ubd_dev->file == NULL){ 9851da177e4SLinus Torvalds CONFIG_CHUNK(str, size, len, "", 1); 9861da177e4SLinus Torvalds goto out; 9871da177e4SLinus Torvalds } 9881da177e4SLinus Torvalds 9897d314e34SPaolo 'Blaisorblade' Giarrusso CONFIG_CHUNK(str, size, len, ubd_dev->file, 0); 9901da177e4SLinus Torvalds 9917d314e34SPaolo 'Blaisorblade' Giarrusso if(ubd_dev->cow.file != NULL){ 9921da177e4SLinus Torvalds CONFIG_CHUNK(str, size, len, ",", 0); 9937d314e34SPaolo 'Blaisorblade' Giarrusso CONFIG_CHUNK(str, size, len, ubd_dev->cow.file, 1); 9941da177e4SLinus Torvalds } 9951da177e4SLinus Torvalds else CONFIG_CHUNK(str, size, len, "", 1); 9961da177e4SLinus Torvalds 9971da177e4SLinus Torvalds out: 998d7fb2c38SPaolo 'Blaisorblade' Giarrusso mutex_unlock(&ubd_lock); 999dc764e50SJeff Dike return len; 10001da177e4SLinus Torvalds } 10011da177e4SLinus Torvalds 100229d56cfeSJeff Dike static int ubd_id(char **str, int *start_out, int *end_out) 100329d56cfeSJeff Dike { 100429d56cfeSJeff Dike int n; 100529d56cfeSJeff Dike 100629d56cfeSJeff Dike n = parse_unit(str); 100729d56cfeSJeff Dike *start_out = 0; 100829d56cfeSJeff Dike *end_out = MAX_DEV - 1; 100929d56cfeSJeff Dike return n; 101029d56cfeSJeff Dike } 101129d56cfeSJeff Dike 1012f28169d2SJeff Dike static int ubd_remove(int n, char **error_out) 10131da177e4SLinus Torvalds { 10142e3f5251SJeff Dike struct gendisk *disk = ubd_gendisk[n]; 10157d314e34SPaolo 'Blaisorblade' Giarrusso struct ubd *ubd_dev; 101629d56cfeSJeff Dike int err = -ENODEV; 10171da177e4SLinus Torvalds 1018d7fb2c38SPaolo 'Blaisorblade' Giarrusso mutex_lock(&ubd_lock); 10191da177e4SLinus Torvalds 10207d314e34SPaolo 'Blaisorblade' Giarrusso ubd_dev = &ubd_devs[n]; 102129d56cfeSJeff Dike 10227d314e34SPaolo 'Blaisorblade' Giarrusso if(ubd_dev->file == NULL) 102329d56cfeSJeff Dike goto out; 102429d56cfeSJeff Dike 102529d56cfeSJeff Dike /* you cannot remove a open disk */ 102629d56cfeSJeff Dike err = -EBUSY; 10277d314e34SPaolo 'Blaisorblade' Giarrusso if(ubd_dev->count > 0) 102829d56cfeSJeff Dike goto out; 102929d56cfeSJeff Dike 10301da177e4SLinus Torvalds ubd_gendisk[n] = NULL; 1031b47d2debSJeff Dike if(disk != NULL){ 1032b47d2debSJeff Dike del_gendisk(disk); 103335efb594SChristoph Hellwig blk_cleanup_disk(disk); 10341da177e4SLinus Torvalds } 10351da177e4SLinus Torvalds 10361da177e4SLinus Torvalds err = 0; 10372e3f5251SJeff Dike platform_device_unregister(&ubd_dev->pdev); 10381da177e4SLinus Torvalds out: 1039d7fb2c38SPaolo 'Blaisorblade' Giarrusso mutex_unlock(&ubd_lock); 104029d56cfeSJeff Dike return err; 10411da177e4SLinus Torvalds } 10421da177e4SLinus Torvalds 1043f28169d2SJeff Dike /* All these are called by mconsole in process context and without 1044b8831a1dSJeff Dike * ubd-specific locks. The structure itself is const except for .list. 1045f28169d2SJeff Dike */ 10461da177e4SLinus Torvalds static struct mc_device ubd_mc = { 104784f48d4fSJeff Dike .list = LIST_HEAD_INIT(ubd_mc.list), 10481da177e4SLinus Torvalds .name = "ubd", 10491da177e4SLinus Torvalds .config = ubd_config, 10501da177e4SLinus Torvalds .get_config = ubd_get_config, 105129d56cfeSJeff Dike .id = ubd_id, 10521da177e4SLinus Torvalds .remove = ubd_remove, 10531da177e4SLinus Torvalds }; 10541da177e4SLinus Torvalds 1055d8d7c28eSPaolo 'Blaisorblade' Giarrusso static int __init ubd_mc_init(void) 10561da177e4SLinus Torvalds { 10571da177e4SLinus Torvalds mconsole_register_dev(&ubd_mc); 10581da177e4SLinus Torvalds return 0; 10591da177e4SLinus Torvalds } 10601da177e4SLinus Torvalds 10611da177e4SLinus Torvalds __initcall(ubd_mc_init); 10621da177e4SLinus Torvalds 1063d8d7c28eSPaolo 'Blaisorblade' Giarrusso static int __init ubd0_init(void) 1064d8d7c28eSPaolo 'Blaisorblade' Giarrusso { 1065d8d7c28eSPaolo 'Blaisorblade' Giarrusso struct ubd *ubd_dev = &ubd_devs[0]; 1066d8d7c28eSPaolo 'Blaisorblade' Giarrusso 1067b8831a1dSJeff Dike mutex_lock(&ubd_lock); 1068d8d7c28eSPaolo 'Blaisorblade' Giarrusso if(ubd_dev->file == NULL) 1069d8d7c28eSPaolo 'Blaisorblade' Giarrusso ubd_dev->file = "root_fs"; 1070b8831a1dSJeff Dike mutex_unlock(&ubd_lock); 1071b8831a1dSJeff Dike 1072dc764e50SJeff Dike return 0; 1073d8d7c28eSPaolo 'Blaisorblade' Giarrusso } 1074d8d7c28eSPaolo 'Blaisorblade' Giarrusso 1075d8d7c28eSPaolo 'Blaisorblade' Giarrusso __initcall(ubd0_init); 1076d8d7c28eSPaolo 'Blaisorblade' Giarrusso 1077b8831a1dSJeff Dike /* Used in ubd_init, which is an initcall */ 10783ae5eaecSRussell King static struct platform_driver ubd_driver = { 10793ae5eaecSRussell King .driver = { 10801da177e4SLinus Torvalds .name = DRIVER_NAME, 10813ae5eaecSRussell King }, 10821da177e4SLinus Torvalds }; 10831da177e4SLinus Torvalds 1084d8d7c28eSPaolo 'Blaisorblade' Giarrusso static int __init ubd_init(void) 10851da177e4SLinus Torvalds { 1086f28169d2SJeff Dike char *error; 1087f28169d2SJeff Dike int i, err; 10881da177e4SLinus Torvalds 1089792dd4fcSChristoph Hellwig if (register_blkdev(UBD_MAJOR, "ubd")) 10901da177e4SLinus Torvalds return -1; 10911da177e4SLinus Torvalds 10926da2ec56SKees Cook irq_req_buffer = kmalloc_array(UBD_REQ_BUFFER_SIZE, 10936da2ec56SKees Cook sizeof(struct io_thread_req *), 1094f88f0bdfSAnton Ivanov GFP_KERNEL 1095f88f0bdfSAnton Ivanov ); 1096f88f0bdfSAnton Ivanov irq_remainder = 0; 1097f88f0bdfSAnton Ivanov 1098f88f0bdfSAnton Ivanov if (irq_req_buffer == NULL) { 1099f88f0bdfSAnton Ivanov printk(KERN_ERR "Failed to initialize ubd buffering\n"); 1100f88f0bdfSAnton Ivanov return -1; 1101f88f0bdfSAnton Ivanov } 11026da2ec56SKees Cook io_req_buffer = kmalloc_array(UBD_REQ_BUFFER_SIZE, 11036da2ec56SKees Cook sizeof(struct io_thread_req *), 1104f88f0bdfSAnton Ivanov GFP_KERNEL 1105f88f0bdfSAnton Ivanov ); 1106f88f0bdfSAnton Ivanov 1107f88f0bdfSAnton Ivanov io_remainder = 0; 1108f88f0bdfSAnton Ivanov 1109f88f0bdfSAnton Ivanov if (io_req_buffer == NULL) { 1110f88f0bdfSAnton Ivanov printk(KERN_ERR "Failed to initialize ubd buffering\n"); 1111f88f0bdfSAnton Ivanov return -1; 1112f88f0bdfSAnton Ivanov } 11133ae5eaecSRussell King platform_driver_register(&ubd_driver); 1114b8831a1dSJeff Dike mutex_lock(&ubd_lock); 1115f28169d2SJeff Dike for (i = 0; i < MAX_DEV; i++){ 1116f28169d2SJeff Dike err = ubd_add(i, &error); 1117f28169d2SJeff Dike if(err) 1118f28169d2SJeff Dike printk(KERN_ERR "Failed to initialize ubd device %d :" 1119f28169d2SJeff Dike "%s\n", i, error); 1120f28169d2SJeff Dike } 1121b8831a1dSJeff Dike mutex_unlock(&ubd_lock); 11221da177e4SLinus Torvalds return 0; 11231da177e4SLinus Torvalds } 11241da177e4SLinus Torvalds 11251da177e4SLinus Torvalds late_initcall(ubd_init); 11261da177e4SLinus Torvalds 1127d8d7c28eSPaolo 'Blaisorblade' Giarrusso static int __init ubd_driver_init(void){ 112891acb21fSJeff Dike unsigned long stack; 112991acb21fSJeff Dike int err; 113091acb21fSJeff Dike 113191acb21fSJeff Dike /* Set by CONFIG_BLK_DEV_UBD_SYNC or ubd=sync.*/ 113291acb21fSJeff Dike if(global_openflags.s){ 113391acb21fSJeff Dike printk(KERN_INFO "ubd: Synchronous mode\n"); 113491acb21fSJeff Dike /* Letting ubd=sync be like using ubd#s= instead of ubd#= is 113591acb21fSJeff Dike * enough. So use anyway the io thread. */ 113691acb21fSJeff Dike } 11372fcb4090SJohannes Berg stack = alloc_stack(0, 0); 1138558f9b2fSYiFei Zhu io_pid = start_io_thread(stack + PAGE_SIZE, &thread_fd); 113991acb21fSJeff Dike if(io_pid < 0){ 114091acb21fSJeff Dike printk(KERN_ERR 114191acb21fSJeff Dike "ubd : Failed to start I/O thread (errno = %d) - " 114291acb21fSJeff Dike "falling back to synchronous I/O\n", -io_pid); 114391acb21fSJeff Dike io_pid = -1; 1144dc764e50SJeff Dike return 0; 114591acb21fSJeff Dike } 114691acb21fSJeff Dike err = um_request_irq(UBD_IRQ, thread_fd, IRQ_READ, ubd_intr, 1147c0b79a90SYong Zhang 0, "ubd", ubd_devs); 114836d46a59SJohannes Berg if(err < 0) 114991acb21fSJeff Dike printk(KERN_ERR "um_request_irq failed - errno = %d\n", -err); 1150f4c57a78SJeff Dike return 0; 115191acb21fSJeff Dike } 115291acb21fSJeff Dike 115391acb21fSJeff Dike device_initcall(ubd_driver_init); 115491acb21fSJeff Dike 1155a625c998SAl Viro static int ubd_open(struct block_device *bdev, fmode_t mode) 11561da177e4SLinus Torvalds { 1157a625c998SAl Viro struct gendisk *disk = bdev->bd_disk; 11587d314e34SPaolo 'Blaisorblade' Giarrusso struct ubd *ubd_dev = disk->private_data; 11591da177e4SLinus Torvalds int err = 0; 11601da177e4SLinus Torvalds 11619a181c58SArnd Bergmann mutex_lock(&ubd_mutex); 11627d314e34SPaolo 'Blaisorblade' Giarrusso if(ubd_dev->count == 0){ 11637d314e34SPaolo 'Blaisorblade' Giarrusso err = ubd_open_dev(ubd_dev); 11641da177e4SLinus Torvalds if(err){ 11651da177e4SLinus Torvalds printk(KERN_ERR "%s: Can't open \"%s\": errno = %d\n", 11667d314e34SPaolo 'Blaisorblade' Giarrusso disk->disk_name, ubd_dev->file, -err); 11671da177e4SLinus Torvalds goto out; 11681da177e4SLinus Torvalds } 11691da177e4SLinus Torvalds } 11707d314e34SPaolo 'Blaisorblade' Giarrusso ubd_dev->count++; 11717d314e34SPaolo 'Blaisorblade' Giarrusso set_disk_ro(disk, !ubd_dev->openflags.w); 11722c49be99SPaolo 'Blaisorblade' Giarrusso 11732c49be99SPaolo 'Blaisorblade' Giarrusso /* This should no more be needed. And it didn't work anyway to exclude 11742c49be99SPaolo 'Blaisorblade' Giarrusso * read-write remounting of filesystems.*/ 1175a625c998SAl Viro /*if((mode & FMODE_WRITE) && !ubd_dev->openflags.w){ 11765f75a4f8SPaolo 'Blaisorblade' Giarrusso if(--ubd_dev->count == 0) ubd_close_dev(ubd_dev); 11771da177e4SLinus Torvalds err = -EROFS; 11782c49be99SPaolo 'Blaisorblade' Giarrusso }*/ 11791da177e4SLinus Torvalds out: 11809a181c58SArnd Bergmann mutex_unlock(&ubd_mutex); 1181dc764e50SJeff Dike return err; 11821da177e4SLinus Torvalds } 11831da177e4SLinus Torvalds 1184db2a144bSAl Viro static void ubd_release(struct gendisk *disk, fmode_t mode) 11851da177e4SLinus Torvalds { 11867d314e34SPaolo 'Blaisorblade' Giarrusso struct ubd *ubd_dev = disk->private_data; 11871da177e4SLinus Torvalds 11889a181c58SArnd Bergmann mutex_lock(&ubd_mutex); 11897d314e34SPaolo 'Blaisorblade' Giarrusso if(--ubd_dev->count == 0) 11905f75a4f8SPaolo 'Blaisorblade' Giarrusso ubd_close_dev(ubd_dev); 11919a181c58SArnd Bergmann mutex_unlock(&ubd_mutex); 11921da177e4SLinus Torvalds } 11931da177e4SLinus Torvalds 119491acb21fSJeff Dike static void cowify_bitmap(__u64 io_offset, int length, unsigned long *cow_mask, 119591acb21fSJeff Dike __u64 *cow_offset, unsigned long *bitmap, 119691acb21fSJeff Dike __u64 bitmap_offset, unsigned long *bitmap_words, 119791acb21fSJeff Dike __u64 bitmap_len) 11981da177e4SLinus Torvalds { 1199a43c8316SAnton Ivanov __u64 sector = io_offset >> SECTOR_SHIFT; 120091acb21fSJeff Dike int i, update_bitmap = 0; 12011da177e4SLinus Torvalds 1202a43c8316SAnton Ivanov for (i = 0; i < length >> SECTOR_SHIFT; i++) { 120391acb21fSJeff Dike if(cow_mask != NULL) 120491acb21fSJeff Dike ubd_set_bit(i, (unsigned char *) cow_mask); 120591acb21fSJeff Dike if(ubd_test_bit(sector + i, (unsigned char *) bitmap)) 120609ace81cSJeff Dike continue; 12071da177e4SLinus Torvalds 120891acb21fSJeff Dike update_bitmap = 1; 120991acb21fSJeff Dike ubd_set_bit(sector + i, (unsigned char *) bitmap); 12101da177e4SLinus Torvalds } 121191acb21fSJeff Dike 121291acb21fSJeff Dike if(!update_bitmap) 121391acb21fSJeff Dike return; 121491acb21fSJeff Dike 121591acb21fSJeff Dike *cow_offset = sector / (sizeof(unsigned long) * 8); 121691acb21fSJeff Dike 121791acb21fSJeff Dike /* This takes care of the case where we're exactly at the end of the 121891acb21fSJeff Dike * device, and *cow_offset + 1 is off the end. So, just back it up 121991acb21fSJeff Dike * by one word. Thanks to Lynn Kerby for the fix and James McMechan 122091acb21fSJeff Dike * for the original diagnosis. 122191acb21fSJeff Dike */ 12226d074242SJiri Olsa if (*cow_offset == (DIV_ROUND_UP(bitmap_len, 12236d074242SJiri Olsa sizeof(unsigned long)) - 1)) 122491acb21fSJeff Dike (*cow_offset)--; 122591acb21fSJeff Dike 122691acb21fSJeff Dike bitmap_words[0] = bitmap[*cow_offset]; 122791acb21fSJeff Dike bitmap_words[1] = bitmap[*cow_offset + 1]; 122891acb21fSJeff Dike 122991acb21fSJeff Dike *cow_offset *= sizeof(unsigned long); 123091acb21fSJeff Dike *cow_offset += bitmap_offset; 123191acb21fSJeff Dike } 123291acb21fSJeff Dike 1233fc6b6a87SGabriel Krisman Bertazi static void cowify_req(struct io_thread_req *req, struct io_desc *segment, 1234fc6b6a87SGabriel Krisman Bertazi unsigned long offset, unsigned long *bitmap, 123591acb21fSJeff Dike __u64 bitmap_offset, __u64 bitmap_len) 123691acb21fSJeff Dike { 1237fc6b6a87SGabriel Krisman Bertazi __u64 sector = offset >> SECTOR_SHIFT; 123891acb21fSJeff Dike int i; 123991acb21fSJeff Dike 1240fc6b6a87SGabriel Krisman Bertazi if (segment->length > (sizeof(segment->sector_mask) * 8) << SECTOR_SHIFT) 124191acb21fSJeff Dike panic("Operation too long"); 124291acb21fSJeff Dike 1243a43c8316SAnton Ivanov if (req_op(req->req) == REQ_OP_READ) { 1244fc6b6a87SGabriel Krisman Bertazi for (i = 0; i < segment->length >> SECTOR_SHIFT; i++) { 124591acb21fSJeff Dike if(ubd_test_bit(sector + i, (unsigned char *) bitmap)) 124691acb21fSJeff Dike ubd_set_bit(i, (unsigned char *) 1247fc6b6a87SGabriel Krisman Bertazi &segment->sector_mask); 124891acb21fSJeff Dike } 1249fc6b6a87SGabriel Krisman Bertazi } else { 1250fc6b6a87SGabriel Krisman Bertazi cowify_bitmap(offset, segment->length, &segment->sector_mask, 1251fc6b6a87SGabriel Krisman Bertazi &segment->cow_offset, bitmap, bitmap_offset, 1252fc6b6a87SGabriel Krisman Bertazi segment->bitmap_words, bitmap_len); 125391acb21fSJeff Dike } 12541da177e4SLinus Torvalds } 12551da177e4SLinus Torvalds 1256fc6b6a87SGabriel Krisman Bertazi static void ubd_map_req(struct ubd *dev, struct io_thread_req *io_req, 1257fc6b6a87SGabriel Krisman Bertazi struct request *req) 12581da177e4SLinus Torvalds { 1259fc6b6a87SGabriel Krisman Bertazi struct bio_vec bvec; 1260fc6b6a87SGabriel Krisman Bertazi struct req_iterator iter; 1261fc6b6a87SGabriel Krisman Bertazi int i = 0; 1262fc6b6a87SGabriel Krisman Bertazi unsigned long byte_offset = io_req->offset; 1263fc6b6a87SGabriel Krisman Bertazi int op = req_op(req); 1264ecb0a83eSChristoph Hellwig 1265fc6b6a87SGabriel Krisman Bertazi if (op == REQ_OP_WRITE_ZEROES || op == REQ_OP_DISCARD) { 1266fc6b6a87SGabriel Krisman Bertazi io_req->io_desc[0].buffer = NULL; 1267fc6b6a87SGabriel Krisman Bertazi io_req->io_desc[0].length = blk_rq_bytes(req); 1268fc6b6a87SGabriel Krisman Bertazi } else { 1269fc6b6a87SGabriel Krisman Bertazi rq_for_each_segment(bvec, req, iter) { 1270fc6b6a87SGabriel Krisman Bertazi BUG_ON(i >= io_req->desc_cnt); 1271fc6b6a87SGabriel Krisman Bertazi 127225d84545SChristoph Hellwig io_req->io_desc[i].buffer = bvec_virt(&bvec); 1273fc6b6a87SGabriel Krisman Bertazi io_req->io_desc[i].length = bvec.bv_len; 1274fc6b6a87SGabriel Krisman Bertazi i++; 1275fc6b6a87SGabriel Krisman Bertazi } 1276fc6b6a87SGabriel Krisman Bertazi } 1277fc6b6a87SGabriel Krisman Bertazi 1278fc6b6a87SGabriel Krisman Bertazi if (dev->cow.file) { 1279fc6b6a87SGabriel Krisman Bertazi for (i = 0; i < io_req->desc_cnt; i++) { 1280fc6b6a87SGabriel Krisman Bertazi cowify_req(io_req, &io_req->io_desc[i], byte_offset, 1281fc6b6a87SGabriel Krisman Bertazi dev->cow.bitmap, dev->cow.bitmap_offset, 1282fc6b6a87SGabriel Krisman Bertazi dev->cow.bitmap_len); 1283fc6b6a87SGabriel Krisman Bertazi byte_offset += io_req->io_desc[i].length; 1284fc6b6a87SGabriel Krisman Bertazi } 1285fc6b6a87SGabriel Krisman Bertazi 1286fc6b6a87SGabriel Krisman Bertazi } 1287fc6b6a87SGabriel Krisman Bertazi } 1288fc6b6a87SGabriel Krisman Bertazi 1289fc6b6a87SGabriel Krisman Bertazi static struct io_thread_req *ubd_alloc_req(struct ubd *dev, struct request *req, 1290fc6b6a87SGabriel Krisman Bertazi int desc_cnt) 1291fc6b6a87SGabriel Krisman Bertazi { 1292fc6b6a87SGabriel Krisman Bertazi struct io_thread_req *io_req; 1293fc6b6a87SGabriel Krisman Bertazi int i; 1294fc6b6a87SGabriel Krisman Bertazi 1295fc6b6a87SGabriel Krisman Bertazi io_req = kmalloc(sizeof(*io_req) + 1296fc6b6a87SGabriel Krisman Bertazi (desc_cnt * sizeof(struct io_desc)), 1297fc6b6a87SGabriel Krisman Bertazi GFP_ATOMIC); 1298ecb0a83eSChristoph Hellwig if (!io_req) 1299fc6b6a87SGabriel Krisman Bertazi return NULL; 130091acb21fSJeff Dike 130162f96cb0SJeff Dike io_req->req = req; 1302ecb0a83eSChristoph Hellwig if (dev->cow.file) 1303ecb0a83eSChristoph Hellwig io_req->fds[0] = dev->cow.fd; 1304ecb0a83eSChristoph Hellwig else 1305ecb0a83eSChristoph Hellwig io_req->fds[0] = dev->fd; 13060033dfd9SAnton Ivanov io_req->error = 0; 130753766defSAnton Ivanov io_req->sectorsize = SECTOR_SIZE; 1308ecb0a83eSChristoph Hellwig io_req->fds[1] = dev->fd; 1309fc6b6a87SGabriel Krisman Bertazi io_req->offset = (u64) blk_rq_pos(req) << SECTOR_SHIFT; 13101da177e4SLinus Torvalds io_req->offsets[0] = 0; 1311ecb0a83eSChristoph Hellwig io_req->offsets[1] = dev->cow.data_offset; 13121da177e4SLinus Torvalds 1313fc6b6a87SGabriel Krisman Bertazi for (i = 0 ; i < desc_cnt; i++) { 1314fc6b6a87SGabriel Krisman Bertazi io_req->io_desc[i].sector_mask = 0; 1315fc6b6a87SGabriel Krisman Bertazi io_req->io_desc[i].cow_offset = -1; 1316fc6b6a87SGabriel Krisman Bertazi } 1317fc6b6a87SGabriel Krisman Bertazi 1318fc6b6a87SGabriel Krisman Bertazi return io_req; 1319fc6b6a87SGabriel Krisman Bertazi } 1320fc6b6a87SGabriel Krisman Bertazi 1321fc6b6a87SGabriel Krisman Bertazi static int ubd_submit_request(struct ubd *dev, struct request *req) 1322fc6b6a87SGabriel Krisman Bertazi { 1323fc6b6a87SGabriel Krisman Bertazi int segs = 0; 1324fc6b6a87SGabriel Krisman Bertazi struct io_thread_req *io_req; 1325fc6b6a87SGabriel Krisman Bertazi int ret; 1326fc6b6a87SGabriel Krisman Bertazi int op = req_op(req); 1327fc6b6a87SGabriel Krisman Bertazi 1328fc6b6a87SGabriel Krisman Bertazi if (op == REQ_OP_FLUSH) 1329fc6b6a87SGabriel Krisman Bertazi segs = 0; 1330fc6b6a87SGabriel Krisman Bertazi else if (op == REQ_OP_WRITE_ZEROES || op == REQ_OP_DISCARD) 1331fc6b6a87SGabriel Krisman Bertazi segs = 1; 1332fc6b6a87SGabriel Krisman Bertazi else 1333fc6b6a87SGabriel Krisman Bertazi segs = blk_rq_nr_phys_segments(req); 1334fc6b6a87SGabriel Krisman Bertazi 1335fc6b6a87SGabriel Krisman Bertazi io_req = ubd_alloc_req(dev, req, segs); 1336fc6b6a87SGabriel Krisman Bertazi if (!io_req) 1337fc6b6a87SGabriel Krisman Bertazi return -ENOMEM; 1338fc6b6a87SGabriel Krisman Bertazi 1339fc6b6a87SGabriel Krisman Bertazi io_req->desc_cnt = segs; 1340fc6b6a87SGabriel Krisman Bertazi if (segs) 1341fc6b6a87SGabriel Krisman Bertazi ubd_map_req(dev, io_req, req); 13421da177e4SLinus Torvalds 1343ecb0a83eSChristoph Hellwig ret = os_write_file(thread_fd, &io_req, sizeof(io_req)); 1344ecb0a83eSChristoph Hellwig if (ret != sizeof(io_req)) { 1345ecb0a83eSChristoph Hellwig if (ret != -EAGAIN) 1346ecb0a83eSChristoph Hellwig pr_err("write to io thread failed: %d\n", -ret); 1347bc1d72e7SRichard Weinberger kfree(io_req); 1348bc1d72e7SRichard Weinberger } 1349ecb0a83eSChristoph Hellwig return ret; 1350bc1d72e7SRichard Weinberger } 1351bc1d72e7SRichard Weinberger 13524e6da0feSRichard Weinberger static blk_status_t ubd_queue_rq(struct blk_mq_hw_ctx *hctx, 13534e6da0feSRichard Weinberger const struct blk_mq_queue_data *bd) 13541da177e4SLinus Torvalds { 13556961cd4dSJens Axboe struct ubd *ubd_dev = hctx->queue->queuedata; 13564e6da0feSRichard Weinberger struct request *req = bd->rq; 135753766defSAnton Ivanov int ret = 0, res = BLK_STS_OK; 13581da177e4SLinus Torvalds 13594e6da0feSRichard Weinberger blk_mq_start_request(req); 1360a0044bdfSJeff Dike 13616961cd4dSJens Axboe spin_lock_irq(&ubd_dev->lock); 13626961cd4dSJens Axboe 136353766defSAnton Ivanov switch (req_op(req)) { 136453766defSAnton Ivanov case REQ_OP_FLUSH: 136553766defSAnton Ivanov case REQ_OP_READ: 136653766defSAnton Ivanov case REQ_OP_WRITE: 136750109b5aSAnton Ivanov case REQ_OP_DISCARD: 136850109b5aSAnton Ivanov case REQ_OP_WRITE_ZEROES: 1369fc6b6a87SGabriel Krisman Bertazi ret = ubd_submit_request(ubd_dev, req); 137050109b5aSAnton Ivanov break; 137153766defSAnton Ivanov default: 137253766defSAnton Ivanov WARN_ON_ONCE(1); 137353766defSAnton Ivanov res = BLK_STS_NOTSUPP; 137453766defSAnton Ivanov } 1375ecb0a83eSChristoph Hellwig 13766961cd4dSJens Axboe spin_unlock_irq(&ubd_dev->lock); 13776961cd4dSJens Axboe 1378d848074bSAnton Ivanov if (ret < 0) { 1379d848074bSAnton Ivanov if (ret == -ENOMEM) 1380d848074bSAnton Ivanov res = BLK_STS_RESOURCE; 1381d848074bSAnton Ivanov else 1382d848074bSAnton Ivanov res = BLK_STS_DEV_RESOURCE; 1383d848074bSAnton Ivanov } 13846961cd4dSJens Axboe 138553766defSAnton Ivanov return res; 138691acb21fSJeff Dike } 13874e6da0feSRichard Weinberger 1388a885c8c4SChristoph Hellwig static int ubd_getgeo(struct block_device *bdev, struct hd_geometry *geo) 1389a885c8c4SChristoph Hellwig { 13907d314e34SPaolo 'Blaisorblade' Giarrusso struct ubd *ubd_dev = bdev->bd_disk->private_data; 1391a885c8c4SChristoph Hellwig 1392a885c8c4SChristoph Hellwig geo->heads = 128; 1393a885c8c4SChristoph Hellwig geo->sectors = 32; 13947d314e34SPaolo 'Blaisorblade' Giarrusso geo->cylinders = ubd_dev->size / (128 * 32 * 512); 1395a885c8c4SChristoph Hellwig return 0; 1396a885c8c4SChristoph Hellwig } 1397a885c8c4SChristoph Hellwig 1398a625c998SAl Viro static int ubd_ioctl(struct block_device *bdev, fmode_t mode, 13991da177e4SLinus Torvalds unsigned int cmd, unsigned long arg) 14001da177e4SLinus Torvalds { 1401a625c998SAl Viro struct ubd *ubd_dev = bdev->bd_disk->private_data; 140273855e13SBartlomiej Zolnierkiewicz u16 ubd_id[ATA_ID_WORDS]; 14031da177e4SLinus Torvalds 14041da177e4SLinus Torvalds switch (cmd) { 14051da177e4SLinus Torvalds struct cdrom_volctrl volume; 14061da177e4SLinus Torvalds case HDIO_GET_IDENTITY: 140773855e13SBartlomiej Zolnierkiewicz memset(&ubd_id, 0, ATA_ID_WORDS * 2); 140873855e13SBartlomiej Zolnierkiewicz ubd_id[ATA_ID_CYLS] = ubd_dev->size / (128 * 32 * 512); 140973855e13SBartlomiej Zolnierkiewicz ubd_id[ATA_ID_HEADS] = 128; 141073855e13SBartlomiej Zolnierkiewicz ubd_id[ATA_ID_SECTORS] = 32; 14111da177e4SLinus Torvalds if(copy_to_user((char __user *) arg, (char *) &ubd_id, 14121da177e4SLinus Torvalds sizeof(ubd_id))) 1413dc764e50SJeff Dike return -EFAULT; 1414dc764e50SJeff Dike return 0; 14151da177e4SLinus Torvalds 14161da177e4SLinus Torvalds case CDROMVOLREAD: 14171da177e4SLinus Torvalds if(copy_from_user(&volume, (char __user *) arg, sizeof(volume))) 1418dc764e50SJeff Dike return -EFAULT; 14191da177e4SLinus Torvalds volume.channel0 = 255; 14201da177e4SLinus Torvalds volume.channel1 = 255; 14211da177e4SLinus Torvalds volume.channel2 = 255; 14221da177e4SLinus Torvalds volume.channel3 = 255; 14231da177e4SLinus Torvalds if(copy_to_user((char __user *) arg, &volume, sizeof(volume))) 1424dc764e50SJeff Dike return -EFAULT; 1425dc764e50SJeff Dike return 0; 14261da177e4SLinus Torvalds } 1427dc764e50SJeff Dike return -EINVAL; 14281da177e4SLinus Torvalds } 14291da177e4SLinus Torvalds 1430a43c8316SAnton Ivanov static int map_error(int error_code) 1431a43c8316SAnton Ivanov { 1432a43c8316SAnton Ivanov switch (error_code) { 1433a43c8316SAnton Ivanov case 0: 1434a43c8316SAnton Ivanov return BLK_STS_OK; 1435a43c8316SAnton Ivanov case ENOSYS: 1436a43c8316SAnton Ivanov case EOPNOTSUPP: 1437a43c8316SAnton Ivanov return BLK_STS_NOTSUPP; 1438a43c8316SAnton Ivanov case ENOSPC: 1439a43c8316SAnton Ivanov return BLK_STS_NOSPC; 1440a43c8316SAnton Ivanov } 1441a43c8316SAnton Ivanov return BLK_STS_IOERR; 1442a43c8316SAnton Ivanov } 1443a43c8316SAnton Ivanov 1444a41421edSAnton Ivanov /* 1445a41421edSAnton Ivanov * Everything from here onwards *IS NOT PART OF THE KERNEL* 1446a41421edSAnton Ivanov * 1447a41421edSAnton Ivanov * The following functions are part of UML hypervisor code. 1448a41421edSAnton Ivanov * All functions from here onwards are executed as a helper 1449a41421edSAnton Ivanov * thread and are not allowed to execute any kernel functions. 1450a41421edSAnton Ivanov * 1451a41421edSAnton Ivanov * Any communication must occur strictly via shared memory and IPC. 1452a41421edSAnton Ivanov * 1453a41421edSAnton Ivanov * Do not add printks, locks, kernel memory operations, etc - it 1454a41421edSAnton Ivanov * will result in unpredictable behaviour and/or crashes. 1455a41421edSAnton Ivanov */ 1456a41421edSAnton Ivanov 1457fc6b6a87SGabriel Krisman Bertazi static int update_bitmap(struct io_thread_req *req, struct io_desc *segment) 14581da177e4SLinus Torvalds { 145991acb21fSJeff Dike int n; 146091acb21fSJeff Dike 1461fc6b6a87SGabriel Krisman Bertazi if (segment->cow_offset == -1) 1462a43c8316SAnton Ivanov return map_error(0); 146391acb21fSJeff Dike 1464fc6b6a87SGabriel Krisman Bertazi n = os_pwrite_file(req->fds[1], &segment->bitmap_words, 1465fc6b6a87SGabriel Krisman Bertazi sizeof(segment->bitmap_words), segment->cow_offset); 1466fc6b6a87SGabriel Krisman Bertazi if (n != sizeof(segment->bitmap_words)) 1467a43c8316SAnton Ivanov return map_error(-n); 146891acb21fSJeff Dike 1469a43c8316SAnton Ivanov return map_error(0); 147091acb21fSJeff Dike } 147191acb21fSJeff Dike 1472fc6b6a87SGabriel Krisman Bertazi static void do_io(struct io_thread_req *req, struct io_desc *desc) 147391acb21fSJeff Dike { 147450109b5aSAnton Ivanov char *buf = NULL; 147591acb21fSJeff Dike unsigned long len; 147691acb21fSJeff Dike int n, nsectors, start, end, bit; 14771da177e4SLinus Torvalds __u64 off; 14781da177e4SLinus Torvalds 147950109b5aSAnton Ivanov /* FLUSH is really a special case, we cannot "case" it with others */ 148050109b5aSAnton Ivanov 1481a43c8316SAnton Ivanov if (req_op(req->req) == REQ_OP_FLUSH) { 1482805f11a0SRichard Weinberger /* fds[0] is always either the rw image or our cow file */ 1483a41421edSAnton Ivanov req->error = map_error(-os_sync_file(req->fds[0])); 1484805f11a0SRichard Weinberger return; 1485805f11a0SRichard Weinberger } 1486805f11a0SRichard Weinberger 1487fc6b6a87SGabriel Krisman Bertazi nsectors = desc->length / req->sectorsize; 14881da177e4SLinus Torvalds start = 0; 14891da177e4SLinus Torvalds do { 1490fc6b6a87SGabriel Krisman Bertazi bit = ubd_test_bit(start, (unsigned char *) &desc->sector_mask); 14911da177e4SLinus Torvalds end = start; 14921da177e4SLinus Torvalds while((end < nsectors) && 1493fc6b6a87SGabriel Krisman Bertazi (ubd_test_bit(end, (unsigned char *) &desc->sector_mask) == bit)) 14941da177e4SLinus Torvalds end++; 14951da177e4SLinus Torvalds 149691acb21fSJeff Dike off = req->offset + req->offsets[bit] + 14971da177e4SLinus Torvalds start * req->sectorsize; 14981da177e4SLinus Torvalds len = (end - start) * req->sectorsize; 1499fc6b6a87SGabriel Krisman Bertazi if (desc->buffer != NULL) 1500fc6b6a87SGabriel Krisman Bertazi buf = &desc->buffer[start * req->sectorsize]; 15011da177e4SLinus Torvalds 150250109b5aSAnton Ivanov switch (req_op(req->req)) { 150350109b5aSAnton Ivanov case REQ_OP_READ: 150491acb21fSJeff Dike n = 0; 150591acb21fSJeff Dike do { 150691acb21fSJeff Dike buf = &buf[n]; 150791acb21fSJeff Dike len -= n; 15088c6157b6SAnton Ivanov n = os_pread_file(req->fds[bit], buf, len, off); 150991acb21fSJeff Dike if (n < 0) { 1510a43c8316SAnton Ivanov req->error = map_error(-n); 15111da177e4SLinus Torvalds return; 15121da177e4SLinus Torvalds } 151391acb21fSJeff Dike } while((n < len) && (n != 0)); 151491acb21fSJeff Dike if (n < len) memset(&buf[n], 0, len - n); 151550109b5aSAnton Ivanov break; 151650109b5aSAnton Ivanov case REQ_OP_WRITE: 15178c6157b6SAnton Ivanov n = os_pwrite_file(req->fds[bit], buf, len, off); 151891acb21fSJeff Dike if(n != len){ 1519a43c8316SAnton Ivanov req->error = map_error(-n); 152091acb21fSJeff Dike return; 152191acb21fSJeff Dike } 152250109b5aSAnton Ivanov break; 152350109b5aSAnton Ivanov case REQ_OP_DISCARD: 152450109b5aSAnton Ivanov case REQ_OP_WRITE_ZEROES: 152550109b5aSAnton Ivanov n = os_falloc_punch(req->fds[bit], off, len); 152650109b5aSAnton Ivanov if (n) { 152750109b5aSAnton Ivanov req->error = map_error(-n); 152850109b5aSAnton Ivanov return; 152950109b5aSAnton Ivanov } 153050109b5aSAnton Ivanov break; 153150109b5aSAnton Ivanov default: 153250109b5aSAnton Ivanov WARN_ON_ONCE(1); 153350109b5aSAnton Ivanov req->error = BLK_STS_NOTSUPP; 153450109b5aSAnton Ivanov return; 153591acb21fSJeff Dike } 15361da177e4SLinus Torvalds 15371da177e4SLinus Torvalds start = end; 15381da177e4SLinus Torvalds } while(start < nsectors); 153991acb21fSJeff Dike 1540fc6b6a87SGabriel Krisman Bertazi req->offset += len; 1541fc6b6a87SGabriel Krisman Bertazi req->error = update_bitmap(req, desc); 15421da177e4SLinus Torvalds } 154391acb21fSJeff Dike 154491acb21fSJeff Dike /* Changed in start_io_thread, which is serialized by being called only 154591acb21fSJeff Dike * from ubd_init, which is an initcall. 154691acb21fSJeff Dike */ 154791acb21fSJeff Dike int kernel_fd = -1; 154891acb21fSJeff Dike 1549d8d7c28eSPaolo 'Blaisorblade' Giarrusso /* Only changed by the io thread. XXX: currently unused. */ 1550d8d7c28eSPaolo 'Blaisorblade' Giarrusso static int io_count = 0; 155191acb21fSJeff Dike 155291acb21fSJeff Dike int io_thread(void *arg) 155391acb21fSJeff Dike { 1554f88f0bdfSAnton Ivanov int n, count, written, res; 155591acb21fSJeff Dike 155691d44ff8SRichard Weinberger os_fix_helper_signals(); 155791d44ff8SRichard Weinberger 155891acb21fSJeff Dike while(1){ 1559f88f0bdfSAnton Ivanov n = bulk_req_safe_read( 1560f88f0bdfSAnton Ivanov kernel_fd, 1561f88f0bdfSAnton Ivanov io_req_buffer, 1562f88f0bdfSAnton Ivanov &io_remainder, 1563f88f0bdfSAnton Ivanov &io_remainder_size, 1564f88f0bdfSAnton Ivanov UBD_REQ_BUFFER_SIZE 1565f88f0bdfSAnton Ivanov ); 1566e355b2f5SGabriel Krisman Bertazi if (n <= 0) { 1567e355b2f5SGabriel Krisman Bertazi if (n == -EAGAIN) 1568f88f0bdfSAnton Ivanov ubd_read_poll(-1); 1569e355b2f5SGabriel Krisman Bertazi 1570f88f0bdfSAnton Ivanov continue; 1571f88f0bdfSAnton Ivanov } 1572f88f0bdfSAnton Ivanov 1573f88f0bdfSAnton Ivanov for (count = 0; count < n/sizeof(struct io_thread_req *); count++) { 1574fc6b6a87SGabriel Krisman Bertazi struct io_thread_req *req = (*io_req_buffer)[count]; 1575fc6b6a87SGabriel Krisman Bertazi int i; 1576fc6b6a87SGabriel Krisman Bertazi 1577f88f0bdfSAnton Ivanov io_count++; 1578fc6b6a87SGabriel Krisman Bertazi for (i = 0; !req->error && i < req->desc_cnt; i++) 1579fc6b6a87SGabriel Krisman Bertazi do_io(req, &(req->io_desc[i])); 1580fc6b6a87SGabriel Krisman Bertazi 1581f88f0bdfSAnton Ivanov } 1582f88f0bdfSAnton Ivanov 1583f88f0bdfSAnton Ivanov written = 0; 1584f88f0bdfSAnton Ivanov 1585f88f0bdfSAnton Ivanov do { 15866e682d53SGabriel Krisman Bertazi res = os_write_file(kernel_fd, 15876e682d53SGabriel Krisman Bertazi ((char *) io_req_buffer) + written, 15886e682d53SGabriel Krisman Bertazi n - written); 1589ff6a1798SAnton Ivanov if (res >= 0) { 1590f88f0bdfSAnton Ivanov written += res; 159191acb21fSJeff Dike } 1592f88f0bdfSAnton Ivanov if (written < n) { 1593f88f0bdfSAnton Ivanov ubd_write_poll(-1); 1594f88f0bdfSAnton Ivanov } 1595f88f0bdfSAnton Ivanov } while (written < n); 159691acb21fSJeff Dike } 159791acb21fSJeff Dike 15981b57e9c2SJeff Dike return 0; 15991b57e9c2SJeff Dike } 1600