]>
Commit | Line | Data |
---|---|---|
6c29256c | 1 | /* |
1da177e4 LT |
2 | * Copyright (C) 2000 Jeff Dike (jdike@karaya.com) |
3 | * Licensed under the GPL | |
4 | */ | |
5 | ||
6 | /* 2001-09-28...2002-04-17 | |
7 | * Partition stuff by James_McMechan@hotmail.com | |
8 | * old style ubd by setting UBD_SHIFT to 0 | |
9 | * 2002-09-27...2002-10-18 massive tinkering for 2.5 | |
10 | * partitions have changed in 2.5 | |
11 | * 2003-01-29 more tinkering for 2.5.59-1 | |
12 | * This should now address the sysfs problems and has | |
13 | * the symlink for devfs to allow for booting with | |
14 | * the common /dev/ubd/discX/... names rather than | |
15 | * only /dev/ubdN/discN this version also has lots of | |
16 | * clean ups preparing for ubd-many. | |
17 | * James McMechan | |
18 | */ | |
19 | ||
1da177e4 LT |
20 | #define UBD_SHIFT 4 |
21 | ||
e16f5350 | 22 | #include "linux/kernel.h" |
1da177e4 LT |
23 | #include "linux/module.h" |
24 | #include "linux/blkdev.h" | |
73855e13 | 25 | #include "linux/ata.h" |
1da177e4 LT |
26 | #include "linux/hdreg.h" |
27 | #include "linux/init.h" | |
1da177e4 LT |
28 | #include "linux/cdrom.h" |
29 | #include "linux/proc_fs.h" | |
30 | #include "linux/ctype.h" | |
31 | #include "linux/capability.h" | |
32 | #include "linux/mm.h" | |
33 | #include "linux/vmalloc.h" | |
34 | #include "linux/blkpg.h" | |
35 | #include "linux/genhd.h" | |
36 | #include "linux/spinlock.h" | |
d052d1be | 37 | #include "linux/platform_device.h" |
23464ffa | 38 | #include "linux/scatterlist.h" |
1da177e4 LT |
39 | #include "asm/segment.h" |
40 | #include "asm/uaccess.h" | |
41 | #include "asm/irq.h" | |
42 | #include "asm/types.h" | |
43 | #include "asm/tlbflush.h" | |
1da177e4 LT |
44 | #include "mem_user.h" |
45 | #include "kern_util.h" | |
46 | #include "kern.h" | |
47 | #include "mconsole_kern.h" | |
48 | #include "init.h" | |
49 | #include "irq_user.h" | |
50 | #include "irq_kern.h" | |
51 | #include "ubd_user.h" | |
1da177e4 LT |
52 | #include "os.h" |
53 | #include "mem.h" | |
54 | #include "mem_kern.h" | |
55 | #include "cow.h" | |
56 | ||
7b9014c1 | 57 | enum ubd_req { UBD_READ, UBD_WRITE }; |
1da177e4 LT |
58 | |
59 | struct io_thread_req { | |
62f96cb0 | 60 | struct request *req; |
91acb21f | 61 | enum ubd_req op; |
1da177e4 LT |
62 | int fds[2]; |
63 | unsigned long offsets[2]; | |
64 | unsigned long long offset; | |
65 | unsigned long length; | |
66 | char *buffer; | |
67 | int sectorsize; | |
91acb21f JD |
68 | unsigned long sector_mask; |
69 | unsigned long long cow_offset; | |
70 | unsigned long bitmap_words[2]; | |
1da177e4 LT |
71 | int error; |
72 | }; | |
73 | ||
91acb21f | 74 | static inline int ubd_test_bit(__u64 bit, unsigned char *data) |
1da177e4 LT |
75 | { |
76 | __u64 n; | |
77 | int bits, off; | |
78 | ||
91acb21f | 79 | bits = sizeof(data[0]) * 8; |
1da177e4 LT |
80 | n = bit / bits; |
81 | off = bit % bits; | |
dc764e50 | 82 | return (data[n] & (1 << off)) != 0; |
1da177e4 LT |
83 | } |
84 | ||
91acb21f | 85 | static inline void ubd_set_bit(__u64 bit, unsigned char *data) |
1da177e4 LT |
86 | { |
87 | __u64 n; | |
88 | int bits, off; | |
89 | ||
91acb21f | 90 | bits = sizeof(data[0]) * 8; |
1da177e4 LT |
91 | n = bit / bits; |
92 | off = bit % bits; | |
91acb21f | 93 | data[n] |= (1 << off); |
1da177e4 LT |
94 | } |
95 | /*End stuff from ubd_user.h*/ | |
96 | ||
97 | #define DRIVER_NAME "uml-blkdev" | |
98 | ||
d7fb2c38 | 99 | static DEFINE_MUTEX(ubd_lock); |
1da177e4 | 100 | |
a625c998 AV |
101 | static int ubd_open(struct block_device *bdev, fmode_t mode); |
102 | static int ubd_release(struct gendisk *disk, fmode_t mode); | |
103 | static int ubd_ioctl(struct block_device *bdev, fmode_t mode, | |
1da177e4 | 104 | unsigned int cmd, unsigned long arg); |
a885c8c4 | 105 | static int ubd_getgeo(struct block_device *bdev, struct hd_geometry *geo); |
1da177e4 | 106 | |
97d88ac8 | 107 | #define MAX_DEV (16) |
1da177e4 | 108 | |
83d5cde4 | 109 | static const struct block_device_operations ubd_blops = { |
1da177e4 | 110 | .owner = THIS_MODULE, |
a625c998 AV |
111 | .open = ubd_open, |
112 | .release = ubd_release, | |
113 | .ioctl = ubd_ioctl, | |
a885c8c4 | 114 | .getgeo = ubd_getgeo, |
1da177e4 LT |
115 | }; |
116 | ||
1da177e4 | 117 | /* Protected by ubd_lock */ |
792dd4fc | 118 | static int fake_major = UBD_MAJOR; |
1da177e4 LT |
119 | static struct gendisk *ubd_gendisk[MAX_DEV]; |
120 | static struct gendisk *fake_gendisk[MAX_DEV]; | |
6c29256c | 121 | |
1da177e4 LT |
122 | #ifdef CONFIG_BLK_DEV_UBD_SYNC |
123 | #define OPEN_FLAGS ((struct openflags) { .r = 1, .w = 1, .s = 1, .c = 0, \ | |
124 | .cl = 1 }) | |
125 | #else | |
126 | #define OPEN_FLAGS ((struct openflags) { .r = 1, .w = 1, .s = 0, .c = 0, \ | |
127 | .cl = 1 }) | |
128 | #endif | |
1da177e4 LT |
129 | static struct openflags global_openflags = OPEN_FLAGS; |
130 | ||
131 | struct cow { | |
2a9d32f6 | 132 | /* backing file name */ |
1da177e4 | 133 | char *file; |
2a9d32f6 | 134 | /* backing file fd */ |
1da177e4 LT |
135 | int fd; |
136 | unsigned long *bitmap; | |
137 | unsigned long bitmap_len; | |
138 | int bitmap_offset; | |
dc764e50 | 139 | int data_offset; |
1da177e4 LT |
140 | }; |
141 | ||
a0044bdf JD |
142 | #define MAX_SG 64 |
143 | ||
1da177e4 | 144 | struct ubd { |
a0044bdf | 145 | struct list_head restart; |
2a9d32f6 PBG |
146 | /* name (and fd, below) of the file opened for writing, either the |
147 | * backing or the cow file. */ | |
1da177e4 LT |
148 | char *file; |
149 | int count; | |
150 | int fd; | |
151 | __u64 size; | |
152 | struct openflags boot_openflags; | |
153 | struct openflags openflags; | |
84e945e3 PBG |
154 | unsigned shared:1; |
155 | unsigned no_cow:1; | |
1da177e4 LT |
156 | struct cow cow; |
157 | struct platform_device pdev; | |
62f96cb0 JD |
158 | struct request_queue *queue; |
159 | spinlock_t lock; | |
a0044bdf JD |
160 | struct scatterlist sg[MAX_SG]; |
161 | struct request *request; | |
162 | int start_sg, end_sg; | |
1da177e4 LT |
163 | }; |
164 | ||
165 | #define DEFAULT_COW { \ | |
166 | .file = NULL, \ | |
dc764e50 JD |
167 | .fd = -1, \ |
168 | .bitmap = NULL, \ | |
1da177e4 | 169 | .bitmap_offset = 0, \ |
dc764e50 | 170 | .data_offset = 0, \ |
1da177e4 LT |
171 | } |
172 | ||
173 | #define DEFAULT_UBD { \ | |
174 | .file = NULL, \ | |
175 | .count = 0, \ | |
176 | .fd = -1, \ | |
177 | .size = -1, \ | |
178 | .boot_openflags = OPEN_FLAGS, \ | |
179 | .openflags = OPEN_FLAGS, \ | |
dc764e50 | 180 | .no_cow = 0, \ |
6c29256c | 181 | .shared = 0, \ |
dc764e50 | 182 | .cow = DEFAULT_COW, \ |
62f96cb0 | 183 | .lock = SPIN_LOCK_UNLOCKED, \ |
a0044bdf JD |
184 | .request = NULL, \ |
185 | .start_sg = 0, \ | |
186 | .end_sg = 0, \ | |
1da177e4 LT |
187 | } |
188 | ||
b8831a1d | 189 | /* Protected by ubd_lock */ |
5dc62b1b | 190 | static struct ubd ubd_devs[MAX_DEV] = { [0 ... MAX_DEV - 1] = DEFAULT_UBD }; |
1da177e4 | 191 | |
1da177e4 LT |
192 | /* Only changed by fake_ide_setup which is a setup */ |
193 | static int fake_ide = 0; | |
194 | static struct proc_dir_entry *proc_ide_root = NULL; | |
195 | static struct proc_dir_entry *proc_ide = NULL; | |
196 | ||
197 | static void make_proc_ide(void) | |
198 | { | |
199 | proc_ide_root = proc_mkdir("ide", NULL); | |
200 | proc_ide = proc_mkdir("ide0", proc_ide_root); | |
201 | } | |
202 | ||
203 | static int proc_ide_read_media(char *page, char **start, off_t off, int count, | |
204 | int *eof, void *data) | |
205 | { | |
206 | int len; | |
207 | ||
208 | strcpy(page, "disk\n"); | |
209 | len = strlen("disk\n"); | |
210 | len -= off; | |
211 | if (len < count){ | |
212 | *eof = 1; | |
213 | if (len <= 0) return 0; | |
214 | } | |
215 | else len = count; | |
216 | *start = page + off; | |
217 | return len; | |
218 | } | |
219 | ||
c0a9290e | 220 | static void make_ide_entries(const char *dev_name) |
1da177e4 LT |
221 | { |
222 | struct proc_dir_entry *dir, *ent; | |
223 | char name[64]; | |
224 | ||
225 | if(proc_ide_root == NULL) make_proc_ide(); | |
226 | ||
227 | dir = proc_mkdir(dev_name, proc_ide); | |
228 | if(!dir) return; | |
229 | ||
230 | ent = create_proc_entry("media", S_IFREG|S_IRUGO, dir); | |
231 | if(!ent) return; | |
1da177e4 LT |
232 | ent->data = NULL; |
233 | ent->read_proc = proc_ide_read_media; | |
234 | ent->write_proc = NULL; | |
c0a9290e | 235 | snprintf(name, sizeof(name), "ide0/%s", dev_name); |
1da177e4 LT |
236 | proc_symlink(dev_name, proc_ide_root, name); |
237 | } | |
238 | ||
239 | static int fake_ide_setup(char *str) | |
240 | { | |
241 | fake_ide = 1; | |
dc764e50 | 242 | return 1; |
1da177e4 LT |
243 | } |
244 | ||
245 | __setup("fake_ide", fake_ide_setup); | |
246 | ||
247 | __uml_help(fake_ide_setup, | |
248 | "fake_ide\n" | |
249 | " Create ide0 entries that map onto ubd devices.\n\n" | |
250 | ); | |
251 | ||
252 | static int parse_unit(char **ptr) | |
253 | { | |
254 | char *str = *ptr, *end; | |
255 | int n = -1; | |
256 | ||
257 | if(isdigit(*str)) { | |
258 | n = simple_strtoul(str, &end, 0); | |
259 | if(end == str) | |
dc764e50 | 260 | return -1; |
1da177e4 LT |
261 | *ptr = end; |
262 | } | |
97d88ac8 | 263 | else if (('a' <= *str) && (*str <= 'z')) { |
1da177e4 LT |
264 | n = *str - 'a'; |
265 | str++; | |
266 | *ptr = str; | |
267 | } | |
dc764e50 | 268 | return n; |
1da177e4 LT |
269 | } |
270 | ||
d8d7c28e PBG |
271 | /* If *index_out == -1 at exit, the passed option was a general one; |
272 | * otherwise, the str pointer is used (and owned) inside ubd_devs array, so it | |
273 | * should not be freed on exit. | |
274 | */ | |
f28169d2 | 275 | static int ubd_setup_common(char *str, int *index_out, char **error_out) |
1da177e4 | 276 | { |
7d314e34 | 277 | struct ubd *ubd_dev; |
1da177e4 LT |
278 | struct openflags flags = global_openflags; |
279 | char *backing_file; | |
b8831a1d | 280 | int n, err = 0, i; |
1da177e4 LT |
281 | |
282 | if(index_out) *index_out = -1; | |
283 | n = *str; | |
284 | if(n == '='){ | |
285 | char *end; | |
286 | int major; | |
287 | ||
288 | str++; | |
1da177e4 LT |
289 | if(!strcmp(str, "sync")){ |
290 | global_openflags = of_sync(global_openflags); | |
b8831a1d | 291 | goto out1; |
1da177e4 | 292 | } |
b8831a1d JD |
293 | |
294 | err = -EINVAL; | |
1da177e4 LT |
295 | major = simple_strtoul(str, &end, 0); |
296 | if((*end != '\0') || (end == str)){ | |
f28169d2 | 297 | *error_out = "Didn't parse major number"; |
b8831a1d | 298 | goto out1; |
1da177e4 LT |
299 | } |
300 | ||
f28169d2 | 301 | mutex_lock(&ubd_lock); |
792dd4fc | 302 | if (fake_major != UBD_MAJOR) { |
f28169d2 JD |
303 | *error_out = "Can't assign a fake major twice"; |
304 | goto out1; | |
305 | } | |
6c29256c | 306 | |
f28169d2 | 307 | fake_major = major; |
1da177e4 LT |
308 | |
309 | printk(KERN_INFO "Setting extra ubd major number to %d\n", | |
310 | major); | |
f28169d2 JD |
311 | err = 0; |
312 | out1: | |
313 | mutex_unlock(&ubd_lock); | |
314 | return err; | |
1da177e4 LT |
315 | } |
316 | ||
317 | n = parse_unit(&str); | |
318 | if(n < 0){ | |
f28169d2 JD |
319 | *error_out = "Couldn't parse device number"; |
320 | return -EINVAL; | |
1da177e4 LT |
321 | } |
322 | if(n >= MAX_DEV){ | |
f28169d2 JD |
323 | *error_out = "Device number out of range"; |
324 | return 1; | |
1da177e4 LT |
325 | } |
326 | ||
f28169d2 | 327 | err = -EBUSY; |
d7fb2c38 | 328 | mutex_lock(&ubd_lock); |
1da177e4 | 329 | |
7d314e34 PBG |
330 | ubd_dev = &ubd_devs[n]; |
331 | if(ubd_dev->file != NULL){ | |
f28169d2 | 332 | *error_out = "Device is already configured"; |
1da177e4 LT |
333 | goto out; |
334 | } | |
335 | ||
336 | if (index_out) | |
337 | *index_out = n; | |
338 | ||
f28169d2 | 339 | err = -EINVAL; |
6c29256c | 340 | for (i = 0; i < sizeof("rscd="); i++) { |
1da177e4 LT |
341 | switch (*str) { |
342 | case 'r': | |
343 | flags.w = 0; | |
344 | break; | |
345 | case 's': | |
346 | flags.s = 1; | |
347 | break; | |
348 | case 'd': | |
7d314e34 | 349 | ubd_dev->no_cow = 1; |
1da177e4 | 350 | break; |
6c29256c | 351 | case 'c': |
7d314e34 | 352 | ubd_dev->shared = 1; |
6c29256c | 353 | break; |
1da177e4 LT |
354 | case '=': |
355 | str++; | |
356 | goto break_loop; | |
357 | default: | |
f28169d2 JD |
358 | *error_out = "Expected '=' or flag letter " |
359 | "(r, s, c, or d)"; | |
1da177e4 LT |
360 | goto out; |
361 | } | |
362 | str++; | |
363 | } | |
364 | ||
f28169d2 JD |
365 | if (*str == '=') |
366 | *error_out = "Too many flags specified"; | |
367 | else | |
368 | *error_out = "Missing '='"; | |
1da177e4 LT |
369 | goto out; |
370 | ||
371 | break_loop: | |
1da177e4 LT |
372 | backing_file = strchr(str, ','); |
373 | ||
f28169d2 | 374 | if (backing_file == NULL) |
1da177e4 | 375 | backing_file = strchr(str, ':'); |
1da177e4 | 376 | |
f28169d2 JD |
377 | if(backing_file != NULL){ |
378 | if(ubd_dev->no_cow){ | |
379 | *error_out = "Can't specify both 'd' and a cow file"; | |
380 | goto out; | |
381 | } | |
1da177e4 LT |
382 | else { |
383 | *backing_file = '\0'; | |
384 | backing_file++; | |
385 | } | |
386 | } | |
f28169d2 | 387 | err = 0; |
7d314e34 PBG |
388 | ubd_dev->file = str; |
389 | ubd_dev->cow.file = backing_file; | |
390 | ubd_dev->boot_openflags = flags; | |
1da177e4 | 391 | out: |
d7fb2c38 | 392 | mutex_unlock(&ubd_lock); |
f28169d2 | 393 | return err; |
1da177e4 LT |
394 | } |
395 | ||
396 | static int ubd_setup(char *str) | |
397 | { | |
f28169d2 JD |
398 | char *error; |
399 | int err; | |
400 | ||
401 | err = ubd_setup_common(str, NULL, &error); | |
402 | if(err) | |
403 | printk(KERN_ERR "Failed to initialize device with \"%s\" : " | |
404 | "%s\n", str, error); | |
405 | return 1; | |
1da177e4 LT |
406 | } |
407 | ||
408 | __setup("ubd", ubd_setup); | |
409 | __uml_help(ubd_setup, | |
410 | "ubd<n><flags>=<filename>[(:|,)<filename2>]\n" | |
411 | " This is used to associate a device with a file in the underlying\n" | |
412 | " filesystem. When specifying two filenames, the first one is the\n" | |
413 | " COW name and the second is the backing file name. As separator you can\n" | |
414 | " use either a ':' or a ',': the first one allows writing things like;\n" | |
415 | " ubd0=~/Uml/root_cow:~/Uml/root_backing_file\n" | |
416 | " while with a ',' the shell would not expand the 2nd '~'.\n" | |
f28169d2 | 417 | " When using only one filename, UML will detect whether to treat it like\n" |
1da177e4 LT |
418 | " a COW file or a backing file. To override this detection, add the 'd'\n" |
419 | " flag:\n" | |
420 | " ubd0d=BackingFile\n" | |
421 | " Usually, there is a filesystem in the file, but \n" | |
422 | " that's not required. Swap devices containing swap files can be\n" | |
423 | " specified like this. Also, a file which doesn't contain a\n" | |
424 | " filesystem can have its contents read in the virtual \n" | |
425 | " machine by running 'dd' on the device. <n> must be in the range\n" | |
426 | " 0 to 7. Appending an 'r' to the number will cause that device\n" | |
427 | " to be mounted read-only. For example ubd1r=./ext_fs. Appending\n" | |
20ede453 JD |
428 | " an 's' will cause data to be written to disk on the host immediately.\n" |
429 | " 'c' will cause the device to be treated as being shared between multiple\n" | |
430 | " UMLs and file locking will be turned off - this is appropriate for a\n" | |
431 | " cluster filesystem and inappropriate at almost all other times.\n\n" | |
1da177e4 LT |
432 | ); |
433 | ||
8299ca5c | 434 | static int udb_setup(char *str) |
1da177e4 LT |
435 | { |
436 | printk("udb%s specified on command line is almost certainly a ubd -> " | |
437 | "udb TYPO\n", str); | |
dc764e50 | 438 | return 1; |
1da177e4 LT |
439 | } |
440 | ||
441 | __setup("udb", udb_setup); | |
442 | __uml_help(udb_setup, | |
443 | "udb\n" | |
0894e27e JD |
444 | " This option is here solely to catch ubd -> udb typos, which can be\n" |
445 | " to impossible to catch visually unless you specifically look for\n" | |
446 | " them. The only result of any option starting with 'udb' is an error\n" | |
1da177e4 LT |
447 | " in the boot output.\n\n" |
448 | ); | |
449 | ||
165125e1 | 450 | static void do_ubd_request(struct request_queue * q); |
91acb21f JD |
451 | |
452 | /* Only changed by ubd_init, which is an initcall. */ | |
5dc62b1b | 453 | static int thread_fd = -1; |
a0044bdf JD |
454 | static LIST_HEAD(restart); |
455 | ||
2fe30a34 | 456 | /* XXX - move this inside ubd_intr. */ |
62f96cb0 | 457 | /* Called without dev->lock held, and only in interrupt context. */ |
91acb21f | 458 | static void ubd_handler(void) |
1da177e4 | 459 | { |
2adcec21 | 460 | struct io_thread_req *req; |
a0044bdf JD |
461 | struct ubd *ubd; |
462 | struct list_head *list, *next_ele; | |
463 | unsigned long flags; | |
91acb21f JD |
464 | int n; |
465 | ||
a0044bdf | 466 | while(1){ |
a6ea4cce JD |
467 | n = os_read_file(thread_fd, &req, |
468 | sizeof(struct io_thread_req *)); | |
a0044bdf JD |
469 | if(n != sizeof(req)){ |
470 | if(n == -EAGAIN) | |
471 | break; | |
472 | printk(KERN_ERR "spurious interrupt in ubd_handler, " | |
473 | "err = %d\n", -n); | |
474 | return; | |
475 | } | |
62f96cb0 | 476 | |
4d6c84d9 | 477 | blk_end_request(req->req, 0, req->length); |
2adcec21 | 478 | kfree(req); |
a0044bdf | 479 | } |
62f96cb0 | 480 | reactivate_fd(thread_fd, UBD_IRQ); |
a0044bdf JD |
481 | |
482 | list_for_each_safe(list, next_ele, &restart){ | |
483 | ubd = container_of(list, struct ubd, restart); | |
484 | list_del_init(&ubd->restart); | |
485 | spin_lock_irqsave(&ubd->lock, flags); | |
486 | do_ubd_request(ubd->queue); | |
487 | spin_unlock_irqrestore(&ubd->lock, flags); | |
488 | } | |
1da177e4 LT |
489 | } |
490 | ||
7bea96fd | 491 | static irqreturn_t ubd_intr(int irq, void *dev) |
1da177e4 | 492 | { |
91acb21f | 493 | ubd_handler(); |
dc764e50 | 494 | return IRQ_HANDLED; |
91acb21f | 495 | } |
09ace81c | 496 | |
91acb21f JD |
497 | /* Only changed by ubd_init, which is an initcall. */ |
498 | static int io_pid = -1; | |
09ace81c | 499 | |
5dc62b1b | 500 | static void kill_io_thread(void) |
91acb21f | 501 | { |
6c29256c | 502 | if(io_pid != -1) |
91acb21f | 503 | os_kill_process(io_pid, 1); |
09ace81c | 504 | } |
1da177e4 | 505 | |
91acb21f JD |
506 | __uml_exitcall(kill_io_thread); |
507 | ||
d8d7c28e | 508 | static inline int ubd_file_size(struct ubd *ubd_dev, __u64 *size_out) |
1da177e4 LT |
509 | { |
510 | char *file; | |
511 | ||
7d314e34 | 512 | file = ubd_dev->cow.file ? ubd_dev->cow.file : ubd_dev->file; |
dc764e50 | 513 | return os_file_size(file, size_out); |
1da177e4 LT |
514 | } |
515 | ||
5dc62b1b WC |
516 | static int read_cow_bitmap(int fd, void *buf, int offset, int len) |
517 | { | |
518 | int err; | |
519 | ||
520 | err = os_seek_file(fd, offset); | |
521 | if (err < 0) | |
522 | return err; | |
523 | ||
524 | err = os_read_file(fd, buf, len); | |
525 | if (err < 0) | |
526 | return err; | |
527 | ||
528 | return 0; | |
529 | } | |
530 | ||
531 | static int backing_file_mismatch(char *file, __u64 size, time_t mtime) | |
532 | { | |
533 | unsigned long modtime; | |
534 | unsigned long long actual; | |
535 | int err; | |
536 | ||
537 | err = os_file_modtime(file, &modtime); | |
538 | if (err < 0) { | |
539 | printk(KERN_ERR "Failed to get modification time of backing " | |
540 | "file \"%s\", err = %d\n", file, -err); | |
541 | return err; | |
542 | } | |
543 | ||
544 | err = os_file_size(file, &actual); | |
545 | if (err < 0) { | |
546 | printk(KERN_ERR "Failed to get size of backing file \"%s\", " | |
547 | "err = %d\n", file, -err); | |
548 | return err; | |
549 | } | |
550 | ||
551 | if (actual != size) { | |
552 | /*__u64 can be a long on AMD64 and with %lu GCC complains; so | |
553 | * the typecast.*/ | |
554 | printk(KERN_ERR "Size mismatch (%llu vs %llu) of COW header " | |
555 | "vs backing file\n", (unsigned long long) size, actual); | |
556 | return -EINVAL; | |
557 | } | |
558 | if (modtime != mtime) { | |
559 | printk(KERN_ERR "mtime mismatch (%ld vs %ld) of COW header vs " | |
560 | "backing file\n", mtime, modtime); | |
561 | return -EINVAL; | |
562 | } | |
563 | return 0; | |
564 | } | |
565 | ||
566 | static int path_requires_switch(char *from_cmdline, char *from_cow, char *cow) | |
567 | { | |
568 | struct uml_stat buf1, buf2; | |
569 | int err; | |
570 | ||
571 | if (from_cmdline == NULL) | |
572 | return 0; | |
573 | if (!strcmp(from_cmdline, from_cow)) | |
574 | return 0; | |
575 | ||
576 | err = os_stat_file(from_cmdline, &buf1); | |
577 | if (err < 0) { | |
578 | printk(KERN_ERR "Couldn't stat '%s', err = %d\n", from_cmdline, | |
579 | -err); | |
580 | return 0; | |
581 | } | |
582 | err = os_stat_file(from_cow, &buf2); | |
583 | if (err < 0) { | |
584 | printk(KERN_ERR "Couldn't stat '%s', err = %d\n", from_cow, | |
585 | -err); | |
586 | return 1; | |
587 | } | |
588 | if ((buf1.ust_dev == buf2.ust_dev) && (buf1.ust_ino == buf2.ust_ino)) | |
589 | return 0; | |
590 | ||
591 | printk(KERN_ERR "Backing file mismatch - \"%s\" requested, " | |
592 | "\"%s\" specified in COW header of \"%s\"\n", | |
593 | from_cmdline, from_cow, cow); | |
594 | return 1; | |
595 | } | |
596 | ||
597 | static int open_ubd_file(char *file, struct openflags *openflags, int shared, | |
598 | char **backing_file_out, int *bitmap_offset_out, | |
599 | unsigned long *bitmap_len_out, int *data_offset_out, | |
600 | int *create_cow_out) | |
601 | { | |
602 | time_t mtime; | |
603 | unsigned long long size; | |
604 | __u32 version, align; | |
605 | char *backing_file; | |
606 | int fd, err, sectorsize, asked_switch, mode = 0644; | |
607 | ||
608 | fd = os_open_file(file, *openflags, mode); | |
609 | if (fd < 0) { | |
610 | if ((fd == -ENOENT) && (create_cow_out != NULL)) | |
611 | *create_cow_out = 1; | |
612 | if (!openflags->w || | |
613 | ((fd != -EROFS) && (fd != -EACCES))) | |
614 | return fd; | |
615 | openflags->w = 0; | |
616 | fd = os_open_file(file, *openflags, mode); | |
617 | if (fd < 0) | |
618 | return fd; | |
619 | } | |
620 | ||
621 | if (shared) | |
622 | printk(KERN_INFO "Not locking \"%s\" on the host\n", file); | |
623 | else { | |
624 | err = os_lock_file(fd, openflags->w); | |
625 | if (err < 0) { | |
626 | printk(KERN_ERR "Failed to lock '%s', err = %d\n", | |
627 | file, -err); | |
628 | goto out_close; | |
629 | } | |
630 | } | |
631 | ||
632 | /* Successful return case! */ | |
633 | if (backing_file_out == NULL) | |
634 | return fd; | |
635 | ||
636 | err = read_cow_header(file_reader, &fd, &version, &backing_file, &mtime, | |
637 | &size, §orsize, &align, bitmap_offset_out); | |
638 | if (err && (*backing_file_out != NULL)) { | |
639 | printk(KERN_ERR "Failed to read COW header from COW file " | |
640 | "\"%s\", errno = %d\n", file, -err); | |
641 | goto out_close; | |
642 | } | |
643 | if (err) | |
644 | return fd; | |
645 | ||
646 | asked_switch = path_requires_switch(*backing_file_out, backing_file, | |
647 | file); | |
648 | ||
649 | /* Allow switching only if no mismatch. */ | |
650 | if (asked_switch && !backing_file_mismatch(*backing_file_out, size, | |
651 | mtime)) { | |
652 | printk(KERN_ERR "Switching backing file to '%s'\n", | |
653 | *backing_file_out); | |
654 | err = write_cow_header(file, fd, *backing_file_out, | |
655 | sectorsize, align, &size); | |
656 | if (err) { | |
657 | printk(KERN_ERR "Switch failed, errno = %d\n", -err); | |
658 | goto out_close; | |
659 | } | |
660 | } else { | |
661 | *backing_file_out = backing_file; | |
662 | err = backing_file_mismatch(*backing_file_out, size, mtime); | |
663 | if (err) | |
664 | goto out_close; | |
665 | } | |
666 | ||
667 | cow_sizes(version, size, sectorsize, align, *bitmap_offset_out, | |
668 | bitmap_len_out, data_offset_out); | |
669 | ||
670 | return fd; | |
671 | out_close: | |
672 | os_close_file(fd); | |
673 | return err; | |
674 | } | |
675 | ||
676 | static int create_cow_file(char *cow_file, char *backing_file, | |
677 | struct openflags flags, | |
678 | int sectorsize, int alignment, int *bitmap_offset_out, | |
679 | unsigned long *bitmap_len_out, int *data_offset_out) | |
680 | { | |
681 | int err, fd; | |
682 | ||
683 | flags.c = 1; | |
684 | fd = open_ubd_file(cow_file, &flags, 0, NULL, NULL, NULL, NULL, NULL); | |
685 | if (fd < 0) { | |
686 | err = fd; | |
687 | printk(KERN_ERR "Open of COW file '%s' failed, errno = %d\n", | |
688 | cow_file, -err); | |
689 | goto out; | |
690 | } | |
691 | ||
692 | err = init_cow_file(fd, cow_file, backing_file, sectorsize, alignment, | |
693 | bitmap_offset_out, bitmap_len_out, | |
694 | data_offset_out); | |
695 | if (!err) | |
696 | return fd; | |
697 | os_close_file(fd); | |
698 | out: | |
699 | return err; | |
700 | } | |
701 | ||
5f75a4f8 | 702 | static void ubd_close_dev(struct ubd *ubd_dev) |
1da177e4 | 703 | { |
7d314e34 PBG |
704 | os_close_file(ubd_dev->fd); |
705 | if(ubd_dev->cow.file == NULL) | |
1da177e4 LT |
706 | return; |
707 | ||
7d314e34 PBG |
708 | os_close_file(ubd_dev->cow.fd); |
709 | vfree(ubd_dev->cow.bitmap); | |
710 | ubd_dev->cow.bitmap = NULL; | |
1da177e4 LT |
711 | } |
712 | ||
7d314e34 | 713 | static int ubd_open_dev(struct ubd *ubd_dev) |
1da177e4 LT |
714 | { |
715 | struct openflags flags; | |
716 | char **back_ptr; | |
717 | int err, create_cow, *create_ptr; | |
0bf16bff | 718 | int fd; |
1da177e4 | 719 | |
7d314e34 | 720 | ubd_dev->openflags = ubd_dev->boot_openflags; |
1da177e4 | 721 | create_cow = 0; |
7d314e34 PBG |
722 | create_ptr = (ubd_dev->cow.file != NULL) ? &create_cow : NULL; |
723 | back_ptr = ubd_dev->no_cow ? NULL : &ubd_dev->cow.file; | |
0bf16bff PBG |
724 | |
725 | fd = open_ubd_file(ubd_dev->file, &ubd_dev->openflags, ubd_dev->shared, | |
7d314e34 PBG |
726 | back_ptr, &ubd_dev->cow.bitmap_offset, |
727 | &ubd_dev->cow.bitmap_len, &ubd_dev->cow.data_offset, | |
6c29256c | 728 | create_ptr); |
1da177e4 | 729 | |
0bf16bff PBG |
730 | if((fd == -ENOENT) && create_cow){ |
731 | fd = create_cow_file(ubd_dev->file, ubd_dev->cow.file, | |
7d314e34 PBG |
732 | ubd_dev->openflags, 1 << 9, PAGE_SIZE, |
733 | &ubd_dev->cow.bitmap_offset, | |
734 | &ubd_dev->cow.bitmap_len, | |
735 | &ubd_dev->cow.data_offset); | |
0bf16bff | 736 | if(fd >= 0){ |
1da177e4 | 737 | printk(KERN_INFO "Creating \"%s\" as COW file for " |
7d314e34 | 738 | "\"%s\"\n", ubd_dev->file, ubd_dev->cow.file); |
1da177e4 LT |
739 | } |
740 | } | |
741 | ||
0bf16bff | 742 | if(fd < 0){ |
7d314e34 | 743 | printk("Failed to open '%s', errno = %d\n", ubd_dev->file, |
0bf16bff PBG |
744 | -fd); |
745 | return fd; | |
1da177e4 | 746 | } |
0bf16bff | 747 | ubd_dev->fd = fd; |
1da177e4 | 748 | |
7d314e34 | 749 | if(ubd_dev->cow.file != NULL){ |
f4768ffd JD |
750 | blk_queue_max_sectors(ubd_dev->queue, 8 * sizeof(long)); |
751 | ||
1da177e4 | 752 | err = -ENOMEM; |
da2486ba | 753 | ubd_dev->cow.bitmap = vmalloc(ubd_dev->cow.bitmap_len); |
7d314e34 | 754 | if(ubd_dev->cow.bitmap == NULL){ |
1da177e4 LT |
755 | printk(KERN_ERR "Failed to vmalloc COW bitmap\n"); |
756 | goto error; | |
757 | } | |
758 | flush_tlb_kernel_vm(); | |
759 | ||
7d314e34 PBG |
760 | err = read_cow_bitmap(ubd_dev->fd, ubd_dev->cow.bitmap, |
761 | ubd_dev->cow.bitmap_offset, | |
762 | ubd_dev->cow.bitmap_len); | |
1da177e4 LT |
763 | if(err < 0) |
764 | goto error; | |
765 | ||
7d314e34 | 766 | flags = ubd_dev->openflags; |
1da177e4 | 767 | flags.w = 0; |
7d314e34 | 768 | err = open_ubd_file(ubd_dev->cow.file, &flags, ubd_dev->shared, NULL, |
6c29256c | 769 | NULL, NULL, NULL, NULL); |
1da177e4 | 770 | if(err < 0) goto error; |
7d314e34 | 771 | ubd_dev->cow.fd = err; |
1da177e4 | 772 | } |
dc764e50 | 773 | return 0; |
1da177e4 | 774 | error: |
7d314e34 | 775 | os_close_file(ubd_dev->fd); |
dc764e50 | 776 | return err; |
1da177e4 LT |
777 | } |
778 | ||
2e3f5251 JD |
779 | static void ubd_device_release(struct device *dev) |
780 | { | |
8691b97b | 781 | struct ubd *ubd_dev = dev_get_drvdata(dev); |
2e3f5251 JD |
782 | |
783 | blk_cleanup_queue(ubd_dev->queue); | |
784 | *ubd_dev = ((struct ubd) DEFAULT_UBD); | |
785 | } | |
786 | ||
5f75a4f8 | 787 | static int ubd_disk_register(int major, u64 size, int unit, |
b8831a1d | 788 | struct gendisk **disk_out) |
1da177e4 LT |
789 | { |
790 | struct gendisk *disk; | |
1da177e4 LT |
791 | |
792 | disk = alloc_disk(1 << UBD_SHIFT); | |
793 | if(disk == NULL) | |
dc764e50 | 794 | return -ENOMEM; |
1da177e4 LT |
795 | |
796 | disk->major = major; | |
797 | disk->first_minor = unit << UBD_SHIFT; | |
798 | disk->fops = &ubd_blops; | |
799 | set_capacity(disk, size / 512); | |
792dd4fc | 800 | if (major == UBD_MAJOR) |
1da177e4 | 801 | sprintf(disk->disk_name, "ubd%c", 'a' + unit); |
ce7b0f46 | 802 | else |
1da177e4 | 803 | sprintf(disk->disk_name, "ubd_fake%d", unit); |
1da177e4 LT |
804 | |
805 | /* sysfs register (not for ide fake devices) */ | |
792dd4fc | 806 | if (major == UBD_MAJOR) { |
7d314e34 PBG |
807 | ubd_devs[unit].pdev.id = unit; |
808 | ubd_devs[unit].pdev.name = DRIVER_NAME; | |
2e3f5251 | 809 | ubd_devs[unit].pdev.dev.release = ubd_device_release; |
8691b97b | 810 | dev_set_drvdata(&ubd_devs[unit].pdev.dev, &ubd_devs[unit]); |
7d314e34 PBG |
811 | platform_device_register(&ubd_devs[unit].pdev); |
812 | disk->driverfs_dev = &ubd_devs[unit].pdev.dev; | |
1da177e4 LT |
813 | } |
814 | ||
7d314e34 | 815 | disk->private_data = &ubd_devs[unit]; |
62f96cb0 | 816 | disk->queue = ubd_devs[unit].queue; |
1da177e4 LT |
817 | add_disk(disk); |
818 | ||
819 | *disk_out = disk; | |
820 | return 0; | |
821 | } | |
822 | ||
823 | #define ROUND_BLOCK(n) ((n + ((1 << 9) - 1)) & (-1 << 9)) | |
824 | ||
f28169d2 | 825 | static int ubd_add(int n, char **error_out) |
1da177e4 | 826 | { |
7d314e34 | 827 | struct ubd *ubd_dev = &ubd_devs[n]; |
f28169d2 | 828 | int err = 0; |
1da177e4 | 829 | |
7d314e34 | 830 | if(ubd_dev->file == NULL) |
ec7cf783 | 831 | goto out; |
1da177e4 | 832 | |
7d314e34 | 833 | err = ubd_file_size(ubd_dev, &ubd_dev->size); |
f28169d2 JD |
834 | if(err < 0){ |
835 | *error_out = "Couldn't determine size of device's file"; | |
80c13749 | 836 | goto out; |
f28169d2 | 837 | } |
1da177e4 | 838 | |
7d314e34 | 839 | ubd_dev->size = ROUND_BLOCK(ubd_dev->size); |
1da177e4 | 840 | |
a0044bdf | 841 | INIT_LIST_HEAD(&ubd_dev->restart); |
4f40c055 | 842 | sg_init_table(ubd_dev->sg, MAX_SG); |
a0044bdf | 843 | |
62f96cb0 JD |
844 | err = -ENOMEM; |
845 | ubd_dev->queue = blk_init_queue(do_ubd_request, &ubd_dev->lock); | |
846 | if (ubd_dev->queue == NULL) { | |
847 | *error_out = "Failed to initialize device queue"; | |
80c13749 | 848 | goto out; |
62f96cb0 JD |
849 | } |
850 | ubd_dev->queue->queuedata = ubd_dev; | |
851 | ||
a0044bdf | 852 | blk_queue_max_hw_segments(ubd_dev->queue, MAX_SG); |
792dd4fc | 853 | err = ubd_disk_register(UBD_MAJOR, ubd_dev->size, n, &ubd_gendisk[n]); |
62f96cb0 JD |
854 | if(err){ |
855 | *error_out = "Failed to register device"; | |
856 | goto out_cleanup; | |
857 | } | |
6c29256c | 858 | |
792dd4fc | 859 | if (fake_major != UBD_MAJOR) |
5f75a4f8 | 860 | ubd_disk_register(fake_major, ubd_dev->size, n, |
62f96cb0 | 861 | &fake_gendisk[n]); |
1da177e4 | 862 | |
83380cc1 JD |
863 | /* |
864 | * Perhaps this should also be under the "if (fake_major)" above | |
865 | * using the fake_disk->disk_name | |
866 | */ | |
1da177e4 LT |
867 | if (fake_ide) |
868 | make_ide_entries(ubd_gendisk[n]->disk_name); | |
869 | ||
ec7cf783 | 870 | err = 0; |
ec7cf783 JD |
871 | out: |
872 | return err; | |
62f96cb0 JD |
873 | |
874 | out_cleanup: | |
875 | blk_cleanup_queue(ubd_dev->queue); | |
876 | goto out; | |
1da177e4 LT |
877 | } |
878 | ||
f28169d2 | 879 | static int ubd_config(char *str, char **error_out) |
1da177e4 | 880 | { |
e7f6552f | 881 | int n, ret; |
1da177e4 | 882 | |
f28169d2 JD |
883 | /* This string is possibly broken up and stored, so it's only |
884 | * freed if ubd_setup_common fails, or if only general options | |
885 | * were set. | |
886 | */ | |
970d6e3a | 887 | str = kstrdup(str, GFP_KERNEL); |
e7f6552f | 888 | if (str == NULL) { |
f28169d2 JD |
889 | *error_out = "Failed to allocate memory"; |
890 | return -ENOMEM; | |
1da177e4 | 891 | } |
f28169d2 JD |
892 | |
893 | ret = ubd_setup_common(str, &n, error_out); | |
894 | if (ret) | |
e7f6552f | 895 | goto err_free; |
f28169d2 | 896 | |
e7f6552f PBG |
897 | if (n == -1) { |
898 | ret = 0; | |
d8d7c28e | 899 | goto err_free; |
1da177e4 | 900 | } |
1da177e4 | 901 | |
dc764e50 | 902 | mutex_lock(&ubd_lock); |
f28169d2 | 903 | ret = ubd_add(n, error_out); |
e7f6552f | 904 | if (ret) |
7d314e34 | 905 | ubd_devs[n].file = NULL; |
dc764e50 | 906 | mutex_unlock(&ubd_lock); |
1da177e4 | 907 | |
e7f6552f | 908 | out: |
dc764e50 | 909 | return ret; |
e7f6552f PBG |
910 | |
911 | err_free: | |
912 | kfree(str); | |
913 | goto out; | |
1da177e4 LT |
914 | } |
915 | ||
916 | static int ubd_get_config(char *name, char *str, int size, char **error_out) | |
917 | { | |
7d314e34 | 918 | struct ubd *ubd_dev; |
1da177e4 LT |
919 | int n, len = 0; |
920 | ||
921 | n = parse_unit(&name); | |
922 | if((n >= MAX_DEV) || (n < 0)){ | |
923 | *error_out = "ubd_get_config : device number out of range"; | |
dc764e50 | 924 | return -1; |
1da177e4 LT |
925 | } |
926 | ||
7d314e34 | 927 | ubd_dev = &ubd_devs[n]; |
d7fb2c38 | 928 | mutex_lock(&ubd_lock); |
1da177e4 | 929 | |
7d314e34 | 930 | if(ubd_dev->file == NULL){ |
1da177e4 LT |
931 | CONFIG_CHUNK(str, size, len, "", 1); |
932 | goto out; | |
933 | } | |
934 | ||
7d314e34 | 935 | CONFIG_CHUNK(str, size, len, ubd_dev->file, 0); |
1da177e4 | 936 | |
7d314e34 | 937 | if(ubd_dev->cow.file != NULL){ |
1da177e4 | 938 | CONFIG_CHUNK(str, size, len, ",", 0); |
7d314e34 | 939 | CONFIG_CHUNK(str, size, len, ubd_dev->cow.file, 1); |
1da177e4 LT |
940 | } |
941 | else CONFIG_CHUNK(str, size, len, "", 1); | |
942 | ||
943 | out: | |
d7fb2c38 | 944 | mutex_unlock(&ubd_lock); |
dc764e50 | 945 | return len; |
1da177e4 LT |
946 | } |
947 | ||
29d56cfe JD |
948 | static int ubd_id(char **str, int *start_out, int *end_out) |
949 | { | |
dc764e50 | 950 | int n; |
29d56cfe JD |
951 | |
952 | n = parse_unit(str); | |
dc764e50 JD |
953 | *start_out = 0; |
954 | *end_out = MAX_DEV - 1; | |
955 | return n; | |
29d56cfe JD |
956 | } |
957 | ||
f28169d2 | 958 | static int ubd_remove(int n, char **error_out) |
1da177e4 | 959 | { |
2e3f5251 | 960 | struct gendisk *disk = ubd_gendisk[n]; |
7d314e34 | 961 | struct ubd *ubd_dev; |
29d56cfe | 962 | int err = -ENODEV; |
1da177e4 | 963 | |
d7fb2c38 | 964 | mutex_lock(&ubd_lock); |
1da177e4 | 965 | |
7d314e34 | 966 | ubd_dev = &ubd_devs[n]; |
1da177e4 | 967 | |
7d314e34 | 968 | if(ubd_dev->file == NULL) |
29d56cfe | 969 | goto out; |
1da177e4 | 970 | |
29d56cfe JD |
971 | /* you cannot remove a open disk */ |
972 | err = -EBUSY; | |
7d314e34 | 973 | if(ubd_dev->count > 0) |
1da177e4 LT |
974 | goto out; |
975 | ||
dc764e50 | 976 | ubd_gendisk[n] = NULL; |
b47d2deb JD |
977 | if(disk != NULL){ |
978 | del_gendisk(disk); | |
979 | put_disk(disk); | |
980 | } | |
1da177e4 LT |
981 | |
982 | if(fake_gendisk[n] != NULL){ | |
983 | del_gendisk(fake_gendisk[n]); | |
984 | put_disk(fake_gendisk[n]); | |
985 | fake_gendisk[n] = NULL; | |
986 | } | |
987 | ||
1da177e4 | 988 | err = 0; |
2e3f5251 | 989 | platform_device_unregister(&ubd_dev->pdev); |
29d56cfe | 990 | out: |
d7fb2c38 | 991 | mutex_unlock(&ubd_lock); |
29d56cfe | 992 | return err; |
1da177e4 LT |
993 | } |
994 | ||
f28169d2 | 995 | /* All these are called by mconsole in process context and without |
b8831a1d | 996 | * ubd-specific locks. The structure itself is const except for .list. |
f28169d2 | 997 | */ |
1da177e4 | 998 | static struct mc_device ubd_mc = { |
84f48d4f | 999 | .list = LIST_HEAD_INIT(ubd_mc.list), |
1da177e4 LT |
1000 | .name = "ubd", |
1001 | .config = ubd_config, | |
dc764e50 | 1002 | .get_config = ubd_get_config, |
29d56cfe | 1003 | .id = ubd_id, |
1da177e4 LT |
1004 | .remove = ubd_remove, |
1005 | }; | |
1006 | ||
d8d7c28e | 1007 | static int __init ubd_mc_init(void) |
1da177e4 LT |
1008 | { |
1009 | mconsole_register_dev(&ubd_mc); | |
1010 | return 0; | |
1011 | } | |
1012 | ||
1013 | __initcall(ubd_mc_init); | |
1014 | ||
d8d7c28e PBG |
1015 | static int __init ubd0_init(void) |
1016 | { | |
1017 | struct ubd *ubd_dev = &ubd_devs[0]; | |
1018 | ||
b8831a1d | 1019 | mutex_lock(&ubd_lock); |
d8d7c28e PBG |
1020 | if(ubd_dev->file == NULL) |
1021 | ubd_dev->file = "root_fs"; | |
b8831a1d JD |
1022 | mutex_unlock(&ubd_lock); |
1023 | ||
dc764e50 | 1024 | return 0; |
d8d7c28e PBG |
1025 | } |
1026 | ||
1027 | __initcall(ubd0_init); | |
1028 | ||
b8831a1d | 1029 | /* Used in ubd_init, which is an initcall */ |
3ae5eaec RK |
1030 | static struct platform_driver ubd_driver = { |
1031 | .driver = { | |
1032 | .name = DRIVER_NAME, | |
1033 | }, | |
1da177e4 LT |
1034 | }; |
1035 | ||
d8d7c28e | 1036 | static int __init ubd_init(void) |
1da177e4 | 1037 | { |
f28169d2 JD |
1038 | char *error; |
1039 | int i, err; | |
1da177e4 | 1040 | |
792dd4fc | 1041 | if (register_blkdev(UBD_MAJOR, "ubd")) |
1da177e4 LT |
1042 | return -1; |
1043 | ||
792dd4fc | 1044 | if (fake_major != UBD_MAJOR) { |
1da177e4 LT |
1045 | char name[sizeof("ubd_nnn\0")]; |
1046 | ||
1047 | snprintf(name, sizeof(name), "ubd_%d", fake_major); | |
1da177e4 LT |
1048 | if (register_blkdev(fake_major, "ubd")) |
1049 | return -1; | |
1050 | } | |
3ae5eaec | 1051 | platform_driver_register(&ubd_driver); |
dc764e50 | 1052 | mutex_lock(&ubd_lock); |
f28169d2 JD |
1053 | for (i = 0; i < MAX_DEV; i++){ |
1054 | err = ubd_add(i, &error); | |
1055 | if(err) | |
1056 | printk(KERN_ERR "Failed to initialize ubd device %d :" | |
1057 | "%s\n", i, error); | |
1058 | } | |
dc764e50 | 1059 | mutex_unlock(&ubd_lock); |
1da177e4 LT |
1060 | return 0; |
1061 | } | |
1062 | ||
1063 | late_initcall(ubd_init); | |
1064 | ||
d8d7c28e | 1065 | static int __init ubd_driver_init(void){ |
91acb21f JD |
1066 | unsigned long stack; |
1067 | int err; | |
1068 | ||
1069 | /* Set by CONFIG_BLK_DEV_UBD_SYNC or ubd=sync.*/ | |
1070 | if(global_openflags.s){ | |
1071 | printk(KERN_INFO "ubd: Synchronous mode\n"); | |
1072 | /* Letting ubd=sync be like using ubd#s= instead of ubd#= is | |
1073 | * enough. So use anyway the io thread. */ | |
1074 | } | |
1075 | stack = alloc_stack(0, 0); | |
6c29256c | 1076 | io_pid = start_io_thread(stack + PAGE_SIZE - sizeof(void *), |
91acb21f JD |
1077 | &thread_fd); |
1078 | if(io_pid < 0){ | |
6c29256c | 1079 | printk(KERN_ERR |
91acb21f JD |
1080 | "ubd : Failed to start I/O thread (errno = %d) - " |
1081 | "falling back to synchronous I/O\n", -io_pid); | |
1082 | io_pid = -1; | |
dc764e50 | 1083 | return 0; |
91acb21f | 1084 | } |
6c29256c | 1085 | err = um_request_irq(UBD_IRQ, thread_fd, IRQ_READ, ubd_intr, |
7d314e34 | 1086 | IRQF_DISABLED, "ubd", ubd_devs); |
91acb21f JD |
1087 | if(err != 0) |
1088 | printk(KERN_ERR "um_request_irq failed - errno = %d\n", -err); | |
f4c57a78 | 1089 | return 0; |
91acb21f JD |
1090 | } |
1091 | ||
1092 | device_initcall(ubd_driver_init); | |
1093 | ||
a625c998 | 1094 | static int ubd_open(struct block_device *bdev, fmode_t mode) |
1da177e4 | 1095 | { |
a625c998 | 1096 | struct gendisk *disk = bdev->bd_disk; |
7d314e34 | 1097 | struct ubd *ubd_dev = disk->private_data; |
1da177e4 LT |
1098 | int err = 0; |
1099 | ||
7d314e34 PBG |
1100 | if(ubd_dev->count == 0){ |
1101 | err = ubd_open_dev(ubd_dev); | |
1da177e4 LT |
1102 | if(err){ |
1103 | printk(KERN_ERR "%s: Can't open \"%s\": errno = %d\n", | |
7d314e34 | 1104 | disk->disk_name, ubd_dev->file, -err); |
1da177e4 LT |
1105 | goto out; |
1106 | } | |
1107 | } | |
7d314e34 PBG |
1108 | ubd_dev->count++; |
1109 | set_disk_ro(disk, !ubd_dev->openflags.w); | |
2c49be99 PBG |
1110 | |
1111 | /* This should no more be needed. And it didn't work anyway to exclude | |
1112 | * read-write remounting of filesystems.*/ | |
a625c998 | 1113 | /*if((mode & FMODE_WRITE) && !ubd_dev->openflags.w){ |
5f75a4f8 | 1114 | if(--ubd_dev->count == 0) ubd_close_dev(ubd_dev); |
1da177e4 | 1115 | err = -EROFS; |
2c49be99 | 1116 | }*/ |
1da177e4 | 1117 | out: |
dc764e50 | 1118 | return err; |
1da177e4 LT |
1119 | } |
1120 | ||
a625c998 | 1121 | static int ubd_release(struct gendisk *disk, fmode_t mode) |
1da177e4 | 1122 | { |
7d314e34 | 1123 | struct ubd *ubd_dev = disk->private_data; |
1da177e4 | 1124 | |
7d314e34 | 1125 | if(--ubd_dev->count == 0) |
5f75a4f8 | 1126 | ubd_close_dev(ubd_dev); |
dc764e50 | 1127 | return 0; |
1da177e4 LT |
1128 | } |
1129 | ||
91acb21f JD |
1130 | static void cowify_bitmap(__u64 io_offset, int length, unsigned long *cow_mask, |
1131 | __u64 *cow_offset, unsigned long *bitmap, | |
1132 | __u64 bitmap_offset, unsigned long *bitmap_words, | |
1133 | __u64 bitmap_len) | |
1da177e4 | 1134 | { |
91acb21f JD |
1135 | __u64 sector = io_offset >> 9; |
1136 | int i, update_bitmap = 0; | |
1137 | ||
1138 | for(i = 0; i < length >> 9; i++){ | |
1139 | if(cow_mask != NULL) | |
1140 | ubd_set_bit(i, (unsigned char *) cow_mask); | |
1141 | if(ubd_test_bit(sector + i, (unsigned char *) bitmap)) | |
1142 | continue; | |
1da177e4 | 1143 | |
91acb21f JD |
1144 | update_bitmap = 1; |
1145 | ubd_set_bit(sector + i, (unsigned char *) bitmap); | |
1146 | } | |
1147 | ||
1148 | if(!update_bitmap) | |
1149 | return; | |
1da177e4 | 1150 | |
91acb21f | 1151 | *cow_offset = sector / (sizeof(unsigned long) * 8); |
1da177e4 | 1152 | |
91acb21f JD |
1153 | /* This takes care of the case where we're exactly at the end of the |
1154 | * device, and *cow_offset + 1 is off the end. So, just back it up | |
1155 | * by one word. Thanks to Lynn Kerby for the fix and James McMechan | |
1156 | * for the original diagnosis. | |
1157 | */ | |
6d074242 JO |
1158 | if (*cow_offset == (DIV_ROUND_UP(bitmap_len, |
1159 | sizeof(unsigned long)) - 1)) | |
91acb21f JD |
1160 | (*cow_offset)--; |
1161 | ||
1162 | bitmap_words[0] = bitmap[*cow_offset]; | |
1163 | bitmap_words[1] = bitmap[*cow_offset + 1]; | |
1164 | ||
1165 | *cow_offset *= sizeof(unsigned long); | |
1166 | *cow_offset += bitmap_offset; | |
1167 | } | |
1168 | ||
1169 | static void cowify_req(struct io_thread_req *req, unsigned long *bitmap, | |
1170 | __u64 bitmap_offset, __u64 bitmap_len) | |
1171 | { | |
1172 | __u64 sector = req->offset >> 9; | |
1173 | int i; | |
1174 | ||
1175 | if(req->length > (sizeof(req->sector_mask) * 8) << 9) | |
1176 | panic("Operation too long"); | |
1177 | ||
1178 | if(req->op == UBD_READ) { | |
1179 | for(i = 0; i < req->length >> 9; i++){ | |
1180 | if(ubd_test_bit(sector + i, (unsigned char *) bitmap)) | |
6c29256c | 1181 | ubd_set_bit(i, (unsigned char *) |
91acb21f | 1182 | &req->sector_mask); |
dc764e50 | 1183 | } |
91acb21f JD |
1184 | } |
1185 | else cowify_bitmap(req->offset, req->length, &req->sector_mask, | |
1186 | &req->cow_offset, bitmap, bitmap_offset, | |
1187 | req->bitmap_words, bitmap_len); | |
1da177e4 LT |
1188 | } |
1189 | ||
62f96cb0 | 1190 | /* Called with dev->lock held */ |
a0044bdf JD |
1191 | static void prepare_request(struct request *req, struct io_thread_req *io_req, |
1192 | unsigned long long offset, int page_offset, | |
1193 | int len, struct page *page) | |
1da177e4 LT |
1194 | { |
1195 | struct gendisk *disk = req->rq_disk; | |
7d314e34 | 1196 | struct ubd *ubd_dev = disk->private_data; |
91acb21f | 1197 | |
62f96cb0 | 1198 | io_req->req = req; |
a0044bdf JD |
1199 | io_req->fds[0] = (ubd_dev->cow.file != NULL) ? ubd_dev->cow.fd : |
1200 | ubd_dev->fd; | |
7d314e34 | 1201 | io_req->fds[1] = ubd_dev->fd; |
91acb21f | 1202 | io_req->cow_offset = -1; |
1da177e4 LT |
1203 | io_req->offset = offset; |
1204 | io_req->length = len; | |
1205 | io_req->error = 0; | |
91acb21f JD |
1206 | io_req->sector_mask = 0; |
1207 | ||
1208 | io_req->op = (rq_data_dir(req) == READ) ? UBD_READ : UBD_WRITE; | |
1da177e4 | 1209 | io_req->offsets[0] = 0; |
7d314e34 | 1210 | io_req->offsets[1] = ubd_dev->cow.data_offset; |
a0044bdf | 1211 | io_req->buffer = page_address(page) + page_offset; |
1da177e4 LT |
1212 | io_req->sectorsize = 1 << 9; |
1213 | ||
7d314e34 | 1214 | if(ubd_dev->cow.file != NULL) |
a0044bdf JD |
1215 | cowify_req(io_req, ubd_dev->cow.bitmap, |
1216 | ubd_dev->cow.bitmap_offset, ubd_dev->cow.bitmap_len); | |
91acb21f | 1217 | |
1da177e4 LT |
1218 | } |
1219 | ||
62f96cb0 | 1220 | /* Called with dev->lock held */ |
165125e1 | 1221 | static void do_ubd_request(struct request_queue *q) |
1da177e4 | 1222 | { |
2adcec21 | 1223 | struct io_thread_req *io_req; |
1da177e4 | 1224 | struct request *req; |
f81f2f7c TH |
1225 | sector_t sector; |
1226 | int n; | |
a0044bdf JD |
1227 | |
1228 | while(1){ | |
2a9529a0 | 1229 | struct ubd *dev = q->queuedata; |
a0044bdf | 1230 | if(dev->end_sg == 0){ |
9934c8c0 | 1231 | struct request *req = blk_fetch_request(q); |
a0044bdf JD |
1232 | if(req == NULL) |
1233 | return; | |
1234 | ||
1235 | dev->request = req; | |
a0044bdf JD |
1236 | dev->start_sg = 0; |
1237 | dev->end_sg = blk_rq_map_sg(q, req, dev->sg); | |
1238 | } | |
1239 | ||
1240 | req = dev->request; | |
83096ebf | 1241 | sector = blk_rq_pos(req); |
a0044bdf JD |
1242 | while(dev->start_sg < dev->end_sg){ |
1243 | struct scatterlist *sg = &dev->sg[dev->start_sg]; | |
1244 | ||
2adcec21 | 1245 | io_req = kmalloc(sizeof(struct io_thread_req), |
990c5587 | 1246 | GFP_ATOMIC); |
2adcec21 JD |
1247 | if(io_req == NULL){ |
1248 | if(list_empty(&dev->restart)) | |
1249 | list_add(&dev->restart, &restart); | |
1250 | return; | |
1251 | } | |
1252 | prepare_request(req, io_req, | |
f81f2f7c | 1253 | (unsigned long long)sector << 9, |
45711f1a | 1254 | sg->offset, sg->length, sg_page(sg)); |
a0044bdf | 1255 | |
f81f2f7c | 1256 | sector += sg->length >> 9; |
a6ea4cce JD |
1257 | n = os_write_file(thread_fd, &io_req, |
1258 | sizeof(struct io_thread_req *)); | |
2adcec21 | 1259 | if(n != sizeof(struct io_thread_req *)){ |
a0044bdf JD |
1260 | if(n != -EAGAIN) |
1261 | printk("write to io thread failed, " | |
1262 | "errno = %d\n", -n); | |
1263 | else if(list_empty(&dev->restart)) | |
1264 | list_add(&dev->restart, &restart); | |
12429bf9 | 1265 | kfree(io_req); |
a0044bdf JD |
1266 | return; |
1267 | } | |
1268 | ||
a0044bdf | 1269 | dev->start_sg++; |
1da177e4 | 1270 | } |
a0044bdf JD |
1271 | dev->end_sg = 0; |
1272 | dev->request = NULL; | |
1da177e4 LT |
1273 | } |
1274 | } | |
1275 | ||
a885c8c4 CH |
1276 | static int ubd_getgeo(struct block_device *bdev, struct hd_geometry *geo) |
1277 | { | |
7d314e34 | 1278 | struct ubd *ubd_dev = bdev->bd_disk->private_data; |
a885c8c4 CH |
1279 | |
1280 | geo->heads = 128; | |
1281 | geo->sectors = 32; | |
7d314e34 | 1282 | geo->cylinders = ubd_dev->size / (128 * 32 * 512); |
a885c8c4 CH |
1283 | return 0; |
1284 | } | |
1285 | ||
a625c998 | 1286 | static int ubd_ioctl(struct block_device *bdev, fmode_t mode, |
1da177e4 LT |
1287 | unsigned int cmd, unsigned long arg) |
1288 | { | |
a625c998 | 1289 | struct ubd *ubd_dev = bdev->bd_disk->private_data; |
73855e13 | 1290 | u16 ubd_id[ATA_ID_WORDS]; |
1da177e4 LT |
1291 | |
1292 | switch (cmd) { | |
1da177e4 | 1293 | struct cdrom_volctrl volume; |
1da177e4 | 1294 | case HDIO_GET_IDENTITY: |
73855e13 BZ |
1295 | memset(&ubd_id, 0, ATA_ID_WORDS * 2); |
1296 | ubd_id[ATA_ID_CYLS] = ubd_dev->size / (128 * 32 * 512); | |
1297 | ubd_id[ATA_ID_HEADS] = 128; | |
1298 | ubd_id[ATA_ID_SECTORS] = 32; | |
1da177e4 LT |
1299 | if(copy_to_user((char __user *) arg, (char *) &ubd_id, |
1300 | sizeof(ubd_id))) | |
dc764e50 JD |
1301 | return -EFAULT; |
1302 | return 0; | |
b8831a1d | 1303 | |
1da177e4 LT |
1304 | case CDROMVOLREAD: |
1305 | if(copy_from_user(&volume, (char __user *) arg, sizeof(volume))) | |
dc764e50 | 1306 | return -EFAULT; |
1da177e4 LT |
1307 | volume.channel0 = 255; |
1308 | volume.channel1 = 255; | |
1309 | volume.channel2 = 255; | |
1310 | volume.channel3 = 255; | |
1311 | if(copy_to_user((char __user *) arg, &volume, sizeof(volume))) | |
dc764e50 JD |
1312 | return -EFAULT; |
1313 | return 0; | |
1da177e4 | 1314 | } |
dc764e50 | 1315 | return -EINVAL; |
1da177e4 LT |
1316 | } |
1317 | ||
91acb21f | 1318 | static int update_bitmap(struct io_thread_req *req) |
1da177e4 | 1319 | { |
91acb21f | 1320 | int n; |
1da177e4 | 1321 | |
91acb21f | 1322 | if(req->cow_offset == -1) |
dc764e50 | 1323 | return 0; |
1da177e4 | 1324 | |
91acb21f JD |
1325 | n = os_seek_file(req->fds[1], req->cow_offset); |
1326 | if(n < 0){ | |
1327 | printk("do_io - bitmap lseek failed : err = %d\n", -n); | |
dc764e50 | 1328 | return 1; |
91acb21f | 1329 | } |
1da177e4 | 1330 | |
a6ea4cce JD |
1331 | n = os_write_file(req->fds[1], &req->bitmap_words, |
1332 | sizeof(req->bitmap_words)); | |
91acb21f JD |
1333 | if(n != sizeof(req->bitmap_words)){ |
1334 | printk("do_io - bitmap update failed, err = %d fd = %d\n", -n, | |
1335 | req->fds[1]); | |
dc764e50 | 1336 | return 1; |
91acb21f | 1337 | } |
1da177e4 | 1338 | |
dc764e50 | 1339 | return 0; |
91acb21f | 1340 | } |
1da177e4 | 1341 | |
5dc62b1b | 1342 | static void do_io(struct io_thread_req *req) |
91acb21f JD |
1343 | { |
1344 | char *buf; | |
1345 | unsigned long len; | |
1346 | int n, nsectors, start, end, bit; | |
1347 | int err; | |
1348 | __u64 off; | |
1349 | ||
1350 | nsectors = req->length / req->sectorsize; | |
1351 | start = 0; | |
1352 | do { | |
1353 | bit = ubd_test_bit(start, (unsigned char *) &req->sector_mask); | |
1354 | end = start; | |
1355 | while((end < nsectors) && | |
1356 | (ubd_test_bit(end, (unsigned char *) | |
1357 | &req->sector_mask) == bit)) | |
1358 | end++; | |
1359 | ||
1360 | off = req->offset + req->offsets[bit] + | |
1361 | start * req->sectorsize; | |
1362 | len = (end - start) * req->sectorsize; | |
1363 | buf = &req->buffer[start * req->sectorsize]; | |
1364 | ||
1365 | err = os_seek_file(req->fds[bit], off); | |
1366 | if(err < 0){ | |
1367 | printk("do_io - lseek failed : err = %d\n", -err); | |
1368 | req->error = 1; | |
1369 | return; | |
1370 | } | |
1371 | if(req->op == UBD_READ){ | |
1372 | n = 0; | |
1373 | do { | |
1374 | buf = &buf[n]; | |
1375 | len -= n; | |
a6ea4cce | 1376 | n = os_read_file(req->fds[bit], buf, len); |
91acb21f JD |
1377 | if (n < 0) { |
1378 | printk("do_io - read failed, err = %d " | |
1379 | "fd = %d\n", -n, req->fds[bit]); | |
1380 | req->error = 1; | |
1381 | return; | |
1382 | } | |
1383 | } while((n < len) && (n != 0)); | |
1384 | if (n < len) memset(&buf[n], 0, len - n); | |
1385 | } else { | |
a6ea4cce | 1386 | n = os_write_file(req->fds[bit], buf, len); |
91acb21f JD |
1387 | if(n != len){ |
1388 | printk("do_io - write failed err = %d " | |
1389 | "fd = %d\n", -n, req->fds[bit]); | |
1390 | req->error = 1; | |
1391 | return; | |
1392 | } | |
1393 | } | |
1394 | ||
1395 | start = end; | |
1396 | } while(start < nsectors); | |
1da177e4 | 1397 | |
91acb21f | 1398 | req->error = update_bitmap(req); |
1da177e4 | 1399 | } |
91acb21f JD |
1400 | |
1401 | /* Changed in start_io_thread, which is serialized by being called only | |
1402 | * from ubd_init, which is an initcall. | |
1403 | */ | |
1404 | int kernel_fd = -1; | |
1405 | ||
d8d7c28e PBG |
1406 | /* Only changed by the io thread. XXX: currently unused. */ |
1407 | static int io_count = 0; | |
91acb21f JD |
1408 | |
1409 | int io_thread(void *arg) | |
1410 | { | |
2adcec21 | 1411 | struct io_thread_req *req; |
91acb21f JD |
1412 | int n; |
1413 | ||
1414 | ignore_sigwinch_sig(); | |
1415 | while(1){ | |
a6ea4cce | 1416 | n = os_read_file(kernel_fd, &req, |
2adcec21 JD |
1417 | sizeof(struct io_thread_req *)); |
1418 | if(n != sizeof(struct io_thread_req *)){ | |
91acb21f JD |
1419 | if(n < 0) |
1420 | printk("io_thread - read failed, fd = %d, " | |
1421 | "err = %d\n", kernel_fd, -n); | |
1422 | else { | |
1423 | printk("io_thread - short read, fd = %d, " | |
1424 | "length = %d\n", kernel_fd, n); | |
1425 | } | |
1426 | continue; | |
1427 | } | |
1428 | io_count++; | |
2adcec21 | 1429 | do_io(req); |
a6ea4cce | 1430 | n = os_write_file(kernel_fd, &req, |
2adcec21 JD |
1431 | sizeof(struct io_thread_req *)); |
1432 | if(n != sizeof(struct io_thread_req *)) | |
91acb21f JD |
1433 | printk("io_thread - write failed, fd = %d, err = %d\n", |
1434 | kernel_fd, -n); | |
1435 | } | |
91acb21f | 1436 | |
1b57e9c2 JD |
1437 | return 0; |
1438 | } |