diff options
author | Jeff Dike <jdike@addtoit.com> | 2005-10-10 23:10:32 -0400 |
---|---|---|
committer | Linus Torvalds <torvalds@g5.osdl.org> | 2005-10-12 11:22:26 -0400 |
commit | 91acb21f084aa244f26839406ae7ed8aa3668058 (patch) | |
tree | 5a186b9b35d2e26aaf1c85441125611c9c88dd18 /arch/um/drivers | |
parent | da64c6ee6bb71bfb3f09d9bb89ce1aa4b1ee7e89 (diff) |
[PATCH] uml: revert block driver use of host AIO
The patch to use host AIO support that I submitted early after 2.6.13 exposed
some problems in the block driver. I have fixes for these, but am not
comfortable putting them into 2.6.14 at this late date. So, this patch reverts
the use of host AIO.
I will resubmit the original patch, plus fixes to the driver after 2.6.14
in order to get a reasonable amount of testing before they're exposed to
the general public.
Signed-off-by: Jeff Dike <jdike@addtoit.com>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
Diffstat (limited to 'arch/um/drivers')
-rw-r--r-- | arch/um/drivers/Makefile | 2 | ||||
-rw-r--r-- | arch/um/drivers/ubd_kern.c | 556 | ||||
-rw-r--r-- | arch/um/drivers/ubd_user.c | 75 |
3 files changed, 386 insertions, 247 deletions
diff --git a/arch/um/drivers/Makefile b/arch/um/drivers/Makefile index 783e18cae090..de17d4c6e02d 100644 --- a/arch/um/drivers/Makefile +++ b/arch/um/drivers/Makefile | |||
@@ -13,7 +13,7 @@ mcast-objs := mcast_kern.o mcast_user.o | |||
13 | net-objs := net_kern.o net_user.o | 13 | net-objs := net_kern.o net_user.o |
14 | mconsole-objs := mconsole_kern.o mconsole_user.o | 14 | mconsole-objs := mconsole_kern.o mconsole_user.o |
15 | hostaudio-objs := hostaudio_kern.o | 15 | hostaudio-objs := hostaudio_kern.o |
16 | ubd-objs := ubd_kern.o | 16 | ubd-objs := ubd_kern.o ubd_user.o |
17 | port-objs := port_kern.o port_user.o | 17 | port-objs := port_kern.o port_user.o |
18 | harddog-objs := harddog_kern.o harddog_user.o | 18 | harddog-objs := harddog_kern.o harddog_user.o |
19 | 19 | ||
diff --git a/arch/um/drivers/ubd_kern.c b/arch/um/drivers/ubd_kern.c index e77a38da4350..f73134333f64 100644 --- a/arch/um/drivers/ubd_kern.c +++ b/arch/um/drivers/ubd_kern.c | |||
@@ -35,7 +35,6 @@ | |||
35 | #include "linux/blkpg.h" | 35 | #include "linux/blkpg.h" |
36 | #include "linux/genhd.h" | 36 | #include "linux/genhd.h" |
37 | #include "linux/spinlock.h" | 37 | #include "linux/spinlock.h" |
38 | #include "asm/atomic.h" | ||
39 | #include "asm/segment.h" | 38 | #include "asm/segment.h" |
40 | #include "asm/uaccess.h" | 39 | #include "asm/uaccess.h" |
41 | #include "asm/irq.h" | 40 | #include "asm/irq.h" |
@@ -54,21 +53,20 @@ | |||
54 | #include "mem.h" | 53 | #include "mem.h" |
55 | #include "mem_kern.h" | 54 | #include "mem_kern.h" |
56 | #include "cow.h" | 55 | #include "cow.h" |
57 | #include "aio.h" | ||
58 | 56 | ||
59 | enum ubd_req { UBD_READ, UBD_WRITE }; | 57 | enum ubd_req { UBD_READ, UBD_WRITE }; |
60 | 58 | ||
61 | struct io_thread_req { | 59 | struct io_thread_req { |
62 | enum aio_type op; | 60 | enum ubd_req op; |
63 | int fds[2]; | 61 | int fds[2]; |
64 | unsigned long offsets[2]; | 62 | unsigned long offsets[2]; |
65 | unsigned long long offset; | 63 | unsigned long long offset; |
66 | unsigned long length; | 64 | unsigned long length; |
67 | char *buffer; | 65 | char *buffer; |
68 | int sectorsize; | 66 | int sectorsize; |
69 | int bitmap_offset; | 67 | unsigned long sector_mask; |
70 | long bitmap_start; | 68 | unsigned long long cow_offset; |
71 | long bitmap_end; | 69 | unsigned long bitmap_words[2]; |
72 | int error; | 70 | int error; |
73 | }; | 71 | }; |
74 | 72 | ||
@@ -82,31 +80,28 @@ extern int create_cow_file(char *cow_file, char *backing_file, | |||
82 | unsigned long *bitmap_len_out, | 80 | unsigned long *bitmap_len_out, |
83 | int *data_offset_out); | 81 | int *data_offset_out); |
84 | extern int read_cow_bitmap(int fd, void *buf, int offset, int len); | 82 | extern int read_cow_bitmap(int fd, void *buf, int offset, int len); |
85 | extern void do_io(struct io_thread_req *req, struct request *r, | 83 | extern void do_io(struct io_thread_req *req); |
86 | unsigned long *bitmap); | ||
87 | 84 | ||
88 | static inline int ubd_test_bit(__u64 bit, void *data) | 85 | static inline int ubd_test_bit(__u64 bit, unsigned char *data) |
89 | { | 86 | { |
90 | unsigned char *buffer = data; | ||
91 | __u64 n; | 87 | __u64 n; |
92 | int bits, off; | 88 | int bits, off; |
93 | 89 | ||
94 | bits = sizeof(buffer[0]) * 8; | 90 | bits = sizeof(data[0]) * 8; |
95 | n = bit / bits; | 91 | n = bit / bits; |
96 | off = bit % bits; | 92 | off = bit % bits; |
97 | return((buffer[n] & (1 << off)) != 0); | 93 | return((data[n] & (1 << off)) != 0); |
98 | } | 94 | } |
99 | 95 | ||
100 | static inline void ubd_set_bit(__u64 bit, void *data) | 96 | static inline void ubd_set_bit(__u64 bit, unsigned char *data) |
101 | { | 97 | { |
102 | unsigned char *buffer = data; | ||
103 | __u64 n; | 98 | __u64 n; |
104 | int bits, off; | 99 | int bits, off; |
105 | 100 | ||
106 | bits = sizeof(buffer[0]) * 8; | 101 | bits = sizeof(data[0]) * 8; |
107 | n = bit / bits; | 102 | n = bit / bits; |
108 | off = bit % bits; | 103 | off = bit % bits; |
109 | buffer[n] |= (1 << off); | 104 | data[n] |= (1 << off); |
110 | } | 105 | } |
111 | /*End stuff from ubd_user.h*/ | 106 | /*End stuff from ubd_user.h*/ |
112 | 107 | ||
@@ -115,6 +110,8 @@ static inline void ubd_set_bit(__u64 bit, void *data) | |||
115 | static DEFINE_SPINLOCK(ubd_io_lock); | 110 | static DEFINE_SPINLOCK(ubd_io_lock); |
116 | static DEFINE_SPINLOCK(ubd_lock); | 111 | static DEFINE_SPINLOCK(ubd_lock); |
117 | 112 | ||
113 | static void (*do_ubd)(void); | ||
114 | |||
118 | static int ubd_open(struct inode * inode, struct file * filp); | 115 | static int ubd_open(struct inode * inode, struct file * filp); |
119 | static int ubd_release(struct inode * inode, struct file * file); | 116 | static int ubd_release(struct inode * inode, struct file * file); |
120 | static int ubd_ioctl(struct inode * inode, struct file * file, | 117 | static int ubd_ioctl(struct inode * inode, struct file * file, |
@@ -161,8 +158,6 @@ struct cow { | |||
161 | int data_offset; | 158 | int data_offset; |
162 | }; | 159 | }; |
163 | 160 | ||
164 | #define MAX_SG 64 | ||
165 | |||
166 | struct ubd { | 161 | struct ubd { |
167 | char *file; | 162 | char *file; |
168 | int count; | 163 | int count; |
@@ -173,7 +168,6 @@ struct ubd { | |||
173 | int no_cow; | 168 | int no_cow; |
174 | struct cow cow; | 169 | struct cow cow; |
175 | struct platform_device pdev; | 170 | struct platform_device pdev; |
176 | struct scatterlist sg[MAX_SG]; | ||
177 | }; | 171 | }; |
178 | 172 | ||
179 | #define DEFAULT_COW { \ | 173 | #define DEFAULT_COW { \ |
@@ -466,114 +460,81 @@ __uml_help(fakehd, | |||
466 | ); | 460 | ); |
467 | 461 | ||
468 | static void do_ubd_request(request_queue_t * q); | 462 | static void do_ubd_request(request_queue_t * q); |
469 | static int in_ubd; | 463 | |
464 | /* Only changed by ubd_init, which is an initcall. */ | ||
465 | int thread_fd = -1; | ||
470 | 466 | ||
471 | /* Changed by ubd_handler, which is serialized because interrupts only | 467 | /* Changed by ubd_handler, which is serialized because interrupts only |
472 | * happen on CPU 0. | 468 | * happen on CPU 0. |
473 | */ | 469 | */ |
474 | int intr_count = 0; | 470 | int intr_count = 0; |
475 | 471 | ||
476 | static void ubd_end_request(struct request *req, int bytes, int uptodate) | 472 | /* call ubd_finish if you need to serialize */ |
473 | static void __ubd_finish(struct request *req, int error) | ||
477 | { | 474 | { |
478 | if (!end_that_request_first(req, uptodate, bytes >> 9)) { | 475 | int nsect; |
479 | add_disk_randomness(req->rq_disk); | 476 | |
480 | end_that_request_last(req); | 477 | if(error){ |
478 | end_request(req, 0); | ||
479 | return; | ||
481 | } | 480 | } |
481 | nsect = req->current_nr_sectors; | ||
482 | req->sector += nsect; | ||
483 | req->buffer += nsect << 9; | ||
484 | req->errors = 0; | ||
485 | req->nr_sectors -= nsect; | ||
486 | req->current_nr_sectors = 0; | ||
487 | end_request(req, 1); | ||
482 | } | 488 | } |
483 | 489 | ||
484 | /* call ubd_finish if you need to serialize */ | 490 | static inline void ubd_finish(struct request *req, int error) |
485 | static void __ubd_finish(struct request *req, int bytes) | ||
486 | { | 491 | { |
487 | if(bytes < 0){ | 492 | spin_lock(&ubd_io_lock); |
488 | ubd_end_request(req, 0, 0); | 493 | __ubd_finish(req, error); |
489 | return; | 494 | spin_unlock(&ubd_io_lock); |
490 | } | ||
491 | |||
492 | ubd_end_request(req, bytes, 1); | ||
493 | } | 495 | } |
494 | 496 | ||
495 | static inline void ubd_finish(struct request *req, int bytes) | 497 | /* Called without ubd_io_lock held */ |
498 | static void ubd_handler(void) | ||
496 | { | 499 | { |
497 | spin_lock(&ubd_io_lock); | 500 | struct io_thread_req req; |
498 | __ubd_finish(req, bytes); | 501 | struct request *rq = elv_next_request(ubd_queue); |
499 | spin_unlock(&ubd_io_lock); | 502 | int n; |
503 | |||
504 | do_ubd = NULL; | ||
505 | intr_count++; | ||
506 | n = os_read_file(thread_fd, &req, sizeof(req)); | ||
507 | if(n != sizeof(req)){ | ||
508 | printk(KERN_ERR "Pid %d - spurious interrupt in ubd_handler, " | ||
509 | "err = %d\n", os_getpid(), -n); | ||
510 | spin_lock(&ubd_io_lock); | ||
511 | end_request(rq, 0); | ||
512 | spin_unlock(&ubd_io_lock); | ||
513 | return; | ||
514 | } | ||
515 | |||
516 | ubd_finish(rq, req.error); | ||
517 | reactivate_fd(thread_fd, UBD_IRQ); | ||
518 | do_ubd_request(ubd_queue); | ||
500 | } | 519 | } |
501 | 520 | ||
502 | struct bitmap_io { | ||
503 | atomic_t count; | ||
504 | struct aio_context aio; | ||
505 | }; | ||
506 | |||
507 | struct ubd_aio { | ||
508 | struct aio_context aio; | ||
509 | struct request *req; | ||
510 | int len; | ||
511 | struct bitmap_io *bitmap; | ||
512 | void *bitmap_buf; | ||
513 | }; | ||
514 | |||
515 | static int ubd_reply_fd = -1; | ||
516 | |||
517 | static irqreturn_t ubd_intr(int irq, void *dev, struct pt_regs *unused) | 521 | static irqreturn_t ubd_intr(int irq, void *dev, struct pt_regs *unused) |
518 | { | 522 | { |
519 | struct aio_thread_reply reply; | 523 | ubd_handler(); |
520 | struct ubd_aio *aio; | 524 | return(IRQ_HANDLED); |
521 | struct request *req; | 525 | } |
522 | int err, n, fd = (int) (long) dev; | ||
523 | |||
524 | while(1){ | ||
525 | err = os_read_file(fd, &reply, sizeof(reply)); | ||
526 | if(err == -EAGAIN) | ||
527 | break; | ||
528 | if(err < 0){ | ||
529 | printk("ubd_aio_handler - read returned err %d\n", | ||
530 | -err); | ||
531 | break; | ||
532 | } | ||
533 | |||
534 | aio = container_of(reply.data, struct ubd_aio, aio); | ||
535 | n = reply.err; | ||
536 | |||
537 | if(n == 0){ | ||
538 | req = aio->req; | ||
539 | req->nr_sectors -= aio->len >> 9; | ||
540 | |||
541 | if((aio->bitmap != NULL) && | ||
542 | (atomic_dec_and_test(&aio->bitmap->count))){ | ||
543 | aio->aio = aio->bitmap->aio; | ||
544 | aio->len = 0; | ||
545 | kfree(aio->bitmap); | ||
546 | aio->bitmap = NULL; | ||
547 | submit_aio(&aio->aio); | ||
548 | } | ||
549 | else { | ||
550 | if((req->nr_sectors == 0) && | ||
551 | (aio->bitmap == NULL)){ | ||
552 | int len = req->hard_nr_sectors << 9; | ||
553 | ubd_finish(req, len); | ||
554 | } | ||
555 | |||
556 | if(aio->bitmap_buf != NULL) | ||
557 | kfree(aio->bitmap_buf); | ||
558 | kfree(aio); | ||
559 | } | ||
560 | } | ||
561 | else if(n < 0){ | ||
562 | ubd_finish(aio->req, n); | ||
563 | if(aio->bitmap != NULL) | ||
564 | kfree(aio->bitmap); | ||
565 | if(aio->bitmap_buf != NULL) | ||
566 | kfree(aio->bitmap_buf); | ||
567 | kfree(aio); | ||
568 | } | ||
569 | } | ||
570 | reactivate_fd(fd, UBD_IRQ); | ||
571 | 526 | ||
572 | do_ubd_request(ubd_queue); | 527 | /* Only changed by ubd_init, which is an initcall. */ |
528 | static int io_pid = -1; | ||
573 | 529 | ||
574 | return(IRQ_HANDLED); | 530 | void kill_io_thread(void) |
531 | { | ||
532 | if(io_pid != -1) | ||
533 | os_kill_process(io_pid, 1); | ||
575 | } | 534 | } |
576 | 535 | ||
536 | __uml_exitcall(kill_io_thread); | ||
537 | |||
577 | static int ubd_file_size(struct ubd *dev, __u64 *size_out) | 538 | static int ubd_file_size(struct ubd *dev, __u64 *size_out) |
578 | { | 539 | { |
579 | char *file; | 540 | char *file; |
@@ -608,7 +569,7 @@ static int ubd_open_dev(struct ubd *dev) | |||
608 | &dev->cow.data_offset, create_ptr); | 569 | &dev->cow.data_offset, create_ptr); |
609 | 570 | ||
610 | if((dev->fd == -ENOENT) && create_cow){ | 571 | if((dev->fd == -ENOENT) && create_cow){ |
611 | dev->fd = create_cow_file(dev->file, dev->cow.file, | 572 | dev->fd = create_cow_file(dev->file, dev->cow.file, |
612 | dev->openflags, 1 << 9, PAGE_SIZE, | 573 | dev->openflags, 1 << 9, PAGE_SIZE, |
613 | &dev->cow.bitmap_offset, | 574 | &dev->cow.bitmap_offset, |
614 | &dev->cow.bitmap_len, | 575 | &dev->cow.bitmap_len, |
@@ -870,10 +831,6 @@ int ubd_init(void) | |||
870 | { | 831 | { |
871 | int i; | 832 | int i; |
872 | 833 | ||
873 | ubd_reply_fd = init_aio_irq(UBD_IRQ, "ubd", ubd_intr); | ||
874 | if(ubd_reply_fd < 0) | ||
875 | printk("Setting up ubd AIO failed, err = %d\n", ubd_reply_fd); | ||
876 | |||
877 | devfs_mk_dir("ubd"); | 834 | devfs_mk_dir("ubd"); |
878 | if (register_blkdev(MAJOR_NR, "ubd")) | 835 | if (register_blkdev(MAJOR_NR, "ubd")) |
879 | return -1; | 836 | return -1; |
@@ -884,7 +841,6 @@ int ubd_init(void) | |||
884 | return -1; | 841 | return -1; |
885 | } | 842 | } |
886 | 843 | ||
887 | blk_queue_max_hw_segments(ubd_queue, MAX_SG); | ||
888 | if (fake_major != MAJOR_NR) { | 844 | if (fake_major != MAJOR_NR) { |
889 | char name[sizeof("ubd_nnn\0")]; | 845 | char name[sizeof("ubd_nnn\0")]; |
890 | 846 | ||
@@ -896,12 +852,40 @@ int ubd_init(void) | |||
896 | driver_register(&ubd_driver); | 852 | driver_register(&ubd_driver); |
897 | for (i = 0; i < MAX_DEV; i++) | 853 | for (i = 0; i < MAX_DEV; i++) |
898 | ubd_add(i); | 854 | ubd_add(i); |
899 | |||
900 | return 0; | 855 | return 0; |
901 | } | 856 | } |
902 | 857 | ||
903 | late_initcall(ubd_init); | 858 | late_initcall(ubd_init); |
904 | 859 | ||
860 | int ubd_driver_init(void){ | ||
861 | unsigned long stack; | ||
862 | int err; | ||
863 | |||
864 | /* Set by CONFIG_BLK_DEV_UBD_SYNC or ubd=sync.*/ | ||
865 | if(global_openflags.s){ | ||
866 | printk(KERN_INFO "ubd: Synchronous mode\n"); | ||
867 | /* Letting ubd=sync be like using ubd#s= instead of ubd#= is | ||
868 | * enough. So use anyway the io thread. */ | ||
869 | } | ||
870 | stack = alloc_stack(0, 0); | ||
871 | io_pid = start_io_thread(stack + PAGE_SIZE - sizeof(void *), | ||
872 | &thread_fd); | ||
873 | if(io_pid < 0){ | ||
874 | printk(KERN_ERR | ||
875 | "ubd : Failed to start I/O thread (errno = %d) - " | ||
876 | "falling back to synchronous I/O\n", -io_pid); | ||
877 | io_pid = -1; | ||
878 | return(0); | ||
879 | } | ||
880 | err = um_request_irq(UBD_IRQ, thread_fd, IRQ_READ, ubd_intr, | ||
881 | SA_INTERRUPT, "ubd", ubd_dev); | ||
882 | if(err != 0) | ||
883 | printk(KERN_ERR "um_request_irq failed - errno = %d\n", -err); | ||
884 | return(err); | ||
885 | } | ||
886 | |||
887 | device_initcall(ubd_driver_init); | ||
888 | |||
905 | static int ubd_open(struct inode *inode, struct file *filp) | 889 | static int ubd_open(struct inode *inode, struct file *filp) |
906 | { | 890 | { |
907 | struct gendisk *disk = inode->i_bdev->bd_disk; | 891 | struct gendisk *disk = inode->i_bdev->bd_disk; |
@@ -939,55 +923,105 @@ static int ubd_release(struct inode * inode, struct file * file) | |||
939 | return(0); | 923 | return(0); |
940 | } | 924 | } |
941 | 925 | ||
942 | static void cowify_bitmap(struct io_thread_req *req, unsigned long *bitmap) | 926 | static void cowify_bitmap(__u64 io_offset, int length, unsigned long *cow_mask, |
927 | __u64 *cow_offset, unsigned long *bitmap, | ||
928 | __u64 bitmap_offset, unsigned long *bitmap_words, | ||
929 | __u64 bitmap_len) | ||
943 | { | 930 | { |
944 | __u64 sector = req->offset / req->sectorsize; | 931 | __u64 sector = io_offset >> 9; |
945 | int i; | 932 | int i, update_bitmap = 0; |
933 | |||
934 | for(i = 0; i < length >> 9; i++){ | ||
935 | if(cow_mask != NULL) | ||
936 | ubd_set_bit(i, (unsigned char *) cow_mask); | ||
937 | if(ubd_test_bit(sector + i, (unsigned char *) bitmap)) | ||
938 | continue; | ||
946 | 939 | ||
947 | for(i = 0; i < req->length / req->sectorsize; i++){ | 940 | update_bitmap = 1; |
948 | if(ubd_test_bit(sector + i, bitmap)) | 941 | ubd_set_bit(sector + i, (unsigned char *) bitmap); |
949 | continue; | 942 | } |
943 | |||
944 | if(!update_bitmap) | ||
945 | return; | ||
950 | 946 | ||
951 | if(req->bitmap_start == -1) | 947 | *cow_offset = sector / (sizeof(unsigned long) * 8); |
952 | req->bitmap_start = sector + i; | ||
953 | req->bitmap_end = sector + i + 1; | ||
954 | 948 | ||
955 | ubd_set_bit(sector + i, bitmap); | 949 | /* This takes care of the case where we're exactly at the end of the |
956 | } | 950 | * device, and *cow_offset + 1 is off the end. So, just back it up |
951 | * by one word. Thanks to Lynn Kerby for the fix and James McMechan | ||
952 | * for the original diagnosis. | ||
953 | */ | ||
954 | if(*cow_offset == ((bitmap_len + sizeof(unsigned long) - 1) / | ||
955 | sizeof(unsigned long) - 1)) | ||
956 | (*cow_offset)--; | ||
957 | |||
958 | bitmap_words[0] = bitmap[*cow_offset]; | ||
959 | bitmap_words[1] = bitmap[*cow_offset + 1]; | ||
960 | |||
961 | *cow_offset *= sizeof(unsigned long); | ||
962 | *cow_offset += bitmap_offset; | ||
963 | } | ||
964 | |||
965 | static void cowify_req(struct io_thread_req *req, unsigned long *bitmap, | ||
966 | __u64 bitmap_offset, __u64 bitmap_len) | ||
967 | { | ||
968 | __u64 sector = req->offset >> 9; | ||
969 | int i; | ||
970 | |||
971 | if(req->length > (sizeof(req->sector_mask) * 8) << 9) | ||
972 | panic("Operation too long"); | ||
973 | |||
974 | if(req->op == UBD_READ) { | ||
975 | for(i = 0; i < req->length >> 9; i++){ | ||
976 | if(ubd_test_bit(sector + i, (unsigned char *) bitmap)) | ||
977 | ubd_set_bit(i, (unsigned char *) | ||
978 | &req->sector_mask); | ||
979 | } | ||
980 | } | ||
981 | else cowify_bitmap(req->offset, req->length, &req->sector_mask, | ||
982 | &req->cow_offset, bitmap, bitmap_offset, | ||
983 | req->bitmap_words, bitmap_len); | ||
957 | } | 984 | } |
958 | 985 | ||
959 | /* Called with ubd_io_lock held */ | 986 | /* Called with ubd_io_lock held */ |
960 | static int prepare_request(struct request *req, struct io_thread_req *io_req, | 987 | static int prepare_request(struct request *req, struct io_thread_req *io_req) |
961 | unsigned long long offset, int page_offset, | ||
962 | int len, struct page *page) | ||
963 | { | 988 | { |
964 | struct gendisk *disk = req->rq_disk; | 989 | struct gendisk *disk = req->rq_disk; |
965 | struct ubd *dev = disk->private_data; | 990 | struct ubd *dev = disk->private_data; |
991 | __u64 offset; | ||
992 | int len; | ||
993 | |||
994 | if(req->rq_status == RQ_INACTIVE) return(1); | ||
966 | 995 | ||
967 | /* This should be impossible now */ | 996 | /* This should be impossible now */ |
968 | if((rq_data_dir(req) == WRITE) && !dev->openflags.w){ | 997 | if((rq_data_dir(req) == WRITE) && !dev->openflags.w){ |
969 | printk("Write attempted on readonly ubd device %s\n", | 998 | printk("Write attempted on readonly ubd device %s\n", |
970 | disk->disk_name); | 999 | disk->disk_name); |
971 | ubd_end_request(req, 0, 0); | 1000 | end_request(req, 0); |
972 | return(1); | 1001 | return(1); |
973 | } | 1002 | } |
974 | 1003 | ||
1004 | offset = ((__u64) req->sector) << 9; | ||
1005 | len = req->current_nr_sectors << 9; | ||
1006 | |||
975 | io_req->fds[0] = (dev->cow.file != NULL) ? dev->cow.fd : dev->fd; | 1007 | io_req->fds[0] = (dev->cow.file != NULL) ? dev->cow.fd : dev->fd; |
976 | io_req->fds[1] = dev->fd; | 1008 | io_req->fds[1] = dev->fd; |
1009 | io_req->cow_offset = -1; | ||
977 | io_req->offset = offset; | 1010 | io_req->offset = offset; |
978 | io_req->length = len; | 1011 | io_req->length = len; |
979 | io_req->error = 0; | 1012 | io_req->error = 0; |
980 | io_req->op = (rq_data_dir(req) == READ) ? AIO_READ : AIO_WRITE; | 1013 | io_req->sector_mask = 0; |
1014 | |||
1015 | io_req->op = (rq_data_dir(req) == READ) ? UBD_READ : UBD_WRITE; | ||
981 | io_req->offsets[0] = 0; | 1016 | io_req->offsets[0] = 0; |
982 | io_req->offsets[1] = dev->cow.data_offset; | 1017 | io_req->offsets[1] = dev->cow.data_offset; |
983 | io_req->buffer = page_address(page) + page_offset; | 1018 | io_req->buffer = req->buffer; |
984 | io_req->sectorsize = 1 << 9; | 1019 | io_req->sectorsize = 1 << 9; |
985 | io_req->bitmap_offset = dev->cow.bitmap_offset; | ||
986 | io_req->bitmap_start = -1; | ||
987 | io_req->bitmap_end = -1; | ||
988 | 1020 | ||
989 | if((dev->cow.file != NULL) && (io_req->op == UBD_WRITE)) | 1021 | if(dev->cow.file != NULL) |
990 | cowify_bitmap(io_req, dev->cow.bitmap); | 1022 | cowify_req(io_req, dev->cow.bitmap, dev->cow.bitmap_offset, |
1023 | dev->cow.bitmap_len); | ||
1024 | |||
991 | return(0); | 1025 | return(0); |
992 | } | 1026 | } |
993 | 1027 | ||
@@ -996,36 +1030,30 @@ static void do_ubd_request(request_queue_t *q) | |||
996 | { | 1030 | { |
997 | struct io_thread_req io_req; | 1031 | struct io_thread_req io_req; |
998 | struct request *req; | 1032 | struct request *req; |
999 | __u64 sector; | 1033 | int err, n; |
1000 | int err; | 1034 | |
1001 | 1035 | if(thread_fd == -1){ | |
1002 | if(in_ubd) | 1036 | while((req = elv_next_request(q)) != NULL){ |
1003 | return; | 1037 | err = prepare_request(req, &io_req); |
1004 | in_ubd = 1; | 1038 | if(!err){ |
1005 | while((req = elv_next_request(q)) != NULL){ | 1039 | do_io(&io_req); |
1006 | struct gendisk *disk = req->rq_disk; | 1040 | __ubd_finish(req, io_req.error); |
1007 | struct ubd *dev = disk->private_data; | 1041 | } |
1008 | int n, i; | 1042 | } |
1009 | 1043 | } | |
1010 | blkdev_dequeue_request(req); | 1044 | else { |
1011 | 1045 | if(do_ubd || (req = elv_next_request(q)) == NULL) | |
1012 | sector = req->sector; | 1046 | return; |
1013 | n = blk_rq_map_sg(q, req, dev->sg); | 1047 | err = prepare_request(req, &io_req); |
1014 | 1048 | if(!err){ | |
1015 | for(i = 0; i < n; i++){ | 1049 | do_ubd = ubd_handler; |
1016 | struct scatterlist *sg = &dev->sg[i]; | 1050 | n = os_write_file(thread_fd, (char *) &io_req, |
1017 | 1051 | sizeof(io_req)); | |
1018 | err = prepare_request(req, &io_req, sector << 9, | 1052 | if(n != sizeof(io_req)) |
1019 | sg->offset, sg->length, | 1053 | printk("write to io thread failed, " |
1020 | sg->page); | 1054 | "errno = %d\n", -n); |
1021 | if(err) | ||
1022 | continue; | ||
1023 | |||
1024 | sector += sg->length >> 9; | ||
1025 | do_io(&io_req, req, dev->cow.bitmap); | ||
1026 | } | 1055 | } |
1027 | } | 1056 | } |
1028 | in_ubd = 0; | ||
1029 | } | 1057 | } |
1030 | 1058 | ||
1031 | static int ubd_ioctl(struct inode * inode, struct file * file, | 1059 | static int ubd_ioctl(struct inode * inode, struct file * file, |
@@ -1241,95 +1269,131 @@ int create_cow_file(char *cow_file, char *backing_file, struct openflags flags, | |||
1241 | return(err); | 1269 | return(err); |
1242 | } | 1270 | } |
1243 | 1271 | ||
1244 | void do_io(struct io_thread_req *req, struct request *r, unsigned long *bitmap) | 1272 | static int update_bitmap(struct io_thread_req *req) |
1245 | { | 1273 | { |
1246 | struct ubd_aio *aio; | 1274 | int n; |
1247 | struct bitmap_io *bitmap_io = NULL; | ||
1248 | char *buf; | ||
1249 | void *bitmap_buf = NULL; | ||
1250 | unsigned long len, sector; | ||
1251 | int nsectors, start, end, bit, err; | ||
1252 | __u64 off; | ||
1253 | |||
1254 | if(req->bitmap_start != -1){ | ||
1255 | /* Round up to the nearest word */ | ||
1256 | int round = sizeof(unsigned long); | ||
1257 | len = (req->bitmap_end - req->bitmap_start + | ||
1258 | round * 8 - 1) / (round * 8); | ||
1259 | len *= round; | ||
1260 | |||
1261 | off = req->bitmap_start / (8 * round); | ||
1262 | off *= round; | ||
1263 | |||
1264 | bitmap_io = kmalloc(sizeof(*bitmap_io), GFP_KERNEL); | ||
1265 | if(bitmap_io == NULL){ | ||
1266 | printk("Failed to kmalloc bitmap IO\n"); | ||
1267 | req->error = 1; | ||
1268 | return; | ||
1269 | } | ||
1270 | 1275 | ||
1271 | bitmap_buf = kmalloc(len, GFP_KERNEL); | 1276 | if(req->cow_offset == -1) |
1272 | if(bitmap_buf == NULL){ | 1277 | return(0); |
1273 | printk("do_io : kmalloc of bitmap chunk " | ||
1274 | "failed\n"); | ||
1275 | kfree(bitmap_io); | ||
1276 | req->error = 1; | ||
1277 | return; | ||
1278 | } | ||
1279 | memcpy(bitmap_buf, &bitmap[off / sizeof(bitmap[0])], len); | ||
1280 | |||
1281 | *bitmap_io = ((struct bitmap_io) | ||
1282 | { .count = ATOMIC_INIT(0), | ||
1283 | .aio = INIT_AIO(AIO_WRITE, req->fds[1], | ||
1284 | bitmap_buf, len, | ||
1285 | req->bitmap_offset + off, | ||
1286 | ubd_reply_fd) } ); | ||
1287 | } | ||
1288 | 1278 | ||
1289 | nsectors = req->length / req->sectorsize; | 1279 | n = os_seek_file(req->fds[1], req->cow_offset); |
1290 | start = 0; | 1280 | if(n < 0){ |
1291 | end = nsectors; | 1281 | printk("do_io - bitmap lseek failed : err = %d\n", -n); |
1292 | bit = 0; | 1282 | return(1); |
1293 | do { | 1283 | } |
1294 | if(bitmap != NULL){ | ||
1295 | sector = req->offset / req->sectorsize; | ||
1296 | bit = ubd_test_bit(sector + start, bitmap); | ||
1297 | end = start; | ||
1298 | while((end < nsectors) && | ||
1299 | (ubd_test_bit(sector + end, bitmap) == bit)) | ||
1300 | end++; | ||
1301 | } | ||
1302 | 1284 | ||
1303 | off = req->offsets[bit] + req->offset + | 1285 | n = os_write_file(req->fds[1], &req->bitmap_words, |
1304 | start * req->sectorsize; | 1286 | sizeof(req->bitmap_words)); |
1305 | len = (end - start) * req->sectorsize; | 1287 | if(n != sizeof(req->bitmap_words)){ |
1306 | buf = &req->buffer[start * req->sectorsize]; | 1288 | printk("do_io - bitmap update failed, err = %d fd = %d\n", -n, |
1289 | req->fds[1]); | ||
1290 | return(1); | ||
1291 | } | ||
1307 | 1292 | ||
1308 | aio = kmalloc(sizeof(*aio), GFP_KERNEL); | 1293 | return(0); |
1309 | if(aio == NULL){ | 1294 | } |
1310 | req->error = 1; | ||
1311 | return; | ||
1312 | } | ||
1313 | 1295 | ||
1314 | *aio = ((struct ubd_aio) | 1296 | void do_io(struct io_thread_req *req) |
1315 | { .aio = INIT_AIO(req->op, req->fds[bit], buf, | 1297 | { |
1316 | len, off, ubd_reply_fd), | 1298 | char *buf; |
1317 | .len = len, | 1299 | unsigned long len; |
1318 | .req = r, | 1300 | int n, nsectors, start, end, bit; |
1319 | .bitmap = bitmap_io, | 1301 | int err; |
1320 | .bitmap_buf = bitmap_buf }); | 1302 | __u64 off; |
1321 | 1303 | ||
1322 | if(aio->bitmap != NULL) | 1304 | nsectors = req->length / req->sectorsize; |
1323 | atomic_inc(&aio->bitmap->count); | 1305 | start = 0; |
1324 | 1306 | do { | |
1325 | err = submit_aio(&aio->aio); | 1307 | bit = ubd_test_bit(start, (unsigned char *) &req->sector_mask); |
1326 | if(err){ | 1308 | end = start; |
1327 | printk("do_io - submit_aio failed, " | 1309 | while((end < nsectors) && |
1328 | "err = %d\n", err); | 1310 | (ubd_test_bit(end, (unsigned char *) |
1329 | req->error = 1; | 1311 | &req->sector_mask) == bit)) |
1330 | return; | 1312 | end++; |
1331 | } | 1313 | |
1314 | off = req->offset + req->offsets[bit] + | ||
1315 | start * req->sectorsize; | ||
1316 | len = (end - start) * req->sectorsize; | ||
1317 | buf = &req->buffer[start * req->sectorsize]; | ||
1318 | |||
1319 | err = os_seek_file(req->fds[bit], off); | ||
1320 | if(err < 0){ | ||
1321 | printk("do_io - lseek failed : err = %d\n", -err); | ||
1322 | req->error = 1; | ||
1323 | return; | ||
1324 | } | ||
1325 | if(req->op == UBD_READ){ | ||
1326 | n = 0; | ||
1327 | do { | ||
1328 | buf = &buf[n]; | ||
1329 | len -= n; | ||
1330 | n = os_read_file(req->fds[bit], buf, len); | ||
1331 | if (n < 0) { | ||
1332 | printk("do_io - read failed, err = %d " | ||
1333 | "fd = %d\n", -n, req->fds[bit]); | ||
1334 | req->error = 1; | ||
1335 | return; | ||
1336 | } | ||
1337 | } while((n < len) && (n != 0)); | ||
1338 | if (n < len) memset(&buf[n], 0, len - n); | ||
1339 | } else { | ||
1340 | n = os_write_file(req->fds[bit], buf, len); | ||
1341 | if(n != len){ | ||
1342 | printk("do_io - write failed err = %d " | ||
1343 | "fd = %d\n", -n, req->fds[bit]); | ||
1344 | req->error = 1; | ||
1345 | return; | ||
1346 | } | ||
1347 | } | ||
1348 | |||
1349 | start = end; | ||
1350 | } while(start < nsectors); | ||
1332 | 1351 | ||
1333 | start = end; | 1352 | req->error = update_bitmap(req); |
1334 | } while(start < nsectors); | ||
1335 | } | 1353 | } |
1354 | |||
1355 | /* Changed in start_io_thread, which is serialized by being called only | ||
1356 | * from ubd_init, which is an initcall. | ||
1357 | */ | ||
1358 | int kernel_fd = -1; | ||
1359 | |||
1360 | /* Only changed by the io thread */ | ||
1361 | int io_count = 0; | ||
1362 | |||
1363 | int io_thread(void *arg) | ||
1364 | { | ||
1365 | struct io_thread_req req; | ||
1366 | int n; | ||
1367 | |||
1368 | ignore_sigwinch_sig(); | ||
1369 | while(1){ | ||
1370 | n = os_read_file(kernel_fd, &req, sizeof(req)); | ||
1371 | if(n != sizeof(req)){ | ||
1372 | if(n < 0) | ||
1373 | printk("io_thread - read failed, fd = %d, " | ||
1374 | "err = %d\n", kernel_fd, -n); | ||
1375 | else { | ||
1376 | printk("io_thread - short read, fd = %d, " | ||
1377 | "length = %d\n", kernel_fd, n); | ||
1378 | } | ||
1379 | continue; | ||
1380 | } | ||
1381 | io_count++; | ||
1382 | do_io(&req); | ||
1383 | n = os_write_file(kernel_fd, &req, sizeof(req)); | ||
1384 | if(n != sizeof(req)) | ||
1385 | printk("io_thread - write failed, fd = %d, err = %d\n", | ||
1386 | kernel_fd, -n); | ||
1387 | } | ||
1388 | } | ||
1389 | |||
1390 | /* | ||
1391 | * Overrides for Emacs so that we follow Linus's tabbing style. | ||
1392 | * Emacs will notice this stuff at the end of the file and automatically | ||
1393 | * adjust the settings for this buffer only. This must remain at the end | ||
1394 | * of the file. | ||
1395 | * --------------------------------------------------------------------------- | ||
1396 | * Local variables: | ||
1397 | * c-file-style: "linux" | ||
1398 | * End: | ||
1399 | */ | ||
diff --git a/arch/um/drivers/ubd_user.c b/arch/um/drivers/ubd_user.c new file mode 100644 index 000000000000..b94d2bc4fe06 --- /dev/null +++ b/arch/um/drivers/ubd_user.c | |||
@@ -0,0 +1,75 @@ | |||
1 | /* | ||
2 | * Copyright (C) 2000, 2001, 2002 Jeff Dike (jdike@karaya.com) | ||
3 | * Copyright (C) 2001 Ridgerun,Inc (glonnon@ridgerun.com) | ||
4 | * Licensed under the GPL | ||
5 | */ | ||
6 | |||
7 | #include <stddef.h> | ||
8 | #include <unistd.h> | ||
9 | #include <errno.h> | ||
10 | #include <sched.h> | ||
11 | #include <signal.h> | ||
12 | #include <string.h> | ||
13 | #include <netinet/in.h> | ||
14 | #include <sys/time.h> | ||
15 | #include <sys/socket.h> | ||
16 | #include <sys/mman.h> | ||
17 | #include <sys/param.h> | ||
18 | #include "asm/types.h" | ||
19 | #include "user_util.h" | ||
20 | #include "kern_util.h" | ||
21 | #include "user.h" | ||
22 | #include "ubd_user.h" | ||
23 | #include "os.h" | ||
24 | #include "cow.h" | ||
25 | |||
26 | #include <endian.h> | ||
27 | #include <byteswap.h> | ||
28 | |||
29 | void ignore_sigwinch_sig(void) | ||
30 | { | ||
31 | signal(SIGWINCH, SIG_IGN); | ||
32 | } | ||
33 | |||
34 | int start_io_thread(unsigned long sp, int *fd_out) | ||
35 | { | ||
36 | int pid, fds[2], err; | ||
37 | |||
38 | err = os_pipe(fds, 1, 1); | ||
39 | if(err < 0){ | ||
40 | printk("start_io_thread - os_pipe failed, err = %d\n", -err); | ||
41 | goto out; | ||
42 | } | ||
43 | |||
44 | kernel_fd = fds[0]; | ||
45 | *fd_out = fds[1]; | ||
46 | |||
47 | pid = clone(io_thread, (void *) sp, CLONE_FILES | CLONE_VM | SIGCHLD, | ||
48 | NULL); | ||
49 | if(pid < 0){ | ||
50 | printk("start_io_thread - clone failed : errno = %d\n", errno); | ||
51 | err = -errno; | ||
52 | goto out_close; | ||
53 | } | ||
54 | |||
55 | return(pid); | ||
56 | |||
57 | out_close: | ||
58 | os_close_file(fds[0]); | ||
59 | os_close_file(fds[1]); | ||
60 | kernel_fd = -1; | ||
61 | *fd_out = -1; | ||
62 | out: | ||
63 | return(err); | ||
64 | } | ||
65 | |||
66 | /* | ||
67 | * Overrides for Emacs so that we follow Linus's tabbing style. | ||
68 | * Emacs will notice this stuff at the end of the file and automatically | ||
69 | * adjust the settings for this buffer only. This must remain at the end | ||
70 | * of the file. | ||
71 | * --------------------------------------------------------------------------- | ||
72 | * Local variables: | ||
73 | * c-file-style: "linux" | ||
74 | * End: | ||
75 | */ | ||