13 #ifdef CONFIG_VALGRIND
14 #include <valgrind/memcheck.h>
17 #include <linux/bio.h>
18 #include <linux/blkdev.h>
19 #include <linux/completion.h>
21 #include <linux/kthread.h>
23 #include "tools-util.h"
27 void (*cleanup)(void);
28 void (*read)(struct bio *bio, struct iovec * iov, unsigned i);
29 void (*write)(struct bio *bio, struct iovec * iov, unsigned i);
32 static struct fops *fops;
33 static io_context_t aio_ctx;
34 static atomic_t running_requests;
36 void generic_make_request(struct bio *bio)
39 struct bvec_iter iter;
44 if (bio->bi_opf & REQ_PREFLUSH) {
45 ret = fdatasync(bio->bi_bdev->bd_fd);
47 fprintf(stderr, "fsync error: %m\n");
48 bio->bi_status = BLK_STS_IOERR;
55 bio_for_each_segment(bv, bio, iter)
58 iov = alloca(sizeof(*iov) * i);
61 bio_for_each_segment(bv, bio, iter) {
62 void *start = page_address(bv.bv_page) + bv.bv_offset;
63 size_t len = bv.bv_len;
65 iov[i++] = (struct iovec) {
70 #ifdef CONFIG_VALGRIND
71 /* To be pedantic it should only be on IO completion. */
72 if (bio_op(bio) == REQ_OP_READ)
73 VALGRIND_MAKE_MEM_DEFINED(start, len);
77 switch (bio_op(bio)) {
79 fops->read(bio, iov, i);
82 fops->write(bio, iov, i);
85 ret = fsync(bio->bi_bdev->bd_fd);
87 die("fsync error: %m");
95 static void submit_bio_wait_endio(struct bio *bio)
97 complete(bio->bi_private);
100 int submit_bio_wait(struct bio *bio)
102 struct completion done;
104 init_completion(&done);
105 bio->bi_private = &done;
106 bio->bi_end_io = submit_bio_wait_endio;
107 bio->bi_opf |= REQ_SYNC;
109 wait_for_completion(&done);
111 return blk_status_to_errno(bio->bi_status);
114 int blkdev_issue_discard(struct block_device *bdev,
115 sector_t sector, sector_t nr_sects,
121 int blkdev_issue_zeroout(struct block_device *bdev,
122 sector_t sector, sector_t nr_sects,
123 gfp_t gfp_mask, unsigned flags)
125 /* Not yet implemented: */
129 unsigned bdev_logical_block_size(struct block_device *bdev)
135 ret = fstat(bdev->bd_fd, &statbuf);
138 if (!S_ISBLK(statbuf.st_mode))
139 return statbuf.st_blksize;
141 xioctl(bdev->bd_fd, BLKPBSZGET, &blksize);
145 sector_t get_capacity(struct gendisk *disk)
147 struct block_device *bdev =
148 container_of(disk, struct block_device, __bd_disk);
153 ret = fstat(bdev->bd_fd, &statbuf);
156 if (!S_ISBLK(statbuf.st_mode))
157 return statbuf.st_size >> 9;
159 ret = ioctl(bdev->bd_fd, BLKGETSIZE64, &bytes);
165 void blkdev_put(struct block_device *bdev, void *holder)
167 fdatasync(bdev->bd_fd);
172 struct block_device *blkdev_get_by_path(const char *path, blk_mode_t mode,
173 void *holder, const struct blk_holder_ops *hop)
175 struct block_device *bdev;
178 if ((mode & (BLK_OPEN_READ|BLK_OPEN_WRITE)) == (BLK_OPEN_READ|BLK_OPEN_WRITE))
180 else if (mode & BLK_OPEN_READ)
182 else if (mode & BLK_OPEN_WRITE)
185 if (!(mode & BLK_OPEN_BUFFERED))
188 if (mode & BLK_OPEN_EXCL)
191 fd = open(path, flags);
193 return ERR_PTR(-errno);
195 bdev = malloc(sizeof(*bdev));
196 memset(bdev, 0, sizeof(*bdev));
198 strncpy(bdev->name, path, sizeof(bdev->name));
199 bdev->name[sizeof(bdev->name) - 1] = '\0';
201 bdev->bd_dev = xfstat(fd).st_rdev;
203 bdev->bd_holder = holder;
204 bdev->bd_disk = &bdev->__bd_disk;
205 bdev->bd_disk->bdi = &bdev->bd_disk->__bdi;
206 bdev->queue.backing_dev_info = bdev->bd_disk->bdi;
211 void bdput(struct block_device *bdev)
216 int lookup_bdev(const char *path, dev_t *dev)
221 static void io_fallback(void)
224 if (fops->init == NULL)
225 die("no fallback possible, something is very wrong");
229 static void sync_check(struct bio *bio, int ret)
231 if (ret != bio->bi_iter.bi_size) {
232 die("IO error: %s\n", strerror(-ret));
235 if (bio->bi_opf & REQ_FUA) {
236 ret = fdatasync(bio->bi_bdev->bd_fd);
238 die("fsync error: %s\n", strerror(-ret));
243 static void sync_init(void) {}
245 static void sync_cleanup(void)
251 static void sync_read(struct bio *bio, struct iovec * iov, unsigned i)
254 ssize_t ret = preadv(bio->bi_bdev->bd_fd, iov, i,
255 bio->bi_iter.bi_sector << 9);
256 sync_check(bio, ret);
259 static void sync_write(struct bio *bio, struct iovec * iov, unsigned i)
261 ssize_t ret = pwritev2(bio->bi_bdev->bd_fd, iov, i,
262 bio->bi_iter.bi_sector << 9,
263 bio->bi_opf & REQ_FUA ? RWF_SYNC : 0);
264 sync_check(bio, ret);
267 static DECLARE_WAIT_QUEUE_HEAD(aio_events_completed);
269 static int aio_completion_thread(void *arg)
271 struct io_event events[8], *ev;
276 ret = io_getevents(aio_ctx, 1, ARRAY_SIZE(events),
279 if (ret < 0 && ret == -EINTR)
282 die("io_getevents() error: %s", strerror(-ret));
284 wake_up(&aio_events_completed);
286 for (ev = events; ev < events + ret; ev++) {
287 struct bio *bio = (struct bio *) ev->data;
289 /* This should only happen during blkdev_cleanup() */
291 BUG_ON(atomic_read(&running_requests) != 0);
296 if (ev->res != bio->bi_iter.bi_size)
297 bio->bi_status = BLK_STS_IOERR;
300 atomic_dec(&running_requests);
307 static struct task_struct *aio_task = NULL;
309 static void aio_init(void)
311 struct task_struct *p;
312 long err = io_setup(256, &aio_ctx);
314 p = kthread_run(aio_completion_thread, NULL, "aio_completion");
319 } else if (err == -ENOSYS) {
322 die("io_setup() error: %s", strerror(err));
326 static void aio_cleanup(void)
328 struct task_struct *p = NULL;
332 /* I mean, really?! IO_CMD_NOOP is even defined, but not implemented. */
336 die("pipe err: %s", strerror(ret));
338 /* Wake up the completion thread with spurious work. */
341 .aio_lio_opcode = IO_CMD_PWRITE,
342 .data = NULL, /* Signal to stop */
343 .aio_fildes = fds[1],
347 ret = io_submit(aio_ctx, 1, &iocbp);
349 die("io_submit cleanup err: %s", strerror(-ret));
351 ret = kthread_stop(p);
360 static void aio_op(struct bio *bio, struct iovec *iov, unsigned i, int opcode)
365 .aio_fildes = bio->bi_bdev->bd_fd,
366 .aio_rw_flags = bio->bi_opf & REQ_FUA ? RWF_SYNC : 0,
367 .aio_lio_opcode = opcode,
370 .u.c.offset = bio->bi_iter.bi_sector << 9,
374 atomic_inc(&running_requests);
376 wait_event(aio_events_completed,
377 (ret = io_submit(aio_ctx, 1, &iocbp)) != -EAGAIN);;
380 die("io_submit err: %s", strerror(-ret));
383 static void aio_read(struct bio *bio, struct iovec *iov, unsigned i)
385 aio_op(bio, iov, i, IO_CMD_PREADV);
388 static void aio_write(struct bio *bio, struct iovec * iov, unsigned i)
390 aio_op(bio, iov, i, IO_CMD_PWRITEV);
394 /* not implemented */
395 static void uring_init(void) {
399 struct fops fops_list[] = {
404 .cleanup = aio_cleanup,
409 .cleanup = sync_cleanup,
417 __attribute__((constructor(102)))
418 static void blkdev_init(void)
424 __attribute__((destructor(102)))
425 static void blkdev_cleanup(void)