* Produced at Lawrence Livermore National Laboratory (cf, DISCLAIMER).
* Rewritten for Linux by Brian Behlendorf <behlendorf1@llnl.gov>.
* LLNL-CODE-403049.
- * Copyright (c) 2013 by Delphix. All rights reserved.
+ * Copyright (c) 2012, 2019 by Delphix. All rights reserved.
*/
#include <sys/zfs_context.h>
-#include <sys/spa.h>
+#include <sys/spa_impl.h>
#include <sys/vdev_disk.h>
#include <sys/vdev_impl.h>
+#include <sys/vdev_trim.h>
+#include <sys/abd.h>
#include <sys/fs/zfs.h>
#include <sys/zio.h>
-#include <sys/sunldi.h>
+#include <linux/mod_compat.h>
+#include <linux/msdos_fs.h>
+#include <linux/vfs_compat.h>
char *zfs_vdev_scheduler = VDEV_SCHEDULER;
static void *zfs_vdev_holder = VDEV_HOLDER;
+/* size of the "reserved" partition, in blocks */
+#define EFI_MIN_RESV_SIZE (16 * 1024)
+
/*
* Virtual device vector for disks.
*/
typedef struct dio_request {
- struct completion dr_comp; /* Completion for sync IO */
- atomic_t dr_ref; /* References */
zio_t *dr_zio; /* Parent ZIO */
- int dr_rw; /* Read/Write */
+ atomic_t dr_ref; /* References */
int dr_error; /* Bio error */
int dr_bio_count; /* Count of bio's */
struct bio *dr_bio[0]; /* Attached bio's */
} dio_request_t;
-#ifdef HAVE_OPEN_BDEV_EXCLUSIVE
+#if defined(HAVE_OPEN_BDEV_EXCLUSIVE) || defined(HAVE_BLKDEV_GET_BY_PATH)
static fmode_t
vdev_bdev_mode(int smode)
{
ASSERT3S(smode & (FREAD | FWRITE), !=, 0);
if ((smode & FREAD) && !(smode & FWRITE))
- mode = MS_RDONLY;
+ mode = SB_RDONLY;
return (mode);
}
#endif /* HAVE_OPEN_BDEV_EXCLUSIVE */
+/*
+ * Returns the usable capacity (in bytes) for the partition or disk.
+ */
static uint64_t
bdev_capacity(struct block_device *bdev)
{
- struct hd_struct *part = bdev->bd_part;
+ return (i_size_read(bdev->bd_inode));
+}
+
+/*
+ * Returns the maximum expansion capacity of the block device (in bytes).
+ *
+ * It is possible to expand a vdev when it has been created as a wholedisk
+ * and the containing block device has increased in capacity. Or when the
+ * partition containing the pool has been manually increased in size.
+ *
+ * This function is only responsible for calculating the potential expansion
+ * size so it can be reported by 'zpool list'. The efi_use_whole_disk() is
+ * responsible for verifying the expected partition layout in the wholedisk
+ * case, and updating the partition table if appropriate. Once the partition
+ * size has been increased the additional capacity will be visible using
+ * bdev_capacity().
+ *
+ * The returned maximum expansion capacity is always expected to be larger, or
+ * at the very least equal, to its usable capacity to prevent overestimating
+ * the pool expandsize.
+ */
+static uint64_t
+bdev_max_capacity(struct block_device *bdev, uint64_t wholedisk)
+{
+ uint64_t psize;
+ int64_t available;
- /* The partition capacity referenced by the block device */
- if (part)
- return (part->nr_sects << 9);
+ if (wholedisk && bdev->bd_part != NULL && bdev != bdev->bd_contains) {
+ /*
+ * When reporting maximum expansion capacity for a wholedisk
+ * deduct any capacity which is expected to be lost due to
+ * alignment restrictions. Over reporting this value isn't
+ * harmful and would only result in slightly less capacity
+ * than expected post expansion.
+ * The estimated available space may be slightly smaller than
+ * bdev_capacity() for devices where the number of sectors is
+ * not a multiple of the alignment size and the partition layout
+ * is keeping less than PARTITION_END_ALIGNMENT bytes after the
+ * "reserved" EFI partition: in such cases return the device
+ * usable capacity.
+ */
+ available = i_size_read(bdev->bd_contains->bd_inode) -
+ ((EFI_MIN_RESV_SIZE + NEW_START_BLOCK +
+ PARTITION_END_ALIGNMENT) << SECTOR_BITS);
+ psize = MAX(available, bdev_capacity(bdev));
+ } else {
+ psize = bdev_capacity(bdev);
+ }
- /* Otherwise assume the full device capacity */
- return (get_capacity(bdev->bd_disk) << 9);
+ return (psize);
}
static void
vdev_disk_error(zio_t *zio)
{
-#ifdef ZFS_DEBUG
- printk("ZFS: zio error=%d type=%d offset=%llu size=%llu "
- "flags=%x delay=%llu\n", zio->io_error, zio->io_type,
+ /*
+ * This function can be called in interrupt context, for instance while
+ * handling IRQs coming from a misbehaving disk device; use printk()
+ * which is safe from any context.
+ */
+ printk(KERN_WARNING "zio pool=%s vdev=%s error=%d type=%d "
+ "offset=%llu size=%llu flags=%x\n", spa_name(zio->io_spa),
+ zio->io_vd->vdev_path, zio->io_error, zio->io_type,
(u_longlong_t)zio->io_offset, (u_longlong_t)zio->io_size,
- zio->io_flags, (u_longlong_t)zio->io_delay);
-#endif
+ zio->io_flags);
}
/*
* physical device. This yields the largest possible requests for
* the device with the lowest total overhead.
*/
-static int
+static void
vdev_elevator_switch(vdev_t *v, char *elevator)
{
vdev_disk_t *vd = v->vdev_tsd;
- struct block_device *bdev = vd->vd_bdev;
- struct request_queue *q = bdev_get_queue(bdev);
- char *device = bdev->bd_disk->disk_name;
+ struct request_queue *q;
+ char *device;
int error;
+ for (int c = 0; c < v->vdev_children; c++)
+ vdev_elevator_switch(v->vdev_child[c], elevator);
+
+ if (!v->vdev_ops->vdev_op_leaf || vd->vd_bdev == NULL)
+ return;
+
+ q = bdev_get_queue(vd->vd_bdev);
+ device = vd->vd_bdev->bd_disk->disk_name;
+
/*
* Skip devices which are not whole disks (partitions).
* Device-mapper devices are excepted since they may be whole
* "Skip devices without schedulers" check below will fail.
*/
if (!v->vdev_wholedisk && strncmp(device, "dm-", 3) != 0)
- return (0);
-
- /* Skip devices without schedulers (loop, ram, dm, etc) */
- if (!q->elevator || !blk_queue_stackable(q))
- return (0);
+ return;
/* Leave existing scheduler when set to "none" */
- if (strncmp(elevator, "none", 4) && (strlen(elevator) == 4) == 0)
- return (0);
+ if ((strncmp(elevator, "none", 4) == 0) && (strlen(elevator) == 4))
+ return;
+ /*
+ * The elevator_change() function was available in kernels from
+ * 2.6.36 to 4.11. When not available fall back to using the user
+ * mode helper functionality to set the elevator via sysfs. This
+ * requires /bin/echo and sysfs to be mounted which may not be true
+ * early in the boot process.
+ */
#ifdef HAVE_ELEVATOR_CHANGE
error = elevator_change(q, elevator);
#else
- /*
- * For pre-2.6.36 kernels elevator_change() is not available.
- * Therefore we fall back to using a usermodehelper to echo the
- * elevator into sysfs; This requires /bin/echo and sysfs to be
- * mounted which may not be true early in the boot process.
- */
#define SET_SCHEDULER_CMD \
"exec 0</dev/null " \
" 1>/sys/block/%s/queue/scheduler " \
" 2>/dev/null; " \
"echo %s"
- {
- char *argv[] = { "/bin/sh", "-c", NULL, NULL };
- char *envp[] = { NULL };
+ char *argv[] = { "/bin/sh", "-c", NULL, NULL };
+ char *envp[] = { NULL };
- argv[2] = kmem_asprintf(SET_SCHEDULER_CMD, device, elevator);
- error = call_usermodehelper(argv[0], argv, envp, UMH_WAIT_PROC);
- strfree(argv[2]);
- }
+ argv[2] = kmem_asprintf(SET_SCHEDULER_CMD, device, elevator);
+ error = call_usermodehelper(argv[0], argv, envp, UMH_WAIT_PROC);
+ strfree(argv[2]);
#endif /* HAVE_ELEVATOR_CHANGE */
- if (error)
- printk("ZFS: Unable to set \"%s\" scheduler for %s (%s): %d\n",
+ if (error) {
+ zfs_dbgmsg("Unable to set \"%s\" scheduler for %s (%s): %d",
elevator, v->vdev_path, device, error);
-
- return (error);
-}
-
-/*
- * Expanding a whole disk vdev involves invoking BLKRRPART on the
- * whole disk device. This poses a problem, because BLKRRPART will
- * return EBUSY if one of the disk's partitions is open. That's why
- * we have to do it here, just before opening the data partition.
- * Unfortunately, BLKRRPART works by dropping all partitions and
- * recreating them, which means that for a short time window, all
- * /dev/sdxN device files disappear (until udev recreates them).
- * This means two things:
- * - When we open the data partition just after a BLKRRPART, we
- * can't do it using the normal device file path because of the
- * obvious race condition with udev. Instead, we use reliable
- * kernel APIs to get a handle to the new partition device from
- * the whole disk device.
- * - Because vdev_disk_open() initially needs to find the device
- * using its path, multiple vdev_disk_open() invocations in
- * short succession on the same disk with BLKRRPARTs in the
- * middle have a high probability of failure (because of the
- * race condition with udev). A typical situation where this
- * might happen is when the zpool userspace tool does a
- * TRYIMPORT immediately followed by an IMPORT. For this
- * reason, we only invoke BLKRRPART in the module when strictly
- * necessary (zpool online -e case), and rely on userspace to
- * do it when possible.
- */
-static struct block_device *
-vdev_disk_rrpart(const char *path, int mode, vdev_disk_t *vd)
-{
-#if defined(HAVE_3ARG_BLKDEV_GET) && defined(HAVE_GET_GENDISK)
- struct block_device *bdev, *result = ERR_PTR(-ENXIO);
- struct gendisk *disk;
- int error, partno;
-
- bdev = vdev_bdev_open(path, vdev_bdev_mode(mode), zfs_vdev_holder);
- if (IS_ERR(bdev))
- return (bdev);
-
- disk = get_gendisk(bdev->bd_dev, &partno);
- vdev_bdev_close(bdev, vdev_bdev_mode(mode));
-
- if (disk) {
- bdev = bdget(disk_devt(disk));
- if (bdev) {
- error = blkdev_get(bdev, vdev_bdev_mode(mode), vd);
- if (error == 0)
- error = ioctl_by_bdev(bdev, BLKRRPART, 0);
- vdev_bdev_close(bdev, vdev_bdev_mode(mode));
- }
-
- bdev = bdget_disk(disk, partno);
- if (bdev) {
- error = blkdev_get(bdev,
- vdev_bdev_mode(mode) | FMODE_EXCL, vd);
- if (error == 0)
- result = bdev;
- }
- put_disk(disk);
}
-
- return (result);
-#else
- return (ERR_PTR(-EOPNOTSUPP));
-#endif /* defined(HAVE_3ARG_BLKDEV_GET) && defined(HAVE_GET_GENDISK) */
}
static int
vdev_disk_open(vdev_t *v, uint64_t *psize, uint64_t *max_psize,
uint64_t *ashift)
{
- struct block_device *bdev = ERR_PTR(-ENXIO);
+ struct block_device *bdev;
+ fmode_t mode = vdev_bdev_mode(spa_mode(v->vdev_spa));
+ int count = 0, block_size;
+ int bdev_retry_count = 50;
vdev_disk_t *vd;
- int mode, block_size;
/* Must have a pathname and it must be absolute. */
if (v->vdev_path == NULL || v->vdev_path[0] != '/') {
v->vdev_stat.vs_aux = VDEV_AUX_BAD_LABEL;
- return (EINVAL);
+ vdev_dbgmsg(v, "invalid vdev_path");
+ return (SET_ERROR(EINVAL));
}
/*
- * Reopen the device if it's not currently open. Otherwise,
- * just update the physical size of the device.
+ * Reopen the device if it is currently open. When expanding a
+ * partition force re-scanning the partition table while closed
+ * in order to get an accurate updated block device size. Then
+ * since udev may need to recreate the device links increase the
+ * open retry count before reporting the device as unavailable.
*/
- if (v->vdev_tsd != NULL) {
- ASSERT(v->vdev_reopening);
- vd = v->vdev_tsd;
- goto skip_open;
- }
+ vd = v->vdev_tsd;
+ if (vd) {
+ char disk_name[BDEVNAME_SIZE + 6] = "/dev/";
+ boolean_t reread_part = B_FALSE;
+
+ rw_enter(&vd->vd_lock, RW_WRITER);
+ bdev = vd->vd_bdev;
+ vd->vd_bdev = NULL;
+
+ if (bdev) {
+ if (v->vdev_expanding && bdev != bdev->bd_contains) {
+ bdevname(bdev->bd_contains, disk_name + 5);
+ reread_part = B_TRUE;
+ }
- vd = kmem_zalloc(sizeof (vdev_disk_t), KM_PUSHPAGE);
- if (vd == NULL)
- return (ENOMEM);
+ vdev_bdev_close(bdev, mode);
+ }
+
+ if (reread_part) {
+ bdev = vdev_bdev_open(disk_name, mode, zfs_vdev_holder);
+ if (!IS_ERR(bdev)) {
+ int error = vdev_bdev_reread_part(bdev);
+ vdev_bdev_close(bdev, mode);
+ if (error == 0)
+ bdev_retry_count = 100;
+ }
+ }
+ } else {
+ vd = kmem_zalloc(sizeof (vdev_disk_t), KM_SLEEP);
+
+ rw_init(&vd->vd_lock, NULL, RW_DEFAULT, NULL);
+ rw_enter(&vd->vd_lock, RW_WRITER);
+ }
/*
* Devices are always opened by the path provided at configuration
* time. This means that if the provided path is a udev by-id path
- * then drives may be recabled without an issue. If the provided
+ * then drives may be re-cabled without an issue. If the provided
* path is a udev by-path path, then the physical location information
* will be preserved. This can be critical for more complicated
* configurations where drives are located in specific physical
- * locations to maximize the systems tolerence to component failure.
+ * locations to maximize the systems tolerance to component failure.
+ *
* Alternatively, you can provide your own udev rule to flexibly map
* the drives as you see fit. It is not advised that you use the
* /dev/[hd]d devices which may be reordered due to probing order.
* Devices in the wrong locations will be detected by the higher
* level vdev validation.
+ *
+ * The specified paths may be briefly removed and recreated in
+ * response to udev events. This should be exceptionally unlikely
+ * because the zpool command makes every effort to verify these paths
+ * have already settled prior to reaching this point. Therefore,
+ * a ENOENT failure at this point is highly likely to be transient
+ * and it is reasonable to sleep and retry before giving up. In
+ * practice delays have been observed to be on the order of 100ms.
*/
- mode = spa_mode(v->vdev_spa);
- if (v->vdev_wholedisk && v->vdev_expanding)
- bdev = vdev_disk_rrpart(v->vdev_path, mode, vd);
- if (IS_ERR(bdev))
- bdev = vdev_bdev_open(v->vdev_path,
- vdev_bdev_mode(mode), zfs_vdev_holder);
+ bdev = ERR_PTR(-ENXIO);
+ while (IS_ERR(bdev) && count < bdev_retry_count) {
+ bdev = vdev_bdev_open(v->vdev_path, mode, zfs_vdev_holder);
+ if (unlikely(PTR_ERR(bdev) == -ENOENT)) {
+ schedule_timeout(MSEC_TO_TICK(10));
+ count++;
+ } else if (IS_ERR(bdev)) {
+ break;
+ }
+ }
+
if (IS_ERR(bdev)) {
- kmem_free(vd, sizeof (vdev_disk_t));
- return (-PTR_ERR(bdev));
+ int error = -PTR_ERR(bdev);
+ vdev_dbgmsg(v, "open error=%d count=%d", error, count);
+ vd->vd_bdev = NULL;
+ v->vdev_tsd = vd;
+ rw_exit(&vd->vd_lock);
+ return (SET_ERROR(error));
+ } else {
+ vd->vd_bdev = bdev;
+ v->vdev_tsd = vd;
+ rw_exit(&vd->vd_lock);
}
- v->vdev_tsd = vd;
- vd->vd_bdev = bdev;
+ struct request_queue *q = bdev_get_queue(vd->vd_bdev);
-skip_open:
/* Determine the physical block size */
block_size = vdev_bdev_block_size(vd->vd_bdev);
/* Clear the nowritecache bit, causes vdev_reopen() to try again. */
v->vdev_nowritecache = B_FALSE;
- /* Physical volume size in bytes */
+ /* Set when device reports it supports TRIM. */
+ v->vdev_has_trim = !!blk_queue_discard(q);
+
+ /* Set when device reports it supports secure TRIM. */
+ v->vdev_has_securetrim = !!blk_queue_discard_secure(q);
+
+ /* Inform the ZIO pipeline that we are non-rotational */
+ v->vdev_nonrot = blk_queue_nonrot(q);
+
+ /* Physical volume size in bytes for the partition */
*psize = bdev_capacity(vd->vd_bdev);
- /* TODO: report possible expansion size */
- *max_psize = *psize;
+ /* Physical volume size in bytes including possible expansion space */
+ *max_psize = bdev_max_capacity(vd->vd_bdev, v->vdev_wholedisk);
/* Based on the minimum sector size set the block size */
- *ashift = highbit(MAX(block_size, SPA_MINBLOCKSIZE)) - 1;
+ *ashift = highbit64(MAX(block_size, SPA_MINBLOCKSIZE)) - 1;
/* Try to set the io scheduler elevator algorithm */
(void) vdev_elevator_switch(v, zfs_vdev_scheduler);
if (v->vdev_reopening || vd == NULL)
return;
- if (vd->vd_bdev != NULL)
+ if (vd->vd_bdev != NULL) {
vdev_bdev_close(vd->vd_bdev,
vdev_bdev_mode(spa_mode(v->vdev_spa)));
+ }
+ rw_destroy(&vd->vd_lock);
kmem_free(vd, sizeof (vdev_disk_t));
v->vdev_tsd = NULL;
}
int i;
dr = kmem_zalloc(sizeof (dio_request_t) +
- sizeof (struct bio *) * bio_count, KM_PUSHPAGE);
+ sizeof (struct bio *) * bio_count, KM_SLEEP);
if (dr) {
- init_completion(&dr->dr_comp);
atomic_set(&dr->dr_ref, 0);
dr->dr_bio_count = bio_count;
dr->dr_error = 0;
sizeof (struct bio *) * dr->dr_bio_count);
}
-static int
-vdev_disk_dio_is_sync(dio_request_t *dr)
-{
-#ifdef HAVE_BIO_RW_SYNC
- /* BIO_RW_SYNC preferred interface from 2.6.12-2.6.29 */
- return (dr->dr_rw & (1 << BIO_RW_SYNC));
-#else
-#ifdef HAVE_BIO_RW_SYNCIO
- /* BIO_RW_SYNCIO preferred interface from 2.6.30-2.6.35 */
- return (dr->dr_rw & (1 << BIO_RW_SYNCIO));
-#else
-#ifdef HAVE_REQ_SYNC
- /* REQ_SYNC preferred interface from 2.6.36-2.6.xx */
- return (dr->dr_rw & REQ_SYNC);
-#else
-#error "Unable to determine bio sync flag"
-#endif /* HAVE_REQ_SYNC */
-#endif /* HAVE_BIO_RW_SYNC */
-#endif /* HAVE_BIO_RW_SYNCIO */
-}
-
static void
vdev_disk_dio_get(dio_request_t *dr)
{
vdev_disk_dio_free(dr);
if (zio) {
- zio->io_delay = jiffies_64 - zio->io_delay;
zio->io_error = error;
ASSERT3S(zio->io_error, >=, 0);
if (zio->io_error)
vdev_disk_error(zio);
- zio_interrupt(zio);
+
+ zio_delay_interrupt(zio);
}
}
return (rc);
}
-BIO_END_IO_PROTO(vdev_disk_physio_completion, bio, size, error)
+BIO_END_IO_PROTO(vdev_disk_physio_completion, bio, error)
{
dio_request_t *dr = bio->bi_private;
int rc;
- /* Fatal error but print some useful debugging before asserting */
- if (dr == NULL)
- PANIC("dr == NULL, bio->bi_private == NULL\n"
- "bi_next: %p, bi_flags: %lx, bi_rw: %lu, bi_vcnt: %d\n"
- "bi_idx: %d, bi_size: %d, bi_end_io: %p, bi_cnt: %d\n",
- bio->bi_next, bio->bi_flags, bio->bi_rw, bio->bi_vcnt,
- BIO_BI_IDX(bio), BIO_BI_SIZE(bio), bio->bi_end_io,
- atomic_read(&bio->bi_cnt));
-
-#ifndef HAVE_2ARGS_BIO_END_IO_T
- if (BIO_BI_SIZE(bio))
- return (1);
-#endif /* HAVE_2ARGS_BIO_END_IO_T */
-
- if (error == 0 && !test_bit(BIO_UPTODATE, &bio->bi_flags))
- error = (-EIO);
-
- if (dr->dr_error == 0)
- dr->dr_error = -error;
+ if (dr->dr_error == 0) {
+#ifdef HAVE_1ARG_BIO_END_IO_T
+ dr->dr_error = BIO_END_IO_ERROR(bio);
+#else
+ if (error)
+ dr->dr_error = -(error);
+ else if (!test_bit(BIO_UPTODATE, &bio->bi_flags))
+ dr->dr_error = EIO;
+#endif
+ }
- /* Drop reference aquired by __vdev_disk_physio */
+ /* Drop reference acquired by __vdev_disk_physio */
rc = vdev_disk_dio_put(dr);
-
- /* Wake up synchronous waiter this is the last outstanding bio */
- if ((rc == 1) && vdev_disk_dio_is_sync(dr))
- complete(&dr->dr_comp);
-
- BIO_END_IO_RETURN(0);
-}
-
-static inline unsigned long
-bio_nr_pages(void *bio_ptr, unsigned int bio_size)
-{
- return ((((unsigned long)bio_ptr + bio_size + PAGE_SIZE - 1) >>
- PAGE_SHIFT) - ((unsigned long)bio_ptr >> PAGE_SHIFT));
}
static unsigned int
if (size > bio_size)
size = bio_size;
- if (kmem_virt(bio_ptr))
+ if (is_vmalloc_addr(bio_ptr))
page = vmalloc_to_page(bio_ptr);
else
page = virt_to_page(bio_ptr);
+ /*
+ * Some network related block device uses tcp_sendpage, which
+ * doesn't behave well when using 0-count page, this is a
+ * safety net to catch them.
+ */
+ ASSERT3S(page_count(page), >, 0);
+
if (bio_add_page(bio, page, size, offset) != size)
break;
return (bio_size);
}
+static unsigned int
+bio_map_abd_off(struct bio *bio, abd_t *abd, unsigned int size, size_t off)
+{
+ if (abd_is_linear(abd))
+ return (bio_map(bio, ((char *)abd_to_buf(abd)) + off, size));
+
+ return (abd_scatter_bio_map_off(bio, abd, size, off));
+}
+
+static inline void
+vdev_submit_bio_impl(struct bio *bio)
+{
+#ifdef HAVE_1ARG_SUBMIT_BIO
+ submit_bio(bio);
+#else
+ submit_bio(0, bio);
+#endif
+}
+
+#ifdef HAVE_BIO_SET_DEV
+#if defined(CONFIG_BLK_CGROUP) && defined(HAVE_BIO_SET_DEV_GPL_ONLY)
+/*
+ * The Linux 5.0 kernel updated the bio_set_dev() macro so it calls the
+ * GPL-only bio_associate_blkg() symbol thus inadvertently converting
+ * the entire macro. Provide a minimal version which always assigns the
+ * request queue's root_blkg to the bio.
+ */
+static inline void
+vdev_bio_associate_blkg(struct bio *bio)
+{
+ struct request_queue *q = bio->bi_disk->queue;
+
+ ASSERT3P(q, !=, NULL);
+ ASSERT3P(bio->bi_blkg, ==, NULL);
+
+ if (blkg_tryget(q->root_blkg))
+ bio->bi_blkg = q->root_blkg;
+}
+#define bio_associate_blkg vdev_bio_associate_blkg
+#endif
+#else
+/*
+ * Provide a bio_set_dev() helper macro for pre-Linux 4.14 kernels.
+ */
+static inline void
+bio_set_dev(struct bio *bio, struct block_device *bdev)
+{
+ bio->bi_bdev = bdev;
+}
+#endif /* HAVE_BIO_SET_DEV */
+
+static inline void
+vdev_submit_bio(struct bio *bio)
+{
+#ifdef HAVE_CURRENT_BIO_TAIL
+ struct bio **bio_tail = current->bio_tail;
+ current->bio_tail = NULL;
+ vdev_submit_bio_impl(bio);
+ current->bio_tail = bio_tail;
+#else
+ struct bio_list *bio_list = current->bio_list;
+ current->bio_list = NULL;
+ vdev_submit_bio_impl(bio);
+ current->bio_list = bio_list;
+#endif
+}
+
static int
-__vdev_disk_physio(struct block_device *bdev, zio_t *zio, caddr_t kbuf_ptr,
- size_t kbuf_size, uint64_t kbuf_offset, int flags)
+__vdev_disk_physio(struct block_device *bdev, zio_t *zio,
+ size_t io_size, uint64_t io_offset, int rw, int flags)
{
dio_request_t *dr;
- caddr_t bio_ptr;
+ uint64_t abd_offset;
uint64_t bio_offset;
int bio_size, bio_count = 16;
int i = 0, error = 0;
-
- ASSERT3U(kbuf_offset + kbuf_size, <=, bdev->bd_inode->i_size);
+#if defined(HAVE_BLK_QUEUE_HAVE_BLK_PLUG)
+ struct blk_plug plug;
+#endif
+ /*
+ * Accessing outside the block device is never allowed.
+ */
+ if (io_offset + io_size > bdev->bd_inode->i_size) {
+ vdev_dbgmsg(zio->io_vd,
+ "Illegal access %llu size %llu, device size %llu",
+ io_offset, io_size, i_size_read(bdev->bd_inode));
+ return (SET_ERROR(EIO));
+ }
retry:
dr = vdev_disk_dio_alloc(bio_count);
if (dr == NULL)
- return (ENOMEM);
+ return (SET_ERROR(ENOMEM));
if (zio && !(zio->io_flags & (ZIO_FLAG_IO_RETRY | ZIO_FLAG_TRYHARD)))
- bio_set_flags_failfast(bdev, &flags);
+ bio_set_flags_failfast(bdev, &flags);
dr->dr_zio = zio;
- dr->dr_rw = flags;
/*
* When the IO size exceeds the maximum bio size for the request
* their volume block size to match the maximum request size and
* the common case will be one bio per vdev IO request.
*/
- bio_ptr = kbuf_ptr;
- bio_offset = kbuf_offset;
- bio_size = kbuf_size;
+
+ abd_offset = 0;
+ bio_offset = io_offset;
+ bio_size = io_size;
for (i = 0; i <= dr->dr_bio_count; i++) {
/* Finished constructing bio's for given buffer */
goto retry;
}
+ /* bio_alloc() with __GFP_WAIT never returns NULL */
dr->dr_bio[i] = bio_alloc(GFP_NOIO,
- bio_nr_pages(bio_ptr, bio_size));
- if (dr->dr_bio[i] == NULL) {
+ MIN(abd_nr_pages_off(zio->io_abd, bio_size, abd_offset),
+ BIO_MAX_PAGES));
+ if (unlikely(dr->dr_bio[i] == NULL)) {
vdev_disk_dio_free(dr);
- return (ENOMEM);
+ return (SET_ERROR(ENOMEM));
}
/* Matching put called by vdev_disk_physio_completion */
vdev_disk_dio_get(dr);
- dr->dr_bio[i]->bi_bdev = bdev;
+ bio_set_dev(dr->dr_bio[i], bdev);
BIO_BI_SECTOR(dr->dr_bio[i]) = bio_offset >> 9;
- dr->dr_bio[i]->bi_rw = dr->dr_rw;
dr->dr_bio[i]->bi_end_io = vdev_disk_physio_completion;
dr->dr_bio[i]->bi_private = dr;
+ bio_set_op_attrs(dr->dr_bio[i], rw, flags);
/* Remaining size is returned to become the new size */
- bio_size = bio_map(dr->dr_bio[i], bio_ptr, bio_size);
+ bio_size = bio_map_abd_off(dr->dr_bio[i], zio->io_abd,
+ bio_size, abd_offset);
/* Advance in buffer and construct another bio if needed */
- bio_ptr += BIO_BI_SIZE(dr->dr_bio[i]);
+ abd_offset += BIO_BI_SIZE(dr->dr_bio[i]);
bio_offset += BIO_BI_SIZE(dr->dr_bio[i]);
}
- /* Extra reference to protect dio_request during submit_bio */
+ /* Extra reference to protect dio_request during vdev_submit_bio */
vdev_disk_dio_get(dr);
- if (zio)
- zio->io_delay = jiffies_64;
+
+#if defined(HAVE_BLK_QUEUE_HAVE_BLK_PLUG)
+ if (dr->dr_bio_count > 1)
+ blk_start_plug(&plug);
+#endif
/* Submit all bio's associated with this dio */
for (i = 0; i < dr->dr_bio_count; i++)
if (dr->dr_bio[i])
- submit_bio(dr->dr_rw, dr->dr_bio[i]);
+ vdev_submit_bio(dr->dr_bio[i]);
- /*
- * On synchronous blocking requests we wait for all bio the completion
- * callbacks to run. We will be woken when the last callback runs
- * for this dio. We are responsible for putting the last dio_request
- * reference will in turn put back the last bio references. The
- * only synchronous consumer is vdev_disk_read_rootlabel() all other
- * IO originating from vdev_disk_io_start() is asynchronous.
- */
- if (vdev_disk_dio_is_sync(dr)) {
- wait_for_completion(&dr->dr_comp);
- error = dr->dr_error;
- ASSERT3S(atomic_read(&dr->dr_ref), ==, 1);
- }
+#if defined(HAVE_BLK_QUEUE_HAVE_BLK_PLUG)
+ if (dr->dr_bio_count > 1)
+ blk_finish_plug(&plug);
+#endif
(void) vdev_disk_dio_put(dr);
return (error);
}
-int
-vdev_disk_physio(struct block_device *bdev, caddr_t kbuf,
- size_t size, uint64_t offset, int flags)
-{
- bio_set_flags_failfast(bdev, &flags);
- return (__vdev_disk_physio(bdev, NULL, kbuf, size, offset, flags));
-}
-
-BIO_END_IO_PROTO(vdev_disk_io_flush_completion, bio, size, rc)
+BIO_END_IO_PROTO(vdev_disk_io_flush_completion, bio, error)
{
zio_t *zio = bio->bi_private;
+#ifdef HAVE_1ARG_BIO_END_IO_T
+ zio->io_error = BIO_END_IO_ERROR(bio);
+#else
+ zio->io_error = -error;
+#endif
- zio->io_delay = jiffies_64 - zio->io_delay;
- zio->io_error = -rc;
- if (rc && (rc == -EOPNOTSUPP))
+ if (zio->io_error && (zio->io_error == EOPNOTSUPP))
zio->io_vd->vdev_nowritecache = B_TRUE;
bio_put(bio);
if (zio->io_error)
vdev_disk_error(zio);
zio_interrupt(zio);
-
- BIO_END_IO_RETURN(0);
}
static int
q = bdev_get_queue(bdev);
if (!q)
- return (ENXIO);
+ return (SET_ERROR(ENXIO));
bio = bio_alloc(GFP_NOIO, 0);
- if (!bio)
- return (ENOMEM);
+ /* bio_alloc() with __GFP_WAIT never returns NULL */
+ if (unlikely(bio == NULL))
+ return (SET_ERROR(ENOMEM));
bio->bi_end_io = vdev_disk_io_flush_completion;
bio->bi_private = zio;
- bio->bi_bdev = bdev;
- zio->io_delay = jiffies_64;
- submit_bio(VDEV_WRITE_FLUSH_FUA, bio);
+ bio_set_dev(bio, bdev);
+ bio_set_flush(bio);
+ vdev_submit_bio(bio);
invalidate_bdev(bdev);
return (0);
}
-static int
+static void
vdev_disk_io_start(zio_t *zio)
{
vdev_t *v = zio->io_vd;
vdev_disk_t *vd = v->vdev_tsd;
- int flags, error;
+ unsigned long trim_flags = 0;
+ int rw, flags, error;
+
+ /*
+ * If the vdev is closed, it's likely in the REMOVED or FAULTED state.
+ * Nothing to be done here but return failure.
+ */
+ if (vd == NULL) {
+ zio->io_error = ENXIO;
+ zio_interrupt(zio);
+ return;
+ }
+
+ rw_enter(&vd->vd_lock, RW_READER);
+
+ /*
+ * If the vdev is closed, it's likely due to a failed reopen and is
+ * in the UNAVAIL state. Nothing to be done here but return failure.
+ */
+ if (vd->vd_bdev == NULL) {
+ rw_exit(&vd->vd_lock);
+ zio->io_error = ENXIO;
+ zio_interrupt(zio);
+ return;
+ }
switch (zio->io_type) {
case ZIO_TYPE_IOCTL:
if (!vdev_readable(v)) {
+ rw_exit(&vd->vd_lock);
zio->io_error = SET_ERROR(ENXIO);
- return (ZIO_PIPELINE_CONTINUE);
+ zio_interrupt(zio);
+ return;
}
switch (zio->io_cmd) {
}
error = vdev_disk_io_flush(vd->vd_bdev, zio);
- if (error == 0)
- return (ZIO_PIPELINE_STOP);
+ if (error == 0) {
+ rw_exit(&vd->vd_lock);
+ return;
+ }
zio->io_error = error;
- if (error == ENOTSUP)
- v->vdev_nowritecache = B_TRUE;
break;
zio->io_error = SET_ERROR(ENOTSUP);
}
- return (ZIO_PIPELINE_CONTINUE);
-
+ rw_exit(&vd->vd_lock);
+ zio_execute(zio);
+ return;
case ZIO_TYPE_WRITE:
- flags = WRITE;
+ rw = WRITE;
+#if defined(HAVE_BLK_QUEUE_HAVE_BIO_RW_UNPLUG)
+ flags = (1 << BIO_RW_UNPLUG);
+#elif defined(REQ_UNPLUG)
+ flags = REQ_UNPLUG;
+#else
+ flags = 0;
+#endif
break;
case ZIO_TYPE_READ:
- flags = READ;
+ rw = READ;
+#if defined(HAVE_BLK_QUEUE_HAVE_BIO_RW_UNPLUG)
+ flags = (1 << BIO_RW_UNPLUG);
+#elif defined(REQ_UNPLUG)
+ flags = REQ_UNPLUG;
+#else
+ flags = 0;
+#endif
break;
+ case ZIO_TYPE_TRIM:
+#if defined(BLKDEV_DISCARD_SECURE)
+ if (zio->io_trim_flags & ZIO_TRIM_SECURE)
+ trim_flags |= BLKDEV_DISCARD_SECURE;
+#endif
+ zio->io_error = -blkdev_issue_discard(vd->vd_bdev,
+ zio->io_offset >> 9, zio->io_size >> 9, GFP_NOFS,
+ trim_flags);
+
+ rw_exit(&vd->vd_lock);
+ zio_interrupt(zio);
+ return;
+
default:
+ rw_exit(&vd->vd_lock);
zio->io_error = SET_ERROR(ENOTSUP);
- return (ZIO_PIPELINE_CONTINUE);
+ zio_interrupt(zio);
+ return;
}
- error = __vdev_disk_physio(vd->vd_bdev, zio, zio->io_data,
- zio->io_size, zio->io_offset, flags);
+ zio->io_target_timestamp = zio_handle_io_delay(zio);
+ error = __vdev_disk_physio(vd->vd_bdev, zio,
+ zio->io_size, zio->io_offset, rw, flags);
+ rw_exit(&vd->vd_lock);
+
if (error) {
zio->io_error = error;
- return (ZIO_PIPELINE_CONTINUE);
+ zio_interrupt(zio);
+ return;
}
-
- return (ZIO_PIPELINE_STOP);
}
static void
/* XXX: Implement me as a vnode rele for the device */
}
+static int
+param_set_vdev_scheduler(const char *val, zfs_kernel_param_t *kp)
+{
+ spa_t *spa = NULL;
+ char *p;
+
+ if (val == NULL)
+ return (SET_ERROR(-EINVAL));
+
+ if ((p = strchr(val, '\n')) != NULL)
+ *p = '\0';
+
+ if (spa_mode_global != 0) {
+ mutex_enter(&spa_namespace_lock);
+ while ((spa = spa_next(spa)) != NULL) {
+ if (spa_state(spa) != POOL_STATE_ACTIVE ||
+ !spa_writeable(spa) || spa_suspended(spa))
+ continue;
+
+ spa_open_ref(spa, FTAG);
+ mutex_exit(&spa_namespace_lock);
+ vdev_elevator_switch(spa->spa_root_vdev, (char *)val);
+ mutex_enter(&spa_namespace_lock);
+ spa_close(spa, FTAG);
+ }
+ mutex_exit(&spa_namespace_lock);
+ }
+
+ return (param_set_charp(val, kp));
+}
+
vdev_ops_t vdev_disk_ops = {
vdev_disk_open,
vdev_disk_close,
vdev_disk_io_start,
vdev_disk_io_done,
NULL,
+ NULL,
vdev_disk_hold,
vdev_disk_rele,
+ NULL,
+ vdev_default_xlate,
VDEV_TYPE_DISK, /* name of this vdev type */
B_TRUE /* leaf vdev */
};
-/*
- * Given the root disk device devid or pathname, read the label from
- * the device, and construct a configuration nvlist.
- */
-int
-vdev_disk_read_rootlabel(char *devpath, char *devid, nvlist_t **config)
-{
- struct block_device *bdev;
- vdev_label_t *label;
- uint64_t s, size;
- int i;
-
- bdev = vdev_bdev_open(devpath, vdev_bdev_mode(FREAD), zfs_vdev_holder);
- if (IS_ERR(bdev))
- return (-PTR_ERR(bdev));
-
- s = bdev_capacity(bdev);
- if (s == 0) {
- vdev_bdev_close(bdev, vdev_bdev_mode(FREAD));
- return (EIO);
- }
-
- size = P2ALIGN_TYPED(s, sizeof (vdev_label_t), uint64_t);
- label = vmem_alloc(sizeof (vdev_label_t), KM_PUSHPAGE);
-
- for (i = 0; i < VDEV_LABELS; i++) {
- uint64_t offset, state, txg = 0;
-
- /* read vdev label */
- offset = vdev_label_offset(size, i, 0);
- if (vdev_disk_physio(bdev, (caddr_t)label,
- VDEV_SKIP_SIZE + VDEV_PHYS_SIZE, offset, READ_SYNC) != 0)
- continue;
-
- if (nvlist_unpack(label->vl_vdev_phys.vp_nvlist,
- sizeof (label->vl_vdev_phys.vp_nvlist), config, 0) != 0) {
- *config = NULL;
- continue;
- }
-
- if (nvlist_lookup_uint64(*config, ZPOOL_CONFIG_POOL_STATE,
- &state) != 0 || state >= POOL_STATE_DESTROYED) {
- nvlist_free(*config);
- *config = NULL;
- continue;
- }
-
- if (nvlist_lookup_uint64(*config, ZPOOL_CONFIG_POOL_TXG,
- &txg) != 0 || txg == 0) {
- nvlist_free(*config);
- *config = NULL;
- continue;
- }
-
- break;
- }
-
- vmem_free(label, sizeof (vdev_label_t));
- vdev_bdev_close(bdev, vdev_bdev_mode(FREAD));
-
- return (0);
-}
-
-module_param(zfs_vdev_scheduler, charp, 0644);
+module_param_call(zfs_vdev_scheduler, param_set_vdev_scheduler,
+ param_get_charp, &zfs_vdev_scheduler, 0644);
MODULE_PARM_DESC(zfs_vdev_scheduler, "I/O scheduler");