#include <mach/vm_map.h>
#include <mach/upl.h>
#include <kern/task.h>
+#include <kern/policy_internal.h>
#include <vm/vm_kern.h>
#include <vm/vm_map.h>
#include <stdbool.h>
+#include <vfs/vfs_disk_conditioner.h>
+
#if 0
#undef KERNEL_DEBUG
#define KERNEL_DEBUG KERNEL_DEBUG_CONSTANT
#define MAX_VECTOR_UPL_ELEMENTS 8
#define MAX_VECTOR_UPL_SIZE (2 * MAX_UPL_SIZE_BYTES)
+#define CLUSTER_IO_WAITING ((buf_t)1)
+
extern upl_t vector_upl_create(vm_offset_t);
extern boolean_t vector_upl_is_valid(upl_t);
extern boolean_t vector_upl_set_subupl(upl_t,upl_t, u_int32_t);
static int cluster_push_now(vnode_t vp, struct cl_extent *, off_t EOF, int flags, int (*)(buf_t, void *), void *callback_arg);
-static int cluster_try_push(struct cl_writebehind *, vnode_t vp, off_t EOF, int push_flag, int flags, int (*)(buf_t, void *), void *callback_arg);
+static int cluster_try_push(struct cl_writebehind *, vnode_t vp, off_t EOF, int push_flag, int flags, int (*)(buf_t, void *), void *callback_arg, int *err);
static void sparse_cluster_switch(struct cl_writebehind *, vnode_t vp, off_t EOF, int (*)(buf_t, void *), void *callback_arg);
-static void sparse_cluster_push(void **cmapp, vnode_t vp, off_t EOF, int push_flag, int io_flags, int (*)(buf_t, void *), void *callback_arg);
+static int sparse_cluster_push(void **cmapp, vnode_t vp, off_t EOF, int push_flag, int io_flags, int (*)(buf_t, void *), void *callback_arg);
static void sparse_cluster_add(void **cmapp, vnode_t vp, struct cl_extent *, off_t EOF, int (*)(buf_t, void *), void *callback_arg);
static kern_return_t vfs_drt_mark_pages(void **cmapp, off_t offset, u_int length, u_int *setcountp);
#define WRITE_BEHIND 1
#define WRITE_BEHIND_SSD 1
+#if CONFIG_EMBEDDED
+#define PREFETCH 1
+#define PREFETCH_SSD 1
+uint32_t speculative_prefetch_max = (2048 * 1024); /* maximum bytes in a specluative read-ahead */
+uint32_t speculative_prefetch_max_iosize = (512 * 1024); /* maximum I/O size to use in a specluative read-ahead */
+#else
#define PREFETCH 3
#define PREFETCH_SSD 2
uint32_t speculative_prefetch_max = (MAX_UPL_SIZE_BYTES * 3); /* maximum bytes in a specluative read-ahead */
uint32_t speculative_prefetch_max_iosize = (512 * 1024); /* maximum I/O size to use in a specluative read-ahead on SSDs*/
+#endif
#define IO_SCALE(vp, base) (vp->v_mount->mnt_ioscale * (base))
#define MAX_CLUSTER_SIZE(vp) (cluster_max_io_size(vp->v_mount, CL_WRITE))
-#define MAX_PREFETCH(vp, size, is_ssd) (size * IO_SCALE(vp, ((is_ssd && !ignore_is_ssd) ? PREFETCH_SSD : PREFETCH)))
+#define MAX_PREFETCH(vp, size, is_ssd) (size * IO_SCALE(vp, ((is_ssd) ? PREFETCH_SSD : PREFETCH)))
-int ignore_is_ssd = 0;
int speculative_reads_disabled = 0;
/*
if (wbp->cl_number) {
lck_mtx_lock(&wbp->cl_lockw);
- cluster_try_push(wbp, vp, newEOF, PUSH_ALL | flags, 0, callback, callback_arg);
+ cluster_try_push(wbp, vp, newEOF, PUSH_ALL | flags, 0, callback, callback_arg, NULL);
lck_mtx_unlock(&wbp->cl_lockw);
}
size_t io_size;
int (*bootcache_check_fn)(dev_t device, u_int64_t blkno) = bootcache_contains_block;
- if (bootcache_check_fn) {
- if (VNOP_BLOCKMAP(vp, f_offset, PAGE_SIZE, &blkno, &io_size, NULL, VNODE_READ, NULL))
+ if (bootcache_check_fn && vp->v_mount && vp->v_mount->mnt_devvp) {
+ if (VNOP_BLOCKMAP(vp, f_offset, PAGE_SIZE, &blkno, &io_size, NULL, VNODE_READ | VNODE_BLOCKMAP_NO_TRACK, NULL))
return(0);
if (io_size == 0)
cbp_head, bp->b_lblkno, bp->b_bcount, bp->b_flags, 0);
if (cbp_head->b_trans_next || !(cbp_head->b_flags & B_EOT)) {
- boolean_t need_wakeup = FALSE;
-
lck_mtx_lock_spin(cl_transaction_mtxp);
bp->b_flags |= B_TDONE;
- if (bp->b_flags & B_TWANTED) {
- CLR(bp->b_flags, B_TWANTED);
- need_wakeup = TRUE;
- }
for (cbp = cbp_head; cbp; cbp = cbp->b_trans_next) {
/*
* all I/O requests that are part of this transaction
lck_mtx_unlock(cl_transaction_mtxp);
- if (need_wakeup == TRUE)
- wakeup(bp);
+ return 0;
+ }
+
+ if (cbp->b_trans_next == CLUSTER_IO_WAITING) {
+ KERNEL_DEBUG((FSDBG_CODE(DBG_FSRW, 20)) | DBG_FUNC_END,
+ cbp_head, cbp, cbp->b_bcount, cbp->b_flags, 0);
+
+ lck_mtx_unlock(cl_transaction_mtxp);
+ wakeup(cbp);
return 0;
}
+
if (cbp->b_flags & B_EOT)
transaction_complete = TRUE;
}
lck_mtx_unlock(cl_transaction_mtxp);
- if (need_wakeup == TRUE)
- wakeup(bp);
-
if (transaction_complete == FALSE) {
KERNEL_DEBUG((FSDBG_CODE(DBG_FSRW, 20)) | DBG_FUNC_END,
cbp_head, 0, 0, 0, 0);
buf_t cbp;
if (async) {
- /*
- * async callback completion will not normally
- * generate a wakeup upon I/O completion...
- * by setting B_TWANTED, we will force a wakeup
- * to occur as any outstanding I/Os complete...
- * I/Os already completed will have B_TDONE already
- * set and we won't cause us to block
- * note that we're actually waiting for the bp to have
- * completed the callback function... only then
- * can we safely take back ownership of the bp
+ /*
+ * Async callback completion will not normally generate a
+ * wakeup upon I/O completion. To get woken up, we set
+ * b_trans_next (which is safe for us to modify) on the last
+ * buffer to CLUSTER_IO_WAITING so that cluster_iodone knows
+ * to wake us up when all buffers as part of this transaction
+ * are completed. This is done under the umbrella of
+ * cl_transaction_mtxp which is also taken in cluster_iodone.
*/
+ bool done = true;
+ buf_t last = NULL;
+
lck_mtx_lock_spin(cl_transaction_mtxp);
- for (cbp = cbp_head; cbp; cbp = cbp->b_trans_next)
- cbp->b_flags |= B_TWANTED;
+ for (cbp = cbp_head; cbp; last = cbp, cbp = cbp->b_trans_next) {
+ if (!ISSET(cbp->b_flags, B_TDONE))
+ done = false;
+ }
- lck_mtx_unlock(cl_transaction_mtxp);
- }
- for (cbp = cbp_head; cbp; cbp = cbp->b_trans_next) {
+ if (!done) {
+ last->b_trans_next = CLUSTER_IO_WAITING;
- if (async) {
- while (!ISSET(cbp->b_flags, B_TDONE)) {
+ DTRACE_IO1(wait__start, buf_t, last);
+ do {
+ msleep(last, cl_transaction_mtxp, PSPIN | (PRIBIO+1), "cluster_wait_IO", NULL);
- lck_mtx_lock_spin(cl_transaction_mtxp);
+ /*
+ * We should only have been woken up if all the
+ * buffers are completed, but just in case...
+ */
+ done = true;
+ for (cbp = cbp_head; cbp != CLUSTER_IO_WAITING; cbp = cbp->b_trans_next) {
+ if (!ISSET(cbp->b_flags, B_TDONE)) {
+ done = false;
+ break;
+ }
+ }
+ } while (!done);
+ DTRACE_IO1(wait__done, buf_t, last);
- if (!ISSET(cbp->b_flags, B_TDONE)) {
- DTRACE_IO1(wait__start, buf_t, cbp);
- (void) msleep(cbp, cl_transaction_mtxp, PDROP | (PRIBIO+1), "cluster_wait_IO", NULL);
- DTRACE_IO1(wait__done, buf_t, cbp);
- } else
- lck_mtx_unlock(cl_transaction_mtxp);
- }
- } else
- buf_biowait(cbp);
+ last->b_trans_next = NULL;
+ }
+
+ lck_mtx_unlock(cl_transaction_mtxp);
+ } else { // !async
+ for (cbp = cbp_head; cbp; cbp = cbp->b_trans_next)
+ buf_biowait(cbp);
}
}
u_int max_cluster_size;
u_int scale;
- max_cluster_size = MAX_CLUSTER_SIZE(vp);
+ if (vp->v_mount->mnt_minsaturationbytecount) {
+ max_cluster_size = vp->v_mount->mnt_minsaturationbytecount;
+ scale = 1;
+ } else {
+ max_cluster_size = MAX_CLUSTER_SIZE(vp);
+
+ if (disk_conditioner_mount_is_ssd(vp->v_mount))
+ scale = WRITE_THROTTLE_SSD;
+ else
+ scale = WRITE_THROTTLE;
+ }
if (max_iosize > max_cluster_size)
max_cluster = max_cluster_size;
else
if (size < max_cluster)
max_cluster = size;
- if ((vp->v_mount->mnt_kern_flag & MNTK_SSD) && !ignore_is_ssd)
- scale = WRITE_THROTTLE_SSD;
- else
- scale = WRITE_THROTTLE;
-
if (flags & CL_CLOSE)
scale += MAX_CLUSTERS;
-
+
async_throttle = min(IO_SCALE(vp, VNODE_ASYNC_THROTTLE), ((scale * max_cluster_size) / max_cluster) - 1);
}
}
* Create a UPL to lock the pages in the cache whilst the
* write is in progress.
*/
- ubc_create_upl(vp, f_offset, non_rounded_size, &cached_upl,
- NULL, UPL_SET_LITE);
+ ubc_create_upl_kernel(vp, f_offset, non_rounded_size, &cached_upl,
+ NULL, UPL_SET_LITE, VM_KERN_MEMORY_FILE);
/*
* Attach this UPL to the other UPL so that we can find it
pageout_flags |= UPL_NOCOMMIT;
if (cbp_head) {
- buf_t last_cbp;
+ buf_t prev_cbp;
+ int bytes_in_last_page;
/*
* first we have to wait for the the current outstanding I/Os
* to complete... EOT hasn't been set yet on this transaction
- * so the pages won't be released just because all of the current
- * I/O linked to this transaction has completed...
+ * so the pages won't be released
*/
cluster_wait_IO(cbp_head, (flags & CL_ASYNC));
- /*
- * we've got a transcation that
- * includes the page we're about to push out through vnode_pageout...
- * find the last bp in the list which will be the one that
- * includes the head of this page and round it's iosize down
- * to a page boundary...
- */
- for (last_cbp = cbp = cbp_head; cbp->b_trans_next; cbp = cbp->b_trans_next)
- last_cbp = cbp;
-
- cbp->b_bcount &= ~PAGE_MASK;
-
- if (cbp->b_bcount == 0) {
- /*
- * this buf no longer has any I/O associated with it
+ bytes_in_last_page = cbp_head->b_uploffset & PAGE_MASK;
+ for (cbp = cbp_head; cbp; cbp = cbp->b_trans_next)
+ bytes_in_last_page += cbp->b_bcount;
+ bytes_in_last_page &= PAGE_MASK;
+
+ while (bytes_in_last_page) {
+ /*
+ * we've got a transcation that
+ * includes the page we're about to push out through vnode_pageout...
+ * find the bp's in the list which intersect this page and either
+ * remove them entirely from the transaction (there could be multiple bp's), or
+ * round it's iosize down to the page boundary (there can only be one)...
+ *
+ * find the last bp in the list and act on it
*/
- free_io_buf(cbp);
+ for (prev_cbp = cbp = cbp_head; cbp->b_trans_next; cbp = cbp->b_trans_next)
+ prev_cbp = cbp;
- if (cbp == cbp_head) {
- /*
- * the buf we just freed was the only buf in
- * this transaction... so there's no I/O to do
+ if (bytes_in_last_page >= cbp->b_bcount) {
+ /*
+ * this buf no longer has any I/O associated with it
*/
- cbp_head = NULL;
+ bytes_in_last_page -= cbp->b_bcount;
+ cbp->b_bcount = 0;
+
+ free_io_buf(cbp);
+
+ if (cbp == cbp_head) {
+ assert(bytes_in_last_page == 0);
+ /*
+ * the buf we just freed was the only buf in
+ * this transaction... so there's no I/O to do
+ */
+ cbp_head = NULL;
+ cbp_tail = NULL;
+ } else {
+ /*
+ * remove the buf we just freed from
+ * the transaction list
+ */
+ prev_cbp->b_trans_next = NULL;
+ cbp_tail = prev_cbp;
+ }
} else {
- /*
- * remove the buf we just freed from
- * the transaction list
+ /*
+ * this is the last bp that has I/O
+ * intersecting the page of interest
+ * only some of the I/O is in the intersection
+ * so clip the size but keep it in the transaction list
*/
- last_cbp->b_trans_next = NULL;
- cbp_tail = last_cbp;
+ cbp->b_bcount -= bytes_in_last_page;
+ cbp_tail = cbp;
+ bytes_in_last_page = 0;
}
}
if (cbp_head) {
return;
}
- max_prefetch = MAX_PREFETCH(vp, cluster_max_io_size(vp->v_mount, CL_READ), (vp->v_mount->mnt_kern_flag & MNTK_SSD));
+ max_prefetch = MAX_PREFETCH(vp, cluster_max_io_size(vp->v_mount, CL_READ), disk_conditioner_mount_is_ssd(vp->v_mount));
if (max_prefetch > speculative_prefetch_max)
max_prefetch = speculative_prefetch_max;
u_int32_t max_io_size;
u_int32_t max_upl_size;
u_int32_t max_vector_size;
+ u_int32_t bytes_outstanding_limit;
boolean_t io_throttled = FALSE;
u_int32_t vector_upl_iosize = 0;
goto wait_for_dwrites;
}
- task_update_logical_writes(current_task(), (io_req_size & ~PAGE_MASK), TASK_WRITE_IMMEDIATE);
+ task_update_logical_writes(current_task(), (io_req_size & ~PAGE_MASK), TASK_WRITE_IMMEDIATE, vp);
while (io_req_size >= PAGE_SIZE && uio->uio_offset < newEOF && retval == 0) {
int throttle_type;
pages_in_pl = 0;
upl_size = upl_needed_size;
upl_flags = UPL_FILE_IO | UPL_COPYOUT_FROM | UPL_NO_SYNC |
- UPL_CLEAN_IN_PLACE | UPL_SET_INTERNAL | UPL_SET_LITE | UPL_SET_IO_WIRE
- | UPL_MEMORY_TAG_MAKE(VM_KERN_MEMORY_FILE);
+ UPL_CLEAN_IN_PLACE | UPL_SET_INTERNAL | UPL_SET_LITE | UPL_SET_IO_WIRE;
kret = vm_map_get_upl(map,
(vm_map_offset_t)(iov_base & ~((user_addr_t)PAGE_MASK)),
NULL,
&pages_in_pl,
&upl_flags,
+ VM_KERN_MEMORY_FILE,
force_data_sync);
if (kret != KERN_SUCCESS) {
* if there are already too many outstanding writes
* wait until some complete before issuing the next
*/
- cluster_iostate_wait(&iostate, max_upl_size * IO_SCALE(vp, 2), "cluster_write_direct");
+ if (vp->v_mount->mnt_minsaturationbytecount)
+ bytes_outstanding_limit = vp->v_mount->mnt_minsaturationbytecount;
+ else
+ bytes_outstanding_limit = max_upl_size * IO_SCALE(vp, 2);
+
+ cluster_iostate_wait(&iostate, bytes_outstanding_limit, "cluster_write_direct");
if (iostate.io_error) {
/*
pages_in_pl = 0;
upl_size = upl_needed_size;
upl_flags = UPL_FILE_IO | UPL_COPYOUT_FROM | UPL_NO_SYNC |
- UPL_CLEAN_IN_PLACE | UPL_SET_INTERNAL | UPL_SET_LITE | UPL_SET_IO_WIRE
- | UPL_MEMORY_TAG_MAKE(VM_KERN_MEMORY_FILE);
+ UPL_CLEAN_IN_PLACE | UPL_SET_INTERNAL | UPL_SET_LITE | UPL_SET_IO_WIRE;
vm_map_t map = UIO_SEG_IS_USER_SPACE(uio->uio_segflg) ? current_map() : kernel_map;
kret = vm_map_get_upl(map,
(vm_map_offset_t)(iov_base & ~((user_addr_t)PAGE_MASK)),
- &upl_size, &upl[cur_upl], NULL, &pages_in_pl, &upl_flags, 0);
+ &upl_size, &upl[cur_upl], NULL, &pages_in_pl, &upl_flags, VM_KERN_MEMORY_FILE, 0);
if (kret != KERN_SUCCESS) {
/*
* The UPL_WILL_MODIFY flag lets the UPL subsystem know
* that we intend to modify these pages.
*/
- kret = ubc_create_upl(vp,
+ kret = ubc_create_upl_kernel(vp,
upl_f_offset,
upl_size,
&upl,
&pl,
- UPL_SET_LITE | (( uio!=NULL && (uio->uio_flags & UIO_FLAGS_IS_COMPRESSED_FILE)) ? 0 : UPL_WILL_MODIFY));
+ UPL_SET_LITE | (( uio!=NULL && (uio->uio_flags & UIO_FLAGS_IS_COMPRESSED_FILE)) ? 0 : UPL_WILL_MODIFY),
+ VM_KERN_MEMORY_FILE);
if (kret != KERN_SUCCESS)
panic("cluster_write_copy: failed to get pagelist");
wbp->cl_seq_written >= (MAX_CLUSTERS * (max_cluster_pgcount * PAGE_SIZE))) {
uint32_t n;
- if (vp->v_mount->mnt_kern_flag & MNTK_SSD)
- n = WRITE_BEHIND_SSD;
- else
- n = WRITE_BEHIND;
+ if (vp->v_mount->mnt_minsaturationbytecount) {
+ n = vp->v_mount->mnt_minsaturationbytecount / MAX_CLUSTER_SIZE(vp);
+
+ if (n > MAX_CLUSTERS)
+ n = MAX_CLUSTERS;
+ } else
+ n = 0;
+ if (n == 0) {
+ if (disk_conditioner_mount_is_ssd(vp->v_mount))
+ n = WRITE_BEHIND_SSD;
+ else
+ n = WRITE_BEHIND;
+ }
while (n--)
- cluster_try_push(wbp, vp, newEOF, 0, 0, callback, callback_arg);
+ cluster_try_push(wbp, vp, newEOF, 0, 0, callback, callback_arg, NULL);
}
if (wbp->cl_number < MAX_CLUSTERS) {
/*
*/
if (!((unsigned int)vfs_flags(vp->v_mount) & MNT_DEFWRITE)) {
- ret_cluster_try_push = cluster_try_push(wbp, vp, newEOF, (flags & IO_NOCACHE) ? 0 : PUSH_DELAY, 0, callback, callback_arg);
+ ret_cluster_try_push = cluster_try_push(wbp, vp, newEOF, (flags & IO_NOCACHE) ? 0 : PUSH_DELAY, 0, callback, callback_arg, NULL);
}
/*
if (flags & IO_SKIP_ENCRYPTION)
flags |= IO_ENCRYPTED;
- /*
- * If we're doing an encrypted IO, then first check to see
- * if the IO requested was page aligned. If not, then bail
- * out immediately.
- */
- if (flags & IO_ENCRYPTED) {
- if (read_length & PAGE_MASK) {
- retval = EINVAL;
- return retval;
- }
- }
/*
* do a read through the cache if one of the following is true....
retval = cluster_io_type(uio, &read_type, &read_length, 0);
}
-
+
while ((cur_resid = uio_resid(uio)) && uio->uio_offset < filesize && retval == 0) {
switch (read_type) {
bflag |= CL_ENCRYPTED;
max_io_size = cluster_max_io_size(vp->v_mount, CL_READ);
- max_prefetch = MAX_PREFETCH(vp, max_io_size, (vp->v_mount->mnt_kern_flag & MNTK_SSD));
+ max_prefetch = MAX_PREFETCH(vp, max_io_size, disk_conditioner_mount_is_ssd(vp->v_mount));
max_rd_size = max_prefetch;
last_request_offset = uio->uio_offset + io_req_size;
KERNEL_DEBUG((FSDBG_CODE(DBG_FSRW, 33)) | DBG_FUNC_START,
upl, (int)upl_f_offset, upl_size, start_offset, 0);
- kret = ubc_create_upl(vp,
+ kret = ubc_create_upl_kernel(vp,
upl_f_offset,
upl_size,
&upl,
&pl,
- UPL_FILE_IO | UPL_SET_LITE);
+ UPL_FILE_IO | UPL_SET_LITE,
+ VM_KERN_MEMORY_FILE);
if (kret != KERN_SUCCESS)
panic("cluster_read_copy: failed to get pagelist");
io_req_size = *read_length;
iov_base = uio_curriovbase(uio);
- max_io_size = filesize - uio->uio_offset;
-
- if ((off_t)io_req_size > max_io_size)
- io_req_size = max_io_size;
-
offset_in_file = (u_int32_t)uio->uio_offset & (devblocksize - 1);
offset_in_iovbase = (u_int32_t)iov_base & mem_alignment_mask;
misaligned = 1;
}
+ max_io_size = filesize - uio->uio_offset;
+
/*
* The user must request IO in aligned chunks. If the
* offset into the file is bad, or the userland pointer
* is non-aligned, then we cannot service the encrypted IO request.
*/
- if ((flags & IO_ENCRYPTED) && (misaligned)) {
- retval = EINVAL;
+ if (flags & IO_ENCRYPTED) {
+ if (misaligned || (io_req_size & (devblocksize - 1)))
+ retval = EINVAL;
+
+ max_io_size = roundup(max_io_size, devblocksize);
}
+ if ((off_t)io_req_size > max_io_size)
+ io_req_size = max_io_size;
+
/*
* When we get to this point, we know...
* -- the offset into the file is on a devblocksize boundary
* (which overlaps the end of the direct read) in order to
* get at the overhang bytes
*/
- if (io_size & (devblocksize - 1)) {
- if (flags & IO_ENCRYPTED) {
- /*
- * Normally, we'd round down to the previous page boundary to
- * let the UBC manage the zero-filling of the file past the EOF.
- * But if we're doing encrypted IO, we can't let any of
- * the data hit the UBC. This means we have to do the full
- * IO to the upper block boundary of the device block that
- * contains the EOF. The user will be responsible for not
- * interpreting data PAST the EOF in its buffer.
- *
- * So just bump the IO back up to a multiple of devblocksize
- */
- io_size = ((io_size + devblocksize) & ~(devblocksize - 1));
- io_min = io_size;
- }
- else {
- /*
- * Clip the request to the previous page size boundary
- * since request does NOT end on a device block boundary
- */
- io_size &= ~PAGE_MASK;
- io_min = PAGE_SIZE;
- }
-
+ if (io_size & (devblocksize - 1)) {
+ assert(!(flags & IO_ENCRYPTED));
+ /*
+ * Clip the request to the previous page size boundary
+ * since request does NOT end on a device block boundary
+ */
+ io_size &= ~PAGE_MASK;
+ io_min = PAGE_SIZE;
}
if (retval || io_size < io_min) {
/*
for (force_data_sync = 0; force_data_sync < 3; force_data_sync++) {
pages_in_pl = 0;
upl_size = upl_needed_size;
- upl_flags = UPL_FILE_IO | UPL_NO_SYNC | UPL_SET_INTERNAL | UPL_SET_LITE | UPL_SET_IO_WIRE
- | UPL_MEMORY_TAG_MAKE(VM_KERN_MEMORY_FILE);
+ upl_flags = UPL_FILE_IO | UPL_NO_SYNC | UPL_SET_INTERNAL | UPL_SET_LITE | UPL_SET_IO_WIRE;
if (no_zero_fill)
upl_flags |= UPL_NOZEROFILL;
if (force_data_sync)
kret = vm_map_create_upl(map,
(vm_map_offset_t)(iov_base & ~((user_addr_t)PAGE_MASK)),
- &upl_size, &upl, NULL, &pages_in_pl, &upl_flags);
+ &upl_size, &upl, NULL, &pages_in_pl, &upl_flags, VM_KERN_MEMORY_FILE);
if (kret != KERN_SUCCESS) {
KERNEL_DEBUG((FSDBG_CODE(DBG_FSRW, 72)) | DBG_FUNC_END,
else {
uio_update(uio, (user_size_t)io_size);
}
- /*
- * Under normal circumstances, the io_size should not be
- * bigger than the io_req_size, but we may have had to round up
- * to the end of the page in the encrypted IO case. In that case only,
- * ensure that we only decrement io_req_size to 0.
- */
- if ((flags & IO_ENCRYPTED) && (io_size > io_req_size)) {
- io_req_size = 0;
- }
- else {
- io_req_size -= io_size;
- }
+
+ io_req_size -= io_size;
KERNEL_DEBUG((FSDBG_CODE(DBG_FSRW, 73)) | DBG_FUNC_END,
upl, (int)uio->uio_offset, io_req_size, retval, 0);
pages_in_pl = 0;
upl_size = upl_needed_size;
- upl_flags = UPL_FILE_IO | UPL_NO_SYNC | UPL_CLEAN_IN_PLACE | UPL_SET_INTERNAL | UPL_SET_LITE | UPL_SET_IO_WIRE
- | UPL_MEMORY_TAG_MAKE(VM_KERN_MEMORY_FILE);
+ upl_flags = UPL_FILE_IO | UPL_NO_SYNC | UPL_CLEAN_IN_PLACE | UPL_SET_INTERNAL | UPL_SET_LITE | UPL_SET_IO_WIRE;
KERNEL_DEBUG((FSDBG_CODE(DBG_FSRW, 92)) | DBG_FUNC_START,
vm_map_t map = UIO_SEG_IS_USER_SPACE(uio->uio_segflg) ? current_map() : kernel_map;
kret = vm_map_get_upl(map,
(vm_map_offset_t)(iov_base & ~((user_addr_t)PAGE_MASK)),
- &upl_size, &upl[cur_upl], NULL, &pages_in_pl, &upl_flags, 0);
+ &upl_size, &upl[cur_upl], NULL, &pages_in_pl, &upl_flags, VM_KERN_MEMORY_FILE, 0);
KERNEL_DEBUG((FSDBG_CODE(DBG_FSRW, 92)) | DBG_FUNC_END,
(int)upl_offset, upl_size, io_size, kret, 0);
else
upl_size = (u_int32_t)iov_len;
- upl_flags = UPL_QUERY_OBJECT_TYPE | UPL_MEMORY_TAG_MAKE(VM_KERN_MEMORY_FILE);
+ upl_flags = UPL_QUERY_OBJECT_TYPE;
vm_map_t map = UIO_SEG_IS_USER_SPACE(uio->uio_segflg) ? current_map() : kernel_map;
if ((vm_map_get_upl(map,
(vm_map_offset_t)(iov_base & ~((user_addr_t)PAGE_MASK)),
- &upl_size, &upl, NULL, NULL, &upl_flags, 0)) != KERN_SUCCESS) {
+ &upl_size, &upl, NULL, NULL, &upl_flags, VM_KERN_MEMORY_FILE, 0)) != KERN_SUCCESS) {
/*
* the user app must have passed in an invalid address
*/
max_io_size = cluster_max_io_size(vp->v_mount, CL_READ);
- if ((vp->v_mount->mnt_kern_flag & MNTK_SSD) && !ignore_is_ssd) {
+#if CONFIG_EMBEDDED
+ if (max_io_size > speculative_prefetch_max_iosize)
+ max_io_size = speculative_prefetch_max_iosize;
+#else
+ if (disk_conditioner_mount_is_ssd(vp->v_mount)) {
if (max_io_size > speculative_prefetch_max_iosize)
max_io_size = speculative_prefetch_max_iosize;
}
+#endif
KERNEL_DEBUG((FSDBG_CODE(DBG_FSRW, 60)) | DBG_FUNC_START,
(int)f_offset, resid, (int)filesize, 0, 0);
KERNEL_DEBUG((FSDBG_CODE(DBG_FSRW, 61)) | DBG_FUNC_START,
upl, (int)upl_f_offset, upl_size, start_offset, 0);
- kret = ubc_create_upl(vp,
+ kret = ubc_create_upl_kernel(vp,
upl_f_offset,
upl_size,
&upl,
&pl,
- UPL_RET_ONLY_ABSENT | UPL_SET_LITE);
+ UPL_RET_ONLY_ABSENT | UPL_SET_LITE,
+ VM_KERN_MEMORY_FILE);
if (kret != KERN_SUCCESS)
return(retval);
issued_io = 0;
int
cluster_push_ext(vnode_t vp, int flags, int (*callback)(buf_t, void *), void *callback_arg)
+{
+ return cluster_push_err(vp, flags, callback, callback_arg, NULL);
+}
+
+/* write errors via err, but return the number of clusters written */
+int
+cluster_push_err(vnode_t vp, int flags, int (*callback)(buf_t, void *), void *callback_arg, int *err)
{
int retval;
int my_sparse_wait = 0;
struct cl_writebehind *wbp;
+ if (err)
+ *err = 0;
+
if ( !UBCINFOEXISTS(vp)) {
- KERNEL_DEBUG((FSDBG_CODE(DBG_FSRW, 53)) | DBG_FUNC_NONE, vp, flags, 0, -1, 0);
+ KERNEL_DEBUG((FSDBG_CODE(DBG_FSRW, 53)) | DBG_FUNC_NONE, kdebug_vnode(vp), flags, 0, -1, 0);
return (0);
}
/* return if deferred write is set */
return (0);
}
if ((wbp = cluster_get_wbp(vp, CLW_RETURNLOCKED)) == NULL) {
- KERNEL_DEBUG((FSDBG_CODE(DBG_FSRW, 53)) | DBG_FUNC_NONE, vp, flags, 0, -2, 0);
+ KERNEL_DEBUG((FSDBG_CODE(DBG_FSRW, 53)) | DBG_FUNC_NONE, kdebug_vnode(vp), flags, 0, -2, 0);
return (0);
}
if (!ISSET(flags, IO_SYNC) && wbp->cl_number == 0 && wbp->cl_scmap == NULL) {
lck_mtx_unlock(&wbp->cl_lockw);
- KERNEL_DEBUG((FSDBG_CODE(DBG_FSRW, 53)) | DBG_FUNC_NONE, vp, flags, 0, -3, 0);
+ KERNEL_DEBUG((FSDBG_CODE(DBG_FSRW, 53)) | DBG_FUNC_NONE, kdebug_vnode(vp), flags, 0, -3, 0);
return(0);
}
KERNEL_DEBUG((FSDBG_CODE(DBG_FSRW, 53)) | DBG_FUNC_START,
* in the sparse map case
*/
while (wbp->cl_sparse_wait) {
- KERNEL_DEBUG((FSDBG_CODE(DBG_FSRW, 97)) | DBG_FUNC_START, vp, 0, 0, 0, 0);
+ KERNEL_DEBUG((FSDBG_CODE(DBG_FSRW, 97)) | DBG_FUNC_START, kdebug_vnode(vp), 0, 0, 0, 0);
msleep((caddr_t)&wbp->cl_sparse_wait, &wbp->cl_lockw, PRIBIO + 1, "cluster_push_ext", NULL);
- KERNEL_DEBUG((FSDBG_CODE(DBG_FSRW, 97)) | DBG_FUNC_END, vp, 0, 0, 0, 0);
+ KERNEL_DEBUG((FSDBG_CODE(DBG_FSRW, 97)) | DBG_FUNC_END, kdebug_vnode(vp), 0, 0, 0, 0);
}
if (flags & IO_SYNC) {
my_sparse_wait = 1;
* fsync actually get cleaned to the disk before this fsync returns
*/
while (wbp->cl_sparse_pushes) {
- KERNEL_DEBUG((FSDBG_CODE(DBG_FSRW, 98)) | DBG_FUNC_START, vp, 0, 0, 0, 0);
+ KERNEL_DEBUG((FSDBG_CODE(DBG_FSRW, 98)) | DBG_FUNC_START, kdebug_vnode(vp), 0, 0, 0, 0);
msleep((caddr_t)&wbp->cl_sparse_pushes, &wbp->cl_lockw, PRIBIO + 1, "cluster_push_ext", NULL);
- KERNEL_DEBUG((FSDBG_CODE(DBG_FSRW, 98)) | DBG_FUNC_END, vp, 0, 0, 0, 0);
+ KERNEL_DEBUG((FSDBG_CODE(DBG_FSRW, 98)) | DBG_FUNC_END, kdebug_vnode(vp), 0, 0, 0, 0);
}
}
if (wbp->cl_scmap) {
lck_mtx_unlock(&wbp->cl_lockw);
- sparse_cluster_push(&scmap, vp, ubc_getsize(vp), PUSH_ALL, flags, callback, callback_arg);
+ retval = sparse_cluster_push(&scmap, vp, ubc_getsize(vp), PUSH_ALL, flags, callback, callback_arg);
lck_mtx_lock(&wbp->cl_lockw);
if (wbp->cl_sparse_wait && wbp->cl_sparse_pushes == 0)
wakeup((caddr_t)&wbp->cl_sparse_pushes);
} else {
- sparse_cluster_push(&(wbp->cl_scmap), vp, ubc_getsize(vp), PUSH_ALL, flags, callback, callback_arg);
+ retval = sparse_cluster_push(&(wbp->cl_scmap), vp, ubc_getsize(vp), PUSH_ALL, flags, callback, callback_arg);
}
+ if (err)
+ *err = retval;
retval = 1;
- } else {
- retval = cluster_try_push(wbp, vp, ubc_getsize(vp), PUSH_ALL, flags, callback, callback_arg);
+ } else {
+ retval = cluster_try_push(wbp, vp, ubc_getsize(vp), PUSH_ALL, flags, callback, callback_arg, err);
}
lck_mtx_unlock(&wbp->cl_lockw);
static int
-cluster_try_push(struct cl_writebehind *wbp, vnode_t vp, off_t EOF, int push_flag, int io_flags, int (*callback)(buf_t, void *), void *callback_arg)
+cluster_try_push(struct cl_writebehind *wbp, vnode_t vp, off_t EOF, int push_flag, int io_flags, int (*callback)(buf_t, void *), void *callback_arg, int *err)
{
int cl_index;
int cl_index1;
int cl_pushed = 0;
struct cl_wextent l_clusters[MAX_CLUSTERS];
u_int max_cluster_pgcount;
-
+ int error = 0;
max_cluster_pgcount = MAX_CLUSTER_SIZE(vp) / PAGE_SIZE;
/*
cl_len = cl_index;
- if ( (push_flag & PUSH_DELAY) && cl_len == MAX_CLUSTERS ) {
+ /* skip switching to the sparse cluster mechanism if on diskimage */
+ if ( ((push_flag & PUSH_DELAY) && cl_len == MAX_CLUSTERS ) &&
+ !(vp->v_mount->mnt_kern_flag & MNTK_VIRTUALDEV) ) {
int i;
/*
for (cl_index = 0; cl_index < cl_len; cl_index++) {
int flags;
struct cl_extent cl;
+ int retval;
flags = io_flags & (IO_PASSIVE|IO_CLOSE);
cl.b_addr = l_clusters[cl_index].b_addr;
cl.e_addr = l_clusters[cl_index].e_addr;
- cluster_push_now(vp, &cl, EOF, flags, callback, callback_arg);
+ retval = cluster_push_now(vp, &cl, EOF, flags, callback, callback_arg);
+
+ if (error == 0 && retval)
+ error = retval;
l_clusters[cl_index].b_addr = 0;
l_clusters[cl_index].e_addr = 0;
if ( !(push_flag & PUSH_ALL) )
break;
}
+ if (err)
+ *err = error;
+
dont_try:
if (cl_len > cl_pushed) {
/*
else
upl_flags = UPL_COPYOUT_FROM | UPL_RET_ONLY_DIRTY | UPL_SET_LITE;
- kret = ubc_create_upl(vp,
+ kret = ubc_create_upl_kernel(vp,
upl_f_offset,
upl_size,
&upl,
&pl,
- upl_flags);
+ upl_flags,
+ VM_KERN_MEMORY_FILE);
if (kret != KERN_SUCCESS)
panic("cluster_push: failed to get pagelist");
{
int cl_index;
- KERNEL_DEBUG((FSDBG_CODE(DBG_FSRW, 78)) | DBG_FUNC_START, vp, wbp->cl_scmap, 0, 0, 0);
+ KERNEL_DEBUG((FSDBG_CODE(DBG_FSRW, 78)) | DBG_FUNC_START, kdebug_vnode(vp), wbp->cl_scmap, 0, 0, 0);
for (cl_index = 0; cl_index < wbp->cl_number; cl_index++) {
int flags;
}
wbp->cl_number = 0;
- KERNEL_DEBUG((FSDBG_CODE(DBG_FSRW, 78)) | DBG_FUNC_END, vp, wbp->cl_scmap, 0, 0, 0);
+ KERNEL_DEBUG((FSDBG_CODE(DBG_FSRW, 78)) | DBG_FUNC_END, kdebug_vnode(vp), wbp->cl_scmap, 0, 0, 0);
}
* still associated with the write-behind context... however, if the scmap has been disassociated
* from the write-behind context (the cluster_push case), the wb lock is not held
*/
-static void
+static int
sparse_cluster_push(void **scmap, vnode_t vp, off_t EOF, int push_flag, int io_flags, int (*callback)(buf_t, void *), void *callback_arg)
{
struct cl_extent cl;
off_t offset;
u_int length;
+ int error = 0;
- KERNEL_DEBUG((FSDBG_CODE(DBG_FSRW, 79)) | DBG_FUNC_START, vp, (*scmap), 0, push_flag, 0);
+ KERNEL_DEBUG((FSDBG_CODE(DBG_FSRW, 79)) | DBG_FUNC_START, kdebug_vnode(vp), (*scmap), 0, push_flag, 0);
if (push_flag & PUSH_ALL)
vfs_drt_control(scmap, 1);
for (;;) {
+ int retval;
if (vfs_drt_get_cluster(scmap, &offset, &length) != KERN_SUCCESS)
break;
cl.b_addr = (daddr64_t)(offset / PAGE_SIZE_64);
cl.e_addr = (daddr64_t)((offset + length) / PAGE_SIZE_64);
- cluster_push_now(vp, &cl, EOF, io_flags & (IO_PASSIVE|IO_CLOSE), callback, callback_arg);
+ retval = cluster_push_now(vp, &cl, EOF, io_flags & (IO_PASSIVE|IO_CLOSE), callback, callback_arg);
+ if (error == 0 && retval)
+ error = retval;
if ( !(push_flag & PUSH_ALL) )
break;
}
- KERNEL_DEBUG((FSDBG_CODE(DBG_FSRW, 79)) | DBG_FUNC_END, vp, (*scmap), 0, 0, 0);
+ KERNEL_DEBUG((FSDBG_CODE(DBG_FSRW, 79)) | DBG_FUNC_END, kdebug_vnode(vp), (*scmap), 0, 0, 0);
+
+ return error;
}
offset += (new_dirty * PAGE_SIZE_64);
length -= (new_dirty * PAGE_SIZE);
}
- KERNEL_DEBUG((FSDBG_CODE(DBG_FSRW, 80)) | DBG_FUNC_END, vp, (*scmap), 0, 0, 0);
+ KERNEL_DEBUG((FSDBG_CODE(DBG_FSRW, 80)) | DBG_FUNC_END, kdebug_vnode(vp), (*scmap), 0, 0, 0);
}
*/
upl_flags |= UPL_FILE_IO;
}
- kret = ubc_create_upl(vp,
+ kret = ubc_create_upl_kernel(vp,
uio->uio_offset & ~PAGE_MASK_64,
PAGE_SIZE,
&upl,
&pl,
- upl_flags);
+ upl_flags,
+ VM_KERN_MEMORY_FILE);
if (kret != KERN_SUCCESS)
return(EINVAL);
return (error);
}
-
-
int
cluster_copy_upl_data(struct uio *uio, upl_t upl, int upl_offset, int *io_resid)
{
uio->uio_segflg = segflg;
- task_update_logical_writes(current_task(), (dirty_count * PAGE_SIZE), TASK_WRITE_DEFERRED);
+ task_update_logical_writes(current_task(), (dirty_count * PAGE_SIZE), TASK_WRITE_DEFERRED, upl_lookup_vnode(upl));
KERNEL_DEBUG((FSDBG_CODE(DBG_FSRW, 34)) | DBG_FUNC_END,
(int)uio->uio_offset, xsize, retval, segflg, 0);
-
+
return (retval);
}
projects / apple / xnu.git / blobdiff