]> git.saurik.com Git - apple/xnu.git/blobdiff - bsd/vm/vnode_pager.c
xnu-7195.101.1.tar.gz
[apple/xnu.git] / bsd / vm / vnode_pager.c
index fdbccff33a8b28c8099f0509d1c37928fb484a32..fc3426fd5976c661098acdda09a13b303fc6f683 100644 (file)
@@ -1,8 +1,8 @@
 /*
- * Copyright (c) 2000-2006 Apple Computer, Inc. All rights reserved.
+ * Copyright (c) 2000-2020 Apple Inc. All rights reserved.
  *
  * @APPLE_OSREFERENCE_LICENSE_HEADER_START@
- * 
+ *
  * This file contains Original Code and/or Modifications of Original Code
  * as defined in and that are subject to the Apple Public Source License
  * Version 2.0 (the 'License'). You may not use this file except in
  * unlawful or unlicensed copies of an Apple operating system, or to
  * circumvent, violate, or enable the circumvention or violation of, any
  * terms of an Apple operating system software license agreement.
- * 
+ *
  * Please obtain a copy of the License at
  * http://www.opensource.apple.com/apsl/ and read it before using this file.
- * 
+ *
  * The Original Code and all software distributed under the License are
  * distributed on an 'AS IS' basis, WITHOUT WARRANTY OF ANY KIND, EITHER
  * EXPRESS OR IMPLIED, AND APPLE HEREBY DISCLAIMS ALL SUCH WARRANTIES,
  * FITNESS FOR A PARTICULAR PURPOSE, QUIET ENJOYMENT OR NON-INFRINGEMENT.
  * Please see the License for the specific language governing rights and
  * limitations under the License.
- * 
+ *
  * @APPLE_OSREFERENCE_LICENSE_HEADER_END@
  */
-/* 
+/*
  * Mach Operating System
  * Copyright (c) 1987 Carnegie-Mellon University
  * All rights reserved.  The CMU software License Agreement specifies
 #include <sys/uio.h>
 #include <sys/vnode_internal.h>
 #include <sys/namei.h>
-#include <sys/mount_internal.h>        /* needs internal due to fhandle_t */
+#include <sys/mount_internal.h> /* needs internal due to fhandle_t */
 #include <sys/ubc_internal.h>
 #include <sys/lock.h>
-#include <sys/disk.h>          /* For DKIOC calls */
+#include <sys/disk.h>           /* For DKIOC calls */
 
 #include <mach/mach_types.h>
 #include <mach/memory_object_types.h>
-#include <mach/memory_object_control.h>
 #include <mach/vm_map.h>
 #include <mach/mach_vm.h>
 #include <mach/upl.h>
@@ -65,7 +64,6 @@
 #include <vm/vm_map.h>
 #include <vm/vm_kern.h>
 #include <kern/zalloc.h>
-#include <kern/kalloc.h>
 #include <libkern/libkern.h>
 
 #include <vm/vnode_pager.h>
@@ -73,6 +71,7 @@
 
 #include <kern/assert.h>
 #include <sys/kdebug.h>
+#include <nfs/nfs_conf.h>
 #include <nfs/rpcv2.h>
 #include <nfs/nfsproto.h>
 #include <nfs/nfs.h>
 #include <vfs/vfs_disk_conditioner.h>
 
 void
-vnode_pager_throttle()
+vnode_pager_throttle(void)
 {
        struct uthread *ut;
 
        ut = get_bsdthread_info(current_thread());
 
-       if (ut->uu_lowpri_window)
+       if (ut->uu_lowpri_window) {
                throttle_lowpri_io(1);
+       }
 }
 
 boolean_t
@@ -104,23 +104,24 @@ vnode_pager_issue_reprioritize_io(struct vnode *devvp, uint64_t blkno, uint32_t
 {
        u_int32_t blocksize = 0;
        dk_extent_t extent;
-        dk_set_tier_t set_tier;
+       dk_set_tier_t set_tier;
        int error = 0;
 
        error = VNOP_IOCTL(devvp, DKIOCGETBLOCKSIZE, (caddr_t)&blocksize, 0, vfs_context_kernel());
-       if (error)
+       if (error) {
                return;
+       }
 
        memset(&extent, 0, sizeof(dk_extent_t));
        memset(&set_tier, 0, sizeof(dk_set_tier_t));
-       
+
        extent.offset = blkno * (u_int64_t) blocksize;
        extent.length = len;
 
-       set_tier.extents = &extent; 
+       set_tier.extents = &extent;
        set_tier.extentsCount = 1;
-       set_tier.tier = priority;
-               
+       set_tier.tier = (uint8_t)priority;
+
        error = VNOP_IOCTL(devvp, DKIOCSETTIER, (caddr_t)&set_tier, 0, vfs_context_kernel());
        return;
 }
@@ -128,25 +129,26 @@ vnode_pager_issue_reprioritize_io(struct vnode *devvp, uint64_t blkno, uint32_t
 
 void
 vnode_pager_was_dirtied(
-       struct vnode            *vp,
-       vm_object_offset_t      s_offset,
-       vm_object_offset_t      e_offset)
+       struct vnode            *vp,
+       vm_object_offset_t      s_offset,
+       vm_object_offset_t      e_offset)
 {
-        cluster_update_state(vp, s_offset, e_offset, TRUE);
+       cluster_update_state(vp, s_offset, e_offset, TRUE);
 }
 
 uint32_t
 vnode_pager_isinuse(struct vnode *vp)
 {
-       if (vp->v_usecount > vp->v_kusecount)
-               return (1);
-       return (0);
+       if (vp->v_usecount > vp->v_kusecount) {
+               return 1;
+       }
+       return 0;
 }
 
 uint32_t
 vnode_pager_return_throttle_io_limit(struct vnode *vp, uint32_t *limit)
 {
-       return(cluster_throttle_io_limit(vp, limit));
+       return cluster_throttle_io_limit(vp, limit);
 }
 
 vm_object_offset_t
@@ -159,19 +161,19 @@ extern int safe_getpath(struct vnode *dvp, char *leafname, char *path, int _len,
 
 kern_return_t
 vnode_pager_get_name(
-       struct vnode    *vp,
-       char            *pathname,
-       vm_size_t       pathname_len,
-       char            *filename,
-       vm_size_t       filename_len,
-       boolean_t       *truncated_path_p)
+       struct vnode    *vp,
+       char            *pathname,
+       vm_size_t       pathname_len,
+       char            *filename,
+       vm_size_t       filename_len,
+       boolean_t       *truncated_path_p)
 {
        *truncated_path_p = FALSE;
        if (pathname != NULL) {
                /* get the path name */
                safe_getpath(vp, NULL,
-                            pathname, (int) pathname_len,
-                            truncated_path_p);
+                   pathname, (int) pathname_len,
+                   truncated_path_p);
        }
        if ((pathname == NULL || *truncated_path_p) &&
            filename != NULL) {
@@ -187,9 +189,9 @@ vnode_pager_get_name(
 
 kern_return_t
 vnode_pager_get_mtime(
-       struct vnode    *vp,
-       struct timespec *current_mtime,
-       struct timespec *cs_mtime)
+       struct vnode    *vp,
+       struct timespec *current_mtime,
+       struct timespec *cs_mtime)
 {
        vnode_mtime(vp, current_mtime, vfs_context_current());
        if (cs_mtime != NULL) {
@@ -200,14 +202,14 @@ vnode_pager_get_mtime(
 
 kern_return_t
 vnode_pager_get_cs_blobs(
-       struct vnode    *vp,
-       void            **blobs)
+       struct vnode    *vp,
+       void            **blobs)
 {
        *blobs = ubc_get_cs_blobs(vp);
        return KERN_SUCCESS;
 }
 
-/* 
+/*
  * vnode_trim:
  * Used to call the DKIOCUNMAP ioctl on the underlying disk device for the specified vnode.
  * Trims the region at offset bytes into the file, for length bytes.
@@ -217,15 +219,16 @@ vnode_pager_get_cs_blobs(
  * This function is non-idempotent in error cases;  We cannot un-discard the blocks if only some of them
  * are successfully discarded.
  */
-u_int32_t vnode_trim (
-               struct vnode *vp,
-               off_t offset,
-               size_t length)
+u_int32_t
+vnode_trim(
+       struct vnode *vp,
+       off_t offset,
+       size_t length)
 {
-       daddr64_t io_blockno;    /* Block number corresponding to the start of the extent */
-       size_t io_bytecount;    /* Number of bytes in current extent for the specified range */
+       daddr64_t io_blockno;    /* Block number corresponding to the start of the extent */
+       size_t io_bytecount;    /* Number of bytes in current extent for the specified range */
        size_t trimmed = 0;
-       off_t current_offset = offset; 
+       off_t current_offset = offset;
        size_t remaining_length = length;
        int error = 0;
        u_int32_t blocksize = 0;
@@ -243,7 +246,7 @@ u_int32_t vnode_trim (
                goto trim_exit;
        }
 
-       /* 
+       /*
         * We may not get the entire range from offset -> offset+length in a single
         * extent from the blockmap call.  Keep looping/going until we are sure we've hit
         * the whole range or if we encounter an error.
@@ -251,22 +254,22 @@ u_int32_t vnode_trim (
        while (trimmed < length) {
                /*
                 * VNOP_BLOCKMAP will tell us the logical to physical block number mapping for the
-                * specified offset.  It returns blocks in contiguous chunks, so if the logical range is 
+                * specified offset.  It returns blocks in contiguous chunks, so if the logical range is
                 * broken into multiple extents, it must be called multiple times, increasing the offset
                 * in each call to ensure that the entire range is covered.
                 */
-               error = VNOP_BLOCKMAP (vp, current_offset, remaining_length, 
-                               &io_blockno, &io_bytecount, NULL, VNODE_READ | VNODE_BLOCKMAP_NO_TRACK, NULL);
+               error = VNOP_BLOCKMAP(vp, current_offset, remaining_length,
+                   &io_blockno, &io_bytecount, NULL, VNODE_READ | VNODE_BLOCKMAP_NO_TRACK, NULL);
 
                if (error) {
                        goto trim_exit;
                }
-               /* 
+               /*
                 * We have a contiguous run.  Prepare & issue the ioctl for the device.
                 * the DKIOCUNMAP ioctl takes offset in bytes from the start of the device.
                 */
-               memset (&extent, 0, sizeof(dk_extent_t));
-               memset (&unmap, 0, sizeof(dk_unmap_t));
+               memset(&extent, 0, sizeof(dk_extent_t));
+               memset(&unmap, 0, sizeof(dk_unmap_t));
                extent.offset = (uint64_t) io_blockno * (u_int64_t) blocksize;
                extent.length = io_bytecount;
                unmap.extents = &extent;
@@ -283,33 +286,52 @@ u_int32_t vnode_trim (
 trim_exit:
 
        return error;
-
 }
 
 pager_return_t
 vnode_pageout(struct vnode *vp,
-       upl_t                   upl,
-       upl_offset_t            upl_offset,
-       vm_object_offset_t      f_offset,
-       upl_size_t              size,
-       int                     flags,
-       int                     *errorp)
+    upl_t                   upl,
+    upl_offset_t            upl_offset,
+    vm_object_offset_t      f_offset,
+    upl_size_t              size,
+    int                     flags,
+    int                     *errorp)
 {
-       int             result = PAGER_SUCCESS;
-       int             error = 0;
-       int             error_ret = 0;
+       int             result = PAGER_SUCCESS;
+       int             error = 0;
+       int             error_ret = 0;
        daddr64_t blkno;
        int isize;
        int pg_index;
        int base_index;
        upl_offset_t offset;
        upl_page_info_t *pl;
-       vfs_context_t ctx = vfs_context_current();      /* pager context */
+       vfs_context_t ctx = vfs_context_current();      /* pager context */
 
        isize = (int)size;
 
+       /*
+        * This call is non-blocking and does not ever fail but it can
+        * only be made when there is other explicit synchronization
+        * with reclaiming of the vnode which, in this path, is provided
+        * by the paging in progress counter.
+        *
+        * In addition, this may also be entered via explicit ubc_msync
+        * calls or vm_swapfile_io where the existing iocount provides
+        * the necessary synchronization. Ideally we would not take an
+        * additional iocount here in the cases where an explcit iocount
+        * has already been taken but this call doesn't cause a deadlock
+        * as other forms of vnode_get* might if this thread has already
+        * taken an iocount.
+        */
+       error = vnode_getalways_from_pager(vp);
+       if (error != 0) {
+               /* This can't happen */
+               panic("vnode_getalways returned %d for vp %p", error, vp);
+       }
+
        if (isize <= 0) {
-               result    = PAGER_ERROR;
+               result    = PAGER_ERROR;
                error_ret = EINVAL;
                goto out;
        }
@@ -318,32 +340,34 @@ vnode_pageout(struct vnode *vp,
                result    = PAGER_ERROR;
                error_ret = EINVAL;
 
-               if (upl && !(flags & UPL_NOCOMMIT))
-                       ubc_upl_abort_range(upl, upl_offset, size, UPL_ABORT_FREE_ON_EMPTY);
+               if (upl && !(flags & UPL_NOCOMMIT)) {
+                       ubc_upl_abort_range(upl, upl_offset, size, UPL_ABORT_FREE_ON_EMPTY);
+               }
                goto out;
        }
-       if ( !(flags & UPL_VNODE_PAGER)) {
+       if (!(flags & UPL_VNODE_PAGER)) {
                /*
                 * This is a pageout from the default pager,
                 * just go ahead and call vnop_pageout since
                 * it has already sorted out the dirty ranges
                 */
                KERNEL_DEBUG_CONSTANT_IST(KDEBUG_TRACE,
-                       (MACHDBG_CODE(DBG_MACH_VM, 1)) | DBG_FUNC_START, 
-                       size, 1, 0, 0, 0);
+                   (MACHDBG_CODE(DBG_MACH_VM, 1)) | DBG_FUNC_START,
+                   size, 1, 0, 0, 0);
 
-               if ( (error_ret = VNOP_PAGEOUT(vp, upl, upl_offset, (off_t)f_offset,
-                                              (size_t)size, flags, ctx)) )
+               if ((error_ret = VNOP_PAGEOUT(vp, upl, upl_offset, (off_t)f_offset,
+                   (size_t)size, flags, ctx))) {
                        result = PAGER_ERROR;
+               }
 
                KERNEL_DEBUG_CONSTANT_IST(KDEBUG_TRACE,
-                       (MACHDBG_CODE(DBG_MACH_VM, 1)) | DBG_FUNC_END, 
-                       size, 1, 0, 0, 0);
+                   (MACHDBG_CODE(DBG_MACH_VM, 1)) | DBG_FUNC_END,
+                   size, 1, 0, 0, 0);
 
                goto out;
        }
        if (upl == NULL) {
-               int                     request_flags;
+               int                     request_flags;
 
                if (vp->v_mount->mnt_vtable->vfc_vfsflags & VFC_VFSVNOP_PAGEOUTV2) {
                        /*
@@ -353,52 +377,56 @@ vnode_pageout(struct vnode *vp,
                         * via 'f_offset' and 'size' into a UPL... this allows the filesystem to first
                         * take any locks it needs, before effectively locking the pages into a UPL...
                         */
-                       KERNEL_DEBUG_CONSTANT_IST(KDEBUG_TRACE, 
-                               (MACHDBG_CODE(DBG_MACH_VM, 1)) | DBG_FUNC_START, 
-                               size, (int)f_offset, 0, 0, 0);
+                       KERNEL_DEBUG_CONSTANT_IST(KDEBUG_TRACE,
+                           (MACHDBG_CODE(DBG_MACH_VM, 1)) | DBG_FUNC_START,
+                           size, (int)f_offset, 0, 0, 0);
 
-                       if ( (error_ret = VNOP_PAGEOUT(vp, NULL, upl_offset, (off_t)f_offset,
-                                                      size, flags, ctx)) ) {
+                       if ((error_ret = VNOP_PAGEOUT(vp, NULL, upl_offset, (off_t)f_offset,
+                           size, flags, ctx))) {
                                result = PAGER_ERROR;
                        }
                        KERNEL_DEBUG_CONSTANT_IST(KDEBUG_TRACE,
-                               (MACHDBG_CODE(DBG_MACH_VM, 1)) | DBG_FUNC_END, 
-                               size, 0, 0, 0, 0);
+                           (MACHDBG_CODE(DBG_MACH_VM, 1)) | DBG_FUNC_END,
+                           size, 0, 0, 0, 0);
 
                        goto out;
                }
-               if (flags & UPL_MSYNC)
+               if (flags & UPL_MSYNC) {
                        request_flags = UPL_UBC_MSYNC | UPL_RET_ONLY_DIRTY;
-               else
+               } else {
                        request_flags = UPL_UBC_PAGEOUT | UPL_RET_ONLY_DIRTY;
-               
-               if (ubc_create_upl_kernel(vp, f_offset, size, &upl, &pl, request_flags, VM_KERN_MEMORY_FILE) != KERN_SUCCESS) {
+               }
+
+               if (ubc_create_upl_kernel(vp, f_offset, size, &upl, &pl, request_flags, VM_KERN_MEMORY_FILE) != KERN_SUCCESS) {
                        result    = PAGER_ERROR;
                        error_ret = EINVAL;
                        goto out;
                }
                upl_offset = 0;
-       } else 
+       } else {
                pl = ubc_upl_pageinfo(upl);
+       }
 
        /*
         * Ignore any non-present pages at the end of the
-        * UPL so that we aren't looking at a upl that 
+        * UPL so that we aren't looking at a upl that
         * may already have been freed by the preceeding
         * aborts/completions.
         */
        base_index = upl_offset / PAGE_SIZE;
 
        for (pg_index = (upl_offset + isize) / PAGE_SIZE; pg_index > base_index;) {
-               if (upl_page_present(pl, --pg_index))
-                       break;
+               if (upl_page_present(pl, --pg_index)) {
+                       break;
+               }
                if (pg_index == base_index) {
-                       /*
+                       /*
                         * no pages were returned, so release
                         * our hold on the upl and leave
                         */
-                       if ( !(flags & UPL_NOCOMMIT))
-                               ubc_upl_abort_range(upl, upl_offset, isize, UPL_ABORT_FREE_ON_EMPTY);
+                       if (!(flags & UPL_NOCOMMIT)) {
+                               ubc_upl_abort_range(upl, upl_offset, isize, UPL_ABORT_FREE_ON_EMPTY);
+                       }
 
                        goto out;
                }
@@ -414,32 +442,33 @@ vnode_pageout(struct vnode *vp,
         */
 
        if (ubc_getsize(vp) == 0) {
-               /*
+               /*
                 * if the file has been effectively deleted, then
                 * we need to go through the UPL and invalidate any
                 * buffer headers we might have that reference any
                 * of it's pages
                 */
                for (offset = upl_offset; isize; isize -= PAGE_SIZE, offset += PAGE_SIZE) {
-#if NFSCLIENT
-                       if (vp->v_tag == VT_NFS)
+#if CONFIG_NFS_CLIENT
+                       if (vp->v_tag == VT_NFS) {
                                /* check with nfs if page is OK to drop */
                                error = nfs_buf_page_inval(vp, (off_t)f_offset);
-                       else
-#endif
+                       else
+#endif /* CONFIG_NFS_CLIENT */
                        {
-                               blkno = ubc_offtoblk(vp, (off_t)f_offset);
-                               error = buf_invalblkno(vp, blkno, 0);
+                               blkno = ubc_offtoblk(vp, (off_t)f_offset);
+                               error = buf_invalblkno(vp, blkno, 0);
                        }
                        if (error) {
-                               if ( !(flags & UPL_NOCOMMIT))
-                                       ubc_upl_abort_range(upl, offset, PAGE_SIZE, UPL_ABORT_FREE_ON_EMPTY);
-                               if (error_ret == 0)
-                                       error_ret = error;
+                               if (!(flags & UPL_NOCOMMIT)) {
+                                       ubc_upl_abort_range(upl, offset, PAGE_SIZE, UPL_ABORT_FREE_ON_EMPTY);
+                               }
+                               if (error_ret == 0) {
+                                       error_ret = error;
+                               }
                                result = PAGER_ERROR;
-
-                       } else if ( !(flags & UPL_NOCOMMIT)) {
-                               ubc_upl_commit_range(upl, offset, PAGE_SIZE, UPL_COMMIT_FREE_ON_EMPTY);
+                       } else if (!(flags & UPL_NOCOMMIT)) {
+                               ubc_upl_commit_range(upl, offset, PAGE_SIZE, UPL_COMMIT_FREE_ON_EMPTY);
                        }
                        f_offset += PAGE_SIZE;
                }
@@ -453,20 +482,20 @@ vnode_pageout(struct vnode *vp,
                int  xsize;
                int  num_of_pages;
 
-               if ( !upl_page_present(pl, pg_index)) {
-                       /*
+               if (!upl_page_present(pl, pg_index)) {
+                       /*
                         * we asked for RET_ONLY_DIRTY, so it's possible
                         * to get back empty slots in the UPL
                         * just skip over them
                         */
-                       f_offset += PAGE_SIZE;
+                       f_offset += PAGE_SIZE;
                        offset   += PAGE_SIZE;
                        isize    -= PAGE_SIZE;
                        pg_index++;
 
                        continue;
                }
-               if ( !upl_dirty_page(pl, pg_index)) {
+               if (!upl_dirty_page(pl, pg_index)) {
                        /*
                         * if the page is not dirty and reached here it is
                         * marked precious or it is due to invalidation in
@@ -477,27 +506,28 @@ vnode_pageout(struct vnode *vp,
                         * Note we must not sleep here if the buffer is busy - that is
                         * a lock inversion which causes deadlock.
                         */
-#if NFSCLIENT
-                       if (vp->v_tag == VT_NFS)
+#if CONFIG_NFS_CLIENT
+                       if (vp->v_tag == VT_NFS) {
                                /* check with nfs if page is OK to drop */
                                error = nfs_buf_page_inval(vp, (off_t)f_offset);
-                       else
-#endif
+                       else
+#endif /* CONFIG_NFS_CLIENT */
                        {
-                               blkno = ubc_offtoblk(vp, (off_t)f_offset);
-                               error = buf_invalblkno(vp, blkno, 0);
+                               blkno = ubc_offtoblk(vp, (off_t)f_offset);
+                               error = buf_invalblkno(vp, blkno, 0);
                        }
                        if (error) {
-                               if ( !(flags & UPL_NOCOMMIT))
-                                       ubc_upl_abort_range(upl, offset, PAGE_SIZE, UPL_ABORT_FREE_ON_EMPTY);
-                               if (error_ret == 0)
-                                       error_ret = error;
+                               if (!(flags & UPL_NOCOMMIT)) {
+                                       ubc_upl_abort_range(upl, offset, PAGE_SIZE, UPL_ABORT_FREE_ON_EMPTY);
+                               }
+                               if (error_ret == 0) {
+                                       error_ret = error;
+                               }
                                result = PAGER_ERROR;
-
-                       } else if ( !(flags & UPL_NOCOMMIT)) {
-                               ubc_upl_commit_range(upl, offset, PAGE_SIZE, UPL_COMMIT_FREE_ON_EMPTY);
+                       } else if (!(flags & UPL_NOCOMMIT)) {
+                               ubc_upl_commit_range(upl, offset, PAGE_SIZE, UPL_COMMIT_FREE_ON_EMPTY);
                        }
-                       f_offset += PAGE_SIZE;
+                       f_offset += PAGE_SIZE;
                        offset   += PAGE_SIZE;
                        isize    -= PAGE_SIZE;
                        pg_index++;
@@ -508,81 +538,108 @@ vnode_pageout(struct vnode *vp,
                xsize = isize - PAGE_SIZE;
 
                while (xsize) {
-                       if ( !upl_dirty_page(pl, pg_index + num_of_pages))
+                       if (!upl_dirty_page(pl, pg_index + num_of_pages)) {
                                break;
+                       }
                        num_of_pages++;
                        xsize -= PAGE_SIZE;
                }
                xsize = num_of_pages * PAGE_SIZE;
 
                KERNEL_DEBUG_CONSTANT_IST(KDEBUG_TRACE,
-                       (MACHDBG_CODE(DBG_MACH_VM, 1)) | DBG_FUNC_START, 
-                       xsize, (int)f_offset, 0, 0, 0);
+                   (MACHDBG_CODE(DBG_MACH_VM, 1)) | DBG_FUNC_START,
+                   xsize, (int)f_offset, 0, 0, 0);
 
-               if ( (error = VNOP_PAGEOUT(vp, upl, offset, (off_t)f_offset,
-                                          xsize, flags, ctx)) ) {
-                       if (error_ret == 0)
-                               error_ret = error;
+               if ((error = VNOP_PAGEOUT(vp, upl, offset, (off_t)f_offset,
+                   xsize, flags, ctx))) {
+                       if (error_ret == 0) {
+                               error_ret = error;
+                       }
                        result = PAGER_ERROR;
                }
                KERNEL_DEBUG_CONSTANT_IST(KDEBUG_TRACE,
-                       (MACHDBG_CODE(DBG_MACH_VM, 1)) | DBG_FUNC_END, 
-                       xsize, 0, 0, 0, 0);
+                   (MACHDBG_CODE(DBG_MACH_VM, 1)) | DBG_FUNC_END,
+                   xsize, 0, 0, 0, 0);
 
-               f_offset += xsize;
+               f_offset += xsize;
                offset   += xsize;
                isize    -= xsize;
                pg_index += num_of_pages;
        }
 out:
-       if (errorp)
+       vnode_put_from_pager(vp);
+
+       if (errorp) {
                *errorp = error_ret;
+       }
 
-       return (result);
+       return result;
 }
 
 
 pager_return_t
 vnode_pagein(
-       struct vnode            *vp,
-       upl_t                   upl,
-       upl_offset_t            upl_offset,
-       vm_object_offset_t      f_offset,
-       upl_size_t              size,
-       int                     flags,
-       int                     *errorp)
+       struct vnode            *vp,
+       upl_t                   upl,
+       upl_offset_t            upl_offset,
+       vm_object_offset_t      f_offset,
+       upl_size_t              size,
+       int                     flags,
+       int                     *errorp)
 {
-        upl_page_info_t *pl;
-       int             result = PAGER_SUCCESS;
-       int             error = 0;
-        int             pages_in_upl;
-        int             start_pg;
-        int             last_pg;
+       upl_page_info_t *pl;
+       int             result = PAGER_SUCCESS;
+       int             error = 0;
+       int             pages_in_upl;
+       int             start_pg;
+       int             last_pg;
        int             first_pg;
-        int             xsize;
-       int             must_commit = 1;
-       int             ignore_valid_page_check = 0;
+       int             xsize;
+       int             must_commit = 1;
+       int             ignore_valid_page_check = 0;
 
-       if (flags & UPL_NOCOMMIT)
-               must_commit = 0;
+       if (flags & UPL_NOCOMMIT) {
+               must_commit = 0;
+       }
 
-       if (flags & UPL_IGNORE_VALID_PAGE_CHECK)
+       if (flags & UPL_IGNORE_VALID_PAGE_CHECK) {
                ignore_valid_page_check = 1;
+       }
+
+       /*
+        * This call is non-blocking and does not ever fail but it can
+        * only be made when there is other explicit synchronization
+        * with reclaiming of the vnode which, in this path, is provided
+        * by the paging in progress counter.
+        *
+        * In addition, this may also be entered via vm_swapfile_io
+        * where the existing iocount provides the necessary synchronization.
+        * Ideally we would not take an additional iocount here in the cases
+        * where an explcit iocount has already been taken but this call
+        * doesn't cause a deadlock as other forms of vnode_get* might if
+        * this thread has already taken an iocount.
+        */
+       error = vnode_getalways_from_pager(vp);
+       if (error != 0) {
+               /* This can't happen */
+               panic("vnode_getalways returned %d for vp %p", error, vp);
+       }
 
        if (UBCINFOEXISTS(vp) == 0) {
                result = PAGER_ERROR;
                error  = PAGER_ERROR;
 
-               if (upl && must_commit)
+               if (upl && must_commit) {
                        ubc_upl_abort_range(upl, upl_offset, size, UPL_ABORT_FREE_ON_EMPTY | UPL_ABORT_ERROR);
+               }
 
                goto out;
        }
        if (upl == (upl_t)NULL) {
                flags &= ~UPL_NOCOMMIT;
 
-               if (size > MAX_UPL_SIZE_BYTES) {
-                       result = PAGER_ERROR;
+               if (size > MAX_UPL_SIZE_BYTES) {
+                       result = PAGER_ERROR;
                        error  = PAGER_ERROR;
                        goto out;
                }
@@ -595,21 +652,22 @@ vnode_pagein(
                         * take any locks it needs, before effectively locking the pages into a UPL...
                         * so we pass a NULL into the filesystem instead of a UPL pointer... the 'upl_offset'
                         * is used to identify the "must have" page in the extent... the filesystem is free
-                        * to clip the extent to better fit the underlying FS blocksize if it desires as 
+                        * to clip the extent to better fit the underlying FS blocksize if it desires as
                         * long as it continues to include the "must have" page... 'f_offset' + 'upl_offset'
                         * identifies that page
                         */
-                       if ( (error = VNOP_PAGEIN(vp, NULL, upl_offset, (off_t)f_offset,
-                                                 size, flags, vfs_context_current())) ) {
+                       if ((error = VNOP_PAGEIN(vp, NULL, upl_offset, (off_t)f_offset,
+                           size, flags, vfs_context_current()))) {
+                               set_thread_pagein_error(current_thread(), error);
                                result = PAGER_ERROR;
                                error  = PAGER_ERROR;
                        }
                        goto out;
                }
-               ubc_create_upl_kernel(vp, f_offset, size, &upl, &pl, UPL_UBC_PAGEIN | UPL_RET_ONLY_ABSENT, VM_KERN_MEMORY_FILE);
+               ubc_create_upl_kernel(vp, f_offset, size, &upl, &pl, UPL_UBC_PAGEIN | UPL_RET_ONLY_ABSENT, VM_KERN_MEMORY_FILE);
 
                if (upl == (upl_t)NULL) {
-                       result =  PAGER_ABSENT;
+                       result =  PAGER_ABSENT;
                        error = PAGER_ABSENT;
                        goto out;
                }
@@ -617,7 +675,7 @@ vnode_pagein(
 
                upl_offset = 0;
                first_pg = 0;
-               
+
                /*
                 * if we get here, we've created the upl and
                 * are responsible for commiting/aborting it
@@ -625,26 +683,28 @@ vnode_pagein(
                 */
                must_commit = 1;
        } else {
-               pl = ubc_upl_pageinfo(upl);
+               pl = ubc_upl_pageinfo(upl);
                first_pg = upl_offset / PAGE_SIZE;
        }
        pages_in_upl = size / PAGE_SIZE;
        DTRACE_VM2(pgpgin, int, pages_in_upl, (uint64_t *), NULL);
 
        /*
-        * before we start marching forward, we must make sure we end on 
+        * before we start marching forward, we must make sure we end on
         * a present page, otherwise we will be working with a freed
-         * upl
+        * upl
         */
        for (last_pg = pages_in_upl - 1; last_pg >= first_pg; last_pg--) {
-               if (upl_page_present(pl, last_pg))
+               if (upl_page_present(pl, last_pg)) {
                        break;
+               }
                if (last_pg == first_pg) {
-                       /*
+                       /*
                         * empty UPL, no pages are present
                         */
-                       if (must_commit)
-                               ubc_upl_abort_range(upl, upl_offset, size, UPL_ABORT_FREE_ON_EMPTY);
+                       if (must_commit) {
+                               ubc_upl_abort_range(upl, upl_offset, size, UPL_ABORT_FREE_ON_EMPTY);
+                       }
                        goto out;
                }
        }
@@ -652,54 +712,58 @@ vnode_pagein(
        last_pg = first_pg;
 
        while (last_pg < pages_in_upl) {
-               /*
+               /*
                 * skip over missing pages...
                 */
-               for ( ; last_pg < pages_in_upl; last_pg++) {
-                       if (upl_page_present(pl, last_pg))
-                               break;
+               for (; last_pg < pages_in_upl; last_pg++) {
+                       if (upl_page_present(pl, last_pg)) {
+                               break;
+                       }
                }
 
                if (ignore_valid_page_check == 1) {
                        start_pg = last_pg;
                } else {
-                       /*
+                       /*
                         * skip over 'valid' pages... we don't want to issue I/O for these
                         */
-                       for (start_pg = last_pg; last_pg < pages_in_upl; last_pg++) {
-                               if (!upl_valid_page(pl, last_pg))
-                                       break;
+                       for (start_pg = last_pg; last_pg < pages_in_upl; last_pg++) {
+                               if (!upl_valid_page(pl, last_pg)) {
+                                       break;
+                               }
                        }
                }
 
                if (last_pg > start_pg) {
-                       /*
+                       /*
                         * we've found a range of valid pages
                         * if we've got COMMIT responsibility
                         * commit this range of pages back to the
                         * cache unchanged
                         */
-                       xsize = (last_pg - start_pg) * PAGE_SIZE;
+                       xsize = (last_pg - start_pg) * PAGE_SIZE;
 
-                       if (must_commit)
-                               ubc_upl_abort_range(upl, start_pg * PAGE_SIZE, xsize, UPL_ABORT_FREE_ON_EMPTY);
+                       if (must_commit) {
+                               ubc_upl_abort_range(upl, start_pg * PAGE_SIZE, xsize, UPL_ABORT_FREE_ON_EMPTY);
+                       }
                }
-               if (last_pg == pages_in_upl)
-                       /*
+               if (last_pg == pages_in_upl) {
+                       /*
                         * we're done... all pages that were present
-                        * have either had I/O issued on them or 
+                        * have either had I/O issued on them or
                         * were aborted unchanged...
                         */
-                       break;
+                       break;
+               }
 
                if (!upl_page_present(pl, last_pg)) {
-                       /*
-                        * we found a range of valid pages 
+                       /*
+                        * we found a range of valid pages
                         * terminated by a missing page...
                         * bump index to the next page and continue on
                         */
-                       last_pg++;
-                       continue;
+                       last_pg++;
+                       continue;
                }
                /*
                 * scan from the found invalid page looking for a valid
@@ -708,51 +772,54 @@ vnode_pagein(
                 * 'cluster_io'
                 */
                for (start_pg = last_pg; last_pg < pages_in_upl; last_pg++) {
-                       if (( !ignore_valid_page_check && upl_valid_page(pl, last_pg)) || !upl_page_present(pl, last_pg))
-                               break;
+                       if ((!ignore_valid_page_check && upl_valid_page(pl, last_pg)) || !upl_page_present(pl, last_pg)) {
+                               break;
+                       }
                }
                if (last_pg > start_pg) {
-                       int xoff;
-                       xsize = (last_pg - start_pg) * PAGE_SIZE;
+                       int xoff;
+                       xsize = (last_pg - start_pg) * PAGE_SIZE;
                        xoff  = start_pg * PAGE_SIZE;
 
-                       if ( (error = VNOP_PAGEIN(vp, upl, (upl_offset_t) xoff,
-                                              (off_t)f_offset + xoff,
-                                              xsize, flags, vfs_context_current())) ) {
-                               /*
+                       if ((error = VNOP_PAGEIN(vp, upl, (upl_offset_t) xoff,
+                           (off_t)f_offset + xoff,
+                           xsize, flags, vfs_context_current()))) {
+                               /*
                                 * Usually this UPL will be aborted/committed by the lower cluster layer.
                                 *
                                 * a)   In the case of decmpfs, however, we may return an error (EAGAIN) to avoid
-                                *      a deadlock with another thread already inflating the file. 
+                                *      a deadlock with another thread already inflating the file.
                                 *
                                 * b)   In the case of content protection, EPERM is a valid error and we should respect it.
                                 *
                                 * In those cases, we must take care of our UPL at this layer itself.
                                 */
                                if (must_commit) {
-                                       if(error == EAGAIN) {
-                                               ubc_upl_abort_range(upl, (upl_offset_t) xoff, xsize, UPL_ABORT_FREE_ON_EMPTY | UPL_ABORT_RESTART);
+                                       if (error == EAGAIN) {
+                                               ubc_upl_abort_range(upl, (upl_offset_t) xoff, xsize, UPL_ABORT_FREE_ON_EMPTY | UPL_ABORT_RESTART);
                                        }
-                                       if(error == EPERM) {
-                                               ubc_upl_abort_range(upl, (upl_offset_t) xoff, xsize, UPL_ABORT_FREE_ON_EMPTY | UPL_ABORT_ERROR);
+                                       if (error == EPERM) {
+                                               ubc_upl_abort_range(upl, (upl_offset_t) xoff, xsize, UPL_ABORT_FREE_ON_EMPTY | UPL_ABORT_ERROR);
                                        }
                                }
+                               set_thread_pagein_error(current_thread(), error);
                                result = PAGER_ERROR;
                                error  = PAGER_ERROR;
-
                        }
                }
-        }
+       }
 out:
-       if (errorp)
+       vnode_put_from_pager(vp);
+
+       if (errorp) {
                *errorp = result;
+       }
 
-       return (error);
+       return error;
 }
 
 void *
 upl_get_internal_page_list(upl_t upl)
 {
-  return(UPL_GET_INTERNAL_PAGE_LIST(upl));
-
+       return UPL_GET_INTERNAL_PAGE_LIST(upl);
 }