X-Git-Url: https://git.saurik.com/apple/xnu.git/blobdiff_plain/89b3af67bb32e691275bf6fa803d1834b2284115..060df5ea7c632b1ac8cc8aac1fb59758165c2084:/bsd/vm/vnode_pager.c diff --git a/bsd/vm/vnode_pager.c b/bsd/vm/vnode_pager.c index 552c0afb0..a15b6dcc4 100644 --- a/bsd/vm/vnode_pager.c +++ b/bsd/vm/vnode_pager.c @@ -1,5 +1,5 @@ /* - * Copyright (c) 2000-2004 Apple Computer, Inc. All rights reserved. + * Copyright (c) 2000-2006 Apple Computer, Inc. All rights reserved. * * @APPLE_OSREFERENCE_LICENSE_HEADER_START@ * @@ -55,6 +55,11 @@ #include #include +#include +#include +#include +#include +#include #include #include @@ -75,11 +80,20 @@ #include -unsigned int vp_pagein=0; -unsigned int vp_pgodirty=0; -unsigned int vp_pgoclean=0; -unsigned int dp_pgouts=0; /* Default pager pageouts */ -unsigned int dp_pgins=0; /* Default pager pageins */ + +uint32_t +vnode_pager_isinuse(struct vnode *vp) +{ + if (vp->v_usecount > vp->v_kusecount) + return (1); + return (0); +} + +uint32_t +vnode_pager_return_hard_throttle_limit(struct vnode *vp, uint32_t *limit, uint32_t hard_throttle) +{ + return(cluster_hard_throttle_limit(vp, limit, hard_throttle)); +} vm_object_offset_t vnode_pager_get_filesize(struct vnode *vp) @@ -108,22 +122,30 @@ vnode_pager_get_pathname( kern_return_t vnode_pager_get_filename( struct vnode *vp, - char **filename) + const char **filename) { *filename = vp->v_name; return KERN_SUCCESS; } +kern_return_t +vnode_pager_get_cs_blobs( + struct vnode *vp, + void **blobs) +{ + *blobs = ubc_get_cs_blobs(vp); + return KERN_SUCCESS; +} + pager_return_t vnode_pageout(struct vnode *vp, upl_t upl, - vm_offset_t upl_offset, + upl_offset_t upl_offset, vm_object_offset_t f_offset, - vm_size_t size, + upl_size_t size, int flags, int *errorp) { - struct proc *p = current_proc(); int result = PAGER_SUCCESS; int error = 0; int error_ret = 0; @@ -131,12 +153,9 @@ vnode_pageout(struct vnode *vp, int isize; int pg_index; int base_index; - int offset; + upl_offset_t offset; upl_page_info_t *pl; - struct vfs_context context; - - context.vc_proc = p; - context.vc_ucred = kauth_cred_get(); + vfs_context_t ctx = vfs_context_current(); /* pager context */ isize = (int)size; @@ -145,9 +164,8 @@ vnode_pageout(struct vnode *vp, error_ret = EINVAL; goto out; } - UBCINFOCHECK("vnode_pageout", vp); - if (UBCINVALID(vp)) { + if (UBCINFOEXISTS(vp) == 0) { result = PAGER_ERROR; error_ret = EINVAL; @@ -161,13 +179,11 @@ vnode_pageout(struct vnode *vp, * just go ahead and call vnop_pageout since * it has already sorted out the dirty ranges */ - dp_pgouts++; - KERNEL_DEBUG_CONSTANT((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, &context)) ) + (size_t)size, flags, ctx)) ) result = PAGER_ERROR; KERNEL_DEBUG_CONSTANT((MACHDBG_CODE(DBG_MACH_VM, 1)) | DBG_FUNC_END, @@ -175,6 +191,45 @@ vnode_pageout(struct vnode *vp, goto out; } + if (upl == NULL) { + int request_flags; + + if (vp->v_mount->mnt_vtable->vfc_vfsflags & VFC_VFSVNOP_PAGEOUTV2) { + /* + * filesystem has requested the new form of VNOP_PAGEOUT for file + * backed objects... we will not grab the UPL befofe calling VNOP_PAGEOUT... + * it is the fileystem's responsibility to grab the range we're denoting + * 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((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)) ) { + result = PAGER_ERROR; + } + KERNEL_DEBUG_CONSTANT((MACHDBG_CODE(DBG_MACH_VM, 1)) | DBG_FUNC_END, + size, 0, 0, 0, 0); + + goto out; + } + if (flags & UPL_MSYNC) + request_flags = UPL_UBC_MSYNC | UPL_RET_ONLY_DIRTY; + else + request_flags = UPL_UBC_PAGEOUT | UPL_RET_ONLY_DIRTY; + + ubc_create_upl(vp, f_offset, size, &upl, &pl, request_flags); + + if (upl == (upl_t)NULL) { + result = PAGER_ERROR; + error_ret = EINVAL; + goto out; + } + upl_offset = 0; + } else + pl = ubc_upl_pageinfo(upl); + /* * we come here for pageouts to 'real' files and * for msyncs... the upl may not contain any @@ -182,8 +237,6 @@ vnode_pageout(struct vnode *vp, * through it and find the 'runs' of dirty pages * to call VNOP_PAGEOUT on... */ - pl = ubc_upl_pageinfo(upl); - if (ubc_getsize(vp) == 0) { /* * if the file has been effectively deleted, then @@ -238,7 +291,7 @@ vnode_pageout(struct vnode *vp, goto out; } } - isize = (pg_index + 1) * PAGE_SIZE; + isize = ((pg_index + 1) - base_index) * PAGE_SIZE; offset = upl_offset; pg_index = base_index; @@ -253,8 +306,9 @@ vnode_pageout(struct vnode *vp, * to get back empty slots in the UPL * just skip over them */ - offset += PAGE_SIZE; - isize -= PAGE_SIZE; + f_offset += PAGE_SIZE; + offset += PAGE_SIZE; + isize -= PAGE_SIZE; pg_index++; continue; @@ -270,16 +324,14 @@ vnode_pageout(struct vnode *vp, * Note we must not sleep here if the buffer is busy - that is * a lock inversion which causes deadlock. */ - vp_pgoclean++; - #if NFSCLIENT 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 + offset)); + error = nfs_buf_page_inval(vp, (off_t)f_offset); else #endif { - blkno = ubc_offtoblk(vp, (off_t)(f_offset + offset)); + blkno = ubc_offtoblk(vp, (off_t)f_offset); error = buf_invalblkno(vp, blkno, 0); } if (error) { @@ -292,14 +344,13 @@ vnode_pageout(struct vnode *vp, } else if ( !(flags & UPL_NOCOMMIT)) { ubc_upl_commit_range(upl, offset, PAGE_SIZE, UPL_COMMIT_FREE_ON_EMPTY); } - offset += PAGE_SIZE; - isize -= PAGE_SIZE; + f_offset += PAGE_SIZE; + offset += PAGE_SIZE; + isize -= PAGE_SIZE; pg_index++; continue; } - vp_pgodirty++; - num_of_pages = 1; xsize = isize - PAGE_SIZE; @@ -312,11 +363,10 @@ vnode_pageout(struct vnode *vp, xsize = num_of_pages * PAGE_SIZE; KERNEL_DEBUG_CONSTANT((MACHDBG_CODE(DBG_MACH_VM, 1)) | DBG_FUNC_START, - xsize, (int)(f_offset + offset), 0, 0, 0); + xsize, (int)f_offset, 0, 0, 0); - if ( (error = VNOP_PAGEOUT(vp, upl, (vm_offset_t)offset, - (off_t)(f_offset + offset), xsize, - flags, &context)) ) { + if ( (error = VNOP_PAGEOUT(vp, upl, offset, (off_t)f_offset, + xsize, flags, ctx)) ) { if (error_ret == 0) error_ret = error; result = PAGER_ERROR; @@ -324,8 +374,9 @@ vnode_pageout(struct vnode *vp, KERNEL_DEBUG_CONSTANT((MACHDBG_CODE(DBG_MACH_VM, 1)) | DBG_FUNC_END, xsize, 0, 0, 0, 0); - offset += xsize; - isize -= xsize; + f_offset += xsize; + offset += xsize; + isize -= xsize; pg_index += num_of_pages; } out: @@ -336,19 +387,16 @@ out: } -void IOSleep(int); - pager_return_t vnode_pagein( struct vnode *vp, upl_t upl, - vm_offset_t upl_offset, + upl_offset_t upl_offset, vm_object_offset_t f_offset, - vm_size_t size, + upl_size_t size, int flags, int *errorp) { - struct proc *p = current_proc(); struct uthread *ut; upl_page_info_t *pl; int result = PAGER_SUCCESS; @@ -358,26 +406,49 @@ vnode_pagein( int last_pg; int first_pg; int xsize; - int abort_needed = 1; - + int must_commit = 1; - UBCINFOCHECK("vnode_pagein", vp); + if (flags & UPL_NOCOMMIT) + must_commit = 0; - if (UBCINVALID(vp)) { + if (UBCINFOEXISTS(vp) == 0) { result = PAGER_ERROR; error = PAGER_ERROR; - if (upl && !(flags & UPL_NOCOMMIT)) { + + 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) { - if (size > (MAX_UPL_TRANSFER * PAGE_SIZE)) { + flags &= ~UPL_NOCOMMIT; + + if (size > (MAX_UPL_SIZE * PAGE_SIZE)) { result = PAGER_ERROR; error = PAGER_ERROR; goto out; } - ubc_create_upl(vp, f_offset, size, &upl, &pl, UPL_RET_ONLY_ABSENT | UPL_SET_LITE); + if (vp->v_mount->mnt_vtable->vfc_vfsflags & VFC_VFSVNOP_PAGEINV2) { + /* + * filesystem has requested the new form of VNOP_PAGEIN for file + * backed objects... we will not grab the UPL befofe calling VNOP_PAGEIN... + * it is the fileystem's responsibility to grab the range we're denoting + * 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... + * 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 + * 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())) ) { + result = PAGER_ERROR; + error = PAGER_ERROR; + } + goto out; + } + ubc_create_upl(vp, f_offset, size, &upl, &pl, UPL_UBC_PAGEIN | UPL_RET_ONLY_ABSENT); if (upl == (upl_t)NULL) { result = PAGER_ABSENT; @@ -385,21 +456,20 @@ vnode_pagein( goto out; } upl_offset = 0; + first_pg = 0; + /* * if we get here, we've created the upl and * are responsible for commiting/aborting it * regardless of what the caller has passed in */ - flags &= ~UPL_NOCOMMIT; - - vp_pagein++; + must_commit = 1; } else { pl = ubc_upl_pageinfo(upl); - - dp_pgins++; + first_pg = upl_offset / PAGE_SIZE; } pages_in_upl = size / PAGE_SIZE; - first_pg = upl_offset / PAGE_SIZE; + DTRACE_VM2(pgpgin, int, pages_in_upl, (uint64_t *), NULL); /* * before we start marching forward, we must make sure we end on @@ -409,30 +479,31 @@ vnode_pagein( for (last_pg = pages_in_upl - 1; last_pg >= first_pg; 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); + goto out; + } } pages_in_upl = last_pg + 1; + last_pg = first_pg; - for (last_pg = first_pg; last_pg < pages_in_upl;) { + while (last_pg < pages_in_upl) { /* - * scan the upl looking for the next - * page that is present.... if all of the - * pages are absent, we're done + * skip over missing pages... */ - for (start_pg = last_pg; last_pg < pages_in_upl; last_pg++) { + for ( ; last_pg < pages_in_upl; last_pg++) { if (upl_page_present(pl, last_pg)) break; } - if (last_pg == pages_in_upl) - break; - /* - * if we get here, we've sitting on a page - * that is present... we want to skip over - * any range of 'valid' pages... if this takes - * us to the end of the request, than we're done + * 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) || !upl_page_present(pl, last_pg)) + if (!upl_valid_page(pl, last_pg)) break; } if (last_pg > start_pg) { @@ -444,22 +515,26 @@ vnode_pagein( */ xsize = (last_pg - start_pg) * PAGE_SIZE; - if (!(flags & UPL_NOCOMMIT)) + if (must_commit) ubc_upl_abort_range(upl, start_pg * PAGE_SIZE, xsize, UPL_ABORT_FREE_ON_EMPTY); - - abort_needed = 0; } if (last_pg == pages_in_upl) + /* + * we're done... all pages that were present + * have either had I/O issued on them or + * were aborted unchanged... + */ break; - if (!upl_page_present(pl, last_pg)) + if (!upl_page_present(pl, last_pg)) { /* - * if we found a range of valid pages - * terminated by a non-present page - * than start over + * we found a range of valid pages + * terminated by a missing page... + * bump index to the next page and continue on */ + last_pg++; continue; - + } /* * scan from the found invalid page looking for a valid * or non-present page before the end of the upl is reached, if we @@ -472,41 +547,44 @@ vnode_pagein( } if (last_pg > start_pg) { int xoff; - struct vfs_context context; - - context.vc_proc = p; - context.vc_ucred = kauth_cred_get(); xsize = (last_pg - start_pg) * PAGE_SIZE; xoff = start_pg * PAGE_SIZE; - if ( (error = VNOP_PAGEIN(vp, upl, (vm_offset_t) xoff, + if ( (error = VNOP_PAGEIN(vp, upl, (upl_offset_t) xoff, (off_t)f_offset + xoff, - xsize, flags, &context)) ) { + xsize, flags, vfs_context_current())) ) { + /* + * Usually this UPL will be aborted/committed by the lower cluster layer. + * In the case of decmpfs, however, we may return an error (EAGAIN) to avoid + * a deadlock with another thread already inflating the file. In that case, + * 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); + } + } result = PAGER_ERROR; error = PAGER_ERROR; } - abort_needed = 0; } } - if (!(flags & UPL_NOCOMMIT) && abort_needed) - ubc_upl_abort_range(upl, upl_offset, size, UPL_ABORT_FREE_ON_EMPTY); out: if (errorp) *errorp = result; ut = get_bsdthread_info(current_thread()); - if (ut->uu_lowpri_delay) { + if (ut->uu_lowpri_window) { /* * task is marked as a low priority I/O type - * and the I/O we issued while in this system call + * and the I/O we issued while in this page fault * collided with normal I/O operations... we'll * delay in order to mitigate the impact of this * task on the normal operation of the system */ - IOSleep(ut->uu_lowpri_delay); - ut->uu_lowpri_delay = 0; + throttle_lowpri_io(TRUE); } return (error); }