X-Git-Url: https://git.saurik.com/apple/xnu.git/blobdiff_plain/3a60a9f5b85abb8c2cf24e1926c5c7b3f608a5e2..527f99514973766e9c0382a4d8550dfb00f54939:/bsd/kern/kern_symfile.c diff --git a/bsd/kern/kern_symfile.c b/bsd/kern/kern_symfile.c index 4dc099a0d..46018b2de 100644 --- a/bsd/kern/kern_symfile.c +++ b/bsd/kern/kern_symfile.c @@ -1,48 +1,35 @@ /* - * Copyright (c) 2000-2004 Apple Computer, Inc. All rights reserved. + * Copyright (c) 2000-2006 Apple Computer, Inc. All rights reserved. * - * @APPLE_LICENSE_HEADER_START@ + * @APPLE_OSREFERENCE_LICENSE_HEADER_START@ * - * The contents of this file constitute Original Code as defined in and - * are subject to the Apple Public Source License Version 1.1 (the - * "License"). You may not use this file except in compliance with the - * License. Please obtain a copy of the License at - * http://www.apple.com/publicsource and read it before using this file. + * 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 + * compliance with the License. The rights granted to you under the License + * may not be used to create, or enable the creation or redistribution of, + * 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. * - * This Original Code and all software distributed under the License are - * distributed on an "AS IS" basis, WITHOUT WARRANTY OF ANY KIND, EITHER + * 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, * INCLUDING WITHOUT LIMITATION, ANY WARRANTIES OF MERCHANTABILITY, - * FITNESS FOR A PARTICULAR PURPOSE OR NON-INFRINGEMENT. Please see the - * License for the specific language governing rights and limitations - * under the License. + * 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_LICENSE_HEADER_END@ + * @APPLE_OSREFERENCE_LICENSE_HEADER_END@ */ /* Copyright (c) 1998 Apple Computer, Inc. All rights reserved. * * File: bsd/kern/kern_symfile.c * - * This file contains creates a dummy symbol file for mach_kernel - * based on the symbol table information passed by the - * SecondaryLoader/PlatformExpert. This allows us to correctly - * link other executables (drivers, etc) against the the kernel in - * cases where the kernel image on the root device does not match - * the live kernel. This can occur during net-booting where the - * actual kernel image is obtained from the network via tftp rather - * than the root device. - * - * If a symbol table is available, then the file /mach.sym will be - * created containing a Mach Header and a LC_SYMTAB load command - * followed by the the symbol table data for mach_kernel. - * - * NOTE: This file supports only 32 bit kernels at the present time; - * adding support for 64 bit kernels is possible, but is not - * necessary at the present time. - * * HISTORY - * - * . */ #include @@ -64,6 +51,7 @@ #include #include #include +#include #include #include @@ -71,320 +59,186 @@ #include #include #include -#include - -extern unsigned char rootdevice[]; -extern struct mach_header _mh_execute_header; - -static int kernel_symfile_opened = 0; -static int error_code = 0; - -extern int IODTGetLoaderInfo(char *key, void **infoAddr, int *infoSize); -extern void IODTFreeLoaderInfo(char *key, void *infoAddr, int infoSize); - -/* - * Can only operate against currently running 32 bit mach_kernel - */ -static int -output_kernel_symbols(struct proc *p) -{ - struct vnode *vp; - kauth_cred_t cred = p->p_ucred; /* XXX */ - struct vnode_attr va; - struct vfs_context context; - struct load_command *cmd; - struct mach_header *orig_mh, *mh; - struct segment_command *orig_ds, *orig_ts, *orig_le, *sg; - struct section *se, *const_text; - struct symtab_command *st, *orig_st; - struct nlist *sym; - vm_size_t orig_mhsize, orig_st_size; - vm_offset_t header; - vm_size_t header_size = 0; /* out: protected by header */ - int error, error1; - unsigned int i, j; - caddr_t addr; - vm_offset_t offset; - int rc_mh, rc_sc; - - error = EFAULT; - - vp = NULL; - header = NULL; - orig_mh = NULL; - orig_st = NULL; - - // Dispose of unnecessary gumf, the booter doesn't need to load these - rc_mh = IODTGetLoaderInfo("Kernel-__HEADER", - (void **)&orig_mh, &orig_mhsize); - if (rc_mh == 0 && orig_mh) - IODTFreeLoaderInfo("Kernel-__HEADER", - (void *)orig_mh, round_page_32(orig_mhsize)); - - rc_sc = IODTGetLoaderInfo("Kernel-__SYMTAB", - (void **) &orig_st, &orig_st_size); - if (rc_sc == 0 && orig_st) - IODTFreeLoaderInfo("Kernel-__SYMTAB", - (void *)orig_st, round_page_32(orig_st_size)); - - if (cred->cr_svuid != cred->cr_ruid || cred->cr_svgid != cred->cr_rgid) - goto out; - - // Check to see if the root is 'e' or 'n', is this a test for network? - if (rootdevice[0] == 'e' && rootdevice[1] == 'n') - goto out; - - context.vc_proc = p; - context.vc_ucred = cred; - - if ((error = vnode_open("mach.sym", (O_CREAT | FWRITE), (S_IRUSR | S_IRGRP | S_IROTH), 0, &vp, &context))) - goto out; - - /* Don't dump to non-regular files or files with links. */ - error = EFAULT; - VATTR_INIT(&va); - VATTR_WANTED(&va, va_nlink); - if ((vp->v_type != VREG) || vnode_getattr(vp, &va, &context) || (va.va_nlink != 1)) - goto out; - - VATTR_INIT(&va); /* better to do it here than waste more stack in vnode_getsize */ - VATTR_SET(&va, va_data_size, 0); - vnode_setattr(vp, &va, &context); - p->p_acflag |= ACORE; - - // If the file type is MH_EXECUTE then this must be a kernel - // as all Kernel extensions must be of type MH_OBJECT - orig_ds = orig_ts = orig_le = NULL; - orig_st = NULL; - orig_mh = &_mh_execute_header; - cmd = (struct load_command *) &orig_mh[1]; - for (i = 0; i < orig_mh->ncmds; i++) { - if (cmd->cmd == LC_SEGMENT) { - struct segment_command *orig_sg = (struct segment_command *) cmd; - - if (!strcmp(SEG_TEXT, orig_sg->segname)) - orig_ts = orig_sg; - else if (!strcmp(SEG_DATA, orig_sg->segname)) - orig_ds = orig_sg; - else if (!strcmp(SEG_LINKEDIT, orig_sg->segname)) - orig_le = orig_sg; - } - else if (cmd->cmd == LC_SYMTAB) - orig_st = (struct symtab_command *) cmd; - - cmd = (struct load_command *) ((caddr_t) cmd + cmd->cmdsize); - } - - if (!orig_ts || !orig_ds || !orig_le || !orig_st) - goto out; - - const_text = NULL; - se = (struct section *) &orig_ts[1]; - for (i = 0; i < orig_ts->nsects; i++, se++) { - if (!strcmp("__const", se->sectname)) { - const_text = se; - break; - } - } - if (!const_text) - goto out; - - header_size = sizeof(struct mach_header) - + orig_ts->cmdsize - + orig_ds->cmdsize - + sizeof(struct symtab_command); - - (void) kmem_alloc(kernel_map, - (vm_offset_t *) &header, - (vm_size_t) header_size); - if (header) - bzero((void *) header, header_size); - else - goto out; - - /* - * Set up Mach-O header. - */ - mh = (struct mach_header *) header; - mh->magic = orig_mh->magic; - mh->cputype = orig_mh->cputype; - mh->cpusubtype = orig_mh->cpusubtype; - mh->filetype = orig_mh->filetype; - mh->ncmds = 3; - mh->sizeofcmds = header_size - sizeof(struct mach_header); - mh->flags = orig_mh->flags; - - // Initialise the current file offset and addr - offset = round_page(header_size); - addr = (caddr_t) const_text->addr; // Load address of __TEXT,__const - - /* - * Construct a TEXT segment load command - * the only part of the TEXT segment we keep is the __TEXT,__const - * which contains the kernel vtables. - */ - sg = (struct segment_command *) &mh[1]; - bcopy(orig_ts, sg, orig_ts->cmdsize); - sg->vmaddr = (unsigned long) addr; - sg->vmsize = const_text->size; - sg->fileoff = 0; - sg->filesize = const_text->size + round_page(header_size); - sg->maxprot = 0; - sg->initprot = 0; - sg->flags = 0; - se = (struct section *)(sg+1); - for ( j = 0; j < sg->nsects; j++, se++ ) { - se->addr = (unsigned long) addr; - se->size = 0; - se->offset = offset; - se->nreloc = 0; - if (!strcmp("__const", se->sectname)) { - se->size = const_text->size; - addr += const_text->size; - offset += const_text->size; - const_text = se; - } - } - offset = round_page(offset); - - // Now copy of the __DATA segment load command, the image need - // not be stored to disk nobody needs it, yet! - sg = (struct segment_command *)((int)sg + sg->cmdsize); - bcopy(orig_ds, sg, orig_ds->cmdsize); - - sg->vmaddr = (unsigned long) addr; - sg->vmsize = 0x1000; // One page for some reason? - sg->fileoff = offset; - sg->filesize = 0; - sg->maxprot = 0; - sg->initprot = 0; - sg->flags = 0; - se = (struct section *)(sg+1); - for ( j = 0; j < sg->nsects; j++, se++ ) { - se->addr = (unsigned long) addr; - se->size = 0; - se->offset = offset; - se->nreloc = 0; - } - offset = round_page(offset); - - - /* - * Set up LC_SYMTAB command - */ - st = (struct symtab_command *)((int)sg + sg->cmdsize); - st->cmd = LC_SYMTAB; - st->cmdsize = sizeof(struct symtab_command); - st->symoff = offset; - st->nsyms = orig_st->nsyms; - st->strsize = orig_st->strsize; - st->stroff = offset + st->nsyms * sizeof(struct nlist); - - /* - * Convert the symbol table in place from section references - * to absolute references. - */ - sym = (struct nlist *) orig_le->vmaddr; - for (i = 0; i < st->nsyms; i++, sym++ ) { - if ( (sym->n_type & N_TYPE) == N_SECT) { - sym->n_sect = NO_SECT; - sym->n_type = (sym->n_type & ~N_TYPE) | N_ABS; - } - } +#include - /* - * Write out the load commands at the beginning of the file. - */ - error = vn_rdwr(UIO_WRITE, vp, (caddr_t) mh, header_size, (off_t) 0, - UIO_SYSSPACE32, IO_NODELOCKED|IO_UNIT, cred, (int *) 0, p); - if (error) - goto out; - - /* - * Write out the __TEXT,__const data segment. - */ - error = vn_rdwr(UIO_WRITE, vp, (caddr_t) const_text->addr, - const_text->size, const_text->offset, - UIO_SYSSPACE32, IO_NODELOCKED|IO_UNIT, cred, (int *) 0, p); - if (error) - goto out; - - /* - * Write out kernel symbols - */ - offset = st->nsyms * sizeof(struct nlist) + st->strsize; // symtab size - error = vn_rdwr(UIO_WRITE, vp, - (caddr_t) orig_le->vmaddr, offset, st->symoff, - UIO_SYSSPACE32, IO_NODELOCKED|IO_UNIT, cred, (int *) 0, p); -out: - if (header) - kmem_free(kernel_map, header, header_size); - - if (vp) { - error1 = vnode_close(vp, FWRITE, &context); - if (!error) error = error1; - } - - return(error); -} -/* - * +/* This function is called from kern_sysctl in the current process context; + * it is exported with the System6.0.exports, but this appears to be a legacy + * export, as there are no internal consumers. */ -int get_kernel_symfile(struct proc *p, char **symfile) +int +get_kernel_symfile(__unused proc_t p, __unused char const **symfile); +int +get_kernel_symfile(__unused proc_t p, __unused char const **symfile) { - if (!kernel_symfile_opened) { - kernel_symfile_opened = 1; - error_code = output_kernel_symbols(p); - } - if (!error_code) - *symfile = "\\mach.sym"; - - return error_code; + return KERN_FAILURE; } struct kern_direct_file_io_ref_t { - struct vfs_context context; - struct vnode *vp; + vfs_context_t ctx; + struct vnode * vp; + dev_t device; + uint32_t blksize; + off_t filelength; + char cf; + char pinned; }; -static int file_ioctl(void * p1, void * p2, int theIoctl, caddr_t result) +static int file_ioctl(void * p1, void * p2, u_long theIoctl, caddr_t result) { - dev_t device = (dev_t) p1; + dev_t device = *(dev_t*) p1; return ((*bdevsw[major(device)].d_ioctl) (device, theIoctl, result, S_IFBLK, p2)); } -static int device_ioctl(void * p1, __unused void * p2, int theIoctl, caddr_t result) +static int device_ioctl(void * p1, __unused void * p2, u_long theIoctl, caddr_t result) { return (VNOP_IOCTL(p1, theIoctl, result, 0, p2)); } +static int +kern_ioctl_file_extents(struct kern_direct_file_io_ref_t * ref, u_long theIoctl, off_t offset, off_t end) +{ + int error = 0; + int (*do_ioctl)(void * p1, void * p2, u_long theIoctl, caddr_t result); + void * p1; + void * p2; + uint64_t fileblk; + size_t filechunk; + dk_extent_t extent; + dk_unmap_t unmap; + _dk_cs_pin_t pin; + + bzero(&extent, sizeof(dk_extent_t)); + bzero(&unmap, sizeof(dk_unmap_t)); + bzero(&pin, sizeof(pin)); + if (ref->vp->v_type == VREG) + { + p1 = &ref->device; + p2 = kernproc; + do_ioctl = &file_ioctl; + } + else + { + /* Partition. */ + p1 = ref->vp; + p2 = ref->ctx; + do_ioctl = &device_ioctl; + } + + if (_DKIOCCSPINEXTENT == theIoctl) { + /* Tell CS the image size, so it knows whether to place the subsequent pins SSD/HDD */ + pin.cp_extent.length = end; + pin.cp_flags = _DKIOCCSHIBERNATEIMGSIZE; + (void) do_ioctl(p1, p2, _DKIOCCSPINEXTENT, (caddr_t)&pin); + } else if (_DKIOCCSUNPINEXTENT == theIoctl) { + /* Tell CS hibernation is done, so it can stop blocking overlapping writes */ + pin.cp_flags = _DKIOCCSPINDISCARDBLACKLIST; + (void) do_ioctl(p1, p2, _DKIOCCSUNPINEXTENT, (caddr_t)&pin); + } + + for (; offset < end; offset += filechunk) + { + if (ref->vp->v_type == VREG) + { + daddr64_t blkno; + filechunk = 1*1024*1024*1024; + if (filechunk > (size_t)(end - offset)) + filechunk = (size_t)(end - offset); + error = VNOP_BLOCKMAP(ref->vp, offset, filechunk, &blkno, + &filechunk, NULL, VNODE_WRITE | VNODE_BLOCKMAP_NO_TRACK, NULL); + if (error) break; + if (-1LL == blkno) continue; + fileblk = blkno * ref->blksize; + } + else if ((ref->vp->v_type == VBLK) || (ref->vp->v_type == VCHR)) + { + fileblk = offset; + filechunk = ref->filelength; + } + + if (DKIOCUNMAP == theIoctl) + { + extent.offset = fileblk; + extent.length = filechunk; + unmap.extents = &extent; + unmap.extentsCount = 1; + error = do_ioctl(p1, p2, theIoctl, (caddr_t)&unmap); +// printf("DKIOCUNMAP(%d) 0x%qx, 0x%qx\n", error, extent.offset, extent.length); + } + else if (_DKIOCCSPINEXTENT == theIoctl) + { + pin.cp_extent.offset = fileblk; + pin.cp_extent.length = filechunk; + pin.cp_flags = _DKIOCCSPINFORHIBERNATION; + error = do_ioctl(p1, p2, theIoctl, (caddr_t)&pin); + if (error && (ENOTTY != error)) + { + printf("_DKIOCCSPINEXTENT(%d) 0x%qx, 0x%qx\n", error, pin.cp_extent.offset, pin.cp_extent.length); + } + } + else if (_DKIOCCSUNPINEXTENT == theIoctl) + { + pin.cp_extent.offset = fileblk; + pin.cp_extent.length = filechunk; + pin.cp_flags = _DKIOCCSPINFORHIBERNATION; + error = do_ioctl(p1, p2, theIoctl, (caddr_t)&pin); + if (error && (ENOTTY != error)) + { + printf("_DKIOCCSUNPINEXTENT(%d) 0x%qx, 0x%qx\n", error, pin.cp_extent.offset, pin.cp_extent.length); + } + } + else error = EINVAL; + + if (error) break; + } + return (error); +} + +extern uint32_t freespace_mb(vnode_t vp); + struct kern_direct_file_io_ref_t * kern_open_file_for_direct_io(const char * name, + boolean_t create_file, kern_get_file_extents_callback_t callback, void * callback_ref, - dev_t * device_result, + off_t set_file_size, + off_t fs_free_size, + off_t write_file_offset, + void * write_file_addr, + size_t write_file_len, + dev_t * partition_device_result, + dev_t * image_device_result, uint64_t * partitionbase_result, - uint64_t * maxiocount_result) + uint64_t * maxiocount_result, + uint32_t * oflags) { struct kern_direct_file_io_ref_t * ref; - struct proc *p; - struct ucred *cred; - struct vnode_attr va; - int error; - off_t f_offset; - uint32_t blksize; - uint64_t size; - dev_t device; - off_t maxiocount, count; - - int (*do_ioctl)(void * p1, void * p2, int theIoctl, caddr_t result); - void * p1; - void * p2; + proc_t p; + struct vnode_attr va; + int error; + off_t f_offset; + uint64_t fileblk; + size_t filechunk; + uint64_t physoffset; + dev_t device; + dev_t target = 0; + int isssd = 0; + uint32_t flags = 0; + uint32_t blksize; + off_t maxiocount, count, segcount; + boolean_t locked = FALSE; + int fmode, cmode; + struct nameidata nd; + u_int32_t ndflags; + off_t mpFree; + + int (*do_ioctl)(void * p1, void * p2, u_long theIoctl, caddr_t result); + void * p1 = NULL; + void * p2 = NULL; error = EFAULT; @@ -395,38 +249,81 @@ kern_open_file_for_direct_io(const char * name, goto out; } - ref->vp = NULL; - p = current_proc(); // kernproc; - cred = p->p_ucred; - ref->context.vc_proc = p; - ref->context.vc_ucred = cred; + bzero(ref, sizeof(*ref)); + p = kernproc; + ref->ctx = vfs_context_kernel(); - if ((error = vnode_open(name, (O_CREAT | FWRITE), (0), 0, &ref->vp, &ref->context))) - goto out; + fmode = (create_file) ? (O_CREAT | FWRITE) : FWRITE; + cmode = S_IRUSR | S_IWUSR; + ndflags = NOFOLLOW; + NDINIT(&nd, LOOKUP, OP_OPEN, ndflags, UIO_SYSSPACE, CAST_USER_ADDR_T(name), ref->ctx); + VATTR_INIT(&va); + VATTR_SET(&va, va_mode, cmode); + VATTR_SET(&va, va_dataprotect_flags, VA_DP_RAWENCRYPTED); + VATTR_SET(&va, va_dataprotect_class, PROTECTION_CLASS_D); + if ((error = vn_open_auth(&nd, &fmode, &va))) { + kprintf("vn_open_auth(fmode: %d, cmode: %d) failed with error: %d\n", fmode, cmode, error); + goto out; + } + + ref->vp = nd.ni_vp; + if (ref->vp->v_type == VREG) + { + vnode_lock_spin(ref->vp); + SET(ref->vp->v_flag, VSWAP); + vnode_unlock(ref->vp); + } + + if (write_file_addr && write_file_len) + { + if ((error = kern_write_file(ref, write_file_offset, write_file_addr, write_file_len, IO_SKIP_ENCRYPTION))) { + kprintf("kern_write_file() failed with error: %d\n", error); + goto out; + } + } VATTR_INIT(&va); VATTR_WANTED(&va, va_rdev); VATTR_WANTED(&va, va_fsid); + VATTR_WANTED(&va, va_devid); VATTR_WANTED(&va, va_data_size); + VATTR_WANTED(&va, va_data_alloc); VATTR_WANTED(&va, va_nlink); error = EFAULT; - if (vnode_getattr(ref->vp, &va, &ref->context)) - goto out; + if (vnode_getattr(ref->vp, &va, ref->ctx)) goto out; - kprintf("vp va_rdev major %d minor %d\n", major(va.va_rdev), minor(va.va_rdev)); - kprintf("vp va_fsid major %d minor %d\n", major(va.va_fsid), minor(va.va_fsid)); - kprintf("vp size %qd\n", va.va_data_size); + mpFree = freespace_mb(ref->vp); + mpFree <<= 20; + kprintf("kern_direct_file(%s): vp size %qd, alloc %qd, mp free %qd, keep free %qd\n", + name, va.va_data_size, va.va_data_alloc, mpFree, fs_free_size); if (ref->vp->v_type == VREG) { - /* Don't dump files with links. */ - if (va.va_nlink != 1) - goto out; + /* Don't dump files with links. */ + if (va.va_nlink != 1) goto out; - device = va.va_fsid; - p1 = (void *) device; + device = (VATTR_IS_SUPPORTED(&va, va_devid)) ? va.va_devid : va.va_fsid; + ref->filelength = va.va_data_size; + + p1 = &device; p2 = p; do_ioctl = &file_ioctl; + + if (set_file_size) + { + if (fs_free_size) + { + mpFree += va.va_data_alloc; + if ((mpFree < set_file_size) || ((mpFree - set_file_size) < fs_free_size)) + { + error = ENOSPC; + goto out; + } + } + error = vnode_setsize(ref->vp, set_file_size, IO_NOZEROFILL | IO_NOAUTH, ref->ctx); + if (error) goto out; + ref->filelength = set_file_size; + } } else if ((ref->vp->v_type == VBLK) || (ref->vp->v_type == VCHR)) { @@ -434,22 +331,118 @@ kern_open_file_for_direct_io(const char * name, device = va.va_rdev; p1 = ref->vp; - p2 = &ref->context; + p2 = ref->ctx; do_ioctl = &device_ioctl; } else { /* Don't dump to non-regular files. */ - error = EFAULT; + error = EFAULT; goto out; } + ref->device = device; - // get partition base + // probe for CF + dk_corestorage_info_t cs_info; + memset(&cs_info, 0, sizeof(dk_corestorage_info_t)); + error = do_ioctl(p1, p2, DKIOCCORESTORAGE, (caddr_t)&cs_info); + ref->cf = (error == 0) && (cs_info.flags & DK_CORESTORAGE_ENABLE_HOTFILES); + + // get block size - error = do_ioctl(p1, p2, DKIOCGETBASE, (caddr_t) partitionbase_result); + error = do_ioctl(p1, p2, DKIOCGETBLOCKSIZE, (caddr_t) &ref->blksize); if (error) goto out; + if (ref->vp->v_type != VREG) + { + error = do_ioctl(p1, p2, DKIOCGETBLOCKCOUNT, (caddr_t) &fileblk); + if (error) goto out; + ref->filelength = fileblk * ref->blksize; + } + + // pin logical extents + + error = kern_ioctl_file_extents(ref, _DKIOCCSPINEXTENT, 0, ref->filelength); + if (error && (ENOTTY != error)) goto out; + ref->pinned = (error == 0); + + // generate the block list + + error = do_ioctl(p1, p2, DKIOCLOCKPHYSICALEXTENTS, NULL); + if (error) goto out; + locked = TRUE; + + f_offset = 0; + for (; f_offset < ref->filelength; f_offset += filechunk) + { + if (ref->vp->v_type == VREG) + { + filechunk = 1*1024*1024*1024; + daddr64_t blkno; + + error = VNOP_BLOCKMAP(ref->vp, f_offset, filechunk, &blkno, + &filechunk, NULL, VNODE_WRITE | VNODE_BLOCKMAP_NO_TRACK, NULL); + if (error) goto out; + if (-1LL == blkno) continue; + fileblk = blkno * ref->blksize; + } + else if ((ref->vp->v_type == VBLK) || (ref->vp->v_type == VCHR)) + { + fileblk = f_offset; + filechunk = f_offset ? 0 : ref->filelength; + } + + physoffset = 0; + while (physoffset < filechunk) + { + dk_physical_extent_t getphysreq; + bzero(&getphysreq, sizeof(getphysreq)); + + getphysreq.offset = fileblk + physoffset; + getphysreq.length = (filechunk - physoffset); + error = do_ioctl(p1, p2, DKIOCGETPHYSICALEXTENT, (caddr_t) &getphysreq); + if (error) goto out; + if (!target) + { + target = getphysreq.dev; + } + else if (target != getphysreq.dev) + { + error = ENOTSUP; + goto out; + } +#if HIBFRAGMENT + uint64_t rev; + for (rev = 4096; rev <= getphysreq.length; rev += 4096) + { + callback(callback_ref, getphysreq.offset + getphysreq.length - rev, 4096); + } +#else + callback(callback_ref, getphysreq.offset, getphysreq.length); +#endif + physoffset += getphysreq.length; + } + } + callback(callback_ref, 0ULL, 0ULL); + + if (ref->vp->v_type == VREG) p1 = ⌖ + else + { + p1 = ⌖ + p2 = p; + do_ioctl = &file_ioctl; + } + + // get partition base + + if (partitionbase_result) + { + error = do_ioctl(p1, p2, DKIOCGETBASE, (caddr_t) partitionbase_result); + if (error) + goto out; + } + // get block size & constraints error = do_ioctl(p1, p2, DKIOCGETBLOCKSIZE, (caddr_t) &blksize); @@ -485,14 +478,20 @@ kern_open_file_for_direct_io(const char * name, maxiocount = count; error = do_ioctl(p1, p2, DKIOCGETMAXSEGMENTBYTECOUNTREAD, (caddr_t) &count); + if (!error) + error = do_ioctl(p1, p2, DKIOCGETMAXSEGMENTCOUNTREAD, (caddr_t) &segcount); if (error) - count = 0; + count = segcount = 0; + count *= segcount; if (count && (count < maxiocount)) maxiocount = count; error = do_ioctl(p1, p2, DKIOCGETMAXSEGMENTBYTECOUNTWRITE, (caddr_t) &count); + if (!error) + error = do_ioctl(p1, p2, DKIOCGETMAXSEGMENTCOUNTWRITE, (caddr_t) &segcount); if (error) - count = 0; + count = segcount = 0; + count *= segcount; if (count && (count < maxiocount)) maxiocount = count; @@ -500,72 +499,136 @@ kern_open_file_for_direct_io(const char * name, if (maxiocount_result) *maxiocount_result = maxiocount; - // generate the block list + error = do_ioctl(p1, p2, DKIOCISSOLIDSTATE, (caddr_t)&isssd); + if (!error && isssd) + flags |= kIOPolledFileSSD; - error = 0; - if (ref->vp->v_type == VREG) - { - f_offset = 0; - while(f_offset < (off_t) va.va_data_size) - { - size_t io_size = 1*1024*1024*1024; - daddr64_t blkno; + if (partition_device_result) + *partition_device_result = device; + if (image_device_result) + *image_device_result = target; + if (oflags) + *oflags = flags; - error = VNOP_BLOCKMAP(ref->vp, f_offset, io_size, &blkno, (size_t *)&io_size, NULL, 0, NULL); - if (error) - goto out; - callback(callback_ref, ((uint64_t) blkno) * blksize, (uint64_t) io_size); - f_offset += io_size; - } - callback(callback_ref, 0ULL, 0ULL); - } - else if ((ref->vp->v_type == VBLK) || (ref->vp->v_type == VCHR)) + if ((ref->vp->v_type == VBLK) || (ref->vp->v_type == VCHR)) { - error = do_ioctl(p1, p2, DKIOCGETBLOCKCOUNT, (caddr_t) &size); - if (error) - goto out; - size *= blksize; - callback(callback_ref, 0ULL, size); - callback(callback_ref, size, 0ULL); + vnode_close(ref->vp, FWRITE, ref->ctx); + ref->vp = NULLVP; + ref->ctx = NULL; } - if (device_result) - *device_result = device; - out: - kprintf("kern_open_file_for_direct_io(%d)\n", error); + printf("kern_open_file_for_direct_io(%p, %d)\n", ref, error); - if (error && ref) { - if (ref->vp) - vnode_close(ref->vp, FWRITE, &ref->context); + if (error && locked) + { + p1 = &device; + (void) do_ioctl(p1, p2, DKIOCUNLOCKPHYSICALEXTENTS, NULL); + } + + if (error && ref) + { + if (ref->vp) + { + (void) kern_ioctl_file_extents(ref, _DKIOCCSUNPINEXTENT, 0, (ref->pinned && ref->cf) ? ref->filelength : 0); + vnode_close(ref->vp, FWRITE, ref->ctx); + ref->vp = NULLVP; + } + ref->ctx = NULL; kfree(ref, sizeof(struct kern_direct_file_io_ref_t)); + ref = NULL; } return(ref); } int -kern_write_file(struct kern_direct_file_io_ref_t * ref, off_t offset, caddr_t addr, vm_size_t len) +kern_write_file(struct kern_direct_file_io_ref_t * ref, off_t offset, void * addr, size_t len, int ioflag) { return (vn_rdwr(UIO_WRITE, ref->vp, addr, len, offset, - UIO_SYSSPACE32, IO_SYNC|IO_NODELOCKED|IO_UNIT, - ref->context.vc_ucred, (int *) 0, ref->context.vc_proc)); + UIO_SYSSPACE, ioflag|IO_SYNC|IO_NODELOCKED|IO_UNIT, + vfs_context_ucred(ref->ctx), (int *) 0, + vfs_context_proc(ref->ctx))); +} + +int +kern_read_file(struct kern_direct_file_io_ref_t * ref, off_t offset, void * addr, size_t len, int ioflag) +{ + return (vn_rdwr(UIO_READ, ref->vp, + addr, len, offset, + UIO_SYSSPACE, ioflag|IO_SYNC|IO_NODELOCKED|IO_UNIT, + vfs_context_ucred(ref->ctx), (int *) 0, + vfs_context_proc(ref->ctx))); +} + + +struct mount * +kern_file_mount(struct kern_direct_file_io_ref_t * ref) +{ + return (ref->vp->v_mount); } void -kern_close_file_for_direct_io(struct kern_direct_file_io_ref_t * ref) +kern_close_file_for_direct_io(struct kern_direct_file_io_ref_t * ref, + off_t write_offset, void * addr, size_t write_length, + off_t discard_offset, off_t discard_end) { - kprintf("kern_close_file_for_direct_io\n"); + int error; + printf("kern_close_file_for_direct_io(%p)\n", ref); - if (ref) { - int error; + if (!ref) return; + + if (ref->vp) + { + int (*do_ioctl)(void * p1, void * p2, u_long theIoctl, caddr_t result); + void * p1; + void * p2; + + if (ref->vp->v_type == VREG) + { + p1 = &ref->device; + p2 = kernproc; + do_ioctl = &file_ioctl; + } + else + { + /* Partition. */ + p1 = ref->vp; + p2 = ref->ctx; + do_ioctl = &device_ioctl; + } + (void) do_ioctl(p1, p2, DKIOCUNLOCKPHYSICALEXTENTS, NULL); + + //XXX If unmapping extents then don't also need to unpin; except ... + //XXX if file unaligned (HFS 4k / Fusion 128k) then pin is superset and + //XXX unmap is subset, so save extra walk over file extents (and the risk + //XXX that CF drain starts) vs leaving partial units pinned to SSD + //XXX (until whatever was sharing also unmaps). Err on cleaning up fully. + boolean_t will_unmap = (!ref->pinned || ref->cf) && (discard_end > discard_offset); + boolean_t will_unpin = (ref->pinned && ref->cf /* && !will_unmap */); + + (void) kern_ioctl_file_extents(ref, _DKIOCCSUNPINEXTENT, 0, (will_unpin) ? ref->filelength : 0); + + if (will_unmap) + { + (void) kern_ioctl_file_extents(ref, DKIOCUNMAP, discard_offset, (ref->cf) ? ref->filelength : discard_end); + } + + if (addr && write_length) + { + (void) kern_write_file(ref, write_offset, addr, write_length, IO_SKIP_ENCRYPTION); + } + + error = vnode_close(ref->vp, FWRITE, ref->ctx); + + ref->vp = NULLVP; + kprintf("vnode_close(%d)\n", error); - if (ref->vp) { - error = vnode_close(ref->vp, FWRITE, &ref->context); - kprintf("vnode_close(%d)\n", error); - } - kfree(ref, sizeof(struct kern_direct_file_io_ref_t)); } + + ref->ctx = NULL; + + kfree(ref, sizeof(struct kern_direct_file_io_ref_t)); }