bsd/vm/vm_compressor_backing_file.c

   1 /*
   2  * Copyright (c) 2000-2013 Apple Inc. All rights reserved.
   3  *
   4  * @APPLE_OSREFERENCE_LICENSE_HEADER_START@
   5  *
   6  * This file contains Original Code and/or Modifications of Original Code
   7  * as defined in and that are subject to the Apple Public Source License
   8  * Version 2.0 (the 'License'). You may not use this file except in
   9  * compliance with the License. The rights granted to you under the License
  10  * may not be used to create, or enable the creation or redistribution of,
  11  * unlawful or unlicensed copies of an Apple operating system, or to
  12  * circumvent, violate, or enable the circumvention or violation of, any
  13  * terms of an Apple operating system software license agreement.
  14  *
  15  * Please obtain a copy of the License at
  16  * http://www.opensource.apple.com/apsl/ and read it before using this file.
  17  *
  18  * The Original Code and all software distributed under the License are
  19  * distributed on an 'AS IS' basis, WITHOUT WARRANTY OF ANY KIND, EITHER
  20  * EXPRESS OR IMPLIED, AND APPLE HEREBY DISCLAIMS ALL SUCH WARRANTIES,
  21  * INCLUDING WITHOUT LIMITATION, ANY WARRANTIES OF MERCHANTABILITY,
  22  * FITNESS FOR A PARTICULAR PURPOSE, QUIET ENJOYMENT OR NON-INFRINGEMENT.
  23  * Please see the License for the specific language governing rights and
  24  * limitations under the License.
  25  *
  26  * @APPLE_OSREFERENCE_LICENSE_HEADER_END@
  27  */
  28
  29 #include <stdint.h>
  30 #include <sys/fcntl.h>
  31 #include <sys/vnode_internal.h>
  32 #include <sys/vnode.h>
  33 #include <sys/kauth.h>
  34 #include <sys/mount_internal.h>
  35 #include <sys/buf_internal.h>
  36 #include <kern/debug.h>
  37 #include <kern/kalloc.h>
  38 #include <sys/cprotect.h>
  39 #include <sys/disk.h>
  40 #include <vm/vm_protos.h>
  41 #include <vm/vm_pageout.h>
  42
  43 void vm_swapfile_open(const char *path, vnode_t *vp);
  44 void vm_swapfile_close(uint64_t path, vnode_t vp);
  45 int vm_swapfile_preallocate(vnode_t vp, uint64_t *size);
  46 uint64_t vm_swapfile_get_blksize(vnode_t vp);
  47 uint64_t vm_swapfile_get_transfer_size(vnode_t vp);
  48 int vm_swapfile_io(vnode_t vp, uint64_t offset, uint64_t start, int npages, int flags);
  49
  50 void
  51 vm_swapfile_open(const char *path, vnode_t *vp)
  52 {
  53         int error = 0;
  54         vfs_context_t   ctx = vfs_context_current();
  55
  56         if ((error = vnode_open(path, (O_CREAT | O_TRUNC | FREAD | FWRITE), S_IRUSR | S_IWUSR, 0, vp, ctx))) {
  57                 printf("Failed to open swap file %d\n", error);
  58                 *vp = NULL;
  59                 return;
  60         }
  61
  62         vnode_put(*vp);
  63 }
  64
  65 uint64_t
  66 vm_swapfile_get_blksize(vnode_t vp)
  67 {
  68         return ((uint64_t)vfs_devblocksize(vnode_mount(vp)));
  69 }
  70
  71 uint64_t
  72 vm_swapfile_get_transfer_size(vnode_t vp)
  73 {
  74         return((uint64_t)vp->v_mount->mnt_vfsstat.f_iosize);
  75 }
  76
  77 int unlink1(vfs_context_t, vnode_t, user_addr_t, enum uio_seg, int);
  78
  79 void
  80 vm_swapfile_close(uint64_t path_addr, vnode_t vp)
  81 {
  82         vfs_context_t context = vfs_context_current();
  83         int error;
  84
  85         vnode_getwithref(vp);
  86         vnode_close(vp, 0, context);
  87
  88         error = unlink1(context, NULLVP, CAST_USER_ADDR_T(path_addr),
  89             UIO_SYSSPACE, 0);
  90
  91 #if DEVELOPMENT || DEBUG
  92         if (error)
  93                 printf("%s : unlink of %s failed with error %d", __FUNCTION__,
  94                     (char *)path_addr, error);
  95 #endif
  96 }
  97
  98 int
  99 vm_swapfile_preallocate(vnode_t vp, uint64_t *size)
 100 {
 101         int             error = 0;
 102         uint64_t        file_size = 0;
 103         vfs_context_t   ctx = NULL;
 104
 105
 106         ctx = vfs_context_current();
 107
 108 #if CONFIG_PROTECT
 109         {
 110 #if 0   // <rdar://11771612>
 111
 112                 if ((error = cp_vnode_setclass(vp, PROTECTION_CLASS_F))) {
 113                         if(config_protect_bug) {
 114                                 printf("swap protection class set failed with %d\n", error);
 115                         } else {
 116                                 panic("swap protection class set failed with %d\n", error);
 117                         }
 118                 }
 119 #endif
 120                 /* initialize content protection keys manually */
 121                 if ((error = cp_handle_vnop(vp, CP_WRITE_ACCESS, 0)) != 0) {
 122                         printf("Content Protection key failure on swap: %d\n", error);
 123                         vnode_put(vp);
 124                         vp = NULL;
 125                         goto done;
 126                 }
 127         }
 128 #endif
 129
 130         error = vnode_setsize(vp, *size, IO_NOZEROFILL, ctx);
 131
 132         if (error) {
 133                 printf("vnode_setsize for swap files failed: %d\n", error);
 134                 goto done;
 135         }
 136
 137         error = vnode_size(vp, (off_t*) &file_size, ctx);
 138
 139         if (error) {
 140                 printf("vnode_size (new file) for swap file failed: %d\n", error);
 141         }
 142
 143         assert(file_size == *size);
 144
 145         vnode_lock_spin(vp);
 146         SET(vp->v_flag, VSWAP);
 147         vnode_unlock(vp);
 148 done:
 149         return error;
 150 }
 151
 152 int
 153 vm_swapfile_io(vnode_t vp, uint64_t offset, uint64_t start, int npages, int flags)
 154 {
 155         int error = 0;
 156         uint64_t io_size = npages * PAGE_SIZE_64;
 157 #if 1
 158         kern_return_t   kr = KERN_SUCCESS;
 159         upl_t           upl = NULL;
 160         unsigned int    count = 0;
 161         int             upl_create_flags = 0, upl_control_flags = 0;
 162         upl_size_t      upl_size = 0;
 163
 164         upl_create_flags = UPL_SET_INTERNAL | UPL_SET_LITE;
 165
 166 #if ENCRYPTED_SWAP
 167         upl_control_flags = UPL_IOSYNC | UPL_PAGING_ENCRYPTED;
 168 #else
 169         upl_control_flags = UPL_IOSYNC;
 170 #endif
 171         if ((flags & SWAP_READ) == FALSE) {
 172                 upl_create_flags |= UPL_COPYOUT_FROM;
 173         }
 174
 175         upl_size = io_size;
 176         kr = vm_map_create_upl( kernel_map,
 177                                 start,
 178                                 &upl_size,
 179                                 &upl,
 180                                 NULL,
 181                                 &count,
 182                                 &upl_create_flags);
 183
 184         if (kr != KERN_SUCCESS || (upl_size != io_size)) {
 185                 panic("vm_map_create_upl failed with %d\n", kr);
 186         }
 187
 188         if (flags & SWAP_READ) {
 189                 vnode_pagein(vp,
 190                               upl,
 191                               0,
 192                               offset,
 193                               io_size,
 194                               upl_control_flags | UPL_IGNORE_VALID_PAGE_CHECK,
 195                               &error);
 196                 if (error) {
 197 #if DEBUG
 198                         printf("vm_swapfile_io: vnode_pagein failed with %d (vp: %p, offset: 0x%llx, size:%llu)\n", error, vp, offset, io_size);
 199 #else /* DEBUG */
 200                         printf("vm_swapfile_io: vnode_pagein failed with %d.\n", error);
 201 #endif /* DEBUG */
 202                 }
 203
 204         } else {
 205                 vnode_pageout(vp,
 206                               upl,
 207                               0,
 208                               offset,
 209                               io_size,
 210                               upl_control_flags,
 211                               &error);
 212                 if (error) {
 213 #if DEBUG
 214                         printf("vm_swapfile_io: vnode_pageout failed with %d (vp: %p, offset: 0x%llx, size:%llu)\n", error, vp, offset, io_size);
 215 #else /* DEBUG */
 216                         printf("vm_swapfile_io: vnode_pageout failed with %d.\n", error);
 217 #endif /* DEBUG */
 218                 }
 219         }
 220         return error;
 221
 222 #else /* 1 */
 223         vfs_context_t ctx;
 224         ctx = vfs_context_kernel();
 225
 226         error = vn_rdwr((flags & SWAP_READ) ? UIO_READ : UIO_WRITE, vp, (caddr_t)start, io_size, offset,
 227                 UIO_SYSSPACE, IO_SYNC | IO_NODELOCKED | IO_UNIT | IO_NOCACHE | IO_SWAP_DISPATCH, vfs_context_ucred(ctx), (int *) 0, vfs_context_proc(ctx));
 228
 229         if (error) {
 230                 printf("vn_rdwr: Swap I/O failed with %d\n", error);
 231         }
 232         return error;
 233 #endif /* 1 */
 234 }
 235
 236
 237 #define MAX_BATCH_TO_TRIM       256
 238
 239 #define ROUTE_ONLY              0x10            /* if corestorage is present, tell it to just pass */
 240                                                 /* the DKIOUNMAP command through w/o acting on it */
 241                                                 /* this is used by the compressed swap system to reclaim empty space */
 242
 243
 244 u_int32_t vnode_trim_list (vnode_t vp, struct trim_list *tl, boolean_t route_only)
 245 {
 246         int             error = 0;
 247         int             trim_index = 0;
 248         u_int32_t       blocksize = 0;
 249         struct vnode    *devvp;
 250         dk_extent_t     *extents;
 251         dk_unmap_t      unmap;
 252         _dk_cs_unmap_t  cs_unmap;
 253
 254         if ( !(vp->v_mount->mnt_ioflags & MNT_IOFLAGS_UNMAP_SUPPORTED))
 255                 return (ENOTSUP);
 256
 257         if (tl == NULL)
 258                 return (0);
 259
 260         /*
 261          * Get the underlying device vnode and physical block size
 262          */
 263         devvp = vp->v_mount->mnt_devvp;
 264         blocksize = vp->v_mount->mnt_devblocksize;
 265
 266         extents = kalloc(sizeof(dk_extent_t) * MAX_BATCH_TO_TRIM);
 267
 268         if (vp->v_mount->mnt_ioflags & MNT_IOFLAGS_CSUNMAP_SUPPORTED) {
 269                 memset (&cs_unmap, 0, sizeof(_dk_cs_unmap_t));
 270                 cs_unmap.extents = extents;
 271
 272                 if (route_only == TRUE)
 273                         cs_unmap.options = ROUTE_ONLY;
 274         } else {
 275                 memset (&unmap, 0, sizeof(dk_unmap_t));
 276                 unmap.extents = extents;
 277         }
 278
 279         while (tl) {
 280                 daddr64_t       io_blockno;     /* Block number corresponding to the start of the extent */
 281                 size_t          io_bytecount;   /* Number of bytes in current extent for the specified range */
 282                 size_t          trimmed;
 283                 size_t          remaining_length;
 284                 off_t           current_offset;
 285
 286                 current_offset = tl->tl_offset;
 287                 remaining_length = tl->tl_length;
 288                 trimmed = 0;
 289
 290                 /*
 291                  * We may not get the entire range from tl_offset -> tl_offset+tl_length in a single
 292                  * extent from the blockmap call.  Keep looping/going until we are sure we've hit
 293                  * the whole range or if we encounter an error.
 294                  */
 295                 while (trimmed < tl->tl_length) {
 296                         /*
 297                          * VNOP_BLOCKMAP will tell us the logical to physical block number mapping for the
 298                          * specified offset.  It returns blocks in contiguous chunks, so if the logical range is
 299                          * broken into multiple extents, it must be called multiple times, increasing the offset
 300                          * in each call to ensure that the entire range is covered.
 301                          */
 302                         error = VNOP_BLOCKMAP (vp, current_offset, remaining_length,
 303                                                &io_blockno, &io_bytecount, NULL, VNODE_READ, NULL);
 304
 305                         if (error) {
 306                                 goto trim_exit;
 307                         }
 308
 309                         extents[trim_index].offset = (uint64_t) io_blockno * (u_int64_t) blocksize;
 310                         extents[trim_index].length = io_bytecount;
 311
 312                         trim_index++;
 313
 314                         if (trim_index == MAX_BATCH_TO_TRIM) {
 315
 316                                 if (vp->v_mount->mnt_ioflags & MNT_IOFLAGS_CSUNMAP_SUPPORTED) {
 317                                         cs_unmap.extentsCount = trim_index;
 318                                         error = VNOP_IOCTL(devvp, _DKIOCCSUNMAP, (caddr_t)&cs_unmap, 0, vfs_context_kernel());
 319                                 } else {
 320                                         unmap.extentsCount = trim_index;
 321                                         error = VNOP_IOCTL(devvp, DKIOCUNMAP, (caddr_t)&unmap, 0, vfs_context_kernel());
 322                                 }
 323                                 if (error) {
 324                                         goto trim_exit;
 325                                 }
 326                                 trim_index = 0;
 327                         }
 328                         trimmed += io_bytecount;
 329                         current_offset += io_bytecount;
 330                         remaining_length -= io_bytecount;
 331                 }
 332                 tl = tl->tl_next;
 333         }
 334         if (trim_index) {
 335                 if (vp->v_mount->mnt_ioflags & MNT_IOFLAGS_CSUNMAP_SUPPORTED) {
 336                         cs_unmap.extentsCount = trim_index;
 337                         error = VNOP_IOCTL(devvp, _DKIOCCSUNMAP, (caddr_t)&cs_unmap, 0, vfs_context_kernel());
 338                 } else {
 339                         unmap.extentsCount = trim_index;
 340                         error = VNOP_IOCTL(devvp, DKIOCUNMAP, (caddr_t)&unmap, 0, vfs_context_kernel());
 341                 }
 342         }
 343 trim_exit:
 344         kfree(extents, sizeof(dk_extent_t) * MAX_BATCH_TO_TRIM);
 345
 346         return error;
 347 }