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, boolean_t *pin);
  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 int vm_record_file_write(struct vnode *vp, uint64_t offset, char *buf, int size);
  50
  51
  52 void
  53 vm_swapfile_open(const char *path, vnode_t *vp)
  54 {
  55         int error = 0;
  56         vfs_context_t   ctx = vfs_context_current();
  57
  58         if ((error = vnode_open(path, (O_CREAT | O_TRUNC | FREAD | FWRITE), S_IRUSR | S_IWUSR, 0, vp, ctx))) {
  59                 printf("Failed to open swap file %d\n", error);
  60                 *vp = NULL;
  61                 return;
  62         }
  63
  64         vnode_put(*vp);
  65 }
  66
  67 uint64_t
  68 vm_swapfile_get_blksize(vnode_t vp)
  69 {
  70         return ((uint64_t)vfs_devblocksize(vnode_mount(vp)));
  71 }
  72
  73 uint64_t
  74 vm_swapfile_get_transfer_size(vnode_t vp)
  75 {
  76         return((uint64_t)vp->v_mount->mnt_vfsstat.f_iosize);
  77 }
  78
  79 int unlink1(vfs_context_t, vnode_t, user_addr_t, enum uio_seg, int);
  80
  81 void
  82 vm_swapfile_close(uint64_t path_addr, vnode_t vp)
  83 {
  84         vfs_context_t context = vfs_context_current();
  85         int error;
  86
  87         vnode_getwithref(vp);
  88         vnode_close(vp, 0, context);
  89
  90         error = unlink1(context, NULLVP, CAST_USER_ADDR_T(path_addr),
  91             UIO_SYSSPACE, 0);
  92
  93 #if DEVELOPMENT || DEBUG
  94         if (error)
  95                 printf("%s : unlink of %s failed with error %d", __FUNCTION__,
  96                     (char *)path_addr, error);
  97 #endif
  98 }
  99
 100 int
 101 vm_swapfile_preallocate(vnode_t vp, uint64_t *size, boolean_t *pin)
 102 {
 103         int             error = 0;
 104         uint64_t        file_size = 0;
 105         vfs_context_t   ctx = NULL;
 106
 107
 108         ctx = vfs_context_current();
 109
 110         error = vnode_setsize(vp, *size, IO_NOZEROFILL, ctx);
 111
 112         if (error) {
 113                 printf("vnode_setsize for swap files failed: %d\n", error);
 114                 goto done;
 115         }
 116
 117         error = vnode_size(vp, (off_t*) &file_size, ctx);
 118
 119         if (error) {
 120                 printf("vnode_size (new file) for swap file failed: %d\n", error);
 121                 goto done;
 122         }
 123         assert(file_size == *size);
 124
 125         if (pin != NULL && *pin != FALSE) {
 126                 error = VNOP_IOCTL(vp, FIOPINSWAP, NULL, 0, ctx);
 127
 128                 if (error) {
 129                         printf("pin for swap files failed: %d,  file_size = %lld\n", error, file_size);
 130                         /* this is not fatal, carry on with files wherever they landed */
 131                         *pin = FALSE;
 132                         error = 0;
 133                 }
 134         }
 135
 136         vnode_lock_spin(vp);
 137         SET(vp->v_flag, VSWAP);
 138         vnode_unlock(vp);
 139 done:
 140         return error;
 141 }
 142
 143
 144 int
 145 vm_record_file_write(vnode_t vp, uint64_t offset, char *buf, int size)
 146 {
 147         int error = 0;
 148         vfs_context_t ctx;
 149
 150         ctx = vfs_context_kernel();
 151
 152         error = vn_rdwr(UIO_WRITE, vp, (caddr_t)buf, size, offset,
 153                 UIO_SYSSPACE, IO_NODELOCKED, vfs_context_ucred(ctx), (int *) 0, vfs_context_proc(ctx));
 154
 155         return (error);
 156 }
 157
 158
 159
 160 int
 161 vm_swapfile_io(vnode_t vp, uint64_t offset, uint64_t start, int npages, int flags)
 162 {
 163         int error = 0;
 164         uint64_t io_size = npages * PAGE_SIZE_64;
 165 #if 1
 166         kern_return_t   kr = KERN_SUCCESS;
 167         upl_t           upl = NULL;
 168         unsigned int    count = 0;
 169         upl_control_flags_t upl_create_flags = 0;
 170         int             upl_control_flags = 0;
 171         upl_size_t      upl_size = 0;
 172
 173         upl_create_flags = UPL_SET_INTERNAL | UPL_SET_LITE
 174                         | UPL_MEMORY_TAG_MAKE(VM_KERN_MEMORY_OSFMK);
 175
 176 #if ENCRYPTED_SWAP
 177         upl_control_flags = UPL_IOSYNC | UPL_PAGING_ENCRYPTED;
 178 #else
 179         upl_control_flags = UPL_IOSYNC;
 180 #endif
 181         if ((flags & SWAP_READ) == FALSE) {
 182                 upl_create_flags |= UPL_COPYOUT_FROM;
 183         }
 184
 185         upl_size = io_size;
 186         kr = vm_map_create_upl( kernel_map,
 187                                 start,
 188                                 &upl_size,
 189                                 &upl,
 190                                 NULL,
 191                                 &count,
 192                                 &upl_create_flags);
 193
 194         if (kr != KERN_SUCCESS || (upl_size != io_size)) {
 195                 panic("vm_map_create_upl failed with %d\n", kr);
 196         }
 197
 198         if (flags & SWAP_READ) {
 199                 vnode_pagein(vp,
 200                               upl,
 201                               0,
 202                               offset,
 203                               io_size,
 204                               upl_control_flags | UPL_IGNORE_VALID_PAGE_CHECK,
 205                               &error);
 206                 if (error) {
 207 #if DEBUG
 208                         printf("vm_swapfile_io: vnode_pagein failed with %d (vp: %p, offset: 0x%llx, size:%llu)\n", error, vp, offset, io_size);
 209 #else /* DEBUG */
 210                         printf("vm_swapfile_io: vnode_pagein failed with %d.\n", error);
 211 #endif /* DEBUG */
 212                 }
 213
 214         } else {
 215                 vnode_pageout(vp,
 216                               upl,
 217                               0,
 218                               offset,
 219                               io_size,
 220                               upl_control_flags,
 221                               &error);
 222                 if (error) {
 223 #if DEBUG
 224                         printf("vm_swapfile_io: vnode_pageout failed with %d (vp: %p, offset: 0x%llx, size:%llu)\n", error, vp, offset, io_size);
 225 #else /* DEBUG */
 226                         printf("vm_swapfile_io: vnode_pageout failed with %d.\n", error);
 227 #endif /* DEBUG */
 228                 }
 229         }
 230         return error;
 231
 232 #else /* 1 */
 233         vfs_context_t ctx;
 234         ctx = vfs_context_kernel();
 235
 236         error = vn_rdwr((flags & SWAP_READ) ? UIO_READ : UIO_WRITE, vp, (caddr_t)start, io_size, offset,
 237                 UIO_SYSSPACE, IO_SYNC | IO_NODELOCKED | IO_UNIT | IO_NOCACHE | IO_SWAP_DISPATCH, vfs_context_ucred(ctx), (int *) 0, vfs_context_proc(ctx));
 238
 239         if (error) {
 240                 printf("vn_rdwr: Swap I/O failed with %d\n", error);
 241         }
 242         return error;
 243 #endif /* 1 */
 244 }
 245
 246
 247 #define MAX_BATCH_TO_TRIM       256
 248
 249 #define ROUTE_ONLY              0x10            /* if corestorage is present, tell it to just pass */
 250                                                 /* the DKIOUNMAP command through w/o acting on it */
 251                                                 /* this is used by the compressed swap system to reclaim empty space */
 252
 253
 254 u_int32_t vnode_trim_list (vnode_t vp, struct trim_list *tl, boolean_t route_only)
 255 {
 256         int             error = 0;
 257         int             trim_index = 0;
 258         u_int32_t       blocksize = 0;
 259         struct vnode    *devvp;
 260         dk_extent_t     *extents;
 261         dk_unmap_t      unmap;
 262         _dk_cs_unmap_t  cs_unmap;
 263
 264         if ( !(vp->v_mount->mnt_ioflags & MNT_IOFLAGS_UNMAP_SUPPORTED))
 265                 return (ENOTSUP);
 266
 267         if (tl == NULL)
 268                 return (0);
 269
 270         /*
 271          * Get the underlying device vnode and physical block size
 272          */
 273         devvp = vp->v_mount->mnt_devvp;
 274         blocksize = vp->v_mount->mnt_devblocksize;
 275
 276         extents = kalloc(sizeof(dk_extent_t) * MAX_BATCH_TO_TRIM);
 277
 278         if (vp->v_mount->mnt_ioflags & MNT_IOFLAGS_CSUNMAP_SUPPORTED) {
 279                 memset (&cs_unmap, 0, sizeof(_dk_cs_unmap_t));
 280                 cs_unmap.extents = extents;
 281
 282                 if (route_only == TRUE)
 283                         cs_unmap.options = ROUTE_ONLY;
 284         } else {
 285                 memset (&unmap, 0, sizeof(dk_unmap_t));
 286                 unmap.extents = extents;
 287         }
 288
 289         while (tl) {
 290                 daddr64_t       io_blockno;     /* Block number corresponding to the start of the extent */
 291                 size_t          io_bytecount;   /* Number of bytes in current extent for the specified range */
 292                 size_t          trimmed;
 293                 size_t          remaining_length;
 294                 off_t           current_offset;
 295
 296                 current_offset = tl->tl_offset;
 297                 remaining_length = tl->tl_length;
 298                 trimmed = 0;
 299
 300                 /*
 301                  * We may not get the entire range from tl_offset -> tl_offset+tl_length in a single
 302                  * extent from the blockmap call.  Keep looping/going until we are sure we've hit
 303                  * the whole range or if we encounter an error.
 304                  */
 305                 while (trimmed < tl->tl_length) {
 306                         /*
 307                          * VNOP_BLOCKMAP will tell us the logical to physical block number mapping for the
 308                          * specified offset.  It returns blocks in contiguous chunks, so if the logical range is
 309                          * broken into multiple extents, it must be called multiple times, increasing the offset
 310                          * in each call to ensure that the entire range is covered.
 311                          */
 312                         error = VNOP_BLOCKMAP (vp, current_offset, remaining_length,
 313                                                &io_blockno, &io_bytecount, NULL, VNODE_READ, NULL);
 314
 315                         if (error) {
 316                                 goto trim_exit;
 317                         }
 318
 319                         extents[trim_index].offset = (uint64_t) io_blockno * (u_int64_t) blocksize;
 320                         extents[trim_index].length = io_bytecount;
 321
 322                         trim_index++;
 323
 324                         if (trim_index == MAX_BATCH_TO_TRIM) {
 325
 326                                 if (vp->v_mount->mnt_ioflags & MNT_IOFLAGS_CSUNMAP_SUPPORTED) {
 327                                         cs_unmap.extentsCount = trim_index;
 328                                         error = VNOP_IOCTL(devvp, _DKIOCCSUNMAP, (caddr_t)&cs_unmap, 0, vfs_context_kernel());
 329                                 } else {
 330                                         unmap.extentsCount = trim_index;
 331                                         error = VNOP_IOCTL(devvp, DKIOCUNMAP, (caddr_t)&unmap, 0, vfs_context_kernel());
 332                                 }
 333                                 if (error) {
 334                                         goto trim_exit;
 335                                 }
 336                                 trim_index = 0;
 337                         }
 338                         trimmed += io_bytecount;
 339                         current_offset += io_bytecount;
 340                         remaining_length -= io_bytecount;
 341                 }
 342                 tl = tl->tl_next;
 343         }
 344         if (trim_index) {
 345                 if (vp->v_mount->mnt_ioflags & MNT_IOFLAGS_CSUNMAP_SUPPORTED) {
 346                         cs_unmap.extentsCount = trim_index;
 347                         error = VNOP_IOCTL(devvp, _DKIOCCSUNMAP, (caddr_t)&cs_unmap, 0, vfs_context_kernel());
 348                 } else {
 349                         unmap.extentsCount = trim_index;
 350                         error = VNOP_IOCTL(devvp, DKIOCUNMAP, (caddr_t)&unmap, 0, vfs_context_kernel());
 351                 }
 352         }
 353 trim_exit:
 354         kfree(extents, sizeof(dk_extent_t) * MAX_BATCH_TO_TRIM);
 355
 356         return error;
 357 }