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 #include <hfs/hfs.h>
  43
  44 void vm_swapfile_open(const char *path, vnode_t *vp);
  45 void vm_swapfile_close(uint64_t path, vnode_t vp);
  46 int vm_swapfile_preallocate(vnode_t vp, uint64_t *size, boolean_t *pin);
  47 uint64_t vm_swapfile_get_blksize(vnode_t vp);
  48 uint64_t vm_swapfile_get_transfer_size(vnode_t vp);
  49 int vm_swapfile_io(vnode_t vp, uint64_t offset, uint64_t start, int npages, int flags);
  50 int vm_record_file_write(struct vnode *vp, uint64_t offset, char *buf, int size);
  51
  52
  53 void
  54 vm_swapfile_open(const char *path, vnode_t *vp)
  55 {
  56         int error = 0;
  57         vfs_context_t   ctx = vfs_context_current();
  58
  59         if ((error = vnode_open(path, (O_CREAT | O_TRUNC | FREAD | FWRITE), S_IRUSR | S_IWUSR, 0, vp, ctx))) {
  60                 printf("Failed to open swap file %d\n", error);
  61                 *vp = NULL;
  62                 return;
  63         }
  64
  65         vnode_put(*vp);
  66 }
  67
  68 uint64_t
  69 vm_swapfile_get_blksize(vnode_t vp)
  70 {
  71         return ((uint64_t)vfs_devblocksize(vnode_mount(vp)));
  72 }
  73
  74 uint64_t
  75 vm_swapfile_get_transfer_size(vnode_t vp)
  76 {
  77         return((uint64_t)vp->v_mount->mnt_vfsstat.f_iosize);
  78 }
  79
  80 int unlink1(vfs_context_t, vnode_t, user_addr_t, enum uio_seg, int);
  81
  82 void
  83 vm_swapfile_close(uint64_t path_addr, vnode_t vp)
  84 {
  85         vfs_context_t context = vfs_context_current();
  86         int error;
  87
  88         vnode_getwithref(vp);
  89         vnode_close(vp, 0, context);
  90
  91         error = unlink1(context, NULLVP, CAST_USER_ADDR_T(path_addr),
  92             UIO_SYSSPACE, 0);
  93
  94 #if DEVELOPMENT || DEBUG
  95         if (error)
  96                 printf("%s : unlink of %s failed with error %d", __FUNCTION__,
  97                     (char *)path_addr, error);
  98 #endif
  99 }
 100
 101 int
 102 vm_swapfile_preallocate(vnode_t vp, uint64_t *size, boolean_t *pin)
 103 {
 104         int             error = 0;
 105         uint64_t        file_size = 0;
 106         vfs_context_t   ctx = NULL;
 107
 108
 109         ctx = vfs_context_current();
 110
 111 #if CONFIG_PROTECT
 112         {
 113 #if 0   // <rdar://11771612>
 114
 115                 if ((error = cp_vnode_setclass(vp, PROTECTION_CLASS_F))) {
 116                         if(config_protect_bug) {
 117                                 printf("swap protection class set failed with %d\n", error);
 118                         } else {
 119                                 panic("swap protection class set failed with %d\n", error);
 120                         }
 121                 }
 122 #endif
 123                 /* initialize content protection keys manually */
 124                 if ((error = cp_handle_vnop(vp, CP_WRITE_ACCESS, 0)) != 0) {
 125                         printf("Content Protection key failure on swap: %d\n", error);
 126                         vnode_put(vp);
 127                         vp = NULL;
 128                         goto done;
 129                 }
 130         }
 131 #endif
 132         error = vnode_setsize(vp, *size, IO_NOZEROFILL, ctx);
 133
 134         if (error) {
 135                 printf("vnode_setsize for swap files failed: %d\n", error);
 136                 goto done;
 137         }
 138
 139         error = vnode_size(vp, (off_t*) &file_size, ctx);
 140
 141         if (error) {
 142                 printf("vnode_size (new file) for swap file failed: %d\n", error);
 143                 goto done;
 144         }
 145         assert(file_size == *size);
 146
 147         if (pin != NULL && *pin != FALSE) {
 148
 149                 assert(vnode_tag(vp) == VT_HFS);
 150
 151                 error = hfs_pin_vnode(VTOHFS(vp), vp, HFS_PIN_IT | HFS_DATALESS_PIN, NULL, ctx);
 152
 153                 if (error) {
 154                         printf("hfs_pin_vnode for swap files failed: %d\n", error);
 155                         /* this is not fatal, carry on with files wherever they landed */
 156                         *pin = FALSE;
 157                         error = 0;
 158                 }
 159         }
 160
 161         vnode_lock_spin(vp);
 162         SET(vp->v_flag, VSWAP);
 163         vnode_unlock(vp);
 164 done:
 165         return error;
 166 }
 167
 168
 169 int
 170 vm_record_file_write(vnode_t vp, uint64_t offset, char *buf, int size)
 171 {
 172         int error = 0;
 173         vfs_context_t ctx;
 174
 175         ctx = vfs_context_kernel();
 176
 177         error = vn_rdwr(UIO_WRITE, vp, (caddr_t)buf, size, offset,
 178                 UIO_SYSSPACE, IO_NODELOCKED, vfs_context_ucred(ctx), (int *) 0, vfs_context_proc(ctx));
 179
 180         return (error);
 181 }
 182
 183
 184
 185 int
 186 vm_swapfile_io(vnode_t vp, uint64_t offset, uint64_t start, int npages, int flags)
 187 {
 188         int error = 0;
 189         uint64_t io_size = npages * PAGE_SIZE_64;
 190 #if 1
 191         kern_return_t   kr = KERN_SUCCESS;
 192         upl_t           upl = NULL;
 193         unsigned int    count = 0;
 194         upl_control_flags_t upl_create_flags = 0;
 195         int             upl_control_flags = 0;
 196         upl_size_t      upl_size = 0;
 197
 198         upl_create_flags = UPL_SET_INTERNAL | UPL_SET_LITE
 199                         | UPL_MEMORY_TAG_MAKE(VM_KERN_MEMORY_OSFMK);
 200
 201 #if ENCRYPTED_SWAP
 202         upl_control_flags = UPL_IOSYNC | UPL_PAGING_ENCRYPTED;
 203 #else
 204         upl_control_flags = UPL_IOSYNC;
 205 #endif
 206         if ((flags & SWAP_READ) == FALSE) {
 207                 upl_create_flags |= UPL_COPYOUT_FROM;
 208         }
 209
 210         upl_size = io_size;
 211         kr = vm_map_create_upl( kernel_map,
 212                                 start,
 213                                 &upl_size,
 214                                 &upl,
 215                                 NULL,
 216                                 &count,
 217                                 &upl_create_flags);
 218
 219         if (kr != KERN_SUCCESS || (upl_size != io_size)) {
 220                 panic("vm_map_create_upl failed with %d\n", kr);
 221         }
 222
 223         if (flags & SWAP_READ) {
 224                 vnode_pagein(vp,
 225                               upl,
 226                               0,
 227                               offset,
 228                               io_size,
 229                               upl_control_flags | UPL_IGNORE_VALID_PAGE_CHECK,
 230                               &error);
 231                 if (error) {
 232 #if DEBUG
 233                         printf("vm_swapfile_io: vnode_pagein failed with %d (vp: %p, offset: 0x%llx, size:%llu)\n", error, vp, offset, io_size);
 234 #else /* DEBUG */
 235                         printf("vm_swapfile_io: vnode_pagein failed with %d.\n", error);
 236 #endif /* DEBUG */
 237                 }
 238
 239         } else {
 240                 vnode_pageout(vp,
 241                               upl,
 242                               0,
 243                               offset,
 244                               io_size,
 245                               upl_control_flags,
 246                               &error);
 247                 if (error) {
 248 #if DEBUG
 249                         printf("vm_swapfile_io: vnode_pageout failed with %d (vp: %p, offset: 0x%llx, size:%llu)\n", error, vp, offset, io_size);
 250 #else /* DEBUG */
 251                         printf("vm_swapfile_io: vnode_pageout failed with %d.\n", error);
 252 #endif /* DEBUG */
 253                 }
 254         }
 255         return error;
 256
 257 #else /* 1 */
 258         vfs_context_t ctx;
 259         ctx = vfs_context_kernel();
 260
 261         error = vn_rdwr((flags & SWAP_READ) ? UIO_READ : UIO_WRITE, vp, (caddr_t)start, io_size, offset,
 262                 UIO_SYSSPACE, IO_SYNC | IO_NODELOCKED | IO_UNIT | IO_NOCACHE | IO_SWAP_DISPATCH, vfs_context_ucred(ctx), (int *) 0, vfs_context_proc(ctx));
 263
 264         if (error) {
 265                 printf("vn_rdwr: Swap I/O failed with %d\n", error);
 266         }
 267         return error;
 268 #endif /* 1 */
 269 }
 270
 271
 272 #define MAX_BATCH_TO_TRIM       256
 273
 274 #define ROUTE_ONLY              0x10            /* if corestorage is present, tell it to just pass */
 275                                                 /* the DKIOUNMAP command through w/o acting on it */
 276                                                 /* this is used by the compressed swap system to reclaim empty space */
 277
 278
 279 u_int32_t vnode_trim_list (vnode_t vp, struct trim_list *tl, boolean_t route_only)
 280 {
 281         int             error = 0;
 282         int             trim_index = 0;
 283         u_int32_t       blocksize = 0;
 284         struct vnode    *devvp;
 285         dk_extent_t     *extents;
 286         dk_unmap_t      unmap;
 287         _dk_cs_unmap_t  cs_unmap;
 288
 289         if ( !(vp->v_mount->mnt_ioflags & MNT_IOFLAGS_UNMAP_SUPPORTED))
 290                 return (ENOTSUP);
 291
 292         if (tl == NULL)
 293                 return (0);
 294
 295         /*
 296          * Get the underlying device vnode and physical block size
 297          */
 298         devvp = vp->v_mount->mnt_devvp;
 299         blocksize = vp->v_mount->mnt_devblocksize;
 300
 301         extents = kalloc(sizeof(dk_extent_t) * MAX_BATCH_TO_TRIM);
 302
 303         if (vp->v_mount->mnt_ioflags & MNT_IOFLAGS_CSUNMAP_SUPPORTED) {
 304                 memset (&cs_unmap, 0, sizeof(_dk_cs_unmap_t));
 305                 cs_unmap.extents = extents;
 306
 307                 if (route_only == TRUE)
 308                         cs_unmap.options = ROUTE_ONLY;
 309         } else {
 310                 memset (&unmap, 0, sizeof(dk_unmap_t));
 311                 unmap.extents = extents;
 312         }
 313
 314         while (tl) {
 315                 daddr64_t       io_blockno;     /* Block number corresponding to the start of the extent */
 316                 size_t          io_bytecount;   /* Number of bytes in current extent for the specified range */
 317                 size_t          trimmed;
 318                 size_t          remaining_length;
 319                 off_t           current_offset;
 320
 321                 current_offset = tl->tl_offset;
 322                 remaining_length = tl->tl_length;
 323                 trimmed = 0;
 324
 325                 /*
 326                  * We may not get the entire range from tl_offset -> tl_offset+tl_length in a single
 327                  * extent from the blockmap call.  Keep looping/going until we are sure we've hit
 328                  * the whole range or if we encounter an error.
 329                  */
 330                 while (trimmed < tl->tl_length) {
 331                         /*
 332                          * VNOP_BLOCKMAP will tell us the logical to physical block number mapping for the
 333                          * specified offset.  It returns blocks in contiguous chunks, so if the logical range is
 334                          * broken into multiple extents, it must be called multiple times, increasing the offset
 335                          * in each call to ensure that the entire range is covered.
 336                          */
 337                         error = VNOP_BLOCKMAP (vp, current_offset, remaining_length,
 338                                                &io_blockno, &io_bytecount, NULL, VNODE_READ, NULL);
 339
 340                         if (error) {
 341                                 goto trim_exit;
 342                         }
 343
 344                         extents[trim_index].offset = (uint64_t) io_blockno * (u_int64_t) blocksize;
 345                         extents[trim_index].length = io_bytecount;
 346
 347                         trim_index++;
 348
 349                         if (trim_index == MAX_BATCH_TO_TRIM) {
 350
 351                                 if (vp->v_mount->mnt_ioflags & MNT_IOFLAGS_CSUNMAP_SUPPORTED) {
 352                                         cs_unmap.extentsCount = trim_index;
 353                                         error = VNOP_IOCTL(devvp, _DKIOCCSUNMAP, (caddr_t)&cs_unmap, 0, vfs_context_kernel());
 354                                 } else {
 355                                         unmap.extentsCount = trim_index;
 356                                         error = VNOP_IOCTL(devvp, DKIOCUNMAP, (caddr_t)&unmap, 0, vfs_context_kernel());
 357                                 }
 358                                 if (error) {
 359                                         goto trim_exit;
 360                                 }
 361                                 trim_index = 0;
 362                         }
 363                         trimmed += io_bytecount;
 364                         current_offset += io_bytecount;
 365                         remaining_length -= io_bytecount;
 366                 }
 367                 tl = tl->tl_next;
 368         }
 369         if (trim_index) {
 370                 if (vp->v_mount->mnt_ioflags & MNT_IOFLAGS_CSUNMAP_SUPPORTED) {
 371                         cs_unmap.extentsCount = trim_index;
 372                         error = VNOP_IOCTL(devvp, _DKIOCCSUNMAP, (caddr_t)&cs_unmap, 0, vfs_context_kernel());
 373                 } else {
 374                         unmap.extentsCount = trim_index;
 375                         error = VNOP_IOCTL(devvp, DKIOCUNMAP, (caddr_t)&unmap, 0, vfs_context_kernel());
 376                 }
 377         }
 378 trim_exit:
 379         kfree(extents, sizeof(dk_extent_t) * MAX_BATCH_TO_TRIM);
 380
 381         return error;
 382 }