module/zfs/vdev_file.c

   1 /*
   2  * CDDL HEADER START
   3  *
   4  * The contents of this file are subject to the terms of the
   5  * Common Development and Distribution License (the "License").
   6  * You may not use this file except in compliance with the License.
   7  *
   8  * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
   9  * or http://www.opensolaris.org/os/licensing.
  10  * See the License for the specific language governing permissions
  11  * and limitations under the License.
  12  *
  13  * When distributing Covered Code, include this CDDL HEADER in each
  14  * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
  15  * If applicable, add the following below this CDDL HEADER, with the
  16  * fields enclosed by brackets "[]" replaced with your own identifying
  17  * information: Portions Copyright [yyyy] [name of copyright owner]
  18  *
  19  * CDDL HEADER END
  20  */
  21 /*
  22  * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved.
  23  * Copyright (c) 2011, 2016 by Delphix. All rights reserved.
  24  */
  25
  26 #include <sys/zfs_context.h>
  27 #include <sys/spa.h>
  28 #include <sys/spa_impl.h>
  29 #include <sys/vdev_file.h>
  30 #include <sys/vdev_impl.h>
  31 #include <sys/zio.h>
  32 #include <sys/fs/zfs.h>
  33 #include <sys/fm/fs/zfs.h>
  34 #include <sys/abd.h>
  35
  36 /*
  37  * Virtual device vector for files.
  38  */
  39
  40 static taskq_t *vdev_file_taskq;
  41
  42 static void
  43 vdev_file_hold(vdev_t *vd)
  44 {
  45         ASSERT(vd->vdev_path != NULL);
  46 }
  47
  48 static void
  49 vdev_file_rele(vdev_t *vd)
  50 {
  51         ASSERT(vd->vdev_path != NULL);
  52 }
  53
  54 static int
  55 vdev_file_open(vdev_t *vd, uint64_t *psize, uint64_t *max_psize,
  56     uint64_t *ashift)
  57 {
  58         vdev_file_t *vf;
  59         vnode_t *vp;
  60         vattr_t vattr;
  61         int error;
  62
  63         /* Rotational optimizations only make sense on block devices */
  64         vd->vdev_nonrot = B_TRUE;
  65
  66         /*
  67          * We must have a pathname, and it must be absolute.
  68          */
  69         if (vd->vdev_path == NULL || vd->vdev_path[0] != '/') {
  70                 vd->vdev_stat.vs_aux = VDEV_AUX_BAD_LABEL;
  71                 return (SET_ERROR(EINVAL));
  72         }
  73
  74         /*
  75          * Reopen the device if it's not currently open.  Otherwise,
  76          * just update the physical size of the device.
  77          */
  78         if (vd->vdev_tsd != NULL) {
  79                 ASSERT(vd->vdev_reopening);
  80                 vf = vd->vdev_tsd;
  81                 goto skip_open;
  82         }
  83
  84         vf = vd->vdev_tsd = kmem_zalloc(sizeof (vdev_file_t), KM_SLEEP);
  85
  86         /*
  87          * We always open the files from the root of the global zone, even if
  88          * we're in a local zone.  If the user has gotten to this point, the
  89          * administrator has already decided that the pool should be available
  90          * to local zone users, so the underlying devices should be as well.
  91          */
  92         ASSERT(vd->vdev_path != NULL && vd->vdev_path[0] == '/');
  93         error = vn_openat(vd->vdev_path + 1, UIO_SYSSPACE,
  94             spa_mode(vd->vdev_spa) | FOFFMAX, 0, &vp, 0, 0, rootdir, -1);
  95
  96         if (error) {
  97                 vd->vdev_stat.vs_aux = VDEV_AUX_OPEN_FAILED;
  98                 return (error);
  99         }
 100
 101         vf->vf_vnode = vp;
 102
 103 #ifdef _KERNEL
 104         /*
 105          * Make sure it's a regular file.
 106          */
 107         if (vp->v_type != VREG) {
 108                 vd->vdev_stat.vs_aux = VDEV_AUX_OPEN_FAILED;
 109                 return (SET_ERROR(ENODEV));
 110         }
 111 #endif
 112
 113 skip_open:
 114         /*
 115          * Determine the physical size of the file.
 116          */
 117         vattr.va_mask = AT_SIZE;
 118         error = VOP_GETATTR(vf->vf_vnode, &vattr, 0, kcred, NULL);
 119         if (error) {
 120                 vd->vdev_stat.vs_aux = VDEV_AUX_OPEN_FAILED;
 121                 return (error);
 122         }
 123
 124         *max_psize = *psize = vattr.va_size;
 125         *ashift = SPA_MINBLOCKSHIFT;
 126
 127         return (0);
 128 }
 129
 130 static void
 131 vdev_file_close(vdev_t *vd)
 132 {
 133         vdev_file_t *vf = vd->vdev_tsd;
 134
 135         if (vd->vdev_reopening || vf == NULL)
 136                 return;
 137
 138         if (vf->vf_vnode != NULL) {
 139                 (void) VOP_PUTPAGE(vf->vf_vnode, 0, 0, B_INVAL, kcred, NULL);
 140                 (void) VOP_CLOSE(vf->vf_vnode, spa_mode(vd->vdev_spa), 1, 0,
 141                     kcred, NULL);
 142         }
 143
 144         vd->vdev_delayed_close = B_FALSE;
 145         kmem_free(vf, sizeof (vdev_file_t));
 146         vd->vdev_tsd = NULL;
 147 }
 148
 149 static void
 150 vdev_file_io_strategy(void *arg)
 151 {
 152         zio_t *zio = (zio_t *)arg;
 153         vdev_t *vd = zio->io_vd;
 154         vdev_file_t *vf = vd->vdev_tsd;
 155         ssize_t resid;
 156         void *buf;
 157
 158         if (zio->io_type == ZIO_TYPE_READ)
 159                 buf = abd_borrow_buf(zio->io_abd, zio->io_size);
 160         else
 161                 buf = abd_borrow_buf_copy(zio->io_abd, zio->io_size);
 162
 163         zio->io_error = vn_rdwr(zio->io_type == ZIO_TYPE_READ ?
 164             UIO_READ : UIO_WRITE, vf->vf_vnode, buf, zio->io_size,
 165             zio->io_offset, UIO_SYSSPACE, 0, RLIM64_INFINITY, kcred, &resid);
 166
 167         if (zio->io_type == ZIO_TYPE_READ)
 168                 abd_return_buf_copy(zio->io_abd, buf, zio->io_size);
 169         else
 170                 abd_return_buf(zio->io_abd, buf, zio->io_size);
 171
 172         if (resid != 0 && zio->io_error == 0)
 173                 zio->io_error = SET_ERROR(ENOSPC);
 174
 175         zio_delay_interrupt(zio);
 176 }
 177
 178 static void
 179 vdev_file_io_fsync(void *arg)
 180 {
 181         zio_t *zio = (zio_t *)arg;
 182         vdev_file_t *vf = zio->io_vd->vdev_tsd;
 183
 184         zio->io_error = VOP_FSYNC(vf->vf_vnode, FSYNC | FDSYNC, kcred, NULL);
 185
 186         zio_interrupt(zio);
 187 }
 188
 189 static void
 190 vdev_file_io_start(zio_t *zio)
 191 {
 192         vdev_t *vd = zio->io_vd;
 193         vdev_file_t *vf = vd->vdev_tsd;
 194
 195         if (zio->io_type == ZIO_TYPE_IOCTL) {
 196                 /* XXPOLICY */
 197                 if (!vdev_readable(vd)) {
 198                         zio->io_error = SET_ERROR(ENXIO);
 199                         zio_interrupt(zio);
 200                         return;
 201                 }
 202
 203                 switch (zio->io_cmd) {
 204                 case DKIOCFLUSHWRITECACHE:
 205
 206                         if (zfs_nocacheflush)
 207                                 break;
 208
 209                         /*
 210                          * We cannot safely call vfs_fsync() when PF_FSTRANS
 211                          * is set in the current context.  Filesystems like
 212                          * XFS include sanity checks to verify it is not
 213                          * already set, see xfs_vm_writepage().  Therefore
 214                          * the sync must be dispatched to a different context.
 215                          */
 216                         if (__spl_pf_fstrans_check()) {
 217                                 VERIFY3U(taskq_dispatch(vdev_file_taskq,
 218                                     vdev_file_io_fsync, zio, TQ_SLEEP), !=,
 219                                     TASKQID_INVALID);
 220                                 return;
 221                         }
 222
 223                         zio->io_error = VOP_FSYNC(vf->vf_vnode, FSYNC | FDSYNC,
 224                             kcred, NULL);
 225                         break;
 226                 default:
 227                         zio->io_error = SET_ERROR(ENOTSUP);
 228                 }
 229
 230                 zio_execute(zio);
 231                 return;
 232         }
 233
 234         zio->io_target_timestamp = zio_handle_io_delay(zio);
 235
 236         VERIFY3U(taskq_dispatch(vdev_file_taskq, vdev_file_io_strategy, zio,
 237             TQ_SLEEP), !=, TASKQID_INVALID);
 238 }
 239
 240 /* ARGSUSED */
 241 static void
 242 vdev_file_io_done(zio_t *zio)
 243 {
 244 }
 245
 246 vdev_ops_t vdev_file_ops = {
 247         vdev_file_open,
 248         vdev_file_close,
 249         vdev_default_asize,
 250         vdev_file_io_start,
 251         vdev_file_io_done,
 252         NULL,
 253         NULL,
 254         vdev_file_hold,
 255         vdev_file_rele,
 256         NULL,
 257         vdev_default_xlate,
 258         VDEV_TYPE_FILE,         /* name of this vdev type */
 259         B_TRUE                  /* leaf vdev */
 260 };
 261
 262 void
 263 vdev_file_init(void)
 264 {
 265         vdev_file_taskq = taskq_create("z_vdev_file", MAX(boot_ncpus, 16),
 266             minclsyspri, boot_ncpus, INT_MAX, TASKQ_DYNAMIC);
 267
 268         VERIFY(vdev_file_taskq);
 269 }
 270
 271 void
 272 vdev_file_fini(void)
 273 {
 274         taskq_destroy(vdev_file_taskq);
 275 }
 276
 277 /*
 278  * From userland we access disks just like files.
 279  */
 280 #ifndef _KERNEL
 281
 282 vdev_ops_t vdev_disk_ops = {
 283         vdev_file_open,
 284         vdev_file_close,
 285         vdev_default_asize,
 286         vdev_file_io_start,
 287         vdev_file_io_done,
 288         NULL,
 289         NULL,
 290         vdev_file_hold,
 291         vdev_file_rele,
 292         NULL,
 293         vdev_default_xlate,
 294         VDEV_TYPE_DISK,         /* name of this vdev type */
 295         B_TRUE                  /* leaf vdev */
 296 };
 297
 298 #endif