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 https://opensource.org/licenses/CDDL-1.0. 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, 2020 by Delphix. All rights reserved. 24 * Copyright (c) 2025, Klara, Inc. 25 */ 26 27 #include <sys/zfs_context.h> 28 #include <sys/spa.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 #include <sys/stat.h> 36 37 /* 38 * Virtual device vector for files. 39 */ 40 41 static taskq_t *vdev_file_taskq; 42 43 /* 44 * By default, the logical/physical ashift for file vdevs is set to 45 * SPA_MINBLOCKSHIFT (9). This allows all file vdevs to use 512B (1 << 9) 46 * blocksizes. Users may opt to change one or both of these for testing 47 * or performance reasons. Care should be taken as these values will 48 * impact the vdev_ashift setting which can only be set at vdev creation 49 * time. 50 */ 51 static uint_t vdev_file_logical_ashift = SPA_MINBLOCKSHIFT; 52 static uint_t vdev_file_physical_ashift = SPA_MINBLOCKSHIFT; 53 54 void 55 vdev_file_init(void) 56 { 57 vdev_file_taskq = taskq_create("z_vdev_file", MAX(boot_ncpus, 16), 58 minclsyspri, boot_ncpus, INT_MAX, TASKQ_DYNAMIC); 59 60 VERIFY(vdev_file_taskq); 61 } 62 63 void 64 vdev_file_fini(void) 65 { 66 taskq_destroy(vdev_file_taskq); 67 } 68 69 static void 70 vdev_file_hold(vdev_t *vd) 71 { 72 ASSERT3P(vd->vdev_path, !=, NULL); 73 } 74 75 static void 76 vdev_file_rele(vdev_t *vd) 77 { 78 ASSERT3P(vd->vdev_path, !=, NULL); 79 } 80 81 static mode_t 82 vdev_file_open_mode(spa_mode_t spa_mode) 83 { 84 mode_t mode = 0; 85 86 if ((spa_mode & SPA_MODE_READ) && (spa_mode & SPA_MODE_WRITE)) { 87 mode = O_RDWR; 88 } else if (spa_mode & SPA_MODE_READ) { 89 mode = O_RDONLY; 90 } else if (spa_mode & SPA_MODE_WRITE) { 91 mode = O_WRONLY; 92 } 93 94 return (mode | O_LARGEFILE); 95 } 96 97 static int 98 vdev_file_open(vdev_t *vd, uint64_t *psize, uint64_t *max_psize, 99 uint64_t *logical_ashift, uint64_t *physical_ashift) 100 { 101 vdev_file_t *vf; 102 zfs_file_t *fp; 103 zfs_file_attr_t zfa; 104 int error; 105 106 /* 107 * Rotational optimizations only make sense on block devices. 108 */ 109 vd->vdev_nonrot = B_TRUE; 110 111 /* 112 * Allow TRIM on file based vdevs. This may not always be supported, 113 * since it depends on your kernel version and underlying filesystem 114 * type but it is always safe to attempt. 115 */ 116 vd->vdev_has_trim = B_TRUE; 117 118 /* 119 * Disable secure TRIM on file based vdevs. There is no way to 120 * request this behavior from the underlying filesystem. 121 */ 122 vd->vdev_has_securetrim = B_FALSE; 123 124 /* 125 * We must have a pathname, and it must be absolute. 126 */ 127 if (vd->vdev_path == NULL || vd->vdev_path[0] != '/') { 128 vd->vdev_stat.vs_aux = VDEV_AUX_BAD_LABEL; 129 return (SET_ERROR(EINVAL)); 130 } 131 132 /* 133 * Reopen the device if it's not currently open. Otherwise, 134 * just update the physical size of the device. 135 */ 136 if (vd->vdev_tsd != NULL) { 137 ASSERT(vd->vdev_reopening); 138 vf = vd->vdev_tsd; 139 goto skip_open; 140 } 141 142 vf = vd->vdev_tsd = kmem_zalloc(sizeof (vdev_file_t), KM_SLEEP); 143 144 /* 145 * We always open the files from the root of the global zone, even if 146 * we're in a local zone. If the user has gotten to this point, the 147 * administrator has already decided that the pool should be available 148 * to local zone users, so the underlying devices should be as well. 149 */ 150 ASSERT3P(vd->vdev_path, !=, NULL); 151 ASSERT3S(vd->vdev_path[0], ==, '/'); 152 153 error = zfs_file_open(vd->vdev_path, 154 vdev_file_open_mode(spa_mode(vd->vdev_spa)), 0, &fp); 155 if (error) { 156 vd->vdev_stat.vs_aux = VDEV_AUX_OPEN_FAILED; 157 return (error); 158 } 159 160 vf->vf_file = fp; 161 162 #ifdef _KERNEL 163 /* 164 * Make sure it's a regular file. 165 */ 166 if (zfs_file_getattr(fp, &zfa)) { 167 return (SET_ERROR(ENODEV)); 168 } 169 if (!S_ISREG(zfa.zfa_mode)) { 170 vd->vdev_stat.vs_aux = VDEV_AUX_OPEN_FAILED; 171 return (SET_ERROR(ENODEV)); 172 } 173 #endif 174 175 skip_open: 176 177 error = zfs_file_getattr(vf->vf_file, &zfa); 178 if (error) { 179 vd->vdev_stat.vs_aux = VDEV_AUX_OPEN_FAILED; 180 return (error); 181 } 182 183 *max_psize = *psize = zfa.zfa_size; 184 *logical_ashift = vdev_file_logical_ashift; 185 *physical_ashift = vdev_file_physical_ashift; 186 187 return (0); 188 } 189 190 static void 191 vdev_file_close(vdev_t *vd) 192 { 193 vdev_file_t *vf = vd->vdev_tsd; 194 195 if (vd->vdev_reopening || vf == NULL) 196 return; 197 198 if (vf->vf_file != NULL) { 199 (void) zfs_file_close(vf->vf_file); 200 } 201 202 vd->vdev_delayed_close = B_FALSE; 203 kmem_free(vf, sizeof (vdev_file_t)); 204 vd->vdev_tsd = NULL; 205 } 206 207 static void 208 vdev_file_io_strategy(void *arg) 209 { 210 zio_t *zio = (zio_t *)arg; 211 vdev_t *vd = zio->io_vd; 212 vdev_file_t *vf = vd->vdev_tsd; 213 void *buf; 214 ssize_t resid; 215 loff_t off; 216 ssize_t size; 217 int err; 218 219 off = zio->io_offset; 220 size = zio->io_size; 221 resid = 0; 222 223 ASSERT(zio->io_type == ZIO_TYPE_READ || zio->io_type == ZIO_TYPE_WRITE); 224 if (zio->io_type == ZIO_TYPE_READ) { 225 buf = abd_borrow_buf(zio->io_abd, zio->io_size); 226 err = zfs_file_pread(vf->vf_file, buf, size, off, &resid); 227 abd_return_buf_copy(zio->io_abd, buf, size); 228 } else { 229 buf = abd_borrow_buf_copy(zio->io_abd, zio->io_size); 230 err = zfs_file_pwrite(vf->vf_file, buf, size, off, &resid); 231 abd_return_buf(zio->io_abd, buf, size); 232 } 233 zio->io_error = err; 234 if (resid != 0 && zio->io_error == 0) 235 zio->io_error = SET_ERROR(ENOSPC); 236 237 zio_delay_interrupt(zio); 238 } 239 240 static void 241 vdev_file_io_fsync(void *arg) 242 { 243 zio_t *zio = (zio_t *)arg; 244 vdev_file_t *vf = zio->io_vd->vdev_tsd; 245 246 zio->io_error = zfs_file_fsync(vf->vf_file, O_SYNC | O_DSYNC); 247 248 zio_interrupt(zio); 249 } 250 251 static void 252 vdev_file_io_deallocate(void *arg) 253 { 254 zio_t *zio = (zio_t *)arg; 255 vdev_file_t *vf = zio->io_vd->vdev_tsd; 256 257 zio->io_error = zfs_file_deallocate(vf->vf_file, 258 zio->io_offset, zio->io_size); 259 260 zio_interrupt(zio); 261 } 262 263 static void 264 vdev_file_io_start(zio_t *zio) 265 { 266 vdev_t *vd = zio->io_vd; 267 268 if (zio->io_type == ZIO_TYPE_FLUSH) { 269 /* XXPOLICY */ 270 if (!vdev_readable(vd)) { 271 zio->io_error = SET_ERROR(ENXIO); 272 zio_interrupt(zio); 273 return; 274 } 275 276 if (zfs_nocacheflush) { 277 zio_interrupt(zio); 278 return; 279 } 280 281 VERIFY3U(taskq_dispatch(vdev_file_taskq, 282 vdev_file_io_fsync, zio, TQ_SLEEP), !=, TASKQID_INVALID); 283 284 return; 285 } 286 287 if (zio->io_type == ZIO_TYPE_TRIM) { 288 ASSERT3U(zio->io_size, !=, 0); 289 290 VERIFY3U(taskq_dispatch(vdev_file_taskq, 291 vdev_file_io_deallocate, zio, TQ_SLEEP), !=, 292 TASKQID_INVALID); 293 294 return; 295 } 296 297 ASSERT(zio->io_type == ZIO_TYPE_READ || zio->io_type == ZIO_TYPE_WRITE); 298 zio->io_target_timestamp = zio_handle_io_delay(zio); 299 300 VERIFY3U(taskq_dispatch(vdev_file_taskq, vdev_file_io_strategy, zio, 301 TQ_SLEEP), !=, TASKQID_INVALID); 302 } 303 304 static void 305 vdev_file_io_done(zio_t *zio) 306 { 307 (void) zio; 308 } 309 310 vdev_ops_t vdev_file_ops = { 311 .vdev_op_init = NULL, 312 .vdev_op_fini = NULL, 313 .vdev_op_open = vdev_file_open, 314 .vdev_op_close = vdev_file_close, 315 .vdev_op_asize = vdev_default_asize, 316 .vdev_op_min_asize = vdev_default_min_asize, 317 .vdev_op_min_alloc = NULL, 318 .vdev_op_io_start = vdev_file_io_start, 319 .vdev_op_io_done = vdev_file_io_done, 320 .vdev_op_state_change = NULL, 321 .vdev_op_need_resilver = NULL, 322 .vdev_op_hold = vdev_file_hold, 323 .vdev_op_rele = vdev_file_rele, 324 .vdev_op_remap = NULL, 325 .vdev_op_xlate = vdev_default_xlate, 326 .vdev_op_rebuild_asize = NULL, 327 .vdev_op_metaslab_init = NULL, 328 .vdev_op_config_generate = NULL, 329 .vdev_op_nparity = NULL, 330 .vdev_op_ndisks = NULL, 331 .vdev_op_type = VDEV_TYPE_FILE, /* name of this vdev type */ 332 .vdev_op_leaf = B_TRUE /* leaf vdev */ 333 }; 334 335 /* 336 * From userland we access disks just like files. 337 */ 338 #ifndef _KERNEL 339 340 vdev_ops_t vdev_disk_ops = { 341 .vdev_op_init = NULL, 342 .vdev_op_fini = NULL, 343 .vdev_op_open = vdev_file_open, 344 .vdev_op_close = vdev_file_close, 345 .vdev_op_asize = vdev_default_asize, 346 .vdev_op_min_asize = vdev_default_min_asize, 347 .vdev_op_min_alloc = NULL, 348 .vdev_op_io_start = vdev_file_io_start, 349 .vdev_op_io_done = vdev_file_io_done, 350 .vdev_op_state_change = NULL, 351 .vdev_op_need_resilver = NULL, 352 .vdev_op_hold = vdev_file_hold, 353 .vdev_op_rele = vdev_file_rele, 354 .vdev_op_remap = NULL, 355 .vdev_op_xlate = vdev_default_xlate, 356 .vdev_op_rebuild_asize = NULL, 357 .vdev_op_metaslab_init = NULL, 358 .vdev_op_config_generate = NULL, 359 .vdev_op_nparity = NULL, 360 .vdev_op_ndisks = NULL, 361 .vdev_op_type = VDEV_TYPE_DISK, /* name of this vdev type */ 362 .vdev_op_leaf = B_TRUE /* leaf vdev */ 363 }; 364 365 #endif 366 367 ZFS_MODULE_PARAM(zfs_vdev_file, vdev_file_, logical_ashift, UINT, ZMOD_RW, 368 "Logical ashift for file-based devices"); 369 ZFS_MODULE_PARAM(zfs_vdev_file, vdev_file_, physical_ashift, UINT, ZMOD_RW, 370 "Physical ashift for file-based devices"); 371