tmpfs_vfsops.c revision 312803
157434Smarkm/* $NetBSD: tmpfs_vfsops.c,v 1.10 2005/12/11 12:24:29 christos Exp $ */ 257434Smarkm 3156813Sru/*- 4156813Sru * Copyright (c) 2005 The NetBSD Foundation, Inc. 557434Smarkm * All rights reserved. 6158519Sdes * 757434Smarkm * This code is derived from software contributed to The NetBSD Foundation 898740Sdes * by Julio M. Merino Vidal, developed as part of Google's Summer of Code 998740Sdes * 2005 program. 1057434Smarkm * 11124242Sdes * Redistribution and use in source and binary forms, with or without 12181111Sdes * modification, are permitted provided that the following conditions 13204917Sdes * are met: 1457434Smarkm * 1. Redistributions of source code must retain the above copyright 15197679Sdes * notice, this list of conditions and the following disclaimer. 16197679Sdes * 2. Redistributions in binary form must reproduce the above copyright 17197679Sdes * notice, this list of conditions and the following disclaimer in the 18255460Sdes * documentation and/or other materials provided with the distribution. 19255460Sdes * 20255386Sdes * THIS SOFTWARE IS PROVIDED BY THE NETBSD FOUNDATION, INC. AND CONTRIBUTORS 21124249Sru * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED 22255460Sdes * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR 23255460Sdes * PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE FOUNDATION OR CONTRIBUTORS 24255460Sdes * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR 25255460Sdes * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF 26255460Sdes * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS 27255460Sdes * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN 28255460Sdes * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) 29156813Sru * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE 30255829Sdes * POSSIBILITY OF SUCH DAMAGE. 31153838Sdfr */ 32153838Sdfr 3395509Sru/* 3457434Smarkm * Efficient memory file system. 35245527Sbz * 36245527Sbz * tmpfs is a file system that uses FreeBSD's virtual memory 37245527Sbz * sub-system to store file data and metadata in an efficient way. 38245527Sbz * This means that it does not follow the structure of an on-disk file 39255460Sdes * system because it simply does not need to. Instead, it uses 40255460Sdes * memory-specific data structures and algorithms to automatically 4165675Skris * allocate and release resources. 42233432Seadler */ 43233432Seadler#include <sys/cdefs.h> 44233432Seadler__FBSDID("$FreeBSD: stable/10/sys/fs/tmpfs/tmpfs_vfsops.c 312803 2017-01-26 10:49:45Z kib $"); 45233432Seadler 4665675Skris#include <sys/param.h> 4774818Sru#include <sys/limits.h> 4874818Sru#include <sys/lock.h> 49158529Sdes#include <sys/mutex.h> 50158529Sdes#include <sys/proc.h> 51255829Sdes#include <sys/jail.h> 52255829Sdes#include <sys/kernel.h> 53255829Sdes#include <sys/rwlock.h> 54#include <sys/stat.h> 55#include <sys/systm.h> 56#include <sys/sysctl.h> 57 58#include <vm/vm.h> 59#include <vm/vm_object.h> 60#include <vm/vm_param.h> 61 62#include <fs/tmpfs/tmpfs.h> 63 64/* 65 * Default permission for root node 66 */ 67#define TMPFS_DEFAULT_ROOT_MODE (S_IRWXU|S_IRGRP|S_IXGRP|S_IROTH|S_IXOTH) 68 69MALLOC_DEFINE(M_TMPFSMNT, "tmpfs mount", "tmpfs mount structures"); 70MALLOC_DEFINE(M_TMPFSNAME, "tmpfs name", "tmpfs file names"); 71 72static int tmpfs_mount(struct mount *); 73static int tmpfs_unmount(struct mount *, int); 74static int tmpfs_root(struct mount *, int flags, struct vnode **); 75static int tmpfs_fhtovp(struct mount *, struct fid *, int, 76 struct vnode **); 77static int tmpfs_statfs(struct mount *, struct statfs *); 78 79static const char *tmpfs_opts[] = { 80 "from", "size", "maxfilesize", "inodes", "uid", "gid", "mode", "export", 81 "union", NULL 82}; 83 84static const char *tmpfs_updateopts[] = { 85 "from", "export", NULL 86}; 87 88static int 89tmpfs_node_ctor(void *mem, int size, void *arg, int flags) 90{ 91 struct tmpfs_node *node = (struct tmpfs_node *)mem; 92 93 node->tn_gen++; 94 node->tn_size = 0; 95 node->tn_status = 0; 96 node->tn_flags = 0; 97 node->tn_links = 0; 98 node->tn_vnode = NULL; 99 node->tn_vpstate = 0; 100 101 return (0); 102} 103 104static void 105tmpfs_node_dtor(void *mem, int size, void *arg) 106{ 107 struct tmpfs_node *node = (struct tmpfs_node *)mem; 108 node->tn_type = VNON; 109} 110 111static int 112tmpfs_node_init(void *mem, int size, int flags) 113{ 114 struct tmpfs_node *node = (struct tmpfs_node *)mem; 115 node->tn_id = 0; 116 117 mtx_init(&node->tn_interlock, "tmpfs node interlock", NULL, MTX_DEF); 118 node->tn_gen = arc4random(); 119 120 return (0); 121} 122 123static void 124tmpfs_node_fini(void *mem, int size) 125{ 126 struct tmpfs_node *node = (struct tmpfs_node *)mem; 127 128 mtx_destroy(&node->tn_interlock); 129} 130 131static int 132tmpfs_mount(struct mount *mp) 133{ 134 const size_t nodes_per_page = howmany(PAGE_SIZE, 135 sizeof(struct tmpfs_dirent) + sizeof(struct tmpfs_node)); 136 struct tmpfs_mount *tmp; 137 struct tmpfs_node *root; 138 struct thread *td = curthread; 139 int error; 140 /* Size counters. */ 141 u_quad_t pages; 142 off_t nodes_max, size_max, maxfilesize; 143 144 /* Root node attributes. */ 145 uid_t root_uid; 146 gid_t root_gid; 147 mode_t root_mode; 148 149 struct vattr va; 150 151 if (!prison_allow(td->td_ucred, PR_ALLOW_MOUNT_TMPFS)) 152 return (EPERM); 153 154 if (vfs_filteropt(mp->mnt_optnew, tmpfs_opts)) 155 return (EINVAL); 156 157 if (mp->mnt_flag & MNT_UPDATE) { 158 /* Only support update mounts for certain options. */ 159 if (vfs_filteropt(mp->mnt_optnew, tmpfs_updateopts) != 0) 160 return (EOPNOTSUPP); 161 if (vfs_flagopt(mp->mnt_optnew, "ro", NULL, 0) != 162 ((struct tmpfs_mount *)mp->mnt_data)->tm_ronly) 163 return (EOPNOTSUPP); 164 return (0); 165 } 166 167 vn_lock(mp->mnt_vnodecovered, LK_SHARED | LK_RETRY); 168 error = VOP_GETATTR(mp->mnt_vnodecovered, &va, mp->mnt_cred); 169 VOP_UNLOCK(mp->mnt_vnodecovered, 0); 170 if (error) 171 return (error); 172 173 if (mp->mnt_cred->cr_ruid != 0 || 174 vfs_scanopt(mp->mnt_optnew, "gid", "%d", &root_gid) != 1) 175 root_gid = va.va_gid; 176 if (mp->mnt_cred->cr_ruid != 0 || 177 vfs_scanopt(mp->mnt_optnew, "uid", "%d", &root_uid) != 1) 178 root_uid = va.va_uid; 179 if (mp->mnt_cred->cr_ruid != 0 || 180 vfs_scanopt(mp->mnt_optnew, "mode", "%ho", &root_mode) != 1) 181 root_mode = va.va_mode; 182 if (vfs_getopt_size(mp->mnt_optnew, "inodes", &nodes_max) != 0) 183 nodes_max = 0; 184 if (vfs_getopt_size(mp->mnt_optnew, "size", &size_max) != 0) 185 size_max = 0; 186 if (vfs_getopt_size(mp->mnt_optnew, "maxfilesize", &maxfilesize) != 0) 187 maxfilesize = 0; 188 189 /* Do not allow mounts if we do not have enough memory to preserve 190 * the minimum reserved pages. */ 191 if (tmpfs_mem_avail() < TMPFS_PAGES_MINRESERVED) 192 return (ENOSPC); 193 194 /* Get the maximum number of memory pages this file system is 195 * allowed to use, based on the maximum size the user passed in 196 * the mount structure. A value of zero is treated as if the 197 * maximum available space was requested. */ 198 if (size_max == 0 || size_max > OFF_MAX - PAGE_SIZE || 199 (SIZE_MAX < OFF_MAX && size_max / PAGE_SIZE >= SIZE_MAX)) 200 pages = SIZE_MAX; 201 else { 202 size_max = roundup(size_max, PAGE_SIZE); 203 pages = howmany(size_max, PAGE_SIZE); 204 } 205 MPASS(pages > 0); 206 207 if (nodes_max <= 3) { 208 if (pages < INT_MAX / nodes_per_page) 209 nodes_max = pages * nodes_per_page; 210 else 211 nodes_max = INT_MAX; 212 } 213 if (nodes_max > INT_MAX) 214 nodes_max = INT_MAX; 215 MPASS(nodes_max >= 3); 216 217 /* Allocate the tmpfs mount structure and fill it. */ 218 tmp = (struct tmpfs_mount *)malloc(sizeof(struct tmpfs_mount), 219 M_TMPFSMNT, M_WAITOK | M_ZERO); 220 221 mtx_init(&tmp->allnode_lock, "tmpfs allnode lock", NULL, MTX_DEF); 222 tmp->tm_nodes_max = nodes_max; 223 tmp->tm_nodes_inuse = 0; 224 tmp->tm_maxfilesize = maxfilesize > 0 ? maxfilesize : OFF_MAX; 225 LIST_INIT(&tmp->tm_nodes_used); 226 227 tmp->tm_pages_max = pages; 228 tmp->tm_pages_used = 0; 229 tmp->tm_ino_unr = new_unrhdr(2, INT_MAX, &tmp->allnode_lock); 230 tmp->tm_dirent_pool = uma_zcreate("TMPFS dirent", 231 sizeof(struct tmpfs_dirent), NULL, NULL, NULL, NULL, 232 UMA_ALIGN_PTR, 0); 233 tmp->tm_node_pool = uma_zcreate("TMPFS node", 234 sizeof(struct tmpfs_node), tmpfs_node_ctor, tmpfs_node_dtor, 235 tmpfs_node_init, tmpfs_node_fini, UMA_ALIGN_PTR, 0); 236 tmp->tm_ronly = (mp->mnt_flag & MNT_RDONLY) != 0; 237 238 /* Allocate the root node. */ 239 error = tmpfs_alloc_node(mp, tmp, VDIR, root_uid, root_gid, 240 root_mode & ALLPERMS, NULL, NULL, VNOVAL, &root); 241 242 if (error != 0 || root == NULL) { 243 uma_zdestroy(tmp->tm_node_pool); 244 uma_zdestroy(tmp->tm_dirent_pool); 245 delete_unrhdr(tmp->tm_ino_unr); 246 free(tmp, M_TMPFSMNT); 247 return (error); 248 } 249 KASSERT(root->tn_id == 2, 250 ("tmpfs root with invalid ino: %ju", (uintmax_t)root->tn_id)); 251 tmp->tm_root = root; 252 253 MNT_ILOCK(mp); 254 mp->mnt_flag |= MNT_LOCAL; 255 MNT_IUNLOCK(mp); 256 257 mp->mnt_data = tmp; 258 mp->mnt_stat.f_namemax = MAXNAMLEN; 259 vfs_getnewfsid(mp); 260 vfs_mountedfrom(mp, "tmpfs"); 261 262 return 0; 263} 264 265/* ARGSUSED2 */ 266static int 267tmpfs_unmount(struct mount *mp, int mntflags) 268{ 269 struct tmpfs_mount *tmp; 270 struct tmpfs_node *node; 271 int error, flags; 272 273 flags = (mntflags & MNT_FORCE) != 0 ? FORCECLOSE : 0; 274 tmp = VFS_TO_TMPFS(mp); 275 276 /* Stop writers */ 277 error = vfs_write_suspend_umnt(mp); 278 if (error != 0) 279 return (error); 280 /* 281 * At this point, nodes cannot be destroyed by any other 282 * thread because write suspension is started. 283 */ 284 285 for (;;) { 286 error = vflush(mp, 0, flags, curthread); 287 if (error != 0) { 288 vfs_write_resume(mp, VR_START_WRITE); 289 return (error); 290 } 291 MNT_ILOCK(mp); 292 if (mp->mnt_nvnodelistsize == 0) { 293 MNT_IUNLOCK(mp); 294 break; 295 } 296 MNT_IUNLOCK(mp); 297 if ((mntflags & MNT_FORCE) == 0) { 298 vfs_write_resume(mp, VR_START_WRITE); 299 return (EBUSY); 300 } 301 } 302 303 TMPFS_LOCK(tmp); 304 while ((node = LIST_FIRST(&tmp->tm_nodes_used)) != NULL) { 305 TMPFS_UNLOCK(tmp); 306 if (node->tn_type == VDIR) 307 tmpfs_dir_destroy(tmp, node); 308 tmpfs_free_node(tmp, node); 309 TMPFS_LOCK(tmp); 310 } 311 TMPFS_UNLOCK(tmp); 312 313 uma_zdestroy(tmp->tm_dirent_pool); 314 uma_zdestroy(tmp->tm_node_pool); 315 delete_unrhdr(tmp->tm_ino_unr); 316 317 mtx_destroy(&tmp->allnode_lock); 318 MPASS(tmp->tm_pages_used == 0); 319 MPASS(tmp->tm_nodes_inuse == 0); 320 321 /* Throw away the tmpfs_mount structure. */ 322 free(mp->mnt_data, M_TMPFSMNT); 323 mp->mnt_data = NULL; 324 vfs_write_resume(mp, VR_START_WRITE); 325 326 MNT_ILOCK(mp); 327 mp->mnt_flag &= ~MNT_LOCAL; 328 MNT_IUNLOCK(mp); 329 330 return (0); 331} 332 333static int 334tmpfs_root(struct mount *mp, int flags, struct vnode **vpp) 335{ 336 int error; 337 338 error = tmpfs_alloc_vp(mp, VFS_TO_TMPFS(mp)->tm_root, flags, vpp); 339 if (error == 0) 340 (*vpp)->v_vflag |= VV_ROOT; 341 return (error); 342} 343 344static int 345tmpfs_fhtovp(struct mount *mp, struct fid *fhp, int flags, 346 struct vnode **vpp) 347{ 348 boolean_t found; 349 struct tmpfs_fid *tfhp; 350 struct tmpfs_mount *tmp; 351 struct tmpfs_node *node; 352 353 tmp = VFS_TO_TMPFS(mp); 354 355 tfhp = (struct tmpfs_fid *)fhp; 356 if (tfhp->tf_len != sizeof(struct tmpfs_fid)) 357 return EINVAL; 358 359 if (tfhp->tf_id >= tmp->tm_nodes_max) 360 return EINVAL; 361 362 found = FALSE; 363 364 TMPFS_LOCK(tmp); 365 LIST_FOREACH(node, &tmp->tm_nodes_used, tn_entries) { 366 if (node->tn_id == tfhp->tf_id && 367 node->tn_gen == tfhp->tf_gen) { 368 found = TRUE; 369 break; 370 } 371 } 372 TMPFS_UNLOCK(tmp); 373 374 if (found) 375 return (tmpfs_alloc_vp(mp, node, LK_EXCLUSIVE, vpp)); 376 377 return (EINVAL); 378} 379 380/* ARGSUSED2 */ 381static int 382tmpfs_statfs(struct mount *mp, struct statfs *sbp) 383{ 384 struct tmpfs_mount *tmp; 385 size_t used; 386 387 tmp = VFS_TO_TMPFS(mp); 388 389 sbp->f_iosize = PAGE_SIZE; 390 sbp->f_bsize = PAGE_SIZE; 391 392 used = tmpfs_pages_used(tmp); 393 if (tmp->tm_pages_max != ULONG_MAX) 394 sbp->f_blocks = tmp->tm_pages_max; 395 else 396 sbp->f_blocks = used + tmpfs_mem_avail(); 397 if (sbp->f_blocks <= used) 398 sbp->f_bavail = 0; 399 else 400 sbp->f_bavail = sbp->f_blocks - used; 401 sbp->f_bfree = sbp->f_bavail; 402 used = tmp->tm_nodes_inuse; 403 sbp->f_files = tmp->tm_nodes_max; 404 if (sbp->f_files <= used) 405 sbp->f_ffree = 0; 406 else 407 sbp->f_ffree = sbp->f_files - used; 408 /* sbp->f_owner = tmp->tn_uid; */ 409 410 return 0; 411} 412 413static int 414tmpfs_sync(struct mount *mp, int waitfor) 415{ 416 struct vnode *vp, *mvp; 417 struct vm_object *obj; 418 419 if (waitfor == MNT_SUSPEND) { 420 MNT_ILOCK(mp); 421 mp->mnt_kern_flag |= MNTK_SUSPEND2 | MNTK_SUSPENDED; 422 MNT_IUNLOCK(mp); 423 } else if (waitfor == MNT_LAZY) { 424 /* 425 * Handle lazy updates of mtime from writes to mmaped 426 * regions. Use MNT_VNODE_FOREACH_ALL instead of 427 * MNT_VNODE_FOREACH_ACTIVE, since unmap of the 428 * tmpfs-backed vnode does not call vinactive(), due 429 * to vm object type is OBJT_SWAP. 430 */ 431 MNT_VNODE_FOREACH_ALL(vp, mp, mvp) { 432 if (vp->v_type != VREG) { 433 VI_UNLOCK(vp); 434 continue; 435 } 436 obj = vp->v_object; 437 KASSERT((obj->flags & (OBJ_TMPFS_NODE | OBJ_TMPFS)) == 438 (OBJ_TMPFS_NODE | OBJ_TMPFS), ("non-tmpfs obj")); 439 440 /* 441 * Unlocked read, avoid taking vnode lock if 442 * not needed. Lost update will be handled on 443 * the next call. 444 */ 445 if ((obj->flags & OBJ_TMPFS_DIRTY) == 0) { 446 VI_UNLOCK(vp); 447 continue; 448 } 449 if (vget(vp, LK_EXCLUSIVE | LK_RETRY | LK_INTERLOCK, 450 curthread) != 0) 451 continue; 452 tmpfs_check_mtime(vp); 453 vput(vp); 454 } 455 } 456 return (0); 457} 458 459/* 460 * tmpfs vfs operations. 461 */ 462 463struct vfsops tmpfs_vfsops = { 464 .vfs_mount = tmpfs_mount, 465 .vfs_unmount = tmpfs_unmount, 466 .vfs_root = tmpfs_root, 467 .vfs_statfs = tmpfs_statfs, 468 .vfs_fhtovp = tmpfs_fhtovp, 469 .vfs_sync = tmpfs_sync, 470}; 471VFS_SET(tmpfs_vfsops, tmpfs, VFCF_JAIL); 472