sys_pipe.c revision 1.4.2.9
1 1.4.2.9 nathanw /* $NetBSD: sys_pipe.c,v 1.4.2.9 2002/01/08 00:32:40 nathanw Exp $ */ 2 1.4.2.2 nathanw 3 1.4.2.2 nathanw /* 4 1.4.2.2 nathanw * Copyright (c) 1996 John S. Dyson 5 1.4.2.2 nathanw * All rights reserved. 6 1.4.2.2 nathanw * 7 1.4.2.2 nathanw * Redistribution and use in source and binary forms, with or without 8 1.4.2.2 nathanw * modification, are permitted provided that the following conditions 9 1.4.2.2 nathanw * are met: 10 1.4.2.2 nathanw * 1. Redistributions of source code must retain the above copyright 11 1.4.2.2 nathanw * notice immediately at the beginning of the file, without modification, 12 1.4.2.2 nathanw * this list of conditions, and the following disclaimer. 13 1.4.2.2 nathanw * 2. Redistributions in binary form must reproduce the above copyright 14 1.4.2.2 nathanw * notice, this list of conditions and the following disclaimer in the 15 1.4.2.2 nathanw * documentation and/or other materials provided with the distribution. 16 1.4.2.2 nathanw * 3. Absolutely no warranty of function or purpose is made by the author 17 1.4.2.2 nathanw * John S. Dyson. 18 1.4.2.2 nathanw * 4. Modifications may be freely made to this file if the above conditions 19 1.4.2.2 nathanw * are met. 20 1.4.2.2 nathanw * 21 1.4.2.2 nathanw * $FreeBSD: src/sys/kern/sys_pipe.c,v 1.82 2001/06/15 20:45:01 jlemon Exp $ 22 1.4.2.2 nathanw */ 23 1.4.2.2 nathanw 24 1.4.2.2 nathanw /* 25 1.4.2.2 nathanw * This file contains a high-performance replacement for the socket-based 26 1.4.2.2 nathanw * pipes scheme originally used in FreeBSD/4.4Lite. It does not support 27 1.4.2.2 nathanw * all features of sockets, but does do everything that pipes normally 28 1.4.2.2 nathanw * do. 29 1.4.2.2 nathanw * 30 1.4.2.2 nathanw * Adaption for NetBSD UVM, including uvm_loan() based direct write, was 31 1.4.2.2 nathanw * written by Jaromir Dolecek. 32 1.4.2.2 nathanw */ 33 1.4.2.2 nathanw 34 1.4.2.2 nathanw /* 35 1.4.2.2 nathanw * This code has two modes of operation, a small write mode and a large 36 1.4.2.2 nathanw * write mode. The small write mode acts like conventional pipes with 37 1.4.2.2 nathanw * a kernel buffer. If the buffer is less than PIPE_MINDIRECT, then the 38 1.4.2.2 nathanw * "normal" pipe buffering is done. If the buffer is between PIPE_MINDIRECT 39 1.4.2.2 nathanw * and PIPE_SIZE in size, it is fully mapped into the kernel (on FreeBSD, 40 1.4.2.2 nathanw * those pages are also wired), and the receiving process can copy it directly 41 1.4.2.2 nathanw * from the pages in the sending process. 42 1.4.2.2 nathanw * 43 1.4.2.2 nathanw * If the sending process receives a signal, it is possible that it will 44 1.4.2.2 nathanw * go away, and certainly its address space can change, because control 45 1.4.2.2 nathanw * is returned back to the user-mode side. In that case, the pipe code 46 1.4.2.2 nathanw * arranges to copy the buffer supplied by the user process on FreeBSD, to 47 1.4.2.2 nathanw * a pageable kernel buffer, and the receiving process will grab the data 48 1.4.2.2 nathanw * from the pageable kernel buffer. Since signals don't happen all that often, 49 1.4.2.2 nathanw * the copy operation is normally eliminated. 50 1.4.2.2 nathanw * For NetBSD, the pages are mapped read-only, COW for kernel by uvm_loan(), 51 1.4.2.2 nathanw * so no explicit handling need to be done, all is handled by standard VM 52 1.4.2.2 nathanw * facilities. 53 1.4.2.2 nathanw * 54 1.4.2.2 nathanw * The constant PIPE_MINDIRECT is chosen to make sure that buffering will 55 1.4.2.2 nathanw * happen for small transfers so that the system will not spend all of 56 1.4.2.2 nathanw * its time context switching. PIPE_SIZE is constrained by the 57 1.4.2.2 nathanw * amount of kernel virtual memory. 58 1.4.2.2 nathanw */ 59 1.4.2.2 nathanw 60 1.4.2.8 nathanw #include <sys/cdefs.h> 61 1.4.2.9 nathanw __KERNEL_RCSID(0, "$NetBSD: sys_pipe.c,v 1.4.2.9 2002/01/08 00:32:40 nathanw Exp $"); 62 1.4.2.8 nathanw 63 1.4.2.2 nathanw #include <sys/param.h> 64 1.4.2.2 nathanw #include <sys/systm.h> 65 1.4.2.2 nathanw #include <sys/proc.h> 66 1.4.2.2 nathanw #include <sys/fcntl.h> 67 1.4.2.2 nathanw #include <sys/file.h> 68 1.4.2.2 nathanw #include <sys/filedesc.h> 69 1.4.2.2 nathanw #include <sys/filio.h> 70 1.4.2.2 nathanw #include <sys/ttycom.h> 71 1.4.2.2 nathanw #include <sys/stat.h> 72 1.4.2.2 nathanw #include <sys/poll.h> 73 1.4.2.2 nathanw #include <sys/signalvar.h> 74 1.4.2.2 nathanw #include <sys/vnode.h> 75 1.4.2.2 nathanw #include <sys/uio.h> 76 1.4.2.2 nathanw #include <sys/lock.h> 77 1.4.2.2 nathanw #ifdef __FreeBSD__ 78 1.4.2.2 nathanw #include <sys/mutex.h> 79 1.4.2.2 nathanw #include <sys/selinfo.h> 80 1.4.2.2 nathanw #include <sys/sysproto.h> 81 1.4.2.2 nathanw #elif defined(__NetBSD__) 82 1.4.2.2 nathanw #include <sys/select.h> 83 1.4.2.2 nathanw #include <sys/malloc.h> 84 1.4.2.2 nathanw #include <sys/mount.h> 85 1.4.2.2 nathanw #include <sys/syscallargs.h> 86 1.4.2.2 nathanw #include <uvm/uvm.h> 87 1.4.2.2 nathanw #include <sys/sysctl.h> 88 1.4.2.8 nathanw #include <sys/kernel.h> 89 1.4.2.2 nathanw #endif /* NetBSD, FreeBSD */ 90 1.4.2.2 nathanw 91 1.4.2.2 nathanw #include <sys/pipe.h> 92 1.4.2.2 nathanw 93 1.4.2.2 nathanw #ifdef __NetBSD__ 94 1.4.2.8 nathanw /* 95 1.4.2.8 nathanw * Avoid microtime(9), it's slow. We don't guard the read from time(9) 96 1.4.2.8 nathanw * with splclock(9) since we don't actually need to be THAT sure the access 97 1.4.2.8 nathanw * is atomic. 98 1.4.2.8 nathanw */ 99 1.4.2.8 nathanw #define vfs_timestamp(tv) (*(tv) = time) 100 1.4.2.2 nathanw #endif 101 1.4.2.2 nathanw 102 1.4.2.2 nathanw /* 103 1.4.2.2 nathanw * Use this define if you want to disable *fancy* VM things. Expect an 104 1.4.2.2 nathanw * approx 30% decrease in transfer rate. This could be useful for 105 1.4.2.2 nathanw * OpenBSD. 106 1.4.2.2 nathanw */ 107 1.4.2.2 nathanw /* #define PIPE_NODIRECT */ 108 1.4.2.2 nathanw 109 1.4.2.2 nathanw /* 110 1.4.2.2 nathanw * interfaces to the outside world 111 1.4.2.2 nathanw */ 112 1.4.2.2 nathanw #ifdef __FreeBSD__ 113 1.4.2.2 nathanw static int pipe_read __P((struct file *fp, struct uio *uio, 114 1.4.2.2 nathanw struct ucred *cred, int flags, struct proc *p)); 115 1.4.2.2 nathanw static int pipe_write __P((struct file *fp, struct uio *uio, 116 1.4.2.2 nathanw struct ucred *cred, int flags, struct proc *p)); 117 1.4.2.2 nathanw static int pipe_close __P((struct file *fp, struct proc *p)); 118 1.4.2.2 nathanw static int pipe_poll __P((struct file *fp, int events, struct ucred *cred, 119 1.4.2.2 nathanw struct proc *p)); 120 1.4.2.2 nathanw static int pipe_kqfilter __P((struct file *fp, struct knote *kn)); 121 1.4.2.2 nathanw static int pipe_stat __P((struct file *fp, struct stat *sb, struct proc *p)); 122 1.4.2.2 nathanw static int pipe_ioctl __P((struct file *fp, u_long cmd, caddr_t data, struct proc *p)); 123 1.4.2.2 nathanw 124 1.4.2.2 nathanw static struct fileops pipeops = { 125 1.4.2.2 nathanw pipe_read, pipe_write, pipe_ioctl, pipe_poll, pipe_kqfilter, 126 1.4.2.2 nathanw pipe_stat, pipe_close 127 1.4.2.2 nathanw }; 128 1.4.2.2 nathanw 129 1.4.2.2 nathanw static void filt_pipedetach(struct knote *kn); 130 1.4.2.2 nathanw static int filt_piperead(struct knote *kn, long hint); 131 1.4.2.2 nathanw static int filt_pipewrite(struct knote *kn, long hint); 132 1.4.2.2 nathanw 133 1.4.2.2 nathanw static struct filterops pipe_rfiltops = 134 1.4.2.2 nathanw { 1, NULL, filt_pipedetach, filt_piperead }; 135 1.4.2.2 nathanw static struct filterops pipe_wfiltops = 136 1.4.2.2 nathanw { 1, NULL, filt_pipedetach, filt_pipewrite }; 137 1.4.2.2 nathanw #endif /* FreeBSD */ 138 1.4.2.2 nathanw 139 1.4.2.2 nathanw #ifdef __NetBSD__ 140 1.4.2.2 nathanw static int pipe_read __P((struct file *fp, off_t *offset, struct uio *uio, 141 1.4.2.2 nathanw struct ucred *cred, int flags)); 142 1.4.2.2 nathanw static int pipe_write __P((struct file *fp, off_t *offset, struct uio *uio, 143 1.4.2.2 nathanw struct ucred *cred, int flags)); 144 1.4.2.2 nathanw static int pipe_close __P((struct file *fp, struct proc *p)); 145 1.4.2.2 nathanw static int pipe_poll __P((struct file *fp, int events, struct proc *p)); 146 1.4.2.2 nathanw static int pipe_fcntl __P((struct file *fp, u_int com, caddr_t data, 147 1.4.2.2 nathanw struct proc *p)); 148 1.4.2.2 nathanw static int pipe_stat __P((struct file *fp, struct stat *sb, struct proc *p)); 149 1.4.2.2 nathanw static int pipe_ioctl __P((struct file *fp, u_long cmd, caddr_t data, struct proc *p)); 150 1.4.2.2 nathanw 151 1.4.2.2 nathanw static struct fileops pipeops = 152 1.4.2.2 nathanw { pipe_read, pipe_write, pipe_ioctl, pipe_fcntl, pipe_poll, 153 1.4.2.2 nathanw pipe_stat, pipe_close }; 154 1.4.2.2 nathanw #endif /* NetBSD */ 155 1.4.2.2 nathanw 156 1.4.2.2 nathanw /* 157 1.4.2.2 nathanw * Default pipe buffer size(s), this can be kind-of large now because pipe 158 1.4.2.2 nathanw * space is pageable. The pipe code will try to maintain locality of 159 1.4.2.2 nathanw * reference for performance reasons, so small amounts of outstanding I/O 160 1.4.2.2 nathanw * will not wipe the cache. 161 1.4.2.2 nathanw */ 162 1.4.2.2 nathanw #define MINPIPESIZE (PIPE_SIZE/3) 163 1.4.2.2 nathanw #define MAXPIPESIZE (2*PIPE_SIZE/3) 164 1.4.2.2 nathanw 165 1.4.2.2 nathanw /* 166 1.4.2.2 nathanw * Maximum amount of kva for pipes -- this is kind-of a soft limit, but 167 1.4.2.2 nathanw * is there so that on large systems, we don't exhaust it. 168 1.4.2.2 nathanw */ 169 1.4.2.2 nathanw #define MAXPIPEKVA (8*1024*1024) 170 1.4.2.2 nathanw static int maxpipekva = MAXPIPEKVA; 171 1.4.2.2 nathanw 172 1.4.2.2 nathanw /* 173 1.4.2.2 nathanw * Limit for direct transfers, we cannot, of course limit 174 1.4.2.2 nathanw * the amount of kva for pipes in general though. 175 1.4.2.2 nathanw */ 176 1.4.2.2 nathanw #define LIMITPIPEKVA (16*1024*1024) 177 1.4.2.2 nathanw static int limitpipekva = LIMITPIPEKVA; 178 1.4.2.2 nathanw 179 1.4.2.2 nathanw /* 180 1.4.2.2 nathanw * Limit the number of "big" pipes 181 1.4.2.2 nathanw */ 182 1.4.2.2 nathanw #define LIMITBIGPIPES 32 183 1.4.2.2 nathanw static int maxbigpipes = LIMITBIGPIPES; 184 1.4.2.2 nathanw static int nbigpipe = 0; 185 1.4.2.2 nathanw 186 1.4.2.2 nathanw /* 187 1.4.2.2 nathanw * Amount of KVA consumed by pipe buffers. 188 1.4.2.2 nathanw */ 189 1.4.2.2 nathanw static int amountpipekva = 0; 190 1.4.2.2 nathanw 191 1.4.2.8 nathanw static void pipeclose __P((struct pipe *)); 192 1.4.2.8 nathanw static void pipe_free_kmem __P((struct pipe *)); 193 1.4.2.8 nathanw static int pipe_create __P((struct pipe **, int)); 194 1.4.2.8 nathanw static __inline int pipelock __P((struct pipe *, int)); 195 1.4.2.8 nathanw static __inline void pipeunlock __P((struct pipe *)); 196 1.4.2.8 nathanw static __inline void pipeselwakeup __P((struct pipe *, struct pipe *)); 197 1.4.2.8 nathanw static int pipespace __P((struct pipe *, int)); 198 1.4.2.2 nathanw 199 1.4.2.2 nathanw #ifdef __FreeBSD__ 200 1.4.2.2 nathanw #ifndef PIPE_NODIRECT 201 1.4.2.2 nathanw static int pipe_build_write_buffer __P((struct pipe *wpipe, struct uio *uio)); 202 1.4.2.2 nathanw static void pipe_destroy_write_buffer __P((struct pipe *wpipe)); 203 1.4.2.2 nathanw static int pipe_direct_write __P((struct pipe *wpipe, struct uio *uio)); 204 1.4.2.2 nathanw static void pipe_clone_write_buffer __P((struct pipe *wpipe)); 205 1.4.2.2 nathanw #endif 206 1.4.2.2 nathanw 207 1.4.2.2 nathanw static vm_zone_t pipe_zone; 208 1.4.2.2 nathanw #endif /* FreeBSD */ 209 1.4.2.2 nathanw 210 1.4.2.2 nathanw #ifdef __NetBSD__ 211 1.4.2.2 nathanw #ifndef PIPE_NODIRECT 212 1.4.2.8 nathanw static int pipe_direct_write __P((struct pipe *, struct uio *)); 213 1.4.2.8 nathanw static int pipe_loan_alloc __P((struct pipe *, int)); 214 1.4.2.8 nathanw static void pipe_loan_free __P((struct pipe *)); 215 1.4.2.2 nathanw #endif /* PIPE_NODIRECT */ 216 1.4.2.2 nathanw 217 1.4.2.2 nathanw static struct pool pipe_pool; 218 1.4.2.2 nathanw #endif /* NetBSD */ 219 1.4.2.2 nathanw 220 1.4.2.2 nathanw /* 221 1.4.2.2 nathanw * The pipe system call for the DTYPE_PIPE type of pipes 222 1.4.2.2 nathanw */ 223 1.4.2.2 nathanw 224 1.4.2.2 nathanw /* ARGSUSED */ 225 1.4.2.2 nathanw #ifdef __FreeBSD__ 226 1.4.2.2 nathanw int 227 1.4.2.2 nathanw pipe(p, uap) 228 1.4.2.2 nathanw struct proc *p; 229 1.4.2.2 nathanw struct pipe_args /* { 230 1.4.2.2 nathanw int dummy; 231 1.4.2.2 nathanw } */ *uap; 232 1.4.2.2 nathanw #elif defined(__NetBSD__) 233 1.4.2.2 nathanw int 234 1.4.2.2 nathanw sys_pipe(p, v, retval) 235 1.4.2.2 nathanw struct proc *p; 236 1.4.2.2 nathanw void *v; 237 1.4.2.2 nathanw register_t *retval; 238 1.4.2.2 nathanw #endif 239 1.4.2.2 nathanw { 240 1.4.2.2 nathanw struct file *rf, *wf; 241 1.4.2.2 nathanw struct pipe *rpipe, *wpipe; 242 1.4.2.2 nathanw int fd, error; 243 1.4.2.2 nathanw 244 1.4.2.2 nathanw #ifdef __FreeBSD__ 245 1.4.2.2 nathanw if (pipe_zone == NULL) 246 1.4.2.2 nathanw pipe_zone = zinit("PIPE", sizeof(struct pipe), 0, 0, 4); 247 1.4.2.2 nathanw 248 1.4.2.2 nathanw rpipe = wpipe = NULL; 249 1.4.2.3 nathanw if (pipe_create(&rpipe, 1) || pipe_create(&wpipe, 1)) { 250 1.4.2.2 nathanw pipeclose(rpipe); 251 1.4.2.2 nathanw pipeclose(wpipe); 252 1.4.2.2 nathanw return (ENFILE); 253 1.4.2.2 nathanw } 254 1.4.2.2 nathanw 255 1.4.2.2 nathanw error = falloc(p, &rf, &fd); 256 1.4.2.2 nathanw if (error) { 257 1.4.2.2 nathanw pipeclose(rpipe); 258 1.4.2.2 nathanw pipeclose(wpipe); 259 1.4.2.2 nathanw return (error); 260 1.4.2.2 nathanw } 261 1.4.2.2 nathanw fhold(rf); 262 1.4.2.2 nathanw p->p_retval[0] = fd; 263 1.4.2.2 nathanw 264 1.4.2.2 nathanw /* 265 1.4.2.2 nathanw * Warning: once we've gotten past allocation of the fd for the 266 1.4.2.2 nathanw * read-side, we can only drop the read side via fdrop() in order 267 1.4.2.2 nathanw * to avoid races against processes which manage to dup() the read 268 1.4.2.2 nathanw * side while we are blocked trying to allocate the write side. 269 1.4.2.2 nathanw */ 270 1.4.2.2 nathanw rf->f_flag = FREAD | FWRITE; 271 1.4.2.2 nathanw rf->f_type = DTYPE_PIPE; 272 1.4.2.2 nathanw rf->f_data = (caddr_t)rpipe; 273 1.4.2.2 nathanw rf->f_ops = &pipeops; 274 1.4.2.2 nathanw error = falloc(p, &wf, &fd); 275 1.4.2.2 nathanw if (error) { 276 1.4.2.3 nathanw struct filedesc *fdp = p->p_fd; 277 1.4.2.3 nathanw 278 1.4.2.2 nathanw if (fdp->fd_ofiles[p->p_retval[0]] == rf) { 279 1.4.2.2 nathanw fdp->fd_ofiles[p->p_retval[0]] = NULL; 280 1.4.2.2 nathanw fdrop(rf, p); 281 1.4.2.2 nathanw } 282 1.4.2.2 nathanw fdrop(rf, p); 283 1.4.2.2 nathanw /* rpipe has been closed by fdrop(). */ 284 1.4.2.2 nathanw pipeclose(wpipe); 285 1.4.2.2 nathanw return (error); 286 1.4.2.2 nathanw } 287 1.4.2.2 nathanw wf->f_flag = FREAD | FWRITE; 288 1.4.2.2 nathanw wf->f_type = DTYPE_PIPE; 289 1.4.2.2 nathanw wf->f_data = (caddr_t)wpipe; 290 1.4.2.2 nathanw wf->f_ops = &pipeops; 291 1.4.2.2 nathanw p->p_retval[1] = fd; 292 1.4.2.2 nathanw 293 1.4.2.2 nathanw rpipe->pipe_peer = wpipe; 294 1.4.2.2 nathanw wpipe->pipe_peer = rpipe; 295 1.4.2.2 nathanw fdrop(rf, p); 296 1.4.2.2 nathanw #endif /* FreeBSD */ 297 1.4.2.2 nathanw 298 1.4.2.2 nathanw #ifdef __NetBSD__ 299 1.4.2.3 nathanw rpipe = wpipe = NULL; 300 1.4.2.3 nathanw if (pipe_create(&rpipe, 1) || pipe_create(&wpipe, 0)) { 301 1.4.2.3 nathanw pipeclose(rpipe); 302 1.4.2.3 nathanw pipeclose(wpipe); 303 1.4.2.3 nathanw return (ENFILE); 304 1.4.2.3 nathanw } 305 1.4.2.3 nathanw 306 1.4.2.2 nathanw /* 307 1.4.2.2 nathanw * Note: the file structure returned from falloc() is marked 308 1.4.2.2 nathanw * as 'larval' initially. Unless we mark it as 'mature' by 309 1.4.2.2 nathanw * FILE_SET_MATURE(), any attempt to do anything with it would 310 1.4.2.2 nathanw * return EBADF, including e.g. dup(2) or close(2). This avoids 311 1.4.2.2 nathanw * file descriptor races if we block in the second falloc(). 312 1.4.2.2 nathanw */ 313 1.4.2.2 nathanw 314 1.4.2.2 nathanw error = falloc(p, &rf, &fd); 315 1.4.2.2 nathanw if (error) 316 1.4.2.2 nathanw goto free2; 317 1.4.2.2 nathanw retval[0] = fd; 318 1.4.2.2 nathanw rf->f_flag = FREAD; 319 1.4.2.2 nathanw rf->f_type = DTYPE_PIPE; 320 1.4.2.2 nathanw rf->f_data = (caddr_t)rpipe; 321 1.4.2.2 nathanw rf->f_ops = &pipeops; 322 1.4.2.2 nathanw 323 1.4.2.2 nathanw error = falloc(p, &wf, &fd); 324 1.4.2.2 nathanw if (error) 325 1.4.2.2 nathanw goto free3; 326 1.4.2.2 nathanw retval[1] = fd; 327 1.4.2.2 nathanw wf->f_flag = FWRITE; 328 1.4.2.2 nathanw wf->f_type = DTYPE_PIPE; 329 1.4.2.2 nathanw wf->f_data = (caddr_t)wpipe; 330 1.4.2.2 nathanw wf->f_ops = &pipeops; 331 1.4.2.2 nathanw 332 1.4.2.2 nathanw rpipe->pipe_peer = wpipe; 333 1.4.2.2 nathanw wpipe->pipe_peer = rpipe; 334 1.4.2.2 nathanw 335 1.4.2.2 nathanw FILE_SET_MATURE(rf); 336 1.4.2.2 nathanw FILE_SET_MATURE(wf); 337 1.4.2.2 nathanw FILE_UNUSE(rf, p); 338 1.4.2.2 nathanw FILE_UNUSE(wf, p); 339 1.4.2.2 nathanw return (0); 340 1.4.2.2 nathanw free3: 341 1.4.2.2 nathanw FILE_UNUSE(rf, p); 342 1.4.2.2 nathanw ffree(rf); 343 1.4.2.3 nathanw fdremove(p->p_fd, retval[0]); 344 1.4.2.2 nathanw free2: 345 1.4.2.2 nathanw pipeclose(wpipe); 346 1.4.2.2 nathanw pipeclose(rpipe); 347 1.4.2.2 nathanw #endif /* NetBSD */ 348 1.4.2.2 nathanw 349 1.4.2.2 nathanw return (error); 350 1.4.2.2 nathanw } 351 1.4.2.2 nathanw 352 1.4.2.2 nathanw /* 353 1.4.2.2 nathanw * Allocate kva for pipe circular buffer, the space is pageable 354 1.4.2.2 nathanw * This routine will 'realloc' the size of a pipe safely, if it fails 355 1.4.2.2 nathanw * it will retain the old buffer. 356 1.4.2.2 nathanw * If it fails it will return ENOMEM. 357 1.4.2.2 nathanw */ 358 1.4.2.2 nathanw static int 359 1.4.2.2 nathanw pipespace(cpipe, size) 360 1.4.2.2 nathanw struct pipe *cpipe; 361 1.4.2.2 nathanw int size; 362 1.4.2.2 nathanw { 363 1.4.2.2 nathanw caddr_t buffer; 364 1.4.2.2 nathanw #ifdef __FreeBSD__ 365 1.4.2.2 nathanw struct vm_object *object; 366 1.4.2.2 nathanw int npages, error; 367 1.4.2.2 nathanw 368 1.4.2.2 nathanw npages = round_page(size)/PAGE_SIZE; 369 1.4.2.2 nathanw /* 370 1.4.2.2 nathanw * Create an object, I don't like the idea of paging to/from 371 1.4.2.2 nathanw * kernel_object. 372 1.4.2.2 nathanw */ 373 1.4.2.2 nathanw mtx_lock(&vm_mtx); 374 1.4.2.2 nathanw object = vm_object_allocate(OBJT_DEFAULT, npages); 375 1.4.2.2 nathanw buffer = (caddr_t) vm_map_min(kernel_map); 376 1.4.2.2 nathanw 377 1.4.2.2 nathanw /* 378 1.4.2.2 nathanw * Insert the object into the kernel map, and allocate kva for it. 379 1.4.2.2 nathanw * The map entry is, by default, pageable. 380 1.4.2.2 nathanw */ 381 1.4.2.2 nathanw error = vm_map_find(kernel_map, object, 0, 382 1.4.2.2 nathanw (vm_offset_t *) &buffer, size, 1, 383 1.4.2.2 nathanw VM_PROT_ALL, VM_PROT_ALL, 0); 384 1.4.2.2 nathanw 385 1.4.2.2 nathanw if (error != KERN_SUCCESS) { 386 1.4.2.2 nathanw vm_object_deallocate(object); 387 1.4.2.2 nathanw mtx_unlock(&vm_mtx); 388 1.4.2.2 nathanw return (ENOMEM); 389 1.4.2.2 nathanw } 390 1.4.2.2 nathanw #endif /* FreeBSD */ 391 1.4.2.2 nathanw 392 1.4.2.2 nathanw #ifdef __NetBSD__ 393 1.4.2.2 nathanw /* 394 1.4.2.2 nathanw * Allocate pageable virtual address space. Physical memory is allocated 395 1.4.2.2 nathanw * on demand. 396 1.4.2.2 nathanw */ 397 1.4.2.2 nathanw buffer = (caddr_t) uvm_km_valloc(kernel_map, round_page(size)); 398 1.4.2.2 nathanw if (buffer == NULL) 399 1.4.2.2 nathanw return (ENOMEM); 400 1.4.2.2 nathanw #endif /* NetBSD */ 401 1.4.2.2 nathanw 402 1.4.2.2 nathanw /* free old resources if we're resizing */ 403 1.4.2.2 nathanw pipe_free_kmem(cpipe); 404 1.4.2.2 nathanw #ifdef __FreeBSD__ 405 1.4.2.2 nathanw mtx_unlock(&vm_mtx); 406 1.4.2.2 nathanw cpipe->pipe_buffer.object = object; 407 1.4.2.2 nathanw #endif 408 1.4.2.2 nathanw cpipe->pipe_buffer.buffer = buffer; 409 1.4.2.2 nathanw cpipe->pipe_buffer.size = size; 410 1.4.2.2 nathanw cpipe->pipe_buffer.in = 0; 411 1.4.2.2 nathanw cpipe->pipe_buffer.out = 0; 412 1.4.2.2 nathanw cpipe->pipe_buffer.cnt = 0; 413 1.4.2.2 nathanw amountpipekva += cpipe->pipe_buffer.size; 414 1.4.2.2 nathanw return (0); 415 1.4.2.2 nathanw } 416 1.4.2.2 nathanw 417 1.4.2.2 nathanw /* 418 1.4.2.2 nathanw * initialize and allocate VM and memory for pipe 419 1.4.2.2 nathanw */ 420 1.4.2.2 nathanw static int 421 1.4.2.3 nathanw pipe_create(cpipep, allockva) 422 1.4.2.2 nathanw struct pipe **cpipep; 423 1.4.2.3 nathanw int allockva; 424 1.4.2.2 nathanw { 425 1.4.2.2 nathanw struct pipe *cpipe; 426 1.4.2.2 nathanw int error; 427 1.4.2.2 nathanw 428 1.4.2.2 nathanw #ifdef __FreeBSD__ 429 1.4.2.2 nathanw *cpipep = zalloc(pipe_zone); 430 1.4.2.2 nathanw #endif 431 1.4.2.2 nathanw #ifdef __NetBSD__ 432 1.4.2.2 nathanw *cpipep = pool_get(&pipe_pool, M_WAITOK); 433 1.4.2.2 nathanw #endif 434 1.4.2.2 nathanw if (*cpipep == NULL) 435 1.4.2.2 nathanw return (ENOMEM); 436 1.4.2.2 nathanw 437 1.4.2.2 nathanw cpipe = *cpipep; 438 1.4.2.2 nathanw 439 1.4.2.3 nathanw /* Initialize */ 440 1.4.2.3 nathanw memset(cpipe, 0, sizeof(*cpipe)); 441 1.4.2.2 nathanw cpipe->pipe_state = PIPE_SIGNALR; 442 1.4.2.2 nathanw 443 1.4.2.3 nathanw if (allockva && (error = pipespace(cpipe, PIPE_SIZE))) 444 1.4.2.2 nathanw return (error); 445 1.4.2.2 nathanw 446 1.4.2.2 nathanw vfs_timestamp(&cpipe->pipe_ctime); 447 1.4.2.2 nathanw cpipe->pipe_atime = cpipe->pipe_ctime; 448 1.4.2.2 nathanw cpipe->pipe_mtime = cpipe->pipe_ctime; 449 1.4.2.2 nathanw #ifdef __NetBSD__ 450 1.4.2.2 nathanw cpipe->pipe_pgid = NO_PID; 451 1.4.2.2 nathanw lockinit(&cpipe->pipe_lock, PRIBIO | PCATCH, "pipelk", 0, 0); 452 1.4.2.2 nathanw #endif 453 1.4.2.2 nathanw 454 1.4.2.2 nathanw return (0); 455 1.4.2.2 nathanw } 456 1.4.2.2 nathanw 457 1.4.2.2 nathanw 458 1.4.2.2 nathanw /* 459 1.4.2.2 nathanw * lock a pipe for I/O, blocking other access 460 1.4.2.2 nathanw */ 461 1.4.2.2 nathanw static __inline int 462 1.4.2.2 nathanw pipelock(cpipe, catch) 463 1.4.2.2 nathanw struct pipe *cpipe; 464 1.4.2.2 nathanw int catch; 465 1.4.2.2 nathanw { 466 1.4.2.2 nathanw int error; 467 1.4.2.2 nathanw 468 1.4.2.2 nathanw #ifdef __FreeBSD__ 469 1.4.2.2 nathanw while (cpipe->pipe_state & PIPE_LOCK) { 470 1.4.2.2 nathanw cpipe->pipe_state |= PIPE_LWANT; 471 1.4.2.2 nathanw error = tsleep(cpipe, catch ? (PRIBIO | PCATCH) : PRIBIO, 472 1.4.2.2 nathanw "pipelk", 0); 473 1.4.2.2 nathanw if (error != 0) 474 1.4.2.2 nathanw return (error); 475 1.4.2.2 nathanw } 476 1.4.2.2 nathanw cpipe->pipe_state |= PIPE_LOCK; 477 1.4.2.2 nathanw return (0); 478 1.4.2.2 nathanw #endif 479 1.4.2.2 nathanw 480 1.4.2.2 nathanw #ifdef __NetBSD__ 481 1.4.2.2 nathanw do { 482 1.4.2.2 nathanw error = lockmgr(&cpipe->pipe_lock, LK_EXCLUSIVE, NULL); 483 1.4.2.2 nathanw } while (!catch && (error == EINTR || error == ERESTART)); 484 1.4.2.2 nathanw return (error); 485 1.4.2.2 nathanw #endif 486 1.4.2.2 nathanw } 487 1.4.2.2 nathanw 488 1.4.2.2 nathanw /* 489 1.4.2.2 nathanw * unlock a pipe I/O lock 490 1.4.2.2 nathanw */ 491 1.4.2.2 nathanw static __inline void 492 1.4.2.2 nathanw pipeunlock(cpipe) 493 1.4.2.2 nathanw struct pipe *cpipe; 494 1.4.2.2 nathanw { 495 1.4.2.2 nathanw #ifdef __FreeBSD__ 496 1.4.2.2 nathanw cpipe->pipe_state &= ~PIPE_LOCK; 497 1.4.2.2 nathanw if (cpipe->pipe_state & PIPE_LWANT) { 498 1.4.2.2 nathanw cpipe->pipe_state &= ~PIPE_LWANT; 499 1.4.2.2 nathanw wakeup(cpipe); 500 1.4.2.2 nathanw } 501 1.4.2.2 nathanw #endif 502 1.4.2.2 nathanw 503 1.4.2.2 nathanw #ifdef __NetBSD__ 504 1.4.2.2 nathanw lockmgr(&cpipe->pipe_lock, LK_RELEASE, NULL); 505 1.4.2.2 nathanw #endif 506 1.4.2.2 nathanw } 507 1.4.2.2 nathanw 508 1.4.2.2 nathanw /* 509 1.4.2.2 nathanw * Select/poll wakup. This also sends SIGIO to peer connected to 510 1.4.2.2 nathanw * 'sigpipe' side of pipe. 511 1.4.2.2 nathanw */ 512 1.4.2.2 nathanw static __inline void 513 1.4.2.3 nathanw pipeselwakeup(selp, sigp) 514 1.4.2.3 nathanw struct pipe *selp, *sigp; 515 1.4.2.2 nathanw { 516 1.4.2.3 nathanw if (selp->pipe_state & PIPE_SEL) { 517 1.4.2.3 nathanw selp->pipe_state &= ~PIPE_SEL; 518 1.4.2.3 nathanw selwakeup(&selp->pipe_sel); 519 1.4.2.2 nathanw } 520 1.4.2.2 nathanw #ifdef __FreeBSD__ 521 1.4.2.3 nathanw if (sigp && (sigp->pipe_state & PIPE_ASYNC) && sigp->pipe_sigio) 522 1.4.2.3 nathanw pgsigio(sigp->pipe_sigio, SIGIO, 0); 523 1.4.2.3 nathanw KNOTE(&selp->pipe_sel.si_note, 0); 524 1.4.2.2 nathanw #endif 525 1.4.2.2 nathanw 526 1.4.2.2 nathanw #ifdef __NetBSD__ 527 1.4.2.3 nathanw if (sigp && (sigp->pipe_state & PIPE_ASYNC) 528 1.4.2.3 nathanw && sigp->pipe_pgid != NO_PID){ 529 1.4.2.2 nathanw struct proc *p; 530 1.4.2.2 nathanw 531 1.4.2.3 nathanw if (sigp->pipe_pgid < 0) 532 1.4.2.3 nathanw gsignal(-sigp->pipe_pgid, SIGIO); 533 1.4.2.3 nathanw else if (sigp->pipe_pgid > 0 && (p = pfind(sigp->pipe_pgid)) != 0) 534 1.4.2.2 nathanw psignal(p, SIGIO); 535 1.4.2.2 nathanw } 536 1.4.2.2 nathanw #endif /* NetBSD */ 537 1.4.2.2 nathanw } 538 1.4.2.2 nathanw 539 1.4.2.2 nathanw /* ARGSUSED */ 540 1.4.2.2 nathanw #ifdef __FreeBSD__ 541 1.4.2.2 nathanw static int 542 1.4.2.2 nathanw pipe_read(fp, uio, cred, flags, p) 543 1.4.2.2 nathanw struct file *fp; 544 1.4.2.2 nathanw struct uio *uio; 545 1.4.2.2 nathanw struct ucred *cred; 546 1.4.2.2 nathanw int flags; 547 1.4.2.2 nathanw struct proc *p; 548 1.4.2.2 nathanw #elif defined(__NetBSD__) 549 1.4.2.2 nathanw static int 550 1.4.2.2 nathanw pipe_read(fp, offset, uio, cred, flags) 551 1.4.2.2 nathanw struct file *fp; 552 1.4.2.2 nathanw off_t *offset; 553 1.4.2.2 nathanw struct uio *uio; 554 1.4.2.2 nathanw struct ucred *cred; 555 1.4.2.2 nathanw int flags; 556 1.4.2.2 nathanw #endif 557 1.4.2.2 nathanw { 558 1.4.2.2 nathanw struct pipe *rpipe = (struct pipe *) fp->f_data; 559 1.4.2.2 nathanw int error; 560 1.4.2.2 nathanw size_t nread = 0; 561 1.4.2.2 nathanw size_t size; 562 1.4.2.2 nathanw size_t ocnt; 563 1.4.2.2 nathanw 564 1.4.2.2 nathanw ++rpipe->pipe_busy; 565 1.4.2.2 nathanw error = pipelock(rpipe, 1); 566 1.4.2.2 nathanw if (error) 567 1.4.2.2 nathanw goto unlocked_error; 568 1.4.2.2 nathanw 569 1.4.2.2 nathanw ocnt = rpipe->pipe_buffer.cnt; 570 1.4.2.2 nathanw 571 1.4.2.2 nathanw while (uio->uio_resid) { 572 1.4.2.2 nathanw /* 573 1.4.2.2 nathanw * normal pipe buffer receive 574 1.4.2.2 nathanw */ 575 1.4.2.2 nathanw if (rpipe->pipe_buffer.cnt > 0) { 576 1.4.2.2 nathanw size = rpipe->pipe_buffer.size - rpipe->pipe_buffer.out; 577 1.4.2.2 nathanw if (size > rpipe->pipe_buffer.cnt) 578 1.4.2.2 nathanw size = rpipe->pipe_buffer.cnt; 579 1.4.2.2 nathanw if (size > uio->uio_resid) 580 1.4.2.2 nathanw size = uio->uio_resid; 581 1.4.2.2 nathanw 582 1.4.2.2 nathanw error = uiomove(&rpipe->pipe_buffer.buffer[rpipe->pipe_buffer.out], 583 1.4.2.2 nathanw size, uio); 584 1.4.2.2 nathanw if (error) 585 1.4.2.2 nathanw break; 586 1.4.2.2 nathanw 587 1.4.2.2 nathanw rpipe->pipe_buffer.out += size; 588 1.4.2.2 nathanw if (rpipe->pipe_buffer.out >= rpipe->pipe_buffer.size) 589 1.4.2.2 nathanw rpipe->pipe_buffer.out = 0; 590 1.4.2.2 nathanw 591 1.4.2.2 nathanw rpipe->pipe_buffer.cnt -= size; 592 1.4.2.2 nathanw 593 1.4.2.2 nathanw /* 594 1.4.2.2 nathanw * If there is no more to read in the pipe, reset 595 1.4.2.2 nathanw * its pointers to the beginning. This improves 596 1.4.2.2 nathanw * cache hit stats. 597 1.4.2.2 nathanw */ 598 1.4.2.2 nathanw if (rpipe->pipe_buffer.cnt == 0) { 599 1.4.2.2 nathanw rpipe->pipe_buffer.in = 0; 600 1.4.2.2 nathanw rpipe->pipe_buffer.out = 0; 601 1.4.2.2 nathanw } 602 1.4.2.2 nathanw nread += size; 603 1.4.2.2 nathanw #ifndef PIPE_NODIRECT 604 1.4.2.2 nathanw /* 605 1.4.2.2 nathanw * Direct copy, bypassing a kernel buffer. 606 1.4.2.2 nathanw */ 607 1.4.2.2 nathanw } else if ((size = rpipe->pipe_map.cnt) && 608 1.4.2.2 nathanw (rpipe->pipe_state & PIPE_DIRECTW)) { 609 1.4.2.2 nathanw caddr_t va; 610 1.4.2.2 nathanw if (size > uio->uio_resid) 611 1.4.2.2 nathanw size = uio->uio_resid; 612 1.4.2.2 nathanw 613 1.4.2.2 nathanw va = (caddr_t) rpipe->pipe_map.kva + 614 1.4.2.2 nathanw rpipe->pipe_map.pos; 615 1.4.2.2 nathanw error = uiomove(va, size, uio); 616 1.4.2.2 nathanw if (error) 617 1.4.2.2 nathanw break; 618 1.4.2.2 nathanw nread += size; 619 1.4.2.2 nathanw rpipe->pipe_map.pos += size; 620 1.4.2.2 nathanw rpipe->pipe_map.cnt -= size; 621 1.4.2.2 nathanw if (rpipe->pipe_map.cnt == 0) { 622 1.4.2.2 nathanw rpipe->pipe_state &= ~PIPE_DIRECTW; 623 1.4.2.2 nathanw wakeup(rpipe); 624 1.4.2.2 nathanw } 625 1.4.2.2 nathanw #endif 626 1.4.2.2 nathanw } else { 627 1.4.2.2 nathanw /* 628 1.4.2.2 nathanw * detect EOF condition 629 1.4.2.2 nathanw * read returns 0 on EOF, no need to set error 630 1.4.2.2 nathanw */ 631 1.4.2.2 nathanw if (rpipe->pipe_state & PIPE_EOF) 632 1.4.2.2 nathanw break; 633 1.4.2.2 nathanw 634 1.4.2.2 nathanw /* 635 1.4.2.2 nathanw * If the "write-side" has been blocked, wake it up now. 636 1.4.2.2 nathanw */ 637 1.4.2.2 nathanw if (rpipe->pipe_state & PIPE_WANTW) { 638 1.4.2.2 nathanw rpipe->pipe_state &= ~PIPE_WANTW; 639 1.4.2.2 nathanw wakeup(rpipe); 640 1.4.2.2 nathanw } 641 1.4.2.2 nathanw 642 1.4.2.2 nathanw /* 643 1.4.2.2 nathanw * Break if some data was read. 644 1.4.2.2 nathanw */ 645 1.4.2.2 nathanw if (nread > 0) 646 1.4.2.2 nathanw break; 647 1.4.2.2 nathanw 648 1.4.2.2 nathanw /* 649 1.4.2.2 nathanw * don't block on non-blocking I/O 650 1.4.2.2 nathanw */ 651 1.4.2.2 nathanw if (fp->f_flag & FNONBLOCK) { 652 1.4.2.2 nathanw error = EAGAIN; 653 1.4.2.2 nathanw break; 654 1.4.2.2 nathanw } 655 1.4.2.2 nathanw 656 1.4.2.2 nathanw /* 657 1.4.2.2 nathanw * Unlock the pipe buffer for our remaining processing. 658 1.4.2.2 nathanw * We will either break out with an error or we will 659 1.4.2.2 nathanw * sleep and relock to loop. 660 1.4.2.2 nathanw */ 661 1.4.2.2 nathanw pipeunlock(rpipe); 662 1.4.2.2 nathanw 663 1.4.2.2 nathanw /* 664 1.4.2.2 nathanw * We want to read more, wake up select/poll. 665 1.4.2.2 nathanw */ 666 1.4.2.3 nathanw pipeselwakeup(rpipe, rpipe->pipe_peer); 667 1.4.2.2 nathanw 668 1.4.2.2 nathanw rpipe->pipe_state |= PIPE_WANTR; 669 1.4.2.2 nathanw error = tsleep(rpipe, PRIBIO | PCATCH, "piperd", 0); 670 1.4.2.2 nathanw if (error != 0 || (error = pipelock(rpipe, 1))) 671 1.4.2.2 nathanw goto unlocked_error; 672 1.4.2.2 nathanw } 673 1.4.2.2 nathanw } 674 1.4.2.2 nathanw pipeunlock(rpipe); 675 1.4.2.2 nathanw 676 1.4.2.2 nathanw if (error == 0) 677 1.4.2.2 nathanw vfs_timestamp(&rpipe->pipe_atime); 678 1.4.2.2 nathanw unlocked_error: 679 1.4.2.2 nathanw --rpipe->pipe_busy; 680 1.4.2.2 nathanw 681 1.4.2.2 nathanw /* 682 1.4.2.2 nathanw * PIPE_WANTCLOSE processing only makes sense if pipe_busy is 0. 683 1.4.2.2 nathanw */ 684 1.4.2.2 nathanw if ((rpipe->pipe_busy == 0) && (rpipe->pipe_state & PIPE_WANTCLOSE)) { 685 1.4.2.2 nathanw rpipe->pipe_state &= ~(PIPE_WANTCLOSE|PIPE_WANTW); 686 1.4.2.2 nathanw wakeup(rpipe); 687 1.4.2.2 nathanw } else if (rpipe->pipe_buffer.cnt < MINPIPESIZE) { 688 1.4.2.2 nathanw /* 689 1.4.2.2 nathanw * Handle write blocking hysteresis. 690 1.4.2.2 nathanw */ 691 1.4.2.2 nathanw if (rpipe->pipe_state & PIPE_WANTW) { 692 1.4.2.2 nathanw rpipe->pipe_state &= ~PIPE_WANTW; 693 1.4.2.2 nathanw wakeup(rpipe); 694 1.4.2.2 nathanw } 695 1.4.2.2 nathanw } 696 1.4.2.2 nathanw 697 1.4.2.2 nathanw /* 698 1.4.2.2 nathanw * If anything was read off the buffer, signal to the writer it's 699 1.4.2.2 nathanw * possible to write more data. Also send signal if we are here for the 700 1.4.2.2 nathanw * first time after last write. 701 1.4.2.2 nathanw */ 702 1.4.2.2 nathanw if ((rpipe->pipe_buffer.size - rpipe->pipe_buffer.cnt) >= PIPE_BUF 703 1.4.2.2 nathanw && (ocnt != rpipe->pipe_buffer.cnt || (rpipe->pipe_state & PIPE_SIGNALR))) { 704 1.4.2.3 nathanw pipeselwakeup(rpipe, rpipe->pipe_peer); 705 1.4.2.2 nathanw rpipe->pipe_state &= ~PIPE_SIGNALR; 706 1.4.2.2 nathanw } 707 1.4.2.2 nathanw 708 1.4.2.2 nathanw return (error); 709 1.4.2.2 nathanw } 710 1.4.2.2 nathanw 711 1.4.2.2 nathanw #ifdef __FreeBSD__ 712 1.4.2.2 nathanw #ifndef PIPE_NODIRECT 713 1.4.2.2 nathanw /* 714 1.4.2.2 nathanw * Map the sending processes' buffer into kernel space and wire it. 715 1.4.2.2 nathanw * This is similar to a physical write operation. 716 1.4.2.2 nathanw */ 717 1.4.2.2 nathanw static int 718 1.4.2.2 nathanw pipe_build_write_buffer(wpipe, uio) 719 1.4.2.2 nathanw struct pipe *wpipe; 720 1.4.2.2 nathanw struct uio *uio; 721 1.4.2.2 nathanw { 722 1.4.2.2 nathanw size_t size; 723 1.4.2.2 nathanw int i; 724 1.4.2.2 nathanw vm_offset_t addr, endaddr, paddr; 725 1.4.2.2 nathanw 726 1.4.2.2 nathanw size = uio->uio_iov->iov_len; 727 1.4.2.2 nathanw if (size > wpipe->pipe_buffer.size) 728 1.4.2.2 nathanw size = wpipe->pipe_buffer.size; 729 1.4.2.2 nathanw 730 1.4.2.2 nathanw endaddr = round_page((vm_offset_t)uio->uio_iov->iov_base + size); 731 1.4.2.2 nathanw mtx_lock(&vm_mtx); 732 1.4.2.2 nathanw addr = trunc_page((vm_offset_t)uio->uio_iov->iov_base); 733 1.4.2.2 nathanw for (i = 0; addr < endaddr; addr += PAGE_SIZE, i++) { 734 1.4.2.2 nathanw vm_page_t m; 735 1.4.2.2 nathanw 736 1.4.2.2 nathanw if (vm_fault_quick((caddr_t)addr, VM_PROT_READ) < 0 || 737 1.4.2.2 nathanw (paddr = pmap_kextract(addr)) == 0) { 738 1.4.2.2 nathanw int j; 739 1.4.2.2 nathanw 740 1.4.2.2 nathanw for (j = 0; j < i; j++) 741 1.4.2.2 nathanw vm_page_unwire(wpipe->pipe_map.ms[j], 1); 742 1.4.2.2 nathanw mtx_unlock(&vm_mtx); 743 1.4.2.2 nathanw return (EFAULT); 744 1.4.2.2 nathanw } 745 1.4.2.2 nathanw 746 1.4.2.2 nathanw m = PHYS_TO_VM_PAGE(paddr); 747 1.4.2.2 nathanw vm_page_wire(m); 748 1.4.2.2 nathanw wpipe->pipe_map.ms[i] = m; 749 1.4.2.2 nathanw } 750 1.4.2.2 nathanw 751 1.4.2.2 nathanw /* 752 1.4.2.2 nathanw * set up the control block 753 1.4.2.2 nathanw */ 754 1.4.2.2 nathanw wpipe->pipe_map.npages = i; 755 1.4.2.2 nathanw wpipe->pipe_map.pos = 756 1.4.2.2 nathanw ((vm_offset_t) uio->uio_iov->iov_base) & PAGE_MASK; 757 1.4.2.2 nathanw wpipe->pipe_map.cnt = size; 758 1.4.2.2 nathanw 759 1.4.2.2 nathanw /* 760 1.4.2.2 nathanw * and map the buffer 761 1.4.2.2 nathanw */ 762 1.4.2.2 nathanw if (wpipe->pipe_map.kva == 0) { 763 1.4.2.2 nathanw /* 764 1.4.2.2 nathanw * We need to allocate space for an extra page because the 765 1.4.2.2 nathanw * address range might (will) span pages at times. 766 1.4.2.2 nathanw */ 767 1.4.2.2 nathanw wpipe->pipe_map.kva = kmem_alloc_pageable(kernel_map, 768 1.4.2.2 nathanw wpipe->pipe_buffer.size + PAGE_SIZE); 769 1.4.2.2 nathanw amountpipekva += wpipe->pipe_buffer.size + PAGE_SIZE; 770 1.4.2.2 nathanw } 771 1.4.2.2 nathanw pmap_qenter(wpipe->pipe_map.kva, wpipe->pipe_map.ms, 772 1.4.2.2 nathanw wpipe->pipe_map.npages); 773 1.4.2.2 nathanw 774 1.4.2.2 nathanw mtx_unlock(&vm_mtx); 775 1.4.2.2 nathanw /* 776 1.4.2.2 nathanw * and update the uio data 777 1.4.2.2 nathanw */ 778 1.4.2.2 nathanw 779 1.4.2.2 nathanw uio->uio_iov->iov_len -= size; 780 1.4.2.2 nathanw uio->uio_iov->iov_base += size; 781 1.4.2.2 nathanw if (uio->uio_iov->iov_len == 0) 782 1.4.2.2 nathanw uio->uio_iov++; 783 1.4.2.2 nathanw uio->uio_resid -= size; 784 1.4.2.2 nathanw uio->uio_offset += size; 785 1.4.2.2 nathanw return (0); 786 1.4.2.2 nathanw } 787 1.4.2.2 nathanw 788 1.4.2.2 nathanw /* 789 1.4.2.2 nathanw * unmap and unwire the process buffer 790 1.4.2.2 nathanw */ 791 1.4.2.2 nathanw static void 792 1.4.2.2 nathanw pipe_destroy_write_buffer(wpipe) 793 1.4.2.2 nathanw struct pipe *wpipe; 794 1.4.2.2 nathanw { 795 1.4.2.2 nathanw int i; 796 1.4.2.2 nathanw 797 1.4.2.2 nathanw mtx_lock(&vm_mtx); 798 1.4.2.2 nathanw if (wpipe->pipe_map.kva) { 799 1.4.2.2 nathanw pmap_qremove(wpipe->pipe_map.kva, wpipe->pipe_map.npages); 800 1.4.2.2 nathanw 801 1.4.2.2 nathanw if (amountpipekva > maxpipekva) { 802 1.4.2.2 nathanw vm_offset_t kva = wpipe->pipe_map.kva; 803 1.4.2.2 nathanw wpipe->pipe_map.kva = 0; 804 1.4.2.2 nathanw kmem_free(kernel_map, kva, 805 1.4.2.2 nathanw wpipe->pipe_buffer.size + PAGE_SIZE); 806 1.4.2.2 nathanw amountpipekva -= wpipe->pipe_buffer.size + PAGE_SIZE; 807 1.4.2.2 nathanw } 808 1.4.2.2 nathanw } 809 1.4.2.2 nathanw for (i = 0; i < wpipe->pipe_map.npages; i++) 810 1.4.2.2 nathanw vm_page_unwire(wpipe->pipe_map.ms[i], 1); 811 1.4.2.2 nathanw mtx_unlock(&vm_mtx); 812 1.4.2.2 nathanw } 813 1.4.2.2 nathanw 814 1.4.2.2 nathanw /* 815 1.4.2.2 nathanw * In the case of a signal, the writing process might go away. This 816 1.4.2.2 nathanw * code copies the data into the circular buffer so that the source 817 1.4.2.2 nathanw * pages can be freed without loss of data. 818 1.4.2.2 nathanw */ 819 1.4.2.2 nathanw static void 820 1.4.2.2 nathanw pipe_clone_write_buffer(wpipe) 821 1.4.2.2 nathanw struct pipe *wpipe; 822 1.4.2.2 nathanw { 823 1.4.2.2 nathanw int size; 824 1.4.2.2 nathanw int pos; 825 1.4.2.2 nathanw 826 1.4.2.2 nathanw size = wpipe->pipe_map.cnt; 827 1.4.2.2 nathanw pos = wpipe->pipe_map.pos; 828 1.4.2.3 nathanw memcpy((caddr_t) wpipe->pipe_buffer.buffer, 829 1.4.2.3 nathanw (caddr_t) wpipe->pipe_map.kva + pos, size); 830 1.4.2.2 nathanw 831 1.4.2.2 nathanw wpipe->pipe_buffer.in = size; 832 1.4.2.2 nathanw wpipe->pipe_buffer.out = 0; 833 1.4.2.2 nathanw wpipe->pipe_buffer.cnt = size; 834 1.4.2.2 nathanw wpipe->pipe_state &= ~PIPE_DIRECTW; 835 1.4.2.2 nathanw 836 1.4.2.2 nathanw pipe_destroy_write_buffer(wpipe); 837 1.4.2.2 nathanw } 838 1.4.2.2 nathanw 839 1.4.2.2 nathanw /* 840 1.4.2.2 nathanw * This implements the pipe buffer write mechanism. Note that only 841 1.4.2.2 nathanw * a direct write OR a normal pipe write can be pending at any given time. 842 1.4.2.2 nathanw * If there are any characters in the pipe buffer, the direct write will 843 1.4.2.2 nathanw * be deferred until the receiving process grabs all of the bytes from 844 1.4.2.2 nathanw * the pipe buffer. Then the direct mapping write is set-up. 845 1.4.2.2 nathanw */ 846 1.4.2.2 nathanw static int 847 1.4.2.2 nathanw pipe_direct_write(wpipe, uio) 848 1.4.2.2 nathanw struct pipe *wpipe; 849 1.4.2.2 nathanw struct uio *uio; 850 1.4.2.2 nathanw { 851 1.4.2.2 nathanw int error; 852 1.4.2.2 nathanw 853 1.4.2.2 nathanw retry: 854 1.4.2.2 nathanw while (wpipe->pipe_state & PIPE_DIRECTW) { 855 1.4.2.2 nathanw if (wpipe->pipe_state & PIPE_WANTR) { 856 1.4.2.2 nathanw wpipe->pipe_state &= ~PIPE_WANTR; 857 1.4.2.2 nathanw wakeup(wpipe); 858 1.4.2.2 nathanw } 859 1.4.2.2 nathanw wpipe->pipe_state |= PIPE_WANTW; 860 1.4.2.2 nathanw error = tsleep(wpipe, PRIBIO | PCATCH, "pipdww", 0); 861 1.4.2.2 nathanw if (error) 862 1.4.2.2 nathanw goto error1; 863 1.4.2.2 nathanw if (wpipe->pipe_state & PIPE_EOF) { 864 1.4.2.2 nathanw error = EPIPE; 865 1.4.2.2 nathanw goto error1; 866 1.4.2.2 nathanw } 867 1.4.2.2 nathanw } 868 1.4.2.2 nathanw wpipe->pipe_map.cnt = 0; /* transfer not ready yet */ 869 1.4.2.2 nathanw if (wpipe->pipe_buffer.cnt > 0) { 870 1.4.2.2 nathanw if (wpipe->pipe_state & PIPE_WANTR) { 871 1.4.2.2 nathanw wpipe->pipe_state &= ~PIPE_WANTR; 872 1.4.2.2 nathanw wakeup(wpipe); 873 1.4.2.2 nathanw } 874 1.4.2.8 nathanw 875 1.4.2.2 nathanw wpipe->pipe_state |= PIPE_WANTW; 876 1.4.2.2 nathanw error = tsleep(wpipe, PRIBIO | PCATCH, "pipdwc", 0); 877 1.4.2.2 nathanw if (error) 878 1.4.2.2 nathanw goto error1; 879 1.4.2.2 nathanw if (wpipe->pipe_state & PIPE_EOF) { 880 1.4.2.2 nathanw error = EPIPE; 881 1.4.2.2 nathanw goto error1; 882 1.4.2.2 nathanw } 883 1.4.2.2 nathanw goto retry; 884 1.4.2.2 nathanw } 885 1.4.2.2 nathanw 886 1.4.2.2 nathanw wpipe->pipe_state |= PIPE_DIRECTW; 887 1.4.2.2 nathanw 888 1.4.2.2 nathanw error = pipe_build_write_buffer(wpipe, uio); 889 1.4.2.2 nathanw if (error) { 890 1.4.2.2 nathanw wpipe->pipe_state &= ~PIPE_DIRECTW; 891 1.4.2.2 nathanw goto error1; 892 1.4.2.2 nathanw } 893 1.4.2.2 nathanw 894 1.4.2.2 nathanw error = 0; 895 1.4.2.2 nathanw while (!error && (wpipe->pipe_state & PIPE_DIRECTW)) { 896 1.4.2.2 nathanw if (wpipe->pipe_state & PIPE_EOF) { 897 1.4.2.2 nathanw pipelock(wpipe, 0); 898 1.4.2.2 nathanw pipe_destroy_write_buffer(wpipe); 899 1.4.2.2 nathanw pipeunlock(wpipe); 900 1.4.2.3 nathanw pipeselwakeup(wpipe, wpipe); 901 1.4.2.2 nathanw error = EPIPE; 902 1.4.2.2 nathanw goto error1; 903 1.4.2.2 nathanw } 904 1.4.2.2 nathanw if (wpipe->pipe_state & PIPE_WANTR) { 905 1.4.2.2 nathanw wpipe->pipe_state &= ~PIPE_WANTR; 906 1.4.2.2 nathanw wakeup(wpipe); 907 1.4.2.2 nathanw } 908 1.4.2.3 nathanw pipeselwakeup(wpipe, wpipe); 909 1.4.2.2 nathanw error = tsleep(wpipe, PRIBIO | PCATCH, "pipdwt", 0); 910 1.4.2.2 nathanw } 911 1.4.2.2 nathanw 912 1.4.2.2 nathanw pipelock(wpipe,0); 913 1.4.2.2 nathanw if (wpipe->pipe_state & PIPE_DIRECTW) { 914 1.4.2.2 nathanw /* 915 1.4.2.2 nathanw * this bit of trickery substitutes a kernel buffer for 916 1.4.2.2 nathanw * the process that might be going away. 917 1.4.2.2 nathanw */ 918 1.4.2.2 nathanw pipe_clone_write_buffer(wpipe); 919 1.4.2.2 nathanw } else { 920 1.4.2.2 nathanw pipe_destroy_write_buffer(wpipe); 921 1.4.2.2 nathanw } 922 1.4.2.2 nathanw pipeunlock(wpipe); 923 1.4.2.2 nathanw return (error); 924 1.4.2.2 nathanw 925 1.4.2.2 nathanw error1: 926 1.4.2.2 nathanw wakeup(wpipe); 927 1.4.2.2 nathanw return (error); 928 1.4.2.2 nathanw } 929 1.4.2.2 nathanw #endif /* !PIPE_NODIRECT */ 930 1.4.2.2 nathanw #endif /* FreeBSD */ 931 1.4.2.2 nathanw 932 1.4.2.2 nathanw #ifdef __NetBSD__ 933 1.4.2.2 nathanw #ifndef PIPE_NODIRECT 934 1.4.2.2 nathanw /* 935 1.4.2.2 nathanw * Allocate structure for loan transfer. 936 1.4.2.2 nathanw */ 937 1.4.2.8 nathanw static int 938 1.4.2.8 nathanw pipe_loan_alloc(wpipe, npages) 939 1.4.2.2 nathanw struct pipe *wpipe; 940 1.4.2.2 nathanw int npages; 941 1.4.2.2 nathanw { 942 1.4.2.8 nathanw vsize_t len; 943 1.4.2.8 nathanw 944 1.4.2.8 nathanw len = (vsize_t)npages << PAGE_SHIFT; 945 1.4.2.8 nathanw wpipe->pipe_map.kva = uvm_km_valloc_wait(kernel_map, len); 946 1.4.2.2 nathanw if (wpipe->pipe_map.kva == NULL) 947 1.4.2.2 nathanw return (ENOMEM); 948 1.4.2.2 nathanw 949 1.4.2.8 nathanw amountpipekva += len; 950 1.4.2.2 nathanw wpipe->pipe_map.npages = npages; 951 1.4.2.8 nathanw wpipe->pipe_map.pgs = malloc(npages * sizeof(struct vm_page *), M_PIPE, 952 1.4.2.8 nathanw M_WAITOK); 953 1.4.2.2 nathanw return (0); 954 1.4.2.2 nathanw } 955 1.4.2.2 nathanw 956 1.4.2.2 nathanw /* 957 1.4.2.2 nathanw * Free resources allocated for loan transfer. 958 1.4.2.2 nathanw */ 959 1.4.2.2 nathanw static void 960 1.4.2.2 nathanw pipe_loan_free(wpipe) 961 1.4.2.2 nathanw struct pipe *wpipe; 962 1.4.2.2 nathanw { 963 1.4.2.8 nathanw vsize_t len; 964 1.4.2.8 nathanw 965 1.4.2.8 nathanw len = (vsize_t)wpipe->pipe_map.npages << PAGE_SHIFT; 966 1.4.2.8 nathanw uvm_km_free(kernel_map, wpipe->pipe_map.kva, len); 967 1.4.2.2 nathanw wpipe->pipe_map.kva = NULL; 968 1.4.2.8 nathanw amountpipekva -= len; 969 1.4.2.8 nathanw free(wpipe->pipe_map.pgs, M_PIPE); 970 1.4.2.8 nathanw wpipe->pipe_map.pgs = NULL; 971 1.4.2.2 nathanw } 972 1.4.2.2 nathanw 973 1.4.2.2 nathanw /* 974 1.4.2.2 nathanw * NetBSD direct write, using uvm_loan() mechanism. 975 1.4.2.2 nathanw * This implements the pipe buffer write mechanism. Note that only 976 1.4.2.2 nathanw * a direct write OR a normal pipe write can be pending at any given time. 977 1.4.2.2 nathanw * If there are any characters in the pipe buffer, the direct write will 978 1.4.2.2 nathanw * be deferred until the receiving process grabs all of the bytes from 979 1.4.2.2 nathanw * the pipe buffer. Then the direct mapping write is set-up. 980 1.4.2.2 nathanw */ 981 1.4.2.8 nathanw static int 982 1.4.2.2 nathanw pipe_direct_write(wpipe, uio) 983 1.4.2.2 nathanw struct pipe *wpipe; 984 1.4.2.2 nathanw struct uio *uio; 985 1.4.2.2 nathanw { 986 1.4.2.3 nathanw int error, npages, j; 987 1.4.2.8 nathanw struct vm_page **pgs; 988 1.4.2.2 nathanw vaddr_t bbase, kva, base, bend; 989 1.4.2.2 nathanw vsize_t blen, bcnt; 990 1.4.2.3 nathanw voff_t bpos; 991 1.4.2.3 nathanw 992 1.4.2.2 nathanw retry: 993 1.4.2.2 nathanw while (wpipe->pipe_state & PIPE_DIRECTW) { 994 1.4.2.2 nathanw if (wpipe->pipe_state & PIPE_WANTR) { 995 1.4.2.2 nathanw wpipe->pipe_state &= ~PIPE_WANTR; 996 1.4.2.2 nathanw wakeup(wpipe); 997 1.4.2.2 nathanw } 998 1.4.2.2 nathanw wpipe->pipe_state |= PIPE_WANTW; 999 1.4.2.2 nathanw error = tsleep(wpipe, PRIBIO | PCATCH, "pipdww", 0); 1000 1.4.2.2 nathanw if (error) 1001 1.4.2.3 nathanw goto error; 1002 1.4.2.2 nathanw if (wpipe->pipe_state & PIPE_EOF) { 1003 1.4.2.2 nathanw error = EPIPE; 1004 1.4.2.3 nathanw goto error; 1005 1.4.2.2 nathanw } 1006 1.4.2.2 nathanw } 1007 1.4.2.2 nathanw wpipe->pipe_map.cnt = 0; /* transfer not ready yet */ 1008 1.4.2.2 nathanw if (wpipe->pipe_buffer.cnt > 0) { 1009 1.4.2.3 nathanw if (wpipe->pipe_state & PIPE_WANTR) { 1010 1.4.2.2 nathanw wpipe->pipe_state &= ~PIPE_WANTR; 1011 1.4.2.2 nathanw wakeup(wpipe); 1012 1.4.2.2 nathanw } 1013 1.4.2.8 nathanw 1014 1.4.2.2 nathanw wpipe->pipe_state |= PIPE_WANTW; 1015 1.4.2.2 nathanw error = tsleep(wpipe, PRIBIO | PCATCH, "pipdwc", 0); 1016 1.4.2.2 nathanw if (error) 1017 1.4.2.3 nathanw goto error; 1018 1.4.2.2 nathanw if (wpipe->pipe_state & PIPE_EOF) { 1019 1.4.2.2 nathanw error = EPIPE; 1020 1.4.2.3 nathanw goto error; 1021 1.4.2.2 nathanw } 1022 1.4.2.2 nathanw goto retry; 1023 1.4.2.2 nathanw } 1024 1.4.2.2 nathanw 1025 1.4.2.2 nathanw /* 1026 1.4.2.5 nathanw * Handle first PIPE_CHUNK_SIZE bytes of buffer. Deal with buffers 1027 1.4.2.5 nathanw * not aligned to PAGE_SIZE. 1028 1.4.2.2 nathanw */ 1029 1.4.2.5 nathanw bbase = (vaddr_t)uio->uio_iov->iov_base; 1030 1.4.2.3 nathanw base = trunc_page(bbase); 1031 1.4.2.5 nathanw bend = round_page(bbase + uio->uio_iov->iov_len); 1032 1.4.2.3 nathanw blen = bend - base; 1033 1.4.2.3 nathanw bpos = bbase - base; 1034 1.4.2.3 nathanw 1035 1.4.2.3 nathanw if (blen > PIPE_DIRECT_CHUNK) { 1036 1.4.2.3 nathanw blen = PIPE_DIRECT_CHUNK; 1037 1.4.2.3 nathanw bend = base + blen; 1038 1.4.2.3 nathanw bcnt = PIPE_DIRECT_CHUNK - bpos; 1039 1.4.2.8 nathanw } else { 1040 1.4.2.5 nathanw bcnt = uio->uio_iov->iov_len; 1041 1.4.2.8 nathanw } 1042 1.4.2.8 nathanw npages = blen >> PAGE_SHIFT; 1043 1.4.2.2 nathanw 1044 1.4.2.3 nathanw wpipe->pipe_map.pos = bpos; 1045 1.4.2.3 nathanw wpipe->pipe_map.cnt = bcnt; 1046 1.4.2.2 nathanw 1047 1.4.2.3 nathanw /* 1048 1.4.2.3 nathanw * Free the old kva if we need more pages than we have 1049 1.4.2.3 nathanw * allocated. 1050 1.4.2.3 nathanw */ 1051 1.4.2.3 nathanw if (wpipe->pipe_map.kva && npages > wpipe->pipe_map.npages) 1052 1.4.2.3 nathanw pipe_loan_free(wpipe); 1053 1.4.2.2 nathanw 1054 1.4.2.3 nathanw /* Allocate new kva. */ 1055 1.4.2.8 nathanw if (wpipe->pipe_map.kva == NULL) { 1056 1.4.2.8 nathanw error = pipe_loan_alloc(wpipe, npages); 1057 1.4.2.8 nathanw if (error) { 1058 1.4.2.8 nathanw goto error; 1059 1.4.2.8 nathanw } 1060 1.4.2.8 nathanw } 1061 1.4.2.8 nathanw 1062 1.4.2.3 nathanw /* Loan the write buffer memory from writer process */ 1063 1.4.2.8 nathanw pgs = wpipe->pipe_map.pgs; 1064 1.4.2.3 nathanw error = uvm_loan(&uio->uio_procp->p_vmspace->vm_map, base, blen, 1065 1.4.2.8 nathanw pgs, UVM_LOAN_TOPAGE); 1066 1.4.2.8 nathanw if (error) { 1067 1.4.2.8 nathanw pgs = NULL; 1068 1.4.2.3 nathanw goto cleanup; 1069 1.4.2.8 nathanw } 1070 1.4.2.8 nathanw 1071 1.4.2.3 nathanw /* Enter the loaned pages to kva */ 1072 1.4.2.3 nathanw kva = wpipe->pipe_map.kva; 1073 1.4.2.8 nathanw for (j = 0; j < npages; j++, kva += PAGE_SIZE) { 1074 1.4.2.8 nathanw pmap_kenter_pa(kva, VM_PAGE_TO_PHYS(pgs[j]), VM_PROT_READ); 1075 1.4.2.8 nathanw } 1076 1.4.2.4 nathanw pmap_update(pmap_kernel()); 1077 1.4.2.2 nathanw 1078 1.4.2.3 nathanw wpipe->pipe_state |= PIPE_DIRECTW; 1079 1.4.2.3 nathanw while (!error && (wpipe->pipe_state & PIPE_DIRECTW)) { 1080 1.4.2.3 nathanw if (wpipe->pipe_state & PIPE_EOF) { 1081 1.4.2.3 nathanw error = EPIPE; 1082 1.4.2.3 nathanw break; 1083 1.4.2.2 nathanw } 1084 1.4.2.3 nathanw if (wpipe->pipe_state & PIPE_WANTR) { 1085 1.4.2.3 nathanw wpipe->pipe_state &= ~PIPE_WANTR; 1086 1.4.2.3 nathanw wakeup(wpipe); 1087 1.4.2.3 nathanw } 1088 1.4.2.3 nathanw pipeselwakeup(wpipe, wpipe); 1089 1.4.2.3 nathanw error = tsleep(wpipe, PRIBIO | PCATCH, "pipdwt", 0); 1090 1.4.2.3 nathanw } 1091 1.4.2.3 nathanw 1092 1.4.2.3 nathanw if (error) 1093 1.4.2.3 nathanw wpipe->pipe_state &= ~PIPE_DIRECTW; 1094 1.4.2.2 nathanw 1095 1.4.2.8 nathanw cleanup: 1096 1.4.2.3 nathanw pipelock(wpipe, 0); 1097 1.4.2.9 nathanw if (pgs != NULL) { 1098 1.4.2.9 nathanw pmap_kremove(wpipe->pipe_map.kva, blen); 1099 1.4.2.8 nathanw uvm_unloan(pgs, npages, UVM_LOAN_TOPAGE); 1100 1.4.2.9 nathanw } 1101 1.4.2.3 nathanw if (error || amountpipekva > maxpipekva) 1102 1.4.2.3 nathanw pipe_loan_free(wpipe); 1103 1.4.2.3 nathanw pipeunlock(wpipe); 1104 1.4.2.2 nathanw 1105 1.4.2.6 nathanw if (error) { 1106 1.4.2.3 nathanw pipeselwakeup(wpipe, wpipe); 1107 1.4.2.3 nathanw 1108 1.4.2.3 nathanw /* 1109 1.4.2.6 nathanw * If nothing was read from what we offered, return error 1110 1.4.2.8 nathanw * straight on. Otherwise update uio resid first. Caller 1111 1.4.2.6 nathanw * will deal with the error condition, returning short 1112 1.4.2.6 nathanw * write, error, or restarting the write(2) as appropriate. 1113 1.4.2.3 nathanw */ 1114 1.4.2.6 nathanw if (wpipe->pipe_map.cnt == bcnt) { 1115 1.4.2.8 nathanw error: 1116 1.4.2.6 nathanw wakeup(wpipe); 1117 1.4.2.6 nathanw return (error); 1118 1.4.2.2 nathanw } 1119 1.4.2.2 nathanw 1120 1.4.2.6 nathanw bcnt -= wpipe->pipe_map.cnt; 1121 1.4.2.3 nathanw } 1122 1.4.2.2 nathanw 1123 1.4.2.8 nathanw uio->uio_resid -= bcnt; 1124 1.4.2.3 nathanw /* uio_offset not updated, not set/used for write(2) */ 1125 1.4.2.8 nathanw uio->uio_iov->iov_base = (char *)uio->uio_iov->iov_base + bcnt; 1126 1.4.2.5 nathanw uio->uio_iov->iov_len -= bcnt; 1127 1.4.2.5 nathanw if (uio->uio_iov->iov_len == 0) { 1128 1.4.2.5 nathanw uio->uio_iov++; 1129 1.4.2.5 nathanw uio->uio_iovcnt--; 1130 1.4.2.5 nathanw } 1131 1.4.2.2 nathanw 1132 1.4.2.6 nathanw return (error); 1133 1.4.2.2 nathanw } 1134 1.4.2.2 nathanw #endif /* !PIPE_NODIRECT */ 1135 1.4.2.2 nathanw #endif /* NetBSD */ 1136 1.4.2.2 nathanw 1137 1.4.2.2 nathanw #ifdef __FreeBSD__ 1138 1.4.2.2 nathanw static int 1139 1.4.2.2 nathanw pipe_write(fp, uio, cred, flags, p) 1140 1.4.2.2 nathanw struct file *fp; 1141 1.4.2.2 nathanw off_t *offset; 1142 1.4.2.2 nathanw struct uio *uio; 1143 1.4.2.2 nathanw struct ucred *cred; 1144 1.4.2.2 nathanw int flags; 1145 1.4.2.2 nathanw struct proc *p; 1146 1.4.2.2 nathanw #elif defined(__NetBSD__) 1147 1.4.2.2 nathanw static int 1148 1.4.2.2 nathanw pipe_write(fp, offset, uio, cred, flags) 1149 1.4.2.2 nathanw struct file *fp; 1150 1.4.2.2 nathanw off_t *offset; 1151 1.4.2.2 nathanw struct uio *uio; 1152 1.4.2.2 nathanw struct ucred *cred; 1153 1.4.2.2 nathanw int flags; 1154 1.4.2.2 nathanw #endif 1155 1.4.2.2 nathanw { 1156 1.4.2.2 nathanw int error = 0; 1157 1.4.2.2 nathanw struct pipe *wpipe, *rpipe; 1158 1.4.2.2 nathanw 1159 1.4.2.2 nathanw rpipe = (struct pipe *) fp->f_data; 1160 1.4.2.2 nathanw wpipe = rpipe->pipe_peer; 1161 1.4.2.2 nathanw 1162 1.4.2.2 nathanw /* 1163 1.4.2.2 nathanw * detect loss of pipe read side, issue SIGPIPE if lost. 1164 1.4.2.2 nathanw */ 1165 1.4.2.2 nathanw if ((wpipe == NULL) || (wpipe->pipe_state & PIPE_EOF)) 1166 1.4.2.2 nathanw return (EPIPE); 1167 1.4.2.2 nathanw 1168 1.4.2.2 nathanw ++wpipe->pipe_busy; 1169 1.4.2.2 nathanw 1170 1.4.2.2 nathanw /* 1171 1.4.2.2 nathanw * If it is advantageous to resize the pipe buffer, do 1172 1.4.2.2 nathanw * so. 1173 1.4.2.2 nathanw */ 1174 1.4.2.2 nathanw if ((uio->uio_resid > PIPE_SIZE) && 1175 1.4.2.2 nathanw (nbigpipe < maxbigpipes) && 1176 1.4.2.2 nathanw #ifndef PIPE_NODIRECT 1177 1.4.2.2 nathanw (wpipe->pipe_state & PIPE_DIRECTW) == 0 && 1178 1.4.2.2 nathanw #endif 1179 1.4.2.2 nathanw (wpipe->pipe_buffer.size <= PIPE_SIZE) && 1180 1.4.2.2 nathanw (wpipe->pipe_buffer.cnt == 0)) { 1181 1.4.2.2 nathanw 1182 1.4.2.2 nathanw if ((error = pipelock(wpipe,1)) == 0) { 1183 1.4.2.2 nathanw if (pipespace(wpipe, BIG_PIPE_SIZE) == 0) 1184 1.4.2.2 nathanw nbigpipe++; 1185 1.4.2.2 nathanw pipeunlock(wpipe); 1186 1.4.2.2 nathanw } else { 1187 1.4.2.2 nathanw /* 1188 1.4.2.3 nathanw * If an error occurred, unbusy and return, waking up 1189 1.4.2.3 nathanw * any waiting readers. 1190 1.4.2.2 nathanw */ 1191 1.4.2.2 nathanw --wpipe->pipe_busy; 1192 1.4.2.2 nathanw if (wpipe->pipe_busy == 0 1193 1.4.2.2 nathanw && (wpipe->pipe_state & PIPE_WANTCLOSE)) { 1194 1.4.2.2 nathanw wpipe->pipe_state &= 1195 1.4.2.2 nathanw ~(PIPE_WANTCLOSE | PIPE_WANTR); 1196 1.4.2.3 nathanw wakeup(wpipe); 1197 1.4.2.2 nathanw } 1198 1.4.2.2 nathanw 1199 1.4.2.2 nathanw return (error); 1200 1.4.2.2 nathanw } 1201 1.4.2.2 nathanw } 1202 1.4.2.8 nathanw 1203 1.4.2.2 nathanw #ifdef __FreeBSD__ 1204 1.4.2.2 nathanw KASSERT(wpipe->pipe_buffer.buffer != NULL, ("pipe buffer gone")); 1205 1.4.2.2 nathanw #endif 1206 1.4.2.2 nathanw 1207 1.4.2.2 nathanw while (uio->uio_resid) { 1208 1.4.2.2 nathanw int space; 1209 1.4.2.2 nathanw 1210 1.4.2.2 nathanw #ifndef PIPE_NODIRECT 1211 1.4.2.2 nathanw /* 1212 1.4.2.2 nathanw * If the transfer is large, we can gain performance if 1213 1.4.2.2 nathanw * we do process-to-process copies directly. 1214 1.4.2.2 nathanw * If the write is non-blocking, we don't use the 1215 1.4.2.2 nathanw * direct write mechanism. 1216 1.4.2.2 nathanw * 1217 1.4.2.2 nathanw * The direct write mechanism will detect the reader going 1218 1.4.2.2 nathanw * away on us. 1219 1.4.2.2 nathanw */ 1220 1.4.2.5 nathanw if ((uio->uio_iov->iov_len >= PIPE_MINDIRECT) && 1221 1.4.2.2 nathanw (fp->f_flag & FNONBLOCK) == 0 && 1222 1.4.2.2 nathanw (wpipe->pipe_map.kva || (amountpipekva < limitpipekva))) { 1223 1.4.2.2 nathanw error = pipe_direct_write(wpipe, uio); 1224 1.4.2.3 nathanw 1225 1.4.2.3 nathanw /* 1226 1.4.2.5 nathanw * Break out if error occured, unless it's ENOMEM. 1227 1.4.2.5 nathanw * ENOMEM means we failed to allocate some resources 1228 1.4.2.5 nathanw * for direct write, so we just fallback to ordinary 1229 1.4.2.5 nathanw * write. If the direct write was successful, 1230 1.4.2.5 nathanw * process rest of data via ordinary write. 1231 1.4.2.3 nathanw */ 1232 1.4.2.5 nathanw if (!error) 1233 1.4.2.5 nathanw continue; 1234 1.4.2.5 nathanw 1235 1.4.2.3 nathanw if (error != ENOMEM) 1236 1.4.2.2 nathanw break; 1237 1.4.2.2 nathanw } 1238 1.4.2.2 nathanw #endif /* PIPE_NODIRECT */ 1239 1.4.2.2 nathanw 1240 1.4.2.2 nathanw /* 1241 1.4.2.2 nathanw * Pipe buffered writes cannot be coincidental with 1242 1.4.2.2 nathanw * direct writes. We wait until the currently executing 1243 1.4.2.2 nathanw * direct write is completed before we start filling the 1244 1.4.2.2 nathanw * pipe buffer. We break out if a signal occurs or the 1245 1.4.2.2 nathanw * reader goes away. 1246 1.4.2.2 nathanw */ 1247 1.4.2.2 nathanw retrywrite: 1248 1.4.2.2 nathanw while (wpipe->pipe_state & PIPE_DIRECTW) { 1249 1.4.2.2 nathanw if (wpipe->pipe_state & PIPE_WANTR) { 1250 1.4.2.2 nathanw wpipe->pipe_state &= ~PIPE_WANTR; 1251 1.4.2.2 nathanw wakeup(wpipe); 1252 1.4.2.2 nathanw } 1253 1.4.2.2 nathanw error = tsleep(wpipe, PRIBIO | PCATCH, "pipbww", 0); 1254 1.4.2.2 nathanw if (wpipe->pipe_state & PIPE_EOF) 1255 1.4.2.2 nathanw break; 1256 1.4.2.2 nathanw if (error) 1257 1.4.2.2 nathanw break; 1258 1.4.2.2 nathanw } 1259 1.4.2.2 nathanw if (wpipe->pipe_state & PIPE_EOF) { 1260 1.4.2.2 nathanw error = EPIPE; 1261 1.4.2.2 nathanw break; 1262 1.4.2.2 nathanw } 1263 1.4.2.2 nathanw 1264 1.4.2.2 nathanw space = wpipe->pipe_buffer.size - wpipe->pipe_buffer.cnt; 1265 1.4.2.2 nathanw 1266 1.4.2.2 nathanw /* Writes of size <= PIPE_BUF must be atomic. */ 1267 1.4.2.5 nathanw if ((space < uio->uio_resid) && (uio->uio_resid <= PIPE_BUF)) 1268 1.4.2.2 nathanw space = 0; 1269 1.4.2.2 nathanw 1270 1.4.2.7 nathanw if (space > 0) { 1271 1.4.2.2 nathanw int size; /* Transfer size */ 1272 1.4.2.2 nathanw int segsize; /* first segment to transfer */ 1273 1.4.2.2 nathanw 1274 1.4.2.2 nathanw if ((error = pipelock(wpipe,1)) != 0) 1275 1.4.2.2 nathanw break; 1276 1.4.2.2 nathanw 1277 1.4.2.2 nathanw /* 1278 1.4.2.2 nathanw * It is possible for a direct write to 1279 1.4.2.2 nathanw * slip in on us... handle it here... 1280 1.4.2.2 nathanw */ 1281 1.4.2.2 nathanw if (wpipe->pipe_state & PIPE_DIRECTW) { 1282 1.4.2.2 nathanw pipeunlock(wpipe); 1283 1.4.2.2 nathanw goto retrywrite; 1284 1.4.2.2 nathanw } 1285 1.4.2.2 nathanw /* 1286 1.4.2.2 nathanw * If a process blocked in uiomove, our 1287 1.4.2.2 nathanw * value for space might be bad. 1288 1.4.2.2 nathanw * 1289 1.4.2.2 nathanw * XXX will we be ok if the reader has gone 1290 1.4.2.2 nathanw * away here? 1291 1.4.2.2 nathanw */ 1292 1.4.2.2 nathanw if (space > wpipe->pipe_buffer.size - 1293 1.4.2.2 nathanw wpipe->pipe_buffer.cnt) { 1294 1.4.2.2 nathanw pipeunlock(wpipe); 1295 1.4.2.2 nathanw goto retrywrite; 1296 1.4.2.2 nathanw } 1297 1.4.2.2 nathanw 1298 1.4.2.2 nathanw /* 1299 1.4.2.2 nathanw * Transfer size is minimum of uio transfer 1300 1.4.2.2 nathanw * and free space in pipe buffer. 1301 1.4.2.2 nathanw */ 1302 1.4.2.2 nathanw if (space > uio->uio_resid) 1303 1.4.2.2 nathanw size = uio->uio_resid; 1304 1.4.2.2 nathanw else 1305 1.4.2.2 nathanw size = space; 1306 1.4.2.2 nathanw /* 1307 1.4.2.2 nathanw * First segment to transfer is minimum of 1308 1.4.2.2 nathanw * transfer size and contiguous space in 1309 1.4.2.2 nathanw * pipe buffer. If first segment to transfer 1310 1.4.2.2 nathanw * is less than the transfer size, we've got 1311 1.4.2.2 nathanw * a wraparound in the buffer. 1312 1.4.2.2 nathanw */ 1313 1.4.2.2 nathanw segsize = wpipe->pipe_buffer.size - 1314 1.4.2.2 nathanw wpipe->pipe_buffer.in; 1315 1.4.2.2 nathanw if (segsize > size) 1316 1.4.2.2 nathanw segsize = size; 1317 1.4.2.8 nathanw 1318 1.4.2.2 nathanw /* Transfer first segment */ 1319 1.4.2.2 nathanw 1320 1.4.2.2 nathanw error = uiomove(&wpipe->pipe_buffer.buffer[wpipe->pipe_buffer.in], 1321 1.4.2.2 nathanw segsize, uio); 1322 1.4.2.8 nathanw 1323 1.4.2.2 nathanw if (error == 0 && segsize < size) { 1324 1.4.2.2 nathanw /* 1325 1.4.2.2 nathanw * Transfer remaining part now, to 1326 1.4.2.2 nathanw * support atomic writes. Wraparound 1327 1.4.2.2 nathanw * happened. 1328 1.4.2.2 nathanw */ 1329 1.4.2.2 nathanw #ifdef DEBUG 1330 1.4.2.2 nathanw if (wpipe->pipe_buffer.in + segsize != 1331 1.4.2.2 nathanw wpipe->pipe_buffer.size) 1332 1.4.2.2 nathanw panic("Expected pipe buffer wraparound disappeared"); 1333 1.4.2.2 nathanw #endif 1334 1.4.2.8 nathanw 1335 1.4.2.2 nathanw error = uiomove(&wpipe->pipe_buffer.buffer[0], 1336 1.4.2.2 nathanw size - segsize, uio); 1337 1.4.2.2 nathanw } 1338 1.4.2.2 nathanw if (error == 0) { 1339 1.4.2.2 nathanw wpipe->pipe_buffer.in += size; 1340 1.4.2.2 nathanw if (wpipe->pipe_buffer.in >= 1341 1.4.2.2 nathanw wpipe->pipe_buffer.size) { 1342 1.4.2.2 nathanw #ifdef DEBUG 1343 1.4.2.2 nathanw if (wpipe->pipe_buffer.in != size - segsize + wpipe->pipe_buffer.size) 1344 1.4.2.2 nathanw panic("Expected wraparound bad"); 1345 1.4.2.2 nathanw #endif 1346 1.4.2.2 nathanw wpipe->pipe_buffer.in = size - segsize; 1347 1.4.2.2 nathanw } 1348 1.4.2.8 nathanw 1349 1.4.2.2 nathanw wpipe->pipe_buffer.cnt += size; 1350 1.4.2.2 nathanw #ifdef DEBUG 1351 1.4.2.2 nathanw if (wpipe->pipe_buffer.cnt > wpipe->pipe_buffer.size) 1352 1.4.2.2 nathanw panic("Pipe buffer overflow"); 1353 1.4.2.2 nathanw #endif 1354 1.4.2.2 nathanw } 1355 1.4.2.2 nathanw pipeunlock(wpipe); 1356 1.4.2.2 nathanw if (error) 1357 1.4.2.2 nathanw break; 1358 1.4.2.2 nathanw } else { 1359 1.4.2.2 nathanw /* 1360 1.4.2.2 nathanw * If the "read-side" has been blocked, wake it up now. 1361 1.4.2.2 nathanw */ 1362 1.4.2.2 nathanw if (wpipe->pipe_state & PIPE_WANTR) { 1363 1.4.2.2 nathanw wpipe->pipe_state &= ~PIPE_WANTR; 1364 1.4.2.2 nathanw wakeup(wpipe); 1365 1.4.2.2 nathanw } 1366 1.4.2.2 nathanw 1367 1.4.2.2 nathanw /* 1368 1.4.2.2 nathanw * don't block on non-blocking I/O 1369 1.4.2.2 nathanw */ 1370 1.4.2.2 nathanw if (fp->f_flag & FNONBLOCK) { 1371 1.4.2.2 nathanw error = EAGAIN; 1372 1.4.2.2 nathanw break; 1373 1.4.2.2 nathanw } 1374 1.4.2.2 nathanw 1375 1.4.2.2 nathanw /* 1376 1.4.2.2 nathanw * We have no more space and have something to offer, 1377 1.4.2.2 nathanw * wake up select/poll. 1378 1.4.2.2 nathanw */ 1379 1.4.2.3 nathanw pipeselwakeup(wpipe, wpipe); 1380 1.4.2.2 nathanw 1381 1.4.2.2 nathanw wpipe->pipe_state |= PIPE_WANTW; 1382 1.4.2.2 nathanw error = tsleep(wpipe, PRIBIO | PCATCH, "pipewr", 0); 1383 1.4.2.2 nathanw if (error != 0) 1384 1.4.2.2 nathanw break; 1385 1.4.2.2 nathanw /* 1386 1.4.2.2 nathanw * If read side wants to go away, we just issue a signal 1387 1.4.2.2 nathanw * to ourselves. 1388 1.4.2.2 nathanw */ 1389 1.4.2.2 nathanw if (wpipe->pipe_state & PIPE_EOF) { 1390 1.4.2.2 nathanw error = EPIPE; 1391 1.4.2.2 nathanw break; 1392 1.4.2.8 nathanw } 1393 1.4.2.2 nathanw } 1394 1.4.2.2 nathanw } 1395 1.4.2.2 nathanw 1396 1.4.2.2 nathanw --wpipe->pipe_busy; 1397 1.4.2.2 nathanw if ((wpipe->pipe_busy == 0) && (wpipe->pipe_state & PIPE_WANTCLOSE)) { 1398 1.4.2.2 nathanw wpipe->pipe_state &= ~(PIPE_WANTCLOSE | PIPE_WANTR); 1399 1.4.2.2 nathanw wakeup(wpipe); 1400 1.4.2.2 nathanw } else if (wpipe->pipe_buffer.cnt > 0) { 1401 1.4.2.2 nathanw /* 1402 1.4.2.2 nathanw * If we have put any characters in the buffer, we wake up 1403 1.4.2.2 nathanw * the reader. 1404 1.4.2.2 nathanw */ 1405 1.4.2.2 nathanw if (wpipe->pipe_state & PIPE_WANTR) { 1406 1.4.2.2 nathanw wpipe->pipe_state &= ~PIPE_WANTR; 1407 1.4.2.2 nathanw wakeup(wpipe); 1408 1.4.2.2 nathanw } 1409 1.4.2.2 nathanw } 1410 1.4.2.2 nathanw 1411 1.4.2.2 nathanw /* 1412 1.4.2.2 nathanw * Don't return EPIPE if I/O was successful 1413 1.4.2.2 nathanw */ 1414 1.4.2.2 nathanw if ((error == EPIPE) && (wpipe->pipe_buffer.cnt == 0) 1415 1.4.2.2 nathanw && (uio->uio_resid == 0)) 1416 1.4.2.2 nathanw error = 0; 1417 1.4.2.2 nathanw 1418 1.4.2.2 nathanw if (error == 0) 1419 1.4.2.2 nathanw vfs_timestamp(&wpipe->pipe_mtime); 1420 1.4.2.2 nathanw 1421 1.4.2.2 nathanw /* 1422 1.4.2.2 nathanw * We have something to offer, wake up select/poll. 1423 1.4.2.2 nathanw * wpipe->pipe_map.cnt is always 0 in this point (direct write 1424 1.4.2.5 nathanw * is only done synchronously), so check only wpipe->pipe_buffer.cnt 1425 1.4.2.2 nathanw */ 1426 1.4.2.2 nathanw if (wpipe->pipe_buffer.cnt) 1427 1.4.2.3 nathanw pipeselwakeup(wpipe, wpipe); 1428 1.4.2.2 nathanw 1429 1.4.2.2 nathanw /* 1430 1.4.2.2 nathanw * Arrange for next read(2) to do a signal. 1431 1.4.2.2 nathanw */ 1432 1.4.2.2 nathanw wpipe->pipe_state |= PIPE_SIGNALR; 1433 1.4.2.2 nathanw 1434 1.4.2.2 nathanw return (error); 1435 1.4.2.2 nathanw } 1436 1.4.2.2 nathanw 1437 1.4.2.2 nathanw /* 1438 1.4.2.2 nathanw * we implement a very minimal set of ioctls for compatibility with sockets. 1439 1.4.2.2 nathanw */ 1440 1.4.2.2 nathanw int 1441 1.4.2.2 nathanw pipe_ioctl(fp, cmd, data, p) 1442 1.4.2.2 nathanw struct file *fp; 1443 1.4.2.2 nathanw u_long cmd; 1444 1.4.2.2 nathanw caddr_t data; 1445 1.4.2.2 nathanw struct proc *p; 1446 1.4.2.2 nathanw { 1447 1.4.2.2 nathanw struct pipe *mpipe = (struct pipe *)fp->f_data; 1448 1.4.2.2 nathanw 1449 1.4.2.2 nathanw switch (cmd) { 1450 1.4.2.2 nathanw 1451 1.4.2.2 nathanw case FIONBIO: 1452 1.4.2.2 nathanw return (0); 1453 1.4.2.2 nathanw 1454 1.4.2.2 nathanw case FIOASYNC: 1455 1.4.2.2 nathanw if (*(int *)data) { 1456 1.4.2.2 nathanw mpipe->pipe_state |= PIPE_ASYNC; 1457 1.4.2.2 nathanw } else { 1458 1.4.2.2 nathanw mpipe->pipe_state &= ~PIPE_ASYNC; 1459 1.4.2.2 nathanw } 1460 1.4.2.2 nathanw return (0); 1461 1.4.2.2 nathanw 1462 1.4.2.2 nathanw case FIONREAD: 1463 1.4.2.2 nathanw #ifndef PIPE_NODIRECT 1464 1.4.2.2 nathanw if (mpipe->pipe_state & PIPE_DIRECTW) 1465 1.4.2.2 nathanw *(int *)data = mpipe->pipe_map.cnt; 1466 1.4.2.2 nathanw else 1467 1.4.2.2 nathanw #endif 1468 1.4.2.2 nathanw *(int *)data = mpipe->pipe_buffer.cnt; 1469 1.4.2.2 nathanw return (0); 1470 1.4.2.2 nathanw 1471 1.4.2.2 nathanw #ifdef __FreeBSD__ 1472 1.4.2.2 nathanw case FIOSETOWN: 1473 1.4.2.2 nathanw return (fsetown(*(int *)data, &mpipe->pipe_sigio)); 1474 1.4.2.2 nathanw 1475 1.4.2.2 nathanw case FIOGETOWN: 1476 1.4.2.2 nathanw *(int *)data = fgetown(mpipe->pipe_sigio); 1477 1.4.2.2 nathanw return (0); 1478 1.4.2.2 nathanw 1479 1.4.2.2 nathanw /* This is deprecated, FIOSETOWN should be used instead. */ 1480 1.4.2.2 nathanw case TIOCSPGRP: 1481 1.4.2.2 nathanw return (fsetown(-(*(int *)data), &mpipe->pipe_sigio)); 1482 1.4.2.2 nathanw 1483 1.4.2.2 nathanw /* This is deprecated, FIOGETOWN should be used instead. */ 1484 1.4.2.2 nathanw case TIOCGPGRP: 1485 1.4.2.2 nathanw *(int *)data = -fgetown(mpipe->pipe_sigio); 1486 1.4.2.2 nathanw return (0); 1487 1.4.2.2 nathanw #endif /* FreeBSD */ 1488 1.4.2.2 nathanw #ifdef __NetBSD__ 1489 1.4.2.2 nathanw case TIOCSPGRP: 1490 1.4.2.2 nathanw mpipe->pipe_pgid = *(int *)data; 1491 1.4.2.2 nathanw return (0); 1492 1.4.2.2 nathanw 1493 1.4.2.2 nathanw case TIOCGPGRP: 1494 1.4.2.2 nathanw *(int *)data = mpipe->pipe_pgid; 1495 1.4.2.2 nathanw return (0); 1496 1.4.2.2 nathanw #endif /* NetBSD */ 1497 1.4.2.2 nathanw 1498 1.4.2.2 nathanw } 1499 1.4.2.2 nathanw return (ENOTTY); 1500 1.4.2.2 nathanw } 1501 1.4.2.2 nathanw 1502 1.4.2.2 nathanw int 1503 1.4.2.2 nathanw pipe_poll(fp, events, p) 1504 1.4.2.2 nathanw struct file *fp; 1505 1.4.2.2 nathanw int events; 1506 1.4.2.2 nathanw struct proc *p; 1507 1.4.2.2 nathanw { 1508 1.4.2.2 nathanw struct pipe *rpipe = (struct pipe *)fp->f_data; 1509 1.4.2.2 nathanw struct pipe *wpipe; 1510 1.4.2.2 nathanw int revents = 0; 1511 1.4.2.2 nathanw 1512 1.4.2.2 nathanw wpipe = rpipe->pipe_peer; 1513 1.4.2.2 nathanw if (events & (POLLIN | POLLRDNORM)) 1514 1.4.2.2 nathanw if ((rpipe->pipe_buffer.cnt > 0) || 1515 1.4.2.2 nathanw #ifndef PIPE_NODIRECT 1516 1.4.2.2 nathanw (rpipe->pipe_state & PIPE_DIRECTW) || 1517 1.4.2.2 nathanw #endif 1518 1.4.2.2 nathanw (rpipe->pipe_state & PIPE_EOF)) 1519 1.4.2.2 nathanw revents |= events & (POLLIN | POLLRDNORM); 1520 1.4.2.2 nathanw 1521 1.4.2.2 nathanw if (events & (POLLOUT | POLLWRNORM)) 1522 1.4.2.2 nathanw if (wpipe == NULL || (wpipe->pipe_state & PIPE_EOF) 1523 1.4.2.2 nathanw || ( 1524 1.4.2.2 nathanw #ifndef PIPE_NODIRECT 1525 1.4.2.2 nathanw ((wpipe->pipe_state & PIPE_DIRECTW) == 0) && 1526 1.4.2.2 nathanw #endif 1527 1.4.2.2 nathanw (wpipe->pipe_buffer.size - wpipe->pipe_buffer.cnt) >= PIPE_BUF)) 1528 1.4.2.2 nathanw revents |= events & (POLLOUT | POLLWRNORM); 1529 1.4.2.2 nathanw 1530 1.4.2.2 nathanw if ((rpipe->pipe_state & PIPE_EOF) || 1531 1.4.2.2 nathanw (wpipe == NULL) || 1532 1.4.2.2 nathanw (wpipe->pipe_state & PIPE_EOF)) 1533 1.4.2.2 nathanw revents |= POLLHUP; 1534 1.4.2.2 nathanw 1535 1.4.2.2 nathanw if (revents == 0) { 1536 1.4.2.2 nathanw if (events & (POLLIN | POLLRDNORM)) { 1537 1.4.2.2 nathanw selrecord(p, &rpipe->pipe_sel); 1538 1.4.2.2 nathanw rpipe->pipe_state |= PIPE_SEL; 1539 1.4.2.2 nathanw } 1540 1.4.2.2 nathanw 1541 1.4.2.2 nathanw if (events & (POLLOUT | POLLWRNORM)) { 1542 1.4.2.2 nathanw selrecord(p, &wpipe->pipe_sel); 1543 1.4.2.2 nathanw wpipe->pipe_state |= PIPE_SEL; 1544 1.4.2.2 nathanw } 1545 1.4.2.2 nathanw } 1546 1.4.2.2 nathanw 1547 1.4.2.2 nathanw return (revents); 1548 1.4.2.2 nathanw } 1549 1.4.2.2 nathanw 1550 1.4.2.2 nathanw static int 1551 1.4.2.2 nathanw pipe_stat(fp, ub, p) 1552 1.4.2.2 nathanw struct file *fp; 1553 1.4.2.2 nathanw struct stat *ub; 1554 1.4.2.2 nathanw struct proc *p; 1555 1.4.2.2 nathanw { 1556 1.4.2.2 nathanw struct pipe *pipe = (struct pipe *)fp->f_data; 1557 1.4.2.2 nathanw 1558 1.4.2.3 nathanw memset((caddr_t)ub, 0, sizeof(*ub)); 1559 1.4.2.2 nathanw ub->st_mode = S_IFIFO; 1560 1.4.2.2 nathanw ub->st_blksize = pipe->pipe_buffer.size; 1561 1.4.2.2 nathanw ub->st_size = pipe->pipe_buffer.cnt; 1562 1.4.2.2 nathanw ub->st_blocks = (ub->st_size) ? 1 : 0; 1563 1.4.2.2 nathanw #ifdef __FreeBSD__ 1564 1.4.2.2 nathanw ub->st_atimespec = pipe->pipe_atime; 1565 1.4.2.2 nathanw ub->st_mtimespec = pipe->pipe_mtime; 1566 1.4.2.2 nathanw ub->st_ctimespec = pipe->pipe_ctime; 1567 1.4.2.2 nathanw #endif /* FreeBSD */ 1568 1.4.2.2 nathanw #ifdef __NetBSD__ 1569 1.4.2.2 nathanw TIMEVAL_TO_TIMESPEC(&pipe->pipe_atime, &ub->st_atimespec) 1570 1.4.2.2 nathanw TIMEVAL_TO_TIMESPEC(&pipe->pipe_mtime, &ub->st_mtimespec); 1571 1.4.2.2 nathanw TIMEVAL_TO_TIMESPEC(&pipe->pipe_ctime, &ub->st_ctimespec); 1572 1.4.2.2 nathanw #endif /* NetBSD */ 1573 1.4.2.2 nathanw ub->st_uid = fp->f_cred->cr_uid; 1574 1.4.2.2 nathanw ub->st_gid = fp->f_cred->cr_gid; 1575 1.4.2.2 nathanw /* 1576 1.4.2.2 nathanw * Left as 0: st_dev, st_ino, st_nlink, st_rdev, st_flags, st_gen. 1577 1.4.2.2 nathanw * XXX (st_dev, st_ino) should be unique. 1578 1.4.2.2 nathanw */ 1579 1.4.2.2 nathanw return (0); 1580 1.4.2.2 nathanw } 1581 1.4.2.2 nathanw 1582 1.4.2.2 nathanw /* ARGSUSED */ 1583 1.4.2.2 nathanw static int 1584 1.4.2.2 nathanw pipe_close(fp, p) 1585 1.4.2.2 nathanw struct file *fp; 1586 1.4.2.2 nathanw struct proc *p; 1587 1.4.2.2 nathanw { 1588 1.4.2.2 nathanw struct pipe *cpipe = (struct pipe *)fp->f_data; 1589 1.4.2.2 nathanw 1590 1.4.2.2 nathanw #ifdef __FreeBSD__ 1591 1.4.2.2 nathanw fp->f_ops = &badfileops; 1592 1.4.2.2 nathanw funsetown(cpipe->pipe_sigio); 1593 1.4.2.2 nathanw #endif 1594 1.4.2.2 nathanw fp->f_data = NULL; 1595 1.4.2.2 nathanw pipeclose(cpipe); 1596 1.4.2.2 nathanw return (0); 1597 1.4.2.2 nathanw } 1598 1.4.2.2 nathanw 1599 1.4.2.2 nathanw static void 1600 1.4.2.2 nathanw pipe_free_kmem(cpipe) 1601 1.4.2.2 nathanw struct pipe *cpipe; 1602 1.4.2.2 nathanw { 1603 1.4.2.2 nathanw 1604 1.4.2.2 nathanw #ifdef __FreeBSD__ 1605 1.4.2.2 nathanw mtx_assert(&vm_mtx, MA_OWNED); 1606 1.4.2.2 nathanw #endif 1607 1.4.2.2 nathanw if (cpipe->pipe_buffer.buffer != NULL) { 1608 1.4.2.2 nathanw if (cpipe->pipe_buffer.size > PIPE_SIZE) 1609 1.4.2.2 nathanw --nbigpipe; 1610 1.4.2.2 nathanw amountpipekva -= cpipe->pipe_buffer.size; 1611 1.4.2.2 nathanw #ifdef __FreeBSD__ 1612 1.4.2.2 nathanw kmem_free(kernel_map, 1613 1.4.2.2 nathanw (vm_offset_t)cpipe->pipe_buffer.buffer, 1614 1.4.2.2 nathanw cpipe->pipe_buffer.size); 1615 1.4.2.2 nathanw #elif defined(__NetBSD__) 1616 1.4.2.2 nathanw uvm_km_free(kernel_map, 1617 1.4.2.2 nathanw (vaddr_t)cpipe->pipe_buffer.buffer, 1618 1.4.2.2 nathanw cpipe->pipe_buffer.size); 1619 1.4.2.2 nathanw #endif /* NetBSD */ 1620 1.4.2.2 nathanw cpipe->pipe_buffer.buffer = NULL; 1621 1.4.2.2 nathanw } 1622 1.4.2.2 nathanw #ifndef PIPE_NODIRECT 1623 1.4.2.2 nathanw if (cpipe->pipe_map.kva != NULL) { 1624 1.4.2.2 nathanw #ifdef __FreeBSD__ 1625 1.4.2.2 nathanw amountpipekva -= cpipe->pipe_buffer.size + PAGE_SIZE; 1626 1.4.2.2 nathanw kmem_free(kernel_map, 1627 1.4.2.2 nathanw cpipe->pipe_map.kva, 1628 1.4.2.2 nathanw cpipe->pipe_buffer.size + PAGE_SIZE); 1629 1.4.2.2 nathanw #elif defined(__NetBSD__) 1630 1.4.2.2 nathanw pipe_loan_free(cpipe); 1631 1.4.2.2 nathanw #endif /* NetBSD */ 1632 1.4.2.2 nathanw cpipe->pipe_map.cnt = 0; 1633 1.4.2.2 nathanw cpipe->pipe_map.kva = NULL; 1634 1.4.2.2 nathanw cpipe->pipe_map.pos = 0; 1635 1.4.2.2 nathanw cpipe->pipe_map.npages = 0; 1636 1.4.2.2 nathanw } 1637 1.4.2.2 nathanw #endif /* !PIPE_NODIRECT */ 1638 1.4.2.2 nathanw } 1639 1.4.2.2 nathanw 1640 1.4.2.2 nathanw /* 1641 1.4.2.2 nathanw * shutdown the pipe 1642 1.4.2.2 nathanw */ 1643 1.4.2.2 nathanw static void 1644 1.4.2.2 nathanw pipeclose(cpipe) 1645 1.4.2.2 nathanw struct pipe *cpipe; 1646 1.4.2.2 nathanw { 1647 1.4.2.2 nathanw struct pipe *ppipe; 1648 1.4.2.2 nathanw 1649 1.4.2.2 nathanw if (!cpipe) 1650 1.4.2.2 nathanw return; 1651 1.4.2.2 nathanw 1652 1.4.2.3 nathanw pipeselwakeup(cpipe, cpipe); 1653 1.4.2.2 nathanw 1654 1.4.2.2 nathanw /* 1655 1.4.2.2 nathanw * If the other side is blocked, wake it up saying that 1656 1.4.2.2 nathanw * we want to close it down. 1657 1.4.2.2 nathanw */ 1658 1.4.2.2 nathanw while (cpipe->pipe_busy) { 1659 1.4.2.2 nathanw wakeup(cpipe); 1660 1.4.2.2 nathanw cpipe->pipe_state |= PIPE_WANTCLOSE | PIPE_EOF; 1661 1.4.2.2 nathanw tsleep(cpipe, PRIBIO, "pipecl", 0); 1662 1.4.2.2 nathanw } 1663 1.4.2.2 nathanw 1664 1.4.2.2 nathanw /* 1665 1.4.2.2 nathanw * Disconnect from peer 1666 1.4.2.2 nathanw */ 1667 1.4.2.2 nathanw if ((ppipe = cpipe->pipe_peer) != NULL) { 1668 1.4.2.3 nathanw pipeselwakeup(ppipe, ppipe); 1669 1.4.2.2 nathanw 1670 1.4.2.2 nathanw ppipe->pipe_state |= PIPE_EOF; 1671 1.4.2.2 nathanw wakeup(ppipe); 1672 1.4.2.2 nathanw ppipe->pipe_peer = NULL; 1673 1.4.2.2 nathanw } 1674 1.4.2.2 nathanw 1675 1.4.2.2 nathanw /* 1676 1.4.2.2 nathanw * free resources 1677 1.4.2.2 nathanw */ 1678 1.4.2.9 nathanw #ifdef __FreeBSD__ 1679 1.4.2.2 nathanw mtx_lock(&vm_mtx); 1680 1.4.2.2 nathanw pipe_free_kmem(cpipe); 1681 1.4.2.2 nathanw /* XXX: erm, doesn't zalloc already have its own locks and 1682 1.4.2.2 nathanw * not need the giant vm lock? 1683 1.4.2.2 nathanw */ 1684 1.4.2.2 nathanw zfree(pipe_zone, cpipe); 1685 1.4.2.2 nathanw mtx_unlock(&vm_mtx); 1686 1.4.2.2 nathanw #endif /* FreeBSD */ 1687 1.4.2.3 nathanw 1688 1.4.2.2 nathanw #ifdef __NetBSD__ 1689 1.4.2.3 nathanw pipe_free_kmem(cpipe); 1690 1.4.2.3 nathanw (void) lockmgr(&cpipe->pipe_lock, LK_DRAIN, NULL); 1691 1.4.2.2 nathanw pool_put(&pipe_pool, cpipe); 1692 1.4.2.2 nathanw #endif 1693 1.4.2.2 nathanw } 1694 1.4.2.2 nathanw 1695 1.4.2.2 nathanw #ifdef __FreeBSD__ 1696 1.4.2.2 nathanw /*ARGSUSED*/ 1697 1.4.2.2 nathanw static int 1698 1.4.2.2 nathanw pipe_kqfilter(struct file *fp, struct knote *kn) 1699 1.4.2.2 nathanw { 1700 1.4.2.2 nathanw struct pipe *cpipe = (struct pipe *)kn->kn_fp->f_data; 1701 1.4.2.2 nathanw 1702 1.4.2.2 nathanw switch (kn->kn_filter) { 1703 1.4.2.2 nathanw case EVFILT_READ: 1704 1.4.2.2 nathanw kn->kn_fop = &pipe_rfiltops; 1705 1.4.2.2 nathanw break; 1706 1.4.2.2 nathanw case EVFILT_WRITE: 1707 1.4.2.2 nathanw kn->kn_fop = &pipe_wfiltops; 1708 1.4.2.2 nathanw cpipe = cpipe->pipe_peer; 1709 1.4.2.2 nathanw break; 1710 1.4.2.2 nathanw default: 1711 1.4.2.2 nathanw return (1); 1712 1.4.2.2 nathanw } 1713 1.4.2.2 nathanw kn->kn_hook = (caddr_t)cpipe; 1714 1.4.2.2 nathanw SLIST_INSERT_HEAD(&cpipe->pipe_sel.si_note, kn, kn_selnext); 1715 1.4.2.2 nathanw return (0); 1716 1.4.2.2 nathanw } 1717 1.4.2.2 nathanw 1718 1.4.2.2 nathanw static void 1719 1.4.2.2 nathanw filt_pipedetach(struct knote *kn) 1720 1.4.2.2 nathanw { 1721 1.4.2.2 nathanw struct pipe *cpipe = (struct pipe *)kn->kn_fp->f_data; 1722 1.4.2.2 nathanw 1723 1.4.2.2 nathanw SLIST_REMOVE(&cpipe->pipe_sel.si_note, kn, knote, kn_selnext); 1724 1.4.2.2 nathanw } 1725 1.4.2.2 nathanw 1726 1.4.2.2 nathanw /*ARGSUSED*/ 1727 1.4.2.2 nathanw static int 1728 1.4.2.2 nathanw filt_piperead(struct knote *kn, long hint) 1729 1.4.2.2 nathanw { 1730 1.4.2.2 nathanw struct pipe *rpipe = (struct pipe *)kn->kn_fp->f_data; 1731 1.4.2.2 nathanw struct pipe *wpipe = rpipe->pipe_peer; 1732 1.4.2.2 nathanw 1733 1.4.2.2 nathanw kn->kn_data = rpipe->pipe_buffer.cnt; 1734 1.4.2.2 nathanw if ((kn->kn_data == 0) && (rpipe->pipe_state & PIPE_DIRECTW)) 1735 1.4.2.2 nathanw kn->kn_data = rpipe->pipe_map.cnt; 1736 1.4.2.2 nathanw 1737 1.4.2.2 nathanw if ((rpipe->pipe_state & PIPE_EOF) || 1738 1.4.2.2 nathanw (wpipe == NULL) || (wpipe->pipe_state & PIPE_EOF)) { 1739 1.4.2.2 nathanw kn->kn_flags |= EV_EOF; 1740 1.4.2.2 nathanw return (1); 1741 1.4.2.2 nathanw } 1742 1.4.2.2 nathanw return (kn->kn_data > 0); 1743 1.4.2.2 nathanw } 1744 1.4.2.2 nathanw 1745 1.4.2.2 nathanw /*ARGSUSED*/ 1746 1.4.2.2 nathanw static int 1747 1.4.2.2 nathanw filt_pipewrite(struct knote *kn, long hint) 1748 1.4.2.2 nathanw { 1749 1.4.2.2 nathanw struct pipe *rpipe = (struct pipe *)kn->kn_fp->f_data; 1750 1.4.2.2 nathanw struct pipe *wpipe = rpipe->pipe_peer; 1751 1.4.2.2 nathanw 1752 1.4.2.2 nathanw if ((wpipe == NULL) || (wpipe->pipe_state & PIPE_EOF)) { 1753 1.4.2.2 nathanw kn->kn_data = 0; 1754 1.4.2.2 nathanw kn->kn_flags |= EV_EOF; 1755 1.4.2.2 nathanw return (1); 1756 1.4.2.2 nathanw } 1757 1.4.2.2 nathanw kn->kn_data = wpipe->pipe_buffer.size - wpipe->pipe_buffer.cnt; 1758 1.4.2.2 nathanw if (wpipe->pipe_state & PIPE_DIRECTW) 1759 1.4.2.2 nathanw kn->kn_data = 0; 1760 1.4.2.2 nathanw 1761 1.4.2.2 nathanw return (kn->kn_data >= PIPE_BUF); 1762 1.4.2.2 nathanw } 1763 1.4.2.2 nathanw #endif /* FreeBSD */ 1764 1.4.2.2 nathanw 1765 1.4.2.2 nathanw #ifdef __NetBSD__ 1766 1.4.2.2 nathanw static int 1767 1.4.2.2 nathanw pipe_fcntl(fp, cmd, data, p) 1768 1.4.2.2 nathanw struct file *fp; 1769 1.4.2.2 nathanw u_int cmd; 1770 1.4.2.2 nathanw caddr_t data; 1771 1.4.2.2 nathanw struct proc *p; 1772 1.4.2.2 nathanw { 1773 1.4.2.2 nathanw if (cmd == F_SETFL) 1774 1.4.2.2 nathanw return (0); 1775 1.4.2.2 nathanw else 1776 1.4.2.2 nathanw return (EOPNOTSUPP); 1777 1.4.2.2 nathanw } 1778 1.4.2.2 nathanw 1779 1.4.2.2 nathanw /* 1780 1.4.2.2 nathanw * Handle pipe sysctls. 1781 1.4.2.2 nathanw */ 1782 1.4.2.2 nathanw int 1783 1.4.2.2 nathanw sysctl_dopipe(name, namelen, oldp, oldlenp, newp, newlen) 1784 1.4.2.2 nathanw int *name; 1785 1.4.2.2 nathanw u_int namelen; 1786 1.4.2.2 nathanw void *oldp; 1787 1.4.2.2 nathanw size_t *oldlenp; 1788 1.4.2.2 nathanw void *newp; 1789 1.4.2.2 nathanw size_t newlen; 1790 1.4.2.2 nathanw { 1791 1.4.2.2 nathanw /* All sysctl names at this level are terminal. */ 1792 1.4.2.2 nathanw if (namelen != 1) 1793 1.4.2.2 nathanw return (ENOTDIR); /* overloaded */ 1794 1.4.2.2 nathanw 1795 1.4.2.2 nathanw switch (name[0]) { 1796 1.4.2.2 nathanw case KERN_PIPE_MAXKVASZ: 1797 1.4.2.2 nathanw return (sysctl_int(oldp, oldlenp, newp, newlen, &maxpipekva)); 1798 1.4.2.2 nathanw case KERN_PIPE_LIMITKVA: 1799 1.4.2.2 nathanw return (sysctl_int(oldp, oldlenp, newp, newlen, &limitpipekva)); 1800 1.4.2.2 nathanw case KERN_PIPE_MAXBIGPIPES: 1801 1.4.2.2 nathanw return (sysctl_int(oldp, oldlenp, newp, newlen, &maxbigpipes)); 1802 1.4.2.2 nathanw case KERN_PIPE_NBIGPIPES: 1803 1.4.2.2 nathanw return (sysctl_rdint(oldp, oldlenp, newp, nbigpipe)); 1804 1.4.2.2 nathanw case KERN_PIPE_KVASIZE: 1805 1.4.2.2 nathanw return (sysctl_rdint(oldp, oldlenp, newp, amountpipekva)); 1806 1.4.2.2 nathanw default: 1807 1.4.2.2 nathanw return (EOPNOTSUPP); 1808 1.4.2.2 nathanw } 1809 1.4.2.2 nathanw /* NOTREACHED */ 1810 1.4.2.2 nathanw } 1811 1.4.2.2 nathanw 1812 1.4.2.2 nathanw /* 1813 1.4.2.2 nathanw * Initialize pipe structs. 1814 1.4.2.2 nathanw */ 1815 1.4.2.2 nathanw void 1816 1.4.2.2 nathanw pipe_init(void) 1817 1.4.2.2 nathanw { 1818 1.4.2.2 nathanw pool_init(&pipe_pool, sizeof(struct pipe), 0, 0, 0, "pipepl", 1819 1.4.2.2 nathanw 0, NULL, NULL, M_PIPE); 1820 1.4.2.2 nathanw } 1821 1.4.2.2 nathanw 1822 1.4.2.2 nathanw #endif /* __NetBSD __ */ 1823
Indexes created Tue Sep 30 11:09:46 GMT 2025