sys/kern/tty_outq.c

   1 /*-
   2  * Copyright (c) 2008 Ed Schouten <ed@FreeBSD.org>
   3  * All rights reserved.
   4  *
   5  * Portions of this software were developed under sponsorship from Snow
   6  * B.V., the Netherlands.
   7  *
   8  * Redistribution and use in source and binary forms, with or without
   9  * modification, are permitted provided that the following conditions
  10  * are met:
  11  * 1. Redistributions of source code must retain the above copyright
  12  *    notice, this list of conditions and the following disclaimer.
  13  * 2. Redistributions in binary form must reproduce the above copyright
  14  *    notice, this list of conditions and the following disclaimer in the
  15  *    documentation and/or other materials provided with the distribution.
  16  *
  17  * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
  18  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
  19  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
  20  * ARE DISCLAIMED.  IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
  21  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
  22  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
  23  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
  24  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
  25  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
  26  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
  27  * SUCH DAMAGE.
  28  */
  29
  30 #include <sys/cdefs.h>
  31 __FBSDID("$FreeBSD$");
  32
  33 #include <sys/param.h>
  34 #include <sys/kernel.h>
  35 #include <sys/lock.h>
  36 #include <sys/queue.h>
  37 #include <sys/sysctl.h>
  38 #include <sys/systm.h>
  39 #include <sys/tty.h>
  40 #include <sys/uio.h>
  41
  42 #include <vm/uma.h>
  43
  44 /*
  45  * TTY output queue buffering.
  46  *
  47  * The previous design of the TTY layer offered the so-called clists.
  48  * These clists were used for both the input queues and the output
  49  * queue. We don't use certain features on the output side, like quoting
  50  * bits for parity marking and such. This mechanism is similar to the
  51  * old clists, but only contains the features we need to buffer the
  52  * output.
  53  */
  54
  55 /* Statistics. */
  56 static long ttyoutq_nfast = 0;
  57 SYSCTL_LONG(_kern, OID_AUTO, tty_outq_nfast, CTLFLAG_RD,
  58         &ttyoutq_nfast, 0, "Unbuffered reads to userspace on output");
  59 static long ttyoutq_nslow = 0;
  60 SYSCTL_LONG(_kern, OID_AUTO, tty_outq_nslow, CTLFLAG_RD,
  61         &ttyoutq_nslow, 0, "Buffered reads to userspace on output");
  62
  63 struct ttyoutq_block {
  64         STAILQ_ENTRY(ttyoutq_block) tob_list;
  65         char    tob_data[TTYOUTQ_DATASIZE];
  66 };
  67
  68 static uma_zone_t ttyoutq_zone;
  69
  70 void
  71 ttyoutq_flush(struct ttyoutq *to)
  72 {
  73
  74         to->to_begin = 0;
  75         to->to_end = 0;
  76 }
  77
  78 void
  79 ttyoutq_setsize(struct ttyoutq *to, struct tty *tp, size_t size)
  80 {
  81         struct ttyoutq_block *tob;
  82
  83         to->to_quota = howmany(size, TTYOUTQ_DATASIZE);
  84
  85         while (to->to_quota > to->to_nblocks) {
  86                 /*
  87                  * List is getting bigger.
  88                  * Add new blocks to the tail of the list.
  89                  *
  90                  * We must unlock the TTY temporarily, because we need
  91                  * to allocate memory. This won't be a problem, because
  92                  * in the worst case, another thread ends up here, which
  93                  * may cause us to allocate too many blocks, but this
  94                  * will be caught by the loop below.
  95                  */
  96                 tty_unlock(tp);
  97                 tob = uma_zalloc(ttyoutq_zone, M_WAITOK);
  98                 tty_lock(tp);
  99
 100                 STAILQ_INSERT_TAIL(&to->to_list, tob, tob_list);
 101                 to->to_nblocks++;
 102         }
 103 }
 104
 105 void
 106 ttyoutq_free(struct ttyoutq *to)
 107 {
 108         struct ttyoutq_block *tob;
 109
 110         ttyoutq_flush(to);
 111         to->to_quota = 0;
 112
 113         while ((tob = STAILQ_FIRST(&to->to_list)) != NULL) {
 114                 STAILQ_REMOVE_HEAD(&to->to_list, tob_list);
 115                 uma_zfree(ttyoutq_zone, tob);
 116                 to->to_nblocks--;
 117         }
 118
 119         MPASS(to->to_nblocks == 0);
 120 }
 121
 122 size_t
 123 ttyoutq_read(struct ttyoutq *to, void *buf, size_t len)
 124 {
 125         char *cbuf = buf;
 126
 127         while (len > 0) {
 128                 struct ttyoutq_block *tob;
 129                 size_t cbegin, cend, clen;
 130
 131                 /* See if there still is data. */
 132                 if (to->to_begin == to->to_end)
 133                         break;
 134                 tob = STAILQ_FIRST(&to->to_list);
 135                 if (tob == NULL)
 136                         break;
 137
 138                 /*
 139                  * The end address should be the lowest of these three:
 140                  * - The write pointer
 141                  * - The blocksize - we can't read beyond the block
 142                  * - The end address if we could perform the full read
 143                  */
 144                 cbegin = to->to_begin;
 145                 cend = MIN(MIN(to->to_end, to->to_begin + len),
 146                     TTYOUTQ_DATASIZE);
 147                 clen = cend - cbegin;
 148
 149                 if (cend == TTYOUTQ_DATASIZE || cend == to->to_end) {
 150                         /* Read the block until the end. */
 151                         STAILQ_REMOVE_HEAD(&to->to_list, tob_list);
 152                         if (to->to_quota < to->to_nblocks) {
 153                                 uma_zfree(ttyoutq_zone, tob);
 154                                 to->to_nblocks--;
 155                         } else {
 156                                 STAILQ_INSERT_TAIL(&to->to_list, tob, tob_list);
 157                         }
 158                         to->to_begin = 0;
 159                         if (to->to_end <= TTYOUTQ_DATASIZE) {
 160                                 to->to_end = 0;
 161                         } else {
 162                                 to->to_end -= TTYOUTQ_DATASIZE;
 163                         }
 164                 } else {
 165                         /* Read the block partially. */
 166                         to->to_begin += clen;
 167                 }
 168
 169                 /* Copy the data out of the buffers. */
 170                 memcpy(cbuf, tob->tob_data + cbegin, clen);
 171                 cbuf += clen;
 172                 len -= clen;
 173         }
 174
 175         return (cbuf - (char *)buf);
 176 }
 177
 178 /*
 179  * An optimized version of ttyoutq_read() which can be used in pseudo
 180  * TTY drivers to directly copy data from the outq to userspace, instead
 181  * of buffering it.
 182  *
 183  * We can only copy data directly if we need to read the entire block
 184  * back to the user, because we temporarily remove the block from the
 185  * queue. Otherwise we need to copy it to a temporary buffer first, to
 186  * make sure data remains in the correct order.
 187  */
 188 int
 189 ttyoutq_read_uio(struct ttyoutq *to, struct tty *tp, struct uio *uio)
 190 {
 191
 192         while (uio->uio_resid > 0) {
 193                 int error;
 194                 struct ttyoutq_block *tob;
 195                 size_t cbegin, cend, clen;
 196
 197                 /* See if there still is data. */
 198                 if (to->to_begin == to->to_end)
 199                         return (0);
 200                 tob = STAILQ_FIRST(&to->to_list);
 201                 if (tob == NULL)
 202                         return (0);
 203
 204                 /*
 205                  * The end address should be the lowest of these three:
 206                  * - The write pointer
 207                  * - The blocksize - we can't read beyond the block
 208                  * - The end address if we could perform the full read
 209                  */
 210                 cbegin = to->to_begin;
 211                 cend = MIN(MIN(to->to_end, to->to_begin + uio->uio_resid),
 212                     TTYOUTQ_DATASIZE);
 213                 clen = cend - cbegin;
 214
 215                 /*
 216                  * We can prevent buffering in some cases:
 217                  * - We need to read the block until the end.
 218                  * - We don't need to read the block until the end, but
 219                  *   there is no data beyond it, which allows us to move
 220                  *   the write pointer to a new block.
 221                  */
 222                 if (cend == TTYOUTQ_DATASIZE || cend == to->to_end) {
 223                         atomic_add_long(&ttyoutq_nfast, 1);
 224
 225                         /*
 226                          * Fast path: zero copy. Remove the first block,
 227                          * so we can unlock the TTY temporarily.
 228                          */
 229                         STAILQ_REMOVE_HEAD(&to->to_list, tob_list);
 230                         to->to_nblocks--;
 231                         to->to_begin = 0;
 232                         if (to->to_end <= TTYOUTQ_DATASIZE) {
 233                                 to->to_end = 0;
 234                         } else {
 235                                 to->to_end -= TTYOUTQ_DATASIZE;
 236                         }
 237
 238                         /* Temporary unlock and copy the data to userspace. */
 239                         tty_unlock(tp);
 240                         error = uiomove(tob->tob_data + cbegin, clen, uio);
 241                         tty_lock(tp);
 242
 243                         /* Block can now be readded to the list. */
 244                         if (to->to_quota <= to->to_nblocks) {
 245                                 uma_zfree(ttyoutq_zone, tob);
 246                         } else {
 247                                 STAILQ_INSERT_TAIL(&to->to_list, tob, tob_list);
 248                                 to->to_nblocks++;
 249                         }
 250                 } else {
 251                         char ob[TTYOUTQ_DATASIZE - 1];
 252                         atomic_add_long(&ttyoutq_nslow, 1);
 253
 254                         /*
 255                          * Slow path: store data in a temporary buffer.
 256                          */
 257                         memcpy(ob, tob->tob_data + cbegin, clen);
 258                         to->to_begin += clen;
 259                         MPASS(to->to_begin < TTYOUTQ_DATASIZE);
 260
 261                         /* Temporary unlock and copy the data to userspace. */
 262                         tty_unlock(tp);
 263                         error = uiomove(ob, clen, uio);
 264                         tty_lock(tp);
 265                 }
 266
 267                 if (error != 0)
 268                         return (error);
 269         }
 270
 271         return (0);
 272 }
 273
 274 size_t
 275 ttyoutq_write(struct ttyoutq *to, const void *buf, size_t nbytes)
 276 {
 277         const char *cbuf = buf;
 278         struct ttyoutq_block *tob;
 279         unsigned int boff;
 280         size_t l;
 281
 282         while (nbytes > 0) {
 283                 /* Offset in current block. */
 284                 tob = to->to_lastblock;
 285                 boff = to->to_end % TTYOUTQ_DATASIZE;
 286
 287                 if (to->to_end == 0) {
 288                         /* First time we're being used or drained. */
 289                         MPASS(to->to_begin == 0);
 290                         tob = to->to_lastblock = STAILQ_FIRST(&to->to_list);
 291                         if (tob == NULL) {
 292                                 /* Queue has no blocks. */
 293                                 break;
 294                         }
 295                 } else if (boff == 0) {
 296                         /* We reached the end of this block on last write. */
 297                         tob = STAILQ_NEXT(tob, tob_list);
 298                         if (tob == NULL) {
 299                                 /* We've reached the watermark. */
 300                                 break;
 301                         }
 302                         to->to_lastblock = tob;
 303                 }
 304
 305                 /* Don't copy more than was requested. */
 306                 l = MIN(nbytes, TTYOUTQ_DATASIZE - boff);
 307                 MPASS(l > 0);
 308                 memcpy(tob->tob_data + boff, cbuf, l);
 309
 310                 cbuf += l;
 311                 nbytes -= l;
 312                 to->to_end += l;
 313         }
 314
 315         return (cbuf - (const char *)buf);
 316 }
 317
 318 int
 319 ttyoutq_write_nofrag(struct ttyoutq *to, const void *buf, size_t nbytes)
 320 {
 321         size_t ret;
 322
 323         if (ttyoutq_bytesleft(to) < nbytes)
 324                 return (-1);
 325
 326         /* We should always be able to write it back. */
 327         ret = ttyoutq_write(to, buf, nbytes);
 328         MPASS(ret == nbytes);
 329
 330         return (0);
 331 }
 332
 333 static void
 334 ttyoutq_startup(void *dummy)
 335 {
 336
 337         ttyoutq_zone = uma_zcreate("ttyoutq", sizeof(struct ttyoutq_block),
 338             NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0);
 339 }
 340
 341 SYSINIT(ttyoutq, SI_SUB_DRIVERS, SI_ORDER_FIRST, ttyoutq_startup, NULL);