The Design and Implementation of the FreeBSD Operating System, Second Edition
Now available: The Design and Implementation of the FreeBSD Operating System (Second Edition)


[ source navigation ] [ diff markup ] [ identifier search ] [ freetext search ] [ file search ] [ list types ] [ track identifier ]

FreeBSD/Linux Kernel Cross Reference
sys/kern/subr_sglist.c

Version: -  FREEBSD  -  FREEBSD-13-STABLE  -  FREEBSD-13-0  -  FREEBSD-12-STABLE  -  FREEBSD-12-0  -  FREEBSD-11-STABLE  -  FREEBSD-11-0  -  FREEBSD-10-STABLE  -  FREEBSD-10-0  -  FREEBSD-9-STABLE  -  FREEBSD-9-0  -  FREEBSD-8-STABLE  -  FREEBSD-8-0  -  FREEBSD-7-STABLE  -  FREEBSD-7-0  -  FREEBSD-6-STABLE  -  FREEBSD-6-0  -  FREEBSD-5-STABLE  -  FREEBSD-5-0  -  FREEBSD-4-STABLE  -  FREEBSD-3-STABLE  -  FREEBSD22  -  l41  -  OPENBSD  -  linux-2.6  -  MK84  -  PLAN9  -  xnu-8792 
SearchContext: -  none  -  3  -  10 

    1 /*-
    2  * SPDX-License-Identifier: BSD-3-Clause
    3  *
    4  * Copyright (c) 2008 Yahoo!, Inc.
    5  * All rights reserved.
    6  * Written by: John Baldwin <jhb@FreeBSD.org>
    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  * 3. Neither the name of the author nor the names of any co-contributors
   17  *    may be used to endorse or promote products derived from this software
   18  *    without specific prior written permission.
   19  *
   20  * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
   21  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
   22  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
   23  * ARE DISCLAIMED.  IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
   24  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
   25  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
   26  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
   27  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
   28  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
   29  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
   30  * SUCH DAMAGE.
   31  */
   32 
   33 #include <sys/cdefs.h>
   34 __FBSDID("$FreeBSD$");
   35 
   36 #include <sys/param.h>
   37 #include <sys/kernel.h>
   38 #include <sys/bio.h>
   39 #include <sys/malloc.h>
   40 #include <sys/mbuf.h>
   41 #include <sys/proc.h>
   42 #include <sys/sglist.h>
   43 #include <sys/uio.h>
   44 
   45 #include <vm/vm.h>
   46 #include <vm/vm_page.h>
   47 #include <vm/pmap.h>
   48 #include <vm/vm_map.h>
   49 
   50 #include <sys/ktr.h>
   51 
   52 static MALLOC_DEFINE(M_SGLIST, "sglist", "scatter/gather lists");
   53 
   54 /*
   55  * Convenience macros to save the state of an sglist so it can be restored
   56  * if an append attempt fails.  Since sglist's only grow we only need to
   57  * save the current count of segments and the length of the ending segment.
   58  * Earlier segments will not be changed by an append, and the only change
   59  * that can occur to the ending segment is that it can be extended.
   60  */
   61 struct sgsave {
   62         u_short sg_nseg;
   63         size_t ss_len;
   64 };
   65 
   66 #define SGLIST_SAVE(sg, sgsave) do {                                    \
   67         (sgsave).sg_nseg = (sg)->sg_nseg;                               \
   68         if ((sgsave).sg_nseg > 0)                                       \
   69                 (sgsave).ss_len = (sg)->sg_segs[(sgsave).sg_nseg - 1].ss_len; \
   70         else                                                            \
   71                 (sgsave).ss_len = 0;                                    \
   72 } while (0)
   73 
   74 #define SGLIST_RESTORE(sg, sgsave) do {                                 \
   75         (sg)->sg_nseg = (sgsave).sg_nseg;                               \
   76         if ((sgsave).sg_nseg > 0)                                       \
   77                 (sg)->sg_segs[(sgsave).sg_nseg - 1].ss_len = (sgsave).ss_len; \
   78 } while (0)
   79 
   80 /*
   81  * Append a single (paddr, len) to a sglist.  sg is the list and ss is
   82  * the current segment in the list.  If we run out of segments then
   83  * EFBIG will be returned.
   84  */
   85 static __inline int
   86 _sglist_append_range(struct sglist *sg, struct sglist_seg **ssp,
   87     vm_paddr_t paddr, size_t len)
   88 {
   89         struct sglist_seg *ss;
   90 
   91         ss = *ssp;
   92         if (ss->ss_paddr + ss->ss_len == paddr)
   93                 ss->ss_len += len;
   94         else {
   95                 if (sg->sg_nseg == sg->sg_maxseg)
   96                         return (EFBIG);
   97                 ss++;
   98                 ss->ss_paddr = paddr;
   99                 ss->ss_len = len;
  100                 sg->sg_nseg++;
  101                 *ssp = ss;
  102         }
  103         return (0);
  104 }
  105 
  106 /*
  107  * Worker routine to append a virtual address range (either kernel or
  108  * user) to a scatter/gather list.
  109  */
  110 static __inline int
  111 _sglist_append_buf(struct sglist *sg, void *buf, size_t len, pmap_t pmap,
  112     size_t *donep)
  113 {
  114         struct sglist_seg *ss;
  115         vm_offset_t vaddr, offset;
  116         vm_paddr_t paddr;
  117         size_t seglen;
  118         int error;
  119 
  120         if (donep)
  121                 *donep = 0;
  122         if (len == 0)
  123                 return (0);
  124 
  125         /* Do the first page.  It may have an offset. */
  126         vaddr = (vm_offset_t)buf;
  127         offset = vaddr & PAGE_MASK;
  128         if (pmap != NULL)
  129                 paddr = pmap_extract(pmap, vaddr);
  130         else
  131                 paddr = pmap_kextract(vaddr);
  132         seglen = MIN(len, PAGE_SIZE - offset);
  133         if (sg->sg_nseg == 0) {
  134                 ss = sg->sg_segs;
  135                 ss->ss_paddr = paddr;
  136                 ss->ss_len = seglen;
  137                 sg->sg_nseg = 1;
  138         } else {
  139                 ss = &sg->sg_segs[sg->sg_nseg - 1];
  140                 error = _sglist_append_range(sg, &ss, paddr, seglen);
  141                 if (error)
  142                         return (error);
  143         }
  144         vaddr += seglen;
  145         len -= seglen;
  146         if (donep)
  147                 *donep += seglen;
  148 
  149         while (len > 0) {
  150                 seglen = MIN(len, PAGE_SIZE);
  151                 if (pmap != NULL)
  152                         paddr = pmap_extract(pmap, vaddr);
  153                 else
  154                         paddr = pmap_kextract(vaddr);
  155                 error = _sglist_append_range(sg, &ss, paddr, seglen);
  156                 if (error)
  157                         return (error);
  158                 vaddr += seglen;
  159                 len -= seglen;
  160                 if (donep)
  161                         *donep += seglen;
  162         }
  163 
  164         return (0);
  165 }
  166 
  167 /*
  168  * Determine the number of scatter/gather list elements needed to
  169  * describe a kernel virtual address range.
  170  */
  171 int
  172 sglist_count(void *buf, size_t len)
  173 {
  174         vm_offset_t vaddr, vendaddr;
  175         vm_paddr_t lastaddr, paddr;
  176         int nsegs;
  177 
  178         if (len == 0)
  179                 return (0);
  180 
  181         vaddr = trunc_page((vm_offset_t)buf);
  182         vendaddr = (vm_offset_t)buf + len;
  183         nsegs = 1;
  184         lastaddr = pmap_kextract(vaddr);
  185         vaddr += PAGE_SIZE;
  186         while (vaddr < vendaddr) {
  187                 paddr = pmap_kextract(vaddr);
  188                 if (lastaddr + PAGE_SIZE != paddr)
  189                         nsegs++;
  190                 lastaddr = paddr;
  191                 vaddr += PAGE_SIZE;
  192         }
  193         return (nsegs);
  194 }
  195 
  196 /*
  197  * Determine the number of scatter/gather list elements needed to
  198  * describe a buffer backed by an array of VM pages.
  199  */
  200 int
  201 sglist_count_vmpages(vm_page_t *m, size_t pgoff, size_t len)
  202 {
  203         vm_paddr_t lastaddr, paddr;
  204         int i, nsegs;
  205 
  206         if (len == 0)
  207                 return (0);
  208 
  209         len += pgoff;
  210         nsegs = 1;
  211         lastaddr = VM_PAGE_TO_PHYS(m[0]);
  212         for (i = 1; len > PAGE_SIZE; len -= PAGE_SIZE, i++) {
  213                 paddr = VM_PAGE_TO_PHYS(m[i]);
  214                 if (lastaddr + PAGE_SIZE != paddr)
  215                         nsegs++;
  216                 lastaddr = paddr;
  217         }
  218         return (nsegs);
  219 }
  220 
  221 /*
  222  * Determine the number of scatter/gather list elements needed to
  223  * describe an M_EXTPG mbuf.
  224  */
  225 int
  226 sglist_count_mbuf_epg(struct mbuf *m, size_t off, size_t len)
  227 {
  228         vm_paddr_t nextaddr, paddr;
  229         size_t seglen, segoff;
  230         int i, nsegs, pglen, pgoff;
  231 
  232         if (len == 0)
  233                 return (0);
  234 
  235         nsegs = 0;
  236         if (m->m_epg_hdrlen != 0) {
  237                 if (off >= m->m_epg_hdrlen) {
  238                         off -= m->m_epg_hdrlen;
  239                 } else {
  240                         seglen = m->m_epg_hdrlen - off;
  241                         segoff = off;
  242                         seglen = MIN(seglen, len);
  243                         off = 0;
  244                         len -= seglen;
  245                         nsegs += sglist_count(&m->m_epg_hdr[segoff],
  246                             seglen);
  247                 }
  248         }
  249         nextaddr = 0;
  250         pgoff = m->m_epg_1st_off;
  251         for (i = 0; i < m->m_epg_npgs && len > 0; i++) {
  252                 pglen = m_epg_pagelen(m, i, pgoff);
  253                 if (off >= pglen) {
  254                         off -= pglen;
  255                         pgoff = 0;
  256                         continue;
  257                 }
  258                 seglen = pglen - off;
  259                 segoff = pgoff + off;
  260                 off = 0;
  261                 seglen = MIN(seglen, len);
  262                 len -= seglen;
  263                 paddr = m->m_epg_pa[i] + segoff;
  264                 if (paddr != nextaddr)
  265                         nsegs++;
  266                 nextaddr = paddr + seglen;
  267                 pgoff = 0;
  268         };
  269         if (len != 0) {
  270                 seglen = MIN(len, m->m_epg_trllen - off);
  271                 len -= seglen;
  272                 nsegs += sglist_count(&m->m_epg_trail[off], seglen);
  273         }
  274         KASSERT(len == 0, ("len != 0"));
  275         return (nsegs);
  276 }
  277 
  278 /*
  279  * Allocate a scatter/gather list along with 'nsegs' segments.  The
  280  * 'mflags' parameters are the same as passed to malloc(9).  The caller
  281  * should use sglist_free() to free this list.
  282  */
  283 struct sglist *
  284 sglist_alloc(int nsegs, int mflags)
  285 {
  286         struct sglist *sg;
  287 
  288         sg = malloc(sizeof(struct sglist) + nsegs * sizeof(struct sglist_seg),
  289             M_SGLIST, mflags);
  290         if (sg == NULL)
  291                 return (NULL);
  292         sglist_init(sg, nsegs, (struct sglist_seg *)(sg + 1));
  293         return (sg);
  294 }
  295 
  296 /*
  297  * Free a scatter/gather list allocated via sglist_allc().
  298  */
  299 void
  300 sglist_free(struct sglist *sg)
  301 {
  302 
  303         if (sg == NULL)
  304                 return;
  305 
  306         if (refcount_release(&sg->sg_refs))
  307                 free(sg, M_SGLIST);
  308 }
  309 
  310 /*
  311  * Append the segments to describe a single kernel virtual address
  312  * range to a scatter/gather list.  If there are insufficient
  313  * segments, then this fails with EFBIG.
  314  */
  315 int
  316 sglist_append(struct sglist *sg, void *buf, size_t len)
  317 {
  318         struct sgsave save;
  319         int error;
  320 
  321         if (sg->sg_maxseg == 0)
  322                 return (EINVAL);
  323         SGLIST_SAVE(sg, save);
  324         error = _sglist_append_buf(sg, buf, len, NULL, NULL);
  325         if (error)
  326                 SGLIST_RESTORE(sg, save);
  327         return (error);
  328 }
  329 
  330 /*
  331  * Append the segments to describe a bio's data to a scatter/gather list.
  332  * If there are insufficient segments, then this fails with EFBIG.
  333  *
  334  * NOTE: This function expects bio_bcount to be initialized.
  335  */
  336 int
  337 sglist_append_bio(struct sglist *sg, struct bio *bp)
  338 {
  339         int error;
  340 
  341         if ((bp->bio_flags & BIO_UNMAPPED) == 0)
  342                 error = sglist_append(sg, bp->bio_data, bp->bio_bcount);
  343         else
  344                 error = sglist_append_vmpages(sg, bp->bio_ma,
  345                     bp->bio_ma_offset, bp->bio_bcount);
  346         return (error);
  347 }
  348 
  349 /*
  350  * Append a single physical address range to a scatter/gather list.
  351  * If there are insufficient segments, then this fails with EFBIG.
  352  */
  353 int
  354 sglist_append_phys(struct sglist *sg, vm_paddr_t paddr, size_t len)
  355 {
  356         struct sglist_seg *ss;
  357         struct sgsave save;
  358         int error;
  359 
  360         if (sg->sg_maxseg == 0)
  361                 return (EINVAL);
  362         if (len == 0)
  363                 return (0);
  364 
  365         if (sg->sg_nseg == 0) {
  366                 sg->sg_segs[0].ss_paddr = paddr;
  367                 sg->sg_segs[0].ss_len = len;
  368                 sg->sg_nseg = 1;
  369                 return (0);
  370         }
  371         ss = &sg->sg_segs[sg->sg_nseg - 1];
  372         SGLIST_SAVE(sg, save);
  373         error = _sglist_append_range(sg, &ss, paddr, len);
  374         if (error)
  375                 SGLIST_RESTORE(sg, save);
  376         return (error);
  377 }
  378 
  379 /*
  380  * Append the segments of single multi-page mbuf.
  381  * If there are insufficient segments, then this fails with EFBIG.
  382  */
  383 int
  384 sglist_append_mbuf_epg(struct sglist *sg, struct mbuf *m, size_t off,
  385     size_t len)
  386 {
  387         size_t seglen, segoff;
  388         vm_paddr_t paddr;
  389         int error, i, pglen, pgoff;
  390 
  391         M_ASSERTEXTPG(m);
  392 
  393         error = 0;
  394         if (m->m_epg_hdrlen != 0) {
  395                 if (off >= m->m_epg_hdrlen) {
  396                         off -= m->m_epg_hdrlen;
  397                 } else {
  398                         seglen = m->m_epg_hdrlen - off;
  399                         segoff = off;
  400                         seglen = MIN(seglen, len);
  401                         off = 0;
  402                         len -= seglen;
  403                         error = sglist_append(sg,
  404                             &m->m_epg_hdr[segoff], seglen);
  405                 }
  406         }
  407         pgoff = m->m_epg_1st_off;
  408         for (i = 0; i < m->m_epg_npgs && error == 0 && len > 0; i++) {
  409                 pglen = m_epg_pagelen(m, i, pgoff);
  410                 if (off >= pglen) {
  411                         off -= pglen;
  412                         pgoff = 0;
  413                         continue;
  414                 }
  415                 seglen = pglen - off;
  416                 segoff = pgoff + off;
  417                 off = 0;
  418                 seglen = MIN(seglen, len);
  419                 len -= seglen;
  420                 paddr = m->m_epg_pa[i] + segoff;
  421                 error = sglist_append_phys(sg, paddr, seglen);
  422                 pgoff = 0;
  423         };
  424         if (error == 0 && len > 0) {
  425                 seglen = MIN(len, m->m_epg_trllen - off);
  426                 len -= seglen;
  427                 error = sglist_append(sg,
  428                     &m->m_epg_trail[off], seglen);
  429         }
  430         if (error == 0)
  431                 KASSERT(len == 0, ("len != 0"));
  432         return (error);
  433 }
  434 
  435 /*
  436  * Append the segments that describe a single mbuf chain to a
  437  * scatter/gather list.  If there are insufficient segments, then this
  438  * fails with EFBIG.
  439  */
  440 int
  441 sglist_append_mbuf(struct sglist *sg, struct mbuf *m0)
  442 {
  443         struct sgsave save;
  444         struct mbuf *m;
  445         int error;
  446 
  447         if (sg->sg_maxseg == 0)
  448                 return (EINVAL);
  449 
  450         error = 0;
  451         SGLIST_SAVE(sg, save);
  452         for (m = m0; m != NULL; m = m->m_next) {
  453                 if (m->m_len > 0) {
  454                         if ((m->m_flags & M_EXTPG) != 0)
  455                                 error = sglist_append_mbuf_epg(sg, m,
  456                                     mtod(m, vm_offset_t), m->m_len);
  457                         else
  458                                 error = sglist_append(sg, m->m_data,
  459                                     m->m_len);
  460                         if (error) {
  461                                 SGLIST_RESTORE(sg, save);
  462                                 return (error);
  463                         }
  464                 }
  465         }
  466         return (0);
  467 }
  468 
  469 /*
  470  * Append the segments that describe a single mbuf to a scatter/gather
  471  * list.  If there are insufficient segments, then this fails with
  472  * EFBIG.
  473  */
  474 int
  475 sglist_append_single_mbuf(struct sglist *sg, struct mbuf *m)
  476 {
  477         if ((m->m_flags & M_EXTPG) != 0)
  478                 return (sglist_append_mbuf_epg(sg, m,
  479                     mtod(m, vm_offset_t), m->m_len));
  480         else
  481                 return (sglist_append(sg, m->m_data, m->m_len));
  482 }
  483 
  484 /*
  485  * Append the segments that describe a buffer spanning an array of VM
  486  * pages.  The buffer begins at an offset of 'pgoff' in the first
  487  * page.
  488  */
  489 int
  490 sglist_append_vmpages(struct sglist *sg, vm_page_t *m, size_t pgoff,
  491     size_t len)
  492 {
  493         struct sgsave save;
  494         struct sglist_seg *ss;
  495         vm_paddr_t paddr;
  496         size_t seglen;
  497         int error, i;
  498 
  499         if (sg->sg_maxseg == 0)
  500                 return (EINVAL);
  501         if (len == 0)
  502                 return (0);
  503 
  504         SGLIST_SAVE(sg, save);
  505         i = 0;
  506         if (sg->sg_nseg == 0) {
  507                 seglen = min(PAGE_SIZE - pgoff, len);
  508                 sg->sg_segs[0].ss_paddr = VM_PAGE_TO_PHYS(m[0]) + pgoff;
  509                 sg->sg_segs[0].ss_len = seglen;
  510                 sg->sg_nseg = 1;
  511                 pgoff = 0;
  512                 len -= seglen;
  513                 i++;
  514         }
  515         ss = &sg->sg_segs[sg->sg_nseg - 1];
  516         for (; len > 0; i++, len -= seglen) {
  517                 seglen = min(PAGE_SIZE - pgoff, len);
  518                 paddr = VM_PAGE_TO_PHYS(m[i]) + pgoff;
  519                 error = _sglist_append_range(sg, &ss, paddr, seglen);
  520                 if (error) {
  521                         SGLIST_RESTORE(sg, save);
  522                         return (error);
  523                 }
  524                 pgoff = 0;
  525         }
  526         return (0);
  527 }
  528 
  529 /*
  530  * Append the segments that describe a single user address range to a
  531  * scatter/gather list.  If there are insufficient segments, then this
  532  * fails with EFBIG.
  533  */
  534 int
  535 sglist_append_user(struct sglist *sg, void *buf, size_t len, struct thread *td)
  536 {
  537         struct sgsave save;
  538         int error;
  539 
  540         if (sg->sg_maxseg == 0)
  541                 return (EINVAL);
  542         SGLIST_SAVE(sg, save);
  543         error = _sglist_append_buf(sg, buf, len,
  544             vmspace_pmap(td->td_proc->p_vmspace), NULL);
  545         if (error)
  546                 SGLIST_RESTORE(sg, save);
  547         return (error);
  548 }
  549 
  550 /*
  551  * Append a subset of an existing scatter/gather list 'source' to a
  552  * the scatter/gather list 'sg'.  If there are insufficient segments,
  553  * then this fails with EFBIG.
  554  */
  555 int
  556 sglist_append_sglist(struct sglist *sg, struct sglist *source, size_t offset,
  557     size_t length)
  558 {
  559         struct sgsave save;
  560         struct sglist_seg *ss;
  561         size_t seglen;
  562         int error, i;
  563 
  564         if (sg->sg_maxseg == 0 || length == 0)
  565                 return (EINVAL);
  566         SGLIST_SAVE(sg, save);
  567         error = EINVAL;
  568         ss = &sg->sg_segs[sg->sg_nseg - 1];
  569         for (i = 0; i < source->sg_nseg; i++) {
  570                 if (offset >= source->sg_segs[i].ss_len) {
  571                         offset -= source->sg_segs[i].ss_len;
  572                         continue;
  573                 }
  574                 seglen = source->sg_segs[i].ss_len - offset;
  575                 if (seglen > length)
  576                         seglen = length;
  577                 error = _sglist_append_range(sg, &ss,
  578                     source->sg_segs[i].ss_paddr + offset, seglen);
  579                 if (error)
  580                         break;
  581                 offset = 0;
  582                 length -= seglen;
  583                 if (length == 0)
  584                         break;
  585         }
  586         if (length != 0)
  587                 error = EINVAL;
  588         if (error)
  589                 SGLIST_RESTORE(sg, save);
  590         return (error);
  591 }
  592 
  593 /*
  594  * Append the segments that describe a single uio to a scatter/gather
  595  * list.  If there are insufficient segments, then this fails with
  596  * EFBIG.
  597  */
  598 int
  599 sglist_append_uio(struct sglist *sg, struct uio *uio)
  600 {
  601         struct iovec *iov;
  602         struct sgsave save;
  603         size_t resid, minlen;
  604         pmap_t pmap;
  605         int error, i;
  606 
  607         if (sg->sg_maxseg == 0)
  608                 return (EINVAL);
  609 
  610         resid = uio->uio_resid;
  611         iov = uio->uio_iov;
  612 
  613         if (uio->uio_segflg == UIO_USERSPACE) {
  614                 KASSERT(uio->uio_td != NULL,
  615                     ("sglist_append_uio: USERSPACE but no thread"));
  616                 pmap = vmspace_pmap(uio->uio_td->td_proc->p_vmspace);
  617         } else
  618                 pmap = NULL;
  619 
  620         error = 0;
  621         SGLIST_SAVE(sg, save);
  622         for (i = 0; i < uio->uio_iovcnt && resid != 0; i++) {
  623                 /*
  624                  * Now at the first iovec to load.  Load each iovec
  625                  * until we have exhausted the residual count.
  626                  */
  627                 minlen = MIN(resid, iov[i].iov_len);
  628                 if (minlen > 0) {
  629                         error = _sglist_append_buf(sg, iov[i].iov_base, minlen,
  630                             pmap, NULL);
  631                         if (error) {
  632                                 SGLIST_RESTORE(sg, save);
  633                                 return (error);
  634                         }
  635                         resid -= minlen;
  636                 }
  637         }
  638         return (0);
  639 }
  640 
  641 /*
  642  * Append the segments that describe at most 'resid' bytes from a
  643  * single uio to a scatter/gather list.  If there are insufficient
  644  * segments, then only the amount that fits is appended.
  645  */
  646 int
  647 sglist_consume_uio(struct sglist *sg, struct uio *uio, size_t resid)
  648 {
  649         struct iovec *iov;
  650         size_t done;
  651         pmap_t pmap;
  652         int error, len;
  653 
  654         if (sg->sg_maxseg == 0)
  655                 return (EINVAL);
  656 
  657         if (uio->uio_segflg == UIO_USERSPACE) {
  658                 KASSERT(uio->uio_td != NULL,
  659                     ("sglist_consume_uio: USERSPACE but no thread"));
  660                 pmap = vmspace_pmap(uio->uio_td->td_proc->p_vmspace);
  661         } else
  662                 pmap = NULL;
  663 
  664         error = 0;
  665         while (resid > 0 && uio->uio_resid) {
  666                 iov = uio->uio_iov;
  667                 len = iov->iov_len;
  668                 if (len == 0) {
  669                         uio->uio_iov++;
  670                         uio->uio_iovcnt--;
  671                         continue;
  672                 }
  673                 if (len > resid)
  674                         len = resid;
  675 
  676                 /*
  677                  * Try to append this iovec.  If we run out of room,
  678                  * then break out of the loop.
  679                  */
  680                 error = _sglist_append_buf(sg, iov->iov_base, len, pmap, &done);
  681                 iov->iov_base = (char *)iov->iov_base + done;
  682                 iov->iov_len -= done;
  683                 uio->uio_resid -= done;
  684                 uio->uio_offset += done;
  685                 resid -= done;
  686                 if (error)
  687                         break;
  688         }
  689         return (0);
  690 }
  691 
  692 /*
  693  * Allocate and populate a scatter/gather list to describe a single
  694  * kernel virtual address range.
  695  */
  696 struct sglist *
  697 sglist_build(void *buf, size_t len, int mflags)
  698 {
  699         struct sglist *sg;
  700         int nsegs;
  701 
  702         if (len == 0)
  703                 return (NULL);
  704 
  705         nsegs = sglist_count(buf, len);
  706         sg = sglist_alloc(nsegs, mflags);
  707         if (sg == NULL)
  708                 return (NULL);
  709         if (sglist_append(sg, buf, len) != 0) {
  710                 sglist_free(sg);
  711                 return (NULL);
  712         }
  713         return (sg);
  714 }
  715 
  716 /*
  717  * Clone a new copy of a scatter/gather list.
  718  */
  719 struct sglist *
  720 sglist_clone(struct sglist *sg, int mflags)
  721 {
  722         struct sglist *new;
  723 
  724         if (sg == NULL)
  725                 return (NULL);
  726         new = sglist_alloc(sg->sg_maxseg, mflags);
  727         if (new == NULL)
  728                 return (NULL);
  729         new->sg_nseg = sg->sg_nseg;
  730         bcopy(sg->sg_segs, new->sg_segs, sizeof(struct sglist_seg) *
  731             sg->sg_nseg);
  732         return (new);
  733 }
  734 
  735 /*
  736  * Calculate the total length of the segments described in a
  737  * scatter/gather list.
  738  */
  739 size_t
  740 sglist_length(struct sglist *sg)
  741 {
  742         size_t space;
  743         int i;
  744 
  745         space = 0;
  746         for (i = 0; i < sg->sg_nseg; i++)
  747                 space += sg->sg_segs[i].ss_len;
  748         return (space);
  749 }
  750 
  751 /*
  752  * Split a scatter/gather list into two lists.  The scatter/gather
  753  * entries for the first 'length' bytes of the 'original' list are
  754  * stored in the '*head' list and are removed from 'original'.
  755  *
  756  * If '*head' is NULL, then a new list will be allocated using
  757  * 'mflags'.  If M_NOWAIT is specified and the allocation fails,
  758  * ENOMEM will be returned.
  759  *
  760  * If '*head' is not NULL, it should point to an empty sglist.  If it
  761  * does not have enough room for the remaining space, then EFBIG will
  762  * be returned.  If '*head' is not empty, then EINVAL will be
  763  * returned.
  764  *
  765  * If 'original' is shared (refcount > 1), then EDOOFUS will be
  766  * returned.
  767  */
  768 int
  769 sglist_split(struct sglist *original, struct sglist **head, size_t length,
  770     int mflags)
  771 {
  772         struct sglist *sg;
  773         size_t space, split;
  774         int count, i;
  775 
  776         if (original->sg_refs > 1)
  777                 return (EDOOFUS);
  778 
  779         /* Figure out how big of a sglist '*head' has to hold. */
  780         count = 0;
  781         space = 0;
  782         split = 0;
  783         for (i = 0; i < original->sg_nseg; i++) {
  784                 space += original->sg_segs[i].ss_len;
  785                 count++;
  786                 if (space >= length) {
  787                         /*
  788                          * If 'length' falls in the middle of a
  789                          * scatter/gather list entry, then 'split'
  790                          * holds how much of that entry will remain in
  791                          * 'original'.
  792                          */
  793                         split = space - length;
  794                         break;
  795                 }
  796         }
  797 
  798         /* Nothing to do, so leave head empty. */
  799         if (count == 0)
  800                 return (0);
  801 
  802         if (*head == NULL) {
  803                 sg = sglist_alloc(count, mflags);
  804                 if (sg == NULL)
  805                         return (ENOMEM);
  806                 *head = sg;
  807         } else {
  808                 sg = *head;
  809                 if (sg->sg_maxseg < count)
  810                         return (EFBIG);
  811                 if (sg->sg_nseg != 0)
  812                         return (EINVAL);
  813         }
  814 
  815         /* Copy 'count' entries to 'sg' from 'original'. */
  816         bcopy(original->sg_segs, sg->sg_segs, count *
  817             sizeof(struct sglist_seg));
  818         sg->sg_nseg = count;
  819 
  820         /*
  821          * If we had to split a list entry, fixup the last entry in
  822          * 'sg' and the new first entry in 'original'.  We also
  823          * decrement 'count' by 1 since we will only be removing
  824          * 'count - 1' segments from 'original' now.
  825          */
  826         if (split != 0) {
  827                 count--;
  828                 sg->sg_segs[count].ss_len -= split;
  829                 original->sg_segs[count].ss_paddr =
  830                     sg->sg_segs[count].ss_paddr + split;
  831                 original->sg_segs[count].ss_len = split;
  832         }
  833 
  834         /* Trim 'count' entries from the front of 'original'. */
  835         original->sg_nseg -= count;
  836         bcopy(original->sg_segs + count, original->sg_segs, count *
  837             sizeof(struct sglist_seg));
  838         return (0);
  839 }
  840 
  841 /*
  842  * Append the scatter/gather list elements in 'second' to the
  843  * scatter/gather list 'first'.  If there is not enough space in
  844  * 'first', EFBIG is returned.
  845  */
  846 int
  847 sglist_join(struct sglist *first, struct sglist *second)
  848 {
  849         struct sglist_seg *flast, *sfirst;
  850         int append;
  851 
  852         /* If 'second' is empty, there is nothing to do. */
  853         if (second->sg_nseg == 0)
  854                 return (0);
  855 
  856         /*
  857          * If the first entry in 'second' can be appended to the last entry
  858          * in 'first' then set append to '1'.
  859          */
  860         append = 0;
  861         flast = &first->sg_segs[first->sg_nseg - 1];
  862         sfirst = &second->sg_segs[0];
  863         if (first->sg_nseg != 0 &&
  864             flast->ss_paddr + flast->ss_len == sfirst->ss_paddr)
  865                 append = 1;
  866 
  867         /* Make sure 'first' has enough room. */
  868         if (first->sg_nseg + second->sg_nseg - append > first->sg_maxseg)
  869                 return (EFBIG);
  870 
  871         /* Merge last in 'first' and first in 'second' if needed. */
  872         if (append)
  873                 flast->ss_len += sfirst->ss_len;
  874 
  875         /* Append new segments from 'second' to 'first'. */
  876         bcopy(first->sg_segs + first->sg_nseg, second->sg_segs + append,
  877             (second->sg_nseg - append) * sizeof(struct sglist_seg));
  878         first->sg_nseg += second->sg_nseg - append;
  879         sglist_reset(second);
  880         return (0);
  881 }
  882 
  883 /*
  884  * Generate a new scatter/gather list from a range of an existing
  885  * scatter/gather list.  The 'offset' and 'length' parameters specify
  886  * the logical range of the 'original' list to extract.  If that range
  887  * is not a subset of the length of 'original', then EINVAL is
  888  * returned.  The new scatter/gather list is stored in '*slice'.
  889  *
  890  * If '*slice' is NULL, then a new list will be allocated using
  891  * 'mflags'.  If M_NOWAIT is specified and the allocation fails,
  892  * ENOMEM will be returned.
  893  *
  894  * If '*slice' is not NULL, it should point to an empty sglist.  If it
  895  * does not have enough room for the remaining space, then EFBIG will
  896  * be returned.  If '*slice' is not empty, then EINVAL will be
  897  * returned.
  898  */
  899 int
  900 sglist_slice(struct sglist *original, struct sglist **slice, size_t offset,
  901     size_t length, int mflags)
  902 {
  903         struct sglist *sg;
  904         size_t space, end, foffs, loffs;
  905         int count, i, fseg;
  906 
  907         /* Nothing to do. */
  908         if (length == 0)
  909                 return (0);
  910 
  911         /* Figure out how many segments '*slice' needs to have. */
  912         end = offset + length;
  913         space = 0;
  914         count = 0;
  915         fseg = 0;
  916         foffs = loffs = 0;
  917         for (i = 0; i < original->sg_nseg; i++) {
  918                 space += original->sg_segs[i].ss_len;
  919                 if (space > offset) {
  920                         /*
  921                          * When we hit the first segment, store its index
  922                          * in 'fseg' and the offset into the first segment
  923                          * of 'offset' in 'foffs'.
  924                          */
  925                         if (count == 0) {
  926                                 fseg = i;
  927                                 foffs = offset - (space -
  928                                     original->sg_segs[i].ss_len);
  929                                 CTR1(KTR_DEV, "sglist_slice: foffs = %08lx",
  930                                     foffs);
  931                         }
  932                         count++;
  933 
  934                         /*
  935                          * When we hit the last segment, break out of
  936                          * the loop.  Store the amount of extra space
  937                          * at the end of this segment in 'loffs'.
  938                          */
  939                         if (space >= end) {
  940                                 loffs = space - end;
  941                                 CTR1(KTR_DEV, "sglist_slice: loffs = %08lx",
  942                                     loffs);
  943                                 break;
  944                         }
  945                 }
  946         }
  947 
  948         /* If we never hit 'end', then 'length' ran off the end, so fail. */
  949         if (space < end)
  950                 return (EINVAL);
  951 
  952         if (*slice == NULL) {
  953                 sg = sglist_alloc(count, mflags);
  954                 if (sg == NULL)
  955                         return (ENOMEM);
  956                 *slice = sg;
  957         } else {
  958                 sg = *slice;
  959                 if (sg->sg_maxseg < count)
  960                         return (EFBIG);
  961                 if (sg->sg_nseg != 0)
  962                         return (EINVAL);
  963         }
  964 
  965         /*
  966          * Copy over 'count' segments from 'original' starting at
  967          * 'fseg' to 'sg'.
  968          */
  969         bcopy(original->sg_segs + fseg, sg->sg_segs,
  970             count * sizeof(struct sglist_seg));
  971         sg->sg_nseg = count;
  972 
  973         /* Fixup first and last segments if needed. */
  974         if (foffs != 0) {
  975                 sg->sg_segs[0].ss_paddr += foffs;
  976                 sg->sg_segs[0].ss_len -= foffs;
  977                 CTR2(KTR_DEV, "sglist_slice seg[0]: %08lx:%08lx",
  978                     (long)sg->sg_segs[0].ss_paddr, sg->sg_segs[0].ss_len);
  979         }
  980         if (loffs != 0) {
  981                 sg->sg_segs[count - 1].ss_len -= loffs;
  982                 CTR2(KTR_DEV, "sglist_slice seg[%d]: len %08x", count - 1,
  983                     sg->sg_segs[count - 1].ss_len);
  984         }
  985         return (0);
  986 }

Cache object: a858c95396a653488d340e894952c462


[ source navigation ] [ diff markup ] [ identifier search ] [ freetext search ] [ file search ] [ list types ] [ track identifier ]


This page is part of the FreeBSD/Linux Linux Kernel Cross-Reference, and was automatically generated using a modified version of the LXR engine.