static void
vfs_mountroot_wait(void)
{
struct root_hold_token *h;
struct timeval lastfail;
int curfail;
curfail = 0;
while (1) {
DROP_GIANT();
g_waitidle();
PICKUP_GIANT();
mtx_lock(&mountlist_mtx);
if (LIST_EMPTY(&root_holds)) {
mtx_unlock(&mountlist_mtx);
break;
}
if (ppsratecheck(&lastfail, &curfail, 1)) {
printf("Root mount waiting for:");
LIST_FOREACH(h, &root_holds, list)
printf(" %s", h->who);
printf("\n");
}
msleep(&root_holds, &mountlist_mtx, PZERO | PDROP, "roothold",
hz);
}
}
/*
* Logging API
*/
void
rlprintf(int pps, const char *fmt, ...)
{
va_list ap;
static struct timeval last_printf;
static int count;
if (ppsratecheck(&last_printf, &count, pps)) {
va_start(ap, fmt);
vprintf(fmt, ap);
va_end(ap);
}
}
void
device_rlprintf(int pps, device_t dev, const char *fmt, ...)
{
va_list ap;
static struct timeval last_printf;
static int count;
if (ppsratecheck(&last_printf, &count, pps)) {
va_start(ap, fmt);
device_print_prettyname(dev);
vprintf(fmt, ap);
va_end(ap);
}
}
/*
* Called by KASSERT, this decides if we will panic
* or if we will log via printf and/or ktr.
*/
void
kassert_panic(const char *fmt, ...)
{
static char buf[256];
va_list ap;
va_start(ap, fmt);
(void)vsnprintf(buf, sizeof(buf), fmt, ap);
va_end(ap);
/*
* panic if we're not just warning, or if we've exceeded
* kassert_log_panic_at warnings.
*/
if (!kassert_warn_only ||
(kassert_log_panic_at > 0 &&
kassert_warnings >= kassert_log_panic_at)) {
va_start(ap, fmt);
vpanic(fmt, ap);
/* NORETURN */
}
#ifdef KTR
if (kassert_do_ktr)
CTR0(ktr_mask, buf);
#endif /* KTR */
/*
* log if we've not yet met the mute limit.
*/
if (kassert_do_log &&
(kassert_log_mute_at == 0 ||
kassert_warnings < kassert_log_mute_at)) {
static struct timeval lasterr;
static int curerr;
if (ppsratecheck(&lasterr, &curerr, kassert_log_pps_limit)) {
printf("KASSERT failed: %s\n", buf);
kdb_backtrace();
}
}
#ifdef KDB
if (kassert_do_kdb) {
kdb_enter(KDB_WHY_KASSERT, buf);
}
#endif
atomic_add_int(&kassert_warnings, 1);
}
static int
codel_should_drop(struct codel *c, class_queue_t *q, struct mbuf *m,
u_int64_t now)
{
struct m_tag *mtag;
uint64_t *enqueue_time;
if (m == NULL) {
c->vars.first_above_time = 0;
return (0);
}
mtag = m_tag_locate(m, MTAG_CODEL, 0, NULL);
if (mtag == NULL) {
/* Only one warning per second. */
if (ppsratecheck(&c->last_log, &c->last_pps, 1))
printf("%s: could not found the packet mtag!\n",
__func__);
c->vars.first_above_time = 0;
return (0);
}
enqueue_time = (uint64_t *)(mtag + 1);
c->vars.ldelay = now - *enqueue_time;
c->stats.maxpacket = MAX(c->stats.maxpacket, m_pktlen(m));
if (codel_time_before(c->vars.ldelay, c->params.target) ||
qsize(q) <= c->stats.maxpacket) {
/* went below - stay below for at least interval */
c->vars.first_above_time = 0;
return (0);
}
if (c->vars.first_above_time == 0) {
/* just went above from below. If we stay above
* for at least interval we'll say it's ok to drop
*/
c->vars.first_above_time = now + c->params.interval;
return (0);
}
if (codel_time_after(now, c->vars.first_above_time))
return (1);
return (0);
}
/*
* Handle a radar detection event on a channel. The channel is
* added to the NOL list and we record the time of the event.
* Entries are aged out after NOL_TIMEOUT. If radar was
* detected while doing CAC we force a state/channel change.
* Otherwise radar triggers a channel switch using the CSA
* mechanism (when the channel is the bss channel).
*/
void
ieee80211_dfs_notify_radar(struct ieee80211com *ic, struct ieee80211_channel *chan)
{
struct ieee80211_dfs_state *dfs = &ic->ic_dfs;
int i, now;
IEEE80211_LOCK_ASSERT(ic);
/*
* If doing DFS debugging (mode 2), don't bother
* running the rest of this function.
*
* Simply announce the presence of the radar and continue
* along merrily.
*/
if (ieee80211_dfs_debug == DFS_DBG_NOCSANOL) {
announce_radar(ic->ic_ifp, chan, chan);
ieee80211_notify_radar(ic, chan);
return;
}
/*
* Don't mark the channel and don't put it into NOL
* if we're doing DFS debugging.
*/
if (ieee80211_dfs_debug == DFS_DBG_NONE) {
/*
* Mark all entries with this frequency. Notify user
* space and arrange for notification when the radar
* indication is cleared. Then kick the NOL processing
* thread if not already running.
*/
now = ticks;
for (i = 0; i < ic->ic_nchans; i++) {
struct ieee80211_channel *c = &ic->ic_channels[i];
if (c->ic_freq == chan->ic_freq) {
c->ic_state &= ~IEEE80211_CHANSTATE_CACDONE;
c->ic_state |= IEEE80211_CHANSTATE_RADAR;
dfs->nol_event[i] = now;
}
}
ieee80211_notify_radar(ic, chan);
chan->ic_state |= IEEE80211_CHANSTATE_NORADAR;
if (!callout_pending(&dfs->nol_timer))
callout_reset(&dfs->nol_timer, NOL_TIMEOUT,
dfs_timeout, ic);
}
/*
* If radar is detected on the bss channel while
* doing CAC; force a state change by scheduling the
* callout to be dispatched asap. Otherwise, if this
* event is for the bss channel then we must quiet
* traffic and schedule a channel switch.
*
* Note this allows us to receive notification about
* channels other than the bss channel; not sure
* that can/will happen but it's simple to support.
*/
if (chan == ic->ic_bsschan) {
/* XXX need a way to defer to user app */
/*
* Don't flip over to a new channel if
* we are currently doing DFS debugging.
*/
if (ieee80211_dfs_debug == DFS_DBG_NONE)
dfs->newchan = ieee80211_dfs_pickchannel(ic);
else
dfs->newchan = chan;
announce_radar(ic->ic_ifp, chan, dfs->newchan);
if (callout_pending(&dfs->cac_timer))
callout_schedule(&dfs->cac_timer, 0);
else if (dfs->newchan != NULL) {
/* XXX mode 1, switch count 2 */
/* XXX calculate switch count based on max
switch time and beacon interval? */
ieee80211_csa_startswitch(ic, dfs->newchan, 1, 2);
} else {
/*
* Spec says to stop all transmissions and
* wait on the current channel for an entry
* on the NOL to expire.
*/
/*XXX*/
if_printf(ic->ic_ifp, "%s: No free channels; waiting for entry "
"on NOL to expire\n", __func__);
}
} else {
/*
* Issue rate-limited console msgs.
*/
if (dfs->lastchan != chan) {
dfs->lastchan = chan;
dfs->cureps = 0;
announce_radar(ic->ic_ifp, chan, NULL);
} else if (ppsratecheck(&dfs->lastevent, &dfs->cureps, 1)) {
announce_radar(ic->ic_ifp, chan, NULL);
}
}
}
/*
* This is now called from local media FS's to operate against their
* own vnodes if they fail to implement VOP_PUTPAGES.
*
* This is typically called indirectly via the pageout daemon and
* clustering has already typically occurred, so in general we ask the
* underlying filesystem to write the data out asynchronously rather
* then delayed.
*/
int
vnode_pager_generic_putpages(struct vnode *vp, vm_page_t *ma, int bytecount,
int flags, int *rtvals)
{
int i;
vm_object_t object;
vm_page_t m;
int count;
int maxsize, ncount;
vm_ooffset_t poffset;
struct uio auio;
struct iovec aiov;
int error;
int ioflags;
int ppscheck = 0;
static struct timeval lastfail;
static int curfail;
object = vp->v_object;
count = bytecount / PAGE_SIZE;
for (i = 0; i < count; i++)
rtvals[i] = VM_PAGER_ERROR;
if ((int64_t)ma[0]->pindex < 0) {
printf("vnode_pager_putpages: attempt to write meta-data!!! -- 0x%lx(%lx)\n",
(long)ma[0]->pindex, (u_long)ma[0]->dirty);
rtvals[0] = VM_PAGER_BAD;
return VM_PAGER_BAD;
}
maxsize = count * PAGE_SIZE;
ncount = count;
poffset = IDX_TO_OFF(ma[0]->pindex);
/*
* If the page-aligned write is larger then the actual file we
* have to invalidate pages occurring beyond the file EOF. However,
* there is an edge case where a file may not be page-aligned where
* the last page is partially invalid. In this case the filesystem
* may not properly clear the dirty bits for the entire page (which
* could be VM_PAGE_BITS_ALL due to the page having been mmap()d).
* With the page locked we are free to fix-up the dirty bits here.
*
* We do not under any circumstances truncate the valid bits, as
* this will screw up bogus page replacement.
*/
VM_OBJECT_WLOCK(object);
if (maxsize + poffset > object->un_pager.vnp.vnp_size) {
if (object->un_pager.vnp.vnp_size > poffset) {
int pgoff;
maxsize = object->un_pager.vnp.vnp_size - poffset;
ncount = btoc(maxsize);
if ((pgoff = (int)maxsize & PAGE_MASK) != 0) {
/*
* If the object is locked and the following
* conditions hold, then the page's dirty
* field cannot be concurrently changed by a
* pmap operation.
*/
m = ma[ncount - 1];
vm_page_assert_sbusied(m);
KASSERT(!pmap_page_is_write_mapped(m),
("vnode_pager_generic_putpages: page %p is not read-only", m));
vm_page_clear_dirty(m, pgoff, PAGE_SIZE -
pgoff);
}
} else {
maxsize = 0;
ncount = 0;
}
if (ncount < count) {
for (i = ncount; i < count; i++) {
rtvals[i] = VM_PAGER_BAD;
}
}
}
VM_OBJECT_WUNLOCK(object);
/*
* pageouts are already clustered, use IO_ASYNC to force a bawrite()
* rather then a bdwrite() to prevent paging I/O from saturating
* the buffer cache. Dummy-up the sequential heuristic to cause
* large ranges to cluster. If neither IO_SYNC or IO_ASYNC is set,
* the system decides how to cluster.
*/
ioflags = IO_VMIO;
if (flags & (VM_PAGER_PUT_SYNC | VM_PAGER_PUT_INVAL))
ioflags |= IO_SYNC;
else if ((flags & VM_PAGER_CLUSTER_OK) == 0)
ioflags |= IO_ASYNC;
ioflags |= (flags & VM_PAGER_PUT_INVAL) ? IO_INVAL: 0;
ioflags |= (flags & VM_PAGER_PUT_NOREUSE) ? IO_NOREUSE : 0;
ioflags |= IO_SEQMAX << IO_SEQSHIFT;
aiov.iov_base = (caddr_t) 0;
aiov.iov_len = maxsize;
auio.uio_iov = &aiov;
auio.uio_iovcnt = 1;
auio.uio_offset = poffset;
auio.uio_segflg = UIO_NOCOPY;
auio.uio_rw = UIO_WRITE;
auio.uio_resid = maxsize;
auio.uio_td = (struct thread *) 0;
error = VOP_WRITE(vp, &auio, ioflags, curthread->td_ucred);
PCPU_INC(cnt.v_vnodeout);
PCPU_ADD(cnt.v_vnodepgsout, ncount);
if (error) {
if ((ppscheck = ppsratecheck(&lastfail, &curfail, 1)))
printf("vnode_pager_putpages: I/O error %d\n", error);
}
if (auio.uio_resid) {
if (ppscheck || ppsratecheck(&lastfail, &curfail, 1))
printf("vnode_pager_putpages: residual I/O %zd at %lu\n",
auio.uio_resid, (u_long)ma[0]->pindex);
}
for (i = 0; i < ncount; i++) {
rtvals[i] = VM_PAGER_OK;
}
return rtvals[0];
}
/*
* malloc:
*
* Allocate a block of memory.
*
* If M_NOWAIT is set, this routine will not block and return NULL if
* the allocation fails.
*/
void *
malloc(unsigned long size, struct malloc_type *mtp, int flags)
{
int indx;
struct malloc_type_internal *mtip;
caddr_t va;
uma_zone_t zone;
#if defined(DIAGNOSTIC) || defined(DEBUG_REDZONE)
unsigned long osize = size;
#endif
#ifdef INVARIANTS
KASSERT(mtp->ks_magic == M_MAGIC, ("malloc: bad malloc type magic"));
/*
* Check that exactly one of M_WAITOK or M_NOWAIT is specified.
*/
indx = flags & (M_WAITOK | M_NOWAIT);
if (indx != M_NOWAIT && indx != M_WAITOK) {
static struct timeval lasterr;
static int curerr, once;
if (once == 0 && ppsratecheck(&lasterr, &curerr, 1)) {
printf("Bad malloc flags: %x\n", indx);
kdb_backtrace();
flags |= M_WAITOK;
once++;
}
}
#endif
#ifdef MALLOC_MAKE_FAILURES
if ((flags & M_NOWAIT) && (malloc_failure_rate != 0)) {
atomic_add_int(&malloc_nowait_count, 1);
if ((malloc_nowait_count % malloc_failure_rate) == 0) {
atomic_add_int(&malloc_failure_count, 1);
t_malloc_fail = time_uptime;
return (NULL);
}
}
#endif
if (flags & M_WAITOK)
KASSERT(curthread->td_intr_nesting_level == 0,
("malloc(M_WAITOK) in interrupt context"));
#ifdef DEBUG_MEMGUARD
if (memguard_cmp_mtp(mtp, size)) {
va = memguard_alloc(size, flags);
if (va != NULL)
return (va);
/* This is unfortunate but should not be fatal. */
}
#endif
#ifdef DEBUG_REDZONE
size = redzone_size_ntor(size);
#endif
if (size <= kmem_zmax) {
mtip = mtp->ks_handle;
if (size & KMEM_ZMASK)
size = (size & ~KMEM_ZMASK) + KMEM_ZBASE;
indx = kmemsize[size >> KMEM_ZSHIFT];
KASSERT(mtip->mti_zone < numzones,
("mti_zone %u out of range %d",
mtip->mti_zone, numzones));
zone = kmemzones[indx].kz_zone[mtip->mti_zone];
#ifdef MALLOC_PROFILE
krequests[size >> KMEM_ZSHIFT]++;
#endif
va = uma_zalloc(zone, flags);
if (va != NULL)
size = zone->uz_size;
malloc_type_zone_allocated(mtp, va == NULL ? 0 : size, indx);
} else {
/*
* Balloc defines the structure of filesystem storage
* by allocating the physical blocks on a device given
* the inode and the logical block number in a file.
* This is the allocation strategy for UFS1. Below is
* the allocation strategy for UFS2.
*/
int
ffs_balloc_ufs1(struct vnode *vp, off_t startoffset, int size,
struct ucred *cred, int flags, struct buf **bpp)
{
struct inode *ip;
struct ufs1_dinode *dp;
ufs_lbn_t lbn, lastlbn;
struct fs *fs;
ufs1_daddr_t nb;
struct buf *bp, *nbp;
struct ufsmount *ump;
struct indir indirs[NIADDR + 2];
int deallocated, osize, nsize, num, i, error;
ufs2_daddr_t newb;
ufs1_daddr_t *bap, pref;
ufs1_daddr_t *allocib, *blkp, *allocblk, allociblk[NIADDR + 1];
ufs2_daddr_t *lbns_remfree, lbns[NIADDR + 1];
int unwindidx = -1;
int saved_inbdflush;
static struct timeval lastfail;
static int curfail;
int reclaimed;
ip = VTOI(vp);
dp = ip->i_din1;
fs = ip->i_fs;
ump = ip->i_ump;
lbn = lblkno(fs, startoffset);
size = blkoff(fs, startoffset) + size;
reclaimed = 0;
if (size > fs->fs_bsize)
panic("ffs_balloc_ufs1: blk too big");
*bpp = NULL;
if (flags & IO_EXT)
return (EOPNOTSUPP);
if (lbn < 0)
return (EFBIG);
if (DOINGSOFTDEP(vp))
softdep_prealloc(vp, MNT_WAIT);
/*
* If the next write will extend the file into a new block,
* and the file is currently composed of a fragment
* this fragment has to be extended to be a full block.
*/
lastlbn = lblkno(fs, ip->i_size);
if (lastlbn < NDADDR && lastlbn < lbn) {
nb = lastlbn;
osize = blksize(fs, ip, nb);
if (osize < fs->fs_bsize && osize > 0) {
UFS_LOCK(ump);
error = ffs_realloccg(ip, nb, dp->di_db[nb],
ffs_blkpref_ufs1(ip, lastlbn, (int)nb,
&dp->di_db[0]), osize, (int)fs->fs_bsize, flags,
cred, &bp);
if (error)
return (error);
if (DOINGSOFTDEP(vp))
softdep_setup_allocdirect(ip, nb,
dbtofsb(fs, bp->b_blkno), dp->di_db[nb],
fs->fs_bsize, osize, bp);
ip->i_size = smalllblktosize(fs, nb + 1);
dp->di_size = ip->i_size;
dp->di_db[nb] = dbtofsb(fs, bp->b_blkno);
ip->i_flag |= IN_CHANGE | IN_UPDATE;
if (flags & IO_SYNC)
bwrite(bp);
else
bawrite(bp);
}
}
/*
* The first NDADDR blocks are direct blocks
*/
if (lbn < NDADDR) {
if (flags & BA_METAONLY)
panic("ffs_balloc_ufs1: BA_METAONLY for direct block");
nb = dp->di_db[lbn];
if (nb != 0 && ip->i_size >= smalllblktosize(fs, lbn + 1)) {
error = bread(vp, lbn, fs->fs_bsize, NOCRED, &bp);
if (error) {
brelse(bp);
return (error);
}
bp->b_blkno = fsbtodb(fs, nb);
*bpp = bp;
return (0);
}
if (nb != 0) {
/*
* Consider need to reallocate a fragment.
*/
osize = fragroundup(fs, blkoff(fs, ip->i_size));
nsize = fragroundup(fs, size);
if (nsize <= osize) {
error = bread(vp, lbn, osize, NOCRED, &bp);
if (error) {
brelse(bp);
return (error);
}
bp->b_blkno = fsbtodb(fs, nb);
} else {
UFS_LOCK(ump);
error = ffs_realloccg(ip, lbn, dp->di_db[lbn],
ffs_blkpref_ufs1(ip, lbn, (int)lbn,
&dp->di_db[0]), osize, nsize, flags,
cred, &bp);
if (error)
return (error);
if (DOINGSOFTDEP(vp))
softdep_setup_allocdirect(ip, lbn,
dbtofsb(fs, bp->b_blkno), nb,
nsize, osize, bp);
}
} else {
if (ip->i_size < smalllblktosize(fs, lbn + 1))
nsize = fragroundup(fs, size);
else
nsize = fs->fs_bsize;
UFS_LOCK(ump);
error = ffs_alloc(ip, lbn,
ffs_blkpref_ufs1(ip, lbn, (int)lbn, &dp->di_db[0]),
nsize, flags, cred, &newb);
if (error)
return (error);
bp = getblk(vp, lbn, nsize, 0, 0, 0);
bp->b_blkno = fsbtodb(fs, newb);
if (flags & BA_CLRBUF)
vfs_bio_clrbuf(bp);
if (DOINGSOFTDEP(vp))
softdep_setup_allocdirect(ip, lbn, newb, 0,
nsize, 0, bp);
}
dp->di_db[lbn] = dbtofsb(fs, bp->b_blkno);
ip->i_flag |= IN_CHANGE | IN_UPDATE;
*bpp = bp;
return (0);
}
/*
* Determine the number of levels of indirection.
*/
pref = 0;
if ((error = ufs_getlbns(vp, lbn, indirs, &num)) != 0)
return(error);
#ifdef INVARIANTS
if (num < 1)
panic ("ffs_balloc_ufs1: ufs_getlbns returned indirect block");
#endif
saved_inbdflush = curthread_pflags_set(TDP_INBDFLUSH);
/*
* Fetch the first indirect block allocating if necessary.
*/
--num;
nb = dp->di_ib[indirs[0].in_off];
allocib = NULL;
allocblk = allociblk;
lbns_remfree = lbns;
if (nb == 0) {
UFS_LOCK(ump);
pref = ffs_blkpref_ufs1(ip, lbn, 0, (ufs1_daddr_t *)0);
if ((error = ffs_alloc(ip, lbn, pref, (int)fs->fs_bsize,
flags, cred, &newb)) != 0) {
curthread_pflags_restore(saved_inbdflush);
return (error);
}
nb = newb;
*allocblk++ = nb;
*lbns_remfree++ = indirs[1].in_lbn;
bp = getblk(vp, indirs[1].in_lbn, fs->fs_bsize, 0, 0, 0);
bp->b_blkno = fsbtodb(fs, nb);
vfs_bio_clrbuf(bp);
if (DOINGSOFTDEP(vp)) {
softdep_setup_allocdirect(ip, NDADDR + indirs[0].in_off,
newb, 0, fs->fs_bsize, 0, bp);
bdwrite(bp);
} else {
/*
* Write synchronously so that indirect blocks
* never point at garbage.
*/
if (DOINGASYNC(vp))
bdwrite(bp);
else if ((error = bwrite(bp)) != 0)
goto fail;
}
allocib = &dp->di_ib[indirs[0].in_off];
*allocib = nb;
ip->i_flag |= IN_CHANGE | IN_UPDATE;
}
/*
* Fetch through the indirect blocks, allocating as necessary.
*/
retry:
for (i = 1;;) {
error = bread(vp,
indirs[i].in_lbn, (int)fs->fs_bsize, NOCRED, &bp);
if (error) {
brelse(bp);
goto fail;
}
bap = (ufs1_daddr_t *)bp->b_data;
nb = bap[indirs[i].in_off];
if (i == num)
break;
i += 1;
if (nb != 0) {
bqrelse(bp);
continue;
}
UFS_LOCK(ump);
if (pref == 0)
pref = ffs_blkpref_ufs1(ip, lbn, 0, (ufs1_daddr_t *)0);
if ((error = ffs_alloc(ip, lbn, pref, (int)fs->fs_bsize,
flags | IO_BUFLOCKED, cred, &newb)) != 0) {
brelse(bp);
if (++reclaimed == 1) {
UFS_LOCK(ump);
softdep_request_cleanup(fs, vp, cred,
FLUSH_BLOCKS_WAIT);
UFS_UNLOCK(ump);
goto retry;
}
if (ppsratecheck(&lastfail, &curfail, 1)) {
ffs_fserr(fs, ip->i_number, "filesystem full");
uprintf("\n%s: write failed, filesystem "
"is full\n", fs->fs_fsmnt);
}
goto fail;
}
nb = newb;
*allocblk++ = nb;
*lbns_remfree++ = indirs[i].in_lbn;
nbp = getblk(vp, indirs[i].in_lbn, fs->fs_bsize, 0, 0, 0);
nbp->b_blkno = fsbtodb(fs, nb);
vfs_bio_clrbuf(nbp);
if (DOINGSOFTDEP(vp)) {
softdep_setup_allocindir_meta(nbp, ip, bp,
indirs[i - 1].in_off, nb);
bdwrite(nbp);
} else {
/*
* Write synchronously so that indirect blocks
* never point at garbage.
*/
if ((error = bwrite(nbp)) != 0) {
brelse(bp);
goto fail;
}
}
bap[indirs[i - 1].in_off] = nb;
if (allocib == NULL && unwindidx < 0)
unwindidx = i - 1;
/*
* If required, write synchronously, otherwise use
* delayed write.
*/
if (flags & IO_SYNC) {
bwrite(bp);
} else {
if (bp->b_bufsize == fs->fs_bsize)
bp->b_flags |= B_CLUSTEROK;
bdwrite(bp);
}
}
/*
* If asked only for the indirect block, then return it.
*/
if (flags & BA_METAONLY) {
curthread_pflags_restore(saved_inbdflush);
*bpp = bp;
return (0);
}
/*
* Get the data block, allocating if necessary.
*/
if (nb == 0) {
UFS_LOCK(ump);
pref = ffs_blkpref_ufs1(ip, lbn, indirs[i].in_off, &bap[0]);
error = ffs_alloc(ip, lbn, pref, (int)fs->fs_bsize,
flags | IO_BUFLOCKED, cred, &newb);
if (error) {
brelse(bp);
if (++reclaimed == 1) {
UFS_LOCK(ump);
softdep_request_cleanup(fs, vp, cred,
FLUSH_BLOCKS_WAIT);
UFS_UNLOCK(ump);
goto retry;
}
if (ppsratecheck(&lastfail, &curfail, 1)) {
ffs_fserr(fs, ip->i_number, "filesystem full");
uprintf("\n%s: write failed, filesystem "
"is full\n", fs->fs_fsmnt);
}
goto fail;
}
nb = newb;
*allocblk++ = nb;
*lbns_remfree++ = lbn;
nbp = getblk(vp, lbn, fs->fs_bsize, 0, 0, 0);
nbp->b_blkno = fsbtodb(fs, nb);
if (flags & BA_CLRBUF)
vfs_bio_clrbuf(nbp);
if (DOINGSOFTDEP(vp))
softdep_setup_allocindir_page(ip, lbn, bp,
indirs[i].in_off, nb, 0, nbp);
bap[indirs[i].in_off] = nb;
/*
* If required, write synchronously, otherwise use
* delayed write.
*/
if (flags & IO_SYNC) {
bwrite(bp);
} else {
if (bp->b_bufsize == fs->fs_bsize)
bp->b_flags |= B_CLUSTEROK;
bdwrite(bp);
}
curthread_pflags_restore(saved_inbdflush);
*bpp = nbp;
return (0);
}
brelse(bp);
if (flags & BA_CLRBUF) {
int seqcount = (flags & BA_SEQMASK) >> BA_SEQSHIFT;
if (seqcount && (vp->v_mount->mnt_flag & MNT_NOCLUSTERR) == 0) {
error = cluster_read(vp, ip->i_size, lbn,
(int)fs->fs_bsize, NOCRED,
MAXBSIZE, seqcount, &nbp);
} else {
error = bread(vp, lbn, (int)fs->fs_bsize, NOCRED, &nbp);
}
if (error) {
brelse(nbp);
goto fail;
}
} else {
int
fork1(struct thread *td, struct fork_req *fr)
{
struct proc *p1, *newproc;
struct thread *td2;
struct vmspace *vm2;
struct file *fp_procdesc;
vm_ooffset_t mem_charged;
int error, nprocs_new, ok;
static int curfail;
static struct timeval lastfail;
int flags, pages;
flags = fr->fr_flags;
pages = fr->fr_pages;
if ((flags & RFSTOPPED) != 0)
MPASS(fr->fr_procp != NULL && fr->fr_pidp == NULL);
else
MPASS(fr->fr_procp == NULL);
/* Check for the undefined or unimplemented flags. */
if ((flags & ~(RFFLAGS | RFTSIGFLAGS(RFTSIGMASK))) != 0)
return (EINVAL);
/* Signal value requires RFTSIGZMB. */
if ((flags & RFTSIGFLAGS(RFTSIGMASK)) != 0 && (flags & RFTSIGZMB) == 0)
return (EINVAL);
/* Can't copy and clear. */
if ((flags & (RFFDG|RFCFDG)) == (RFFDG|RFCFDG))
return (EINVAL);
/* Check the validity of the signal number. */
if ((flags & RFTSIGZMB) != 0 && (u_int)RFTSIGNUM(flags) > _SIG_MAXSIG)
return (EINVAL);
if ((flags & RFPROCDESC) != 0) {
/* Can't not create a process yet get a process descriptor. */
if ((flags & RFPROC) == 0)
return (EINVAL);
/* Must provide a place to put a procdesc if creating one. */
if (fr->fr_pd_fd == NULL)
return (EINVAL);
/* Check if we are using supported flags. */
if ((fr->fr_pd_flags & ~PD_ALLOWED_AT_FORK) != 0)
return (EINVAL);
}
p1 = td->td_proc;
/*
* Here we don't create a new process, but we divorce
* certain parts of a process from itself.
*/
if ((flags & RFPROC) == 0) {
if (fr->fr_procp != NULL)
*fr->fr_procp = NULL;
else if (fr->fr_pidp != NULL)
*fr->fr_pidp = 0;
return (fork_norfproc(td, flags));
}
fp_procdesc = NULL;
newproc = NULL;
vm2 = NULL;
/*
* Increment the nprocs resource before allocations occur.
* Although process entries are dynamically created, we still
* keep a global limit on the maximum number we will
* create. There are hard-limits as to the number of processes
* that can run, established by the KVA and memory usage for
* the process data.
*
* Don't allow a nonprivileged user to use the last ten
* processes; don't let root exceed the limit.
*/
nprocs_new = atomic_fetchadd_int(&nprocs, 1) + 1;
if ((nprocs_new >= maxproc - 10 && priv_check_cred(td->td_ucred,
PRIV_MAXPROC, 0) != 0) || nprocs_new >= maxproc) {
error = EAGAIN;
sx_xlock(&allproc_lock);
if (ppsratecheck(&lastfail, &curfail, 1)) {
printf("maxproc limit exceeded by uid %u (pid %d); "
"see tuning(7) and login.conf(5)\n",
td->td_ucred->cr_ruid, p1->p_pid);
}
sx_xunlock(&allproc_lock);
goto fail2;
}
/*
* If required, create a process descriptor in the parent first; we
* will abandon it if something goes wrong. We don't finit() until
* later.
*/
if (flags & RFPROCDESC) {
error = procdesc_falloc(td, &fp_procdesc, fr->fr_pd_fd,
fr->fr_pd_flags, fr->fr_pd_fcaps);
if (error != 0)
goto fail2;
}
mem_charged = 0;
if (pages == 0)
pages = kstack_pages;
/* Allocate new proc. */
newproc = uma_zalloc(proc_zone, M_WAITOK);
td2 = FIRST_THREAD_IN_PROC(newproc);
if (td2 == NULL) {
td2 = thread_alloc(pages);
if (td2 == NULL) {
error = ENOMEM;
goto fail2;
}
proc_linkup(newproc, td2);
} else {
if (td2->td_kstack == 0 || td2->td_kstack_pages != pages) {
if (td2->td_kstack != 0)
vm_thread_dispose(td2);
if (!thread_alloc_stack(td2, pages)) {
error = ENOMEM;
goto fail2;
}
}
}
if ((flags & RFMEM) == 0) {
vm2 = vmspace_fork(p1->p_vmspace, &mem_charged);
if (vm2 == NULL) {
error = ENOMEM;
goto fail2;
}
if (!swap_reserve(mem_charged)) {
/*
* The swap reservation failed. The accounting
* from the entries of the copied vm2 will be
* subtracted in vmspace_free(), so force the
* reservation there.
*/
swap_reserve_force(mem_charged);
error = ENOMEM;
goto fail2;
}
} else
vm2 = NULL;
/*
* XXX: This is ugly; when we copy resource usage, we need to bump
* per-cred resource counters.
*/
proc_set_cred_init(newproc, crhold(td->td_ucred));
/*
* Initialize resource accounting for the child process.
*/
error = racct_proc_fork(p1, newproc);
if (error != 0) {
error = EAGAIN;
goto fail1;
}
#ifdef MAC
mac_proc_init(newproc);
#endif
newproc->p_klist = knlist_alloc(&newproc->p_mtx);
STAILQ_INIT(&newproc->p_ktr);
/* We have to lock the process tree while we look for a pid. */
sx_slock(&proctree_lock);
sx_xlock(&allproc_lock);
/*
* Increment the count of procs running with this uid. Don't allow
* a nonprivileged user to exceed their current limit.
*
* XXXRW: Can we avoid privilege here if it's not needed?
*/
error = priv_check_cred(td->td_ucred, PRIV_PROC_LIMIT, 0);
if (error == 0)
ok = chgproccnt(td->td_ucred->cr_ruidinfo, 1, 0);
else {
ok = chgproccnt(td->td_ucred->cr_ruidinfo, 1,
lim_cur(td, RLIMIT_NPROC));
}
if (ok) {
do_fork(td, fr, newproc, td2, vm2, fp_procdesc);
return (0);
}
error = EAGAIN;
sx_sunlock(&proctree_lock);
sx_xunlock(&allproc_lock);
#ifdef MAC
mac_proc_destroy(newproc);
#endif
racct_proc_exit(newproc);
fail1:
crfree(newproc->p_ucred);
newproc->p_ucred = NULL;
fail2:
if (vm2 != NULL)
vmspace_free(vm2);
uma_zfree(proc_zone, newproc);
if ((flags & RFPROCDESC) != 0 && fp_procdesc != NULL) {
fdclose(td, fp_procdesc, *fr->fr_pd_fd);
fdrop(fp_procdesc, td);
}
atomic_add_int(&nprocs, -1);
pause("fork", hz / 2);
return (error);
}
/*
* Balloc defines the structure of file system storage
* by allocating the physical blocks on a device given
* the inode and the logical block number in a file.
* This is the allocation strategy for UFS2. Above is
* the allocation strategy for UFS1.
*/
int
ffs_balloc_ufs2(vnode *vp, off_t startoffset, int size,
Ucred *cred, int flags, Buf **bpp)
{
int error = 0;
print("HARVEY TODO: %s\n", __func__);
#if 0
struct inode *ip;
struct ufs2_dinode *dp;
ufs_lbn_t lbn, lastlbn;
struct fs *fs;
struct buf *bp, *nbp;
struct ufsmount *ump;
struct indir indirs[UFS_NIADDR + 2];
ufs2_daddr_t nb, newb, *bap, pref;
ufs2_daddr_t *allocib, *blkp, *allocblk, allociblk[UFS_NIADDR + 1];
ufs2_daddr_t *lbns_remfree, lbns[UFS_NIADDR + 1];
int deallocated, osize, nsize, num, i, error;
int unwindidx = -1;
int saved_inbdflush;
static struct timeval lastfail;
static int curfail;
int gbflags, reclaimed;
ip = VTOI(vp);
dp = ip->i_din2;
fs = ITOFS(ip);
ump = ITOUMP(ip);
lbn = lblkno(fs, startoffset);
size = blkoff(fs, startoffset) + size;
reclaimed = 0;
if (size > fs->fs_bsize)
panic("ffs_balloc_ufs2: blk too big");
*bpp = nil;
if (lbn < 0)
return (EFBIG);
gbflags = (flags & BA_UNMAPPED) != 0 ? GB_UNMAPPED : 0;
if (DOINGSOFTDEP(vp))
softdep_prealloc(vp, MNT_WAIT);
/*
* Check for allocating external data.
*/
if (flags & IO_EXT) {
if (lbn >= UFS_NXADDR)
return (EFBIG);
/*
* If the next write will extend the data into a new block,
* and the data is currently composed of a fragment
* this fragment has to be extended to be a full block.
*/
lastlbn = lblkno(fs, dp->di_extsize);
if (lastlbn < lbn) {
nb = lastlbn;
osize = sblksize(fs, dp->di_extsize, nb);
if (osize < fs->fs_bsize && osize > 0) {
UFS_LOCK(ump);
error = ffs_realloccg(ip, -1 - nb,
dp->di_extb[nb],
ffs_blkpref_ufs2(ip, lastlbn, (int)nb,
&dp->di_extb[0]), osize,
(int)fs->fs_bsize, flags, cred, &bp);
if (error)
return (error);
if (DOINGSOFTDEP(vp))
softdep_setup_allocext(ip, nb,
dbtofsb(fs, bp->b_blkno),
dp->di_extb[nb],
fs->fs_bsize, osize, bp);
dp->di_extsize = smalllblktosize(fs, nb + 1);
dp->di_extb[nb] = dbtofsb(fs, bp->b_blkno);
bp->b_xflags |= BX_ALTDATA;
ip->i_flag |= IN_CHANGE;
if (flags & IO_SYNC)
bwrite(bp);
else
bawrite(bp);
}
}
/*
* All blocks are direct blocks
*/
if (flags & BA_METAONLY)
panic("ffs_balloc_ufs2: BA_METAONLY for ext block");
nb = dp->di_extb[lbn];
if (nb != 0 && dp->di_extsize >= smalllblktosize(fs, lbn + 1)) {
error = bread_gb(vp, -1 - lbn, fs->fs_bsize, NOCRED,
gbflags, &bp);
if (error) {
brelse(bp);
return (error);
}
bp->b_blkno = fsbtodb(fs, nb);
bp->b_xflags |= BX_ALTDATA;
*bpp = bp;
return (0);
}
if (nb != 0) {
/*
* Consider need to reallocate a fragment.
*/
osize = fragroundup(fs, blkoff(fs, dp->di_extsize));
nsize = fragroundup(fs, size);
if (nsize <= osize) {
error = bread_gb(vp, -1 - lbn, osize, NOCRED,
gbflags, &bp);
if (error) {
brelse(bp);
return (error);
}
bp->b_blkno = fsbtodb(fs, nb);
bp->b_xflags |= BX_ALTDATA;
} else {
UFS_LOCK(ump);
error = ffs_realloccg(ip, -1 - lbn,
dp->di_extb[lbn],
ffs_blkpref_ufs2(ip, lbn, (int)lbn,
&dp->di_extb[0]), osize, nsize, flags,
cred, &bp);
if (error)
return (error);
bp->b_xflags |= BX_ALTDATA;
if (DOINGSOFTDEP(vp))
softdep_setup_allocext(ip, lbn,
dbtofsb(fs, bp->b_blkno), nb,
nsize, osize, bp);
}
} else {
if (dp->di_extsize < smalllblktosize(fs, lbn + 1))
nsize = fragroundup(fs, size);
else
nsize = fs->fs_bsize;
UFS_LOCK(ump);
error = ffs_alloc(ip, lbn,
ffs_blkpref_ufs2(ip, lbn, (int)lbn, &dp->di_extb[0]),
nsize, flags, cred, &newb);
if (error)
return (error);
bp = getblk(vp, -1 - lbn, nsize, 0, 0, gbflags);
bp->b_blkno = fsbtodb(fs, newb);
bp->b_xflags |= BX_ALTDATA;
if (flags & BA_CLRBUF)
vfs_bio_clrbuf(bp);
if (DOINGSOFTDEP(vp))
softdep_setup_allocext(ip, lbn, newb, 0,
nsize, 0, bp);
}
dp->di_extb[lbn] = dbtofsb(fs, bp->b_blkno);
ip->i_flag |= IN_CHANGE;
*bpp = bp;
return (0);
}
/*
* If the next write will extend the file into a new block,
* and the file is currently composed of a fragment
* this fragment has to be extended to be a full block.
*/
lastlbn = lblkno(fs, ip->i_size);
if (lastlbn < UFS_NDADDR && lastlbn < lbn) {
nb = lastlbn;
osize = blksize(fs, ip, nb);
if (osize < fs->fs_bsize && osize > 0) {
UFS_LOCK(ump);
error = ffs_realloccg(ip, nb, dp->di_db[nb],
ffs_blkpref_ufs2(ip, lastlbn, (int)nb,
&dp->di_db[0]), osize, (int)fs->fs_bsize,
flags, cred, &bp);
if (error)
return (error);
if (DOINGSOFTDEP(vp))
softdep_setup_allocdirect(ip, nb,
dbtofsb(fs, bp->b_blkno),
dp->di_db[nb],
fs->fs_bsize, osize, bp);
ip->i_size = smalllblktosize(fs, nb + 1);
dp->di_size = ip->i_size;
dp->di_db[nb] = dbtofsb(fs, bp->b_blkno);
ip->i_flag |= IN_CHANGE | IN_UPDATE;
if (flags & IO_SYNC)
bwrite(bp);
else
bawrite(bp);
}
}
/*
* The first UFS_NDADDR blocks are direct blocks
*/
if (lbn < UFS_NDADDR) {
if (flags & BA_METAONLY)
panic("ffs_balloc_ufs2: BA_METAONLY for direct block");
nb = dp->di_db[lbn];
if (nb != 0 && ip->i_size >= smalllblktosize(fs, lbn + 1)) {
error = bread_gb(vp, lbn, fs->fs_bsize, NOCRED,
gbflags, &bp);
if (error) {
brelse(bp);
return (error);
}
bp->b_blkno = fsbtodb(fs, nb);
*bpp = bp;
return (0);
}
if (nb != 0) {
/*
* Consider need to reallocate a fragment.
*/
osize = fragroundup(fs, blkoff(fs, ip->i_size));
nsize = fragroundup(fs, size);
if (nsize <= osize) {
error = bread_gb(vp, lbn, osize, NOCRED,
gbflags, &bp);
if (error) {
brelse(bp);
return (error);
}
bp->b_blkno = fsbtodb(fs, nb);
} else {
UFS_LOCK(ump);
error = ffs_realloccg(ip, lbn, dp->di_db[lbn],
ffs_blkpref_ufs2(ip, lbn, (int)lbn,
&dp->di_db[0]), osize, nsize, flags,
cred, &bp);
if (error)
return (error);
if (DOINGSOFTDEP(vp))
softdep_setup_allocdirect(ip, lbn,
dbtofsb(fs, bp->b_blkno), nb,
nsize, osize, bp);
}
} else {
if (ip->i_size < smalllblktosize(fs, lbn + 1))
nsize = fragroundup(fs, size);
else
nsize = fs->fs_bsize;
UFS_LOCK(ump);
error = ffs_alloc(ip, lbn,
ffs_blkpref_ufs2(ip, lbn, (int)lbn,
&dp->di_db[0]), nsize, flags, cred, &newb);
if (error)
return (error);
bp = getblk(vp, lbn, nsize, 0, 0, gbflags);
bp->b_blkno = fsbtodb(fs, newb);
if (flags & BA_CLRBUF)
vfs_bio_clrbuf(bp);
if (DOINGSOFTDEP(vp))
softdep_setup_allocdirect(ip, lbn, newb, 0,
nsize, 0, bp);
}
dp->di_db[lbn] = dbtofsb(fs, bp->b_blkno);
ip->i_flag |= IN_CHANGE | IN_UPDATE;
*bpp = bp;
return (0);
}
/*
* Determine the number of levels of indirection.
*/
pref = 0;
if ((error = ufs_getlbns(vp, lbn, indirs, &num)) != 0)
return(error);
#ifdef INVARIANTS
if (num < 1)
panic ("ffs_balloc_ufs2: ufs_getlbns returned indirect block");
#endif
saved_inbdflush = curthread_pflags_set(TDP_INBDFLUSH);
/*
* Fetch the first indirect block allocating if necessary.
*/
--num;
nb = dp->di_ib[indirs[0].in_off];
allocib = nil;
allocblk = allociblk;
lbns_remfree = lbns;
if (nb == 0) {
UFS_LOCK(ump);
pref = ffs_blkpref_ufs2(ip, lbn, -indirs[0].in_off - 1,
(ufs2_daddr_t *)0);
if ((error = ffs_alloc(ip, lbn, pref, (int)fs->fs_bsize,
flags, cred, &newb)) != 0) {
curthread_pflags_restore(saved_inbdflush);
return (error);
}
pref = newb + fs->fs_frag;
nb = newb;
MPASS(allocblk < allociblk + nitems(allociblk));
MPASS(lbns_remfree < lbns + nitems(lbns));
*allocblk++ = nb;
*lbns_remfree++ = indirs[1].in_lbn;
bp = getblk(vp, indirs[1].in_lbn, fs->fs_bsize, 0, 0,
GB_UNMAPPED);
bp->b_blkno = fsbtodb(fs, nb);
vfs_bio_clrbuf(bp);
if (DOINGSOFTDEP(vp)) {
softdep_setup_allocdirect(ip,
UFS_NDADDR + indirs[0].in_off, newb, 0,
fs->fs_bsize, 0, bp);
bdwrite(bp);
} else if ((flags & IO_SYNC) == 0 && DOINGASYNC(vp)) {
if (bp->b_bufsize == fs->fs_bsize)
bp->b_flags |= B_CLUSTEROK;
bdwrite(bp);
} else {
if ((error = bwrite(bp)) != 0)
goto fail;
}
allocib = &dp->di_ib[indirs[0].in_off];
*allocib = nb;
ip->i_flag |= IN_CHANGE | IN_UPDATE;
}
/*
* Fetch through the indirect blocks, allocating as necessary.
*/
retry:
for (i = 1;;) {
error = bread(vp,
indirs[i].in_lbn, (int)fs->fs_bsize, NOCRED, &bp);
if (error) {
brelse(bp);
goto fail;
}
bap = (ufs2_daddr_t *)bp->b_data;
nb = bap[indirs[i].in_off];
if (i == num)
break;
i += 1;
if (nb != 0) {
bqrelse(bp);
continue;
}
UFS_LOCK(ump);
/*
* If parent indirect has just been allocated, try to cluster
* immediately following it.
*/
if (pref == 0)
pref = ffs_blkpref_ufs2(ip, lbn, i - num - 1,
(ufs2_daddr_t *)0);
if ((error = ffs_alloc(ip, lbn, pref, (int)fs->fs_bsize,
flags | IO_BUFLOCKED, cred, &newb)) != 0) {
brelse(bp);
if (DOINGSOFTDEP(vp) && ++reclaimed == 1) {
UFS_LOCK(ump);
softdep_request_cleanup(fs, vp, cred,
FLUSH_BLOCKS_WAIT);
UFS_UNLOCK(ump);
goto retry;
}
if (ppsratecheck(&lastfail, &curfail, 1)) {
ffs_fserr(fs, ip->i_number, "filesystem full");
uprintf("\n%s: write failed, filesystem "
"is full\n", fs->fs_fsmnt);
}
goto fail;
}
pref = newb + fs->fs_frag;
nb = newb;
MPASS(allocblk < allociblk + nitems(allociblk));
MPASS(lbns_remfree < lbns + nitems(lbns));
*allocblk++ = nb;
*lbns_remfree++ = indirs[i].in_lbn;
nbp = getblk(vp, indirs[i].in_lbn, fs->fs_bsize, 0, 0,
GB_UNMAPPED);
nbp->b_blkno = fsbtodb(fs, nb);
vfs_bio_clrbuf(nbp);
if (DOINGSOFTDEP(vp)) {
softdep_setup_allocindir_meta(nbp, ip, bp,
indirs[i - 1].in_off, nb);
bdwrite(nbp);
} else if ((flags & IO_SYNC) == 0 && DOINGASYNC(vp)) {
if (nbp->b_bufsize == fs->fs_bsize)
nbp->b_flags |= B_CLUSTEROK;
bdwrite(nbp);
} else {
if ((error = bwrite(nbp)) != 0) {
brelse(bp);
goto fail;
}
}
bap[indirs[i - 1].in_off] = nb;
if (allocib == nil && unwindidx < 0)
unwindidx = i - 1;
/*
* If required, write synchronously, otherwise use
* delayed write.
*/
if (flags & IO_SYNC) {
bwrite(bp);
} else {
if (bp->b_bufsize == fs->fs_bsize)
bp->b_flags |= B_CLUSTEROK;
bdwrite(bp);
}
}
/*
* If asked only for the indirect block, then return it.
*/
if (flags & BA_METAONLY) {
curthread_pflags_restore(saved_inbdflush);
*bpp = bp;
return (0);
}
/*
* Get the data block, allocating if necessary.
*/
if (nb == 0) {
UFS_LOCK(ump);
/*
* If allocating metadata at the front of the cylinder
* group and parent indirect block has just been allocated,
* then cluster next to it if it is the first indirect in
* the file. Otherwise it has been allocated in the metadata
* area, so we want to find our own place out in the data area.
*/
if (pref == 0 || (lbn > UFS_NDADDR && fs->fs_metaspace != 0))
pref = ffs_blkpref_ufs2(ip, lbn, indirs[i].in_off,
&bap[0]);
error = ffs_alloc(ip, lbn, pref, (int)fs->fs_bsize,
flags | IO_BUFLOCKED, cred, &newb);
if (error) {
brelse(bp);
if (DOINGSOFTDEP(vp) && ++reclaimed == 1) {
UFS_LOCK(ump);
softdep_request_cleanup(fs, vp, cred,
FLUSH_BLOCKS_WAIT);
UFS_UNLOCK(ump);
goto retry;
}
if (ppsratecheck(&lastfail, &curfail, 1)) {
ffs_fserr(fs, ip->i_number, "filesystem full");
uprintf("\n%s: write failed, filesystem "
"is full\n", fs->fs_fsmnt);
}
goto fail;
}
nb = newb;
MPASS(allocblk < allociblk + nitems(allociblk));
MPASS(lbns_remfree < lbns + nitems(lbns));
*allocblk++ = nb;
*lbns_remfree++ = lbn;
nbp = getblk(vp, lbn, fs->fs_bsize, 0, 0, gbflags);
nbp->b_blkno = fsbtodb(fs, nb);
if (flags & BA_CLRBUF)
vfs_bio_clrbuf(nbp);
if (DOINGSOFTDEP(vp))
softdep_setup_allocindir_page(ip, lbn, bp,
indirs[i].in_off, nb, 0, nbp);
bap[indirs[i].in_off] = nb;
/*
* If required, write synchronously, otherwise use
* delayed write.
*/
if (flags & IO_SYNC) {
bwrite(bp);
} else {
if (bp->b_bufsize == fs->fs_bsize)
bp->b_flags |= B_CLUSTEROK;
bdwrite(bp);
}
curthread_pflags_restore(saved_inbdflush);
*bpp = nbp;
return (0);
}
brelse(bp);
/*
* If requested clear invalid portions of the buffer. If we
* have to do a read-before-write (typical if BA_CLRBUF is set),
* try to do some read-ahead in the sequential case to reduce
* the number of I/O transactions.
*/
if (flags & BA_CLRBUF) {
int seqcount = (flags & BA_SEQMASK) >> BA_SEQSHIFT;
if (seqcount != 0 &&
(vp->v_mount->mnt_flag & MNT_NOCLUSTERR) == 0 &&
!(vm_page_count_severe() || buf_dirty_count_severe())) {
error = cluster_read(vp, ip->i_size, lbn,
(int)fs->fs_bsize, NOCRED,
MAXBSIZE, seqcount, gbflags, &nbp);
} else {
error = bread_gb(vp, lbn, (int)fs->fs_bsize,
NOCRED, gbflags, &nbp);
}
if (error) {
brelse(nbp);
goto fail;
}
} else {
/*
* This is now called from local media FS's to operate against their
* own vnodes if they fail to implement VOP_PUTPAGES.
*
* This is typically called indirectly via the pageout daemon and
* clustering has already typically occurred, so in general we ask the
* underlying filesystem to write the data out asynchronously rather
* then delayed.
*/
int
vnode_pager_generic_putpages(struct vnode *vp, vm_page_t *ma, int bytecount,
int flags, int *rtvals)
{
vm_object_t object;
vm_page_t m;
vm_ooffset_t poffset;
struct uio auio;
struct iovec aiov;
int count, error, i, maxsize, ncount, pgoff, ppscheck;
static struct timeval lastfail;
static int curfail;
object = vp->v_object;
count = bytecount / PAGE_SIZE;
for (i = 0; i < count; i++)
rtvals[i] = VM_PAGER_ERROR;
if ((int64_t)ma[0]->pindex < 0) {
printf("vnode_pager_generic_putpages: "
"attempt to write meta-data 0x%jx(%lx)\n",
(uintmax_t)ma[0]->pindex, (u_long)ma[0]->dirty);
rtvals[0] = VM_PAGER_BAD;
return (VM_PAGER_BAD);
}
maxsize = count * PAGE_SIZE;
ncount = count;
poffset = IDX_TO_OFF(ma[0]->pindex);
/*
* If the page-aligned write is larger then the actual file we
* have to invalidate pages occurring beyond the file EOF. However,
* there is an edge case where a file may not be page-aligned where
* the last page is partially invalid. In this case the filesystem
* may not properly clear the dirty bits for the entire page (which
* could be VM_PAGE_BITS_ALL due to the page having been mmap()d).
* With the page locked we are free to fix-up the dirty bits here.
*
* We do not under any circumstances truncate the valid bits, as
* this will screw up bogus page replacement.
*/
VM_OBJECT_WLOCK(object);
if (maxsize + poffset > object->un_pager.vnp.vnp_size) {
if (object->un_pager.vnp.vnp_size > poffset) {
maxsize = object->un_pager.vnp.vnp_size - poffset;
ncount = btoc(maxsize);
if ((pgoff = (int)maxsize & PAGE_MASK) != 0) {
/*
* If the object is locked and the following
* conditions hold, then the page's dirty
* field cannot be concurrently changed by a
* pmap operation.
*/
m = ma[ncount - 1];
vm_page_assert_sbusied(m);
KASSERT(!pmap_page_is_write_mapped(m),
("vnode_pager_generic_putpages: page %p is not read-only", m));
MPASS(m->dirty != 0);
vm_page_clear_dirty(m, pgoff, PAGE_SIZE -
pgoff);
}
} else {
maxsize = 0;
ncount = 0;
}
for (i = ncount; i < count; i++)
rtvals[i] = VM_PAGER_BAD;
}
for (i = 0; i < ncount - ((btoc(maxsize) & PAGE_MASK) != 0); i++)
MPASS(ma[i]->dirty == VM_PAGE_BITS_ALL);
VM_OBJECT_WUNLOCK(object);
aiov.iov_base = NULL;
aiov.iov_len = maxsize;
auio.uio_iov = &aiov;
auio.uio_iovcnt = 1;
auio.uio_offset = poffset;
auio.uio_segflg = UIO_NOCOPY;
auio.uio_rw = UIO_WRITE;
auio.uio_resid = maxsize;
auio.uio_td = NULL;
error = VOP_WRITE(vp, &auio, vnode_pager_putpages_ioflags(flags),
curthread->td_ucred);
VM_CNT_INC(v_vnodeout);
VM_CNT_ADD(v_vnodepgsout, ncount);
ppscheck = 0;
if (error != 0 && (ppscheck = ppsratecheck(&lastfail, &curfail, 1))
!= 0)
printf("vnode_pager_putpages: I/O error %d\n", error);
if (auio.uio_resid != 0 && (ppscheck != 0 ||
ppsratecheck(&lastfail, &curfail, 1) != 0))
printf("vnode_pager_putpages: residual I/O %zd at %ju\n",
auio.uio_resid, (uintmax_t)ma[0]->pindex);
for (i = 0; i < ncount; i++)
rtvals[i] = VM_PAGER_OK;
return (rtvals[0]);
}
/*
* Handle a radar detection event on a channel. The channel is
* added to the NOL list and we record the time of the event.
* Entries are aged out after NOL_TIMEOUT. If radar was
* detected while doing CAC we force a state/channel change.
* Otherwise radar triggers a channel switch using the CSA
* mechanism (when the channel is the bss channel).
*/
void
ieee80211_dfs_notify_radar(struct ieee80211com *ic, struct ieee80211_channel *chan)
{
struct ieee80211_dfs_state *dfs = &ic->ic_dfs;
int i, now;
/*
* Mark all entries with this frequency. Notify user
* space and arrange for notification when the radar
* indication is cleared. Then kick the NOL processing
* thread if not already running.
*/
now = ticks;
for (i = 0; i < ic->ic_nchans; i++) {
struct ieee80211_channel *c = &ic->ic_channels[i];
if (c->ic_freq == chan->ic_freq) {
c->ic_state &= ~IEEE80211_CHANSTATE_CACDONE;
c->ic_state |= IEEE80211_CHANSTATE_RADAR;
dfs->nol_event[i] = now;
}
}
ieee80211_notify_radar(ic, chan);
chan->ic_state |= IEEE80211_CHANSTATE_NORADAR;
if (!callout_pending(&dfs->nol_timer)) {
callout_reset(&dfs->nol_timer, NOL_TIMEOUT,
dfs_timeout_callout, ic);
}
/*
* If radar is detected on the bss channel while
* doing CAC; force a state change by scheduling the
* callout to be dispatched asap. Otherwise, if this
* event is for the bss channel then we must quiet
* traffic and schedule a channel switch.
*
* Note this allows us to receive notification about
* channels other than the bss channel; not sure
* that can/will happen but it's simple to support.
*/
if (chan == ic->ic_bsschan) {
/* XXX need a way to defer to user app */
dfs->newchan = ieee80211_dfs_pickchannel(ic);
announce_radar(ic->ic_ifp, chan, dfs->newchan);
#ifdef notyet
if (callout_pending(&dfs->cac_timer)) {
callout_reset(&dfs->cac_timer, 0,
cac_timeout_callout, vap);
}
else if (dfs->newchan != NULL) {
/* XXX mode 1, switch count 2 */
/* XXX calculate switch count based on max
switch time and beacon interval? */
ieee80211_csa_startswitch(ic, dfs->newchan, 1, 2);
} else {
/*
* Spec says to stop all transmissions and
* wait on the current channel for an entry
* on the NOL to expire.
*/
/*XXX*/
}
#endif
} else {
/*
* Issue rate-limited console msgs.
*/
if (dfs->lastchan != chan) {
dfs->lastchan = chan;
dfs->cureps = 0;
announce_radar(ic->ic_ifp, chan, NULL);
} else if (ppsratecheck(&dfs->lastevent, &dfs->cureps, 1)) {
announce_radar(ic->ic_ifp, chan, NULL);
}
}
}
