upstream/opnsense-src
1
0
Fork 0
mirror of https://github.com/opnsense/src.git synced 2026-06-09 00:32:25 -04:00
Code Issues Projects Releases Packages Wiki Activity Actions

Changes to make the NFS client MP safe.

Browse source 
Thanks to Kris Kennaway for testing and sending lots of bugs my way.
This commit is contained in:
Mohan Srinivasan 2006-05-19 00:04:24 +00:00
parent 067eb35dd0
commit f1cdf89911
11 changed files with 923 additions and 450 deletions
Show stats Download patch file Download diff file Expand all files Collapse all files

4
sys/nfs4client/nfs4_vfsops.c
View file

@ -563,6 +563,8 @@ mountnfs(struct nfs_args *argp, struct mount *mp, struct sockaddr *nam,
vfs_getnewfsid(mp);
nmp->nm_mountp = mp;
mtx_init(&nmp->nm_mtx, "NFS4mount lock", NULL, MTX_DEF);
nmp->nm_maxfilesize = 0xffffffffLL;
nmp->nm_timeo = NFS_TIMEO;
nmp->nm_retry = NFS_RETRANS;
@ -652,6 +654,7 @@ mountnfs(struct nfs_args *argp, struct mount *mp, struct sockaddr *nam,
if (mrep != NULL)
m_freem(mrep);
bad:
mtx_destroy(&nmp->nm_mtx);
nfs4_disconnect(nmp);
uma_zfree(nfsmount_zone, nmp);
FREE(nam, M_SONAME);
@ -698,6 +701,7 @@ nfs_unmount(struct mount *mp, int mntflags, struct thread *td)
/* XXX there's a race condition here for SMP */
wakeup(&nfs4_daemonproc);
mtx_destroy(&nmp->nm_mtx);
uma_zfree(nfsmount_zone, nmp);
return (0);
}

4
sys/nfsclient/nfs.h
View file

@ -131,6 +131,7 @@ extern struct uma_zone *nfsmount_zone;
extern struct callout nfs_callout;
extern struct nfsstats nfsstats;
extern struct mtx nfs_iod_mtx;
extern int nfs_numasync;
extern unsigned int nfs_iodmax;
@ -178,6 +179,7 @@ struct nfsreq {
int r_rtt; /* RTT for rpc */
int r_lastmsg; /* last tprintf */
struct thread *r_td; /* Proc that did I/O system call */
struct mtx r_mtx; /* Protects nfsreq fields */
};
/*
@ -310,8 +312,6 @@ int nfs_meta_setsize (struct vnode *, struct ucred *,
void nfs_set_sigmask __P((struct thread *td, sigset_t *oldset));
void nfs_restore_sigmask __P((struct thread *td, sigset_t *set));
int nfs_tsleep __P((struct thread *td, void *ident, int priority, char *wmesg,
int timo));
int nfs_msleep __P((struct thread *td, void *ident, struct mtx *mtx, int priority,
char *wmesg, int timo));

330
sys/nfsclient/nfs_bio.c
View file

@ -71,6 +71,7 @@ static int nfs_directio_write(struct vnode *vp, struct uio *uiop,
extern int nfs_directio_enable;
extern int nfs_directio_allow_mmap;
/*
* Vnode op for VM getpages.
*/
@ -90,8 +91,6 @@ nfs_getpages(struct vop_getpages_args *ap)
vm_page_t *pages;
struct nfsnode *np;
GIANT_REQUIRED;
vp = ap->a_vp;
np = VTONFS(vp);
td = curthread; /* XXX */
@ -101,22 +100,28 @@ nfs_getpages(struct vop_getpages_args *ap)
count = ap->a_count;
if ((object = vp->v_object) == NULL) {
printf("nfs_getpages: called with non-merged cache vnode??\n");
nfs_printf("nfs_getpages: called with non-merged cache vnode??\n");
return VM_PAGER_ERROR;
}
if (nfs_directio_enable && !nfs_directio_allow_mmap &&
(np->n_flag & NNONCACHE) &&
(vp->v_type == VREG)) {
printf("nfs_getpages: called on non-cacheable vnode??\n");
return VM_PAGER_ERROR;
if (nfs_directio_enable && !nfs_directio_allow_mmap) {
mtx_lock(&np->n_mtx);
if ((np->n_flag & NNONCACHE) && (vp->v_type == VREG)) {
mtx_unlock(&np->n_mtx);
nfs_printf("nfs_getpages: called on non-cacheable vnode??\n");
return VM_PAGER_ERROR;
} else
mtx_unlock(&np->n_mtx);
}
mtx_lock(&nmp->nm_mtx);
if ((nmp->nm_flag & NFSMNT_NFSV3) != 0 &&
(nmp->nm_state & NFSSTA_GOTFSINFO) == 0) {
(nmp->nm_state & NFSSTA_GOTFSINFO) == 0) {
mtx_unlock(&nmp->nm_mtx);
/* We'll never get here for v4, because we always have fsinfo */
(void)nfs_fsinfo(nmp, vp, cred, td);
}
} else
mtx_unlock(&nmp->nm_mtx);
npages = btoc(count);
@ -173,7 +178,7 @@ nfs_getpages(struct vop_getpages_args *ap)
relpbuf(bp, &nfs_pbuf_freecnt);
if (error && (uio.uio_resid == count)) {
printf("nfs_getpages: error %d\n", error);
nfs_printf("nfs_getpages: error %d\n", error);
VM_OBJECT_LOCK(object);
vm_page_lock_queues();
for (i = 0; i < npages; ++i) {
@ -270,8 +275,6 @@ nfs_putpages(struct vop_putpages_args *ap)
struct nfsnode *np;
vm_page_t *pages;
GIANT_REQUIRED;
vp = ap->a_vp;
np = VTONFS(vp);
td = curthread; /* XXX */
@ -282,15 +285,22 @@ nfs_putpages(struct vop_putpages_args *ap)
rtvals = ap->a_rtvals;
npages = btoc(count);
offset = IDX_TO_OFF(pages[0]->pindex);
mtx_lock(&nmp->nm_mtx);
if ((nmp->nm_flag & NFSMNT_NFSV3) != 0 &&
(nmp->nm_state & NFSSTA_GOTFSINFO) == 0) {
mtx_unlock(&nmp->nm_mtx);
(void)nfs_fsinfo(nmp, vp, cred, td);
}
} else
mtx_unlock(&nmp->nm_mtx);
mtx_lock(&np->n_mtx);
if (nfs_directio_enable && !nfs_directio_allow_mmap &&
(np->n_flag & NNONCACHE) && (vp->v_type == VREG))
printf("nfs_putpages: called on noncache-able vnode??\n");
(np->n_flag & NNONCACHE) && (vp->v_type == VREG)) {
mtx_unlock(&np->n_mtx);
nfs_printf("nfs_putpages: called on noncache-able vnode??\n");
mtx_lock(&np->n_mtx);
}
for (i = 0; i < npages; i++)
rtvals[i] = VM_PAGER_AGAIN;
@ -298,12 +308,12 @@ nfs_putpages(struct vop_putpages_args *ap)
/*
* When putting pages, do not extend file past EOF.
*/
if (offset + count > np->n_size) {
count = np->n_size - offset;
if (count < 0)
count = 0;
}
mtx_unlock(&np->n_mtx);
/*
* We use only the kva address for the buffer, but this is extremely
@ -349,6 +359,81 @@ nfs_putpages(struct vop_putpages_args *ap)
return rtvals[0];
}
/*
* For nfs, cache consistency can only be maintained approximately.
* Although RFC1094 does not specify the criteria, the following is
* believed to be compatible with the reference port.
* For nfs:
* If the file's modify time on the server has changed since the
* last read rpc or you have written to the file,
* you may have lost data cache consistency with the
* server, so flush all of the file's data out of the cache.
* Then force a getattr rpc to ensure that you have up to date
* attributes.
* NB: This implies that cache data can be read when up to
* NFS_ATTRTIMEO seconds out of date. If you find that you need current
* attributes this could be forced by setting n_attrstamp to 0 before
* the VOP_GETATTR() call.
*/
static inline int
nfs_bioread_check_cons(struct vnode *vp, struct thread *td, struct ucred *cred)
{
int error = 0;
struct vattr vattr;
struct nfsnode *np = VTONFS(vp);
int old_lock;
struct nfsmount *nmp = VFSTONFS(vp->v_mount);
/*
* Grab the exclusive lock before checking whether the cache is
* consistent.
* XXX - We can make this cheaper later (by acquiring cheaper locks).
* But for now, this suffices.
*/
old_lock = nfs_upgrade_vnlock(vp, td);
mtx_lock(&np->n_mtx);
if (np->n_flag & NMODIFIED) {
mtx_unlock(&np->n_mtx);
if (vp->v_type != VREG) {
if (vp->v_type != VDIR)
panic("nfs: bioread, not dir");
(nmp->nm_rpcops->nr_invaldir)(vp);
error = nfs_vinvalbuf(vp, V_SAVE, td, 1);
if (error)
goto out;
}
np->n_attrstamp = 0;
error = VOP_GETATTR(vp, &vattr, cred, td);
if (error)
goto out;
mtx_lock(&np->n_mtx);
np->n_mtime = vattr.va_mtime;
mtx_unlock(&np->n_mtx);
} else {
mtx_unlock(&np->n_mtx);
error = VOP_GETATTR(vp, &vattr, cred, td);
if (error)
return (error);
mtx_lock(&np->n_mtx);
if ((np->n_flag & NSIZECHANGED)
|| (NFS_TIMESPEC_COMPARE(&np->n_mtime, &vattr.va_mtime))) {
mtx_unlock(&np->n_mtx);
if (vp->v_type == VDIR)
(nmp->nm_rpcops->nr_invaldir)(vp);
error = nfs_vinvalbuf(vp, V_SAVE, td, 1);
if (error)
goto out;
mtx_lock(&np->n_mtx);
np->n_mtime = vattr.va_mtime;
np->n_flag &= ~NSIZECHANGED;
}
mtx_unlock(&np->n_mtx);
}
out:
nfs_downgrade_vnlock(vp, td, old_lock);
return error;
}
/*
* Vnode op for read using bio
*/
@ -358,7 +443,6 @@ nfs_bioread(struct vnode *vp, struct uio *uio, int ioflag, struct ucred *cred)
struct nfsnode *np = VTONFS(vp);
int biosize, i;
struct buf *bp, *rabp;
struct vattr vattr;
struct thread *td;
struct nfsmount *nmp = VFSTONFS(vp->v_mount);
daddr_t lbn, rabn;
@ -376,9 +460,14 @@ nfs_bioread(struct vnode *vp, struct uio *uio, int ioflag, struct ucred *cred)
return (EINVAL);
td = uio->uio_td;
mtx_lock(&nmp->nm_mtx);
if ((nmp->nm_flag & NFSMNT_NFSV3) != 0 &&
(nmp->nm_state & NFSSTA_GOTFSINFO) == 0)
(nmp->nm_state & NFSSTA_GOTFSINFO) == 0) {
mtx_unlock(&nmp->nm_mtx);
(void)nfs_fsinfo(nmp, vp, cred, td);
} else
mtx_unlock(&nmp->nm_mtx);
if (vp->v_type != VDIR &&
(uio->uio_offset + uio->uio_resid) > nmp->nm_maxfilesize)
return (EFBIG);
@ -389,52 +478,18 @@ nfs_bioread(struct vnode *vp, struct uio *uio, int ioflag, struct ucred *cred)
biosize = vp->v_mount->mnt_stat.f_iosize;
seqcount = (int)((off_t)(ioflag >> IO_SEQSHIFT) * biosize / BKVASIZE);
/*
* For nfs, cache consistency can only be maintained approximately.
* Although RFC1094 does not specify the criteria, the following is
* believed to be compatible with the reference port.
* For nfs:
* If the file's modify time on the server has changed since the
* last read rpc or you have written to the file,
* you may have lost data cache consistency with the
* server, so flush all of the file's data out of the cache.
* Then force a getattr rpc to ensure that you have up to date
* attributes.
* NB: This implies that cache data can be read when up to
* NFS_ATTRTIMEO seconds out of date. If you find that you need current
* attributes this could be forced by setting n_attrstamp to 0 before
* the VOP_GETATTR() call.
*/
if (np->n_flag & NMODIFIED) {
if (vp->v_type != VREG) {
if (vp->v_type != VDIR)
panic("nfs: bioread, not dir");
(nmp->nm_rpcops->nr_invaldir)(vp);
error = nfs_vinvalbuf(vp, V_SAVE, td, 1);
if (error)
return (error);
}
np->n_attrstamp = 0;
error = VOP_GETATTR(vp, &vattr, cred, td);
if (error)
return (error);
np->n_mtime = vattr.va_mtime;
} else {
error = VOP_GETATTR(vp, &vattr, cred, td);
if (error)
return (error);
if ((np->n_flag & NSIZECHANGED)
|| (NFS_TIMESPEC_COMPARE(&np->n_mtime, &vattr.va_mtime))) {
if (vp->v_type == VDIR)
(nmp->nm_rpcops->nr_invaldir)(vp);
error = nfs_vinvalbuf(vp, V_SAVE, td, 1);
if (error)
return (error);
np->n_mtime = vattr.va_mtime;
np->n_flag &= ~NSIZECHANGED;
}
}
error = nfs_bioread_check_cons(vp, td, cred);
if (error)
return error;
do {
u_quad_t nsize;
mtx_lock(&np->n_mtx);
nsize = np->n_size;
mtx_unlock(&np->n_mtx);
switch (vp->v_type) {
case VREG:
nfsstats.biocache_reads++;
@ -443,12 +498,10 @@ nfs_bioread(struct vnode *vp, struct uio *uio, int ioflag, struct ucred *cred)
/*
* Start the read ahead(s), as required.
* The readahead is kicked off only if sequential access
* is detected, based on the readahead hint (ra_expect_lbn).
*/
if (nmp->nm_readahead > 0 && np->ra_expect_lbn == lbn) {
if (nmp->nm_readahead > 0) {
for (nra = 0; nra < nmp->nm_readahead && nra < seqcount &&
(off_t)(lbn + 1 + nra) * biosize < np->n_size; nra++) {
(off_t)(lbn + 1 + nra) * biosize < nsize; nra++) {
rabn = lbn + 1 + nra;
if (incore(&vp->v_bufobj, rabn) == NULL) {
rabp = nfs_getcacheblk(vp, rabn, biosize, td);
@ -472,15 +525,14 @@ nfs_bioread(struct vnode *vp, struct uio *uio, int ioflag, struct ucred *cred)
}
}
}
np->ra_expect_lbn = lbn + 1;
}
/* Note that bcount is *not* DEV_BSIZE aligned. */
bcount = biosize;
if ((off_t)lbn * biosize >= np->n_size) {
if ((off_t)lbn * biosize >= nsize) {
bcount = 0;
} else if ((off_t)(lbn + 1) * biosize > np->n_size) {
bcount = np->n_size - (off_t)lbn * biosize;
} else if ((off_t)(lbn + 1) * biosize > nsize) {
bcount = nsize - (off_t)lbn * biosize;
}
bp = nfs_getcacheblk(vp, lbn, bcount, td);
@ -652,7 +704,7 @@ nfs_bioread(struct vnode *vp, struct uio *uio, int ioflag, struct ucred *cred)
n = np->n_direofoffset - uio->uio_offset;
break;
default:
printf(" nfs_bioread: type %x unexpected\n", vp->v_type);
nfs_printf(" nfs_bioread: type %x unexpected\n", vp->v_type);
bp = NULL;
break;
};
@ -690,14 +742,18 @@ nfs_directio_write(vp, uiop, cred, ioflag)
struct nfsmount *nmp = VFSTONFS(vp->v_mount);
struct thread *td = uiop->uio_td;
int size;
int wsize;
mtx_lock(&nmp->nm_mtx);
wsize = nmp->nm_wsize;
mtx_unlock(&nmp->nm_mtx);
if (ioflag & IO_SYNC) {
int iomode, must_commit;
struct uio uio;
struct iovec iov;
do_sync:
while (uiop->uio_resid > 0) {
size = min(uiop->uio_resid, nmp->nm_wsize);
size = min(uiop->uio_resid, wsize);
size = min(uiop->uio_iov->iov_len, size);
iov.iov_base = uiop->uio_iov->iov_base;
iov.iov_len = size;
@ -746,7 +802,7 @@ do_sync:
* in NFS directio access.
*/
while (uiop->uio_resid > 0) {
size = min(uiop->uio_resid, nmp->nm_wsize);
size = min(uiop->uio_resid, wsize);
size = min(uiop->uio_iov->iov_len, size);
bp = getpbuf(&nfs_pbuf_freecnt);
t_uio = malloc(sizeof(struct uio), M_NFSDIRECTIO, M_WAITOK);
@ -819,8 +875,6 @@ nfs_write(struct vop_write_args *ap)
int n, on, error = 0;
struct proc *p = td?td->td_proc:NULL;
GIANT_REQUIRED;
#ifdef DIAGNOSTIC
if (uio->uio_rw != UIO_WRITE)
panic("nfs_write mode");
@ -829,20 +883,29 @@ nfs_write(struct vop_write_args *ap)
#endif
if (vp->v_type != VREG)
return (EIO);
mtx_lock(&np->n_mtx);
if (np->n_flag & NWRITEERR) {
np->n_flag &= ~NWRITEERR;
mtx_unlock(&np->n_mtx);
return (np->n_error);
}
} else
mtx_unlock(&np->n_mtx);
mtx_lock(&nmp->nm_mtx);
if ((nmp->nm_flag & NFSMNT_NFSV3) != 0 &&
(nmp->nm_state & NFSSTA_GOTFSINFO) == 0)
(nmp->nm_state & NFSSTA_GOTFSINFO) == 0) {
mtx_unlock(&nmp->nm_mtx);
(void)nfs_fsinfo(nmp, vp, cred, td);
} else
mtx_unlock(&nmp->nm_mtx);
/*
* Synchronously flush pending buffers if we are in synchronous
* mode or if we are appending.
*/
if (ioflag & (IO_APPEND | IO_SYNC)) {
mtx_lock(&np->n_mtx);
if (np->n_flag & NMODIFIED) {
mtx_unlock(&np->n_mtx);
#ifdef notyet /* Needs matching nonblock semantics elsewhere, too. */
/*
* Require non-blocking, synchronous writes to
@ -857,7 +920,8 @@ flush_and_restart:
error = nfs_vinvalbuf(vp, V_SAVE, td, 1);
if (error)
return (error);
}
} else
mtx_unlock(&np->n_mtx);
}
/*
@ -869,7 +933,9 @@ flush_and_restart:
error = VOP_GETATTR(vp, &vattr, cred, td);
if (error)
return (error);
mtx_lock(&np->n_mtx);
uio->uio_offset = np->n_size;
mtx_unlock(&np->n_mtx);
}
if (uio->uio_offset < 0)
@ -907,6 +973,11 @@ flush_and_restart:
* no point optimizing for something that really won't ever happen.
*/
if (!(ioflag & IO_SYNC)) {
int nflag;
mtx_lock(&np->n_mtx);
nflag = np->n_flag;
mtx_unlock(&np->n_mtx);
int needrestart = 0;
if (nmp->nm_wcommitsize < uio->uio_resid) {
/*
@ -918,9 +989,9 @@ flush_and_restart:
if (ioflag & IO_NDELAY)
return (EAGAIN);
ioflag |= IO_SYNC;
if (np->n_flag & NMODIFIED)
if (nflag & NMODIFIED)
needrestart = 1;
} else if (np->n_flag & NMODIFIED) {
} else if (nflag & NMODIFIED) {
int wouldcommit = 0;
BO_LOCK(&vp->v_bufobj);
if (vp->v_bufobj.bo_dirty.bv_cnt != 0) {
@ -961,8 +1032,9 @@ again:
* Handle direct append and file extension cases, calculate
* unaligned buffer size.
*/
mtx_lock(&np->n_mtx);
if (uio->uio_offset == np->n_size && n) {
mtx_unlock(&np->n_mtx);
/*
* Get the buffer (in its pre-append state to maintain
* B_CACHE if it was previously set). Resize the
@ -975,9 +1047,11 @@ again:
if (bp != NULL) {
long save;
mtx_lock(&np->n_mtx);
np->n_size = uio->uio_offset + n;
np->n_flag |= NMODIFIED;
vnode_pager_setsize(vp, np->n_size);
mtx_unlock(&np->n_mtx);
save = bp->b_flags & B_CACHE;
bcount += n;
@ -996,12 +1070,15 @@ again:
else
bcount = np->n_size - (off_t)lbn * biosize;
}
mtx_unlock(&np->n_mtx);
bp = nfs_getcacheblk(vp, lbn, bcount, td);
mtx_lock(&np->n_mtx);
if (uio->uio_offset + n > np->n_size) {
np->n_size = uio->uio_offset + n;
np->n_flag |= NMODIFIED;
vnode_pager_setsize(vp, np->n_size);
}
mtx_unlock(&np->n_mtx);
}
if (!bp) {
@ -1047,7 +1124,9 @@ again:
}
if (bp->b_wcred == NOCRED)
bp->b_wcred = crhold(cred);
mtx_lock(&np->n_mtx);
np->n_flag |= NMODIFIED;
mtx_unlock(&np->n_mtx);
/*
* If dirtyend exceeds file size, chop it down. This should
@ -1059,7 +1138,7 @@ again:
*/
if (bp->b_dirtyend > bcount) {
printf("NFS append race @%lx:%d\n",
nfs_printf("NFS append race @%lx:%d\n",
(long)bp->b_blkno * DEV_BSIZE,
bp->b_dirtyend - bcount);
bp->b_dirtyend = bcount;
@ -1139,7 +1218,7 @@ again:
break;
} else if ((n + on) == biosize) {
bp->b_flags |= B_ASYNC;
(void) (nmp->nm_rpcops->nr_writebp)(bp, 0, 0);
(void) (nmp->nm_rpcops->nr_writebp)(bp, 0, NULL);
} else {
bdwrite(bp);
}
@ -1229,15 +1308,7 @@ nfs_vinvalbuf(struct vnode *vp, int flags, struct thread *td, int intrflg)
slptimeo = 0;
}
if ((old_lock = VOP_ISLOCKED(vp, td)) != LK_EXCLUSIVE) {
if (old_lock == LK_SHARED) {
/* Upgrade to exclusive lock, this might block */
vn_lock(vp, LK_UPGRADE | LK_RETRY, td);
} else {
vn_lock(vp, LK_EXCLUSIVE | LK_RETRY, td);
}
}
old_lock = nfs_upgrade_vnlock(vp, td);
/*
* Now, flush as required.
*/
@ -1247,17 +1318,12 @@ nfs_vinvalbuf(struct vnode *vp, int flags, struct thread *td, int intrflg)
goto out;
error = vinvalbuf(vp, flags, td, 0, slptimeo);
}
mtx_lock(&np->n_mtx);
if (np->n_directio_asyncwr == 0)
np->n_flag &= ~NMODIFIED;
mtx_unlock(&np->n_mtx);
out:
if (old_lock != LK_EXCLUSIVE) {
if (old_lock == LK_SHARED) {
/* Downgrade from exclusive lock, this might block */
vn_lock(vp, LK_DOWNGRADE, td);
} else {
VOP_UNLOCK(vp, 0, td);
}
}
nfs_downgrade_vnlock(vp, td, old_lock);
return error;
}
@ -1283,11 +1349,12 @@ nfs_asyncio(struct nfsmount *nmp, struct buf *bp, struct ucred *cred, struct thr
* leave the async daemons for more important rpc's (such as reads
* and writes).
*/
mtx_lock(&nfs_iod_mtx);
if (bp->b_iocmd == BIO_WRITE && (bp->b_flags & B_NEEDCOMMIT) &&
(nmp->nm_bufqiods > nfs_numasync / 2)) {
mtx_unlock(&nfs_iod_mtx);
return(EIO);
}
again:
if (nmp->nm_flag & NFSMNT_INT)
slpflag = PCATCH;
@ -1350,12 +1417,15 @@ again:
NFS_DPF(ASYNCIO,
("nfs_asyncio: waiting for mount %p queue to drain\n", nmp));
nmp->nm_bufqwant = TRUE;
error = nfs_tsleep(td, &nmp->nm_bufq, slpflag | PRIBIO,
error = nfs_msleep(td, &nmp->nm_bufq, &nfs_iod_mtx,
slpflag | PRIBIO,
"nfsaio", slptimeo);
if (error) {
error2 = nfs_sigintr(nmp, NULL, td);
if (error2)
if (error2) {
mtx_unlock(&nfs_iod_mtx);
return (error2);
}
if (slpflag == PCATCH) {
slpflag = 0;
slptimeo = 2 * hz;
@ -1385,11 +1455,17 @@ again:
BUF_KERNPROC(bp);
TAILQ_INSERT_TAIL(&nmp->nm_bufq, bp, b_freelist);
nmp->nm_bufqlen++;
if ((bp->b_flags & B_DIRECT) && bp->b_iocmd == BIO_WRITE)
if ((bp->b_flags & B_DIRECT) && bp->b_iocmd == BIO_WRITE) {
mtx_lock(&(VTONFS(bp->b_vp))->n_mtx);
VTONFS(bp->b_vp)->n_directio_asyncwr++;
mtx_unlock(&(VTONFS(bp->b_vp))->n_mtx);
}
mtx_unlock(&nfs_iod_mtx);
return (0);
}
mtx_unlock(&nfs_iod_mtx);
/*
* All the iods are busy on other mounts, so return EIO to
* force the caller to process the i/o synchronously.
@ -1415,12 +1491,13 @@ nfs_doio_directwrite(struct buf *bp)
free(uiop, M_NFSDIRECTIO);
if ((bp->b_flags & B_DIRECT) && bp->b_iocmd == BIO_WRITE) {
struct nfsnode *np = VTONFS(bp->b_vp);
mtx_lock(&np->n_mtx);
np->n_directio_asyncwr--;
if ((np->n_flag & NFSYNCWAIT) && np->n_directio_asyncwr == 0) {
np->n_flag &= ~NFSYNCWAIT;
wakeup((caddr_t)&np->n_directio_asyncwr);
}
mtx_unlock(&np->n_mtx);
}
vdrop(bp->b_vp);
bp->b_vp = NULL;
@ -1441,7 +1518,8 @@ nfs_doio(struct vnode *vp, struct buf *bp, struct ucred *cr, struct thread *td)
struct uio uio;
struct iovec io;
struct proc *p = td ? td->td_proc : NULL;
uint8_t iocmd;
np = VTONFS(vp);
nmp = VFSTONFS(vp->v_mount);
uiop = &uio;
@ -1459,8 +1537,8 @@ nfs_doio(struct vnode *vp, struct buf *bp, struct ucred *cr, struct thread *td)
bp->b_ioflags &= ~BIO_ERROR;
KASSERT(!(bp->b_flags & B_DONE), ("nfs_doio: bp %p already marked done", bp));
if (bp->b_iocmd == BIO_READ) {
iocmd = bp->b_iocmd;
if (iocmd == BIO_READ) {
io.iov_len = uiop->uio_resid = bp->b_bcount;
io.iov_base = bp->b_data;
uiop->uio_rw = UIO_READ;
@ -1490,11 +1568,15 @@ nfs_doio(struct vnode *vp, struct buf *bp, struct ucred *cr, struct thread *td)
}
}
/* ASSERT_VOP_LOCKED(vp, "nfs_doio"); */
if (p && (vp->v_vflag & VV_TEXT) &&
(NFS_TIMESPEC_COMPARE(&np->n_mtime, &np->n_vattr.va_mtime))) {
PROC_LOCK(p);
killproc(p, "text file modification");
PROC_UNLOCK(p);
if (p && (vp->v_vflag & VV_TEXT)) {
mtx_lock(&np->n_mtx);
if (NFS_TIMESPEC_COMPARE(&np->n_mtime, &np->n_vattr.va_mtime)) {
mtx_unlock(&np->n_mtx);
PROC_LOCK(p);
killproc(p, "text file modification");
PROC_UNLOCK(p);
} else
mtx_unlock(&np->n_mtx);
}
break;
case VLNK:
@ -1524,7 +1606,7 @@ nfs_doio(struct vnode *vp, struct buf *bp, struct ucred *cr, struct thread *td)
bp->b_flags |= B_INVAL;
break;
default:
printf("nfs_doio: type %x unexpected\n", vp->v_type);
nfs_printf("nfs_doio: type %x unexpected\n", vp->v_type);
break;
};
if (error) {
@ -1558,9 +1640,10 @@ nfs_doio(struct vnode *vp, struct buf *bp, struct ucred *cr, struct thread *td)
/*
* Setup for actual write
*/
mtx_lock(&np->n_mtx);
if ((off_t)bp->b_blkno * DEV_BSIZE + bp->b_dirtyend > np->n_size)
bp->b_dirtyend = np->n_size - (off_t)bp->b_blkno * DEV_BSIZE;
mtx_unlock(&np->n_mtx);
if (bp->b_dirtyend > bp->b_dirtyoff) {
io.iov_len = uiop->uio_resid = bp->b_dirtyend
@ -1635,7 +1718,9 @@ nfs_doio(struct vnode *vp, struct buf *bp, struct ucred *cr, struct thread *td)
if (error) {
bp->b_ioflags |= BIO_ERROR;
bp->b_error = np->n_error = error;
mtx_lock(&np->n_mtx);
np->n_flag |= NWRITEERR;
mtx_unlock(&np->n_mtx);
}
bp->b_dirtyoff = bp->b_dirtyend = 0;
}
@ -1664,13 +1749,16 @@ int
nfs_meta_setsize(struct vnode *vp, struct ucred *cred, struct thread *td, u_quad_t nsize)
{
struct nfsnode *np = VTONFS(vp);
u_quad_t tsize = np->n_size;
u_quad_t tsize;
int biosize = vp->v_mount->mnt_stat.f_iosize;
int error = 0;
mtx_lock(&np->n_mtx);
tsize = np->n_size;
np->n_size = nsize;
mtx_unlock(&np->n_mtx);
if (np->n_size < tsize) {
if (nsize < tsize) {
struct buf *bp;
daddr_t lbn;
int bufsize;

29
sys/nfsclient/nfs_nfsiod.c
View file

@ -102,17 +102,22 @@ sysctl_iodmin(SYSCTL_HANDLER_ARGS)
error = sysctl_handle_int(oidp, &newmin, 0, req);
if (error || (req->newptr == NULL))
return (error);
if (newmin > nfs_iodmax)
return (EINVAL);
mtx_lock(&nfs_iod_mtx);
if (newmin > nfs_iodmax) {
error = EINVAL;
goto out;
}
nfs_iodmin = newmin;
if (nfs_numasync >= nfs_iodmin)
return (0);
goto out;
/*
* If the current number of nfsiod is lower
* than the new minimum, create some more.
*/
for (i = nfs_iodmin - nfs_numasync; i > 0; i--)
nfs_nfsiodnew();
out:
mtx_unlock(&nfs_iod_mtx);
return (0);
}
SYSCTL_PROC(_vfs_nfs, OID_AUTO, iodmin, CTLTYPE_UINT | CTLFLAG_RW, 0,
@ -131,9 +136,10 @@ sysctl_iodmax(SYSCTL_HANDLER_ARGS)
return (error);
if (newmax > NFS_MAXASYNCDAEMON)
return (EINVAL);
mtx_lock(&nfs_iod_mtx);
nfs_iodmax = newmax;
if (nfs_numasync <= nfs_iodmax)
return (0);
goto out;
/*
* If there are some asleep nfsiods that should
* exit, wakeup() them so that they check nfs_iodmax
@ -146,6 +152,8 @@ sysctl_iodmax(SYSCTL_HANDLER_ARGS)
wakeup(&nfs_iodwant[iod]);
iod--;
}
out:
mtx_unlock(&nfs_iod_mtx);
return (0);
}
SYSCTL_PROC(_vfs_nfs, OID_AUTO, iodmax, CTLTYPE_UINT | CTLFLAG_RW, 0,
@ -168,8 +176,10 @@ nfs_nfsiodnew(void)
}
if (newiod == -1)
return (-1);
mtx_unlock(&nfs_iod_mtx);
error = kthread_create(nfssvc_iod, nfs_asyncdaemon + i, NULL, RFHIGHPID,
0, "nfsiod %d", newiod);
mtx_lock(&nfs_iod_mtx);
if (error)
return (-1);
nfs_numasync++;
@ -183,6 +193,7 @@ nfsiod_setup(void *dummy)
int error;
TUNABLE_INT_FETCH("vfs.nfs.iodmin", &nfs_iodmin);
mtx_lock(&nfs_iod_mtx);
/* Silently limit the start number of nfsiod's */
if (nfs_iodmin > NFS_MAXASYNCDAEMON)
nfs_iodmin = NFS_MAXASYNCDAEMON;
@ -192,6 +203,7 @@ nfsiod_setup(void *dummy)
if (error == -1)
panic("nfsiod_setup: nfs_nfsiodnew failed");
}
mtx_unlock(&nfs_iod_mtx);
}
SYSINIT(nfsiod, SI_SUB_KTHREAD_IDLE, SI_ORDER_ANY, nfsiod_setup, NULL);
@ -211,7 +223,7 @@ nfssvc_iod(void *instance)
int myiod, timo;
int error = 0;
mtx_lock(&Giant);
mtx_lock(&nfs_iod_mtx);
myiod = (int *)instance - nfs_asyncdaemon;
/*
* Main loop
@ -230,7 +242,7 @@ nfssvc_iod(void *instance)
* Always keep at least nfs_iodmin kthreads.
*/
timo = (myiod < nfs_iodmin) ? 0 : nfs_iodmaxidle * hz;
error = tsleep(&nfs_iodwant[myiod], PWAIT | PCATCH,
error = msleep(&nfs_iodwant[myiod], &nfs_iod_mtx, PWAIT | PCATCH,
"-", timo);
}
if (error)
@ -243,6 +255,7 @@ nfssvc_iod(void *instance)
nmp->nm_bufqwant = 0;
wakeup(&nmp->nm_bufq);
}
mtx_unlock(&nfs_iod_mtx);
if (bp->b_flags & B_DIRECT) {
KASSERT((bp->b_iocmd == BIO_WRITE), ("nfscvs_iod: BIO_WRITE not set"));
(void)nfs_doio_directwrite(bp);
@ -252,7 +265,7 @@ nfssvc_iod(void *instance)
else
(void) nfs_doio(bp->b_vp, bp, bp->b_wcred, NULL);
}
mtx_lock(&nfs_iod_mtx);
/*
* If there are more than one iod on this mount, then defect
* so that the iods can be shared out fairly between the mounts
@ -276,7 +289,7 @@ finish:
/* Someone may be waiting for the last nfsiod to terminate. */
if (--nfs_numasync == 0)
wakeup(&nfs_numasync);
mtx_unlock(&Giant);
mtx_unlock(&nfs_iod_mtx);
if ((error == 0) || (error == EWOULDBLOCK))
kthread_exit(0);
/* Abnormal termination */

3
sys/nfsclient/nfs_node.c
View file

@ -164,6 +164,7 @@ nfs_nget(struct mount *mntp, nfsfh_t *fhp, int fhsize, struct nfsnode **npp)
np->n_fhp = &np->n_fh;
bcopy((caddr_t)fhp, (caddr_t)np->n_fhp, fhsize);
np->n_fhsize = fhsize;
mtx_init(&np->n_mtx, "NFSnode lock", NULL, MTX_DEF);
*npp = np;
return (0);
@ -234,7 +235,7 @@ nfs_reclaim(struct vop_reclaim_args *ap)
if (np->n_fhsize > NFS_SMALLFH) {
FREE((caddr_t)np->n_fhp, M_NFSBIGFH);
}
mtx_destroy(&np->n_mtx);
uma_zfree(nfsnode_zone, vp->v_data);
vp->v_data = NULL;
return (0);

443
sys/nfsclient/nfs_socket.c
View file

@ -115,7 +115,7 @@ static int nfs_realign_test;
static int nfs_realign_count;
static int nfs_bufpackets = 4;
static int nfs_reconnects;
static int nfs3_jukebox_delay = 10;
static int nfs3_jukebox_delay = 10;
SYSCTL_DECL(_vfs_nfs);
@ -125,8 +125,7 @@ SYSCTL_INT(_vfs_nfs, OID_AUTO, bufpackets, CTLFLAG_RW, &nfs_bufpackets, 0, "");
SYSCTL_INT(_vfs_nfs, OID_AUTO, reconnects, CTLFLAG_RD, &nfs_reconnects, 0,
"number of times the nfs client has had to reconnect");
SYSCTL_INT(_vfs_nfs, OID_AUTO, nfs3_jukebox_delay, CTLFLAG_RW, &nfs3_jukebox_delay, 0,
"number of seconds to delay a retry after receiving EJUKEBOX");
"number of seconds to delay a retry after receiving EJUKEBOX");
/*
* There is a congestion window for outstanding rpcs maintained per mount
@ -154,10 +153,8 @@ static void nfs_softterm(struct nfsreq *rep);
static int nfs_reconnect(struct nfsreq *rep);
static void nfs_clnt_tcp_soupcall(struct socket *so, void *arg, int waitflag);
static void nfs_clnt_udp_soupcall(struct socket *so, void *arg, int waitflag);
static void wakeup_nfsreq(struct nfsreq *req);
extern struct mtx nfs_reqq_mtx;
extern struct mtx nfs_reply_mtx;
/*
* Initialize sockets and congestion for a new NFS connection.
@ -172,13 +169,13 @@ nfs_connect(struct nfsmount *nmp, struct nfsreq *rep)
struct sockaddr *saddr;
struct thread *td = &thread0; /* only used for socreate and sobind */
NET_ASSERT_GIANT();
NET_LOCK_GIANT();
if (nmp->nm_sotype == SOCK_STREAM) {
mtx_lock(&nmp->nm_nfstcpstate.mtx);
mtx_lock(&nmp->nm_mtx);
nmp->nm_nfstcpstate.flags |= NFS_TCP_EXPECT_RPCMARKER;
nmp->nm_nfstcpstate.rpcresid = 0;
mtx_unlock(&nmp->nm_nfstcpstate.mtx);
mtx_unlock(&nmp->nm_mtx);
}
nmp->nm_so = NULL;
saddr = nmp->nm_nam;
@ -243,12 +240,16 @@ nfs_connect(struct nfsmount *nmp, struct nfsreq *rep)
* Protocols that do not require connections may be optionally left
* unconnected for servers that reply from a port other than NFS_PORT.
*/
mtx_lock(&nmp->nm_mtx);
if (nmp->nm_flag & NFSMNT_NOCONN) {
if (nmp->nm_soflags & PR_CONNREQUIRED) {
error = ENOTCONN;
mtx_unlock(&nmp->nm_mtx);
goto bad;
}
} else
mtx_unlock(&nmp->nm_mtx);
} else {
mtx_unlock(&nmp->nm_mtx);
error = soconnect(so, nmp->nm_nam, td);
if (error)
goto bad;
@ -290,7 +291,7 @@ nfs_connect(struct nfsmount *nmp, struct nfsreq *rep)
pktscale = 2;
if (pktscale > 64)
pktscale = 64;
mtx_lock(&nmp->nm_mtx);
if (nmp->nm_sotype == SOCK_DGRAM) {
sndreserve = (nmp->nm_wsize + NFS_MAXPKTHDR) * pktscale;
rcvreserve = (max(nmp->nm_rsize, nmp->nm_readdirsize) +
@ -313,7 +314,9 @@ nfs_connect(struct nfsmount *nmp, struct nfsreq *rep)
sopt.sopt_val = &val;
sopt.sopt_valsize = sizeof val;
val = 1;
mtx_unlock(&nmp->nm_mtx);
sosetopt(so, &sopt);
mtx_lock(&nmp->nm_mtx);
}
if (so->so_proto->pr_protocol == IPPROTO_TCP) {
struct sockopt sopt;
@ -326,13 +329,16 @@ nfs_connect(struct nfsmount *nmp, struct nfsreq *rep)
sopt.sopt_val = &val;
sopt.sopt_valsize = sizeof val;
val = 1;
mtx_unlock(&nmp->nm_mtx);
sosetopt(so, &sopt);
mtx_lock(&nmp->nm_mtx);
}
sndreserve = (nmp->nm_wsize + NFS_MAXPKTHDR +
sizeof (u_int32_t)) * pktscale;
rcvreserve = (nmp->nm_rsize + NFS_MAXPKTHDR +
sizeof (u_int32_t)) * pktscale;
}
mtx_unlock(&nmp->nm_mtx);
error = soreserve(so, sndreserve, rcvreserve);
if (error)
goto bad;
@ -349,6 +355,7 @@ nfs_connect(struct nfsmount *nmp, struct nfsreq *rep)
so->so_snd.sb_flags |= SB_NOINTR;
SOCKBUF_UNLOCK(&so->so_snd);
mtx_lock(&nmp->nm_mtx);
/* Initialize other non-zero congestion variables */
nmp->nm_srtt[0] = nmp->nm_srtt[1] = nmp->nm_srtt[2] =
nmp->nm_srtt[3] = (NFS_TIMEO << 3);
@ -357,10 +364,13 @@ nfs_connect(struct nfsmount *nmp, struct nfsreq *rep)
nmp->nm_cwnd = NFS_MAXCWND / 2; /* Initial send window */
nmp->nm_sent = 0;
nmp->nm_timeouts = 0;
mtx_unlock(&nmp->nm_mtx);
NET_UNLOCK_GIANT();
return (0);
bad:
nfs_disconnect(nmp);
NET_UNLOCK_GIANT();
return (error);
}
@ -387,7 +397,9 @@ nfs_reconnect(struct nfsreq *rep)
error = EINTR;
if (error == EIO || error == EINTR)
return (error);
mtx_lock(&Giant);
(void) tsleep(&lbolt, PSOCK, "nfscon", 0);
mtx_unlock(&Giant);
}
/*
@ -399,9 +411,10 @@ nfs_reconnect(struct nfsreq *rep)
* until the connection is established successfully, and
* then re-transmit the request.
*/
mtx_lock(&nmp->nm_nfstcpstate.mtx);
mtx_lock(&nmp->nm_mtx);
nmp->nm_nfstcpstate.flags &= ~NFS_TCP_FORCE_RECONNECT;
mtx_unlock(&nmp->nm_nfstcpstate.mtx);
nmp->nm_nfstcpstate.rpcresid = 0;
mtx_unlock(&nmp->nm_mtx);
/*
* Loop through outstanding request list and fix up all requests
@ -409,8 +422,11 @@ nfs_reconnect(struct nfsreq *rep)
*/
mtx_lock(&nfs_reqq_mtx);
TAILQ_FOREACH(rp, &nfs_reqq, r_chain) {
if (rp->r_nmp == nmp)
if (rp->r_nmp == nmp) {
mtx_lock(&rp->r_mtx);
rp->r_flags |= R_MUSTRESEND;
mtx_unlock(&rp->r_mtx);
}
}
mtx_unlock(&nfs_reqq_mtx);
return (0);
@ -426,9 +442,11 @@ nfs_disconnect(struct nfsmount *nmp)
NET_ASSERT_GIANT();
mtx_lock(&nmp->nm_mtx);
if (nmp->nm_so) {
so = nmp->nm_so;
nmp->nm_so = NULL;
mtx_unlock(&nmp->nm_mtx);
SOCKBUF_LOCK(&so->so_rcv);
so->so_upcallarg = NULL;
so->so_upcall = NULL;
@ -436,7 +454,8 @@ nfs_disconnect(struct nfsmount *nmp)
SOCKBUF_UNLOCK(&so->so_rcv);
soshutdown(so, SHUT_WR);
soclose(so);
}
} else
mtx_unlock(&nmp->nm_mtx);
}
void
@ -463,22 +482,29 @@ nfs_send(struct socket *so, struct sockaddr *nam, struct mbuf *top,
struct sockaddr *sendnam;
int error, error2, soflags, flags;
NET_ASSERT_GIANT();
NET_LOCK_GIANT();
KASSERT(rep, ("nfs_send: called with rep == NULL"));
error = nfs_sigintr(rep->r_nmp, rep, rep->r_td);
if (error) {
m_freem(top);
return (error);
goto out;
}
mtx_lock(&rep->r_nmp->nm_mtx);
mtx_lock(&rep->r_mtx);
if ((so = rep->r_nmp->nm_so) == NULL) {
rep->r_flags |= R_MUSTRESEND;
mtx_unlock(&rep->r_mtx);
mtx_unlock(&rep->r_nmp->nm_mtx);
m_freem(top);
return (0);
error = 0;
goto out;
}
rep->r_flags &= ~R_MUSTRESEND;
soflags = rep->r_nmp->nm_soflags;
mtx_unlock(&rep->r_mtx);
mtx_unlock(&rep->r_nmp->nm_mtx);
if ((soflags & PR_CONNREQUIRED) || (so->so_state & SS_ISCONNECTED))
sendnam = NULL;
@ -493,7 +519,9 @@ nfs_send(struct socket *so, struct sockaddr *nam, struct mbuf *top,
flags, curthread /*XXX*/);
if (error == ENOBUFS && so->so_type == SOCK_DGRAM) {
error = 0;
mtx_lock(&rep->r_mtx);
rep->r_flags |= R_MUSTRESEND;
mtx_unlock(&rep->r_mtx);
}
if (error) {
@ -513,8 +541,11 @@ nfs_send(struct socket *so, struct sockaddr *nam, struct mbuf *top,
error2 = NFS_SIGREP(rep);
if (error2)
error = error2;
else
else {
mtx_lock(&rep->r_mtx);
rep->r_flags |= R_MUSTRESEND;
mtx_unlock(&rep->r_mtx);
}
/*
* Handle any recoverable (soft) socket errors here. (?)
@ -523,6 +554,8 @@ nfs_send(struct socket *so, struct sockaddr *nam, struct mbuf *top,
error != EWOULDBLOCK && error != EPIPE)
error = 0;
}
out:
NET_UNLOCK_GIANT();
return (error);
}
@ -533,7 +566,7 @@ nfs_reply(struct nfsreq *rep)
register struct mbuf *m;
int error = 0, sotype, slpflag;
NET_ASSERT_GIANT();
NET_LOCK_GIANT();
sotype = rep->r_nmp->nm_sotype;
/*
@ -543,30 +576,39 @@ nfs_reply(struct nfsreq *rep)
if (sotype != SOCK_DGRAM) {
error = nfs_sndlock(rep);
if (error)
return (error);
goto out;
tryagain:
mtx_lock(&rep->r_nmp->nm_mtx);
mtx_lock(&rep->r_mtx);
if (rep->r_mrep) {
mtx_unlock(&rep->r_mtx);
mtx_unlock(&rep->r_nmp->nm_mtx);
nfs_sndunlock(rep);
return (0);
error = 0;
goto out;
}
if (rep->r_flags & R_SOFTTERM) {
mtx_unlock(&rep->r_mtx);
mtx_unlock(&rep->r_nmp->nm_mtx);
nfs_sndunlock(rep);
return (EINTR);
error = EINTR;
goto out;
}
so = rep->r_nmp->nm_so;
mtx_lock(&rep->r_nmp->nm_nfstcpstate.mtx);
if (!so ||
(rep->r_nmp->nm_nfstcpstate.flags & NFS_TCP_FORCE_RECONNECT)) {
mtx_unlock(&rep->r_nmp->nm_nfstcpstate.mtx);
mtx_unlock(&rep->r_mtx);
mtx_unlock(&rep->r_nmp->nm_mtx);
error = nfs_reconnect(rep);
if (error) {
nfs_sndunlock(rep);
return (error);
goto out;
}
goto tryagain;
} else
mtx_unlock(&rep->r_nmp->nm_nfstcpstate.mtx);
}
while (rep->r_flags & R_MUSTRESEND) {
mtx_unlock(&rep->r_mtx);
mtx_unlock(&rep->r_nmp->nm_mtx);
m = m_copym(rep->r_mreq, 0, M_COPYALL, M_WAIT);
nfsstats.rpcretries++;
error = nfs_send(so, rep->r_nmp->nm_nam, m, rep);
@ -574,41 +616,59 @@ tryagain:
if (error == EINTR || error == ERESTART ||
(error = nfs_reconnect(rep)) != 0) {
nfs_sndunlock(rep);
return (error);
goto out;
}
goto tryagain;
}
mtx_lock(&rep->r_nmp->nm_mtx);
mtx_lock(&rep->r_mtx);
}
mtx_unlock(&rep->r_nmp->nm_mtx);
mtx_unlock(&rep->r_mtx);
nfs_sndunlock(rep);
}
slpflag = 0;
mtx_lock(&rep->r_nmp->nm_mtx);
if (rep->r_nmp->nm_flag & NFSMNT_INT)
slpflag = PCATCH;
mtx_lock(&nfs_reply_mtx);
mtx_unlock(&rep->r_nmp->nm_mtx);
mtx_lock(&rep->r_mtx);
while ((rep->r_mrep == NULL) && (error == 0) &&
((rep->r_flags & R_SOFTTERM) == 0) &&
((sotype == SOCK_DGRAM) || ((rep->r_flags & R_MUSTRESEND) == 0)))
error = msleep((caddr_t)rep, &nfs_reply_mtx,
error = msleep((caddr_t)rep, &rep->r_mtx,
slpflag | (PZERO - 1), "nfsreq", 0);
mtx_unlock(&nfs_reply_mtx);
if (error == EINTR || error == ERESTART)
if (error == EINTR || error == ERESTART) {
/* NFS operations aren't restartable. Map ERESTART to EINTR */
return (EINTR);
if (rep->r_flags & R_SOFTTERM)
error = EINTR;
mtx_unlock(&rep->r_mtx);
goto out;
}
if (rep->r_flags & R_SOFTTERM) {
/* Request was terminated because we exceeded the retries (soft mount) */
return (ETIMEDOUT);
error = ETIMEDOUT;
mtx_unlock(&rep->r_mtx);
goto out;
}
mtx_unlock(&rep->r_mtx);
if (sotype == SOCK_STREAM) {
mtx_lock(&rep->r_nmp->nm_nfstcpstate.mtx);
mtx_lock(&rep->r_nmp->nm_mtx);
mtx_lock(&rep->r_mtx);
if (((rep->r_nmp->nm_nfstcpstate.flags & NFS_TCP_FORCE_RECONNECT) ||
(rep->r_flags & R_MUSTRESEND))) {
mtx_unlock(&rep->r_nmp->nm_nfstcpstate.mtx);
mtx_unlock(&rep->r_mtx);
mtx_unlock(&rep->r_nmp->nm_mtx);
error = nfs_sndlock(rep);
if (error)
return (error);
goto out;
goto tryagain;
} else
mtx_unlock(&rep->r_nmp->nm_nfstcpstate.mtx);
} else {
mtx_unlock(&rep->r_mtx);
mtx_unlock(&rep->r_nmp->nm_mtx);
}
}
out:
NET_UNLOCK_GIANT();
return (error);
}
@ -660,6 +720,8 @@ nfsmout:
* Iff no match, just drop the datagram
*/
TAILQ_FOREACH(rep, &nfs_reqq, r_chain) {
mtx_lock(&nmp->nm_mtx);
mtx_lock(&rep->r_mtx);
if (rep->r_mrep == NULL && rxid == rep->r_xid) {
/* Found it.. */
rep->r_mrep = mrep;
@ -703,8 +765,13 @@ nfsmout:
NFS_SDRTT(rep) += t1;
}
nmp->nm_timeouts = 0;
wakeup((caddr_t)rep);
mtx_unlock(&rep->r_mtx);
mtx_unlock(&nmp->nm_mtx);
break;
}
mtx_unlock(&rep->r_mtx);
mtx_unlock(&nmp->nm_mtx);
}
/*
* If not matched to a request, drop it.
@ -713,31 +780,18 @@ nfsmout:
if (rep == 0) {
nfsstats.rpcunexpected++;
m_freem(mrep);
} else
wakeup_nfsreq(rep);
}
mtx_unlock(&nfs_reqq_mtx);
}
/*
* The wakeup of the requestor should be done under the mutex
* to avoid potential missed wakeups.
*/
static void
wakeup_nfsreq(struct nfsreq *req)
{
mtx_lock(&nfs_reply_mtx);
wakeup((caddr_t)req);
mtx_unlock(&nfs_reply_mtx);
}
static void
nfs_mark_for_reconnect(struct nfsmount *nmp)
{
struct nfsreq *rp;
mtx_lock(&nmp->nm_nfstcpstate.mtx);
mtx_lock(&nmp->nm_mtx);
nmp->nm_nfstcpstate.flags |= NFS_TCP_FORCE_RECONNECT;
mtx_unlock(&nmp->nm_nfstcpstate.mtx);
mtx_unlock(&nmp->nm_mtx);
/*
* Wakeup all processes that are waiting for replies
* on this mount point. One of them does the reconnect.
@ -745,8 +799,10 @@ nfs_mark_for_reconnect(struct nfsmount *nmp)
mtx_lock(&nfs_reqq_mtx);
TAILQ_FOREACH(rp, &nfs_reqq, r_chain) {
if (rp->r_nmp == nmp) {
mtx_lock(&rp->r_mtx);
rp->r_flags |= R_MUSTRESEND;
wakeup_nfsreq(rp);
wakeup((caddr_t)rp);
mtx_unlock(&rp->r_mtx);
}
}
mtx_unlock(&nfs_reqq_mtx);
@ -795,19 +851,21 @@ nfs_clnt_tcp_soupcall(struct socket *so, void *arg, int waitflag)
* Don't pick any more data from the socket if we've marked the
* mountpoint for reconnect.
*/
mtx_lock(&nmp->nm_nfstcpstate.mtx);
mtx_lock(&nmp->nm_mtx);
if (nmp->nm_nfstcpstate.flags & NFS_TCP_FORCE_RECONNECT) {
mtx_unlock(&nmp->nm_nfstcpstate.mtx);
mtx_unlock(&nmp->nm_mtx);
return;
} else
mtx_unlock(&nmp->nm_nfstcpstate.mtx);
mtx_unlock(&nmp->nm_mtx);
auio.uio_td = curthread;
auio.uio_segflg = UIO_SYSSPACE;
auio.uio_rw = UIO_READ;
for ( ; ; ) {
mtx_lock(&nmp->nm_mtx);
if (nmp->nm_nfstcpstate.flags & NFS_TCP_EXPECT_RPCMARKER) {
int resid;
mtx_unlock(&nmp->nm_mtx);
if (!nfstcp_marker_readable(so)) {
/* Marker is not readable */
return;
@ -864,14 +922,20 @@ nfs_clnt_tcp_soupcall(struct socket *so, void *arg, int waitflag)
nmp->nm_mountp->mnt_stat.f_mntfromname);
goto mark_reconnect;
}
mtx_lock(&nmp->nm_mtx);
nmp->nm_nfstcpstate.rpcresid = len;
nmp->nm_nfstcpstate.flags &= ~(NFS_TCP_EXPECT_RPCMARKER);
}
mtx_unlock(&nmp->nm_mtx);
} else
mtx_unlock(&nmp->nm_mtx);
/*
* Processed RPC marker or no RPC marker to process.
* Pull in and process data.
*/
mtx_lock(&nmp->nm_mtx);
if (nmp->nm_nfstcpstate.rpcresid > 0) {
mtx_unlock(&nmp->nm_mtx);
if (!nfstcp_readable(so, nmp->nm_nfstcpstate.rpcresid)) {
/* All data not readable */
return;
@ -894,11 +958,14 @@ nfs_clnt_tcp_soupcall(struct socket *so, void *arg, int waitflag)
}
if (mp == NULL)
panic("nfs_clnt_tcp_soupcall: Got empty mbuf chain from sorecv\n");
mtx_lock(&nmp->nm_mtx);
nmp->nm_nfstcpstate.rpcresid = 0;
nmp->nm_nfstcpstate.flags |= NFS_TCP_EXPECT_RPCMARKER;
mtx_unlock(&nmp->nm_mtx);
/* We got the entire RPC reply. Match XIDs and wake up requestor */
nfs_clnt_match_xid(so, nmp, mp);
}
} else
mtx_unlock(&nmp->nm_mtx);
}
mark_reconnect:
@ -953,7 +1020,7 @@ nfs_request(struct vnode *vp, struct mbuf *mrest, int procnum,
struct mbuf *m, *md, *mheadend;
time_t waituntil;
caddr_t dpos;
int s, error = 0, mrest_len, auth_len, auth_type;
int error = 0, mrest_len, auth_len, auth_type;
struct timeval now;
u_int32_t *xidp;
@ -966,11 +1033,12 @@ nfs_request(struct vnode *vp, struct mbuf *mrest, int procnum,
if ((nmp->nm_flag & NFSMNT_NFSV4) != 0)
return nfs4_request(vp, mrest, procnum, td, cred, mrp, mdp, dposp);
MALLOC(rep, struct nfsreq *, sizeof(struct nfsreq), M_NFSREQ, M_WAITOK);
rep->r_mrep = rep->r_md = NULL;
bzero(rep, sizeof(struct nfsreq));
rep->r_nmp = nmp;
rep->r_vp = vp;
rep->r_td = td;
rep->r_procnum = procnum;
mtx_init(&rep->r_mtx, "NFSrep lock", NULL, MTX_DEF);
getmicrouptime(&now);
rep->r_lastmsg = now.tv_sec -
@ -1019,7 +1087,6 @@ tryagain:
* Chain request into list of outstanding requests. Be sure
* to put it LAST so timer finds oldest requests first.
*/
s = splsoftclock();
mtx_lock(&nfs_reqq_mtx);
if (TAILQ_EMPTY(&nfs_reqq))
callout_reset(&nfs_callout, nfs_ticks, nfs_timer, NULL);
@ -1031,10 +1098,11 @@ tryagain:
* send this one now but let timer do it. If not timing a request,
* do it now.
*/
mtx_lock(&nmp->nm_mtx);
if (nmp->nm_so && (nmp->nm_sotype != SOCK_DGRAM ||
(nmp->nm_flag & NFSMNT_DUMBTIMR) ||
nmp->nm_sent < nmp->nm_cwnd)) {
splx(s);
mtx_unlock(&nmp->nm_mtx);
error = nfs_sndlock(rep);
if (!error) {
m2 = m_copym(m, 0, M_COPYALL, M_TRYWAIT);
@ -1047,12 +1115,14 @@ tryagain:
* blocking on nfs_send() too long, so check for R_SENT here.
*/
if (!error && (rep->r_flags & (R_SENT | R_MUSTRESEND)) == 0) {
mtx_lock(&nmp->nm_mtx);
nmp->nm_sent += NFS_CWNDSCALE;
mtx_unlock(&nmp->nm_mtx);
rep->r_flags |= R_SENT;
}
mtx_unlock(&nfs_reqq_mtx);
} else {
splx(s);
mtx_unlock(&nmp->nm_mtx);
rep->r_rtt = -1;
}
@ -1065,7 +1135,6 @@ tryagain:
/*
* RPC done, unlink the request.
*/
s = splsoftclock();
mtx_lock(&nfs_reqq_mtx);
/*
* nfs_timer() may be in the process of re-transmitting this request.
@ -1086,10 +1155,11 @@ tryagain:
*/
if (rep->r_flags & R_SENT) {
rep->r_flags &= ~R_SENT; /* paranoia */
mtx_lock(&nmp->nm_mtx);
nmp->nm_sent -= NFS_CWNDSCALE;
mtx_unlock(&nmp->nm_mtx);
}
mtx_unlock(&nfs_reqq_mtx);
splx(s);
/*
* If there was a successful reply and a tprintf msg.
@ -1113,6 +1183,7 @@ tryagain:
if (rep->r_mrep != NULL)
m_freem(rep->r_mrep);
m_freem(rep->r_mreq);
mtx_destroy(&rep->r_mtx);
free((caddr_t)rep, M_NFSREQ);
return (error);
}
@ -1131,6 +1202,7 @@ tryagain:
error = EACCES;
m_freem(mrep);
m_freem(rep->r_mreq);
mtx_destroy(&rep->r_mtx);
free((caddr_t)rep, M_NFSREQ);
return (error);
}
@ -1153,12 +1225,16 @@ tryagain:
m_freem(mrep);
error = 0;
waituntil = time_second + nfs3_jukebox_delay;
while (time_second < waituntil)
(void) tsleep(&lbolt,
PSOCK, "nqnfstry", 0);
while (time_second < waituntil) {
mtx_lock(&Giant);
(void) tsleep(&lbolt, PSOCK, "nqnfstry", 0);
mtx_unlock(&Giant);
}
mtx_lock(&nfs_reqq_mtx);
if (++nfs_xid == 0)
nfs_xid++;
rep->r_xid = *xidp = txdr_unsigned(nfs_xid);
mtx_unlock(&nfs_reqq_mtx);
goto tryagain;
}
@ -1176,6 +1252,7 @@ tryagain:
} else
m_freem(mrep);
m_freem(rep->r_mreq);
mtx_destroy(&rep->r_mtx);
free((caddr_t)rep, M_NFSREQ);
return (error);
}
@ -1184,6 +1261,7 @@ tryagain:
*mdp = md;
*dposp = dpos;
m_freem(rep->r_mreq);
mtx_destroy(&rep->r_mtx);
FREE((caddr_t)rep, M_NFSREQ);
return (0);
}
@ -1191,6 +1269,7 @@ tryagain:
error = EPROTONOSUPPORT;
nfsmout:
m_freem(rep->r_mreq);
mtx_destroy(&rep->r_mtx);
free((caddr_t)rep, M_NFSREQ);
return (error);
}
@ -1215,34 +1294,34 @@ nfs_timer(void *arg)
struct socket *so;
struct nfsmount *nmp;
int timeo;
int s, error;
int error;
struct timeval now;
getmicrouptime(&now);
s = splnet();
mtx_lock(&Giant); /* nfs_down -> tprintf */
mtx_lock(&nfs_reqq_mtx);
TAILQ_FOREACH(rep, &nfs_reqq, r_chain) {
nmp = rep->r_nmp;
if (rep->r_mrep || (rep->r_flags & R_SOFTTERM))
mtx_lock(&rep->r_mtx);
if (rep->r_mrep || (rep->r_flags & R_SOFTTERM)) {
mtx_unlock(&rep->r_mtx);
continue;
} else
mtx_unlock(&rep->r_mtx);
if (nfs_sigintr(nmp, rep, rep->r_td))
continue;
mtx_lock(&nmp->nm_mtx);
mtx_lock(&rep->r_mtx);
if (nmp->nm_tprintf_initial_delay != 0 &&
(rep->r_rexmit > 2 || (rep->r_flags & R_RESENDERR)) &&
rep->r_lastmsg + nmp->nm_tprintf_delay < now.tv_sec) {
mtx_unlock(&rep->r_mtx);
mtx_unlock(&nmp->nm_mtx);
rep->r_lastmsg = now.tv_sec;
nfs_down(rep, nmp, rep->r_td, "not responding",
0, NFSSTA_TIMEO);
#if 0
if (!(nmp->nm_state & NFSSTA_MOUNTED)) {
/* we're not yet completely mounted and */
/* we can't complete an RPC, so we fail */
nfsstats.rpctimeouts++;
nfs_softterm(rep);
continue;
}
#endif
0, NFSSTA_TIMEO);
mtx_lock(&nmp->nm_mtx);
mtx_lock(&rep->r_mtx);
}
if (rep->r_rtt >= 0) {
rep->r_rtt++;
@ -1252,14 +1331,19 @@ nfs_timer(void *arg)
timeo = NFS_RTO(nmp, proct[rep->r_procnum]);
if (nmp->nm_timeouts > 0)
timeo *= nfs_backoff[nmp->nm_timeouts - 1];
if (rep->r_rtt <= timeo)
if (rep->r_rtt <= timeo) {
mtx_unlock(&rep->r_mtx);
mtx_unlock(&nmp->nm_mtx);
continue;
}
if (nmp->nm_timeouts < NFS_NBACKOFF)
nmp->nm_timeouts++;
}
if (rep->r_rexmit >= rep->r_retry) { /* too many */
nfsstats.rpctimeouts++;
nfs_softterm(rep);
mtx_unlock(&rep->r_mtx);
mtx_unlock(&nmp->nm_mtx);
continue;
}
if (nmp->nm_sotype != SOCK_DGRAM) {
@ -1272,12 +1356,17 @@ nfs_timer(void *arg)
* if necessary.
*/
rep->r_flags |= R_MUSTRESEND;
wakeup_nfsreq(rep);
wakeup((caddr_t)rep);
rep->r_rtt = 0;
mtx_unlock(&rep->r_mtx);
mtx_unlock(&nmp->nm_mtx);
continue;
}
if ((so = nmp->nm_so) == NULL)
if ((so = nmp->nm_so) == NULL) {
mtx_unlock(&rep->r_mtx);
mtx_unlock(&nmp->nm_mtx);
continue;
}
/*
* If there is enough space and the window allows..
* Resend it
@ -1285,57 +1374,69 @@ nfs_timer(void *arg)
*/
rep->r_rtt = -1;
if (sbspace(&so->so_snd) >= rep->r_mreq->m_pkthdr.len &&
((nmp->nm_flag & NFSMNT_DUMBTIMR) ||
(rep->r_flags & R_SENT) ||
nmp->nm_sent < nmp->nm_cwnd) &&
(m = m_copym(rep->r_mreq, 0, M_COPYALL, M_DONTWAIT))) {
/*
* Mark the request to indicate that a XMIT is in progress
* to prevent the req structure being removed in nfs_request().
*/
rep->r_flags |= R_REXMIT_INPROG;
mtx_unlock(&nfs_reqq_mtx);
NET_LOCK_GIANT();
if ((nmp->nm_flag & NFSMNT_NOCONN) == 0)
error = (*so->so_proto->pr_usrreqs->pru_send)
(so, 0, m, NULL, NULL, curthread);
else
error = (*so->so_proto->pr_usrreqs->pru_send)
(so, 0, m, nmp->nm_nam, NULL, curthread);
NET_UNLOCK_GIANT();
mtx_lock(&nfs_reqq_mtx);
rep->r_flags &= ~R_REXMIT_INPROG;
wakeup((caddr_t)&rep->r_flags);
if (error) {
if (NFSIGNORE_SOERROR(nmp->nm_soflags, error))
so->so_error = 0;
rep->r_flags |= R_RESENDERR;
} else {
((nmp->nm_flag & NFSMNT_DUMBTIMR) || (rep->r_flags & R_SENT) ||
nmp->nm_sent < nmp->nm_cwnd)) {
mtx_unlock(&rep->r_mtx);
mtx_unlock(&nmp->nm_mtx);
if ((m = m_copym(rep->r_mreq, 0, M_COPYALL, M_DONTWAIT))) {
/*
* Iff first send, start timing
* else turn timing off, backoff timer
* and divide congestion window by 2.
* Mark the request to indicate that a XMIT is in
* progress to prevent the req structure being
* removed in nfs_request().
*/
rep->r_flags &= ~R_RESENDERR;
if (rep->r_flags & R_SENT) {
rep->r_flags &= ~R_TIMING;
if (++rep->r_rexmit > NFS_MAXREXMIT)
rep->r_rexmit = NFS_MAXREXMIT;
nmp->nm_cwnd >>= 1;
if (nmp->nm_cwnd < NFS_CWNDSCALE)
nmp->nm_cwnd = NFS_CWNDSCALE;
nfsstats.rpcretries++;
mtx_lock(&rep->r_mtx);
rep->r_flags |= R_REXMIT_INPROG;
mtx_unlock(&rep->r_mtx);
mtx_unlock(&nfs_reqq_mtx);
NET_LOCK_GIANT();
if ((nmp->nm_flag & NFSMNT_NOCONN) == 0)
error = (*so->so_proto->pr_usrreqs->pru_send)
(so, 0, m, NULL, NULL, curthread);
else
error = (*so->so_proto->pr_usrreqs->pru_send)
(so, 0, m, nmp->nm_nam, NULL,
curthread);
NET_UNLOCK_GIANT();
mtx_lock(&nfs_reqq_mtx);
mtx_lock(&nmp->nm_mtx);
mtx_lock(&rep->r_mtx);
rep->r_flags &= ~R_REXMIT_INPROG;
wakeup((caddr_t)&rep->r_flags);
if (error) {
if (NFSIGNORE_SOERROR(nmp->nm_soflags, error))
so->so_error = 0;
rep->r_flags |= R_RESENDERR;
} else {
rep->r_flags |= R_SENT;
nmp->nm_sent += NFS_CWNDSCALE;
/*
* Iff first send, start timing
* else turn timing off, backoff timer
* and divide congestion window by 2.
*/
rep->r_flags &= ~R_RESENDERR;
if (rep->r_flags & R_SENT) {
rep->r_flags &= ~R_TIMING;
if (++rep->r_rexmit > NFS_MAXREXMIT)
rep->r_rexmit = NFS_MAXREXMIT;
nmp->nm_cwnd >>= 1;
if (nmp->nm_cwnd < NFS_CWNDSCALE)
nmp->nm_cwnd = NFS_CWNDSCALE;
nfsstats.rpcretries++;
} else {
rep->r_flags |= R_SENT;
nmp->nm_sent += NFS_CWNDSCALE;
}
rep->r_rtt = 0;
}
rep->r_rtt = 0;
mtx_unlock(&rep->r_mtx);
mtx_unlock(&nmp->nm_mtx);
}
} else {
mtx_unlock(&rep->r_mtx);
mtx_unlock(&nmp->nm_mtx);
}
}
mtx_unlock(&nfs_reqq_mtx);
mtx_unlock(&Giant); /* nfs_down -> tprintf */
splx(s);
callout_reset(&nfs_callout, nfs_ticks, nfs_timer, NULL);
}
@ -1349,31 +1450,33 @@ nfs_nmcancelreqs(nmp)
struct nfsmount *nmp;
{
struct nfsreq *req;
int i, s;
int i;
s = splnet();
mtx_lock(&nfs_reqq_mtx);
TAILQ_FOREACH(req, &nfs_reqq, r_chain) {
mtx_lock(&req->r_mtx);
if (nmp != req->r_nmp || req->r_mrep != NULL ||
(req->r_flags & R_SOFTTERM))
(req->r_flags & R_SOFTTERM)) {
mtx_unlock(&req->r_mtx);
continue;
}
nfs_softterm(req);
mtx_unlock(&req->r_mtx);
}
mtx_unlock(&nfs_reqq_mtx);
splx(s);
for (i = 0; i < 30; i++) {
s = splnet();
mtx_lock(&nfs_reqq_mtx);
TAILQ_FOREACH(req, &nfs_reqq, r_chain) {
if (nmp == req->r_nmp)
break;
}
mtx_unlock(&nfs_reqq_mtx);
splx(s);
if (req == NULL)
return (0);
mtx_lock(&Giant);
tsleep(&lbolt, PSOCK, "nfscancel", 0);
mtx_unlock(&Giant);
}
return (EBUSY);
}
@ -1387,7 +1490,7 @@ nfs_nmcancelreqs(nmp)
static void
nfs_softterm(struct nfsreq *rep)
{
KASSERT(mtx_owned(&rep->r_mtx), ("NFS req lock not owned !"));
rep->r_flags |= R_SOFTTERM;
if (rep->r_flags & R_SENT) {
rep->r_nmp->nm_sent -= NFS_CWNDSCALE;
@ -1397,7 +1500,7 @@ nfs_softterm(struct nfsreq *rep)
* Request terminated, wakeup the blocked process, so that we
* can return EINTR back.
*/
wakeup_nfsreq(rep);
wakeup((caddr_t)rep);
}
/*
@ -1493,28 +1596,6 @@ nfs_msleep(struct thread *td, void *ident, struct mtx *mtx, int priority, char *
return (error);
}
/*
* NFS wrapper to tsleep(), that shoves a new p_sigmask and restores the
* old one after tsleep() returns.
*/
int
nfs_tsleep(struct thread *td, void *ident, int priority, char *wmesg, int timo)
{
sigset_t oldset;
int error;
struct proc *p;
if ((priority & PCATCH) == 0)
return tsleep(ident, priority, wmesg, timo);
if (td == NULL)
td = curthread; /* XXX */
nfs_set_sigmask(td, &oldset);
error = tsleep(ident, priority, wmesg, timo);
nfs_restore_sigmask(td, &oldset);
p = td->td_proc;
return (error);
}
/*
* Test for a termination condition pending on the process.
* This is used for NFSMNT_INT mounts.
@ -1524,19 +1605,28 @@ nfs_sigintr(struct nfsmount *nmp, struct nfsreq *rep, struct thread *td)
{
struct proc *p;
sigset_t tmpset;
int error = 0;
if ((nmp->nm_flag & NFSMNT_NFSV4) != 0)
return nfs4_sigintr(nmp, rep, td);
if (rep && (rep->r_flags & R_SOFTTERM))
return (EIO);
if (rep) {
mtx_lock(&rep->r_mtx);
if (rep->r_flags & R_SOFTTERM) {
mtx_unlock(&rep->r_mtx);
error = EIO;
goto out;
} else
mtx_unlock(&rep->r_mtx);
}
/* Terminate all requests while attempting a forced unmount. */
if (nmp->nm_mountp->mnt_kern_flag & MNTK_UNMOUNTF)
return (EIO);
if (nmp->nm_mountp->mnt_kern_flag & MNTK_UNMOUNTF) {
error = EIO;
goto out;
}
if (!(nmp->nm_flag & NFSMNT_INT))
return (0);
goto out;
if (td == NULL)
return (0);
p = td->td_proc;
PROC_LOCK(p);
tmpset = p->p_siglist;
@ -1551,6 +1641,8 @@ nfs_sigintr(struct nfsmount *nmp, struct nfsreq *rep, struct thread *td)
PROC_UNLOCK(p);
return (0);
out:
return(error);
}
/*
@ -1567,21 +1659,25 @@ nfs_sndlock(struct nfsreq *rep)
int error, slpflag = 0, slptimeo = 0;
td = rep->r_td;
mtx_lock(&rep->r_nmp->nm_mtx);
if (rep->r_nmp->nm_flag & NFSMNT_INT)
slpflag = PCATCH;
while (*statep & NFSSTA_SNDLOCK) {
error = nfs_sigintr(rep->r_nmp, rep, td);
if (error)
if (error) {
mtx_unlock(&rep->r_nmp->nm_mtx);
return (error);
}
*statep |= NFSSTA_WANTSND;
(void) tsleep(statep, slpflag | (PZERO - 1),
"nfsndlck", slptimeo);
(void) msleep(statep, &rep->r_nmp->nm_mtx,
slpflag | (PZERO - 1), "nfsndlck", slptimeo);
if (slpflag == PCATCH) {
slpflag = 0;
slptimeo = 2 * hz;
}
}
*statep |= NFSSTA_SNDLOCK;
mtx_unlock(&rep->r_nmp->nm_mtx);
return (0);
}
@ -1593,6 +1689,7 @@ nfs_sndunlock(struct nfsreq *rep)
{
int *statep = &rep->r_nmp->nm_state;
mtx_lock(&rep->r_nmp->nm_mtx);
if ((*statep & NFSSTA_SNDLOCK) == 0)
panic("nfs sndunlock");
*statep &= ~NFSSTA_SNDLOCK;
@ -1600,6 +1697,7 @@ nfs_sndunlock(struct nfsreq *rep)
*statep &= ~NFSSTA_WANTSND;
wakeup(statep);
}
mtx_unlock(&rep->r_nmp->nm_mtx);
}
/*
@ -1703,8 +1801,10 @@ nfs_down(rep, nmp, td, msg, error, flags)
nmp->nm_state |= NFSSTA_LOCKTIMEO;
}
#endif
mtx_lock(&rep->r_mtx);
if (rep)
rep->r_flags |= R_TPRINTFMSG;
mtx_unlock(&rep->r_mtx);
nfs_msg(td, nmp->nm_mountp->mnt_stat.f_mntfromname, msg, error);
}
@ -1721,8 +1821,10 @@ nfs_up(rep, nmp, td, msg, flags)
if (nmp == NULL)
return;
mtx_lock(&rep->r_mtx);
if ((rep == NULL) || (rep->r_flags & R_TPRINTFMSG) != 0)
nfs_msg(td, nmp->nm_mountp->mnt_stat.f_mntfromname, msg, 0);
mtx_unlock(&rep->r_mtx);
if ((flags & NFSSTA_TIMEO) && (nmp->nm_state & NFSSTA_TIMEO)) {
nmp->nm_state &= ~NFSSTA_TIMEO;
vfs_event_signal(&nmp->nm_mountp->mnt_stat.f_fsid,
@ -1736,4 +1838,3 @@ nfs_up(rep, nmp, td, msg, flags)
}
#endif
}

132
sys/nfsclient/nfs_subs.c
View file

@ -75,6 +75,12 @@ __FBSDID("$FreeBSD$");
#include <netinet/in.h>
/*
* Note that stdarg.h and the ANSI style va_start macro is used for both
* ANSI and traditional C compilers.
*/
#include <machine/stdarg.h>
/*
* Data items converted to xdr at startup, since they are constant
* This is kinda hokey, but may save a little time doing byte swaps
@ -95,7 +101,6 @@ int nfs_pbuf_freecnt = -1; /* start out unlimited */
struct nfs_reqq nfs_reqq;
struct mtx nfs_reqq_mtx;
struct mtx nfs_reply_mtx;
struct nfs_bufq nfs_bufq;
/*
@ -182,6 +187,7 @@ nfsm_rpchead(struct ucred *cr, int nmflag, int procid, int auth_type,
*/
tl = nfsm_build(u_int32_t *, 8 * NFSX_UNSIGNED);
mtx_lock(&nfs_reqq_mtx);
/* Get a pretty random xid to start with */
if (!nfs_xid)
nfs_xid = random();
@ -193,6 +199,7 @@ nfsm_rpchead(struct ucred *cr, int nmflag, int procid, int auth_type,
*xidpp = tl;
*tl++ = txdr_unsigned(nfs_xid);
mtx_unlock(&nfs_reqq_mtx);
*tl++ = rpc_call;
*tl++ = rpc_vers;
*tl++ = txdr_unsigned(NFS_PROG);
@ -416,7 +423,7 @@ nfs_init(struct vfsconf *vfsp)
TAILQ_INIT(&nfs_reqq);
callout_init(&nfs_callout, CALLOUT_MPSAFE);
mtx_init(&nfs_reqq_mtx, "NFS reqq lock", NULL, MTX_DEF);
mtx_init(&nfs_reply_mtx, "Synch NFS reply posting", NULL, MTX_DEF);
mtx_init(&nfs_iod_mtx, "NFS iod lock", NULL, MTX_DEF);
nfs_pbuf_freecnt = nswbuf / 2 + 1;
@ -437,19 +444,80 @@ nfs_uninit(struct vfsconf *vfsp)
* Tell all nfsiod processes to exit. Clear nfs_iodmax, and wakeup
* any sleeping nfsiods so they check nfs_iodmax and exit.
*/
mtx_lock(&nfs_iod_mtx);
nfs_iodmax = 0;
for (i = 0; i < nfs_numasync; i++)
if (nfs_iodwant[i])
wakeup(&nfs_iodwant[i]);
/* The last nfsiod to exit will wake us up when nfs_numasync hits 0 */
while (nfs_numasync)
tsleep(&nfs_numasync, PWAIT, "ioddie", 0);
msleep(&nfs_numasync, &nfs_iod_mtx, PWAIT, "ioddie", 0);
mtx_unlock(&nfs_iod_mtx);
nfs_nhuninit();
uma_zdestroy(nfsmount_zone);
return (0);
}
void
nfs_dircookie_lock(struct nfsnode *np)
{
mtx_lock(&np->n_mtx);
while (np->n_flag & NDIRCOOKIELK)
(void) msleep(&np->n_flag, &np->n_mtx, PZERO, "nfsdirlk", 0);
np->n_flag |= NDIRCOOKIELK;
mtx_unlock(&np->n_mtx);
}
void
nfs_dircookie_unlock(struct nfsnode *np)
{
mtx_lock(&np->n_mtx);
np->n_flag &= ~NDIRCOOKIELK;
wakeup(&np->n_flag);
mtx_unlock(&np->n_mtx);
}
int
nfs_upgrade_vnlock(struct vnode *vp, struct thread *td)
{
int old_lock;
if ((old_lock = VOP_ISLOCKED(vp, td)) != LK_EXCLUSIVE) {
if (old_lock == LK_SHARED) {
/* Upgrade to exclusive lock, this might block */
vn_lock(vp, LK_UPGRADE | LK_RETRY, td);
} else {
vn_lock(vp, LK_EXCLUSIVE | LK_RETRY, td);
}
}
return old_lock;
}
void
nfs_downgrade_vnlock(struct vnode *vp, struct thread *td, int old_lock)
{
if (old_lock != LK_EXCLUSIVE) {
if (old_lock == LK_SHARED) {
/* Downgrade from exclusive lock, this might block */
vn_lock(vp, LK_DOWNGRADE, td);
} else {
VOP_UNLOCK(vp, 0, td);
}
}
}
void
nfs_printf(const char *fmt, ...)
{
va_list ap;
mtx_lock(&Giant);
va_start(ap, fmt);
printf(fmt, ap);
va_end(ap);
mtx_unlock(&Giant);
}
/*
* Attribute cache routines.
* nfs_loadattrcache() - loads or updates the cache contents from attributes
@ -466,7 +534,7 @@ nfs_uninit(struct vfsconf *vfsp)
*/
int
nfs_loadattrcache(struct vnode **vpp, struct mbuf **mdp, caddr_t *dposp,
struct vattr *vaper, int dontshrink)
struct vattr *vaper, int dontshrink)
{
struct vnode *vp = *vpp;
struct vattr *vap;
@ -535,6 +603,7 @@ nfs_loadattrcache(struct vnode **vpp, struct mbuf **mdp, caddr_t *dposp,
* information.
*/
np = VTONFS(vp);
mtx_lock(&np->n_mtx);
if (vp->v_type != vtyp) {
vp->v_type = vtyp;
if (vp->v_type == VFIFO)
@ -617,6 +686,7 @@ nfs_loadattrcache(struct vnode **vpp, struct mbuf **mdp, caddr_t *dposp,
vaper->va_mtime = np->n_mtim;
}
}
mtx_unlock(&np->n_mtx);
return (0);
}
@ -639,16 +709,20 @@ nfs_getattrcache(struct vnode *vp, struct vattr *vaper)
struct vattr *vap;
struct nfsmount *nmp;
int timeo;
np = VTONFS(vp);
vap = &np->n_vattr;
nmp = VFSTONFS(vp->v_mount);
#ifdef NFS_ACDEBUG
mtx_lock(&Giant); /* nfs_printf() */
#endif
mtx_lock(&np->n_mtx);
/* XXX n_mtime doesn't seem to be updated on a miss-and-reload */
timeo = (time_second - np->n_mtime.tv_sec) / 10;
#ifdef NFS_ACDEBUG
if (nfs_acdebug>1)
printf("nfs_getattrcache: initial timeo = %d\n", timeo);
nfs_printf("nfs_getattrcache: initial timeo = %d\n", timeo);
#endif
if (vap->va_type == VDIR) {
@ -665,18 +739,19 @@ nfs_getattrcache(struct vnode *vp, struct vattr *vaper)
#ifdef NFS_ACDEBUG
if (nfs_acdebug > 2)
printf("acregmin %d; acregmax %d; acdirmin %d; acdirmax %d\n",
nmp->nm_acregmin, nmp->nm_acregmax,
nmp->nm_acdirmin, nmp->nm_acdirmax);
nfs_printf("acregmin %d; acregmax %d; acdirmin %d; acdirmax %d\n",
nmp->nm_acregmin, nmp->nm_acregmax,
nmp->nm_acdirmin, nmp->nm_acdirmax);
if (nfs_acdebug)
printf("nfs_getattrcache: age = %d; final timeo = %d\n",
(time_second - np->n_attrstamp), timeo);
nfs_printf("nfs_getattrcache: age = %d; final timeo = %d\n",
(time_second - np->n_attrstamp), timeo);
#endif
if ((time_second - np->n_attrstamp) >= timeo) {
nfsstats.attrcache_misses++;
return (ENOENT);
mtx_unlock(&np->n_mtx);
return( ENOENT);
}
nfsstats.attrcache_hits++;
if (vap->va_size != np->n_size) {
@ -701,6 +776,10 @@ nfs_getattrcache(struct vnode *vp, struct vattr *vaper)
if (np->n_flag & NUPD)
vaper->va_mtime = np->n_mtim;
}
mtx_unlock(&np->n_mtx);
#ifdef NFS_ACDEBUG
mtx_unlock(&Giant); /* nfs_printf() */
#endif
return (0);
}
@ -714,7 +793,8 @@ nfs_getcookie(struct nfsnode *np, off_t off, int add)
{
struct nfsdmap *dp, *dp2;
int pos;
nfsuint64 *retval = NULL;
pos = (uoff_t)off / NFS_DIRBLKSIZ;
if (pos == 0 || off < 0) {
#ifdef DIAGNOSTIC
@ -732,14 +812,14 @@ nfs_getcookie(struct nfsnode *np, off_t off, int add)
dp->ndm_eocookie = 0;
LIST_INSERT_HEAD(&np->n_cookies, dp, ndm_list);
} else
return (NULL);
goto out;
}
while (pos >= NFSNUMCOOKIES) {
pos -= NFSNUMCOOKIES;
if (LIST_NEXT(dp, ndm_list)) {
if (!add && dp->ndm_eocookie < NFSNUMCOOKIES &&
pos >= dp->ndm_eocookie)
return (NULL);
pos >= dp->ndm_eocookie)
goto out;
dp = LIST_NEXT(dp, ndm_list);
} else if (add) {
MALLOC(dp2, struct nfsdmap *, sizeof (struct nfsdmap),
@ -748,15 +828,17 @@ nfs_getcookie(struct nfsnode *np, off_t off, int add)
LIST_INSERT_AFTER(dp, dp2, ndm_list);
dp = dp2;
} else
return (NULL);
goto out;
}
if (pos >= dp->ndm_eocookie) {
if (add)
dp->ndm_eocookie = pos + 1;
else
return (NULL);
goto out;
}
return (&dp->ndm_cookies[pos]);
retval = &dp->ndm_cookies[pos];
out:
return (retval);
}
/*
@ -773,11 +855,13 @@ nfs_invaldir(struct vnode *vp)
if (vp->v_type != VDIR)
panic("nfs: invaldir not dir");
#endif
nfs_dircookie_lock(np);
np->n_direofoffset = 0;
np->n_cookieverf.nfsuquad[0] = 0;
np->n_cookieverf.nfsuquad[1] = 0;
if (LIST_FIRST(&np->n_cookies))
LIST_FIRST(&np->n_cookies)->ndm_eocookie = 0;
nfs_dircookie_unlock(np);
}
/*
@ -797,8 +881,6 @@ nfs_clearcommit(struct mount *mp)
struct buf *bp, *nbp;
int s;
GIANT_REQUIRED;
s = splbio();
MNT_ILOCK(mp);
MNT_VNODE_FOREACH(vp, mp, nvp) {
@ -896,7 +978,7 @@ nfsm_getfh_xx(nfsfh_t **f, int *s, int v3, struct mbuf **md, caddr_t *dpos)
int
nfsm_loadattr_xx(struct vnode **v, struct vattr *va, struct mbuf **md,
caddr_t *dpos)
caddr_t *dpos)
{
int t1;
@ -910,7 +992,7 @@ nfsm_loadattr_xx(struct vnode **v, struct vattr *va, struct mbuf **md,
int
nfsm_postop_attr_xx(struct vnode **v, int *f, struct mbuf **md,
caddr_t *dpos)
caddr_t *dpos)
{
u_int32_t *tl;
int t1;
@ -945,9 +1027,11 @@ nfsm_wcc_data_xx(struct vnode **v, int *f, struct mbuf **md, caddr_t *dpos)
tl = nfsm_dissect_xx(6 * NFSX_UNSIGNED, md, dpos);
if (tl == NULL)
return EBADRPC;
mtx_lock(&(VTONFS(*v))->n_mtx);
if (*f)
ttretf = (VTONFS(*v)->n_mtime.tv_sec == fxdr_unsigned(u_int32_t, *(tl + 2)) &&
VTONFS(*v)->n_mtime.tv_nsec == fxdr_unsigned(u_int32_t, *(tl + 3)));
mtx_unlock(&(VTONFS(*v))->n_mtx);
}
t1 = nfsm_postop_attr_xx(v, &ttattrf, md, dpos);
if (t1)

92
sys/nfsclient/nfs_vfsops.c
View file

@ -35,6 +35,7 @@
#include <sys/cdefs.h>
__FBSDID("$FreeBSD$");
#include "opt_bootp.h"
#include "opt_nfsroot.h"
@ -84,6 +85,7 @@ MALLOC_DEFINE(M_NFSDIRECTIO, "nfsclient_directio", "NFS Direct IO async write st
uma_zone_t nfsmount_zone;
struct nfsstats nfsstats;
SYSCTL_NODE(_vfs, OID_AUTO, nfs, CTLFLAG_RW, 0, "NFS filesystem");
SYSCTL_STRUCT(_vfs_nfs, NFS_NFSSTATS, nfsstats, CTLFLAG_RD,
&nfsstats, nfsstats, "S,nfsstats");
@ -183,7 +185,8 @@ nfs_iosize(struct nfsmount *nmp)
* space.
*/
iosize = max(nmp->nm_rsize, nmp->nm_wsize);
if (iosize < PAGE_SIZE) iosize = PAGE_SIZE;
if (iosize < PAGE_SIZE)
iosize = PAGE_SIZE;
return iosize;
}
@ -257,8 +260,12 @@ nfs_statfs(struct mount *mp, struct statfs *sbp, struct thread *td)
return (error);
}
vp = NFSTOV(np);
if (v3 && (nmp->nm_state & NFSSTA_GOTFSINFO) == 0)
mtx_lock(&nmp->nm_mtx);
if (v3 && (nmp->nm_state & NFSSTA_GOTFSINFO) == 0) {
mtx_unlock(&nmp->nm_mtx);
(void)nfs_fsinfo(nmp, vp, td->td_ucred, td);
} else
mtx_unlock(&nmp->nm_mtx);
nfsstats.rpccnt[NFSPROC_FSSTAT]++;
mreq = nfsm_reqhead(vp, NFSPROC_FSSTAT, NFSX_FH(v3));
mb = mreq;
@ -273,7 +280,9 @@ nfs_statfs(struct mount *mp, struct statfs *sbp, struct thread *td)
goto nfsmout;
}
sfp = nfsm_dissect(struct nfs_statfs *, NFSX_STATFS(v3));
mtx_lock(&nmp->nm_mtx);
sbp->f_iosize = nfs_iosize(nmp);
mtx_unlock(&nmp->nm_mtx);
if (v3) {
sbp->f_bsize = NFS_FABLKSIZE;
tquad = fxdr_hyper(&sfp->sf_tbytes);
@ -314,7 +323,7 @@ nfs_fsinfo(struct nfsmount *nmp, struct vnode *vp, struct ucred *cred,
int error = 0, retattr;
struct mbuf *mreq, *mrep, *md, *mb;
u_int64_t maxfsize;
nfsstats.rpccnt[NFSPROC_FSINFO]++;
mreq = nfsm_reqhead(vp, NFSPROC_FSINFO, NFSX_FH(1));
mb = mreq;
@ -323,6 +332,7 @@ nfs_fsinfo(struct nfsmount *nmp, struct vnode *vp, struct ucred *cred,
nfsm_request(vp, NFSPROC_FSINFO, td, cred);
nfsm_postop_attr(vp, retattr);
if (!error) {
mtx_lock(&nmp->nm_mtx);
fsp = nfsm_dissect(struct nfsv3_fsinfo *, NFSX_V3FSINFO);
pref = fxdr_unsigned(u_int32_t, fsp->fs_wtpref);
if (pref < nmp->nm_wsize && pref >= NFS_FABLKSIZE)
@ -358,6 +368,7 @@ nfs_fsinfo(struct nfsmount *nmp, struct vnode *vp, struct ucred *cred,
nmp->nm_maxfilesize = maxfsize;
nmp->nm_mountp->mnt_stat.f_iosize = nfs_iosize(nmp);
nmp->nm_state |= NFSSTA_GOTFSINFO;
mtx_unlock(&nmp->nm_mtx);
}
m_freem(mrep);
nfsmout:
@ -664,8 +675,7 @@ nfs_decode_args(struct mount *mp, struct nfsmount *nmp, struct nfs_args *argp)
if (nmp->nm_sotype == SOCK_DGRAM)
while (nfs_connect(nmp, NULL)) {
printf("nfs_args: retrying connect\n");
(void) tsleep((caddr_t)&lbolt,
PSOCK, "nfscon", 0);
(void) tsleep((caddr_t)&lbolt, PSOCK, "nfscon", 0);
}
}
}
@ -693,24 +703,31 @@ nfs_mount(struct mount *mp, struct thread *td)
size_t len;
u_char nfh[NFSX_V3FHMAX];
if (vfs_filteropt(mp->mnt_optnew, nfs_opts))
return (EINVAL);
if (vfs_filteropt(mp->mnt_optnew, nfs_opts)) {
error = EINVAL;
goto out;
}
if (mp->mnt_flag & MNT_ROOTFS)
return (nfs_mountroot(mp, td));
if (mp->mnt_flag & MNT_ROOTFS) {
error = nfs_mountroot(mp, td);
goto out;
}
error = vfs_copyopt(mp->mnt_optnew, "nfs_args", &args, sizeof args);
if (error)
return (error);
goto out;
if (args.version != NFS_ARGSVERSION) {
return (EPROGMISMATCH);
error = EPROGMISMATCH;
goto out;
}
if (mp->mnt_flag & MNT_UPDATE) {
struct nfsmount *nmp = VFSTONFS(mp);
if (nmp == NULL)
return (EIO);
if (nmp == NULL) {
error = EIO;
goto out;
}
/*
* When doing an update, we can't change from or to
* v3, switch lockd strategies or change cookie translation
@ -720,7 +737,7 @@ nfs_mount(struct mount *mp, struct thread *td)
(nmp->nm_flag &
(NFSMNT_NFSV3 | NFSMNT_NOLOCKD /*|NFSMNT_XLATECOOKIE*/));
nfs_decode_args(mp, nmp, &args);
return (0);
goto out;
}
/*
@ -734,21 +751,25 @@ nfs_mount(struct mount *mp, struct thread *td)
*/
if (nfs_ip_paranoia == 0)
args.flags |= NFSMNT_NOCONN;
if (args.fhsize < 0 || args.fhsize > NFSX_V3FHMAX)
return (EINVAL);
if (args.fhsize < 0 || args.fhsize > NFSX_V3FHMAX) {
error = EINVAL;
goto out;
}
error = copyin((caddr_t)args.fh, (caddr_t)nfh, args.fhsize);
if (error)
return (error);
goto out;
error = copyinstr(args.hostname, hst, MNAMELEN-1, &len);
if (error)
return (error);
goto out;
bzero(&hst[len], MNAMELEN - len);
/* sockargs() call must be after above copyin() calls */
error = getsockaddr(&nam, (caddr_t)args.addr, args.addrlen);
if (error)
return (error);
goto out;
args.fh = nfh;
error = mountnfs(&args, mp, nam, hst, &vp, td->td_ucred);
mp->mnt_kern_flag |= MNTK_MPSAFE;
out:
return (error);
}
@ -771,12 +792,11 @@ nfs_cmount(struct mntarg *ma, void *data, int flags, struct thread *td)
error = copyin(data, &args, sizeof (struct nfs_args));
if (error)
return (error);
return error;
ma = mount_arg(ma, "nfs_args", &args, sizeof args);
error = kernel_mount(ma, flags);
return (error);
}
@ -805,6 +825,7 @@ mountnfs(struct nfs_args *argp, struct mount *mp, struct sockaddr *nam,
}
vfs_getnewfsid(mp);
nmp->nm_mountp = mp;
mtx_init(&nmp->nm_mtx, "NFSmount lock", NULL, MTX_DEF);
/*
* V2 can only handle 32 bit filesizes. A 4GB-1 limit may be too
@ -851,10 +872,6 @@ mountnfs(struct nfs_args *argp, struct mount *mp, struct sockaddr *nam,
nfs_decode_args(mp, nmp, argp);
if (nmp->nm_sotype == SOCK_STREAM)
mtx_init(&nmp->nm_nfstcpstate.mtx, "NFS/TCP state lock",
NULL, MTX_DEF);
/*
* For Connection based sockets (TCP,...) defer the connect until
* the first request, in case the server is not responding.
@ -869,7 +886,9 @@ mountnfs(struct nfs_args *argp, struct mount *mp, struct sockaddr *nam,
* stuck on a dead server and we are holding a lock on the mount
* point.
*/
mtx_lock(&nmp->nm_mtx);
mp->mnt_stat.f_iosize = nfs_iosize(nmp);
mtx_unlock(&nmp->nm_mtx);
/*
* A reference count is needed on the nfsnode representing the
* remote root. If this object is not persistent, then backward
@ -900,8 +919,7 @@ mountnfs(struct nfs_args *argp, struct mount *mp, struct sockaddr *nam,
return (0);
bad:
if (nmp->nm_sotype == SOCK_STREAM)
mtx_destroy(&nmp->nm_nfstcpstate.mtx);
mtx_destroy(&nmp->nm_mtx);
nfs_disconnect(nmp);
uma_zfree(nfsmount_zone, nmp);
FREE(nam, M_SONAME);
@ -930,12 +948,12 @@ nfs_unmount(struct mount *mp, int mntflags, struct thread *td)
if (flags & FORCECLOSE) {
error = nfs_nmcancelreqs(nmp);
if (error)
return (error);
goto out;
}
/* We hold 1 extra ref on the root vnode; see comment in mountnfs(). */
error = vflush(mp, 1, flags, td);
if (error)
return (error);
goto out;
/*
* We are now committed to the unmount.
@ -943,11 +961,10 @@ nfs_unmount(struct mount *mp, int mntflags, struct thread *td)
nfs_disconnect(nmp);
FREE(nmp->nm_nam, M_SONAME);
if (nmp->nm_sotype == SOCK_STREAM)
mtx_destroy(&nmp->nm_nfstcpstate.mtx);
mtx_destroy(&nmp->nm_mtx);
uma_zfree(nfsmount_zone, nmp);
return (0);
out:
return (error);
}
/*
@ -964,15 +981,18 @@ nfs_root(struct mount *mp, int flags, struct vnode **vpp, struct thread *td)
nmp = VFSTONFS(mp);
error = nfs_nget(mp, (nfsfh_t *)nmp->nm_fh, nmp->nm_fhsize, &np);
if (error)
return (error);
return error;
vp = NFSTOV(np);
/*
* Get transfer parameters and attributes for root vnode once.
*/
mtx_lock(&nmp->nm_mtx);
if ((nmp->nm_state & NFSSTA_GOTFSINFO) == 0 &&
(nmp->nm_flag & NFSMNT_NFSV3)) {
mtx_unlock(&nmp->nm_mtx);
nfs_fsinfo(nmp, vp, curthread->td_ucred, curthread);
}
} else
mtx_unlock(&nmp->nm_mtx);
if (vp->v_type == VNON)
vp->v_type = VDIR;
vp->v_vflag |= VV_ROOT;
@ -1051,8 +1071,10 @@ nfs_sysctl(struct mount *mp, fsctlop_t op, struct sysctl_req *req)
break;
#endif
case VFS_CTL_QUERY:
mtx_lock(&nmp->nm_mtx);
if (nmp->nm_state & NFSSTA_TIMEO)
vq.vq_flags |= VQ_NOTRESP;
mtx_unlock(&nmp->nm_mtx);
#if 0
if (!(nmp->nm_flag & NFSMNT_NOLOCKS) &&
(nmp->nm_state & NFSSTA_LOCKTIMEO))

319
sys/nfsclient/nfs_vnops.c
View file

@ -192,6 +192,7 @@ static int nfs_renameit(struct vnode *sdvp, struct componentname *scnp,
/*
* Global variables
*/
struct mtx nfs_iod_mtx;
struct proc *nfs_iodwant[NFS_MAXASYNCDAEMON];
struct nfsmount *nfs_iodmount[NFS_MAXASYNCDAEMON];
int nfs_numasync = 0;
@ -241,6 +242,23 @@ SYSCTL_INT(_vfs_nfs, OID_AUTO, access_cache_misses, CTLFLAG_RD,
#define NFSV3ACCESS_ALL (NFSV3ACCESS_READ | NFSV3ACCESS_MODIFY \
| NFSV3ACCESS_EXTEND | NFSV3ACCESS_EXECUTE \
| NFSV3ACCESS_DELETE | NFSV3ACCESS_LOOKUP)
/*
* SMP Locking Note :
* The list of locks after the description of the lock is the ordering
* of other locks acquired with the lock held.
* np->n_mtx : Protects the fields in the nfsnode.
VM Object Lock
VI_MTX (acquired indirectly)
* nmp->nm_mtx : Protects the fields in the nfsmount.
rep->r_mtx
* nfs_iod_mtx : Global lock, protects shared nfsiod state.
* nfs_reqq_mtx : Global lock, protects the nfs_reqq list.
nmp->nm_mtx
rep->r_mtx
* rep->r_mtx : Protects the fields in an nfsreq.
*/
static int
nfs3_access_otw(struct vnode *vp, int wmode, struct thread *td,
struct ucred *cred)
@ -266,9 +284,11 @@ nfs3_access_otw(struct vnode *vp, int wmode, struct thread *td,
if (!error) {
tl = nfsm_dissect(u_int32_t *, NFSX_UNSIGNED);
rmode = fxdr_unsigned(u_int32_t, *tl);
mtx_lock(&np->n_mtx);
np->n_mode = rmode;
np->n_modeuid = cred->cr_uid;
np->n_modestamp = time_second;
mtx_unlock(&np->n_mtx);
}
m_freem(mrep);
nfsmout:
@ -343,6 +363,7 @@ nfs_access(struct vop_access_args *ap)
* Does our cached result allow us to give a definite yes to
* this request?
*/
mtx_lock(&np->n_mtx);
if ((time_second < (np->n_modestamp + nfsaccess_cache_timeout)) &&
(ap->a_cred->cr_uid == np->n_modeuid) &&
((np->n_mode & mode) == mode)) {
@ -352,18 +373,21 @@ nfs_access(struct vop_access_args *ap)
* Either a no, or a don't know. Go to the wire.
*/
nfsstats.accesscache_misses++;
mtx_unlock(&np->n_mtx);
error = nfs3_access_otw(vp, wmode, ap->a_td,ap->a_cred);
mtx_lock(&np->n_mtx);
if (!error) {
if ((np->n_mode & mode) != mode) {
error = EACCES;
}
}
}
mtx_unlock(&np->n_mtx);
return (error);
} else {
if ((error = nfsspec_access(ap)) != 0)
if ((error = nfsspec_access(ap)) != 0) {
return (error);
}
/*
* Attempt to prevent a mapped root from accessing a file
* which it shouldn't. We try to read a byte from the file
@ -371,12 +395,14 @@ nfs_access(struct vop_access_args *ap)
* After calling nfsspec_access, we should have the correct
* file size cached.
*/
mtx_lock(&np->n_mtx);
if (ap->a_cred->cr_uid == 0 && (ap->a_mode & VREAD)
&& VTONFS(vp)->n_size > 0) {
struct iovec aiov;
struct uio auio;
char buf[1];
mtx_unlock(&np->n_mtx);
aiov.iov_base = buf;
aiov.iov_len = 1;
auio.uio_iov = &aiov;
@ -400,7 +426,8 @@ nfs_access(struct vop_access_args *ap)
error = nfs_readlinkrpc(vp, &auio, ap->a_cred);
else
error = EACCES;
}
} else
mtx_unlock(&np->n_mtx);
return (error);
}
}
@ -428,7 +455,9 @@ nfs_open(struct vop_open_args *ap)
/*
* Get a valid lease. If cached data is stale, flush it.
*/
mtx_lock(&np->n_mtx);
if (np->n_flag & NMODIFIED) {
mtx_unlock(&np->n_mtx);
error = nfs_vinvalbuf(vp, V_SAVE, ap->a_td, 1);
if (error == EINTR || error == EIO)
return (error);
@ -438,20 +467,28 @@ nfs_open(struct vop_open_args *ap)
error = VOP_GETATTR(vp, &vattr, ap->a_cred, ap->a_td);
if (error)
return (error);
mtx_lock(&np->n_mtx);
np->n_mtime = vattr.va_mtime;
mtx_unlock(&np->n_mtx);
} else {
np->n_attrstamp = 0;
mtx_unlock(&np->n_mtx);
error = VOP_GETATTR(vp, &vattr, ap->a_cred, ap->a_td);
if (error)
return (error);
mtx_lock(&np->n_mtx);
if (NFS_TIMESPEC_COMPARE(&np->n_mtime, &vattr.va_mtime)) {
if (vp->v_type == VDIR)
np->n_direofoffset = 0;
mtx_unlock(&np->n_mtx);
error = nfs_vinvalbuf(vp, V_SAVE, ap->a_td, 1);
if (error == EINTR || error == EIO)
if (error == EINTR || error == EIO) {
return (error);
}
mtx_lock(&np->n_mtx);
np->n_mtime = vattr.va_mtime;
}
mtx_unlock(&np->n_mtx);
}
/*
* If the object has >= 1 O_DIRECT active opens, we disable caching.
@ -461,11 +498,12 @@ nfs_open(struct vop_open_args *ap)
error = nfs_vinvalbuf(vp, V_SAVE, ap->a_td, 1);
if (error)
return (error);
mtx_lock(&np->n_mtx);
np->n_flag |= NNONCACHE;
mtx_unlock(&np->n_mtx);
}
np->n_directio_opens++;
}
np->ra_expect_lbn = 0;
vnode_create_vobject(vp, vattr.va_size, ap->a_td);
return (0);
}
@ -519,7 +557,9 @@ nfs_close(struct vop_close_args *ap)
vm_object_page_clean(vp->v_object, 0, 0, 0);
VM_OBJECT_UNLOCK(vp->v_object);
}
mtx_lock(&np->n_mtx);
if (np->n_flag & NMODIFIED) {
mtx_unlock(&np->n_mtx);
if (NFS_ISV3(vp)) {
/*
* Under NFSv3 we have dirty buffers to dispose of. We
@ -539,6 +579,7 @@ nfs_close(struct vop_close_args *ap)
/* np->n_flag &= ~NMODIFIED; */
} else
error = nfs_vinvalbuf(vp, V_SAVE, ap->a_td, 1);
mtx_lock(&np->n_mtx);
}
/*
* Invalidate the attribute cache in all cases.
@ -551,13 +592,16 @@ nfs_close(struct vop_close_args *ap)
np->n_flag &= ~NWRITEERR;
error = np->n_error;
}
mtx_unlock(&np->n_mtx);
}
if (nfs_directio_enable && (fmode & O_DIRECT) && (vp->v_type == VREG)) {
mtx_lock(&np->n_mtx);
KASSERT((np->n_directio_opens > 0),
("nfs_close: unexpectedly value (0) of n_directio_opens\n"));
("nfs_close: unexpectedly value (0) of n_directio_opens\n"));
np->n_directio_opens--;
if (np->n_directio_opens == 0)
np->n_flag &= ~NNONCACHE;
mtx_unlock(&np->n_mtx);
}
return (error);
}
@ -578,21 +622,21 @@ nfs_getattr(struct vop_getattr_args *ap)
/*
* Update local times for special files.
*/
mtx_lock(&np->n_mtx);
if (np->n_flag & (NACC | NUPD))
np->n_flag |= NCHG;
mtx_unlock(&np->n_mtx);
/*
* First look in the cache.
*/
if (nfs_getattrcache(vp, ap->a_vap) == 0)
return (0);
goto nfsmout;
if (v3 && nfsaccess_cache_timeout > 0) {
nfsstats.accesscache_misses++;
nfs3_access_otw(vp, NFSV3ACCESS_ALL, ap->a_td, ap->a_cred);
if (nfs_getattrcache(vp, ap->a_vap) == 0)
return (0);
goto nfsmout;
}
nfsstats.rpccnt[NFSPROC_GETATTR]++;
mreq = nfsm_reqhead(vp, NFSPROC_GETATTR, NFSX_FH(v3));
mb = mreq;
@ -635,8 +679,10 @@ nfs_setattr(struct vop_setattr_args *ap)
if ((vap->va_flags != VNOVAL || vap->va_uid != (uid_t)VNOVAL ||
vap->va_gid != (gid_t)VNOVAL || vap->va_atime.tv_sec != VNOVAL ||
vap->va_mtime.tv_sec != VNOVAL || vap->va_mode != (mode_t)VNOVAL) &&
(vp->v_mount->mnt_flag & MNT_RDONLY))
return (EROFS);
(vp->v_mount->mnt_flag & MNT_RDONLY)) {
error = EROFS;
goto out;
}
if (vap->va_size != VNOVAL) {
switch (vp->v_type) {
case VDIR:
@ -650,7 +696,7 @@ nfs_setattr(struct vop_setattr_args *ap)
vap->va_mode == (mode_t)VNOVAL &&
vap->va_uid == (uid_t)VNOVAL &&
vap->va_gid == (gid_t)VNOVAL)
return (0);
return (0);
vap->va_size = VNOVAL;
break;
default:
@ -660,47 +706,60 @@ nfs_setattr(struct vop_setattr_args *ap)
*/
if (vp->v_mount->mnt_flag & MNT_RDONLY)
return (EROFS);
/*
* We run vnode_pager_setsize() early (why?),
* we must set np->n_size now to avoid vinvalbuf
* V_SAVE races that might setsize a lower
* value.
*/
mtx_lock(&np->n_mtx);
tsize = np->n_size;
mtx_unlock(&np->n_mtx);
error = nfs_meta_setsize(vp, ap->a_cred,
ap->a_td, vap->va_size);
ap->a_td, vap->va_size);
mtx_lock(&np->n_mtx);
if (np->n_flag & NMODIFIED) {
tsize = np->n_size;
mtx_unlock(&np->n_mtx);
if (vap->va_size == 0)
error = nfs_vinvalbuf(vp, 0, ap->a_td, 1);
else
error = nfs_vinvalbuf(vp, V_SAVE, ap->a_td, 1);
if (error) {
vnode_pager_setsize(vp, np->n_size);
return (error);
vnode_pager_setsize(vp, tsize);
goto out;
}
}
} else
mtx_unlock(&np->n_mtx);
/*
* np->n_size has already been set to vap->va_size
* in nfs_meta_setsize(). We must set it again since
* nfs_loadattrcache() could be called through
* nfs_meta_setsize() and could modify np->n_size.
*/
mtx_lock(&np->n_mtx);
np->n_vattr.va_size = np->n_size = vap->va_size;
mtx_unlock(&np->n_mtx);
};
} else if ((vap->va_mtime.tv_sec != VNOVAL ||
vap->va_atime.tv_sec != VNOVAL) && (np->n_flag & NMODIFIED) &&
vp->v_type == VREG &&
(error = nfs_vinvalbuf(vp, V_SAVE, ap->a_td, 1)) != 0 &&
(error == EINTR || error == EIO))
return (error);
} else {
mtx_lock(&np->n_mtx);
if ((vap->va_mtime.tv_sec != VNOVAL || vap->va_atime.tv_sec != VNOVAL) &&
(np->n_flag & NMODIFIED) && vp->v_type == VREG) {
mtx_unlock(&np->n_mtx);
if ((error = nfs_vinvalbuf(vp, V_SAVE, ap->a_td, 1)) != 0 &&
(error == EINTR || error == EIO))
return error;
} else
mtx_unlock(&np->n_mtx);
}
error = nfs_setattrrpc(vp, vap, ap->a_cred, ap->a_td);
if (error && vap->va_size != VNOVAL) {
mtx_lock(&np->n_mtx);
np->n_size = np->n_vattr.va_size = tsize;
vnode_pager_setsize(vp, np->n_size);
vnode_pager_setsize(vp, tsize);
mtx_unlock(&np->n_mtx);
}
out:
return (error);
}
@ -779,7 +838,7 @@ nfs_lookup(struct vop_lookup_args *ap)
int error = 0, attrflag, fhsize;
int v3 = NFS_ISV3(dvp);
struct thread *td = cnp->cn_thread;
*vpp = NULLVP;
if ((flags & ISLASTCN) && (dvp->v_mount->mnt_flag & MNT_RDONLY) &&
(cnp->cn_nameiop == DELETE || cnp->cn_nameiop == RENAME))
@ -964,8 +1023,10 @@ nfs_readlinkrpc(struct vnode *vp, struct uio *uiop, struct ucred *cred)
nfsm_strsiz(len, NFS_MAXPATHLEN);
if (len == NFS_MAXPATHLEN) {
struct nfsnode *np = VTONFS(vp);
mtx_lock(&np->n_mtx);
if (np->n_size && np->n_size < NFS_MAXPATHLEN)
len = np->n_size;
mtx_unlock(&np->n_mtx);
}
nfsm_mtouio(uiop, len);
}
@ -987,17 +1048,23 @@ nfs_readrpc(struct vnode *vp, struct uio *uiop, struct ucred *cred)
struct nfsmount *nmp;
int error = 0, len, retlen, tsiz, eof, attrflag;
int v3 = NFS_ISV3(vp);
int rsize;
#ifndef nolint
eof = 0;
#endif
nmp = VFSTONFS(vp->v_mount);
tsiz = uiop->uio_resid;
if (uiop->uio_offset + tsiz > nmp->nm_maxfilesize)
mtx_lock(&nmp->nm_mtx);
if (uiop->uio_offset + tsiz > nmp->nm_maxfilesize) {
mtx_unlock(&nmp->nm_mtx);
return (EFBIG);
}
rsize = nmp->nm_rsize;
mtx_unlock(&nmp->nm_mtx);
while (tsiz > 0) {
nfsstats.rpccnt[NFSPROC_READ]++;
len = (tsiz > nmp->nm_rsize) ? nmp->nm_rsize : tsiz;
len = (tsiz > rsize) ? rsize : tsiz;
mreq = nfsm_reqhead(vp, NFSPROC_READ, NFSX_FH(v3) + NFSX_UNSIGNED * 3);
mb = mreq;
bpos = mtod(mb, caddr_t);
@ -1020,9 +1087,10 @@ nfs_readrpc(struct vnode *vp, struct uio *uiop, struct ucred *cred)
}
tl = nfsm_dissect(u_int32_t *, 2 * NFSX_UNSIGNED);
eof = fxdr_unsigned(int, *(tl + 1));
} else
} else {
nfsm_loadattr(vp, NULL);
nfsm_strsiz(retlen, nmp->nm_rsize);
}
nfsm_strsiz(retlen, rsize);
nfsm_mtouio(uiop, retlen);
m_freem(mrep);
tsiz -= retlen;
@ -1043,7 +1111,7 @@ nfsmout:
*/
int
nfs_writerpc(struct vnode *vp, struct uio *uiop, struct ucred *cred,
int *iomode, int *must_commit)
int *iomode, int *must_commit)
{
u_int32_t *tl;
int32_t backup;
@ -1052,18 +1120,24 @@ nfs_writerpc(struct vnode *vp, struct uio *uiop, struct ucred *cred,
struct nfsmount *nmp = VFSTONFS(vp->v_mount);
int error = 0, len, tsiz, wccflag = NFSV3_WCCRATTR, rlen, commit;
int v3 = NFS_ISV3(vp), committed = NFSV3WRITE_FILESYNC;
int wsize;
#ifndef DIAGNOSTIC
if (uiop->uio_iovcnt != 1)
panic("nfs: writerpc iovcnt > 1");
#endif
*must_commit = 0;
tsiz = uiop->uio_resid;
if (uiop->uio_offset + tsiz > nmp->nm_maxfilesize)
mtx_lock(&nmp->nm_mtx);
if (uiop->uio_offset + tsiz > nmp->nm_maxfilesize) {
mtx_unlock(&nmp->nm_mtx);
return (EFBIG);
}
wsize = nmp->nm_wsize;
mtx_unlock(&nmp->nm_mtx);
while (tsiz > 0) {
nfsstats.rpccnt[NFSPROC_WRITE]++;
len = (tsiz > nmp->nm_wsize) ? nmp->nm_wsize : tsiz;
len = (tsiz > wsize) ? wsize : tsiz;
mreq = nfsm_reqhead(vp, NFSPROC_WRITE,
NFSX_FH(v3) + 5 * NFSX_UNSIGNED + nfsm_rndup(len));
mb = mreq;
@ -1122,6 +1196,7 @@ nfs_writerpc(struct vnode *vp, struct uio *uiop, struct ucred *cred,
else if (committed == NFSV3WRITE_DATASYNC &&
commit == NFSV3WRITE_UNSTABLE)
committed = commit;
mtx_lock(&nmp->nm_mtx);
if ((nmp->nm_state & NFSSTA_HASWRITEVERF) == 0){
bcopy((caddr_t)tl, (caddr_t)nmp->nm_verf,
NFSX_V3WRITEVERF);
@ -1132,11 +1207,16 @@ nfs_writerpc(struct vnode *vp, struct uio *uiop, struct ucred *cred,
bcopy((caddr_t)tl, (caddr_t)nmp->nm_verf,
NFSX_V3WRITEVERF);
}
mtx_unlock(&nmp->nm_mtx);
}
} else
nfsm_loadattr(vp, NULL);
if (wccflag)
VTONFS(vp)->n_mtime = VTONFS(vp)->n_vattr.va_mtime;
} else {
nfsm_loadattr(vp, NULL);
}
if (wccflag) {
mtx_lock(&(VTONFS(vp))->n_mtx);
VTONFS(vp)->n_mtime = VTONFS(vp)->n_vattr.va_mtime;
mtx_unlock(&(VTONFS(vp))->n_mtx);
}
m_freem(mrep);
if (error)
break;
@ -1232,9 +1312,11 @@ nfsmout:
cache_enter(dvp, newvp, cnp);
*vpp = newvp;
}
mtx_lock(&(VTONFS(dvp))->n_mtx);
VTONFS(dvp)->n_flag |= NMODIFIED;
if (!wccflag)
VTONFS(dvp)->n_attrstamp = 0;
mtx_unlock(&(VTONFS(dvp))->n_mtx);
return (error);
}
@ -1246,7 +1328,6 @@ nfsmout:
static int
nfs_mknod(struct vop_mknod_args *ap)
{
return (nfs_mknodrpc(ap->a_dvp, ap->a_vpp, ap->a_cnp, ap->a_vap));
}
@ -1359,9 +1440,11 @@ nfsmout:
cache_enter(dvp, newvp, cnp);
*ap->a_vpp = newvp;
}
mtx_lock(&(VTONFS(dvp))->n_mtx);
VTONFS(dvp)->n_flag |= NMODIFIED;
if (!wccflag)
VTONFS(dvp)->n_attrstamp = 0;
mtx_unlock(&(VTONFS(dvp))->n_mtx);
return (error);
}
@ -1434,7 +1517,6 @@ nfs_remove(struct vop_remove_args *ap)
int
nfs_removeit(struct sillyrename *sp)
{
/*
* Make sure that the directory vnode is still valid.
* XXX we should lock sp->s_dvp here.
@ -1469,9 +1551,11 @@ nfs_removerpc(struct vnode *dvp, const char *name, int namelen,
nfsm_wcc_data(dvp, wccflag);
m_freem(mrep);
nfsmout:
mtx_lock(&(VTONFS(dvp))->n_mtx);
VTONFS(dvp)->n_flag |= NMODIFIED;
if (!wccflag)
VTONFS(dvp)->n_attrstamp = 0;
mtx_unlock(&(VTONFS(dvp))->n_mtx);
return (error);
}
@ -1502,7 +1586,7 @@ nfs_rename(struct vop_rename_args *ap)
}
if (fvp == tvp) {
printf("nfs_rename: fvp == tvp (can't happen)\n");
nfs_printf("nfs_rename: fvp == tvp (can't happen)\n");
error = 0;
goto out;
}
@ -1609,8 +1693,12 @@ nfs_renamerpc(struct vnode *fdvp, const char *fnameptr, int fnamelen,
}
m_freem(mrep);
nfsmout:
mtx_lock(&(VTONFS(fdvp))->n_mtx);
VTONFS(fdvp)->n_flag |= NMODIFIED;
mtx_unlock(&(VTONFS(fdvp))->n_mtx);
mtx_lock(&(VTONFS(tdvp))->n_mtx);
VTONFS(tdvp)->n_flag |= NMODIFIED;
mtx_unlock(&(VTONFS(tdvp))->n_mtx);
if (!fwccflag)
VTONFS(fdvp)->n_attrstamp = 0;
if (!twccflag)
@ -1659,7 +1747,9 @@ nfs_link(struct vop_link_args *ap)
}
m_freem(mrep);
nfsmout:
mtx_lock(&(VTONFS(tdvp))->n_mtx);
VTONFS(tdvp)->n_flag |= NMODIFIED;
mtx_unlock(&(VTONFS(tdvp))->n_mtx);
if (!attrflag)
VTONFS(vp)->n_attrstamp = 0;
if (!wccflag)
@ -1758,7 +1848,9 @@ nfsmout:
} else {
*ap->a_vpp = newvp;
}
mtx_lock(&(VTONFS(dvp))->n_mtx);
VTONFS(dvp)->n_flag |= NMODIFIED;
mtx_unlock(&(VTONFS(dvp))->n_mtx);
if (!wccflag)
VTONFS(dvp)->n_attrstamp = 0;
return (error);
@ -1813,7 +1905,9 @@ nfs_mkdir(struct vop_mkdir_args *ap)
nfsm_wcc_data(dvp, wccflag);
m_freem(mrep);
nfsmout:
mtx_lock(&(VTONFS(dvp))->n_mtx);
VTONFS(dvp)->n_flag |= NMODIFIED;
mtx_unlock(&(VTONFS(dvp))->n_mtx);
if (!wccflag)
VTONFS(dvp)->n_attrstamp = 0;
/*
@ -1869,7 +1963,9 @@ nfs_rmdir(struct vop_rmdir_args *ap)
nfsm_wcc_data(dvp, wccflag);
m_freem(mrep);
nfsmout:
mtx_lock(&(VTONFS(dvp))->n_mtx);
VTONFS(dvp)->n_flag |= NMODIFIED;
mtx_unlock(&(VTONFS(dvp))->n_mtx);
if (!wccflag)
VTONFS(dvp)->n_attrstamp = 0;
cache_purge(dvp);
@ -1891,20 +1987,25 @@ nfs_readdir(struct vop_readdir_args *ap)
struct vnode *vp = ap->a_vp;
struct nfsnode *np = VTONFS(vp);
struct uio *uio = ap->a_uio;
int tresid, error;
int tresid, error = 0;
struct vattr vattr;
if (vp->v_type != VDIR)
return(EPERM);
if (vp->v_type != VDIR)
return (EPERM);
/*
* First, check for hit on the EOF offset cache
*/
if (np->n_direofoffset > 0 && uio->uio_offset >= np->n_direofoffset &&
(np->n_flag & NMODIFIED) == 0) {
if (VOP_GETATTR(vp, &vattr, ap->a_cred, uio->uio_td) == 0 &&
!NFS_TIMESPEC_COMPARE(&np->n_mtime, &vattr.va_mtime)) {
nfsstats.direofcache_hits++;
return (0);
if (VOP_GETATTR(vp, &vattr, ap->a_cred, uio->uio_td) == 0) {
mtx_lock(&np->n_mtx);
if (!NFS_TIMESPEC_COMPARE(&np->n_mtime, &vattr.va_mtime)) {
mtx_unlock(&np->n_mtx);
nfsstats.direofcache_hits++;
goto out;
} else
mtx_unlock(&np->n_mtx);
}
}
@ -1914,8 +2015,10 @@ nfs_readdir(struct vop_readdir_args *ap)
tresid = uio->uio_resid;
error = nfs_bioread(vp, uio, 0, ap->a_cred);
if (!error && uio->uio_resid == tresid)
if (!error && uio->uio_resid == tresid) {
nfsstats.direofcache_misses++;
}
out:
return (error);
}
@ -1950,11 +2053,16 @@ nfs_readdirrpc(struct vnode *vp, struct uio *uiop, struct ucred *cred)
/*
* If there is no cookie, assume directory was stale.
*/
nfs_dircookie_lock(dnp);
cookiep = nfs_getcookie(dnp, uiop->uio_offset, 0);
if (cookiep)
if (cookiep) {
cookie = *cookiep;
else
nfs_dircookie_unlock(dnp);
} else {
nfs_dircookie_unlock(dnp);
return (NFSERR_BAD_COOKIE);
}
/*
* Loop around doing readdir rpc's of size nm_readdirsize
* truncated to a multiple of DIRBLKSIZ.
@ -1971,8 +2079,10 @@ nfs_readdirrpc(struct vnode *vp, struct uio *uiop, struct ucred *cred)
tl = nfsm_build(u_int32_t *, 5 * NFSX_UNSIGNED);
*tl++ = cookie.nfsuquad[0];
*tl++ = cookie.nfsuquad[1];
mtx_lock(&dnp->n_mtx);
*tl++ = dnp->n_cookieverf.nfsuquad[0];
*tl++ = dnp->n_cookieverf.nfsuquad[1];
mtx_unlock(&dnp->n_mtx);
} else {
tl = nfsm_build(u_int32_t *, 2 * NFSX_UNSIGNED);
*tl++ = cookie.nfsuquad[0];
@ -1984,8 +2094,10 @@ nfs_readdirrpc(struct vnode *vp, struct uio *uiop, struct ucred *cred)
if (!error) {
tl = nfsm_dissect(u_int32_t *,
2 * NFSX_UNSIGNED);
mtx_lock(&dnp->n_mtx);
dnp->n_cookieverf.nfsuquad[0] = *tl++;
dnp->n_cookieverf.nfsuquad[1] = *tl;
mtx_unlock(&dnp->n_mtx);
} else {
m_freem(mrep);
goto nfsmout;
@ -2100,9 +2212,11 @@ nfs_readdirrpc(struct vnode *vp, struct uio *uiop, struct ucred *cred)
dnp->n_direofoffset = uiop->uio_offset;
else {
if (uiop->uio_resid > 0)
printf("EEK! readdirrpc resid > 0\n");
nfs_printf("EEK! readdirrpc resid > 0\n");
nfs_dircookie_lock(dnp);
cookiep = nfs_getcookie(dnp, uiop->uio_offset, 1);
*cookiep = cookie;
nfs_dircookie_unlock(dnp);
}
nfsmout:
return (error);
@ -2146,11 +2260,15 @@ nfs_readdirplusrpc(struct vnode *vp, struct uio *uiop, struct ucred *cred)
/*
* If there is no cookie, assume directory was stale.
*/
nfs_dircookie_lock(dnp);
cookiep = nfs_getcookie(dnp, uiop->uio_offset, 0);
if (cookiep)
if (cookiep) {
cookie = *cookiep;
else
nfs_dircookie_unlock(dnp);
} else {
nfs_dircookie_unlock(dnp);
return (NFSERR_BAD_COOKIE);
}
/*
* Loop around doing readdir rpc's of size nm_readdirsize
* truncated to a multiple of DIRBLKSIZ.
@ -2166,8 +2284,10 @@ nfs_readdirplusrpc(struct vnode *vp, struct uio *uiop, struct ucred *cred)
tl = nfsm_build(u_int32_t *, 6 * NFSX_UNSIGNED);
*tl++ = cookie.nfsuquad[0];
*tl++ = cookie.nfsuquad[1];
mtx_lock(&dnp->n_mtx);
*tl++ = dnp->n_cookieverf.nfsuquad[0];
*tl++ = dnp->n_cookieverf.nfsuquad[1];
mtx_unlock(&dnp->n_mtx);
*tl++ = txdr_unsigned(nmp->nm_readdirsize);
*tl = txdr_unsigned(nmp->nm_rsize);
nfsm_request(vp, NFSPROC_READDIRPLUS, uiop->uio_td, cred);
@ -2177,8 +2297,10 @@ nfs_readdirplusrpc(struct vnode *vp, struct uio *uiop, struct ucred *cred)
goto nfsmout;
}
tl = nfsm_dissect(u_int32_t *, 3 * NFSX_UNSIGNED);
mtx_lock(&dnp->n_mtx);
dnp->n_cookieverf.nfsuquad[0] = *tl++;
dnp->n_cookieverf.nfsuquad[1] = *tl++;
mtx_unlock(&dnp->n_mtx);
more_dirs = fxdr_unsigned(int, *tl);
/* loop thru the dir entries, doctoring them to 4bsd form */
@ -2313,9 +2435,9 @@ nfs_readdirplusrpc(struct vnode *vp, struct uio *uiop, struct ucred *cred)
tl = nfsm_dissect(u_int32_t *, NFSX_UNSIGNED);
i = fxdr_unsigned(int, *tl);
if (i) {
tl = nfsm_dissect(u_int32_t *, NFSX_UNSIGNED);
fhsize = fxdr_unsigned(int, *tl);
nfsm_adv(nfsm_rndup(fhsize));
tl = nfsm_dissect(u_int32_t *, NFSX_UNSIGNED);
fhsize = fxdr_unsigned(int, *tl);
nfsm_adv(nfsm_rndup(fhsize));
}
}
if (newvp != NULLVP) {
@ -2359,9 +2481,11 @@ nfs_readdirplusrpc(struct vnode *vp, struct uio *uiop, struct ucred *cred)
dnp->n_direofoffset = uiop->uio_offset;
else {
if (uiop->uio_resid > 0)
printf("EEK! readdirplusrpc resid > 0\n");
nfs_printf("EEK! readdirplusrpc resid > 0\n");
nfs_dircookie_lock(dnp);
cookiep = nfs_getcookie(dnp, uiop->uio_offset, 1);
*cookiep = cookie;
nfs_dircookie_unlock(dnp);
}
nfsmout:
if (newvp != NULLVP) {
@ -2521,7 +2645,7 @@ nfsmout:
*/
int
nfs_commit(struct vnode *vp, u_quad_t offset, int cnt, struct ucred *cred,
struct thread *td)
struct thread *td)
{
u_int32_t *tl;
struct nfsmount *nmp = VFSTONFS(vp->v_mount);
@ -2529,8 +2653,12 @@ nfs_commit(struct vnode *vp, u_quad_t offset, int cnt, struct ucred *cred,
int error = 0, wccflag = NFSV3_WCCRATTR;
struct mbuf *mreq, *mrep, *md, *mb;
if ((nmp->nm_state & NFSSTA_HASWRITEVERF) == 0)
mtx_lock(&nmp->nm_mtx);
if ((nmp->nm_state & NFSSTA_HASWRITEVERF) == 0) {
mtx_unlock(&nmp->nm_mtx);
return (0);
}
mtx_unlock(&nmp->nm_mtx);
nfsstats.rpccnt[NFSPROC_COMMIT]++;
mreq = nfsm_reqhead(vp, NFSPROC_COMMIT, NFSX_FH(1));
mb = mreq;
@ -2600,7 +2728,6 @@ nfs_strategy(struct vop_strategy_args *ap)
static int
nfs_fsync(struct vop_fsync_args *ap)
{
return (nfs_flush(ap->a_vp, ap->a_waitfor, ap->a_td, 1));
}
@ -2821,8 +2948,10 @@ loop:
LK_EXCLUSIVE | LK_SLEEPFAIL | LK_INTERLOCK,
VI_MTX(vp), "nfsfsync", slpflag, slptimeo);
splx(s);
if (error == 0)
panic("nfs_fsync: inconsistent lock");
if (error == 0) {
BUF_UNLOCK(bp);
goto loop;
}
if (error == ENOLCK)
goto loop;
if (nfs_sigintr(nmp, NULL, td)) {
@ -2880,23 +3009,28 @@ loop:
VI_UNLOCK(vp);
goto loop;
}
/*
* Wait for all the async IO requests to drain
/*
* Wait for all the async IO requests to drain
*/
VI_UNLOCK(vp);
mtx_lock(&np->n_mtx);
while (np->n_directio_asyncwr > 0) {
np->n_flag |= NFSYNCWAIT;
error = nfs_tsleep(td, (caddr_t)&np->n_directio_asyncwr,
slpflag | (PRIBIO + 1), "nfsfsync", 0);
error = nfs_msleep(td, (caddr_t)&np->n_directio_asyncwr,
&np->n_mtx, slpflag | (PRIBIO + 1),
"nfsfsync", 0);
if (error) {
if (nfs_sigintr(nmp, (struct nfsreq *)0, td)) {
error = EINTR;
mtx_unlock(&np->n_mtx);
error = EINTR;
goto done;
}
}
}
}
VI_UNLOCK(vp);
mtx_unlock(&np->n_mtx);
} else
VI_UNLOCK(vp);
mtx_lock(&np->n_mtx);
if (np->n_flag & NWRITEERR) {
error = np->n_error;
np->n_flag &= ~NWRITEERR;
@ -2904,6 +3038,7 @@ loop:
if (commit && vp->v_bufobj.bo_dirty.bv_cnt == 0 &&
vp->v_bufobj.bo_numoutput == 0 && np->n_directio_asyncwr == 0)
np->n_flag &= ~NMODIFIED;
mtx_unlock(&np->n_mtx);
done:
if (bvec != NULL && bvec != bvec_on_stack)
free(bvec, M_TEMP);
@ -2916,13 +3051,19 @@ done:
static int
nfs_advlock(struct vop_advlock_args *ap)
{
int error;
mtx_lock(&Giant);
if ((VFSTONFS(ap->a_vp->v_mount)->nm_flag & NFSMNT_NOLOCKD) != 0) {
struct nfsnode *np = VTONFS(ap->a_vp);
return (lf_advlock(ap, &(np->n_lockf), np->n_size));
error = lf_advlock(ap, &(np->n_lockf), np->n_size);
goto out;
}
return (nfs_dolock(ap));
error = nfs_dolock(ap);
out:
mtx_unlock(&Giant);
return (error);
}
/*
@ -2934,7 +3075,7 @@ nfs_print(struct vop_print_args *ap)
struct vnode *vp = ap->a_vp;
struct nfsnode *np = VTONFS(vp);
printf("\tfileid %ld fsid 0x%x",
nfs_printf("\tfileid %ld fsid 0x%x",
np->n_vattr.va_fileid, np->n_vattr.va_fsid);
if (vp->v_type == VFIFO)
fifo_printinfo(vp);
@ -2998,7 +3139,6 @@ nfs_writebp(struct buf *bp, int force __unused, struct thread *td)
reassignbuf(bp);
splx(s);
}
brelse(bp);
return (rtval);
}
@ -3039,9 +3179,11 @@ nfsspec_access(struct vop_access_args *ap)
vap = &vattr;
error = VOP_GETATTR(vp, vap, cred, ap->a_td);
if (error)
return (error);
return (vaccess(vp->v_type, vap->va_mode, vap->va_uid, vap->va_gid,
mode, cred, NULL));
goto out;
error = vaccess(vp->v_type, vap->va_mode, vap->va_uid, vap->va_gid,
mode, cred, NULL);
out:
return error;
}
/*
@ -3051,13 +3193,17 @@ static int
nfsfifo_read(struct vop_read_args *ap)
{
struct nfsnode *np = VTONFS(ap->a_vp);
int error;
/*
* Set access flag.
*/
mtx_lock(&np->n_mtx);
np->n_flag |= NACC;
getnanotime(&np->n_atim);
return (fifo_specops.vop_read(ap));
mtx_unlock(&np->n_mtx);
error = fifo_specops.vop_read(ap);
return error;
}
/*
@ -3071,9 +3217,11 @@ nfsfifo_write(struct vop_write_args *ap)
/*
* Set update flag.
*/
mtx_lock(&np->n_mtx);
np->n_flag |= NUPD;
getnanotime(&np->n_mtim);
return (fifo_specops.vop_write(ap));
mtx_unlock(&np->n_mtx);
return(fifo_specops.vop_write(ap));
}
/*
@ -3089,6 +3237,7 @@ nfsfifo_close(struct vop_close_args *ap)
struct vattr vattr;
struct timespec ts;
mtx_lock(&np->n_mtx);
if (np->n_flag & (NACC | NUPD)) {
getnanotime(&ts);
if (np->n_flag & NACC)
@ -3103,9 +3252,13 @@ nfsfifo_close(struct vop_close_args *ap)
vattr.va_atime = np->n_atim;
if (np->n_flag & NUPD)
vattr.va_mtime = np->n_mtim;
mtx_unlock(&np->n_mtx);
(void)VOP_SETATTR(vp, &vattr, ap->a_cred, ap->a_td);
goto out;
}
}
mtx_unlock(&np->n_mtx);
out:
return (fifo_specops.vop_close(ap));
}

2
sys/nfsclient/nfsmount.h
View file

@ -41,7 +41,6 @@ struct nfs_tcp_mountstate {
#define NFS_TCP_EXPECT_RPCMARKER 0x0001 /* Expect to see a RPC/TCP marker next */
#define NFS_TCP_FORCE_RECONNECT 0x0002 /* Force a TCP reconnect */
int flags;
struct mtx mtx;
};
/*
@ -50,6 +49,7 @@ struct nfs_tcp_mountstate {
* Holds NFS specific information for mount.
*/
struct nfsmount {
struct mtx nm_mtx;
int nm_flag; /* Flags for soft/hard... */
int nm_state; /* Internal state flags */
struct mount *nm_mountp; /* Vfs structure for this filesystem */

15
sys/nfsclient/nfsnode.h
View file

@ -88,6 +88,7 @@ struct nfsdmap {
* be well aligned and, therefore, tightly packed.
*/
struct nfsnode {
struct mtx n_mtx; /* Protects all of these members */
u_quad_t n_size; /* Current size of file */
u_quad_t n_brev; /* Modify rev when cached */
u_quad_t n_lrev; /* Modify rev for lease */
@ -124,9 +125,8 @@ struct nfsnode {
struct nfs4_fctx n_wfc;
u_char *n_name; /* leaf name, for v4 OPEN op */
uint32_t n_namelen;
daddr_t ra_expect_lbn;
int n_directio_opens;
int n_directio_asyncwr;
int n_directio_asyncwr;
};
#define n_atim n_un1.nf_atim
@ -140,6 +140,8 @@ struct nfsnode {
/*
* Flags for n_flag
*/
#define NFSYNCWAIT 0x0002 /* fsync waiting for all directio async writes
to drain */
#define NMODIFIED 0x0004 /* Might have a modified buffer in bio */
#define NWRITEERR 0x0008 /* Flag write errors so close will know */
/* 0x20, 0x40, 0x80 free */
@ -150,8 +152,7 @@ struct nfsnode {
#define NTRUNCATE 0x1000 /* Opened by nfs_setattr() */
#define NSIZECHANGED 0x2000 /* File size has changed: need cache inval */
#define NNONCACHE 0x4000 /* Node marked as noncacheable */
#define NFSYNCWAIT 0x8000 /* fsync waiting for all directio async writes
to drain */
#define NDIRCOOKIELK 0x8000 /* Lock to serialize access to directory cookies */
/*
* Convert between nfsnode pointers and vnode pointers
@ -193,6 +194,12 @@ nfsuint64 *nfs_getcookie(struct nfsnode *, off_t, int);
uint64_t *nfs4_getcookie(struct nfsnode *, off_t, int);
void nfs_invaldir(struct vnode *);
void nfs4_invaldir(struct vnode *);
int nfs_upgrade_vnlock(struct vnode *vp, struct thread *td);
void nfs_downgrade_vnlock(struct vnode *vp, struct thread *td, int old_lock);
void nfs_printf(const char *fmt, ...);
void nfs_dircookie_lock(struct nfsnode *np);
void nfs_dircookie_unlock(struct nfsnode *np);
#endif /* _KERNEL */