/*-
* Copyright (c) 2008, 2009, 2023 The NetBSD Foundation, Inc.
* All rights reserved.
*
* This code is derived from software contributed to The NetBSD Foundation
* by Andrew Doran.
*
* Redistribution and use in source and binary forms, with or without
* modification, are permitted provided that the following conditions
* are met:
* 1. Redistributions of source code must retain the above copyright
* notice, this list of conditions and the following disclaimer.
* 2. Redistributions in binary form must reproduce the above copyright
* notice, this list of conditions and the following disclaimer in the
* documentation and/or other materials provided with the distribution.
*
* THIS SOFTWARE IS PROVIDED BY THE NETBSD FOUNDATION, INC. AND CONTRIBUTORS
* ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED
* TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR
* PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE FOUNDATION OR CONTRIBUTORS
* BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
* CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
* SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
* INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
* CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
* ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
* POSSIBILITY OF SUCH DAMAGE.
*/
/*
* Copyright (c) 1982, 1986, 1989, 1991, 1993
* The Regents of the University of California. All rights reserved.
* (c) UNIX System Laboratories, Inc.
* All or some portions of this file are derived from material licensed
* to the University of California by American Telephone and Telegraph
* Co. or Unix System Laboratories, Inc. and are reproduced herein with
* the permission of UNIX System Laboratories, Inc.
*
* Redistribution and use in source and binary forms, with or without
* modification, are permitted provided that the following conditions
* are met:
* 1. Redistributions of source code must retain the above copyright
* notice, this list of conditions and the following disclaimer.
* 2. Redistributions in binary form must reproduce the above copyright
* notice, this list of conditions and the following disclaimer in the
* documentation and/or other materials provided with the distribution.
* 3. Neither the name of the University nor the names of its contributors
* may be used to endorse or promote products derived from this software
* without specific prior written permission.
*
* THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
* ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
* IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
* ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
* FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
* DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
* OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
* HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
* LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
* OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
* SUCH DAMAGE.
*
* @(#)kern_descrip.c 8.8 (Berkeley) 2/14/95
*/
/*
* A list (head) of open files, counter, and lock protecting them.
*/
struct filelist filehead __cacheline_aligned;
static u_int nfiles __cacheline_aligned;
kmutex_t filelist_lock __cacheline_aligned;
/*
* Look up the file structure corresponding to a file descriptor
* and return the file, holding a reference on the descriptor.
*/
file_t *
fd_getfile(unsigned fd)
{
filedesc_t *fdp;
fdfile_t *ff;
file_t *fp;
fdtab_t *dt;
/*
* Look up the fdfile structure representing this descriptor.
* We are doing this unlocked. See fd_tryexpand().
*/
fdp = curlwp->l_fd;
dt = atomic_load_consume(&fdp->fd_dt);
if (__predict_false(fd >= dt->dt_nfiles)) {
return NULL;
}
ff = dt->dt_ff[fd];
KASSERT(fd >= NDFDFILE || ff == (fdfile_t *)fdp->fd_dfdfile[fd]);
if (__predict_false(ff == NULL)) {
return NULL;
}
/* Now get a reference to the descriptor. */
if (fdp->fd_refcnt == 1) {
/*
* Single threaded: don't need to worry about concurrent
* access (other than earlier calls to kqueue, which may
* hold a reference to the descriptor).
*/
ff->ff_refcnt++;
} else {
/*
* Multi threaded: issue a memory barrier to ensure that we
* acquire the file pointer _after_ adding a reference. If
* no memory barrier, we could fetch a stale pointer.
*
* In particular, we must coordinate the following four
* memory operations:
*
* A. fd_close store ff->ff_file = NULL
* B. fd_close refcnt = atomic_dec_uint_nv(&ff->ff_refcnt)
* C. fd_getfile atomic_inc_uint(&ff->ff_refcnt)
* D. fd_getfile load fp = ff->ff_file
*
* If the order is D;A;B;C:
*
* 1. D: fp = ff->ff_file
* 2. A: ff->ff_file = NULL
* 3. B: refcnt = atomic_dec_uint_nv(&ff->ff_refcnt)
* 4. C: atomic_inc_uint(&ff->ff_refcnt)
*
* then fd_close determines that there are no more
* references and decides to free fp immediately, at
* the same that fd_getfile ends up with an fp that's
* about to be freed. *boom*
*
* By making B a release operation in fd_close, and by
* making C an acquire operation in fd_getfile, since
* they are atomic operations on the same object, which
* has a total modification order, we guarantee either:
*
* - B happens before C. Then since A is
* sequenced before B in fd_close, and C is
* sequenced before D in fd_getfile, we
* guarantee A happens before D, so fd_getfile
* reads a null fp and safely fails.
*
* - C happens before B. Then fd_getfile may read
* null or nonnull, but either way, fd_close
* will safely wait for references to drain.
*/
atomic_inc_uint(&ff->ff_refcnt);
membar_acquire();
}
/*
* If the file is not open or is being closed then put the
* reference back.
*/
fp = atomic_load_consume(&ff->ff_file);
if (__predict_true(fp != NULL)) {
return fp;
}
fd_putfile(fd);
return NULL;
}
/*
* Release a reference to a file descriptor acquired with fd_getfile().
*/
void
fd_putfile(unsigned fd)
{
filedesc_t *fdp;
fdfile_t *ff;
u_int u, v;
fdp = curlwp->l_fd;
KASSERT(fd < atomic_load_consume(&fdp->fd_dt)->dt_nfiles);
ff = atomic_load_consume(&fdp->fd_dt)->dt_ff[fd];
if (fdp->fd_refcnt == 1) {
/*
* Single threaded: don't need to worry about concurrent
* access (other than earlier calls to kqueue, which may
* hold a reference to the descriptor).
*/
if (__predict_false((ff->ff_refcnt & FR_CLOSING) != 0)) {
fd_close(fd);
return;
}
ff->ff_refcnt--;
return;
}
/*
* Ensure that any use of the file is complete and globally
* visible before dropping the final reference. If no membar,
* the current CPU could still access memory associated with
* the file after it has been freed or recycled by another
* CPU.
*/
membar_release();
/*
* Be optimistic and start out with the assumption that no other
* threads are trying to close the descriptor. If the CAS fails,
* we lost a race and/or it's being closed.
*/
for (u = ff->ff_refcnt & FR_MASK;; u = v) {
v = atomic_cas_uint(&ff->ff_refcnt, u, u - 1);
if (__predict_true(u == v)) {
return;
}
if (__predict_false((v & FR_CLOSING) != 0)) {
break;
}
}
/* Another thread is waiting to close the file: join it. */
(void)fd_close(fd);
}
/*
* Convenience wrapper around fd_getfile() that returns reference
* to a vnode.
*/
int
fd_getvnode(unsigned fd, file_t **fpp)
{
vnode_t *vp;
file_t *fp;
fp = fd_getfile(fd);
if (__predict_false(fp == NULL)) {
return EBADF;
}
if (__predict_false(fp->f_type != DTYPE_VNODE)) {
fd_putfile(fd);
return EINVAL;
}
vp = fp->f_vnode;
if (__predict_false(vp->v_type == VBAD)) {
/* XXX Is this case really necessary? */
fd_putfile(fd);
return EBADF;
}
*fpp = fp;
return 0;
}
/*
* Convenience wrapper around fd_getfile() that returns reference
* to a socket.
*/
int
fd_getsock1(unsigned fd, struct socket **sop, file_t **fp)
{
*fp = fd_getfile(fd);
if (__predict_false(*fp == NULL)) {
return EBADF;
}
if (__predict_false((*fp)->f_type != DTYPE_SOCKET)) {
fd_putfile(fd);
return ENOTSOCK;
}
*sop = (*fp)->f_socket;
return 0;
}
/*
* Look up the file structure corresponding to a file descriptor
* and return it with a reference held on the file, not the
* descriptor.
*
* This is heavyweight and only used when accessing descriptors
* from a foreign process. The caller must ensure that `p' does
* not exit or fork across this call.
*
* To release the file (not descriptor) reference, use closef().
*/
file_t *
fd_getfile2(proc_t *p, unsigned fd)
{
filedesc_t *fdp;
fdfile_t *ff;
file_t *fp;
fdtab_t *dt;
/*
* Internal form of close. Must be called with a reference to the
* descriptor, and will drop the reference. When all descriptor
* references are dropped, releases the descriptor slot and a single
* reference to the file structure.
*/
int
fd_close(unsigned fd)
{
struct flock lf;
filedesc_t *fdp;
fdfile_t *ff;
file_t *fp;
proc_t *p;
lwp_t *l;
u_int refcnt;
l = curlwp;
p = l->l_proc;
fdp = l->l_fd;
ff = atomic_load_consume(&fdp->fd_dt)->dt_ff[fd];
mutex_enter(&fdp->fd_lock);
KASSERT((ff->ff_refcnt & FR_MASK) > 0);
fp = atomic_load_consume(&ff->ff_file);
if (__predict_false(fp == NULL)) {
/*
* Another user of the file is already closing, and is
* waiting for other users of the file to drain. Release
* our reference, and wake up the closer.
*/
membar_release();
atomic_dec_uint(&ff->ff_refcnt);
cv_broadcast(&ff->ff_closing);
mutex_exit(&fdp->fd_lock);
/*
* An application error, so pretend that the descriptor
* was already closed. We can't safely wait for it to
* be closed without potentially deadlocking.
*/
return (EBADF);
}
KASSERT((ff->ff_refcnt & FR_CLOSING) == 0);
/*
* There may be multiple users of this file within the process.
* Notify existing and new users that the file is closing. This
* will prevent them from adding additional uses to this file
* while we are closing it.
*/
atomic_store_relaxed(&ff->ff_file, NULL);
ff->ff_exclose = false;
/*
* We expect the caller to hold a descriptor reference - drop it.
* The reference count may increase beyond zero at this point due
* to an erroneous descriptor reference by an application, but
* fd_getfile() will notice that the file is being closed and drop
* the reference again.
*/
if (fdp->fd_refcnt == 1) {
/* Single threaded. */
refcnt = --(ff->ff_refcnt);
} else {
/* Multi threaded. */
membar_release();
refcnt = atomic_dec_uint_nv(&ff->ff_refcnt);
membar_acquire();
}
if (__predict_false(refcnt != 0)) {
/*
* Wait for other references to drain. This is typically
* an application error - the descriptor is being closed
* while still in use.
* (Or just a threaded application trying to unblock its
* thread that sleeps in (say) accept()).
*/
atomic_or_uint(&ff->ff_refcnt, FR_CLOSING);
/*
* Remove any knotes attached to the file. A knote
* attached to the descriptor can hold references on it.
*/
mutex_exit(&fdp->fd_lock);
if (!SLIST_EMPTY(&ff->ff_knlist)) {
knote_fdclose(fd);
}
/*
* Since the file system code doesn't know which fd
* each request came from (think dup()), we have to
* ask it to return ERESTART for any long-term blocks.
* The re-entry through read/write/etc will detect the
* closed fd and return EBAFD.
* Blocked partial writes may return a short length.
*/
(*fp->f_ops->fo_restart)(fp);
mutex_enter(&fdp->fd_lock);
/*
* We need to see the count drop to zero at least once,
* in order to ensure that all pre-existing references
* have been drained. New references past this point are
* of no interest.
* XXX (dsl) this may need to call fo_restart() after a
* timeout to guarantee that all the system calls exit.
*/
while ((ff->ff_refcnt & FR_MASK) != 0) {
cv_wait(&ff->ff_closing, &fdp->fd_lock);
}
atomic_and_uint(&ff->ff_refcnt, ~FR_CLOSING);
} else {
/* If no references, there must be no knotes. */
KASSERT(SLIST_EMPTY(&ff->ff_knlist));
}
/*
* POSIX record locking dictates that any close releases ALL
* locks owned by this process. This is handled by setting
* a flag in the unlock to free ONLY locks obeying POSIX
* semantics, and not to free BSD-style file locks.
* If the descriptor was in a message, POSIX-style locks
* aren't passed with the descriptor.
*/
if (__predict_false((p->p_flag & PK_ADVLOCK) != 0) &&
fp->f_ops->fo_advlock != NULL) {
lf.l_whence = SEEK_SET;
lf.l_start = 0;
lf.l_len = 0;
lf.l_type = F_UNLCK;
mutex_exit(&fdp->fd_lock);
(void)(*fp->f_ops->fo_advlock)(fp, p, F_UNLCK, &lf, F_POSIX);
mutex_enter(&fdp->fd_lock);
}
if (flags & ~(O_CLOEXEC|O_NONBLOCK|O_NOSIGPIPE))
return EINVAL;
/*
* Ensure there are enough slots in the descriptor table,
* and allocate an fdfile_t up front in case we need it.
*/
while (newfd >= atomic_load_consume(&fdp->fd_dt)->dt_nfiles) {
fd_tryexpand(curproc);
}
ff = kmem_alloc(sizeof(*ff), KM_SLEEP);
fdfile_ctor(ff);
/*
* If there is already a file open, close it. If the file is
* half open, wait for it to be constructed before closing it.
* XXX Potential for deadlock here?
*/
mutex_enter(&fdp->fd_lock);
while (fd_isused(fdp, newfd)) {
mutex_exit(&fdp->fd_lock);
if (fd_getfile(newfd) != NULL) {
(void)fd_close(newfd);
} else {
/*
* Crummy, but unlikely to happen.
* Can occur if we interrupt another
* thread while it is opening a file.
*/
kpause("dup2", false, 1, NULL);
}
mutex_enter(&fdp->fd_lock);
}
dt = fdp->fd_dt;
if (dt->dt_ff[newfd] == NULL) {
KASSERT(newfd >= NDFDFILE);
dt->dt_ff[newfd] = ff;
ff = NULL;
}
fd_used(fdp, newfd);
mutex_exit(&fdp->fd_lock);
fd_set_exclose(curlwp, newfd, (flags & O_CLOEXEC) != 0);
fp->f_flag |= flags & (FNONBLOCK|FNOSIGPIPE);
/* Slot is now allocated. Insert copy of the file. */
fd_affix(curproc, fp, newfd);
if (ff != NULL) {
cv_destroy(&ff->ff_closing);
kmem_free(ff, sizeof(*ff));
}
return 0;
}
/*
* Drop reference to a file structure.
*/
int
closef(file_t *fp)
{
struct flock lf;
int error;
/*
* Drop reference. If referenced elsewhere it's still open
* and we have nothing more to do.
*/
mutex_enter(&fp->f_lock);
KASSERT(fp->f_count > 0);
if (--fp->f_count > 0) {
mutex_exit(&fp->f_lock);
return 0;
}
KASSERT(fp->f_count == 0);
mutex_exit(&fp->f_lock);
/* We held the last reference - release locks, close and free. */
if (fp->f_ops->fo_advlock == NULL) {
KASSERT((fp->f_flag & FHASLOCK) == 0);
} else if (fp->f_flag & FHASLOCK) {
lf.l_whence = SEEK_SET;
lf.l_start = 0;
lf.l_len = 0;
lf.l_type = F_UNLCK;
(void)(*fp->f_ops->fo_advlock)(fp, fp, F_UNLCK, &lf, F_FLOCK);
}
if (fp->f_ops != NULL) {
error = (*fp->f_ops->fo_close)(fp);
/*
* .fo_close is final, so real errors are frowned on
* (but allowed and passed on to close(2)), and
* ERESTART is absolutely forbidden because the file
* descriptor is gone and there is no chance to retry.
*/
KASSERTMSG(error != ERESTART,
"file %p f_ops %p fo_close %p returned ERESTART",
fp, fp->f_ops, fp->f_ops->fo_close);
} else {
error = 0;
}
KASSERT(fp->f_count == 0);
KASSERT(fp->f_cred != NULL);
pool_cache_put(file_cache, fp);
return error;
}
/*
* Allocate a file descriptor for the process.
*
* Future idea for experimentation: replace all of this with radixtree.
*/
int
fd_alloc(proc_t *p, int want, int *result)
{
filedesc_t *fdp = p->p_fd;
int i, lim, last, error, hi;
u_int off;
fdtab_t *dt;
KASSERT(p == curproc || p == &proc0);
/*
* Search for a free descriptor starting at the higher
* of want or fd_freefile.
*/
mutex_enter(&fdp->fd_lock);
fd_checkmaps(fdp);
dt = fdp->fd_dt;
KASSERT(dt->dt_ff[0] == (fdfile_t *)fdp->fd_dfdfile[0]);
lim = uimin((int)p->p_rlimit[RLIMIT_NOFILE].rlim_cur, maxfiles);
last = uimin(dt->dt_nfiles, lim);
for (;;) {
if ((i = want) < fdp->fd_freefile)
i = fdp->fd_freefile;
off = i >> NDENTRYSHIFT;
hi = fd_next_zero(fdp, fdp->fd_himap, off,
(last + NDENTRIES - 1) >> NDENTRYSHIFT);
if (hi == -1)
break;
i = fd_next_zero(fdp, &fdp->fd_lomap[hi],
hi > off ? 0 : i & NDENTRYMASK, NDENTRIES);
if (i == -1) {
/*
* Free file descriptor in this block was
* below want, try again with higher want.
*/
want = (hi + 1) << NDENTRYSHIFT;
continue;
}
i += (hi << NDENTRYSHIFT);
if (i >= last) {
break;
}
if (dt->dt_ff[i] == NULL) {
KASSERT(i >= NDFDFILE);
dt->dt_ff[i] = kmem_alloc(sizeof(fdfile_t), KM_SLEEP);
fdfile_ctor(dt->dt_ff[i]);
}
KASSERT(dt->dt_ff[i]->ff_file == NULL);
fd_used(fdp, i);
if (want <= fdp->fd_freefile) {
fdp->fd_freefile = i;
}
*result = i;
KASSERT(i >= NDFDFILE ||
dt->dt_ff[i] == (fdfile_t *)fdp->fd_dfdfile[i]);
fd_checkmaps(fdp);
mutex_exit(&fdp->fd_lock);
return 0;
}
/* No space in current array. Let the caller expand and retry. */
error = (dt->dt_nfiles >= lim) ? EMFILE : ENOSPC;
mutex_exit(&fdp->fd_lock);
return error;
}
/*
* Allocate memory for a descriptor table.
*/
static fdtab_t *
fd_dtab_alloc(int n)
{
fdtab_t *dt;
size_t sz;
mutex_enter(&fdp->fd_lock);
dt = fdp->fd_dt;
KASSERT(dt->dt_ff[0] == (fdfile_t *)fdp->fd_dfdfile[0]);
if (dt->dt_nfiles != oldnfiles) {
/* fdp changed; caller must retry */
mutex_exit(&fdp->fd_lock);
fd_dtab_free(newdt);
if (NDHISLOTS(numfiles) > NDHISLOTS(oldnfiles)) {
fd_map_free(numfiles, newlomap, newhimap);
}
return;
}
/* Copy the existing descriptor table and zero the new portion. */
i = sizeof(fdfile_t *) * oldnfiles;
memcpy(newdt->dt_ff, dt->dt_ff, i);
memset((uint8_t *)newdt->dt_ff + i, 0,
numfiles * sizeof(fdfile_t *) - i);
/*
* Link old descriptor array into list to be discarded. We defer
* freeing until the last reference to the descriptor table goes
* away (usually process exit). This allows us to do lockless
* lookups in fd_getfile().
*/
if (oldnfiles > NDFILE) {
if (fdp->fd_refcnt > 1) {
newdt->dt_link = dt;
} else {
fd_dtab_free(dt);
}
}
if (NDHISLOTS(numfiles) > NDHISLOTS(oldnfiles)) {
i = NDHISLOTS(oldnfiles) * sizeof(uint32_t);
memcpy(newhimap, fdp->fd_himap, i);
memset((uint8_t *)newhimap + i, 0,
NDHISLOTS(numfiles) * sizeof(uint32_t) - i);
i = NDLOSLOTS(oldnfiles) * sizeof(uint32_t);
memcpy(newlomap, fdp->fd_lomap, i);
memset((uint8_t *)newlomap + i, 0,
NDLOSLOTS(numfiles) * sizeof(uint32_t) - i);
/*
* All other modifications must become globally visible before
* the change to fd_dt. See fd_getfile().
*/
atomic_store_release(&fdp->fd_dt, newdt);
KASSERT(newdt->dt_ff[0] == (fdfile_t *)fdp->fd_dfdfile[0]);
fd_checkmaps(fdp);
mutex_exit(&fdp->fd_lock);
}
/*
* Create a new open file structure and allocate a file descriptor
* for the current process.
*/
int
fd_allocfile(file_t **resultfp, int *resultfd)
{
proc_t *p = curproc;
kauth_cred_t cred;
file_t *fp;
int error;
while ((error = fd_alloc(p, 0, resultfd)) != 0) {
if (error != ENOSPC) {
return error;
}
fd_tryexpand(p);
}
/* Replace cached credentials if not what we need. */
cred = curlwp->l_cred;
if (__predict_false(cred != fp->f_cred)) {
kauth_cred_free(fp->f_cred);
fp->f_cred = kauth_cred_hold(cred);
}
/*
* Don't allow recycled files to be scanned.
* See uipc_usrreq.c.
*/
if (__predict_false((fp->f_flag & FSCAN) != 0)) {
mutex_enter(&fp->f_lock);
atomic_and_uint(&fp->f_flag, ~FSCAN);
mutex_exit(&fp->f_lock);
}
/*
* Successful creation of a new descriptor: make visible to the process.
*/
void
fd_affix(proc_t *p, file_t *fp, unsigned fd)
{
fdfile_t *ff;
filedesc_t *fdp;
fdtab_t *dt;
KASSERT(p == curproc || p == &proc0);
/* Add a reference to the file structure. */
mutex_enter(&fp->f_lock);
fp->f_count++;
mutex_exit(&fp->f_lock);
/*
* Insert the new file into the descriptor slot.
*/
fdp = p->p_fd;
dt = atomic_load_consume(&fdp->fd_dt);
ff = dt->dt_ff[fd];
static int
file_ctor(void *arg, void *obj, int flags)
{
/*
* It's easy to exhaust the open file limit on a system with many
* CPUs due to caching. Allow a bit of leeway to reduce the element
* of surprise.
*/
u_int slop = PCG_NOBJECTS_NORMAL * (ncpu - 1);
file_t *fp = obj;
/*
* If the number of open files fits in the internal arrays
* of the open file structure, use them, otherwise allocate
* additional memory for the number of descriptors currently
* in use.
*/
if (lastfile < NDFILE) {
i = NDFILE;
newdt = newfdp->fd_dt;
KASSERT(newfdp->fd_dt == &newfdp->fd_dtbuiltin);
} else {
/*
* Compute the smallest multiple of NDEXTENT needed
* for the file descriptors currently in use,
* allowing the table to shrink.
*/
i = numfiles;
while (i >= 2 * NDEXTENT && i > lastfile * 2) {
i /= 2;
}
KASSERT(i > NDFILE);
newdt = fd_dtab_alloc(i);
newfdp->fd_dt = newdt;
memcpy(newdt->dt_ff, newfdp->fd_dtbuiltin.dt_ff,
NDFDFILE * sizeof(fdfile_t **));
memset(newdt->dt_ff + NDFDFILE, 0,
(i - NDFDFILE) * sizeof(fdfile_t **));
}
if (NDHISLOTS(i) <= NDHISLOTS(NDFILE)) {
newfdp->fd_himap = newfdp->fd_dhimap;
newfdp->fd_lomap = newfdp->fd_dlomap;
} else {
fd_map_alloc(i, &newfdp->fd_lomap, &newfdp->fd_himap);
KASSERT(i >= NDENTRIES * NDENTRIES);
memset(newfdp->fd_himap, 0, NDHISLOTS(i)*sizeof(uint32_t));
memset(newfdp->fd_lomap, 0, NDLOSLOTS(i)*sizeof(uint32_t));
}
newfdp->fd_freefile = fdp->fd_freefile;
newfdp->fd_exclose = fdp->fd_exclose;
ffp = fdp->fd_dt->dt_ff;
nffp = newdt->dt_ff;
newlast = -1;
for (i = 0; i <= lastfile; i++, ffp++, nffp++) {
KASSERT(i >= NDFDFILE ||
*nffp == (fdfile_t *)newfdp->fd_dfdfile[i]);
ff = *ffp;
if (ff == NULL ||
(fp = atomic_load_consume(&ff->ff_file)) == NULL) {
/* Descriptor unused, or descriptor half open. */
KASSERT(!fd_isused(newfdp, i));
continue;
}
if (__predict_false(fp->f_type == DTYPE_KQUEUE)) {
/* kqueue descriptors cannot be copied. */
if (i < newfdp->fd_freefile) {
newfdp->fd_freefile = i;
}
continue;
}
/* It's active: add a reference to the file. */
mutex_enter(&fp->f_lock);
fp->f_count++;
mutex_exit(&fp->f_lock);
/* Allocate an fdfile_t to represent it. */
if (i >= NDFDFILE) {
ff2 = kmem_alloc(sizeof(*ff2), KM_SLEEP);
fdfile_ctor(ff2);
*nffp = ff2;
} else {
ff2 = newdt->dt_ff[i];
}
ff2->ff_file = fp;
ff2->ff_exclose = ff->ff_exclose;
ff2->ff_allocated = true;
membar_release();
if (atomic_dec_uint_nv(&fdp->fd_refcnt) > 0)
return;
membar_acquire();
/*
* Close any files that the process holds open.
*/
dt = fdp->fd_dt;
fd_checkmaps(fdp);
#ifdef DEBUG
fdp->fd_refcnt = -1; /* see fd_checkmaps */
#endif
for (fd = 0, nf = dt->dt_nfiles; fd < nf; fd++) {
ff = dt->dt_ff[fd];
KASSERT(fd >= NDFDFILE ||
ff == (fdfile_t *)fdp->fd_dfdfile[fd]);
if (ff == NULL)
continue;
if ((fp = atomic_load_consume(&ff->ff_file)) != NULL) {
/*
* Must use fd_close() here if there is
* a reference from kqueue or we might have posix
* advisory locks.
*/
if (__predict_true(ff->ff_refcnt == 0) &&
(noadvlock || fp->f_type != DTYPE_VNODE)) {
ff->ff_file = NULL;
ff->ff_exclose = false;
ff->ff_allocated = false;
closef(fp);
} else {
ff->ff_refcnt++;
fd_close(fd);
}
}
KASSERT(ff->ff_refcnt == 0);
KASSERT(ff->ff_file == NULL);
KASSERT(!ff->ff_exclose);
KASSERT(!ff->ff_allocated);
if (fd >= NDFDFILE) {
cv_destroy(&ff->ff_closing);
kmem_free(ff, sizeof(*ff));
dt->dt_ff[fd] = NULL;
}
}
/*
* Clean out the descriptor table for the next user and return
* to the cache.
*/
if (__predict_false(dt != &fdp->fd_dtbuiltin)) {
fd_dtab_free(fdp->fd_dt);
/* Otherwise, done above. */
memset(&fdp->fd_dtbuiltin.dt_ff[NDFDFILE], 0,
(NDFILE - NDFDFILE) * sizeof(fdp->fd_dtbuiltin.dt_ff[0]));
fdp->fd_dt = &fdp->fd_dtbuiltin;
}
if (__predict_false(NDHISLOTS(nf) > NDHISLOTS(NDFILE))) {
KASSERT(fdp->fd_himap != fdp->fd_dhimap);
KASSERT(fdp->fd_lomap != fdp->fd_dlomap);
fd_map_free(nf, fdp->fd_lomap, fdp->fd_himap);
}
if (__predict_false(fdp->fd_knhash != NULL)) {
hashdone(fdp->fd_knhash, HASH_LIST, fdp->fd_knhashmask);
fdp->fd_knhash = NULL;
fdp->fd_knhashmask = 0;
} else {
KASSERT(fdp->fd_knhashmask == 0);
}
fdp->fd_dt = &fdp->fd_dtbuiltin;
fdp->fd_lastkqfile = -1;
fdp->fd_lastfile = -1;
fdp->fd_freefile = 0;
fdp->fd_exclose = false;
memset(&fdp->fd_startzero, 0, sizeof(*fdp) -
offsetof(filedesc_t, fd_startzero));
fdp->fd_himap = fdp->fd_dhimap;
fdp->fd_lomap = fdp->fd_dlomap;
KASSERT(fdp->fd_dtbuiltin.dt_nfiles == NDFILE);
KASSERT(fdp->fd_dtbuiltin.dt_link == NULL);
KASSERT(fdp->fd_dt == &fdp->fd_dtbuiltin);
#ifdef DEBUG
fdp->fd_refcnt = 0; /* see fd_checkmaps */
#endif
fd_checkmaps(fdp);
pool_cache_put(filedesc_cache, fdp);
}
/*
* File Descriptor pseudo-device driver (/dev/fd/).
*
* Opening minor device N dup()s the file (if any) connected to file
* descriptor N belonging to the calling process. Note that this driver
* consists of only the ``open()'' routine, because all subsequent
* references to this file will be direct to the other driver.
*/
static int
filedescopen(dev_t dev, int mode, int type, lwp_t *l)
{
/*
* XXX Kludge: set dupfd to contain the value of the
* the file descriptor being sought for duplication. The error
* return ensures that the vnode for this device will be released
* by vn_open. Open will detect this special error and take the
* actions in fd_dupopen below. Other callers of vn_open or VOP_OPEN
* will simply report the error.
*/
l->l_dupfd = minor(dev); /* XXX */
return EDUPFD;
}
/*
* Duplicate the specified descriptor to a free descriptor.
*
* old is the original fd.
* moveit is true if we should move rather than duplicate.
* flags are the open flags (converted from O_* to F*).
* newp returns the new fd on success.
*
* These two cases are produced by the EDUPFD and EMOVEFD magic
* errnos, but in the interest of removing that regrettable interface,
* vn_open has been changed to intercept them. Now vn_open returns
* either a vnode or a filehandle, and the filehandle is accompanied
* by a boolean that says whether we should dup (moveit == false) or
* move (moveit == true) the fd.
*
* The dup case is used by /dev/stderr, /proc/self/fd, and such. The
* move case is used by cloner devices that allocate a fd of their
* own (a layering violation that should go away eventually) that
* then needs to be put in the place open() expects it.
*/
int
fd_dupopen(int old, bool moveit, int flags, int *newp)
{
filedesc_t *fdp;
fdfile_t *ff;
file_t *fp;
fdtab_t *dt;
int error;
if ((fp = fd_getfile(old)) == NULL) {
return EBADF;
}
fdp = curlwp->l_fd;
dt = atomic_load_consume(&fdp->fd_dt);
ff = dt->dt_ff[old];
/*
* There are two cases of interest here.
*
* 1. moveit == false (used to be the EDUPFD magic errno):
* simply dup (old) to file descriptor (new) and return.
*
* 2. moveit == true (used to be the EMOVEFD magic errno):
* steal away the file structure from (old) and store it in
* (new). (old) is effectively closed by this operation.
*/
if (moveit == false) {
/*
* Check that the mode the file is being opened for is a
* subset of the mode of the existing descriptor.
*/
if (((flags & (FREAD|FWRITE)) | fp->f_flag) != fp->f_flag) {
error = EACCES;
goto out;
}
/* Steal away the file pointer from 'old'. */
(void)fd_close(old);
return 0;
}
out:
fd_putfile(old);
return error;
}
/*
* Close open files on exec.
*/
void
fd_closeexec(void)
{
proc_t *p;
filedesc_t *fdp;
fdfile_t *ff;
lwp_t *l;
fdtab_t *dt;
int fd;
l = curlwp;
p = l->l_proc;
fdp = p->p_fd;
if (fdp->fd_refcnt > 1) {
fdp = fd_copy();
fd_free();
p->p_fd = fdp;
l->l_fd = fdp;
}
if (!fdp->fd_exclose) {
return;
}
fdp->fd_exclose = false;
dt = atomic_load_consume(&fdp->fd_dt);
for (fd = 0; fd <= fdp->fd_lastfile; fd++) {
if ((ff = dt->dt_ff[fd]) == NULL) {
KASSERT(fd >= NDFDFILE);
continue;
}
KASSERT(fd >= NDFDFILE ||
ff == (fdfile_t *)fdp->fd_dfdfile[fd]);
if (ff->ff_file == NULL)
continue;
if (ff->ff_exclose) {
/*
* We need a reference to close the file.
* No other threads can see the fdfile_t at
* this point, so don't bother locking.
*/
KASSERT((ff->ff_refcnt & FR_CLOSING) == 0);
ff->ff_refcnt++;
fd_close(fd);
}
}
}
/*
* Sets descriptor owner. If the owner is a process, 'pgid'
* is set to positive value, process ID. If the owner is process group,
* 'pgid' is set to -pg_id.
*/
int
fsetown(pid_t *pgid, u_long cmd, const void *data)
{
pid_t id = *(const pid_t *)data;
int error;
if (id <= INT_MIN)
return EINVAL;
switch (cmd) {
case TIOCSPGRP:
if (id < 0)
return EINVAL;
id = -id;
break;
default:
break;
}
if (id > 0) {
mutex_enter(&proc_lock);
error = proc_find(id) ? 0 : ESRCH;
mutex_exit(&proc_lock);
} else if (id < 0) {
error = pgid_in_session(curproc, -id);
} else {
error = 0;
}
if (!error) {
*pgid = id;
}
return error;
}
ff->ff_exclose = exclose;
if (exclose)
fdp->fd_exclose = true;
}
/*
* Return descriptor owner information. If the value is positive,
* it's process ID. If it's negative, it's process group ID and
* needs the sign removed before use.
*/
int
fgetown(pid_t pgid, u_long cmd, void *data)
{
/*
* Send signal to descriptor owner, either process or process group.
*/
void
fownsignal(pid_t pgid, int signo, int code, int band, void *fdescdata)
{
ksiginfo_t ksi;
if (elem_size < 1 || elem_count < 0)
return EINVAL;
switch (op) {
case KERN_FILE_BYFILE:
case KERN_FILE_BYPID:
/*
* We're traversing the process list in both cases; the BYFILE
* case does additional work of keeping track of files already
* looked at.
*/
/* doesn't use arg so it must be zero */
if ((op == KERN_FILE_BYFILE) && (arg != 0))
return EINVAL;
if ((op == KERN_FILE_BYPID) && (arg < -1))
/* -1 means all processes */
return EINVAL;
sysctl_unlock();
if (op == KERN_FILE_BYFILE)
mutex_enter(&sysctl_file_marker_lock);
mutex_enter(&proc_lock);
PROCLIST_FOREACH(p, &allproc) {
if (p->p_stat == SIDL) {
/* skip embryonic processes */
continue;
}
if (arg > 0 && p->p_pid != arg) {
/* pick only the one we want */
/* XXX want 0 to mean "kernel files" */
continue;
}
mutex_enter(p->p_lock);
error = kauth_authorize_process(l->l_cred,
KAUTH_PROCESS_CANSEE, p,
KAUTH_ARG(KAUTH_REQ_PROCESS_CANSEE_OPENFILES),
NULL, NULL);
mutex_exit(p->p_lock);
if (error != 0) {
/*
* Don't leak kauth retval if we're silently
* skipping this entry.
*/
error = 0;
continue;
}
/*
* Grab a hold on the process.
*/
if (!rw_tryenter(&p->p_reflock, RW_READER)) {
continue;
}
mutex_exit(&proc_lock);
fd = p->p_fd;
mutex_enter(&fd->fd_lock);
dt = fd->fd_dt;
for (i = 0; i < dt->dt_nfiles; i++) {
if ((ff = dt->dt_ff[i]) == NULL) {
continue;
}
if ((fp = atomic_load_consume(&ff->ff_file)) ==
NULL) {
continue;
}
if ((op == KERN_FILE_BYFILE) &&
(fp->f_marker == sysctl_file_marker)) {
continue;
}
if (len >= elem_size && elem_count > 0) {
mutex_enter(&fp->f_lock);
fill_file2(&kf, fp, ff, i, p->p_pid);
mutex_exit(&fp->f_lock);
mutex_exit(&fd->fd_lock);
error = sysctl_copyout(l,
&kf, dp, out_size);
mutex_enter(&fd->fd_lock);
if (error)
break;
dp += elem_size;
len -= elem_size;
}
if (op == KERN_FILE_BYFILE)
fp->f_marker = sysctl_file_marker;
needed += elem_size;
if (elem_count > 0 && elem_count != INT_MAX)
elem_count--;
}
mutex_exit(&fd->fd_lock);
/*
* Release reference to process.
*/
mutex_enter(&proc_lock);
rw_exit(&p->p_reflock);
}
if (op == KERN_FILE_BYFILE) {
sysctl_file_marker++;
/* Reset all markers if wrapped. */
if (sysctl_file_marker == 0) {
sysctl_file_marker_reset();
sysctl_file_marker++;
}
}
mutex_exit(&proc_lock);
if (op == KERN_FILE_BYFILE)
mutex_exit(&sysctl_file_marker_lock);
sysctl_relock();
break;
default:
return EINVAL;
}
