/*
- * Copyright (c) 2000-2008 Apple Inc. All rights reserved.
+ * Copyright (c) 2000-2014 Apple Inc. All rights reserved.
*
* @APPLE_OSREFERENCE_LICENSE_HEADER_START@
*
#include <security/mac_framework.h>
#endif
+#if CONFIG_PROTECT
+#include <sys/cprotect.h>
+#endif
+
+extern void sigpup_attach_vnode(vnode_t); /* XXX */
static int vn_closefile(struct fileglob *fp, vfs_context_t ctx);
static int vn_ioctl(struct fileproc *fp, u_long com, caddr_t data,
vfs_context_t ctx);
static void filt_vndetach(struct knote *kn);
static int filt_vnode(struct knote *kn, long hint);
+static int vn_open_auth_finish(vnode_t vp, int fmode, vfs_context_t ctx);
#if 0
static int vn_kqfilt_remove(struct vnode *vp, uintptr_t ident,
vfs_context_t ctx);
#endif
-struct fileops vnops =
- { vn_read, vn_write, vn_ioctl, vn_select, vn_closefile, vn_kqfilt_add, NULL };
+const struct fileops vnops = {
+ DTYPE_VNODE,
+ vn_read,
+ vn_write,
+ vn_ioctl,
+ vn_select,
+ vn_closefile,
+ vn_kqfilt_add,
+ NULL
+};
struct filterops vnode_filtops = {
.f_isfd = 1,
return(vn_open_auth(ndp, fmodep, &va));
}
+static int
+vn_open_auth_finish(vnode_t vp, int fmode, vfs_context_t ctx)
+{
+ int error;
+
+ if ((error = vnode_ref_ext(vp, fmode, 0)) != 0) {
+ goto bad;
+ }
+
+ /* Call out to allow 3rd party notification of open.
+ * Ignore result of kauth_authorize_fileop call.
+ */
+#if CONFIG_MACF
+ mac_vnode_notify_open(ctx, vp, fmode);
+#endif
+ kauth_authorize_fileop(vfs_context_ucred(ctx), KAUTH_FILEOP_OPEN,
+ (uintptr_t)vp, 0);
+
+ sigpup_attach_vnode(vp);
+
+ return 0;
+
+bad:
+ return error;
+
+}
+
+/*
+ * May do nameidone() to allow safely adding an FSEvent. Cue off of ni_dvp to
+ * determine whether that has happened.
+ */
+static int
+vn_open_auth_do_create(struct nameidata *ndp, struct vnode_attr *vap, int fmode, boolean_t *did_create, boolean_t *did_open, vfs_context_t ctx)
+{
+ uint32_t status = 0;
+ vnode_t dvp = ndp->ni_dvp;
+ int batched;
+ int error;
+ vnode_t vp;
+
+ batched = vnode_compound_open_available(ndp->ni_dvp);
+ *did_open = FALSE;
+
+ VATTR_SET(vap, va_type, VREG);
+ if (fmode & O_EXCL)
+ vap->va_vaflags |= VA_EXCLUSIVE;
+
+#if NAMEDRSRCFORK
+ if (ndp->ni_cnd.cn_flags & CN_WANTSRSRCFORK) {
+ if ((error = vn_authorize_create(dvp, &ndp->ni_cnd, vap, ctx, NULL)) != 0)
+ goto out;
+ if ((error = vnode_makenamedstream(dvp, &ndp->ni_vp, XATTR_RESOURCEFORK_NAME, 0, ctx)) != 0)
+ goto out;
+ *did_create = TRUE;
+ } else {
+#endif
+ if (!batched) {
+ if ((error = vn_authorize_create(dvp, &ndp->ni_cnd, vap, ctx, NULL)) != 0)
+ goto out;
+ }
+
+ error = vn_create(dvp, &ndp->ni_vp, ndp, vap, VN_CREATE_DOOPEN, fmode, &status, ctx);
+ if (error != 0) {
+ if (batched) {
+ *did_create = (status & COMPOUND_OPEN_STATUS_DID_CREATE) ? TRUE : FALSE;
+ } else {
+ *did_create = FALSE;
+ }
+
+ if (error == EKEEPLOOKING) {
+ if (*did_create) {
+ panic("EKEEPLOOKING, but we did a create?");
+ }
+ if (!batched) {
+ panic("EKEEPLOOKING from filesystem that doesn't support compound vnops?");
+ }
+ if ((ndp->ni_flag & NAMEI_CONTLOOKUP) == 0) {
+ panic("EKEEPLOOKING, but continue flag not set?");
+ }
+
+ /*
+ * Do NOT drop the dvp: we need everything to continue the lookup.
+ */
+ return error;
+ }
+ } else {
+ if (batched) {
+ *did_create = (status & COMPOUND_OPEN_STATUS_DID_CREATE) ? 1 : 0;
+ *did_open = TRUE;
+ } else {
+ *did_create = TRUE;
+ }
+ }
+#if NAMEDRSRCFORK
+ }
+#endif
+
+ vp = ndp->ni_vp;
+
+ if (*did_create) {
+ int update_flags = 0;
+
+ // Make sure the name & parent pointers are hooked up
+ if (vp->v_name == NULL)
+ update_flags |= VNODE_UPDATE_NAME;
+ if (vp->v_parent == NULLVP)
+ update_flags |= VNODE_UPDATE_PARENT;
+
+ if (update_flags)
+ vnode_update_identity(vp, dvp, ndp->ni_cnd.cn_nameptr, ndp->ni_cnd.cn_namelen, ndp->ni_cnd.cn_hash, update_flags);
+
+ vnode_put(dvp);
+ ndp->ni_dvp = NULLVP;
+
+#if CONFIG_FSE
+ if (need_fsevent(FSE_CREATE_FILE, vp)) {
+ add_fsevent(FSE_CREATE_FILE, ctx,
+ FSE_ARG_VNODE, vp,
+ FSE_ARG_DONE);
+ }
+#endif
+ }
+out:
+ if (ndp->ni_dvp != NULLVP) {
+ vnode_put(dvp);
+ ndp->ni_dvp = NULLVP;
+ }
+
+ return error;
+}
+
/*
* Open a file with authorization, updating the contents of the structures
* pointed to by ndp, fmodep, and vap as necessary to perform the requested
int error;
int fmode;
uint32_t origcnflags;
- kauth_action_t action;
+ boolean_t did_create;
+ boolean_t did_open;
+ boolean_t need_vnop_open;
+ boolean_t batched;
+ boolean_t ref_failed;
again:
vp = NULL;
dvp = NULL;
+ batched = FALSE;
+ did_create = FALSE;
+ need_vnop_open = TRUE;
+ ref_failed = FALSE;
fmode = *fmodep;
origcnflags = ndp->ni_cnd.cn_flags;
+
+ /*
+ * O_CREAT
+ */
if (fmode & O_CREAT) {
if ( (fmode & O_DIRECTORY) ) {
error = EINVAL;
goto out;
}
ndp->ni_cnd.cn_nameiop = CREATE;
+#if CONFIG_TRIGGERS
+ ndp->ni_op = OP_LINK;
+#endif
/* Inherit USEDVP, vnode_open() supported flags only */
- ndp->ni_cnd.cn_flags &= (USEDVP | NOCROSSMOUNT | DOWHITEOUT);
+ ndp->ni_cnd.cn_flags &= (USEDVP | NOCROSSMOUNT);
ndp->ni_cnd.cn_flags |= LOCKPARENT | LOCKLEAF | AUDITVNPATH1;
+ ndp->ni_flag = NAMEI_COMPOUNDOPEN;
#if NAMEDRSRCFORK
/* open calls are allowed for resource forks. */
ndp->ni_cnd.cn_flags |= CN_ALLOWRSRCFORK;
#endif
if ((fmode & O_EXCL) == 0 && (fmode & O_NOFOLLOW) == 0 && (origcnflags & FOLLOW) != 0)
ndp->ni_cnd.cn_flags |= FOLLOW;
+
+continue_create_lookup:
if ( (error = namei(ndp)) )
goto out;
+
dvp = ndp->ni_dvp;
vp = ndp->ni_vp;
- /* not found, create */
- if (vp == NULL) {
- /* must have attributes for a new file */
- if (vap == NULL) {
- error = EINVAL;
- goto badcreate;
- }
-
- VATTR_SET(vap, va_type, VREG);
-#if CONFIG_MACF
- error = mac_vnode_check_create(ctx,
- dvp, &ndp->ni_cnd, vap);
- if (error)
- goto badcreate;
-#endif /* MAC */
+ batched = vnode_compound_open_available(dvp);
- /* authorize before creating */
- if ((error = vnode_authorize(dvp, NULL, KAUTH_VNODE_ADD_FILE, ctx)) != 0)
- goto badcreate;
+ /* not found, create */
+ if (vp == NULL) {
+ /* must have attributes for a new file */
+ if (vap == NULL) {
+ vnode_put(dvp);
+ error = EINVAL;
+ goto out;
+ }
+ /*
+ * Attempt a create. For a system supporting compound VNOPs, we may
+ * find an existing file or create one; in either case, we will already
+ * have the file open and no VNOP_OPEN() will be needed.
+ */
+ error = vn_open_auth_do_create(ndp, vap, fmode, &did_create, &did_open, ctx);
- if (fmode & O_EXCL)
- vap->va_vaflags |= VA_EXCLUSIVE;
-#if NAMEDRSRCFORK
- if (ndp->ni_cnd.cn_flags & CN_WANTSRSRCFORK) {
- if ((error = vnode_makenamedstream(dvp, &ndp->ni_vp, XATTR_RESOURCEFORK_NAME, 0, ctx)) != 0)
- goto badcreate;
- } else
-#endif
- if ((error = vn_create(dvp, &ndp->ni_vp, &ndp->ni_cnd, vap, 0, ctx)) != 0)
- goto badcreate;
-
+ dvp = ndp->ni_dvp;
vp = ndp->ni_vp;
- if (vp) {
- int update_flags = 0;
-
- // Make sure the name & parent pointers are hooked up
- if (vp->v_name == NULL)
- update_flags |= VNODE_UPDATE_NAME;
- if (vp->v_parent == NULLVP)
- update_flags |= VNODE_UPDATE_PARENT;
-
- if (update_flags)
- vnode_update_identity(vp, dvp, ndp->ni_cnd.cn_nameptr, ndp->ni_cnd.cn_namelen, ndp->ni_cnd.cn_hash, update_flags);
-
-#if CONFIG_FSE
- if (need_fsevent(FSE_CREATE_FILE, vp)) {
- vnode_put(dvp);
- dvp = NULL;
- add_fsevent(FSE_CREATE_FILE, ctx,
- FSE_ARG_VNODE, vp,
- FSE_ARG_DONE);
+ /*
+ * Detected a node that the filesystem couldn't handle. Don't call
+ * nameidone() yet, because we need that path buffer.
+ */
+ if (error == EKEEPLOOKING) {
+ if (!batched) {
+ panic("EKEEPLOOKING from a filesystem that doesn't support compound VNOPs?");
}
-#endif
-
+ goto continue_create_lookup;
}
- /*
- * nameidone has to happen before we vnode_put(dvp)
- * and clear the ni_dvp field, since it may need
- * to release the fs_nodelock on the dvp
- */
-badcreate:
- nameidone(ndp);
- ndp->ni_dvp = NULL;
+ nameidone(ndp);
if (dvp) {
- vnode_put(dvp);
+ panic("Shouldn't have a dvp here.");
}
if (error) {
* Check for a creation or unlink race.
*/
if (((error == EEXIST) && !(fmode & O_EXCL)) ||
- ((error == ENOENT) && (fmode & O_CREAT))){
+ ((error == ENOENT) && (fmode & O_CREAT))){
+ if (vp)
+ vnode_put(vp);
goto again;
}
goto bad;
}
- fmode &= ~O_TRUNC;
- } else {
+
+ need_vnop_open = !did_open;
+ }
+ else {
+ if (fmode & O_EXCL)
+ error = EEXIST;
+
+ /*
+ * We have a vnode. Use compound open if available
+ * or else fall through to "traditional" path. Note: can't
+ * do a compound open for root, because the parent belongs
+ * to a different FS.
+ */
+ if (error == 0 && batched && (vnode_mount(dvp) == vnode_mount(vp))) {
+ error = VNOP_COMPOUND_OPEN(dvp, &ndp->ni_vp, ndp, 0, fmode, NULL, NULL, ctx);
+
+ if (error == 0) {
+ vp = ndp->ni_vp;
+ need_vnop_open = FALSE;
+ } else if (error == EKEEPLOOKING) {
+ if ((ndp->ni_flag & NAMEI_CONTLOOKUP) == 0) {
+ panic("EKEEPLOOKING, but continue flag not set?");
+ }
+ goto continue_create_lookup;
+ }
+ }
nameidone(ndp);
- ndp->ni_dvp = NULL;
vnode_put(dvp);
+ ndp->ni_dvp = NULLVP;
- if (fmode & O_EXCL) {
- error = EEXIST;
+ if (error) {
goto bad;
}
+
fmode &= ~O_CREAT;
+
+ /* Fall through */
}
- } else {
+ }
+ else {
+ /*
+ * Not O_CREAT
+ */
ndp->ni_cnd.cn_nameiop = LOOKUP;
/* Inherit USEDVP, vnode_open() supported flags only */
- ndp->ni_cnd.cn_flags &= (USEDVP | NOCROSSMOUNT | DOWHITEOUT);
- ndp->ni_cnd.cn_flags |= FOLLOW | LOCKLEAF | AUDITVNPATH1;
+ ndp->ni_cnd.cn_flags &= (USEDVP | NOCROSSMOUNT);
+ ndp->ni_cnd.cn_flags |= FOLLOW | LOCKLEAF | AUDITVNPATH1 | WANTPARENT;
#if NAMEDRSRCFORK
/* open calls are allowed for resource forks. */
ndp->ni_cnd.cn_flags |= CN_ALLOWRSRCFORK;
#endif
+ ndp->ni_flag = NAMEI_COMPOUNDOPEN;
+
/* preserve NOFOLLOW from vnode_open() */
if (fmode & O_NOFOLLOW || fmode & O_SYMLINK || (origcnflags & FOLLOW) == 0) {
- ndp->ni_cnd.cn_flags &= ~FOLLOW;
+ ndp->ni_cnd.cn_flags &= ~FOLLOW;
}
- if ( (error = namei(ndp)) )
- goto out;
- vp = ndp->ni_vp;
+ /* Do a lookup, possibly going directly to filesystem for compound operation */
+ do {
+ if ( (error = namei(ndp)) )
+ goto out;
+ vp = ndp->ni_vp;
+ dvp = ndp->ni_dvp;
+
+ /* Check for batched lookup-open */
+ batched = vnode_compound_open_available(dvp);
+ if (batched && ((vp == NULLVP) || (vnode_mount(dvp) == vnode_mount(vp)))) {
+ error = VNOP_COMPOUND_OPEN(dvp, &ndp->ni_vp, ndp, 0, fmode, NULL, NULL, ctx);
+ vp = ndp->ni_vp;
+ if (error == 0) {
+ need_vnop_open = FALSE;
+ } else if (error == EKEEPLOOKING) {
+ if ((ndp->ni_flag & NAMEI_CONTLOOKUP) == 0) {
+ panic("EKEEPLOOKING, but continue flag not set?");
+ }
+ }
+ }
+ } while (error == EKEEPLOOKING);
+
nameidone(ndp);
- ndp->ni_dvp = NULL;
+ vnode_put(dvp);
+ ndp->ni_dvp = NULLVP;
- if ( (fmode & O_DIRECTORY) && vp->v_type != VDIR ) {
- error = ENOTDIR;
+ if (error) {
goto bad;
}
}
- if (vp->v_type == VSOCK && vp->v_tag != VT_FDESC) {
- error = EOPNOTSUPP; /* Operation not supported on socket */
- goto bad;
- }
-
- if (vp->v_type == VLNK && (fmode & O_NOFOLLOW) != 0) {
- error = ELOOP; /* O_NOFOLLOW was specified and the target is a symbolic link */
- goto bad;
+ /*
+ * By this point, nameidone() is called, dvp iocount is dropped,
+ * and dvp pointer is cleared.
+ */
+ if (ndp->ni_dvp != NULLVP) {
+ panic("Haven't cleaned up adequately in vn_open_auth()");
}
- /* authorize open of an existing file */
- if ((fmode & O_CREAT) == 0) {
-
- /* disallow write operations on directories */
- if (vnode_isdir(vp) && (fmode & (FWRITE | O_TRUNC))) {
- error = EISDIR;
- goto bad;
+ /*
+ * Expect to use this code for filesystems without compound VNOPs, for the root
+ * of a filesystem, which can't be "looked up" in the sense of VNOP_LOOKUP(),
+ * and for shadow files, which do not live on the same filesystems as their "parents."
+ */
+ if (need_vnop_open) {
+ if (batched && !vnode_isvroot(vp) && !vnode_isnamedstream(vp)) {
+ panic("Why am I trying to use VNOP_OPEN() on anything other than the root or a named stream?");
}
-#if CONFIG_MACF
- error = mac_vnode_check_open(ctx, vp, fmode);
- if (error)
- goto bad;
-#endif
-
- /* compute action to be authorized */
- action = 0;
- if (fmode & FREAD) {
- action |= KAUTH_VNODE_READ_DATA;
+ if (!did_create) {
+ error = vn_authorize_open_existing(vp, &ndp->ni_cnd, fmode, ctx, NULL);
+ if (error) {
+ goto bad;
+ }
}
- if (fmode & (FWRITE | O_TRUNC)) {
- /*
- * If we are writing, appending, and not truncating,
- * indicate that we are appending so that if the
- * UF_APPEND or SF_APPEND bits are set, we do not deny
- * the open.
- */
- if ((fmode & O_APPEND) && !(fmode & O_TRUNC)) {
- action |= KAUTH_VNODE_APPEND_DATA;
- } else {
- action |= KAUTH_VNODE_WRITE_DATA;
+
+#if CONFIG_PROTECT
+ /*
+ * Perform any content protection access checks prior to calling
+ * into the filesystem, if the raw encrypted mode was not
+ * requested.
+ *
+ * If the va_dataprotect_flags are NOT active, or if they are,
+ * but they do not have the VA_DP_RAWENCRYPTED bit set, then we need
+ * to perform the checks.
+ */
+ if (!(VATTR_IS_ACTIVE (vap, va_dataprotect_flags)) ||
+ ((vap->va_dataprotect_flags & VA_DP_RAWENCRYPTED) == 0)) {
+ error = cp_handle_open (vp, fmode);
+ if (error) {
+ goto bad;
}
}
- if ((error = vnode_authorize(vp, NULL, action, ctx)) != 0)
- goto bad;
-
+#endif
- //
- // if the vnode is tagged VOPENEVT and the current process
- // has the P_CHECKOPENEVT flag set, then we or in the O_EVTONLY
- // flag to the open mode so that this open won't count against
- // the vnode when carbon delete() does a vnode_isinuse() to see
- // if a file is currently in use. this allows spotlight
- // importers to not interfere with carbon apps that depend on
- // the no-delete-if-busy semantics of carbon delete().
- //
- if ((vp->v_flag & VOPENEVT) && (current_proc()->p_flag & P_CHECKOPENEVT)) {
- fmode |= O_EVTONLY;
+ error = VNOP_OPEN(vp, fmode, ctx);
+ if (error) {
+ goto bad;
}
+ need_vnop_open = FALSE;
+ }
+ // if the vnode is tagged VOPENEVT and the current process
+ // has the P_CHECKOPENEVT flag set, then we or in the O_EVTONLY
+ // flag to the open mode so that this open won't count against
+ // the vnode when carbon delete() does a vnode_isinuse() to see
+ // if a file is currently in use. this allows spotlight
+ // importers to not interfere with carbon apps that depend on
+ // the no-delete-if-busy semantics of carbon delete().
+ //
+ if (!did_create && (vp->v_flag & VOPENEVT) && (current_proc()->p_flag & P_CHECKOPENEVT)) {
+ fmode |= O_EVTONLY;
}
- if ( (error = VNOP_OPEN(vp, fmode, ctx)) ) {
+ /*
+ * Grab reference, etc.
+ */
+ error = vn_open_auth_finish(vp, fmode, ctx);
+ if (error) {
+ ref_failed = TRUE;
goto bad;
}
- if ( (error = vnode_ref_ext(vp, fmode)) ) {
- goto bad2;
- }
- /* call out to allow 3rd party notification of open.
- * Ignore result of kauth_authorize_fileop call.
- */
- kauth_authorize_fileop(vfs_context_ucred(ctx), KAUTH_FILEOP_OPEN,
- (uintptr_t)vp, 0);
+ /* Compound VNOP open is responsible for doing the truncate */
+ if (batched || did_create)
+ fmode &= ~O_TRUNC;
*fmodep = fmode;
return (0);
-bad2:
- VNOP_CLOSE(vp, fmode, ctx);
+
bad:
+ /* Opened either explicitly or by a batched create */
+ if (!need_vnop_open) {
+ VNOP_CLOSE(vp, fmode, ctx);
+ }
+
ndp->ni_vp = NULL;
if (vp) {
#if NAMEDRSRCFORK
*
* EREDRIVEOPEN: means that we were hit by the tty allocation race.
*/
- if (((error == ENOENT) && (*fmodep & O_CREAT)) || (error == EREDRIVEOPEN)) {
+ if (((error == ENOENT) && (*fmodep & O_CREAT)) || (error == EREDRIVEOPEN) || ref_failed) {
goto again;
}
}
+
out:
return (error);
}
vn_close(struct vnode *vp, int flags, vfs_context_t ctx)
{
int error;
-
-#if CONFIG_FSE
- if (flags & FWASWRITTEN) {
- if (need_fsevent(FSE_CONTENT_MODIFIED, vp)) {
- add_fsevent(FSE_CONTENT_MODIFIED, ctx,
- FSE_ARG_VNODE, vp,
- FSE_ARG_DONE);
- }
- }
-#endif
+ int flusherror = 0;
#if NAMEDRSRCFORK
/* Sync data from resource fork shadow file if needed. */
(vp->v_parent != NULLVP) &&
vnode_isshadow(vp)) {
if (flags & FWASWRITTEN) {
- (void) vnode_flushnamedstream(vp->v_parent, vp, ctx);
+ flusherror = vnode_flushnamedstream(vp->v_parent, vp, ctx);
}
}
#endif
if (vnode_isspec(vp))
(void)vnode_rele_ext(vp, flags, 0);
+ /*
+ * On HFS, we flush when the last writer closes. We do this
+ * because resource fork vnodes hold a reference on data fork
+ * vnodes and that will prevent them from getting VNOP_INACTIVE
+ * which will delay when we flush cached data. In future, we
+ * might find it beneficial to do this for all file systems.
+ * Note that it's OK to access v_writecount without the lock
+ * in this context.
+ */
+ if (vp->v_tag == VT_HFS && (flags & FWRITE) && vp->v_writecount == 1)
+ VNOP_FSYNC(vp, MNT_NOWAIT, ctx);
+
error = VNOP_CLOSE(vp, flags, ctx);
+#if CONFIG_FSE
+ if (flags & FWASWRITTEN) {
+ if (need_fsevent(FSE_CONTENT_MODIFIED, vp)) {
+ add_fsevent(FSE_CONTENT_MODIFIED, ctx,
+ FSE_ARG_VNODE, vp,
+ FSE_ARG_DONE);
+ }
+ }
+#endif
+
if (!vnode_isspec(vp))
(void)vnode_rele_ext(vp, flags, 0);
+ if (flusherror) {
+ error = flusherror;
+ }
return (error);
}
if (error == 0) {
if (rw == UIO_READ) {
- if (vnode_isswap(vp)) {
+ if (vnode_isswap(vp) && ((ioflg & IO_SWAP_DISPATCH) == 0)) {
error = vn_read_swapfile(vp, auio);
} else {
error = VNOP_READ(vp, auio, ioflg, &context);
return (error);
}
+static inline void
+vn_offset_lock(struct fileglob *fg)
+{
+ lck_mtx_lock_spin(&fg->fg_lock);
+ while (fg->fg_lflags & FG_OFF_LOCKED) {
+ fg->fg_lflags |= FG_OFF_LOCKWANT;
+ msleep(&fg->fg_lflags, &fg->fg_lock, PVFS | PSPIN,
+ "fg_offset_lock_wait", 0);
+ }
+ fg->fg_lflags |= FG_OFF_LOCKED;
+ lck_mtx_unlock(&fg->fg_lock);
+}
+
+static inline void
+vn_offset_unlock(struct fileglob *fg)
+{
+ int lock_wanted = 0;
+
+ lck_mtx_lock_spin(&fg->fg_lock);
+ if (fg->fg_lflags & FG_OFF_LOCKWANT) {
+ lock_wanted = 1;
+ }
+ fg->fg_lflags &= ~(FG_OFF_LOCKED | FG_OFF_LOCKWANT);
+ lck_mtx_unlock(&fg->fg_lock);
+ if (lock_wanted) {
+ wakeup(&fg->fg_lflags);
+ }
+}
+
/*
* File table vnode read routine.
*/
vn_read(struct fileproc *fp, struct uio *uio, int flags, vfs_context_t ctx)
{
struct vnode *vp;
- int error, ioflag;
+ int error;
+ int ioflag;
off_t count;
+ int offset_locked = 0;
vp = (struct vnode *)fp->f_fglob->fg_data;
if ( (error = vnode_getwithref(vp)) ) {
}
#endif
- ioflag = 0;
+ /* This signals to VNOP handlers that this read came from a file table read */
+ ioflag = IO_SYSCALL_DISPATCH;
+
if (fp->f_fglob->fg_flag & FNONBLOCK)
ioflag |= IO_NDELAY;
if ((fp->f_fglob->fg_flag & FNOCACHE) || vnode_isnocache(vp))
- ioflag |= IO_NOCACHE;
+ ioflag |= IO_NOCACHE;
+ if (fp->f_fglob->fg_flag & FENCRYPTED) {
+ ioflag |= IO_ENCRYPTED;
+ }
if (fp->f_fglob->fg_flag & FNORDAHEAD)
- ioflag |= IO_RAOFF;
+ ioflag |= IO_RAOFF;
- if ((flags & FOF_OFFSET) == 0)
+ if ((flags & FOF_OFFSET) == 0) {
+ if ((vnode_vtype(vp) == VREG) && !vnode_isswap(vp)) {
+ vn_offset_lock(fp->f_fglob);
+ offset_locked = 1;
+ }
uio->uio_offset = fp->f_fglob->fg_offset;
+ }
count = uio_resid(uio);
if (vnode_isswap(vp)) {
} else {
error = VNOP_READ(vp, uio, ioflag, ctx);
}
- if ((flags & FOF_OFFSET) == 0)
+ if ((flags & FOF_OFFSET) == 0) {
fp->f_fglob->fg_offset += count - uio_resid(uio);
+ if (offset_locked) {
+ vn_offset_unlock(fp->f_fglob);
+ offset_locked = 0;
+ }
+ }
(void)vnode_put(vp);
return (error);
int clippedsize = 0;
int partialwrite=0;
int residcount, oldcount;
+ int offset_locked = 0;
proc_t p = vfs_context_proc(ctx);
count = 0;
}
#endif
- ioflag = IO_UNIT;
+ /*
+ * IO_SYSCALL_DISPATCH signals to VNOP handlers that this write came from
+ * a file table write
+ */
+ ioflag = (IO_UNIT | IO_SYSCALL_DISPATCH);
+
if (vp->v_type == VREG && (fp->f_fglob->fg_flag & O_APPEND))
ioflag |= IO_APPEND;
if (fp->f_fglob->fg_flag & FNONBLOCK)
ioflag |= IO_NDELAY;
if ((fp->f_fglob->fg_flag & FNOCACHE) || vnode_isnocache(vp))
ioflag |= IO_NOCACHE;
+ if (fp->f_fglob->fg_flag & FNODIRECT)
+ ioflag |= IO_NODIRECT;
+ if (fp->f_fglob->fg_flag & FSINGLE_WRITER)
+ ioflag |= IO_SINGLE_WRITER;
+
/*
* Treat synchronous mounts and O_FSYNC on the fd as equivalent.
*
}
if ((flags & FOF_OFFSET) == 0) {
+ if ((vnode_vtype(vp) == VREG) && !vnode_isswap(vp)) {
+ vn_offset_lock(fp->f_fglob);
+ offset_locked = 1;
+ }
uio->uio_offset = fp->f_fglob->fg_offset;
count = uio_resid(uio);
}
}
if (clippedsize >= residcount) {
psignal(p, SIGXFSZ);
- vnode_put(vp);
- return (EFBIG);
+ error = EFBIG;
+ goto error_out;
}
partialwrite = 1;
uio_setresid(uio, residcount-clippedsize);
if (p && (vp->v_type == VREG) &&
((rlim_t)uio->uio_offset >= p->p_rlimit[RLIMIT_FSIZE].rlim_cur)) {
psignal(p, SIGXFSZ);
- vnode_put(vp);
- return (EFBIG);
+ error = EFBIG;
+ goto error_out;
}
if (p && (vp->v_type == VREG) &&
((rlim_t)(uio->uio_offset + uio_resid(uio)) > p->p_rlimit[RLIMIT_FSIZE].rlim_cur)) {
fp->f_fglob->fg_offset = uio->uio_offset;
else
fp->f_fglob->fg_offset += count - uio_resid(uio);
+ if (offset_locked) {
+ vn_offset_unlock(fp->f_fglob);
+ offset_locked = 0;
+ }
}
/*
}
(void)vnode_put(vp);
return (error);
+
+error_out:
+ if (offset_locked) {
+ vn_offset_unlock(fp->f_fglob);
+ }
+ (void)vnode_put(vp);
+ return (error);
}
/*
sb->st_blocks = roundup(va.va_total_alloc, 512) / 512;
}
- /* if we're interested in exended security data and we got an ACL */
+ /* if we're interested in extended security data and we got an ACL */
if (xsec != NULL) {
if (!VATTR_IS_SUPPORTED(&va, va_acl) &&
!VATTR_IS_SUPPORTED(&va, va_uuuid) &&
off_t file_size;
int error;
struct vnode *ttyvp;
- int funnel_state;
struct session * sessp;
if ( (error = vnode_getwithref(vp)) ) {
error = VNOP_IOCTL(vp, com, data, fp->f_fglob->fg_flag, ctx);
if (error == 0 && com == TIOCSCTTY) {
- vnode_ref(vp);
-
- funnel_state = thread_funnel_set(kernel_flock, TRUE);
sessp = proc_session(vfs_context_proc(ctx));
session_lock(sessp);
sessp->s_ttyvid = vnode_vid(vp);
session_unlock(sessp);
session_rele(sessp);
- thread_funnel_set(kernel_flock, funnel_state);
-
- if (ttyvp)
- vnode_rele(ttyvp);
}
}
out:
if ( (error = vnode_getwithref(vp)) == 0 ) {
- if ((fg->fg_flag & FHASLOCK) && fg->fg_type == DTYPE_VNODE) {
+ if ((fg->fg_flag & FHASLOCK) &&
+ FILEGLOB_DTYPE(fg) == DTYPE_VNODE) {
lf.l_whence = SEEK_SET;
lf.l_start = 0;
lf.l_len = 0;
lf.l_type = F_UNLCK;
- (void)VNOP_ADVLOCK(vp, (caddr_t)fg, F_UNLCK, &lf, F_FLOCK, ctx);
+ (void)VNOP_ADVLOCK(vp, (caddr_t)fg, F_UNLCK, &lf, F_FLOCK, ctx, NULL);
}
error = vn_close(vp, fg->fg_flag, ctx);
vn_pathconf(vnode_t vp, int name, int32_t *retval, vfs_context_t ctx)
{
int error = 0;
+ struct vfs_attr vfa;
switch(name) {
case _PC_EXTENDED_SECURITY_NP:
case _PC_SYNC_IO: /* unistd.h: _POSIX_SYNCHRONIZED_IO */
*retval = 0; /* [SIO] option is not supported */
break;
+ case _PC_XATTR_SIZE_BITS:
+ /* The number of bits used to store maximum extended
+ * attribute size in bytes. For example, if the maximum
+ * attribute size supported by a file system is 128K, the
+ * value returned will be 18. However a value 18 can mean
+ * that the maximum attribute size can be anywhere from
+ * (256KB - 1) to 128KB. As a special case, the resource
+ * fork can have much larger size, and some file system
+ * specific extended attributes can have smaller and preset
+ * size; for example, Finder Info is always 32 bytes.
+ */
+ memset(&vfa, 0, sizeof(vfa));
+ VFSATTR_INIT(&vfa);
+ VFSATTR_WANTED(&vfa, f_capabilities);
+ if (vfs_getattr(vnode_mount(vp), &vfa, ctx) == 0 &&
+ (VFSATTR_IS_SUPPORTED(&vfa, f_capabilities)) &&
+ (vfa.f_capabilities.capabilities[VOL_CAPABILITIES_INTERFACES] & VOL_CAP_INT_EXTENDED_ATTR) &&
+ (vfa.f_capabilities.valid[VOL_CAPABILITIES_INTERFACES] & VOL_CAP_INT_EXTENDED_ATTR)) {
+ /* Supports native extended attributes */
+ error = VNOP_PATHCONF(vp, name, retval, ctx);
+ } else {
+ /* Number of bits used to represent the maximum size of
+ * extended attribute stored in an Apple Double file.
+ */
+ *retval = AD_XATTR_SIZE_BITS;
+ }
+ break;
default:
error = VNOP_PATHCONF(vp, name, retval, ctx);
break;
}
} else if (!vnode_isreg(vp)) {
- if (vnode_isspec(vp) &&
+ if (vnode_ischr(vp) &&
(error = spec_kqfilter(vp, kn)) == 0) {
/* claimed by a special device */
vnode_put(vp);
vnode_readable_data_count(vnode_t vp, off_t current_offset, int ispoll)
{
if (vnode_isfifo(vp)) {
+#if FIFO
int cnt;
int err = fifo_charcount(vp, &cnt);
if (err == 0) {
return (intptr_t)cnt;
- } else {
+ } else
+#endif
+ {
return (intptr_t)0;
}
} else if (vnode_isreg(vp)) {
vnode_writable_space_count(vnode_t vp)
{
if (vnode_isfifo(vp)) {
+#if FIFO
long spc;
int err = fifo_freespace(vp, &spc);
if (err == 0) {
return (intptr_t)spc;
- } else {
+ } else
+#endif
+ {
return (intptr_t)0;
}
} else if (vnode_isreg(vp)) {
static int
filt_vnode(struct knote *kn, long hint)
{
- struct vnode *vp = (struct vnode *)kn->kn_hook;
+ vnode_t vp = (struct vnode *)kn->kn_hook;
int activate = 0;
+ long orig_hint = hint;
if (0 == hint) {
- if ((vnode_getwithvid(vp, kn->kn_hookid) != 0)) {
+ vnode_lock(vp);
+
+ if (vnode_getiocount(vp, kn->kn_hookid, VNODE_NODEAD | VNODE_WITHID) != 0) {
+ /* Is recycled */
hint = NOTE_REVOKE;
- } else {
- vnode_put(vp);
- }
- }
+ }
+ } else {
+ lck_mtx_assert(&vp->v_lock, LCK_MTX_ASSERT_OWNED);
+ }
- /* NOTE_REVOKE is special, as it is only sent during vnode reclaim */
+ /* Special handling for vnodes that are in recycle or already gone */
if (NOTE_REVOKE == hint) {
kn->kn_flags |= (EV_EOF | EV_ONESHOT);
activate = 1;
}
}
+ if (orig_hint == 0) {
+ /*
+ * Definitely need to unlock, may need to put
+ */
+ if (hint == 0) {
+ vnode_put_locked(vp);
+ }
+ vnode_unlock(vp);
+ }
+
return (activate);
}
projects / apple / xnu.git / blobdiff