1/*-
2 * SPDX-License-Identifier: BSD-3-Clause
3 *
4 * Copyright (c) 1990, 1993
5 * The Regents of the University of California. All rights reserved.
6 *
7 * Redistribution and use in source and binary forms, with or without
8 * modification, are permitted provided that the following conditions
9 * are met:
10 * 1. Redistributions of source code must retain the above copyright
11 * notice, this list of conditions and the following disclaimer.
12 * 2. Redistributions in binary form must reproduce the above copyright
13 * notice, this list of conditions and the following disclaimer in the
14 * documentation and/or other materials provided with the distribution.
15 * 3. Neither the name of the University nor the names of its contributors
16 * may be used to endorse or promote products derived from this software
17 * without specific prior written permission.
18 *
19 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
20 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
21 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
22 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
23 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
24 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
25 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
26 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
27 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
28 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
29 * SUCH DAMAGE.
30 */
31
32#ifndef _SYS_FILEDESC_H_
33#define _SYS_FILEDESC_H_
34
35#include <sys/types.h>
36#include <sys/caprights.h>
37#include <sys/queue.h>
38#include <sys/event.h>
39#include <sys/lock.h>
40#include <sys/mutex.h>
41#include <sys/priority.h>
42#include <sys/seqc.h>
43#include <sys/sx.h>
44#include <sys/_smr.h>
45#include <sys/smr_types.h>
46
47#include <machine/_limits.h>
48
49struct filecaps {
50 cap_rights_t fc_rights; /* per-descriptor capability rights */
51 u_long *fc_ioctls; /* per-descriptor allowed ioctls */
52 int16_t fc_nioctls; /* fc_ioctls array size */
53 uint32_t fc_fcntls; /* per-descriptor allowed fcntls */
54};
55
56struct filedescent {
57 struct file *fde_file; /* file structure for open file */
58 struct filecaps fde_caps; /* per-descriptor rights */
59 uint8_t fde_flags; /* per-process open file flags */
60 seqc_t fde_seqc; /* keep file and caps in sync */
61};
62#define fde_rights fde_caps.fc_rights
63#define fde_fcntls fde_caps.fc_fcntls
64#define fde_ioctls fde_caps.fc_ioctls
65#define fde_nioctls fde_caps.fc_nioctls
66
67#ifdef _KERNEL
68static inline void
69fde_copy(struct filedescent *from, struct filedescent *to)
70{
71
72 to->fde_file = from->fde_file;
73 to->fde_caps = from->fde_caps;
74 to->fde_flags = from->fde_flags;
75}
76#endif
77
78struct fdescenttbl {
79 int fdt_nfiles; /* number of open files allocated */
80 struct filedescent fdt_ofiles[0]; /* open files */
81};
82#define fd_seqc(fdt, fd) (&(fdt)->fdt_ofiles[(fd)].fde_seqc)
83
84#define NDSLOTTYPE u_long
85
86/*
87 * This struct is copy-on-write and allocated from an SMR zone.
88 * All fields are constant after initialization apart from the reference count.
89 * The ABI root directory is initialized as the root directory and changed
90 * during process transiting to or from non-native ABI.
91 *
92 * Check pwd_* routines for usage.
93 */
94struct pwd {
95 u_int pwd_refcount;
96 struct vnode *pwd_cdir; /* current directory */
97 struct vnode *pwd_rdir; /* root directory */
98 struct vnode *pwd_jdir; /* jail root directory */
99 struct vnode *pwd_adir; /* abi root directory */
100};
101typedef SMR_POINTER(struct pwd *) smrpwd_t;
102
103struct pwddesc {
104 struct mtx pd_lock; /* protects members of this struct */
105 smrpwd_t pd_pwd; /* directories */
106 u_int pd_refcount;
107 u_short pd_cmask; /* mask for file creation */
108};
109
110/*
111 * This structure is used for the management of descriptors. It may be
112 * shared by multiple processes.
113 */
114struct filedesc {
115 struct fdescenttbl *fd_files; /* open files table */
116 NDSLOTTYPE *fd_map; /* bitmap of free fds */
117 int fd_freefile; /* approx. next free file */
118 int fd_refcnt; /* thread reference count */
119 int fd_holdcnt; /* hold count on structure + mutex */
120 struct sx fd_sx; /* protects members of this struct */
121 struct kqlist fd_kqlist; /* list of kqueues on this filedesc */
122 int fd_holdleaderscount; /* block fdfree() for shared close() */
123 int fd_holdleaderswakeup; /* fdfree() needs wakeup */
124};
125
126/*
127 * Structure to keep track of (process leader, struct fildedesc) tuples.
128 * Each process has a pointer to such a structure when detailed tracking
129 * is needed, e.g., when rfork(RFPROC | RFMEM) causes a file descriptor
130 * table to be shared by processes having different "p_leader" pointers
131 * and thus distinct POSIX style locks.
132 *
133 * fdl_refcount and fdl_holdcount are protected by struct filedesc mtx.
134 */
135struct filedesc_to_leader {
136 int fdl_refcount; /* references from struct proc */
137 int fdl_holdcount; /* temporary hold during closef */
138 int fdl_wakeup; /* fdfree() waits on closef() */
139 struct proc *fdl_leader; /* owner of POSIX locks */
140 /* Circular list: */
141 struct filedesc_to_leader *fdl_prev;
142 struct filedesc_to_leader *fdl_next;
143};
144#define fd_nfiles fd_files->fdt_nfiles
145#define fd_ofiles fd_files->fdt_ofiles
146
147/*
148 * Per-process open flags.
149 */
150#define UF_EXCLOSE 0x01 /* auto-close on exec */
151#define UF_RESOLVE_BENEATH 0x02 /* lookups must be beneath this dir */
152#define UF_FOCLOSE 0x04 /* auto-close on fork */
153
154#ifdef _KERNEL
155
156/* Lock a paths descriptor table. */
157#define PWDDESC_LOCK(pdp) (&(pdp)->pd_lock)
158#define PWDDESC_LOCK_INIT(pdp) \
159 mtx_init(PWDDESC_LOCK(pdp), "pwddesc", NULL, MTX_DEF)
160#define PWDDESC_LOCK_DESTROY(pdp) mtx_destroy(PWDDESC_LOCK(pdp))
161#define PWDDESC_XLOCK(pdp) mtx_lock(PWDDESC_LOCK(pdp))
162#define PWDDESC_XUNLOCK(pdp) mtx_unlock(PWDDESC_LOCK(pdp))
163#define PWDDESC_LOCK_ASSERT(pdp, what) \
164 mtx_assert(PWDDESC_LOCK(pdp), (what))
165#define PWDDESC_ASSERT_XLOCKED(pdp) \
166 PWDDESC_LOCK_ASSERT((pdp), MA_OWNED)
167#define PWDDESC_ASSERT_UNLOCKED(pdp) \
168 PWDDESC_LOCK_ASSERT((pdp), MA_NOTOWNED)
169
170#define PWDDESC_XLOCKED_LOAD_PWD(pdp) ({ \
171 struct pwddesc *_pdp = (pdp); \
172 struct pwd *_pwd; \
173 _pwd = smr_serialized_load(&(_pdp)->pd_pwd, \
174 (PWDDESC_ASSERT_XLOCKED(_pdp), true)); \
175 _pwd; \
176})
177
178/* Lock a file descriptor table. */
179#define FILEDESC_LOCK_INIT(fdp) sx_init(&(fdp)->fd_sx, "filedesc structure")
180#define FILEDESC_LOCK_DESTROY(fdp) sx_destroy(&(fdp)->fd_sx)
181#define FILEDESC_LOCK(fdp) (&(fdp)->fd_sx)
182#define FILEDESC_XLOCK(fdp) sx_xlock(&(fdp)->fd_sx)
183#define FILEDESC_XUNLOCK(fdp) sx_xunlock(&(fdp)->fd_sx)
184#define FILEDESC_SLOCK(fdp) sx_slock(&(fdp)->fd_sx)
185#define FILEDESC_SUNLOCK(fdp) sx_sunlock(&(fdp)->fd_sx)
186
187#define FILEDESC_LOCK_ASSERT(fdp) sx_assert(&(fdp)->fd_sx, SX_LOCKED | \
188 SX_NOTRECURSED)
189#define FILEDESC_XLOCK_ASSERT(fdp) sx_assert(&(fdp)->fd_sx, SX_XLOCKED | \
190 SX_NOTRECURSED)
191#define FILEDESC_UNLOCK_ASSERT(fdp) sx_assert(&(fdp)->fd_sx, SX_UNLOCKED)
192
193#define FILEDESC_IS_ONLY_USER(fdp) ({ \
194 struct filedesc *_fdp = (fdp); \
195 MPASS(curproc->p_fd == _fdp); \
196 (curproc->p_numthreads == 1 && refcount_load(&_fdp->fd_refcnt) == 1); \
197})
198
199#else
200
201/*
202 * Accessor for libkvm et al.
203 */
204#define PWDDESC_KVM_LOAD_PWD(pdp) ({ \
205 struct pwddesc *_pdp = (pdp); \
206 struct pwd *_pwd; \
207 _pwd = smr_kvm_load(&(_pdp)->pd_pwd); \
208 _pwd; \
209})
210
211#endif
212
213#ifdef _KERNEL
214
215/* Operation types for kern_dup(). */
216enum {
217 FDDUP_NORMAL, /* dup() behavior. */
218 FDDUP_FCNTL, /* fcntl()-style errors. */
219 FDDUP_FIXED, /* Force fixed allocation. */
220 FDDUP_LASTMODE,
221};
222
223/* Flags for kern_dup(). */
224#define FDDUP_FLAG_CLOEXEC 0x1 /* Atomically set UF_EXCLOSE. */
225#define FDDUP_FLAG_CLOFORK 0x2 /* Atomically set UF_FOCLOSE. */
226
227/* For backward compatibility. */
228#define falloc(td, resultfp, resultfd, flags) \
229 falloc_caps(td, resultfp, resultfd, flags, NULL)
230
231struct mount;
232struct thread;
233
234static __inline void
235filecaps_init(struct filecaps *fcaps)
236{
237
238 bzero(fcaps, sizeof(*fcaps));
239 fcaps->fc_nioctls = -1;
240}
241bool filecaps_copy(const struct filecaps *src, struct filecaps *dst,
242 bool locked);
243void filecaps_move(struct filecaps *src, struct filecaps *dst);
244void filecaps_free(struct filecaps *fcaps);
245
246int closef(struct file *fp, struct thread *td);
247void closef_nothread(struct file *fp);
248int descrip_check_write_mp(struct filedesc *fdp, struct mount *mp);
249int dupfdopen(struct thread *td, struct filedesc *fdp, int dfd, int mode,
250 int openerror, int *indxp);
251int falloc_caps(struct thread *td, struct file **resultfp, int *resultfd,
252 int flags, struct filecaps *fcaps);
253void falloc_abort(struct thread *td, struct file *fp);
254int _falloc_noinstall(struct thread *td, struct file **resultfp, u_int n);
255#define falloc_noinstall(td, resultfp) _falloc_noinstall(td, resultfp, 1)
256void _finstall(struct filedesc *fdp, struct file *fp, int fd, int flags,
257 struct filecaps *fcaps);
258int finstall(struct thread *td, struct file *fp, int *resultfd, int flags,
259 struct filecaps *fcaps);
260int finstall_refed(struct thread *td, struct file *fp, int *resultfd, int flags,
261 struct filecaps *fcaps);
262int fdalloc(struct thread *td, int minfd, int *result);
263int fdallocn(struct thread *td, int minfd, int *fds, int n);
264int fdcheckstd(struct thread *td);
265void fdclose(struct thread *td, struct file *fp, int idx);
266void fdcloseexec(struct thread *td);
267void fdsetugidsafety(struct thread *td);
268struct filedesc *fdcopy(struct filedesc *fdp);
269void fdunshare(struct thread *td);
270void fdescfree(struct thread *td);
271int fdlastfile(struct filedesc *fdp);
272int fdlastfile_single(struct filedesc *fdp);
273struct filedesc *fdinit(void);
274struct filedesc *fdshare(struct filedesc *fdp);
275struct filedesc_to_leader *
276 filedesc_to_leader_alloc(struct filedesc_to_leader *old,
277 struct filedesc *fdp, struct proc *leader);
278struct filedesc_to_leader *
279 filedesc_to_leader_share(struct filedesc_to_leader *fdtol,
280 struct filedesc *fdp);
281int getvnode(struct thread *td, int fd, const cap_rights_t *rightsp,
282 struct file **fpp);
283int getvnode_path(struct thread *td, int fd, const cap_rights_t *rightsp,
284 uint8_t *flagsp, struct file **fpp);
285void mountcheckdirs(struct vnode *olddp, struct vnode *newdp);
286
287int fget_cap_noref(struct filedesc *fdp, int fd,
288 const cap_rights_t *needrightsp, struct file **fpp,
289 struct filecaps *havecapsp);
290int fget_cap(struct thread *td, int fd, const cap_rights_t *needrightsp,
291 uint8_t *flagsp, struct file **fpp, struct filecaps *havecapsp);
292/* Return a referenced file from an unlocked descriptor. */
293int fget_unlocked(struct thread *td, int fd,
294 const cap_rights_t *needrightsp, struct file **fpp);
295int fget_unlocked_flags(struct thread *td, int fd,
296 const cap_rights_t *needrightsp, uint8_t *flagsp,
297 struct file **fpp);
298/* Return a file pointer without a ref. FILEDESC_IS_ONLY_USER must be true. */
299int fget_only_user(struct filedesc *fdp, int fd,
300 const cap_rights_t *needrightsp, struct file **fpp);
301#define fput_only_user(fdp, fp) ({ \
302 MPASS(FILEDESC_IS_ONLY_USER(fdp)); \
303 MPASS(refcount_load(&fp->f_count) > 0); \
304})
305
306/* Requires a FILEDESC_{S,X}LOCK held and returns without a ref. */
307static __inline struct file *
308fget_noref(struct filedesc *fdp, int fd)
309{
310
311 FILEDESC_LOCK_ASSERT(fdp);
312
313 if (__predict_false((u_int)fd >= (u_int)fdp->fd_nfiles))
314 return (NULL);
315
316 return (fdp->fd_ofiles[fd].fde_file);
317}
318
319static __inline struct filedescent *
320fdeget_noref(struct filedesc *fdp, int fd)
321{
322 struct filedescent *fde;
323
324 FILEDESC_LOCK_ASSERT(fdp);
325
326 if (__predict_false((u_int)fd >= (u_int)fdp->fd_nfiles))
327 return (NULL);
328
329 fde = &fdp->fd_ofiles[fd];
330 if (__predict_false(fde->fde_file == NULL))
331 return (NULL);
332
333 return (fde);
334}
335
336#ifdef CAPABILITIES
337static __inline bool
338fd_modified(struct filedesc *fdp, int fd, seqc_t seqc)
339{
340
341 return (!seqc_consistent(fd_seqc(fdp->fd_files, fd), seqc));
342}
343#endif
344
345int proc_nfiles(struct proc *p);
346
347/* cdir/rdir/jdir manipulation functions. */
348struct pwddesc *pdcopy(struct pwddesc *pdp);
349void pdescfree(struct thread *td);
350struct pwddesc *pdinit(struct pwddesc *pdp, bool keeplock);
351struct pwddesc *pdshare(struct pwddesc *pdp);
352void pdunshare(struct thread *td);
353
354void pwd_altroot(struct thread *td, struct vnode *altroot_vp);
355void pwd_chdir(struct thread *td, struct vnode *vp);
356int pwd_chroot(struct thread *td, struct vnode *vp);
357int pwd_chroot_chdir(struct thread *td, struct vnode *vp);
358void pwd_ensure_dirs(void);
359void pwd_set_rootvnode(void);
360
361struct pwd *pwd_hold_pwddesc(struct pwddesc *pdp);
362bool pwd_hold_smr(struct pwd *pwd);
363struct pwd *pwd_hold_proc(struct proc *p);
364struct pwd *pwd_hold(struct thread *td);
365void pwd_drop(struct pwd *pwd);
366static inline void
367pwd_set(struct pwddesc *pdp, struct pwd *newpwd)
368{
369 smr_serialized_store(&pdp->pd_pwd, newpwd,
370 (PWDDESC_ASSERT_XLOCKED(pdp), true));
371}
372#define pwd_get_smr() vfs_smr_entered_load(&curproc->p_pd->pd_pwd)
373
374#endif /* _KERNEL */
375
376#endif /* !_SYS_FILEDESC_H_ */