xref: /titanic_41/usr/src/uts/common/fs/tmpfs/tmp_vfsops.c (revision f841f6ad96ea6675d6c6b35c749eaac601799fdf)
1 /*
2  * CDDL HEADER START
3  *
4  * The contents of this file are subject to the terms of the
5  * Common Development and Distribution License (the "License").
6  * You may not use this file except in compliance with the License.
7  *
8  * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
9  * or http://www.opensolaris.org/os/licensing.
10  * See the License for the specific language governing permissions
11  * and limitations under the License.
12  *
13  * When distributing Covered Code, include this CDDL HEADER in each
14  * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
15  * If applicable, add the following below this CDDL HEADER, with the
16  * fields enclosed by brackets "[]" replaced with your own identifying
17  * information: Portions Copyright [yyyy] [name of copyright owner]
18  *
19  * CDDL HEADER END
20  */
21 /*
22  * Copyright 2006 Sun Microsystems, Inc.  All rights reserved.
23  * Use is subject to license terms.
24  */
25 
26 #pragma ident	"%Z%%M%	%I%	%E% SMI"
27 
28 #include <sys/types.h>
29 #include <sys/param.h>
30 #include <sys/sysmacros.h>
31 #include <sys/kmem.h>
32 #include <sys/time.h>
33 #include <sys/pathname.h>
34 #include <sys/vfs.h>
35 #include <sys/vnode.h>
36 #include <sys/stat.h>
37 #include <sys/uio.h>
38 #include <sys/stat.h>
39 #include <sys/errno.h>
40 #include <sys/cmn_err.h>
41 #include <sys/cred.h>
42 #include <sys/statvfs.h>
43 #include <sys/mount.h>
44 #include <sys/debug.h>
45 #include <sys/systm.h>
46 #include <sys/mntent.h>
47 #include <fs/fs_subr.h>
48 #include <vm/page.h>
49 #include <vm/anon.h>
50 #include <sys/model.h>
51 #include <sys/policy.h>
52 
53 #include <sys/fs/swapnode.h>
54 #include <sys/fs/tmp.h>
55 #include <sys/fs/tmpnode.h>
56 
57 static int tmpfsfstype;
58 
59 /*
60  * tmpfs vfs operations.
61  */
62 static int tmpfsinit(int, char *);
63 static int tmp_mount(struct vfs *, struct vnode *,
64 	struct mounta *, struct cred *);
65 static int tmp_unmount(struct vfs *, int, struct cred *);
66 static int tmp_root(struct vfs *, struct vnode **);
67 static int tmp_statvfs(struct vfs *, struct statvfs64 *);
68 static int tmp_vget(struct vfs *, struct vnode **, struct fid *);
69 
70 /*
71  * Loadable module wrapper
72  */
73 #include <sys/modctl.h>
74 
75 static mntopts_t tmpfs_proto_opttbl;
76 
77 static vfsdef_t vfw = {
78 	VFSDEF_VERSION,
79 	"tmpfs",
80 	tmpfsinit,
81 	VSW_HASPROTO|VSW_STATS,
82 	&tmpfs_proto_opttbl
83 };
84 
85 /*
86  * in-kernel mnttab options
87  */
88 static char *xattr_cancel[] = { MNTOPT_NOXATTR, NULL };
89 static char *noxattr_cancel[] = { MNTOPT_XATTR, NULL };
90 
91 static mntopt_t tmpfs_options[] = {
92 	/* Option name		Cancel Opt	Arg	Flags		Data */
93 	{ MNTOPT_XATTR,		xattr_cancel,	NULL,	MO_DEFAULT,	NULL},
94 	{ MNTOPT_NOXATTR,	noxattr_cancel,	NULL,	NULL,		NULL},
95 	{ "size",		NULL,		"0",	MO_HASVALUE,	NULL}
96 };
97 
98 
99 static mntopts_t tmpfs_proto_opttbl = {
100 	sizeof (tmpfs_options) / sizeof (mntopt_t),
101 	tmpfs_options
102 };
103 
104 /*
105  * Module linkage information
106  */
107 static struct modlfs modlfs = {
108 	&mod_fsops, "filesystem for tmpfs", &vfw
109 };
110 
111 static struct modlinkage modlinkage = {
112 	MODREV_1, &modlfs, NULL
113 };
114 
115 int
116 _init()
117 {
118 	return (mod_install(&modlinkage));
119 }
120 
121 int
122 _fini()
123 {
124 	int error;
125 
126 	error = mod_remove(&modlinkage);
127 	if (error)
128 		return (error);
129 	/*
130 	 * Tear down the operations vectors
131 	 */
132 	(void) vfs_freevfsops_by_type(tmpfsfstype);
133 	vn_freevnodeops(tmp_vnodeops);
134 	return (0);
135 }
136 
137 int
138 _info(struct modinfo *modinfop)
139 {
140 	return (mod_info(&modlinkage, modinfop));
141 }
142 
143 /*
144  * The following are patchable variables limiting the amount of system
145  * resources tmpfs can use.
146  *
147  * tmpfs_maxkmem limits the amount of kernel kmem_alloc memory
148  * tmpfs can use for it's data structures (e.g. tmpnodes, directory entries)
149  * It is not determined by setting a hard limit but rather as a percentage of
150  * physical memory which is determined when tmpfs is first used in the system.
151  *
152  * tmpfs_minfree is the minimum amount of swap space that tmpfs leaves for
153  * the rest of the system.  In other words, if the amount of free swap space
154  * in the system (i.e. anoninfo.ani_free) drops below tmpfs_minfree, tmpfs
155  * anon allocations will fail.
156  *
157  * There is also a per mount limit on the amount of swap space
158  * (tmount.tm_anonmax) settable via a mount option.
159  */
160 size_t tmpfs_maxkmem = 0;
161 size_t tmpfs_minfree = 0;
162 size_t tmp_kmemspace;		/* bytes of kernel heap used by all tmpfs */
163 
164 static major_t tmpfs_major;
165 static minor_t tmpfs_minor;
166 static kmutex_t	tmpfs_minor_lock;
167 
168 /*
169  * initialize global tmpfs locks and such
170  * called when loading tmpfs module
171  */
172 static int
173 tmpfsinit(int fstype, char *name)
174 {
175 	static const fs_operation_def_t tmp_vfsops_template[] = {
176 		VFSNAME_MOUNT, tmp_mount,
177 		VFSNAME_UNMOUNT, tmp_unmount,
178 		VFSNAME_ROOT, tmp_root,
179 		VFSNAME_STATVFS, tmp_statvfs,
180 		VFSNAME_VGET, tmp_vget,
181 		NULL, NULL
182 	};
183 	int error;
184 	extern  void    tmpfs_hash_init();
185 
186 	tmpfs_hash_init();
187 	tmpfsfstype = fstype;
188 	ASSERT(tmpfsfstype != 0);
189 
190 	error = vfs_setfsops(fstype, tmp_vfsops_template, NULL);
191 	if (error != 0) {
192 		cmn_err(CE_WARN, "tmpfsinit: bad vfs ops template");
193 		return (error);
194 	}
195 
196 	error = vn_make_ops(name, tmp_vnodeops_template, &tmp_vnodeops);
197 	if (error != 0) {
198 		(void) vfs_freevfsops_by_type(fstype);
199 		cmn_err(CE_WARN, "tmpfsinit: bad vnode ops template");
200 		return (error);
201 	}
202 
203 	/*
204 	 * tmpfs_minfree doesn't need to be some function of configured
205 	 * swap space since it really is an absolute limit of swap space
206 	 * which still allows other processes to execute.
207 	 */
208 	if (tmpfs_minfree == 0) {
209 		/*
210 		 * Set if not patched
211 		 */
212 		tmpfs_minfree = btopr(TMPMINFREE);
213 	}
214 
215 	/*
216 	 * The maximum amount of space tmpfs can allocate is
217 	 * TMPMAXPROCKMEM percent of kernel memory
218 	 */
219 	if (tmpfs_maxkmem == 0)
220 		tmpfs_maxkmem = MAX(PAGESIZE, kmem_maxavail() / TMPMAXFRACKMEM);
221 
222 	if ((tmpfs_major = getudev()) == (major_t)-1) {
223 		cmn_err(CE_WARN, "tmpfsinit: Can't get unique device number.");
224 		tmpfs_major = 0;
225 	}
226 	mutex_init(&tmpfs_minor_lock, NULL, MUTEX_DEFAULT, NULL);
227 	return (0);
228 }
229 
230 static int
231 tmp_mount(
232 	struct vfs *vfsp,
233 	struct vnode *mvp,
234 	struct mounta *uap,
235 	struct cred *cr)
236 {
237 	struct tmount *tm = NULL;
238 	struct tmpnode *tp;
239 	struct pathname dpn;
240 	int error;
241 	pgcnt_t anonmax;
242 	struct vattr rattr;
243 	int got_attrs;
244 
245 	char *sizestr;
246 
247 	if ((error = secpolicy_fs_mount(cr, mvp, vfsp)) != 0)
248 		return (error);
249 
250 	if (mvp->v_type != VDIR)
251 		return (ENOTDIR);
252 
253 	mutex_enter(&mvp->v_lock);
254 	if ((uap->flags & MS_OVERLAY) == 0 &&
255 	    (mvp->v_count != 1 || (mvp->v_flag & VROOT))) {
256 		mutex_exit(&mvp->v_lock);
257 		return (EBUSY);
258 	}
259 	mutex_exit(&mvp->v_lock);
260 
261 	/*
262 	 * Having the resource be anything but "swap" doesn't make sense.
263 	 */
264 	vfs_setresource(vfsp, "swap");
265 
266 	/*
267 	 * now look for options we understand...
268 	 */
269 
270 	/* tmpfs doesn't support read-only mounts */
271 	if (vfs_optionisset(vfsp, MNTOPT_RO, NULL)) {
272 		error = EINVAL;
273 		goto out;
274 	}
275 
276 	/*
277 	 * tm_anonmax is set according to the mount arguments
278 	 * if any.  Otherwise, it is set to a maximum value.
279 	 */
280 	if (vfs_optionisset(vfsp, "size", &sizestr)) {
281 		if ((error = tmp_convnum(sizestr, &anonmax)) != 0)
282 			goto out;
283 	} else {
284 		anonmax = ULONG_MAX;
285 	}
286 
287 	if (error = pn_get(uap->dir,
288 	    (uap->flags & MS_SYSSPACE) ? UIO_SYSSPACE : UIO_USERSPACE, &dpn))
289 		goto out;
290 
291 	if ((tm = tmp_memalloc(sizeof (struct tmount), 0)) == NULL) {
292 		pn_free(&dpn);
293 		error = ENOMEM;
294 		goto out;
295 	}
296 
297 	/*
298 	 * find an available minor device number for this mount
299 	 */
300 	mutex_enter(&tmpfs_minor_lock);
301 	do {
302 		tmpfs_minor = (tmpfs_minor + 1) & L_MAXMIN32;
303 		tm->tm_dev = makedevice(tmpfs_major, tmpfs_minor);
304 	} while (vfs_devismounted(tm->tm_dev));
305 	mutex_exit(&tmpfs_minor_lock);
306 
307 	/*
308 	 * Set but don't bother entering the mutex
309 	 * (tmount not on mount list yet)
310 	 */
311 	mutex_init(&tm->tm_contents, NULL, MUTEX_DEFAULT, NULL);
312 	mutex_init(&tm->tm_renamelck, NULL, MUTEX_DEFAULT, NULL);
313 
314 	tm->tm_vfsp = vfsp;
315 	tm->tm_anonmax = anonmax;
316 
317 	vfsp->vfs_data = (caddr_t)tm;
318 	vfsp->vfs_fstype = tmpfsfstype;
319 	vfsp->vfs_dev = tm->tm_dev;
320 	vfsp->vfs_bsize = PAGESIZE;
321 	vfsp->vfs_flag |= VFS_NOTRUNC;
322 	vfs_make_fsid(&vfsp->vfs_fsid, tm->tm_dev, tmpfsfstype);
323 	tm->tm_mntpath = tmp_memalloc(dpn.pn_pathlen + 1, TMP_MUSTHAVE);
324 	(void) strcpy(tm->tm_mntpath, dpn.pn_path);
325 
326 	/*
327 	 * allocate and initialize root tmpnode structure
328 	 */
329 	bzero(&rattr, sizeof (struct vattr));
330 	rattr.va_mode = (mode_t)(S_IFDIR | 0777);	/* XXX modes */
331 	rattr.va_type = VDIR;
332 	rattr.va_rdev = 0;
333 	tp = tmp_memalloc(sizeof (struct tmpnode), TMP_MUSTHAVE);
334 	tmpnode_init(tm, tp, &rattr, cr);
335 
336 	/*
337 	 * Get the mode, uid, and gid from the underlying mount point.
338 	 */
339 	rattr.va_mask = AT_MODE|AT_UID|AT_GID;	/* Hint to getattr */
340 	got_attrs = VOP_GETATTR(mvp, &rattr, 0, cr);
341 
342 	rw_enter(&tp->tn_rwlock, RW_WRITER);
343 	TNTOV(tp)->v_flag |= VROOT;
344 
345 	/*
346 	 * If the getattr succeeded, use its results.  Otherwise allow
347 	 * the previously set hardwired defaults to prevail.
348 	 */
349 	if (got_attrs == 0) {
350 		tp->tn_mode = rattr.va_mode;
351 		tp->tn_uid = rattr.va_uid;
352 		tp->tn_gid = rattr.va_gid;
353 	}
354 
355 	/*
356 	 * initialize linked list of tmpnodes so that the back pointer of
357 	 * the root tmpnode always points to the last one on the list
358 	 * and the forward pointer of the last node is null
359 	 */
360 	tp->tn_back = tp;
361 	tp->tn_forw = NULL;
362 	tp->tn_nlink = 0;
363 	tm->tm_rootnode = tp;
364 
365 	tdirinit(tp, tp);
366 
367 	rw_exit(&tp->tn_rwlock);
368 
369 	pn_free(&dpn);
370 	error = 0;
371 
372 out:
373 	return (error);
374 }
375 
376 static int
377 tmp_unmount(struct vfs *vfsp, int flag, struct cred *cr)
378 {
379 	struct tmount *tm = (struct tmount *)VFSTOTM(vfsp);
380 	struct tmpnode *tnp, *cancel;
381 	struct vnode	*vp;
382 	int error;
383 
384 	if ((error = secpolicy_fs_unmount(cr, vfsp)) != 0)
385 		return (error);
386 
387 	/*
388 	 * forced unmount is not supported by this file system
389 	 * and thus, ENOTSUP, is being returned.
390 	 */
391 	if (flag & MS_FORCE)
392 		return (ENOTSUP);
393 
394 	mutex_enter(&tm->tm_contents);
395 
396 	/*
397 	 * If there are no open files, only the root node should have
398 	 * a reference count.
399 	 * With tm_contents held, nothing can be added or removed.
400 	 * There may be some dirty pages.  To prevent fsflush from
401 	 * disrupting the unmount, put a hold on each node while scanning.
402 	 * If we find a previously referenced node, undo the holds we have
403 	 * placed and fail EBUSY.
404 	 */
405 	tnp = tm->tm_rootnode;
406 	if (TNTOV(tnp)->v_count > 1) {
407 		mutex_exit(&tm->tm_contents);
408 		return (EBUSY);
409 	}
410 
411 	for (tnp = tnp->tn_forw; tnp; tnp = tnp->tn_forw) {
412 		if ((vp = TNTOV(tnp))->v_count > 0) {
413 			cancel = tm->tm_rootnode->tn_forw;
414 			while (cancel != tnp) {
415 				vp = TNTOV(cancel);
416 				ASSERT(vp->v_count > 0);
417 				VN_RELE(vp);
418 				cancel = cancel->tn_forw;
419 			}
420 			mutex_exit(&tm->tm_contents);
421 			return (EBUSY);
422 		}
423 		VN_HOLD(vp);
424 	}
425 
426 	/*
427 	 * We can drop the mutex now because no one can find this mount
428 	 */
429 	mutex_exit(&tm->tm_contents);
430 
431 	/*
432 	 * Free all kmemalloc'd and anonalloc'd memory associated with
433 	 * this filesystem.  To do this, we go through the file list twice,
434 	 * once to remove all the directory entries, and then to remove
435 	 * all the files.  We do this because there is useful code in
436 	 * tmpnode_free which assumes that the directory entry has been
437 	 * removed before the file.
438 	 */
439 	/*
440 	 * Remove all directory entries
441 	 */
442 	for (tnp = tm->tm_rootnode; tnp; tnp = tnp->tn_forw) {
443 		rw_enter(&tnp->tn_rwlock, RW_WRITER);
444 		if (tnp->tn_type == VDIR)
445 			tdirtrunc(tnp);
446 		if (tnp->tn_vnode->v_flag & V_XATTRDIR) {
447 			/*
448 			 * Account for implicit attrdir reference.
449 			 */
450 			ASSERT(tnp->tn_nlink > 0);
451 			DECR_COUNT(&tnp->tn_nlink, &tnp->tn_tlock);
452 		}
453 		rw_exit(&tnp->tn_rwlock);
454 	}
455 
456 	ASSERT(tm->tm_rootnode);
457 
458 	/*
459 	 * All links are gone, v_count is keeping nodes in place.
460 	 * VN_RELE should make the node disappear, unless somebody
461 	 * is holding pages against it.  Nap and retry until it disappears.
462 	 *
463 	 * We re-acquire the lock to prevent others who have a HOLD on
464 	 * a tmpnode via its pages or anon slots from blowing it away
465 	 * (in tmp_inactive) while we're trying to get to it here. Once
466 	 * we have a HOLD on it we know it'll stick around.
467 	 *
468 	 */
469 	mutex_enter(&tm->tm_contents);
470 	/*
471 	 * Remove all the files (except the rootnode) backwards.
472 	 */
473 	while ((tnp = tm->tm_rootnode->tn_back) != tm->tm_rootnode) {
474 		mutex_exit(&tm->tm_contents);
475 		/*
476 		 * Inhibit tmp_inactive from touching attribute directory
477 		 * as all nodes will be released here.
478 		 * Note we handled the link count in pass 2 above.
479 		 */
480 		rw_enter(&tnp->tn_rwlock, RW_WRITER);
481 		tnp->tn_xattrdp = NULL;
482 		rw_exit(&tnp->tn_rwlock);
483 		vp = TNTOV(tnp);
484 		VN_RELE(vp);
485 		mutex_enter(&tm->tm_contents);
486 		/*
487 		 * It's still there after the RELE. Someone else like pageout
488 		 * has a hold on it so wait a bit and then try again - we know
489 		 * they'll give it up soon.
490 		 */
491 		if (tnp == tm->tm_rootnode->tn_back) {
492 			VN_HOLD(vp);
493 			mutex_exit(&tm->tm_contents);
494 			delay(hz / 4);
495 			mutex_enter(&tm->tm_contents);
496 		}
497 	}
498 	mutex_exit(&tm->tm_contents);
499 
500 	tm->tm_rootnode->tn_xattrdp = NULL;
501 	VN_RELE(TNTOV(tm->tm_rootnode));
502 
503 	ASSERT(tm->tm_mntpath);
504 
505 	tmp_memfree(tm->tm_mntpath, strlen(tm->tm_mntpath) + 1);
506 
507 	ASSERT(tm->tm_anonmem == 0);
508 
509 	mutex_destroy(&tm->tm_contents);
510 	mutex_destroy(&tm->tm_renamelck);
511 	tmp_memfree(tm, sizeof (struct tmount));
512 
513 	return (0);
514 }
515 
516 /*
517  * return root tmpnode for given vnode
518  */
519 static int
520 tmp_root(struct vfs *vfsp, struct vnode **vpp)
521 {
522 	struct tmount *tm = (struct tmount *)VFSTOTM(vfsp);
523 	struct tmpnode *tp = tm->tm_rootnode;
524 	struct vnode *vp;
525 
526 	ASSERT(tp);
527 
528 	vp = TNTOV(tp);
529 	VN_HOLD(vp);
530 	*vpp = vp;
531 	return (0);
532 }
533 
534 static int
535 tmp_statvfs(struct vfs *vfsp, struct statvfs64 *sbp)
536 {
537 	struct tmount	*tm = (struct tmount *)VFSTOTM(vfsp);
538 	ulong_t	blocks;
539 	dev32_t d32;
540 
541 	sbp->f_bsize = PAGESIZE;
542 	sbp->f_frsize = PAGESIZE;
543 
544 	/*
545 	 * Find the amount of available physical and memory swap
546 	 */
547 	mutex_enter(&anoninfo_lock);
548 	ASSERT(k_anoninfo.ani_max >= k_anoninfo.ani_phys_resv);
549 	blocks = (ulong_t)CURRENT_TOTAL_AVAILABLE_SWAP;
550 	mutex_exit(&anoninfo_lock);
551 
552 	/*
553 	 * If tm_anonmax for this mount is less than the available swap space
554 	 * (minus the amount tmpfs can't use), use that instead
555 	 */
556 	if (blocks > tmpfs_minfree)
557 		sbp->f_bfree = MIN(blocks - tmpfs_minfree,
558 		    tm->tm_anonmax - tm->tm_anonmem);
559 	else
560 		sbp->f_bfree = 0;
561 
562 	sbp->f_bavail = sbp->f_bfree;
563 
564 	/*
565 	 * Total number of blocks is what's available plus what's been used
566 	 */
567 	sbp->f_blocks = (fsblkcnt64_t)(sbp->f_bfree + tm->tm_anonmem);
568 
569 	/*
570 	 * The maximum number of files available is approximately the number
571 	 * of tmpnodes we can allocate from the remaining kernel memory
572 	 * available to tmpfs.  This is fairly inaccurate since it doesn't
573 	 * take into account the names stored in the directory entries.
574 	 */
575 	if (tmpfs_maxkmem > tmp_kmemspace)
576 		sbp->f_ffree = (tmpfs_maxkmem - tmp_kmemspace) /
577 		    (sizeof (struct tmpnode) + sizeof (struct tdirent));
578 	else
579 		sbp->f_ffree = 0;
580 
581 	sbp->f_files = tmpfs_maxkmem /
582 	    (sizeof (struct tmpnode) + sizeof (struct tdirent));
583 	sbp->f_favail = (fsfilcnt64_t)(sbp->f_ffree);
584 	(void) cmpldev(&d32, vfsp->vfs_dev);
585 	sbp->f_fsid = d32;
586 	(void) strcpy(sbp->f_basetype, vfssw[tmpfsfstype].vsw_name);
587 	(void) strncpy(sbp->f_fstr, tm->tm_mntpath, sizeof (sbp->f_fstr));
588 	/*
589 	 * ensure null termination
590 	 */
591 	sbp->f_fstr[sizeof (sbp->f_fstr) - 1] = '\0';
592 	sbp->f_flag = vf_to_stf(vfsp->vfs_flag);
593 	sbp->f_namemax = MAXNAMELEN - 1;
594 	return (0);
595 }
596 
597 static int
598 tmp_vget(struct vfs *vfsp, struct vnode **vpp, struct fid *fidp)
599 {
600 	struct tfid *tfid;
601 	struct tmount *tm = (struct tmount *)VFSTOTM(vfsp);
602 	struct tmpnode *tp = NULL;
603 
604 	tfid = (struct tfid *)fidp;
605 	*vpp = NULL;
606 
607 	mutex_enter(&tm->tm_contents);
608 	for (tp = tm->tm_rootnode; tp; tp = tp->tn_forw) {
609 		mutex_enter(&tp->tn_tlock);
610 		if (tp->tn_nodeid == tfid->tfid_ino) {
611 			/*
612 			 * If the gen numbers don't match we know the
613 			 * file won't be found since only one tmpnode
614 			 * can have this number at a time.
615 			 */
616 			if (tp->tn_gen != tfid->tfid_gen || tp->tn_nlink == 0) {
617 				mutex_exit(&tp->tn_tlock);
618 				mutex_exit(&tm->tm_contents);
619 				return (0);
620 			}
621 			*vpp = (struct vnode *)TNTOV(tp);
622 
623 			VN_HOLD(*vpp);
624 
625 			if ((tp->tn_mode & S_ISVTX) &&
626 			    !(tp->tn_mode & (S_IXUSR | S_IFDIR))) {
627 				mutex_enter(&(*vpp)->v_lock);
628 				(*vpp)->v_flag |= VISSWAP;
629 				mutex_exit(&(*vpp)->v_lock);
630 			}
631 			mutex_exit(&tp->tn_tlock);
632 			mutex_exit(&tm->tm_contents);
633 			return (0);
634 		}
635 		mutex_exit(&tp->tn_tlock);
636 	}
637 	mutex_exit(&tm->tm_contents);
638 	return (0);
639 }
640