xref: /titanic_41/usr/src/uts/common/fs/tmpfs/tmp_tnode.c (revision 1d4b38e0077763e7c9b20768eacb841957e787bc)
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, Version 1.0 only
6  * (the "License").  You may not use this file except in compliance
7  * with the License.
8  *
9  * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
10  * or http://www.opensolaris.org/os/licensing.
11  * See the License for the specific language governing permissions
12  * and limitations under the License.
13  *
14  * When distributing Covered Code, include this CDDL HEADER in each
15  * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
16  * If applicable, add the following below this CDDL HEADER, with the
17  * fields enclosed by brackets "[]" replaced with your own identifying
18  * information: Portions Copyright [yyyy] [name of copyright owner]
19  *
20  * CDDL HEADER END
21  */
22 /*
23  * Copyright 2004 Sun Microsystems, Inc.  All rights reserved.
24  * Use is subject to license terms.
25  */
26 
27 #pragma ident	"%Z%%M%	%I%	%E% SMI"
28 
29 #include <sys/types.h>
30 #include <sys/param.h>
31 #include <sys/t_lock.h>
32 #include <sys/systm.h>
33 #include <sys/time.h>
34 #include <sys/sysmacros.h>
35 #include <sys/proc.h>
36 #include <sys/disp.h>
37 #include <sys/user.h>
38 #include <sys/time.h>
39 #include <sys/vfs.h>
40 #include <sys/vnode.h>
41 #include <sys/stat.h>
42 #include <sys/mode.h>
43 #include <sys/errno.h>
44 #include <sys/kmem.h>
45 #include <vm/seg.h>
46 #include <vm/seg_map.h>
47 #include <vm/anon.h>
48 #include <vm/page.h>
49 #include <vm/pvn.h>
50 #include <sys/fs/tmp.h>
51 #include <sys/fs/tmpnode.h>
52 #include <sys/debug.h>
53 #include <sys/cmn_err.h>
54 #include <sys/swap.h>
55 #include <sys/vtrace.h>
56 
57 /*
58  * Reserve swap space for the size of the file.
59  * Called before growing a file (i.e. ftruncate, write)
60  * Returns 0 on success.
61  */
62 int
63 tmp_resv(
64 	struct tmount *tm,
65 	struct tmpnode *tp,
66 	size_t delta,		/* size needed */
67 	int pagecreate)		/* call anon_resv if set */
68 {
69 	pgcnt_t pages = btopr(delta);
70 
71 	ASSERT(RW_WRITE_HELD(&tp->tn_rwlock));
72 	ASSERT(tp->tn_type == VREG);
73 	/*
74 	 * pagecreate is set only if we actually need to call anon_resv
75 	 * to reserve an additional page of anonymous memory.
76 	 * Since anon_resv always reserves a page at a time,
77 	 * it should only get called when we know we're growing the
78 	 * file into a new page or filling a hole.
79 	 *
80 	 * Deny if trying to reserve more than tmpfs can allocate
81 	 */
82 	if (pagecreate && ((tm->tm_anonmem + pages > tm->tm_anonmax) ||
83 	    (!anon_checkspace(ptob(pages + tmpfs_minfree))) ||
84 	    (anon_resv(delta) == 0))) {
85 		return (1);
86 	}
87 
88 	/*
89 	 * update statistics
90 	 */
91 	if (pagecreate) {
92 		mutex_enter(&tm->tm_contents);
93 		tm->tm_anonmem += pages;
94 		mutex_exit(&tm->tm_contents);
95 
96 		TRACE_2(TR_FAC_VM, TR_ANON_TMPFS, "anon tmpfs:%p %lu",
97 		    tp, delta);
98 	}
99 
100 	return (0);
101 }
102 
103 /*
104  * tmp_unresv - called when truncating a file
105  * Only called if we're freeing at least pagesize bytes
106  * because anon_unresv does a btopr(delta)
107  */
108 static void
109 tmp_unresv(
110 	struct tmount *tm,
111 	struct tmpnode *tp,
112 	size_t delta)
113 {
114 	ASSERT(RW_WRITE_HELD(&tp->tn_rwlock));
115 	ASSERT(tp->tn_type == VREG);
116 
117 	anon_unresv(delta);
118 
119 	mutex_enter(&tm->tm_contents);
120 	tm->tm_anonmem -= btopr(delta);
121 	mutex_exit(&tm->tm_contents);
122 
123 	TRACE_2(TR_FAC_VM, TR_ANON_TMPFS, "anon tmpfs:%p %lu", tp, delta);
124 }
125 
126 #define	TMP_INIT_SZ	128
127 
128 /*
129  * Grow the anon pointer array to cover 'newsize' bytes plus slack.
130  */
131 void
132 tmpnode_growmap(struct tmpnode *tp, ulong_t newsize)
133 {
134 	pgcnt_t np = btopr(newsize);
135 
136 	ASSERT(RW_WRITE_HELD(&tp->tn_rwlock));
137 	ASSERT(RW_WRITE_HELD(&tp->tn_contents));
138 	ASSERT(tp->tn_type == VREG);
139 
140 	if (tp->tn_asize >= np)
141 		return;
142 
143 	if (newsize > MAXOFF_T)
144 		np = btopr(MAXOFF_T);
145 
146 	if (tp->tn_anon == NULL) {
147 		tp->tn_anon = anon_create(MAX(np, TMP_INIT_SZ), ANON_SLEEP);
148 		tp->tn_asize = tp->tn_anon->size;
149 		return;
150 	}
151 
152 	tp->tn_asize = anon_grow(tp->tn_anon, NULL, tp->tn_asize,
153 	    np - tp->tn_asize, ANON_SLEEP);
154 	ASSERT(tp->tn_asize >= np);
155 }
156 
157 /*
158  * Initialize a tmpnode and add it to file list under mount point.
159  */
160 void
161 tmpnode_init(struct tmount *tm, struct tmpnode *t, vattr_t *vap, cred_t *cred)
162 {
163 	struct vnode *vp;
164 	timestruc_t now;
165 
166 	ASSERT(vap != NULL);
167 
168 	rw_init(&t->tn_rwlock, NULL, RW_DEFAULT, NULL);
169 	mutex_init(&t->tn_tlock, NULL, MUTEX_DEFAULT, NULL);
170 	t->tn_mode = MAKEIMODE(vap->va_type, vap->va_mode);
171 	t->tn_mask = 0;
172 	t->tn_type = vap->va_type;
173 	t->tn_nodeid = (ino64_t)(uint32_t)((uintptr_t)t >> 3);
174 	t->tn_nlink = 1;
175 	t->tn_size = 0;
176 
177 	if (cred == NULL) {
178 		t->tn_uid = vap->va_uid;
179 		t->tn_gid = vap->va_gid;
180 	} else {
181 		t->tn_uid = crgetuid(cred);
182 		t->tn_gid = crgetgid(cred);
183 	}
184 
185 	t->tn_fsid = tm->tm_dev;
186 	t->tn_rdev = vap->va_rdev;
187 	t->tn_blksize = PAGESIZE;
188 	t->tn_nblocks = 0;
189 	gethrestime(&now);
190 	t->tn_atime = now;
191 	t->tn_mtime = now;
192 	t->tn_ctime = now;
193 	t->tn_seq = 0;
194 	t->tn_dir = NULL;
195 
196 	t->tn_vnode = vn_alloc(KM_SLEEP);
197 	vp = TNTOV(t);
198 	vn_setops(vp, tmp_vnodeops);
199 	vp->v_vfsp = tm->tm_vfsp;
200 	vp->v_type = vap->va_type;
201 	vp->v_rdev = vap->va_rdev;
202 	vp->v_data = (caddr_t)t;
203 	mutex_enter(&tm->tm_contents);
204 	/*
205 	 * Increment the pseudo generation number for this tmpnode.
206 	 * Since tmpnodes are allocated and freed, there really is no
207 	 * particular generation number for a new tmpnode.  Just fake it
208 	 * by using a counter in each file system.
209 	 */
210 	t->tn_gen = tm->tm_gen++;
211 
212 	/*
213 	 * Add new tmpnode to end of linked list of tmpnodes for this tmpfs
214 	 * Root directory is handled specially in tmp_mount.
215 	 */
216 	if (tm->tm_rootnode != (struct tmpnode *)NULL) {
217 		t->tn_forw = NULL;
218 		t->tn_back = tm->tm_rootnode->tn_back;
219 		t->tn_back->tn_forw = tm->tm_rootnode->tn_back = t;
220 	}
221 	mutex_exit(&tm->tm_contents);
222 	vn_exists(vp);
223 }
224 
225 /*
226  * tmpnode_trunc - set length of tmpnode and deal with resources
227  */
228 int
229 tmpnode_trunc(
230 	struct tmount *tm,
231 	struct tmpnode *tp,
232 	ulong_t newsize)
233 {
234 	size_t oldsize = tp->tn_size;
235 	size_t delta;
236 	struct vnode *vp = TNTOV(tp);
237 	timestruc_t now;
238 	int error = 0;
239 
240 	ASSERT(RW_WRITE_HELD(&tp->tn_rwlock));
241 	ASSERT(RW_WRITE_HELD(&tp->tn_contents));
242 
243 	if (newsize == oldsize) {
244 		/* Required by POSIX */
245 		goto stamp_out;
246 	}
247 
248 	switch (tp->tn_type) {
249 	case VREG:
250 		/* Growing the file */
251 		if (newsize > oldsize) {
252 			delta = P2ROUNDUP(newsize, PAGESIZE) -
253 			    P2ROUNDUP(oldsize, PAGESIZE);
254 			/*
255 			 * Grow the size of the anon array to the new size
256 			 * Reserve the space for the growth here.
257 			 * We do it this way for now because this is how
258 			 * tmpfs used to do it, and this way the reserved
259 			 * space is alway equal to the file size.
260 			 * Alternatively, we could wait to reserve space 'til
261 			 * someone tries to store into one of the newly
262 			 * trunc'ed up pages. This would give us behavior
263 			 * identical to ufs; i.e., you could fail a
264 			 * fault on storing into a holey region of a file
265 			 * if there is no space in the filesystem to fill
266 			 * the hole at that time.
267 			 */
268 			/*
269 			 * tmp_resv calls anon_resv only if we're extending
270 			 * the file into a new page
271 			 */
272 			if (tmp_resv(tm, tp, delta,
273 			    (btopr(newsize) != btopr(oldsize)))) {
274 				error = ENOSPC;
275 				goto out;
276 			}
277 			tmpnode_growmap(tp, newsize);
278 			tp->tn_size = newsize;
279 			break;
280 		}
281 
282 		/* Free anon pages if shrinking file over page boundary. */
283 		if (btopr(newsize) != btopr(oldsize)) {
284 			pgcnt_t freed;
285 			delta = P2ROUNDUP(oldsize, PAGESIZE) -
286 			    P2ROUNDUP(newsize, PAGESIZE);
287 			freed = anon_pages(tp->tn_anon, btopr(newsize),
288 			    btopr(delta));
289 			tp->tn_nblocks -= freed;
290 			anon_free(tp->tn_anon, btopr(newsize), delta);
291 			tmp_unresv(tm, tp, delta);
292 		}
293 
294 		/*
295 		 * Update the file size now to reflect the pages we just
296 		 * blew away as we're about to drop the
297 		 * contents lock to zero the partial page (which could
298 		 * re-enter tmpfs via getpage and try to reacquire the lock)
299 		 * Once we drop the lock, faulters can fill in holes in
300 		 * the file and if we haven't updated the size they
301 		 * may fill in holes that are beyond EOF, which will then
302 		 * never get cleared.
303 		 */
304 		tp->tn_size = newsize;
305 
306 		/* Zero new size of file to page boundary. */
307 		if (anon_get_ptr(tp->tn_anon, btop(newsize)) != NULL) {
308 			size_t zlen;
309 
310 			zlen = PAGESIZE - ((ulong_t)newsize & PAGEOFFSET);
311 			rw_exit(&tp->tn_contents);
312 			pvn_vpzero(TNTOV(tp), (u_offset_t)newsize, zlen);
313 			rw_enter(&tp->tn_contents, RW_WRITER);
314 		}
315 
316 		if (newsize == 0) {
317 			/* Delete anon array for tmpnode */
318 			ASSERT(tp->tn_nblocks == 0);
319 			ASSERT(anon_get_ptr(tp->tn_anon, 0) == NULL);
320 			ASSERT(!vn_has_cached_data(vp));
321 
322 			anon_release(tp->tn_anon, tp->tn_asize);
323 			tp->tn_anon = NULL;
324 			tp->tn_asize = 0;
325 		}
326 		break;
327 	case VLNK:
328 		/*
329 		 * Don't do anything here
330 		 * tmp_inactive frees the memory
331 		 */
332 		if (newsize != 0)
333 			error = EINVAL;
334 		goto out;
335 	case VDIR:
336 		/*
337 		 * Remove all the directory entries under this directory.
338 		 */
339 		if (newsize != 0) {
340 			error = EINVAL;
341 			goto out;
342 		}
343 		tdirtrunc(tp);
344 		ASSERT(tp->tn_nlink == 0);
345 		break;
346 	default:
347 		goto out;
348 	}
349 
350 stamp_out:
351 	gethrestime(&now);
352 	tp->tn_mtime = now;
353 	tp->tn_ctime = now;
354 out:
355 	/*
356 	 * tmpnode_trunc() cannot fail when newsize == 0.
357 	 */
358 	ASSERT(error == 0 || newsize != 0);
359 	return (error);
360 }
361