xref: /illumos-gate/usr/src/uts/common/fs/ufs/ufs_snap.c (revision d48be21240dfd051b689384ce2b23479d757f2d8)
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 #include <sys/systm.h>
28 #include <sys/types.h>
29 #include <sys/vnode.h>
30 #include <sys/file.h>
31 #include <sys/buf.h>
32 #include <sys/ddi.h>
33 #include <sys/errno.h>
34 #include <sys/cmn_err.h>
35 #include <sys/fs/ufs_inode.h>
36 #include <sys/fs/ufs_filio.h>
37 #include <sys/fs/ufs_snap.h>
38 #include <sys/fssnap_if.h>
39 #include <sys/sysmacros.h>
40 #include <sys/modctl.h>
41 #include <sys/fs/ufs_bio.h>
42 #include <sys/debug.h>
43 #include <sys/kmem.h>
44 #include <sys/inttypes.h>
45 #include <sys/vfs.h>
46 #include <sys/disp.h>
47 #include <sys/atomic.h>
48 #include <sys/conf.h>
49 #include <sys/param.h>
50 #include <sys/policy.h>
51 
52 static int ufs_snap_init_backfile(int  *, int, vnode_t ***, struct ufsvfs *);
53 static void release_backing_vnodes(vnode_t ***, int);
54 static int ufs_snap_find_candidates(void *, struct ufsvfs *, int);
55 
56 /*
57  * Create a snapshot on a file system
58  */
59 int
60 ufs_snap_create(struct vnode *vp, struct fiosnapcreate_multi *fiosnapp,
61     cred_t *cr)
62 {
63 	int		error = 0;
64 	struct ufsvfs	*ufsvfsp = VTOI(vp)->i_ufsvfs;
65 	struct fs	*fs = ufsvfsp->vfs_fs;
66 	vnode_t		**bfvpp = NULL;
67 	struct lockfs	lf;
68 	void		*snapid = NULL;
69 
70 	u_offset_t	nchunks;
71 	uint_t		chunksize, fragsperchunk;
72 
73 	/*
74 	 * Only privilege processes can create a snapshot for now.  This
75 	 * would be better if it was based on the permissions of the device
76 	 * file.
77 	 */
78 	if (secpolicy_fs_config(cr, ufsvfsp->vfs_vfs) != 0)
79 		return (EPERM);
80 
81 	/*
82 	 * There is no reason to make a snapshot of a read-only file system
83 	 */
84 	if (fs->fs_ronly) {
85 		fiosnapp->error = FIOCOW_EREADONLY;
86 		return (EROFS);
87 	}
88 
89 	/*
90 	 * Initialize the backing files to store old data.  This assumes any
91 	 * preallocation and setup has been done already.
92 	 * ufs_snap_init_backfile() allocates and returns a pointer to
93 	 * a null-terminated array of vnodes in bfvpp.
94 	 */
95 	error = ufs_snap_init_backfile(fiosnapp->backfiledesc,
96 	    fiosnapp->backfilecount, &bfvpp, ufsvfsp);
97 	if (error) {
98 		fiosnapp->error = FIOCOW_EBACKFILE;
99 		return (error);
100 	}
101 
102 	/*
103 	 * File system must be write locked to prevent updates while
104 	 * the snapshot is being established.
105 	 */
106 	if ((error = ufs_fiolfss(vp, &lf)) != 0) {
107 		release_backing_vnodes(&bfvpp, fiosnapp->backfilecount);
108 		return (error);
109 	}
110 
111 	if (!LOCKFS_IS_ULOCK(&lf)) {
112 		release_backing_vnodes(&bfvpp, fiosnapp->backfilecount);
113 		fiosnapp->error = FIOCOW_EULOCK;
114 		return (EINVAL);
115 	}
116 
117 	lf.lf_lock = LOCKFS_WLOCK;
118 	lf.lf_flags = 0;
119 	lf.lf_comment = NULL;
120 	if ((error = ufs_fiolfs(vp, &lf, 1)) != 0) {
121 		release_backing_vnodes(&bfvpp, fiosnapp->backfilecount);
122 		fiosnapp->error = FIOCOW_EWLOCK;
123 		return (EINVAL);
124 	}
125 
126 	/*
127 	 * File system must be fairly consistent to enable snapshots
128 	 */
129 	if (fs->fs_clean != FSACTIVE &&
130 	    fs->fs_clean != FSSTABLE &&
131 	    fs->fs_clean != FSCLEAN &&
132 	    fs->fs_clean != FSLOG) {
133 		fiosnapp->error = FIOCOW_ECLEAN;
134 		error = EINVAL;
135 		goto unlockout;
136 	}
137 
138 	/*
139 	 * Only one snapshot is allowed per file system, so error if
140 	 * a snapshot is already enabled.
141 	 */
142 	if (ufsvfsp->vfs_snapshot) {
143 		fiosnapp->error = FIOCOW_EBUSY;
144 		error = EBUSY;
145 		goto unlockout;
146 	}
147 
148 	/* Tell bio.c how to call our strategy routine.  XXX ugly hack */
149 	if (bio_snapshot_strategy == NULL)
150 		bio_snapshot_strategy =
151 		    (void (*) (void *, buf_t *))fssnap_strategy;
152 
153 	/*
154 	 * use chunk size that is passed in, or the file system
155 	 * block size if it is zero.  For most cases, the file system
156 	 * block size will be reasonably efficient.  A larger
157 	 * chunksize uses less memory but may potentially induce more
158 	 * I/O copying the larger chunks aside.
159 	 */
160 	if (fiosnapp->chunksize != 0)
161 		chunksize = fiosnapp->chunksize;
162 	else
163 		chunksize = fs->fs_bsize * 4;
164 
165 
166 	/*
167 	 * compute the number of chunks in this whole file system.  Since
168 	 * the UFS allocation bitmaps are in units of fragments, we first
169 	 * compute the number of fragments per chunk.  Things work out
170 	 * nicer if the chunk size is a power-of-two multiple of the
171 	 * fragment size.
172 	 */
173 	if ((chunksize < fs->fs_fsize) || (chunksize % fs->fs_fsize != 0)) {
174 		fiosnapp->error = FIOCOW_ECHUNKSZ;
175 		error = EINVAL;
176 		goto unlockout;
177 	}
178 	fragsperchunk = chunksize >> fs->fs_fshift;
179 	nchunks = (fs->fs_size + fragsperchunk) / fragsperchunk;
180 
181 	/*
182 	 * Create and initialize snapshot state and allocate/initialize
183 	 * translation table.  This does the real work of taking the snapshot.
184 	 */
185 	snapid = fssnap_create(nchunks, chunksize, fiosnapp->maxsize, vp,
186 	    fiosnapp->backfilecount, bfvpp, fiosnapp->backfilename,
187 	    fiosnapp->backfilesize);
188 	if (snapid == NULL) {
189 		fiosnapp->error = FIOCOW_ECREATE;
190 		error = EINVAL;
191 		goto unlockout;
192 	}
193 
194 	error = ufs_snap_find_candidates(snapid, ufsvfsp, chunksize);
195 	fiosnapp->snapshotnumber = fssnap_create_done(snapid);
196 
197 	if (error) {
198 		cmn_err(CE_WARN, "ufs_snap_create: failed scanning bitmaps, "
199 		    "error = %d.", error);
200 		fiosnapp->error = FIOCOW_EBITMAP;
201 		goto unlockout;
202 	}
203 
204 	ufsvfsp->vfs_snapshot = snapid;
205 
206 unlockout:
207 	/*
208 	 * Unlock the file system
209 	 */
210 	lf.lf_lock = LOCKFS_ULOCK;
211 	lf.lf_flags = 0;
212 	if ((ufs_fiolfs(vp, &lf, 1) != 0) && !error) {
213 		fiosnapp->error = FIOCOW_ENOULOCK;
214 		error = EINVAL;
215 	} else {
216 		fiosnapp->error = 0;
217 	}
218 
219 	/* clean up the snapshot if an error occurred. */
220 	if (error && snapid != NULL)
221 		(void) fssnap_delete(&snapid);
222 	else if (error && bfvpp != NULL)
223 		release_backing_vnodes(&bfvpp, fiosnapp->backfilecount);
224 
225 	return (error);
226 }
227 
228 static int
229 ufs_snap_init_backfile(int *filedesc, int count, vnode_t ***vppp,
230     struct ufsvfs *ufsvfsp)
231 {
232 	file_t *fp;
233 	vnode_t **vpp;
234 	int i;
235 
236 	vpp = (vnode_t **)kmem_zalloc((count  + 1) * sizeof (vnode_t *),
237 	    KM_SLEEP);
238 	*vppp = vpp;
239 	for (i = 0; i < count; i++) {
240 		if ((fp = getf(*filedesc)) == NULL) {
241 			release_backing_vnodes(vppp, count);
242 			*vppp = NULL;
243 			return (EBADF);
244 		}
245 
246 		ASSERT(fp->f_vnode != NULL);
247 		VN_HOLD(fp->f_vnode);
248 
249 		*vpp = fp->f_vnode;
250 		releasef(*filedesc);
251 		filedesc++;
252 
253 		/* make sure the backing file is on a different file system */
254 		if ((*vpp)->v_vfsp == ufsvfsp->vfs_vfs) {
255 			release_backing_vnodes(vppp, count);
256 			*vppp = NULL;
257 			return (EINVAL);
258 		}
259 		vpp++;
260 	}
261 	return (0);
262 }
263 
264 static void
265 release_backing_vnodes(vnode_t ***bvppp, int count)
266 {
267 	vnode_t **vpp;
268 
269 	vpp = *bvppp;
270 	while (*vpp) {
271 		VN_RELE(*vpp);
272 		*vpp++ = NULL;
273 	}
274 	kmem_free(*bvppp, (count + 1) * sizeof (vnode_t *));
275 	*bvppp = NULL;
276 }
277 
278 static int
279 ufs_snap_find_candidates(void *snapid, struct ufsvfs *ufsvfsp, int chunksize)
280 {
281 	struct fs	*fs = ufsvfsp->vfs_fs;
282 	struct buf	*cgbp;	/* cylinder group buffer */
283 	struct cg	*cgp;	/* cylinder group data */
284 	ulong_t		cg;
285 	ulong_t		cgbase;
286 	ulong_t		chunk;
287 	uchar_t		*blksfree;
288 
289 	ulong_t		curfrag;
290 	int		error = 0;
291 
292 	/*
293 	 * read through each ufs cylinder group and fetch the fragment
294 	 * allocation bitmap.  UFS indicates a fragment is allocated by
295 	 * a zero bit (not a one bit) in the fragment offset.
296 	 */
297 	cgbase = 0LL;
298 	for (cg = 0; cg < fs->fs_ncg; cg++) {
299 		/* read the cylinder group in */
300 		cgbp = BREAD(ufsvfsp->vfs_dev,
301 		    (daddr_t)fsbtodb(fs, cgtod(fs, cg)), (int)fs->fs_cgsize);
302 		if ((error = geterror(cgbp)) != 0) {
303 			brelse(cgbp);
304 			goto errout;
305 		}
306 		cgp = cgbp->b_un.b_cg;
307 
308 		/* check the magic number */
309 		if (cgp->cg_magic != CG_MAGIC) {
310 			cmn_err(CE_WARN, "ufs_snap_find_candidates: cg %lu "
311 			    "magic number (0x%x) does not match expected "
312 			    "magic number (0x%x)", cg, cgp->cg_magic, CG_MAGIC);
313 			error = EIO;
314 			goto errout;
315 		}
316 
317 		blksfree = cg_blksfree(cgp);
318 
319 		/*
320 		 * go through the allocation bitmap and set the
321 		 * corresponding bit in the candidate map.
322 		 */
323 		for (curfrag = 0; curfrag < cgp->cg_ndblk; curfrag++) {
324 			if (isclr(blksfree, curfrag)) {
325 				/*
326 				 * this assumes chunksize is a multiple of
327 				 * the fragment size
328 				 */
329 				chunk = (ulong_t)((cgbase + curfrag) /
330 				    (chunksize >> fs->fs_fshift));
331 
332 				fssnap_set_candidate(snapid, chunk);
333 				/*
334 				 * no need to scan the rest of this chunk since
335 				 * it is already marked, so skip to the next
336 				 */
337 				curfrag += ((chunksize >> fs->fs_fshift) -
338 				    ((cgbase + curfrag) %
339 				    (chunksize >> fs->fs_fshift))) - 1;
340 			}
341 		}
342 
343 		cgbase += cgp->cg_ndblk;
344 		ASSERT(cgbase <= fs->fs_size);
345 		brelse(cgbp);
346 	} /* cylinder group loop */
347 
348 	ASSERT(cgbase == fs->fs_size);
349 
350 errout:
351 	return (error);
352 }
353 
354 
355 int
356 ufs_snap_delete(struct vnode *vp, struct fiosnapdelete *fiosnapp, cred_t *cr)
357 {
358 	struct ufsvfs	*ufsvfsp = VTOI(vp)->i_ufsvfs;
359 	struct fs	*fs = ufsvfsp->vfs_fs;
360 
361 	/*
362 	 * Initialize fields in the user's buffer
363 	 */
364 	fiosnapp->error = 0;
365 
366 	/*
367 	 * No snapshot exists, we're done.
368 	 */
369 	if (ufsvfsp->vfs_snapshot == NULL)
370 		return (ENOENT);
371 
372 	/*
373 	 * must have sufficient privileges.
374 	 */
375 	if (secpolicy_fs_config(cr, ufsvfsp->vfs_vfs) != 0)
376 		return (EPERM);
377 
378 	/*
379 	 * Readonly file system
380 	 */
381 	if (fs->fs_ronly) {
382 		fiosnapp->error = FIOCOW_EREADONLY;
383 		return (EROFS);
384 	}
385 
386 	/* free the data structures and clear the vfs_snapshot field. */
387 	fiosnapp->snapshotnumber = fssnap_delete(&ufsvfsp->vfs_snapshot);
388 
389 	if (fiosnapp->snapshotnumber == -1)
390 		return (EINVAL);
391 
392 	return (0);
393 }
394