35 */
36
37#include "opt_quota.h"
38
39#include <sys/param.h>
40#include <sys/systm.h>
41#include <sys/buf.h>
42#include <sys/proc.h>
43#include <sys/vnode.h>
44#include <sys/mount.h>
45#include <sys/kernel.h>
46#include <sys/sysctl.h>
47#include <sys/syslog.h>
48
49#include <ufs/ufs/quota.h>
50#include <ufs/ufs/inode.h>
51#include <ufs/ufs/ufsmount.h>
52
53#include <ufs/ffs/fs.h>
54#include <ufs/ffs/ffs_extern.h>
55
56typedef ufs_daddr_t allocfcn_t __P((struct inode *ip, int cg, ufs_daddr_t bpref,
57 int size));
58
59static ufs_daddr_t ffs_alloccg __P((struct inode *, int, ufs_daddr_t, int));
60static ufs_daddr_t ffs_alloccgblk __P((struct fs *, struct cg *, ufs_daddr_t));
61static void ffs_clusteracct __P((struct fs *, struct cg *, ufs_daddr_t,
62 int));
63static ufs_daddr_t ffs_clusteralloc __P((struct inode *, int, ufs_daddr_t,
64 int));
65static ino_t ffs_dirpref __P((struct fs *));
66static ufs_daddr_t ffs_fragextend __P((struct inode *, int, long, int, int));
67static void ffs_fserr __P((struct fs *, u_int, char *));
68static u_long ffs_hashalloc
69 __P((struct inode *, int, long, int, allocfcn_t *));
70static ino_t ffs_nodealloccg __P((struct inode *, int, ufs_daddr_t, int));
71static ufs_daddr_t ffs_mapsearch __P((struct fs *, struct cg *, ufs_daddr_t,
72 int));
73
74/*
75 * Allocate a block in the file system.
76 *
77 * The size of the requested block is given, which must be some
78 * multiple of fs_fsize and <= fs_bsize.
79 * A preference may be optionally specified. If a preference is given
80 * the following hierarchy is used to allocate a block:
81 * 1) allocate the requested block.
82 * 2) allocate a rotationally optimal block in the same cylinder.
83 * 3) allocate a block in the same cylinder group.
84 * 4) quadradically rehash into other cylinder groups, until an
85 * available block is located.
86 * If no block preference is given the following heirarchy is used
87 * to allocate a block:
88 * 1) allocate a block in the cylinder group that contains the
89 * inode for the file.
90 * 2) quadradically rehash into other cylinder groups, until an
91 * available block is located.
92 */
93int
94ffs_alloc(ip, lbn, bpref, size, cred, bnp)
95 register struct inode *ip;
96 ufs_daddr_t lbn, bpref;
97 int size;
98 struct ucred *cred;
99 ufs_daddr_t *bnp;
100{
101 register struct fs *fs;
102 ufs_daddr_t bno;
103 int cg;
104#ifdef QUOTA
105 int error;
106#endif
107
108 *bnp = 0;
109 fs = ip->i_fs;
110#ifdef DIAGNOSTIC
111 if ((u_int)size > fs->fs_bsize || fragoff(fs, size) != 0) {
112 printf("dev = 0x%lx, bsize = %ld, size = %d, fs = %s\n",
113 (u_long)ip->i_dev, fs->fs_bsize, size, fs->fs_fsmnt);
114 panic("ffs_alloc: bad size");
115 }
116 if (cred == NOCRED)
117 panic("ffs_alloc: missing credential");
118#endif /* DIAGNOSTIC */
119 if (size == fs->fs_bsize && fs->fs_cstotal.cs_nbfree == 0)
120 goto nospace;
121 if (cred->cr_uid != 0 &&
122 freespace(fs, fs->fs_minfree) - numfrags(fs, size) < 0)
123 goto nospace;
124#ifdef QUOTA
125 error = chkdq(ip, (long)btodb(size), cred, 0);
126 if (error)
127 return (error);
128#endif
129 if (bpref >= fs->fs_size)
130 bpref = 0;
131 if (bpref == 0)
132 cg = ino_to_cg(fs, ip->i_number);
133 else
134 cg = dtog(fs, bpref);
135 bno = (ufs_daddr_t)ffs_hashalloc(ip, cg, (long)bpref, size,
136 ffs_alloccg);
137 if (bno > 0) {
138 ip->i_blocks += btodb(size);
139 ip->i_flag |= IN_CHANGE | IN_UPDATE;
140 *bnp = bno;
141 return (0);
142 }
143#ifdef QUOTA
144 /*
145 * Restore user's disk quota because allocation failed.
146 */
147 (void) chkdq(ip, (long)-btodb(size), cred, FORCE);
148#endif
149nospace:
150 ffs_fserr(fs, cred->cr_uid, "file system full");
151 uprintf("\n%s: write failed, file system is full\n", fs->fs_fsmnt);
152 return (ENOSPC);
153}
154
155/*
156 * Reallocate a fragment to a bigger size
157 *
158 * The number and size of the old block is given, and a preference
159 * and new size is also specified. The allocator attempts to extend
160 * the original block. Failing that, the regular block allocator is
161 * invoked to get an appropriate block.
162 */
163int
164ffs_realloccg(ip, lbprev, bpref, osize, nsize, cred, bpp)
165 register struct inode *ip;
166 ufs_daddr_t lbprev;
167 ufs_daddr_t bpref;
168 int osize, nsize;
169 struct ucred *cred;
170 struct buf **bpp;
171{
172 register struct fs *fs;
173 struct buf *bp;
174 int cg, request, error;
175 ufs_daddr_t bprev, bno;
176
177 *bpp = 0;
178 fs = ip->i_fs;
179#ifdef DIAGNOSTIC
180 if ((u_int)osize > fs->fs_bsize || fragoff(fs, osize) != 0 ||
181 (u_int)nsize > fs->fs_bsize || fragoff(fs, nsize) != 0) {
182 printf(
183 "dev = 0x%lx, bsize = %ld, osize = %d, "
184 "nsize = %d, fs = %s\n",
185 (u_long)ip->i_dev, fs->fs_bsize, osize,
186 nsize, fs->fs_fsmnt);
187 panic("ffs_realloccg: bad size");
188 }
189 if (cred == NOCRED)
190 panic("ffs_realloccg: missing credential");
191#endif /* DIAGNOSTIC */
192 if (cred->cr_uid != 0 &&
193 freespace(fs, fs->fs_minfree) - numfrags(fs, nsize - osize) < 0)
194 goto nospace;
195 if ((bprev = ip->i_db[lbprev]) == 0) {
196 printf("dev = 0x%lx, bsize = %ld, bprev = %ld, fs = %s\n",
197 (u_long) ip->i_dev, fs->fs_bsize, bprev, fs->fs_fsmnt);
198 panic("ffs_realloccg: bad bprev");
199 }
200 /*
201 * Allocate the extra space in the buffer.
202 */
203 error = bread(ITOV(ip), lbprev, osize, NOCRED, &bp);
204 if (error) {
205 brelse(bp);
206 return (error);
207 }
208
209 if( bp->b_blkno == bp->b_lblkno) {
210 if( lbprev >= NDADDR)
211 panic("ffs_realloccg: lbprev out of range");
212 bp->b_blkno = fsbtodb(fs, bprev);
213 }
214
215#ifdef QUOTA
216 error = chkdq(ip, (long)btodb(nsize - osize), cred, 0);
217 if (error) {
218 brelse(bp);
219 return (error);
220 }
221#endif
222 /*
223 * Check for extension in the existing location.
224 */
225 cg = dtog(fs, bprev);
226 bno = ffs_fragextend(ip, cg, (long)bprev, osize, nsize);
227 if (bno) {
228 if (bp->b_blkno != fsbtodb(fs, bno))
229 panic("ffs_realloccg: bad blockno");
230 ip->i_blocks += btodb(nsize - osize);
231 ip->i_flag |= IN_CHANGE | IN_UPDATE;
232 allocbuf(bp, nsize);
233 bp->b_flags |= B_DONE;
234 bzero((char *)bp->b_data + osize, (u_int)nsize - osize);
235 *bpp = bp;
236 return (0);
237 }
238 /*
239 * Allocate a new disk location.
240 */
241 if (bpref >= fs->fs_size)
242 bpref = 0;
243 switch ((int)fs->fs_optim) {
244 case FS_OPTSPACE:
245 /*
246 * Allocate an exact sized fragment. Although this makes
247 * best use of space, we will waste time relocating it if
248 * the file continues to grow. If the fragmentation is
249 * less than half of the minimum free reserve, we choose
250 * to begin optimizing for time.
251 */
252 request = nsize;
253 if (fs->fs_minfree <= 5 ||
254 fs->fs_cstotal.cs_nffree >
255 fs->fs_dsize * fs->fs_minfree / (2 * 100))
256 break;
257 log(LOG_NOTICE, "%s: optimization changed from SPACE to TIME\n",
258 fs->fs_fsmnt);
259 fs->fs_optim = FS_OPTTIME;
260 break;
261 case FS_OPTTIME:
262 /*
263 * At this point we have discovered a file that is trying to
264 * grow a small fragment to a larger fragment. To save time,
265 * we allocate a full sized block, then free the unused portion.
266 * If the file continues to grow, the `ffs_fragextend' call
267 * above will be able to grow it in place without further
268 * copying. If aberrant programs cause disk fragmentation to
269 * grow within 2% of the free reserve, we choose to begin
270 * optimizing for space.
271 */
272 request = fs->fs_bsize;
273 if (fs->fs_cstotal.cs_nffree <
274 fs->fs_dsize * (fs->fs_minfree - 2) / 100)
275 break;
276 log(LOG_NOTICE, "%s: optimization changed from TIME to SPACE\n",
277 fs->fs_fsmnt);
278 fs->fs_optim = FS_OPTSPACE;
279 break;
280 default:
281 printf("dev = 0x%lx, optim = %ld, fs = %s\n",
282 (u_long)ip->i_dev, fs->fs_optim, fs->fs_fsmnt);
283 panic("ffs_realloccg: bad optim");
284 /* NOTREACHED */
285 }
286 bno = (ufs_daddr_t)ffs_hashalloc(ip, cg, (long)bpref, request,
287 ffs_alloccg);
288 if (bno > 0) {
289 bp->b_blkno = fsbtodb(fs, bno);
290 ffs_blkfree(ip, bprev, (long)osize);
291 if (nsize < request)
292 ffs_blkfree(ip, bno + numfrags(fs, nsize),
293 (long)(request - nsize));
294 ip->i_blocks += btodb(nsize - osize);
295 ip->i_flag |= IN_CHANGE | IN_UPDATE;
296 allocbuf(bp, nsize);
297 bp->b_flags |= B_DONE;
298 bzero((char *)bp->b_data + osize, (u_int)nsize - osize);
299 *bpp = bp;
300 return (0);
301 }
302#ifdef QUOTA
303 /*
304 * Restore user's disk quota because allocation failed.
305 */
306 (void) chkdq(ip, (long)-btodb(nsize - osize), cred, FORCE);
307#endif
308 brelse(bp);
309nospace:
310 /*
311 * no space available
312 */
313 ffs_fserr(fs, cred->cr_uid, "file system full");
314 uprintf("\n%s: write failed, file system is full\n", fs->fs_fsmnt);
315 return (ENOSPC);
316}
317
318/*
319 * Reallocate a sequence of blocks into a contiguous sequence of blocks.
320 *
321 * The vnode and an array of buffer pointers for a range of sequential
322 * logical blocks to be made contiguous is given. The allocator attempts
323 * to find a range of sequential blocks starting as close as possible to
324 * an fs_rotdelay offset from the end of the allocation for the logical
325 * block immediately preceeding the current range. If successful, the
326 * physical block numbers in the buffer pointers and in the inode are
327 * changed to reflect the new allocation. If unsuccessful, the allocation
328 * is left unchanged. The success in doing the reallocation is returned.
329 * Note that the error return is not reflected back to the user. Rather
330 * the previous block allocation will be used.
331 */
332static int doasyncfree = 1;
333SYSCTL_INT(_vfs_ffs, FFS_ASYNCFREE, doasyncfree, CTLFLAG_RW, &doasyncfree, 0, "");
334
335int doreallocblks = 1;
336SYSCTL_INT(_vfs_ffs, FFS_REALLOCBLKS, doreallocblks, CTLFLAG_RW, &doreallocblks, 0, "");
337
338static int prtrealloc = 0;
339
340int
341ffs_reallocblks(ap)
342 struct vop_reallocblks_args /* {
343 struct vnode *a_vp;
344 struct cluster_save *a_buflist;
345 } */ *ap;
346{
347#if !defined (not_yes)
348 return (ENOSPC);
349#else
350 struct fs *fs;
351 struct inode *ip;
352 struct vnode *vp;
353 struct buf *sbp, *ebp;
354 ufs_daddr_t *bap, *sbap, *ebap = 0;
355 struct cluster_save *buflist;
356 ufs_daddr_t start_lbn, end_lbn, soff, newblk, blkno;
357 struct indir start_ap[NIADDR + 1], end_ap[NIADDR + 1], *idp;
358 int i, len, start_lvl, end_lvl, pref, ssize;
359 struct timeval tv;
360
361 if (doreallocblks == 0)
362 return (ENOSPC);
363 vp = ap->a_vp;
364 ip = VTOI(vp);
365 fs = ip->i_fs;
366 if (fs->fs_contigsumsize <= 0)
367 return (ENOSPC);
368 buflist = ap->a_buflist;
369 len = buflist->bs_nchildren;
370 start_lbn = buflist->bs_children[0]->b_lblkno;
371 end_lbn = start_lbn + len - 1;
372#ifdef DIAGNOSTIC
373 for (i = 0; i < len; i++)
374 if (!ffs_checkblk(ip,
375 dbtofsb(fs, buflist->bs_children[i]->b_blkno), fs->fs_bsize))
376 panic("ffs_reallocblks: unallocated block 1");
377 for (i = 1; i < len; i++)
378 if (buflist->bs_children[i]->b_lblkno != start_lbn + i)
379 panic("ffs_reallocblks: non-logical cluster");
380 blkno = buflist->bs_children[0]->b_blkno;
381 ssize = fsbtodb(fs, fs->fs_frag);
382 for (i = 1; i < len - 1; i++)
383 if (buflist->bs_children[i]->b_blkno != blkno + (i * ssize))
384 panic("ffs_reallocblks: non-physical cluster %d", i);
385#endif
386 /*
387 * If the latest allocation is in a new cylinder group, assume that
388 * the filesystem has decided to move and do not force it back to
389 * the previous cylinder group.
390 */
391 if (dtog(fs, dbtofsb(fs, buflist->bs_children[0]->b_blkno)) !=
392 dtog(fs, dbtofsb(fs, buflist->bs_children[len - 1]->b_blkno)))
393 return (ENOSPC);
394 if (ufs_getlbns(vp, start_lbn, start_ap, &start_lvl) ||
395 ufs_getlbns(vp, end_lbn, end_ap, &end_lvl))
396 return (ENOSPC);
397 /*
398 * Get the starting offset and block map for the first block.
399 */
400 if (start_lvl == 0) {
401 sbap = &ip->i_db[0];
402 soff = start_lbn;
403 } else {
404 idp = &start_ap[start_lvl - 1];
405 if (bread(vp, idp->in_lbn, (int)fs->fs_bsize, NOCRED, &sbp)) {
406 brelse(sbp);
407 return (ENOSPC);
408 }
409 sbap = (ufs_daddr_t *)sbp->b_data;
410 soff = idp->in_off;
411 }
412 /*
413 * Find the preferred location for the cluster.
414 */
415 pref = ffs_blkpref(ip, start_lbn, soff, sbap);
416 /*
417 * If the block range spans two block maps, get the second map.
418 */
419 if (end_lvl == 0 || (idp = &end_ap[end_lvl - 1])->in_off + 1 >= len) {
420 ssize = len;
421 } else {
422#ifdef DIAGNOSTIC
423 if (start_ap[start_lvl-1].in_lbn == idp->in_lbn)
424 panic("ffs_reallocblk: start == end");
425#endif
426 ssize = len - (idp->in_off + 1);
427 if (bread(vp, idp->in_lbn, (int)fs->fs_bsize, NOCRED, &ebp))
428 goto fail;
429 ebap = (ufs_daddr_t *)ebp->b_data;
430 }
431 /*
432 * Search the block map looking for an allocation of the desired size.
433 */
434 if ((newblk = (ufs_daddr_t)ffs_hashalloc(ip, dtog(fs, pref), (long)pref,
435 len, ffs_clusteralloc)) == 0)
436 goto fail;
437 /*
438 * We have found a new contiguous block.
439 *
440 * First we have to replace the old block pointers with the new
441 * block pointers in the inode and indirect blocks associated
442 * with the file.
443 */
444#ifdef DEBUG
445 if (prtrealloc)
446 printf("realloc: ino %d, lbns %d-%d\n\told:", ip->i_number,
447 start_lbn, end_lbn);
448#endif
449 blkno = newblk;
450 for (bap = &sbap[soff], i = 0; i < len; i++, blkno += fs->fs_frag) {
451 if (i == ssize)
452 bap = ebap;
453#ifdef DIAGNOSTIC
454 if (!ffs_checkblk(ip,
455 dbtofsb(fs, buflist->bs_children[i]->b_blkno), fs->fs_bsize))
456 panic("ffs_reallocblks: unallocated block 2");
457 if (dbtofsb(fs, buflist->bs_children[i]->b_blkno) != *bap)
458 panic("ffs_reallocblks: alloc mismatch");
459#endif
460#ifdef DEBUG
461 if (prtrealloc)
462 printf(" %d,", *bap);
463#endif
464 *bap++ = blkno;
465 }
466 /*
467 * Next we must write out the modified inode and indirect blocks.
468 * For strict correctness, the writes should be synchronous since
469 * the old block values may have been written to disk. In practise
470 * they are almost never written, but if we are concerned about
471 * strict correctness, the `doasyncfree' flag should be set to zero.
472 *
473 * The test on `doasyncfree' should be changed to test a flag
474 * that shows whether the associated buffers and inodes have
475 * been written. The flag should be set when the cluster is
476 * started and cleared whenever the buffer or inode is flushed.
477 * We can then check below to see if it is set, and do the
478 * synchronous write only when it has been cleared.
479 */
480 if (sbap != &ip->i_db[0]) {
481 if (doasyncfree)
482 bdwrite(sbp);
483 else
484 bwrite(sbp);
485 } else {
486 ip->i_flag |= IN_CHANGE | IN_UPDATE;
487 if (!doasyncfree) {
488 gettime(&tv);
| 35 */
36
37#include "opt_quota.h"
38
39#include <sys/param.h>
40#include <sys/systm.h>
41#include <sys/buf.h>
42#include <sys/proc.h>
43#include <sys/vnode.h>
44#include <sys/mount.h>
45#include <sys/kernel.h>
46#include <sys/sysctl.h>
47#include <sys/syslog.h>
48
49#include <ufs/ufs/quota.h>
50#include <ufs/ufs/inode.h>
51#include <ufs/ufs/ufsmount.h>
52
53#include <ufs/ffs/fs.h>
54#include <ufs/ffs/ffs_extern.h>
55
56typedef ufs_daddr_t allocfcn_t __P((struct inode *ip, int cg, ufs_daddr_t bpref,
57 int size));
58
59static ufs_daddr_t ffs_alloccg __P((struct inode *, int, ufs_daddr_t, int));
60static ufs_daddr_t ffs_alloccgblk __P((struct fs *, struct cg *, ufs_daddr_t));
61static void ffs_clusteracct __P((struct fs *, struct cg *, ufs_daddr_t,
62 int));
63static ufs_daddr_t ffs_clusteralloc __P((struct inode *, int, ufs_daddr_t,
64 int));
65static ino_t ffs_dirpref __P((struct fs *));
66static ufs_daddr_t ffs_fragextend __P((struct inode *, int, long, int, int));
67static void ffs_fserr __P((struct fs *, u_int, char *));
68static u_long ffs_hashalloc
69 __P((struct inode *, int, long, int, allocfcn_t *));
70static ino_t ffs_nodealloccg __P((struct inode *, int, ufs_daddr_t, int));
71static ufs_daddr_t ffs_mapsearch __P((struct fs *, struct cg *, ufs_daddr_t,
72 int));
73
74/*
75 * Allocate a block in the file system.
76 *
77 * The size of the requested block is given, which must be some
78 * multiple of fs_fsize and <= fs_bsize.
79 * A preference may be optionally specified. If a preference is given
80 * the following hierarchy is used to allocate a block:
81 * 1) allocate the requested block.
82 * 2) allocate a rotationally optimal block in the same cylinder.
83 * 3) allocate a block in the same cylinder group.
84 * 4) quadradically rehash into other cylinder groups, until an
85 * available block is located.
86 * If no block preference is given the following heirarchy is used
87 * to allocate a block:
88 * 1) allocate a block in the cylinder group that contains the
89 * inode for the file.
90 * 2) quadradically rehash into other cylinder groups, until an
91 * available block is located.
92 */
93int
94ffs_alloc(ip, lbn, bpref, size, cred, bnp)
95 register struct inode *ip;
96 ufs_daddr_t lbn, bpref;
97 int size;
98 struct ucred *cred;
99 ufs_daddr_t *bnp;
100{
101 register struct fs *fs;
102 ufs_daddr_t bno;
103 int cg;
104#ifdef QUOTA
105 int error;
106#endif
107
108 *bnp = 0;
109 fs = ip->i_fs;
110#ifdef DIAGNOSTIC
111 if ((u_int)size > fs->fs_bsize || fragoff(fs, size) != 0) {
112 printf("dev = 0x%lx, bsize = %ld, size = %d, fs = %s\n",
113 (u_long)ip->i_dev, fs->fs_bsize, size, fs->fs_fsmnt);
114 panic("ffs_alloc: bad size");
115 }
116 if (cred == NOCRED)
117 panic("ffs_alloc: missing credential");
118#endif /* DIAGNOSTIC */
119 if (size == fs->fs_bsize && fs->fs_cstotal.cs_nbfree == 0)
120 goto nospace;
121 if (cred->cr_uid != 0 &&
122 freespace(fs, fs->fs_minfree) - numfrags(fs, size) < 0)
123 goto nospace;
124#ifdef QUOTA
125 error = chkdq(ip, (long)btodb(size), cred, 0);
126 if (error)
127 return (error);
128#endif
129 if (bpref >= fs->fs_size)
130 bpref = 0;
131 if (bpref == 0)
132 cg = ino_to_cg(fs, ip->i_number);
133 else
134 cg = dtog(fs, bpref);
135 bno = (ufs_daddr_t)ffs_hashalloc(ip, cg, (long)bpref, size,
136 ffs_alloccg);
137 if (bno > 0) {
138 ip->i_blocks += btodb(size);
139 ip->i_flag |= IN_CHANGE | IN_UPDATE;
140 *bnp = bno;
141 return (0);
142 }
143#ifdef QUOTA
144 /*
145 * Restore user's disk quota because allocation failed.
146 */
147 (void) chkdq(ip, (long)-btodb(size), cred, FORCE);
148#endif
149nospace:
150 ffs_fserr(fs, cred->cr_uid, "file system full");
151 uprintf("\n%s: write failed, file system is full\n", fs->fs_fsmnt);
152 return (ENOSPC);
153}
154
155/*
156 * Reallocate a fragment to a bigger size
157 *
158 * The number and size of the old block is given, and a preference
159 * and new size is also specified. The allocator attempts to extend
160 * the original block. Failing that, the regular block allocator is
161 * invoked to get an appropriate block.
162 */
163int
164ffs_realloccg(ip, lbprev, bpref, osize, nsize, cred, bpp)
165 register struct inode *ip;
166 ufs_daddr_t lbprev;
167 ufs_daddr_t bpref;
168 int osize, nsize;
169 struct ucred *cred;
170 struct buf **bpp;
171{
172 register struct fs *fs;
173 struct buf *bp;
174 int cg, request, error;
175 ufs_daddr_t bprev, bno;
176
177 *bpp = 0;
178 fs = ip->i_fs;
179#ifdef DIAGNOSTIC
180 if ((u_int)osize > fs->fs_bsize || fragoff(fs, osize) != 0 ||
181 (u_int)nsize > fs->fs_bsize || fragoff(fs, nsize) != 0) {
182 printf(
183 "dev = 0x%lx, bsize = %ld, osize = %d, "
184 "nsize = %d, fs = %s\n",
185 (u_long)ip->i_dev, fs->fs_bsize, osize,
186 nsize, fs->fs_fsmnt);
187 panic("ffs_realloccg: bad size");
188 }
189 if (cred == NOCRED)
190 panic("ffs_realloccg: missing credential");
191#endif /* DIAGNOSTIC */
192 if (cred->cr_uid != 0 &&
193 freespace(fs, fs->fs_minfree) - numfrags(fs, nsize - osize) < 0)
194 goto nospace;
195 if ((bprev = ip->i_db[lbprev]) == 0) {
196 printf("dev = 0x%lx, bsize = %ld, bprev = %ld, fs = %s\n",
197 (u_long) ip->i_dev, fs->fs_bsize, bprev, fs->fs_fsmnt);
198 panic("ffs_realloccg: bad bprev");
199 }
200 /*
201 * Allocate the extra space in the buffer.
202 */
203 error = bread(ITOV(ip), lbprev, osize, NOCRED, &bp);
204 if (error) {
205 brelse(bp);
206 return (error);
207 }
208
209 if( bp->b_blkno == bp->b_lblkno) {
210 if( lbprev >= NDADDR)
211 panic("ffs_realloccg: lbprev out of range");
212 bp->b_blkno = fsbtodb(fs, bprev);
213 }
214
215#ifdef QUOTA
216 error = chkdq(ip, (long)btodb(nsize - osize), cred, 0);
217 if (error) {
218 brelse(bp);
219 return (error);
220 }
221#endif
222 /*
223 * Check for extension in the existing location.
224 */
225 cg = dtog(fs, bprev);
226 bno = ffs_fragextend(ip, cg, (long)bprev, osize, nsize);
227 if (bno) {
228 if (bp->b_blkno != fsbtodb(fs, bno))
229 panic("ffs_realloccg: bad blockno");
230 ip->i_blocks += btodb(nsize - osize);
231 ip->i_flag |= IN_CHANGE | IN_UPDATE;
232 allocbuf(bp, nsize);
233 bp->b_flags |= B_DONE;
234 bzero((char *)bp->b_data + osize, (u_int)nsize - osize);
235 *bpp = bp;
236 return (0);
237 }
238 /*
239 * Allocate a new disk location.
240 */
241 if (bpref >= fs->fs_size)
242 bpref = 0;
243 switch ((int)fs->fs_optim) {
244 case FS_OPTSPACE:
245 /*
246 * Allocate an exact sized fragment. Although this makes
247 * best use of space, we will waste time relocating it if
248 * the file continues to grow. If the fragmentation is
249 * less than half of the minimum free reserve, we choose
250 * to begin optimizing for time.
251 */
252 request = nsize;
253 if (fs->fs_minfree <= 5 ||
254 fs->fs_cstotal.cs_nffree >
255 fs->fs_dsize * fs->fs_minfree / (2 * 100))
256 break;
257 log(LOG_NOTICE, "%s: optimization changed from SPACE to TIME\n",
258 fs->fs_fsmnt);
259 fs->fs_optim = FS_OPTTIME;
260 break;
261 case FS_OPTTIME:
262 /*
263 * At this point we have discovered a file that is trying to
264 * grow a small fragment to a larger fragment. To save time,
265 * we allocate a full sized block, then free the unused portion.
266 * If the file continues to grow, the `ffs_fragextend' call
267 * above will be able to grow it in place without further
268 * copying. If aberrant programs cause disk fragmentation to
269 * grow within 2% of the free reserve, we choose to begin
270 * optimizing for space.
271 */
272 request = fs->fs_bsize;
273 if (fs->fs_cstotal.cs_nffree <
274 fs->fs_dsize * (fs->fs_minfree - 2) / 100)
275 break;
276 log(LOG_NOTICE, "%s: optimization changed from TIME to SPACE\n",
277 fs->fs_fsmnt);
278 fs->fs_optim = FS_OPTSPACE;
279 break;
280 default:
281 printf("dev = 0x%lx, optim = %ld, fs = %s\n",
282 (u_long)ip->i_dev, fs->fs_optim, fs->fs_fsmnt);
283 panic("ffs_realloccg: bad optim");
284 /* NOTREACHED */
285 }
286 bno = (ufs_daddr_t)ffs_hashalloc(ip, cg, (long)bpref, request,
287 ffs_alloccg);
288 if (bno > 0) {
289 bp->b_blkno = fsbtodb(fs, bno);
290 ffs_blkfree(ip, bprev, (long)osize);
291 if (nsize < request)
292 ffs_blkfree(ip, bno + numfrags(fs, nsize),
293 (long)(request - nsize));
294 ip->i_blocks += btodb(nsize - osize);
295 ip->i_flag |= IN_CHANGE | IN_UPDATE;
296 allocbuf(bp, nsize);
297 bp->b_flags |= B_DONE;
298 bzero((char *)bp->b_data + osize, (u_int)nsize - osize);
299 *bpp = bp;
300 return (0);
301 }
302#ifdef QUOTA
303 /*
304 * Restore user's disk quota because allocation failed.
305 */
306 (void) chkdq(ip, (long)-btodb(nsize - osize), cred, FORCE);
307#endif
308 brelse(bp);
309nospace:
310 /*
311 * no space available
312 */
313 ffs_fserr(fs, cred->cr_uid, "file system full");
314 uprintf("\n%s: write failed, file system is full\n", fs->fs_fsmnt);
315 return (ENOSPC);
316}
317
318/*
319 * Reallocate a sequence of blocks into a contiguous sequence of blocks.
320 *
321 * The vnode and an array of buffer pointers for a range of sequential
322 * logical blocks to be made contiguous is given. The allocator attempts
323 * to find a range of sequential blocks starting as close as possible to
324 * an fs_rotdelay offset from the end of the allocation for the logical
325 * block immediately preceeding the current range. If successful, the
326 * physical block numbers in the buffer pointers and in the inode are
327 * changed to reflect the new allocation. If unsuccessful, the allocation
328 * is left unchanged. The success in doing the reallocation is returned.
329 * Note that the error return is not reflected back to the user. Rather
330 * the previous block allocation will be used.
331 */
332static int doasyncfree = 1;
333SYSCTL_INT(_vfs_ffs, FFS_ASYNCFREE, doasyncfree, CTLFLAG_RW, &doasyncfree, 0, "");
334
335int doreallocblks = 1;
336SYSCTL_INT(_vfs_ffs, FFS_REALLOCBLKS, doreallocblks, CTLFLAG_RW, &doreallocblks, 0, "");
337
338static int prtrealloc = 0;
339
340int
341ffs_reallocblks(ap)
342 struct vop_reallocblks_args /* {
343 struct vnode *a_vp;
344 struct cluster_save *a_buflist;
345 } */ *ap;
346{
347#if !defined (not_yes)
348 return (ENOSPC);
349#else
350 struct fs *fs;
351 struct inode *ip;
352 struct vnode *vp;
353 struct buf *sbp, *ebp;
354 ufs_daddr_t *bap, *sbap, *ebap = 0;
355 struct cluster_save *buflist;
356 ufs_daddr_t start_lbn, end_lbn, soff, newblk, blkno;
357 struct indir start_ap[NIADDR + 1], end_ap[NIADDR + 1], *idp;
358 int i, len, start_lvl, end_lvl, pref, ssize;
359 struct timeval tv;
360
361 if (doreallocblks == 0)
362 return (ENOSPC);
363 vp = ap->a_vp;
364 ip = VTOI(vp);
365 fs = ip->i_fs;
366 if (fs->fs_contigsumsize <= 0)
367 return (ENOSPC);
368 buflist = ap->a_buflist;
369 len = buflist->bs_nchildren;
370 start_lbn = buflist->bs_children[0]->b_lblkno;
371 end_lbn = start_lbn + len - 1;
372#ifdef DIAGNOSTIC
373 for (i = 0; i < len; i++)
374 if (!ffs_checkblk(ip,
375 dbtofsb(fs, buflist->bs_children[i]->b_blkno), fs->fs_bsize))
376 panic("ffs_reallocblks: unallocated block 1");
377 for (i = 1; i < len; i++)
378 if (buflist->bs_children[i]->b_lblkno != start_lbn + i)
379 panic("ffs_reallocblks: non-logical cluster");
380 blkno = buflist->bs_children[0]->b_blkno;
381 ssize = fsbtodb(fs, fs->fs_frag);
382 for (i = 1; i < len - 1; i++)
383 if (buflist->bs_children[i]->b_blkno != blkno + (i * ssize))
384 panic("ffs_reallocblks: non-physical cluster %d", i);
385#endif
386 /*
387 * If the latest allocation is in a new cylinder group, assume that
388 * the filesystem has decided to move and do not force it back to
389 * the previous cylinder group.
390 */
391 if (dtog(fs, dbtofsb(fs, buflist->bs_children[0]->b_blkno)) !=
392 dtog(fs, dbtofsb(fs, buflist->bs_children[len - 1]->b_blkno)))
393 return (ENOSPC);
394 if (ufs_getlbns(vp, start_lbn, start_ap, &start_lvl) ||
395 ufs_getlbns(vp, end_lbn, end_ap, &end_lvl))
396 return (ENOSPC);
397 /*
398 * Get the starting offset and block map for the first block.
399 */
400 if (start_lvl == 0) {
401 sbap = &ip->i_db[0];
402 soff = start_lbn;
403 } else {
404 idp = &start_ap[start_lvl - 1];
405 if (bread(vp, idp->in_lbn, (int)fs->fs_bsize, NOCRED, &sbp)) {
406 brelse(sbp);
407 return (ENOSPC);
408 }
409 sbap = (ufs_daddr_t *)sbp->b_data;
410 soff = idp->in_off;
411 }
412 /*
413 * Find the preferred location for the cluster.
414 */
415 pref = ffs_blkpref(ip, start_lbn, soff, sbap);
416 /*
417 * If the block range spans two block maps, get the second map.
418 */
419 if (end_lvl == 0 || (idp = &end_ap[end_lvl - 1])->in_off + 1 >= len) {
420 ssize = len;
421 } else {
422#ifdef DIAGNOSTIC
423 if (start_ap[start_lvl-1].in_lbn == idp->in_lbn)
424 panic("ffs_reallocblk: start == end");
425#endif
426 ssize = len - (idp->in_off + 1);
427 if (bread(vp, idp->in_lbn, (int)fs->fs_bsize, NOCRED, &ebp))
428 goto fail;
429 ebap = (ufs_daddr_t *)ebp->b_data;
430 }
431 /*
432 * Search the block map looking for an allocation of the desired size.
433 */
434 if ((newblk = (ufs_daddr_t)ffs_hashalloc(ip, dtog(fs, pref), (long)pref,
435 len, ffs_clusteralloc)) == 0)
436 goto fail;
437 /*
438 * We have found a new contiguous block.
439 *
440 * First we have to replace the old block pointers with the new
441 * block pointers in the inode and indirect blocks associated
442 * with the file.
443 */
444#ifdef DEBUG
445 if (prtrealloc)
446 printf("realloc: ino %d, lbns %d-%d\n\told:", ip->i_number,
447 start_lbn, end_lbn);
448#endif
449 blkno = newblk;
450 for (bap = &sbap[soff], i = 0; i < len; i++, blkno += fs->fs_frag) {
451 if (i == ssize)
452 bap = ebap;
453#ifdef DIAGNOSTIC
454 if (!ffs_checkblk(ip,
455 dbtofsb(fs, buflist->bs_children[i]->b_blkno), fs->fs_bsize))
456 panic("ffs_reallocblks: unallocated block 2");
457 if (dbtofsb(fs, buflist->bs_children[i]->b_blkno) != *bap)
458 panic("ffs_reallocblks: alloc mismatch");
459#endif
460#ifdef DEBUG
461 if (prtrealloc)
462 printf(" %d,", *bap);
463#endif
464 *bap++ = blkno;
465 }
466 /*
467 * Next we must write out the modified inode and indirect blocks.
468 * For strict correctness, the writes should be synchronous since
469 * the old block values may have been written to disk. In practise
470 * they are almost never written, but if we are concerned about
471 * strict correctness, the `doasyncfree' flag should be set to zero.
472 *
473 * The test on `doasyncfree' should be changed to test a flag
474 * that shows whether the associated buffers and inodes have
475 * been written. The flag should be set when the cluster is
476 * started and cleared whenever the buffer or inode is flushed.
477 * We can then check below to see if it is set, and do the
478 * synchronous write only when it has been cleared.
479 */
480 if (sbap != &ip->i_db[0]) {
481 if (doasyncfree)
482 bdwrite(sbp);
483 else
484 bwrite(sbp);
485 } else {
486 ip->i_flag |= IN_CHANGE | IN_UPDATE;
487 if (!doasyncfree) {
488 gettime(&tv);
|
490 }
491 }
492 if (ssize < len)
493 if (doasyncfree)
494 bdwrite(ebp);
495 else
496 bwrite(ebp);
497 /*
498 * Last, free the old blocks and assign the new blocks to the buffers.
499 */
500#ifdef DEBUG
501 if (prtrealloc)
502 printf("\n\tnew:");
503#endif
504 for (blkno = newblk, i = 0; i < len; i++, blkno += fs->fs_frag) {
505 ffs_blkfree(ip, dbtofsb(fs, buflist->bs_children[i]->b_blkno),
506 fs->fs_bsize);
507 buflist->bs_children[i]->b_blkno = fsbtodb(fs, blkno);
508#ifdef DEBUG
509 if (!ffs_checkblk(ip,
510 dbtofsb(fs, buflist->bs_children[i]->b_blkno), fs->fs_bsize))
511 panic("ffs_reallocblks: unallocated block 3");
512 if (prtrealloc)
513 printf(" %d,", blkno);
514#endif
515 }
516#ifdef DEBUG
517 if (prtrealloc) {
518 prtrealloc--;
519 printf("\n");
520 }
521#endif
522 return (0);
523
524fail:
525 if (ssize < len)
526 brelse(ebp);
527 if (sbap != &ip->i_db[0])
528 brelse(sbp);
529 return (ENOSPC);
530#endif
531}
532
533/*
534 * Allocate an inode in the file system.
535 *
536 * If allocating a directory, use ffs_dirpref to select the inode.
537 * If allocating in a directory, the following hierarchy is followed:
538 * 1) allocate the preferred inode.
539 * 2) allocate an inode in the same cylinder group.
540 * 3) quadradically rehash into other cylinder groups, until an
541 * available inode is located.
542 * If no inode preference is given the following heirarchy is used
543 * to allocate an inode:
544 * 1) allocate an inode in cylinder group 0.
545 * 2) quadradically rehash into other cylinder groups, until an
546 * available inode is located.
547 */
548int
549ffs_valloc(pvp, mode, cred, vpp)
550 struct vnode *pvp;
551 int mode;
552 struct ucred *cred;
553 struct vnode **vpp;
554{
555 register struct inode *pip;
556 register struct fs *fs;
557 register struct inode *ip;
558 ino_t ino, ipref;
559 int cg, error;
560
561 *vpp = NULL;
562 pip = VTOI(pvp);
563 fs = pip->i_fs;
564 if (fs->fs_cstotal.cs_nifree == 0)
565 goto noinodes;
566
567 if ((mode & IFMT) == IFDIR)
568 ipref = ffs_dirpref(fs);
569 else
570 ipref = pip->i_number;
571 if (ipref >= fs->fs_ncg * fs->fs_ipg)
572 ipref = 0;
573 cg = ino_to_cg(fs, ipref);
574 ino = (ino_t)ffs_hashalloc(pip, cg, (long)ipref, mode,
575 (allocfcn_t *)ffs_nodealloccg);
576 if (ino == 0)
577 goto noinodes;
578 error = VFS_VGET(pvp->v_mount, ino, vpp);
579 if (error) {
580 UFS_VFREE(pvp, ino, mode);
581 return (error);
582 }
583 ip = VTOI(*vpp);
584 if (ip->i_mode) {
585 printf("mode = 0%o, inum = %ld, fs = %s\n",
586 ip->i_mode, ip->i_number, fs->fs_fsmnt);
587 panic("ffs_valloc: dup alloc");
588 }
589 if (ip->i_blocks) { /* XXX */
590 printf("free inode %s/%ld had %ld blocks\n",
591 fs->fs_fsmnt, ino, ip->i_blocks);
592 ip->i_blocks = 0;
593 }
594 ip->i_flags = 0;
595 /*
596 * Set up a new generation number for this inode.
597 */
598 if (ip->i_gen == 0 || ++(ip->i_gen) == 0)
599 ip->i_gen = random() / 2 + 1;
600 return (0);
601noinodes:
602 ffs_fserr(fs, cred->cr_uid, "out of inodes");
603 uprintf("\n%s: create/symlink failed, no inodes free\n", fs->fs_fsmnt);
604 return (ENOSPC);
605}
606
607/*
608 * Find a cylinder to place a directory.
609 *
610 * The policy implemented by this algorithm is to select from
611 * among those cylinder groups with above the average number of
612 * free inodes, the one with the smallest number of directories.
613 */
614static ino_t
615ffs_dirpref(fs)
616 register struct fs *fs;
617{
618 int cg, minndir, mincg, avgifree;
619
620 avgifree = fs->fs_cstotal.cs_nifree / fs->fs_ncg;
621 minndir = fs->fs_ipg;
622 mincg = 0;
623 for (cg = 0; cg < fs->fs_ncg; cg++)
624 if (fs->fs_cs(fs, cg).cs_ndir < minndir &&
625 fs->fs_cs(fs, cg).cs_nifree >= avgifree) {
626 mincg = cg;
627 minndir = fs->fs_cs(fs, cg).cs_ndir;
628 }
629 return ((ino_t)(fs->fs_ipg * mincg));
630}
631
632/*
633 * Select the desired position for the next block in a file. The file is
634 * logically divided into sections. The first section is composed of the
635 * direct blocks. Each additional section contains fs_maxbpg blocks.
636 *
637 * If no blocks have been allocated in the first section, the policy is to
638 * request a block in the same cylinder group as the inode that describes
639 * the file. If no blocks have been allocated in any other section, the
640 * policy is to place the section in a cylinder group with a greater than
641 * average number of free blocks. An appropriate cylinder group is found
642 * by using a rotor that sweeps the cylinder groups. When a new group of
643 * blocks is needed, the sweep begins in the cylinder group following the
644 * cylinder group from which the previous allocation was made. The sweep
645 * continues until a cylinder group with greater than the average number
646 * of free blocks is found. If the allocation is for the first block in an
647 * indirect block, the information on the previous allocation is unavailable;
648 * here a best guess is made based upon the logical block number being
649 * allocated.
650 *
651 * If a section is already partially allocated, the policy is to
652 * contiguously allocate fs_maxcontig blocks. The end of one of these
653 * contiguous blocks and the beginning of the next is physically separated
654 * so that the disk head will be in transit between them for at least
655 * fs_rotdelay milliseconds. This is to allow time for the processor to
656 * schedule another I/O transfer.
657 */
658ufs_daddr_t
659ffs_blkpref(ip, lbn, indx, bap)
660 struct inode *ip;
661 ufs_daddr_t lbn;
662 int indx;
663 ufs_daddr_t *bap;
664{
665 register struct fs *fs;
666 register int cg;
667 int avgbfree, startcg;
668 ufs_daddr_t nextblk;
669
670 fs = ip->i_fs;
671 if (indx % fs->fs_maxbpg == 0 || bap[indx - 1] == 0) {
672 if (lbn < NDADDR) {
673 cg = ino_to_cg(fs, ip->i_number);
674 return (fs->fs_fpg * cg + fs->fs_frag);
675 }
676 /*
677 * Find a cylinder with greater than average number of
678 * unused data blocks.
679 */
680 if (indx == 0 || bap[indx - 1] == 0)
681 startcg =
682 ino_to_cg(fs, ip->i_number) + lbn / fs->fs_maxbpg;
683 else
684 startcg = dtog(fs, bap[indx - 1]) + 1;
685 startcg %= fs->fs_ncg;
686 avgbfree = fs->fs_cstotal.cs_nbfree / fs->fs_ncg;
687 for (cg = startcg; cg < fs->fs_ncg; cg++)
688 if (fs->fs_cs(fs, cg).cs_nbfree >= avgbfree) {
689 fs->fs_cgrotor = cg;
690 return (fs->fs_fpg * cg + fs->fs_frag);
691 }
692 for (cg = 0; cg <= startcg; cg++)
693 if (fs->fs_cs(fs, cg).cs_nbfree >= avgbfree) {
694 fs->fs_cgrotor = cg;
695 return (fs->fs_fpg * cg + fs->fs_frag);
696 }
697 return (0);
698 }
699 /*
700 * One or more previous blocks have been laid out. If less
701 * than fs_maxcontig previous blocks are contiguous, the
702 * next block is requested contiguously, otherwise it is
703 * requested rotationally delayed by fs_rotdelay milliseconds.
704 */
705 nextblk = bap[indx - 1] + fs->fs_frag;
706 if (fs->fs_rotdelay == 0 || indx < fs->fs_maxcontig ||
707 bap[indx - fs->fs_maxcontig] +
708 blkstofrags(fs, fs->fs_maxcontig) != nextblk)
709 return (nextblk);
710 /*
711 * Here we convert ms of delay to frags as:
712 * (frags) = (ms) * (rev/sec) * (sect/rev) /
713 * ((sect/frag) * (ms/sec))
714 * then round up to the next block.
715 */
716 nextblk += roundup(fs->fs_rotdelay * fs->fs_rps * fs->fs_nsect /
717 (NSPF(fs) * 1000), fs->fs_frag);
718 return (nextblk);
719}
720
721/*
722 * Implement the cylinder overflow algorithm.
723 *
724 * The policy implemented by this algorithm is:
725 * 1) allocate the block in its requested cylinder group.
726 * 2) quadradically rehash on the cylinder group number.
727 * 3) brute force search for a free block.
728 */
729/*VARARGS5*/
730static u_long
731ffs_hashalloc(ip, cg, pref, size, allocator)
732 struct inode *ip;
733 int cg;
734 long pref;
735 int size; /* size for data blocks, mode for inodes */
736 allocfcn_t *allocator;
737{
738 register struct fs *fs;
739 long result; /* XXX why not same type as we return? */
740 int i, icg = cg;
741
742 fs = ip->i_fs;
743 /*
744 * 1: preferred cylinder group
745 */
746 result = (*allocator)(ip, cg, pref, size);
747 if (result)
748 return (result);
749 /*
750 * 2: quadratic rehash
751 */
752 for (i = 1; i < fs->fs_ncg; i *= 2) {
753 cg += i;
754 if (cg >= fs->fs_ncg)
755 cg -= fs->fs_ncg;
756 result = (*allocator)(ip, cg, 0, size);
757 if (result)
758 return (result);
759 }
760 /*
761 * 3: brute force search
762 * Note that we start at i == 2, since 0 was checked initially,
763 * and 1 is always checked in the quadratic rehash.
764 */
765 cg = (icg + 2) % fs->fs_ncg;
766 for (i = 2; i < fs->fs_ncg; i++) {
767 result = (*allocator)(ip, cg, 0, size);
768 if (result)
769 return (result);
770 cg++;
771 if (cg == fs->fs_ncg)
772 cg = 0;
773 }
774 return (0);
775}
776
777/*
778 * Determine whether a fragment can be extended.
779 *
780 * Check to see if the necessary fragments are available, and
781 * if they are, allocate them.
782 */
783static ufs_daddr_t
784ffs_fragextend(ip, cg, bprev, osize, nsize)
785 struct inode *ip;
786 int cg;
787 long bprev;
788 int osize, nsize;
789{
790 register struct fs *fs;
791 register struct cg *cgp;
792 struct buf *bp;
793 long bno;
794 int frags, bbase;
795 int i, error;
796
797 fs = ip->i_fs;
798 if (fs->fs_cs(fs, cg).cs_nffree < numfrags(fs, nsize - osize))
799 return (0);
800 frags = numfrags(fs, nsize);
801 bbase = fragnum(fs, bprev);
802 if (bbase > fragnum(fs, (bprev + frags - 1))) {
803 /* cannot extend across a block boundary */
804 return (0);
805 }
806 error = bread(ip->i_devvp, fsbtodb(fs, cgtod(fs, cg)),
807 (int)fs->fs_cgsize, NOCRED, &bp);
808 if (error) {
809 brelse(bp);
810 return (0);
811 }
812 cgp = (struct cg *)bp->b_data;
813 if (!cg_chkmagic(cgp)) {
814 brelse(bp);
815 return (0);
816 }
817 cgp->cg_time = time.tv_sec;
818 bno = dtogd(fs, bprev);
819 for (i = numfrags(fs, osize); i < frags; i++)
820 if (isclr(cg_blksfree(cgp), bno + i)) {
821 brelse(bp);
822 return (0);
823 }
824 /*
825 * the current fragment can be extended
826 * deduct the count on fragment being extended into
827 * increase the count on the remaining fragment (if any)
828 * allocate the extended piece
829 */
830 for (i = frags; i < fs->fs_frag - bbase; i++)
831 if (isclr(cg_blksfree(cgp), bno + i))
832 break;
833 cgp->cg_frsum[i - numfrags(fs, osize)]--;
834 if (i != frags)
835 cgp->cg_frsum[i - frags]++;
836 for (i = numfrags(fs, osize); i < frags; i++) {
837 clrbit(cg_blksfree(cgp), bno + i);
838 cgp->cg_cs.cs_nffree--;
839 fs->fs_cstotal.cs_nffree--;
840 fs->fs_cs(fs, cg).cs_nffree--;
841 }
842 fs->fs_fmod = 1;
843 bdwrite(bp);
844 return (bprev);
845}
846
847/*
848 * Determine whether a block can be allocated.
849 *
850 * Check to see if a block of the appropriate size is available,
851 * and if it is, allocate it.
852 */
853static ufs_daddr_t
854ffs_alloccg(ip, cg, bpref, size)
855 struct inode *ip;
856 int cg;
857 ufs_daddr_t bpref;
858 int size;
859{
860 register struct fs *fs;
861 register struct cg *cgp;
862 struct buf *bp;
863 register int i;
864 int error, bno, frags, allocsiz;
865
866 fs = ip->i_fs;
867 if (fs->fs_cs(fs, cg).cs_nbfree == 0 && size == fs->fs_bsize)
868 return (0);
869 error = bread(ip->i_devvp, fsbtodb(fs, cgtod(fs, cg)),
870 (int)fs->fs_cgsize, NOCRED, &bp);
871 if (error) {
872 brelse(bp);
873 return (0);
874 }
875 cgp = (struct cg *)bp->b_data;
876 if (!cg_chkmagic(cgp) ||
877 (cgp->cg_cs.cs_nbfree == 0 && size == fs->fs_bsize)) {
878 brelse(bp);
879 return (0);
880 }
881 cgp->cg_time = time.tv_sec;
882 if (size == fs->fs_bsize) {
883 bno = ffs_alloccgblk(fs, cgp, bpref);
884 bdwrite(bp);
885 return (bno);
886 }
887 /*
888 * check to see if any fragments are already available
889 * allocsiz is the size which will be allocated, hacking
890 * it down to a smaller size if necessary
891 */
892 frags = numfrags(fs, size);
893 for (allocsiz = frags; allocsiz < fs->fs_frag; allocsiz++)
894 if (cgp->cg_frsum[allocsiz] != 0)
895 break;
896 if (allocsiz == fs->fs_frag) {
897 /*
898 * no fragments were available, so a block will be
899 * allocated, and hacked up
900 */
901 if (cgp->cg_cs.cs_nbfree == 0) {
902 brelse(bp);
903 return (0);
904 }
905 bno = ffs_alloccgblk(fs, cgp, bpref);
906 bpref = dtogd(fs, bno);
907 for (i = frags; i < fs->fs_frag; i++)
908 setbit(cg_blksfree(cgp), bpref + i);
909 i = fs->fs_frag - frags;
910 cgp->cg_cs.cs_nffree += i;
911 fs->fs_cstotal.cs_nffree += i;
912 fs->fs_cs(fs, cg).cs_nffree += i;
913 fs->fs_fmod = 1;
914 cgp->cg_frsum[i]++;
915 bdwrite(bp);
916 return (bno);
917 }
918 bno = ffs_mapsearch(fs, cgp, bpref, allocsiz);
919 if (bno < 0) {
920 brelse(bp);
921 return (0);
922 }
923 for (i = 0; i < frags; i++)
924 clrbit(cg_blksfree(cgp), bno + i);
925 cgp->cg_cs.cs_nffree -= frags;
926 fs->fs_cstotal.cs_nffree -= frags;
927 fs->fs_cs(fs, cg).cs_nffree -= frags;
928 fs->fs_fmod = 1;
929 cgp->cg_frsum[allocsiz]--;
930 if (frags != allocsiz)
931 cgp->cg_frsum[allocsiz - frags]++;
932 bdwrite(bp);
933 return (cg * fs->fs_fpg + bno);
934}
935
936/*
937 * Allocate a block in a cylinder group.
938 *
939 * This algorithm implements the following policy:
940 * 1) allocate the requested block.
941 * 2) allocate a rotationally optimal block in the same cylinder.
942 * 3) allocate the next available block on the block rotor for the
943 * specified cylinder group.
944 * Note that this routine only allocates fs_bsize blocks; these
945 * blocks may be fragmented by the routine that allocates them.
946 */
947static ufs_daddr_t
948ffs_alloccgblk(fs, cgp, bpref)
949 register struct fs *fs;
950 register struct cg *cgp;
951 ufs_daddr_t bpref;
952{
953 ufs_daddr_t bno, blkno;
954 int cylno, pos, delta;
955 short *cylbp;
956 register int i;
957
958 if (bpref == 0 || dtog(fs, bpref) != cgp->cg_cgx) {
959 bpref = cgp->cg_rotor;
960 goto norot;
961 }
962 bpref = blknum(fs, bpref);
963 bpref = dtogd(fs, bpref);
964 /*
965 * if the requested block is available, use it
966 */
967 if (ffs_isblock(fs, cg_blksfree(cgp), fragstoblks(fs, bpref))) {
968 bno = bpref;
969 goto gotit;
970 }
971 if (fs->fs_nrpos <= 1 || fs->fs_cpc == 0) {
972 /*
973 * Block layout information is not available.
974 * Leaving bpref unchanged means we take the
975 * next available free block following the one
976 * we just allocated. Hopefully this will at
977 * least hit a track cache on drives of unknown
978 * geometry (e.g. SCSI).
979 */
980 goto norot;
981 }
982 /*
983 * check for a block available on the same cylinder
984 */
985 cylno = cbtocylno(fs, bpref);
986 if (cg_blktot(cgp)[cylno] == 0)
987 goto norot;
988 /*
989 * check the summary information to see if a block is
990 * available in the requested cylinder starting at the
991 * requested rotational position and proceeding around.
992 */
993 cylbp = cg_blks(fs, cgp, cylno);
994 pos = cbtorpos(fs, bpref);
995 for (i = pos; i < fs->fs_nrpos; i++)
996 if (cylbp[i] > 0)
997 break;
998 if (i == fs->fs_nrpos)
999 for (i = 0; i < pos; i++)
1000 if (cylbp[i] > 0)
1001 break;
1002 if (cylbp[i] > 0) {
1003 /*
1004 * found a rotational position, now find the actual
1005 * block. A panic if none is actually there.
1006 */
1007 pos = cylno % fs->fs_cpc;
1008 bno = (cylno - pos) * fs->fs_spc / NSPB(fs);
1009 if (fs_postbl(fs, pos)[i] == -1) {
1010 printf("pos = %d, i = %d, fs = %s\n",
1011 pos, i, fs->fs_fsmnt);
1012 panic("ffs_alloccgblk: cyl groups corrupted");
1013 }
1014 for (i = fs_postbl(fs, pos)[i];; ) {
1015 if (ffs_isblock(fs, cg_blksfree(cgp), bno + i)) {
1016 bno = blkstofrags(fs, (bno + i));
1017 goto gotit;
1018 }
1019 delta = fs_rotbl(fs)[i];
1020 if (delta <= 0 ||
1021 delta + i > fragstoblks(fs, fs->fs_fpg))
1022 break;
1023 i += delta;
1024 }
1025 printf("pos = %d, i = %d, fs = %s\n", pos, i, fs->fs_fsmnt);
1026 panic("ffs_alloccgblk: can't find blk in cyl");
1027 }
1028norot:
1029 /*
1030 * no blocks in the requested cylinder, so take next
1031 * available one in this cylinder group.
1032 */
1033 bno = ffs_mapsearch(fs, cgp, bpref, (int)fs->fs_frag);
1034 if (bno < 0)
1035 return (0);
1036 cgp->cg_rotor = bno;
1037gotit:
1038 blkno = fragstoblks(fs, bno);
1039 ffs_clrblock(fs, cg_blksfree(cgp), (long)blkno);
1040 ffs_clusteracct(fs, cgp, blkno, -1);
1041 cgp->cg_cs.cs_nbfree--;
1042 fs->fs_cstotal.cs_nbfree--;
1043 fs->fs_cs(fs, cgp->cg_cgx).cs_nbfree--;
1044 cylno = cbtocylno(fs, bno);
1045 cg_blks(fs, cgp, cylno)[cbtorpos(fs, bno)]--;
1046 cg_blktot(cgp)[cylno]--;
1047 fs->fs_fmod = 1;
1048 return (cgp->cg_cgx * fs->fs_fpg + bno);
1049}
1050
1051#ifdef notyet
1052/*
1053 * Determine whether a cluster can be allocated.
1054 *
1055 * We do not currently check for optimal rotational layout if there
1056 * are multiple choices in the same cylinder group. Instead we just
1057 * take the first one that we find following bpref.
1058 */
1059static ufs_daddr_t
1060ffs_clusteralloc(ip, cg, bpref, len)
1061 struct inode *ip;
1062 int cg;
1063 ufs_daddr_t bpref;
1064 int len;
1065{
1066 register struct fs *fs;
1067 register struct cg *cgp;
1068 struct buf *bp;
1069 int i, got, run, bno, bit, map;
1070 u_char *mapp;
1071 int32_t *lp;
1072
1073 fs = ip->i_fs;
1074 if (fs->fs_maxcluster[cg] < len)
1075 return (NULL);
1076 if (bread(ip->i_devvp, fsbtodb(fs, cgtod(fs, cg)), (int)fs->fs_cgsize,
1077 NOCRED, &bp))
1078 goto fail;
1079 cgp = (struct cg *)bp->b_data;
1080 if (!cg_chkmagic(cgp))
1081 goto fail;
1082 /*
1083 * Check to see if a cluster of the needed size (or bigger) is
1084 * available in this cylinder group.
1085 */
1086 lp = &cg_clustersum(cgp)[len];
1087 for (i = len; i <= fs->fs_contigsumsize; i++)
1088 if (*lp++ > 0)
1089 break;
1090 if (i > fs->fs_contigsumsize) {
1091 /*
1092 * This is the first time looking for a cluster in this
1093 * cylinder group. Update the cluster summary information
1094 * to reflect the true maximum sized cluster so that
1095 * future cluster allocation requests can avoid reading
1096 * the cylinder group map only to find no clusters.
1097 */
1098 lp = &cg_clustersum(cgp)[len - 1];
1099 for (i = len - 1; i > 0; i--)
1100 if (*lp-- > 0)
1101 break;
1102 fs->fs_maxcluster[cg] = i;
1103 goto fail;
1104 }
1105 /*
1106 * Search the cluster map to find a big enough cluster.
1107 * We take the first one that we find, even if it is larger
1108 * than we need as we prefer to get one close to the previous
1109 * block allocation. We do not search before the current
1110 * preference point as we do not want to allocate a block
1111 * that is allocated before the previous one (as we will
1112 * then have to wait for another pass of the elevator
1113 * algorithm before it will be read). We prefer to fail and
1114 * be recalled to try an allocation in the next cylinder group.
1115 */
1116 if (dtog(fs, bpref) != cg)
1117 bpref = 0;
1118 else
1119 bpref = fragstoblks(fs, dtogd(fs, blknum(fs, bpref)));
1120 mapp = &cg_clustersfree(cgp)[bpref / NBBY];
1121 map = *mapp++;
1122 bit = 1 << (bpref % NBBY);
1123 for (run = 0, got = bpref; got < cgp->cg_nclusterblks; got++) {
1124 if ((map & bit) == 0) {
1125 run = 0;
1126 } else {
1127 run++;
1128 if (run == len)
1129 break;
1130 }
1131 if ((got & (NBBY - 1)) != (NBBY - 1)) {
1132 bit <<= 1;
1133 } else {
1134 map = *mapp++;
1135 bit = 1;
1136 }
1137 }
1138 if (got >= cgp->cg_nclusterblks)
1139 goto fail;
1140 /*
1141 * Allocate the cluster that we have found.
1142 */
1143 for (i = 1; i <= len; i++)
1144 if (!ffs_isblock(fs, cg_blksfree(cgp), got - run + i))
1145 panic("ffs_clusteralloc: map mismatch");
1146 bno = cg * fs->fs_fpg + blkstofrags(fs, got - run + 1);
1147 if (dtog(fs, bno) != cg)
1148 panic("ffs_clusteralloc: allocated out of group");
1149 len = blkstofrags(fs, len);
1150 for (i = 0; i < len; i += fs->fs_frag)
1151 if ((got = ffs_alloccgblk(fs, cgp, bno + i)) != bno + i)
1152 panic("ffs_clusteralloc: lost block");
1153 bdwrite(bp);
1154 return (bno);
1155
1156fail:
1157 brelse(bp);
1158 return (0);
1159}
1160#endif
1161
1162/*
1163 * Determine whether an inode can be allocated.
1164 *
1165 * Check to see if an inode is available, and if it is,
1166 * allocate it using the following policy:
1167 * 1) allocate the requested inode.
1168 * 2) allocate the next available inode after the requested
1169 * inode in the specified cylinder group.
1170 */
1171static ino_t
1172ffs_nodealloccg(ip, cg, ipref, mode)
1173 struct inode *ip;
1174 int cg;
1175 ufs_daddr_t ipref;
1176 int mode;
1177{
1178 register struct fs *fs;
1179 register struct cg *cgp;
1180 struct buf *bp;
1181 int error, start, len, loc, map, i;
1182
1183 fs = ip->i_fs;
1184 if (fs->fs_cs(fs, cg).cs_nifree == 0)
1185 return (0);
1186 error = bread(ip->i_devvp, fsbtodb(fs, cgtod(fs, cg)),
1187 (int)fs->fs_cgsize, NOCRED, &bp);
1188 if (error) {
1189 brelse(bp);
1190 return (0);
1191 }
1192 cgp = (struct cg *)bp->b_data;
1193 if (!cg_chkmagic(cgp) || cgp->cg_cs.cs_nifree == 0) {
1194 brelse(bp);
1195 return (0);
1196 }
1197 cgp->cg_time = time.tv_sec;
1198 if (ipref) {
1199 ipref %= fs->fs_ipg;
1200 if (isclr(cg_inosused(cgp), ipref))
1201 goto gotit;
1202 }
1203 start = cgp->cg_irotor / NBBY;
1204 len = howmany(fs->fs_ipg - cgp->cg_irotor, NBBY);
1205 loc = skpc(0xff, len, &cg_inosused(cgp)[start]);
1206 if (loc == 0) {
1207 len = start + 1;
1208 start = 0;
1209 loc = skpc(0xff, len, &cg_inosused(cgp)[0]);
1210 if (loc == 0) {
1211 printf("cg = %d, irotor = %ld, fs = %s\n",
1212 cg, cgp->cg_irotor, fs->fs_fsmnt);
1213 panic("ffs_nodealloccg: map corrupted");
1214 /* NOTREACHED */
1215 }
1216 }
1217 i = start + len - loc;
1218 map = cg_inosused(cgp)[i];
1219 ipref = i * NBBY;
1220 for (i = 1; i < (1 << NBBY); i <<= 1, ipref++) {
1221 if ((map & i) == 0) {
1222 cgp->cg_irotor = ipref;
1223 goto gotit;
1224 }
1225 }
1226 printf("fs = %s\n", fs->fs_fsmnt);
1227 panic("ffs_nodealloccg: block not in map");
1228 /* NOTREACHED */
1229gotit:
1230 setbit(cg_inosused(cgp), ipref);
1231 cgp->cg_cs.cs_nifree--;
1232 fs->fs_cstotal.cs_nifree--;
1233 fs->fs_cs(fs, cg).cs_nifree--;
1234 fs->fs_fmod = 1;
1235 if ((mode & IFMT) == IFDIR) {
1236 cgp->cg_cs.cs_ndir++;
1237 fs->fs_cstotal.cs_ndir++;
1238 fs->fs_cs(fs, cg).cs_ndir++;
1239 }
1240 bdwrite(bp);
1241 return (cg * fs->fs_ipg + ipref);
1242}
1243
1244/*
1245 * Free a block or fragment.
1246 *
1247 * The specified block or fragment is placed back in the
1248 * free map. If a fragment is deallocated, a possible
1249 * block reassembly is checked.
1250 */
1251void
1252ffs_blkfree(ip, bno, size)
1253 register struct inode *ip;
1254 ufs_daddr_t bno;
1255 long size;
1256{
1257 register struct fs *fs;
1258 register struct cg *cgp;
1259 struct buf *bp;
1260 ufs_daddr_t blkno;
1261 int i, error, cg, blk, frags, bbase;
1262
1263 fs = ip->i_fs;
1264 if ((u_int)size > fs->fs_bsize || fragoff(fs, size) != 0) {
1265 printf("dev = 0x%lx, bsize = %ld, size = %ld, fs = %s\n",
1266 (u_long)ip->i_dev, fs->fs_bsize, size, fs->fs_fsmnt);
1267 panic("ffs_blkfree: bad size");
1268 }
1269 cg = dtog(fs, bno);
1270 if ((u_int)bno >= fs->fs_size) {
1271 printf("bad block %ld, ino %ld\n", bno, ip->i_number);
1272 ffs_fserr(fs, ip->i_uid, "bad block");
1273 return;
1274 }
1275 error = bread(ip->i_devvp, fsbtodb(fs, cgtod(fs, cg)),
1276 (int)fs->fs_cgsize, NOCRED, &bp);
1277 if (error) {
1278 brelse(bp);
1279 return;
1280 }
1281 cgp = (struct cg *)bp->b_data;
1282 if (!cg_chkmagic(cgp)) {
1283 brelse(bp);
1284 return;
1285 }
1286 cgp->cg_time = time.tv_sec;
1287 bno = dtogd(fs, bno);
1288 if (size == fs->fs_bsize) {
1289 blkno = fragstoblks(fs, bno);
1290 if (ffs_isblock(fs, cg_blksfree(cgp), blkno)) {
1291 printf("dev = 0x%lx, block = %ld, fs = %s\n",
1292 (u_long) ip->i_dev, bno, fs->fs_fsmnt);
1293 panic("ffs_blkfree: freeing free block");
1294 }
1295 ffs_setblock(fs, cg_blksfree(cgp), blkno);
1296 ffs_clusteracct(fs, cgp, blkno, 1);
1297 cgp->cg_cs.cs_nbfree++;
1298 fs->fs_cstotal.cs_nbfree++;
1299 fs->fs_cs(fs, cg).cs_nbfree++;
1300 i = cbtocylno(fs, bno);
1301 cg_blks(fs, cgp, i)[cbtorpos(fs, bno)]++;
1302 cg_blktot(cgp)[i]++;
1303 } else {
1304 bbase = bno - fragnum(fs, bno);
1305 /*
1306 * decrement the counts associated with the old frags
1307 */
1308 blk = blkmap(fs, cg_blksfree(cgp), bbase);
1309 ffs_fragacct(fs, blk, cgp->cg_frsum, -1);
1310 /*
1311 * deallocate the fragment
1312 */
1313 frags = numfrags(fs, size);
1314 for (i = 0; i < frags; i++) {
1315 if (isset(cg_blksfree(cgp), bno + i)) {
1316 printf("dev = 0x%lx, block = %ld, fs = %s\n",
1317 (u_long) ip->i_dev, bno + i, fs->fs_fsmnt);
1318 panic("ffs_blkfree: freeing free frag");
1319 }
1320 setbit(cg_blksfree(cgp), bno + i);
1321 }
1322 cgp->cg_cs.cs_nffree += i;
1323 fs->fs_cstotal.cs_nffree += i;
1324 fs->fs_cs(fs, cg).cs_nffree += i;
1325 /*
1326 * add back in counts associated with the new frags
1327 */
1328 blk = blkmap(fs, cg_blksfree(cgp), bbase);
1329 ffs_fragacct(fs, blk, cgp->cg_frsum, 1);
1330 /*
1331 * if a complete block has been reassembled, account for it
1332 */
1333 blkno = fragstoblks(fs, bbase);
1334 if (ffs_isblock(fs, cg_blksfree(cgp), blkno)) {
1335 cgp->cg_cs.cs_nffree -= fs->fs_frag;
1336 fs->fs_cstotal.cs_nffree -= fs->fs_frag;
1337 fs->fs_cs(fs, cg).cs_nffree -= fs->fs_frag;
1338 ffs_clusteracct(fs, cgp, blkno, 1);
1339 cgp->cg_cs.cs_nbfree++;
1340 fs->fs_cstotal.cs_nbfree++;
1341 fs->fs_cs(fs, cg).cs_nbfree++;
1342 i = cbtocylno(fs, bbase);
1343 cg_blks(fs, cgp, i)[cbtorpos(fs, bbase)]++;
1344 cg_blktot(cgp)[i]++;
1345 }
1346 }
1347 fs->fs_fmod = 1;
1348 bdwrite(bp);
1349}
1350
1351#ifdef DIAGNOSTIC
1352/*
1353 * Verify allocation of a block or fragment. Returns true if block or
1354 * fragment is allocated, false if it is free.
1355 */
1356int
1357ffs_checkblk(ip, bno, size)
1358 struct inode *ip;
1359 ufs_daddr_t bno;
1360 long size;
1361{
1362 struct fs *fs;
1363 struct cg *cgp;
1364 struct buf *bp;
1365 int i, error, frags, free;
1366
1367 fs = ip->i_fs;
1368 if ((u_int)size > fs->fs_bsize || fragoff(fs, size) != 0) {
1369 printf("bsize = %d, size = %d, fs = %s\n",
1370 fs->fs_bsize, size, fs->fs_fsmnt);
1371 panic("ffs_checkblk: bad size");
1372 }
1373 if ((u_int)bno >= fs->fs_size)
1374 panic("ffs_checkblk: bad block %d", bno);
1375 error = bread(ip->i_devvp, fsbtodb(fs, cgtod(fs, dtog(fs, bno))),
1376 (int)fs->fs_cgsize, NOCRED, &bp);
1377 if (error)
1378 panic("ffs_checkblk: cg bread failed");
1379 cgp = (struct cg *)bp->b_data;
1380 if (!cg_chkmagic(cgp))
1381 panic("ffs_checkblk: cg magic mismatch");
1382 bno = dtogd(fs, bno);
1383 if (size == fs->fs_bsize) {
1384 free = ffs_isblock(fs, cg_blksfree(cgp), fragstoblks(fs, bno));
1385 } else {
1386 frags = numfrags(fs, size);
1387 for (free = 0, i = 0; i < frags; i++)
1388 if (isset(cg_blksfree(cgp), bno + i))
1389 free++;
1390 if (free != 0 && free != frags)
1391 panic("ffs_checkblk: partially free fragment");
1392 }
1393 brelse(bp);
1394 return (!free);
1395}
1396#endif /* DIAGNOSTIC */
1397
1398/*
1399 * Free an inode.
1400 *
1401 * The specified inode is placed back in the free map.
1402 */
1403int
1404ffs_vfree(pvp, ino, mode)
1405 struct vnode *pvp;
1406 ino_t ino;
1407 int mode;
1408{
1409 register struct fs *fs;
1410 register struct cg *cgp;
1411 register struct inode *pip;
1412 struct buf *bp;
1413 int error, cg;
1414
1415 pip = VTOI(pvp);
1416 fs = pip->i_fs;
1417 if ((u_int)ino >= fs->fs_ipg * fs->fs_ncg)
1418 panic("ffs_vfree: range: dev = 0x%x, ino = %d, fs = %s",
1419 pip->i_dev, ino, fs->fs_fsmnt);
1420 cg = ino_to_cg(fs, ino);
1421 error = bread(pip->i_devvp, fsbtodb(fs, cgtod(fs, cg)),
1422 (int)fs->fs_cgsize, NOCRED, &bp);
1423 if (error) {
1424 brelse(bp);
1425 return (0);
1426 }
1427 cgp = (struct cg *)bp->b_data;
1428 if (!cg_chkmagic(cgp)) {
1429 brelse(bp);
1430 return (0);
1431 }
1432 cgp->cg_time = time.tv_sec;
1433 ino %= fs->fs_ipg;
1434 if (isclr(cg_inosused(cgp), ino)) {
1435 printf("dev = 0x%lx, ino = %ld, fs = %s\n",
1436 (u_long)pip->i_dev, ino, fs->fs_fsmnt);
1437 if (fs->fs_ronly == 0)
1438 panic("ffs_vfree: freeing free inode");
1439 }
1440 clrbit(cg_inosused(cgp), ino);
1441 if (ino < cgp->cg_irotor)
1442 cgp->cg_irotor = ino;
1443 cgp->cg_cs.cs_nifree++;
1444 fs->fs_cstotal.cs_nifree++;
1445 fs->fs_cs(fs, cg).cs_nifree++;
1446 if ((mode & IFMT) == IFDIR) {
1447 cgp->cg_cs.cs_ndir--;
1448 fs->fs_cstotal.cs_ndir--;
1449 fs->fs_cs(fs, cg).cs_ndir--;
1450 }
1451 fs->fs_fmod = 1;
1452 bdwrite(bp);
1453 return (0);
1454}
1455
1456/*
1457 * Find a block of the specified size in the specified cylinder group.
1458 *
1459 * It is a panic if a request is made to find a block if none are
1460 * available.
1461 */
1462static ufs_daddr_t
1463ffs_mapsearch(fs, cgp, bpref, allocsiz)
1464 register struct fs *fs;
1465 register struct cg *cgp;
1466 ufs_daddr_t bpref;
1467 int allocsiz;
1468{
1469 ufs_daddr_t bno;
1470 int start, len, loc, i;
1471 int blk, field, subfield, pos;
1472
1473 /*
1474 * find the fragment by searching through the free block
1475 * map for an appropriate bit pattern
1476 */
1477 if (bpref)
1478 start = dtogd(fs, bpref) / NBBY;
1479 else
1480 start = cgp->cg_frotor / NBBY;
1481 len = howmany(fs->fs_fpg, NBBY) - start;
1482 loc = scanc((u_int)len, (u_char *)&cg_blksfree(cgp)[start],
1483 (u_char *)fragtbl[fs->fs_frag],
1484 (u_char)(1 << (allocsiz - 1 + (fs->fs_frag % NBBY))));
1485 if (loc == 0) {
1486 len = start + 1;
1487 start = 0;
1488 loc = scanc((u_int)len, (u_char *)&cg_blksfree(cgp)[0],
1489 (u_char *)fragtbl[fs->fs_frag],
1490 (u_char)(1 << (allocsiz - 1 + (fs->fs_frag % NBBY))));
1491 if (loc == 0) {
1492 printf("start = %d, len = %d, fs = %s\n",
1493 start, len, fs->fs_fsmnt);
1494 panic("ffs_alloccg: map corrupted");
1495 /* NOTREACHED */
1496 }
1497 }
1498 bno = (start + len - loc) * NBBY;
1499 cgp->cg_frotor = bno;
1500 /*
1501 * found the byte in the map
1502 * sift through the bits to find the selected frag
1503 */
1504 for (i = bno + NBBY; bno < i; bno += fs->fs_frag) {
1505 blk = blkmap(fs, cg_blksfree(cgp), bno);
1506 blk <<= 1;
1507 field = around[allocsiz];
1508 subfield = inside[allocsiz];
1509 for (pos = 0; pos <= fs->fs_frag - allocsiz; pos++) {
1510 if ((blk & field) == subfield)
1511 return (bno + pos);
1512 field <<= 1;
1513 subfield <<= 1;
1514 }
1515 }
1516 printf("bno = %lu, fs = %s\n", (u_long)bno, fs->fs_fsmnt);
1517 panic("ffs_alloccg: block not in map");
1518 return (-1);
1519}
1520
1521/*
1522 * Update the cluster map because of an allocation or free.
1523 *
1524 * Cnt == 1 means free; cnt == -1 means allocating.
1525 */
1526static void
1527ffs_clusteracct(fs, cgp, blkno, cnt)
1528 struct fs *fs;
1529 struct cg *cgp;
1530 ufs_daddr_t blkno;
1531 int cnt;
1532{
1533 int32_t *sump;
1534 int32_t *lp;
1535 u_char *freemapp, *mapp;
1536 int i, start, end, forw, back, map, bit;
1537
1538 if (fs->fs_contigsumsize <= 0)
1539 return;
1540 freemapp = cg_clustersfree(cgp);
1541 sump = cg_clustersum(cgp);
1542 /*
1543 * Allocate or clear the actual block.
1544 */
1545 if (cnt > 0)
1546 setbit(freemapp, blkno);
1547 else
1548 clrbit(freemapp, blkno);
1549 /*
1550 * Find the size of the cluster going forward.
1551 */
1552 start = blkno + 1;
1553 end = start + fs->fs_contigsumsize;
1554 if (end >= cgp->cg_nclusterblks)
1555 end = cgp->cg_nclusterblks;
1556 mapp = &freemapp[start / NBBY];
1557 map = *mapp++;
1558 bit = 1 << (start % NBBY);
1559 for (i = start; i < end; i++) {
1560 if ((map & bit) == 0)
1561 break;
1562 if ((i & (NBBY - 1)) != (NBBY - 1)) {
1563 bit <<= 1;
1564 } else {
1565 map = *mapp++;
1566 bit = 1;
1567 }
1568 }
1569 forw = i - start;
1570 /*
1571 * Find the size of the cluster going backward.
1572 */
1573 start = blkno - 1;
1574 end = start - fs->fs_contigsumsize;
1575 if (end < 0)
1576 end = -1;
1577 mapp = &freemapp[start / NBBY];
1578 map = *mapp--;
1579 bit = 1 << (start % NBBY);
1580 for (i = start; i > end; i--) {
1581 if ((map & bit) == 0)
1582 break;
1583 if ((i & (NBBY - 1)) != 0) {
1584 bit >>= 1;
1585 } else {
1586 map = *mapp--;
1587 bit = 1 << (NBBY - 1);
1588 }
1589 }
1590 back = start - i;
1591 /*
1592 * Account for old cluster and the possibly new forward and
1593 * back clusters.
1594 */
1595 i = back + forw + 1;
1596 if (i > fs->fs_contigsumsize)
1597 i = fs->fs_contigsumsize;
1598 sump[i] += cnt;
1599 if (back > 0)
1600 sump[back] -= cnt;
1601 if (forw > 0)
1602 sump[forw] -= cnt;
1603 /*
1604 * Update cluster summary information.
1605 */
1606 lp = &sump[fs->fs_contigsumsize];
1607 for (i = fs->fs_contigsumsize; i > 0; i--)
1608 if (*lp-- > 0)
1609 break;
1610 fs->fs_maxcluster[cgp->cg_cgx] = i;
1611}
1612
1613/*
1614 * Fserr prints the name of a file system with an error diagnostic.
1615 *
1616 * The form of the error message is:
1617 * fs: error message
1618 */
1619static void
1620ffs_fserr(fs, uid, cp)
1621 struct fs *fs;
1622 u_int uid;
1623 char *cp;
1624{
1625 struct proc *p = curproc; /* XXX */
1626
1627 log(LOG_ERR, "pid %d (%s), uid %d on %s: %s\n", p ? p->p_pid : -1,
1628 p ? p->p_comm : "-", uid, fs->fs_fsmnt, cp);
1629}
| 490 }
491 }
492 if (ssize < len)
493 if (doasyncfree)
494 bdwrite(ebp);
495 else
496 bwrite(ebp);
497 /*
498 * Last, free the old blocks and assign the new blocks to the buffers.
499 */
500#ifdef DEBUG
501 if (prtrealloc)
502 printf("\n\tnew:");
503#endif
504 for (blkno = newblk, i = 0; i < len; i++, blkno += fs->fs_frag) {
505 ffs_blkfree(ip, dbtofsb(fs, buflist->bs_children[i]->b_blkno),
506 fs->fs_bsize);
507 buflist->bs_children[i]->b_blkno = fsbtodb(fs, blkno);
508#ifdef DEBUG
509 if (!ffs_checkblk(ip,
510 dbtofsb(fs, buflist->bs_children[i]->b_blkno), fs->fs_bsize))
511 panic("ffs_reallocblks: unallocated block 3");
512 if (prtrealloc)
513 printf(" %d,", blkno);
514#endif
515 }
516#ifdef DEBUG
517 if (prtrealloc) {
518 prtrealloc--;
519 printf("\n");
520 }
521#endif
522 return (0);
523
524fail:
525 if (ssize < len)
526 brelse(ebp);
527 if (sbap != &ip->i_db[0])
528 brelse(sbp);
529 return (ENOSPC);
530#endif
531}
532
533/*
534 * Allocate an inode in the file system.
535 *
536 * If allocating a directory, use ffs_dirpref to select the inode.
537 * If allocating in a directory, the following hierarchy is followed:
538 * 1) allocate the preferred inode.
539 * 2) allocate an inode in the same cylinder group.
540 * 3) quadradically rehash into other cylinder groups, until an
541 * available inode is located.
542 * If no inode preference is given the following heirarchy is used
543 * to allocate an inode:
544 * 1) allocate an inode in cylinder group 0.
545 * 2) quadradically rehash into other cylinder groups, until an
546 * available inode is located.
547 */
548int
549ffs_valloc(pvp, mode, cred, vpp)
550 struct vnode *pvp;
551 int mode;
552 struct ucred *cred;
553 struct vnode **vpp;
554{
555 register struct inode *pip;
556 register struct fs *fs;
557 register struct inode *ip;
558 ino_t ino, ipref;
559 int cg, error;
560
561 *vpp = NULL;
562 pip = VTOI(pvp);
563 fs = pip->i_fs;
564 if (fs->fs_cstotal.cs_nifree == 0)
565 goto noinodes;
566
567 if ((mode & IFMT) == IFDIR)
568 ipref = ffs_dirpref(fs);
569 else
570 ipref = pip->i_number;
571 if (ipref >= fs->fs_ncg * fs->fs_ipg)
572 ipref = 0;
573 cg = ino_to_cg(fs, ipref);
574 ino = (ino_t)ffs_hashalloc(pip, cg, (long)ipref, mode,
575 (allocfcn_t *)ffs_nodealloccg);
576 if (ino == 0)
577 goto noinodes;
578 error = VFS_VGET(pvp->v_mount, ino, vpp);
579 if (error) {
580 UFS_VFREE(pvp, ino, mode);
581 return (error);
582 }
583 ip = VTOI(*vpp);
584 if (ip->i_mode) {
585 printf("mode = 0%o, inum = %ld, fs = %s\n",
586 ip->i_mode, ip->i_number, fs->fs_fsmnt);
587 panic("ffs_valloc: dup alloc");
588 }
589 if (ip->i_blocks) { /* XXX */
590 printf("free inode %s/%ld had %ld blocks\n",
591 fs->fs_fsmnt, ino, ip->i_blocks);
592 ip->i_blocks = 0;
593 }
594 ip->i_flags = 0;
595 /*
596 * Set up a new generation number for this inode.
597 */
598 if (ip->i_gen == 0 || ++(ip->i_gen) == 0)
599 ip->i_gen = random() / 2 + 1;
600 return (0);
601noinodes:
602 ffs_fserr(fs, cred->cr_uid, "out of inodes");
603 uprintf("\n%s: create/symlink failed, no inodes free\n", fs->fs_fsmnt);
604 return (ENOSPC);
605}
606
607/*
608 * Find a cylinder to place a directory.
609 *
610 * The policy implemented by this algorithm is to select from
611 * among those cylinder groups with above the average number of
612 * free inodes, the one with the smallest number of directories.
613 */
614static ino_t
615ffs_dirpref(fs)
616 register struct fs *fs;
617{
618 int cg, minndir, mincg, avgifree;
619
620 avgifree = fs->fs_cstotal.cs_nifree / fs->fs_ncg;
621 minndir = fs->fs_ipg;
622 mincg = 0;
623 for (cg = 0; cg < fs->fs_ncg; cg++)
624 if (fs->fs_cs(fs, cg).cs_ndir < minndir &&
625 fs->fs_cs(fs, cg).cs_nifree >= avgifree) {
626 mincg = cg;
627 minndir = fs->fs_cs(fs, cg).cs_ndir;
628 }
629 return ((ino_t)(fs->fs_ipg * mincg));
630}
631
632/*
633 * Select the desired position for the next block in a file. The file is
634 * logically divided into sections. The first section is composed of the
635 * direct blocks. Each additional section contains fs_maxbpg blocks.
636 *
637 * If no blocks have been allocated in the first section, the policy is to
638 * request a block in the same cylinder group as the inode that describes
639 * the file. If no blocks have been allocated in any other section, the
640 * policy is to place the section in a cylinder group with a greater than
641 * average number of free blocks. An appropriate cylinder group is found
642 * by using a rotor that sweeps the cylinder groups. When a new group of
643 * blocks is needed, the sweep begins in the cylinder group following the
644 * cylinder group from which the previous allocation was made. The sweep
645 * continues until a cylinder group with greater than the average number
646 * of free blocks is found. If the allocation is for the first block in an
647 * indirect block, the information on the previous allocation is unavailable;
648 * here a best guess is made based upon the logical block number being
649 * allocated.
650 *
651 * If a section is already partially allocated, the policy is to
652 * contiguously allocate fs_maxcontig blocks. The end of one of these
653 * contiguous blocks and the beginning of the next is physically separated
654 * so that the disk head will be in transit between them for at least
655 * fs_rotdelay milliseconds. This is to allow time for the processor to
656 * schedule another I/O transfer.
657 */
658ufs_daddr_t
659ffs_blkpref(ip, lbn, indx, bap)
660 struct inode *ip;
661 ufs_daddr_t lbn;
662 int indx;
663 ufs_daddr_t *bap;
664{
665 register struct fs *fs;
666 register int cg;
667 int avgbfree, startcg;
668 ufs_daddr_t nextblk;
669
670 fs = ip->i_fs;
671 if (indx % fs->fs_maxbpg == 0 || bap[indx - 1] == 0) {
672 if (lbn < NDADDR) {
673 cg = ino_to_cg(fs, ip->i_number);
674 return (fs->fs_fpg * cg + fs->fs_frag);
675 }
676 /*
677 * Find a cylinder with greater than average number of
678 * unused data blocks.
679 */
680 if (indx == 0 || bap[indx - 1] == 0)
681 startcg =
682 ino_to_cg(fs, ip->i_number) + lbn / fs->fs_maxbpg;
683 else
684 startcg = dtog(fs, bap[indx - 1]) + 1;
685 startcg %= fs->fs_ncg;
686 avgbfree = fs->fs_cstotal.cs_nbfree / fs->fs_ncg;
687 for (cg = startcg; cg < fs->fs_ncg; cg++)
688 if (fs->fs_cs(fs, cg).cs_nbfree >= avgbfree) {
689 fs->fs_cgrotor = cg;
690 return (fs->fs_fpg * cg + fs->fs_frag);
691 }
692 for (cg = 0; cg <= startcg; cg++)
693 if (fs->fs_cs(fs, cg).cs_nbfree >= avgbfree) {
694 fs->fs_cgrotor = cg;
695 return (fs->fs_fpg * cg + fs->fs_frag);
696 }
697 return (0);
698 }
699 /*
700 * One or more previous blocks have been laid out. If less
701 * than fs_maxcontig previous blocks are contiguous, the
702 * next block is requested contiguously, otherwise it is
703 * requested rotationally delayed by fs_rotdelay milliseconds.
704 */
705 nextblk = bap[indx - 1] + fs->fs_frag;
706 if (fs->fs_rotdelay == 0 || indx < fs->fs_maxcontig ||
707 bap[indx - fs->fs_maxcontig] +
708 blkstofrags(fs, fs->fs_maxcontig) != nextblk)
709 return (nextblk);
710 /*
711 * Here we convert ms of delay to frags as:
712 * (frags) = (ms) * (rev/sec) * (sect/rev) /
713 * ((sect/frag) * (ms/sec))
714 * then round up to the next block.
715 */
716 nextblk += roundup(fs->fs_rotdelay * fs->fs_rps * fs->fs_nsect /
717 (NSPF(fs) * 1000), fs->fs_frag);
718 return (nextblk);
719}
720
721/*
722 * Implement the cylinder overflow algorithm.
723 *
724 * The policy implemented by this algorithm is:
725 * 1) allocate the block in its requested cylinder group.
726 * 2) quadradically rehash on the cylinder group number.
727 * 3) brute force search for a free block.
728 */
729/*VARARGS5*/
730static u_long
731ffs_hashalloc(ip, cg, pref, size, allocator)
732 struct inode *ip;
733 int cg;
734 long pref;
735 int size; /* size for data blocks, mode for inodes */
736 allocfcn_t *allocator;
737{
738 register struct fs *fs;
739 long result; /* XXX why not same type as we return? */
740 int i, icg = cg;
741
742 fs = ip->i_fs;
743 /*
744 * 1: preferred cylinder group
745 */
746 result = (*allocator)(ip, cg, pref, size);
747 if (result)
748 return (result);
749 /*
750 * 2: quadratic rehash
751 */
752 for (i = 1; i < fs->fs_ncg; i *= 2) {
753 cg += i;
754 if (cg >= fs->fs_ncg)
755 cg -= fs->fs_ncg;
756 result = (*allocator)(ip, cg, 0, size);
757 if (result)
758 return (result);
759 }
760 /*
761 * 3: brute force search
762 * Note that we start at i == 2, since 0 was checked initially,
763 * and 1 is always checked in the quadratic rehash.
764 */
765 cg = (icg + 2) % fs->fs_ncg;
766 for (i = 2; i < fs->fs_ncg; i++) {
767 result = (*allocator)(ip, cg, 0, size);
768 if (result)
769 return (result);
770 cg++;
771 if (cg == fs->fs_ncg)
772 cg = 0;
773 }
774 return (0);
775}
776
777/*
778 * Determine whether a fragment can be extended.
779 *
780 * Check to see if the necessary fragments are available, and
781 * if they are, allocate them.
782 */
783static ufs_daddr_t
784ffs_fragextend(ip, cg, bprev, osize, nsize)
785 struct inode *ip;
786 int cg;
787 long bprev;
788 int osize, nsize;
789{
790 register struct fs *fs;
791 register struct cg *cgp;
792 struct buf *bp;
793 long bno;
794 int frags, bbase;
795 int i, error;
796
797 fs = ip->i_fs;
798 if (fs->fs_cs(fs, cg).cs_nffree < numfrags(fs, nsize - osize))
799 return (0);
800 frags = numfrags(fs, nsize);
801 bbase = fragnum(fs, bprev);
802 if (bbase > fragnum(fs, (bprev + frags - 1))) {
803 /* cannot extend across a block boundary */
804 return (0);
805 }
806 error = bread(ip->i_devvp, fsbtodb(fs, cgtod(fs, cg)),
807 (int)fs->fs_cgsize, NOCRED, &bp);
808 if (error) {
809 brelse(bp);
810 return (0);
811 }
812 cgp = (struct cg *)bp->b_data;
813 if (!cg_chkmagic(cgp)) {
814 brelse(bp);
815 return (0);
816 }
817 cgp->cg_time = time.tv_sec;
818 bno = dtogd(fs, bprev);
819 for (i = numfrags(fs, osize); i < frags; i++)
820 if (isclr(cg_blksfree(cgp), bno + i)) {
821 brelse(bp);
822 return (0);
823 }
824 /*
825 * the current fragment can be extended
826 * deduct the count on fragment being extended into
827 * increase the count on the remaining fragment (if any)
828 * allocate the extended piece
829 */
830 for (i = frags; i < fs->fs_frag - bbase; i++)
831 if (isclr(cg_blksfree(cgp), bno + i))
832 break;
833 cgp->cg_frsum[i - numfrags(fs, osize)]--;
834 if (i != frags)
835 cgp->cg_frsum[i - frags]++;
836 for (i = numfrags(fs, osize); i < frags; i++) {
837 clrbit(cg_blksfree(cgp), bno + i);
838 cgp->cg_cs.cs_nffree--;
839 fs->fs_cstotal.cs_nffree--;
840 fs->fs_cs(fs, cg).cs_nffree--;
841 }
842 fs->fs_fmod = 1;
843 bdwrite(bp);
844 return (bprev);
845}
846
847/*
848 * Determine whether a block can be allocated.
849 *
850 * Check to see if a block of the appropriate size is available,
851 * and if it is, allocate it.
852 */
853static ufs_daddr_t
854ffs_alloccg(ip, cg, bpref, size)
855 struct inode *ip;
856 int cg;
857 ufs_daddr_t bpref;
858 int size;
859{
860 register struct fs *fs;
861 register struct cg *cgp;
862 struct buf *bp;
863 register int i;
864 int error, bno, frags, allocsiz;
865
866 fs = ip->i_fs;
867 if (fs->fs_cs(fs, cg).cs_nbfree == 0 && size == fs->fs_bsize)
868 return (0);
869 error = bread(ip->i_devvp, fsbtodb(fs, cgtod(fs, cg)),
870 (int)fs->fs_cgsize, NOCRED, &bp);
871 if (error) {
872 brelse(bp);
873 return (0);
874 }
875 cgp = (struct cg *)bp->b_data;
876 if (!cg_chkmagic(cgp) ||
877 (cgp->cg_cs.cs_nbfree == 0 && size == fs->fs_bsize)) {
878 brelse(bp);
879 return (0);
880 }
881 cgp->cg_time = time.tv_sec;
882 if (size == fs->fs_bsize) {
883 bno = ffs_alloccgblk(fs, cgp, bpref);
884 bdwrite(bp);
885 return (bno);
886 }
887 /*
888 * check to see if any fragments are already available
889 * allocsiz is the size which will be allocated, hacking
890 * it down to a smaller size if necessary
891 */
892 frags = numfrags(fs, size);
893 for (allocsiz = frags; allocsiz < fs->fs_frag; allocsiz++)
894 if (cgp->cg_frsum[allocsiz] != 0)
895 break;
896 if (allocsiz == fs->fs_frag) {
897 /*
898 * no fragments were available, so a block will be
899 * allocated, and hacked up
900 */
901 if (cgp->cg_cs.cs_nbfree == 0) {
902 brelse(bp);
903 return (0);
904 }
905 bno = ffs_alloccgblk(fs, cgp, bpref);
906 bpref = dtogd(fs, bno);
907 for (i = frags; i < fs->fs_frag; i++)
908 setbit(cg_blksfree(cgp), bpref + i);
909 i = fs->fs_frag - frags;
910 cgp->cg_cs.cs_nffree += i;
911 fs->fs_cstotal.cs_nffree += i;
912 fs->fs_cs(fs, cg).cs_nffree += i;
913 fs->fs_fmod = 1;
914 cgp->cg_frsum[i]++;
915 bdwrite(bp);
916 return (bno);
917 }
918 bno = ffs_mapsearch(fs, cgp, bpref, allocsiz);
919 if (bno < 0) {
920 brelse(bp);
921 return (0);
922 }
923 for (i = 0; i < frags; i++)
924 clrbit(cg_blksfree(cgp), bno + i);
925 cgp->cg_cs.cs_nffree -= frags;
926 fs->fs_cstotal.cs_nffree -= frags;
927 fs->fs_cs(fs, cg).cs_nffree -= frags;
928 fs->fs_fmod = 1;
929 cgp->cg_frsum[allocsiz]--;
930 if (frags != allocsiz)
931 cgp->cg_frsum[allocsiz - frags]++;
932 bdwrite(bp);
933 return (cg * fs->fs_fpg + bno);
934}
935
936/*
937 * Allocate a block in a cylinder group.
938 *
939 * This algorithm implements the following policy:
940 * 1) allocate the requested block.
941 * 2) allocate a rotationally optimal block in the same cylinder.
942 * 3) allocate the next available block on the block rotor for the
943 * specified cylinder group.
944 * Note that this routine only allocates fs_bsize blocks; these
945 * blocks may be fragmented by the routine that allocates them.
946 */
947static ufs_daddr_t
948ffs_alloccgblk(fs, cgp, bpref)
949 register struct fs *fs;
950 register struct cg *cgp;
951 ufs_daddr_t bpref;
952{
953 ufs_daddr_t bno, blkno;
954 int cylno, pos, delta;
955 short *cylbp;
956 register int i;
957
958 if (bpref == 0 || dtog(fs, bpref) != cgp->cg_cgx) {
959 bpref = cgp->cg_rotor;
960 goto norot;
961 }
962 bpref = blknum(fs, bpref);
963 bpref = dtogd(fs, bpref);
964 /*
965 * if the requested block is available, use it
966 */
967 if (ffs_isblock(fs, cg_blksfree(cgp), fragstoblks(fs, bpref))) {
968 bno = bpref;
969 goto gotit;
970 }
971 if (fs->fs_nrpos <= 1 || fs->fs_cpc == 0) {
972 /*
973 * Block layout information is not available.
974 * Leaving bpref unchanged means we take the
975 * next available free block following the one
976 * we just allocated. Hopefully this will at
977 * least hit a track cache on drives of unknown
978 * geometry (e.g. SCSI).
979 */
980 goto norot;
981 }
982 /*
983 * check for a block available on the same cylinder
984 */
985 cylno = cbtocylno(fs, bpref);
986 if (cg_blktot(cgp)[cylno] == 0)
987 goto norot;
988 /*
989 * check the summary information to see if a block is
990 * available in the requested cylinder starting at the
991 * requested rotational position and proceeding around.
992 */
993 cylbp = cg_blks(fs, cgp, cylno);
994 pos = cbtorpos(fs, bpref);
995 for (i = pos; i < fs->fs_nrpos; i++)
996 if (cylbp[i] > 0)
997 break;
998 if (i == fs->fs_nrpos)
999 for (i = 0; i < pos; i++)
1000 if (cylbp[i] > 0)
1001 break;
1002 if (cylbp[i] > 0) {
1003 /*
1004 * found a rotational position, now find the actual
1005 * block. A panic if none is actually there.
1006 */
1007 pos = cylno % fs->fs_cpc;
1008 bno = (cylno - pos) * fs->fs_spc / NSPB(fs);
1009 if (fs_postbl(fs, pos)[i] == -1) {
1010 printf("pos = %d, i = %d, fs = %s\n",
1011 pos, i, fs->fs_fsmnt);
1012 panic("ffs_alloccgblk: cyl groups corrupted");
1013 }
1014 for (i = fs_postbl(fs, pos)[i];; ) {
1015 if (ffs_isblock(fs, cg_blksfree(cgp), bno + i)) {
1016 bno = blkstofrags(fs, (bno + i));
1017 goto gotit;
1018 }
1019 delta = fs_rotbl(fs)[i];
1020 if (delta <= 0 ||
1021 delta + i > fragstoblks(fs, fs->fs_fpg))
1022 break;
1023 i += delta;
1024 }
1025 printf("pos = %d, i = %d, fs = %s\n", pos, i, fs->fs_fsmnt);
1026 panic("ffs_alloccgblk: can't find blk in cyl");
1027 }
1028norot:
1029 /*
1030 * no blocks in the requested cylinder, so take next
1031 * available one in this cylinder group.
1032 */
1033 bno = ffs_mapsearch(fs, cgp, bpref, (int)fs->fs_frag);
1034 if (bno < 0)
1035 return (0);
1036 cgp->cg_rotor = bno;
1037gotit:
1038 blkno = fragstoblks(fs, bno);
1039 ffs_clrblock(fs, cg_blksfree(cgp), (long)blkno);
1040 ffs_clusteracct(fs, cgp, blkno, -1);
1041 cgp->cg_cs.cs_nbfree--;
1042 fs->fs_cstotal.cs_nbfree--;
1043 fs->fs_cs(fs, cgp->cg_cgx).cs_nbfree--;
1044 cylno = cbtocylno(fs, bno);
1045 cg_blks(fs, cgp, cylno)[cbtorpos(fs, bno)]--;
1046 cg_blktot(cgp)[cylno]--;
1047 fs->fs_fmod = 1;
1048 return (cgp->cg_cgx * fs->fs_fpg + bno);
1049}
1050
1051#ifdef notyet
1052/*
1053 * Determine whether a cluster can be allocated.
1054 *
1055 * We do not currently check for optimal rotational layout if there
1056 * are multiple choices in the same cylinder group. Instead we just
1057 * take the first one that we find following bpref.
1058 */
1059static ufs_daddr_t
1060ffs_clusteralloc(ip, cg, bpref, len)
1061 struct inode *ip;
1062 int cg;
1063 ufs_daddr_t bpref;
1064 int len;
1065{
1066 register struct fs *fs;
1067 register struct cg *cgp;
1068 struct buf *bp;
1069 int i, got, run, bno, bit, map;
1070 u_char *mapp;
1071 int32_t *lp;
1072
1073 fs = ip->i_fs;
1074 if (fs->fs_maxcluster[cg] < len)
1075 return (NULL);
1076 if (bread(ip->i_devvp, fsbtodb(fs, cgtod(fs, cg)), (int)fs->fs_cgsize,
1077 NOCRED, &bp))
1078 goto fail;
1079 cgp = (struct cg *)bp->b_data;
1080 if (!cg_chkmagic(cgp))
1081 goto fail;
1082 /*
1083 * Check to see if a cluster of the needed size (or bigger) is
1084 * available in this cylinder group.
1085 */
1086 lp = &cg_clustersum(cgp)[len];
1087 for (i = len; i <= fs->fs_contigsumsize; i++)
1088 if (*lp++ > 0)
1089 break;
1090 if (i > fs->fs_contigsumsize) {
1091 /*
1092 * This is the first time looking for a cluster in this
1093 * cylinder group. Update the cluster summary information
1094 * to reflect the true maximum sized cluster so that
1095 * future cluster allocation requests can avoid reading
1096 * the cylinder group map only to find no clusters.
1097 */
1098 lp = &cg_clustersum(cgp)[len - 1];
1099 for (i = len - 1; i > 0; i--)
1100 if (*lp-- > 0)
1101 break;
1102 fs->fs_maxcluster[cg] = i;
1103 goto fail;
1104 }
1105 /*
1106 * Search the cluster map to find a big enough cluster.
1107 * We take the first one that we find, even if it is larger
1108 * than we need as we prefer to get one close to the previous
1109 * block allocation. We do not search before the current
1110 * preference point as we do not want to allocate a block
1111 * that is allocated before the previous one (as we will
1112 * then have to wait for another pass of the elevator
1113 * algorithm before it will be read). We prefer to fail and
1114 * be recalled to try an allocation in the next cylinder group.
1115 */
1116 if (dtog(fs, bpref) != cg)
1117 bpref = 0;
1118 else
1119 bpref = fragstoblks(fs, dtogd(fs, blknum(fs, bpref)));
1120 mapp = &cg_clustersfree(cgp)[bpref / NBBY];
1121 map = *mapp++;
1122 bit = 1 << (bpref % NBBY);
1123 for (run = 0, got = bpref; got < cgp->cg_nclusterblks; got++) {
1124 if ((map & bit) == 0) {
1125 run = 0;
1126 } else {
1127 run++;
1128 if (run == len)
1129 break;
1130 }
1131 if ((got & (NBBY - 1)) != (NBBY - 1)) {
1132 bit <<= 1;
1133 } else {
1134 map = *mapp++;
1135 bit = 1;
1136 }
1137 }
1138 if (got >= cgp->cg_nclusterblks)
1139 goto fail;
1140 /*
1141 * Allocate the cluster that we have found.
1142 */
1143 for (i = 1; i <= len; i++)
1144 if (!ffs_isblock(fs, cg_blksfree(cgp), got - run + i))
1145 panic("ffs_clusteralloc: map mismatch");
1146 bno = cg * fs->fs_fpg + blkstofrags(fs, got - run + 1);
1147 if (dtog(fs, bno) != cg)
1148 panic("ffs_clusteralloc: allocated out of group");
1149 len = blkstofrags(fs, len);
1150 for (i = 0; i < len; i += fs->fs_frag)
1151 if ((got = ffs_alloccgblk(fs, cgp, bno + i)) != bno + i)
1152 panic("ffs_clusteralloc: lost block");
1153 bdwrite(bp);
1154 return (bno);
1155
1156fail:
1157 brelse(bp);
1158 return (0);
1159}
1160#endif
1161
1162/*
1163 * Determine whether an inode can be allocated.
1164 *
1165 * Check to see if an inode is available, and if it is,
1166 * allocate it using the following policy:
1167 * 1) allocate the requested inode.
1168 * 2) allocate the next available inode after the requested
1169 * inode in the specified cylinder group.
1170 */
1171static ino_t
1172ffs_nodealloccg(ip, cg, ipref, mode)
1173 struct inode *ip;
1174 int cg;
1175 ufs_daddr_t ipref;
1176 int mode;
1177{
1178 register struct fs *fs;
1179 register struct cg *cgp;
1180 struct buf *bp;
1181 int error, start, len, loc, map, i;
1182
1183 fs = ip->i_fs;
1184 if (fs->fs_cs(fs, cg).cs_nifree == 0)
1185 return (0);
1186 error = bread(ip->i_devvp, fsbtodb(fs, cgtod(fs, cg)),
1187 (int)fs->fs_cgsize, NOCRED, &bp);
1188 if (error) {
1189 brelse(bp);
1190 return (0);
1191 }
1192 cgp = (struct cg *)bp->b_data;
1193 if (!cg_chkmagic(cgp) || cgp->cg_cs.cs_nifree == 0) {
1194 brelse(bp);
1195 return (0);
1196 }
1197 cgp->cg_time = time.tv_sec;
1198 if (ipref) {
1199 ipref %= fs->fs_ipg;
1200 if (isclr(cg_inosused(cgp), ipref))
1201 goto gotit;
1202 }
1203 start = cgp->cg_irotor / NBBY;
1204 len = howmany(fs->fs_ipg - cgp->cg_irotor, NBBY);
1205 loc = skpc(0xff, len, &cg_inosused(cgp)[start]);
1206 if (loc == 0) {
1207 len = start + 1;
1208 start = 0;
1209 loc = skpc(0xff, len, &cg_inosused(cgp)[0]);
1210 if (loc == 0) {
1211 printf("cg = %d, irotor = %ld, fs = %s\n",
1212 cg, cgp->cg_irotor, fs->fs_fsmnt);
1213 panic("ffs_nodealloccg: map corrupted");
1214 /* NOTREACHED */
1215 }
1216 }
1217 i = start + len - loc;
1218 map = cg_inosused(cgp)[i];
1219 ipref = i * NBBY;
1220 for (i = 1; i < (1 << NBBY); i <<= 1, ipref++) {
1221 if ((map & i) == 0) {
1222 cgp->cg_irotor = ipref;
1223 goto gotit;
1224 }
1225 }
1226 printf("fs = %s\n", fs->fs_fsmnt);
1227 panic("ffs_nodealloccg: block not in map");
1228 /* NOTREACHED */
1229gotit:
1230 setbit(cg_inosused(cgp), ipref);
1231 cgp->cg_cs.cs_nifree--;
1232 fs->fs_cstotal.cs_nifree--;
1233 fs->fs_cs(fs, cg).cs_nifree--;
1234 fs->fs_fmod = 1;
1235 if ((mode & IFMT) == IFDIR) {
1236 cgp->cg_cs.cs_ndir++;
1237 fs->fs_cstotal.cs_ndir++;
1238 fs->fs_cs(fs, cg).cs_ndir++;
1239 }
1240 bdwrite(bp);
1241 return (cg * fs->fs_ipg + ipref);
1242}
1243
1244/*
1245 * Free a block or fragment.
1246 *
1247 * The specified block or fragment is placed back in the
1248 * free map. If a fragment is deallocated, a possible
1249 * block reassembly is checked.
1250 */
1251void
1252ffs_blkfree(ip, bno, size)
1253 register struct inode *ip;
1254 ufs_daddr_t bno;
1255 long size;
1256{
1257 register struct fs *fs;
1258 register struct cg *cgp;
1259 struct buf *bp;
1260 ufs_daddr_t blkno;
1261 int i, error, cg, blk, frags, bbase;
1262
1263 fs = ip->i_fs;
1264 if ((u_int)size > fs->fs_bsize || fragoff(fs, size) != 0) {
1265 printf("dev = 0x%lx, bsize = %ld, size = %ld, fs = %s\n",
1266 (u_long)ip->i_dev, fs->fs_bsize, size, fs->fs_fsmnt);
1267 panic("ffs_blkfree: bad size");
1268 }
1269 cg = dtog(fs, bno);
1270 if ((u_int)bno >= fs->fs_size) {
1271 printf("bad block %ld, ino %ld\n", bno, ip->i_number);
1272 ffs_fserr(fs, ip->i_uid, "bad block");
1273 return;
1274 }
1275 error = bread(ip->i_devvp, fsbtodb(fs, cgtod(fs, cg)),
1276 (int)fs->fs_cgsize, NOCRED, &bp);
1277 if (error) {
1278 brelse(bp);
1279 return;
1280 }
1281 cgp = (struct cg *)bp->b_data;
1282 if (!cg_chkmagic(cgp)) {
1283 brelse(bp);
1284 return;
1285 }
1286 cgp->cg_time = time.tv_sec;
1287 bno = dtogd(fs, bno);
1288 if (size == fs->fs_bsize) {
1289 blkno = fragstoblks(fs, bno);
1290 if (ffs_isblock(fs, cg_blksfree(cgp), blkno)) {
1291 printf("dev = 0x%lx, block = %ld, fs = %s\n",
1292 (u_long) ip->i_dev, bno, fs->fs_fsmnt);
1293 panic("ffs_blkfree: freeing free block");
1294 }
1295 ffs_setblock(fs, cg_blksfree(cgp), blkno);
1296 ffs_clusteracct(fs, cgp, blkno, 1);
1297 cgp->cg_cs.cs_nbfree++;
1298 fs->fs_cstotal.cs_nbfree++;
1299 fs->fs_cs(fs, cg).cs_nbfree++;
1300 i = cbtocylno(fs, bno);
1301 cg_blks(fs, cgp, i)[cbtorpos(fs, bno)]++;
1302 cg_blktot(cgp)[i]++;
1303 } else {
1304 bbase = bno - fragnum(fs, bno);
1305 /*
1306 * decrement the counts associated with the old frags
1307 */
1308 blk = blkmap(fs, cg_blksfree(cgp), bbase);
1309 ffs_fragacct(fs, blk, cgp->cg_frsum, -1);
1310 /*
1311 * deallocate the fragment
1312 */
1313 frags = numfrags(fs, size);
1314 for (i = 0; i < frags; i++) {
1315 if (isset(cg_blksfree(cgp), bno + i)) {
1316 printf("dev = 0x%lx, block = %ld, fs = %s\n",
1317 (u_long) ip->i_dev, bno + i, fs->fs_fsmnt);
1318 panic("ffs_blkfree: freeing free frag");
1319 }
1320 setbit(cg_blksfree(cgp), bno + i);
1321 }
1322 cgp->cg_cs.cs_nffree += i;
1323 fs->fs_cstotal.cs_nffree += i;
1324 fs->fs_cs(fs, cg).cs_nffree += i;
1325 /*
1326 * add back in counts associated with the new frags
1327 */
1328 blk = blkmap(fs, cg_blksfree(cgp), bbase);
1329 ffs_fragacct(fs, blk, cgp->cg_frsum, 1);
1330 /*
1331 * if a complete block has been reassembled, account for it
1332 */
1333 blkno = fragstoblks(fs, bbase);
1334 if (ffs_isblock(fs, cg_blksfree(cgp), blkno)) {
1335 cgp->cg_cs.cs_nffree -= fs->fs_frag;
1336 fs->fs_cstotal.cs_nffree -= fs->fs_frag;
1337 fs->fs_cs(fs, cg).cs_nffree -= fs->fs_frag;
1338 ffs_clusteracct(fs, cgp, blkno, 1);
1339 cgp->cg_cs.cs_nbfree++;
1340 fs->fs_cstotal.cs_nbfree++;
1341 fs->fs_cs(fs, cg).cs_nbfree++;
1342 i = cbtocylno(fs, bbase);
1343 cg_blks(fs, cgp, i)[cbtorpos(fs, bbase)]++;
1344 cg_blktot(cgp)[i]++;
1345 }
1346 }
1347 fs->fs_fmod = 1;
1348 bdwrite(bp);
1349}
1350
1351#ifdef DIAGNOSTIC
1352/*
1353 * Verify allocation of a block or fragment. Returns true if block or
1354 * fragment is allocated, false if it is free.
1355 */
1356int
1357ffs_checkblk(ip, bno, size)
1358 struct inode *ip;
1359 ufs_daddr_t bno;
1360 long size;
1361{
1362 struct fs *fs;
1363 struct cg *cgp;
1364 struct buf *bp;
1365 int i, error, frags, free;
1366
1367 fs = ip->i_fs;
1368 if ((u_int)size > fs->fs_bsize || fragoff(fs, size) != 0) {
1369 printf("bsize = %d, size = %d, fs = %s\n",
1370 fs->fs_bsize, size, fs->fs_fsmnt);
1371 panic("ffs_checkblk: bad size");
1372 }
1373 if ((u_int)bno >= fs->fs_size)
1374 panic("ffs_checkblk: bad block %d", bno);
1375 error = bread(ip->i_devvp, fsbtodb(fs, cgtod(fs, dtog(fs, bno))),
1376 (int)fs->fs_cgsize, NOCRED, &bp);
1377 if (error)
1378 panic("ffs_checkblk: cg bread failed");
1379 cgp = (struct cg *)bp->b_data;
1380 if (!cg_chkmagic(cgp))
1381 panic("ffs_checkblk: cg magic mismatch");
1382 bno = dtogd(fs, bno);
1383 if (size == fs->fs_bsize) {
1384 free = ffs_isblock(fs, cg_blksfree(cgp), fragstoblks(fs, bno));
1385 } else {
1386 frags = numfrags(fs, size);
1387 for (free = 0, i = 0; i < frags; i++)
1388 if (isset(cg_blksfree(cgp), bno + i))
1389 free++;
1390 if (free != 0 && free != frags)
1391 panic("ffs_checkblk: partially free fragment");
1392 }
1393 brelse(bp);
1394 return (!free);
1395}
1396#endif /* DIAGNOSTIC */
1397
1398/*
1399 * Free an inode.
1400 *
1401 * The specified inode is placed back in the free map.
1402 */
1403int
1404ffs_vfree(pvp, ino, mode)
1405 struct vnode *pvp;
1406 ino_t ino;
1407 int mode;
1408{
1409 register struct fs *fs;
1410 register struct cg *cgp;
1411 register struct inode *pip;
1412 struct buf *bp;
1413 int error, cg;
1414
1415 pip = VTOI(pvp);
1416 fs = pip->i_fs;
1417 if ((u_int)ino >= fs->fs_ipg * fs->fs_ncg)
1418 panic("ffs_vfree: range: dev = 0x%x, ino = %d, fs = %s",
1419 pip->i_dev, ino, fs->fs_fsmnt);
1420 cg = ino_to_cg(fs, ino);
1421 error = bread(pip->i_devvp, fsbtodb(fs, cgtod(fs, cg)),
1422 (int)fs->fs_cgsize, NOCRED, &bp);
1423 if (error) {
1424 brelse(bp);
1425 return (0);
1426 }
1427 cgp = (struct cg *)bp->b_data;
1428 if (!cg_chkmagic(cgp)) {
1429 brelse(bp);
1430 return (0);
1431 }
1432 cgp->cg_time = time.tv_sec;
1433 ino %= fs->fs_ipg;
1434 if (isclr(cg_inosused(cgp), ino)) {
1435 printf("dev = 0x%lx, ino = %ld, fs = %s\n",
1436 (u_long)pip->i_dev, ino, fs->fs_fsmnt);
1437 if (fs->fs_ronly == 0)
1438 panic("ffs_vfree: freeing free inode");
1439 }
1440 clrbit(cg_inosused(cgp), ino);
1441 if (ino < cgp->cg_irotor)
1442 cgp->cg_irotor = ino;
1443 cgp->cg_cs.cs_nifree++;
1444 fs->fs_cstotal.cs_nifree++;
1445 fs->fs_cs(fs, cg).cs_nifree++;
1446 if ((mode & IFMT) == IFDIR) {
1447 cgp->cg_cs.cs_ndir--;
1448 fs->fs_cstotal.cs_ndir--;
1449 fs->fs_cs(fs, cg).cs_ndir--;
1450 }
1451 fs->fs_fmod = 1;
1452 bdwrite(bp);
1453 return (0);
1454}
1455
1456/*
1457 * Find a block of the specified size in the specified cylinder group.
1458 *
1459 * It is a panic if a request is made to find a block if none are
1460 * available.
1461 */
1462static ufs_daddr_t
1463ffs_mapsearch(fs, cgp, bpref, allocsiz)
1464 register struct fs *fs;
1465 register struct cg *cgp;
1466 ufs_daddr_t bpref;
1467 int allocsiz;
1468{
1469 ufs_daddr_t bno;
1470 int start, len, loc, i;
1471 int blk, field, subfield, pos;
1472
1473 /*
1474 * find the fragment by searching through the free block
1475 * map for an appropriate bit pattern
1476 */
1477 if (bpref)
1478 start = dtogd(fs, bpref) / NBBY;
1479 else
1480 start = cgp->cg_frotor / NBBY;
1481 len = howmany(fs->fs_fpg, NBBY) - start;
1482 loc = scanc((u_int)len, (u_char *)&cg_blksfree(cgp)[start],
1483 (u_char *)fragtbl[fs->fs_frag],
1484 (u_char)(1 << (allocsiz - 1 + (fs->fs_frag % NBBY))));
1485 if (loc == 0) {
1486 len = start + 1;
1487 start = 0;
1488 loc = scanc((u_int)len, (u_char *)&cg_blksfree(cgp)[0],
1489 (u_char *)fragtbl[fs->fs_frag],
1490 (u_char)(1 << (allocsiz - 1 + (fs->fs_frag % NBBY))));
1491 if (loc == 0) {
1492 printf("start = %d, len = %d, fs = %s\n",
1493 start, len, fs->fs_fsmnt);
1494 panic("ffs_alloccg: map corrupted");
1495 /* NOTREACHED */
1496 }
1497 }
1498 bno = (start + len - loc) * NBBY;
1499 cgp->cg_frotor = bno;
1500 /*
1501 * found the byte in the map
1502 * sift through the bits to find the selected frag
1503 */
1504 for (i = bno + NBBY; bno < i; bno += fs->fs_frag) {
1505 blk = blkmap(fs, cg_blksfree(cgp), bno);
1506 blk <<= 1;
1507 field = around[allocsiz];
1508 subfield = inside[allocsiz];
1509 for (pos = 0; pos <= fs->fs_frag - allocsiz; pos++) {
1510 if ((blk & field) == subfield)
1511 return (bno + pos);
1512 field <<= 1;
1513 subfield <<= 1;
1514 }
1515 }
1516 printf("bno = %lu, fs = %s\n", (u_long)bno, fs->fs_fsmnt);
1517 panic("ffs_alloccg: block not in map");
1518 return (-1);
1519}
1520
1521/*
1522 * Update the cluster map because of an allocation or free.
1523 *
1524 * Cnt == 1 means free; cnt == -1 means allocating.
1525 */
1526static void
1527ffs_clusteracct(fs, cgp, blkno, cnt)
1528 struct fs *fs;
1529 struct cg *cgp;
1530 ufs_daddr_t blkno;
1531 int cnt;
1532{
1533 int32_t *sump;
1534 int32_t *lp;
1535 u_char *freemapp, *mapp;
1536 int i, start, end, forw, back, map, bit;
1537
1538 if (fs->fs_contigsumsize <= 0)
1539 return;
1540 freemapp = cg_clustersfree(cgp);
1541 sump = cg_clustersum(cgp);
1542 /*
1543 * Allocate or clear the actual block.
1544 */
1545 if (cnt > 0)
1546 setbit(freemapp, blkno);
1547 else
1548 clrbit(freemapp, blkno);
1549 /*
1550 * Find the size of the cluster going forward.
1551 */
1552 start = blkno + 1;
1553 end = start + fs->fs_contigsumsize;
1554 if (end >= cgp->cg_nclusterblks)
1555 end = cgp->cg_nclusterblks;
1556 mapp = &freemapp[start / NBBY];
1557 map = *mapp++;
1558 bit = 1 << (start % NBBY);
1559 for (i = start; i < end; i++) {
1560 if ((map & bit) == 0)
1561 break;
1562 if ((i & (NBBY - 1)) != (NBBY - 1)) {
1563 bit <<= 1;
1564 } else {
1565 map = *mapp++;
1566 bit = 1;
1567 }
1568 }
1569 forw = i - start;
1570 /*
1571 * Find the size of the cluster going backward.
1572 */
1573 start = blkno - 1;
1574 end = start - fs->fs_contigsumsize;
1575 if (end < 0)
1576 end = -1;
1577 mapp = &freemapp[start / NBBY];
1578 map = *mapp--;
1579 bit = 1 << (start % NBBY);
1580 for (i = start; i > end; i--) {
1581 if ((map & bit) == 0)
1582 break;
1583 if ((i & (NBBY - 1)) != 0) {
1584 bit >>= 1;
1585 } else {
1586 map = *mapp--;
1587 bit = 1 << (NBBY - 1);
1588 }
1589 }
1590 back = start - i;
1591 /*
1592 * Account for old cluster and the possibly new forward and
1593 * back clusters.
1594 */
1595 i = back + forw + 1;
1596 if (i > fs->fs_contigsumsize)
1597 i = fs->fs_contigsumsize;
1598 sump[i] += cnt;
1599 if (back > 0)
1600 sump[back] -= cnt;
1601 if (forw > 0)
1602 sump[forw] -= cnt;
1603 /*
1604 * Update cluster summary information.
1605 */
1606 lp = &sump[fs->fs_contigsumsize];
1607 for (i = fs->fs_contigsumsize; i > 0; i--)
1608 if (*lp-- > 0)
1609 break;
1610 fs->fs_maxcluster[cgp->cg_cgx] = i;
1611}
1612
1613/*
1614 * Fserr prints the name of a file system with an error diagnostic.
1615 *
1616 * The form of the error message is:
1617 * fs: error message
1618 */
1619static void
1620ffs_fserr(fs, uid, cp)
1621 struct fs *fs;
1622 u_int uid;
1623 char *cp;
1624{
1625 struct proc *p = curproc; /* XXX */
1626
1627 log(LOG_ERR, "pid %d (%s), uid %d on %s: %s\n", p ? p->p_pid : -1,
1628 p ? p->p_comm : "-", uid, fs->fs_fsmnt, cp);
1629}
|