2 * Copyright 2000, International Business Machines Corporation and others.
5 * This software has been released under the terms of the IBM Public
6 * License. For details, see the LICENSE file in the top-level source
7 * directory or online at http://www.openafs.org/dl/license10.html
16 /* Clone a volume. Assumes the new volume is already created */
18 #include <afsconfig.h>
19 #include <afs/param.h>
23 #include <sys/types.h>
25 #include <afs/afs_assert.h>
42 #include <afs/afsint.h>
46 #include <afs/afssyscalls.h>
50 #include "partition.h"
51 #include "viceinode.h"
52 #include "vol_prototypes.h"
55 int (*vol_PollProc) (void) = 0; /* someone must init this */
57 #define ERROR_EXIT(code) do { \
62 /* parameters for idec call - this could just be an IHandle_t, but leaving
63 * open the possibility of decrementing the special files as well.
70 #define CLONE_MAXITEMS 100
72 struct clone_items *next;
74 Inode data[CLONE_MAXITEMS];
78 struct clone_items *first;
79 struct clone_items *last;
82 void CloneVolume(Error *, Volume *, Volume *, Volume *);
85 ci_AddItem(struct clone_head *ah, Inode aino)
87 struct clone_items *ti;
89 /* if no last elt (first call) or last item full, get a new one */
90 if ((!ah->last) || ah->last->nitems >= CLONE_MAXITEMS) {
91 ti = (struct clone_items *)malloc(sizeof(struct clone_items));
93 Log("ci_AddItem: malloc failed\n");
94 osi_Panic("ci_AddItem: malloc failed\n");
97 ti->next = (struct clone_items *)0;
102 /* first dude in the list */
103 ah->first = ah->last = ti;
108 /* now ti points to the end of the list, to a clone_item with room
109 * for at least one more element. Add it.
111 ti->data[ti->nitems++] = aino;
115 /* initialize a clone header */
117 ci_InitHead(struct clone_head *ah)
119 memset(ah, 0, sizeof(*ah));
123 /* apply a function to all dudes in the set */
125 ci_Apply(struct clone_head *ah, int (*aproc) (Inode, void *), void *arock)
127 struct clone_items *ti;
130 for (ti = ah->first; ti; ti = ti->next) {
131 for (i = 0; i < ti->nitems; i++) {
132 (*aproc) (ti->data[i], arock);
138 /* free all dudes in the list */
140 ci_Destroy(struct clone_head *ah)
142 struct clone_items *ti, *ni;
144 for (ti = ah->first; ti; ti = ni) {
145 ni = ti->next; /* guard against freeing */
152 IDecProc(Inode adata, void *arock)
154 struct clone_rock *aparm = (struct clone_rock *)arock;
155 IH_DEC(aparm->h, adata, aparm->vol);
161 DoCloneIndex(Volume * rwvp, Volume * clvp, VnodeClass class, int reclone)
163 afs_int32 code, error = 0;
164 FdHandle_t *rwFd = 0, *clFdIn = 0, *clFdOut = 0;
165 StreamHandle_t *rwfile = 0, *clfilein = 0, *clfileout = 0;
166 IHandle_t *rwH = 0, *clHin = 0, *clHout = 0;
167 char buf[SIZEOF_LARGEDISKVNODE], dbuf[SIZEOF_LARGEDISKVNODE];
168 struct VnodeDiskObject *rwvnode = (struct VnodeDiskObject *)buf;
169 struct VnodeDiskObject *clvnode = (struct VnodeDiskObject *)dbuf;
172 struct clone_head decHead;
173 struct clone_rock decRock;
174 afs_foff_t offset = 0;
175 afs_int32 dircloned, inodeinced;
176 afs_int32 filecount = 0, diskused = 0;
179 struct VnodeClassInfo *vcp = &VnodeClassInfo[class];
180 int ReadWriteOriginal = VolumeWriteable(rwvp);
182 /* Correct number of files in volume: this assumes indexes are always
183 cloned starting with vLarge */
184 if (ReadWriteOriginal && class != vLarge) {
185 filecount = V_filecount(rwvp);
186 diskused = V_diskused(rwvp);
189 /* Open the RW volume's index file and seek to beginning */
190 IH_COPY(rwH, rwvp->vnodeIndex[class].handle);
194 rwfile = FDH_FDOPEN(rwFd, ReadWriteOriginal ? "r+" : "r");
197 STREAM_ASEEK(rwfile, vcp->diskSize); /* Will fail if no vnodes */
199 /* Open the clone volume's index file and seek to beginning */
200 IH_COPY(clHout, clvp->vnodeIndex[class].handle);
201 clFdOut = IH_OPEN(clHout);
204 clfileout = FDH_FDOPEN(clFdOut, "a");
207 code = STREAM_ASEEK(clfileout, vcp->diskSize);
211 /* If recloning, open the new volume's index; this time for
212 * reading. We never read anything that we're simultaneously
213 * writing, so this all works.
216 IH_COPY(clHin, clvp->vnodeIndex[class].handle);
217 clFdIn = IH_OPEN(clHin);
220 clfilein = FDH_FDOPEN(clFdIn, "r");
223 STREAM_ASEEK(clfilein, vcp->diskSize); /* Will fail if no vnodes */
226 /* Initialize list of inodes to nuke */
227 ci_InitHead(&decHead);
228 decRock.h = V_linkHandle(rwvp);
229 decRock.vol = V_parentId(rwvp);
231 /* Read each vnode in the old volume's index file */
232 for (offset = vcp->diskSize;
233 STREAM_READ(rwvnode, vcp->diskSize, 1, rwfile) == 1;
234 offset += vcp->diskSize) {
235 dircloned = inodeinced = 0;
237 /* If we are recloning the volume, read the corresponding vnode
238 * from the clone and determine its inode number.
240 if (reclone && !STREAM_EOF(clfilein)
241 && (STREAM_READ(clvnode, vcp->diskSize, 1, clfilein) == 1)) {
242 clinode = VNDISK_GET_INO(clvnode);
247 if (rwvnode->type != vNull) {
250 if (rwvnode->vnodeMagic != vcp->magic)
252 rwinode = VNDISK_GET_INO(rwvnode);
254 VNDISK_GET_LEN(ll, rwvnode);
255 diskused += nBlocks(ll);
257 /* Increment the inode if not already */
258 if (clinode && (clinode == rwinode)) {
259 clinode = 0; /* already cloned - don't delete later */
260 } else if (rwinode) {
261 if (IH_INC(V_linkHandle(rwvp), rwinode, V_parentId(rwvp)) ==
263 Log("IH_INC failed: %"AFS_PTR_FMT", %s, %u errno %d\n",
264 V_linkHandle(rwvp), PrintInode(stmp, rwinode),
265 V_parentId(rwvp), errno);
272 /* If a directory, mark vnode in old volume as cloned */
273 if ((rwvnode->type == vDirectory) && ReadWriteOriginal) {
276 * It is my firmly held belief that immediately after
277 * copy-on-write, the two directories can be identical.
278 * If the new copy is changed (presumably, that is the very
279 * next thing that will happen) then the dataVersion will
282 /* NOTE: the dataVersion++ is incredibly important!!!.
283 * This will cause the inode created by the file server
284 * on copy-on-write to be stamped with a dataVersion bigger
285 * than the current one. The salvager will then do the
287 rwvnode->dataVersion++;
290 code = STREAM_ASEEK(rwfile, offset);
293 code = STREAM_WRITE(rwvnode, vcp->diskSize, 1, rwfile);
297 code = STREAM_ASEEK(rwfile, offset + vcp->diskSize);
301 rwvnode->dataVersion--; /* Really needs to be set to the value in the inode,
302 * for the read-only volume */
307 /* Overwrite the vnode entry in the clone volume */
309 code = STREAM_WRITE(rwvnode, vcp->diskSize, 1, clfileout);
312 /* Couldn't clone, go back and decrement the inode's link count */
314 if (IH_DEC(V_linkHandle(rwvp), rwinode, V_parentId(rwvp)) ==
316 Log("IH_DEC failed: %"AFS_PTR_FMT", %s, %u errno %d\n",
317 V_linkHandle(rwvp), PrintInode(stmp, rwinode),
318 V_parentId(rwvp), errno);
323 /* And if the directory was marked clone, unmark it */
326 if (STREAM_ASEEK(rwfile, offset) != -1)
327 (void)STREAM_WRITE(rwvnode, vcp->diskSize, 1, rwfile);
332 /* Removal of the old cloned inode */
334 ci_AddItem(&decHead, clinode); /* just queue it */
339 if (STREAM_ERROR(clfileout))
342 /* Clean out any junk at end of clone file */
344 STREAM_ASEEK(clfilein, offset);
345 while (STREAM_READ(clvnode, vcp->diskSize, 1, clfilein) == 1) {
346 if (clvnode->type != vNull && VNDISK_GET_INO(clvnode) != 0) {
347 ci_AddItem(&decHead, VNDISK_GET_INO(clvnode));
353 /* come here to finish up. If code is non-zero, we've already run into problems,
354 * and shouldn't do the idecs.
358 STREAM_CLOSE(rwfile);
360 STREAM_CLOSE(clfilein);
362 STREAM_CLOSE(clfileout);
378 /* Next, we sync the disk. We have to reopen in case we're truncating,
379 * since we were using stdio above, and don't know when the buffers
380 * would otherwise be flushed. There's no stdio fftruncate call.
382 rwFd = IH_OPEN(clvp->vnodeIndex[class].handle);
388 /* If doing a reclone, we're keeping the clone. We need to
389 * truncate the file to offset bytes.
391 if (reclone && !error) {
392 error = FDH_TRUNC(rwFd, offset);
399 /* Now finally do the idec's. At this point, all potential
400 * references have been cleaned up and sent to the disk
401 * (see above fclose and fsync). No matter what happens, we
402 * no longer need to keep these references around.
404 code = ci_Apply(&decHead, IDecProc, (char *)&decRock);
407 ci_Destroy(&decHead);
409 if (ReadWriteOriginal && filecount > 0)
410 V_filecount(rwvp) = filecount;
411 if (ReadWriteOriginal && diskused > 0)
412 V_diskused(rwvp) = diskused;
417 CloneVolume(Error * rerror, Volume * original, Volume * new, Volume * old)
419 afs_int32 code, error = 0;
421 afs_int32 filecount = V_filecount(original), diskused = V_diskused(original);
424 reclone = ((new == old) ? 1 : 0);
426 code = DoCloneIndex(original, new, vLarge, reclone);
429 code = DoCloneIndex(original, new, vSmall, reclone);
432 if (filecount != V_filecount(original) || diskused != V_diskused(original))
433 Log("Clone %u: filecount %d -> %d diskused %d -> %d\n",
434 V_id(original), filecount, V_filecount(original), diskused, V_diskused(original));
436 code = CopyVolumeHeader(&V_disk(original), &V_disk(new));