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
10 /* I/O operations for the Unix open by name (namei) interface. */
12 #include <afsconfig.h>
13 #include <afs/param.h>
20 #ifdef HAVE_SYS_FILE_H
21 # include <sys/file.h>
33 #include <rx/rx_queue.h>
34 #ifdef AFS_PTHREAD_ENV
35 # include <opr/lock.h>
38 #include <afs/afsutil.h>
41 #include <afs/afsint.h>
45 #include "viceinode.h"
47 #include "partition.h"
49 #include "volume_inline.h"
51 #include <afs/errors.h>
54 #include <afs/errmap_nt.h>
58 #define LOCK_SH 1 /* shared lock */
59 #define LOCK_EX 2 /* exclusive lock */
60 #define LOCK_NB 4 /* don't block when locking */
61 #define LOCK_UN 8 /* unlock */
66 #include <afs/work_queue.h>
67 #include <afs/thread_pool.h>
68 #include <vol/vol-salvage.h>
75 namei_iread(IHandle_t * h, afs_foff_t offset, char *buf, afs_fsize_t size)
84 nBytes = FDH_PREAD(fdP, buf, size, offset);
93 namei_iwrite(IHandle_t * h, afs_foff_t offset, char *buf, afs_fsize_t size)
102 nBytes = FDH_PWRITE(fdP, buf, size, offset);
104 FDH_REALLYCLOSE(fdP);
111 /* Inode number format:
112 * low 32 bits - if a regular file or directory, the vnode; else the type.
113 * 32-36 - uniquifier tag and index into counts array for this vnode. Only
114 * two of the available bits are currently used. The rest are
115 * present in case we ever increase the number of types of volumes
116 * in the volume group.
117 * bit 37 : 1 == special, 0 == regular
119 # define NAMEI_VNODEMASK 0x00ffffffff
120 # define NAMEI_TAGSHIFT 32
121 # define NAMEI_INODESPECIAL 0x2000000000
122 # define NAMEI_SPECDIR "R"
123 # define NAMEI_SPECDIRC 'R'
124 #else /* !AFS_NT40_ENV */
125 /* Inode number format:
126 * low 26 bits - vnode number - all 1's if volume special file.
128 * next 3 bits spare (0's)
129 * high 32 bits - uniquifier (regular) or type if spare
131 # define NAMEI_VNODEMASK 0x003ffffff
132 # define NAMEI_TAGSHIFT 26
133 # define NAMEI_UNIQMASK 0xffffffff
134 # define NAMEI_UNIQSHIFT 32
135 # define NAMEI_INODESPECIAL ((Inode)NAMEI_VNODEMASK)
136 /* dir1 is the high 8 bits of the 26 bit vnode */
137 # define VNO_DIR1(vno) ((vno >> 14) & 0xff)
138 /* dir2 is the next 9 bits */
139 # define VNO_DIR2(vno) ((vno >> 9) & 0x1ff)
140 /* "name" is the low 9 bits of the vnode, the 3 bit tag and the uniq */
141 # define NAMEI_SPECDIR "special"
142 #endif /* !AFS_NT40_ENV */
143 #define NAMEI_TAGMASK 0x7
144 #define NAMEI_VNODESPECIAL NAMEI_VNODEMASK
146 #define NAMEI_SPECDIRLEN (sizeof(NAMEI_SPECDIR)-1)
148 #define NAMEI_MAXVOLS 5 /* Maximum supported number of volumes per volume
149 * group, not counting temporary (move) volumes.
150 * This is the number of separate files, all having
151 * the same vnode number, which can occur in a volume
163 static int GetFreeTag(IHandle_t * ih, int vno);
165 /* namei_HandleToInodeDir
167 * Construct the path name of the directory holding the inode data.
168 * Format: /<vicepx>/INODEDIR
173 namei_HandleToInodeDir(namei_t * name, IHandle_t * ih)
175 memset(name, '\0', sizeof(*name));
176 nt_DevToDrive(name->n_drive, ih->ih_dev);
177 strlcpy(name->n_path, name->n_drive, sizeof(name->n_path));
181 /* Format: /<vicepx>/INODEDIR */
182 #define PNAME_BLEN 64
184 namei_HandleToInodeDir(namei_t * name, IHandle_t * ih)
188 memset(name, '\0', sizeof(*name));
191 * Add the /vicepXX string to the start of name->n_base and then calculate
192 * offset as the number of bytes we know we added.
194 * FIXME: This embeds knowledge of the vice partition naming scheme and
195 * mapping from device numbers. There needs to be an API that tells us
198 volutil_PartitionName_r(ih->ih_dev, name->n_base, sizeof(name->n_base));
199 offset = VICE_PREFIX_SIZE + (ih->ih_dev > 25 ? 2 : 1);
200 name->n_base[offset] = OS_DIRSEPC;
202 strlcpy(name->n_base + offset, INODEDIR, sizeof(name->n_base) - offset);
203 strlcpy(name->n_path, name->n_base, sizeof(name->n_path));
207 #define addtoname(N, C) \
209 if ((N)->n_path[strlen((N)->n_path)-1] != OS_DIRSEPC) \
210 strlcat((N)->n_path, OS_DIRSEP, sizeof((N)->n_path)); \
211 strlcat((N)->n_path, (C), sizeof((N)->n_path)); \
217 namei_HandleToVolDir(namei_t * name, IHandle_t * ih)
219 /* X:\Vol_XXXXXXX.data */
223 namei_HandleToInodeDir(name, ih);
224 /* nt_drive added to name by namei_HandleToInodeDir() */
225 namep = name->n_voldir;
226 (void)memcpy(namep, "\\Vol_", 5);
228 (void)strcpy(namep, int_to_base32(str1, ih->ih_vid));
229 namep += strlen(namep);
230 memcpy(namep, ".data", 5);
233 addtoname(name, name->n_voldir);
237 namei_HandleToVolDir(namei_t * name, IHandle_t * ih)
241 namei_HandleToInodeDir(name, ih);
242 (void)int32_to_flipbase64(tmp, (int64_t) (ih->ih_vid & 0xff));
243 strlcpy(name->n_voldir1, tmp, sizeof(name->n_voldir1));
244 addtoname(name, name->n_voldir1);
245 (void)int32_to_flipbase64(tmp, (int64_t) ih->ih_vid);
246 strlcpy(name->n_voldir2, tmp, sizeof(name->n_voldir2));
247 addtoname(name, name->n_voldir2);
251 /* namei_HandleToName
253 * Constructs a file name for the fully qualified handle.
256 /* Note that special files end up in X:\Vol_XXXXXXX.data\R */
258 namei_HandleToName(namei_t * name, IHandle_t * ih)
260 int vno = (int)(ih->ih_ino & NAMEI_VNODEMASK);
261 int special = (ih->ih_ino & NAMEI_INODESPECIAL)?1:0;
262 int tag = (int)((ih->ih_ino >> NAMEI_TAGSHIFT) & NAMEI_TAGMASK);
265 namei_HandleToVolDir(name, ih);
268 name->n_dir[0] = NAMEI_SPECDIRC;
271 name->n_dir[0] = 'Q';
273 name->n_dir[0] = ((vno & 0x1f) >> 1) + 'A';
276 name->n_dir[1] = '\0';
277 addtoname(name, name->n_dir);
278 /* X:\Vol_XXXXXXX.data\X\V_XXXXXXX.XXX */
279 namep = name->n_inode;
280 (void)memcpy(namep, "\\V_", 3);
282 (void)strcpy(namep, int_to_base32(str1, vno));
283 namep += strlen(namep);
285 (void)strcpy(namep, int_to_base32(str1, tag));
286 namep += strlen(namep);
287 addtoname(name, name->n_inode);
290 /* Note that special files end up in /vicepX/InodeDir/Vxx/V*.data/special */
292 namei_HandleToName(namei_t * name, IHandle_t * ih)
294 int vno = (int)(ih->ih_ino & NAMEI_VNODEMASK);
297 namei_HandleToVolDir(name, ih);
299 if (vno == NAMEI_VNODESPECIAL) {
300 strlcpy(name->n_dir1, NAMEI_SPECDIR, sizeof(name->n_dir1));
301 addtoname(name, name->n_dir1);
302 name->n_dir2[0] = '\0';
304 (void)int32_to_flipbase64(str, VNO_DIR1(vno));
305 strlcpy(name->n_dir1, str, sizeof(name->n_dir1));
306 addtoname(name, name->n_dir1);
307 (void)int32_to_flipbase64(str, VNO_DIR2(vno));
308 strlcpy(name->n_dir2, str, sizeof(name->n_dir2));
309 addtoname(name, name->n_dir2);
311 (void)int64_to_flipbase64(str, (int64_t) ih->ih_ino);
312 strlcpy(name->n_inode, str, sizeof(name->n_inode));
313 addtoname(name, name->n_inode);
318 /* The following is a warning to tell sys-admins to not muck about in this
321 #define VICE_README "These files and directories are a part of the AFS \
322 namespace. Modifying them\nin any way will result in loss of AFS data,\n\
323 ownership and permissions included.\n"
325 namei_ViceREADME(char *partition)
330 /* Create the inode directory if we're starting for the first time */
331 snprintf(filename, sizeof filename, "%s" OS_DIRSEP "%s", partition,
333 mkdir(filename, 0700);
335 snprintf(filename, sizeof filename,
336 "%s" OS_DIRSEP "%s" OS_DIRSEP "README",
337 partition, INODEDIR);
338 fd = OS_OPEN(filename, O_WRONLY | O_CREAT | O_TRUNC, 0444);
339 if (fd != INVALID_FD) {
340 (void)OS_WRITE(fd, VICE_README, strlen(VICE_README));
347 /* namei_CreateDataDirectories
349 * If creating the file failed because of ENOENT or ENOTDIR, try
350 * creating all the directories first.
354 namei_CreateDataDirectories(namei_t * name, int *created)
361 snprintf(tmp, 256, "%s" OS_DIRSEP "%s", name->n_drive, name->n_voldir);
363 if (mkdir(tmp) < 0) {
374 for (i = 'A'; i <= NAMEI_SPECDIRC; i++) {
376 if (mkdir(tmp) < 0 && errno != EEXIST)
382 #define create_dir() \
384 if (mkdir(tmp, 0700)<0) { \
385 if (errno != EEXIST) \
393 #define create_nextdir(A) \
395 strcat(tmp, OS_DIRSEP); strcat(tmp, A); create_dir(); \
399 namei_CreateDataDirectories(namei_t * name, int *created)
405 strlcpy(tmp, name->n_base, sizeof(tmp));
408 create_nextdir(name->n_voldir1);
409 create_nextdir(name->n_voldir2);
410 create_nextdir(name->n_dir1);
411 if (name->n_dir2[0]) {
412 create_nextdir(name->n_dir2);
419 /* delTree(): Deletes an entire tree of directories (no files)
421 * root : Full path to the subtree. Should be big enough for PATH_MAX
422 * tree : the subtree to be deleted is rooted here. Specifies only the
423 * subtree beginning at tree (not the entire path). It should be
424 * a pointer into the "root" buffer.
426 * errp : errno of the first error encountered during the directory cleanup.
427 * *errp should have been initialized to 0.
430 * -1 : If errors were encountered during cleanup and error is set to
434 * If there are errors, we try to work around them and delete as many
435 * directories as possible. We don't attempt to remove directories that still
436 * have non-dir entries in them.
439 delTree(char *root, char *tree, int *errp)
447 /* delete the children first */
448 cp = strchr(tree, OS_DIRSEPC);
450 delTree(root, cp + 1, errp);
453 cp = tree + strlen(tree); /* move cp to the end of string tree */
455 /* now delete all entries in this dir */
456 if ((ds = opendir(root)) != NULL) {
458 while ((dirp = readdir(ds))) {
459 /* ignore . and .. */
460 if (!strcmp(dirp->d_name, ".") || !strcmp(dirp->d_name, ".."))
462 /* since root is big enough, we reuse the space to
463 * concatenate the dirname to the current tree
465 strcat(root, OS_DIRSEP);
466 strcat(root, dirp->d_name);
467 if (afs_stat(root, &st) == 0 && S_ISDIR(st.st_mode)) {
468 /* delete this subtree */
469 delTree(root, cp + 1, errp);
471 *errp = *errp ? *errp : errno;
473 /* recover path to our cur tree by truncating it to
478 /* if (!errno) -- closedir not implicit if we got an error */
482 /* finally axe the current dir */
484 *errp = *errp ? *errp : errno;
486 #ifndef AFS_PTHREAD_ENV /* let rx get some work done */
488 #endif /* !AFS_PTHREAD_ENV */
493 /* if we encountered errors during cleanup, we return a -1 */
502 /* namei_RemoveDataDirectories
504 * Returns 0 on success.
505 * Returns -1 on error. Typically, callers ignore this error because we
506 * can continue running if the removes fail. The salvage process will
507 * finish tidying up for us.
512 namei_RemoveDataDirectories(namei_t * name)
519 snprintf(tmp, 256, "%s" OS_DIRSEP "%s", name->n_drive, name->n_voldir);
522 path += strlen(path);
523 *path++ = OS_DIRSEPC;
525 for (i = 'A'; i <= NAMEI_SPECDIRC; i++) {
527 if (rmdir(name->n_path) < 0 && errno != ENOENT)
532 /* Delete the Vol_NNNNNN.data directory. */
535 if (rmdir(name->n_path) < 0 && errno != ENOENT) {
543 * We only use the n_base and n_voldir1 entries
544 * and only do rmdir's.
547 namei_RemoveDataDirectories(namei_t * name)
551 int prefixlen = strlen(name->n_base), err = 0;
552 int vollen = strlen(name->n_voldir1);
553 char pbuf[MAXPATHLEN];
557 strlcpy(path, name->n_path, sizeof(pbuf));
559 /* move past the prefix and n_voldir1 */
560 path = path + prefixlen + 1 + vollen + 1; /* skip over the trailing / */
562 /* now delete all dirs upto path */
563 code = delTree(pbuf, path, &err);
565 /* We've now deleted everything under /n_base/n_voldir1/n_voldir2 that
566 * we could. Do not delete /n_base/n_voldir1, since doing such might
567 * interrupt another thread trying to create a volume. We could introduce
568 * some locking to make this safe (or only remove it for whole-partition
569 * salvages), but by not deleting it we only leave behind a maximum of
570 * 256 empty directories. So at least for now, don't bother. */
575 /* Create the file in the name space.
577 * Parameters stored as follows:
579 * p1 - volid - implied in containing directory.
580 * p2 - vnode - name is <vno:31-23>/<vno:22-15>/<vno:15-0><uniq:31-5><tag:2-0>
581 * p3 - uniq -- bits 4-0 are in mode bits 4-0
582 * p4 - dv ---- dv:15-0 in uid, dv:29-16 in gid, dv:31-30 in mode:6-5
584 * p1 - volid - creation time - dwHighDateTime
585 * p2 - vnode - -1 means special, file goes in "S" subdirectory.
586 * p3 - type -- name is <type>.<tag> where tag is a file name unqiquifier.
587 * p4 - parid - parent volume id - implied in containing directory.
589 * Return value is the inode number or (Inode)-1 if error.
590 * We "know" there is only one link table, so return EEXIST if there already
591 * is a link table. It's up to the calling code to test errno and increment
597 * This function is called by VCreateVolume to hide the implementation
598 * details of the inode numbers. This only allows for 7 volume special
599 * types, but if we get that far, this could should be dead by then.
602 namei_MakeSpecIno(VolumeId volid, int type)
605 ino = NAMEI_INODESPECIAL;
608 /* tag is always 0 for special */
610 type &= NAMEI_TAGMASK;
611 ino |= ((Inode) type) << NAMEI_TAGSHIFT;
612 ino |= ((Inode) volid) << NAMEI_UNIQSHIFT;
620 SetOGM(FD_t fd, int parm, int tag)
626 SetWinOGM(FD_t fd, int p1, int p2)
631 ftime.dwHighDateTime = p1;
632 ftime.dwLowDateTime = p2;
634 code = SetFileTime(fd, &ftime, NULL /*access*/, NULL /*write*/);
641 GetWinOGM(FD_t fd, int *p1, int *p2)
646 code = GetFileTime(fd, &ftime, NULL /*access*/, NULL /*write*/);
650 *p1 = ftime.dwHighDateTime;
651 *p2 = ftime.dwLowDateTime;
657 CheckOGM(FdHandle_t *fdP, int p1)
661 if (GetWinOGM(fdP->fd_fd, &ogm_p1, &ogm_p2)) {
673 FixSpecialOGM(FdHandle_t *fdP, int check)
675 Inode ino = fdP->fd_ih->ih_ino;
676 VnodeId vno = NAMEI_VNODESPECIAL, ogm_vno;
679 if (GetWinOGM(fdP->fd_fd, &ogm_volid, &ogm_vno)) {
683 /* the only thing we can check is the vnode number; for the volid we have
684 * nothing else to compare against */
685 if (vno != ogm_vno) {
689 if (SetWinOGM(fdP->fd_fd, ogm_volid, vno)) {
696 #else /* AFS_NT40_ENV */
697 /* SetOGM - set owner group and mode bits from parm and tag */
699 SetOGM(FD_t fd, int parm, int tag)
702 * owner - low 15 bits of parm.
703 * group - next 15 bits of parm.
704 * mode - 2 bits of parm, then lowest = 3 bits of tag.
706 int owner, group, mode;
708 owner = parm & 0x7fff;
709 group = (parm >> 15) & 0x7fff;
710 if (fchown(fd, owner, group) < 0)
713 mode = (parm >> 27) & 0x18;
715 if (fchmod(fd, mode) < 0)
720 /* GetOGM - get parm and tag from owner, group and mode bits. */
722 GetOGMFromStat(struct afs_stat_st *status, int *parm, int *tag)
724 *parm = status->st_uid | (status->st_gid << 15);
725 *parm |= (status->st_mode & 0x18) << 27;
726 *tag = status->st_mode & 0x7;
730 GetOGM(FdHandle_t *fdP, int *parm, int *tag)
732 struct afs_stat_st status;
733 if (afs_fstat(fdP->fd_fd, &status) < 0)
735 GetOGMFromStat(&status, parm, tag);
740 CheckOGM(FdHandle_t *fdP, int p1)
744 if (GetOGM(fdP, &parm, &tag) < 0)
753 FixSpecialOGM(FdHandle_t *fdP, int check)
755 int inode_volid, ogm_volid;
756 int inode_type, ogm_type;
757 Inode ino = fdP->fd_ih->ih_ino;
759 inode_volid = ((ino >> NAMEI_UNIQSHIFT) & NAMEI_UNIQMASK);
760 inode_type = (int)((ino >> NAMEI_TAGSHIFT) & NAMEI_TAGMASK);
762 if (GetOGM(fdP, &ogm_volid, &ogm_type) < 0) {
763 Log("Error retrieving OGM info\n");
767 if (inode_volid != ogm_volid || inode_type != ogm_type) {
768 Log("%sIncorrect OGM data (ino: vol %u type %d) (ogm: vol %u type %d)\n",
769 check?"":"Fixing ", inode_volid, inode_type, ogm_volid, ogm_type);
775 if (SetOGM(fdP->fd_fd, inode_volid, inode_type) < 0) {
776 Log("Error setting OGM data\n");
783 #endif /* !AFS_NT40_ENV */
786 * Check/fix the OGM data for an inode
788 * @param[in] fdP Open file handle for the inode to check
789 * @param[in] check 1 to just check the OGM data, and return an error if it
790 * is incorrect. 0 to fix the OGM data if it is incorrect.
792 * @pre fdP must be for a special inode
799 namei_FixSpecialOGM(FdHandle_t *fdP, int check)
802 Inode ino = fdP->fd_ih->ih_ino;
804 vnode = (int)(ino & NAMEI_VNODEMASK);
805 if (vnode != NAMEI_VNODESPECIAL) {
806 Log("FixSpecialOGM: non-special vnode %u\n", vnode);
810 return FixSpecialOGM(fdP, check);
813 int big_vno = 0; /* Just in case we ever do 64 bit vnodes. */
815 /* Derive the name and create it O_EXCL. If that fails we have an error.
816 * Get the tag from a free column in the link table.
820 namei_icreate(IHandle_t * lh, char *part, afs_uint32 p1, afs_uint32 p2, afs_uint32 p3, afs_uint32 p4)
823 FD_t fd = INVALID_FD;
834 memset((void *)&tmp, 0, sizeof(IHandle_t));
835 memset(&tfd, 0, sizeof(FdHandle_t));
837 tmp.ih_dev = nt_DriveToDev(part);
838 if (tmp.ih_dev == -1) {
843 if (p2 == INODESPECIAL) {
844 /* Parameters for special file:
845 * p1 - volume id - goes into owner/group/mode
846 * p2 - vnode == INODESPECIAL
848 * p4 - parent volume id
853 tmp.ih_vid = p4; /* Use parent volume id, where this file will be. */
854 tmp.ih_ino = namei_MakeSpecIno(p1, p3);
856 int vno = p2 & NAMEI_VNODEMASK;
857 /* Parameters for regular file:
871 tmp.ih_ino = (Inode) p2;
876 namei_HandleToName(&name, &tmp);
877 p = strrchr((char *)&name.n_path, '.');
879 for (tag = 0; tag < NAMEI_MAXVOLS; tag++) {
880 *p = *int_to_base32(str1, tag);
881 fd = OS_OPEN((char *)&name.n_path, O_CREAT | O_RDWR | O_EXCL, 0666);
882 if (fd == INVALID_FD) {
883 if (errno == ENOTDIR || errno == ENOENT) {
884 if (namei_CreateDataDirectories(&name, &created_dir) == 0)
885 fd = OS_OPEN((char *)&name.n_path, O_CREAT | O_RDWR | O_EXCL, 0666);
889 if (fd != INVALID_FD)
891 if (p2 == INODESPECIAL && p3 == VI_LINKTABLE)
894 if (fd == INVALID_FD) {
898 tmp.ih_ino &= ~(((Inode) NAMEI_TAGMASK) << NAMEI_TAGSHIFT);
899 tmp.ih_ino |= (((Inode) tag) << NAMEI_TAGSHIFT);
902 if (SetWinOGM(fd, ogm_p1, ogm_p2)) {
903 errno = OS_ERROR(EBADF);
909 if (p2 != INODESPECIAL) {
910 if (fd == INVALID_FD) {
912 code = nt_unlink((char *)&name.n_path);
921 code = namei_SetLinkCount(fdP, tmp.ih_ino, 1, 0);
923 } else if (p2 == INODESPECIAL && p3 == VI_LINKTABLE) {
924 if (fd == INVALID_FD)
926 /* hack at tmp to setup for set link count call. */
928 code = namei_SetLinkCount(&tfd, (Inode) 0, 1, 0);
933 if (fd != INVALID_FD)
936 if (code || (fd == INVALID_FD)) {
937 if (p2 != INODESPECIAL) {
940 namei_SetLinkCount(fdP, tmp.ih_ino, 0, 0);
946 int save_errno = errno;
947 namei_RemoveDataDirectories(&name);
951 return (code || (fd == INVALID_FD)) ? (Inode) -1 : tmp.ih_ino;
953 #else /* !AFS_NT40_ENV */
955 icreate(IHandle_t * lh, char *part, afs_uint32 p1, afs_uint32 p2, afs_uint32 p3, afs_uint32 p4,
956 FD_t *afd, Inode *ainode)
968 memset((void *)&tmp, 0, sizeof(IHandle_t));
969 memset(&tfd, 0, sizeof(FdHandle_t));
971 tmp.ih_dev = volutil_GetPartitionID(part);
972 if (tmp.ih_dev == -1) {
978 /* Parameters for special file:
979 * p1 - volume id - goes into owner/group/mode
982 * p4 - parent volume id
987 tmp.ih_vid = p4; /* Use parent volume id, where this file will be. */
988 tmp.ih_ino = namei_MakeSpecIno(p1, p3);
990 int vno = p2 & NAMEI_VNODEMASK;
991 /* Parameters for regular file:
1003 /* If GetFreeTag succeeds, it atomically sets link count to 1. */
1004 tag = GetFreeTag(lh, p2);
1009 tmp.ih_ino = (Inode) p2;
1010 /* name is <uniq(p3)><tag><vno(p2)> */
1011 tmp.ih_ino |= ((Inode) tag) << NAMEI_TAGSHIFT;
1012 tmp.ih_ino |= ((Inode) p3) << NAMEI_UNIQSHIFT;
1017 namei_HandleToName(&name, &tmp);
1018 fd = OS_OPEN(name.n_path, O_CREAT | O_EXCL | O_RDWR, 0);
1019 if (fd == INVALID_FD) {
1020 if (errno == ENOTDIR || errno == ENOENT) {
1021 if (namei_CreateDataDirectories(&name, &created_dir) < 0)
1023 fd = OS_OPEN(name.n_path, O_CREAT | O_EXCL | O_RDWR,
1025 if (fd == INVALID_FD)
1031 if (SetOGM(fd, ogm_parm, tag) < 0) {
1037 if (p2 == (afs_uint32)-1 && p3 == VI_LINKTABLE) {
1038 /* hack at tmp to setup for set link count call. */
1039 memset((void *)&tfd, 0, sizeof(FdHandle_t)); /* minimalistic still, but a little cleaner */
1042 code = namei_SetLinkCount(&tfd, (Inode) 0, 1, 0);
1046 if (code || (fd == INVALID_FD)) {
1050 namei_SetLinkCount(fdP, tmp.ih_ino, 0, 0);
1057 *ainode = tmp.ih_ino;
1063 namei_icreate(IHandle_t * lh, char *part,
1064 afs_uint32 p1, afs_uint32 p2, afs_uint32 p3, afs_uint32 p4)
1067 int fd = INVALID_FD;
1070 code = icreate(lh, part, p1, p2, p3, p4, &fd, &ino);
1071 if (fd != INVALID_FD) {
1074 return (code || (fd == INVALID_FD)) ? (Inode) - 1 : ino;
1078 namei_icreate_init(IHandle_t * lh, int dev, char *part,
1079 afs_uint32 p1, afs_uint32 p2, afs_uint32 p3, afs_uint32 p4)
1082 int fd = INVALID_FD;
1087 code = icreate(lh, part, p1, p2, p3, p4, &fd, &ino);
1088 if (fd == INVALID_FD) {
1096 IH_INIT(ihP, dev, p1, ino);
1097 fdP = ih_attachfd(ihP, fd);
1110 namei_iopen(IHandle_t * h)
1115 /* Convert handle to file name. */
1116 namei_HandleToName(&name, h);
1117 fd = OS_OPEN((char *)&name.n_path, O_RDWR, 0666);
1121 /* Need to detect vol special file and just unlink. In those cases, the
1122 * handle passed in _is_ for the inode. We only check p1 for the special
1126 namei_dec(IHandle_t * ih, Inode ino, int p1)
1133 if ((ino & NAMEI_INODESPECIAL) == NAMEI_INODESPECIAL) {
1135 int type = (int)((ino >> NAMEI_TAGSHIFT) & NAMEI_TAGMASK);
1137 /* Verify this is the right file. */
1138 IH_INIT(tmp, ih->ih_dev, ih->ih_vid, ino);
1140 namei_HandleToName(&name, tmp);
1145 errno = OS_ERROR(ENOENT);
1149 if (CheckOGM(fdP, p1) < 0) {
1150 FDH_REALLYCLOSE(fdP);
1152 errno = OS_ERROR(EINVAL);
1156 /* If it's the link table itself, decrement the link count. */
1157 if (type == VI_LINKTABLE) {
1158 if ((count = namei_GetLinkCount(fdP, (Inode) 0, 1, 0, 1)) < 0) {
1159 FDH_REALLYCLOSE(fdP);
1165 if (namei_SetLinkCount(fdP, (Inode) 0, count < 0 ? 0 : count, 1) <
1167 FDH_REALLYCLOSE(fdP);
1179 if ((code = OS_UNLINK(name.n_path)) == 0) {
1180 if (type == VI_LINKTABLE) {
1181 /* Try to remove directory. If it fails, that's ok.
1182 * Salvage will clean up.
1184 char *slash = strrchr(name.n_path, OS_DIRSEPC);
1186 /* avoid an rmdir() on the file we just unlinked */
1189 (void)namei_RemoveDataDirectories(&name);
1192 FDH_REALLYCLOSE(fdP);
1195 /* Get a file descriptor handle for this Inode */
1201 if ((count = namei_GetLinkCount(fdP, ino, 1, 0, 1)) < 0) {
1202 FDH_REALLYCLOSE(fdP);
1208 if (namei_SetLinkCount(fdP, ino, count, 1) < 0) {
1209 FDH_REALLYCLOSE(fdP);
1214 IH_INIT(th, ih->ih_dev, ih->ih_vid, ino);
1215 Log("Warning: Lost ref on ihandle dev %d vid %" AFS_VOLID_FMT " ino %lld\n",
1216 th->ih_dev, afs_printable_VolumeId_lu(th->ih_vid), (afs_int64)th->ih_ino);
1219 /* If we're less than 0, someone presumably unlinked;
1220 don't bother setting count to 0, but we need to drop a lock */
1221 if (namei_SetLinkCount(fdP, ino, 0, 1) < 0) {
1222 FDH_REALLYCLOSE(fdP);
1228 IH_INIT(th, ih->ih_dev, ih->ih_vid, ino);
1230 namei_HandleToName(&name, th);
1232 code = OS_UNLINK(name.n_path);
1241 namei_inc(IHandle_t * h, Inode ino, int p1)
1247 if ((ino & NAMEI_INODESPECIAL) == NAMEI_INODESPECIAL) {
1248 int type = (int)((ino >> NAMEI_TAGSHIFT) & NAMEI_TAGMASK);
1249 if (type != VI_LINKTABLE)
1254 /* Get a file descriptor handle for this Inode */
1260 if ((count = namei_GetLinkCount(fdP, ino, 1, 0, 1)) < 0)
1265 errno = OS_ERROR(EINVAL);
1269 if (namei_SetLinkCount(fdP, ino, count, 1) < 0)
1273 FDH_REALLYCLOSE(fdP);
1280 #ifndef AFS_NT40_ENV
1282 namei_replace_file_by_hardlink(IHandle_t *hLink, IHandle_t *hTarget)
1288 /* Convert handle to file name. */
1289 namei_HandleToName(&nameLink, hLink);
1290 namei_HandleToName(&nameTarget, hTarget);
1292 OS_UNLINK(nameLink.n_path);
1293 code = link(nameTarget.n_path, nameLink.n_path);
1298 namei_copy_on_write(IHandle_t *h)
1304 struct afs_stat_st tstat;
1307 namei_HandleToName(&name, h);
1308 if (afs_stat(name.n_path, &tstat) < 0)
1310 if (tstat.st_nlink > 1) { /* do a copy on write */
1319 snprintf(path, sizeof(path), "%s-tmp", name.n_path);
1320 fd = OS_OPEN(path, O_CREAT | O_EXCL | O_RDWR, 0);
1321 if (fd == INVALID_FD) {
1332 size = tstat.st_size;
1335 tlen = size > 8192 ? 8192 : size;
1336 if (FDH_PREAD(fdP, buf, tlen, offset) != tlen)
1338 if (OS_WRITE(fd, buf, tlen) != tlen)
1344 FDH_REALLYCLOSE(fdP);
1349 OS_UNLINK(name.n_path);
1350 code = rename(path, name.n_path);
1357 /************************************************************************
1358 * File Name Structure
1359 ************************************************************************
1361 * Each AFS file needs a unique name and it needs to be findable with
1362 * minimal lookup time. Note that the constraint on the number of files and
1363 * directories in a volume is the size of the vnode index files and the
1364 * max file size AFS supports (for internal files) of 2^31. Since a record
1365 * in the small vnode index file is 64 bytes long, we can have at most
1366 * (2^31)/64 or 33554432 files. A record in the large index file is
1367 * 256 bytes long, giving a maximum of (2^31)/256 = 8388608 directories.
1368 * Another layout parameter is that there is roughly a 16 to 1 ratio between
1369 * the number of files and the number of directories.
1371 * Using this information we can see that a layout of 256 directories, each
1372 * with 512 subdirectories and each of those having 512 files gives us
1373 * 256*512*512 = 67108864 AFS files and directories.
1375 * The volume, vnode, uniquifier and data version, as well as the tag
1376 * are required, either for finding the file or for salvaging. It's best to
1377 * restrict the name to something that can be mapped into 64 bits so the
1378 * "Inode" is easily comparable (using "==") to other "Inodes". The tag
1379 * is used to distinguish between different versions of the same file
1380 * which are currently in the RW and clones of a volume. See "Link Table
1381 * Organization" below for more information on the tag. The tag is
1382 * required in the name of the file to ensure a unique name.
1384 * ifdef AFS_NT40_ENV
1385 * The data for each volume group is in a separate directory. The name of the
1386 * volume is of the form: Vol_NNNNNN.data, where NNNNNN is a base 32
1387 * representation of the RW volume ID (even where the RO is the only volume
1388 * on the partition). Below that are separate subdirectories for the
1389 * AFS directories and special files. There are also 16 directories for files,
1390 * hashed on the low 5 bits (recall bit0 is always 0) of the vnode number.
1391 * These directories are named:
1392 * A - P - 16 file directories.
1393 * Q ----- data directory
1394 * R ----- special files directory
1396 * The vnode is hashed into the directory using the low bits of the
1399 * The format of a file name for a regular file is:
1400 * Y:\Vol_NNNNNN.data\X\V_IIIIII.J
1401 * Y - partition encoded as drive letter, starting with D
1402 * NNNNNN - base 32 encoded volume number of RW volume
1403 * X - hash directory, as above
1404 * IIIIII - base 32 encoded vnode number
1405 * J - base 32 encoded tag
1407 * uniq is stored in the dwHighDateTime creation time field
1408 * dv is stored in the dwLowDateTime creation time field
1410 * Special inodes are always in the R directory, as above, and are
1412 * True child volid is stored in the dwHighDateTime creation time field
1413 * vnode number is always -1 (Special)
1414 * type is the IIIIII part of the filename
1415 * uniq is the J part of the filename
1416 * parent volume id is implied in the containing directory
1419 * We can store data in the uid, gid and mode bits of the files, provided
1420 * the directories have root only access. This gives us 15 bits for each
1421 * of uid and gid (GNU chown considers 65535 to mean "don't change").
1422 * There are 9 available mode bits. Adn we need to store a total of
1423 * 32 (volume id) + 26 (vnode) + 32 (uniquifier) + 32 (data-version) + 3 (tag)
1424 * or 131 bits somewhere.
1426 * The format of a file name for a regular file is:
1427 * /vicepX/AFSIDat/V1/V2/AA/BB/<tag><uniq><vno>
1428 * V1 - low 8 bits of RW volume id
1429 * V2 - all bits of RW volume id
1430 * AA - high 8 bits of vnode number.
1431 * BB - next 9 bits of vnode number.
1432 * <tag><uniq><vno> - file name
1434 * Volume special files are stored in a separate directory:
1435 * /vicepX/AFSIDat/V1/V2/special/<tag><uniq><vno>
1438 * The vnode is hashed into the directory using the high bits of the
1439 * vnode number. This is so that consecutively created vnodes are in
1440 * roughly the same area on the disk. This will at least be optimal if
1441 * the user is creating many files in the same AFS directory. The name
1442 * should be formed so that the leading characters are different as quickly
1443 * as possible, leading to faster discards of incorrect matches in the
1451 /************************************************************************
1452 * Link Table Organization
1453 ************************************************************************
1455 * The link table volume special file is used to hold the link counts that
1456 * are held in the inodes in inode based AFS vice filesystems. For user
1457 * space access, the link counts are being kept in a separate
1458 * volume special file. The file begins with the usual version stamp
1459 * information and is then followed by one row per vnode number. vnode 0
1460 * is used to hold the link count of the link table itself. That is because
1461 * the same link table is shared among all the volumes of the volume group
1462 * and is deleted only when the last volume of a volume group is deleted.
1464 * Within each row, the columns are 3 bits wide. They can each hold a 0 based
1465 * link count from 0 through 7. Each colume represents a unique instance of
1466 * that vnode. Say we have a file shared between the RW and a RO and a
1467 * different version of the file (or a different uniquifer) for the BU volume.
1468 * Then one column would be holding the link count of 2 for the RW and RO
1469 * and a different column would hold the link count of 1 for the BU volume.
1470 * # ifdef AFS_NT40_ENV
1471 * The column used is determined for NT by the uniquifier tag applied to
1472 * generate a unique file name in the NTFS namespace. The file name is
1473 * of the form "V_<vno>.<tag>" . And the <tag> is also the column number
1474 * in the link table.
1476 * Note that we allow only 5 volumes per file, giving 15 bits used in the
1480 #define LINKTABLE_WIDTH 2
1481 #define LINKTABLE_SHIFT 1 /* log 2 = 1 */
1484 * compute namei link table file and bit offset from inode number.
1486 * @param[in] ino inode number
1487 * @param[out] offset link table file offset
1488 * @param[out] index bit offset within 2-byte record
1493 namei_GetLCOffsetAndIndexFromIno(Inode ino, afs_foff_t * offset, int *index)
1495 int toff = (int)(ino & NAMEI_VNODEMASK);
1496 int tindex = (int)((ino >> NAMEI_TAGSHIFT) & NAMEI_TAGMASK);
1498 *offset = (afs_foff_t) ((toff << LINKTABLE_SHIFT) + 8); /* * 2 + sizeof stamp */
1499 *index = (tindex << 1) + tindex;
1502 #ifdef AFS_PTHREAD_ENV
1503 /* XXX do static initializers work for WINNT/pthread? */
1504 pthread_mutex_t _namei_glc_lock = PTHREAD_MUTEX_INITIALIZER;
1505 #define NAMEI_GLC_LOCK opr_mutex_enter(&_namei_glc_lock)
1506 #define NAMEI_GLC_UNLOCK opr_mutex_exit(&_namei_glc_lock)
1507 #else /* !AFS_PTHREAD_ENV */
1508 #define NAMEI_GLC_LOCK
1509 #define NAMEI_GLC_UNLOCK
1510 #endif /* !AFS_PTHREAD_ENV */
1513 * get the link count of an inode.
1515 * @param[in] h namei link count table file handle
1516 * @param[in] ino inode number for which we are requesting a link count
1517 * @param[in] lockit if asserted, return with lock held on link table file
1518 * @param[in] fixup if asserted, write 1 to link count when read() returns
1520 * @param[in] nowrite return success on zero byte read or ZLC
1522 * @post if lockit asserted and lookup was successful, will return with write
1523 * lock on link table file descriptor
1525 * @return link count
1526 * @retval -1 namei link table i/o error
1531 namei_GetLinkCount(FdHandle_t * h, Inode ino, int lockit, int fixup, int nowrite)
1533 unsigned short row = 0;
1538 /* there's no linktable yet. the salvager will create one later */
1539 if (h->fd_fd == INVALID_FD && fixup)
1541 namei_GetLCOffsetAndIndexFromIno(ino, &offset, &index);
1544 if (FDH_LOCKFILE(h, offset) != 0)
1548 rc = FDH_PREAD(h, (char*)&row, sizeof(row), offset);
1550 goto bad_getLinkByte;
1552 if ((rc == 0 || !((row >> index) & NAMEI_TAGMASK)) && fixup && nowrite)
1554 if (rc == 0 && fixup) {
1556 * extend link table and write a link count of 1 for ino
1558 * in order to make MT-safe, truncation (extension really)
1559 * must happen under a mutex
1562 if (FDH_SIZE(h) >= offset+sizeof(row)) {
1564 goto bad_getLinkByte;
1566 FDH_TRUNC(h, offset+sizeof(row));
1568 rc = FDH_PWRITE(h, (char *)&row, sizeof(row), offset);
1571 if (rc != sizeof(row)) {
1572 goto bad_getLinkByte;
1575 if (fixup && !((row >> index) & NAMEI_TAGMASK)) {
1579 * in order to make this mt-safe, we need to do the read-modify-write
1580 * under a mutex. thus, we repeat the read inside the lock.
1583 rc = FDH_PREAD(h, (char *)&row, sizeof(row), offset);
1584 if (rc == sizeof(row)) {
1586 rc = FDH_PWRITE(h, (char *)&row, sizeof(row), offset);
1589 if (rc != sizeof(row))
1590 goto bad_getLinkByte;
1593 return (int)((row >> index) & NAMEI_TAGMASK);
1597 FDH_UNLOCKFILE(h, offset);
1602 namei_SetNonZLC(FdHandle_t * h, Inode ino)
1604 return namei_GetLinkCount(h, ino, 0, 1, 0);
1607 /* Return a free column index for this vnode. */
1609 GetFreeTag(IHandle_t * ih, int vno)
1623 /* Only one manipulates at a time. */
1624 if (FDH_LOCKFILE(fdP, offset) != 0) {
1625 FDH_REALLYCLOSE(fdP);
1629 offset = (vno << LINKTABLE_SHIFT) + 8; /* * 2 + sizeof stamp */
1631 nBytes = FDH_PREAD(fdP, (char *)&row, sizeof(row), offset);
1632 if (nBytes != sizeof(row)) {
1638 /* Now find a free column in this row and claim it. */
1639 for (col = 0; col < NAMEI_MAXVOLS; col++) {
1640 coldata = 7 << (col * 3);
1641 if ((row & coldata) == 0)
1644 if (col >= NAMEI_MAXVOLS) {
1649 coldata = 1 << (col * 3);
1652 if (FDH_PWRITE(fdP, (char *)&row, sizeof(row), offset) != sizeof(row)) {
1655 (void)FDH_SYNC(fdP);
1656 FDH_UNLOCKFILE(fdP, offset);
1661 FDH_UNLOCKFILE(fdP, offset);
1662 FDH_REALLYCLOSE(fdP);
1668 /* namei_SetLinkCount
1669 * If locked is set, assume file is locked. Otherwise, lock file before
1670 * proceeding to modify it.
1673 namei_SetLinkCount(FdHandle_t * fdP, Inode ino, int count, int locked)
1679 ssize_t nBytes = -1;
1681 namei_GetLCOffsetAndIndexFromIno(ino, &offset, &index);
1684 if (FDH_LOCKFILE(fdP, offset) != 0) {
1689 nBytes = FDH_PREAD(fdP, (char *)&row, sizeof(row), offset);
1690 if (nBytes != sizeof(row)) {
1692 errno = OS_ERROR(EBADF);
1693 goto bad_SetLinkCount;
1698 bytesRead = 7 << index;
1700 row &= (unsigned short)~bytesRead;
1701 row |= (unsigned short)count;
1703 if (FDH_PWRITE(fdP, (char *)&row, sizeof(short), offset) != sizeof(short)) {
1704 errno = OS_ERROR(EBADF);
1705 goto bad_SetLinkCount;
1707 (void)FDH_SYNC(fdP);
1713 FDH_UNLOCKFILE(fdP, offset);
1715 /* disallowed above 7, so... */
1720 /* ListViceInodes - write inode data to a results file. */
1721 static int DecodeInode(char *dpath, char *name, struct ViceInodeInfo *info,
1723 static int DecodeVolumeName(char *name, VolumeId *vid);
1724 static int namei_ListAFSSubDirs(IHandle_t * dirIH,
1725 int (*write_fun) (FD_t,
1726 struct ViceInodeInfo *,
1727 char *, char *), FD_t fp,
1728 int (*judgeFun) (struct ViceInodeInfo *,
1729 VolumeId vid, void *),
1730 VolumeId singleVolumeNumber, void *rock);
1735 * Write the inode data to the results file.
1737 * Returns -2 on error, 0 on success.
1739 * This is written as a callback simply so that other listing routines
1740 * can use the same inode reading code.
1743 WriteInodeInfo(FD_t fp, struct ViceInodeInfo *info, char *dir, char *name)
1746 n = OS_WRITE(fp, info, sizeof(*info));
1747 return (n == sizeof(*info)) ? 0 : -2;
1751 int mode_errors; /* Number of errors found in mode bits on directories. */
1753 VerifyDirPerms(char *path)
1755 struct afs_stat_st status;
1757 if (afs_stat(path, &status) < 0) {
1758 Log("Unable to stat %s. Please manually verify mode bits for this"
1759 " directory\n", path);
1761 if (((status.st_mode & 0777) != 0700) || (status.st_uid != 0))
1767 * Fill the results file with the requested inode information.
1769 * This code optimizes single volume salvages by just looking at that one
1770 * volume's directory.
1772 * @param[in] devname device name string
1773 * @param[in] moutnedOn vice partition mount point
1774 * @param[in] resultFile result file in which to write inode
1775 * metadata. If NULL, write routine is not
1777 * @param[in] judgeInode filter function pointer. if not NULL, only
1778 * inodes for which this routine returns non-
1779 * zero will be written to the results file.
1780 * @param[in] singleVolumeNumber volume id filter
1781 * @param[out] forcep always set to 0 for namei impl
1782 * @param[in] forceR not used by namei impl
1783 * @param[in] wpath not used by namei impl
1784 * @param[in] rock opaque pointer passed to judgeInode
1786 * @return operation status
1788 * @retval -1 complete failure, salvage should terminate.
1789 * @retval -2 not enough space on partition, salvager has error message
1793 ListViceInodes(char *devname, char *mountedOn, FD_t inodeFile,
1794 int (*judgeInode) (struct ViceInodeInfo * info, VolumeId vid, void *rock),
1795 VolumeId singleVolumeNumber, int *forcep, int forceR, char *wpath,
1800 *forcep = 0; /* no need to salvage until further notice */
1802 /* Verify protections on directories. */
1804 VerifyDirPerms(mountedOn);
1807 namei_ListAFSFiles(mountedOn, WriteInodeInfo, inodeFile, judgeInode,
1808 singleVolumeNumber, rock);
1810 if (inodeFile == INVALID_FD)
1817 if (OS_SYNC(inodeFile) == -1) {
1818 Log("Unable to successfully fsync inode file for %s\n", mountedOn);
1823 * Paranoia: check that the file is really the right size
1825 if (OS_SIZE(inodeFile) != ninodes * sizeof(struct ViceInodeInfo)) {
1826 Log("Wrong size (%d instead of %lu) in inode file for %s\n",
1827 (int) OS_SIZE(inodeFile),
1828 (long unsigned int) ninodes * sizeof(struct ViceInodeInfo),
1837 * Collect all the matching AFS files on the drive.
1838 * If singleVolumeNumber is non-zero, just return files for that volume.
1840 * @param[in] dev vice partition path
1841 * @param[in] writeFun function pointer to a function which
1842 * writes inode information to FILE fp
1843 * @param[in] fp file stream where inode metadata is sent
1844 * @param[in] judgeFun filter function pointer. if not NULL,
1845 * only entries for which a non-zero value
1846 * is returned are written to fp
1847 * @param[in] singleVolumeNumber volume id filter. if nonzero, only
1848 * process files for that specific volume id
1849 * @param[in] rock opaque pointer passed into writeFun and
1852 * @return operation status
1854 * @retval >=0 number of matching files found
1857 namei_ListAFSFiles(char *dev,
1858 int (*writeFun) (FD_t, struct ViceInodeInfo *, char *,
1861 int (*judgeFun) (struct ViceInodeInfo *, VolumeId, void *),
1862 VolumeId singleVolumeNumber, void *rock)
1869 #ifndef AFS_NT40_ENV
1875 static void FreeZLCList(void);
1878 memset((void *)&ih, 0, sizeof(IHandle_t));
1880 ih.ih_dev = nt_DriveToDev(dev);
1882 ih.ih_dev = volutil_GetPartitionID(dev);
1885 if (singleVolumeNumber) {
1886 ih.ih_vid = singleVolumeNumber;
1887 namei_HandleToVolDir(&name, &ih);
1889 namei_ListAFSSubDirs(&ih, writeFun, fp, judgeFun,
1890 singleVolumeNumber, rock);
1894 /* Find all volume data directories and descend through them. */
1895 namei_HandleToInodeDir(&name, &ih);
1897 dirp1 = opendir(name.n_path);
1900 while ((dp1 = readdir(dirp1))) {
1902 /* Heirarchy is one level on Windows */
1903 if (!DecodeVolumeName(dp1->d_name, &ih.ih_vid)) {
1905 namei_ListAFSSubDirs(&ih, writeFun, fp, judgeFun,
1909 if (*dp1->d_name == '.')
1911 snprintf(path2, sizeof(path2), "%s" OS_DIRSEP "%s", name.n_path,
1913 dirp2 = opendir(path2);
1915 while ((dp2 = readdir(dirp2))) {
1916 if (*dp2->d_name == '.')
1918 if (!DecodeVolumeName(dp2->d_name, &ih.ih_vid)) {
1920 namei_ListAFSSubDirs(&ih, writeFun, fp, judgeFun,
1937 static void AddToZLCDeleteList(char dir, char *name);
1938 static void DeleteZLCFiles(char *path);
1942 * examine a namei volume special file.
1944 * @param[in] path1 volume special directory path
1945 * @param[in] dname directory entry name
1946 * @param[in] myIH inode handle to volume directory
1947 * @param[out] linkHandle namei link count fd handle. if
1948 * the inode in question is the link
1949 * table, then the FdHandle is populated
1950 * @param[in] writeFun metadata write function pointer
1951 * @param[in] fp file pointer where inode metadata
1952 * is written by (*writeFun)()
1953 * @param[in] judgeFun inode filter function pointer. if
1954 * not NULL, only inodes for which this
1955 * function returns non-zero are recorded
1956 * into fp by writeFun
1957 * @param[in] singleVolumeNumer volume id filter. if non-zero, only
1958 * inodes associated with this volume id
1959 * are recorded by writeFun
1960 * @param[in] rock opaque pointer passed to writeFun and
1963 * @return operation status
1964 * @retval 1 count this inode
1965 * @retval 0 don't count this inode
1966 * @retval -1 failure
1971 _namei_examine_special(char * path1,
1974 FdHandle_t * linkHandle,
1975 int (*writeFun) (FD_t, struct ViceInodeInfo *, char *,
1978 int (*judgeFun) (struct ViceInodeInfo *, VolumeId, void *),
1979 VolumeId singleVolumeNumber,
1983 struct ViceInodeInfo info;
1985 if (DecodeInode(path1, dname, &info, myIH) < 0) {
1990 if (info.u.param[2] != VI_LINKTABLE) {
1992 } else if (info.u.param[0] != myIH->ih_vid) {
1993 /* VGID encoded in linktable filename and/or OGM data isn't
1994 * consistent with VGID encoded in namei path */
1995 Log("namei_ListAFSSubDirs: warning: inconsistent linktable "
1996 "filename \"%s" OS_DIRSEP "%s\"; salvager will delete it "
1997 "(dir_vgid=%" AFS_VOLID_FMT ", inode_vgid=%" AFS_VOLID_FMT ")\n",
1998 path1, dname, afs_printable_VolumeId_lu(myIH->ih_vid),
1999 afs_printable_VolumeId_lu(info.u.param[0]));
2000 /* We need to set the linkCount to _something_, so linkCount
2001 * doesn't just contain stack garbage. Set it to 0, so in case
2002 * the salvager or whatever our caller is does try to process
2003 * this like a normal file, we won't try to INC or DEC it. */
2007 /* Open this handle */
2008 snprintf(path2, sizeof(path2),
2009 "%s" OS_DIRSEP "%s", path1, dname);
2010 linkHandle->fd_fd = OS_OPEN(path2, Testing ? O_RDONLY : O_RDWR, 0666);
2012 namei_GetLinkCount(linkHandle, (Inode) 0, 1, 1, Testing);
2016 (*judgeFun) (&info, singleVolumeNumber, rock)) {
2017 ret = (*writeFun) (fp, &info, path1, dname);
2019 Log("_namei_examine_special: writeFun returned %d\n", ret);
2031 * examine a namei file.
2033 * @param[in] path3 volume special directory path
2034 * @param[in] dname directory entry name
2035 * @param[in] myIH inode handle to volume directory
2036 * @param[in] linkHandle namei link count fd handle.
2037 * @param[in] writeFun metadata write function pointer
2038 * @param[in] fp file pointer where inode metadata
2039 * is written by (*writeFun)()
2040 * @param[in] judgeFun inode filter function pointer. if
2041 * not NULL, only inodes for which this
2042 * function returns non-zero are recorded
2043 * into fp by writeFun
2044 * @param[in] singleVolumeNumer volume id filter. if non-zero, only
2045 * inodes associated with this volume id
2046 * are recorded by writeFun
2047 * @param[in] rock opaque pointer passed to writeFun and
2050 * @return operation status
2051 * @retval 1 count this inode
2052 * @retval 0 don't count this inode
2053 * @retval -1 failure
2054 * @retval -2 request ZLC delete
2059 _namei_examine_reg(char * path3,
2062 FdHandle_t * linkHandle,
2063 int (*writeFun) (FD_t, struct ViceInodeInfo *, char *,
2066 int (*judgeFun) (struct ViceInodeInfo *, VolumeId, void *),
2067 VolumeId singleVolumeNumber,
2071 struct ViceInodeInfo info;
2073 int dirl; /* Windows-only (one level hash dir) */
2076 if (DecodeInode(path3, dname, &info, myIH) < 0) {
2081 namei_GetLinkCount(linkHandle,
2082 info.inodeNumber, 1, 1, Testing);
2083 if (info.linkCount == 0) {
2085 Log("Found 0 link count file %s" OS_DIRSEP "%s, deleting it.\n", path3, dname);
2086 #ifdef AFS_SALSRV_ENV
2087 /* defer -- the AddToZLCDeleteList() interface is not MT-safe */
2089 #else /* !AFS_SALSRV_ENV */
2090 dirl = path3[strlen(path3)-1];
2091 AddToZLCDeleteList((char)dirl, dname);
2092 #endif /* !AFS_SALSRV_ENV */
2093 #else /* !DELETE_ZLC */
2094 Log("Found 0 link count file %s" OS_DIRSEP "%s.\n", path3,
2101 (*judgeFun) (&info, singleVolumeNumber, rock)) {
2102 ret = (*writeFun) (fp, &info, path3, dname);
2104 Log("_namei_examine_reg: writeFun returned %d\n", ret);
2116 * listsubdirs work queue node.
2118 struct listsubdirs_work_node {
2119 #ifdef AFS_SALSRV_ENV
2120 int *error; /**< *error set if an error was
2121 * encountered in any listsubdirs
2125 IHandle_t * IH; /**< volume directory handle */
2126 FdHandle_t *linkHandle; /**< namei link count fd handle. when
2127 * examinining the link table special
2128 * inode, this will be pointed at the
2131 FD_t fp; /**< file pointer for writeFun */
2133 /** function which will write inode metadata to fp */
2134 int (*writeFun) (FD_t, struct ViceInodeInfo *, char *, char *);
2136 /** inode filter function */
2137 int (*judgeFun) (struct ViceInodeInfo *, VolumeId, void *);
2138 VolumeId singleVolumeNumber; /**< volume id filter */
2139 void * rock; /**< pointer passed to writeFun and judgeFun */
2140 int code; /**< return code from examine function */
2141 int special; /**< asserted when this is a volume
2146 * simple wrapper around _namei_examine_special and _namei_examine_reg.
2148 * @param[in] work the struct listsubdirs_work_node for the associated
2149 * "list subdirs" job
2150 * @param[in] dir the directory to examine
2151 * @param[in] filename the filename in 'dir' to examine
2153 * @return operation status
2154 * @retval 1 count this inode
2155 * @retval 0 don't count this inode
2156 * @retval -1 failure
2159 _namei_examine_file(const struct listsubdirs_work_node *work, char *dir,
2162 if (work->special) {
2163 return _namei_examine_special(dir, filename, work->IH,
2164 work->linkHandle, work->writeFun, work->fp,
2165 work->judgeFun, work->singleVolumeNumber,
2168 return _namei_examine_reg(dir, filename, work->IH,
2169 work->linkHandle, work->writeFun, work->fp,
2170 work->judgeFun, work->singleVolumeNumber,
2176 #ifdef AFS_SALSRV_ENV
2177 /** @addtogroup afs_vol_salsrv_pario */
2181 * arguments for the _namei_examine_file_cbk callback function.
2183 struct listsubdirs_args {
2184 const struct listsubdirs_work_node *work; /**< arguments that are the same
2185 * for all invocations of
2186 * _namei_examine_file_cbk, in
2188 int *result; /**< where we can store the return code of _namei_examine_file */
2190 char dir[512]; /**< directory to examine */
2191 char filename[256]; /**< filename in 'dir' to examine */
2195 * a node in the list of results of listsubdir jobs.
2197 struct listsubdirs_result {
2199 int inodes; /**< return value from _namei_examine_file */
2203 * clean up a list of 'struct listsubdirs_result's and interpret the results.
2205 * @param[in] resultlist a list of 'struct listsubdirs_result's
2207 * @return number of inodes found
2211 _namei_listsubdirs_cleanup_results(struct rx_queue *resultlist)
2213 struct listsubdirs_result *res, *nres;
2216 for(queue_Scan(resultlist, res, nres, listsubdirs_result)) {
2218 /* noop, retain erroneous error code */
2219 } else if (res->inodes < 0) {
2234 * wait for the spawned listsubdirs jobs to finish, and return how many inodes
2237 * @param[in] queue queue to wait to finish
2238 * @param[in] resultlist list of 'struct listsubdirs_result's that the queued
2239 * jobs are storing their results in
2241 * @return number of inodes found
2245 _namei_listsubdirs_wait(struct afs_work_queue *queue, struct rx_queue *resultlist)
2249 code = afs_wq_wait_all(queue);
2254 return _namei_listsubdirs_cleanup_results(resultlist);
2258 * work queue entry point for examining namei files.
2260 * @param[in] queue pointer to struct Vwork_queue
2261 * @param[in] node pointer to struct Vwork_queue_node
2262 * @param[in] queue_rock opaque pointer to struct salsrv_pool_state
2263 * @param[in] node_rock opaque pointer to struct listsubdirs_work_node
2264 * @param[in] caller_rock opaque pointer to struct salsrv_worker_thread_state
2266 * @return operation status
2268 * @see Vwork_queue_callback_func_t
2273 _namei_examine_file_cbk(struct afs_work_queue *queue,
2274 struct afs_work_queue_node *node,
2280 struct listsubdirs_args *args = node_rock;
2281 const struct listsubdirs_work_node * work = args->work;
2282 char *dir = args->dir;
2283 char *filename = args->filename;
2285 code = _namei_examine_file(work, dir, filename);
2287 *(args->result) = code;
2291 /* we've errored, so no point in letting more jobs continue */
2292 afs_wq_shutdown(queue);
2298 static pthread_once_t wq_once = PTHREAD_ONCE_INIT;
2299 static pthread_key_t wq_key;
2302 * create the wq_key key for storing a work queue.
2305 _namei_wq_keycreate(void)
2307 opr_Verify(pthread_key_create(&wq_key, NULL) == 0);
2311 * set the work queue for this thread to use for backgrounding namei ops.
2313 * The work queue will be used in ListAFSSubdirs (called indirectly by
2314 * ListViceInodes) to examine files in parallel.
2316 * @param[in] wq the work queue to use
2319 namei_SetWorkQueue(struct afs_work_queue *wq)
2321 opr_Verify(pthread_once(&wq_once, _namei_wq_keycreate) == 0);
2323 opr_Verify(pthread_setspecific(wq_key, wq) == 0);
2327 * enqueue an examine file work unit.
2329 * @param[in] work the _namei_examine_file arguments that are common to
2330 * all callers within the same ListAFSFiles operation
2331 * @param[in] dir the specific directory to look at (string will be
2332 * copied; can be stack/temporary memory)
2333 * @param[in] filename the filename to look at (string will be copied; can be
2334 * stack/temporary memory)
2335 * @param[in] wq work queue to enqueue this work unit to
2336 * @param[in] resultlist the list to append the 'struct listsubdirs_result' to
2337 * for the enqueued work unit
2339 * @return operation status
2341 * @retval -1 fatal error
2343 * @note errors MUST be indicated by a -1 error code and nothing else, to be
2344 * compatible with _namei_examine_reg and _namei_examine_special
2349 _namei_examine_file_spawn(const struct listsubdirs_work_node *work,
2350 const char *dir, const char *filename,
2351 struct afs_work_queue *wq,
2352 struct rx_queue *resultlist)
2355 struct listsubdirs_args *args = NULL;
2356 struct listsubdirs_result *result = NULL;
2357 struct afs_work_queue_node *node = NULL;
2358 struct afs_work_queue_add_opts opts;
2360 args = malloc(sizeof(*args));
2366 result = malloc(sizeof(*result));
2367 if (result == NULL) {
2372 code = afs_wq_node_alloc(&node);
2377 code = afs_wq_node_set_detached(node);
2384 args->result = &result->inodes;
2385 strlcpy(args->dir, dir, sizeof(args->dir));
2386 strlcpy(args->filename, filename, sizeof(args->filename));
2388 code = afs_wq_node_set_callback(node,
2389 &_namei_examine_file_cbk,
2397 afs_wq_add_opts_init(&opts);
2400 code = afs_wq_add(wq, node, &opts);
2407 queue_Append(resultlist, result);
2412 afs_wq_node_put(node);
2428 #else /* !AFS_SALSRV_ENV */
2429 # define _namei_examine_file_spawn(work, dir, file, wq, resultlist) \
2430 _namei_examine_file(work, dir, file)
2431 #endif /* !AFS_SALSRV_ENV */
2434 * traverse and check inodes.
2436 * @param[in] dirIH volume group directory handle
2437 * @param[in] writeFun function pointer which will write inode
2438 * metadata to FILE stream fp
2439 * @param[in] fp file stream where inode metadata gets
2441 * @param[in] judgeFun inode filter function. if not NULL, only
2442 * inodes for which the filter returns non-zero
2443 * will be written out by writeFun
2444 * @param[in] singleVolumeNumber volume id filter. only inodes matching this
2445 * filter are written out by writeFun
2446 * @param[in] rock opaque pointer passed to judgeFun and writeFun
2448 * @return operation status
2450 * @retval >=0 number of matching inodes found
2452 * @todo the salsrv implementation may consume a lot of
2453 * memory for a large volume. at some point we should
2454 * probably write a background thread to asynchronously
2455 * clean up the resultlist nodes to reduce memory footprint
2460 namei_ListAFSSubDirs(IHandle_t * dirIH,
2461 int (*writeFun) (FD_t, struct ViceInodeInfo *, char *,
2464 int (*judgeFun) (struct ViceInodeInfo *, VolumeId, void *),
2465 VolumeId singleVolumeNumber, void *rock)
2467 int code = 0, ret = 0;
2468 IHandle_t myIH = *dirIH;
2470 char path1[512], path3[512];
2472 #ifndef AFS_NT40_ENV
2477 struct dirent *dp1, *dp3;
2478 FdHandle_t linkHandle;
2480 struct listsubdirs_work_node work;
2481 #ifdef AFS_SALSRV_ENV
2483 struct afs_work_queue *wq;
2485 struct rx_queue resultlist;
2488 namei_HandleToVolDir(&name, &myIH);
2489 strlcpy(path1, name.n_path, sizeof(path1));
2491 /* Do the directory containing the special files first to pick up link
2494 (void)strcat(path1, OS_DIRSEP);
2495 (void)strcat(path1, NAMEI_SPECDIR);
2497 linkHandle.fd_fd = INVALID_FD;
2498 #ifdef AFS_SALSRV_ENV
2499 opr_Verify(pthread_once(&wq_once, _namei_wq_keycreate) == 0);
2501 wq = pthread_getspecific(wq_key);
2507 queue_Init(&resultlist);
2510 memset(&work, 0, sizeof(work));
2511 work.linkHandle = &linkHandle;
2514 work.writeFun = writeFun;
2515 work.judgeFun = judgeFun;
2516 work.singleVolumeNumber = singleVolumeNumber;
2519 #ifdef AFS_SALSRV_ENV
2520 work.error = &error;
2523 dirp1 = opendir(path1);
2525 while ((dp1 = readdir(dirp1))) {
2526 if (*dp1->d_name == '.')
2529 #ifdef AFS_SALSRV_ENV
2535 #endif /* AFS_SALSRV_ENV */
2537 code = _namei_examine_file_spawn(&work, path1, dp1->d_name, wq, &resultlist);
2547 /* count this inode */
2548 #ifndef AFS_SALSRV_ENV
2557 #ifdef AFS_SALSRV_ENV
2558 /* effectively a barrier */
2559 code = _namei_listsubdirs_wait(wq, &resultlist);
2560 if (code < 0 || error) {
2568 if (linkHandle.fd_fd == INVALID_FD) {
2569 Log("namei_ListAFSSubDirs: warning: VG %" AFS_VOLID_FMT " does not have a link table; "
2570 "salvager will recreate it.\n", afs_printable_VolumeId_lu(dirIH->ih_vid));
2573 /* Now run through all the other subdirs */
2574 namei_HandleToVolDir(&name, &myIH);
2575 strlcpy(path1, name.n_path, sizeof(path1));
2579 dirp1 = opendir(path1);
2581 while ((dp1 = readdir(dirp1))) {
2582 #ifndef AFS_NT40_ENV
2583 if (*dp1->d_name == '.')
2586 if (!strcmp(dp1->d_name, NAMEI_SPECDIR))
2589 #ifndef AFS_NT40_ENV /* This level missing on Windows */
2590 /* Now we've got a next level subdir. */
2591 snprintf(path2, sizeof(path2), "%s" OS_DIRSEP "%s",
2592 path1, dp1->d_name);
2593 dirp2 = opendir(path2);
2595 while ((dp2 = readdir(dirp2))) {
2596 if (*dp2->d_name == '.')
2599 /* Now we've got to the actual data */
2600 snprintf(path3, sizeof(path3), "%s" OS_DIRSEP "%s",
2601 path2, dp2->d_name);
2603 /* Now we've got to the actual data */
2604 snprintf(path3, sizeof(path3), "%s" OS_DIRSEP "%s",
2605 path1, dp1->d_name);
2607 dirp3 = opendir(path3);
2609 while ((dp3 = readdir(dirp3))) {
2610 #ifndef AFS_NT40_ENV
2611 if (*dp3->d_name == '.')
2615 #ifdef AFS_SALSRV_ENV
2618 #ifndef AFS_NT40_ENV
2625 #endif /* AFS_SALSRV_ENV */
2627 code = _namei_examine_file_spawn(&work, path3,
2634 #ifndef AFS_NT40_ENV
2642 #ifndef AFS_SALSRV_ENV
2650 #ifndef AFS_NT40_ENV /* This level missing on Windows */
2659 #ifdef AFS_SALSRV_ENV
2660 /* effectively a barrier */
2661 code = _namei_listsubdirs_wait(wq, &resultlist);
2662 if (code < 0 || error) {
2672 /* Then why does this directory exist? Blow it away. */
2673 namei_HandleToVolDir(&name, dirIH);
2674 namei_RemoveDataDirectories(&name);
2678 #ifdef AFS_SALSRV_ENV
2680 afs_wq_wait_all(wq);
2682 _namei_listsubdirs_cleanup_results(&resultlist);
2684 if (linkHandle.fd_fd != INVALID_FD)
2685 OS_CLOSE(linkHandle.fd_fd);
2697 DecodeVolumeName(char *name, VolumeId *vid)
2699 /* Name begins with "Vol_" and ends with .data. See nt_HandleToVolDir() */
2706 if (strncmp(name, "Vol_", 4))
2708 if (strcmp(name + len - 5, ".data"))
2711 stmp[len - 5] = '\0';
2712 *vid = base32_to_int(stmp + 4);
2717 DecodeVolumeName(char *name, VolumeId *vid)
2719 if (strlen(name) < 1)
2721 *vid = (unsigned int)flipbase64_to_int64(name);
2729 * Get the inode number from the name.
2734 DecodeInode(char *dpath, char *name, struct ViceInodeInfo *info,
2739 WIN32_FIND_DATA data;
2743 FdHandle_t linkHandle;
2745 VolumeId volid = myIH->ih_vid;
2747 snprintf(fpath, sizeof(fpath), "%s" OS_DIRSEP "%s", dpath, name);
2749 dirH = FindFirstFileEx(fpath, FindExInfoStandard, &data,
2750 FindExSearchNameMatch, NULL,
2751 FIND_FIRST_EX_CASE_SENSITIVE);
2752 if (dirH == INVALID_HANDLE_VALUE)
2755 (void)strcpy(stmp, name);
2756 s = strrchr(stmp, '_');
2760 t = strrchr(s, '.');
2765 vno = base32_to_int(s); /* type for special files */
2766 tag = base32_to_int(t+1);
2767 info->inodeNumber = ((Inode) tag) << NAMEI_TAGSHIFT;
2768 info->inodeNumber |= vno;
2769 info->byteCount = data.nFileSizeLow;
2771 dirl = dpath[strlen(dpath)-1];
2772 if (dirl == NAMEI_SPECDIRC) { /* Special inode. */
2773 info->inodeNumber |= NAMEI_INODESPECIAL;
2774 info->u.param[0] = data.ftCreationTime.dwHighDateTime;
2775 info->u.param[1] = data.ftCreationTime.dwLowDateTime;
2776 info->u.param[2] = vno; /* type */
2777 info->u.param[3] = volid;
2778 if (vno != VI_LINKTABLE)
2779 info->linkCount = 1;
2781 /* Open this handle */
2783 (void)sprintf(lpath, "%s" OS_DIRSEP "%s", fpath, data.cFileName);
2784 linkHandle.fd_fd = OS_OPEN(lpath, O_RDONLY, 0666);
2786 namei_GetLinkCount(&linkHandle, (Inode) 0, 0, 0, 0);
2790 namei_GetLinkCount(&linkHandle, info->inodeNumber, 0, 0, 0);
2791 if (info->linkCount == 0) {
2793 Log("Found 0 link count file %s" OS_DIRSEP "%s, deleting it.\n",
2794 fpath, data.cFileName);
2795 AddToZLCDeleteList(dirl, data.cFileName);
2797 Log("Found 0 link count file %s" OS_DIRSEP "%s.\n", path,
2801 info->u.param[2] = data.ftCreationTime.dwHighDateTime;
2802 info->u.param[3] = data.ftCreationTime.dwLowDateTime;
2803 info->u.param[1] = vno;
2804 info->u.param[0] = volid;
2811 DecodeInode(char *dpath, char *name, struct ViceInodeInfo *info,
2815 struct afs_stat_st status;
2816 struct afs_stat_st checkstatus;
2818 lb64_string_t check;
2819 VolumeId volid = myIH->ih_vid;
2823 snprintf(fpath, sizeof(fpath), "%s" OS_DIRSEP "%s", dpath, name);
2825 if (afs_stat(fpath, &status) < 0) {
2829 info->byteCount = status.st_size;
2830 info->inodeNumber = (Inode) flipbase64_to_int64(name);
2832 int64_to_flipbase64(check, info->inodeNumber);
2833 if (strcmp(name, check))
2836 /* Check if the _full_ path is correct, to ensure we can actually open this
2837 * file later. Otherwise, the salvager can choke. */
2838 memset(&tmpih, 0, sizeof(tmpih));
2839 tmpih.ih_dev = myIH->ih_dev;
2840 tmpih.ih_vid = myIH->ih_vid;
2841 tmpih.ih_ino = info->inodeNumber;
2842 namei_HandleToName(&nameiname, &tmpih);
2843 if ((afs_stat(nameiname.n_path, &checkstatus) < 0) ||
2844 checkstatus.st_ino != status.st_ino ||
2845 checkstatus.st_size != status.st_size) {
2847 /* log something for this case, since this means the filename looks
2848 * like a valid inode, but it's just in the wrong place. That's pretty
2853 Log(" Seemingly-misplaced files have been found, which I am\n");
2854 Log(" ignoring for now. If you cannot salvage the relevant volume,\n");
2855 Log(" you may try manually moving them to their correct location.\n");
2856 Log(" If the relevant volume seems fine, and these files do not\n");
2857 Log(" appear to contain important data, you can probably manually\n");
2858 Log(" delete them, or leave them alone. Contact your local OpenAFS\n");
2859 Log(" expert if you are unsure.\n");
2861 Log("Ignoring misplaced file in volume group %u: %s (should be %s)\n",
2862 (unsigned)myIH->ih_vid, fpath, nameiname.n_path);
2866 if ((info->inodeNumber & NAMEI_INODESPECIAL) == NAMEI_INODESPECIAL) {
2867 parm = ((info->inodeNumber >> NAMEI_UNIQSHIFT) & NAMEI_UNIQMASK);
2868 tag = (int)((info->inodeNumber >> NAMEI_TAGSHIFT) & NAMEI_TAGMASK);
2870 /* p1 - vid, p2 - -1, p3 - type, p4 - rwvid */
2871 info->u.param[0] = parm;
2872 info->u.param[1] = -1;
2873 info->u.param[2] = tag;
2874 info->u.param[3] = volid;
2876 GetOGMFromStat(&status, &parm, &tag);
2877 /* p1 - vid, p2 - vno, p3 - uniq, p4 - dv */
2878 info->u.param[0] = volid;
2879 info->u.param[1] = (int)(info->inodeNumber & NAMEI_VNODEMASK);
2880 info->u.param[2] = (int)((info->inodeNumber >> NAMEI_UNIQSHIFT)
2881 & (Inode) NAMEI_UNIQMASK);
2882 info->u.param[3] = parm;
2889 * Convert the VolumeInfo file from RO to RW
2890 * this routine is called by namei_convertROtoRWvolume()
2893 #ifdef FSSYNC_BUILD_CLIENT
2895 convertVolumeInfo(FD_t fdr, FD_t fdw, VolumeId vid)
2897 struct VolumeDiskData vd;
2900 if (OS_READ(fdr, &vd, sizeof(struct VolumeDiskData)) !=
2901 sizeof(struct VolumeDiskData)) {
2902 Log("1 convertVolumeInfo: read failed for %" AFS_VOLID_FMT " with code %d\n",
2903 afs_printable_VolumeId_lu(vid),
2907 vd.restoredFromId = vd.id; /* remember the RO volume here */
2909 vd.id = vd.parentId;
2913 vd.uniquifier += 5000; /* just in case there are still file copies from
2914 * the old RW volume around */
2916 /* For ROs, the copyDate contains the time that the RO volume was actually
2917 * created, and the creationDate just contains the last time the RO was
2918 * copied from the RW data. So, make the new RW creationDate more accurate
2919 * by setting it to copyDate, if copyDate is older. Since, we know the
2920 * volume is at least as old as copyDate. */
2921 if (vd.copyDate < vd.creationDate) {
2922 vd.creationDate = vd.copyDate;
2924 /* If copyDate is newer, just make copyDate and creationDate the same,
2925 * for consistency with other RWs */
2926 vd.copyDate = vd.creationDate;
2929 p = strrchr(vd.name, '.');
2930 if (p && !strcmp(p, ".readonly")) {
2933 if (OS_WRITE(fdw, &vd, sizeof(struct VolumeDiskData)) !=
2934 sizeof(struct VolumeDiskData)) {
2935 Log("1 convertVolumeInfo: write failed for %lu with code %d\n",
2936 afs_printable_uint32_lu(vid),
2945 * Convert a RO-volume into a RW-volume
2947 * This function allows to recover very fast from the loss of a partition
2948 * from RO-copies if all RO-Copies exist on another partition.
2949 * Then these RO-volumes can be made to the new RW-volumes.
2950 * Backup of RW-volumes then consists in "vos release".
2952 * We must make sure in this partition exists only the RO-volume which
2953 * is typical for remote replicas.
2955 * Then the linktable is already ok,
2956 * the vnode files need to be renamed
2957 * the volinfo file needs to be replaced by another one with
2958 * slightly different contents and new name.
2959 * The volume header file of the RO-volume in the /vicep<x> directory
2960 * is destroyed by this call. A new header file for the RW-volume must
2961 * be created after return from this routine.
2965 namei_ConvertROtoRWvolume(char *pname, VolumeId volumeId)
2968 #ifdef FSSYNC_BUILD_CLIENT
2970 char dir_name[512], oldpath[512], newpath[512];
2985 struct DiskPartition64 *partP;
2986 struct ViceInodeInfo info;
2987 struct VolumeDiskHeader h;
2988 char *rwpart, *rwname;
2990 # ifdef AFS_DEMAND_ATTACH_FS
2992 # endif /* AFS_DEMAND_ATTACH_FS */
2994 for (partP = DiskPartitionList; partP && strcmp(partP->name, pname);
2995 partP = partP->next);
2997 Log("1 namei_ConvertROtoRWvolume: Couldn't find DiskPartition for %s\n", pname);
3002 # ifdef AFS_DEMAND_ATTACH_FS
3003 locktype = VVolLockType(V_VOLUPD, 1);
3004 code = VLockVolumeByIdNB(volumeId, partP, locktype);
3010 # endif /* AFS_DEMAND_ATTACH_FS */
3012 if (VReadVolumeDiskHeader(volumeId, partP, &h)) {
3013 Log("1 namei_ConvertROtoRWvolume: Couldn't read header for RO-volume %lu.\n",
3014 afs_printable_uint32_lu(volumeId));
3019 /* check for existing RW on any partition on this server; */
3020 /* if found, return EXDEV - invalid cross-device link */
3022 VGetVolumePath(&ec, h.parent, &rwpart, &rwname);
3024 Log("1 namei_ConvertROtoRWvolume: RW volume %lu already exists on server partition %s.\n",
3025 afs_printable_uint32_lu(h.parent), rwpart);
3032 FSYNC_VolOp(volumeId, pname, FSYNC_VOL_BREAKCBKS, 0, NULL);
3034 ino = namei_MakeSpecIno(h.parent, VI_LINKTABLE);
3035 IH_INIT(ih, partP->device, h.parent, ino);
3037 namei_HandleToName(&n, ih);
3038 strlcpy(dir_name, n.n_path, sizeof(dir_name));
3039 p = strrchr(dir_name, OS_DIRSEPC);
3041 dirp = opendir(dir_name);
3043 Log("1 namei_ConvertROtoRWvolume: Could not opendir(%s)\n", dir_name);
3048 while ((dp = readdir(dirp))) {
3049 /* struct ViceInodeInfo info; */
3050 #ifndef AFS_NT40_ENV
3051 if (*dp->d_name == '.')
3054 if (DecodeInode(dir_name, dp->d_name, &info, ih) < 0) {
3055 Log("1 namei_ConvertROtoRWvolume: DecodeInode failed for %s" OS_DIRSEP "%s\n",
3056 dir_name, dp->d_name);
3061 if (info.u.param[1] != -1) {
3062 Log("1 namei_ConvertROtoRWvolume: found other than volume special file %s" OS_DIRSEP "%s\n", dir_name, dp->d_name);
3067 if (info.u.param[0] != volumeId) {
3068 if (info.u.param[0] == ih->ih_vid) {
3069 if (info.u.param[2] == VI_LINKTABLE) { /* link table */
3074 Log("1 namei_ConvertROtoRWvolume: found special file %s" OS_DIRSEP "%s"
3075 " for volume %lu\n", dir_name, dp->d_name,
3076 afs_printable_uint32_lu(info.u.param[0]));
3081 if (info.u.param[2] == VI_VOLINFO) { /* volume info file */
3082 strlcpy(infoName, dp->d_name, sizeof(infoName));
3084 } else if (info.u.param[2] == VI_SMALLINDEX) { /* small vnodes file */
3085 strlcpy(smallName, dp->d_name, sizeof(smallName));
3087 } else if (info.u.param[2] == VI_LARGEINDEX) { /* large vnodes file */
3088 strlcpy(largeName, dp->d_name, sizeof(largeName));
3092 Log("1 namei_ConvertROtoRWvolume: unknown type %u of special file found : %s" OS_DIRSEP "%s\n", info.u.param[2], dir_name, dp->d_name);
3099 if (!infoSeen || !smallSeen || !largeSeen || !linkSeen) {
3100 Log("1 namei_ConvertROtoRWvolume: not all special files found in %s\n", dir_name);
3106 * If we come here then there was only a RO-volume and we can safely
3110 memset(&t_ih, 0, sizeof(t_ih));
3111 t_ih.ih_dev = ih->ih_dev;
3112 t_ih.ih_vid = ih->ih_vid;
3114 snprintf(oldpath, sizeof oldpath, "%s" OS_DIRSEP "%s", dir_name,
3116 fd = OS_OPEN(oldpath, O_RDWR, 0);
3117 if (fd == INVALID_FD) {
3118 Log("1 namei_ConvertROtoRWvolume: could not open RO info file: %s\n",
3123 t_ih.ih_ino = namei_MakeSpecIno(ih->ih_vid, VI_VOLINFO);
3124 namei_HandleToName(&n, &t_ih);
3125 fd2 = OS_OPEN(n.n_path, O_CREAT | O_EXCL | O_RDWR, 0);
3126 if (fd2 == INVALID_FD) {
3127 Log("1 namei_ConvertROtoRWvolume: could not create RW info file: %s\n", n.n_path);
3132 code = convertVolumeInfo(fd, fd2, ih->ih_vid);
3136 OS_UNLINK(n.n_path);
3140 SetOGM(fd2, ih->ih_vid, 1);
3143 t_ih.ih_ino = namei_MakeSpecIno(ih->ih_vid, VI_SMALLINDEX);
3144 namei_HandleToName(&n, &t_ih);
3145 snprintf(newpath, sizeof newpath, "%s" OS_DIRSEP "%s", dir_name,
3147 fd = OS_OPEN(newpath, O_RDWR, 0);
3148 if (fd == INVALID_FD) {
3149 Log("1 namei_ConvertROtoRWvolume: could not open SmallIndex file: %s\n", newpath);
3153 SetOGM(fd, ih->ih_vid, 2);
3156 MoveFileEx(n.n_path, newpath, MOVEFILE_WRITE_THROUGH);
3158 link(newpath, n.n_path);
3162 t_ih.ih_ino = namei_MakeSpecIno(ih->ih_vid, VI_LARGEINDEX);
3163 namei_HandleToName(&n, &t_ih);
3164 snprintf(newpath, sizeof newpath, "%s" OS_DIRSEP "%s", dir_name,
3166 fd = OS_OPEN(newpath, O_RDWR, 0);
3167 if (fd == INVALID_FD) {
3168 Log("1 namei_ConvertROtoRWvolume: could not open LargeIndex file: %s\n", newpath);
3172 SetOGM(fd, ih->ih_vid, 3);
3175 MoveFileEx(n.n_path, newpath, MOVEFILE_WRITE_THROUGH);
3177 link(newpath, n.n_path);
3184 h.volumeInfo_hi = h.id;
3185 h.smallVnodeIndex_hi = h.id;
3186 h.largeVnodeIndex_hi = h.id;
3187 h.linkTable_hi = h.id;
3189 if (VCreateVolumeDiskHeader(&h, partP)) {
3190 Log("1 namei_ConvertROtoRWvolume: Couldn't write header for RW-volume %lu\n",
3191 afs_printable_uint32_lu(h.id));
3196 if (VDestroyVolumeDiskHeader(partP, volumeId, h.parent)) {
3197 Log("1 namei_ConvertROtoRWvolume: Couldn't unlink header for RO-volume %lu\n",
3198 afs_printable_uint32_lu(volumeId));
3201 FSYNC_VolOp(volumeId, pname, FSYNC_VOL_DONE, 0, NULL);
3202 FSYNC_VolOp(h.id, pname, FSYNC_VOL_ON, 0, NULL);
3205 # ifdef AFS_DEMAND_ATTACH_FS
3207 VUnlockVolumeById(volumeId, partP);
3209 # endif /* AFS_DEMAND_ATTACH_FS */
3217 * returns a static string used to print either 32 or 64 bit inode numbers.
3220 PrintInode(char *s, Inode ino)
3222 static afs_ino_str_t result;
3226 snprintf(s, sizeof(afs_ino_str_t), "%llu", (afs_uintmax_t) ino);
3233 /* Routines to facilitate removing zero link count files. */
3234 #define MAX_ZLC_NAMES 32
3235 #define MAX_ZLC_NAMELEN 16
3236 typedef struct zlcList_s {
3237 struct zlcList_s *zlc_next;
3239 char zlc_names[MAX_ZLC_NAMES][MAX_ZLC_NAMELEN];
3242 static zlcList_t *zlcAnchor = NULL;
3243 static zlcList_t *zlcCur = NULL;
3246 AddToZLCDeleteList(char dir, char *name)
3248 opr_Assert(strlen(name) <= MAX_ZLC_NAMELEN - 3);
3250 if (!zlcCur || zlcCur->zlc_n >= MAX_ZLC_NAMES) {
3251 if (zlcCur && zlcCur->zlc_next)
3252 zlcCur = zlcCur->zlc_next;
3254 zlcList_t *tmp = malloc(sizeof(zlcList_t));
3260 zlcCur->zlc_next = tmp;
3264 zlcCur->zlc_next = NULL;
3269 (void)sprintf(zlcCur->zlc_names[zlcCur->zlc_n], "%c" OS_DIRSEP "%s", dir, name);
3271 (void)sprintf(zlcCur->zlc_names[zlcCur->zlc_n], "%s", name);
3277 DeleteZLCFiles(char *path)
3283 for (z = zlcAnchor; z; z = z->zlc_next) {
3284 for (i = 0; i < z->zlc_n; i++) {
3286 (void)sprintf(fname, "%s" OS_DIRSEP "%s", path, z->zlc_names[i]);
3288 (void)sprintf(fname, "%s", z->zlc_names[i]);
3289 if (namei_unlink(fname) < 0) {
3290 Log("ZLC: Can't unlink %s, dos error = %d\n", fname,
3294 z->zlc_n = 0; /* Can reuse space. */
3307 tnext = i->zlc_next;
3311 zlcCur = zlcAnchor = NULL;
3315 #endif /* AFS_NAMEI_ENV */