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>
73 static void namei_UnlockLinkCount(FdHandle_t * fdP, Inode ino);
76 namei_iread(IHandle_t * h, afs_foff_t offset, char *buf, afs_fsize_t size)
85 nBytes = FDH_PREAD(fdP, buf, size, offset);
94 namei_iwrite(IHandle_t * h, afs_foff_t offset, char *buf, afs_fsize_t size)
103 nBytes = FDH_PWRITE(fdP, buf, size, offset);
105 FDH_REALLYCLOSE(fdP);
112 /* Inode number format:
113 * low 32 bits - if a regular file or directory, the vnode; else the type.
114 * 32-36 - uniquifier tag and index into counts array for this vnode. Only
115 * two of the available bits are currently used. The rest are
116 * present in case we ever increase the number of types of volumes
117 * in the volume group.
118 * bit 37 : 1 == special, 0 == regular
120 # define NAMEI_VNODEMASK 0x00ffffffff
121 # define NAMEI_TAGSHIFT 32
122 # define NAMEI_INODESPECIAL 0x2000000000
123 # define NAMEI_SPECDIR "R"
124 # define NAMEI_SPECDIRC 'R'
125 #else /* !AFS_NT40_ENV */
126 /* Inode number format:
127 * low 26 bits - vnode number - all 1's if volume special file.
129 * next 3 bits spare (0's)
130 * high 32 bits - uniquifier (regular) or type if spare
132 # define NAMEI_VNODEMASK 0x003ffffff
133 # define NAMEI_TAGSHIFT 26
134 # define NAMEI_UNIQMASK 0xffffffff
135 # define NAMEI_UNIQSHIFT 32
136 # define NAMEI_INODESPECIAL ((Inode)NAMEI_VNODEMASK)
137 /* dir1 is the high 8 bits of the 26 bit vnode */
138 # define VNO_DIR1(vno) ((vno >> 14) & 0xff)
139 /* dir2 is the next 9 bits */
140 # define VNO_DIR2(vno) ((vno >> 9) & 0x1ff)
141 /* "name" is the low 9 bits of the vnode, the 3 bit tag and the uniq */
142 # define NAMEI_SPECDIR "special"
143 #endif /* !AFS_NT40_ENV */
144 #define NAMEI_TAGMASK 0x7
145 #define NAMEI_VNODESPECIAL NAMEI_VNODEMASK
147 #define NAMEI_SPECDIRLEN (sizeof(NAMEI_SPECDIR)-1)
149 #define NAMEI_MAXVOLS 5 /* Maximum supported number of volumes per volume
150 * group, not counting temporary (move) volumes.
151 * This is the number of separate files, all having
152 * the same vnode number, which can occur in a volume
164 static int GetFreeTag(IHandle_t * ih, int vno);
166 /* namei_HandleToInodeDir
168 * Construct the path name of the directory holding the inode data.
169 * Format: /<vicepx>/INODEDIR
174 namei_HandleToInodeDir(namei_t * name, IHandle_t * ih)
176 memset(name, '\0', sizeof(*name));
177 nt_DevToDrive(name->n_drive, ih->ih_dev);
178 strlcpy(name->n_path, name->n_drive, sizeof(name->n_path));
182 /* Format: /<vicepx>/INODEDIR */
183 #define PNAME_BLEN 64
185 namei_HandleToInodeDir(namei_t * name, IHandle_t * ih)
189 memset(name, '\0', sizeof(*name));
192 * Add the /vicepXX string to the start of name->n_base and then calculate
193 * offset as the number of bytes we know we added.
195 * FIXME: This embeds knowledge of the vice partition naming scheme and
196 * mapping from device numbers. There needs to be an API that tells us
199 volutil_PartitionName_r(ih->ih_dev, name->n_base, sizeof(name->n_base));
200 offset = VICE_PREFIX_SIZE + (ih->ih_dev > 25 ? 2 : 1);
201 name->n_base[offset] = OS_DIRSEPC;
203 strlcpy(name->n_base + offset, INODEDIR, sizeof(name->n_base) - offset);
204 strlcpy(name->n_path, name->n_base, sizeof(name->n_path));
208 #define addtoname(N, C) \
210 if ((N)->n_path[strlen((N)->n_path)-1] != OS_DIRSEPC) \
211 strlcat((N)->n_path, OS_DIRSEP, sizeof((N)->n_path)); \
212 strlcat((N)->n_path, (C), sizeof((N)->n_path)); \
218 namei_HandleToVolDir(namei_t * name, IHandle_t * ih)
220 /* X:\Vol_XXXXXXX.data */
224 namei_HandleToInodeDir(name, ih);
225 /* nt_drive added to name by namei_HandleToInodeDir() */
226 namep = name->n_voldir;
227 (void)memcpy(namep, "\\Vol_", 5);
229 (void)strcpy(namep, int_to_base32(str1, ih->ih_vid));
230 namep += strlen(namep);
231 memcpy(namep, ".data", 5);
234 addtoname(name, name->n_voldir);
238 namei_HandleToVolDir(namei_t * name, IHandle_t * ih)
242 namei_HandleToInodeDir(name, ih);
243 (void)int32_to_flipbase64(tmp, (int64_t) (ih->ih_vid & 0xff));
244 strlcpy(name->n_voldir1, tmp, sizeof(name->n_voldir1));
245 addtoname(name, name->n_voldir1);
246 (void)int32_to_flipbase64(tmp, (int64_t) ih->ih_vid);
247 strlcpy(name->n_voldir2, tmp, sizeof(name->n_voldir2));
248 addtoname(name, name->n_voldir2);
252 /* namei_HandleToName
254 * Constructs a file name for the fully qualified handle.
257 /* Note that special files end up in X:\Vol_XXXXXXX.data\R */
259 namei_HandleToName(namei_t * name, IHandle_t * ih)
261 int vno = (int)(ih->ih_ino & NAMEI_VNODEMASK);
262 int special = (ih->ih_ino & NAMEI_INODESPECIAL)?1:0;
263 int tag = (int)((ih->ih_ino >> NAMEI_TAGSHIFT) & NAMEI_TAGMASK);
266 namei_HandleToVolDir(name, ih);
269 name->n_dir[0] = NAMEI_SPECDIRC;
272 name->n_dir[0] = 'Q';
274 name->n_dir[0] = ((vno & 0x1f) >> 1) + 'A';
277 name->n_dir[1] = '\0';
278 addtoname(name, name->n_dir);
279 /* X:\Vol_XXXXXXX.data\X\V_XXXXXXX.XXX */
280 namep = name->n_inode;
281 (void)memcpy(namep, "\\V_", 3);
283 (void)strcpy(namep, int_to_base32(str1, vno));
284 namep += strlen(namep);
286 (void)strcpy(namep, int_to_base32(str1, tag));
287 namep += strlen(namep);
288 addtoname(name, name->n_inode);
291 /* Note that special files end up in /vicepX/InodeDir/Vxx/V*.data/special */
293 namei_HandleToName(namei_t * name, IHandle_t * ih)
295 int vno = (int)(ih->ih_ino & NAMEI_VNODEMASK);
298 namei_HandleToVolDir(name, ih);
300 if (vno == NAMEI_VNODESPECIAL) {
301 strlcpy(name->n_dir1, NAMEI_SPECDIR, sizeof(name->n_dir1));
302 addtoname(name, name->n_dir1);
303 name->n_dir2[0] = '\0';
305 (void)int32_to_flipbase64(str, VNO_DIR1(vno));
306 strlcpy(name->n_dir1, str, sizeof(name->n_dir1));
307 addtoname(name, name->n_dir1);
308 (void)int32_to_flipbase64(str, VNO_DIR2(vno));
309 strlcpy(name->n_dir2, str, sizeof(name->n_dir2));
310 addtoname(name, name->n_dir2);
312 (void)int64_to_flipbase64(str, (int64_t) ih->ih_ino);
313 strlcpy(name->n_inode, str, sizeof(name->n_inode));
314 addtoname(name, name->n_inode);
319 /* The following is a warning to tell sys-admins to not muck about in this
322 #define VICE_README "These files and directories are a part of the AFS \
323 namespace. Modifying them\nin any way will result in loss of AFS data,\n\
324 ownership and permissions included.\n"
326 namei_ViceREADME(char *partition)
331 /* Create the inode directory if we're starting for the first time */
332 snprintf(filename, sizeof filename, "%s" OS_DIRSEP "%s", partition,
334 mkdir(filename, 0700);
336 snprintf(filename, sizeof filename,
337 "%s" OS_DIRSEP "%s" OS_DIRSEP "README",
338 partition, INODEDIR);
339 fd = OS_OPEN(filename, O_WRONLY | O_CREAT | O_TRUNC, 0444);
340 if (fd != INVALID_FD) {
341 (void)OS_WRITE(fd, VICE_README, strlen(VICE_README));
348 /* namei_CreateDataDirectories
350 * If creating the file failed because of ENOENT or ENOTDIR, try
351 * creating all the directories first.
355 namei_CreateDataDirectories(namei_t * name, int *created)
362 snprintf(tmp, 256, "%s" OS_DIRSEP "%s", name->n_drive, name->n_voldir);
364 if (mkdir(tmp) < 0) {
375 for (i = 'A'; i <= NAMEI_SPECDIRC; i++) {
377 if (mkdir(tmp) < 0 && errno != EEXIST)
383 #define create_dir() \
385 if (mkdir(tmp, 0700)<0) { \
386 if (errno != EEXIST) \
394 #define create_nextdir(A) \
396 strcat(tmp, OS_DIRSEP); strcat(tmp, A); create_dir(); \
400 namei_CreateDataDirectories(namei_t * name, int *created)
406 strlcpy(tmp, name->n_base, sizeof(tmp));
409 create_nextdir(name->n_voldir1);
410 create_nextdir(name->n_voldir2);
411 create_nextdir(name->n_dir1);
412 if (name->n_dir2[0]) {
413 create_nextdir(name->n_dir2);
420 /* delTree(): Deletes an entire tree of directories (no files)
422 * root : Full path to the subtree. Should be big enough for PATH_MAX
423 * tree : the subtree to be deleted is rooted here. Specifies only the
424 * subtree beginning at tree (not the entire path). It should be
425 * a pointer into the "root" buffer.
427 * errp : errno of the first error encountered during the directory cleanup.
428 * *errp should have been initialized to 0.
431 * -1 : If errors were encountered during cleanup and error is set to
435 * If there are errors, we try to work around them and delete as many
436 * directories as possible. We don't attempt to remove directories that still
437 * have non-dir entries in them.
440 delTree(char *root, char *tree, int *errp)
448 /* delete the children first */
449 cp = strchr(tree, OS_DIRSEPC);
451 delTree(root, cp + 1, errp);
454 cp = tree + strlen(tree); /* move cp to the end of string tree */
456 /* now delete all entries in this dir */
457 if ((ds = opendir(root)) != NULL) {
459 while ((dirp = readdir(ds))) {
460 /* ignore . and .. */
461 if (!strcmp(dirp->d_name, ".") || !strcmp(dirp->d_name, ".."))
463 /* since root is big enough, we reuse the space to
464 * concatenate the dirname to the current tree
466 strcat(root, OS_DIRSEP);
467 strcat(root, dirp->d_name);
468 if (afs_stat(root, &st) == 0 && S_ISDIR(st.st_mode)) {
469 /* delete this subtree */
470 delTree(root, cp + 1, errp);
472 *errp = *errp ? *errp : errno;
474 /* recover path to our cur tree by truncating it to
479 /* if (!errno) -- closedir not implicit if we got an error */
483 /* finally axe the current dir */
485 *errp = *errp ? *errp : errno;
487 #ifndef AFS_PTHREAD_ENV /* let rx get some work done */
489 #endif /* !AFS_PTHREAD_ENV */
494 /* if we encountered errors during cleanup, we return a -1 */
503 /* namei_RemoveDataDirectories
505 * Returns 0 on success.
506 * Returns -1 on error. Typically, callers ignore this error because we
507 * can continue running if the removes fail. The salvage process will
508 * finish tidying up for us.
513 namei_RemoveDataDirectories(namei_t * name)
520 snprintf(tmp, 256, "%s" OS_DIRSEP "%s", name->n_drive, name->n_voldir);
523 path += strlen(path);
524 *path++ = OS_DIRSEPC;
526 for (i = 'A'; i <= NAMEI_SPECDIRC; i++) {
528 if (rmdir(name->n_path) < 0 && errno != ENOENT)
533 /* Delete the Vol_NNNNNN.data directory. */
536 if (rmdir(name->n_path) < 0 && errno != ENOENT) {
544 * We only use the n_base and n_voldir1 entries
545 * and only do rmdir's.
548 namei_RemoveDataDirectories(namei_t * name)
552 int prefixlen = strlen(name->n_base), err = 0;
553 int vollen = strlen(name->n_voldir1);
554 char pbuf[MAXPATHLEN];
558 strlcpy(path, name->n_path, sizeof(pbuf));
560 /* move past the prefix and n_voldir1 */
561 path = path + prefixlen + 1 + vollen + 1; /* skip over the trailing / */
563 /* now delete all dirs upto path */
564 code = delTree(pbuf, path, &err);
566 /* We've now deleted everything under /n_base/n_voldir1/n_voldir2 that
567 * we could. Do not delete /n_base/n_voldir1, since doing such might
568 * interrupt another thread trying to create a volume. We could introduce
569 * some locking to make this safe (or only remove it for whole-partition
570 * salvages), but by not deleting it we only leave behind a maximum of
571 * 256 empty directories. So at least for now, don't bother. */
576 /* Create the file in the name space.
578 * Parameters stored as follows:
580 * p1 - volid - implied in containing directory.
581 * p2 - vnode - name is <vno:31-23>/<vno:22-15>/<vno:15-0><uniq:31-5><tag:2-0>
582 * p3 - uniq -- bits 4-0 are in mode bits 4-0
583 * p4 - dv ---- dv:15-0 in uid, dv:29-16 in gid, dv:31-30 in mode:6-5
585 * p1 - volid - creation time - dwHighDateTime
586 * p2 - vnode - -1 means special, file goes in "S" subdirectory.
587 * p3 - type -- name is <type>.<tag> where tag is a file name unqiquifier.
588 * p4 - parid - parent volume id - implied in containing directory.
590 * Return value is the inode number or (Inode)-1 if error.
591 * We "know" there is only one link table, so return EEXIST if there already
592 * is a link table. It's up to the calling code to test errno and increment
598 * This function is called by VCreateVolume to hide the implementation
599 * details of the inode numbers. This only allows for 7 volume special
600 * types, but if we get that far, this could should be dead by then.
603 namei_MakeSpecIno(VolumeId volid, int type)
606 ino = NAMEI_INODESPECIAL;
609 /* tag is always 0 for special */
611 type &= NAMEI_TAGMASK;
612 ino |= ((Inode) type) << NAMEI_TAGSHIFT;
613 ino |= ((Inode) volid) << NAMEI_UNIQSHIFT;
621 SetOGM(FD_t fd, int parm, int tag)
627 SetWinOGM(FD_t fd, int p1, int p2)
632 ftime.dwHighDateTime = p1;
633 ftime.dwLowDateTime = p2;
635 code = SetFileTime(fd, &ftime, NULL /*access*/, NULL /*write*/);
642 GetWinOGM(FD_t fd, int *p1, int *p2)
647 code = GetFileTime(fd, &ftime, NULL /*access*/, NULL /*write*/);
651 *p1 = ftime.dwHighDateTime;
652 *p2 = ftime.dwLowDateTime;
658 CheckOGM(FdHandle_t *fdP, int p1)
662 if (GetWinOGM(fdP->fd_fd, &ogm_p1, &ogm_p2)) {
674 FixSpecialOGM(FdHandle_t *fdP, int check)
676 Inode ino = fdP->fd_ih->ih_ino;
677 VnodeId vno = NAMEI_VNODESPECIAL, ogm_vno;
680 if (GetWinOGM(fdP->fd_fd, &ogm_volid, &ogm_vno)) {
684 /* the only thing we can check is the vnode number; for the volid we have
685 * nothing else to compare against */
686 if (vno != ogm_vno) {
690 if (SetWinOGM(fdP->fd_fd, ogm_volid, vno)) {
697 #else /* AFS_NT40_ENV */
698 /* SetOGM - set owner group and mode bits from parm and tag */
700 SetOGM(FD_t fd, int parm, int tag)
703 * owner - low 15 bits of parm.
704 * group - next 15 bits of parm.
705 * mode - 2 bits of parm, then lowest = 3 bits of tag.
707 int owner, group, mode;
709 owner = parm & 0x7fff;
710 group = (parm >> 15) & 0x7fff;
711 if (fchown(fd, owner, group) < 0)
714 mode = (parm >> 27) & 0x18;
716 if (fchmod(fd, mode) < 0)
721 /* GetOGM - get parm and tag from owner, group and mode bits. */
723 GetOGMFromStat(struct afs_stat_st *status, int *parm, int *tag)
725 *parm = status->st_uid | (status->st_gid << 15);
726 *parm |= (status->st_mode & 0x18) << 27;
727 *tag = status->st_mode & 0x7;
731 GetOGM(FdHandle_t *fdP, int *parm, int *tag)
733 struct afs_stat_st status;
734 if (afs_fstat(fdP->fd_fd, &status) < 0)
736 GetOGMFromStat(&status, parm, tag);
741 CheckOGM(FdHandle_t *fdP, int p1)
745 if (GetOGM(fdP, &parm, &tag) < 0)
754 FixSpecialOGM(FdHandle_t *fdP, int check)
756 int inode_volid, ogm_volid;
757 int inode_type, ogm_type;
758 Inode ino = fdP->fd_ih->ih_ino;
760 inode_volid = ((ino >> NAMEI_UNIQSHIFT) & NAMEI_UNIQMASK);
761 inode_type = (int)((ino >> NAMEI_TAGSHIFT) & NAMEI_TAGMASK);
763 if (GetOGM(fdP, &ogm_volid, &ogm_type) < 0) {
764 Log("Error retrieving OGM info\n");
768 if (inode_volid != ogm_volid || inode_type != ogm_type) {
769 Log("%sIncorrect OGM data (ino: vol %u type %d) (ogm: vol %u type %d)\n",
770 check?"":"Fixing ", inode_volid, inode_type, ogm_volid, ogm_type);
776 if (SetOGM(fdP->fd_fd, inode_volid, inode_type) < 0) {
777 Log("Error setting OGM data\n");
784 #endif /* !AFS_NT40_ENV */
787 * Check/fix the OGM data for an inode
789 * @param[in] fdP Open file handle for the inode to check
790 * @param[in] check 1 to just check the OGM data, and return an error if it
791 * is incorrect. 0 to fix the OGM data if it is incorrect.
793 * @pre fdP must be for a special inode
800 namei_FixSpecialOGM(FdHandle_t *fdP, int check)
803 Inode ino = fdP->fd_ih->ih_ino;
805 vnode = (int)(ino & NAMEI_VNODEMASK);
806 if (vnode != NAMEI_VNODESPECIAL) {
807 Log("FixSpecialOGM: non-special vnode %u\n", vnode);
811 return FixSpecialOGM(fdP, check);
814 int big_vno = 0; /* Just in case we ever do 64 bit vnodes. */
816 /* Derive the name and create it O_EXCL. If that fails we have an error.
817 * Get the tag from a free column in the link table.
821 namei_icreate(IHandle_t * lh, char *part, afs_uint32 p1, afs_uint32 p2, afs_uint32 p3, afs_uint32 p4)
824 FD_t fd = INVALID_FD;
835 memset((void *)&tmp, 0, sizeof(IHandle_t));
836 memset(&tfd, 0, sizeof(FdHandle_t));
838 tmp.ih_dev = nt_DriveToDev(part);
839 if (tmp.ih_dev == -1) {
844 if (p2 == INODESPECIAL) {
845 /* Parameters for special file:
846 * p1 - volume id - goes into owner/group/mode
847 * p2 - vnode == INODESPECIAL
849 * p4 - parent volume id
854 tmp.ih_vid = p4; /* Use parent volume id, where this file will be. */
855 tmp.ih_ino = namei_MakeSpecIno(p1, p3);
857 int vno = p2 & NAMEI_VNODEMASK;
858 /* Parameters for regular file:
872 tmp.ih_ino = (Inode) p2;
877 namei_HandleToName(&name, &tmp);
878 p = strrchr((char *)&name.n_path, '.');
880 for (tag = 0; tag < NAMEI_MAXVOLS; tag++) {
881 *p = *int_to_base32(str1, tag);
882 fd = OS_OPEN((char *)&name.n_path, O_CREAT | O_RDWR | O_EXCL, 0666);
883 if (fd == INVALID_FD) {
884 if (errno == ENOTDIR || errno == ENOENT) {
885 if (namei_CreateDataDirectories(&name, &created_dir) == 0)
886 fd = OS_OPEN((char *)&name.n_path, O_CREAT | O_RDWR | O_EXCL, 0666);
890 if (fd != INVALID_FD)
892 if (p2 == INODESPECIAL && p3 == VI_LINKTABLE)
895 if (fd == INVALID_FD) {
899 tmp.ih_ino &= ~(((Inode) NAMEI_TAGMASK) << NAMEI_TAGSHIFT);
900 tmp.ih_ino |= (((Inode) tag) << NAMEI_TAGSHIFT);
903 if (SetWinOGM(fd, ogm_p1, ogm_p2)) {
904 errno = OS_ERROR(EBADF);
910 if (p2 != INODESPECIAL) {
911 if (fd == INVALID_FD) {
913 code = nt_unlink((char *)&name.n_path);
922 code = namei_SetLinkCount(fdP, tmp.ih_ino, 1, 0);
924 } else if (p2 == INODESPECIAL && p3 == VI_LINKTABLE) {
925 if (fd == INVALID_FD)
927 /* hack at tmp to setup for set link count call. */
929 code = namei_SetLinkCount(&tfd, (Inode) 0, 1, 0);
934 if (fd != INVALID_FD)
937 if (code || (fd == INVALID_FD)) {
938 if (p2 != INODESPECIAL) {
941 namei_SetLinkCount(fdP, tmp.ih_ino, 0, 0);
947 int save_errno = errno;
948 namei_RemoveDataDirectories(&name);
952 return (code || (fd == INVALID_FD)) ? (Inode) -1 : tmp.ih_ino;
954 #else /* !AFS_NT40_ENV */
956 icreate(IHandle_t * lh, char *part, afs_uint32 p1, afs_uint32 p2, afs_uint32 p3, afs_uint32 p4,
964 IHandle_t *realh = NULL;
969 memset((void *)&tmp, 0, sizeof(IHandle_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 IH_INIT(realh, tmp.ih_dev, tmp.ih_vid, tmp.ih_ino);
1038 fdP = ih_attachfd(realh, fd);
1040 /* ih_attachfd can only return NULL if we give it an invalid fd; our fd
1041 * must be valid by this point. */
1044 if (p2 == (afs_uint32)-1 && p3 == VI_LINKTABLE) {
1045 code = namei_SetLinkCount(fdP, (Inode) 0, 1, 0);
1051 if (code || (fd == INVALID_FD)) {
1055 namei_SetLinkCount(fdP, tmp.ih_ino, 0, 0);
1068 namei_icreate(IHandle_t * lh, char *part,
1069 afs_uint32 p1, afs_uint32 p2, afs_uint32 p3, afs_uint32 p4)
1072 IHandle_t *ihP = NULL;
1075 code = icreate(lh, part, p1, p2, p3, p4, &ihP);
1087 namei_icreate_init(IHandle_t * lh, int dev, char *part,
1088 afs_uint32 p1, afs_uint32 p2, afs_uint32 p3, afs_uint32 p4)
1091 IHandle_t *ihP = NULL;
1093 code = icreate(lh, part, p1, p2, p3, p4, &ihP);
1103 namei_iopen(IHandle_t * h)
1108 /* Convert handle to file name. */
1109 namei_HandleToName(&name, h);
1110 fd = OS_OPEN((char *)&name.n_path, O_RDWR, 0666);
1114 /* Need to detect vol special file and just unlink. In those cases, the
1115 * handle passed in _is_ for the inode. We only check p1 for the special
1119 namei_dec(IHandle_t * ih, Inode ino, int p1)
1126 if ((ino & NAMEI_INODESPECIAL) == NAMEI_INODESPECIAL) {
1128 int type = (int)((ino >> NAMEI_TAGSHIFT) & NAMEI_TAGMASK);
1130 /* Verify this is the right file. */
1131 IH_INIT(tmp, ih->ih_dev, ih->ih_vid, ino);
1133 namei_HandleToName(&name, tmp);
1138 errno = OS_ERROR(ENOENT);
1142 if (CheckOGM(fdP, p1) < 0) {
1143 FDH_REALLYCLOSE(fdP);
1145 errno = OS_ERROR(EINVAL);
1149 /* If it's the link table itself, decrement the link count. */
1150 if (type == VI_LINKTABLE) {
1151 if ((count = namei_GetLinkCount(fdP, (Inode) 0, 1, 0, 1)) < 0) {
1152 FDH_REALLYCLOSE(fdP);
1159 /* if our count is non-zero, we just set our new linkcount and
1160 * return. But if our count is 0, don't bother updating the
1161 * linktable, since we're about to delete the link table,
1163 if (namei_SetLinkCount(fdP, (Inode) 0, count < 0 ? 0 : count, 1) < 0) {
1164 FDH_REALLYCLOSE(fdP);
1174 namei_UnlockLinkCount(fdP, (Inode) 0);
1177 /* We should IH_REALLYCLOSE right before deleting the special file from
1178 * disk, to ensure that somebody else cannot create a special inode,
1179 * then IH_OPEN that special inode and get back a cached fd for the
1180 * file we are deleting here (instead of an fd for the file they just
1182 IH_REALLYCLOSE(tmp);
1183 FDH_REALLYCLOSE(fdP);
1186 if ((code = OS_UNLINK(name.n_path)) == 0) {
1187 if (type == VI_LINKTABLE) {
1188 /* Try to remove directory. If it fails, that's ok.
1189 * Salvage will clean up.
1191 char *slash = strrchr(name.n_path, OS_DIRSEPC);
1193 /* avoid an rmdir() on the file we just unlinked */
1196 (void)namei_RemoveDataDirectories(&name);
1200 /* Get a file descriptor handle for this Inode */
1206 if ((count = namei_GetLinkCount(fdP, ino, 1, 0, 1)) < 0) {
1207 FDH_REALLYCLOSE(fdP);
1213 if (namei_SetLinkCount(fdP, ino, count, 1) < 0) {
1214 FDH_REALLYCLOSE(fdP);
1219 IH_INIT(th, ih->ih_dev, ih->ih_vid, ino);
1220 Log("Warning: Lost ref on ihandle dev %d vid %" AFS_VOLID_FMT " ino %lld\n",
1221 th->ih_dev, afs_printable_VolumeId_lu(th->ih_vid), (afs_int64)th->ih_ino);
1224 /* If we're less than 0, someone presumably unlinked;
1225 don't bother setting count to 0, but we need to drop a lock */
1226 if (namei_SetLinkCount(fdP, ino, 0, 1) < 0) {
1227 FDH_REALLYCLOSE(fdP);
1233 IH_INIT(th, ih->ih_dev, ih->ih_vid, ino);
1235 namei_HandleToName(&name, th);
1237 code = OS_UNLINK(name.n_path);
1246 namei_inc(IHandle_t * h, Inode ino, int p1)
1252 if ((ino & NAMEI_INODESPECIAL) == NAMEI_INODESPECIAL) {
1253 int type = (int)((ino >> NAMEI_TAGSHIFT) & NAMEI_TAGMASK);
1254 if (type != VI_LINKTABLE)
1259 /* Get a file descriptor handle for this Inode */
1265 if ((count = namei_GetLinkCount(fdP, ino, 1, 0, 1)) < 0)
1270 errno = OS_ERROR(EINVAL);
1274 if (namei_SetLinkCount(fdP, ino, count, 1) < 0)
1278 FDH_REALLYCLOSE(fdP);
1285 #ifndef AFS_NT40_ENV
1287 namei_replace_file_by_hardlink(IHandle_t *hLink, IHandle_t *hTarget)
1293 /* Convert handle to file name. */
1294 namei_HandleToName(&nameLink, hLink);
1295 namei_HandleToName(&nameTarget, hTarget);
1297 OS_UNLINK(nameLink.n_path);
1298 code = link(nameTarget.n_path, nameLink.n_path);
1303 namei_copy_on_write(IHandle_t *h)
1309 struct afs_stat_st tstat;
1312 namei_HandleToName(&name, h);
1313 if (afs_stat(name.n_path, &tstat) < 0)
1315 if (tstat.st_nlink > 1) { /* do a copy on write */
1324 snprintf(path, sizeof(path), "%s-tmp", name.n_path);
1325 fd = OS_OPEN(path, O_CREAT | O_EXCL | O_RDWR, 0);
1326 if (fd == INVALID_FD) {
1337 size = tstat.st_size;
1340 tlen = size > 8192 ? 8192 : size;
1341 if (FDH_PREAD(fdP, buf, tlen, offset) != tlen)
1343 if (OS_WRITE(fd, buf, tlen) != tlen)
1349 FDH_REALLYCLOSE(fdP);
1354 OS_UNLINK(name.n_path);
1355 code = rename(path, name.n_path);
1362 /************************************************************************
1363 * File Name Structure
1364 ************************************************************************
1366 * Each AFS file needs a unique name and it needs to be findable with
1367 * minimal lookup time. Note that the constraint on the number of files and
1368 * directories in a volume is the size of the vnode index files and the
1369 * max file size AFS supports (for internal files) of 2^31. Since a record
1370 * in the small vnode index file is 64 bytes long, we can have at most
1371 * (2^31)/64 or 33554432 files. A record in the large index file is
1372 * 256 bytes long, giving a maximum of (2^31)/256 = 8388608 directories.
1373 * Another layout parameter is that there is roughly a 16 to 1 ratio between
1374 * the number of files and the number of directories.
1376 * Using this information we can see that a layout of 256 directories, each
1377 * with 512 subdirectories and each of those having 512 files gives us
1378 * 256*512*512 = 67108864 AFS files and directories.
1380 * The volume, vnode, uniquifier and data version, as well as the tag
1381 * are required, either for finding the file or for salvaging. It's best to
1382 * restrict the name to something that can be mapped into 64 bits so the
1383 * "Inode" is easily comparable (using "==") to other "Inodes". The tag
1384 * is used to distinguish between different versions of the same file
1385 * which are currently in the RW and clones of a volume. See "Link Table
1386 * Organization" below for more information on the tag. The tag is
1387 * required in the name of the file to ensure a unique name.
1389 * ifdef AFS_NT40_ENV
1390 * The data for each volume group is in a separate directory. The name of the
1391 * volume is of the form: Vol_NNNNNN.data, where NNNNNN is a base 32
1392 * representation of the RW volume ID (even where the RO is the only volume
1393 * on the partition). Below that are separate subdirectories for the
1394 * AFS directories and special files. There are also 16 directories for files,
1395 * hashed on the low 5 bits (recall bit0 is always 0) of the vnode number.
1396 * These directories are named:
1397 * A - P - 16 file directories.
1398 * Q ----- data directory
1399 * R ----- special files directory
1401 * The vnode is hashed into the directory using the low bits of the
1404 * The format of a file name for a regular file is:
1405 * Y:\Vol_NNNNNN.data\X\V_IIIIII.J
1406 * Y - partition encoded as drive letter, starting with D
1407 * NNNNNN - base 32 encoded volume number of RW volume
1408 * X - hash directory, as above
1409 * IIIIII - base 32 encoded vnode number
1410 * J - base 32 encoded tag
1412 * uniq is stored in the dwHighDateTime creation time field
1413 * dv is stored in the dwLowDateTime creation time field
1415 * Special inodes are always in the R directory, as above, and are
1417 * True child volid is stored in the dwHighDateTime creation time field
1418 * vnode number is always -1 (Special)
1419 * type is the IIIIII part of the filename
1420 * uniq is the J part of the filename
1421 * parent volume id is implied in the containing directory
1424 * We can store data in the uid, gid and mode bits of the files, provided
1425 * the directories have root only access. This gives us 15 bits for each
1426 * of uid and gid (GNU chown considers 65535 to mean "don't change").
1427 * There are 9 available mode bits. Adn we need to store a total of
1428 * 32 (volume id) + 26 (vnode) + 32 (uniquifier) + 32 (data-version) + 3 (tag)
1429 * or 131 bits somewhere.
1431 * The format of a file name for a regular file is:
1432 * /vicepX/AFSIDat/V1/V2/AA/BB/<tag><uniq><vno>
1433 * V1 - low 8 bits of RW volume id
1434 * V2 - all bits of RW volume id
1435 * AA - high 8 bits of vnode number.
1436 * BB - next 9 bits of vnode number.
1437 * <tag><uniq><vno> - file name
1439 * Volume special files are stored in a separate directory:
1440 * /vicepX/AFSIDat/V1/V2/special/<tag><uniq><vno>
1443 * The vnode is hashed into the directory using the high bits of the
1444 * vnode number. This is so that consecutively created vnodes are in
1445 * roughly the same area on the disk. This will at least be optimal if
1446 * the user is creating many files in the same AFS directory. The name
1447 * should be formed so that the leading characters are different as quickly
1448 * as possible, leading to faster discards of incorrect matches in the
1456 /************************************************************************
1457 * Link Table Organization
1458 ************************************************************************
1460 * The link table volume special file is used to hold the link counts that
1461 * are held in the inodes in inode based AFS vice filesystems. For user
1462 * space access, the link counts are being kept in a separate
1463 * volume special file. The file begins with the usual version stamp
1464 * information and is then followed by one row per vnode number. vnode 0
1465 * is used to hold the link count of the link table itself. That is because
1466 * the same link table is shared among all the volumes of the volume group
1467 * and is deleted only when the last volume of a volume group is deleted.
1469 * Within each row, the columns are 3 bits wide. They can each hold a 0 based
1470 * link count from 0 through 7. Each colume represents a unique instance of
1471 * that vnode. Say we have a file shared between the RW and a RO and a
1472 * different version of the file (or a different uniquifer) for the BU volume.
1473 * Then one column would be holding the link count of 2 for the RW and RO
1474 * and a different column would hold the link count of 1 for the BU volume.
1475 * # ifdef AFS_NT40_ENV
1476 * The column used is determined for NT by the uniquifier tag applied to
1477 * generate a unique file name in the NTFS namespace. The file name is
1478 * of the form "V_<vno>.<tag>" . And the <tag> is also the column number
1479 * in the link table.
1481 * Note that we allow only 5 volumes per file, giving 15 bits used in the
1485 #define LINKTABLE_WIDTH 2
1486 #define LINKTABLE_SHIFT 1 /* log 2 = 1 */
1489 * compute namei link table file and bit offset from inode number.
1491 * @param[in] ino inode number
1492 * @param[out] offset link table file offset
1493 * @param[out] index bit offset within 2-byte record
1498 namei_GetLCOffsetAndIndexFromIno(Inode ino, afs_foff_t * offset, int *index)
1500 int toff = (int)(ino & NAMEI_VNODEMASK);
1501 int tindex = (int)((ino >> NAMEI_TAGSHIFT) & NAMEI_TAGMASK);
1503 *offset = (afs_foff_t) ((toff << LINKTABLE_SHIFT) + 8); /* * 2 + sizeof stamp */
1504 *index = (tindex << 1) + tindex;
1507 #ifdef AFS_PTHREAD_ENV
1508 /* XXX do static initializers work for WINNT/pthread? */
1509 pthread_mutex_t _namei_glc_lock = PTHREAD_MUTEX_INITIALIZER;
1510 #define NAMEI_GLC_LOCK opr_mutex_enter(&_namei_glc_lock)
1511 #define NAMEI_GLC_UNLOCK opr_mutex_exit(&_namei_glc_lock)
1512 #else /* !AFS_PTHREAD_ENV */
1513 #define NAMEI_GLC_LOCK
1514 #define NAMEI_GLC_UNLOCK
1515 #endif /* !AFS_PTHREAD_ENV */
1518 * get the link count of an inode.
1520 * @param[in] h namei link count table file handle
1521 * @param[in] ino inode number for which we are requesting a link count
1522 * @param[in] lockit if asserted, return with lock held on link table file
1523 * @param[in] fixup if asserted, write 1 to link count when read() returns
1525 * @param[in] nowrite return success on zero byte read or ZLC
1527 * @post if lockit asserted and lookup was successful, will return with write
1528 * lock on link table file descriptor
1530 * @return link count
1531 * @retval -1 namei link table i/o error
1536 namei_GetLinkCount(FdHandle_t * h, Inode ino, int lockit, int fixup, int nowrite)
1538 unsigned short row = 0;
1543 /* there's no linktable yet. the salvager will create one later */
1544 if (h->fd_fd == INVALID_FD && fixup)
1546 namei_GetLCOffsetAndIndexFromIno(ino, &offset, &index);
1549 if (FDH_LOCKFILE(h, offset) != 0)
1553 rc = FDH_PREAD(h, (char*)&row, sizeof(row), offset);
1555 goto bad_getLinkByte;
1557 if ((rc == 0 || !((row >> index) & NAMEI_TAGMASK)) && fixup && nowrite)
1559 if (rc == 0 && fixup) {
1561 * extend link table and write a link count of 1 for ino
1563 * in order to make MT-safe, truncation (extension really)
1564 * must happen under a mutex
1567 if (FDH_SIZE(h) >= offset+sizeof(row)) {
1569 goto bad_getLinkByte;
1571 FDH_TRUNC(h, offset+sizeof(row));
1573 rc = FDH_PWRITE(h, (char *)&row, sizeof(row), offset);
1576 if (rc != sizeof(row)) {
1577 goto bad_getLinkByte;
1580 if (fixup && !((row >> index) & NAMEI_TAGMASK)) {
1584 * in order to make this mt-safe, we need to do the read-modify-write
1585 * under a mutex. thus, we repeat the read inside the lock.
1588 rc = FDH_PREAD(h, (char *)&row, sizeof(row), offset);
1589 if (rc == sizeof(row)) {
1591 rc = FDH_PWRITE(h, (char *)&row, sizeof(row), offset);
1594 if (rc != sizeof(row))
1595 goto bad_getLinkByte;
1598 return (int)((row >> index) & NAMEI_TAGMASK);
1602 FDH_UNLOCKFILE(h, offset);
1606 /* Return a free column index for this vnode. */
1608 GetFreeTag(IHandle_t * ih, int vno)
1622 /* Only one manipulates at a time. */
1623 if (FDH_LOCKFILE(fdP, offset) != 0) {
1624 FDH_REALLYCLOSE(fdP);
1628 offset = (vno << LINKTABLE_SHIFT) + 8; /* * 2 + sizeof stamp */
1630 nBytes = FDH_PREAD(fdP, (char *)&row, sizeof(row), offset);
1631 if (nBytes != sizeof(row)) {
1637 /* Now find a free column in this row and claim it. */
1638 for (col = 0; col < NAMEI_MAXVOLS; col++) {
1639 coldata = 7 << (col * 3);
1640 if ((row & coldata) == 0)
1643 if (col >= NAMEI_MAXVOLS) {
1648 coldata = 1 << (col * 3);
1651 if (FDH_PWRITE(fdP, (char *)&row, sizeof(row), offset) != sizeof(row)) {
1654 (void)FDH_SYNC(fdP);
1655 FDH_UNLOCKFILE(fdP, offset);
1660 FDH_UNLOCKFILE(fdP, offset);
1661 FDH_REALLYCLOSE(fdP);
1667 /* namei_SetLinkCount
1668 * If locked is set, assume file is locked. Otherwise, lock file before
1669 * proceeding to modify it.
1672 namei_SetLinkCount(FdHandle_t * fdP, Inode ino, int count, int locked)
1678 ssize_t nBytes = -1;
1680 namei_GetLCOffsetAndIndexFromIno(ino, &offset, &index);
1683 if (FDH_LOCKFILE(fdP, offset) != 0) {
1688 nBytes = FDH_PREAD(fdP, (char *)&row, sizeof(row), offset);
1689 if (nBytes != sizeof(row)) {
1691 errno = OS_ERROR(EBADF);
1692 goto bad_SetLinkCount;
1697 bytesRead = 7 << index;
1699 row &= (unsigned short)~bytesRead;
1700 row |= (unsigned short)count;
1702 if (FDH_PWRITE(fdP, (char *)&row, sizeof(short), offset) != sizeof(short)) {
1703 errno = OS_ERROR(EBADF);
1704 goto bad_SetLinkCount;
1706 (void)FDH_SYNC(fdP);
1712 FDH_UNLOCKFILE(fdP, offset);
1714 /* disallowed above 7, so... */
1719 namei_UnlockLinkCount(FdHandle_t * fdP, Inode ino)
1724 namei_GetLCOffsetAndIndexFromIno(ino, &offset, &index);
1726 FDH_UNLOCKFILE(fdP, offset);
1730 /* ListViceInodes - write inode data to a results file. */
1731 static int DecodeInode(char *dpath, char *name, struct ViceInodeInfo *info,
1733 static int DecodeVolumeName(char *name, VolumeId *vid);
1734 static int namei_ListAFSSubDirs(IHandle_t * dirIH,
1735 int (*write_fun) (FD_t,
1736 struct ViceInodeInfo *,
1737 char *, char *), FD_t fp,
1738 int (*judgeFun) (struct ViceInodeInfo *,
1739 VolumeId vid, void *),
1740 VolumeId singleVolumeNumber, void *rock);
1745 * Write the inode data to the results file.
1747 * Returns -2 on error, 0 on success.
1749 * This is written as a callback simply so that other listing routines
1750 * can use the same inode reading code.
1753 WriteInodeInfo(FD_t fp, struct ViceInodeInfo *info, char *dir, char *name)
1756 n = OS_WRITE(fp, info, sizeof(*info));
1757 return (n == sizeof(*info)) ? 0 : -2;
1761 int mode_errors; /* Number of errors found in mode bits on directories. */
1763 VerifyDirPerms(char *path)
1765 struct afs_stat_st status;
1767 if (afs_stat(path, &status) < 0) {
1768 Log("Unable to stat %s. Please manually verify mode bits for this"
1769 " directory\n", path);
1771 if (((status.st_mode & 0777) != 0700) || (status.st_uid != 0))
1777 * Fill the results file with the requested inode information.
1779 * This code optimizes single volume salvages by just looking at that one
1780 * volume's directory.
1782 * @param[in] devname device name string
1783 * @param[in] moutnedOn vice partition mount point
1784 * @param[in] resultFile result file in which to write inode
1785 * metadata. If NULL, write routine is not
1787 * @param[in] judgeInode filter function pointer. if not NULL, only
1788 * inodes for which this routine returns non-
1789 * zero will be written to the results file.
1790 * @param[in] singleVolumeNumber volume id filter
1791 * @param[out] forcep always set to 0 for namei impl
1792 * @param[in] forceR not used by namei impl
1793 * @param[in] wpath not used by namei impl
1794 * @param[in] rock opaque pointer passed to judgeInode
1796 * @return operation status
1798 * @retval -1 complete failure, salvage should terminate.
1799 * @retval -2 not enough space on partition, salvager has error message
1803 ListViceInodes(char *devname, char *mountedOn, FD_t inodeFile,
1804 int (*judgeInode) (struct ViceInodeInfo * info, VolumeId vid, void *rock),
1805 VolumeId singleVolumeNumber, int *forcep, int forceR, char *wpath,
1810 *forcep = 0; /* no need to salvage until further notice */
1812 /* Verify protections on directories. */
1814 VerifyDirPerms(mountedOn);
1817 namei_ListAFSFiles(mountedOn, WriteInodeInfo, inodeFile, judgeInode,
1818 singleVolumeNumber, rock);
1820 if (inodeFile == INVALID_FD)
1827 if (OS_SYNC(inodeFile) == -1) {
1828 Log("Unable to successfully fsync inode file for %s\n", mountedOn);
1833 * Paranoia: check that the file is really the right size
1835 if (OS_SIZE(inodeFile) != ninodes * sizeof(struct ViceInodeInfo)) {
1836 Log("Wrong size (%d instead of %lu) in inode file for %s\n",
1837 (int) OS_SIZE(inodeFile),
1838 (long unsigned int) ninodes * sizeof(struct ViceInodeInfo),
1847 * Collect all the matching AFS files on the drive.
1848 * If singleVolumeNumber is non-zero, just return files for that volume.
1850 * @param[in] dev vice partition path
1851 * @param[in] writeFun function pointer to a function which
1852 * writes inode information to FILE fp
1853 * @param[in] fp file stream where inode metadata is sent
1854 * @param[in] judgeFun filter function pointer. if not NULL,
1855 * only entries for which a non-zero value
1856 * is returned are written to fp
1857 * @param[in] singleVolumeNumber volume id filter. if nonzero, only
1858 * process files for that specific volume id
1859 * @param[in] rock opaque pointer passed into writeFun and
1862 * @return operation status
1864 * @retval >=0 number of matching files found
1867 namei_ListAFSFiles(char *dev,
1868 int (*writeFun) (FD_t, struct ViceInodeInfo *, char *,
1871 int (*judgeFun) (struct ViceInodeInfo *, VolumeId, void *),
1872 VolumeId singleVolumeNumber, void *rock)
1879 #ifndef AFS_NT40_ENV
1885 static void FreeZLCList(void);
1888 memset((void *)&ih, 0, sizeof(IHandle_t));
1890 ih.ih_dev = nt_DriveToDev(dev);
1892 ih.ih_dev = volutil_GetPartitionID(dev);
1895 if (singleVolumeNumber) {
1896 ih.ih_vid = singleVolumeNumber;
1897 namei_HandleToVolDir(&name, &ih);
1899 namei_ListAFSSubDirs(&ih, writeFun, fp, judgeFun,
1900 singleVolumeNumber, rock);
1904 /* Find all volume data directories and descend through them. */
1905 namei_HandleToInodeDir(&name, &ih);
1907 dirp1 = opendir(name.n_path);
1910 while ((dp1 = readdir(dirp1))) {
1912 /* Heirarchy is one level on Windows */
1913 if (!DecodeVolumeName(dp1->d_name, &ih.ih_vid)) {
1915 namei_ListAFSSubDirs(&ih, writeFun, fp, judgeFun,
1919 if (*dp1->d_name == '.')
1921 snprintf(path2, sizeof(path2), "%s" OS_DIRSEP "%s", name.n_path,
1923 dirp2 = opendir(path2);
1925 while ((dp2 = readdir(dirp2))) {
1926 if (*dp2->d_name == '.')
1928 if (!DecodeVolumeName(dp2->d_name, &ih.ih_vid)) {
1930 namei_ListAFSSubDirs(&ih, writeFun, fp, judgeFun,
1947 static void AddToZLCDeleteList(char dir, char *name);
1948 static void DeleteZLCFiles(char *path);
1952 * examine a namei volume special file.
1954 * @param[in] path1 volume special directory path
1955 * @param[in] dname directory entry name
1956 * @param[in] myIH inode handle to volume directory
1957 * @param[out] linkHandle namei link count fd handle. if
1958 * the inode in question is the link
1959 * table, then the FdHandle is populated
1960 * @param[in] writeFun metadata write function pointer
1961 * @param[in] fp file pointer where inode metadata
1962 * is written by (*writeFun)()
1963 * @param[in] judgeFun inode filter function pointer. if
1964 * not NULL, only inodes for which this
1965 * function returns non-zero are recorded
1966 * into fp by writeFun
1967 * @param[in] singleVolumeNumer volume id filter. if non-zero, only
1968 * inodes associated with this volume id
1969 * are recorded by writeFun
1970 * @param[in] rock opaque pointer passed to writeFun and
1973 * @return operation status
1974 * @retval 1 count this inode
1975 * @retval 0 don't count this inode
1976 * @retval -1 failure
1981 _namei_examine_special(char * path1,
1984 FdHandle_t * linkHandle,
1985 int (*writeFun) (FD_t, struct ViceInodeInfo *, char *,
1988 int (*judgeFun) (struct ViceInodeInfo *, VolumeId, void *),
1989 VolumeId singleVolumeNumber,
1993 struct ViceInodeInfo info;
1995 if (DecodeInode(path1, dname, &info, myIH) < 0) {
2000 if (info.u.param[2] != VI_LINKTABLE) {
2002 } else if (info.u.param[0] != myIH->ih_vid) {
2003 /* VGID encoded in linktable filename and/or OGM data isn't
2004 * consistent with VGID encoded in namei path */
2005 Log("namei_ListAFSSubDirs: warning: inconsistent linktable "
2006 "filename \"%s" OS_DIRSEP "%s\"; salvager will delete it "
2007 "(dir_vgid=%" AFS_VOLID_FMT ", inode_vgid=%" AFS_VOLID_FMT ")\n",
2008 path1, dname, afs_printable_VolumeId_lu(myIH->ih_vid),
2009 afs_printable_VolumeId_lu(info.u.param[0]));
2010 /* We need to set the linkCount to _something_, so linkCount
2011 * doesn't just contain stack garbage. Set it to 0, so in case
2012 * the salvager or whatever our caller is does try to process
2013 * this like a normal file, we won't try to INC or DEC it. */
2017 /* Open this handle */
2018 snprintf(path2, sizeof(path2),
2019 "%s" OS_DIRSEP "%s", path1, dname);
2020 linkHandle->fd_fd = OS_OPEN(path2, Testing ? O_RDONLY : O_RDWR, 0666);
2022 namei_GetLinkCount(linkHandle, (Inode) 0, 1, 1, Testing);
2026 (*judgeFun) (&info, singleVolumeNumber, rock)) {
2027 ret = (*writeFun) (fp, &info, path1, dname);
2029 Log("_namei_examine_special: writeFun returned %d\n", ret);
2041 * examine a namei file.
2043 * @param[in] path3 volume special directory path
2044 * @param[in] dname directory entry name
2045 * @param[in] myIH inode handle to volume directory
2046 * @param[in] linkHandle namei link count fd handle.
2047 * @param[in] writeFun metadata write function pointer
2048 * @param[in] fp file pointer where inode metadata
2049 * is written by (*writeFun)()
2050 * @param[in] judgeFun inode filter function pointer. if
2051 * not NULL, only inodes for which this
2052 * function returns non-zero are recorded
2053 * into fp by writeFun
2054 * @param[in] singleVolumeNumer volume id filter. if non-zero, only
2055 * inodes associated with this volume id
2056 * are recorded by writeFun
2057 * @param[in] rock opaque pointer passed to writeFun and
2060 * @return operation status
2061 * @retval 1 count this inode
2062 * @retval 0 don't count this inode
2063 * @retval -1 failure
2064 * @retval -2 request ZLC delete
2069 _namei_examine_reg(char * path3,
2072 FdHandle_t * linkHandle,
2073 int (*writeFun) (FD_t, struct ViceInodeInfo *, char *,
2076 int (*judgeFun) (struct ViceInodeInfo *, VolumeId, void *),
2077 VolumeId singleVolumeNumber,
2081 struct ViceInodeInfo info;
2083 int dirl; /* Windows-only (one level hash dir) */
2086 if (DecodeInode(path3, dname, &info, myIH) < 0) {
2091 namei_GetLinkCount(linkHandle,
2092 info.inodeNumber, 1, 1, Testing);
2093 if (info.linkCount == 0) {
2095 Log("Found 0 link count file %s" OS_DIRSEP "%s, deleting it.\n", path3, dname);
2096 #ifdef AFS_SALSRV_ENV
2097 /* defer -- the AddToZLCDeleteList() interface is not MT-safe */
2099 #else /* !AFS_SALSRV_ENV */
2100 dirl = path3[strlen(path3)-1];
2101 AddToZLCDeleteList((char)dirl, dname);
2102 #endif /* !AFS_SALSRV_ENV */
2103 #else /* !DELETE_ZLC */
2104 Log("Found 0 link count file %s" OS_DIRSEP "%s.\n", path3,
2111 (*judgeFun) (&info, singleVolumeNumber, rock)) {
2112 ret = (*writeFun) (fp, &info, path3, dname);
2114 Log("_namei_examine_reg: writeFun returned %d\n", ret);
2126 * listsubdirs work queue node.
2128 struct listsubdirs_work_node {
2129 #ifdef AFS_SALSRV_ENV
2130 int *error; /**< *error set if an error was
2131 * encountered in any listsubdirs
2135 IHandle_t * IH; /**< volume directory handle */
2136 FdHandle_t *linkHandle; /**< namei link count fd handle. when
2137 * examinining the link table special
2138 * inode, this will be pointed at the
2141 FD_t fp; /**< file pointer for writeFun */
2143 /** function which will write inode metadata to fp */
2144 int (*writeFun) (FD_t, struct ViceInodeInfo *, char *, char *);
2146 /** inode filter function */
2147 int (*judgeFun) (struct ViceInodeInfo *, VolumeId, void *);
2148 VolumeId singleVolumeNumber; /**< volume id filter */
2149 void * rock; /**< pointer passed to writeFun and judgeFun */
2150 int code; /**< return code from examine function */
2151 int special; /**< asserted when this is a volume
2156 * simple wrapper around _namei_examine_special and _namei_examine_reg.
2158 * @param[in] work the struct listsubdirs_work_node for the associated
2159 * "list subdirs" job
2160 * @param[in] dir the directory to examine
2161 * @param[in] filename the filename in 'dir' to examine
2163 * @return operation status
2164 * @retval 1 count this inode
2165 * @retval 0 don't count this inode
2166 * @retval -1 failure
2169 _namei_examine_file(const struct listsubdirs_work_node *work, char *dir,
2172 if (work->special) {
2173 return _namei_examine_special(dir, filename, work->IH,
2174 work->linkHandle, work->writeFun, work->fp,
2175 work->judgeFun, work->singleVolumeNumber,
2178 return _namei_examine_reg(dir, filename, work->IH,
2179 work->linkHandle, work->writeFun, work->fp,
2180 work->judgeFun, work->singleVolumeNumber,
2186 #ifdef AFS_SALSRV_ENV
2187 /** @addtogroup afs_vol_salsrv_pario */
2191 * arguments for the _namei_examine_file_cbk callback function.
2193 struct listsubdirs_args {
2194 const struct listsubdirs_work_node *work; /**< arguments that are the same
2195 * for all invocations of
2196 * _namei_examine_file_cbk, in
2198 int *result; /**< where we can store the return code of _namei_examine_file */
2200 char dir[512]; /**< directory to examine */
2201 char filename[256]; /**< filename in 'dir' to examine */
2205 * a node in the list of results of listsubdir jobs.
2207 struct listsubdirs_result {
2209 int inodes; /**< return value from _namei_examine_file */
2213 * clean up a list of 'struct listsubdirs_result's and interpret the results.
2215 * @param[in] resultlist a list of 'struct listsubdirs_result's
2217 * @return number of inodes found
2221 _namei_listsubdirs_cleanup_results(struct rx_queue *resultlist)
2223 struct listsubdirs_result *res, *nres;
2226 for(queue_Scan(resultlist, res, nres, listsubdirs_result)) {
2228 /* noop, retain erroneous error code */
2229 } else if (res->inodes < 0) {
2244 * wait for the spawned listsubdirs jobs to finish, and return how many inodes
2247 * @param[in] queue queue to wait to finish
2248 * @param[in] resultlist list of 'struct listsubdirs_result's that the queued
2249 * jobs are storing their results in
2251 * @return number of inodes found
2255 _namei_listsubdirs_wait(struct afs_work_queue *queue, struct rx_queue *resultlist)
2259 code = afs_wq_wait_all(queue);
2264 return _namei_listsubdirs_cleanup_results(resultlist);
2268 * work queue entry point for examining namei files.
2270 * @param[in] queue pointer to struct Vwork_queue
2271 * @param[in] node pointer to struct Vwork_queue_node
2272 * @param[in] queue_rock opaque pointer to struct salsrv_pool_state
2273 * @param[in] node_rock opaque pointer to struct listsubdirs_work_node
2274 * @param[in] caller_rock opaque pointer to struct salsrv_worker_thread_state
2276 * @return operation status
2278 * @see Vwork_queue_callback_func_t
2283 _namei_examine_file_cbk(struct afs_work_queue *queue,
2284 struct afs_work_queue_node *node,
2290 struct listsubdirs_args *args = node_rock;
2291 const struct listsubdirs_work_node * work = args->work;
2292 char *dir = args->dir;
2293 char *filename = args->filename;
2295 code = _namei_examine_file(work, dir, filename);
2297 *(args->result) = code;
2301 /* we've errored, so no point in letting more jobs continue */
2302 afs_wq_shutdown(queue);
2308 static pthread_once_t wq_once = PTHREAD_ONCE_INIT;
2309 static pthread_key_t wq_key;
2312 * create the wq_key key for storing a work queue.
2315 _namei_wq_keycreate(void)
2317 opr_Verify(pthread_key_create(&wq_key, NULL) == 0);
2321 * set the work queue for this thread to use for backgrounding namei ops.
2323 * The work queue will be used in ListAFSSubdirs (called indirectly by
2324 * ListViceInodes) to examine files in parallel.
2326 * @param[in] wq the work queue to use
2329 namei_SetWorkQueue(struct afs_work_queue *wq)
2331 opr_Verify(pthread_once(&wq_once, _namei_wq_keycreate) == 0);
2333 opr_Verify(pthread_setspecific(wq_key, wq) == 0);
2337 * enqueue an examine file work unit.
2339 * @param[in] work the _namei_examine_file arguments that are common to
2340 * all callers within the same ListAFSFiles operation
2341 * @param[in] dir the specific directory to look at (string will be
2342 * copied; can be stack/temporary memory)
2343 * @param[in] filename the filename to look at (string will be copied; can be
2344 * stack/temporary memory)
2345 * @param[in] wq work queue to enqueue this work unit to
2346 * @param[in] resultlist the list to append the 'struct listsubdirs_result' to
2347 * for the enqueued work unit
2349 * @return operation status
2351 * @retval -1 fatal error
2353 * @note errors MUST be indicated by a -1 error code and nothing else, to be
2354 * compatible with _namei_examine_reg and _namei_examine_special
2359 _namei_examine_file_spawn(const struct listsubdirs_work_node *work,
2360 const char *dir, const char *filename,
2361 struct afs_work_queue *wq,
2362 struct rx_queue *resultlist)
2365 struct listsubdirs_args *args = NULL;
2366 struct listsubdirs_result *result = NULL;
2367 struct afs_work_queue_node *node = NULL;
2368 struct afs_work_queue_add_opts opts;
2370 args = malloc(sizeof(*args));
2376 result = malloc(sizeof(*result));
2377 if (result == NULL) {
2382 code = afs_wq_node_alloc(&node);
2387 code = afs_wq_node_set_detached(node);
2394 args->result = &result->inodes;
2395 strlcpy(args->dir, dir, sizeof(args->dir));
2396 strlcpy(args->filename, filename, sizeof(args->filename));
2398 code = afs_wq_node_set_callback(node,
2399 &_namei_examine_file_cbk,
2407 afs_wq_add_opts_init(&opts);
2410 code = afs_wq_add(wq, node, &opts);
2417 queue_Append(resultlist, result);
2422 afs_wq_node_put(node);
2438 #else /* !AFS_SALSRV_ENV */
2439 # define _namei_examine_file_spawn(work, dir, file, wq, resultlist) \
2440 _namei_examine_file(work, dir, file)
2441 #endif /* !AFS_SALSRV_ENV */
2444 * traverse and check inodes.
2446 * @param[in] dirIH volume group directory handle
2447 * @param[in] writeFun function pointer which will write inode
2448 * metadata to FILE stream fp
2449 * @param[in] fp file stream where inode metadata gets
2451 * @param[in] judgeFun inode filter function. if not NULL, only
2452 * inodes for which the filter returns non-zero
2453 * will be written out by writeFun
2454 * @param[in] singleVolumeNumber volume id filter. only inodes matching this
2455 * filter are written out by writeFun
2456 * @param[in] rock opaque pointer passed to judgeFun and writeFun
2458 * @return operation status
2460 * @retval >=0 number of matching inodes found
2462 * @todo the salsrv implementation may consume a lot of
2463 * memory for a large volume. at some point we should
2464 * probably write a background thread to asynchronously
2465 * clean up the resultlist nodes to reduce memory footprint
2470 namei_ListAFSSubDirs(IHandle_t * dirIH,
2471 int (*writeFun) (FD_t, struct ViceInodeInfo *, char *,
2474 int (*judgeFun) (struct ViceInodeInfo *, VolumeId, void *),
2475 VolumeId singleVolumeNumber, void *rock)
2477 int code = 0, ret = 0;
2478 IHandle_t myIH = *dirIH;
2480 char path1[512], path3[512];
2482 #ifndef AFS_NT40_ENV
2487 struct dirent *dp1, *dp3;
2488 FdHandle_t linkHandle;
2490 struct listsubdirs_work_node work;
2491 #ifdef AFS_SALSRV_ENV
2493 struct afs_work_queue *wq;
2495 struct rx_queue resultlist;
2498 namei_HandleToVolDir(&name, &myIH);
2499 strlcpy(path1, name.n_path, sizeof(path1));
2501 /* Do the directory containing the special files first to pick up link
2504 (void)strcat(path1, OS_DIRSEP);
2505 (void)strcat(path1, NAMEI_SPECDIR);
2507 linkHandle.fd_fd = INVALID_FD;
2508 #ifdef AFS_SALSRV_ENV
2509 opr_Verify(pthread_once(&wq_once, _namei_wq_keycreate) == 0);
2511 wq = pthread_getspecific(wq_key);
2517 queue_Init(&resultlist);
2520 memset(&work, 0, sizeof(work));
2521 work.linkHandle = &linkHandle;
2524 work.writeFun = writeFun;
2525 work.judgeFun = judgeFun;
2526 work.singleVolumeNumber = singleVolumeNumber;
2529 #ifdef AFS_SALSRV_ENV
2530 work.error = &error;
2533 dirp1 = opendir(path1);
2535 while ((dp1 = readdir(dirp1))) {
2536 if (*dp1->d_name == '.')
2539 #ifdef AFS_SALSRV_ENV
2545 #endif /* AFS_SALSRV_ENV */
2547 code = _namei_examine_file_spawn(&work, path1, dp1->d_name, wq, &resultlist);
2557 /* count this inode */
2558 #ifndef AFS_SALSRV_ENV
2567 #ifdef AFS_SALSRV_ENV
2568 /* effectively a barrier */
2569 code = _namei_listsubdirs_wait(wq, &resultlist);
2570 if (code < 0 || error) {
2578 if (linkHandle.fd_fd == INVALID_FD) {
2579 Log("namei_ListAFSSubDirs: warning: VG %" AFS_VOLID_FMT " does not have a link table; "
2580 "salvager will recreate it.\n", afs_printable_VolumeId_lu(dirIH->ih_vid));
2583 /* Now run through all the other subdirs */
2584 namei_HandleToVolDir(&name, &myIH);
2585 strlcpy(path1, name.n_path, sizeof(path1));
2589 dirp1 = opendir(path1);
2591 while ((dp1 = readdir(dirp1))) {
2592 #ifndef AFS_NT40_ENV
2593 if (*dp1->d_name == '.')
2596 if (!strcmp(dp1->d_name, NAMEI_SPECDIR))
2599 #ifndef AFS_NT40_ENV /* This level missing on Windows */
2600 /* Now we've got a next level subdir. */
2601 snprintf(path2, sizeof(path2), "%s" OS_DIRSEP "%s",
2602 path1, dp1->d_name);
2603 dirp2 = opendir(path2);
2605 while ((dp2 = readdir(dirp2))) {
2606 if (*dp2->d_name == '.')
2609 /* Now we've got to the actual data */
2610 snprintf(path3, sizeof(path3), "%s" OS_DIRSEP "%s",
2611 path2, dp2->d_name);
2613 /* Now we've got to the actual data */
2614 snprintf(path3, sizeof(path3), "%s" OS_DIRSEP "%s",
2615 path1, dp1->d_name);
2617 dirp3 = opendir(path3);
2619 while ((dp3 = readdir(dirp3))) {
2620 #ifndef AFS_NT40_ENV
2621 if (*dp3->d_name == '.')
2625 #ifdef AFS_SALSRV_ENV
2628 #ifndef AFS_NT40_ENV
2635 #endif /* AFS_SALSRV_ENV */
2637 code = _namei_examine_file_spawn(&work, path3,
2644 #ifndef AFS_NT40_ENV
2652 #ifndef AFS_SALSRV_ENV
2660 #ifndef AFS_NT40_ENV /* This level missing on Windows */
2669 #ifdef AFS_SALSRV_ENV
2670 /* effectively a barrier */
2671 code = _namei_listsubdirs_wait(wq, &resultlist);
2672 if (code < 0 || error) {
2682 /* Then why does this directory exist? Blow it away. */
2683 namei_HandleToVolDir(&name, dirIH);
2684 namei_RemoveDataDirectories(&name);
2688 #ifdef AFS_SALSRV_ENV
2690 afs_wq_wait_all(wq);
2692 _namei_listsubdirs_cleanup_results(&resultlist);
2694 if (linkHandle.fd_fd != INVALID_FD)
2695 OS_CLOSE(linkHandle.fd_fd);
2707 DecodeVolumeName(char *name, VolumeId *vid)
2709 /* Name begins with "Vol_" and ends with .data. See nt_HandleToVolDir() */
2716 if (strncmp(name, "Vol_", 4))
2718 if (strcmp(name + len - 5, ".data"))
2721 stmp[len - 5] = '\0';
2722 *vid = base32_to_int(stmp + 4);
2727 DecodeVolumeName(char *name, VolumeId *vid)
2729 if (strlen(name) < 1)
2731 *vid = (unsigned int)flipbase64_to_int64(name);
2739 * Get the inode number from the name.
2744 DecodeInode(char *dpath, char *name, struct ViceInodeInfo *info,
2749 WIN32_FIND_DATA data;
2753 FdHandle_t linkHandle;
2755 VolumeId volid = myIH->ih_vid;
2757 snprintf(fpath, sizeof(fpath), "%s" OS_DIRSEP "%s", dpath, name);
2759 dirH = FindFirstFileEx(fpath, FindExInfoStandard, &data,
2760 FindExSearchNameMatch, NULL,
2761 FIND_FIRST_EX_CASE_SENSITIVE);
2762 if (dirH == INVALID_HANDLE_VALUE)
2765 (void)strcpy(stmp, name);
2766 s = strrchr(stmp, '_');
2770 t = strrchr(s, '.');
2775 vno = base32_to_int(s); /* type for special files */
2776 tag = base32_to_int(t+1);
2777 info->inodeNumber = ((Inode) tag) << NAMEI_TAGSHIFT;
2778 info->inodeNumber |= vno;
2779 info->byteCount = data.nFileSizeLow;
2781 dirl = dpath[strlen(dpath)-1];
2782 if (dirl == NAMEI_SPECDIRC) { /* Special inode. */
2783 info->inodeNumber |= NAMEI_INODESPECIAL;
2784 info->u.param[0] = data.ftCreationTime.dwHighDateTime;
2785 info->u.param[1] = data.ftCreationTime.dwLowDateTime;
2786 info->u.param[2] = vno; /* type */
2787 info->u.param[3] = volid;
2788 if (vno != VI_LINKTABLE)
2789 info->linkCount = 1;
2791 /* Open this handle */
2793 (void)sprintf(lpath, "%s" OS_DIRSEP "%s", fpath, data.cFileName);
2794 linkHandle.fd_fd = OS_OPEN(lpath, O_RDONLY, 0666);
2796 namei_GetLinkCount(&linkHandle, (Inode) 0, 0, 0, 0);
2800 namei_GetLinkCount(&linkHandle, info->inodeNumber, 0, 0, 0);
2801 if (info->linkCount == 0) {
2803 Log("Found 0 link count file %s" OS_DIRSEP "%s, deleting it.\n",
2804 fpath, data.cFileName);
2805 AddToZLCDeleteList(dirl, data.cFileName);
2807 Log("Found 0 link count file %s" OS_DIRSEP "%s.\n", path,
2811 info->u.param[2] = data.ftCreationTime.dwHighDateTime;
2812 info->u.param[3] = data.ftCreationTime.dwLowDateTime;
2813 info->u.param[1] = vno;
2814 info->u.param[0] = volid;
2821 DecodeInode(char *dpath, char *name, struct ViceInodeInfo *info,
2825 struct afs_stat_st status;
2826 struct afs_stat_st checkstatus;
2828 lb64_string_t check;
2829 VolumeId volid = myIH->ih_vid;
2833 snprintf(fpath, sizeof(fpath), "%s" OS_DIRSEP "%s", dpath, name);
2835 if (afs_stat(fpath, &status) < 0) {
2839 info->byteCount = status.st_size;
2840 info->inodeNumber = (Inode) flipbase64_to_int64(name);
2842 int64_to_flipbase64(check, info->inodeNumber);
2843 if (strcmp(name, check))
2846 /* Check if the _full_ path is correct, to ensure we can actually open this
2847 * file later. Otherwise, the salvager can choke. */
2848 memset(&tmpih, 0, sizeof(tmpih));
2849 tmpih.ih_dev = myIH->ih_dev;
2850 tmpih.ih_vid = myIH->ih_vid;
2851 tmpih.ih_ino = info->inodeNumber;
2852 namei_HandleToName(&nameiname, &tmpih);
2853 if ((afs_stat(nameiname.n_path, &checkstatus) < 0) ||
2854 checkstatus.st_ino != status.st_ino ||
2855 checkstatus.st_size != status.st_size) {
2857 /* log something for this case, since this means the filename looks
2858 * like a valid inode, but it's just in the wrong place. That's pretty
2863 Log(" Seemingly-misplaced files have been found, which I am\n");
2864 Log(" ignoring for now. If you cannot salvage the relevant volume,\n");
2865 Log(" you may try manually moving them to their correct location.\n");
2866 Log(" If the relevant volume seems fine, and these files do not\n");
2867 Log(" appear to contain important data, you can probably manually\n");
2868 Log(" delete them, or leave them alone. Contact your local OpenAFS\n");
2869 Log(" expert if you are unsure.\n");
2871 Log("Ignoring misplaced file in volume group %u: %s (should be %s)\n",
2872 (unsigned)myIH->ih_vid, fpath, nameiname.n_path);
2876 if ((info->inodeNumber & NAMEI_INODESPECIAL) == NAMEI_INODESPECIAL) {
2877 parm = ((info->inodeNumber >> NAMEI_UNIQSHIFT) & NAMEI_UNIQMASK);
2878 tag = (int)((info->inodeNumber >> NAMEI_TAGSHIFT) & NAMEI_TAGMASK);
2880 /* p1 - vid, p2 - -1, p3 - type, p4 - rwvid */
2881 info->u.param[0] = parm;
2882 info->u.param[1] = -1;
2883 info->u.param[2] = tag;
2884 info->u.param[3] = volid;
2886 GetOGMFromStat(&status, &parm, &tag);
2887 /* p1 - vid, p2 - vno, p3 - uniq, p4 - dv */
2888 info->u.param[0] = volid;
2889 info->u.param[1] = (int)(info->inodeNumber & NAMEI_VNODEMASK);
2890 info->u.param[2] = (int)((info->inodeNumber >> NAMEI_UNIQSHIFT)
2891 & (Inode) NAMEI_UNIQMASK);
2892 info->u.param[3] = parm;
2899 * Convert the VolumeInfo file from RO to RW
2900 * this routine is called by namei_convertROtoRWvolume()
2903 #ifdef FSSYNC_BUILD_CLIENT
2905 convertVolumeInfo(FD_t fdr, FD_t fdw, VolumeId vid)
2907 struct VolumeDiskData vd;
2910 if (OS_READ(fdr, &vd, sizeof(struct VolumeDiskData)) !=
2911 sizeof(struct VolumeDiskData)) {
2912 Log("1 convertVolumeInfo: read failed for %" AFS_VOLID_FMT " with code %d\n",
2913 afs_printable_VolumeId_lu(vid),
2917 vd.restoredFromId = vd.id; /* remember the RO volume here */
2919 vd.id = vd.parentId;
2923 vd.uniquifier += 5000; /* just in case there are still file copies from
2924 * the old RW volume around */
2926 /* For ROs, the copyDate contains the time that the RO volume was actually
2927 * created, and the creationDate just contains the last time the RO was
2928 * copied from the RW data. So, make the new RW creationDate more accurate
2929 * by setting it to copyDate, if copyDate is older. Since, we know the
2930 * volume is at least as old as copyDate. */
2931 if (vd.copyDate < vd.creationDate) {
2932 vd.creationDate = vd.copyDate;
2934 /* If copyDate is newer, just make copyDate and creationDate the same,
2935 * for consistency with other RWs */
2936 vd.copyDate = vd.creationDate;
2939 p = strrchr(vd.name, '.');
2940 if (p && !strcmp(p, ".readonly")) {
2943 if (OS_WRITE(fdw, &vd, sizeof(struct VolumeDiskData)) !=
2944 sizeof(struct VolumeDiskData)) {
2945 Log("1 convertVolumeInfo: write failed for %lu with code %d\n",
2946 afs_printable_uint32_lu(vid),
2955 * Convert a RO-volume into a RW-volume
2957 * This function allows to recover very fast from the loss of a partition
2958 * from RO-copies if all RO-Copies exist on another partition.
2959 * Then these RO-volumes can be made to the new RW-volumes.
2960 * Backup of RW-volumes then consists in "vos release".
2962 * We must make sure in this partition exists only the RO-volume which
2963 * is typical for remote replicas.
2965 * Then the linktable is already ok,
2966 * the vnode files need to be renamed
2967 * the volinfo file needs to be replaced by another one with
2968 * slightly different contents and new name.
2969 * The volume header file of the RO-volume in the /vicep<x> directory
2970 * is destroyed by this call. A new header file for the RW-volume must
2971 * be created after return from this routine.
2975 namei_ConvertROtoRWvolume(char *pname, VolumeId volumeId)
2978 #ifdef FSSYNC_BUILD_CLIENT
2980 char dir_name[512], oldpath[512], newpath[512];
2995 struct DiskPartition64 *partP;
2996 struct ViceInodeInfo info;
2997 struct VolumeDiskHeader h;
2998 char *rwpart, *rwname;
3000 # ifdef AFS_DEMAND_ATTACH_FS
3002 # endif /* AFS_DEMAND_ATTACH_FS */
3004 for (partP = DiskPartitionList; partP && strcmp(partP->name, pname);
3005 partP = partP->next);
3007 Log("1 namei_ConvertROtoRWvolume: Couldn't find DiskPartition for %s\n", pname);
3012 # ifdef AFS_DEMAND_ATTACH_FS
3013 locktype = VVolLockType(V_VOLUPD, 1);
3014 code = VLockVolumeByIdNB(volumeId, partP, locktype);
3020 # endif /* AFS_DEMAND_ATTACH_FS */
3022 if (VReadVolumeDiskHeader(volumeId, partP, &h)) {
3023 Log("1 namei_ConvertROtoRWvolume: Couldn't read header for RO-volume %lu.\n",
3024 afs_printable_uint32_lu(volumeId));
3029 /* check for existing RW on any partition on this server; */
3030 /* if found, return EXDEV - invalid cross-device link */
3032 VGetVolumePath(&ec, h.parent, &rwpart, &rwname);
3034 Log("1 namei_ConvertROtoRWvolume: RW volume %lu already exists on server partition %s.\n",
3035 afs_printable_uint32_lu(h.parent), rwpart);
3042 FSYNC_VolOp(volumeId, pname, FSYNC_VOL_BREAKCBKS, 0, NULL);
3044 ino = namei_MakeSpecIno(h.parent, VI_LINKTABLE);
3045 IH_INIT(ih, partP->device, h.parent, ino);
3047 namei_HandleToName(&n, ih);
3048 strlcpy(dir_name, n.n_path, sizeof(dir_name));
3049 p = strrchr(dir_name, OS_DIRSEPC);
3051 dirp = opendir(dir_name);
3053 Log("1 namei_ConvertROtoRWvolume: Could not opendir(%s)\n", dir_name);
3058 while ((dp = readdir(dirp))) {
3059 /* struct ViceInodeInfo info; */
3060 #ifndef AFS_NT40_ENV
3061 if (*dp->d_name == '.')
3064 if (DecodeInode(dir_name, dp->d_name, &info, ih) < 0) {
3065 Log("1 namei_ConvertROtoRWvolume: DecodeInode failed for %s" OS_DIRSEP "%s\n",
3066 dir_name, dp->d_name);
3071 if (info.u.param[1] != -1) {
3072 Log("1 namei_ConvertROtoRWvolume: found other than volume special file %s" OS_DIRSEP "%s\n", dir_name, dp->d_name);
3077 if (info.u.param[0] != volumeId) {
3078 if (info.u.param[0] == ih->ih_vid) {
3079 if (info.u.param[2] == VI_LINKTABLE) { /* link table */
3084 Log("1 namei_ConvertROtoRWvolume: found special file %s" OS_DIRSEP "%s"
3085 " for volume %lu\n", dir_name, dp->d_name,
3086 afs_printable_uint32_lu(info.u.param[0]));
3091 if (info.u.param[2] == VI_VOLINFO) { /* volume info file */
3092 strlcpy(infoName, dp->d_name, sizeof(infoName));
3094 } else if (info.u.param[2] == VI_SMALLINDEX) { /* small vnodes file */
3095 strlcpy(smallName, dp->d_name, sizeof(smallName));
3097 } else if (info.u.param[2] == VI_LARGEINDEX) { /* large vnodes file */
3098 strlcpy(largeName, dp->d_name, sizeof(largeName));
3102 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);
3109 if (!infoSeen || !smallSeen || !largeSeen || !linkSeen) {
3110 Log("1 namei_ConvertROtoRWvolume: not all special files found in %s\n", dir_name);
3116 * If we come here then there was only a RO-volume and we can safely
3120 memset(&t_ih, 0, sizeof(t_ih));
3121 t_ih.ih_dev = ih->ih_dev;
3122 t_ih.ih_vid = ih->ih_vid;
3124 snprintf(oldpath, sizeof oldpath, "%s" OS_DIRSEP "%s", dir_name,
3126 fd = OS_OPEN(oldpath, O_RDWR, 0);
3127 if (fd == INVALID_FD) {
3128 Log("1 namei_ConvertROtoRWvolume: could not open RO info file: %s\n",
3133 t_ih.ih_ino = namei_MakeSpecIno(ih->ih_vid, VI_VOLINFO);
3134 namei_HandleToName(&n, &t_ih);
3135 fd2 = OS_OPEN(n.n_path, O_CREAT | O_EXCL | O_RDWR, 0);
3136 if (fd2 == INVALID_FD) {
3137 Log("1 namei_ConvertROtoRWvolume: could not create RW info file: %s\n", n.n_path);
3142 code = convertVolumeInfo(fd, fd2, ih->ih_vid);
3146 OS_UNLINK(n.n_path);
3150 SetOGM(fd2, ih->ih_vid, 1);
3153 t_ih.ih_ino = namei_MakeSpecIno(ih->ih_vid, VI_SMALLINDEX);
3154 namei_HandleToName(&n, &t_ih);
3155 snprintf(newpath, sizeof newpath, "%s" OS_DIRSEP "%s", dir_name,
3157 fd = OS_OPEN(newpath, O_RDWR, 0);
3158 if (fd == INVALID_FD) {
3159 Log("1 namei_ConvertROtoRWvolume: could not open SmallIndex file: %s\n", newpath);
3163 SetOGM(fd, ih->ih_vid, 2);
3166 MoveFileEx(n.n_path, newpath, MOVEFILE_WRITE_THROUGH);
3168 link(newpath, n.n_path);
3172 t_ih.ih_ino = namei_MakeSpecIno(ih->ih_vid, VI_LARGEINDEX);
3173 namei_HandleToName(&n, &t_ih);
3174 snprintf(newpath, sizeof newpath, "%s" OS_DIRSEP "%s", dir_name,
3176 fd = OS_OPEN(newpath, O_RDWR, 0);
3177 if (fd == INVALID_FD) {
3178 Log("1 namei_ConvertROtoRWvolume: could not open LargeIndex file: %s\n", newpath);
3182 SetOGM(fd, ih->ih_vid, 3);
3185 MoveFileEx(n.n_path, newpath, MOVEFILE_WRITE_THROUGH);
3187 link(newpath, n.n_path);
3194 h.volumeInfo_hi = h.id;
3195 h.smallVnodeIndex_hi = h.id;
3196 h.largeVnodeIndex_hi = h.id;
3197 h.linkTable_hi = h.id;
3199 if (VCreateVolumeDiskHeader(&h, partP)) {
3200 Log("1 namei_ConvertROtoRWvolume: Couldn't write header for RW-volume %lu\n",
3201 afs_printable_uint32_lu(h.id));
3206 if (VDestroyVolumeDiskHeader(partP, volumeId, h.parent)) {
3207 Log("1 namei_ConvertROtoRWvolume: Couldn't unlink header for RO-volume %lu\n",
3208 afs_printable_uint32_lu(volumeId));
3211 FSYNC_VolOp(volumeId, pname, FSYNC_VOL_DONE, 0, NULL);
3212 FSYNC_VolOp(h.id, pname, FSYNC_VOL_ON, 0, NULL);
3215 # ifdef AFS_DEMAND_ATTACH_FS
3217 VUnlockVolumeById(volumeId, partP);
3219 # endif /* AFS_DEMAND_ATTACH_FS */
3227 * returns a static string used to print either 32 or 64 bit inode numbers.
3230 PrintInode(char *s, Inode ino)
3232 static afs_ino_str_t result;
3236 snprintf(s, sizeof(afs_ino_str_t), "%llu", (afs_uintmax_t) ino);
3243 /* Routines to facilitate removing zero link count files. */
3244 #define MAX_ZLC_NAMES 32
3245 #define MAX_ZLC_NAMELEN 16
3246 typedef struct zlcList_s {
3247 struct zlcList_s *zlc_next;
3249 char zlc_names[MAX_ZLC_NAMES][MAX_ZLC_NAMELEN];
3252 static zlcList_t *zlcAnchor = NULL;
3253 static zlcList_t *zlcCur = NULL;
3256 AddToZLCDeleteList(char dir, char *name)
3258 opr_Assert(strlen(name) <= MAX_ZLC_NAMELEN - 3);
3260 if (!zlcCur || zlcCur->zlc_n >= MAX_ZLC_NAMES) {
3261 if (zlcCur && zlcCur->zlc_next)
3262 zlcCur = zlcCur->zlc_next;
3264 zlcList_t *tmp = malloc(sizeof(zlcList_t));
3270 zlcCur->zlc_next = tmp;
3274 zlcCur->zlc_next = NULL;
3279 (void)sprintf(zlcCur->zlc_names[zlcCur->zlc_n], "%c" OS_DIRSEP "%s", dir, name);
3281 (void)sprintf(zlcCur->zlc_names[zlcCur->zlc_n], "%s", name);
3287 DeleteZLCFiles(char *path)
3293 for (z = zlcAnchor; z; z = z->zlc_next) {
3294 for (i = 0; i < z->zlc_n; i++) {
3296 (void)sprintf(fname, "%s" OS_DIRSEP "%s", path, z->zlc_names[i]);
3298 (void)sprintf(fname, "%s", z->zlc_names[i]);
3299 if (namei_unlink(fname) < 0) {
3300 Log("ZLC: Can't unlink %s, dos error = %d\n", fname,
3304 z->zlc_n = 0; /* Can reuse space. */
3317 tnext = i->zlc_next;
3321 zlcCur = zlcAnchor = NULL;
3325 #endif /* AFS_NAMEI_ENV */