4 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
6 * This program is free software; you can redistribute it and/or modify
7 * it under the terms of the GNU General Public License version 2 only,
8 * as published by the Free Software Foundation.
10 * This program is distributed in the hope that it will be useful, but
11 * WITHOUT ANY WARRANTY; without even the implied warranty of
12 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
13 * General Public License version 2 for more details (a copy is included
14 * in the LICENSE file that accompanied this code).
16 * You should have received a copy of the GNU General Public License
17 * version 2 along with this program; If not, see
18 * http://www.sun.com/software/products/lustre/docs/GPLv2.pdf
20 * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara,
21 * CA 95054 USA or visit www.sun.com if you need additional information or
27 * Copyright (c) 2003, 2010, Oracle and/or its affiliates. All rights reserved.
28 * Use is subject to license terms.
30 * Copyright (c) 2011, 2013, Intel Corporation.
33 * This file is part of Lustre, http://www.lustre.org/
34 * Lustre is a trademark of Sun Microsystems, Inc.
36 * lustre/utils/liblustreapi.c
38 * Author: Peter J. Braam <braam@clusterfs.com>
39 * Author: Phil Schwan <phil@clusterfs.com>
40 * Author: Robert Read <rread@clusterfs.com>
52 #include <sys/ioctl.h>
59 #include <sys/types.h>
60 #include <sys/syscall.h>
61 #include <sys/xattr.h>
64 #include <libgen.h> /* for dirname() */
65 #ifdef HAVE_LINUX_UNISTD_H
66 #include <linux/unistd.h>
72 #include <liblustre.h>
73 #include <lnet/lnetctl.h>
76 #include <lustre/lustreapi.h>
77 #include "lustreapi_internal.h"
79 static unsigned llapi_dir_filetype_table[] = {
88 #if defined(DT_DOOR) && defined(S_IFDOOR)
93 #if defined(DT_DOOR) && defined(S_IFDOOR)
94 static const int DT_MAX = DT_DOOR;
96 static const int DT_MAX = DT_SOCK;
99 static unsigned llapi_filetype_dir_table[] = {
108 #if defined(DT_DOOR) && defined(S_IFDOOR)
113 #if defined(DT_DOOR) && defined(S_IFDOOR)
114 static const int S_IFMAX = DT_DOOR;
116 static const int S_IFMAX = DT_SOCK;
119 /* liblustreapi message level */
120 static int llapi_msg_level = LLAPI_MSG_MAX;
122 void llapi_msg_set_level(int level)
124 /* ensure level is in the good range */
125 if (level < LLAPI_MSG_OFF)
126 llapi_msg_level = LLAPI_MSG_OFF;
127 else if (level > LLAPI_MSG_MAX)
128 llapi_msg_level = LLAPI_MSG_MAX;
130 llapi_msg_level = level;
133 static void error_callback_default(enum llapi_message_level level, int err,
134 const char *fmt, va_list ap)
136 vfprintf(stderr, fmt, ap);
137 if (level & LLAPI_MSG_NO_ERRNO)
138 fprintf(stderr, "\n");
140 fprintf(stderr, ": %s (%d)\n", strerror(err), err);
143 static void info_callback_default(enum llapi_message_level level, int err,
144 const char *fmt, va_list ap)
146 vfprintf(stdout, fmt, ap);
149 static llapi_log_callback_t llapi_error_callback = error_callback_default;
150 static llapi_log_callback_t llapi_info_callback = info_callback_default;
153 /* llapi_error will preserve errno */
154 void llapi_error(enum llapi_message_level level, int err, const char *fmt, ...)
157 int tmp_errno = errno;
159 if ((level & LLAPI_MSG_MASK) > llapi_msg_level)
163 llapi_error_callback(level, abs(err), fmt, args);
168 /* llapi_printf will preserve errno */
169 void llapi_printf(enum llapi_message_level level, const char *fmt, ...)
172 int tmp_errno = errno;
174 if ((level & LLAPI_MSG_MASK) > llapi_msg_level)
178 llapi_info_callback(level, 0, fmt, args);
184 * Set a custom error logging function. Passing in NULL will reset the logging
185 * callback to its default value.
187 * This function returns the value of the old callback.
189 llapi_log_callback_t llapi_error_callback_set(llapi_log_callback_t cb)
191 llapi_log_callback_t old = llapi_error_callback;
194 llapi_error_callback = cb;
196 llapi_error_callback = error_callback_default;
202 * Set a custom info logging function. Passing in NULL will reset the logging
203 * callback to its default value.
205 * This function returns the value of the old callback.
207 llapi_log_callback_t llapi_info_callback_set(llapi_log_callback_t cb)
209 llapi_log_callback_t old = llapi_info_callback;
212 llapi_info_callback = cb;
214 llapi_info_callback = info_callback_default;
220 * size_units is to be initialized (or zeroed) by caller.
222 int llapi_parse_size(const char *optarg, unsigned long long *size,
223 unsigned long long *size_units, int bytes_spec)
227 if (strncmp(optarg, "-", 1) == 0)
230 if (*size_units == 0)
233 *size = strtoull(optarg, &end, 0);
236 if ((*end == 'b') && *(end + 1) == '\0' &&
237 (*size & (~0ULL << (64 - 9))) == 0 &&
239 *size_units = 1 << 9;
240 } else if ((*end == 'b') &&
241 *(end + 1) == '\0' &&
244 } else if ((*end == 'k' || *end == 'K') &&
245 *(end + 1) == '\0' &&
246 (*size & (~0ULL << (64 - 10))) == 0) {
247 *size_units = 1 << 10;
248 } else if ((*end == 'm' || *end == 'M') &&
249 *(end + 1) == '\0' &&
250 (*size & (~0ULL << (64 - 20))) == 0) {
251 *size_units = 1 << 20;
252 } else if ((*end == 'g' || *end == 'G') &&
253 *(end + 1) == '\0' &&
254 (*size & (~0ULL << (64 - 30))) == 0) {
255 *size_units = 1 << 30;
256 } else if ((*end == 't' || *end == 'T') &&
257 *(end + 1) == '\0' &&
258 (*size & (~0ULL << (64 - 40))) == 0) {
259 *size_units = 1ULL << 40;
260 } else if ((*end == 'p' || *end == 'P') &&
261 *(end + 1) == '\0' &&
262 (*size & (~0ULL << (64 - 50))) == 0) {
263 *size_units = 1ULL << 50;
264 } else if ((*end == 'e' || *end == 'E') &&
265 *(end + 1) == '\0' &&
266 (*size & (~0ULL << (64 - 60))) == 0) {
267 *size_units = 1ULL << 60;
272 *size *= *size_units;
276 /* XXX: llapi_xxx() functions return negative values upon failure */
278 int llapi_stripe_limit_check(unsigned long long stripe_size, int stripe_offset,
279 int stripe_count, int stripe_pattern)
283 /* 64 KB is the largest common page size I'm aware of (on ia64), but
284 * check the local page size just in case. */
285 page_size = LOV_MIN_STRIPE_SIZE;
286 if (getpagesize() > page_size) {
287 page_size = getpagesize();
288 llapi_err_noerrno(LLAPI_MSG_WARN,
289 "warning: your page size (%u) is "
290 "larger than expected (%u)", page_size,
291 LOV_MIN_STRIPE_SIZE);
293 if ((stripe_size & (LOV_MIN_STRIPE_SIZE - 1))) {
295 llapi_error(LLAPI_MSG_ERROR, rc, "error: bad stripe_size %llu, "
296 "must be an even multiple of %d bytes",
297 stripe_size, page_size);
300 if (stripe_offset < -1 || stripe_offset > MAX_OBD_DEVICES) {
302 llapi_error(LLAPI_MSG_ERROR, rc, "error: bad stripe offset %d",
306 if (stripe_count < -1 || stripe_count > LOV_MAX_STRIPE_COUNT) {
308 llapi_error(LLAPI_MSG_ERROR, rc, "error: bad stripe count %d",
312 if (stripe_size >= (1ULL << 32)) {
314 llapi_error(LLAPI_MSG_ERROR, rc,
315 "warning: stripe size 4G or larger "
316 "is not currently supported and would wrap");
322 /* return the first file matching this pattern */
323 static int first_match(char *pattern, char *buffer)
327 if (glob(pattern, GLOB_BRACE, NULL, &glob_info))
330 if (glob_info.gl_pathc < 1) {
331 globfree(&glob_info);
335 strcpy(buffer, glob_info.gl_pathv[0]);
337 globfree(&glob_info);
341 static int find_target_obdpath(char *fsname, char *path)
344 char pattern[PATH_MAX + 1];
347 snprintf(pattern, PATH_MAX,
348 "/proc/fs/lustre/lov/%s-*/target_obd",
350 rc = glob(pattern, GLOB_BRACE, NULL, &glob_info);
351 if (rc == GLOB_NOMATCH)
356 strcpy(path, glob_info.gl_pathv[0]);
357 globfree(&glob_info);
361 static int find_poolpath(char *fsname, char *poolname, char *poolpath)
364 char pattern[PATH_MAX + 1];
367 snprintf(pattern, PATH_MAX,
368 "/proc/fs/lustre/lov/%s-*/pools/%s",
370 rc = glob(pattern, GLOB_BRACE, NULL, &glob_info);
371 /* If no pools, make sure the lov is available */
372 if ((rc == GLOB_NOMATCH) &&
373 (find_target_obdpath(fsname, poolpath) == -ENODEV))
378 strcpy(poolpath, glob_info.gl_pathv[0]);
379 globfree(&glob_info);
384 * return a parameter string for a specific device type or mountpoint
386 * \param param_path the path to the file containing parameter data
387 * \param result buffer for parameter value string
388 * \param result_size size of buffer for return value
390 * The \param param_path is appended to /proc/{fs,sys}/{lnet,lustre} to
391 * complete the absolute path to the file containing the parameter data
392 * the user is requesting. If that file exist then the data is read from
393 * the file and placed into the \param result buffer that is passed by
394 * the user. Data is only copied up to the \param result_size to prevent
395 * overflow of the array.
397 * Return 0 for success, with a NUL-terminated string in \param result.
398 * Return -ve value for error.
400 int get_param(const char *param_path, char *result,
401 unsigned int result_size)
403 char file[PATH_MAX + 1], pattern[PATH_MAX + 1], buf[result_size];
407 snprintf(pattern, PATH_MAX, "/proc/{fs,sys}/{lnet,lustre}/%s",
409 rc = first_match(pattern, file);
410 if (rc != 0 || result == NULL)
413 fp = fopen(file, "r");
415 while (fgets(buf, result_size, fp) != NULL)
424 #define DEVICES_LIST "/proc/fs/lustre/devices"
427 * return a parameter string for a specific device type or mountpoint
429 * \param fsname Lustre filesystem name (optional)
430 * \param file_path path to file in filesystem (optional, if fsname unset)
431 * \param obd_type Lustre OBD device type
432 * \param param_name parameter name to fetch
433 * \param value return buffer for parameter value string
434 * \param val_len size of buffer for return value
436 * If fsname is specified then the parameter will be from that filesystem
437 * (if it exists). If file_path is given and it is in a mounted Lustre
438 * filesystem, then the parameter will be otherwise the value may be
439 * from any mounted filesystem (if there is more than one).
441 * If "obd_type" matches a Lustre device then the first matching device
442 * (as with "lctl dl", constrained by \param fsname or \param mount_path)
443 * will be used to provide the return value, otherwise the first such
444 * device found will be used.
446 * Return 0 for success, with a NUL-terminated string in \param buffer.
447 * Return -ve value for error.
449 static int get_param_obdvar(const char *fsname, const char *file_path,
450 const char *obd_type, const char *param_name,
451 char *value, unsigned int val_len)
453 char devices[PATH_MAX + 1], dev[PATH_MAX + 1] = "*", fs[PATH_MAX + 1];
454 FILE *fp = fopen(DEVICES_LIST, "r");
457 if (!fsname && file_path) {
458 rc = llapi_search_fsname(file_path, fs);
460 llapi_error(LLAPI_MSG_ERROR, rc,
461 "'%s' is not on a Lustre filesystem",
468 if (strlen(fsname) > sizeof(fs)-1) {
473 strncpy(fs, fsname, sizeof(fs));
478 llapi_error(LLAPI_MSG_ERROR, rc, "error: opening "DEVICES_LIST);
482 while (fgets(devices, sizeof(devices), fp) != NULL) {
483 char *bufp = devices, *tmp;
485 while (bufp[0] == ' ')
488 tmp = strstr(bufp, obd_type);
490 tmp += strlen(obd_type) + 1;
493 if (strlen(tmp) > sizeof(dev)-1) {
497 strncpy(dev, tmp, sizeof(dev));
498 tmp = strchr(dev, ' ');
505 if (dev[0] == '*' && strlen(fs))
506 snprintf(dev, PATH_MAX, "%s-*", fs);
507 snprintf(devices, PATH_MAX, "%s/%s/%s", obd_type, dev, param_name);
509 return get_param(devices, value, val_len);
513 * TYPE one of llite, lmv, lov.
514 * /proc/fs/lustre/TYPE/INST the directory of interest.
516 static int get_param_cli(const char *type, const char *inst,
517 const char *param, char *buf, size_t buf_size)
519 char param_path[PATH_MAX + 1];
520 FILE *param_file = NULL;
523 snprintf(param_path, sizeof(param_path),
524 "/proc/fs/lustre/%s/%s/%s", type, inst, param);
526 param_file = fopen(param_path, "r");
527 if (param_file == NULL) {
532 if (fgets(buf, buf_size, param_file) == NULL) {
539 if (param_file != NULL)
545 static int get_param_llite(const char *path,
546 const char *param, char *buf, size_t buf_size)
551 rc = llapi_getname(path, inst, sizeof(inst));
555 return get_param_cli("llite", inst, param, buf, buf_size);
558 static int get_param_lov(const char *path,
559 const char *param, char *buf, size_t buf_size)
561 struct obd_uuid uuid;
564 rc = llapi_file_get_lov_uuid(path, &uuid);
568 return get_param_cli("lov", uuid.uuid, param, buf, buf_size);
571 static int get_param_lmv(const char *path,
572 const char *param, char *buf, size_t buf_size)
574 struct obd_uuid uuid;
577 rc = llapi_file_get_lmv_uuid(path, &uuid);
581 return get_param_cli("lmv", uuid.uuid, param, buf, buf_size);
584 static int get_mds_md_size(const char *path)
586 int md_size = lov_user_md_size(LOV_MAX_STRIPE_COUNT, LOV_USER_MAGIC_V3);
590 /* Get the max ea size from llite proc. */
591 rc = get_param_llite(path, "max_easize", buf, sizeof(buf));
604 * if pool is NULL, search ostname in target_obd
605 * if pool is not NULL:
606 * if pool not found returns errno < 0
607 * if ostname is NULL, returns 1 if pool is not empty and 0 if pool empty
608 * if ostname is not NULL, returns 1 if OST is in pool and 0 if not
610 int llapi_search_ost(char *fsname, char *poolname, char *ostname)
613 char buffer[PATH_MAX + 1];
617 len = strlen(ostname);
619 if (poolname == NULL) {
623 rc = find_target_obdpath(fsname, buffer);
625 rc = find_poolpath(fsname, poolname, buffer);
630 fd = fopen(buffer, "r");
634 while (fgets(buffer, sizeof(buffer), fd) != NULL) {
635 if (poolname == NULL) {
637 /* Search for an ostname in the list of OSTs
638 Line format is IDX: fsname-OSTxxxx_UUID STATUS */
639 ptr = strchr(buffer, ' ');
641 (strncmp(ptr + 1, ostname, len) == 0)) {
646 /* Search for an ostname in a pool,
647 (or an existing non-empty pool if no ostname) */
648 if ((ostname == NULL) ||
649 (strncmp(buffer, ostname, len) == 0)) {
659 int llapi_file_open_pool(const char *name, int flags, int mode,
660 unsigned long long stripe_size, int stripe_offset,
661 int stripe_count, int stripe_pattern, char *pool_name)
663 struct lov_user_md_v3 lum = { 0 };
666 /* Make sure we have a good pool */
667 if (pool_name != NULL) {
668 char fsname[MAX_OBD_NAME + 1], *ptr;
670 rc = llapi_search_fsname(name, fsname);
672 llapi_error(LLAPI_MSG_ERROR, rc,
673 "'%s' is not on a Lustre filesystem",
678 /* in case user gives the full pool name <fsname>.<poolname>,
679 * strip the fsname */
680 ptr = strchr(pool_name, '.');
683 if (strcmp(pool_name, fsname) != 0) {
685 llapi_err_noerrno(LLAPI_MSG_ERROR,
686 "Pool '%s' is not on filesystem '%s'",
693 /* Make sure the pool exists and is non-empty */
694 rc = llapi_search_ost(fsname, pool_name, NULL);
696 llapi_err_noerrno(LLAPI_MSG_ERROR,
697 "pool '%s.%s' %s", fsname, pool_name,
698 rc == 0 ? "has no OSTs" : "does not exist");
704 fd = open(name, flags | O_LOV_DELAY_CREATE, mode);
706 if (errno == EISDIR && !(flags & O_DIRECTORY)) {
707 flags = O_DIRECTORY | O_RDONLY;
714 llapi_error(LLAPI_MSG_ERROR, rc, "unable to open '%s'", name);
718 rc = llapi_stripe_limit_check(stripe_size, stripe_offset, stripe_count,
723 /* Initialize IOCTL striping pattern structure */
724 lum.lmm_magic = LOV_USER_MAGIC_V3;
725 lum.lmm_pattern = stripe_pattern;
726 lum.lmm_stripe_size = stripe_size;
727 lum.lmm_stripe_count = stripe_count;
728 lum.lmm_stripe_offset = stripe_offset;
729 if (pool_name != NULL) {
730 strncpy(lum.lmm_pool_name, pool_name, LOV_MAXPOOLNAME);
732 /* If no pool is specified at all, use V1 request */
733 lum.lmm_magic = LOV_USER_MAGIC_V1;
736 if (ioctl(fd, LL_IOC_LOV_SETSTRIPE, &lum)) {
737 char *errmsg = "stripe already set";
739 if (errno != EEXIST && errno != EALREADY)
740 errmsg = strerror(errno);
742 llapi_err_noerrno(LLAPI_MSG_ERROR,
743 "error on ioctl "LPX64" for '%s' (%d): %s",
744 (__u64)LL_IOC_LOV_SETSTRIPE, name, fd,errmsg);
755 int llapi_file_open(const char *name, int flags, int mode,
756 unsigned long long stripe_size, int stripe_offset,
757 int stripe_count, int stripe_pattern)
759 return llapi_file_open_pool(name, flags, mode, stripe_size,
760 stripe_offset, stripe_count,
761 stripe_pattern, NULL);
764 int llapi_file_create(const char *name, unsigned long long stripe_size,
765 int stripe_offset, int stripe_count, int stripe_pattern)
769 fd = llapi_file_open_pool(name, O_CREAT | O_WRONLY, 0644, stripe_size,
770 stripe_offset, stripe_count, stripe_pattern,
779 int llapi_file_create_pool(const char *name, unsigned long long stripe_size,
780 int stripe_offset, int stripe_count,
781 int stripe_pattern, char *pool_name)
785 fd = llapi_file_open_pool(name, O_CREAT | O_WRONLY, 0644, stripe_size,
786 stripe_offset, stripe_count, stripe_pattern,
795 int llapi_dir_create_pool(const char *name, int flags, int stripe_offset,
796 int stripe_count, int stripe_pattern,
797 const char *pool_name)
799 struct lmv_user_md lmu = { 0 };
800 struct obd_ioctl_data data = { 0 };
803 char *dirpath = NULL;
804 char *namepath = NULL;
810 dirpath = strdup(name);
811 namepath = strdup(name);
812 if (!dirpath || !namepath)
815 lmu.lum_magic = LMV_USER_MAGIC;
816 lmu.lum_stripe_offset = stripe_offset;
817 lmu.lum_stripe_count = stripe_count;
818 lmu.lum_hash_type = stripe_pattern;
819 if (pool_name != NULL) {
820 if (strlen(pool_name) >= LOV_MAXPOOLNAME) {
821 llapi_err_noerrno(LLAPI_MSG_ERROR,
822 "error LL_IOC_LMV_SETSTRIPE '%s' : too large"
823 "pool name: %s", name, pool_name);
827 memcpy(lmu.lum_pool_name, pool_name, strlen(pool_name));
830 filename = basename(namepath);
831 dir = dirname(dirpath);
833 data.ioc_inlbuf1 = (char *)filename;
834 data.ioc_inllen1 = strlen(filename) + 1;
835 data.ioc_inlbuf2 = (char *)&lmu;
836 data.ioc_inllen2 = sizeof(struct lmv_user_md);
837 rc = obd_ioctl_pack(&data, &buf, sizeof(rawbuf));
839 llapi_error(LLAPI_MSG_ERROR, rc,
840 "error: LL_IOC_LMV_SETSTRIPE pack failed '%s'.",
845 fd = open(dir, O_DIRECTORY | O_RDONLY);
848 llapi_error(LLAPI_MSG_ERROR, rc, "unable to open '%s'", name);
852 if (ioctl(fd, LL_IOC_LMV_SETSTRIPE, buf)) {
853 char *errmsg = "stripe already set";
855 if (errno != EEXIST && errno != EALREADY)
856 errmsg = strerror(errno);
858 llapi_err_noerrno(LLAPI_MSG_ERROR,
859 "error on LL_IOC_LMV_SETSTRIPE '%s' (%d): %s",
869 int llapi_direntry_remove(char *dname)
871 char *dirpath = NULL;
872 char *namepath = NULL;
878 dirpath = strdup(dname);
879 namepath = strdup(dname);
880 if (!dirpath || !namepath)
883 filename = basename(namepath);
885 dir = dirname(dirpath);
887 fd = open(dir, O_DIRECTORY | O_RDONLY);
890 llapi_error(LLAPI_MSG_ERROR, rc, "unable to open '%s'",
895 if (ioctl(fd, LL_IOC_REMOVE_ENTRY, filename)) {
896 char *errmsg = strerror(errno);
897 llapi_err_noerrno(LLAPI_MSG_ERROR,
898 "error on ioctl "LPX64" for '%s' (%d): %s",
899 (__u64)LL_IOC_LMV_SETSTRIPE, filename,
911 * Find the fsname, the full path, and/or an open fd.
912 * Either the fsname or path must not be NULL
914 int get_root_path(int want, char *fsname, int *outfd, char *path, int index)
917 char buf[PATH_MAX], mntdir[PATH_MAX];
920 int idx = 0, len = 0, mntlen, fd;
923 /* get the mount point */
924 fp = setmntent(MOUNTED, "r");
927 llapi_error(LLAPI_MSG_ERROR, rc,
928 "setmntent(%s) failed", MOUNTED);
932 if (getmntent_r(fp, &mnt, buf, sizeof(buf)) == NULL)
935 if (!llapi_is_lustre_mnt(&mnt))
938 if ((want & WANT_INDEX) && (idx++ != index))
941 mntlen = strlen(mnt.mnt_dir);
942 ptr = strrchr(mnt.mnt_fsname, '/');
943 /* thanks to the call to llapi_is_lustre_mnt() above,
944 * we are sure that mnt.mnt_fsname contains ":/",
945 * so ptr should never be NULL */
950 /* Check the fsname for a match, if given */
951 if (!(want & WANT_FSNAME) && fsname != NULL &&
952 (strlen(fsname) > 0) && (strcmp(ptr, fsname) != 0))
955 /* If the path isn't set return the first one we find */
956 if (path == NULL || strlen(path) == 0) {
957 strcpy(mntdir, mnt.mnt_dir);
958 if ((want & WANT_FSNAME) && fsname != NULL)
962 /* Otherwise find the longest matching path */
963 } else if ((strlen(path) >= mntlen) && (mntlen >= len) &&
964 (strncmp(mnt.mnt_dir, path, mntlen) == 0)) {
965 strcpy(mntdir, mnt.mnt_dir);
967 if ((want & WANT_FSNAME) && fsname != NULL)
976 if ((want & WANT_PATH) && path != NULL)
977 strcpy(path, mntdir);
978 if (want & WANT_FD) {
979 fd = open(mntdir, O_RDONLY | O_DIRECTORY | O_NONBLOCK);
982 llapi_error(LLAPI_MSG_ERROR, rc,
983 "error opening '%s'", mntdir);
989 } else if (want & WANT_ERROR)
990 llapi_err_noerrno(LLAPI_MSG_ERROR,
991 "can't find fs root for '%s': %d",
992 (want & WANT_PATH) ? fsname : path, rc);
997 * search lustre mounts
999 * Calling this function will return to the user the mount point, mntdir, and
1000 * the file system name, fsname, if the user passed a buffer to this routine.
1002 * The user inputs are pathname and index. If the pathname is supplied then
1003 * the value of the index will be ignored. The pathname will return data if
1004 * the pathname is located on a lustre mount. Index is used to pick which
1005 * mount point you want in the case of multiple mounted lustre file systems.
1006 * See function lfs_osts in lfs.c for a example of the index use.
1008 int llapi_search_mounts(const char *pathname, int index, char *mntdir,
1011 int want = WANT_PATH, idx = -1;
1013 if (!pathname || pathname[0] == '\0') {
1017 strcpy(mntdir, pathname);
1020 want |= WANT_FSNAME;
1021 return get_root_path(want, fsname, NULL, mntdir, idx);
1024 /* Given a path, find the corresponding Lustre fsname */
1025 int llapi_search_fsname(const char *pathname, char *fsname)
1030 path = realpath(pathname, NULL);
1032 char buf[PATH_MAX + 1], *ptr;
1035 if (pathname[0] != '/') {
1036 /* Need an absolute path, but realpath() only works for
1037 * pathnames that actually exist. We go through the
1038 * extra hurdle of dirname(getcwd() + pathname) in
1039 * case the relative pathname contains ".." in it. */
1040 if (getcwd(buf, sizeof(buf) - 1) == NULL)
1044 strncat(buf, pathname, sizeof(buf) - strlen(buf));
1045 path = realpath(buf, NULL);
1047 ptr = strrchr(buf, '/');
1051 path = realpath(buf, NULL);
1054 llapi_error(LLAPI_MSG_ERROR, rc,
1055 "pathname '%s' cannot expand",
1061 rc = get_root_path(WANT_FSNAME | WANT_ERROR, fsname, NULL, path, -1);
1066 int llapi_search_rootpath(char *pathname, const char *fsname)
1068 return get_root_path(WANT_PATH, (char *)fsname, NULL, pathname, -1);
1071 int llapi_getname(const char *path, char *buf, size_t size)
1073 struct obd_uuid uuid_buf;
1074 char *uuid = uuid_buf.uuid;
1077 memset(&uuid_buf, 0, sizeof(uuid_buf));
1078 rc = llapi_file_get_lov_uuid(path, &uuid_buf);
1082 /* We want to turn lustre-clilov-ffff88002738bc00 into
1083 * lustre-ffff88002738bc00. */
1085 nr = snprintf(buf, size, "%.*s-%s",
1086 (int) (strlen(uuid) - 24), uuid,
1087 uuid + strlen(uuid) - 16);
1097 * find the pool directory path under /proc
1098 * (can be also used to test if a fsname is known)
1100 static int poolpath(char *fsname, char *pathname, char *pool_pathname)
1103 char pattern[PATH_MAX + 1];
1104 char buffer[PATH_MAX];
1106 if (fsname == NULL) {
1107 rc = llapi_search_fsname(pathname, buffer);
1111 strcpy(pathname, fsname);
1114 snprintf(pattern, PATH_MAX, "/proc/fs/lustre/lov/%s-*/pools", fsname);
1115 rc = first_match(pattern, buffer);
1119 /* in fsname test mode, pool_pathname is NULL */
1120 if (pool_pathname != NULL)
1121 strcpy(pool_pathname, buffer);
1127 * Get the list of pool members.
1128 * \param poolname string of format \<fsname\>.\<poolname\>
1129 * \param members caller-allocated array of char*
1130 * \param list_size size of the members array
1131 * \param buffer caller-allocated buffer for storing OST names
1132 * \param buffer_size size of the buffer
1134 * \return number of members retrieved for this pool
1135 * \retval -error failure
1137 int llapi_get_poolmembers(const char *poolname, char **members,
1138 int list_size, char *buffer, int buffer_size)
1140 char fsname[PATH_MAX + 1];
1142 char pathname[PATH_MAX + 1];
1143 char path[PATH_MAX + 1];
1150 /* name is FSNAME.POOLNAME */
1151 if (strlen(poolname) > PATH_MAX)
1153 strcpy(fsname, poolname);
1154 pool = strchr(fsname, '.');
1161 rc = poolpath(fsname, NULL, pathname);
1163 llapi_error(LLAPI_MSG_ERROR, rc,
1164 "Lustre filesystem '%s' not found",
1169 llapi_printf(LLAPI_MSG_NORMAL, "Pool: %s.%s\n", fsname, pool);
1170 sprintf(path, "%s/%s", pathname, pool);
1171 fd = fopen(path, "r");
1174 llapi_error(LLAPI_MSG_ERROR, rc, "Cannot open %s", path);
1179 while (fgets(buf, sizeof(buf), fd) != NULL) {
1180 if (nb_entries >= list_size) {
1185 tmp = strchr(buf, '\n');
1188 if (used + strlen(buf) + 1 > buffer_size) {
1193 strcpy(buffer + used, buf);
1194 members[nb_entries] = buffer + used;
1195 used += strlen(buf) + 1;
1205 * Get the list of pools in a filesystem.
1206 * \param name filesystem name or path
1207 * \param poollist caller-allocated array of char*
1208 * \param list_size size of the poollist array
1209 * \param buffer caller-allocated buffer for storing pool names
1210 * \param buffer_size size of the buffer
1212 * \return number of pools retrieved for this filesystem
1213 * \retval -error failure
1215 int llapi_get_poollist(const char *name, char **poollist, int list_size,
1216 char *buffer, int buffer_size)
1218 char fsname[PATH_MAX + 1], rname[PATH_MAX + 1], pathname[PATH_MAX + 1];
1222 struct dirent *cookie = NULL;
1224 unsigned int nb_entries = 0;
1225 unsigned int used = 0;
1228 /* initilize output array */
1229 for (i = 0; i < list_size; i++)
1232 /* is name a pathname ? */
1233 ptr = strchr(name, '/');
1235 /* only absolute pathname is supported */
1239 if (!realpath(name, rname)) {
1241 llapi_error(LLAPI_MSG_ERROR, rc, "invalid path '%s'",
1246 rc = poolpath(NULL, rname, pathname);
1248 llapi_error(LLAPI_MSG_ERROR, rc, "'%s' is not"
1249 " a Lustre filesystem", name);
1252 if (strlen(rname) > sizeof(fsname)-1)
1254 strncpy(fsname, rname, sizeof(fsname));
1256 /* name is FSNAME */
1257 if (strlen(name) > sizeof(fsname)-1)
1259 strncpy(fsname, name, sizeof(fsname));
1260 rc = poolpath(fsname, NULL, pathname);
1263 llapi_error(LLAPI_MSG_ERROR, rc,
1264 "Lustre filesystem '%s' not found", name);
1268 llapi_printf(LLAPI_MSG_NORMAL, "Pools from %s:\n", fsname);
1269 dir = opendir(pathname);
1272 llapi_error(LLAPI_MSG_ERROR, rc,
1273 "Could not open pool list for '%s'",
1279 rc = readdir_r(dir, &pool, &cookie);
1283 llapi_error(LLAPI_MSG_ERROR, rc,
1284 "Error reading pool list for '%s'", name);
1286 } else if ((rc == 0) && (cookie == NULL)) {
1287 /* end of directory */
1291 /* ignore . and .. */
1292 if (!strcmp(pool.d_name, ".") || !strcmp(pool.d_name, ".."))
1295 /* check output bounds */
1296 if (nb_entries >= list_size) {
1301 /* +2 for '.' and final '\0' */
1302 if (used + strlen(pool.d_name) + strlen(fsname) + 2
1308 sprintf(buffer + used, "%s.%s", fsname, pool.d_name);
1309 poollist[nb_entries] = buffer + used;
1310 used += strlen(pool.d_name) + strlen(fsname) + 2;
1316 return ((rc != 0) ? rc : nb_entries);
1319 /* wrapper for lfs.c and obd.c */
1320 int llapi_poollist(const char *name)
1322 /* list of pool names (assume that pool count is smaller
1324 char **list, *buffer = NULL, *path = NULL, *fsname = NULL;
1325 int obdcount, bufsize, rc, nb, i;
1326 char *poolname = NULL, *tmp = NULL, data[16];
1328 if (name[0] != '/') {
1329 fsname = strdup(name);
1330 poolname = strchr(fsname, '.');
1334 path = (char *) name;
1337 rc = get_param_obdvar(fsname, path, "lov", "numobd",
1338 data, sizeof(data));
1341 obdcount = atoi(data);
1343 /* Allocate space for each fsname-OST0000_UUID, 1 per OST,
1344 * and also an array to store the pointers for all that
1345 * allocated space. */
1347 bufsize = sizeof(struct obd_uuid) * obdcount;
1348 buffer = realloc(tmp, bufsize + sizeof(*list) * obdcount);
1349 if (buffer == NULL) {
1353 list = (char **) (buffer + bufsize);
1356 /* name is a path or fsname */
1357 nb = llapi_get_poollist(name, list, obdcount,
1360 /* name is a pool name (<fsname>.<poolname>) */
1361 nb = llapi_get_poolmembers(name, list, obdcount,
1365 if (nb == -EOVERFLOW) {
1368 goto retry_get_pools;
1371 for (i = 0; i < nb; i++)
1372 llapi_printf(LLAPI_MSG_NORMAL, "%s\n", list[i]);
1373 rc = (nb < 0 ? nb : 0);
1382 typedef int (semantic_func_t)(char *path, DIR *parent, DIR *d,
1383 void *data, struct dirent64 *de);
1385 #define OBD_NOT_FOUND (-1)
1387 static int common_param_init(struct find_param *param, char *path)
1389 int lumlen = get_mds_md_size(path);
1391 if (lumlen < PATH_MAX + 1)
1392 lumlen = PATH_MAX + 1;
1394 param->lumlen = lumlen;
1395 param->lmd = malloc(sizeof(lstat_t) + param->lumlen);
1396 if (param->lmd == NULL) {
1397 llapi_error(LLAPI_MSG_ERROR, -ENOMEM,
1398 "error: allocation of %zu bytes for ioctl",
1399 sizeof(lstat_t) + param->lumlen);
1403 param->fp_lmv_count = 256;
1404 param->fp_lmv_md = malloc(lmv_user_md_size(256, LMV_MAGIC_V1));
1405 if (param->fp_lmv_md == NULL) {
1406 llapi_error(LLAPI_MSG_ERROR, -ENOMEM,
1407 "error: allocation of %d bytes for ioctl",
1408 lmv_user_md_size(256, LMV_MAGIC_V1));
1412 param->got_uuids = 0;
1413 param->obdindexes = NULL;
1414 param->obdindex = OBD_NOT_FOUND;
1415 param->mdtindex = OBD_NOT_FOUND;
1419 static void find_param_fini(struct find_param *param)
1421 if (param->obdindexes)
1422 free(param->obdindexes);
1427 if (param->fp_lmv_md)
1428 free(param->fp_lmv_md);
1431 static int cb_common_fini(char *path, DIR *parent, DIR *d, void *data,
1432 struct dirent64 *de)
1434 struct find_param *param = (struct find_param *)data;
1439 /* set errno upon failure */
1440 static DIR *opendir_parent(char *path)
1446 fname = strrchr(path, '/');
1448 return opendir(".");
1452 parent = opendir(path);
1457 static int cb_get_dirstripe(char *path, DIR *d, struct find_param *param)
1459 struct lmv_user_md *lmv = (struct lmv_user_md *)param->fp_lmv_md;
1462 lmv->lum_stripe_count = param->fp_lmv_count;
1463 lmv->lum_magic = LMV_MAGIC_V1;
1464 ret = ioctl(dirfd(d), LL_IOC_LMV_GETSTRIPE, lmv);
1468 static int get_lmd_info(char *path, DIR *parent, DIR *dir,
1469 struct lov_user_mds_data *lmd, int lumlen)
1471 lstat_t *st = &lmd->lmd_st;
1474 if (parent == NULL && dir == NULL)
1478 ret = ioctl(dirfd(dir), LL_IOC_MDC_GETINFO, (void *)lmd);
1479 } else if (parent) {
1480 char *fname = strrchr(path, '/');
1482 fname = (fname == NULL ? path : fname + 1);
1483 /* retrieve needed file info */
1484 strncpy((char *)lmd, fname, lumlen);
1485 ret = ioctl(dirfd(parent), IOC_MDC_GETFILEINFO, (void *)lmd);
1489 if (errno == ENOTTY) {
1490 /* ioctl is not supported, it is not a lustre fs.
1491 * Do the regular lstat(2) instead. */
1492 ret = lstat_f(path, st);
1495 llapi_error(LLAPI_MSG_ERROR, ret,
1496 "error: %s: lstat failed for %s",
1499 } else if (errno == ENOENT) {
1501 llapi_error(LLAPI_MSG_WARN, ret,
1502 "warning: %s: %s does not exist",
1504 } else if (errno != EISDIR) {
1506 llapi_error(LLAPI_MSG_ERROR, ret,
1507 "%s ioctl failed for %s.",
1508 dir ? "LL_IOC_MDC_GETINFO" :
1509 "IOC_MDC_GETFILEINFO", path);
1512 llapi_error(LLAPI_MSG_ERROR, ret,
1513 "error: %s: IOC_MDC_GETFILEINFO failed for %s",
1520 static int llapi_semantic_traverse(char *path, int size, DIR *parent,
1521 semantic_func_t sem_init,
1522 semantic_func_t sem_fini, void *data,
1523 struct dirent64 *de)
1525 struct find_param *param = (struct find_param *)data;
1526 struct dirent64 *dent;
1534 if (!d && errno != ENOTDIR) {
1536 llapi_error(LLAPI_MSG_ERROR, ret, "%s: Failed to open '%s'",
1539 } else if (!d && !parent) {
1540 /* ENOTDIR. Open the parent dir. */
1541 p = opendir_parent(path);
1548 if (sem_init && (ret = sem_init(path, parent ?: p, d, data, de)))
1551 if (!d || (param->get_lmv && !param->recursive)) {
1556 while ((dent = readdir64(d)) != NULL) {
1557 param->have_fileinfo = 0;
1559 if (!strcmp(dent->d_name, ".") || !strcmp(dent->d_name, ".."))
1562 /* Don't traverse .lustre directory */
1563 if (!(strcmp(dent->d_name, dot_lustre_name)))
1567 if ((len + dent->d_reclen + 2) > size) {
1568 llapi_err_noerrno(LLAPI_MSG_ERROR,
1569 "error: %s: string buffer is too small",
1574 strcat(path, dent->d_name);
1576 if (dent->d_type == DT_UNKNOWN) {
1577 lstat_t *st = ¶m->lmd->lmd_st;
1579 ret = get_lmd_info(path, d, NULL, param->lmd,
1583 llapi_filetype_dir_table[st->st_mode &
1589 switch (dent->d_type) {
1591 llapi_err_noerrno(LLAPI_MSG_ERROR,
1592 "error: %s: '%s' is UNKNOWN type %d",
1593 __func__, dent->d_name, dent->d_type);
1596 ret = llapi_semantic_traverse(path, size, d, sem_init,
1597 sem_fini, data, dent);
1604 ret = sem_init(path, d, NULL, data, dent);
1608 if (sem_fini && ret == 0)
1609 sem_fini(path, d, NULL, data, dent);
1617 sem_fini(path, parent, d, data, de);
1626 static int param_callback(char *path, semantic_func_t sem_init,
1627 semantic_func_t sem_fini, struct find_param *param)
1629 int ret, len = strlen(path);
1632 if (len > PATH_MAX) {
1634 llapi_error(LLAPI_MSG_ERROR, ret,
1635 "Path name '%s' is too long", path);
1639 buf = (char *)malloc(PATH_MAX + 1);
1643 strncpy(buf, path, PATH_MAX + 1);
1644 ret = common_param_init(param, buf);
1649 ret = llapi_semantic_traverse(buf, PATH_MAX + 1, NULL, sem_init,
1650 sem_fini, param, NULL);
1652 find_param_fini(param);
1654 return ret < 0 ? ret : 0;
1657 int llapi_file_fget_lov_uuid(int fd, struct obd_uuid *lov_name)
1659 int rc = ioctl(fd, OBD_IOC_GETNAME, lov_name);
1662 llapi_error(LLAPI_MSG_ERROR, rc, "error: can't get lov name.");
1667 int llapi_file_fget_lmv_uuid(int fd, struct obd_uuid *lov_name)
1669 int rc = ioctl(fd, OBD_IOC_GETMDNAME, lov_name);
1672 llapi_error(LLAPI_MSG_ERROR, rc, "error: can't get lmv name.");
1677 int llapi_file_get_lov_uuid(const char *path, struct obd_uuid *lov_uuid)
1681 fd = open(path, O_RDONLY);
1684 llapi_error(LLAPI_MSG_ERROR, rc, "error opening %s", path);
1688 rc = llapi_file_fget_lov_uuid(fd, lov_uuid);
1694 int llapi_file_get_lmv_uuid(const char *path, struct obd_uuid *lov_uuid)
1698 fd = open(path, O_RDONLY);
1701 llapi_error(LLAPI_MSG_ERROR, rc, "error opening %s", path);
1705 rc = llapi_file_fget_lmv_uuid(fd, lov_uuid);
1716 * If uuidp is NULL, return the number of available obd uuids.
1717 * If uuidp is non-NULL, then it will return the uuids of the obds. If
1718 * there are more OSTs then allocated to uuidp, then an error is returned with
1719 * the ost_count set to number of available obd uuids.
1721 static int llapi_get_target_uuids(int fd, struct obd_uuid *uuidp,
1722 int *ost_count, enum tgt_type type)
1724 struct obd_uuid name;
1727 int rc = 0, index = 0;
1729 /* Get the lov name */
1730 if (type == LOV_TYPE) {
1731 rc = llapi_file_fget_lov_uuid(fd, &name);
1735 rc = llapi_file_fget_lmv_uuid(fd, &name);
1740 /* Now get the ost uuids from /proc */
1741 snprintf(buf, sizeof(buf), "/proc/fs/lustre/%s/%s/target_obd",
1742 type == LOV_TYPE ? "lov" : "lmv", name.uuid);
1743 fp = fopen(buf, "r");
1746 llapi_error(LLAPI_MSG_ERROR, rc, "error: opening '%s'", buf);
1750 while (fgets(buf, sizeof(buf), fp) != NULL) {
1751 if (uuidp && (index < *ost_count)) {
1752 if (sscanf(buf, "%d: %s", &index, uuidp[index].uuid) <2)
1760 if (uuidp && (index > *ost_count))
1767 int llapi_lov_get_uuids(int fd, struct obd_uuid *uuidp, int *ost_count)
1769 return llapi_get_target_uuids(fd, uuidp, ost_count, LOV_TYPE);
1772 int llapi_get_obd_count(char *mnt, int *count, int is_mdt)
1777 root = opendir(mnt);
1780 llapi_error(LLAPI_MSG_ERROR, rc, "open %s failed", mnt);
1785 rc = ioctl(dirfd(root), LL_IOC_GETOBDCOUNT, count);
1793 /* Check if user specified value matches a real uuid. Ignore _UUID,
1794 * -osc-4ba41334, other trailing gunk in comparison.
1795 * @param real_uuid ends in "_UUID"
1796 * @param search_uuid may or may not end in "_UUID"
1798 int llapi_uuid_match(char *real_uuid, char *search_uuid)
1800 int cmplen = strlen(real_uuid);
1801 int searchlen = strlen(search_uuid);
1803 if (cmplen > 5 && strcmp(real_uuid + cmplen - 5, "_UUID") == 0)
1805 if (searchlen > 5 && strcmp(search_uuid + searchlen - 5, "_UUID") == 0)
1808 /* The UUIDs may legitimately be different lengths, if
1809 * the system was upgraded from an older version. */
1810 if (cmplen != searchlen)
1813 return (strncmp(search_uuid, real_uuid, cmplen) == 0);
1816 /* Here, param->obduuid points to a single obduuid, the index of which is
1817 * returned in param->obdindex */
1818 static int setup_obd_uuid(DIR *dir, char *dname, struct find_param *param)
1820 struct obd_uuid obd_uuid;
1821 char uuid[sizeof(struct obd_uuid)];
1826 if (param->got_uuids)
1829 /* Get the lov/lmv name */
1831 rc = llapi_file_fget_lmv_uuid(dirfd(dir), &obd_uuid);
1833 rc = llapi_file_fget_lov_uuid(dirfd(dir), &obd_uuid);
1835 if (rc != -ENOTTY) {
1836 llapi_error(LLAPI_MSG_ERROR, rc,
1837 "error: can't get lov name: %s", dname);
1844 param->got_uuids = 1;
1846 /* Now get the ost uuids from /proc */
1847 snprintf(buf, sizeof(buf), "/proc/fs/lustre/%s/%s/target_obd",
1848 param->get_lmv ? "lmv" : "lov", obd_uuid.uuid);
1849 fp = fopen(buf, "r");
1852 llapi_error(LLAPI_MSG_ERROR, rc, "error: opening '%s'", buf);
1856 if (!param->obduuid && !param->quiet && !param->obds_printed)
1857 llapi_printf(LLAPI_MSG_NORMAL, "%s:\n",
1858 param->get_lmv ? "MDTS" : "OBDS:");
1860 while (fgets(buf, sizeof(buf), fp) != NULL) {
1861 if (sscanf(buf, "%d: %s", &index, uuid) < 2)
1864 if (param->obduuid) {
1865 if (llapi_uuid_match(uuid, param->obduuid->uuid)) {
1866 param->obdindex = index;
1869 } else if (!param->quiet && !param->obds_printed) {
1870 /* Print everything */
1871 llapi_printf(LLAPI_MSG_NORMAL, "%s", buf);
1874 param->obds_printed = 1;
1878 if (param->obduuid && (param->obdindex == OBD_NOT_FOUND)) {
1879 llapi_err_noerrno(LLAPI_MSG_ERROR,
1880 "error: %s: unknown obduuid: %s",
1881 __func__, param->obduuid->uuid);
1888 /* In this case, param->obduuid will be an array of obduuids and
1889 * obd index for all these obduuids will be returned in
1890 * param->obdindexes */
1891 static int setup_indexes(DIR *dir, char *path, struct obd_uuid *obduuids,
1892 int num_obds, int **obdindexes, int *obdindex,
1895 int ret, obdcount, obd_valid = 0, obdnum;
1897 struct obd_uuid *uuids = NULL;
1901 if (type == LOV_TYPE)
1902 ret = get_param_lov(path, "numobd", buf, sizeof(buf));
1904 ret = get_param_lmv(path, "numobd", buf, sizeof(buf));
1908 obdcount = atoi(buf);
1909 uuids = (struct obd_uuid *)malloc(obdcount *
1910 sizeof(struct obd_uuid));
1915 ret = llapi_get_target_uuids(dirfd(dir), uuids, &obdcount, type);
1917 struct obd_uuid *uuids_temp;
1919 if (ret == -EOVERFLOW) {
1920 uuids_temp = realloc(uuids, obdcount *
1921 sizeof(struct obd_uuid));
1922 if (uuids_temp != NULL) {
1924 goto retry_get_uuids;
1930 llapi_error(LLAPI_MSG_ERROR, ret, "get ost uuid failed");
1934 indexes = malloc(num_obds * sizeof(*obdindex));
1935 if (indexes == NULL) {
1940 for (obdnum = 0; obdnum < num_obds; obdnum++) {
1943 /* The user may have specified a simple index */
1944 i = strtol(obduuids[obdnum].uuid, &end, 0);
1945 if (end && *end == '\0' && i < obdcount) {
1946 indexes[obdnum] = i;
1949 for (i = 0; i < obdcount; i++) {
1950 if (llapi_uuid_match(uuids[i].uuid,
1951 obduuids[obdnum].uuid)) {
1952 indexes[obdnum] = i;
1958 if (i >= obdcount) {
1959 indexes[obdnum] = OBD_NOT_FOUND;
1960 llapi_err_noerrno(LLAPI_MSG_ERROR,
1961 "error: %s: unknown obduuid: %s",
1962 __func__, obduuids[obdnum].uuid);
1968 *obdindex = OBD_NOT_FOUND;
1970 *obdindex = obd_valid;
1972 *obdindexes = indexes;
1980 static int setup_target_indexes(DIR *dir, char *path, struct find_param *param)
1984 if (param->mdtuuid) {
1985 ret = setup_indexes(dir, path, param->mdtuuid, param->num_mdts,
1986 ¶m->mdtindexes, ¶m->mdtindex, LMV_TYPE);
1990 if (param->obduuid) {
1991 ret = setup_indexes(dir, path, param->obduuid, param->num_obds,
1992 ¶m->obdindexes, ¶m->obdindex, LOV_TYPE);
1996 param->got_uuids = 1;
2000 int llapi_ostlist(char *path, struct find_param *param)
2005 dir = opendir(path);
2009 ret = setup_obd_uuid(dir, path, param);
2016 * Given a filesystem name, or a pathname of a file on a lustre filesystem,
2017 * tries to determine the path to the filesystem's clilov directory under /proc
2019 * fsname is limited to MTI_NAME_MAXLEN in lustre_idl.h
2020 * The NUL terminator is compensated by the additional "%s" bytes. */
2021 #define LOV_LEN (sizeof("/proc/fs/lustre/lov/%s-clilov-*") + MTI_NAME_MAXLEN)
2022 static int clilovpath(const char *fsname, const char *const pathname,
2026 char pattern[LOV_LEN];
2027 char buffer[PATH_MAX + 1];
2029 if (fsname == NULL) {
2030 rc = llapi_search_fsname(pathname, buffer);
2036 snprintf(pattern, sizeof(pattern), "/proc/fs/lustre/lov/%s-clilov-*",
2039 rc = first_match(pattern, buffer);
2043 strncpy(clilovpath, buffer, sizeof(buffer));
2049 * Given the path to a stripe attribute proc file, tries to open and
2050 * read the attribute and return the value using the attr parameter
2052 static int sattr_read_attr(const char *const fpath,
2057 char line[PATH_MAX + 1];
2060 f = fopen(fpath, "r");
2063 llapi_error(LLAPI_MSG_ERROR, rc, "Cannot open '%s'", fpath);
2067 if (fgets(line, sizeof(line), f) != NULL) {
2070 llapi_error(LLAPI_MSG_ERROR, errno, "Cannot read from '%s'", fpath);
2079 * Tries to determine the default stripe attributes for a given filesystem. The
2080 * filesystem to check should be specified by fsname, or will be determined
2083 static int sattr_get_defaults(const char *const fsname,
2084 const char *const pathname,
2085 unsigned int *scount,
2086 unsigned int *ssize,
2087 unsigned int *soffset)
2090 char dpath[PATH_MAX + 1];
2091 char fpath[PATH_MAX + 1];
2093 rc = clilovpath(fsname, pathname, dpath);
2098 snprintf(fpath, PATH_MAX, "%s/stripecount", dpath);
2099 rc = sattr_read_attr(fpath, scount);
2105 snprintf(fpath, PATH_MAX, "%s/stripesize", dpath);
2106 rc = sattr_read_attr(fpath, ssize);
2112 snprintf(fpath, PATH_MAX, "%s/stripeoffset", dpath);
2113 rc = sattr_read_attr(fpath, soffset);
2122 * Tries to gather the default stripe attributes for a given filesystem. If
2123 * the attributes can be determined, they are cached for easy retreival the
2124 * next time they are needed. Only a single filesystem's attributes are
2127 static int sattr_cache_get_defaults(const char *const fsname,
2128 const char *const pathname,
2129 unsigned int *scount,
2130 unsigned int *ssize,
2131 unsigned int *soffset)
2134 char fsname[PATH_MAX + 1];
2135 unsigned int stripecount;
2136 unsigned int stripesize;
2137 unsigned int stripeoffset;
2143 char fsname_buf[PATH_MAX + 1];
2144 unsigned int tmp[3];
2146 if (fsname == NULL) {
2147 rc = llapi_search_fsname(pathname, fsname_buf);
2151 strncpy(fsname_buf, fsname, PATH_MAX);
2154 if (strncmp(fsname_buf, cache.fsname, PATH_MAX) != 0) {
2156 * Ensure all 3 sattrs (count, size, and offset) are
2157 * successfully retrieved and stored in tmp before writing to
2160 rc = sattr_get_defaults(fsname_buf, NULL, &tmp[0], &tmp[1],
2165 cache.stripecount = tmp[0];
2166 cache.stripesize = tmp[1];
2167 cache.stripeoffset = tmp[2];
2168 strncpy(cache.fsname, fsname_buf, PATH_MAX);
2172 *scount = cache.stripecount;
2174 *ssize = cache.stripesize;
2176 *soffset = cache.stripeoffset;
2181 static void lov_dump_user_lmm_header(struct lov_user_md *lum, char *path,
2182 struct lov_user_ost_data_v1 *objects,
2183 int is_dir, int verbose, int depth,
2184 int raw, char *pool_name)
2186 char *prefix = is_dir ? "" : "lmm_";
2187 char *seperator = "";
2190 if (is_dir && lmm_oi_seq(&lum->lmm_oi) == FID_SEQ_LOV_DEFAULT) {
2191 lmm_oi_set_seq(&lum->lmm_oi, 0);
2192 if (verbose & VERBOSE_DETAIL)
2193 llapi_printf(LLAPI_MSG_NORMAL, "(Default) ");
2196 if (depth && path && ((verbose != VERBOSE_OBJID) || !is_dir))
2197 llapi_printf(LLAPI_MSG_NORMAL, "%s\n", path);
2199 if ((verbose & VERBOSE_DETAIL) && !is_dir) {
2200 llapi_printf(LLAPI_MSG_NORMAL, "lmm_magic: 0x%08X\n",
2202 llapi_printf(LLAPI_MSG_NORMAL, "lmm_seq: "LPX64"\n",
2203 lmm_oi_seq(&lum->lmm_oi));
2204 llapi_printf(LLAPI_MSG_NORMAL, "lmm_object_id: "LPX64"\n",
2205 lmm_oi_id(&lum->lmm_oi));
2208 if (verbose & VERBOSE_COUNT) {
2209 if (verbose & ~VERBOSE_COUNT)
2210 llapi_printf(LLAPI_MSG_NORMAL, "%sstripe_count: ",
2213 if (!raw && lum->lmm_stripe_count == 0) {
2214 unsigned int scount;
2215 rc = sattr_cache_get_defaults(NULL, path,
2219 llapi_printf(LLAPI_MSG_NORMAL, "%d",
2222 llapi_error(LLAPI_MSG_ERROR, rc,
2223 "Cannot determine default"
2226 llapi_printf(LLAPI_MSG_NORMAL, "%d",
2227 lum->lmm_stripe_count ==
2228 (typeof(lum->lmm_stripe_count))(-1)
2229 ? -1 : lum->lmm_stripe_count);
2232 llapi_printf(LLAPI_MSG_NORMAL, "%hd",
2233 (__s16)lum->lmm_stripe_count);
2235 seperator = is_dir ? " " : "\n";
2238 if (verbose & VERBOSE_SIZE) {
2239 llapi_printf(LLAPI_MSG_NORMAL, "%s", seperator);
2240 if (verbose & ~VERBOSE_SIZE)
2241 llapi_printf(LLAPI_MSG_NORMAL, "%sstripe_size: ",
2243 if (is_dir && !raw && lum->lmm_stripe_size == 0) {
2245 rc = sattr_cache_get_defaults(NULL, path, NULL, &ssize,
2248 llapi_printf(LLAPI_MSG_NORMAL, "%u", ssize);
2250 llapi_error(LLAPI_MSG_ERROR, rc,
2251 "Cannot determine default"
2254 llapi_printf(LLAPI_MSG_NORMAL, "%u",
2255 lum->lmm_stripe_size);
2257 seperator = is_dir ? " " : "\n";
2260 if ((verbose & VERBOSE_LAYOUT) && !is_dir) {
2261 llapi_printf(LLAPI_MSG_NORMAL, "%s", seperator);
2262 if (verbose & ~VERBOSE_LAYOUT)
2263 llapi_printf(LLAPI_MSG_NORMAL, "%spattern: ",
2265 llapi_printf(LLAPI_MSG_NORMAL, "%.x", lum->lmm_pattern);
2269 if ((verbose & VERBOSE_GENERATION) && !is_dir) {
2270 llapi_printf(LLAPI_MSG_NORMAL, "%s", seperator);
2271 if (verbose & ~VERBOSE_GENERATION)
2272 llapi_printf(LLAPI_MSG_NORMAL, "%slayout_gen: ",
2274 llapi_printf(LLAPI_MSG_NORMAL, "%u",
2275 (int)lum->lmm_layout_gen);
2279 if (verbose & VERBOSE_OFFSET) {
2280 llapi_printf(LLAPI_MSG_NORMAL, "%s", seperator);
2281 if (verbose & ~VERBOSE_OFFSET)
2282 llapi_printf(LLAPI_MSG_NORMAL, "%sstripe_offset: ",
2285 llapi_printf(LLAPI_MSG_NORMAL, "%d",
2286 lum->lmm_stripe_offset ==
2287 (typeof(lum->lmm_stripe_offset))(-1) ? -1 :
2288 lum->lmm_stripe_offset);
2290 llapi_printf(LLAPI_MSG_NORMAL, "%u",
2291 objects[0].l_ost_idx);
2292 seperator = is_dir ? " " : "\n";
2295 if ((verbose & VERBOSE_POOL) && (pool_name != NULL)) {
2296 llapi_printf(LLAPI_MSG_NORMAL, "%s", seperator);
2297 if (verbose & ~VERBOSE_POOL)
2298 llapi_printf(LLAPI_MSG_NORMAL, "%spool: ",
2300 llapi_printf(LLAPI_MSG_NORMAL, "%s", pool_name);
2303 if (!is_dir || (is_dir && (verbose != VERBOSE_OBJID)))
2304 llapi_printf(LLAPI_MSG_NORMAL, "\n");
2307 void lov_dump_user_lmm_v1v3(struct lov_user_md *lum, char *pool_name,
2308 struct lov_user_ost_data_v1 *objects,
2309 char *path, int is_dir, int obdindex,
2310 int depth, int header, int raw)
2312 int i, obdstripe = (obdindex != OBD_NOT_FOUND) ? 0 : 1;
2315 for (i = 0; !is_dir && i < lum->lmm_stripe_count; i++) {
2316 if (obdindex == objects[i].l_ost_idx) {
2324 lov_dump_user_lmm_header(lum, path, objects, is_dir, header,
2325 depth, raw, pool_name);
2327 if (!is_dir && (header & VERBOSE_OBJID) &&
2328 !(lum->lmm_pattern & LOV_PATTERN_F_RELEASED)) {
2330 llapi_printf(LLAPI_MSG_NORMAL,
2331 "\tobdidx\t\t objid\t\t objid\t\t group\n");
2333 for (i = 0; i < lum->lmm_stripe_count; i++) {
2334 int idx = objects[i].l_ost_idx;
2335 long long oid = ostid_id(&objects[i].l_ost_oi);
2336 long long gr = ostid_seq(&objects[i].l_ost_oi);
2337 if ((obdindex == OBD_NOT_FOUND) || (obdindex == idx)) {
2339 sprintf(fmt, "%s%s%s\n",
2340 "\t%6u\t%14llu\t%#13llx\t",
2341 (fid_seq_is_rsvd(gr) ||
2342 fid_seq_is_mdt0(gr)) ?
2343 "%14llu" : "%#14llx", "%s");
2344 llapi_printf(LLAPI_MSG_NORMAL, fmt, idx, oid,
2346 obdindex == idx ? " *" : "");
2350 llapi_printf(LLAPI_MSG_NORMAL, "\n");
2354 void lmv_dump_user_lmm(struct lmv_user_md *lum, char *pool_name,
2355 char *path, int obdindex, int depth, int verbose)
2357 struct lmv_user_mds_data *objects = lum->lum_objects;
2358 char *prefix = lum->lum_magic == LMV_USER_MAGIC ? "(Default)" : "";
2359 int i, obdstripe = 0;
2361 if (obdindex != OBD_NOT_FOUND) {
2362 for (i = 0; i < lum->lum_stripe_count; i++) {
2363 if (obdindex == objects[i].lum_mds) {
2364 llapi_printf(LLAPI_MSG_NORMAL, "%s%s\n", prefix,
2374 /* show all information default */
2376 if (lum->lum_magic == LMV_USER_MAGIC)
2377 verbose = VERBOSE_POOL | VERBOSE_COUNT | VERBOSE_OFFSET;
2379 verbose = VERBOSE_OBJID;
2382 if (lum->lum_magic == LMV_USER_MAGIC)
2383 verbose &= ~VERBOSE_OBJID;
2385 if (depth && path && ((verbose != VERBOSE_OBJID)))
2386 llapi_printf(LLAPI_MSG_NORMAL, "%s\n", path);
2388 if (verbose & VERBOSE_COUNT) {
2389 if (verbose & ~VERBOSE_COUNT)
2390 llapi_printf(LLAPI_MSG_NORMAL, "lmv_stripe_count: ");
2391 llapi_printf(LLAPI_MSG_NORMAL, "%u\n",
2392 (int)lum->lum_stripe_count);
2395 if (verbose & VERBOSE_OFFSET) {
2396 if (verbose & ~VERBOSE_OFFSET)
2397 llapi_printf(LLAPI_MSG_NORMAL, "lmv_stripe_offset: ");
2398 llapi_printf(LLAPI_MSG_NORMAL, "%d\n",
2399 (int)lum->lum_stripe_offset);
2402 if (verbose & VERBOSE_OBJID) {
2403 if ((obdstripe == 1))
2404 llapi_printf(LLAPI_MSG_NORMAL,
2405 "\tmdtidx\t\t FID[seq:oid:ver]\n");
2406 for (i = 0; i < lum->lum_stripe_count; i++) {
2407 int idx = objects[i].lum_mds;
2408 struct lu_fid *fid = &objects[i].lum_fid;
2409 if ((obdindex == OBD_NOT_FOUND) || (obdindex == idx))
2410 llapi_printf(LLAPI_MSG_NORMAL,
2411 "\t%6u\t\t "DFID"\t\t%s\n",
2413 obdindex == idx ? " *" : "");
2418 if ((verbose & VERBOSE_POOL) && (pool_name[0] != '\0')) {
2419 if (verbose & ~VERBOSE_POOL)
2420 llapi_printf(LLAPI_MSG_NORMAL, "%slmv_pool: ",
2422 llapi_printf(LLAPI_MSG_NORMAL, "%s%c ", pool_name, ' ');
2424 llapi_printf(LLAPI_MSG_NORMAL, "\n");
2427 void llapi_lov_dump_user_lmm(struct find_param *param, char *path, int is_dir)
2432 magic = (__u32)param->fp_lmv_md->lum_magic;
2434 magic = *(__u32 *)¶m->lmd->lmd_lmm; /* lum->lmm_magic */
2437 case LOV_USER_MAGIC_V1:
2438 lov_dump_user_lmm_v1v3(¶m->lmd->lmd_lmm, NULL,
2439 param->lmd->lmd_lmm.lmm_objects,
2441 param->obdindex, param->maxdepth,
2442 param->verbose, param->raw);
2444 case LOV_USER_MAGIC_V3: {
2445 char pool_name[LOV_MAXPOOLNAME + 1];
2446 struct lov_user_ost_data_v1 *objects;
2447 struct lov_user_md_v3 *lmmv3 = (void *)¶m->lmd->lmd_lmm;
2449 strncpy(pool_name, lmmv3->lmm_pool_name, LOV_MAXPOOLNAME);
2450 pool_name[LOV_MAXPOOLNAME] = '\0';
2451 objects = lmmv3->lmm_objects;
2452 lov_dump_user_lmm_v1v3(¶m->lmd->lmd_lmm, pool_name,
2453 objects, path, is_dir,
2454 param->obdindex, param->maxdepth,
2455 param->verbose, param->raw);
2459 case LMV_USER_MAGIC: {
2460 char pool_name[LOV_MAXPOOLNAME + 1];
2461 struct lmv_user_md *lum;
2463 lum = (struct lmv_user_md *)param->fp_lmv_md;
2464 strncpy(pool_name, lum->lum_pool_name, LOV_MAXPOOLNAME);
2465 lmv_dump_user_lmm(lum, pool_name, path,
2466 param->obdindex, param->maxdepth,
2471 llapi_printf(LLAPI_MSG_NORMAL, "unknown lmm_magic: %#x "
2472 "(expecting one of %#x %#x %#x %#x)\n",
2473 *(__u32 *)¶m->lmd->lmd_lmm,
2474 LOV_USER_MAGIC_V1, LOV_USER_MAGIC_V3,
2475 LMV_USER_MAGIC, LMV_MAGIC_V1);
2480 int llapi_file_get_stripe(const char *path, struct lov_user_md *lum)
2486 fname = strrchr(path, '/');
2488 /* It should be a file (or other non-directory) */
2489 if (fname == NULL) {
2490 dname = (char *)malloc(2);
2494 fname = (char *)path;
2496 dname = (char *)malloc(fname - path + 1);
2499 strncpy(dname, path, fname - path);
2500 dname[fname - path] = '\0';
2504 fd = open(dname, O_RDONLY);
2511 strcpy((char *)lum, fname);
2512 if (ioctl(fd, IOC_MDC_GETFILESTRIPE, (void *)lum) == -1)
2515 if (close(fd) == -1 && rc == 0)
2522 int llapi_file_lookup(int dirfd, const char *name)
2524 struct obd_ioctl_data data = { 0 };
2529 if (dirfd < 0 || name == NULL)
2532 data.ioc_version = OBD_IOCTL_VERSION;
2533 data.ioc_len = sizeof(data);
2534 data.ioc_inlbuf1 = (char *)name;
2535 data.ioc_inllen1 = strlen(name) + 1;
2537 rc = obd_ioctl_pack(&data, &buf, sizeof(rawbuf));
2539 llapi_error(LLAPI_MSG_ERROR, rc,
2540 "error: IOC_MDC_LOOKUP pack failed for '%s': rc %d",
2545 rc = ioctl(dirfd, IOC_MDC_LOOKUP, buf);
2551 /* Check if the value matches 1 of the given criteria (e.g. --atime +/-N).
2552 * @mds indicates if this is MDS timestamps and there are attributes on OSTs.
2554 * The result is -1 if it does not match, 0 if not yet clear, 1 if matches.
2555 * The table below gives the answers for the specified parameters (value and
2556 * sign), 1st column is the answer for the MDS value, the 2nd is for the OST:
2557 * --------------------------------------
2558 * 1 | file > limit; sign > 0 | -1 / -1 |
2559 * 2 | file = limit; sign > 0 | -1 / -1 |
2560 * 3 | file < limit; sign > 0 | ? / 1 |
2561 * 4 | file > limit; sign = 0 | -1 / -1 |
2562 * 5 | file = limit; sign = 0 | ? / 1 | <- (see the Note below)
2563 * 6 | file < limit; sign = 0 | ? / -1 |
2564 * 7 | file > limit; sign < 0 | 1 / 1 |
2565 * 8 | file = limit; sign < 0 | ? / -1 |
2566 * 9 | file < limit; sign < 0 | ? / -1 |
2567 * --------------------------------------
2568 * Note: 5th actually means that the value is within the interval
2569 * (limit - margin, limit]. */
2570 static int find_value_cmp(unsigned long long file, unsigned long long limit,
2571 int sign, int negopt, unsigned long long margin,
2577 /* Drop the fraction of margin (of days). */
2578 if (file + margin <= limit)
2580 } else if (sign == 0) {
2581 if (file <= limit && file + margin > limit)
2583 else if (file + margin <= limit)
2585 } else if (sign < 0) {
2592 return negopt ? ~ret + 1 : ret;
2595 /* Check if the file time matches all the given criteria (e.g. --atime +/-N).
2596 * Return -1 or 1 if file timestamp does not or does match the given criteria
2597 * correspondingly. Return 0 if the MDS time is being checked and there are
2598 * attributes on OSTs and it is not yet clear if the timespamp matches.
2600 * If 0 is returned, we need to do another RPC to the OSTs to obtain the
2601 * updated timestamps. */
2602 static int find_time_check(lstat_t *st, struct find_param *param, int mds)
2607 /* Check if file is accepted. */
2609 ret = find_value_cmp(st->st_atime, param->atime,
2610 param->asign, param->exclude_atime,
2618 ret = find_value_cmp(st->st_mtime, param->mtime,
2619 param->msign, param->exclude_mtime,
2624 /* If the previous check matches, but this one is not yet clear,
2625 * we should return 0 to do an RPC on OSTs. */
2631 ret = find_value_cmp(st->st_ctime, param->ctime,
2632 param->csign, param->exclude_ctime,
2637 /* If the previous check matches, but this one is not yet clear,
2638 * we should return 0 to do an RPC on OSTs. */
2647 * Check whether the stripes matches the indexes user provided
2651 static int check_obd_match(struct find_param *param)
2653 lstat_t *st = ¶m->lmd->lmd_st;
2654 struct lov_user_ost_data_v1 *lmm_objects;
2657 if (param->obduuid && param->obdindex == OBD_NOT_FOUND)
2660 if (!S_ISREG(st->st_mode))
2663 /* Only those files should be accepted, which have a
2664 * stripe on the specified OST. */
2665 if (!param->lmd->lmd_lmm.lmm_stripe_count)
2668 if (param->lmd->lmd_lmm.lmm_magic ==
2669 LOV_USER_MAGIC_V3) {
2670 struct lov_user_md_v3 *lmmv3 = (void *)¶m->lmd->lmd_lmm;
2672 lmm_objects = lmmv3->lmm_objects;
2673 } else if (param->lmd->lmd_lmm.lmm_magic == LOV_USER_MAGIC_V1) {
2674 lmm_objects = param->lmd->lmd_lmm.lmm_objects;
2676 llapi_err_noerrno(LLAPI_MSG_ERROR, "%s:Unknown magic: 0x%08X\n",
2677 __func__, param->lmd->lmd_lmm.lmm_magic);
2681 for (i = 0; i < param->lmd->lmd_lmm.lmm_stripe_count; i++) {
2682 for (j = 0; j < param->num_obds; j++) {
2683 if (param->obdindexes[j] ==
2684 lmm_objects[i].l_ost_idx) {
2685 if (param->exclude_obd)
2692 if (param->exclude_obd)
2697 static int check_mdt_match(struct find_param *param)
2701 if (param->mdtuuid && param->mdtindex == OBD_NOT_FOUND)
2704 /* FIXME: For striped dir, we should get stripe information and check */
2705 for (i = 0; i < param->num_mdts; i++) {
2706 if (param->mdtindexes[i] == param->file_mdtindex)
2707 return !param->exclude_mdt;
2710 if (param->exclude_mdt)
2716 * Check whether the obd is active or not, if it is
2717 * not active, just print the object affected by this
2720 static int print_failed_tgt(struct find_param *param, char *path, int type)
2722 struct obd_statfs stat_buf;
2723 struct obd_uuid uuid_buf;
2726 LASSERT(type == LL_STATFS_LOV || type == LL_STATFS_LMV);
2728 memset(&stat_buf, 0, sizeof(struct obd_statfs));
2729 memset(&uuid_buf, 0, sizeof(struct obd_uuid));
2730 ret = llapi_obd_statfs(path, type,
2731 param->obdindex, &stat_buf,
2734 llapi_printf(LLAPI_MSG_NORMAL,
2735 "obd_uuid: %s failed %s ",
2736 param->obduuid->uuid,
2742 static int cb_find_init(char *path, DIR *parent, DIR *dir,
2743 void *data, struct dirent64 *de)
2745 struct find_param *param = (struct find_param *)data;
2746 int decision = 1; /* 1 is accepted; -1 is rejected. */
2747 lstat_t *st = ¶m->lmd->lmd_st;
2749 int checked_type = 0;
2752 LASSERT(parent != NULL || dir != NULL);
2754 if (param->have_fileinfo == 0)
2755 param->lmd->lmd_lmm.lmm_stripe_count = 0;
2757 /* If a regular expression is presented, make the initial decision */
2758 if (param->pattern != NULL) {
2759 char *fname = strrchr(path, '/');
2760 fname = (fname == NULL ? path : fname + 1);
2761 ret = fnmatch(param->pattern, fname, 0);
2762 if ((ret == FNM_NOMATCH && !param->exclude_pattern) ||
2763 (ret == 0 && param->exclude_pattern))
2767 /* See if we can check the file type from the dirent. */
2768 if (param->type && de != NULL && de->d_type != DT_UNKNOWN &&
2769 de->d_type < DT_MAX) {
2771 if (llapi_dir_filetype_table[de->d_type] == param->type) {
2772 if (param->exclude_type)
2775 if (!param->exclude_type)
2782 /* Request MDS for the stat info if some of these parameters need
2783 * to be compared. */
2784 if (param->obduuid || param->mdtuuid || param->check_uid ||
2785 param->check_gid || param->check_pool || param->atime ||
2786 param->ctime || param->mtime || param->check_size ||
2787 param->check_stripecount || param->check_stripesize ||
2788 param->check_layout)
2791 if (param->type && checked_type == 0)
2794 if (param->have_fileinfo == 0 && decision == 0) {
2795 ret = get_lmd_info(path, parent, dir, param->lmd,
2799 ret = llapi_file_fget_mdtidx(dirfd(dir),
2800 ¶m->file_mdtindex);
2805 ret = lstat_f(path, &tmp_st);
2808 llapi_error(LLAPI_MSG_ERROR, ret,
2809 "error: %s: lstat failed"
2810 "for %s", __func__, path);
2813 if (S_ISREG(tmp_st.st_mode)) {
2814 fd = open(path, O_RDONLY);
2816 ret = llapi_file_fget_mdtidx(fd,
2817 ¶m->file_mdtindex);
2823 /* For special inode, it assumes to
2824 * reside on the same MDT with the
2827 ret = llapi_file_fget_mdtidx(fd,
2828 ¶m->file_mdtindex);
2841 if (param->type && !checked_type) {
2842 if ((st->st_mode & S_IFMT) == param->type) {
2843 if (param->exclude_type)
2846 if (!param->exclude_type)
2852 if (param->obduuid || param->mdtuuid) {
2853 if (lustre_fs && param->got_uuids &&
2854 param->st_dev != st->st_dev) {
2855 /* A lustre/lustre mount point is crossed. */
2856 param->got_uuids = 0;
2857 param->obds_printed = 0;
2858 param->obdindex = param->mdtindex = OBD_NOT_FOUND;
2861 if (lustre_fs && !param->got_uuids) {
2862 ret = setup_target_indexes(dir ? dir : parent, path,
2867 param->st_dev = st->st_dev;
2868 } else if (!lustre_fs && param->got_uuids) {
2869 /* A lustre/non-lustre mount point is crossed. */
2870 param->got_uuids = 0;
2871 param->obdindex = param->mdtindex = OBD_NOT_FOUND;
2875 if (param->check_stripesize) {
2876 decision = find_value_cmp(param->lmd->lmd_lmm.lmm_stripe_size,
2878 param->stripesize_sign,
2879 param->exclude_stripesize,
2880 param->stripesize_units, 0);
2885 if (param->check_stripecount) {
2886 decision = find_value_cmp(param->lmd->lmd_lmm.lmm_stripe_count,
2888 param->stripecount_sign,
2889 param->exclude_stripecount, 1, 0);
2894 if (param->check_layout) {
2897 found = (param->lmd->lmd_lmm.lmm_pattern & param->layout);
2898 if ((param->lmd->lmd_lmm.lmm_pattern == 0xFFFFFFFF) ||
2899 (found && param->exclude_layout) ||
2900 (!found && !param->exclude_layout)) {
2906 /* If an OBD UUID is specified but none matches, skip this file. */
2907 if ((param->obduuid && param->obdindex == OBD_NOT_FOUND) ||
2908 (param->mdtuuid && param->mdtindex == OBD_NOT_FOUND))
2911 /* If a OST or MDT UUID is given, and some OST matches,
2913 if (param->obdindex != OBD_NOT_FOUND ||
2914 param->mdtindex != OBD_NOT_FOUND) {
2915 if (param->obduuid) {
2916 if (check_obd_match(param)) {
2917 /* If no mdtuuid is given, we are done.
2918 * Otherwise, fall through to the mdtuuid
2920 if (!param->mdtuuid)
2926 if (param->mdtuuid) {
2927 if (check_mdt_match(param))
2933 if (param->check_uid) {
2934 if (st->st_uid == param->uid) {
2935 if (param->exclude_uid)
2938 if (!param->exclude_uid)
2943 if (param->check_gid) {
2944 if (st->st_gid == param->gid) {
2945 if (param->exclude_gid)
2948 if (!param->exclude_gid)
2953 if (param->check_pool) {
2954 struct lov_user_md_v3 *lmmv3 = (void *)¶m->lmd->lmd_lmm;
2956 /* empty requested pool is taken as no pool search => V1 */
2957 if (((param->lmd->lmd_lmm.lmm_magic == LOV_USER_MAGIC_V1) &&
2958 (param->poolname[0] == '\0')) ||
2959 ((param->lmd->lmd_lmm.lmm_magic == LOV_USER_MAGIC_V3) &&
2960 (strncmp(lmmv3->lmm_pool_name,
2961 param->poolname, LOV_MAXPOOLNAME) == 0)) ||
2962 ((param->lmd->lmd_lmm.lmm_magic == LOV_USER_MAGIC_V3) &&
2963 (strcmp(param->poolname, "*") == 0))) {
2964 if (param->exclude_pool)
2967 if (!param->exclude_pool)
2972 /* Check the time on mds. */
2974 if (param->atime || param->ctime || param->mtime) {
2977 for_mds = lustre_fs ? (S_ISREG(st->st_mode) &&
2978 param->lmd->lmd_lmm.lmm_stripe_count)
2980 decision = find_time_check(st, param, for_mds);
2985 /* If file still fits the request, ask ost for updated info.
2986 The regular stat is almost of the same speed as some new
2987 'glimpse-size-ioctl'. */
2989 if (param->check_size && S_ISREG(st->st_mode) &&
2990 param->lmd->lmd_lmm.lmm_stripe_count)
2994 /* For regular files with the stripe the decision may have not
2995 * been taken yet if *time or size is to be checked. */
2996 LASSERT((S_ISREG(st->st_mode) &&
2997 param->lmd->lmd_lmm.lmm_stripe_count) ||
2998 param->mdtindex != OBD_NOT_FOUND);
3000 if (param->obdindex != OBD_NOT_FOUND)
3001 print_failed_tgt(param, path, LL_STATFS_LOV);
3003 if (param->mdtindex != OBD_NOT_FOUND)
3004 print_failed_tgt(param, path, LL_STATFS_LMV);
3007 ret = ioctl(dirfd(dir), IOC_LOV_GETINFO,
3008 (void *)param->lmd);
3009 } else if (parent) {
3010 ret = ioctl(dirfd(parent), IOC_LOV_GETINFO,
3011 (void *)param->lmd);
3015 if (errno == ENOENT) {
3016 llapi_error(LLAPI_MSG_ERROR, -ENOENT,
3017 "warning: %s: %s does not exist",
3022 llapi_error(LLAPI_MSG_ERROR, ret,
3023 "%s: IOC_LOV_GETINFO on %s failed",
3029 /* Check the time on osc. */
3030 decision = find_time_check(st, param, 0);
3037 if (param->check_size)
3038 decision = find_value_cmp(st->st_size, param->size,
3039 param->size_sign, param->exclude_size,
3040 param->size_units, 0);
3042 if (decision != -1) {
3043 llapi_printf(LLAPI_MSG_NORMAL, "%s", path);
3045 llapi_printf(LLAPI_MSG_NORMAL, "%c", '\0');
3047 llapi_printf(LLAPI_MSG_NORMAL, "\n");
3051 /* Do not get down anymore? */
3052 if (param->depth == param->maxdepth)
3059 int llapi_find(char *path, struct find_param *param)
3061 return param_callback(path, cb_find_init, cb_common_fini, param);
3065 * Get MDT number that the file/directory inode referenced
3066 * by the open fd resides on.
3067 * Return 0 and mdtidx on success, or -ve errno.
3069 int llapi_file_fget_mdtidx(int fd, int *mdtidx)
3071 if (ioctl(fd, LL_IOC_GET_MDTIDX, mdtidx) < 0)
3076 static int cb_get_mdt_index(char *path, DIR *parent, DIR *d, void *data,
3077 struct dirent64 *de)
3079 struct find_param *param = (struct find_param *)data;
3083 LASSERT(parent != NULL || d != NULL);
3086 ret = llapi_file_fget_mdtidx(dirfd(d), &mdtidx);
3087 } else if (parent) {
3090 fd = open(path, O_RDONLY);
3092 ret = llapi_file_fget_mdtidx(fd, &mdtidx);
3100 if (ret == -ENODATA) {
3101 if (!param->obduuid)
3102 llapi_printf(LLAPI_MSG_NORMAL,
3103 "%s has no stripe info\n", path);
3105 } else if (ret == -ENOENT) {
3106 llapi_error(LLAPI_MSG_WARN, ret,
3107 "warning: %s: %s does not exist",
3110 } else if (ret == -ENOTTY) {
3111 llapi_error(LLAPI_MSG_ERROR, ret,
3112 "%s: '%s' not on a Lustre fs?",
3115 llapi_error(LLAPI_MSG_ERROR, ret,
3116 "error: %s: LL_IOC_GET_MDTIDX failed for %s",
3122 /* The 'LASSERT(parent != NULL || d != NULL);' guarantees
3123 * that either 'd' or 'parent' is not null.
3124 * So in all cases llapi_file_fget_mdtidx() is called,
3125 * thus initializing 'mdtidx'. */
3126 if (param->quiet || !(param->verbose & VERBOSE_DETAIL))
3127 /* coverity[uninit_use_in_call] */
3128 llapi_printf(LLAPI_MSG_NORMAL, "%d\n", mdtidx);
3130 /* coverity[uninit_use_in_call] */
3131 llapi_printf(LLAPI_MSG_NORMAL, "%s\nmdt_index:\t%d\n",
3135 /* Do not get down anymore? */
3136 if (param->depth == param->maxdepth)
3143 static int cb_getstripe(char *path, DIR *parent, DIR *d, void *data,
3144 struct dirent64 *de)
3146 struct find_param *param = (struct find_param *)data;
3149 LASSERT(parent != NULL || d != NULL);
3151 if (param->obduuid) {
3153 ret = setup_obd_uuid(d ? d : parent, path, param);
3159 if (param->get_lmv) {
3160 ret = cb_get_dirstripe(path, d, param);
3162 ret = ioctl(dirfd(d), LL_IOC_LOV_GETSTRIPE,
3163 (void *)¶m->lmd->lmd_lmm);
3166 } else if (parent) {
3167 char *fname = strrchr(path, '/');
3168 fname = (fname == NULL ? path : fname + 1);
3170 if (param->get_lmv) {
3171 llapi_printf(LLAPI_MSG_NORMAL,
3172 "%s get dirstripe information for file\n",
3177 strncpy((char *)¶m->lmd->lmd_lmm, fname, param->lumlen);
3179 ret = ioctl(dirfd(parent), IOC_MDC_GETFILESTRIPE,
3180 (void *)¶m->lmd->lmd_lmm);
3184 if (errno == ENODATA && d != NULL) {
3185 /* We need to "fake" the "use the default" values
3186 * since the lmm struct is zeroed out at this point.
3187 * The magic needs to be set in order to satisfy
3188 * a check later on in the code path.
3189 * The object_seq needs to be set for the "(Default)"
3190 * prefix to be displayed. */
3191 struct lov_user_md *lmm = ¶m->lmd->lmd_lmm;
3192 lmm->lmm_magic = LOV_USER_MAGIC_V1;
3194 ostid_set_seq(&lmm->lmm_oi,
3195 FID_SEQ_LOV_DEFAULT);
3196 lmm->lmm_stripe_count = 0;
3197 lmm->lmm_stripe_size = 0;
3198 lmm->lmm_stripe_offset = -1;
3200 } else if (errno == ENODATA && parent != NULL) {
3201 if (!param->obduuid && !param->mdtuuid)
3202 llapi_printf(LLAPI_MSG_NORMAL,
3203 "%s has no stripe info\n", path);
3205 } else if (errno == ENOENT) {
3206 llapi_error(LLAPI_MSG_WARN, -ENOENT,
3207 "warning: %s: %s does not exist",
3210 } else if (errno == ENOTTY) {
3212 llapi_error(LLAPI_MSG_ERROR, ret,
3213 "%s: '%s' not on a Lustre fs?",
3217 llapi_error(LLAPI_MSG_ERROR, ret,
3218 "error: %s: %s failed for %s",
3219 __func__, d ? "LL_IOC_LOV_GETSTRIPE" :
3220 "IOC_MDC_GETFILESTRIPE", path);
3227 if (!(param->verbose & VERBOSE_MDTINDEX))
3228 llapi_lov_dump_user_lmm(param, path, d ? 1 : 0);
3231 /* Do not get down anymore? */
3232 if (param->depth == param->maxdepth)
3239 int llapi_getstripe(char *path, struct find_param *param)
3241 return param_callback(path, (param->verbose & VERBOSE_MDTINDEX) ?
3242 cb_get_mdt_index : cb_getstripe,
3243 cb_common_fini, param);
3246 int llapi_obd_statfs(char *path, __u32 type, __u32 index,
3247 struct obd_statfs *stat_buf,
3248 struct obd_uuid *uuid_buf)
3251 char raw[OBD_MAX_IOCTL_BUFFER] = {'\0'};
3253 struct obd_ioctl_data data = { 0 };
3256 data.ioc_inlbuf1 = (char *)&type;
3257 data.ioc_inllen1 = sizeof(__u32);
3258 data.ioc_inlbuf2 = (char *)&index;
3259 data.ioc_inllen2 = sizeof(__u32);
3260 data.ioc_pbuf1 = (char *)stat_buf;
3261 data.ioc_plen1 = sizeof(struct obd_statfs);
3262 data.ioc_pbuf2 = (char *)uuid_buf;
3263 data.ioc_plen2 = sizeof(struct obd_uuid);
3265 rc = obd_ioctl_pack(&data, &rawbuf, sizeof(raw));
3267 llapi_error(LLAPI_MSG_ERROR, rc,
3268 "llapi_obd_statfs: error packing ioctl data");
3272 fd = open(path, O_RDONLY);
3273 if (errno == EISDIR)
3274 fd = open(path, O_DIRECTORY | O_RDONLY);
3277 rc = errno ? -errno : -EBADF;
3278 llapi_error(LLAPI_MSG_ERROR, rc, "error: %s: opening '%s'",
3282 rc = ioctl(fd, IOC_OBD_STATFS, (void *)rawbuf);
3284 rc = errno ? -errno : -EINVAL;
3290 #define MAX_STRING_SIZE 128
3292 int llapi_ping(char *obd_type, char *obd_name)
3294 char path[MAX_STRING_SIZE];
3298 snprintf(path, MAX_STRING_SIZE, "/proc/fs/lustre/%s/%s/ping",
3299 obd_type, obd_name);
3301 fd = open(path, O_WRONLY);
3304 llapi_error(LLAPI_MSG_ERROR, rc, "error opening %s", path);
3308 /* The purpose is to send a byte as a ping, whatever this byte is. */
3309 /* coverity[uninit_use_in_call] */
3310 rc = write(fd, buf, 1);
3320 int llapi_target_iterate(int type_num, char **obd_type,
3321 void *args, llapi_cb_t cb)
3323 char buf[MAX_STRING_SIZE];
3324 FILE *fp = fopen(DEVICES_LIST, "r");
3329 llapi_error(LLAPI_MSG_ERROR, rc, "error: opening "DEVICES_LIST);
3333 while (fgets(buf, sizeof(buf), fp) != NULL) {
3334 char *obd_type_name = NULL;
3335 char *obd_name = NULL;
3336 char *obd_uuid = NULL;
3338 struct obd_statfs osfs_buffer;
3340 while(bufp[0] == ' ')
3343 for(i = 0; i < 3; i++) {
3344 obd_type_name = strsep(&bufp, " ");
3346 obd_name = strsep(&bufp, " ");
3347 obd_uuid = strsep(&bufp, " ");
3349 memset(&osfs_buffer, 0, sizeof (osfs_buffer));
3351 for (i = 0; i < type_num; i++) {
3352 if (strcmp(obd_type_name, obd_type[i]) != 0)
3355 cb(obd_type_name, obd_name, obd_uuid, args);
3362 static void do_target_check(char *obd_type_name, char *obd_name,
3363 char *obd_uuid, void *args)
3367 rc = llapi_ping(obd_type_name, obd_name);
3368 if (rc == ENOTCONN) {
3369 llapi_printf(LLAPI_MSG_NORMAL, "%s inactive.\n", obd_name);
3371 llapi_error(LLAPI_MSG_ERROR, rc, "error: check '%s'", obd_name);
3373 llapi_printf(LLAPI_MSG_NORMAL, "%s active.\n", obd_name);
3377 int llapi_target_check(int type_num, char **obd_type, char *dir)
3379 return llapi_target_iterate(type_num, obd_type, NULL, do_target_check);
3382 #undef MAX_STRING_SIZE
3384 /* Is this a lustre fs? */
3385 int llapi_is_lustre_mnttype(const char *type)
3387 return (strcmp(type, "lustre") == 0 || strcmp(type,"lustre_lite") == 0);
3390 /* Is this a lustre client fs? */
3391 int llapi_is_lustre_mnt(struct mntent *mnt)
3393 return (llapi_is_lustre_mnttype(mnt->mnt_type) &&
3394 strstr(mnt->mnt_fsname, ":/") != NULL);
3397 int llapi_quotacheck(char *mnt, int check_type)
3402 root = opendir(mnt);
3405 llapi_error(LLAPI_MSG_ERROR, rc, "open %s failed", mnt);
3409 rc = ioctl(dirfd(root), LL_IOC_QUOTACHECK, check_type);
3417 int llapi_poll_quotacheck(char *mnt, struct if_quotacheck *qchk)
3423 root = opendir(mnt);
3426 llapi_error(LLAPI_MSG_ERROR, rc, "open %s failed", mnt);
3431 rc = ioctl(dirfd(root), LL_IOC_POLL_QUOTACHECK, qchk);
3435 if (poll_intvl < 30)
3443 int llapi_quotactl(char *mnt, struct if_quotactl *qctl)
3448 root = opendir(mnt);
3451 llapi_error(LLAPI_MSG_ERROR, rc, "open %s failed", mnt);
3455 rc = ioctl(dirfd(root), LL_IOC_QUOTACTL, qctl);
3463 static int cb_quotachown(char *path, DIR *parent, DIR *d, void *data,
3464 struct dirent64 *de)
3466 struct find_param *param = (struct find_param *)data;
3470 LASSERT(parent != NULL || d != NULL);
3472 rc = get_lmd_info(path, parent, d, param->lmd, param->lumlen);
3474 if (rc == -ENODATA) {
3475 if (!param->obduuid && !param->quiet)
3476 llapi_error(LLAPI_MSG_ERROR, -ENODATA,
3477 "%s has no stripe info", path);
3479 } else if (rc == -ENOENT) {
3485 st = ¶m->lmd->lmd_st;
3487 /* libc chown() will do extra check, and if the real owner is
3488 * the same as the ones to set, it won't fall into kernel, so
3489 * invoke syscall directly. */
3490 rc = syscall(SYS_chown, path, -1, -1);
3492 llapi_error(LLAPI_MSG_ERROR, errno,
3493 "error: chown %s", path);
3495 rc = chmod(path, st->st_mode);
3498 llapi_error(LLAPI_MSG_ERROR, rc, "error: chmod %s (%hu)",
3505 int llapi_quotachown(char *path, int flag)
3507 struct find_param param;
3509 memset(¶m, 0, sizeof(param));
3510 param.recursive = 1;
3514 return param_callback(path, cb_quotachown, NULL, ¶m);
3520 #include <sys/wait.h>
3524 static int rmtacl_notify(int ops)
3528 int found = 0, fd = 0, rc = 0;
3530 fp = setmntent(MOUNTED, "r");
3533 llapi_error(LLAPI_MSG_ERROR, rc,
3534 "error setmntent(%s)", MOUNTED);
3539 mnt = getmntent(fp);
3543 if (!llapi_is_lustre_mnt(mnt))
3546 fd = open(mnt->mnt_dir, O_RDONLY | O_DIRECTORY);
3549 llapi_error(LLAPI_MSG_ERROR, rc,
3550 "Can't open '%s'\n", mnt->mnt_dir);
3554 rc = ioctl(fd, LL_IOC_RMTACL, ops);
3558 llapi_error(LLAPI_MSG_ERROR, rc,
3559 "ioctl RMTACL on '%s' err %d\n",
3569 return ((rc != 0) ? rc : found);
3572 static char *next_token(char *p, int div)
3578 while (*p && *p != ':' && !isspace(*p))
3581 while (*p == ':' || isspace(*p))
3584 return *p ? p : NULL;
3587 static int rmtacl_name2id(char *name, int is_user)
3592 pw = getpwnam(name);
3596 return (int)(pw->pw_uid);
3600 gr = getgrnam(name);
3604 return (int)(gr->gr_gid);
3608 static int isodigit(int c)
3610 return (c >= '0' && c <= '7') ? 1 : 0;
3614 * Whether the name is just digits string (uid/gid) already or not.
3619 static int str_is_id(char *str)
3626 if (*str == 'x' || *str == 'X') { /* for Hex. */
3627 if (!isxdigit(*(++str)))
3630 while (isxdigit(*(++str)));
3631 } else if (isodigit(*str)) { /* for Oct. */
3632 while (isodigit(*(++str)));
3634 } else if (isdigit(*str)) { /* for Dec. */
3635 while (isdigit(*(++str)));
3638 return (*str == 0) ? 1 : 0;
3648 #define RMTACL_OPTNAME(name) name, sizeof(name) - 1
3650 static rmtacl_name_t rmtacl_namelist[] = {
3651 { RMTACL_OPTNAME("user:"), 1, 0 },
3652 { RMTACL_OPTNAME("group:"), 0, 0 },
3653 { RMTACL_OPTNAME("default:user:"), 1, 0 },
3654 { RMTACL_OPTNAME("default:group:"), 0, 0 },
3655 /* for --tabular option */
3656 { RMTACL_OPTNAME("user"), 1, 1 },
3657 { RMTACL_OPTNAME("group"), 0, 1 },
3661 static int rgetfacl_output(char *str)
3663 char *start = NULL, *end = NULL;
3664 int is_user = 0, n, id;
3671 for (rn = rmtacl_namelist; rn->name; rn++) {
3672 if(strncmp(str, rn->name, rn->length) == 0) {
3673 if (!rn->next_token)
3674 start = str + rn->length;
3676 start = next_token(str + rn->length, 0);
3677 is_user = rn->is_user;
3682 end = next_token(start, 1);
3683 if (end == NULL || start == end) {
3684 n = printf("%s", str);
3690 id = rmtacl_name2id(start, is_user);
3691 if (id == INVALID_ID) {
3692 if (str_is_id(start)) {
3694 n = printf("%s", str);
3697 } else if ((id == NOBODY_UID && is_user) ||
3698 (id == NOBODY_GID && !is_user)) {
3700 n = printf("%s", str);
3704 n = printf("%s%d%s", str, id, end);
3709 static int child_status(int status)
3711 return WIFEXITED(status) ? WEXITSTATUS(status) : -1;
3714 static int do_rmtacl(int argc, char *argv[], int ops, int (output_func)(char *))
3717 int fd[2], status, rc;
3724 llapi_error(LLAPI_MSG_ERROR, rc, "Can't create pipe\n");
3731 llapi_error(LLAPI_MSG_ERROR, rc, "Can't fork\n");
3736 /* child process redirects its output. */
3739 if (dup2(fd[1], 1) < 0) {
3741 llapi_error(LLAPI_MSG_ERROR, rc,
3742 "Can't dup2 %d\n", fd[1]);
3752 status = rmtacl_notify(ops);
3756 exit(execvp(argv[0], argv));
3759 /* the following is parent process */
3760 fp = fdopen(fd[0], "r");
3763 llapi_error(LLAPI_MSG_ERROR, rc, "fdopen %d failed\n", fd[0]);
3769 while (fgets(buf, PIPE_BUF, fp) != NULL) {
3770 if (output_func(buf) < 0)
3771 fprintf(stderr, "WARNING: unexpected error!\n[%s]\n",
3777 if (waitpid(pid, &status, 0) < 0) {
3779 llapi_error(LLAPI_MSG_ERROR, rc, "waitpid %d failed\n", pid);
3783 return child_status(status);
3786 int llapi_lsetfacl(int argc, char *argv[])
3788 return do_rmtacl(argc, argv, RMT_LSETFACL, NULL);
3791 int llapi_lgetfacl(int argc, char *argv[])
3793 return do_rmtacl(argc, argv, RMT_LGETFACL, NULL);
3796 int llapi_rsetfacl(int argc, char *argv[])
3798 return do_rmtacl(argc, argv, RMT_RSETFACL, NULL);
3801 int llapi_rgetfacl(int argc, char *argv[])
3803 return do_rmtacl(argc, argv, RMT_RGETFACL, rgetfacl_output);
3806 int llapi_cp(int argc, char *argv[])
3810 rc = rmtacl_notify(RMT_RSETFACL);
3814 exit(execvp(argv[0], argv));
3817 int llapi_ls(int argc, char *argv[])
3821 rc = rmtacl_notify(RMT_LGETFACL);
3825 exit(execvp(argv[0], argv));
3828 /* Print mdtname 'name' into 'buf' using 'format'. Add -MDT0000 if needed.
3829 * format must have %s%s, buf must be > 16
3830 * Eg: if name = "lustre-MDT0000", "lustre", or "lustre-MDT0000_UUID"
3831 * then buf = "lustre-MDT0000"
3833 static int get_mdtname(char *name, char *format, char *buf)
3835 char suffix[]="-MDT0000";
3836 int len = strlen(name);
3838 if ((len > 5) && (strncmp(name + len - 5, "_UUID", 5) == 0)) {
3839 name[len - 5] = '\0';
3844 if ((len <= 16) && strncmp(name + len - 8, "-MDT", 4) == 0) {
3847 /* Not enough room to add suffix */
3848 llapi_err_noerrno(LLAPI_MSG_ERROR,
3849 "MDT name too long |%s|", name);
3854 return sprintf(buf, format, name, suffix);
3857 /** ioctl on filsystem root, with mdtindex sent as data
3858 * \param mdtname path, fsname, or mdtname (lutre-MDT0004)
3859 * \param mdtidxp pointer to integer within data to be filled in with the
3860 * mdt index (0 if no mdt is specified). NULL won't be filled.
3862 int root_ioctl(const char *mdtname, int opc, void *data, int *mdtidxp,
3870 /* Take path, fsname, or MDTname. Assume MDT0000 in the former cases.
3871 Open root and parse mdt index. */
3872 if (mdtname[0] == '/') {
3874 rc = get_root_path(WANT_FD | want_error, NULL, &fd,
3875 (char *)mdtname, -1);
3877 if (get_mdtname((char *)mdtname, "%s%s", fsname) < 0)
3879 ptr = fsname + strlen(fsname) - 8;
3881 index = strtol(ptr + 4, NULL, 10);
3882 rc = get_root_path(WANT_FD | want_error, fsname, &fd, NULL, -1);
3886 llapi_err_noerrno(LLAPI_MSG_ERROR,
3887 "Can't open %s: %d\n", mdtname, rc);
3894 rc = ioctl(fd, opc, data);
3899 if (rc && want_error)
3900 llapi_error(LLAPI_MSG_ERROR, rc, "ioctl %d err %d", opc, rc);
3906 /****** Changelog API ********/
3908 static int changelog_ioctl(const char *mdtname, int opc, int id,
3909 long long recno, int flags)
3911 struct ioc_changelog data;
3915 data.icc_recno = recno;
3916 data.icc_flags = flags;
3917 idx = (int *)(&data.icc_mdtindex);
3919 return root_ioctl(mdtname, opc, &data, idx, WANT_ERROR);
3922 #define CHANGELOG_PRIV_MAGIC 0xCA8E1080
3923 struct changelog_private {
3926 lustre_kernelcomm kuc;
3929 /** Start reading from a changelog
3930 * @param priv Opaque private control structure
3931 * @param flags Start flags (e.g. CHANGELOG_FLAG_BLOCK)
3932 * @param device Report changes recorded on this MDT
3933 * @param startrec Report changes beginning with this record number
3934 * (just call llapi_changelog_fini when done; don't need an endrec)
3936 int llapi_changelog_start(void **priv, int flags, const char *device,
3939 struct changelog_private *cp;
3942 /* Set up the receiver control struct */
3943 cp = calloc(1, sizeof(*cp));
3947 cp->magic = CHANGELOG_PRIV_MAGIC;
3950 /* Set up the receiver */
3951 rc = libcfs_ukuc_start(&cp->kuc, 0 /* no group registration */);
3957 /* Tell the kernel to start sending */
3958 rc = changelog_ioctl(device, OBD_IOC_CHANGELOG_SEND, cp->kuc.lk_wfd,
3960 /* Only the kernel reference keeps the write side open */
3961 close(cp->kuc.lk_wfd);
3962 cp->kuc.lk_wfd = LK_NOFD;
3964 /* frees and clears priv */
3965 llapi_changelog_fini(priv);
3976 /** Finish reading from a changelog */
3977 int llapi_changelog_fini(void **priv)
3979 struct changelog_private *cp = (struct changelog_private *)*priv;
3981 if (!cp || (cp->magic != CHANGELOG_PRIV_MAGIC))
3984 libcfs_ukuc_stop(&cp->kuc);
3990 /** Convert a changelog_rec to changelog_ext_rec, in this way client can treat
3991 * all records in the format of changelog_ext_rec, this can make record
3994 static inline int changelog_extend_rec(struct changelog_ext_rec *ext)
3996 if (!CHANGELOG_REC_EXTENDED(ext)) {
3997 struct changelog_rec *rec = (struct changelog_rec *)ext;
3999 memmove(ext->cr_name, rec->cr_name, rec->cr_namelen);
4000 fid_zero(&ext->cr_sfid);
4001 fid_zero(&ext->cr_spfid);
4008 /** Read the next changelog entry
4009 * @param priv Opaque private control structure
4010 * @param rech Changelog record handle; record will be allocated here
4011 * @return 0 valid message received; rec is set
4015 int llapi_changelog_recv(void *priv, struct changelog_ext_rec **rech)
4017 struct changelog_private *cp = (struct changelog_private *)priv;
4018 struct kuc_hdr *kuch;
4021 if (!cp || (cp->magic != CHANGELOG_PRIV_MAGIC))
4025 kuch = malloc(KUC_CHANGELOG_MSG_MAXSIZE);
4030 rc = libcfs_ukuc_msg_get(&cp->kuc, (char *)kuch,
4031 KUC_CHANGELOG_MSG_MAXSIZE,
4032 KUC_TRANSPORT_CHANGELOG);
4036 if ((kuch->kuc_transport != KUC_TRANSPORT_CHANGELOG) ||
4037 ((kuch->kuc_msgtype != CL_RECORD) &&
4038 (kuch->kuc_msgtype != CL_EOF))) {
4039 llapi_err_noerrno(LLAPI_MSG_ERROR,
4040 "Unknown changelog message type %d:%d\n",
4041 kuch->kuc_transport, kuch->kuc_msgtype);
4046 if (kuch->kuc_msgtype == CL_EOF) {
4047 if (cp->flags & CHANGELOG_FLAG_FOLLOW) {
4056 /* Our message is a changelog_ext_rec. Use pointer math to skip
4057 * kuch_hdr and point directly to the message payload.
4059 *rech = (struct changelog_ext_rec *)(kuch + 1);
4060 changelog_extend_rec(*rech);
4070 /** Release the changelog record when done with it. */
4071 int llapi_changelog_free(struct changelog_ext_rec **rech)
4074 /* We allocated memory starting at the kuc_hdr, but passed
4075 * the consumer a pointer to the payload.
4076 * Use pointer math to get back to the header.
4078 struct kuc_hdr *kuch = (struct kuc_hdr *)*rech - 1;
4085 int llapi_changelog_clear(const char *mdtname, const char *idstr,
4091 llapi_err_noerrno(LLAPI_MSG_ERROR,
4092 "can't purge negative records\n");
4096 id = strtol(idstr + strlen(CHANGELOG_USER_PREFIX), NULL, 10);
4097 if ((id == 0) || (strncmp(idstr, CHANGELOG_USER_PREFIX,
4098 strlen(CHANGELOG_USER_PREFIX)) != 0)) {
4099 llapi_err_noerrno(LLAPI_MSG_ERROR,
4100 "expecting id of the form '"
4101 CHANGELOG_USER_PREFIX
4102 "<num>'; got '%s'\n", idstr);
4106 return changelog_ioctl(mdtname, OBD_IOC_CHANGELOG_CLEAR, id, endrec, 0);
4109 int llapi_fid2path(const char *device, const char *fidstr, char *buf,
4110 int buflen, long long *recno, int *linkno)
4113 struct getinfo_fid2path *gf;
4116 while (*fidstr == '[')
4119 sscanf(fidstr, SFID, RFID(&fid));
4120 if (!fid_is_sane(&fid)) {
4121 llapi_err_noerrno(LLAPI_MSG_ERROR,
4122 "bad FID format [%s], should be "DFID"\n",
4123 fidstr, (__u64)1, 2, 0);
4127 gf = malloc(sizeof(*gf) + buflen);
4131 gf->gf_recno = *recno;
4132 gf->gf_linkno = *linkno;
4133 gf->gf_pathlen = buflen;
4135 /* Take path or fsname */
4136 rc = root_ioctl(device, OBD_IOC_FID2PATH, gf, NULL, 0);
4139 llapi_error(LLAPI_MSG_ERROR, rc, "ioctl err %d", rc);
4141 memcpy(buf, gf->gf_path, gf->gf_pathlen);
4142 *recno = gf->gf_recno;
4143 *linkno = gf->gf_linkno;
4150 static int fid_from_lma(const char *path, const int fd, lustre_fid *fid)
4153 struct lustre_mdt_attrs *lma;
4157 rc = fgetxattr(fd, XATTR_NAME_LMA, buf, sizeof(buf));
4159 rc = lgetxattr(path, XATTR_NAME_LMA, buf, sizeof(buf));
4162 lma = (struct lustre_mdt_attrs *)buf;
4163 fid_le_to_cpu(fid, &lma->lma_self_fid);
4167 int llapi_fd2fid(const int fd, lustre_fid *fid)
4171 memset(fid, 0, sizeof(*fid));
4173 rc = ioctl(fd, LL_IOC_PATH2FID, fid) < 0 ? -errno : 0;
4174 if (rc == -EINVAL || rc == -ENOTTY)
4175 rc = fid_from_lma(NULL, fd, fid);
4180 int llapi_path2fid(const char *path, lustre_fid *fid)
4184 memset(fid, 0, sizeof(*fid));
4185 fd = open(path, O_RDONLY | O_NONBLOCK | O_NOFOLLOW);
4187 if (errno == ELOOP || errno == ENXIO)
4188 return fid_from_lma(path, -1, fid);
4192 rc = llapi_fd2fid(fd, fid);
4193 if (rc == -EINVAL || rc == -ENOTTY)
4194 rc = fid_from_lma(path, -1, fid);
4200 int llapi_get_connect_flags(const char *mnt, __u64 *flags)
4205 root = opendir(mnt);
4208 llapi_error(LLAPI_MSG_ERROR, rc, "open %s failed", mnt);
4212 rc = ioctl(dirfd(root), LL_IOC_GET_CONNECT_FLAGS, flags);
4215 llapi_error(LLAPI_MSG_ERROR, rc,
4216 "ioctl on %s for getting connect flags failed", mnt);
4222 int llapi_get_version(char *buffer, int buffer_size,
4227 struct obd_ioctl_data *data = (struct obd_ioctl_data *)buffer;
4229 fd = open(OBD_DEV_PATH, O_RDONLY);
4233 memset(buffer, 0, buffer_size);
4234 data->ioc_version = OBD_IOCTL_VERSION;
4235 data->ioc_inllen1 = buffer_size - cfs_size_round(sizeof(*data));
4236 data->ioc_inlbuf1 = buffer + cfs_size_round(sizeof(*data));
4237 data->ioc_len = obd_ioctl_packlen(data);
4239 rc = ioctl(fd, OBD_GET_VERSION, buffer);
4246 *version = data->ioc_bulk;
4251 * Get a 64-bit value representing the version of file data pointed by fd.
4253 * Each write or truncate, flushed on OST, will change this value. You can use
4254 * this value to verify if file data was modified. This only checks the file
4255 * data, not metadata.
4257 * \param flags 0: no flush pages, usually used it the process has already
4259 * LL_DV_RD_FLUSH: OSTs will take LCK_PR to flush dirty pages
4261 * LL_DV_WR_FLUSH: OSTs will take LCK_PW to flush all caching
4262 * pages from clients.
4264 * \retval 0 on success.
4265 * \retval -errno on error.
4267 int llapi_get_data_version(int fd, __u64 *data_version, __u64 flags)
4270 struct ioc_data_version idv;
4272 idv.idv_flags = flags;
4274 rc = ioctl(fd, LL_IOC_DATA_VERSION, &idv);
4278 *data_version = idv.idv_version;
4284 * Create a volatile file and open it for write:
4285 * - file is created as a standard file in the directory
4286 * - file does not appears in directory and directory mtime does not change
4287 * - file is removed at close
4288 * - file modes are rw-------, if user wants another one it must use fchmod()
4289 * \param directory Directory where the file is created
4290 * \param idx MDT index on which the file is created
4291 * \param open_flags Standard open flags
4293 * \retval 0 on success.
4294 * \retval -errno on error.
4296 int llapi_create_volatile_idx(char *directory, int idx, int open_flags)
4298 char file_path[PATH_MAX];
4299 char filename[PATH_MAX];
4304 fd = open("/dev/urandom", O_RDONLY);
4306 llapi_error(LLAPI_MSG_ERROR, errno,
4307 "Cannot open /dev/urandom\n");
4310 rc = read(fd, &random, sizeof(random));
4312 if (rc < sizeof(random)) {
4313 llapi_error(LLAPI_MSG_ERROR, errno,
4314 "cannot read %zu bytes from /dev/urandom",
4319 snprintf(filename, sizeof(filename),
4320 LUSTRE_VOLATILE_HDR"::%.4X", random);
4322 snprintf(filename, sizeof(filename),
4323 LUSTRE_VOLATILE_HDR":%.4X:%.4X", idx, random);
4325 rc = snprintf(file_path, sizeof(file_path),
4326 "%s/%s", directory, filename);
4327 if (rc >= sizeof(file_path))
4330 fd = open(file_path, O_RDWR | O_CREAT | open_flags, S_IRUSR | S_IWUSR);
4332 llapi_error(LLAPI_MSG_ERROR, errno,
4333 "Cannot create volatile file %s in %s\n",
4334 filename + LUSTRE_VOLATILE_HDR_LEN,
4342 * Swap the layouts between 2 file descriptors
4343 * the 2 files must be open in write
4344 * first fd received the ioctl, second fd is passed as arg
4345 * this is assymetric but avoid use of root path for ioctl
4347 int llapi_fswap_layouts(int fd1, int fd2, __u64 dv1, __u64 dv2, __u64 flags)
4349 struct lustre_swap_layouts lsl;
4352 srandom(time(NULL));
4354 lsl.sl_flags = flags;
4355 lsl.sl_gid = random();
4358 rc = ioctl(fd1, LL_IOC_LOV_SWAP_LAYOUTS, &lsl);
4365 * Swap the layouts between 2 files
4366 * the 2 files are open in write
4368 int llapi_swap_layouts(const char *path1, const char *path2,
4369 __u64 dv1, __u64 dv2, __u64 flags)
4373 fd1 = open(path1, O_WRONLY | O_LOV_DELAY_CREATE);
4376 llapi_error(LLAPI_MSG_ERROR, rc,
4377 "error: cannot open '%s' for write", path1);
4381 fd2 = open(path2, O_WRONLY | O_LOV_DELAY_CREATE);
4384 llapi_error(LLAPI_MSG_ERROR, rc,
4385 "error: cannot open '%s' for write", path2);
4389 rc = llapi_fswap_layouts(fd1, fd2, dv1, dv2, flags);
4391 llapi_error(LLAPI_MSG_ERROR, rc,
4392 "error: cannot swap layout between '%s' and '%s'\n",