X-Git-Url: https://git.whamcloud.com/?p=fs%2Flustre-release.git;a=blobdiff_plain;f=lustre%2Futils%2Flfs.c;h=767c5d4a1c444e79a4870f9c49d7bfc5b580d8e7;hp=3e42beb045de45fe76836029895e805d156d9bff;hb=849d7d5b1b4cabb7578c3ab5aaf271e90dd33864;hpb=cbade14c3f853c4f61f7878c34af6c2cf0baafc8 diff --git a/lustre/utils/lfs.c b/lustre/utils/lfs.c index 3e42beb..767c5d4 100644 --- a/lustre/utils/lfs.c +++ b/lustre/utils/lfs.c @@ -27,7 +27,7 @@ * Copyright (c) 2003, 2010, Oracle and/or its affiliates. All rights reserved. * Use is subject to license terms. * - * Copyright (c) 2011, 2013, Intel Corporation. + * Copyright (c) 2011, 2014, Intel Corporation. */ /* * This file is part of Lustre, http://www.lustre.org/ @@ -50,11 +50,14 @@ #include #include #include +#include #include #include #include #include +#include #include +#include #include #include #include @@ -65,12 +68,11 @@ # include #endif -#include -#include -#include +#include +#include +#include #include #include -#include "obdctl.h" /* all functions */ static int lfs_setstripe(int argc, char **argv); @@ -85,9 +87,6 @@ static int lfs_df(int argc, char **argv); static int lfs_getname(int argc, char **argv); static int lfs_check(int argc, char **argv); #ifdef HAVE_SYS_QUOTA_H -static int lfs_quotacheck(int argc, char **argv); -static int lfs_quotaon(int argc, char **argv); -static int lfs_quotaoff(int argc, char **argv); static int lfs_setquota(int argc, char **argv); static int lfs_quota(int argc, char **argv); #endif @@ -117,21 +116,46 @@ static int lfs_hsm_cancel(int argc, char **argv); static int lfs_swap_layouts(int argc, char **argv); static int lfs_mv(int argc, char **argv); -#define SETSTRIPE_USAGE(_cmd, _tgt) \ - "usage: "_cmd" [--stripe-count|-c ]\n"\ - " [--stripe-index|-i ]\n"\ - " [--stripe-size|-S ]\n"\ - " [--pool|-p ]\n"\ - " [--block|-b] "_tgt"\n"\ - "\tstripe_size: Number of bytes on each OST (0 filesystem default)\n"\ - "\t Can be specified with k, m or g (in KB, MB and GB\n"\ - "\t respectively)\n"\ - "\tstart_ost_idx: OST index of first stripe (-1 default)\n"\ - "\tstripe_count: Number of OSTs to stripe over (0 default, -1 all)\n"\ - "\tpool_name: Name of OST pool to use (default none)\n"\ - "\tblock: Block file access during data migration" - -/* all avaialable commands */ +/* Setstripe and migrate share mostly the same parameters */ +#define SSM_CMD_COMMON(cmd) \ + "usage: "cmd" [--stripe-count|-c ]\n" \ + " [--stripe-index|-i ]\n" \ + " [--stripe-size|-S ]\n" \ + " [--pool|-p ]\n" \ + " [--ost-list|-o ]\n" + +#define SSM_HELP_COMMON \ + "\tstripe_size: Number of bytes on each OST (0 filesystem default)\n" \ + "\t Can be specified with k, m or g (in KB, MB and GB\n" \ + "\t respectively)\n" \ + "\tstart_ost_idx: OST index of first stripe (-1 default)\n" \ + "\tstripe_count: Number of OSTs to stripe over (0 default, -1 all)\n" \ + "\tpool_name: Name of OST pool to use (default none)\n" \ + "\tost_indices: List of OST indices, can be repeated multiple times\n"\ + "\t Indices be specified in a format of:\n" \ + "\t -o ,-,\n" \ + "\t Or:\n" \ + "\t -o -o - -o \n" \ + "\t If --pool is set with --ost-list, then the OSTs\n" \ + "\t must be the members of the pool." + +#define SETSTRIPE_USAGE \ + SSM_CMD_COMMON("setstripe") \ + " \n" \ + SSM_HELP_COMMON + +#define MIGRATE_USAGE \ + SSM_CMD_COMMON("migrate ") \ + " [--block|-b]\n" \ + " \n" \ + SSM_HELP_COMMON \ + "\n" \ + "\tblock: Block file access during data migration\n" \ + +static const char *progname; +static bool file_lease_supported = true; + +/* all available commands */ command_t cmdlist[] = { {"setstripe", lfs_setstripe, 0, "Create a new file with a specific striping pattern or\n" @@ -139,7 +163,7 @@ command_t cmdlist[] = { "delete the default striping pattern from an existing directory\n" "usage: setstripe -d (to delete default striping)\n"\ " or\n" - SETSTRIPE_USAGE("setstripe", "")}, + SETSTRIPE_USAGE}, {"getstripe", lfs_getstripe, 0, "To list the striping info for a given file or files in a\n" "directory or recursively for all files in a directory tree.\n" @@ -150,14 +174,18 @@ command_t cmdlist[] = { " [--layout|-L]\n" " ..."}, {"setdirstripe", lfs_setdirstripe, 0, - "To create a remote directory on a specified MDT.\n" + "To create a striped directory on a specified MDT. This can only\n" + "be done on MDT0 with the right of administrator.\n" "usage: setdirstripe <--count|-c stripe_count>\n" - "[--index|-i mdt_index] [--hash-type|-t hash_type]\n" - "[--default_stripe|-D ] \n" + " [--index|-i mdt_index] [--hash-type|-t hash_type]\n" + " [--default_stripe|-D ] [--mode|-m mode] \n" "\tstripe_count: stripe count of the striped directory\n" "\tmdt_index: MDT index of first stripe\n" - "\thash_type: hash type of the striped directory\n" - "\tdefault_stripe: set default dirstripe of the directory\n"}, + "\thash_type: hash type of the striped directory. Hash types:\n" + " fnv_1a_64 FNV-1a hash algorithm (default)\n" + " all_char sum of characters % MDT_COUNT (not recommended)\n" + "\tdefault_stripe: set default dirstripe of the directory\n" + "\tmode: the mode of the directory\n"}, {"getdirstripe", lfs_getdirstripe, 0, "To list the striping info for a given directory\n" "or recursively for all directories in a directory tree.\n" @@ -165,10 +193,18 @@ command_t cmdlist[] = { " [--count|-c ] [--index|-i ] [--raw|-R]\n" " [--recursive | -r] [ --default_stripe | -D ] "}, {"mkdir", lfs_setdirstripe, 0, - "To create a remote directory on a specified MDT. And this can only\n" - "be done on MDT0 by administrator.\n" - "usage: mkdir <--index|-i mdt_index> \n" - "\tmdt_index: MDT index of the remote directory.\n"}, + "To create a striped directory on a specified MDT. This can only\n" + "be done on MDT0 with the right of administrator.\n" + "usage: mkdir <--count|-c stripe_count>\n" + " [--index|-i mdt_index] [--hash-type|-t hash_type]\n" + " [--default_stripe|-D ] [--mode|-m mode] \n" + "\tstripe_count: stripe count of the striped directory\n" + "\tmdt_index: MDT index of first stripe\n" + "\thash_type: hash type of the striped directory. Hash types:\n" + " fnv_1a_64 FNV-1a hash algorithm (default)\n" + " all_char sum of characters % MDT_COUNT (not recommended)\n" + "\tdefault_stripe: set default dirstripe of the directory\n" + "\tmode: the mode of the directory\n"}, {"rm_entry", lfs_rmentry, 0, "To remove the name entry of the remote directory. Note: This\n" "command will only delete the name entry, i.e. the remote directory\n" @@ -214,16 +250,6 @@ command_t cmdlist[] = { "[for specified path only]\n" "Usage: getname [-h]|[path ...] "}, #ifdef HAVE_SYS_QUOTA_H - {"quotacheck", lfs_quotacheck, 0, - "Scan the specified filesystem for disk usage, and create,\n" - "or update quota files. Deprecated as of 2.4.0.\n" - "usage: quotacheck [ -ug ] "}, - {"quotaon", lfs_quotaon, 0, "Turn filesystem" - " quotas on. Deprecated as of 2.4.0.\n" - "usage: quotaon [ -ugf ] "}, - {"quotaoff", lfs_quotaoff, 0, "Turn filesystem" - " quotas off. Deprecated as of 2.4.0.\n" - "usage: quotaoff [ -ug ] "}, {"setquota", lfs_setquota, 0, "Set filesystem quotas.\n" "usage: setquota <-u|-g> |||\n" " -b -B \n" @@ -288,7 +314,7 @@ command_t cmdlist[] = { "usage: fid2path [--link ] ..." /* [ --rec ] */ }, {"path2fid", lfs_path2fid, 0, "Display the fid(s) for a given path(s).\n" - "usage: path2fid ..."}, + "usage: path2fid [--parents] ..."}, {"data_version", lfs_data_version, 0, "Display file data version for " "a given path.\n" "usage: data_version -[n|r|w] "}, {"hsm_state", lfs_hsm_state, 0, "Display the HSM information (states, " @@ -314,102 +340,354 @@ command_t cmdlist[] = { "usage: hsm_release [--filelist FILELIST] [--data DATA] ..."}, {"hsm_remove", lfs_hsm_remove, 0, "Remove file copy from external storage.\n" - "usage: hsm_remove [--filelist FILELIST] [--data DATA] ..."}, + "usage: hsm_remove [--filelist FILELIST] [--data DATA]\n" + " [--mntpath MOUNTPATH] [--archive NUM] ...\n" + "\n" + "Note: To remove files from the archive that have been deleted on\n" + "Lustre, set mntpath and optionally archive. In that case, all the\n" + "positional arguments and entries in the file list must be FIDs." + }, {"hsm_cancel", lfs_hsm_cancel, 0, "Cancel requests related to specified files.\n" "usage: hsm_cancel [--filelist FILELIST] [--data DATA] ..."}, {"swap_layouts", lfs_swap_layouts, 0, "Swap layouts between 2 files.\n" "usage: swap_layouts "}, - {"migrate", lfs_setstripe, 0, "migrate file from one OST layout to " - "another (may be not safe with concurent writes).\n" - SETSTRIPE_USAGE("migrate ", "")}, + {"migrate", lfs_setstripe, 0, + "migrate file/directory between MDTs, or migrate file from one OST " + "layout\nto another (may be not safe with concurent writes).\n" + "usage: migrate [--mdt-index|-m ] ]\n" + "\tmdt_idx: MDT index to migrate to\n" + " or\n" + MIGRATE_USAGE}, {"mv", lfs_mv, 0, - "To move directories between MDTs.\n" + "To move directories between MDTs. This command is deprecated, " + "use \"migrate\" instead.\n" "usage: mv [--mdt-index|-M] " "[--verbose|-v]\n"}, {"help", Parser_help, 0, "help"}, {"exit", Parser_quit, 0, "quit"}, {"quit", Parser_quit, 0, "quit"}, + {"--version", Parser_version, 0, + "output build version of the utility and exit"}, { 0, 0, 0, NULL } }; + #define MIGRATION_BLOCKS 1 -static int lfs_migrate(char *name, unsigned long long stripe_size, - int stripe_offset, int stripe_count, - int stripe_pattern, char *pool_name, - __u64 migration_flags) +/** + * Internal helper for migrate_copy_data(). Check lease and report error if + * need be. + * + * \param[in] fd File descriptor on which to check the lease. + * \param[out] lease_broken Set to true if the lease was broken. + * \param[in] group_locked Whether a group lock was taken or not. + * \param[in] path Name of the file being processed, for error + * reporting + * + * \retval 0 Migration can keep on going. + * \retval -errno Error occurred, abort migration. + */ +static int check_lease(int fd, bool *lease_broken, bool group_locked, + const char *path) +{ + int rc; + + if (!file_lease_supported) + return 0; + + rc = llapi_lease_check(fd); + if (rc > 0) + return 0; /* llapi_check_lease returns > 0 on success. */ + + if (!group_locked) { + fprintf(stderr, "%s: cannot migrate '%s': file busy\n", + progname, path); + rc = rc ? rc : -EAGAIN; + } else { + fprintf(stderr, "%s: external attempt to access file '%s' " + "blocked until migration ends.\n", progname, path); + rc = 0; + } + *lease_broken = true; + return rc; +} + +static int migrate_copy_data(int fd_src, int fd_dst, size_t buf_size, + bool group_locked, const char *fname) { - int fd, fdv; - char volatile_file[PATH_MAX]; + void *buf = NULL; + ssize_t rsize = -1; + ssize_t wsize = 0; + size_t rpos = 0; + size_t wpos = 0; + off_t bufoff = 0; + int rc; + bool lease_broken = false; + + /* Use a page-aligned buffer for direct I/O */ + rc = posix_memalign(&buf, getpagesize(), buf_size); + if (rc != 0) + return -rc; + + while (1) { + /* read new data only if we have written all + * previously read data */ + if (wpos == rpos) { + if (!lease_broken) { + rc = check_lease(fd_src, &lease_broken, + group_locked, fname); + if (rc < 0) + goto out; + } + rsize = read(fd_src, buf, buf_size); + if (rsize < 0) { + rc = -errno; + fprintf(stderr, "%s: %s: read failed: %s\n", + progname, fname, strerror(-rc)); + goto out; + } + rpos += rsize; + bufoff = 0; + } + /* eof ? */ + if (rsize == 0) + break; + + wsize = write(fd_dst, buf + bufoff, rpos - wpos); + if (wsize < 0) { + rc = -errno; + fprintf(stderr, + "%s: %s: write failed on volatile: %s\n", + progname, fname, strerror(-rc)); + goto out; + } + wpos += wsize; + bufoff += wsize; + } + + rc = fsync(fd_dst); + if (rc < 0) { + rc = -errno; + fprintf(stderr, "%s: %s: fsync failed: %s\n", + progname, fname, strerror(-rc)); + } + +out: + free(buf); + return rc; +} + +static int migrate_copy_timestamps(int fdv, const struct stat *st) +{ + struct timeval tv[2] = { + {.tv_sec = st->st_atime}, + {.tv_sec = st->st_mtime} + }; + + return futimes(fdv, tv); +} + +static int migrate_block(int fd, int fdv, const struct stat *st, + size_t buf_size, const char *name) +{ + __u64 dv1; + int gid; + int rc; + int rc2; + + rc = llapi_get_data_version(fd, &dv1, LL_DV_RD_FLUSH); + if (rc < 0) { + fprintf(stderr, "%s: %s: cannot get dataversion: %s\n", + progname, name, strerror(-rc)); + return rc; + } + + do + gid = random(); + while (gid == 0); + + /* The grouplock blocks all concurrent accesses to the file. + * It has to be taken after llapi_get_data_version as it would + * block it too. */ + rc = llapi_group_lock(fd, gid); + if (rc < 0) { + fprintf(stderr, "%s: %s: cannot get group lock: %s\n", + progname, name, strerror(-rc)); + return rc; + } + + rc = migrate_copy_data(fd, fdv, buf_size, true, name); + if (rc < 0) { + fprintf(stderr, "%s: %s: data copy failed\n", progname, name); + goto out_unlock; + } + + /* Make sure we keep original atime/mtime values */ + rc = migrate_copy_timestamps(fdv, st); + if (rc < 0) { + fprintf(stderr, "%s: %s: timestamp copy failed\n", + progname, name); + goto out_unlock; + } + + /* swap layouts + * for a migration we need to check data version on file did + * not change. + * + * Pass in gid=0 since we already own grouplock. */ + rc = llapi_fswap_layouts_grouplock(fd, fdv, dv1, 0, 0, + SWAP_LAYOUTS_CHECK_DV1); + if (rc == -EAGAIN) { + fprintf(stderr, "%s: %s: dataversion changed during copy, " + "migration aborted\n", progname, name); + goto out_unlock; + } else if (rc < 0) { + fprintf(stderr, "%s: %s: cannot swap layouts: %s\n", progname, + name, strerror(-rc)); + goto out_unlock; + } + +out_unlock: + rc2 = llapi_group_unlock(fd, gid); + if (rc2 < 0 && rc == 0) { + fprintf(stderr, "%s: %s: putting group lock failed: %s\n", + progname, name, strerror(-rc2)); + rc = rc2; + } + + return rc; +} + +static int migrate_nonblock(int fd, int fdv, const struct stat *st, + size_t buf_size, const char *name) +{ + __u64 dv1; + __u64 dv2; + int rc; + + rc = llapi_get_data_version(fd, &dv1, LL_DV_RD_FLUSH); + if (rc < 0) { + fprintf(stderr, "%s: %s: cannot get data version: %s\n", + progname, name, strerror(-rc)); + return rc; + } + + rc = migrate_copy_data(fd, fdv, buf_size, false, name); + if (rc < 0) { + fprintf(stderr, "%s: %s: data copy failed\n", progname, name); + return rc; + } + + rc = llapi_get_data_version(fd, &dv2, LL_DV_RD_FLUSH); + if (rc != 0) { + fprintf(stderr, "%s: %s: cannot get data version: %s\n", + progname, name, strerror(-rc)); + return rc; + } + + if (dv1 != dv2) { + rc = -EAGAIN; + fprintf(stderr, "%s: %s: data version changed during " + "migration\n", + progname, name); + return rc; + } + + /* Make sure we keep original atime/mtime values */ + rc = migrate_copy_timestamps(fdv, st); + if (rc < 0) { + fprintf(stderr, "%s: %s: timestamp copy failed\n", + progname, name); + return rc; + } + + /* Atomically put lease, swap layouts and close. + * for a migration we need to check data version on file did + * not change. */ + rc = llapi_fswap_layouts(fd, fdv, 0, 0, SWAP_LAYOUTS_CLOSE); + if (rc < 0) { + fprintf(stderr, "%s: %s: cannot swap layouts: %s\n", + progname, name, strerror(-rc)); + return rc; + } + + return 0; +} + +static int lfs_migrate(char *name, __u64 migration_flags, + struct llapi_stripe_param *param) +{ + int fd = -1; + int fdv = -1; + char volatile_file[PATH_MAX + + LUSTRE_VOLATILE_HDR_LEN + 4]; char parent[PATH_MAX]; char *ptr; int rc; - __u64 dv1; struct lov_user_md *lum = NULL; - int lumsz; - int bufsz; - void *buf = NULL; - int rsize, wsize; - __u64 rpos, wpos, bufoff; - int gid = 0, sz; - int have_gl = 0; - struct stat st, stv; + int lum_size; + int buf_size; + bool have_lease_rdlck = false; + struct stat st; + struct stat stv; /* find the right size for the IO and allocate the buffer */ - lumsz = lov_user_md_size(LOV_MAX_STRIPE_COUNT, LOV_USER_MAGIC_V3); - lum = malloc(lumsz); + lum_size = lov_user_md_size(LOV_MAX_STRIPE_COUNT, LOV_USER_MAGIC_V3); + lum = malloc(lum_size); if (lum == NULL) { rc = -ENOMEM; goto free; } rc = llapi_file_get_stripe(name, lum); - /* failure can come from may case and some may be not real error + /* failure can happen for many reasons and some may be not real errors * (eg: no stripe) - * in case of a real error, a later call will failed with a better + * in case of a real error, a later call will fail with better * error management */ if (rc < 0) - bufsz = 1024*1024; + buf_size = 1024 * 1024; else - bufsz = lum->lmm_stripe_size; - rc = posix_memalign(&buf, getpagesize(), bufsz); - if (rc != 0) { - rc = -rc; + buf_size = lum->lmm_stripe_size; + + /* open file, direct io */ + /* even if the file is only read, WR mode is nedeed to allow + * layout swap on fd */ + fd = open(name, O_RDWR | O_DIRECT); + if (fd == -1) { + rc = -errno; + fprintf(stderr, "%s: %s: cannot open: %s\n", progname, name, + strerror(-rc)); goto free; } - if (migration_flags & MIGRATION_BLOCKS) { - /* generate a random id for the grouplock */ - fd = open("/dev/urandom", O_RDONLY); - if (fd == -1) { - rc = -errno; - fprintf(stderr, "cannot open /dev/urandom (%s)\n", - strerror(-rc)); - goto free; - } - sz = sizeof(gid); - rc = read(fd, &gid, sz); - close(fd); - if (rc < sz) { - rc = -errno; - fprintf(stderr, "cannot read %d bytes from" - " /dev/urandom (%s)\n", sz, strerror(-rc)); - goto free; + if (file_lease_supported) { + rc = llapi_lease_get(fd, LL_LEASE_RDLCK); + if (rc == -EOPNOTSUPP) { + /* Older servers do not support file lease. + * Disable related checks. This opens race conditions + * as explained in LU-4840 */ + file_lease_supported = false; + } else if (rc < 0) { + fprintf(stderr, "%s: %s: cannot get open lease: %s\n", + progname, name, strerror(-rc)); + goto error; + } else { + have_lease_rdlck = true; } } /* search for file directory pathname */ if (strlen(name) > sizeof(parent)-1) { rc = -E2BIG; - goto free; + goto error; } strncpy(parent, name, sizeof(parent)); ptr = strrchr(parent, '/'); if (ptr == NULL) { if (getcwd(parent, sizeof(parent)) == NULL) { rc = -errno; - goto free; + goto error; } } else { if (ptr == parent) @@ -417,30 +695,25 @@ static int lfs_migrate(char *name, unsigned long long stripe_size, else *ptr = '\0'; } - sprintf(volatile_file, "%s/%s::", parent, LUSTRE_VOLATILE_HDR); + + rc = snprintf(volatile_file, sizeof(volatile_file), "%s/%s::", parent, + LUSTRE_VOLATILE_HDR); + if (rc >= sizeof(volatile_file)) { + rc = -E2BIG; + goto error; + } /* create, open a volatile file, use caching (ie no directio) */ /* exclusive create is not needed because volatile files cannot * conflict on name by construction */ - fdv = llapi_file_open_pool(volatile_file, O_CREAT | O_WRONLY, - 0644, stripe_size, stripe_offset, - stripe_count, stripe_pattern, pool_name); + fdv = llapi_file_open_param(volatile_file, O_CREAT | O_WRONLY, 0644, + param); if (fdv < 0) { rc = fdv; - fprintf(stderr, "cannot create volatile file in %s (%s)\n", - parent, strerror(-rc)); - goto free; - } - - /* open file, direct io */ - /* even if the file is only read, WR mode is nedeed to allow - * layout swap on fd */ - fd = open(name, O_RDWR | O_DIRECT); - if (fd == -1) { - rc = -errno; - fprintf(stderr, "cannot open %s (%s)\n", name, strerror(-rc)); - close(fdv); - goto free; + fprintf(stderr, "%s: %s: cannot create volatile file in" + " directory: %s\n", + progname, parent, strerror(-rc)); + goto error; } /* Not-owner (root?) special case. @@ -450,152 +723,166 @@ static int lfs_migrate(char *name, unsigned long long stripe_size, rc = fstat(fd, &st); if (rc != 0) { rc = -errno; - fprintf(stderr, "cannot stat %s (%s)\n", name, + fprintf(stderr, "%s: %s: cannot stat: %s\n", progname, name, strerror(errno)); goto error; } rc = fstat(fdv, &stv); if (rc != 0) { rc = -errno; - fprintf(stderr, "cannot stat %s (%s)\n", volatile_file, - strerror(errno)); + fprintf(stderr, "%s: %s: cannot stat: %s\n", progname, + volatile_file, strerror(errno)); goto error; } if (st.st_uid != stv.st_uid || st.st_gid != stv.st_gid) { rc = fchown(fdv, st.st_uid, st.st_gid); if (rc != 0) { rc = -errno; - fprintf(stderr, "cannot chown %s (%s)\n", name, - strerror(errno)); + fprintf(stderr, "%s: %s: cannot chown: %s\n", progname, + name, strerror(errno)); goto error; } } - /* get file data version */ - rc = llapi_get_data_version(fd, &dv1, LL_DV_RD_FLUSH); - if (rc != 0) { - fprintf(stderr, "cannot get dataversion on %s (%s)\n", - name, strerror(-rc)); - goto error; - } - - if (migration_flags & MIGRATION_BLOCKS) { - /* take group lock to limit concurent access - * this will be no more needed when exclusive access will - * be implemented (see LU-2919) */ - /* group lock is taken after data version read because it - * blocks data version call */ - if (ioctl(fd, LL_IOC_GROUP_LOCK, gid) == -1) { - rc = -errno; - fprintf(stderr, "cannot get group lock on %s (%s)\n", - name, strerror(-rc)); - goto error; + if (migration_flags & MIGRATION_BLOCKS || !file_lease_supported) { + /* Blocking mode, forced if servers do not support file lease */ + rc = migrate_block(fd, fdv, &st, buf_size, name); + } else { + rc = migrate_nonblock(fd, fdv, &st, buf_size, name); + if (rc == 0) { + have_lease_rdlck = false; + fdv = -1; /* The volatile file is closed as we put the + * lease in non-blocking mode. */ } - have_gl = 1; } - /* copy data */ - rpos = 0; - wpos = 0; - bufoff = 0; - rsize = -1; - do { - /* read new data only if we have written all - * previously read data */ - if (wpos == rpos) { - rsize = read(fd, buf, bufsz); - if (rsize < 0) { - rc = -errno; - fprintf(stderr, "read failed on %s" - " (%s)\n", name, - strerror(-rc)); - goto error; - } - rpos += rsize; - bufoff = 0; - } - /* eof ? */ - if (rsize == 0) +error: + if (have_lease_rdlck) + llapi_lease_put(fd); + + if (fd >= 0) + close(fd); + + if (fdv >= 0) + close(fdv); + +free: + if (lum) + free(lum); + + return rc; +} + +/** + * Parse a string containing an OST index list into an array of integers. + * + * The input string contains a comma delimited list of individual + * indices and ranges, for example "1,2-4,7". Add the indices into the + * \a osts array and remove duplicates. + * + * \param[out] osts array to store indices in + * \param[in] size size of \a osts array + * \param[in] offset starting index in \a osts + * \param[in] arg string containing OST index list + * + * \retval positive number of indices in \a osts + * \retval -EINVAL unable to parse \a arg + */ +static int parse_targets(__u32 *osts, int size, int offset, char *arg) +{ + int rc; + int nr = offset; + int slots = size - offset; + char *ptr = NULL; + bool end_of_loop; + + if (arg == NULL) + return -EINVAL; + + end_of_loop = false; + while (!end_of_loop) { + int start_index; + int end_index; + int i; + char *endptr = NULL; + + rc = -EINVAL; + + ptr = strchrnul(arg, ','); + + end_of_loop = *ptr == '\0'; + *ptr = '\0'; + + start_index = strtol(arg, &endptr, 0); + if (endptr == arg) /* no data at all */ break; - wsize = write(fdv, buf + bufoff, rpos - wpos); - if (wsize < 0) { - rc = -errno; - fprintf(stderr, "write failed on volatile" - " for %s (%s)\n", name, strerror(-rc)); - goto error; + if (*endptr != '-' && *endptr != '\0') /* has invalid data */ + break; + if (start_index < 0) + break; + + end_index = start_index; + if (*endptr == '-') { + end_index = strtol(endptr + 1, &endptr, 0); + if (*endptr != '\0') + break; + if (end_index < start_index) + break; } - wpos += wsize; - bufoff += wsize; - } while (1); - /* flush data */ - fsync(fdv); + for (i = start_index; i <= end_index && slots > 0; i++) { + int j; - if (migration_flags & MIGRATION_BLOCKS) { - /* give back group lock */ - if (ioctl(fd, LL_IOC_GROUP_UNLOCK, gid) == -1) { - rc = -errno; - fprintf(stderr, "cannot put group lock on %s (%s)\n", - name, strerror(-rc)); + /* remove duplicate */ + for (j = 0; j < offset; j++) { + if (osts[j] == i) + break; + } + if (j == offset) { /* no duplicate */ + osts[nr++] = i; + --slots; + } } - have_gl = 0; - } - - /* swap layouts - * for a migration we need to: - * - check data version on file did not change - * - keep file mtime - * - keep file atime - */ - rc = llapi_fswap_layouts(fd, fdv, dv1, 0, - SWAP_LAYOUTS_CHECK_DV1 | - SWAP_LAYOUTS_KEEP_MTIME | - SWAP_LAYOUTS_KEEP_ATIME); - if (rc == -EAGAIN) { - fprintf(stderr, "%s: dataversion changed during copy, " - "migration aborted\n", name); - goto error; - } - if (rc != 0) - fprintf(stderr, "%s: swap layout to new file failed: %s\n", - name, strerror(-rc)); + if (slots == 0 && i < end_index) + break; -error: - /* give back group lock */ - if ((migration_flags & MIGRATION_BLOCKS) && have_gl && - (ioctl(fd, LL_IOC_GROUP_UNLOCK, gid) == -1)) { - /* we keep in rc the original error */ - fprintf(stderr, "cannot put group lock on %s (%s)\n", - name, strerror(-errno)); + *ptr = ','; + arg = ++ptr; + offset = nr; + rc = 0; } + if (!end_of_loop && ptr != NULL) + *ptr = ','; - close(fdv); - close(fd); -free: - if (lum) - free(lum); - if (buf) - free(buf); - return rc; + return rc < 0 ? rc : nr; } /* functions */ static int lfs_setstripe(int argc, char **argv) { - char *fname; - int result; - unsigned long long st_size; - int st_offset, st_count; - char *end; - int c; - int delete = 0; - char *stripe_size_arg = NULL; - char *stripe_off_arg = NULL; - char *stripe_count_arg = NULL; - char *pool_name_arg = NULL; - unsigned long long size_units = 1; - int migrate_mode = 0; - __u64 migration_flags = 0; + struct llapi_stripe_param *param = NULL; + struct find_param migrate_mdt_param = { + .fp_max_depth = -1, + .fp_mdt_index = -1, + }; + char *fname; + int result; + int result2 = 0; + unsigned long long st_size; + int st_offset, st_count; + char *end; + int c; + int delete = 0; + char *stripe_size_arg = NULL; + char *stripe_off_arg = NULL; + char *stripe_count_arg = NULL; + char *pool_name_arg = NULL; + char *mdt_idx_arg = NULL; + unsigned long long size_units = 1; + bool migrate_mode = false; + __u64 migration_flags = 0; + __u32 osts[LOV_MAX_STRIPE_COUNT] = { 0 }; + int nr_osts = 0; struct option long_opts[] = { /* valid only in migrate mode */ @@ -617,12 +904,10 @@ static int lfs_setstripe(int argc, char **argv) #endif {"stripe-index", required_argument, 0, 'i'}, {"stripe_index", required_argument, 0, 'i'}, -#if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(2, 9, 53, 0) - /* This formerly implied "stripe-index", but was confusing - * with "file offset" (which will eventually be needed for - * with different layouts by offset), so deprecate it. */ - {"offset", required_argument, 0, 'o'}, -#endif + {"mdt-index", required_argument, 0, 'm'}, + {"mdt_index", required_argument, 0, 'm'}, + {"ost-list", required_argument, 0, 'o'}, + {"ost_list", required_argument, 0, 'o'}, {"pool", required_argument, 0, 'p'}, #if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(2, 9, 53, 0) /* This formerly implied "--stripe-size", but was confusing @@ -635,24 +920,23 @@ static int lfs_setstripe(int argc, char **argv) {0, 0, 0, 0} }; - st_size = 0; - st_offset = -1; - st_count = 0; + st_size = 0; + st_offset = -1; + st_count = 0; if (strcmp(argv[0], "migrate") == 0) - migrate_mode = 1; + migrate_mode = true; - optind = 0; - while ((c = getopt_long(argc, argv, "c:di:o:p:s:S:", + while ((c = getopt_long(argc, argv, "bc:di:m:o:p:s:S:", long_opts, NULL)) >= 0) { switch (c) { case 0: /* Long options. */ break; case 'b': - if (migrate_mode == 0) { + if (!migrate_mode) { fprintf(stderr, "--block is valid only for" - " migrate mode"); + " migrate mode\n"); return CMD_HELP; } migration_flags |= MIGRATION_BLOCKS; @@ -669,11 +953,20 @@ static int lfs_setstripe(int argc, char **argv) /* delete the default striping pattern */ delete = 1; break; -#if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(2, 9, 53, 0) case 'o': - fprintf(stderr, "warning: '--offset|-o' deprecated, " - "use '--stripe-index|-i' instead\n"); -#endif + nr_osts = parse_targets(osts, + sizeof(osts) / sizeof(__u32), + nr_osts, optarg); + if (nr_osts < 0) { + fprintf(stderr, + "error: %s: bad OST indices '%s'\n", + argv[0], optarg); + return CMD_HELP; + } + + if (st_offset == -1) /* first in the command line */ + st_offset = osts[0]; + break; case 'i': #if LUSTRE_VERSION_CODE >= OBD_OCD_VERSION(2, 6, 53, 0) if (strcmp(argv[optind - 1], "--index") == 0) @@ -682,6 +975,14 @@ static int lfs_setstripe(int argc, char **argv) #endif stripe_off_arg = optarg; break; + case 'm': + if (!migrate_mode) { + fprintf(stderr, "--mdt-index is valid only for" + " migrate mode\n"); + return CMD_HELP; + } + mdt_idx_arg = optarg; + break; #if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(2, 9, 53, 0) case 's': #if LUSTRE_VERSION_CODE >= OBD_OCD_VERSION(2, 6, 53, 0) @@ -717,6 +1018,19 @@ static int lfs_setstripe(int argc, char **argv) return CMD_HELP; } + if (mdt_idx_arg != NULL && optind > 3) { + fprintf(stderr, "error: %s: cannot specify -m with other " + "options\n", argv[0]); + return CMD_HELP; + } + + if (pool_name_arg && strlen(pool_name_arg) > LOV_MAXPOOLNAME) { + fprintf(stderr, + "error: %s: pool name '%s' is too long (max is %d characters)\n", + argv[0], pool_name_arg, LOV_MAXPOOLNAME); + return CMD_HELP; + } + /* get the stripe size */ if (stripe_size_arg != NULL) { result = llapi_parse_size(stripe_size_arg, &st_size, @@ -724,7 +1038,7 @@ static int lfs_setstripe(int argc, char **argv) if (result) { fprintf(stderr, "error: %s: bad stripe size '%s'\n", argv[0], stripe_size_arg); - return result; + return CMD_HELP; } } /* get the stripe offset */ @@ -746,26 +1060,73 @@ static int lfs_setstripe(int argc, char **argv) } } - do { - if (migrate_mode) - result = lfs_migrate(fname, st_size, st_offset, - st_count, 0, pool_name_arg, - migration_flags); - else - result = llapi_file_create_pool(fname, st_size, - st_offset, st_count, - 0, pool_name_arg); + if (mdt_idx_arg != NULL) { + /* initialize migrate mdt parameters */ + migrate_mdt_param.fp_mdt_index = strtoul(mdt_idx_arg, &end, 0); + if (*end != '\0') { + fprintf(stderr, "error: %s: bad MDT index '%s'\n", + argv[0], mdt_idx_arg); + return CMD_HELP; + } + migrate_mdt_param.fp_migrate = 1; + } else { + /* initialize stripe parameters */ + param = calloc(1, offsetof(typeof(*param), lsp_osts[nr_osts])); + if (param == NULL) { + fprintf(stderr, "error: %s: run out of memory\n", + argv[0]); + return CMD_HELP; + } + + param->lsp_stripe_size = st_size; + param->lsp_stripe_offset = st_offset; + param->lsp_stripe_count = st_count; + param->lsp_stripe_pattern = 0; + param->lsp_pool = pool_name_arg; + param->lsp_is_specific = false; + if (nr_osts > 0) { + if (st_count > 0 && nr_osts != st_count) { + fprintf(stderr, "error: %s: stripe count '%d' " + "doesn't match the number of OSTs: %d\n" + , argv[0], st_count, nr_osts); + free(param); + return CMD_HELP; + } + + param->lsp_is_specific = true; + param->lsp_stripe_count = nr_osts; + memcpy(param->lsp_osts, osts, sizeof(*osts) * nr_osts); + } + } + + for (fname = argv[optind]; fname != NULL; fname = argv[++optind]) { + if (!migrate_mode) { + result = llapi_file_open_param(fname, + O_CREAT | O_WRONLY, + 0644, param); + if (result >= 0) { + close(result); + result = 0; + } + } else if (mdt_idx_arg != NULL) { + result = llapi_migrate_mdt(fname, &migrate_mdt_param); + } else { + result = lfs_migrate(fname, migration_flags, param); + } if (result) { + /* Save the first error encountered. */ + if (result2 == 0) + result2 = result; fprintf(stderr, - "error: %s: %s stripe file '%s' failed\n", + "error: %s: %s file '%s' failed\n", argv[0], migrate_mode ? "migrate" : "create", fname); - break; + continue; } - fname = argv[++optind]; - } while (fname != NULL); + } - return result; + free(param); + return result2; } static int lfs_poollist(int argc, char **argv) @@ -880,11 +1241,12 @@ static int name2layout(__u32 *layout, char *name) #define FIND_POOL_OPT 3 static int lfs_find(int argc, char **argv) { - int c, ret; + int c, rc; + int ret = 0; time_t t; struct find_param param = { .fp_max_depth = -1, - .quiet = 1, + .fp_quiet = 1, }; struct option long_opts[] = { {"atime", required_argument, 0, 'A'}, @@ -925,7 +1287,6 @@ static int lfs_find(int argc, char **argv) time(&t); - optind = 0; /* when getopt_long_only() hits '!' it returns 1, puts "!" in optarg */ while ((c = getopt_long_only(argc, argv, "-A:c:C:D:g:G:i:L:m:M:n:O:Ppqrs:S:t:u:U:v", @@ -982,40 +1343,40 @@ static int lfs_find(int argc, char **argv) xsign = ¶m.fp_msign; param.fp_exclude_mtime = !!neg_opt; } - ret = set_time(&t, xtime, optarg); - if (ret == INT_MAX) { - ret = -1; - goto err; - } - if (ret) - *xsign = ret; - break; + rc = set_time(&t, xtime, optarg); + if (rc == INT_MAX) { + ret = -1; + goto err; + } + if (rc) + *xsign = rc; + break; case 'c': if (optarg[0] == '+') { - param.stripecount_sign = -1; + param.fp_stripe_count_sign = -1; optarg++; } else if (optarg[0] == '-') { - param.stripecount_sign = 1; + param.fp_stripe_count_sign = 1; optarg++; } - param.stripecount = strtoul(optarg, &endptr, 0); + param.fp_stripe_count = strtoul(optarg, &endptr, 0); if (*endptr != '\0') { fprintf(stderr,"error: bad stripe_count '%s'\n", optarg); ret = -1; goto err; } - param.check_stripecount = 1; - param.exclude_stripecount = !!neg_opt; + param.fp_check_stripe_count = 1; + param.fp_exclude_stripe_count = !!neg_opt; break; case 'D': param.fp_max_depth = strtol(optarg, 0, 0); break; case 'g': case 'G': - ret = name2id(¶m.fp_gid, optarg, GROUP); - if (ret) { + rc = name2id(¶m.fp_gid, optarg, GROUP); + if (rc) { param.fp_gid = strtoul(optarg, &endptr, 10); if (*endptr != '\0') { fprintf(stderr, "Group/GID: %s cannot " @@ -1028,16 +1389,16 @@ static int lfs_find(int argc, char **argv) param.fp_check_gid = 1; break; case 'L': - ret = name2layout(¶m.layout, optarg); + ret = name2layout(¶m.fp_layout, optarg); if (ret) goto err; - param.exclude_layout = !!neg_opt; - param.check_layout = 1; + param.fp_exclude_layout = !!neg_opt; + param.fp_check_layout = 1; break; case 'u': case 'U': - ret = name2id(¶m.fp_uid, optarg, USER); - if (ret) { + rc = name2id(¶m.fp_uid, optarg, USER); + if (rc) { param.fp_uid = strtoul(optarg, &endptr, 10); if (*endptr != '\0') { fprintf(stderr, "User/UID: %s cannot " @@ -1060,14 +1421,14 @@ static int lfs_find(int argc, char **argv) } /* we do check for empty pool because empty pool * is used to find V1 lov attributes */ - strncpy(param.poolname, optarg, LOV_MAXPOOLNAME); - param.poolname[LOV_MAXPOOLNAME] = '\0'; - param.exclude_pool = !!neg_opt; - param.check_pool = 1; + strncpy(param.fp_poolname, optarg, LOV_MAXPOOLNAME); + param.fp_poolname[LOV_MAXPOOLNAME] = '\0'; + param.fp_exclude_pool = !!neg_opt; + param.fp_check_pool = 1; break; case 'n': - param.pattern = (char *)optarg; - param.exclude_pattern = !!neg_opt; + param.fp_pattern = (char *)optarg; + param.fp_exclude_pattern = !!neg_opt; break; case 'm': case 'i': @@ -1082,7 +1443,7 @@ static int lfs_find(int argc, char **argv) goto err; } - param.exclude_obd = !!neg_opt; + param.fp_exclude_obd = !!neg_opt; token = buf; while (token && *token) { @@ -1093,32 +1454,38 @@ static int lfs_find(int argc, char **argv) } } if (c == 'm') { - param.exclude_mdt = !!neg_opt; - param.num_alloc_mdts += len; - tmp = realloc(param.mdtuuid, - param.num_alloc_mdts * - sizeof(*param.mdtuuid)); - if (tmp == NULL) - GOTO(err_free, ret = -ENOMEM); - param.mdtuuid = tmp; + param.fp_exclude_mdt = !!neg_opt; + param.fp_num_alloc_mdts += len; + tmp = realloc(param.fp_mdt_uuid, + param.fp_num_alloc_mdts * + sizeof(*param.fp_mdt_uuid)); + if (tmp == NULL) { + ret = -ENOMEM; + goto err_free; + } + + param.fp_mdt_uuid = tmp; } else { - param.exclude_obd = !!neg_opt; - param.num_alloc_obds += len; - tmp = realloc(param.obduuid, - param.num_alloc_obds * - sizeof(*param.obduuid)); - if (tmp == NULL) - GOTO(err_free, ret = -ENOMEM); - param.obduuid = tmp; + param.fp_exclude_obd = !!neg_opt; + param.fp_num_alloc_obds += len; + tmp = realloc(param.fp_obd_uuid, + param.fp_num_alloc_obds * + sizeof(*param.fp_obd_uuid)); + if (tmp == NULL) { + ret = -ENOMEM; + goto err_free; + } + + param.fp_obd_uuid = tmp; } for (token = buf; token && *token; token = next) { struct obd_uuid *puuid; if (c == 'm') { puuid = - ¶m.mdtuuid[param.num_mdts++]; + ¶m.fp_mdt_uuid[param.fp_num_mdts++]; } else { puuid = - ¶m.obduuid[param.num_obds++]; + ¶m.fp_obd_uuid[param.fp_num_obds++]; } p = strchr(token, ','); next = 0; @@ -1126,8 +1493,12 @@ static int lfs_find(int argc, char **argv) *p = 0; next = p+1; } - if (strlen(token) > sizeof(puuid->uuid)-1) - GOTO(err_free, ret = -E2BIG); + + if (strlen(token) > sizeof(puuid->uuid) - 1) { + ret = -E2BIG; + goto err_free; + } + strncpy(puuid->uuid, token, sizeof(puuid->uuid)); } @@ -1137,47 +1508,47 @@ err_free: break; } case 'p': - param.zeroend = 1; + param.fp_zero_end = 1; break; case 'P': break; case 's': if (optarg[0] == '+') { - param.size_sign = -1; + param.fp_size_sign = -1; optarg++; } else if (optarg[0] == '-') { - param.size_sign = 1; + param.fp_size_sign = 1; optarg++; } - ret = llapi_parse_size(optarg, ¶m.size, - ¶m.size_units, 0); + ret = llapi_parse_size(optarg, ¶m.fp_size, + ¶m.fp_size_units, 0); if (ret) { fprintf(stderr, "error: bad file size '%s'\n", optarg); goto err; } - param.check_size = 1; - param.exclude_size = !!neg_opt; + param.fp_check_size = 1; + param.fp_exclude_size = !!neg_opt; break; case 'S': if (optarg[0] == '+') { - param.stripesize_sign = -1; + param.fp_stripe_size_sign = -1; optarg++; } else if (optarg[0] == '-') { - param.stripesize_sign = 1; + param.fp_stripe_size_sign = 1; optarg++; } - ret = llapi_parse_size(optarg, ¶m.stripesize, - ¶m.stripesize_units, 0); + ret = llapi_parse_size(optarg, ¶m.fp_stripe_size, + ¶m.fp_stripe_size_units, 0); if (ret) { fprintf(stderr, "error: bad stripe_size '%s'\n", optarg); goto err; } - param.check_stripesize = 1; - param.exclude_stripesize = !!neg_opt; + param.fp_check_stripe_size = 1; + param.fp_exclude_stripe_size = !!neg_opt; break; case 't': param.fp_exclude_type = !!neg_opt; @@ -1226,19 +1597,21 @@ err_free: pathend = argc; } - do { - ret = llapi_find(argv[pathstart], ¶m); - } while (++pathstart < pathend && !ret); + do { + rc = llapi_find(argv[pathstart], ¶m); + if (rc != 0 && ret == 0) + ret = rc; + } while (++pathstart < pathend); if (ret) fprintf(stderr, "error: %s failed for %s.\n", argv[0], argv[optind - 1]); err: - if (param.obduuid && param.num_alloc_obds) - free(param.obduuid); + if (param.fp_obd_uuid && param.fp_num_alloc_obds) + free(param.fp_obd_uuid); - if (param.mdtuuid && param.num_alloc_mdts) - free(param.mdtuuid); + if (param.fp_mdt_uuid && param.fp_num_alloc_mdts) + free(param.fp_mdt_uuid); return ret; } @@ -1295,33 +1668,32 @@ static int lfs_getstripe_internal(int argc, char **argv, int c, rc; param->fp_max_depth = 1; - optind = 0; while ((c = getopt_long(argc, argv, "cdDghiLMoO:pqrRsSv", long_opts, NULL)) != -1) { switch (c) { case 'O': - if (param->obduuid) { + if (param->fp_obd_uuid) { fprintf(stderr, "error: %s: only one obduuid allowed", argv[0]); return CMD_HELP; } - param->obduuid = (struct obd_uuid *)optarg; + param->fp_obd_uuid = (struct obd_uuid *)optarg; break; case 'q': - param->quiet++; + param->fp_quiet++; break; case 'd': param->fp_max_depth = 0; break; case 'D': - param->get_default_lmv = 1; + param->fp_get_default_lmv = 1; break; case 'r': - param->recursive = 1; + param->fp_recursive = 1; break; case 'v': - param->verbose = VERBOSE_ALL | VERBOSE_DETAIL; + param->fp_verbose = VERBOSE_ALL | VERBOSE_DETAIL; break; case 'c': #if LUSTRE_VERSION_CODE >= OBD_OCD_VERSION(2, 6, 53, 0) @@ -1329,8 +1701,8 @@ static int lfs_getstripe_internal(int argc, char **argv, fprintf(stderr, "warning: '--count' deprecated," " use '--stripe-count' instead\n"); #endif - if (!(param->verbose & VERBOSE_DETAIL)) { - param->verbose |= VERBOSE_COUNT; + if (!(param->fp_verbose & VERBOSE_DETAIL)) { + param->fp_verbose |= VERBOSE_COUNT; param->fp_max_depth = 0; } break; @@ -1342,8 +1714,8 @@ static int lfs_getstripe_internal(int argc, char **argv, #endif #endif /* LUSTRE_VERSION_CODE < OBD_OCD_VERSION(2, 9, 53, 0) */ case 'S': - if (!(param->verbose & VERBOSE_DETAIL)) { - param->verbose |= VERBOSE_SIZE; + if (!(param->fp_verbose & VERBOSE_DETAIL)) { + param->fp_verbose |= VERBOSE_SIZE; param->fp_max_depth = 0; } break; @@ -1358,36 +1730,36 @@ static int lfs_getstripe_internal(int argc, char **argv, fprintf(stderr, "warning: '--index' deprecated" ", use '--stripe-index' instead\n"); #endif - if (!(param->verbose & VERBOSE_DETAIL)) { - param->verbose |= VERBOSE_OFFSET; + if (!(param->fp_verbose & VERBOSE_DETAIL)) { + param->fp_verbose |= VERBOSE_OFFSET; param->fp_max_depth = 0; } break; case 'p': - if (!(param->verbose & VERBOSE_DETAIL)) { - param->verbose |= VERBOSE_POOL; + if (!(param->fp_verbose & VERBOSE_DETAIL)) { + param->fp_verbose |= VERBOSE_POOL; param->fp_max_depth = 0; } break; case 'g': - if (!(param->verbose & VERBOSE_DETAIL)) { - param->verbose |= VERBOSE_GENERATION; + if (!(param->fp_verbose & VERBOSE_DETAIL)) { + param->fp_verbose |= VERBOSE_GENERATION; param->fp_max_depth = 0; } break; case 'L': - if (!(param->verbose & VERBOSE_DETAIL)) { - param->verbose |= VERBOSE_LAYOUT; + if (!(param->fp_verbose & VERBOSE_DETAIL)) { + param->fp_verbose |= VERBOSE_LAYOUT; param->fp_max_depth = 0; } break; case 'M': - if (!(param->verbose & VERBOSE_DETAIL)) + if (!(param->fp_verbose & VERBOSE_DETAIL)) param->fp_max_depth = 0; - param->verbose |= VERBOSE_MDTINDEX; + param->fp_verbose |= VERBOSE_MDTINDEX; break; case 'R': - param->raw = 1; + param->fp_raw = 1; break; default: return CMD_HELP; @@ -1397,13 +1769,13 @@ static int lfs_getstripe_internal(int argc, char **argv, if (optind >= argc) return CMD_HELP; - if (param->recursive) + if (param->fp_recursive) param->fp_max_depth = -1; - if (!param->verbose) - param->verbose = VERBOSE_ALL; - if (param->quiet) - param->verbose = VERBOSE_OBJID; + if (!param->fp_verbose) + param->fp_verbose = VERBOSE_ALL; + if (param->fp_quiet) + param->fp_verbose = VERBOSE_OBJID; do { rc = llapi_getstripe(argv[optind], param); @@ -1438,7 +1810,7 @@ static int lfs_tgts(int argc, char **argv) memset(¶m, 0, sizeof(param)); if (!strcmp(argv[0], "mdts")) - param.get_lmv = 1; + param.fp_get_lmv = 1; rc = llapi_ostlist(mntdir, ¶m); if (rc) { @@ -1464,7 +1836,7 @@ static int lfs_getdirstripe(int argc, char **argv) { struct find_param param = { 0 }; - param.get_lmv = 1; + param.fp_get_lmv = 1; return lfs_getstripe_internal(argc, argv, ¶m); } @@ -1481,19 +1853,24 @@ static int lfs_setdirstripe(int argc, char **argv) char *stripe_offset_opt = NULL; char *stripe_count_opt = NULL; char *stripe_hash_opt = NULL; - int default_stripe = 0; + char *mode_opt = NULL; + bool default_stripe = false; + mode_t mode = S_IRWXU | S_IRWXG | S_IRWXO; + mode_t previous_mode = 0; + bool delete = false; struct option long_opts[] = { {"count", required_argument, 0, 'c'}, + {"delete", no_argument, 0, 'd'}, {"index", required_argument, 0, 'i'}, + {"mode", required_argument, 0, 'm'}, {"hash-type", required_argument, 0, 't'}, - {"default_stripe", required_argument, 0, 'D'}, + {"default_stripe", no_argument, 0, 'D'}, {0, 0, 0, 0} }; - optind = 0; - - while ((c = getopt_long(argc, argv, "c:Di:t:", long_opts, NULL)) >= 0) { + while ((c = getopt_long(argc, argv, "c:dDi:m:t:", long_opts, + NULL)) >= 0) { switch (c) { case 0: /* Long options. */ @@ -1501,12 +1878,19 @@ static int lfs_setdirstripe(int argc, char **argv) case 'c': stripe_count_opt = optarg; break; + case 'd': + delete = true; + default_stripe = true; + break; case 'D': - default_stripe = 1; + default_stripe = true; break; case 'i': stripe_offset_opt = optarg; break; + case 'm': + mode_opt = optarg; + break; case 't': stripe_hash_opt = optarg; break; @@ -1524,7 +1908,7 @@ static int lfs_setdirstripe(int argc, char **argv) return CMD_HELP; } - if (stripe_offset_opt == NULL && stripe_count_opt == NULL) { + if (!delete && stripe_offset_opt == NULL && stripe_count_opt == NULL) { fprintf(stderr, "error: %s: missing stripe offset and count.\n", argv[0]); return CMD_HELP; @@ -1540,6 +1924,27 @@ static int lfs_setdirstripe(int argc, char **argv) } } + if (delete) { + if (stripe_offset_opt != NULL || stripe_count_opt != NULL) { + fprintf(stderr, "error: %s: cannot specify -d with -s," + " or -i options.\n", argv[0]); + return CMD_HELP; + } else { + stripe_count = 0; + } + } + + + if (mode_opt != NULL) { + mode = strtoul(mode_opt, &end, 8); + if (*end != '\0') { + fprintf(stderr, "error: %s: bad mode '%s'\n", + argv[0], mode_opt); + return CMD_HELP; + } + previous_mode = umask(0); + } + if (stripe_hash_opt == NULL || strcmp(stripe_hash_opt, LMV_HASH_NAME_FNV_1A_64) == 0) { hash_type = LMV_HASH_TYPE_FNV_1A_64; @@ -1563,12 +1968,13 @@ static int lfs_setdirstripe(int argc, char **argv) dname = argv[optind]; do { - if (default_stripe == 1) { + if (default_stripe) { result = llapi_dir_set_default_lmv_stripe(dname, stripe_offset, stripe_count, hash_type, NULL); } else { - result = llapi_dir_create_pool(dname, 0, stripe_offset, + result = llapi_dir_create_pool(dname, mode, + stripe_offset, stripe_count, hash_type, NULL); } @@ -1581,6 +1987,9 @@ static int lfs_setdirstripe(int argc, char **argv) dname = argv[++optind]; } while (dname != NULL); + if (mode_opt != NULL) + umask(previous_mode); + return result; } @@ -1615,13 +2024,13 @@ static int lfs_mv(int argc, char **argv) { struct find_param param = { .fp_max_depth = -1, - .mdtindex = -1, + .fp_mdt_index = -1, }; char *end; int c; int rc = 0; struct option long_opts[] = { - {"--mdt-index", required_argument, 0, 'M'}, + {"mdt-index", required_argument, 0, 'M'}, {"verbose", no_argument, 0, 'v'}, {0, 0, 0, 0} }; @@ -1629,7 +2038,7 @@ static int lfs_mv(int argc, char **argv) while ((c = getopt_long(argc, argv, "M:v", long_opts, NULL)) != -1) { switch (c) { case 'M': { - param.mdtindex = strtoul(optarg, &end, 0); + param.fp_mdt_index = strtoul(optarg, &end, 0); if (*end != '\0') { fprintf(stderr, "%s: invalid MDT index'%s'\n", argv[0], optarg); @@ -1638,7 +2047,7 @@ static int lfs_mv(int argc, char **argv) break; } case 'v': { - param.verbose = VERBOSE_DETAIL; + param.fp_verbose = VERBOSE_DETAIL; break; } default: @@ -1648,21 +2057,22 @@ static int lfs_mv(int argc, char **argv) } } - if (param.mdtindex == -1) { - fprintf(stderr, "%s MDT index must be indicated\n", argv[0]); + if (param.fp_mdt_index == -1) { + fprintf(stderr, "%s: MDT index must be specified\n", argv[0]); return CMD_HELP; } if (optind >= argc) { - fprintf(stderr, "%s missing operand path\n", argv[0]); + fprintf(stderr, "%s: missing operand path\n", argv[0]); return CMD_HELP; } - param.migrate = 1; - rc = llapi_mv(argv[optind], ¶m); + param.fp_migrate = 1; + rc = llapi_migrate_mdt(argv[optind], ¶m); if (rc != 0) - fprintf(stderr, "cannot migrate '%s' to MDT%04x: %s\n", - argv[optind], param.mdtindex, strerror(-rc)); + fprintf(stderr, "%s: cannot migrate '%s' to MDT%04x: %s\n", + argv[0], argv[optind], param.fp_mdt_index, + strerror(-rc)); return rc; } @@ -1700,7 +2110,10 @@ static int showdf(char *mntdir, struct obd_statfs *stat, double ratio = 0; char *suffix = "KMGTPEZY"; /* Note if we have >2^64 bytes/fs these buffers will need to be grown */ - char tbuf[20], ubuf[20], abuf[20], rbuf[20]; + char tbuf[3 * sizeof(__u64)]; + char ubuf[3 * sizeof(__u64)]; + char abuf[3 * sizeof(__u64)]; + char rbuf[3 * sizeof(__u64)]; if (!uuid || !stat) return -EINVAL; @@ -1824,6 +2237,9 @@ static int mntdf(char *mntdir, char *fsname, char *pool, int ishow, if (rc == -ENODEV) break; + if (rc == -EAGAIN) + continue; + if (poolname && tp->st_op == LL_STATFS_LOV && llapi_search_ost(fsname, poolname, obd_uuid2str(&uuid_buf)) != 1) @@ -1886,7 +2302,6 @@ static int lfs_df(int argc, char **argv) {0, 0, 0, 0} }; - optind = 0; while ((c = getopt_long(argc, argv, "hilp:", long_opts, NULL)) != -1) { switch (c) { case 'i': @@ -1933,7 +2348,6 @@ static int lfs_getname(int argc, char **argv) int rc = 0, index = 0, c; char buf[sizeof(struct obd_uuid)]; - optind = 0; while ((c = getopt(argc, argv, "h")) != -1) return CMD_HELP; @@ -2002,9 +2416,7 @@ static int lfs_check(int argc, char **argv) return rc; } - rc = llapi_target_iterate(num_types, obd_types, - mntdir, llapi_ping_target); - + rc = llapi_target_check(num_types, obd_types, mntdir); if (rc) fprintf(stderr, "error: %s: %s status failed\n", argv[0],argv[1]); @@ -2021,209 +2433,6 @@ static int lfs_join(int argc, char **argv) } #ifdef HAVE_SYS_QUOTA_H -static int lfs_quotacheck(int argc, char **argv) -{ - int c, check_type = 0; - char *mnt; - struct if_quotacheck qchk; - struct if_quotactl qctl; - char *obd_type = (char *)qchk.obd_type; - int rc; - - memset(&qchk, 0, sizeof(qchk)); - - optind = 0; - while ((c = getopt(argc, argv, "gu")) != -1) { - switch (c) { - case 'u': - check_type |= 0x01; - break; - case 'g': - check_type |= 0x02; - break; - default: - fprintf(stderr, "error: %s: option '-%c' " - "unrecognized\n", argv[0], c); - return CMD_HELP; - } - } - - if (check_type) - check_type--; - else /* do quotacheck for both user & group quota by default */ - check_type = 0x02; - - if (argc == optind) - return CMD_HELP; - - mnt = argv[optind]; - - rc = llapi_quotacheck(mnt, check_type); - if (rc == -EOPNOTSUPP) { - fprintf(stderr, "error: quotacheck not supported by the quota " - "master.\nPlease note that quotacheck is deprecated as " - "of lustre 2.4.0 since space accounting is always " - "enabled.\nFilesystems not formatted with 2.4 utils or " - "beyond can be upgraded with tunefs.lustre --quota.\n"); - return rc; - } else if (rc) { - fprintf(stderr, "quotacheck failed: %s\n", strerror(-rc)); - return rc; - } - - rc = llapi_poll_quotacheck(mnt, &qchk); - if (rc) { - if (*obd_type) - fprintf(stderr, "%s %s ", obd_type, - obd_uuid2str(&qchk.obd_uuid)); - fprintf(stderr, "quota check failed: %s\n", strerror(-rc)); - return rc; - } - - memset(&qctl, 0, sizeof(qctl)); - qctl.qc_cmd = LUSTRE_Q_QUOTAON; - qctl.qc_type = check_type; - rc = llapi_quotactl(mnt, &qctl); - if (rc && rc != -EALREADY) { - if (*obd_type) - fprintf(stderr, "%s %s ", (char *)qctl.obd_type, - obd_uuid2str(&qctl.obd_uuid)); - fprintf(stderr, "%s turn on quota failed: %s\n", - argv[0], strerror(-rc)); - return rc; - } - - return 0; -} - -static int lfs_quotaon(int argc, char **argv) -{ - int c; - char *mnt; - struct if_quotactl qctl; - char *obd_type = (char *)qctl.obd_type; - int rc; - - memset(&qctl, 0, sizeof(qctl)); - qctl.qc_cmd = LUSTRE_Q_QUOTAON; - - optind = 0; - while ((c = getopt(argc, argv, "fgu")) != -1) { - switch (c) { - case 'u': - qctl.qc_type |= 0x01; - break; - case 'g': - qctl.qc_type |= 0x02; - break; - case 'f': - qctl.qc_cmd = LUSTRE_Q_QUOTAOFF; - break; - default: - fprintf(stderr, "error: %s: option '-%c' " - "unrecognized\n", argv[0], c); - return CMD_HELP; - } - } - - if (qctl.qc_type) - qctl.qc_type--; - else /* by default, enable quota for both user & group */ - qctl.qc_type = 0x02; - - if (argc == optind) - return CMD_HELP; - - mnt = argv[optind]; - - rc = llapi_quotactl(mnt, &qctl); - if (rc) { - if (rc == -EOPNOTSUPP) { - fprintf(stderr, "error: quotaon not supported by the " - "quota master.\nPlease note that quotaon/off is" - " deprecated as of lustre 2.4.0.\nQuota " - "enforcement should now be enabled on the MGS " - "via:\nmgs# lctl conf_param ${FSNAME}.quota." - "=\n(ost for block quota, mdt " - "for inode quota, u for user and g for group" - "\n"); - } else if (rc == -EALREADY) { - rc = 0; - } else if (rc == -ENOENT) { - fprintf(stderr, "error: cannot find quota database, " - "make sure you have run quotacheck\n"); - } else { - if (*obd_type) - fprintf(stderr, "%s %s ", obd_type, - obd_uuid2str(&qctl.obd_uuid)); - fprintf(stderr, "%s failed: %s\n", argv[0], - strerror(-rc)); - } - } - - return rc; -} - -static int lfs_quotaoff(int argc, char **argv) -{ - int c; - char *mnt; - struct if_quotactl qctl; - char *obd_type = (char *)qctl.obd_type; - int rc; - - memset(&qctl, 0, sizeof(qctl)); - qctl.qc_cmd = LUSTRE_Q_QUOTAOFF; - - optind = 0; - while ((c = getopt(argc, argv, "gu")) != -1) { - switch (c) { - case 'u': - qctl.qc_type |= 0x01; - break; - case 'g': - qctl.qc_type |= 0x02; - break; - default: - fprintf(stderr, "error: %s: option '-%c' " - "unrecognized\n", argv[0], c); - return CMD_HELP; - } - } - - if (qctl.qc_type) - qctl.qc_type--; - else /* by default, disable quota for both user & group */ - qctl.qc_type = 0x02; - - if (argc == optind) - return CMD_HELP; - - mnt = argv[optind]; - - rc = llapi_quotactl(mnt, &qctl); - if (rc) { - if (rc == -EOPNOTSUPP) { - fprintf(stderr, "error: quotaoff not supported by the " - "quota master.\nPlease note that quotaon/off is" - " deprecated as of lustre 2.4.0.\nQuota " - "enforcement can be disabled on the MGS via:\n" - "mgs# lctl conf_param ${FSNAME}.quota." - "=\"\"\n"); - } else if (rc == -EALREADY) { - rc = 0; - } else { - if (*obd_type) - fprintf(stderr, "%s %s ", obd_type, - obd_uuid2str(&qctl.obd_uuid)); - fprintf(stderr, "quotaoff failed: %s\n", - strerror(-rc)); - } - } - - return rc; -} - #define ARG2INT(nr, str, msg) \ do { \ char *endp; \ @@ -2337,7 +2546,6 @@ int lfs_setquota_times(int argc, char **argv) qctl.qc_cmd = LUSTRE_Q_SETINFO; qctl.qc_type = UGQUOTA; - optind = 0; while ((c = getopt_long(argc, argv, "b:gi:tu", long_opts, NULL)) != -1) { switch (c) { case 'u': @@ -2427,7 +2635,6 @@ int lfs_setquota(int argc, char **argv) * so it can be used as a marker that qc_type * isn't reinitialized from command line */ - optind = 0; while ((c = getopt_long(argc, argv, "b:B:g:i:I:u:", long_opts, NULL)) != -1) { switch (c) { case 'u': @@ -2814,7 +3021,6 @@ static int lfs_quota(int argc, char **argv) __u64 total_ialloc = 0, total_balloc = 0; bool human_readable = false; - optind = 0; while ((c = getopt(argc, argv, "gi:I:o:qtuvh")) != -1) { switch (c) { case 'u': @@ -2836,7 +3042,7 @@ static int lfs_quota(int argc, char **argv) break; case 'o': valid = qctl.qc_valid = QC_UUID; - strncpy(obd_uuid, optarg, sizeof(qctl.obd_uuid)); + strlcpy(obd_uuid, optarg, sizeof(qctl.obd_uuid)); break; case 'i': valid = qctl.qc_valid = QC_MDTIDX; @@ -2988,12 +3194,11 @@ static int flushctx_ioctl(char *mp) static int lfs_flushctx(int argc, char **argv) { - int kdestroy = 0, c; - FILE *proc = NULL; - char procline[PATH_MAX], *line; - int rc = 0; + int kdestroy = 0, c; + char mntdir[PATH_MAX] = {'\0'}; + int index = 0; + int rc = 0; - optind = 0; while ((c = getopt(argc, argv, "k")) != -1) { switch (c) { case 'k': @@ -3007,46 +3212,24 @@ static int lfs_flushctx(int argc, char **argv) } if (kdestroy) { - int rc; if ((rc = system("kdestroy > /dev/null")) != 0) { rc = WEXITSTATUS(rc); fprintf(stderr, "error destroying tickets: %d, continuing\n", rc); } } - if (optind >= argc) { - /* flush for all mounted lustre fs. */ - proc = fopen("/proc/mounts", "r"); - if (!proc) { - fprintf(stderr, "error: %s: can't open /proc/mounts\n", - argv[0]); - return -1; - } - - while ((line = fgets(procline, PATH_MAX, proc)) != NULL) { - char dev[PATH_MAX]; - char mp[PATH_MAX]; - char fs[PATH_MAX]; + if (optind >= argc) { + /* flush for all mounted lustre fs. */ + while (!llapi_search_mounts(NULL, index++, mntdir, NULL)) { + /* Check if we have a mount point */ + if (mntdir[0] == '\0') + continue; - if (sscanf(line, "%s %s %s", dev, mp, fs) != 3) { - fprintf(stderr, "%s: unexpected format in " - "/proc/mounts\n", - argv[0]); + if (flushctx_ioctl(mntdir)) rc = -1; - goto out; - } - if (strcmp(fs, "lustre") != 0) - continue; - /* we use '@' to determine it's a client. are there - * any other better way? - */ - if (strchr(dev, '@') == NULL) - continue; - - if (flushctx_ioctl(mp)) - rc = -1; - } + mntdir[0] = '\0'; /* avoid matching in next loop */ + } } else { /* flush fs as specified */ while (optind < argc) { @@ -3054,10 +3237,6 @@ static int lfs_flushctx(int argc, char **argv) rc = -1; } } - -out: - if (proc != NULL) - fclose(proc); return rc; } @@ -3098,7 +3277,7 @@ static int lfs_ls(int argc, char **argv) static int lfs_changelog(int argc, char **argv) { void *changelog_priv; - struct changelog_ext_rec *rec; + struct changelog_rec *rec; long long startrec = 0, endrec = 0; char *mdd; struct option long_opts[] = { @@ -3108,7 +3287,6 @@ static int lfs_changelog(int argc, char **argv) char short_opts[] = "f"; int rc, follow = 0; - optind = 0; while ((rc = getopt_long(argc, argv, short_opts, long_opts, NULL)) != -1) { switch (rc) { @@ -3132,15 +3310,16 @@ static int lfs_changelog(int argc, char **argv) if (argc > optind) endrec = strtoll(argv[optind++], NULL, 10); - rc = llapi_changelog_start(&changelog_priv, - CHANGELOG_FLAG_BLOCK | - (follow ? CHANGELOG_FLAG_FOLLOW : 0), - mdd, startrec); - if (rc < 0) { - fprintf(stderr, "Can't start changelog: %s\n", - strerror(errno = -rc)); - return rc; - } + rc = llapi_changelog_start(&changelog_priv, + CHANGELOG_FLAG_BLOCK | + CHANGELOG_FLAG_JOBID | + (follow ? CHANGELOG_FLAG_FOLLOW : 0), + mdd, startrec); + if (rc < 0) { + fprintf(stderr, "Can't start changelog: %s\n", + strerror(errno = -rc)); + return rc; + } while ((rc = llapi_changelog_recv(changelog_priv, &rec)) == 0) { time_t secs; @@ -3155,28 +3334,42 @@ static int lfs_changelog(int argc, char **argv) continue; } - secs = rec->cr_time >> 30; - gmtime_r(&secs, &ts); - printf(LPU64" %02d%-5s %02d:%02d:%02d.%06d %04d.%02d.%02d " - "0x%x t="DFID, rec->cr_index, rec->cr_type, - changelog_type2str(rec->cr_type), - ts.tm_hour, ts.tm_min, ts.tm_sec, - (int)(rec->cr_time & ((1<<30) - 1)), - ts.tm_year + 1900, ts.tm_mon + 1, ts.tm_mday, - rec->cr_flags & CLF_FLAGMASK, PFID(&rec->cr_tfid)); - if (rec->cr_namelen) - /* namespace rec includes parent and filename */ - printf(" p="DFID" %.*s", PFID(&rec->cr_pfid), - rec->cr_namelen, rec->cr_name); + secs = rec->cr_time >> 30; + gmtime_r(&secs, &ts); + printf(LPU64" %02d%-5s %02d:%02d:%02d.%06d %04d.%02d.%02d " + "0x%x t="DFID, rec->cr_index, rec->cr_type, + changelog_type2str(rec->cr_type), + ts.tm_hour, ts.tm_min, ts.tm_sec, + (int)(rec->cr_time & ((1<<30) - 1)), + ts.tm_year + 1900, ts.tm_mon + 1, ts.tm_mday, + rec->cr_flags & CLF_FLAGMASK, PFID(&rec->cr_tfid)); + + if (rec->cr_flags & CLF_JOBID) { + struct changelog_ext_jobid *jid = + changelog_rec_jobid(rec); + + if (jid->cr_jobid[0] != '\0') + printf(" j=%s", jid->cr_jobid); + } - if (!fid_is_zero(&rec->cr_sfid)) - printf(" s="DFID" sp="DFID" %.*s", - PFID(&rec->cr_sfid), PFID(&rec->cr_spfid), - changelog_rec_snamelen(rec), - changelog_rec_sname(rec)); + if (rec->cr_namelen) + printf(" p="DFID" %.*s", PFID(&rec->cr_pfid), + rec->cr_namelen, changelog_rec_name(rec)); + + if (rec->cr_flags & CLF_RENAME) { + struct changelog_ext_rename *rnm = + changelog_rec_rename(rec); + + if (!fid_is_zero(&rnm->cr_sfid)) + printf(" s="DFID" sp="DFID" %.*s", + PFID(&rnm->cr_sfid), + PFID(&rnm->cr_spfid), + (int)changelog_rec_snamelen(rec), + changelog_rec_sname(rec)); + } printf("\n"); - llapi_changelog_free(&rec); + llapi_changelog_free(&rec); } llapi_changelog_fini(&changelog_priv); @@ -3220,8 +3413,6 @@ static int lfs_fid2path(int argc, char **argv) int printcur = 0; int rc = 0; - optind = 0; - while ((rc = getopt_long(argc, argv, short_opts, long_opts, NULL)) != -1) { switch (rc) { @@ -3248,6 +3439,10 @@ static int lfs_fid2path(int argc, char **argv) device = argv[optind++]; path = calloc(1, PATH_MAX); + if (path == NULL) { + fprintf(stderr, "error: Not enough memory\n"); + return -errno; + } rc = 0; while (optind < argc) { @@ -3294,33 +3489,74 @@ static int lfs_fid2path(int argc, char **argv) static int lfs_path2fid(int argc, char **argv) { - char **path; - const char *sep = ""; - lustre_fid fid; - int rc = 0; + struct option long_opts[] = { + {"parents", no_argument, 0, 'p'}, + {0, 0, 0, 0} + }; + char **path; + const char short_opts[] = "p"; + const char *sep = ""; + lustre_fid fid; + int rc = 0; + bool show_parents = false; + + while ((rc = getopt_long(argc, argv, short_opts, + long_opts, NULL)) != -1) { + switch (rc) { + case 'p': + show_parents = true; + break; + default: + fprintf(stderr, "error: %s: option '%s' unrecognized\n", + argv[0], argv[optind - 1]); + return CMD_HELP; + } + } - if (argc < 2) + if (optind > argc - 1) return CMD_HELP; - else if (argc > 2) + else if (optind < argc - 1) sep = ": "; - path = argv + 1; - while (*path != NULL) { - int err = llapi_path2fid(*path, &fid); + rc = 0; + for (path = argv + optind; *path != NULL; path++) { + int err = 0; + if (!show_parents) { + err = llapi_path2fid(*path, &fid); + if (!err) + printf("%s%s"DFID"\n", + *sep != '\0' ? *path : "", sep, + PFID(&fid)); + } else { + char name[NAME_MAX + 1]; + unsigned int linkno = 0; + + while ((err = llapi_path2parent(*path, linkno, &fid, + name, sizeof(name))) == 0) { + if (*sep != '\0' && linkno == 0) + printf("%s%s", *path, sep); + + printf("%s"DFID"/%s", linkno != 0 ? "\t" : "", + PFID(&fid), name); + linkno++; + } + + /* err == -ENODATA is end-of-loop */ + if (linkno > 0 && err == -ENODATA) { + printf("\n"); + err = 0; + } + } if (err) { - fprintf(stderr, "%s: can't get fid for %s: %s\n", - argv[0], *path, strerror(-err)); + fprintf(stderr, "%s: can't get %sfid for %s: %s\n", + argv[0], show_parents ? "parent " : "", *path, + strerror(-err)); if (rc == 0) { rc = err; errno = -err; } - goto out; } - printf("%s%s"DFID"\n", *sep != '\0' ? *path : "", sep, - PFID(&fid)); -out: - path++; } return rc; @@ -3338,7 +3574,6 @@ static int lfs_data_version(int argc, char **argv) if (argc < 2) return CMD_HELP; - optind = 0; while ((c = getopt(argc, argv, "nrw")) != -1) { switch (c) { case 'n': @@ -3448,7 +3683,6 @@ static int lfs_hsm_change_flags(int argc, char **argv, int mode) if (argc < 3) return CMD_HELP; - optind = 0; while ((c = getopt_long(argc, argv, short_opts, long_opts, NULL)) != -1) { switch (c) { @@ -3542,12 +3776,15 @@ static int lfs_hsm_action(int argc, char **argv) if ((hps == HPS_RUNNING) && (hua == HUA_ARCHIVE || hua == HUA_RESTORE)) - printf("("LPX64 " bytes moved)\n", he.length); + printf("(%llu bytes moved)\n", + (unsigned long long)he.length); else if ((he.offset + he.length) == LUSTRE_EOF) - printf("(from "LPX64 " to EOF)\n", he.offset); + printf("(from %llu to EOF)\n", + (unsigned long long)he.offset); else - printf("(from "LPX64 " to "LPX64")\n", - he.offset, he.offset + he.length); + printf("(from %llu to %llu)\n", + (unsigned long long)he.offset, + (unsigned long long)(he.offset + he.length)); } while (++i < argc); @@ -3573,7 +3810,7 @@ static int lfs_hsm_clear(int argc, char **argv) * * \return 0 on success. */ -static int lfs_hsm_prepare_file(char *file, struct lu_fid *fid, +static int lfs_hsm_prepare_file(const char *file, struct lu_fid *fid, dev_t *last_dev) { struct stat st; @@ -3584,6 +3821,13 @@ static int lfs_hsm_prepare_file(char *file, struct lu_fid *fid, fprintf(stderr, "Cannot stat %s: %s\n", file, strerror(errno)); return -errno; } + /* Checking for regular file as archiving as posix copytool + * rejects archiving files other than regular files + */ + if (!S_ISREG(st.st_mode)) { + fprintf(stderr, "error: \"%s\" is not a regular file\n", file); + return CMD_HELP; + } /* A request should be ... */ if (*last_dev != st.st_dev && *last_dev != 0) { fprintf(stderr, "All files should be " @@ -3601,16 +3845,62 @@ static int lfs_hsm_prepare_file(char *file, struct lu_fid *fid, return 0; } +/* Fill an HSM HUR item with a given file name. + * + * If mntpath is set, then the filename is actually a FID, and no + * lookup on the filesystem will be performed. + * + * \param[in] hur the user request to fill + * \param[in] idx index of the item inside the HUR to fill + * \param[in] mntpath mountpoint of Lustre + * \param[in] fname filename (if mtnpath is NULL) + * or FID (if mntpath is set) + * \param[in] last_dev pointer to last device id used + * + * \retval 0 on success + * \retval CMD_HELP or a negative errno on error + */ +static int fill_hur_item(struct hsm_user_request *hur, unsigned int idx, + const char *mntpath, const char *fname, + dev_t *last_dev) +{ + struct hsm_user_item *hui = &hur->hur_user_item[idx]; + int rc; + + hui->hui_extent.length = -1; + + if (mntpath != NULL) { + if (*fname == '[') + fname++; + rc = sscanf(fname, SFID, RFID(&hui->hui_fid)); + if (rc == 3) { + rc = 0; + } else { + fprintf(stderr, "hsm: '%s' is not a valid FID\n", + fname); + rc = -EINVAL; + } + } else { + rc = lfs_hsm_prepare_file(fname, &hui->hui_fid, last_dev); + } + + if (rc == 0) + hur->hur_request.hr_itemcount++; + + return rc; +} + static int lfs_hsm_request(int argc, char **argv, int action) { struct option long_opts[] = { {"filelist", 1, 0, 'l'}, {"data", 1, 0, 'D'}, {"archive", 1, 0, 'a'}, + {"mntpath", 1, 0, 'm'}, {0, 0, 0, 0} }; dev_t last_dev = 0; - char short_opts[] = "l:D:a:"; + char short_opts[] = "l:D:a:m:"; struct hsm_user_request *hur, *oldhur; int c, i; size_t len; @@ -3623,13 +3913,13 @@ static int lfs_hsm_request(int argc, char **argv, int action) int archive_id = 0; FILE *fp; int nbfile_alloc = 0; - char some_file[PATH_MAX+1] = ""; + char *some_file = NULL; + char *mntpath = NULL; int rc; if (argc < 2) return CMD_HELP; - optind = 0; while ((c = getopt_long(argc, argv, short_opts, long_opts, NULL)) != -1) { switch (c) { @@ -3640,14 +3930,21 @@ static int lfs_hsm_request(int argc, char **argv, int action) opaque = optarg; break; case 'a': - if (action != HUA_ARCHIVE) { + if (action != HUA_ARCHIVE && + action != HUA_REMOVE) { fprintf(stderr, "error: -a is supported only " - "when archiving\n"); + "when archiving or removing\n"); return CMD_HELP; } archive_id = atoi(optarg); break; + case 'm': + if (some_file == NULL) { + mntpath = optarg; + some_file = strdup(optarg); + } + break; case '?': return CMD_HELP; default: @@ -3681,20 +3978,12 @@ static int lfs_hsm_request(int argc, char **argv, int action) hur->hur_request.hr_flags = 0; /* All remaining args are files, add them */ - if (nbfile != 0) { - if (strlen(argv[optind]) > sizeof(some_file)-1) { - free(hur); - return -E2BIG; - } - strncpy(some_file, argv[optind], sizeof(some_file)); - } + if (nbfile != 0 && some_file == NULL) + some_file = strdup(argv[optind]); for (i = 0; i < nbfile; i++) { - hur->hur_user_item[i].hui_extent.length = -1; - rc = lfs_hsm_prepare_file(argv[optind + i], - &hur->hur_user_item[i].hui_fid, - &last_dev); - hur->hur_request.hr_itemcount++; + rc = fill_hur_item(hur, i, mntpath, argv[optind + i], + &last_dev); if (rc) goto out_free; } @@ -3712,17 +4001,17 @@ static int lfs_hsm_request(int argc, char **argv, int action) } while ((rc = getline(&line, &len, fp)) != -1) { - struct hsm_user_item *hui; - - /* If allocated buffer was too small, gets something - * bigger */ + /* If allocated buffer was too small, get something + * larger */ if (nbfile_alloc <= hur->hur_request.hr_itemcount) { + ssize_t size; + nbfile_alloc = nbfile_alloc * 2 + 1; oldhur = hur; hur = llapi_hsm_user_request_alloc(nbfile_alloc, opaque_len); if (hur == NULL) { - fprintf(stderr, "Cannot allocate " + fprintf(stderr, "hsm: cannot allocate " "the request: %s\n", strerror(errno)); hur = oldhur; @@ -3730,7 +4019,19 @@ static int lfs_hsm_request(int argc, char **argv, int action) fclose(fp); goto out_free; } - memcpy(hur, oldhur, hur_len(oldhur)); + size = hur_len(oldhur); + if (size < 0) { + fprintf(stderr, "hsm: cannot allocate " + "%u files + %u bytes data\n", + oldhur->hur_request.hr_itemcount, + oldhur->hur_request.hr_data_len); + free(hur); + hur = oldhur; + rc = -E2BIG; + fclose(fp); + goto out_free; + } + memcpy(hur, oldhur, size); free(oldhur); } @@ -3738,25 +4039,21 @@ static int lfs_hsm_request(int argc, char **argv, int action) if (line[strlen(line) - 1] == '\n') line[strlen(line) - 1] = '\0'; - hui = - &hur->hur_user_item[hur->hur_request.hr_itemcount]; - hui->hui_extent.length = -1; - rc = lfs_hsm_prepare_file(line, &hui->hui_fid, - &last_dev); - hur->hur_request.hr_itemcount++; + rc = fill_hur_item(hur, hur->hur_request.hr_itemcount, + mntpath, line, &last_dev); if (rc) { fclose(fp); goto out_free; } - if ((some_file[0] == '\0') && - (strlen(line) < sizeof(some_file))) - strcpy(some_file, line); + if (some_file == NULL) { + some_file = line; + line = NULL; + } } rc = fclose(fp); - if (line) - free(line); + free(line); } /* If a --data was used, add it to the request */ @@ -3777,6 +4074,7 @@ static int lfs_hsm_request(int argc, char **argv, int action) } out_free: + free(some_file); free(hur); return rc; } @@ -3820,21 +4118,21 @@ int main(int argc, char **argv) { int rc; - setlinebuf(stdout); + /* Ensure that liblustreapi constructor has run */ + if (!liblustreapi_initialized) + fprintf(stderr, "liblustreapi was not properly initialized\n"); - ptl_initialize(argc, argv); - if (obd_initialize(argc, argv) < 0) - exit(2); + setlinebuf(stdout); - Parser_init("lfs > ", cmdlist); + Parser_init("lfs > ", cmdlist); + progname = argv[0]; /* Used in error messages */ if (argc > 1) { rc = Parser_execarg(argc - 1, argv + 1, cmdlist); } else { rc = Parser_commands(); } - obd_finalize(argc, argv); return rc < 0 ? -rc : rc; }