* (C) Copyright 2012 Commissariat a l'energie atomique et aux energies
* alternatives
*
- * Copyright (c) 2013, 2016, Intel Corporation.
+ * Copyright (c) 2013, 2017, Intel Corporation.
*
* All rights reserved. This program and the accompanying materials
* are made available under the terms of the GNU Lesser General Public License
/****** HSM Copytool API ********/
#define CT_PRIV_MAGIC 0xC0BE2001
struct hsm_copytool_private {
- int magic;
- char *mnt;
- struct kuc_hdr *kuch;
- int mnt_fd;
- int open_by_fid_fd;
- struct lustre_kernelcomm kuc;
- __u32 archives;
+ int magic;
+ char *mnt;
+ struct kuc_hdr *kuch;
+ int mnt_fd;
+ int open_by_fid_fd;
+ struct lustre_kernelcomm *kuc;
};
#define CP_PRIV_MAGIC 0x19880429
struct hsm_copyaction_private {
__u32 magic;
+ __u32 source_fd;
__s32 data_fd;
const struct hsm_copytool_private *ct_priv;
struct hsm_copy copy;
- lstat_t stat;
+ lstatx_t statx;
};
enum ct_progress_type {
goto err;
rc = llapi_json_add_item(&json_items, "archive", LLAPI_JSON_INTEGER,
- &ct->archives);
+ &ct->kuc->lk_data_count);
if (rc < 0)
goto err;
out_free:
if (json_items != NULL)
llapi_json_destroy_list(&json_items);
-
- return;
}
/** Register a copytool
return -EINVAL;
}
- if (archive_count > LL_HSM_MAX_ARCHIVE) {
- llapi_err_noerrno(LLAPI_MSG_ERROR, "%d requested when maximum "
- "of %zu archives supported", archive_count,
- LL_HSM_MAX_ARCHIVE);
- return -EINVAL;
+ for (rc = 0; rc < archive_count; rc++) {
+ /* in the list we have an all archive wildcard
+ * so move to all archives mode
+ */
+ if (archives[rc] == 0) {
+ archive_count = 0;
+ break;
+ }
}
ct = calloc(1, sizeof(*ct));
ct->magic = CT_PRIV_MAGIC;
ct->mnt_fd = -1;
ct->open_by_fid_fd = -1;
- ct->kuc.lk_rfd = LK_NOFD;
- ct->kuc.lk_wfd = LK_NOFD;
ct->mnt = strdup(mnt);
if (ct->mnt == NULL) {
goto out_err;
}
- ct->kuch = malloc(HAL_MAXSIZE + sizeof(*ct->kuch));
+ ct->kuch = calloc(1, HAL_MAXSIZE + sizeof(*ct->kuch));
if (ct->kuch == NULL) {
rc = -ENOMEM;
goto out_err;
goto out_err;
}
- /* no archives specified means "match all". */
- ct->archives = 0;
+ ct->kuc = malloc(sizeof(*ct) + archive_count * sizeof(__u32));
+ if (ct->kuc == NULL) {
+ rc = -ENOMEM;
+ goto out_err;
+ }
+
+ ct->kuc->lk_rfd = LK_NOFD;
+ ct->kuc->lk_wfd = LK_NOFD;
+
+ rc = libcfs_ukuc_start(ct->kuc, KUC_GRP_HSM, rfd_flags);
+ if (rc < 0)
+ goto out_free_kuc;
+
+ ct->kuc->lk_flags = LK_FLG_DATANR;
+ ct->kuc->lk_data_count = archive_count;
for (rc = 0; rc < archive_count; rc++) {
- if ((archives[rc] > LL_HSM_MAX_ARCHIVE) || (archives[rc] < 0)) {
+ if (archives[rc] < 0) {
llapi_err_noerrno(LLAPI_MSG_ERROR, "%d requested when "
- "archive id [0 - %zu] is supported",
- archives[rc], LL_HSM_MAX_ARCHIVE);
+ "archive id >= 0 is supported",
+ archives[rc]);
rc = -EINVAL;
- goto out_err;
- }
- /* in the list we have an all archive wildcard
- * so move to all archives mode
- */
- if (archives[rc] == 0) {
- ct->archives = 0;
- archive_count = 0;
- break;
+ goto out_kuc;
}
- ct->archives |= (1 << (archives[rc] - 1));
- }
- rc = libcfs_ukuc_start(&ct->kuc, KUC_GRP_HSM, rfd_flags);
- if (rc < 0)
- goto out_err;
+ ct->kuc->lk_data[rc] = archives[rc];
+ }
- /* Storing archive(s) in lk_data; see mdc_ioc_hsm_ct_start */
- ct->kuc.lk_data = ct->archives;
- rc = ioctl(ct->mnt_fd, LL_IOC_HSM_CT_START, &ct->kuc);
+ rc = ioctl(ct->mnt_fd, LL_IOC_HSM_CT_START, ct->kuc);
if (rc < 0) {
rc = -errno;
llapi_error(LLAPI_MSG_ERROR, rc,
llapi_hsm_log_ct_registration(&ct, CT_REGISTER);
/* Only the kernel reference keeps the write side open */
- close(ct->kuc.lk_wfd);
- ct->kuc.lk_wfd = LK_NOFD;
+ close(ct->kuc->lk_wfd);
+ ct->kuc->lk_wfd = LK_NOFD;
*priv = ct;
return 0;
out_kuc:
/* cleanup the kuc channel */
- libcfs_ukuc_stop(&ct->kuc);
+ libcfs_ukuc_stop(ct->kuc);
+
+out_free_kuc:
+ free(ct->kuc);
out_err:
if (!(ct->mnt_fd < 0))
* enters libcfs_kkuc_group_put() acquires kg_sem and blocks
* in pipe_write() due to full pipe; then we attempt to
* unregister and block on kg_sem. */
- libcfs_ukuc_stop(&ct->kuc);
+ libcfs_ukuc_stop(ct->kuc);
/* Tell the kernel to stop sending us messages */
- ct->kuc.lk_flags = LK_FLG_STOP;
- ioctl(ct->mnt_fd, LL_IOC_HSM_CT_START, &ct->kuc);
+ ct->kuc->lk_flags = LK_FLG_STOP;
+ ioctl(ct->mnt_fd, LL_IOC_HSM_CT_START, ct->kuc);
llapi_hsm_log_ct_registration(&ct, CT_UNREGISTER);
close(ct->mnt_fd);
free(ct->mnt);
free(ct->kuch);
+ free(ct->kuc);
free(ct);
*priv = NULL;
if (ct == NULL || ct->magic != CT_PRIV_MAGIC)
return -EINVAL;
- return libcfs_ukuc_get_rfd(&ct->kuc);
+ return libcfs_ukuc_get_rfd(ct->kuc);
}
/** Wait for the next hsm_action_list
kuch = ct->kuch;
repeat:
- rc = libcfs_ukuc_msg_get(&ct->kuc, (char *)kuch,
+ rc = libcfs_ukuc_msg_get(ct->kuc, (char *)kuch,
HAL_MAXSIZE + sizeof(*kuch),
KUC_TRANSPORT_HSM);
if (rc < 0)
/* Check that we have registered for this archive #
* if 0 registered, we serve any archive */
- if (ct->archives &&
- ((1 << (hal->hal_archive_id - 1)) & ct->archives) == 0) {
- llapi_err_noerrno(LLAPI_MSG_INFO,
- "This copytool does not service archive #%d,"
- " ignoring this request."
- " Mask of served archive is 0x%.8X",
- hal->hal_archive_id, ct->archives);
-
- goto repeat;
+ if (ct->kuc != NULL && ct->kuc->lk_data_count != 0) {
+ int i;
+
+ for (i = 0; i < ct->kuc->lk_data_count; i++) {
+ if (hal->hal_archive_id == ct->kuc->lk_data[i])
+ break;
+ }
+
+ if (i >= ct->kuc->lk_data_count)
+ goto repeat;
}
*halh = hal;
*/
static int ct_md_getattr(const struct hsm_copytool_private *ct,
const struct lu_fid *fid,
- lstat_t *st)
+ lstatx_t *stx)
{
struct lov_user_mds_data *lmd;
+ char fname[FID_NOBRACE_LEN + 1] = "";
size_t lmd_size;
int rc;
- lmd_size = sizeof(lmd->lmd_st) +
- lov_user_md_size(LOV_MAX_STRIPE_COUNT, LOV_USER_MAGIC_V3);
+ rc = snprintf(fname, sizeof(fname), DFID_NOBRACE, PFID(fid));
+ if (rc < 0)
+ return rc;
+ if (rc >= sizeof(fname) || rc == 0)
+ return -EINVAL;
- if (lmd_size < sizeof(lmd->lmd_st) + XATTR_SIZE_MAX)
- lmd_size = sizeof(lmd->lmd_st) + XATTR_SIZE_MAX;
+ lmd_size = offsetof(typeof(*lmd), lmd_lmm) +
+ lov_user_md_size(LOV_MAX_STRIPE_COUNT, LOV_USER_MAGIC_V3);
- if (lmd_size < FID_NOBRACE_LEN + 1)
- lmd_size = FID_NOBRACE_LEN + 1;
+ if (lmd_size < offsetof(typeof(*lmd), lmd_lmm) + XATTR_SIZE_MAX)
+ lmd_size = offsetof(typeof(*lmd), lmd_lmm) + XATTR_SIZE_MAX;
lmd = malloc(lmd_size);
if (lmd == NULL)
return -ENOMEM;
- snprintf((char *)lmd, lmd_size, DFID_NOBRACE, PFID(fid));
-
- rc = ioctl(ct->open_by_fid_fd, IOC_MDC_GETFILEINFO, lmd);
- if (rc != 0) {
- rc = -errno;
- llapi_error(LLAPI_MSG_ERROR, rc,
- "cannot get metadata attributes of "DFID" in '%s'",
- PFID(fid), ct->mnt);
+ rc = get_lmd_info_fd(fname, ct->open_by_fid_fd, -1,
+ lmd, lmd_size, GET_LMD_INFO);
+ if (rc)
goto out;
- }
- *st = lmd->lmd_st;
+ *stx = lmd->lmd_stx;
out:
free(lmd);
if (fd < 0)
return fd;
- rc = fchown(fd, hcp->stat.st_uid, hcp->stat.st_gid);
+ rc = fchown(fd, hcp->statx.stx_uid, hcp->statx.stx_gid);
if (rc < 0)
goto err_cleanup;
int restore_mdt_index, int restore_open_flags,
bool is_error)
{
- struct hsm_copyaction_private *hcp;
- int rc;
+ struct hsm_copyaction_private *hcp;
+ int fd;
+ int rc;
hcp = calloc(1, sizeof(*hcp));
if (hcp == NULL)
return -ENOMEM;
+ hcp->source_fd = -1;
hcp->data_fd = -1;
hcp->ct_priv = ct;
hcp->copy.hc_hai = *hai;
if (is_error)
goto ok_out;
- if (hai->hai_action == HSMA_RESTORE) {
- rc = ct_md_getattr(hcp->ct_priv, &hai->hai_fid, &hcp->stat);
+ if (hai->hai_action == HSMA_ARCHIVE) {
+ fd = ct_open_by_fid(hcp->ct_priv, &hai->hai_dfid,
+ O_RDONLY | O_NOATIME | O_NOFOLLOW | O_NONBLOCK);
+ if (fd < 0) {
+ rc = fd;
+ goto err_out;
+ }
+
+ hcp->source_fd = fd;
+ } else if (hai->hai_action == HSMA_RESTORE) {
+ rc = ct_md_getattr(hcp->ct_priv, &hai->hai_fid, &hcp->statx);
if (rc < 0)
goto err_out;
restore_open_flags);
if (rc < 0)
goto err_out;
- }
+ } else if (hai->hai_action == HSMA_REMOVE) {
+ /* Since remove is atomic there is no need to send an
+ * initial MDS_HSM_PROGRESS RPC.
+ * RW-PCC uses Lustre HSM mechanism for data synchronization.
+ * At the beginning of RW-PCC attach, the client tries to
+ * exclusively open the file by using a lease lock. A
+ * successful lease open ensures that the current attach
+ * process is the unique opener for the file.
+ * After taking the lease, the file data is then copied from
+ * OSTs into PCC and then the client closes the lease with
+ * with a PCC attach intent.
+ * However, for a file with HSM exists, archived state (i.e. a
+ * cached file just was detached from PCC and restore into
+ * OST), a HSM REMOVE request may delete the above PCC copy
+ * during RW-PCC attach wrongly.
+ * Thus, a open/close on the corresponding Lustre file is added
+ * for HSMA_REMOVE here to solve this conflict.
+ */
+ fd = ct_open_by_fid(hcp->ct_priv, &hai->hai_fid,
+ O_RDONLY | O_NOATIME | O_NOFOLLOW | O_NONBLOCK);
+ if (fd < 0) {
+ rc = fd;
+ /* ignore the error in case of Remove Archive on Last
+ * Unlink (RAoLU).
+ */
+ if (rc == -ENOENT) {
+ rc = 0;
+ goto out_log;
+ }
+ goto err_out;
+ }
- /* Since remove is atomic there is no need to send an initial
- * MDS_HSM_PROGRESS RPC. */
- if (hai->hai_action == HSMA_REMOVE)
+ hcp->source_fd = fd;
goto out_log;
+ }
rc = ioctl(ct->mnt_fd, LL_IOC_HSM_COPY_START, &hcp->copy);
if (rc < 0) {
return 0;
err_out:
+ if (!(hcp->source_fd < 0))
+ close(hcp->source_fd);
+
if (!(hcp->data_fd < 0))
close(hcp->data_fd);
if (hai->hai_action == HSMA_RESTORE && errval == 0) {
struct ll_futimes_3 lfu = {
- .lfu_atime_sec = hcp->stat.st_atim.tv_sec,
- .lfu_atime_nsec = hcp->stat.st_atim.tv_nsec,
- .lfu_mtime_sec = hcp->stat.st_mtim.tv_sec,
- .lfu_mtime_nsec = hcp->stat.st_mtim.tv_nsec,
- .lfu_ctime_sec = hcp->stat.st_ctim.tv_sec,
- .lfu_ctime_nsec = hcp->stat.st_ctim.tv_nsec,
+ .lfu_atime_sec = hcp->statx.stx_atime.tv_sec,
+ .lfu_atime_nsec = hcp->statx.stx_atime.tv_nsec,
+ .lfu_mtime_sec = hcp->statx.stx_mtime.tv_sec,
+ .lfu_mtime_nsec = hcp->statx.stx_mtime.tv_nsec,
+ .lfu_ctime_sec = hcp->statx.stx_ctime.tv_sec,
+ .lfu_ctime_nsec = hcp->statx.stx_ctime.tv_nsec,
};
rc = fsync(hcp->data_fd);
llapi_hsm_log_ct_progress(&hcp, hai, CT_FINISH, 0, 0);
err_cleanup:
+ if (!(hcp->source_fd < 0))
+ close(hcp->source_fd);
+
if (!(hcp->data_fd < 0))
close(hcp->data_fd);
return -EINVAL;
if (hai->hai_action == HSMA_ARCHIVE) {
- return ct_open_by_fid(hcp->ct_priv, &hai->hai_dfid,
- O_RDONLY | O_NOATIME | O_NOFOLLOW | O_NONBLOCK);
+ fd = dup(hcp->source_fd);
+ return fd < 0 ? -errno : fd;
} else if (hai->hai_action == HSMA_RESTORE) {
fd = dup(hcp->data_fd);
return fd < 0 ? -errno : fd;