Whamcloud - gitweb
LU-5095 hsm: Allow receiving messages to be non-blocking
[fs/lustre-release.git] / lustre / utils / liblustreapi_hsm.c
index 3f1b8b9..e47a7d9 100644 (file)
@@ -53,9 +53,9 @@
 #include <unistd.h>
 #endif
 
-#include <liblustre.h>
+#include <libcfs/libcfs.h>
 #include <lnet/lnetctl.h>
-#include <obd.h>
+#include <lustre/lustre_idl.h>
 #include <lustre/lustreapi.h>
 #include "lustreapi_internal.h"
 
@@ -66,6 +66,7 @@
 struct hsm_copytool_private {
        int                      magic;
        char                    *mnt;
+       struct kuc_hdr          *kuch;
        int                      mnt_fd;
        int                      open_by_fid_fd;
        lustre_kernelcomm        kuc;
@@ -113,7 +114,7 @@ enum ct_event {
 };
 
 /* initialized in llapi_hsm_register_event_fifo() */
-FILE *llapi_hsm_event_fp;
+int llapi_hsm_event_fd = -1;
 
 static inline const char *llapi_hsm_ct_ev2str(int type)
 {
@@ -173,12 +174,14 @@ int llapi_hsm_write_json_event(struct llapi_json_item_list **event)
 {
        int                             rc;
        char                            time_string[40];
+       char                            json_buf[PIPE_BUF];
+       FILE                            *buf_file;
        time_t                          event_time = time(0);
        struct tm                       time_components;
        struct llapi_json_item_list     *json_items;
 
-       /* Noop unless the event fp was initialized */
-       if (llapi_hsm_event_fp == NULL)
+       /* Noop unless the event fd was initialized */
+       if (llapi_hsm_event_fd < 0)
                return 0;
 
        if (event == NULL || *event == NULL)
@@ -203,21 +206,24 @@ int llapi_hsm_write_json_event(struct llapi_json_item_list **event)
                return rc;
        }
 
-       rc = llapi_json_write_list(event, llapi_hsm_event_fp);
-       if (rc < 0) {
-               /* Ignore write failures due to missing reader. */
-               if (rc == -EPIPE)
-                       return 0;
+       buf_file = fmemopen(json_buf, sizeof(json_buf), "w");
+       if (buf_file == NULL)
+               return -errno;
 
-               /* Skip llapi_error() here because there's no point
-                * in creating a JSON-formatted error message about
-                * failing to write a JSON-formatted message.
-                */
-               fprintf(stderr,
-                       "\nFATAL ERROR IN llapi_hsm_write_list(): rc %d", rc);
+       rc = llapi_json_write_list(event, buf_file);
+       if (rc < 0) {
+               fclose(buf_file);
                return rc;
        }
 
+       fclose(buf_file);
+
+       if (write(llapi_hsm_event_fd, json_buf, strlen(json_buf)) < 0) {
+               /* Ignore write failures due to missing reader. */
+               if (errno != EPIPE)
+                       return -errno;
+       }
+
        return 0;
 }
 
@@ -304,16 +310,28 @@ out_free:
  * llapi_hsm_write_json_event:
  *
  * Copytool registration and deregistration:
- * {"event_time": "2014-02-26 14:58:01 -0500", "event_type": "REGISTER", "archive": 0, "mount_point": "/mnt/lustre", "uuid": "80379a60-1f8a-743f-daf2-307cde793ec2"}
- * {"event_time": "2014-02-26 14:58:01 -0500", "event_type": "UNREGISTER", "archive": 0, "mount_point": "/mnt/lustre", "uuid": "80379a60-1f8a-743f-daf2-307cde793ec2"}
+ * {"event_time": "2014-02-26 14:58:01 -0500", "event_type": "REGISTER",
+ *  "archive": 0, "mount_point": "/mnt/lustre",
+ *  "uuid": "80379a60-1f8a-743f-daf2-307cde793ec2"}
+ * {"event_time": "2014-02-26 14:58:01 -0500", "event_type": "UNREGISTER",
+ *  "archive": 0, "mount_point": "/mnt/lustre",
+ *  "uuid": "80379a60-1f8a-743f-daf2-307cde793ec2"}
  *
  * An archive action, start to completion:
- * {"event_time": "2014-02-26 14:50:13 -0500", "event_type": "ARCHIVE_START", "total_bytes": 0, "lustre_path": "d71.sanity-hsm/f71.sanity-hsm", "source_fid": "0x2000013a1:0x2:0x0", "data_fid": "0x2000013a1:0x2:0x0"}
- * {"event_time": "2014-02-26 14:50:18 -0500", "event_type": "ARCHIVE_RUNNING", "current_bytes": 5242880, "total_bytes": 39000000, "lustre_path": "d71.sanity-hsm/f71.sanity-hsm", "source_fid": "0x2000013a1:0x2:0x0", "data_fid": "0x2000013a1:0x2:0x0"}
- * {"event_time": "2014-02-26 14:50:50 -0500", "event_type": "ARCHIVE_FINISH", "source_fid": "0x2000013a1:0x2:0x0", "data_fid": "0x2000013a1:0x2:0x0"}
+ * {"event_time": "2014-02-26 14:50:13 -0500", "event_type": "ARCHIVE_START",
+ *  "total_bytes": 0, "lustre_path": "d71.sanity-hsm/f71.sanity-hsm",
+ *  "source_fid": "0x2000013a1:0x2:0x0", "data_fid": "0x2000013a1:0x2:0x0"}
+ * {"event_time": "2014-02-26 14:50:18 -0500", "event_type": "ARCHIVE_RUNNING",
+ *  "current_bytes": 5242880, "total_bytes": 39000000,
+ *  "lustre_path": "d71.sanity-hsm/f71.sanity-hsm",
+ *  "source_fid": "0x2000013a1:0x2:0x0", "data_fid": "0x2000013a1:0x2:0x0"}
+ * {"event_time": "2014-02-26 14:50:50 -0500", "event_type": "ARCHIVE_FINISH",
+ *  "source_fid": "0x2000013a1:0x2:0x0", "data_fid": "0x2000013a1:0x2:0x0"}
  *
  * A log message:
- * {"event_time": "2014-02-26 14:50:13 -0500", "event_type": "LOGGED_MESSAGE", "level": "INFO", "message": "lhsmtool_posix[59401]: copytool fs=lustre archive#=2 item_count=1"}
+ * {"event_time": "2014-02-26 14:50:13 -0500", "event_type": "LOGGED_MESSAGE",
+ *  "level": "INFO",
+ *  "message": "lhsmtool_posix[42]: copytool fs=lustre archive#=2 item_count=1"}
  *
  * \param hcp                Opaque action handle returned by
  *                           llapi_hsm_action_start.
@@ -443,7 +461,7 @@ out_free:
  */
 int llapi_hsm_register_event_fifo(char *path)
 {
-       int read_fd, write_fd;
+       int read_fd;
        struct stat statbuf;
 
        /* Create the FIFO if necessary. */
@@ -476,8 +494,8 @@ int llapi_hsm_register_event_fifo(char *path)
 
        /* Open the FIFO for writes, but don't block on waiting
         * for a reader. */
-       write_fd = open(path, O_WRONLY | O_NONBLOCK);
-       if (write_fd < 0) {
+       llapi_hsm_event_fd = open(path, O_WRONLY | O_NONBLOCK);
+       if (llapi_hsm_event_fd < 0) {
                llapi_error(LLAPI_MSG_ERROR, errno,
                            "cannot open(%s) for write", path);
                return -errno;
@@ -488,19 +506,9 @@ int llapi_hsm_register_event_fifo(char *path)
         * events are lost. NOTE: Only one reader at a time! */
        close(read_fd);
 
-       llapi_hsm_event_fp = fdopen(write_fd, "w");
-       if (llapi_hsm_event_fp == NULL) {
-               llapi_error(LLAPI_MSG_ERROR, errno,
-                           "cannot fdopen(%s) for write", path);
-               return -errno;
-       }
-
        /* Ignore SIGPIPEs -- can occur if the reader goes away. */
        signal(SIGPIPE, SIG_IGN);
 
-       /* Don't buffer the event stream. */
-       setbuf(llapi_hsm_event_fp, NULL);
-
        return 0;
 }
 
@@ -514,16 +522,16 @@ int llapi_hsm_register_event_fifo(char *path)
  */
 int llapi_hsm_unregister_event_fifo(char *path)
 {
-       /* Noop unless the event fp was initialized */
-       if (llapi_hsm_event_fp == NULL)
+       /* Noop unless the event fd was initialized */
+       if (llapi_hsm_event_fd < 0)
                return 0;
 
-       if (fclose(llapi_hsm_event_fp) != 0)
+       if (close(llapi_hsm_event_fd) < 0)
                return -errno;
 
        unlink(path);
 
-       llapi_hsm_event_fp = NULL;
+       llapi_hsm_event_fd = -1;
 
        return 0;
 }
@@ -550,8 +558,8 @@ void llapi_hsm_log_error(enum llapi_message_level level, int _rc,
        va_list                         args2;
        struct llapi_json_item_list     *json_items;
 
-       /* Noop unless the event fp was initialized */
-       if (llapi_hsm_event_fp == NULL)
+       /* Noop unless the event fd was initialized */
+       if (llapi_hsm_event_fd < 0)
                return;
 
        rc = llapi_json_init_list(&json_items);
@@ -633,13 +641,13 @@ out_free:
 /** Register a copytool
  * \param[out] priv Opaque private control structure
  * \param mnt Lustre filesystem mount point
- * \param flags Open flags, currently unused (e.g. O_NONBLOCK)
  * \param archive_count
  * \param archives Which archive numbers this copytool is responsible for
+ * \param rfd_flags flags applied to read fd of pipe (e.g. O_NONBLOCK)
  */
 int llapi_hsm_copytool_register(struct hsm_copytool_private **priv,
-                               const char *mnt, int flags, int archive_count,
-                               int *archives)
+                               const char *mnt, int archive_count,
+                               int *archives, int rfd_flags)
 {
        struct hsm_copytool_private     *ct;
        int                              rc;
@@ -666,6 +674,12 @@ int llapi_hsm_copytool_register(struct hsm_copytool_private **priv,
                goto out_err;
        }
 
+       ct->kuch = malloc(HAL_MAXSIZE + sizeof(*ct->kuch));
+       if (ct->kuch == NULL) {
+               rc = -ENOMEM;
+               goto out_err;
+       }
+
        ct->mnt_fd = open(ct->mnt, O_RDONLY);
        if (ct->mnt_fd < 0) {
                rc = -errno;
@@ -698,7 +712,7 @@ int llapi_hsm_copytool_register(struct hsm_copytool_private **priv,
                ct->archives |= (1 << (archives[rc] - 1));
        }
 
-       rc = libcfs_ukuc_start(&ct->kuc, KUC_GRP_HSM);
+       rc = libcfs_ukuc_start(&ct->kuc, KUC_GRP_HSM, rfd_flags);
        if (rc < 0)
                goto out_err;
 
@@ -736,8 +750,9 @@ out_err:
        if (!(ct->open_by_fid_fd < 0))
                close(ct->open_by_fid_fd);
 
-       if (ct->mnt != NULL)
-               free(ct->mnt);
+       free(ct->mnt);
+
+       free(ct->kuch);
 
        free(ct);
 
@@ -772,18 +787,34 @@ int llapi_hsm_copytool_unregister(struct hsm_copytool_private **priv)
        close(ct->open_by_fid_fd);
        close(ct->mnt_fd);
        free(ct->mnt);
+       free(ct->kuch);
        free(ct);
        *priv = NULL;
 
        return 0;
 }
 
+/** Returns a file descriptor to poll/select on.
+ * \param ct Opaque private control structure
+ * \retval -EINVAL on error
+ * \retval the file descriptor for reading HSM events from the kernel
+ */
+int llapi_hsm_copytool_get_fd(struct hsm_copytool_private *ct)
+{
+       if (ct == NULL || ct->magic != CT_PRIV_MAGIC)
+               return -EINVAL;
+
+       return libcfs_ukuc_get_rfd(&ct->kuc);
+}
+
 /** Wait for the next hsm_action_list
  * \param ct Opaque private control structure
  * \param halh Action list handle, will be allocated here
  * \param msgsize Number of bytes in the message, will be set here
  * \return 0 valid message received; halh and msgsize are set
  *        <0 error code
+ * Note: The application must not call llapi_hsm_copytool_recv until it has
+ * cleared the data in ct->kuch from the previous call.
  */
 int llapi_hsm_copytool_recv(struct hsm_copytool_private *ct,
                            struct hsm_action_list **halh, int *msgsize)
@@ -798,21 +829,20 @@ int llapi_hsm_copytool_recv(struct hsm_copytool_private *ct,
        if (halh == NULL || msgsize == NULL)
                return -EINVAL;
 
-       kuch = malloc(HAL_MAXSIZE + sizeof(*kuch));
-       if (kuch == NULL)
-               return -ENOMEM;
+       kuch = ct->kuch;
 
+repeat:
        rc = libcfs_ukuc_msg_get(&ct->kuc, (char *)kuch,
                                 HAL_MAXSIZE + sizeof(*kuch),
                                 KUC_TRANSPORT_HSM);
        if (rc < 0)
-               goto out_free;
+               goto out_err;
 
        /* Handle generic messages */
        if (kuch->kuc_transport == KUC_TRANSPORT_GENERIC &&
            kuch->kuc_msgtype == KUC_MSG_SHUTDOWN) {
                rc = -ESHUTDOWN;
-               goto out_free;
+               goto out_err;
        }
 
        if (kuch->kuc_transport != KUC_TRANSPORT_HSM ||
@@ -821,14 +851,14 @@ int llapi_hsm_copytool_recv(struct hsm_copytool_private *ct,
                                  "Unknown HSM message type %d:%d\n",
                                  kuch->kuc_transport, kuch->kuc_msgtype);
                rc = -EPROTO;
-               goto out_free;
+               goto out_err;
        }
 
        if (kuch->kuc_msglen < sizeof(*kuch) + sizeof(*hal)) {
                llapi_err_noerrno(LLAPI_MSG_ERROR, "Short HSM message %d",
                                  kuch->kuc_msglen);
                rc = -EPROTO;
-               goto out_free;
+               goto out_err;
        }
 
        /* Our message is a hsm_action_list. Use pointer math to skip
@@ -845,29 +875,20 @@ int llapi_hsm_copytool_recv(struct hsm_copytool_private *ct,
                                  " ignoring this request."
                                  " Mask of served archive is 0x%.8X",
                                  hal->hal_archive_id, ct->archives);
-               rc = -EAGAIN;
 
-               goto out_free;
+               goto repeat;
        }
 
        *halh = hal;
        *msgsize = kuch->kuc_msglen - sizeof(*kuch);
        return 0;
 
-out_free:
+out_err:
        *halh = NULL;
        *msgsize = 0;
-       free(kuch);
        return rc;
 }
 
-/** Release the action list when done with it. */
-void llapi_hsm_action_list_free(struct hsm_action_list **hal)
-{
-       /* Reuse the llapi_changelog_free function */
-       llapi_changelog_free((struct changelog_ext_rec **)hal);
-}
-
 /** Get parent path from mount point and fid.
  *
  * \param mnt        Filesystem root path.
@@ -914,10 +935,12 @@ static int ct_open_by_fid(const struct hsm_copytool_private *ct,
                          const struct lu_fid *fid, int open_flags)
 {
        char fid_name[FID_NOBRACE_LEN + 1];
+       int fd;
 
        snprintf(fid_name, sizeof(fid_name), DFID_NOBRACE, PFID(fid));
 
-       return openat(ct->open_by_fid_fd, fid_name, open_flags);
+       fd = openat(ct->open_by_fid_fd, fid_name, open_flags);
+       return fd < 0 ? -errno : fd;
 }
 
 static int ct_stat_by_fid(const struct hsm_copytool_private *ct,
@@ -925,10 +948,12 @@ static int ct_stat_by_fid(const struct hsm_copytool_private *ct,
                          struct stat *buf)
 {
        char fid_name[FID_NOBRACE_LEN + 1];
+       int rc;
 
        snprintf(fid_name, sizeof(fid_name), DFID_NOBRACE, PFID(fid));
 
-       return fstatat(ct->open_by_fid_fd, fid_name, buf, 0);
+       rc = fstatat(ct->open_by_fid_fd, fid_name, buf, 0);
+       return rc ? -errno : 0;
 }
 
 /** Create the destination volatile file for a restore operation.
@@ -1199,17 +1224,20 @@ int llapi_hsm_action_get_dfid(const struct hsm_copyaction_private *hcp,
 int llapi_hsm_action_get_fd(const struct hsm_copyaction_private *hcp)
 {
        const struct hsm_action_item    *hai = &hcp->copy.hc_hai;
+       int fd;
 
        if (hcp->magic != CP_PRIV_MAGIC)
                return -EINVAL;
 
-       if (hai->hai_action == HSMA_ARCHIVE)
+       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);
-       else if (hai->hai_action == HSMA_RESTORE)
-               return dup(hcp->data_fd);
-       else
+       } else if (hai->hai_action == HSMA_RESTORE) {
+               fd = dup(hcp->data_fd);
+               return fd < 0 ? -errno : fd;
+       } else {
                return -EINVAL;
+       }
 }
 
 /**
@@ -1243,9 +1271,9 @@ int llapi_hsm_import(const char *dst, int archive, const struct stat *st,
                                  stripe_pattern | LOV_PATTERN_F_RELEASED,
                                  pool_name);
        if (fd < 0) {
-               llapi_error(LLAPI_MSG_ERROR, -errno,
+               llapi_error(LLAPI_MSG_ERROR, fd,
                            "cannot create '%s' for import", dst);
-               return -errno;
+               return fd;
        }
 
        /* Get the new fid in Lustre. Caller needs to use this fid
@@ -1268,8 +1296,8 @@ int llapi_hsm_import(const char *dst, int archive, const struct stat *st,
        hui.hui_mtime_ns = st->st_mtim.tv_nsec;
        rc = ioctl(fd, LL_IOC_HSM_IMPORT, &hui);
        if (rc != 0) {
-               llapi_error(LLAPI_MSG_ERROR, rc, "cannot import '%s'", dst);
                rc = -errno;
+               llapi_error(LLAPI_MSG_ERROR, rc, "cannot import '%s'", dst);
                goto out_unlink;
        }