X-Git-Url: http://git.ipfire.org/?a=blobdiff_plain;f=monitor.c;h=bd67e1ad1e29798a117181cacc4c5d51ad89164e;hb=7801ac209240ca5d5159d2ab990dd8d5573e2195;hp=e941a1abd8352548660122ae6a3770cac265e35d;hpb=0af73f61a25904edc7da24e2da9786b48bb8bec6;p=thirdparty%2Fmdadm.git diff --git a/monitor.c b/monitor.c index e941a1ab..bd67e1ad 100644 --- a/monitor.c +++ b/monitor.c @@ -1,9 +1,9 @@ #include "mdadm.h" #include "mdmon.h" - +#include #include - +#include static char *array_states[] = { "clear", "inactive", "suspended", "readonly", "read-auto", @@ -47,24 +47,7 @@ static int read_attr(char *buf, int len, int fd) return n; } -static int get_sync_pos(struct active_array *a) -{ - char buf[30]; - int n; - - n = read_attr(buf, 30, a->sync_pos_fd); - if (n <= 0) - return n; - - if (strncmp(buf, "max", 3) == 0) { - a->sync_pos = ~(unsigned long long)0; - return 1; - } - a->sync_pos = strtoull(buf, NULL, 10); - return 1; -} - -static int get_resync_start(struct active_array *a) +int get_resync_start(struct active_array *a) { char buf[30]; int n; @@ -78,30 +61,6 @@ static int get_resync_start(struct active_array *a) return 1; } -static int attr_match(const char *attr, const char *str) -{ - /* See if attr, read from a sysfs file, matches - * str. They must either be the same, or attr can - * have a trailing newline or comma - */ - while (*attr && *str && *attr == *str) { - attr++; - str++; - } - - if (*str || (*attr && *attr != ',' && *attr != '\n')) - return 0; - return 1; -} - -static int match_word(const char *word, char **list) -{ - int n; - for (n=0; list[n]; n++) - if (attr_match(word, list[n])) - break; - return n; -} static enum array_state read_state(int fd) { @@ -110,7 +69,7 @@ static enum array_state read_state(int fd) if (n <= 0) return bad_word; - return (enum array_state) match_word(buf, array_states); + return (enum array_state) sysfs_match_word(buf, array_states); } static enum sync_action read_action( int fd) @@ -120,7 +79,7 @@ static enum sync_action read_action( int fd) if (n <= 0) return bad_action; - return (enum sync_action) match_word(buf, sync_actions); + return (enum sync_action) sysfs_match_word(buf, sync_actions); } int read_dev_state(int fd) @@ -135,15 +94,15 @@ int read_dev_state(int fd) cp = buf; while (cp) { - if (attr_match(cp, "faulty")) + if (sysfs_attr_match(cp, "faulty")) rv |= DS_FAULTY; - if (attr_match(cp, "in_sync")) + if (sysfs_attr_match(cp, "in_sync")) rv |= DS_INSYNC; - if (attr_match(cp, "write_mostly")) + if (sysfs_attr_match(cp, "write_mostly")) rv |= DS_WRITE_MOSTLY; - if (attr_match(cp, "spare")) + if (sysfs_attr_match(cp, "spare")) rv |= DS_SPARE; - if (attr_match(cp, "blocked")) + if (sysfs_attr_match(cp, "blocked")) rv |= DS_BLOCKED; cp = strchr(cp, ','); if (cp) @@ -152,6 +111,12 @@ int read_dev_state(int fd) return rv; } +static void signal_manager(void) +{ + /* tgkill(getpid(), mon_tid, SIGUSR1); */ + int pid = getpid(); + syscall(SYS_tgkill, pid, mgr_tid, SIGUSR1); +} /* Monitor a set of active md arrays - all of which share the * same metadata - and respond to events that require @@ -183,24 +148,10 @@ int read_dev_state(int fd) * sync_action was 'resync' and becomes 'idle' and resync_start becomes * MaxSector * Notify metadata that sync is complete. - * "Deal with Degraded" * * recovery completes * sync_action changes from 'recover' to 'idle' * Check each device state and mark metadata if 'faulty' or 'in_sync'. - * "Deal with Degraded" - * - * deal with degraded array - * We only do this when first noticing the array is degraded. - * This can be when we first see the array, when sync completes or - * when recovery completes. - * - * Check if number of failed devices suggests recovery is needed, and - * skip if not. - * Ask metadata for a spare device - * Add device as not in_sync and give a role - * Update metadata. - * Start recovery. * * deal with resync * This only happens on finding a new array... mdadm will have set @@ -226,7 +177,7 @@ int read_dev_state(int fd) static int read_and_act(struct active_array *a) { - int check_degraded; + int check_degraded = 0; int deactivate = 0; struct mdinfo *mdi; @@ -237,53 +188,71 @@ static int read_and_act(struct active_array *a) a->curr_action = read_action(a->action_fd); for (mdi = a->info.devs; mdi ; mdi = mdi->next) { mdi->next_state = 0; - if (mdi->state_fd > 0) + if (mdi->state_fd >= 0) mdi->curr_state = read_dev_state(mdi->state_fd); } if (a->curr_state <= inactive && a->prev_state > inactive) { /* array has been stopped */ - get_sync_pos(a); - a->container->ss->mark_clean(a, a->sync_pos); + get_resync_start(a); + a->container->ss->set_array_state(a, 1); a->next_state = clear; deactivate = 1; } if (a->curr_state == write_pending) { - a->container->ss->mark_dirty(a); + get_resync_start(a); + a->container->ss->set_array_state(a, 0); a->next_state = active; } if (a->curr_state == active_idle) { - /* Set array to 'clean' FIRST, then - * a->ss->mark_clean(a); - * just ignore for now. + /* Set array to 'clean' FIRST, then mark clean + * in the metadata */ + a->next_state = clean; + } + if (a->curr_state == clean) { + get_resync_start(a); + a->container->ss->set_array_state(a, 1); } if (a->curr_state == readonly) { - /* Well, I'm ready to handle things, so - * read-auto is OK. FIXME what if we really want - * readonly ??? + /* Well, I'm ready to handle things. If readonly + * wasn't requested, transition to read-auto. */ - get_resync_start(a); - if (a->resync_start == ~0ULL) - a->next_state = read_auto; /* array is clean */ - else { - a->container->ss->mark_dirty(a); - a->next_state = active; + char buf[64]; + read_attr(buf, sizeof(buf), a->metadata_fd); + if (strncmp(buf, "external:-", 10) == 0) { + /* explicit request for readonly array. Leave it alone */ + ; + } else { + get_resync_start(a); + if (a->container->ss->set_array_state(a, 2)) + a->next_state = read_auto; /* array is clean */ + else + a->next_state = active; /* Now active for recovery etc */ } } - if (a->curr_action == idle && + if (!deactivate && + a->curr_action == idle && a->prev_action == resync) { - /* check resync_start to see if it is 'max' */ + /* A resync has finished. The endpoint is recorded in + * 'sync_start'. We don't update the metadata + * until the array goes inactive or readonly though. + * Just check if we need to fiddle spares. + */ get_resync_start(a); - a->container->ss->mark_sync(a, a->resync_start); + a->container->ss->set_array_state(a, a->curr_state <= clean); check_degraded = 1; } - if (a->curr_action == idle && + if (!deactivate && + a->curr_action == idle && a->prev_action == recover) { + /* A recovery has finished. Some disks may be in sync now, + * and the array may no longer be degraded + */ for (mdi = a->info.devs ; mdi ; mdi = mdi->next) { a->container->ss->set_disk(a, mdi->disk.raid_disk, mdi->curr_state); @@ -292,45 +261,67 @@ static int read_and_act(struct active_array *a) } } - + /* Check for failures and if found: + * 1/ Record the failure in the metadata and unblock the device. + * FIXME update the kernel to stop notifying on failed drives when + * the array is readonly and we have cleared 'blocked' + * 2/ Try to remove the device if the array is writable, or can be + * made writable. + */ for (mdi = a->info.devs ; mdi ; mdi = mdi->next) { if (mdi->curr_state & DS_FAULTY) { a->container->ss->set_disk(a, mdi->disk.raid_disk, mdi->curr_state); check_degraded = 1; - mdi->next_state = DS_REMOVE; + mdi->next_state |= DS_UNBLOCK; + if (a->curr_state == read_auto) { + a->container->ss->set_array_state(a, 0); + a->next_state = active; + } + if (a->curr_state > readonly) + mdi->next_state |= DS_REMOVE; } } - if (check_degraded) { - // FIXME; - } - - a->container->ss->sync_metadata(a); + a->container->ss->sync_metadata(a->container); + dprintf("%s(%d): state:%s action:%s next(", __func__, a->info.container_member, + array_states[a->curr_state], sync_actions[a->curr_action]); /* Effect state changes in the array */ - if (a->next_state != bad_word) + if (a->next_state != bad_word) { + dprintf(" state:%s", array_states[a->next_state]); write_attr(array_states[a->next_state], a->info.state_fd); - if (a->next_action != bad_action) + } + if (a->next_action != bad_action) { write_attr(sync_actions[a->next_action], a->action_fd); + dprintf(" action:%s", sync_actions[a->next_action]); + } for (mdi = a->info.devs; mdi ; mdi = mdi->next) { - if (mdi->next_state == DS_REMOVE && mdi->state_fd > 0) { - int remove_err; - + if (mdi->next_state & DS_UNBLOCK) { + dprintf(" %d:-blocked", mdi->disk.raid_disk); write_attr("-blocked", mdi->state_fd); + } + + if ((mdi->next_state & DS_REMOVE) && mdi->state_fd >= 0) { + int remove_result; + /* the kernel may not be able to immediately remove the * disk, we can simply wait until the next event to try * again. */ - remove_err = write_attr("remove", mdi->state_fd); - if (!remove_err) { + remove_result = write_attr("remove", mdi->state_fd); + if (remove_result > 0) { + dprintf(" %d:removed", mdi->disk.raid_disk); close(mdi->state_fd); mdi->state_fd = -1; } } - if (mdi->next_state & DS_INSYNC) + if (mdi->next_state & DS_INSYNC) { write_attr("+in_sync", mdi->state_fd); + dprintf(" %d:+in_sync", mdi->disk.raid_disk); + } } + dprintf(" )\n"); /* move curr_ to prev_ */ a->prev_state = a->curr_state; @@ -342,6 +333,12 @@ static int read_and_act(struct active_array *a) mdi->next_state = 0; } + if (check_degraded) { + /* manager will do the actual check */ + a->check_degraded = 1; + signal_manager(); + } + if (deactivate) a->container = NULL; @@ -377,44 +374,119 @@ static void reconcile_failed(struct active_array *aa, struct mdinfo *failed) } } -static int wait_and_act(struct active_array *aa, int pfd, int nowait) +#ifdef DEBUG +static void dprint_wake_reasons(fd_set *fds) +{ + int i; + char proc_path[256]; + char link[256]; + char *basename; + int rv; + + fprintf(stderr, "monitor: wake ( "); + for (i = 0; i < FD_SETSIZE; i++) { + if (FD_ISSET(i, fds)) { + sprintf(proc_path, "/proc/%d/fd/%d", + (int) getpid(), i); + + rv = readlink(proc_path, link, sizeof(link) - 1); + if (rv < 0) { + fprintf(stderr, "%d:unknown ", i); + continue; + } + link[rv] = '\0'; + basename = strrchr(link, '/'); + fprintf(stderr, "%d:%s ", + i, basename ? ++basename : link); + } + } + fprintf(stderr, ")\n"); +} +#endif + +int monitor_loop_cnt; + +static int wait_and_act(struct supertype *container, int nowait) { fd_set rfds; int maxfd = 0; - struct active_array *a; + struct active_array **aap = &container->arrays; + struct active_array *a, **ap; int rv; struct mdinfo *mdi; FD_ZERO(&rfds); - add_fd(&rfds, &maxfd, pfd); - for (a = aa ; a ; a = a->next) { - /* once an array has been deactivated only the manager - * thread can make us care about it again + for (ap = aap ; *ap ;) { + a = *ap; + /* once an array has been deactivated we want to + * ask the manager to discard it. */ - if (!a->container) + if (!a->container) { + if (discard_this) { + ap = &(*ap)->next; + continue; + } + *ap = a->next; + a->next = NULL; + discard_this = a; + signal_manager(); continue; + } add_fd(&rfds, &maxfd, a->info.state_fd); add_fd(&rfds, &maxfd, a->action_fd); for (mdi = a->info.devs ; mdi ; mdi = mdi->next) add_fd(&rfds, &maxfd, mdi->state_fd); + + ap = &(*ap)->next; + } + + if (manager_ready && *aap == NULL) { + /* No interesting arrays. Lets see about exiting. + * Note that blocking at this point is not a problem + * as there are no active arrays, there is nothing that + * we need to be ready to do. + */ + int fd = open(container->device_name, O_RDONLY|O_EXCL); + if (fd >= 0 || errno != EBUSY) { + /* OK, we are safe to leave */ + dprintf("no arrays to monitor... exiting\n"); + remove_pidfile(container->devname); + exit_now = 1; + signal_manager(); + exit(0); + } } if (!nowait) { - rv = select(maxfd+1, &rfds, NULL, NULL, NULL); + sigset_t set; + sigprocmask(SIG_UNBLOCK, NULL, &set); + sigdelset(&set, SIGUSR1); + monitor_loop_cnt |= 1; + rv = pselect(maxfd+1, &rfds, NULL, NULL, NULL, &set); + monitor_loop_cnt += 1; + if (rv == -1 && errno == EINTR) + rv = 0; + #ifdef DEBUG + dprint_wake_reasons(&rfds); + #endif - if (rv <= 0) - return rv; + } - if (FD_ISSET(pfd, &rfds)) { - char buf[4]; - read(pfd, buf, 4); - ; // FIXME read from the pipe - } + if (update_queue) { + struct metadata_update *this; + + for (this = update_queue; this ; this = this->next) + container->ss->process_update(container, this); + + update_queue_handled = update_queue; + update_queue = NULL; + signal_manager(); + container->ss->sync_metadata(container); } - for (a = aa; a ; a = a->next) { + for (a = *aap; a ; a = a->next) { if (a->replaces && !discard_this) { struct active_array **ap; for (ap = &a->next; *ap && *ap != a->replaces; @@ -424,18 +496,20 @@ static int wait_and_act(struct active_array *aa, int pfd, int nowait) *ap = (*ap)->next; discard_this = a->replaces; a->replaces = NULL; + /* FIXME check if device->state_fd need to be cleared?*/ + signal_manager(); } if (a->container) rv += read_and_act(a); } /* propagate failures across container members */ - for (a = aa; a ; a = a->next) { + for (a = *aap; a ; a = a->next) { if (!a->container) continue; for (mdi = a->info.devs ; mdi ; mdi = mdi->next) if (mdi->curr_state & DS_FAULTY) - reconcile_failed(aa, mdi); + reconcile_failed(*aap, mdi); } return rv; @@ -446,7 +520,7 @@ void do_monitor(struct supertype *container) int rv; int first = 1; do { - rv = wait_and_act(container->arrays, container->pipe[0], first); + rv = wait_and_act(container, first); first = 0; } while (rv >= 0); }