diff options
Diffstat (limited to 'fs/xfs/xfs_log.c')
| -rw-r--r-- | fs/xfs/xfs_log.c | 3779 |
1 files changed, 2036 insertions, 1743 deletions
diff --git a/fs/xfs/xfs_log.c b/fs/xfs/xfs_log.c index a75edca1860..292308dede6 100644 --- a/fs/xfs/xfs_log.c +++ b/fs/xfs/xfs_log.c @@ -17,239 +17,327 @@ */ #include "xfs.h" #include "xfs_fs.h" -#include "xfs_types.h" -#include "xfs_bit.h" -#include "xfs_log.h" -#include "xfs_inum.h" -#include "xfs_trans.h" +#include "xfs_shared.h" +#include "xfs_format.h" +#include "xfs_log_format.h" +#include "xfs_trans_resv.h" #include "xfs_sb.h" #include "xfs_ag.h" -#include "xfs_dir2.h" -#include "xfs_dmapi.h" #include "xfs_mount.h" #include "xfs_error.h" +#include "xfs_trans.h" +#include "xfs_trans_priv.h" +#include "xfs_log.h" #include "xfs_log_priv.h" -#include "xfs_buf_item.h" -#include "xfs_bmap_btree.h" -#include "xfs_alloc_btree.h" -#include "xfs_ialloc_btree.h" #include "xfs_log_recover.h" -#include "xfs_trans_priv.h" -#include "xfs_dir2_sf.h" -#include "xfs_attr_sf.h" -#include "xfs_dinode.h" #include "xfs_inode.h" -#include "xfs_rw.h" +#include "xfs_trace.h" +#include "xfs_fsops.h" +#include "xfs_cksum.h" - -#define xlog_write_adv_cnt(ptr, len, off, bytes) \ - { (ptr) += (bytes); \ - (len) -= (bytes); \ - (off) += (bytes);} +kmem_zone_t *xfs_log_ticket_zone; /* Local miscellaneous function prototypes */ -STATIC int xlog_bdstrat_cb(struct xfs_buf *); -STATIC int xlog_commit_record(xfs_mount_t *mp, xlog_ticket_t *ticket, - xlog_in_core_t **, xfs_lsn_t *); -STATIC xlog_t * xlog_alloc_log(xfs_mount_t *mp, - xfs_buftarg_t *log_target, - xfs_daddr_t blk_offset, - int num_bblks); -STATIC int xlog_space_left(xlog_t *log, int cycle, int bytes); -STATIC int xlog_sync(xlog_t *log, xlog_in_core_t *iclog); -STATIC void xlog_dealloc_log(xlog_t *log); -STATIC int xlog_write(xfs_mount_t *mp, xfs_log_iovec_t region[], - int nentries, xfs_log_ticket_t tic, - xfs_lsn_t *start_lsn, - xlog_in_core_t **commit_iclog, - uint flags); +STATIC int +xlog_commit_record( + struct xlog *log, + struct xlog_ticket *ticket, + struct xlog_in_core **iclog, + xfs_lsn_t *commitlsnp); + +STATIC struct xlog * +xlog_alloc_log( + struct xfs_mount *mp, + struct xfs_buftarg *log_target, + xfs_daddr_t blk_offset, + int num_bblks); +STATIC int +xlog_space_left( + struct xlog *log, + atomic64_t *head); +STATIC int +xlog_sync( + struct xlog *log, + struct xlog_in_core *iclog); +STATIC void +xlog_dealloc_log( + struct xlog *log); /* local state machine functions */ STATIC void xlog_state_done_syncing(xlog_in_core_t *iclog, int); -STATIC void xlog_state_do_callback(xlog_t *log,int aborted, xlog_in_core_t *iclog); -STATIC int xlog_state_get_iclog_space(xlog_t *log, - int len, - xlog_in_core_t **iclog, - xlog_ticket_t *ticket, - int *continued_write, - int *logoffsetp); -STATIC void xlog_state_put_ticket(xlog_t *log, - xlog_ticket_t *tic); -STATIC int xlog_state_release_iclog(xlog_t *log, - xlog_in_core_t *iclog); -STATIC void xlog_state_switch_iclogs(xlog_t *log, - xlog_in_core_t *iclog, - int eventual_size); -STATIC int xlog_state_sync(xlog_t *log, - xfs_lsn_t lsn, - uint flags, - int *log_flushed); -STATIC int xlog_state_sync_all(xlog_t *log, uint flags, int *log_flushed); -STATIC void xlog_state_want_sync(xlog_t *log, xlog_in_core_t *iclog); - -/* local functions to manipulate grant head */ -STATIC int xlog_grant_log_space(xlog_t *log, - xlog_ticket_t *xtic); -STATIC void xlog_grant_push_ail(xfs_mount_t *mp, - int need_bytes); -STATIC void xlog_regrant_reserve_log_space(xlog_t *log, - xlog_ticket_t *ticket); -STATIC int xlog_regrant_write_log_space(xlog_t *log, - xlog_ticket_t *ticket); -STATIC void xlog_ungrant_log_space(xlog_t *log, - xlog_ticket_t *ticket); - - -/* local ticket functions */ -STATIC void xlog_state_ticket_alloc(xlog_t *log); -STATIC xlog_ticket_t *xlog_ticket_get(xlog_t *log, - int unit_bytes, - int count, - char clientid, - uint flags); -STATIC void xlog_ticket_put(xlog_t *log, xlog_ticket_t *ticket); +STATIC void +xlog_state_do_callback( + struct xlog *log, + int aborted, + struct xlog_in_core *iclog); +STATIC int +xlog_state_get_iclog_space( + struct xlog *log, + int len, + struct xlog_in_core **iclog, + struct xlog_ticket *ticket, + int *continued_write, + int *logoffsetp); +STATIC int +xlog_state_release_iclog( + struct xlog *log, + struct xlog_in_core *iclog); +STATIC void +xlog_state_switch_iclogs( + struct xlog *log, + struct xlog_in_core *iclog, + int eventual_size); +STATIC void +xlog_state_want_sync( + struct xlog *log, + struct xlog_in_core *iclog); + +STATIC void +xlog_grant_push_ail( + struct xlog *log, + int need_bytes); +STATIC void +xlog_regrant_reserve_log_space( + struct xlog *log, + struct xlog_ticket *ticket); +STATIC void +xlog_ungrant_log_space( + struct xlog *log, + struct xlog_ticket *ticket); #if defined(DEBUG) -STATIC void xlog_verify_dest_ptr(xlog_t *log, __psint_t ptr); -STATIC void xlog_verify_grant_head(xlog_t *log, int equals); -STATIC void xlog_verify_iclog(xlog_t *log, xlog_in_core_t *iclog, - int count, boolean_t syncing); -STATIC void xlog_verify_tail_lsn(xlog_t *log, xlog_in_core_t *iclog, - xfs_lsn_t tail_lsn); +STATIC void +xlog_verify_dest_ptr( + struct xlog *log, + char *ptr); +STATIC void +xlog_verify_grant_tail( + struct xlog *log); +STATIC void +xlog_verify_iclog( + struct xlog *log, + struct xlog_in_core *iclog, + int count, + bool syncing); +STATIC void +xlog_verify_tail_lsn( + struct xlog *log, + struct xlog_in_core *iclog, + xfs_lsn_t tail_lsn); #else #define xlog_verify_dest_ptr(a,b) -#define xlog_verify_grant_head(a,b) +#define xlog_verify_grant_tail(a) #define xlog_verify_iclog(a,b,c,d) #define xlog_verify_tail_lsn(a,b,c) #endif -STATIC int xlog_iclogs_empty(xlog_t *log); +STATIC int +xlog_iclogs_empty( + struct xlog *log); -#if defined(XFS_LOG_TRACE) -void -xlog_trace_loggrant(xlog_t *log, xlog_ticket_t *tic, xfs_caddr_t string) +static void +xlog_grant_sub_space( + struct xlog *log, + atomic64_t *head, + int bytes) { - unsigned long cnts; + int64_t head_val = atomic64_read(head); + int64_t new, old; - if (!log->l_grant_trace) { - log->l_grant_trace = ktrace_alloc(2048, KM_NOSLEEP); - if (!log->l_grant_trace) - return; - } - /* ticket counts are 1 byte each */ - cnts = ((unsigned long)tic->t_ocnt) | ((unsigned long)tic->t_cnt) << 8; - - ktrace_enter(log->l_grant_trace, - (void *)tic, - (void *)log->l_reserve_headq, - (void *)log->l_write_headq, - (void *)((unsigned long)log->l_grant_reserve_cycle), - (void *)((unsigned long)log->l_grant_reserve_bytes), - (void *)((unsigned long)log->l_grant_write_cycle), - (void *)((unsigned long)log->l_grant_write_bytes), - (void *)((unsigned long)log->l_curr_cycle), - (void *)((unsigned long)log->l_curr_block), - (void *)((unsigned long)CYCLE_LSN(log->l_tail_lsn)), - (void *)((unsigned long)BLOCK_LSN(log->l_tail_lsn)), - (void *)string, - (void *)((unsigned long)tic->t_trans_type), - (void *)cnts, - (void *)((unsigned long)tic->t_curr_res), - (void *)((unsigned long)tic->t_unit_res)); + do { + int cycle, space; + + xlog_crack_grant_head_val(head_val, &cycle, &space); + + space -= bytes; + if (space < 0) { + space += log->l_logsize; + cycle--; + } + + old = head_val; + new = xlog_assign_grant_head_val(cycle, space); + head_val = atomic64_cmpxchg(head, old, new); + } while (head_val != old); } -void -xlog_trace_iclog(xlog_in_core_t *iclog, uint state) -{ - if (!iclog->ic_trace) - iclog->ic_trace = ktrace_alloc(256, KM_SLEEP); - ktrace_enter(iclog->ic_trace, - (void *)((unsigned long)state), - (void *)((unsigned long)current_pid()), - (void *)NULL, (void *)NULL, (void *)NULL, (void *)NULL, - (void *)NULL, (void *)NULL, (void *)NULL, (void *)NULL, - (void *)NULL, (void *)NULL, (void *)NULL, (void *)NULL, - (void *)NULL, (void *)NULL); +static void +xlog_grant_add_space( + struct xlog *log, + atomic64_t *head, + int bytes) +{ + int64_t head_val = atomic64_read(head); + int64_t new, old; + + do { + int tmp; + int cycle, space; + + xlog_crack_grant_head_val(head_val, &cycle, &space); + + tmp = log->l_logsize - space; + if (tmp > bytes) + space += bytes; + else { + space = bytes - tmp; + cycle++; + } + + old = head_val; + new = xlog_assign_grant_head_val(cycle, space); + head_val = atomic64_cmpxchg(head, old, new); + } while (head_val != old); } -#else -#define xlog_trace_loggrant(log,tic,string) -#define xlog_trace_iclog(iclog,state) -#endif /* XFS_LOG_TRACE */ +STATIC void +xlog_grant_head_init( + struct xlog_grant_head *head) +{ + xlog_assign_grant_head(&head->grant, 1, 0); + INIT_LIST_HEAD(&head->waiters); + spin_lock_init(&head->lock); +} -static void -xlog_ins_ticketq(struct xlog_ticket **qp, struct xlog_ticket *tic) +STATIC void +xlog_grant_head_wake_all( + struct xlog_grant_head *head) { - if (*qp) { - tic->t_next = (*qp); - tic->t_prev = (*qp)->t_prev; - (*qp)->t_prev->t_next = tic; - (*qp)->t_prev = tic; - } else { - tic->t_prev = tic->t_next = tic; - *qp = tic; - } + struct xlog_ticket *tic; - tic->t_flags |= XLOG_TIC_IN_Q; + spin_lock(&head->lock); + list_for_each_entry(tic, &head->waiters, t_queue) + wake_up_process(tic->t_task); + spin_unlock(&head->lock); } -static void -xlog_del_ticketq(struct xlog_ticket **qp, struct xlog_ticket *tic) +static inline int +xlog_ticket_reservation( + struct xlog *log, + struct xlog_grant_head *head, + struct xlog_ticket *tic) { - if (tic == tic->t_next) { - *qp = NULL; + if (head == &log->l_write_head) { + ASSERT(tic->t_flags & XLOG_TIC_PERM_RESERV); + return tic->t_unit_res; } else { - *qp = tic->t_next; - tic->t_next->t_prev = tic->t_prev; - tic->t_prev->t_next = tic->t_next; + if (tic->t_flags & XLOG_TIC_PERM_RESERV) + return tic->t_unit_res * tic->t_cnt; + else + return tic->t_unit_res; } - - tic->t_next = tic->t_prev = NULL; - tic->t_flags &= ~XLOG_TIC_IN_Q; } -static void -xlog_grant_sub_space(struct log *log, int bytes) +STATIC bool +xlog_grant_head_wake( + struct xlog *log, + struct xlog_grant_head *head, + int *free_bytes) { - log->l_grant_write_bytes -= bytes; - if (log->l_grant_write_bytes < 0) { - log->l_grant_write_bytes += log->l_logsize; - log->l_grant_write_cycle--; - } + struct xlog_ticket *tic; + int need_bytes; - log->l_grant_reserve_bytes -= bytes; - if ((log)->l_grant_reserve_bytes < 0) { - log->l_grant_reserve_bytes += log->l_logsize; - log->l_grant_reserve_cycle--; + list_for_each_entry(tic, &head->waiters, t_queue) { + need_bytes = xlog_ticket_reservation(log, head, tic); + if (*free_bytes < need_bytes) + return false; + + *free_bytes -= need_bytes; + trace_xfs_log_grant_wake_up(log, tic); + wake_up_process(tic->t_task); } + return true; } -static void -xlog_grant_add_space_write(struct log *log, int bytes) +STATIC int +xlog_grant_head_wait( + struct xlog *log, + struct xlog_grant_head *head, + struct xlog_ticket *tic, + int need_bytes) __releases(&head->lock) + __acquires(&head->lock) { - log->l_grant_write_bytes += bytes; - if (log->l_grant_write_bytes > log->l_logsize) { - log->l_grant_write_bytes -= log->l_logsize; - log->l_grant_write_cycle++; - } + list_add_tail(&tic->t_queue, &head->waiters); + + do { + if (XLOG_FORCED_SHUTDOWN(log)) + goto shutdown; + xlog_grant_push_ail(log, need_bytes); + + __set_current_state(TASK_UNINTERRUPTIBLE); + spin_unlock(&head->lock); + + XFS_STATS_INC(xs_sleep_logspace); + + trace_xfs_log_grant_sleep(log, tic); + schedule(); + trace_xfs_log_grant_wake(log, tic); + + spin_lock(&head->lock); + if (XLOG_FORCED_SHUTDOWN(log)) + goto shutdown; + } while (xlog_space_left(log, &head->grant) < need_bytes); + + list_del_init(&tic->t_queue); + return 0; +shutdown: + list_del_init(&tic->t_queue); + return XFS_ERROR(EIO); } -static void -xlog_grant_add_space_reserve(struct log *log, int bytes) +/* + * Atomically get the log space required for a log ticket. + * + * Once a ticket gets put onto head->waiters, it will only return after the + * needed reservation is satisfied. + * + * This function is structured so that it has a lock free fast path. This is + * necessary because every new transaction reservation will come through this + * path. Hence any lock will be globally hot if we take it unconditionally on + * every pass. + * + * As tickets are only ever moved on and off head->waiters under head->lock, we + * only need to take that lock if we are going to add the ticket to the queue + * and sleep. We can avoid taking the lock if the ticket was never added to + * head->waiters because the t_queue list head will be empty and we hold the + * only reference to it so it can safely be checked unlocked. + */ +STATIC int +xlog_grant_head_check( + struct xlog *log, + struct xlog_grant_head *head, + struct xlog_ticket *tic, + int *need_bytes) { - log->l_grant_reserve_bytes += bytes; - if (log->l_grant_reserve_bytes > log->l_logsize) { - log->l_grant_reserve_bytes -= log->l_logsize; - log->l_grant_reserve_cycle++; + int free_bytes; + int error = 0; + + ASSERT(!(log->l_flags & XLOG_ACTIVE_RECOVERY)); + + /* + * If there are other waiters on the queue then give them a chance at + * logspace before us. Wake up the first waiters, if we do not wake + * up all the waiters then go to sleep waiting for more free space, + * otherwise try to get some space for this transaction. + */ + *need_bytes = xlog_ticket_reservation(log, head, tic); + free_bytes = xlog_space_left(log, &head->grant); + if (!list_empty_careful(&head->waiters)) { + spin_lock(&head->lock); + if (!xlog_grant_head_wake(log, head, &free_bytes) || + free_bytes < *need_bytes) { + error = xlog_grant_head_wait(log, head, tic, + *need_bytes); + } + spin_unlock(&head->lock); + } else if (free_bytes < *need_bytes) { + spin_lock(&head->lock); + error = xlog_grant_head_wait(log, head, tic, *need_bytes); + spin_unlock(&head->lock); } -} -static inline void -xlog_grant_add_space(struct log *log, int bytes) -{ - xlog_grant_add_space_write(log, bytes); - xlog_grant_add_space_reserve(log, bytes); + return error; } static void @@ -277,6 +365,129 @@ xlog_tic_add_region(xlog_ticket_t *tic, uint len, uint type) } /* + * Replenish the byte reservation required by moving the grant write head. + */ +int +xfs_log_regrant( + struct xfs_mount *mp, + struct xlog_ticket *tic) +{ + struct xlog *log = mp->m_log; + int need_bytes; + int error = 0; + + if (XLOG_FORCED_SHUTDOWN(log)) + return XFS_ERROR(EIO); + + XFS_STATS_INC(xs_try_logspace); + + /* + * This is a new transaction on the ticket, so we need to change the + * transaction ID so that the next transaction has a different TID in + * the log. Just add one to the existing tid so that we can see chains + * of rolling transactions in the log easily. + */ + tic->t_tid++; + + xlog_grant_push_ail(log, tic->t_unit_res); + + tic->t_curr_res = tic->t_unit_res; + xlog_tic_reset_res(tic); + + if (tic->t_cnt > 0) + return 0; + + trace_xfs_log_regrant(log, tic); + + error = xlog_grant_head_check(log, &log->l_write_head, tic, + &need_bytes); + if (error) + goto out_error; + + xlog_grant_add_space(log, &log->l_write_head.grant, need_bytes); + trace_xfs_log_regrant_exit(log, tic); + xlog_verify_grant_tail(log); + return 0; + +out_error: + /* + * If we are failing, make sure the ticket doesn't have any current + * reservations. We don't want to add this back when the ticket/ + * transaction gets cancelled. + */ + tic->t_curr_res = 0; + tic->t_cnt = 0; /* ungrant will give back unit_res * t_cnt. */ + return error; +} + +/* + * Reserve log space and return a ticket corresponding the reservation. + * + * Each reservation is going to reserve extra space for a log record header. + * When writes happen to the on-disk log, we don't subtract the length of the + * log record header from any reservation. By wasting space in each + * reservation, we prevent over allocation problems. + */ +int +xfs_log_reserve( + struct xfs_mount *mp, + int unit_bytes, + int cnt, + struct xlog_ticket **ticp, + __uint8_t client, + bool permanent, + uint t_type) +{ + struct xlog *log = mp->m_log; + struct xlog_ticket *tic; + int need_bytes; + int error = 0; + + ASSERT(client == XFS_TRANSACTION || client == XFS_LOG); + + if (XLOG_FORCED_SHUTDOWN(log)) + return XFS_ERROR(EIO); + + XFS_STATS_INC(xs_try_logspace); + + ASSERT(*ticp == NULL); + tic = xlog_ticket_alloc(log, unit_bytes, cnt, client, permanent, + KM_SLEEP | KM_MAYFAIL); + if (!tic) + return XFS_ERROR(ENOMEM); + + tic->t_trans_type = t_type; + *ticp = tic; + + xlog_grant_push_ail(log, tic->t_cnt ? tic->t_unit_res * tic->t_cnt + : tic->t_unit_res); + + trace_xfs_log_reserve(log, tic); + + error = xlog_grant_head_check(log, &log->l_reserve_head, tic, + &need_bytes); + if (error) + goto out_error; + + xlog_grant_add_space(log, &log->l_reserve_head.grant, need_bytes); + xlog_grant_add_space(log, &log->l_write_head.grant, need_bytes); + trace_xfs_log_reserve_exit(log, tic); + xlog_verify_grant_tail(log); + return 0; + +out_error: + /* + * If we are failing, make sure the ticket doesn't have any current + * reservations. We don't want to add this back when the ticket/ + * transaction gets cancelled. + */ + tic->t_curr_res = 0; + tic->t_cnt = 0; /* ungrant will give back unit_res * t_cnt. */ + return error; +} + + +/* * NOTES: * * 1. currblock field gets updated at startup and after in-core logs @@ -298,14 +509,14 @@ xlog_tic_add_region(xlog_ticket_t *tic, uint len, uint type) * out when the next write occurs. */ xfs_lsn_t -xfs_log_done(xfs_mount_t *mp, - xfs_log_ticket_t xtic, - void **iclog, - uint flags) +xfs_log_done( + struct xfs_mount *mp, + struct xlog_ticket *ticket, + struct xlog_in_core **iclog, + uint flags) { - xlog_t *log = mp->m_log; - xlog_ticket_t *ticket = (xfs_log_ticket_t) xtic; - xfs_lsn_t lsn = 0; + struct xlog *log = mp->m_log; + xfs_lsn_t lsn = 0; if (XLOG_FORCED_SHUTDOWN(log) || /* @@ -313,8 +524,7 @@ xfs_log_done(xfs_mount_t *mp, * If we get an error, just continue and give back the log ticket. */ (((ticket->t_flags & XLOG_TIC_INITED) == 0) && - (xlog_commit_record(mp, ticket, - (xlog_in_core_t **)iclog, &lsn)))) { + (xlog_commit_record(log, ticket, iclog, &lsn)))) { lsn = (xfs_lsn_t) -1; if (ticket->t_flags & XLOG_TIC_PERM_RESERV) { flags |= XFS_LOG_REL_PERM_RESERV; @@ -324,67 +534,27 @@ xfs_log_done(xfs_mount_t *mp, if ((ticket->t_flags & XLOG_TIC_PERM_RESERV) == 0 || (flags & XFS_LOG_REL_PERM_RESERV)) { + trace_xfs_log_done_nonperm(log, ticket); + /* * Release ticket if not permanent reservation or a specific * request has been made to release a permanent reservation. */ - xlog_trace_loggrant(log, ticket, "xfs_log_done: (non-permanent)"); xlog_ungrant_log_space(log, ticket); - xlog_state_put_ticket(log, ticket); + xfs_log_ticket_put(ticket); } else { - xlog_trace_loggrant(log, ticket, "xfs_log_done: (permanent)"); - xlog_regrant_reserve_log_space(log, ticket); - } + trace_xfs_log_done_perm(log, ticket); - /* If this ticket was a permanent reservation and we aren't - * trying to release it, reset the inited flags; so next time - * we write, a start record will be written out. - */ - if ((ticket->t_flags & XLOG_TIC_PERM_RESERV) && - (flags & XFS_LOG_REL_PERM_RESERV) == 0) + xlog_regrant_reserve_log_space(log, ticket); + /* If this ticket was a permanent reservation and we aren't + * trying to release it, reset the inited flags; so next time + * we write, a start record will be written out. + */ ticket->t_flags |= XLOG_TIC_INITED; + } return lsn; -} /* xfs_log_done */ - - -/* - * Force the in-core log to disk. If flags == XFS_LOG_SYNC, - * the force is done synchronously. - * - * Asynchronous forces are implemented by setting the WANT_SYNC - * bit in the appropriate in-core log and then returning. - * - * Synchronous forces are implemented with a semaphore. All callers - * to force a given lsn to disk will wait on a semaphore attached to the - * specific in-core log. When given in-core log finally completes its - * write to disk, that thread will wake up all threads waiting on the - * semaphore. - */ -int -_xfs_log_force( - xfs_mount_t *mp, - xfs_lsn_t lsn, - uint flags, - int *log_flushed) -{ - xlog_t *log = mp->m_log; - int dummy; - - if (!log_flushed) - log_flushed = &dummy; - - ASSERT(flags & XFS_LOG_FORCE); - - XFS_STATS_INC(xs_log_force); - - if (log->l_flags & XLOG_IO_ERROR) - return XFS_ERROR(EIO); - if (lsn == 0) - return xlog_state_sync_all(log, flags, log_flushed); - else - return xlog_state_sync(log, lsn, flags, log_flushed); -} /* xfs_log_force */ +} /* * Attaches a new iclog I/O completion callback routine during @@ -393,16 +563,14 @@ _xfs_log_force( * executing the callback at an appropriate time. */ int -xfs_log_notify(xfs_mount_t *mp, /* mount of partition */ - void *iclog_hndl, /* iclog to hang callback off */ - xfs_log_callback_t *cb) +xfs_log_notify( + struct xfs_mount *mp, + struct xlog_in_core *iclog, + xfs_log_callback_t *cb) { - xlog_t *log = mp->m_log; - xlog_in_core_t *iclog = (xlog_in_core_t *)iclog_hndl; int abortflg; - cb->cb_next = NULL; - spin_lock(&log->l_icloglock); + spin_lock(&iclog->ic_callback_lock); abortflg = (iclog->ic_state & XLOG_STATE_IOERROR); if (!abortflg) { ASSERT_ALWAYS((iclog->ic_state == XLOG_STATE_ACTIVE) || @@ -411,18 +579,16 @@ xfs_log_notify(xfs_mount_t *mp, /* mount of partition */ *(iclog->ic_callback_tail) = cb; iclog->ic_callback_tail = &(cb->cb_next); } - spin_unlock(&log->l_icloglock); + spin_unlock(&iclog->ic_callback_lock); return abortflg; -} /* xfs_log_notify */ +} int -xfs_log_release_iclog(xfs_mount_t *mp, - void *iclog_hndl) +xfs_log_release_iclog( + struct xfs_mount *mp, + struct xlog_in_core *iclog) { - xlog_t *log = mp->m_log; - xlog_in_core_t *iclog = (xlog_in_core_t *)iclog_hndl; - - if (xlog_state_release_iclog(log, iclog)) { + if (xlog_state_release_iclog(mp->m_log, iclog)) { xfs_force_shutdown(mp, SHUTDOWN_LOG_IO_ERROR); return EIO; } @@ -431,63 +597,6 @@ xfs_log_release_iclog(xfs_mount_t *mp, } /* - * 1. Reserve an amount of on-disk log space and return a ticket corresponding - * to the reservation. - * 2. Potentially, push buffers at tail of log to disk. - * - * Each reservation is going to reserve extra space for a log record header. - * When writes happen to the on-disk log, we don't subtract the length of the - * log record header from any reservation. By wasting space in each - * reservation, we prevent over allocation problems. - */ -int -xfs_log_reserve(xfs_mount_t *mp, - int unit_bytes, - int cnt, - xfs_log_ticket_t *ticket, - __uint8_t client, - uint flags, - uint t_type) -{ - xlog_t *log = mp->m_log; - xlog_ticket_t *internal_ticket; - int retval = 0; - - ASSERT(client == XFS_TRANSACTION || client == XFS_LOG); - ASSERT((flags & XFS_LOG_NOSLEEP) == 0); - - if (XLOG_FORCED_SHUTDOWN(log)) - return XFS_ERROR(EIO); - - XFS_STATS_INC(xs_try_logspace); - - if (*ticket != NULL) { - ASSERT(flags & XFS_LOG_PERM_RESERV); - internal_ticket = (xlog_ticket_t *)*ticket; - xlog_trace_loggrant(log, internal_ticket, "xfs_log_reserve: existing ticket (permanent trans)"); - xlog_grant_push_ail(mp, internal_ticket->t_unit_res); - retval = xlog_regrant_write_log_space(log, internal_ticket); - } else { - /* may sleep if need to allocate more tickets */ - internal_ticket = xlog_ticket_get(log, unit_bytes, cnt, - client, flags); - internal_ticket->t_trans_type = t_type; - *ticket = internal_ticket; - xlog_trace_loggrant(log, internal_ticket, - (internal_ticket->t_flags & XLOG_TIC_PERM_RESERV) ? - "xfs_log_reserve: create new ticket (permanent trans)" : - "xfs_log_reserve: create new ticket"); - xlog_grant_push_ail(mp, - (internal_ticket->t_unit_res * - internal_ticket->t_cnt)); - retval = xlog_grant_log_space(log, internal_ticket); - } - - return retval; -} /* xfs_log_reserve */ - - -/* * Mount a log filesystem * * mp - ubiquitous xfs mount point structure @@ -504,28 +613,78 @@ xfs_log_mount( xfs_daddr_t blk_offset, int num_bblks) { - int error; + int error = 0; + int min_logfsbs; - if (!(mp->m_flags & XFS_MOUNT_NORECOVERY)) - cmn_err(CE_NOTE, "XFS mounting filesystem %s", mp->m_fsname); - else { - cmn_err(CE_NOTE, - "!Mounting filesystem \"%s\" in no-recovery mode. Filesystem will be inconsistent.", - mp->m_fsname); + if (!(mp->m_flags & XFS_MOUNT_NORECOVERY)) { + xfs_notice(mp, "Mounting V%d Filesystem", + XFS_SB_VERSION_NUM(&mp->m_sb)); + } else { + xfs_notice(mp, +"Mounting V%d filesystem in no-recovery mode. Filesystem will be inconsistent.", + XFS_SB_VERSION_NUM(&mp->m_sb)); ASSERT(mp->m_flags & XFS_MOUNT_RDONLY); } mp->m_log = xlog_alloc_log(mp, log_target, blk_offset, num_bblks); + if (IS_ERR(mp->m_log)) { + error = -PTR_ERR(mp->m_log); + goto out; + } + + /* + * Validate the given log space and drop a critical message via syslog + * if the log size is too small that would lead to some unexpected + * situations in transaction log space reservation stage. + * + * Note: we can't just reject the mount if the validation fails. This + * would mean that people would have to downgrade their kernel just to + * remedy the situation as there is no way to grow the log (short of + * black magic surgery with xfs_db). + * + * We can, however, reject mounts for CRC format filesystems, as the + * mkfs binary being used to make the filesystem should never create a + * filesystem with a log that is too small. + */ + min_logfsbs = xfs_log_calc_minimum_size(mp); + + if (mp->m_sb.sb_logblocks < min_logfsbs) { + xfs_warn(mp, + "Log size %d blocks too small, minimum size is %d blocks", + mp->m_sb.sb_logblocks, min_logfsbs); + error = EINVAL; + } else if (mp->m_sb.sb_logblocks > XFS_MAX_LOG_BLOCKS) { + xfs_warn(mp, + "Log size %d blocks too large, maximum size is %lld blocks", + mp->m_sb.sb_logblocks, XFS_MAX_LOG_BLOCKS); + error = EINVAL; + } else if (XFS_FSB_TO_B(mp, mp->m_sb.sb_logblocks) > XFS_MAX_LOG_BYTES) { + xfs_warn(mp, + "log size %lld bytes too large, maximum size is %lld bytes", + XFS_FSB_TO_B(mp, mp->m_sb.sb_logblocks), + XFS_MAX_LOG_BYTES); + error = EINVAL; + } + if (error) { + if (xfs_sb_version_hascrc(&mp->m_sb)) { + xfs_crit(mp, "AAIEEE! Log failed size checks. Abort!"); + ASSERT(0); + goto out_free_log; + } + xfs_crit(mp, +"Log size out of supported range. Continuing onwards, but if log hangs are\n" +"experienced then please report this message in the bug report."); + } /* * Initialize the AIL now we have a log. */ - spin_lock_init(&mp->m_ail_lock); error = xfs_trans_ail_init(mp); if (error) { - cmn_err(CE_WARN, "XFS: AIL initialisation failed: error %d", error); - goto error; + xfs_warn(mp, "AIL initialisation failed: error %d", error); + goto out_free_log; } + mp->m_log->l_ailp = mp->m_ail; /* * skip log recovery on a norecovery mount. pretend it all @@ -542,54 +701,56 @@ xfs_log_mount( if (readonly) mp->m_flags |= XFS_MOUNT_RDONLY; if (error) { - cmn_err(CE_WARN, "XFS: log mount/recovery failed: error %d", error); - goto error; + xfs_warn(mp, "log mount/recovery failed: error %d", + error); + goto out_destroy_ail; } } /* Normal transactions can now occur */ mp->m_log->l_flags &= ~XLOG_ACTIVE_RECOVERY; - /* End mounting message in xfs_log_mount_finish */ + /* + * Now the log has been fully initialised and we know were our + * space grant counters are, we can initialise the permanent ticket + * needed for delayed logging to work. + */ + xlog_cil_init_post_recovery(mp->m_log); + return 0; -error: - xfs_log_unmount_dealloc(mp); + +out_destroy_ail: + xfs_trans_ail_destroy(mp); +out_free_log: + xlog_dealloc_log(mp->m_log); +out: return error; -} /* xfs_log_mount */ +} /* - * Finish the recovery of the file system. This is separate from - * the xfs_log_mount() call, because it depends on the code in - * xfs_mountfs() to read in the root and real-time bitmap inodes - * between calling xfs_log_mount() and here. + * Finish the recovery of the file system. This is separate from the + * xfs_log_mount() call, because it depends on the code in xfs_mountfs() to read + * in the root and real-time bitmap inodes between calling xfs_log_mount() and + * here. * - * mp - ubiquitous xfs mount point structure + * If we finish recovery successfully, start the background log work. If we are + * not doing recovery, then we have a RO filesystem and we don't need to start + * it. */ int -xfs_log_mount_finish(xfs_mount_t *mp, int mfsi_flags) +xfs_log_mount_finish(xfs_mount_t *mp) { - int error; + int error = 0; - if (!(mp->m_flags & XFS_MOUNT_NORECOVERY)) - error = xlog_recover_finish(mp->m_log, mfsi_flags); - else { - error = 0; + if (!(mp->m_flags & XFS_MOUNT_NORECOVERY)) { + error = xlog_recover_finish(mp->m_log); + if (!error) + xfs_log_work_queue(mp); + } else { ASSERT(mp->m_flags & XFS_MOUNT_RDONLY); } - return error; -} -/* - * Unmount processing for the log. - */ -int -xfs_log_unmount(xfs_mount_t *mp) -{ - int error; - - error = xfs_log_unmount_write(mp); - xfs_log_unmount_dealloc(mp); return error; } @@ -605,30 +766,22 @@ xfs_log_unmount(xfs_mount_t *mp) * Unmount record used to have a string "Unmount filesystem--" in the * data section where the "Un" was really a magic number (XLOG_UNMOUNT_TYPE). * We just write the magic number now since that particular field isn't - * currently architecture converted and "nUmount" is a bit foo. + * currently architecture converted and "Unmount" is a bit foo. * As far as I know, there weren't any dependencies on the old behaviour. */ int xfs_log_unmount_write(xfs_mount_t *mp) { - xlog_t *log = mp->m_log; + struct xlog *log = mp->m_log; xlog_in_core_t *iclog; #ifdef DEBUG xlog_in_core_t *first_iclog; #endif - xfs_log_iovec_t reg[1]; - xfs_log_ticket_t tic = NULL; + xlog_ticket_t *tic = NULL; xfs_lsn_t lsn; int error; - /* the data section must be 32 bit size aligned */ - struct { - __uint16_t magic; - __uint16_t pad1; - __uint32_t pad2; /* may as well make it 64 bits */ - } magic = { XLOG_UNMOUNT_TYPE, 0, 0 }; - /* * Don't write out unmount record on read-only mounts. * Or, if we are doing a forced umount (typically because of IO errors). @@ -636,7 +789,8 @@ xfs_log_unmount_write(xfs_mount_t *mp) if (mp->m_flags & XFS_MOUNT_RDONLY) return 0; - xfs_log_force(mp, 0, XFS_LOG_FORCE|XFS_LOG_SYNC); + error = _xfs_log_force(mp, XFS_LOG_SYNC, NULL); + ASSERT(error || !(XLOG_FORCED_SHUTDOWN(log))); #ifdef DEBUG first_iclog = iclog = log->l_iclog; @@ -649,16 +803,31 @@ xfs_log_unmount_write(xfs_mount_t *mp) } while (iclog != first_iclog); #endif if (! (XLOG_FORCED_SHUTDOWN(log))) { - reg[0].i_addr = (void*)&magic; - reg[0].i_len = sizeof(magic); - XLOG_VEC_SET_TYPE(®[0], XLOG_REG_TYPE_UNMOUNT); - error = xfs_log_reserve(mp, 600, 1, &tic, XFS_LOG, 0, XLOG_UNMOUNT_REC_TYPE); if (!error) { - /* remove inited flag */ - ((xlog_ticket_t *)tic)->t_flags = 0; - error = xlog_write(mp, reg, 1, tic, &lsn, + /* the data section must be 32 bit size aligned */ + struct { + __uint16_t magic; + __uint16_t pad1; + __uint32_t pad2; /* may as well make it 64 bits */ + } magic = { + .magic = XLOG_UNMOUNT_TYPE, + }; + struct xfs_log_iovec reg = { + .i_addr = &magic, + .i_len = sizeof(magic), + .i_type = XLOG_REG_TYPE_UNMOUNT, + }; + struct xfs_log_vec vec = { + .lv_niovecs = 1, + .lv_iovecp = ®, + }; + + /* remove inited flag, and account for space used */ + tic->t_flags = 0; + tic->t_curr_res -= sizeof(magic); + error = xlog_write(log, &vec, tic, &lsn, NULL, XLOG_UNMOUNT_TRANS); /* * At this point, we're umounting anyway, @@ -667,25 +836,23 @@ xfs_log_unmount_write(xfs_mount_t *mp) */ } - if (error) { - xfs_fs_cmn_err(CE_ALERT, mp, - "xfs_log_unmount: unmount record failed"); - } + if (error) + xfs_alert(mp, "%s: unmount record failed", __func__); spin_lock(&log->l_icloglock); iclog = log->l_iclog; - iclog->ic_refcnt++; - spin_unlock(&log->l_icloglock); + atomic_inc(&iclog->ic_refcnt); xlog_state_want_sync(log, iclog); - (void) xlog_state_release_iclog(log, iclog); + spin_unlock(&log->l_icloglock); + error = xlog_state_release_iclog(log, iclog); spin_lock(&log->l_icloglock); if (!(iclog->ic_state == XLOG_STATE_ACTIVE || iclog->ic_state == XLOG_STATE_DIRTY)) { if (!XLOG_FORCED_SHUTDOWN(log)) { - sv_wait(&iclog->ic_forcesema, PMEM, - &log->l_icloglock, s); + xlog_wait(&iclog->ic_force_wait, + &log->l_icloglock); } else { spin_unlock(&log->l_icloglock); } @@ -693,9 +860,9 @@ xfs_log_unmount_write(xfs_mount_t *mp) spin_unlock(&log->l_icloglock); } if (tic) { - xlog_trace_loggrant(log, tic, "unmount rec"); + trace_xfs_log_umount_write(log, tic); xlog_ungrant_log_space(log, tic); - xlog_state_put_ticket(log, tic); + xfs_log_ticket_put(tic); } } else { /* @@ -713,11 +880,11 @@ xfs_log_unmount_write(xfs_mount_t *mp) */ spin_lock(&log->l_icloglock); iclog = log->l_iclog; - iclog->ic_refcnt++; - spin_unlock(&log->l_icloglock); + atomic_inc(&iclog->ic_refcnt); xlog_state_want_sync(log, iclog); - (void) xlog_state_release_iclog(log, iclog); + spin_unlock(&log->l_icloglock); + error = xlog_state_release_iclog(log, iclog); spin_lock(&log->l_icloglock); @@ -725,191 +892,209 @@ xfs_log_unmount_write(xfs_mount_t *mp) || iclog->ic_state == XLOG_STATE_DIRTY || iclog->ic_state == XLOG_STATE_IOERROR) ) { - sv_wait(&iclog->ic_forcesema, PMEM, - &log->l_icloglock, s); + xlog_wait(&iclog->ic_force_wait, + &log->l_icloglock); } else { spin_unlock(&log->l_icloglock); } } - return 0; + return error; } /* xfs_log_unmount_write */ /* - * Deallocate log structures for unmount/relocation. + * Empty the log for unmount/freeze. * - * We need to stop the aild from running before we destroy - * and deallocate the log as the aild references the log. + * To do this, we first need to shut down the background log work so it is not + * trying to cover the log as we clean up. We then need to unpin all objects in + * the log so we can then flush them out. Once they have completed their IO and + * run the callbacks removing themselves from the AIL, we can write the unmount + * record. */ void -xfs_log_unmount_dealloc(xfs_mount_t *mp) +xfs_log_quiesce( + struct xfs_mount *mp) { - xfs_trans_ail_destroy(mp); - xlog_dealloc_log(mp->m_log); + cancel_delayed_work_sync(&mp->m_log->l_work); + xfs_log_force(mp, XFS_LOG_SYNC); + + /* + * The superblock buffer is uncached and while xfs_ail_push_all_sync() + * will push it, xfs_wait_buftarg() will not wait for it. Further, + * xfs_buf_iowait() cannot be used because it was pushed with the + * XBF_ASYNC flag set, so we need to use a lock/unlock pair to wait for + * the IO to complete. + */ + xfs_ail_push_all_sync(mp->m_ail); + xfs_wait_buftarg(mp->m_ddev_targp); + xfs_buf_lock(mp->m_sb_bp); + xfs_buf_unlock(mp->m_sb_bp); + + xfs_log_unmount_write(mp); } /* - * Write region vectors to log. The write happens using the space reservation - * of the ticket (tic). It is not a requirement that all writes for a given - * transaction occur with one call to xfs_log_write(). + * Shut down and release the AIL and Log. + * + * During unmount, we need to ensure we flush all the dirty metadata objects + * from the AIL so that the log is empty before we write the unmount record to + * the log. Once this is done, we can tear down the AIL and the log. */ -int -xfs_log_write(xfs_mount_t * mp, - xfs_log_iovec_t reg[], - int nentries, - xfs_log_ticket_t tic, - xfs_lsn_t *start_lsn) +void +xfs_log_unmount( + struct xfs_mount *mp) { - int error; - xlog_t *log = mp->m_log; - - if (XLOG_FORCED_SHUTDOWN(log)) - return XFS_ERROR(EIO); + xfs_log_quiesce(mp); - if ((error = xlog_write(mp, reg, nentries, tic, start_lsn, NULL, 0))) { - xfs_force_shutdown(mp, SHUTDOWN_LOG_IO_ERROR); - } - return error; -} /* xfs_log_write */ + xfs_trans_ail_destroy(mp); + xlog_dealloc_log(mp->m_log); +} +void +xfs_log_item_init( + struct xfs_mount *mp, + struct xfs_log_item *item, + int type, + const struct xfs_item_ops *ops) +{ + item->li_mountp = mp; + item->li_ailp = mp->m_ail; + item->li_type = type; + item->li_ops = ops; + item->li_lv = NULL; + + INIT_LIST_HEAD(&item->li_ail); + INIT_LIST_HEAD(&item->li_cil); +} +/* + * Wake up processes waiting for log space after we have moved the log tail. + */ void -xfs_log_move_tail(xfs_mount_t *mp, - xfs_lsn_t tail_lsn) +xfs_log_space_wake( + struct xfs_mount *mp) { - xlog_ticket_t *tic; - xlog_t *log = mp->m_log; - int need_bytes, free_bytes, cycle, bytes; + struct xlog *log = mp->m_log; + int free_bytes; if (XLOG_FORCED_SHUTDOWN(log)) return; - if (tail_lsn == 0) { - /* needed since sync_lsn is 64 bits */ - spin_lock(&log->l_icloglock); - tail_lsn = log->l_last_sync_lsn; - spin_unlock(&log->l_icloglock); - } - - spin_lock(&log->l_grant_lock); + if (!list_empty_careful(&log->l_write_head.waiters)) { + ASSERT(!(log->l_flags & XLOG_ACTIVE_RECOVERY)); - /* Also an invalid lsn. 1 implies that we aren't passing in a valid - * tail_lsn. - */ - if (tail_lsn != 1) { - log->l_tail_lsn = tail_lsn; + spin_lock(&log->l_write_head.lock); + free_bytes = xlog_space_left(log, &log->l_write_head.grant); + xlog_grant_head_wake(log, &log->l_write_head, &free_bytes); + spin_unlock(&log->l_write_head.lock); } - if ((tic = log->l_write_headq)) { -#ifdef DEBUG - if (log->l_flags & XLOG_ACTIVE_RECOVERY) - panic("Recovery problem"); -#endif - cycle = log->l_grant_write_cycle; - bytes = log->l_grant_write_bytes; - free_bytes = xlog_space_left(log, cycle, bytes); - do { - ASSERT(tic->t_flags & XLOG_TIC_PERM_RESERV); + if (!list_empty_careful(&log->l_reserve_head.waiters)) { + ASSERT(!(log->l_flags & XLOG_ACTIVE_RECOVERY)); - if (free_bytes < tic->t_unit_res && tail_lsn != 1) - break; - tail_lsn = 0; - free_bytes -= tic->t_unit_res; - sv_signal(&tic->t_sema); - tic = tic->t_next; - } while (tic != log->l_write_headq); + spin_lock(&log->l_reserve_head.lock); + free_bytes = xlog_space_left(log, &log->l_reserve_head.grant); + xlog_grant_head_wake(log, &log->l_reserve_head, &free_bytes); + spin_unlock(&log->l_reserve_head.lock); } - if ((tic = log->l_reserve_headq)) { -#ifdef DEBUG - if (log->l_flags & XLOG_ACTIVE_RECOVERY) - panic("Recovery problem"); -#endif - cycle = log->l_grant_reserve_cycle; - bytes = log->l_grant_reserve_bytes; - free_bytes = xlog_space_left(log, cycle, bytes); - do { - if (tic->t_flags & XLOG_TIC_PERM_RESERV) - need_bytes = tic->t_unit_res*tic->t_cnt; - else - need_bytes = tic->t_unit_res; - if (free_bytes < need_bytes && tail_lsn != 1) - break; - tail_lsn = 0; - free_bytes -= need_bytes; - sv_signal(&tic->t_sema); - tic = tic->t_next; - } while (tic != log->l_reserve_headq); - } - spin_unlock(&log->l_grant_lock); -} /* xfs_log_move_tail */ +} /* - * Determine if we have a transaction that has gone to disk - * that needs to be covered. Log activity needs to be idle (no AIL and - * nothing in the iclogs). And, we need to be in the right state indicating - * something has gone out. + * Determine if we have a transaction that has gone to disk that needs to be + * covered. To begin the transition to the idle state firstly the log needs to + * be idle. That means the CIL, the AIL and the iclogs needs to be empty before + * we start attempting to cover the log. + * + * Only if we are then in a state where covering is needed, the caller is + * informed that dummy transactions are required to move the log into the idle + * state. + * + * If there are any items in the AIl or CIL, then we do not want to attempt to + * cover the log as we may be in a situation where there isn't log space + * available to run a dummy transaction and this can lead to deadlocks when the + * tail of the log is pinned by an item that is modified in the CIL. Hence + * there's no point in running a dummy transaction at this point because we + * can't start trying to idle the log until both the CIL and AIL are empty. */ int xfs_log_need_covered(xfs_mount_t *mp) { - int needed = 0, gen; - xlog_t *log = mp->m_log; + struct xlog *log = mp->m_log; + int needed = 0; if (!xfs_fs_writable(mp)) return 0; + if (!xlog_cil_empty(log)) + return 0; + spin_lock(&log->l_icloglock); - if (((log->l_covered_state == XLOG_STATE_COVER_NEED) || - (log->l_covered_state == XLOG_STATE_COVER_NEED2)) - && !xfs_trans_first_ail(mp, &gen) - && xlog_iclogs_empty(log)) { + switch (log->l_covered_state) { + case XLOG_STATE_COVER_DONE: + case XLOG_STATE_COVER_DONE2: + case XLOG_STATE_COVER_IDLE: + break; + case XLOG_STATE_COVER_NEED: + case XLOG_STATE_COVER_NEED2: + if (xfs_ail_min_lsn(log->l_ailp)) + break; + if (!xlog_iclogs_empty(log)) + break; + + needed = 1; if (log->l_covered_state == XLOG_STATE_COVER_NEED) log->l_covered_state = XLOG_STATE_COVER_DONE; - else { - ASSERT(log->l_covered_state == XLOG_STATE_COVER_NEED2); + else log->l_covered_state = XLOG_STATE_COVER_DONE2; - } + break; + default: needed = 1; + break; } spin_unlock(&log->l_icloglock); return needed; } -/****************************************************************************** - * - * local routines - * - ****************************************************************************** - */ - -/* xfs_trans_tail_ail returns 0 when there is nothing in the list. - * The log manager must keep track of the last LR which was committed - * to disk. The lsn of this LR will become the new tail_lsn whenever - * xfs_trans_tail_ail returns 0. If we don't do this, we run into - * the situation where stuff could be written into the log but nothing - * was ever in the AIL when asked. Eventually, we panic since the - * tail hits the head. - * +/* * We may be holding the log iclog lock upon entering this routine. */ xfs_lsn_t -xlog_assign_tail_lsn(xfs_mount_t *mp) +xlog_assign_tail_lsn_locked( + struct xfs_mount *mp) { - xfs_lsn_t tail_lsn; - xlog_t *log = mp->m_log; + struct xlog *log = mp->m_log; + struct xfs_log_item *lip; + xfs_lsn_t tail_lsn; - tail_lsn = xfs_trans_tail_ail(mp); - spin_lock(&log->l_grant_lock); - if (tail_lsn != 0) { - log->l_tail_lsn = tail_lsn; - } else { - tail_lsn = log->l_tail_lsn = log->l_last_sync_lsn; - } - spin_unlock(&log->l_grant_lock); + assert_spin_locked(&mp->m_ail->xa_lock); + /* + * To make sure we always have a valid LSN for the log tail we keep + * track of the last LSN which was committed in log->l_last_sync_lsn, + * and use that when the AIL was empty. + */ + lip = xfs_ail_min(mp->m_ail); + if (lip) + tail_lsn = lip->li_lsn; + else + tail_lsn = atomic64_read(&log->l_last_sync_lsn); + trace_xfs_log_assign_tail_lsn(log, tail_lsn); + atomic64_set(&log->l_tail_lsn, tail_lsn); return tail_lsn; -} /* xlog_assign_tail_lsn */ +} +xfs_lsn_t +xlog_assign_tail_lsn( + struct xfs_mount *mp) +{ + xfs_lsn_t tail_lsn; + + spin_lock(&mp->m_ail->xa_lock); + tail_lsn = xlog_assign_tail_lsn_locked(mp); + spin_unlock(&mp->m_ail->xa_lock); + + return tail_lsn; +} /* * Return the space in the log between the tail and the head. The head @@ -926,37 +1111,42 @@ xlog_assign_tail_lsn(xfs_mount_t *mp) * result is that we return the size of the log as the amount of space left. */ STATIC int -xlog_space_left(xlog_t *log, int cycle, int bytes) +xlog_space_left( + struct xlog *log, + atomic64_t *head) { - int free_bytes; - int tail_bytes; - int tail_cycle; - - tail_bytes = BBTOB(BLOCK_LSN(log->l_tail_lsn)); - tail_cycle = CYCLE_LSN(log->l_tail_lsn); - if ((tail_cycle == cycle) && (bytes >= tail_bytes)) { - free_bytes = log->l_logsize - (bytes - tail_bytes); - } else if ((tail_cycle + 1) < cycle) { + int free_bytes; + int tail_bytes; + int tail_cycle; + int head_cycle; + int head_bytes; + + xlog_crack_grant_head(head, &head_cycle, &head_bytes); + xlog_crack_atomic_lsn(&log->l_tail_lsn, &tail_cycle, &tail_bytes); + tail_bytes = BBTOB(tail_bytes); + if (tail_cycle == head_cycle && head_bytes >= tail_bytes) + free_bytes = log->l_logsize - (head_bytes - tail_bytes); + else if (tail_cycle + 1 < head_cycle) return 0; - } else if (tail_cycle < cycle) { - ASSERT(tail_cycle == (cycle - 1)); - free_bytes = tail_bytes - bytes; + else if (tail_cycle < head_cycle) { + ASSERT(tail_cycle == (head_cycle - 1)); + free_bytes = tail_bytes - head_bytes; } else { /* * The reservation head is behind the tail. * In this case we just want to return the size of the * log as the amount of space left. */ - xfs_fs_cmn_err(CE_ALERT, log->l_mp, + xfs_alert(log->l_mp, "xlog_space_left: head behind tail\n" " tail_cycle = %d, tail_bytes = %d\n" " GH cycle = %d, GH bytes = %d", - tail_cycle, tail_bytes, cycle, bytes); + tail_cycle, tail_bytes, head_cycle, head_bytes); ASSERT(0); free_bytes = log->l_logsize; } return free_bytes; -} /* xlog_space_left */ +} /* @@ -968,42 +1158,17 @@ xlog_space_left(xlog_t *log, int cycle, int bytes) void xlog_iodone(xfs_buf_t *bp) { - xlog_in_core_t *iclog; - xlog_t *l; - int aborted; - - iclog = XFS_BUF_FSPRIVATE(bp, xlog_in_core_t *); - ASSERT(XFS_BUF_FSPRIVATE2(bp, unsigned long) == (unsigned long) 2); - XFS_BUF_SET_FSPRIVATE2(bp, (unsigned long)1); - aborted = 0; - - /* - * Some versions of cpp barf on the recursive definition of - * ic_log -> hic_fields.ic_log and expand ic_log twice when - * it is passed through two macros. Workaround broken cpp. - */ - l = iclog->ic_log; - - /* - * If the ordered flag has been removed by a lower - * layer, it means the underlyin device no longer supports - * barrier I/O. Warn loudly and turn off barriers. - */ - if ((l->l_mp->m_flags & XFS_MOUNT_BARRIER) && !XFS_BUF_ORDERED(bp)) { - l->l_mp->m_flags &= ~XFS_MOUNT_BARRIER; - xfs_fs_cmn_err(CE_WARN, l->l_mp, - "xlog_iodone: Barriers are no longer supported" - " by device. Disabling barriers\n"); - xfs_buftrace("XLOG_IODONE BARRIERS OFF", bp); - } + struct xlog_in_core *iclog = bp->b_fspriv; + struct xlog *l = iclog->ic_log; + int aborted = 0; /* * Race to shutdown the filesystem if we see an error. */ - if (XFS_TEST_ERROR((XFS_BUF_GETERROR(bp)), l->l_mp, + if (XFS_TEST_ERROR(bp->b_error, l->l_mp, XFS_ERRTAG_IODONE_IOERR, XFS_RANDOM_IODONE_IOERR)) { - xfs_ioerror_alert("xlog_iodone", l->l_mp, bp, XFS_BUF_ADDR(bp)); - XFS_BUF_STALE(bp); + xfs_buf_ioerror_alert(bp, __func__); + xfs_buf_stale(bp); xfs_force_shutdown(l->l_mp, SHUTDOWN_LOG_IO_ERROR); /* * This flag will be propagated to the trans-committed @@ -1018,58 +1183,29 @@ xlog_iodone(xfs_buf_t *bp) /* log I/O is always issued ASYNC */ ASSERT(XFS_BUF_ISASYNC(bp)); xlog_state_done_syncing(iclog, aborted); + /* - * do not reference the buffer (bp) here as we could race - * with it being freed after writing the unmount record to the - * log. + * drop the buffer lock now that we are done. Nothing references + * the buffer after this, so an unmount waiting on this lock can now + * tear it down safely. As such, it is unsafe to reference the buffer + * (bp) after the unlock as we could race with it being freed. */ - -} /* xlog_iodone */ - -/* - * The bdstrat callback function for log bufs. This gives us a central - * place to trap bufs in case we get hit by a log I/O error and need to - * shutdown. Actually, in practice, even when we didn't get a log error, - * we transition the iclogs to IOERROR state *after* flushing all existing - * iclogs to disk. This is because we don't want anymore new transactions to be - * started or completed afterwards. - */ -STATIC int -xlog_bdstrat_cb(struct xfs_buf *bp) -{ - xlog_in_core_t *iclog; - - iclog = XFS_BUF_FSPRIVATE(bp, xlog_in_core_t *); - - if ((iclog->ic_state & XLOG_STATE_IOERROR) == 0) { - /* note for irix bstrat will need struct bdevsw passed - * Fix the following macro if the code ever is merged - */ - XFS_bdstrat(bp); - return 0; - } - - xfs_buftrace("XLOG__BDSTRAT IOERROR", bp); - XFS_BUF_ERROR(bp, EIO); - XFS_BUF_STALE(bp); - xfs_biodone(bp); - return XFS_ERROR(EIO); - - + xfs_buf_unlock(bp); } /* * Return size of each in-core log record buffer. * - * All machines get 8 x 32KB buffers by default, unless tuned otherwise. + * All machines get 8 x 32kB buffers by default, unless tuned otherwise. * * If the filesystem blocksize is too large, we may need to choose a * larger size since the directory code currently logs entire blocks. */ STATIC void -xlog_get_iclog_buffer_size(xfs_mount_t *mp, - xlog_t *log) +xlog_get_iclog_buffer_size( + struct xfs_mount *mp, + struct xlog *log) { int size; int xhdrs; @@ -1090,9 +1226,9 @@ xlog_get_iclog_buffer_size(xfs_mount_t *mp, size >>= 1; } - if (XFS_SB_VERSION_HASLOGV2(&mp->m_sb)) { - /* # headers = size / 32K - * one header holds cycles from 32K of data + if (xfs_sb_version_haslogv2(&mp->m_sb)) { + /* # headers = size / 32k + * one header holds cycles from 32k of data */ xhdrs = mp->m_logbsize / XLOG_HEADER_CYCLE_SIZE; @@ -1108,7 +1244,7 @@ xlog_get_iclog_buffer_size(xfs_mount_t *mp, goto done; } - /* All machines use 32KB buffers by default. */ + /* All machines use 32kB buffers by default. */ log->l_iclog_size = XLOG_BIG_RECORD_BSIZE; log->l_iclog_size_log = XLOG_BIG_RECORD_BSHIFT; @@ -1116,32 +1252,8 @@ xlog_get_iclog_buffer_size(xfs_mount_t *mp, log->l_iclog_hsize = BBSIZE; log->l_iclog_heads = 1; - /* - * For 16KB, we use 3 32KB buffers. For 32KB block sizes, we use - * 4 32KB buffers. For 64KB block sizes, we use 8 32KB buffers. - */ - if (mp->m_sb.sb_blocksize >= 16*1024) { - log->l_iclog_size = XLOG_BIG_RECORD_BSIZE; - log->l_iclog_size_log = XLOG_BIG_RECORD_BSHIFT; - if (mp->m_logbufs <= 0) { - switch (mp->m_sb.sb_blocksize) { - case 16*1024: /* 16 KB */ - log->l_iclog_bufs = 3; - break; - case 32*1024: /* 32 KB */ - log->l_iclog_bufs = 4; - break; - case 64*1024: /* 64 KB */ - log->l_iclog_bufs = 8; - break; - default: - xlog_panic("XFS: Invalid blocksize"); - break; - } - } - } - -done: /* are we being asked to make the sizes selected above visible? */ +done: + /* are we being asked to make the sizes selected above visible? */ if (mp->m_logbufs == 0) mp->m_logbufs = log->l_iclog_bufs; if (mp->m_logbsize == 0) @@ -1149,26 +1261,66 @@ done: /* are we being asked to make the sizes selected above visible? */ } /* xlog_get_iclog_buffer_size */ +void +xfs_log_work_queue( + struct xfs_mount *mp) +{ + queue_delayed_work(mp->m_log_workqueue, &mp->m_log->l_work, + msecs_to_jiffies(xfs_syncd_centisecs * 10)); +} + +/* + * Every sync period we need to unpin all items in the AIL and push them to + * disk. If there is nothing dirty, then we might need to cover the log to + * indicate that the filesystem is idle. + */ +void +xfs_log_worker( + struct work_struct *work) +{ + struct xlog *log = container_of(to_delayed_work(work), + struct xlog, l_work); + struct xfs_mount *mp = log->l_mp; + + /* dgc: errors ignored - not fatal and nowhere to report them */ + if (xfs_log_need_covered(mp)) + xfs_fs_log_dummy(mp); + else + xfs_log_force(mp, 0); + + /* start pushing all the metadata that is currently dirty */ + xfs_ail_push_all(mp->m_ail); + + /* queue us up again */ + xfs_log_work_queue(mp); +} + /* * This routine initializes some of the log structure for a given mount point. * Its primary purpose is to fill in enough, so recovery can occur. However, * some other stuff may be filled in too. */ -STATIC xlog_t * -xlog_alloc_log(xfs_mount_t *mp, - xfs_buftarg_t *log_target, - xfs_daddr_t blk_offset, - int num_bblks) +STATIC struct xlog * +xlog_alloc_log( + struct xfs_mount *mp, + struct xfs_buftarg *log_target, + xfs_daddr_t blk_offset, + int num_bblks) { - xlog_t *log; + struct xlog *log; xlog_rec_header_t *head; xlog_in_core_t **iclogp; xlog_in_core_t *iclog, *prev_iclog=NULL; xfs_buf_t *bp; int i; - int iclogsize; + int error = ENOMEM; + uint log2_size = 0; - log = (xlog_t *)kmem_zalloc(sizeof(xlog_t), KM_SLEEP); + log = kmem_zalloc(sizeof(struct xlog), KM_MAYFAIL); + if (!log) { + xfs_warn(mp, "Log allocation failed: No memory!"); + goto out; + } log->l_mp = mp; log->l_targ = log_target; @@ -1177,43 +1329,64 @@ xlog_alloc_log(xfs_mount_t *mp, log->l_logBBsize = num_bblks; log->l_covered_state = XLOG_STATE_COVER_IDLE; log->l_flags |= XLOG_ACTIVE_RECOVERY; + INIT_DELAYED_WORK(&log->l_work, xfs_log_worker); log->l_prev_block = -1; - log->l_tail_lsn = xlog_assign_lsn(1, 0); /* log->l_tail_lsn = 0x100000000LL; cycle = 1; current block = 0 */ - log->l_last_sync_lsn = log->l_tail_lsn; + xlog_assign_atomic_lsn(&log->l_tail_lsn, 1, 0); + xlog_assign_atomic_lsn(&log->l_last_sync_lsn, 1, 0); log->l_curr_cycle = 1; /* 0 is bad since this is initial value */ - log->l_grant_reserve_cycle = 1; - log->l_grant_write_cycle = 1; - if (XFS_SB_VERSION_HASSECTOR(&mp->m_sb)) { - log->l_sectbb_log = mp->m_sb.sb_logsectlog - BBSHIFT; - ASSERT(log->l_sectbb_log <= mp->m_sectbb_log); + xlog_grant_head_init(&log->l_reserve_head); + xlog_grant_head_init(&log->l_write_head); + + error = EFSCORRUPTED; + if (xfs_sb_version_hassector(&mp->m_sb)) { + log2_size = mp->m_sb.sb_logsectlog; + if (log2_size < BBSHIFT) { + xfs_warn(mp, "Log sector size too small (0x%x < 0x%x)", + log2_size, BBSHIFT); + goto out_free_log; + } + + log2_size -= BBSHIFT; + if (log2_size > mp->m_sectbb_log) { + xfs_warn(mp, "Log sector size too large (0x%x > 0x%x)", + log2_size, mp->m_sectbb_log); + goto out_free_log; + } + /* for larger sector sizes, must have v2 or external log */ - ASSERT(log->l_sectbb_log == 0 || - log->l_logBBstart == 0 || - XFS_SB_VERSION_HASLOGV2(&mp->m_sb)); - ASSERT(mp->m_sb.sb_logsectlog >= BBSHIFT); + if (log2_size && log->l_logBBstart > 0 && + !xfs_sb_version_haslogv2(&mp->m_sb)) { + xfs_warn(mp, + "log sector size (0x%x) invalid for configuration.", + log2_size); + goto out_free_log; + } } - log->l_sectbb_mask = (1 << log->l_sectbb_log) - 1; + log->l_sectBBsize = 1 << log2_size; xlog_get_iclog_buffer_size(mp, log); - bp = xfs_buf_get_empty(log->l_iclog_size, mp->m_logdev_targp); - XFS_BUF_SET_IODONE_FUNC(bp, xlog_iodone); - XFS_BUF_SET_BDSTRAT_FUNC(bp, xlog_bdstrat_cb); - XFS_BUF_SET_FSPRIVATE2(bp, (unsigned long)1); - ASSERT(XFS_BUF_ISBUSY(bp)); - ASSERT(XFS_BUF_VALUSEMA(bp) <= 0); + error = ENOMEM; + bp = xfs_buf_alloc(mp->m_logdev_targp, 0, BTOBB(log->l_iclog_size), 0); + if (!bp) + goto out_free_log; + + /* + * The iclogbuf buffer locks are held over IO but we are not going to do + * IO yet. Hence unlock the buffer so that the log IO path can grab it + * when appropriately. + */ + ASSERT(xfs_buf_islocked(bp)); + xfs_buf_unlock(bp); + + bp->b_iodone = xlog_iodone; log->l_xbuf = bp; spin_lock_init(&log->l_icloglock); - spin_lock_init(&log->l_grant_lock); - initnsema(&log->l_flushsema, 0, "ic-flush"); - xlog_state_ticket_alloc(log); /* wait until after icloglock inited */ - - /* log record size must be multiple of BBSIZE; see xlog_rec_header_t */ - ASSERT((XFS_BUF_SIZE(bp) & BBMASK) == 0); + init_waitqueue_head(&log->l_flush_wait); iclogp = &log->l_iclog; /* @@ -1223,54 +1396,74 @@ xlog_alloc_log(xfs_mount_t *mp, * with different amounts of memory. See the definition of * xlog_in_core_t in xfs_log_priv.h for details. */ - iclogsize = log->l_iclog_size; ASSERT(log->l_iclog_size >= 4096); for (i=0; i < log->l_iclog_bufs; i++) { - *iclogp = (xlog_in_core_t *) - kmem_zalloc(sizeof(xlog_in_core_t), KM_SLEEP); + *iclogp = kmem_zalloc(sizeof(xlog_in_core_t), KM_MAYFAIL); + if (!*iclogp) + goto out_free_iclog; + iclog = *iclogp; iclog->ic_prev = prev_iclog; prev_iclog = iclog; - bp = xfs_buf_get_noaddr(log->l_iclog_size, mp->m_logdev_targp); - if (!XFS_BUF_CPSEMA(bp)) - ASSERT(0); - XFS_BUF_SET_IODONE_FUNC(bp, xlog_iodone); - XFS_BUF_SET_BDSTRAT_FUNC(bp, xlog_bdstrat_cb); - XFS_BUF_SET_FSPRIVATE2(bp, (unsigned long)1); - iclog->ic_bp = bp; - iclog->hic_data = bp->b_addr; + bp = xfs_buf_get_uncached(mp->m_logdev_targp, + BTOBB(log->l_iclog_size), 0); + if (!bp) + goto out_free_iclog; - log->l_iclog_bak[i] = (xfs_caddr_t)&(iclog->ic_header); + ASSERT(xfs_buf_islocked(bp)); + xfs_buf_unlock(bp); + bp->b_iodone = xlog_iodone; + iclog->ic_bp = bp; + iclog->ic_data = bp->b_addr; +#ifdef DEBUG + log->l_iclog_bak[i] = (xfs_caddr_t)&(iclog->ic_header); +#endif head = &iclog->ic_header; memset(head, 0, sizeof(xlog_rec_header_t)); head->h_magicno = cpu_to_be32(XLOG_HEADER_MAGIC_NUM); head->h_version = cpu_to_be32( - XFS_SB_VERSION_HASLOGV2(&log->l_mp->m_sb) ? 2 : 1); + xfs_sb_version_haslogv2(&log->l_mp->m_sb) ? 2 : 1); head->h_size = cpu_to_be32(log->l_iclog_size); /* new fields */ head->h_fmt = cpu_to_be32(XLOG_FMT); memcpy(&head->h_fs_uuid, &mp->m_sb.sb_uuid, sizeof(uuid_t)); - - iclog->ic_size = XFS_BUF_SIZE(bp) - log->l_iclog_hsize; + iclog->ic_size = BBTOB(bp->b_length) - log->l_iclog_hsize; iclog->ic_state = XLOG_STATE_ACTIVE; iclog->ic_log = log; + atomic_set(&iclog->ic_refcnt, 0); + spin_lock_init(&iclog->ic_callback_lock); iclog->ic_callback_tail = &(iclog->ic_callback); - iclog->ic_datap = (char *)iclog->hic_data + log->l_iclog_hsize; + iclog->ic_datap = (char *)iclog->ic_data + log->l_iclog_hsize; - ASSERT(XFS_BUF_ISBUSY(iclog->ic_bp)); - ASSERT(XFS_BUF_VALUSEMA(iclog->ic_bp) <= 0); - sv_init(&iclog->ic_forcesema, SV_DEFAULT, "iclog-force"); - sv_init(&iclog->ic_writesema, SV_DEFAULT, "iclog-write"); + init_waitqueue_head(&iclog->ic_force_wait); + init_waitqueue_head(&iclog->ic_write_wait); iclogp = &iclog->ic_next; } *iclogp = log->l_iclog; /* complete ring */ log->l_iclog->ic_prev = prev_iclog; /* re-write 1st prev ptr */ + error = xlog_cil_init(log); + if (error) + goto out_free_iclog; return log; + +out_free_iclog: + for (iclog = log->l_iclog; iclog; iclog = prev_iclog) { + prev_iclog = iclog->ic_next; + if (iclog->ic_bp) + xfs_buf_free(iclog->ic_bp); + kmem_free(iclog); + } + spinlock_destroy(&log->l_icloglock); + xfs_buf_free(log->l_xbuf); +out_free_log: + kmem_free(log); +out: + return ERR_PTR(-error); } /* xlog_alloc_log */ @@ -1279,26 +1472,31 @@ xlog_alloc_log(xfs_mount_t *mp, * ticket. Return the lsn of the commit record. */ STATIC int -xlog_commit_record(xfs_mount_t *mp, - xlog_ticket_t *ticket, - xlog_in_core_t **iclog, - xfs_lsn_t *commitlsnp) +xlog_commit_record( + struct xlog *log, + struct xlog_ticket *ticket, + struct xlog_in_core **iclog, + xfs_lsn_t *commitlsnp) { - int error; - xfs_log_iovec_t reg[1]; - - reg[0].i_addr = NULL; - reg[0].i_len = 0; - XLOG_VEC_SET_TYPE(®[0], XLOG_REG_TYPE_COMMIT); + struct xfs_mount *mp = log->l_mp; + int error; + struct xfs_log_iovec reg = { + .i_addr = NULL, + .i_len = 0, + .i_type = XLOG_REG_TYPE_COMMIT, + }; + struct xfs_log_vec vec = { + .lv_niovecs = 1, + .lv_iovecp = ®, + }; ASSERT_ALWAYS(iclog); - if ((error = xlog_write(mp, reg, 1, ticket, commitlsnp, - iclog, XLOG_COMMIT_TRANS))) { + error = xlog_write(log, &vec, ticket, commitlsnp, iclog, + XLOG_COMMIT_TRANS); + if (error) xfs_force_shutdown(mp, SHUTDOWN_LOG_IO_ERROR); - } return error; -} /* xlog_commit_record */ - +} /* * Push on the buffer cache code if we ever use more than 75% of the on-disk @@ -1308,62 +1506,176 @@ xlog_commit_record(xfs_mount_t *mp, * water mark. In this manner, we would be creating a low water mark. */ STATIC void -xlog_grant_push_ail(xfs_mount_t *mp, - int need_bytes) -{ - xlog_t *log = mp->m_log; /* pointer to the log */ - xfs_lsn_t tail_lsn; /* lsn of the log tail */ - xfs_lsn_t threshold_lsn = 0; /* lsn we'd like to be at */ - int free_blocks; /* free blocks left to write to */ - int free_bytes; /* free bytes left to write to */ - int threshold_block; /* block in lsn we'd like to be at */ - int threshold_cycle; /* lsn cycle we'd like to be at */ - int free_threshold; - - ASSERT(BTOBB(need_bytes) < log->l_logBBsize); - - spin_lock(&log->l_grant_lock); - free_bytes = xlog_space_left(log, - log->l_grant_reserve_cycle, - log->l_grant_reserve_bytes); - tail_lsn = log->l_tail_lsn; - free_blocks = BTOBBT(free_bytes); - - /* - * Set the threshold for the minimum number of free blocks in the - * log to the maximum of what the caller needs, one quarter of the - * log, and 256 blocks. - */ - free_threshold = BTOBB(need_bytes); - free_threshold = MAX(free_threshold, (log->l_logBBsize >> 2)); - free_threshold = MAX(free_threshold, 256); - if (free_blocks < free_threshold) { - threshold_block = BLOCK_LSN(tail_lsn) + free_threshold; - threshold_cycle = CYCLE_LSN(tail_lsn); +xlog_grant_push_ail( + struct xlog *log, + int need_bytes) +{ + xfs_lsn_t threshold_lsn = 0; + xfs_lsn_t last_sync_lsn; + int free_blocks; + int free_bytes; + int threshold_block; + int threshold_cycle; + int free_threshold; + + ASSERT(BTOBB(need_bytes) < log->l_logBBsize); + + free_bytes = xlog_space_left(log, &log->l_reserve_head.grant); + free_blocks = BTOBBT(free_bytes); + + /* + * Set the threshold for the minimum number of free blocks in the + * log to the maximum of what the caller needs, one quarter of the + * log, and 256 blocks. + */ + free_threshold = BTOBB(need_bytes); + free_threshold = MAX(free_threshold, (log->l_logBBsize >> 2)); + free_threshold = MAX(free_threshold, 256); + if (free_blocks >= free_threshold) + return; + + xlog_crack_atomic_lsn(&log->l_tail_lsn, &threshold_cycle, + &threshold_block); + threshold_block += free_threshold; if (threshold_block >= log->l_logBBsize) { - threshold_block -= log->l_logBBsize; - threshold_cycle += 1; + threshold_block -= log->l_logBBsize; + threshold_cycle += 1; } - threshold_lsn = xlog_assign_lsn(threshold_cycle, threshold_block); + threshold_lsn = xlog_assign_lsn(threshold_cycle, + threshold_block); + /* + * Don't pass in an lsn greater than the lsn of the last + * log record known to be on disk. Use a snapshot of the last sync lsn + * so that it doesn't change between the compare and the set. + */ + last_sync_lsn = atomic64_read(&log->l_last_sync_lsn); + if (XFS_LSN_CMP(threshold_lsn, last_sync_lsn) > 0) + threshold_lsn = last_sync_lsn; - /* Don't pass in an lsn greater than the lsn of the last - * log record known to be on disk. + /* + * Get the transaction layer to kick the dirty buffers out to + * disk asynchronously. No point in trying to do this if + * the filesystem is shutting down. */ - if (XFS_LSN_CMP(threshold_lsn, log->l_last_sync_lsn) > 0) - threshold_lsn = log->l_last_sync_lsn; - } - spin_unlock(&log->l_grant_lock); + if (!XLOG_FORCED_SHUTDOWN(log)) + xfs_ail_push(log->l_ailp, threshold_lsn); +} + +/* + * Stamp cycle number in every block + */ +STATIC void +xlog_pack_data( + struct xlog *log, + struct xlog_in_core *iclog, + int roundoff) +{ + int i, j, k; + int size = iclog->ic_offset + roundoff; + __be32 cycle_lsn; + xfs_caddr_t dp; + + cycle_lsn = CYCLE_LSN_DISK(iclog->ic_header.h_lsn); + + dp = iclog->ic_datap; + for (i = 0; i < BTOBB(size); i++) { + if (i >= (XLOG_HEADER_CYCLE_SIZE / BBSIZE)) + break; + iclog->ic_header.h_cycle_data[i] = *(__be32 *)dp; + *(__be32 *)dp = cycle_lsn; + dp += BBSIZE; + } - /* - * Get the transaction layer to kick the dirty buffers out to - * disk asynchronously. No point in trying to do this if - * the filesystem is shutting down. - */ - if (threshold_lsn && - !XLOG_FORCED_SHUTDOWN(log)) - xfs_trans_push_ail(mp, threshold_lsn); -} /* xlog_grant_push_ail */ + if (xfs_sb_version_haslogv2(&log->l_mp->m_sb)) { + xlog_in_core_2_t *xhdr = iclog->ic_data; + + for ( ; i < BTOBB(size); i++) { + j = i / (XLOG_HEADER_CYCLE_SIZE / BBSIZE); + k = i % (XLOG_HEADER_CYCLE_SIZE / BBSIZE); + xhdr[j].hic_xheader.xh_cycle_data[k] = *(__be32 *)dp; + *(__be32 *)dp = cycle_lsn; + dp += BBSIZE; + } + for (i = 1; i < log->l_iclog_heads; i++) + xhdr[i].hic_xheader.xh_cycle = cycle_lsn; + } +} + +/* + * Calculate the checksum for a log buffer. + * + * This is a little more complicated than it should be because the various + * headers and the actual data are non-contiguous. + */ +__le32 +xlog_cksum( + struct xlog *log, + struct xlog_rec_header *rhead, + char *dp, + int size) +{ + __uint32_t crc; + + /* first generate the crc for the record header ... */ + crc = xfs_start_cksum((char *)rhead, + sizeof(struct xlog_rec_header), + offsetof(struct xlog_rec_header, h_crc)); + + /* ... then for additional cycle data for v2 logs ... */ + if (xfs_sb_version_haslogv2(&log->l_mp->m_sb)) { + union xlog_in_core2 *xhdr = (union xlog_in_core2 *)rhead; + int i; + + for (i = 1; i < log->l_iclog_heads; i++) { + crc = crc32c(crc, &xhdr[i].hic_xheader, + sizeof(struct xlog_rec_ext_header)); + } + } + + /* ... and finally for the payload */ + crc = crc32c(crc, dp, size); + + return xfs_end_cksum(crc); +} + +/* + * The bdstrat callback function for log bufs. This gives us a central + * place to trap bufs in case we get hit by a log I/O error and need to + * shutdown. Actually, in practice, even when we didn't get a log error, + * we transition the iclogs to IOERROR state *after* flushing all existing + * iclogs to disk. This is because we don't want anymore new transactions to be + * started or completed afterwards. + * + * We lock the iclogbufs here so that we can serialise against IO completion + * during unmount. We might be processing a shutdown triggered during unmount, + * and that can occur asynchronously to the unmount thread, and hence we need to + * ensure that completes before tearing down the iclogbufs. Hence we need to + * hold the buffer lock across the log IO to acheive that. + */ +STATIC int +xlog_bdstrat( + struct xfs_buf *bp) +{ + struct xlog_in_core *iclog = bp->b_fspriv; + + xfs_buf_lock(bp); + if (iclog->ic_state & XLOG_STATE_IOERROR) { + xfs_buf_ioerror(bp, EIO); + xfs_buf_stale(bp); + xfs_buf_ioend(bp, 0); + /* + * It would seem logical to return EIO here, but we rely on + * the log state machine to propagate I/O errors instead of + * doing it here. Similarly, IO completion will unlock the + * buffer, so we don't do it here. + */ + return 0; + } + + xfs_buf_iorequest(bp); + return 0; +} /* * Flush out the in-core log (iclog) to the on-disk log in an asynchronous @@ -1391,10 +1703,10 @@ xlog_grant_push_ail(xfs_mount_t *mp, */ STATIC int -xlog_sync(xlog_t *log, - xlog_in_core_t *iclog) +xlog_sync( + struct xlog *log, + struct xlog_in_core *iclog) { - xfs_caddr_t dptr; /* pointer to byte sized element */ xfs_buf_t *bp; int i; uint count; /* byte count of bwrite */ @@ -1402,10 +1714,11 @@ xlog_sync(xlog_t *log, int roundoff; /* roundoff to BB or stripe */ int split = 0; /* split write into two regions */ int error; - int v2 = XFS_SB_VERSION_HASLOGV2(&log->l_mp->m_sb); + int v2 = xfs_sb_version_haslogv2(&log->l_mp->m_sb); + int size; XFS_STATS_INC(xs_log_writes); - ASSERT(iclog->ic_refcnt == 0); + ASSERT(atomic_read(&iclog->ic_refcnt) == 0); /* Add for LR header */ count_init = log->l_iclog_hsize + iclog->ic_offset; @@ -1426,53 +1739,83 @@ xlog_sync(xlog_t *log, roundoff < BBTOB(1))); /* move grant heads by roundoff in sync */ - spin_lock(&log->l_grant_lock); - xlog_grant_add_space(log, roundoff); - spin_unlock(&log->l_grant_lock); + xlog_grant_add_space(log, &log->l_reserve_head.grant, roundoff); + xlog_grant_add_space(log, &log->l_write_head.grant, roundoff); /* put cycle number in every block */ xlog_pack_data(log, iclog, roundoff); /* real byte length */ - if (v2) { - iclog->ic_header.h_len = - cpu_to_be32(iclog->ic_offset + roundoff); - } else { - iclog->ic_header.h_len = - cpu_to_be32(iclog->ic_offset); - } + size = iclog->ic_offset; + if (v2) + size += roundoff; + iclog->ic_header.h_len = cpu_to_be32(size); bp = iclog->ic_bp; - ASSERT(XFS_BUF_FSPRIVATE2(bp, unsigned long) == (unsigned long)1); - XFS_BUF_SET_FSPRIVATE2(bp, (unsigned long)2); XFS_BUF_SET_ADDR(bp, BLOCK_LSN(be64_to_cpu(iclog->ic_header.h_lsn))); XFS_STATS_ADD(xs_log_blocks, BTOBB(count)); /* Do we need to split this write into 2 parts? */ if (XFS_BUF_ADDR(bp) + BTOBB(count) > log->l_logBBsize) { + char *dptr; + split = count - (BBTOB(log->l_logBBsize - XFS_BUF_ADDR(bp))); count = BBTOB(log->l_logBBsize - XFS_BUF_ADDR(bp)); - iclog->ic_bwritecnt = 2; /* split into 2 writes */ + iclog->ic_bwritecnt = 2; + + /* + * Bump the cycle numbers at the start of each block in the + * part of the iclog that ends up in the buffer that gets + * written to the start of the log. + * + * Watch out for the header magic number case, though. + */ + dptr = (char *)&iclog->ic_header + count; + for (i = 0; i < split; i += BBSIZE) { + __uint32_t cycle = be32_to_cpu(*(__be32 *)dptr); + if (++cycle == XLOG_HEADER_MAGIC_NUM) + cycle++; + *(__be32 *)dptr = cpu_to_be32(cycle); + + dptr += BBSIZE; + } } else { iclog->ic_bwritecnt = 1; } - XFS_BUF_SET_COUNT(bp, count); - XFS_BUF_SET_FSPRIVATE(bp, iclog); /* save for later */ + + /* calculcate the checksum */ + iclog->ic_header.h_crc = xlog_cksum(log, &iclog->ic_header, + iclog->ic_datap, size); + + bp->b_io_length = BTOBB(count); + bp->b_fspriv = iclog; XFS_BUF_ZEROFLAGS(bp); - XFS_BUF_BUSY(bp); XFS_BUF_ASYNC(bp); - /* - * Do an ordered write for the log block. - * Its unnecessary to flush the first split block in the log wrap case. - */ - if (!split && (log->l_mp->m_flags & XFS_MOUNT_BARRIER)) - XFS_BUF_ORDERED(bp); + bp->b_flags |= XBF_SYNCIO; + + if (log->l_mp->m_flags & XFS_MOUNT_BARRIER) { + bp->b_flags |= XBF_FUA; + + /* + * Flush the data device before flushing the log to make + * sure all meta data written back from the AIL actually made + * it to disk before stamping the new log tail LSN into the + * log buffer. For an external log we need to issue the + * flush explicitly, and unfortunately synchronously here; + * for an internal log we can simply use the block layer + * state machine for preflushes. + */ + if (log->l_mp->m_logdev_targp != log->l_mp->m_ddev_targp) + xfs_blkdev_issue_flush(log->l_mp->m_ddev_targp); + else + bp->b_flags |= XBF_FLUSH; + } ASSERT(XFS_BUF_ADDR(bp) <= log->l_logBBsize-1); ASSERT(XFS_BUF_ADDR(bp) + BTOBB(count) <= log->l_logBBsize); - xlog_verify_iclog(log, iclog, count, B_TRUE); + xlog_verify_iclog(log, iclog, count, true); /* account for log which doesn't start at block #0 */ XFS_BUF_SET_ADDR(bp, XFS_BUF_ADDR(bp) + log->l_logBBstart); @@ -1482,38 +1825,22 @@ xlog_sync(xlog_t *log, */ XFS_BUF_WRITE(bp); - if ((error = XFS_bwrite(bp))) { - xfs_ioerror_alert("xlog_sync", log->l_mp, bp, - XFS_BUF_ADDR(bp)); + error = xlog_bdstrat(bp); + if (error) { + xfs_buf_ioerror_alert(bp, "xlog_sync"); return error; } if (split) { bp = iclog->ic_log->l_xbuf; - ASSERT(XFS_BUF_FSPRIVATE2(bp, unsigned long) == - (unsigned long)1); - XFS_BUF_SET_FSPRIVATE2(bp, (unsigned long)2); XFS_BUF_SET_ADDR(bp, 0); /* logical 0 */ - XFS_BUF_SET_PTR(bp, (xfs_caddr_t)((__psint_t)&(iclog->ic_header)+ - (__psint_t)count), split); - XFS_BUF_SET_FSPRIVATE(bp, iclog); + xfs_buf_associate_memory(bp, + (char *)&iclog->ic_header + count, split); + bp->b_fspriv = iclog; XFS_BUF_ZEROFLAGS(bp); - XFS_BUF_BUSY(bp); XFS_BUF_ASYNC(bp); + bp->b_flags |= XBF_SYNCIO; if (log->l_mp->m_flags & XFS_MOUNT_BARRIER) - XFS_BUF_ORDERED(bp); - dptr = XFS_BUF_PTR(bp); - /* - * Bump the cycle numbers at the start of each block - * since this part of the buffer is at the start of - * a new cycle. Watch out for the header magic number - * case, though. - */ - for (i = 0; i < split; i += BBSIZE) { - be32_add_cpu((__be32 *)dptr, 1); - if (be32_to_cpu(*(__be32 *)dptr) == XLOG_HEADER_MAGIC_NUM) - be32_add_cpu((__be32 *)dptr, 1); - dptr += BBSIZE; - } + bp->b_flags |= XBF_FUA; ASSERT(XFS_BUF_ADDR(bp) <= log->l_logBBsize-1); ASSERT(XFS_BUF_ADDR(bp) + BTOBB(count) <= log->l_logBBsize); @@ -1521,71 +1848,59 @@ xlog_sync(xlog_t *log, /* account for internal log which doesn't start at block #0 */ XFS_BUF_SET_ADDR(bp, XFS_BUF_ADDR(bp) + log->l_logBBstart); XFS_BUF_WRITE(bp); - if ((error = XFS_bwrite(bp))) { - xfs_ioerror_alert("xlog_sync (split)", log->l_mp, - bp, XFS_BUF_ADDR(bp)); + error = xlog_bdstrat(bp); + if (error) { + xfs_buf_ioerror_alert(bp, "xlog_sync (split)"); return error; } } return 0; } /* xlog_sync */ - /* * Deallocate a log structure */ STATIC void -xlog_dealloc_log(xlog_t *log) +xlog_dealloc_log( + struct xlog *log) { xlog_in_core_t *iclog, *next_iclog; - xlog_ticket_t *tic, *next_tic; int i; + xlog_cil_destroy(log); + + /* + * Cycle all the iclogbuf locks to make sure all log IO completion + * is done before we tear down these buffers. + */ iclog = log->l_iclog; - for (i=0; i<log->l_iclog_bufs; i++) { - sv_destroy(&iclog->ic_forcesema); - sv_destroy(&iclog->ic_writesema); + for (i = 0; i < log->l_iclog_bufs; i++) { + xfs_buf_lock(iclog->ic_bp); + xfs_buf_unlock(iclog->ic_bp); + iclog = iclog->ic_next; + } + + /* + * Always need to ensure that the extra buffer does not point to memory + * owned by another log buffer before we free it. Also, cycle the lock + * first to ensure we've completed IO on it. + */ + xfs_buf_lock(log->l_xbuf); + xfs_buf_unlock(log->l_xbuf); + xfs_buf_set_empty(log->l_xbuf, BTOBB(log->l_iclog_size)); + xfs_buf_free(log->l_xbuf); + + iclog = log->l_iclog; + for (i = 0; i < log->l_iclog_bufs; i++) { xfs_buf_free(iclog->ic_bp); -#ifdef XFS_LOG_TRACE - if (iclog->ic_trace != NULL) { - ktrace_free(iclog->ic_trace); - } -#endif next_iclog = iclog->ic_next; - kmem_free(iclog, sizeof(xlog_in_core_t)); + kmem_free(iclog); iclog = next_iclog; } - freesema(&log->l_flushsema); spinlock_destroy(&log->l_icloglock); - spinlock_destroy(&log->l_grant_lock); - - /* XXXsup take a look at this again. */ - if ((log->l_ticket_cnt != log->l_ticket_tcnt) && - !XLOG_FORCED_SHUTDOWN(log)) { - xfs_fs_cmn_err(CE_WARN, log->l_mp, - "xlog_dealloc_log: (cnt: %d, total: %d)", - log->l_ticket_cnt, log->l_ticket_tcnt); - /* ASSERT(log->l_ticket_cnt == log->l_ticket_tcnt); */ - } else { - tic = log->l_unmount_free; - while (tic) { - next_tic = tic->t_next; - kmem_free(tic, PAGE_SIZE); - tic = next_tic; - } - } - xfs_buf_free(log->l_xbuf); -#ifdef XFS_LOG_TRACE - if (log->l_trace != NULL) { - ktrace_free(log->l_trace); - } - if (log->l_grant_trace != NULL) { - ktrace_free(log->l_grant_trace); - } -#endif log->l_mp->m_log = NULL; - kmem_free(log, sizeof(xlog_t)); + kmem_free(log); } /* xlog_dealloc_log */ /* @@ -1593,10 +1908,11 @@ xlog_dealloc_log(xlog_t *log) */ /* ARGSUSED */ static inline void -xlog_state_finish_copy(xlog_t *log, - xlog_in_core_t *iclog, - int record_cnt, - int copy_bytes) +xlog_state_finish_copy( + struct xlog *log, + struct xlog_in_core *iclog, + int record_cnt, + int copy_bytes) { spin_lock(&log->l_icloglock); @@ -1613,8 +1929,10 @@ xlog_state_finish_copy(xlog_t *log, * print out info relating to regions written which consume * the reservation */ -STATIC void -xlog_print_tic_res(xfs_mount_t *mp, xlog_ticket_t *ticket) +void +xlog_print_tic_res( + struct xfs_mount *mp, + struct xlog_ticket *ticket) { uint i; uint ophdr_spc = ticket->t_res_num_ophdrs * (uint)sizeof(xlog_op_header_t); @@ -1684,36 +2002,228 @@ xlog_print_tic_res(xfs_mount_t *mp, xlog_ticket_t *ticket) "SWAPEXT" }; - xfs_fs_cmn_err(CE_WARN, mp, - "xfs_log_write: reservation summary:\n" - " trans type = %s (%u)\n" - " unit res = %d bytes\n" - " current res = %d bytes\n" - " total reg = %u bytes (o/flow = %u bytes)\n" - " ophdrs = %u (ophdr space = %u bytes)\n" - " ophdr + reg = %u bytes\n" - " num regions = %u\n", - ((ticket->t_trans_type <= 0 || - ticket->t_trans_type > XFS_TRANS_TYPE_MAX) ? - "bad-trans-type" : trans_type_str[ticket->t_trans_type-1]), - ticket->t_trans_type, - ticket->t_unit_res, - ticket->t_curr_res, - ticket->t_res_arr_sum, ticket->t_res_o_flow, - ticket->t_res_num_ophdrs, ophdr_spc, - ticket->t_res_arr_sum + - ticket->t_res_o_flow + ophdr_spc, - ticket->t_res_num); + xfs_warn(mp, + "xlog_write: reservation summary:\n" + " trans type = %s (%u)\n" + " unit res = %d bytes\n" + " current res = %d bytes\n" + " total reg = %u bytes (o/flow = %u bytes)\n" + " ophdrs = %u (ophdr space = %u bytes)\n" + " ophdr + reg = %u bytes\n" + " num regions = %u\n", + ((ticket->t_trans_type <= 0 || + ticket->t_trans_type > XFS_TRANS_TYPE_MAX) ? + "bad-trans-type" : trans_type_str[ticket->t_trans_type-1]), + ticket->t_trans_type, + ticket->t_unit_res, + ticket->t_curr_res, + ticket->t_res_arr_sum, ticket->t_res_o_flow, + ticket->t_res_num_ophdrs, ophdr_spc, + ticket->t_res_arr_sum + + ticket->t_res_o_flow + ophdr_spc, + ticket->t_res_num); for (i = 0; i < ticket->t_res_num; i++) { - uint r_type = ticket->t_res_arr[i].r_type; - cmn_err(CE_WARN, - "region[%u]: %s - %u bytes\n", - i, + uint r_type = ticket->t_res_arr[i].r_type; + xfs_warn(mp, "region[%u]: %s - %u bytes", i, ((r_type <= 0 || r_type > XLOG_REG_TYPE_MAX) ? "bad-rtype" : res_type_str[r_type-1]), ticket->t_res_arr[i].r_len); } + + xfs_alert_tag(mp, XFS_PTAG_LOGRES, + "xlog_write: reservation ran out. Need to up reservation"); + xfs_force_shutdown(mp, SHUTDOWN_LOG_IO_ERROR); +} + +/* + * Calculate the potential space needed by the log vector. Each region gets + * its own xlog_op_header_t and may need to be double word aligned. + */ +static int +xlog_write_calc_vec_length( + struct xlog_ticket *ticket, + struct xfs_log_vec *log_vector) +{ + struct xfs_log_vec *lv; + int headers = 0; + int len = 0; + int i; + + /* acct for start rec of xact */ + if (ticket->t_flags & XLOG_TIC_INITED) + headers++; + + for (lv = log_vector; lv; lv = lv->lv_next) { + /* we don't write ordered log vectors */ + if (lv->lv_buf_len == XFS_LOG_VEC_ORDERED) + continue; + + headers += lv->lv_niovecs; + + for (i = 0; i < lv->lv_niovecs; i++) { + struct xfs_log_iovec *vecp = &lv->lv_iovecp[i]; + + len += vecp->i_len; + xlog_tic_add_region(ticket, vecp->i_len, vecp->i_type); + } + } + + ticket->t_res_num_ophdrs += headers; + len += headers * sizeof(struct xlog_op_header); + + return len; +} + +/* + * If first write for transaction, insert start record We can't be trying to + * commit if we are inited. We can't have any "partial_copy" if we are inited. + */ +static int +xlog_write_start_rec( + struct xlog_op_header *ophdr, + struct xlog_ticket *ticket) +{ + if (!(ticket->t_flags & XLOG_TIC_INITED)) + return 0; + + ophdr->oh_tid = cpu_to_be32(ticket->t_tid); + ophdr->oh_clientid = ticket->t_clientid; + ophdr->oh_len = 0; + ophdr->oh_flags = XLOG_START_TRANS; + ophdr->oh_res2 = 0; + + ticket->t_flags &= ~XLOG_TIC_INITED; + + return sizeof(struct xlog_op_header); +} + +static xlog_op_header_t * +xlog_write_setup_ophdr( + struct xlog *log, + struct xlog_op_header *ophdr, + struct xlog_ticket *ticket, + uint flags) +{ + ophdr->oh_tid = cpu_to_be32(ticket->t_tid); + ophdr->oh_clientid = ticket->t_clientid; + ophdr->oh_res2 = 0; + + /* are we copying a commit or unmount record? */ + ophdr->oh_flags = flags; + + /* + * We've seen logs corrupted with bad transaction client ids. This + * makes sure that XFS doesn't generate them on. Turn this into an EIO + * and shut down the filesystem. + */ + switch (ophdr->oh_clientid) { + case XFS_TRANSACTION: + case XFS_VOLUME: + case XFS_LOG: + break; + default: + xfs_warn(log->l_mp, + "Bad XFS transaction clientid 0x%x in ticket 0x%p", + ophdr->oh_clientid, ticket); + return NULL; + } + + return ophdr; +} + +/* + * Set up the parameters of the region copy into the log. This has + * to handle region write split across multiple log buffers - this + * state is kept external to this function so that this code can + * be written in an obvious, self documenting manner. + */ +static int +xlog_write_setup_copy( + struct xlog_ticket *ticket, + struct xlog_op_header *ophdr, + int space_available, + int space_required, + int *copy_off, + int *copy_len, + int *last_was_partial_copy, + int *bytes_consumed) +{ + int still_to_copy; + + still_to_copy = space_required - *bytes_consumed; + *copy_off = *bytes_consumed; + + if (still_to_copy <= space_available) { + /* write of region completes here */ + *copy_len = still_to_copy; + ophdr->oh_len = cpu_to_be32(*copy_len); + if (*last_was_partial_copy) + ophdr->oh_flags |= (XLOG_END_TRANS|XLOG_WAS_CONT_TRANS); + *last_was_partial_copy = 0; + *bytes_consumed = 0; + return 0; + } + + /* partial write of region, needs extra log op header reservation */ + *copy_len = space_available; + ophdr->oh_len = cpu_to_be32(*copy_len); + ophdr->oh_flags |= XLOG_CONTINUE_TRANS; + if (*last_was_partial_copy) + ophdr->oh_flags |= XLOG_WAS_CONT_TRANS; + *bytes_consumed += *copy_len; + (*last_was_partial_copy)++; + + /* account for new log op header */ + ticket->t_curr_res -= sizeof(struct xlog_op_header); + ticket->t_res_num_ophdrs++; + + return sizeof(struct xlog_op_header); +} + +static int +xlog_write_copy_finish( + struct xlog *log, + struct xlog_in_core *iclog, + uint flags, + int *record_cnt, + int *data_cnt, + int *partial_copy, + int *partial_copy_len, + int log_offset, + struct xlog_in_core **commit_iclog) +{ + if (*partial_copy) { + /* + * This iclog has already been marked WANT_SYNC by + * xlog_state_get_iclog_space. + */ + xlog_state_finish_copy(log, iclog, *record_cnt, *data_cnt); + *record_cnt = 0; + *data_cnt = 0; + return xlog_state_release_iclog(log, iclog); + } + + *partial_copy = 0; + *partial_copy_len = 0; + + if (iclog->ic_size - log_offset <= sizeof(xlog_op_header_t)) { + /* no more space in this iclog - push it. */ + xlog_state_finish_copy(log, iclog, *record_cnt, *data_cnt); + *record_cnt = 0; + *data_cnt = 0; + + spin_lock(&log->l_icloglock); + xlog_state_want_sync(log, iclog); + spin_unlock(&log->l_icloglock); + + if (!commit_iclog) + return xlog_state_release_iclog(log, iclog); + ASSERT(flags & XLOG_COMMIT_TRANS); + *commit_iclog = iclog; + } + + return 0; } /* @@ -1756,209 +2266,171 @@ xlog_print_tic_res(xfs_mount_t *mp, xlog_ticket_t *ticket) * we don't update ic_offset until the end when we know exactly how many * bytes have been written out. */ -STATIC int -xlog_write(xfs_mount_t * mp, - xfs_log_iovec_t reg[], - int nentries, - xfs_log_ticket_t tic, - xfs_lsn_t *start_lsn, - xlog_in_core_t **commit_iclog, - uint flags) -{ - xlog_t *log = mp->m_log; - xlog_ticket_t *ticket = (xlog_ticket_t *)tic; - xlog_in_core_t *iclog = NULL; /* ptr to current in-core log */ - xlog_op_header_t *logop_head; /* ptr to log operation header */ - __psint_t ptr; /* copy address into data region */ - int len; /* # xlog_write() bytes 2 still copy */ - int index; /* region index currently copying */ - int log_offset; /* offset (from 0) into data region */ - int start_rec_copy; /* # bytes to copy for start record */ - int partial_copy; /* did we split a region? */ - int partial_copy_len;/* # bytes copied if split region */ - int need_copy; /* # bytes need to memcpy this region */ - int copy_len; /* # bytes actually memcpy'ing */ - int copy_off; /* # bytes from entry start */ - int contwr; /* continued write of in-core log? */ - int error; - int record_cnt = 0, data_cnt = 0; - - partial_copy_len = partial_copy = 0; - - /* Calculate potential maximum space. Each region gets its own - * xlog_op_header_t and may need to be double word aligned. - */ - len = 0; - if (ticket->t_flags & XLOG_TIC_INITED) { /* acct for start rec of xact */ - len += sizeof(xlog_op_header_t); - ticket->t_res_num_ophdrs++; - } +int +xlog_write( + struct xlog *log, + struct xfs_log_vec *log_vector, + struct xlog_ticket *ticket, + xfs_lsn_t *start_lsn, + struct xlog_in_core **commit_iclog, + uint flags) +{ + struct xlog_in_core *iclog = NULL; + struct xfs_log_iovec *vecp; + struct xfs_log_vec *lv; + int len; + int index; + int partial_copy = 0; + int partial_copy_len = 0; + int contwr = 0; + int record_cnt = 0; + int data_cnt = 0; + int error; + + *start_lsn = 0; + + len = xlog_write_calc_vec_length(ticket, log_vector); - for (index = 0; index < nentries; index++) { - len += sizeof(xlog_op_header_t); /* each region gets >= 1 */ - ticket->t_res_num_ophdrs++; - len += reg[index].i_len; - xlog_tic_add_region(ticket, reg[index].i_len, reg[index].i_type); - } - contwr = *start_lsn = 0; + /* + * Region headers and bytes are already accounted for. + * We only need to take into account start records and + * split regions in this function. + */ + if (ticket->t_flags & XLOG_TIC_INITED) + ticket->t_curr_res -= sizeof(xlog_op_header_t); - if (ticket->t_curr_res < len) { - xlog_print_tic_res(mp, ticket); -#ifdef DEBUG - xlog_panic( - "xfs_log_write: reservation ran out. Need to up reservation"); -#else - /* Customer configurable panic */ - xfs_cmn_err(XFS_PTAG_LOGRES, CE_ALERT, mp, - "xfs_log_write: reservation ran out. Need to up reservation"); - /* If we did not panic, shutdown the filesystem */ - xfs_force_shutdown(mp, SHUTDOWN_CORRUPT_INCORE); -#endif - } else - ticket->t_curr_res -= len; + /* + * Commit record headers need to be accounted for. These + * come in as separate writes so are easy to detect. + */ + if (flags & (XLOG_COMMIT_TRANS | XLOG_UNMOUNT_TRANS)) + ticket->t_curr_res -= sizeof(xlog_op_header_t); - for (index = 0; index < nentries; ) { - if ((error = xlog_state_get_iclog_space(log, len, &iclog, ticket, - &contwr, &log_offset))) - return error; + if (ticket->t_curr_res < 0) + xlog_print_tic_res(log->l_mp, ticket); - ASSERT(log_offset <= iclog->ic_size - 1); - ptr = (__psint_t) ((char *)iclog->ic_datap+log_offset); + index = 0; + lv = log_vector; + vecp = lv->lv_iovecp; + while (lv && (!lv->lv_niovecs || index < lv->lv_niovecs)) { + void *ptr; + int log_offset; - /* start_lsn is the first lsn written to. That's all we need. */ - if (! *start_lsn) - *start_lsn = be64_to_cpu(iclog->ic_header.h_lsn); + error = xlog_state_get_iclog_space(log, len, &iclog, ticket, + &contwr, &log_offset); + if (error) + return error; - /* This loop writes out as many regions as can fit in the amount - * of space which was allocated by xlog_state_get_iclog_space(). - */ - while (index < nentries) { - ASSERT(reg[index].i_len % sizeof(__int32_t) == 0); - ASSERT((__psint_t)ptr % sizeof(__int32_t) == 0); - start_rec_copy = 0; - - /* If first write for transaction, insert start record. - * We can't be trying to commit if we are inited. We can't - * have any "partial_copy" if we are inited. - */ - if (ticket->t_flags & XLOG_TIC_INITED) { - logop_head = (xlog_op_header_t *)ptr; - logop_head->oh_tid = cpu_to_be32(ticket->t_tid); - logop_head->oh_clientid = ticket->t_clientid; - logop_head->oh_len = 0; - logop_head->oh_flags = XLOG_START_TRANS; - logop_head->oh_res2 = 0; - ticket->t_flags &= ~XLOG_TIC_INITED; /* clear bit */ - record_cnt++; - - start_rec_copy = sizeof(xlog_op_header_t); - xlog_write_adv_cnt(ptr, len, log_offset, start_rec_copy); - } + ASSERT(log_offset <= iclog->ic_size - 1); + ptr = iclog->ic_datap + log_offset; + + /* start_lsn is the first lsn written to. That's all we need. */ + if (!*start_lsn) + *start_lsn = be64_to_cpu(iclog->ic_header.h_lsn); - /* Copy log operation header directly into data section */ - logop_head = (xlog_op_header_t *)ptr; - logop_head->oh_tid = cpu_to_be32(ticket->t_tid); - logop_head->oh_clientid = ticket->t_clientid; - logop_head->oh_res2 = 0; + /* + * This loop writes out as many regions as can fit in the amount + * of space which was allocated by xlog_state_get_iclog_space(). + */ + while (lv && (!lv->lv_niovecs || index < lv->lv_niovecs)) { + struct xfs_log_iovec *reg; + struct xlog_op_header *ophdr; + int start_rec_copy; + int copy_len; + int copy_off; + bool ordered = false; + + /* ordered log vectors have no regions to write */ + if (lv->lv_buf_len == XFS_LOG_VEC_ORDERED) { + ASSERT(lv->lv_niovecs == 0); + ordered = true; + goto next_lv; + } - /* header copied directly */ - xlog_write_adv_cnt(ptr, len, log_offset, sizeof(xlog_op_header_t)); + reg = &vecp[index]; + ASSERT(reg->i_len % sizeof(__int32_t) == 0); + ASSERT((unsigned long)ptr % sizeof(__int32_t) == 0); - /* are we copying a commit or unmount record? */ - logop_head->oh_flags = flags; + start_rec_copy = xlog_write_start_rec(ptr, ticket); + if (start_rec_copy) { + record_cnt++; + xlog_write_adv_cnt(&ptr, &len, &log_offset, + start_rec_copy); + } - /* - * We've seen logs corrupted with bad transaction client - * ids. This makes sure that XFS doesn't generate them on. - * Turn this into an EIO and shut down the filesystem. - */ - switch (logop_head->oh_clientid) { - case XFS_TRANSACTION: - case XFS_VOLUME: - case XFS_LOG: - break; - default: - xfs_fs_cmn_err(CE_WARN, mp, - "Bad XFS transaction clientid 0x%x in ticket 0x%p", - logop_head->oh_clientid, tic); - return XFS_ERROR(EIO); - } + ophdr = xlog_write_setup_ophdr(log, ptr, ticket, flags); + if (!ophdr) + return XFS_ERROR(EIO); - /* Partial write last time? => (partial_copy != 0) - * need_copy is the amount we'd like to copy if everything could - * fit in the current memcpy. - */ - need_copy = reg[index].i_len - partial_copy_len; - - copy_off = partial_copy_len; - if (need_copy <= iclog->ic_size - log_offset) { /*complete write */ - copy_len = need_copy; - logop_head->oh_len = cpu_to_be32(copy_len); - if (partial_copy) - logop_head->oh_flags|= (XLOG_END_TRANS|XLOG_WAS_CONT_TRANS); - partial_copy_len = partial_copy = 0; - } else { /* partial write */ - copy_len = iclog->ic_size - log_offset; - logop_head->oh_len = cpu_to_be32(copy_len); - logop_head->oh_flags |= XLOG_CONTINUE_TRANS; - if (partial_copy) - logop_head->oh_flags |= XLOG_WAS_CONT_TRANS; - partial_copy_len += copy_len; - partial_copy++; - len += sizeof(xlog_op_header_t); /* from splitting of region */ - /* account for new log op header */ - ticket->t_curr_res -= sizeof(xlog_op_header_t); - ticket->t_res_num_ophdrs++; - } - xlog_verify_dest_ptr(log, ptr); - - /* copy region */ - ASSERT(copy_len >= 0); - memcpy((xfs_caddr_t)ptr, reg[index].i_addr + copy_off, copy_len); - xlog_write_adv_cnt(ptr, len, log_offset, copy_len); - - /* make copy_len total bytes copied, including headers */ - copy_len += start_rec_copy + sizeof(xlog_op_header_t); - record_cnt++; - data_cnt += contwr ? copy_len : 0; - if (partial_copy) { /* copied partial region */ - /* already marked WANT_SYNC by xlog_state_get_iclog_space */ - xlog_state_finish_copy(log, iclog, record_cnt, data_cnt); - record_cnt = data_cnt = 0; - if ((error = xlog_state_release_iclog(log, iclog))) - return error; - break; /* don't increment index */ - } else { /* copied entire region */ - index++; - partial_copy_len = partial_copy = 0; - - if (iclog->ic_size - log_offset <= sizeof(xlog_op_header_t)) { - xlog_state_finish_copy(log, iclog, record_cnt, data_cnt); - record_cnt = data_cnt = 0; - xlog_state_want_sync(log, iclog); - if (commit_iclog) { - ASSERT(flags & XLOG_COMMIT_TRANS); - *commit_iclog = iclog; - } else if ((error = xlog_state_release_iclog(log, iclog))) - return error; - if (index == nentries) - return 0; /* we are done */ - else - break; + xlog_write_adv_cnt(&ptr, &len, &log_offset, + sizeof(struct xlog_op_header)); + + len += xlog_write_setup_copy(ticket, ophdr, + iclog->ic_size-log_offset, + reg->i_len, + ©_off, ©_len, + &partial_copy, + &partial_copy_len); + xlog_verify_dest_ptr(log, ptr); + + /* copy region */ + ASSERT(copy_len >= 0); + memcpy(ptr, reg->i_addr + copy_off, copy_len); + xlog_write_adv_cnt(&ptr, &len, &log_offset, copy_len); + + copy_len += start_rec_copy + sizeof(xlog_op_header_t); + record_cnt++; + data_cnt += contwr ? copy_len : 0; + + error = xlog_write_copy_finish(log, iclog, flags, + &record_cnt, &data_cnt, + &partial_copy, + &partial_copy_len, + log_offset, + commit_iclog); + if (error) + return error; + + /* + * if we had a partial copy, we need to get more iclog + * space but we don't want to increment the region + * index because there is still more is this region to + * write. + * + * If we completed writing this region, and we flushed + * the iclog (indicated by resetting of the record + * count), then we also need to get more log space. If + * this was the last record, though, we are done and + * can just return. + */ + if (partial_copy) + break; + + if (++index == lv->lv_niovecs) { +next_lv: + lv = lv->lv_next; + index = 0; + if (lv) + vecp = lv->lv_iovecp; + } + if (record_cnt == 0 && ordered == false) { + if (!lv) + return 0; + break; + } } - } /* if (partial_copy) */ - } /* while (index < nentries) */ - } /* for (index = 0; index < nentries; ) */ - ASSERT(len == 0); + } + + ASSERT(len == 0); + + xlog_state_finish_copy(log, iclog, record_cnt, data_cnt); + if (!commit_iclog) + return xlog_state_release_iclog(log, iclog); - xlog_state_finish_copy(log, iclog, record_cnt, data_cnt); - if (commit_iclog) { ASSERT(flags & XLOG_COMMIT_TRANS); *commit_iclog = iclog; return 0; - } - return xlog_state_release_iclog(log, iclog); -} /* xlog_write */ +} /***************************************************************************** @@ -1971,13 +2443,14 @@ xlog_write(xfs_mount_t * mp, /* Clean iclogs starting from the head. This ordering must be * maintained, so an iclog doesn't become ACTIVE beyond one that * is SYNCING. This is also required to maintain the notion that we use - * a counting semaphore to hold off would be writers to the log when every + * a ordered wait queue to hold off would be writers to the log when every * iclog is trying to sync to disk. * * State Change: DIRTY -> ACTIVE */ STATIC void -xlog_state_clean_log(xlog_t *log) +xlog_state_clean_log( + struct xlog *log) { xlog_in_core_t *iclog; int changed = 0; @@ -1987,7 +2460,7 @@ xlog_state_clean_log(xlog_t *log) if (iclog->ic_state == XLOG_STATE_DIRTY) { iclog->ic_state = XLOG_STATE_ACTIVE; iclog->ic_offset = 0; - iclog->ic_callback = NULL; /* don't need to free */ + ASSERT(iclog->ic_callback == NULL); /* * If the number of ops in this iclog indicate it just * contains the dummy transaction, we can @@ -2057,7 +2530,7 @@ xlog_state_clean_log(xlog_t *log) STATIC xfs_lsn_t xlog_get_lowest_lsn( - xlog_t *log) + struct xlog *log) { xlog_in_core_t *lsn_log; xfs_lsn_t lowest_lsn, lsn; @@ -2080,9 +2553,9 @@ xlog_get_lowest_lsn( STATIC void xlog_state_do_callback( - xlog_t *log, - int aborted, - xlog_in_core_t *ciclog) + struct xlog *log, + int aborted, + struct xlog_in_core *ciclog) { xlog_in_core_t *iclog; xlog_in_core_t *first_iclog; /* used to know when we've @@ -2095,6 +2568,7 @@ xlog_state_do_callback( int funcdidcallbacks; /* flag: function did callbacks */ int repeats; /* for issuing console warnings if * looping too many times */ + int wake = 0; spin_lock(&log->l_icloglock); first_iclog = iclog = log->l_iclog; @@ -2169,7 +2643,7 @@ xlog_state_do_callback( lowest_lsn = xlog_get_lowest_lsn(log); if (lowest_lsn && XFS_LSN_CMP(lowest_lsn, - be64_to_cpu(iclog->ic_header.h_lsn)) < 0) { + be64_to_cpu(iclog->ic_header.h_lsn)) < 0) { iclog = iclog->ic_next; continue; /* Leave this iclog for * another thread */ @@ -2177,50 +2651,64 @@ xlog_state_do_callback( iclog->ic_state = XLOG_STATE_CALLBACK; - spin_unlock(&log->l_icloglock); - - /* l_last_sync_lsn field protected by - * l_grant_lock. Don't worry about iclog's lsn. - * No one else can be here except us. - */ - spin_lock(&log->l_grant_lock); - ASSERT(XFS_LSN_CMP(log->l_last_sync_lsn, - be64_to_cpu(iclog->ic_header.h_lsn)) <= 0); - log->l_last_sync_lsn = - be64_to_cpu(iclog->ic_header.h_lsn); - spin_unlock(&log->l_grant_lock); /* - * Keep processing entries in the callback list - * until we come around and it is empty. We - * need to atomically see that the list is - * empty and change the state to DIRTY so that - * we don't miss any more callbacks being added. + * Completion of a iclog IO does not imply that + * a transaction has completed, as transactions + * can be large enough to span many iclogs. We + * cannot change the tail of the log half way + * through a transaction as this may be the only + * transaction in the log and moving th etail to + * point to the middle of it will prevent + * recovery from finding the start of the + * transaction. Hence we should only update the + * last_sync_lsn if this iclog contains + * transaction completion callbacks on it. + * + * We have to do this before we drop the + * icloglock to ensure we are the only one that + * can update it. */ - spin_lock(&log->l_icloglock); - } else { + ASSERT(XFS_LSN_CMP(atomic64_read(&log->l_last_sync_lsn), + be64_to_cpu(iclog->ic_header.h_lsn)) <= 0); + if (iclog->ic_callback) + atomic64_set(&log->l_last_sync_lsn, + be64_to_cpu(iclog->ic_header.h_lsn)); + + } else ioerrors++; - } - cb = iclog->ic_callback; + spin_unlock(&log->l_icloglock); + + /* + * Keep processing entries in the callback list until + * we come around and it is empty. We need to + * atomically see that the list is empty and change the + * state to DIRTY so that we don't miss any more + * callbacks being added. + */ + spin_lock(&iclog->ic_callback_lock); + cb = iclog->ic_callback; while (cb) { iclog->ic_callback_tail = &(iclog->ic_callback); iclog->ic_callback = NULL; - spin_unlock(&log->l_icloglock); + spin_unlock(&iclog->ic_callback_lock); /* perform callbacks in the order given */ for (; cb; cb = cb_next) { cb_next = cb->cb_next; cb->cb_func(cb->cb_arg, aborted); } - spin_lock(&log->l_icloglock); + spin_lock(&iclog->ic_callback_lock); cb = iclog->ic_callback; } loopdidcallbacks++; funcdidcallbacks++; + spin_lock(&log->l_icloglock); ASSERT(iclog->ic_callback == NULL); + spin_unlock(&iclog->ic_callback_lock); if (!(iclog->ic_state & XLOG_STATE_IOERROR)) iclog->ic_state = XLOG_STATE_DIRTY; @@ -2231,7 +2719,7 @@ xlog_state_do_callback( xlog_state_clean_log(log); /* wake up threads waiting in xfs_log_force() */ - sv_broadcast(&iclog->ic_forcesema); + wake_up_all(&iclog->ic_force_wait); iclog = iclog->ic_next; } while (first_iclog != iclog); @@ -2239,9 +2727,9 @@ xlog_state_do_callback( if (repeats > 5000) { flushcnt += repeats; repeats = 0; - xfs_fs_cmn_err(CE_WARN, log->l_mp, + xfs_warn(log->l_mp, "%s: possible infinite loop (%d iterations)", - __FUNCTION__, flushcnt); + __func__, flushcnt); } } while (!ioerrors && loopdidcallbacks); @@ -2273,15 +2761,13 @@ xlog_state_do_callback( } #endif - flushcnt = 0; - if (log->l_iclog->ic_state & (XLOG_STATE_ACTIVE|XLOG_STATE_IOERROR)) { - flushcnt = log->l_flushcnt; - log->l_flushcnt = 0; - } + if (log->l_iclog->ic_state & (XLOG_STATE_ACTIVE|XLOG_STATE_IOERROR)) + wake = 1; spin_unlock(&log->l_icloglock); - while (flushcnt--) - vsema(&log->l_flushsema); -} /* xlog_state_do_callback */ + + if (wake) + wake_up_all(&log->l_flush_wait); +} /* @@ -2295,21 +2781,20 @@ xlog_state_do_callback( * the second completion goes through. * * Callbacks could take time, so they are done outside the scope of the - * global state machine log lock. Assume that the calls to cvsema won't - * take a long time. At least we know it won't sleep. + * global state machine log lock. */ STATIC void xlog_state_done_syncing( xlog_in_core_t *iclog, int aborted) { - xlog_t *log = iclog->ic_log; + struct xlog *log = iclog->ic_log; spin_lock(&log->l_icloglock); ASSERT(iclog->ic_state == XLOG_STATE_SYNCING || iclog->ic_state == XLOG_STATE_IOERROR); - ASSERT(iclog->ic_refcnt == 0); + ASSERT(atomic_read(&iclog->ic_refcnt) == 0); ASSERT(iclog->ic_bwritecnt == 1 || iclog->ic_bwritecnt == 2); @@ -2332,7 +2817,7 @@ xlog_state_done_syncing( * iclog buffer, we wake them all, one will get to do the * I/O, the others get to wait for the result. */ - sv_broadcast(&iclog->ic_writesema); + wake_up_all(&iclog->ic_write_wait); spin_unlock(&log->l_icloglock); xlog_state_do_callback(log, aborted, iclog); /* also cleans log */ } /* xlog_state_done_syncing */ @@ -2340,11 +2825,9 @@ xlog_state_done_syncing( /* * If the head of the in-core log ring is not (ACTIVE or DIRTY), then we must - * sleep. The flush semaphore is set to the number of in-core buffers and - * decremented around disk syncing. Therefore, if all buffers are syncing, - * this semaphore will cause new writes to sleep until a sync completes. - * Otherwise, this code just does p() followed by v(). This approximates - * a sleep/wakeup except we can't race. + * sleep. We wait on the flush queue on the head iclog as that should be + * the first iclog to complete flushing. Hence if all iclogs are syncing, + * we will wait here and all new writes will sleep until a sync completes. * * The in-core logs are used in a circular fashion. They are not used * out-of-order even when an iclog past the head is free. @@ -2359,12 +2842,13 @@ xlog_state_done_syncing( * is copied. */ STATIC int -xlog_state_get_iclog_space(xlog_t *log, - int len, - xlog_in_core_t **iclogp, - xlog_ticket_t *ticket, - int *continued_write, - int *logoffsetp) +xlog_state_get_iclog_space( + struct xlog *log, + int len, + struct xlog_in_core **iclogp, + struct xlog_ticket *ticket, + int *continued_write, + int *logoffsetp) { int log_offset; xlog_rec_header_t *head; @@ -2379,19 +2863,17 @@ restart: } iclog = log->l_iclog; - if (! (iclog->ic_state == XLOG_STATE_ACTIVE)) { - log->l_flushcnt++; - spin_unlock(&log->l_icloglock); - xlog_trace_iclog(iclog, XLOG_TRACE_SLEEP_FLUSH); + if (iclog->ic_state != XLOG_STATE_ACTIVE) { XFS_STATS_INC(xs_log_noiclogs); - /* Ensure that log writes happen */ - psema(&log->l_flushsema, PINOD); + + /* Wait for log writes to have flushed */ + xlog_wait(&log->l_flush_wait, &log->l_icloglock); goto restart; } - ASSERT(iclog->ic_state == XLOG_STATE_ACTIVE); + head = &iclog->ic_header; - iclog->ic_refcnt++; /* prevents sync */ + atomic_inc(&iclog->ic_refcnt); /* prevents sync */ log_offset = iclog->ic_offset; /* On the 1st write to an iclog, figure out lsn. This works @@ -2422,13 +2904,20 @@ restart: if (iclog->ic_size - iclog->ic_offset < 2*sizeof(xlog_op_header_t)) { xlog_state_switch_iclogs(log, iclog, iclog->ic_size); - /* If I'm the only one writing to this iclog, sync it to disk */ - if (iclog->ic_refcnt == 1) { + /* + * If I'm the only one writing to this iclog, sync it to disk. + * We need to do an atomic compare and decrement here to avoid + * racing with concurrent atomic_dec_and_lock() calls in + * xlog_state_release_iclog() when there is more than one + * reference to the iclog. + */ + if (!atomic_add_unless(&iclog->ic_refcnt, -1, 1)) { + /* we are the only one */ spin_unlock(&log->l_icloglock); - if ((error = xlog_state_release_iclog(log, iclog))) + error = xlog_state_release_iclog(log, iclog); + if (error) return error; } else { - iclog->ic_refcnt--; spin_unlock(&log->l_icloglock); } goto restart; @@ -2456,258 +2945,6 @@ restart: return 0; } /* xlog_state_get_iclog_space */ -/* - * Atomically get the log space required for a log ticket. - * - * Once a ticket gets put onto the reserveq, it will only return after - * the needed reservation is satisfied. - */ -STATIC int -xlog_grant_log_space(xlog_t *log, - xlog_ticket_t *tic) -{ - int free_bytes; - int need_bytes; -#ifdef DEBUG - xfs_lsn_t tail_lsn; -#endif - - -#ifdef DEBUG - if (log->l_flags & XLOG_ACTIVE_RECOVERY) - panic("grant Recovery problem"); -#endif - - /* Is there space or do we need to sleep? */ - spin_lock(&log->l_grant_lock); - xlog_trace_loggrant(log, tic, "xlog_grant_log_space: enter"); - - /* something is already sleeping; insert new transaction at end */ - if (log->l_reserve_headq) { - xlog_ins_ticketq(&log->l_reserve_headq, tic); - xlog_trace_loggrant(log, tic, - "xlog_grant_log_space: sleep 1"); - /* - * Gotta check this before going to sleep, while we're - * holding the grant lock. - */ - if (XLOG_FORCED_SHUTDOWN(log)) - goto error_return; - - XFS_STATS_INC(xs_sleep_logspace); - sv_wait(&tic->t_sema, PINOD|PLTWAIT, &log->l_grant_lock, s); - /* - * If we got an error, and the filesystem is shutting down, - * we'll catch it down below. So just continue... - */ - xlog_trace_loggrant(log, tic, - "xlog_grant_log_space: wake 1"); - spin_lock(&log->l_grant_lock); - } - if (tic->t_flags & XFS_LOG_PERM_RESERV) - need_bytes = tic->t_unit_res*tic->t_ocnt; - else - need_bytes = tic->t_unit_res; - -redo: - if (XLOG_FORCED_SHUTDOWN(log)) - goto error_return; - - free_bytes = xlog_space_left(log, log->l_grant_reserve_cycle, - log->l_grant_reserve_bytes); - if (free_bytes < need_bytes) { - if ((tic->t_flags & XLOG_TIC_IN_Q) == 0) - xlog_ins_ticketq(&log->l_reserve_headq, tic); - xlog_trace_loggrant(log, tic, - "xlog_grant_log_space: sleep 2"); - XFS_STATS_INC(xs_sleep_logspace); - sv_wait(&tic->t_sema, PINOD|PLTWAIT, &log->l_grant_lock, s); - - if (XLOG_FORCED_SHUTDOWN(log)) { - spin_lock(&log->l_grant_lock); - goto error_return; - } - - xlog_trace_loggrant(log, tic, - "xlog_grant_log_space: wake 2"); - xlog_grant_push_ail(log->l_mp, need_bytes); - spin_lock(&log->l_grant_lock); - goto redo; - } else if (tic->t_flags & XLOG_TIC_IN_Q) - xlog_del_ticketq(&log->l_reserve_headq, tic); - - /* we've got enough space */ - xlog_grant_add_space(log, need_bytes); -#ifdef DEBUG - tail_lsn = log->l_tail_lsn; - /* - * Check to make sure the grant write head didn't just over lap the - * tail. If the cycles are the same, we can't be overlapping. - * Otherwise, make sure that the cycles differ by exactly one and - * check the byte count. - */ - if (CYCLE_LSN(tail_lsn) != log->l_grant_write_cycle) { - ASSERT(log->l_grant_write_cycle-1 == CYCLE_LSN(tail_lsn)); - ASSERT(log->l_grant_write_bytes <= BBTOB(BLOCK_LSN(tail_lsn))); - } -#endif - xlog_trace_loggrant(log, tic, "xlog_grant_log_space: exit"); - xlog_verify_grant_head(log, 1); - spin_unlock(&log->l_grant_lock); - return 0; - - error_return: - if (tic->t_flags & XLOG_TIC_IN_Q) - xlog_del_ticketq(&log->l_reserve_headq, tic); - xlog_trace_loggrant(log, tic, "xlog_grant_log_space: err_ret"); - /* - * If we are failing, make sure the ticket doesn't have any - * current reservations. We don't want to add this back when - * the ticket/transaction gets cancelled. - */ - tic->t_curr_res = 0; - tic->t_cnt = 0; /* ungrant will give back unit_res * t_cnt. */ - spin_unlock(&log->l_grant_lock); - return XFS_ERROR(EIO); -} /* xlog_grant_log_space */ - - -/* - * Replenish the byte reservation required by moving the grant write head. - * - * - */ -STATIC int -xlog_regrant_write_log_space(xlog_t *log, - xlog_ticket_t *tic) -{ - int free_bytes, need_bytes; - xlog_ticket_t *ntic; -#ifdef DEBUG - xfs_lsn_t tail_lsn; -#endif - - tic->t_curr_res = tic->t_unit_res; - xlog_tic_reset_res(tic); - - if (tic->t_cnt > 0) - return 0; - -#ifdef DEBUG - if (log->l_flags & XLOG_ACTIVE_RECOVERY) - panic("regrant Recovery problem"); -#endif - - spin_lock(&log->l_grant_lock); - xlog_trace_loggrant(log, tic, "xlog_regrant_write_log_space: enter"); - - if (XLOG_FORCED_SHUTDOWN(log)) - goto error_return; - - /* If there are other waiters on the queue then give them a - * chance at logspace before us. Wake up the first waiters, - * if we do not wake up all the waiters then go to sleep waiting - * for more free space, otherwise try to get some space for - * this transaction. - */ - - if ((ntic = log->l_write_headq)) { - free_bytes = xlog_space_left(log, log->l_grant_write_cycle, - log->l_grant_write_bytes); - do { - ASSERT(ntic->t_flags & XLOG_TIC_PERM_RESERV); - - if (free_bytes < ntic->t_unit_res) - break; - free_bytes -= ntic->t_unit_res; - sv_signal(&ntic->t_sema); - ntic = ntic->t_next; - } while (ntic != log->l_write_headq); - - if (ntic != log->l_write_headq) { - if ((tic->t_flags & XLOG_TIC_IN_Q) == 0) - xlog_ins_ticketq(&log->l_write_headq, tic); - - xlog_trace_loggrant(log, tic, - "xlog_regrant_write_log_space: sleep 1"); - XFS_STATS_INC(xs_sleep_logspace); - sv_wait(&tic->t_sema, PINOD|PLTWAIT, - &log->l_grant_lock, s); - - /* If we're shutting down, this tic is already - * off the queue */ - if (XLOG_FORCED_SHUTDOWN(log)) { - spin_lock(&log->l_grant_lock); - goto error_return; - } - - xlog_trace_loggrant(log, tic, - "xlog_regrant_write_log_space: wake 1"); - xlog_grant_push_ail(log->l_mp, tic->t_unit_res); - spin_lock(&log->l_grant_lock); - } - } - - need_bytes = tic->t_unit_res; - -redo: - if (XLOG_FORCED_SHUTDOWN(log)) - goto error_return; - - free_bytes = xlog_space_left(log, log->l_grant_write_cycle, - log->l_grant_write_bytes); - if (free_bytes < need_bytes) { - if ((tic->t_flags & XLOG_TIC_IN_Q) == 0) - xlog_ins_ticketq(&log->l_write_headq, tic); - XFS_STATS_INC(xs_sleep_logspace); - sv_wait(&tic->t_sema, PINOD|PLTWAIT, &log->l_grant_lock, s); - - /* If we're shutting down, this tic is already off the queue */ - if (XLOG_FORCED_SHUTDOWN(log)) { - spin_lock(&log->l_grant_lock); - goto error_return; - } - - xlog_trace_loggrant(log, tic, - "xlog_regrant_write_log_space: wake 2"); - xlog_grant_push_ail(log->l_mp, need_bytes); - spin_lock(&log->l_grant_lock); - goto redo; - } else if (tic->t_flags & XLOG_TIC_IN_Q) - xlog_del_ticketq(&log->l_write_headq, tic); - - /* we've got enough space */ - xlog_grant_add_space_write(log, need_bytes); -#ifdef DEBUG - tail_lsn = log->l_tail_lsn; - if (CYCLE_LSN(tail_lsn) != log->l_grant_write_cycle) { - ASSERT(log->l_grant_write_cycle-1 == CYCLE_LSN(tail_lsn)); - ASSERT(log->l_grant_write_bytes <= BBTOB(BLOCK_LSN(tail_lsn))); - } -#endif - - xlog_trace_loggrant(log, tic, "xlog_regrant_write_log_space: exit"); - xlog_verify_grant_head(log, 1); - spin_unlock(&log->l_grant_lock); - return 0; - - - error_return: - if (tic->t_flags & XLOG_TIC_IN_Q) - xlog_del_ticketq(&log->l_reserve_headq, tic); - xlog_trace_loggrant(log, tic, "xlog_regrant_write_log_space: err_ret"); - /* - * If we are failing, make sure the ticket doesn't have any - * current reservations. We don't want to add this back when - * the ticket/transaction gets cancelled. - */ - tic->t_curr_res = 0; - tic->t_cnt = 0; /* ungrant will give back unit_res * t_cnt. */ - spin_unlock(&log->l_grant_lock); - return XFS_ERROR(EIO); -} /* xlog_regrant_write_log_space */ - - /* The first cnt-1 times through here we don't need to * move the grant write head because the permanent * reservation has reserved cnt times the unit amount. @@ -2716,33 +2953,33 @@ redo: * move grant reservation head forward. */ STATIC void -xlog_regrant_reserve_log_space(xlog_t *log, - xlog_ticket_t *ticket) +xlog_regrant_reserve_log_space( + struct xlog *log, + struct xlog_ticket *ticket) { - xlog_trace_loggrant(log, ticket, - "xlog_regrant_reserve_log_space: enter"); + trace_xfs_log_regrant_reserve_enter(log, ticket); + if (ticket->t_cnt > 0) ticket->t_cnt--; - spin_lock(&log->l_grant_lock); - xlog_grant_sub_space(log, ticket->t_curr_res); + xlog_grant_sub_space(log, &log->l_reserve_head.grant, + ticket->t_curr_res); + xlog_grant_sub_space(log, &log->l_write_head.grant, + ticket->t_curr_res); ticket->t_curr_res = ticket->t_unit_res; xlog_tic_reset_res(ticket); - xlog_trace_loggrant(log, ticket, - "xlog_regrant_reserve_log_space: sub current res"); - xlog_verify_grant_head(log, 1); + + trace_xfs_log_regrant_reserve_sub(log, ticket); /* just return if we still have some of the pre-reserved space */ - if (ticket->t_cnt > 0) { - spin_unlock(&log->l_grant_lock); + if (ticket->t_cnt > 0) return; - } - xlog_grant_add_space_reserve(log, ticket->t_unit_res); - xlog_trace_loggrant(log, ticket, - "xlog_regrant_reserve_log_space: exit"); - xlog_verify_grant_head(log, 0); - spin_unlock(&log->l_grant_lock); + xlog_grant_add_space(log, &log->l_reserve_head.grant, + ticket->t_unit_res); + + trace_xfs_log_regrant_reserve_exit(log, ticket); + ticket->t_curr_res = ticket->t_unit_res; xlog_tic_reset_res(ticket); } /* xlog_regrant_reserve_log_space */ @@ -2763,45 +3000,35 @@ xlog_regrant_reserve_log_space(xlog_t *log, * in the current reservation field. */ STATIC void -xlog_ungrant_log_space(xlog_t *log, - xlog_ticket_t *ticket) +xlog_ungrant_log_space( + struct xlog *log, + struct xlog_ticket *ticket) { + int bytes; + if (ticket->t_cnt > 0) ticket->t_cnt--; - spin_lock(&log->l_grant_lock); - xlog_trace_loggrant(log, ticket, "xlog_ungrant_log_space: enter"); - - xlog_grant_sub_space(log, ticket->t_curr_res); + trace_xfs_log_ungrant_enter(log, ticket); + trace_xfs_log_ungrant_sub(log, ticket); - xlog_trace_loggrant(log, ticket, "xlog_ungrant_log_space: sub current"); - - /* If this is a permanent reservation ticket, we may be able to free + /* + * If this is a permanent reservation ticket, we may be able to free * up more space based on the remaining count. */ + bytes = ticket->t_curr_res; if (ticket->t_cnt > 0) { ASSERT(ticket->t_flags & XLOG_TIC_PERM_RESERV); - xlog_grant_sub_space(log, ticket->t_unit_res*ticket->t_cnt); + bytes += ticket->t_unit_res*ticket->t_cnt; } - xlog_trace_loggrant(log, ticket, "xlog_ungrant_log_space: exit"); - xlog_verify_grant_head(log, 1); - spin_unlock(&log->l_grant_lock); - xfs_log_move_tail(log->l_mp, 1); -} /* xlog_ungrant_log_space */ + xlog_grant_sub_space(log, &log->l_reserve_head.grant, bytes); + xlog_grant_sub_space(log, &log->l_write_head.grant, bytes); + trace_xfs_log_ungrant_exit(log, ticket); -/* - * Atomically put back used ticket. - */ -STATIC void -xlog_state_put_ticket(xlog_t *log, - xlog_ticket_t *tic) -{ - spin_lock(&log->l_icloglock); - xlog_ticket_put(log, tic); - spin_unlock(&log->l_icloglock); -} /* xlog_state_put_ticket */ + xfs_log_space_wake(log->l_mp); +} /* * Flush iclog to disk if this is the last reference to the given iclog and @@ -2813,33 +3040,35 @@ xlog_state_put_ticket(xlog_t *log, * */ STATIC int -xlog_state_release_iclog(xlog_t *log, - xlog_in_core_t *iclog) +xlog_state_release_iclog( + struct xlog *log, + struct xlog_in_core *iclog) { int sync = 0; /* do we sync? */ - xlog_assign_tail_lsn(log->l_mp); + if (iclog->ic_state & XLOG_STATE_IOERROR) + return XFS_ERROR(EIO); - spin_lock(&log->l_icloglock); + ASSERT(atomic_read(&iclog->ic_refcnt) > 0); + if (!atomic_dec_and_lock(&iclog->ic_refcnt, &log->l_icloglock)) + return 0; if (iclog->ic_state & XLOG_STATE_IOERROR) { spin_unlock(&log->l_icloglock); return XFS_ERROR(EIO); } - - ASSERT(iclog->ic_refcnt > 0); ASSERT(iclog->ic_state == XLOG_STATE_ACTIVE || iclog->ic_state == XLOG_STATE_WANT_SYNC); - if (--iclog->ic_refcnt == 0 && - iclog->ic_state == XLOG_STATE_WANT_SYNC) { + if (iclog->ic_state == XLOG_STATE_WANT_SYNC) { + /* update tail before writing to iclog */ + xfs_lsn_t tail_lsn = xlog_assign_tail_lsn(log->l_mp); sync++; iclog->ic_state = XLOG_STATE_SYNCING; - iclog->ic_header.h_tail_lsn = cpu_to_be64(log->l_tail_lsn); - xlog_verify_tail_lsn(log, iclog, log->l_tail_lsn); + iclog->ic_header.h_tail_lsn = cpu_to_be64(tail_lsn); + xlog_verify_tail_lsn(log, iclog, tail_lsn); /* cycle incremented when incrementing curr_block */ } - spin_unlock(&log->l_icloglock); /* @@ -2849,11 +3078,9 @@ xlog_state_release_iclog(xlog_t *log, * this iclog has consistent data, so we ignore IOERROR * flags after this point. */ - if (sync) { + if (sync) return xlog_sync(log, iclog); - } return 0; - } /* xlog_state_release_iclog */ @@ -2865,9 +3092,10 @@ xlog_state_release_iclog(xlog_t *log, * that every data block. We have run out of space in this log record. */ STATIC void -xlog_state_switch_iclogs(xlog_t *log, - xlog_in_core_t *iclog, - int eventual_size) +xlog_state_switch_iclogs( + struct xlog *log, + struct xlog_in_core *iclog, + int eventual_size) { ASSERT(iclog->ic_state == XLOG_STATE_ACTIVE); if (!eventual_size) @@ -2881,7 +3109,7 @@ xlog_state_switch_iclogs(xlog_t *log, log->l_curr_block += BTOBB(eventual_size)+BTOBB(log->l_iclog_hsize); /* Round up to next log-sunit */ - if (XFS_SB_VERSION_HASLOGV2(&log->l_mp->m_sb) && + if (xfs_sb_version_haslogv2(&log->l_mp->m_sb) && log->l_mp->m_sb.sb_logsunit > 1) { __uint32_t sunit_bb = BTOBB(log->l_mp->m_sb.sb_logsunit); log->l_curr_block = roundup(log->l_curr_block, sunit_bb); @@ -2898,7 +3126,6 @@ xlog_state_switch_iclogs(xlog_t *log, log->l_iclog = iclog->ic_next; } /* xlog_state_switch_iclogs */ - /* * Write out all data in the in-core log as of this exact moment in time. * @@ -2915,7 +3142,7 @@ xlog_state_switch_iclogs(xlog_t *log, * 2. the current iclog is drity, and the previous iclog is in the * active or dirty state. * - * We may sleep (call psema) if: + * We may sleep if: * * 1. the current iclog is not in the active nor dirty state. * 2. the current iclog dirty, and the previous iclog is not in the @@ -2926,11 +3153,19 @@ xlog_state_switch_iclogs(xlog_t *log, * b) when we return from flushing out this iclog, it is still * not in the active nor dirty state. */ -STATIC int -xlog_state_sync_all(xlog_t *log, uint flags, int *log_flushed) +int +_xfs_log_force( + struct xfs_mount *mp, + uint flags, + int *log_flushed) { - xlog_in_core_t *iclog; - xfs_lsn_t lsn; + struct xlog *log = mp->m_log; + struct xlog_in_core *iclog; + xfs_lsn_t lsn; + + XFS_STATS_INC(xs_log_force); + + xlog_cil_force(log); spin_lock(&log->l_icloglock); @@ -2953,7 +3188,8 @@ xlog_state_sync_all(xlog_t *log, uint flags, int *log_flushed) * previous iclog and go to sleep. */ if (iclog->ic_state == XLOG_STATE_DIRTY || - (iclog->ic_refcnt == 0 && iclog->ic_offset == 0)) { + (atomic_read(&iclog->ic_refcnt) == 0 + && iclog->ic_offset == 0)) { iclog = iclog->ic_prev; if (iclog->ic_state == XLOG_STATE_ACTIVE || iclog->ic_state == XLOG_STATE_DIRTY) @@ -2961,21 +3197,23 @@ xlog_state_sync_all(xlog_t *log, uint flags, int *log_flushed) else goto maybe_sleep; } else { - if (iclog->ic_refcnt == 0) { + if (atomic_read(&iclog->ic_refcnt) == 0) { /* We are the only one with access to this * iclog. Flush it out now. There should * be a roundoff of zero to show that someone * has already taken care of the roundoff from * the previous sync. */ - iclog->ic_refcnt++; + atomic_inc(&iclog->ic_refcnt); lsn = be64_to_cpu(iclog->ic_header.h_lsn); xlog_state_switch_iclogs(log, iclog, 0); spin_unlock(&log->l_icloglock); if (xlog_state_release_iclog(log, iclog)) return XFS_ERROR(EIO); - *log_flushed = 1; + + if (log_flushed) + *log_flushed = 1; spin_lock(&log->l_icloglock); if (be64_to_cpu(iclog->ic_header.h_lsn) == lsn && iclog->ic_state != XLOG_STATE_DIRTY) @@ -3011,7 +3249,7 @@ maybe_sleep: return XFS_ERROR(EIO); } XFS_STATS_INC(xs_log_force_sleep); - sv_wait(&iclog->ic_forcesema, PINOD, &log->l_icloglock, s); + xlog_wait(&iclog->ic_force_wait, &log->l_icloglock); /* * No need to grab the log lock here since we're * only deciding whether or not to return EIO @@ -3019,19 +3257,36 @@ maybe_sleep: */ if (iclog->ic_state & XLOG_STATE_IOERROR) return XFS_ERROR(EIO); - *log_flushed = 1; - + if (log_flushed) + *log_flushed = 1; } else { no_sleep: spin_unlock(&log->l_icloglock); } return 0; -} /* xlog_state_sync_all */ +} + +/* + * Wrapper for _xfs_log_force(), to be used when caller doesn't care + * about errors or whether the log was flushed or not. This is the normal + * interface to use when trying to unpin items or move the log forward. + */ +void +xfs_log_force( + xfs_mount_t *mp, + uint flags) +{ + int error; + trace_xfs_log_force(mp, 0); + error = _xfs_log_force(mp, flags, NULL); + if (error) + xfs_warn(mp, "%s: error %d returned.", __func__, error); +} /* - * Used by code which implements synchronous log forces. + * Force the in-core log to disk for a specific LSN. * * Find in-core log with lsn. * If it is in the DIRTY state, just return. @@ -3039,118 +3294,156 @@ no_sleep: * state and go to sleep or return. * If it is in any other state, go to sleep or return. * - * If filesystem activity goes to zero, the iclog will get flushed only by - * bdflush(). + * Synchronous forces are implemented with a signal variable. All callers + * to force a given lsn to disk will wait on a the sv attached to the + * specific in-core log. When given in-core log finally completes its + * write to disk, that thread will wake up all threads waiting on the + * sv. */ -STATIC int -xlog_state_sync(xlog_t *log, - xfs_lsn_t lsn, - uint flags, - int *log_flushed) +int +_xfs_log_force_lsn( + struct xfs_mount *mp, + xfs_lsn_t lsn, + uint flags, + int *log_flushed) { - xlog_in_core_t *iclog; - int already_slept = 0; + struct xlog *log = mp->m_log; + struct xlog_in_core *iclog; + int already_slept = 0; -try_again: - spin_lock(&log->l_icloglock); - iclog = log->l_iclog; + ASSERT(lsn != 0); - if (iclog->ic_state & XLOG_STATE_IOERROR) { - spin_unlock(&log->l_icloglock); - return XFS_ERROR(EIO); - } + XFS_STATS_INC(xs_log_force); - do { - if (be64_to_cpu(iclog->ic_header.h_lsn) != lsn) { - iclog = iclog->ic_next; - continue; - } + lsn = xlog_cil_force_lsn(log, lsn); + if (lsn == NULLCOMMITLSN) + return 0; - if (iclog->ic_state == XLOG_STATE_DIRTY) { +try_again: + spin_lock(&log->l_icloglock); + iclog = log->l_iclog; + if (iclog->ic_state & XLOG_STATE_IOERROR) { spin_unlock(&log->l_icloglock); - return 0; + return XFS_ERROR(EIO); } - if (iclog->ic_state == XLOG_STATE_ACTIVE) { - /* - * We sleep here if we haven't already slept (e.g. - * this is the first time we've looked at the correct - * iclog buf) and the buffer before us is going to - * be sync'ed. The reason for this is that if we - * are doing sync transactions here, by waiting for - * the previous I/O to complete, we can allow a few - * more transactions into this iclog before we close - * it down. - * - * Otherwise, we mark the buffer WANT_SYNC, and bump - * up the refcnt so we can release the log (which drops - * the ref count). The state switch keeps new transaction - * commits from using this buffer. When the current commits - * finish writing into the buffer, the refcount will drop to - * zero and the buffer will go out then. - */ - if (!already_slept && - (iclog->ic_prev->ic_state & (XLOG_STATE_WANT_SYNC | - XLOG_STATE_SYNCING))) { - ASSERT(!(iclog->ic_state & XLOG_STATE_IOERROR)); - XFS_STATS_INC(xs_log_force_sleep); - sv_wait(&iclog->ic_prev->ic_writesema, PSWP, - &log->l_icloglock, s); - *log_flushed = 1; - already_slept = 1; - goto try_again; - } else { - iclog->ic_refcnt++; + do { + if (be64_to_cpu(iclog->ic_header.h_lsn) != lsn) { + iclog = iclog->ic_next; + continue; + } + + if (iclog->ic_state == XLOG_STATE_DIRTY) { + spin_unlock(&log->l_icloglock); + return 0; + } + + if (iclog->ic_state == XLOG_STATE_ACTIVE) { + /* + * We sleep here if we haven't already slept (e.g. + * this is the first time we've looked at the correct + * iclog buf) and the buffer before us is going to + * be sync'ed. The reason for this is that if we + * are doing sync transactions here, by waiting for + * the previous I/O to complete, we can allow a few + * more transactions into this iclog before we close + * it down. + * + * Otherwise, we mark the buffer WANT_SYNC, and bump + * up the refcnt so we can release the log (which + * drops the ref count). The state switch keeps new + * transaction commits from using this buffer. When + * the current commits finish writing into the buffer, + * the refcount will drop to zero and the buffer will + * go out then. + */ + if (!already_slept && + (iclog->ic_prev->ic_state & + (XLOG_STATE_WANT_SYNC | XLOG_STATE_SYNCING))) { + ASSERT(!(iclog->ic_state & XLOG_STATE_IOERROR)); + + XFS_STATS_INC(xs_log_force_sleep); + + xlog_wait(&iclog->ic_prev->ic_write_wait, + &log->l_icloglock); + if (log_flushed) + *log_flushed = 1; + already_slept = 1; + goto try_again; + } + atomic_inc(&iclog->ic_refcnt); xlog_state_switch_iclogs(log, iclog, 0); spin_unlock(&log->l_icloglock); if (xlog_state_release_iclog(log, iclog)) return XFS_ERROR(EIO); - *log_flushed = 1; + if (log_flushed) + *log_flushed = 1; spin_lock(&log->l_icloglock); } - } - if ((flags & XFS_LOG_SYNC) && /* sleep */ - !(iclog->ic_state & (XLOG_STATE_ACTIVE | XLOG_STATE_DIRTY))) { + if ((flags & XFS_LOG_SYNC) && /* sleep */ + !(iclog->ic_state & + (XLOG_STATE_ACTIVE | XLOG_STATE_DIRTY))) { + /* + * Don't wait on completion if we know that we've + * gotten a log write error. + */ + if (iclog->ic_state & XLOG_STATE_IOERROR) { + spin_unlock(&log->l_icloglock); + return XFS_ERROR(EIO); + } + XFS_STATS_INC(xs_log_force_sleep); + xlog_wait(&iclog->ic_force_wait, &log->l_icloglock); + /* + * No need to grab the log lock here since we're + * only deciding whether or not to return EIO + * and the memory read should be atomic. + */ + if (iclog->ic_state & XLOG_STATE_IOERROR) + return XFS_ERROR(EIO); - /* - * Don't wait on the forcesema if we know that we've - * gotten a log write error. - */ - if (iclog->ic_state & XLOG_STATE_IOERROR) { + if (log_flushed) + *log_flushed = 1; + } else { /* just return */ spin_unlock(&log->l_icloglock); - return XFS_ERROR(EIO); } - XFS_STATS_INC(xs_log_force_sleep); - sv_wait(&iclog->ic_forcesema, PSWP, &log->l_icloglock, s); - /* - * No need to grab the log lock here since we're - * only deciding whether or not to return EIO - * and the memory read should be atomic. - */ - if (iclog->ic_state & XLOG_STATE_IOERROR) - return XFS_ERROR(EIO); - *log_flushed = 1; - } else { /* just return */ - spin_unlock(&log->l_icloglock); - } - return 0; - } while (iclog != log->l_iclog); + return 0; + } while (iclog != log->l_iclog); - spin_unlock(&log->l_icloglock); - return 0; -} /* xlog_state_sync */ + spin_unlock(&log->l_icloglock); + return 0; +} +/* + * Wrapper for _xfs_log_force_lsn(), to be used when caller doesn't care + * about errors or whether the log was flushed or not. This is the normal + * interface to use when trying to unpin items or move the log forward. + */ +void +xfs_log_force_lsn( + xfs_mount_t *mp, + xfs_lsn_t lsn, + uint flags) +{ + int error; + + trace_xfs_log_force(mp, lsn); + error = _xfs_log_force_lsn(mp, lsn, flags, NULL); + if (error) + xfs_warn(mp, "%s: error %d returned.", __func__, error); +} /* * Called when we want to mark the current iclog as being ready to sync to * disk. */ STATIC void -xlog_state_want_sync(xlog_t *log, xlog_in_core_t *iclog) +xlog_state_want_sync( + struct xlog *log, + struct xlog_in_core *iclog) { - spin_lock(&log->l_icloglock); + assert_spin_locked(&log->l_icloglock); if (iclog->ic_state == XLOG_STATE_ACTIVE) { xlog_state_switch_iclogs(log, iclog, 0); @@ -3158,10 +3451,7 @@ xlog_state_want_sync(xlog_t *log, xlog_in_core_t *iclog) ASSERT(iclog->ic_state & (XLOG_STATE_WANT_SYNC|XLOG_STATE_IOERROR)); } - - spin_unlock(&log->l_icloglock); -} /* xlog_state_want_sync */ - +} /***************************************************************************** @@ -3172,118 +3462,38 @@ xlog_state_want_sync(xlog_t *log, xlog_in_core_t *iclog) */ /* - * Algorithm doesn't take into account page size. ;-( + * Free a used ticket when its refcount falls to zero. */ -STATIC void -xlog_state_ticket_alloc(xlog_t *log) +void +xfs_log_ticket_put( + xlog_ticket_t *ticket) { - xlog_ticket_t *t_list; - xlog_ticket_t *next; - xfs_caddr_t buf; - uint i = (PAGE_SIZE / sizeof(xlog_ticket_t)) - 2; - - /* - * The kmem_zalloc may sleep, so we shouldn't be holding the - * global lock. XXXmiken: may want to use zone allocator. - */ - buf = (xfs_caddr_t) kmem_zalloc(PAGE_SIZE, KM_SLEEP); - - spin_lock(&log->l_icloglock); - - /* Attach 1st ticket to Q, so we can keep track of allocated memory */ - t_list = (xlog_ticket_t *)buf; - t_list->t_next = log->l_unmount_free; - log->l_unmount_free = t_list++; - log->l_ticket_cnt++; - log->l_ticket_tcnt++; - - /* Next ticket becomes first ticket attached to ticket free list */ - if (log->l_freelist != NULL) { - ASSERT(log->l_tail != NULL); - log->l_tail->t_next = t_list; - } else { - log->l_freelist = t_list; - } - log->l_ticket_cnt++; - log->l_ticket_tcnt++; - - /* Cycle through rest of alloc'ed memory, building up free Q */ - for ( ; i > 0; i--) { - next = t_list + 1; - t_list->t_next = next; - t_list = next; - log->l_ticket_cnt++; - log->l_ticket_tcnt++; - } - t_list->t_next = NULL; - log->l_tail = t_list; - spin_unlock(&log->l_icloglock); -} /* xlog_state_ticket_alloc */ - + ASSERT(atomic_read(&ticket->t_ref) > 0); + if (atomic_dec_and_test(&ticket->t_ref)) + kmem_zone_free(xfs_log_ticket_zone, ticket); +} -/* - * Put ticket into free list - * - * Assumption: log lock is held around this call. - */ -STATIC void -xlog_ticket_put(xlog_t *log, - xlog_ticket_t *ticket) +xlog_ticket_t * +xfs_log_ticket_get( + xlog_ticket_t *ticket) { - sv_destroy(&ticket->t_sema); - - /* - * Don't think caching will make that much difference. It's - * more important to make debug easier. - */ -#if 0 - /* real code will want to use LIFO for caching */ - ticket->t_next = log->l_freelist; - log->l_freelist = ticket; - /* no need to clear fields */ -#else - /* When we debug, it is easier if tickets are cycled */ - ticket->t_next = NULL; - if (log->l_tail) { - log->l_tail->t_next = ticket; - } else { - ASSERT(log->l_freelist == NULL); - log->l_freelist = ticket; - } - log->l_tail = ticket; -#endif /* DEBUG */ - log->l_ticket_cnt++; -} /* xlog_ticket_put */ - + ASSERT(atomic_read(&ticket->t_ref) > 0); + atomic_inc(&ticket->t_ref); + return ticket; +} /* - * Grab ticket off freelist or allocation some more + * Figure out the total log space unit (in bytes) that would be + * required for a log ticket. */ -STATIC xlog_ticket_t * -xlog_ticket_get(xlog_t *log, - int unit_bytes, - int cnt, - char client, - uint xflags) +int +xfs_log_calc_unit_res( + struct xfs_mount *mp, + int unit_bytes) { - xlog_ticket_t *tic; - uint num_headers; - - alloc: - if (log->l_freelist == NULL) - xlog_state_ticket_alloc(log); /* potentially sleep */ - - spin_lock(&log->l_icloglock); - if (log->l_freelist == NULL) { - spin_unlock(&log->l_icloglock); - goto alloc; - } - tic = log->l_freelist; - log->l_freelist = tic->t_next; - if (log->l_freelist == NULL) - log->l_tail = NULL; - log->l_ticket_cnt--; - spin_unlock(&log->l_icloglock); + struct xlog *log = mp->m_log; + int iclog_space; + uint num_headers; /* * Permanent reservations have up to 'cnt'-1 active log operations @@ -3323,42 +3533,91 @@ xlog_ticket_get(xlog_t *log, /* for start-rec */ unit_bytes += sizeof(xlog_op_header_t); - /* for LR headers */ - num_headers = ((unit_bytes + log->l_iclog_size-1) >> log->l_iclog_size_log); + /* + * for LR headers - the space for data in an iclog is the size minus + * the space used for the headers. If we use the iclog size, then we + * undercalculate the number of headers required. + * + * Furthermore - the addition of op headers for split-recs might + * increase the space required enough to require more log and op + * headers, so take that into account too. + * + * IMPORTANT: This reservation makes the assumption that if this + * transaction is the first in an iclog and hence has the LR headers + * accounted to it, then the remaining space in the iclog is + * exclusively for this transaction. i.e. if the transaction is larger + * than the iclog, it will be the only thing in that iclog. + * Fundamentally, this means we must pass the entire log vector to + * xlog_write to guarantee this. + */ + iclog_space = log->l_iclog_size - log->l_iclog_hsize; + num_headers = howmany(unit_bytes, iclog_space); + + /* for split-recs - ophdrs added when data split over LRs */ + unit_bytes += sizeof(xlog_op_header_t) * num_headers; + + /* add extra header reservations if we overrun */ + while (!num_headers || + howmany(unit_bytes, iclog_space) > num_headers) { + unit_bytes += sizeof(xlog_op_header_t); + num_headers++; + } unit_bytes += log->l_iclog_hsize * num_headers; /* for commit-rec LR header - note: padding will subsume the ophdr */ unit_bytes += log->l_iclog_hsize; - /* for split-recs - ophdrs added when data split over LRs */ - unit_bytes += sizeof(xlog_op_header_t) * num_headers; - /* for roundoff padding for transaction data and one for commit record */ - if (XFS_SB_VERSION_HASLOGV2(&log->l_mp->m_sb) && - log->l_mp->m_sb.sb_logsunit > 1) { + if (xfs_sb_version_haslogv2(&mp->m_sb) && mp->m_sb.sb_logsunit > 1) { /* log su roundoff */ - unit_bytes += 2*log->l_mp->m_sb.sb_logsunit; + unit_bytes += 2 * mp->m_sb.sb_logsunit; } else { /* BB roundoff */ - unit_bytes += 2*BBSIZE; + unit_bytes += 2 * BBSIZE; } - tic->t_unit_res = unit_bytes; - tic->t_curr_res = unit_bytes; + return unit_bytes; +} + +/* + * Allocate and initialise a new log ticket. + */ +struct xlog_ticket * +xlog_ticket_alloc( + struct xlog *log, + int unit_bytes, + int cnt, + char client, + bool permanent, + xfs_km_flags_t alloc_flags) +{ + struct xlog_ticket *tic; + int unit_res; + + tic = kmem_zone_zalloc(xfs_log_ticket_zone, alloc_flags); + if (!tic) + return NULL; + + unit_res = xfs_log_calc_unit_res(log->l_mp, unit_bytes); + + atomic_set(&tic->t_ref, 1); + tic->t_task = current; + INIT_LIST_HEAD(&tic->t_queue); + tic->t_unit_res = unit_res; + tic->t_curr_res = unit_res; tic->t_cnt = cnt; tic->t_ocnt = cnt; - tic->t_tid = (xlog_tid_t)((__psint_t)tic & 0xffffffff); + tic->t_tid = prandom_u32(); tic->t_clientid = client; tic->t_flags = XLOG_TIC_INITED; tic->t_trans_type = 0; - if (xflags & XFS_LOG_PERM_RESERV) + if (permanent) tic->t_flags |= XLOG_TIC_PERM_RESERV; - sv_init(&(tic->t_sema), SV_DEFAULT, "logtick"); xlog_tic_reset_res(tic); return tic; -} /* xlog_ticket_get */ +} /****************************************************************************** @@ -3374,40 +3633,66 @@ xlog_ticket_get(xlog_t *log, * part of the log in case we trash the log structure. */ void -xlog_verify_dest_ptr(xlog_t *log, - __psint_t ptr) +xlog_verify_dest_ptr( + struct xlog *log, + char *ptr) { int i; int good_ptr = 0; - for (i=0; i < log->l_iclog_bufs; i++) { - if (ptr >= (__psint_t)log->l_iclog_bak[i] && - ptr <= (__psint_t)log->l_iclog_bak[i]+log->l_iclog_size) + for (i = 0; i < log->l_iclog_bufs; i++) { + if (ptr >= log->l_iclog_bak[i] && + ptr <= log->l_iclog_bak[i] + log->l_iclog_size) good_ptr++; } - if (! good_ptr) - xlog_panic("xlog_verify_dest_ptr: invalid ptr"); -} /* xlog_verify_dest_ptr */ + if (!good_ptr) + xfs_emerg(log->l_mp, "%s: invalid ptr", __func__); +} + +/* + * Check to make sure the grant write head didn't just over lap the tail. If + * the cycles are the same, we can't be overlapping. Otherwise, make sure that + * the cycles differ by exactly one and check the byte count. + * + * This check is run unlocked, so can give false positives. Rather than assert + * on failures, use a warn-once flag and a panic tag to allow the admin to + * determine if they want to panic the machine when such an error occurs. For + * debug kernels this will have the same effect as using an assert but, unlinke + * an assert, it can be turned off at runtime. + */ STATIC void -xlog_verify_grant_head(xlog_t *log, int equals) +xlog_verify_grant_tail( + struct xlog *log) { - if (log->l_grant_reserve_cycle == log->l_grant_write_cycle) { - if (equals) - ASSERT(log->l_grant_reserve_bytes >= log->l_grant_write_bytes); - else - ASSERT(log->l_grant_reserve_bytes > log->l_grant_write_bytes); - } else { - ASSERT(log->l_grant_reserve_cycle-1 == log->l_grant_write_cycle); - ASSERT(log->l_grant_write_bytes >= log->l_grant_reserve_bytes); - } -} /* xlog_verify_grant_head */ + int tail_cycle, tail_blocks; + int cycle, space; + + xlog_crack_grant_head(&log->l_write_head.grant, &cycle, &space); + xlog_crack_atomic_lsn(&log->l_tail_lsn, &tail_cycle, &tail_blocks); + if (tail_cycle != cycle) { + if (cycle - 1 != tail_cycle && + !(log->l_flags & XLOG_TAIL_WARN)) { + xfs_alert_tag(log->l_mp, XFS_PTAG_LOGRES, + "%s: cycle - 1 != tail_cycle", __func__); + log->l_flags |= XLOG_TAIL_WARN; + } + + if (space > BBTOB(tail_blocks) && + !(log->l_flags & XLOG_TAIL_WARN)) { + xfs_alert_tag(log->l_mp, XFS_PTAG_LOGRES, + "%s: space > BBTOB(tail_blocks)", __func__); + log->l_flags |= XLOG_TAIL_WARN; + } + } +} /* check if it will fit */ STATIC void -xlog_verify_tail_lsn(xlog_t *log, - xlog_in_core_t *iclog, - xfs_lsn_t tail_lsn) +xlog_verify_tail_lsn( + struct xlog *log, + struct xlog_in_core *iclog, + xfs_lsn_t tail_lsn) { int blocks; @@ -3415,16 +3700,16 @@ xlog_verify_tail_lsn(xlog_t *log, blocks = log->l_logBBsize - (log->l_prev_block - BLOCK_LSN(tail_lsn)); if (blocks < BTOBB(iclog->ic_offset)+BTOBB(log->l_iclog_hsize)) - xlog_panic("xlog_verify_tail_lsn: ran out of log space"); + xfs_emerg(log->l_mp, "%s: ran out of log space", __func__); } else { ASSERT(CYCLE_LSN(tail_lsn)+1 == log->l_prev_cycle); if (BLOCK_LSN(tail_lsn) == log->l_prev_block) - xlog_panic("xlog_verify_tail_lsn: tail wrapped"); + xfs_emerg(log->l_mp, "%s: tail wrapped", __func__); blocks = BLOCK_LSN(tail_lsn) - log->l_prev_block; if (blocks < BTOBB(iclog->ic_offset) + 1) - xlog_panic("xlog_verify_tail_lsn: ran out of log space"); + xfs_emerg(log->l_mp, "%s: ran out of log space", __func__); } } /* xlog_verify_tail_lsn */ @@ -3444,10 +3729,11 @@ xlog_verify_tail_lsn(xlog_t *log, * the cycle numbers agree with the current cycle number. */ STATIC void -xlog_verify_iclog(xlog_t *log, - xlog_in_core_t *iclog, - int count, - boolean_t syncing) +xlog_verify_iclog( + struct xlog *log, + struct xlog_in_core *iclog, + int count, + bool syncing) { xlog_op_header_t *ophead; xlog_in_core_t *icptr; @@ -3462,24 +3748,23 @@ xlog_verify_iclog(xlog_t *log, /* check validity of iclog pointers */ spin_lock(&log->l_icloglock); icptr = log->l_iclog; - for (i=0; i < log->l_iclog_bufs; i++) { - if (icptr == NULL) - xlog_panic("xlog_verify_iclog: invalid ptr"); - icptr = icptr->ic_next; - } + for (i = 0; i < log->l_iclog_bufs; i++, icptr = icptr->ic_next) + ASSERT(icptr); + if (icptr != log->l_iclog) - xlog_panic("xlog_verify_iclog: corrupt iclog ring"); + xfs_emerg(log->l_mp, "%s: corrupt iclog ring", __func__); spin_unlock(&log->l_icloglock); /* check log magic numbers */ - if (be32_to_cpu(iclog->ic_header.h_magicno) != XLOG_HEADER_MAGIC_NUM) - xlog_panic("xlog_verify_iclog: invalid magic num"); + if (iclog->ic_header.h_magicno != cpu_to_be32(XLOG_HEADER_MAGIC_NUM)) + xfs_emerg(log->l_mp, "%s: invalid magic num", __func__); ptr = (xfs_caddr_t) &iclog->ic_header; for (ptr += BBSIZE; ptr < ((xfs_caddr_t)&iclog->ic_header) + count; ptr += BBSIZE) { - if (be32_to_cpu(*(__be32 *)ptr) == XLOG_HEADER_MAGIC_NUM) - xlog_panic("xlog_verify_iclog: unexpected magic num"); + if (*(__be32 *)ptr == cpu_to_be32(XLOG_HEADER_MAGIC_NUM)) + xfs_emerg(log->l_mp, "%s: unexpected magic num", + __func__); } /* check fields */ @@ -3487,14 +3772,14 @@ xlog_verify_iclog(xlog_t *log, ptr = iclog->ic_datap; base_ptr = ptr; ophead = (xlog_op_header_t *)ptr; - xhdr = (xlog_in_core_2_t *)&iclog->ic_header; + xhdr = iclog->ic_data; for (i = 0; i < len; i++) { ophead = (xlog_op_header_t *)ptr; /* clientid is only 1 byte */ field_offset = (__psint_t) ((xfs_caddr_t)&(ophead->oh_clientid) - base_ptr); - if (syncing == B_FALSE || (field_offset & 0x1ff)) { + if (!syncing || (field_offset & 0x1ff)) { clientid = ophead->oh_clientid; } else { idx = BTOBBT((xfs_caddr_t)&(ophead->oh_clientid) - iclog->ic_datap); @@ -3509,14 +3794,15 @@ xlog_verify_iclog(xlog_t *log, } } if (clientid != XFS_TRANSACTION && clientid != XFS_LOG) - cmn_err(CE_WARN, "xlog_verify_iclog: " - "invalid clientid %d op 0x%p offset 0x%lx", - clientid, ophead, (unsigned long)field_offset); + xfs_warn(log->l_mp, + "%s: invalid clientid %d op 0x%p offset 0x%lx", + __func__, clientid, ophead, + (unsigned long)field_offset); /* check length */ field_offset = (__psint_t) ((xfs_caddr_t)&(ophead->oh_len) - base_ptr); - if (syncing == B_FALSE || (field_offset & 0x1ff)) { + if (!syncing || (field_offset & 0x1ff)) { op_len = be32_to_cpu(ophead->oh_len); } else { idx = BTOBBT((__psint_t)&ophead->oh_len - @@ -3539,7 +3825,7 @@ xlog_verify_iclog(xlog_t *log, */ STATIC int xlog_state_ioerror( - xlog_t *log) + struct xlog *log) { xlog_in_core_t *iclog, *ic; @@ -3573,16 +3859,19 @@ xlog_state_ioerror( * c. nothing new gets queued up after (a) and (b) are done. * d. if !logerror, flush the iclogs to disk, then seal them off * for business. + * + * Note: for delayed logging the !logerror case needs to flush the regions + * held in memory out to the iclogs before flushing them to disk. This needs + * to be done before the log is marked as shutdown, otherwise the flush to the + * iclogs will fail. */ int xfs_log_force_umount( struct xfs_mount *mp, int logerror) { - xlog_ticket_t *tic; - xlog_t *log; + struct xlog *log; int retval; - int dummy; log = mp->m_log; @@ -3593,7 +3882,8 @@ xfs_log_force_umount( if (!log || log->l_flags & XLOG_ACTIVE_RECOVERY) { mp->m_flags |= XFS_MOUNT_FS_SHUTDOWN; - XFS_BUF_DONE(mp->m_sb_bp); + if (mp->m_sb_bp) + XFS_BUF_DONE(mp->m_sb_bp); return 0; } @@ -3606,15 +3896,25 @@ xfs_log_force_umount( return 1; } retval = 0; + + /* + * Flush the in memory commit item list before marking the log as + * being shut down. We need to do it in this order to ensure all the + * completed transactions are flushed to disk with the xfs_log_force() + * call below. + */ + if (!logerror) + xlog_cil_force(log); + /* - * We must hold both the GRANT lock and the LOG lock, - * before we mark the filesystem SHUTDOWN and wake - * everybody up to tell the bad news. + * mark the filesystem and the as in a shutdown state and wake + * everybody up to tell them the bad news. */ - spin_lock(&log->l_grant_lock); spin_lock(&log->l_icloglock); mp->m_flags |= XFS_MOUNT_FS_SHUTDOWN; - XFS_BUF_DONE(mp->m_sb_bp); + if (mp->m_sb_bp) + XFS_BUF_DONE(mp->m_sb_bp); + /* * This flag is sort of redundant because of the mount flag, but * it's good to maintain the separation between the log and the rest @@ -3631,44 +3931,35 @@ xfs_log_force_umount( spin_unlock(&log->l_icloglock); /* - * We don't want anybody waiting for log reservations - * after this. That means we have to wake up everybody - * queued up on reserve_headq as well as write_headq. - * In addition, we make sure in xlog_{re}grant_log_space - * that we don't enqueue anything once the SHUTDOWN flag - * is set, and this action is protected by the GRANTLOCK. + * We don't want anybody waiting for log reservations after this. That + * means we have to wake up everybody queued up on reserveq as well as + * writeq. In addition, we make sure in xlog_{re}grant_log_space that + * we don't enqueue anything once the SHUTDOWN flag is set, and this + * action is protected by the grant locks. */ - if ((tic = log->l_reserve_headq)) { - do { - sv_signal(&tic->t_sema); - tic = tic->t_next; - } while (tic != log->l_reserve_headq); - } - - if ((tic = log->l_write_headq)) { - do { - sv_signal(&tic->t_sema); - tic = tic->t_next; - } while (tic != log->l_write_headq); - } - spin_unlock(&log->l_grant_lock); + xlog_grant_head_wake_all(&log->l_reserve_head); + xlog_grant_head_wake_all(&log->l_write_head); - if (! (log->l_iclog->ic_state & XLOG_STATE_IOERROR)) { + if (!(log->l_iclog->ic_state & XLOG_STATE_IOERROR)) { ASSERT(!logerror); /* * Force the incore logs to disk before shutting the * log down completely. */ - xlog_state_sync_all(log, XFS_LOG_FORCE|XFS_LOG_SYNC, &dummy); + _xfs_log_force(mp, XFS_LOG_SYNC, NULL); + spin_lock(&log->l_icloglock); retval = xlog_state_ioerror(log); spin_unlock(&log->l_icloglock); } + /* - * Wake up everybody waiting on xfs_log_force. - * Callback all log item committed functions as if the - * log writes were completed. + * Wake up everybody waiting on xfs_log_force. Wake the CIL push first + * as if the log writes were completed. The abort handling in the log + * item committed callback functions will do this again under lock to + * avoid races. */ + wake_up_all(&log->l_cilp->xc_commit_wait); xlog_state_do_callback(log, XFS_LI_ABORTED, NULL); #ifdef XFSERRORDEBUG @@ -3689,7 +3980,8 @@ xfs_log_force_umount( } STATIC int -xlog_iclogs_empty(xlog_t *log) +xlog_iclogs_empty( + struct xlog *log) { xlog_in_core_t *iclog; @@ -3704,3 +3996,4 @@ xlog_iclogs_empty(xlog_t *log) } while (iclog != log->l_iclog); return 1; } + |
