brintos

brintos / linux-shallow public Read only

0
0
Text · 100.9 KiB · 26b2f58 Raw
3597 lines · c
1// SPDX-License-Identifier: GPL-2.02/*3 * Copyright (c) 2000-2005 Silicon Graphics, Inc.4 * All Rights Reserved.5 */6#include "xfs.h"7#include "xfs_fs.h"8#include "xfs_shared.h"9#include "xfs_format.h"10#include "xfs_log_format.h"11#include "xfs_trans_resv.h"12#include "xfs_mount.h"13#include "xfs_errortag.h"14#include "xfs_error.h"15#include "xfs_trans.h"16#include "xfs_trans_priv.h"17#include "xfs_log.h"18#include "xfs_log_priv.h"19#include "xfs_trace.h"20#include "xfs_sysfs.h"21#include "xfs_sb.h"22#include "xfs_health.h"23 24struct kmem_cache	*xfs_log_ticket_cache;25 26/* Local miscellaneous function prototypes */27STATIC struct xlog *28xlog_alloc_log(29	struct xfs_mount	*mp,30	struct xfs_buftarg	*log_target,31	xfs_daddr_t		blk_offset,32	int			num_bblks);33STATIC void34xlog_dealloc_log(35	struct xlog		*log);36 37/* local state machine functions */38STATIC void xlog_state_done_syncing(39	struct xlog_in_core	*iclog);40STATIC void xlog_state_do_callback(41	struct xlog		*log);42STATIC int43xlog_state_get_iclog_space(44	struct xlog		*log,45	int			len,46	struct xlog_in_core	**iclog,47	struct xlog_ticket	*ticket,48	int			*logoffsetp);49STATIC void50xlog_sync(51	struct xlog		*log,52	struct xlog_in_core	*iclog,53	struct xlog_ticket	*ticket);54#if defined(DEBUG)55STATIC void56xlog_verify_iclog(57	struct xlog		*log,58	struct xlog_in_core	*iclog,59	int			count);60STATIC void61xlog_verify_tail_lsn(62	struct xlog		*log,63	struct xlog_in_core	*iclog);64#else65#define xlog_verify_iclog(a,b,c)66#define xlog_verify_tail_lsn(a,b)67#endif68 69STATIC int70xlog_iclogs_empty(71	struct xlog		*log);72 73static int74xfs_log_cover(struct xfs_mount *);75 76/*77 * We need to make sure the buffer pointer returned is naturally aligned for the78 * biggest basic data type we put into it. We have already accounted for this79 * padding when sizing the buffer.80 *81 * However, this padding does not get written into the log, and hence we have to82 * track the space used by the log vectors separately to prevent log space hangs83 * due to inaccurate accounting (i.e. a leak) of the used log space through the84 * CIL context ticket.85 *86 * We also add space for the xlog_op_header that describes this region in the87 * log. This prepends the data region we return to the caller to copy their data88 * into, so do all the static initialisation of the ophdr now. Because the ophdr89 * is not 8 byte aligned, we have to be careful to ensure that we align the90 * start of the buffer such that the region we return to the call is 8 byte91 * aligned and packed against the tail of the ophdr.92 */93void *94xlog_prepare_iovec(95	struct xfs_log_vec	*lv,96	struct xfs_log_iovec	**vecp,97	uint			type)98{99	struct xfs_log_iovec	*vec = *vecp;100	struct xlog_op_header	*oph;101	uint32_t		len;102	void			*buf;103 104	if (vec) {105		ASSERT(vec - lv->lv_iovecp < lv->lv_niovecs);106		vec++;107	} else {108		vec = &lv->lv_iovecp[0];109	}110 111	len = lv->lv_buf_len + sizeof(struct xlog_op_header);112	if (!IS_ALIGNED(len, sizeof(uint64_t))) {113		lv->lv_buf_len = round_up(len, sizeof(uint64_t)) -114					sizeof(struct xlog_op_header);115	}116 117	vec->i_type = type;118	vec->i_addr = lv->lv_buf + lv->lv_buf_len;119 120	oph = vec->i_addr;121	oph->oh_clientid = XFS_TRANSACTION;122	oph->oh_res2 = 0;123	oph->oh_flags = 0;124 125	buf = vec->i_addr + sizeof(struct xlog_op_header);126	ASSERT(IS_ALIGNED((unsigned long)buf, sizeof(uint64_t)));127 128	*vecp = vec;129	return buf;130}131 132static inline void133xlog_grant_sub_space(134	struct xlog_grant_head	*head,135	int64_t			bytes)136{137	atomic64_sub(bytes, &head->grant);138}139 140static inline void141xlog_grant_add_space(142	struct xlog_grant_head	*head,143	int64_t			bytes)144{145	atomic64_add(bytes, &head->grant);146}147 148static void149xlog_grant_head_init(150	struct xlog_grant_head	*head)151{152	atomic64_set(&head->grant, 0);153	INIT_LIST_HEAD(&head->waiters);154	spin_lock_init(&head->lock);155}156 157void158xlog_grant_return_space(159	struct xlog	*log,160	xfs_lsn_t	old_head,161	xfs_lsn_t	new_head)162{163	int64_t		diff = xlog_lsn_sub(log, new_head, old_head);164 165	xlog_grant_sub_space(&log->l_reserve_head, diff);166	xlog_grant_sub_space(&log->l_write_head, diff);167}168 169/*170 * Return the space in the log between the tail and the head.  In the case where171 * we have overrun available reservation space, return 0. The memory barrier172 * pairs with the smp_wmb() in xlog_cil_ail_insert() to ensure that grant head173 * vs tail space updates are seen in the correct order and hence avoid174 * transients as space is transferred from the grant heads to the AIL on commit175 * completion.176 */177static uint64_t178xlog_grant_space_left(179	struct xlog		*log,180	struct xlog_grant_head	*head)181{182	int64_t			free_bytes;183 184	smp_rmb();	/* paired with smp_wmb in xlog_cil_ail_insert() */185	free_bytes = log->l_logsize - READ_ONCE(log->l_tail_space) -186			atomic64_read(&head->grant);187	if (free_bytes > 0)188		return free_bytes;189	return 0;190}191 192STATIC void193xlog_grant_head_wake_all(194	struct xlog_grant_head	*head)195{196	struct xlog_ticket	*tic;197 198	spin_lock(&head->lock);199	list_for_each_entry(tic, &head->waiters, t_queue)200		wake_up_process(tic->t_task);201	spin_unlock(&head->lock);202}203 204static inline int205xlog_ticket_reservation(206	struct xlog		*log,207	struct xlog_grant_head	*head,208	struct xlog_ticket	*tic)209{210	if (head == &log->l_write_head) {211		ASSERT(tic->t_flags & XLOG_TIC_PERM_RESERV);212		return tic->t_unit_res;213	}214 215	if (tic->t_flags & XLOG_TIC_PERM_RESERV)216		return tic->t_unit_res * tic->t_cnt;217 218	return tic->t_unit_res;219}220 221STATIC bool222xlog_grant_head_wake(223	struct xlog		*log,224	struct xlog_grant_head	*head,225	int			*free_bytes)226{227	struct xlog_ticket	*tic;228	int			need_bytes;229 230	list_for_each_entry(tic, &head->waiters, t_queue) {231		need_bytes = xlog_ticket_reservation(log, head, tic);232		if (*free_bytes < need_bytes)233			return false;234 235		*free_bytes -= need_bytes;236		trace_xfs_log_grant_wake_up(log, tic);237		wake_up_process(tic->t_task);238	}239 240	return true;241}242 243STATIC int244xlog_grant_head_wait(245	struct xlog		*log,246	struct xlog_grant_head	*head,247	struct xlog_ticket	*tic,248	int			need_bytes) __releases(&head->lock)249					    __acquires(&head->lock)250{251	list_add_tail(&tic->t_queue, &head->waiters);252 253	do {254		if (xlog_is_shutdown(log))255			goto shutdown;256 257		__set_current_state(TASK_UNINTERRUPTIBLE);258		spin_unlock(&head->lock);259 260		XFS_STATS_INC(log->l_mp, xs_sleep_logspace);261 262		/* Push on the AIL to free up all the log space. */263		xfs_ail_push_all(log->l_ailp);264 265		trace_xfs_log_grant_sleep(log, tic);266		schedule();267		trace_xfs_log_grant_wake(log, tic);268 269		spin_lock(&head->lock);270		if (xlog_is_shutdown(log))271			goto shutdown;272	} while (xlog_grant_space_left(log, head) < need_bytes);273 274	list_del_init(&tic->t_queue);275	return 0;276shutdown:277	list_del_init(&tic->t_queue);278	return -EIO;279}280 281/*282 * Atomically get the log space required for a log ticket.283 *284 * Once a ticket gets put onto head->waiters, it will only return after the285 * needed reservation is satisfied.286 *287 * This function is structured so that it has a lock free fast path. This is288 * necessary because every new transaction reservation will come through this289 * path. Hence any lock will be globally hot if we take it unconditionally on290 * every pass.291 *292 * As tickets are only ever moved on and off head->waiters under head->lock, we293 * only need to take that lock if we are going to add the ticket to the queue294 * and sleep. We can avoid taking the lock if the ticket was never added to295 * head->waiters because the t_queue list head will be empty and we hold the296 * only reference to it so it can safely be checked unlocked.297 */298STATIC int299xlog_grant_head_check(300	struct xlog		*log,301	struct xlog_grant_head	*head,302	struct xlog_ticket	*tic,303	int			*need_bytes)304{305	int			free_bytes;306	int			error = 0;307 308	ASSERT(!xlog_in_recovery(log));309 310	/*311	 * If there are other waiters on the queue then give them a chance at312	 * logspace before us.  Wake up the first waiters, if we do not wake313	 * up all the waiters then go to sleep waiting for more free space,314	 * otherwise try to get some space for this transaction.315	 */316	*need_bytes = xlog_ticket_reservation(log, head, tic);317	free_bytes = xlog_grant_space_left(log, head);318	if (!list_empty_careful(&head->waiters)) {319		spin_lock(&head->lock);320		if (!xlog_grant_head_wake(log, head, &free_bytes) ||321		    free_bytes < *need_bytes) {322			error = xlog_grant_head_wait(log, head, tic,323						     *need_bytes);324		}325		spin_unlock(&head->lock);326	} else if (free_bytes < *need_bytes) {327		spin_lock(&head->lock);328		error = xlog_grant_head_wait(log, head, tic, *need_bytes);329		spin_unlock(&head->lock);330	}331 332	return error;333}334 335bool336xfs_log_writable(337	struct xfs_mount	*mp)338{339	/*340	 * Do not write to the log on norecovery mounts, if the data or log341	 * devices are read-only, or if the filesystem is shutdown. Read-only342	 * mounts allow internal writes for log recovery and unmount purposes,343	 * so don't restrict that case.344	 */345	if (xfs_has_norecovery(mp))346		return false;347	if (xfs_readonly_buftarg(mp->m_ddev_targp))348		return false;349	if (xfs_readonly_buftarg(mp->m_log->l_targ))350		return false;351	if (xlog_is_shutdown(mp->m_log))352		return false;353	return true;354}355 356/*357 * Replenish the byte reservation required by moving the grant write head.358 */359int360xfs_log_regrant(361	struct xfs_mount	*mp,362	struct xlog_ticket	*tic)363{364	struct xlog		*log = mp->m_log;365	int			need_bytes;366	int			error = 0;367 368	if (xlog_is_shutdown(log))369		return -EIO;370 371	XFS_STATS_INC(mp, xs_try_logspace);372 373	/*374	 * This is a new transaction on the ticket, so we need to change the375	 * transaction ID so that the next transaction has a different TID in376	 * the log. Just add one to the existing tid so that we can see chains377	 * of rolling transactions in the log easily.378	 */379	tic->t_tid++;380	tic->t_curr_res = tic->t_unit_res;381	if (tic->t_cnt > 0)382		return 0;383 384	trace_xfs_log_regrant(log, tic);385 386	error = xlog_grant_head_check(log, &log->l_write_head, tic,387				      &need_bytes);388	if (error)389		goto out_error;390 391	xlog_grant_add_space(&log->l_write_head, need_bytes);392	trace_xfs_log_regrant_exit(log, tic);393	return 0;394 395out_error:396	/*397	 * If we are failing, make sure the ticket doesn't have any current398	 * reservations.  We don't want to add this back when the ticket/399	 * transaction gets cancelled.400	 */401	tic->t_curr_res = 0;402	tic->t_cnt = 0;	/* ungrant will give back unit_res * t_cnt. */403	return error;404}405 406/*407 * Reserve log space and return a ticket corresponding to the reservation.408 *409 * Each reservation is going to reserve extra space for a log record header.410 * When writes happen to the on-disk log, we don't subtract the length of the411 * log record header from any reservation.  By wasting space in each412 * reservation, we prevent over allocation problems.413 */414int415xfs_log_reserve(416	struct xfs_mount	*mp,417	int			unit_bytes,418	int			cnt,419	struct xlog_ticket	**ticp,420	bool			permanent)421{422	struct xlog		*log = mp->m_log;423	struct xlog_ticket	*tic;424	int			need_bytes;425	int			error = 0;426 427	if (xlog_is_shutdown(log))428		return -EIO;429 430	XFS_STATS_INC(mp, xs_try_logspace);431 432	ASSERT(*ticp == NULL);433	tic = xlog_ticket_alloc(log, unit_bytes, cnt, permanent);434	*ticp = tic;435	trace_xfs_log_reserve(log, tic);436	error = xlog_grant_head_check(log, &log->l_reserve_head, tic,437				      &need_bytes);438	if (error)439		goto out_error;440 441	xlog_grant_add_space(&log->l_reserve_head, need_bytes);442	xlog_grant_add_space(&log->l_write_head, need_bytes);443	trace_xfs_log_reserve_exit(log, tic);444	return 0;445 446out_error:447	/*448	 * If we are failing, make sure the ticket doesn't have any current449	 * reservations.  We don't want to add this back when the ticket/450	 * transaction gets cancelled.451	 */452	tic->t_curr_res = 0;453	tic->t_cnt = 0;	/* ungrant will give back unit_res * t_cnt. */454	return error;455}456 457/*458 * Run all the pending iclog callbacks and wake log force waiters and iclog459 * space waiters so they can process the newly set shutdown state. We really460 * don't care what order we process callbacks here because the log is shut down461 * and so state cannot change on disk anymore. However, we cannot wake waiters462 * until the callbacks have been processed because we may be in unmount and463 * we must ensure that all AIL operations the callbacks perform have completed464 * before we tear down the AIL.465 *466 * We avoid processing actively referenced iclogs so that we don't run callbacks467 * while the iclog owner might still be preparing the iclog for IO submssion.468 * These will be caught by xlog_state_iclog_release() and call this function469 * again to process any callbacks that may have been added to that iclog.470 */471static void472xlog_state_shutdown_callbacks(473	struct xlog		*log)474{475	struct xlog_in_core	*iclog;476	LIST_HEAD(cb_list);477 478	iclog = log->l_iclog;479	do {480		if (atomic_read(&iclog->ic_refcnt)) {481			/* Reference holder will re-run iclog callbacks. */482			continue;483		}484		list_splice_init(&iclog->ic_callbacks, &cb_list);485		spin_unlock(&log->l_icloglock);486 487		xlog_cil_process_committed(&cb_list);488 489		spin_lock(&log->l_icloglock);490		wake_up_all(&iclog->ic_write_wait);491		wake_up_all(&iclog->ic_force_wait);492	} while ((iclog = iclog->ic_next) != log->l_iclog);493 494	wake_up_all(&log->l_flush_wait);495}496 497/*498 * Flush iclog to disk if this is the last reference to the given iclog and the499 * it is in the WANT_SYNC state.500 *501 * If XLOG_ICL_NEED_FUA is already set on the iclog, we need to ensure that the502 * log tail is updated correctly. NEED_FUA indicates that the iclog will be503 * written to stable storage, and implies that a commit record is contained504 * within the iclog. We need to ensure that the log tail does not move beyond505 * the tail that the first commit record in the iclog ordered against, otherwise506 * correct recovery of that checkpoint becomes dependent on future operations507 * performed on this iclog.508 *509 * Hence if NEED_FUA is set and the current iclog tail lsn is empty, write the510 * current tail into iclog. Once the iclog tail is set, future operations must511 * not modify it, otherwise they potentially violate ordering constraints for512 * the checkpoint commit that wrote the initial tail lsn value. The tail lsn in513 * the iclog will get zeroed on activation of the iclog after sync, so we514 * always capture the tail lsn on the iclog on the first NEED_FUA release515 * regardless of the number of active reference counts on this iclog.516 */517int518xlog_state_release_iclog(519	struct xlog		*log,520	struct xlog_in_core	*iclog,521	struct xlog_ticket	*ticket)522{523	bool			last_ref;524 525	lockdep_assert_held(&log->l_icloglock);526 527	trace_xlog_iclog_release(iclog, _RET_IP_);528	/*529	 * Grabbing the current log tail needs to be atomic w.r.t. the writing530	 * of the tail LSN into the iclog so we guarantee that the log tail does531	 * not move between the first time we know that the iclog needs to be532	 * made stable and when we eventually submit it.533	 */534	if ((iclog->ic_state == XLOG_STATE_WANT_SYNC ||535	     (iclog->ic_flags & XLOG_ICL_NEED_FUA)) &&536	    !iclog->ic_header.h_tail_lsn) {537		iclog->ic_header.h_tail_lsn =538				cpu_to_be64(atomic64_read(&log->l_tail_lsn));539	}540 541	last_ref = atomic_dec_and_test(&iclog->ic_refcnt);542 543	if (xlog_is_shutdown(log)) {544		/*545		 * If there are no more references to this iclog, process the546		 * pending iclog callbacks that were waiting on the release of547		 * this iclog.548		 */549		if (last_ref)550			xlog_state_shutdown_callbacks(log);551		return -EIO;552	}553 554	if (!last_ref)555		return 0;556 557	if (iclog->ic_state != XLOG_STATE_WANT_SYNC) {558		ASSERT(iclog->ic_state == XLOG_STATE_ACTIVE);559		return 0;560	}561 562	iclog->ic_state = XLOG_STATE_SYNCING;563	xlog_verify_tail_lsn(log, iclog);564	trace_xlog_iclog_syncing(iclog, _RET_IP_);565 566	spin_unlock(&log->l_icloglock);567	xlog_sync(log, iclog, ticket);568	spin_lock(&log->l_icloglock);569	return 0;570}571 572/*573 * Mount a log filesystem574 *575 * mp		- ubiquitous xfs mount point structure576 * log_target	- buftarg of on-disk log device577 * blk_offset	- Start block # where block size is 512 bytes (BBSIZE)578 * num_bblocks	- Number of BBSIZE blocks in on-disk log579 *580 * Return error or zero.581 */582int583xfs_log_mount(584	xfs_mount_t		*mp,585	struct xfs_buftarg	*log_target,586	xfs_daddr_t		blk_offset,587	int			num_bblks)588{589	struct xlog		*log;590	int			error = 0;591	int			min_logfsbs;592 593	if (!xfs_has_norecovery(mp)) {594		xfs_notice(mp, "Mounting V%d Filesystem %pU",595			   XFS_SB_VERSION_NUM(&mp->m_sb),596			   &mp->m_sb.sb_uuid);597	} else {598		xfs_notice(mp,599"Mounting V%d filesystem %pU in no-recovery mode. Filesystem will be inconsistent.",600			   XFS_SB_VERSION_NUM(&mp->m_sb),601			   &mp->m_sb.sb_uuid);602		ASSERT(xfs_is_readonly(mp));603	}604 605	log = xlog_alloc_log(mp, log_target, blk_offset, num_bblks);606	if (IS_ERR(log)) {607		error = PTR_ERR(log);608		goto out;609	}610	mp->m_log = log;611 612	/*613	 * Now that we have set up the log and it's internal geometry614	 * parameters, we can validate the given log space and drop a critical615	 * message via syslog if the log size is too small. A log that is too616	 * small can lead to unexpected situations in transaction log space617	 * reservation stage. The superblock verifier has already validated all618	 * the other log geometry constraints, so we don't have to check those619	 * here.620	 *621	 * Note: For v4 filesystems, we can't just reject the mount if the622	 * validation fails.  This would mean that people would have to623	 * downgrade their kernel just to remedy the situation as there is no624	 * way to grow the log (short of black magic surgery with xfs_db).625	 *626	 * We can, however, reject mounts for V5 format filesystems, as the627	 * mkfs binary being used to make the filesystem should never create a628	 * filesystem with a log that is too small.629	 */630	min_logfsbs = xfs_log_calc_minimum_size(mp);631	if (mp->m_sb.sb_logblocks < min_logfsbs) {632		xfs_warn(mp,633		"Log size %d blocks too small, minimum size is %d blocks",634			 mp->m_sb.sb_logblocks, min_logfsbs);635 636		/*637		 * Log check errors are always fatal on v5; or whenever bad638		 * metadata leads to a crash.639		 */640		if (xfs_has_crc(mp)) {641			xfs_crit(mp, "AAIEEE! Log failed size checks. Abort!");642			ASSERT(0);643			error = -EINVAL;644			goto out_free_log;645		}646		xfs_crit(mp, "Log size out of supported range.");647		xfs_crit(mp,648"Continuing onwards, but if log hangs are experienced then please report this message in the bug report.");649	}650 651	/*652	 * Initialize the AIL now we have a log.653	 */654	error = xfs_trans_ail_init(mp);655	if (error) {656		xfs_warn(mp, "AIL initialisation failed: error %d", error);657		goto out_free_log;658	}659	log->l_ailp = mp->m_ail;660 661	/*662	 * skip log recovery on a norecovery mount.  pretend it all663	 * just worked.664	 */665	if (!xfs_has_norecovery(mp)) {666		error = xlog_recover(log);667		if (error) {668			xfs_warn(mp, "log mount/recovery failed: error %d",669				error);670			xlog_recover_cancel(log);671			goto out_destroy_ail;672		}673	}674 675	error = xfs_sysfs_init(&log->l_kobj, &xfs_log_ktype, &mp->m_kobj,676			       "log");677	if (error)678		goto out_destroy_ail;679 680	/* Normal transactions can now occur */681	clear_bit(XLOG_ACTIVE_RECOVERY, &log->l_opstate);682 683	/*684	 * Now the log has been fully initialised and we know were our685	 * space grant counters are, we can initialise the permanent ticket686	 * needed for delayed logging to work.687	 */688	xlog_cil_init_post_recovery(log);689 690	return 0;691 692out_destroy_ail:693	xfs_trans_ail_destroy(mp);694out_free_log:695	xlog_dealloc_log(log);696out:697	return error;698}699 700/*701 * Finish the recovery of the file system.  This is separate from the702 * xfs_log_mount() call, because it depends on the code in xfs_mountfs() to read703 * in the root and real-time bitmap inodes between calling xfs_log_mount() and704 * here.705 *706 * If we finish recovery successfully, start the background log work. If we are707 * not doing recovery, then we have a RO filesystem and we don't need to start708 * it.709 */710int711xfs_log_mount_finish(712	struct xfs_mount	*mp)713{714	struct xlog		*log = mp->m_log;715	int			error = 0;716 717	if (xfs_has_norecovery(mp)) {718		ASSERT(xfs_is_readonly(mp));719		return 0;720	}721 722	/*723	 * During the second phase of log recovery, we need iget and724	 * iput to behave like they do for an active filesystem.725	 * xfs_fs_drop_inode needs to be able to prevent the deletion726	 * of inodes before we're done replaying log items on those727	 * inodes.  Turn it off immediately after recovery finishes728	 * so that we don't leak the quota inodes if subsequent mount729	 * activities fail.730	 *731	 * We let all inodes involved in redo item processing end up on732	 * the LRU instead of being evicted immediately so that if we do733	 * something to an unlinked inode, the irele won't cause734	 * premature truncation and freeing of the inode, which results735	 * in log recovery failure.  We have to evict the unreferenced736	 * lru inodes after clearing SB_ACTIVE because we don't737	 * otherwise clean up the lru if there's a subsequent failure in738	 * xfs_mountfs, which leads to us leaking the inodes if nothing739	 * else (e.g. quotacheck) references the inodes before the740	 * mount failure occurs.741	 */742	mp->m_super->s_flags |= SB_ACTIVE;743	xfs_log_work_queue(mp);744	if (xlog_recovery_needed(log))745		error = xlog_recover_finish(log);746	mp->m_super->s_flags &= ~SB_ACTIVE;747	evict_inodes(mp->m_super);748 749	/*750	 * Drain the buffer LRU after log recovery. This is required for v4751	 * filesystems to avoid leaving around buffers with NULL verifier ops,752	 * but we do it unconditionally to make sure we're always in a clean753	 * cache state after mount.754	 *755	 * Don't push in the error case because the AIL may have pending intents756	 * that aren't removed until recovery is cancelled.757	 */758	if (xlog_recovery_needed(log)) {759		if (!error) {760			xfs_log_force(mp, XFS_LOG_SYNC);761			xfs_ail_push_all_sync(mp->m_ail);762		}763		xfs_notice(mp, "Ending recovery (logdev: %s)",764				mp->m_logname ? mp->m_logname : "internal");765	} else {766		xfs_info(mp, "Ending clean mount");767	}768	xfs_buftarg_drain(mp->m_ddev_targp);769 770	clear_bit(XLOG_RECOVERY_NEEDED, &log->l_opstate);771 772	/* Make sure the log is dead if we're returning failure. */773	ASSERT(!error || xlog_is_shutdown(log));774 775	return error;776}777 778/*779 * The mount has failed. Cancel the recovery if it hasn't completed and destroy780 * the log.781 */782void783xfs_log_mount_cancel(784	struct xfs_mount	*mp)785{786	xlog_recover_cancel(mp->m_log);787	xfs_log_unmount(mp);788}789 790/*791 * Flush out the iclog to disk ensuring that device caches are flushed and792 * the iclog hits stable storage before any completion waiters are woken.793 */794static inline int795xlog_force_iclog(796	struct xlog_in_core	*iclog)797{798	atomic_inc(&iclog->ic_refcnt);799	iclog->ic_flags |= XLOG_ICL_NEED_FLUSH | XLOG_ICL_NEED_FUA;800	if (iclog->ic_state == XLOG_STATE_ACTIVE)801		xlog_state_switch_iclogs(iclog->ic_log, iclog, 0);802	return xlog_state_release_iclog(iclog->ic_log, iclog, NULL);803}804 805/*806 * Cycle all the iclogbuf locks to make sure all log IO completion807 * is done before we tear down these buffers.808 */809static void810xlog_wait_iclog_completion(struct xlog *log)811{812	int		i;813	struct xlog_in_core	*iclog = log->l_iclog;814 815	for (i = 0; i < log->l_iclog_bufs; i++) {816		down(&iclog->ic_sema);817		up(&iclog->ic_sema);818		iclog = iclog->ic_next;819	}820}821 822/*823 * Wait for the iclog and all prior iclogs to be written disk as required by the824 * log force state machine. Waiting on ic_force_wait ensures iclog completions825 * have been ordered and callbacks run before we are woken here, hence826 * guaranteeing that all the iclogs up to this one are on stable storage.827 */828int829xlog_wait_on_iclog(830	struct xlog_in_core	*iclog)831		__releases(iclog->ic_log->l_icloglock)832{833	struct xlog		*log = iclog->ic_log;834 835	trace_xlog_iclog_wait_on(iclog, _RET_IP_);836	if (!xlog_is_shutdown(log) &&837	    iclog->ic_state != XLOG_STATE_ACTIVE &&838	    iclog->ic_state != XLOG_STATE_DIRTY) {839		XFS_STATS_INC(log->l_mp, xs_log_force_sleep);840		xlog_wait(&iclog->ic_force_wait, &log->l_icloglock);841	} else {842		spin_unlock(&log->l_icloglock);843	}844 845	if (xlog_is_shutdown(log))846		return -EIO;847	return 0;848}849 850/*851 * Write out an unmount record using the ticket provided. We have to account for852 * the data space used in the unmount ticket as this write is not done from a853 * transaction context that has already done the accounting for us.854 */855static int856xlog_write_unmount_record(857	struct xlog		*log,858	struct xlog_ticket	*ticket)859{860	struct  {861		struct xlog_op_header ophdr;862		struct xfs_unmount_log_format ulf;863	} unmount_rec = {864		.ophdr = {865			.oh_clientid = XFS_LOG,866			.oh_tid = cpu_to_be32(ticket->t_tid),867			.oh_flags = XLOG_UNMOUNT_TRANS,868		},869		.ulf = {870			.magic = XLOG_UNMOUNT_TYPE,871		},872	};873	struct xfs_log_iovec reg = {874		.i_addr = &unmount_rec,875		.i_len = sizeof(unmount_rec),876		.i_type = XLOG_REG_TYPE_UNMOUNT,877	};878	struct xfs_log_vec vec = {879		.lv_niovecs = 1,880		.lv_iovecp = &reg,881	};882	LIST_HEAD(lv_chain);883	list_add(&vec.lv_list, &lv_chain);884 885	BUILD_BUG_ON((sizeof(struct xlog_op_header) +886		      sizeof(struct xfs_unmount_log_format)) !=887							sizeof(unmount_rec));888 889	/* account for space used by record data */890	ticket->t_curr_res -= sizeof(unmount_rec);891 892	return xlog_write(log, NULL, &lv_chain, ticket, reg.i_len);893}894 895/*896 * Mark the filesystem clean by writing an unmount record to the head of the897 * log.898 */899static void900xlog_unmount_write(901	struct xlog		*log)902{903	struct xfs_mount	*mp = log->l_mp;904	struct xlog_in_core	*iclog;905	struct xlog_ticket	*tic = NULL;906	int			error;907 908	error = xfs_log_reserve(mp, 600, 1, &tic, 0);909	if (error)910		goto out_err;911 912	error = xlog_write_unmount_record(log, tic);913	/*914	 * At this point, we're umounting anyway, so there's no point in915	 * transitioning log state to shutdown. Just continue...916	 */917out_err:918	if (error)919		xfs_alert(mp, "%s: unmount record failed", __func__);920 921	spin_lock(&log->l_icloglock);922	iclog = log->l_iclog;923	error = xlog_force_iclog(iclog);924	xlog_wait_on_iclog(iclog);925 926	if (tic) {927		trace_xfs_log_umount_write(log, tic);928		xfs_log_ticket_ungrant(log, tic);929	}930}931 932static void933xfs_log_unmount_verify_iclog(934	struct xlog		*log)935{936	struct xlog_in_core	*iclog = log->l_iclog;937 938	do {939		ASSERT(iclog->ic_state == XLOG_STATE_ACTIVE);940		ASSERT(iclog->ic_offset == 0);941	} while ((iclog = iclog->ic_next) != log->l_iclog);942}943 944/*945 * Unmount record used to have a string "Unmount filesystem--" in the946 * data section where the "Un" was really a magic number (XLOG_UNMOUNT_TYPE).947 * We just write the magic number now since that particular field isn't948 * currently architecture converted and "Unmount" is a bit foo.949 * As far as I know, there weren't any dependencies on the old behaviour.950 */951static void952xfs_log_unmount_write(953	struct xfs_mount	*mp)954{955	struct xlog		*log = mp->m_log;956 957	if (!xfs_log_writable(mp))958		return;959 960	xfs_log_force(mp, XFS_LOG_SYNC);961 962	if (xlog_is_shutdown(log))963		return;964 965	/*966	 * If we think the summary counters are bad, avoid writing the unmount967	 * record to force log recovery at next mount, after which the summary968	 * counters will be recalculated.  Refer to xlog_check_unmount_rec for969	 * more details.970	 */971	if (XFS_TEST_ERROR(xfs_fs_has_sickness(mp, XFS_SICK_FS_COUNTERS), mp,972			XFS_ERRTAG_FORCE_SUMMARY_RECALC)) {973		xfs_alert(mp, "%s: will fix summary counters at next mount",974				__func__);975		return;976	}977 978	xfs_log_unmount_verify_iclog(log);979	xlog_unmount_write(log);980}981 982/*983 * Empty the log for unmount/freeze.984 *985 * To do this, we first need to shut down the background log work so it is not986 * trying to cover the log as we clean up. We then need to unpin all objects in987 * the log so we can then flush them out. Once they have completed their IO and988 * run the callbacks removing themselves from the AIL, we can cover the log.989 */990int991xfs_log_quiesce(992	struct xfs_mount	*mp)993{994	/*995	 * Clear log incompat features since we're quiescing the log.  Report996	 * failures, though it's not fatal to have a higher log feature997	 * protection level than the log contents actually require.998	 */999	if (xfs_clear_incompat_log_features(mp)) {1000		int error;1001 1002		error = xfs_sync_sb(mp, false);1003		if (error)1004			xfs_warn(mp,1005	"Failed to clear log incompat features on quiesce");1006	}1007 1008	cancel_delayed_work_sync(&mp->m_log->l_work);1009	xfs_log_force(mp, XFS_LOG_SYNC);1010 1011	/*1012	 * The superblock buffer is uncached and while xfs_ail_push_all_sync()1013	 * will push it, xfs_buftarg_wait() will not wait for it. Further,1014	 * xfs_buf_iowait() cannot be used because it was pushed with the1015	 * XBF_ASYNC flag set, so we need to use a lock/unlock pair to wait for1016	 * the IO to complete.1017	 */1018	xfs_ail_push_all_sync(mp->m_ail);1019	xfs_buftarg_wait(mp->m_ddev_targp);1020	xfs_buf_lock(mp->m_sb_bp);1021	xfs_buf_unlock(mp->m_sb_bp);1022 1023	return xfs_log_cover(mp);1024}1025 1026void1027xfs_log_clean(1028	struct xfs_mount	*mp)1029{1030	xfs_log_quiesce(mp);1031	xfs_log_unmount_write(mp);1032}1033 1034/*1035 * Shut down and release the AIL and Log.1036 *1037 * During unmount, we need to ensure we flush all the dirty metadata objects1038 * from the AIL so that the log is empty before we write the unmount record to1039 * the log. Once this is done, we can tear down the AIL and the log.1040 */1041void1042xfs_log_unmount(1043	struct xfs_mount	*mp)1044{1045	xfs_log_clean(mp);1046 1047	/*1048	 * If shutdown has come from iclog IO context, the log1049	 * cleaning will have been skipped and so we need to wait1050	 * for the iclog to complete shutdown processing before we1051	 * tear anything down.1052	 */1053	xlog_wait_iclog_completion(mp->m_log);1054 1055	xfs_buftarg_drain(mp->m_ddev_targp);1056 1057	xfs_trans_ail_destroy(mp);1058 1059	xfs_sysfs_del(&mp->m_log->l_kobj);1060 1061	xlog_dealloc_log(mp->m_log);1062}1063 1064void1065xfs_log_item_init(1066	struct xfs_mount	*mp,1067	struct xfs_log_item	*item,1068	int			type,1069	const struct xfs_item_ops *ops)1070{1071	item->li_log = mp->m_log;1072	item->li_ailp = mp->m_ail;1073	item->li_type = type;1074	item->li_ops = ops;1075	item->li_lv = NULL;1076 1077	INIT_LIST_HEAD(&item->li_ail);1078	INIT_LIST_HEAD(&item->li_cil);1079	INIT_LIST_HEAD(&item->li_bio_list);1080	INIT_LIST_HEAD(&item->li_trans);1081}1082 1083/*1084 * Wake up processes waiting for log space after we have moved the log tail.1085 */1086void1087xfs_log_space_wake(1088	struct xfs_mount	*mp)1089{1090	struct xlog		*log = mp->m_log;1091	int			free_bytes;1092 1093	if (xlog_is_shutdown(log))1094		return;1095 1096	if (!list_empty_careful(&log->l_write_head.waiters)) {1097		ASSERT(!xlog_in_recovery(log));1098 1099		spin_lock(&log->l_write_head.lock);1100		free_bytes = xlog_grant_space_left(log, &log->l_write_head);1101		xlog_grant_head_wake(log, &log->l_write_head, &free_bytes);1102		spin_unlock(&log->l_write_head.lock);1103	}1104 1105	if (!list_empty_careful(&log->l_reserve_head.waiters)) {1106		ASSERT(!xlog_in_recovery(log));1107 1108		spin_lock(&log->l_reserve_head.lock);1109		free_bytes = xlog_grant_space_left(log, &log->l_reserve_head);1110		xlog_grant_head_wake(log, &log->l_reserve_head, &free_bytes);1111		spin_unlock(&log->l_reserve_head.lock);1112	}1113}1114 1115/*1116 * Determine if we have a transaction that has gone to disk that needs to be1117 * covered. To begin the transition to the idle state firstly the log needs to1118 * be idle. That means the CIL, the AIL and the iclogs needs to be empty before1119 * we start attempting to cover the log.1120 *1121 * Only if we are then in a state where covering is needed, the caller is1122 * informed that dummy transactions are required to move the log into the idle1123 * state.1124 *1125 * If there are any items in the AIl or CIL, then we do not want to attempt to1126 * cover the log as we may be in a situation where there isn't log space1127 * available to run a dummy transaction and this can lead to deadlocks when the1128 * tail of the log is pinned by an item that is modified in the CIL.  Hence1129 * there's no point in running a dummy transaction at this point because we1130 * can't start trying to idle the log until both the CIL and AIL are empty.1131 */1132static bool1133xfs_log_need_covered(1134	struct xfs_mount	*mp)1135{1136	struct xlog		*log = mp->m_log;1137	bool			needed = false;1138 1139	if (!xlog_cil_empty(log))1140		return false;1141 1142	spin_lock(&log->l_icloglock);1143	switch (log->l_covered_state) {1144	case XLOG_STATE_COVER_DONE:1145	case XLOG_STATE_COVER_DONE2:1146	case XLOG_STATE_COVER_IDLE:1147		break;1148	case XLOG_STATE_COVER_NEED:1149	case XLOG_STATE_COVER_NEED2:1150		if (xfs_ail_min_lsn(log->l_ailp))1151			break;1152		if (!xlog_iclogs_empty(log))1153			break;1154 1155		needed = true;1156		if (log->l_covered_state == XLOG_STATE_COVER_NEED)1157			log->l_covered_state = XLOG_STATE_COVER_DONE;1158		else1159			log->l_covered_state = XLOG_STATE_COVER_DONE2;1160		break;1161	default:1162		needed = true;1163		break;1164	}1165	spin_unlock(&log->l_icloglock);1166	return needed;1167}1168 1169/*1170 * Explicitly cover the log. This is similar to background log covering but1171 * intended for usage in quiesce codepaths. The caller is responsible to ensure1172 * the log is idle and suitable for covering. The CIL, iclog buffers and AIL1173 * must all be empty.1174 */1175static int1176xfs_log_cover(1177	struct xfs_mount	*mp)1178{1179	int			error = 0;1180	bool			need_covered;1181 1182	ASSERT((xlog_cil_empty(mp->m_log) && xlog_iclogs_empty(mp->m_log) &&1183	        !xfs_ail_min_lsn(mp->m_log->l_ailp)) ||1184		xlog_is_shutdown(mp->m_log));1185 1186	if (!xfs_log_writable(mp))1187		return 0;1188 1189	/*1190	 * xfs_log_need_covered() is not idempotent because it progresses the1191	 * state machine if the log requires covering. Therefore, we must call1192	 * this function once and use the result until we've issued an sb sync.1193	 * Do so first to make that abundantly clear.1194	 *1195	 * Fall into the covering sequence if the log needs covering or the1196	 * mount has lazy superblock accounting to sync to disk. The sb sync1197	 * used for covering accumulates the in-core counters, so covering1198	 * handles this for us.1199	 */1200	need_covered = xfs_log_need_covered(mp);1201	if (!need_covered && !xfs_has_lazysbcount(mp))1202		return 0;1203 1204	/*1205	 * To cover the log, commit the superblock twice (at most) in1206	 * independent checkpoints. The first serves as a reference for the1207	 * tail pointer. The sync transaction and AIL push empties the AIL and1208	 * updates the in-core tail to the LSN of the first checkpoint. The1209	 * second commit updates the on-disk tail with the in-core LSN,1210	 * covering the log. Push the AIL one more time to leave it empty, as1211	 * we found it.1212	 */1213	do {1214		error = xfs_sync_sb(mp, true);1215		if (error)1216			break;1217		xfs_ail_push_all_sync(mp->m_ail);1218	} while (xfs_log_need_covered(mp));1219 1220	return error;1221}1222 1223static void1224xlog_ioend_work(1225	struct work_struct	*work)1226{1227	struct xlog_in_core     *iclog =1228		container_of(work, struct xlog_in_core, ic_end_io_work);1229	struct xlog		*log = iclog->ic_log;1230	int			error;1231 1232	error = blk_status_to_errno(iclog->ic_bio.bi_status);1233#ifdef DEBUG1234	/* treat writes with injected CRC errors as failed */1235	if (iclog->ic_fail_crc)1236		error = -EIO;1237#endif1238 1239	/*1240	 * Race to shutdown the filesystem if we see an error.1241	 */1242	if (XFS_TEST_ERROR(error, log->l_mp, XFS_ERRTAG_IODONE_IOERR)) {1243		xfs_alert(log->l_mp, "log I/O error %d", error);1244		xlog_force_shutdown(log, SHUTDOWN_LOG_IO_ERROR);1245	}1246 1247	xlog_state_done_syncing(iclog);1248	bio_uninit(&iclog->ic_bio);1249 1250	/*1251	 * Drop the lock to signal that we are done. Nothing references the1252	 * iclog after this, so an unmount waiting on this lock can now tear it1253	 * down safely. As such, it is unsafe to reference the iclog after the1254	 * unlock as we could race with it being freed.1255	 */1256	up(&iclog->ic_sema);1257}1258 1259/*1260 * Return size of each in-core log record buffer.1261 *1262 * All machines get 8 x 32kB buffers by default, unless tuned otherwise.1263 *1264 * If the filesystem blocksize is too large, we may need to choose a1265 * larger size since the directory code currently logs entire blocks.1266 */1267STATIC void1268xlog_get_iclog_buffer_size(1269	struct xfs_mount	*mp,1270	struct xlog		*log)1271{1272	if (mp->m_logbufs <= 0)1273		mp->m_logbufs = XLOG_MAX_ICLOGS;1274	if (mp->m_logbsize <= 0)1275		mp->m_logbsize = XLOG_BIG_RECORD_BSIZE;1276 1277	log->l_iclog_bufs = mp->m_logbufs;1278	log->l_iclog_size = mp->m_logbsize;1279 1280	/*1281	 * # headers = size / 32k - one header holds cycles from 32k of data.1282	 */1283	log->l_iclog_heads =1284		DIV_ROUND_UP(mp->m_logbsize, XLOG_HEADER_CYCLE_SIZE);1285	log->l_iclog_hsize = log->l_iclog_heads << BBSHIFT;1286}1287 1288void1289xfs_log_work_queue(1290	struct xfs_mount        *mp)1291{1292	queue_delayed_work(mp->m_sync_workqueue, &mp->m_log->l_work,1293				msecs_to_jiffies(xfs_syncd_centisecs * 10));1294}1295 1296/*1297 * Clear the log incompat flags if we have the opportunity.1298 *1299 * This only happens if we're about to log the second dummy transaction as part1300 * of covering the log.1301 */1302static inline void1303xlog_clear_incompat(1304	struct xlog		*log)1305{1306	struct xfs_mount	*mp = log->l_mp;1307 1308	if (!xfs_sb_has_incompat_log_feature(&mp->m_sb,1309				XFS_SB_FEAT_INCOMPAT_LOG_ALL))1310		return;1311 1312	if (log->l_covered_state != XLOG_STATE_COVER_DONE2)1313		return;1314 1315	xfs_clear_incompat_log_features(mp);1316}1317 1318/*1319 * Every sync period we need to unpin all items in the AIL and push them to1320 * disk. If there is nothing dirty, then we might need to cover the log to1321 * indicate that the filesystem is idle.1322 */1323static void1324xfs_log_worker(1325	struct work_struct	*work)1326{1327	struct xlog		*log = container_of(to_delayed_work(work),1328						struct xlog, l_work);1329	struct xfs_mount	*mp = log->l_mp;1330 1331	/* dgc: errors ignored - not fatal and nowhere to report them */1332	if (xfs_fs_writable(mp, SB_FREEZE_WRITE) && xfs_log_need_covered(mp)) {1333		/*1334		 * Dump a transaction into the log that contains no real change.1335		 * This is needed to stamp the current tail LSN into the log1336		 * during the covering operation.1337		 *1338		 * We cannot use an inode here for this - that will push dirty1339		 * state back up into the VFS and then periodic inode flushing1340		 * will prevent log covering from making progress. Hence we1341		 * synchronously log the superblock instead to ensure the1342		 * superblock is immediately unpinned and can be written back.1343		 */1344		xlog_clear_incompat(log);1345		xfs_sync_sb(mp, true);1346	} else1347		xfs_log_force(mp, 0);1348 1349	/* start pushing all the metadata that is currently dirty */1350	xfs_ail_push_all(mp->m_ail);1351 1352	/* queue us up again */1353	xfs_log_work_queue(mp);1354}1355 1356/*1357 * This routine initializes some of the log structure for a given mount point.1358 * Its primary purpose is to fill in enough, so recovery can occur.  However,1359 * some other stuff may be filled in too.1360 */1361STATIC struct xlog *1362xlog_alloc_log(1363	struct xfs_mount	*mp,1364	struct xfs_buftarg	*log_target,1365	xfs_daddr_t		blk_offset,1366	int			num_bblks)1367{1368	struct xlog		*log;1369	xlog_rec_header_t	*head;1370	xlog_in_core_t		**iclogp;1371	xlog_in_core_t		*iclog, *prev_iclog=NULL;1372	int			i;1373	int			error = -ENOMEM;1374	uint			log2_size = 0;1375 1376	log = kzalloc(sizeof(struct xlog), GFP_KERNEL | __GFP_RETRY_MAYFAIL);1377	if (!log) {1378		xfs_warn(mp, "Log allocation failed: No memory!");1379		goto out;1380	}1381 1382	log->l_mp	   = mp;1383	log->l_targ	   = log_target;1384	log->l_logsize     = BBTOB(num_bblks);1385	log->l_logBBstart  = blk_offset;1386	log->l_logBBsize   = num_bblks;1387	log->l_covered_state = XLOG_STATE_COVER_IDLE;1388	set_bit(XLOG_ACTIVE_RECOVERY, &log->l_opstate);1389	INIT_DELAYED_WORK(&log->l_work, xfs_log_worker);1390	INIT_LIST_HEAD(&log->r_dfops);1391 1392	log->l_prev_block  = -1;1393	/* log->l_tail_lsn = 0x100000000LL; cycle = 1; current block = 0 */1394	xlog_assign_atomic_lsn(&log->l_tail_lsn, 1, 0);1395	log->l_curr_cycle  = 1;	    /* 0 is bad since this is initial value */1396 1397	if (xfs_has_logv2(mp) && mp->m_sb.sb_logsunit > 1)1398		log->l_iclog_roundoff = mp->m_sb.sb_logsunit;1399	else1400		log->l_iclog_roundoff = BBSIZE;1401 1402	xlog_grant_head_init(&log->l_reserve_head);1403	xlog_grant_head_init(&log->l_write_head);1404 1405	error = -EFSCORRUPTED;1406	if (xfs_has_sector(mp)) {1407	        log2_size = mp->m_sb.sb_logsectlog;1408		if (log2_size < BBSHIFT) {1409			xfs_warn(mp, "Log sector size too small (0x%x < 0x%x)",1410				log2_size, BBSHIFT);1411			goto out_free_log;1412		}1413 1414	        log2_size -= BBSHIFT;1415		if (log2_size > mp->m_sectbb_log) {1416			xfs_warn(mp, "Log sector size too large (0x%x > 0x%x)",1417				log2_size, mp->m_sectbb_log);1418			goto out_free_log;1419		}1420 1421		/* for larger sector sizes, must have v2 or external log */1422		if (log2_size && log->l_logBBstart > 0 &&1423			    !xfs_has_logv2(mp)) {1424			xfs_warn(mp,1425		"log sector size (0x%x) invalid for configuration.",1426				log2_size);1427			goto out_free_log;1428		}1429	}1430	log->l_sectBBsize = 1 << log2_size;1431 1432	xlog_get_iclog_buffer_size(mp, log);1433 1434	spin_lock_init(&log->l_icloglock);1435	init_waitqueue_head(&log->l_flush_wait);1436 1437	iclogp = &log->l_iclog;1438	/*1439	 * The amount of memory to allocate for the iclog structure is1440	 * rather funky due to the way the structure is defined.  It is1441	 * done this way so that we can use different sizes for machines1442	 * with different amounts of memory.  See the definition of1443	 * xlog_in_core_t in xfs_log_priv.h for details.1444	 */1445	ASSERT(log->l_iclog_size >= 4096);1446	for (i = 0; i < log->l_iclog_bufs; i++) {1447		size_t bvec_size = howmany(log->l_iclog_size, PAGE_SIZE) *1448				sizeof(struct bio_vec);1449 1450		iclog = kzalloc(sizeof(*iclog) + bvec_size,1451				GFP_KERNEL | __GFP_RETRY_MAYFAIL);1452		if (!iclog)1453			goto out_free_iclog;1454 1455		*iclogp = iclog;1456		iclog->ic_prev = prev_iclog;1457		prev_iclog = iclog;1458 1459		iclog->ic_data = kvzalloc(log->l_iclog_size,1460				GFP_KERNEL | __GFP_RETRY_MAYFAIL);1461		if (!iclog->ic_data)1462			goto out_free_iclog;1463		head = &iclog->ic_header;1464		memset(head, 0, sizeof(xlog_rec_header_t));1465		head->h_magicno = cpu_to_be32(XLOG_HEADER_MAGIC_NUM);1466		head->h_version = cpu_to_be32(1467			xfs_has_logv2(log->l_mp) ? 2 : 1);1468		head->h_size = cpu_to_be32(log->l_iclog_size);1469		/* new fields */1470		head->h_fmt = cpu_to_be32(XLOG_FMT);1471		memcpy(&head->h_fs_uuid, &mp->m_sb.sb_uuid, sizeof(uuid_t));1472 1473		iclog->ic_size = log->l_iclog_size - log->l_iclog_hsize;1474		iclog->ic_state = XLOG_STATE_ACTIVE;1475		iclog->ic_log = log;1476		atomic_set(&iclog->ic_refcnt, 0);1477		INIT_LIST_HEAD(&iclog->ic_callbacks);1478		iclog->ic_datap = (void *)iclog->ic_data + log->l_iclog_hsize;1479 1480		init_waitqueue_head(&iclog->ic_force_wait);1481		init_waitqueue_head(&iclog->ic_write_wait);1482		INIT_WORK(&iclog->ic_end_io_work, xlog_ioend_work);1483		sema_init(&iclog->ic_sema, 1);1484 1485		iclogp = &iclog->ic_next;1486	}1487	*iclogp = log->l_iclog;			/* complete ring */1488	log->l_iclog->ic_prev = prev_iclog;	/* re-write 1st prev ptr */1489 1490	log->l_ioend_workqueue = alloc_workqueue("xfs-log/%s",1491			XFS_WQFLAGS(WQ_FREEZABLE | WQ_MEM_RECLAIM |1492				    WQ_HIGHPRI),1493			0, mp->m_super->s_id);1494	if (!log->l_ioend_workqueue)1495		goto out_free_iclog;1496 1497	error = xlog_cil_init(log);1498	if (error)1499		goto out_destroy_workqueue;1500	return log;1501 1502out_destroy_workqueue:1503	destroy_workqueue(log->l_ioend_workqueue);1504out_free_iclog:1505	for (iclog = log->l_iclog; iclog; iclog = prev_iclog) {1506		prev_iclog = iclog->ic_next;1507		kvfree(iclog->ic_data);1508		kfree(iclog);1509		if (prev_iclog == log->l_iclog)1510			break;1511	}1512out_free_log:1513	kfree(log);1514out:1515	return ERR_PTR(error);1516}	/* xlog_alloc_log */1517 1518/*1519 * Stamp cycle number in every block1520 */1521STATIC void1522xlog_pack_data(1523	struct xlog		*log,1524	struct xlog_in_core	*iclog,1525	int			roundoff)1526{1527	int			i, j, k;1528	int			size = iclog->ic_offset + roundoff;1529	__be32			cycle_lsn;1530	char			*dp;1531 1532	cycle_lsn = CYCLE_LSN_DISK(iclog->ic_header.h_lsn);1533 1534	dp = iclog->ic_datap;1535	for (i = 0; i < BTOBB(size); i++) {1536		if (i >= (XLOG_HEADER_CYCLE_SIZE / BBSIZE))1537			break;1538		iclog->ic_header.h_cycle_data[i] = *(__be32 *)dp;1539		*(__be32 *)dp = cycle_lsn;1540		dp += BBSIZE;1541	}1542 1543	if (xfs_has_logv2(log->l_mp)) {1544		xlog_in_core_2_t *xhdr = iclog->ic_data;1545 1546		for ( ; i < BTOBB(size); i++) {1547			j = i / (XLOG_HEADER_CYCLE_SIZE / BBSIZE);1548			k = i % (XLOG_HEADER_CYCLE_SIZE / BBSIZE);1549			xhdr[j].hic_xheader.xh_cycle_data[k] = *(__be32 *)dp;1550			*(__be32 *)dp = cycle_lsn;1551			dp += BBSIZE;1552		}1553 1554		for (i = 1; i < log->l_iclog_heads; i++)1555			xhdr[i].hic_xheader.xh_cycle = cycle_lsn;1556	}1557}1558 1559/*1560 * Calculate the checksum for a log buffer.1561 *1562 * This is a little more complicated than it should be because the various1563 * headers and the actual data are non-contiguous.1564 */1565__le321566xlog_cksum(1567	struct xlog		*log,1568	struct xlog_rec_header	*rhead,1569	char			*dp,1570	int			size)1571{1572	uint32_t		crc;1573 1574	/* first generate the crc for the record header ... */1575	crc = xfs_start_cksum_update((char *)rhead,1576			      sizeof(struct xlog_rec_header),1577			      offsetof(struct xlog_rec_header, h_crc));1578 1579	/* ... then for additional cycle data for v2 logs ... */1580	if (xfs_has_logv2(log->l_mp)) {1581		union xlog_in_core2 *xhdr = (union xlog_in_core2 *)rhead;1582		int		i;1583		int		xheads;1584 1585		xheads = DIV_ROUND_UP(size, XLOG_HEADER_CYCLE_SIZE);1586 1587		for (i = 1; i < xheads; i++) {1588			crc = crc32c(crc, &xhdr[i].hic_xheader,1589				     sizeof(struct xlog_rec_ext_header));1590		}1591	}1592 1593	/* ... and finally for the payload */1594	crc = crc32c(crc, dp, size);1595 1596	return xfs_end_cksum(crc);1597}1598 1599static void1600xlog_bio_end_io(1601	struct bio		*bio)1602{1603	struct xlog_in_core	*iclog = bio->bi_private;1604 1605	queue_work(iclog->ic_log->l_ioend_workqueue,1606		   &iclog->ic_end_io_work);1607}1608 1609static int1610xlog_map_iclog_data(1611	struct bio		*bio,1612	void			*data,1613	size_t			count)1614{1615	do {1616		struct page	*page = kmem_to_page(data);1617		unsigned int	off = offset_in_page(data);1618		size_t		len = min_t(size_t, count, PAGE_SIZE - off);1619 1620		if (bio_add_page(bio, page, len, off) != len)1621			return -EIO;1622 1623		data += len;1624		count -= len;1625	} while (count);1626 1627	return 0;1628}1629 1630STATIC void1631xlog_write_iclog(1632	struct xlog		*log,1633	struct xlog_in_core	*iclog,1634	uint64_t		bno,1635	unsigned int		count)1636{1637	ASSERT(bno < log->l_logBBsize);1638	trace_xlog_iclog_write(iclog, _RET_IP_);1639 1640	/*1641	 * We lock the iclogbufs here so that we can serialise against I/O1642	 * completion during unmount.  We might be processing a shutdown1643	 * triggered during unmount, and that can occur asynchronously to the1644	 * unmount thread, and hence we need to ensure that completes before1645	 * tearing down the iclogbufs.  Hence we need to hold the buffer lock1646	 * across the log IO to archieve that.1647	 */1648	down(&iclog->ic_sema);1649	if (xlog_is_shutdown(log)) {1650		/*1651		 * It would seem logical to return EIO here, but we rely on1652		 * the log state machine to propagate I/O errors instead of1653		 * doing it here.  We kick of the state machine and unlock1654		 * the buffer manually, the code needs to be kept in sync1655		 * with the I/O completion path.1656		 */1657		goto sync;1658	}1659 1660	/*1661	 * We use REQ_SYNC | REQ_IDLE here to tell the block layer the are more1662	 * IOs coming immediately after this one. This prevents the block layer1663	 * writeback throttle from throttling log writes behind background1664	 * metadata writeback and causing priority inversions.1665	 */1666	bio_init(&iclog->ic_bio, log->l_targ->bt_bdev, iclog->ic_bvec,1667		 howmany(count, PAGE_SIZE),1668		 REQ_OP_WRITE | REQ_META | REQ_SYNC | REQ_IDLE);1669	iclog->ic_bio.bi_iter.bi_sector = log->l_logBBstart + bno;1670	iclog->ic_bio.bi_end_io = xlog_bio_end_io;1671	iclog->ic_bio.bi_private = iclog;1672 1673	if (iclog->ic_flags & XLOG_ICL_NEED_FLUSH) {1674		iclog->ic_bio.bi_opf |= REQ_PREFLUSH;1675		/*1676		 * For external log devices, we also need to flush the data1677		 * device cache first to ensure all metadata writeback covered1678		 * by the LSN in this iclog is on stable storage. This is slow,1679		 * but it *must* complete before we issue the external log IO.1680		 *1681		 * If the flush fails, we cannot conclude that past metadata1682		 * writeback from the log succeeded.  Repeating the flush is1683		 * not possible, hence we must shut down with log IO error to1684		 * avoid shutdown re-entering this path and erroring out again.1685		 */1686		if (log->l_targ != log->l_mp->m_ddev_targp &&1687		    blkdev_issue_flush(log->l_mp->m_ddev_targp->bt_bdev))1688			goto shutdown;1689	}1690	if (iclog->ic_flags & XLOG_ICL_NEED_FUA)1691		iclog->ic_bio.bi_opf |= REQ_FUA;1692 1693	iclog->ic_flags &= ~(XLOG_ICL_NEED_FLUSH | XLOG_ICL_NEED_FUA);1694 1695	if (xlog_map_iclog_data(&iclog->ic_bio, iclog->ic_data, count))1696		goto shutdown;1697 1698	if (is_vmalloc_addr(iclog->ic_data))1699		flush_kernel_vmap_range(iclog->ic_data, count);1700 1701	/*1702	 * If this log buffer would straddle the end of the log we will have1703	 * to split it up into two bios, so that we can continue at the start.1704	 */1705	if (bno + BTOBB(count) > log->l_logBBsize) {1706		struct bio *split;1707 1708		split = bio_split(&iclog->ic_bio, log->l_logBBsize - bno,1709				  GFP_NOIO, &fs_bio_set);1710		bio_chain(split, &iclog->ic_bio);1711		submit_bio(split);1712 1713		/* restart at logical offset zero for the remainder */1714		iclog->ic_bio.bi_iter.bi_sector = log->l_logBBstart;1715	}1716 1717	submit_bio(&iclog->ic_bio);1718	return;1719shutdown:1720	xlog_force_shutdown(log, SHUTDOWN_LOG_IO_ERROR);1721sync:1722	xlog_state_done_syncing(iclog);1723	up(&iclog->ic_sema);1724}1725 1726/*1727 * We need to bump cycle number for the part of the iclog that is1728 * written to the start of the log. Watch out for the header magic1729 * number case, though.1730 */1731static void1732xlog_split_iclog(1733	struct xlog		*log,1734	void			*data,1735	uint64_t		bno,1736	unsigned int		count)1737{1738	unsigned int		split_offset = BBTOB(log->l_logBBsize - bno);1739	unsigned int		i;1740 1741	for (i = split_offset; i < count; i += BBSIZE) {1742		uint32_t cycle = get_unaligned_be32(data + i);1743 1744		if (++cycle == XLOG_HEADER_MAGIC_NUM)1745			cycle++;1746		put_unaligned_be32(cycle, data + i);1747	}1748}1749 1750static int1751xlog_calc_iclog_size(1752	struct xlog		*log,1753	struct xlog_in_core	*iclog,1754	uint32_t		*roundoff)1755{1756	uint32_t		count_init, count;1757 1758	/* Add for LR header */1759	count_init = log->l_iclog_hsize + iclog->ic_offset;1760	count = roundup(count_init, log->l_iclog_roundoff);1761 1762	*roundoff = count - count_init;1763 1764	ASSERT(count >= count_init);1765	ASSERT(*roundoff < log->l_iclog_roundoff);1766	return count;1767}1768 1769/*1770 * Flush out the in-core log (iclog) to the on-disk log in an asynchronous1771 * fashion.  Previously, we should have moved the current iclog1772 * ptr in the log to point to the next available iclog.  This allows further1773 * write to continue while this code syncs out an iclog ready to go.1774 * Before an in-core log can be written out, the data section must be scanned1775 * to save away the 1st word of each BBSIZE block into the header.  We replace1776 * it with the current cycle count.  Each BBSIZE block is tagged with the1777 * cycle count because there in an implicit assumption that drives will1778 * guarantee that entire 512 byte blocks get written at once.  In other words,1779 * we can't have part of a 512 byte block written and part not written.  By1780 * tagging each block, we will know which blocks are valid when recovering1781 * after an unclean shutdown.1782 *1783 * This routine is single threaded on the iclog.  No other thread can be in1784 * this routine with the same iclog.  Changing contents of iclog can there-1785 * fore be done without grabbing the state machine lock.  Updating the global1786 * log will require grabbing the lock though.1787 *1788 * The entire log manager uses a logical block numbering scheme.  Only1789 * xlog_write_iclog knows about the fact that the log may not start with1790 * block zero on a given device.1791 */1792STATIC void1793xlog_sync(1794	struct xlog		*log,1795	struct xlog_in_core	*iclog,1796	struct xlog_ticket	*ticket)1797{1798	unsigned int		count;		/* byte count of bwrite */1799	unsigned int		roundoff;       /* roundoff to BB or stripe */1800	uint64_t		bno;1801	unsigned int		size;1802 1803	ASSERT(atomic_read(&iclog->ic_refcnt) == 0);1804	trace_xlog_iclog_sync(iclog, _RET_IP_);1805 1806	count = xlog_calc_iclog_size(log, iclog, &roundoff);1807 1808	/*1809	 * If we have a ticket, account for the roundoff via the ticket1810	 * reservation to avoid touching the hot grant heads needlessly.1811	 * Otherwise, we have to move grant heads directly.1812	 */1813	if (ticket) {1814		ticket->t_curr_res -= roundoff;1815	} else {1816		xlog_grant_add_space(&log->l_reserve_head, roundoff);1817		xlog_grant_add_space(&log->l_write_head, roundoff);1818	}1819 1820	/* put cycle number in every block */1821	xlog_pack_data(log, iclog, roundoff);1822 1823	/* real byte length */1824	size = iclog->ic_offset;1825	if (xfs_has_logv2(log->l_mp))1826		size += roundoff;1827	iclog->ic_header.h_len = cpu_to_be32(size);1828 1829	XFS_STATS_INC(log->l_mp, xs_log_writes);1830	XFS_STATS_ADD(log->l_mp, xs_log_blocks, BTOBB(count));1831 1832	bno = BLOCK_LSN(be64_to_cpu(iclog->ic_header.h_lsn));1833 1834	/* Do we need to split this write into 2 parts? */1835	if (bno + BTOBB(count) > log->l_logBBsize)1836		xlog_split_iclog(log, &iclog->ic_header, bno, count);1837 1838	/* calculcate the checksum */1839	iclog->ic_header.h_crc = xlog_cksum(log, &iclog->ic_header,1840					    iclog->ic_datap, size);1841	/*1842	 * Intentionally corrupt the log record CRC based on the error injection1843	 * frequency, if defined. This facilitates testing log recovery in the1844	 * event of torn writes. Hence, set the IOABORT state to abort the log1845	 * write on I/O completion and shutdown the fs. The subsequent mount1846	 * detects the bad CRC and attempts to recover.1847	 */1848#ifdef DEBUG1849	if (XFS_TEST_ERROR(false, log->l_mp, XFS_ERRTAG_LOG_BAD_CRC)) {1850		iclog->ic_header.h_crc &= cpu_to_le32(0xAAAAAAAA);1851		iclog->ic_fail_crc = true;1852		xfs_warn(log->l_mp,1853	"Intentionally corrupted log record at LSN 0x%llx. Shutdown imminent.",1854			 be64_to_cpu(iclog->ic_header.h_lsn));1855	}1856#endif1857	xlog_verify_iclog(log, iclog, count);1858	xlog_write_iclog(log, iclog, bno, count);1859}1860 1861/*1862 * Deallocate a log structure1863 */1864STATIC void1865xlog_dealloc_log(1866	struct xlog	*log)1867{1868	xlog_in_core_t	*iclog, *next_iclog;1869	int		i;1870 1871	/*1872	 * Destroy the CIL after waiting for iclog IO completion because an1873	 * iclog EIO error will try to shut down the log, which accesses the1874	 * CIL to wake up the waiters.1875	 */1876	xlog_cil_destroy(log);1877 1878	iclog = log->l_iclog;1879	for (i = 0; i < log->l_iclog_bufs; i++) {1880		next_iclog = iclog->ic_next;1881		kvfree(iclog->ic_data);1882		kfree(iclog);1883		iclog = next_iclog;1884	}1885 1886	log->l_mp->m_log = NULL;1887	destroy_workqueue(log->l_ioend_workqueue);1888	kfree(log);1889}1890 1891/*1892 * Update counters atomically now that memcpy is done.1893 */1894static inline void1895xlog_state_finish_copy(1896	struct xlog		*log,1897	struct xlog_in_core	*iclog,1898	int			record_cnt,1899	int			copy_bytes)1900{1901	lockdep_assert_held(&log->l_icloglock);1902 1903	be32_add_cpu(&iclog->ic_header.h_num_logops, record_cnt);1904	iclog->ic_offset += copy_bytes;1905}1906 1907/*1908 * print out info relating to regions written which consume1909 * the reservation1910 */1911void1912xlog_print_tic_res(1913	struct xfs_mount	*mp,1914	struct xlog_ticket	*ticket)1915{1916	xfs_warn(mp, "ticket reservation summary:");1917	xfs_warn(mp, "  unit res    = %d bytes", ticket->t_unit_res);1918	xfs_warn(mp, "  current res = %d bytes", ticket->t_curr_res);1919	xfs_warn(mp, "  original count  = %d", ticket->t_ocnt);1920	xfs_warn(mp, "  remaining count = %d", ticket->t_cnt);1921}1922 1923/*1924 * Print a summary of the transaction.1925 */1926void1927xlog_print_trans(1928	struct xfs_trans	*tp)1929{1930	struct xfs_mount	*mp = tp->t_mountp;1931	struct xfs_log_item	*lip;1932 1933	/* dump core transaction and ticket info */1934	xfs_warn(mp, "transaction summary:");1935	xfs_warn(mp, "  log res   = %d", tp->t_log_res);1936	xfs_warn(mp, "  log count = %d", tp->t_log_count);1937	xfs_warn(mp, "  flags     = 0x%x", tp->t_flags);1938 1939	xlog_print_tic_res(mp, tp->t_ticket);1940 1941	/* dump each log item */1942	list_for_each_entry(lip, &tp->t_items, li_trans) {1943		struct xfs_log_vec	*lv = lip->li_lv;1944		struct xfs_log_iovec	*vec;1945		int			i;1946 1947		xfs_warn(mp, "log item: ");1948		xfs_warn(mp, "  type	= 0x%x", lip->li_type);1949		xfs_warn(mp, "  flags	= 0x%lx", lip->li_flags);1950		if (!lv)1951			continue;1952		xfs_warn(mp, "  niovecs	= %d", lv->lv_niovecs);1953		xfs_warn(mp, "  size	= %d", lv->lv_size);1954		xfs_warn(mp, "  bytes	= %d", lv->lv_bytes);1955		xfs_warn(mp, "  buf len	= %d", lv->lv_buf_len);1956 1957		/* dump each iovec for the log item */1958		vec = lv->lv_iovecp;1959		for (i = 0; i < lv->lv_niovecs; i++) {1960			int dumplen = min(vec->i_len, 32);1961 1962			xfs_warn(mp, "  iovec[%d]", i);1963			xfs_warn(mp, "    type	= 0x%x", vec->i_type);1964			xfs_warn(mp, "    len	= %d", vec->i_len);1965			xfs_warn(mp, "    first %d bytes of iovec[%d]:", dumplen, i);1966			xfs_hex_dump(vec->i_addr, dumplen);1967 1968			vec++;1969		}1970	}1971}1972 1973static inline void1974xlog_write_iovec(1975	struct xlog_in_core	*iclog,1976	uint32_t		*log_offset,1977	void			*data,1978	uint32_t		write_len,1979	int			*bytes_left,1980	uint32_t		*record_cnt,1981	uint32_t		*data_cnt)1982{1983	ASSERT(*log_offset < iclog->ic_log->l_iclog_size);1984	ASSERT(*log_offset % sizeof(int32_t) == 0);1985	ASSERT(write_len % sizeof(int32_t) == 0);1986 1987	memcpy(iclog->ic_datap + *log_offset, data, write_len);1988	*log_offset += write_len;1989	*bytes_left -= write_len;1990	(*record_cnt)++;1991	*data_cnt += write_len;1992}1993 1994/*1995 * Write log vectors into a single iclog which is guaranteed by the caller1996 * to have enough space to write the entire log vector into.1997 */1998static void1999xlog_write_full(2000	struct xfs_log_vec	*lv,2001	struct xlog_ticket	*ticket,2002	struct xlog_in_core	*iclog,2003	uint32_t		*log_offset,2004	uint32_t		*len,2005	uint32_t		*record_cnt,2006	uint32_t		*data_cnt)2007{2008	int			index;2009 2010	ASSERT(*log_offset + *len <= iclog->ic_size ||2011		iclog->ic_state == XLOG_STATE_WANT_SYNC);2012 2013	/*2014	 * Ordered log vectors have no regions to write so this2015	 * loop will naturally skip them.2016	 */2017	for (index = 0; index < lv->lv_niovecs; index++) {2018		struct xfs_log_iovec	*reg = &lv->lv_iovecp[index];2019		struct xlog_op_header	*ophdr = reg->i_addr;2020 2021		ophdr->oh_tid = cpu_to_be32(ticket->t_tid);2022		xlog_write_iovec(iclog, log_offset, reg->i_addr,2023				reg->i_len, len, record_cnt, data_cnt);2024	}2025}2026 2027static int2028xlog_write_get_more_iclog_space(2029	struct xlog_ticket	*ticket,2030	struct xlog_in_core	**iclogp,2031	uint32_t		*log_offset,2032	uint32_t		len,2033	uint32_t		*record_cnt,2034	uint32_t		*data_cnt)2035{2036	struct xlog_in_core	*iclog = *iclogp;2037	struct xlog		*log = iclog->ic_log;2038	int			error;2039 2040	spin_lock(&log->l_icloglock);2041	ASSERT(iclog->ic_state == XLOG_STATE_WANT_SYNC);2042	xlog_state_finish_copy(log, iclog, *record_cnt, *data_cnt);2043	error = xlog_state_release_iclog(log, iclog, ticket);2044	spin_unlock(&log->l_icloglock);2045	if (error)2046		return error;2047 2048	error = xlog_state_get_iclog_space(log, len, &iclog, ticket,2049					log_offset);2050	if (error)2051		return error;2052	*record_cnt = 0;2053	*data_cnt = 0;2054	*iclogp = iclog;2055	return 0;2056}2057 2058/*2059 * Write log vectors into a single iclog which is smaller than the current chain2060 * length. We write until we cannot fit a full record into the remaining space2061 * and then stop. We return the log vector that is to be written that cannot2062 * wholly fit in the iclog.2063 */2064static int2065xlog_write_partial(2066	struct xfs_log_vec	*lv,2067	struct xlog_ticket	*ticket,2068	struct xlog_in_core	**iclogp,2069	uint32_t		*log_offset,2070	uint32_t		*len,2071	uint32_t		*record_cnt,2072	uint32_t		*data_cnt)2073{2074	struct xlog_in_core	*iclog = *iclogp;2075	struct xlog_op_header	*ophdr;2076	int			index = 0;2077	uint32_t		rlen;2078	int			error;2079 2080	/* walk the logvec, copying until we run out of space in the iclog */2081	for (index = 0; index < lv->lv_niovecs; index++) {2082		struct xfs_log_iovec	*reg = &lv->lv_iovecp[index];2083		uint32_t		reg_offset = 0;2084 2085		/*2086		 * The first region of a continuation must have a non-zero2087		 * length otherwise log recovery will just skip over it and2088		 * start recovering from the next opheader it finds. Because we2089		 * mark the next opheader as a continuation, recovery will then2090		 * incorrectly add the continuation to the previous region and2091		 * that breaks stuff.2092		 *2093		 * Hence if there isn't space for region data after the2094		 * opheader, then we need to start afresh with a new iclog.2095		 */2096		if (iclog->ic_size - *log_offset <=2097					sizeof(struct xlog_op_header)) {2098			error = xlog_write_get_more_iclog_space(ticket,2099					&iclog, log_offset, *len, record_cnt,2100					data_cnt);2101			if (error)2102				return error;2103		}2104 2105		ophdr = reg->i_addr;2106		rlen = min_t(uint32_t, reg->i_len, iclog->ic_size - *log_offset);2107 2108		ophdr->oh_tid = cpu_to_be32(ticket->t_tid);2109		ophdr->oh_len = cpu_to_be32(rlen - sizeof(struct xlog_op_header));2110		if (rlen != reg->i_len)2111			ophdr->oh_flags |= XLOG_CONTINUE_TRANS;2112 2113		xlog_write_iovec(iclog, log_offset, reg->i_addr,2114				rlen, len, record_cnt, data_cnt);2115 2116		/* If we wrote the whole region, move to the next. */2117		if (rlen == reg->i_len)2118			continue;2119 2120		/*2121		 * We now have a partially written iovec, but it can span2122		 * multiple iclogs so we loop here. First we release the iclog2123		 * we currently have, then we get a new iclog and add a new2124		 * opheader. Then we continue copying from where we were until2125		 * we either complete the iovec or fill the iclog. If we2126		 * complete the iovec, then we increment the index and go right2127		 * back to the top of the outer loop. if we fill the iclog, we2128		 * run the inner loop again.2129		 *2130		 * This is complicated by the tail of a region using all the2131		 * space in an iclog and hence requiring us to release the iclog2132		 * and get a new one before returning to the outer loop. We must2133		 * always guarantee that we exit this inner loop with at least2134		 * space for log transaction opheaders left in the current2135		 * iclog, hence we cannot just terminate the loop at the end2136		 * of the of the continuation. So we loop while there is no2137		 * space left in the current iclog, and check for the end of the2138		 * continuation after getting a new iclog.2139		 */2140		do {2141			/*2142			 * Ensure we include the continuation opheader in the2143			 * space we need in the new iclog by adding that size2144			 * to the length we require. This continuation opheader2145			 * needs to be accounted to the ticket as the space it2146			 * consumes hasn't been accounted to the lv we are2147			 * writing.2148			 */2149			error = xlog_write_get_more_iclog_space(ticket,2150					&iclog, log_offset,2151					*len + sizeof(struct xlog_op_header),2152					record_cnt, data_cnt);2153			if (error)2154				return error;2155 2156			ophdr = iclog->ic_datap + *log_offset;2157			ophdr->oh_tid = cpu_to_be32(ticket->t_tid);2158			ophdr->oh_clientid = XFS_TRANSACTION;2159			ophdr->oh_res2 = 0;2160			ophdr->oh_flags = XLOG_WAS_CONT_TRANS;2161 2162			ticket->t_curr_res -= sizeof(struct xlog_op_header);2163			*log_offset += sizeof(struct xlog_op_header);2164			*data_cnt += sizeof(struct xlog_op_header);2165 2166			/*2167			 * If rlen fits in the iclog, then end the region2168			 * continuation. Otherwise we're going around again.2169			 */2170			reg_offset += rlen;2171			rlen = reg->i_len - reg_offset;2172			if (rlen <= iclog->ic_size - *log_offset)2173				ophdr->oh_flags |= XLOG_END_TRANS;2174			else2175				ophdr->oh_flags |= XLOG_CONTINUE_TRANS;2176 2177			rlen = min_t(uint32_t, rlen, iclog->ic_size - *log_offset);2178			ophdr->oh_len = cpu_to_be32(rlen);2179 2180			xlog_write_iovec(iclog, log_offset,2181					reg->i_addr + reg_offset,2182					rlen, len, record_cnt, data_cnt);2183 2184		} while (ophdr->oh_flags & XLOG_CONTINUE_TRANS);2185	}2186 2187	/*2188	 * No more iovecs remain in this logvec so return the next log vec to2189	 * the caller so it can go back to fast path copying.2190	 */2191	*iclogp = iclog;2192	return 0;2193}2194 2195/*2196 * Write some region out to in-core log2197 *2198 * This will be called when writing externally provided regions or when2199 * writing out a commit record for a given transaction.2200 *2201 * General algorithm:2202 *	1. Find total length of this write.  This may include adding to the2203 *		lengths passed in.2204 *	2. Check whether we violate the tickets reservation.2205 *	3. While writing to this iclog2206 *	    A. Reserve as much space in this iclog as can get2207 *	    B. If this is first write, save away start lsn2208 *	    C. While writing this region:2209 *		1. If first write of transaction, write start record2210 *		2. Write log operation header (header per region)2211 *		3. Find out if we can fit entire region into this iclog2212 *		4. Potentially, verify destination memcpy ptr2213 *		5. Memcpy (partial) region2214 *		6. If partial copy, release iclog; otherwise, continue2215 *			copying more regions into current iclog2216 *	4. Mark want sync bit (in simulation mode)2217 *	5. Release iclog for potential flush to on-disk log.2218 *2219 * ERRORS:2220 * 1.	Panic if reservation is overrun.  This should never happen since2221 *	reservation amounts are generated internal to the filesystem.2222 * NOTES:2223 * 1. Tickets are single threaded data structures.2224 * 2. The XLOG_END_TRANS & XLOG_CONTINUE_TRANS flags are passed down to the2225 *	syncing routine.  When a single log_write region needs to span2226 *	multiple in-core logs, the XLOG_CONTINUE_TRANS bit should be set2227 *	on all log operation writes which don't contain the end of the2228 *	region.  The XLOG_END_TRANS bit is used for the in-core log2229 *	operation which contains the end of the continued log_write region.2230 * 3. When xlog_state_get_iclog_space() grabs the rest of the current iclog,2231 *	we don't really know exactly how much space will be used.  As a result,2232 *	we don't update ic_offset until the end when we know exactly how many2233 *	bytes have been written out.2234 */2235int2236xlog_write(2237	struct xlog		*log,2238	struct xfs_cil_ctx	*ctx,2239	struct list_head	*lv_chain,2240	struct xlog_ticket	*ticket,2241	uint32_t		len)2242 2243{2244	struct xlog_in_core	*iclog = NULL;2245	struct xfs_log_vec	*lv;2246	uint32_t		record_cnt = 0;2247	uint32_t		data_cnt = 0;2248	int			error = 0;2249	int			log_offset;2250 2251	if (ticket->t_curr_res < 0) {2252		xfs_alert_tag(log->l_mp, XFS_PTAG_LOGRES,2253		     "ctx ticket reservation ran out. Need to up reservation");2254		xlog_print_tic_res(log->l_mp, ticket);2255		xlog_force_shutdown(log, SHUTDOWN_LOG_IO_ERROR);2256	}2257 2258	error = xlog_state_get_iclog_space(log, len, &iclog, ticket,2259					   &log_offset);2260	if (error)2261		return error;2262 2263	ASSERT(log_offset <= iclog->ic_size - 1);2264 2265	/*2266	 * If we have a context pointer, pass it the first iclog we are2267	 * writing to so it can record state needed for iclog write2268	 * ordering.2269	 */2270	if (ctx)2271		xlog_cil_set_ctx_write_state(ctx, iclog);2272 2273	list_for_each_entry(lv, lv_chain, lv_list) {2274		/*2275		 * If the entire log vec does not fit in the iclog, punt it to2276		 * the partial copy loop which can handle this case.2277		 */2278		if (lv->lv_niovecs &&2279		    lv->lv_bytes > iclog->ic_size - log_offset) {2280			error = xlog_write_partial(lv, ticket, &iclog,2281					&log_offset, &len, &record_cnt,2282					&data_cnt);2283			if (error) {2284				/*2285				 * We have no iclog to release, so just return2286				 * the error immediately.2287				 */2288				return error;2289			}2290		} else {2291			xlog_write_full(lv, ticket, iclog, &log_offset,2292					 &len, &record_cnt, &data_cnt);2293		}2294	}2295	ASSERT(len == 0);2296 2297	/*2298	 * We've already been guaranteed that the last writes will fit inside2299	 * the current iclog, and hence it will already have the space used by2300	 * those writes accounted to it. Hence we do not need to update the2301	 * iclog with the number of bytes written here.2302	 */2303	spin_lock(&log->l_icloglock);2304	xlog_state_finish_copy(log, iclog, record_cnt, 0);2305	error = xlog_state_release_iclog(log, iclog, ticket);2306	spin_unlock(&log->l_icloglock);2307 2308	return error;2309}2310 2311static void2312xlog_state_activate_iclog(2313	struct xlog_in_core	*iclog,2314	int			*iclogs_changed)2315{2316	ASSERT(list_empty_careful(&iclog->ic_callbacks));2317	trace_xlog_iclog_activate(iclog, _RET_IP_);2318 2319	/*2320	 * If the number of ops in this iclog indicate it just contains the2321	 * dummy transaction, we can change state into IDLE (the second time2322	 * around). Otherwise we should change the state into NEED a dummy.2323	 * We don't need to cover the dummy.2324	 */2325	if (*iclogs_changed == 0 &&2326	    iclog->ic_header.h_num_logops == cpu_to_be32(XLOG_COVER_OPS)) {2327		*iclogs_changed = 1;2328	} else {2329		/*2330		 * We have two dirty iclogs so start over.  This could also be2331		 * num of ops indicating this is not the dummy going out.2332		 */2333		*iclogs_changed = 2;2334	}2335 2336	iclog->ic_state	= XLOG_STATE_ACTIVE;2337	iclog->ic_offset = 0;2338	iclog->ic_header.h_num_logops = 0;2339	memset(iclog->ic_header.h_cycle_data, 0,2340		sizeof(iclog->ic_header.h_cycle_data));2341	iclog->ic_header.h_lsn = 0;2342	iclog->ic_header.h_tail_lsn = 0;2343}2344 2345/*2346 * Loop through all iclogs and mark all iclogs currently marked DIRTY as2347 * ACTIVE after iclog I/O has completed.2348 */2349static void2350xlog_state_activate_iclogs(2351	struct xlog		*log,2352	int			*iclogs_changed)2353{2354	struct xlog_in_core	*iclog = log->l_iclog;2355 2356	do {2357		if (iclog->ic_state == XLOG_STATE_DIRTY)2358			xlog_state_activate_iclog(iclog, iclogs_changed);2359		/*2360		 * The ordering of marking iclogs ACTIVE must be maintained, so2361		 * an iclog doesn't become ACTIVE beyond one that is SYNCING.2362		 */2363		else if (iclog->ic_state != XLOG_STATE_ACTIVE)2364			break;2365	} while ((iclog = iclog->ic_next) != log->l_iclog);2366}2367 2368static int2369xlog_covered_state(2370	int			prev_state,2371	int			iclogs_changed)2372{2373	/*2374	 * We go to NEED for any non-covering writes. We go to NEED2 if we just2375	 * wrote the first covering record (DONE). We go to IDLE if we just2376	 * wrote the second covering record (DONE2) and remain in IDLE until a2377	 * non-covering write occurs.2378	 */2379	switch (prev_state) {2380	case XLOG_STATE_COVER_IDLE:2381		if (iclogs_changed == 1)2382			return XLOG_STATE_COVER_IDLE;2383		fallthrough;2384	case XLOG_STATE_COVER_NEED:2385	case XLOG_STATE_COVER_NEED2:2386		break;2387	case XLOG_STATE_COVER_DONE:2388		if (iclogs_changed == 1)2389			return XLOG_STATE_COVER_NEED2;2390		break;2391	case XLOG_STATE_COVER_DONE2:2392		if (iclogs_changed == 1)2393			return XLOG_STATE_COVER_IDLE;2394		break;2395	default:2396		ASSERT(0);2397	}2398 2399	return XLOG_STATE_COVER_NEED;2400}2401 2402STATIC void2403xlog_state_clean_iclog(2404	struct xlog		*log,2405	struct xlog_in_core	*dirty_iclog)2406{2407	int			iclogs_changed = 0;2408 2409	trace_xlog_iclog_clean(dirty_iclog, _RET_IP_);2410 2411	dirty_iclog->ic_state = XLOG_STATE_DIRTY;2412 2413	xlog_state_activate_iclogs(log, &iclogs_changed);2414	wake_up_all(&dirty_iclog->ic_force_wait);2415 2416	if (iclogs_changed) {2417		log->l_covered_state = xlog_covered_state(log->l_covered_state,2418				iclogs_changed);2419	}2420}2421 2422STATIC xfs_lsn_t2423xlog_get_lowest_lsn(2424	struct xlog		*log)2425{2426	struct xlog_in_core	*iclog = log->l_iclog;2427	xfs_lsn_t		lowest_lsn = 0, lsn;2428 2429	do {2430		if (iclog->ic_state == XLOG_STATE_ACTIVE ||2431		    iclog->ic_state == XLOG_STATE_DIRTY)2432			continue;2433 2434		lsn = be64_to_cpu(iclog->ic_header.h_lsn);2435		if ((lsn && !lowest_lsn) || XFS_LSN_CMP(lsn, lowest_lsn) < 0)2436			lowest_lsn = lsn;2437	} while ((iclog = iclog->ic_next) != log->l_iclog);2438 2439	return lowest_lsn;2440}2441 2442/*2443 * Return true if we need to stop processing, false to continue to the next2444 * iclog. The caller will need to run callbacks if the iclog is returned in the2445 * XLOG_STATE_CALLBACK state.2446 */2447static bool2448xlog_state_iodone_process_iclog(2449	struct xlog		*log,2450	struct xlog_in_core	*iclog)2451{2452	xfs_lsn_t		lowest_lsn;2453	xfs_lsn_t		header_lsn;2454 2455	switch (iclog->ic_state) {2456	case XLOG_STATE_ACTIVE:2457	case XLOG_STATE_DIRTY:2458		/*2459		 * Skip all iclogs in the ACTIVE & DIRTY states:2460		 */2461		return false;2462	case XLOG_STATE_DONE_SYNC:2463		/*2464		 * Now that we have an iclog that is in the DONE_SYNC state, do2465		 * one more check here to see if we have chased our tail around.2466		 * If this is not the lowest lsn iclog, then we will leave it2467		 * for another completion to process.2468		 */2469		header_lsn = be64_to_cpu(iclog->ic_header.h_lsn);2470		lowest_lsn = xlog_get_lowest_lsn(log);2471		if (lowest_lsn && XFS_LSN_CMP(lowest_lsn, header_lsn) < 0)2472			return false;2473		/*2474		 * If there are no callbacks on this iclog, we can mark it clean2475		 * immediately and return. Otherwise we need to run the2476		 * callbacks.2477		 */2478		if (list_empty(&iclog->ic_callbacks)) {2479			xlog_state_clean_iclog(log, iclog);2480			return false;2481		}2482		trace_xlog_iclog_callback(iclog, _RET_IP_);2483		iclog->ic_state = XLOG_STATE_CALLBACK;2484		return false;2485	default:2486		/*2487		 * Can only perform callbacks in order.  Since this iclog is not2488		 * in the DONE_SYNC state, we skip the rest and just try to2489		 * clean up.2490		 */2491		return true;2492	}2493}2494 2495/*2496 * Loop over all the iclogs, running attached callbacks on them. Return true if2497 * we ran any callbacks, indicating that we dropped the icloglock. We don't need2498 * to handle transient shutdown state here at all because2499 * xlog_state_shutdown_callbacks() will be run to do the necessary shutdown2500 * cleanup of the callbacks.2501 */2502static bool2503xlog_state_do_iclog_callbacks(2504	struct xlog		*log)2505		__releases(&log->l_icloglock)2506		__acquires(&log->l_icloglock)2507{2508	struct xlog_in_core	*first_iclog = log->l_iclog;2509	struct xlog_in_core	*iclog = first_iclog;2510	bool			ran_callback = false;2511 2512	do {2513		LIST_HEAD(cb_list);2514 2515		if (xlog_state_iodone_process_iclog(log, iclog))2516			break;2517		if (iclog->ic_state != XLOG_STATE_CALLBACK) {2518			iclog = iclog->ic_next;2519			continue;2520		}2521		list_splice_init(&iclog->ic_callbacks, &cb_list);2522		spin_unlock(&log->l_icloglock);2523 2524		trace_xlog_iclog_callbacks_start(iclog, _RET_IP_);2525		xlog_cil_process_committed(&cb_list);2526		trace_xlog_iclog_callbacks_done(iclog, _RET_IP_);2527		ran_callback = true;2528 2529		spin_lock(&log->l_icloglock);2530		xlog_state_clean_iclog(log, iclog);2531		iclog = iclog->ic_next;2532	} while (iclog != first_iclog);2533 2534	return ran_callback;2535}2536 2537 2538/*2539 * Loop running iclog completion callbacks until there are no more iclogs in a2540 * state that can run callbacks.2541 */2542STATIC void2543xlog_state_do_callback(2544	struct xlog		*log)2545{2546	int			flushcnt = 0;2547	int			repeats = 0;2548 2549	spin_lock(&log->l_icloglock);2550	while (xlog_state_do_iclog_callbacks(log)) {2551		if (xlog_is_shutdown(log))2552			break;2553 2554		if (++repeats > 5000) {2555			flushcnt += repeats;2556			repeats = 0;2557			xfs_warn(log->l_mp,2558				"%s: possible infinite loop (%d iterations)",2559				__func__, flushcnt);2560		}2561	}2562 2563	if (log->l_iclog->ic_state == XLOG_STATE_ACTIVE)2564		wake_up_all(&log->l_flush_wait);2565 2566	spin_unlock(&log->l_icloglock);2567}2568 2569 2570/*2571 * Finish transitioning this iclog to the dirty state.2572 *2573 * Callbacks could take time, so they are done outside the scope of the2574 * global state machine log lock.2575 */2576STATIC void2577xlog_state_done_syncing(2578	struct xlog_in_core	*iclog)2579{2580	struct xlog		*log = iclog->ic_log;2581 2582	spin_lock(&log->l_icloglock);2583	ASSERT(atomic_read(&iclog->ic_refcnt) == 0);2584	trace_xlog_iclog_sync_done(iclog, _RET_IP_);2585 2586	/*2587	 * If we got an error, either on the first buffer, or in the case of2588	 * split log writes, on the second, we shut down the file system and2589	 * no iclogs should ever be attempted to be written to disk again.2590	 */2591	if (!xlog_is_shutdown(log)) {2592		ASSERT(iclog->ic_state == XLOG_STATE_SYNCING);2593		iclog->ic_state = XLOG_STATE_DONE_SYNC;2594	}2595 2596	/*2597	 * Someone could be sleeping prior to writing out the next2598	 * iclog buffer, we wake them all, one will get to do the2599	 * I/O, the others get to wait for the result.2600	 */2601	wake_up_all(&iclog->ic_write_wait);2602	spin_unlock(&log->l_icloglock);2603	xlog_state_do_callback(log);2604}2605 2606/*2607 * If the head of the in-core log ring is not (ACTIVE or DIRTY), then we must2608 * sleep.  We wait on the flush queue on the head iclog as that should be2609 * the first iclog to complete flushing. Hence if all iclogs are syncing,2610 * we will wait here and all new writes will sleep until a sync completes.2611 *2612 * The in-core logs are used in a circular fashion. They are not used2613 * out-of-order even when an iclog past the head is free.2614 *2615 * return:2616 *	* log_offset where xlog_write() can start writing into the in-core2617 *		log's data space.2618 *	* in-core log pointer to which xlog_write() should write.2619 *	* boolean indicating this is a continued write to an in-core log.2620 *		If this is the last write, then the in-core log's offset field2621 *		needs to be incremented, depending on the amount of data which2622 *		is copied.2623 */2624STATIC int2625xlog_state_get_iclog_space(2626	struct xlog		*log,2627	int			len,2628	struct xlog_in_core	**iclogp,2629	struct xlog_ticket	*ticket,2630	int			*logoffsetp)2631{2632	int		  log_offset;2633	xlog_rec_header_t *head;2634	xlog_in_core_t	  *iclog;2635 2636restart:2637	spin_lock(&log->l_icloglock);2638	if (xlog_is_shutdown(log)) {2639		spin_unlock(&log->l_icloglock);2640		return -EIO;2641	}2642 2643	iclog = log->l_iclog;2644	if (iclog->ic_state != XLOG_STATE_ACTIVE) {2645		XFS_STATS_INC(log->l_mp, xs_log_noiclogs);2646 2647		/* Wait for log writes to have flushed */2648		xlog_wait(&log->l_flush_wait, &log->l_icloglock);2649		goto restart;2650	}2651 2652	head = &iclog->ic_header;2653 2654	atomic_inc(&iclog->ic_refcnt);	/* prevents sync */2655	log_offset = iclog->ic_offset;2656 2657	trace_xlog_iclog_get_space(iclog, _RET_IP_);2658 2659	/* On the 1st write to an iclog, figure out lsn.  This works2660	 * if iclogs marked XLOG_STATE_WANT_SYNC always write out what they are2661	 * committing to.  If the offset is set, that's how many blocks2662	 * must be written.2663	 */2664	if (log_offset == 0) {2665		ticket->t_curr_res -= log->l_iclog_hsize;2666		head->h_cycle = cpu_to_be32(log->l_curr_cycle);2667		head->h_lsn = cpu_to_be64(2668			xlog_assign_lsn(log->l_curr_cycle, log->l_curr_block));2669		ASSERT(log->l_curr_block >= 0);2670	}2671 2672	/* If there is enough room to write everything, then do it.  Otherwise,2673	 * claim the rest of the region and make sure the XLOG_STATE_WANT_SYNC2674	 * bit is on, so this will get flushed out.  Don't update ic_offset2675	 * until you know exactly how many bytes get copied.  Therefore, wait2676	 * until later to update ic_offset.2677	 *2678	 * xlog_write() algorithm assumes that at least 2 xlog_op_header_t's2679	 * can fit into remaining data section.2680	 */2681	if (iclog->ic_size - iclog->ic_offset < 2*sizeof(xlog_op_header_t)) {2682		int		error = 0;2683 2684		xlog_state_switch_iclogs(log, iclog, iclog->ic_size);2685 2686		/*2687		 * If we are the only one writing to this iclog, sync it to2688		 * disk.  We need to do an atomic compare and decrement here to2689		 * avoid racing with concurrent atomic_dec_and_lock() calls in2690		 * xlog_state_release_iclog() when there is more than one2691		 * reference to the iclog.2692		 */2693		if (!atomic_add_unless(&iclog->ic_refcnt, -1, 1))2694			error = xlog_state_release_iclog(log, iclog, ticket);2695		spin_unlock(&log->l_icloglock);2696		if (error)2697			return error;2698		goto restart;2699	}2700 2701	/* Do we have enough room to write the full amount in the remainder2702	 * of this iclog?  Or must we continue a write on the next iclog and2703	 * mark this iclog as completely taken?  In the case where we switch2704	 * iclogs (to mark it taken), this particular iclog will release/sync2705	 * to disk in xlog_write().2706	 */2707	if (len <= iclog->ic_size - iclog->ic_offset)2708		iclog->ic_offset += len;2709	else2710		xlog_state_switch_iclogs(log, iclog, iclog->ic_size);2711	*iclogp = iclog;2712 2713	ASSERT(iclog->ic_offset <= iclog->ic_size);2714	spin_unlock(&log->l_icloglock);2715 2716	*logoffsetp = log_offset;2717	return 0;2718}2719 2720/*2721 * The first cnt-1 times a ticket goes through here we don't need to move the2722 * grant write head because the permanent reservation has reserved cnt times the2723 * unit amount.  Release part of current permanent unit reservation and reset2724 * current reservation to be one units worth.  Also move grant reservation head2725 * forward.2726 */2727void2728xfs_log_ticket_regrant(2729	struct xlog		*log,2730	struct xlog_ticket	*ticket)2731{2732	trace_xfs_log_ticket_regrant(log, ticket);2733 2734	if (ticket->t_cnt > 0)2735		ticket->t_cnt--;2736 2737	xlog_grant_sub_space(&log->l_reserve_head, ticket->t_curr_res);2738	xlog_grant_sub_space(&log->l_write_head, ticket->t_curr_res);2739	ticket->t_curr_res = ticket->t_unit_res;2740 2741	trace_xfs_log_ticket_regrant_sub(log, ticket);2742 2743	/* just return if we still have some of the pre-reserved space */2744	if (!ticket->t_cnt) {2745		xlog_grant_add_space(&log->l_reserve_head, ticket->t_unit_res);2746		trace_xfs_log_ticket_regrant_exit(log, ticket);2747 2748		ticket->t_curr_res = ticket->t_unit_res;2749	}2750 2751	xfs_log_ticket_put(ticket);2752}2753 2754/*2755 * Give back the space left from a reservation.2756 *2757 * All the information we need to make a correct determination of space left2758 * is present.  For non-permanent reservations, things are quite easy.  The2759 * count should have been decremented to zero.  We only need to deal with the2760 * space remaining in the current reservation part of the ticket.  If the2761 * ticket contains a permanent reservation, there may be left over space which2762 * needs to be released.  A count of N means that N-1 refills of the current2763 * reservation can be done before we need to ask for more space.  The first2764 * one goes to fill up the first current reservation.  Once we run out of2765 * space, the count will stay at zero and the only space remaining will be2766 * in the current reservation field.2767 */2768void2769xfs_log_ticket_ungrant(2770	struct xlog		*log,2771	struct xlog_ticket	*ticket)2772{2773	int			bytes;2774 2775	trace_xfs_log_ticket_ungrant(log, ticket);2776 2777	if (ticket->t_cnt > 0)2778		ticket->t_cnt--;2779 2780	trace_xfs_log_ticket_ungrant_sub(log, ticket);2781 2782	/*2783	 * If this is a permanent reservation ticket, we may be able to free2784	 * up more space based on the remaining count.2785	 */2786	bytes = ticket->t_curr_res;2787	if (ticket->t_cnt > 0) {2788		ASSERT(ticket->t_flags & XLOG_TIC_PERM_RESERV);2789		bytes += ticket->t_unit_res*ticket->t_cnt;2790	}2791 2792	xlog_grant_sub_space(&log->l_reserve_head, bytes);2793	xlog_grant_sub_space(&log->l_write_head, bytes);2794 2795	trace_xfs_log_ticket_ungrant_exit(log, ticket);2796 2797	xfs_log_space_wake(log->l_mp);2798	xfs_log_ticket_put(ticket);2799}2800 2801/*2802 * This routine will mark the current iclog in the ring as WANT_SYNC and move2803 * the current iclog pointer to the next iclog in the ring.2804 */2805void2806xlog_state_switch_iclogs(2807	struct xlog		*log,2808	struct xlog_in_core	*iclog,2809	int			eventual_size)2810{2811	ASSERT(iclog->ic_state == XLOG_STATE_ACTIVE);2812	assert_spin_locked(&log->l_icloglock);2813	trace_xlog_iclog_switch(iclog, _RET_IP_);2814 2815	if (!eventual_size)2816		eventual_size = iclog->ic_offset;2817	iclog->ic_state = XLOG_STATE_WANT_SYNC;2818	iclog->ic_header.h_prev_block = cpu_to_be32(log->l_prev_block);2819	log->l_prev_block = log->l_curr_block;2820	log->l_prev_cycle = log->l_curr_cycle;2821 2822	/* roll log?: ic_offset changed later */2823	log->l_curr_block += BTOBB(eventual_size)+BTOBB(log->l_iclog_hsize);2824 2825	/* Round up to next log-sunit */2826	if (log->l_iclog_roundoff > BBSIZE) {2827		uint32_t sunit_bb = BTOBB(log->l_iclog_roundoff);2828		log->l_curr_block = roundup(log->l_curr_block, sunit_bb);2829	}2830 2831	if (log->l_curr_block >= log->l_logBBsize) {2832		/*2833		 * Rewind the current block before the cycle is bumped to make2834		 * sure that the combined LSN never transiently moves forward2835		 * when the log wraps to the next cycle. This is to support the2836		 * unlocked sample of these fields from xlog_valid_lsn(). Most2837		 * other cases should acquire l_icloglock.2838		 */2839		log->l_curr_block -= log->l_logBBsize;2840		ASSERT(log->l_curr_block >= 0);2841		smp_wmb();2842		log->l_curr_cycle++;2843		if (log->l_curr_cycle == XLOG_HEADER_MAGIC_NUM)2844			log->l_curr_cycle++;2845	}2846	ASSERT(iclog == log->l_iclog);2847	log->l_iclog = iclog->ic_next;2848}2849 2850/*2851 * Force the iclog to disk and check if the iclog has been completed before2852 * xlog_force_iclog() returns. This can happen on synchronous (e.g.2853 * pmem) or fast async storage because we drop the icloglock to issue the IO.2854 * If completion has already occurred, tell the caller so that it can avoid an2855 * unnecessary wait on the iclog.2856 */2857static int2858xlog_force_and_check_iclog(2859	struct xlog_in_core	*iclog,2860	bool			*completed)2861{2862	xfs_lsn_t		lsn = be64_to_cpu(iclog->ic_header.h_lsn);2863	int			error;2864 2865	*completed = false;2866	error = xlog_force_iclog(iclog);2867	if (error)2868		return error;2869 2870	/*2871	 * If the iclog has already been completed and reused the header LSN2872	 * will have been rewritten by completion2873	 */2874	if (be64_to_cpu(iclog->ic_header.h_lsn) != lsn)2875		*completed = true;2876	return 0;2877}2878 2879/*2880 * Write out all data in the in-core log as of this exact moment in time.2881 *2882 * Data may be written to the in-core log during this call.  However,2883 * we don't guarantee this data will be written out.  A change from past2884 * implementation means this routine will *not* write out zero length LRs.2885 *2886 * Basically, we try and perform an intelligent scan of the in-core logs.2887 * If we determine there is no flushable data, we just return.  There is no2888 * flushable data if:2889 *2890 *	1. the current iclog is active and has no data; the previous iclog2891 *		is in the active or dirty state.2892 *	2. the current iclog is drity, and the previous iclog is in the2893 *		active or dirty state.2894 *2895 * We may sleep if:2896 *2897 *	1. the current iclog is not in the active nor dirty state.2898 *	2. the current iclog dirty, and the previous iclog is not in the2899 *		active nor dirty state.2900 *	3. the current iclog is active, and there is another thread writing2901 *		to this particular iclog.2902 *	4. a) the current iclog is active and has no other writers2903 *	   b) when we return from flushing out this iclog, it is still2904 *		not in the active nor dirty state.2905 */2906int2907xfs_log_force(2908	struct xfs_mount	*mp,2909	uint			flags)2910{2911	struct xlog		*log = mp->m_log;2912	struct xlog_in_core	*iclog;2913 2914	XFS_STATS_INC(mp, xs_log_force);2915	trace_xfs_log_force(mp, 0, _RET_IP_);2916 2917	xlog_cil_force(log);2918 2919	spin_lock(&log->l_icloglock);2920	if (xlog_is_shutdown(log))2921		goto out_error;2922 2923	iclog = log->l_iclog;2924	trace_xlog_iclog_force(iclog, _RET_IP_);2925 2926	if (iclog->ic_state == XLOG_STATE_DIRTY ||2927	    (iclog->ic_state == XLOG_STATE_ACTIVE &&2928	     atomic_read(&iclog->ic_refcnt) == 0 && iclog->ic_offset == 0)) {2929		/*2930		 * If the head is dirty or (active and empty), then we need to2931		 * look at the previous iclog.2932		 *2933		 * If the previous iclog is active or dirty we are done.  There2934		 * is nothing to sync out. Otherwise, we attach ourselves to the2935		 * previous iclog and go to sleep.2936		 */2937		iclog = iclog->ic_prev;2938	} else if (iclog->ic_state == XLOG_STATE_ACTIVE) {2939		if (atomic_read(&iclog->ic_refcnt) == 0) {2940			/* We have exclusive access to this iclog. */2941			bool	completed;2942 2943			if (xlog_force_and_check_iclog(iclog, &completed))2944				goto out_error;2945 2946			if (completed)2947				goto out_unlock;2948		} else {2949			/*2950			 * Someone else is still writing to this iclog, so we2951			 * need to ensure that when they release the iclog it2952			 * gets synced immediately as we may be waiting on it.2953			 */2954			xlog_state_switch_iclogs(log, iclog, 0);2955		}2956	}2957 2958	/*2959	 * The iclog we are about to wait on may contain the checkpoint pushed2960	 * by the above xlog_cil_force() call, but it may not have been pushed2961	 * to disk yet. Like the ACTIVE case above, we need to make sure caches2962	 * are flushed when this iclog is written.2963	 */2964	if (iclog->ic_state == XLOG_STATE_WANT_SYNC)2965		iclog->ic_flags |= XLOG_ICL_NEED_FLUSH | XLOG_ICL_NEED_FUA;2966 2967	if (flags & XFS_LOG_SYNC)2968		return xlog_wait_on_iclog(iclog);2969out_unlock:2970	spin_unlock(&log->l_icloglock);2971	return 0;2972out_error:2973	spin_unlock(&log->l_icloglock);2974	return -EIO;2975}2976 2977/*2978 * Force the log to a specific LSN.2979 *2980 * If an iclog with that lsn can be found:2981 *	If it is in the DIRTY state, just return.2982 *	If it is in the ACTIVE state, move the in-core log into the WANT_SYNC2983 *		state and go to sleep or return.2984 *	If it is in any other state, go to sleep or return.2985 *2986 * Synchronous forces are implemented with a wait queue.  All callers trying2987 * to force a given lsn to disk must wait on the queue attached to the2988 * specific in-core log.  When given in-core log finally completes its write2989 * to disk, that thread will wake up all threads waiting on the queue.2990 */2991static int2992xlog_force_lsn(2993	struct xlog		*log,2994	xfs_lsn_t		lsn,2995	uint			flags,2996	int			*log_flushed,2997	bool			already_slept)2998{2999	struct xlog_in_core	*iclog;3000	bool			completed;3001 3002	spin_lock(&log->l_icloglock);3003	if (xlog_is_shutdown(log))3004		goto out_error;3005 3006	iclog = log->l_iclog;3007	while (be64_to_cpu(iclog->ic_header.h_lsn) != lsn) {3008		trace_xlog_iclog_force_lsn(iclog, _RET_IP_);3009		iclog = iclog->ic_next;3010		if (iclog == log->l_iclog)3011			goto out_unlock;3012	}3013 3014	switch (iclog->ic_state) {3015	case XLOG_STATE_ACTIVE:3016		/*3017		 * We sleep here if we haven't already slept (e.g. this is the3018		 * first time we've looked at the correct iclog buf) and the3019		 * buffer before us is going to be sync'ed.  The reason for this3020		 * is that if we are doing sync transactions here, by waiting3021		 * for the previous I/O to complete, we can allow a few more3022		 * transactions into this iclog before we close it down.3023		 *3024		 * Otherwise, we mark the buffer WANT_SYNC, and bump up the3025		 * refcnt so we can release the log (which drops the ref count).3026		 * The state switch keeps new transaction commits from using3027		 * this buffer.  When the current commits finish writing into3028		 * the buffer, the refcount will drop to zero and the buffer3029		 * will go out then.3030		 */3031		if (!already_slept &&3032		    (iclog->ic_prev->ic_state == XLOG_STATE_WANT_SYNC ||3033		     iclog->ic_prev->ic_state == XLOG_STATE_SYNCING)) {3034			xlog_wait(&iclog->ic_prev->ic_write_wait,3035					&log->l_icloglock);3036			return -EAGAIN;3037		}3038		if (xlog_force_and_check_iclog(iclog, &completed))3039			goto out_error;3040		if (log_flushed)3041			*log_flushed = 1;3042		if (completed)3043			goto out_unlock;3044		break;3045	case XLOG_STATE_WANT_SYNC:3046		/*3047		 * This iclog may contain the checkpoint pushed by the3048		 * xlog_cil_force_seq() call, but there are other writers still3049		 * accessing it so it hasn't been pushed to disk yet. Like the3050		 * ACTIVE case above, we need to make sure caches are flushed3051		 * when this iclog is written.3052		 */3053		iclog->ic_flags |= XLOG_ICL_NEED_FLUSH | XLOG_ICL_NEED_FUA;3054		break;3055	default:3056		/*3057		 * The entire checkpoint was written by the CIL force and is on3058		 * its way to disk already. It will be stable when it3059		 * completes, so we don't need to manipulate caches here at all.3060		 * We just need to wait for completion if necessary.3061		 */3062		break;3063	}3064 3065	if (flags & XFS_LOG_SYNC)3066		return xlog_wait_on_iclog(iclog);3067out_unlock:3068	spin_unlock(&log->l_icloglock);3069	return 0;3070out_error:3071	spin_unlock(&log->l_icloglock);3072	return -EIO;3073}3074 3075/*3076 * Force the log to a specific checkpoint sequence.3077 *3078 * First force the CIL so that all the required changes have been flushed to the3079 * iclogs. If the CIL force completed it will return a commit LSN that indicates3080 * the iclog that needs to be flushed to stable storage. If the caller needs3081 * a synchronous log force, we will wait on the iclog with the LSN returned by3082 * xlog_cil_force_seq() to be completed.3083 */3084int3085xfs_log_force_seq(3086	struct xfs_mount	*mp,3087	xfs_csn_t		seq,3088	uint			flags,3089	int			*log_flushed)3090{3091	struct xlog		*log = mp->m_log;3092	xfs_lsn_t		lsn;3093	int			ret;3094	ASSERT(seq != 0);3095 3096	XFS_STATS_INC(mp, xs_log_force);3097	trace_xfs_log_force(mp, seq, _RET_IP_);3098 3099	lsn = xlog_cil_force_seq(log, seq);3100	if (lsn == NULLCOMMITLSN)3101		return 0;3102 3103	ret = xlog_force_lsn(log, lsn, flags, log_flushed, false);3104	if (ret == -EAGAIN) {3105		XFS_STATS_INC(mp, xs_log_force_sleep);3106		ret = xlog_force_lsn(log, lsn, flags, log_flushed, true);3107	}3108	return ret;3109}3110 3111/*3112 * Free a used ticket when its refcount falls to zero.3113 */3114void3115xfs_log_ticket_put(3116	xlog_ticket_t	*ticket)3117{3118	ASSERT(atomic_read(&ticket->t_ref) > 0);3119	if (atomic_dec_and_test(&ticket->t_ref))3120		kmem_cache_free(xfs_log_ticket_cache, ticket);3121}3122 3123xlog_ticket_t *3124xfs_log_ticket_get(3125	xlog_ticket_t	*ticket)3126{3127	ASSERT(atomic_read(&ticket->t_ref) > 0);3128	atomic_inc(&ticket->t_ref);3129	return ticket;3130}3131 3132/*3133 * Figure out the total log space unit (in bytes) that would be3134 * required for a log ticket.3135 */3136static int3137xlog_calc_unit_res(3138	struct xlog		*log,3139	int			unit_bytes,3140	int			*niclogs)3141{3142	int			iclog_space;3143	uint			num_headers;3144 3145	/*3146	 * Permanent reservations have up to 'cnt'-1 active log operations3147	 * in the log.  A unit in this case is the amount of space for one3148	 * of these log operations.  Normal reservations have a cnt of 13149	 * and their unit amount is the total amount of space required.3150	 *3151	 * The following lines of code account for non-transaction data3152	 * which occupy space in the on-disk log.3153	 *3154	 * Normal form of a transaction is:3155	 * <oph><trans-hdr><start-oph><reg1-oph><reg1><reg2-oph>...<commit-oph>3156	 * and then there are LR hdrs, split-recs and roundoff at end of syncs.3157	 *3158	 * We need to account for all the leadup data and trailer data3159	 * around the transaction data.3160	 * And then we need to account for the worst case in terms of using3161	 * more space.3162	 * The worst case will happen if:3163	 * - the placement of the transaction happens to be such that the3164	 *   roundoff is at its maximum3165	 * - the transaction data is synced before the commit record is synced3166	 *   i.e. <transaction-data><roundoff> | <commit-rec><roundoff>3167	 *   Therefore the commit record is in its own Log Record.3168	 *   This can happen as the commit record is called with its3169	 *   own region to xlog_write().3170	 *   This then means that in the worst case, roundoff can happen for3171	 *   the commit-rec as well.3172	 *   The commit-rec is smaller than padding in this scenario and so it is3173	 *   not added separately.3174	 */3175 3176	/* for trans header */3177	unit_bytes += sizeof(xlog_op_header_t);3178	unit_bytes += sizeof(xfs_trans_header_t);3179 3180	/* for start-rec */3181	unit_bytes += sizeof(xlog_op_header_t);3182 3183	/*3184	 * for LR headers - the space for data in an iclog is the size minus3185	 * the space used for the headers. If we use the iclog size, then we3186	 * undercalculate the number of headers required.3187	 *3188	 * Furthermore - the addition of op headers for split-recs might3189	 * increase the space required enough to require more log and op3190	 * headers, so take that into account too.3191	 *3192	 * IMPORTANT: This reservation makes the assumption that if this3193	 * transaction is the first in an iclog and hence has the LR headers3194	 * accounted to it, then the remaining space in the iclog is3195	 * exclusively for this transaction.  i.e. if the transaction is larger3196	 * than the iclog, it will be the only thing in that iclog.3197	 * Fundamentally, this means we must pass the entire log vector to3198	 * xlog_write to guarantee this.3199	 */3200	iclog_space = log->l_iclog_size - log->l_iclog_hsize;3201	num_headers = howmany(unit_bytes, iclog_space);3202 3203	/* for split-recs - ophdrs added when data split over LRs */3204	unit_bytes += sizeof(xlog_op_header_t) * num_headers;3205 3206	/* add extra header reservations if we overrun */3207	while (!num_headers ||3208	       howmany(unit_bytes, iclog_space) > num_headers) {3209		unit_bytes += sizeof(xlog_op_header_t);3210		num_headers++;3211	}3212	unit_bytes += log->l_iclog_hsize * num_headers;3213 3214	/* for commit-rec LR header - note: padding will subsume the ophdr */3215	unit_bytes += log->l_iclog_hsize;3216 3217	/* roundoff padding for transaction data and one for commit record */3218	unit_bytes += 2 * log->l_iclog_roundoff;3219 3220	if (niclogs)3221		*niclogs = num_headers;3222	return unit_bytes;3223}3224 3225int3226xfs_log_calc_unit_res(3227	struct xfs_mount	*mp,3228	int			unit_bytes)3229{3230	return xlog_calc_unit_res(mp->m_log, unit_bytes, NULL);3231}3232 3233/*3234 * Allocate and initialise a new log ticket.3235 */3236struct xlog_ticket *3237xlog_ticket_alloc(3238	struct xlog		*log,3239	int			unit_bytes,3240	int			cnt,3241	bool			permanent)3242{3243	struct xlog_ticket	*tic;3244	int			unit_res;3245 3246	tic = kmem_cache_zalloc(xfs_log_ticket_cache,3247			GFP_KERNEL | __GFP_NOFAIL);3248 3249	unit_res = xlog_calc_unit_res(log, unit_bytes, &tic->t_iclog_hdrs);3250 3251	atomic_set(&tic->t_ref, 1);3252	tic->t_task		= current;3253	INIT_LIST_HEAD(&tic->t_queue);3254	tic->t_unit_res		= unit_res;3255	tic->t_curr_res		= unit_res;3256	tic->t_cnt		= cnt;3257	tic->t_ocnt		= cnt;3258	tic->t_tid		= get_random_u32();3259	if (permanent)3260		tic->t_flags |= XLOG_TIC_PERM_RESERV;3261 3262	return tic;3263}3264 3265#if defined(DEBUG)3266static void3267xlog_verify_dump_tail(3268	struct xlog		*log,3269	struct xlog_in_core	*iclog)3270{3271	xfs_alert(log->l_mp,3272"ran out of log space tail 0x%llx/0x%llx, head lsn 0x%llx, head 0x%x/0x%x, prev head 0x%x/0x%x",3273			iclog ? be64_to_cpu(iclog->ic_header.h_tail_lsn) : -1,3274			atomic64_read(&log->l_tail_lsn),3275			log->l_ailp->ail_head_lsn,3276			log->l_curr_cycle, log->l_curr_block,3277			log->l_prev_cycle, log->l_prev_block);3278	xfs_alert(log->l_mp,3279"write grant 0x%llx, reserve grant 0x%llx, tail_space 0x%llx, size 0x%x, iclog flags 0x%x",3280			atomic64_read(&log->l_write_head.grant),3281			atomic64_read(&log->l_reserve_head.grant),3282			log->l_tail_space, log->l_logsize,3283			iclog ? iclog->ic_flags : -1);3284}3285 3286/* Check if the new iclog will fit in the log. */3287STATIC void3288xlog_verify_tail_lsn(3289	struct xlog		*log,3290	struct xlog_in_core	*iclog)3291{3292	xfs_lsn_t	tail_lsn = be64_to_cpu(iclog->ic_header.h_tail_lsn);3293	int		blocks;3294 3295	if (CYCLE_LSN(tail_lsn) == log->l_prev_cycle) {3296		blocks = log->l_logBBsize -3297				(log->l_prev_block - BLOCK_LSN(tail_lsn));3298		if (blocks < BTOBB(iclog->ic_offset) +3299					BTOBB(log->l_iclog_hsize)) {3300			xfs_emerg(log->l_mp,3301					"%s: ran out of log space", __func__);3302			xlog_verify_dump_tail(log, iclog);3303		}3304		return;3305	}3306 3307	if (CYCLE_LSN(tail_lsn) + 1 != log->l_prev_cycle) {3308		xfs_emerg(log->l_mp, "%s: head has wrapped tail.", __func__);3309		xlog_verify_dump_tail(log, iclog);3310		return;3311	}3312	if (BLOCK_LSN(tail_lsn) == log->l_prev_block) {3313		xfs_emerg(log->l_mp, "%s: tail wrapped", __func__);3314		xlog_verify_dump_tail(log, iclog);3315		return;3316	}3317 3318	blocks = BLOCK_LSN(tail_lsn) - log->l_prev_block;3319	if (blocks < BTOBB(iclog->ic_offset) + 1) {3320		xfs_emerg(log->l_mp, "%s: ran out of iclog space", __func__);3321		xlog_verify_dump_tail(log, iclog);3322	}3323}3324 3325/*3326 * Perform a number of checks on the iclog before writing to disk.3327 *3328 * 1. Make sure the iclogs are still circular3329 * 2. Make sure we have a good magic number3330 * 3. Make sure we don't have magic numbers in the data3331 * 4. Check fields of each log operation header for:3332 *	A. Valid client identifier3333 *	B. tid ptr value falls in valid ptr space (user space code)3334 *	C. Length in log record header is correct according to the3335 *		individual operation headers within record.3336 * 5. When a bwrite will occur within 5 blocks of the front of the physical3337 *	log, check the preceding blocks of the physical log to make sure all3338 *	the cycle numbers agree with the current cycle number.3339 */3340STATIC void3341xlog_verify_iclog(3342	struct xlog		*log,3343	struct xlog_in_core	*iclog,3344	int			count)3345{3346	xlog_op_header_t	*ophead;3347	xlog_in_core_t		*icptr;3348	xlog_in_core_2_t	*xhdr;3349	void			*base_ptr, *ptr, *p;3350	ptrdiff_t		field_offset;3351	uint8_t			clientid;3352	int			len, i, j, k, op_len;3353	int			idx;3354 3355	/* check validity of iclog pointers */3356	spin_lock(&log->l_icloglock);3357	icptr = log->l_iclog;3358	for (i = 0; i < log->l_iclog_bufs; i++, icptr = icptr->ic_next)3359		ASSERT(icptr);3360 3361	if (icptr != log->l_iclog)3362		xfs_emerg(log->l_mp, "%s: corrupt iclog ring", __func__);3363	spin_unlock(&log->l_icloglock);3364 3365	/* check log magic numbers */3366	if (iclog->ic_header.h_magicno != cpu_to_be32(XLOG_HEADER_MAGIC_NUM))3367		xfs_emerg(log->l_mp, "%s: invalid magic num", __func__);3368 3369	base_ptr = ptr = &iclog->ic_header;3370	p = &iclog->ic_header;3371	for (ptr += BBSIZE; ptr < base_ptr + count; ptr += BBSIZE) {3372		if (*(__be32 *)ptr == cpu_to_be32(XLOG_HEADER_MAGIC_NUM))3373			xfs_emerg(log->l_mp, "%s: unexpected magic num",3374				__func__);3375	}3376 3377	/* check fields */3378	len = be32_to_cpu(iclog->ic_header.h_num_logops);3379	base_ptr = ptr = iclog->ic_datap;3380	ophead = ptr;3381	xhdr = iclog->ic_data;3382	for (i = 0; i < len; i++) {3383		ophead = ptr;3384 3385		/* clientid is only 1 byte */3386		p = &ophead->oh_clientid;3387		field_offset = p - base_ptr;3388		if (field_offset & 0x1ff) {3389			clientid = ophead->oh_clientid;3390		} else {3391			idx = BTOBBT((void *)&ophead->oh_clientid - iclog->ic_datap);3392			if (idx >= (XLOG_HEADER_CYCLE_SIZE / BBSIZE)) {3393				j = idx / (XLOG_HEADER_CYCLE_SIZE / BBSIZE);3394				k = idx % (XLOG_HEADER_CYCLE_SIZE / BBSIZE);3395				clientid = xlog_get_client_id(3396					xhdr[j].hic_xheader.xh_cycle_data[k]);3397			} else {3398				clientid = xlog_get_client_id(3399					iclog->ic_header.h_cycle_data[idx]);3400			}3401		}3402		if (clientid != XFS_TRANSACTION && clientid != XFS_LOG) {3403			xfs_warn(log->l_mp,3404				"%s: op %d invalid clientid %d op "PTR_FMT" offset 0x%lx",3405				__func__, i, clientid, ophead,3406				(unsigned long)field_offset);3407		}3408 3409		/* check length */3410		p = &ophead->oh_len;3411		field_offset = p - base_ptr;3412		if (field_offset & 0x1ff) {3413			op_len = be32_to_cpu(ophead->oh_len);3414		} else {3415			idx = BTOBBT((void *)&ophead->oh_len - iclog->ic_datap);3416			if (idx >= (XLOG_HEADER_CYCLE_SIZE / BBSIZE)) {3417				j = idx / (XLOG_HEADER_CYCLE_SIZE / BBSIZE);3418				k = idx % (XLOG_HEADER_CYCLE_SIZE / BBSIZE);3419				op_len = be32_to_cpu(xhdr[j].hic_xheader.xh_cycle_data[k]);3420			} else {3421				op_len = be32_to_cpu(iclog->ic_header.h_cycle_data[idx]);3422			}3423		}3424		ptr += sizeof(xlog_op_header_t) + op_len;3425	}3426}3427#endif3428 3429/*3430 * Perform a forced shutdown on the log.3431 *3432 * This can be called from low level log code to trigger a shutdown, or from the3433 * high level mount shutdown code when the mount shuts down.3434 *3435 * Our main objectives here are to make sure that:3436 *	a. if the shutdown was not due to a log IO error, flush the logs to3437 *	   disk. Anything modified after this is ignored.3438 *	b. the log gets atomically marked 'XLOG_IO_ERROR' for all interested3439 *	   parties to find out. Nothing new gets queued after this is done.3440 *	c. Tasks sleeping on log reservations, pinned objects and3441 *	   other resources get woken up.3442 *	d. The mount is also marked as shut down so that log triggered shutdowns3443 *	   still behave the same as if they called xfs_forced_shutdown().3444 *3445 * Return true if the shutdown cause was a log IO error and we actually shut the3446 * log down.3447 */3448bool3449xlog_force_shutdown(3450	struct xlog	*log,3451	uint32_t	shutdown_flags)3452{3453	bool		log_error = (shutdown_flags & SHUTDOWN_LOG_IO_ERROR);3454 3455	if (!log)3456		return false;3457 3458	/*3459	 * Flush all the completed transactions to disk before marking the log3460	 * being shut down. We need to do this first as shutting down the log3461	 * before the force will prevent the log force from flushing the iclogs3462	 * to disk.3463	 *3464	 * When we are in recovery, there are no transactions to flush, and3465	 * we don't want to touch the log because we don't want to perturb the3466	 * current head/tail for future recovery attempts. Hence we need to3467	 * avoid a log force in this case.3468	 *3469	 * If we are shutting down due to a log IO error, then we must avoid3470	 * trying to write the log as that may just result in more IO errors and3471	 * an endless shutdown/force loop.3472	 */3473	if (!log_error && !xlog_in_recovery(log))3474		xfs_log_force(log->l_mp, XFS_LOG_SYNC);3475 3476	/*3477	 * Atomically set the shutdown state. If the shutdown state is already3478	 * set, there someone else is performing the shutdown and so we are done3479	 * here. This should never happen because we should only ever get called3480	 * once by the first shutdown caller.3481	 *3482	 * Much of the log state machine transitions assume that shutdown state3483	 * cannot change once they hold the log->l_icloglock. Hence we need to3484	 * hold that lock here, even though we use the atomic test_and_set_bit()3485	 * operation to set the shutdown state.3486	 */3487	spin_lock(&log->l_icloglock);3488	if (test_and_set_bit(XLOG_IO_ERROR, &log->l_opstate)) {3489		spin_unlock(&log->l_icloglock);3490		return false;3491	}3492	spin_unlock(&log->l_icloglock);3493 3494	/*3495	 * If this log shutdown also sets the mount shutdown state, issue a3496	 * shutdown warning message.3497	 */3498	if (!xfs_set_shutdown(log->l_mp)) {3499		xfs_alert_tag(log->l_mp, XFS_PTAG_SHUTDOWN_LOGERROR,3500"Filesystem has been shut down due to log error (0x%x).",3501				shutdown_flags);3502		xfs_alert(log->l_mp,3503"Please unmount the filesystem and rectify the problem(s).");3504		if (xfs_error_level >= XFS_ERRLEVEL_HIGH)3505			xfs_stack_trace();3506	}3507 3508	/*3509	 * We don't want anybody waiting for log reservations after this. That3510	 * means we have to wake up everybody queued up on reserveq as well as3511	 * writeq.  In addition, we make sure in xlog_{re}grant_log_space that3512	 * we don't enqueue anything once the SHUTDOWN flag is set, and this3513	 * action is protected by the grant locks.3514	 */3515	xlog_grant_head_wake_all(&log->l_reserve_head);3516	xlog_grant_head_wake_all(&log->l_write_head);3517 3518	/*3519	 * Wake up everybody waiting on xfs_log_force. Wake the CIL push first3520	 * as if the log writes were completed. The abort handling in the log3521	 * item committed callback functions will do this again under lock to3522	 * avoid races.3523	 */3524	spin_lock(&log->l_cilp->xc_push_lock);3525	wake_up_all(&log->l_cilp->xc_start_wait);3526	wake_up_all(&log->l_cilp->xc_commit_wait);3527	spin_unlock(&log->l_cilp->xc_push_lock);3528 3529	spin_lock(&log->l_icloglock);3530	xlog_state_shutdown_callbacks(log);3531	spin_unlock(&log->l_icloglock);3532 3533	wake_up_var(&log->l_opstate);3534	return log_error;3535}3536 3537STATIC int3538xlog_iclogs_empty(3539	struct xlog	*log)3540{3541	xlog_in_core_t	*iclog;3542 3543	iclog = log->l_iclog;3544	do {3545		/* endianness does not matter here, zero is zero in3546		 * any language.3547		 */3548		if (iclog->ic_header.h_num_logops)3549			return 0;3550		iclog = iclog->ic_next;3551	} while (iclog != log->l_iclog);3552	return 1;3553}3554 3555/*3556 * Verify that an LSN stamped into a piece of metadata is valid. This is3557 * intended for use in read verifiers on v5 superblocks.3558 */3559bool3560xfs_log_check_lsn(3561	struct xfs_mount	*mp,3562	xfs_lsn_t		lsn)3563{3564	struct xlog		*log = mp->m_log;3565	bool			valid;3566 3567	/*3568	 * norecovery mode skips mount-time log processing and unconditionally3569	 * resets the in-core LSN. We can't validate in this mode, but3570	 * modifications are not allowed anyways so just return true.3571	 */3572	if (xfs_has_norecovery(mp))3573		return true;3574 3575	/*3576	 * Some metadata LSNs are initialized to NULL (e.g., the agfl). This is3577	 * handled by recovery and thus safe to ignore here.3578	 */3579	if (lsn == NULLCOMMITLSN)3580		return true;3581 3582	valid = xlog_valid_lsn(mp->m_log, lsn);3583 3584	/* warn the user about what's gone wrong before verifier failure */3585	if (!valid) {3586		spin_lock(&log->l_icloglock);3587		xfs_warn(mp,3588"Corruption warning: Metadata has LSN (%d:%d) ahead of current LSN (%d:%d). "3589"Please unmount and run xfs_repair (>= v4.3) to resolve.",3590			 CYCLE_LSN(lsn), BLOCK_LSN(lsn),3591			 log->l_curr_cycle, log->l_curr_block);3592		spin_unlock(&log->l_icloglock);3593	}3594 3595	return valid;3596}3597