xfs_log.c 105.0 KB
Newer Older
L
Linus Torvalds 已提交
1
/*
2 3
 * Copyright (c) 2000-2005 Silicon Graphics, Inc.
 * All Rights Reserved.
L
Linus Torvalds 已提交
4
 *
5 6
 * This program is free software; you can redistribute it and/or
 * modify it under the terms of the GNU General Public License as
L
Linus Torvalds 已提交
7 8
 * published by the Free Software Foundation.
 *
9 10 11 12
 * This program is distributed in the hope that it would be useful,
 * but WITHOUT ANY WARRANTY; without even the implied warranty of
 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
 * GNU General Public License for more details.
L
Linus Torvalds 已提交
13
 *
14 15 16
 * You should have received a copy of the GNU General Public License
 * along with this program; if not, write the Free Software Foundation,
 * Inc.,  51 Franklin St, Fifth Floor, Boston, MA  02110-1301  USA
L
Linus Torvalds 已提交
17 18
 */
#include "xfs.h"
19
#include "xfs_fs.h"
L
Linus Torvalds 已提交
20
#include "xfs_types.h"
21
#include "xfs_bit.h"
L
Linus Torvalds 已提交
22
#include "xfs_log.h"
23
#include "xfs_inum.h"
L
Linus Torvalds 已提交
24
#include "xfs_trans.h"
25 26 27
#include "xfs_sb.h"
#include "xfs_ag.h"
#include "xfs_dir2.h"
L
Linus Torvalds 已提交
28 29 30 31 32
#include "xfs_dmapi.h"
#include "xfs_mount.h"
#include "xfs_error.h"
#include "xfs_log_priv.h"
#include "xfs_buf_item.h"
33
#include "xfs_bmap_btree.h"
L
Linus Torvalds 已提交
34
#include "xfs_alloc_btree.h"
35
#include "xfs_ialloc_btree.h"
L
Linus Torvalds 已提交
36 37
#include "xfs_log_recover.h"
#include "xfs_trans_priv.h"
38 39 40 41 42
#include "xfs_dir2_sf.h"
#include "xfs_attr_sf.h"
#include "xfs_dinode.h"
#include "xfs_inode.h"
#include "xfs_rw.h"
L
Linus Torvalds 已提交
43

44
kmem_zone_t	*xfs_log_ticket_zone;
L
Linus Torvalds 已提交
45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60

#define xlog_write_adv_cnt(ptr, len, off, bytes) \
	{ (ptr) += (bytes); \
	  (len) -= (bytes); \
	  (off) += (bytes);}

/* Local miscellaneous function prototypes */
STATIC int	 xlog_bdstrat_cb(struct xfs_buf *);
STATIC int	 xlog_commit_record(xfs_mount_t *mp, xlog_ticket_t *ticket,
				    xlog_in_core_t **, xfs_lsn_t *);
STATIC xlog_t *  xlog_alloc_log(xfs_mount_t	*mp,
				xfs_buftarg_t	*log_target,
				xfs_daddr_t	blk_offset,
				int		num_bblks);
STATIC int	 xlog_space_left(xlog_t *log, int cycle, int bytes);
STATIC int	 xlog_sync(xlog_t *log, xlog_in_core_t *iclog);
61
STATIC void	 xlog_dealloc_log(xlog_t *log);
L
Linus Torvalds 已提交
62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81
STATIC int	 xlog_write(xfs_mount_t *mp, xfs_log_iovec_t region[],
			    int nentries, xfs_log_ticket_t tic,
			    xfs_lsn_t *start_lsn,
			    xlog_in_core_t **commit_iclog,
			    uint flags);

/* local state machine functions */
STATIC void xlog_state_done_syncing(xlog_in_core_t *iclog, int);
STATIC void xlog_state_do_callback(xlog_t *log,int aborted, xlog_in_core_t *iclog);
STATIC int  xlog_state_get_iclog_space(xlog_t		*log,
				       int		len,
				       xlog_in_core_t	**iclog,
				       xlog_ticket_t	*ticket,
				       int		*continued_write,
				       int		*logoffsetp);
STATIC int  xlog_state_release_iclog(xlog_t		*log,
				     xlog_in_core_t	*iclog);
STATIC void xlog_state_switch_iclogs(xlog_t		*log,
				     xlog_in_core_t *iclog,
				     int		eventual_size);
82 83 84 85 86
STATIC int  xlog_state_sync(xlog_t			*log,
			    xfs_lsn_t 			lsn,
			    uint			flags,
			    int				*log_flushed);
STATIC int  xlog_state_sync_all(xlog_t *log, uint flags, int *log_flushed);
L
Linus Torvalds 已提交
87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109
STATIC void xlog_state_want_sync(xlog_t	*log, xlog_in_core_t *iclog);

/* local functions to manipulate grant head */
STATIC int  xlog_grant_log_space(xlog_t		*log,
				 xlog_ticket_t	*xtic);
STATIC void xlog_grant_push_ail(xfs_mount_t	*mp,
				int		need_bytes);
STATIC void xlog_regrant_reserve_log_space(xlog_t	 *log,
					   xlog_ticket_t *ticket);
STATIC int xlog_regrant_write_log_space(xlog_t		*log,
					 xlog_ticket_t  *ticket);
STATIC void xlog_ungrant_log_space(xlog_t	 *log,
				   xlog_ticket_t *ticket);


/* local ticket functions */
STATIC xlog_ticket_t	*xlog_ticket_get(xlog_t *log,
					 int	unit_bytes,
					 int	count,
					 char	clientid,
					 uint	flags);
STATIC void		xlog_ticket_put(xlog_t *log, xlog_ticket_t *ticket);

110
#if defined(DEBUG)
L
Linus Torvalds 已提交
111 112 113 114 115 116 117 118 119 120 121 122 123
STATIC void	xlog_verify_dest_ptr(xlog_t *log, __psint_t ptr);
STATIC void	xlog_verify_grant_head(xlog_t *log, int equals);
STATIC void	xlog_verify_iclog(xlog_t *log, xlog_in_core_t *iclog,
				  int count, boolean_t syncing);
STATIC void	xlog_verify_tail_lsn(xlog_t *log, xlog_in_core_t *iclog,
				     xfs_lsn_t tail_lsn);
#else
#define xlog_verify_dest_ptr(a,b)
#define xlog_verify_grant_head(a,b)
#define xlog_verify_iclog(a,b,c,d)
#define xlog_verify_tail_lsn(a,b,c)
#endif

124
STATIC int	xlog_iclogs_empty(xlog_t *log);
L
Linus Torvalds 已提交
125 126 127 128 129

#if defined(XFS_LOG_TRACE)
void
xlog_trace_loggrant(xlog_t *log, xlog_ticket_t *tic, xfs_caddr_t string)
{
130 131 132 133 134
	unsigned long cnts;

	if (!log->l_grant_trace) {
		log->l_grant_trace = ktrace_alloc(2048, KM_NOSLEEP);
		if (!log->l_grant_trace)
L
Linus Torvalds 已提交
135 136
			return;
	}
137 138
	/* ticket counts are 1 byte each */
	cnts = ((unsigned long)tic->t_ocnt) | ((unsigned long)tic->t_cnt) << 8;
L
Linus Torvalds 已提交
139 140 141 142 143 144 145 146 147 148 149 150 151 152

	ktrace_enter(log->l_grant_trace,
		     (void *)tic,
		     (void *)log->l_reserve_headq,
		     (void *)log->l_write_headq,
		     (void *)((unsigned long)log->l_grant_reserve_cycle),
		     (void *)((unsigned long)log->l_grant_reserve_bytes),
		     (void *)((unsigned long)log->l_grant_write_cycle),
		     (void *)((unsigned long)log->l_grant_write_bytes),
		     (void *)((unsigned long)log->l_curr_cycle),
		     (void *)((unsigned long)log->l_curr_block),
		     (void *)((unsigned long)CYCLE_LSN(log->l_tail_lsn)),
		     (void *)((unsigned long)BLOCK_LSN(log->l_tail_lsn)),
		     (void *)string,
153 154 155 156
		     (void *)((unsigned long)tic->t_trans_type),
		     (void *)cnts,
		     (void *)((unsigned long)tic->t_curr_res),
		     (void *)((unsigned long)tic->t_unit_res));
L
Linus Torvalds 已提交
157 158 159 160 161 162 163 164 165
}

void
xlog_trace_iclog(xlog_in_core_t *iclog, uint state)
{
	if (!iclog->ic_trace)
		iclog->ic_trace = ktrace_alloc(256, KM_SLEEP);
	ktrace_enter(iclog->ic_trace,
		     (void *)((unsigned long)state),
166 167 168 169 170
		     (void *)((unsigned long)current_pid()),
		     (void *)NULL, (void *)NULL, (void *)NULL, (void *)NULL,
		     (void *)NULL, (void *)NULL, (void *)NULL, (void *)NULL,
		     (void *)NULL, (void *)NULL, (void *)NULL, (void *)NULL,
		     (void *)NULL, (void *)NULL);
L
Linus Torvalds 已提交
171 172 173 174 175 176
}
#else
#define	xlog_trace_loggrant(log,tic,string)
#define	xlog_trace_iclog(iclog,state)
#endif /* XFS_LOG_TRACE */

177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252

static void
xlog_ins_ticketq(struct xlog_ticket **qp, struct xlog_ticket *tic)
{
	if (*qp) {
		tic->t_next	    = (*qp);
		tic->t_prev	    = (*qp)->t_prev;
		(*qp)->t_prev->t_next = tic;
		(*qp)->t_prev	    = tic;
	} else {
		tic->t_prev = tic->t_next = tic;
		*qp = tic;
	}

	tic->t_flags |= XLOG_TIC_IN_Q;
}

static void
xlog_del_ticketq(struct xlog_ticket **qp, struct xlog_ticket *tic)
{
	if (tic == tic->t_next) {
		*qp = NULL;
	} else {
		*qp = tic->t_next;
		tic->t_next->t_prev = tic->t_prev;
		tic->t_prev->t_next = tic->t_next;
	}

	tic->t_next = tic->t_prev = NULL;
	tic->t_flags &= ~XLOG_TIC_IN_Q;
}

static void
xlog_grant_sub_space(struct log *log, int bytes)
{
	log->l_grant_write_bytes -= bytes;
	if (log->l_grant_write_bytes < 0) {
		log->l_grant_write_bytes += log->l_logsize;
		log->l_grant_write_cycle--;
	}

	log->l_grant_reserve_bytes -= bytes;
	if ((log)->l_grant_reserve_bytes < 0) {
		log->l_grant_reserve_bytes += log->l_logsize;
		log->l_grant_reserve_cycle--;
	}

}

static void
xlog_grant_add_space_write(struct log *log, int bytes)
{
	log->l_grant_write_bytes += bytes;
	if (log->l_grant_write_bytes > log->l_logsize) {
		log->l_grant_write_bytes -= log->l_logsize;
		log->l_grant_write_cycle++;
	}
}

static void
xlog_grant_add_space_reserve(struct log *log, int bytes)
{
	log->l_grant_reserve_bytes += bytes;
	if (log->l_grant_reserve_bytes > log->l_logsize) {
		log->l_grant_reserve_bytes -= log->l_logsize;
		log->l_grant_reserve_cycle++;
	}
}

static inline void
xlog_grant_add_space(struct log *log, int bytes)
{
	xlog_grant_add_space_write(log, bytes);
	xlog_grant_add_space_reserve(log, bytes);
}

253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275
static void
xlog_tic_reset_res(xlog_ticket_t *tic)
{
	tic->t_res_num = 0;
	tic->t_res_arr_sum = 0;
	tic->t_res_num_ophdrs = 0;
}

static void
xlog_tic_add_region(xlog_ticket_t *tic, uint len, uint type)
{
	if (tic->t_res_num == XLOG_TIC_LEN_MAX) {
		/* add to overflow and start again */
		tic->t_res_o_flow += tic->t_res_arr_sum;
		tic->t_res_num = 0;
		tic->t_res_arr_sum = 0;
	}

	tic->t_res_arr[tic->t_res_num].r_len = len;
	tic->t_res_arr[tic->t_res_num].r_type = type;
	tic->t_res_arr_sum += len;
	tic->t_res_num++;
}
276

L
Linus Torvalds 已提交
277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325
/*
 * NOTES:
 *
 *	1. currblock field gets updated at startup and after in-core logs
 *		marked as with WANT_SYNC.
 */

/*
 * This routine is called when a user of a log manager ticket is done with
 * the reservation.  If the ticket was ever used, then a commit record for
 * the associated transaction is written out as a log operation header with
 * no data.  The flag XLOG_TIC_INITED is set when the first write occurs with
 * a given ticket.  If the ticket was one with a permanent reservation, then
 * a few operations are done differently.  Permanent reservation tickets by
 * default don't release the reservation.  They just commit the current
 * transaction with the belief that the reservation is still needed.  A flag
 * must be passed in before permanent reservations are actually released.
 * When these type of tickets are not released, they need to be set into
 * the inited state again.  By doing this, a start record will be written
 * out when the next write occurs.
 */
xfs_lsn_t
xfs_log_done(xfs_mount_t	*mp,
	     xfs_log_ticket_t	xtic,
	     void		**iclog,
	     uint		flags)
{
	xlog_t		*log    = mp->m_log;
	xlog_ticket_t	*ticket = (xfs_log_ticket_t) xtic;
	xfs_lsn_t	lsn	= 0;

	if (XLOG_FORCED_SHUTDOWN(log) ||
	    /*
	     * If nothing was ever written, don't write out commit record.
	     * If we get an error, just continue and give back the log ticket.
	     */
	    (((ticket->t_flags & XLOG_TIC_INITED) == 0) &&
	     (xlog_commit_record(mp, ticket,
				 (xlog_in_core_t **)iclog, &lsn)))) {
		lsn = (xfs_lsn_t) -1;
		if (ticket->t_flags & XLOG_TIC_PERM_RESERV) {
			flags |= XFS_LOG_REL_PERM_RESERV;
		}
	}


	if ((ticket->t_flags & XLOG_TIC_PERM_RESERV) == 0 ||
	    (flags & XFS_LOG_REL_PERM_RESERV)) {
		/*
326
		 * Release ticket if not permanent reservation or a specific
L
Linus Torvalds 已提交
327 328
		 * request has been made to release a permanent reservation.
		 */
329
		xlog_trace_loggrant(log, ticket, "xfs_log_done: (non-permanent)");
L
Linus Torvalds 已提交
330
		xlog_ungrant_log_space(log, ticket);
331
		xlog_ticket_put(log, ticket);
L
Linus Torvalds 已提交
332
	} else {
333
		xlog_trace_loggrant(log, ticket, "xfs_log_done: (permanent)");
L
Linus Torvalds 已提交
334 335 336 337 338 339 340 341 342 343 344 345 346 347 348 349 350 351 352 353 354 355 356 357 358 359 360 361 362
		xlog_regrant_reserve_log_space(log, ticket);
	}

	/* If this ticket was a permanent reservation and we aren't
	 * trying to release it, reset the inited flags; so next time
	 * we write, a start record will be written out.
	 */
	if ((ticket->t_flags & XLOG_TIC_PERM_RESERV) &&
	    (flags & XFS_LOG_REL_PERM_RESERV) == 0)
		ticket->t_flags |= XLOG_TIC_INITED;

	return lsn;
}	/* xfs_log_done */


/*
 * Force the in-core log to disk.  If flags == XFS_LOG_SYNC,
 *	the force is done synchronously.
 *
 * Asynchronous forces are implemented by setting the WANT_SYNC
 * bit in the appropriate in-core log and then returning.
 *
 * Synchronous forces are implemented with a semaphore.  All callers
 * to force a given lsn to disk will wait on a semaphore attached to the
 * specific in-core log.  When given in-core log finally completes its
 * write to disk, that thread will wake up all threads waiting on the
 * semaphore.
 */
int
363 364 365 366 367
_xfs_log_force(
	xfs_mount_t	*mp,
	xfs_lsn_t	lsn,
	uint		flags,
	int		*log_flushed)
L
Linus Torvalds 已提交
368
{
369 370 371 372 373
	xlog_t		*log = mp->m_log;
	int		dummy;

	if (!log_flushed)
		log_flushed = &dummy;
L
Linus Torvalds 已提交
374 375 376 377 378

	ASSERT(flags & XFS_LOG_FORCE);

	XFS_STATS_INC(xs_log_force);

379 380 381 382 383 384
	if (log->l_flags & XLOG_IO_ERROR)
		return XFS_ERROR(EIO);
	if (lsn == 0)
		return xlog_state_sync_all(log, flags, log_flushed);
	else
		return xlog_state_sync(log, lsn, flags, log_flushed);
385 386 387 388 389 390 391 392 393 394 395 396 397 398 399 400 401 402 403 404 405
}	/* _xfs_log_force */

/*
 * Wrapper for _xfs_log_force(), to be used when caller doesn't care
 * about errors or whether the log was flushed or not. This is the normal
 * interface to use when trying to unpin items or move the log forward.
 */
void
xfs_log_force(
	xfs_mount_t	*mp,
	xfs_lsn_t	lsn,
	uint		flags)
{
	int	error;
	error = _xfs_log_force(mp, lsn, flags, NULL);
	if (error) {
		xfs_fs_cmn_err(CE_WARN, mp, "xfs_log_force: "
			"error %d returned.", error);
	}
}

L
Linus Torvalds 已提交
406 407 408 409 410 411 412 413 414 415 416 417 418

/*
 * Attaches a new iclog I/O completion callback routine during
 * transaction commit.  If the log is in error state, a non-zero
 * return code is handed back and the caller is responsible for
 * executing the callback at an appropriate time.
 */
int
xfs_log_notify(xfs_mount_t	  *mp,		/* mount of partition */
	       void		  *iclog_hndl,	/* iclog to hang callback off */
	       xfs_log_callback_t *cb)
{
	xlog_in_core_t	  *iclog = (xlog_in_core_t *)iclog_hndl;
E
Eric Sandeen 已提交
419
	int	abortflg;
L
Linus Torvalds 已提交
420

421
	spin_lock(&iclog->ic_callback_lock);
L
Linus Torvalds 已提交
422 423 424 425 426 427 428 429
	abortflg = (iclog->ic_state & XLOG_STATE_IOERROR);
	if (!abortflg) {
		ASSERT_ALWAYS((iclog->ic_state == XLOG_STATE_ACTIVE) ||
			      (iclog->ic_state == XLOG_STATE_WANT_SYNC));
		cb->cb_next = NULL;
		*(iclog->ic_callback_tail) = cb;
		iclog->ic_callback_tail = &(cb->cb_next);
	}
430
	spin_unlock(&iclog->ic_callback_lock);
L
Linus Torvalds 已提交
431 432 433 434 435 436 437 438 439 440 441
	return abortflg;
}	/* xfs_log_notify */

int
xfs_log_release_iclog(xfs_mount_t *mp,
		      void	  *iclog_hndl)
{
	xlog_t *log = mp->m_log;
	xlog_in_core_t	  *iclog = (xlog_in_core_t *)iclog_hndl;

	if (xlog_state_release_iclog(log, iclog)) {
442
		xfs_force_shutdown(mp, SHUTDOWN_LOG_IO_ERROR);
443
		return EIO;
L
Linus Torvalds 已提交
444 445 446 447 448 449 450 451 452 453 454 455 456 457 458 459 460 461 462 463 464
	}

	return 0;
}

/*
 *  1. Reserve an amount of on-disk log space and return a ticket corresponding
 *	to the reservation.
 *  2. Potentially, push buffers at tail of log to disk.
 *
 * Each reservation is going to reserve extra space for a log record header.
 * When writes happen to the on-disk log, we don't subtract the length of the
 * log record header from any reservation.  By wasting space in each
 * reservation, we prevent over allocation problems.
 */
int
xfs_log_reserve(xfs_mount_t	 *mp,
		int		 unit_bytes,
		int		 cnt,
		xfs_log_ticket_t *ticket,
		__uint8_t	 client,
465 466
		uint		 flags,
		uint		 t_type)
L
Linus Torvalds 已提交
467 468 469
{
	xlog_t		*log = mp->m_log;
	xlog_ticket_t	*internal_ticket;
470
	int		retval = 0;
L
Linus Torvalds 已提交
471 472 473 474 475 476 477 478 479 480 481 482

	ASSERT(client == XFS_TRANSACTION || client == XFS_LOG);
	ASSERT((flags & XFS_LOG_NOSLEEP) == 0);

	if (XLOG_FORCED_SHUTDOWN(log))
		return XFS_ERROR(EIO);

	XFS_STATS_INC(xs_try_logspace);

	if (*ticket != NULL) {
		ASSERT(flags & XFS_LOG_PERM_RESERV);
		internal_ticket = (xlog_ticket_t *)*ticket;
483
		xlog_trace_loggrant(log, internal_ticket, "xfs_log_reserve: existing ticket (permanent trans)");
L
Linus Torvalds 已提交
484 485 486 487 488 489
		xlog_grant_push_ail(mp, internal_ticket->t_unit_res);
		retval = xlog_regrant_write_log_space(log, internal_ticket);
	} else {
		/* may sleep if need to allocate more tickets */
		internal_ticket = xlog_ticket_get(log, unit_bytes, cnt,
						  client, flags);
490 491
		if (!internal_ticket)
			return XFS_ERROR(ENOMEM);
492
		internal_ticket->t_trans_type = t_type;
L
Linus Torvalds 已提交
493
		*ticket = internal_ticket;
494 495 496 497
		xlog_trace_loggrant(log, internal_ticket, 
			(internal_ticket->t_flags & XLOG_TIC_PERM_RESERV) ?
			"xfs_log_reserve: create new ticket (permanent trans)" :
			"xfs_log_reserve: create new ticket");
L
Linus Torvalds 已提交
498 499 500 501 502 503 504 505 506 507 508 509 510 511 512 513 514 515 516 517 518
		xlog_grant_push_ail(mp,
				    (internal_ticket->t_unit_res *
				     internal_ticket->t_cnt));
		retval = xlog_grant_log_space(log, internal_ticket);
	}

	return retval;
}	/* xfs_log_reserve */


/*
 * Mount a log filesystem
 *
 * mp		- ubiquitous xfs mount point structure
 * log_target	- buftarg of on-disk log device
 * blk_offset	- Start block # where block size is 512 bytes (BBSIZE)
 * num_bblocks	- Number of BBSIZE blocks in on-disk log
 *
 * Return error or zero.
 */
int
519 520 521 522 523
xfs_log_mount(
	xfs_mount_t	*mp,
	xfs_buftarg_t	*log_target,
	xfs_daddr_t	blk_offset,
	int		num_bblks)
L
Linus Torvalds 已提交
524
{
525 526
	int		error;

L
Linus Torvalds 已提交
527 528 529 530 531 532
	if (!(mp->m_flags & XFS_MOUNT_NORECOVERY))
		cmn_err(CE_NOTE, "XFS mounting filesystem %s", mp->m_fsname);
	else {
		cmn_err(CE_NOTE,
			"!Mounting filesystem \"%s\" in no-recovery mode.  Filesystem will be inconsistent.",
			mp->m_fsname);
533
		ASSERT(mp->m_flags & XFS_MOUNT_RDONLY);
L
Linus Torvalds 已提交
534 535 536 537
	}

	mp->m_log = xlog_alloc_log(mp, log_target, blk_offset, num_bblks);

538 539 540 541 542 543 544 545 546 547
	/*
	 * Initialize the AIL now we have a log.
	 */
	spin_lock_init(&mp->m_ail_lock);
	error = xfs_trans_ail_init(mp);
	if (error) {
		cmn_err(CE_WARN, "XFS: AIL initialisation failed: error %d", error);
		goto error;
	}

L
Linus Torvalds 已提交
548 549 550 551 552
	/*
	 * skip log recovery on a norecovery mount.  pretend it all
	 * just worked.
	 */
	if (!(mp->m_flags & XFS_MOUNT_NORECOVERY)) {
553
		int	readonly = (mp->m_flags & XFS_MOUNT_RDONLY);
L
Linus Torvalds 已提交
554 555

		if (readonly)
556
			mp->m_flags &= ~XFS_MOUNT_RDONLY;
L
Linus Torvalds 已提交
557

558
		error = xlog_recover(mp->m_log);
L
Linus Torvalds 已提交
559 560

		if (readonly)
561
			mp->m_flags |= XFS_MOUNT_RDONLY;
L
Linus Torvalds 已提交
562 563
		if (error) {
			cmn_err(CE_WARN, "XFS: log mount/recovery failed: error %d", error);
564
			goto error;
L
Linus Torvalds 已提交
565 566 567 568 569 570 571 572
		}
	}

	/* Normal transactions can now occur */
	mp->m_log->l_flags &= ~XLOG_ACTIVE_RECOVERY;

	/* End mounting message in xfs_log_mount_finish */
	return 0;
573 574 575
error:
	xfs_log_unmount_dealloc(mp);
	return error;
L
Linus Torvalds 已提交
576 577 578 579 580 581 582 583 584 585 586 587 588 589 590 591 592 593 594
}	/* xfs_log_mount */

/*
 * Finish the recovery of the file system.  This is separate from
 * the xfs_log_mount() call, because it depends on the code in
 * xfs_mountfs() to read in the root and real-time bitmap inodes
 * between calling xfs_log_mount() and here.
 *
 * mp		- ubiquitous xfs mount point structure
 */
int
xfs_log_mount_finish(xfs_mount_t *mp, int mfsi_flags)
{
	int	error;

	if (!(mp->m_flags & XFS_MOUNT_NORECOVERY))
		error = xlog_recover_finish(mp->m_log, mfsi_flags);
	else {
		error = 0;
595
		ASSERT(mp->m_flags & XFS_MOUNT_RDONLY);
L
Linus Torvalds 已提交
596 597 598 599 600 601 602 603 604 605 606 607 608 609 610
	}

	return error;
}

/*
 * Unmount processing for the log.
 */
int
xfs_log_unmount(xfs_mount_t *mp)
{
	int		error;

	error = xfs_log_unmount_write(mp);
	xfs_log_unmount_dealloc(mp);
611
	return error;
L
Linus Torvalds 已提交
612 613 614 615 616 617 618 619 620 621 622 623 624 625 626 627 628 629 630 631 632 633 634 635 636 637 638 639 640 641 642 643 644 645 646 647 648 649 650 651 652 653
}

/*
 * Final log writes as part of unmount.
 *
 * Mark the filesystem clean as unmount happens.  Note that during relocation
 * this routine needs to be executed as part of source-bag while the
 * deallocation must not be done until source-end.
 */

/*
 * Unmount record used to have a string "Unmount filesystem--" in the
 * data section where the "Un" was really a magic number (XLOG_UNMOUNT_TYPE).
 * We just write the magic number now since that particular field isn't
 * currently architecture converted and "nUmount" is a bit foo.
 * As far as I know, there weren't any dependencies on the old behaviour.
 */

int
xfs_log_unmount_write(xfs_mount_t *mp)
{
	xlog_t		 *log = mp->m_log;
	xlog_in_core_t	 *iclog;
#ifdef DEBUG
	xlog_in_core_t	 *first_iclog;
#endif
	xfs_log_iovec_t  reg[1];
	xfs_log_ticket_t tic = NULL;
	xfs_lsn_t	 lsn;
	int		 error;

	/* the data section must be 32 bit size aligned */
	struct {
	    __uint16_t magic;
	    __uint16_t pad1;
	    __uint32_t pad2; /* may as well make it 64 bits */
	} magic = { XLOG_UNMOUNT_TYPE, 0, 0 };

	/*
	 * Don't write out unmount record on read-only mounts.
	 * Or, if we are doing a forced umount (typically because of IO errors).
	 */
654
	if (mp->m_flags & XFS_MOUNT_RDONLY)
L
Linus Torvalds 已提交
655 656
		return 0;

657 658
	error = _xfs_log_force(mp, 0, XFS_LOG_FORCE|XFS_LOG_SYNC, NULL);
	ASSERT(error || !(XLOG_FORCED_SHUTDOWN(log)));
L
Linus Torvalds 已提交
659 660 661 662 663 664 665 666 667 668 669 670 671 672

#ifdef DEBUG
	first_iclog = iclog = log->l_iclog;
	do {
		if (!(iclog->ic_state & XLOG_STATE_IOERROR)) {
			ASSERT(iclog->ic_state & XLOG_STATE_ACTIVE);
			ASSERT(iclog->ic_offset == 0);
		}
		iclog = iclog->ic_next;
	} while (iclog != first_iclog);
#endif
	if (! (XLOG_FORCED_SHUTDOWN(log))) {
		reg[0].i_addr = (void*)&magic;
		reg[0].i_len  = sizeof(magic);
673
		XLOG_VEC_SET_TYPE(&reg[0], XLOG_REG_TYPE_UNMOUNT);
L
Linus Torvalds 已提交
674

675 676
		error = xfs_log_reserve(mp, 600, 1, &tic,
					XFS_LOG, 0, XLOG_UNMOUNT_REC_TYPE);
L
Linus Torvalds 已提交
677 678 679 680 681 682 683 684 685 686 687 688 689 690 691 692 693 694
		if (!error) {
			/* remove inited flag */
			((xlog_ticket_t *)tic)->t_flags = 0;
			error = xlog_write(mp, reg, 1, tic, &lsn,
					   NULL, XLOG_UNMOUNT_TRANS);
			/*
			 * At this point, we're umounting anyway,
			 * so there's no point in transitioning log state
			 * to IOERROR. Just continue...
			 */
		}

		if (error) {
			xfs_fs_cmn_err(CE_ALERT, mp,
				"xfs_log_unmount: unmount record failed");
		}


E
Eric Sandeen 已提交
695
		spin_lock(&log->l_icloglock);
L
Linus Torvalds 已提交
696
		iclog = log->l_iclog;
697
		atomic_inc(&iclog->ic_refcnt);
E
Eric Sandeen 已提交
698
		spin_unlock(&log->l_icloglock);
L
Linus Torvalds 已提交
699
		xlog_state_want_sync(log, iclog);
700
		error = xlog_state_release_iclog(log, iclog);
L
Linus Torvalds 已提交
701

E
Eric Sandeen 已提交
702
		spin_lock(&log->l_icloglock);
L
Linus Torvalds 已提交
703 704 705 706 707 708
		if (!(iclog->ic_state == XLOG_STATE_ACTIVE ||
		      iclog->ic_state == XLOG_STATE_DIRTY)) {
			if (!XLOG_FORCED_SHUTDOWN(log)) {
				sv_wait(&iclog->ic_forcesema, PMEM,
					&log->l_icloglock, s);
			} else {
E
Eric Sandeen 已提交
709
				spin_unlock(&log->l_icloglock);
L
Linus Torvalds 已提交
710 711
			}
		} else {
E
Eric Sandeen 已提交
712
			spin_unlock(&log->l_icloglock);
L
Linus Torvalds 已提交
713
		}
714 715 716
		if (tic) {
			xlog_trace_loggrant(log, tic, "unmount rec");
			xlog_ungrant_log_space(log, tic);
717
			xlog_ticket_put(log, tic);
718
		}
L
Linus Torvalds 已提交
719 720 721 722 723 724 725
	} else {
		/*
		 * We're already in forced_shutdown mode, couldn't
		 * even attempt to write out the unmount transaction.
		 *
		 * Go through the motions of sync'ing and releasing
		 * the iclog, even though no I/O will actually happen,
726
		 * we need to wait for other log I/Os that may already
L
Linus Torvalds 已提交
727 728 729 730 731 732
		 * be in progress.  Do this as a separate section of
		 * code so we'll know if we ever get stuck here that
		 * we're in this odd situation of trying to unmount
		 * a file system that went into forced_shutdown as
		 * the result of an unmount..
		 */
E
Eric Sandeen 已提交
733
		spin_lock(&log->l_icloglock);
L
Linus Torvalds 已提交
734
		iclog = log->l_iclog;
735
		atomic_inc(&iclog->ic_refcnt);
E
Eric Sandeen 已提交
736
		spin_unlock(&log->l_icloglock);
L
Linus Torvalds 已提交
737 738

		xlog_state_want_sync(log, iclog);
739
		error =  xlog_state_release_iclog(log, iclog);
L
Linus Torvalds 已提交
740

E
Eric Sandeen 已提交
741
		spin_lock(&log->l_icloglock);
L
Linus Torvalds 已提交
742 743 744 745 746 747 748 749

		if ( ! (   iclog->ic_state == XLOG_STATE_ACTIVE
			|| iclog->ic_state == XLOG_STATE_DIRTY
			|| iclog->ic_state == XLOG_STATE_IOERROR) ) {

				sv_wait(&iclog->ic_forcesema, PMEM,
					&log->l_icloglock, s);
		} else {
E
Eric Sandeen 已提交
750
			spin_unlock(&log->l_icloglock);
L
Linus Torvalds 已提交
751 752 753
		}
	}

754
	return error;
L
Linus Torvalds 已提交
755 756 757 758
}	/* xfs_log_unmount_write */

/*
 * Deallocate log structures for unmount/relocation.
759 760 761
 *
 * We need to stop the aild from running before we destroy
 * and deallocate the log as the aild references the log.
L
Linus Torvalds 已提交
762 763 764 765
 */
void
xfs_log_unmount_dealloc(xfs_mount_t *mp)
{
766
	xfs_trans_ail_destroy(mp);
767
	xlog_dealloc_log(mp->m_log);
L
Linus Torvalds 已提交
768 769 770 771 772 773 774 775 776 777 778 779 780 781 782 783 784 785 786 787 788
}

/*
 * Write region vectors to log.  The write happens using the space reservation
 * of the ticket (tic).  It is not a requirement that all writes for a given
 * transaction occur with one call to xfs_log_write().
 */
int
xfs_log_write(xfs_mount_t *	mp,
	      xfs_log_iovec_t	reg[],
	      int		nentries,
	      xfs_log_ticket_t	tic,
	      xfs_lsn_t		*start_lsn)
{
	int	error;
	xlog_t *log = mp->m_log;

	if (XLOG_FORCED_SHUTDOWN(log))
		return XFS_ERROR(EIO);

	if ((error = xlog_write(mp, reg, nentries, tic, start_lsn, NULL, 0))) {
789
		xfs_force_shutdown(mp, SHUTDOWN_LOG_IO_ERROR);
L
Linus Torvalds 已提交
790
	}
791
	return error;
L
Linus Torvalds 已提交
792 793 794 795 796 797 798 799 800 801 802 803 804 805 806 807
}	/* xfs_log_write */


void
xfs_log_move_tail(xfs_mount_t	*mp,
		  xfs_lsn_t	tail_lsn)
{
	xlog_ticket_t	*tic;
	xlog_t		*log = mp->m_log;
	int		need_bytes, free_bytes, cycle, bytes;

	if (XLOG_FORCED_SHUTDOWN(log))
		return;

	if (tail_lsn == 0) {
		/* needed since sync_lsn is 64 bits */
E
Eric Sandeen 已提交
808
		spin_lock(&log->l_icloglock);
L
Linus Torvalds 已提交
809
		tail_lsn = log->l_last_sync_lsn;
E
Eric Sandeen 已提交
810
		spin_unlock(&log->l_icloglock);
L
Linus Torvalds 已提交
811 812
	}

E
Eric Sandeen 已提交
813
	spin_lock(&log->l_grant_lock);
L
Linus Torvalds 已提交
814 815 816 817 818 819 820 821 822 823 824 825 826 827 828 829 830 831 832 833 834 835 836 837 838 839 840 841 842 843 844 845 846 847 848 849 850 851 852 853 854 855 856 857 858 859 860 861

	/* Also an invalid lsn.  1 implies that we aren't passing in a valid
	 * tail_lsn.
	 */
	if (tail_lsn != 1) {
		log->l_tail_lsn = tail_lsn;
	}

	if ((tic = log->l_write_headq)) {
#ifdef DEBUG
		if (log->l_flags & XLOG_ACTIVE_RECOVERY)
			panic("Recovery problem");
#endif
		cycle = log->l_grant_write_cycle;
		bytes = log->l_grant_write_bytes;
		free_bytes = xlog_space_left(log, cycle, bytes);
		do {
			ASSERT(tic->t_flags & XLOG_TIC_PERM_RESERV);

			if (free_bytes < tic->t_unit_res && tail_lsn != 1)
				break;
			tail_lsn = 0;
			free_bytes -= tic->t_unit_res;
			sv_signal(&tic->t_sema);
			tic = tic->t_next;
		} while (tic != log->l_write_headq);
	}
	if ((tic = log->l_reserve_headq)) {
#ifdef DEBUG
		if (log->l_flags & XLOG_ACTIVE_RECOVERY)
			panic("Recovery problem");
#endif
		cycle = log->l_grant_reserve_cycle;
		bytes = log->l_grant_reserve_bytes;
		free_bytes = xlog_space_left(log, cycle, bytes);
		do {
			if (tic->t_flags & XLOG_TIC_PERM_RESERV)
				need_bytes = tic->t_unit_res*tic->t_cnt;
			else
				need_bytes = tic->t_unit_res;
			if (free_bytes < need_bytes && tail_lsn != 1)
				break;
			tail_lsn = 0;
			free_bytes -= need_bytes;
			sv_signal(&tic->t_sema);
			tic = tic->t_next;
		} while (tic != log->l_reserve_headq);
	}
E
Eric Sandeen 已提交
862
	spin_unlock(&log->l_grant_lock);
L
Linus Torvalds 已提交
863 864 865 866 867 868 869 870 871 872 873 874 875 876
}	/* xfs_log_move_tail */

/*
 * Determine if we have a transaction that has gone to disk
 * that needs to be covered. Log activity needs to be idle (no AIL and
 * nothing in the iclogs). And, we need to be in the right state indicating
 * something has gone out.
 */
int
xfs_log_need_covered(xfs_mount_t *mp)
{
	int		needed = 0, gen;
	xlog_t		*log = mp->m_log;

D
David Chinner 已提交
877
	if (!xfs_fs_writable(mp))
L
Linus Torvalds 已提交
878 879
		return 0;

E
Eric Sandeen 已提交
880
	spin_lock(&log->l_icloglock);
L
Linus Torvalds 已提交
881 882 883 884 885 886 887 888 889 890 891 892
	if (((log->l_covered_state == XLOG_STATE_COVER_NEED) ||
		(log->l_covered_state == XLOG_STATE_COVER_NEED2))
			&& !xfs_trans_first_ail(mp, &gen)
			&& xlog_iclogs_empty(log)) {
		if (log->l_covered_state == XLOG_STATE_COVER_NEED)
			log->l_covered_state = XLOG_STATE_COVER_DONE;
		else {
			ASSERT(log->l_covered_state == XLOG_STATE_COVER_NEED2);
			log->l_covered_state = XLOG_STATE_COVER_DONE2;
		}
		needed = 1;
	}
E
Eric Sandeen 已提交
893
	spin_unlock(&log->l_icloglock);
894
	return needed;
L
Linus Torvalds 已提交
895 896 897 898 899 900 901 902 903 904 905 906 907 908 909 910 911 912 913 914 915 916 917 918 919 920
}

/******************************************************************************
 *
 *	local routines
 *
 ******************************************************************************
 */

/* xfs_trans_tail_ail returns 0 when there is nothing in the list.
 * The log manager must keep track of the last LR which was committed
 * to disk.  The lsn of this LR will become the new tail_lsn whenever
 * xfs_trans_tail_ail returns 0.  If we don't do this, we run into
 * the situation where stuff could be written into the log but nothing
 * was ever in the AIL when asked.  Eventually, we panic since the
 * tail hits the head.
 *
 * We may be holding the log iclog lock upon entering this routine.
 */
xfs_lsn_t
xlog_assign_tail_lsn(xfs_mount_t *mp)
{
	xfs_lsn_t tail_lsn;
	xlog_t	  *log = mp->m_log;

	tail_lsn = xfs_trans_tail_ail(mp);
E
Eric Sandeen 已提交
921
	spin_lock(&log->l_grant_lock);
L
Linus Torvalds 已提交
922 923 924 925 926
	if (tail_lsn != 0) {
		log->l_tail_lsn = tail_lsn;
	} else {
		tail_lsn = log->l_tail_lsn = log->l_last_sync_lsn;
	}
E
Eric Sandeen 已提交
927
	spin_unlock(&log->l_grant_lock);
L
Linus Torvalds 已提交
928 929 930 931 932 933 934 935 936 937 938 939 940 941 942 943 944 945 946

	return tail_lsn;
}	/* xlog_assign_tail_lsn */


/*
 * Return the space in the log between the tail and the head.  The head
 * is passed in the cycle/bytes formal parms.  In the special case where
 * the reserve head has wrapped passed the tail, this calculation is no
 * longer valid.  In this case, just return 0 which means there is no space
 * in the log.  This works for all places where this function is called
 * with the reserve head.  Of course, if the write head were to ever
 * wrap the tail, we should blow up.  Rather than catch this case here,
 * we depend on other ASSERTions in other parts of the code.   XXXmiken
 *
 * This code also handles the case where the reservation head is behind
 * the tail.  The details of this case are described below, but the end
 * result is that we return the size of the log as the amount of space left.
 */
D
David Chinner 已提交
947
STATIC int
L
Linus Torvalds 已提交
948 949 950 951 952 953 954 955 956 957 958 959 960 961 962 963 964 965 966 967 968 969 970 971 972 973 974 975 976 977 978 979 980 981 982 983 984 985 986 987 988 989 990 991 992 993 994 995 996 997 998 999 1000 1001 1002 1003 1004 1005
xlog_space_left(xlog_t *log, int cycle, int bytes)
{
	int free_bytes;
	int tail_bytes;
	int tail_cycle;

	tail_bytes = BBTOB(BLOCK_LSN(log->l_tail_lsn));
	tail_cycle = CYCLE_LSN(log->l_tail_lsn);
	if ((tail_cycle == cycle) && (bytes >= tail_bytes)) {
		free_bytes = log->l_logsize - (bytes - tail_bytes);
	} else if ((tail_cycle + 1) < cycle) {
		return 0;
	} else if (tail_cycle < cycle) {
		ASSERT(tail_cycle == (cycle - 1));
		free_bytes = tail_bytes - bytes;
	} else {
		/*
		 * The reservation head is behind the tail.
		 * In this case we just want to return the size of the
		 * log as the amount of space left.
		 */
		xfs_fs_cmn_err(CE_ALERT, log->l_mp,
			"xlog_space_left: head behind tail\n"
			"  tail_cycle = %d, tail_bytes = %d\n"
			"  GH   cycle = %d, GH   bytes = %d",
			tail_cycle, tail_bytes, cycle, bytes);
		ASSERT(0);
		free_bytes = log->l_logsize;
	}
	return free_bytes;
}	/* xlog_space_left */


/*
 * Log function which is called when an io completes.
 *
 * The log manager needs its own routine, in order to control what
 * happens with the buffer after the write completes.
 */
void
xlog_iodone(xfs_buf_t *bp)
{
	xlog_in_core_t	*iclog;
	xlog_t		*l;
	int		aborted;

	iclog = XFS_BUF_FSPRIVATE(bp, xlog_in_core_t *);
	ASSERT(XFS_BUF_FSPRIVATE2(bp, unsigned long) == (unsigned long) 2);
	XFS_BUF_SET_FSPRIVATE2(bp, (unsigned long)1);
	aborted = 0;

	/*
	 * Some versions of cpp barf on the recursive definition of
	 * ic_log -> hic_fields.ic_log and expand ic_log twice when
	 * it is passed through two macros.  Workaround broken cpp.
	 */
	l = iclog->ic_log;

1006 1007 1008 1009 1010 1011 1012 1013 1014 1015 1016 1017 1018
	/*
	 * If the ordered flag has been removed by a lower
	 * layer, it means the underlyin device no longer supports
	 * barrier I/O. Warn loudly and turn off barriers.
	 */
	if ((l->l_mp->m_flags & XFS_MOUNT_BARRIER) && !XFS_BUF_ORDERED(bp)) {
		l->l_mp->m_flags &= ~XFS_MOUNT_BARRIER;
		xfs_fs_cmn_err(CE_WARN, l->l_mp,
				"xlog_iodone: Barriers are no longer supported"
				" by device. Disabling barriers\n");
		xfs_buftrace("XLOG_IODONE BARRIERS OFF", bp);
	}

L
Linus Torvalds 已提交
1019 1020 1021 1022 1023 1024 1025
	/*
	 * Race to shutdown the filesystem if we see an error.
	 */
	if (XFS_TEST_ERROR((XFS_BUF_GETERROR(bp)), l->l_mp,
			XFS_ERRTAG_IODONE_IOERR, XFS_RANDOM_IODONE_IOERR)) {
		xfs_ioerror_alert("xlog_iodone", l->l_mp, bp, XFS_BUF_ADDR(bp));
		XFS_BUF_STALE(bp);
1026
		xfs_force_shutdown(l->l_mp, SHUTDOWN_LOG_IO_ERROR);
L
Linus Torvalds 已提交
1027 1028 1029 1030 1031 1032 1033 1034 1035
		/*
		 * This flag will be propagated to the trans-committed
		 * callback routines to let them know that the log-commit
		 * didn't succeed.
		 */
		aborted = XFS_LI_ABORTED;
	} else if (iclog->ic_state & XLOG_STATE_IOERROR) {
		aborted = XFS_LI_ABORTED;
	}
1036 1037 1038

	/* log I/O is always issued ASYNC */
	ASSERT(XFS_BUF_ISASYNC(bp));
L
Linus Torvalds 已提交
1039
	xlog_state_done_syncing(iclog, aborted);
1040 1041 1042 1043 1044 1045
	/*
	 * do not reference the buffer (bp) here as we could race
	 * with it being freed after writing the unmount record to the
	 * log.
	 */

L
Linus Torvalds 已提交
1046 1047 1048 1049 1050 1051 1052 1053 1054 1055 1056 1057 1058 1059 1060 1061 1062 1063 1064 1065 1066 1067 1068 1069 1070 1071 1072 1073 1074
}	/* xlog_iodone */

/*
 * The bdstrat callback function for log bufs. This gives us a central
 * place to trap bufs in case we get hit by a log I/O error and need to
 * shutdown. Actually, in practice, even when we didn't get a log error,
 * we transition the iclogs to IOERROR state *after* flushing all existing
 * iclogs to disk. This is because we don't want anymore new transactions to be
 * started or completed afterwards.
 */
STATIC int
xlog_bdstrat_cb(struct xfs_buf *bp)
{
	xlog_in_core_t *iclog;

	iclog = XFS_BUF_FSPRIVATE(bp, xlog_in_core_t *);

	if ((iclog->ic_state & XLOG_STATE_IOERROR) == 0) {
	  /* note for irix bstrat will need  struct bdevsw passed
	   * Fix the following macro if the code ever is merged
	   */
	    XFS_bdstrat(bp);
		return 0;
	}

	xfs_buftrace("XLOG__BDSTRAT IOERROR", bp);
	XFS_BUF_ERROR(bp, EIO);
	XFS_BUF_STALE(bp);
	xfs_biodone(bp);
1075
	return XFS_ERROR(EIO);
L
Linus Torvalds 已提交
1076 1077 1078 1079 1080 1081 1082


}

/*
 * Return size of each in-core log record buffer.
 *
1083
 * All machines get 8 x 32KB buffers by default, unless tuned otherwise.
L
Linus Torvalds 已提交
1084 1085 1086 1087 1088 1089 1090 1091 1092 1093 1094 1095
 *
 * If the filesystem blocksize is too large, we may need to choose a
 * larger size since the directory code currently logs entire blocks.
 */

STATIC void
xlog_get_iclog_buffer_size(xfs_mount_t	*mp,
			   xlog_t	*log)
{
	int size;
	int xhdrs;

1096 1097 1098
	if (mp->m_logbufs <= 0)
		log->l_iclog_bufs = XLOG_MAX_ICLOGS;
	else
1099
		log->l_iclog_bufs = mp->m_logbufs;
L
Linus Torvalds 已提交
1100 1101 1102 1103

	/*
	 * Buffer size passed in from mount system call.
	 */
1104
	if (mp->m_logbsize > 0) {
L
Linus Torvalds 已提交
1105 1106 1107 1108 1109 1110 1111
		size = log->l_iclog_size = mp->m_logbsize;
		log->l_iclog_size_log = 0;
		while (size != 1) {
			log->l_iclog_size_log++;
			size >>= 1;
		}

1112
		if (xfs_sb_version_haslogv2(&mp->m_sb)) {
L
Linus Torvalds 已提交
1113 1114 1115 1116 1117 1118 1119 1120 1121 1122 1123 1124 1125 1126
			/* # headers = size / 32K
			 * one header holds cycles from 32K of data
			 */

			xhdrs = mp->m_logbsize / XLOG_HEADER_CYCLE_SIZE;
			if (mp->m_logbsize % XLOG_HEADER_CYCLE_SIZE)
				xhdrs++;
			log->l_iclog_hsize = xhdrs << BBSHIFT;
			log->l_iclog_heads = xhdrs;
		} else {
			ASSERT(mp->m_logbsize <= XLOG_BIG_RECORD_BSIZE);
			log->l_iclog_hsize = BBSIZE;
			log->l_iclog_heads = 1;
		}
1127
		goto done;
L
Linus Torvalds 已提交
1128 1129
	}

1130 1131 1132
	/* All machines use 32KB buffers by default. */
	log->l_iclog_size = XLOG_BIG_RECORD_BSIZE;
	log->l_iclog_size_log = XLOG_BIG_RECORD_BSHIFT;
L
Linus Torvalds 已提交
1133 1134 1135 1136 1137 1138 1139 1140 1141 1142 1143 1144

	/* the default log size is 16k or 32k which is one header sector */
	log->l_iclog_hsize = BBSIZE;
	log->l_iclog_heads = 1;

	/*
	 * For 16KB, we use 3 32KB buffers.  For 32KB block sizes, we use
	 * 4 32KB buffers.  For 64KB block sizes, we use 8 32KB buffers.
	 */
	if (mp->m_sb.sb_blocksize >= 16*1024) {
		log->l_iclog_size = XLOG_BIG_RECORD_BSIZE;
		log->l_iclog_size_log = XLOG_BIG_RECORD_BSHIFT;
1145
		if (mp->m_logbufs <= 0) {
L
Linus Torvalds 已提交
1146 1147 1148 1149 1150 1151 1152 1153 1154 1155 1156 1157 1158 1159 1160 1161
			switch (mp->m_sb.sb_blocksize) {
			    case 16*1024:			/* 16 KB */
				log->l_iclog_bufs = 3;
				break;
			    case 32*1024:			/* 32 KB */
				log->l_iclog_bufs = 4;
				break;
			    case 64*1024:			/* 64 KB */
				log->l_iclog_bufs = 8;
				break;
			    default:
				xlog_panic("XFS: Invalid blocksize");
				break;
			}
		}
	}
1162 1163 1164 1165 1166 1167

done:	/* are we being asked to make the sizes selected above visible? */
	if (mp->m_logbufs == 0)
		mp->m_logbufs = log->l_iclog_bufs;
	if (mp->m_logbsize == 0)
		mp->m_logbsize = log->l_iclog_size;
L
Linus Torvalds 已提交
1168 1169 1170 1171 1172 1173 1174 1175 1176 1177 1178 1179 1180 1181 1182 1183 1184 1185 1186 1187 1188 1189 1190 1191 1192 1193 1194 1195 1196 1197 1198 1199 1200
}	/* xlog_get_iclog_buffer_size */


/*
 * This routine initializes some of the log structure for a given mount point.
 * Its primary purpose is to fill in enough, so recovery can occur.  However,
 * some other stuff may be filled in too.
 */
STATIC xlog_t *
xlog_alloc_log(xfs_mount_t	*mp,
	       xfs_buftarg_t	*log_target,
	       xfs_daddr_t	blk_offset,
	       int		num_bblks)
{
	xlog_t			*log;
	xlog_rec_header_t	*head;
	xlog_in_core_t		**iclogp;
	xlog_in_core_t		*iclog, *prev_iclog=NULL;
	xfs_buf_t		*bp;
	int			i;
	int			iclogsize;

	log = (xlog_t *)kmem_zalloc(sizeof(xlog_t), KM_SLEEP);

	log->l_mp	   = mp;
	log->l_targ	   = log_target;
	log->l_logsize     = BBTOB(num_bblks);
	log->l_logBBstart  = blk_offset;
	log->l_logBBsize   = num_bblks;
	log->l_covered_state = XLOG_STATE_COVER_IDLE;
	log->l_flags	   |= XLOG_ACTIVE_RECOVERY;

	log->l_prev_block  = -1;
1201
	log->l_tail_lsn	   = xlog_assign_lsn(1, 0);
L
Linus Torvalds 已提交
1202 1203 1204 1205 1206 1207
	/* log->l_tail_lsn = 0x100000000LL; cycle = 1; current block = 0 */
	log->l_last_sync_lsn = log->l_tail_lsn;
	log->l_curr_cycle  = 1;	    /* 0 is bad since this is initial value */
	log->l_grant_reserve_cycle = 1;
	log->l_grant_write_cycle = 1;

1208
	if (xfs_sb_version_hassector(&mp->m_sb)) {
L
Linus Torvalds 已提交
1209 1210 1211 1212 1213
		log->l_sectbb_log = mp->m_sb.sb_logsectlog - BBSHIFT;
		ASSERT(log->l_sectbb_log <= mp->m_sectbb_log);
		/* for larger sector sizes, must have v2 or external log */
		ASSERT(log->l_sectbb_log == 0 ||
			log->l_logBBstart == 0 ||
1214
			xfs_sb_version_haslogv2(&mp->m_sb));
L
Linus Torvalds 已提交
1215 1216 1217 1218 1219 1220 1221 1222 1223 1224 1225 1226 1227 1228
		ASSERT(mp->m_sb.sb_logsectlog >= BBSHIFT);
	}
	log->l_sectbb_mask = (1 << log->l_sectbb_log) - 1;

	xlog_get_iclog_buffer_size(mp, log);

	bp = xfs_buf_get_empty(log->l_iclog_size, mp->m_logdev_targp);
	XFS_BUF_SET_IODONE_FUNC(bp, xlog_iodone);
	XFS_BUF_SET_BDSTRAT_FUNC(bp, xlog_bdstrat_cb);
	XFS_BUF_SET_FSPRIVATE2(bp, (unsigned long)1);
	ASSERT(XFS_BUF_ISBUSY(bp));
	ASSERT(XFS_BUF_VALUSEMA(bp) <= 0);
	log->l_xbuf = bp;

E
Eric Sandeen 已提交
1229 1230
	spin_lock_init(&log->l_icloglock);
	spin_lock_init(&log->l_grant_lock);
L
Linus Torvalds 已提交
1231 1232 1233 1234 1235 1236 1237 1238 1239 1240 1241 1242 1243 1244 1245 1246 1247 1248 1249 1250 1251
	initnsema(&log->l_flushsema, 0, "ic-flush");

	/* log record size must be multiple of BBSIZE; see xlog_rec_header_t */
	ASSERT((XFS_BUF_SIZE(bp) & BBMASK) == 0);

	iclogp = &log->l_iclog;
	/*
	 * The amount of memory to allocate for the iclog structure is
	 * rather funky due to the way the structure is defined.  It is
	 * done this way so that we can use different sizes for machines
	 * with different amounts of memory.  See the definition of
	 * xlog_in_core_t in xfs_log_priv.h for details.
	 */
	iclogsize = log->l_iclog_size;
	ASSERT(log->l_iclog_size >= 4096);
	for (i=0; i < log->l_iclog_bufs; i++) {
		*iclogp = (xlog_in_core_t *)
			  kmem_zalloc(sizeof(xlog_in_core_t), KM_SLEEP);
		iclog = *iclogp;
		iclog->ic_prev = prev_iclog;
		prev_iclog = iclog;
1252 1253 1254 1255 1256 1257 1258 1259 1260

		bp = xfs_buf_get_noaddr(log->l_iclog_size, mp->m_logdev_targp);
		if (!XFS_BUF_CPSEMA(bp))
			ASSERT(0);
		XFS_BUF_SET_IODONE_FUNC(bp, xlog_iodone);
		XFS_BUF_SET_BDSTRAT_FUNC(bp, xlog_bdstrat_cb);
		XFS_BUF_SET_FSPRIVATE2(bp, (unsigned long)1);
		iclog->ic_bp = bp;
		iclog->hic_data = bp->b_addr;
1261
#ifdef DEBUG
L
Linus Torvalds 已提交
1262
		log->l_iclog_bak[i] = (xfs_caddr_t)&(iclog->ic_header);
1263
#endif
L
Linus Torvalds 已提交
1264 1265
		head = &iclog->ic_header;
		memset(head, 0, sizeof(xlog_rec_header_t));
1266 1267
		head->h_magicno = cpu_to_be32(XLOG_HEADER_MAGIC_NUM);
		head->h_version = cpu_to_be32(
1268
			xfs_sb_version_haslogv2(&log->l_mp->m_sb) ? 2 : 1);
1269
		head->h_size = cpu_to_be32(log->l_iclog_size);
L
Linus Torvalds 已提交
1270
		/* new fields */
1271
		head->h_fmt = cpu_to_be32(XLOG_FMT);
L
Linus Torvalds 已提交
1272 1273 1274 1275 1276
		memcpy(&head->h_fs_uuid, &mp->m_sb.sb_uuid, sizeof(uuid_t));

		iclog->ic_size = XFS_BUF_SIZE(bp) - log->l_iclog_hsize;
		iclog->ic_state = XLOG_STATE_ACTIVE;
		iclog->ic_log = log;
1277 1278
		atomic_set(&iclog->ic_refcnt, 0);
		spin_lock_init(&iclog->ic_callback_lock);
L
Linus Torvalds 已提交
1279 1280 1281 1282 1283 1284 1285 1286 1287 1288 1289 1290 1291 1292 1293 1294 1295 1296 1297 1298 1299 1300 1301 1302 1303 1304 1305 1306 1307 1308 1309 1310
		iclog->ic_callback_tail = &(iclog->ic_callback);
		iclog->ic_datap = (char *)iclog->hic_data + log->l_iclog_hsize;

		ASSERT(XFS_BUF_ISBUSY(iclog->ic_bp));
		ASSERT(XFS_BUF_VALUSEMA(iclog->ic_bp) <= 0);
		sv_init(&iclog->ic_forcesema, SV_DEFAULT, "iclog-force");
		sv_init(&iclog->ic_writesema, SV_DEFAULT, "iclog-write");

		iclogp = &iclog->ic_next;
	}
	*iclogp = log->l_iclog;			/* complete ring */
	log->l_iclog->ic_prev = prev_iclog;	/* re-write 1st prev ptr */

	return log;
}	/* xlog_alloc_log */


/*
 * Write out the commit record of a transaction associated with the given
 * ticket.  Return the lsn of the commit record.
 */
STATIC int
xlog_commit_record(xfs_mount_t  *mp,
		   xlog_ticket_t *ticket,
		   xlog_in_core_t **iclog,
		   xfs_lsn_t	*commitlsnp)
{
	int		error;
	xfs_log_iovec_t	reg[1];

	reg[0].i_addr = NULL;
	reg[0].i_len = 0;
1311
	XLOG_VEC_SET_TYPE(&reg[0], XLOG_REG_TYPE_COMMIT);
L
Linus Torvalds 已提交
1312 1313 1314 1315

	ASSERT_ALWAYS(iclog);
	if ((error = xlog_write(mp, reg, 1, ticket, commitlsnp,
			       iclog, XLOG_COMMIT_TRANS))) {
1316
		xfs_force_shutdown(mp, SHUTDOWN_LOG_IO_ERROR);
L
Linus Torvalds 已提交
1317
	}
1318
	return error;
L
Linus Torvalds 已提交
1319 1320 1321 1322 1323 1324 1325 1326 1327 1328
}	/* xlog_commit_record */


/*
 * Push on the buffer cache code if we ever use more than 75% of the on-disk
 * log space.  This code pushes on the lsn which would supposedly free up
 * the 25% which we want to leave free.  We may need to adopt a policy which
 * pushes on an lsn which is further along in the log once we reach the high
 * water mark.  In this manner, we would be creating a low water mark.
 */
D
David Chinner 已提交
1329
STATIC void
L
Linus Torvalds 已提交
1330 1331 1332 1333 1334 1335 1336 1337 1338 1339 1340 1341 1342 1343
xlog_grant_push_ail(xfs_mount_t	*mp,
		    int		need_bytes)
{
    xlog_t	*log = mp->m_log;	/* pointer to the log */
    xfs_lsn_t	tail_lsn;		/* lsn of the log tail */
    xfs_lsn_t	threshold_lsn = 0;	/* lsn we'd like to be at */
    int		free_blocks;		/* free blocks left to write to */
    int		free_bytes;		/* free bytes left to write to */
    int		threshold_block;	/* block in lsn we'd like to be at */
    int		threshold_cycle;	/* lsn cycle we'd like to be at */
    int		free_threshold;

    ASSERT(BTOBB(need_bytes) < log->l_logBBsize);

E
Eric Sandeen 已提交
1344
    spin_lock(&log->l_grant_lock);
L
Linus Torvalds 已提交
1345 1346 1347 1348 1349 1350 1351 1352 1353 1354 1355 1356 1357 1358 1359 1360 1361 1362 1363 1364 1365
    free_bytes = xlog_space_left(log,
				 log->l_grant_reserve_cycle,
				 log->l_grant_reserve_bytes);
    tail_lsn = log->l_tail_lsn;
    free_blocks = BTOBBT(free_bytes);

    /*
     * Set the threshold for the minimum number of free blocks in the
     * log to the maximum of what the caller needs, one quarter of the
     * log, and 256 blocks.
     */
    free_threshold = BTOBB(need_bytes);
    free_threshold = MAX(free_threshold, (log->l_logBBsize >> 2));
    free_threshold = MAX(free_threshold, 256);
    if (free_blocks < free_threshold) {
	threshold_block = BLOCK_LSN(tail_lsn) + free_threshold;
	threshold_cycle = CYCLE_LSN(tail_lsn);
	if (threshold_block >= log->l_logBBsize) {
	    threshold_block -= log->l_logBBsize;
	    threshold_cycle += 1;
	}
1366
	threshold_lsn = xlog_assign_lsn(threshold_cycle, threshold_block);
L
Linus Torvalds 已提交
1367 1368 1369 1370 1371 1372 1373

	/* Don't pass in an lsn greater than the lsn of the last
	 * log record known to be on disk.
	 */
	if (XFS_LSN_CMP(threshold_lsn, log->l_last_sync_lsn) > 0)
	    threshold_lsn = log->l_last_sync_lsn;
    }
E
Eric Sandeen 已提交
1374
    spin_unlock(&log->l_grant_lock);
L
Linus Torvalds 已提交
1375 1376 1377 1378 1379 1380 1381 1382 1383 1384 1385 1386 1387 1388 1389 1390 1391 1392 1393 1394 1395 1396 1397 1398 1399 1400 1401 1402 1403 1404 1405 1406 1407 1408 1409 1410 1411

    /*
     * Get the transaction layer to kick the dirty buffers out to
     * disk asynchronously. No point in trying to do this if
     * the filesystem is shutting down.
     */
    if (threshold_lsn &&
	!XLOG_FORCED_SHUTDOWN(log))
	    xfs_trans_push_ail(mp, threshold_lsn);
}	/* xlog_grant_push_ail */


/*
 * Flush out the in-core log (iclog) to the on-disk log in an asynchronous 
 * fashion.  Previously, we should have moved the current iclog
 * ptr in the log to point to the next available iclog.  This allows further
 * write to continue while this code syncs out an iclog ready to go.
 * Before an in-core log can be written out, the data section must be scanned
 * to save away the 1st word of each BBSIZE block into the header.  We replace
 * it with the current cycle count.  Each BBSIZE block is tagged with the
 * cycle count because there in an implicit assumption that drives will
 * guarantee that entire 512 byte blocks get written at once.  In other words,
 * we can't have part of a 512 byte block written and part not written.  By
 * tagging each block, we will know which blocks are valid when recovering
 * after an unclean shutdown.
 *
 * This routine is single threaded on the iclog.  No other thread can be in
 * this routine with the same iclog.  Changing contents of iclog can there-
 * fore be done without grabbing the state machine lock.  Updating the global
 * log will require grabbing the lock though.
 *
 * The entire log manager uses a logical block numbering scheme.  Only
 * log_sync (and then only bwrite()) know about the fact that the log may
 * not start with block zero on a given device.  The log block start offset
 * is added immediately before calling bwrite().
 */

D
David Chinner 已提交
1412
STATIC int
L
Linus Torvalds 已提交
1413 1414 1415 1416 1417
xlog_sync(xlog_t		*log,
	  xlog_in_core_t	*iclog)
{
	xfs_caddr_t	dptr;		/* pointer to byte sized element */
	xfs_buf_t	*bp;
1418
	int		i;
L
Linus Torvalds 已提交
1419 1420 1421 1422 1423
	uint		count;		/* byte count of bwrite */
	uint		count_init;	/* initial count before roundup */
	int		roundoff;       /* roundoff to BB or stripe */
	int		split = 0;	/* split write into two regions */
	int		error;
1424
	int		v2 = xfs_sb_version_haslogv2(&log->l_mp->m_sb);
L
Linus Torvalds 已提交
1425 1426

	XFS_STATS_INC(xs_log_writes);
1427
	ASSERT(atomic_read(&iclog->ic_refcnt) == 0);
L
Linus Torvalds 已提交
1428 1429 1430 1431 1432 1433 1434 1435 1436 1437 1438 1439 1440 1441 1442 1443 1444 1445 1446 1447

	/* Add for LR header */
	count_init = log->l_iclog_hsize + iclog->ic_offset;

	/* Round out the log write size */
	if (v2 && log->l_mp->m_sb.sb_logsunit > 1) {
		/* we have a v2 stripe unit to use */
		count = XLOG_LSUNITTOB(log, XLOG_BTOLSUNIT(log, count_init));
	} else {
		count = BBTOB(BTOBB(count_init));
	}
	roundoff = count - count_init;
	ASSERT(roundoff >= 0);
	ASSERT((v2 && log->l_mp->m_sb.sb_logsunit > 1 && 
                roundoff < log->l_mp->m_sb.sb_logsunit)
		|| 
		(log->l_mp->m_sb.sb_logsunit <= 1 && 
		 roundoff < BBTOB(1)));

	/* move grant heads by roundoff in sync */
E
Eric Sandeen 已提交
1448
	spin_lock(&log->l_grant_lock);
1449
	xlog_grant_add_space(log, roundoff);
E
Eric Sandeen 已提交
1450
	spin_unlock(&log->l_grant_lock);
L
Linus Torvalds 已提交
1451 1452 1453 1454 1455 1456

	/* put cycle number in every block */
	xlog_pack_data(log, iclog, roundoff); 

	/* real byte length */
	if (v2) {
1457 1458
		iclog->ic_header.h_len =
			cpu_to_be32(iclog->ic_offset + roundoff);
L
Linus Torvalds 已提交
1459
	} else {
1460 1461
		iclog->ic_header.h_len =
			cpu_to_be32(iclog->ic_offset);
L
Linus Torvalds 已提交
1462 1463
	}

1464
	bp = iclog->ic_bp;
L
Linus Torvalds 已提交
1465 1466
	ASSERT(XFS_BUF_FSPRIVATE2(bp, unsigned long) == (unsigned long)1);
	XFS_BUF_SET_FSPRIVATE2(bp, (unsigned long)2);
1467
	XFS_BUF_SET_ADDR(bp, BLOCK_LSN(be64_to_cpu(iclog->ic_header.h_lsn)));
L
Linus Torvalds 已提交
1468 1469 1470 1471 1472 1473 1474 1475 1476 1477 1478

	XFS_STATS_ADD(xs_log_blocks, BTOBB(count));

	/* Do we need to split this write into 2 parts? */
	if (XFS_BUF_ADDR(bp) + BTOBB(count) > log->l_logBBsize) {
		split = count - (BBTOB(log->l_logBBsize - XFS_BUF_ADDR(bp)));
		count = BBTOB(log->l_logBBsize - XFS_BUF_ADDR(bp));
		iclog->ic_bwritecnt = 2;	/* split into 2 writes */
	} else {
		iclog->ic_bwritecnt = 1;
	}
1479
	XFS_BUF_SET_COUNT(bp, count);
L
Linus Torvalds 已提交
1480
	XFS_BUF_SET_FSPRIVATE(bp, iclog);	/* save for later */
1481
	XFS_BUF_ZEROFLAGS(bp);
L
Linus Torvalds 已提交
1482 1483 1484
	XFS_BUF_BUSY(bp);
	XFS_BUF_ASYNC(bp);
	/*
1485
	 * Do an ordered write for the log block.
1486
	 * Its unnecessary to flush the first split block in the log wrap case.
L
Linus Torvalds 已提交
1487
	 */
1488
	if (!split && (log->l_mp->m_flags & XFS_MOUNT_BARRIER))
1489
		XFS_BUF_ORDERED(bp);
L
Linus Torvalds 已提交
1490 1491 1492 1493 1494 1495 1496 1497 1498 1499 1500 1501 1502 1503 1504 1505 1506

	ASSERT(XFS_BUF_ADDR(bp) <= log->l_logBBsize-1);
	ASSERT(XFS_BUF_ADDR(bp) + BTOBB(count) <= log->l_logBBsize);

	xlog_verify_iclog(log, iclog, count, B_TRUE);

	/* account for log which doesn't start at block #0 */
	XFS_BUF_SET_ADDR(bp, XFS_BUF_ADDR(bp) + log->l_logBBstart);
	/*
	 * Don't call xfs_bwrite here. We do log-syncs even when the filesystem
	 * is shutting down.
	 */
	XFS_BUF_WRITE(bp);

	if ((error = XFS_bwrite(bp))) {
		xfs_ioerror_alert("xlog_sync", log->l_mp, bp,
				  XFS_BUF_ADDR(bp));
1507
		return error;
L
Linus Torvalds 已提交
1508 1509
	}
	if (split) {
1510
		bp = iclog->ic_log->l_xbuf;
L
Linus Torvalds 已提交
1511 1512 1513 1514 1515 1516 1517
		ASSERT(XFS_BUF_FSPRIVATE2(bp, unsigned long) ==
							(unsigned long)1);
		XFS_BUF_SET_FSPRIVATE2(bp, (unsigned long)2);
		XFS_BUF_SET_ADDR(bp, 0);	     /* logical 0 */
		XFS_BUF_SET_PTR(bp, (xfs_caddr_t)((__psint_t)&(iclog->ic_header)+
					    (__psint_t)count), split);
		XFS_BUF_SET_FSPRIVATE(bp, iclog);
1518
		XFS_BUF_ZEROFLAGS(bp);
L
Linus Torvalds 已提交
1519 1520
		XFS_BUF_BUSY(bp);
		XFS_BUF_ASYNC(bp);
1521 1522
		if (log->l_mp->m_flags & XFS_MOUNT_BARRIER)
			XFS_BUF_ORDERED(bp);
L
Linus Torvalds 已提交
1523 1524 1525 1526 1527 1528 1529
		dptr = XFS_BUF_PTR(bp);
		/*
		 * Bump the cycle numbers at the start of each block
		 * since this part of the buffer is at the start of
		 * a new cycle.  Watch out for the header magic number
		 * case, though.
		 */
1530
		for (i = 0; i < split; i += BBSIZE) {
1531
			be32_add_cpu((__be32 *)dptr, 1);
1532
			if (be32_to_cpu(*(__be32 *)dptr) == XLOG_HEADER_MAGIC_NUM)
1533
				be32_add_cpu((__be32 *)dptr, 1);
L
Linus Torvalds 已提交
1534 1535 1536 1537 1538 1539
			dptr += BBSIZE;
		}

		ASSERT(XFS_BUF_ADDR(bp) <= log->l_logBBsize-1);
		ASSERT(XFS_BUF_ADDR(bp) + BTOBB(count) <= log->l_logBBsize);

1540
		/* account for internal log which doesn't start at block #0 */
L
Linus Torvalds 已提交
1541 1542 1543 1544 1545
		XFS_BUF_SET_ADDR(bp, XFS_BUF_ADDR(bp) + log->l_logBBstart);
		XFS_BUF_WRITE(bp);
		if ((error = XFS_bwrite(bp))) {
			xfs_ioerror_alert("xlog_sync (split)", log->l_mp,
					  bp, XFS_BUF_ADDR(bp));
1546
			return error;
L
Linus Torvalds 已提交
1547 1548
		}
	}
1549
	return 0;
L
Linus Torvalds 已提交
1550 1551 1552 1553
}	/* xlog_sync */


/*
1554
 * Deallocate a log structure
L
Linus Torvalds 已提交
1555
 */
D
David Chinner 已提交
1556
STATIC void
1557
xlog_dealloc_log(xlog_t *log)
L
Linus Torvalds 已提交
1558 1559 1560 1561 1562 1563 1564 1565 1566 1567 1568 1569 1570 1571 1572 1573 1574 1575 1576 1577 1578 1579 1580 1581 1582 1583 1584 1585 1586 1587 1588 1589 1590
{
	xlog_in_core_t	*iclog, *next_iclog;
	int		i;

	iclog = log->l_iclog;
	for (i=0; i<log->l_iclog_bufs; i++) {
		sv_destroy(&iclog->ic_forcesema);
		sv_destroy(&iclog->ic_writesema);
		xfs_buf_free(iclog->ic_bp);
#ifdef XFS_LOG_TRACE
		if (iclog->ic_trace != NULL) {
			ktrace_free(iclog->ic_trace);
		}
#endif
		next_iclog = iclog->ic_next;
		kmem_free(iclog, sizeof(xlog_in_core_t));
		iclog = next_iclog;
	}
	freesema(&log->l_flushsema);
	spinlock_destroy(&log->l_icloglock);
	spinlock_destroy(&log->l_grant_lock);

	xfs_buf_free(log->l_xbuf);
#ifdef XFS_LOG_TRACE
	if (log->l_trace != NULL) {
		ktrace_free(log->l_trace);
	}
	if (log->l_grant_trace != NULL) {
		ktrace_free(log->l_grant_trace);
	}
#endif
	log->l_mp->m_log = NULL;
	kmem_free(log, sizeof(xlog_t));
1591
}	/* xlog_dealloc_log */
L
Linus Torvalds 已提交
1592 1593 1594 1595 1596 1597 1598 1599 1600 1601 1602

/*
 * Update counters atomically now that memcpy is done.
 */
/* ARGSUSED */
static inline void
xlog_state_finish_copy(xlog_t		*log,
		       xlog_in_core_t	*iclog,
		       int		record_cnt,
		       int		copy_bytes)
{
E
Eric Sandeen 已提交
1603
	spin_lock(&log->l_icloglock);
L
Linus Torvalds 已提交
1604

1605
	be32_add_cpu(&iclog->ic_header.h_num_logops, record_cnt);
L
Linus Torvalds 已提交
1606 1607
	iclog->ic_offset += copy_bytes;

E
Eric Sandeen 已提交
1608
	spin_unlock(&log->l_icloglock);
L
Linus Torvalds 已提交
1609 1610 1611 1612 1613
}	/* xlog_state_finish_copy */




1614 1615 1616 1617 1618 1619 1620 1621 1622 1623 1624 1625 1626 1627 1628 1629 1630 1631 1632 1633 1634 1635 1636 1637 1638 1639 1640 1641 1642 1643 1644 1645 1646 1647 1648 1649 1650 1651 1652 1653 1654 1655 1656 1657 1658 1659 1660 1661 1662 1663 1664 1665 1666 1667 1668 1669 1670 1671 1672 1673 1674 1675 1676 1677 1678 1679 1680 1681 1682 1683 1684 1685 1686 1687 1688 1689 1690 1691 1692 1693 1694 1695 1696 1697 1698 1699 1700 1701 1702 1703 1704 1705 1706
/*
 * print out info relating to regions written which consume
 * the reservation
 */
STATIC void
xlog_print_tic_res(xfs_mount_t *mp, xlog_ticket_t *ticket)
{
	uint i;
	uint ophdr_spc = ticket->t_res_num_ophdrs * (uint)sizeof(xlog_op_header_t);

	/* match with XLOG_REG_TYPE_* in xfs_log.h */
	static char *res_type_str[XLOG_REG_TYPE_MAX] = {
	    "bformat",
	    "bchunk",
	    "efi_format",
	    "efd_format",
	    "iformat",
	    "icore",
	    "iext",
	    "ibroot",
	    "ilocal",
	    "iattr_ext",
	    "iattr_broot",
	    "iattr_local",
	    "qformat",
	    "dquot",
	    "quotaoff",
	    "LR header",
	    "unmount",
	    "commit",
	    "trans header"
	};
	static char *trans_type_str[XFS_TRANS_TYPE_MAX] = {
	    "SETATTR_NOT_SIZE",
	    "SETATTR_SIZE",
	    "INACTIVE",
	    "CREATE",
	    "CREATE_TRUNC",
	    "TRUNCATE_FILE",
	    "REMOVE",
	    "LINK",
	    "RENAME",
	    "MKDIR",
	    "RMDIR",
	    "SYMLINK",
	    "SET_DMATTRS",
	    "GROWFS",
	    "STRAT_WRITE",
	    "DIOSTRAT",
	    "WRITE_SYNC",
	    "WRITEID",
	    "ADDAFORK",
	    "ATTRINVAL",
	    "ATRUNCATE",
	    "ATTR_SET",
	    "ATTR_RM",
	    "ATTR_FLAG",
	    "CLEAR_AGI_BUCKET",
	    "QM_SBCHANGE",
	    "DUMMY1",
	    "DUMMY2",
	    "QM_QUOTAOFF",
	    "QM_DQALLOC",
	    "QM_SETQLIM",
	    "QM_DQCLUSTER",
	    "QM_QINOCREATE",
	    "QM_QUOTAOFF_END",
	    "SB_UNIT",
	    "FSYNC_TS",
	    "GROWFSRT_ALLOC",
	    "GROWFSRT_ZERO",
	    "GROWFSRT_FREE",
	    "SWAPEXT"
	};

	xfs_fs_cmn_err(CE_WARN, mp,
			"xfs_log_write: reservation summary:\n"
			"  trans type  = %s (%u)\n"
			"  unit res    = %d bytes\n"
			"  current res = %d bytes\n"
			"  total reg   = %u bytes (o/flow = %u bytes)\n"
			"  ophdrs      = %u (ophdr space = %u bytes)\n"
			"  ophdr + reg = %u bytes\n"
			"  num regions = %u\n",
			((ticket->t_trans_type <= 0 ||
			  ticket->t_trans_type > XFS_TRANS_TYPE_MAX) ?
			  "bad-trans-type" : trans_type_str[ticket->t_trans_type-1]),
			ticket->t_trans_type,
			ticket->t_unit_res,
			ticket->t_curr_res,
			ticket->t_res_arr_sum, ticket->t_res_o_flow,
			ticket->t_res_num_ophdrs, ophdr_spc,
			ticket->t_res_arr_sum + 
1707
			ticket->t_res_o_flow + ophdr_spc,
1708 1709 1710
			ticket->t_res_num);

	for (i = 0; i < ticket->t_res_num; i++) {
1711
		uint r_type = ticket->t_res_arr[i].r_type; 
1712 1713 1714 1715 1716 1717 1718 1719 1720
		cmn_err(CE_WARN,
			    "region[%u]: %s - %u bytes\n",
			    i, 
			    ((r_type <= 0 || r_type > XLOG_REG_TYPE_MAX) ?
			    "bad-rtype" : res_type_str[r_type-1]),
			    ticket->t_res_arr[i].r_len);
	}
}

L
Linus Torvalds 已提交
1721 1722 1723 1724 1725 1726 1727 1728 1729 1730 1731 1732 1733 1734 1735 1736 1737 1738 1739 1740 1741 1742 1743 1744 1745 1746 1747 1748 1749 1750 1751 1752 1753 1754 1755 1756 1757 1758 1759 1760
/*
 * Write some region out to in-core log
 *
 * This will be called when writing externally provided regions or when
 * writing out a commit record for a given transaction.
 *
 * General algorithm:
 *	1. Find total length of this write.  This may include adding to the
 *		lengths passed in.
 *	2. Check whether we violate the tickets reservation.
 *	3. While writing to this iclog
 *	    A. Reserve as much space in this iclog as can get
 *	    B. If this is first write, save away start lsn
 *	    C. While writing this region:
 *		1. If first write of transaction, write start record
 *		2. Write log operation header (header per region)
 *		3. Find out if we can fit entire region into this iclog
 *		4. Potentially, verify destination memcpy ptr
 *		5. Memcpy (partial) region
 *		6. If partial copy, release iclog; otherwise, continue
 *			copying more regions into current iclog
 *	4. Mark want sync bit (in simulation mode)
 *	5. Release iclog for potential flush to on-disk log.
 *
 * ERRORS:
 * 1.	Panic if reservation is overrun.  This should never happen since
 *	reservation amounts are generated internal to the filesystem.
 * NOTES:
 * 1. Tickets are single threaded data structures.
 * 2. The XLOG_END_TRANS & XLOG_CONTINUE_TRANS flags are passed down to the
 *	syncing routine.  When a single log_write region needs to span
 *	multiple in-core logs, the XLOG_CONTINUE_TRANS bit should be set
 *	on all log operation writes which don't contain the end of the
 *	region.  The XLOG_END_TRANS bit is used for the in-core log
 *	operation which contains the end of the continued log_write region.
 * 3. When xlog_state_get_iclog_space() grabs the rest of the current iclog,
 *	we don't really know exactly how much space will be used.  As a result,
 *	we don't update ic_offset until the end when we know exactly how many
 *	bytes have been written out.
 */
D
David Chinner 已提交
1761
STATIC int
L
Linus Torvalds 已提交
1762 1763 1764 1765 1766 1767 1768 1769
xlog_write(xfs_mount_t *	mp,
	   xfs_log_iovec_t	reg[],
	   int			nentries,
	   xfs_log_ticket_t	tic,
	   xfs_lsn_t		*start_lsn,
	   xlog_in_core_t	**commit_iclog,
	   uint			flags)
{
1770
    xlog_t	     *log = mp->m_log;
L
Linus Torvalds 已提交
1771
    xlog_ticket_t    *ticket = (xlog_ticket_t *)tic;
1772
    xlog_in_core_t   *iclog = NULL;  /* ptr to current in-core log */
L
Linus Torvalds 已提交
1773 1774 1775 1776 1777 1778 1779 1780 1781 1782 1783 1784 1785 1786 1787 1788 1789 1790 1791 1792 1793
    xlog_op_header_t *logop_head;    /* ptr to log operation header */
    __psint_t	     ptr;	     /* copy address into data region */
    int		     len;	     /* # xlog_write() bytes 2 still copy */
    int		     index;	     /* region index currently copying */
    int		     log_offset;     /* offset (from 0) into data region */
    int		     start_rec_copy; /* # bytes to copy for start record */
    int		     partial_copy;   /* did we split a region? */
    int		     partial_copy_len;/* # bytes copied if split region */
    int		     need_copy;	     /* # bytes need to memcpy this region */
    int		     copy_len;	     /* # bytes actually memcpy'ing */
    int		     copy_off;	     /* # bytes from entry start */
    int		     contwr;	     /* continued write of in-core log? */
    int		     error;
    int		     record_cnt = 0, data_cnt = 0;

    partial_copy_len = partial_copy = 0;

    /* Calculate potential maximum space.  Each region gets its own
     * xlog_op_header_t and may need to be double word aligned.
     */
    len = 0;
1794
    if (ticket->t_flags & XLOG_TIC_INITED) {    /* acct for start rec of xact */
L
Linus Torvalds 已提交
1795
	len += sizeof(xlog_op_header_t);
1796
	ticket->t_res_num_ophdrs++;
1797
    }
L
Linus Torvalds 已提交
1798 1799 1800

    for (index = 0; index < nentries; index++) {
	len += sizeof(xlog_op_header_t);	    /* each region gets >= 1 */
1801
	ticket->t_res_num_ophdrs++;
L
Linus Torvalds 已提交
1802
	len += reg[index].i_len;
1803
	xlog_tic_add_region(ticket, reg[index].i_len, reg[index].i_type);
L
Linus Torvalds 已提交
1804 1805 1806 1807
    }
    contwr = *start_lsn = 0;

    if (ticket->t_curr_res < len) {
1808
	xlog_print_tic_res(mp, ticket);
L
Linus Torvalds 已提交
1809 1810 1811 1812 1813 1814 1815 1816
#ifdef DEBUG
	xlog_panic(
		"xfs_log_write: reservation ran out. Need to up reservation");
#else
	/* Customer configurable panic */
	xfs_cmn_err(XFS_PTAG_LOGRES, CE_ALERT, mp,
		"xfs_log_write: reservation ran out. Need to up reservation");
	/* If we did not panic, shutdown the filesystem */
1817
	xfs_force_shutdown(mp, SHUTDOWN_CORRUPT_INCORE);
L
Linus Torvalds 已提交
1818 1819 1820 1821 1822 1823 1824
#endif
    } else
	ticket->t_curr_res -= len;

    for (index = 0; index < nentries; ) {
	if ((error = xlog_state_get_iclog_space(log, len, &iclog, ticket,
					       &contwr, &log_offset)))
1825
		return error;
L
Linus Torvalds 已提交
1826 1827 1828 1829 1830 1831

	ASSERT(log_offset <= iclog->ic_size - 1);
	ptr = (__psint_t) ((char *)iclog->ic_datap+log_offset);

	/* start_lsn is the first lsn written to. That's all we need. */
	if (! *start_lsn)
1832
	    *start_lsn = be64_to_cpu(iclog->ic_header.h_lsn);
L
Linus Torvalds 已提交
1833 1834 1835 1836 1837 1838 1839 1840 1841 1842 1843 1844 1845 1846 1847

	/* This loop writes out as many regions as can fit in the amount
	 * of space which was allocated by xlog_state_get_iclog_space().
	 */
	while (index < nentries) {
	    ASSERT(reg[index].i_len % sizeof(__int32_t) == 0);
	    ASSERT((__psint_t)ptr % sizeof(__int32_t) == 0);
	    start_rec_copy = 0;

	    /* If first write for transaction, insert start record.
	     * We can't be trying to commit if we are inited.  We can't
	     * have any "partial_copy" if we are inited.
	     */
	    if (ticket->t_flags & XLOG_TIC_INITED) {
		logop_head		= (xlog_op_header_t *)ptr;
1848
		logop_head->oh_tid	= cpu_to_be32(ticket->t_tid);
L
Linus Torvalds 已提交
1849 1850 1851 1852 1853 1854 1855 1856 1857 1858 1859 1860 1861
		logop_head->oh_clientid = ticket->t_clientid;
		logop_head->oh_len	= 0;
		logop_head->oh_flags    = XLOG_START_TRANS;
		logop_head->oh_res2	= 0;
		ticket->t_flags		&= ~XLOG_TIC_INITED;	/* clear bit */
		record_cnt++;

		start_rec_copy = sizeof(xlog_op_header_t);
		xlog_write_adv_cnt(ptr, len, log_offset, start_rec_copy);
	    }

	    /* Copy log operation header directly into data section */
	    logop_head			= (xlog_op_header_t *)ptr;
1862
	    logop_head->oh_tid		= cpu_to_be32(ticket->t_tid);
L
Linus Torvalds 已提交
1863 1864 1865 1866 1867 1868 1869 1870 1871 1872 1873 1874 1875 1876 1877 1878 1879 1880 1881 1882 1883 1884 1885 1886 1887 1888 1889 1890 1891 1892 1893 1894 1895 1896
	    logop_head->oh_clientid	= ticket->t_clientid;
	    logop_head->oh_res2		= 0;

	    /* header copied directly */
	    xlog_write_adv_cnt(ptr, len, log_offset, sizeof(xlog_op_header_t));

	    /* are we copying a commit or unmount record? */
	    logop_head->oh_flags = flags;

	    /*
	     * We've seen logs corrupted with bad transaction client
	     * ids.  This makes sure that XFS doesn't generate them on.
	     * Turn this into an EIO and shut down the filesystem.
	     */
	    switch (logop_head->oh_clientid)  {
	    case XFS_TRANSACTION:
	    case XFS_VOLUME:
	    case XFS_LOG:
		break;
	    default:
		xfs_fs_cmn_err(CE_WARN, mp,
		    "Bad XFS transaction clientid 0x%x in ticket 0x%p",
		    logop_head->oh_clientid, tic);
		return XFS_ERROR(EIO);
	    }

	    /* Partial write last time? => (partial_copy != 0)
	     * need_copy is the amount we'd like to copy if everything could
	     * fit in the current memcpy.
	     */
	    need_copy =	reg[index].i_len - partial_copy_len;

	    copy_off = partial_copy_len;
	    if (need_copy <= iclog->ic_size - log_offset) { /*complete write */
1897 1898
	        copy_len = need_copy;
		logop_head->oh_len = cpu_to_be32(copy_len);
L
Linus Torvalds 已提交
1899 1900 1901 1902 1903
		if (partial_copy)
		    logop_head->oh_flags|= (XLOG_END_TRANS|XLOG_WAS_CONT_TRANS);
		partial_copy_len = partial_copy = 0;
	    } else {					    /* partial write */
		copy_len = iclog->ic_size - log_offset;
1904
		logop_head->oh_len = cpu_to_be32(copy_len);
L
Linus Torvalds 已提交
1905 1906 1907 1908 1909 1910 1911 1912
		logop_head->oh_flags |= XLOG_CONTINUE_TRANS;
		if (partial_copy)
			logop_head->oh_flags |= XLOG_WAS_CONT_TRANS;
		partial_copy_len += copy_len;
		partial_copy++;
		len += sizeof(xlog_op_header_t); /* from splitting of region */
		/* account for new log op header */
		ticket->t_curr_res -= sizeof(xlog_op_header_t);
1913
		ticket->t_res_num_ophdrs++;
L
Linus Torvalds 已提交
1914 1915 1916 1917 1918 1919 1920 1921 1922 1923 1924 1925 1926 1927 1928 1929 1930
	    }
	    xlog_verify_dest_ptr(log, ptr);

	    /* copy region */
	    ASSERT(copy_len >= 0);
	    memcpy((xfs_caddr_t)ptr, reg[index].i_addr + copy_off, copy_len);
	    xlog_write_adv_cnt(ptr, len, log_offset, copy_len);

	    /* make copy_len total bytes copied, including headers */
	    copy_len += start_rec_copy + sizeof(xlog_op_header_t);
	    record_cnt++;
	    data_cnt += contwr ? copy_len : 0;
	    if (partial_copy) {			/* copied partial region */
		    /* already marked WANT_SYNC by xlog_state_get_iclog_space */
		    xlog_state_finish_copy(log, iclog, record_cnt, data_cnt);
		    record_cnt = data_cnt = 0;
		    if ((error = xlog_state_release_iclog(log, iclog)))
1931
			    return error;
L
Linus Torvalds 已提交
1932 1933 1934 1935 1936 1937 1938 1939 1940 1941 1942 1943 1944
		    break;			/* don't increment index */
	    } else {				/* copied entire region */
		index++;
		partial_copy_len = partial_copy = 0;

		if (iclog->ic_size - log_offset <= sizeof(xlog_op_header_t)) {
		    xlog_state_finish_copy(log, iclog, record_cnt, data_cnt);
		    record_cnt = data_cnt = 0;
		    xlog_state_want_sync(log, iclog);
		    if (commit_iclog) {
			ASSERT(flags & XLOG_COMMIT_TRANS);
			*commit_iclog = iclog;
		    } else if ((error = xlog_state_release_iclog(log, iclog)))
1945
			   return error;
L
Linus Torvalds 已提交
1946 1947 1948 1949 1950 1951 1952 1953 1954 1955 1956 1957 1958 1959 1960 1961
		    if (index == nentries)
			    return 0;		/* we are done */
		    else
			    break;
		}
	    } /* if (partial_copy) */
	} /* while (index < nentries) */
    } /* for (index = 0; index < nentries; ) */
    ASSERT(len == 0);

    xlog_state_finish_copy(log, iclog, record_cnt, data_cnt);
    if (commit_iclog) {
	ASSERT(flags & XLOG_COMMIT_TRANS);
	*commit_iclog = iclog;
	return 0;
    }
1962
    return xlog_state_release_iclog(log, iclog);
L
Linus Torvalds 已提交
1963 1964 1965 1966 1967 1968 1969 1970 1971 1972 1973 1974 1975 1976 1977 1978 1979 1980
}	/* xlog_write */


/*****************************************************************************
 *
 *		State Machine functions
 *
 *****************************************************************************
 */

/* Clean iclogs starting from the head.  This ordering must be
 * maintained, so an iclog doesn't become ACTIVE beyond one that
 * is SYNCING.  This is also required to maintain the notion that we use
 * a counting semaphore to hold off would be writers to the log when every
 * iclog is trying to sync to disk.
 *
 * State Change: DIRTY -> ACTIVE
 */
1981
STATIC void
L
Linus Torvalds 已提交
1982 1983 1984 1985 1986 1987 1988 1989 1990 1991
xlog_state_clean_log(xlog_t *log)
{
	xlog_in_core_t	*iclog;
	int changed = 0;

	iclog = log->l_iclog;
	do {
		if (iclog->ic_state == XLOG_STATE_DIRTY) {
			iclog->ic_state	= XLOG_STATE_ACTIVE;
			iclog->ic_offset       = 0;
1992
			ASSERT(iclog->ic_callback == NULL);
L
Linus Torvalds 已提交
1993 1994 1995 1996 1997 1998 1999 2000 2001
			/*
			 * If the number of ops in this iclog indicate it just
			 * contains the dummy transaction, we can
			 * change state into IDLE (the second time around).
			 * Otherwise we should change the state into
			 * NEED a dummy.
			 * We don't need to cover the dummy.
			 */
			if (!changed &&
2002 2003
			   (be32_to_cpu(iclog->ic_header.h_num_logops) ==
			   		XLOG_COVER_OPS)) {
L
Linus Torvalds 已提交
2004 2005 2006 2007 2008 2009 2010 2011 2012 2013 2014 2015 2016 2017 2018 2019 2020 2021 2022 2023 2024 2025 2026 2027 2028 2029 2030 2031 2032 2033 2034 2035 2036 2037 2038 2039 2040 2041 2042 2043 2044 2045 2046 2047 2048 2049 2050 2051 2052 2053 2054 2055 2056 2057 2058 2059 2060 2061 2062 2063 2064 2065 2066 2067 2068 2069 2070
				changed = 1;
			} else {
				/*
				 * We have two dirty iclogs so start over
				 * This could also be num of ops indicates
				 * this is not the dummy going out.
				 */
				changed = 2;
			}
			iclog->ic_header.h_num_logops = 0;
			memset(iclog->ic_header.h_cycle_data, 0,
			      sizeof(iclog->ic_header.h_cycle_data));
			iclog->ic_header.h_lsn = 0;
		} else if (iclog->ic_state == XLOG_STATE_ACTIVE)
			/* do nothing */;
		else
			break;	/* stop cleaning */
		iclog = iclog->ic_next;
	} while (iclog != log->l_iclog);

	/* log is locked when we are called */
	/*
	 * Change state for the dummy log recording.
	 * We usually go to NEED. But we go to NEED2 if the changed indicates
	 * we are done writing the dummy record.
	 * If we are done with the second dummy recored (DONE2), then
	 * we go to IDLE.
	 */
	if (changed) {
		switch (log->l_covered_state) {
		case XLOG_STATE_COVER_IDLE:
		case XLOG_STATE_COVER_NEED:
		case XLOG_STATE_COVER_NEED2:
			log->l_covered_state = XLOG_STATE_COVER_NEED;
			break;

		case XLOG_STATE_COVER_DONE:
			if (changed == 1)
				log->l_covered_state = XLOG_STATE_COVER_NEED2;
			else
				log->l_covered_state = XLOG_STATE_COVER_NEED;
			break;

		case XLOG_STATE_COVER_DONE2:
			if (changed == 1)
				log->l_covered_state = XLOG_STATE_COVER_IDLE;
			else
				log->l_covered_state = XLOG_STATE_COVER_NEED;
			break;

		default:
			ASSERT(0);
		}
	}
}	/* xlog_state_clean_log */

STATIC xfs_lsn_t
xlog_get_lowest_lsn(
	xlog_t		*log)
{
	xlog_in_core_t  *lsn_log;
	xfs_lsn_t	lowest_lsn, lsn;

	lsn_log = log->l_iclog;
	lowest_lsn = 0;
	do {
	    if (!(lsn_log->ic_state & (XLOG_STATE_ACTIVE|XLOG_STATE_DIRTY))) {
2071
		lsn = be64_to_cpu(lsn_log->ic_header.h_lsn);
L
Linus Torvalds 已提交
2072 2073 2074 2075 2076 2077 2078
		if ((lsn && !lowest_lsn) ||
		    (XFS_LSN_CMP(lsn, lowest_lsn) < 0)) {
			lowest_lsn = lsn;
		}
	    }
	    lsn_log = lsn_log->ic_next;
	} while (lsn_log != log->l_iclog);
2079
	return lowest_lsn;
L
Linus Torvalds 已提交
2080 2081 2082 2083 2084 2085 2086 2087 2088 2089 2090 2091 2092 2093 2094 2095 2096 2097 2098 2099 2100
}


STATIC void
xlog_state_do_callback(
	xlog_t		*log,
	int		aborted,
	xlog_in_core_t	*ciclog)
{
	xlog_in_core_t	   *iclog;
	xlog_in_core_t	   *first_iclog;	/* used to know when we've
						 * processed all iclogs once */
	xfs_log_callback_t *cb, *cb_next;
	int		   flushcnt = 0;
	xfs_lsn_t	   lowest_lsn;
	int		   ioerrors;	/* counter: iclogs with errors */
	int		   loopdidcallbacks; /* flag: inner loop did callbacks*/
	int		   funcdidcallbacks; /* flag: function did callbacks */
	int		   repeats;	/* for issuing console warnings if
					 * looping too many times */

E
Eric Sandeen 已提交
2101
	spin_lock(&log->l_icloglock);
L
Linus Torvalds 已提交
2102 2103 2104 2105 2106 2107 2108 2109 2110 2111 2112 2113 2114 2115 2116 2117 2118 2119 2120 2121 2122 2123 2124 2125 2126 2127 2128 2129 2130 2131 2132 2133 2134 2135 2136 2137 2138 2139 2140 2141 2142 2143 2144 2145
	first_iclog = iclog = log->l_iclog;
	ioerrors = 0;
	funcdidcallbacks = 0;
	repeats = 0;

	do {
		/*
		 * Scan all iclogs starting with the one pointed to by the
		 * log.  Reset this starting point each time the log is
		 * unlocked (during callbacks).
		 *
		 * Keep looping through iclogs until one full pass is made
		 * without running any callbacks.
		 */
		first_iclog = log->l_iclog;
		iclog = log->l_iclog;
		loopdidcallbacks = 0;
		repeats++;

		do {

			/* skip all iclogs in the ACTIVE & DIRTY states */
			if (iclog->ic_state &
			    (XLOG_STATE_ACTIVE|XLOG_STATE_DIRTY)) {
				iclog = iclog->ic_next;
				continue;
			}

			/*
			 * Between marking a filesystem SHUTDOWN and stopping
			 * the log, we do flush all iclogs to disk (if there
			 * wasn't a log I/O error). So, we do want things to
			 * go smoothly in case of just a SHUTDOWN  w/o a
			 * LOG_IO_ERROR.
			 */
			if (!(iclog->ic_state & XLOG_STATE_IOERROR)) {
				/*
				 * Can only perform callbacks in order.  Since
				 * this iclog is not in the DONE_SYNC/
				 * DO_CALLBACK state, we skip the rest and
				 * just try to clean up.  If we set our iclog
				 * to DO_CALLBACK, we will not process it when
				 * we retry since a previous iclog is in the
				 * CALLBACK and the state cannot change since
E
Eric Sandeen 已提交
2146
				 * we are holding the l_icloglock.
L
Linus Torvalds 已提交
2147 2148 2149 2150 2151 2152 2153 2154 2155 2156 2157 2158 2159 2160 2161 2162 2163 2164 2165 2166 2167 2168 2169 2170 2171
				 */
				if (!(iclog->ic_state &
					(XLOG_STATE_DONE_SYNC |
						 XLOG_STATE_DO_CALLBACK))) {
					if (ciclog && (ciclog->ic_state ==
							XLOG_STATE_DONE_SYNC)) {
						ciclog->ic_state = XLOG_STATE_DO_CALLBACK;
					}
					break;
				}
				/*
				 * We now have an iclog that is in either the
				 * DO_CALLBACK or DONE_SYNC states. The other
				 * states (WANT_SYNC, SYNCING, or CALLBACK were
				 * caught by the above if and are going to
				 * clean (i.e. we aren't doing their callbacks)
				 * see the above if.
				 */

				/*
				 * We will do one more check here to see if we
				 * have chased our tail around.
				 */

				lowest_lsn = xlog_get_lowest_lsn(log);
2172 2173 2174
				if (lowest_lsn &&
				    XFS_LSN_CMP(lowest_lsn,
				    		be64_to_cpu(iclog->ic_header.h_lsn)) < 0) {
L
Linus Torvalds 已提交
2175 2176 2177 2178 2179 2180 2181
					iclog = iclog->ic_next;
					continue; /* Leave this iclog for
						   * another thread */
				}

				iclog->ic_state = XLOG_STATE_CALLBACK;

E
Eric Sandeen 已提交
2182
				spin_unlock(&log->l_icloglock);
L
Linus Torvalds 已提交
2183 2184

				/* l_last_sync_lsn field protected by
E
Eric Sandeen 已提交
2185
				 * l_grant_lock. Don't worry about iclog's lsn.
L
Linus Torvalds 已提交
2186 2187
				 * No one else can be here except us.
				 */
E
Eric Sandeen 已提交
2188
				spin_lock(&log->l_grant_lock);
2189 2190 2191 2192
				ASSERT(XFS_LSN_CMP(log->l_last_sync_lsn,
				       be64_to_cpu(iclog->ic_header.h_lsn)) <= 0);
				log->l_last_sync_lsn =
					be64_to_cpu(iclog->ic_header.h_lsn);
E
Eric Sandeen 已提交
2193
				spin_unlock(&log->l_grant_lock);
L
Linus Torvalds 已提交
2194 2195

			} else {
2196
				spin_unlock(&log->l_icloglock);
L
Linus Torvalds 已提交
2197 2198 2199
				ioerrors++;
			}

2200 2201 2202 2203 2204 2205 2206 2207 2208
			/*
			 * Keep processing entries in the callback list until
			 * we come around and it is empty.  We need to
			 * atomically see that the list is empty and change the
			 * state to DIRTY so that we don't miss any more
			 * callbacks being added.
			 */
			spin_lock(&iclog->ic_callback_lock);
			cb = iclog->ic_callback;
2209
			while (cb) {
L
Linus Torvalds 已提交
2210 2211
				iclog->ic_callback_tail = &(iclog->ic_callback);
				iclog->ic_callback = NULL;
2212
				spin_unlock(&iclog->ic_callback_lock);
L
Linus Torvalds 已提交
2213 2214

				/* perform callbacks in the order given */
2215
				for (; cb; cb = cb_next) {
L
Linus Torvalds 已提交
2216 2217 2218
					cb_next = cb->cb_next;
					cb->cb_func(cb->cb_arg, aborted);
				}
2219
				spin_lock(&iclog->ic_callback_lock);
L
Linus Torvalds 已提交
2220 2221 2222 2223 2224 2225
				cb = iclog->ic_callback;
			}

			loopdidcallbacks++;
			funcdidcallbacks++;

2226
			spin_lock(&log->l_icloglock);
2227
			ASSERT(iclog->ic_callback == NULL);
2228
			spin_unlock(&iclog->ic_callback_lock);
L
Linus Torvalds 已提交
2229 2230 2231 2232 2233 2234 2235 2236 2237 2238 2239 2240 2241 2242
			if (!(iclog->ic_state & XLOG_STATE_IOERROR))
				iclog->ic_state = XLOG_STATE_DIRTY;

			/*
			 * Transition from DIRTY to ACTIVE if applicable.
			 * NOP if STATE_IOERROR.
			 */
			xlog_state_clean_log(log);

			/* wake up threads waiting in xfs_log_force() */
			sv_broadcast(&iclog->ic_forcesema);

			iclog = iclog->ic_next;
		} while (first_iclog != iclog);
2243 2244 2245 2246

		if (repeats > 5000) {
			flushcnt += repeats;
			repeats = 0;
L
Linus Torvalds 已提交
2247
			xfs_fs_cmn_err(CE_WARN, log->l_mp,
2248
				"%s: possible infinite loop (%d iterations)",
2249
				__func__, flushcnt);
L
Linus Torvalds 已提交
2250 2251 2252 2253 2254 2255 2256 2257 2258 2259 2260 2261 2262 2263 2264 2265 2266 2267
		}
	} while (!ioerrors && loopdidcallbacks);

	/*
	 * make one last gasp attempt to see if iclogs are being left in
	 * limbo..
	 */
#ifdef DEBUG
	if (funcdidcallbacks) {
		first_iclog = iclog = log->l_iclog;
		do {
			ASSERT(iclog->ic_state != XLOG_STATE_DO_CALLBACK);
			/*
			 * Terminate the loop if iclogs are found in states
			 * which will cause other threads to clean up iclogs.
			 *
			 * SYNCING - i/o completion will go through logs
			 * DONE_SYNC - interrupt thread should be waiting for
E
Eric Sandeen 已提交
2268
			 *              l_icloglock
L
Linus Torvalds 已提交
2269 2270 2271 2272 2273 2274 2275 2276 2277 2278 2279 2280
			 * IOERROR - give up hope all ye who enter here
			 */
			if (iclog->ic_state == XLOG_STATE_WANT_SYNC ||
			    iclog->ic_state == XLOG_STATE_SYNCING ||
			    iclog->ic_state == XLOG_STATE_DONE_SYNC ||
			    iclog->ic_state == XLOG_STATE_IOERROR )
				break;
			iclog = iclog->ic_next;
		} while (first_iclog != iclog);
	}
#endif

2281
	flushcnt = 0;
L
Linus Torvalds 已提交
2282 2283 2284 2285
	if (log->l_iclog->ic_state & (XLOG_STATE_ACTIVE|XLOG_STATE_IOERROR)) {
		flushcnt = log->l_flushcnt;
		log->l_flushcnt = 0;
	}
E
Eric Sandeen 已提交
2286
	spin_unlock(&log->l_icloglock);
L
Linus Torvalds 已提交
2287 2288 2289 2290 2291 2292 2293 2294 2295 2296 2297 2298 2299 2300 2301 2302 2303 2304 2305
	while (flushcnt--)
		vsema(&log->l_flushsema);
}	/* xlog_state_do_callback */


/*
 * Finish transitioning this iclog to the dirty state.
 *
 * Make sure that we completely execute this routine only when this is
 * the last call to the iclog.  There is a good chance that iclog flushes,
 * when we reach the end of the physical log, get turned into 2 separate
 * calls to bwrite.  Hence, one iclog flush could generate two calls to this
 * routine.  By using the reference count bwritecnt, we guarantee that only
 * the second completion goes through.
 *
 * Callbacks could take time, so they are done outside the scope of the
 * global state machine log lock.  Assume that the calls to cvsema won't
 * take a long time.  At least we know it won't sleep.
 */
D
David Chinner 已提交
2306
STATIC void
L
Linus Torvalds 已提交
2307 2308 2309 2310 2311 2312
xlog_state_done_syncing(
	xlog_in_core_t	*iclog,
	int		aborted)
{
	xlog_t		   *log = iclog->ic_log;

E
Eric Sandeen 已提交
2313
	spin_lock(&log->l_icloglock);
L
Linus Torvalds 已提交
2314 2315 2316

	ASSERT(iclog->ic_state == XLOG_STATE_SYNCING ||
	       iclog->ic_state == XLOG_STATE_IOERROR);
2317
	ASSERT(atomic_read(&iclog->ic_refcnt) == 0);
L
Linus Torvalds 已提交
2318 2319 2320 2321 2322 2323 2324 2325 2326 2327 2328
	ASSERT(iclog->ic_bwritecnt == 1 || iclog->ic_bwritecnt == 2);


	/*
	 * If we got an error, either on the first buffer, or in the case of
	 * split log writes, on the second, we mark ALL iclogs STATE_IOERROR,
	 * and none should ever be attempted to be written to disk
	 * again.
	 */
	if (iclog->ic_state != XLOG_STATE_IOERROR) {
		if (--iclog->ic_bwritecnt == 1) {
E
Eric Sandeen 已提交
2329
			spin_unlock(&log->l_icloglock);
L
Linus Torvalds 已提交
2330 2331 2332 2333 2334 2335 2336 2337 2338 2339 2340
			return;
		}
		iclog->ic_state = XLOG_STATE_DONE_SYNC;
	}

	/*
	 * Someone could be sleeping prior to writing out the next
	 * iclog buffer, we wake them all, one will get to do the
	 * I/O, the others get to wait for the result.
	 */
	sv_broadcast(&iclog->ic_writesema);
E
Eric Sandeen 已提交
2341
	spin_unlock(&log->l_icloglock);
L
Linus Torvalds 已提交
2342 2343 2344 2345 2346 2347 2348 2349 2350 2351 2352 2353 2354 2355 2356 2357 2358 2359 2360 2361 2362 2363 2364 2365
	xlog_state_do_callback(log, aborted, iclog);	/* also cleans log */
}	/* xlog_state_done_syncing */


/*
 * If the head of the in-core log ring is not (ACTIVE or DIRTY), then we must
 * sleep.  The flush semaphore is set to the number of in-core buffers and
 * decremented around disk syncing.  Therefore, if all buffers are syncing,
 * this semaphore will cause new writes to sleep until a sync completes.
 * Otherwise, this code just does p() followed by v().  This approximates
 * a sleep/wakeup except we can't race.
 *
 * The in-core logs are used in a circular fashion. They are not used
 * out-of-order even when an iclog past the head is free.
 *
 * return:
 *	* log_offset where xlog_write() can start writing into the in-core
 *		log's data space.
 *	* in-core log pointer to which xlog_write() should write.
 *	* boolean indicating this is a continued write to an in-core log.
 *		If this is the last write, then the in-core log's offset field
 *		needs to be incremented, depending on the amount of data which
 *		is copied.
 */
D
David Chinner 已提交
2366
STATIC int
L
Linus Torvalds 已提交
2367 2368 2369 2370 2371 2372 2373 2374 2375 2376 2377 2378 2379
xlog_state_get_iclog_space(xlog_t	  *log,
			   int		  len,
			   xlog_in_core_t **iclogp,
			   xlog_ticket_t  *ticket,
			   int		  *continued_write,
			   int		  *logoffsetp)
{
	int		  log_offset;
	xlog_rec_header_t *head;
	xlog_in_core_t	  *iclog;
	int		  error;

restart:
E
Eric Sandeen 已提交
2380
	spin_lock(&log->l_icloglock);
L
Linus Torvalds 已提交
2381
	if (XLOG_FORCED_SHUTDOWN(log)) {
E
Eric Sandeen 已提交
2382
		spin_unlock(&log->l_icloglock);
L
Linus Torvalds 已提交
2383 2384 2385 2386 2387 2388
		return XFS_ERROR(EIO);
	}

	iclog = log->l_iclog;
	if (! (iclog->ic_state == XLOG_STATE_ACTIVE)) {
		log->l_flushcnt++;
E
Eric Sandeen 已提交
2389
		spin_unlock(&log->l_icloglock);
L
Linus Torvalds 已提交
2390 2391 2392 2393 2394 2395 2396 2397 2398
		xlog_trace_iclog(iclog, XLOG_TRACE_SLEEP_FLUSH);
		XFS_STATS_INC(xs_log_noiclogs);
		/* Ensure that log writes happen */
		psema(&log->l_flushsema, PINOD);
		goto restart;
	}
	ASSERT(iclog->ic_state == XLOG_STATE_ACTIVE);
	head = &iclog->ic_header;

2399
	atomic_inc(&iclog->ic_refcnt);	/* prevents sync */
L
Linus Torvalds 已提交
2400 2401 2402 2403 2404 2405 2406 2407 2408
	log_offset = iclog->ic_offset;

	/* On the 1st write to an iclog, figure out lsn.  This works
	 * if iclogs marked XLOG_STATE_WANT_SYNC always write out what they are
	 * committing to.  If the offset is set, that's how many blocks
	 * must be written.
	 */
	if (log_offset == 0) {
		ticket->t_curr_res -= log->l_iclog_hsize;
2409
		xlog_tic_add_region(ticket,
2410 2411
				    log->l_iclog_hsize,
				    XLOG_REG_TYPE_LRHEADER);
2412 2413
		head->h_cycle = cpu_to_be32(log->l_curr_cycle);
		head->h_lsn = cpu_to_be64(
2414
			xlog_assign_lsn(log->l_curr_cycle, log->l_curr_block));
L
Linus Torvalds 已提交
2415 2416 2417 2418 2419 2420 2421 2422 2423 2424 2425 2426 2427 2428 2429 2430
		ASSERT(log->l_curr_block >= 0);
	}

	/* If there is enough room to write everything, then do it.  Otherwise,
	 * claim the rest of the region and make sure the XLOG_STATE_WANT_SYNC
	 * bit is on, so this will get flushed out.  Don't update ic_offset
	 * until you know exactly how many bytes get copied.  Therefore, wait
	 * until later to update ic_offset.
	 *
	 * xlog_write() algorithm assumes that at least 2 xlog_op_header_t's
	 * can fit into remaining data section.
	 */
	if (iclog->ic_size - iclog->ic_offset < 2*sizeof(xlog_op_header_t)) {
		xlog_state_switch_iclogs(log, iclog, iclog->ic_size);

		/* If I'm the only one writing to this iclog, sync it to disk */
2431
		if (atomic_read(&iclog->ic_refcnt) == 1) {
E
Eric Sandeen 已提交
2432
			spin_unlock(&log->l_icloglock);
L
Linus Torvalds 已提交
2433
			if ((error = xlog_state_release_iclog(log, iclog)))
2434
				return error;
L
Linus Torvalds 已提交
2435
		} else {
2436
			atomic_dec(&iclog->ic_refcnt);
E
Eric Sandeen 已提交
2437
			spin_unlock(&log->l_icloglock);
L
Linus Torvalds 已提交
2438 2439 2440 2441 2442 2443 2444 2445 2446 2447 2448 2449 2450 2451 2452 2453 2454 2455 2456 2457
		}
		goto restart;
	}

	/* Do we have enough room to write the full amount in the remainder
	 * of this iclog?  Or must we continue a write on the next iclog and
	 * mark this iclog as completely taken?  In the case where we switch
	 * iclogs (to mark it taken), this particular iclog will release/sync
	 * to disk in xlog_write().
	 */
	if (len <= iclog->ic_size - iclog->ic_offset) {
		*continued_write = 0;
		iclog->ic_offset += len;
	} else {
		*continued_write = 1;
		xlog_state_switch_iclogs(log, iclog, iclog->ic_size);
	}
	*iclogp = iclog;

	ASSERT(iclog->ic_offset <= iclog->ic_size);
E
Eric Sandeen 已提交
2458
	spin_unlock(&log->l_icloglock);
L
Linus Torvalds 已提交
2459 2460 2461 2462 2463 2464 2465 2466 2467 2468 2469 2470 2471 2472 2473 2474 2475 2476 2477 2478 2479 2480 2481 2482 2483 2484 2485 2486

	*logoffsetp = log_offset;
	return 0;
}	/* xlog_state_get_iclog_space */

/*
 * Atomically get the log space required for a log ticket.
 *
 * Once a ticket gets put onto the reserveq, it will only return after
 * the needed reservation is satisfied.
 */
STATIC int
xlog_grant_log_space(xlog_t	   *log,
		     xlog_ticket_t *tic)
{
	int		 free_bytes;
	int		 need_bytes;
#ifdef DEBUG
	xfs_lsn_t	 tail_lsn;
#endif


#ifdef DEBUG
	if (log->l_flags & XLOG_ACTIVE_RECOVERY)
		panic("grant Recovery problem");
#endif

	/* Is there space or do we need to sleep? */
E
Eric Sandeen 已提交
2487
	spin_lock(&log->l_grant_lock);
L
Linus Torvalds 已提交
2488 2489 2490 2491
	xlog_trace_loggrant(log, tic, "xlog_grant_log_space: enter");

	/* something is already sleeping; insert new transaction at end */
	if (log->l_reserve_headq) {
2492
		xlog_ins_ticketq(&log->l_reserve_headq, tic);
L
Linus Torvalds 已提交
2493 2494 2495 2496 2497 2498 2499 2500 2501 2502 2503 2504 2505 2506 2507 2508 2509
		xlog_trace_loggrant(log, tic,
				    "xlog_grant_log_space: sleep 1");
		/*
		 * Gotta check this before going to sleep, while we're
		 * holding the grant lock.
		 */
		if (XLOG_FORCED_SHUTDOWN(log))
			goto error_return;

		XFS_STATS_INC(xs_sleep_logspace);
		sv_wait(&tic->t_sema, PINOD|PLTWAIT, &log->l_grant_lock, s);
		/*
		 * If we got an error, and the filesystem is shutting down,
		 * we'll catch it down below. So just continue...
		 */
		xlog_trace_loggrant(log, tic,
				    "xlog_grant_log_space: wake 1");
E
Eric Sandeen 已提交
2510
		spin_lock(&log->l_grant_lock);
L
Linus Torvalds 已提交
2511 2512 2513 2514 2515 2516 2517 2518 2519 2520 2521 2522 2523 2524
	}
	if (tic->t_flags & XFS_LOG_PERM_RESERV)
		need_bytes = tic->t_unit_res*tic->t_ocnt;
	else
		need_bytes = tic->t_unit_res;

redo:
	if (XLOG_FORCED_SHUTDOWN(log))
		goto error_return;

	free_bytes = xlog_space_left(log, log->l_grant_reserve_cycle,
				     log->l_grant_reserve_bytes);
	if (free_bytes < need_bytes) {
		if ((tic->t_flags & XLOG_TIC_IN_Q) == 0)
2525
			xlog_ins_ticketq(&log->l_reserve_headq, tic);
L
Linus Torvalds 已提交
2526 2527 2528 2529 2530 2531
		xlog_trace_loggrant(log, tic,
				    "xlog_grant_log_space: sleep 2");
		XFS_STATS_INC(xs_sleep_logspace);
		sv_wait(&tic->t_sema, PINOD|PLTWAIT, &log->l_grant_lock, s);

		if (XLOG_FORCED_SHUTDOWN(log)) {
E
Eric Sandeen 已提交
2532
			spin_lock(&log->l_grant_lock);
L
Linus Torvalds 已提交
2533 2534 2535 2536 2537 2538
			goto error_return;
		}

		xlog_trace_loggrant(log, tic,
				    "xlog_grant_log_space: wake 2");
		xlog_grant_push_ail(log->l_mp, need_bytes);
E
Eric Sandeen 已提交
2539
		spin_lock(&log->l_grant_lock);
L
Linus Torvalds 已提交
2540 2541
		goto redo;
	} else if (tic->t_flags & XLOG_TIC_IN_Q)
2542
		xlog_del_ticketq(&log->l_reserve_headq, tic);
L
Linus Torvalds 已提交
2543 2544

	/* we've got enough space */
2545
	xlog_grant_add_space(log, need_bytes);
L
Linus Torvalds 已提交
2546 2547 2548 2549 2550 2551 2552 2553 2554 2555 2556 2557 2558 2559 2560
#ifdef DEBUG
	tail_lsn = log->l_tail_lsn;
	/*
	 * Check to make sure the grant write head didn't just over lap the
	 * tail.  If the cycles are the same, we can't be overlapping.
	 * Otherwise, make sure that the cycles differ by exactly one and
	 * check the byte count.
	 */
	if (CYCLE_LSN(tail_lsn) != log->l_grant_write_cycle) {
		ASSERT(log->l_grant_write_cycle-1 == CYCLE_LSN(tail_lsn));
		ASSERT(log->l_grant_write_bytes <= BBTOB(BLOCK_LSN(tail_lsn)));
	}
#endif
	xlog_trace_loggrant(log, tic, "xlog_grant_log_space: exit");
	xlog_verify_grant_head(log, 1);
E
Eric Sandeen 已提交
2561
	spin_unlock(&log->l_grant_lock);
L
Linus Torvalds 已提交
2562 2563 2564 2565
	return 0;

 error_return:
	if (tic->t_flags & XLOG_TIC_IN_Q)
2566
		xlog_del_ticketq(&log->l_reserve_headq, tic);
L
Linus Torvalds 已提交
2567 2568 2569 2570 2571 2572 2573 2574
	xlog_trace_loggrant(log, tic, "xlog_grant_log_space: err_ret");
	/*
	 * If we are failing, make sure the ticket doesn't have any
	 * current reservations. We don't want to add this back when
	 * the ticket/transaction gets cancelled.
	 */
	tic->t_curr_res = 0;
	tic->t_cnt = 0; /* ungrant will give back unit_res * t_cnt. */
E
Eric Sandeen 已提交
2575
	spin_unlock(&log->l_grant_lock);
L
Linus Torvalds 已提交
2576 2577 2578 2579 2580 2581 2582 2583 2584 2585 2586 2587 2588 2589 2590 2591 2592 2593 2594 2595
	return XFS_ERROR(EIO);
}	/* xlog_grant_log_space */


/*
 * Replenish the byte reservation required by moving the grant write head.
 *
 *
 */
STATIC int
xlog_regrant_write_log_space(xlog_t	   *log,
			     xlog_ticket_t *tic)
{
	int		free_bytes, need_bytes;
	xlog_ticket_t	*ntic;
#ifdef DEBUG
	xfs_lsn_t	tail_lsn;
#endif

	tic->t_curr_res = tic->t_unit_res;
2596
	xlog_tic_reset_res(tic);
L
Linus Torvalds 已提交
2597 2598

	if (tic->t_cnt > 0)
2599
		return 0;
L
Linus Torvalds 已提交
2600 2601 2602 2603 2604 2605

#ifdef DEBUG
	if (log->l_flags & XLOG_ACTIVE_RECOVERY)
		panic("regrant Recovery problem");
#endif

E
Eric Sandeen 已提交
2606
	spin_lock(&log->l_grant_lock);
L
Linus Torvalds 已提交
2607 2608 2609 2610 2611 2612 2613 2614 2615 2616 2617 2618 2619 2620 2621 2622 2623 2624 2625 2626 2627 2628 2629 2630 2631 2632 2633
	xlog_trace_loggrant(log, tic, "xlog_regrant_write_log_space: enter");

	if (XLOG_FORCED_SHUTDOWN(log))
		goto error_return;

	/* If there are other waiters on the queue then give them a
	 * chance at logspace before us. Wake up the first waiters,
	 * if we do not wake up all the waiters then go to sleep waiting
	 * for more free space, otherwise try to get some space for
	 * this transaction.
	 */

	if ((ntic = log->l_write_headq)) {
		free_bytes = xlog_space_left(log, log->l_grant_write_cycle,
					     log->l_grant_write_bytes);
		do {
			ASSERT(ntic->t_flags & XLOG_TIC_PERM_RESERV);

			if (free_bytes < ntic->t_unit_res)
				break;
			free_bytes -= ntic->t_unit_res;
			sv_signal(&ntic->t_sema);
			ntic = ntic->t_next;
		} while (ntic != log->l_write_headq);

		if (ntic != log->l_write_headq) {
			if ((tic->t_flags & XLOG_TIC_IN_Q) == 0)
2634
				xlog_ins_ticketq(&log->l_write_headq, tic);
L
Linus Torvalds 已提交
2635 2636 2637 2638 2639 2640 2641 2642 2643 2644

			xlog_trace_loggrant(log, tic,
				    "xlog_regrant_write_log_space: sleep 1");
			XFS_STATS_INC(xs_sleep_logspace);
			sv_wait(&tic->t_sema, PINOD|PLTWAIT,
				&log->l_grant_lock, s);

			/* If we're shutting down, this tic is already
			 * off the queue */
			if (XLOG_FORCED_SHUTDOWN(log)) {
E
Eric Sandeen 已提交
2645
				spin_lock(&log->l_grant_lock);
L
Linus Torvalds 已提交
2646 2647 2648 2649 2650 2651
				goto error_return;
			}

			xlog_trace_loggrant(log, tic,
				    "xlog_regrant_write_log_space: wake 1");
			xlog_grant_push_ail(log->l_mp, tic->t_unit_res);
E
Eric Sandeen 已提交
2652
			spin_lock(&log->l_grant_lock);
L
Linus Torvalds 已提交
2653 2654 2655 2656 2657 2658 2659 2660 2661 2662 2663 2664 2665
		}
	}

	need_bytes = tic->t_unit_res;

redo:
	if (XLOG_FORCED_SHUTDOWN(log))
		goto error_return;

	free_bytes = xlog_space_left(log, log->l_grant_write_cycle,
				     log->l_grant_write_bytes);
	if (free_bytes < need_bytes) {
		if ((tic->t_flags & XLOG_TIC_IN_Q) == 0)
2666
			xlog_ins_ticketq(&log->l_write_headq, tic);
L
Linus Torvalds 已提交
2667 2668 2669 2670 2671
		XFS_STATS_INC(xs_sleep_logspace);
		sv_wait(&tic->t_sema, PINOD|PLTWAIT, &log->l_grant_lock, s);

		/* If we're shutting down, this tic is already off the queue */
		if (XLOG_FORCED_SHUTDOWN(log)) {
E
Eric Sandeen 已提交
2672
			spin_lock(&log->l_grant_lock);
L
Linus Torvalds 已提交
2673 2674 2675 2676 2677 2678
			goto error_return;
		}

		xlog_trace_loggrant(log, tic,
				    "xlog_regrant_write_log_space: wake 2");
		xlog_grant_push_ail(log->l_mp, need_bytes);
E
Eric Sandeen 已提交
2679
		spin_lock(&log->l_grant_lock);
L
Linus Torvalds 已提交
2680 2681
		goto redo;
	} else if (tic->t_flags & XLOG_TIC_IN_Q)
2682
		xlog_del_ticketq(&log->l_write_headq, tic);
L
Linus Torvalds 已提交
2683

2684 2685
	/* we've got enough space */
	xlog_grant_add_space_write(log, need_bytes);
L
Linus Torvalds 已提交
2686 2687 2688 2689 2690 2691 2692 2693 2694 2695
#ifdef DEBUG
	tail_lsn = log->l_tail_lsn;
	if (CYCLE_LSN(tail_lsn) != log->l_grant_write_cycle) {
		ASSERT(log->l_grant_write_cycle-1 == CYCLE_LSN(tail_lsn));
		ASSERT(log->l_grant_write_bytes <= BBTOB(BLOCK_LSN(tail_lsn)));
	}
#endif

	xlog_trace_loggrant(log, tic, "xlog_regrant_write_log_space: exit");
	xlog_verify_grant_head(log, 1);
E
Eric Sandeen 已提交
2696
	spin_unlock(&log->l_grant_lock);
2697
	return 0;
L
Linus Torvalds 已提交
2698 2699 2700 2701


 error_return:
	if (tic->t_flags & XLOG_TIC_IN_Q)
2702
		xlog_del_ticketq(&log->l_reserve_headq, tic);
L
Linus Torvalds 已提交
2703 2704 2705 2706 2707 2708 2709 2710
	xlog_trace_loggrant(log, tic, "xlog_regrant_write_log_space: err_ret");
	/*
	 * If we are failing, make sure the ticket doesn't have any
	 * current reservations. We don't want to add this back when
	 * the ticket/transaction gets cancelled.
	 */
	tic->t_curr_res = 0;
	tic->t_cnt = 0; /* ungrant will give back unit_res * t_cnt. */
E
Eric Sandeen 已提交
2711
	spin_unlock(&log->l_grant_lock);
L
Linus Torvalds 已提交
2712 2713 2714 2715 2716 2717 2718 2719 2720 2721 2722 2723 2724 2725 2726 2727 2728 2729 2730 2731
	return XFS_ERROR(EIO);
}	/* xlog_regrant_write_log_space */


/* The first cnt-1 times through here we don't need to
 * move the grant write head because the permanent
 * reservation has reserved cnt times the unit amount.
 * Release part of current permanent unit reservation and
 * reset current reservation to be one units worth.  Also
 * move grant reservation head forward.
 */
STATIC void
xlog_regrant_reserve_log_space(xlog_t	     *log,
			       xlog_ticket_t *ticket)
{
	xlog_trace_loggrant(log, ticket,
			    "xlog_regrant_reserve_log_space: enter");
	if (ticket->t_cnt > 0)
		ticket->t_cnt--;

E
Eric Sandeen 已提交
2732
	spin_lock(&log->l_grant_lock);
2733
	xlog_grant_sub_space(log, ticket->t_curr_res);
L
Linus Torvalds 已提交
2734
	ticket->t_curr_res = ticket->t_unit_res;
2735
	xlog_tic_reset_res(ticket);
L
Linus Torvalds 已提交
2736 2737 2738 2739 2740 2741
	xlog_trace_loggrant(log, ticket,
			    "xlog_regrant_reserve_log_space: sub current res");
	xlog_verify_grant_head(log, 1);

	/* just return if we still have some of the pre-reserved space */
	if (ticket->t_cnt > 0) {
E
Eric Sandeen 已提交
2742
		spin_unlock(&log->l_grant_lock);
L
Linus Torvalds 已提交
2743 2744 2745
		return;
	}

2746
	xlog_grant_add_space_reserve(log, ticket->t_unit_res);
L
Linus Torvalds 已提交
2747 2748 2749
	xlog_trace_loggrant(log, ticket,
			    "xlog_regrant_reserve_log_space: exit");
	xlog_verify_grant_head(log, 0);
E
Eric Sandeen 已提交
2750
	spin_unlock(&log->l_grant_lock);
L
Linus Torvalds 已提交
2751
	ticket->t_curr_res = ticket->t_unit_res;
2752
	xlog_tic_reset_res(ticket);
L
Linus Torvalds 已提交
2753 2754 2755 2756 2757 2758 2759 2760 2761 2762 2763 2764 2765 2766 2767 2768 2769 2770 2771 2772 2773 2774 2775 2776
}	/* xlog_regrant_reserve_log_space */


/*
 * Give back the space left from a reservation.
 *
 * All the information we need to make a correct determination of space left
 * is present.  For non-permanent reservations, things are quite easy.  The
 * count should have been decremented to zero.  We only need to deal with the
 * space remaining in the current reservation part of the ticket.  If the
 * ticket contains a permanent reservation, there may be left over space which
 * needs to be released.  A count of N means that N-1 refills of the current
 * reservation can be done before we need to ask for more space.  The first
 * one goes to fill up the first current reservation.  Once we run out of
 * space, the count will stay at zero and the only space remaining will be
 * in the current reservation field.
 */
STATIC void
xlog_ungrant_log_space(xlog_t	     *log,
		       xlog_ticket_t *ticket)
{
	if (ticket->t_cnt > 0)
		ticket->t_cnt--;

E
Eric Sandeen 已提交
2777
	spin_lock(&log->l_grant_lock);
L
Linus Torvalds 已提交
2778 2779
	xlog_trace_loggrant(log, ticket, "xlog_ungrant_log_space: enter");

2780
	xlog_grant_sub_space(log, ticket->t_curr_res);
L
Linus Torvalds 已提交
2781 2782 2783 2784 2785 2786 2787 2788

	xlog_trace_loggrant(log, ticket, "xlog_ungrant_log_space: sub current");

	/* If this is a permanent reservation ticket, we may be able to free
	 * up more space based on the remaining count.
	 */
	if (ticket->t_cnt > 0) {
		ASSERT(ticket->t_flags & XLOG_TIC_PERM_RESERV);
2789
		xlog_grant_sub_space(log, ticket->t_unit_res*ticket->t_cnt);
L
Linus Torvalds 已提交
2790 2791 2792 2793
	}

	xlog_trace_loggrant(log, ticket, "xlog_ungrant_log_space: exit");
	xlog_verify_grant_head(log, 1);
E
Eric Sandeen 已提交
2794
	spin_unlock(&log->l_grant_lock);
L
Linus Torvalds 已提交
2795 2796 2797 2798 2799 2800 2801 2802 2803 2804 2805 2806 2807
	xfs_log_move_tail(log->l_mp, 1);
}	/* xlog_ungrant_log_space */


/*
 * Flush iclog to disk if this is the last reference to the given iclog and
 * the WANT_SYNC bit is set.
 *
 * When this function is entered, the iclog is not necessarily in the
 * WANT_SYNC state.  It may be sitting around waiting to get filled.
 *
 *
 */
D
David Chinner 已提交
2808
STATIC int
2809 2810 2811
xlog_state_release_iclog(
	xlog_t		*log,
	xlog_in_core_t	*iclog)
L
Linus Torvalds 已提交
2812 2813 2814
{
	int		sync = 0;	/* do we sync? */

2815 2816 2817 2818 2819 2820 2821
	if (iclog->ic_state & XLOG_STATE_IOERROR)
		return XFS_ERROR(EIO);

	ASSERT(atomic_read(&iclog->ic_refcnt) > 0);
	if (!atomic_dec_and_lock(&iclog->ic_refcnt, &log->l_icloglock))
		return 0;

L
Linus Torvalds 已提交
2822
	if (iclog->ic_state & XLOG_STATE_IOERROR) {
E
Eric Sandeen 已提交
2823
		spin_unlock(&log->l_icloglock);
L
Linus Torvalds 已提交
2824 2825 2826 2827 2828
		return XFS_ERROR(EIO);
	}
	ASSERT(iclog->ic_state == XLOG_STATE_ACTIVE ||
	       iclog->ic_state == XLOG_STATE_WANT_SYNC);

2829
	if (iclog->ic_state == XLOG_STATE_WANT_SYNC) {
2830 2831
		/* update tail before writing to iclog */
		xlog_assign_tail_lsn(log->l_mp);
L
Linus Torvalds 已提交
2832 2833
		sync++;
		iclog->ic_state = XLOG_STATE_SYNCING;
2834
		iclog->ic_header.h_tail_lsn = cpu_to_be64(log->l_tail_lsn);
L
Linus Torvalds 已提交
2835 2836 2837
		xlog_verify_tail_lsn(log, iclog, log->l_tail_lsn);
		/* cycle incremented when incrementing curr_block */
	}
E
Eric Sandeen 已提交
2838
	spin_unlock(&log->l_icloglock);
L
Linus Torvalds 已提交
2839 2840 2841

	/*
	 * We let the log lock go, so it's possible that we hit a log I/O
2842
	 * error or some other SHUTDOWN condition that marks the iclog
L
Linus Torvalds 已提交
2843 2844 2845 2846
	 * as XLOG_STATE_IOERROR before the bwrite. However, we know that
	 * this iclog has consistent data, so we ignore IOERROR
	 * flags after this point.
	 */
2847
	if (sync)
L
Linus Torvalds 已提交
2848
		return xlog_sync(log, iclog);
2849
	return 0;
L
Linus Torvalds 已提交
2850 2851 2852 2853 2854 2855 2856 2857 2858 2859 2860 2861 2862 2863 2864 2865 2866 2867 2868
}	/* xlog_state_release_iclog */


/*
 * This routine will mark the current iclog in the ring as WANT_SYNC
 * and move the current iclog pointer to the next iclog in the ring.
 * When this routine is called from xlog_state_get_iclog_space(), the
 * exact size of the iclog has not yet been determined.  All we know is
 * that every data block.  We have run out of space in this log record.
 */
STATIC void
xlog_state_switch_iclogs(xlog_t		*log,
			 xlog_in_core_t *iclog,
			 int		eventual_size)
{
	ASSERT(iclog->ic_state == XLOG_STATE_ACTIVE);
	if (!eventual_size)
		eventual_size = iclog->ic_offset;
	iclog->ic_state = XLOG_STATE_WANT_SYNC;
2869
	iclog->ic_header.h_prev_block = cpu_to_be32(log->l_prev_block);
L
Linus Torvalds 已提交
2870 2871 2872 2873 2874 2875 2876
	log->l_prev_block = log->l_curr_block;
	log->l_prev_cycle = log->l_curr_cycle;

	/* roll log?: ic_offset changed later */
	log->l_curr_block += BTOBB(eventual_size)+BTOBB(log->l_iclog_hsize);

	/* Round up to next log-sunit */
2877
	if (xfs_sb_version_haslogv2(&log->l_mp->m_sb) &&
L
Linus Torvalds 已提交
2878 2879 2880 2881 2882 2883 2884 2885 2886 2887 2888 2889 2890 2891 2892 2893 2894 2895 2896 2897 2898 2899 2900 2901 2902 2903 2904 2905 2906 2907 2908 2909 2910 2911 2912 2913 2914 2915 2916 2917 2918 2919 2920 2921 2922
	    log->l_mp->m_sb.sb_logsunit > 1) {
		__uint32_t sunit_bb = BTOBB(log->l_mp->m_sb.sb_logsunit);
		log->l_curr_block = roundup(log->l_curr_block, sunit_bb);
	}

	if (log->l_curr_block >= log->l_logBBsize) {
		log->l_curr_cycle++;
		if (log->l_curr_cycle == XLOG_HEADER_MAGIC_NUM)
			log->l_curr_cycle++;
		log->l_curr_block -= log->l_logBBsize;
		ASSERT(log->l_curr_block >= 0);
	}
	ASSERT(iclog == log->l_iclog);
	log->l_iclog = iclog->ic_next;
}	/* xlog_state_switch_iclogs */


/*
 * Write out all data in the in-core log as of this exact moment in time.
 *
 * Data may be written to the in-core log during this call.  However,
 * we don't guarantee this data will be written out.  A change from past
 * implementation means this routine will *not* write out zero length LRs.
 *
 * Basically, we try and perform an intelligent scan of the in-core logs.
 * If we determine there is no flushable data, we just return.  There is no
 * flushable data if:
 *
 *	1. the current iclog is active and has no data; the previous iclog
 *		is in the active or dirty state.
 *	2. the current iclog is drity, and the previous iclog is in the
 *		active or dirty state.
 *
 * We may sleep (call psema) if:
 *
 *	1. the current iclog is not in the active nor dirty state.
 *	2. the current iclog dirty, and the previous iclog is not in the
 *		active nor dirty state.
 *	3. the current iclog is active, and there is another thread writing
 *		to this particular iclog.
 *	4. a) the current iclog is active and has no other writers
 *	   b) when we return from flushing out this iclog, it is still
 *		not in the active nor dirty state.
 */
STATIC int
2923
xlog_state_sync_all(xlog_t *log, uint flags, int *log_flushed)
L
Linus Torvalds 已提交
2924 2925 2926 2927
{
	xlog_in_core_t	*iclog;
	xfs_lsn_t	lsn;

E
Eric Sandeen 已提交
2928
	spin_lock(&log->l_icloglock);
L
Linus Torvalds 已提交
2929 2930 2931

	iclog = log->l_iclog;
	if (iclog->ic_state & XLOG_STATE_IOERROR) {
E
Eric Sandeen 已提交
2932
		spin_unlock(&log->l_icloglock);
L
Linus Torvalds 已提交
2933 2934 2935 2936 2937 2938 2939 2940 2941 2942 2943 2944 2945 2946 2947 2948
		return XFS_ERROR(EIO);
	}

	/* If the head iclog is not active nor dirty, we just attach
	 * ourselves to the head and go to sleep.
	 */
	if (iclog->ic_state == XLOG_STATE_ACTIVE ||
	    iclog->ic_state == XLOG_STATE_DIRTY) {
		/*
		 * If the head is dirty or (active and empty), then
		 * we need to look at the previous iclog.  If the previous
		 * iclog is active or dirty we are done.  There is nothing
		 * to sync out.  Otherwise, we attach ourselves to the
		 * previous iclog and go to sleep.
		 */
		if (iclog->ic_state == XLOG_STATE_DIRTY ||
2949 2950
		    (atomic_read(&iclog->ic_refcnt) == 0
		     && iclog->ic_offset == 0)) {
L
Linus Torvalds 已提交
2951 2952 2953 2954 2955 2956 2957
			iclog = iclog->ic_prev;
			if (iclog->ic_state == XLOG_STATE_ACTIVE ||
			    iclog->ic_state == XLOG_STATE_DIRTY)
				goto no_sleep;
			else
				goto maybe_sleep;
		} else {
2958
			if (atomic_read(&iclog->ic_refcnt) == 0) {
L
Linus Torvalds 已提交
2959 2960 2961 2962 2963 2964
				/* We are the only one with access to this
				 * iclog.  Flush it out now.  There should
				 * be a roundoff of zero to show that someone
				 * has already taken care of the roundoff from
				 * the previous sync.
				 */
2965
				atomic_inc(&iclog->ic_refcnt);
2966
				lsn = be64_to_cpu(iclog->ic_header.h_lsn);
L
Linus Torvalds 已提交
2967
				xlog_state_switch_iclogs(log, iclog, 0);
E
Eric Sandeen 已提交
2968
				spin_unlock(&log->l_icloglock);
L
Linus Torvalds 已提交
2969 2970 2971

				if (xlog_state_release_iclog(log, iclog))
					return XFS_ERROR(EIO);
2972
				*log_flushed = 1;
E
Eric Sandeen 已提交
2973
				spin_lock(&log->l_icloglock);
2974
				if (be64_to_cpu(iclog->ic_header.h_lsn) == lsn &&
L
Linus Torvalds 已提交
2975 2976 2977 2978 2979 2980 2981 2982 2983 2984 2985 2986 2987 2988 2989 2990 2991 2992 2993 2994 2995 2996 2997 2998
				    iclog->ic_state != XLOG_STATE_DIRTY)
					goto maybe_sleep;
				else
					goto no_sleep;
			} else {
				/* Someone else is writing to this iclog.
				 * Use its call to flush out the data.  However,
				 * the other thread may not force out this LR,
				 * so we mark it WANT_SYNC.
				 */
				xlog_state_switch_iclogs(log, iclog, 0);
				goto maybe_sleep;
			}
		}
	}

	/* By the time we come around again, the iclog could've been filled
	 * which would give it another lsn.  If we have a new lsn, just
	 * return because the relevant data has been flushed.
	 */
maybe_sleep:
	if (flags & XFS_LOG_SYNC) {
		/*
		 * We must check if we're shutting down here, before
E
Eric Sandeen 已提交
2999
		 * we wait, while we're holding the l_icloglock.
L
Linus Torvalds 已提交
3000 3001 3002 3003
		 * Then we check again after waking up, in case our
		 * sleep was disturbed by a bad news.
		 */
		if (iclog->ic_state & XLOG_STATE_IOERROR) {
E
Eric Sandeen 已提交
3004
			spin_unlock(&log->l_icloglock);
L
Linus Torvalds 已提交
3005 3006 3007 3008 3009 3010 3011 3012 3013 3014 3015
			return XFS_ERROR(EIO);
		}
		XFS_STATS_INC(xs_log_force_sleep);
		sv_wait(&iclog->ic_forcesema, PINOD, &log->l_icloglock, s);
		/*
		 * No need to grab the log lock here since we're
		 * only deciding whether or not to return EIO
		 * and the memory read should be atomic.
		 */
		if (iclog->ic_state & XLOG_STATE_IOERROR)
			return XFS_ERROR(EIO);
3016
		*log_flushed = 1;
L
Linus Torvalds 已提交
3017 3018 3019 3020

	} else {

no_sleep:
E
Eric Sandeen 已提交
3021
		spin_unlock(&log->l_icloglock);
L
Linus Torvalds 已提交
3022 3023 3024 3025 3026 3027 3028 3029 3030 3031 3032 3033 3034 3035 3036 3037 3038
	}
	return 0;
}	/* xlog_state_sync_all */


/*
 * Used by code which implements synchronous log forces.
 *
 * Find in-core log with lsn.
 *	If it is in the DIRTY state, just return.
 *	If it is in the ACTIVE state, move the in-core log into the WANT_SYNC
 *		state and go to sleep or return.
 *	If it is in any other state, go to sleep or return.
 *
 * If filesystem activity goes to zero, the iclog will get flushed only by
 * bdflush().
 */
D
David Chinner 已提交
3039
STATIC int
L
Linus Torvalds 已提交
3040 3041
xlog_state_sync(xlog_t	  *log,
		xfs_lsn_t lsn,
3042 3043
		uint	  flags,
		int	  *log_flushed)
L
Linus Torvalds 已提交
3044 3045 3046 3047 3048
{
    xlog_in_core_t	*iclog;
    int			already_slept = 0;

try_again:
E
Eric Sandeen 已提交
3049
    spin_lock(&log->l_icloglock);
L
Linus Torvalds 已提交
3050 3051 3052
    iclog = log->l_iclog;

    if (iclog->ic_state & XLOG_STATE_IOERROR) {
E
Eric Sandeen 已提交
3053
	    spin_unlock(&log->l_icloglock);
L
Linus Torvalds 已提交
3054 3055 3056 3057
	    return XFS_ERROR(EIO);
    }

    do {
3058 3059 3060
	if (be64_to_cpu(iclog->ic_header.h_lsn) != lsn) {
		iclog = iclog->ic_next;
		continue;
L
Linus Torvalds 已提交
3061 3062 3063
	}

	if (iclog->ic_state == XLOG_STATE_DIRTY) {
E
Eric Sandeen 已提交
3064
		spin_unlock(&log->l_icloglock);
L
Linus Torvalds 已提交
3065 3066 3067 3068 3069 3070 3071 3072 3073 3074 3075 3076 3077 3078 3079 3080 3081 3082 3083 3084 3085 3086 3087 3088 3089 3090 3091 3092
		return 0;
	}

	if (iclog->ic_state == XLOG_STATE_ACTIVE) {
		/*
		 * We sleep here if we haven't already slept (e.g.
		 * this is the first time we've looked at the correct
		 * iclog buf) and the buffer before us is going to
		 * be sync'ed. The reason for this is that if we
		 * are doing sync transactions here, by waiting for
		 * the previous I/O to complete, we can allow a few
		 * more transactions into this iclog before we close
		 * it down.
		 *
		 * Otherwise, we mark the buffer WANT_SYNC, and bump
		 * up the refcnt so we can release the log (which drops
		 * the ref count).  The state switch keeps new transaction
		 * commits from using this buffer.  When the current commits
		 * finish writing into the buffer, the refcount will drop to
		 * zero and the buffer will go out then.
		 */
		if (!already_slept &&
		    (iclog->ic_prev->ic_state & (XLOG_STATE_WANT_SYNC |
						 XLOG_STATE_SYNCING))) {
			ASSERT(!(iclog->ic_state & XLOG_STATE_IOERROR));
			XFS_STATS_INC(xs_log_force_sleep);
			sv_wait(&iclog->ic_prev->ic_writesema, PSWP,
				&log->l_icloglock, s);
3093
			*log_flushed = 1;
L
Linus Torvalds 已提交
3094 3095 3096
			already_slept = 1;
			goto try_again;
		} else {
3097
			atomic_inc(&iclog->ic_refcnt);
L
Linus Torvalds 已提交
3098
			xlog_state_switch_iclogs(log, iclog, 0);
E
Eric Sandeen 已提交
3099
			spin_unlock(&log->l_icloglock);
L
Linus Torvalds 已提交
3100 3101
			if (xlog_state_release_iclog(log, iclog))
				return XFS_ERROR(EIO);
3102
			*log_flushed = 1;
E
Eric Sandeen 已提交
3103
			spin_lock(&log->l_icloglock);
L
Linus Torvalds 已提交
3104 3105 3106 3107 3108 3109 3110 3111 3112 3113 3114
		}
	}

	if ((flags & XFS_LOG_SYNC) && /* sleep */
	    !(iclog->ic_state & (XLOG_STATE_ACTIVE | XLOG_STATE_DIRTY))) {

		/*
		 * Don't wait on the forcesema if we know that we've
		 * gotten a log write error.
		 */
		if (iclog->ic_state & XLOG_STATE_IOERROR) {
E
Eric Sandeen 已提交
3115
			spin_unlock(&log->l_icloglock);
L
Linus Torvalds 已提交
3116 3117 3118 3119 3120 3121 3122 3123 3124 3125 3126
			return XFS_ERROR(EIO);
		}
		XFS_STATS_INC(xs_log_force_sleep);
		sv_wait(&iclog->ic_forcesema, PSWP, &log->l_icloglock, s);
		/*
		 * No need to grab the log lock here since we're
		 * only deciding whether or not to return EIO
		 * and the memory read should be atomic.
		 */
		if (iclog->ic_state & XLOG_STATE_IOERROR)
			return XFS_ERROR(EIO);
3127
		*log_flushed = 1;
L
Linus Torvalds 已提交
3128
	} else {		/* just return */
E
Eric Sandeen 已提交
3129
		spin_unlock(&log->l_icloglock);
L
Linus Torvalds 已提交
3130 3131 3132 3133 3134
	}
	return 0;

    } while (iclog != log->l_iclog);

E
Eric Sandeen 已提交
3135
    spin_unlock(&log->l_icloglock);
3136
    return 0;
L
Linus Torvalds 已提交
3137 3138 3139 3140 3141 3142 3143
}	/* xlog_state_sync */


/*
 * Called when we want to mark the current iclog as being ready to sync to
 * disk.
 */
D
David Chinner 已提交
3144
STATIC void
L
Linus Torvalds 已提交
3145 3146
xlog_state_want_sync(xlog_t *log, xlog_in_core_t *iclog)
{
E
Eric Sandeen 已提交
3147
	spin_lock(&log->l_icloglock);
L
Linus Torvalds 已提交
3148 3149 3150 3151 3152 3153 3154 3155

	if (iclog->ic_state == XLOG_STATE_ACTIVE) {
		xlog_state_switch_iclogs(log, iclog, 0);
	} else {
		ASSERT(iclog->ic_state &
			(XLOG_STATE_WANT_SYNC|XLOG_STATE_IOERROR));
	}

E
Eric Sandeen 已提交
3156
	spin_unlock(&log->l_icloglock);
L
Linus Torvalds 已提交
3157 3158 3159 3160 3161 3162 3163 3164 3165 3166 3167 3168
}	/* xlog_state_want_sync */



/*****************************************************************************
 *
 *		TICKET functions
 *
 *****************************************************************************
 */

/*
3169
 * Free a used ticket.
L
Linus Torvalds 已提交
3170 3171 3172 3173 3174 3175
 */
STATIC void
xlog_ticket_put(xlog_t		*log,
		xlog_ticket_t	*ticket)
{
	sv_destroy(&ticket->t_sema);
3176
	kmem_zone_free(xfs_log_ticket_zone, ticket);
L
Linus Torvalds 已提交
3177 3178 3179 3180
}	/* xlog_ticket_put */


/*
3181
 * Allocate and initialise a new log ticket.
L
Linus Torvalds 已提交
3182
 */
D
David Chinner 已提交
3183
STATIC xlog_ticket_t *
L
Linus Torvalds 已提交
3184 3185 3186 3187 3188 3189 3190 3191 3192
xlog_ticket_get(xlog_t		*log,
		int		unit_bytes,
		int		cnt,
		char		client,
		uint		xflags)
{
	xlog_ticket_t	*tic;
	uint		num_headers;

3193 3194 3195
	tic = kmem_zone_zalloc(xfs_log_ticket_zone, KM_SLEEP|KM_MAYFAIL);
	if (!tic)
		return NULL;
L
Linus Torvalds 已提交
3196 3197 3198 3199 3200 3201 3202 3203

	/*
	 * Permanent reservations have up to 'cnt'-1 active log operations
	 * in the log.  A unit in this case is the amount of space for one
	 * of these log operations.  Normal reservations have a cnt of 1
	 * and their unit amount is the total amount of space required.
	 *
	 * The following lines of code account for non-transaction data
3204 3205 3206 3207 3208 3209 3210 3211 3212 3213 3214 3215 3216 3217 3218 3219 3220 3221 3222 3223 3224 3225
	 * which occupy space in the on-disk log.
	 *
	 * Normal form of a transaction is:
	 * <oph><trans-hdr><start-oph><reg1-oph><reg1><reg2-oph>...<commit-oph>
	 * and then there are LR hdrs, split-recs and roundoff at end of syncs.
	 *
	 * We need to account for all the leadup data and trailer data
	 * around the transaction data.
	 * And then we need to account for the worst case in terms of using
	 * more space.
	 * The worst case will happen if:
	 * - the placement of the transaction happens to be such that the
	 *   roundoff is at its maximum
	 * - the transaction data is synced before the commit record is synced
	 *   i.e. <transaction-data><roundoff> | <commit-rec><roundoff>
	 *   Therefore the commit record is in its own Log Record.
	 *   This can happen as the commit record is called with its
	 *   own region to xlog_write().
	 *   This then means that in the worst case, roundoff can happen for
	 *   the commit-rec as well.
	 *   The commit-rec is smaller than padding in this scenario and so it is
	 *   not added separately.
L
Linus Torvalds 已提交
3226 3227
	 */

3228 3229 3230 3231
	/* for trans header */
	unit_bytes += sizeof(xlog_op_header_t);
	unit_bytes += sizeof(xfs_trans_header_t);

L
Linus Torvalds 已提交
3232
	/* for start-rec */
3233 3234 3235 3236 3237
	unit_bytes += sizeof(xlog_op_header_t);

	/* for LR headers */
	num_headers = ((unit_bytes + log->l_iclog_size-1) >> log->l_iclog_size_log);
	unit_bytes += log->l_iclog_hsize * num_headers;
L
Linus Torvalds 已提交
3238

3239 3240 3241 3242 3243 3244 3245
	/* for commit-rec LR header - note: padding will subsume the ophdr */
	unit_bytes += log->l_iclog_hsize;

	/* for split-recs - ophdrs added when data split over LRs */
	unit_bytes += sizeof(xlog_op_header_t) * num_headers;

	/* for roundoff padding for transaction data and one for commit record */
3246
	if (xfs_sb_version_haslogv2(&log->l_mp->m_sb) &&
3247
	    log->l_mp->m_sb.sb_logsunit > 1) {
L
Linus Torvalds 已提交
3248
		/* log su roundoff */
3249
		unit_bytes += 2*log->l_mp->m_sb.sb_logsunit;
L
Linus Torvalds 已提交
3250 3251
	} else {
		/* BB roundoff */
3252
		unit_bytes += 2*BBSIZE;
L
Linus Torvalds 已提交
3253 3254 3255 3256 3257 3258 3259 3260 3261
        }

	tic->t_unit_res		= unit_bytes;
	tic->t_curr_res		= unit_bytes;
	tic->t_cnt		= cnt;
	tic->t_ocnt		= cnt;
	tic->t_tid		= (xlog_tid_t)((__psint_t)tic & 0xffffffff);
	tic->t_clientid		= client;
	tic->t_flags		= XLOG_TIC_INITED;
3262
	tic->t_trans_type	= 0;
L
Linus Torvalds 已提交
3263 3264 3265 3266
	if (xflags & XFS_LOG_PERM_RESERV)
		tic->t_flags |= XLOG_TIC_PERM_RESERV;
	sv_init(&(tic->t_sema), SV_DEFAULT, "logtick");

3267
	xlog_tic_reset_res(tic);
3268

L
Linus Torvalds 已提交
3269 3270 3271 3272 3273 3274 3275 3276 3277 3278
	return tic;
}	/* xlog_ticket_get */


/******************************************************************************
 *
 *		Log debug routines
 *
 ******************************************************************************
 */
3279
#if defined(DEBUG)
L
Linus Torvalds 已提交
3280 3281 3282 3283 3284 3285 3286 3287 3288 3289 3290 3291 3292 3293 3294 3295 3296 3297 3298 3299 3300 3301 3302 3303 3304 3305 3306 3307 3308 3309 3310 3311 3312 3313 3314 3315 3316 3317 3318 3319 3320 3321 3322 3323 3324 3325 3326 3327 3328 3329 3330 3331 3332 3333 3334 3335 3336 3337 3338 3339 3340 3341 3342 3343 3344 3345 3346 3347 3348 3349 3350 3351 3352 3353 3354 3355 3356 3357 3358 3359 3360 3361 3362 3363 3364 3365 3366 3367 3368 3369 3370 3371
/*
 * Make sure that the destination ptr is within the valid data region of
 * one of the iclogs.  This uses backup pointers stored in a different
 * part of the log in case we trash the log structure.
 */
void
xlog_verify_dest_ptr(xlog_t     *log,
		     __psint_t  ptr)
{
	int i;
	int good_ptr = 0;

	for (i=0; i < log->l_iclog_bufs; i++) {
		if (ptr >= (__psint_t)log->l_iclog_bak[i] &&
		    ptr <= (__psint_t)log->l_iclog_bak[i]+log->l_iclog_size)
			good_ptr++;
	}
	if (! good_ptr)
		xlog_panic("xlog_verify_dest_ptr: invalid ptr");
}	/* xlog_verify_dest_ptr */

STATIC void
xlog_verify_grant_head(xlog_t *log, int equals)
{
    if (log->l_grant_reserve_cycle == log->l_grant_write_cycle) {
	if (equals)
	    ASSERT(log->l_grant_reserve_bytes >= log->l_grant_write_bytes);
	else
	    ASSERT(log->l_grant_reserve_bytes > log->l_grant_write_bytes);
    } else {
	ASSERT(log->l_grant_reserve_cycle-1 == log->l_grant_write_cycle);
	ASSERT(log->l_grant_write_bytes >= log->l_grant_reserve_bytes);
    }
}	/* xlog_verify_grant_head */

/* check if it will fit */
STATIC void
xlog_verify_tail_lsn(xlog_t	    *log,
		     xlog_in_core_t *iclog,
		     xfs_lsn_t	    tail_lsn)
{
    int blocks;

    if (CYCLE_LSN(tail_lsn) == log->l_prev_cycle) {
	blocks =
	    log->l_logBBsize - (log->l_prev_block - BLOCK_LSN(tail_lsn));
	if (blocks < BTOBB(iclog->ic_offset)+BTOBB(log->l_iclog_hsize))
	    xlog_panic("xlog_verify_tail_lsn: ran out of log space");
    } else {
	ASSERT(CYCLE_LSN(tail_lsn)+1 == log->l_prev_cycle);

	if (BLOCK_LSN(tail_lsn) == log->l_prev_block)
	    xlog_panic("xlog_verify_tail_lsn: tail wrapped");

	blocks = BLOCK_LSN(tail_lsn) - log->l_prev_block;
	if (blocks < BTOBB(iclog->ic_offset) + 1)
	    xlog_panic("xlog_verify_tail_lsn: ran out of log space");
    }
}	/* xlog_verify_tail_lsn */

/*
 * Perform a number of checks on the iclog before writing to disk.
 *
 * 1. Make sure the iclogs are still circular
 * 2. Make sure we have a good magic number
 * 3. Make sure we don't have magic numbers in the data
 * 4. Check fields of each log operation header for:
 *	A. Valid client identifier
 *	B. tid ptr value falls in valid ptr space (user space code)
 *	C. Length in log record header is correct according to the
 *		individual operation headers within record.
 * 5. When a bwrite will occur within 5 blocks of the front of the physical
 *	log, check the preceding blocks of the physical log to make sure all
 *	the cycle numbers agree with the current cycle number.
 */
STATIC void
xlog_verify_iclog(xlog_t	 *log,
		  xlog_in_core_t *iclog,
		  int		 count,
		  boolean_t	 syncing)
{
	xlog_op_header_t	*ophead;
	xlog_in_core_t		*icptr;
	xlog_in_core_2_t	*xhdr;
	xfs_caddr_t		ptr;
	xfs_caddr_t		base_ptr;
	__psint_t		field_offset;
	__uint8_t		clientid;
	int			len, i, j, k, op_len;
	int			idx;

	/* check validity of iclog pointers */
E
Eric Sandeen 已提交
3372
	spin_lock(&log->l_icloglock);
L
Linus Torvalds 已提交
3373 3374
	icptr = log->l_iclog;
	for (i=0; i < log->l_iclog_bufs; i++) {
3375
		if (icptr == NULL)
L
Linus Torvalds 已提交
3376 3377 3378 3379 3380
			xlog_panic("xlog_verify_iclog: invalid ptr");
		icptr = icptr->ic_next;
	}
	if (icptr != log->l_iclog)
		xlog_panic("xlog_verify_iclog: corrupt iclog ring");
E
Eric Sandeen 已提交
3381
	spin_unlock(&log->l_icloglock);
L
Linus Torvalds 已提交
3382 3383

	/* check log magic numbers */
3384
	if (be32_to_cpu(iclog->ic_header.h_magicno) != XLOG_HEADER_MAGIC_NUM)
L
Linus Torvalds 已提交
3385 3386
		xlog_panic("xlog_verify_iclog: invalid magic num");

3387 3388
	ptr = (xfs_caddr_t) &iclog->ic_header;
	for (ptr += BBSIZE; ptr < ((xfs_caddr_t)&iclog->ic_header) + count;
L
Linus Torvalds 已提交
3389
	     ptr += BBSIZE) {
3390
		if (be32_to_cpu(*(__be32 *)ptr) == XLOG_HEADER_MAGIC_NUM)
L
Linus Torvalds 已提交
3391 3392 3393 3394
			xlog_panic("xlog_verify_iclog: unexpected magic num");
	}

	/* check fields */
3395
	len = be32_to_cpu(iclog->ic_header.h_num_logops);
L
Linus Torvalds 已提交
3396 3397 3398 3399 3400 3401 3402 3403 3404 3405 3406 3407 3408 3409 3410 3411 3412
	ptr = iclog->ic_datap;
	base_ptr = ptr;
	ophead = (xlog_op_header_t *)ptr;
	xhdr = (xlog_in_core_2_t *)&iclog->ic_header;
	for (i = 0; i < len; i++) {
		ophead = (xlog_op_header_t *)ptr;

		/* clientid is only 1 byte */
		field_offset = (__psint_t)
			       ((xfs_caddr_t)&(ophead->oh_clientid) - base_ptr);
		if (syncing == B_FALSE || (field_offset & 0x1ff)) {
			clientid = ophead->oh_clientid;
		} else {
			idx = BTOBBT((xfs_caddr_t)&(ophead->oh_clientid) - iclog->ic_datap);
			if (idx >= (XLOG_HEADER_CYCLE_SIZE / BBSIZE)) {
				j = idx / (XLOG_HEADER_CYCLE_SIZE / BBSIZE);
				k = idx % (XLOG_HEADER_CYCLE_SIZE / BBSIZE);
3413 3414
				clientid = xlog_get_client_id(
					xhdr[j].hic_xheader.xh_cycle_data[k]);
L
Linus Torvalds 已提交
3415
			} else {
3416 3417
				clientid = xlog_get_client_id(
					iclog->ic_header.h_cycle_data[idx]);
L
Linus Torvalds 已提交
3418 3419 3420
			}
		}
		if (clientid != XFS_TRANSACTION && clientid != XFS_LOG)
3421 3422 3423
			cmn_err(CE_WARN, "xlog_verify_iclog: "
				"invalid clientid %d op 0x%p offset 0x%lx",
				clientid, ophead, (unsigned long)field_offset);
L
Linus Torvalds 已提交
3424 3425 3426 3427 3428

		/* check length */
		field_offset = (__psint_t)
			       ((xfs_caddr_t)&(ophead->oh_len) - base_ptr);
		if (syncing == B_FALSE || (field_offset & 0x1ff)) {
3429
			op_len = be32_to_cpu(ophead->oh_len);
L
Linus Torvalds 已提交
3430 3431 3432 3433 3434 3435
		} else {
			idx = BTOBBT((__psint_t)&ophead->oh_len -
				    (__psint_t)iclog->ic_datap);
			if (idx >= (XLOG_HEADER_CYCLE_SIZE / BBSIZE)) {
				j = idx / (XLOG_HEADER_CYCLE_SIZE / BBSIZE);
				k = idx % (XLOG_HEADER_CYCLE_SIZE / BBSIZE);
3436
				op_len = be32_to_cpu(xhdr[j].hic_xheader.xh_cycle_data[k]);
L
Linus Torvalds 已提交
3437
			} else {
3438
				op_len = be32_to_cpu(iclog->ic_header.h_cycle_data[idx]);
L
Linus Torvalds 已提交
3439 3440 3441 3442 3443
			}
		}
		ptr += sizeof(xlog_op_header_t) + op_len;
	}
}	/* xlog_verify_iclog */
3444
#endif
L
Linus Torvalds 已提交
3445 3446

/*
E
Eric Sandeen 已提交
3447
 * Mark all iclogs IOERROR. l_icloglock is held by the caller.
L
Linus Torvalds 已提交
3448 3449 3450 3451 3452 3453 3454 3455 3456 3457 3458 3459 3460 3461 3462 3463 3464 3465
 */
STATIC int
xlog_state_ioerror(
	xlog_t	*log)
{
	xlog_in_core_t	*iclog, *ic;

	iclog = log->l_iclog;
	if (! (iclog->ic_state & XLOG_STATE_IOERROR)) {
		/*
		 * Mark all the incore logs IOERROR.
		 * From now on, no log flushes will result.
		 */
		ic = iclog;
		do {
			ic->ic_state = XLOG_STATE_IOERROR;
			ic = ic->ic_next;
		} while (ic != iclog);
3466
		return 0;
L
Linus Torvalds 已提交
3467 3468 3469 3470
	}
	/*
	 * Return non-zero, if state transition has already happened.
	 */
3471
	return 1;
L
Linus Torvalds 已提交
3472 3473 3474 3475 3476 3477 3478 3479 3480 3481 3482 3483 3484 3485 3486 3487 3488 3489 3490 3491 3492 3493
}

/*
 * This is called from xfs_force_shutdown, when we're forcibly
 * shutting down the filesystem, typically because of an IO error.
 * Our main objectives here are to make sure that:
 *	a. the filesystem gets marked 'SHUTDOWN' for all interested
 *	   parties to find out, 'atomically'.
 *	b. those who're sleeping on log reservations, pinned objects and
 *	    other resources get woken up, and be told the bad news.
 *	c. nothing new gets queued up after (a) and (b) are done.
 *	d. if !logerror, flush the iclogs to disk, then seal them off
 *	   for business.
 */
int
xfs_log_force_umount(
	struct xfs_mount	*mp,
	int			logerror)
{
	xlog_ticket_t	*tic;
	xlog_t		*log;
	int		retval;
3494
	int		dummy;
L
Linus Torvalds 已提交
3495 3496 3497 3498 3499 3500 3501 3502 3503 3504 3505

	log = mp->m_log;

	/*
	 * If this happens during log recovery, don't worry about
	 * locking; the log isn't open for business yet.
	 */
	if (!log ||
	    log->l_flags & XLOG_ACTIVE_RECOVERY) {
		mp->m_flags |= XFS_MOUNT_FS_SHUTDOWN;
		XFS_BUF_DONE(mp->m_sb_bp);
3506
		return 0;
L
Linus Torvalds 已提交
3507 3508 3509 3510 3511 3512 3513 3514
	}

	/*
	 * Somebody could've already done the hard work for us.
	 * No need to get locks for this.
	 */
	if (logerror && log->l_iclog->ic_state & XLOG_STATE_IOERROR) {
		ASSERT(XLOG_FORCED_SHUTDOWN(log));
3515
		return 1;
L
Linus Torvalds 已提交
3516 3517 3518 3519 3520 3521 3522
	}
	retval = 0;
	/*
	 * We must hold both the GRANT lock and the LOG lock,
	 * before we mark the filesystem SHUTDOWN and wake
	 * everybody up to tell the bad news.
	 */
E
Eric Sandeen 已提交
3523
	spin_lock(&log->l_icloglock);
3524
	spin_lock(&log->l_grant_lock);
L
Linus Torvalds 已提交
3525 3526 3527 3528 3529 3530 3531 3532 3533 3534 3535 3536 3537 3538 3539
	mp->m_flags |= XFS_MOUNT_FS_SHUTDOWN;
	XFS_BUF_DONE(mp->m_sb_bp);
	/*
	 * This flag is sort of redundant because of the mount flag, but
	 * it's good to maintain the separation between the log and the rest
	 * of XFS.
	 */
	log->l_flags |= XLOG_IO_ERROR;

	/*
	 * If we hit a log error, we want to mark all the iclogs IOERROR
	 * while we're still holding the loglock.
	 */
	if (logerror)
		retval = xlog_state_ioerror(log);
E
Eric Sandeen 已提交
3540
	spin_unlock(&log->l_icloglock);
L
Linus Torvalds 已提交
3541 3542 3543 3544 3545 3546 3547 3548 3549 3550 3551 3552 3553 3554 3555 3556 3557 3558 3559 3560 3561 3562

	/*
	 * We don't want anybody waiting for log reservations
	 * after this. That means we have to wake up everybody
	 * queued up on reserve_headq as well as write_headq.
	 * In addition, we make sure in xlog_{re}grant_log_space
	 * that we don't enqueue anything once the SHUTDOWN flag
	 * is set, and this action is protected by the GRANTLOCK.
	 */
	if ((tic = log->l_reserve_headq)) {
		do {
			sv_signal(&tic->t_sema);
			tic = tic->t_next;
		} while (tic != log->l_reserve_headq);
	}

	if ((tic = log->l_write_headq)) {
		do {
			sv_signal(&tic->t_sema);
			tic = tic->t_next;
		} while (tic != log->l_write_headq);
	}
E
Eric Sandeen 已提交
3563
	spin_unlock(&log->l_grant_lock);
L
Linus Torvalds 已提交
3564 3565 3566 3567 3568 3569 3570

	if (! (log->l_iclog->ic_state & XLOG_STATE_IOERROR)) {
		ASSERT(!logerror);
		/*
		 * Force the incore logs to disk before shutting the
		 * log down completely.
		 */
3571
		xlog_state_sync_all(log, XFS_LOG_FORCE|XFS_LOG_SYNC, &dummy);
E
Eric Sandeen 已提交
3572
		spin_lock(&log->l_icloglock);
L
Linus Torvalds 已提交
3573
		retval = xlog_state_ioerror(log);
E
Eric Sandeen 已提交
3574
		spin_unlock(&log->l_icloglock);
L
Linus Torvalds 已提交
3575 3576 3577 3578 3579 3580 3581 3582 3583 3584 3585 3586
	}
	/*
	 * Wake up everybody waiting on xfs_log_force.
	 * Callback all log item committed functions as if the
	 * log writes were completed.
	 */
	xlog_state_do_callback(log, XFS_LI_ABORTED, NULL);

#ifdef XFSERRORDEBUG
	{
		xlog_in_core_t	*iclog;

E
Eric Sandeen 已提交
3587
		spin_lock(&log->l_icloglock);
L
Linus Torvalds 已提交
3588 3589 3590 3591 3592
		iclog = log->l_iclog;
		do {
			ASSERT(iclog->ic_callback == 0);
			iclog = iclog->ic_next;
		} while (iclog != log->l_iclog);
E
Eric Sandeen 已提交
3593
		spin_unlock(&log->l_icloglock);
L
Linus Torvalds 已提交
3594 3595 3596
	}
#endif
	/* return non-zero if log IOERROR transition had already happened */
3597
	return retval;
L
Linus Torvalds 已提交
3598 3599
}

3600
STATIC int
L
Linus Torvalds 已提交
3601 3602 3603 3604 3605 3606 3607 3608 3609 3610
xlog_iclogs_empty(xlog_t *log)
{
	xlog_in_core_t	*iclog;

	iclog = log->l_iclog;
	do {
		/* endianness does not matter here, zero is zero in
		 * any language.
		 */
		if (iclog->ic_header.h_num_logops)
3611
			return 0;
L
Linus Torvalds 已提交
3612 3613
		iclog = iclog->ic_next;
	} while (iclog != log->l_iclog);
3614
	return 1;
L
Linus Torvalds 已提交
3615
}