xref: /openbmc/linux/fs/gfs2/lock_dlm.c (revision 601ef0d5)
1 // SPDX-License-Identifier: GPL-2.0-only
2 /*
3  * Copyright (C) Sistina Software, Inc.  1997-2003 All rights reserved.
4  * Copyright 2004-2011 Red Hat, Inc.
5  */
6 
7 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
8 
9 #include <linux/fs.h>
10 #include <linux/dlm.h>
11 #include <linux/slab.h>
12 #include <linux/types.h>
13 #include <linux/delay.h>
14 #include <linux/gfs2_ondisk.h>
15 #include <linux/sched/signal.h>
16 
17 #include "incore.h"
18 #include "glock.h"
19 #include "glops.h"
20 #include "recovery.h"
21 #include "util.h"
22 #include "sys.h"
23 #include "trace_gfs2.h"
24 
25 /**
26  * gfs2_update_stats - Update time based stats
27  * @mv: Pointer to mean/variance structure to update
28  * @sample: New data to include
29  *
30  * @delta is the difference between the current rtt sample and the
31  * running average srtt. We add 1/8 of that to the srtt in order to
32  * update the current srtt estimate. The variance estimate is a bit
33  * more complicated. We subtract the current variance estimate from
34  * the abs value of the @delta and add 1/4 of that to the running
35  * total.  That's equivalent to 3/4 of the current variance
36  * estimate plus 1/4 of the abs of @delta.
37  *
38  * Note that the index points at the array entry containing the smoothed
39  * mean value, and the variance is always in the following entry
40  *
41  * Reference: TCP/IP Illustrated, vol 2, p. 831,832
42  * All times are in units of integer nanoseconds. Unlike the TCP/IP case,
43  * they are not scaled fixed point.
44  */
45 
46 static inline void gfs2_update_stats(struct gfs2_lkstats *s, unsigned index,
47 				     s64 sample)
48 {
49 	s64 delta = sample - s->stats[index];
50 	s->stats[index] += (delta >> 3);
51 	index++;
52 	s->stats[index] += (s64)(abs(delta) - s->stats[index]) >> 2;
53 }
54 
55 /**
56  * gfs2_update_reply_times - Update locking statistics
57  * @gl: The glock to update
58  *
59  * This assumes that gl->gl_dstamp has been set earlier.
60  *
61  * The rtt (lock round trip time) is an estimate of the time
62  * taken to perform a dlm lock request. We update it on each
63  * reply from the dlm.
64  *
65  * The blocking flag is set on the glock for all dlm requests
66  * which may potentially block due to lock requests from other nodes.
67  * DLM requests where the current lock state is exclusive, the
68  * requested state is null (or unlocked) or where the TRY or
69  * TRY_1CB flags are set are classified as non-blocking. All
70  * other DLM requests are counted as (potentially) blocking.
71  */
72 static inline void gfs2_update_reply_times(struct gfs2_glock *gl)
73 {
74 	struct gfs2_pcpu_lkstats *lks;
75 	const unsigned gltype = gl->gl_name.ln_type;
76 	unsigned index = test_bit(GLF_BLOCKING, &gl->gl_flags) ?
77 			 GFS2_LKS_SRTTB : GFS2_LKS_SRTT;
78 	s64 rtt;
79 
80 	preempt_disable();
81 	rtt = ktime_to_ns(ktime_sub(ktime_get_real(), gl->gl_dstamp));
82 	lks = this_cpu_ptr(gl->gl_name.ln_sbd->sd_lkstats);
83 	gfs2_update_stats(&gl->gl_stats, index, rtt);		/* Local */
84 	gfs2_update_stats(&lks->lkstats[gltype], index, rtt);	/* Global */
85 	preempt_enable();
86 
87 	trace_gfs2_glock_lock_time(gl, rtt);
88 }
89 
90 /**
91  * gfs2_update_request_times - Update locking statistics
92  * @gl: The glock to update
93  *
94  * The irt (lock inter-request times) measures the average time
95  * between requests to the dlm. It is updated immediately before
96  * each dlm call.
97  */
98 
99 static inline void gfs2_update_request_times(struct gfs2_glock *gl)
100 {
101 	struct gfs2_pcpu_lkstats *lks;
102 	const unsigned gltype = gl->gl_name.ln_type;
103 	ktime_t dstamp;
104 	s64 irt;
105 
106 	preempt_disable();
107 	dstamp = gl->gl_dstamp;
108 	gl->gl_dstamp = ktime_get_real();
109 	irt = ktime_to_ns(ktime_sub(gl->gl_dstamp, dstamp));
110 	lks = this_cpu_ptr(gl->gl_name.ln_sbd->sd_lkstats);
111 	gfs2_update_stats(&gl->gl_stats, GFS2_LKS_SIRT, irt);		/* Local */
112 	gfs2_update_stats(&lks->lkstats[gltype], GFS2_LKS_SIRT, irt);	/* Global */
113 	preempt_enable();
114 }
115 
116 static void gdlm_ast(void *arg)
117 {
118 	struct gfs2_glock *gl = arg;
119 	unsigned ret = gl->gl_state;
120 
121 	gfs2_update_reply_times(gl);
122 	BUG_ON(gl->gl_lksb.sb_flags & DLM_SBF_DEMOTED);
123 
124 	if ((gl->gl_lksb.sb_flags & DLM_SBF_VALNOTVALID) && gl->gl_lksb.sb_lvbptr)
125 		memset(gl->gl_lksb.sb_lvbptr, 0, GDLM_LVB_SIZE);
126 
127 	switch (gl->gl_lksb.sb_status) {
128 	case -DLM_EUNLOCK: /* Unlocked, so glock can be freed */
129 		if (gl->gl_ops->go_free)
130 			gl->gl_ops->go_free(gl);
131 		gfs2_glock_free(gl);
132 		return;
133 	case -DLM_ECANCEL: /* Cancel while getting lock */
134 		ret |= LM_OUT_CANCELED;
135 		goto out;
136 	case -EAGAIN: /* Try lock fails */
137 	case -EDEADLK: /* Deadlock detected */
138 		goto out;
139 	case -ETIMEDOUT: /* Canceled due to timeout */
140 		ret |= LM_OUT_ERROR;
141 		goto out;
142 	case 0: /* Success */
143 		break;
144 	default: /* Something unexpected */
145 		BUG();
146 	}
147 
148 	ret = gl->gl_req;
149 	if (gl->gl_lksb.sb_flags & DLM_SBF_ALTMODE) {
150 		if (gl->gl_req == LM_ST_SHARED)
151 			ret = LM_ST_DEFERRED;
152 		else if (gl->gl_req == LM_ST_DEFERRED)
153 			ret = LM_ST_SHARED;
154 		else
155 			BUG();
156 	}
157 
158 	set_bit(GLF_INITIAL, &gl->gl_flags);
159 	gfs2_glock_complete(gl, ret);
160 	return;
161 out:
162 	if (!test_bit(GLF_INITIAL, &gl->gl_flags))
163 		gl->gl_lksb.sb_lkid = 0;
164 	gfs2_glock_complete(gl, ret);
165 }
166 
167 static void gdlm_bast(void *arg, int mode)
168 {
169 	struct gfs2_glock *gl = arg;
170 
171 	switch (mode) {
172 	case DLM_LOCK_EX:
173 		gfs2_glock_cb(gl, LM_ST_UNLOCKED);
174 		break;
175 	case DLM_LOCK_CW:
176 		gfs2_glock_cb(gl, LM_ST_DEFERRED);
177 		break;
178 	case DLM_LOCK_PR:
179 		gfs2_glock_cb(gl, LM_ST_SHARED);
180 		break;
181 	default:
182 		fs_err(gl->gl_name.ln_sbd, "unknown bast mode %d\n", mode);
183 		BUG();
184 	}
185 }
186 
187 /* convert gfs lock-state to dlm lock-mode */
188 
189 static int make_mode(struct gfs2_sbd *sdp, const unsigned int lmstate)
190 {
191 	switch (lmstate) {
192 	case LM_ST_UNLOCKED:
193 		return DLM_LOCK_NL;
194 	case LM_ST_EXCLUSIVE:
195 		return DLM_LOCK_EX;
196 	case LM_ST_DEFERRED:
197 		return DLM_LOCK_CW;
198 	case LM_ST_SHARED:
199 		return DLM_LOCK_PR;
200 	}
201 	fs_err(sdp, "unknown LM state %d\n", lmstate);
202 	BUG();
203 	return -1;
204 }
205 
206 static u32 make_flags(struct gfs2_glock *gl, const unsigned int gfs_flags,
207 		      const int req)
208 {
209 	u32 lkf = 0;
210 
211 	if (gl->gl_lksb.sb_lvbptr)
212 		lkf |= DLM_LKF_VALBLK;
213 
214 	if (gfs_flags & LM_FLAG_TRY)
215 		lkf |= DLM_LKF_NOQUEUE;
216 
217 	if (gfs_flags & LM_FLAG_TRY_1CB) {
218 		lkf |= DLM_LKF_NOQUEUE;
219 		lkf |= DLM_LKF_NOQUEUEBAST;
220 	}
221 
222 	if (gfs_flags & LM_FLAG_PRIORITY) {
223 		lkf |= DLM_LKF_NOORDER;
224 		lkf |= DLM_LKF_HEADQUE;
225 	}
226 
227 	if (gfs_flags & LM_FLAG_ANY) {
228 		if (req == DLM_LOCK_PR)
229 			lkf |= DLM_LKF_ALTCW;
230 		else if (req == DLM_LOCK_CW)
231 			lkf |= DLM_LKF_ALTPR;
232 		else
233 			BUG();
234 	}
235 
236 	if (gl->gl_lksb.sb_lkid != 0) {
237 		lkf |= DLM_LKF_CONVERT;
238 		if (test_bit(GLF_BLOCKING, &gl->gl_flags))
239 			lkf |= DLM_LKF_QUECVT;
240 	}
241 
242 	return lkf;
243 }
244 
245 static void gfs2_reverse_hex(char *c, u64 value)
246 {
247 	*c = '0';
248 	while (value) {
249 		*c-- = hex_asc[value & 0x0f];
250 		value >>= 4;
251 	}
252 }
253 
254 static int gdlm_lock(struct gfs2_glock *gl, unsigned int req_state,
255 		     unsigned int flags)
256 {
257 	struct lm_lockstruct *ls = &gl->gl_name.ln_sbd->sd_lockstruct;
258 	int req;
259 	u32 lkf;
260 	char strname[GDLM_STRNAME_BYTES] = "";
261 
262 	req = make_mode(gl->gl_name.ln_sbd, req_state);
263 	lkf = make_flags(gl, flags, req);
264 	gfs2_glstats_inc(gl, GFS2_LKS_DCOUNT);
265 	gfs2_sbstats_inc(gl, GFS2_LKS_DCOUNT);
266 	if (gl->gl_lksb.sb_lkid) {
267 		gfs2_update_request_times(gl);
268 	} else {
269 		memset(strname, ' ', GDLM_STRNAME_BYTES - 1);
270 		strname[GDLM_STRNAME_BYTES - 1] = '\0';
271 		gfs2_reverse_hex(strname + 7, gl->gl_name.ln_type);
272 		gfs2_reverse_hex(strname + 23, gl->gl_name.ln_number);
273 		gl->gl_dstamp = ktime_get_real();
274 	}
275 	/*
276 	 * Submit the actual lock request.
277 	 */
278 
279 	return dlm_lock(ls->ls_dlm, req, &gl->gl_lksb, lkf, strname,
280 			GDLM_STRNAME_BYTES - 1, 0, gdlm_ast, gl, gdlm_bast);
281 }
282 
283 static void gdlm_put_lock(struct gfs2_glock *gl)
284 {
285 	struct gfs2_sbd *sdp = gl->gl_name.ln_sbd;
286 	struct lm_lockstruct *ls = &sdp->sd_lockstruct;
287 	int lvb_needs_unlock = 0;
288 	int error;
289 
290 	if (gl->gl_lksb.sb_lkid == 0) {
291 		gfs2_glock_free(gl);
292 		return;
293 	}
294 
295 	clear_bit(GLF_BLOCKING, &gl->gl_flags);
296 	gfs2_glstats_inc(gl, GFS2_LKS_DCOUNT);
297 	gfs2_sbstats_inc(gl, GFS2_LKS_DCOUNT);
298 	gfs2_update_request_times(gl);
299 
300 	/* don't want to skip dlm_unlock writing the lvb when lock is ex */
301 
302 	if (gl->gl_lksb.sb_lvbptr && (gl->gl_state == LM_ST_EXCLUSIVE))
303 		lvb_needs_unlock = 1;
304 
305 	if (test_bit(SDF_SKIP_DLM_UNLOCK, &sdp->sd_flags) &&
306 	    !lvb_needs_unlock) {
307 		gfs2_glock_free(gl);
308 		return;
309 	}
310 
311 	error = dlm_unlock(ls->ls_dlm, gl->gl_lksb.sb_lkid, DLM_LKF_VALBLK,
312 			   NULL, gl);
313 	if (error) {
314 		fs_err(sdp, "gdlm_unlock %x,%llx err=%d\n",
315 		       gl->gl_name.ln_type,
316 		       (unsigned long long)gl->gl_name.ln_number, error);
317 		return;
318 	}
319 }
320 
321 static void gdlm_cancel(struct gfs2_glock *gl)
322 {
323 	struct lm_lockstruct *ls = &gl->gl_name.ln_sbd->sd_lockstruct;
324 	dlm_unlock(ls->ls_dlm, gl->gl_lksb.sb_lkid, DLM_LKF_CANCEL, NULL, gl);
325 }
326 
327 /*
328  * dlm/gfs2 recovery coordination using dlm_recover callbacks
329  *
330  *  0. gfs2 checks for another cluster node withdraw, needing journal replay
331  *  1. dlm_controld sees lockspace members change
332  *  2. dlm_controld blocks dlm-kernel locking activity
333  *  3. dlm_controld within dlm-kernel notifies gfs2 (recover_prep)
334  *  4. dlm_controld starts and finishes its own user level recovery
335  *  5. dlm_controld starts dlm-kernel dlm_recoverd to do kernel recovery
336  *  6. dlm_recoverd notifies gfs2 of failed nodes (recover_slot)
337  *  7. dlm_recoverd does its own lock recovery
338  *  8. dlm_recoverd unblocks dlm-kernel locking activity
339  *  9. dlm_recoverd notifies gfs2 when done (recover_done with new generation)
340  * 10. gfs2_control updates control_lock lvb with new generation and jid bits
341  * 11. gfs2_control enqueues journals for gfs2_recover to recover (maybe none)
342  * 12. gfs2_recover dequeues and recovers journals of failed nodes
343  * 13. gfs2_recover provides recovery results to gfs2_control (recovery_result)
344  * 14. gfs2_control updates control_lock lvb jid bits for recovered journals
345  * 15. gfs2_control unblocks normal locking when all journals are recovered
346  *
347  * - failures during recovery
348  *
349  * recover_prep() may set BLOCK_LOCKS (step 3) again before gfs2_control
350  * clears BLOCK_LOCKS (step 15), e.g. another node fails while still
351  * recovering for a prior failure.  gfs2_control needs a way to detect
352  * this so it can leave BLOCK_LOCKS set in step 15.  This is managed using
353  * the recover_block and recover_start values.
354  *
355  * recover_done() provides a new lockspace generation number each time it
356  * is called (step 9).  This generation number is saved as recover_start.
357  * When recover_prep() is called, it sets BLOCK_LOCKS and sets
358  * recover_block = recover_start.  So, while recover_block is equal to
359  * recover_start, BLOCK_LOCKS should remain set.  (recover_spin must
360  * be held around the BLOCK_LOCKS/recover_block/recover_start logic.)
361  *
362  * - more specific gfs2 steps in sequence above
363  *
364  *  3. recover_prep sets BLOCK_LOCKS and sets recover_block = recover_start
365  *  6. recover_slot records any failed jids (maybe none)
366  *  9. recover_done sets recover_start = new generation number
367  * 10. gfs2_control sets control_lock lvb = new gen + bits for failed jids
368  * 12. gfs2_recover does journal recoveries for failed jids identified above
369  * 14. gfs2_control clears control_lock lvb bits for recovered jids
370  * 15. gfs2_control checks if recover_block == recover_start (step 3 occured
371  *     again) then do nothing, otherwise if recover_start > recover_block
372  *     then clear BLOCK_LOCKS.
373  *
374  * - parallel recovery steps across all nodes
375  *
376  * All nodes attempt to update the control_lock lvb with the new generation
377  * number and jid bits, but only the first to get the control_lock EX will
378  * do so; others will see that it's already done (lvb already contains new
379  * generation number.)
380  *
381  * . All nodes get the same recover_prep/recover_slot/recover_done callbacks
382  * . All nodes attempt to set control_lock lvb gen + bits for the new gen
383  * . One node gets control_lock first and writes the lvb, others see it's done
384  * . All nodes attempt to recover jids for which they see control_lock bits set
385  * . One node succeeds for a jid, and that one clears the jid bit in the lvb
386  * . All nodes will eventually see all lvb bits clear and unblock locks
387  *
388  * - is there a problem with clearing an lvb bit that should be set
389  *   and missing a journal recovery?
390  *
391  * 1. jid fails
392  * 2. lvb bit set for step 1
393  * 3. jid recovered for step 1
394  * 4. jid taken again (new mount)
395  * 5. jid fails (for step 4)
396  * 6. lvb bit set for step 5 (will already be set)
397  * 7. lvb bit cleared for step 3
398  *
399  * This is not a problem because the failure in step 5 does not
400  * require recovery, because the mount in step 4 could not have
401  * progressed far enough to unblock locks and access the fs.  The
402  * control_mount() function waits for all recoveries to be complete
403  * for the latest lockspace generation before ever unblocking locks
404  * and returning.  The mount in step 4 waits until the recovery in
405  * step 1 is done.
406  *
407  * - special case of first mounter: first node to mount the fs
408  *
409  * The first node to mount a gfs2 fs needs to check all the journals
410  * and recover any that need recovery before other nodes are allowed
411  * to mount the fs.  (Others may begin mounting, but they must wait
412  * for the first mounter to be done before taking locks on the fs
413  * or accessing the fs.)  This has two parts:
414  *
415  * 1. The mounted_lock tells a node it's the first to mount the fs.
416  * Each node holds the mounted_lock in PR while it's mounted.
417  * Each node tries to acquire the mounted_lock in EX when it mounts.
418  * If a node is granted the mounted_lock EX it means there are no
419  * other mounted nodes (no PR locks exist), and it is the first mounter.
420  * The mounted_lock is demoted to PR when first recovery is done, so
421  * others will fail to get an EX lock, but will get a PR lock.
422  *
423  * 2. The control_lock blocks others in control_mount() while the first
424  * mounter is doing first mount recovery of all journals.
425  * A mounting node needs to acquire control_lock in EX mode before
426  * it can proceed.  The first mounter holds control_lock in EX while doing
427  * the first mount recovery, blocking mounts from other nodes, then demotes
428  * control_lock to NL when it's done (others_may_mount/first_done),
429  * allowing other nodes to continue mounting.
430  *
431  * first mounter:
432  * control_lock EX/NOQUEUE success
433  * mounted_lock EX/NOQUEUE success (no other PR, so no other mounters)
434  * set first=1
435  * do first mounter recovery
436  * mounted_lock EX->PR
437  * control_lock EX->NL, write lvb generation
438  *
439  * other mounter:
440  * control_lock EX/NOQUEUE success (if fail -EAGAIN, retry)
441  * mounted_lock EX/NOQUEUE fail -EAGAIN (expected due to other mounters PR)
442  * mounted_lock PR/NOQUEUE success
443  * read lvb generation
444  * control_lock EX->NL
445  * set first=0
446  *
447  * - mount during recovery
448  *
449  * If a node mounts while others are doing recovery (not first mounter),
450  * the mounting node will get its initial recover_done() callback without
451  * having seen any previous failures/callbacks.
452  *
453  * It must wait for all recoveries preceding its mount to be finished
454  * before it unblocks locks.  It does this by repeating the "other mounter"
455  * steps above until the lvb generation number is >= its mount generation
456  * number (from initial recover_done) and all lvb bits are clear.
457  *
458  * - control_lock lvb format
459  *
460  * 4 bytes generation number: the latest dlm lockspace generation number
461  * from recover_done callback.  Indicates the jid bitmap has been updated
462  * to reflect all slot failures through that generation.
463  * 4 bytes unused.
464  * GDLM_LVB_SIZE-8 bytes of jid bit map. If bit N is set, it indicates
465  * that jid N needs recovery.
466  */
467 
468 #define JID_BITMAP_OFFSET 8 /* 4 byte generation number + 4 byte unused */
469 
470 static void control_lvb_read(struct lm_lockstruct *ls, uint32_t *lvb_gen,
471 			     char *lvb_bits)
472 {
473 	__le32 gen;
474 	memcpy(lvb_bits, ls->ls_control_lvb, GDLM_LVB_SIZE);
475 	memcpy(&gen, lvb_bits, sizeof(__le32));
476 	*lvb_gen = le32_to_cpu(gen);
477 }
478 
479 static void control_lvb_write(struct lm_lockstruct *ls, uint32_t lvb_gen,
480 			      char *lvb_bits)
481 {
482 	__le32 gen;
483 	memcpy(ls->ls_control_lvb, lvb_bits, GDLM_LVB_SIZE);
484 	gen = cpu_to_le32(lvb_gen);
485 	memcpy(ls->ls_control_lvb, &gen, sizeof(__le32));
486 }
487 
488 static int all_jid_bits_clear(char *lvb)
489 {
490 	return !memchr_inv(lvb + JID_BITMAP_OFFSET, 0,
491 			GDLM_LVB_SIZE - JID_BITMAP_OFFSET);
492 }
493 
494 static void sync_wait_cb(void *arg)
495 {
496 	struct lm_lockstruct *ls = arg;
497 	complete(&ls->ls_sync_wait);
498 }
499 
500 static int sync_unlock(struct gfs2_sbd *sdp, struct dlm_lksb *lksb, char *name)
501 {
502 	struct lm_lockstruct *ls = &sdp->sd_lockstruct;
503 	int error;
504 
505 	error = dlm_unlock(ls->ls_dlm, lksb->sb_lkid, 0, lksb, ls);
506 	if (error) {
507 		fs_err(sdp, "%s lkid %x error %d\n",
508 		       name, lksb->sb_lkid, error);
509 		return error;
510 	}
511 
512 	wait_for_completion(&ls->ls_sync_wait);
513 
514 	if (lksb->sb_status != -DLM_EUNLOCK) {
515 		fs_err(sdp, "%s lkid %x status %d\n",
516 		       name, lksb->sb_lkid, lksb->sb_status);
517 		return -1;
518 	}
519 	return 0;
520 }
521 
522 static int sync_lock(struct gfs2_sbd *sdp, int mode, uint32_t flags,
523 		     unsigned int num, struct dlm_lksb *lksb, char *name)
524 {
525 	struct lm_lockstruct *ls = &sdp->sd_lockstruct;
526 	char strname[GDLM_STRNAME_BYTES];
527 	int error, status;
528 
529 	memset(strname, 0, GDLM_STRNAME_BYTES);
530 	snprintf(strname, GDLM_STRNAME_BYTES, "%8x%16x", LM_TYPE_NONDISK, num);
531 
532 	error = dlm_lock(ls->ls_dlm, mode, lksb, flags,
533 			 strname, GDLM_STRNAME_BYTES - 1,
534 			 0, sync_wait_cb, ls, NULL);
535 	if (error) {
536 		fs_err(sdp, "%s lkid %x flags %x mode %d error %d\n",
537 		       name, lksb->sb_lkid, flags, mode, error);
538 		return error;
539 	}
540 
541 	wait_for_completion(&ls->ls_sync_wait);
542 
543 	status = lksb->sb_status;
544 
545 	if (status && status != -EAGAIN) {
546 		fs_err(sdp, "%s lkid %x flags %x mode %d status %d\n",
547 		       name, lksb->sb_lkid, flags, mode, status);
548 	}
549 
550 	return status;
551 }
552 
553 static int mounted_unlock(struct gfs2_sbd *sdp)
554 {
555 	struct lm_lockstruct *ls = &sdp->sd_lockstruct;
556 	return sync_unlock(sdp, &ls->ls_mounted_lksb, "mounted_lock");
557 }
558 
559 static int mounted_lock(struct gfs2_sbd *sdp, int mode, uint32_t flags)
560 {
561 	struct lm_lockstruct *ls = &sdp->sd_lockstruct;
562 	return sync_lock(sdp, mode, flags, GFS2_MOUNTED_LOCK,
563 			 &ls->ls_mounted_lksb, "mounted_lock");
564 }
565 
566 static int control_unlock(struct gfs2_sbd *sdp)
567 {
568 	struct lm_lockstruct *ls = &sdp->sd_lockstruct;
569 	return sync_unlock(sdp, &ls->ls_control_lksb, "control_lock");
570 }
571 
572 static int control_lock(struct gfs2_sbd *sdp, int mode, uint32_t flags)
573 {
574 	struct lm_lockstruct *ls = &sdp->sd_lockstruct;
575 	return sync_lock(sdp, mode, flags, GFS2_CONTROL_LOCK,
576 			 &ls->ls_control_lksb, "control_lock");
577 }
578 
579 /**
580  * remote_withdraw - react to a node withdrawing from the file system
581  * @sdp: The superblock
582  */
583 static void remote_withdraw(struct gfs2_sbd *sdp)
584 {
585 	struct gfs2_jdesc *jd;
586 	int ret = 0, count = 0;
587 
588 	list_for_each_entry(jd, &sdp->sd_jindex_list, jd_list) {
589 		if (jd->jd_jid == sdp->sd_lockstruct.ls_jid)
590 			continue;
591 		ret = gfs2_recover_journal(jd, true);
592 		if (ret)
593 			break;
594 		count++;
595 	}
596 
597 	/* Now drop the additional reference we acquired */
598 	fs_err(sdp, "Journals checked: %d, ret = %d.\n", count, ret);
599 }
600 
601 static void gfs2_control_func(struct work_struct *work)
602 {
603 	struct gfs2_sbd *sdp = container_of(work, struct gfs2_sbd, sd_control_work.work);
604 	struct lm_lockstruct *ls = &sdp->sd_lockstruct;
605 	uint32_t block_gen, start_gen, lvb_gen, flags;
606 	int recover_set = 0;
607 	int write_lvb = 0;
608 	int recover_size;
609 	int i, error;
610 
611 	/* First check for other nodes that may have done a withdraw. */
612 	if (test_bit(SDF_REMOTE_WITHDRAW, &sdp->sd_flags)) {
613 		remote_withdraw(sdp);
614 		clear_bit(SDF_REMOTE_WITHDRAW, &sdp->sd_flags);
615 		return;
616 	}
617 
618 	spin_lock(&ls->ls_recover_spin);
619 	/*
620 	 * No MOUNT_DONE means we're still mounting; control_mount()
621 	 * will set this flag, after which this thread will take over
622 	 * all further clearing of BLOCK_LOCKS.
623 	 *
624 	 * FIRST_MOUNT means this node is doing first mounter recovery,
625 	 * for which recovery control is handled by
626 	 * control_mount()/control_first_done(), not this thread.
627 	 */
628 	if (!test_bit(DFL_MOUNT_DONE, &ls->ls_recover_flags) ||
629 	     test_bit(DFL_FIRST_MOUNT, &ls->ls_recover_flags)) {
630 		spin_unlock(&ls->ls_recover_spin);
631 		return;
632 	}
633 	block_gen = ls->ls_recover_block;
634 	start_gen = ls->ls_recover_start;
635 	spin_unlock(&ls->ls_recover_spin);
636 
637 	/*
638 	 * Equal block_gen and start_gen implies we are between
639 	 * recover_prep and recover_done callbacks, which means
640 	 * dlm recovery is in progress and dlm locking is blocked.
641 	 * There's no point trying to do any work until recover_done.
642 	 */
643 
644 	if (block_gen == start_gen)
645 		return;
646 
647 	/*
648 	 * Propagate recover_submit[] and recover_result[] to lvb:
649 	 * dlm_recoverd adds to recover_submit[] jids needing recovery
650 	 * gfs2_recover adds to recover_result[] journal recovery results
651 	 *
652 	 * set lvb bit for jids in recover_submit[] if the lvb has not
653 	 * yet been updated for the generation of the failure
654 	 *
655 	 * clear lvb bit for jids in recover_result[] if the result of
656 	 * the journal recovery is SUCCESS
657 	 */
658 
659 	error = control_lock(sdp, DLM_LOCK_EX, DLM_LKF_CONVERT|DLM_LKF_VALBLK);
660 	if (error) {
661 		fs_err(sdp, "control lock EX error %d\n", error);
662 		return;
663 	}
664 
665 	control_lvb_read(ls, &lvb_gen, ls->ls_lvb_bits);
666 
667 	spin_lock(&ls->ls_recover_spin);
668 	if (block_gen != ls->ls_recover_block ||
669 	    start_gen != ls->ls_recover_start) {
670 		fs_info(sdp, "recover generation %u block1 %u %u\n",
671 			start_gen, block_gen, ls->ls_recover_block);
672 		spin_unlock(&ls->ls_recover_spin);
673 		control_lock(sdp, DLM_LOCK_NL, DLM_LKF_CONVERT);
674 		return;
675 	}
676 
677 	recover_size = ls->ls_recover_size;
678 
679 	if (lvb_gen <= start_gen) {
680 		/*
681 		 * Clear lvb bits for jids we've successfully recovered.
682 		 * Because all nodes attempt to recover failed journals,
683 		 * a journal can be recovered multiple times successfully
684 		 * in succession.  Only the first will really do recovery,
685 		 * the others find it clean, but still report a successful
686 		 * recovery.  So, another node may have already recovered
687 		 * the jid and cleared the lvb bit for it.
688 		 */
689 		for (i = 0; i < recover_size; i++) {
690 			if (ls->ls_recover_result[i] != LM_RD_SUCCESS)
691 				continue;
692 
693 			ls->ls_recover_result[i] = 0;
694 
695 			if (!test_bit_le(i, ls->ls_lvb_bits + JID_BITMAP_OFFSET))
696 				continue;
697 
698 			__clear_bit_le(i, ls->ls_lvb_bits + JID_BITMAP_OFFSET);
699 			write_lvb = 1;
700 		}
701 	}
702 
703 	if (lvb_gen == start_gen) {
704 		/*
705 		 * Failed slots before start_gen are already set in lvb.
706 		 */
707 		for (i = 0; i < recover_size; i++) {
708 			if (!ls->ls_recover_submit[i])
709 				continue;
710 			if (ls->ls_recover_submit[i] < lvb_gen)
711 				ls->ls_recover_submit[i] = 0;
712 		}
713 	} else if (lvb_gen < start_gen) {
714 		/*
715 		 * Failed slots before start_gen are not yet set in lvb.
716 		 */
717 		for (i = 0; i < recover_size; i++) {
718 			if (!ls->ls_recover_submit[i])
719 				continue;
720 			if (ls->ls_recover_submit[i] < start_gen) {
721 				ls->ls_recover_submit[i] = 0;
722 				__set_bit_le(i, ls->ls_lvb_bits + JID_BITMAP_OFFSET);
723 			}
724 		}
725 		/* even if there are no bits to set, we need to write the
726 		   latest generation to the lvb */
727 		write_lvb = 1;
728 	} else {
729 		/*
730 		 * we should be getting a recover_done() for lvb_gen soon
731 		 */
732 	}
733 	spin_unlock(&ls->ls_recover_spin);
734 
735 	if (write_lvb) {
736 		control_lvb_write(ls, start_gen, ls->ls_lvb_bits);
737 		flags = DLM_LKF_CONVERT | DLM_LKF_VALBLK;
738 	} else {
739 		flags = DLM_LKF_CONVERT;
740 	}
741 
742 	error = control_lock(sdp, DLM_LOCK_NL, flags);
743 	if (error) {
744 		fs_err(sdp, "control lock NL error %d\n", error);
745 		return;
746 	}
747 
748 	/*
749 	 * Everyone will see jid bits set in the lvb, run gfs2_recover_set(),
750 	 * and clear a jid bit in the lvb if the recovery is a success.
751 	 * Eventually all journals will be recovered, all jid bits will
752 	 * be cleared in the lvb, and everyone will clear BLOCK_LOCKS.
753 	 */
754 
755 	for (i = 0; i < recover_size; i++) {
756 		if (test_bit_le(i, ls->ls_lvb_bits + JID_BITMAP_OFFSET)) {
757 			fs_info(sdp, "recover generation %u jid %d\n",
758 				start_gen, i);
759 			gfs2_recover_set(sdp, i);
760 			recover_set++;
761 		}
762 	}
763 	if (recover_set)
764 		return;
765 
766 	/*
767 	 * No more jid bits set in lvb, all recovery is done, unblock locks
768 	 * (unless a new recover_prep callback has occured blocking locks
769 	 * again while working above)
770 	 */
771 
772 	spin_lock(&ls->ls_recover_spin);
773 	if (ls->ls_recover_block == block_gen &&
774 	    ls->ls_recover_start == start_gen) {
775 		clear_bit(DFL_BLOCK_LOCKS, &ls->ls_recover_flags);
776 		spin_unlock(&ls->ls_recover_spin);
777 		fs_info(sdp, "recover generation %u done\n", start_gen);
778 		gfs2_glock_thaw(sdp);
779 	} else {
780 		fs_info(sdp, "recover generation %u block2 %u %u\n",
781 			start_gen, block_gen, ls->ls_recover_block);
782 		spin_unlock(&ls->ls_recover_spin);
783 	}
784 }
785 
786 static int control_mount(struct gfs2_sbd *sdp)
787 {
788 	struct lm_lockstruct *ls = &sdp->sd_lockstruct;
789 	uint32_t start_gen, block_gen, mount_gen, lvb_gen;
790 	int mounted_mode;
791 	int retries = 0;
792 	int error;
793 
794 	memset(&ls->ls_mounted_lksb, 0, sizeof(struct dlm_lksb));
795 	memset(&ls->ls_control_lksb, 0, sizeof(struct dlm_lksb));
796 	memset(&ls->ls_control_lvb, 0, GDLM_LVB_SIZE);
797 	ls->ls_control_lksb.sb_lvbptr = ls->ls_control_lvb;
798 	init_completion(&ls->ls_sync_wait);
799 
800 	set_bit(DFL_BLOCK_LOCKS, &ls->ls_recover_flags);
801 
802 	error = control_lock(sdp, DLM_LOCK_NL, DLM_LKF_VALBLK);
803 	if (error) {
804 		fs_err(sdp, "control_mount control_lock NL error %d\n", error);
805 		return error;
806 	}
807 
808 	error = mounted_lock(sdp, DLM_LOCK_NL, 0);
809 	if (error) {
810 		fs_err(sdp, "control_mount mounted_lock NL error %d\n", error);
811 		control_unlock(sdp);
812 		return error;
813 	}
814 	mounted_mode = DLM_LOCK_NL;
815 
816 restart:
817 	if (retries++ && signal_pending(current)) {
818 		error = -EINTR;
819 		goto fail;
820 	}
821 
822 	/*
823 	 * We always start with both locks in NL. control_lock is
824 	 * demoted to NL below so we don't need to do it here.
825 	 */
826 
827 	if (mounted_mode != DLM_LOCK_NL) {
828 		error = mounted_lock(sdp, DLM_LOCK_NL, DLM_LKF_CONVERT);
829 		if (error)
830 			goto fail;
831 		mounted_mode = DLM_LOCK_NL;
832 	}
833 
834 	/*
835 	 * Other nodes need to do some work in dlm recovery and gfs2_control
836 	 * before the recover_done and control_lock will be ready for us below.
837 	 * A delay here is not required but often avoids having to retry.
838 	 */
839 
840 	msleep_interruptible(500);
841 
842 	/*
843 	 * Acquire control_lock in EX and mounted_lock in either EX or PR.
844 	 * control_lock lvb keeps track of any pending journal recoveries.
845 	 * mounted_lock indicates if any other nodes have the fs mounted.
846 	 */
847 
848 	error = control_lock(sdp, DLM_LOCK_EX, DLM_LKF_CONVERT|DLM_LKF_NOQUEUE|DLM_LKF_VALBLK);
849 	if (error == -EAGAIN) {
850 		goto restart;
851 	} else if (error) {
852 		fs_err(sdp, "control_mount control_lock EX error %d\n", error);
853 		goto fail;
854 	}
855 
856 	/**
857 	 * If we're a spectator, we don't want to take the lock in EX because
858 	 * we cannot do the first-mount responsibility it implies: recovery.
859 	 */
860 	if (sdp->sd_args.ar_spectator)
861 		goto locks_done;
862 
863 	error = mounted_lock(sdp, DLM_LOCK_EX, DLM_LKF_CONVERT|DLM_LKF_NOQUEUE);
864 	if (!error) {
865 		mounted_mode = DLM_LOCK_EX;
866 		goto locks_done;
867 	} else if (error != -EAGAIN) {
868 		fs_err(sdp, "control_mount mounted_lock EX error %d\n", error);
869 		goto fail;
870 	}
871 
872 	error = mounted_lock(sdp, DLM_LOCK_PR, DLM_LKF_CONVERT|DLM_LKF_NOQUEUE);
873 	if (!error) {
874 		mounted_mode = DLM_LOCK_PR;
875 		goto locks_done;
876 	} else {
877 		/* not even -EAGAIN should happen here */
878 		fs_err(sdp, "control_mount mounted_lock PR error %d\n", error);
879 		goto fail;
880 	}
881 
882 locks_done:
883 	/*
884 	 * If we got both locks above in EX, then we're the first mounter.
885 	 * If not, then we need to wait for the control_lock lvb to be
886 	 * updated by other mounted nodes to reflect our mount generation.
887 	 *
888 	 * In simple first mounter cases, first mounter will see zero lvb_gen,
889 	 * but in cases where all existing nodes leave/fail before mounting
890 	 * nodes finish control_mount, then all nodes will be mounting and
891 	 * lvb_gen will be non-zero.
892 	 */
893 
894 	control_lvb_read(ls, &lvb_gen, ls->ls_lvb_bits);
895 
896 	if (lvb_gen == 0xFFFFFFFF) {
897 		/* special value to force mount attempts to fail */
898 		fs_err(sdp, "control_mount control_lock disabled\n");
899 		error = -EINVAL;
900 		goto fail;
901 	}
902 
903 	if (mounted_mode == DLM_LOCK_EX) {
904 		/* first mounter, keep both EX while doing first recovery */
905 		spin_lock(&ls->ls_recover_spin);
906 		clear_bit(DFL_BLOCK_LOCKS, &ls->ls_recover_flags);
907 		set_bit(DFL_MOUNT_DONE, &ls->ls_recover_flags);
908 		set_bit(DFL_FIRST_MOUNT, &ls->ls_recover_flags);
909 		spin_unlock(&ls->ls_recover_spin);
910 		fs_info(sdp, "first mounter control generation %u\n", lvb_gen);
911 		return 0;
912 	}
913 
914 	error = control_lock(sdp, DLM_LOCK_NL, DLM_LKF_CONVERT);
915 	if (error)
916 		goto fail;
917 
918 	/*
919 	 * We are not first mounter, now we need to wait for the control_lock
920 	 * lvb generation to be >= the generation from our first recover_done
921 	 * and all lvb bits to be clear (no pending journal recoveries.)
922 	 */
923 
924 	if (!all_jid_bits_clear(ls->ls_lvb_bits)) {
925 		/* journals need recovery, wait until all are clear */
926 		fs_info(sdp, "control_mount wait for journal recovery\n");
927 		goto restart;
928 	}
929 
930 	spin_lock(&ls->ls_recover_spin);
931 	block_gen = ls->ls_recover_block;
932 	start_gen = ls->ls_recover_start;
933 	mount_gen = ls->ls_recover_mount;
934 
935 	if (lvb_gen < mount_gen) {
936 		/* wait for mounted nodes to update control_lock lvb to our
937 		   generation, which might include new recovery bits set */
938 		if (sdp->sd_args.ar_spectator) {
939 			fs_info(sdp, "Recovery is required. Waiting for a "
940 				"non-spectator to mount.\n");
941 			msleep_interruptible(1000);
942 		} else {
943 			fs_info(sdp, "control_mount wait1 block %u start %u "
944 				"mount %u lvb %u flags %lx\n", block_gen,
945 				start_gen, mount_gen, lvb_gen,
946 				ls->ls_recover_flags);
947 		}
948 		spin_unlock(&ls->ls_recover_spin);
949 		goto restart;
950 	}
951 
952 	if (lvb_gen != start_gen) {
953 		/* wait for mounted nodes to update control_lock lvb to the
954 		   latest recovery generation */
955 		fs_info(sdp, "control_mount wait2 block %u start %u mount %u "
956 			"lvb %u flags %lx\n", block_gen, start_gen, mount_gen,
957 			lvb_gen, ls->ls_recover_flags);
958 		spin_unlock(&ls->ls_recover_spin);
959 		goto restart;
960 	}
961 
962 	if (block_gen == start_gen) {
963 		/* dlm recovery in progress, wait for it to finish */
964 		fs_info(sdp, "control_mount wait3 block %u start %u mount %u "
965 			"lvb %u flags %lx\n", block_gen, start_gen, mount_gen,
966 			lvb_gen, ls->ls_recover_flags);
967 		spin_unlock(&ls->ls_recover_spin);
968 		goto restart;
969 	}
970 
971 	clear_bit(DFL_BLOCK_LOCKS, &ls->ls_recover_flags);
972 	set_bit(DFL_MOUNT_DONE, &ls->ls_recover_flags);
973 	memset(ls->ls_recover_submit, 0, ls->ls_recover_size*sizeof(uint32_t));
974 	memset(ls->ls_recover_result, 0, ls->ls_recover_size*sizeof(uint32_t));
975 	spin_unlock(&ls->ls_recover_spin);
976 	return 0;
977 
978 fail:
979 	mounted_unlock(sdp);
980 	control_unlock(sdp);
981 	return error;
982 }
983 
984 static int control_first_done(struct gfs2_sbd *sdp)
985 {
986 	struct lm_lockstruct *ls = &sdp->sd_lockstruct;
987 	uint32_t start_gen, block_gen;
988 	int error;
989 
990 restart:
991 	spin_lock(&ls->ls_recover_spin);
992 	start_gen = ls->ls_recover_start;
993 	block_gen = ls->ls_recover_block;
994 
995 	if (test_bit(DFL_BLOCK_LOCKS, &ls->ls_recover_flags) ||
996 	    !test_bit(DFL_MOUNT_DONE, &ls->ls_recover_flags) ||
997 	    !test_bit(DFL_FIRST_MOUNT, &ls->ls_recover_flags)) {
998 		/* sanity check, should not happen */
999 		fs_err(sdp, "control_first_done start %u block %u flags %lx\n",
1000 		       start_gen, block_gen, ls->ls_recover_flags);
1001 		spin_unlock(&ls->ls_recover_spin);
1002 		control_unlock(sdp);
1003 		return -1;
1004 	}
1005 
1006 	if (start_gen == block_gen) {
1007 		/*
1008 		 * Wait for the end of a dlm recovery cycle to switch from
1009 		 * first mounter recovery.  We can ignore any recover_slot
1010 		 * callbacks between the recover_prep and next recover_done
1011 		 * because we are still the first mounter and any failed nodes
1012 		 * have not fully mounted, so they don't need recovery.
1013 		 */
1014 		spin_unlock(&ls->ls_recover_spin);
1015 		fs_info(sdp, "control_first_done wait gen %u\n", start_gen);
1016 
1017 		wait_on_bit(&ls->ls_recover_flags, DFL_DLM_RECOVERY,
1018 			    TASK_UNINTERRUPTIBLE);
1019 		goto restart;
1020 	}
1021 
1022 	clear_bit(DFL_FIRST_MOUNT, &ls->ls_recover_flags);
1023 	set_bit(DFL_FIRST_MOUNT_DONE, &ls->ls_recover_flags);
1024 	memset(ls->ls_recover_submit, 0, ls->ls_recover_size*sizeof(uint32_t));
1025 	memset(ls->ls_recover_result, 0, ls->ls_recover_size*sizeof(uint32_t));
1026 	spin_unlock(&ls->ls_recover_spin);
1027 
1028 	memset(ls->ls_lvb_bits, 0, GDLM_LVB_SIZE);
1029 	control_lvb_write(ls, start_gen, ls->ls_lvb_bits);
1030 
1031 	error = mounted_lock(sdp, DLM_LOCK_PR, DLM_LKF_CONVERT);
1032 	if (error)
1033 		fs_err(sdp, "control_first_done mounted PR error %d\n", error);
1034 
1035 	error = control_lock(sdp, DLM_LOCK_NL, DLM_LKF_CONVERT|DLM_LKF_VALBLK);
1036 	if (error)
1037 		fs_err(sdp, "control_first_done control NL error %d\n", error);
1038 
1039 	return error;
1040 }
1041 
1042 /*
1043  * Expand static jid arrays if necessary (by increments of RECOVER_SIZE_INC)
1044  * to accomodate the largest slot number.  (NB dlm slot numbers start at 1,
1045  * gfs2 jids start at 0, so jid = slot - 1)
1046  */
1047 
1048 #define RECOVER_SIZE_INC 16
1049 
1050 static int set_recover_size(struct gfs2_sbd *sdp, struct dlm_slot *slots,
1051 			    int num_slots)
1052 {
1053 	struct lm_lockstruct *ls = &sdp->sd_lockstruct;
1054 	uint32_t *submit = NULL;
1055 	uint32_t *result = NULL;
1056 	uint32_t old_size, new_size;
1057 	int i, max_jid;
1058 
1059 	if (!ls->ls_lvb_bits) {
1060 		ls->ls_lvb_bits = kzalloc(GDLM_LVB_SIZE, GFP_NOFS);
1061 		if (!ls->ls_lvb_bits)
1062 			return -ENOMEM;
1063 	}
1064 
1065 	max_jid = 0;
1066 	for (i = 0; i < num_slots; i++) {
1067 		if (max_jid < slots[i].slot - 1)
1068 			max_jid = slots[i].slot - 1;
1069 	}
1070 
1071 	old_size = ls->ls_recover_size;
1072 	new_size = old_size;
1073 	while (new_size < max_jid + 1)
1074 		new_size += RECOVER_SIZE_INC;
1075 	if (new_size == old_size)
1076 		return 0;
1077 
1078 	submit = kcalloc(new_size, sizeof(uint32_t), GFP_NOFS);
1079 	result = kcalloc(new_size, sizeof(uint32_t), GFP_NOFS);
1080 	if (!submit || !result) {
1081 		kfree(submit);
1082 		kfree(result);
1083 		return -ENOMEM;
1084 	}
1085 
1086 	spin_lock(&ls->ls_recover_spin);
1087 	memcpy(submit, ls->ls_recover_submit, old_size * sizeof(uint32_t));
1088 	memcpy(result, ls->ls_recover_result, old_size * sizeof(uint32_t));
1089 	kfree(ls->ls_recover_submit);
1090 	kfree(ls->ls_recover_result);
1091 	ls->ls_recover_submit = submit;
1092 	ls->ls_recover_result = result;
1093 	ls->ls_recover_size = new_size;
1094 	spin_unlock(&ls->ls_recover_spin);
1095 	return 0;
1096 }
1097 
1098 static void free_recover_size(struct lm_lockstruct *ls)
1099 {
1100 	kfree(ls->ls_lvb_bits);
1101 	kfree(ls->ls_recover_submit);
1102 	kfree(ls->ls_recover_result);
1103 	ls->ls_recover_submit = NULL;
1104 	ls->ls_recover_result = NULL;
1105 	ls->ls_recover_size = 0;
1106 	ls->ls_lvb_bits = NULL;
1107 }
1108 
1109 /* dlm calls before it does lock recovery */
1110 
1111 static void gdlm_recover_prep(void *arg)
1112 {
1113 	struct gfs2_sbd *sdp = arg;
1114 	struct lm_lockstruct *ls = &sdp->sd_lockstruct;
1115 
1116 	if (gfs2_withdrawn(sdp)) {
1117 		fs_err(sdp, "recover_prep ignored due to withdraw.\n");
1118 		return;
1119 	}
1120 	spin_lock(&ls->ls_recover_spin);
1121 	ls->ls_recover_block = ls->ls_recover_start;
1122 	set_bit(DFL_DLM_RECOVERY, &ls->ls_recover_flags);
1123 
1124 	if (!test_bit(DFL_MOUNT_DONE, &ls->ls_recover_flags) ||
1125 	     test_bit(DFL_FIRST_MOUNT, &ls->ls_recover_flags)) {
1126 		spin_unlock(&ls->ls_recover_spin);
1127 		return;
1128 	}
1129 	set_bit(DFL_BLOCK_LOCKS, &ls->ls_recover_flags);
1130 	spin_unlock(&ls->ls_recover_spin);
1131 }
1132 
1133 /* dlm calls after recover_prep has been completed on all lockspace members;
1134    identifies slot/jid of failed member */
1135 
1136 static void gdlm_recover_slot(void *arg, struct dlm_slot *slot)
1137 {
1138 	struct gfs2_sbd *sdp = arg;
1139 	struct lm_lockstruct *ls = &sdp->sd_lockstruct;
1140 	int jid = slot->slot - 1;
1141 
1142 	if (gfs2_withdrawn(sdp)) {
1143 		fs_err(sdp, "recover_slot jid %d ignored due to withdraw.\n",
1144 		       jid);
1145 		return;
1146 	}
1147 	spin_lock(&ls->ls_recover_spin);
1148 	if (ls->ls_recover_size < jid + 1) {
1149 		fs_err(sdp, "recover_slot jid %d gen %u short size %d\n",
1150 		       jid, ls->ls_recover_block, ls->ls_recover_size);
1151 		spin_unlock(&ls->ls_recover_spin);
1152 		return;
1153 	}
1154 
1155 	if (ls->ls_recover_submit[jid]) {
1156 		fs_info(sdp, "recover_slot jid %d gen %u prev %u\n",
1157 			jid, ls->ls_recover_block, ls->ls_recover_submit[jid]);
1158 	}
1159 	ls->ls_recover_submit[jid] = ls->ls_recover_block;
1160 	spin_unlock(&ls->ls_recover_spin);
1161 }
1162 
1163 /* dlm calls after recover_slot and after it completes lock recovery */
1164 
1165 static void gdlm_recover_done(void *arg, struct dlm_slot *slots, int num_slots,
1166 			      int our_slot, uint32_t generation)
1167 {
1168 	struct gfs2_sbd *sdp = arg;
1169 	struct lm_lockstruct *ls = &sdp->sd_lockstruct;
1170 
1171 	if (gfs2_withdrawn(sdp)) {
1172 		fs_err(sdp, "recover_done ignored due to withdraw.\n");
1173 		return;
1174 	}
1175 	/* ensure the ls jid arrays are large enough */
1176 	set_recover_size(sdp, slots, num_slots);
1177 
1178 	spin_lock(&ls->ls_recover_spin);
1179 	ls->ls_recover_start = generation;
1180 
1181 	if (!ls->ls_recover_mount) {
1182 		ls->ls_recover_mount = generation;
1183 		ls->ls_jid = our_slot - 1;
1184 	}
1185 
1186 	if (!test_bit(DFL_UNMOUNT, &ls->ls_recover_flags))
1187 		queue_delayed_work(gfs2_control_wq, &sdp->sd_control_work, 0);
1188 
1189 	clear_bit(DFL_DLM_RECOVERY, &ls->ls_recover_flags);
1190 	smp_mb__after_atomic();
1191 	wake_up_bit(&ls->ls_recover_flags, DFL_DLM_RECOVERY);
1192 	spin_unlock(&ls->ls_recover_spin);
1193 }
1194 
1195 /* gfs2_recover thread has a journal recovery result */
1196 
1197 static void gdlm_recovery_result(struct gfs2_sbd *sdp, unsigned int jid,
1198 				 unsigned int result)
1199 {
1200 	struct lm_lockstruct *ls = &sdp->sd_lockstruct;
1201 
1202 	if (gfs2_withdrawn(sdp)) {
1203 		fs_err(sdp, "recovery_result jid %d ignored due to withdraw.\n",
1204 		       jid);
1205 		return;
1206 	}
1207 	if (test_bit(DFL_NO_DLM_OPS, &ls->ls_recover_flags))
1208 		return;
1209 
1210 	/* don't care about the recovery of own journal during mount */
1211 	if (jid == ls->ls_jid)
1212 		return;
1213 
1214 	spin_lock(&ls->ls_recover_spin);
1215 	if (test_bit(DFL_FIRST_MOUNT, &ls->ls_recover_flags)) {
1216 		spin_unlock(&ls->ls_recover_spin);
1217 		return;
1218 	}
1219 	if (ls->ls_recover_size < jid + 1) {
1220 		fs_err(sdp, "recovery_result jid %d short size %d\n",
1221 		       jid, ls->ls_recover_size);
1222 		spin_unlock(&ls->ls_recover_spin);
1223 		return;
1224 	}
1225 
1226 	fs_info(sdp, "recover jid %d result %s\n", jid,
1227 		result == LM_RD_GAVEUP ? "busy" : "success");
1228 
1229 	ls->ls_recover_result[jid] = result;
1230 
1231 	/* GAVEUP means another node is recovering the journal; delay our
1232 	   next attempt to recover it, to give the other node a chance to
1233 	   finish before trying again */
1234 
1235 	if (!test_bit(DFL_UNMOUNT, &ls->ls_recover_flags))
1236 		queue_delayed_work(gfs2_control_wq, &sdp->sd_control_work,
1237 				   result == LM_RD_GAVEUP ? HZ : 0);
1238 	spin_unlock(&ls->ls_recover_spin);
1239 }
1240 
1241 static const struct dlm_lockspace_ops gdlm_lockspace_ops = {
1242 	.recover_prep = gdlm_recover_prep,
1243 	.recover_slot = gdlm_recover_slot,
1244 	.recover_done = gdlm_recover_done,
1245 };
1246 
1247 static int gdlm_mount(struct gfs2_sbd *sdp, const char *table)
1248 {
1249 	struct lm_lockstruct *ls = &sdp->sd_lockstruct;
1250 	char cluster[GFS2_LOCKNAME_LEN];
1251 	const char *fsname;
1252 	uint32_t flags;
1253 	int error, ops_result;
1254 
1255 	/*
1256 	 * initialize everything
1257 	 */
1258 
1259 	INIT_DELAYED_WORK(&sdp->sd_control_work, gfs2_control_func);
1260 	spin_lock_init(&ls->ls_recover_spin);
1261 	ls->ls_recover_flags = 0;
1262 	ls->ls_recover_mount = 0;
1263 	ls->ls_recover_start = 0;
1264 	ls->ls_recover_block = 0;
1265 	ls->ls_recover_size = 0;
1266 	ls->ls_recover_submit = NULL;
1267 	ls->ls_recover_result = NULL;
1268 	ls->ls_lvb_bits = NULL;
1269 
1270 	error = set_recover_size(sdp, NULL, 0);
1271 	if (error)
1272 		goto fail;
1273 
1274 	/*
1275 	 * prepare dlm_new_lockspace args
1276 	 */
1277 
1278 	fsname = strchr(table, ':');
1279 	if (!fsname) {
1280 		fs_info(sdp, "no fsname found\n");
1281 		error = -EINVAL;
1282 		goto fail_free;
1283 	}
1284 	memset(cluster, 0, sizeof(cluster));
1285 	memcpy(cluster, table, strlen(table) - strlen(fsname));
1286 	fsname++;
1287 
1288 	flags = DLM_LSFL_FS | DLM_LSFL_NEWEXCL;
1289 
1290 	/*
1291 	 * create/join lockspace
1292 	 */
1293 
1294 	error = dlm_new_lockspace(fsname, cluster, flags, GDLM_LVB_SIZE,
1295 				  &gdlm_lockspace_ops, sdp, &ops_result,
1296 				  &ls->ls_dlm);
1297 	if (error) {
1298 		fs_err(sdp, "dlm_new_lockspace error %d\n", error);
1299 		goto fail_free;
1300 	}
1301 
1302 	if (ops_result < 0) {
1303 		/*
1304 		 * dlm does not support ops callbacks,
1305 		 * old dlm_controld/gfs_controld are used, try without ops.
1306 		 */
1307 		fs_info(sdp, "dlm lockspace ops not used\n");
1308 		free_recover_size(ls);
1309 		set_bit(DFL_NO_DLM_OPS, &ls->ls_recover_flags);
1310 		return 0;
1311 	}
1312 
1313 	if (!test_bit(SDF_NOJOURNALID, &sdp->sd_flags)) {
1314 		fs_err(sdp, "dlm lockspace ops disallow jid preset\n");
1315 		error = -EINVAL;
1316 		goto fail_release;
1317 	}
1318 
1319 	/*
1320 	 * control_mount() uses control_lock to determine first mounter,
1321 	 * and for later mounts, waits for any recoveries to be cleared.
1322 	 */
1323 
1324 	error = control_mount(sdp);
1325 	if (error) {
1326 		fs_err(sdp, "mount control error %d\n", error);
1327 		goto fail_release;
1328 	}
1329 
1330 	ls->ls_first = !!test_bit(DFL_FIRST_MOUNT, &ls->ls_recover_flags);
1331 	clear_bit(SDF_NOJOURNALID, &sdp->sd_flags);
1332 	smp_mb__after_atomic();
1333 	wake_up_bit(&sdp->sd_flags, SDF_NOJOURNALID);
1334 	return 0;
1335 
1336 fail_release:
1337 	dlm_release_lockspace(ls->ls_dlm, 2);
1338 fail_free:
1339 	free_recover_size(ls);
1340 fail:
1341 	return error;
1342 }
1343 
1344 static void gdlm_first_done(struct gfs2_sbd *sdp)
1345 {
1346 	struct lm_lockstruct *ls = &sdp->sd_lockstruct;
1347 	int error;
1348 
1349 	if (test_bit(DFL_NO_DLM_OPS, &ls->ls_recover_flags))
1350 		return;
1351 
1352 	error = control_first_done(sdp);
1353 	if (error)
1354 		fs_err(sdp, "mount first_done error %d\n", error);
1355 }
1356 
1357 static void gdlm_unmount(struct gfs2_sbd *sdp)
1358 {
1359 	struct lm_lockstruct *ls = &sdp->sd_lockstruct;
1360 
1361 	if (test_bit(DFL_NO_DLM_OPS, &ls->ls_recover_flags))
1362 		goto release;
1363 
1364 	/* wait for gfs2_control_wq to be done with this mount */
1365 
1366 	spin_lock(&ls->ls_recover_spin);
1367 	set_bit(DFL_UNMOUNT, &ls->ls_recover_flags);
1368 	spin_unlock(&ls->ls_recover_spin);
1369 	flush_delayed_work(&sdp->sd_control_work);
1370 
1371 	/* mounted_lock and control_lock will be purged in dlm recovery */
1372 release:
1373 	if (ls->ls_dlm) {
1374 		dlm_release_lockspace(ls->ls_dlm, 2);
1375 		ls->ls_dlm = NULL;
1376 	}
1377 
1378 	free_recover_size(ls);
1379 }
1380 
1381 static const match_table_t dlm_tokens = {
1382 	{ Opt_jid, "jid=%d"},
1383 	{ Opt_id, "id=%d"},
1384 	{ Opt_first, "first=%d"},
1385 	{ Opt_nodir, "nodir=%d"},
1386 	{ Opt_err, NULL },
1387 };
1388 
1389 const struct lm_lockops gfs2_dlm_ops = {
1390 	.lm_proto_name = "lock_dlm",
1391 	.lm_mount = gdlm_mount,
1392 	.lm_first_done = gdlm_first_done,
1393 	.lm_recovery_result = gdlm_recovery_result,
1394 	.lm_unmount = gdlm_unmount,
1395 	.lm_put_lock = gdlm_put_lock,
1396 	.lm_lock = gdlm_lock,
1397 	.lm_cancel = gdlm_cancel,
1398 	.lm_tokens = &dlm_tokens,
1399 };
1400 
1401