xref: /linux/fs/ocfs2/slot_map.c (revision 37744feebc086908fd89760650f458ab19071750) 
1 // SPDX-License-Identifier: GPL-2.0-or-later
2 /* -*- mode: c; c-basic-offset: 8; -*-
3  * vim: noexpandtab sw=8 ts=8 sts=0:
4  *
5  * slot_map.c
6  *
7  * Copyright (C) 2002, 2004 Oracle.  All rights reserved.
8  */
9 
10 #include <linux/types.h>
11 #include <linux/slab.h>
12 #include <linux/highmem.h>
13 
14 #include <cluster/masklog.h>
15 
16 #include "ocfs2.h"
17 
18 #include "dlmglue.h"
19 #include "extent_map.h"
20 #include "heartbeat.h"
21 #include "inode.h"
22 #include "slot_map.h"
23 #include "super.h"
24 #include "sysfile.h"
25 #include "ocfs2_trace.h"
26 
27 #include "buffer_head_io.h"
28 
29 
30 struct ocfs2_slot {
31 	int sl_valid;
32 	unsigned int sl_node_num;
33 };
34 
35 struct ocfs2_slot_info {
36 	int si_extended;
37 	int si_slots_per_block;
38 	struct inode *si_inode;
39 	unsigned int si_blocks;
40 	struct buffer_head **si_bh;
41 	unsigned int si_num_slots;
42 	struct ocfs2_slot si_slots[];
43 };
44 
45 
46 static int __ocfs2_node_num_to_slot(struct ocfs2_slot_info *si,
47 				    unsigned int node_num);
48 
49 static void ocfs2_invalidate_slot(struct ocfs2_slot_info *si,
50 				  int slot_num)
51 {
52 	BUG_ON((slot_num < 0) || (slot_num >= si->si_num_slots));
53 	si->si_slots[slot_num].sl_valid = 0;
54 }
55 
56 static void ocfs2_set_slot(struct ocfs2_slot_info *si,
57 			   int slot_num, unsigned int node_num)
58 {
59 	BUG_ON((slot_num < 0) || (slot_num >= si->si_num_slots));
60 
61 	si->si_slots[slot_num].sl_valid = 1;
62 	si->si_slots[slot_num].sl_node_num = node_num;
63 }
64 
65 /* This version is for the extended slot map */
66 static void ocfs2_update_slot_info_extended(struct ocfs2_slot_info *si)
67 {
68 	int b, i, slotno;
69 	struct ocfs2_slot_map_extended *se;
70 
71 	slotno = 0;
72 	for (b = 0; b < si->si_blocks; b++) {
73 		se = (struct ocfs2_slot_map_extended *)si->si_bh[b]->b_data;
74 		for (i = 0;
75 		     (i < si->si_slots_per_block) &&
76 		     (slotno < si->si_num_slots);
77 		     i++, slotno++) {
78 			if (se->se_slots[i].es_valid)
79 				ocfs2_set_slot(si, slotno,
80 					       le32_to_cpu(se->se_slots[i].es_node_num));
81 			else
82 				ocfs2_invalidate_slot(si, slotno);
83 		}
84 	}
85 }
86 
87 /*
88  * Post the slot information on disk into our slot_info struct.
89  * Must be protected by osb_lock.
90  */
91 static void ocfs2_update_slot_info_old(struct ocfs2_slot_info *si)
92 {
93 	int i;
94 	struct ocfs2_slot_map *sm;
95 
96 	sm = (struct ocfs2_slot_map *)si->si_bh[0]->b_data;
97 
98 	for (i = 0; i < si->si_num_slots; i++) {
99 		if (le16_to_cpu(sm->sm_slots[i]) == (u16)OCFS2_INVALID_SLOT)
100 			ocfs2_invalidate_slot(si, i);
101 		else
102 			ocfs2_set_slot(si, i, le16_to_cpu(sm->sm_slots[i]));
103 	}
104 }
105 
106 static void ocfs2_update_slot_info(struct ocfs2_slot_info *si)
107 {
108 	/*
109 	 * The slot data will have been refreshed when ocfs2_super_lock
110 	 * was taken.
111 	 */
112 	if (si->si_extended)
113 		ocfs2_update_slot_info_extended(si);
114 	else
115 		ocfs2_update_slot_info_old(si);
116 }
117 
118 int ocfs2_refresh_slot_info(struct ocfs2_super *osb)
119 {
120 	int ret;
121 	struct ocfs2_slot_info *si = osb->slot_info;
122 
123 	if (si == NULL)
124 		return 0;
125 
126 	BUG_ON(si->si_blocks == 0);
127 	BUG_ON(si->si_bh == NULL);
128 
129 	trace_ocfs2_refresh_slot_info(si->si_blocks);
130 
131 	/*
132 	 * We pass -1 as blocknr because we expect all of si->si_bh to
133 	 * be !NULL.  Thus, ocfs2_read_blocks() will ignore blocknr.  If
134 	 * this is not true, the read of -1 (UINT64_MAX) will fail.
135 	 */
136 	ret = ocfs2_read_blocks(INODE_CACHE(si->si_inode), -1, si->si_blocks,
137 				si->si_bh, OCFS2_BH_IGNORE_CACHE, NULL);
138 	if (ret == 0) {
139 		spin_lock(&osb->osb_lock);
140 		ocfs2_update_slot_info(si);
141 		spin_unlock(&osb->osb_lock);
142 	}
143 
144 	return ret;
145 }
146 
147 /* post the our slot info stuff into it's destination bh and write it
148  * out. */
149 static void ocfs2_update_disk_slot_extended(struct ocfs2_slot_info *si,
150 					    int slot_num,
151 					    struct buffer_head **bh)
152 {
153 	int blkind = slot_num / si->si_slots_per_block;
154 	int slotno = slot_num % si->si_slots_per_block;
155 	struct ocfs2_slot_map_extended *se;
156 
157 	BUG_ON(blkind >= si->si_blocks);
158 
159 	se = (struct ocfs2_slot_map_extended *)si->si_bh[blkind]->b_data;
160 	se->se_slots[slotno].es_valid = si->si_slots[slot_num].sl_valid;
161 	if (si->si_slots[slot_num].sl_valid)
162 		se->se_slots[slotno].es_node_num =
163 			cpu_to_le32(si->si_slots[slot_num].sl_node_num);
164 	*bh = si->si_bh[blkind];
165 }
166 
167 static void ocfs2_update_disk_slot_old(struct ocfs2_slot_info *si,
168 				       int slot_num,
169 				       struct buffer_head **bh)
170 {
171 	int i;
172 	struct ocfs2_slot_map *sm;
173 
174 	sm = (struct ocfs2_slot_map *)si->si_bh[0]->b_data;
175 	for (i = 0; i < si->si_num_slots; i++) {
176 		if (si->si_slots[i].sl_valid)
177 			sm->sm_slots[i] =
178 				cpu_to_le16(si->si_slots[i].sl_node_num);
179 		else
180 			sm->sm_slots[i] = cpu_to_le16(OCFS2_INVALID_SLOT);
181 	}
182 	*bh = si->si_bh[0];
183 }
184 
185 static int ocfs2_update_disk_slot(struct ocfs2_super *osb,
186 				  struct ocfs2_slot_info *si,
187 				  int slot_num)
188 {
189 	int status;
190 	struct buffer_head *bh;
191 
192 	spin_lock(&osb->osb_lock);
193 	if (si->si_extended)
194 		ocfs2_update_disk_slot_extended(si, slot_num, &bh);
195 	else
196 		ocfs2_update_disk_slot_old(si, slot_num, &bh);
197 	spin_unlock(&osb->osb_lock);
198 
199 	status = ocfs2_write_block(osb, bh, INODE_CACHE(si->si_inode));
200 	if (status < 0)
201 		mlog_errno(status);
202 
203 	return status;
204 }
205 
206 /*
207  * Calculate how many bytes are needed by the slot map.  Returns
208  * an error if the slot map file is too small.
209  */
210 static int ocfs2_slot_map_physical_size(struct ocfs2_super *osb,
211 					struct inode *inode,
212 					unsigned long long *bytes)
213 {
214 	unsigned long long bytes_needed;
215 
216 	if (ocfs2_uses_extended_slot_map(osb)) {
217 		bytes_needed = osb->max_slots *
218 			sizeof(struct ocfs2_extended_slot);
219 	} else {
220 		bytes_needed = osb->max_slots * sizeof(__le16);
221 	}
222 	if (bytes_needed > i_size_read(inode)) {
223 		mlog(ML_ERROR,
224 		     "Slot map file is too small!  (size %llu, needed %llu)\n",
225 		     i_size_read(inode), bytes_needed);
226 		return -ENOSPC;
227 	}
228 
229 	*bytes = bytes_needed;
230 	return 0;
231 }
232 
233 /* try to find global node in the slot info. Returns -ENOENT
234  * if nothing is found. */
235 static int __ocfs2_node_num_to_slot(struct ocfs2_slot_info *si,
236 				    unsigned int node_num)
237 {
238 	int i, ret = -ENOENT;
239 
240 	for(i = 0; i < si->si_num_slots; i++) {
241 		if (si->si_slots[i].sl_valid &&
242 		    (node_num == si->si_slots[i].sl_node_num)) {
243 			ret = i;
244 			break;
245 		}
246 	}
247 
248 	return ret;
249 }
250 
251 static int __ocfs2_find_empty_slot(struct ocfs2_slot_info *si,
252 				   int preferred)
253 {
254 	int i, ret = -ENOSPC;
255 
256 	if ((preferred >= 0) && (preferred < si->si_num_slots)) {
257 		if (!si->si_slots[preferred].sl_valid) {
258 			ret = preferred;
259 			goto out;
260 		}
261 	}
262 
263 	for(i = 0; i < si->si_num_slots; i++) {
264 		if (!si->si_slots[i].sl_valid) {
265 			ret = i;
266 			break;
267 		}
268 	}
269 out:
270 	return ret;
271 }
272 
273 int ocfs2_node_num_to_slot(struct ocfs2_super *osb, unsigned int node_num)
274 {
275 	int slot;
276 	struct ocfs2_slot_info *si = osb->slot_info;
277 
278 	spin_lock(&osb->osb_lock);
279 	slot = __ocfs2_node_num_to_slot(si, node_num);
280 	spin_unlock(&osb->osb_lock);
281 
282 	return slot;
283 }
284 
285 int ocfs2_slot_to_node_num_locked(struct ocfs2_super *osb, int slot_num,
286 				  unsigned int *node_num)
287 {
288 	struct ocfs2_slot_info *si = osb->slot_info;
289 
290 	assert_spin_locked(&osb->osb_lock);
291 
292 	BUG_ON(slot_num < 0);
293 	BUG_ON(slot_num >= osb->max_slots);
294 
295 	if (!si->si_slots[slot_num].sl_valid)
296 		return -ENOENT;
297 
298 	*node_num = si->si_slots[slot_num].sl_node_num;
299 	return 0;
300 }
301 
302 static void __ocfs2_free_slot_info(struct ocfs2_slot_info *si)
303 {
304 	unsigned int i;
305 
306 	if (si == NULL)
307 		return;
308 
309 	iput(si->si_inode);
310 	if (si->si_bh) {
311 		for (i = 0; i < si->si_blocks; i++) {
312 			if (si->si_bh[i]) {
313 				brelse(si->si_bh[i]);
314 				si->si_bh[i] = NULL;
315 			}
316 		}
317 		kfree(si->si_bh);
318 	}
319 
320 	kfree(si);
321 }
322 
323 int ocfs2_clear_slot(struct ocfs2_super *osb, int slot_num)
324 {
325 	struct ocfs2_slot_info *si = osb->slot_info;
326 
327 	if (si == NULL)
328 		return 0;
329 
330 	spin_lock(&osb->osb_lock);
331 	ocfs2_invalidate_slot(si, slot_num);
332 	spin_unlock(&osb->osb_lock);
333 
334 	return ocfs2_update_disk_slot(osb, osb->slot_info, slot_num);
335 }
336 
337 static int ocfs2_map_slot_buffers(struct ocfs2_super *osb,
338 				  struct ocfs2_slot_info *si)
339 {
340 	int status = 0;
341 	u64 blkno;
342 	unsigned long long blocks, bytes = 0;
343 	unsigned int i;
344 	struct buffer_head *bh;
345 
346 	status = ocfs2_slot_map_physical_size(osb, si->si_inode, &bytes);
347 	if (status)
348 		goto bail;
349 
350 	blocks = ocfs2_blocks_for_bytes(si->si_inode->i_sb, bytes);
351 	BUG_ON(blocks > UINT_MAX);
352 	si->si_blocks = blocks;
353 	if (!si->si_blocks)
354 		goto bail;
355 
356 	if (si->si_extended)
357 		si->si_slots_per_block =
358 			(osb->sb->s_blocksize /
359 			 sizeof(struct ocfs2_extended_slot));
360 	else
361 		si->si_slots_per_block = osb->sb->s_blocksize / sizeof(__le16);
362 
363 	/* The size checks above should ensure this */
364 	BUG_ON((osb->max_slots / si->si_slots_per_block) > blocks);
365 
366 	trace_ocfs2_map_slot_buffers(bytes, si->si_blocks);
367 
368 	si->si_bh = kcalloc(si->si_blocks, sizeof(struct buffer_head *),
369 			    GFP_KERNEL);
370 	if (!si->si_bh) {
371 		status = -ENOMEM;
372 		mlog_errno(status);
373 		goto bail;
374 	}
375 
376 	for (i = 0; i < si->si_blocks; i++) {
377 		status = ocfs2_extent_map_get_blocks(si->si_inode, i,
378 						     &blkno, NULL, NULL);
379 		if (status < 0) {
380 			mlog_errno(status);
381 			goto bail;
382 		}
383 
384 		trace_ocfs2_map_slot_buffers_block((unsigned long long)blkno, i);
385 
386 		bh = NULL;  /* Acquire a fresh bh */
387 		status = ocfs2_read_blocks(INODE_CACHE(si->si_inode), blkno,
388 					   1, &bh, OCFS2_BH_IGNORE_CACHE, NULL);
389 		if (status < 0) {
390 			mlog_errno(status);
391 			goto bail;
392 		}
393 
394 		si->si_bh[i] = bh;
395 	}
396 
397 bail:
398 	return status;
399 }
400 
401 int ocfs2_init_slot_info(struct ocfs2_super *osb)
402 {
403 	int status;
404 	struct inode *inode = NULL;
405 	struct ocfs2_slot_info *si;
406 
407 	si = kzalloc(struct_size(si, si_slots, osb->max_slots), GFP_KERNEL);
408 	if (!si) {
409 		status = -ENOMEM;
410 		mlog_errno(status);
411 		return status;
412 	}
413 
414 	si->si_extended = ocfs2_uses_extended_slot_map(osb);
415 	si->si_num_slots = osb->max_slots;
416 
417 	inode = ocfs2_get_system_file_inode(osb, SLOT_MAP_SYSTEM_INODE,
418 					    OCFS2_INVALID_SLOT);
419 	if (!inode) {
420 		status = -EINVAL;
421 		mlog_errno(status);
422 		goto bail;
423 	}
424 
425 	si->si_inode = inode;
426 	status = ocfs2_map_slot_buffers(osb, si);
427 	if (status < 0) {
428 		mlog_errno(status);
429 		goto bail;
430 	}
431 
432 	osb->slot_info = (struct ocfs2_slot_info *)si;
433 bail:
434 	if (status < 0)
435 		__ocfs2_free_slot_info(si);
436 
437 	return status;
438 }
439 
440 void ocfs2_free_slot_info(struct ocfs2_super *osb)
441 {
442 	struct ocfs2_slot_info *si = osb->slot_info;
443 
444 	osb->slot_info = NULL;
445 	__ocfs2_free_slot_info(si);
446 }
447 
448 int ocfs2_find_slot(struct ocfs2_super *osb)
449 {
450 	int status;
451 	int slot;
452 	struct ocfs2_slot_info *si;
453 
454 	si = osb->slot_info;
455 
456 	spin_lock(&osb->osb_lock);
457 	ocfs2_update_slot_info(si);
458 
459 	/* search for ourselves first and take the slot if it already
460 	 * exists. Perhaps we need to mark this in a variable for our
461 	 * own journal recovery? Possibly not, though we certainly
462 	 * need to warn to the user */
463 	slot = __ocfs2_node_num_to_slot(si, osb->node_num);
464 	if (slot < 0) {
465 		/* if no slot yet, then just take 1st available
466 		 * one. */
467 		slot = __ocfs2_find_empty_slot(si, osb->preferred_slot);
468 		if (slot < 0) {
469 			spin_unlock(&osb->osb_lock);
470 			mlog(ML_ERROR, "no free slots available!\n");
471 			status = -EINVAL;
472 			goto bail;
473 		}
474 	} else
475 		printk(KERN_INFO "ocfs2: Slot %d on device (%s) was already "
476 		       "allocated to this node!\n", slot, osb->dev_str);
477 
478 	ocfs2_set_slot(si, slot, osb->node_num);
479 	osb->slot_num = slot;
480 	spin_unlock(&osb->osb_lock);
481 
482 	trace_ocfs2_find_slot(osb->slot_num);
483 
484 	status = ocfs2_update_disk_slot(osb, si, osb->slot_num);
485 	if (status < 0) {
486 		mlog_errno(status);
487 		/*
488 		 * if write block failed, invalidate slot to avoid overwrite
489 		 * slot during dismount in case another node rightly has mounted
490 		 */
491 		spin_lock(&osb->osb_lock);
492 		ocfs2_invalidate_slot(si, osb->slot_num);
493 		osb->slot_num = OCFS2_INVALID_SLOT;
494 		spin_unlock(&osb->osb_lock);
495 	}
496 
497 bail:
498 	return status;
499 }
500 
501 void ocfs2_put_slot(struct ocfs2_super *osb)
502 {
503 	int status, slot_num;
504 	struct ocfs2_slot_info *si = osb->slot_info;
505 
506 	if (!si)
507 		return;
508 
509 	spin_lock(&osb->osb_lock);
510 	ocfs2_update_slot_info(si);
511 
512 	slot_num = osb->slot_num;
513 	ocfs2_invalidate_slot(si, osb->slot_num);
514 	osb->slot_num = OCFS2_INVALID_SLOT;
515 	spin_unlock(&osb->osb_lock);
516 
517 	status = ocfs2_update_disk_slot(osb, si, slot_num);
518 	if (status < 0)
519 		mlog_errno(status);
520 
521 	ocfs2_free_slot_info(osb);
522 }
523