xref: /linux/fs/ocfs2/slot_map.c (revision 0be3ff0c) 
1 // SPDX-License-Identifier: GPL-2.0-or-later
2 /*
3  * slot_map.c
4  *
5  * Copyright (C) 2002, 2004 Oracle.  All rights reserved.
6  */
7 
8 #include <linux/types.h>
9 #include <linux/slab.h>
10 #include <linux/highmem.h>
11 
12 #include <cluster/masklog.h>
13 
14 #include "ocfs2.h"
15 
16 #include "dlmglue.h"
17 #include "extent_map.h"
18 #include "heartbeat.h"
19 #include "inode.h"
20 #include "slot_map.h"
21 #include "super.h"
22 #include "sysfile.h"
23 #include "ocfs2_trace.h"
24 
25 #include "buffer_head_io.h"
26 
27 
28 struct ocfs2_slot {
29 	int sl_valid;
30 	unsigned int sl_node_num;
31 };
32 
33 struct ocfs2_slot_info {
34 	int si_extended;
35 	int si_slots_per_block;
36 	struct inode *si_inode;
37 	unsigned int si_blocks;
38 	struct buffer_head **si_bh;
39 	unsigned int si_num_slots;
40 	struct ocfs2_slot si_slots[];
41 };
42 
43 
44 static int __ocfs2_node_num_to_slot(struct ocfs2_slot_info *si,
45 				    unsigned int node_num);
46 
47 static void ocfs2_invalidate_slot(struct ocfs2_slot_info *si,
48 				  int slot_num)
49 {
50 	BUG_ON((slot_num < 0) || (slot_num >= si->si_num_slots));
51 	si->si_slots[slot_num].sl_valid = 0;
52 }
53 
54 static void ocfs2_set_slot(struct ocfs2_slot_info *si,
55 			   int slot_num, unsigned int node_num)
56 {
57 	BUG_ON((slot_num < 0) || (slot_num >= si->si_num_slots));
58 
59 	si->si_slots[slot_num].sl_valid = 1;
60 	si->si_slots[slot_num].sl_node_num = node_num;
61 }
62 
63 /* This version is for the extended slot map */
64 static void ocfs2_update_slot_info_extended(struct ocfs2_slot_info *si)
65 {
66 	int b, i, slotno;
67 	struct ocfs2_slot_map_extended *se;
68 
69 	slotno = 0;
70 	for (b = 0; b < si->si_blocks; b++) {
71 		se = (struct ocfs2_slot_map_extended *)si->si_bh[b]->b_data;
72 		for (i = 0;
73 		     (i < si->si_slots_per_block) &&
74 		     (slotno < si->si_num_slots);
75 		     i++, slotno++) {
76 			if (se->se_slots[i].es_valid)
77 				ocfs2_set_slot(si, slotno,
78 					       le32_to_cpu(se->se_slots[i].es_node_num));
79 			else
80 				ocfs2_invalidate_slot(si, slotno);
81 		}
82 	}
83 }
84 
85 /*
86  * Post the slot information on disk into our slot_info struct.
87  * Must be protected by osb_lock.
88  */
89 static void ocfs2_update_slot_info_old(struct ocfs2_slot_info *si)
90 {
91 	int i;
92 	struct ocfs2_slot_map *sm;
93 
94 	sm = (struct ocfs2_slot_map *)si->si_bh[0]->b_data;
95 
96 	for (i = 0; i < si->si_num_slots; i++) {
97 		if (le16_to_cpu(sm->sm_slots[i]) == (u16)OCFS2_INVALID_SLOT)
98 			ocfs2_invalidate_slot(si, i);
99 		else
100 			ocfs2_set_slot(si, i, le16_to_cpu(sm->sm_slots[i]));
101 	}
102 }
103 
104 static void ocfs2_update_slot_info(struct ocfs2_slot_info *si)
105 {
106 	/*
107 	 * The slot data will have been refreshed when ocfs2_super_lock
108 	 * was taken.
109 	 */
110 	if (si->si_extended)
111 		ocfs2_update_slot_info_extended(si);
112 	else
113 		ocfs2_update_slot_info_old(si);
114 }
115 
116 int ocfs2_refresh_slot_info(struct ocfs2_super *osb)
117 {
118 	int ret;
119 	struct ocfs2_slot_info *si = osb->slot_info;
120 
121 	if (si == NULL)
122 		return 0;
123 
124 	BUG_ON(si->si_blocks == 0);
125 	BUG_ON(si->si_bh == NULL);
126 
127 	trace_ocfs2_refresh_slot_info(si->si_blocks);
128 
129 	/*
130 	 * We pass -1 as blocknr because we expect all of si->si_bh to
131 	 * be !NULL.  Thus, ocfs2_read_blocks() will ignore blocknr.  If
132 	 * this is not true, the read of -1 (UINT64_MAX) will fail.
133 	 */
134 	ret = ocfs2_read_blocks(INODE_CACHE(si->si_inode), -1, si->si_blocks,
135 				si->si_bh, OCFS2_BH_IGNORE_CACHE, NULL);
136 	if (ret == 0) {
137 		spin_lock(&osb->osb_lock);
138 		ocfs2_update_slot_info(si);
139 		spin_unlock(&osb->osb_lock);
140 	}
141 
142 	return ret;
143 }
144 
145 /* post the our slot info stuff into it's destination bh and write it
146  * out. */
147 static void ocfs2_update_disk_slot_extended(struct ocfs2_slot_info *si,
148 					    int slot_num,
149 					    struct buffer_head **bh)
150 {
151 	int blkind = slot_num / si->si_slots_per_block;
152 	int slotno = slot_num % si->si_slots_per_block;
153 	struct ocfs2_slot_map_extended *se;
154 
155 	BUG_ON(blkind >= si->si_blocks);
156 
157 	se = (struct ocfs2_slot_map_extended *)si->si_bh[blkind]->b_data;
158 	se->se_slots[slotno].es_valid = si->si_slots[slot_num].sl_valid;
159 	if (si->si_slots[slot_num].sl_valid)
160 		se->se_slots[slotno].es_node_num =
161 			cpu_to_le32(si->si_slots[slot_num].sl_node_num);
162 	*bh = si->si_bh[blkind];
163 }
164 
165 static void ocfs2_update_disk_slot_old(struct ocfs2_slot_info *si,
166 				       int slot_num,
167 				       struct buffer_head **bh)
168 {
169 	int i;
170 	struct ocfs2_slot_map *sm;
171 
172 	sm = (struct ocfs2_slot_map *)si->si_bh[0]->b_data;
173 	for (i = 0; i < si->si_num_slots; i++) {
174 		if (si->si_slots[i].sl_valid)
175 			sm->sm_slots[i] =
176 				cpu_to_le16(si->si_slots[i].sl_node_num);
177 		else
178 			sm->sm_slots[i] = cpu_to_le16(OCFS2_INVALID_SLOT);
179 	}
180 	*bh = si->si_bh[0];
181 }
182 
183 static int ocfs2_update_disk_slot(struct ocfs2_super *osb,
184 				  struct ocfs2_slot_info *si,
185 				  int slot_num)
186 {
187 	int status;
188 	struct buffer_head *bh;
189 
190 	spin_lock(&osb->osb_lock);
191 	if (si->si_extended)
192 		ocfs2_update_disk_slot_extended(si, slot_num, &bh);
193 	else
194 		ocfs2_update_disk_slot_old(si, slot_num, &bh);
195 	spin_unlock(&osb->osb_lock);
196 
197 	status = ocfs2_write_block(osb, bh, INODE_CACHE(si->si_inode));
198 	if (status < 0)
199 		mlog_errno(status);
200 
201 	return status;
202 }
203 
204 /*
205  * Calculate how many bytes are needed by the slot map.  Returns
206  * an error if the slot map file is too small.
207  */
208 static int ocfs2_slot_map_physical_size(struct ocfs2_super *osb,
209 					struct inode *inode,
210 					unsigned long long *bytes)
211 {
212 	unsigned long long bytes_needed;
213 
214 	if (ocfs2_uses_extended_slot_map(osb)) {
215 		bytes_needed = osb->max_slots *
216 			sizeof(struct ocfs2_extended_slot);
217 	} else {
218 		bytes_needed = osb->max_slots * sizeof(__le16);
219 	}
220 	if (bytes_needed > i_size_read(inode)) {
221 		mlog(ML_ERROR,
222 		     "Slot map file is too small!  (size %llu, needed %llu)\n",
223 		     i_size_read(inode), bytes_needed);
224 		return -ENOSPC;
225 	}
226 
227 	*bytes = bytes_needed;
228 	return 0;
229 }
230 
231 /* try to find global node in the slot info. Returns -ENOENT
232  * if nothing is found. */
233 static int __ocfs2_node_num_to_slot(struct ocfs2_slot_info *si,
234 				    unsigned int node_num)
235 {
236 	int i, ret = -ENOENT;
237 
238 	for(i = 0; i < si->si_num_slots; i++) {
239 		if (si->si_slots[i].sl_valid &&
240 		    (node_num == si->si_slots[i].sl_node_num)) {
241 			ret = i;
242 			break;
243 		}
244 	}
245 
246 	return ret;
247 }
248 
249 static int __ocfs2_find_empty_slot(struct ocfs2_slot_info *si,
250 				   int preferred)
251 {
252 	int i, ret = -ENOSPC;
253 
254 	if ((preferred >= 0) && (preferred < si->si_num_slots)) {
255 		if (!si->si_slots[preferred].sl_valid ||
256 		    !si->si_slots[preferred].sl_node_num) {
257 			ret = preferred;
258 			goto out;
259 		}
260 	}
261 
262 	for(i = 0; i < si->si_num_slots; i++) {
263 		if (!si->si_slots[i].sl_valid ||
264 		    !si->si_slots[i].sl_node_num) {
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 	if (ocfs2_mount_local(osb))
460 		/* use slot 0 directly in local mode */
461 		slot = 0;
462 	else {
463 		/* search for ourselves first and take the slot if it already
464 		 * exists. Perhaps we need to mark this in a variable for our
465 		 * own journal recovery? Possibly not, though we certainly
466 		 * need to warn to the user */
467 		slot = __ocfs2_node_num_to_slot(si, osb->node_num);
468 		if (slot < 0) {
469 			/* if no slot yet, then just take 1st available
470 			 * one. */
471 			slot = __ocfs2_find_empty_slot(si, osb->preferred_slot);
472 			if (slot < 0) {
473 				spin_unlock(&osb->osb_lock);
474 				mlog(ML_ERROR, "no free slots available!\n");
475 				status = -EINVAL;
476 				goto bail;
477 			}
478 		} else
479 			printk(KERN_INFO "ocfs2: Slot %d on device (%s) was "
480 			       "already allocated to this node!\n",
481 			       slot, osb->dev_str);
482 	}
483 
484 	ocfs2_set_slot(si, slot, osb->node_num);
485 	osb->slot_num = slot;
486 	spin_unlock(&osb->osb_lock);
487 
488 	trace_ocfs2_find_slot(osb->slot_num);
489 
490 	status = ocfs2_update_disk_slot(osb, si, osb->slot_num);
491 	if (status < 0) {
492 		mlog_errno(status);
493 		/*
494 		 * if write block failed, invalidate slot to avoid overwrite
495 		 * slot during dismount in case another node rightly has mounted
496 		 */
497 		spin_lock(&osb->osb_lock);
498 		ocfs2_invalidate_slot(si, osb->slot_num);
499 		osb->slot_num = OCFS2_INVALID_SLOT;
500 		spin_unlock(&osb->osb_lock);
501 	}
502 
503 bail:
504 	return status;
505 }
506 
507 void ocfs2_put_slot(struct ocfs2_super *osb)
508 {
509 	int status, slot_num;
510 	struct ocfs2_slot_info *si = osb->slot_info;
511 
512 	if (!si)
513 		return;
514 
515 	spin_lock(&osb->osb_lock);
516 	ocfs2_update_slot_info(si);
517 
518 	slot_num = osb->slot_num;
519 	ocfs2_invalidate_slot(si, osb->slot_num);
520 	osb->slot_num = OCFS2_INVALID_SLOT;
521 	spin_unlock(&osb->osb_lock);
522 
523 	status = ocfs2_update_disk_slot(osb, si, slot_num);
524 	if (status < 0)
525 		mlog_errno(status);
526 
527 	ocfs2_free_slot_info(osb);
528 }
529