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[] __counted_by(si_num_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, slot_num: slotno, |
78 | le32_to_cpu(se->se_slots[i].es_node_num)); |
79 | else |
80 | ocfs2_invalidate_slot(si, slot_num: 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, slot_num: i); |
99 | else |
100 | ocfs2_set_slot(si, slot_num: 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(num: 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(ci: INODE_CACHE(inode: si->si_inode), block: -1, nr: si->si_blocks, |
135 | bhs: si->si_bh, OCFS2_BH_IGNORE_CACHE, NULL); |
136 | if (ret == 0) { |
137 | spin_lock(lock: &osb->osb_lock); |
138 | ocfs2_update_slot_info(si); |
139 | spin_unlock(lock: &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(lock: &osb->osb_lock); |
191 | if (si->si_extended) |
192 | ocfs2_update_disk_slot_extended(si, slot_num, bh: &bh); |
193 | else |
194 | ocfs2_update_disk_slot_old(si, slot_num, bh: &bh); |
195 | spin_unlock(lock: &osb->osb_lock); |
196 | |
197 | status = ocfs2_write_block(osb, bh, ci: INODE_CACHE(inode: 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 | ret = preferred; |
257 | goto out; |
258 | } |
259 | } |
260 | |
261 | for(i = 0; i < si->si_num_slots; i++) { |
262 | if (!si->si_slots[i].sl_valid) { |
263 | ret = i; |
264 | break; |
265 | } |
266 | } |
267 | out: |
268 | return ret; |
269 | } |
270 | |
271 | int ocfs2_node_num_to_slot(struct ocfs2_super *osb, unsigned int node_num) |
272 | { |
273 | int slot; |
274 | struct ocfs2_slot_info *si = osb->slot_info; |
275 | |
276 | spin_lock(lock: &osb->osb_lock); |
277 | slot = __ocfs2_node_num_to_slot(si, node_num); |
278 | spin_unlock(lock: &osb->osb_lock); |
279 | |
280 | return slot; |
281 | } |
282 | |
283 | int ocfs2_slot_to_node_num_locked(struct ocfs2_super *osb, int slot_num, |
284 | unsigned int *node_num) |
285 | { |
286 | struct ocfs2_slot_info *si = osb->slot_info; |
287 | |
288 | assert_spin_locked(&osb->osb_lock); |
289 | |
290 | BUG_ON(slot_num < 0); |
291 | BUG_ON(slot_num >= osb->max_slots); |
292 | |
293 | if (!si->si_slots[slot_num].sl_valid) |
294 | return -ENOENT; |
295 | |
296 | *node_num = si->si_slots[slot_num].sl_node_num; |
297 | return 0; |
298 | } |
299 | |
300 | static void __ocfs2_free_slot_info(struct ocfs2_slot_info *si) |
301 | { |
302 | unsigned int i; |
303 | |
304 | if (si == NULL) |
305 | return; |
306 | |
307 | iput(si->si_inode); |
308 | if (si->si_bh) { |
309 | for (i = 0; i < si->si_blocks; i++) { |
310 | if (si->si_bh[i]) { |
311 | brelse(bh: si->si_bh[i]); |
312 | si->si_bh[i] = NULL; |
313 | } |
314 | } |
315 | kfree(objp: si->si_bh); |
316 | } |
317 | |
318 | kfree(objp: si); |
319 | } |
320 | |
321 | int ocfs2_clear_slot(struct ocfs2_super *osb, int slot_num) |
322 | { |
323 | struct ocfs2_slot_info *si = osb->slot_info; |
324 | |
325 | if (si == NULL) |
326 | return 0; |
327 | |
328 | spin_lock(lock: &osb->osb_lock); |
329 | ocfs2_invalidate_slot(si, slot_num); |
330 | spin_unlock(lock: &osb->osb_lock); |
331 | |
332 | return ocfs2_update_disk_slot(osb, si: osb->slot_info, slot_num); |
333 | } |
334 | |
335 | static int ocfs2_map_slot_buffers(struct ocfs2_super *osb, |
336 | struct ocfs2_slot_info *si) |
337 | { |
338 | int status = 0; |
339 | u64 blkno; |
340 | unsigned long long blocks, bytes = 0; |
341 | unsigned int i; |
342 | struct buffer_head *bh; |
343 | |
344 | status = ocfs2_slot_map_physical_size(osb, inode: si->si_inode, bytes: &bytes); |
345 | if (status) |
346 | goto bail; |
347 | |
348 | blocks = ocfs2_blocks_for_bytes(sb: si->si_inode->i_sb, bytes); |
349 | BUG_ON(blocks > UINT_MAX); |
350 | si->si_blocks = blocks; |
351 | if (!si->si_blocks) |
352 | goto bail; |
353 | |
354 | if (si->si_extended) |
355 | si->si_slots_per_block = |
356 | (osb->sb->s_blocksize / |
357 | sizeof(struct ocfs2_extended_slot)); |
358 | else |
359 | si->si_slots_per_block = osb->sb->s_blocksize / sizeof(__le16); |
360 | |
361 | /* The size checks above should ensure this */ |
362 | BUG_ON((osb->max_slots / si->si_slots_per_block) > blocks); |
363 | |
364 | trace_ocfs2_map_slot_buffers(val1: bytes, val2: si->si_blocks); |
365 | |
366 | si->si_bh = kcalloc(n: si->si_blocks, size: sizeof(struct buffer_head *), |
367 | GFP_KERNEL); |
368 | if (!si->si_bh) { |
369 | status = -ENOMEM; |
370 | mlog_errno(status); |
371 | goto bail; |
372 | } |
373 | |
374 | for (i = 0; i < si->si_blocks; i++) { |
375 | status = ocfs2_extent_map_get_blocks(inode: si->si_inode, v_blkno: i, |
376 | p_blkno: &blkno, NULL, NULL); |
377 | if (status < 0) { |
378 | mlog_errno(status); |
379 | goto bail; |
380 | } |
381 | |
382 | trace_ocfs2_map_slot_buffers_block(val1: (unsigned long long)blkno, val2: i); |
383 | |
384 | bh = NULL; /* Acquire a fresh bh */ |
385 | status = ocfs2_read_blocks(ci: INODE_CACHE(inode: si->si_inode), block: blkno, |
386 | nr: 1, bhs: &bh, OCFS2_BH_IGNORE_CACHE, NULL); |
387 | if (status < 0) { |
388 | mlog_errno(status); |
389 | goto bail; |
390 | } |
391 | |
392 | si->si_bh[i] = bh; |
393 | } |
394 | |
395 | bail: |
396 | return status; |
397 | } |
398 | |
399 | int ocfs2_init_slot_info(struct ocfs2_super *osb) |
400 | { |
401 | int status; |
402 | struct inode *inode = NULL; |
403 | struct ocfs2_slot_info *si; |
404 | |
405 | si = kzalloc(struct_size(si, si_slots, osb->max_slots), GFP_KERNEL); |
406 | if (!si) { |
407 | status = -ENOMEM; |
408 | mlog_errno(status); |
409 | return status; |
410 | } |
411 | |
412 | si->si_extended = ocfs2_uses_extended_slot_map(osb); |
413 | si->si_num_slots = osb->max_slots; |
414 | |
415 | inode = ocfs2_get_system_file_inode(osb, type: SLOT_MAP_SYSTEM_INODE, |
416 | OCFS2_INVALID_SLOT); |
417 | if (!inode) { |
418 | status = -EINVAL; |
419 | mlog_errno(status); |
420 | goto bail; |
421 | } |
422 | |
423 | si->si_inode = inode; |
424 | status = ocfs2_map_slot_buffers(osb, si); |
425 | if (status < 0) { |
426 | mlog_errno(status); |
427 | goto bail; |
428 | } |
429 | |
430 | osb->slot_info = (struct ocfs2_slot_info *)si; |
431 | bail: |
432 | if (status < 0) |
433 | __ocfs2_free_slot_info(si); |
434 | |
435 | return status; |
436 | } |
437 | |
438 | void ocfs2_free_slot_info(struct ocfs2_super *osb) |
439 | { |
440 | struct ocfs2_slot_info *si = osb->slot_info; |
441 | |
442 | osb->slot_info = NULL; |
443 | __ocfs2_free_slot_info(si); |
444 | } |
445 | |
446 | int ocfs2_find_slot(struct ocfs2_super *osb) |
447 | { |
448 | int status; |
449 | int slot; |
450 | struct ocfs2_slot_info *si; |
451 | |
452 | si = osb->slot_info; |
453 | |
454 | spin_lock(lock: &osb->osb_lock); |
455 | ocfs2_update_slot_info(si); |
456 | |
457 | /* search for ourselves first and take the slot if it already |
458 | * exists. Perhaps we need to mark this in a variable for our |
459 | * own journal recovery? Possibly not, though we certainly |
460 | * need to warn to the user */ |
461 | slot = __ocfs2_node_num_to_slot(si, node_num: osb->node_num); |
462 | if (slot < 0) { |
463 | /* if no slot yet, then just take 1st available |
464 | * one. */ |
465 | slot = __ocfs2_find_empty_slot(si, preferred: osb->preferred_slot); |
466 | if (slot < 0) { |
467 | spin_unlock(lock: &osb->osb_lock); |
468 | mlog(ML_ERROR, "no free slots available!\n" ); |
469 | status = -EINVAL; |
470 | goto bail; |
471 | } |
472 | } else |
473 | printk(KERN_INFO "ocfs2: Slot %d on device (%s) was already " |
474 | "allocated to this node!\n" , slot, osb->dev_str); |
475 | |
476 | ocfs2_set_slot(si, slot_num: slot, node_num: osb->node_num); |
477 | osb->slot_num = slot; |
478 | spin_unlock(lock: &osb->osb_lock); |
479 | |
480 | trace_ocfs2_find_slot(num: osb->slot_num); |
481 | |
482 | status = ocfs2_update_disk_slot(osb, si, slot_num: osb->slot_num); |
483 | if (status < 0) { |
484 | mlog_errno(status); |
485 | /* |
486 | * if write block failed, invalidate slot to avoid overwrite |
487 | * slot during dismount in case another node rightly has mounted |
488 | */ |
489 | spin_lock(lock: &osb->osb_lock); |
490 | ocfs2_invalidate_slot(si, slot_num: osb->slot_num); |
491 | osb->slot_num = OCFS2_INVALID_SLOT; |
492 | spin_unlock(lock: &osb->osb_lock); |
493 | } |
494 | |
495 | bail: |
496 | return status; |
497 | } |
498 | |
499 | void ocfs2_put_slot(struct ocfs2_super *osb) |
500 | { |
501 | int status, slot_num; |
502 | struct ocfs2_slot_info *si = osb->slot_info; |
503 | |
504 | if (!si) |
505 | return; |
506 | |
507 | spin_lock(lock: &osb->osb_lock); |
508 | ocfs2_update_slot_info(si); |
509 | |
510 | slot_num = osb->slot_num; |
511 | ocfs2_invalidate_slot(si, slot_num: osb->slot_num); |
512 | osb->slot_num = OCFS2_INVALID_SLOT; |
513 | spin_unlock(lock: &osb->osb_lock); |
514 | |
515 | status = ocfs2_update_disk_slot(osb, si, slot_num); |
516 | if (status < 0) |
517 | mlog_errno(status); |
518 | |
519 | ocfs2_free_slot_info(osb); |
520 | } |
521 | |