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:
7 * Copyright (C) 2002, 2004 Oracle. All rights reserved.
10 #include <linux/types.h>
11 #include <linux/slab.h>
12 #include <linux/highmem.h>
14 #include <cluster/masklog.h>
19 #include "extent_map.h"
20 #include "heartbeat.h"
25 #include "ocfs2_trace.h"
27 #include "buffer_head_io.h"
32 unsigned int sl_node_num;
35 struct ocfs2_slot_info {
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[];
46 static int __ocfs2_node_num_to_slot(struct ocfs2_slot_info *si,
47 unsigned int node_num);
49 static void ocfs2_invalidate_slot(struct ocfs2_slot_info *si,
52 BUG_ON((slot_num < 0) || (slot_num >= si->si_num_slots));
53 si->si_slots[slot_num].sl_valid = 0;
56 static void ocfs2_set_slot(struct ocfs2_slot_info *si,
57 int slot_num, unsigned int node_num)
59 BUG_ON((slot_num < 0) || (slot_num >= si->si_num_slots));
61 si->si_slots[slot_num].sl_valid = 1;
62 si->si_slots[slot_num].sl_node_num = node_num;
65 /* This version is for the extended slot map */
66 static void ocfs2_update_slot_info_extended(struct ocfs2_slot_info *si)
69 struct ocfs2_slot_map_extended *se;
72 for (b = 0; b < si->si_blocks; b++) {
73 se = (struct ocfs2_slot_map_extended *)si->si_bh[b]->b_data;
75 (i < si->si_slots_per_block) &&
76 (slotno < si->si_num_slots);
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));
82 ocfs2_invalidate_slot(si, slotno);
88 * Post the slot information on disk into our slot_info struct.
89 * Must be protected by osb_lock.
91 static void ocfs2_update_slot_info_old(struct ocfs2_slot_info *si)
94 struct ocfs2_slot_map *sm;
96 sm = (struct ocfs2_slot_map *)si->si_bh[0]->b_data;
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);
102 ocfs2_set_slot(si, i, le16_to_cpu(sm->sm_slots[i]));
106 static void ocfs2_update_slot_info(struct ocfs2_slot_info *si)
109 * The slot data will have been refreshed when ocfs2_super_lock
113 ocfs2_update_slot_info_extended(si);
115 ocfs2_update_slot_info_old(si);
118 int ocfs2_refresh_slot_info(struct ocfs2_super *osb)
121 struct ocfs2_slot_info *si = osb->slot_info;
126 BUG_ON(si->si_blocks == 0);
127 BUG_ON(si->si_bh == NULL);
129 trace_ocfs2_refresh_slot_info(si->si_blocks);
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.
136 ret = ocfs2_read_blocks(INODE_CACHE(si->si_inode), -1, si->si_blocks,
137 si->si_bh, OCFS2_BH_IGNORE_CACHE, NULL);
139 spin_lock(&osb->osb_lock);
140 ocfs2_update_slot_info(si);
141 spin_unlock(&osb->osb_lock);
147 /* post the our slot info stuff into it's destination bh and write it
149 static void ocfs2_update_disk_slot_extended(struct ocfs2_slot_info *si,
151 struct buffer_head **bh)
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;
157 BUG_ON(blkind >= si->si_blocks);
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];
167 static void ocfs2_update_disk_slot_old(struct ocfs2_slot_info *si,
169 struct buffer_head **bh)
172 struct ocfs2_slot_map *sm;
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)
178 cpu_to_le16(si->si_slots[i].sl_node_num);
180 sm->sm_slots[i] = cpu_to_le16(OCFS2_INVALID_SLOT);
185 static int ocfs2_update_disk_slot(struct ocfs2_super *osb,
186 struct ocfs2_slot_info *si,
190 struct buffer_head *bh;
192 spin_lock(&osb->osb_lock);
194 ocfs2_update_disk_slot_extended(si, slot_num, &bh);
196 ocfs2_update_disk_slot_old(si, slot_num, &bh);
197 spin_unlock(&osb->osb_lock);
199 status = ocfs2_write_block(osb, bh, INODE_CACHE(si->si_inode));
207 * Calculate how many bytes are needed by the slot map. Returns
208 * an error if the slot map file is too small.
210 static int ocfs2_slot_map_physical_size(struct ocfs2_super *osb,
212 unsigned long long *bytes)
214 unsigned long long bytes_needed;
216 if (ocfs2_uses_extended_slot_map(osb)) {
217 bytes_needed = osb->max_slots *
218 sizeof(struct ocfs2_extended_slot);
220 bytes_needed = osb->max_slots * sizeof(__le16);
222 if (bytes_needed > i_size_read(inode)) {
224 "Slot map file is too small! (size %llu, needed %llu)\n",
225 i_size_read(inode), bytes_needed);
229 *bytes = bytes_needed;
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)
238 int i, ret = -ENOENT;
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)) {
251 static int __ocfs2_find_empty_slot(struct ocfs2_slot_info *si,
254 int i, ret = -ENOSPC;
256 if ((preferred >= 0) && (preferred < si->si_num_slots)) {
257 if (!si->si_slots[preferred].sl_valid) {
263 for(i = 0; i < si->si_num_slots; i++) {
264 if (!si->si_slots[i].sl_valid) {
273 int ocfs2_node_num_to_slot(struct ocfs2_super *osb, unsigned int node_num)
276 struct ocfs2_slot_info *si = osb->slot_info;
278 spin_lock(&osb->osb_lock);
279 slot = __ocfs2_node_num_to_slot(si, node_num);
280 spin_unlock(&osb->osb_lock);
285 int ocfs2_slot_to_node_num_locked(struct ocfs2_super *osb, int slot_num,
286 unsigned int *node_num)
288 struct ocfs2_slot_info *si = osb->slot_info;
290 assert_spin_locked(&osb->osb_lock);
292 BUG_ON(slot_num < 0);
293 BUG_ON(slot_num >= osb->max_slots);
295 if (!si->si_slots[slot_num].sl_valid)
298 *node_num = si->si_slots[slot_num].sl_node_num;
302 static void __ocfs2_free_slot_info(struct ocfs2_slot_info *si)
311 for (i = 0; i < si->si_blocks; i++) {
313 brelse(si->si_bh[i]);
323 int ocfs2_clear_slot(struct ocfs2_super *osb, int slot_num)
325 struct ocfs2_slot_info *si = osb->slot_info;
330 spin_lock(&osb->osb_lock);
331 ocfs2_invalidate_slot(si, slot_num);
332 spin_unlock(&osb->osb_lock);
334 return ocfs2_update_disk_slot(osb, osb->slot_info, slot_num);
337 static int ocfs2_map_slot_buffers(struct ocfs2_super *osb,
338 struct ocfs2_slot_info *si)
342 unsigned long long blocks, bytes = 0;
344 struct buffer_head *bh;
346 status = ocfs2_slot_map_physical_size(osb, si->si_inode, &bytes);
350 blocks = ocfs2_blocks_for_bytes(si->si_inode->i_sb, bytes);
351 BUG_ON(blocks > UINT_MAX);
352 si->si_blocks = blocks;
357 si->si_slots_per_block =
358 (osb->sb->s_blocksize /
359 sizeof(struct ocfs2_extended_slot));
361 si->si_slots_per_block = osb->sb->s_blocksize / sizeof(__le16);
363 /* The size checks above should ensure this */
364 BUG_ON((osb->max_slots / si->si_slots_per_block) > blocks);
366 trace_ocfs2_map_slot_buffers(bytes, si->si_blocks);
368 si->si_bh = kcalloc(si->si_blocks, sizeof(struct buffer_head *),
376 for (i = 0; i < si->si_blocks; i++) {
377 status = ocfs2_extent_map_get_blocks(si->si_inode, i,
384 trace_ocfs2_map_slot_buffers_block((unsigned long long)blkno, i);
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);
401 int ocfs2_init_slot_info(struct ocfs2_super *osb)
404 struct inode *inode = NULL;
405 struct ocfs2_slot_info *si;
407 si = kzalloc(struct_size(si, si_slots, osb->max_slots), GFP_KERNEL);
414 si->si_extended = ocfs2_uses_extended_slot_map(osb);
415 si->si_num_slots = osb->max_slots;
417 inode = ocfs2_get_system_file_inode(osb, SLOT_MAP_SYSTEM_INODE,
425 si->si_inode = inode;
426 status = ocfs2_map_slot_buffers(osb, si);
432 osb->slot_info = (struct ocfs2_slot_info *)si;
435 __ocfs2_free_slot_info(si);
440 void ocfs2_free_slot_info(struct ocfs2_super *osb)
442 struct ocfs2_slot_info *si = osb->slot_info;
444 osb->slot_info = NULL;
445 __ocfs2_free_slot_info(si);
448 int ocfs2_find_slot(struct ocfs2_super *osb)
452 struct ocfs2_slot_info *si;
456 spin_lock(&osb->osb_lock);
457 ocfs2_update_slot_info(si);
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);
465 /* if no slot yet, then just take 1st available
467 slot = __ocfs2_find_empty_slot(si, osb->preferred_slot);
469 spin_unlock(&osb->osb_lock);
470 mlog(ML_ERROR, "no free slots available!\n");
475 printk(KERN_INFO "ocfs2: Slot %d on device (%s) was already "
476 "allocated to this node!\n", slot, osb->dev_str);
478 ocfs2_set_slot(si, slot, osb->node_num);
479 osb->slot_num = slot;
480 spin_unlock(&osb->osb_lock);
482 trace_ocfs2_find_slot(osb->slot_num);
484 status = ocfs2_update_disk_slot(osb, si, osb->slot_num);
488 * if write block failed, invalidate slot to avoid overwrite
489 * slot during dismount in case another node rightly has mounted
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);
501 void ocfs2_put_slot(struct ocfs2_super *osb)
503 int status, slot_num;
504 struct ocfs2_slot_info *si = osb->slot_info;
509 spin_lock(&osb->osb_lock);
510 ocfs2_update_slot_info(si);
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);
517 status = ocfs2_update_disk_slot(osb, si, slot_num);
521 ocfs2_free_slot_info(osb);