Merge branch 'akpm' (fixes from Andrew)
[linux-2.6/cjktty.git] / fs / dlm / dir.c
blob278a75cda4463006d4b2c2e85d75c7728d5b4526
1 /******************************************************************************
2 *******************************************************************************
3 **
4 ** Copyright (C) Sistina Software, Inc. 1997-2003 All rights reserved.
5 ** Copyright (C) 2004-2005 Red Hat, Inc. All rights reserved.
6 **
7 ** This copyrighted material is made available to anyone wishing to use,
8 ** modify, copy, or redistribute it subject to the terms and conditions
9 ** of the GNU General Public License v.2.
11 *******************************************************************************
12 ******************************************************************************/
14 #include "dlm_internal.h"
15 #include "lockspace.h"
16 #include "member.h"
17 #include "lowcomms.h"
18 #include "rcom.h"
19 #include "config.h"
20 #include "memory.h"
21 #include "recover.h"
22 #include "util.h"
23 #include "lock.h"
24 #include "dir.h"
27 * We use the upper 16 bits of the hash value to select the directory node.
28 * Low bits are used for distribution of rsb's among hash buckets on each node.
30 * To give the exact range wanted (0 to num_nodes-1), we apply a modulus of
31 * num_nodes to the hash value. This value in the desired range is used as an
32 * offset into the sorted list of nodeid's to give the particular nodeid.
35 int dlm_hash2nodeid(struct dlm_ls *ls, uint32_t hash)
37 uint32_t node;
39 if (ls->ls_num_nodes == 1)
40 return dlm_our_nodeid();
41 else {
42 node = (hash >> 16) % ls->ls_total_weight;
43 return ls->ls_node_array[node];
47 int dlm_dir_nodeid(struct dlm_rsb *r)
49 return r->res_dir_nodeid;
52 void dlm_recover_dir_nodeid(struct dlm_ls *ls)
54 struct dlm_rsb *r;
56 down_read(&ls->ls_root_sem);
57 list_for_each_entry(r, &ls->ls_root_list, res_root_list) {
58 r->res_dir_nodeid = dlm_hash2nodeid(ls, r->res_hash);
60 up_read(&ls->ls_root_sem);
63 int dlm_recover_directory(struct dlm_ls *ls)
65 struct dlm_member *memb;
66 char *b, *last_name = NULL;
67 int error = -ENOMEM, last_len, nodeid, result;
68 uint16_t namelen;
69 unsigned int count = 0, count_match = 0, count_bad = 0, count_add = 0;
71 log_debug(ls, "dlm_recover_directory");
73 if (dlm_no_directory(ls))
74 goto out_status;
76 last_name = kmalloc(DLM_RESNAME_MAXLEN, GFP_NOFS);
77 if (!last_name)
78 goto out;
80 list_for_each_entry(memb, &ls->ls_nodes, list) {
81 if (memb->nodeid == dlm_our_nodeid())
82 continue;
84 memset(last_name, 0, DLM_RESNAME_MAXLEN);
85 last_len = 0;
87 for (;;) {
88 int left;
89 error = dlm_recovery_stopped(ls);
90 if (error)
91 goto out_free;
93 error = dlm_rcom_names(ls, memb->nodeid,
94 last_name, last_len);
95 if (error)
96 goto out_free;
98 cond_resched();
101 * pick namelen/name pairs out of received buffer
104 b = ls->ls_recover_buf->rc_buf;
105 left = ls->ls_recover_buf->rc_header.h_length;
106 left -= sizeof(struct dlm_rcom);
108 for (;;) {
109 __be16 v;
111 error = -EINVAL;
112 if (left < sizeof(__be16))
113 goto out_free;
115 memcpy(&v, b, sizeof(__be16));
116 namelen = be16_to_cpu(v);
117 b += sizeof(__be16);
118 left -= sizeof(__be16);
120 /* namelen of 0xFFFFF marks end of names for
121 this node; namelen of 0 marks end of the
122 buffer */
124 if (namelen == 0xFFFF)
125 goto done;
126 if (!namelen)
127 break;
129 if (namelen > left)
130 goto out_free;
132 if (namelen > DLM_RESNAME_MAXLEN)
133 goto out_free;
135 error = dlm_master_lookup(ls, memb->nodeid,
136 b, namelen,
137 DLM_LU_RECOVER_DIR,
138 &nodeid, &result);
139 if (error) {
140 log_error(ls, "recover_dir lookup %d",
141 error);
142 goto out_free;
145 /* The name was found in rsbtbl, but the
146 * master nodeid is different from
147 * memb->nodeid which says it is the master.
148 * This should not happen. */
150 if (result == DLM_LU_MATCH &&
151 nodeid != memb->nodeid) {
152 count_bad++;
153 log_error(ls, "recover_dir lookup %d "
154 "nodeid %d memb %d bad %u",
155 result, nodeid, memb->nodeid,
156 count_bad);
157 print_hex_dump_bytes("dlm_recover_dir ",
158 DUMP_PREFIX_NONE,
159 b, namelen);
162 /* The name was found in rsbtbl, and the
163 * master nodeid matches memb->nodeid. */
165 if (result == DLM_LU_MATCH &&
166 nodeid == memb->nodeid) {
167 count_match++;
170 /* The name was not found in rsbtbl and was
171 * added with memb->nodeid as the master. */
173 if (result == DLM_LU_ADD) {
174 count_add++;
177 last_len = namelen;
178 memcpy(last_name, b, namelen);
179 b += namelen;
180 left -= namelen;
181 count++;
184 done:
188 out_status:
189 error = 0;
190 dlm_set_recover_status(ls, DLM_RS_DIR);
192 log_debug(ls, "dlm_recover_directory %u in %u new",
193 count, count_add);
194 out_free:
195 kfree(last_name);
196 out:
197 return error;
200 static struct dlm_rsb *find_rsb_root(struct dlm_ls *ls, char *name, int len)
202 struct dlm_rsb *r;
203 uint32_t hash, bucket;
204 int rv;
206 hash = jhash(name, len, 0);
207 bucket = hash & (ls->ls_rsbtbl_size - 1);
209 spin_lock(&ls->ls_rsbtbl[bucket].lock);
210 rv = dlm_search_rsb_tree(&ls->ls_rsbtbl[bucket].keep, name, len, &r);
211 if (rv)
212 rv = dlm_search_rsb_tree(&ls->ls_rsbtbl[bucket].toss,
213 name, len, &r);
214 spin_unlock(&ls->ls_rsbtbl[bucket].lock);
216 if (!rv)
217 return r;
219 down_read(&ls->ls_root_sem);
220 list_for_each_entry(r, &ls->ls_root_list, res_root_list) {
221 if (len == r->res_length && !memcmp(name, r->res_name, len)) {
222 up_read(&ls->ls_root_sem);
223 log_debug(ls, "find_rsb_root revert to root_list %s",
224 r->res_name);
225 return r;
228 up_read(&ls->ls_root_sem);
229 return NULL;
232 /* Find the rsb where we left off (or start again), then send rsb names
233 for rsb's we're master of and whose directory node matches the requesting
234 node. inbuf is the rsb name last sent, inlen is the name's length */
236 void dlm_copy_master_names(struct dlm_ls *ls, char *inbuf, int inlen,
237 char *outbuf, int outlen, int nodeid)
239 struct list_head *list;
240 struct dlm_rsb *r;
241 int offset = 0, dir_nodeid;
242 __be16 be_namelen;
244 down_read(&ls->ls_root_sem);
246 if (inlen > 1) {
247 r = find_rsb_root(ls, inbuf, inlen);
248 if (!r) {
249 inbuf[inlen - 1] = '\0';
250 log_error(ls, "copy_master_names from %d start %d %s",
251 nodeid, inlen, inbuf);
252 goto out;
254 list = r->res_root_list.next;
255 } else {
256 list = ls->ls_root_list.next;
259 for (offset = 0; list != &ls->ls_root_list; list = list->next) {
260 r = list_entry(list, struct dlm_rsb, res_root_list);
261 if (r->res_nodeid)
262 continue;
264 dir_nodeid = dlm_dir_nodeid(r);
265 if (dir_nodeid != nodeid)
266 continue;
269 * The block ends when we can't fit the following in the
270 * remaining buffer space:
271 * namelen (uint16_t) +
272 * name (r->res_length) +
273 * end-of-block record 0x0000 (uint16_t)
276 if (offset + sizeof(uint16_t)*2 + r->res_length > outlen) {
277 /* Write end-of-block record */
278 be_namelen = cpu_to_be16(0);
279 memcpy(outbuf + offset, &be_namelen, sizeof(__be16));
280 offset += sizeof(__be16);
281 ls->ls_recover_dir_sent_msg++;
282 goto out;
285 be_namelen = cpu_to_be16(r->res_length);
286 memcpy(outbuf + offset, &be_namelen, sizeof(__be16));
287 offset += sizeof(__be16);
288 memcpy(outbuf + offset, r->res_name, r->res_length);
289 offset += r->res_length;
290 ls->ls_recover_dir_sent_res++;
294 * If we've reached the end of the list (and there's room) write a
295 * terminating record.
298 if ((list == &ls->ls_root_list) &&
299 (offset + sizeof(uint16_t) <= outlen)) {
300 be_namelen = cpu_to_be16(0xFFFF);
301 memcpy(outbuf + offset, &be_namelen, sizeof(__be16));
302 offset += sizeof(__be16);
303 ls->ls_recover_dir_sent_msg++;
305 out:
306 up_read(&ls->ls_root_sem);