4 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
6 * This program is free software; you can redistribute it and/or modify
7 * it under the terms of the GNU General Public License version 2 only,
8 * as published by the Free Software Foundation.
10 * This program is distributed in the hope that it will be useful, but
11 * WITHOUT ANY WARRANTY; without even the implied warranty of
12 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
13 * General Public License version 2 for more details (a copy is included
14 * in the LICENSE file that accompanied this code).
16 * You should have received a copy of the GNU General Public License
17 * version 2 along with this program; if not, write to the
18 * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
19 * Boston, MA 021110-1307, USA
24 * Copyright (c) 2012, 2016, Intel Corporation.
27 * lustre/target/tgt_main.c
29 * Lustre Unified Target main initialization code
31 * Author: Mikhail Pershin <mike.pershin@intel.com>
34 #define DEBUG_SUBSYSTEM S_CLASS
37 #include "tgt_internal.h"
38 #include "../ptlrpc/ptlrpc_internal.h"
41 * Save cross-MDT lock in lut_slc_locks.
43 * Lock R/W count is not saved, but released in unlock (not canceled remotely),
44 * instead only a refcount is taken, so that the remote MDT where the object
45 * resides can detect conflict with this lock there.
48 * \param lock cross-MDT lock to save
49 * \param transno when the transaction with this transno is committed, this lock
52 void tgt_save_slc_lock(struct lu_target *lut, struct ldlm_lock *lock,
55 spin_lock(&lut->lut_slc_locks_guard);
56 lock_res_and_lock(lock);
57 if (ldlm_is_cbpending(lock)) {
58 /* if it was canceld by server, don't save, because remote MDT
59 * will do Sync-on-Cancel. */
62 lock->l_transno = transno;
63 /* if this lock is in the list already, there are two operations
64 * both use this lock, and save it after use, so for the second
65 * one, just put the refcount. */
66 if (list_empty(&lock->l_slc_link))
67 list_add_tail(&lock->l_slc_link, &lut->lut_slc_locks);
71 unlock_res_and_lock(lock);
72 spin_unlock(&lut->lut_slc_locks_guard);
74 EXPORT_SYMBOL(tgt_save_slc_lock);
77 * Discard cross-MDT lock from lut_slc_locks.
79 * This is called upon BAST, just remove lock from lut_slc_locks and put lock
80 * refcount. The BAST will cancel this lock.
83 * \param lock cross-MDT lock to discard
85 void tgt_discard_slc_lock(struct lu_target *lut, struct ldlm_lock *lock)
87 spin_lock(&lut->lut_slc_locks_guard);
88 lock_res_and_lock(lock);
89 /* may race with tgt_cancel_slc_locks() */
90 if (lock->l_transno != 0) {
91 LASSERT(!list_empty(&lock->l_slc_link));
92 LASSERT(ldlm_is_cbpending(lock));
93 list_del_init(&lock->l_slc_link);
97 unlock_res_and_lock(lock);
98 spin_unlock(&lut->lut_slc_locks_guard);
100 EXPORT_SYMBOL(tgt_discard_slc_lock);
103 * Cancel cross-MDT locks upon transaction commit.
105 * Remove cross-MDT locks from lut_slc_locks, cancel them and put lock refcount.
108 * \param transno transaction with this number was committed.
110 void tgt_cancel_slc_locks(struct lu_target *lut, __u64 transno)
112 struct ldlm_lock *lock, *next;
114 struct lustre_handle lockh;
116 spin_lock(&lut->lut_slc_locks_guard);
117 list_for_each_entry_safe(lock, next, &lut->lut_slc_locks,
119 lock_res_and_lock(lock);
120 LASSERT(lock->l_transno != 0);
121 if (lock->l_transno > transno) {
122 unlock_res_and_lock(lock);
125 /* ouch, another operation is using it after it's saved */
126 if (lock->l_readers != 0 || lock->l_writers != 0) {
127 unlock_res_and_lock(lock);
130 /* set CBPENDING so that this lock won't be used again */
131 ldlm_set_cbpending(lock);
133 list_move(&lock->l_slc_link, &list);
134 unlock_res_and_lock(lock);
136 spin_unlock(&lut->lut_slc_locks_guard);
138 list_for_each_entry_safe(lock, next, &list, l_slc_link) {
139 list_del_init(&lock->l_slc_link);
140 ldlm_lock2handle(lock, &lockh);
141 ldlm_cli_cancel(&lockh, LCF_ASYNC);
146 int tgt_init(const struct lu_env *env, struct lu_target *lut,
147 struct obd_device *obd, struct dt_device *dt,
148 struct tgt_opc_slice *slice, int request_fail_id,
151 struct dt_object_format dof;
162 lut->lut_bottom = dt;
163 lut->lut_last_rcvd = NULL;
164 lut->lut_client_bitmap = NULL;
165 atomic_set(&lut->lut_num_clients, 0);
166 atomic_set(&lut->lut_client_generation, 0);
167 lut->lut_reply_data = NULL;
168 lut->lut_reply_bitmap = NULL;
169 obd->u.obt.obt_lut = lut;
170 obd->u.obt.obt_magic = OBT_MAGIC;
172 /* set request handler slice and parameters */
173 lut->lut_slice = slice;
174 lut->lut_reply_fail_id = reply_fail_id;
175 lut->lut_request_fail_id = request_fail_id;
177 /* sptlrcp variables init */
178 rwlock_init(&lut->lut_sptlrpc_lock);
179 sptlrpc_rule_set_init(&lut->lut_sptlrpc_rset);
181 spin_lock_init(&lut->lut_flags_lock);
182 lut->lut_sync_lock_cancel = NEVER_SYNC_ON_CANCEL;
184 spin_lock_init(&lut->lut_slc_locks_guard);
185 INIT_LIST_HEAD(&lut->lut_slc_locks);
187 /* last_rcvd initialization is needed by replayable targets only */
188 if (!obd->obd_replayable)
191 spin_lock_init(&lut->lut_translock);
192 spin_lock_init(&lut->lut_client_bitmap_lock);
194 OBD_ALLOC(lut->lut_client_bitmap, LR_MAX_CLIENTS >> 3);
195 if (lut->lut_client_bitmap == NULL)
198 memset(&attr, 0, sizeof(attr));
199 attr.la_valid = LA_MODE;
200 attr.la_mode = S_IFREG | S_IRUGO | S_IWUSR;
201 dof.dof_type = dt_mode_to_dft(S_IFREG);
203 lu_local_obj_fid(&fid, LAST_RECV_OID);
205 o = dt_find_or_create(env, lut->lut_bottom, &fid, &dof, &attr);
208 CERROR("%s: cannot open LAST_RCVD: rc = %d\n", tgt_name(lut),
213 lut->lut_last_rcvd = o;
214 rc = tgt_server_data_init(env, lut);
218 /* prepare transactions callbacks */
219 lut->lut_txn_cb.dtc_txn_start = tgt_txn_start_cb;
220 lut->lut_txn_cb.dtc_txn_stop = tgt_txn_stop_cb;
221 lut->lut_txn_cb.dtc_txn_commit = NULL;
222 lut->lut_txn_cb.dtc_cookie = lut;
223 lut->lut_txn_cb.dtc_tag = LCT_DT_THREAD | LCT_MD_THREAD;
224 INIT_LIST_HEAD(&lut->lut_txn_cb.dtc_linkage);
226 dt_txn_callback_add(lut->lut_bottom, &lut->lut_txn_cb);
227 lut->lut_bottom->dd_lu_dev.ld_site->ls_tgt = lut;
229 /* reply_data is supported by MDT targets only for now */
230 if (strncmp(obd->obd_type->typ_name, LUSTRE_MDT_NAME, 3) != 0)
233 OBD_ALLOC(lut->lut_reply_bitmap,
234 LUT_REPLY_SLOTS_MAX_CHUNKS * sizeof(unsigned long *));
235 if (lut->lut_reply_bitmap == NULL)
236 GOTO(out, rc = -ENOMEM);
238 memset(&attr, 0, sizeof(attr));
239 attr.la_valid = LA_MODE;
240 attr.la_mode = S_IFREG | S_IRUGO | S_IWUSR;
241 dof.dof_type = dt_mode_to_dft(S_IFREG);
243 lu_local_obj_fid(&fid, REPLY_DATA_OID);
245 o = dt_find_or_create(env, lut->lut_bottom, &fid, &dof, &attr);
248 CERROR("%s: cannot open REPLY_DATA: rc = %d\n", tgt_name(lut),
252 lut->lut_reply_data = o;
254 rc = tgt_reply_data_init(env, lut);
258 atomic_set(&lut->lut_sync_count, 0);
263 dt_txn_callback_del(lut->lut_bottom, &lut->lut_txn_cb);
265 obd->u.obt.obt_magic = 0;
266 obd->u.obt.obt_lut = NULL;
267 if (lut->lut_last_rcvd != NULL) {
268 lu_object_put(env, &lut->lut_last_rcvd->do_lu);
269 lut->lut_last_rcvd = NULL;
271 if (lut->lut_client_bitmap != NULL)
272 OBD_FREE(lut->lut_client_bitmap, LR_MAX_CLIENTS >> 3);
273 lut->lut_client_bitmap = NULL;
274 if (lut->lut_reply_data != NULL)
275 lu_object_put(env, &lut->lut_reply_data->do_lu);
276 lut->lut_reply_data = NULL;
277 if (lut->lut_reply_bitmap != NULL) {
278 for (i = 0; i < LUT_REPLY_SLOTS_MAX_CHUNKS; i++) {
279 if (lut->lut_reply_bitmap[i] != NULL)
280 OBD_FREE_LARGE(lut->lut_reply_bitmap[i],
281 BITS_TO_LONGS(LUT_REPLY_SLOTS_PER_CHUNK) *
283 lut->lut_reply_bitmap[i] = NULL;
285 OBD_FREE(lut->lut_reply_bitmap,
286 LUT_REPLY_SLOTS_MAX_CHUNKS * sizeof(unsigned long *));
288 lut->lut_reply_bitmap = NULL;
291 EXPORT_SYMBOL(tgt_init);
293 void tgt_fini(const struct lu_env *env, struct lu_target *lut)
299 if (lut->lut_lsd.lsd_feature_incompat & OBD_INCOMPAT_MULTI_RPCS &&
300 atomic_read(&lut->lut_num_clients) == 0) {
301 /* Clear MULTI RPCS incompatibility flag that prevents previous
302 * Lustre versions to mount a target with reply_data file */
303 lut->lut_lsd.lsd_feature_incompat &= ~OBD_INCOMPAT_MULTI_RPCS;
304 rc = tgt_server_data_update(env, lut, 1);
306 CERROR("%s: unable to clear MULTI RPCS "
307 "incompatibility flag\n",
308 lut->lut_obd->obd_name);
311 sptlrpc_rule_set_free(&lut->lut_sptlrpc_rset);
313 if (lut->lut_reply_data != NULL)
314 lu_object_put(env, &lut->lut_reply_data->do_lu);
315 lut->lut_reply_data = NULL;
316 if (lut->lut_reply_bitmap != NULL) {
317 for (i = 0; i < LUT_REPLY_SLOTS_MAX_CHUNKS; i++) {
318 if (lut->lut_reply_bitmap[i] != NULL)
319 OBD_FREE_LARGE(lut->lut_reply_bitmap[i],
320 BITS_TO_LONGS(LUT_REPLY_SLOTS_PER_CHUNK) *
322 lut->lut_reply_bitmap[i] = NULL;
324 OBD_FREE(lut->lut_reply_bitmap,
325 LUT_REPLY_SLOTS_MAX_CHUNKS * sizeof(unsigned long *));
327 lut->lut_reply_bitmap = NULL;
328 if (lut->lut_client_bitmap) {
329 OBD_FREE(lut->lut_client_bitmap, LR_MAX_CLIENTS >> 3);
330 lut->lut_client_bitmap = NULL;
332 if (lut->lut_last_rcvd) {
333 dt_txn_callback_del(lut->lut_bottom, &lut->lut_txn_cb);
334 lu_object_put(env, &lut->lut_last_rcvd->do_lu);
335 lut->lut_last_rcvd = NULL;
339 EXPORT_SYMBOL(tgt_fini);
341 /* context key constructor/destructor: tg_key_init, tg_key_fini */
342 LU_KEY_INIT(tgt, struct tgt_thread_info);
344 static void tgt_key_fini(const struct lu_context *ctx,
345 struct lu_context_key *key, void *data)
347 struct tgt_thread_info *info = data;
348 struct thandle_exec_args *args = &info->tti_tea;
351 for (i = 0; i < args->ta_alloc_args; i++) {
352 if (args->ta_args[i] != NULL)
353 OBD_FREE_PTR(args->ta_args[i]);
356 if (args->ta_args != NULL)
357 OBD_FREE(args->ta_args, sizeof(args->ta_args[0]) *
358 args->ta_alloc_args);
362 static void tgt_key_exit(const struct lu_context *ctx,
363 struct lu_context_key *key, void *data)
365 struct tgt_thread_info *tti = data;
367 tti->tti_has_trans = 0;
368 tti->tti_mult_trans = 0;
371 /* context key: tg_thread_key */
372 struct lu_context_key tgt_thread_key = {
373 .lct_tags = LCT_MD_THREAD | LCT_DT_THREAD,
374 .lct_init = tgt_key_init,
375 .lct_fini = tgt_key_fini,
376 .lct_exit = tgt_key_exit,
379 LU_KEY_INIT_GENERIC(tgt);
381 /* context key constructor/destructor: tgt_ses_key_init, tgt_ses_key_fini */
382 LU_KEY_INIT_FINI(tgt_ses, struct tgt_session_info);
384 /* context key: tgt_session_key */
385 struct lu_context_key tgt_session_key = {
386 .lct_tags = LCT_SERVER_SESSION,
387 .lct_init = tgt_ses_key_init,
388 .lct_fini = tgt_ses_key_fini,
390 EXPORT_SYMBOL(tgt_session_key);
392 LU_KEY_INIT_GENERIC(tgt_ses);
395 * this page is allocated statically when module is initializing
396 * it is used to simulate data corruptions, see ost_checksum_bulk()
397 * for details. as the original pages provided by the layers below
398 * can be remain in the internal cache, we do not want to modify
401 struct page *tgt_page_to_corrupt;
403 int tgt_mod_init(void)
407 tgt_page_to_corrupt = alloc_page(GFP_KERNEL);
409 tgt_key_init_generic(&tgt_thread_key, NULL);
410 lu_context_key_register_many(&tgt_thread_key, NULL);
412 tgt_ses_key_init_generic(&tgt_session_key, NULL);
413 lu_context_key_register_many(&tgt_session_key, NULL);
420 void tgt_mod_exit(void)
422 if (tgt_page_to_corrupt != NULL)
423 put_page(tgt_page_to_corrupt);
425 lu_context_key_degister(&tgt_thread_key);
426 lu_context_key_degister(&tgt_session_key);