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, see
18 * http://www.gnu.org/licenses/gpl-2.0.html
23 * Copyright (c) 2009, 2010, Oracle and/or its affiliates. All rights reserved.
24 * Use is subject to license terms.
26 * Copyright (c) 2012, 2017, Intel Corporation.
29 * This file is part of Lustre, http://www.lustre.org/
31 * lustre/osd-zfs/osd_handler.c
32 * Top-level entry points into osd module
34 * Author: Alex Zhuravlev <bzzz@whamcloud.com>
35 * Author: Mike Pershin <tappro@whamcloud.com>
36 * Author: Johann Lombardi <johann@whamcloud.com>
39 #define DEBUG_SUBSYSTEM S_OSD
41 #include <libcfs/libcfs.h>
42 #include <obd_support.h>
43 #include <lustre_net.h>
45 #include <obd_class.h>
46 #include <lustre_disk.h>
47 #include <lustre_fid.h>
48 #include <uapi/linux/lustre/lustre_param.h>
49 #include <md_object.h>
51 #include "osd_internal.h"
53 #include <sys/dnode.h>
58 #include <sys/spa_impl.h>
59 #include <sys/zfs_znode.h>
60 #include <sys/dmu_tx.h>
61 #include <sys/dmu_objset.h>
62 #include <sys/dsl_prop.h>
63 #include <sys/sa_impl.h>
66 struct lu_context_key osd_key;
68 static int osd_txg_sync_delay_us = -1;
70 /* Slab for OSD object allocation */
71 struct kmem_cache *osd_object_kmem;
73 /* Slab to allocate osd_zap_it */
74 struct kmem_cache *osd_zapit_cachep;
76 static struct lu_kmem_descr osd_caches[] = {
78 .ckd_cache = &osd_object_kmem,
79 .ckd_name = "zfs_osd_obj",
80 .ckd_size = sizeof(struct osd_object)
83 .ckd_cache = &osd_zapit_cachep,
84 .ckd_name = "osd_zapit_cache",
85 .ckd_size = sizeof(struct osd_zap_it)
92 static void arc_prune_func(int64_t bytes, void *private)
94 struct osd_device *od = private;
95 struct lu_site *site = &od->od_site;
99 rc = lu_env_init(&env, LCT_SHRINKER);
101 CERROR("%s: can't initialize shrinker env: rc = %d\n",
106 lu_site_purge(&env, site, (bytes >> 10));
112 * Concurrency: doesn't access mutable data
114 static int osd_root_get(const struct lu_env *env,
115 struct dt_device *dev, struct lu_fid *f)
117 lu_local_obj_fid(f, OSD_FS_ROOT_OID);
122 * OSD object methods.
126 * Concurrency: shouldn't matter.
128 static void osd_trans_commit_cb(void *cb_data, int error)
130 struct osd_thandle *oh = cb_data;
131 struct thandle *th = &oh->ot_super;
132 struct osd_device *osd = osd_dt_dev(th->th_dev);
133 struct lu_device *lud = &th->th_dev->dd_lu_dev;
134 struct dt_txn_commit_cb *dcb, *tmp;
139 if (error == ECANCELED)
140 CWARN("%s: transaction @0x%p was aborted\n",
141 osd_dt_dev(th->th_dev)->od_svname, th);
143 CERROR("%s: transaction @0x%p commit error: rc = %d\n",
144 osd_dt_dev(th->th_dev)->od_svname, th, error);
147 /* call per-transaction callbacks if any */
148 list_for_each_entry_safe(dcb, tmp, &oh->ot_dcb_list, dcb_linkage) {
149 LASSERTF(dcb->dcb_magic == TRANS_COMMIT_CB_MAGIC,
150 "commit callback entry: magic=%x name='%s'\n",
151 dcb->dcb_magic, dcb->dcb_name);
152 list_del_init(&dcb->dcb_linkage);
153 dcb->dcb_func(NULL, th, dcb, error);
156 /* Unlike ldiskfs, zfs updates space accounting at commit time.
157 * As a consequence, op_end is called only now to inform the quota slave
158 * component that reserved quota space is now accounted in usage and
159 * should be released. Quota space won't be adjusted at this point since
160 * we can't provide a suitable environment. It will be performed
161 * asynchronously by a lquota thread. */
162 qsd_op_end(NULL, osd->od_quota_slave_dt, &oh->ot_quota_trans);
163 if (osd->od_quota_slave_md != NULL)
164 qsd_op_end(NULL, osd->od_quota_slave_md, &oh->ot_quota_trans);
173 static int osd_trans_cb_add(struct thandle *th, struct dt_txn_commit_cb *dcb)
175 struct osd_thandle *oh = container_of(th, struct osd_thandle,
178 LASSERT(dcb->dcb_magic == TRANS_COMMIT_CB_MAGIC);
179 LASSERT(&dcb->dcb_func != NULL);
180 if (dcb->dcb_flags & DCB_TRANS_STOP)
181 list_add(&dcb->dcb_linkage, &oh->ot_stop_dcb_list);
183 list_add(&dcb->dcb_linkage, &oh->ot_dcb_list);
189 * Concurrency: shouldn't matter.
191 static int osd_trans_start(const struct lu_env *env, struct dt_device *d,
194 struct osd_device *osd = osd_dt_dev(d);
195 struct osd_thandle *oh;
200 oh = container_of(th, struct osd_thandle, ot_super);
204 rc = dt_txn_hook_start(env, d, th);
206 CERROR("%s: dt_txn_hook_start failed: rc = %d\n",
211 if (OBD_FAIL_CHECK(OBD_FAIL_OSD_TXN_START))
212 /* Unlike ldiskfs, ZFS checks for available space and returns
213 * -ENOSPC when assigning txg */
216 rc = -dmu_tx_assign(oh->ot_tx, TXG_WAIT);
217 if (unlikely(rc != 0)) {
218 /* dmu will call commit callback with error code during abort */
219 if (!lu_device_is_md(&d->dd_lu_dev) && rc == -ENOSPC)
220 CERROR("%s: failed to start transaction due to ENOSPC"
221 "\n", osd->od_svname);
223 CERROR("%s: can't assign tx: rc = %d\n",
226 /* add commit callback */
227 dmu_tx_callback_register(oh->ot_tx, osd_trans_commit_cb, oh);
229 osd_oti_get(env)->oti_in_trans = 1;
230 lu_device_get(&d->dd_lu_dev);
236 static void osd_unlinked_list_emptify(const struct lu_env *env,
237 struct osd_device *osd,
238 struct list_head *list, bool free)
240 struct osd_object *obj;
243 while (!list_empty(list)) {
244 obj = list_entry(list->next,
245 struct osd_object, oo_unlinked_linkage);
246 LASSERT(obj->oo_dn != NULL);
247 oid = obj->oo_dn->dn_object;
249 list_del_init(&obj->oo_unlinked_linkage);
251 (void)osd_unlinked_object_free(env, osd, oid);
255 static void osd_trans_stop_cb(struct osd_thandle *oth, int result)
257 struct dt_txn_commit_cb *dcb;
258 struct dt_txn_commit_cb *tmp;
260 /* call per-transaction stop callbacks if any */
261 list_for_each_entry_safe(dcb, tmp, &oth->ot_stop_dcb_list,
263 LASSERTF(dcb->dcb_magic == TRANS_COMMIT_CB_MAGIC,
264 "commit callback entry: magic=%x name='%s'\n",
265 dcb->dcb_magic, dcb->dcb_name);
266 list_del_init(&dcb->dcb_linkage);
267 dcb->dcb_func(NULL, &oth->ot_super, dcb, result);
272 * Concurrency: shouldn't matter.
274 static int osd_trans_stop(const struct lu_env *env, struct dt_device *dt,
277 struct osd_device *osd = osd_dt_dev(th->th_dev);
278 bool sync = (th->th_sync != 0);
279 struct osd_thandle *oh;
285 oh = container_of(th, struct osd_thandle, ot_super);
286 list_splice_init(&oh->ot_unlinked_list, &unlinked);
288 osd_oti_get(env)->oti_ins_cache_depth--;
289 /* reset OI cache for safety */
290 if (osd_oti_get(env)->oti_ins_cache_depth == 0)
291 osd_oti_get(env)->oti_ins_cache_used = 0;
293 if (oh->ot_assigned == 0) {
295 CDEBUG(D_OTHER, "%s: transaction is aborted\n", osd->od_svname);
296 osd_trans_stop_cb(oh, th->th_result);
297 dmu_tx_abort(oh->ot_tx);
298 osd_object_sa_dirty_rele(env, oh);
299 osd_unlinked_list_emptify(env, osd, &unlinked, false);
300 /* there won't be any commit, release reserved quota space now,
302 qsd_op_end(env, osd->od_quota_slave_dt, &oh->ot_quota_trans);
303 if (osd->od_quota_slave_md != NULL)
304 qsd_op_end(env, osd->od_quota_slave_md,
305 &oh->ot_quota_trans);
310 rc = dt_txn_hook_stop(env, th);
312 CDEBUG(D_OTHER, "%s: transaction hook failed: rc = %d\n",
315 osd_trans_stop_cb(oh, rc);
318 txg = oh->ot_tx->tx_txg;
320 osd_object_sa_dirty_rele(env, oh);
321 /* XXX: Once dmu_tx_commit() called, oh/th could have been freed
322 * by osd_trans_commit_cb already. */
323 dmu_tx_commit(oh->ot_tx);
324 osd_oti_get(env)->oti_in_trans = 0;
326 osd_unlinked_list_emptify(env, osd, &unlinked, true);
329 if (osd_txg_sync_delay_us < 0)
330 txg_wait_synced(dmu_objset_pool(osd->od_os), txg);
332 udelay(osd_txg_sync_delay_us);
338 static struct thandle *osd_trans_create(const struct lu_env *env,
339 struct dt_device *dt)
341 struct osd_device *osd = osd_dt_dev(dt);
342 struct osd_thandle *oh;
348 CERROR("%s: someone try to start transaction under "
349 "readonly mode, should be disabled.\n",
350 osd_name(osd_dt_dev(dt)));
352 RETURN(ERR_PTR(-EROFS));
355 tx = dmu_tx_create(osd->od_os);
357 RETURN(ERR_PTR(-ENOMEM));
359 /* alloc callback data */
363 RETURN(ERR_PTR(-ENOMEM));
367 INIT_LIST_HEAD(&oh->ot_dcb_list);
368 INIT_LIST_HEAD(&oh->ot_stop_dcb_list);
369 INIT_LIST_HEAD(&oh->ot_unlinked_list);
370 INIT_LIST_HEAD(&oh->ot_sa_list);
371 memset(&oh->ot_quota_trans, 0, sizeof(oh->ot_quota_trans));
376 osd_oti_get(env)->oti_ins_cache_depth++;
381 /* Estimate the total number of objects from a number of blocks */
382 uint64_t osd_objs_count_estimate(uint64_t usedbytes, uint64_t usedobjs,
383 uint64_t nrblocks, uint64_t est_maxblockshift)
385 uint64_t est_totobjs, est_usedblocks, est_usedobjs;
388 * If blocksize is below 64KB (e.g. MDT with recordsize=4096) then
389 * bump the free dnode estimate to assume blocks at least 64KB in
390 * case of a directory-heavy MDT (at 32KB/directory).
392 if (est_maxblockshift < 16) {
393 nrblocks >>= (16 - est_maxblockshift);
394 est_maxblockshift = 16;
398 * Estimate the total number of dnodes from the total blocks count
399 * and the space used per dnode. Since we don't know the overhead
400 * associated with each dnode (xattrs, SAs, VDEV overhead, etc.)
401 * just using DNODE_SHIFT isn't going to give a good estimate.
402 * Instead, compute the current average space usage per dnode, with
403 * an upper and lower cap to avoid unrealistic estimates..
405 * In case there aren't many dnodes or blocks used yet, add a small
406 * correction factor (OSD_DNODE_EST_{COUNT,BLKSHIFT}). This factor
407 * gradually disappears as the number of real dnodes grows. It also
408 * avoids the need to check for divide-by-zero computing dn_per_block.
410 BUILD_BUG_ON(OSD_DNODE_MIN_BLKSHIFT <= 0);
411 BUILD_BUG_ON(OSD_DNODE_EST_BLKSHIFT <= 0);
413 est_usedblocks = ((OSD_DNODE_EST_COUNT << OSD_DNODE_EST_BLKSHIFT) +
414 usedbytes) >> est_maxblockshift;
415 est_usedobjs = OSD_DNODE_EST_COUNT + usedobjs;
417 if (est_usedobjs <= est_usedblocks) {
419 * Average space/dnode more than maximum block size, use max
420 * block size to estimate free dnodes from adjusted free blocks
421 * count. OSTs typically use multiple blocks per dnode so this
424 est_totobjs = nrblocks;
426 } else if (est_usedobjs >= (est_usedblocks << OSD_DNODE_MIN_BLKSHIFT)) {
428 * Average space/dnode smaller than min dnode size (probably
429 * due to metadnode compression), use min dnode size to
430 * estimate object count. MDTs may use only one block per node
431 * so this case applies.
433 est_totobjs = nrblocks << OSD_DNODE_MIN_BLKSHIFT;
437 * Between the extremes, use average space per existing dnode
438 * to compute the number of dnodes that will fit into nrblocks:
440 * est_totobjs = nrblocks * (est_usedobjs / est_usedblocks)
442 * this may overflow 64 bits or become 0 if not handled well.
444 * We know nrblocks is below 2^(64 - blkbits) bits, and
445 * est_usedobjs is under 48 bits due to DN_MAX_OBJECT_SHIFT,
446 * which means that multiplying them may get as large as
447 * 2 ^ 96 for the minimum blocksize of 64KB allowed above.
449 * The ratio of dnodes per block (est_usedobjs / est_usedblocks)
450 * is under 2^(blkbits - DNODE_SHIFT) = blocksize / 512 due to
451 * the limit checks above, so we can safely compute this first.
452 * We care more about accuracy on the MDT (many dnodes/block)
453 * which is good because this is where truncation errors are
454 * smallest. Since both nrblocks and dn_per_block are a
455 * function of blkbits, their product is at most:
457 * 2^(64 - blkbits) * 2^(blkbits - DNODE_SHIFT) = 2^(64 - 9)
459 * so we can safely use 7 bits to compute a fixed-point
460 * fraction and est_totobjs can still fit in 64 bits.
462 unsigned dn_per_block = (est_usedobjs << 7) / est_usedblocks;
464 est_totobjs = (nrblocks * dn_per_block) >> 7;
469 static int osd_objset_statfs(struct osd_device *osd, struct obd_statfs *osfs)
471 struct objset *os = osd->od_os;
472 uint64_t usedbytes, availbytes, usedobjs, availobjs;
473 uint64_t est_availobjs;
477 dmu_objset_space(os, &usedbytes, &availbytes, &usedobjs, &availobjs);
479 memset(osfs, 0, sizeof(*osfs));
481 /* We're a zfs filesystem. */
482 osfs->os_type = UBERBLOCK_MAGIC;
485 * ZFS allows multiple block sizes. For statfs, Linux makes no
486 * proper distinction between bsize and frsize. For calculations
487 * of free and used blocks incorrectly uses bsize instead of frsize,
488 * but bsize is also used as the optimal blocksize. We return the
489 * largest possible block size as IO size for the optimum performance
490 * and scale the free and used blocks count appropriately.
492 osfs->os_bsize = osd->od_max_blksz;
493 bshift = fls64(osfs->os_bsize) - 1;
495 osfs->os_blocks = (usedbytes + availbytes) >> bshift;
496 osfs->os_bfree = availbytes >> bshift;
497 osfs->os_bavail = osfs->os_bfree; /* no extra root reservation */
499 /* Take replication (i.e. number of copies) into account */
500 if (os->os_copies != 0)
501 osfs->os_bavail /= os->os_copies;
504 * Reserve some space so we don't run into ENOSPC due to grants not
505 * accounting for metadata overhead in ZFS, and to avoid fragmentation.
506 * Rather than report this via os_bavail (which makes users unhappy if
507 * they can't fill the filesystem 100%), reduce os_blocks as well.
509 * Reserve 0.78% of total space, at least 16MB for small filesystems,
510 * for internal files to be created/unlinked when space is tight.
512 BUILD_BUG_ON(OSD_STATFS_RESERVED_SIZE <= 0);
513 reserved = OSD_STATFS_RESERVED_SIZE >> bshift;
514 if (likely(osfs->os_blocks >= reserved << OSD_STATFS_RESERVED_SHIFT))
515 reserved = osfs->os_blocks >> OSD_STATFS_RESERVED_SHIFT;
517 osfs->os_blocks -= reserved;
518 osfs->os_bfree -= min(reserved, osfs->os_bfree);
519 osfs->os_bavail -= min(reserved, osfs->os_bavail);
522 * The availobjs value returned from dmu_objset_space() is largely
523 * useless, since it reports the number of objects that might
524 * theoretically still fit into the dataset, independent of minor
525 * issues like how much space is actually available in the pool.
526 * Compute a better estimate in udmu_objs_count_estimate().
528 est_availobjs = osd_objs_count_estimate(usedbytes, usedobjs,
529 osfs->os_bfree, bshift);
531 osfs->os_ffree = min(availobjs, est_availobjs);
532 osfs->os_files = osfs->os_ffree + usedobjs;
534 /* ZFS XXX: fill in backing dataset FSID/UUID
535 memcpy(osfs->os_fsid, .... );*/
537 osfs->os_namelen = MAXNAMELEN;
538 osfs->os_maxbytes = OBD_OBJECT_EOF;
540 if (!spa_writeable(dmu_objset_spa(os)) ||
541 osd->od_dev_set_rdonly || osd->od_prop_rdonly)
542 osfs->os_state |= OS_STATFS_READONLY;
548 * Concurrency: shouldn't matter.
550 int osd_statfs(const struct lu_env *env, struct dt_device *d,
551 struct obd_statfs *osfs, struct obd_statfs_info *info)
553 struct osd_device *osd = osd_dt_dev(d);
557 rc = osd_objset_statfs(osd, osfs);
558 if (unlikely(rc != 0))
561 osfs->os_bavail -= min_t(u64,
562 OSD_GRANT_FOR_LOCAL_OIDS / osfs->os_bsize,
565 /* ZFS does not support reporting nonrotional status yet, so return
566 * flag only if user has set nonrotational.
568 osfs->os_state |= osd->od_nonrotational ? OS_STATFS_NONROT : 0;
573 static int osd_blk_insert_cost(struct osd_device *osd)
575 int max_blockshift, nr_blkptrshift, bshift;
577 /* max_blockshift is the log2 of the number of blocks needed to reach
578 * the maximum filesize (that's to say 2^64) */
579 bshift = osd_spa_maxblockshift(dmu_objset_spa(osd->od_os));
580 max_blockshift = DN_MAX_OFFSET_SHIFT - bshift;
582 /* nr_blkptrshift is the log2 of the number of block pointers that can
583 * be stored in an indirect block */
584 BUILD_BUG_ON(DN_MAX_INDBLKSHIFT <= SPA_BLKPTRSHIFT);
585 nr_blkptrshift = DN_MAX_INDBLKSHIFT - SPA_BLKPTRSHIFT;
587 /* max_blockshift / nr_blkptrshift is thus the maximum depth of the
588 * tree. We add +1 for rounding purpose.
589 * The tree depth times the indirect block size gives us the maximum
590 * cost of inserting a block in the tree */
591 return (max_blockshift / nr_blkptrshift + 1) * (1<<DN_MAX_INDBLKSHIFT);
595 * Concurrency: doesn't access mutable data.
597 static void osd_conf_get(const struct lu_env *env,
598 const struct dt_device *dev,
599 struct dt_device_param *param)
601 struct osd_device *osd = osd_dt_dev(dev);
604 * XXX should be taken from not-yet-existing fs abstraction layer.
606 param->ddp_max_name_len = MAXNAMELEN;
607 param->ddp_max_nlink = 1 << 31; /* it's 8byte on a disk */
608 param->ddp_symlink_max = PATH_MAX;
609 param->ddp_mount_type = LDD_MT_ZFS;
611 param->ddp_mntopts = MNTOPT_USERXATTR;
612 if (osd->od_posix_acl)
613 param->ddp_mntopts |= MNTOPT_ACL;
614 /* Previously DXATTR_MAX_ENTRY_SIZE */
615 param->ddp_max_ea_size = OBD_MAX_EA_SIZE;
617 /* for maxbytes, report same value as ZPL */
618 param->ddp_maxbytes = MAX_LFS_FILESIZE;
620 /* inodes are dynamically allocated, so we report the per-inode space
621 * consumption to upper layers. This static value is not really accurate
622 * and we should use the same logic as in udmu_objset_statfs() to
623 * estimate the real size consumed by an object */
624 param->ddp_inodespace = OSD_DNODE_EST_COUNT;
625 /* Although ZFS isn't an extent-based filesystem, the metadata overhead
626 * (i.e. 7 levels of indirect blocks, see osd_blk_insert_cost()) should
627 * not be accounted for every single new block insertion.
628 * Instead, the maximum extent size is set to the number of blocks that
629 * can fit into a single contiguous indirect block. There would be some
630 * cases where this crosses indirect blocks, but it also won't have 7
631 * new levels of indirect blocks in that case either, so it will still
632 * have enough reserved space for the extra indirect block */
633 param->ddp_max_extent_blks =
634 (1 << (DN_MAX_INDBLKSHIFT - SPA_BLKPTRSHIFT));
635 param->ddp_extent_tax = osd_blk_insert_cost(osd);
637 /* Preferred RPC size for efficient disk IO. 1MB shows good
638 * all-around performance for ZFS, but use blocksize (recordsize)
639 * by default if larger to avoid read-modify-write. */
640 if (osd->od_max_blksz > ONE_MB_BRW_SIZE)
641 param->ddp_brw_size = osd->od_max_blksz;
643 param->ddp_brw_size = ONE_MB_BRW_SIZE;
645 #ifdef HAVE_DMU_OFFSET_NEXT
646 param->ddp_has_lseek_data_hole = osd->od_sync_on_lseek;
648 param->ddp_has_lseek_data_hole = false;
653 * Concurrency: shouldn't matter.
655 static int osd_sync(const struct lu_env *env, struct dt_device *d)
658 struct osd_device *osd = osd_dt_dev(d);
660 CDEBUG(D_CACHE, "syncing OSD %s\n", LUSTRE_OSD_ZFS_NAME);
661 txg_wait_synced(dmu_objset_pool(osd->od_os), 0ULL);
662 CDEBUG(D_CACHE, "synced OSD %s\n", LUSTRE_OSD_ZFS_NAME);
668 static int osd_commit_async(const struct lu_env *env, struct dt_device *dev)
670 struct osd_device *osd = osd_dt_dev(dev);
671 tx_state_t *tx = &dmu_objset_pool(osd->od_os)->dp_tx;
674 mutex_enter(&tx->tx_sync_lock);
675 txg = tx->tx_open_txg + 1;
676 if (tx->tx_quiesce_txg_waiting < txg) {
677 tx->tx_quiesce_txg_waiting = txg;
678 cv_broadcast(&tx->tx_quiesce_more_cv);
680 mutex_exit(&tx->tx_sync_lock);
686 * Concurrency: shouldn't matter.
688 static int osd_ro(const struct lu_env *env, struct dt_device *d)
690 struct osd_device *osd = osd_dt_dev(d);
693 CERROR("%s: *** setting device %s read-only ***\n",
694 osd->od_svname, LUSTRE_OSD_ZFS_NAME);
695 osd->od_dev_set_rdonly = 1;
696 spa_freeze(dmu_objset_spa(osd->od_os));
701 static const struct dt_device_operations osd_dt_ops = {
702 .dt_root_get = osd_root_get,
703 .dt_statfs = osd_statfs,
704 .dt_trans_create = osd_trans_create,
705 .dt_trans_start = osd_trans_start,
706 .dt_trans_stop = osd_trans_stop,
707 .dt_trans_cb_add = osd_trans_cb_add,
708 .dt_conf_get = osd_conf_get,
710 .dt_commit_async = osd_commit_async,
715 * DMU OSD device type methods
717 static int osd_type_init(struct lu_device_type *t)
719 LU_CONTEXT_KEY_INIT(&osd_key);
720 return lu_context_key_register(&osd_key);
723 static void osd_type_fini(struct lu_device_type *t)
725 lu_context_key_degister(&osd_key);
728 static void *osd_key_init(const struct lu_context *ctx,
729 struct lu_context_key *key)
731 struct osd_thread_info *info;
735 info->oti_env = container_of(ctx, struct lu_env, le_ctx);
737 info = ERR_PTR(-ENOMEM);
741 static void osd_key_fini(const struct lu_context *ctx,
742 struct lu_context_key *key, void *data)
744 struct osd_thread_info *info = data;
745 struct osd_idmap_cache *idc = info->oti_ins_cache;
748 LASSERT(info->oti_ins_cache_size > 0);
749 OBD_FREE_PTR_ARRAY_LARGE(idc, info->oti_ins_cache_size);
750 info->oti_ins_cache = NULL;
751 info->oti_ins_cache_size = 0;
753 lu_buf_free(&info->oti_xattr_lbuf);
757 static void osd_key_exit(const struct lu_context *ctx,
758 struct lu_context_key *key, void *data)
762 struct lu_context_key osd_key = {
763 .lct_tags = LCT_DT_THREAD | LCT_MD_THREAD | LCT_MG_THREAD | LCT_LOCAL,
764 .lct_init = osd_key_init,
765 .lct_fini = osd_key_fini,
766 .lct_exit = osd_key_exit
769 static void osd_fid_fini(const struct lu_env *env, struct osd_device *osd)
771 if (osd->od_cl_seq == NULL)
774 seq_client_fini(osd->od_cl_seq);
775 OBD_FREE_PTR(osd->od_cl_seq);
776 osd->od_cl_seq = NULL;
779 static int osd_shutdown(const struct lu_env *env, struct osd_device *o)
783 /* shutdown quota slave instance associated with the device */
784 if (o->od_quota_slave_md != NULL) {
785 /* complete all in-flight callbacks */
786 osd_sync(env, &o->od_dt_dev);
787 txg_wait_callbacks(spa_get_dsl(dmu_objset_spa(o->od_os)));
788 qsd_fini(env, o->od_quota_slave_md);
789 o->od_quota_slave_md = NULL;
792 if (o->od_quota_slave_dt != NULL) {
793 /* complete all in-flight callbacks */
794 osd_sync(env, &o->od_dt_dev);
795 txg_wait_callbacks(spa_get_dsl(dmu_objset_spa(o->od_os)));
796 qsd_fini(env, o->od_quota_slave_dt);
797 o->od_quota_slave_dt = NULL;
799 osd_fid_fini(env, o);
804 static void osd_xattr_changed_cb(void *arg, uint64_t newval)
806 struct osd_device *osd = arg;
808 osd->od_xattr_in_sa = (newval == ZFS_XATTR_SA);
811 static void osd_recordsize_changed_cb(void *arg, uint64_t newval)
813 struct osd_device *osd = arg;
815 LASSERT(newval <= osd_spa_maxblocksize(dmu_objset_spa(osd->od_os)));
816 LASSERT(newval >= SPA_MINBLOCKSIZE);
817 LASSERT(ISP2(newval));
819 osd->od_max_blksz = newval;
822 static void osd_readonly_changed_cb(void *arg, uint64_t newval)
824 struct osd_device *osd = arg;
826 osd->od_prop_rdonly = !!newval;
829 #ifdef HAVE_DMU_OBJECT_ALLOC_DNSIZE
830 static void osd_dnodesize_changed_cb(void *arg, uint64_t newval)
832 struct osd_device *osd = arg;
834 osd->od_dnsize = newval;
838 * This function unregisters all registered callbacks. It's harmless to
839 * unregister callbacks that were never registered so it is used to safely
840 * unwind a partially completed call to osd_objset_register_callbacks().
842 static void osd_objset_unregister_callbacks(struct osd_device *o)
844 struct dsl_dataset *ds = dmu_objset_ds(o->od_os);
846 (void) dsl_prop_unregister(ds, zfs_prop_to_name(ZFS_PROP_XATTR),
847 osd_xattr_changed_cb, o);
848 (void) dsl_prop_unregister(ds, zfs_prop_to_name(ZFS_PROP_RECORDSIZE),
849 osd_recordsize_changed_cb, o);
850 (void) dsl_prop_unregister(ds, zfs_prop_to_name(ZFS_PROP_READONLY),
851 osd_readonly_changed_cb, o);
852 #ifdef HAVE_DMU_OBJECT_ALLOC_DNSIZE
853 (void) dsl_prop_unregister(ds, zfs_prop_to_name(ZFS_PROP_DNODESIZE),
854 osd_dnodesize_changed_cb, o);
857 if (o->arc_prune_cb != NULL) {
858 arc_remove_prune_callback(o->arc_prune_cb);
859 o->arc_prune_cb = NULL;
864 * Register the required callbacks to be notified when zfs properties
865 * are modified using the 'zfs(8)' command line utility.
867 static int osd_objset_register_callbacks(struct osd_device *o)
869 struct dsl_dataset *ds = dmu_objset_ds(o->od_os);
870 dsl_pool_t *dp = dmu_objset_pool(o->od_os);
876 dsl_pool_config_enter(dp, FTAG);
877 rc = -dsl_prop_register(ds, zfs_prop_to_name(ZFS_PROP_XATTR),
878 osd_xattr_changed_cb, o);
882 rc = -dsl_prop_register(ds, zfs_prop_to_name(ZFS_PROP_RECORDSIZE),
883 osd_recordsize_changed_cb, o);
887 rc = -dsl_prop_register(ds, zfs_prop_to_name(ZFS_PROP_READONLY),
888 osd_readonly_changed_cb, o);
892 #ifdef HAVE_DMU_OBJECT_ALLOC_DNSIZE
893 rc = -dsl_prop_register(ds, zfs_prop_to_name(ZFS_PROP_DNODESIZE),
894 osd_dnodesize_changed_cb, o);
899 o->arc_prune_cb = arc_add_prune_callback(arc_prune_func, o);
901 dsl_pool_config_exit(dp, FTAG);
903 osd_objset_unregister_callbacks(o);
908 static int osd_objset_open(struct osd_device *o)
910 uint64_t version = ZPL_VERSION;
911 uint64_t sa_obj, unlink_obj;
915 rc = -osd_dmu_objset_own(o->od_mntdev, DMU_OST_ZFS,
916 o->od_dt_dev.dd_rdonly ? B_TRUE : B_FALSE,
917 B_TRUE, o, &o->od_os);
920 CERROR("%s: can't open %s\n", o->od_svname, o->od_mntdev);
926 /* Check ZFS version */
927 rc = -zap_lookup(o->od_os, MASTER_NODE_OBJ,
928 ZPL_VERSION_STR, 8, 1, &version);
930 CERROR("%s: Error looking up ZPL VERSION\n", o->od_mntdev);
932 * We can't return ENOENT because that would mean the objset
935 GOTO(out, rc = -EIO);
938 rc = -zap_lookup(o->od_os, MASTER_NODE_OBJ,
939 ZFS_SA_ATTRS, 8, 1, &sa_obj);
943 rc = -sa_setup(o->od_os, sa_obj, zfs_attr_table,
944 ZPL_END, &o->z_attr_table);
948 rc = -zap_lookup(o->od_os, MASTER_NODE_OBJ, ZFS_ROOT_OBJ,
949 8, 1, &o->od_rootid);
951 CERROR("%s: lookup for root failed: rc = %d\n",
956 rc = -zap_lookup(o->od_os, MASTER_NODE_OBJ, ZFS_UNLINKED_SET,
959 CERROR("%s: lookup for %s failed: rc = %d\n",
960 o->od_svname, ZFS_UNLINKED_SET, rc);
964 /* Check that user/group usage tracking is supported */
965 if (!dmu_objset_userused_enabled(o->od_os) ||
966 DMU_USERUSED_DNODE(o->od_os)->dn_type != DMU_OT_USERGROUP_USED ||
967 DMU_GROUPUSED_DNODE(o->od_os)->dn_type != DMU_OT_USERGROUP_USED) {
968 CERROR("%s: Space accounting not supported by this target, "
969 "aborting\n", o->od_svname);
970 GOTO(out, rc = -ENOTSUPP);
973 rc = __osd_obj2dnode(o->od_os, unlink_obj, &o->od_unlinked);
975 CERROR("%s: can't get dnode for unlinked: rc = %d\n",
981 if (rc != 0 && o->od_os != NULL) {
982 osd_dmu_objset_disown(o->od_os, B_TRUE, o);
989 int osd_unlinked_object_free(const struct lu_env *env, struct osd_device *osd,
992 char *key = osd_oti_get(env)->oti_str;
996 if (osd->od_dt_dev.dd_rdonly) {
997 CERROR("%s: someone try to free objects under "
998 "readonly mode, should be disabled.\n", osd_name(osd));
1004 rc = -dmu_free_long_range(osd->od_os, oid, 0, DMU_OBJECT_END);
1006 CWARN("%s: Cannot truncate %llu: rc = %d\n",
1007 osd->od_svname, oid, rc);
1011 tx = dmu_tx_create(osd->od_os);
1012 dmu_tx_mark_netfree(tx);
1013 dmu_tx_hold_free(tx, oid, 0, DMU_OBJECT_END);
1014 osd_tx_hold_zap(tx, osd->od_unlinked->dn_object, osd->od_unlinked,
1016 rc = -dmu_tx_assign(tx, TXG_WAIT);
1018 CWARN("%s: Cannot assign tx for %llu: rc = %d\n",
1019 osd->od_svname, oid, rc);
1023 snprintf(key, sizeof(osd_oti_get(env)->oti_str), "%llx", oid);
1024 rc = osd_zap_remove(osd, osd->od_unlinked->dn_object,
1025 osd->od_unlinked, key, tx);
1027 CWARN("%s: Cannot remove %llu from unlinked set: rc = %d\n",
1028 osd->od_svname, oid, rc);
1032 rc = -dmu_object_free(osd->od_os, oid, tx);
1034 CWARN("%s: Cannot free %llu: rc = %d\n",
1035 osd->od_svname, oid, rc);
1050 osd_unlinked_drain(const struct lu_env *env, struct osd_device *osd)
1053 zap_attribute_t *za = &osd_oti_get(env)->oti_za;
1055 zap_cursor_init(&zc, osd->od_os, osd->od_unlinked->dn_object);
1057 while (zap_cursor_retrieve(&zc, za) == 0) {
1058 /* If cannot free the object, leave it in the unlinked set,
1059 * until the OSD is mounted again when obd_unlinked_drain()
1060 * will be called. */
1061 if (osd_unlinked_object_free(env, osd, za->za_first_integer))
1063 zap_cursor_advance(&zc);
1066 zap_cursor_fini(&zc);
1069 static int osd_mount(const struct lu_env *env,
1070 struct osd_device *o, struct lustre_cfg *cfg)
1072 char *mntdev = lustre_cfg_string(cfg, 1);
1073 char *str = lustre_cfg_string(cfg, 2);
1074 char *svname = lustre_cfg_string(cfg, 4);
1077 bool resetoi = false;
1082 if (o->od_os != NULL)
1085 if (mntdev == NULL || svname == NULL)
1088 rc = strlcpy(o->od_mntdev, mntdev, sizeof(o->od_mntdev));
1089 if (rc >= sizeof(o->od_mntdev))
1092 rc = strlcpy(o->od_svname, svname, sizeof(o->od_svname));
1093 if (rc >= sizeof(o->od_svname))
1096 opts = lustre_cfg_string(cfg, 3);
1098 o->od_index_backup_stop = 0;
1099 o->od_index = -1; /* -1 means index is invalid */
1100 rc = server_name2index(o->od_svname, &o->od_index, NULL);
1101 str = strstr(str, ":");
1103 unsigned long flags;
1105 rc = kstrtoul(str + 1, 10, &flags);
1109 if (flags & LMD_FLG_DEV_RDONLY) {
1110 o->od_dt_dev.dd_rdonly = 1;
1111 LCONSOLE_WARN("%s: set dev_rdonly on this device\n",
1115 if (flags & LMD_FLG_NOSCRUB)
1116 o->od_auto_scrub_interval = AS_NEVER;
1119 if (server_name_is_ost(o->od_svname))
1122 rc = osd_objset_open(o);
1126 o->od_xattr_in_sa = B_TRUE;
1127 o->od_max_blksz = osd_spa_maxblocksize(o->od_os->os_spa);
1128 o->od_readcache_max_filesize = OSD_MAX_CACHE_SIZE;
1130 rc = __osd_obj2dnode(o->od_os, o->od_rootid, &rootdn);
1133 o->od_root = rootdn->dn_object;
1134 osd_dnode_rele(rootdn);
1136 rc = __osd_obj2dnode(o->od_os, DMU_USERUSED_OBJECT,
1137 &o->od_userused_dn);
1141 rc = __osd_obj2dnode(o->od_os, DMU_GROUPUSED_OBJECT,
1142 &o->od_groupused_dn);
1146 #ifdef ZFS_PROJINHERIT
1147 if (dmu_objset_projectquota_enabled(o->od_os)) {
1148 rc = __osd_obj2dnode(o->od_os, DMU_PROJECTUSED_OBJECT,
1149 &o->od_projectused_dn);
1150 if (rc && rc != -ENOENT)
1155 rc = lu_site_init(&o->od_site, osd2lu_dev(o));
1158 o->od_site.ls_bottom_dev = osd2lu_dev(o);
1160 rc = lu_site_init_finish(&o->od_site);
1164 rc = osd_objset_register_callbacks(o);
1168 if (opts && strstr(opts, "resetoi"))
1172 rc = osd_scrub_setup(env, o, resetoi);
1177 rc = osd_procfs_init(o, o->od_svname);
1181 /* currently it's no need to prepare qsd_instance_md for OST */
1182 if (!o->od_is_ost) {
1183 o->od_quota_slave_md = qsd_init(env, o->od_svname,
1185 o->od_proc_entry, true);
1186 if (IS_ERR(o->od_quota_slave_md)) {
1187 rc = PTR_ERR(o->od_quota_slave_md);
1188 o->od_quota_slave_md = NULL;
1193 o->od_quota_slave_dt = qsd_init(env, o->od_svname, &o->od_dt_dev,
1194 o->od_proc_entry, false);
1196 if (IS_ERR(o->od_quota_slave_dt)) {
1197 if (o->od_quota_slave_md != NULL) {
1198 qsd_fini(env, o->od_quota_slave_md);
1199 o->od_quota_slave_md = NULL;
1202 rc = PTR_ERR(o->od_quota_slave_dt);
1203 o->od_quota_slave_dt = NULL;
1207 #ifdef HAVE_DMU_USEROBJ_ACCOUNTING
1208 if (!osd_dmu_userobj_accounting_available(o))
1209 CWARN("%s: dnode accounting not enabled: "
1210 "enable feature@userobj_accounting in pool\n",
1214 /* parse mount option "noacl", and enable ACL by default */
1215 if (opts == NULL || strstr(opts, "noacl") == NULL)
1216 o->od_posix_acl = 1;
1218 osd_unlinked_drain(env, o);
1220 if (rc && o->od_os) {
1221 osd_dmu_objset_disown(o->od_os, B_TRUE, o);
1228 static void osd_umount(const struct lu_env *env, struct osd_device *o)
1232 if (atomic_read(&o->od_zerocopy_alloc))
1233 CERROR("%s: lost %d allocated page(s)\n", o->od_svname,
1234 atomic_read(&o->od_zerocopy_alloc));
1235 if (atomic_read(&o->od_zerocopy_loan))
1236 CERROR("%s: lost %d loaned abuf(s)\n", o->od_svname,
1237 atomic_read(&o->od_zerocopy_loan));
1238 if (atomic_read(&o->od_zerocopy_pin))
1239 CERROR("%s: lost %d pinned dbuf(s)\n", o->od_svname,
1240 atomic_read(&o->od_zerocopy_pin));
1242 if (o->od_unlinked) {
1243 osd_dnode_rele(o->od_unlinked);
1244 o->od_unlinked = NULL;
1246 if (o->od_userused_dn) {
1247 osd_dnode_rele(o->od_userused_dn);
1248 o->od_userused_dn = NULL;
1250 if (o->od_groupused_dn) {
1251 osd_dnode_rele(o->od_groupused_dn);
1252 o->od_groupused_dn = NULL;
1255 #ifdef ZFS_PROJINHERIT
1256 if (o->od_projectused_dn) {
1257 osd_dnode_rele(o->od_projectused_dn);
1258 o->od_projectused_dn = NULL;
1262 if (o->od_os != NULL) {
1263 if (!o->od_dt_dev.dd_rdonly)
1264 /* force a txg sync to get all commit callbacks */
1265 txg_wait_synced(dmu_objset_pool(o->od_os), 0ULL);
1267 /* close the object set */
1268 osd_dmu_objset_disown(o->od_os, B_TRUE, o);
1275 static int osd_device_init0(const struct lu_env *env,
1276 struct osd_device *o,
1277 struct lustre_cfg *cfg)
1279 struct lu_device *l = osd2lu_dev(o);
1282 /* if the module was re-loaded, env can loose its keys */
1283 rc = lu_env_refill((struct lu_env *) env);
1287 l->ld_ops = &osd_lu_ops;
1288 o->od_dt_dev.dd_ops = &osd_dt_ops;
1289 sema_init(&o->od_otable_sem, 1);
1290 INIT_LIST_HEAD(&o->od_ios_list);
1291 o->od_auto_scrub_interval = AS_DEFAULT;
1292 o->od_sync_on_lseek = B_TRUE;
1294 /* ZFS does not support reporting nonrotional status yet, so this flag
1295 * is only set if explicitly set by the user.
1297 o->od_nonrotational = 0;
1303 static struct lu_device *osd_device_fini(const struct lu_env *env,
1304 struct lu_device *dev);
1306 static struct lu_device *osd_device_alloc(const struct lu_env *env,
1307 struct lu_device_type *type,
1308 struct lustre_cfg *cfg)
1310 struct osd_device *dev;
1311 struct osd_seq_list *osl;
1316 return ERR_PTR(-ENOMEM);
1318 osl = &dev->od_seq_list;
1319 INIT_LIST_HEAD(&osl->osl_seq_list);
1320 rwlock_init(&osl->osl_seq_list_lock);
1321 sema_init(&osl->osl_seq_init_sem, 1);
1322 INIT_LIST_HEAD(&dev->od_index_backup_list);
1323 INIT_LIST_HEAD(&dev->od_index_restore_list);
1324 spin_lock_init(&dev->od_lock);
1325 dev->od_index_backup_policy = LIBP_NONE;
1327 rc = dt_device_init(&dev->od_dt_dev, type);
1329 rc = osd_device_init0(env, dev, cfg);
1331 rc = osd_mount(env, dev, cfg);
1333 osd_device_fini(env, osd2lu_dev(dev));
1336 dt_device_fini(&dev->od_dt_dev);
1339 if (unlikely(rc != 0))
1342 return rc == 0 ? osd2lu_dev(dev) : ERR_PTR(rc);
1345 static struct lu_device *osd_device_free(const struct lu_env *env,
1346 struct lu_device *d)
1348 struct osd_device *o = osd_dev(d);
1351 /* XXX: make osd top device in order to release reference */
1352 d->ld_site->ls_top_dev = d;
1353 lu_site_purge(env, d->ld_site, -1);
1354 lu_site_print(env, d->ld_site, &d->ld_site->ls_obj_hash.nelems,
1355 D_ERROR, lu_cdebug_printer);
1356 lu_site_fini(&o->od_site);
1357 dt_device_fini(&o->od_dt_dev);
1363 static struct lu_device *osd_device_fini(const struct lu_env *env,
1364 struct lu_device *d)
1366 struct osd_device *o = osd_dev(d);
1370 osd_index_backup(env, o, false);
1372 osd_objset_unregister_callbacks(o);
1373 if (!o->od_dt_dev.dd_rdonly) {
1374 osd_sync(env, lu2dt_dev(d));
1376 spa_get_dsl(dmu_objset_spa(o->od_os)));
1380 /* now with all the callbacks completed we can cleanup the remainings */
1381 osd_shutdown(env, o);
1382 osd_scrub_cleanup(env, o);
1384 rc = osd_procfs_fini(o);
1386 CERROR("proc fini error %d\n", rc);
1387 RETURN(ERR_PTR(rc));
1396 static int osd_device_init(const struct lu_env *env, struct lu_device *d,
1397 const char *name, struct lu_device *next)
1403 * To be removed, setup is performed by osd_device_{init,alloc} and
1404 * cleanup is performed by osd_device_{fini,free).
1406 static int osd_process_config(const struct lu_env *env,
1407 struct lu_device *d, struct lustre_cfg *cfg)
1409 struct osd_device *o = osd_dev(d);
1414 switch(cfg->lcfg_command) {
1416 rc = osd_mount(env, o, cfg);
1419 /* For the case LCFG_PRE_CLEANUP is not called in advance,
1420 * that may happend if hit failure during mount process. */
1421 osd_index_backup(env, o, false);
1422 rc = osd_shutdown(env, o);
1425 LASSERT(&o->od_dt_dev);
1426 count = class_modify_config(cfg, PARAM_OSD,
1427 &o->od_dt_dev.dd_kobj);
1429 count = class_modify_config(cfg, PARAM_OST,
1430 &o->od_dt_dev.dd_kobj);
1431 rc = count > 0 ? 0 : count;
1434 case LCFG_PRE_CLEANUP:
1436 osd_index_backup(env, o,
1437 o->od_index_backup_policy != LIBP_NONE);
1447 static int osd_recovery_complete(const struct lu_env *env, struct lu_device *d)
1449 struct osd_device *osd = osd_dev(d);
1453 if (osd->od_quota_slave_md == NULL && osd->od_quota_slave_dt == NULL)
1456 /* start qsd instance on recovery completion, this notifies the quota
1457 * slave code that we are about to process new requests now */
1458 rc = qsd_start(env, osd->od_quota_slave_dt);
1459 if (rc == 0 && osd->od_quota_slave_md != NULL)
1460 rc = qsd_start(env, osd->od_quota_slave_md);
1465 * we use exports to track all osd users
1467 static int osd_obd_connect(const struct lu_env *env, struct obd_export **exp,
1468 struct obd_device *obd, struct obd_uuid *cluuid,
1469 struct obd_connect_data *data, void *localdata)
1471 struct osd_device *osd = osd_dev(obd->obd_lu_dev);
1472 struct lustre_handle conn;
1476 CDEBUG(D_CONFIG, "connect #%d\n", osd->od_connects);
1478 rc = class_connect(&conn, obd, cluuid);
1482 *exp = class_conn2export(&conn);
1484 spin_lock(&obd->obd_dev_lock);
1486 spin_unlock(&obd->obd_dev_lock);
1492 * once last export (we don't count self-export) disappeared
1493 * osd can be released
1495 static int osd_obd_disconnect(struct obd_export *exp)
1497 struct obd_device *obd = exp->exp_obd;
1498 struct osd_device *osd = osd_dev(obd->obd_lu_dev);
1499 int rc, release = 0;
1502 /* Only disconnect the underlying layers on the final disconnect. */
1503 spin_lock(&obd->obd_dev_lock);
1505 if (osd->od_connects == 0)
1507 spin_unlock(&obd->obd_dev_lock);
1509 rc = class_disconnect(exp); /* bz 9811 */
1511 if (rc == 0 && release)
1512 class_manual_cleanup(obd);
1516 static int osd_fid_init(const struct lu_env *env, struct osd_device *osd)
1518 struct seq_server_site *ss = osd_seq_site(osd);
1522 if (osd->od_is_ost || osd->od_cl_seq != NULL)
1525 if (unlikely(ss == NULL))
1528 OBD_ALLOC_PTR(osd->od_cl_seq);
1529 if (osd->od_cl_seq == NULL)
1532 seq_client_init(osd->od_cl_seq, NULL, LUSTRE_SEQ_METADATA,
1533 osd->od_svname, ss->ss_server_seq);
1535 if (ss->ss_node_id == 0) {
1537 * If the OSD on the sequence controller(MDT0), then allocate
1538 * sequence here, otherwise allocate sequence after connected
1539 * to MDT0 (see mdt_register_lwp_callback()).
1541 rc = seq_server_alloc_meta(osd->od_cl_seq->lcs_srv,
1542 &osd->od_cl_seq->lcs_space, env);
1548 static int osd_prepare(const struct lu_env *env, struct lu_device *pdev,
1549 struct lu_device *dev)
1551 struct osd_device *osd = osd_dev(dev);
1555 if (osd->od_quota_slave_md != NULL) {
1556 /* set up quota slave objects */
1557 rc = qsd_prepare(env, osd->od_quota_slave_md);
1562 if (osd->od_quota_slave_dt != NULL) {
1563 /* set up quota slave objects */
1564 rc = qsd_prepare(env, osd->od_quota_slave_dt);
1569 rc = osd_fid_init(env, osd);
1575 * Implementation of lu_device_operations::ldo_fid_alloc() for OSD
1579 * see include/lu_object.h for the details.
1581 static int osd_fid_alloc(const struct lu_env *env, struct lu_device *d,
1582 struct lu_fid *fid, struct lu_object *parent,
1583 const struct lu_name *name)
1585 struct osd_device *osd = osd_dev(d);
1587 return seq_client_alloc_fid(env, osd->od_cl_seq, fid);
1590 const struct lu_device_operations osd_lu_ops = {
1591 .ldo_object_alloc = osd_object_alloc,
1592 .ldo_process_config = osd_process_config,
1593 .ldo_recovery_complete = osd_recovery_complete,
1594 .ldo_prepare = osd_prepare,
1595 .ldo_fid_alloc = osd_fid_alloc,
1598 static void osd_type_start(struct lu_device_type *t)
1602 static void osd_type_stop(struct lu_device_type *t)
1606 static const struct lu_device_type_operations osd_device_type_ops = {
1607 .ldto_init = osd_type_init,
1608 .ldto_fini = osd_type_fini,
1610 .ldto_start = osd_type_start,
1611 .ldto_stop = osd_type_stop,
1613 .ldto_device_alloc = osd_device_alloc,
1614 .ldto_device_free = osd_device_free,
1616 .ldto_device_init = osd_device_init,
1617 .ldto_device_fini = osd_device_fini
1620 static struct lu_device_type osd_device_type = {
1621 .ldt_tags = LU_DEVICE_DT,
1622 .ldt_name = LUSTRE_OSD_ZFS_NAME,
1623 .ldt_ops = &osd_device_type_ops,
1624 .ldt_ctx_tags = LCT_LOCAL
1628 static const struct obd_ops osd_obd_device_ops = {
1629 .o_owner = THIS_MODULE,
1630 .o_connect = osd_obd_connect,
1631 .o_disconnect = osd_obd_disconnect,
1634 static int __init osd_init(void)
1638 rc = osd_options_init();
1642 rc = lu_kmem_init(osd_caches);
1646 rc = class_register_type(&osd_obd_device_ops, NULL, true,
1647 LUSTRE_OSD_ZFS_NAME, &osd_device_type);
1649 lu_kmem_fini(osd_caches);
1653 static void __exit osd_exit(void)
1655 class_unregister_type(LUSTRE_OSD_ZFS_NAME);
1656 lu_kmem_fini(osd_caches);
1659 module_param(osd_oi_count, int, 0444);
1660 MODULE_PARM_DESC(osd_oi_count, "Number of Object Index containers to be created, it's only valid for new filesystem.");
1662 module_param(osd_txg_sync_delay_us, int, 0644);
1663 MODULE_PARM_DESC(osd_txg_sync_delay_us,
1664 "When zero or larger delay N usec instead of doing TXG sync");
1666 MODULE_AUTHOR("OpenSFS, Inc. <http://www.lustre.org/>");
1667 MODULE_DESCRIPTION("Lustre Object Storage Device ("LUSTRE_OSD_ZFS_NAME")");
1668 MODULE_VERSION(LUSTRE_VERSION_STRING);
1669 MODULE_LICENSE("GPL");
1671 module_init(osd_init);
1672 module_exit(osd_exit);