*
* You should have received a copy of the GNU General Public License
* version 2 along with this program; If not, see
- * http://www.sun.com/software/products/lustre/docs/GPLv2.pdf
- *
- * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara,
- * CA 95054 USA or visit www.sun.com if you need additional information or
- * have any questions.
+ * http://www.gnu.org/licenses/gpl-2.0.html
*
* GPL HEADER END
*/
* Copyright (c) 2009, 2010, Oracle and/or its affiliates. All rights reserved.
* Use is subject to license terms.
*
- * Copyright (c) 2012, 2013, Intel Corporation.
+ * Copyright (c) 2012, 2014 Intel Corporation.
*/
/*
* This file is part of Lustre, http://www.lustre.org/
* Lustre is a trademark of Sun Microsystems, Inc.
*
- * lustre/ofd/ofd.c
+ * lustre/ofd/ofd_dev.c
+ *
+ * This file contains OSD API methods for OBD Filter Device (OFD),
+ * request handlers and supplemental functions to set OFD up and clean it up.
*
- * Author: Alex Zhuravlev <bzzz@whamcloud.com>
- * Author: Mike Pershin <tappro@whamcloud.com>
- * Author: Johann Lombardi <johann@whamcloud.com>
+ * Author: Alex Zhuravlev <alexey.zhuravlev@intel.com>
+ * Author: Mike Pershin <mike.pershin@intel.com>
+ * Author: Johann Lombardi <johann.lombardi@intel.com>
+ */
+/*
+ * The OBD Filter Device (OFD) module belongs to the Object Storage
+ * Server stack and connects the RPC oriented Unified Target (TGT)
+ * layer (see lustre/include/lu_target.h) to the storage oriented OSD
+ * layer (see lustre/doc/osd-api.txt).
+ *
+ * TGT
+ * | DT and OBD APIs
+ * OFD
+ * | DT API
+ * OSD
+ *
+ * OFD implements the LU and OBD device APIs and is responsible for:
+ *
+ * - Handling client requests (create, destroy, bulk IO, setattr,
+ * get_info, set_info, statfs) for the objects belonging to the OST
+ * (together with TGT).
+ *
+ * - Providing grant space management which allows clients to reserve
+ * disk space for data writeback. OFD tracks grants on global and
+ * per client levels.
+ *
+ * - Handling object precreation requests from MDTs.
+ *
+ * - Operating the LDLM service that allows clients to maintain object
+ * data cache coherence.
*/
#define DEBUG_SUBSYSTEM S_FILTER
#include <lustre/lustre_idl.h>
#include <lustre_dlm.h>
#include <lustre_quota.h>
+#include <lustre_nodemap.h>
#include "ofd_internal.h"
}
};
+/**
+ * Connect OFD to the next device in the stack.
+ *
+ * This function is used for device stack configuration and links OFD
+ * device with bottom OSD device.
+ *
+ * \param[in] env execution environment
+ * \param[in] m OFD device
+ * \param[in] next name of next device in the stack
+ * \param[out] exp export to return
+ *
+ * \retval 0 and export in \a exp if successful
+ * \retval negative value on error
+ */
static int ofd_connect_to_next(const struct lu_env *env, struct ofd_device *m,
const char *next, struct obd_export **exp)
{
obd = class_name2obd(next);
if (obd == NULL) {
CERROR("%s: can't locate next device: %s\n",
- m->ofd_dt_dev.dd_lu_dev.ld_obd->obd_name, next);
+ ofd_name(m), next);
GOTO(out, rc = -ENOTCONN);
}
rc = obd_connect(NULL, exp, obd, &obd->obd_uuid, data, NULL);
if (rc) {
CERROR("%s: cannot connect to next dev %s: rc = %d\n",
- m->ofd_dt_dev.dd_lu_dev.ld_obd->obd_name, next, rc);
+ ofd_name(m), next, rc);
GOTO(out, rc);
}
RETURN(rc);
}
+/**
+ * Initialize stack of devices.
+ *
+ * This function initializes OFD-OSD device stack to serve OST requests
+ *
+ * \param[in] env execution environment
+ * \param[in] m OFD device
+ * \param[in] cfg Lustre config for this server
+ *
+ * \retval 0 if successful
+ * \retval negative value on error
+ */
static int ofd_stack_init(const struct lu_env *env,
struct ofd_device *m, struct lustre_cfg *cfg)
{
RETURN(rc);
}
+/**
+ * Finalize the device stack OFD-OSD.
+ *
+ * This function cleans OFD-OSD device stack and
+ * disconnects OFD from the OSD.
+ *
+ * \param[in] env execution environment
+ * \param[in] m OFD device
+ * \param[in] top top device of stack
+ *
+ * \retval 0 if successful
+ * \retval negative value on error
+ */
static void ofd_stack_fini(const struct lu_env *env, struct ofd_device *m,
struct lu_device *top)
{
strcat(flags, "A");
lustre_cfg_bufs_set_string(&bufs, 1, flags);
lcfg = lustre_cfg_new(LCFG_CLEANUP, &bufs);
- if (!lcfg) {
- CERROR("Cannot alloc lcfg!\n");
+ if (lcfg == NULL)
RETURN_EXIT;
- }
LASSERT(top);
top->ld_ops->ldo_process_config(env, top, lcfg);
LASSERT(m->ofd_osd_exp);
obd_disconnect(m->ofd_osd_exp);
- m->ofd_osd = NULL;
EXIT;
}
{ NULL }
};
-/* Some parameters were moved from ofd to osd and only their
+/**
+ * Check if parameters are symlinks to the OSD.
+ *
+ * Some parameters were moved from ofd to osd and only their
* symlinks were kept in ofd by LU-3106. They are:
* -writehthrough_cache_enable
- * -readcache_max_filese
+ * -readcache_max_filesize
* -read_cache_enable
* -brw_stats
- * Since they are not included by the static lprocfs var list,
- * a pre-check is added for them to avoid "unknown param" error
- * message confuses the customer. If they are matched in this
- * check, they will be passed to the osd directly.
+ *
+ * Since they are not included by the static lprocfs var list, a pre-check
+ * is added for them to avoid "unknown param" errors. If they are matched
+ * in this check, they will be passed to the OSD directly.
+ *
+ * \param[in] param parameters to check
+ *
+ * \retval true if param is symlink to OSD param
+ * false otherwise
*/
static bool match_symlink_param(char *param)
{
return false;
}
-/* used by MGS to process specific configurations */
+/**
+ * Process various configuration parameters.
+ *
+ * This function is used by MGS to process specific configurations and
+ * pass them through to the next device in server stack, i.e. the OSD.
+ *
+ * \param[in] env execution environment
+ * \param[in] d LU device of OFD
+ * \param[in] cfg parameters to process
+ *
+ * \retval 0 if successful
+ * \retval negative value on error
+ */
static int ofd_process_config(const struct lu_env *env, struct lu_device *d,
struct lustre_cfg *cfg)
{
switch (cfg->lcfg_command) {
case LCFG_PARAM: {
- struct lprocfs_static_vars lvars;
-
+ struct obd_device *obd = ofd_obd(m);
/* For interoperability */
struct cfg_interop_param *ptr = NULL;
struct lustre_cfg *old_cfg = NULL;
break;
}
- lprocfs_ofd_init_vars(&lvars);
- rc = class_process_proc_param(PARAM_OST, lvars.obd_vars, cfg,
+ rc = class_process_proc_param(PARAM_OST, obd->obd_vars, cfg,
d->ld_obd);
if (rc > 0 || rc == -ENOSYS) {
CDEBUG(D_CONFIG, "pass param %s down the stack.\n",
RETURN(rc);
}
+/**
+ * Implementation of lu_object_operations::loo_object_init for OFD
+ *
+ * Allocate just the next object (OSD) in stack.
+ *
+ * \param[in] env execution environment
+ * \param[in] o lu_object of OFD object
+ * \param[in] conf additional configuration parameters, not used here
+ *
+ * \retval 0 if successful
+ * \retval negative value on error
+ */
static int ofd_object_init(const struct lu_env *env, struct lu_object *o,
const struct lu_object_conf *conf)
{
RETURN(rc);
}
+/**
+ * Implementation of lu_object_operations::loo_object_free.
+ *
+ * Finish OFD object lifecycle and free its memory.
+ *
+ * \param[in] env execution environment
+ * \param[in] o LU object of OFD object
+ */
static void ofd_object_free(const struct lu_env *env, struct lu_object *o)
{
struct ofd_object *of = ofd_obj(o);
EXIT;
}
+/**
+ * Implementation of lu_object_operations::loo_object_print.
+ *
+ * Print OFD part of compound OFD-OSD object. See lu_object_print() and
+ * LU_OBJECT_DEBUG() for more details about the compound object printing.
+ *
+ * \param[in] env execution environment
+ * \param[in] cookie opaque data passed to the printer function
+ * \param[in] p printer function to use
+ * \param[in] o LU object of OFD object
+ *
+ * \retval 0 if successful
+ * \retval negative value on error
+ */
static int ofd_object_print(const struct lu_env *env, void *cookie,
lu_printer_t p, const struct lu_object *o)
{
return (*p)(env, cookie, LUSTRE_OST_NAME"-object@%p", o);
}
-struct lu_object_operations ofd_obj_ops = {
+static struct lu_object_operations ofd_obj_ops = {
.loo_object_init = ofd_object_init,
.loo_object_free = ofd_object_free,
.loo_object_print = ofd_object_print
};
+/**
+ * Implementation of lu_device_operations::lod_object_alloc.
+ *
+ * This function allocates OFD part of compound OFD-OSD object and
+ * initializes its header, because OFD is the top device in stack
+ *
+ * \param[in] env execution environment
+ * \param[in] hdr object header, NULL for OFD
+ * \param[in] d lu_device
+ *
+ * \retval allocated object if successful
+ * \retval NULL value on failed allocation
+ */
static struct lu_object *ofd_object_alloc(const struct lu_env *env,
const struct lu_object_header *hdr,
struct lu_device *d)
ENTRY;
- OBD_SLAB_ALLOC_PTR_GFP(of, ofd_object_kmem, __GFP_IO);
+ OBD_SLAB_ALLOC_PTR_GFP(of, ofd_object_kmem, GFP_NOFS);
if (of != NULL) {
struct lu_object *o;
struct lu_object_header *h;
}
}
-extern int ost_handle(struct ptlrpc_request *req);
-
+/**
+ * Return the result of LFSCK run to the OFD.
+ *
+ * Notify OFD about result of LFSCK run. That may block the new object
+ * creation until problem is fixed by LFSCK.
+ *
+ * \param[in] env execution environment
+ * \param[in] data pointer to the OFD device
+ * \param[in] event LFSCK event type
+ *
+ * \retval 0 if successful
+ * \retval negative value on unknown event
+ */
static int ofd_lfsck_out_notify(const struct lu_env *env, void *data,
enum lfsck_events event)
{
ofd->ofd_lastid_rebuilding = 0;
ofd->ofd_lastid_gen++;
up_write(&ofd->ofd_lastid_rwsem);
+ CWARN("%s: Rebuilt crashed LAST_ID files successfully.\n",
+ obd->obd_name);
break;
}
default:
CERROR("%s: unknown lfsck event: rc = %d\n",
- ofd_obd(ofd)->obd_name, event);
+ ofd_name(ofd), event);
return -EINVAL;
}
return 0;
}
+/**
+ * Implementation of lu_device_operations::ldo_prepare.
+ *
+ * This method is called after layer has been initialized and before it starts
+ * serving user requests. In OFD it starts lfsk check routines and initializes
+ * recovery.
+ *
+ * \param[in] env execution environment
+ * \param[in] pdev higher device in stack, NULL for OFD
+ * \param[in] dev lu_device of OFD device
+ *
+ * \retval 0 if successful
+ * \retval negative value on error
+ */
static int ofd_prepare(const struct lu_env *env, struct lu_device *pdev,
struct lu_device *dev)
{
struct ofd_device *ofd = ofd_dev(dev);
struct obd_device *obd = ofd_obd(ofd);
struct lu_device *next = &ofd->ofd_osd->dd_lu_dev;
- struct lfsck_start_param lsp;
int rc;
ENTRY;
if (rc != 0)
RETURN(rc);
- rc = lfsck_register(env, ofd->ofd_osd, ofd->ofd_osd,
+ rc = lfsck_register(env, ofd->ofd_osd, ofd->ofd_osd, obd,
ofd_lfsck_out_notify, ofd, false);
if (rc != 0) {
CERROR("%s: failed to initialize lfsck: rc = %d\n",
RETURN(rc);
}
- lsp.lsp_start = NULL;
- lsp.lsp_namespace = ofd->ofd_namespace;
- rc = lfsck_start(env, ofd->ofd_osd, &lsp);
- if (rc != 0) {
- CWARN("%s: auto trigger paused LFSCK failed: rc = %d\n",
- obd->obd_name, rc);
- rc = 0;
- }
+ rc = lfsck_register_namespace(env, ofd->ofd_osd, ofd->ofd_namespace);
+ /* The LFSCK instance is registered just now, so it must be there when
+ * register the namespace to such instance. */
+ LASSERTF(rc == 0, "register namespace failed: rc = %d\n", rc);
target_recovery_init(&ofd->ofd_lut, tgt_request_handle);
LASSERT(obd->obd_no_conn);
RETURN(rc);
}
+/**
+ * Implementation of lu_device_operations::ldo_recovery_complete.
+ *
+ * This method notifies all layers about 'recovery complete' event. That means
+ * device is in full state and consistent. An OFD calculates available grant
+ * space upon this event.
+ *
+ * \param[in] env execution environment
+ * \param[in] dev lu_device of OFD device
+ *
+ * \retval 0 if successful
+ * \retval negative value on error
+ */
static int ofd_recovery_complete(const struct lu_env *env,
struct lu_device *dev)
{
ENTRY;
- /* Grant space for object precreation on the self export.
+ /*
+ * Grant space for object precreation on the self export.
* This initial reserved space (i.e. 10MB for zfs and 280KB for ldiskfs)
* is enough to create 10k objects. More space is then acquired for
* precreation in ofd_grant_create().
RETURN(rc);
}
+/**
+ * lu_device_operations matrix for OFD device.
+ */
static struct lu_device_operations ofd_lu_ops = {
.ldo_object_alloc = ofd_object_alloc,
.ldo_process_config = ofd_process_config,
.ldo_prepare = ofd_prepare,
};
+LPROC_SEQ_FOPS(lprocfs_nid_stats_clear);
+
+/**
+ * Initialize all needed procfs entries for OFD device.
+ *
+ * \param[in] ofd OFD device
+ *
+ * \retval 0 if successful
+ * \retval negative value on error
+ */
static int ofd_procfs_init(struct ofd_device *ofd)
{
- struct lprocfs_static_vars lvars;
struct obd_device *obd = ofd_obd(ofd);
- cfs_proc_dir_entry_t *entry;
+ struct proc_dir_entry *entry;
int rc = 0;
ENTRY;
/* lprocfs must be setup before the ofd so state can be safely added
* to /proc incrementally as the ofd is setup */
- lprocfs_ofd_init_vars(&lvars);
- rc = lprocfs_obd_setup(obd, lvars.obd_vars);
+ obd->obd_vars = lprocfs_ofd_obd_vars;
+ rc = lprocfs_obd_setup(obd);
if (rc) {
CERROR("%s: lprocfs_obd_setup failed: %d.\n",
obd->obd_name, rc);
obd->obd_uses_nid_stats = 1;
- entry = lprocfs_register("exports", obd->obd_proc_entry, NULL, NULL);
+ entry = lprocfs_seq_register("exports", obd->obd_proc_entry, NULL,
+ NULL);
if (IS_ERR(entry)) {
rc = PTR_ERR(entry);
CERROR("%s: error %d setting up lprocfs for %s\n",
obd->obd_proc_exports_entry = entry;
entry = lprocfs_add_simple(obd->obd_proc_exports_entry, "clear",
- lprocfs_nid_stats_clear_read,
- lprocfs_nid_stats_clear_write, obd, NULL);
+ obd, &lprocfs_nid_stats_clear_fops);
if (IS_ERR(entry)) {
rc = PTR_ERR(entry);
CERROR("%s: add proc entry 'clear' failed: %d.\n",
rc = lprocfs_job_stats_init(obd, LPROC_OFD_STATS_LAST,
ofd_stats_counter_init);
if (rc)
- GOTO(remove_entry_clear, rc);
+ GOTO(obd_cleanup, rc);
RETURN(0);
-remove_entry_clear:
- lprocfs_remove_proc_entry("clear", obd->obd_proc_exports_entry);
obd_cleanup:
lprocfs_obd_cleanup(obd);
lprocfs_free_obd_stats(obd);
return rc;
}
+/**
+ * Expose OSD statistics to OFD layer.
+ *
+ * The osd interfaces to the backend file system exposes useful data
+ * such as brw_stats and read or write cache states. This same data
+ * needs to be exposed into the obdfilter (ofd) layer to maintain
+ * backwards compatibility. This function creates the symlinks in the
+ * proc layer to enable this.
+ *
+ * \param[in] ofd OFD device
+ */
static void ofd_procfs_add_brw_stats_symlink(struct ofd_device *ofd)
{
struct obd_device *obd = ofd_obd(ofd);
struct obd_device *osd_obd = ofd->ofd_osd_exp->exp_obd;
- cfs_proc_dir_entry_t *osd_root = osd_obd->obd_type->typ_procroot;
- cfs_proc_dir_entry_t *osd_dir;
- osd_dir = lprocfs_srch(osd_root, obd->obd_name);
- if (osd_dir == NULL)
+ if (obd->obd_proc_entry == NULL)
return;
- if (lprocfs_srch(osd_dir, "brw_stats") != NULL)
- lprocfs_add_symlink("brw_stats", obd->obd_proc_entry,
- "../../%s/%s/brw_stats",
- osd_root->name, osd_dir->name);
-
- if (lprocfs_srch(osd_dir, "read_cache_enable") != NULL)
- lprocfs_add_symlink("read_cache_enable", obd->obd_proc_entry,
- "../../%s/%s/read_cache_enable",
- osd_root->name, osd_dir->name);
-
- if (lprocfs_srch(osd_dir, "readcache_max_filesize") != NULL)
- lprocfs_add_symlink("readcache_max_filesize",
- obd->obd_proc_entry,
- "../../%s/%s/readcache_max_filesize",
- osd_root->name, osd_dir->name);
-
- if (lprocfs_srch(osd_dir, "writethrough_cache_enable") != NULL)
- lprocfs_add_symlink("writethrough_cache_enable",
- obd->obd_proc_entry,
- "../../%s/%s/writethrough_cache_enable",
- osd_root->name, osd_dir->name);
+ lprocfs_add_symlink("brw_stats", obd->obd_proc_entry,
+ "../../%s/%s/brw_stats",
+ osd_obd->obd_type->typ_name, obd->obd_name);
+
+ lprocfs_add_symlink("read_cache_enable", obd->obd_proc_entry,
+ "../../%s/%s/read_cache_enable",
+ osd_obd->obd_type->typ_name, obd->obd_name);
+
+ lprocfs_add_symlink("readcache_max_filesize",
+ obd->obd_proc_entry,
+ "../../%s/%s/readcache_max_filesize",
+ osd_obd->obd_type->typ_name, obd->obd_name);
+
+ lprocfs_add_symlink("writethrough_cache_enable",
+ obd->obd_proc_entry,
+ "../../%s/%s/writethrough_cache_enable",
+ osd_obd->obd_type->typ_name, obd->obd_name);
}
+/**
+ * Cleanup all procfs entries in OFD.
+ *
+ * \param[in] ofd OFD device
+ */
static void ofd_procfs_fini(struct ofd_device *ofd)
{
struct obd_device *obd = ofd_obd(ofd);
- lprocfs_remove_proc_entry("writethrough_cache_enable",
- obd->obd_proc_entry);
- lprocfs_remove_proc_entry("readcache_max_filesize",
- obd->obd_proc_entry);
- lprocfs_remove_proc_entry("read_cache_enable", obd->obd_proc_entry);
- lprocfs_remove_proc_entry("brw_stats", obd->obd_proc_entry);
- lprocfs_remove_proc_entry("clear", obd->obd_proc_exports_entry);
lprocfs_free_per_client_stats(obd);
lprocfs_obd_cleanup(obd);
lprocfs_free_obd_stats(obd);
lprocfs_job_stats_fini(obd);
}
-extern int ost_handle(struct ptlrpc_request *req);
-
+/**
+ * Stop SEQ/FID server on OFD.
+ *
+ * \param[in] env execution environment
+ * \param[in] ofd OFD device
+ *
+ * \retval 0 if successful
+ * \retval negative value on error
+ */
int ofd_fid_fini(const struct lu_env *env, struct ofd_device *ofd)
{
return seq_site_fini(env, &ofd->ofd_seq_site);
}
+/**
+ * Start SEQ/FID server on OFD.
+ *
+ * The SEQ/FID server on OFD is needed to allocate FIDs for new objects.
+ * It also connects to the master server to get own FID sequence (SEQ) range
+ * to this particular OFD. Typically that happens when the OST is first
+ * formatted or in the rare case that it exhausts the local sequence range.
+ *
+ * The sequence range is allocated out to the MDTs for OST object allocations,
+ * and not directly to the clients.
+ *
+ * \param[in] env execution environment
+ * \param[in] ofd OFD device
+ *
+ * \retval 0 if successful
+ * \retval negative value on error
+ */
int ofd_fid_init(const struct lu_env *env, struct ofd_device *ofd)
{
struct seq_server_site *ss = &ofd->ofd_seq_site;
return rc;
}
-int ofd_set_info_hdl(struct tgt_session_info *tsi)
+/**
+ * OFD request handler for OST_SET_INFO RPC.
+ *
+ * This is OFD-specific part of request handling
+ *
+ * \param[in] tsi target session environment for this request
+ *
+ * \retval 0 if successful
+ * \retval negative value on error
+ */
+static int ofd_set_info_hdl(struct tgt_session_info *tsi)
{
struct ptlrpc_request *req = tgt_ses_req(tsi);
struct ost_body *body = NULL, *repbody;
RETURN(rc);
}
+/**
+ * Get FIEMAP (FIle Extent MAPping) for object with the given FID.
+ *
+ * This function returns a list of extents which describes how a file's
+ * blocks are laid out on the disk.
+ *
+ * \param[in] env execution environment
+ * \param[in] ofd OFD device
+ * \param[in] fid FID of object
+ * \param[in] fiemap fiemap structure to fill with data
+ *
+ * \retval 0 if \a fiemap is filled with data successfully
+ * \retval negative value on error
+ */
int ofd_fiemap_get(const struct lu_env *env, struct ofd_device *ofd,
struct lu_fid *fid, struct ll_user_fiemap *fiemap)
{
}
struct locked_region {
- cfs_list_t list;
+ struct list_head list;
struct lustre_handle lh;
};
+/**
+ * Lock single extent and save lock handle in the list.
+ *
+ * This is supplemental function for lock_zero_regions(). It allocates
+ * new locked_region structure and locks it with extent lock, then adds
+ * it to the list of all such regions.
+ *
+ * \param[in] ns LDLM namespace
+ * \param[in] res_id resource ID
+ * \param[in] begin start of region
+ * \param[in] end end of region
+ * \param[in] locked list head of regions list
+ *
+ * \retval 0 if successful locking
+ * \retval negative value on error
+ */
static int lock_region(struct ldlm_namespace *ns, struct ldlm_res_id *res_id,
unsigned long long begin, unsigned long long end,
- cfs_list_t *locked)
+ struct list_head *locked)
{
struct locked_region *region = NULL;
__u64 flags = 0;
CDEBUG(D_OTHER, "ost lock [%llu,%llu], lh=%p\n", begin, end,
®ion->lh);
- cfs_list_add(®ion->list, locked);
+ list_add(®ion->list, locked);
return 0;
}
+/**
+ * Lock the sparse areas of given resource.
+ *
+ * The locking of sparse areas will cause dirty data to be flushed back from
+ * clients. This is used when getting the FIEMAP of an object to make sure
+ * there is no unaccounted cached data on clients.
+ *
+ * This function goes through \a fiemap list of extents and locks only sparse
+ * areas between extents.
+ *
+ * \param[in] ns LDLM namespace
+ * \param[in] res_id resource ID
+ * \param[in] fiemap file extents mapping on disk
+ * \param[in] locked list head of regions list
+ *
+ * \retval 0 if successful
+ * \retval negative value on error
+ */
static int lock_zero_regions(struct ldlm_namespace *ns,
struct ldlm_res_id *res_id,
struct ll_user_fiemap *fiemap,
- cfs_list_t *locked)
+ struct list_head *locked)
{
__u64 begin = fiemap->fm_start;
unsigned int i;
RETURN(rc);
}
-static void unlock_zero_regions(struct ldlm_namespace *ns, cfs_list_t *locked)
+/**
+ * Unlock all previously locked sparse areas for given resource.
+ *
+ * This function goes through list of locked regions, unlocking and freeing
+ * them one-by-one.
+ *
+ * \param[in] ns LDLM namespace
+ * \param[in] locked list head of regions list
+ */
+static void
+unlock_zero_regions(struct ldlm_namespace *ns, struct list_head *locked)
{
struct locked_region *entry, *temp;
- cfs_list_for_each_entry_safe(entry, temp, locked, list) {
+ list_for_each_entry_safe(entry, temp, locked, list) {
CDEBUG(D_OTHER, "ost unlock lh=%p\n", &entry->lh);
tgt_extent_unlock(&entry->lh, LCK_PR);
- cfs_list_del(&entry->list);
+ list_del(&entry->list);
OBD_FREE_PTR(entry);
}
}
-int ofd_get_info_hdl(struct tgt_session_info *tsi)
+/**
+ * OFD request handler for OST_GET_INFO RPC.
+ *
+ * This is OFD-specific part of request handling. The OFD-specific keys are:
+ * - KEY_LAST_ID (obsolete)
+ * - KEY_FIEMAP
+ * - KEY_LAST_FID
+ *
+ * This function reads needed data from storage and fills reply with it.
+ *
+ * Note: the KEY_LAST_ID is obsolete, replaced by KEY_LAST_FID on newer MDTs,
+ * and is kept for compatibility.
+ *
+ * \param[in] tsi target session environment for this request
+ *
+ * \retval 0 if successful
+ * \retval negative value on error
+ */
+static int ofd_get_info_hdl(struct tgt_session_info *tsi)
{
struct obd_export *exp = tsi->tsi_exp;
struct ofd_device *ofd = ofd_exp(exp);
RCL_CLIENT);
if (KEY_IS(KEY_LAST_ID)) {
- obd_id *last_id;
+ u64 *last_id;
struct ofd_seq *oseq;
req_capsule_extend(tsi->tsi_pill, &RQF_OST_GET_INFO_LAST_ID);
last_id = req_capsule_server_get(tsi->tsi_pill, &RMF_OBD_ID);
oseq = ofd_seq_load(tsi->tsi_env, ofd,
- (obd_seq)exp->exp_filter_data.fed_group);
+ (u64)exp->exp_filter_data.fed_group);
if (IS_ERR(oseq))
rc = -EFAULT;
else
* flushed back from client, then call fiemap again. */
if (fm_key->oa.o_valid & OBD_MD_FLFLAGS &&
fm_key->oa.o_flags & OBD_FL_SRVLOCK) {
- cfs_list_t locked = CFS_LIST_HEAD_INIT(locked);
+ struct list_head locked;
+ INIT_LIST_HEAD(&locked);
ost_fid_build_resid(fid, &fti->fti_resid);
rc = lock_zero_regions(ofd->ofd_namespace,
&fti->fti_resid, fiemap,
&locked);
- if (rc == 0 && !cfs_list_empty(&locked)) {
+ if (rc == 0 && !list_empty(&locked)) {
rc = ofd_fiemap_get(tsi->tsi_env, ofd, fid,
fiemap);
unlock_zero_regions(ofd->ofd_namespace,
RETURN(rc);
}
+/**
+ * OFD request handler for OST_GETATTR RPC.
+ *
+ * This is OFD-specific part of request handling. It finds the OFD object
+ * by its FID, gets attributes from storage and packs result to the reply.
+ *
+ * \param[in] tsi target session environment for this request
+ *
+ * \retval 0 if successful
+ * \retval negative value on error
+ */
static int ofd_getattr_hdl(struct tgt_session_info *tsi)
{
struct ofd_thread_info *fti = tsi2ofd_info(tsi);
RETURN(rc);
}
+/**
+ * OFD request handler for OST_SETATTR RPC.
+ *
+ * This is OFD-specific part of request handling. It finds the OFD object
+ * by its FID, sets attributes from request and packs result to the reply.
+ *
+ * \param[in] tsi target session environment for this request
+ *
+ * \retval 0 if successful
+ * \retval negative value on error
+ */
static int ofd_setattr_hdl(struct tgt_session_info *tsi)
{
struct ofd_thread_info *fti = tsi2ofd_info(tsi);
* to go... deadlock! */
res = ldlm_resource_get(ofd->ofd_namespace, NULL,
&tsi->tsi_resid, LDLM_EXTENT, 0);
- if (res != NULL) {
+ if (!IS_ERR(res)) {
ldlm_res_lvbo_update(res, NULL, 0);
ldlm_resource_putref(res);
}
return rc;
}
+/**
+ * Destroy OST orphans.
+ *
+ * This is part of OST_CREATE RPC handling. If there is flag OBD_FL_DELORPHAN
+ * set then we must destroy possible orphaned objects.
+ *
+ * \param[in] env execution environment
+ * \param[in] exp OBD export
+ * \param[in] ofd OFD device
+ * \param[in] oa obdo structure for reply
+ *
+ * \retval 0 if successful
+ * \retval negative value on error
+ */
static int ofd_orphans_destroy(const struct lu_env *env,
struct obd_export *exp,
struct ofd_device *ofd, struct obdo *oa)
struct lu_fid *fid = &info->fti_fid;
struct ost_id *oi = &oa->o_oi;
struct ofd_seq *oseq;
- obd_seq seq = ostid_seq(oi);
- obd_id end_id = ostid_id(oi);
- obd_id last;
- obd_id oid;
+ u64 seq = ostid_seq(oi);
+ u64 end_id = ostid_id(oi);
+ u64 last;
+ u64 oid;
int skip_orphan;
int rc = 0;
LASSERT(exp != NULL);
skip_orphan = !!(exp_connect_flags(exp) & OBD_CONNECT_SKIP_ORPHAN);
+ if (OBD_FAIL_CHECK(OBD_FAIL_OST_NODESTROY))
+ goto done;
+
LCONSOLE(D_INFO, "%s: deleting orphan objects from "DOSTID
" to "DOSTID"\n", ofd_name(ofd), seq, end_id + 1, seq, last);
GOTO(out_put, rc);
rc = ofd_destroy_by_fid(env, ofd, fid, 1);
- if (rc != 0 && rc != -ENOENT) /* this is pretty fatal... */
- CEMERG("%s: error destroying precreated id "DFID
- ": rc = %d\n", ofd_name(ofd), PFID(fid), rc);
+ if (rc != 0 && rc != -ENOENT && rc != -ESTALE &&
+ likely(rc != -EREMCHG && rc != -EINPROGRESS))
+ /* this is pretty fatal... */
+ CEMERG("%s: error destroying precreated id "
+ DFID": rc = %d\n",
+ ofd_name(ofd), PFID(fid), rc);
oid--;
if (!skip_orphan) {
CDEBUG(D_HA, "%s: after destroy: set last_id to "DOSTID"\n",
ofd_name(ofd), seq, oid);
+done:
if (!skip_orphan) {
+ ofd_seq_last_oid_set(oseq, oid);
rc = ofd_seq_last_oid_write(env, ofd, oseq);
} else {
/* don't reuse orphan object, return last used objid */
return rc;
}
+/**
+ * OFD request handler for OST_CREATE RPC.
+ *
+ * This is OFD-specific part of request handling. Its main purpose is to
+ * create new data objects on OST, but it also used to destroy orphans.
+ *
+ * \param[in] tsi target session environment for this request
+ *
+ * \retval 0 if successful
+ * \retval negative value on error
+ */
static int ofd_create_hdl(struct tgt_session_info *tsi)
{
+ struct ptlrpc_request *req = tgt_ses_req(tsi);
struct ost_body *repbody;
const struct obdo *oa = &tsi->tsi_ost_body->oa;
struct obdo *rep_oa;
struct obd_export *exp = tsi->tsi_exp;
struct ofd_device *ofd = ofd_exp(exp);
- obd_seq seq = ostid_seq(&oa->o_oi);
- obd_id oid = ostid_id(&oa->o_oi);
+ u64 seq = ostid_seq(&oa->o_oi);
+ u64 oid = ostid_id(&oa->o_oi);
struct ofd_seq *oseq;
int rc = 0, diff;
int sync_trans = 0;
CDEBUG(D_HA, "ofd_last_id() = "LPU64" -> diff = %d\n",
ofd_seq_last_oid(oseq), diff);
if (-diff > OST_MAX_PRECREATE) {
- /* FIXME: should reset precreate_next_id on MDS */
+ /* Let MDS know that we are so far ahead. */
+ ostid_set_id(&rep_oa->o_oi, ofd_seq_last_oid(oseq) + 1);
rc = 0;
} else if (diff < 0) {
rc = ofd_orphans_destroy(tsi->tsi_env, exp,
} else {
if (unlikely(exp->exp_filter_data.fed_lastid_gen !=
ofd->ofd_lastid_gen)) {
- ofd_obd_disconnect(exp);
+ /* Keep the export ref so we can send the reply. */
+ ofd_obd_disconnect(class_export_get(exp));
GOTO(out_nolock, rc = -ENOTCONN);
}
ofd_name(ofd), POSTID(&oa->o_oi));
GOTO(out, rc = -EINVAL);
}
+
+ if (diff < 0) {
+ /* LU-5648 */
+ CERROR("%s: invalid precreate request for "
+ DOSTID", last_id " LPU64 ". "
+ "Likely MDS last_id corruption\n",
+ ofd_name(ofd), POSTID(&oa->o_oi),
+ ofd_seq_last_oid(oseq));
+ GOTO(out, rc = -EINVAL);
+ }
}
}
if (diff > 0) {
cfs_time_t enough_time = cfs_time_shift(DISK_TIMEOUT);
- obd_id next_id;
+ u64 next_id;
int created = 0;
int count;
" at "LPU64"\n", ofd_name(ofd),
count, seq, next_id);
- if (cfs_time_after(jiffies, enough_time)) {
- LCONSOLE_WARN("%s: Slow creates, %d/%d objects"
- " created at a rate of %d/s\n",
- ofd_name(ofd), created,
- diff + created,
- created / DISK_TIMEOUT);
+ if (!(lustre_msg_get_flags(req->rq_reqmsg) & MSG_REPLAY)
+ && cfs_time_after(jiffies, enough_time)) {
+ CDEBUG(D_HA, "%s: Slow creates, %d/%d objects"
+ " created at a rate of %d/s\n",
+ ofd_name(ofd), created, diff + created,
+ created / DISK_TIMEOUT);
break;
}
break;
}
}
+
+ if (diff > 0 &&
+ lustre_msg_get_flags(req->rq_reqmsg) & MSG_REPLAY)
+ LCONSOLE_WARN("%s: can't create the same count of"
+ " objects when replaying the request"
+ " (diff is %d). see LU-4621\n",
+ ofd_name(ofd), diff);
+
if (created > 0)
/* some objects got created, we can return
* them, even if last creation failed */
out:
mutex_unlock(&oseq->os_create_lock);
out_nolock:
- if (rc == 0)
+ if (rc == 0) {
+#if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(2, 8, 53, 0)
+ struct ofd_thread_info *info = ofd_info(tsi->tsi_env);
+ struct lu_fid *fid = &info->fti_fid;
+
+ /* For compatible purpose, it needs to convert back to
+ * OST ID before put it on wire. */
+ *fid = rep_oa->o_oi.oi_fid;
+ fid_to_ostid(fid, &rep_oa->o_oi);
+#endif
rep_oa->o_valid |= OBD_MD_FLID | OBD_MD_FLGROUP;
-
+ }
ofd_seq_put(tsi->tsi_env, oseq);
out_sem:
return rc;
}
+/**
+ * OFD request handler for OST_DESTROY RPC.
+ *
+ * This is OFD-specific part of request handling. It destroys data objects
+ * related to destroyed object on MDT.
+ *
+ * \param[in] tsi target session environment for this request
+ *
+ * \retval 0 if successful
+ * \retval negative value on error
+ */
static int ofd_destroy_hdl(struct tgt_session_info *tsi)
{
const struct ost_body *body = tsi->tsi_ost_body;
struct ofd_device *ofd = ofd_exp(tsi->tsi_exp);
struct ofd_thread_info *fti = tsi2ofd_info(tsi);
struct lu_fid *fid = &fti->fti_fid;
- obd_id oid;
- obd_count count;
+ u64 oid;
+ u32 count;
int rc = 0;
ENTRY;
dlm = req_capsule_client_get(tsi->tsi_pill, &RMF_DLM_REQ);
if (dlm == NULL)
RETURN(-EFAULT);
- ldlm_request_cancel(tgt_ses_req(tsi), dlm, 0);
+ ldlm_request_cancel(tgt_ses_req(tsi), dlm, 0, LATF_SKIP);
}
*fid = body->oa.o_oi.oi_fid;
return rc;
}
+/**
+ * OFD request handler for OST_STATFS RPC.
+ *
+ * This function gets statfs data from storage as part of request
+ * processing.
+ *
+ * \param[in] tsi target session environment for this request
+ *
+ * \retval 0 if successful
+ * \retval negative value on error
+ */
static int ofd_statfs_hdl(struct tgt_session_info *tsi)
{
struct obd_statfs *osfs;
RETURN(rc);
}
+/**
+ * OFD request handler for OST_SYNC RPC.
+ *
+ * Sync object data or all filesystem data to the disk and pack the
+ * result in reply.
+ *
+ * \param[in] tsi target session environment for this request
+ *
+ * \retval 0 if successful
+ * \retval negative value on error
+ */
static int ofd_sync_hdl(struct tgt_session_info *tsi)
{
struct ost_body *body = tsi->tsi_ost_body;
}
rc = tgt_sync(tsi->tsi_env, tsi->tsi_tgt,
- fo != NULL ? ofd_object_child(fo) : NULL);
+ fo != NULL ? ofd_object_child(fo) : NULL,
+ repbody->oa.o_size, repbody->oa.o_blocks);
if (rc)
GOTO(put, rc);
return rc;
}
+/**
+ * OFD request handler for OST_PUNCH RPC.
+ *
+ * This is part of request processing. Validate request fields,
+ * punch (truncate) the given OFD object and pack reply.
+ *
+ * \param[in] tsi target session environment for this request
+ *
+ * \retval 0 if successful
+ * \retval negative value on error
+ */
static int ofd_punch_hdl(struct tgt_session_info *tsi)
{
const struct obdo *oa = &tsi->tsi_ost_body->oa;
}
rc = ofd_object_punch(tsi->tsi_env, fo, start, end, &info->fti_attr,
- ff);
+ ff, (struct obdo *)oa);
if (rc)
GOTO(out_put, rc);
* to go... deadlock! */
res = ldlm_resource_get(ns, NULL, &tsi->tsi_resid,
LDLM_EXTENT, 0);
- if (res != NULL) {
+ if (!IS_ERR(res)) {
ldlm_res_lvbo_update(res, NULL, 0);
ldlm_resource_putref(res);
}
return rc;
}
+/**
+ * OFD request handler for OST_QUOTACTL RPC.
+ *
+ * This is part of request processing to validate incoming request fields,
+ * get the requested data from OSD and pack reply.
+ *
+ * \param[in] tsi target session environment for this request
+ *
+ * \retval 0 if successful
+ * \retval negative value on error
+ */
static int ofd_quotactl(struct tgt_session_info *tsi)
{
struct obd_quotactl *oqctl, *repoqc;
+ struct lu_nodemap *nodemap =
+ tsi->tsi_exp->exp_target_data.ted_nodemap;
+ int id;
int rc;
ENTRY;
RETURN(0);
*repoqc = *oqctl;
+
+ id = repoqc->qc_id;
+ if (oqctl->qc_type == USRQUOTA)
+ id = nodemap_map_id(nodemap, NODEMAP_UID,
+ NODEMAP_CLIENT_TO_FS,
+ repoqc->qc_id);
+ else if (oqctl->qc_type == GRPQUOTA)
+ id = nodemap_map_id(nodemap, NODEMAP_GID,
+ NODEMAP_CLIENT_TO_FS,
+ repoqc->qc_id);
+
+ if (repoqc->qc_id != id)
+ swap(repoqc->qc_id, id);
+
rc = lquotactl_slv(tsi->tsi_env, tsi->tsi_tgt->lut_bottom, repoqc);
ofd_counter_incr(tsi->tsi_exp, LPROC_OFD_STATS_QUOTACTL,
tsi->tsi_jobid, 1);
+ if (repoqc->qc_id != id)
+ swap(repoqc->qc_id, id);
+
RETURN(rc);
}
-/* High priority request handlers for OFD */
-
-/* prolong locks for the current service time of the corresponding
- * portal (= OST_IO_PORTAL)
+/**
+ * Calculate the amount of time for lock prolongation.
+ *
+ * This is helper for ofd_prolong_extent_locks() function to get
+ * the timeout extra time.
+ *
+ * \param[in] req current request
+ *
+ * \retval amount of time to extend the timeout with
*/
-static inline int prolong_timeout(struct ptlrpc_request *req)
+static inline int prolong_timeout(struct ptlrpc_request *req,
+ struct ldlm_lock *lock)
{
struct ptlrpc_service_part *svcpt = req->rq_rqbd->rqbd_svcpt;
if (AT_OFF)
return obd_timeout / 2;
- return max(at_est2timeout(at_get(&svcpt->scp_at_estimate)),
- ldlm_timeout);
+ /* We are in the middle of the process - BL AST is sent, CANCEL
+ is ahead. Take half of AT + IO process time. */
+ return at_est2timeout(at_get(&svcpt->scp_at_estimate)) +
+ (ldlm_bl_timeout(lock) >> 1);
}
+/**
+ * Prolong single lock timeout.
+ *
+ * This is supplemental function to the ofd_prolong_locks(). It prolongs
+ * a single lock.
+ *
+ * \param[in] tsi target session environment for this request
+ * \param[in] lock LDLM lock to prolong
+ * \param[in] extent related extent
+ * \param[in] timeout timeout value to add
+ *
+ * \retval 0 if lock is not suitable for prolongation
+ * \retval 1 if lock was prolonged successfully
+ */
static int ofd_prolong_one_lock(struct tgt_session_info *tsi,
struct ldlm_lock *lock,
- struct ldlm_extent *extent, int timeout)
+ struct ldlm_extent *extent)
{
+ int timeout = prolong_timeout(tgt_ses_req(tsi), lock);
if (lock->l_flags & LDLM_FL_DESTROYED) /* lock already cancelled */
return 0;
return 1;
}
+/**
+ * Prolong lock timeout for the given extent.
+ *
+ * This function finds all locks related with incoming request and
+ * prolongs their timeout.
+ *
+ * If a client is holding a lock for a long time while it sends
+ * read or write RPCs to the OST for the object under this lock,
+ * then we don't want the OST to evict the client. Otherwise,
+ * if the network or disk is very busy then the client may not
+ * be able to make any progress to clear out dirty pages under
+ * the lock and the application will fail.
+ *
+ * Every time a Bulk Read/Write (BRW) request arrives for the object
+ * covered by the lock, extend the timeout on that lock. The RPC should
+ * contain a lock handle for the lock it is using, but this
+ * isn't handled correctly by all client versions, and the
+ * request may cover multiple locks.
+ *
+ * \param[in] tsi target session environment for this request
+ * \param[in] start start of extent
+ * \param[in] end end of extent
+ *
+ * \retval number of prolonged locks
+ */
static int ofd_prolong_extent_locks(struct tgt_session_info *tsi,
__u64 start, __u64 end)
{
.end = end
};
struct ldlm_lock *lock;
- int timeout = prolong_timeout(tgt_ses_req(tsi));
int lock_count = 0;
ENTRY;
/* bingo */
LASSERT(lock->l_export == exp);
lock_count = ofd_prolong_one_lock(tsi, lock,
- &extent, timeout);
+ &extent);
LDLM_LOCK_PUT(lock);
RETURN(lock_count);
}
&extent))
continue;
- lock_count += ofd_prolong_one_lock(tsi, lock, &extent, timeout);
+ lock_count += ofd_prolong_one_lock(tsi, lock, &extent);
}
spin_unlock_bh(&exp->exp_bl_list_lock);
}
/**
- * Returns 1 if the given PTLRPC matches the given LDLM lock, or 0 if it does
- * not.
+ * Implementation of ptlrpc_hpreq_ops::hpreq_lock_match for OFD RW requests.
+ *
+ * Determine if \a lock and the lock from request \a req are equivalent
+ * by comparing their resource names, modes, and extents.
+ *
+ * It is used to give priority to read and write RPCs being done
+ * under this lock so that the client can drop the contended
+ * lock more quickly and let other clients use it. This improves
+ * overall performance in the case where the first client gets a
+ * very large lock extent that prevents other clients from
+ * submitting their writes.
+ *
+ * \param[in] req ptlrpc_request being processed
+ * \param[in] lock contended lock to match
+ *
+ * \retval 1 if lock is matched
+ * \retval 0 otherwise
*/
static int ofd_rw_hpreq_lock_match(struct ptlrpc_request *req,
struct ldlm_lock *lock)
rnb = req_capsule_client_get(&req->rq_pill, &RMF_NIOBUF_REMOTE);
LASSERT(rnb != NULL);
- ext.start = rnb->offset;
+ ext.start = rnb->rnb_offset;
rnb += ioo->ioo_bufcnt - 1;
- ext.end = rnb->offset + rnb->len - 1;
+ ext.end = rnb->rnb_offset + rnb->rnb_len - 1;
LASSERT(lock->l_resource != NULL);
if (!ostid_res_name_eq(&ioo->ioo_oid, &lock->l_resource->lr_name))
RETURN(0);
+ /* a bulk write can only hold a reference on a PW extent lock */
mode = LCK_PW;
if (opc == OST_READ)
+ /* whereas a bulk read can be protected by either a PR or PW
+ * extent lock */
mode |= LCK_PR;
if (!(lock->l_granted_mode & mode))
}
/**
- * High-priority queue request check for whether the given PTLRPC request
- * (\a req) is blocking an LDLM lock cancel.
+ * Implementation of ptlrpc_hpreq_ops::hpreq_lock_check for OFD RW requests.
*
- * Returns 1 if the given given PTLRPC request (\a req) is blocking an LDLM lock
- * cancel, 0 if it is not, and -EFAULT if the request is malformed.
+ * Check for whether the given PTLRPC request (\a req) is blocking
+ * an LDLM lock cancel.
*
- * Only OST_READs, OST_WRITEs and OST_PUNCHes go on the h-p RPC queue. This
- * function looks only at OST_READs and OST_WRITEs.
+ * \param[in] req the incoming request
+ *
+ * \retval 1 if \a req is blocking an LDLM lock cancel
+ * \retval 0 if it is not
*/
static int ofd_rw_hpreq_check(struct ptlrpc_request *req)
{
/* Don't use tgt_ses_info() to get session info, because lock_match()
* can be called while request has no processing thread yet. */
tsi = lu_context_key_get(&req->rq_session, &tgt_session_key);
- LASSERT(tsi != NULL);
/*
* Use LASSERT below because malformed RPCs should have
rnb = req_capsule_client_get(&req->rq_pill, &RMF_NIOBUF_REMOTE);
LASSERT(rnb != NULL);
- LASSERT(!(rnb->flags & OBD_BRW_SRVLOCK));
+ LASSERT(!(rnb->rnb_flags & OBD_BRW_SRVLOCK));
- start = rnb->offset;
+ start = rnb->rnb_offset;
rnb += ioo->ioo_bufcnt - 1;
- end = rnb->offset + rnb->len - 1;
+ end = rnb->rnb_offset + rnb->rnb_len - 1;
DEBUG_REQ(D_RPCTRACE, req, "%s %s: refresh rw locks: "DFID
" ("LPU64"->"LPU64")\n",
RETURN(lock_count > 0);
}
+/**
+ * Implementation of ptlrpc_hpreq_ops::hpreq_lock_fini for OFD RW requests.
+ *
+ * Called after the request has been handled. It refreshes lock timeout again
+ * so that client has more time to send lock cancel RPC.
+ *
+ * \param[in] req request which is being processed.
+ */
static void ofd_rw_hpreq_fini(struct ptlrpc_request *req)
{
ofd_rw_hpreq_check(req);
}
/**
- * Like tgt_rw_hpreq_lock_match(), but for OST_PUNCH RPCs.
+ * Implementation of ptlrpc_hpreq_ops::hpreq_lock_match for OST_PUNCH request.
+ *
+ * This function checks if the given lock is the same by its resname, mode
+ * and extent as one taken from the request.
+ * It is used to give priority to punch/truncate RPCs that might lead to
+ * the fastest release of that lock when a lock is contended.
+ *
+ * \param[in] req ptlrpc_request being processed
+ * \param[in] lock contended lock to match
+ *
+ * \retval 1 if lock is matched
+ * \retval 0 otherwise
*/
static int ofd_punch_hpreq_lock_match(struct ptlrpc_request *req,
struct ldlm_lock *lock)
/* Don't use tgt_ses_info() to get session info, because lock_match()
* can be called while request has no processing thread yet. */
tsi = lu_context_key_get(&req->rq_session, &tgt_session_key);
- LASSERT(tsi != NULL);
+ /*
+ * Use LASSERT below because malformed RPCs should have
+ * been filtered out in tgt_hpreq_handler().
+ */
LASSERT(tsi->tsi_ost_body != NULL);
if (tsi->tsi_ost_body->oa.o_valid & OBD_MD_FLHANDLE &&
tsi->tsi_ost_body->oa.o_handle.cookie == lock->l_handle.h_cookie)
}
/**
- * Like ost_rw_hpreq_check(), but for OST_PUNCH RPCs.
+ * Implementation of ptlrpc_hpreq_ops::hpreq_lock_check for OST_PUNCH request.
+ *
+ * High-priority queue request check for whether the given punch request
+ * (\a req) is blocking an LDLM lock cancel.
+ *
+ * \param[in] req the incoming request
+ *
+ * \retval 1 if \a req is blocking an LDLM lock cancel
+ * \retval 0 if it is not
*/
static int ofd_punch_hpreq_check(struct ptlrpc_request *req)
{
RETURN(lock_count > 0);
}
+/**
+ * Implementation of ptlrpc_hpreq_ops::hpreq_lock_fini for OST_PUNCH request.
+ *
+ * Called after the request has been handled. It refreshes lock timeout again
+ * so that client has more time to send lock cancel RPC.
+ *
+ * \param[in] req request which is being processed.
+ */
static void ofd_punch_hpreq_fini(struct ptlrpc_request *req)
{
ofd_punch_hpreq_check(req);
}
-struct ptlrpc_hpreq_ops ofd_hpreq_rw = {
+static struct ptlrpc_hpreq_ops ofd_hpreq_rw = {
.hpreq_lock_match = ofd_rw_hpreq_lock_match,
.hpreq_check = ofd_rw_hpreq_check,
.hpreq_fini = ofd_rw_hpreq_fini
};
-struct ptlrpc_hpreq_ops ofd_hpreq_punch = {
+static struct ptlrpc_hpreq_ops ofd_hpreq_punch = {
.hpreq_lock_match = ofd_punch_hpreq_lock_match,
.hpreq_check = ofd_punch_hpreq_check,
.hpreq_fini = ofd_punch_hpreq_fini
};
-/** Assign high priority operations to the IO requests */
+/**
+ * Assign high priority operations to an IO request.
+ *
+ * Check if the incoming request is a candidate for
+ * high-priority processing. If it is, assign it a high
+ * priority operations table.
+ *
+ * \param[in] tsi target session environment for this request
+ */
static void ofd_hp_brw(struct tgt_session_info *tsi)
{
struct niobuf_remote *rnb;
LASSERT(rnb != NULL); /* must exist after request preprocessing */
/* no high priority if server lock is needed */
- if (rnb->flags & OBD_BRW_SRVLOCK)
+ if (rnb->rnb_flags & OBD_BRW_SRVLOCK)
return;
}
tgt_ses_req(tsi)->rq_ops = &ofd_hpreq_rw;
}
+/**
+ * Assign high priority operations to an punch request.
+ *
+ * Check if the incoming request is a candidate for
+ * high-priority processing. If it is, assign it a high
+ * priority operations table.
+ *
+ * \param[in] tsi target session environment for this request
+ */
static void ofd_hp_punch(struct tgt_session_info *tsi)
{
LASSERT(tsi->tsi_ost_body != NULL); /* must exists if we are here */
#define OST_BRW_READ OST_READ
#define OST_BRW_WRITE OST_WRITE
+/**
+ * Table of OFD-specific request handlers
+ *
+ * This table contains all opcodes accepted by OFD and
+ * specifies handlers for them. The tgt_request_handler()
+ * uses such table from each target to process incoming
+ * requests.
+ */
static struct tgt_handler ofd_tgt_handlers[] = {
TGT_RPC_HANDLER(OST_FIRST_OPC,
0, OST_CONNECT, tgt_connect,
.tos_hs = tgt_dlm_handlers
},
{
- .tos_opc_start = UPDATE_OBJ,
- .tos_opc_end = UPDATE_LAST_OPC,
+ .tos_opc_start = OUT_UPDATE_FIRST_OPC,
+ .tos_opc_end = OUT_UPDATE_LAST_OPC,
.tos_hs = tgt_out_handlers
},
{
.tos_hs = seq_handlers
},
{
+ .tos_opc_start = LFSCK_FIRST_OPC,
+ .tos_opc_end = LFSCK_LAST_OPC,
+ .tos_hs = tgt_lfsck_handlers
+ },
+ {
.tos_hs = NULL
}
};
+/* context key constructor/destructor: ofd_key_init(), ofd_key_fini() */
+LU_KEY_INIT_FINI(ofd, struct ofd_thread_info);
+
+/**
+ * Implementation of lu_context_key::lct_key_exit.
+ *
+ * Optional method called on lu_context_exit() for all allocated
+ * keys.
+ * It is used in OFD to sanitize context values which may be re-used
+ * during another request processing by the same thread.
+ *
+ * \param[in] ctx execution context
+ * \param[in] key context key
+ * \param[in] data ofd_thread_info
+ */
+static void ofd_key_exit(const struct lu_context *ctx,
+ struct lu_context_key *key, void *data)
+{
+ struct ofd_thread_info *info = data;
+
+ info->fti_env = NULL;
+ info->fti_exp = NULL;
+
+ info->fti_xid = 0;
+ info->fti_pre_version = 0;
+ info->fti_used = 0;
+
+ memset(&info->fti_attr, 0, sizeof info->fti_attr);
+}
+
+struct lu_context_key ofd_thread_key = {
+ .lct_tags = LCT_DT_THREAD,
+ .lct_init = ofd_key_init,
+ .lct_fini = ofd_key_fini,
+ .lct_exit = ofd_key_exit
+};
+
+/**
+ * Initialize OFD device according to parameters in the config log \a cfg.
+ *
+ * This is the main starting point of OFD initialization. It fills all OFD
+ * parameters with their initial values and calls other initializing functions
+ * to set up all OFD subsystems.
+ *
+ * \param[in] env execution environment
+ * \param[in] m OFD device
+ * \param[in] ldt LU device type of OFD
+ * \param[in] cfg configuration log
+ *
+ * \retval 0 if successful
+ * \retval negative value on error
+ */
static int ofd_init0(const struct lu_env *env, struct ofd_device *m,
struct lu_device_type *ldt, struct lustre_cfg *cfg)
{
m->ofd_tot_granted = 0;
m->ofd_tot_pending = 0;
m->ofd_seq_count = 0;
+ init_waitqueue_head(&m->ofd_inconsistency_thread.t_ctl_waitq);
+ INIT_LIST_HEAD(&m->ofd_inconsistency_list);
+ spin_lock_init(&m->ofd_inconsistency_lock);
spin_lock_init(&m->ofd_batch_lock);
- rwlock_init(&obd->u.filter.fo_sptlrpc_lock);
- sptlrpc_rule_set_init(&obd->u.filter.fo_sptlrpc_rset);
init_rwsem(&m->ofd_lastid_rwsem);
obd->u.filter.fo_fl_oss_capa = 0;
- CFS_INIT_LIST_HEAD(&obd->u.filter.fo_capa_keys);
+ INIT_LIST_HEAD(&obd->u.filter.fo_capa_keys);
obd->u.filter.fo_capa_hash = init_capa_hash();
if (obd->u.filter.fo_capa_hash == NULL)
RETURN(-ENOMEM);
if (rc)
GOTO(err_fini_lut, rc);
+ rc = ofd_start_inconsistency_verification_thread(m);
+ if (rc != 0)
+ GOTO(err_fini_fs, rc);
+
RETURN(0);
+
+err_fini_fs:
+ ofd_fs_cleanup(env, m);
err_fini_lut:
tgt_fini(env, &m->ofd_lut);
err_free_ns:
return rc;
}
+/**
+ * Stop the OFD device
+ *
+ * This function stops the OFD device and all its subsystems.
+ * This is the end of OFD lifecycle.
+ *
+ * \param[in] env execution environment
+ * \param[in] m OFD device
+ */
static void ofd_fini(const struct lu_env *env, struct ofd_device *m)
{
- struct obd_device *obd = ofd_obd(m);
- struct lu_device *d = &m->ofd_dt_dev.dd_lu_dev;
+ struct obd_device *obd = ofd_obd(m);
+ struct lu_device *d = &m->ofd_dt_dev.dd_lu_dev;
+ struct lfsck_stop stop;
- lfsck_stop(env, m->ofd_osd, true);
- lfsck_degister(env, m->ofd_osd);
+ stop.ls_status = LS_PAUSED;
+ stop.ls_flags = 0;
+ lfsck_stop(env, m->ofd_osd, &stop);
target_recovery_fini(obd);
obd_exports_barrier(obd);
obd_zombie_barrier();
tgt_fini(env, &m->ofd_lut);
+ ofd_stop_inconsistency_verification_thread(m);
+ lfsck_degister(env, m->ofd_osd);
ofd_fs_cleanup(env, m);
ofd_free_capa_keys(m);
ofd_stack_fini(env, m, &m->ofd_dt_dev.dd_lu_dev);
ofd_procfs_fini(m);
LASSERT(atomic_read(&d->ld_ref) == 0);
- server_put_mount(obd->obd_name);
+ server_put_mount(obd->obd_name, true);
EXIT;
}
+/**
+ * Implementation of lu_device_type_operations::ldto_device_fini.
+ *
+ * Finalize device. Dual to ofd_device_init(). It is called from
+ * obd_precleanup() and stops the current device.
+ *
+ * \param[in] env execution environment
+ * \param[in] d LU device of OFD
+ *
+ * \retval NULL
+ */
static struct lu_device *ofd_device_fini(const struct lu_env *env,
struct lu_device *d)
{
RETURN(NULL);
}
+/**
+ * Implementation of lu_device_type_operations::ldto_device_free.
+ *
+ * Free OFD device. Dual to ofd_device_alloc().
+ *
+ * \param[in] env execution environment
+ * \param[in] d LU device of OFD
+ *
+ * \retval NULL
+ */
static struct lu_device *ofd_device_free(const struct lu_env *env,
struct lu_device *d)
{
RETURN(NULL);
}
+/**
+ * Implementation of lu_device_type_operations::ldto_device_alloc.
+ *
+ * This function allocates the new OFD device. It is called from
+ * obd_setup() if OBD device had lu_device_type defined.
+ *
+ * \param[in] env execution environment
+ * \param[in] t lu_device_type of OFD device
+ * \param[in] cfg configuration log
+ *
+ * \retval pointer to the lu_device of just allocated OFD
+ * \retval ERR_PTR of return value on error
+ */
static struct lu_device *ofd_device_alloc(const struct lu_env *env,
struct lu_device_type *t,
struct lustre_cfg *cfg)
return l;
}
-/* thread context key constructor/destructor */
-LU_KEY_INIT_FINI(ofd, struct ofd_thread_info);
-
-static void ofd_key_exit(const struct lu_context *ctx,
- struct lu_context_key *key, void *data)
-{
- struct ofd_thread_info *info = data;
-
- info->fti_env = NULL;
- info->fti_exp = NULL;
-
- info->fti_xid = 0;
- info->fti_pre_version = 0;
- info->fti_used = 0;
-
- memset(&info->fti_attr, 0, sizeof info->fti_attr);
-}
-
-struct lu_context_key ofd_thread_key = {
- .lct_tags = LCT_DT_THREAD,
- .lct_init = ofd_key_init,
- .lct_fini = ofd_key_fini,
- .lct_exit = ofd_key_exit
-};
-
-/* type constructor/destructor: mdt_type_init, mdt_type_fini */
+/* type constructor/destructor: ofd_type_init(), ofd_type_fini() */
LU_TYPE_INIT_FINI(ofd, &ofd_thread_key);
static struct lu_device_type_operations ofd_device_type_ops = {
.ldt_ctx_tags = LCT_DT_THREAD
};
-int __init ofd_init(void)
+/**
+ * Initialize OFD module.
+ *
+ * This function is called upon module loading. It registers OFD device type
+ * and prepares all in-memory structures used by all OFD devices.
+ *
+ * \retval 0 if successful
+ * \retval negative value on error
+ */
+static int __init ofd_init(void)
{
- struct lprocfs_static_vars lvars;
int rc;
rc = lu_kmem_init(ofd_caches);
return(rc);
}
- lprocfs_ofd_init_vars(&lvars);
-
- rc = class_register_type(&ofd_obd_ops, NULL, NULL,
-#ifndef HAVE_ONLY_PROCFS_SEQ
- lvars.module_vars,
-#endif
- LUSTRE_OST_NAME, &ofd_device_type);
+ rc = class_register_type(&ofd_obd_ops, NULL, true, NULL,
+ LUSTRE_OST_NAME, &ofd_device_type);
return rc;
}
-void __exit ofd_exit(void)
+/**
+ * Stop OFD module.
+ *
+ * This function is called upon OFD module unloading.
+ * It frees all related structures and unregisters OFD device type.
+ */
+static void __exit ofd_exit(void)
{
ofd_fmd_exit();
lu_kmem_fini(ofd_caches);