+#include "../ptlrpc/ptlrpc_internal.h"
+
+/* This must be longer than the longest string below */
+#define SYNC_STATES_MAXLEN 16
+static const char * const sync_lock_cancel_states[] = {
+ [SYNC_LOCK_CANCEL_NEVER] = "never",
+ [SYNC_LOCK_CANCEL_BLOCKING] = "blocking",
+ [SYNC_LOCK_CANCEL_ALWAYS] = "always",
+};
+
+/**
+ * Show policy for handling dirty data under a lock being cancelled.
+ *
+ * \param[in] kobj sysfs kobject
+ * \param[in] attr sysfs attribute
+ * \param[in] buf buffer for data
+ *
+ * \retval 0 and buffer filled with data on success
+ * \retval negative value on error
+ */
+ssize_t sync_lock_cancel_show(struct kobject *kobj,
+ struct attribute *attr, char *buf)
+{
+ struct obd_device *obd = container_of(kobj, struct obd_device,
+ obd_kset.kobj);
+ struct lu_target *tgt = obd->u.obt.obt_lut;
+
+ return sprintf(buf, "%s\n",
+ sync_lock_cancel_states[tgt->lut_sync_lock_cancel]);
+}
+EXPORT_SYMBOL(sync_lock_cancel_show);
+
+/**
+ * Change policy for handling dirty data under a lock being cancelled.
+ *
+ * This variable defines what action target takes upon lock cancel
+ * There are three possible modes:
+ * 1) never - never do sync upon lock cancel. This can lead to data
+ * inconsistencies if both the OST and client crash while writing a file
+ * that is also concurrently being read by another client. In these cases,
+ * this may allow the file data to "rewind" to an earlier state.
+ * 2) blocking - do sync only if there is blocking lock, e.g. if another
+ * client is trying to access this same object
+ * 3) always - do sync always
+ *
+ * \param[in] kobj kobject
+ * \param[in] attr attribute to show
+ * \param[in] buf buffer for data
+ * \param[in] count buffer size
+ *
+ * \retval \a count on success
+ * \retval negative value on error
+ */
+ssize_t sync_lock_cancel_store(struct kobject *kobj, struct attribute *attr,
+ const char *buffer, size_t count)
+{
+ struct obd_device *obd = container_of(kobj, struct obd_device,
+ obd_kset.kobj);
+ struct lu_target *tgt = obd->u.obt.obt_lut;
+ int val = -1;
+ enum tgt_sync_lock_cancel slc;
+
+ if (count == 0 || count >= SYNC_STATES_MAXLEN)
+ return -EINVAL;
+
+ for (slc = 0; slc < ARRAY_SIZE(sync_lock_cancel_states); slc++) {
+ if (strcmp(buffer, sync_lock_cancel_states[slc]) == 0) {
+ val = slc;
+ break;
+ }
+ }
+
+ /* Legacy numeric codes */
+ if (val == -1) {
+ int rc = kstrtoint(buffer, 0, &val);
+ if (rc)
+ return rc;
+ }
+
+ if (val < 0 || val > 2)
+ return -EINVAL;
+
+ spin_lock(&tgt->lut_flags_lock);
+ tgt->lut_sync_lock_cancel = val;
+ spin_unlock(&tgt->lut_flags_lock);
+ return count;
+}
+EXPORT_SYMBOL(sync_lock_cancel_store);
+LUSTRE_RW_ATTR(sync_lock_cancel);
+
+/**
+ * Show maximum number of Filter Modification Data (FMD) maintained.
+ *
+ * \param[in] kobj kobject
+ * \param[in] attr attribute to show
+ * \param[in] buf buffer for data
+ *
+ * \retval 0 and buffer filled with data on success
+ * \retval negative value on error
+ */
+ssize_t tgt_fmd_count_show(struct kobject *kobj, struct attribute *attr,
+ char *buf)
+{
+ struct obd_device *obd = container_of(kobj, struct obd_device,
+ obd_kset.kobj);
+ struct lu_target *lut = obd->u.obt.obt_lut;
+
+ return sprintf(buf, "%u\n", lut->lut_fmd_max_num);
+}
+
+/**
+ * Change number of FMDs maintained by target.
+ *
+ * This defines how large the list of FMDs can be.
+ *
+ * \param[in] kobj kobject
+ * \param[in] attr attribute to show
+ * \param[in] buf buffer for data
+ * \param[in] count buffer size
+ *
+ * \retval \a count on success
+ * \retval negative value on error
+ */
+ssize_t tgt_fmd_count_store(struct kobject *kobj, struct attribute *attr,
+ const char *buffer, size_t count)
+{
+ struct obd_device *obd = container_of(kobj, struct obd_device,
+ obd_kset.kobj);
+ struct lu_target *lut = obd->u.obt.obt_lut;
+ int val, rc;
+
+ rc = kstrtoint(buffer, 0, &val);
+ if (rc)
+ return rc;
+
+ if (val < 1 || val > 65536)
+ return -EINVAL;
+
+ lut->lut_fmd_max_num = val;
+
+ return count;
+}
+LUSTRE_RW_ATTR(tgt_fmd_count);
+
+/**
+ * Show the maximum age of FMD data in seconds.
+ *
+ * \param[in] kobj kobject
+ * \param[in] attr attribute to show
+ * \param[in] buf buffer for data
+ *
+ * \retval 0 and buffer filled with data on success
+ * \retval negative value on error
+ */
+ssize_t tgt_fmd_seconds_show(struct kobject *kobj, struct attribute *attr,
+ char *buf)
+{
+ struct obd_device *obd = container_of(kobj, struct obd_device,
+ obd_kset.kobj);
+ struct lu_target *lut = obd->u.obt.obt_lut;
+
+ return sprintf(buf, "%lld\n", lut->lut_fmd_max_age);
+}
+
+/**
+ * Set the maximum age of FMD data in seconds.
+ *
+ * This defines how long FMD data stays in the FMD list.
+ *
+ * \param[in] kobj kobject
+ * \param[in] attr attribute to show
+ * \param[in] buf buffer for data
+ * \param[in] count buffer size
+ *
+ * \retval \a count on success
+ * \retval negative number on error
+ */
+ssize_t tgt_fmd_seconds_store(struct kobject *kobj, struct attribute *attr,
+ const char *buffer, size_t count)
+{
+ struct obd_device *obd = container_of(kobj, struct obd_device,
+ obd_kset.kobj);
+ struct lu_target *lut = obd->u.obt.obt_lut;
+ time64_t val;
+ int rc;
+
+ rc = kstrtoll(buffer, 0, &val);
+ if (rc)
+ return rc;
+
+ if (val < 1 || val > 65536) /* ~ 18 hour max */
+ return -EINVAL;
+
+ lut->lut_fmd_max_age = val;
+
+ return count;
+}
+LUSTRE_RW_ATTR(tgt_fmd_seconds);
+
+/* These two aliases are old names and kept for compatibility, they were
+ * changed to 'tgt_fmd_count' and 'tgt_fmd_seconds'.
+ * This change was made in Lustre 2.13, so these aliases can be removed
+ * when back compatibility is not needed with any Lustre version prior 2.13
+ */
+static struct lustre_attr tgt_fmd_count_compat = __ATTR(client_cache_count,
+ 0644, tgt_fmd_count_show, tgt_fmd_count_store);
+static struct lustre_attr tgt_fmd_seconds_compat = __ATTR(client_cache_seconds,
+ 0644, tgt_fmd_seconds_show, tgt_fmd_seconds_store);
+
+static const struct attribute *tgt_attrs[] = {
+ &lustre_attr_sync_lock_cancel.attr,
+ &lustre_attr_tgt_fmd_count.attr,
+ &lustre_attr_tgt_fmd_seconds.attr,
+ &tgt_fmd_count_compat.attr,
+ &tgt_fmd_seconds_compat.attr,
+ NULL,
+};
+
+/**
+ * Decide which checksums both client and OST support, possibly forcing
+ * the use of T10PI checksums if the hardware supports this.
+ *
+ * The clients that have no T10-PI RPC checksum support will use the same
+ * mechanism to select checksum type as before, and will not be affected by
+ * the following logic.
+ *
+ * For the clients that have T10-PI RPC checksum support:
+ *
+ * If the target supports T10-PI feature and T10-PI checksum is enforced,
+ * clients will have no other choice for RPC checksum type other than using
+ * the T10PI checksum type. This is useful for enforcing end-to-end integrity
+ * in the whole system.
+ *
+ * If the target doesn't support T10-PI feature and T10-PI checksum is
+ * enforced, together with other checksum with reasonably good speeds (e.g.
+ * crc32, crc32c, adler, etc.), all T10-PI checksum types understood by the
+ * client (t10ip512, t10ip4K, t10crc512, t10crc4K) will be added to the
+ * available checksum types, regardless of the speeds of T10-PI checksums.
+ * This is useful for testing T10-PI checksum of RPC.
+ *
+ * If the target supports T10-PI feature and T10-PI checksum is NOT enforced,
+ * the corresponding T10-PI checksum type will be added to the checksum type
+ * list, regardless of the speed of the T10-PI checksum. This provides clients
+ * the flexibility to choose whether to enable end-to-end integrity or not.
+ *
+ * If the target does NOT supports T10-PI feature and T10-PI checksum is NOT
+ * enforced, together with other checksums with reasonably good speeds,
+ * all the T10-PI checksum types with good speeds will be added into the
+ * checksum type list. Note that a T10-PI checksum type with a speed worse
+ * than half of Alder will NOT be added as a option. In this circumstance,
+ * T10-PI checksum types has the same behavior like other normal checksum
+ * types.
+ */
+void tgt_mask_cksum_types(struct lu_target *lut, enum cksum_types *cksum_types)
+{
+ bool enforce = lut->lut_cksum_t10pi_enforce;
+ enum cksum_types tgt_t10_cksum_type;
+ enum cksum_types client_t10_types = *cksum_types & OBD_CKSUM_T10_ALL;
+ enum cksum_types server_t10_types;
+
+ /*
+ * The client set in ocd_cksum_types the checksum types it
+ * supports. We have to mask off the algorithms that we don't
+ * support. T10PI checksum types will be added later.
+ */
+ *cksum_types &= (lut->lut_cksum_types_supported & ~OBD_CKSUM_T10_ALL);
+ server_t10_types = lut->lut_cksum_types_supported & OBD_CKSUM_T10_ALL;
+ tgt_t10_cksum_type = lut->lut_dt_conf.ddp_t10_cksum_type;
+
+ /* Quick exit if no T10-PI support on client */
+ if (!client_t10_types)
+ return;
+
+ /*
+ * This OST has NO T10-PI feature. Add all supported T10-PI checksums
+ * as options if T10-PI checksum is enforced. If the T10-PI checksum is
+ * not enforced, only add them as options when speed is good.
+ */
+ if (tgt_t10_cksum_type == 0) {
+ /*
+ * Server allows all T10PI checksums, and server_t10_types
+ * include quick ones.
+ */
+ if (enforce)
+ *cksum_types |= client_t10_types;
+ else
+ *cksum_types |= client_t10_types & server_t10_types;
+ return;
+ }
+
+ /*
+ * This OST has T10-PI feature. Disable all other checksum types if
+ * T10-PI checksum is enforced. If the T10-PI checksum is not enforced,
+ * add the checksum type as an option.
+ */
+ if (client_t10_types & tgt_t10_cksum_type) {
+ if (enforce)
+ *cksum_types = tgt_t10_cksum_type;
+ else
+ *cksum_types |= tgt_t10_cksum_type;
+ }
+}
+EXPORT_SYMBOL(tgt_mask_cksum_types);
+
+int tgt_tunables_init(struct lu_target *lut)
+{
+ int rc;
+
+ rc = sysfs_create_files(&lut->lut_obd->obd_kset.kobj, tgt_attrs);
+ if (!rc)
+ lut->lut_attrs = tgt_attrs;
+ return rc;
+}
+EXPORT_SYMBOL(tgt_tunables_init);
+
+void tgt_tunables_fini(struct lu_target *lut)
+{
+ if (lut->lut_attrs) {
+ sysfs_remove_files(&lut->lut_obd->obd_kset.kobj,
+ lut->lut_attrs);
+ lut->lut_attrs = NULL;
+ }
+}
+EXPORT_SYMBOL(tgt_tunables_fini);
+
+/*
+ * Save cross-MDT lock in lut_slc_locks.
+ *
+ * Lock R/W count is not saved, but released in unlock (not canceled remotely),
+ * instead only a refcount is taken, so that the remote MDT where the object
+ * resides can detect conflict with this lock there.
+ *
+ * \param lut target
+ * \param lock cross-MDT lock to save
+ * \param transno when the transaction with this transno is committed, this lock
+ * can be canceled.
+ */
+void tgt_save_slc_lock(struct lu_target *lut, struct ldlm_lock *lock,
+ __u64 transno)
+{
+ spin_lock(&lut->lut_slc_locks_guard);
+ lock_res_and_lock(lock);
+ if (ldlm_is_cbpending(lock)) {
+ /* if it was canceld by server, don't save, because remote MDT
+ * will do Sync-on-Cancel. */
+ LDLM_LOCK_PUT(lock);
+ } else {
+ lock->l_transno = transno;
+ /* if this lock is in the list already, there are two operations
+ * both use this lock, and save it after use, so for the second
+ * one, just put the refcount. */
+ if (list_empty(&lock->l_slc_link))
+ list_add_tail(&lock->l_slc_link, &lut->lut_slc_locks);
+ else
+ LDLM_LOCK_PUT(lock);
+ }
+ unlock_res_and_lock(lock);
+ spin_unlock(&lut->lut_slc_locks_guard);
+}
+EXPORT_SYMBOL(tgt_save_slc_lock);
+
+/*
+ * Discard cross-MDT lock from lut_slc_locks.
+ *
+ * This is called upon BAST, just remove lock from lut_slc_locks and put lock
+ * refcount. The BAST will cancel this lock.
+ *
+ * \param lut target
+ * \param lock cross-MDT lock to discard
+ */
+void tgt_discard_slc_lock(struct lu_target *lut, struct ldlm_lock *lock)
+{
+ spin_lock(&lut->lut_slc_locks_guard);
+ lock_res_and_lock(lock);
+ /* may race with tgt_cancel_slc_locks() */
+ if (lock->l_transno != 0) {
+ LASSERT(!list_empty(&lock->l_slc_link));
+ LASSERT(ldlm_is_cbpending(lock));
+ list_del_init(&lock->l_slc_link);
+ lock->l_transno = 0;
+ LDLM_LOCK_PUT(lock);
+ }
+ unlock_res_and_lock(lock);
+ spin_unlock(&lut->lut_slc_locks_guard);
+}
+EXPORT_SYMBOL(tgt_discard_slc_lock);
+
+/*
+ * Cancel cross-MDT locks upon transaction commit.
+ *
+ * Remove cross-MDT locks from lut_slc_locks, cancel them and put lock refcount.
+ *
+ * \param lut target
+ * \param transno transaction with this number was committed.
+ */
+void tgt_cancel_slc_locks(struct lu_target *lut, __u64 transno)
+{
+ struct ldlm_lock *lock, *next;
+ LIST_HEAD(list);
+ struct lustre_handle lockh;
+
+ spin_lock(&lut->lut_slc_locks_guard);
+ list_for_each_entry_safe(lock, next, &lut->lut_slc_locks,
+ l_slc_link) {
+ lock_res_and_lock(lock);
+ LASSERT(lock->l_transno != 0);
+ if (lock->l_transno > transno) {
+ unlock_res_and_lock(lock);
+ continue;
+ }
+ /* ouch, another operation is using it after it's saved */
+ if (lock->l_readers != 0 || lock->l_writers != 0) {
+ unlock_res_and_lock(lock);
+ continue;
+ }
+ /* set CBPENDING so that this lock won't be used again */
+ ldlm_set_cbpending(lock);
+ lock->l_transno = 0;
+ list_move(&lock->l_slc_link, &list);
+ unlock_res_and_lock(lock);
+ }
+ spin_unlock(&lut->lut_slc_locks_guard);
+
+ list_for_each_entry_safe(lock, next, &list, l_slc_link) {
+ list_del_init(&lock->l_slc_link);
+ ldlm_lock2handle(lock, &lockh);
+ ldlm_cli_cancel(&lockh, LCF_ASYNC);
+ LDLM_LOCK_PUT(lock);
+ }
+}