c/mirror_zfs
1
0
Fork 0
mirror of https://git.proxmox.com/git/mirror_zfs.git synced 2024-12-27 03:19:35 +03:00
Code Issues Packages Projects Releases Wiki Activity

assertion in arc_release() during encrypted receive

Browse Source 
In the existing code, when doing a raw (encrypted) zfs receive, 
we call arc_convert_to_raw() from open context. This creates a 
race condition between arc_release()/arc_change_state() and 
writing out the block from syncing context (arc_write_ready/done()).

This change makes it so that when we are doing a raw (encrypted) 
zfs receive, we save the crypt parameters (salt, iv, mac) of dnode 
blocks in the dbuf_dirty_record_t, and call arc_convert_to_raw() 
from syncing context when writing out the block of dnodes.

Additionally, we can eliminate dr_raw and associated setters, and 
instead know that dnode blocks are always raw when doing a zfs 
receive (see the new field os_raw_receive).

Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov>
Reviewed-by: Tom Caputi <tcaputi@datto.com>
Signed-off-by: Matthew Ahrens <mahrens@delphix.com>
Closes #7424 
Closes #7429
This commit is contained in:
Matthew Ahrens 2018-04-17 11:06:54 -07:00 committed by Brian Behlendorf
parent b40d45bc6c
commit 0c03d21ac9
7 changed files with 105 additions and 218 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

11
include/sys/dbuf.h
View File

@ -147,7 +147,16 @@ typedef struct dbuf_dirty_record {
override_states_t dr_override_state; override_states_t dr_override_state;
uint8_t dr_copies; uint8_t dr_copies;
boolean_t dr_nopwrite; boolean_t dr_nopwrite;
boolean_t dr_raw; boolean_t dr_has_raw_params;
/*
* If dr_has_raw_params is set, the following crypt
* params will be set on the BP that's written.
*/
boolean_t dr_byteorder;
uint8_t dr_salt[ZIO_DATA_SALT_LEN];
uint8_t dr_iv[ZIO_DATA_IV_LEN];
uint8_t dr_mac[ZIO_DATA_MAC_LEN];
} dl; } dl;
} dt; } dt;
} dbuf_dirty_record_t; } dbuf_dirty_record_t;

10
include/sys/dmu.h
View File

@ -472,7 +472,6 @@ void dmu_object_set_checksum(objset_t *os, uint64_t object, uint8_t checksum,
void dmu_object_set_compress(objset_t *os, uint64_t object, uint8_t compress, void dmu_object_set_compress(objset_t *os, uint64_t object, uint8_t compress,
dmu_tx_t *tx); dmu_tx_t *tx);
int dmu_object_dirty_raw(objset_t *os, uint64_t object, dmu_tx_t *tx);
int dmu_object_remap_indirects(objset_t *os, uint64_t object, uint64_t txg); int dmu_object_remap_indirects(objset_t *os, uint64_t object, uint64_t txg);
@ -715,7 +714,8 @@ struct blkptr *dmu_buf_get_blkptr(dmu_buf_t *db);
* (ie. you've called dmu_tx_hold_object(tx, db->db_object)). * (ie. you've called dmu_tx_hold_object(tx, db->db_object)).
*/ */
void dmu_buf_will_dirty(dmu_buf_t *db, dmu_tx_t *tx); void dmu_buf_will_dirty(dmu_buf_t *db, dmu_tx_t *tx);
void dmu_buf_will_change_crypt_params(dmu_buf_t *db, dmu_tx_t *tx); void dmu_buf_set_crypt_params(dmu_buf_t *db_fake, boolean_t byteorder,
const uint8_t *salt, const uint8_t *iv, const uint8_t *mac, dmu_tx_t *tx);
/* /*
* You must create a transaction, then hold the objects which you will * You must create a transaction, then hold the objects which you will
@ -794,10 +794,7 @@ int dmu_free_range(objset_t *os, uint64_t object, uint64_t offset,
uint64_t size, dmu_tx_t *tx); uint64_t size, dmu_tx_t *tx);
int dmu_free_long_range(objset_t *os, uint64_t object, uint64_t offset, int dmu_free_long_range(objset_t *os, uint64_t object, uint64_t offset,
uint64_t size); uint64_t size);
int dmu_free_long_range_raw(objset_t *os, uint64_t object, uint64_t offset,
uint64_t size);
int dmu_free_long_object(objset_t *os, uint64_t object); int dmu_free_long_object(objset_t *os, uint64_t object);
int dmu_free_long_object_raw(objset_t *os, uint64_t object);
/* /*
* Convenience functions. * Convenience functions.
@ -837,9 +834,6 @@ void dmu_assign_arcbuf_by_dnode(dnode_t *dn, uint64_t offset,
void dmu_assign_arcbuf_by_dbuf(dmu_buf_t *handle, uint64_t offset, void dmu_assign_arcbuf_by_dbuf(dmu_buf_t *handle, uint64_t offset,
struct arc_buf *buf, dmu_tx_t *tx); struct arc_buf *buf, dmu_tx_t *tx);
#define dmu_assign_arcbuf dmu_assign_arcbuf_by_dbuf #define dmu_assign_arcbuf dmu_assign_arcbuf_by_dbuf
int dmu_convert_mdn_block_to_raw(objset_t *os, uint64_t firstobj,
boolean_t byteorder, const uint8_t *salt, const uint8_t *iv,
const uint8_t *mac, dmu_tx_t *tx);
void dmu_copy_from_buf(objset_t *os, uint64_t object, uint64_t offset, void dmu_copy_from_buf(objset_t *os, uint64_t object, uint64_t offset,
dmu_buf_t *handle, dmu_tx_t *tx); dmu_buf_t *handle, dmu_tx_t *tx);
#ifdef HAVE_UIO_ZEROCOPY #ifdef HAVE_UIO_ZEROCOPY

1
include/sys/dmu_objset.h
View File

@ -139,6 +139,7 @@ struct objset {
uint64_t os_flags; uint64_t os_flags;
uint64_t os_freed_dnodes; uint64_t os_freed_dnodes;
boolean_t os_rescan_dnodes; boolean_t os_rescan_dnodes;
boolean_t os_raw_receive;
/* os_phys_buf should be written raw next txg */ /* os_phys_buf should be written raw next txg */
boolean_t os_next_write_raw[TXG_SIZE]; boolean_t os_next_write_raw[TXG_SIZE];

59
module/zfs/dbuf.c
View File

@ -1538,7 +1538,7 @@ dbuf_unoverride(dbuf_dirty_record_t *dr)
dr->dt.dl.dr_override_state = DR_NOT_OVERRIDDEN; dr->dt.dl.dr_override_state = DR_NOT_OVERRIDDEN;
dr->dt.dl.dr_nopwrite = B_FALSE; dr->dt.dl.dr_nopwrite = B_FALSE;
dr->dt.dl.dr_raw = B_FALSE; dr->dt.dl.dr_has_raw_params = B_FALSE;
/* /*
* Release the already-written buffer, so we leave it in * Release the already-written buffer, so we leave it in
@ -2211,15 +2211,26 @@ dmu_buf_will_fill(dmu_buf_t *db_fake, dmu_tx_t *tx)
/* /*
* This function is effectively the same as dmu_buf_will_dirty(), but * This function is effectively the same as dmu_buf_will_dirty(), but
* indicates the caller expects raw encrypted data in the db. It will * indicates the caller expects raw encrypted data in the db, and provides
* also set the raw flag on the created dirty record. * the crypt params (byteorder, salt, iv, mac) which should be stored in the
* blkptr_t when this dbuf is written. This is only used for blocks of
* dnodes, during raw receive.
*/ */
void void
dmu_buf_will_change_crypt_params(dmu_buf_t *db_fake, dmu_tx_t *tx) dmu_buf_set_crypt_params(dmu_buf_t *db_fake, boolean_t byteorder,
const uint8_t *salt, const uint8_t *iv, const uint8_t *mac, dmu_tx_t *tx)
{ {
dmu_buf_impl_t *db = (dmu_buf_impl_t *)db_fake; dmu_buf_impl_t *db = (dmu_buf_impl_t *)db_fake;
dbuf_dirty_record_t *dr; dbuf_dirty_record_t *dr;
/*
* dr_has_raw_params is only processed for blocks of dnodes
* (see dbuf_sync_dnode_leaf_crypt()).
*/
ASSERT3U(db->db.db_object, ==, DMU_META_DNODE_OBJECT);
ASSERT3U(db->db_level, ==, 0);
ASSERT(db->db_objset->os_raw_receive);
dmu_buf_will_dirty_impl(db_fake, dmu_buf_will_dirty_impl(db_fake,
DB_RF_MUST_SUCCEED | DB_RF_NOPREFETCH | DB_RF_NO_DECRYPT, tx); DB_RF_MUST_SUCCEED | DB_RF_NOPREFETCH | DB_RF_NO_DECRYPT, tx);
@ -2229,8 +2240,12 @@ dmu_buf_will_change_crypt_params(dmu_buf_t *db_fake, dmu_tx_t *tx)
ASSERT3P(dr, !=, NULL); ASSERT3P(dr, !=, NULL);
ASSERT3U(dr->dr_txg, ==, tx->tx_txg); ASSERT3U(dr->dr_txg, ==, tx->tx_txg);
dr->dt.dl.dr_raw = B_TRUE;
db->db_objset->os_next_write_raw[tx->tx_txg & TXG_MASK] = B_TRUE; dr->dt.dl.dr_has_raw_params = B_TRUE;
dr->dt.dl.dr_byteorder = byteorder;
bcopy(salt, dr->dt.dl.dr_salt, ZIO_DATA_SALT_LEN);
bcopy(iv, dr->dt.dl.dr_iv, ZIO_DATA_IV_LEN);
bcopy(mac, dr->dt.dl.dr_mac, ZIO_DATA_MAC_LEN);
} }
#pragma weak dmu_buf_fill_done = dbuf_fill_done #pragma weak dmu_buf_fill_done = dbuf_fill_done
@ -2341,7 +2356,6 @@ dbuf_assign_arcbuf(dmu_buf_impl_t *db, arc_buf_t *buf, dmu_tx_t *tx)
ASSERT(db->db_buf != NULL); ASSERT(db->db_buf != NULL);
if (dr != NULL && dr->dr_txg == tx->tx_txg) { if (dr != NULL && dr->dr_txg == tx->tx_txg) {
ASSERT(dr->dt.dl.dr_data == db->db_buf); ASSERT(dr->dt.dl.dr_data == db->db_buf);
IMPLY(arc_is_encrypted(buf), dr->dt.dl.dr_raw);
if (!arc_released(db->db_buf)) { if (!arc_released(db->db_buf)) {
ASSERT(dr->dt.dl.dr_override_state == ASSERT(dr->dt.dl.dr_override_state ==
@ -3452,20 +3466,23 @@ dbuf_check_blkptr(dnode_t *dn, dmu_buf_impl_t *db)
} }
/* /*
* Ensure the dbuf's data is untransformed if the associated dirty * When syncing out a blocks of dnodes, adjust the block to deal with
* record requires it. This is used by dbuf_sync_leaf() to ensure * encryption. Normally, we make sure the block is decrypted before writing
* that a dnode block is decrypted before we write new data to it. * it. If we have crypt params, then we are writing a raw (encrypted) block,
* For raw writes we assert that the buffer is already encrypted. * from a raw receive. In this case, set the ARC buf's crypt params so
* that the BP will be filled with the correct byteorder, salt, iv, and mac.
*/ */
static void static void
dbuf_check_crypt(dbuf_dirty_record_t *dr) dbuf_prepare_encrypted_dnode_leaf(dbuf_dirty_record_t *dr)
{ {
int err; int err;
dmu_buf_impl_t *db = dr->dr_dbuf; dmu_buf_impl_t *db = dr->dr_dbuf;
ASSERT(MUTEX_HELD(&db->db_mtx)); ASSERT(MUTEX_HELD(&db->db_mtx));
ASSERT3U(db->db.db_object, ==, DMU_META_DNODE_OBJECT);
ASSERT3U(db->db_level, ==, 0);
if (!dr->dt.dl.dr_raw && arc_is_encrypted(db->db_buf)) { if (!db->db_objset->os_raw_receive && arc_is_encrypted(db->db_buf)) {
zbookmark_phys_t zb; zbookmark_phys_t zb;
/* /*
@ -3481,12 +3498,12 @@ dbuf_check_crypt(dbuf_dirty_record_t *dr)
&zb, B_TRUE); &zb, B_TRUE);
if (err) if (err)
panic("Invalid dnode block MAC"); panic("Invalid dnode block MAC");
} else if (dr->dt.dl.dr_raw) { } else if (dr->dt.dl.dr_has_raw_params) {
/* (void) arc_release(dr->dt.dl.dr_data, db);
* Writing raw encrypted data requires the db's arc buffer arc_convert_to_raw(dr->dt.dl.dr_data,
* to be converted to raw by the caller. dmu_objset_id(db->db_objset),
*/ dr->dt.dl.dr_byteorder, DMU_OT_DNODE,
ASSERT(arc_is_encrypted(db->db_buf)); dr->dt.dl.dr_salt, dr->dt.dl.dr_iv, dr->dt.dl.dr_mac);
} }
} }
@ -3667,7 +3684,7 @@ dbuf_sync_leaf(dbuf_dirty_record_t *dr, dmu_tx_t *tx)
* or decrypted, depending on what we are writing to it this txg. * or decrypted, depending on what we are writing to it this txg.
*/ */
if (os->os_encrypted && dn->dn_object == DMU_META_DNODE_OBJECT) if (os->os_encrypted && dn->dn_object == DMU_META_DNODE_OBJECT)
dbuf_check_crypt(dr); dbuf_prepare_encrypted_dnode_leaf(dr);
if (db->db_state != DB_NOFILL && if (db->db_state != DB_NOFILL &&
dn->dn_object != DMU_META_DNODE_OBJECT && dn->dn_object != DMU_META_DNODE_OBJECT &&
@ -4336,7 +4353,7 @@ EXPORT_SYMBOL(dbuf_free_range);
EXPORT_SYMBOL(dbuf_new_size); EXPORT_SYMBOL(dbuf_new_size);
EXPORT_SYMBOL(dbuf_release_bp); EXPORT_SYMBOL(dbuf_release_bp);
EXPORT_SYMBOL(dbuf_dirty); EXPORT_SYMBOL(dbuf_dirty);
EXPORT_SYMBOL(dmu_buf_will_change_crypt_params); EXPORT_SYMBOL(dmu_buf_set_crypt_params);
EXPORT_SYMBOL(dmu_buf_will_dirty); EXPORT_SYMBOL(dmu_buf_will_dirty);
EXPORT_SYMBOL(dmu_buf_will_not_fill); EXPORT_SYMBOL(dmu_buf_will_not_fill);
EXPORT_SYMBOL(dmu_buf_will_fill); EXPORT_SYMBOL(dmu_buf_will_fill);

119
module/zfs/dmu.c
View File

@ -765,7 +765,7 @@ dmu_objset_zfs_unmounting(objset_t *os)
static int static int
dmu_free_long_range_impl(objset_t *os, dnode_t *dn, uint64_t offset, dmu_free_long_range_impl(objset_t *os, dnode_t *dn, uint64_t offset,
uint64_t length, boolean_t raw) uint64_t length)
{ {
uint64_t object_size; uint64_t object_size;
int err; int err;
@ -848,19 +848,6 @@ dmu_free_long_range_impl(objset_t *os, dnode_t *dn, uint64_t offset,
uint64_t, dmu_tx_get_txg(tx)); uint64_t, dmu_tx_get_txg(tx));
dnode_free_range(dn, chunk_begin, chunk_len, tx); dnode_free_range(dn, chunk_begin, chunk_len, tx);
/* if this is a raw free, mark the dirty record as such */
if (raw) {
dbuf_dirty_record_t *dr = dn->dn_dbuf->db_last_dirty;
while (dr != NULL && dr->dr_txg > tx->tx_txg)
dr = dr->dr_next;
if (dr != NULL && dr->dr_txg == tx->tx_txg) {
dr->dt.dl.dr_raw = B_TRUE;
dn->dn_objset->os_next_write_raw
[tx->tx_txg & TXG_MASK] = B_TRUE;
}
}
dmu_tx_commit(tx); dmu_tx_commit(tx);
length -= chunk_len; length -= chunk_len;
@ -878,7 +865,7 @@ dmu_free_long_range(objset_t *os, uint64_t object,
err = dnode_hold(os, object, FTAG, &dn); err = dnode_hold(os, object, FTAG, &dn);
if (err != 0) if (err != 0)
return (err); return (err);
err = dmu_free_long_range_impl(os, dn, offset, length, B_FALSE); err = dmu_free_long_range_impl(os, dn, offset, length);
/* /*
* It is important to zero out the maxblkid when freeing the entire * It is important to zero out the maxblkid when freeing the entire
@ -893,37 +880,8 @@ dmu_free_long_range(objset_t *os, uint64_t object,
return (err); return (err);
} }
/*
* This function is equivalent to dmu_free_long_range(), but also
* marks the new dirty record as a raw write.
*/
int int
dmu_free_long_range_raw(objset_t *os, uint64_t object, dmu_free_long_object(objset_t *os, uint64_t object)
uint64_t offset, uint64_t length)
{
dnode_t *dn;
int err;
err = dnode_hold(os, object, FTAG, &dn);
if (err != 0)
return (err);
err = dmu_free_long_range_impl(os, dn, offset, length, B_TRUE);
/*
* It is important to zero out the maxblkid when freeing the entire
* file, so that (a) subsequent calls to dmu_free_long_range_impl()
* will take the fast path, and (b) dnode_reallocate() can verify
* that the entire file has been freed.
*/
if (err == 0 && offset == 0 && length == DMU_OBJECT_END)
dn->dn_maxblkid = 0;
dnode_rele(dn, FTAG);
return (err);
}
static int
dmu_free_long_object_impl(objset_t *os, uint64_t object, boolean_t raw)
{ {
dmu_tx_t *tx; dmu_tx_t *tx;
int err; int err;
@ -938,8 +896,6 @@ dmu_free_long_object_impl(objset_t *os, uint64_t object, boolean_t raw)
dmu_tx_mark_netfree(tx); dmu_tx_mark_netfree(tx);
err = dmu_tx_assign(tx, TXG_WAIT); err = dmu_tx_assign(tx, TXG_WAIT);
if (err == 0) { if (err == 0) {
if (raw)
err = dmu_object_dirty_raw(os, object, tx);
if (err == 0) if (err == 0)
err = dmu_object_free(os, object, tx); err = dmu_object_free(os, object, tx);
@ -951,19 +907,6 @@ dmu_free_long_object_impl(objset_t *os, uint64_t object, boolean_t raw)
return (err); return (err);
} }
int
dmu_free_long_object(objset_t *os, uint64_t object)
{
return (dmu_free_long_object_impl(os, object, B_FALSE));
}
int
dmu_free_long_object_raw(objset_t *os, uint64_t object)
{
return (dmu_free_long_object_impl(os, object, B_TRUE));
}
int int
dmu_free_range(objset_t *os, uint64_t object, uint64_t offset, dmu_free_range(objset_t *os, uint64_t object, uint64_t offset,
uint64_t size, dmu_tx_t *tx) uint64_t size, dmu_tx_t *tx)
@ -1666,41 +1609,6 @@ dmu_return_arcbuf(arc_buf_t *buf)
arc_buf_destroy(buf, FTAG); arc_buf_destroy(buf, FTAG);
} }
int
dmu_convert_mdn_block_to_raw(objset_t *os, uint64_t firstobj,
boolean_t byteorder, const uint8_t *salt, const uint8_t *iv,
const uint8_t *mac, dmu_tx_t *tx)
{
int ret;
dmu_buf_t *handle = NULL;
dmu_buf_impl_t *db = NULL;
uint64_t offset = firstobj * DNODE_MIN_SIZE;
uint64_t dsobj = dmu_objset_id(os);
ret = dmu_buf_hold_by_dnode(DMU_META_DNODE(os), offset, FTAG, &handle,
DMU_READ_PREFETCH | DMU_READ_NO_DECRYPT);
if (ret != 0)
return (ret);
dmu_buf_will_change_crypt_params(handle, tx);
db = (dmu_buf_impl_t *)handle;
ASSERT3P(db->db_buf, !=, NULL);
ASSERT3U(dsobj, !=, 0);
/*
* This technically violates the assumption the dmu code makes
* that dnode blocks are only released in syncing context.
*/
(void) arc_release(db->db_buf, db);
arc_convert_to_raw(db->db_buf, dsobj, byteorder, DMU_OT_DNODE,
salt, iv, mac);
dmu_buf_rele(handle, FTAG);
return (0);
}
void void
dmu_copy_from_buf(objset_t *os, uint64_t object, uint64_t offset, dmu_copy_from_buf(objset_t *os, uint64_t object, uint64_t offset,
dmu_buf_t *handle, dmu_tx_t *tx) dmu_buf_t *handle, dmu_tx_t *tx)
@ -2222,25 +2130,6 @@ dmu_object_set_compress(objset_t *os, uint64_t object, uint8_t compress,
dnode_rele(dn, FTAG); dnode_rele(dn, FTAG);
} }
/*
* Dirty an object and set the dirty record's raw flag. This is used
* when writing raw data to an object that will not effect the
* encryption parameters, specifically during raw receives.
*/
int
dmu_object_dirty_raw(objset_t *os, uint64_t object, dmu_tx_t *tx)
{
dnode_t *dn;
int err;
err = dnode_hold(os, object, FTAG, &dn);
if (err)
return (err);
dmu_buf_will_change_crypt_params((dmu_buf_t *)dn->dn_dbuf, tx);
dnode_rele(dn, FTAG);
return (err);
}
/* /*
* When the "redundant_metadata" property is set to "most", only indirect * When the "redundant_metadata" property is set to "most", only indirect
* blocks of this level and higher will have an additional ditto block. * blocks of this level and higher will have an additional ditto block.
@ -2623,9 +2512,7 @@ EXPORT_SYMBOL(dmu_buf_rele_array);
EXPORT_SYMBOL(dmu_prefetch); EXPORT_SYMBOL(dmu_prefetch);
EXPORT_SYMBOL(dmu_free_range); EXPORT_SYMBOL(dmu_free_range);
EXPORT_SYMBOL(dmu_free_long_range); EXPORT_SYMBOL(dmu_free_long_range);
EXPORT_SYMBOL(dmu_free_long_range_raw);
EXPORT_SYMBOL(dmu_free_long_object); EXPORT_SYMBOL(dmu_free_long_object);
EXPORT_SYMBOL(dmu_free_long_object_raw);
EXPORT_SYMBOL(dmu_read); EXPORT_SYMBOL(dmu_read);
EXPORT_SYMBOL(dmu_read_by_dnode); EXPORT_SYMBOL(dmu_read_by_dnode);
EXPORT_SYMBOL(dmu_write); EXPORT_SYMBOL(dmu_write);

9
module/zfs/dmu_objset.c
View File

@ -1663,11 +1663,12 @@ dmu_objset_sync(objset_t *os, zio_t *pio, dmu_tx_t *tx)
dmu_write_policy(os, NULL, 0, 0, &zp); dmu_write_policy(os, NULL, 0, 0, &zp);
/* /*
* If we are either claiming the ZIL or doing a raw receive write out * If we are either claiming the ZIL or doing a raw receive, write
* the os_phys_buf raw. Neither of these actions will effect the MAC * out the os_phys_buf raw. Neither of these actions will effect the
* at this point. * MAC at this point.
*/ */
if (os->os_next_write_raw[tx->tx_txg & TXG_MASK]) { if (os->os_raw_receive ||
os->os_next_write_raw[tx->tx_txg & TXG_MASK]) {
ASSERT(os->os_encrypted); ASSERT(os->os_encrypted);
os->os_next_write_raw[tx->tx_txg & TXG_MASK] = B_FALSE; os->os_next_write_raw[tx->tx_txg & TXG_MASK] = B_FALSE;
arc_convert_to_raw(os->os_phys_buf, arc_convert_to_raw(os->os_phys_buf,

102
module/zfs/dmu_send.c
View File

@ -2174,6 +2174,7 @@ struct receive_writer_arg {
uint64_t bytes_read; /* bytes read when current record created */ uint64_t bytes_read; /* bytes read when current record created */
/* Encryption parameters for the last received DRR_OBJECT_RANGE */ /* Encryption parameters for the last received DRR_OBJECT_RANGE */
boolean_t or_crypt_params_present;
uint64_t or_firstobj; uint64_t or_firstobj;
uint64_t or_numslots; uint64_t or_numslots;
uint8_t or_salt[ZIO_DATA_SALT_LEN]; uint8_t or_salt[ZIO_DATA_SALT_LEN];
@ -2507,23 +2508,16 @@ receive_object(struct receive_writer_arg *rwa, struct drr_object *drro,
if (rwa->raw && nblkptr != drro->drr_nblkptr) if (rwa->raw && nblkptr != drro->drr_nblkptr)
return (SET_ERROR(EINVAL)); return (SET_ERROR(EINVAL));
if (rwa->raw && if (drro->drr_blksz != doi.doi_data_block_size ||
(drro->drr_blksz != doi.doi_data_block_size ||
nblkptr < doi.doi_nblkptr || nblkptr < doi.doi_nblkptr ||
indblksz != doi.doi_metadata_block_size || drro->drr_dn_slots != doi.doi_dnodesize >> DNODE_SHIFT ||
drro->drr_nlevels < doi.doi_indirection || (rwa->raw &&
drro->drr_dn_slots != doi.doi_dnodesize >> DNODE_SHIFT)) { (indblksz != doi.doi_metadata_block_size ||
err = dmu_free_long_range_raw(rwa->os, drro->drr_nlevels < doi.doi_indirection))) {
err = dmu_free_long_range(rwa->os,
drro->drr_object, 0, DMU_OBJECT_END); drro->drr_object, 0, DMU_OBJECT_END);
if (err != 0) if (err != 0)
return (SET_ERROR(EINVAL)); return (SET_ERROR(EINVAL));
} else if (drro->drr_blksz != doi.doi_data_block_size ||
nblkptr < doi.doi_nblkptr ||
drro->drr_dn_slots != doi.doi_dnodesize >> DNODE_SHIFT) {
err = dmu_free_long_range(rwa->os, drro->drr_object,
0, DMU_OBJECT_END);
if (err != 0)
return (SET_ERROR(EINVAL));
} }
/* /*
@ -2538,13 +2532,7 @@ receive_object(struct receive_writer_arg *rwa, struct drr_object *drro,
*/ */
if ((rwa->raw && drro->drr_nlevels < doi.doi_indirection) || if ((rwa->raw && drro->drr_nlevels < doi.doi_indirection) ||
drro->drr_dn_slots != doi.doi_dnodesize >> DNODE_SHIFT) { drro->drr_dn_slots != doi.doi_dnodesize >> DNODE_SHIFT) {
if (rwa->raw) { err = dmu_free_long_object(rwa->os, drro->drr_object);
err = dmu_free_long_object_raw(rwa->os,
drro->drr_object);
} else {
err = dmu_free_long_object(rwa->os,
drro->drr_object);
}
if (err != 0) if (err != 0)
return (SET_ERROR(EINVAL)); return (SET_ERROR(EINVAL));
@ -2586,9 +2574,6 @@ receive_object(struct receive_writer_arg *rwa, struct drr_object *drro,
else if (err != 0) else if (err != 0)
return (err); return (err);
if (rwa->raw)
err = dmu_free_long_object_raw(rwa->os, slot);
else
err = dmu_free_long_object(rwa->os, slot); err = dmu_free_long_object(rwa->os, slot);
if (err != 0) if (err != 0)
@ -2630,26 +2615,38 @@ receive_object(struct receive_writer_arg *rwa, struct drr_object *drro,
return (SET_ERROR(EINVAL)); return (SET_ERROR(EINVAL));
} }
if (rwa->raw) { if (rwa->or_crypt_params_present) {
/* /*
* Convert the buffer associated with this range of dnodes * Set the crypt params for the buffer associated with this
* to a raw buffer. This ensures that it will be written out * range of dnodes. This causes the blkptr_t to have the
* as a raw buffer when we fill in the dnode object. Since we * same crypt params (byteorder, salt, iv, mac) as on the
* are committing this tx now, it is possible for the dnode * sending side.
* block to end up on-disk with the incorrect MAC. Despite *
* this, the dataset is marked as inconsistent so no other * Since we are committing this tx now, it is possible for
* code paths (apart from scrubs) will attempt to read this * the dnode block to end up on-disk with the incorrect MAC,
* data. Scrubs will not be effected by this either since * if subsequent objects in this block are received in a
* scrubs only read raw data and do not attempt to check * different txg. However, since the dataset is marked as
* the MAC. * inconsistent, no code paths will do a non-raw read (or
* decrypt the block / verify the MAC). The receive code and
* scrub code can safely do raw reads and verify the
* checksum. They don't need to verify the MAC.
*/ */
err = dmu_convert_mdn_block_to_raw(rwa->os, rwa->or_firstobj, dmu_buf_t *db = NULL;
rwa->or_byteorder, rwa->or_salt, rwa->or_iv, rwa->or_mac, uint64_t offset = rwa->or_firstobj * DNODE_MIN_SIZE;
tx);
err = dmu_buf_hold_by_dnode(DMU_META_DNODE(rwa->os),
offset, FTAG, &db, DMU_READ_PREFETCH | DMU_READ_NO_DECRYPT);
if (err != 0) { if (err != 0) {
dmu_tx_commit(tx); dmu_tx_commit(tx);
return (SET_ERROR(EINVAL)); return (SET_ERROR(EINVAL));
} }
dmu_buf_set_crypt_params(db, rwa->or_byteorder,
rwa->or_salt, rwa->or_iv, rwa->or_mac, tx);
dmu_buf_rele(db, FTAG);
rwa->or_crypt_params_present = B_FALSE;
} }
dmu_object_set_checksum(rwa->os, drro->drr_object, dmu_object_set_checksum(rwa->os, drro->drr_object,
@ -2726,9 +2723,6 @@ receive_freeobjects(struct receive_writer_arg *rwa,
else if (err != 0) else if (err != 0)
return (err); return (err);
if (rwa->raw)
err = dmu_free_long_object_raw(rwa->os, obj);
else
err = dmu_free_long_object(rwa->os, obj); err = dmu_free_long_object(rwa->os, obj);
if (err != 0) if (err != 0)
@ -2781,9 +2775,6 @@ receive_write(struct receive_writer_arg *rwa, struct drr_write *drrw,
return (err); return (err);
} }
if (rwa->raw)
VERIFY0(dmu_object_dirty_raw(rwa->os, drrw->drr_object, tx));
if (rwa->byteswap && !arc_is_encrypted(abuf) && if (rwa->byteswap && !arc_is_encrypted(abuf) &&
arc_get_compression(abuf) == ZIO_COMPRESS_OFF) { arc_get_compression(abuf) == ZIO_COMPRESS_OFF) {
dmu_object_byteswap_t byteswap = dmu_object_byteswap_t byteswap =
@ -2870,7 +2861,6 @@ receive_write_byref(struct receive_writer_arg *rwa,
} }
if (rwa->raw) { if (rwa->raw) {
VERIFY0(dmu_object_dirty_raw(rwa->os, drrwbr->drr_object, tx));
dmu_copy_from_buf(rwa->os, drrwbr->drr_object, dmu_copy_from_buf(rwa->os, drrwbr->drr_object,
drrwbr->drr_offset, dbp, tx); drrwbr->drr_offset, dbp, tx);
} else { } else {
@ -2971,13 +2961,7 @@ receive_spill(struct receive_writer_arg *rwa, struct drr_spill *drrs,
dmu_tx_abort(tx); dmu_tx_abort(tx);
return (err); return (err);
} }
if (rwa->raw) {
VERIFY0(dmu_object_dirty_raw(rwa->os, drrs->drr_object, tx));
dmu_buf_will_change_crypt_params(db_spill, tx);
} else {
dmu_buf_will_dirty(db_spill, tx); dmu_buf_will_dirty(db_spill, tx);
}
if (db_spill->db_size < drrs->drr_length) if (db_spill->db_size < drrs->drr_length)
VERIFY(0 == dbuf_spill_set_blksz(db_spill, VERIFY(0 == dbuf_spill_set_blksz(db_spill,
@ -3016,13 +3000,8 @@ receive_free(struct receive_writer_arg *rwa, struct drr_free *drrf)
if (drrf->drr_object > rwa->max_object) if (drrf->drr_object > rwa->max_object)
rwa->max_object = drrf->drr_object; rwa->max_object = drrf->drr_object;
if (rwa->raw) {
err = dmu_free_long_range_raw(rwa->os, drrf->drr_object,
drrf->drr_offset, drrf->drr_length);
} else {
err = dmu_free_long_range(rwa->os, drrf->drr_object, err = dmu_free_long_range(rwa->os, drrf->drr_object,
drrf->drr_offset, drrf->drr_length); drrf->drr_offset, drrf->drr_length);
}
return (err); return (err);
} }
@ -3062,9 +3041,10 @@ receive_object_range(struct receive_writer_arg *rwa,
/* /*
* The DRR_OBJECT_RANGE handling must be deferred to receive_object() * The DRR_OBJECT_RANGE handling must be deferred to receive_object()
* so that the encryption parameters are set with each object that is * so that the block of dnodes is not written out when it's empty,
* written into that block. * and converted to a HOLE BP.
*/ */
rwa->or_crypt_params_present = B_TRUE;
rwa->or_firstobj = drror->drr_firstobj; rwa->or_firstobj = drror->drr_firstobj;
rwa->or_numslots = drror->drr_numslots; rwa->or_numslots = drror->drr_numslots;
bcopy(drror->drr_salt, rwa->or_salt, ZIO_DATA_SALT_LEN); bcopy(drror->drr_salt, rwa->or_salt, ZIO_DATA_SALT_LEN);
@ -3090,6 +3070,7 @@ dmu_recv_cleanup_ds(dmu_recv_cookie_t *drc)
* after we stopped receiving the dataset. * after we stopped receiving the dataset.
*/ */
txg_wait_synced(ds->ds_dir->dd_pool, 0); txg_wait_synced(ds->ds_dir->dd_pool, 0);
ds->ds_objset->os_raw_receive = B_FALSE;
rrw_enter(&ds->ds_bp_rwlock, RW_READER, FTAG); rrw_enter(&ds->ds_bp_rwlock, RW_READER, FTAG);
if (drc->drc_resumable && !BP_IS_HOLE(dsl_dataset_get_blkptr(ds))) { if (drc->drc_resumable && !BP_IS_HOLE(dsl_dataset_get_blkptr(ds))) {
@ -3841,6 +3822,7 @@ dmu_recv_stream(dmu_recv_cookie_t *drc, vnode_t *vp, offset_t *voffp,
rwa->byteswap = drc->drc_byteswap; rwa->byteswap = drc->drc_byteswap;
rwa->resumable = drc->drc_resumable; rwa->resumable = drc->drc_resumable;
rwa->raw = drc->drc_raw; rwa->raw = drc->drc_raw;
rwa->os->os_raw_receive = drc->drc_raw;
(void) thread_create(NULL, 0, receive_writer_thread, rwa, 0, curproc, (void) thread_create(NULL, 0, receive_writer_thread, rwa, 0, curproc,
TS_RUN, minclsyspri); TS_RUN, minclsyspri);
@ -3903,12 +3885,7 @@ dmu_recv_stream(dmu_recv_cookie_t *drc, vnode_t *vp, offset_t *voffp,
int next_err = 0; int next_err = 0;
while (next_err == 0) { while (next_err == 0) {
if (drc->drc_raw) {
free_err = dmu_free_long_object_raw(rwa->os,
obj);
} else {
free_err = dmu_free_long_object(rwa->os, obj); free_err = dmu_free_long_object(rwa->os, obj);
}
if (free_err != 0 && free_err != ENOENT) if (free_err != 0 && free_err != ENOENT)
break; break;
@ -4037,6 +4014,7 @@ dmu_recv_end_sync(void *arg, dmu_tx_t *tx)
spa_history_log_internal_ds(drc->drc_ds, "finish receiving", spa_history_log_internal_ds(drc->drc_ds, "finish receiving",
tx, "snap=%s", drc->drc_tosnap); tx, "snap=%s", drc->drc_tosnap);
drc->drc_ds->ds_objset->os_raw_receive = B_FALSE;
if (!drc->drc_newfs) { if (!drc->drc_newfs) {
dsl_dataset_t *origin_head; dsl_dataset_t *origin_head;