/*ARGSUSED*/
static int
zfs_znode_cache_constructor(void *buf, void *arg, int kmflags)
{
znode_t *zp = buf;
ASSERT(!POINTER_IS_VALID(zp->z_zfsvfs));
zp->z_vnode = vn_alloc(kmflags);
if (zp->z_vnode == NULL) {
return (-1);
}
ZTOV(zp)->v_data = zp;
list_link_init(&zp->z_link_node);
mutex_init(&zp->z_lock, NULL, MUTEX_DEFAULT, NULL);
rw_init(&zp->z_parent_lock, NULL, RW_DEFAULT, NULL);
rw_init(&zp->z_name_lock, NULL, RW_DEFAULT, NULL);
mutex_init(&zp->z_acl_lock, NULL, MUTEX_DEFAULT, NULL);
mutex_init(&zp->z_range_lock, NULL, MUTEX_DEFAULT, NULL);
avl_create(&zp->z_range_avl, zfs_range_compare,
sizeof (rl_t), offsetof(rl_t, r_node));
zp->z_dbuf = NULL;
zp->z_dirlocks = NULL;
zp->z_acl_cached = NULL;
return (0);
}
static void
zfs_znode_dmu_init(zfsvfs_t *zfsvfs, znode_t *zp, dmu_buf_t *db)
{
znode_t *nzp;
ASSERT(!POINTER_IS_VALID(zp->z_zfsvfs) || (zfsvfs == zp->z_zfsvfs));
ASSERT(MUTEX_HELD(ZFS_OBJ_MUTEX(zfsvfs, zp->z_id)));
mutex_enter(&zp->z_lock);
ASSERT(zp->z_dbuf == NULL);
zp->z_dbuf = db;
nzp = dmu_buf_set_user_ie(db, zp, &zp->z_phys, znode_evict_error);
/*
* there should be no
* concurrent zgets on this object.
*/
if (nzp != NULL)
panic("existing znode %p for dbuf %p", (void *)nzp, (void *)db);
/*
* Slap on VROOT if we are the root znode
*/
if (zp->z_id == zfsvfs->z_root)
ZTOV(zp)->v_flag |= VROOT;
mutex_exit(&zp->z_lock);
vn_exists(ZTOV(zp));
}
/**
* Retrieve the blocks of the given file currently in the Linux Page Cache.
*
* @param str IN The string array.
* @param max IN The number of elements in the array.
* @param value IN The search value.
* @returns index of the array where the value is located or -1 if not found.
*/
void
PrimeCache (void)
{
sqlite3 *dbh = NULL; /* Database Handle */
sqlite3_stmt *sth = NULL; /* SQLite Statement Handle */
int rc = 0; /* SQLite Return Code */
printf("\nBeginning Cache Priming\n");
fflush(stdout);
/* Enable the SQLite shared cache */
// sqlite3_enable_shared_cache(1);
/* Open the SQLite database */
rc = sqlite3_open(dbFileName, &dbh);
sqlite3_busy_handler(dbh, busyHandler, NULL);
/* Prepare the SQL */
rc = sqlite3_prepare(dbh, SQL_CACHED_FILES_QUERY, -1, &sth, NULL);
/*
* Iterate over the block ranges, reading them into cache.
*/
while ((rc = sqlite3_step(sth)) != SQLITE_ERROR)
{
if (rc == SQLITE_ROW)
{
int fileId = sqlite3_column_int(sth, 0);
if (POINTER_IS_VALID(thp))
tpool_add_work(thp, PrimeCacheForFile, (void *)fileId);
else
PrimeCacheForFile(fileId);
}
else if (rc == SQLITE_DONE)
{
break;
}
else if (rc == SQLITE_BUSY)
{
continue;
}
}
sqlite3_finalize(sth);
sqlite3_close(dbh);
} /* PrimeCache() */
/*ARGSUSED*/
static void
zfs_znode_cache_destructor(void *buf, void *arg)
{
znode_t *zp = buf;
ASSERT(!POINTER_IS_VALID(zp->z_zfsvfs));
ASSERT(ZTOV(zp) == NULL);
vn_free(ZTOV(zp));
ASSERT(!list_link_active(&zp->z_link_node));
mutex_destroy(&zp->z_lock);
rw_destroy(&zp->z_map_lock);
rw_destroy(&zp->z_parent_lock);
rw_destroy(&zp->z_name_lock);
mutex_destroy(&zp->z_acl_lock);
avl_destroy(&zp->z_range_avl);
mutex_destroy(&zp->z_range_lock);
ASSERT(zp->z_dbuf == NULL);
ASSERT(zp->z_dirlocks == NULL);
}
/*
* XXX: We cannot use this function as a cache constructor, because
* there is one global cache for all file systems and we need
* to pass vfsp here, which is not possible, because argument
* 'cdrarg' is defined at kmem_cache_create() time.
*/
static int
zfs_znode_cache_constructor(void *buf, void *arg, int kmflags)
{
znode_t *zp = buf;
vnode_t *vp;
vfs_t *vfsp = arg;
int error;
POINTER_INVALIDATE(&zp->z_zfsvfs);
ASSERT(!POINTER_IS_VALID(zp->z_zfsvfs));
if (vfsp != NULL) {
error = getnewvnode("zfs", vfsp, &zfs_vnodeops, &vp);
if (error != 0 && (kmflags & KM_NOSLEEP))
return (-1);
ASSERT(error == 0);
vn_lock(vp, LK_EXCLUSIVE | LK_RETRY, curthread);
zp->z_vnode = vp;
vp->v_data = (caddr_t)zp;
VN_LOCK_AREC(vp);
} else {
zp->z_vnode = NULL;
}
list_link_init(&zp->z_link_node);
mutex_init(&zp->z_lock, NULL, MUTEX_DEFAULT, NULL);
rw_init(&zp->z_map_lock, NULL, RW_DEFAULT, NULL);
rw_init(&zp->z_parent_lock, NULL, RW_DEFAULT, NULL);
rw_init(&zp->z_name_lock, NULL, RW_DEFAULT, NULL);
mutex_init(&zp->z_acl_lock, NULL, MUTEX_DEFAULT, NULL);
mutex_init(&zp->z_range_lock, NULL, MUTEX_DEFAULT, NULL);
avl_create(&zp->z_range_avl, zfs_range_compare,
sizeof (rl_t), offsetof(rl_t, r_node));
zp->z_dbuf = NULL;
zp->z_dirlocks = NULL;
return (0);
}
/*
* Construct a new znode/vnode and intialize.
*
* This does not do a call to dmu_set_user() that is
* up to the caller to do, in case you don't want to
* return the znode
*/
static znode_t *
zfs_znode_alloc(zfsvfs_t *zfsvfs, dmu_buf_t *db, int blksz)
{
znode_t *zp;
vnode_t *vp;
zp = kmem_cache_alloc(znode_cache, KM_SLEEP);
zfs_znode_cache_constructor(zp, zfsvfs->z_parent->z_vfs, 0);
ASSERT(zp->z_dirlocks == NULL);
ASSERT(zp->z_dbuf == NULL);
ASSERT(!POINTER_IS_VALID(zp->z_zfsvfs));
/*
* Defer setting z_zfsvfs until the znode is ready to be a candidate for
* the zfs_znode_move() callback.
*/
zp->z_phys = NULL;
zp->z_unlinked = 0;
zp->z_atime_dirty = 0;
zp->z_mapcnt = 0;
zp->z_last_itx = 0;
zp->z_id = db->db_object;
zp->z_blksz = blksz;
zp->z_seq = 0x7A4653;
zp->z_sync_cnt = 0;
vp = ZTOV(zp);
#ifdef TODO
vn_reinit(vp);
#endif
zfs_znode_dmu_init(zfsvfs, zp, db);
zp->z_gen = zp->z_phys->zp_gen;
#if 0
if (vp == NULL)
return (zp);
#endif
vp->v_type = IFTOVT((mode_t)zp->z_phys->zp_mode);
switch (vp->v_type) {
case VDIR:
zp->z_zn_prefetch = B_TRUE; /* z_prefetch default is enabled */
break;
case VFIFO:
vp->v_op = &zfs_fifoops;
break;
}
if (vp->v_type != VFIFO)
VN_LOCK_ASHARE(vp);
mutex_enter(&zfsvfs->z_znodes_lock);
list_insert_tail(&zfsvfs->z_all_znodes, zp);
membar_producer();
/*
* Everything else must be valid before assigning z_zfsvfs makes the
* znode eligible for zfs_znode_move().
*/
zp->z_zfsvfs = zfsvfs;
mutex_exit(&zfsvfs->z_znodes_lock);
VFS_HOLD(zfsvfs->z_vfs);
return (zp);
}
/*ARGSUSED*/
static kmem_cbrc_t
zfs_znode_move(void *buf, void *newbuf, size_t size, void *arg)
{
znode_t *ozp = buf, *nzp = newbuf;
zfsvfs_t *zfsvfs;
vnode_t *vp;
/*
* The znode is on the file system's list of known znodes if the vfs
* pointer is valid. We set the low bit of the vfs pointer when freeing
* the znode to invalidate it, and the memory patterns written by kmem
* (baddcafe and deadbeef) set at least one of the two low bits. A newly
* created znode sets the vfs pointer last of all to indicate that the
* znode is known and in a valid state to be moved by this function.
*/
zfsvfs = ozp->z_zfsvfs;
if (!POINTER_IS_VALID(zfsvfs)) {
ZNODE_STAT_ADD(znode_move_stats.zms_zfsvfs_invalid);
return (KMEM_CBRC_DONT_KNOW);
}
/*
* Ensure that the filesystem is not unmounted during the move.
*/
if (zfs_enter(zfsvfs) != 0) { /* ZFS_ENTER */
ZNODE_STAT_ADD(znode_move_stats.zms_zfsvfs_unmounted);
return (KMEM_CBRC_DONT_KNOW);
}
mutex_enter(&zfsvfs->z_znodes_lock);
/*
* Recheck the vfs pointer in case the znode was removed just before
* acquiring the lock.
*/
if (zfsvfs != ozp->z_zfsvfs) {
mutex_exit(&zfsvfs->z_znodes_lock);
ZFS_EXIT(zfsvfs);
ZNODE_STAT_ADD(znode_move_stats.zms_zfsvfs_recheck_invalid);
return (KMEM_CBRC_DONT_KNOW);
}
/*
* At this point we know that as long as we hold z_znodes_lock, the
* znode cannot be freed and fields within the znode can be safely
* accessed. Now, prevent a race with zfs_zget().
*/
if (ZFS_OBJ_HOLD_TRYENTER(zfsvfs, ozp->z_id) == 0) {
mutex_exit(&zfsvfs->z_znodes_lock);
ZFS_EXIT(zfsvfs);
ZNODE_STAT_ADD(znode_move_stats.zms_obj_held);
return (KMEM_CBRC_LATER);
}
vp = ZTOV(ozp);
if (mutex_tryenter(&vp->v_lock) == 0) {
ZFS_OBJ_HOLD_EXIT(zfsvfs, ozp->z_id);
mutex_exit(&zfsvfs->z_znodes_lock);
ZFS_EXIT(zfsvfs);
ZNODE_STAT_ADD(znode_move_stats.zms_vnode_locked);
return (KMEM_CBRC_LATER);
}
/* Only move znodes that are referenced _only_ by the DNLC. */
if (vp->v_count != 1 || !vn_in_dnlc(vp)) {
mutex_exit(&vp->v_lock);
ZFS_OBJ_HOLD_EXIT(zfsvfs, ozp->z_id);
mutex_exit(&zfsvfs->z_znodes_lock);
ZFS_EXIT(zfsvfs);
ZNODE_STAT_ADD(znode_move_stats.zms_not_only_dnlc);
return (KMEM_CBRC_LATER);
}
/*
* The znode is known and in a valid state to move. We're holding the
* locks needed to execute the critical section.
*/
zfs_znode_move_impl(ozp, nzp);
mutex_exit(&vp->v_lock);
ZFS_OBJ_HOLD_EXIT(zfsvfs, ozp->z_id);
list_link_replace(&ozp->z_link_node, &nzp->z_link_node);
mutex_exit(&zfsvfs->z_znodes_lock);
ZFS_EXIT(zfsvfs);
return (KMEM_CBRC_YES);
}
/**
* Retrieve the blocks of the given file currently in the Linux Page Cache.
*
* @param str IN The string array.
* @param max IN The number of elements in the array.
* @param value IN The search value.
* @returns index of the array where the value is located or -1 if not found.
*/
void
BuildCacheProfile (void)
{
PGconn *pgh = NULL; /* Postgres Connection Handle */
PGresult *pgrsh = NULL;
sqlite3 *dbh = NULL; /* SQLite Database Handle */
sqlite3_stmt *sth = NULL; /* SQLite Statement Handle */
int i = 0; /* Generic Iterator */
int rc = 0; /* SQLite Return Code */
int dbOid = 0; /* Postgres Database OID */
int dbTablespaceOid = 0; /* Postgres Default Tablespace OID */
char *baseDir = get_current_dir_name();
/* Open the SQLite database */
rc = sqlite3_open(dbFileName, &dbh);
sqlite3_busy_handler(dbh, busyHandler, NULL);
/* Connect to Postgres */
pgh = PQsetdbLogin(NULL, NULL, NULL, NULL,
pgConnectString == NULL ? "postgres"
: pgConnectString,
NULL, pgPassword);
if (PQstatus(pgh) == CONNECTION_BAD)
{
ERROR_PRINT("%s", "Connection Test Failed\n");
ERROR_PRINT("SQLERRMC: %s\n", PQerrorMessage(pgh));
PQfinish(pgh);
exit(EXIT_FAILURE);
}
/* Retrieve the Postgres block size */
pgrsh = PQexec(pgh, PG_BLCKSZ_QUERY);
if (PQresultStatus(pgrsh) == PGRES_TUPLES_OK)
{
db_block_size = atoi(PQgetvalue(pgrsh, 0, 0));
PQclear(pgrsh);
/* Inform user of block sizes */
printf("PG BLCKSZ........ %u\n", db_block_size);
printf("OS Page Size..... %u\n", (uint32_t)getpagesize());
}
else
{
ERROR_PRINT("%s", "Could not retrieve database settings\n");
ERROR_PRINT("SQLERRMC: %s\n", PQerrorMessage(pgh));
PQclear(pgrsh);
PQfinish(pgh);
}
/* Retrieve the Postgres block size */
pgrsh = PQexec(pgh, PG_DEFAULT_TABLESPACE_QUERY);
if (PQresultStatus(pgrsh) == PGRES_TUPLES_OK)
{
dbOid = atoi(PQgetvalue(pgrsh, 0, 0));
dbTablespaceOid = atoi(PQgetvalue(pgrsh, 0, 1));
PQclear(pgrsh);
}
else
{
ERROR_PRINT("%s", "Could not retrieve database settings\n");
ERROR_PRINT("SQLERRMC: %s\n", PQerrorMessage(pgh));
PQclear(pgrsh);
PQfinish(pgh);
}
/* Retrieve the Postgres tablespace information */
pgrsh = PQexec(pgh, PG_OBJECT_QUERY);
if (PQresultStatus(pgrsh) == PGRES_TUPLES_OK)
{
int numRows = PQntuples(pgrsh);
char filePath[1024];
printf("\nBeginning cache profiling for %d database objects.\n",
numRows);
for (i = 0; i < numRows; i++)
{
struct dirent *dp = NULL; /* Directory Entry Pointer */
DIR *dfd = NULL; /* Directory Stream */
char pattern1[64];
char pattern2[64];
char *name = PQgetvalue(pgrsh, i, 0);
char *kind = PQgetvalue(pgrsh, i, 1);
int tableSpace = atoi(PQgetvalue(pgrsh, i, 2));
int fileNode = atoi(PQgetvalue(pgrsh, i, 3));
snprintf(pattern1, sizeof(pattern1), "%d", fileNode);
snprintf(pattern2, sizeof(pattern2), "%d.*", fileNode);
if (tableSpace == 0 || tableSpace == dbTablespaceOid)
snprintf(filePath, sizeof(filePath),
"%s/pg_tblspc/%d/%d", pgDataDir,
dbTablespaceOid, dbOid);
dfd = opendir(filePath);
if (dfd != NULL)
{
chdir(filePath);
while ((dp = readdir(dfd)) != NULL)
{
struct stat sb;
/* Attempt to parse the given PG log file */
if (stat(dp->d_name, &sb) == -1)
continue;
/* Make sure this is a normal file */
if (!S_ISREG(sb.st_mode))
continue;
/* Does this file match our pattern */
DEBUG_PRINT("Checking pattern [%s] and [%s] for file [%s]\n",
pattern1, pattern2, dp->d_name);
if (fnmatch(pattern1, dp->d_name, 0) == 0
|| fnmatch(pattern2, dp->d_name, 0) == 0)
{
char fullFilePath[1024];
snprintf(fullFilePath, sizeof(fullFilePath),
"%s/%s", filePath, dp->d_name);
if (POINTER_IS_VALID(thp))
tpool_add_work(thp, ProfileCacheForFile,
(void *)strdup(fullFilePath));
else
ProfileCacheForFile(strdup(fullFilePath));
}
}
/* We're done, close the directory stream */
closedir(dfd);
chdir(baseDir);
}
}
PQclear(pgrsh);
}
else
{
ERROR_PRINT("%s", "Could not retrieve database settings\n");
ERROR_PRINT("SQLERRMC: %s\n", PQerrorMessage(pgh));
PQclear(pgrsh);
PQfinish(pgh);
}
/* We're done with the connection */
PQfinish(pgh);
} /* BuildCacheProfile() */
int
main (int argc, char **argv)
{
int opt = 0; /* Option Identifier */
int optindex = 0; /* Option Index */
bool isProfiling = false; /* Are we profiling the cache? */
bool isPriming = false; /* Are we priming the cache? */
long numCPUs = 0; /* The number of online CPUs */
struct dirent *dp = NULL; /* Directory Entry Pointer */
DIR *dfd = NULL; /* Directory Stream */
double loadAverages[3] = { 0.00 }; /* System Load Averages */
PGconn *pgh = NULL; /* Postgres Connection Handle */
bool isPWDRequired = false; /* Is Postgres Password Reqd? */
struct option long_options[] = /* Options for getopt() */
{
{"connect-string", required_argument, NULL, 'c'},
{"profile", no_argument, NULL, 'p'},
{"prime", no_argument, NULL, 'w'},
{"data-dir", required_argument, NULL, 'D'},
{"postgres-only", no_argument, NULL, 'o'},
{"sqlite-db", required_argument, NULL, 's'},
{"help", no_argument, NULL, 'h'},
{"debug", no_argument, NULL, 'd'},
{NULL, 0, NULL, 0}
};
/* Go for the glory! */
fprintf(stderr, "\n%s: Release %s - %s\n", PACKAGE_NAME,
PACKAGE_VERSION, APP_RELEASE);
fprintf(stderr, "\n%s\n\n", APP_COPYRIGHT);
fflush(stdout);
/* Process command-line options */
while ((opt = getopt_long(argc, argv, "c:s:D:awhdp",
long_options, &optindex)) != -1)
{
switch (opt)
{
case 'h':
usage();
exit(EXIT_SUCCESS);
break;
case 'p':
if (isPriming == false)
isProfiling = true;
else
{
fprintf(stderr,
"Profiling and warming are mutually exlusive!\n");
exit(EXIT_FAILURE);
}
break;
case 'w':
if (isProfiling == false)
isPriming = true;
else
{
fprintf(stderr,
"Profiling and warming are mutually exlusive!\n");
exit(EXIT_FAILURE);
}
break;
case 'd':
is_debug = true;
break;
case 's':
dbFileName = xstrdup(optarg);
break;
case 'c':
pgConnectString = xstrdup(optarg);
break;
case 'D':
pgDataDir = optarg;
break;
default:
usage();
exit(EXIT_FAILURE);
}
}
/* Make sure user requested profile OR prime */
if (isProfiling == false && isPriming == false)
{
fprintf(stderr, "Expected either -p or -w\n");
usage();
exit(EXIT_FAILURE);
}
/* Make sure the database name is set */
/* Get the PG log file name from the end of the command line */
if (optind < (argc - 1))
{
fprintf(stderr, "too many command-line arguments (first is \"%s\")\n",
argv[optind + 1]);
usage();
exit(EXIT_FAILURE);
}
/* Perform a Postgres connection test & get password (if required) */
do
{
isPWDRequired = false;
pgh = PQsetdbLogin(NULL, NULL, NULL, NULL,
pgConnectString == NULL ? "postgres"
: pgConnectString,
NULL, pgPassword);
if (PQstatus(pgh) == CONNECTION_BAD)
{
if (PQconnectionNeedsPassword(pgh) && pgPassword == NULL)
{
printf("\nTesting Postgres Connection\n");
PQfinish(pgh);
pgPassword = simple_prompt("Password: "******"%s", "Connection Test Failed\n");
ERROR_PRINT("SQLERRMC: %s\n", PQerrorMessage(pgh));
PQfinish(pgh);
exit(EXIT_FAILURE);
}
}
} while (isPWDRequired);
PQfinish(pgh);
/* Get the number of available CPUs */
numCPUs = sysconf(_SC_NPROCESSORS_ONLN);
if (numCPUs < 1)
numCPUs = 1;
/*
* Choose the number of CPUs to use in the thread pool based on load
* average. It only makes sense to do this if we have more than one CPU
* to play with.
*/
if ((numCPUs > 1)
&& (getloadavg(loadAverages, 3) == 3))
{
long idleCPUs = 0; /* The number of idle CPUs */
/* Show what we got */
printf("load averages.... %3.2f %3.2f %3.2f\n",
loadAverages[0], loadAverages[1], loadAverages[2]);
/*
* We're going to base the number of usable CPUs by subtracting
* the sum of 1 (to account for OS and I/O overhead) plus the 1 minute
* load average from the number of available CPUs.
*/
idleCPUs = numCPUs - (1 + (int)(loadAverages[0] + 0.5));
/* Assign # of available CPUs with some sanity checking */
if (idleCPUs < numCPUs)
numCPUs = idleCPUs;
if (numCPUs < 1)
numCPUs = 1;
}
/* Inform user of # of CPUs that will be used */
printf("usable CPUs...... %d\n", numCPUs);
/* If we have more than one CPU, multi-thread our operations */
if (numCPUs > 1)
{
/* Initialize the thread pool */
thp = tpool_init(numCPUs, 1024, true);
}
if (isProfiling)
BuildCacheProfile();
else /* isPriming */
PrimeCache();
/* If we have more than one CPU, multi-thread our operations */
if (POINTER_IS_VALID(thp))
{
/* Destroy the thread pool */
tpool_destroy(thp, 1);
}
/* Cleanup */
free(dbFileName);
return EXIT_SUCCESS;
} /* main() */
/*ARGSUSED*/
static kmem_cbrc_t
zfs_znode_move(void *buf, void *newbuf, size_t size, void *arg)
{
znode_t *ozp = buf, *nzp = newbuf;
zfsvfs_t *zfsvfs;
vnode_t *vp;
/*
* The znode is on the file system's list of known znodes if the vfs
* pointer is valid. We set the low bit of the vfs pointer when freeing
* the znode to invalidate it, and the memory patterns written by kmem
* (baddcafe and deadbeef) set at least one of the two low bits. A newly
* created znode sets the vfs pointer last of all to indicate that the
* znode is known and in a valid state to be moved by this function.
*/
zfsvfs = ozp->z_zfsvfs;
if (!POINTER_IS_VALID(zfsvfs)) {
ZNODE_STAT_ADD(znode_move_stats.zms_zfsvfs_invalid);
return (KMEM_CBRC_DONT_KNOW);
}
/*
* Close a small window in which it's possible that the filesystem could
* be unmounted and freed, and zfsvfs, though valid in the previous
* statement, could point to unrelated memory by the time we try to
* prevent the filesystem from being unmounted.
*/
rw_enter(&zfsvfs_lock, RW_WRITER);
if (zfsvfs != ozp->z_zfsvfs) {
rw_exit(&zfsvfs_lock);
ZNODE_STAT_ADD(znode_move_stats.zms_zfsvfs_recheck1);
return (KMEM_CBRC_DONT_KNOW);
}
/*
* If the znode is still valid, then so is the file system. We know that
* no valid file system can be freed while we hold zfsvfs_lock, so we
* can safely ensure that the filesystem is not and will not be
* unmounted. The next statement is equivalent to ZFS_ENTER().
*/
rrw_enter(&zfsvfs->z_teardown_lock, RW_READER, FTAG);
if (zfsvfs->z_unmounted) {
ZFS_EXIT(zfsvfs);
rw_exit(&zfsvfs_lock);
ZNODE_STAT_ADD(znode_move_stats.zms_zfsvfs_unmounted);
return (KMEM_CBRC_DONT_KNOW);
}
rw_exit(&zfsvfs_lock);
mutex_enter(&zfsvfs->z_znodes_lock);
/*
* Recheck the vfs pointer in case the znode was removed just before
* acquiring the lock.
*/
if (zfsvfs != ozp->z_zfsvfs) {
mutex_exit(&zfsvfs->z_znodes_lock);
ZFS_EXIT(zfsvfs);
ZNODE_STAT_ADD(znode_move_stats.zms_zfsvfs_recheck2);
return (KMEM_CBRC_DONT_KNOW);
}
/*
* At this point we know that as long as we hold z_znodes_lock, the
* znode cannot be freed and fields within the znode can be safely
* accessed. Now, prevent a race with zfs_zget().
*/
if (ZFS_OBJ_HOLD_TRYENTER(zfsvfs, ozp->z_id) == 0) {
mutex_exit(&zfsvfs->z_znodes_lock);
ZFS_EXIT(zfsvfs);
ZNODE_STAT_ADD(znode_move_stats.zms_obj_held);
return (KMEM_CBRC_LATER);
}
vp = ZTOV(ozp);
if (mutex_tryenter(&vp->v_lock) == 0) {
ZFS_OBJ_HOLD_EXIT(zfsvfs, ozp->z_id);
mutex_exit(&zfsvfs->z_znodes_lock);
ZFS_EXIT(zfsvfs);
ZNODE_STAT_ADD(znode_move_stats.zms_vnode_locked);
return (KMEM_CBRC_LATER);
}
/* Only move znodes that are referenced _only_ by the DNLC. */
if (vp->v_count != 1 || !vn_in_dnlc(vp)) {
mutex_exit(&vp->v_lock);
ZFS_OBJ_HOLD_EXIT(zfsvfs, ozp->z_id);
mutex_exit(&zfsvfs->z_znodes_lock);
ZFS_EXIT(zfsvfs);
ZNODE_STAT_ADD(znode_move_stats.zms_not_only_dnlc);
return (KMEM_CBRC_LATER);
}
/*
* The znode is known and in a valid state to move. We're holding the
* locks needed to execute the critical section.
*/
zfs_znode_move_impl(ozp, nzp);
mutex_exit(&vp->v_lock);
ZFS_OBJ_HOLD_EXIT(zfsvfs, ozp->z_id);
list_link_replace(&ozp->z_link_node, &nzp->z_link_node);
mutex_exit(&zfsvfs->z_znodes_lock);
ZFS_EXIT(zfsvfs);
return (KMEM_CBRC_YES);
}