From 95fd54a1c5b93bb2aa3e7dffc28c784b1e21a8bb Mon Sep 17 00:00:00 2001 From: Steven Hartland Date: Sat, 25 May 2013 02:06:23 +0000 Subject: Illumos #3740 3740 Poor ZFS send / receive performance due to snapshot hold / release processing Reviewed by: Matthew Ahrens Approved by: Christopher Siden References: https://www.illumos.org/issues/3740 illumos/illumos-gate@a7a845e4bf22fd1b2a284729ccd95c7370a0438c Ported-by: Richard Yao Signed-off-by: Brian Behlendorf Issue #1775 Porting notes: 1. 13fe019870c8779bf2f5b3ff731b512cf89133ef introduced a merge conflict in dsl_dataset_user_release_tmp where some variables were moved outside of the preprocessor directive. 2. dea9dfefdd747534b3846845629d2200f0616dad made the previous merge conflict worse by switching KM_SLEEP to KM_PUSHPAGE. This is notable because this commit refactors the code, adding a new KM_SLEEP allocation. It is not clear to me whether this should be converted to KM_PUSHPAGE. 3. We had a merge conflict in libzfs_sendrecv.c because of copyright notices. 4. Several small C99 compatibility fixed were made. --- lib/libzfs/libzfs_sendrecv.c | 148 +++++++++++++++++++++---------------------- 1 file changed, 74 insertions(+), 74 deletions(-) (limited to 'lib/libzfs/libzfs_sendrecv.c') diff --git a/lib/libzfs/libzfs_sendrecv.c b/lib/libzfs/libzfs_sendrecv.c index 28751b215..6f0d27f5b 100644 --- a/lib/libzfs/libzfs_sendrecv.c +++ b/lib/libzfs/libzfs_sendrecv.c @@ -22,9 +22,10 @@ /* * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved. * Copyright (c) 2012 by Delphix. All rights reserved. - * Copyright (c) 2012 Pawel Jakub Dawidek . * Copyright (c) 2012, Joyent, Inc. All rights reserved. + * Copyright (c) 2012 Pawel Jakub Dawidek . * All rights reserved + * Copyright (c) 2013 Steven Hartland. All rights reserved. */ #include @@ -799,6 +800,7 @@ typedef struct send_dump_data { int outfd; boolean_t err; nvlist_t *fss; + nvlist_t *snapholds; avl_tree_t *fsavl; snapfilter_cb_t *filter_cb; void *filter_cb_arg; @@ -948,41 +950,19 @@ dump_ioctl(zfs_handle_t *zhp, const char *fromsnap, uint64_t fromsnap_obj, return (0); } -static int -hold_for_send(zfs_handle_t *zhp, send_dump_data_t *sdd) +static void +gather_holds(zfs_handle_t *zhp, send_dump_data_t *sdd) { - zfs_handle_t *pzhp; - int error = 0; - char *thissnap; - assert(zhp->zfs_type == ZFS_TYPE_SNAPSHOT); - if (sdd->dryrun) - return (0); - /* - * zfs_send() only opens a cleanup_fd for sends that need it, + * zfs_send() only sets snapholds for sends that need them, * e.g. replication and doall. */ - if (sdd->cleanup_fd == -1) - return (0); - - thissnap = strchr(zhp->zfs_name, '@') + 1; - *(thissnap - 1) = '\0'; - pzhp = zfs_open(zhp->zfs_hdl, zhp->zfs_name, ZFS_TYPE_DATASET); - *(thissnap - 1) = '@'; - - /* - * It's OK if the parent no longer exists. The send code will - * handle that error. - */ - if (pzhp) { - error = zfs_hold(pzhp, thissnap, sdd->holdtag, - B_FALSE, B_TRUE, sdd->cleanup_fd); - zfs_close(pzhp); - } + if (sdd->snapholds == NULL) + return; - return (error); + fnvlist_add_string(sdd->snapholds, zhp->zfs_name, sdd->holdtag); } static void * @@ -1038,28 +1018,23 @@ dump_snapshot(zfs_handle_t *zhp, void *arg) send_dump_data_t *sdd = arg; progress_arg_t pa = { 0 }; pthread_t tid; - char *thissnap; int err; boolean_t isfromsnap, istosnap, fromorigin; boolean_t exclude = B_FALSE; + err = 0; thissnap = strchr(zhp->zfs_name, '@') + 1; isfromsnap = (sdd->fromsnap != NULL && strcmp(sdd->fromsnap, thissnap) == 0); if (!sdd->seenfrom && isfromsnap) { - err = hold_for_send(zhp, sdd); - if (err == 0) { - sdd->seenfrom = B_TRUE; - (void) strcpy(sdd->prevsnap, thissnap); - sdd->prevsnap_obj = zfs_prop_get_int(zhp, - ZFS_PROP_OBJSETID); - } else if (err == ENOENT) { - err = 0; - } + gather_holds(zhp, sdd); + sdd->seenfrom = B_TRUE; + (void) strcpy(sdd->prevsnap, thissnap); + sdd->prevsnap_obj = zfs_prop_get_int(zhp, ZFS_PROP_OBJSETID); zfs_close(zhp); - return (err); + return (0); } if (sdd->seento || !sdd->seenfrom) { @@ -1110,14 +1085,7 @@ dump_snapshot(zfs_handle_t *zhp, void *arg) return (0); } - err = hold_for_send(zhp, sdd); - if (err) { - if (err == ENOENT) - err = 0; - zfs_close(zhp); - return (err); - } - + gather_holds(zhp, sdd); fromorigin = sdd->prevsnap[0] == '\0' && (sdd->fromorigin || sdd->replicate); @@ -1385,7 +1353,7 @@ zfs_send(zfs_handle_t *zhp, const char *fromsnap, const char *tosnap, avl_tree_t *fsavl = NULL; static uint64_t holdseq; int spa_version; - pthread_t tid; + pthread_t tid = 0; int pipefd[2]; dedup_arg_t dda = { 0 }; int featureflags = 0; @@ -1458,11 +1426,8 @@ zfs_send(zfs_handle_t *zhp, const char *fromsnap, const char *tosnap, *debugnvp = hdrnv; else nvlist_free(hdrnv); - if (err) { - fsavl_destroy(fsavl); - nvlist_free(fss); + if (err) goto stderr_out; - } } if (!flags->dryrun) { @@ -1486,8 +1451,6 @@ zfs_send(zfs_handle_t *zhp, const char *fromsnap, const char *tosnap, } free(packbuf); if (err == -1) { - fsavl_destroy(fsavl); - nvlist_free(fss); err = errno; goto stderr_out; } @@ -1498,8 +1461,6 @@ zfs_send(zfs_handle_t *zhp, const char *fromsnap, const char *tosnap, drr.drr_u.drr_end.drr_checksum = zc; err = write(outfd, &drr, sizeof (drr)); if (err == -1) { - fsavl_destroy(fsavl); - nvlist_free(fss); err = errno; goto stderr_out; } @@ -1511,7 +1472,7 @@ zfs_send(zfs_handle_t *zhp, const char *fromsnap, const char *tosnap, /* dump each stream */ sdd.fromsnap = fromsnap; sdd.tosnap = tosnap; - if (flags->dedup) + if (tid != 0) sdd.outfd = pipefd[0]; else sdd.outfd = outfd; @@ -1548,36 +1509,71 @@ zfs_send(zfs_handle_t *zhp, const char *fromsnap, const char *tosnap, err = errno; goto stderr_out; } + sdd.snapholds = fnvlist_alloc(); } else { sdd.cleanup_fd = -1; + sdd.snapholds = NULL; } - if (flags->verbose) { + if (flags->verbose || sdd.snapholds != NULL) { /* * Do a verbose no-op dry run to get all the verbose output - * before generating any data. Then do a non-verbose real - * run to generate the streams. + * or to gather snapshot hold's before generating any data, + * then do a non-verbose real run to generate the streams. */ sdd.dryrun = B_TRUE; err = dump_filesystems(zhp, &sdd); - sdd.dryrun = flags->dryrun; - sdd.verbose = B_FALSE; - if (flags->parsable) { - (void) fprintf(stderr, "size\t%llu\n", - (longlong_t)sdd.size); - } else { - char buf[16]; - zfs_nicenum(sdd.size, buf, sizeof (buf)); - (void) fprintf(stderr, dgettext(TEXT_DOMAIN, - "total estimated size is %s\n"), buf); + + if (err != 0) + goto stderr_out; + + if (flags->verbose) { + if (flags->parsable) { + (void) fprintf(stderr, "size\t%llu\n", + (longlong_t)sdd.size); + } else { + char buf[16]; + zfs_nicenum(sdd.size, buf, sizeof (buf)); + (void) fprintf(stderr, dgettext(TEXT_DOMAIN, + "total estimated size is %s\n"), buf); + } + } + + /* Ensure no snaps found is treated as an error. */ + if (!sdd.seento) { + err = ENOENT; + goto err_out; } + + /* Skip the second run if dryrun was requested. */ + if (flags->dryrun) + goto err_out; + + if (sdd.snapholds != NULL) { + err = zfs_hold_nvl(zhp, sdd.cleanup_fd, sdd.snapholds); + if (err != 0) + goto stderr_out; + + fnvlist_free(sdd.snapholds); + sdd.snapholds = NULL; + } + + sdd.dryrun = B_FALSE; + sdd.verbose = B_FALSE; } + err = dump_filesystems(zhp, &sdd); fsavl_destroy(fsavl); nvlist_free(fss); - if (flags->dedup) { - (void) close(pipefd[0]); + /* Ensure no snaps found is treated as an error. */ + if (err == 0 && !sdd.seento) + err = ENOENT; + + if (tid != 0) { + if (err != 0) + (void) pthread_cancel(tid); (void) pthread_join(tid, NULL); + (void) close(pipefd[0]); } if (sdd.cleanup_fd != -1) { @@ -1605,9 +1601,13 @@ zfs_send(zfs_handle_t *zhp, const char *fromsnap, const char *tosnap, stderr_out: err = zfs_standard_error(zhp->zfs_hdl, err, errbuf); err_out: + fsavl_destroy(fsavl); + nvlist_free(fss); + fnvlist_free(sdd.snapholds); + if (sdd.cleanup_fd != -1) VERIFY(0 == close(sdd.cleanup_fd)); - if (flags->dedup) { + if (tid != 0) { (void) pthread_cancel(tid); (void) pthread_join(tid, NULL); (void) close(pipefd[0]); -- cgit v1.2.3