[CRIU] [PATCH] deduplication: bunch neighbour data to punch together
Pavel Emelyanov
xemul at parallels.com
Fri Mar 7 00:38:05 PST 2014
On 03/07/2014 12:24 PM, Pavel Tikhomirov wrote:
> 2014-03-04 9:30 GMT+04:00 Pavel Tikhomirov <snorcht at gmail.com <mailto:snorcht at gmail.com>>:
>
> 2014-03-03 17:51 GMT+04:00 Tikhomirov Pavel <snorcht at gmail.com <mailto:snorcht at gmail.com>>:
>
> when decide that data is no longer needed, there are two cases:
> 1. if data neighbours previous block of "no needed" data, extend bunch
> block(it holds begining and size of concequent "no needed" data) by
> length of curent block and go next.
> 2. if data not neighbours bunch block(or bunch block size is bigger
> than MAX_BUNCH_SIZE), than punch bunch block and set bunch block
> to curent block.
>
> in the end make cleanup to punch last bunch block.
>
> Signed-off-by: Tikhomirov Pavel <snorcht at gmail.com <mailto:snorcht at gmail.com>>
> ---
> cr-dedup.c | 34 ++++++++++++++++++++++++++--------
> include/page-read.h | 5 ++++-
> page-read.c | 14 +++++++++++++-
> 3 files changed, 43 insertions(+), 10 deletions(-)
>
> diff --git a/cr-dedup.c b/cr-dedup.c
> index 6f059d1..9d269d2 100644
> --- a/cr-dedup.c
> +++ b/cr-dedup.c
> @@ -7,6 +7,8 @@
> #include "page-read.h"
> #include "restorer.h"
>
> +#define MAX_BUNCH_SIZE 256
> +
> static int cr_dedup_one_pagemap(int pid);
>
> int cr_dedup(void)
> @@ -101,15 +103,30 @@ exit:
> return 0;
> }
>
> -int punch_hole(int fd, unsigned long off, unsigned long len)
> +int punch_hole(int fd, unsigned long off, unsigned long len,
> + struct iovec * bunch, bool cleanup, int pr_id)
> {
> int ret;
> - pr_debug("Punch!/%lu/%lu/\n", off, len);
> - ret = fallocate(fd, FALLOC_FL_PUNCH_HOLE | FALLOC_FL_KEEP_SIZE,
> - off, len);
> - if (ret != 0) {
> - pr_perror("Error punching hole");
> - return -1;
> +
> + if ((unsigned long)bunch->iov_base + bunch->iov_len == off && !cleanup
> + && bunch->iov_len < MAX_BUNCH_SIZE * PAGE_SIZE) {
> + pr_debug("pr%d:Extend bunch len from %lx to %lx\n", pr_id,
> + bunch->iov_len, bunch->iov_len + len);
> + bunch->iov_len += len;
> + } else {
> + if (bunch->iov_len > 0) {
> + pr_debug("Punch!/%lx/%lx/\n", (unsigned long)bunch->iov_base, bunch->iov_len);
> + ret = fallocate(fd, FALLOC_FL_PUNCH_HOLE | FALLOC_FL_KEEP_SIZE,
> + (unsigned long)bunch->iov_base, bunch->iov_len);
> + if (ret != 0) {
> + pr_perror("Error punching hole");
> + return -1;
> + }
> + }
> + bunch->iov_base = (void *)off;
> + bunch->iov_len = len;
> + pr_debug("pr%d:New bunch/%lx/%lx/\n", pr_id,
> + (unsigned long)bunch->iov_base, bunch->iov_len);
> }
> return 0;
> }
> @@ -146,7 +163,8 @@ int dedup_one_iovec(struct page_read *pr, struct iovec *iov)
> piov_end = (unsigned long)piov.iov_base + piov.iov_len;
> off_real = lseek(pr->fd_pg, 0, SEEK_CUR);
> if (!pr->pe->in_parent) {
> - ret = punch_hole(pr->fd_pg, off_real, min(piov_end, iov_end) - off);
> + ret = punch_hole(pr->fd_pg, off_real, min(piov_end, iov_end) - off,
> + &pr->bunch, false, pr->id);
> if (ret == -1)
> return ret;
> }
> diff --git a/include/page-read.h b/include/page-read.h
> index 8b467c6..28435c7 100644
> --- a/include/page-read.h
> +++ b/include/page-read.h
> @@ -63,6 +63,8 @@ struct page_read {
> read_pagemap_page */
> unsigned long cvaddr; /* vaddr we are on */
>
> + struct iovec bunch; /* record consequent neighbour
> + iovecs to punch together */
> unsigned id; /* for logging */
> };
>
> @@ -72,5 +74,6 @@ extern void pagemap2iovec(PagemapEntry *pe, struct iovec *iov);
> extern int seek_pagemap_page(struct page_read *pr, unsigned long vaddr, bool warn);
>
> extern int dedup_one_iovec(struct page_read *pr, struct iovec *iov);
> -extern int punch_hole(int fd, unsigned long off, unsigned long len);
> +extern int punch_hole(int fd, unsigned long off, unsigned long len,
> + struct iovec * bunch, bool cleanup, int pr_id);
> #endif /* __CR_PAGE_READ_H__ */
> diff --git a/page-read.c b/page-read.c
> index e2514d6..0fd3464 100644
> --- a/page-read.c
> +++ b/page-read.c
> @@ -149,7 +149,8 @@ static int read_pagemap_page(struct page_read *pr, unsigned long vaddr, void *bu
> }
>
> if (opts.auto_dedup) {
> - ret = punch_hole(pr->fd_pg, current_vaddr, (unsigned int)PAGE_SIZE);
> + ret = punch_hole(pr->fd_pg, current_vaddr, (unsigned int)PAGE_SIZE,
> + &pr->bunch, false, pr->id);
> if (ret == -1) {
> return -1;
> }
> @@ -163,6 +164,15 @@ static int read_pagemap_page(struct page_read *pr, unsigned long vaddr, void *bu
>
> static void close_page_read(struct page_read *pr)
> {
> + int ret;
> + if (pr->bunch.iov_len > 0) {
> + ret = punch_hole(pr->fd_pg, 0, 0, &pr->bunch, true, pr->id);
> + if (ret == -1)
> + return;
>
>
> Here I'm not sure if I can do something like this, and what will be better way to do it.
> But this place is really suitable to make cleanup. So give me an advice, Thanks!
>
>
> Make close_page_read return int? Or it is unappropriate to make cleanup in close func?
If close fails there's nothing we can do about it :) Thus void is OK.
BTW, I tend to think, that even if punch-hole fails, we should only
emit a warning (pr_warn) but not fail the whole process.
>
>
>
> +
> + pr->bunch.iov_len = 0;
> + }
> +
> if (pr->parent) {
> close_page_read(pr->parent);
> xfree(pr->parent);
> @@ -207,6 +217,8 @@ err_cl:
> int open_page_read_at(int dfd, int pid, struct page_read *pr, int flags)
> {
> pr->pe = NULL;
> + pr->bunch.iov_len = 0;
> + pr->bunch.iov_base = NULL;
>
> pr->fd = open_image_at(dfd, CR_FD_PAGEMAP, O_RSTR, (long)pid);
> if (pr->fd < 0) {
> --
> 1.8.3.2
>
>
>
More information about the CRIU
mailing list