| Message ID | alpine.LSU.2.11.2104221347240.1170@eggly.anvils (mailing list archive) |
|---|---|
| State | New, archived |
| Headers | show |
| Series | None | expand |
On Thu, 22 Apr 2021 13:48:57 -0700 (PDT) Hugh Dickins <hughd@google.com> wrote: > Andrew, I'd have just sent a -fix.patch to remove the unnecessary u64s, > but need to reword the commit message: so please replace yesterday's > mm-filemap-fix-mapping_seek_hole_data-on-thp-32-bit.patch > by this one - thanks. Actually, I routinely update the base patch's changelog when queueing a -fix.
On Thu, 22 Apr 2021, Andrew Morton wrote: > On Thu, 22 Apr 2021 13:48:57 -0700 (PDT) Hugh Dickins <hughd@google.com> wrote: > > > Andrew, I'd have just sent a -fix.patch to remove the unnecessary u64s, > > but need to reword the commit message: so please replace yesterday's > > mm-filemap-fix-mapping_seek_hole_data-on-thp-32-bit.patch > > by this one - thanks. > > Actually, I routinely update the base patch's changelog when queueing a -fix. And thank you for that, but if there's time, I think we would still prefer the final commit message to include corrections where Matthew enlightened me (that "sign-extension" claim came from my confusion): -u64 casts added to stop unfortunate sign-extension when shifting (and -let's use shifts throughout, rather than mixed with * and /). - -Use round_up() when advancing pos, to stop assuming that pos was already -THP-aligned when advancing it by THP-size. (But I believe this use of -round_up() assumes that any THP must be THP-aligned: true while tmpfs -enforces that alignment, and is the only fs with FS_THP_SUPPORT; but might -need to be generalized in the future? If I try to generalize it right -now, I'm sure to get it wrong!) +u64 cast to stop losing bits when converting unsigned long to loff_t +(and let's use shifts throughout, rather than mixed with * and /). + +Use round_up() when advancing pos, to stop assuming that pos was +already THP-aligned when advancing it by THP-size. (This use of +round_up() assumes that any THP has THP-aligned index: true at present +and true going forward, but could be recoded to avoid the assumption.) Thanks, Hugh
On Fri, 23 Apr 2021 10:22:51 -0700 (PDT) Hugh Dickins <hughd@google.com> wrote: > On Thu, 22 Apr 2021, Andrew Morton wrote: > > On Thu, 22 Apr 2021 13:48:57 -0700 (PDT) Hugh Dickins <hughd@google.com> wrote: > > > > > Andrew, I'd have just sent a -fix.patch to remove the unnecessary u64s, > > > but need to reword the commit message: so please replace yesterday's > > > mm-filemap-fix-mapping_seek_hole_data-on-thp-32-bit.patch > > > by this one - thanks. > > > > Actually, I routinely update the base patch's changelog when queueing a -fix. > > And thank you for that, but if there's time, I think we would still > prefer the final commit message to include corrections where Matthew > enlightened me (that "sign-extension" claim came from my confusion): That's my point. When I merge a -v2 as a -fix, I replace the v1 patch's changelog with v2's changelog so everything works out after folding.
On Fri, 23 Apr 2021, Andrew Morton wrote: > On Fri, 23 Apr 2021 10:22:51 -0700 (PDT) Hugh Dickins <hughd@google.com> wrote: > > On Thu, 22 Apr 2021, Andrew Morton wrote: > > > On Thu, 22 Apr 2021 13:48:57 -0700 (PDT) Hugh Dickins <hughd@google.com> wrote: > > > > > > > Andrew, I'd have just sent a -fix.patch to remove the unnecessary u64s, > > > > but need to reword the commit message: so please replace yesterday's > > > > mm-filemap-fix-mapping_seek_hole_data-on-thp-32-bit.patch > > > > by this one - thanks. > > > > > > Actually, I routinely update the base patch's changelog when queueing a -fix. > > > > And thank you for that, but if there's time, I think we would still > > prefer the final commit message to include corrections where Matthew > > enlightened me (that "sign-extension" claim came from my confusion): > > That's my point. When I merge a -v2 as a -fix, I replace the v1 > patch's changelog with v2's changelog so everything works out after > folding. Oh, great, thanks: that was not clear to me, I feared you meant adding "v2: remove unneeded u64 casts" to v1 when merging. Hugh
--- 5.12-rc8/mm/filemap.c 2021-02-26 19:42:39.812156085 -0800 +++ linux/mm/filemap.c 2021-04-21 22:58:03.699655576 -0700 @@ -2672,7 +2672,7 @@ loff_t mapping_seek_hole_data(struct add loff_t end, int whence) { XA_STATE(xas, &mapping->i_pages, start >> PAGE_SHIFT); - pgoff_t max = (end - 1) / PAGE_SIZE; + pgoff_t max = (end - 1) >> PAGE_SHIFT; bool seek_data = (whence == SEEK_DATA); struct page *page; @@ -2681,7 +2681,8 @@ loff_t mapping_seek_hole_data(struct add rcu_read_lock(); while ((page = find_get_entry(&xas, max, XA_PRESENT))) { - loff_t pos = xas.xa_index * PAGE_SIZE; + loff_t pos = (u64)xas.xa_index << PAGE_SHIFT; + unsigned int seek_size; if (start < pos) { if (!seek_data) @@ -2689,25 +2690,25 @@ loff_t mapping_seek_hole_data(struct add start = pos; } - pos += seek_page_size(&xas, page); + seek_size = seek_page_size(&xas, page); + pos = round_up(pos + 1, seek_size); start = page_seek_hole_data(&xas, mapping, page, start, pos, seek_data); if (start < pos) goto unlock; + if (start >= end) + break; + if (seek_size > PAGE_SIZE) + xas_set(&xas, pos >> PAGE_SHIFT); if (!xa_is_value(page)) put_page(page); } - rcu_read_unlock(); - if (seek_data) - return -ENXIO; - goto out; - + start = -ENXIO; unlock: rcu_read_unlock(); - if (!xa_is_value(page)) + if (page && !xa_is_value(page)) put_page(page); -out: if (start > end) return end; return start;
No problem on 64-bit without huge pages, but xfstests generic/285 and other SEEK_HOLE/SEEK_DATA tests have regressed on huge tmpfs, and on 32-bit architectures, with the new mapping_seek_hole_data(). Several different bugs turned out to need fixing. u64 cast to stop losing bits when converting unsigned long to loff_t (and let's use shifts throughout, rather than mixed with * and /). Use round_up() when advancing pos, to stop assuming that pos was already THP-aligned when advancing it by THP-size. (This use of round_up() assumes that any THP has THP-aligned index: true at present and true going forward, but could be recoded to avoid the assumption.) Use xas_set() when iterating away from a THP, so that xa_index stays in synch with start, instead of drifting away to return bogus offset. Check start against end to avoid wrapping 32-bit xa_index to 0 (and to handle these additional cases, seek_data or not, it's easier to break the loop than goto: so rearrange exit from the function). Fixes: 41139aa4c3a3 ("mm/filemap: add mapping_seek_hole_data") Signed-off-by: Hugh Dickins <hughd@google.com> --- v2: Removed all but one of v1's u64 casts, as suggested my Matthew. Updated commit message on u64 cast and THP alignment, per Matthew. Andrew, I'd have just sent a -fix.patch to remove the unnecessary u64s, but need to reword the commit message: so please replace yesterday's mm-filemap-fix-mapping_seek_hole_data-on-thp-32-bit.patch by this one - thanks. mm/filemap.c | 21 +++++++++++---------- 1 file changed, 11 insertions(+), 10 deletions(-)