Allow nbdkit-file-plugin to zero and trim block devices

resolves: RHEL-89353

0019-file-Fix-minor-typo-in-debug-message.patch

@@ -0,0 +1,26 @@
+From a79bf57c8ec805516e8dbe7995aa2bd46b83ade3 Mon Sep 17 00:00:00 2001
+From: "Richard W.M. Jones" <rjones@redhat.com>
+Date: Thu, 1 May 2025 10:03:06 +0100
+Subject: [PATCH] file: Fix minor typo in debug message
+
+(cherry picked from commit a75db5636b94c9184f8eb02fd51182d935df64a6)
+---
+ plugins/file/file.c | 2 +-
+ 1 file changed, 1 insertion(+), 1 deletion(-)
+
+diff --git a/plugins/file/file.c b/plugins/file/file.c
+index 6bcc5537..71b349ac 100644
+--- a/plugins/file/file.c
++++ b/plugins/file/file.c
+@@ -924,7 +924,7 @@ file_zero (void *handle, uint32_t count, uint64_t offset, uint32_t flags)
+     r = do_fallocate (h->fd, FALLOC_FL_ZERO_RANGE, offset, count);
+     if (r == 0) {
+       if (file_debug_zero)
+-        nbdkit_debug ("h->can_zero-range: "
++        nbdkit_debug ("h->can_zero_range: "
+                       "zero succeeded using fallocate");
+       goto out;
+     }
+-- 
+2.47.1
+

0020-file-Add-more-debugging-when-D-file.zero-1-is-used.patch

@@ -0,0 +1,36 @@
+From 1cb341e75c1a17553b69ea8d9889662e6d09ae78 Mon Sep 17 00:00:00 2001
+From: "Richard W.M. Jones" <rjones@redhat.com>
+Date: Thu, 1 May 2025 10:21:23 +0100
+Subject: [PATCH] file: Add more debugging when -D file.zero=1 is used
+
+(cherry picked from commit ecf6b15fa84a02b74ea969f06552c82ee418b9b4)
+---
+ plugins/file/file.c | 12 +++++++++++-
+ 1 file changed, 11 insertions(+), 1 deletion(-)
+
+diff --git a/plugins/file/file.c b/plugins/file/file.c
+index 71b349ac..378f6988 100644
+--- a/plugins/file/file.c
++++ b/plugins/file/file.c
+@@ -879,7 +879,17 @@ file_pwrite (void *handle, const void *buf, uint32_t count, uint64_t offset,
+ static int
+ do_fallocate (int fd, int mode_, off_t offset, off_t len)
+ {
+-  int r = fallocate (fd, mode_, offset, len);
++  int r;
++
++  r = fallocate (fd, mode_, offset, len);
++
++  if (file_debug_zero)
++    nbdkit_debug ("fallocate ([%s%s ], %" PRIu64 ", %" PRIu64") => %d (%d)",
++                  mode_ & FALLOC_FL_PUNCH_HOLE ? " FALLOC_FL_PUNCH_HOLE" : "",
++                  mode_ & FALLOC_FL_ZERO_RANGE ? " FALLOC_FL_ZERO_RANGE" : "",
++                  (uint64_t) offset, (uint64_t) len, r,
++                  r == -1 ? errno : 0);
++
+   if (r == -1 && errno == ENODEV) {
+     /* kernel 3.10 fails with ENODEV for block device. Kernel >= 4.9 fails
+        with EOPNOTSUPP in this case. Normalize errno to simplify callers. */
+-- 
+2.47.1
+

0021-file-Fix-comment-style-in-a-few-places.patch

@@ -0,0 +1,43 @@
+From 664e447d858a21304610db3023cc728db0c974bd Mon Sep 17 00:00:00 2001
+From: "Richard W.M. Jones" <rjones@redhat.com>
+Date: Thu, 1 May 2025 10:32:17 +0100
+Subject: [PATCH] file: Fix comment style in a few places
+
+No actual change here.
+
+(cherry picked from commit 0df4142c4be2b059c4d17aae0ec71f16ffc9ba35)
+---
+ plugins/file/file.c | 10 ++++++----
+ 1 file changed, 6 insertions(+), 4 deletions(-)
+
+diff --git a/plugins/file/file.c b/plugins/file/file.c
+index 378f6988..01ad1ef2 100644
+--- a/plugins/file/file.c
++++ b/plugins/file/file.c
+@@ -892,7 +892,8 @@ do_fallocate (int fd, int mode_, off_t offset, off_t len)
+ 
+   if (r == -1 && errno == ENODEV) {
+     /* kernel 3.10 fails with ENODEV for block device. Kernel >= 4.9 fails
+-       with EOPNOTSUPP in this case. Normalize errno to simplify callers. */
++     * with EOPNOTSUPP in this case. Normalize errno to simplify callers.
++     */
+     errno = EOPNOTSUPP;
+   }
+   return r;
+@@ -949,9 +950,10 @@ file_zero (void *handle, uint32_t count, uint64_t offset, uint32_t flags)
+ #endif
+ 
+ #ifdef FALLOC_FL_PUNCH_HOLE
+-  /* If we can punch hole but may not trim, we can combine punching hole and
+-   * fallocate to zero a range. This is expected to be more efficient than
+-   * writing zeroes manually. */
++  /* If we can punch hole but may not trim, we can combine punching
++   * hole and fallocate to zero a range. This is expected to be more
++   * efficient than writing zeroes manually.
++   */
+   if (h->can_punch_hole && h->can_fallocate) {
+     int r;
+ 
+-- 
+2.47.1
+

0022-file-Fix-do_fallocate-debugging-on-Alpine.patch

@@ -0,0 +1,60 @@
+From 4c02ff62f40497335da185cc4b45c2ba43fb609b Mon Sep 17 00:00:00 2001
+From: "Richard W.M. Jones" <rjones@redhat.com>
+Date: Thu, 1 May 2025 10:59:17 +0100
+Subject: [PATCH] file: Fix do_fallocate debugging on Alpine
+
+Alpine has some weird/old kernel that doesn't support
+FALLOC_FL_ZERO_RANGE but does support FALLOC_FL_PUNCH_HOLE, so the
+debugging I added in commit ecf6b15fa8 failed to compile with:
+
+  file.c: In function 'do_fallocate':
+  file.c:958:27: error: 'FALLOC_FL_ZERO_RANGE' undeclared (first use in this function)
+    958 |                   mode_ & FALLOC_FL_ZERO_RANGE ? " FALLOC_FL_ZERO_RANGE" : "",
+        |                           ^~~~~~~~~~~~~~~~~~~~
+  file.c:958:27: note: each undeclared identifier is reported only once for each function it appears in
+  make[3]: *** [Makefile:666: nbdkit_file_plugin_la-file.lo] Error 1
+
+Fixes: commit ecf6b15fa84a02b74ea969f06552c82ee418b9b4
+(cherry picked from commit 419a347054f81c53706637feddbc5008beab77d3)
+---
+ plugins/file/file.c | 15 +++++++++++++--
+ 1 file changed, 13 insertions(+), 2 deletions(-)
+
+diff --git a/plugins/file/file.c b/plugins/file/file.c
+index 01ad1ef2..32c5e2b7 100644
+--- a/plugins/file/file.c
++++ b/plugins/file/file.c
+@@ -875,7 +875,7 @@ file_pwrite (void *handle, const void *buf, uint32_t count, uint64_t offset,
+   return 0;
+ }
+ 
+-#if defined (FALLOC_FL_PUNCH_HOLE) || defined (FALLOC_FL_ZERO_RANGE)
++#if defined(FALLOC_FL_PUNCH_HOLE) || defined(FALLOC_FL_ZERO_RANGE)
+ static int
+ do_fallocate (int fd, int mode_, off_t offset, off_t len)
+ {
+@@ -884,9 +884,20 @@ do_fallocate (int fd, int mode_, off_t offset, off_t len)
+   r = fallocate (fd, mode_, offset, len);
+ 
+   if (file_debug_zero)
+-    nbdkit_debug ("fallocate ([%s%s ], %" PRIu64 ", %" PRIu64") => %d (%d)",
++    nbdkit_debug ("fallocate (["
++#if defined(FALLOC_FL_PUNCH_HOLE)
++                  "%s"
++#endif
++#if defined(FALLOC_FL_ZERO_RANGE)
++                  "%s"
++#endif
++                  " ], %" PRIu64 ", %" PRIu64") => %d (%d)",
++#if defined(FALLOC_FL_PUNCH_HOLE)
+                   mode_ & FALLOC_FL_PUNCH_HOLE ? " FALLOC_FL_PUNCH_HOLE" : "",
++#endif
++#if defined(FALLOC_FL_ZERO_RANGE)
+                   mode_ & FALLOC_FL_ZERO_RANGE ? " FALLOC_FL_ZERO_RANGE" : "",
++#endif
+                   (uint64_t) offset, (uint64_t) len, r,
+                   r == -1 ? errno : 0);
+ 
+-- 
+2.47.1
+

0023-file-Rename-h-can_zeroout-to-h-can_blkzeroout-to-ref.patch

@@ -0,0 +1,65 @@
+From bc4598f3d2d1ef2f4ebdf5b365ed08eff14d5654 Mon Sep 17 00:00:00 2001
+From: "Richard W.M. Jones" <rjones@redhat.com>
+Date: Thu, 1 May 2025 10:26:41 +0100
+Subject: [PATCH] file: Rename h->can_zeroout to h->can_blkzeroout to reflect
+ ioctl
+
+Since we're calling the blockdev-specific BLKZEROOUT ioctl when this
+flag is set, rename the flag.
+
+(cherry picked from commit fba20ce06c2f0e7c4be7e52e8e1934933851dfbc)
+---
+ plugins/file/file.c | 10 +++++-----
+ 1 file changed, 5 insertions(+), 5 deletions(-)
+
+diff --git a/plugins/file/file.c b/plugins/file/file.c
+index 32c5e2b7..70805bd7 100644
+--- a/plugins/file/file.c
++++ b/plugins/file/file.c
+@@ -497,7 +497,7 @@ struct handle {
+   bool can_punch_hole;
+   bool can_zero_range;
+   bool can_fallocate;
+-  bool can_zeroout;
++  bool can_blkzeroout;
+ };
+ 
+ /* Common code for opening a file by name, used by mode_filename and
+@@ -703,7 +703,7 @@ file_open (int readonly)
+ #endif
+ 
+   h->can_fallocate = true;
+-  h->can_zeroout = h->is_block_device;
++  h->can_blkzeroout = h->is_block_device;
+ 
+   return h;
+ }
+@@ -998,14 +998,14 @@ file_zero (void *handle, uint32_t count, uint64_t offset, uint32_t flags)
+ 
+ #ifdef BLKZEROOUT
+   /* For aligned range and block device, we can use BLKZEROOUT. */
+-  if (h->can_zeroout && IS_ALIGNED (offset | count, h->sector_size)) {
++  if (h->can_blkzeroout && IS_ALIGNED (offset | count, h->sector_size)) {
+     int r;
+     uint64_t range[2] = {offset, count};
+ 
+     r = ioctl (h->fd, BLKZEROOUT, &range);
+     if (r == 0) {
+       if (file_debug_zero)
+-        nbdkit_debug ("h->can_zeroout && IS_ALIGNED: "
++        nbdkit_debug ("h->can_blkzeroout && IS_ALIGNED: "
+                       "zero succeeded using BLKZEROOUT");
+       goto out;
+     }
+@@ -1015,7 +1015,7 @@ file_zero (void *handle, uint32_t count, uint64_t offset, uint32_t flags)
+       return -1;
+     }
+ 
+-    h->can_zeroout = false;
++    h->can_blkzeroout = false;
+   }
+ #endif
+ 
+-- 
+2.47.1
+

0024-file-zero-Document-implicit-order-that-we-will-try-z.patch

@@ -0,0 +1,38 @@
+From c1984ddcc6497c4446d1bf0e8828d1259852eb74 Mon Sep 17 00:00:00 2001
+From: "Richard W.M. Jones" <rjones@redhat.com>
+Date: Thu, 1 May 2025 10:30:41 +0100
+Subject: [PATCH] file: zero: Document implicit order that we will try zeroing
+ methods
+
+There's no substantive change here.  I just pulled out the test (flags
+& NBDKIT_FLAG_MAY_TRIM) into a boolean variable, and documented that
+we (will) try zero-with-trim methods first.
+
+(cherry picked from commit 61fc023f235b17f8a19302885d1613dd0a7a3793)
+---
+ plugins/file/file.c | 7 ++++++-
+ 1 file changed, 6 insertions(+), 1 deletion(-)
+
+diff --git a/plugins/file/file.c b/plugins/file/file.c
+index 70805bd7..3b82e02d 100644
+--- a/plugins/file/file.c
++++ b/plugins/file/file.c
+@@ -916,9 +916,14 @@ static int
+ file_zero (void *handle, uint32_t count, uint64_t offset, uint32_t flags)
+ {
+   struct handle *h __attribute__ ((unused)) = handle;
++  const bool may_trim __attribute__ ((unused)) = flags & NBDKIT_FLAG_MAY_TRIM;
+ 
++  /* These alternate zeroing methods are ordered.  Methods which can
++   * trim (if may_trim is set) are tried first.  Methods which can
++   * only zero are tried last.
++   */
+ #ifdef FALLOC_FL_PUNCH_HOLE
+-  if (h->can_punch_hole && (flags & NBDKIT_FLAG_MAY_TRIM)) {
++  if (may_trim && h->can_punch_hole) {
+     int r;
+ 
+     r = do_fallocate (h->fd, FALLOC_FL_PUNCH_HOLE | FALLOC_FL_KEEP_SIZE,
+-- 
+2.47.1
+

0025-file-zero-Use-BLKDISCARD-method-if-may_trim-is-set.patch

@@ -0,0 +1,122 @@
+From 396e8a97835155a620cabbcf1aabaaa1fa4a08f1 Mon Sep 17 00:00:00 2001
+From: "Richard W.M. Jones" <rjones@redhat.com>
+Date: Thu, 1 May 2025 10:36:23 +0100
+Subject: [PATCH] file: zero: Use BLKDISCARD method if may_trim is set
+
+If we're allowed to trim and we're writing to a block device,
+previously we hit the case fallocate(FALLOC_FL_ZERO_RANGE) first.
+This succeeds in Linux, zeroing (not trimming) the range.
+
+However it would be better to trim in this case.  Linux supports
+ioctl(BLKDISCARD) on block devices, so try this method first.
+
+Fixes: https://issues.redhat.com/browse/RHEL-89353
+Reported-by: Germano Veit Michel
+Thanks: Eric Blake
+(cherry picked from commit 7a9ecda24906c64d9f8c7238a96cb3f686e894eb)
+---
+ plugins/file/file.c                 | 50 +++++++++++++++++++++++++++++
+ plugins/file/nbdkit-file-plugin.pod |  5 +++
+ 2 files changed, 55 insertions(+)
+
+diff --git a/plugins/file/file.c b/plugins/file/file.c
+index 3b82e02d..b4dec3c5 100644
+--- a/plugins/file/file.c
++++ b/plugins/file/file.c
+@@ -397,6 +397,9 @@ file_dump_plugin (void)
+ #ifdef BLKSSZGET
+   printf ("file_blksszget=yes\n");
+ #endif
++#ifdef BLKDISCARD
++  printf ("file_blkdiscard=yes\n");
++#endif
+ #ifdef BLKZEROOUT
+   printf ("file_blkzeroout=yes\n");
+ #endif
+@@ -497,6 +500,7 @@ struct handle {
+   bool can_punch_hole;
+   bool can_zero_range;
+   bool can_fallocate;
++  bool can_blkdiscard;
+   bool can_blkzeroout;
+ };
+ 
+@@ -704,6 +708,7 @@ file_open (int readonly)
+ 
+   h->can_fallocate = true;
+   h->can_blkzeroout = h->is_block_device;
++  h->can_blkdiscard = h->is_block_device;
+ 
+   return h;
+ }
+@@ -944,6 +949,51 @@ file_zero (void *handle, uint32_t count, uint64_t offset, uint32_t flags)
+   }
+ #endif
+ 
++#if defined(BLKDISCARD) && defined(FALLOC_FL_ZERO_RANGE)
++  /* For aligned range and block device, we can use BLKDISCARD to
++   * trim.  However BLKDISCARD doesn't necessarily zero (eg for local
++   * disk) so we have to zero first and then discard.
++   *
++   * In future all Linux block devices may understand
++   * FALLOC_FL_PUNCH_HOLE which means this case would no longer be
++   * necessary, since the case above will handle it.
++   */
++  if (may_trim && h->can_blkdiscard && h->can_zero_range &&
++      IS_ALIGNED (offset | count, h->sector_size)) {
++    int r;
++    uint64_t range[2] = {offset, count};
++
++    r = do_fallocate (h->fd, FALLOC_FL_ZERO_RANGE, offset, count);
++    if (r == 0) {
++      /* We could use FALLOC_FL_PUNCH_HOLE here instead, but currently
++       * thin LVs do not support it (XXX 2025-04).
++       */
++      r = ioctl (h->fd, BLKDISCARD, &range);
++      if (r == 0) {
++        if (file_debug_zero)
++          nbdkit_debug ("h->can_blkdiscard && may_trim && IS_ALIGNED: "
++                        "zero succeeded using BLKDISCARD");
++        goto out;
++      }
++
++      if (!is_enotsup (errno)) {
++        nbdkit_error ("zero: %m");
++        return -1;
++      }
++
++      h->can_blkdiscard = false;
++    }
++    else {
++      if (!is_enotsup (errno)) {
++        nbdkit_error ("zero: %m");
++        return -1;
++      }
++
++      h->can_fallocate = false;
++    }
++  }
++#endif
++
+ #ifdef FALLOC_FL_ZERO_RANGE
+   if (h->can_zero_range) {
+     int r;
+diff --git a/plugins/file/nbdkit-file-plugin.pod b/plugins/file/nbdkit-file-plugin.pod
+index a50bef2d..0e260b7f 100644
+--- a/plugins/file/nbdkit-file-plugin.pod
++++ b/plugins/file/nbdkit-file-plugin.pod
+@@ -227,6 +227,11 @@ future.
+ If both set, the plugin may be able to efficiently zero ranges of
+ block devices, where the driver and block device itself supports this.
+ 
++=item C<file_blkdiscard=yes>
++
++If set, the plugin may be able to efficiently trim ranges of block
++devices, where the driver and block device itself supports this.
++
+ =item C<file_extents=yes>
+ 
+ If set, the plugin can read file extents.
+-- 
+2.47.1
+

nbdkit.spec

@@ -99,6 +99,13 @@ Patch0015:     0015-file-Hard-error-if-sync_file_range-fails.patch
 Patch0016:     0016-file-Reduce-the-size-of-the-lock-around-write-evicti.patch
 Patch0017:     0017-file-Document-implicit-assumption-about-eviction-win.patch
 Patch0018:     0018-server-Turn-flush-into-a-controlpath-message.patch
+Patch0019:     0019-file-Fix-minor-typo-in-debug-message.patch
+Patch0020:     0020-file-Add-more-debugging-when-D-file.zero-1-is-used.patch
+Patch0021:     0021-file-Fix-comment-style-in-a-few-places.patch
+Patch0022:     0022-file-Fix-do_fallocate-debugging-on-Alpine.patch
+Patch0023:     0023-file-Rename-h-can_zeroout-to-h-can_blkzeroout-to-ref.patch
+Patch0024:     0024-file-zero-Document-implicit-order-that-we-will-try-z.patch
+Patch0025:     0025-file-zero-Use-BLKDISCARD-method-if-may_trim-is-set.patch
 
 # For automatic RPM Provides generation.
 # See: https://rpm-software-management.github.io/rpm/manual/dependency_generators.html
@@ -1517,9 +1524,11 @@ fi
 
 
 %changelog
-* Mon Apr 07 2025 Richard W.M. Jones <rjones@redhat.com> - 1.38.5-5
+* Thu May 01 2025 Richard W.M. Jones <rjones@redhat.com> - 1.38.5-6
 - Add extra system call checking and debugging to nbdkit-file-plugin
   resolves: RHEL-85510
+- Allow nbdkit-file-plugin to zero and trim block devices
+  resolves: RHEL-89353
 
 * Mon Jan 06 2025 Richard W.M. Jones <rjones@redhat.com> - 1.38.5-2
 - vddk: Avoid reading partial chunk beyond the end of the disk