file-access errors on ghostscript >= 9.50

[stuart-little]

For newer gs versions every run of pdf2archive results in this error being thrown right at the top of the output:

GPL Ghostscript 9.54.0: Unrecoverable error, exit code 1

After some investigation, I tracked it down to a file-access error related to the "safe" mode that is the default in gs versions 9.50 and higher.

The following patch fixes it (it's literally a one-line change). I'd be happy to make that into a PR if you think it useful.

Patch:

Date:   Sat Jun 12 13:47:26 2021 -0400

    add '-dNOSAFER' option to eliminate gs file-access error

diff --git a/pdf2archive b/pdf2archive
index bac84be..e808079 100755
--- a/pdf2archive
+++ b/pdf2archive
@@ -249,7 +249,7 @@ quit
 #     use 'LC_CTYPE=C && LANG=C && echo "$METADUMP" ...' in the
 #     variable assignments; however, this produces bad PDF files.
 #
-METADUMP=$(gs -dNODISPLAY -q -sFile="$INPUT" $INFOTMPFILE | iconv -f utf-8 -t utf-8 -c)
+METADUMP=$(gs -dNOSAFER -dNODISPLAY -q -sFile="$INPUT" $INFOTMPFILE | iconv -f utf-8 -t utf-8 -c)
 [ -z ${PDFTITLE+x} ] && PDFTITLE=$(echo "$METADUMP" | grep "__knowninfoTitle: " | sed "s/^__knowninfoTitle: //g")
 [ -z ${PDFAUTHOR+x} ] && PDFAUTHOR=$(echo "$METADUMP" | grep "__knowninfoAuthor: " | sed "s/^__knowninfoAuthor: //g")
 [ -z ${PDFSUBJECT+x} ] && PDFSUBJECT=$(echo "$METADUMP" | grep "__knowninfoSubject: " | sed "s/^__knowninfoSubject: //g")

[matteosecli]

Hi @stuart-little, thanks for reporting.

I can confirm the issue with Ghostscript 9.53.3, as well as the cause. The PostScript commands in that file are based on the Ghostscript utility lib/pdf_info.ps; if I indeed try to run the command

$ gs -dNODISPLAY -sFile="$INPUT" <path_to_ghostscript_prefix>/lib/pdf_info.ps

I get the same error.

Can you please try to confirm that if you modify the metadump line into

METADUMP=$(gs -dNODISPLAY -q --permit-file-read="$INPUT" -sFile="$INPUT" $INFOTMPFILE | iconv -f utf-8 -t utf-8 -c)

that error goes away? It doesn't (completely) do so with lib/pdf_info.ps, but it should work correctly with the subset of commands I'm using in pdf2archive.

Unfortunately, it seems that the option --permit-file-read= breaks the compatibility with older Ghostscript versions, so right now I would momentarily go for the -dNOSAFER option, that albeit being "unsafe" it does indeed solve the issue. In the long term, however, we should detect the version of Ghostscript and use the --permit-file-read= flag in newer versions.

Can you please send a PR if you already have it as a commit on your local repo? Otherwise I'll just make a quick fix and credit you in the commit message.

Anyways, the version detection of Ghostscript is something I definitely have to do at some point, also to fix the annoying warning about -dUseCIEColor, that is necessary in older Ghostscript versions but it has to be replaced with different options in newer versions in order to get properly compliant files (blindly removing -dUseCIEColor can give non-compliant files).

---

More references on this issue:

• http://git.ghostscript.com/?p=ghostpdl.git;a=blob;f=lib/pdf_info.ps;hb=HEAD
• https://stackoverflow.com/questions/62524446/how-to-list-pdf-page-sizes-in-the-command-line
• https://stackoverflow.com/questions/40156190/ghostscript-error-invalidfileaccess-in-file
• https://stackoverflow.com/questions/54159444/accessing-files-from-ghostscript-commands-with-dsafer
• Ghostscript versions >9.50 have problem to get page count pimcore/pimcore#6911
• Add --permit-file-read flag to ghostscript command. pimcore/pimcore#6953

[stuart-little]

Hi @stuart-little, thanks for reporting.

Can you please try to confirm that if you modify the metadump line into

METADUMP=$(gs -dNODISPLAY -q --permit-file-read="$INPUT" -sFile="$INPUT" $INFOTMPFILE | iconv -f utf-8 -t utf-8 -c)

that error goes away?

Confirmed: that silences the error on the pdf I tested (which did exhibit the error before).

[matteosecli]

Ok, that's great! We'll then use -dNOSAFER as a temporary stopgap, until we conditionally add --permit-file-read="$INPUT" on newer Ghostscript versions in order to avoid that potentially unsafe option. 🙂

[matteosecli]

I've merged #18 but I'll keep this one open as a reminder that this should be a temporary fix. Thank you! 🙂

[stuart-little]

Thanks!