ssh breaks out of while-loop in bash [duplicate]

问题:

This question already has an answer here:

• Shell script while read line loop stops after the first line 4 answers

I use this bash-code to upload files to a remote server, for normal files this works fine:

for i in `find devel/ -newer $UPLOAD_FILE`
do
    echo \"Upload:\" $i
    if [ -d $i ]
    then
        echo \"Creating directory\" $i
        ssh $USER@$SERVER \"cd ${REMOTE_PATH}; mkdir -p $i\"
        continue
    fi
    if scp -Cp $i $USER@$SERVER:$REMOTE_PATH/$i
    then
        echo \"$i OK\"
    else
        echo \"$i NOK\"
        rm ${UPLOAD_FILE}_tmp
    fi
done

The only problem is that for files with a space in the name, the for-loop fails, so I replaced the first line like this:

find devel/ -newer $UPLOAD_FILE | while read i
do
    echo \"Upload:\" $i
    if [ -d $i ]
    then
        echo \"Creating directory\" $i
        ssh $USER@$SERVER \"cd ${REMOTE_PATH}; mkdir -p $i\"
        continue
    fi
    if scp -Cp $i $USER@$SERVER:$REMOTE_PATH/$i
    then
        echo \"$i OK\"
    else
        echo \"$i NOK\"
        rm ${UPLOAD_FILE}_tmp
    fi
done

For some strange reason, the ssh-command breaks out of the while-loop, therefore the first missing directory is created fine, but all subsequent missing files/directories are ignored.

I guess this has something to do with ssh writing something to stdout which confuses the \"read\" command. Commenting out the ssh-command makes the loop work as it should.

Does anybody know why this happens and how one can prevent ssh from breaking the while-loop?

回答1:

The problem is that ssh reads from standard input, therefore it eats all your remaining lines. You can just connect its standard input to nowhere:

ssh $USER@$SERVER \"cd ${REMOTE_PATH}; mkdir -p $i\" < /dev/null

You can also use ssh -n instead of the redirection.

回答2:

Another approach is to loop over a FD other than stdin:

while IFS= read -u 3 -r -d \'\' filename; do
  if [[ -d $filename ]]; then
    printf -v cmd_str \'cd %q; mkdir -p %q\' \"$REMOTE_PATH\" \"$filename\"
    ssh \"$USER@$SERVER\" \"$cmd_str\"
  else
    printf -v remote_path_str \'%q@%q:%q/%q\' \"$USER\" \"$SERVER\" \"$REMOTE_PATH\" \"$filename\"
    scp -Cp \"$filename\" \"$remote_path_str\"
  fi
done 3< <(find devel/ -newer \"$UPLOAD_FILE\" -print0)

The -u 3 and 3< operators are critical here, using FD 3 rather than the default FD 0 (stdin).

The approach given here -- using -print0, a cleared IFS value, and the like -- is also less buggy than the original code and the existing answer, which can\'t handle interesting filenames correctly. (Glenn Jackman\'s answer is close, but even that can\'t deal with filenames with newlines or filenames with trailing whitespace).

The use of printf %q is critical to generate commands which can\'t be used to attack the remote machine. Consider what would happen with a file named devel/$(rm -rf /)/hello with code which didn\'t have this paranoia.

回答3:

In addition to choroba\'s answer, don\'t use a for loop to read filenames:

find devel/ -newer $UPLOAD_FILE | 
while read -r i
do ...